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.
26 #include <sys/zfs_context.h>
27 #include <sys/fm/fs/zfs.h>
29 #include <sys/spa_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/vdev_impl.h>
33 #include <sys/uberblock_impl.h>
34 #include <sys/metaslab.h>
35 #include <sys/metaslab_impl.h>
36 #include <sys/space_map.h>
39 #include <sys/fs/zfs.h>
42 #include <sys/dsl_scan.h>
44 SYSCTL_DECL(_vfs_zfs);
45 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
48 * Virtual device management.
51 static vdev_ops_t *vdev_ops_table[] = {
68 /* maximum scrub/resilver I/O queue per leaf vdev */
69 int zfs_scrub_limit = 10;
71 TUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit);
72 SYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0,
73 "Maximum scrub/resilver I/O queue");
76 * Given a vdev type, return the appropriate ops vector.
79 vdev_getops(const char *type)
81 vdev_ops_t *ops, **opspp;
83 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
84 if (strcmp(ops->vdev_op_type, type) == 0)
91 * Default asize function: return the MAX of psize with the asize of
92 * all children. This is what's used by anything other than RAID-Z.
95 vdev_default_asize(vdev_t *vd, uint64_t psize)
97 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
100 for (int c = 0; c < vd->vdev_children; c++) {
101 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
102 asize = MAX(asize, csize);
109 * Get the minimum allocatable size. We define the allocatable size as
110 * the vdev's asize rounded to the nearest metaslab. This allows us to
111 * replace or attach devices which don't have the same physical size but
112 * can still satisfy the same number of allocations.
115 vdev_get_min_asize(vdev_t *vd)
117 vdev_t *pvd = vd->vdev_parent;
120 * The our parent is NULL (inactive spare or cache) or is the root,
121 * just return our own asize.
124 return (vd->vdev_asize);
127 * The top-level vdev just returns the allocatable size rounded
128 * to the nearest metaslab.
130 if (vd == vd->vdev_top)
131 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
134 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
135 * so each child must provide at least 1/Nth of its asize.
137 if (pvd->vdev_ops == &vdev_raidz_ops)
138 return (pvd->vdev_min_asize / pvd->vdev_children);
140 return (pvd->vdev_min_asize);
144 vdev_set_min_asize(vdev_t *vd)
146 vd->vdev_min_asize = vdev_get_min_asize(vd);
148 for (int c = 0; c < vd->vdev_children; c++)
149 vdev_set_min_asize(vd->vdev_child[c]);
153 vdev_lookup_top(spa_t *spa, uint64_t vdev)
155 vdev_t *rvd = spa->spa_root_vdev;
157 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
159 if (vdev < rvd->vdev_children) {
160 ASSERT(rvd->vdev_child[vdev] != NULL);
161 return (rvd->vdev_child[vdev]);
168 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
172 if (vd->vdev_guid == guid)
175 for (int c = 0; c < vd->vdev_children; c++)
176 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
184 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
186 size_t oldsize, newsize;
187 uint64_t id = cvd->vdev_id;
190 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
191 ASSERT(cvd->vdev_parent == NULL);
193 cvd->vdev_parent = pvd;
198 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
200 oldsize = pvd->vdev_children * sizeof (vdev_t *);
201 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
202 newsize = pvd->vdev_children * sizeof (vdev_t *);
204 newchild = kmem_zalloc(newsize, KM_SLEEP);
205 if (pvd->vdev_child != NULL) {
206 bcopy(pvd->vdev_child, newchild, oldsize);
207 kmem_free(pvd->vdev_child, oldsize);
210 pvd->vdev_child = newchild;
211 pvd->vdev_child[id] = cvd;
213 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
214 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
217 * Walk up all ancestors to update guid sum.
219 for (; pvd != NULL; pvd = pvd->vdev_parent)
220 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
224 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
227 uint_t id = cvd->vdev_id;
229 ASSERT(cvd->vdev_parent == pvd);
234 ASSERT(id < pvd->vdev_children);
235 ASSERT(pvd->vdev_child[id] == cvd);
237 pvd->vdev_child[id] = NULL;
238 cvd->vdev_parent = NULL;
240 for (c = 0; c < pvd->vdev_children; c++)
241 if (pvd->vdev_child[c])
244 if (c == pvd->vdev_children) {
245 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
246 pvd->vdev_child = NULL;
247 pvd->vdev_children = 0;
251 * Walk up all ancestors to update guid sum.
253 for (; pvd != NULL; pvd = pvd->vdev_parent)
254 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
258 * Remove any holes in the child array.
261 vdev_compact_children(vdev_t *pvd)
263 vdev_t **newchild, *cvd;
264 int oldc = pvd->vdev_children;
267 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
269 for (int c = newc = 0; c < oldc; c++)
270 if (pvd->vdev_child[c])
273 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
275 for (int c = newc = 0; c < oldc; c++) {
276 if ((cvd = pvd->vdev_child[c]) != NULL) {
277 newchild[newc] = cvd;
278 cvd->vdev_id = newc++;
282 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
283 pvd->vdev_child = newchild;
284 pvd->vdev_children = newc;
288 * Allocate and minimally initialize a vdev_t.
291 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
295 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
297 if (spa->spa_root_vdev == NULL) {
298 ASSERT(ops == &vdev_root_ops);
299 spa->spa_root_vdev = vd;
302 if (guid == 0 && ops != &vdev_hole_ops) {
303 if (spa->spa_root_vdev == vd) {
305 * The root vdev's guid will also be the pool guid,
306 * which must be unique among all pools.
308 guid = spa_generate_guid(NULL);
311 * Any other vdev's guid must be unique within the pool.
313 guid = spa_generate_guid(spa);
315 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
320 vd->vdev_guid = guid;
321 vd->vdev_guid_sum = guid;
323 vd->vdev_state = VDEV_STATE_CLOSED;
324 vd->vdev_ishole = (ops == &vdev_hole_ops);
326 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
327 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
328 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
329 for (int t = 0; t < DTL_TYPES; t++) {
330 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
333 txg_list_create(&vd->vdev_ms_list,
334 offsetof(struct metaslab, ms_txg_node));
335 txg_list_create(&vd->vdev_dtl_list,
336 offsetof(struct vdev, vdev_dtl_node));
337 vd->vdev_stat.vs_timestamp = gethrtime();
345 * Allocate a new vdev. The 'alloctype' is used to control whether we are
346 * creating a new vdev or loading an existing one - the behavior is slightly
347 * different for each case.
350 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
355 uint64_t guid = 0, islog, nparity;
358 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
360 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
363 if ((ops = vdev_getops(type)) == NULL)
367 * If this is a load, get the vdev guid from the nvlist.
368 * Otherwise, vdev_alloc_common() will generate one for us.
370 if (alloctype == VDEV_ALLOC_LOAD) {
373 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
377 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
379 } else if (alloctype == VDEV_ALLOC_SPARE) {
380 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
382 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
383 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
385 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
386 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
391 * The first allocated vdev must be of type 'root'.
393 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
397 * Determine whether we're a log vdev.
400 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
401 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
404 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
408 * Set the nparity property for RAID-Z vdevs.
411 if (ops == &vdev_raidz_ops) {
412 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
414 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
417 * Previous versions could only support 1 or 2 parity
421 spa_version(spa) < SPA_VERSION_RAIDZ2)
424 spa_version(spa) < SPA_VERSION_RAIDZ3)
428 * We require the parity to be specified for SPAs that
429 * support multiple parity levels.
431 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
434 * Otherwise, we default to 1 parity device for RAID-Z.
441 ASSERT(nparity != -1ULL);
443 vd = vdev_alloc_common(spa, id, guid, ops);
445 vd->vdev_islog = islog;
446 vd->vdev_nparity = nparity;
448 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
449 vd->vdev_path = spa_strdup(vd->vdev_path);
450 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
451 vd->vdev_devid = spa_strdup(vd->vdev_devid);
452 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
453 &vd->vdev_physpath) == 0)
454 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
455 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
456 vd->vdev_fru = spa_strdup(vd->vdev_fru);
459 * Set the whole_disk property. If it's not specified, leave the value
462 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
463 &vd->vdev_wholedisk) != 0)
464 vd->vdev_wholedisk = -1ULL;
467 * Look for the 'not present' flag. This will only be set if the device
468 * was not present at the time of import.
470 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
471 &vd->vdev_not_present);
474 * Get the alignment requirement.
476 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
479 * Retrieve the vdev creation time.
481 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
485 * If we're a top-level vdev, try to load the allocation parameters.
487 if (parent && !parent->vdev_parent &&
488 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
489 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
491 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
493 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
495 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
499 if (parent && !parent->vdev_parent) {
500 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
501 alloctype == VDEV_ALLOC_ADD ||
502 alloctype == VDEV_ALLOC_SPLIT ||
503 alloctype == VDEV_ALLOC_ROOTPOOL);
504 vd->vdev_mg = metaslab_group_create(islog ?
505 spa_log_class(spa) : spa_normal_class(spa), vd);
509 * If we're a leaf vdev, try to load the DTL object and other state.
