xref: /titanic_41/usr/src/uts/common/fs/zfs/spa.c (revision 9404882939d18ddd3c94a5bd3da7a0449c195a5d)
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 	error = vdev_open(rvd);
488 	if (error != 0)
489 		goto out;
490 
491 	/*
492 	 * Validate the labels for all leaf vdevs.  We need to grab the config
493 	 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD
494 	 * flag.
495 	 */
496 	spa_config_enter(spa, RW_READER, FTAG);
497 	error = vdev_validate(rvd);
498 	spa_config_exit(spa, FTAG);
499 
500 	if (error != 0)
501 		goto out;
502 
503 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
504 		error = ENXIO;
505 		goto out;
506 	}
507 
508 	/*
509 	 * Find the best uberblock.
510 	 */
511 	bzero(ub, sizeof (uberblock_t));
512 
513 	zio = zio_root(spa, NULL, NULL,
514 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
515 	vdev_uberblock_load(zio, rvd, ub);
516 	error = zio_wait(zio);
517 
518 	/*
519 	 * If we weren't able to find a single valid uberblock, return failure.
520 	 */
521 	if (ub->ub_txg == 0) {
522 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
523 		    VDEV_AUX_CORRUPT_DATA);
524 		error = ENXIO;
525 		goto out;
526 	}
527 
528 	/*
529 	 * If the pool is newer than the code, we can't open it.
530 	 */
531 	if (ub->ub_version > ZFS_VERSION) {
532 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
533 		    VDEV_AUX_VERSION_NEWER);
534 		error = ENOTSUP;
535 		goto out;
536 	}
537 
538 	/*
539 	 * If the vdev guid sum doesn't match the uberblock, we have an
540 	 * incomplete configuration.
541 	 */
542 	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
543 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
544 		    VDEV_AUX_BAD_GUID_SUM);
545 		error = ENXIO;
546 		goto out;
547 	}
548 
549 	/*
550 	 * Initialize internal SPA structures.
551 	 */
552 	spa->spa_state = POOL_STATE_ACTIVE;
553 	spa->spa_ubsync = spa->spa_uberblock;
554 	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
555 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
556 	if (error) {
557 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
558 		    VDEV_AUX_CORRUPT_DATA);
559 		goto out;
560 	}
561 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
562 
563 	if (zap_lookup(spa->spa_meta_objset,
564 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
565 	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
566 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
567 		    VDEV_AUX_CORRUPT_DATA);
568 		error = EIO;
569 		goto out;
570 	}
571 
572 	if (!mosconfig) {
573 		nvlist_t *newconfig;
574 		uint64_t hostid;
575 
576 		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
577 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
578 			    VDEV_AUX_CORRUPT_DATA);
579 			error = EIO;
580 			goto out;
581 		}
582 
583 		if (nvlist_lookup_uint64(newconfig, ZPOOL_CONFIG_HOSTID,
584 		    &hostid) == 0) {
585 			char *hostname;
586 			unsigned long myhostid = 0;
587 
588 			VERIFY(nvlist_lookup_string(newconfig,
589 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
590 
591 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
592 			if (hostid != 0 && myhostid != 0 &&
593 			    (unsigned long)hostid != myhostid) {
594 				cmn_err(CE_WARN, "pool '%s' could not be "
595 				    "loaded as it was last accessed by "
596 				    "another system (host: %s hostid: 0x%lx).  "
597 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
598 				    spa->spa_name, hostname,
599 				    (unsigned long)hostid);
600 				error = EBADF;
601 				goto out;
602 			}
603 		}
604 
605 		spa_config_set(spa, newconfig);
606 		spa_unload(spa);
607 		spa_deactivate(spa);
608 		spa_activate(spa);
609 
610 		return (spa_load(spa, newconfig, state, B_TRUE));
611 	}
612 
613 	if (zap_lookup(spa->spa_meta_objset,
614 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
615 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
616 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
617 		    VDEV_AUX_CORRUPT_DATA);
618 		error = EIO;
619 		goto out;
620 	}
621 
622 	/*
623 	 * Load the bit that tells us to use the new accounting function
624 	 * (raid-z deflation).  If we have an older pool, this will not
625 	 * be present.
626 	 */
627 	error = zap_lookup(spa->spa_meta_objset,
628 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
629 	    sizeof (uint64_t), 1, &spa->spa_deflate);
630 	if (error != 0 && error != ENOENT) {
631 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
632 		    VDEV_AUX_CORRUPT_DATA);
633 		error = EIO;
634 		goto out;
635 	}
636 
637 	/*
638 	 * Load the persistent error log.  If we have an older pool, this will
639 	 * not be present.
640 	 */
641 	error = zap_lookup(spa->spa_meta_objset,
642 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
643 	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
644 	if (error != 0 && error != ENOENT) {
645 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
646 		    VDEV_AUX_CORRUPT_DATA);
647 		error = EIO;
648 		goto out;
649 	}
650 
651 	error = zap_lookup(spa->spa_meta_objset,
652 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
653 	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
654 	if (error != 0 && error != ENOENT) {
655 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
656 		    VDEV_AUX_CORRUPT_DATA);
657 		error = EIO;
658 		goto out;
659 	}
660 
661 	/*
662 	 * Load the history object.  If we have an older pool, this
663 	 * will not be present.
664 	 */
665 	error = zap_lookup(spa->spa_meta_objset,
666 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
667 	    sizeof (uint64_t), 1, &spa->spa_history);
668 	if (error != 0 && error != ENOENT) {
669 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
670 		    VDEV_AUX_CORRUPT_DATA);
671 		error = EIO;
672 		goto out;
673 	}
674 
675 	/*
676 	 * Load any hot spares for this pool.
677 	 */
678 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
679 	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object);
680 	if (error != 0 && error != ENOENT) {
681 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
682 		    VDEV_AUX_CORRUPT_DATA);
683 		error = EIO;
684 		goto out;
685 	}
686 	if (error == 0) {
687 		ASSERT(spa_version(spa) >= ZFS_VERSION_SPARES);
688 		if (load_nvlist(spa, spa->spa_spares_object,
689 		    &spa->spa_sparelist) != 0) {
690 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
691 			    VDEV_AUX_CORRUPT_DATA);
692 			error = EIO;
693 			goto out;
694 		}
695 
696 		spa_config_enter(spa, RW_WRITER, FTAG);
697 		spa_load_spares(spa);
698 		spa_config_exit(spa, FTAG);
699 	}
700 
701 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
702 	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
703 
704 	if (error && error != ENOENT) {
705 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
706 		    VDEV_AUX_CORRUPT_DATA);
707 		error = EIO;
708 		goto out;
709 	}
710 
711 	if (error == 0) {
712 		(void) zap_lookup(spa->spa_meta_objset,
713 		    spa->spa_pool_props_object,
714 		    zpool_prop_to_name(ZFS_PROP_BOOTFS),
715 		    sizeof (uint64_t), 1, &spa->spa_bootfs);
716 	}
717 
718 	/*
719 	 * Load the vdev state for all toplevel vdevs.
720 	 */
721 	vdev_load(rvd);
722 
723 	/*
724 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
725 	 */
726 	spa_config_enter(spa, RW_WRITER, FTAG);
727 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
728 	spa_config_exit(spa, FTAG);
729 
730 	/*
731 	 * Check the state of the root vdev.  If it can't be opened, it
732 	 * indicates one or more toplevel vdevs are faulted.
733 	 */
734 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
735 		error = ENXIO;
736 		goto out;
737 	}
738 
739 	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
740 		dmu_tx_t *tx;
741 		int need_update = B_FALSE;
742 		int c;
743 
744 		/*
745 		 * Claim log blocks that haven't been committed yet.
746 		 * This must all happen in a single txg.
747 		 */
748 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
749 		    spa_first_txg(spa));
750 		(void) dmu_objset_find(spa->spa_name,
751 		    zil_claim, tx, DS_FIND_CHILDREN);
752 		dmu_tx_commit(tx);
753 
754 		spa->spa_sync_on = B_TRUE;
755 		txg_sync_start(spa->spa_dsl_pool);
756 
757 		/*
758 		 * Wait for all claims to sync.
759 		 */
760 		txg_wait_synced(spa->spa_dsl_pool, 0);
761 
762 		/*
763 		 * If the config cache is stale, or we have uninitialized
764 		 * metaslabs (see spa_vdev_add()), then update the config.
765 		 */
766 		if (config_cache_txg != spa->spa_config_txg ||
767 		    state == SPA_LOAD_IMPORT)
768 			need_update = B_TRUE;
769 
770 		for (c = 0; c < rvd->vdev_children; c++)
771 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
772 				need_update = B_TRUE;
773 
774 		/*
775 		 * Update the config cache asychronously in case we're the
776 		 * root pool, in which case the config cache isn't writable yet.
777 		 */
778 		if (need_update)
779 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
780 	}
781 
782 	error = 0;
783 out:
784 	if (error && error != EBADF)
785 		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
786 	spa->spa_load_state = SPA_LOAD_NONE;
787 	spa->spa_ena = 0;
788 
789 	return (error);
790 }
791 
792 /*
793  * Pool Open/Import
794  *
795  * The import case is identical to an open except that the configuration is sent
796  * down from userland, instead of grabbed from the configuration cache.  For the
797  * case of an open, the pool configuration will exist in the
798  * POOL_STATE_UNITIALIZED state.
799  *
800  * The stats information (gen/count/ustats) is used to gather vdev statistics at
801  * the same time open the pool, without having to keep around the spa_t in some
802  * ambiguous state.
803  */
804 static int
805 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
806 {
807 	spa_t *spa;
808 	int error;
809 	int loaded = B_FALSE;
810 	int locked = B_FALSE;
811 
812 	*spapp = NULL;
813 
814 	/*
815 	 * As disgusting as this is, we need to support recursive calls to this
816 	 * function because dsl_dir_open() is called during spa_load(), and ends
817 	 * up calling spa_open() again.  The real fix is to figure out how to
818 	 * avoid dsl_dir_open() calling this in the first place.
819 	 */
820 	if (mutex_owner(&spa_namespace_lock) != curthread) {
821 		mutex_enter(&spa_namespace_lock);
822 		locked = B_TRUE;
823 	}
824 
825 	if ((spa = spa_lookup(pool)) == NULL) {
826 		if (locked)
827 			mutex_exit(&spa_namespace_lock);
828 		return (ENOENT);
829 	}
830 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
831 
832 		spa_activate(spa);
833 
834 		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
835 
836 		if (error == EBADF) {
837 			/*
838 			 * If vdev_validate() returns failure (indicated by
839 			 * EBADF), it indicates that one of the vdevs indicates
840 			 * that the pool has been exported or destroyed.  If
841 			 * this is the case, the config cache is out of sync and
842 			 * we should remove the pool from the namespace.
843 			 */
844 			zfs_post_ok(spa, NULL);
845 			spa_unload(spa);
846 			spa_deactivate(spa);
847 			spa_remove(spa);
848 			spa_config_sync();
849 			if (locked)
850 				mutex_exit(&spa_namespace_lock);
851 			return (ENOENT);
852 		}
853 
854 		if (error) {
855 			/*
856 			 * We can't open the pool, but we still have useful
857 			 * information: the state of each vdev after the
858 			 * attempted vdev_open().  Return this to the user.
