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) 2013 by Delphix. All rights reserved.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/fm/fs/zfs.h>
32 #include <sys/spa_impl.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/vdev_impl.h>
36 #include <sys/uberblock_impl.h>
37 #include <sys/metaslab.h>
38 #include <sys/metaslab_impl.h>
39 #include <sys/space_map.h>
42 #include <sys/fs/zfs.h>
45 #include <sys/dsl_scan.h>
46 #include <sys/trim_map.h>
48 SYSCTL_DECL(_vfs_zfs);
49 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
52 * Virtual device management.
55 static vdev_ops_t *vdev_ops_table[] = {
74 * Given a vdev type, return the appropriate ops vector.
77 vdev_getops(const char *type)
79 vdev_ops_t *ops, **opspp;
81 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
82 if (strcmp(ops->vdev_op_type, type) == 0)
89 * Default asize function: return the MAX of psize with the asize of
90 * all children. This is what's used by anything other than RAID-Z.
93 vdev_default_asize(vdev_t *vd, uint64_t psize)
95 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
98 for (int c = 0; c < vd->vdev_children; c++) {
99 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
100 asize = MAX(asize, csize);
107 * Get the minimum allocatable size. We define the allocatable size as
108 * the vdev's asize rounded to the nearest metaslab. This allows us to
109 * replace or attach devices which don't have the same physical size but
110 * can still satisfy the same number of allocations.
113 vdev_get_min_asize(vdev_t *vd)
115 vdev_t *pvd = vd->vdev_parent;
118 * If our parent is NULL (inactive spare or cache) or is the root,
119 * just return our own asize.
122 return (vd->vdev_asize);
125 * The top-level vdev just returns the allocatable size rounded
126 * to the nearest metaslab.
128 if (vd == vd->vdev_top)
129 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
132 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
133 * so each child must provide at least 1/Nth of its asize.
135 if (pvd->vdev_ops == &vdev_raidz_ops)
136 return (pvd->vdev_min_asize / pvd->vdev_children);
138 return (pvd->vdev_min_asize);
142 vdev_set_min_asize(vdev_t *vd)
144 vd->vdev_min_asize = vdev_get_min_asize(vd);
146 for (int 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)
170 if (vd->vdev_guid == guid)
173 for (int c = 0; c < vd->vdev_children; c++)
174 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
182 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
184 size_t oldsize, newsize;
185 uint64_t id = cvd->vdev_id;
188 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
189 ASSERT(cvd->vdev_parent == NULL);
191 cvd->vdev_parent = pvd;
196 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
198 oldsize = pvd->vdev_children * sizeof (vdev_t *);
199 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
200 newsize = pvd->vdev_children * sizeof (vdev_t *);
202 newchild = kmem_zalloc(newsize, KM_SLEEP);
203 if (pvd->vdev_child != NULL) {
204 bcopy(pvd->vdev_child, newchild, oldsize);
205 kmem_free(pvd->vdev_child, oldsize);
208 pvd->vdev_child = newchild;
209 pvd->vdev_child[id] = cvd;
211 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
212 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
215 * Walk up all ancestors to update guid sum.
217 for (; pvd != NULL; pvd = pvd->vdev_parent)
218 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
222 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
225 uint_t id = cvd->vdev_id;
227 ASSERT(cvd->vdev_parent == pvd);
232 ASSERT(id < pvd->vdev_children);
233 ASSERT(pvd->vdev_child[id] == cvd);
235 pvd->vdev_child[id] = NULL;
236 cvd->vdev_parent = NULL;
238 for (c = 0; c < pvd->vdev_children; c++)
239 if (pvd->vdev_child[c])
242 if (c == pvd->vdev_children) {
243 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
244 pvd->vdev_child = NULL;
245 pvd->vdev_children = 0;
249 * Walk up all ancestors to update guid sum.
251 for (; pvd != NULL; pvd = pvd->vdev_parent)
252 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
256 * Remove any holes in the child array.
259 vdev_compact_children(vdev_t *pvd)
261 vdev_t **newchild, *cvd;
262 int oldc = pvd->vdev_children;
265 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
267 for (int c = newc = 0; c < oldc; c++)
268 if (pvd->vdev_child[c])
271 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
273 for (int c = newc = 0; c < oldc; c++) {
274 if ((cvd = pvd->vdev_child[c]) != NULL) {
275 newchild[newc] = cvd;
276 cvd->vdev_id = newc++;
280 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
281 pvd->vdev_child = newchild;
282 pvd->vdev_children = newc;
286 * Allocate and minimally initialize a vdev_t.
289 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
293 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
295 if (spa->spa_root_vdev == NULL) {
296 ASSERT(ops == &vdev_root_ops);
297 spa->spa_root_vdev = vd;
298 spa->spa_load_guid = spa_generate_guid(NULL);
301 if (guid == 0 && ops != &vdev_hole_ops) {
302 if (spa->spa_root_vdev == vd) {
304 * The root vdev's guid will also be the pool guid,
305 * which must be unique among all pools.
307 guid = spa_generate_guid(NULL);
310 * Any other vdev's guid must be unique within the pool.
312 guid = spa_generate_guid(spa);
314 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
319 vd->vdev_guid = guid;
320 vd->vdev_guid_sum = guid;
322 vd->vdev_state = VDEV_STATE_CLOSED;
323 vd->vdev_ishole = (ops == &vdev_hole_ops);
325 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
326 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
327 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
328 for (int t = 0; t < DTL_TYPES; t++) {
329 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
332 txg_list_create(&vd->vdev_ms_list,
333 offsetof(struct metaslab, ms_txg_node));
334 txg_list_create(&vd->vdev_dtl_list,
335 offsetof(struct vdev, vdev_dtl_node));
336 vd->vdev_stat.vs_timestamp = gethrtime();
344 * Allocate a new vdev. The 'alloctype' is used to control whether we are
345 * creating a new vdev or loading an existing one - the behavior is slightly
346 * different for each case.
349 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
354 uint64_t guid = 0, islog, nparity;
357 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
359 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
360 return (SET_ERROR(EINVAL));
362 if ((ops = vdev_getops(type)) == NULL)
363 return (SET_ERROR(EINVAL));
366 * If this is a load, get the vdev guid from the nvlist.
367 * Otherwise, vdev_alloc_common() will generate one for us.
369 if (alloctype == VDEV_ALLOC_LOAD) {
372 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
374 return (SET_ERROR(EINVAL));
376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
377 return (SET_ERROR(EINVAL));
378 } else if (alloctype == VDEV_ALLOC_SPARE) {
379 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
380 return (SET_ERROR(EINVAL));
381 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
382 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
383 return (SET_ERROR(EINVAL));
384 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
385 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
386 return (SET_ERROR(EINVAL));
390 * The first allocated vdev must be of type 'root'.
392 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
393 return (SET_ERROR(EINVAL));
396 * Determine whether we're a log vdev.
399 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
400 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
401 return (SET_ERROR(ENOTSUP));
403 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
404 return (SET_ERROR(ENOTSUP));
407 * Set the nparity property for RAID-Z vdevs.
410 if (ops == &vdev_raidz_ops) {
411 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
413 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
414 return (SET_ERROR(EINVAL));
416 * Previous versions could only support 1 or 2 parity
420 spa_version(spa) < SPA_VERSION_RAIDZ2)
421 return (SET_ERROR(ENOTSUP));
423 spa_version(spa) < SPA_VERSION_RAIDZ3)
424 return (SET_ERROR(ENOTSUP));
427 * We require the parity to be specified for SPAs that
428 * support multiple parity levels.
430 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
431 return (SET_ERROR(EINVAL));
433 * Otherwise, we default to 1 parity device for RAID-Z.
440 ASSERT(nparity != -1ULL);
442 vd = vdev_alloc_common(spa, id, guid, ops);
444 vd->vdev_islog = islog;
445 vd->vdev_nparity = nparity;
447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
448 vd->vdev_path = spa_strdup(vd->vdev_path);
449 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
450 vd->vdev_devid = spa_strdup(vd->vdev_devid);
451 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
452 &vd->vdev_physpath) == 0)
453 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
454 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
455 vd->vdev_fru = spa_strdup(vd->vdev_fru);
458 * Set the whole_disk property. If it's not specified, leave the value
461 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
462 &vd->vdev_wholedisk) != 0)
463 vd->vdev_wholedisk = -1ULL;
466 * Look for the 'not present' flag. This will only be set if the device
467 * was not present at the time of import.
469 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
470 &vd->vdev_not_present);
473 * Get the alignment requirement.
475 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
478 * Retrieve the vdev creation time.
480 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
484 * If we're a top-level vdev, try to load the allocation parameters.
486 if (parent && !parent->vdev_parent &&
487 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
488 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
490 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
494 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
498 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
499 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
500 alloctype == VDEV_ALLOC_ADD ||
501 alloctype == VDEV_ALLOC_SPLIT ||
502 alloctype == VDEV_ALLOC_ROOTPOOL);
503 vd->vdev_mg = metaslab_group_create(islog ?
504 spa_log_class(spa) : spa_normal_class(spa), vd);
508 * If we're a leaf vdev, try to load the DTL object and other state.