511 if (vd->vdev_ops->vdev_op_leaf &&
512 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
513 alloctype == VDEV_ALLOC_ROOTPOOL)) {
514 if (alloctype == VDEV_ALLOC_LOAD) {
515 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
516 &vd->vdev_dtl_smo.smo_object);
517 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
521 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
524 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
525 &spare) == 0 && spare)
529 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
532 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
533 &vd->vdev_resilvering);
536 * When importing a pool, we want to ignore the persistent fault
537 * state, as the diagnosis made on another system may not be
538 * valid in the current context. Local vdevs will
539 * remain in the faulted state.
541 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
542 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
544 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
546 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
549 if (vd->vdev_faulted || vd->vdev_degraded) {
553 VDEV_AUX_ERR_EXCEEDED;
554 if (nvlist_lookup_string(nv,
555 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
556 strcmp(aux, "external") == 0)
557 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
563 * Add ourselves to the parent's list of children.
565 vdev_add_child(parent, vd);
573 vdev_free(vdev_t *vd)
575 spa_t *spa = vd->vdev_spa;
578 * vdev_free() implies closing the vdev first. This is simpler than
579 * trying to ensure complicated semantics for all callers.
583 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
584 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
589 for (int c = 0; c < vd->vdev_children; c++)
590 vdev_free(vd->vdev_child[c]);
592 ASSERT(vd->vdev_child == NULL);
593 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
596 * Discard allocation state.
598 if (vd->vdev_mg != NULL) {
599 vdev_metaslab_fini(vd);
600 metaslab_group_destroy(vd->vdev_mg);
603 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
604 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
605 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
608 * Remove this vdev from its parent's child list.
610 vdev_remove_child(vd->vdev_parent, vd);
612 ASSERT(vd->vdev_parent == NULL);
615 * Clean up vdev structure.
621 spa_strfree(vd->vdev_path);
623 spa_strfree(vd->vdev_devid);
624 if (vd->vdev_physpath)
625 spa_strfree(vd->vdev_physpath);
627 spa_strfree(vd->vdev_fru);
629 if (vd->vdev_isspare)
630 spa_spare_remove(vd);
631 if (vd->vdev_isl2cache)
632 spa_l2cache_remove(vd);
634 txg_list_destroy(&vd->vdev_ms_list);
635 txg_list_destroy(&vd->vdev_dtl_list);
637 mutex_enter(&vd->vdev_dtl_lock);
638 for (int t = 0; t < DTL_TYPES; t++) {
639 space_map_unload(&vd->vdev_dtl[t]);
640 space_map_destroy(&vd->vdev_dtl[t]);
642 mutex_exit(&vd->vdev_dtl_lock);
644 mutex_destroy(&vd->vdev_dtl_lock);
645 mutex_destroy(&vd->vdev_stat_lock);
646 mutex_destroy(&vd->vdev_probe_lock);
648 if (vd == spa->spa_root_vdev)
649 spa->spa_root_vdev = NULL;
651 kmem_free(vd, sizeof (vdev_t));
655 * Transfer top-level vdev state from svd to tvd.
658 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
660 spa_t *spa = svd->vdev_spa;
665 ASSERT(tvd == tvd->vdev_top);
667 tvd->vdev_ms_array = svd->vdev_ms_array;
668 tvd->vdev_ms_shift = svd->vdev_ms_shift;
669 tvd->vdev_ms_count = svd->vdev_ms_count;
671 svd->vdev_ms_array = 0;
672 svd->vdev_ms_shift = 0;
673 svd->vdev_ms_count = 0;
675 tvd->vdev_mg = svd->vdev_mg;
676 tvd->vdev_ms = svd->vdev_ms;
681 if (tvd->vdev_mg != NULL)
682 tvd->vdev_mg->mg_vd = tvd;
684 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
685 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
686 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
688 svd->vdev_stat.vs_alloc = 0;
689 svd->vdev_stat.vs_space = 0;
690 svd->vdev_stat.vs_dspace = 0;
692 for (t = 0; t < TXG_SIZE; t++) {
693 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
694 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
695 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
696 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
697 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
698 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
701 if (list_link_active(&svd->vdev_config_dirty_node)) {
702 vdev_config_clean(svd);
703 vdev_config_dirty(tvd);
706 if (list_link_active(&svd->vdev_state_dirty_node)) {
707 vdev_state_clean(svd);
708 vdev_state_dirty(tvd);
711 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
712 svd->vdev_deflate_ratio = 0;
714 tvd->vdev_islog = svd->vdev_islog;
719 vdev_top_update(vdev_t *tvd, vdev_t *vd)
726 for (int c = 0; c < vd->vdev_children; c++)
727 vdev_top_update(tvd, vd->vdev_child[c]);
731 * Add a mirror/replacing vdev above an existing vdev.
734 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
736 spa_t *spa = cvd->vdev_spa;
737 vdev_t *pvd = cvd->vdev_parent;
740 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
742 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
744 mvd->vdev_asize = cvd->vdev_asize;
745 mvd->vdev_min_asize = cvd->vdev_min_asize;
746 mvd->vdev_ashift = cvd->vdev_ashift;
747 mvd->vdev_state = cvd->vdev_state;
748 mvd->vdev_crtxg = cvd->vdev_crtxg;
750 vdev_remove_child(pvd, cvd);
751 vdev_add_child(pvd, mvd);
752 cvd->vdev_id = mvd->vdev_children;
753 vdev_add_child(mvd, cvd);
754 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
756 if (mvd == mvd->vdev_top)
757 vdev_top_transfer(cvd, mvd);
763 * Remove a 1-way mirror/replacing vdev from the tree.
766 vdev_remove_parent(vdev_t *cvd)
768 vdev_t *mvd = cvd->vdev_parent;
769 vdev_t *pvd = mvd->vdev_parent;
771 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
773 ASSERT(mvd->vdev_children == 1);
774 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
775 mvd->vdev_ops == &vdev_replacing_ops ||
776 mvd->vdev_ops == &vdev_spare_ops);
777 cvd->vdev_ashift = mvd->vdev_ashift;
779 vdev_remove_child(mvd, cvd);
780 vdev_remove_child(pvd, mvd);
783 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
784 * Otherwise, we could have detached an offline device, and when we
785 * go to import the pool we'll think we have two top-level vdevs,
786 * instead of a different version of the same top-level vdev.
788 if (mvd->vdev_top == mvd) {
789 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
790 cvd->vdev_orig_guid = cvd->vdev_guid;
791 cvd->vdev_guid += guid_delta;
792 cvd->vdev_guid_sum += guid_delta;
794 cvd->vdev_id = mvd->vdev_id;
795 vdev_add_child(pvd, cvd);
796 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
798 if (cvd == cvd->vdev_top)
799 vdev_top_transfer(mvd, cvd);
801 ASSERT(mvd->vdev_children == 0);
806 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
808 spa_t *spa = vd->vdev_spa;
809 objset_t *mos = spa->spa_meta_objset;
811 uint64_t oldc = vd->vdev_ms_count;
812 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
816 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
819 * This vdev is not being allocated from yet or is a hole.
821 if (vd->vdev_ms_shift == 0)
824 ASSERT(!vd->vdev_ishole);
827 * Compute the raidz-deflation ratio. Note, we hard-code
828 * in 128k (1 << 17) because it is the current "typical" blocksize.
829 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
830 * or we will inconsistently account for existing bp's.
832 vd->vdev_deflate_ratio = (1 << 17) /
833 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
835 ASSERT(oldc <= newc);
837 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
840 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
841 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
845 vd->vdev_ms_count = newc;
847 for (m = oldc; m < newc; m++) {
848 space_map_obj_t smo = { 0, 0, 0 };
851 error = dmu_read(mos, vd->vdev_ms_array,
852 m * sizeof (uint64_t), sizeof (uint64_t), &object,
858 error = dmu_bonus_hold(mos, object, FTAG, &db);
861 ASSERT3U(db->db_size, >=, sizeof (smo));
862 bcopy(db->db_data, &smo, sizeof (smo));
863 ASSERT3U(smo.smo_object, ==, object);
864 dmu_buf_rele(db, FTAG);
867 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
868 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
872 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
875 * If the vdev is being removed we don't activate
876 * the metaslabs since we want to ensure that no new
877 * allocations are performed on this device.
