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 (c) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright 2016 Toomas Soome <tsoome@me.com>
29 * Copyright 2017 Joyent, Inc.
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/bpobj.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/uberblock_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/metaslab_impl.h>
44 #include <sys/space_map.h>
45 #include <sys/space_reftree.h>
48 #include <sys/fs/zfs.h>
51 #include <sys/dsl_scan.h>
53 #include <sys/trim_map.h>
55 SYSCTL_DECL(_vfs_zfs);
56 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
59 * Virtual device management.
63 * The limit for ZFS to automatically increase a top-level vdev's ashift
64 * from logical ashift to physical ashift.
66 * Example: one or more 512B emulation child vdevs
67 * child->vdev_ashift = 9 (512 bytes)
68 * child->vdev_physical_ashift = 12 (4096 bytes)
69 * zfs_max_auto_ashift = 11 (2048 bytes)
70 * zfs_min_auto_ashift = 9 (512 bytes)
72 * On pool creation or the addition of a new top-level vdev, ZFS will
73 * increase the ashift of the top-level vdev to 2048 as limited by
74 * zfs_max_auto_ashift.
76 * Example: one or more 512B emulation child vdevs
77 * child->vdev_ashift = 9 (512 bytes)
78 * child->vdev_physical_ashift = 12 (4096 bytes)
79 * zfs_max_auto_ashift = 13 (8192 bytes)
80 * zfs_min_auto_ashift = 9 (512 bytes)
82 * On pool creation or the addition of a new top-level vdev, ZFS will
83 * increase the ashift of the top-level vdev to 4096 to match the
84 * max vdev_physical_ashift.
86 * Example: one or more 512B emulation child vdevs
87 * child->vdev_ashift = 9 (512 bytes)
88 * child->vdev_physical_ashift = 9 (512 bytes)
89 * zfs_max_auto_ashift = 13 (8192 bytes)
90 * zfs_min_auto_ashift = 12 (4096 bytes)
92 * On pool creation or the addition of a new top-level vdev, ZFS will
93 * increase the ashift of the top-level vdev to 4096 to match the
94 * zfs_min_auto_ashift.
96 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
97 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
100 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
105 val = zfs_max_auto_ashift;
106 err = sysctl_handle_64(oidp, &val, 0, req);
107 if (err != 0 || req->newptr == NULL)
110 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
113 zfs_max_auto_ashift = val;
117 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
118 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
119 sysctl_vfs_zfs_max_auto_ashift, "QU",
120 "Max ashift used when optimising for logical -> physical sectors size on "
121 "new top-level vdevs.");
124 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
129 val = zfs_min_auto_ashift;
130 err = sysctl_handle_64(oidp, &val, 0, req);
131 if (err != 0 || req->newptr == NULL)
134 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
137 zfs_min_auto_ashift = val;
141 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
142 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
143 sysctl_vfs_zfs_min_auto_ashift, "QU",
144 "Min ashift used when creating new top-level vdevs.");
146 static vdev_ops_t *vdev_ops_table[] = {
166 * When a vdev is added, it will be divided into approximately (but no
167 * more than) this number of metaslabs.
169 int metaslabs_per_vdev = 200;
170 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
171 &metaslabs_per_vdev, 0,
172 "When a vdev is added, how many metaslabs the vdev should be divided into");
175 * Given a vdev type, return the appropriate ops vector.
178 vdev_getops(const char *type)
180 vdev_ops_t *ops, **opspp;
182 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
183 if (strcmp(ops->vdev_op_type, type) == 0)
190 * Default asize function: return the MAX of psize with the asize of
191 * all children. This is what's used by anything other than RAID-Z.
194 vdev_default_asize(vdev_t *vd, uint64_t psize)
196 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
199 for (int c = 0; c < vd->vdev_children; c++) {
200 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
201 asize = MAX(asize, csize);
208 * Get the minimum allocatable size. We define the allocatable size as
209 * the vdev's asize rounded to the nearest metaslab. This allows us to
210 * replace or attach devices which don't have the same physical size but
211 * can still satisfy the same number of allocations.
