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 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>
41 #include <sys/fs/zfs.h>
44 #include <sys/dsl_scan.h>
46 SYSCTL_DECL(_vfs_zfs);
47 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
50 * Virtual device management.
53 static vdev_ops_t *vdev_ops_table[] = {
70 /* maximum scrub/resilver I/O queue per leaf vdev */
71 int zfs_scrub_limit = 10;
73 TUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit);
74 SYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0,
75 "Maximum scrub/resilver I/O queue");
78 * Given a vdev type, return the appropriate ops vector.
81 vdev_getops(const char *type)
83 vdev_ops_t *ops, **opspp;
85 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
86 if (strcmp(ops->vdev_op_type, type) == 0)
93 * Default asize function: return the MAX of psize with the asize of
94 * all children. This is what's used by anything other than RAID-Z.
97 vdev_default_asize(vdev_t *vd, uint64_t psize)
99 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
102 for (int c = 0; c < vd->vdev_children; c++) {
103 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
104 asize = MAX(asize, csize);
111 * Get the minimum allocatable size. We define the allocatable size as
112 * the vdev's asize rounded to the nearest metaslab. This allows us to
113 * replace or attach devices which don't have the same physical size but
114 * can still satisfy the same number of allocations.
117 vdev_get_min_asize(vdev_t *vd)
119 vdev_t *pvd = vd->vdev_parent;
122 * The our parent is NULL (inactive spare or cache) or is the root,
123 * just return our own asize.
126 return (vd->vdev_asize);
129 * The top-level vdev just returns the allocatable size rounded
130 * to the nearest metaslab.
132 if (vd == vd->vdev_top)
133 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
136 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
137 * so each child must provide at least 1/Nth of its asize.
139 if (pvd->vdev_ops == &vdev_raidz_ops)
140 return (pvd->vdev_min_asize / pvd->vdev_children);
142 return (pvd->vdev_min_asize);
146 vdev_set_min_asize(vdev_t *vd)
148 vd->vdev_min_asize = vdev_get_min_asize(vd);
150 for (int c = 0; c < vd->vdev_children; c++)
151 vdev_set_min_asize(vd->vdev_child[c]);
155 vdev_lookup_top(spa_t *spa, uint64_t vdev)
157 vdev_t *rvd = spa->spa_root_vdev;
159 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
161 if (vdev < rvd->vdev_children) {
162 ASSERT(rvd->vdev_child[vdev] != NULL);
163 return (rvd->vdev_child[vdev]);
170 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
174 if (vd->vdev_guid == guid)
177 for (int c = 0; c < vd->vdev_children; c++)
178 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
186 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
188 size_t oldsize, newsize;
189 uint64_t id = cvd->vdev_id;
192 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
193 ASSERT(cvd->vdev_parent == NULL);
195 cvd->vdev_parent = pvd;
200 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
202 oldsize = pvd->vdev_children * sizeof (vdev_t *);
203 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
204 newsize = pvd->vdev_children * sizeof (vdev_t *);
206 newchild = kmem_zalloc(newsize, KM_SLEEP);
207 if (pvd->vdev_child != NULL) {
208 bcopy(pvd->vdev_child, newchild, oldsize);
209 kmem_free(pvd->vdev_child, oldsize);
212 pvd->vdev_child = newchild;
213 pvd->vdev_child[id] = cvd;
215 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
216 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
219 * Walk up all ancestors to update guid sum.
221 for (; pvd != NULL; pvd = pvd->vdev_parent)
222 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
226 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
229 uint_t id = cvd->vdev_id;
231 ASSERT(cvd->vdev_parent == pvd);
236 ASSERT(id < pvd->vdev_children);
237 ASSERT(pvd->vdev_child[id] == cvd);
239 pvd->vdev_child[id] = NULL;
240 cvd->vdev_parent = NULL;
242 for (c = 0; c < pvd->vdev_children; c++)
243 if (pvd->vdev_child[c])
246 if (c == pvd->vdev_children) {
247 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
248 pvd->vdev_child = NULL;
249 pvd->vdev_children = 0;
253 * Walk up all ancestors to update guid sum.
255 for (; pvd != NULL; pvd = pvd->vdev_parent)
256 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
260 * Remove any holes in the child array.
263 vdev_compact_children(vdev_t *pvd)
265 vdev_t **newchild, *cvd;
266 int oldc = pvd->vdev_children;
269 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
271 for (int c = newc = 0; c < oldc; c++)
272 if (pvd->vdev_child[c])
275 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
277 for (int c = newc = 0; c < oldc; c++) {
278 if ((cvd = pvd->vdev_child[c]) != NULL) {
279 newchild[newc] = cvd;
280 cvd->vdev_id = newc++;
284 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
285 pvd->vdev_child = newchild;
286 pvd->vdev_children = newc;
290 * Allocate and minimally initialize a vdev_t.
293 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
297 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
299 if (spa->spa_root_vdev == NULL) {
300 ASSERT(ops == &vdev_root_ops);
301 spa->spa_root_vdev = vd;
302 spa->spa_load_guid = spa_generate_guid(NULL);
305 if (guid == 0 && ops != &vdev_hole_ops) {
306 if (spa->spa_root_vdev == vd) {
308 * The root vdev's guid will also be the pool guid,
309 * which must be unique among all pools.
311 guid = spa_generate_guid(NULL);
314 * Any other vdev's guid must be unique within the pool.
316 guid = spa_generate_guid(spa);
318 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
323 vd->vdev_guid = guid;
324 vd->vdev_guid_sum = guid;
326 vd->vdev_state = VDEV_STATE_CLOSED;
327 vd->vdev_ishole = (ops == &vdev_hole_ops);
329 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
330 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
331 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
332 for (int t = 0; t < DTL_TYPES; t++) {
333 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
336 txg_list_create(&vd->vdev_ms_list,
337 offsetof(struct metaslab, ms_txg_node));
338 txg_list_create(&vd->vdev_dtl_list,
339 offsetof(struct vdev, vdev_dtl_node));
340 vd->vdev_stat.vs_timestamp = gethrtime();
348 * Allocate a new vdev. The 'alloctype' is used to control whether we are
349 * creating a new vdev or loading an existing one - the behavior is slightly
350 * different for each case.
353 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
358 uint64_t guid = 0, islog, nparity;
361 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
363 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
366 if ((ops = vdev_getops(type)) == NULL)
370 * If this is a load, get the vdev guid from the nvlist.
371 * Otherwise, vdev_alloc_common() will generate one for us.
373 if (alloctype == VDEV_ALLOC_LOAD) {
376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
380 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
382 } else if (alloctype == VDEV_ALLOC_SPARE) {
383 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
385 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
386 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
388 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
389 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
394 * The first allocated vdev must be of type 'root'.
396 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
400 * Determine whether we're a log vdev.
403 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
404 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
407 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
411 * Set the nparity property for RAID-Z vdevs.
414 if (ops == &vdev_raidz_ops) {
415 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
417 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
420 * Previous versions could only support 1 or 2 parity
424 spa_version(spa) < SPA_VERSION_RAIDZ2)
427 spa_version(spa) < SPA_VERSION_RAIDZ3)
431 * We require the parity to be specified for SPAs that
432 * support multiple parity levels.
434 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
437 * Otherwise, we default to 1 parity device for RAID-Z.
444 ASSERT(nparity != -1ULL);
446 vd = vdev_alloc_common(spa, id, guid, ops);
448 vd->vdev_islog = islog;
449 vd->vdev_nparity = nparity;
451 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
452 vd->vdev_path = spa_strdup(vd->vdev_path);
453 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
454 vd->vdev_devid = spa_strdup(vd->vdev_devid);
455 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
456 &vd->vdev_physpath) == 0)
457 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
458 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
459 vd->vdev_fru = spa_strdup(vd->vdev_fru);
462 * Set the whole_disk property. If it's not specified, leave the value
465 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
466 &vd->vdev_wholedisk) != 0)
467 vd->vdev_wholedisk = -1ULL;
470 * Look for the 'not present' flag. This will only be set if the device
471 * was not present at the time of import.
473 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
474 &vd->vdev_not_present);
477 * Get the alignment requirement.
479 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
482 * Retrieve the vdev creation time.
484 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
488 * If we're a top-level vdev, try to load the allocation parameters.
490 if (parent && !parent->vdev_parent &&
491 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
494 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
496 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
498 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
502 if (parent && !parent->vdev_parent) {
503 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
504 alloctype == VDEV_ALLOC_ADD ||
505 alloctype == VDEV_ALLOC_SPLIT ||
506 alloctype == VDEV_ALLOC_ROOTPOOL);
507 vd->vdev_mg = metaslab_group_create(islog ?
508 spa_log_class(spa) : spa_normal_class(spa), vd);
512 * If we're a leaf vdev, try to load the DTL object and other state.