879 if (oldc == 0 && !vd->vdev_removing)
880 metaslab_group_activate(vd->vdev_mg);
883 spa_config_exit(spa, SCL_ALLOC, FTAG);
889 vdev_metaslab_fini(vdev_t *vd)
892 uint64_t count = vd->vdev_ms_count;
894 if (vd->vdev_ms != NULL) {
895 metaslab_group_passivate(vd->vdev_mg);
896 for (m = 0; m < count; m++)
897 if (vd->vdev_ms[m] != NULL)
898 metaslab_fini(vd->vdev_ms[m]);
899 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
904 typedef struct vdev_probe_stats {
905 boolean_t vps_readable;
906 boolean_t vps_writeable;
908 } vdev_probe_stats_t;
911 vdev_probe_done(zio_t *zio)
913 spa_t *spa = zio->io_spa;
914 vdev_t *vd = zio->io_vd;
915 vdev_probe_stats_t *vps = zio->io_private;
917 ASSERT(vd->vdev_probe_zio != NULL);
919 if (zio->io_type == ZIO_TYPE_READ) {
920 if (zio->io_error == 0)
921 vps->vps_readable = 1;
922 if (zio->io_error == 0 && spa_writeable(spa)) {
923 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
924 zio->io_offset, zio->io_size, zio->io_data,
925 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
926 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
928 zio_buf_free(zio->io_data, zio->io_size);
930 } else if (zio->io_type == ZIO_TYPE_WRITE) {
931 if (zio->io_error == 0)
932 vps->vps_writeable = 1;
933 zio_buf_free(zio->io_data, zio->io_size);
934 } else if (zio->io_type == ZIO_TYPE_NULL) {
937 vd->vdev_cant_read |= !vps->vps_readable;
938 vd->vdev_cant_write |= !vps->vps_writeable;
940 if (vdev_readable(vd) &&
941 (vdev_writeable(vd) || !spa_writeable(spa))) {
944 ASSERT(zio->io_error != 0);
945 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
946 spa, vd, NULL, 0, 0);
947 zio->io_error = ENXIO;
950 mutex_enter(&vd->vdev_probe_lock);
951 ASSERT(vd->vdev_probe_zio == zio);
952 vd->vdev_probe_zio = NULL;
953 mutex_exit(&vd->vdev_probe_lock);
955 while ((pio = zio_walk_parents(zio)) != NULL)
956 if (!vdev_accessible(vd, pio))
957 pio->io_error = ENXIO;
959 kmem_free(vps, sizeof (*vps));
964 * Determine whether this device is accessible by reading and writing
965 * to several known locations: the pad regions of each vdev label
966 * but the first (which we leave alone in case it contains a VTOC).
969 vdev_probe(vdev_t *vd, zio_t *zio)
971 spa_t *spa = vd->vdev_spa;
972 vdev_probe_stats_t *vps = NULL;
975 ASSERT(vd->vdev_ops->vdev_op_leaf);
978 * Don't probe the probe.
980 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
984 * To prevent 'probe storms' when a device fails, we create
985 * just one probe i/o at a time. All zios that want to probe
986 * this vdev will become parents of the probe io.
988 mutex_enter(&vd->vdev_probe_lock);
990 if ((pio = vd->vdev_probe_zio) == NULL) {
991 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
993 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
994 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
997 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
999 * vdev_cant_read and vdev_cant_write can only
1000 * transition from TRUE to FALSE when we have the
1001 * SCL_ZIO lock as writer; otherwise they can only
1002 * transition from FALSE to TRUE. This ensures that
1003 * any zio looking at these values can assume that
1004 * failures persist for the life of the I/O. That's
1005 * important because when a device has intermittent
1006 * connectivity problems, we want to ensure that
1007 * they're ascribed to the device (ENXIO) and not
1010 * Since we hold SCL_ZIO as writer here, clear both
1011 * values so the probe can reevaluate from first
1014 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1015 vd->vdev_cant_read = B_FALSE;
1016 vd->vdev_cant_write = B_FALSE;
1019 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1020 vdev_probe_done, vps,
1021 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1024 * We can't change the vdev state in this context, so we
1025 * kick off an async task to do it on our behalf.
1028 vd->vdev_probe_wanted = B_TRUE;
1029 spa_async_request(spa, SPA_ASYNC_PROBE);
1034 zio_add_child(zio, pio);
1036 mutex_exit(&vd->vdev_probe_lock);
1039 ASSERT(zio != NULL);
1043 for (int l = 1; l < VDEV_LABELS; l++) {
1044 zio_nowait(zio_read_phys(pio, vd,
1045 vdev_label_offset(vd->vdev_psize, l,
1046 offsetof(vdev_label_t, vl_pad2)),
1047 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1048 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1049 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1060 vdev_open_child(void *arg)
1064 vd->vdev_open_thread = curthread;
1065 vd->vdev_open_error = vdev_open(vd);
1066 vd->vdev_open_thread = NULL;
1070 vdev_uses_zvols(vdev_t *vd)
1072 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1073 strlen(ZVOL_DIR)) == 0)
1075 for (int c = 0; c < vd->vdev_children; c++)
1076 if (vdev_uses_zvols(vd->vdev_child[c]))
1082 vdev_open_children(vdev_t *vd)
1085 int children = vd->vdev_children;
1088 * in order to handle pools on top of zvols, do the opens
1089 * in a single thread so that the same thread holds the
1090 * spa_namespace_lock
1092 if (B_TRUE || vdev_uses_zvols(vd)) {
1093 for (int c = 0; c < children; c++)
1094 vd->vdev_child[c]->vdev_open_error =
1095 vdev_open(vd->vdev_child[c]);
1098 tq = taskq_create("vdev_open", children, minclsyspri,
1099 children, children, TASKQ_PREPOPULATE);
1101 for (int c = 0; c < children; c++)
1102 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1109 * Prepare a virtual device for access.
1112 vdev_open(vdev_t *vd)
1114 spa_t *spa = vd->vdev_spa;
1117 uint64_t asize, psize;
1118 uint64_t ashift = 0;
1120 ASSERT(vd->vdev_open_thread == curthread ||
1121 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1122 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1123 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1124 vd->vdev_state == VDEV_STATE_OFFLINE);
1126 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1127 vd->vdev_cant_read = B_FALSE;
1128 vd->vdev_cant_write = B_FALSE;
1129 vd->vdev_min_asize = vdev_get_min_asize(vd);
1132 * If this vdev is not removed, check its fault status. If it's
1133 * faulted, bail out of the open.
1135 if (!vd->vdev_removed && vd->vdev_faulted) {
1136 ASSERT(vd->vdev_children == 0);
1137 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1138 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1139 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1140 vd->vdev_label_aux);
1142 } else if (vd->vdev_offline) {
1143 ASSERT(vd->vdev_children == 0);
1144 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1148 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1151 * Reset the vdev_reopening flag so that we actually close
1152 * the vdev on error.
1154 vd->vdev_reopening = B_FALSE;
1155 if (zio_injection_enabled && error == 0)
1156 error = zio_handle_device_injection(vd, NULL, ENXIO);
1159 if (vd->vdev_removed &&
1160 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1161 vd->vdev_removed = B_FALSE;
1163 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1164 vd->vdev_stat.vs_aux);
1168 vd->vdev_removed = B_FALSE;
1171 * Recheck the faulted flag now that we have confirmed that
1172 * the vdev is accessible. If we're faulted, bail.
1174 if (vd->vdev_faulted) {
1175 ASSERT(vd->vdev_children == 0);
1176 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1177 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1178 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1179 vd->vdev_label_aux);
1183 if (vd->vdev_degraded) {
1184 ASSERT(vd->vdev_children == 0);
1185 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1186 VDEV_AUX_ERR_EXCEEDED);
1188 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1192 * For hole or missing vdevs we just return success.
1194 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1197 for (int c = 0; c < vd->vdev_children; c++) {
1198 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1199 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1205 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1207 if (vd->vdev_children == 0) {
1208 if (osize < SPA_MINDEVSIZE) {
1209 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1210 VDEV_AUX_TOO_SMALL);
1214 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1216 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1217 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1218 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1219 VDEV_AUX_TOO_SMALL);
1226 vd->vdev_psize = psize;
1229 * Make sure the allocatable size hasn't shrunk.
1231 if (asize < vd->vdev_min_asize) {
1232 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1233 VDEV_AUX_BAD_LABEL);
1237 if (vd->vdev_asize == 0) {
1239 * This is the first-ever open, so use the computed values.
1240 * For testing purposes, a higher ashift can be requested.
1242 vd->vdev_asize = asize;
1243 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1246 * Make sure the alignment requirement hasn't increased.
1248 if (ashift > vd->vdev_top->vdev_ashift) {
1249 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1250 VDEV_AUX_BAD_LABEL);
1256 * If all children are healthy and the asize has increased,
1257 * then we've experienced dynamic LUN growth. If automatic
1258 * expansion is enabled then use the additional space.
1260 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1261 (vd->vdev_expanding || spa->spa_autoexpand))
1262 vd->vdev_asize = asize;
1264 vdev_set_min_asize(vd);
1267 * Ensure we can issue some IO before declaring the
1268 * vdev open for business.
1270 if (vd->vdev_ops->vdev_op_leaf &&
1271 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1272 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1273 VDEV_AUX_ERR_EXCEEDED);
1278 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1279 * resilver. But don't do this if we are doing a reopen for a scrub,
1280 * since this would just restart the scrub we are already doing.
1282 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1283 vdev_resilver_needed(vd, NULL, NULL))
1284 spa_async_request(spa, SPA_ASYNC_RESILVER);
1290 * Called once the vdevs are all opened, this routine validates the label
1291 * contents. This needs to be done before vdev_load() so that we don't
1292 * inadvertently do repair I/Os to the wrong device.
