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.
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.
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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa_impl.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
39 #include <sys/space_reftree.h>
42 #include <sys/fs/zfs.h>
45 #include <sys/dsl_scan.h>
49 * When a vdev is added, it will be divided into approximately (but no
50 * more than) this number of metaslabs.
52 int metaslabs_per_vdev = 200;
55 * Virtual device management.
58 static vdev_ops_t *vdev_ops_table[] = {
72 * Given a vdev type, return the appropriate ops vector.
75 vdev_getops(const char *type)
77 vdev_ops_t *ops, **opspp;
79 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
80 if (strcmp(ops->vdev_op_type, type) == 0)
87 * Default asize function: return the MAX of psize with the asize of
88 * all children. This is what's used by anything other than RAID-Z.
91 vdev_default_asize(vdev_t *vd, uint64_t psize)
93 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
97 for (c = 0; c < vd->vdev_children; c++) {
98 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
99 asize = MAX(asize, csize);
106 * Get the minimum allocatable size. We define the allocatable size as
107 * the vdev's asize rounded to the nearest metaslab. This allows us to
108 * replace or attach devices which don't have the same physical size but
109 * can still satisfy the same number of allocations.
112 vdev_get_min_asize(vdev_t *vd)
114 vdev_t *pvd = vd->vdev_parent;
117 * If our parent is NULL (inactive spare or cache) or is the root,
118 * just return our own asize.
121 return (vd->vdev_asize);
124 * The top-level vdev just returns the allocatable size rounded
125 * to the nearest metaslab.
127 if (vd == vd->vdev_top)
128 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
131 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
132 * so each child must provide at least 1/Nth of its asize.
134 if (pvd->vdev_ops == &vdev_raidz_ops)
135 return (pvd->vdev_min_asize / pvd->vdev_children);
137 return (pvd->vdev_min_asize);
141 vdev_set_min_asize(vdev_t *vd)
144 vd->vdev_min_asize = vdev_get_min_asize(vd);
146 for (c = 0; c < vd->vdev_children; c++)
147 vdev_set_min_asize(vd->vdev_child[c]);
151 vdev_lookup_top(spa_t *spa, uint64_t vdev)
153 vdev_t *rvd = spa->spa_root_vdev;
155 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
157 if (vdev < rvd->vdev_children) {
158 ASSERT(rvd->vdev_child[vdev] != NULL);
159 return (rvd->vdev_child[vdev]);
166 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
171 if (vd->vdev_guid == guid)
174 for (c = 0; c < vd->vdev_children; c++)
175 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
183 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
185 size_t oldsize, newsize;
186 uint64_t id = cvd->vdev_id;
189 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
190 ASSERT(cvd->vdev_parent == NULL);
192 cvd->vdev_parent = pvd;
197 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
199 oldsize = pvd->vdev_children * sizeof (vdev_t *);
200 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
201 newsize = pvd->vdev_children * sizeof (vdev_t *);
203 newchild = kmem_alloc(newsize, KM_PUSHPAGE);
204 if (pvd->vdev_child != NULL) {
205 bcopy(pvd->vdev_child, newchild, oldsize);
206 kmem_free(pvd->vdev_child, oldsize);
209 pvd->vdev_child = newchild;
210 pvd->vdev_child[id] = cvd;
212 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
213 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
216 * Walk up all ancestors to update guid sum.
218 for (; pvd != NULL; pvd = pvd->vdev_parent)
219 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
223 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
226 uint_t id = cvd->vdev_id;
228 ASSERT(cvd->vdev_parent == pvd);
233 ASSERT(id < pvd->vdev_children);
234 ASSERT(pvd->vdev_child[id] == cvd);
236 pvd->vdev_child[id] = NULL;
237 cvd->vdev_parent = NULL;
239 for (c = 0; c < pvd->vdev_children; c++)
240 if (pvd->vdev_child[c])
243 if (c == pvd->vdev_children) {
244 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
245 pvd->vdev_child = NULL;
246 pvd->vdev_children = 0;
250 * Walk up all ancestors to update guid sum.
252 for (; pvd != NULL; pvd = pvd->vdev_parent)
253 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
257 * Remove any holes in the child array.
260 vdev_compact_children(vdev_t *pvd)
262 vdev_t **newchild, *cvd;
263 int oldc = pvd->vdev_children;
267 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
269 for (c = newc = 0; c < oldc; c++)
270 if (pvd->vdev_child[c])
273 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_PUSHPAGE);
275 for (c = newc = 0; c < oldc; c++) {
276 if ((cvd = pvd->vdev_child[c]) != NULL) {
277 newchild[newc] = cvd;
278 cvd->vdev_id = newc++;
282 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
283 pvd->vdev_child = newchild;
284 pvd->vdev_children = newc;
288 * Allocate and minimally initialize a vdev_t.
291 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
296 vd = kmem_zalloc(sizeof (vdev_t), KM_PUSHPAGE);
298 if (spa->spa_root_vdev == NULL) {
299 ASSERT(ops == &vdev_root_ops);
300 spa->spa_root_vdev = vd;
301 spa->spa_load_guid = spa_generate_guid(NULL);
304 if (guid == 0 && ops != &vdev_hole_ops) {
305 if (spa->spa_root_vdev == vd) {
307 * The root vdev's guid will also be the pool guid,
308 * which must be unique among all pools.
310 guid = spa_generate_guid(NULL);
313 * Any other vdev's guid must be unique within the pool.
315 guid = spa_generate_guid(spa);
317 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
322 vd->vdev_guid = guid;
323 vd->vdev_guid_sum = guid;
325 vd->vdev_state = VDEV_STATE_CLOSED;
326 vd->vdev_ishole = (ops == &vdev_hole_ops);
328 list_link_init(&vd->vdev_config_dirty_node);
329 list_link_init(&vd->vdev_state_dirty_node);
330 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
331 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
332 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
333 for (t = 0; t < DTL_TYPES; t++) {
334 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
337 txg_list_create(&vd->vdev_ms_list,
338 offsetof(struct metaslab, ms_txg_node));
339 txg_list_create(&vd->vdev_dtl_list,
340 offsetof(struct vdev, vdev_dtl_node));
341 vd->vdev_stat.vs_timestamp = gethrtime();
349 * Allocate a new vdev. The 'alloctype' is used to control whether we are
350 * creating a new vdev or loading an existing one - the behavior is slightly
351 * different for each case.
354 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
359 uint64_t guid = 0, islog, nparity;
362 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
364 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
365 return (SET_ERROR(EINVAL));
367 if ((ops = vdev_getops(type)) == NULL)
368 return (SET_ERROR(EINVAL));
371 * If this is a load, get the vdev guid from the nvlist.
372 * Otherwise, vdev_alloc_common() will generate one for us.
374 if (alloctype == VDEV_ALLOC_LOAD) {
377 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
379 return (SET_ERROR(EINVAL));
381 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
382 return (SET_ERROR(EINVAL));
383 } else if (alloctype == VDEV_ALLOC_SPARE) {
384 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
385 return (SET_ERROR(EINVAL));
386 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
387 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
388 return (SET_ERROR(EINVAL));
389 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
390 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
391 return (SET_ERROR(EINVAL));
395 * The first allocated vdev must be of type 'root'.
397 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
398 return (SET_ERROR(EINVAL));
401 * Determine whether we're a log vdev.
404 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
405 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
406 return (SET_ERROR(ENOTSUP));
408 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
409 return (SET_ERROR(ENOTSUP));
412 * Set the nparity property for RAID-Z vdevs.
415 if (ops == &vdev_raidz_ops) {
416 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
418 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
419 return (SET_ERROR(EINVAL));
421 * Previous versions could only support 1 or 2 parity
425 spa_version(spa) < SPA_VERSION_RAIDZ2)
426 return (SET_ERROR(ENOTSUP));
428 spa_version(spa) < SPA_VERSION_RAIDZ3)
429 return (SET_ERROR(ENOTSUP));
432 * We require the parity to be specified for SPAs that
433 * support multiple parity levels.
435 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
436 return (SET_ERROR(EINVAL));
438 * Otherwise, we default to 1 parity device for RAID-Z.
445 ASSERT(nparity != -1ULL);
447 vd = vdev_alloc_common(spa, id, guid, ops);
449 vd->vdev_islog = islog;
450 vd->vdev_nparity = nparity;
452 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
453 vd->vdev_path = spa_strdup(vd->vdev_path);
454 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
455 vd->vdev_devid = spa_strdup(vd->vdev_devid);
456 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
457 &vd->vdev_physpath) == 0)
458 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
459 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
460 vd->vdev_fru = spa_strdup(vd->vdev_fru);
463 * Set the whole_disk property. If it's not specified, leave the value
466 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
467 &vd->vdev_wholedisk) != 0)
468 vd->vdev_wholedisk = -1ULL;
471 * Look for the 'not present' flag. This will only be set if the device
472 * was not present at the time of import.
474 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
475 &vd->vdev_not_present);
478 * Get the alignment requirement.
480 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
483 * Retrieve the vdev creation time.
485 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
489 * If we're a top-level vdev, try to load the allocation parameters.
491 if (parent && !parent->vdev_parent &&
492 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
493 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
495 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
497 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
499 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
503 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
504 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
505 alloctype == VDEV_ALLOC_ADD ||
506 alloctype == VDEV_ALLOC_SPLIT ||
507 alloctype == VDEV_ALLOC_ROOTPOOL);
508 vd->vdev_mg = metaslab_group_create(islog ?
509 spa_log_class(spa) : spa_normal_class(spa), vd);
513 * If we're a leaf vdev, try to load the DTL object and other state.
515 if (vd->vdev_ops->vdev_op_leaf &&
516 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
517 alloctype == VDEV_ALLOC_ROOTPOOL)) {
518 if (alloctype == VDEV_ALLOC_LOAD) {
519 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
520 &vd->vdev_dtl_object);
521 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
525 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
528 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
529 &spare) == 0 && spare)
533 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
536 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
537 &vd->vdev_resilver_txg);
540 * When importing a pool, we want to ignore the persistent fault
541 * state, as the diagnosis made on another system may not be
542 * valid in the current context. Local vdevs will
543 * remain in the faulted state.