510 if (vd->vdev_ops->vdev_op_leaf &&
511 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
512 alloctype == VDEV_ALLOC_ROOTPOOL)) {
513 if (alloctype == VDEV_ALLOC_LOAD) {
514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
515 &vd->vdev_dtl_smo.smo_object);
516 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
520 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
523 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
524 &spare) == 0 && spare)
528 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
531 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
532 &vd->vdev_resilvering);
535 * When importing a pool, we want to ignore the persistent fault
536 * state, as the diagnosis made on another system may not be
537 * valid in the current context. Local vdevs will
538 * remain in the faulted state.
540 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
541 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
543 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
545 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
548 if (vd->vdev_faulted || vd->vdev_degraded) {
552 VDEV_AUX_ERR_EXCEEDED;
553 if (nvlist_lookup_string(nv,
554 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
555 strcmp(aux, "external") == 0)
556 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
562 * Add ourselves to the parent's list of children.
564 vdev_add_child(parent, vd);
572 vdev_free(vdev_t *vd)
574 spa_t *spa = vd->vdev_spa;
577 * vdev_free() implies closing the vdev first. This is simpler than
578 * trying to ensure complicated semantics for all callers.
582 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
583 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
588 for (int c = 0; c < vd->vdev_children; c++)
589 vdev_free(vd->vdev_child[c]);
591 ASSERT(vd->vdev_child == NULL);
592 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
595 * Discard allocation state.
597 if (vd->vdev_mg != NULL) {
598 vdev_metaslab_fini(vd);
599 metaslab_group_destroy(vd->vdev_mg);
602 ASSERT0(vd->vdev_stat.vs_space);
603 ASSERT0(vd->vdev_stat.vs_dspace);
604 ASSERT0(vd->vdev_stat.vs_alloc);
607 * Remove this vdev from its parent's child list.
609 vdev_remove_child(vd->vdev_parent, vd);
611 ASSERT(vd->vdev_parent == NULL);
614 * Clean up vdev structure.
620 spa_strfree(vd->vdev_path);
622 spa_strfree(vd->vdev_devid);
623 if (vd->vdev_physpath)
624 spa_strfree(vd->vdev_physpath);
626 spa_strfree(vd->vdev_fru);
628 if (vd->vdev_isspare)
629 spa_spare_remove(vd);
630 if (vd->vdev_isl2cache)
631 spa_l2cache_remove(vd);
633 txg_list_destroy(&vd->vdev_ms_list);
634 txg_list_destroy(&vd->vdev_dtl_list);
636 mutex_enter(&vd->vdev_dtl_lock);
637 for (int t = 0; t < DTL_TYPES; t++) {
638 space_map_unload(&vd->vdev_dtl[t]);
639 space_map_destroy(&vd->vdev_dtl[t]);
641 mutex_exit(&vd->vdev_dtl_lock);
643 mutex_destroy(&vd->vdev_dtl_lock);
644 mutex_destroy(&vd->vdev_stat_lock);
645 mutex_destroy(&vd->vdev_probe_lock);
647 if (vd == spa->spa_root_vdev)
648 spa->spa_root_vdev = NULL;
650 kmem_free(vd, sizeof (vdev_t));
654 * Transfer top-level vdev state from svd to tvd.
657 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
659 spa_t *spa = svd->vdev_spa;
664 ASSERT(tvd == tvd->vdev_top);
666 tvd->vdev_ms_array = svd->vdev_ms_array;
667 tvd->vdev_ms_shift = svd->vdev_ms_shift;
668 tvd->vdev_ms_count = svd->vdev_ms_count;
670 svd->vdev_ms_array = 0;
671 svd->vdev_ms_shift = 0;
672 svd->vdev_ms_count = 0;
675 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
676 tvd->vdev_mg = svd->vdev_mg;
677 tvd->vdev_ms = svd->vdev_ms;
682 if (tvd->vdev_mg != NULL)
683 tvd->vdev_mg->mg_vd = tvd;
685 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
686 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
687 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
689 svd->vdev_stat.vs_alloc = 0;
690 svd->vdev_stat.vs_space = 0;
691 svd->vdev_stat.vs_dspace = 0;
693 for (t = 0; t < TXG_SIZE; t++) {
694 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
695 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
696 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
697 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
698 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
699 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
702 if (list_link_active(&svd->vdev_config_dirty_node)) {
703 vdev_config_clean(svd);
704 vdev_config_dirty(tvd);
707 if (list_link_active(&svd->vdev_state_dirty_node)) {
708 vdev_state_clean(svd);
709 vdev_state_dirty(tvd);
712 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
713 svd->vdev_deflate_ratio = 0;
715 tvd->vdev_islog = svd->vdev_islog;
720 vdev_top_update(vdev_t *tvd, vdev_t *vd)
727 for (int c = 0; c < vd->vdev_children; c++)
728 vdev_top_update(tvd, vd->vdev_child[c]);
732 * Add a mirror/replacing vdev above an existing vdev.
735 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
737 spa_t *spa = cvd->vdev_spa;
738 vdev_t *pvd = cvd->vdev_parent;
741 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
743 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
745 mvd->vdev_asize = cvd->vdev_asize;
746 mvd->vdev_min_asize = cvd->vdev_min_asize;
747 mvd->vdev_max_asize = cvd->vdev_max_asize;
748 mvd->vdev_ashift = cvd->vdev_ashift;
749 mvd->vdev_state = cvd->vdev_state;
750 mvd->vdev_crtxg = cvd->vdev_crtxg;
752 vdev_remove_child(pvd, cvd);
753 vdev_add_child(pvd, mvd);
754 cvd->vdev_id = mvd->vdev_children;
755 vdev_add_child(mvd, cvd);
756 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
758 if (mvd == mvd->vdev_top)
759 vdev_top_transfer(cvd, mvd);
765 * Remove a 1-way mirror/replacing vdev from the tree.
768 vdev_remove_parent(vdev_t *cvd)
770 vdev_t *mvd = cvd->vdev_parent;
771 vdev_t *pvd = mvd->vdev_parent;
773 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
775 ASSERT(mvd->vdev_children == 1);
776 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
777 mvd->vdev_ops == &vdev_replacing_ops ||
778 mvd->vdev_ops == &vdev_spare_ops);
779 cvd->vdev_ashift = mvd->vdev_ashift;
781 vdev_remove_child(mvd, cvd);
782 vdev_remove_child(pvd, mvd);
785 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
786 * Otherwise, we could have detached an offline device, and when we
787 * go to import the pool we'll think we have two top-level vdevs,
788 * instead of a different version of the same top-level vdev.
790 if (mvd->vdev_top == mvd) {
791 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
792 cvd->vdev_orig_guid = cvd->vdev_guid;
793 cvd->vdev_guid += guid_delta;
794 cvd->vdev_guid_sum += guid_delta;
796 cvd->vdev_id = mvd->vdev_id;
797 vdev_add_child(pvd, cvd);
798 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
800 if (cvd == cvd->vdev_top)
801 vdev_top_transfer(mvd, cvd);
803 ASSERT(mvd->vdev_children == 0);
808 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
810 spa_t *spa = vd->vdev_spa;
811 objset_t *mos = spa->spa_meta_objset;
813 uint64_t oldc = vd->vdev_ms_count;
814 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
818 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
821 * This vdev is not being allocated from yet or is a hole.
823 if (vd->vdev_ms_shift == 0)
826 ASSERT(!vd->vdev_ishole);
829 * Compute the raidz-deflation ratio. Note, we hard-code
830 * in 128k (1 << 17) because it is the current "typical" blocksize.
831 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
832 * or we will inconsistently account for existing bp's.
834 vd->vdev_deflate_ratio = (1 << 17) /
835 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
837 ASSERT(oldc <= newc);
839 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
842 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
843 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
847 vd->vdev_ms_count = newc;
849 for (m = oldc; m < newc; m++) {
850 space_map_obj_t smo = { 0, 0, 0 };
853 error = dmu_read(mos, vd->vdev_ms_array,
854 m * sizeof (uint64_t), sizeof (uint64_t), &object,
860 error = dmu_bonus_hold(mos, object, FTAG, &db);
863 ASSERT3U(db->db_size, >=, sizeof (smo));
864 bcopy(db->db_data, &smo, sizeof (smo));
865 ASSERT3U(smo.smo_object, ==, object);
866 dmu_buf_rele(db, FTAG);
869 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
870 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
874 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
877 * If the vdev is being removed we don't activate
878 * the metaslabs since we want to ensure that no new
879 * allocations are performed on this device.