1294 * This function will only return failure if one of the vdevs indicates that it
1295 * has since been destroyed or exported. This is only possible if
1296 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1297 * will be updated but the function will return 0.
1300 vdev_validate(vdev_t *vd)
1302 spa_t *spa = vd->vdev_spa;
1304 uint64_t guid = 0, top_guid;
1307 for (int c = 0; c < vd->vdev_children; c++)
1308 if (vdev_validate(vd->vdev_child[c]) != 0)
1312 * If the device has already failed, or was marked offline, don't do
1313 * any further validation. Otherwise, label I/O will fail and we will
1314 * overwrite the previous state.
1316 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1317 uint64_t aux_guid = 0;
1320 if ((label = vdev_label_read_config(vd)) == NULL) {
1321 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1322 VDEV_AUX_BAD_LABEL);
1327 * Determine if this vdev has been split off into another
1328 * pool. If so, then refuse to open it.
1330 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1331 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1332 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1333 VDEV_AUX_SPLIT_POOL);
1338 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1339 &guid) != 0 || guid != spa_guid(spa)) {
1340 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1341 VDEV_AUX_CORRUPT_DATA);
1346 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1347 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1352 * If this vdev just became a top-level vdev because its
1353 * sibling was detached, it will have adopted the parent's
1354 * vdev guid -- but the label may or may not be on disk yet.
1355 * Fortunately, either version of the label will have the
1356 * same top guid, so if we're a top-level vdev, we can
1357 * safely compare to that instead.
1359 * If we split this vdev off instead, then we also check the
1360 * original pool's guid. We don't want to consider the vdev
1361 * corrupt if it is partway through a split operation.
1363 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1365 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1367 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1368 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1369 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1370 VDEV_AUX_CORRUPT_DATA);
1375 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1377 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1378 VDEV_AUX_CORRUPT_DATA);
1386 * If this is a verbatim import, no need to check the
1387 * state of the pool.
1389 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1390 spa_load_state(spa) == SPA_LOAD_OPEN &&
1391 state != POOL_STATE_ACTIVE)
1395 * If we were able to open and validate a vdev that was
1396 * previously marked permanently unavailable, clear that state
1399 if (vd->vdev_not_present)
1400 vd->vdev_not_present = 0;
1407 * Close a virtual device.
1410 vdev_close(vdev_t *vd)
1412 spa_t *spa = vd->vdev_spa;
1413 vdev_t *pvd = vd->vdev_parent;
1415 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1418 * If our parent is reopening, then we are as well, unless we are
1421 if (pvd != NULL && pvd->vdev_reopening)
1422 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1424 vd->vdev_ops->vdev_op_close(vd);
1426 vdev_cache_purge(vd);
1429 * We record the previous state before we close it, so that if we are
1430 * doing a reopen(), we don't generate FMA ereports if we notice that
1431 * it's still faulted.
1433 vd->vdev_prevstate = vd->vdev_state;
1435 if (vd->vdev_offline)
1436 vd->vdev_state = VDEV_STATE_OFFLINE;
1438 vd->vdev_state = VDEV_STATE_CLOSED;
1439 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1443 vdev_hold(vdev_t *vd)
1445 spa_t *spa = vd->vdev_spa;
1447 ASSERT(spa_is_root(spa));
1448 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1451 for (int c = 0; c < vd->vdev_children; c++)
1452 vdev_hold(vd->vdev_child[c]);
1454 if (vd->vdev_ops->vdev_op_leaf)
1455 vd->vdev_ops->vdev_op_hold(vd);
1459 vdev_rele(vdev_t *vd)
1461 spa_t *spa = vd->vdev_spa;
1463 ASSERT(spa_is_root(spa));
1464 for (int c = 0; c < vd->vdev_children; c++)
1465 vdev_rele(vd->vdev_child[c]);
1467 if (vd->vdev_ops->vdev_op_leaf)
1468 vd->vdev_ops->vdev_op_rele(vd);
1472 * Reopen all interior vdevs and any unopened leaves. We don't actually
1473 * reopen leaf vdevs which had previously been opened as they might deadlock
1474 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1475 * If the leaf has never been opened then open it, as usual.
1478 vdev_reopen(vdev_t *vd)
1480 spa_t *spa = vd->vdev_spa;
1482 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1484 /* set the reopening flag unless we're taking the vdev offline */
1485 vd->vdev_reopening = !vd->vdev_offline;
1487 (void) vdev_open(vd);
1490 * Call vdev_validate() here to make sure we have the same device.
1491 * Otherwise, a device with an invalid label could be successfully
1492 * opened in response to vdev_reopen().
1495 (void) vdev_validate_aux(vd);
1496 if (vdev_readable(vd) && vdev_writeable(vd) &&
1497 vd->vdev_aux == &spa->spa_l2cache &&
1498 !l2arc_vdev_present(vd))
1499 l2arc_add_vdev(spa, vd);
1501 (void) vdev_validate(vd);
1505 * Reassess parent vdev's health.
1507 vdev_propagate_state(vd);
1511 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1516 * Normally, partial opens (e.g. of a mirror) are allowed.
1517 * For a create, however, we want to fail the request if
1518 * there are any components we can't open.
1520 error = vdev_open(vd);
1522 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1524 return (error ? error : ENXIO);
1528 * Recursively initialize all labels.
1530 if ((error = vdev_label_init(vd, txg, isreplacing ?
1531 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1540 vdev_metaslab_set_size(vdev_t *vd)
1543 * Aim for roughly 200 metaslabs per vdev.
1545 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1546 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1550 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1552 ASSERT(vd == vd->vdev_top);
1553 ASSERT(!vd->vdev_ishole);
1554 ASSERT(ISP2(flags));
1555 ASSERT(spa_writeable(vd->vdev_spa));
1557 if (flags & VDD_METASLAB)
1558 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1560 if (flags & VDD_DTL)
1561 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1563 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1569 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1570 * the vdev has less than perfect replication. There are four kinds of DTL:
1572 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1574 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1576 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1577 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1578 * txgs that was scrubbed.
1580 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1581 * persistent errors or just some device being offline.
1582 * Unlike the other three, the DTL_OUTAGE map is not generally
1583 * maintained; it's only computed when needed, typically to
1584 * determine whether a device can be detached.
1586 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1587 * either has the data or it doesn't.
1589 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1590 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1591 * if any child is less than fully replicated, then so is its parent.
1592 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1593 * comprising only those txgs which appear in 'maxfaults' or more children;
1594 * those are the txgs we don't have enough replication to read. For example,
1595 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1596 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1597 * two child DTL_MISSING maps.
1599 * It should be clear from the above that to compute the DTLs and outage maps
1600 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1601 * Therefore, that is all we keep on disk. When loading the pool, or after
1602 * a configuration change, we generate all other DTLs from first principles.
1605 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1607 space_map_t *sm = &vd->vdev_dtl[t];
1609 ASSERT(t < DTL_TYPES);
1610 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1611 ASSERT(spa_writeable(vd->vdev_spa));
1613 mutex_enter(sm->sm_lock);
1614 if (!space_map_contains(sm, txg, size))
1615 space_map_add(sm, txg, size);
1616 mutex_exit(sm->sm_lock);
1620 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1622 space_map_t *sm = &vd->vdev_dtl[t];
1623 boolean_t dirty = B_FALSE;
1625 ASSERT(t < DTL_TYPES);
1626 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1628 mutex_enter(sm->sm_lock);
1629 if (sm->sm_space != 0)
1630 dirty = space_map_contains(sm, txg, size);
1631 mutex_exit(sm->sm_lock);
1637 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1639 space_map_t *sm = &vd->vdev_dtl[t];
1642 mutex_enter(sm->sm_lock);
1643 empty = (sm->sm_space == 0);
1644 mutex_exit(sm->sm_lock);
1650 * Reassess DTLs after a config change or scrub completion.
1653 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1655 spa_t *spa = vd->vdev_spa;
1659 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1661 for (int c = 0; c < vd->vdev_children; c++)
1662 vdev_dtl_reassess(vd->vdev_child[c], txg,
1663 scrub_txg, scrub_done);
1665 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1668 if (vd->vdev_ops->vdev_op_leaf) {
1669 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1671 mutex_enter(&vd->vdev_dtl_lock);
1672 if (scrub_txg != 0 &&
1673 (spa->spa_scrub_started ||
1674 (scn && scn->scn_phys.scn_errors == 0))) {
1676 * We completed a scrub up to scrub_txg. If we
1677 * did it without rebooting, then the scrub dtl
1678 * will be valid, so excise the old region and
1679 * fold in the scrub dtl. Otherwise, leave the
1680 * dtl as-is if there was an error.
1682 * There's little trick here: to excise the beginning
1683 * of the DTL_MISSING map, we put it into a reference
1684 * tree and then add a segment with refcnt -1 that
1685 * covers the range [0, scrub_txg). This means
1686 * that each txg in that range has refcnt -1 or 0.