514 if (vd->vdev_ops->vdev_op_leaf &&
515 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
516 alloctype == VDEV_ALLOC_ROOTPOOL)) {
517 if (alloctype == VDEV_ALLOC_LOAD) {
518 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
519 &vd->vdev_dtl_smo.smo_object);
520 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
524 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
527 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
528 &spare) == 0 && spare)
532 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
535 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
536 &vd->vdev_resilvering);
539 * When importing a pool, we want to ignore the persistent fault
540 * state, as the diagnosis made on another system may not be
541 * valid in the current context. Local vdevs will
542 * remain in the faulted state.
544 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
545 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
547 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
549 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
552 if (vd->vdev_faulted || vd->vdev_degraded) {
556 VDEV_AUX_ERR_EXCEEDED;
557 if (nvlist_lookup_string(nv,
558 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
559 strcmp(aux, "external") == 0)
560 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
566 * Add ourselves to the parent's list of children.
568 vdev_add_child(parent, vd);
576 vdev_free(vdev_t *vd)
578 spa_t *spa = vd->vdev_spa;
581 * vdev_free() implies closing the vdev first. This is simpler than
582 * trying to ensure complicated semantics for all callers.
586 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
587 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
592 for (int c = 0; c < vd->vdev_children; c++)
593 vdev_free(vd->vdev_child[c]);
595 ASSERT(vd->vdev_child == NULL);
596 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
599 * Discard allocation state.
601 if (vd->vdev_mg != NULL) {
602 vdev_metaslab_fini(vd);
603 metaslab_group_destroy(vd->vdev_mg);
606 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
607 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
608 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
611 * Remove this vdev from its parent's child list.
613 vdev_remove_child(vd->vdev_parent, vd);
615 ASSERT(vd->vdev_parent == NULL);
618 * Clean up vdev structure.
624 spa_strfree(vd->vdev_path);
626 spa_strfree(vd->vdev_devid);
627 if (vd->vdev_physpath)
628 spa_strfree(vd->vdev_physpath);
630 spa_strfree(vd->vdev_fru);
632 if (vd->vdev_isspare)
633 spa_spare_remove(vd);
634 if (vd->vdev_isl2cache)
635 spa_l2cache_remove(vd);
637 txg_list_destroy(&vd->vdev_ms_list);
638 txg_list_destroy(&vd->vdev_dtl_list);
640 mutex_enter(&vd->vdev_dtl_lock);
641 for (int t = 0; t < DTL_TYPES; t++) {
642 space_map_unload(&vd->vdev_dtl[t]);
643 space_map_destroy(&vd->vdev_dtl[t]);
645 mutex_exit(&vd->vdev_dtl_lock);
647 mutex_destroy(&vd->vdev_dtl_lock);
648 mutex_destroy(&vd->vdev_stat_lock);
649 mutex_destroy(&vd->vdev_probe_lock);
651 if (vd == spa->spa_root_vdev)
652 spa->spa_root_vdev = NULL;
654 kmem_free(vd, sizeof (vdev_t));
658 * Transfer top-level vdev state from svd to tvd.
661 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
663 spa_t *spa = svd->vdev_spa;
668 ASSERT(tvd == tvd->vdev_top);
670 tvd->vdev_ms_array = svd->vdev_ms_array;
671 tvd->vdev_ms_shift = svd->vdev_ms_shift;
672 tvd->vdev_ms_count = svd->vdev_ms_count;
674 svd->vdev_ms_array = 0;
675 svd->vdev_ms_shift = 0;
676 svd->vdev_ms_count = 0;
678 tvd->vdev_mg = svd->vdev_mg;
679 tvd->vdev_ms = svd->vdev_ms;
684 if (tvd->vdev_mg != NULL)
685 tvd->vdev_mg->mg_vd = tvd;
687 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
688 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
689 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
691 svd->vdev_stat.vs_alloc = 0;
692 svd->vdev_stat.vs_space = 0;
693 svd->vdev_stat.vs_dspace = 0;
695 for (t = 0; t < TXG_SIZE; t++) {
696 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
697 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
698 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
699 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
700 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
701 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
704 if (list_link_active(&svd->vdev_config_dirty_node)) {
705 vdev_config_clean(svd);
706 vdev_config_dirty(tvd);
709 if (list_link_active(&svd->vdev_state_dirty_node)) {
710 vdev_state_clean(svd);
711 vdev_state_dirty(tvd);
714 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
715 svd->vdev_deflate_ratio = 0;
717 tvd->vdev_islog = svd->vdev_islog;
722 vdev_top_update(vdev_t *tvd, vdev_t *vd)
729 for (int c = 0; c < vd->vdev_children; c++)
730 vdev_top_update(tvd, vd->vdev_child[c]);
734 * Add a mirror/replacing vdev above an existing vdev.
737 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
739 spa_t *spa = cvd->vdev_spa;
740 vdev_t *pvd = cvd->vdev_parent;
743 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
745 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
747 mvd->vdev_asize = cvd->vdev_asize;
748 mvd->vdev_min_asize = cvd->vdev_min_asize;
749 mvd->vdev_ashift = cvd->vdev_ashift;
750 mvd->vdev_state = cvd->vdev_state;
751 mvd->vdev_crtxg = cvd->vdev_crtxg;
753 vdev_remove_child(pvd, cvd);
754 vdev_add_child(pvd, mvd);
755 cvd->vdev_id = mvd->vdev_children;
756 vdev_add_child(mvd, cvd);
757 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
759 if (mvd == mvd->vdev_top)
760 vdev_top_transfer(cvd, mvd);
766 * Remove a 1-way mirror/replacing vdev from the tree.
769 vdev_remove_parent(vdev_t *cvd)
771 vdev_t *mvd = cvd->vdev_parent;
772 vdev_t *pvd = mvd->vdev_parent;
774 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
776 ASSERT(mvd->vdev_children == 1);
777 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
778 mvd->vdev_ops == &vdev_replacing_ops ||
779 mvd->vdev_ops == &vdev_spare_ops);
780 cvd->vdev_ashift = mvd->vdev_ashift;
782 vdev_remove_child(mvd, cvd);
783 vdev_remove_child(pvd, mvd);
786 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
787 * Otherwise, we could have detached an offline device, and when we
788 * go to import the pool we'll think we have two top-level vdevs,
789 * instead of a different version of the same top-level vdev.
791 if (mvd->vdev_top == mvd) {
792 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
793 cvd->vdev_orig_guid = cvd->vdev_guid;
794 cvd->vdev_guid += guid_delta;
795 cvd->vdev_guid_sum += guid_delta;
797 cvd->vdev_id = mvd->vdev_id;
798 vdev_add_child(pvd, cvd);
799 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
801 if (cvd == cvd->vdev_top)
802 vdev_top_transfer(mvd, cvd);
804 ASSERT(mvd->vdev_children == 0);
809 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
811 spa_t *spa = vd->vdev_spa;
812 objset_t *mos = spa->spa_meta_objset;
814 uint64_t oldc = vd->vdev_ms_count;
815 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
819 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
822 * This vdev is not being allocated from yet or is a hole.
824 if (vd->vdev_ms_shift == 0)
827 ASSERT(!vd->vdev_ishole);
830 * Compute the raidz-deflation ratio. Note, we hard-code
831 * in 128k (1 << 17) because it is the current "typical" blocksize.
832 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
833 * or we will inconsistently account for existing bp's.
835 vd->vdev_deflate_ratio = (1 << 17) /
836 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
838 ASSERT(oldc <= newc);
840 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
843 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
844 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
848 vd->vdev_ms_count = newc;
850 for (m = oldc; m < newc; m++) {
851 space_map_obj_t smo = { 0, 0, 0 };
854 error = dmu_read(mos, vd->vdev_ms_array,
855 m * sizeof (uint64_t), sizeof (uint64_t), &object,
861 error = dmu_bonus_hold(mos, object, FTAG, &db);
864 ASSERT3U(db->db_size, >=, sizeof (smo));
865 bcopy(db->db_data, &smo, sizeof (smo));
866 ASSERT3U(smo.smo_object, ==, object);
867 dmu_buf_rele(db, FTAG);
870 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
871 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
875 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
878 * If the vdev is being removed we don't activate
879 * the metaslabs since we want to ensure that no new
880 * allocations are performed on this device.