881 if (oldc == 0 && !vd->vdev_removing)
882 metaslab_group_activate(vd->vdev_mg);
885 spa_config_exit(spa, SCL_ALLOC, FTAG);
891 vdev_metaslab_fini(vdev_t *vd)
894 uint64_t count = vd->vdev_ms_count;
896 if (vd->vdev_ms != NULL) {
897 metaslab_group_passivate(vd->vdev_mg);
898 for (m = 0; m < count; m++)
899 if (vd->vdev_ms[m] != NULL)
900 metaslab_fini(vd->vdev_ms[m]);
901 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
906 typedef struct vdev_probe_stats {
907 boolean_t vps_readable;
908 boolean_t vps_writeable;
910 } vdev_probe_stats_t;
913 vdev_probe_done(zio_t *zio)
915 spa_t *spa = zio->io_spa;
916 vdev_t *vd = zio->io_vd;
917 vdev_probe_stats_t *vps = zio->io_private;
919 ASSERT(vd->vdev_probe_zio != NULL);
921 if (zio->io_type == ZIO_TYPE_READ) {
922 if (zio->io_error == 0)
923 vps->vps_readable = 1;
924 if (zio->io_error == 0 && spa_writeable(spa)) {
925 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
926 zio->io_offset, zio->io_size, zio->io_data,
927 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
928 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
930 zio_buf_free(zio->io_data, zio->io_size);
932 } else if (zio->io_type == ZIO_TYPE_WRITE) {
933 if (zio->io_error == 0)
934 vps->vps_writeable = 1;
935 zio_buf_free(zio->io_data, zio->io_size);
936 } else if (zio->io_type == ZIO_TYPE_NULL) {
939 vd->vdev_cant_read |= !vps->vps_readable;
940 vd->vdev_cant_write |= !vps->vps_writeable;
942 if (vdev_readable(vd) &&
943 (vdev_writeable(vd) || !spa_writeable(spa))) {
946 ASSERT(zio->io_error != 0);
947 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
948 spa, vd, NULL, 0, 0);
949 zio->io_error = SET_ERROR(ENXIO);
952 mutex_enter(&vd->vdev_probe_lock);
953 ASSERT(vd->vdev_probe_zio == zio);
954 vd->vdev_probe_zio = NULL;
955 mutex_exit(&vd->vdev_probe_lock);
957 while ((pio = zio_walk_parents(zio)) != NULL)
958 if (!vdev_accessible(vd, pio))
959 pio->io_error = SET_ERROR(ENXIO);
961 kmem_free(vps, sizeof (*vps));
966 * Determine whether this device is accessible.
968 * Read and write to several known locations: the pad regions of each
969 * vdev label but the first, which we leave alone in case it contains
973 vdev_probe(vdev_t *vd, zio_t *zio)
975 spa_t *spa = vd->vdev_spa;
976 vdev_probe_stats_t *vps = NULL;
979 ASSERT(vd->vdev_ops->vdev_op_leaf);
982 * Don't probe the probe.
984 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
988 * To prevent 'probe storms' when a device fails, we create
989 * just one probe i/o at a time. All zios that want to probe
990 * this vdev will become parents of the probe io.
992 mutex_enter(&vd->vdev_probe_lock);
994 if ((pio = vd->vdev_probe_zio) == NULL) {
995 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
997 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
998 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1001 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1003 * vdev_cant_read and vdev_cant_write can only
1004 * transition from TRUE to FALSE when we have the
1005 * SCL_ZIO lock as writer; otherwise they can only
1006 * transition from FALSE to TRUE. This ensures that
1007 * any zio looking at these values can assume that
1008 * failures persist for the life of the I/O. That's
1009 * important because when a device has intermittent
1010 * connectivity problems, we want to ensure that
1011 * they're ascribed to the device (ENXIO) and not
1014 * Since we hold SCL_ZIO as writer here, clear both
1015 * values so the probe can reevaluate from first
1018 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1019 vd->vdev_cant_read = B_FALSE;
1020 vd->vdev_cant_write = B_FALSE;
1023 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1024 vdev_probe_done, vps,
1025 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1028 * We can't change the vdev state in this context, so we
1029 * kick off an async task to do it on our behalf.
1032 vd->vdev_probe_wanted = B_TRUE;
1033 spa_async_request(spa, SPA_ASYNC_PROBE);
1038 zio_add_child(zio, pio);
1040 mutex_exit(&vd->vdev_probe_lock);
1043 ASSERT(zio != NULL);
1047 for (int l = 1; l < VDEV_LABELS; l++) {
1048 zio_nowait(zio_read_phys(pio, vd,
1049 vdev_label_offset(vd->vdev_psize, l,
1050 offsetof(vdev_label_t, vl_pad2)),
1051 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1052 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1053 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1064 vdev_open_child(void *arg)
1068 vd->vdev_open_thread = curthread;
1069 vd->vdev_open_error = vdev_open(vd);
1070 vd->vdev_open_thread = NULL;
1074 vdev_uses_zvols(vdev_t *vd)
1076 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1077 strlen(ZVOL_DIR)) == 0)
1079 for (int c = 0; c < vd->vdev_children; c++)
1080 if (vdev_uses_zvols(vd->vdev_child[c]))
1086 vdev_open_children(vdev_t *vd)
1089 int children = vd->vdev_children;
1092 * in order to handle pools on top of zvols, do the opens
1093 * in a single thread so that the same thread holds the
1094 * spa_namespace_lock
1096 if (B_TRUE || vdev_uses_zvols(vd)) {
1097 for (int c = 0; c < children; c++)
1098 vd->vdev_child[c]->vdev_open_error =
1099 vdev_open(vd->vdev_child[c]);
1102 tq = taskq_create("vdev_open", children, minclsyspri,
1103 children, children, TASKQ_PREPOPULATE);
1105 for (int c = 0; c < children; c++)
1106 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1113 * Prepare a virtual device for access.
1116 vdev_open(vdev_t *vd)
1118 spa_t *spa = vd->vdev_spa;
1121 uint64_t max_osize = 0;
1122 uint64_t asize, max_asize, psize;
1123 uint64_t ashift = 0;
1125 ASSERT(vd->vdev_open_thread == curthread ||
1126 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1127 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1128 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1129 vd->vdev_state == VDEV_STATE_OFFLINE);
1131 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1132 vd->vdev_cant_read = B_FALSE;
1133 vd->vdev_cant_write = B_FALSE;
1134 vd->vdev_min_asize = vdev_get_min_asize(vd);
1137 * If this vdev is not removed, check its fault status. If it's
1138 * faulted, bail out of the open.
1140 if (!vd->vdev_removed && vd->vdev_faulted) {
1141 ASSERT(vd->vdev_children == 0);
1142 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1143 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1144 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1145 vd->vdev_label_aux);
1146 return (SET_ERROR(ENXIO));
1147 } else if (vd->vdev_offline) {
1148 ASSERT(vd->vdev_children == 0);
1149 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1150 return (SET_ERROR(ENXIO));
1153 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1156 * Reset the vdev_reopening flag so that we actually close
1157 * the vdev on error.
1159 vd->vdev_reopening = B_FALSE;
1160 if (zio_injection_enabled && error == 0)
1161 error = zio_handle_device_injection(vd, NULL, ENXIO);
1164 if (vd->vdev_removed &&
1165 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1166 vd->vdev_removed = B_FALSE;
1168 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1169 vd->vdev_stat.vs_aux);
1173 vd->vdev_removed = B_FALSE;
1176 * Recheck the faulted flag now that we have confirmed that
1177 * the vdev is accessible. If we're faulted, bail.
1179 if (vd->vdev_faulted) {
1180 ASSERT(vd->vdev_children == 0);
1181 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1182 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1183 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1184 vd->vdev_label_aux);
1185 return (SET_ERROR(ENXIO));
1188 if (vd->vdev_degraded) {
1189 ASSERT(vd->vdev_children == 0);
1190 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1191 VDEV_AUX_ERR_EXCEEDED);
1193 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1197 * For hole or missing vdevs we just return success.
1199 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1202 if (vd->vdev_ops->vdev_op_leaf) {
1203 vd->vdev_notrim = B_FALSE;
1204 trim_map_create(vd);
1207 for (int c = 0; c < vd->vdev_children; c++) {
1208 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1209 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1215 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1216 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1218 if (vd->vdev_children == 0) {
1219 if (osize < SPA_MINDEVSIZE) {
1220 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1221 VDEV_AUX_TOO_SMALL);
1222 return (SET_ERROR(EOVERFLOW));
1225 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1226 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1227 VDEV_LABEL_END_SIZE);
1229 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1230 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1231 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1232 VDEV_AUX_TOO_SMALL);
1233 return (SET_ERROR(EOVERFLOW));
1237 max_asize = max_osize;
1240 vd->vdev_psize = psize;
1243 * Make sure the allocatable size hasn't shrunk.
1245 if (asize < vd->vdev_min_asize) {
1246 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1247 VDEV_AUX_BAD_LABEL);
1248 return (SET_ERROR(EINVAL));
1251 if (vd->vdev_asize == 0) {
1253 * This is the first-ever open, so use the computed values.
1254 * For testing purposes, a higher ashift can be requested.
1256 vd->vdev_asize = asize;
1257 vd->vdev_max_asize = max_asize;
1258 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1261 * Make sure the alignment requirement hasn't increased.
1263 if (ashift > vd->vdev_top->vdev_ashift) {
1264 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1265 VDEV_AUX_BAD_LABEL);
1268 vd->vdev_max_asize = max_asize;
1272 * If all children are healthy and the asize has increased,
1273 * then we've experienced dynamic LUN growth. If automatic
1274 * expansion is enabled then use the additional space.
1276 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1277 (vd->vdev_expanding || spa->spa_autoexpand))
1278 vd->vdev_asize = asize;
1280 vdev_set_min_asize(vd);
1283 * Ensure we can issue some IO before declaring the
1284 * vdev open for business.
1286 if (vd->vdev_ops->vdev_op_leaf &&
1287 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1288 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1289 VDEV_AUX_ERR_EXCEEDED);
1294 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1295 * resilver. But don't do this if we are doing a reopen for a scrub,
1296 * since this would just restart the scrub we are already doing.
1298 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1299 vdev_resilver_needed(vd, NULL, NULL))
1300 spa_async_request(spa, SPA_ASYNC_RESILVER);
1306 * Called once the vdevs are all opened, this routine validates the label
1307 * contents. This needs to be done before vdev_load() so that we don't
1308 * inadvertently do repair I/Os to the wrong device.