545 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
546 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
548 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
550 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
553 if (vd->vdev_faulted || vd->vdev_degraded) {
557 VDEV_AUX_ERR_EXCEEDED;
558 if (nvlist_lookup_string(nv,
559 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
560 strcmp(aux, "external") == 0)
561 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
567 * Add ourselves to the parent's list of children.
569 vdev_add_child(parent, vd);
577 vdev_free(vdev_t *vd)
580 spa_t *spa = vd->vdev_spa;
583 * vdev_free() implies closing the vdev first. This is simpler than
584 * trying to ensure complicated semantics for all callers.
588 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
589 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
594 for (c = 0; c < vd->vdev_children; c++)
595 vdev_free(vd->vdev_child[c]);
597 ASSERT(vd->vdev_child == NULL);
598 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
601 * Discard allocation state.
603 if (vd->vdev_mg != NULL) {
604 vdev_metaslab_fini(vd);
605 metaslab_group_destroy(vd->vdev_mg);
608 ASSERT0(vd->vdev_stat.vs_space);
609 ASSERT0(vd->vdev_stat.vs_dspace);
610 ASSERT0(vd->vdev_stat.vs_alloc);
613 * Remove this vdev from its parent's child list.
615 vdev_remove_child(vd->vdev_parent, vd);
617 ASSERT(vd->vdev_parent == NULL);
620 * Clean up vdev structure.
626 spa_strfree(vd->vdev_path);
628 spa_strfree(vd->vdev_devid);
629 if (vd->vdev_physpath)
630 spa_strfree(vd->vdev_physpath);
632 spa_strfree(vd->vdev_fru);
634 if (vd->vdev_isspare)
635 spa_spare_remove(vd);
636 if (vd->vdev_isl2cache)
637 spa_l2cache_remove(vd);
639 txg_list_destroy(&vd->vdev_ms_list);
640 txg_list_destroy(&vd->vdev_dtl_list);
642 mutex_enter(&vd->vdev_dtl_lock);
643 space_map_close(vd->vdev_dtl_sm);
644 for (t = 0; t < DTL_TYPES; t++) {
645 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
646 range_tree_destroy(vd->vdev_dtl[t]);
648 mutex_exit(&vd->vdev_dtl_lock);
650 mutex_destroy(&vd->vdev_dtl_lock);
651 mutex_destroy(&vd->vdev_stat_lock);
652 mutex_destroy(&vd->vdev_probe_lock);
654 if (vd == spa->spa_root_vdev)
655 spa->spa_root_vdev = NULL;
657 kmem_free(vd, sizeof (vdev_t));
661 * Transfer top-level vdev state from svd to tvd.
664 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
666 spa_t *spa = svd->vdev_spa;
671 ASSERT(tvd == tvd->vdev_top);
673 tvd->vdev_ms_array = svd->vdev_ms_array;
674 tvd->vdev_ms_shift = svd->vdev_ms_shift;
675 tvd->vdev_ms_count = svd->vdev_ms_count;
677 svd->vdev_ms_array = 0;
678 svd->vdev_ms_shift = 0;
679 svd->vdev_ms_count = 0;
682 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
683 tvd->vdev_mg = svd->vdev_mg;
684 tvd->vdev_ms = svd->vdev_ms;
689 if (tvd->vdev_mg != NULL)
690 tvd->vdev_mg->mg_vd = tvd;
692 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
693 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
694 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
696 svd->vdev_stat.vs_alloc = 0;
697 svd->vdev_stat.vs_space = 0;
698 svd->vdev_stat.vs_dspace = 0;
700 for (t = 0; t < TXG_SIZE; t++) {
701 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
702 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
703 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
704 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
705 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
706 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
709 if (list_link_active(&svd->vdev_config_dirty_node)) {
710 vdev_config_clean(svd);
711 vdev_config_dirty(tvd);
714 if (list_link_active(&svd->vdev_state_dirty_node)) {
715 vdev_state_clean(svd);
716 vdev_state_dirty(tvd);
719 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
720 svd->vdev_deflate_ratio = 0;
722 tvd->vdev_islog = svd->vdev_islog;
727 vdev_top_update(vdev_t *tvd, vdev_t *vd)
736 for (c = 0; c < vd->vdev_children; c++)
737 vdev_top_update(tvd, vd->vdev_child[c]);
741 * Add a mirror/replacing vdev above an existing vdev.
744 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
746 spa_t *spa = cvd->vdev_spa;
747 vdev_t *pvd = cvd->vdev_parent;
750 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
752 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
754 mvd->vdev_asize = cvd->vdev_asize;
755 mvd->vdev_min_asize = cvd->vdev_min_asize;
756 mvd->vdev_max_asize = cvd->vdev_max_asize;
757 mvd->vdev_ashift = cvd->vdev_ashift;
758 mvd->vdev_state = cvd->vdev_state;
759 mvd->vdev_crtxg = cvd->vdev_crtxg;
761 vdev_remove_child(pvd, cvd);
762 vdev_add_child(pvd, mvd);
763 cvd->vdev_id = mvd->vdev_children;
764 vdev_add_child(mvd, cvd);
765 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
767 if (mvd == mvd->vdev_top)
768 vdev_top_transfer(cvd, mvd);
774 * Remove a 1-way mirror/replacing vdev from the tree.
777 vdev_remove_parent(vdev_t *cvd)
779 vdev_t *mvd = cvd->vdev_parent;
780 vdev_t *pvd = mvd->vdev_parent;
782 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
784 ASSERT(mvd->vdev_children == 1);
785 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
786 mvd->vdev_ops == &vdev_replacing_ops ||
787 mvd->vdev_ops == &vdev_spare_ops);
788 cvd->vdev_ashift = mvd->vdev_ashift;
790 vdev_remove_child(mvd, cvd);
791 vdev_remove_child(pvd, mvd);
794 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
795 * Otherwise, we could have detached an offline device, and when we
796 * go to import the pool we'll think we have two top-level vdevs,
797 * instead of a different version of the same top-level vdev.
799 if (mvd->vdev_top == mvd) {
800 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
801 cvd->vdev_orig_guid = cvd->vdev_guid;
802 cvd->vdev_guid += guid_delta;
803 cvd->vdev_guid_sum += guid_delta;
806 * If pool not set for autoexpand, we need to also preserve
807 * mvd's asize to prevent automatic expansion of cvd.
808 * Otherwise if we are adjusting the mirror by attaching and
809 * detaching children of non-uniform sizes, the mirror could
810 * autoexpand, unexpectedly requiring larger devices to
811 * re-establish the mirror.
813 if (!cvd->vdev_spa->spa_autoexpand)
814 cvd->vdev_asize = mvd->vdev_asize;
816 cvd->vdev_id = mvd->vdev_id;
817 vdev_add_child(pvd, cvd);
818 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
820 if (cvd == cvd->vdev_top)
821 vdev_top_transfer(mvd, cvd);
823 ASSERT(mvd->vdev_children == 0);
828 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
830 spa_t *spa = vd->vdev_spa;
831 objset_t *mos = spa->spa_meta_objset;
833 uint64_t oldc = vd->vdev_ms_count;
834 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
838 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
841 * This vdev is not being allocated from yet or is a hole.
843 if (vd->vdev_ms_shift == 0)
846 ASSERT(!vd->vdev_ishole);
849 * Compute the raidz-deflation ratio. Note, we hard-code
850 * in 128k (1 << 17) because it is the current "typical" blocksize.
851 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
852 * or we will inconsistently account for existing bp's.
854 vd->vdev_deflate_ratio = (1 << 17) /
855 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
857 ASSERT(oldc <= newc);
859 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_PUSHPAGE | KM_NODEBUG);
862 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
863 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
867 vd->vdev_ms_count = newc;
869 for (m = oldc; m < newc; m++) {
873 error = dmu_read(mos, vd->vdev_ms_array,
874 m * sizeof (uint64_t), sizeof (uint64_t), &object,
880 error = metaslab_init(vd->vdev_mg, m, object, txg,
887 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
890 * If the vdev is being removed we don't activate
891 * the metaslabs since we want to ensure that no new
892 * allocations are performed on this device.
894 if (oldc == 0 && !vd->vdev_removing)
895 metaslab_group_activate(vd->vdev_mg);
898 spa_config_exit(spa, SCL_ALLOC, FTAG);
904 vdev_metaslab_fini(vdev_t *vd)
907 uint64_t count = vd->vdev_ms_count;
909 if (vd->vdev_ms != NULL) {
910 metaslab_group_passivate(vd->vdev_mg);
911 for (m = 0; m < count; m++) {
912 metaslab_t *msp = vd->vdev_ms[m];
917 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
921 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
924 typedef struct vdev_probe_stats {
925 boolean_t vps_readable;
926 boolean_t vps_writeable;
928 } vdev_probe_stats_t;
931 vdev_probe_done(zio_t *zio)
933 spa_t *spa = zio->io_spa;
934 vdev_t *vd = zio->io_vd;
935 vdev_probe_stats_t *vps = zio->io_private;
937 ASSERT(vd->vdev_probe_zio != NULL);
939 if (zio->io_type == ZIO_TYPE_READ) {
940 if (zio->io_error == 0)
941 vps->vps_readable = 1;
942 if (zio->io_error == 0 && spa_writeable(spa)) {
943 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
944 zio->io_offset, zio->io_size, zio->io_data,
945 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
946 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
948 zio_buf_free(zio->io_data, zio->io_size);
950 } else if (zio->io_type == ZIO_TYPE_WRITE) {
951 if (zio->io_error == 0)
952 vps->vps_writeable = 1;
953 zio_buf_free(zio->io_data, zio->io_size);
954 } else if (zio->io_type == ZIO_TYPE_NULL) {
957 vd->vdev_cant_read |= !vps->vps_readable;
958 vd->vdev_cant_write |= !vps->vps_writeable;
960 if (vdev_readable(vd) &&
961 (vdev_writeable(vd) || !spa_writeable(spa))) {
964 ASSERT(zio->io_error != 0);
965 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
966 spa, vd, NULL, 0, 0);
967 zio->io_error = SET_ERROR(ENXIO);
970 mutex_enter(&vd->vdev_probe_lock);
971 ASSERT(vd->vdev_probe_zio == zio);
972 vd->vdev_probe_zio = NULL;
973 mutex_exit(&vd->vdev_probe_lock);
975 while ((pio = zio_walk_parents(zio)) != NULL)
976 if (!vdev_accessible(vd, pio))
977 pio->io_error = SET_ERROR(ENXIO);
979 kmem_free(vps, sizeof (*vps));
984 * Determine whether this device is accessible.