859 			 */
860 			if (config != NULL && spa->spa_root_vdev != NULL) {
861 				spa_config_enter(spa, RW_READER, FTAG);
862 				*config = spa_config_generate(spa, NULL, -1ULL,
863 				    B_TRUE);
864 				spa_config_exit(spa, FTAG);
865 			}
866 			spa_unload(spa);
867 			spa_deactivate(spa);
868 			spa->spa_last_open_failed = B_TRUE;
869 			if (locked)
870 				mutex_exit(&spa_namespace_lock);
871 			*spapp = NULL;
872 			return (error);
873 		} else {
874 			zfs_post_ok(spa, NULL);
875 			spa->spa_last_open_failed = B_FALSE;
876 		}
877 
878 		loaded = B_TRUE;
879 	}
880 
881 	spa_open_ref(spa, tag);
882 	if (locked)
883 		mutex_exit(&spa_namespace_lock);
884 
885 	*spapp = spa;
886 
887 	if (config != NULL) {
888 		spa_config_enter(spa, RW_READER, FTAG);
889 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
890 		spa_config_exit(spa, FTAG);
891 	}
892 
893 	/*
894 	 * If we just loaded the pool, resilver anything that's out of date.
895 	 */
896 	if (loaded && (spa_mode & FWRITE))
897 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
898 
899 	return (0);
900 }
901 
902 int
903 spa_open(const char *name, spa_t **spapp, void *tag)
904 {
905 	return (spa_open_common(name, spapp, tag, NULL));
906 }
907 
908 /*
909  * Lookup the given spa_t, incrementing the inject count in the process,
910  * preventing it from being exported or destroyed.
911  */
912 spa_t *
913 spa_inject_addref(char *name)
914 {
915 	spa_t *spa;
916 
917 	mutex_enter(&spa_namespace_lock);
918 	if ((spa = spa_lookup(name)) == NULL) {
919 		mutex_exit(&spa_namespace_lock);
920 		return (NULL);
921 	}
922 	spa->spa_inject_ref++;
923 	mutex_exit(&spa_namespace_lock);
924 
925 	return (spa);
926 }
927 
928 void
929 spa_inject_delref(spa_t *spa)
930 {
931 	mutex_enter(&spa_namespace_lock);
932 	spa->spa_inject_ref--;
933 	mutex_exit(&spa_namespace_lock);
934 }
935 
936 static void
937 spa_add_spares(spa_t *spa, nvlist_t *config)
938 {
939 	nvlist_t **spares;
940 	uint_t i, nspares;
941 	nvlist_t *nvroot;
942 	uint64_t guid;
943 	vdev_stat_t *vs;
944 	uint_t vsc;
945 	uint64_t pool;
946 
947 	if (spa->spa_nspares == 0)
948 		return;
949 
950 	VERIFY(nvlist_lookup_nvlist(config,
951 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
952 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
953 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
954 	if (nspares != 0) {
955 		VERIFY(nvlist_add_nvlist_array(nvroot,
956 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
957 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
958 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
959 
960 		/*
961 		 * Go through and find any spares which have since been
962 		 * repurposed as an active spare.  If this is the case, update
963 		 * their status appropriately.
964 		 */
965 		for (i = 0; i < nspares; i++) {
966 			VERIFY(nvlist_lookup_uint64(spares[i],
967 			    ZPOOL_CONFIG_GUID, &guid) == 0);
968 			if (spa_spare_exists(guid, &pool) && pool != 0ULL) {
969 				VERIFY(nvlist_lookup_uint64_array(
970 				    spares[i], ZPOOL_CONFIG_STATS,
971 				    (uint64_t **)&vs, &vsc) == 0);
972 				vs->vs_state = VDEV_STATE_CANT_OPEN;
973 				vs->vs_aux = VDEV_AUX_SPARED;
974 			}
975 		}
976 	}
977 }
978 
979 int
980 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
981 {
982 	int error;
983 	spa_t *spa;
984 
985 	*config = NULL;
986 	error = spa_open_common(name, &spa, FTAG, config);
987 
988 	if (spa && *config != NULL) {
989 		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
990 		    spa_get_errlog_size(spa)) == 0);
991 
992 		spa_add_spares(spa, *config);
993 	}
994 
995 	/*
996 	 * We want to get the alternate root even for faulted pools, so we cheat
997 	 * and call spa_lookup() directly.
998 	 */
999 	if (altroot) {
1000 		if (spa == NULL) {
1001 			mutex_enter(&spa_namespace_lock);
1002 			spa = spa_lookup(name);
1003 			if (spa)
1004 				spa_altroot(spa, altroot, buflen);
1005 			else
1006 				altroot[0] = '\0';
1007 			spa = NULL;
1008 			mutex_exit(&spa_namespace_lock);
1009 		} else {
1010 			spa_altroot(spa, altroot, buflen);
1011 		}
1012 	}
1013 
1014 	if (spa != NULL)
1015 		spa_close(spa, FTAG);
1016 
1017 	return (error);
1018 }
1019 
1020 /*
1021  * Validate that the 'spares' array is well formed.  We must have an array of
1022  * nvlists, each which describes a valid leaf vdev.  If this is an import (mode
1023  * is VDEV_ALLOC_SPARE), then we allow corrupted spares to be specified, as long
1024  * as they are well-formed.
1025  */
1026 static int
1027 spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1028 {
1029 	nvlist_t **spares;
1030 	uint_t i, nspares;
1031 	vdev_t *vd;
1032 	int error;
1033 
1034 	/*
1035 	 * It's acceptable to have no spares specified.
1036 	 */
1037 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1038 	    &spares, &nspares) != 0)
1039 		return (0);
1040 
1041 	if (nspares == 0)
1042 		return (EINVAL);
1043 
1044 	/*
1045 	 * Make sure the pool is formatted with a version that supports hot
1046 	 * spares.
1047 	 */
1048 	if (spa_version(spa) < ZFS_VERSION_SPARES)
1049 		return (ENOTSUP);
1050 
1051 	/*
1052 	 * Set the pending spare list so we correctly handle device in-use
1053 	 * checking.
1054 	 */
1055 	spa->spa_pending_spares = spares;
1056 	spa->spa_pending_nspares = nspares;
1057 
1058 	for (i = 0; i < nspares; i++) {
1059 		if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0,
1060 		    mode)) != 0)
1061 			goto out;
1062 
1063 		if (!vd->vdev_ops->vdev_op_leaf) {
1064 			vdev_free(vd);
1065 			error = EINVAL;
1066 			goto out;
1067 		}
1068 
1069 		vd->vdev_top = vd;
1070 
1071 		if ((error = vdev_open(vd)) == 0 &&
1072 		    (error = vdev_label_init(vd, crtxg,
1073 		    VDEV_LABEL_SPARE)) == 0) {
1074 			VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID,
1075 			    vd->vdev_guid) == 0);
1076 		}
1077 
1078 		vdev_free(vd);
1079 
1080 		if (error && mode != VDEV_ALLOC_SPARE)
1081 			goto out;
1082 		else
1083 			error = 0;
1084 	}
1085 
1086 out:
1087 	spa->spa_pending_spares = NULL;
1088 	spa->spa_pending_nspares = 0;
1089 	return (error);
1090 }
1091 
1092 /*
1093  * Pool Creation
1094  */
1095 int
1096 spa_create(const char *pool, nvlist_t *nvroot, const char *altroot)
1097 {
1098 	spa_t *spa;
1099 	vdev_t *rvd;
1100 	dsl_pool_t *dp;
1101 	dmu_tx_t *tx;
1102 	int c, error = 0;
1103 	uint64_t txg = TXG_INITIAL;
1104 	nvlist_t **spares;
1105 	uint_t nspares;
1106 
1107 	/*
1108 	 * If this pool already exists, return failure.
1109 	 */
1110 	mutex_enter(&spa_namespace_lock);
1111 	if (spa_lookup(pool) != NULL) {
1112 		mutex_exit(&spa_namespace_lock);
1113 		return (EEXIST);
1114 	}
1115 
1116 	/*
1117 	 * Allocate a new spa_t structure.
1118 	 */
1119 	spa = spa_add(pool, altroot);
1120 	spa_activate(spa);
1121 
1122 	spa->spa_uberblock.ub_txg = txg - 1;
1123 	spa->spa_uberblock.ub_version = ZFS_VERSION;
1124 	spa->spa_ubsync = spa->spa_uberblock;
1125 
1126 	/*
1127 	 * Create the root vdev.
1128 	 */
1129 	spa_config_enter(spa, RW_WRITER, FTAG);
1130 
1131 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1132 
1133 	ASSERT(error != 0 || rvd != NULL);
1134 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1135 
1136 	if (error == 0 && rvd->vdev_children == 0)
1137 		error = EINVAL;
1138 
1139 	if (error == 0 &&
1140 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1141 	    (error = spa_validate_spares(spa, nvroot, txg,
1142 	    VDEV_ALLOC_ADD)) == 0) {
1143 		for (c = 0; c < rvd->vdev_children; c++)
1144 			vdev_init(rvd->vdev_child[c], txg);
1145 		vdev_config_dirty(rvd);
1146 	}
1147 
1148 	spa_config_exit(spa, FTAG);
1149 
1150 	if (error != 0) {
1151 		spa_unload(spa);
1152 		spa_deactivate(spa);
1153 		spa_remove(spa);
1154 		mutex_exit(&spa_namespace_lock);
1155 		return (error);
1156 	}
1157 
1158 	/*
1159 	 * Get the list of spares, if specified.
1160 	 */
1161 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1162 	    &spares, &nspares) == 0) {
1163 		VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME,
1164 		    KM_SLEEP) == 0);
1165 		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1166 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1167 		spa_config_enter(spa, RW_WRITER, FTAG);
1168 		spa_load_spares(spa);
1169 		spa_config_exit(spa, FTAG);
1170 		spa->spa_sync_spares = B_TRUE;
1171 	}
1172 
1173 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
1174 	spa->spa_meta_objset = dp->dp_meta_objset;
1175 
1176 	tx = dmu_tx_create_assigned(dp, txg);
1177 
1178 	/*
1179 	 * Create the pool config object.
1180 	 */
1181 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
1182 	    DMU_OT_PACKED_NVLIST, 1 << 14,
1183 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
1184 
1185 	if (zap_add(spa->spa_meta_objset,
1186 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1187 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
1188 		cmn_err(CE_PANIC, "failed to add pool config");
1189 	}
1190 
1191 	/* Newly created pools are always deflated. */
1192 	spa->spa_deflate = TRUE;
1193 	if (zap_add(spa->spa_meta_objset,
1194 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1195 	    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
1196 		cmn_err(CE_PANIC, "failed to add deflate");
1197 	}
1198 
1199 	/*
1200 	 * Create the deferred-free bplist object.  Turn off compression
1201 	 * because sync-to-convergence takes longer if the blocksize
1202 	 * keeps changing.
1203 	 */
1204 	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
1205 	    1 << 14, tx);
1206 	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
1207 	    ZIO_COMPRESS_OFF, tx);
1208 
1209 	if (zap_add(spa->spa_meta_objset,
1210 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1211 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
1212 		cmn_err(CE_PANIC, "failed to add bplist");
1213 	}
1214 
1215 	/*
1216 	 * Create the pool's history object.
1217 	 */
1218 	spa_history_create_obj(spa, tx);
1219 
1220 	dmu_tx_commit(tx);
1221 
1222 	spa->spa_bootfs = zfs_prop_default_numeric(ZFS_PROP_BOOTFS);
1223 	spa->spa_sync_on = B_TRUE;
1224 	txg_sync_start(spa->spa_dsl_pool);
1225 
1226 	/*
1227 	 * We explicitly wait for the first transaction to complete so that our
1228 	 * bean counters are appropriately updated.