1687 * We then add DTL_SCRUB with a refcnt of 2, so that
1688 * entries in the range [0, scrub_txg) will have a
1689 * positive refcnt -- either 1 or 2. We then convert
1690 * the reference tree into the new DTL_MISSING map.
1692 space_map_ref_create(&reftree);
1693 space_map_ref_add_map(&reftree,
1694 &vd->vdev_dtl[DTL_MISSING], 1);
1695 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1696 space_map_ref_add_map(&reftree,
1697 &vd->vdev_dtl[DTL_SCRUB], 2);
1698 space_map_ref_generate_map(&reftree,
1699 &vd->vdev_dtl[DTL_MISSING], 1);
1700 space_map_ref_destroy(&reftree);
1702 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1703 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1704 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1706 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1707 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1708 if (!vdev_readable(vd))
1709 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1711 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1712 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1713 mutex_exit(&vd->vdev_dtl_lock);
1716 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1720 mutex_enter(&vd->vdev_dtl_lock);
1721 for (int t = 0; t < DTL_TYPES; t++) {
1722 /* account for child's outage in parent's missing map */
1723 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1725 continue; /* leaf vdevs only */
1726 if (t == DTL_PARTIAL)
1727 minref = 1; /* i.e. non-zero */
1728 else if (vd->vdev_nparity != 0)
1729 minref = vd->vdev_nparity + 1; /* RAID-Z */
1731 minref = vd->vdev_children; /* any kind of mirror */
1732 space_map_ref_create(&reftree);
1733 for (int c = 0; c < vd->vdev_children; c++) {
1734 vdev_t *cvd = vd->vdev_child[c];
1735 mutex_enter(&cvd->vdev_dtl_lock);
1736 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1737 mutex_exit(&cvd->vdev_dtl_lock);
1739 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1740 space_map_ref_destroy(&reftree);
1742 mutex_exit(&vd->vdev_dtl_lock);
1746 vdev_dtl_load(vdev_t *vd)
1748 spa_t *spa = vd->vdev_spa;
1749 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1750 objset_t *mos = spa->spa_meta_objset;
1754 ASSERT(vd->vdev_children == 0);
1756 if (smo->smo_object == 0)
1759 ASSERT(!vd->vdev_ishole);
1761 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1764 ASSERT3U(db->db_size, >=, sizeof (*smo));
1765 bcopy(db->db_data, smo, sizeof (*smo));
1766 dmu_buf_rele(db, FTAG);
1768 mutex_enter(&vd->vdev_dtl_lock);
1769 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1770 NULL, SM_ALLOC, smo, mos);
1771 mutex_exit(&vd->vdev_dtl_lock);
1777 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1779 spa_t *spa = vd->vdev_spa;
1780 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1781 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1782 objset_t *mos = spa->spa_meta_objset;
1788 ASSERT(!vd->vdev_ishole);
1790 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1792 if (vd->vdev_detached) {
1793 if (smo->smo_object != 0) {
1794 int err = dmu_object_free(mos, smo->smo_object, tx);
1795 ASSERT3U(err, ==, 0);
1796 smo->smo_object = 0;
1802 if (smo->smo_object == 0) {
1803 ASSERT(smo->smo_objsize == 0);
1804 ASSERT(smo->smo_alloc == 0);
1805 smo->smo_object = dmu_object_alloc(mos,
1806 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1807 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1808 ASSERT(smo->smo_object != 0);
1809 vdev_config_dirty(vd->vdev_top);
1812 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1814 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1817 mutex_enter(&smlock);
1819 mutex_enter(&vd->vdev_dtl_lock);
1820 space_map_walk(sm, space_map_add, &smsync);
1821 mutex_exit(&vd->vdev_dtl_lock);
1823 space_map_truncate(smo, mos, tx);
1824 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1826 space_map_destroy(&smsync);
1828 mutex_exit(&smlock);
1829 mutex_destroy(&smlock);
1831 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1832 dmu_buf_will_dirty(db, tx);
1833 ASSERT3U(db->db_size, >=, sizeof (*smo));
1834 bcopy(smo, db->db_data, sizeof (*smo));
1835 dmu_buf_rele(db, FTAG);
1841 * Determine whether the specified vdev can be offlined/detached/removed
1842 * without losing data.
1845 vdev_dtl_required(vdev_t *vd)
1847 spa_t *spa = vd->vdev_spa;
1848 vdev_t *tvd = vd->vdev_top;
1849 uint8_t cant_read = vd->vdev_cant_read;
1852 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1854 if (vd == spa->spa_root_vdev || vd == tvd)
1858 * Temporarily mark the device as unreadable, and then determine
1859 * whether this results in any DTL outages in the top-level vdev.
1860 * If not, we can safely offline/detach/remove the device.
1862 vd->vdev_cant_read = B_TRUE;
1863 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1864 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1865 vd->vdev_cant_read = cant_read;
1866 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1868 if (!required && zio_injection_enabled)
1869 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1875 * Determine if resilver is needed, and if so the txg range.
1878 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1880 boolean_t needed = B_FALSE;
1881 uint64_t thismin = UINT64_MAX;
1882 uint64_t thismax = 0;
1884 if (vd->vdev_children == 0) {
1885 mutex_enter(&vd->vdev_dtl_lock);
1886 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1887 vdev_writeable(vd)) {
1890 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1891 thismin = ss->ss_start - 1;
1892 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1893 thismax = ss->ss_end;
1896 mutex_exit(&vd->vdev_dtl_lock);
1898 for (int c = 0; c < vd->vdev_children; c++) {
1899 vdev_t *cvd = vd->vdev_child[c];
1900 uint64_t cmin, cmax;
1902 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1903 thismin = MIN(thismin, cmin);
1904 thismax = MAX(thismax, cmax);
1910 if (needed && minp) {
1918 vdev_load(vdev_t *vd)
1921 * Recursively load all children.
1923 for (int c = 0; c < vd->vdev_children; c++)
1924 vdev_load(vd->vdev_child[c]);
1927 * If this is a top-level vdev, initialize its metaslabs.
1929 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1930 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1931 vdev_metaslab_init(vd, 0) != 0))
1932 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1933 VDEV_AUX_CORRUPT_DATA);
1936 * If this is a leaf vdev, load its DTL.
1938 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1939 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1940 VDEV_AUX_CORRUPT_DATA);
1944 * The special vdev case is used for hot spares and l2cache devices. Its
1945 * sole purpose it to set the vdev state for the associated vdev. To do this,
1946 * we make sure that we can open the underlying device, then try to read the
1947 * label, and make sure that the label is sane and that it hasn't been
1948 * repurposed to another pool.
1951 vdev_validate_aux(vdev_t *vd)
1954 uint64_t guid, version;
1957 if (!vdev_readable(vd))
1960 if ((label = vdev_label_read_config(vd)) == NULL) {
1961 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1962 VDEV_AUX_CORRUPT_DATA);
1966 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1967 version > SPA_VERSION ||
1968 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1969 guid != vd->vdev_guid ||
1970 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1971 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1972 VDEV_AUX_CORRUPT_DATA);
1978 * We don't actually check the pool state here. If it's in fact in
1979 * use by another pool, we update this fact on the fly when requested.
1986 vdev_remove(vdev_t *vd, uint64_t txg)
1988 spa_t *spa = vd->vdev_spa;
1989 objset_t *mos = spa->spa_meta_objset;
1992 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
1994 if (vd->vdev_dtl_smo.smo_object) {
1995 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
1996 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
1997 vd->vdev_dtl_smo.smo_object = 0;
2000 if (vd->vdev_ms != NULL) {
2001 for (int m = 0; m < vd->vdev_ms_count; m++) {
2002 metaslab_t *msp = vd->vdev_ms[m];
2004 if (msp == NULL || msp->ms_smo.smo_object == 0)
2007 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2008 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2009 msp->ms_smo.smo_object = 0;
2013 if (vd->vdev_ms_array) {
2014 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2015 vd->vdev_ms_array = 0;
2016 vd->vdev_ms_shift = 0;
2022 vdev_sync_done(vdev_t *vd, uint64_t txg)
2025 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2027 ASSERT(!vd->vdev_ishole);
2029 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2030 metaslab_sync_done(msp, txg);
2033 metaslab_sync_reassess(vd->vdev_mg);
2037 vdev_sync(vdev_t *vd, uint64_t txg)
2039 spa_t *spa = vd->vdev_spa;
2044 ASSERT(!vd->vdev_ishole);
2046 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2047 ASSERT(vd == vd->vdev_top);
2048 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2049 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2050 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2051 ASSERT(vd->vdev_ms_array != 0);
2052 vdev_config_dirty(vd);
2057 * Remove the metadata associated with this vdev once it's empty.
2059 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2060 vdev_remove(vd, txg);
2062 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2063 metaslab_sync(msp, txg);
2064 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2067 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2068 vdev_dtl_sync(lvd, txg);
2070 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2074 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2076 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2080 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2081 * not be opened, and no I/O is attempted.