986 * Read and write to several known locations: the pad regions of each
987 * vdev label but the first, which we leave alone in case it contains
991 vdev_probe(vdev_t *vd, zio_t *zio)
993 spa_t *spa = vd->vdev_spa;
994 vdev_probe_stats_t *vps = NULL;
998 ASSERT(vd->vdev_ops->vdev_op_leaf);
1001 * Don't probe the probe.
1003 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1007 * To prevent 'probe storms' when a device fails, we create
1008 * just one probe i/o at a time. All zios that want to probe
1009 * this vdev will become parents of the probe io.
1011 mutex_enter(&vd->vdev_probe_lock);
1013 if ((pio = vd->vdev_probe_zio) == NULL) {
1014 vps = kmem_zalloc(sizeof (*vps), KM_PUSHPAGE);
1016 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1017 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1020 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1022 * vdev_cant_read and vdev_cant_write can only
1023 * transition from TRUE to FALSE when we have the
1024 * SCL_ZIO lock as writer; otherwise they can only
1025 * transition from FALSE to TRUE. This ensures that
1026 * any zio looking at these values can assume that
1027 * failures persist for the life of the I/O. That's
1028 * important because when a device has intermittent
1029 * connectivity problems, we want to ensure that
1030 * they're ascribed to the device (ENXIO) and not
1033 * Since we hold SCL_ZIO as writer here, clear both
1034 * values so the probe can reevaluate from first
1037 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1038 vd->vdev_cant_read = B_FALSE;
1039 vd->vdev_cant_write = B_FALSE;
1042 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1043 vdev_probe_done, vps,
1044 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1047 * We can't change the vdev state in this context, so we
1048 * kick off an async task to do it on our behalf.
1051 vd->vdev_probe_wanted = B_TRUE;
1052 spa_async_request(spa, SPA_ASYNC_PROBE);
1057 zio_add_child(zio, pio);
1059 mutex_exit(&vd->vdev_probe_lock);
1062 ASSERT(zio != NULL);
1066 for (l = 1; l < VDEV_LABELS; l++) {
1067 zio_nowait(zio_read_phys(pio, vd,
1068 vdev_label_offset(vd->vdev_psize, l,
1069 offsetof(vdev_label_t, vl_pad2)),
1070 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1071 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1072 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1083 vdev_open_child(void *arg)
1087 vd->vdev_open_thread = curthread;
1088 vd->vdev_open_error = vdev_open(vd);
1089 vd->vdev_open_thread = NULL;
1093 vdev_uses_zvols(vdev_t *vd)
1098 if (zvol_is_zvol(vd->vdev_path))
1102 for (c = 0; c < vd->vdev_children; c++)
1103 if (vdev_uses_zvols(vd->vdev_child[c]))
1110 vdev_open_children(vdev_t *vd)
1113 int children = vd->vdev_children;
1117 * in order to handle pools on top of zvols, do the opens
1118 * in a single thread so that the same thread holds the
1119 * spa_namespace_lock
1121 if (vdev_uses_zvols(vd)) {
1122 for (c = 0; c < children; c++)
1123 vd->vdev_child[c]->vdev_open_error =
1124 vdev_open(vd->vdev_child[c]);
1127 tq = taskq_create("vdev_open", children, minclsyspri,
1128 children, children, TASKQ_PREPOPULATE);
1130 for (c = 0; c < children; c++)
1131 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1138 * Prepare a virtual device for access.
1141 vdev_open(vdev_t *vd)
1143 spa_t *spa = vd->vdev_spa;
1146 uint64_t max_osize = 0;
1147 uint64_t asize, max_asize, psize;
1148 uint64_t ashift = 0;
1151 ASSERT(vd->vdev_open_thread == curthread ||
1152 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1153 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1154 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1155 vd->vdev_state == VDEV_STATE_OFFLINE);
1157 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1158 vd->vdev_cant_read = B_FALSE;
1159 vd->vdev_cant_write = B_FALSE;
1160 vd->vdev_min_asize = vdev_get_min_asize(vd);
1163 * If this vdev is not removed, check its fault status. If it's
1164 * faulted, bail out of the open.
1166 if (!vd->vdev_removed && vd->vdev_faulted) {
1167 ASSERT(vd->vdev_children == 0);
1168 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1169 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1170 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1171 vd->vdev_label_aux);
1172 return (SET_ERROR(ENXIO));
1173 } else if (vd->vdev_offline) {
1174 ASSERT(vd->vdev_children == 0);
1175 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1176 return (SET_ERROR(ENXIO));
1179 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1182 * Reset the vdev_reopening flag so that we actually close
1183 * the vdev on error.
1185 vd->vdev_reopening = B_FALSE;
1186 if (zio_injection_enabled && error == 0)
1187 error = zio_handle_device_injection(vd, NULL, ENXIO);
1190 if (vd->vdev_removed &&
1191 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1192 vd->vdev_removed = B_FALSE;
1194 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1195 vd->vdev_stat.vs_aux);
1199 vd->vdev_removed = B_FALSE;
1202 * Recheck the faulted flag now that we have confirmed that
1203 * the vdev is accessible. If we're faulted, bail.
1205 if (vd->vdev_faulted) {
1206 ASSERT(vd->vdev_children == 0);
1207 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1208 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1209 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1210 vd->vdev_label_aux);
1211 return (SET_ERROR(ENXIO));
1214 if (vd->vdev_degraded) {
1215 ASSERT(vd->vdev_children == 0);
1216 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1217 VDEV_AUX_ERR_EXCEEDED);
1219 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1223 * For hole or missing vdevs we just return success.
1225 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1228 for (c = 0; c < vd->vdev_children; c++) {
1229 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1230 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1236 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1237 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1239 if (vd->vdev_children == 0) {
1240 if (osize < SPA_MINDEVSIZE) {
1241 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1242 VDEV_AUX_TOO_SMALL);
1243 return (SET_ERROR(EOVERFLOW));
1246 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1247 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1248 VDEV_LABEL_END_SIZE);
1250 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1251 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1252 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1253 VDEV_AUX_TOO_SMALL);
1254 return (SET_ERROR(EOVERFLOW));
1258 max_asize = max_osize;
1261 vd->vdev_psize = psize;
1264 * Make sure the allocatable size hasn't shrunk.
1266 if (asize < vd->vdev_min_asize) {
1267 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1268 VDEV_AUX_BAD_LABEL);
1269 return (SET_ERROR(EINVAL));
1272 if (vd->vdev_asize == 0) {
1274 * This is the first-ever open, so use the computed values.
1275 * For compatibility, a different ashift can be requested.
1277 vd->vdev_asize = asize;
1278 vd->vdev_max_asize = max_asize;
1279 if (vd->vdev_ashift == 0)
1280 vd->vdev_ashift = ashift;
1283 * Detect if the alignment requirement has increased.
1284 * We don't want to make the pool unavailable, just
1285 * post an event instead.
1287 if (ashift > vd->vdev_top->vdev_ashift &&
1288 vd->vdev_ops->vdev_op_leaf) {
1289 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
1290 spa, vd, NULL, 0, 0);
1293 vd->vdev_max_asize = max_asize;
1297 * If all children are healthy and the asize has increased,
1298 * then we've experienced dynamic LUN growth. If automatic
1299 * expansion is enabled then use the additional space.
1301 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1302 (vd->vdev_expanding || spa->spa_autoexpand))
1303 vd->vdev_asize = asize;
1305 vdev_set_min_asize(vd);
1308 * Ensure we can issue some IO before declaring the
1309 * vdev open for business.
1311 if (vd->vdev_ops->vdev_op_leaf &&
1312 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1313 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1314 VDEV_AUX_ERR_EXCEEDED);
1319 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1320 * resilver. But don't do this if we are doing a reopen for a scrub,
1321 * since this would just restart the scrub we are already doing.
1323 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1324 vdev_resilver_needed(vd, NULL, NULL))
1325 spa_async_request(spa, SPA_ASYNC_RESILVER);
1331 * Called once the vdevs are all opened, this routine validates the label
1332 * contents. This needs to be done before vdev_load() so that we don't
1333 * inadvertently do repair I/Os to the wrong device.
1335 * If 'strict' is false ignore the spa guid check. This is necessary because
1336 * if the machine crashed during a re-guid the new guid might have been written
1337 * to all of the vdev labels, but not the cached config. The strict check
1338 * will be performed when the pool is opened again using the mos config.
1340 * This function will only return failure if one of the vdevs indicates that it
1341 * has since been destroyed or exported. This is only possible if
1342 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1343 * will be updated but the function will return 0.
1346 vdev_validate(vdev_t *vd, boolean_t strict)
1348 spa_t *spa = vd->vdev_spa;
1350 uint64_t guid = 0, top_guid;
1354 for (c = 0; c < vd->vdev_children; c++)
1355 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1356 return (SET_ERROR(EBADF));
1359 * If the device has already failed, or was marked offline, don't do
1360 * any further validation. Otherwise, label I/O will fail and we will
1361 * overwrite the previous state.
