xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa.c (revision 69a0bf0c658b89d78f2def6997ee0e7d96f13524)
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  * Checks to see if the given vdev could not be opened, in which case we post a
428  * sysevent to notify the autoreplace code that the device has been removed.
429  */
430 static void
431 spa_check_removed(vdev_t *vd)
432 {
433 	int c;
434 
435 	for (c = 0; c < vd->vdev_children; c++)
436 		spa_check_removed(vd->vdev_child[c]);
437 
438 	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
439 		zfs_post_autoreplace(vd->vdev_spa, vd);
440 		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
441 	}
442 }
443 
444 /*
445  * Load an existing storage pool, using the pool's builtin spa_config as a
446  * source of configuration information.
447  */
448 static int
449 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
450 {
451 	int error = 0;
452 	nvlist_t *nvroot = NULL;
453 	vdev_t *rvd;
454 	uberblock_t *ub = &spa->spa_uberblock;
455 	uint64_t config_cache_txg = spa->spa_config_txg;
456 	uint64_t pool_guid;
457 	uint64_t version;
458 	zio_t *zio;
459 	uint64_t autoreplace = 0;
460 
461 	spa->spa_load_state = state;
462 
463 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
464 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
465 		error = EINVAL;
466 		goto out;
467 	}
468 
469 	/*
470 	 * Versioning wasn't explicitly added to the label until later, so if
471 	 * it's not present treat it as the initial version.
472 	 */
473 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
474 		version = ZFS_VERSION_INITIAL;
475 
476 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
477 	    &spa->spa_config_txg);
478 
479 	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
480 	    spa_guid_exists(pool_guid, 0)) {
481 		error = EEXIST;
482 		goto out;
483 	}
484 
485 	spa->spa_load_guid = pool_guid;
486 
487 	/*
488 	 * Parse the configuration into a vdev tree.  We explicitly set the
489 	 * value that will be returned by spa_version() since parsing the
490 	 * configuration requires knowing the version number.
491 	 */
492 	spa_config_enter(spa, RW_WRITER, FTAG);
493 	spa->spa_ubsync.ub_version = version;
494 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
495 	spa_config_exit(spa, FTAG);
496 
497 	if (error != 0)
498 		goto out;
499 
500 	ASSERT(spa->spa_root_vdev == rvd);
501 	ASSERT(spa_guid(spa) == pool_guid);
502 
503 	/*
504 	 * Try to open all vdevs, loading each label in the process.
505 	 */
506 	error = vdev_open(rvd);
507 	if (error != 0)
508 		goto out;
509 
510 	/*
511 	 * Validate the labels for all leaf vdevs.  We need to grab the config
512 	 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD
513 	 * flag.
514 	 */
515 	spa_config_enter(spa, RW_READER, FTAG);
516 	error = vdev_validate(rvd);
517 	spa_config_exit(spa, FTAG);
518 
519 	if (error != 0)
520 		goto out;
521 
522 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
523 		error = ENXIO;
524 		goto out;
525 	}
526 
527 	/*
528 	 * Find the best uberblock.
529 	 */
530 	bzero(ub, sizeof (uberblock_t));
531 
532 	zio = zio_root(spa, NULL, NULL,
533 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
534 	vdev_uberblock_load(zio, rvd, ub);
535 	error = zio_wait(zio);
536 
537 	/*
538 	 * If we weren't able to find a single valid uberblock, return failure.
539 	 */
540 	if (ub->ub_txg == 0) {
541 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
542 		    VDEV_AUX_CORRUPT_DATA);
543 		error = ENXIO;
544 		goto out;
545 	}
546 
547 	/*
548 	 * If the pool is newer than the code, we can't open it.
549 	 */
550 	if (ub->ub_version > ZFS_VERSION) {
551 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
552 		    VDEV_AUX_VERSION_NEWER);
553 		error = ENOTSUP;
554 		goto out;
555 	}
556 
557 	/*
558 	 * If the vdev guid sum doesn't match the uberblock, we have an
559 	 * incomplete configuration.
560 	 */
561 	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
562 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
563 		    VDEV_AUX_BAD_GUID_SUM);
564 		error = ENXIO;
565 		goto out;
566 	}
567 
568 	/*
569 	 * Initialize internal SPA structures.
570 	 */
571 	spa->spa_state = POOL_STATE_ACTIVE;
572 	spa->spa_ubsync = spa->spa_uberblock;
573 	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
574 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
575 	if (error) {
576 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
577 		    VDEV_AUX_CORRUPT_DATA);
578 		goto out;
579 	}
580 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
581 
582 	if (zap_lookup(spa->spa_meta_objset,
583 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
584 	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
585 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
586 		    VDEV_AUX_CORRUPT_DATA);
587 		error = EIO;
588 		goto out;
589 	}
590 
591 	if (!mosconfig) {
592 		nvlist_t *newconfig;
593 		uint64_t hostid;
594 
595 		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
596 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
597 			    VDEV_AUX_CORRUPT_DATA);
598 			error = EIO;
599 			goto out;
600 		}
601 
602 		if (nvlist_lookup_uint64(newconfig, ZPOOL_CONFIG_HOSTID,
603 		    &hostid) == 0) {
604 			char *hostname;
605 			unsigned long myhostid = 0;
606 
607 			VERIFY(nvlist_lookup_string(newconfig,
608 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
609 
610 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
611 			if (hostid != 0 && myhostid != 0 &&
612 			    (unsigned long)hostid != myhostid) {
613 				cmn_err(CE_WARN, "pool '%s' could not be "
614 				    "loaded as it was last accessed by "
615 				    "another system (host: %s hostid: 0x%lx).  "
616 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
617 				    spa->spa_name, hostname,
618 				    (unsigned long)hostid);
619 				error = EBADF;
620 				goto out;
621 			}
622 		}
623 
624 		spa_config_set(spa, newconfig);
625 		spa_unload(spa);
626 		spa_deactivate(spa);
627 		spa_activate(spa);
628 
629 		return (spa_load(spa, newconfig, state, B_TRUE));
630 	}
631 
632 	if (zap_lookup(spa->spa_meta_objset,
633 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
634 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
635 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
636 		    VDEV_AUX_CORRUPT_DATA);
637 		error = EIO;
638 		goto out;
639 	}
640 
641 	/*
642 	 * Load the bit that tells us to use the new accounting function
643 	 * (raid-z deflation).  If we have an older pool, this will not
644 	 * be present.
645 	 */
646 	error = zap_lookup(spa->spa_meta_objset,
647 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
648 	    sizeof (uint64_t), 1, &spa->spa_deflate);
649 	if (error != 0 && error != ENOENT) {
650 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
651 		    VDEV_AUX_CORRUPT_DATA);
652 		error = EIO;
653 		goto out;
654 	}
655 
656 	/*
657 	 * Load the persistent error log.  If we have an older pool, this will
658 	 * not be present.
659 	 */
660 	error = zap_lookup(spa->spa_meta_objset,
661 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
662 	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
663 	if (error != 0 && error != ENOENT) {
664 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
665 		    VDEV_AUX_CORRUPT_DATA);
666 		error = EIO;
667 		goto out;
668 	}
669 
670 	error = zap_lookup(spa->spa_meta_objset,
671 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
672 	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
673 	if (error != 0 && error != ENOENT) {
674 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
675 		    VDEV_AUX_CORRUPT_DATA);
676 		error = EIO;
677 		goto out;
678 	}
679 
680 	/*
681 	 * Load the history object.  If we have an older pool, this
682 	 * will not be present.
683 	 */
684 	error = zap_lookup(spa->spa_meta_objset,
685 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
686 	    sizeof (uint64_t), 1, &spa->spa_history);
687 	if (error != 0 && error != ENOENT) {
688 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
689 		    VDEV_AUX_CORRUPT_DATA);
690 		error = EIO;
691 		goto out;
692 	}
693 
694 	/*
695 	 * Load any hot spares for this pool.
696 	 */
697 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
698 	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object);
699 	if (error != 0 && error != ENOENT) {
700 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
701 		    VDEV_AUX_CORRUPT_DATA);
702 		error = EIO;
703 		goto out;
704 	}
705 	if (error == 0) {
706 		ASSERT(spa_version(spa) >= ZFS_VERSION_SPARES);
707 		if (load_nvlist(spa, spa->spa_spares_object,
708 		    &spa->spa_sparelist) != 0) {
709 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
710 			    VDEV_AUX_CORRUPT_DATA);
711 			error = EIO;
712 			goto out;
713 		}
714 
715 		spa_config_enter(spa, RW_WRITER, FTAG);
716 		spa_load_spares(spa);
717 		spa_config_exit(spa, FTAG);
718 	}
719 
720 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
721 	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
722 
723 	if (error && error != ENOENT) {
724 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
725 		    VDEV_AUX_CORRUPT_DATA);
726 		error = EIO;
727 		goto out;
728 	}
729 
730 	if (error == 0) {
731 		(void) zap_lookup(spa->spa_meta_objset,
732 		    spa->spa_pool_props_object,
733 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
734 		    sizeof (uint64_t), 1, &spa->spa_bootfs);
735 		(void) zap_lookup(spa->spa_meta_objset,
736 		    spa->spa_pool_props_object,
737 		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
738 		    sizeof (uint64_t), 1, &autoreplace);
739 	}
740 
741 	/*
742 	 * If the 'autoreplace' property is set, then post a resource notifying
743 	 * the ZFS DE that it should not issue any faults for unopenable
744 	 * devices.  We also iterate over the vdevs, and post a sysevent for any
745 	 * unopenable vdevs so that the normal autoreplace handler can take
746 	 * over.
747 	 */
748 	if (autoreplace)
749 		spa_check_removed(spa->spa_root_vdev);
750 
751 	/*
752 	 * Load the vdev state for all toplevel vdevs.
753 	 */
754 	vdev_load(rvd);
755 
756 	/*
757 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
758 	 */
759 	spa_config_enter(spa, RW_WRITER, FTAG);
760 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
761 	spa_config_exit(spa, FTAG);
762 
763 	/*
764 	 * Check the state of the root vdev.  If it can't be opened, it
765 	 * indicates one or more toplevel vdevs are faulted.
766 	 */
767 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
768 		error = ENXIO;
769 		goto out;
770 	}
771 
772 	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
773 		dmu_tx_t *tx;
774 		int need_update = B_FALSE;
775 		int c;
776 
777 		/*
778 		 * Claim log blocks that haven't been committed yet.
779 		 * This must all happen in a single txg.
780 		 */
781 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
782 		    spa_first_txg(spa));
783 		(void) dmu_objset_find(spa->spa_name,
784 		    zil_claim, tx, DS_FIND_CHILDREN);
785 		dmu_tx_commit(tx);
786 
787 		spa->spa_sync_on = B_TRUE;
788 		txg_sync_start(spa->spa_dsl_pool);
789 
790 		/*
791 		 * Wait for all claims to sync.
792 		 */
793 		txg_wait_synced(spa->spa_dsl_pool, 0);
794 
795 		/*
796 		 * If the config cache is stale, or we have uninitialized
797 		 * metaslabs (see spa_vdev_add()), then update the config.
798 		 */
799 		if (config_cache_txg != spa->spa_config_txg ||
800 		    state == SPA_LOAD_IMPORT)
801 			need_update = B_TRUE;
802 
803 		for (c = 0; c < rvd->vdev_children; c++)
804 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
805 				need_update = B_TRUE;
806 
807 		/*
808 		 * Update the config cache asychronously in case we're the
809 		 * root pool, in which case the config cache isn't writable yet.
810 		 */
811 		if (need_update)
812 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
813 	}
814 
815 	error = 0;
816 out:
817 	if (error && error != EBADF)
818 		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
819 	spa->spa_load_state = SPA_LOAD_NONE;
820 	spa->spa_ena = 0;
821 
822 	return (error);
823 }
824 
825 /*
826  * Pool Open/Import
827  *
828  * The import case is identical to an open except that the configuration is sent
829  * down from userland, instead of grabbed from the configuration cache.  For the
830  * case of an open, the pool configuration will exist in the
831  * POOL_STATE_UNINITIALIZED state.
832  *
833  * The stats information (gen/count/ustats) is used to gather vdev statistics at
834  * the same time open the pool, without having to keep around the spa_t in some
835  * ambiguous state.