2084 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2088 spa_vdev_state_enter(spa, SCL_NONE);
2090 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2091 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2093 if (!vd->vdev_ops->vdev_op_leaf)
2094 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2099 * We don't directly use the aux state here, but if we do a
2100 * vdev_reopen(), we need this value to be present to remember why we
2103 vd->vdev_label_aux = aux;
2106 * Faulted state takes precedence over degraded.
2108 vd->vdev_delayed_close = B_FALSE;
2109 vd->vdev_faulted = 1ULL;
2110 vd->vdev_degraded = 0ULL;
2111 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2114 * If this device has the only valid copy of the data, then
2115 * back off and simply mark the vdev as degraded instead.
2117 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2118 vd->vdev_degraded = 1ULL;
2119 vd->vdev_faulted = 0ULL;
2122 * If we reopen the device and it's not dead, only then do we
2127 if (vdev_readable(vd))
2128 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2131 return (spa_vdev_state_exit(spa, vd, 0));
2135 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2136 * user that something is wrong. The vdev continues to operate as normal as far
2137 * as I/O is concerned.
2140 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2144 spa_vdev_state_enter(spa, SCL_NONE);
2146 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2147 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2149 if (!vd->vdev_ops->vdev_op_leaf)
2150 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2153 * If the vdev is already faulted, then don't do anything.
2155 if (vd->vdev_faulted || vd->vdev_degraded)
2156 return (spa_vdev_state_exit(spa, NULL, 0));
2158 vd->vdev_degraded = 1ULL;
2159 if (!vdev_is_dead(vd))
2160 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2163 return (spa_vdev_state_exit(spa, vd, 0));
2167 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2168 * any attached spare device should be detached when the device finishes
2169 * resilvering. Second, the online should be treated like a 'test' online case,
2170 * so no FMA events are generated if the device fails to open.
2173 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2175 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2177 spa_vdev_state_enter(spa, SCL_NONE);
2179 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2180 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2182 if (!vd->vdev_ops->vdev_op_leaf)
2183 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2186 vd->vdev_offline = B_FALSE;
2187 vd->vdev_tmpoffline = B_FALSE;
2188 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2189 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2191 /* XXX - L2ARC 1.0 does not support expansion */
2192 if (!vd->vdev_aux) {
2193 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2194 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2198 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2200 if (!vd->vdev_aux) {
2201 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2202 pvd->vdev_expanding = B_FALSE;
2206 *newstate = vd->vdev_state;
2207 if ((flags & ZFS_ONLINE_UNSPARE) &&
2208 !vdev_is_dead(vd) && vd->vdev_parent &&
2209 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2210 vd->vdev_parent->vdev_child[0] == vd)
2211 vd->vdev_unspare = B_TRUE;
2213 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2215 /* XXX - L2ARC 1.0 does not support expansion */
2217 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2218 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2220 return (spa_vdev_state_exit(spa, vd, 0));
2224 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2228 uint64_t generation;
2229 metaslab_group_t *mg;
2232 spa_vdev_state_enter(spa, SCL_ALLOC);
2234 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2235 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2237 if (!vd->vdev_ops->vdev_op_leaf)
2238 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2242 generation = spa->spa_config_generation + 1;
2245 * If the device isn't already offline, try to offline it.
2247 if (!vd->vdev_offline) {
2249 * If this device has the only valid copy of some data,
2250 * don't allow it to be offlined. Log devices are always
2253 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2254 vdev_dtl_required(vd))
2255 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2258 * If the top-level is a slog and it has had allocations
2259 * then proceed. We check that the vdev's metaslab group
2260 * is not NULL since it's possible that we may have just
2261 * added this vdev but not yet initialized its metaslabs.
2263 if (tvd->vdev_islog && mg != NULL) {
2265 * Prevent any future allocations.
2267 metaslab_group_passivate(mg);
2268 (void) spa_vdev_state_exit(spa, vd, 0);
2270 error = spa_offline_log(spa);
2272 spa_vdev_state_enter(spa, SCL_ALLOC);
2275 * Check to see if the config has changed.
2277 if (error || generation != spa->spa_config_generation) {
2278 metaslab_group_activate(mg);
2280 return (spa_vdev_state_exit(spa,
2282 (void) spa_vdev_state_exit(spa, vd, 0);
2285 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2289 * Offline this device and reopen its top-level vdev.
2290 * If the top-level vdev is a log device then just offline
2291 * it. Otherwise, if this action results in the top-level
2292 * vdev becoming unusable, undo it and fail the request.
2294 vd->vdev_offline = B_TRUE;
2297 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2298 vdev_is_dead(tvd)) {
2299 vd->vdev_offline = B_FALSE;
2301 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2305 * Add the device back into the metaslab rotor so that
2306 * once we online the device it's open for business.
2308 if (tvd->vdev_islog && mg != NULL)
2309 metaslab_group_activate(mg);
2312 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2314 return (spa_vdev_state_exit(spa, vd, 0));
2318 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2322 mutex_enter(&spa->spa_vdev_top_lock);
2323 error = vdev_offline_locked(spa, guid, flags);
2324 mutex_exit(&spa->spa_vdev_top_lock);
2330 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2331 * vdev_offline(), we assume the spa config is locked. We also clear all
2332 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2335 vdev_clear(spa_t *spa, vdev_t *vd)
2337 vdev_t *rvd = spa->spa_root_vdev;
2339 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2344 vd->vdev_stat.vs_read_errors = 0;
2345 vd->vdev_stat.vs_write_errors = 0;
2346 vd->vdev_stat.vs_checksum_errors = 0;
2348 for (int c = 0; c < vd->vdev_children; c++)
2349 vdev_clear(spa, vd->vdev_child[c]);
2352 * If we're in the FAULTED state or have experienced failed I/O, then
2353 * clear the persistent state and attempt to reopen the device. We
2354 * also mark the vdev config dirty, so that the new faulted state is
2355 * written out to disk.
2357 if (vd->vdev_faulted || vd->vdev_degraded ||
2358 !vdev_readable(vd) || !vdev_writeable(vd)) {
2361 * When reopening in reponse to a clear event, it may be due to
2362 * a fmadm repair request. In this case, if the device is
2363 * still broken, we want to still post the ereport again.
2365 vd->vdev_forcefault = B_TRUE;
2367 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2368 vd->vdev_cant_read = B_FALSE;
2369 vd->vdev_cant_write = B_FALSE;
2371 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2373 vd->vdev_forcefault = B_FALSE;
2375 if (vd != rvd && vdev_writeable(vd->vdev_top))
2376 vdev_state_dirty(vd->vdev_top);
2378 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2379 spa_async_request(spa, SPA_ASYNC_RESILVER);
2381 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2385 * When clearing a FMA-diagnosed fault, we always want to
2386 * unspare the device, as we assume that the original spare was
2387 * done in response to the FMA fault.
2389 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2390 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2391 vd->vdev_parent->vdev_child[0] == vd)
2392 vd->vdev_unspare = B_TRUE;
2396 vdev_is_dead(vdev_t *vd)
2399 * Holes and missing devices are always considered "dead".
2400 * This simplifies the code since we don't have to check for
2401 * these types of devices in the various code paths.
2402 * Instead we rely on the fact that we skip over dead devices
2403 * before issuing I/O to them.
2405 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2406 vd->vdev_ops == &vdev_missing_ops);
2410 vdev_readable(vdev_t *vd)
2412 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2416 vdev_writeable(vdev_t *vd)
2418 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2422 vdev_allocatable(vdev_t *vd)
2424 uint64_t state = vd->vdev_state;
2427 * We currently allow allocations from vdevs which may be in the
2428 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2429 * fails to reopen then we'll catch it later when we're holding
2430 * the proper locks. Note that we have to get the vdev state
2431 * in a local variable because although it changes atomically,
2432 * we're asking two separate questions about it.
2434 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2435 !vd->vdev_cant_write && !vd->vdev_ishole);
2439 vdev_accessible(vdev_t *vd, zio_t *zio)
2441 ASSERT(zio->io_vd == vd);
2443 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2446 if (zio->io_type == ZIO_TYPE_READ)
2447 return (!vd->vdev_cant_read);
2449 if (zio->io_type == ZIO_TYPE_WRITE)
2450 return (!vd->vdev_cant_write);
2456 * Get statistics for the given vdev.
2459 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2461 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2463 mutex_enter(&vd->vdev_stat_lock);
2464 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2465 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2466 vs->vs_state = vd->vdev_state;
2467 vs->vs_rsize = vdev_get_min_asize(vd);
2468 if (vd->vdev_ops->vdev_op_leaf)
2469 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2470 mutex_exit(&vd->vdev_stat_lock);
2473 * If we're getting stats on the root vdev, aggregate the I/O counts
2474 * over all top-level vdevs (i.e. the direct children of the root).