1229 	 */
1230 	txg_wait_synced(spa->spa_dsl_pool, txg);
1231 
1232 	spa_config_sync();
1233 
1234 	mutex_exit(&spa_namespace_lock);
1235 
1236 	return (0);
1237 }
1238 
1239 /*
1240  * Import the given pool into the system.  We set up the necessary spa_t and
1241  * then call spa_load() to do the dirty work.
1242  */
1243 int
1244 spa_import(const char *pool, nvlist_t *config, const char *altroot)
1245 {
1246 	spa_t *spa;
1247 	int error;
1248 	nvlist_t *nvroot;
1249 	nvlist_t **spares;
1250 	uint_t nspares;
1251 
1252 	if (!(spa_mode & FWRITE))
1253 		return (EROFS);
1254 
1255 	/*
1256 	 * If a pool with this name exists, return failure.
1257 	 */
1258 	mutex_enter(&spa_namespace_lock);
1259 	if (spa_lookup(pool) != NULL) {
1260 		mutex_exit(&spa_namespace_lock);
1261 		return (EEXIST);
1262 	}
1263 
1264 	/*
1265 	 * Create and initialize the spa structure.
1266 	 */
1267 	spa = spa_add(pool, altroot);
1268 	spa_activate(spa);
1269 
1270 	/*
1271 	 * Pass off the heavy lifting to spa_load().
1272 	 * Pass TRUE for mosconfig because the user-supplied config
1273 	 * is actually the one to trust when doing an import.
1274 	 */
1275 	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
1276 
1277 	spa_config_enter(spa, RW_WRITER, FTAG);
1278 	/*
1279 	 * Toss any existing sparelist, as it doesn't have any validity anymore,
1280 	 * and conflicts with spa_has_spare().
1281 	 */
1282 	if (spa->spa_sparelist) {
1283 		nvlist_free(spa->spa_sparelist);
1284 		spa->spa_sparelist = NULL;
1285 		spa_load_spares(spa);
1286 	}
1287 
1288 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1289 	    &nvroot) == 0);
1290 	if (error == 0)
1291 		error = spa_validate_spares(spa, nvroot, -1ULL,
1292 		    VDEV_ALLOC_SPARE);
1293 	spa_config_exit(spa, FTAG);
1294 
1295 	if (error != 0) {
1296 		spa_unload(spa);
1297 		spa_deactivate(spa);
1298 		spa_remove(spa);
1299 		mutex_exit(&spa_namespace_lock);
1300 		return (error);
1301 	}
1302 
1303 	/*
1304 	 * Override any spares as specified by the user, as these may have
1305 	 * correct device names/devids, etc.
1306 	 */
1307 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1308 	    &spares, &nspares) == 0) {
1309 		if (spa->spa_sparelist)
1310 			VERIFY(nvlist_remove(spa->spa_sparelist,
1311 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1312 		else
1313 			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1314 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1315 		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1316 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1317 		spa_config_enter(spa, RW_WRITER, FTAG);
1318 		spa_load_spares(spa);
1319 		spa_config_exit(spa, FTAG);
1320 		spa->spa_sync_spares = B_TRUE;
1321 	}
1322 
1323 	/*
1324 	 * Update the config cache to include the newly-imported pool.
1325 	 */
1326 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1327 
1328 	mutex_exit(&spa_namespace_lock);
1329 
1330 	/*
1331 	 * Resilver anything that's out of date.
1332 	 */
1333 	if (spa_mode & FWRITE)
1334 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1335 
1336 	return (0);
1337 }
1338 
1339 /*
1340  * This (illegal) pool name is used when temporarily importing a spa_t in order
1341  * to get the vdev stats associated with the imported devices.
1342  */
1343 #define	TRYIMPORT_NAME	"$import"
1344 
1345 nvlist_t *
1346 spa_tryimport(nvlist_t *tryconfig)
1347 {
1348 	nvlist_t *config = NULL;
1349 	char *poolname;
1350 	spa_t *spa;
1351 	uint64_t state;
1352 
1353 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
1354 		return (NULL);
1355 
1356 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
1357 		return (NULL);
1358 
1359 	/*
1360 	 * Create and initialize the spa structure.
1361 	 */
1362 	mutex_enter(&spa_namespace_lock);
1363 	spa = spa_add(TRYIMPORT_NAME, NULL);
1364 	spa_activate(spa);
1365 
1366 	/*
1367 	 * Pass off the heavy lifting to spa_load().
1368 	 * Pass TRUE for mosconfig because the user-supplied config
1369 	 * is actually the one to trust when doing an import.
1370 	 */
1371 	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
1372 
1373 	/*
1374 	 * If 'tryconfig' was at least parsable, return the current config.
1375 	 */
1376 	if (spa->spa_root_vdev != NULL) {
1377 		spa_config_enter(spa, RW_READER, FTAG);
1378 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1379 		spa_config_exit(spa, FTAG);
1380 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
1381 		    poolname) == 0);
1382 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1383 		    state) == 0);
1384 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
1385 		    spa->spa_uberblock.ub_timestamp) == 0);
1386 
1387 		/*
1388 		 * Add the list of hot spares.
1389 		 */
1390 		spa_add_spares(spa, config);
1391 	}
1392 
1393 	spa_unload(spa);
1394 	spa_deactivate(spa);
1395 	spa_remove(spa);
1396 	mutex_exit(&spa_namespace_lock);
1397 
1398 	return (config);
1399 }
1400 
1401 /*
1402  * Pool export/destroy
1403  *
1404  * The act of destroying or exporting a pool is very simple.  We make sure there
1405  * is no more pending I/O and any references to the pool are gone.  Then, we
1406  * update the pool state and sync all the labels to disk, removing the
1407  * configuration from the cache afterwards.
1408  */
1409 static int
1410 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig)
1411 {
1412 	spa_t *spa;
1413 
1414 	if (oldconfig)
1415 		*oldconfig = NULL;
1416 
1417 	if (!(spa_mode & FWRITE))
1418 		return (EROFS);
1419 
1420 	mutex_enter(&spa_namespace_lock);
1421 	if ((spa = spa_lookup(pool)) == NULL) {
1422 		mutex_exit(&spa_namespace_lock);
1423 		return (ENOENT);
1424 	}
1425 
1426 	/*
1427 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
1428 	 * reacquire the namespace lock, and see if we can export.
1429 	 */
1430 	spa_open_ref(spa, FTAG);
1431 	mutex_exit(&spa_namespace_lock);
1432 	spa_async_suspend(spa);
1433 	mutex_enter(&spa_namespace_lock);
1434 	spa_close(spa, FTAG);
1435 
1436 	/*
1437 	 * The pool will be in core if it's openable,
1438 	 * in which case we can modify its state.
1439 	 */
1440 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
1441 		/*
1442 		 * Objsets may be open only because they're dirty, so we
1443 		 * have to force it to sync before checking spa_refcnt.
1444 		 */
1445 		spa_scrub_suspend(spa);
1446 		txg_wait_synced(spa->spa_dsl_pool, 0);
1447 
1448 		/*
1449 		 * A pool cannot be exported or destroyed if there are active
1450 		 * references.  If we are resetting a pool, allow references by
1451 		 * fault injection handlers.
1452 		 */
1453 		if (!spa_refcount_zero(spa) ||
1454 		    (spa->spa_inject_ref != 0 &&
1455 		    new_state != POOL_STATE_UNINITIALIZED)) {
1456 			spa_scrub_resume(spa);
1457 			spa_async_resume(spa);
1458 			mutex_exit(&spa_namespace_lock);
1459 			return (EBUSY);
1460 		}
1461 
1462 		spa_scrub_resume(spa);
1463 		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
1464 
1465 		/*
1466 		 * We want this to be reflected on every label,
1467 		 * so mark them all dirty.  spa_unload() will do the
1468 		 * final sync that pushes these changes out.
1469 		 */
1470 		if (new_state != POOL_STATE_UNINITIALIZED) {
1471 			spa_config_enter(spa, RW_WRITER, FTAG);
1472 			spa->spa_state = new_state;
1473 			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
1474 			vdev_config_dirty(spa->spa_root_vdev);
1475 			spa_config_exit(spa, FTAG);
1476 		}
1477 	}
1478 
1479 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1480 		spa_unload(spa);
1481 		spa_deactivate(spa);
1482 	}
1483 
1484 	if (oldconfig && spa->spa_config)
1485 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
1486 
1487 	if (new_state != POOL_STATE_UNINITIALIZED) {
1488 		spa_remove(spa);
1489 		spa_config_sync();
1490 	}
1491 	mutex_exit(&spa_namespace_lock);
1492 
1493 	return (0);
1494 }
1495 
1496 /*
1497  * Destroy a storage pool.
1498  */
1499 int
1500 spa_destroy(char *pool)
1501 {
1502 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL));
1503 }
1504 
1505 /*
1506  * Export a storage pool.
1507  */
1508 int
1509 spa_export(char *pool, nvlist_t **oldconfig)
1510 {
1511 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig));
1512 }
1513 
1514 /*
1515  * Similar to spa_export(), this unloads the spa_t without actually removing it
1516  * from the namespace in any way.
1517  */
1518 int
1519 spa_reset(char *pool)
1520 {
1521 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL));
1522 }
1523 
1524 
1525 /*
1526  * ==========================================================================
1527  * Device manipulation
1528  * ==========================================================================
1529  */
1530 
1531 /*
1532  * Add capacity to a storage pool.
1533  */
1534 int
1535 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1536 {
1537 	uint64_t txg;
1538 	int c, error;
1539 	vdev_t *rvd = spa->spa_root_vdev;
1540 	vdev_t *vd, *tvd;
1541 	nvlist_t **spares;
1542 	uint_t i, nspares;
1543 
1544 	txg = spa_vdev_enter(spa);
1545 
1546 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
1547 	    VDEV_ALLOC_ADD)) != 0)
1548 		return (spa_vdev_exit(spa, NULL, txg, error));
1549 
1550 	spa->spa_pending_vdev = vd;
1551 
1552 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1553 	    &spares, &nspares) != 0)
1554 		nspares = 0;
1555 
1556 	if (vd->vdev_children == 0 && nspares == 0) {
1557 		spa->spa_pending_vdev = NULL;
1558 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1559 	}
1560 
1561 	if (vd->vdev_children != 0) {
1562 		if ((error = vdev_create(vd, txg, B_FALSE)) != 0) {
1563 			spa->spa_pending_vdev = NULL;
1564 			return (spa_vdev_exit(spa, vd, txg, error));
1565 		}
1566 	}
1567 
1568 	/*
1569 	 * We must validate the spares after checking the children.  Otherwise,
1570 	 * vdev_inuse() will blindly overwrite the spare.
1571 	 */
1572 	if ((error = spa_validate_spares(spa, nvroot, txg,
1573 	    VDEV_ALLOC_ADD)) != 0) {
1574 		spa->spa_pending_vdev = NULL;
1575 		return (spa_vdev_exit(spa, vd, txg, error));
1576 	}
1577 
1578 	spa->spa_pending_vdev = NULL;
1579 
1580 	/*
1581 	 * Transfer each new top-level vdev from vd to rvd.
