xref: /titanic_44/usr/src/uts/common/fs/zfs/spa.c (revision 5151fb1220e0ceafdc172203863c73da4285c170)
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 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * This file contains all the routines used when modifying on-disk SPA state.
31  * This includes opening, importing, destroying, exporting a pool, and syncing a
32  * pool.
33  */
34 
35 #include <sys/zfs_context.h>
36 #include <sys/fm/fs/zfs.h>
37 #include <sys/spa_impl.h>
38 #include <sys/zio.h>
39 #include <sys/zio_checksum.h>
40 #include <sys/zio_compress.h>
41 #include <sys/dmu.h>
42 #include <sys/dmu_tx.h>
43 #include <sys/zap.h>
44 #include <sys/zil.h>
45 #include <sys/vdev_impl.h>
46 #include <sys/metaslab.h>
47 #include <sys/uberblock_impl.h>
48 #include <sys/txg.h>
49 #include <sys/avl.h>
50 #include <sys/dmu_traverse.h>
51 #include <sys/dmu_objset.h>
52 #include <sys/unique.h>
53 #include <sys/dsl_pool.h>
54 #include <sys/dsl_dataset.h>
55 #include <sys/dsl_dir.h>
56 #include <sys/dsl_prop.h>
57 #include <sys/dsl_synctask.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/callb.h>
60 #include <sys/systeminfo.h>
61 #include <sys/sunddi.h>
62 
63 int zio_taskq_threads = 8;
64 
65 /*
66  * ==========================================================================
67  * SPA state manipulation (open/create/destroy/import/export)
68  * ==========================================================================
69  */
70 
71 static int
72 spa_error_entry_compare(const void *a, const void *b)
73 {
74 	spa_error_entry_t *sa = (spa_error_entry_t *)a;
75 	spa_error_entry_t *sb = (spa_error_entry_t *)b;
76 	int ret;
77 
78 	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
79 	    sizeof (zbookmark_t));
80 
81 	if (ret < 0)
82 		return (-1);
83 	else if (ret > 0)
84 		return (1);
85 	else
86 		return (0);
87 }
88 
89 /*
90  * Utility function which retrieves copies of the current logs and
91  * re-initializes them in the process.
92  */
93 void
94 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
95 {
96 	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
97 
98 	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
99 	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
100 
101 	avl_create(&spa->spa_errlist_scrub,
102 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
103 	    offsetof(spa_error_entry_t, se_avl));
104 	avl_create(&spa->spa_errlist_last,
105 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
106 	    offsetof(spa_error_entry_t, se_avl));
107 }
108 
109 /*
110  * Activate an uninitialized pool.
111  */
112 static void
113 spa_activate(spa_t *spa)
114 {
115 	int t;
116 
117 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
118 
119 	spa->spa_state = POOL_STATE_ACTIVE;
120 
121 	spa->spa_normal_class = metaslab_class_create();
122 
123 	for (t = 0; t < ZIO_TYPES; t++) {
124 		spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue",
125 		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
126 		    TASKQ_PREPOPULATE);
127 		spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr",
128 		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
129 		    TASKQ_PREPOPULATE);
130 	}
131 
132 	rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL);
133 
134 	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
135 	mutex_init(&spa->spa_config_cache_lock, NULL, MUTEX_DEFAULT, NULL);
136 	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
137 	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
138 	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
139 	mutex_init(&spa->spa_config_lock.scl_lock, NULL, MUTEX_DEFAULT, NULL);
140 	mutex_init(&spa->spa_sync_bplist.bpl_lock, NULL, MUTEX_DEFAULT, NULL);
141 	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
142 	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
143 
144 	list_create(&spa->spa_dirty_list, sizeof (vdev_t),
145 	    offsetof(vdev_t, vdev_dirty_node));
146 
147 	txg_list_create(&spa->spa_vdev_txg_list,
148 	    offsetof(struct vdev, vdev_txg_node));
149 
150 	avl_create(&spa->spa_errlist_scrub,
151 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
152 	    offsetof(spa_error_entry_t, se_avl));
153 	avl_create(&spa->spa_errlist_last,
154 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
155 	    offsetof(spa_error_entry_t, se_avl));
156 }
157 
158 /*
159  * Opposite of spa_activate().
160  */
161 static void
162 spa_deactivate(spa_t *spa)
163 {
164 	int t;
165 
166 	ASSERT(spa->spa_sync_on == B_FALSE);
167 	ASSERT(spa->spa_dsl_pool == NULL);
168 	ASSERT(spa->spa_root_vdev == NULL);
169 
170 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
171 
172 	txg_list_destroy(&spa->spa_vdev_txg_list);
173 
174 	list_destroy(&spa->spa_dirty_list);
175 
176 	rw_destroy(&spa->spa_traverse_lock);
177 
178 	for (t = 0; t < ZIO_TYPES; t++) {
179 		taskq_destroy(spa->spa_zio_issue_taskq[t]);
180 		taskq_destroy(spa->spa_zio_intr_taskq[t]);
181 		spa->spa_zio_issue_taskq[t] = NULL;
182 		spa->spa_zio_intr_taskq[t] = NULL;
183 	}
184 
185 	metaslab_class_destroy(spa->spa_normal_class);
186 	spa->spa_normal_class = NULL;
187 
188 	/*
189 	 * If this was part of an import or the open otherwise failed, we may
190 	 * still have errors left in the queues.  Empty them just in case.
191 	 */
192 	spa_errlog_drain(spa);
193 
194 	avl_destroy(&spa->spa_errlist_scrub);
195 	avl_destroy(&spa->spa_errlist_last);
196 
197 	spa->spa_state = POOL_STATE_UNINITIALIZED;
198 }
199 
200 /*
201  * Verify a pool configuration, and construct the vdev tree appropriately.  This
202  * will create all the necessary vdevs in the appropriate layout, with each vdev
203  * in the CLOSED state.  This will prep the pool before open/creation/import.
204  * All vdev validation is done by the vdev_alloc() routine.
205  */
206 static int
207 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
208     uint_t id, int atype)
209 {
210 	nvlist_t **child;
211 	uint_t c, children;
212 	int error;
213 
214 	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
215 		return (error);
216 
217 	if ((*vdp)->vdev_ops->vdev_op_leaf)
218 		return (0);
219 
220 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
221 	    &child, &children) != 0) {
222 		vdev_free(*vdp);
223 		*vdp = NULL;
224 		return (EINVAL);
225 	}
226 
227 	for (c = 0; c < children; c++) {
228 		vdev_t *vd;
229 		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
230 		    atype)) != 0) {
231 			vdev_free(*vdp);
232 			*vdp = NULL;
233 			return (error);
234 		}
235 	}
236 
237 	ASSERT(*vdp != NULL);
238 
239 	return (0);
240 }
241 
242 /*
243  * Opposite of spa_load().
244  */
245 static void
246 spa_unload(spa_t *spa)
247 {
248 	int i;
249 
250 	/*
251 	 * Stop async tasks.
252 	 */
253 	spa_async_suspend(spa);
254 
255 	/*
256 	 * Stop syncing.
257 	 */
258 	if (spa->spa_sync_on) {
259 		txg_sync_stop(spa->spa_dsl_pool);
260 		spa->spa_sync_on = B_FALSE;
261 	}
262 
263 	/*
264 	 * Wait for any outstanding prefetch I/O to complete.
265 	 */
266 	spa_config_enter(spa, RW_WRITER, FTAG);
267 	spa_config_exit(spa, FTAG);
268 
269 	/*
270 	 * Close the dsl pool.
271 	 */
272 	if (spa->spa_dsl_pool) {
273 		dsl_pool_close(spa->spa_dsl_pool);
274 		spa->spa_dsl_pool = NULL;
275 	}
276 
277 	/*
278 	 * Close all vdevs.
279 	 */
280 	if (spa->spa_root_vdev)
281 		vdev_free(spa->spa_root_vdev);
282 	ASSERT(spa->spa_root_vdev == NULL);
283 
284 	for (i = 0; i < spa->spa_nspares; i++)
285 		vdev_free(spa->spa_spares[i]);
286 	if (spa->spa_spares) {
287 		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
288 		spa->spa_spares = NULL;
289 	}
290 	if (spa->spa_sparelist) {
291 		nvlist_free(spa->spa_sparelist);
292 		spa->spa_sparelist = NULL;
293 	}
294 
295 	spa->spa_async_suspended = 0;
296 }
297 
298 /*
299  * Load (or re-load) the current list of vdevs describing the active spares for
300  * this pool.  When this is called, we have some form of basic information in
301  * 'spa_sparelist'.  We parse this into vdevs, try to open them, and then
302  * re-generate a more complete list including status information.
303  */
304 static void
305 spa_load_spares(spa_t *spa)
306 {
307 	nvlist_t **spares;
308 	uint_t nspares;
309 	int i;
310 	vdev_t *vd, *tvd;
311 
312 	/*
313 	 * First, close and free any existing spare vdevs.
314 	 */
315 	for (i = 0; i < spa->spa_nspares; i++) {
316 		vd = spa->spa_spares[i];
317 
318 		/* Undo the call to spa_activate() below */
319 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL &&
320 		    tvd->vdev_isspare)
321 			spa_spare_remove(tvd);
322 		vdev_close(vd);
323 		vdev_free(vd);
324 	}
325 
326 	if (spa->spa_spares)
327 		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
328 
329 	if (spa->spa_sparelist == NULL)
330 		nspares = 0;
331 	else
332 		VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
333 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
334 
335 	spa->spa_nspares = (int)nspares;
336 	spa->spa_spares = NULL;
337 
338 	if (nspares == 0)
339 		return;
340 
341 	/*
342 	 * Construct the array of vdevs, opening them to get status in the
343 	 * process.   For each spare, there is potentially two different vdev_t
344 	 * structures associated with it: one in the list of spares (used only
345 	 * for basic validation purposes) and one in the active vdev
346 	 * configuration (if it's spared in).  During this phase we open and
347 	 * validate each vdev on the spare list.  If the vdev also exists in the
348 	 * active configuration, then we also mark this vdev as an active spare.
349 	 */
350 	spa->spa_spares = kmem_alloc(nspares * sizeof (void *), KM_SLEEP);
351 	for (i = 0; i < spa->spa_nspares; i++) {
352 		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
353 		    VDEV_ALLOC_SPARE) == 0);
354 		ASSERT(vd != NULL);
355 
356 		spa->spa_spares[i] = vd;
357 
358 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL) {
359 			if (!tvd->vdev_isspare)
360 				spa_spare_add(tvd);
361 
362 			/*
363 			 * We only mark the spare active if we were successfully
364 			 * able to load the vdev.  Otherwise, importing a pool
365 			 * with a bad active spare would result in strange
366 			 * behavior, because multiple pool would think the spare
367 			 * is actively in use.
368 			 *
369 			 * There is a vulnerability here to an equally bizarre
370 			 * circumstance, where a dead active spare is later
371 			 * brought back to life (onlined or otherwise).  Given
372 			 * the rarity of this scenario, and the extra complexity
373 			 * it adds, we ignore the possibility.
374 			 */
375 			if (!vdev_is_dead(tvd))
376 				spa_spare_activate(tvd);
377 		}
378 
379 		if (vdev_open(vd) != 0)
380 			continue;
381 
382 		vd->vdev_top = vd;
383 		(void) vdev_validate_spare(vd);
384 	}
385 
386 	/*
387 	 * Recompute the stashed list of spares, with status information
388 	 * this time.
389 	 */
390 	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
391 	    DATA_TYPE_NVLIST_ARRAY) == 0);
392 
393 	spares = kmem_alloc(spa->spa_nspares * sizeof (void *), KM_SLEEP);
394 	for (i = 0; i < spa->spa_nspares; i++)
395 		spares[i] = vdev_config_generate(spa, spa->spa_spares[i],
396 		    B_TRUE, B_TRUE);
397 	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
398 	    spares, spa->spa_nspares) == 0);
399 	for (i = 0; i < spa->spa_nspares; i++)
400 		nvlist_free(spares[i]);
401 	kmem_free(spares, spa->spa_nspares * sizeof (void *));
402 }
403 
404 static int
405 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
406 {
407 	dmu_buf_t *db;
408 	char *packed = NULL;
409 	size_t nvsize = 0;
410 	int error;
411 	*value = NULL;
412 
413 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
414 	nvsize = *(uint64_t *)db->db_data;
415 	dmu_buf_rele(db, FTAG);
416 
417 	packed = kmem_alloc(nvsize, KM_SLEEP);
418 	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
419 	if (error == 0)
420 		error = nvlist_unpack(packed, nvsize, value, 0);
421 	kmem_free(packed, nvsize);
422 
423 	return (error);
424 }
425 
426 /*
427  * Load an existing storage pool, using the pool's builtin spa_config as a
428  * source of configuration information.
429  */
430 static int
431 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
432 {
433 	int error = 0;
434 	nvlist_t *nvroot = NULL;
435 	vdev_t *rvd;
436 	uberblock_t *ub = &spa->spa_uberblock;
437 	uint64_t config_cache_txg = spa->spa_config_txg;
438 	uint64_t pool_guid;
439 	uint64_t version;
440 	zio_t *zio;
441 
442 	spa->spa_load_state = state;
443 
444 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
445 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
446 		error = EINVAL;
447 		goto out;
448 	}
449 
450 	/*
451 	 * Versioning wasn't explicitly added to the label until later, so if
452 	 * it's not present treat it as the initial version.