1310 * If 'strict' is false ignore the spa guid check. This is necessary because
1311 * if the machine crashed during a re-guid the new guid might have been written
1312 * to all of the vdev labels, but not the cached config. The strict check
1313 * will be performed when the pool is opened again using the mos config.
1315 * This function will only return failure if one of the vdevs indicates that it
1316 * has since been destroyed or exported. This is only possible if
1317 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1318 * will be updated but the function will return 0.
1321 vdev_validate(vdev_t *vd, boolean_t strict)
1323 spa_t *spa = vd->vdev_spa;
1325 uint64_t guid = 0, top_guid;
1328 for (int c = 0; c < vd->vdev_children; c++)
1329 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1330 return (SET_ERROR(EBADF));
1333 * If the device has already failed, or was marked offline, don't do
1334 * any further validation. Otherwise, label I/O will fail and we will
1335 * overwrite the previous state.
1337 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1338 uint64_t aux_guid = 0;
1340 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1341 spa_last_synced_txg(spa) : -1ULL;
1343 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1344 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1345 VDEV_AUX_BAD_LABEL);
1350 * Determine if this vdev has been split off into another
1351 * pool. If so, then refuse to open it.
1353 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1354 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1355 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1356 VDEV_AUX_SPLIT_POOL);
1361 if (strict && (nvlist_lookup_uint64(label,
1362 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1363 guid != spa_guid(spa))) {
1364 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1365 VDEV_AUX_CORRUPT_DATA);
1370 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1371 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1376 * If this vdev just became a top-level vdev because its
1377 * sibling was detached, it will have adopted the parent's
1378 * vdev guid -- but the label may or may not be on disk yet.
1379 * Fortunately, either version of the label will have the
1380 * same top guid, so if we're a top-level vdev, we can
1381 * safely compare to that instead.
1383 * If we split this vdev off instead, then we also check the
1384 * original pool's guid. We don't want to consider the vdev
1385 * corrupt if it is partway through a split operation.
1387 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1389 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1391 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1392 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1393 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1394 VDEV_AUX_CORRUPT_DATA);
1399 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1401 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1402 VDEV_AUX_CORRUPT_DATA);
1410 * If this is a verbatim import, no need to check the
1411 * state of the pool.
1413 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1414 spa_load_state(spa) == SPA_LOAD_OPEN &&
1415 state != POOL_STATE_ACTIVE)
1416 return (SET_ERROR(EBADF));
1419 * If we were able to open and validate a vdev that was
1420 * previously marked permanently unavailable, clear that state
1423 if (vd->vdev_not_present)
1424 vd->vdev_not_present = 0;
1431 * Close a virtual device.
1434 vdev_close(vdev_t *vd)
1436 spa_t *spa = vd->vdev_spa;
1437 vdev_t *pvd = vd->vdev_parent;
1439 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1442 * If our parent is reopening, then we are as well, unless we are
1445 if (pvd != NULL && pvd->vdev_reopening)
1446 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1448 vd->vdev_ops->vdev_op_close(vd);
1450 vdev_cache_purge(vd);
1452 if (vd->vdev_ops->vdev_op_leaf)
1453 trim_map_destroy(vd);
1456 * We record the previous state before we close it, so that if we are
1457 * doing a reopen(), we don't generate FMA ereports if we notice that
1458 * it's still faulted.
1460 vd->vdev_prevstate = vd->vdev_state;
1462 if (vd->vdev_offline)
1463 vd->vdev_state = VDEV_STATE_OFFLINE;
1465 vd->vdev_state = VDEV_STATE_CLOSED;
1466 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1470 vdev_hold(vdev_t *vd)
1472 spa_t *spa = vd->vdev_spa;
1474 ASSERT(spa_is_root(spa));
1475 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1478 for (int c = 0; c < vd->vdev_children; c++)
1479 vdev_hold(vd->vdev_child[c]);
1481 if (vd->vdev_ops->vdev_op_leaf)
1482 vd->vdev_ops->vdev_op_hold(vd);
1486 vdev_rele(vdev_t *vd)
1488 spa_t *spa = vd->vdev_spa;
1490 ASSERT(spa_is_root(spa));
1491 for (int c = 0; c < vd->vdev_children; c++)
1492 vdev_rele(vd->vdev_child[c]);
1494 if (vd->vdev_ops->vdev_op_leaf)
1495 vd->vdev_ops->vdev_op_rele(vd);
1499 * Reopen all interior vdevs and any unopened leaves. We don't actually
1500 * reopen leaf vdevs which had previously been opened as they might deadlock
1501 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1502 * If the leaf has never been opened then open it, as usual.
1505 vdev_reopen(vdev_t *vd)
1507 spa_t *spa = vd->vdev_spa;
1509 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1511 /* set the reopening flag unless we're taking the vdev offline */
1512 vd->vdev_reopening = !vd->vdev_offline;
1514 (void) vdev_open(vd);
1517 * Call vdev_validate() here to make sure we have the same device.
1518 * Otherwise, a device with an invalid label could be successfully
1519 * opened in response to vdev_reopen().
1522 (void) vdev_validate_aux(vd);
1523 if (vdev_readable(vd) && vdev_writeable(vd) &&
1524 vd->vdev_aux == &spa->spa_l2cache &&
1525 !l2arc_vdev_present(vd))
1526 l2arc_add_vdev(spa, vd);
1528 (void) vdev_validate(vd, B_TRUE);
1532 * Reassess parent vdev's health.
1534 vdev_propagate_state(vd);
1538 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1543 * Normally, partial opens (e.g. of a mirror) are allowed.
1544 * For a create, however, we want to fail the request if
1545 * there are any components we can't open.
1547 error = vdev_open(vd);
1549 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1551 return (error ? error : ENXIO);
1555 * Recursively initialize all labels.
1557 if ((error = vdev_label_init(vd, txg, isreplacing ?
1558 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1567 vdev_metaslab_set_size(vdev_t *vd)
1570 * Aim for roughly 200 metaslabs per vdev.
1572 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1573 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1577 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1579 ASSERT(vd == vd->vdev_top);
1580 ASSERT(!vd->vdev_ishole);
1581 ASSERT(ISP2(flags));
1582 ASSERT(spa_writeable(vd->vdev_spa));
1584 if (flags & VDD_METASLAB)
1585 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1587 if (flags & VDD_DTL)
1588 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1590 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1596 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1597 * the vdev has less than perfect replication. There are four kinds of DTL:
1599 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1601 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1603 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1604 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1605 * txgs that was scrubbed.
1607 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1608 * persistent errors or just some device being offline.
1609 * Unlike the other three, the DTL_OUTAGE map is not generally
1610 * maintained; it's only computed when needed, typically to
1611 * determine whether a device can be detached.
1613 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1614 * either has the data or it doesn't.
1616 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1617 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1618 * if any child is less than fully replicated, then so is its parent.
1619 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1620 * comprising only those txgs which appear in 'maxfaults' or more children;
1621 * those are the txgs we don't have enough replication to read. For example,
1622 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1623 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1624 * two child DTL_MISSING maps.
1626 * It should be clear from the above that to compute the DTLs and outage maps
1627 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1628 * Therefore, that is all we keep on disk. When loading the pool, or after
1629 * a configuration change, we generate all other DTLs from first principles.
1632 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1634 space_map_t *sm = &vd->vdev_dtl[t];
1636 ASSERT(t < DTL_TYPES);
1637 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1638 ASSERT(spa_writeable(vd->vdev_spa));
1640 mutex_enter(sm->sm_lock);
1641 if (!space_map_contains(sm, txg, size))
1642 space_map_add(sm, txg, size);
1643 mutex_exit(sm->sm_lock);
1647 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1649 space_map_t *sm = &vd->vdev_dtl[t];
1650 boolean_t dirty = B_FALSE;
1652 ASSERT(t < DTL_TYPES);
1653 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1655 mutex_enter(sm->sm_lock);
1656 if (sm->sm_space != 0)
1657 dirty = space_map_contains(sm, txg, size);
1658 mutex_exit(sm->sm_lock);
1664 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1666 space_map_t *sm = &vd->vdev_dtl[t];
1669 mutex_enter(sm->sm_lock);
1670 empty = (sm->sm_space == 0);
1671 mutex_exit(sm->sm_lock);
1677 * Reassess DTLs after a config change or scrub completion.
1680 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1682 spa_t *spa = vd->vdev_spa;
1686 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1688 for (int c = 0; c < vd->vdev_children; c++)
1689 vdev_dtl_reassess(vd->vdev_child[c], txg,
1690 scrub_txg, scrub_done);
1692 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1695 if (vd->vdev_ops->vdev_op_leaf) {
1696 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1698 mutex_enter(&vd->vdev_dtl_lock);
1699 if (scrub_txg != 0 &&
1700 (spa->spa_scrub_started ||
1701 (scn && scn->scn_phys.scn_errors == 0))) {
1703 * We completed a scrub up to scrub_txg. If we
1704 * did it without rebooting, then the scrub dtl
1705 * will be valid, so excise the old region and
1706 * fold in the scrub dtl. Otherwise, leave the
1707 * dtl as-is if there was an error.
1709 * There's little trick here: to excise the beginning
1710 * of the DTL_MISSING map, we put it into a reference
1711 * tree and then add a segment with refcnt -1 that
1712 * covers the range [0, scrub_txg). This means
1713 * that each txg in that range has refcnt -1 or 0.