836  */
837 static int
838 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
839 {
840 	spa_t *spa;
841 	int error;
842 	int loaded = B_FALSE;
843 	int locked = B_FALSE;
844 
845 	*spapp = NULL;
846 
847 	/*
848 	 * As disgusting as this is, we need to support recursive calls to this
849 	 * function because dsl_dir_open() is called during spa_load(), and ends
850 	 * up calling spa_open() again.  The real fix is to figure out how to
851 	 * avoid dsl_dir_open() calling this in the first place.
852 	 */
853 	if (mutex_owner(&spa_namespace_lock) != curthread) {
854 		mutex_enter(&spa_namespace_lock);
855 		locked = B_TRUE;
856 	}
857 
858 	if ((spa = spa_lookup(pool)) == NULL) {
859 		if (locked)
860 			mutex_exit(&spa_namespace_lock);
861 		return (ENOENT);
862 	}
863 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
864 
865 		spa_activate(spa);
866 
867 		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
868 
869 		if (error == EBADF) {
870 			/*
871 			 * If vdev_validate() returns failure (indicated by
872 			 * EBADF), it indicates that one of the vdevs indicates
873 			 * that the pool has been exported or destroyed.  If
874 			 * this is the case, the config cache is out of sync and
875 			 * we should remove the pool from the namespace.
876 			 */
877 			zfs_post_ok(spa, NULL);
878 			spa_unload(spa);
879 			spa_deactivate(spa);
880 			spa_remove(spa);
881 			spa_config_sync();
882 			if (locked)
883 				mutex_exit(&spa_namespace_lock);
884 			return (ENOENT);
885 		}
886 
887 		if (error) {
888 			/*
889 			 * We can't open the pool, but we still have useful
890 			 * information: the state of each vdev after the
891 			 * attempted vdev_open().  Return this to the user.
892 			 */
893 			if (config != NULL && spa->spa_root_vdev != NULL) {
894 				spa_config_enter(spa, RW_READER, FTAG);
895 				*config = spa_config_generate(spa, NULL, -1ULL,
896 				    B_TRUE);
897 				spa_config_exit(spa, FTAG);
898 			}
899 			spa_unload(spa);
900 			spa_deactivate(spa);
901 			spa->spa_last_open_failed = B_TRUE;
902 			if (locked)
903 				mutex_exit(&spa_namespace_lock);
904 			*spapp = NULL;
905 			return (error);
906 		} else {
907 			zfs_post_ok(spa, NULL);
908 			spa->spa_last_open_failed = B_FALSE;
909 		}
910 
911 		loaded = B_TRUE;
912 	}
913 
914 	spa_open_ref(spa, tag);
915 
916 	/*
917 	 * If we just loaded the pool, resilver anything that's out of date.
918 	 */
919 	if (loaded && (spa_mode & FWRITE))
920 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
921 
922 	if (locked)
923 		mutex_exit(&spa_namespace_lock);
924 
925 	*spapp = spa;
926 
927 	if (config != NULL) {
928 		spa_config_enter(spa, RW_READER, FTAG);
929 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
930 		spa_config_exit(spa, FTAG);
931 	}
932 
933 	return (0);
934 }
935 
936 int
937 spa_open(const char *name, spa_t **spapp, void *tag)
938 {
939 	return (spa_open_common(name, spapp, tag, NULL));
940 }
941 
942 /*
943  * Lookup the given spa_t, incrementing the inject count in the process,
944  * preventing it from being exported or destroyed.
945  */
946 spa_t *
947 spa_inject_addref(char *name)
948 {
949 	spa_t *spa;
950 
951 	mutex_enter(&spa_namespace_lock);
952 	if ((spa = spa_lookup(name)) == NULL) {
953 		mutex_exit(&spa_namespace_lock);
954 		return (NULL);
955 	}
956 	spa->spa_inject_ref++;
957 	mutex_exit(&spa_namespace_lock);
958 
959 	return (spa);
960 }
961 
962 void
963 spa_inject_delref(spa_t *spa)
964 {
965 	mutex_enter(&spa_namespace_lock);
966 	spa->spa_inject_ref--;
967 	mutex_exit(&spa_namespace_lock);
968 }
969 
970 static void
971 spa_add_spares(spa_t *spa, nvlist_t *config)
972 {
973 	nvlist_t **spares;
974 	uint_t i, nspares;
975 	nvlist_t *nvroot;
976 	uint64_t guid;
977 	vdev_stat_t *vs;
978 	uint_t vsc;
979 	uint64_t pool;
980 
981 	if (spa->spa_nspares == 0)
982 		return;
983 
984 	VERIFY(nvlist_lookup_nvlist(config,
985 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
986 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
987 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
988 	if (nspares != 0) {
989 		VERIFY(nvlist_add_nvlist_array(nvroot,
990 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
991 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
992 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
993 
994 		/*
995 		 * Go through and find any spares which have since been
996 		 * repurposed as an active spare.  If this is the case, update
997 		 * their status appropriately.
998 		 */
999 		for (i = 0; i < nspares; i++) {
1000 			VERIFY(nvlist_lookup_uint64(spares[i],
1001 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1002 			if (spa_spare_exists(guid, &pool) && pool != 0ULL) {
1003 				VERIFY(nvlist_lookup_uint64_array(
1004 				    spares[i], ZPOOL_CONFIG_STATS,
1005 				    (uint64_t **)&vs, &vsc) == 0);
1006 				vs->vs_state = VDEV_STATE_CANT_OPEN;
1007 				vs->vs_aux = VDEV_AUX_SPARED;
1008 			}
1009 		}
1010 	}
1011 }
1012 
1013 int
1014 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1015 {
1016 	int error;
1017 	spa_t *spa;
1018 
1019 	*config = NULL;
1020 	error = spa_open_common(name, &spa, FTAG, config);
1021 
1022 	if (spa && *config != NULL) {
1023 		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
1024 		    spa_get_errlog_size(spa)) == 0);
1025 
1026 		spa_add_spares(spa, *config);
1027 	}
1028 
1029 	/*
1030 	 * We want to get the alternate root even for faulted pools, so we cheat
1031 	 * and call spa_lookup() directly.
1032 	 */
1033 	if (altroot) {
1034 		if (spa == NULL) {
1035 			mutex_enter(&spa_namespace_lock);
1036 			spa = spa_lookup(name);
1037 			if (spa)
1038 				spa_altroot(spa, altroot, buflen);
1039 			else
1040 				altroot[0] = '\0';
1041 			spa = NULL;
1042 			mutex_exit(&spa_namespace_lock);
1043 		} else {
1044 			spa_altroot(spa, altroot, buflen);
1045 		}
1046 	}
1047 
1048 	if (spa != NULL)
1049 		spa_close(spa, FTAG);
1050 
1051 	return (error);
1052 }
1053 
1054 /*
1055  * Validate that the 'spares' array is well formed.  We must have an array of
1056  * nvlists, each which describes a valid leaf vdev.  If this is an import (mode
1057  * is VDEV_ALLOC_SPARE), then we allow corrupted spares to be specified, as long
1058  * as they are well-formed.
1059  */
1060 static int
1061 spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1062 {
1063 	nvlist_t **spares;
1064 	uint_t i, nspares;
1065 	vdev_t *vd;
1066 	int error;
1067 
1068 	/*
1069 	 * It's acceptable to have no spares specified.
1070 	 */
1071 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1072 	    &spares, &nspares) != 0)
1073 		return (0);
1074 
1075 	if (nspares == 0)
1076 		return (EINVAL);
1077 
1078 	/*
1079 	 * Make sure the pool is formatted with a version that supports hot
1080 	 * spares.
1081 	 */
1082 	if (spa_version(spa) < ZFS_VERSION_SPARES)
1083 		return (ENOTSUP);
1084 
1085 	/*
1086 	 * Set the pending spare list so we correctly handle device in-use
1087 	 * checking.
1088 	 */
1089 	spa->spa_pending_spares = spares;
1090 	spa->spa_pending_nspares = nspares;
1091 
1092 	for (i = 0; i < nspares; i++) {
1093 		if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0,
1094 		    mode)) != 0)
1095 			goto out;
1096 
1097 		if (!vd->vdev_ops->vdev_op_leaf) {
1098 			vdev_free(vd);
1099 			error = EINVAL;
1100 			goto out;
1101 		}
1102 
1103 		vd->vdev_top = vd;
1104 
1105 		if ((error = vdev_open(vd)) == 0 &&
1106 		    (error = vdev_label_init(vd, crtxg,
1107 		    VDEV_LABEL_SPARE)) == 0) {
1108 			VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID,
1109 			    vd->vdev_guid) == 0);
1110 		}
1111 
1112 		vdev_free(vd);
1113 
1114 		if (error && mode != VDEV_ALLOC_SPARE)
1115 			goto out;
1116 		else
1117 			error = 0;
1118 	}
1119 
1120 out:
1121 	spa->spa_pending_spares = NULL;
1122 	spa->spa_pending_nspares = 0;
1123 	return (error);
1124 }
1125 
1126 /*
1127  * Pool Creation
1128  */
1129 int
1130 spa_create(const char *pool, nvlist_t *nvroot, const char *altroot)
1131 {
1132 	spa_t *spa;
1133 	vdev_t *rvd;
1134 	dsl_pool_t *dp;
1135 	dmu_tx_t *tx;
1136 	int c, error = 0;
1137 	uint64_t txg = TXG_INITIAL;
1138 	nvlist_t **spares;
1139 	uint_t nspares;
1140 
1141 	/*
1142 	 * If this pool already exists, return failure.
1143 	 */
1144 	mutex_enter(&spa_namespace_lock);
1145 	if (spa_lookup(pool) != NULL) {
1146 		mutex_exit(&spa_namespace_lock);
1147 		return (EEXIST);
1148 	}
1149 
1150 	/*
1151 	 * Allocate a new spa_t structure.
1152 	 */
1153 	spa = spa_add(pool, altroot);
1154 	spa_activate(spa);
1155 
1156 	spa->spa_uberblock.ub_txg = txg - 1;
1157 	spa->spa_uberblock.ub_version = ZFS_VERSION;
1158 	spa->spa_ubsync = spa->spa_uberblock;
1159 
1160 	/*
1161 	 * Create the root vdev.
1162 	 */
1163 	spa_config_enter(spa, RW_WRITER, FTAG);
1164 
1165 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1166 
1167 	ASSERT(error != 0 || rvd != NULL);
1168 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1169 
1170 	if (error == 0 && rvd->vdev_children == 0)
1171 		error = EINVAL;
1172 
1173 	if (error == 0 &&
1174 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1175 	    (error = spa_validate_spares(spa, nvroot, txg,
1176 	    VDEV_ALLOC_ADD)) == 0) {
1177 		for (c = 0; c < rvd->vdev_children; c++)
1178 			vdev_init(rvd->vdev_child[c], txg);
1179 		vdev_config_dirty(rvd);
1180 	}
1181 
1182 	spa_config_exit(spa, FTAG);
1183 
1184 	if (error != 0) {
1185 		spa_unload(spa);
1186 		spa_deactivate(spa);
1187 		spa_remove(spa);
1188 		mutex_exit(&spa_namespace_lock);
1189 		return (error);
1190 	}
1191 
1192 	/*
1193 	 * Get the list of spares, if specified.
1194 	 */
1195 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1196 	    &spares, &nspares) == 0) {
1197 		VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME,
1198 		    KM_SLEEP) == 0);
1199 		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1200 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1201 		spa_config_enter(spa, RW_WRITER, FTAG);
1202 		spa_load_spares(spa);
1203 		spa_config_exit(spa, FTAG);
1204 		spa->spa_sync_spares = B_TRUE;
1205 	}
1206 
1207 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
1208 	spa->spa_meta_objset = dp->dp_meta_objset;
1209 
1210 	tx = dmu_tx_create_assigned(dp, txg);
1211 
1212 	/*
1213 	 * Create the pool config object.
1214 	 */
1215 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
1216 	    DMU_OT_PACKED_NVLIST, 1 << 14,
1217 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
1218 
1219 	if (zap_add(spa->spa_meta_objset,
1220 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1221 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
1222 		cmn_err(CE_PANIC, "failed to add pool config");
1223 	}
1224 
1225 	/* Newly created pools are always deflated. */
1226 	spa->spa_deflate = TRUE;
1227 	if (zap_add(spa->spa_meta_objset,
1228 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1229 	    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
1230 		cmn_err(CE_PANIC, "failed to add deflate");
1231 	}
1232 
1233 	/*
1234 	 * Create the deferred-free bplist object.  Turn off compression
1235 	 * because sync-to-convergence takes longer if the blocksize
1236 	 * keeps changing.