882 if (oldc == 0 && !vd->vdev_removing)
883 metaslab_group_activate(vd->vdev_mg);
886 spa_config_exit(spa, SCL_ALLOC, FTAG);
892 vdev_metaslab_fini(vdev_t *vd)
895 uint64_t count = vd->vdev_ms_count;
897 if (vd->vdev_ms != NULL) {
898 metaslab_group_passivate(vd->vdev_mg);
899 for (m = 0; m < count; m++)
900 if (vd->vdev_ms[m] != NULL)
901 metaslab_fini(vd->vdev_ms[m]);
902 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
907 typedef struct vdev_probe_stats {
908 boolean_t vps_readable;
909 boolean_t vps_writeable;
911 } vdev_probe_stats_t;
914 vdev_probe_done(zio_t *zio)
916 spa_t *spa = zio->io_spa;
917 vdev_t *vd = zio->io_vd;
918 vdev_probe_stats_t *vps = zio->io_private;
920 ASSERT(vd->vdev_probe_zio != NULL);
922 if (zio->io_type == ZIO_TYPE_READ) {
923 if (zio->io_error == 0)
924 vps->vps_readable = 1;
925 if (zio->io_error == 0 && spa_writeable(spa)) {
926 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
927 zio->io_offset, zio->io_size, zio->io_data,
928 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
929 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
931 zio_buf_free(zio->io_data, zio->io_size);
933 } else if (zio->io_type == ZIO_TYPE_WRITE) {
934 if (zio->io_error == 0)
935 vps->vps_writeable = 1;
936 zio_buf_free(zio->io_data, zio->io_size);
937 } else if (zio->io_type == ZIO_TYPE_NULL) {
940 vd->vdev_cant_read |= !vps->vps_readable;
941 vd->vdev_cant_write |= !vps->vps_writeable;
943 if (vdev_readable(vd) &&
944 (vdev_writeable(vd) || !spa_writeable(spa))) {
947 ASSERT(zio->io_error != 0);
948 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
949 spa, vd, NULL, 0, 0);
950 zio->io_error = ENXIO;
953 mutex_enter(&vd->vdev_probe_lock);
954 ASSERT(vd->vdev_probe_zio == zio);
955 vd->vdev_probe_zio = NULL;
956 mutex_exit(&vd->vdev_probe_lock);
958 while ((pio = zio_walk_parents(zio)) != NULL)
959 if (!vdev_accessible(vd, pio))
960 pio->io_error = ENXIO;
962 kmem_free(vps, sizeof (*vps));
967 * Determine whether this device is accessible by reading and writing
968 * to several known locations: the pad regions of each vdev label
969 * but the first (which we leave alone in case it contains a VTOC).
972 vdev_probe(vdev_t *vd, zio_t *zio)
974 spa_t *spa = vd->vdev_spa;
975 vdev_probe_stats_t *vps = NULL;
978 ASSERT(vd->vdev_ops->vdev_op_leaf);
981 * Don't probe the probe.
983 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
987 * To prevent 'probe storms' when a device fails, we create
988 * just one probe i/o at a time. All zios that want to probe
989 * this vdev will become parents of the probe io.
991 mutex_enter(&vd->vdev_probe_lock);
993 if ((pio = vd->vdev_probe_zio) == NULL) {
994 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
996 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
997 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1000 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1002 * vdev_cant_read and vdev_cant_write can only
1003 * transition from TRUE to FALSE when we have the
1004 * SCL_ZIO lock as writer; otherwise they can only
1005 * transition from FALSE to TRUE. This ensures that
1006 * any zio looking at these values can assume that
1007 * failures persist for the life of the I/O. That's
1008 * important because when a device has intermittent
1009 * connectivity problems, we want to ensure that
1010 * they're ascribed to the device (ENXIO) and not
1013 * Since we hold SCL_ZIO as writer here, clear both
1014 * values so the probe can reevaluate from first
1017 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1018 vd->vdev_cant_read = B_FALSE;
1019 vd->vdev_cant_write = B_FALSE;
1022 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1023 vdev_probe_done, vps,
1024 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1027 * We can't change the vdev state in this context, so we
1028 * kick off an async task to do it on our behalf.
1031 vd->vdev_probe_wanted = B_TRUE;
1032 spa_async_request(spa, SPA_ASYNC_PROBE);
1037 zio_add_child(zio, pio);
1039 mutex_exit(&vd->vdev_probe_lock);
1042 ASSERT(zio != NULL);
1046 for (int l = 1; l < VDEV_LABELS; l++) {
1047 zio_nowait(zio_read_phys(pio, vd,
1048 vdev_label_offset(vd->vdev_psize, l,
1049 offsetof(vdev_label_t, vl_pad2)),
1050 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1051 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1052 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1063 vdev_open_child(void *arg)
1067 vd->vdev_open_thread = curthread;
1068 vd->vdev_open_error = vdev_open(vd);
1069 vd->vdev_open_thread = NULL;
1073 vdev_uses_zvols(vdev_t *vd)
1075 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1076 strlen(ZVOL_DIR)) == 0)
1078 for (int c = 0; c < vd->vdev_children; c++)
1079 if (vdev_uses_zvols(vd->vdev_child[c]))
1085 vdev_open_children(vdev_t *vd)
1088 int children = vd->vdev_children;
1091 * in order to handle pools on top of zvols, do the opens
1092 * in a single thread so that the same thread holds the
1093 * spa_namespace_lock
1095 if (B_TRUE || vdev_uses_zvols(vd)) {
1096 for (int c = 0; c < children; c++)
1097 vd->vdev_child[c]->vdev_open_error =
1098 vdev_open(vd->vdev_child[c]);
1101 tq = taskq_create("vdev_open", children, minclsyspri,
1102 children, children, TASKQ_PREPOPULATE);
1104 for (int c = 0; c < children; c++)
1105 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1112 * Prepare a virtual device for access.
1115 vdev_open(vdev_t *vd)
1117 spa_t *spa = vd->vdev_spa;
1120 uint64_t asize, psize;
1121 uint64_t ashift = 0;
1123 ASSERT(vd->vdev_open_thread == curthread ||
1124 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1125 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1126 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1127 vd->vdev_state == VDEV_STATE_OFFLINE);
1129 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1130 vd->vdev_cant_read = B_FALSE;
1131 vd->vdev_cant_write = B_FALSE;
1132 vd->vdev_min_asize = vdev_get_min_asize(vd);
1135 * If this vdev is not removed, check its fault status. If it's
1136 * faulted, bail out of the open.
1138 if (!vd->vdev_removed && vd->vdev_faulted) {
1139 ASSERT(vd->vdev_children == 0);
1140 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1141 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1142 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1143 vd->vdev_label_aux);
1145 } else if (vd->vdev_offline) {
1146 ASSERT(vd->vdev_children == 0);
1147 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1151 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1154 * Reset the vdev_reopening flag so that we actually close
1155 * the vdev on error.
1157 vd->vdev_reopening = B_FALSE;
1158 if (zio_injection_enabled && error == 0)
1159 error = zio_handle_device_injection(vd, NULL, ENXIO);
1162 if (vd->vdev_removed &&
1163 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1164 vd->vdev_removed = B_FALSE;
1166 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1167 vd->vdev_stat.vs_aux);
1171 vd->vdev_removed = B_FALSE;
1174 * Recheck the faulted flag now that we have confirmed that
1175 * the vdev is accessible. If we're faulted, bail.
1177 if (vd->vdev_faulted) {
1178 ASSERT(vd->vdev_children == 0);
1179 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1180 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1181 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1182 vd->vdev_label_aux);
1186 if (vd->vdev_degraded) {
1187 ASSERT(vd->vdev_children == 0);
1188 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1189 VDEV_AUX_ERR_EXCEEDED);
1191 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1195 * For hole or missing vdevs we just return success.
1197 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1200 for (int c = 0; c < vd->vdev_children; c++) {
1201 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1202 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1208 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1210 if (vd->vdev_children == 0) {
1211 if (osize < SPA_MINDEVSIZE) {
1212 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1213 VDEV_AUX_TOO_SMALL);
1217 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1219 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1220 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1221 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1222 VDEV_AUX_TOO_SMALL);
1229 vd->vdev_psize = psize;
1232 * Make sure the allocatable size hasn't shrunk.
1234 if (asize < vd->vdev_min_asize) {
1235 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1236 VDEV_AUX_BAD_LABEL);
1240 if (vd->vdev_asize == 0) {
1242 * This is the first-ever open, so use the computed values.
1243 * For testing purposes, a higher ashift can be requested.
1245 vd->vdev_asize = asize;
1246 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1249 * Make sure the alignment requirement hasn't increased.
1251 if (ashift > vd->vdev_top->vdev_ashift) {
1252 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1253 VDEV_AUX_BAD_LABEL);
1259 * If all children are healthy and the asize has increased,
1260 * then we've experienced dynamic LUN growth. If automatic
1261 * expansion is enabled then use the additional space.
1263 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1264 (vd->vdev_expanding || spa->spa_autoexpand))
1265 vd->vdev_asize = asize;
1267 vdev_set_min_asize(vd);
1270 * Ensure we can issue some IO before declaring the
1271 * vdev open for business.
1273 if (vd->vdev_ops->vdev_op_leaf &&
1274 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1275 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1276 VDEV_AUX_ERR_EXCEEDED);
1281 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1282 * resilver. But don't do this if we are doing a reopen for a scrub,
1283 * since this would just restart the scrub we are already doing.
1285 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1286 vdev_resilver_needed(vd, NULL, NULL))
1287 spa_async_request(spa, SPA_ASYNC_RESILVER);
1293 * Called once the vdevs are all opened, this routine validates the label
1294 * contents. This needs to be done before vdev_load() so that we don't
1295 * inadvertently do repair I/Os to the wrong device.
1297 * This function will only return failure if one of the vdevs indicates that it
1298 * has since been destroyed or exported. This is only possible if
1299 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1300 * will be updated but the function will return 0.