1363 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1364 uint64_t aux_guid = 0;
1366 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1367 spa_last_synced_txg(spa) : -1ULL;
1369 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1370 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1371 VDEV_AUX_BAD_LABEL);
1376 * Determine if this vdev has been split off into another
1377 * pool. If so, then refuse to open it.
1379 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1380 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1381 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1382 VDEV_AUX_SPLIT_POOL);
1387 if (strict && (nvlist_lookup_uint64(label,
1388 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1389 guid != spa_guid(spa))) {
1390 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1391 VDEV_AUX_CORRUPT_DATA);
1396 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1397 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1402 * If this vdev just became a top-level vdev because its
1403 * sibling was detached, it will have adopted the parent's
1404 * vdev guid -- but the label may or may not be on disk yet.
1405 * Fortunately, either version of the label will have the
1406 * same top guid, so if we're a top-level vdev, we can
1407 * safely compare to that instead.
1409 * If we split this vdev off instead, then we also check the
1410 * original pool's guid. We don't want to consider the vdev
1411 * corrupt if it is partway through a split operation.
1413 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1415 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1417 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1418 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1419 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1420 VDEV_AUX_CORRUPT_DATA);
1425 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1427 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1428 VDEV_AUX_CORRUPT_DATA);
1436 * If this is a verbatim import, no need to check the
1437 * state of the pool.
1439 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1440 spa_load_state(spa) == SPA_LOAD_OPEN &&
1441 state != POOL_STATE_ACTIVE)
1442 return (SET_ERROR(EBADF));
1445 * If we were able to open and validate a vdev that was
1446 * previously marked permanently unavailable, clear that state
1449 if (vd->vdev_not_present)
1450 vd->vdev_not_present = 0;
1457 * Close a virtual device.
1460 vdev_close(vdev_t *vd)
1462 vdev_t *pvd = vd->vdev_parent;
1463 ASSERTV(spa_t *spa = vd->vdev_spa);
1465 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1468 * If our parent is reopening, then we are as well, unless we are
1471 if (pvd != NULL && pvd->vdev_reopening)
1472 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1474 vd->vdev_ops->vdev_op_close(vd);
1476 vdev_cache_purge(vd);
1479 * We record the previous state before we close it, so that if we are
1480 * doing a reopen(), we don't generate FMA ereports if we notice that
1481 * it's still faulted.
1483 vd->vdev_prevstate = vd->vdev_state;
1485 if (vd->vdev_offline)
1486 vd->vdev_state = VDEV_STATE_OFFLINE;
1488 vd->vdev_state = VDEV_STATE_CLOSED;
1489 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1493 vdev_hold(vdev_t *vd)
1495 spa_t *spa = vd->vdev_spa;
1498 ASSERT(spa_is_root(spa));
1499 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1502 for (c = 0; c < vd->vdev_children; c++)
1503 vdev_hold(vd->vdev_child[c]);
1505 if (vd->vdev_ops->vdev_op_leaf)
1506 vd->vdev_ops->vdev_op_hold(vd);
1510 vdev_rele(vdev_t *vd)
1514 ASSERT(spa_is_root(vd->vdev_spa));
1515 for (c = 0; c < vd->vdev_children; c++)
1516 vdev_rele(vd->vdev_child[c]);
1518 if (vd->vdev_ops->vdev_op_leaf)
1519 vd->vdev_ops->vdev_op_rele(vd);
1523 * Reopen all interior vdevs and any unopened leaves. We don't actually
1524 * reopen leaf vdevs which had previously been opened as they might deadlock
1525 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1526 * If the leaf has never been opened then open it, as usual.
1529 vdev_reopen(vdev_t *vd)
1531 spa_t *spa = vd->vdev_spa;
1533 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1535 /* set the reopening flag unless we're taking the vdev offline */
1536 vd->vdev_reopening = !vd->vdev_offline;
1538 (void) vdev_open(vd);
1541 * Call vdev_validate() here to make sure we have the same device.
1542 * Otherwise, a device with an invalid label could be successfully
1543 * opened in response to vdev_reopen().
1546 (void) vdev_validate_aux(vd);
1547 if (vdev_readable(vd) && vdev_writeable(vd) &&
1548 vd->vdev_aux == &spa->spa_l2cache &&
1549 !l2arc_vdev_present(vd))
1550 l2arc_add_vdev(spa, vd);
1552 (void) vdev_validate(vd, B_TRUE);
1556 * Reassess parent vdev's health.
1558 vdev_propagate_state(vd);
1562 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1567 * Normally, partial opens (e.g. of a mirror) are allowed.
1568 * For a create, however, we want to fail the request if
1569 * there are any components we can't open.
1571 error = vdev_open(vd);
1573 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1575 return (error ? error : ENXIO);
1579 * Recursively load DTLs and initialize all labels.
1581 if ((error = vdev_dtl_load(vd)) != 0 ||
1582 (error = vdev_label_init(vd, txg, isreplacing ?
1583 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1592 vdev_metaslab_set_size(vdev_t *vd)
1595 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1597 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1598 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1602 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1604 ASSERT(vd == vd->vdev_top);
1605 ASSERT(!vd->vdev_ishole);
1606 ASSERT(ISP2(flags));
1607 ASSERT(spa_writeable(vd->vdev_spa));
1609 if (flags & VDD_METASLAB)
1610 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1612 if (flags & VDD_DTL)
1613 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1615 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1619 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1623 for (c = 0; c < vd->vdev_children; c++)
1624 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1626 if (vd->vdev_ops->vdev_op_leaf)
1627 vdev_dirty(vd->vdev_top, flags, vd, txg);
1633 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1634 * the vdev has less than perfect replication. There are four kinds of DTL:
1636 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1638 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1640 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1641 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1642 * txgs that was scrubbed.
1644 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1645 * persistent errors or just some device being offline.
1646 * Unlike the other three, the DTL_OUTAGE map is not generally
1647 * maintained; it's only computed when needed, typically to
1648 * determine whether a device can be detached.
1650 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1651 * either has the data or it doesn't.
1653 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1654 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1655 * if any child is less than fully replicated, then so is its parent.
1656 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1657 * comprising only those txgs which appear in 'maxfaults' or more children;
1658 * those are the txgs we don't have enough replication to read. For example,
1659 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1660 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1661 * two child DTL_MISSING maps.
1663 * It should be clear from the above that to compute the DTLs and outage maps
1664 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1665 * Therefore, that is all we keep on disk. When loading the pool, or after
1666 * a configuration change, we generate all other DTLs from first principles.
1669 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1671 range_tree_t *rt = vd->vdev_dtl[t];
1673 ASSERT(t < DTL_TYPES);
1674 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1675 ASSERT(spa_writeable(vd->vdev_spa));
1677 mutex_enter(rt->rt_lock);
1678 if (!range_tree_contains(rt, txg, size))
1679 range_tree_add(rt, txg, size);
1680 mutex_exit(rt->rt_lock);
1684 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1686 range_tree_t *rt = vd->vdev_dtl[t];
1687 boolean_t dirty = B_FALSE;
1689 ASSERT(t < DTL_TYPES);
1690 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1692 mutex_enter(rt->rt_lock);
1693 if (range_tree_space(rt) != 0)
1694 dirty = range_tree_contains(rt, txg, size);
1695 mutex_exit(rt->rt_lock);
1701 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1703 range_tree_t *rt = vd->vdev_dtl[t];
1706 mutex_enter(rt->rt_lock);
1707 empty = (range_tree_space(rt) == 0);
1708 mutex_exit(rt->rt_lock);
1714 * Returns the lowest txg in the DTL range.
1717 vdev_dtl_min(vdev_t *vd)
1721 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1722 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1723 ASSERT0(vd->vdev_children);
1725 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1726 return (rs->rs_start - 1);
1730 * Returns the highest txg in the DTL.
1733 vdev_dtl_max(vdev_t *vd)
1737 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1738 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1739 ASSERT0(vd->vdev_children);
1741 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1742 return (rs->rs_end);
1746 * Determine if a resilvering vdev should remove any DTL entries from
1747 * its range. If the vdev was resilvering for the entire duration of the
1748 * scan then it should excise that range from its DTLs. Otherwise, this
1749 * vdev is considered partially resilvered and should leave its DTL
1750 * entries intact. The comment in vdev_dtl_reassess() describes how we
1754 vdev_dtl_should_excise(vdev_t *vd)
1756 spa_t *spa = vd->vdev_spa;
1757 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1759 ASSERT0(scn->scn_phys.scn_errors);
1760 ASSERT0(vd->vdev_children);
1762 if (vd->vdev_resilver_txg == 0 ||
1763 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1767 * When a resilver is initiated the scan will assign the scn_max_txg
1768 * value to the highest txg value that exists in all DTLs. If this
1769 * device's max DTL is not part of this scan (i.e. it is not in
1770 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1773 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1774 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1775 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1776 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1783 * Reassess DTLs after a config change or scrub completion.
1786 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1788 spa_t *spa = vd->vdev_spa;
1792 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1794 for (c = 0; c < vd->vdev_children; c++)
1795 vdev_dtl_reassess(vd->vdev_child[c], txg,
1796 scrub_txg, scrub_done);
1798 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1801 if (vd->vdev_ops->vdev_op_leaf) {
1802 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1804 mutex_enter(&vd->vdev_dtl_lock);
1807 * If we've completed a scan cleanly then determine
1808 * if this vdev should remove any DTLs. We only want to
1809 * excise regions on vdevs that were available during
1810 * the entire duration of this scan.
1812 if (scrub_txg != 0 &&
1813 (spa->spa_scrub_started ||
1814 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1815 vdev_dtl_should_excise(vd)) {
1817 * We completed a scrub up to scrub_txg. If we
1818 * did it without rebooting, then the scrub dtl
1819 * will be valid, so excise the old region and
1820 * fold in the scrub dtl. Otherwise, leave the
1821 * dtl as-is if there was an error.