214 vdev_get_min_asize(vdev_t *vd)
216 vdev_t *pvd = vd->vdev_parent;
219 * If our parent is NULL (inactive spare or cache) or is the root,
220 * just return our own asize.
223 return (vd->vdev_asize);
226 * The top-level vdev just returns the allocatable size rounded
227 * to the nearest metaslab.
229 if (vd == vd->vdev_top)
230 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
233 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
234 * so each child must provide at least 1/Nth of its asize.
236 if (pvd->vdev_ops == &vdev_raidz_ops)
237 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
240 return (pvd->vdev_min_asize);
244 vdev_set_min_asize(vdev_t *vd)
246 vd->vdev_min_asize = vdev_get_min_asize(vd);
248 for (int c = 0; c < vd->vdev_children; c++)
249 vdev_set_min_asize(vd->vdev_child[c]);
253 vdev_lookup_top(spa_t *spa, uint64_t vdev)
255 vdev_t *rvd = spa->spa_root_vdev;
257 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
259 if (vdev < rvd->vdev_children) {
260 ASSERT(rvd->vdev_child[vdev] != NULL);
261 return (rvd->vdev_child[vdev]);
268 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
272 if (vd->vdev_guid == guid)
275 for (int c = 0; c < vd->vdev_children; c++)
276 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
284 vdev_count_leaves_impl(vdev_t *vd)
288 if (vd->vdev_ops->vdev_op_leaf)
291 for (int c = 0; c < vd->vdev_children; c++)
292 n += vdev_count_leaves_impl(vd->vdev_child[c]);
298 vdev_count_leaves(spa_t *spa)
300 return (vdev_count_leaves_impl(spa->spa_root_vdev));
304 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
306 size_t oldsize, newsize;
307 uint64_t id = cvd->vdev_id;
309 spa_t *spa = cvd->vdev_spa;
311 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
312 ASSERT(cvd->vdev_parent == NULL);
314 cvd->vdev_parent = pvd;
319 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
321 oldsize = pvd->vdev_children * sizeof (vdev_t *);
322 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
323 newsize = pvd->vdev_children * sizeof (vdev_t *);
325 newchild = kmem_zalloc(newsize, KM_SLEEP);
326 if (pvd->vdev_child != NULL) {
327 bcopy(pvd->vdev_child, newchild, oldsize);
328 kmem_free(pvd->vdev_child, oldsize);
331 pvd->vdev_child = newchild;
332 pvd->vdev_child[id] = cvd;
334 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
335 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
338 * Walk up all ancestors to update guid sum.
340 for (; pvd != NULL; pvd = pvd->vdev_parent)
341 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
345 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
348 uint_t id = cvd->vdev_id;
350 ASSERT(cvd->vdev_parent == pvd);
355 ASSERT(id < pvd->vdev_children);
356 ASSERT(pvd->vdev_child[id] == cvd);
358 pvd->vdev_child[id] = NULL;
359 cvd->vdev_parent = NULL;
361 for (c = 0; c < pvd->vdev_children; c++)
362 if (pvd->vdev_child[c])
365 if (c == pvd->vdev_children) {
366 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
367 pvd->vdev_child = NULL;
368 pvd->vdev_children = 0;
372 * Walk up all ancestors to update guid sum.
374 for (; pvd != NULL; pvd = pvd->vdev_parent)
375 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
379 * Remove any holes in the child array.
382 vdev_compact_children(vdev_t *pvd)
384 vdev_t **newchild, *cvd;
385 int oldc = pvd->vdev_children;
388 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
390 for (int c = newc = 0; c < oldc; c++)
391 if (pvd->vdev_child[c])
394 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
396 for (int c = newc = 0; c < oldc; c++) {
397 if ((cvd = pvd->vdev_child[c]) != NULL) {
398 newchild[newc] = cvd;
399 cvd->vdev_id = newc++;
403 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
404 pvd->vdev_child = newchild;
405 pvd->vdev_children = newc;
409 * Allocate and minimally initialize a vdev_t.