1237 	 */
1238 	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
1239 	    1 << 14, tx);
1240 	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
1241 	    ZIO_COMPRESS_OFF, tx);
1242 
1243 	if (zap_add(spa->spa_meta_objset,
1244 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1245 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
1246 		cmn_err(CE_PANIC, "failed to add bplist");
1247 	}
1248 
1249 	/*
1250 	 * Create the pool's history object.
1251 	 */
1252 	spa_history_create_obj(spa, tx);
1253 
1254 	dmu_tx_commit(tx);
1255 
1256 	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
1257 	spa->spa_sync_on = B_TRUE;
1258 	txg_sync_start(spa->spa_dsl_pool);
1259 
1260 	/*
1261 	 * We explicitly wait for the first transaction to complete so that our
1262 	 * bean counters are appropriately updated.
1263 	 */
1264 	txg_wait_synced(spa->spa_dsl_pool, txg);
1265 
1266 	spa_config_sync();
1267 
1268 	mutex_exit(&spa_namespace_lock);
1269 
1270 	return (0);
1271 }
1272 
1273 /*
1274  * Import the given pool into the system.  We set up the necessary spa_t and
1275  * then call spa_load() to do the dirty work.
1276  */
1277 int
1278 spa_import(const char *pool, nvlist_t *config, const char *altroot)
1279 {
1280 	spa_t *spa;
1281 	int error;
1282 	nvlist_t *nvroot;
1283 	nvlist_t **spares;
1284 	uint_t nspares;
1285 
1286 	if (!(spa_mode & FWRITE))
1287 		return (EROFS);
1288 
1289 	/*
1290 	 * If a pool with this name exists, return failure.
1291 	 */
1292 	mutex_enter(&spa_namespace_lock);
1293 	if (spa_lookup(pool) != NULL) {
1294 		mutex_exit(&spa_namespace_lock);
1295 		return (EEXIST);
1296 	}
1297 
1298 	/*
1299 	 * Create and initialize the spa structure.
1300 	 */
1301 	spa = spa_add(pool, altroot);
1302 	spa_activate(spa);
1303 
1304 	/*
1305 	 * Pass off the heavy lifting to spa_load().
1306 	 * Pass TRUE for mosconfig because the user-supplied config
1307 	 * is actually the one to trust when doing an import.
1308 	 */
1309 	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
1310 
1311 	spa_config_enter(spa, RW_WRITER, FTAG);
1312 	/*
1313 	 * Toss any existing sparelist, as it doesn't have any validity anymore,
1314 	 * and conflicts with spa_has_spare().
1315 	 */
1316 	if (spa->spa_sparelist) {
1317 		nvlist_free(spa->spa_sparelist);
1318 		spa->spa_sparelist = NULL;
1319 		spa_load_spares(spa);
1320 	}
1321 
1322 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1323 	    &nvroot) == 0);
1324 	if (error == 0)
1325 		error = spa_validate_spares(spa, nvroot, -1ULL,
1326 		    VDEV_ALLOC_SPARE);
1327 	spa_config_exit(spa, FTAG);
1328 
1329 	if (error != 0) {
1330 		spa_unload(spa);
1331 		spa_deactivate(spa);
1332 		spa_remove(spa);
1333 		mutex_exit(&spa_namespace_lock);
1334 		return (error);
1335 	}
1336 
1337 	/*
1338 	 * Override any spares as specified by the user, as these may have
1339 	 * correct device names/devids, etc.
1340 	 */
1341 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1342 	    &spares, &nspares) == 0) {
1343 		if (spa->spa_sparelist)
1344 			VERIFY(nvlist_remove(spa->spa_sparelist,
1345 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1346 		else
1347 			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1348 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1349 		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1350 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1351 		spa_config_enter(spa, RW_WRITER, FTAG);
1352 		spa_load_spares(spa);
1353 		spa_config_exit(spa, FTAG);
1354 		spa->spa_sync_spares = B_TRUE;
1355 	}
1356 
1357 	/*
1358 	 * Update the config cache to include the newly-imported pool.
1359 	 */
1360 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1361 
1362 	/*
1363 	 * Resilver anything that's out of date.
1364 	 */
1365 	if (spa_mode & FWRITE)
1366 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1367 
1368 	mutex_exit(&spa_namespace_lock);
1369 
1370 	return (0);
1371 }
1372 
1373 /*
1374  * This (illegal) pool name is used when temporarily importing a spa_t in order
1375  * to get the vdev stats associated with the imported devices.
1376  */
1377 #define	TRYIMPORT_NAME	"$import"
1378 
1379 nvlist_t *
1380 spa_tryimport(nvlist_t *tryconfig)
1381 {
1382 	nvlist_t *config = NULL;
1383 	char *poolname;
1384 	spa_t *spa;
1385 	uint64_t state;
1386 
1387 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
1388 		return (NULL);
1389 
1390 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
1391 		return (NULL);
1392 
1393 	/*
1394 	 * Create and initialize the spa structure.
1395 	 */
1396 	mutex_enter(&spa_namespace_lock);
1397 	spa = spa_add(TRYIMPORT_NAME, NULL);
1398 	spa_activate(spa);
1399 
1400 	/*
1401 	 * Pass off the heavy lifting to spa_load().
1402 	 * Pass TRUE for mosconfig because the user-supplied config
1403 	 * is actually the one to trust when doing an import.
1404 	 */
1405 	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
1406 
1407 	/*
1408 	 * If 'tryconfig' was at least parsable, return the current config.
1409 	 */
1410 	if (spa->spa_root_vdev != NULL) {
1411 		spa_config_enter(spa, RW_READER, FTAG);
1412 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1413 		spa_config_exit(spa, FTAG);
1414 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
1415 		    poolname) == 0);
1416 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1417 		    state) == 0);
1418 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
1419 		    spa->spa_uberblock.ub_timestamp) == 0);
1420 
1421 		/*
1422 		 * Add the list of hot spares.
1423 		 */
1424 		spa_add_spares(spa, config);
1425 	}
1426 
1427 	spa_unload(spa);
1428 	spa_deactivate(spa);
1429 	spa_remove(spa);
1430 	mutex_exit(&spa_namespace_lock);
1431 
1432 	return (config);
1433 }
1434 
1435 /*
1436  * Pool export/destroy
1437  *
1438  * The act of destroying or exporting a pool is very simple.  We make sure there
1439  * is no more pending I/O and any references to the pool are gone.  Then, we
1440  * update the pool state and sync all the labels to disk, removing the
1441  * configuration from the cache afterwards.
1442  */
1443 static int
1444 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig)
1445 {
1446 	spa_t *spa;
1447 
1448 	if (oldconfig)
1449 		*oldconfig = NULL;
1450 
1451 	if (!(spa_mode & FWRITE))
1452 		return (EROFS);
1453 
1454 	mutex_enter(&spa_namespace_lock);
1455 	if ((spa = spa_lookup(pool)) == NULL) {
1456 		mutex_exit(&spa_namespace_lock);
1457 		return (ENOENT);
1458 	}
1459 
1460 	/*
1461 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
1462 	 * reacquire the namespace lock, and see if we can export.
1463 	 */
1464 	spa_open_ref(spa, FTAG);
1465 	mutex_exit(&spa_namespace_lock);
1466 	spa_async_suspend(spa);
1467 	mutex_enter(&spa_namespace_lock);
1468 	spa_close(spa, FTAG);
1469 
1470 	/*
1471 	 * The pool will be in core if it's openable,
1472 	 * in which case we can modify its state.
1473 	 */
1474 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
1475 		/*
1476 		 * Objsets may be open only because they're dirty, so we
1477 		 * have to force it to sync before checking spa_refcnt.
1478 		 */
1479 		spa_scrub_suspend(spa);
1480 		txg_wait_synced(spa->spa_dsl_pool, 0);
1481 
1482 		/*
1483 		 * A pool cannot be exported or destroyed if there are active
1484 		 * references.  If we are resetting a pool, allow references by
1485 		 * fault injection handlers.
1486 		 */
1487 		if (!spa_refcount_zero(spa) ||
1488 		    (spa->spa_inject_ref != 0 &&
1489 		    new_state != POOL_STATE_UNINITIALIZED)) {
1490 			spa_scrub_resume(spa);
1491 			spa_async_resume(spa);
1492 			mutex_exit(&spa_namespace_lock);
1493 			return (EBUSY);
1494 		}
1495 
1496 		spa_scrub_resume(spa);
1497 		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
1498 
1499 		/*
1500 		 * We want this to be reflected on every label,
1501 		 * so mark them all dirty.  spa_unload() will do the
1502 		 * final sync that pushes these changes out.
1503 		 */
1504 		if (new_state != POOL_STATE_UNINITIALIZED) {
1505 			spa_config_enter(spa, RW_WRITER, FTAG);
1506 			spa->spa_state = new_state;
1507 			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
1508 			vdev_config_dirty(spa->spa_root_vdev);
1509 			spa_config_exit(spa, FTAG);
1510 		}
1511 	}
1512 
1513 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
1514 
1515 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1516 		spa_unload(spa);
1517 		spa_deactivate(spa);
1518 	}
1519 
1520 	if (oldconfig && spa->spa_config)
1521 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
1522 
1523 	if (new_state != POOL_STATE_UNINITIALIZED) {
1524 		spa_remove(spa);
1525 		spa_config_sync();
1526 	}
1527 	mutex_exit(&spa_namespace_lock);
1528 
1529 	return (0);
1530 }
1531 
1532 /*
1533  * Destroy a storage pool.
1534  */
1535 int
1536 spa_destroy(char *pool)
1537 {
1538 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL));
1539 }
1540 
1541 /*
1542  * Export a storage pool.
1543  */
1544 int
1545 spa_export(char *pool, nvlist_t **oldconfig)
1546 {
1547 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig));
1548 }
1549 
1550 /*
1551  * Similar to spa_export(), this unloads the spa_t without actually removing it
1552  * from the namespace in any way.
1553  */
1554 int
1555 spa_reset(char *pool)
1556 {
1557 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL));
1558 }
1559 
1560 
1561 /*
1562  * ==========================================================================
1563  * Device manipulation
1564  * ==========================================================================
1565  */
1566 
1567 /*
1568  * Add capacity to a storage pool.
1569  */
1570 int
1571 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1572 {
1573 	uint64_t txg;
1574 	int c, error;
1575 	vdev_t *rvd = spa->spa_root_vdev;
1576 	vdev_t *vd, *tvd;
1577 	nvlist_t **spares;
1578 	uint_t i, nspares;
1579 
1580 	txg = spa_vdev_enter(spa);
1581 
1582 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
1583 	    VDEV_ALLOC_ADD)) != 0)
1584 		return (spa_vdev_exit(spa, NULL, txg, error));
1585 
1586 	spa->spa_pending_vdev = vd;
1587 
1588 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1589 	    &spares, &nspares) != 0)
1590 		nspares = 0;
1591 
1592 	if (vd->vdev_children == 0 && nspares == 0) {
1593 		spa->spa_pending_vdev = NULL;
1594 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1595 	}
1596 
1597 	if (vd->vdev_children != 0) {
1598 		if ((error = vdev_create(vd, txg, B_FALSE)) != 0) {
1599 			spa->spa_pending_vdev = NULL;
1600 			return (spa_vdev_exit(spa, vd, txg, error));
1601 		}
1602 	}
1603 
1604 	/*
1605 	 * We must validate the spares after checking the children.  Otherwise,
1606 	 * vdev_inuse() will blindly overwrite the spare.
1607 	 */
1608 	if ((error = spa_validate_spares(spa, nvroot, txg,
1609 	    VDEV_ALLOC_ADD)) != 0) {
1610 		spa->spa_pending_vdev = NULL;
1611 		return (spa_vdev_exit(spa, vd, txg, error));
1612 	}
1613 
1614 	spa->spa_pending_vdev = NULL;
1615 
1616 	/*
1617 	 * Transfer each new top-level vdev from vd to rvd.