453 	 */
454 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
455 		version = ZFS_VERSION_INITIAL;
456 
457 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
458 	    &spa->spa_config_txg);
459 
460 	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
461 	    spa_guid_exists(pool_guid, 0)) {
462 		error = EEXIST;
463 		goto out;
464 	}
465 
466 	spa->spa_load_guid = pool_guid;
467 
468 	/*
469 	 * Parse the configuration into a vdev tree.  We explicitly set the
470 	 * value that will be returned by spa_version() since parsing the
471 	 * configuration requires knowing the version number.
472 	 */
473 	spa_config_enter(spa, RW_WRITER, FTAG);
474 	spa->spa_ubsync.ub_version = version;
475 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
476 	spa_config_exit(spa, FTAG);
477 
478 	if (error != 0)
479 		goto out;
480 
481 	ASSERT(spa->spa_root_vdev == rvd);
482 	ASSERT(spa_guid(spa) == pool_guid);
483 
484 	/*
485 	 * Try to open all vdevs, loading each label in the process.
486 	 */
487 	if (vdev_open(rvd) != 0) {
488 		error = ENXIO;
489 		goto out;
490 	}
491 
492 	/*
493 	 * Validate the labels for all leaf vdevs.  We need to grab the config
494 	 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD
495 	 * flag.
496 	 */
497 	spa_config_enter(spa, RW_READER, FTAG);
498 	error = vdev_validate(rvd);
499 	spa_config_exit(spa, FTAG);
500 
501 	if (error != 0) {
502 		error = EBADF;
503 		goto out;
504 	}
505 
506 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
507 		error = ENXIO;
508 		goto out;
509 	}
510 
511 	/*
512 	 * Find the best uberblock.
513 	 */
514 	bzero(ub, sizeof (uberblock_t));
515 
516 	zio = zio_root(spa, NULL, NULL,
517 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
518 	vdev_uberblock_load(zio, rvd, ub);
519 	error = zio_wait(zio);
520 
521 	/*
522 	 * If we weren't able to find a single valid uberblock, return failure.
523 	 */
524 	if (ub->ub_txg == 0) {
525 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
526 		    VDEV_AUX_CORRUPT_DATA);
527 		error = ENXIO;
528 		goto out;
529 	}
530 
531 	/*
532 	 * If the pool is newer than the code, we can't open it.
533 	 */
534 	if (ub->ub_version > ZFS_VERSION) {
535 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
536 		    VDEV_AUX_VERSION_NEWER);
537 		error = ENOTSUP;
538 		goto out;
539 	}
540 
541 	/*
542 	 * If the vdev guid sum doesn't match the uberblock, we have an
543 	 * incomplete configuration.
544 	 */
545 	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
546 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
547 		    VDEV_AUX_BAD_GUID_SUM);
548 		error = ENXIO;
549 		goto out;
550 	}
551 
552 	/*
553 	 * Initialize internal SPA structures.
554 	 */
555 	spa->spa_state = POOL_STATE_ACTIVE;
556 	spa->spa_ubsync = spa->spa_uberblock;
557 	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
558 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
559 	if (error) {
560 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
561 		    VDEV_AUX_CORRUPT_DATA);
562 		goto out;
563 	}
564 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
565 
566 	if (zap_lookup(spa->spa_meta_objset,
567 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
568 	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
569 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
570 		    VDEV_AUX_CORRUPT_DATA);
571 		error = EIO;
572 		goto out;
573 	}
574 
575 	if (!mosconfig) {
576 		nvlist_t *newconfig;
577 		uint64_t hostid;
578 
579 		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
580 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
581 			    VDEV_AUX_CORRUPT_DATA);
582 			error = EIO;
583 			goto out;
584 		}
585 
586 		if (nvlist_lookup_uint64(newconfig, ZPOOL_CONFIG_HOSTID,
587 		    &hostid) == 0) {
588 			char *hostname;
589 			unsigned long myhostid = 0;
590 
591 			VERIFY(nvlist_lookup_string(newconfig,
592 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
593 
594 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
595 			if ((unsigned long)hostid != myhostid) {
596 				cmn_err(CE_WARN, "pool '%s' could not be "
597 				    "loaded as it was last accessed by "
598 				    "another system (host: %s hostid: 0x%lx).  "
599 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
600 				    spa->spa_name, hostname,
601 				    (unsigned long)hostid);
602 				error = EBADF;
603 				goto out;
604 			}
605 		}
606 
607 		spa_config_set(spa, newconfig);
608 		spa_unload(spa);
609 		spa_deactivate(spa);
610 		spa_activate(spa);
611 
612 		return (spa_load(spa, newconfig, state, B_TRUE));
613 	}
614 
615 	if (zap_lookup(spa->spa_meta_objset,
616 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
617 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
618 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
619 		    VDEV_AUX_CORRUPT_DATA);
620 		error = EIO;
621 		goto out;
622 	}
623 
624 	/*
625 	 * Load the bit that tells us to use the new accounting function
626 	 * (raid-z deflation).  If we have an older pool, this will not
627 	 * be present.
628 	 */
629 	error = zap_lookup(spa->spa_meta_objset,
630 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
631 	    sizeof (uint64_t), 1, &spa->spa_deflate);
632 	if (error != 0 && error != ENOENT) {
633 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
634 		    VDEV_AUX_CORRUPT_DATA);
635 		error = EIO;
636 		goto out;
637 	}
638 
639 	/*
640 	 * Load the persistent error log.  If we have an older pool, this will
641 	 * not be present.
642 	 */
643 	error = zap_lookup(spa->spa_meta_objset,
644 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
645 	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
646 	if (error != 0 && error != ENOENT) {
647 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
648 		    VDEV_AUX_CORRUPT_DATA);
649 		error = EIO;
650 		goto out;
651 	}
652 
653 	error = zap_lookup(spa->spa_meta_objset,
654 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
655 	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
656 	if (error != 0 && error != ENOENT) {
657 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
658 		    VDEV_AUX_CORRUPT_DATA);
659 		error = EIO;
660 		goto out;
661 	}
662 
663 	/*
664 	 * Load the history object.  If we have an older pool, this
665 	 * will not be present.
666 	 */
667 	error = zap_lookup(spa->spa_meta_objset,
668 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
669 	    sizeof (uint64_t), 1, &spa->spa_history);
670 	if (error != 0 && error != ENOENT) {
671 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
672 		    VDEV_AUX_CORRUPT_DATA);
673 		error = EIO;
674 		goto out;
675 	}
676 
677 	/*
678 	 * Load any hot spares for this pool.
679 	 */
680 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
681 	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object);
682 	if (error != 0 && error != ENOENT) {
683 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
684 		    VDEV_AUX_CORRUPT_DATA);
685 		error = EIO;
686 		goto out;
687 	}
688 	if (error == 0) {
689 		ASSERT(spa_version(spa) >= ZFS_VERSION_SPARES);
690 		if (load_nvlist(spa, spa->spa_spares_object,
691 		    &spa->spa_sparelist) != 0) {
692 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
693 			    VDEV_AUX_CORRUPT_DATA);
694 			error = EIO;
695 			goto out;
696 		}
697 
698 		spa_config_enter(spa, RW_WRITER, FTAG);
699 		spa_load_spares(spa);
700 		spa_config_exit(spa, FTAG);
701 	}
702 
703 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
704 	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
705 
706 	if (error && error != ENOENT) {
707 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
708 		    VDEV_AUX_CORRUPT_DATA);
709 		error = EIO;
710 		goto out;
711 	}
712 
713 	if (error == 0) {
714 		(void) zap_lookup(spa->spa_meta_objset,
715 		    spa->spa_pool_props_object,
716 		    zpool_prop_to_name(ZFS_PROP_BOOTFS),
717 		    sizeof (uint64_t), 1, &spa->spa_bootfs);
718 	}
719 
720 	/*
721 	 * Load the vdev state for all toplevel vdevs.
722 	 */
723 	vdev_load(rvd);
724 
725 	/*
726 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
727 	 */
728 	spa_config_enter(spa, RW_WRITER, FTAG);
729 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
730 	spa_config_exit(spa, FTAG);
731 
732 	/*
733 	 * Check the state of the root vdev.  If it can't be opened, it
734 	 * indicates one or more toplevel vdevs are faulted.
735 	 */
736 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
737 		error = ENXIO;
738 		goto out;
739 	}
740 
741 	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
742 		dmu_tx_t *tx;
743 		int need_update = B_FALSE;
744 		int c;
745 
746 		/*
747 		 * Claim log blocks that haven't been committed yet.
748 		 * This must all happen in a single txg.
749 		 */
750 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
751 		    spa_first_txg(spa));
752 		(void) dmu_objset_find(spa->spa_name,
753 		    zil_claim, tx, DS_FIND_CHILDREN);
754 		dmu_tx_commit(tx);
755 
756 		spa->spa_sync_on = B_TRUE;
757 		txg_sync_start(spa->spa_dsl_pool);
758 
759 		/*
760 		 * Wait for all claims to sync.
761 		 */
762 		txg_wait_synced(spa->spa_dsl_pool, 0);
763 
764 		/*
765 		 * If the config cache is stale, or we have uninitialized
766 		 * metaslabs (see spa_vdev_add()), then update the config.
767 		 */
768 		if (config_cache_txg != spa->spa_config_txg ||
769 		    state == SPA_LOAD_IMPORT)
770 			need_update = B_TRUE;
771 
772 		for (c = 0; c < rvd->vdev_children; c++)
773 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
774 				need_update = B_TRUE;
775 
776 		/*
777 		 * Update the config cache asychronously in case we're the
778 		 * root pool, in which case the config cache isn't writable yet.
779 		 */
780 		if (need_update)
781 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
782 	}
783 
784 	error = 0;
785 out:
786 	if (error && error != EBADF)
787 		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
788 	spa->spa_load_state = SPA_LOAD_NONE;
789 	spa->spa_ena = 0;
790 
791 	return (error);
792 }
793 
794 /*
795  * Pool Open/Import
796  *
797  * The import case is identical to an open except that the configuration is sent
798  * down from userland, instead of grabbed from the configuration cache.  For the
799  * case of an open, the pool configuration will exist in the
800  * POOL_STATE_UNITIALIZED state.
801  *
802  * The stats information (gen/count/ustats) is used to gather vdev statistics at
803  * the same time open the pool, without having to keep around the spa_t in some
804  * ambiguous state.
805  */
806 static int
807 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
808 {
809 	spa_t *spa;
810 	int error;
811 	int loaded = B_FALSE;
812 	int locked = B_FALSE;
813 
814 	*spapp = NULL;
815 
816 	/*
817 	 * As disgusting as this is, we need to support recursive calls to this
818 	 * function because dsl_dir_open() is called during spa_load(), and ends
819 	 * up calling spa_open() again.  The real fix is to figure out how to
820 	 * avoid dsl_dir_open() calling this in the first place.
821 	 */
822 	if (mutex_owner(&spa_namespace_lock) != curthread) {
823 		mutex_enter(&spa_namespace_lock);
824 		locked = B_TRUE;
825 	}
826 
827 	if ((spa = spa_lookup(pool)) == NULL) {
828 		if (locked)
829 			mutex_exit(&spa_namespace_lock);
830 		return (ENOENT);
831 	}
832 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
833 
834 		spa_activate(spa);
835 
836 		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
837 
838 		if (error == EBADF) {
839 			/*
840 			 * If vdev_validate() returns failure (indicated by
841 			 * EBADF), it indicates that one of the vdevs indicates
842 			 * that the pool has been exported or destroyed.  If
843 			 * this is the case, the config cache is out of sync and
844 			 * we should remove the pool from the namespace.
845 			 */
846 			zfs_post_ok(spa, NULL);
847 			spa_unload(spa);
848 			spa_deactivate(spa);
849 			spa_remove(spa);
850 			spa_config_sync();
851 			if (locked)
852 				mutex_exit(&spa_namespace_lock);
853 			return (ENOENT);
854 		}
855 
856 		if (error) {
857 			/*
858 			 * We can't open the pool, but we still have useful
859 			 * information: the state of each vdev after the
860 			 * attempted vdev_open().  Return this to the user.
861 			 */
862 			if (config != NULL && spa->spa_root_vdev != NULL) {
863 				spa_config_enter(spa, RW_READER, FTAG);
864 				*config = spa_config_generate(spa, NULL, -1ULL,
865 				    B_TRUE);
866 				spa_config_exit(spa, FTAG);
867 			}
868 			spa_unload(spa);
869 			spa_deactivate(spa);
870 			spa->spa_last_open_failed = B_TRUE;
871 			if (locked)
872 				mutex_exit(&spa_namespace_lock);
873 			*spapp = NULL;
874 			return (error);
875 		} else {
876 			zfs_post_ok(spa, NULL);
877 			spa->spa_last_open_failed = B_FALSE;
878 		}
879 
880 		loaded = B_TRUE;
881 	}
882 
883 	spa_open_ref(spa, tag);
884 	if (locked)
885 		mutex_exit(&spa_namespace_lock);
886 
887 	*spapp = spa;
888 
889 	if (config != NULL) {
890 		spa_config_enter(spa, RW_READER, FTAG);
891 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
892 		spa_config_exit(spa, FTAG);
893 	}
894 
895 	/*
896 	 * If we just loaded the pool, resilver anything that's out of date.