1582 	 */
1583 	for (c = 0; c < vd->vdev_children; c++) {
1584 		tvd = vd->vdev_child[c];
1585 		vdev_remove_child(vd, tvd);
1586 		tvd->vdev_id = rvd->vdev_children;
1587 		vdev_add_child(rvd, tvd);
1588 		vdev_config_dirty(tvd);
1589 	}
1590 
1591 	if (nspares != 0) {
1592 		if (spa->spa_sparelist != NULL) {
1593 			nvlist_t **oldspares;
1594 			uint_t oldnspares;
1595 			nvlist_t **newspares;
1596 
1597 			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
1598 			    ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0);
1599 
1600 			newspares = kmem_alloc(sizeof (void *) *
1601 			    (nspares + oldnspares), KM_SLEEP);
1602 			for (i = 0; i < oldnspares; i++)
1603 				VERIFY(nvlist_dup(oldspares[i],
1604 				    &newspares[i], KM_SLEEP) == 0);
1605 			for (i = 0; i < nspares; i++)
1606 				VERIFY(nvlist_dup(spares[i],
1607 				    &newspares[i + oldnspares],
1608 				    KM_SLEEP) == 0);
1609 
1610 			VERIFY(nvlist_remove(spa->spa_sparelist,
1611 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1612 
1613 			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1614 			    ZPOOL_CONFIG_SPARES, newspares,
1615 			    nspares + oldnspares) == 0);
1616 			for (i = 0; i < oldnspares + nspares; i++)
1617 				nvlist_free(newspares[i]);
1618 			kmem_free(newspares, (oldnspares + nspares) *
1619 			    sizeof (void *));
1620 		} else {
1621 			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1622 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1623 			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1624 			    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1625 		}
1626 
1627 		spa_load_spares(spa);
1628 		spa->spa_sync_spares = B_TRUE;
1629 	}
1630 
1631 	/*
1632 	 * We have to be careful when adding new vdevs to an existing pool.
1633 	 * If other threads start allocating from these vdevs before we
1634 	 * sync the config cache, and we lose power, then upon reboot we may
1635 	 * fail to open the pool because there are DVAs that the config cache
1636 	 * can't translate.  Therefore, we first add the vdevs without
1637 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
1638 	 * and then let spa_config_update() initialize the new metaslabs.
1639 	 *
1640 	 * spa_load() checks for added-but-not-initialized vdevs, so that
1641 	 * if we lose power at any point in this sequence, the remaining
1642 	 * steps will be completed the next time we load the pool.
1643 	 */
1644 	(void) spa_vdev_exit(spa, vd, txg, 0);
1645 
1646 	mutex_enter(&spa_namespace_lock);
1647 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1648 	mutex_exit(&spa_namespace_lock);
1649 
1650 	return (0);
1651 }
1652 
1653 /*
1654  * Attach a device to a mirror.  The arguments are the path to any device
1655  * in the mirror, and the nvroot for the new device.  If the path specifies
1656  * a device that is not mirrored, we automatically insert the mirror vdev.
1657  *
1658  * If 'replacing' is specified, the new device is intended to replace the
1659  * existing device; in this case the two devices are made into their own
1660  * mirror using the 'replacing' vdev, which is functionally idendical to
1661  * the mirror vdev (it actually reuses all the same ops) but has a few
1662  * extra rules: you can't attach to it after it's been created, and upon
1663  * completion of resilvering, the first disk (the one being replaced)
1664  * is automatically detached.
1665  */
1666 int
1667 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
1668 {
1669 	uint64_t txg, open_txg;
1670 	int error;
1671 	vdev_t *rvd = spa->spa_root_vdev;
1672 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
1673 	vdev_ops_t *pvops;
1674 
1675 	txg = spa_vdev_enter(spa);
1676 
1677 	oldvd = vdev_lookup_by_guid(rvd, guid);
1678 
1679 	if (oldvd == NULL)
1680 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1681 
1682 	if (!oldvd->vdev_ops->vdev_op_leaf)
1683 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1684 
1685 	pvd = oldvd->vdev_parent;
1686 
1687 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
1688 	    VDEV_ALLOC_ADD)) != 0 || newrootvd->vdev_children != 1)
1689 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1690 
1691 	newvd = newrootvd->vdev_child[0];
1692 
1693 	if (!newvd->vdev_ops->vdev_op_leaf)
1694 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1695 
1696 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
1697 		return (spa_vdev_exit(spa, newrootvd, txg, error));
1698 
1699 	if (!replacing) {
1700 		/*
1701 		 * For attach, the only allowable parent is a mirror or the root
1702 		 * vdev.
1703 		 */
1704 		if (pvd->vdev_ops != &vdev_mirror_ops &&
1705 		    pvd->vdev_ops != &vdev_root_ops)
1706 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1707 
1708 		pvops = &vdev_mirror_ops;
1709 	} else {
1710 		/*
1711 		 * Active hot spares can only be replaced by inactive hot
1712 		 * spares.
1713 		 */
1714 		if (pvd->vdev_ops == &vdev_spare_ops &&
1715 		    pvd->vdev_child[1] == oldvd &&
1716 		    !spa_has_spare(spa, newvd->vdev_guid))
1717 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1718 
1719 		/*
1720 		 * If the source is a hot spare, and the parent isn't already a
1721 		 * spare, then we want to create a new hot spare.  Otherwise, we
1722 		 * want to create a replacing vdev.  The user is not allowed to
1723 		 * attach to a spared vdev child unless the 'isspare' state is
1724 		 * the same (spare replaces spare, non-spare replaces
1725 		 * non-spare).
1726 		 */
1727 		if (pvd->vdev_ops == &vdev_replacing_ops)
1728 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1729 		else if (pvd->vdev_ops == &vdev_spare_ops &&
1730 		    newvd->vdev_isspare != oldvd->vdev_isspare)
1731 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1732 		else if (pvd->vdev_ops != &vdev_spare_ops &&
1733 		    newvd->vdev_isspare)
1734 			pvops = &vdev_spare_ops;
1735 		else
1736 			pvops = &vdev_replacing_ops;
1737 	}
1738 
1739 	/*
1740 	 * Compare the new device size with the replaceable/attachable
1741 	 * device size.
1742 	 */
1743 	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
1744 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
1745 
1746 	/*
1747 	 * The new device cannot have a higher alignment requirement
1748 	 * than the top-level vdev.
1749 	 */
1750 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
1751 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
1752 
1753 	/*
1754 	 * If this is an in-place replacement, update oldvd's path and devid
1755 	 * to make it distinguishable from newvd, and unopenable from now on.
1756 	 */
1757 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
1758 		spa_strfree(oldvd->vdev_path);
1759 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
1760 		    KM_SLEEP);
1761 		(void) sprintf(oldvd->vdev_path, "%s/%s",
1762 		    newvd->vdev_path, "old");
1763 		if (oldvd->vdev_devid != NULL) {
1764 			spa_strfree(oldvd->vdev_devid);
1765 			oldvd->vdev_devid = NULL;
1766 		}
1767 	}
1768 
1769 	/*
1770 	 * If the parent is not a mirror, or if we're replacing, insert the new
1771 	 * mirror/replacing/spare vdev above oldvd.
1772 	 */
1773 	if (pvd->vdev_ops != pvops)
1774 		pvd = vdev_add_parent(oldvd, pvops);
1775 
1776 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
1777 	ASSERT(pvd->vdev_ops == pvops);
1778 	ASSERT(oldvd->vdev_parent == pvd);
1779 
1780 	/*
1781 	 * Extract the new device from its root and add it to pvd.
1782 	 */
1783 	vdev_remove_child(newrootvd, newvd);
1784 	newvd->vdev_id = pvd->vdev_children;
1785 	vdev_add_child(pvd, newvd);
1786 
1787 	/*
1788 	 * If newvd is smaller than oldvd, but larger than its rsize,
1789 	 * the addition of newvd may have decreased our parent's asize.
1790 	 */
1791 	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
1792 
1793 	tvd = newvd->vdev_top;
1794 	ASSERT(pvd->vdev_top == tvd);
1795 	ASSERT(tvd->vdev_parent == rvd);
1796 
1797 	vdev_config_dirty(tvd);
1798 
1799 	/*
1800 	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
1801 	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
1802 	 */
1803 	open_txg = txg + TXG_CONCURRENT_STATES - 1;
1804 
1805 	mutex_enter(&newvd->vdev_dtl_lock);
1806 	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
1807 	    open_txg - TXG_INITIAL + 1);
1808 	mutex_exit(&newvd->vdev_dtl_lock);
1809 
1810 	if (newvd->vdev_isspare)
1811 		spa_spare_activate(newvd);
1812 
1813 	/*
1814 	 * Mark newvd's DTL dirty in this txg.
1815 	 */
1816 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
1817 
1818 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
1819 
1820 	/*
1821 	 * Kick off a resilver to update newvd.
1822 	 */
1823 	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1824 
1825 	return (0);
1826 }
1827 
1828 /*
1829  * Detach a device from a mirror or replacing vdev.
1830  * If 'replace_done' is specified, only detach if the parent
1831  * is a replacing vdev.
1832  */
1833 int
1834 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
1835 {
1836 	uint64_t txg;
1837 	int c, t, error;
1838 	vdev_t *rvd = spa->spa_root_vdev;
1839 	vdev_t *vd, *pvd, *cvd, *tvd;
1840 	boolean_t unspare = B_FALSE;
1841 	uint64_t unspare_guid;
1842 
1843 	txg = spa_vdev_enter(spa);
1844 
1845 	vd = vdev_lookup_by_guid(rvd, guid);
1846 
1847 	if (vd == NULL)
1848 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1849 
1850 	if (!vd->vdev_ops->vdev_op_leaf)
1851 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1852 
1853 	pvd = vd->vdev_parent;
1854 
1855 	/*
1856 	 * If replace_done is specified, only remove this device if it's
1857 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
1858 	 * disk can be removed.
1859 	 */
1860 	if (replace_done) {
1861 		if (pvd->vdev_ops == &vdev_replacing_ops) {
1862 			if (vd->vdev_id != 0)
1863 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1864 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
1865 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1866 		}
1867 	}
1868 
1869 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
1870 	    spa_version(spa) >= ZFS_VERSION_SPARES);
1871 
1872 	/*
1873 	 * Only mirror, replacing, and spare vdevs support detach.
1874 	 */
1875 	if (pvd->vdev_ops != &vdev_replacing_ops &&
1876 	    pvd->vdev_ops != &vdev_mirror_ops &&
1877 	    pvd->vdev_ops != &vdev_spare_ops)
1878 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1879 
1880 	/*
1881 	 * If there's only one replica, you can't detach it.
1882 	 */
1883 	if (pvd->vdev_children <= 1)
1884 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1885 
1886 	/*
1887 	 * If all siblings have non-empty DTLs, this device may have the only
1888 	 * valid copy of the data, which means we cannot safely detach it.
1889 	 *
1890 	 * XXX -- as in the vdev_offline() case, we really want a more
1891 	 * precise DTL check.
1892 	 */
1893 	for (c = 0; c < pvd->vdev_children; c++) {
1894 		uint64_t dirty;
1895 
1896 		cvd = pvd->vdev_child[c];
1897 		if (cvd == vd)
1898 			continue;
1899 		if (vdev_is_dead(cvd))
1900 			continue;
1901 		mutex_enter(&cvd->vdev_dtl_lock);
1902 		dirty = cvd->vdev_dtl_map.sm_space |
1903 		    cvd->vdev_dtl_scrub.sm_space;
1904 		mutex_exit(&cvd->vdev_dtl_lock);
1905 		if (!dirty)
1906 			break;
1907 	}
1908 
1909 	/*
1910 	 * If we are a replacing or spare vdev, then we can always detach the
1911 	 * latter child, as that is how one cancels the operation.
1912 	 */
1913 	if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) &&
1914 	    c == pvd->vdev_children)
1915 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1916 
1917 	/*
1918 	 * If we are detaching the original disk from a spare, then it implies
1919 	 * that the spare should become a real disk, and be removed from the
1920 	 * active spare list for the pool.