1618 	 */
1619 	for (c = 0; c < vd->vdev_children; c++) {
1620 		tvd = vd->vdev_child[c];
1621 		vdev_remove_child(vd, tvd);
1622 		tvd->vdev_id = rvd->vdev_children;
1623 		vdev_add_child(rvd, tvd);
1624 		vdev_config_dirty(tvd);
1625 	}
1626 
1627 	if (nspares != 0) {
1628 		if (spa->spa_sparelist != NULL) {
1629 			nvlist_t **oldspares;
1630 			uint_t oldnspares;
1631 			nvlist_t **newspares;
1632 
1633 			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
1634 			    ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0);
1635 
1636 			newspares = kmem_alloc(sizeof (void *) *
1637 			    (nspares + oldnspares), KM_SLEEP);
1638 			for (i = 0; i < oldnspares; i++)
1639 				VERIFY(nvlist_dup(oldspares[i],
1640 				    &newspares[i], KM_SLEEP) == 0);
1641 			for (i = 0; i < nspares; i++)
1642 				VERIFY(nvlist_dup(spares[i],
1643 				    &newspares[i + oldnspares],
1644 				    KM_SLEEP) == 0);
1645 
1646 			VERIFY(nvlist_remove(spa->spa_sparelist,
1647 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1648 
1649 			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1650 			    ZPOOL_CONFIG_SPARES, newspares,
1651 			    nspares + oldnspares) == 0);
1652 			for (i = 0; i < oldnspares + nspares; i++)
1653 				nvlist_free(newspares[i]);
1654 			kmem_free(newspares, (oldnspares + nspares) *
1655 			    sizeof (void *));
1656 		} else {
1657 			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1658 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1659 			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1660 			    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1661 		}
1662 
1663 		spa_load_spares(spa);
1664 		spa->spa_sync_spares = B_TRUE;
1665 	}
1666 
1667 	/*
1668 	 * We have to be careful when adding new vdevs to an existing pool.
1669 	 * If other threads start allocating from these vdevs before we
1670 	 * sync the config cache, and we lose power, then upon reboot we may
1671 	 * fail to open the pool because there are DVAs that the config cache
1672 	 * can't translate.  Therefore, we first add the vdevs without
1673 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
1674 	 * and then let spa_config_update() initialize the new metaslabs.
1675 	 *
1676 	 * spa_load() checks for added-but-not-initialized vdevs, so that
1677 	 * if we lose power at any point in this sequence, the remaining
1678 	 * steps will be completed the next time we load the pool.
1679 	 */
1680 	(void) spa_vdev_exit(spa, vd, txg, 0);
1681 
1682 	mutex_enter(&spa_namespace_lock);
1683 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1684 	mutex_exit(&spa_namespace_lock);
1685 
1686 	return (0);
1687 }
1688 
1689 /*
1690  * Attach a device to a mirror.  The arguments are the path to any device
1691  * in the mirror, and the nvroot for the new device.  If the path specifies
1692  * a device that is not mirrored, we automatically insert the mirror vdev.
1693  *
1694  * If 'replacing' is specified, the new device is intended to replace the
1695  * existing device; in this case the two devices are made into their own
1696  * mirror using the 'replacing' vdev, which is functionally identical to
1697  * the mirror vdev (it actually reuses all the same ops) but has a few
1698  * extra rules: you can't attach to it after it's been created, and upon
1699  * completion of resilvering, the first disk (the one being replaced)
1700  * is automatically detached.
1701  */
1702 int
1703 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
1704 {
1705 	uint64_t txg, open_txg;
1706 	int error;
1707 	vdev_t *rvd = spa->spa_root_vdev;
1708 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
1709 	vdev_ops_t *pvops;
1710 
1711 	txg = spa_vdev_enter(spa);
1712 
1713 	oldvd = vdev_lookup_by_guid(rvd, guid);
1714 
1715 	if (oldvd == NULL)
1716 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1717 
1718 	if (!oldvd->vdev_ops->vdev_op_leaf)
1719 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1720 
1721 	pvd = oldvd->vdev_parent;
1722 
1723 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
1724 	    VDEV_ALLOC_ADD)) != 0)
1725 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
1726 
1727 	if (newrootvd->vdev_children != 1)
1728 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1729 
1730 	newvd = newrootvd->vdev_child[0];
1731 
1732 	if (!newvd->vdev_ops->vdev_op_leaf)
1733 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1734 
1735 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
1736 		return (spa_vdev_exit(spa, newrootvd, txg, error));
1737 
1738 	if (!replacing) {
1739 		/*
1740 		 * For attach, the only allowable parent is a mirror or the root
1741 		 * vdev.
1742 		 */
1743 		if (pvd->vdev_ops != &vdev_mirror_ops &&
1744 		    pvd->vdev_ops != &vdev_root_ops)
1745 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1746 
1747 		pvops = &vdev_mirror_ops;
1748 	} else {
1749 		/*
1750 		 * Active hot spares can only be replaced by inactive hot
1751 		 * spares.
1752 		 */
1753 		if (pvd->vdev_ops == &vdev_spare_ops &&
1754 		    pvd->vdev_child[1] == oldvd &&
1755 		    !spa_has_spare(spa, newvd->vdev_guid))
1756 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1757 
1758 		/*
1759 		 * If the source is a hot spare, and the parent isn't already a
1760 		 * spare, then we want to create a new hot spare.  Otherwise, we
1761 		 * want to create a replacing vdev.  The user is not allowed to
1762 		 * attach to a spared vdev child unless the 'isspare' state is
1763 		 * the same (spare replaces spare, non-spare replaces
1764 		 * non-spare).
1765 		 */
1766 		if (pvd->vdev_ops == &vdev_replacing_ops)
1767 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1768 		else if (pvd->vdev_ops == &vdev_spare_ops &&
1769 		    newvd->vdev_isspare != oldvd->vdev_isspare)
1770 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1771 		else if (pvd->vdev_ops != &vdev_spare_ops &&
1772 		    newvd->vdev_isspare)
1773 			pvops = &vdev_spare_ops;
1774 		else
1775 			pvops = &vdev_replacing_ops;
1776 	}
1777 
1778 	/*
1779 	 * Compare the new device size with the replaceable/attachable
1780 	 * device size.
1781 	 */
1782 	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
1783 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
1784 
1785 	/*
1786 	 * The new device cannot have a higher alignment requirement
1787 	 * than the top-level vdev.
1788 	 */
1789 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
1790 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
1791 
1792 	/*
1793 	 * If this is an in-place replacement, update oldvd's path and devid
1794 	 * to make it distinguishable from newvd, and unopenable from now on.
1795 	 */
1796 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
1797 		spa_strfree(oldvd->vdev_path);
1798 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
1799 		    KM_SLEEP);
1800 		(void) sprintf(oldvd->vdev_path, "%s/%s",
1801 		    newvd->vdev_path, "old");
1802 		if (oldvd->vdev_devid != NULL) {
1803 			spa_strfree(oldvd->vdev_devid);
1804 			oldvd->vdev_devid = NULL;
1805 		}
1806 	}
1807 
1808 	/*
1809 	 * If the parent is not a mirror, or if we're replacing, insert the new
1810 	 * mirror/replacing/spare vdev above oldvd.
1811 	 */
1812 	if (pvd->vdev_ops != pvops)
1813 		pvd = vdev_add_parent(oldvd, pvops);
1814 
1815 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
1816 	ASSERT(pvd->vdev_ops == pvops);
1817 	ASSERT(oldvd->vdev_parent == pvd);
1818 
1819 	/*
1820 	 * Extract the new device from its root and add it to pvd.
1821 	 */
1822 	vdev_remove_child(newrootvd, newvd);
1823 	newvd->vdev_id = pvd->vdev_children;
1824 	vdev_add_child(pvd, newvd);
1825 
1826 	/*
1827 	 * If newvd is smaller than oldvd, but larger than its rsize,
1828 	 * the addition of newvd may have decreased our parent's asize.
1829 	 */
1830 	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
1831 
1832 	tvd = newvd->vdev_top;
1833 	ASSERT(pvd->vdev_top == tvd);
1834 	ASSERT(tvd->vdev_parent == rvd);
1835 
1836 	vdev_config_dirty(tvd);
1837 
1838 	/*
1839 	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
1840 	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
1841 	 */
1842 	open_txg = txg + TXG_CONCURRENT_STATES - 1;
1843 
1844 	mutex_enter(&newvd->vdev_dtl_lock);
1845 	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
1846 	    open_txg - TXG_INITIAL + 1);
1847 	mutex_exit(&newvd->vdev_dtl_lock);
1848 
1849 	if (newvd->vdev_isspare)
1850 		spa_spare_activate(newvd);
1851 
1852 	/*
1853 	 * Mark newvd's DTL dirty in this txg.
1854 	 */
1855 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
1856 
1857 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
1858 
1859 	/*
1860 	 * Kick off a resilver to update newvd.  We need to grab the namespace
1861 	 * lock because spa_scrub() needs to post a sysevent with the pool name.
1862 	 */
1863 	mutex_enter(&spa_namespace_lock);
1864 	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1865 	mutex_exit(&spa_namespace_lock);
1866 
1867 	return (0);
1868 }
1869 
1870 /*
1871  * Detach a device from a mirror or replacing vdev.
1872  * If 'replace_done' is specified, only detach if the parent
1873  * is a replacing vdev.
1874  */
1875 int
1876 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
1877 {
1878 	uint64_t txg;
1879 	int c, t, error;
1880 	vdev_t *rvd = spa->spa_root_vdev;
1881 	vdev_t *vd, *pvd, *cvd, *tvd;
1882 	boolean_t unspare = B_FALSE;
1883 	uint64_t unspare_guid;
1884 
1885 	txg = spa_vdev_enter(spa);
1886 
1887 	vd = vdev_lookup_by_guid(rvd, guid);
1888 
1889 	if (vd == NULL)
1890 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1891 
1892 	if (!vd->vdev_ops->vdev_op_leaf)
1893 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1894 
1895 	pvd = vd->vdev_parent;
1896 
1897 	/*
1898 	 * If replace_done is specified, only remove this device if it's
1899 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
1900 	 * disk can be removed.
1901 	 */
1902 	if (replace_done) {
1903 		if (pvd->vdev_ops == &vdev_replacing_ops) {
1904 			if (vd->vdev_id != 0)
1905 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1906 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
1907 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1908 		}
1909 	}
1910 
1911 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
1912 	    spa_version(spa) >= ZFS_VERSION_SPARES);
1913 
1914 	/*
1915 	 * Only mirror, replacing, and spare vdevs support detach.
1916 	 */
1917 	if (pvd->vdev_ops != &vdev_replacing_ops &&
1918 	    pvd->vdev_ops != &vdev_mirror_ops &&
1919 	    pvd->vdev_ops != &vdev_spare_ops)
1920 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1921 
1922 	/*
1923 	 * If there's only one replica, you can't detach it.
1924 	 */
1925 	if (pvd->vdev_children <= 1)
1926 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1927 
1928 	/*
1929 	 * If all siblings have non-empty DTLs, this device may have the only
1930 	 * valid copy of the data, which means we cannot safely detach it.
1931 	 *
1932 	 * XXX -- as in the vdev_offline() case, we really want a more
1933 	 * precise DTL check.
1934 	 */
1935 	for (c = 0; c < pvd->vdev_children; c++) {
1936 		uint64_t dirty;
1937 
1938 		cvd = pvd->vdev_child[c];
1939 		if (cvd == vd)
1940 			continue;
1941 		if (vdev_is_dead(cvd))
1942 			continue;
1943 		mutex_enter(&cvd->vdev_dtl_lock);
1944 		dirty = cvd->vdev_dtl_map.sm_space |
1945 		    cvd->vdev_dtl_scrub.sm_space;
1946 		mutex_exit(&cvd->vdev_dtl_lock);
1947 		if (!dirty)
1948 			break;
1949 	}
1950 
1951 	/*
1952 	 * If we are a replacing or spare vdev, then we can always detach the
1953 	 * latter child, as that is how one cancels the operation.
1954 	 */
1955 	if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) &&
1956 	    c == pvd->vdev_children)
1957 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1958 
1959 	/*
1960 	 * If we are detaching the original disk from a spare, then it implies
1961 	 * that the spare should become a real disk, and be removed from the
1962 	 * active spare list for the pool.
1963 	 */
1964 	if (pvd->vdev_ops == &vdev_spare_ops &&
1965 	    vd->vdev_id == 0)
1966 		unspare = B_TRUE;
1967 
1968 	/*
1969 	 * Erase the disk labels so the disk can be used for other things.
1970 	 * This must be done after all other error cases are handled,
1971 	 * but before we disembowel vd (so we can still do I/O to it).
1972 	 * But if we can't do it, don't treat the error as fatal --
1973 	 * it may be that the unwritability of the disk is the reason
1974 	 * it's being detached!