897 	 */
898 	if (loaded && (spa_mode & FWRITE))
899 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
900 
901 	return (0);
902 }
903 
904 int
905 spa_open(const char *name, spa_t **spapp, void *tag)
906 {
907 	return (spa_open_common(name, spapp, tag, NULL));
908 }
909 
910 /*
911  * Lookup the given spa_t, incrementing the inject count in the process,
912  * preventing it from being exported or destroyed.
913  */
914 spa_t *
915 spa_inject_addref(char *name)
916 {
917 	spa_t *spa;
918 
919 	mutex_enter(&spa_namespace_lock);
920 	if ((spa = spa_lookup(name)) == NULL) {
921 		mutex_exit(&spa_namespace_lock);
922 		return (NULL);
923 	}
924 	spa->spa_inject_ref++;
925 	mutex_exit(&spa_namespace_lock);
926 
927 	return (spa);
928 }
929 
930 void
931 spa_inject_delref(spa_t *spa)
932 {
933 	mutex_enter(&spa_namespace_lock);
934 	spa->spa_inject_ref--;
935 	mutex_exit(&spa_namespace_lock);
936 }
937 
938 static void
939 spa_add_spares(spa_t *spa, nvlist_t *config)
940 {
941 	nvlist_t **spares;
942 	uint_t i, nspares;
943 	nvlist_t *nvroot;
944 	uint64_t guid;
945 	vdev_stat_t *vs;
946 	uint_t vsc;
947 	uint64_t pool;
948 
949 	if (spa->spa_nspares == 0)
950 		return;
951 
952 	VERIFY(nvlist_lookup_nvlist(config,
953 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
954 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
955 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
956 	if (nspares != 0) {
957 		VERIFY(nvlist_add_nvlist_array(nvroot,
958 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
959 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
960 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
961 
962 		/*
963 		 * Go through and find any spares which have since been
964 		 * repurposed as an active spare.  If this is the case, update
965 		 * their status appropriately.
966 		 */
967 		for (i = 0; i < nspares; i++) {
968 			VERIFY(nvlist_lookup_uint64(spares[i],
969 			    ZPOOL_CONFIG_GUID, &guid) == 0);
970 			if (spa_spare_exists(guid, &pool) && pool != 0ULL) {
971 				VERIFY(nvlist_lookup_uint64_array(
972 				    spares[i], ZPOOL_CONFIG_STATS,
973 				    (uint64_t **)&vs, &vsc) == 0);
974 				vs->vs_state = VDEV_STATE_CANT_OPEN;
975 				vs->vs_aux = VDEV_AUX_SPARED;
976 			}
977 		}
978 	}
979 }
980 
981 int
982 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
983 {
984 	int error;
985 	spa_t *spa;
986 
987 	*config = NULL;
988 	error = spa_open_common(name, &spa, FTAG, config);
989 
990 	if (spa && *config != NULL) {
991 		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
992 		    spa_get_errlog_size(spa)) == 0);
993 
994 		spa_add_spares(spa, *config);
995 	}
996 
997 	/*
998 	 * We want to get the alternate root even for faulted pools, so we cheat
999 	 * and call spa_lookup() directly.
1000 	 */
1001 	if (altroot) {
1002 		if (spa == NULL) {
1003 			mutex_enter(&spa_namespace_lock);
1004 			spa = spa_lookup(name);
1005 			if (spa)
1006 				spa_altroot(spa, altroot, buflen);
1007 			else
1008 				altroot[0] = '\0';
1009 			spa = NULL;
1010 			mutex_exit(&spa_namespace_lock);
1011 		} else {
1012 			spa_altroot(spa, altroot, buflen);
1013 		}
1014 	}
1015 
1016 	if (spa != NULL)
1017 		spa_close(spa, FTAG);
1018 
1019 	return (error);
1020 }
1021 
1022 /*
1023  * Validate that the 'spares' array is well formed.  We must have an array of
1024  * nvlists, each which describes a valid leaf vdev.  If this is an import (mode
1025  * is VDEV_ALLOC_SPARE), then we allow corrupted spares to be specified, as long
1026  * as they are well-formed.
1027  */
1028 static int
1029 spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1030 {
1031 	nvlist_t **spares;
1032 	uint_t i, nspares;
1033 	vdev_t *vd;
1034 	int error;
1035 
1036 	/*
1037 	 * It's acceptable to have no spares specified.
1038 	 */
1039 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1040 	    &spares, &nspares) != 0)
1041 		return (0);
1042 
1043 	if (nspares == 0)
1044 		return (EINVAL);
1045 
1046 	/*
1047 	 * Make sure the pool is formatted with a version that supports hot
1048 	 * spares.
1049 	 */
1050 	if (spa_version(spa) < ZFS_VERSION_SPARES)
1051 		return (ENOTSUP);
1052 
1053 	/*
1054 	 * Set the pending spare list so we correctly handle device in-use
1055 	 * checking.
1056 	 */
1057 	spa->spa_pending_spares = spares;
1058 	spa->spa_pending_nspares = nspares;
1059 
1060 	for (i = 0; i < nspares; i++) {
1061 		if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0,
1062 		    mode)) != 0)
1063 			goto out;
1064 
1065 		if (!vd->vdev_ops->vdev_op_leaf) {
1066 			vdev_free(vd);
1067 			error = EINVAL;
1068 			goto out;
1069 		}
1070 
1071 		vd->vdev_top = vd;
1072 
1073 		if ((error = vdev_open(vd)) == 0 &&
1074 		    (error = vdev_label_init(vd, crtxg,
1075 		    VDEV_LABEL_SPARE)) == 0) {
1076 			VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID,
1077 			    vd->vdev_guid) == 0);
1078 		}
1079 
1080 		vdev_free(vd);
1081 
1082 		if (error && mode != VDEV_ALLOC_SPARE)
1083 			goto out;
1084 		else
1085 			error = 0;
1086 	}
1087 
1088 out:
1089 	spa->spa_pending_spares = NULL;
1090 	spa->spa_pending_nspares = 0;
1091 	return (error);
1092 }
1093 
1094 /*
1095  * Pool Creation
1096  */
1097 int
1098 spa_create(const char *pool, nvlist_t *nvroot, const char *altroot)
1099 {
1100 	spa_t *spa;
1101 	vdev_t *rvd;
1102 	dsl_pool_t *dp;
1103 	dmu_tx_t *tx;
1104 	int c, error = 0;
1105 	uint64_t txg = TXG_INITIAL;
1106 	nvlist_t **spares;
1107 	uint_t nspares;
1108 
1109 	/*
1110 	 * If this pool already exists, return failure.
1111 	 */
1112 	mutex_enter(&spa_namespace_lock);
1113 	if (spa_lookup(pool) != NULL) {
1114 		mutex_exit(&spa_namespace_lock);
1115 		return (EEXIST);
1116 	}
1117 
1118 	/*
1119 	 * Allocate a new spa_t structure.
1120 	 */
1121 	spa = spa_add(pool, altroot);
1122 	spa_activate(spa);
1123 
1124 	spa->spa_uberblock.ub_txg = txg - 1;
1125 	spa->spa_uberblock.ub_version = ZFS_VERSION;
1126 	spa->spa_ubsync = spa->spa_uberblock;
1127 
1128 	/*
1129 	 * Create the root vdev.
1130 	 */
1131 	spa_config_enter(spa, RW_WRITER, FTAG);
1132 
1133 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1134 
1135 	ASSERT(error != 0 || rvd != NULL);
1136 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1137 
1138 	if (error == 0 && rvd->vdev_children == 0)
1139 		error = EINVAL;
1140 
1141 	if (error == 0 &&
1142 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1143 	    (error = spa_validate_spares(spa, nvroot, txg,
1144 	    VDEV_ALLOC_ADD)) == 0) {
1145 		for (c = 0; c < rvd->vdev_children; c++)
1146 			vdev_init(rvd->vdev_child[c], txg);
1147 		vdev_config_dirty(rvd);
1148 	}
1149 
1150 	spa_config_exit(spa, FTAG);
1151 
1152 	if (error != 0) {
1153 		spa_unload(spa);
1154 		spa_deactivate(spa);
1155 		spa_remove(spa);
1156 		mutex_exit(&spa_namespace_lock);
1157 		return (error);
1158 	}
1159 
1160 	/*
1161 	 * Get the list of spares, if specified.
1162 	 */
1163 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1164 	    &spares, &nspares) == 0) {
1165 		VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME,
1166 		    KM_SLEEP) == 0);
1167 		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1168 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1169 		spa_config_enter(spa, RW_WRITER, FTAG);
1170 		spa_load_spares(spa);
1171 		spa_config_exit(spa, FTAG);
1172 		spa->spa_sync_spares = B_TRUE;
1173 	}
1174 
1175 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
1176 	spa->spa_meta_objset = dp->dp_meta_objset;
1177 
1178 	tx = dmu_tx_create_assigned(dp, txg);
1179 
1180 	/*
1181 	 * Create the pool config object.
1182 	 */
1183 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
1184 	    DMU_OT_PACKED_NVLIST, 1 << 14,
1185 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
1186 
1187 	if (zap_add(spa->spa_meta_objset,
1188 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1189 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
1190 		cmn_err(CE_PANIC, "failed to add pool config");
1191 	}
1192 
1193 	/* Newly created pools are always deflated. */
1194 	spa->spa_deflate = TRUE;
1195 	if (zap_add(spa->spa_meta_objset,
1196 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1197 	    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
1198 		cmn_err(CE_PANIC, "failed to add deflate");
1199 	}
1200 
1201 	/*
1202 	 * Create the deferred-free bplist object.  Turn off compression
1203 	 * because sync-to-convergence takes longer if the blocksize
1204 	 * keeps changing.
1205 	 */
1206 	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
1207 	    1 << 14, tx);
1208 	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
1209 	    ZIO_COMPRESS_OFF, tx);
1210 
1211 	if (zap_add(spa->spa_meta_objset,
1212 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1213 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
1214 		cmn_err(CE_PANIC, "failed to add bplist");
1215 	}
1216 
1217 	/*
1218 	 * Create the pool's history object.
1219 	 */
1220 	spa_history_create_obj(spa, tx);
1221 
1222 	dmu_tx_commit(tx);
1223 
1224 	spa->spa_bootfs = zfs_prop_default_numeric(ZFS_PROP_BOOTFS);
1225 	spa->spa_sync_on = B_TRUE;
1226 	txg_sync_start(spa->spa_dsl_pool);
1227 
1228 	/*
1229 	 * We explicitly wait for the first transaction to complete so that our
1230 	 * bean counters are appropriately updated.
1231 	 */
1232 	txg_wait_synced(spa->spa_dsl_pool, txg);
1233 
1234 	spa_config_sync();
1235 
1236 	mutex_exit(&spa_namespace_lock);
1237 
1238 	return (0);
1239 }
1240 
1241 /*
1242  * Import the given pool into the system.  We set up the necessary spa_t and
1243  * then call spa_load() to do the dirty work.
1244  */
1245 int
1246 spa_import(const char *pool, nvlist_t *config, const char *altroot)
1247 {
1248 	spa_t *spa;
1249 	int error;
1250 	nvlist_t *nvroot;
1251 	nvlist_t **spares;
1252 	uint_t nspares;
1253 
1254 	if (!(spa_mode & FWRITE))
1255 		return (EROFS);
1256 
1257 	/*
1258 	 * If a pool with this name exists, return failure.
1259 	 */
1260 	mutex_enter(&spa_namespace_lock);
1261 	if (spa_lookup(pool) != NULL) {
1262 		mutex_exit(&spa_namespace_lock);
1263 		return (EEXIST);
1264 	}
1265 
1266 	/*
1267 	 * Create and initialize the spa structure.
1268 	 */
1269 	spa = spa_add(pool, altroot);
1270 	spa_activate(spa);
1271 
1272 	/*
1273 	 * Pass off the heavy lifting to spa_load().
1274 	 * Pass TRUE for mosconfig because the user-supplied config
1275 	 * is actually the one to trust when doing an import.
1276 	 */
1277 	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
1278 
1279 	spa_config_enter(spa, RW_WRITER, FTAG);
1280 	/*
1281 	 * Toss any existing sparelist, as it doesn't have any validity anymore,
1282 	 * and conflicts with spa_has_spare().
1283 	 */
1284 	if (spa->spa_sparelist) {
1285 		nvlist_free(spa->spa_sparelist);
1286 		spa->spa_sparelist = NULL;
1287 		spa_load_spares(spa);
1288 	}
1289 
1290 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1291 	    &nvroot) == 0);
1292 	if (error == 0)
1293 		error = spa_validate_spares(spa, nvroot, -1ULL,
1294 		    VDEV_ALLOC_SPARE);
1295 	spa_config_exit(spa, FTAG);
1296 
1297 	if (error != 0) {
1298 		spa_unload(spa);
1299 		spa_deactivate(spa);
1300 		spa_remove(spa);
1301 		mutex_exit(&spa_namespace_lock);
1302 		return (error);
1303 	}
1304 
1305 	/*
1306 	 * Override any spares as specified by the user, as these may have
1307 	 * correct device names/devids, etc.