1823 * There's little trick here: to excise the beginning
1824 * of the DTL_MISSING map, we put it into a reference
1825 * tree and then add a segment with refcnt -1 that
1826 * covers the range [0, scrub_txg). This means
1827 * that each txg in that range has refcnt -1 or 0.
1828 * We then add DTL_SCRUB with a refcnt of 2, so that
1829 * entries in the range [0, scrub_txg) will have a
1830 * positive refcnt -- either 1 or 2. We then convert
1831 * the reference tree into the new DTL_MISSING map.
1833 space_reftree_create(&reftree);
1834 space_reftree_add_map(&reftree,
1835 vd->vdev_dtl[DTL_MISSING], 1);
1836 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1837 space_reftree_add_map(&reftree,
1838 vd->vdev_dtl[DTL_SCRUB], 2);
1839 space_reftree_generate_map(&reftree,
1840 vd->vdev_dtl[DTL_MISSING], 1);
1841 space_reftree_destroy(&reftree);
1843 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1844 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1845 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1847 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1848 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1849 if (!vdev_readable(vd))
1850 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1852 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1853 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1856 * If the vdev was resilvering and no longer has any
1857 * DTLs then reset its resilvering flag.
1859 if (vd->vdev_resilver_txg != 0 &&
1860 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1861 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0)
1862 vd->vdev_resilver_txg = 0;
1864 mutex_exit(&vd->vdev_dtl_lock);
1867 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1871 mutex_enter(&vd->vdev_dtl_lock);
1872 for (t = 0; t < DTL_TYPES; t++) {
1875 /* account for child's outage in parent's missing map */
1876 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1878 continue; /* leaf vdevs only */
1879 if (t == DTL_PARTIAL)
1880 minref = 1; /* i.e. non-zero */
1881 else if (vd->vdev_nparity != 0)
1882 minref = vd->vdev_nparity + 1; /* RAID-Z */
1884 minref = vd->vdev_children; /* any kind of mirror */
1885 space_reftree_create(&reftree);
1886 for (c = 0; c < vd->vdev_children; c++) {
1887 vdev_t *cvd = vd->vdev_child[c];
1888 mutex_enter(&cvd->vdev_dtl_lock);
1889 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1890 mutex_exit(&cvd->vdev_dtl_lock);
1892 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
1893 space_reftree_destroy(&reftree);
1895 mutex_exit(&vd->vdev_dtl_lock);
1899 vdev_dtl_load(vdev_t *vd)
1901 spa_t *spa = vd->vdev_spa;
1902 objset_t *mos = spa->spa_meta_objset;
1906 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
1907 ASSERT(!vd->vdev_ishole);
1909 error = space_map_open(&vd->vdev_dtl_sm, mos,
1910 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
1913 ASSERT(vd->vdev_dtl_sm != NULL);
1915 mutex_enter(&vd->vdev_dtl_lock);
1918 * Now that we've opened the space_map we need to update
1921 space_map_update(vd->vdev_dtl_sm);
1923 error = space_map_load(vd->vdev_dtl_sm,
1924 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
1925 mutex_exit(&vd->vdev_dtl_lock);
1930 for (c = 0; c < vd->vdev_children; c++) {
1931 error = vdev_dtl_load(vd->vdev_child[c]);
1940 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1942 spa_t *spa = vd->vdev_spa;
1943 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
1944 objset_t *mos = spa->spa_meta_objset;
1945 range_tree_t *rtsync;
1948 uint64_t object = space_map_object(vd->vdev_dtl_sm);
1950 ASSERT(!vd->vdev_ishole);
1951 ASSERT(vd->vdev_ops->vdev_op_leaf);
1953 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1955 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
1956 mutex_enter(&vd->vdev_dtl_lock);
1957 space_map_free(vd->vdev_dtl_sm, tx);
1958 space_map_close(vd->vdev_dtl_sm);
1959 vd->vdev_dtl_sm = NULL;
1960 mutex_exit(&vd->vdev_dtl_lock);
1965 if (vd->vdev_dtl_sm == NULL) {
1966 uint64_t new_object;
1968 new_object = space_map_alloc(mos, tx);
1969 VERIFY3U(new_object, !=, 0);
1971 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
1972 0, -1ULL, 0, &vd->vdev_dtl_lock));
1973 ASSERT(vd->vdev_dtl_sm != NULL);
1976 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
1978 rtsync = range_tree_create(NULL, NULL, &rtlock);
1980 mutex_enter(&rtlock);
1982 mutex_enter(&vd->vdev_dtl_lock);
1983 range_tree_walk(rt, range_tree_add, rtsync);
1984 mutex_exit(&vd->vdev_dtl_lock);
1986 space_map_truncate(vd->vdev_dtl_sm, tx);
1987 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
1988 range_tree_vacate(rtsync, NULL, NULL);
1990 range_tree_destroy(rtsync);
1992 mutex_exit(&rtlock);
1993 mutex_destroy(&rtlock);
1996 * If the object for the space map has changed then dirty
1997 * the top level so that we update the config.
1999 if (object != space_map_object(vd->vdev_dtl_sm)) {
2000 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2001 "new object %llu", txg, spa_name(spa), object,
2002 space_map_object(vd->vdev_dtl_sm));
2003 vdev_config_dirty(vd->vdev_top);
2008 mutex_enter(&vd->vdev_dtl_lock);
2009 space_map_update(vd->vdev_dtl_sm);
2010 mutex_exit(&vd->vdev_dtl_lock);
2014 * Determine whether the specified vdev can be offlined/detached/removed
2015 * without losing data.
2018 vdev_dtl_required(vdev_t *vd)
2020 spa_t *spa = vd->vdev_spa;
2021 vdev_t *tvd = vd->vdev_top;
2022 uint8_t cant_read = vd->vdev_cant_read;
2025 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2027 if (vd == spa->spa_root_vdev || vd == tvd)
2031 * Temporarily mark the device as unreadable, and then determine
2032 * whether this results in any DTL outages in the top-level vdev.
2033 * If not, we can safely offline/detach/remove the device.
2035 vd->vdev_cant_read = B_TRUE;
2036 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2037 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2038 vd->vdev_cant_read = cant_read;
2039 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2041 if (!required && zio_injection_enabled)
2042 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2048 * Determine if resilver is needed, and if so the txg range.
2051 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2053 boolean_t needed = B_FALSE;
2054 uint64_t thismin = UINT64_MAX;
2055 uint64_t thismax = 0;
2058 if (vd->vdev_children == 0) {
2059 mutex_enter(&vd->vdev_dtl_lock);
2060 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2061 vdev_writeable(vd)) {
2063 thismin = vdev_dtl_min(vd);
2064 thismax = vdev_dtl_max(vd);
2067 mutex_exit(&vd->vdev_dtl_lock);
2069 for (c = 0; c < vd->vdev_children; c++) {
2070 vdev_t *cvd = vd->vdev_child[c];
2071 uint64_t cmin, cmax;
2073 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2074 thismin = MIN(thismin, cmin);
2075 thismax = MAX(thismax, cmax);
2081 if (needed && minp) {
2089 vdev_load(vdev_t *vd)
2094 * Recursively load all children.
2096 for (c = 0; c < vd->vdev_children; c++)
2097 vdev_load(vd->vdev_child[c]);
2100 * If this is a top-level vdev, initialize its metaslabs.
2102 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2103 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2104 vdev_metaslab_init(vd, 0) != 0))
2105 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2106 VDEV_AUX_CORRUPT_DATA);
2109 * If this is a leaf vdev, load its DTL.
2111 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2112 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2113 VDEV_AUX_CORRUPT_DATA);
2117 * The special vdev case is used for hot spares and l2cache devices. Its
2118 * sole purpose it to set the vdev state for the associated vdev. To do this,
2119 * we make sure that we can open the underlying device, then try to read the
2120 * label, and make sure that the label is sane and that it hasn't been
2121 * repurposed to another pool.
2124 vdev_validate_aux(vdev_t *vd)
2127 uint64_t guid, version;
2130 if (!vdev_readable(vd))
2133 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2134 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2135 VDEV_AUX_CORRUPT_DATA);
2139 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2140 !SPA_VERSION_IS_SUPPORTED(version) ||
2141 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2142 guid != vd->vdev_guid ||
2143 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2144 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2145 VDEV_AUX_CORRUPT_DATA);
2151 * We don't actually check the pool state here. If it's in fact in
2152 * use by another pool, we update this fact on the fly when requested.
2159 vdev_remove(vdev_t *vd, uint64_t txg)
2161 spa_t *spa = vd->vdev_spa;
2162 objset_t *mos = spa->spa_meta_objset;
2166 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2168 if (vd->vdev_ms != NULL) {
2169 metaslab_group_t *mg = vd->vdev_mg;
2171 metaslab_group_histogram_verify(mg);
2172 metaslab_class_histogram_verify(mg->mg_class);
2174 for (m = 0; m < vd->vdev_ms_count; m++) {
2175 metaslab_t *msp = vd->vdev_ms[m];
2177 if (msp == NULL || msp->ms_sm == NULL)
2180 mutex_enter(&msp->ms_lock);
2182 * If the metaslab was not loaded when the vdev
2183 * was removed then the histogram accounting may
2184 * not be accurate. Update the histogram information
2185 * here so that we ensure that the metaslab group
2186 * and metaslab class are up-to-date.