1975 	 */
1976 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
1977 
1978 	/*
1979 	 * Remove vd from its parent and compact the parent's children.
1980 	 */
1981 	vdev_remove_child(pvd, vd);
1982 	vdev_compact_children(pvd);
1983 
1984 	/*
1985 	 * Remember one of the remaining children so we can get tvd below.
1986 	 */
1987 	cvd = pvd->vdev_child[0];
1988 
1989 	/*
1990 	 * If we need to remove the remaining child from the list of hot spares,
1991 	 * do it now, marking the vdev as no longer a spare in the process.  We
1992 	 * must do this before vdev_remove_parent(), because that can change the
1993 	 * GUID if it creates a new toplevel GUID.
1994 	 */
1995 	if (unspare) {
1996 		ASSERT(cvd->vdev_isspare);
1997 		spa_spare_remove(cvd);
1998 		unspare_guid = cvd->vdev_guid;
1999 	}
2000 
2001 	/*
2002 	 * If the parent mirror/replacing vdev only has one child,
2003 	 * the parent is no longer needed.  Remove it from the tree.
2004 	 */
2005 	if (pvd->vdev_children == 1)
2006 		vdev_remove_parent(cvd);
2007 
2008 	/*
2009 	 * We don't set tvd until now because the parent we just removed
2010 	 * may have been the previous top-level vdev.
2011 	 */
2012 	tvd = cvd->vdev_top;
2013 	ASSERT(tvd->vdev_parent == rvd);
2014 
2015 	/*
2016 	 * Reevaluate the parent vdev state.
2017 	 */
2018 	vdev_propagate_state(cvd);
2019 
2020 	/*
2021 	 * If the device we just detached was smaller than the others, it may be
2022 	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
2023 	 * can't fail because the existing metaslabs are already in core, so
2024 	 * there's nothing to read from disk.
2025 	 */
2026 	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
2027 
2028 	vdev_config_dirty(tvd);
2029 
2030 	/*
2031 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
2032 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
2033 	 * But first make sure we're not on any *other* txg's DTL list, to
2034 	 * prevent vd from being accessed after it's freed.
2035 	 */
2036 	for (t = 0; t < TXG_SIZE; t++)
2037 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
2038 	vd->vdev_detached = B_TRUE;
2039 	vdev_dirty(tvd, VDD_DTL, vd, txg);
2040 
2041 	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
2042 
2043 	error = spa_vdev_exit(spa, vd, txg, 0);
2044 
2045 	/*
2046 	 * If this was the removal of the original device in a hot spare vdev,
2047 	 * then we want to go through and remove the device from the hot spare
2048 	 * list of every other pool.
2049 	 */
2050 	if (unspare) {
2051 		spa = NULL;
2052 		mutex_enter(&spa_namespace_lock);
2053 		while ((spa = spa_next(spa)) != NULL) {
2054 			if (spa->spa_state != POOL_STATE_ACTIVE)
2055 				continue;
2056 
2057 			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
2058 		}
2059 		mutex_exit(&spa_namespace_lock);
2060 	}
2061 
2062 	return (error);
2063 }
2064 
2065 /*
2066  * Remove a device from the pool.  Currently, this supports removing only hot
2067  * spares.
2068  */
2069 int
2070 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
2071 {
2072 	vdev_t *vd;
2073 	nvlist_t **spares, *nv, **newspares;
2074 	uint_t i, j, nspares;
2075 	int ret = 0;
2076 
2077 	spa_config_enter(spa, RW_WRITER, FTAG);
2078 
2079 	vd = spa_lookup_by_guid(spa, guid);
2080 
2081 	nv = NULL;
2082 	if (spa->spa_spares != NULL &&
2083 	    nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2084 	    &spares, &nspares) == 0) {
2085 		for (i = 0; i < nspares; i++) {
2086 			uint64_t theguid;
2087 
2088 			VERIFY(nvlist_lookup_uint64(spares[i],
2089 			    ZPOOL_CONFIG_GUID, &theguid) == 0);
2090 			if (theguid == guid) {
2091 				nv = spares[i];
2092 				break;
2093 			}
2094 		}
2095 	}
2096 
2097 	/*
2098 	 * We only support removing a hot spare, and only if it's not currently
2099 	 * in use in this pool.
2100 	 */
2101 	if (nv == NULL && vd == NULL) {
2102 		ret = ENOENT;
2103 		goto out;
2104 	}
2105 
2106 	if (nv == NULL && vd != NULL) {
2107 		ret = ENOTSUP;
2108 		goto out;
2109 	}
2110 
2111 	if (!unspare && nv != NULL && vd != NULL) {
2112 		ret = EBUSY;
2113 		goto out;
2114 	}
2115 
2116 	if (nspares == 1) {
2117 		newspares = NULL;
2118 	} else {
2119 		newspares = kmem_alloc((nspares - 1) * sizeof (void *),
2120 		    KM_SLEEP);
2121 		for (i = 0, j = 0; i < nspares; i++) {
2122 			if (spares[i] != nv)
2123 				VERIFY(nvlist_dup(spares[i],
2124 				    &newspares[j++], KM_SLEEP) == 0);
2125 		}
2126 	}
2127 
2128 	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2129 	    DATA_TYPE_NVLIST_ARRAY) == 0);
2130 	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2131 	    newspares, nspares - 1) == 0);
2132 	for (i = 0; i < nspares - 1; i++)
2133 		nvlist_free(newspares[i]);
2134 	kmem_free(newspares, (nspares - 1) * sizeof (void *));
2135 	spa_load_spares(spa);
2136 	spa->spa_sync_spares = B_TRUE;
2137 
2138 out:
2139 	spa_config_exit(spa, FTAG);
2140 
2141 	return (ret);
2142 }
2143 
2144 /*
2145  * Find any device that's done replacing, or a vdev marked 'unspare' that's
2146  * current spared, so we can detach it.
2147  */
2148 static vdev_t *
2149 spa_vdev_resilver_done_hunt(vdev_t *vd)
2150 {
2151 	vdev_t *newvd, *oldvd;
2152 	int c;
2153 
2154 	for (c = 0; c < vd->vdev_children; c++) {
2155 		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
2156 		if (oldvd != NULL)
2157 			return (oldvd);
2158 	}
2159 
2160 	/*
2161 	 * Check for a completed replacement.
2162 	 */
2163 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
2164 		oldvd = vd->vdev_child[0];
2165 		newvd = vd->vdev_child[1];
2166 
2167 		mutex_enter(&newvd->vdev_dtl_lock);
2168 		if (newvd->vdev_dtl_map.sm_space == 0 &&
2169 		    newvd->vdev_dtl_scrub.sm_space == 0) {
2170 			mutex_exit(&newvd->vdev_dtl_lock);
2171 			return (oldvd);
2172 		}
2173 		mutex_exit(&newvd->vdev_dtl_lock);
2174 	}
2175 
2176 	/*
2177 	 * Check for a completed resilver with the 'unspare' flag set.
2178 	 */
2179 	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
2180 		newvd = vd->vdev_child[0];
2181 		oldvd = vd->vdev_child[1];
2182 
2183 		mutex_enter(&newvd->vdev_dtl_lock);
2184 		if (newvd->vdev_unspare &&
2185 		    newvd->vdev_dtl_map.sm_space == 0 &&
2186 		    newvd->vdev_dtl_scrub.sm_space == 0) {
2187 			newvd->vdev_unspare = 0;
2188 			mutex_exit(&newvd->vdev_dtl_lock);
2189 			return (oldvd);
2190 		}
2191 		mutex_exit(&newvd->vdev_dtl_lock);
2192 	}
2193 
2194 	return (NULL);
2195 }
2196 
2197 static void
2198 spa_vdev_resilver_done(spa_t *spa)
2199 {
2200 	vdev_t *vd;
2201 	vdev_t *pvd;
2202 	uint64_t guid;
2203 	uint64_t pguid = 0;
2204 
2205 	spa_config_enter(spa, RW_READER, FTAG);
2206 
2207 	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
2208 		guid = vd->vdev_guid;
2209 		/*
2210 		 * If we have just finished replacing a hot spared device, then
2211 		 * we need to detach the parent's first child (the original hot
2212 		 * spare) as well.
2213 		 */
2214 		pvd = vd->vdev_parent;
2215 		if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2216 		    pvd->vdev_id == 0) {
2217 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
2218 			ASSERT(pvd->vdev_parent->vdev_children == 2);
2219 			pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
2220 		}
2221 		spa_config_exit(spa, FTAG);
2222 		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
2223 			return;
2224 		if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
2225 			return;
2226 		spa_config_enter(spa, RW_READER, FTAG);
2227 	}
2228 
2229 	spa_config_exit(spa, FTAG);
2230 }
2231 
2232 /*
2233  * Update the stored path for this vdev.  Dirty the vdev configuration, relying
2234  * on spa_vdev_enter/exit() to synchronize the labels and cache.
2235  */
2236 int
2237 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
2238 {
2239 	vdev_t *rvd, *vd;
2240 	uint64_t txg;
2241 
2242 	rvd = spa->spa_root_vdev;
2243 
2244 	txg = spa_vdev_enter(spa);
2245 
2246 	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2247 		/*
2248 		 * Determine if this is a reference to a hot spare.  In that
2249 		 * case, update the path as stored in the spare list.
2250 		 */
2251 		nvlist_t **spares;
2252 		uint_t i, nspares;
2253 		if (spa->spa_sparelist != NULL) {
2254 			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
2255 			    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2256 			for (i = 0; i < nspares; i++) {
2257 				uint64_t theguid;
2258 				VERIFY(nvlist_lookup_uint64(spares[i],
2259 				    ZPOOL_CONFIG_GUID, &theguid) == 0);
2260 				if (theguid == guid)
2261 					break;
2262 			}
2263 
2264 			if (i == nspares)
2265 				return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2266 
2267 			VERIFY(nvlist_add_string(spares[i],
2268 			    ZPOOL_CONFIG_PATH, newpath) == 0);
2269 			spa_load_spares(spa);
2270 			spa->spa_sync_spares = B_TRUE;
2271 			return (spa_vdev_exit(spa, NULL, txg, 0));
2272 		} else {
2273 			return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2274 		}
2275 	}
2276 
2277 	if (!vd->vdev_ops->vdev_op_leaf)
2278 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2279 
2280 	spa_strfree(vd->vdev_path);
2281 	vd->vdev_path = spa_strdup(newpath);
2282 
2283 	vdev_config_dirty(vd->vdev_top);
2284 
2285 	return (spa_vdev_exit(spa, NULL, txg, 0));
2286 }
2287 
2288 /*
2289  * ==========================================================================
2290  * SPA Scrubbing
2291  * ==========================================================================
2292  */
2293 
2294 static void
2295 spa_scrub_io_done(zio_t *zio)
2296 {
2297 	spa_t *spa = zio->io_spa;
2298 
2299 	arc_data_buf_free(zio->io_data, zio->io_size);
2300 
2301 	mutex_enter(&spa->spa_scrub_lock);
2302 	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
2303 		vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev;
2304 		spa->spa_scrub_errors++;
2305 		mutex_enter(&vd->vdev_stat_lock);
2306 		vd->vdev_stat.vs_scrub_errors++;
2307 		mutex_exit(&vd->vdev_stat_lock);
2308 	}
2309 
2310 	if (--spa->spa_scrub_inflight < spa->spa_scrub_maxinflight)
2311 		cv_broadcast(&spa->spa_scrub_io_cv);
2312 
2313 	ASSERT(spa->spa_scrub_inflight >= 0);
2314 
2315 	mutex_exit(&spa->spa_scrub_lock);
2316 }
2317 
2318 static void
2319 spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
2320     zbookmark_t *zb)
2321 {
2322 	size_t size = BP_GET_LSIZE(bp);
2323 	void *data;
2324 
2325 	mutex_enter(&spa->spa_scrub_lock);
2326 	/*
2327 	 * Do not give too much work to vdev(s).