1308 	 */
1309 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1310 	    &spares, &nspares) == 0) {
1311 		if (spa->spa_sparelist)
1312 			VERIFY(nvlist_remove(spa->spa_sparelist,
1313 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1314 		else
1315 			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1316 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1317 		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1318 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1319 		spa_config_enter(spa, RW_WRITER, FTAG);
1320 		spa_load_spares(spa);
1321 		spa_config_exit(spa, FTAG);
1322 		spa->spa_sync_spares = B_TRUE;
1323 	}
1324 
1325 	/*
1326 	 * Update the config cache to include the newly-imported pool.
1327 	 */
1328 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1329 
1330 	mutex_exit(&spa_namespace_lock);
1331 
1332 	/*
1333 	 * Resilver anything that's out of date.
1334 	 */
1335 	if (spa_mode & FWRITE)
1336 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1337 
1338 	return (0);
1339 }
1340 
1341 /*
1342  * This (illegal) pool name is used when temporarily importing a spa_t in order
1343  * to get the vdev stats associated with the imported devices.
1344  */
1345 #define	TRYIMPORT_NAME	"$import"
1346 
1347 nvlist_t *
1348 spa_tryimport(nvlist_t *tryconfig)
1349 {
1350 	nvlist_t *config = NULL;
1351 	char *poolname;
1352 	spa_t *spa;
1353 	uint64_t state;
1354 
1355 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
1356 		return (NULL);
1357 
1358 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
1359 		return (NULL);
1360 
1361 	/*
1362 	 * Create and initialize the spa structure.
1363 	 */
1364 	mutex_enter(&spa_namespace_lock);
1365 	spa = spa_add(TRYIMPORT_NAME, NULL);
1366 	spa_activate(spa);
1367 
1368 	/*
1369 	 * Pass off the heavy lifting to spa_load().
1370 	 * Pass TRUE for mosconfig because the user-supplied config
1371 	 * is actually the one to trust when doing an import.
1372 	 */
1373 	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
1374 
1375 	/*
1376 	 * If 'tryconfig' was at least parsable, return the current config.
1377 	 */
1378 	if (spa->spa_root_vdev != NULL) {
1379 		spa_config_enter(spa, RW_READER, FTAG);
1380 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1381 		spa_config_exit(spa, FTAG);
1382 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
1383 		    poolname) == 0);
1384 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1385 		    state) == 0);
1386 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
1387 		    spa->spa_uberblock.ub_timestamp) == 0);
1388 
1389 		/*
1390 		 * Add the list of hot spares.
1391 		 */
1392 		spa_add_spares(spa, config);
1393 	}
1394 
1395 	spa_unload(spa);
1396 	spa_deactivate(spa);
1397 	spa_remove(spa);
1398 	mutex_exit(&spa_namespace_lock);
1399 
1400 	return (config);
1401 }
1402 
1403 /*
1404  * Pool export/destroy
1405  *
1406  * The act of destroying or exporting a pool is very simple.  We make sure there
1407  * is no more pending I/O and any references to the pool are gone.  Then, we
1408  * update the pool state and sync all the labels to disk, removing the
1409  * configuration from the cache afterwards.
1410  */
1411 static int
1412 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig)
1413 {
1414 	spa_t *spa;
1415 
1416 	if (oldconfig)
1417 		*oldconfig = NULL;
1418 
1419 	if (!(spa_mode & FWRITE))
1420 		return (EROFS);
1421 
1422 	mutex_enter(&spa_namespace_lock);
1423 	if ((spa = spa_lookup(pool)) == NULL) {
1424 		mutex_exit(&spa_namespace_lock);
1425 		return (ENOENT);
1426 	}
1427 
1428 	/*
1429 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
1430 	 * reacquire the namespace lock, and see if we can export.
1431 	 */
1432 	spa_open_ref(spa, FTAG);
1433 	mutex_exit(&spa_namespace_lock);
1434 	spa_async_suspend(spa);
1435 	mutex_enter(&spa_namespace_lock);
1436 	spa_close(spa, FTAG);
1437 
1438 	/*
1439 	 * The pool will be in core if it's openable,
1440 	 * in which case we can modify its state.
1441 	 */
1442 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
1443 		/*
1444 		 * Objsets may be open only because they're dirty, so we
1445 		 * have to force it to sync before checking spa_refcnt.
1446 		 */
1447 		spa_scrub_suspend(spa);
1448 		txg_wait_synced(spa->spa_dsl_pool, 0);
1449 
1450 		/*
1451 		 * A pool cannot be exported or destroyed if there are active
1452 		 * references.  If we are resetting a pool, allow references by
1453 		 * fault injection handlers.
1454 		 */
1455 		if (!spa_refcount_zero(spa) ||
1456 		    (spa->spa_inject_ref != 0 &&
1457 		    new_state != POOL_STATE_UNINITIALIZED)) {
1458 			spa_scrub_resume(spa);
1459 			spa_async_resume(spa);
1460 			mutex_exit(&spa_namespace_lock);
1461 			return (EBUSY);
1462 		}
1463 
1464 		spa_scrub_resume(spa);
1465 		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
1466 
1467 		/*
1468 		 * We want this to be reflected on every label,
1469 		 * so mark them all dirty.  spa_unload() will do the
1470 		 * final sync that pushes these changes out.
1471 		 */
1472 		if (new_state != POOL_STATE_UNINITIALIZED) {
1473 			spa_config_enter(spa, RW_WRITER, FTAG);
1474 			spa->spa_state = new_state;
1475 			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
1476 			vdev_config_dirty(spa->spa_root_vdev);
1477 			spa_config_exit(spa, FTAG);
1478 		}
1479 	}
1480 
1481 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1482 		spa_unload(spa);
1483 		spa_deactivate(spa);
1484 	}
1485 
1486 	if (oldconfig && spa->spa_config)
1487 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
1488 
1489 	if (new_state != POOL_STATE_UNINITIALIZED) {
1490 		spa_remove(spa);
1491 		spa_config_sync();
1492 	}
1493 	mutex_exit(&spa_namespace_lock);
1494 
1495 	return (0);
1496 }
1497 
1498 /*
1499  * Destroy a storage pool.
1500  */
1501 int
1502 spa_destroy(char *pool)
1503 {
1504 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL));
1505 }
1506 
1507 /*
1508  * Export a storage pool.
1509  */
1510 int
1511 spa_export(char *pool, nvlist_t **oldconfig)
1512 {
1513 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig));
1514 }
1515 
1516 /*
1517  * Similar to spa_export(), this unloads the spa_t without actually removing it
1518  * from the namespace in any way.
1519  */
1520 int
1521 spa_reset(char *pool)
1522 {
1523 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL));
1524 }
1525 
1526 
1527 /*
1528  * ==========================================================================
1529  * Device manipulation
1530  * ==========================================================================
1531  */
1532 
1533 /*
1534  * Add capacity to a storage pool.
1535  */
1536 int
1537 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1538 {
1539 	uint64_t txg;
1540 	int c, error;
1541 	vdev_t *rvd = spa->spa_root_vdev;
1542 	vdev_t *vd, *tvd;
1543 	nvlist_t **spares;
1544 	uint_t i, nspares;
1545 
1546 	txg = spa_vdev_enter(spa);
1547 
1548 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
1549 	    VDEV_ALLOC_ADD)) != 0)
1550 		return (spa_vdev_exit(spa, NULL, txg, error));
1551 
1552 	spa->spa_pending_vdev = vd;
1553 
1554 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1555 	    &spares, &nspares) != 0)
1556 		nspares = 0;
1557 
1558 	if (vd->vdev_children == 0 && nspares == 0) {
1559 		spa->spa_pending_vdev = NULL;
1560 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1561 	}
1562 
1563 	if (vd->vdev_children != 0) {
1564 		if ((error = vdev_create(vd, txg, B_FALSE)) != 0) {
1565 			spa->spa_pending_vdev = NULL;
1566 			return (spa_vdev_exit(spa, vd, txg, error));
1567 		}
1568 	}
1569 
1570 	/*
1571 	 * We must validate the spares after checking the children.  Otherwise,
1572 	 * vdev_inuse() will blindly overwrite the spare.
1573 	 */
1574 	if ((error = spa_validate_spares(spa, nvroot, txg,
1575 	    VDEV_ALLOC_ADD)) != 0) {
1576 		spa->spa_pending_vdev = NULL;
1577 		return (spa_vdev_exit(spa, vd, txg, error));
1578 	}
1579 
1580 	spa->spa_pending_vdev = NULL;
1581 
1582 	/*
1583 	 * Transfer each new top-level vdev from vd to rvd.
1584 	 */
1585 	for (c = 0; c < vd->vdev_children; c++) {
1586 		tvd = vd->vdev_child[c];
1587 		vdev_remove_child(vd, tvd);
1588 		tvd->vdev_id = rvd->vdev_children;
1589 		vdev_add_child(rvd, tvd);
1590 		vdev_config_dirty(tvd);
1591 	}
1592 
1593 	if (nspares != 0) {
1594 		if (spa->spa_sparelist != NULL) {
1595 			nvlist_t **oldspares;
1596 			uint_t oldnspares;
1597 			nvlist_t **newspares;
1598 
1599 			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
1600 			    ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0);
1601 
1602 			newspares = kmem_alloc(sizeof (void *) *
1603 			    (nspares + oldnspares), KM_SLEEP);
1604 			for (i = 0; i < oldnspares; i++)
1605 				VERIFY(nvlist_dup(oldspares[i],
1606 				    &newspares[i], KM_SLEEP) == 0);
1607 			for (i = 0; i < nspares; i++)
1608 				VERIFY(nvlist_dup(spares[i],
1609 				    &newspares[i + oldnspares],
1610 				    KM_SLEEP) == 0);
1611 
1612 			VERIFY(nvlist_remove(spa->spa_sparelist,
1613 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1614 
1615 			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1616 			    ZPOOL_CONFIG_SPARES, newspares,
1617 			    nspares + oldnspares) == 0);
1618 			for (i = 0; i < oldnspares + nspares; i++)
1619 				nvlist_free(newspares[i]);
1620 			kmem_free(newspares, (oldnspares + nspares) *
1621 			    sizeof (void *));
1622 		} else {
1623 			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1624 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1625 			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1626 			    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1627 		}
1628 
1629 		spa_load_spares(spa);
1630 		spa->spa_sync_spares = B_TRUE;
1631 	}
1632 
1633 	/*
1634 	 * We have to be careful when adding new vdevs to an existing pool.
1635 	 * If other threads start allocating from these vdevs before we
1636 	 * sync the config cache, and we lose power, then upon reboot we may
1637 	 * fail to open the pool because there are DVAs that the config cache
1638 	 * can't translate.  Therefore, we first add the vdevs without
1639 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
1640 	 * and then let spa_config_update() initialize the new metaslabs.
1641 	 *
1642 	 * spa_load() checks for added-but-not-initialized vdevs, so that
1643 	 * if we lose power at any point in this sequence, the remaining
1644 	 * steps will be completed the next time we load the pool.
1645 	 */
1646 	(void) spa_vdev_exit(spa, vd, txg, 0);
1647 
1648 	mutex_enter(&spa_namespace_lock);
1649 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1650 	mutex_exit(&spa_namespace_lock);
1651 
1652 	return (0);
1653 }
1654 
1655 /*
1656  * Attach a device to a mirror.  The arguments are the path to any device
1657  * in the mirror, and the nvroot for the new device.  If the path specifies
1658  * a device that is not mirrored, we automatically insert the mirror vdev.
1659  *
1660  * If 'replacing' is specified, the new device is intended to replace the
1661  * existing device; in this case the two devices are made into their own
1662  * mirror using the 'replacing' vdev, which is functionally idendical to
1663  * the mirror vdev (it actually reuses all the same ops) but has a few
1664  * extra rules: you can't attach to it after it's been created, and upon
1665  * completion of resilvering, the first disk (the one being replaced)
1666  * is automatically detached.
1667  */
1668 int
1669 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
1670 {
1671 	uint64_t txg, open_txg;
1672 	int error;
1673 	vdev_t *rvd = spa->spa_root_vdev;
1674 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
1675 	vdev_ops_t *pvops;
1676 
1677 	txg = spa_vdev_enter(spa);
1678 
1679 	oldvd = vdev_lookup_by_guid(rvd, guid);
1680 
1681 	if (oldvd == NULL)
1682 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1683 
1684 	if (!oldvd->vdev_ops->vdev_op_leaf)
1685 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1686 
1687 	pvd = oldvd->vdev_parent;
1688 
1689 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
1690 	    VDEV_ALLOC_ADD)) != 0 || newrootvd->vdev_children != 1)
1691 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1692 
1693 	newvd = newrootvd->vdev_child[0];
1694 
1695 	if (!newvd->vdev_ops->vdev_op_leaf)
1696 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1697 
1698 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
1699 		return (spa_vdev_exit(spa, newrootvd, txg, error));
1700 
1701 	if (!replacing) {
1702 		/*
1703 		 * For attach, the only allowable parent is a mirror or the root
1704 		 * vdev.
1705 		 */
1706 		if (pvd->vdev_ops != &vdev_mirror_ops &&
1707 		    pvd->vdev_ops != &vdev_root_ops)
1708 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1709 
1710 		pvops = &vdev_mirror_ops;
1711 	} else {
1712 		/*
1713 		 * Active hot spares can only be replaced by inactive hot
1714 		 * spares.