1921 	 */
1922 	if (pvd->vdev_ops == &vdev_spare_ops &&
1923 	    vd->vdev_id == 0)
1924 		unspare = B_TRUE;
1925 
1926 	/*
1927 	 * Erase the disk labels so the disk can be used for other things.
1928 	 * This must be done after all other error cases are handled,
1929 	 * but before we disembowel vd (so we can still do I/O to it).
1930 	 * But if we can't do it, don't treat the error as fatal --
1931 	 * it may be that the unwritability of the disk is the reason
1932 	 * it's being detached!
1933 	 */
1934 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
1935 
1936 	/*
1937 	 * Remove vd from its parent and compact the parent's children.
1938 	 */
1939 	vdev_remove_child(pvd, vd);
1940 	vdev_compact_children(pvd);
1941 
1942 	/*
1943 	 * Remember one of the remaining children so we can get tvd below.
1944 	 */
1945 	cvd = pvd->vdev_child[0];
1946 
1947 	/*
1948 	 * If we need to remove the remaining child from the list of hot spares,
1949 	 * do it now, marking the vdev as no longer a spare in the process.  We
1950 	 * must do this before vdev_remove_parent(), because that can change the
1951 	 * GUID if it creates a new toplevel GUID.
1952 	 */
1953 	if (unspare) {
1954 		ASSERT(cvd->vdev_isspare);
1955 		spa_spare_remove(cvd);
1956 		unspare_guid = cvd->vdev_guid;
1957 	}
1958 
1959 	/*
1960 	 * If the parent mirror/replacing vdev only has one child,
1961 	 * the parent is no longer needed.  Remove it from the tree.
1962 	 */
1963 	if (pvd->vdev_children == 1)
1964 		vdev_remove_parent(cvd);
1965 
1966 	/*
1967 	 * We don't set tvd until now because the parent we just removed
1968 	 * may have been the previous top-level vdev.
1969 	 */
1970 	tvd = cvd->vdev_top;
1971 	ASSERT(tvd->vdev_parent == rvd);
1972 
1973 	/*
1974 	 * Reevaluate the parent vdev state.
1975 	 */
1976 	vdev_propagate_state(cvd->vdev_parent);
1977 
1978 	/*
1979 	 * If the device we just detached was smaller than the others, it may be
1980 	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
1981 	 * can't fail because the existing metaslabs are already in core, so
1982 	 * there's nothing to read from disk.
1983 	 */
1984 	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
1985 
1986 	vdev_config_dirty(tvd);
1987 
1988 	/*
1989 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
1990 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
1991 	 * But first make sure we're not on any *other* txg's DTL list, to
1992 	 * prevent vd from being accessed after it's freed.
1993 	 */
1994 	for (t = 0; t < TXG_SIZE; t++)
1995 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
1996 	vd->vdev_detached = B_TRUE;
1997 	vdev_dirty(tvd, VDD_DTL, vd, txg);
1998 
1999 	error = spa_vdev_exit(spa, vd, txg, 0);
2000 
2001 	/*
2002 	 * If this was the removal of the original device in a hot spare vdev,
2003 	 * then we want to go through and remove the device from the hot spare
2004 	 * list of every other pool.
2005 	 */
2006 	if (unspare) {
2007 		spa = NULL;
2008 		mutex_enter(&spa_namespace_lock);
2009 		while ((spa = spa_next(spa)) != NULL) {
2010 			if (spa->spa_state != POOL_STATE_ACTIVE)
2011 				continue;
2012 
2013 			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
2014 		}
2015 		mutex_exit(&spa_namespace_lock);
2016 	}
2017 
2018 	return (error);
2019 }
2020 
2021 /*
2022  * Remove a device from the pool.  Currently, this supports removing only hot
2023  * spares.
2024  */
2025 int
2026 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
2027 {
2028 	vdev_t *vd;
2029 	nvlist_t **spares, *nv, **newspares;
2030 	uint_t i, j, nspares;
2031 	int ret = 0;
2032 
2033 	spa_config_enter(spa, RW_WRITER, FTAG);
2034 
2035 	vd = spa_lookup_by_guid(spa, guid);
2036 
2037 	nv = NULL;
2038 	if (spa->spa_spares != NULL &&
2039 	    nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2040 	    &spares, &nspares) == 0) {
2041 		for (i = 0; i < nspares; i++) {
2042 			uint64_t theguid;
2043 
2044 			VERIFY(nvlist_lookup_uint64(spares[i],
2045 			    ZPOOL_CONFIG_GUID, &theguid) == 0);
2046 			if (theguid == guid) {
2047 				nv = spares[i];
2048 				break;
2049 			}
2050 		}
2051 	}
2052 
2053 	/*
2054 	 * We only support removing a hot spare, and only if it's not currently
2055 	 * in use in this pool.
2056 	 */
2057 	if (nv == NULL && vd == NULL) {
2058 		ret = ENOENT;
2059 		goto out;
2060 	}
2061 
2062 	if (nv == NULL && vd != NULL) {
2063 		ret = ENOTSUP;
2064 		goto out;
2065 	}
2066 
2067 	if (!unspare && nv != NULL && vd != NULL) {
2068 		ret = EBUSY;
2069 		goto out;
2070 	}
2071 
2072 	if (nspares == 1) {
2073 		newspares = NULL;
2074 	} else {
2075 		newspares = kmem_alloc((nspares - 1) * sizeof (void *),
2076 		    KM_SLEEP);
2077 		for (i = 0, j = 0; i < nspares; i++) {
2078 			if (spares[i] != nv)
2079 				VERIFY(nvlist_dup(spares[i],
2080 				    &newspares[j++], KM_SLEEP) == 0);
2081 		}
2082 	}
2083 
2084 	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2085 	    DATA_TYPE_NVLIST_ARRAY) == 0);
2086 	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2087 	    newspares, nspares - 1) == 0);
2088 	for (i = 0; i < nspares - 1; i++)
2089 		nvlist_free(newspares[i]);
2090 	kmem_free(newspares, (nspares - 1) * sizeof (void *));
2091 	spa_load_spares(spa);
2092 	spa->spa_sync_spares = B_TRUE;
2093 
2094 out:
2095 	spa_config_exit(spa, FTAG);
2096 
2097 	return (ret);
2098 }
2099 
2100 /*
2101  * Find any device that's done replacing, so we can detach it.
2102  */
2103 static vdev_t *
2104 spa_vdev_replace_done_hunt(vdev_t *vd)
2105 {
2106 	vdev_t *newvd, *oldvd;
2107 	int c;
2108 
2109 	for (c = 0; c < vd->vdev_children; c++) {
2110 		oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]);
2111 		if (oldvd != NULL)
2112 			return (oldvd);
2113 	}
2114 
2115 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
2116 		oldvd = vd->vdev_child[0];
2117 		newvd = vd->vdev_child[1];
2118 
2119 		mutex_enter(&newvd->vdev_dtl_lock);
2120 		if (newvd->vdev_dtl_map.sm_space == 0 &&
2121 		    newvd->vdev_dtl_scrub.sm_space == 0) {
2122 			mutex_exit(&newvd->vdev_dtl_lock);
2123 			return (oldvd);
2124 		}
2125 		mutex_exit(&newvd->vdev_dtl_lock);
2126 	}
2127 
2128 	return (NULL);
2129 }
2130 
2131 static void
2132 spa_vdev_replace_done(spa_t *spa)
2133 {
2134 	vdev_t *vd;
2135 	vdev_t *pvd;
2136 	uint64_t guid;
2137 	uint64_t pguid = 0;
2138 
2139 	spa_config_enter(spa, RW_READER, FTAG);
2140 
2141 	while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) {
2142 		guid = vd->vdev_guid;
2143 		/*
2144 		 * If we have just finished replacing a hot spared device, then
2145 		 * we need to detach the parent's first child (the original hot
2146 		 * spare) as well.
2147 		 */
2148 		pvd = vd->vdev_parent;
2149 		if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2150 		    pvd->vdev_id == 0) {
2151 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
2152 			ASSERT(pvd->vdev_parent->vdev_children == 2);
2153 			pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
2154 		}
2155 		spa_config_exit(spa, FTAG);
2156 		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
2157 			return;
2158 		if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
2159 			return;
2160 		spa_config_enter(spa, RW_READER, FTAG);
2161 	}
2162 
2163 	spa_config_exit(spa, FTAG);
2164 }
2165 
2166 /*
2167  * Update the stored path for this vdev.  Dirty the vdev configuration, relying
2168  * on spa_vdev_enter/exit() to synchronize the labels and cache.
2169  */
2170 int
2171 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
2172 {
2173 	vdev_t *rvd, *vd;
2174 	uint64_t txg;
2175 
2176 	rvd = spa->spa_root_vdev;
2177 
2178 	txg = spa_vdev_enter(spa);
2179 
2180 	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2181 		/*
2182 		 * Determine if this is a reference to a hot spare.  In that
2183 		 * case, update the path as stored in the spare list.
2184 		 */
2185 		nvlist_t **spares;
2186 		uint_t i, nspares;
2187 		if (spa->spa_sparelist != NULL) {
2188 			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
2189 			    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2190 			for (i = 0; i < nspares; i++) {
2191 				uint64_t theguid;
2192 				VERIFY(nvlist_lookup_uint64(spares[i],
2193 				    ZPOOL_CONFIG_GUID, &theguid) == 0);
2194 				if (theguid == guid)
2195 					break;
2196 			}
2197 
2198 			if (i == nspares)
2199 				return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2200 
2201 			VERIFY(nvlist_add_string(spares[i],
2202 			    ZPOOL_CONFIG_PATH, newpath) == 0);
2203 			spa_load_spares(spa);
2204 			spa->spa_sync_spares = B_TRUE;
2205 			return (spa_vdev_exit(spa, NULL, txg, 0));
2206 		} else {
2207 			return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2208 		}
2209 	}
2210 
2211 	if (!vd->vdev_ops->vdev_op_leaf)
2212 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2213 
2214 	spa_strfree(vd->vdev_path);
2215 	vd->vdev_path = spa_strdup(newpath);
2216 
2217 	vdev_config_dirty(vd->vdev_top);
2218 
2219 	return (spa_vdev_exit(spa, NULL, txg, 0));
2220 }
2221 
2222 /*
2223  * ==========================================================================
2224  * SPA Scrubbing
2225  * ==========================================================================
2226  */
2227 
2228 static void
2229 spa_scrub_io_done(zio_t *zio)
2230 {
2231 	spa_t *spa = zio->io_spa;
2232 
2233 	arc_data_buf_free(zio->io_data, zio->io_size);
2234 
2235 	mutex_enter(&spa->spa_scrub_lock);
2236 	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
2237 		vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev;
2238 		spa->spa_scrub_errors++;
2239 		mutex_enter(&vd->vdev_stat_lock);
2240 		vd->vdev_stat.vs_scrub_errors++;
2241 		mutex_exit(&vd->vdev_stat_lock);
2242 	}
2243 
2244 	if (--spa->spa_scrub_inflight < spa->spa_scrub_maxinflight)
2245 		cv_broadcast(&spa->spa_scrub_io_cv);
2246 
2247 	ASSERT(spa->spa_scrub_inflight >= 0);
2248 
2249 	mutex_exit(&spa->spa_scrub_lock);
2250 }
2251 
2252 static void
2253 spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
2254     zbookmark_t *zb)
2255 {
2256 	size_t size = BP_GET_LSIZE(bp);
2257 	void *data;
2258 
2259 	mutex_enter(&spa->spa_scrub_lock);
2260 	/*
2261 	 * Do not give too much work to vdev(s).