1714 * We then add DTL_SCRUB with a refcnt of 2, so that
1715 * entries in the range [0, scrub_txg) will have a
1716 * positive refcnt -- either 1 or 2. We then convert
1717 * the reference tree into the new DTL_MISSING map.
1719 space_map_ref_create(&reftree);
1720 space_map_ref_add_map(&reftree,
1721 &vd->vdev_dtl[DTL_MISSING], 1);
1722 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1723 space_map_ref_add_map(&reftree,
1724 &vd->vdev_dtl[DTL_SCRUB], 2);
1725 space_map_ref_generate_map(&reftree,
1726 &vd->vdev_dtl[DTL_MISSING], 1);
1727 space_map_ref_destroy(&reftree);
1729 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1730 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1731 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1733 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1734 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1735 if (!vdev_readable(vd))
1736 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1738 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1739 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1740 mutex_exit(&vd->vdev_dtl_lock);
1743 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1747 mutex_enter(&vd->vdev_dtl_lock);
1748 for (int t = 0; t < DTL_TYPES; t++) {
1749 /* account for child's outage in parent's missing map */
1750 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1752 continue; /* leaf vdevs only */
1753 if (t == DTL_PARTIAL)
1754 minref = 1; /* i.e. non-zero */
1755 else if (vd->vdev_nparity != 0)
1756 minref = vd->vdev_nparity + 1; /* RAID-Z */
1758 minref = vd->vdev_children; /* any kind of mirror */
1759 space_map_ref_create(&reftree);
1760 for (int c = 0; c < vd->vdev_children; c++) {
1761 vdev_t *cvd = vd->vdev_child[c];
1762 mutex_enter(&cvd->vdev_dtl_lock);
1763 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1764 mutex_exit(&cvd->vdev_dtl_lock);
1766 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1767 space_map_ref_destroy(&reftree);
1769 mutex_exit(&vd->vdev_dtl_lock);
1773 vdev_dtl_load(vdev_t *vd)
1775 spa_t *spa = vd->vdev_spa;
1776 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1777 objset_t *mos = spa->spa_meta_objset;
1781 ASSERT(vd->vdev_children == 0);
1783 if (smo->smo_object == 0)
1786 ASSERT(!vd->vdev_ishole);
1788 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1791 ASSERT3U(db->db_size, >=, sizeof (*smo));
1792 bcopy(db->db_data, smo, sizeof (*smo));
1793 dmu_buf_rele(db, FTAG);
1795 mutex_enter(&vd->vdev_dtl_lock);
1796 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1797 NULL, SM_ALLOC, smo, mos);
1798 mutex_exit(&vd->vdev_dtl_lock);
1804 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1806 spa_t *spa = vd->vdev_spa;
1807 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1808 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1809 objset_t *mos = spa->spa_meta_objset;
1815 ASSERT(!vd->vdev_ishole);
1817 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1819 if (vd->vdev_detached) {
1820 if (smo->smo_object != 0) {
1821 int err = dmu_object_free(mos, smo->smo_object, tx);
1823 smo->smo_object = 0;
1829 if (smo->smo_object == 0) {
1830 ASSERT(smo->smo_objsize == 0);
1831 ASSERT(smo->smo_alloc == 0);
1832 smo->smo_object = dmu_object_alloc(mos,
1833 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1834 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1835 ASSERT(smo->smo_object != 0);
1836 vdev_config_dirty(vd->vdev_top);
1839 bzero(&smlock, sizeof (smlock));
1840 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1842 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1845 mutex_enter(&smlock);
1847 mutex_enter(&vd->vdev_dtl_lock);
1848 space_map_walk(sm, space_map_add, &smsync);
1849 mutex_exit(&vd->vdev_dtl_lock);
1851 space_map_truncate(smo, mos, tx);
1852 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1853 space_map_vacate(&smsync, NULL, NULL);
1855 space_map_destroy(&smsync);
1857 mutex_exit(&smlock);
1858 mutex_destroy(&smlock);
1860 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1861 dmu_buf_will_dirty(db, tx);
1862 ASSERT3U(db->db_size, >=, sizeof (*smo));
1863 bcopy(smo, db->db_data, sizeof (*smo));
1864 dmu_buf_rele(db, FTAG);
1870 * Determine whether the specified vdev can be offlined/detached/removed
1871 * without losing data.
1874 vdev_dtl_required(vdev_t *vd)
1876 spa_t *spa = vd->vdev_spa;
1877 vdev_t *tvd = vd->vdev_top;
1878 uint8_t cant_read = vd->vdev_cant_read;
1881 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1883 if (vd == spa->spa_root_vdev || vd == tvd)
1887 * Temporarily mark the device as unreadable, and then determine
1888 * whether this results in any DTL outages in the top-level vdev.
1889 * If not, we can safely offline/detach/remove the device.
1891 vd->vdev_cant_read = B_TRUE;
1892 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1893 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1894 vd->vdev_cant_read = cant_read;
1895 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1897 if (!required && zio_injection_enabled)
1898 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1904 * Determine if resilver is needed, and if so the txg range.
1907 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1909 boolean_t needed = B_FALSE;
1910 uint64_t thismin = UINT64_MAX;
1911 uint64_t thismax = 0;
1913 if (vd->vdev_children == 0) {
1914 mutex_enter(&vd->vdev_dtl_lock);
1915 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1916 vdev_writeable(vd)) {
1919 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1920 thismin = ss->ss_start - 1;
1921 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1922 thismax = ss->ss_end;
1925 mutex_exit(&vd->vdev_dtl_lock);
1927 for (int c = 0; c < vd->vdev_children; c++) {
1928 vdev_t *cvd = vd->vdev_child[c];
1929 uint64_t cmin, cmax;
1931 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1932 thismin = MIN(thismin, cmin);
1933 thismax = MAX(thismax, cmax);
1939 if (needed && minp) {
1947 vdev_load(vdev_t *vd)
1950 * Recursively load all children.
1952 for (int c = 0; c < vd->vdev_children; c++)
1953 vdev_load(vd->vdev_child[c]);
1956 * If this is a top-level vdev, initialize its metaslabs.
1958 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1959 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1960 vdev_metaslab_init(vd, 0) != 0))
1961 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1962 VDEV_AUX_CORRUPT_DATA);
1965 * If this is a leaf vdev, load its DTL.
1967 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1968 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1969 VDEV_AUX_CORRUPT_DATA);
1973 * The special vdev case is used for hot spares and l2cache devices. Its
1974 * sole purpose it to set the vdev state for the associated vdev. To do this,
1975 * we make sure that we can open the underlying device, then try to read the
1976 * label, and make sure that the label is sane and that it hasn't been
1977 * repurposed to another pool.
1980 vdev_validate_aux(vdev_t *vd)
1983 uint64_t guid, version;
1986 if (!vdev_readable(vd))
1989 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
1990 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1991 VDEV_AUX_CORRUPT_DATA);
1995 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1996 !SPA_VERSION_IS_SUPPORTED(version) ||
1997 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1998 guid != vd->vdev_guid ||
1999 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2000 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2001 VDEV_AUX_CORRUPT_DATA);
2007 * We don't actually check the pool state here. If it's in fact in
2008 * use by another pool, we update this fact on the fly when requested.
2015 vdev_remove(vdev_t *vd, uint64_t txg)
2017 spa_t *spa = vd->vdev_spa;
2018 objset_t *mos = spa->spa_meta_objset;
2021 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2023 if (vd->vdev_dtl_smo.smo_object) {
2024 ASSERT0(vd->vdev_dtl_smo.smo_alloc);
2025 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2026 vd->vdev_dtl_smo.smo_object = 0;
2029 if (vd->vdev_ms != NULL) {
2030 for (int m = 0; m < vd->vdev_ms_count; m++) {
2031 metaslab_t *msp = vd->vdev_ms[m];
2033 if (msp == NULL || msp->ms_smo.smo_object == 0)
2036 ASSERT0(msp->ms_smo.smo_alloc);
2037 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2038 msp->ms_smo.smo_object = 0;
2042 if (vd->vdev_ms_array) {
2043 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2044 vd->vdev_ms_array = 0;
2045 vd->vdev_ms_shift = 0;
2051 vdev_sync_done(vdev_t *vd, uint64_t txg)
2054 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2056 ASSERT(!vd->vdev_ishole);
2058 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2059 metaslab_sync_done(msp, txg);
2062 metaslab_sync_reassess(vd->vdev_mg);
2066 vdev_sync(vdev_t *vd, uint64_t txg)
2068 spa_t *spa = vd->vdev_spa;
2073 ASSERT(!vd->vdev_ishole);
2075 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2076 ASSERT(vd == vd->vdev_top);
2077 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2078 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2079 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2080 ASSERT(vd->vdev_ms_array != 0);
2081 vdev_config_dirty(vd);
2086 * Remove the metadata associated with this vdev once it's empty.
2088 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2089 vdev_remove(vd, txg);
2091 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2092 metaslab_sync(msp, txg);
2093 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2096 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2097 vdev_dtl_sync(lvd, txg);
2099 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2103 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2105 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2109 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2110 * not be opened, and no I/O is attempted.
2113 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2117 spa_vdev_state_enter(spa, SCL_NONE);
2119 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2120 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2122 if (!vd->vdev_ops->vdev_op_leaf)
2123 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2128 * We don't directly use the aux state here, but if we do a
2129 * vdev_reopen(), we need this value to be present to remember why we
2132 vd->vdev_label_aux = aux;
2135 * Faulted state takes precedence over degraded.