412 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
415 vdev_indirect_config_t *vic;
417 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
418 vic = &vd->vdev_indirect_config;
420 if (spa->spa_root_vdev == NULL) {
421 ASSERT(ops == &vdev_root_ops);
422 spa->spa_root_vdev = vd;
423 spa->spa_load_guid = spa_generate_guid(NULL);
426 if (guid == 0 && ops != &vdev_hole_ops) {
427 if (spa->spa_root_vdev == vd) {
429 * The root vdev's guid will also be the pool guid,
430 * which must be unique among all pools.
432 guid = spa_generate_guid(NULL);
435 * Any other vdev's guid must be unique within the pool.
437 guid = spa_generate_guid(spa);
439 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
444 vd->vdev_guid = guid;
445 vd->vdev_guid_sum = guid;
447 vd->vdev_state = VDEV_STATE_CLOSED;
448 vd->vdev_ishole = (ops == &vdev_hole_ops);
449 vic->vic_prev_indirect_vdev = UINT64_MAX;
451 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
452 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
453 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
455 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
456 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
457 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
458 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
459 for (int t = 0; t < DTL_TYPES; t++) {
460 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
462 txg_list_create(&vd->vdev_ms_list, spa,
463 offsetof(struct metaslab, ms_txg_node));
464 txg_list_create(&vd->vdev_dtl_list, spa,
465 offsetof(struct vdev, vdev_dtl_node));
466 vd->vdev_stat.vs_timestamp = gethrtime();
474 * Allocate a new vdev. The 'alloctype' is used to control whether we are
475 * creating a new vdev or loading an existing one - the behavior is slightly
476 * different for each case.
479 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
484 uint64_t guid = 0, islog, nparity;
486 vdev_indirect_config_t *vic;
488 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
490 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
491 return (SET_ERROR(EINVAL));
493 if ((ops = vdev_getops(type)) == NULL)
494 return (SET_ERROR(EINVAL));
497 * If this is a load, get the vdev guid from the nvlist.
498 * Otherwise, vdev_alloc_common() will generate one for us.
500 if (alloctype == VDEV_ALLOC_LOAD) {
503 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
505 return (SET_ERROR(EINVAL));
507 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
508 return (SET_ERROR(EINVAL));
509 } else if (alloctype == VDEV_ALLOC_SPARE) {
510 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
511 return (SET_ERROR(EINVAL));
512 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
513 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
514 return (SET_ERROR(EINVAL));
515 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
516 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
517 return (SET_ERROR(EINVAL));
521 * The first allocated vdev must be of type 'root'.
523 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
524 return (SET_ERROR(EINVAL));
527 * Determine whether we're a log vdev.
530 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
531 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
532 return (SET_ERROR(ENOTSUP));
534 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
535 return (SET_ERROR(ENOTSUP));
538 * Set the nparity property for RAID-Z vdevs.
541 if (ops == &vdev_raidz_ops) {
542 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
544 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
545 return (SET_ERROR(EINVAL));
547 * Previous versions could only support 1 or 2 parity
551 spa_version(spa) < SPA_VERSION_RAIDZ2)
552 return (SET_ERROR(ENOTSUP));
554 spa_version(spa) < SPA_VERSION_RAIDZ3)
555 return (SET_ERROR(ENOTSUP));
558 * We require the parity to be specified for SPAs that
559 * support multiple parity levels.
561 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
562 return (SET_ERROR(EINVAL));
564 * Otherwise, we default to 1 parity device for RAID-Z.