2262 	 */
2263 	while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight) {
2264 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2265 	}
2266 	spa->spa_scrub_inflight++;
2267 	mutex_exit(&spa->spa_scrub_lock);
2268 
2269 	data = arc_data_buf_alloc(size);
2270 
2271 	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
2272 		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
2273 
2274 	flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL;
2275 
2276 	zio_nowait(zio_read(NULL, spa, bp, data, size,
2277 	    spa_scrub_io_done, NULL, priority, flags, zb));
2278 }
2279 
2280 /* ARGSUSED */
2281 static int
2282 spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
2283 {
2284 	blkptr_t *bp = &bc->bc_blkptr;
2285 	vdev_t *vd = spa->spa_root_vdev;
2286 	dva_t *dva = bp->blk_dva;
2287 	int needs_resilver = B_FALSE;
2288 	int d;
2289 
2290 	if (bc->bc_errno) {
2291 		/*
2292 		 * We can't scrub this block, but we can continue to scrub
2293 		 * the rest of the pool.  Note the error and move along.
2294 		 */
2295 		mutex_enter(&spa->spa_scrub_lock);
2296 		spa->spa_scrub_errors++;
2297 		mutex_exit(&spa->spa_scrub_lock);
2298 
2299 		mutex_enter(&vd->vdev_stat_lock);
2300 		vd->vdev_stat.vs_scrub_errors++;
2301 		mutex_exit(&vd->vdev_stat_lock);
2302 
2303 		return (ERESTART);
2304 	}
2305 
2306 	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
2307 
2308 	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
2309 		vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]));
2310 
2311 		ASSERT(vd != NULL);
2312 
2313 		/*
2314 		 * Keep track of how much data we've examined so that
2315 		 * zpool(1M) status can make useful progress reports.
2316 		 */
2317 		mutex_enter(&vd->vdev_stat_lock);
2318 		vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]);
2319 		mutex_exit(&vd->vdev_stat_lock);
2320 
2321 		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
2322 			if (DVA_GET_GANG(&dva[d])) {
2323 				/*
2324 				 * Gang members may be spread across multiple
2325 				 * vdevs, so the best we can do is look at the
2326 				 * pool-wide DTL.
2327 				 * XXX -- it would be better to change our
2328 				 * allocation policy to ensure that this can't
2329 				 * happen.
2330 				 */
2331 				vd = spa->spa_root_vdev;
2332 			}
2333 			if (vdev_dtl_contains(&vd->vdev_dtl_map,
2334 			    bp->blk_birth, 1))
2335 				needs_resilver = B_TRUE;
2336 		}
2337 	}
2338 
2339 	if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING)
2340 		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
2341 		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
2342 	else if (needs_resilver)
2343 		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
2344 		    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
2345 
2346 	return (0);
2347 }
2348 
2349 static void
2350 spa_scrub_thread(spa_t *spa)
2351 {
2352 	callb_cpr_t cprinfo;
2353 	traverse_handle_t *th = spa->spa_scrub_th;
2354 	vdev_t *rvd = spa->spa_root_vdev;
2355 	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
2356 	int error = 0;
2357 	boolean_t complete;
2358 
2359 	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
2360 
2361 	/*
2362 	 * If we're restarting due to a snapshot create/delete,
2363 	 * wait for that to complete.
2364 	 */
2365 	txg_wait_synced(spa_get_dsl(spa), 0);
2366 
2367 	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
2368 	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2369 	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
2370 
2371 	spa_config_enter(spa, RW_WRITER, FTAG);
2372 	vdev_reopen(rvd);		/* purge all vdev caches */
2373 	vdev_config_dirty(rvd);		/* rewrite all disk labels */
2374 	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
2375 	spa_config_exit(spa, FTAG);
2376 
2377 	mutex_enter(&spa->spa_scrub_lock);
2378 	spa->spa_scrub_errors = 0;
2379 	spa->spa_scrub_active = 1;
2380 	ASSERT(spa->spa_scrub_inflight == 0);
2381 
2382 	while (!spa->spa_scrub_stop) {
2383 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2384 		while (spa->spa_scrub_suspended) {
2385 			spa->spa_scrub_active = 0;
2386 			cv_broadcast(&spa->spa_scrub_cv);
2387 			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2388 			spa->spa_scrub_active = 1;
2389 		}
2390 		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
2391 
2392 		if (spa->spa_scrub_restart_txg != 0)
2393 			break;
2394 
2395 		mutex_exit(&spa->spa_scrub_lock);
2396 		error = traverse_more(th);
2397 		mutex_enter(&spa->spa_scrub_lock);
2398 		if (error != EAGAIN)
2399 			break;
2400 	}
2401 
2402 	while (spa->spa_scrub_inflight)
2403 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2404 
2405 	spa->spa_scrub_active = 0;
2406 	cv_broadcast(&spa->spa_scrub_cv);
2407 
2408 	mutex_exit(&spa->spa_scrub_lock);
2409 
2410 	spa_config_enter(spa, RW_WRITER, FTAG);
2411 
2412 	mutex_enter(&spa->spa_scrub_lock);
2413 
2414 	/*
2415 	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
2416 	 * AND the spa config lock to synchronize with any config changes
2417 	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
2418 	 */
2419 	if (spa->spa_scrub_restart_txg != 0)
2420 		error = ERESTART;
2421 
2422 	if (spa->spa_scrub_stop)
2423 		error = EINTR;
2424 
2425 	/*
2426 	 * Even if there were uncorrectable errors, we consider the scrub
2427 	 * completed.  The downside is that if there is a transient error during
2428 	 * a resilver, we won't resilver the data properly to the target.  But
2429 	 * if the damage is permanent (more likely) we will resilver forever,
2430 	 * which isn't really acceptable.  Since there is enough information for
2431 	 * the user to know what has failed and why, this seems like a more
2432 	 * tractable approach.
2433 	 */
2434 	complete = (error == 0);
2435 
2436 	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
2437 	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2438 	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
2439 	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
2440 
2441 	mutex_exit(&spa->spa_scrub_lock);
2442 
2443 	/*
2444 	 * If the scrub/resilver completed, update all DTLs to reflect this.
2445 	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
2446 	 */
2447 	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
2448 	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
2449 	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
2450 	spa_errlog_rotate(spa);
2451 
2452 	spa_config_exit(spa, FTAG);
2453 
2454 	mutex_enter(&spa->spa_scrub_lock);
2455 
2456 	/*
2457 	 * We may have finished replacing a device.
2458 	 * Let the async thread assess this and handle the detach.
2459 	 */
2460 	spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
2461 
2462 	/*
2463 	 * If we were told to restart, our final act is to start a new scrub.
2464 	 */
2465 	if (error == ERESTART)
2466 		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
2467 		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
2468 
2469 	spa->spa_scrub_type = POOL_SCRUB_NONE;
2470 	spa->spa_scrub_active = 0;
2471 	spa->spa_scrub_thread = NULL;
2472 	cv_broadcast(&spa->spa_scrub_cv);
2473 	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
2474 	thread_exit();
2475 }
2476 
2477 void
2478 spa_scrub_suspend(spa_t *spa)
2479 {
2480 	mutex_enter(&spa->spa_scrub_lock);
2481 	spa->spa_scrub_suspended++;
2482 	while (spa->spa_scrub_active) {
2483 		cv_broadcast(&spa->spa_scrub_cv);
2484 		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2485 	}
2486 	while (spa->spa_scrub_inflight)
2487 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2488 	mutex_exit(&spa->spa_scrub_lock);
2489 }
2490 
2491 void
2492 spa_scrub_resume(spa_t *spa)
2493 {
2494 	mutex_enter(&spa->spa_scrub_lock);
2495 	ASSERT(spa->spa_scrub_suspended != 0);
2496 	if (--spa->spa_scrub_suspended == 0)
2497 		cv_broadcast(&spa->spa_scrub_cv);
2498 	mutex_exit(&spa->spa_scrub_lock);
2499 }
2500 
2501 void
2502 spa_scrub_restart(spa_t *spa, uint64_t txg)
2503 {
2504 	/*
2505 	 * Something happened (e.g. snapshot create/delete) that means
2506 	 * we must restart any in-progress scrubs.  The itinerary will
2507 	 * fix this properly.
2508 	 */
2509 	mutex_enter(&spa->spa_scrub_lock);
2510 	spa->spa_scrub_restart_txg = txg;
2511 	mutex_exit(&spa->spa_scrub_lock);
2512 }
2513 
2514 int
2515 spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
2516 {
2517 	space_seg_t *ss;
2518 	uint64_t mintxg, maxtxg;
2519 	vdev_t *rvd = spa->spa_root_vdev;
2520 
2521 	if ((uint_t)type >= POOL_SCRUB_TYPES)
2522 		return (ENOTSUP);
2523 
2524 	mutex_enter(&spa->spa_scrub_lock);
2525 
2526 	/*
2527 	 * If there's a scrub or resilver already in progress, stop it.
2528 	 */
2529 	while (spa->spa_scrub_thread != NULL) {
2530 		/*
2531 		 * Don't stop a resilver unless forced.
2532 		 */
2533 		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
2534 			mutex_exit(&spa->spa_scrub_lock);
2535 			return (EBUSY);
2536 		}
2537 		spa->spa_scrub_stop = 1;
2538 		cv_broadcast(&spa->spa_scrub_cv);
2539 		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2540 	}
2541 
2542 	/*
2543 	 * Terminate the previous traverse.
2544 	 */
2545 	if (spa->spa_scrub_th != NULL) {
2546 		traverse_fini(spa->spa_scrub_th);
2547 		spa->spa_scrub_th = NULL;
2548 	}
2549 
2550 	if (rvd == NULL) {
2551 		ASSERT(spa->spa_scrub_stop == 0);
2552 		ASSERT(spa->spa_scrub_type == type);
2553 		ASSERT(spa->spa_scrub_restart_txg == 0);
2554 		mutex_exit(&spa->spa_scrub_lock);
2555 		return (0);
2556 	}
2557 
2558 	mintxg = TXG_INITIAL - 1;
2559 	maxtxg = spa_last_synced_txg(spa) + 1;
2560 
2561 	mutex_enter(&rvd->vdev_dtl_lock);
2562 
2563 	if (rvd->vdev_dtl_map.sm_space == 0) {
2564 		/*
2565 		 * The pool-wide DTL is empty.
2566 		 * If this is a resilver, there's nothing to do except
2567 		 * check whether any in-progress replacements have completed.
2568 		 */
2569 		if (type == POOL_SCRUB_RESILVER) {
2570 			type = POOL_SCRUB_NONE;
2571 			spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
2572 		}
2573 	} else {
2574 		/*
2575 		 * The pool-wide DTL is non-empty.
2576 		 * If this is a normal scrub, upgrade to a resilver instead.
2577 		 */
2578 		if (type == POOL_SCRUB_EVERYTHING)
2579 			type = POOL_SCRUB_RESILVER;
2580 	}
2581 
2582 	if (type == POOL_SCRUB_RESILVER) {
2583 		/*
2584 		 * Determine the resilvering boundaries.
2585 		 *
2586 		 * Note: (mintxg, maxtxg) is an open interval,
2587 		 * i.e. mintxg and maxtxg themselves are not included.
2588 		 *
2589 		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
2590 		 * so we don't claim to resilver a txg that's still changing.