1303 vdev_validate(vdev_t *vd)
1305 spa_t *spa = vd->vdev_spa;
1307 uint64_t guid = 0, top_guid;
1310 for (int c = 0; c < vd->vdev_children; c++)
1311 if (vdev_validate(vd->vdev_child[c]) != 0)
1315 * If the device has already failed, or was marked offline, don't do
1316 * any further validation. Otherwise, label I/O will fail and we will
1317 * overwrite the previous state.
1319 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1320 uint64_t aux_guid = 0;
1323 if ((label = vdev_label_read_config(vd)) == NULL) {
1324 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1325 VDEV_AUX_BAD_LABEL);
1330 * Determine if this vdev has been split off into another
1331 * pool. If so, then refuse to open it.
1333 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1334 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1335 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1336 VDEV_AUX_SPLIT_POOL);
1341 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1342 &guid) != 0 || guid != spa_guid(spa)) {
1343 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1344 VDEV_AUX_CORRUPT_DATA);
1349 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1350 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1355 * If this vdev just became a top-level vdev because its
1356 * sibling was detached, it will have adopted the parent's
1357 * vdev guid -- but the label may or may not be on disk yet.
1358 * Fortunately, either version of the label will have the
1359 * same top guid, so if we're a top-level vdev, we can
1360 * safely compare to that instead.
1362 * If we split this vdev off instead, then we also check the
1363 * original pool's guid. We don't want to consider the vdev
1364 * corrupt if it is partway through a split operation.
1366 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1368 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1370 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1371 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1372 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1373 VDEV_AUX_CORRUPT_DATA);
1378 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1380 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1381 VDEV_AUX_CORRUPT_DATA);
1389 * If this is a verbatim import, no need to check the
1390 * state of the pool.
1392 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1393 spa_load_state(spa) == SPA_LOAD_OPEN &&
1394 state != POOL_STATE_ACTIVE)
1398 * If we were able to open and validate a vdev that was
1399 * previously marked permanently unavailable, clear that state
1402 if (vd->vdev_not_present)
1403 vd->vdev_not_present = 0;
1410 * Close a virtual device.
1413 vdev_close(vdev_t *vd)
1415 spa_t *spa = vd->vdev_spa;
1416 vdev_t *pvd = vd->vdev_parent;
1418 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1421 * If our parent is reopening, then we are as well, unless we are
1424 if (pvd != NULL && pvd->vdev_reopening)
1425 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1427 vd->vdev_ops->vdev_op_close(vd);
1429 vdev_cache_purge(vd);
1432 * We record the previous state before we close it, so that if we are
1433 * doing a reopen(), we don't generate FMA ereports if we notice that
1434 * it's still faulted.
1436 vd->vdev_prevstate = vd->vdev_state;
1438 if (vd->vdev_offline)
1439 vd->vdev_state = VDEV_STATE_OFFLINE;
1441 vd->vdev_state = VDEV_STATE_CLOSED;
1442 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1446 vdev_hold(vdev_t *vd)
1448 spa_t *spa = vd->vdev_spa;
1450 ASSERT(spa_is_root(spa));
1451 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1454 for (int c = 0; c < vd->vdev_children; c++)
1455 vdev_hold(vd->vdev_child[c]);
1457 if (vd->vdev_ops->vdev_op_leaf)
1458 vd->vdev_ops->vdev_op_hold(vd);
1462 vdev_rele(vdev_t *vd)
1464 spa_t *spa = vd->vdev_spa;
1466 ASSERT(spa_is_root(spa));
1467 for (int c = 0; c < vd->vdev_children; c++)
1468 vdev_rele(vd->vdev_child[c]);
1470 if (vd->vdev_ops->vdev_op_leaf)
1471 vd->vdev_ops->vdev_op_rele(vd);
1475 * Reopen all interior vdevs and any unopened leaves. We don't actually
1476 * reopen leaf vdevs which had previously been opened as they might deadlock
1477 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1478 * If the leaf has never been opened then open it, as usual.
1481 vdev_reopen(vdev_t *vd)
1483 spa_t *spa = vd->vdev_spa;
1485 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1487 /* set the reopening flag unless we're taking the vdev offline */
1488 vd->vdev_reopening = !vd->vdev_offline;
1490 (void) vdev_open(vd);
1493 * Call vdev_validate() here to make sure we have the same device.
1494 * Otherwise, a device with an invalid label could be successfully
1495 * opened in response to vdev_reopen().
1498 (void) vdev_validate_aux(vd);
1499 if (vdev_readable(vd) && vdev_writeable(vd) &&
1500 vd->vdev_aux == &spa->spa_l2cache &&
1501 !l2arc_vdev_present(vd))
1502 l2arc_add_vdev(spa, vd);
1504 (void) vdev_validate(vd);
1508 * Reassess parent vdev's health.
1510 vdev_propagate_state(vd);
1514 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1519 * Normally, partial opens (e.g. of a mirror) are allowed.
1520 * For a create, however, we want to fail the request if
1521 * there are any components we can't open.
1523 error = vdev_open(vd);
1525 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1527 return (error ? error : ENXIO);
1531 * Recursively initialize all labels.
1533 if ((error = vdev_label_init(vd, txg, isreplacing ?
1534 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1543 vdev_metaslab_set_size(vdev_t *vd)
1546 * Aim for roughly 200 metaslabs per vdev.
1548 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1549 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1553 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1555 ASSERT(vd == vd->vdev_top);
1556 ASSERT(!vd->vdev_ishole);
1557 ASSERT(ISP2(flags));
1558 ASSERT(spa_writeable(vd->vdev_spa));
1560 if (flags & VDD_METASLAB)
1561 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1563 if (flags & VDD_DTL)
1564 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1566 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1572 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1573 * the vdev has less than perfect replication. There are four kinds of DTL:
1575 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1577 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1579 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1580 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1581 * txgs that was scrubbed.
1583 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1584 * persistent errors or just some device being offline.
1585 * Unlike the other three, the DTL_OUTAGE map is not generally
1586 * maintained; it's only computed when needed, typically to
1587 * determine whether a device can be detached.
1589 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1590 * either has the data or it doesn't.
1592 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1593 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1594 * if any child is less than fully replicated, then so is its parent.
1595 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1596 * comprising only those txgs which appear in 'maxfaults' or more children;
1597 * those are the txgs we don't have enough replication to read. For example,
1598 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1599 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1600 * two child DTL_MISSING maps.
1602 * It should be clear from the above that to compute the DTLs and outage maps
1603 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1604 * Therefore, that is all we keep on disk. When loading the pool, or after
1605 * a configuration change, we generate all other DTLs from first principles.
1608 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1610 space_map_t *sm = &vd->vdev_dtl[t];
1612 ASSERT(t < DTL_TYPES);
1613 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1614 ASSERT(spa_writeable(vd->vdev_spa));
1616 mutex_enter(sm->sm_lock);
1617 if (!space_map_contains(sm, txg, size))
1618 space_map_add(sm, txg, size);
1619 mutex_exit(sm->sm_lock);
1623 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1625 space_map_t *sm = &vd->vdev_dtl[t];
1626 boolean_t dirty = B_FALSE;
1628 ASSERT(t < DTL_TYPES);
1629 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1631 mutex_enter(sm->sm_lock);
1632 if (sm->sm_space != 0)
1633 dirty = space_map_contains(sm, txg, size);
1634 mutex_exit(sm->sm_lock);
1640 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1642 space_map_t *sm = &vd->vdev_dtl[t];
1645 mutex_enter(sm->sm_lock);
1646 empty = (sm->sm_space == 0);
1647 mutex_exit(sm->sm_lock);
1653 * Reassess DTLs after a config change or scrub completion.
1656 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1658 spa_t *spa = vd->vdev_spa;
1662 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1664 for (int c = 0; c < vd->vdev_children; c++)
1665 vdev_dtl_reassess(vd->vdev_child[c], txg,
1666 scrub_txg, scrub_done);
1668 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1671 if (vd->vdev_ops->vdev_op_leaf) {
1672 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1674 mutex_enter(&vd->vdev_dtl_lock);
1675 if (scrub_txg != 0 &&
1676 (spa->spa_scrub_started ||
1677 (scn && scn->scn_phys.scn_errors == 0))) {
1679 * We completed a scrub up to scrub_txg. If we
1680 * did it without rebooting, then the scrub dtl
1681 * will be valid, so excise the old region and
1682 * fold in the scrub dtl. Otherwise, leave the
1683 * dtl as-is if there was an error.
1685 * There's little trick here: to excise the beginning
1686 * of the DTL_MISSING map, we put it into a reference
1687 * tree and then add a segment with refcnt -1 that
1688 * covers the range [0, scrub_txg). This means
1689 * that each txg in that range has refcnt -1 or 0.
1690 * We then add DTL_SCRUB with a refcnt of 2, so that
1691 * entries in the range [0, scrub_txg) will have a
1692 * positive refcnt -- either 1 or 2. We then convert
1693 * the reference tree into the new DTL_MISSING map.