2137 vd->vdev_delayed_close = B_FALSE;
2138 vd->vdev_faulted = 1ULL;
2139 vd->vdev_degraded = 0ULL;
2140 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2143 * If this device has the only valid copy of the data, then
2144 * back off and simply mark the vdev as degraded instead.
2146 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2147 vd->vdev_degraded = 1ULL;
2148 vd->vdev_faulted = 0ULL;
2151 * If we reopen the device and it's not dead, only then do we
2156 if (vdev_readable(vd))
2157 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2160 return (spa_vdev_state_exit(spa, vd, 0));
2164 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2165 * user that something is wrong. The vdev continues to operate as normal as far
2166 * as I/O is concerned.
2169 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2173 spa_vdev_state_enter(spa, SCL_NONE);
2175 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2176 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2178 if (!vd->vdev_ops->vdev_op_leaf)
2179 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2182 * If the vdev is already faulted, then don't do anything.
2184 if (vd->vdev_faulted || vd->vdev_degraded)
2185 return (spa_vdev_state_exit(spa, NULL, 0));
2187 vd->vdev_degraded = 1ULL;
2188 if (!vdev_is_dead(vd))
2189 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2192 return (spa_vdev_state_exit(spa, vd, 0));
2196 * Online the given vdev.
2198 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2199 * spare device should be detached when the device finishes resilvering.
2200 * Second, the online should be treated like a 'test' online case, so no FMA
2201 * events are generated if the device fails to open.
2204 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2206 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2208 spa_vdev_state_enter(spa, SCL_NONE);
2210 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2211 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2213 if (!vd->vdev_ops->vdev_op_leaf)
2214 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2217 vd->vdev_offline = B_FALSE;
2218 vd->vdev_tmpoffline = B_FALSE;
2219 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2220 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2222 /* XXX - L2ARC 1.0 does not support expansion */
2223 if (!vd->vdev_aux) {
2224 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2225 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2229 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2231 if (!vd->vdev_aux) {
2232 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2233 pvd->vdev_expanding = B_FALSE;
2237 *newstate = vd->vdev_state;
2238 if ((flags & ZFS_ONLINE_UNSPARE) &&
2239 !vdev_is_dead(vd) && vd->vdev_parent &&
2240 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2241 vd->vdev_parent->vdev_child[0] == vd)
2242 vd->vdev_unspare = B_TRUE;
2244 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2246 /* XXX - L2ARC 1.0 does not support expansion */
2248 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2249 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2251 return (spa_vdev_state_exit(spa, vd, 0));
2255 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2259 uint64_t generation;
2260 metaslab_group_t *mg;
2263 spa_vdev_state_enter(spa, SCL_ALLOC);
2265 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2266 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2268 if (!vd->vdev_ops->vdev_op_leaf)
2269 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2273 generation = spa->spa_config_generation + 1;
2276 * If the device isn't already offline, try to offline it.
2278 if (!vd->vdev_offline) {
2280 * If this device has the only valid copy of some data,
2281 * don't allow it to be offlined. Log devices are always
2284 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2285 vdev_dtl_required(vd))
2286 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2289 * If the top-level is a slog and it has had allocations
2290 * then proceed. We check that the vdev's metaslab group
2291 * is not NULL since it's possible that we may have just
2292 * added this vdev but not yet initialized its metaslabs.
2294 if (tvd->vdev_islog && mg != NULL) {
2296 * Prevent any future allocations.
2298 metaslab_group_passivate(mg);
2299 (void) spa_vdev_state_exit(spa, vd, 0);
2301 error = spa_offline_log(spa);
2303 spa_vdev_state_enter(spa, SCL_ALLOC);
2306 * Check to see if the config has changed.
2308 if (error || generation != spa->spa_config_generation) {
2309 metaslab_group_activate(mg);
2311 return (spa_vdev_state_exit(spa,
2313 (void) spa_vdev_state_exit(spa, vd, 0);
2316 ASSERT0(tvd->vdev_stat.vs_alloc);
2320 * Offline this device and reopen its top-level vdev.
2321 * If the top-level vdev is a log device then just offline
2322 * it. Otherwise, if this action results in the top-level
2323 * vdev becoming unusable, undo it and fail the request.
2325 vd->vdev_offline = B_TRUE;
2328 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2329 vdev_is_dead(tvd)) {
2330 vd->vdev_offline = B_FALSE;
2332 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2336 * Add the device back into the metaslab rotor so that
2337 * once we online the device it's open for business.
2339 if (tvd->vdev_islog && mg != NULL)
2340 metaslab_group_activate(mg);
2343 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2345 return (spa_vdev_state_exit(spa, vd, 0));
2349 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2353 mutex_enter(&spa->spa_vdev_top_lock);
2354 error = vdev_offline_locked(spa, guid, flags);
2355 mutex_exit(&spa->spa_vdev_top_lock);
2361 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2362 * vdev_offline(), we assume the spa config is locked. We also clear all
2363 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2366 vdev_clear(spa_t *spa, vdev_t *vd)
2368 vdev_t *rvd = spa->spa_root_vdev;
2370 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2375 vd->vdev_stat.vs_read_errors = 0;
2376 vd->vdev_stat.vs_write_errors = 0;
2377 vd->vdev_stat.vs_checksum_errors = 0;
2379 for (int c = 0; c < vd->vdev_children; c++)
2380 vdev_clear(spa, vd->vdev_child[c]);
2383 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2384 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2386 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2387 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2391 * If we're in the FAULTED state or have experienced failed I/O, then
2392 * clear the persistent state and attempt to reopen the device. We
2393 * also mark the vdev config dirty, so that the new faulted state is
2394 * written out to disk.
2396 if (vd->vdev_faulted || vd->vdev_degraded ||
2397 !vdev_readable(vd) || !vdev_writeable(vd)) {
2400 * When reopening in reponse to a clear event, it may be due to
2401 * a fmadm repair request. In this case, if the device is
2402 * still broken, we want to still post the ereport again.
2404 vd->vdev_forcefault = B_TRUE;
2406 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2407 vd->vdev_cant_read = B_FALSE;
2408 vd->vdev_cant_write = B_FALSE;
2410 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2412 vd->vdev_forcefault = B_FALSE;
2414 if (vd != rvd && vdev_writeable(vd->vdev_top))
2415 vdev_state_dirty(vd->vdev_top);
2417 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2418 spa_async_request(spa, SPA_ASYNC_RESILVER);
2420 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2424 * When clearing a FMA-diagnosed fault, we always want to
2425 * unspare the device, as we assume that the original spare was
2426 * done in response to the FMA fault.
2428 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2429 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2430 vd->vdev_parent->vdev_child[0] == vd)
2431 vd->vdev_unspare = B_TRUE;
2435 vdev_is_dead(vdev_t *vd)
2438 * Holes and missing devices are always considered "dead".
2439 * This simplifies the code since we don't have to check for
2440 * these types of devices in the various code paths.
2441 * Instead we rely on the fact that we skip over dead devices
2442 * before issuing I/O to them.
2444 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2445 vd->vdev_ops == &vdev_missing_ops);
2449 vdev_readable(vdev_t *vd)
2451 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2455 vdev_writeable(vdev_t *vd)
2457 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2461 vdev_allocatable(vdev_t *vd)
2463 uint64_t state = vd->vdev_state;
2466 * We currently allow allocations from vdevs which may be in the
2467 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2468 * fails to reopen then we'll catch it later when we're holding
2469 * the proper locks. Note that we have to get the vdev state
2470 * in a local variable because although it changes atomically,
2471 * we're asking two separate questions about it.
2473 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2474 !vd->vdev_cant_write && !vd->vdev_ishole);
2478 vdev_accessible(vdev_t *vd, zio_t *zio)
2480 ASSERT(zio->io_vd == vd);
2482 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2485 if (zio->io_type == ZIO_TYPE_READ)
2486 return (!vd->vdev_cant_read);
2488 if (zio->io_type == ZIO_TYPE_WRITE)
2489 return (!vd->vdev_cant_write);
2495 * Get statistics for the given vdev.
2498 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2500 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2502 mutex_enter(&vd->vdev_stat_lock);
2503 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2504 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2505 vs->vs_state = vd->vdev_state;
2506 vs->vs_rsize = vdev_get_min_asize(vd);
2507 if (vd->vdev_ops->vdev_op_leaf)
2508 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2509 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2510 mutex_exit(&vd->vdev_stat_lock);
2513 * If we're getting stats on the root vdev, aggregate the I/O counts
2514 * over all top-level vdevs (i.e. the direct children of the root).