571 ASSERT(nparity != -1ULL);
573 vd = vdev_alloc_common(spa, id, guid, ops);
574 vic = &vd->vdev_indirect_config;
576 vd->vdev_islog = islog;
577 vd->vdev_nparity = nparity;
579 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
580 vd->vdev_path = spa_strdup(vd->vdev_path);
581 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
582 vd->vdev_devid = spa_strdup(vd->vdev_devid);
583 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
584 &vd->vdev_physpath) == 0)
585 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
586 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
587 vd->vdev_fru = spa_strdup(vd->vdev_fru);
590 * Set the whole_disk property. If it's not specified, leave the value
593 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
594 &vd->vdev_wholedisk) != 0)
595 vd->vdev_wholedisk = -1ULL;
597 ASSERT0(vic->vic_mapping_object);
598 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
599 &vic->vic_mapping_object);
600 ASSERT0(vic->vic_births_object);
601 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
602 &vic->vic_births_object);
603 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
604 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
605 &vic->vic_prev_indirect_vdev);
608 * Look for the 'not present' flag. This will only be set if the device
609 * was not present at the time of import.
611 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
612 &vd->vdev_not_present);
615 * Get the alignment requirement.
617 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
620 * Retrieve the vdev creation time.
622 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
626 * If we're a top-level vdev, try to load the allocation parameters.
628 if (parent && !parent->vdev_parent &&
629 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
630 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
632 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
634 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
636 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
638 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
641 ASSERT0(vd->vdev_top_zap);
644 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
645 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
646 alloctype == VDEV_ALLOC_ADD ||
647 alloctype == VDEV_ALLOC_SPLIT ||
648 alloctype == VDEV_ALLOC_ROOTPOOL);
649 vd->vdev_mg = metaslab_group_create(islog ?
650 spa_log_class(spa) : spa_normal_class(spa), vd);
653 if (vd->vdev_ops->vdev_op_leaf &&
654 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
655 (void) nvlist_lookup_uint64(nv,
656 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
658 ASSERT0(vd->vdev_leaf_zap);
662 * If we're a leaf vdev, try to load the DTL object and other state.
665 if (vd->vdev_ops->vdev_op_leaf &&
666 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
667 alloctype == VDEV_ALLOC_ROOTPOOL)) {
668 if (alloctype == VDEV_ALLOC_LOAD) {
669 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
670 &vd->vdev_dtl_object);
671 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
675 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
678 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
679 &spare) == 0 && spare)
683 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
686 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
687 &vd->vdev_resilver_txg);
690 * When importing a pool, we want to ignore the persistent fault
691 * state, as the diagnosis made on another system may not be
692 * valid in the current context. Local vdevs will
693 * remain in the faulted state.
695 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
696 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
698 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
700 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
703 if (vd->vdev_faulted || vd->vdev_degraded) {
707 VDEV_AUX_ERR_EXCEEDED;
708 if (nvlist_lookup_string(nv,
709 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
710 strcmp(aux, "external") == 0)
711 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
717 * Add ourselves to the parent's list of children.
719 vdev_add_child(parent, vd);
727 vdev_free(vdev_t *vd)
729 spa_t *spa = vd->vdev_spa;
732 * vdev_free() implies closing the vdev first. This is simpler than
733 * trying to ensure complicated semantics for all callers.
737 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
738 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
743 for (int c = 0; c < vd->vdev_children; c++)
744 vdev_free(vd->vdev_child[c]);
746 ASSERT(vd->vdev_child == NULL);
747 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
750 * Discard allocation state.
752 if (vd->vdev_mg != NULL) {
753 vdev_metaslab_fini(vd);
754 metaslab_group_destroy(vd->vdev_mg);
757 ASSERT0(vd->vdev_stat.vs_space);
758 ASSERT0(vd->vdev_stat.vs_dspace);
759 ASSERT0(vd->vdev_stat.vs_alloc);
762 * Remove this vdev from its parent's child list.
764 vdev_remove_child(vd->vdev_parent, vd);
766 ASSERT(vd->vdev_parent == NULL);
769 * Clean up vdev structure.