2188 metaslab_group_histogram_remove(mg, msp);
2190 VERIFY0(space_map_allocated(msp->ms_sm));
2191 space_map_free(msp->ms_sm, tx);
2192 space_map_close(msp->ms_sm);
2194 mutex_exit(&msp->ms_lock);
2197 metaslab_group_histogram_verify(mg);
2198 metaslab_class_histogram_verify(mg->mg_class);
2199 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2200 ASSERT0(mg->mg_histogram[i]);
2204 if (vd->vdev_ms_array) {
2205 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2206 vd->vdev_ms_array = 0;
2212 vdev_sync_done(vdev_t *vd, uint64_t txg)
2215 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2217 ASSERT(!vd->vdev_ishole);
2219 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
2220 metaslab_sync_done(msp, txg);
2223 metaslab_sync_reassess(vd->vdev_mg);
2227 vdev_sync(vdev_t *vd, uint64_t txg)
2229 spa_t *spa = vd->vdev_spa;
2234 ASSERT(!vd->vdev_ishole);
2236 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2237 ASSERT(vd == vd->vdev_top);
2238 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2239 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2240 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2241 ASSERT(vd->vdev_ms_array != 0);
2242 vdev_config_dirty(vd);
2247 * Remove the metadata associated with this vdev once it's empty.
2249 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2250 vdev_remove(vd, txg);
2252 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2253 metaslab_sync(msp, txg);
2254 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2257 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2258 vdev_dtl_sync(lvd, txg);
2260 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2264 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2266 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2270 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2271 * not be opened, and no I/O is attempted.
2274 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2278 spa_vdev_state_enter(spa, SCL_NONE);
2280 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2281 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2283 if (!vd->vdev_ops->vdev_op_leaf)
2284 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2289 * We don't directly use the aux state here, but if we do a
2290 * vdev_reopen(), we need this value to be present to remember why we
2293 vd->vdev_label_aux = aux;
2296 * Faulted state takes precedence over degraded.
2298 vd->vdev_delayed_close = B_FALSE;
2299 vd->vdev_faulted = 1ULL;
2300 vd->vdev_degraded = 0ULL;
2301 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2304 * If this device has the only valid copy of the data, then
2305 * back off and simply mark the vdev as degraded instead.
2307 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2308 vd->vdev_degraded = 1ULL;
2309 vd->vdev_faulted = 0ULL;
2312 * If we reopen the device and it's not dead, only then do we
2317 if (vdev_readable(vd))
2318 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2321 return (spa_vdev_state_exit(spa, vd, 0));
2325 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2326 * user that something is wrong. The vdev continues to operate as normal as far
2327 * as I/O is concerned.
2330 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2334 spa_vdev_state_enter(spa, SCL_NONE);
2336 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2337 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2339 if (!vd->vdev_ops->vdev_op_leaf)
2340 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2343 * If the vdev is already faulted, then don't do anything.
2345 if (vd->vdev_faulted || vd->vdev_degraded)
2346 return (spa_vdev_state_exit(spa, NULL, 0));
2348 vd->vdev_degraded = 1ULL;
2349 if (!vdev_is_dead(vd))
2350 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2353 return (spa_vdev_state_exit(spa, vd, 0));
2357 * Online the given vdev.
2359 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2360 * spare device should be detached when the device finishes resilvering.
2361 * Second, the online should be treated like a 'test' online case, so no FMA
2362 * events are generated if the device fails to open.
2365 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2367 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2369 spa_vdev_state_enter(spa, SCL_NONE);
2371 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2372 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2374 if (!vd->vdev_ops->vdev_op_leaf)
2375 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2378 vd->vdev_offline = B_FALSE;
2379 vd->vdev_tmpoffline = B_FALSE;
2380 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2381 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2383 /* XXX - L2ARC 1.0 does not support expansion */
2384 if (!vd->vdev_aux) {
2385 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2386 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2390 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2392 if (!vd->vdev_aux) {
2393 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2394 pvd->vdev_expanding = B_FALSE;
2398 *newstate = vd->vdev_state;
2399 if ((flags & ZFS_ONLINE_UNSPARE) &&
2400 !vdev_is_dead(vd) && vd->vdev_parent &&
2401 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2402 vd->vdev_parent->vdev_child[0] == vd)
2403 vd->vdev_unspare = B_TRUE;
2405 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2407 /* XXX - L2ARC 1.0 does not support expansion */
2409 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2410 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2412 return (spa_vdev_state_exit(spa, vd, 0));
2416 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2420 uint64_t generation;
2421 metaslab_group_t *mg;
2424 spa_vdev_state_enter(spa, SCL_ALLOC);
2426 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2427 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2429 if (!vd->vdev_ops->vdev_op_leaf)
2430 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2434 generation = spa->spa_config_generation + 1;
2437 * If the device isn't already offline, try to offline it.
2439 if (!vd->vdev_offline) {
2441 * If this device has the only valid copy of some data,
2442 * don't allow it to be offlined. Log devices are always
2445 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2446 vdev_dtl_required(vd))
2447 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2450 * If the top-level is a slog and it has had allocations
2451 * then proceed. We check that the vdev's metaslab group
2452 * is not NULL since it's possible that we may have just
2453 * added this vdev but not yet initialized its metaslabs.
2455 if (tvd->vdev_islog && mg != NULL) {
2457 * Prevent any future allocations.
2459 metaslab_group_passivate(mg);
2460 (void) spa_vdev_state_exit(spa, vd, 0);
2462 error = spa_offline_log(spa);
2464 spa_vdev_state_enter(spa, SCL_ALLOC);
2467 * Check to see if the config has changed.
2469 if (error || generation != spa->spa_config_generation) {
2470 metaslab_group_activate(mg);
2472 return (spa_vdev_state_exit(spa,
2474 (void) spa_vdev_state_exit(spa, vd, 0);
2477 ASSERT0(tvd->vdev_stat.vs_alloc);
2481 * Offline this device and reopen its top-level vdev.
2482 * If the top-level vdev is a log device then just offline
2483 * it. Otherwise, if this action results in the top-level
2484 * vdev becoming unusable, undo it and fail the request.
2486 vd->vdev_offline = B_TRUE;
2489 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2490 vdev_is_dead(tvd)) {
2491 vd->vdev_offline = B_FALSE;
2493 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2497 * Add the device back into the metaslab rotor so that
2498 * once we online the device it's open for business.
2500 if (tvd->vdev_islog && mg != NULL)
2501 metaslab_group_activate(mg);
2504 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2506 return (spa_vdev_state_exit(spa, vd, 0));
2510 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2514 mutex_enter(&spa->spa_vdev_top_lock);
2515 error = vdev_offline_locked(spa, guid, flags);
2516 mutex_exit(&spa->spa_vdev_top_lock);
2522 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2523 * vdev_offline(), we assume the spa config is locked. We also clear all
2524 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2527 vdev_clear(spa_t *spa, vdev_t *vd)
2529 vdev_t *rvd = spa->spa_root_vdev;
2532 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2537 vd->vdev_stat.vs_read_errors = 0;
2538 vd->vdev_stat.vs_write_errors = 0;
2539 vd->vdev_stat.vs_checksum_errors = 0;
2541 for (c = 0; c < vd->vdev_children; c++)
2542 vdev_clear(spa, vd->vdev_child[c]);
2545 * If we're in the FAULTED state or have experienced failed I/O, then
2546 * clear the persistent state and attempt to reopen the device. We
2547 * also mark the vdev config dirty, so that the new faulted state is
2548 * written out to disk.
2550 if (vd->vdev_faulted || vd->vdev_degraded ||
2551 !vdev_readable(vd) || !vdev_writeable(vd)) {
2554 * When reopening in reponse to a clear event, it may be due to
2555 * a fmadm repair request. In this case, if the device is
2556 * still broken, we want to still post the ereport again.
2558 vd->vdev_forcefault = B_TRUE;
2560 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2561 vd->vdev_cant_read = B_FALSE;
2562 vd->vdev_cant_write = B_FALSE;
2564 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2566 vd->vdev_forcefault = B_FALSE;
2568 if (vd != rvd && vdev_writeable(vd->vdev_top))
2569 vdev_state_dirty(vd->vdev_top);
2571 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2572 spa_async_request(spa, SPA_ASYNC_RESILVER);
2574 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR);
2578 * When clearing a FMA-diagnosed fault, we always want to
2579 * unspare the device, as we assume that the original spare was
2580 * done in response to the FMA fault.
2582 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2583 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2584 vd->vdev_parent->vdev_child[0] == vd)
2585 vd->vdev_unspare = B_TRUE;
2589 vdev_is_dead(vdev_t *vd)
2592 * Holes and missing devices are always considered "dead".
2593 * This simplifies the code since we don't have to check for
2594 * these types of devices in the various code paths.
2595 * Instead we rely on the fact that we skip over dead devices
2596 * before issuing I/O to them.
2598 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2599 vd->vdev_ops == &vdev_missing_ops);
2603 vdev_readable(vdev_t *vd)
2605 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2609 vdev_writeable(vdev_t *vd)
2611 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2615 vdev_allocatable(vdev_t *vd)
2617 uint64_t state = vd->vdev_state;
2620 * We currently allow allocations from vdevs which may be in the
2621 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2622 * fails to reopen then we'll catch it later when we're holding
2623 * the proper locks. Note that we have to get the vdev state
2624 * in a local variable because although it changes atomically,
2625 * we're asking two separate questions about it.
2627 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2628 !vd->vdev_cant_write && !vd->vdev_ishole);
2632 vdev_accessible(vdev_t *vd, zio_t *zio)
2634 ASSERT(zio->io_vd == vd);
2636 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2639 if (zio->io_type == ZIO_TYPE_READ)
2640 return (!vd->vdev_cant_read);
2642 if (zio->io_type == ZIO_TYPE_WRITE)
2643 return (!vd->vdev_cant_write);
2649 * Get statistics for the given vdev.
2652 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2654 spa_t *spa = vd->vdev_spa;
2655 vdev_t *rvd = spa->spa_root_vdev;
2658 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2660 mutex_enter(&vd->vdev_stat_lock);
2661 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2662 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2663 vs->vs_state = vd->vdev_state;
2664 vs->vs_rsize = vdev_get_min_asize(vd);
2665 if (vd->vdev_ops->vdev_op_leaf)
2666 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2667 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2668 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2669 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2673 * If we're getting stats on the root vdev, aggregate the I/O counts
2674 * over all top-level vdevs (i.e. the direct children of the root).