1695 space_map_ref_create(&reftree);
1696 space_map_ref_add_map(&reftree,
1697 &vd->vdev_dtl[DTL_MISSING], 1);
1698 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1699 space_map_ref_add_map(&reftree,
1700 &vd->vdev_dtl[DTL_SCRUB], 2);
1701 space_map_ref_generate_map(&reftree,
1702 &vd->vdev_dtl[DTL_MISSING], 1);
1703 space_map_ref_destroy(&reftree);
1705 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1706 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1707 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1709 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1710 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1711 if (!vdev_readable(vd))
1712 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1714 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1715 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1716 mutex_exit(&vd->vdev_dtl_lock);
1719 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1723 mutex_enter(&vd->vdev_dtl_lock);
1724 for (int t = 0; t < DTL_TYPES; t++) {
1725 /* account for child's outage in parent's missing map */
1726 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1728 continue; /* leaf vdevs only */
1729 if (t == DTL_PARTIAL)
1730 minref = 1; /* i.e. non-zero */
1731 else if (vd->vdev_nparity != 0)
1732 minref = vd->vdev_nparity + 1; /* RAID-Z */
1734 minref = vd->vdev_children; /* any kind of mirror */
1735 space_map_ref_create(&reftree);
1736 for (int c = 0; c < vd->vdev_children; c++) {
1737 vdev_t *cvd = vd->vdev_child[c];
1738 mutex_enter(&cvd->vdev_dtl_lock);
1739 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1740 mutex_exit(&cvd->vdev_dtl_lock);
1742 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1743 space_map_ref_destroy(&reftree);
1745 mutex_exit(&vd->vdev_dtl_lock);
1749 vdev_dtl_load(vdev_t *vd)
1751 spa_t *spa = vd->vdev_spa;
1752 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1753 objset_t *mos = spa->spa_meta_objset;
1757 ASSERT(vd->vdev_children == 0);
1759 if (smo->smo_object == 0)
1762 ASSERT(!vd->vdev_ishole);
1764 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1767 ASSERT3U(db->db_size, >=, sizeof (*smo));
1768 bcopy(db->db_data, smo, sizeof (*smo));
1769 dmu_buf_rele(db, FTAG);
1771 mutex_enter(&vd->vdev_dtl_lock);
1772 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1773 NULL, SM_ALLOC, smo, mos);
1774 mutex_exit(&vd->vdev_dtl_lock);
1780 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1782 spa_t *spa = vd->vdev_spa;
1783 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1784 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1785 objset_t *mos = spa->spa_meta_objset;
1791 ASSERT(!vd->vdev_ishole);
1793 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1795 if (vd->vdev_detached) {
1796 if (smo->smo_object != 0) {
1797 int err = dmu_object_free(mos, smo->smo_object, tx);
1798 ASSERT3U(err, ==, 0);
1799 smo->smo_object = 0;
1805 if (smo->smo_object == 0) {
1806 ASSERT(smo->smo_objsize == 0);
1807 ASSERT(smo->smo_alloc == 0);
1808 smo->smo_object = dmu_object_alloc(mos,
1809 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1810 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1811 ASSERT(smo->smo_object != 0);
1812 vdev_config_dirty(vd->vdev_top);
1815 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1817 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1820 mutex_enter(&smlock);
1822 mutex_enter(&vd->vdev_dtl_lock);
1823 space_map_walk(sm, space_map_add, &smsync);
1824 mutex_exit(&vd->vdev_dtl_lock);
1826 space_map_truncate(smo, mos, tx);
1827 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1829 space_map_destroy(&smsync);
1831 mutex_exit(&smlock);
1832 mutex_destroy(&smlock);
1834 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1835 dmu_buf_will_dirty(db, tx);
1836 ASSERT3U(db->db_size, >=, sizeof (*smo));
1837 bcopy(smo, db->db_data, sizeof (*smo));
1838 dmu_buf_rele(db, FTAG);
1844 * Determine whether the specified vdev can be offlined/detached/removed
1845 * without losing data.
1848 vdev_dtl_required(vdev_t *vd)
1850 spa_t *spa = vd->vdev_spa;
1851 vdev_t *tvd = vd->vdev_top;
1852 uint8_t cant_read = vd->vdev_cant_read;
1855 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1857 if (vd == spa->spa_root_vdev || vd == tvd)
1861 * Temporarily mark the device as unreadable, and then determine
1862 * whether this results in any DTL outages in the top-level vdev.
1863 * If not, we can safely offline/detach/remove the device.
1865 vd->vdev_cant_read = B_TRUE;
1866 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1867 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1868 vd->vdev_cant_read = cant_read;
1869 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1871 if (!required && zio_injection_enabled)
1872 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1878 * Determine if resilver is needed, and if so the txg range.
1881 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1883 boolean_t needed = B_FALSE;
1884 uint64_t thismin = UINT64_MAX;
1885 uint64_t thismax = 0;
1887 if (vd->vdev_children == 0) {
1888 mutex_enter(&vd->vdev_dtl_lock);
1889 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1890 vdev_writeable(vd)) {
1893 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1894 thismin = ss->ss_start - 1;
1895 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1896 thismax = ss->ss_end;
1899 mutex_exit(&vd->vdev_dtl_lock);
1901 for (int c = 0; c < vd->vdev_children; c++) {
1902 vdev_t *cvd = vd->vdev_child[c];
1903 uint64_t cmin, cmax;
1905 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1906 thismin = MIN(thismin, cmin);
1907 thismax = MAX(thismax, cmax);
1913 if (needed && minp) {
1921 vdev_load(vdev_t *vd)
1924 * Recursively load all children.
1926 for (int c = 0; c < vd->vdev_children; c++)
1927 vdev_load(vd->vdev_child[c]);
1930 * If this is a top-level vdev, initialize its metaslabs.
1932 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1933 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1934 vdev_metaslab_init(vd, 0) != 0))
1935 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1936 VDEV_AUX_CORRUPT_DATA);
1939 * If this is a leaf vdev, load its DTL.
1941 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1942 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1943 VDEV_AUX_CORRUPT_DATA);
1947 * The special vdev case is used for hot spares and l2cache devices. Its
1948 * sole purpose it to set the vdev state for the associated vdev. To do this,
1949 * we make sure that we can open the underlying device, then try to read the
1950 * label, and make sure that the label is sane and that it hasn't been
1951 * repurposed to another pool.
1954 vdev_validate_aux(vdev_t *vd)
1957 uint64_t guid, version;
1960 if (!vdev_readable(vd))
1963 if ((label = vdev_label_read_config(vd)) == NULL) {
1964 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1965 VDEV_AUX_CORRUPT_DATA);
1969 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1970 version > SPA_VERSION ||
1971 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1972 guid != vd->vdev_guid ||
1973 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1974 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1975 VDEV_AUX_CORRUPT_DATA);
1981 * We don't actually check the pool state here. If it's in fact in
1982 * use by another pool, we update this fact on the fly when requested.
1989 vdev_remove(vdev_t *vd, uint64_t txg)
1991 spa_t *spa = vd->vdev_spa;
1992 objset_t *mos = spa->spa_meta_objset;
1995 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
1997 if (vd->vdev_dtl_smo.smo_object) {
1998 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
1999 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2000 vd->vdev_dtl_smo.smo_object = 0;
2003 if (vd->vdev_ms != NULL) {
2004 for (int m = 0; m < vd->vdev_ms_count; m++) {
2005 metaslab_t *msp = vd->vdev_ms[m];
2007 if (msp == NULL || msp->ms_smo.smo_object == 0)
2010 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2011 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2012 msp->ms_smo.smo_object = 0;
2016 if (vd->vdev_ms_array) {
2017 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2018 vd->vdev_ms_array = 0;
2019 vd->vdev_ms_shift = 0;
2025 vdev_sync_done(vdev_t *vd, uint64_t txg)
2028 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2030 ASSERT(!vd->vdev_ishole);
2032 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2033 metaslab_sync_done(msp, txg);
2036 metaslab_sync_reassess(vd->vdev_mg);
2040 vdev_sync(vdev_t *vd, uint64_t txg)
2042 spa_t *spa = vd->vdev_spa;
2047 ASSERT(!vd->vdev_ishole);
2049 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2050 ASSERT(vd == vd->vdev_top);
2051 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2052 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2053 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2054 ASSERT(vd->vdev_ms_array != 0);
2055 vdev_config_dirty(vd);
2060 * Remove the metadata associated with this vdev once it's empty.
2062 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2063 vdev_remove(vd, txg);
2065 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2066 metaslab_sync(msp, txg);
2067 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2070 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2071 vdev_dtl_sync(lvd, txg);
2073 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2077 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2079 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2083 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2084 * not be opened, and no I/O is attempted.
2087 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2091 spa_vdev_state_enter(spa, SCL_NONE);
2093 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2094 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2096 if (!vd->vdev_ops->vdev_op_leaf)
2097 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2102 * We don't directly use the aux state here, but if we do a
2103 * vdev_reopen(), we need this value to be present to remember why we
2106 vd->vdev_label_aux = aux;
2109 * Faulted state takes precedence over degraded.