775 spa_strfree(vd->vdev_path);
777 spa_strfree(vd->vdev_devid);
778 if (vd->vdev_physpath)
779 spa_strfree(vd->vdev_physpath);
781 spa_strfree(vd->vdev_fru);
783 if (vd->vdev_isspare)
784 spa_spare_remove(vd);
785 if (vd->vdev_isl2cache)
786 spa_l2cache_remove(vd);
788 txg_list_destroy(&vd->vdev_ms_list);
789 txg_list_destroy(&vd->vdev_dtl_list);
791 mutex_enter(&vd->vdev_dtl_lock);
792 space_map_close(vd->vdev_dtl_sm);
793 for (int t = 0; t < DTL_TYPES; t++) {
794 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
795 range_tree_destroy(vd->vdev_dtl[t]);
797 mutex_exit(&vd->vdev_dtl_lock);
799 EQUIV(vd->vdev_indirect_births != NULL,
800 vd->vdev_indirect_mapping != NULL);
801 if (vd->vdev_indirect_births != NULL) {
802 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
803 vdev_indirect_births_close(vd->vdev_indirect_births);
806 if (vd->vdev_obsolete_sm != NULL) {
807 ASSERT(vd->vdev_removing ||
808 vd->vdev_ops == &vdev_indirect_ops);
809 space_map_close(vd->vdev_obsolete_sm);
810 vd->vdev_obsolete_sm = NULL;
812 range_tree_destroy(vd->vdev_obsolete_segments);
813 rw_destroy(&vd->vdev_indirect_rwlock);
814 mutex_destroy(&vd->vdev_obsolete_lock);
816 mutex_destroy(&vd->vdev_queue_lock);
817 mutex_destroy(&vd->vdev_dtl_lock);
818 mutex_destroy(&vd->vdev_stat_lock);
819 mutex_destroy(&vd->vdev_probe_lock);
821 if (vd == spa->spa_root_vdev)
822 spa->spa_root_vdev = NULL;
824 kmem_free(vd, sizeof (vdev_t));
828 * Transfer top-level vdev state from svd to tvd.
831 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
833 spa_t *spa = svd->vdev_spa;
838 ASSERT(tvd == tvd->vdev_top);
840 tvd->vdev_ms_array = svd->vdev_ms_array;
841 tvd->vdev_ms_shift = svd->vdev_ms_shift;
842 tvd->vdev_ms_count = svd->vdev_ms_count;
843 tvd->vdev_top_zap = svd->vdev_top_zap;
845 svd->vdev_ms_array = 0;
846 svd->vdev_ms_shift = 0;
847 svd->vdev_ms_count = 0;
848 svd->vdev_top_zap = 0;
851 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
852 tvd->vdev_mg = svd->vdev_mg;
853 tvd->vdev_ms = svd->vdev_ms;
858 if (tvd->vdev_mg != NULL)
859 tvd->vdev_mg->mg_vd = tvd;
861 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
862 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
863 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
865 svd->vdev_stat.vs_alloc = 0;
866 svd->vdev_stat.vs_space = 0;
867 svd->vdev_stat.vs_dspace = 0;
869 for (t = 0; t < TXG_SIZE; t++) {
870 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
871 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
872 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
873 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
874 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
875 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
878 if (list_link_active(&svd->vdev_config_dirty_node)) {
879 vdev_config_clean(svd);
880 vdev_config_dirty(tvd);
883 if (list_link_active(&svd->vdev_state_dirty_node)) {
884 vdev_state_clean(svd);
885 vdev_state_dirty(tvd);
888 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
889 svd->vdev_deflate_ratio = 0;
891 tvd->vdev_islog = svd->vdev_islog;
896 vdev_top_update(vdev_t *tvd, vdev_t *vd)
903 for (int c = 0; c < vd->vdev_children; c++)
904 vdev_top_update(tvd, vd->vdev_child[c]);
908 * Add a mirror/replacing vdev above an existing vdev.