2328 	 */
2329 	while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight) {
2330 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2331 	}
2332 	spa->spa_scrub_inflight++;
2333 	mutex_exit(&spa->spa_scrub_lock);
2334 
2335 	data = arc_data_buf_alloc(size);
2336 
2337 	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
2338 		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
2339 
2340 	flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL;
2341 
2342 	zio_nowait(zio_read(NULL, spa, bp, data, size,
2343 	    spa_scrub_io_done, NULL, priority, flags, zb));
2344 }
2345 
2346 /* ARGSUSED */
2347 static int
2348 spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
2349 {
2350 	blkptr_t *bp = &bc->bc_blkptr;
2351 	vdev_t *vd = spa->spa_root_vdev;
2352 	dva_t *dva = bp->blk_dva;
2353 	int needs_resilver = B_FALSE;
2354 	int d;
2355 
2356 	if (bc->bc_errno) {
2357 		/*
2358 		 * We can't scrub this block, but we can continue to scrub
2359 		 * the rest of the pool.  Note the error and move along.
2360 		 */
2361 		mutex_enter(&spa->spa_scrub_lock);
2362 		spa->spa_scrub_errors++;
2363 		mutex_exit(&spa->spa_scrub_lock);
2364 
2365 		mutex_enter(&vd->vdev_stat_lock);
2366 		vd->vdev_stat.vs_scrub_errors++;
2367 		mutex_exit(&vd->vdev_stat_lock);
2368 
2369 		return (ERESTART);
2370 	}
2371 
2372 	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
2373 
2374 	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
2375 		vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]));
2376 
2377 		ASSERT(vd != NULL);
2378 
2379 		/*
2380 		 * Keep track of how much data we've examined so that
2381 		 * zpool(1M) status can make useful progress reports.
2382 		 */
2383 		mutex_enter(&vd->vdev_stat_lock);
2384 		vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]);
2385 		mutex_exit(&vd->vdev_stat_lock);
2386 
2387 		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
2388 			if (DVA_GET_GANG(&dva[d])) {
2389 				/*
2390 				 * Gang members may be spread across multiple
2391 				 * vdevs, so the best we can do is look at the
2392 				 * pool-wide DTL.
2393 				 * XXX -- it would be better to change our
2394 				 * allocation policy to ensure that this can't
2395 				 * happen.
2396 				 */
2397 				vd = spa->spa_root_vdev;
2398 			}
2399 			if (vdev_dtl_contains(&vd->vdev_dtl_map,
2400 			    bp->blk_birth, 1))
2401 				needs_resilver = B_TRUE;
2402 		}
2403 	}
2404 
2405 	if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING)
2406 		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
2407 		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
2408 	else if (needs_resilver)
2409 		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
2410 		    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
2411 
2412 	return (0);
2413 }
2414 
2415 static void
2416 spa_scrub_thread(spa_t *spa)
2417 {
2418 	callb_cpr_t cprinfo;
2419 	traverse_handle_t *th = spa->spa_scrub_th;
2420 	vdev_t *rvd = spa->spa_root_vdev;
2421 	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
2422 	int error = 0;
2423 	boolean_t complete;
2424 
2425 	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
2426 
2427 	/*
2428 	 * If we're restarting due to a snapshot create/delete,
2429 	 * wait for that to complete.
2430 	 */
2431 	txg_wait_synced(spa_get_dsl(spa), 0);
2432 
2433 	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
2434 	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2435 	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
2436 
2437 	spa_config_enter(spa, RW_WRITER, FTAG);
2438 	vdev_reopen(rvd);		/* purge all vdev caches */
2439 	vdev_config_dirty(rvd);		/* rewrite all disk labels */
2440 	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
2441 	spa_config_exit(spa, FTAG);
2442 
2443 	mutex_enter(&spa->spa_scrub_lock);
2444 	spa->spa_scrub_errors = 0;
2445 	spa->spa_scrub_active = 1;
2446 	ASSERT(spa->spa_scrub_inflight == 0);
2447 
2448 	while (!spa->spa_scrub_stop) {
2449 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2450 		while (spa->spa_scrub_suspended) {
2451 			spa->spa_scrub_active = 0;
2452 			cv_broadcast(&spa->spa_scrub_cv);
2453 			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2454 			spa->spa_scrub_active = 1;
2455 		}
2456 		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
2457 
2458 		if (spa->spa_scrub_restart_txg != 0)
2459 			break;
2460 
2461 		mutex_exit(&spa->spa_scrub_lock);
2462 		error = traverse_more(th);
2463 		mutex_enter(&spa->spa_scrub_lock);
2464 		if (error != EAGAIN)
2465 			break;
2466 	}
2467 
2468 	while (spa->spa_scrub_inflight)
2469 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2470 
2471 	spa->spa_scrub_active = 0;
2472 	cv_broadcast(&spa->spa_scrub_cv);
2473 
2474 	mutex_exit(&spa->spa_scrub_lock);
2475 
2476 	spa_config_enter(spa, RW_WRITER, FTAG);
2477 
2478 	mutex_enter(&spa->spa_scrub_lock);
2479 
2480 	/*
2481 	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
2482 	 * AND the spa config lock to synchronize with any config changes
2483 	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
2484 	 */
2485 	if (spa->spa_scrub_restart_txg != 0)
2486 		error = ERESTART;
2487 
2488 	if (spa->spa_scrub_stop)
2489 		error = EINTR;
2490 
2491 	/*
2492 	 * Even if there were uncorrectable errors, we consider the scrub
2493 	 * completed.  The downside is that if there is a transient error during
2494 	 * a resilver, we won't resilver the data properly to the target.  But
2495 	 * if the damage is permanent (more likely) we will resilver forever,
2496 	 * which isn't really acceptable.  Since there is enough information for
2497 	 * the user to know what has failed and why, this seems like a more
2498 	 * tractable approach.
2499 	 */
2500 	complete = (error == 0);
2501 
2502 	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
2503 	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2504 	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
2505 	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
2506 
2507 	mutex_exit(&spa->spa_scrub_lock);
2508 
2509 	/*
2510 	 * If the scrub/resilver completed, update all DTLs to reflect this.
2511 	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
2512 	 */
2513 	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
2514 	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
2515 	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
2516 	spa_errlog_rotate(spa);
2517 
2518 	if (scrub_type == POOL_SCRUB_RESILVER && complete)
2519 		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_FINISH);
2520 
2521 	spa_config_exit(spa, FTAG);
2522 
2523 	mutex_enter(&spa->spa_scrub_lock);
2524 
2525 	/*
2526 	 * We may have finished replacing a device.
2527 	 * Let the async thread assess this and handle the detach.
2528 	 */
2529 	spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
2530 
2531 	/*
2532 	 * If we were told to restart, our final act is to start a new scrub.
2533 	 */
2534 	if (error == ERESTART)
2535 		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
2536 		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
2537 
2538 	spa->spa_scrub_type = POOL_SCRUB_NONE;
2539 	spa->spa_scrub_active = 0;
2540 	spa->spa_scrub_thread = NULL;
2541 	cv_broadcast(&spa->spa_scrub_cv);
2542 	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
2543 	thread_exit();
2544 }
2545 
2546 void
2547 spa_scrub_suspend(spa_t *spa)
2548 {
2549 	mutex_enter(&spa->spa_scrub_lock);
2550 	spa->spa_scrub_suspended++;
2551 	while (spa->spa_scrub_active) {
2552 		cv_broadcast(&spa->spa_scrub_cv);
2553 		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2554 	}
2555 	while (spa->spa_scrub_inflight)
2556 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2557 	mutex_exit(&spa->spa_scrub_lock);
2558 }
2559 
2560 void
2561 spa_scrub_resume(spa_t *spa)
2562 {
2563 	mutex_enter(&spa->spa_scrub_lock);
2564 	ASSERT(spa->spa_scrub_suspended != 0);
2565 	if (--spa->spa_scrub_suspended == 0)
2566 		cv_broadcast(&spa->spa_scrub_cv);
2567 	mutex_exit(&spa->spa_scrub_lock);
2568 }
2569 
2570 void
2571 spa_scrub_restart(spa_t *spa, uint64_t txg)
2572 {
2573 	/*
2574 	 * Something happened (e.g. snapshot create/delete) that means
2575 	 * we must restart any in-progress scrubs.  The itinerary will
2576 	 * fix this properly.
2577 	 */
2578 	mutex_enter(&spa->spa_scrub_lock);
2579 	spa->spa_scrub_restart_txg = txg;
2580 	mutex_exit(&spa->spa_scrub_lock);
2581 }
2582 
2583 int
2584 spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
2585 {
2586 	space_seg_t *ss;
2587 	uint64_t mintxg, maxtxg;
2588 	vdev_t *rvd = spa->spa_root_vdev;
2589 
2590 	if ((uint_t)type >= POOL_SCRUB_TYPES)
2591 		return (ENOTSUP);
2592 
2593 	mutex_enter(&spa->spa_scrub_lock);
2594 
2595 	/*
2596 	 * If there's a scrub or resilver already in progress, stop it.
2597 	 */
2598 	while (spa->spa_scrub_thread != NULL) {
2599 		/*
2600 		 * Don't stop a resilver unless forced.
2601 		 */
2602 		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
2603 			mutex_exit(&spa->spa_scrub_lock);
2604 			return (EBUSY);
2605 		}
2606 		spa->spa_scrub_stop = 1;
2607 		cv_broadcast(&spa->spa_scrub_cv);
2608 		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2609 	}
2610 
2611 	/*
2612 	 * Terminate the previous traverse.
2613 	 */
2614 	if (spa->spa_scrub_th != NULL) {
2615 		traverse_fini(spa->spa_scrub_th);
2616 		spa->spa_scrub_th = NULL;
2617 	}
2618 
2619 	if (rvd == NULL) {
2620 		ASSERT(spa->spa_scrub_stop == 0);
2621 		ASSERT(spa->spa_scrub_type == type);
2622 		ASSERT(spa->spa_scrub_restart_txg == 0);
2623 		mutex_exit(&spa->spa_scrub_lock);
2624 		return (0);
2625 	}
2626 
2627 	mintxg = TXG_INITIAL - 1;
2628 	maxtxg = spa_last_synced_txg(spa) + 1;
2629 
2630 	mutex_enter(&rvd->vdev_dtl_lock);
2631 
2632 	if (rvd->vdev_dtl_map.sm_space == 0) {
2633 		/*
2634 		 * The pool-wide DTL is empty.
2635 		 * If this is a resilver, there's nothing to do except
2636 		 * check whether any in-progress replacements have completed.
2637 		 */
2638 		if (type == POOL_SCRUB_RESILVER) {
2639 			type = POOL_SCRUB_NONE;
2640 			spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
2641 		}
2642 	} else {
2643 		/*
2644 		 * The pool-wide DTL is non-empty.
2645 		 * If this is a normal scrub, upgrade to a resilver instead.
2646 		 */
2647 		if (type == POOL_SCRUB_EVERYTHING)
2648 			type = POOL_SCRUB_RESILVER;
2649 	}
2650 
2651 	if (type == POOL_SCRUB_RESILVER) {
2652 		/*
2653 		 * Determine the resilvering boundaries.
2654 		 *
2655 		 * Note: (mintxg, maxtxg) is an open interval,
2656 		 * i.e. mintxg and maxtxg themselves are not included.
2657 		 *
2658 		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
2659 		 * so we don't claim to resilver a txg that's still changing.
2660 		 */
2661 		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
2662 		mintxg = ss->ss_start - 1;
2663 		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
2664 		maxtxg = MIN(ss->ss_end, maxtxg);
2665 
2666 		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
2667 	}
2668 
2669 	mutex_exit(&rvd->vdev_dtl_lock);
2670 
2671 	spa->spa_scrub_stop = 0;
2672 	spa->spa_scrub_type = type;
2673 	spa->spa_scrub_restart_txg = 0;
2674 
2675 	if (type != POOL_SCRUB_NONE) {
2676 		spa->spa_scrub_mintxg = mintxg;
2677 		spa->spa_scrub_maxtxg = maxtxg;
2678 		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
2679 		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
2680 		    ZIO_FLAG_CANFAIL);
2681 		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
2682 		spa->spa_scrub_thread = thread_create(NULL, 0,
2683 		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
2684 	}
2685 
2686 	mutex_exit(&spa->spa_scrub_lock);
2687 
2688 	return (0);
2689 }
2690 
2691 /*
2692  * ==========================================================================
2693  * SPA async task processing
2694  * ==========================================================================
2695  */
2696 
2697 static void
2698 spa_async_remove(spa_t *spa, vdev_t *vd)
2699 {
2700 	vdev_t *tvd;
2701 	int c;
2702 
2703 	for (c = 0; c < vd->vdev_children; c++) {
2704 		tvd = vd->vdev_child[c];
2705 		if (tvd->vdev_remove_wanted) {
2706 			tvd->vdev_remove_wanted = 0;
2707 			vdev_set_state(tvd, B_FALSE, VDEV_STATE_REMOVED,
2708 			    VDEV_AUX_NONE);
2709 			vdev_clear(spa, tvd);
2710 			vdev_config_dirty(tvd->vdev_top);
2711 		}
2712 		spa_async_remove(spa, tvd);
2713 	}
2714 }
2715 
2716 static void
2717 spa_async_thread(spa_t *spa)
2718 {
2719 	int tasks;
2720 	uint64_t txg;
2721 
2722 	ASSERT(spa->spa_sync_on);
2723 
2724 	mutex_enter(&spa->spa_async_lock);
2725 	tasks = spa->spa_async_tasks;
2726 	spa->spa_async_tasks = 0;
2727 	mutex_exit(&spa->spa_async_lock);
2728 
2729 	/*
2730 	 * See if the config needs to be updated.