2111 vd->vdev_delayed_close = B_FALSE;
2112 vd->vdev_faulted = 1ULL;
2113 vd->vdev_degraded = 0ULL;
2114 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2117 * If this device has the only valid copy of the data, then
2118 * back off and simply mark the vdev as degraded instead.
2120 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2121 vd->vdev_degraded = 1ULL;
2122 vd->vdev_faulted = 0ULL;
2125 * If we reopen the device and it's not dead, only then do we
2130 if (vdev_readable(vd))
2131 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2134 return (spa_vdev_state_exit(spa, vd, 0));
2138 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2139 * user that something is wrong. The vdev continues to operate as normal as far
2140 * as I/O is concerned.
2143 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2147 spa_vdev_state_enter(spa, SCL_NONE);
2149 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2150 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2152 if (!vd->vdev_ops->vdev_op_leaf)
2153 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2156 * If the vdev is already faulted, then don't do anything.
2158 if (vd->vdev_faulted || vd->vdev_degraded)
2159 return (spa_vdev_state_exit(spa, NULL, 0));
2161 vd->vdev_degraded = 1ULL;
2162 if (!vdev_is_dead(vd))
2163 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2166 return (spa_vdev_state_exit(spa, vd, 0));
2170 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2171 * any attached spare device should be detached when the device finishes
2172 * resilvering. Second, the online should be treated like a 'test' online case,
2173 * so no FMA events are generated if the device fails to open.
2176 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2178 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2180 spa_vdev_state_enter(spa, SCL_NONE);
2182 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2183 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2185 if (!vd->vdev_ops->vdev_op_leaf)
2186 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2189 vd->vdev_offline = B_FALSE;
2190 vd->vdev_tmpoffline = B_FALSE;
2191 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2192 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2194 /* XXX - L2ARC 1.0 does not support expansion */
2195 if (!vd->vdev_aux) {
2196 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2197 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2201 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2203 if (!vd->vdev_aux) {
2204 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2205 pvd->vdev_expanding = B_FALSE;
2209 *newstate = vd->vdev_state;
2210 if ((flags & ZFS_ONLINE_UNSPARE) &&
2211 !vdev_is_dead(vd) && vd->vdev_parent &&
2212 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2213 vd->vdev_parent->vdev_child[0] == vd)
2214 vd->vdev_unspare = B_TRUE;
2216 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2218 /* XXX - L2ARC 1.0 does not support expansion */
2220 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2221 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2223 return (spa_vdev_state_exit(spa, vd, 0));
2227 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2231 uint64_t generation;
2232 metaslab_group_t *mg;
2235 spa_vdev_state_enter(spa, SCL_ALLOC);
2237 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2238 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2240 if (!vd->vdev_ops->vdev_op_leaf)
2241 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2245 generation = spa->spa_config_generation + 1;
2248 * If the device isn't already offline, try to offline it.
2250 if (!vd->vdev_offline) {
2252 * If this device has the only valid copy of some data,
2253 * don't allow it to be offlined. Log devices are always
2256 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2257 vdev_dtl_required(vd))
2258 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2261 * If the top-level is a slog and it has had allocations
2262 * then proceed. We check that the vdev's metaslab group
2263 * is not NULL since it's possible that we may have just
2264 * added this vdev but not yet initialized its metaslabs.
2266 if (tvd->vdev_islog && mg != NULL) {
2268 * Prevent any future allocations.
2270 metaslab_group_passivate(mg);
2271 (void) spa_vdev_state_exit(spa, vd, 0);
2273 error = spa_offline_log(spa);
2275 spa_vdev_state_enter(spa, SCL_ALLOC);
2278 * Check to see if the config has changed.
2280 if (error || generation != spa->spa_config_generation) {
2281 metaslab_group_activate(mg);
2283 return (spa_vdev_state_exit(spa,
2285 (void) spa_vdev_state_exit(spa, vd, 0);
2288 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2292 * Offline this device and reopen its top-level vdev.
2293 * If the top-level vdev is a log device then just offline
2294 * it. Otherwise, if this action results in the top-level
2295 * vdev becoming unusable, undo it and fail the request.
2297 vd->vdev_offline = B_TRUE;
2300 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2301 vdev_is_dead(tvd)) {
2302 vd->vdev_offline = B_FALSE;
2304 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2308 * Add the device back into the metaslab rotor so that
2309 * once we online the device it's open for business.
2311 if (tvd->vdev_islog && mg != NULL)
2312 metaslab_group_activate(mg);
2315 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2317 return (spa_vdev_state_exit(spa, vd, 0));
2321 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2325 mutex_enter(&spa->spa_vdev_top_lock);
2326 error = vdev_offline_locked(spa, guid, flags);
2327 mutex_exit(&spa->spa_vdev_top_lock);
2333 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2334 * vdev_offline(), we assume the spa config is locked. We also clear all
2335 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2338 vdev_clear(spa_t *spa, vdev_t *vd)
2340 vdev_t *rvd = spa->spa_root_vdev;
2342 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2347 vd->vdev_stat.vs_read_errors = 0;
2348 vd->vdev_stat.vs_write_errors = 0;
2349 vd->vdev_stat.vs_checksum_errors = 0;
2351 for (int c = 0; c < vd->vdev_children; c++)
2352 vdev_clear(spa, vd->vdev_child[c]);
2355 * If we're in the FAULTED state or have experienced failed I/O, then
2356 * clear the persistent state and attempt to reopen the device. We
2357 * also mark the vdev config dirty, so that the new faulted state is
2358 * written out to disk.
2360 if (vd->vdev_faulted || vd->vdev_degraded ||
2361 !vdev_readable(vd) || !vdev_writeable(vd)) {
2364 * When reopening in reponse to a clear event, it may be due to
2365 * a fmadm repair request. In this case, if the device is
2366 * still broken, we want to still post the ereport again.
2368 vd->vdev_forcefault = B_TRUE;
2370 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2371 vd->vdev_cant_read = B_FALSE;
2372 vd->vdev_cant_write = B_FALSE;
2374 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2376 vd->vdev_forcefault = B_FALSE;
2378 if (vd != rvd && vdev_writeable(vd->vdev_top))
2379 vdev_state_dirty(vd->vdev_top);
2381 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2382 spa_async_request(spa, SPA_ASYNC_RESILVER);
2384 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2388 * When clearing a FMA-diagnosed fault, we always want to
2389 * unspare the device, as we assume that the original spare was
2390 * done in response to the FMA fault.
2392 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2393 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2394 vd->vdev_parent->vdev_child[0] == vd)
2395 vd->vdev_unspare = B_TRUE;
2399 vdev_is_dead(vdev_t *vd)
2402 * Holes and missing devices are always considered "dead".
2403 * This simplifies the code since we don't have to check for
2404 * these types of devices in the various code paths.
2405 * Instead we rely on the fact that we skip over dead devices
2406 * before issuing I/O to them.
2408 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2409 vd->vdev_ops == &vdev_missing_ops);
2413 vdev_readable(vdev_t *vd)
2415 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2419 vdev_writeable(vdev_t *vd)
2421 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2425 vdev_allocatable(vdev_t *vd)
2427 uint64_t state = vd->vdev_state;
2430 * We currently allow allocations from vdevs which may be in the
2431 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2432 * fails to reopen then we'll catch it later when we're holding
2433 * the proper locks. Note that we have to get the vdev state
2434 * in a local variable because although it changes atomically,
2435 * we're asking two separate questions about it.
2437 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2438 !vd->vdev_cant_write && !vd->vdev_ishole);
2442 vdev_accessible(vdev_t *vd, zio_t *zio)
2444 ASSERT(zio->io_vd == vd);
2446 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2449 if (zio->io_type == ZIO_TYPE_READ)
2450 return (!vd->vdev_cant_read);
2452 if (zio->io_type == ZIO_TYPE_WRITE)
2453 return (!vd->vdev_cant_write);
2459 * Get statistics for the given vdev.
2462 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2464 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2466 mutex_enter(&vd->vdev_stat_lock);
2467 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2468 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2469 vs->vs_state = vd->vdev_state;
2470 vs->vs_rsize = vdev_get_min_asize(vd);
2471 if (vd->vdev_ops->vdev_op_leaf)
2472 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2473 mutex_exit(&vd->vdev_stat_lock);
2476 * If we're getting stats on the root vdev, aggregate the I/O counts
2477 * over all top-level vdevs (i.e. the direct children of the root).