1715 		 */
1716 		if (pvd->vdev_ops == &vdev_spare_ops &&
1717 		    pvd->vdev_child[1] == oldvd &&
1718 		    !spa_has_spare(spa, newvd->vdev_guid))
1719 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1720 
1721 		/*
1722 		 * If the source is a hot spare, and the parent isn't already a
1723 		 * spare, then we want to create a new hot spare.  Otherwise, we
1724 		 * want to create a replacing vdev.  The user is not allowed to
1725 		 * attach to a spared vdev child unless the 'isspare' state is
1726 		 * the same (spare replaces spare, non-spare replaces
1727 		 * non-spare).
1728 		 */
1729 		if (pvd->vdev_ops == &vdev_replacing_ops)
1730 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1731 		else if (pvd->vdev_ops == &vdev_spare_ops &&
1732 		    newvd->vdev_isspare != oldvd->vdev_isspare)
1733 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1734 		else if (pvd->vdev_ops != &vdev_spare_ops &&
1735 		    newvd->vdev_isspare)
1736 			pvops = &vdev_spare_ops;
1737 		else
1738 			pvops = &vdev_replacing_ops;
1739 	}
1740 
1741 	/*
1742 	 * Compare the new device size with the replaceable/attachable
1743 	 * device size.
1744 	 */
1745 	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
1746 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
1747 
1748 	/*
1749 	 * The new device cannot have a higher alignment requirement
1750 	 * than the top-level vdev.
1751 	 */
1752 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
1753 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
1754 
1755 	/*
1756 	 * If this is an in-place replacement, update oldvd's path and devid
1757 	 * to make it distinguishable from newvd, and unopenable from now on.
1758 	 */
1759 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
1760 		spa_strfree(oldvd->vdev_path);
1761 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
1762 		    KM_SLEEP);
1763 		(void) sprintf(oldvd->vdev_path, "%s/%s",
1764 		    newvd->vdev_path, "old");
1765 		if (oldvd->vdev_devid != NULL) {
1766 			spa_strfree(oldvd->vdev_devid);
1767 			oldvd->vdev_devid = NULL;
1768 		}
1769 	}
1770 
1771 	/*
1772 	 * If the parent is not a mirror, or if we're replacing, insert the new
1773 	 * mirror/replacing/spare vdev above oldvd.
1774 	 */
1775 	if (pvd->vdev_ops != pvops)
1776 		pvd = vdev_add_parent(oldvd, pvops);
1777 
1778 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
1779 	ASSERT(pvd->vdev_ops == pvops);
1780 	ASSERT(oldvd->vdev_parent == pvd);
1781 
1782 	/*
1783 	 * Extract the new device from its root and add it to pvd.
1784 	 */
1785 	vdev_remove_child(newrootvd, newvd);
1786 	newvd->vdev_id = pvd->vdev_children;
1787 	vdev_add_child(pvd, newvd);
1788 
1789 	/*
1790 	 * If newvd is smaller than oldvd, but larger than its rsize,
1791 	 * the addition of newvd may have decreased our parent's asize.
1792 	 */
1793 	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
1794 
1795 	tvd = newvd->vdev_top;
1796 	ASSERT(pvd->vdev_top == tvd);
1797 	ASSERT(tvd->vdev_parent == rvd);
1798 
1799 	vdev_config_dirty(tvd);
1800 
1801 	/*
1802 	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
1803 	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
1804 	 */
1805 	open_txg = txg + TXG_CONCURRENT_STATES - 1;
1806 
1807 	mutex_enter(&newvd->vdev_dtl_lock);
1808 	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
1809 	    open_txg - TXG_INITIAL + 1);
1810 	mutex_exit(&newvd->vdev_dtl_lock);
1811 
1812 	if (newvd->vdev_isspare)
1813 		spa_spare_activate(newvd);
1814 
1815 	/*
1816 	 * Mark newvd's DTL dirty in this txg.
1817 	 */
1818 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
1819 
1820 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
1821 
1822 	/*
1823 	 * Kick off a resilver to update newvd.
1824 	 */
1825 	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1826 
1827 	return (0);
1828 }
1829 
1830 /*
1831  * Detach a device from a mirror or replacing vdev.
1832  * If 'replace_done' is specified, only detach if the parent
1833  * is a replacing vdev.
1834  */
1835 int
1836 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
1837 {
1838 	uint64_t txg;
1839 	int c, t, error;
1840 	vdev_t *rvd = spa->spa_root_vdev;
1841 	vdev_t *vd, *pvd, *cvd, *tvd;
1842 	boolean_t unspare = B_FALSE;
1843 	uint64_t unspare_guid;
1844 
1845 	txg = spa_vdev_enter(spa);
1846 
1847 	vd = vdev_lookup_by_guid(rvd, guid);
1848 
1849 	if (vd == NULL)
1850 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1851 
1852 	if (!vd->vdev_ops->vdev_op_leaf)
1853 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1854 
1855 	pvd = vd->vdev_parent;
1856 
1857 	/*
1858 	 * If replace_done is specified, only remove this device if it's
1859 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
1860 	 * disk can be removed.
1861 	 */
1862 	if (replace_done) {
1863 		if (pvd->vdev_ops == &vdev_replacing_ops) {
1864 			if (vd->vdev_id != 0)
1865 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1866 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
1867 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1868 		}
1869 	}
1870 
1871 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
1872 	    spa_version(spa) >= ZFS_VERSION_SPARES);
1873 
1874 	/*
1875 	 * Only mirror, replacing, and spare vdevs support detach.
1876 	 */
1877 	if (pvd->vdev_ops != &vdev_replacing_ops &&
1878 	    pvd->vdev_ops != &vdev_mirror_ops &&
1879 	    pvd->vdev_ops != &vdev_spare_ops)
1880 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1881 
1882 	/*
1883 	 * If there's only one replica, you can't detach it.
1884 	 */
1885 	if (pvd->vdev_children <= 1)
1886 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1887 
1888 	/*
1889 	 * If all siblings have non-empty DTLs, this device may have the only
1890 	 * valid copy of the data, which means we cannot safely detach it.
1891 	 *
1892 	 * XXX -- as in the vdev_offline() case, we really want a more
1893 	 * precise DTL check.
1894 	 */
1895 	for (c = 0; c < pvd->vdev_children; c++) {
1896 		uint64_t dirty;
1897 
1898 		cvd = pvd->vdev_child[c];
1899 		if (cvd == vd)
1900 			continue;
1901 		if (vdev_is_dead(cvd))
1902 			continue;
1903 		mutex_enter(&cvd->vdev_dtl_lock);
1904 		dirty = cvd->vdev_dtl_map.sm_space |
1905 		    cvd->vdev_dtl_scrub.sm_space;
1906 		mutex_exit(&cvd->vdev_dtl_lock);
1907 		if (!dirty)
1908 			break;
1909 	}
1910 
1911 	/*
1912 	 * If we are a replacing or spare vdev, then we can always detach the
1913 	 * latter child, as that is how one cancels the operation.
1914 	 */
1915 	if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) &&
1916 	    c == pvd->vdev_children)
1917 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1918 
1919 	/*
1920 	 * If we are detaching the original disk from a spare, then it implies
1921 	 * that the spare should become a real disk, and be removed from the
1922 	 * active spare list for the pool.
1923 	 */
1924 	if (pvd->vdev_ops == &vdev_spare_ops &&
1925 	    vd->vdev_id == 0)
1926 		unspare = B_TRUE;
1927 
1928 	/*
1929 	 * Erase the disk labels so the disk can be used for other things.
1930 	 * This must be done after all other error cases are handled,
1931 	 * but before we disembowel vd (so we can still do I/O to it).
1932 	 * But if we can't do it, don't treat the error as fatal --
1933 	 * it may be that the unwritability of the disk is the reason
1934 	 * it's being detached!
1935 	 */
1936 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
1937 
1938 	/*
1939 	 * Remove vd from its parent and compact the parent's children.
1940 	 */
1941 	vdev_remove_child(pvd, vd);
1942 	vdev_compact_children(pvd);
1943 
1944 	/*
1945 	 * Remember one of the remaining children so we can get tvd below.
1946 	 */
1947 	cvd = pvd->vdev_child[0];
1948 
1949 	/*
1950 	 * If we need to remove the remaining child from the list of hot spares,
1951 	 * do it now, marking the vdev as no longer a spare in the process.  We
1952 	 * must do this before vdev_remove_parent(), because that can change the
1953 	 * GUID if it creates a new toplevel GUID.
1954 	 */
1955 	if (unspare) {
1956 		ASSERT(cvd->vdev_isspare);
1957 		spa_spare_remove(cvd);
1958 		unspare_guid = cvd->vdev_guid;
1959 	}
1960 
1961 	/*
1962 	 * If the parent mirror/replacing vdev only has one child,
1963 	 * the parent is no longer needed.  Remove it from the tree.
1964 	 */
1965 	if (pvd->vdev_children == 1)
1966 		vdev_remove_parent(cvd);
1967 
1968 	/*
1969 	 * We don't set tvd until now because the parent we just removed
1970 	 * may have been the previous top-level vdev.
1971 	 */
1972 	tvd = cvd->vdev_top;
1973 	ASSERT(tvd->vdev_parent == rvd);
1974 
1975 	/*
1976 	 * Reevaluate the parent vdev state.
1977 	 */
1978 	vdev_propagate_state(cvd->vdev_parent);
1979 
1980 	/*
1981 	 * If the device we just detached was smaller than the others, it may be
1982 	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
1983 	 * can't fail because the existing metaslabs are already in core, so
1984 	 * there's nothing to read from disk.
1985 	 */
1986 	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
1987 
1988 	vdev_config_dirty(tvd);
1989 
1990 	/*
1991 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
1992 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
1993 	 * But first make sure we're not on any *other* txg's DTL list, to
1994 	 * prevent vd from being accessed after it's freed.
1995 	 */
1996 	for (t = 0; t < TXG_SIZE; t++)
1997 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
1998 	vd->vdev_detached = B_TRUE;
1999 	vdev_dirty(tvd, VDD_DTL, vd, txg);
2000 
2001 	error = spa_vdev_exit(spa, vd, txg, 0);
2002 
2003 	/*
2004 	 * If this was the removal of the original device in a hot spare vdev,
2005 	 * then we want to go through and remove the device from the hot spare
2006 	 * list of every other pool.
2007 	 */
2008 	if (unspare) {
2009 		spa = NULL;
2010 		mutex_enter(&spa_namespace_lock);
2011 		while ((spa = spa_next(spa)) != NULL) {
2012 			if (spa->spa_state != POOL_STATE_ACTIVE)
2013 				continue;
2014 
2015 			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
2016 		}
2017 		mutex_exit(&spa_namespace_lock);
2018 	}
2019 
2020 	return (error);
2021 }
2022 
2023 /*
2024  * Remove a device from the pool.  Currently, this supports removing only hot
2025  * spares.
2026  */
2027 int
2028 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
2029 {
2030 	vdev_t *vd;
2031 	nvlist_t **spares, *nv, **newspares;
2032 	uint_t i, j, nspares;
2033 	int ret = 0;
2034 
2035 	spa_config_enter(spa, RW_WRITER, FTAG);
2036 
2037 	vd = spa_lookup_by_guid(spa, guid);
2038 
2039 	nv = NULL;
2040 	if (spa->spa_spares != NULL &&
2041 	    nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2042 	    &spares, &nspares) == 0) {
2043 		for (i = 0; i < nspares; i++) {
2044 			uint64_t theguid;
2045 
2046 			VERIFY(nvlist_lookup_uint64(spares[i],
2047 			    ZPOOL_CONFIG_GUID, &theguid) == 0);
2048 			if (theguid == guid) {
2049 				nv = spares[i];
2050 				break;
2051 			}
2052 		}
2053 	}
2054 
2055 	/*
2056 	 * We only support removing a hot spare, and only if it's not currently
2057 	 * in use in this pool.
2058 	 */
2059 	if (nv == NULL && vd == NULL) {
2060 		ret = ENOENT;
2061 		goto out;
2062 	}
2063 
2064 	if (nv == NULL && vd != NULL) {
2065 		ret = ENOTSUP;
2066 		goto out;
2067 	}
2068 
2069 	if (!unspare && nv != NULL && vd != NULL) {
2070 		ret = EBUSY;
2071 		goto out;
2072 	}
2073 
2074 	if (nspares == 1) {
2075 		newspares = NULL;
2076 	} else {
2077 		newspares = kmem_alloc((nspares - 1) * sizeof (void *),
2078 		    KM_SLEEP);
2079 		for (i = 0, j = 0; i < nspares; i++) {
2080 			if (spares[i] != nv)
2081 				VERIFY(nvlist_dup(spares[i],
2082 				    &newspares[j++], KM_SLEEP) == 0);
2083 		}
2084 	}
2085 
2086 	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2087 	    DATA_TYPE_NVLIST_ARRAY) == 0);
2088 	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2089 	    newspares, nspares - 1) == 0);
2090 	for (i = 0; i < nspares - 1; i++)
2091 		nvlist_free(newspares[i]);
2092 	kmem_free(newspares, (nspares - 1) * sizeof (void *));
2093 	spa_load_spares(spa);
2094 	spa->spa_sync_spares = B_TRUE;
2095 
2096 out:
2097 	spa_config_exit(spa, FTAG);
2098 
2099 	return (ret);
2100 }
2101 
2102 /*
2103  * Find any device that's done replacing, so we can detach it.