2591 		 */
2592 		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
2593 		mintxg = ss->ss_start - 1;
2594 		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
2595 		maxtxg = MIN(ss->ss_end, maxtxg);
2596 	}
2597 
2598 	mutex_exit(&rvd->vdev_dtl_lock);
2599 
2600 	spa->spa_scrub_stop = 0;
2601 	spa->spa_scrub_type = type;
2602 	spa->spa_scrub_restart_txg = 0;
2603 
2604 	if (type != POOL_SCRUB_NONE) {
2605 		spa->spa_scrub_mintxg = mintxg;
2606 		spa->spa_scrub_maxtxg = maxtxg;
2607 		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
2608 		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
2609 		    ZIO_FLAG_CANFAIL);
2610 		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
2611 		spa->spa_scrub_thread = thread_create(NULL, 0,
2612 		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
2613 	}
2614 
2615 	mutex_exit(&spa->spa_scrub_lock);
2616 
2617 	return (0);
2618 }
2619 
2620 /*
2621  * ==========================================================================
2622  * SPA async task processing
2623  * ==========================================================================
2624  */
2625 
2626 static void
2627 spa_async_reopen(spa_t *spa)
2628 {
2629 	vdev_t *rvd = spa->spa_root_vdev;
2630 	vdev_t *tvd;
2631 	int c;
2632 
2633 	spa_config_enter(spa, RW_WRITER, FTAG);
2634 
2635 	for (c = 0; c < rvd->vdev_children; c++) {
2636 		tvd = rvd->vdev_child[c];
2637 		if (tvd->vdev_reopen_wanted) {
2638 			tvd->vdev_reopen_wanted = 0;
2639 			vdev_reopen(tvd);
2640 		}
2641 	}
2642 
2643 	spa_config_exit(spa, FTAG);
2644 }
2645 
2646 static void
2647 spa_async_thread(spa_t *spa)
2648 {
2649 	int tasks;
2650 
2651 	ASSERT(spa->spa_sync_on);
2652 
2653 	mutex_enter(&spa->spa_async_lock);
2654 	tasks = spa->spa_async_tasks;
2655 	spa->spa_async_tasks = 0;
2656 	mutex_exit(&spa->spa_async_lock);
2657 
2658 	/*
2659 	 * See if the config needs to be updated.
2660 	 */
2661 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
2662 		mutex_enter(&spa_namespace_lock);
2663 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2664 		mutex_exit(&spa_namespace_lock);
2665 	}
2666 
2667 	/*
2668 	 * See if any devices need to be reopened.
2669 	 */
2670 	if (tasks & SPA_ASYNC_REOPEN)
2671 		spa_async_reopen(spa);
2672 
2673 	/*
2674 	 * If any devices are done replacing, detach them.
2675 	 */
2676 	if (tasks & SPA_ASYNC_REPLACE_DONE)
2677 		spa_vdev_replace_done(spa);
2678 
2679 	/*
2680 	 * Kick off a scrub.
2681 	 */
2682 	if (tasks & SPA_ASYNC_SCRUB)
2683 		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
2684 
2685 	/*
2686 	 * Kick off a resilver.
2687 	 */
2688 	if (tasks & SPA_ASYNC_RESILVER)
2689 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
2690 
2691 	/*
2692 	 * Let the world know that we're done.
2693 	 */
2694 	mutex_enter(&spa->spa_async_lock);
2695 	spa->spa_async_thread = NULL;
2696 	cv_broadcast(&spa->spa_async_cv);
2697 	mutex_exit(&spa->spa_async_lock);
2698 	thread_exit();
2699 }
2700 
2701 void
2702 spa_async_suspend(spa_t *spa)
2703 {
2704 	mutex_enter(&spa->spa_async_lock);
2705 	spa->spa_async_suspended++;
2706 	while (spa->spa_async_thread != NULL)
2707 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
2708 	mutex_exit(&spa->spa_async_lock);
2709 }
2710 
2711 void
2712 spa_async_resume(spa_t *spa)
2713 {
2714 	mutex_enter(&spa->spa_async_lock);
2715 	ASSERT(spa->spa_async_suspended != 0);
2716 	spa->spa_async_suspended--;
2717 	mutex_exit(&spa->spa_async_lock);
2718 }
2719 
2720 static void
2721 spa_async_dispatch(spa_t *spa)
2722 {
2723 	mutex_enter(&spa->spa_async_lock);
2724 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
2725 	    spa->spa_async_thread == NULL &&
2726 	    rootdir != NULL && !vn_is_readonly(rootdir))
2727 		spa->spa_async_thread = thread_create(NULL, 0,
2728 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
2729 	mutex_exit(&spa->spa_async_lock);
2730 }
2731 
2732 void
2733 spa_async_request(spa_t *spa, int task)
2734 {
2735 	mutex_enter(&spa->spa_async_lock);
2736 	spa->spa_async_tasks |= task;
2737 	mutex_exit(&spa->spa_async_lock);
2738 }
2739 
2740 /*
2741  * ==========================================================================
2742  * SPA syncing routines
2743  * ==========================================================================
2744  */
2745 
2746 static void
2747 spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
2748 {
2749 	bplist_t *bpl = &spa->spa_sync_bplist;
2750 	dmu_tx_t *tx;
2751 	blkptr_t blk;
2752 	uint64_t itor = 0;
2753 	zio_t *zio;
2754 	int error;
2755 	uint8_t c = 1;
2756 
2757 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
2758 
2759 	while (bplist_iterate(bpl, &itor, &blk) == 0)
2760 		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
2761 
2762 	error = zio_wait(zio);
2763 	ASSERT3U(error, ==, 0);
2764 
2765 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2766 	bplist_vacate(bpl, tx);
2767 
2768 	/*
2769 	 * Pre-dirty the first block so we sync to convergence faster.
2770 	 * (Usually only the first block is needed.)
2771 	 */
2772 	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
2773 	dmu_tx_commit(tx);
2774 }
2775 
2776 static void
2777 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
2778 {
2779 	char *packed = NULL;
2780 	size_t nvsize = 0;
2781 	dmu_buf_t *db;
2782 
2783 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
2784 
2785 	packed = kmem_alloc(nvsize, KM_SLEEP);
2786 
2787 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
2788 	    KM_SLEEP) == 0);
2789 
2790 	dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx);
2791 
2792 	kmem_free(packed, nvsize);
2793 
2794 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
2795 	dmu_buf_will_dirty(db, tx);
2796 	*(uint64_t *)db->db_data = nvsize;
2797 	dmu_buf_rele(db, FTAG);
2798 }
2799 
2800 static void
2801 spa_sync_spares(spa_t *spa, dmu_tx_t *tx)
2802 {
2803 	nvlist_t *nvroot;
2804 	nvlist_t **spares;
2805 	int i;
2806 
2807 	if (!spa->spa_sync_spares)
2808 		return;
2809 
2810 	/*
2811 	 * Update the MOS nvlist describing the list of available spares.
2812 	 * spa_validate_spares() will have already made sure this nvlist is
2813 	 * valid and the vdevs are labelled appropriately.
2814 	 */
2815 	if (spa->spa_spares_object == 0) {
2816 		spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset,
2817 		    DMU_OT_PACKED_NVLIST, 1 << 14,
2818 		    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2819 		VERIFY(zap_update(spa->spa_meta_objset,
2820 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES,
2821 		    sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0);
2822 	}
2823 
2824 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2825 	if (spa->spa_nspares == 0) {
2826 		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2827 		    NULL, 0) == 0);
2828 	} else {
2829 		spares = kmem_alloc(spa->spa_nspares * sizeof (void *),
2830 		    KM_SLEEP);
2831 		for (i = 0; i < spa->spa_nspares; i++)
2832 			spares[i] = vdev_config_generate(spa,
2833 			    spa->spa_spares[i], B_FALSE, B_TRUE);
2834 		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2835 		    spares, spa->spa_nspares) == 0);
2836 		for (i = 0; i < spa->spa_nspares; i++)
2837 			nvlist_free(spares[i]);
2838 		kmem_free(spares, spa->spa_nspares * sizeof (void *));
2839 	}
2840 
2841 	spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx);
2842 	nvlist_free(nvroot);
2843 
2844 	spa->spa_sync_spares = B_FALSE;
2845 }
2846 
2847 static void
2848 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
2849 {
2850 	nvlist_t *config;
2851 
2852 	if (list_is_empty(&spa->spa_dirty_list))
2853 		return;
2854 
2855 	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
2856 
2857 	if (spa->spa_config_syncing)
2858 		nvlist_free(spa->spa_config_syncing);
2859 	spa->spa_config_syncing = config;
2860 
2861 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
2862 }
2863 
2864 static void
2865 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
2866 {
2867 	spa_t *spa = arg1;
2868 	nvlist_t *nvp = arg2;
2869 	nvpair_t *nvpair;
2870 	objset_t *mos = spa->spa_meta_objset;
2871 	uint64_t zapobj;
2872 
2873 	mutex_enter(&spa->spa_props_lock);
2874 	if (spa->spa_pool_props_object == 0) {
2875 		zapobj = zap_create(mos, DMU_OT_POOL_PROPS, DMU_OT_NONE, 0, tx);
2876 		VERIFY(zapobj > 0);
2877 
2878 		spa->spa_pool_props_object = zapobj;
2879 
2880 		VERIFY(zap_update(mos, DMU_POOL_DIRECTORY_OBJECT,
2881 		    DMU_POOL_PROPS, 8, 1,
2882 		    &spa->spa_pool_props_object, tx) == 0);
2883 	}
2884 	mutex_exit(&spa->spa_props_lock);
2885 
2886 	nvpair = NULL;
2887 	while ((nvpair = nvlist_next_nvpair(nvp, nvpair))) {
2888 		switch (zpool_name_to_prop(nvpair_name(nvpair))) {
2889 		case ZFS_PROP_BOOTFS:
2890 			VERIFY(nvlist_lookup_uint64(nvp,
2891 			    nvpair_name(nvpair), &spa->spa_bootfs) == 0);
2892 			VERIFY(zap_update(mos,
2893 			    spa->spa_pool_props_object,
2894 			    zpool_prop_to_name(ZFS_PROP_BOOTFS), 8, 1,
2895 			    &spa->spa_bootfs, tx) == 0);
2896 			break;
2897 		}
2898 	}
2899 }
2900 
2901 /*
2902  * Sync the specified transaction group.  New blocks may be dirtied as
2903  * part of the process, so we iterate until it converges.
2904  */
2905 void
2906 spa_sync(spa_t *spa, uint64_t txg)
2907 {
2908 	dsl_pool_t *dp = spa->spa_dsl_pool;
2909 	objset_t *mos = spa->spa_meta_objset;
2910 	bplist_t *bpl = &spa->spa_sync_bplist;
2911 	vdev_t *rvd = spa->spa_root_vdev;
2912 	vdev_t *vd;
2913 	dmu_tx_t *tx;
2914 	int dirty_vdevs;
2915 
2916 	/*
2917 	 * Lock out configuration changes.
2918 	 */
2919 	spa_config_enter(spa, RW_READER, FTAG);
2920 
2921 	spa->spa_syncing_txg = txg;
2922 	spa->spa_sync_pass = 0;
2923 
2924 	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
2925 
2926 	tx = dmu_tx_create_assigned(dp, txg);
2927 
2928 	/*
2929 	 * If we are upgrading to ZFS_VERSION_RAIDZ_DEFLATE this txg,
2930 	 * set spa_deflate if we have no raid-z vdevs.
2931 	 */
2932 	if (spa->spa_ubsync.ub_version < ZFS_VERSION_RAIDZ_DEFLATE &&
2933 	    spa->spa_uberblock.ub_version >= ZFS_VERSION_RAIDZ_DEFLATE) {
2934 		int i;
2935 
2936 		for (i = 0; i < rvd->vdev_children; i++) {
2937 			vd = rvd->vdev_child[i];
2938 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
2939 				break;
2940 		}
2941 		if (i == rvd->vdev_children) {
2942 			spa->spa_deflate = TRUE;
2943 			VERIFY(0 == zap_add(spa->spa_meta_objset,
2944 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2945 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
2946 		}
2947 	}
2948 
2949 	/*
2950 	 * If anything has changed in this txg, push the deferred frees
2951 	 * from the previous txg.  If not, leave them alone so that we
2952 	 * don't generate work on an otherwise idle system.