2480 for (int c = 0; c < rvd->vdev_children; c++) {
2481 vdev_t *cvd = rvd->vdev_child[c];
2482 vdev_stat_t *cvs = &cvd->vdev_stat;
2484 mutex_enter(&vd->vdev_stat_lock);
2485 for (int t = 0; t < ZIO_TYPES; t++) {
2486 vs->vs_ops[t] += cvs->vs_ops[t];
2487 vs->vs_bytes[t] += cvs->vs_bytes[t];
2489 cvs->vs_scan_removing = cvd->vdev_removing;
2490 mutex_exit(&vd->vdev_stat_lock);
2496 vdev_clear_stats(vdev_t *vd)
2498 mutex_enter(&vd->vdev_stat_lock);
2499 vd->vdev_stat.vs_space = 0;
2500 vd->vdev_stat.vs_dspace = 0;
2501 vd->vdev_stat.vs_alloc = 0;
2502 mutex_exit(&vd->vdev_stat_lock);
2506 vdev_scan_stat_init(vdev_t *vd)
2508 vdev_stat_t *vs = &vd->vdev_stat;
2510 for (int c = 0; c < vd->vdev_children; c++)
2511 vdev_scan_stat_init(vd->vdev_child[c]);
2513 mutex_enter(&vd->vdev_stat_lock);
2514 vs->vs_scan_processed = 0;
2515 mutex_exit(&vd->vdev_stat_lock);
2519 vdev_stat_update(zio_t *zio, uint64_t psize)
2521 spa_t *spa = zio->io_spa;
2522 vdev_t *rvd = spa->spa_root_vdev;
2523 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2525 uint64_t txg = zio->io_txg;
2526 vdev_stat_t *vs = &vd->vdev_stat;
2527 zio_type_t type = zio->io_type;
2528 int flags = zio->io_flags;
2531 * If this i/o is a gang leader, it didn't do any actual work.
2533 if (zio->io_gang_tree)
2536 if (zio->io_error == 0) {
2538 * If this is a root i/o, don't count it -- we've already
2539 * counted the top-level vdevs, and vdev_get_stats() will
2540 * aggregate them when asked. This reduces contention on
2541 * the root vdev_stat_lock and implicitly handles blocks
2542 * that compress away to holes, for which there is no i/o.
2543 * (Holes never create vdev children, so all the counters
2544 * remain zero, which is what we want.)
2546 * Note: this only applies to successful i/o (io_error == 0)
2547 * because unlike i/o counts, errors are not additive.
2548 * When reading a ditto block, for example, failure of
2549 * one top-level vdev does not imply a root-level error.
2554 ASSERT(vd == zio->io_vd);
2556 if (flags & ZIO_FLAG_IO_BYPASS)
2559 mutex_enter(&vd->vdev_stat_lock);
2561 if (flags & ZIO_FLAG_IO_REPAIR) {
2562 if (flags & ZIO_FLAG_SCAN_THREAD) {
2563 dsl_scan_phys_t *scn_phys =
2564 &spa->spa_dsl_pool->dp_scan->scn_phys;
2565 uint64_t *processed = &scn_phys->scn_processed;
2568 if (vd->vdev_ops->vdev_op_leaf)
2569 atomic_add_64(processed, psize);
2570 vs->vs_scan_processed += psize;
2573 if (flags & ZIO_FLAG_SELF_HEAL)
2574 vs->vs_self_healed += psize;
2578 vs->vs_bytes[type] += psize;
2580 mutex_exit(&vd->vdev_stat_lock);
2584 if (flags & ZIO_FLAG_SPECULATIVE)
2588 * If this is an I/O error that is going to be retried, then ignore the
2589 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2590 * hard errors, when in reality they can happen for any number of
2591 * innocuous reasons (bus resets, MPxIO link failure, etc).
2593 if (zio->io_error == EIO &&
2594 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2598 * Intent logs writes won't propagate their error to the root
2599 * I/O so don't mark these types of failures as pool-level
2602 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2605 mutex_enter(&vd->vdev_stat_lock);
2606 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2607 if (zio->io_error == ECKSUM)
2608 vs->vs_checksum_errors++;
2610 vs->vs_read_errors++;
2612 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2613 vs->vs_write_errors++;
2614 mutex_exit(&vd->vdev_stat_lock);
2616 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2617 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2618 (flags & ZIO_FLAG_SCAN_THREAD) ||
2619 spa->spa_claiming)) {
2621 * This is either a normal write (not a repair), or it's
2622 * a repair induced by the scrub thread, or it's a repair
2623 * made by zil_claim() during spa_load() in the first txg.
2624 * In the normal case, we commit the DTL change in the same
2625 * txg as the block was born. In the scrub-induced repair
2626 * case, we know that scrubs run in first-pass syncing context,
2627 * so we commit the DTL change in spa_syncing_txg(spa).
2628 * In the zil_claim() case, we commit in spa_first_txg(spa).
2630 * We currently do not make DTL entries for failed spontaneous
2631 * self-healing writes triggered by normal (non-scrubbing)
2632 * reads, because we have no transactional context in which to
2633 * do so -- and it's not clear that it'd be desirable anyway.
2635 if (vd->vdev_ops->vdev_op_leaf) {
2636 uint64_t commit_txg = txg;
2637 if (flags & ZIO_FLAG_SCAN_THREAD) {
2638 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2639 ASSERT(spa_sync_pass(spa) == 1);
2640 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2641 commit_txg = spa_syncing_txg(spa);
2642 } else if (spa->spa_claiming) {
2643 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2644 commit_txg = spa_first_txg(spa);
2646 ASSERT(commit_txg >= spa_syncing_txg(spa));
2647 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2649 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2650 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2651 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2654 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2659 * Update the in-core space usage stats for this vdev, its metaslab class,
2660 * and the root vdev.
2663 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2664 int64_t space_delta)
2666 int64_t dspace_delta = space_delta;
2667 spa_t *spa = vd->vdev_spa;
2668 vdev_t *rvd = spa->spa_root_vdev;
2669 metaslab_group_t *mg = vd->vdev_mg;
2670 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2672 ASSERT(vd == vd->vdev_top);
2675 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2676 * factor. We must calculate this here and not at the root vdev
2677 * because the root vdev's psize-to-asize is simply the max of its
2678 * childrens', thus not accurate enough for us.
2680 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2681 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2682 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2683 vd->vdev_deflate_ratio;
2685 mutex_enter(&vd->vdev_stat_lock);
2686 vd->vdev_stat.vs_alloc += alloc_delta;
2687 vd->vdev_stat.vs_space += space_delta;
2688 vd->vdev_stat.vs_dspace += dspace_delta;
2689 mutex_exit(&vd->vdev_stat_lock);
2691 if (mc == spa_normal_class(spa)) {
2692 mutex_enter(&rvd->vdev_stat_lock);
2693 rvd->vdev_stat.vs_alloc += alloc_delta;
2694 rvd->vdev_stat.vs_space += space_delta;
2695 rvd->vdev_stat.vs_dspace += dspace_delta;
2696 mutex_exit(&rvd->vdev_stat_lock);
2700 ASSERT(rvd == vd->vdev_parent);
2701 ASSERT(vd->vdev_ms_count != 0);
2703 metaslab_class_space_update(mc,
2704 alloc_delta, defer_delta, space_delta, dspace_delta);
2709 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2710 * so that it will be written out next time the vdev configuration is synced.
2711 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2714 vdev_config_dirty(vdev_t *vd)
2716 spa_t *spa = vd->vdev_spa;
2717 vdev_t *rvd = spa->spa_root_vdev;
2720 ASSERT(spa_writeable(spa));
2723 * If this is an aux vdev (as with l2cache and spare devices), then we
2724 * update the vdev config manually and set the sync flag.
2726 if (vd->vdev_aux != NULL) {
2727 spa_aux_vdev_t *sav = vd->vdev_aux;
2731 for (c = 0; c < sav->sav_count; c++) {
2732 if (sav->sav_vdevs[c] == vd)
2736 if (c == sav->sav_count) {
2738 * We're being removed. There's nothing more to do.
2740 ASSERT(sav->sav_sync == B_TRUE);
2744 sav->sav_sync = B_TRUE;
2746 if (nvlist_lookup_nvlist_array(sav->sav_config,
2747 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2748 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2749 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2755 * Setting the nvlist in the middle if the array is a little
2756 * sketchy, but it will work.
2758 nvlist_free(aux[c]);
2759 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2765 * The dirty list is protected by the SCL_CONFIG lock. The caller
2766 * must either hold SCL_CONFIG as writer, or must be the sync thread
2767 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2768 * so this is sufficient to ensure mutual exclusion.
2770 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2771 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2772 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2775 for (c = 0; c < rvd->vdev_children; c++)
2776 vdev_config_dirty(rvd->vdev_child[c]);
2778 ASSERT(vd == vd->vdev_top);
2780 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2782 list_insert_head(&spa->spa_config_dirty_list, vd);
2787 vdev_config_clean(vdev_t *vd)
2789 spa_t *spa = vd->vdev_spa;
2791 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2792 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2793 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2795 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2796 list_remove(&spa->spa_config_dirty_list, vd);
2800 * Mark a top-level vdev's state as dirty, so that the next pass of
2801 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2802 * the state changes from larger config changes because they require
2803 * much less locking, and are often needed for administrative actions.