2677 for (c = 0; c < rvd->vdev_children; c++) {
2678 vdev_t *cvd = rvd->vdev_child[c];
2679 vdev_stat_t *cvs = &cvd->vdev_stat;
2681 for (t = 0; t < ZIO_TYPES; t++) {
2682 vs->vs_ops[t] += cvs->vs_ops[t];
2683 vs->vs_bytes[t] += cvs->vs_bytes[t];
2685 cvs->vs_scan_removing = cvd->vdev_removing;
2688 mutex_exit(&vd->vdev_stat_lock);
2692 vdev_clear_stats(vdev_t *vd)
2694 mutex_enter(&vd->vdev_stat_lock);
2695 vd->vdev_stat.vs_space = 0;
2696 vd->vdev_stat.vs_dspace = 0;
2697 vd->vdev_stat.vs_alloc = 0;
2698 mutex_exit(&vd->vdev_stat_lock);
2702 vdev_scan_stat_init(vdev_t *vd)
2704 vdev_stat_t *vs = &vd->vdev_stat;
2707 for (c = 0; c < vd->vdev_children; c++)
2708 vdev_scan_stat_init(vd->vdev_child[c]);
2710 mutex_enter(&vd->vdev_stat_lock);
2711 vs->vs_scan_processed = 0;
2712 mutex_exit(&vd->vdev_stat_lock);
2716 vdev_stat_update(zio_t *zio, uint64_t psize)
2718 spa_t *spa = zio->io_spa;
2719 vdev_t *rvd = spa->spa_root_vdev;
2720 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2722 uint64_t txg = zio->io_txg;
2723 vdev_stat_t *vs = &vd->vdev_stat;
2724 zio_type_t type = zio->io_type;
2725 int flags = zio->io_flags;
2728 * If this i/o is a gang leader, it didn't do any actual work.
2730 if (zio->io_gang_tree)
2733 if (zio->io_error == 0) {
2735 * If this is a root i/o, don't count it -- we've already
2736 * counted the top-level vdevs, and vdev_get_stats() will
2737 * aggregate them when asked. This reduces contention on
2738 * the root vdev_stat_lock and implicitly handles blocks
2739 * that compress away to holes, for which there is no i/o.
2740 * (Holes never create vdev children, so all the counters
2741 * remain zero, which is what we want.)
2743 * Note: this only applies to successful i/o (io_error == 0)
2744 * because unlike i/o counts, errors are not additive.
2745 * When reading a ditto block, for example, failure of
2746 * one top-level vdev does not imply a root-level error.
2751 ASSERT(vd == zio->io_vd);
2753 if (flags & ZIO_FLAG_IO_BYPASS)
2756 mutex_enter(&vd->vdev_stat_lock);
2758 if (flags & ZIO_FLAG_IO_REPAIR) {
2759 if (flags & ZIO_FLAG_SCAN_THREAD) {
2760 dsl_scan_phys_t *scn_phys =
2761 &spa->spa_dsl_pool->dp_scan->scn_phys;
2762 uint64_t *processed = &scn_phys->scn_processed;
2765 if (vd->vdev_ops->vdev_op_leaf)
2766 atomic_add_64(processed, psize);
2767 vs->vs_scan_processed += psize;
2770 if (flags & ZIO_FLAG_SELF_HEAL)
2771 vs->vs_self_healed += psize;
2775 vs->vs_bytes[type] += psize;
2777 mutex_exit(&vd->vdev_stat_lock);
2781 if (flags & ZIO_FLAG_SPECULATIVE)
2785 * If this is an I/O error that is going to be retried, then ignore the
2786 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2787 * hard errors, when in reality they can happen for any number of
2788 * innocuous reasons (bus resets, MPxIO link failure, etc).
2790 if (zio->io_error == EIO &&
2791 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2795 * Intent logs writes won't propagate their error to the root
2796 * I/O so don't mark these types of failures as pool-level
2799 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2802 mutex_enter(&vd->vdev_stat_lock);
2803 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2804 if (zio->io_error == ECKSUM)
2805 vs->vs_checksum_errors++;
2807 vs->vs_read_errors++;
2809 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2810 vs->vs_write_errors++;
2811 mutex_exit(&vd->vdev_stat_lock);
2813 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2814 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2815 (flags & ZIO_FLAG_SCAN_THREAD) ||
2816 spa->spa_claiming)) {
2818 * This is either a normal write (not a repair), or it's
2819 * a repair induced by the scrub thread, or it's a repair
2820 * made by zil_claim() during spa_load() in the first txg.
2821 * In the normal case, we commit the DTL change in the same
2822 * txg as the block was born. In the scrub-induced repair
2823 * case, we know that scrubs run in first-pass syncing context,
2824 * so we commit the DTL change in spa_syncing_txg(spa).
2825 * In the zil_claim() case, we commit in spa_first_txg(spa).
2827 * We currently do not make DTL entries for failed spontaneous
2828 * self-healing writes triggered by normal (non-scrubbing)
2829 * reads, because we have no transactional context in which to
2830 * do so -- and it's not clear that it'd be desirable anyway.
2832 if (vd->vdev_ops->vdev_op_leaf) {
2833 uint64_t commit_txg = txg;
2834 if (flags & ZIO_FLAG_SCAN_THREAD) {
2835 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2836 ASSERT(spa_sync_pass(spa) == 1);
2837 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2838 commit_txg = spa_syncing_txg(spa);
2839 } else if (spa->spa_claiming) {
2840 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2841 commit_txg = spa_first_txg(spa);
2843 ASSERT(commit_txg >= spa_syncing_txg(spa));
2844 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2846 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2847 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2848 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2851 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2856 * Update the in-core space usage stats for this vdev, its metaslab class,
2857 * and the root vdev.
2860 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2861 int64_t space_delta)
2863 int64_t dspace_delta = space_delta;
2864 spa_t *spa = vd->vdev_spa;
2865 vdev_t *rvd = spa->spa_root_vdev;
2866 metaslab_group_t *mg = vd->vdev_mg;
2867 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2869 ASSERT(vd == vd->vdev_top);
2872 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2873 * factor. We must calculate this here and not at the root vdev
2874 * because the root vdev's psize-to-asize is simply the max of its
2875 * childrens', thus not accurate enough for us.
2877 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2878 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2879 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2880 vd->vdev_deflate_ratio;
2882 mutex_enter(&vd->vdev_stat_lock);
2883 vd->vdev_stat.vs_alloc += alloc_delta;
2884 vd->vdev_stat.vs_space += space_delta;
2885 vd->vdev_stat.vs_dspace += dspace_delta;
2886 mutex_exit(&vd->vdev_stat_lock);
2888 if (mc == spa_normal_class(spa)) {
2889 mutex_enter(&rvd->vdev_stat_lock);
2890 rvd->vdev_stat.vs_alloc += alloc_delta;
2891 rvd->vdev_stat.vs_space += space_delta;
2892 rvd->vdev_stat.vs_dspace += dspace_delta;
2893 mutex_exit(&rvd->vdev_stat_lock);
2897 ASSERT(rvd == vd->vdev_parent);
2898 ASSERT(vd->vdev_ms_count != 0);
2900 metaslab_class_space_update(mc,
2901 alloc_delta, defer_delta, space_delta, dspace_delta);
2906 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2907 * so that it will be written out next time the vdev configuration is synced.
2908 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2911 vdev_config_dirty(vdev_t *vd)
2913 spa_t *spa = vd->vdev_spa;
2914 vdev_t *rvd = spa->spa_root_vdev;
2917 ASSERT(spa_writeable(spa));
2920 * If this is an aux vdev (as with l2cache and spare devices), then we
2921 * update the vdev config manually and set the sync flag.
2923 if (vd->vdev_aux != NULL) {
2924 spa_aux_vdev_t *sav = vd->vdev_aux;
2928 for (c = 0; c < sav->sav_count; c++) {
2929 if (sav->sav_vdevs[c] == vd)
2933 if (c == sav->sav_count) {
2935 * We're being removed. There's nothing more to do.
2937 ASSERT(sav->sav_sync == B_TRUE);
2941 sav->sav_sync = B_TRUE;
2943 if (nvlist_lookup_nvlist_array(sav->sav_config,
2944 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2945 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2946 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2952 * Setting the nvlist in the middle if the array is a little
2953 * sketchy, but it will work.
2955 nvlist_free(aux[c]);
2956 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2962 * The dirty list is protected by the SCL_CONFIG lock. The caller
2963 * must either hold SCL_CONFIG as writer, or must be the sync thread
2964 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2965 * so this is sufficient to ensure mutual exclusion.
2967 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2968 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2969 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2972 for (c = 0; c < rvd->vdev_children; c++)
2973 vdev_config_dirty(rvd->vdev_child[c]);
2975 ASSERT(vd == vd->vdev_top);
2977 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2979 list_insert_head(&spa->spa_config_dirty_list, vd);
2984 vdev_config_clean(vdev_t *vd)
2986 spa_t *spa = vd->vdev_spa;
2988 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2989 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2990 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2992 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2993 list_remove(&spa->spa_config_dirty_list, vd);
2997 * Mark a top-level vdev's state as dirty, so that the next pass of
2998 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2999 * the state changes from larger config changes because they require
3000 * much less locking, and are often needed for administrative actions.
3003 vdev_state_dirty(vdev_t *vd)
3005 spa_t *spa = vd->vdev_spa;
3007 ASSERT(spa_writeable(spa));
3008 ASSERT(vd == vd->vdev_top);
3011 * The state list is protected by the SCL_STATE lock. The caller
3012 * must either hold SCL_STATE as writer, or must be the sync thread
3013 * (which holds SCL_STATE as reader). There's only one sync thread,
3014 * so this is sufficient to ensure mutual exclusion.