911 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
913 spa_t *spa = cvd->vdev_spa;
914 vdev_t *pvd = cvd->vdev_parent;
917 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
919 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
921 mvd->vdev_asize = cvd->vdev_asize;
922 mvd->vdev_min_asize = cvd->vdev_min_asize;
923 mvd->vdev_max_asize = cvd->vdev_max_asize;
924 mvd->vdev_psize = cvd->vdev_psize;
925 mvd->vdev_ashift = cvd->vdev_ashift;
926 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
927 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
928 mvd->vdev_state = cvd->vdev_state;
929 mvd->vdev_crtxg = cvd->vdev_crtxg;
931 vdev_remove_child(pvd, cvd);
932 vdev_add_child(pvd, mvd);
933 cvd->vdev_id = mvd->vdev_children;
934 vdev_add_child(mvd, cvd);
935 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
937 if (mvd == mvd->vdev_top)
938 vdev_top_transfer(cvd, mvd);
944 * Remove a 1-way mirror/replacing vdev from the tree.
947 vdev_remove_parent(vdev_t *cvd)
949 vdev_t *mvd = cvd->vdev_parent;
950 vdev_t *pvd = mvd->vdev_parent;
952 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
954 ASSERT(mvd->vdev_children == 1);
955 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
956 mvd->vdev_ops == &vdev_replacing_ops ||
957 mvd->vdev_ops == &vdev_spare_ops);
958 cvd->vdev_ashift = mvd->vdev_ashift;
959 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
960 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
962 vdev_remove_child(mvd, cvd);
963 vdev_remove_child(pvd, mvd);
966 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
967 * Otherwise, we could have detached an offline device, and when we
968 * go to import the pool we'll think we have two top-level vdevs,
969 * instead of a different version of the same top-level vdev.
971 if (mvd->vdev_top == mvd) {
972 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
973 cvd->vdev_orig_guid = cvd->vdev_guid;
974 cvd->vdev_guid += guid_delta;
975 cvd->vdev_guid_sum += guid_delta;
977 cvd->vdev_id = mvd->vdev_id;
978 vdev_add_child(pvd, cvd);
979 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
981 if (cvd == cvd->vdev_top)
982 vdev_top_transfer(mvd, cvd);
984 ASSERT(mvd->vdev_children == 0);
989 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
991 spa_t *spa = vd->vdev_spa;
992 objset_t *mos = spa->spa_meta_objset;
994 uint64_t oldc = vd->vdev_ms_count;
995 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
999 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1002 * This vdev is not being allocated from yet or is a hole.
1004 if (vd->vdev_ms_shift == 0)
1007 ASSERT(!vd->vdev_ishole);
1009 ASSERT(oldc <= newc);
1011 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1014 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1015 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1019 vd->vdev_ms_count = newc;
1021 for (m = oldc; m < newc; m++) {
1022 uint64_t object = 0;
1025 * vdev_ms_array may be 0 if we are creating the "fake"
1026 * metaslabs for an indirect vdev for zdb's leak detection.
1027 * See zdb_leak_init().
1029 if (txg == 0 && vd->vdev_ms_array != 0) {
1030 error = dmu_read(mos, vd->vdev_ms_array,
1031 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1037 error = metaslab_init(vd->vdev_mg, m, object, txg,
1044 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1047 * If the vdev is being removed we don't activate
1048 * the metaslabs since we want to ensure that no new
1049 * allocations are performed on this device.