2104  */
2105 static vdev_t *
2106 spa_vdev_replace_done_hunt(vdev_t *vd)
2107 {
2108 	vdev_t *newvd, *oldvd;
2109 	int c;
2110 
2111 	for (c = 0; c < vd->vdev_children; c++) {
2112 		oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]);
2113 		if (oldvd != NULL)
2114 			return (oldvd);
2115 	}
2116 
2117 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
2118 		oldvd = vd->vdev_child[0];
2119 		newvd = vd->vdev_child[1];
2120 
2121 		mutex_enter(&newvd->vdev_dtl_lock);
2122 		if (newvd->vdev_dtl_map.sm_space == 0 &&
2123 		    newvd->vdev_dtl_scrub.sm_space == 0) {
2124 			mutex_exit(&newvd->vdev_dtl_lock);
2125 			return (oldvd);
2126 		}
2127 		mutex_exit(&newvd->vdev_dtl_lock);
2128 	}
2129 
2130 	return (NULL);
2131 }
2132 
2133 static void
2134 spa_vdev_replace_done(spa_t *spa)
2135 {
2136 	vdev_t *vd;
2137 	vdev_t *pvd;
2138 	uint64_t guid;
2139 	uint64_t pguid = 0;
2140 
2141 	spa_config_enter(spa, RW_READER, FTAG);
2142 
2143 	while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) {
2144 		guid = vd->vdev_guid;
2145 		/*
2146 		 * If we have just finished replacing a hot spared device, then
2147 		 * we need to detach the parent's first child (the original hot
2148 		 * spare) as well.
2149 		 */
2150 		pvd = vd->vdev_parent;
2151 		if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2152 		    pvd->vdev_id == 0) {
2153 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
2154 			ASSERT(pvd->vdev_parent->vdev_children == 2);
2155 			pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
2156 		}
2157 		spa_config_exit(spa, FTAG);
2158 		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
2159 			return;
2160 		if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
2161 			return;
2162 		spa_config_enter(spa, RW_READER, FTAG);
2163 	}
2164 
2165 	spa_config_exit(spa, FTAG);
2166 }
2167 
2168 /*
2169  * Update the stored path for this vdev.  Dirty the vdev configuration, relying
2170  * on spa_vdev_enter/exit() to synchronize the labels and cache.
2171  */
2172 int
2173 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
2174 {
2175 	vdev_t *rvd, *vd;
2176 	uint64_t txg;
2177 
2178 	rvd = spa->spa_root_vdev;
2179 
2180 	txg = spa_vdev_enter(spa);
2181 
2182 	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2183 		/*
2184 		 * Determine if this is a reference to a hot spare.  In that
2185 		 * case, update the path as stored in the spare list.
2186 		 */
2187 		nvlist_t **spares;
2188 		uint_t i, nspares;
2189 		if (spa->spa_sparelist != NULL) {
2190 			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
2191 			    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2192 			for (i = 0; i < nspares; i++) {
2193 				uint64_t theguid;
2194 				VERIFY(nvlist_lookup_uint64(spares[i],
2195 				    ZPOOL_CONFIG_GUID, &theguid) == 0);
2196 				if (theguid == guid)
2197 					break;
2198 			}
2199 
2200 			if (i == nspares)
2201 				return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2202 
2203 			VERIFY(nvlist_add_string(spares[i],
2204 			    ZPOOL_CONFIG_PATH, newpath) == 0);
2205 			spa_load_spares(spa);
2206 			spa->spa_sync_spares = B_TRUE;
2207 			return (spa_vdev_exit(spa, NULL, txg, 0));
2208 		} else {
2209 			return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2210 		}
2211 	}
2212 
2213 	if (!vd->vdev_ops->vdev_op_leaf)
2214 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2215 
2216 	spa_strfree(vd->vdev_path);
2217 	vd->vdev_path = spa_strdup(newpath);
2218 
2219 	vdev_config_dirty(vd->vdev_top);
2220 
2221 	return (spa_vdev_exit(spa, NULL, txg, 0));
2222 }
2223 
2224 /*
2225  * ==========================================================================
2226  * SPA Scrubbing
2227  * ==========================================================================
2228  */
2229 
2230 static void
2231 spa_scrub_io_done(zio_t *zio)
2232 {
2233 	spa_t *spa = zio->io_spa;
2234 
2235 	zio_data_buf_free(zio->io_data, zio->io_size);
2236 
2237 	mutex_enter(&spa->spa_scrub_lock);
2238 	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
2239 		vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev;
2240 		spa->spa_scrub_errors++;
2241 		mutex_enter(&vd->vdev_stat_lock);
2242 		vd->vdev_stat.vs_scrub_errors++;
2243 		mutex_exit(&vd->vdev_stat_lock);
2244 	}
2245 
2246 	if (--spa->spa_scrub_inflight < spa->spa_scrub_maxinflight)
2247 		cv_broadcast(&spa->spa_scrub_io_cv);
2248 
2249 	ASSERT(spa->spa_scrub_inflight >= 0);
2250 
2251 	mutex_exit(&spa->spa_scrub_lock);
2252 }
2253 
2254 static void
2255 spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
2256     zbookmark_t *zb)
2257 {
2258 	size_t size = BP_GET_LSIZE(bp);
2259 	void *data;
2260 
2261 	mutex_enter(&spa->spa_scrub_lock);
2262 	/*
2263 	 * Do not give too much work to vdev(s).
2264 	 */
2265 	while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight) {
2266 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2267 	}
2268 	spa->spa_scrub_inflight++;
2269 	mutex_exit(&spa->spa_scrub_lock);
2270 
2271 	data = zio_data_buf_alloc(size);
2272 
2273 	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
2274 		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
2275 
2276 	flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL;
2277 
2278 	zio_nowait(zio_read(NULL, spa, bp, data, size,
2279 	    spa_scrub_io_done, NULL, priority, flags, zb));
2280 }
2281 
2282 /* ARGSUSED */
2283 static int
2284 spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
2285 {
2286 	blkptr_t *bp = &bc->bc_blkptr;
2287 	vdev_t *vd = spa->spa_root_vdev;
2288 	dva_t *dva = bp->blk_dva;
2289 	int needs_resilver = B_FALSE;
2290 	int d;
2291 
2292 	if (bc->bc_errno) {
2293 		/*
2294 		 * We can't scrub this block, but we can continue to scrub
2295 		 * the rest of the pool.  Note the error and move along.
2296 		 */
2297 		mutex_enter(&spa->spa_scrub_lock);
2298 		spa->spa_scrub_errors++;
2299 		mutex_exit(&spa->spa_scrub_lock);
2300 
2301 		mutex_enter(&vd->vdev_stat_lock);
2302 		vd->vdev_stat.vs_scrub_errors++;
2303 		mutex_exit(&vd->vdev_stat_lock);
2304 
2305 		return (ERESTART);
2306 	}
2307 
2308 	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
2309 
2310 	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
2311 		vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]));
2312 
2313 		ASSERT(vd != NULL);
2314 
2315 		/*
2316 		 * Keep track of how much data we've examined so that
2317 		 * zpool(1M) status can make useful progress reports.
2318 		 */
2319 		mutex_enter(&vd->vdev_stat_lock);
2320 		vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]);
2321 		mutex_exit(&vd->vdev_stat_lock);
2322 
2323 		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
2324 			if (DVA_GET_GANG(&dva[d])) {
2325 				/*
2326 				 * Gang members may be spread across multiple
2327 				 * vdevs, so the best we can do is look at the
2328 				 * pool-wide DTL.
2329 				 * XXX -- it would be better to change our
2330 				 * allocation policy to ensure that this can't
2331 				 * happen.
2332 				 */
2333 				vd = spa->spa_root_vdev;
2334 			}
2335 			if (vdev_dtl_contains(&vd->vdev_dtl_map,
2336 			    bp->blk_birth, 1))
2337 				needs_resilver = B_TRUE;
2338 		}
2339 	}
2340 
2341 	if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING)
2342 		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
2343 		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
2344 	else if (needs_resilver)
2345 		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
2346 		    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
2347 
2348 	return (0);
2349 }
2350 
2351 static void
2352 spa_scrub_thread(spa_t *spa)
2353 {
2354 	callb_cpr_t cprinfo;
2355 	traverse_handle_t *th = spa->spa_scrub_th;
2356 	vdev_t *rvd = spa->spa_root_vdev;
2357 	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
2358 	int error = 0;
2359 	boolean_t complete;
2360 
2361 	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
2362 
2363 	/*
2364 	 * If we're restarting due to a snapshot create/delete,
2365 	 * wait for that to complete.
2366 	 */
2367 	txg_wait_synced(spa_get_dsl(spa), 0);
2368 
2369 	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
2370 	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2371 	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
2372 
2373 	spa_config_enter(spa, RW_WRITER, FTAG);
2374 	vdev_reopen(rvd);		/* purge all vdev caches */
2375 	vdev_config_dirty(rvd);		/* rewrite all disk labels */
2376 	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
2377 	spa_config_exit(spa, FTAG);
2378 
2379 	mutex_enter(&spa->spa_scrub_lock);
2380 	spa->spa_scrub_errors = 0;
2381 	spa->spa_scrub_active = 1;
2382 	ASSERT(spa->spa_scrub_inflight == 0);
2383 
2384 	while (!spa->spa_scrub_stop) {
2385 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2386 		while (spa->spa_scrub_suspended) {
2387 			spa->spa_scrub_active = 0;
2388 			cv_broadcast(&spa->spa_scrub_cv);
2389 			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2390 			spa->spa_scrub_active = 1;
2391 		}
2392 		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
2393 
2394 		if (spa->spa_scrub_restart_txg != 0)
2395 			break;
2396 
2397 		mutex_exit(&spa->spa_scrub_lock);
2398 		error = traverse_more(th);
2399 		mutex_enter(&spa->spa_scrub_lock);
2400 		if (error != EAGAIN)
2401 			break;
2402 	}
2403 
2404 	while (spa->spa_scrub_inflight)
2405 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2406 
2407 	spa->spa_scrub_active = 0;
2408 	cv_broadcast(&spa->spa_scrub_cv);
2409 
2410 	mutex_exit(&spa->spa_scrub_lock);
2411 
2412 	spa_config_enter(spa, RW_WRITER, FTAG);
2413 
2414 	mutex_enter(&spa->spa_scrub_lock);
2415 
2416 	/*
2417 	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
2418 	 * AND the spa config lock to synchronize with any config changes
2419 	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
2420 	 */
2421 	if (spa->spa_scrub_restart_txg != 0)
2422 		error = ERESTART;
2423 
2424 	if (spa->spa_scrub_stop)
2425 		error = EINTR;
2426 
2427 	/*
2428 	 * Even if there were uncorrectable errors, we consider the scrub
2429 	 * completed.  The downside is that if there is a transient error during
2430 	 * a resilver, we won't resilver the data properly to the target.  But
2431 	 * if the damage is permanent (more likely) we will resilver forever,
2432 	 * which isn't really acceptable.  Since there is enough information for
2433 	 * the user to know what has failed and why, this seems like a more
2434 	 * tractable approach.
2435 	 */
2436 	complete = (error == 0);
2437 
2438 	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
2439 	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2440 	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
2441 	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
2442 
2443 	mutex_exit(&spa->spa_scrub_lock);
2444 
2445 	/*
2446 	 * If the scrub/resilver completed, update all DTLs to reflect this.
2447 	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
2448 	 */
2449 	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
2450 	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
2451 	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
2452 	spa_errlog_rotate(spa);
2453 
2454 	spa_config_exit(spa, FTAG);
2455 
2456 	mutex_enter(&spa->spa_scrub_lock);
2457 
2458 	/*
2459 	 * We may have finished replacing a device.
2460 	 * Let the async thread assess this and handle the detach.
2461 	 */
2462 	spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
2463 
2464 	/*
2465 	 * If we were told to restart, our final act is to start a new scrub.
2466 	 */
2467 	if (error == ERESTART)
2468 		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
2469 		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
2470 
2471 	spa->spa_scrub_type = POOL_SCRUB_NONE;
2472 	spa->spa_scrub_active = 0;
2473 	spa->spa_scrub_thread = NULL;
2474 	cv_broadcast(&spa->spa_scrub_cv);
2475 	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
2476 	thread_exit();
2477 }
2478 
2479 void
2480 spa_scrub_suspend(spa_t *spa)
2481 {
2482 	mutex_enter(&spa->spa_scrub_lock);
2483 	spa->spa_scrub_suspended++;
2484 	while (spa->spa_scrub_active) {
2485 		cv_broadcast(&spa->spa_scrub_cv);
2486 		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2487 	}
2488 	while (spa->spa_scrub_inflight)
2489 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2490 	mutex_exit(&spa->spa_scrub_lock);
2491 }
2492 
2493 void
2494 spa_scrub_resume(spa_t *spa)
2495 {
2496 	mutex_enter(&spa->spa_scrub_lock);
2497 	ASSERT(spa->spa_scrub_suspended != 0);
2498 	if (--spa->spa_scrub_suspended == 0)
2499 		cv_broadcast(&spa->spa_scrub_cv);
2500 	mutex_exit(&spa->spa_scrub_lock);
2501 }
2502 
2503 void
2504 spa_scrub_restart(spa_t *spa, uint64_t txg)
2505 {
2506 	/*
2507 	 * Something happened (e.g. snapshot create/delete) that means
2508 	 * we must restart any in-progress scrubs.  The itinerary will
2509 	 * fix this properly.