2517 for (int c = 0; c < rvd->vdev_children; c++) {
2518 vdev_t *cvd = rvd->vdev_child[c];
2519 vdev_stat_t *cvs = &cvd->vdev_stat;
2521 mutex_enter(&vd->vdev_stat_lock);
2522 for (int t = 0; t < ZIO_TYPES; t++) {
2523 vs->vs_ops[t] += cvs->vs_ops[t];
2524 vs->vs_bytes[t] += cvs->vs_bytes[t];
2526 cvs->vs_scan_removing = cvd->vdev_removing;
2527 mutex_exit(&vd->vdev_stat_lock);
2533 vdev_clear_stats(vdev_t *vd)
2535 mutex_enter(&vd->vdev_stat_lock);
2536 vd->vdev_stat.vs_space = 0;
2537 vd->vdev_stat.vs_dspace = 0;
2538 vd->vdev_stat.vs_alloc = 0;
2539 mutex_exit(&vd->vdev_stat_lock);
2543 vdev_scan_stat_init(vdev_t *vd)
2545 vdev_stat_t *vs = &vd->vdev_stat;
2547 for (int c = 0; c < vd->vdev_children; c++)
2548 vdev_scan_stat_init(vd->vdev_child[c]);
2550 mutex_enter(&vd->vdev_stat_lock);
2551 vs->vs_scan_processed = 0;
2552 mutex_exit(&vd->vdev_stat_lock);
2556 vdev_stat_update(zio_t *zio, uint64_t psize)
2558 spa_t *spa = zio->io_spa;
2559 vdev_t *rvd = spa->spa_root_vdev;
2560 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2562 uint64_t txg = zio->io_txg;
2563 vdev_stat_t *vs = &vd->vdev_stat;
2564 zio_type_t type = zio->io_type;
2565 int flags = zio->io_flags;
2568 * If this i/o is a gang leader, it didn't do any actual work.
2570 if (zio->io_gang_tree)
2573 if (zio->io_error == 0) {
2575 * If this is a root i/o, don't count it -- we've already
2576 * counted the top-level vdevs, and vdev_get_stats() will
2577 * aggregate them when asked. This reduces contention on
2578 * the root vdev_stat_lock and implicitly handles blocks
2579 * that compress away to holes, for which there is no i/o.
2580 * (Holes never create vdev children, so all the counters
2581 * remain zero, which is what we want.)
2583 * Note: this only applies to successful i/o (io_error == 0)
2584 * because unlike i/o counts, errors are not additive.
2585 * When reading a ditto block, for example, failure of
2586 * one top-level vdev does not imply a root-level error.
2591 ASSERT(vd == zio->io_vd);
2593 if (flags & ZIO_FLAG_IO_BYPASS)
2596 mutex_enter(&vd->vdev_stat_lock);
2598 if (flags & ZIO_FLAG_IO_REPAIR) {
2599 if (flags & ZIO_FLAG_SCAN_THREAD) {
2600 dsl_scan_phys_t *scn_phys =
2601 &spa->spa_dsl_pool->dp_scan->scn_phys;
2602 uint64_t *processed = &scn_phys->scn_processed;
2605 if (vd->vdev_ops->vdev_op_leaf)
2606 atomic_add_64(processed, psize);
2607 vs->vs_scan_processed += psize;
2610 if (flags & ZIO_FLAG_SELF_HEAL)
2611 vs->vs_self_healed += psize;
2615 vs->vs_bytes[type] += psize;
2617 mutex_exit(&vd->vdev_stat_lock);
2621 if (flags & ZIO_FLAG_SPECULATIVE)
2625 * If this is an I/O error that is going to be retried, then ignore the
2626 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2627 * hard errors, when in reality they can happen for any number of
2628 * innocuous reasons (bus resets, MPxIO link failure, etc).
2630 if (zio->io_error == EIO &&
2631 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2635 * Intent logs writes won't propagate their error to the root
2636 * I/O so don't mark these types of failures as pool-level
2639 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2642 mutex_enter(&vd->vdev_stat_lock);
2643 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2644 if (zio->io_error == ECKSUM)
2645 vs->vs_checksum_errors++;
2647 vs->vs_read_errors++;
2649 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2650 vs->vs_write_errors++;
2651 mutex_exit(&vd->vdev_stat_lock);
2653 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2654 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2655 (flags & ZIO_FLAG_SCAN_THREAD) ||
2656 spa->spa_claiming)) {
2658 * This is either a normal write (not a repair), or it's
2659 * a repair induced by the scrub thread, or it's a repair
2660 * made by zil_claim() during spa_load() in the first txg.
2661 * In the normal case, we commit the DTL change in the same
2662 * txg as the block was born. In the scrub-induced repair
2663 * case, we know that scrubs run in first-pass syncing context,
2664 * so we commit the DTL change in spa_syncing_txg(spa).
2665 * In the zil_claim() case, we commit in spa_first_txg(spa).
2667 * We currently do not make DTL entries for failed spontaneous
2668 * self-healing writes triggered by normal (non-scrubbing)
2669 * reads, because we have no transactional context in which to
2670 * do so -- and it's not clear that it'd be desirable anyway.
2672 if (vd->vdev_ops->vdev_op_leaf) {
2673 uint64_t commit_txg = txg;
2674 if (flags & ZIO_FLAG_SCAN_THREAD) {
2675 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2676 ASSERT(spa_sync_pass(spa) == 1);
2677 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2678 commit_txg = spa_syncing_txg(spa);
2679 } else if (spa->spa_claiming) {
2680 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2681 commit_txg = spa_first_txg(spa);
2683 ASSERT(commit_txg >= spa_syncing_txg(spa));
2684 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2686 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2687 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2688 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2691 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2696 * Update the in-core space usage stats for this vdev, its metaslab class,
2697 * and the root vdev.
2700 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2701 int64_t space_delta)
2703 int64_t dspace_delta = space_delta;
2704 spa_t *spa = vd->vdev_spa;
2705 vdev_t *rvd = spa->spa_root_vdev;
2706 metaslab_group_t *mg = vd->vdev_mg;
2707 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2709 ASSERT(vd == vd->vdev_top);
2712 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2713 * factor. We must calculate this here and not at the root vdev
2714 * because the root vdev's psize-to-asize is simply the max of its
2715 * childrens', thus not accurate enough for us.
2717 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2718 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2719 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2720 vd->vdev_deflate_ratio;
2722 mutex_enter(&vd->vdev_stat_lock);
2723 vd->vdev_stat.vs_alloc += alloc_delta;
2724 vd->vdev_stat.vs_space += space_delta;
2725 vd->vdev_stat.vs_dspace += dspace_delta;
2726 mutex_exit(&vd->vdev_stat_lock);
2728 if (mc == spa_normal_class(spa)) {
2729 mutex_enter(&rvd->vdev_stat_lock);
2730 rvd->vdev_stat.vs_alloc += alloc_delta;
2731 rvd->vdev_stat.vs_space += space_delta;
2732 rvd->vdev_stat.vs_dspace += dspace_delta;
2733 mutex_exit(&rvd->vdev_stat_lock);
2737 ASSERT(rvd == vd->vdev_parent);
2738 ASSERT(vd->vdev_ms_count != 0);
2740 metaslab_class_space_update(mc,
2741 alloc_delta, defer_delta, space_delta, dspace_delta);
2746 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2747 * so that it will be written out next time the vdev configuration is synced.
2748 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2751 vdev_config_dirty(vdev_t *vd)
2753 spa_t *spa = vd->vdev_spa;
2754 vdev_t *rvd = spa->spa_root_vdev;
2757 ASSERT(spa_writeable(spa));
2760 * If this is an aux vdev (as with l2cache and spare devices), then we
2761 * update the vdev config manually and set the sync flag.
2763 if (vd->vdev_aux != NULL) {
2764 spa_aux_vdev_t *sav = vd->vdev_aux;
2768 for (c = 0; c < sav->sav_count; c++) {
2769 if (sav->sav_vdevs[c] == vd)
2773 if (c == sav->sav_count) {
2775 * We're being removed. There's nothing more to do.
2777 ASSERT(sav->sav_sync == B_TRUE);
2781 sav->sav_sync = B_TRUE;
2783 if (nvlist_lookup_nvlist_array(sav->sav_config,
2784 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2785 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2786 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2792 * Setting the nvlist in the middle if the array is a little
2793 * sketchy, but it will work.
2795 nvlist_free(aux[c]);
2796 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2802 * The dirty list is protected by the SCL_CONFIG lock. The caller
2803 * must either hold SCL_CONFIG as writer, or must be the sync thread
2804 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2805 * so this is sufficient to ensure mutual exclusion.
2807 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2808 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2809 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2812 for (c = 0; c < rvd->vdev_children; c++)
2813 vdev_config_dirty(rvd->vdev_child[c]);
2815 ASSERT(vd == vd->vdev_top);
2817 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2819 list_insert_head(&spa->spa_config_dirty_list, vd);
2824 vdev_config_clean(vdev_t *vd)
2826 spa_t *spa = vd->vdev_spa;
2828 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2829 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2830 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2832 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2833 list_remove(&spa->spa_config_dirty_list, vd);
2837 * Mark a top-level vdev's state as dirty, so that the next pass of
2838 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2839 * the state changes from larger config changes because they require
2840 * much less locking, and are often needed for administrative actions.
2843 vdev_state_dirty(vdev_t *vd)
2845 spa_t *spa = vd->vdev_spa;
2847 ASSERT(spa_writeable(spa));
2848 ASSERT(vd == vd->vdev_top);
2851 * The state list is protected by the SCL_STATE lock. The caller
2852 * must either hold SCL_STATE as writer, or must be the sync thread
2853 * (which holds SCL_STATE as reader). There's only one sync thread,
2854 * so this is sufficient to ensure mutual exclusion.
2856 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2857 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2858 spa_config_held(spa, SCL_STATE, RW_READER)));
2860 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2861 list_insert_head(&spa->spa_state_dirty_list, vd);
2865 vdev_state_clean(vdev_t *vd)
2867 spa_t *spa = vd->vdev_spa;
2869 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2870 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2871 spa_config_held(spa, SCL_STATE, RW_READER)));
2873 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2874 list_remove(&spa->spa_state_dirty_list, vd);
2878 * Propagate vdev state up from children to parent.