1051 if (oldc == 0 && !vd->vdev_removing)
1052 metaslab_group_activate(vd->vdev_mg);
1055 spa_config_exit(spa, SCL_ALLOC, FTAG);
1061 vdev_metaslab_fini(vdev_t *vd)
1063 if (vd->vdev_ms != NULL) {
1064 uint64_t count = vd->vdev_ms_count;
1066 metaslab_group_passivate(vd->vdev_mg);
1067 for (uint64_t m = 0; m < count; m++) {
1068 metaslab_t *msp = vd->vdev_ms[m];
1073 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1076 vd->vdev_ms_count = 0;
1078 ASSERT0(vd->vdev_ms_count);
1081 typedef struct vdev_probe_stats {
1082 boolean_t vps_readable;
1083 boolean_t vps_writeable;
1085 } vdev_probe_stats_t;
1088 vdev_probe_done(zio_t *zio)
1090 spa_t *spa = zio->io_spa;
1091 vdev_t *vd = zio->io_vd;
1092 vdev_probe_stats_t *vps = zio->io_private;
1094 ASSERT(vd->vdev_probe_zio != NULL);
1096 if (zio->io_type == ZIO_TYPE_READ) {
1097 if (zio->io_error == 0)
1098 vps->vps_readable = 1;
1099 if (zio->io_error == 0 && spa_writeable(spa)) {
1100 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1101 zio->io_offset, zio->io_size, zio->io_abd,
1102 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1103 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1105 abd_free(zio->io_abd);
1107 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1108 if (zio->io_error == 0)
1109 vps->vps_writeable = 1;
1110 abd_free(zio->io_abd);
1111 } else if (zio->io_type == ZIO_TYPE_NULL) {
1114 vd->vdev_cant_read |= !vps->vps_readable;
1115 vd->vdev_cant_write |= !vps->vps_writeable;
1117 if (vdev_readable(vd) &&
1118 (vdev_writeable(vd) || !spa_writeable(spa))) {
1121 ASSERT(zio->io_error != 0);
1122 zfs_dbgmsg("failed probe on vdev %llu",
1123 (longlong_t)vd->vdev_id);
1124 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1125 spa, vd, NULL, 0, 0);
1126 zio->io_error = SET_ERROR(ENXIO);
1129 mutex_enter(&vd->vdev_probe_lock);
1130 ASSERT(vd->vdev_probe_zio == zio);
1131 vd->vdev_probe_zio = NULL;
1132 mutex_exit(&vd->vdev_probe_lock);
1134 zio_link_t *zl = NULL;
1135 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1136 if (!vdev_accessible(vd, pio))
1137 pio->io_error = SET_ERROR(ENXIO);
1139 kmem_free(vps, sizeof (*vps));
1144 * Determine whether this device is accessible.
1146 * Read and write to several known locations: the pad regions of each
1147 * vdev label but the first, which we leave alone in case it contains
1151 vdev_probe(vdev_t *vd, zio_t *zio)
1153 spa_t *spa = vd->vdev_spa;
1154 vdev_probe_stats_t *vps = NULL;
1157 ASSERT(vd->vdev_ops->vdev_op_leaf);
1160 * Don't probe the probe.
1162 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1166 * To prevent 'probe storms' when a device fails, we create
1167 * just one probe i/o at a time. All zios that want to probe
1168 * this vdev will become parents of the probe io.
1170 mutex_enter(&vd->vdev_probe_lock);
1172 if ((pio = vd->vdev_probe_zio) == NULL) {
1173 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1175 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1176 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1179 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1181 * vdev_cant_read and vdev_cant_write can only
1182 * transition from TRUE to FALSE when we have the
1183 * SCL_ZIO lock as writer; otherwise they can only
1184 * transition from FALSE to TRUE. This ensures that
1185 * any zio looking at these values can assume that
1186 * failures persist for the life of the I/O. That's
1187 * important because when a device has intermittent
1188 * connectivity problems, we want to ensure that
1189 * they're ascribed to the device (ENXIO) and not
1192 * Since we hold SCL_ZIO as writer here, clear both
1193 * values so the probe can reevaluate from first
1196 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1197 vd->vdev_cant_read = B_FALSE;
1198 vd->vdev_cant_write = B_FALSE;
1201 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1202 vdev_probe_done, vps,
1203 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1206 * We can't change the vdev state in this context, so we
1207 * kick off an async task to do it on our behalf.