2477 for (int c = 0; c < rvd->vdev_children; c++) {
2478 vdev_t *cvd = rvd->vdev_child[c];
2479 vdev_stat_t *cvs = &cvd->vdev_stat;
2481 mutex_enter(&vd->vdev_stat_lock);
2482 for (int t = 0; t < ZIO_TYPES; t++) {
2483 vs->vs_ops[t] += cvs->vs_ops[t];
2484 vs->vs_bytes[t] += cvs->vs_bytes[t];
2486 cvs->vs_scan_removing = cvd->vdev_removing;
2487 mutex_exit(&vd->vdev_stat_lock);
2493 vdev_clear_stats(vdev_t *vd)
2495 mutex_enter(&vd->vdev_stat_lock);
2496 vd->vdev_stat.vs_space = 0;
2497 vd->vdev_stat.vs_dspace = 0;
2498 vd->vdev_stat.vs_alloc = 0;
2499 mutex_exit(&vd->vdev_stat_lock);
2503 vdev_scan_stat_init(vdev_t *vd)
2505 vdev_stat_t *vs = &vd->vdev_stat;
2507 for (int c = 0; c < vd->vdev_children; c++)
2508 vdev_scan_stat_init(vd->vdev_child[c]);
2510 mutex_enter(&vd->vdev_stat_lock);
2511 vs->vs_scan_processed = 0;
2512 mutex_exit(&vd->vdev_stat_lock);
2516 vdev_stat_update(zio_t *zio, uint64_t psize)
2518 spa_t *spa = zio->io_spa;
2519 vdev_t *rvd = spa->spa_root_vdev;
2520 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2522 uint64_t txg = zio->io_txg;
2523 vdev_stat_t *vs = &vd->vdev_stat;
2524 zio_type_t type = zio->io_type;
2525 int flags = zio->io_flags;
2528 * If this i/o is a gang leader, it didn't do any actual work.
2530 if (zio->io_gang_tree)
2533 if (zio->io_error == 0) {
2535 * If this is a root i/o, don't count it -- we've already
2536 * counted the top-level vdevs, and vdev_get_stats() will
2537 * aggregate them when asked. This reduces contention on
2538 * the root vdev_stat_lock and implicitly handles blocks
2539 * that compress away to holes, for which there is no i/o.
2540 * (Holes never create vdev children, so all the counters
2541 * remain zero, which is what we want.)
2543 * Note: this only applies to successful i/o (io_error == 0)
2544 * because unlike i/o counts, errors are not additive.
2545 * When reading a ditto block, for example, failure of
2546 * one top-level vdev does not imply a root-level error.
2551 ASSERT(vd == zio->io_vd);
2553 if (flags & ZIO_FLAG_IO_BYPASS)
2556 mutex_enter(&vd->vdev_stat_lock);
2558 if (flags & ZIO_FLAG_IO_REPAIR) {
2559 if (flags & ZIO_FLAG_SCAN_THREAD) {
2560 dsl_scan_phys_t *scn_phys =
2561 &spa->spa_dsl_pool->dp_scan->scn_phys;
2562 uint64_t *processed = &scn_phys->scn_processed;
2565 if (vd->vdev_ops->vdev_op_leaf)
2566 atomic_add_64(processed, psize);
2567 vs->vs_scan_processed += psize;
2570 if (flags & ZIO_FLAG_SELF_HEAL)
2571 vs->vs_self_healed += psize;
2575 vs->vs_bytes[type] += psize;
2577 mutex_exit(&vd->vdev_stat_lock);
2581 if (flags & ZIO_FLAG_SPECULATIVE)
2585 * If this is an I/O error that is going to be retried, then ignore the
2586 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2587 * hard errors, when in reality they can happen for any number of
2588 * innocuous reasons (bus resets, MPxIO link failure, etc).
2590 if (zio->io_error == EIO &&
2591 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2595 * Intent logs writes won't propagate their error to the root
2596 * I/O so don't mark these types of failures as pool-level
2599 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2602 mutex_enter(&vd->vdev_stat_lock);
2603 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2604 if (zio->io_error == ECKSUM)
2605 vs->vs_checksum_errors++;
2607 vs->vs_read_errors++;
2609 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2610 vs->vs_write_errors++;
2611 mutex_exit(&vd->vdev_stat_lock);
2613 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2614 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2615 (flags & ZIO_FLAG_SCAN_THREAD) ||
2616 spa->spa_claiming)) {
2618 * This is either a normal write (not a repair), or it's
2619 * a repair induced by the scrub thread, or it's a repair
2620 * made by zil_claim() during spa_load() in the first txg.
2621 * In the normal case, we commit the DTL change in the same
2622 * txg as the block was born. In the scrub-induced repair
2623 * case, we know that scrubs run in first-pass syncing context,
2624 * so we commit the DTL change in spa_syncing_txg(spa).
2625 * In the zil_claim() case, we commit in spa_first_txg(spa).
2627 * We currently do not make DTL entries for failed spontaneous
2628 * self-healing writes triggered by normal (non-scrubbing)
2629 * reads, because we have no transactional context in which to
2630 * do so -- and it's not clear that it'd be desirable anyway.
2632 if (vd->vdev_ops->vdev_op_leaf) {
2633 uint64_t commit_txg = txg;
2634 if (flags & ZIO_FLAG_SCAN_THREAD) {
2635 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2636 ASSERT(spa_sync_pass(spa) == 1);
2637 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2638 commit_txg = spa_syncing_txg(spa);
2639 } else if (spa->spa_claiming) {
2640 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2641 commit_txg = spa_first_txg(spa);
2643 ASSERT(commit_txg >= spa_syncing_txg(spa));
2644 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2646 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2647 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2648 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2651 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2656 * Update the in-core space usage stats for this vdev, its metaslab class,
2657 * and the root vdev.
2660 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2661 int64_t space_delta)
2663 int64_t dspace_delta = space_delta;
2664 spa_t *spa = vd->vdev_spa;
2665 vdev_t *rvd = spa->spa_root_vdev;
2666 metaslab_group_t *mg = vd->vdev_mg;
2667 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2669 ASSERT(vd == vd->vdev_top);
2672 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2673 * factor. We must calculate this here and not at the root vdev
2674 * because the root vdev's psize-to-asize is simply the max of its
2675 * childrens', thus not accurate enough for us.
2677 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2678 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2679 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2680 vd->vdev_deflate_ratio;
2682 mutex_enter(&vd->vdev_stat_lock);
2683 vd->vdev_stat.vs_alloc += alloc_delta;
2684 vd->vdev_stat.vs_space += space_delta;
2685 vd->vdev_stat.vs_dspace += dspace_delta;
2686 mutex_exit(&vd->vdev_stat_lock);
2688 if (mc == spa_normal_class(spa)) {
2689 mutex_enter(&rvd->vdev_stat_lock);
2690 rvd->vdev_stat.vs_alloc += alloc_delta;
2691 rvd->vdev_stat.vs_space += space_delta;
2692 rvd->vdev_stat.vs_dspace += dspace_delta;
2693 mutex_exit(&rvd->vdev_stat_lock);
2697 ASSERT(rvd == vd->vdev_parent);
2698 ASSERT(vd->vdev_ms_count != 0);
2700 metaslab_class_space_update(mc,
2701 alloc_delta, defer_delta, space_delta, dspace_delta);
2706 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2707 * so that it will be written out next time the vdev configuration is synced.
2708 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2711 vdev_config_dirty(vdev_t *vd)
2713 spa_t *spa = vd->vdev_spa;
2714 vdev_t *rvd = spa->spa_root_vdev;
2717 ASSERT(spa_writeable(spa));
2720 * If this is an aux vdev (as with l2cache and spare devices), then we
2721 * update the vdev config manually and set the sync flag.
2723 if (vd->vdev_aux != NULL) {
2724 spa_aux_vdev_t *sav = vd->vdev_aux;
2728 for (c = 0; c < sav->sav_count; c++) {
2729 if (sav->sav_vdevs[c] == vd)
2733 if (c == sav->sav_count) {
2735 * We're being removed. There's nothing more to do.
2737 ASSERT(sav->sav_sync == B_TRUE);
2741 sav->sav_sync = B_TRUE;
2743 if (nvlist_lookup_nvlist_array(sav->sav_config,
2744 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2745 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2746 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2752 * Setting the nvlist in the middle if the array is a little
2753 * sketchy, but it will work.
2755 nvlist_free(aux[c]);
2756 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2762 * The dirty list is protected by the SCL_CONFIG lock. The caller
2763 * must either hold SCL_CONFIG as writer, or must be the sync thread
2764 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2765 * so this is sufficient to ensure mutual exclusion.
2767 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2768 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2769 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2772 for (c = 0; c < rvd->vdev_children; c++)
2773 vdev_config_dirty(rvd->vdev_child[c]);
2775 ASSERT(vd == vd->vdev_top);
2777 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2779 list_insert_head(&spa->spa_config_dirty_list, vd);
2784 vdev_config_clean(vdev_t *vd)
2786 spa_t *spa = vd->vdev_spa;
2788 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2789 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2790 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2792 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2793 list_remove(&spa->spa_config_dirty_list, vd);
2797 * Mark a top-level vdev's state as dirty, so that the next pass of
2798 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2799 * the state changes from larger config changes because they require
2800 * much less locking, and are often needed for administrative actions.