2881 vdev_propagate_state(vdev_t *vd)
2883 spa_t *spa = vd->vdev_spa;
2884 vdev_t *rvd = spa->spa_root_vdev;
2885 int degraded = 0, faulted = 0;
2889 if (vd->vdev_children > 0) {
2890 for (int c = 0; c < vd->vdev_children; c++) {
2891 child = vd->vdev_child[c];
2894 * Don't factor holes into the decision.
2896 if (child->vdev_ishole)
2899 if (!vdev_readable(child) ||
2900 (!vdev_writeable(child) && spa_writeable(spa))) {
2902 * Root special: if there is a top-level log
2903 * device, treat the root vdev as if it were
2906 if (child->vdev_islog && vd == rvd)
2910 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2914 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2918 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2921 * Root special: if there is a top-level vdev that cannot be
2922 * opened due to corrupted metadata, then propagate the root
2923 * vdev's aux state as 'corrupt' rather than 'insufficient
2926 if (corrupted && vd == rvd &&
2927 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2928 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2929 VDEV_AUX_CORRUPT_DATA);
2932 if (vd->vdev_parent)
2933 vdev_propagate_state(vd->vdev_parent);
2937 * Set a vdev's state. If this is during an open, we don't update the parent
2938 * state, because we're in the process of opening children depth-first.
2939 * Otherwise, we propagate the change to the parent.
2941 * If this routine places a device in a faulted state, an appropriate ereport is
2945 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2947 uint64_t save_state;
2948 spa_t *spa = vd->vdev_spa;
2950 if (state == vd->vdev_state) {
2951 vd->vdev_stat.vs_aux = aux;
2955 save_state = vd->vdev_state;
2957 vd->vdev_state = state;
2958 vd->vdev_stat.vs_aux = aux;
2961 * If we are setting the vdev state to anything but an open state, then
2962 * always close the underlying device unless the device has requested
2963 * a delayed close (i.e. we're about to remove or fault the device).
2964 * Otherwise, we keep accessible but invalid devices open forever.
2965 * We don't call vdev_close() itself, because that implies some extra
2966 * checks (offline, etc) that we don't want here. This is limited to
2967 * leaf devices, because otherwise closing the device will affect other
2970 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2971 vd->vdev_ops->vdev_op_leaf)
2972 vd->vdev_ops->vdev_op_close(vd);
2975 * If we have brought this vdev back into service, we need
2976 * to notify fmd so that it can gracefully repair any outstanding
2977 * cases due to a missing device. We do this in all cases, even those
2978 * that probably don't correlate to a repaired fault. This is sure to
2979 * catch all cases, and we let the zfs-retire agent sort it out. If
2980 * this is a transient state it's OK, as the retire agent will
2981 * double-check the state of the vdev before repairing it.
2983 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2984 vd->vdev_prevstate != state)
2985 zfs_post_state_change(spa, vd);
2987 if (vd->vdev_removed &&
2988 state == VDEV_STATE_CANT_OPEN &&
2989 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2991 * If the previous state is set to VDEV_STATE_REMOVED, then this
2992 * device was previously marked removed and someone attempted to
2993 * reopen it. If this failed due to a nonexistent device, then
2994 * keep the device in the REMOVED state. We also let this be if
2995 * it is one of our special test online cases, which is only
2996 * attempting to online the device and shouldn't generate an FMA
2999 vd->vdev_state = VDEV_STATE_REMOVED;
3000 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3001 } else if (state == VDEV_STATE_REMOVED) {
3002 vd->vdev_removed = B_TRUE;
3003 } else if (state == VDEV_STATE_CANT_OPEN) {
3005 * If we fail to open a vdev during an import or recovery, we
3006 * mark it as "not available", which signifies that it was
3007 * never there to begin with. Failure to open such a device
3008 * is not considered an error.
3010 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3011 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3012 vd->vdev_ops->vdev_op_leaf)
3013 vd->vdev_not_present = 1;
3016 * Post the appropriate ereport. If the 'prevstate' field is
3017 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3018 * that this is part of a vdev_reopen(). In this case, we don't
3019 * want to post the ereport if the device was already in the
3020 * CANT_OPEN state beforehand.
3022 * If the 'checkremove' flag is set, then this is an attempt to
3023 * online the device in response to an insertion event. If we
3024 * hit this case, then we have detected an insertion event for a
3025 * faulted or offline device that wasn't in the removed state.
3026 * In this scenario, we don't post an ereport because we are
3027 * about to replace the device, or attempt an online with
3028 * vdev_forcefault, which will generate the fault for us.
3030 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3031 !vd->vdev_not_present && !vd->vdev_checkremove &&
3032 vd != spa->spa_root_vdev) {
3036 case VDEV_AUX_OPEN_FAILED:
3037 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3039 case VDEV_AUX_CORRUPT_DATA:
3040 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3042 case VDEV_AUX_NO_REPLICAS:
3043 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3045 case VDEV_AUX_BAD_GUID_SUM:
3046 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3048 case VDEV_AUX_TOO_SMALL:
3049 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3051 case VDEV_AUX_BAD_LABEL:
3052 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3055 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3058 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3061 /* Erase any notion of persistent removed state */
3062 vd->vdev_removed = B_FALSE;
3064 vd->vdev_removed = B_FALSE;
3067 if (!isopen && vd->vdev_parent)
3068 vdev_propagate_state(vd->vdev_parent);
3072 * Check the vdev configuration to ensure that it's capable of supporting
3075 * On Solaris, we do not support RAID-Z or partial configuration. In
3076 * addition, only a single top-level vdev is allowed and none of the
3077 * leaves can be wholedisks.
3079 * For FreeBSD, we can boot from any configuration. There is a
3080 * limitation that the boot filesystem must be either uncompressed or
3081 * compresses with lzjb compression but I'm not sure how to enforce
3085 vdev_is_bootable(vdev_t *vd)
3088 if (!vd->vdev_ops->vdev_op_leaf) {
3089 char *vdev_type = vd->vdev_ops->vdev_op_type;
3091 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3092 vd->vdev_children > 1) {
3094 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3095 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3098 } else if (vd->vdev_wholedisk == 1) {
3102 for (int c = 0; c < vd->vdev_children; c++) {
3103 if (!vdev_is_bootable(vd->vdev_child[c]))
3111 * Load the state from the original vdev tree (ovd) which
3112 * we've retrieved from the MOS config object. If the original
3113 * vdev was offline or faulted then we transfer that state to the
3114 * device in the current vdev tree (nvd).
3117 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3119 spa_t *spa = nvd->vdev_spa;
3121 ASSERT(nvd->vdev_top->vdev_islog);
3122 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3123 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3125 for (int c = 0; c < nvd->vdev_children; c++)
3126 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3128 if (nvd->vdev_ops->vdev_op_leaf) {
3130 * Restore the persistent vdev state
3132 nvd->vdev_offline = ovd->vdev_offline;
3133 nvd->vdev_faulted = ovd->vdev_faulted;
3134 nvd->vdev_degraded = ovd->vdev_degraded;
3135 nvd->vdev_removed = ovd->vdev_removed;
3140 * Determine if a log device has valid content. If the vdev was
3141 * removed or faulted in the MOS config then we know that
3142 * the content on the log device has already been written to the pool.
3145 vdev_log_state_valid(vdev_t *vd)
3147 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3151 for (int c = 0; c < vd->vdev_children; c++)
3152 if (vdev_log_state_valid(vd->vdev_child[c]))
3159 * Expand a vdev if possible.
3162 vdev_expand(vdev_t *vd, uint64_t txg)
3164 ASSERT(vd->vdev_top == vd);
3165 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3167 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3168 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3169 vdev_config_dirty(vd);
3177 vdev_split(vdev_t *vd)
3179 vdev_t *cvd, *pvd = vd->vdev_parent;
3181 vdev_remove_child(pvd, vd);
3182 vdev_compact_children(pvd);
3184 cvd = pvd->vdev_child[0];
3185 if (pvd->vdev_children == 1) {
3186 vdev_remove_parent(cvd);
3187 cvd->vdev_splitting = B_TRUE;
3189 vdev_propagate_state(cvd);
3193 vdev_deadman(vdev_t *vd)
3195 for (int c = 0; c < vd->vdev_children; c++) {
3196 vdev_t *cvd = vd->vdev_child[c];
3201 if (vd->vdev_ops->vdev_op_leaf) {
3202 vdev_queue_t *vq = &vd->vdev_queue;
3204 mutex_enter(&vq->vq_lock);
3205 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3206 spa_t *spa = vd->vdev_spa;
3211 * Look at the head of all the pending queues,
3212 * if any I/O has been outstanding for longer than
3213 * the spa_deadman_synctime we panic the system.
3215 fio = avl_first(&vq->vq_active_tree);
3216 delta = gethrtime() - fio->io_timestamp;
3217 if (delta > spa_deadman_synctime(spa)) {
3218 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3219 "delta %lluns, last io %lluns",
3220 fio->io_timestamp, delta,
3221 vq->vq_io_complete_ts);
3222 fm_panic("I/O to pool '%s' appears to be "
3223 "hung on vdev guid %llu at '%s'.",
3225 (long long unsigned int) vd->vdev_guid,
3229 mutex_exit(&vq->vq_lock);