2731 	 */
2732 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
2733 		mutex_enter(&spa_namespace_lock);
2734 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2735 		mutex_exit(&spa_namespace_lock);
2736 	}
2737 
2738 	/*
2739 	 * See if any devices need to be marked REMOVED.
2740 	 */
2741 	if (tasks & SPA_ASYNC_REMOVE) {
2742 		txg = spa_vdev_enter(spa);
2743 		spa_async_remove(spa, spa->spa_root_vdev);
2744 		(void) spa_vdev_exit(spa, NULL, txg, 0);
2745 	}
2746 
2747 	/*
2748 	 * If any devices are done replacing, detach them.
2749 	 */
2750 	if (tasks & SPA_ASYNC_RESILVER_DONE)
2751 		spa_vdev_resilver_done(spa);
2752 
2753 	/*
2754 	 * Kick off a scrub.  When starting a RESILVER scrub (or an EVERYTHING
2755 	 * scrub which can become a resilver), we need to hold
2756 	 * spa_namespace_lock() because the sysevent we post via
2757 	 * spa_event_notify() needs to get the name of the pool.
2758 	 */
2759 	if (tasks & SPA_ASYNC_SCRUB) {
2760 		mutex_enter(&spa_namespace_lock);
2761 		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
2762 		mutex_exit(&spa_namespace_lock);
2763 	}
2764 
2765 	/*
2766 	 * Kick off a resilver.
2767 	 */
2768 	if (tasks & SPA_ASYNC_RESILVER) {
2769 		mutex_enter(&spa_namespace_lock);
2770 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
2771 		mutex_exit(&spa_namespace_lock);
2772 	}
2773 
2774 	/*
2775 	 * Let the world know that we're done.
2776 	 */
2777 	mutex_enter(&spa->spa_async_lock);
2778 	spa->spa_async_thread = NULL;
2779 	cv_broadcast(&spa->spa_async_cv);
2780 	mutex_exit(&spa->spa_async_lock);
2781 	thread_exit();
2782 }
2783 
2784 void
2785 spa_async_suspend(spa_t *spa)
2786 {
2787 	mutex_enter(&spa->spa_async_lock);
2788 	spa->spa_async_suspended++;
2789 	while (spa->spa_async_thread != NULL)
2790 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
2791 	mutex_exit(&spa->spa_async_lock);
2792 }
2793 
2794 void
2795 spa_async_resume(spa_t *spa)
2796 {
2797 	mutex_enter(&spa->spa_async_lock);
2798 	ASSERT(spa->spa_async_suspended != 0);
2799 	spa->spa_async_suspended--;
2800 	mutex_exit(&spa->spa_async_lock);
2801 }
2802 
2803 static void
2804 spa_async_dispatch(spa_t *spa)
2805 {
2806 	mutex_enter(&spa->spa_async_lock);
2807 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
2808 	    spa->spa_async_thread == NULL &&
2809 	    rootdir != NULL && !vn_is_readonly(rootdir))
2810 		spa->spa_async_thread = thread_create(NULL, 0,
2811 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
2812 	mutex_exit(&spa->spa_async_lock);
2813 }
2814 
2815 void
2816 spa_async_request(spa_t *spa, int task)
2817 {
2818 	mutex_enter(&spa->spa_async_lock);
2819 	spa->spa_async_tasks |= task;
2820 	mutex_exit(&spa->spa_async_lock);
2821 }
2822 
2823 /*
2824  * ==========================================================================
2825  * SPA syncing routines
2826  * ==========================================================================
2827  */
2828 
2829 static void
2830 spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
2831 {
2832 	bplist_t *bpl = &spa->spa_sync_bplist;
2833 	dmu_tx_t *tx;
2834 	blkptr_t blk;
2835 	uint64_t itor = 0;
2836 	zio_t *zio;
2837 	int error;
2838 	uint8_t c = 1;
2839 
2840 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
2841 
2842 	while (bplist_iterate(bpl, &itor, &blk) == 0)
2843 		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
2844 
2845 	error = zio_wait(zio);
2846 	ASSERT3U(error, ==, 0);
2847 
2848 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2849 	bplist_vacate(bpl, tx);
2850 
2851 	/*
2852 	 * Pre-dirty the first block so we sync to convergence faster.
2853 	 * (Usually only the first block is needed.)
2854 	 */
2855 	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
2856 	dmu_tx_commit(tx);
2857 }
2858 
2859 static void
2860 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
2861 {
2862 	char *packed = NULL;
2863 	size_t nvsize = 0;
2864 	dmu_buf_t *db;
2865 
2866 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
2867 
2868 	packed = kmem_alloc(nvsize, KM_SLEEP);
2869 
2870 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
2871 	    KM_SLEEP) == 0);
2872 
2873 	dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx);
2874 
2875 	kmem_free(packed, nvsize);
2876 
2877 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
2878 	dmu_buf_will_dirty(db, tx);
2879 	*(uint64_t *)db->db_data = nvsize;
2880 	dmu_buf_rele(db, FTAG);
2881 }
2882 
2883 static void
2884 spa_sync_spares(spa_t *spa, dmu_tx_t *tx)
2885 {
2886 	nvlist_t *nvroot;
2887 	nvlist_t **spares;
2888 	int i;
2889 
2890 	if (!spa->spa_sync_spares)
2891 		return;
2892 
2893 	/*
2894 	 * Update the MOS nvlist describing the list of available spares.
2895 	 * spa_validate_spares() will have already made sure this nvlist is
2896 	 * valid and the vdevs are labeled appropriately.
2897 	 */
2898 	if (spa->spa_spares_object == 0) {
2899 		spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset,
2900 		    DMU_OT_PACKED_NVLIST, 1 << 14,
2901 		    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2902 		VERIFY(zap_update(spa->spa_meta_objset,
2903 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES,
2904 		    sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0);
2905 	}
2906 
2907 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2908 	if (spa->spa_nspares == 0) {
2909 		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2910 		    NULL, 0) == 0);
2911 	} else {
2912 		spares = kmem_alloc(spa->spa_nspares * sizeof (void *),
2913 		    KM_SLEEP);
2914 		for (i = 0; i < spa->spa_nspares; i++)
2915 			spares[i] = vdev_config_generate(spa,
2916 			    spa->spa_spares[i], B_FALSE, B_TRUE);
2917 		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2918 		    spares, spa->spa_nspares) == 0);
2919 		for (i = 0; i < spa->spa_nspares; i++)
2920 			nvlist_free(spares[i]);
2921 		kmem_free(spares, spa->spa_nspares * sizeof (void *));
2922 	}
2923 
2924 	spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx);
2925 	nvlist_free(nvroot);
2926 
2927 	spa->spa_sync_spares = B_FALSE;
2928 }
2929 
2930 static void
2931 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
2932 {
2933 	nvlist_t *config;
2934 
2935 	if (list_is_empty(&spa->spa_dirty_list))
2936 		return;
2937 
2938 	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
2939 
2940 	if (spa->spa_config_syncing)
2941 		nvlist_free(spa->spa_config_syncing);
2942 	spa->spa_config_syncing = config;
2943 
2944 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
2945 }
2946 
2947 static void
2948 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
2949 {
2950 	spa_t *spa = arg1;
2951 	nvlist_t *nvp = arg2;
2952 	nvpair_t *nvpair;
2953 	objset_t *mos = spa->spa_meta_objset;
2954 	uint64_t zapobj;
2955 	uint64_t intval;
2956 
2957 	mutex_enter(&spa->spa_props_lock);
2958 	if (spa->spa_pool_props_object == 0) {
2959 		zapobj = zap_create(mos, DMU_OT_POOL_PROPS, DMU_OT_NONE, 0, tx);
2960 		VERIFY(zapobj > 0);
2961 
2962 		spa->spa_pool_props_object = zapobj;
2963 
2964 		VERIFY(zap_update(mos, DMU_POOL_DIRECTORY_OBJECT,
2965 		    DMU_POOL_PROPS, 8, 1,
2966 		    &spa->spa_pool_props_object, tx) == 0);
2967 	}
2968 	mutex_exit(&spa->spa_props_lock);
2969 
2970 	nvpair = NULL;
2971 	while ((nvpair = nvlist_next_nvpair(nvp, nvpair))) {
2972 		switch (zpool_name_to_prop(nvpair_name(nvpair))) {
2973 		case ZPOOL_PROP_BOOTFS:
2974 			VERIFY(nvlist_lookup_uint64(nvp,
2975 			    nvpair_name(nvpair), &spa->spa_bootfs) == 0);
2976 			VERIFY(zap_update(mos,
2977 			    spa->spa_pool_props_object,
2978 			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), 8, 1,
2979 			    &spa->spa_bootfs, tx) == 0);
2980 			break;
2981 
2982 		case ZPOOL_PROP_AUTOREPLACE:
2983 			VERIFY(nvlist_lookup_uint64(nvp,
2984 			    nvpair_name(nvpair), &intval) == 0);
2985 			VERIFY(zap_update(mos,
2986 			    spa->spa_pool_props_object,
2987 			    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 8, 1,
2988 			    &intval, tx) == 0);
2989 			break;
2990 		}
2991 	}
2992 }
2993 
2994 /*
2995  * Sync the specified transaction group.  New blocks may be dirtied as
2996  * part of the process, so we iterate until it converges.
2997  */
2998 void
2999 spa_sync(spa_t *spa, uint64_t txg)
3000 {
3001 	dsl_pool_t *dp = spa->spa_dsl_pool;
3002 	objset_t *mos = spa->spa_meta_objset;
3003 	bplist_t *bpl = &spa->spa_sync_bplist;
3004 	vdev_t *rvd = spa->spa_root_vdev;
3005 	vdev_t *vd;
3006 	dmu_tx_t *tx;
3007 	int dirty_vdevs;
3008 
3009 	/*
3010 	 * Lock out configuration changes.
3011 	 */
3012 	spa_config_enter(spa, RW_READER, FTAG);
3013 
3014 	spa->spa_syncing_txg = txg;
3015 	spa->spa_sync_pass = 0;
3016 
3017 	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
3018 
3019 	tx = dmu_tx_create_assigned(dp, txg);
3020 
3021 	/*
3022 	 * If we are upgrading to ZFS_VERSION_RAIDZ_DEFLATE this txg,
3023 	 * set spa_deflate if we have no raid-z vdevs.
3024 	 */
3025 	if (spa->spa_ubsync.ub_version < ZFS_VERSION_RAIDZ_DEFLATE &&
3026 	    spa->spa_uberblock.ub_version >= ZFS_VERSION_RAIDZ_DEFLATE) {
3027 		int i;
3028 
3029 		for (i = 0; i < rvd->vdev_children; i++) {
3030 			vd = rvd->vdev_child[i];
3031 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
3032 				break;
3033 		}
3034 		if (i == rvd->vdev_children) {
3035 			spa->spa_deflate = TRUE;
3036 			VERIFY(0 == zap_add(spa->spa_meta_objset,
3037 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3038 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
3039 		}
3040 	}
3041 
3042 	/*
3043 	 * If anything has changed in this txg, push the deferred frees
3044 	 * from the previous txg.  If not, leave them alone so that we
3045 	 * don't generate work on an otherwise idle system.
3046 	 */
3047 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
3048 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
3049 	    !txg_list_empty(&dp->dp_sync_tasks, txg))
3050 		spa_sync_deferred_frees(spa, txg);
3051 
3052 	/*
3053 	 * Iterate to convergence.