2510 	 */
2511 	mutex_enter(&spa->spa_scrub_lock);
2512 	spa->spa_scrub_restart_txg = txg;
2513 	mutex_exit(&spa->spa_scrub_lock);
2514 }
2515 
2516 int
2517 spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
2518 {
2519 	space_seg_t *ss;
2520 	uint64_t mintxg, maxtxg;
2521 	vdev_t *rvd = spa->spa_root_vdev;
2522 
2523 	if ((uint_t)type >= POOL_SCRUB_TYPES)
2524 		return (ENOTSUP);
2525 
2526 	mutex_enter(&spa->spa_scrub_lock);
2527 
2528 	/*
2529 	 * If there's a scrub or resilver already in progress, stop it.
2530 	 */
2531 	while (spa->spa_scrub_thread != NULL) {
2532 		/*
2533 		 * Don't stop a resilver unless forced.
2534 		 */
2535 		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
2536 			mutex_exit(&spa->spa_scrub_lock);
2537 			return (EBUSY);
2538 		}
2539 		spa->spa_scrub_stop = 1;
2540 		cv_broadcast(&spa->spa_scrub_cv);
2541 		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2542 	}
2543 
2544 	/*
2545 	 * Terminate the previous traverse.
2546 	 */
2547 	if (spa->spa_scrub_th != NULL) {
2548 		traverse_fini(spa->spa_scrub_th);
2549 		spa->spa_scrub_th = NULL;
2550 	}
2551 
2552 	if (rvd == NULL) {
2553 		ASSERT(spa->spa_scrub_stop == 0);
2554 		ASSERT(spa->spa_scrub_type == type);
2555 		ASSERT(spa->spa_scrub_restart_txg == 0);
2556 		mutex_exit(&spa->spa_scrub_lock);
2557 		return (0);
2558 	}
2559 
2560 	mintxg = TXG_INITIAL - 1;
2561 	maxtxg = spa_last_synced_txg(spa) + 1;
2562 
2563 	mutex_enter(&rvd->vdev_dtl_lock);
2564 
2565 	if (rvd->vdev_dtl_map.sm_space == 0) {
2566 		/*
2567 		 * The pool-wide DTL is empty.
2568 		 * If this is a resilver, there's nothing to do except
2569 		 * check whether any in-progress replacements have completed.
2570 		 */
2571 		if (type == POOL_SCRUB_RESILVER) {
2572 			type = POOL_SCRUB_NONE;
2573 			spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
2574 		}
2575 	} else {
2576 		/*
2577 		 * The pool-wide DTL is non-empty.
2578 		 * If this is a normal scrub, upgrade to a resilver instead.
2579 		 */
2580 		if (type == POOL_SCRUB_EVERYTHING)
2581 			type = POOL_SCRUB_RESILVER;
2582 	}
2583 
2584 	if (type == POOL_SCRUB_RESILVER) {
2585 		/*
2586 		 * Determine the resilvering boundaries.
2587 		 *
2588 		 * Note: (mintxg, maxtxg) is an open interval,
2589 		 * i.e. mintxg and maxtxg themselves are not included.
2590 		 *
2591 		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
2592 		 * so we don't claim to resilver a txg that's still changing.
2593 		 */
2594 		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
2595 		mintxg = ss->ss_start - 1;
2596 		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
2597 		maxtxg = MIN(ss->ss_end, maxtxg);
2598 	}
2599 
2600 	mutex_exit(&rvd->vdev_dtl_lock);
2601 
2602 	spa->spa_scrub_stop = 0;
2603 	spa->spa_scrub_type = type;
2604 	spa->spa_scrub_restart_txg = 0;
2605 
2606 	if (type != POOL_SCRUB_NONE) {
2607 		spa->spa_scrub_mintxg = mintxg;
2608 		spa->spa_scrub_maxtxg = maxtxg;
2609 		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
2610 		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
2611 		    ZIO_FLAG_CANFAIL);
2612 		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
2613 		spa->spa_scrub_thread = thread_create(NULL, 0,
2614 		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
2615 	}
2616 
2617 	mutex_exit(&spa->spa_scrub_lock);
2618 
2619 	return (0);
2620 }
2621 
2622 /*
2623  * ==========================================================================
2624  * SPA async task processing
2625  * ==========================================================================
2626  */
2627 
2628 static void
2629 spa_async_reopen(spa_t *spa)
2630 {
2631 	vdev_t *rvd = spa->spa_root_vdev;
2632 	vdev_t *tvd;
2633 	int c;
2634 
2635 	spa_config_enter(spa, RW_WRITER, FTAG);
2636 
2637 	for (c = 0; c < rvd->vdev_children; c++) {
2638 		tvd = rvd->vdev_child[c];
2639 		if (tvd->vdev_reopen_wanted) {
2640 			tvd->vdev_reopen_wanted = 0;
2641 			vdev_reopen(tvd);
2642 		}
2643 	}
2644 
2645 	spa_config_exit(spa, FTAG);
2646 }
2647 
2648 static void
2649 spa_async_thread(spa_t *spa)
2650 {
2651 	int tasks;
2652 
2653 	ASSERT(spa->spa_sync_on);
2654 
2655 	mutex_enter(&spa->spa_async_lock);
2656 	tasks = spa->spa_async_tasks;
2657 	spa->spa_async_tasks = 0;
2658 	mutex_exit(&spa->spa_async_lock);
2659 
2660 	/*
2661 	 * See if the config needs to be updated.
2662 	 */
2663 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
2664 		mutex_enter(&spa_namespace_lock);
2665 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2666 		mutex_exit(&spa_namespace_lock);
2667 	}
2668 
2669 	/*
2670 	 * See if any devices need to be reopened.
2671 	 */
2672 	if (tasks & SPA_ASYNC_REOPEN)
2673 		spa_async_reopen(spa);
2674 
2675 	/*
2676 	 * If any devices are done replacing, detach them.
2677 	 */
2678 	if (tasks & SPA_ASYNC_REPLACE_DONE)
2679 		spa_vdev_replace_done(spa);
2680 
2681 	/*
2682 	 * Kick off a scrub.
2683 	 */
2684 	if (tasks & SPA_ASYNC_SCRUB)
2685 		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
2686 
2687 	/*
2688 	 * Kick off a resilver.
2689 	 */
2690 	if (tasks & SPA_ASYNC_RESILVER)
2691 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
2692 
2693 	/*
2694 	 * Let the world know that we're done.
2695 	 */
2696 	mutex_enter(&spa->spa_async_lock);
2697 	spa->spa_async_thread = NULL;
2698 	cv_broadcast(&spa->spa_async_cv);
2699 	mutex_exit(&spa->spa_async_lock);
2700 	thread_exit();
2701 }
2702 
2703 void
2704 spa_async_suspend(spa_t *spa)
2705 {
2706 	mutex_enter(&spa->spa_async_lock);
2707 	spa->spa_async_suspended++;
2708 	while (spa->spa_async_thread != NULL)
2709 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
2710 	mutex_exit(&spa->spa_async_lock);
2711 }
2712 
2713 void
2714 spa_async_resume(spa_t *spa)
2715 {
2716 	mutex_enter(&spa->spa_async_lock);
2717 	ASSERT(spa->spa_async_suspended != 0);
2718 	spa->spa_async_suspended--;
2719 	mutex_exit(&spa->spa_async_lock);
2720 }
2721 
2722 static void
2723 spa_async_dispatch(spa_t *spa)
2724 {
2725 	mutex_enter(&spa->spa_async_lock);
2726 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
2727 	    spa->spa_async_thread == NULL &&
2728 	    rootdir != NULL && !vn_is_readonly(rootdir))
2729 		spa->spa_async_thread = thread_create(NULL, 0,
2730 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
2731 	mutex_exit(&spa->spa_async_lock);
2732 }
2733 
2734 void
2735 spa_async_request(spa_t *spa, int task)
2736 {
2737 	mutex_enter(&spa->spa_async_lock);
2738 	spa->spa_async_tasks |= task;
2739 	mutex_exit(&spa->spa_async_lock);
2740 }
2741 
2742 /*
2743  * ==========================================================================
2744  * SPA syncing routines
2745  * ==========================================================================
2746  */
2747 
2748 static void
2749 spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
2750 {
2751 	bplist_t *bpl = &spa->spa_sync_bplist;
2752 	dmu_tx_t *tx;
2753 	blkptr_t blk;
2754 	uint64_t itor = 0;
2755 	zio_t *zio;
2756 	int error;
2757 	uint8_t c = 1;
2758 
2759 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
2760 
2761 	while (bplist_iterate(bpl, &itor, &blk) == 0)
2762 		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
2763 
2764 	error = zio_wait(zio);
2765 	ASSERT3U(error, ==, 0);
2766 
2767 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2768 	bplist_vacate(bpl, tx);
2769 
2770 	/*
2771 	 * Pre-dirty the first block so we sync to convergence faster.
2772 	 * (Usually only the first block is needed.)
2773 	 */
2774 	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
2775 	dmu_tx_commit(tx);
2776 }
2777 
2778 static void
2779 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
2780 {
2781 	char *packed = NULL;
2782 	size_t nvsize = 0;
2783 	dmu_buf_t *db;
2784 
2785 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
2786 
2787 	packed = kmem_alloc(nvsize, KM_SLEEP);
2788 
2789 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
2790 	    KM_SLEEP) == 0);
2791 
2792 	dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx);
2793 
2794 	kmem_free(packed, nvsize);
2795 
2796 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
2797 	dmu_buf_will_dirty(db, tx);
2798 	*(uint64_t *)db->db_data = nvsize;
2799 	dmu_buf_rele(db, FTAG);
2800 }
2801 
2802 static void
2803 spa_sync_spares(spa_t *spa, dmu_tx_t *tx)
2804 {
2805 	nvlist_t *nvroot;
2806 	nvlist_t **spares;
2807 	int i;
2808 
2809 	if (!spa->spa_sync_spares)
2810 		return;
2811 
2812 	/*
2813 	 * Update the MOS nvlist describing the list of available spares.
2814 	 * spa_validate_spares() will have already made sure this nvlist is
2815 	 * valid and the vdevs are labelled appropriately.
2816 	 */
2817 	if (spa->spa_spares_object == 0) {
2818 		spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset,
2819 		    DMU_OT_PACKED_NVLIST, 1 << 14,
2820 		    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2821 		VERIFY(zap_update(spa->spa_meta_objset,
2822 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES,
2823 		    sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0);
2824 	}
2825 
2826 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2827 	if (spa->spa_nspares == 0) {
2828 		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2829 		    NULL, 0) == 0);
2830 	} else {
2831 		spares = kmem_alloc(spa->spa_nspares * sizeof (void *),
2832 		    KM_SLEEP);
2833 		for (i = 0; i < spa->spa_nspares; i++)
2834 			spares[i] = vdev_config_generate(spa,
2835 			    spa->spa_spares[i], B_FALSE, B_TRUE);
2836 		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2837 		    spares, spa->spa_nspares) == 0);
2838 		for (i = 0; i < spa->spa_nspares; i++)
2839 			nvlist_free(spares[i]);
2840 		kmem_free(spares, spa->spa_nspares * sizeof (void *));
2841 	}
2842 
2843 	spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx);
2844 	nvlist_free(nvroot);
2845 
2846 	spa->spa_sync_spares = B_FALSE;
2847 }
2848 
2849 static void
2850 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
2851 {
2852 	nvlist_t *config;
2853 
2854 	if (list_is_empty(&spa->spa_dirty_list))
2855 		return;
2856 
2857 	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
2858 
2859 	if (spa->spa_config_syncing)
2860 		nvlist_free(spa->spa_config_syncing);
2861 	spa->spa_config_syncing = config;
2862 
2863 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
2864 }
2865 
2866 static void
2867 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
2868 {
2869 	spa_t *spa = arg1;
2870 	nvlist_t *nvp = arg2;
2871 	nvpair_t *nvpair;
2872 	objset_t *mos = spa->spa_meta_objset;
2873 	uint64_t zapobj;
2874 
2875 	mutex_enter(&spa->spa_props_lock);
2876 	if (spa->spa_pool_props_object == 0) {
2877 		zapobj = zap_create(mos, DMU_OT_POOL_PROPS, DMU_OT_NONE, 0, tx);
2878 		VERIFY(zapobj > 0);
2879 
2880 		spa->spa_pool_props_object = zapobj;
2881 
2882 		VERIFY(zap_update(mos, DMU_POOL_DIRECTORY_OBJECT,
2883 		    DMU_POOL_PROPS, 8, 1,
2884 		    &spa->spa_pool_props_object, tx) == 0);
2885 	}
2886 	mutex_exit(&spa->spa_props_lock);
2887 
2888 	nvpair = NULL;
2889 	while ((nvpair = nvlist_next_nvpair(nvp, nvpair))) {
2890 		switch (zpool_name_to_prop(nvpair_name(nvpair))) {
2891 		case ZFS_PROP_BOOTFS:
2892 			VERIFY(nvlist_lookup_uint64(nvp,
2893 			    nvpair_name(nvpair), &spa->spa_bootfs) == 0);
2894 			VERIFY(zap_update(mos,
2895 			    spa->spa_pool_props_object,
2896 			    zpool_prop_to_name(ZFS_PROP_BOOTFS), 8, 1,
2897 			    &spa->spa_bootfs, tx) == 0);
2898 			break;
2899 		}
2900 	}
2901 }
2902 
2903 /*
2904  * Sync the specified transaction group.  New blocks may be dirtied as
2905  * part of the process, so we iterate until it converges.