2806 vdev_state_dirty(vdev_t *vd)
2808 spa_t *spa = vd->vdev_spa;
2810 ASSERT(spa_writeable(spa));
2811 ASSERT(vd == vd->vdev_top);
2814 * The state list is protected by the SCL_STATE lock. The caller
2815 * must either hold SCL_STATE as writer, or must be the sync thread
2816 * (which holds SCL_STATE as reader). There's only one sync thread,
2817 * so this is sufficient to ensure mutual exclusion.
2819 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2820 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2821 spa_config_held(spa, SCL_STATE, RW_READER)));
2823 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2824 list_insert_head(&spa->spa_state_dirty_list, vd);
2828 vdev_state_clean(vdev_t *vd)
2830 spa_t *spa = vd->vdev_spa;
2832 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2833 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2834 spa_config_held(spa, SCL_STATE, RW_READER)));
2836 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2837 list_remove(&spa->spa_state_dirty_list, vd);
2841 * Propagate vdev state up from children to parent.
2844 vdev_propagate_state(vdev_t *vd)
2846 spa_t *spa = vd->vdev_spa;
2847 vdev_t *rvd = spa->spa_root_vdev;
2848 int degraded = 0, faulted = 0;
2852 if (vd->vdev_children > 0) {
2853 for (int c = 0; c < vd->vdev_children; c++) {
2854 child = vd->vdev_child[c];
2857 * Don't factor holes into the decision.
2859 if (child->vdev_ishole)
2862 if (!vdev_readable(child) ||
2863 (!vdev_writeable(child) && spa_writeable(spa))) {
2865 * Root special: if there is a top-level log
2866 * device, treat the root vdev as if it were
2869 if (child->vdev_islog && vd == rvd)
2873 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2877 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2881 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2884 * Root special: if there is a top-level vdev that cannot be
2885 * opened due to corrupted metadata, then propagate the root
2886 * vdev's aux state as 'corrupt' rather than 'insufficient
2889 if (corrupted && vd == rvd &&
2890 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2891 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2892 VDEV_AUX_CORRUPT_DATA);
2895 if (vd->vdev_parent)
2896 vdev_propagate_state(vd->vdev_parent);
2900 * Set a vdev's state. If this is during an open, we don't update the parent
2901 * state, because we're in the process of opening children depth-first.
2902 * Otherwise, we propagate the change to the parent.
2904 * If this routine places a device in a faulted state, an appropriate ereport is
2908 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2910 uint64_t save_state;
2911 spa_t *spa = vd->vdev_spa;
2913 if (state == vd->vdev_state) {
2914 vd->vdev_stat.vs_aux = aux;
2918 save_state = vd->vdev_state;
2920 vd->vdev_state = state;
2921 vd->vdev_stat.vs_aux = aux;
2924 * If we are setting the vdev state to anything but an open state, then
2925 * always close the underlying device unless the device has requested
2926 * a delayed close (i.e. we're about to remove or fault the device).
2927 * Otherwise, we keep accessible but invalid devices open forever.
2928 * We don't call vdev_close() itself, because that implies some extra
2929 * checks (offline, etc) that we don't want here. This is limited to
2930 * leaf devices, because otherwise closing the device will affect other
2933 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2934 vd->vdev_ops->vdev_op_leaf)
2935 vd->vdev_ops->vdev_op_close(vd);
2938 * If we have brought this vdev back into service, we need
2939 * to notify fmd so that it can gracefully repair any outstanding
2940 * cases due to a missing device. We do this in all cases, even those
2941 * that probably don't correlate to a repaired fault. This is sure to
2942 * catch all cases, and we let the zfs-retire agent sort it out. If
2943 * this is a transient state it's OK, as the retire agent will
2944 * double-check the state of the vdev before repairing it.
2946 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2947 vd->vdev_prevstate != state)
2948 zfs_post_state_change(spa, vd);
2950 if (vd->vdev_removed &&
2951 state == VDEV_STATE_CANT_OPEN &&
2952 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2954 * If the previous state is set to VDEV_STATE_REMOVED, then this
2955 * device was previously marked removed and someone attempted to
2956 * reopen it. If this failed due to a nonexistent device, then
2957 * keep the device in the REMOVED state. We also let this be if
2958 * it is one of our special test online cases, which is only
2959 * attempting to online the device and shouldn't generate an FMA
2962 vd->vdev_state = VDEV_STATE_REMOVED;
2963 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2964 } else if (state == VDEV_STATE_REMOVED) {
2965 vd->vdev_removed = B_TRUE;
2966 } else if (state == VDEV_STATE_CANT_OPEN) {
2968 * If we fail to open a vdev during an import or recovery, we
2969 * mark it as "not available", which signifies that it was
2970 * never there to begin with. Failure to open such a device
2971 * is not considered an error.
2973 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2974 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2975 vd->vdev_ops->vdev_op_leaf)
2976 vd->vdev_not_present = 1;
2979 * Post the appropriate ereport. If the 'prevstate' field is
2980 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2981 * that this is part of a vdev_reopen(). In this case, we don't
2982 * want to post the ereport if the device was already in the
2983 * CANT_OPEN state beforehand.
2985 * If the 'checkremove' flag is set, then this is an attempt to
2986 * online the device in response to an insertion event. If we
2987 * hit this case, then we have detected an insertion event for a
2988 * faulted or offline device that wasn't in the removed state.
2989 * In this scenario, we don't post an ereport because we are
2990 * about to replace the device, or attempt an online with
2991 * vdev_forcefault, which will generate the fault for us.
2993 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
2994 !vd->vdev_not_present && !vd->vdev_checkremove &&
2995 vd != spa->spa_root_vdev) {
2999 case VDEV_AUX_OPEN_FAILED:
3000 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3002 case VDEV_AUX_CORRUPT_DATA:
3003 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3005 case VDEV_AUX_NO_REPLICAS:
3006 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3008 case VDEV_AUX_BAD_GUID_SUM:
3009 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3011 case VDEV_AUX_TOO_SMALL:
3012 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3014 case VDEV_AUX_BAD_LABEL:
3015 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3018 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3021 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3024 /* Erase any notion of persistent removed state */
3025 vd->vdev_removed = B_FALSE;
3027 vd->vdev_removed = B_FALSE;
3030 if (!isopen && vd->vdev_parent)
3031 vdev_propagate_state(vd->vdev_parent);
3035 * Check the vdev configuration to ensure that it's capable of supporting
3038 * On Solaris, we do not support RAID-Z or partial configuration. In
3039 * addition, only a single top-level vdev is allowed and none of the
3040 * leaves can be wholedisks.
3042 * For FreeBSD, we can boot from any configuration. There is a
3043 * limitation that the boot filesystem must be either uncompressed or
3044 * compresses with lzjb compression but I'm not sure how to enforce
3048 vdev_is_bootable(vdev_t *vd)
3051 if (!vd->vdev_ops->vdev_op_leaf) {
3052 char *vdev_type = vd->vdev_ops->vdev_op_type;
3054 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3055 vd->vdev_children > 1) {
3057 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3058 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3061 } else if (vd->vdev_wholedisk == 1) {
3065 for (int c = 0; c < vd->vdev_children; c++) {
3066 if (!vdev_is_bootable(vd->vdev_child[c]))
3074 * Load the state from the original vdev tree (ovd) which
3075 * we've retrieved from the MOS config object. If the original
3076 * vdev was offline or faulted then we transfer that state to the
3077 * device in the current vdev tree (nvd).
3080 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3082 spa_t *spa = nvd->vdev_spa;
3084 ASSERT(nvd->vdev_top->vdev_islog);
3085 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3086 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3088 for (int c = 0; c < nvd->vdev_children; c++)
3089 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3091 if (nvd->vdev_ops->vdev_op_leaf) {
3093 * Restore the persistent vdev state
3095 nvd->vdev_offline = ovd->vdev_offline;
3096 nvd->vdev_faulted = ovd->vdev_faulted;
3097 nvd->vdev_degraded = ovd->vdev_degraded;
3098 nvd->vdev_removed = ovd->vdev_removed;
3103 * Determine if a log device has valid content. If the vdev was
3104 * removed or faulted in the MOS config then we know that
3105 * the content on the log device has already been written to the pool.
3108 vdev_log_state_valid(vdev_t *vd)
3110 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3114 for (int c = 0; c < vd->vdev_children; c++)
3115 if (vdev_log_state_valid(vd->vdev_child[c]))
3122 * Expand a vdev if possible.
3125 vdev_expand(vdev_t *vd, uint64_t txg)
3127 ASSERT(vd->vdev_top == vd);
3128 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3130 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3131 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3132 vdev_config_dirty(vd);
3140 vdev_split(vdev_t *vd)
3142 vdev_t *cvd, *pvd = vd->vdev_parent;
3144 vdev_remove_child(pvd, vd);
3145 vdev_compact_children(pvd);
3147 cvd = pvd->vdev_child[0];
3148 if (pvd->vdev_children == 1) {
3149 vdev_remove_parent(cvd);
3150 cvd->vdev_splitting = B_TRUE;
3152 vdev_propagate_state(cvd);