2953 	 */
2954 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
2955 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
2956 	    !txg_list_empty(&dp->dp_sync_tasks, txg))
2957 		spa_sync_deferred_frees(spa, txg);
2958 
2959 	/*
2960 	 * Iterate to convergence.
2961 	 */
2962 	do {
2963 		spa->spa_sync_pass++;
2964 
2965 		spa_sync_config_object(spa, tx);
2966 		spa_sync_spares(spa, tx);
2967 		spa_errlog_sync(spa, txg);
2968 		dsl_pool_sync(dp, txg);
2969 
2970 		dirty_vdevs = 0;
2971 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
2972 			vdev_sync(vd, txg);
2973 			dirty_vdevs++;
2974 		}
2975 
2976 		bplist_sync(bpl, tx);
2977 	} while (dirty_vdevs);
2978 
2979 	bplist_close(bpl);
2980 
2981 	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
2982 
2983 	/*
2984 	 * Rewrite the vdev configuration (which includes the uberblock)
2985 	 * to commit the transaction group.
2986 	 *
2987 	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
2988 	 * Otherwise, pick a random top-level vdev that's known to be
2989 	 * visible in the config cache (see spa_vdev_add() for details).
2990 	 * If the write fails, try the next vdev until we're tried them all.
2991 	 */
2992 	if (!list_is_empty(&spa->spa_dirty_list)) {
2993 		VERIFY(vdev_config_sync(rvd, txg) == 0);
2994 	} else {
2995 		int children = rvd->vdev_children;
2996 		int c0 = spa_get_random(children);
2997 		int c;
2998 
2999 		for (c = 0; c < children; c++) {
3000 			vd = rvd->vdev_child[(c0 + c) % children];
3001 			if (vd->vdev_ms_array == 0)
3002 				continue;
3003 			if (vdev_config_sync(vd, txg) == 0)
3004 				break;
3005 		}
3006 		if (c == children)
3007 			VERIFY(vdev_config_sync(rvd, txg) == 0);
3008 	}
3009 
3010 	dmu_tx_commit(tx);
3011 
3012 	/*
3013 	 * Clear the dirty config list.
3014 	 */
3015 	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
3016 		vdev_config_clean(vd);
3017 
3018 	/*
3019 	 * Now that the new config has synced transactionally,
3020 	 * let it become visible to the config cache.
3021 	 */
3022 	if (spa->spa_config_syncing != NULL) {
3023 		spa_config_set(spa, spa->spa_config_syncing);
3024 		spa->spa_config_txg = txg;
3025 		spa->spa_config_syncing = NULL;
3026 	}
3027 
3028 	/*
3029 	 * Make a stable copy of the fully synced uberblock.
3030 	 * We use this as the root for pool traversals.
3031 	 */
3032 	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
3033 
3034 	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
3035 
3036 	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
3037 	spa->spa_traverse_wanted = 0;
3038 	spa->spa_ubsync = spa->spa_uberblock;
3039 	rw_exit(&spa->spa_traverse_lock);
3040 
3041 	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
3042 
3043 	/*
3044 	 * Clean up the ZIL records for the synced txg.
3045 	 */
3046 	dsl_pool_zil_clean(dp);
3047 
3048 	/*
3049 	 * Update usable space statistics.
3050 	 */
3051 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
3052 		vdev_sync_done(vd, txg);
3053 
3054 	/*
3055 	 * It had better be the case that we didn't dirty anything
3056 	 * since vdev_config_sync().
3057 	 */
3058 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
3059 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
3060 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
3061 	ASSERT(bpl->bpl_queue == NULL);
3062 
3063 	spa_config_exit(spa, FTAG);
3064 
3065 	/*
3066 	 * If any async tasks have been requested, kick them off.
3067 	 */
3068 	spa_async_dispatch(spa);
3069 }
3070 
3071 /*
3072  * Sync all pools.  We don't want to hold the namespace lock across these
3073  * operations, so we take a reference on the spa_t and drop the lock during the
3074  * sync.
3075  */
3076 void
3077 spa_sync_allpools(void)
3078 {
3079 	spa_t *spa = NULL;
3080 	mutex_enter(&spa_namespace_lock);
3081 	while ((spa = spa_next(spa)) != NULL) {
3082 		if (spa_state(spa) != POOL_STATE_ACTIVE)
3083 			continue;
3084 		spa_open_ref(spa, FTAG);
3085 		mutex_exit(&spa_namespace_lock);
3086 		txg_wait_synced(spa_get_dsl(spa), 0);
3087 		mutex_enter(&spa_namespace_lock);
3088 		spa_close(spa, FTAG);
3089 	}
3090 	mutex_exit(&spa_namespace_lock);
3091 }
3092 
3093 /*
3094  * ==========================================================================
3095  * Miscellaneous routines
3096  * ==========================================================================
3097  */
3098 
3099 /*
3100  * Remove all pools in the system.
3101  */
3102 void
3103 spa_evict_all(void)
3104 {
3105 	spa_t *spa;
3106 
3107 	/*
3108 	 * Remove all cached state.  All pools should be closed now,
3109 	 * so every spa in the AVL tree should be unreferenced.
3110 	 */
3111 	mutex_enter(&spa_namespace_lock);
3112 	while ((spa = spa_next(NULL)) != NULL) {
3113 		/*
3114 		 * Stop async tasks.  The async thread may need to detach
3115 		 * a device that's been replaced, which requires grabbing
3116 		 * spa_namespace_lock, so we must drop it here.
3117 		 */
3118 		spa_open_ref(spa, FTAG);
3119 		mutex_exit(&spa_namespace_lock);
3120 		spa_async_suspend(spa);
3121 		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
3122 		mutex_enter(&spa_namespace_lock);
3123 		spa_close(spa, FTAG);
3124 
3125 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3126 			spa_unload(spa);
3127 			spa_deactivate(spa);
3128 		}
3129 		spa_remove(spa);
3130 	}
3131 	mutex_exit(&spa_namespace_lock);
3132 }
3133 
3134 vdev_t *
3135 spa_lookup_by_guid(spa_t *spa, uint64_t guid)
3136 {
3137 	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
3138 }
3139 
3140 void
3141 spa_upgrade(spa_t *spa)
3142 {
3143 	spa_config_enter(spa, RW_WRITER, FTAG);
3144 
3145 	/*
3146 	 * This should only be called for a non-faulted pool, and since a
3147 	 * future version would result in an unopenable pool, this shouldn't be
3148 	 * possible.
3149 	 */
3150 	ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION);
3151 
3152 	spa->spa_uberblock.ub_version = ZFS_VERSION;
3153 	vdev_config_dirty(spa->spa_root_vdev);
3154 
3155 	spa_config_exit(spa, FTAG);
3156 
3157 	txg_wait_synced(spa_get_dsl(spa), 0);
3158 }
3159 
3160 boolean_t
3161 spa_has_spare(spa_t *spa, uint64_t guid)
3162 {
3163 	int i;
3164 	uint64_t spareguid;
3165 
3166 	for (i = 0; i < spa->spa_nspares; i++)
3167 		if (spa->spa_spares[i]->vdev_guid == guid)
3168 			return (B_TRUE);
3169 
3170 	for (i = 0; i < spa->spa_pending_nspares; i++) {
3171 		if (nvlist_lookup_uint64(spa->spa_pending_spares[i],
3172 		    ZPOOL_CONFIG_GUID, &spareguid) == 0 &&
3173 		    spareguid == guid)
3174 			return (B_TRUE);
3175 	}
3176 
3177 	return (B_FALSE);
3178 }
3179 
3180 int
3181 spa_set_props(spa_t *spa, nvlist_t *nvp)
3182 {
3183 	return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
3184 	    spa, nvp, 3));
3185 }
3186 
3187 int
3188 spa_get_props(spa_t *spa, nvlist_t **nvp)
3189 {
3190 	zap_cursor_t zc;
3191 	zap_attribute_t za;
3192 	objset_t *mos = spa->spa_meta_objset;
3193 	zfs_source_t src;
3194 	zfs_prop_t prop;
3195 	nvlist_t *propval;
3196 	uint64_t value;
3197 	int err;
3198 
3199 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3200 
3201 	mutex_enter(&spa->spa_props_lock);
3202 	/* If no props object, then just return empty nvlist */
3203 	if (spa->spa_pool_props_object == 0) {
3204 		mutex_exit(&spa->spa_props_lock);
3205 		return (0);
3206 	}
3207 
3208 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
3209 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
3210 	    zap_cursor_advance(&zc)) {
3211 
3212 		if ((prop = zpool_name_to_prop(za.za_name)) == ZFS_PROP_INVAL)
3213 			continue;
3214 
3215 		VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3216 		switch (za.za_integer_length) {
3217 		case 8:
3218 			if (zfs_prop_default_numeric(prop) ==
3219 			    za.za_first_integer)
3220 				src = ZFS_SRC_DEFAULT;
3221 			else
3222 				src = ZFS_SRC_LOCAL;
3223 			value = za.za_first_integer;
3224 
3225 			if (prop == ZFS_PROP_BOOTFS) {
3226 				dsl_pool_t *dp;
3227 				dsl_dataset_t *ds = NULL;
3228 				char strval[MAXPATHLEN];
3229 
3230 				dp = spa_get_dsl(spa);
3231 				rw_enter(&dp->dp_config_rwlock, RW_READER);
3232 				if ((err = dsl_dataset_open_obj(dp,
3233 				    za.za_first_integer, NULL, DS_MODE_NONE,
3234 				    FTAG, &ds)) != 0) {
3235 					rw_exit(&dp->dp_config_rwlock);
3236 					break;
3237 				}
3238 				dsl_dataset_name(ds, strval);
3239 				dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
3240 				rw_exit(&dp->dp_config_rwlock);
3241 
3242 				VERIFY(nvlist_add_uint64(propval,
3243 				    ZFS_PROP_SOURCE, src) == 0);
3244 				VERIFY(nvlist_add_string(propval,
3245 				    ZFS_PROP_VALUE, strval) == 0);
3246 			} else {
3247 				VERIFY(nvlist_add_uint64(propval,
3248 				    ZFS_PROP_SOURCE, src) == 0);
3249 				VERIFY(nvlist_add_uint64(propval,
3250 				    ZFS_PROP_VALUE, value) == 0);
3251 			}
3252 			VERIFY(nvlist_add_nvlist(*nvp, za.za_name,
3253 			    propval) == 0);
3254 			break;
3255 		}
3256 		nvlist_free(propval);
3257 	}
3258 	zap_cursor_fini(&zc);
3259 	mutex_exit(&spa->spa_props_lock);
3260 	if (err && err != ENOENT) {
3261 		nvlist_free(*nvp);
3262 		return (err);
3263 	}
3264 
3265 	return (0);
3266 }
3267 
3268 /*
3269  * If the bootfs property value is dsobj, clear it.
3270  */
3271 void
3272 spa_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
3273 {
3274 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
3275 		VERIFY(zap_remove(spa->spa_meta_objset,
3276 		    spa->spa_pool_props_object,
3277 		    zpool_prop_to_name(ZFS_PROP_BOOTFS), tx) == 0);
3278 		spa->spa_bootfs = 0;
3279 	}
3280 }
3281