2803 vdev_state_dirty(vdev_t *vd)
2805 spa_t *spa = vd->vdev_spa;
2807 ASSERT(spa_writeable(spa));
2808 ASSERT(vd == vd->vdev_top);
2811 * The state list is protected by the SCL_STATE lock. The caller
2812 * must either hold SCL_STATE as writer, or must be the sync thread
2813 * (which holds SCL_STATE as reader). There's only one sync thread,
2814 * so this is sufficient to ensure mutual exclusion.
2816 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2817 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2818 spa_config_held(spa, SCL_STATE, RW_READER)));
2820 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2821 list_insert_head(&spa->spa_state_dirty_list, vd);
2825 vdev_state_clean(vdev_t *vd)
2827 spa_t *spa = vd->vdev_spa;
2829 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2830 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2831 spa_config_held(spa, SCL_STATE, RW_READER)));
2833 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2834 list_remove(&spa->spa_state_dirty_list, vd);
2838 * Propagate vdev state up from children to parent.
2841 vdev_propagate_state(vdev_t *vd)
2843 spa_t *spa = vd->vdev_spa;
2844 vdev_t *rvd = spa->spa_root_vdev;
2845 int degraded = 0, faulted = 0;
2849 if (vd->vdev_children > 0) {
2850 for (int c = 0; c < vd->vdev_children; c++) {
2851 child = vd->vdev_child[c];
2854 * Don't factor holes into the decision.
2856 if (child->vdev_ishole)
2859 if (!vdev_readable(child) ||
2860 (!vdev_writeable(child) && spa_writeable(spa))) {
2862 * Root special: if there is a top-level log
2863 * device, treat the root vdev as if it were
2866 if (child->vdev_islog && vd == rvd)
2870 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2874 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2878 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2881 * Root special: if there is a top-level vdev that cannot be
2882 * opened due to corrupted metadata, then propagate the root
2883 * vdev's aux state as 'corrupt' rather than 'insufficient
2886 if (corrupted && vd == rvd &&
2887 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2888 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2889 VDEV_AUX_CORRUPT_DATA);
2892 if (vd->vdev_parent)
2893 vdev_propagate_state(vd->vdev_parent);
2897 * Set a vdev's state. If this is during an open, we don't update the parent
2898 * state, because we're in the process of opening children depth-first.
2899 * Otherwise, we propagate the change to the parent.
2901 * If this routine places a device in a faulted state, an appropriate ereport is
2905 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2907 uint64_t save_state;
2908 spa_t *spa = vd->vdev_spa;
2910 if (state == vd->vdev_state) {
2911 vd->vdev_stat.vs_aux = aux;
2915 save_state = vd->vdev_state;
2917 vd->vdev_state = state;
2918 vd->vdev_stat.vs_aux = aux;
2921 * If we are setting the vdev state to anything but an open state, then
2922 * always close the underlying device unless the device has requested
2923 * a delayed close (i.e. we're about to remove or fault the device).
2924 * Otherwise, we keep accessible but invalid devices open forever.
2925 * We don't call vdev_close() itself, because that implies some extra
2926 * checks (offline, etc) that we don't want here. This is limited to
2927 * leaf devices, because otherwise closing the device will affect other
2930 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2931 vd->vdev_ops->vdev_op_leaf)
2932 vd->vdev_ops->vdev_op_close(vd);
2935 * If we have brought this vdev back into service, we need
2936 * to notify fmd so that it can gracefully repair any outstanding
2937 * cases due to a missing device. We do this in all cases, even those
2938 * that probably don't correlate to a repaired fault. This is sure to
2939 * catch all cases, and we let the zfs-retire agent sort it out. If
2940 * this is a transient state it's OK, as the retire agent will
2941 * double-check the state of the vdev before repairing it.
2943 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2944 vd->vdev_prevstate != state)
2945 zfs_post_state_change(spa, vd);
2947 if (vd->vdev_removed &&
2948 state == VDEV_STATE_CANT_OPEN &&
2949 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2951 * If the previous state is set to VDEV_STATE_REMOVED, then this
2952 * device was previously marked removed and someone attempted to
2953 * reopen it. If this failed due to a nonexistent device, then
2954 * keep the device in the REMOVED state. We also let this be if
2955 * it is one of our special test online cases, which is only
2956 * attempting to online the device and shouldn't generate an FMA
2959 vd->vdev_state = VDEV_STATE_REMOVED;
2960 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2961 } else if (state == VDEV_STATE_REMOVED) {
2962 vd->vdev_removed = B_TRUE;
2963 } else if (state == VDEV_STATE_CANT_OPEN) {
2965 * If we fail to open a vdev during an import or recovery, we
2966 * mark it as "not available", which signifies that it was
2967 * never there to begin with. Failure to open such a device
2968 * is not considered an error.
2970 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2971 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2972 vd->vdev_ops->vdev_op_leaf)
2973 vd->vdev_not_present = 1;
2976 * Post the appropriate ereport. If the 'prevstate' field is
2977 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2978 * that this is part of a vdev_reopen(). In this case, we don't
2979 * want to post the ereport if the device was already in the
2980 * CANT_OPEN state beforehand.
2982 * If the 'checkremove' flag is set, then this is an attempt to
2983 * online the device in response to an insertion event. If we
2984 * hit this case, then we have detected an insertion event for a
2985 * faulted or offline device that wasn't in the removed state.
2986 * In this scenario, we don't post an ereport because we are
2987 * about to replace the device, or attempt an online with
2988 * vdev_forcefault, which will generate the fault for us.
2990 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
2991 !vd->vdev_not_present && !vd->vdev_checkremove &&
2992 vd != spa->spa_root_vdev) {
2996 case VDEV_AUX_OPEN_FAILED:
2997 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
2999 case VDEV_AUX_CORRUPT_DATA:
3000 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3002 case VDEV_AUX_NO_REPLICAS:
3003 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3005 case VDEV_AUX_BAD_GUID_SUM:
3006 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3008 case VDEV_AUX_TOO_SMALL:
3009 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3011 case VDEV_AUX_BAD_LABEL:
3012 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3015 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3018 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3021 /* Erase any notion of persistent removed state */
3022 vd->vdev_removed = B_FALSE;
3024 vd->vdev_removed = B_FALSE;
3027 if (!isopen && vd->vdev_parent)
3028 vdev_propagate_state(vd->vdev_parent);
3032 * Check the vdev configuration to ensure that it's capable of supporting
3035 * On Solaris, we do not support RAID-Z or partial configuration. In
3036 * addition, only a single top-level vdev is allowed and none of the
3037 * leaves can be wholedisks.
3039 * For FreeBSD, we can boot from any configuration. There is a
3040 * limitation that the boot filesystem must be either uncompressed or
3041 * compresses with lzjb compression but I'm not sure how to enforce
3045 vdev_is_bootable(vdev_t *vd)
3048 if (!vd->vdev_ops->vdev_op_leaf) {
3049 char *vdev_type = vd->vdev_ops->vdev_op_type;
3051 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3052 vd->vdev_children > 1) {
3054 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3055 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3058 } else if (vd->vdev_wholedisk == 1) {
3062 for (int c = 0; c < vd->vdev_children; c++) {
3063 if (!vdev_is_bootable(vd->vdev_child[c]))
3071 * Load the state from the original vdev tree (ovd) which
3072 * we've retrieved from the MOS config object. If the original
3073 * vdev was offline or faulted then we transfer that state to the
3074 * device in the current vdev tree (nvd).
3077 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3079 spa_t *spa = nvd->vdev_spa;
3081 ASSERT(nvd->vdev_top->vdev_islog);
3082 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3083 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3085 for (int c = 0; c < nvd->vdev_children; c++)
3086 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3088 if (nvd->vdev_ops->vdev_op_leaf) {
3090 * Restore the persistent vdev state
3092 nvd->vdev_offline = ovd->vdev_offline;
3093 nvd->vdev_faulted = ovd->vdev_faulted;
3094 nvd->vdev_degraded = ovd->vdev_degraded;
3095 nvd->vdev_removed = ovd->vdev_removed;
3100 * Determine if a log device has valid content. If the vdev was
3101 * removed or faulted in the MOS config then we know that
3102 * the content on the log device has already been written to the pool.
3105 vdev_log_state_valid(vdev_t *vd)
3107 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3111 for (int c = 0; c < vd->vdev_children; c++)
3112 if (vdev_log_state_valid(vd->vdev_child[c]))
3119 * Expand a vdev if possible.
3122 vdev_expand(vdev_t *vd, uint64_t txg)
3124 ASSERT(vd->vdev_top == vd);
3125 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3127 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3128 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3129 vdev_config_dirty(vd);
3137 vdev_split(vdev_t *vd)
3139 vdev_t *cvd, *pvd = vd->vdev_parent;
3141 vdev_remove_child(pvd, vd);
3142 vdev_compact_children(pvd);
3144 cvd = pvd->vdev_child[0];
3145 if (pvd->vdev_children == 1) {
3146 vdev_remove_parent(cvd);
3147 cvd->vdev_splitting = B_TRUE;
3149 vdev_propagate_state(cvd);