3054 	 */
3055 	do {
3056 		spa->spa_sync_pass++;
3057 
3058 		spa_sync_config_object(spa, tx);
3059 		spa_sync_spares(spa, tx);
3060 		spa_errlog_sync(spa, txg);
3061 		dsl_pool_sync(dp, txg);
3062 
3063 		dirty_vdevs = 0;
3064 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
3065 			vdev_sync(vd, txg);
3066 			dirty_vdevs++;
3067 		}
3068 
3069 		bplist_sync(bpl, tx);
3070 	} while (dirty_vdevs);
3071 
3072 	bplist_close(bpl);
3073 
3074 	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
3075 
3076 	/*
3077 	 * Rewrite the vdev configuration (which includes the uberblock)
3078 	 * to commit the transaction group.
3079 	 *
3080 	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
3081 	 * Otherwise, pick a random top-level vdev that's known to be
3082 	 * visible in the config cache (see spa_vdev_add() for details).
3083 	 * If the write fails, try the next vdev until we're tried them all.
3084 	 */
3085 	if (!list_is_empty(&spa->spa_dirty_list)) {
3086 		VERIFY(vdev_config_sync(rvd, txg) == 0);
3087 	} else {
3088 		int children = rvd->vdev_children;
3089 		int c0 = spa_get_random(children);
3090 		int c;
3091 
3092 		for (c = 0; c < children; c++) {
3093 			vd = rvd->vdev_child[(c0 + c) % children];
3094 			if (vd->vdev_ms_array == 0)
3095 				continue;
3096 			if (vdev_config_sync(vd, txg) == 0)
3097 				break;
3098 		}
3099 		if (c == children)
3100 			VERIFY(vdev_config_sync(rvd, txg) == 0);
3101 	}
3102 
3103 	dmu_tx_commit(tx);
3104 
3105 	/*
3106 	 * Clear the dirty config list.
3107 	 */
3108 	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
3109 		vdev_config_clean(vd);
3110 
3111 	/*
3112 	 * Now that the new config has synced transactionally,
3113 	 * let it become visible to the config cache.
3114 	 */
3115 	if (spa->spa_config_syncing != NULL) {
3116 		spa_config_set(spa, spa->spa_config_syncing);
3117 		spa->spa_config_txg = txg;
3118 		spa->spa_config_syncing = NULL;
3119 	}
3120 
3121 	/*
3122 	 * Make a stable copy of the fully synced uberblock.
3123 	 * We use this as the root for pool traversals.
3124 	 */
3125 	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
3126 
3127 	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
3128 
3129 	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
3130 	spa->spa_traverse_wanted = 0;
3131 	spa->spa_ubsync = spa->spa_uberblock;
3132 	rw_exit(&spa->spa_traverse_lock);
3133 
3134 	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
3135 
3136 	/*
3137 	 * Clean up the ZIL records for the synced txg.
3138 	 */
3139 	dsl_pool_zil_clean(dp);
3140 
3141 	/*
3142 	 * Update usable space statistics.
3143 	 */
3144 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
3145 		vdev_sync_done(vd, txg);
3146 
3147 	/*
3148 	 * It had better be the case that we didn't dirty anything
3149 	 * since vdev_config_sync().
3150 	 */
3151 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
3152 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
3153 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
3154 	ASSERT(bpl->bpl_queue == NULL);
3155 
3156 	spa_config_exit(spa, FTAG);
3157 
3158 	/*
3159 	 * If any async tasks have been requested, kick them off.
3160 	 */
3161 	spa_async_dispatch(spa);
3162 }
3163 
3164 /*
3165  * Sync all pools.  We don't want to hold the namespace lock across these
3166  * operations, so we take a reference on the spa_t and drop the lock during the
3167  * sync.
3168  */
3169 void
3170 spa_sync_allpools(void)
3171 {
3172 	spa_t *spa = NULL;
3173 	mutex_enter(&spa_namespace_lock);
3174 	while ((spa = spa_next(spa)) != NULL) {
3175 		if (spa_state(spa) != POOL_STATE_ACTIVE)
3176 			continue;
3177 		spa_open_ref(spa, FTAG);
3178 		mutex_exit(&spa_namespace_lock);
3179 		txg_wait_synced(spa_get_dsl(spa), 0);
3180 		mutex_enter(&spa_namespace_lock);
3181 		spa_close(spa, FTAG);
3182 	}
3183 	mutex_exit(&spa_namespace_lock);
3184 }
3185 
3186 /*
3187  * ==========================================================================
3188  * Miscellaneous routines
3189  * ==========================================================================
3190  */
3191 
3192 /*
3193  * Remove all pools in the system.
3194  */
3195 void
3196 spa_evict_all(void)
3197 {
3198 	spa_t *spa;
3199 
3200 	/*
3201 	 * Remove all cached state.  All pools should be closed now,
3202 	 * so every spa in the AVL tree should be unreferenced.
3203 	 */
3204 	mutex_enter(&spa_namespace_lock);
3205 	while ((spa = spa_next(NULL)) != NULL) {
3206 		/*
3207 		 * Stop async tasks.  The async thread may need to detach
3208 		 * a device that's been replaced, which requires grabbing
3209 		 * spa_namespace_lock, so we must drop it here.
3210 		 */
3211 		spa_open_ref(spa, FTAG);
3212 		mutex_exit(&spa_namespace_lock);
3213 		spa_async_suspend(spa);
3214 		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
3215 		mutex_enter(&spa_namespace_lock);
3216 		spa_close(spa, FTAG);
3217 
3218 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3219 			spa_unload(spa);
3220 			spa_deactivate(spa);
3221 		}
3222 		spa_remove(spa);
3223 	}
3224 	mutex_exit(&spa_namespace_lock);
3225 }
3226 
3227 vdev_t *
3228 spa_lookup_by_guid(spa_t *spa, uint64_t guid)
3229 {
3230 	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
3231 }
3232 
3233 void
3234 spa_upgrade(spa_t *spa)
3235 {
3236 	spa_config_enter(spa, RW_WRITER, FTAG);
3237 
3238 	/*
3239 	 * This should only be called for a non-faulted pool, and since a
3240 	 * future version would result in an unopenable pool, this shouldn't be
3241 	 * possible.
3242 	 */
3243 	ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION);
3244 
3245 	spa->spa_uberblock.ub_version = ZFS_VERSION;
3246 	vdev_config_dirty(spa->spa_root_vdev);
3247 
3248 	spa_config_exit(spa, FTAG);
3249 
3250 	txg_wait_synced(spa_get_dsl(spa), 0);
3251 }
3252 
3253 boolean_t
3254 spa_has_spare(spa_t *spa, uint64_t guid)
3255 {
3256 	int i;
3257 	uint64_t spareguid;
3258 
3259 	for (i = 0; i < spa->spa_nspares; i++)
3260 		if (spa->spa_spares[i]->vdev_guid == guid)
3261 			return (B_TRUE);
3262 
3263 	for (i = 0; i < spa->spa_pending_nspares; i++) {
3264 		if (nvlist_lookup_uint64(spa->spa_pending_spares[i],
3265 		    ZPOOL_CONFIG_GUID, &spareguid) == 0 &&
3266 		    spareguid == guid)
3267 			return (B_TRUE);
3268 	}
3269 
3270 	return (B_FALSE);
3271 }
3272 
3273 int
3274 spa_set_props(spa_t *spa, nvlist_t *nvp)
3275 {
3276 	return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
3277 	    spa, nvp, 3));
3278 }
3279 
3280 int
3281 spa_get_props(spa_t *spa, nvlist_t **nvp)
3282 {
3283 	zap_cursor_t zc;
3284 	zap_attribute_t za;
3285 	objset_t *mos = spa->spa_meta_objset;
3286 	zfs_source_t src;
3287 	zpool_prop_t prop;
3288 	nvlist_t *propval;
3289 	uint64_t value;
3290 	int err;
3291 
3292 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3293 
3294 	mutex_enter(&spa->spa_props_lock);
3295 	/* If no props object, then just return empty nvlist */
3296 	if (spa->spa_pool_props_object == 0) {
3297 		mutex_exit(&spa->spa_props_lock);
3298 		return (0);
3299 	}
3300 
3301 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
3302 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
3303 	    zap_cursor_advance(&zc)) {
3304 
3305 		if ((prop = zpool_name_to_prop(za.za_name)) == ZFS_PROP_INVAL)
3306 			continue;
3307 
3308 		VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3309 		switch (za.za_integer_length) {
3310 		case 8:
3311 			if (zpool_prop_default_numeric(prop) ==
3312 			    za.za_first_integer)
3313 				src = ZFS_SRC_DEFAULT;
3314 			else
3315 				src = ZFS_SRC_LOCAL;
3316 			value = za.za_first_integer;
3317 
3318 			if (prop == ZPOOL_PROP_BOOTFS) {
3319 				dsl_pool_t *dp;
3320 				dsl_dataset_t *ds = NULL;
3321 				char strval[MAXPATHLEN];
3322 
3323 				dp = spa_get_dsl(spa);
3324 				rw_enter(&dp->dp_config_rwlock, RW_READER);
3325 				if ((err = dsl_dataset_open_obj(dp,
3326 				    za.za_first_integer, NULL, DS_MODE_NONE,
3327 				    FTAG, &ds)) != 0) {
3328 					rw_exit(&dp->dp_config_rwlock);
3329 					break;
3330 				}
3331 				dsl_dataset_name(ds, strval);
3332 				dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
3333 				rw_exit(&dp->dp_config_rwlock);
3334 
3335 				VERIFY(nvlist_add_uint64(propval,
3336 				    ZFS_PROP_SOURCE, src) == 0);
3337 				VERIFY(nvlist_add_string(propval,
3338 				    ZFS_PROP_VALUE, strval) == 0);
3339 			} else {
3340 				VERIFY(nvlist_add_uint64(propval,
3341 				    ZFS_PROP_SOURCE, src) == 0);
3342 				VERIFY(nvlist_add_uint64(propval,
3343 				    ZFS_PROP_VALUE, value) == 0);
3344 			}
3345 			VERIFY(nvlist_add_nvlist(*nvp, za.za_name,
3346 			    propval) == 0);
3347 			break;
3348 		}
3349 		nvlist_free(propval);
3350 	}
3351 	zap_cursor_fini(&zc);
3352 	mutex_exit(&spa->spa_props_lock);
3353 	if (err && err != ENOENT) {
3354 		nvlist_free(*nvp);
3355 		return (err);
3356 	}
3357 
3358 	return (0);
3359 }
3360 
3361 /*
3362  * If the bootfs property value is dsobj, clear it.
3363  */
3364 void
3365 spa_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
3366 {
3367 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
3368 		VERIFY(zap_remove(spa->spa_meta_objset,
3369 		    spa->spa_pool_props_object,
3370 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
3371 		spa->spa_bootfs = 0;
3372 	}
3373 }
3374 
3375 /*
3376  * Post a sysevent corresponding to the given event.  The 'name' must be one of
3377  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
3378  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
3379  * in the userland libzpool, as we don't want consumers to misinterpret ztest
3380  * or zdb as real changes.
3381  */
3382 void
3383 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
3384 {
3385 #ifdef _KERNEL
3386 	sysevent_t		*ev;
3387 	sysevent_attr_list_t	*attr = NULL;
3388 	sysevent_value_t	value;
3389 	sysevent_id_t		eid;
3390 
3391 	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
3392 	    SE_SLEEP);
3393 
3394 	value.value_type = SE_DATA_TYPE_STRING;
3395 	value.value.sv_string = spa_name(spa);
3396 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
3397 		goto done;
3398 
3399 	value.value_type = SE_DATA_TYPE_UINT64;
3400 	value.value.sv_uint64 = spa_guid(spa);
3401 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
3402 		goto done;
3403 
3404 	if (vd) {
3405 		value.value_type = SE_DATA_TYPE_UINT64;
3406 		value.value.sv_uint64 = vd->vdev_guid;
3407 		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
3408 		    SE_SLEEP) != 0)
3409 			goto done;
3410 
3411 		if (vd->vdev_path) {
3412 			value.value_type = SE_DATA_TYPE_STRING;
3413 			value.value.sv_string = vd->vdev_path;
3414 			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
3415 			    &value, SE_SLEEP) != 0)
3416 				goto done;
3417 		}
3418 	}
3419 
3420 	(void) log_sysevent(ev, SE_SLEEP, &eid);
3421 
3422 done:
3423 	if (attr)
3424 		sysevent_free_attr(attr);
3425 	sysevent_free(ev);
3426 #endif
3427 }
3428