2906  */
2907 void
2908 spa_sync(spa_t *spa, uint64_t txg)
2909 {
2910 	dsl_pool_t *dp = spa->spa_dsl_pool;
2911 	objset_t *mos = spa->spa_meta_objset;
2912 	bplist_t *bpl = &spa->spa_sync_bplist;
2913 	vdev_t *rvd = spa->spa_root_vdev;
2914 	vdev_t *vd;
2915 	dmu_tx_t *tx;
2916 	int dirty_vdevs;
2917 
2918 	/*
2919 	 * Lock out configuration changes.
2920 	 */
2921 	spa_config_enter(spa, RW_READER, FTAG);
2922 
2923 	spa->spa_syncing_txg = txg;
2924 	spa->spa_sync_pass = 0;
2925 
2926 	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
2927 
2928 	tx = dmu_tx_create_assigned(dp, txg);
2929 
2930 	/*
2931 	 * If we are upgrading to ZFS_VERSION_RAIDZ_DEFLATE this txg,
2932 	 * set spa_deflate if we have no raid-z vdevs.
2933 	 */
2934 	if (spa->spa_ubsync.ub_version < ZFS_VERSION_RAIDZ_DEFLATE &&
2935 	    spa->spa_uberblock.ub_version >= ZFS_VERSION_RAIDZ_DEFLATE) {
2936 		int i;
2937 
2938 		for (i = 0; i < rvd->vdev_children; i++) {
2939 			vd = rvd->vdev_child[i];
2940 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
2941 				break;
2942 		}
2943 		if (i == rvd->vdev_children) {
2944 			spa->spa_deflate = TRUE;
2945 			VERIFY(0 == zap_add(spa->spa_meta_objset,
2946 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2947 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
2948 		}
2949 	}
2950 
2951 	/*
2952 	 * If anything has changed in this txg, push the deferred frees
2953 	 * from the previous txg.  If not, leave them alone so that we
2954 	 * don't generate work on an otherwise idle system.
2955 	 */
2956 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
2957 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
2958 	    !txg_list_empty(&dp->dp_sync_tasks, txg))
2959 		spa_sync_deferred_frees(spa, txg);
2960 
2961 	/*
2962 	 * Iterate to convergence.
2963 	 */
2964 	do {
2965 		spa->spa_sync_pass++;
2966 
2967 		spa_sync_config_object(spa, tx);
2968 		spa_sync_spares(spa, tx);
2969 		spa_errlog_sync(spa, txg);
2970 		dsl_pool_sync(dp, txg);
2971 
2972 		dirty_vdevs = 0;
2973 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
2974 			vdev_sync(vd, txg);
2975 			dirty_vdevs++;
2976 		}
2977 
2978 		bplist_sync(bpl, tx);
2979 	} while (dirty_vdevs);
2980 
2981 	bplist_close(bpl);
2982 
2983 	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
2984 
2985 	/*
2986 	 * Rewrite the vdev configuration (which includes the uberblock)
2987 	 * to commit the transaction group.
2988 	 *
2989 	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
2990 	 * Otherwise, pick a random top-level vdev that's known to be
2991 	 * visible in the config cache (see spa_vdev_add() for details).
2992 	 * If the write fails, try the next vdev until we're tried them all.
2993 	 */
2994 	if (!list_is_empty(&spa->spa_dirty_list)) {
2995 		VERIFY(vdev_config_sync(rvd, txg) == 0);
2996 	} else {
2997 		int children = rvd->vdev_children;
2998 		int c0 = spa_get_random(children);
2999 		int c;
3000 
3001 		for (c = 0; c < children; c++) {
3002 			vd = rvd->vdev_child[(c0 + c) % children];
3003 			if (vd->vdev_ms_array == 0)
3004 				continue;
3005 			if (vdev_config_sync(vd, txg) == 0)
3006 				break;
3007 		}
3008 		if (c == children)
3009 			VERIFY(vdev_config_sync(rvd, txg) == 0);
3010 	}
3011 
3012 	dmu_tx_commit(tx);
3013 
3014 	/*
3015 	 * Clear the dirty config list.
3016 	 */
3017 	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
3018 		vdev_config_clean(vd);
3019 
3020 	/*
3021 	 * Now that the new config has synced transactionally,
3022 	 * let it become visible to the config cache.
3023 	 */
3024 	if (spa->spa_config_syncing != NULL) {
3025 		spa_config_set(spa, spa->spa_config_syncing);
3026 		spa->spa_config_txg = txg;
3027 		spa->spa_config_syncing = NULL;
3028 	}
3029 
3030 	/*
3031 	 * Make a stable copy of the fully synced uberblock.
3032 	 * We use this as the root for pool traversals.
3033 	 */
3034 	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
3035 
3036 	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
3037 
3038 	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
3039 	spa->spa_traverse_wanted = 0;
3040 	spa->spa_ubsync = spa->spa_uberblock;
3041 	rw_exit(&spa->spa_traverse_lock);
3042 
3043 	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
3044 
3045 	/*
3046 	 * Clean up the ZIL records for the synced txg.
3047 	 */
3048 	dsl_pool_zil_clean(dp);
3049 
3050 	/*
3051 	 * Update usable space statistics.
3052 	 */
3053 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
3054 		vdev_sync_done(vd, txg);
3055 
3056 	/*
3057 	 * It had better be the case that we didn't dirty anything
3058 	 * since vdev_config_sync().
3059 	 */
3060 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
3061 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
3062 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
3063 	ASSERT(bpl->bpl_queue == NULL);
3064 
3065 	spa_config_exit(spa, FTAG);
3066 
3067 	/*
3068 	 * If any async tasks have been requested, kick them off.
3069 	 */
3070 	spa_async_dispatch(spa);
3071 }
3072 
3073 /*
3074  * Sync all pools.  We don't want to hold the namespace lock across these
3075  * operations, so we take a reference on the spa_t and drop the lock during the
3076  * sync.
3077  */
3078 void
3079 spa_sync_allpools(void)
3080 {
3081 	spa_t *spa = NULL;
3082 	mutex_enter(&spa_namespace_lock);
3083 	while ((spa = spa_next(spa)) != NULL) {
3084 		if (spa_state(spa) != POOL_STATE_ACTIVE)
3085 			continue;
3086 		spa_open_ref(spa, FTAG);
3087 		mutex_exit(&spa_namespace_lock);
3088 		txg_wait_synced(spa_get_dsl(spa), 0);
3089 		mutex_enter(&spa_namespace_lock);
3090 		spa_close(spa, FTAG);
3091 	}
3092 	mutex_exit(&spa_namespace_lock);
3093 }
3094 
3095 /*
3096  * ==========================================================================
3097  * Miscellaneous routines
3098  * ==========================================================================
3099  */
3100 
3101 /*
3102  * Remove all pools in the system.
3103  */
3104 void
3105 spa_evict_all(void)
3106 {
3107 	spa_t *spa;
3108 
3109 	/*
3110 	 * Remove all cached state.  All pools should be closed now,
3111 	 * so every spa in the AVL tree should be unreferenced.
3112 	 */
3113 	mutex_enter(&spa_namespace_lock);
3114 	while ((spa = spa_next(NULL)) != NULL) {
3115 		/*
3116 		 * Stop async tasks.  The async thread may need to detach
3117 		 * a device that's been replaced, which requires grabbing
3118 		 * spa_namespace_lock, so we must drop it here.
3119 		 */
3120 		spa_open_ref(spa, FTAG);
3121 		mutex_exit(&spa_namespace_lock);
3122 		spa_async_suspend(spa);
3123 		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
3124 		mutex_enter(&spa_namespace_lock);
3125 		spa_close(spa, FTAG);
3126 
3127 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3128 			spa_unload(spa);
3129 			spa_deactivate(spa);
3130 		}
3131 		spa_remove(spa);
3132 	}
3133 	mutex_exit(&spa_namespace_lock);
3134 }
3135 
3136 vdev_t *
3137 spa_lookup_by_guid(spa_t *spa, uint64_t guid)
3138 {
3139 	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
3140 }
3141 
3142 void
3143 spa_upgrade(spa_t *spa)
3144 {
3145 	spa_config_enter(spa, RW_WRITER, FTAG);
3146 
3147 	/*
3148 	 * This should only be called for a non-faulted pool, and since a
3149 	 * future version would result in an unopenable pool, this shouldn't be
3150 	 * possible.
3151 	 */
3152 	ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION);
3153 
3154 	spa->spa_uberblock.ub_version = ZFS_VERSION;
3155 	vdev_config_dirty(spa->spa_root_vdev);
3156 
3157 	spa_config_exit(spa, FTAG);
3158 
3159 	txg_wait_synced(spa_get_dsl(spa), 0);
3160 }
3161 
3162 boolean_t
3163 spa_has_spare(spa_t *spa, uint64_t guid)
3164 {
3165 	int i;
3166 	uint64_t spareguid;
3167 
3168 	for (i = 0; i < spa->spa_nspares; i++)
3169 		if (spa->spa_spares[i]->vdev_guid == guid)
3170 			return (B_TRUE);
3171 
3172 	for (i = 0; i < spa->spa_pending_nspares; i++) {
3173 		if (nvlist_lookup_uint64(spa->spa_pending_spares[i],
3174 		    ZPOOL_CONFIG_GUID, &spareguid) == 0 &&
3175 		    spareguid == guid)
3176 			return (B_TRUE);
3177 	}
3178 
3179 	return (B_FALSE);
3180 }
3181 
3182 int
3183 spa_set_props(spa_t *spa, nvlist_t *nvp)
3184 {
3185 	return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
3186 	    spa, nvp, 3));
3187 }
3188 
3189 int
3190 spa_get_props(spa_t *spa, nvlist_t **nvp)
3191 {
3192 	zap_cursor_t zc;
3193 	zap_attribute_t za;
3194 	objset_t *mos = spa->spa_meta_objset;
3195 	zfs_source_t src;
3196 	zfs_prop_t prop;
3197 	nvlist_t *propval;
3198 	uint64_t value;
3199 	int err;
3200 
3201 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3202 
3203 	mutex_enter(&spa->spa_props_lock);
3204 	/* If no props object, then just return empty nvlist */
3205 	if (spa->spa_pool_props_object == 0) {
3206 		mutex_exit(&spa->spa_props_lock);
3207 		return (0);
3208 	}
3209 
3210 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
3211 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
3212 	    zap_cursor_advance(&zc)) {
3213 
3214 		if ((prop = zpool_name_to_prop(za.za_name)) == ZFS_PROP_INVAL)
3215 			continue;
3216 
3217 		VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3218 		switch (za.za_integer_length) {
3219 		case 8:
3220 			if (zfs_prop_default_numeric(prop) ==
3221 			    za.za_first_integer)
3222 				src = ZFS_SRC_DEFAULT;
3223 			else
3224 				src = ZFS_SRC_LOCAL;
3225 			value = za.za_first_integer;
3226 
3227 			if (prop == ZFS_PROP_BOOTFS) {
3228 				dsl_pool_t *dp;
3229 				dsl_dataset_t *ds = NULL;
3230 				char strval[MAXPATHLEN];
3231 
3232 				dp = spa_get_dsl(spa);
3233 				rw_enter(&dp->dp_config_rwlock, RW_READER);
3234 				if ((err = dsl_dataset_open_obj(dp,
3235 				    za.za_first_integer, NULL, DS_MODE_NONE,
3236 				    FTAG, &ds)) != 0) {
3237 					rw_exit(&dp->dp_config_rwlock);
3238 					break;
3239 				}
3240 				dsl_dataset_name(ds, strval);
3241 				dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
3242 				rw_exit(&dp->dp_config_rwlock);
3243 
3244 				VERIFY(nvlist_add_uint64(propval,
3245 				    ZFS_PROP_SOURCE, src) == 0);
3246 				VERIFY(nvlist_add_string(propval,
3247 				    ZFS_PROP_VALUE, strval) == 0);
3248 			} else {
3249 				VERIFY(nvlist_add_uint64(propval,
3250 				    ZFS_PROP_SOURCE, src) == 0);
3251 				VERIFY(nvlist_add_uint64(propval,
3252 				    ZFS_PROP_VALUE, value) == 0);
3253 			}
3254 			VERIFY(nvlist_add_nvlist(*nvp, za.za_name,
3255 			    propval) == 0);
3256 			break;
3257 		}
3258 		nvlist_free(propval);
3259 	}
3260 	zap_cursor_fini(&zc);
3261 	mutex_exit(&spa->spa_props_lock);
3262 	if (err && err != ENOENT) {
3263 		nvlist_free(*nvp);
3264 		return (err);
3265 	}
3266 
3267 	return (0);
3268 }
3269 
3270 /*
3271  * If the bootfs property value is dsobj, clear it.
3272  */
3273 void
3274 spa_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
3275 {
3276 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
3277 		VERIFY(zap_remove(spa->spa_meta_objset,
3278 		    spa->spa_pool_props_object,
3279 		    zpool_prop_to_name(ZFS_PROP_BOOTFS), tx) == 0);
3280 		spa->spa_bootfs = 0;
3281 	}
3282 }
3283