3016 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3017 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3018 spa_config_held(spa, SCL_STATE, RW_READER)));
3020 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3021 list_insert_head(&spa->spa_state_dirty_list, vd);
3025 vdev_state_clean(vdev_t *vd)
3027 spa_t *spa = vd->vdev_spa;
3029 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3030 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3031 spa_config_held(spa, SCL_STATE, RW_READER)));
3033 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3034 list_remove(&spa->spa_state_dirty_list, vd);
3038 * Propagate vdev state up from children to parent.
3041 vdev_propagate_state(vdev_t *vd)
3043 spa_t *spa = vd->vdev_spa;
3044 vdev_t *rvd = spa->spa_root_vdev;
3045 int degraded = 0, faulted = 0;
3050 if (vd->vdev_children > 0) {
3051 for (c = 0; c < vd->vdev_children; c++) {
3052 child = vd->vdev_child[c];
3055 * Don't factor holes into the decision.
3057 if (child->vdev_ishole)
3060 if (!vdev_readable(child) ||
3061 (!vdev_writeable(child) && spa_writeable(spa))) {
3063 * Root special: if there is a top-level log
3064 * device, treat the root vdev as if it were
3067 if (child->vdev_islog && vd == rvd)
3071 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3075 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3079 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3082 * Root special: if there is a top-level vdev that cannot be
3083 * opened due to corrupted metadata, then propagate the root
3084 * vdev's aux state as 'corrupt' rather than 'insufficient
3087 if (corrupted && vd == rvd &&
3088 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3089 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3090 VDEV_AUX_CORRUPT_DATA);
3093 if (vd->vdev_parent)
3094 vdev_propagate_state(vd->vdev_parent);
3098 * Set a vdev's state. If this is during an open, we don't update the parent
3099 * state, because we're in the process of opening children depth-first.
3100 * Otherwise, we propagate the change to the parent.
3102 * If this routine places a device in a faulted state, an appropriate ereport is
3106 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3108 uint64_t save_state;
3109 spa_t *spa = vd->vdev_spa;
3111 if (state == vd->vdev_state) {
3112 vd->vdev_stat.vs_aux = aux;
3116 save_state = vd->vdev_state;
3118 vd->vdev_state = state;
3119 vd->vdev_stat.vs_aux = aux;
3122 * If we are setting the vdev state to anything but an open state, then
3123 * always close the underlying device unless the device has requested
3124 * a delayed close (i.e. we're about to remove or fault the device).
3125 * Otherwise, we keep accessible but invalid devices open forever.
3126 * We don't call vdev_close() itself, because that implies some extra
3127 * checks (offline, etc) that we don't want here. This is limited to
3128 * leaf devices, because otherwise closing the device will affect other
3131 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3132 vd->vdev_ops->vdev_op_leaf)
3133 vd->vdev_ops->vdev_op_close(vd);
3136 * If we have brought this vdev back into service, we need
3137 * to notify fmd so that it can gracefully repair any outstanding
3138 * cases due to a missing device. We do this in all cases, even those
3139 * that probably don't correlate to a repaired fault. This is sure to
3140 * catch all cases, and we let the zfs-retire agent sort it out. If
3141 * this is a transient state it's OK, as the retire agent will
3142 * double-check the state of the vdev before repairing it.
3144 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3145 vd->vdev_prevstate != state)
3146 zfs_post_state_change(spa, vd);
3148 if (vd->vdev_removed &&
3149 state == VDEV_STATE_CANT_OPEN &&
3150 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3152 * If the previous state is set to VDEV_STATE_REMOVED, then this
3153 * device was previously marked removed and someone attempted to
3154 * reopen it. If this failed due to a nonexistent device, then
3155 * keep the device in the REMOVED state. We also let this be if
3156 * it is one of our special test online cases, which is only
3157 * attempting to online the device and shouldn't generate an FMA
3160 vd->vdev_state = VDEV_STATE_REMOVED;
3161 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3162 } else if (state == VDEV_STATE_REMOVED) {
3163 vd->vdev_removed = B_TRUE;
3164 } else if (state == VDEV_STATE_CANT_OPEN) {
3166 * If we fail to open a vdev during an import or recovery, we
3167 * mark it as "not available", which signifies that it was
3168 * never there to begin with. Failure to open such a device
3169 * is not considered an error.
3171 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3172 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3173 vd->vdev_ops->vdev_op_leaf)
3174 vd->vdev_not_present = 1;
3177 * Post the appropriate ereport. If the 'prevstate' field is
3178 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3179 * that this is part of a vdev_reopen(). In this case, we don't
3180 * want to post the ereport if the device was already in the
3181 * CANT_OPEN state beforehand.
3183 * If the 'checkremove' flag is set, then this is an attempt to
3184 * online the device in response to an insertion event. If we
3185 * hit this case, then we have detected an insertion event for a
3186 * faulted or offline device that wasn't in the removed state.
3187 * In this scenario, we don't post an ereport because we are
3188 * about to replace the device, or attempt an online with
3189 * vdev_forcefault, which will generate the fault for us.
3191 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3192 !vd->vdev_not_present && !vd->vdev_checkremove &&
3193 vd != spa->spa_root_vdev) {
3197 case VDEV_AUX_OPEN_FAILED:
3198 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3200 case VDEV_AUX_CORRUPT_DATA:
3201 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3203 case VDEV_AUX_NO_REPLICAS:
3204 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3206 case VDEV_AUX_BAD_GUID_SUM:
3207 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3209 case VDEV_AUX_TOO_SMALL:
3210 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3212 case VDEV_AUX_BAD_LABEL:
3213 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3216 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3219 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3222 /* Erase any notion of persistent removed state */
3223 vd->vdev_removed = B_FALSE;
3225 vd->vdev_removed = B_FALSE;
3228 if (!isopen && vd->vdev_parent)
3229 vdev_propagate_state(vd->vdev_parent);
3233 * Check the vdev configuration to ensure that it's capable of supporting
3237 vdev_is_bootable(vdev_t *vd)
3239 #if defined(__sun__) || defined(__sun)
3241 * Currently, we do not support RAID-Z or partial configuration.
3242 * In addition, only a single top-level vdev is allowed and none of the
3243 * leaves can be wholedisks.
3247 if (!vd->vdev_ops->vdev_op_leaf) {
3248 char *vdev_type = vd->vdev_ops->vdev_op_type;
3250 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3251 vd->vdev_children > 1) {
3253 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3254 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3257 } else if (vd->vdev_wholedisk == 1) {
3261 for (c = 0; c < vd->vdev_children; c++) {
3262 if (!vdev_is_bootable(vd->vdev_child[c]))
3265 #endif /* __sun__ || __sun */
3270 * Load the state from the original vdev tree (ovd) which
3271 * we've retrieved from the MOS config object. If the original
3272 * vdev was offline or faulted then we transfer that state to the
3273 * device in the current vdev tree (nvd).
3276 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3280 ASSERT(nvd->vdev_top->vdev_islog);
3281 ASSERT(spa_config_held(nvd->vdev_spa,
3282 SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3283 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3285 for (c = 0; c < nvd->vdev_children; c++)
3286 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3288 if (nvd->vdev_ops->vdev_op_leaf) {
3290 * Restore the persistent vdev state
3292 nvd->vdev_offline = ovd->vdev_offline;
3293 nvd->vdev_faulted = ovd->vdev_faulted;
3294 nvd->vdev_degraded = ovd->vdev_degraded;
3295 nvd->vdev_removed = ovd->vdev_removed;
3300 * Determine if a log device has valid content. If the vdev was
3301 * removed or faulted in the MOS config then we know that
3302 * the content on the log device has already been written to the pool.
3305 vdev_log_state_valid(vdev_t *vd)
3309 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3313 for (c = 0; c < vd->vdev_children; c++)
3314 if (vdev_log_state_valid(vd->vdev_child[c]))
3321 * Expand a vdev if possible.
3324 vdev_expand(vdev_t *vd, uint64_t txg)
3326 ASSERT(vd->vdev_top == vd);
3327 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3329 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3330 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3331 vdev_config_dirty(vd);
3339 vdev_split(vdev_t *vd)
3341 vdev_t *cvd, *pvd = vd->vdev_parent;
3343 vdev_remove_child(pvd, vd);
3344 vdev_compact_children(pvd);
3346 cvd = pvd->vdev_child[0];
3347 if (pvd->vdev_children == 1) {
3348 vdev_remove_parent(cvd);
3349 cvd->vdev_splitting = B_TRUE;
3351 vdev_propagate_state(cvd);
3355 vdev_deadman(vdev_t *vd)
3359 for (c = 0; c < vd->vdev_children; c++) {
3360 vdev_t *cvd = vd->vdev_child[c];
3365 if (vd->vdev_ops->vdev_op_leaf) {
3366 vdev_queue_t *vq = &vd->vdev_queue;
3368 mutex_enter(&vq->vq_lock);
3369 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3370 spa_t *spa = vd->vdev_spa;
3375 * Look at the head of all the pending queues,
3376 * if any I/O has been outstanding for longer than
3377 * the spa_deadman_synctime we log a zevent.
3379 fio = avl_first(&vq->vq_active_tree);
3380 delta = gethrtime() - fio->io_timestamp;
3381 if (delta > spa_deadman_synctime(spa)) {
3382 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3383 "delta %lluns, last io %lluns",
3384 fio->io_timestamp, delta,
3385 vq->vq_io_complete_ts);
3386 zfs_ereport_post(FM_EREPORT_ZFS_DELAY,
3387 spa, vd, fio, 0, 0);
3390 mutex_exit(&vq->vq_lock);
3394 #if defined(_KERNEL) && defined(HAVE_SPL)
3395 EXPORT_SYMBOL(vdev_fault);
3396 EXPORT_SYMBOL(vdev_degrade);
3397 EXPORT_SYMBOL(vdev_online);
3398 EXPORT_SYMBOL(vdev_offline);
3399 EXPORT_SYMBOL(vdev_clear);
3401 module_param(metaslabs_per_vdev, int, 0644);
3402 MODULE_PARM_DESC(metaslabs_per_vdev,
3403 "Divide added vdev into approximately (but no more than) this number "