1210 vd->vdev_probe_wanted = B_TRUE;
1211 spa_async_request(spa, SPA_ASYNC_PROBE);
1216 zio_add_child(zio, pio);
1218 mutex_exit(&vd->vdev_probe_lock);
1221 ASSERT(zio != NULL);
1225 for (int l = 1; l < VDEV_LABELS; l++) {
1226 zio_nowait(zio_read_phys(pio, vd,
1227 vdev_label_offset(vd->vdev_psize, l,
1228 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1229 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1230 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1231 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1242 vdev_open_child(void *arg)
1246 vd->vdev_open_thread = curthread;
1247 vd->vdev_open_error = vdev_open(vd);
1248 vd->vdev_open_thread = NULL;
1252 vdev_uses_zvols(vdev_t *vd)
1254 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1255 strlen(ZVOL_DIR)) == 0)
1257 for (int c = 0; c < vd->vdev_children; c++)
1258 if (vdev_uses_zvols(vd->vdev_child[c]))
1264 vdev_open_children(vdev_t *vd)
1267 int children = vd->vdev_children;
1270 * in order to handle pools on top of zvols, do the opens
1271 * in a single thread so that the same thread holds the
1272 * spa_namespace_lock
1274 if (B_TRUE || vdev_uses_zvols(vd)) {
1275 for (int c = 0; c < children; c++)
1276 vd->vdev_child[c]->vdev_open_error =
1277 vdev_open(vd->vdev_child[c]);
1280 tq = taskq_create("vdev_open", children, minclsyspri,
1281 children, children, TASKQ_PREPOPULATE);
1283 for (int c = 0; c < children; c++)
1284 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1291 * Compute the raidz-deflation ratio. Note, we hard-code
1292 * in 128k (1 << 17) because it is the "typical" blocksize.
1293 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1294 * otherwise it would inconsistently account for existing bp's.
1297 vdev_set_deflate_ratio(vdev_t *vd)
1299 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1300 vd->vdev_deflate_ratio = (1 << 17) /
1301 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1306 * Prepare a virtual device for access.
1309 vdev_open(vdev_t *vd)
1311 spa_t *spa = vd->vdev_spa;
1314 uint64_t max_osize = 0;
1315 uint64_t asize, max_asize, psize;
1316 uint64_t logical_ashift = 0;
1317 uint64_t physical_ashift = 0;
1319 ASSERT(vd->vdev_open_thread == curthread ||
1320 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1321 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1322 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1323 vd->vdev_state == VDEV_STATE_OFFLINE);
1325 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1326 vd->vdev_cant_read = B_FALSE;
1327 vd->vdev_cant_write = B_FALSE;
1328 vd->vdev_notrim = B_FALSE;
1329 vd->vdev_min_asize = vdev_get_min_asize(vd);
1332 * If this vdev is not removed, check its fault status. If it's
1333 * faulted, bail out of the open.
1335 if (!vd->vdev_removed && vd->vdev_faulted) {
1336 ASSERT(vd->vdev_children == 0);
1337 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1338 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1339 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1340 vd->vdev_label_aux);
1341 return (SET_ERROR(ENXIO));
1342 } else if (vd->vdev_offline) {
1343 ASSERT(vd->vdev_children == 0);
1344 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1345 return (SET_ERROR(ENXIO));
1348 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1349 &logical_ashift, &physical_ashift);
1352 * Reset the vdev_reopening flag so that we actually close
1353 * the vdev on error.
1355 vd->vdev_reopening = B_FALSE;
1356 if (zio_injection_enabled && error == 0)
1357 error = zio_handle_device_injection(vd, NULL, ENXIO);
1360 if (vd->vdev_removed &&
1361 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1362 vd->vdev_removed = B_FALSE;
1364 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1365 vd->vdev_stat.vs_aux);
1369 vd->vdev_removed = B_FALSE;
1372 * Recheck the faulted flag now that we have confirmed that
1373 * the vdev is accessible. If we're faulted, bail.
1375 if (vd->vdev_faulted) {
1376 ASSERT(vd->vdev_children == 0);
1377 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1378 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1379 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1380 vd->vdev_label_aux);
1381 return (SET_ERROR(ENXIO));
1384 if (vd->vdev_degraded) {
1385 ASSERT(vd->vdev_children == 0);
1386 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1387 VDEV_AUX_ERR_EXCEEDED);
1389 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1393 * For hole or missing vdevs we just return success.
1395 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1398 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1399 trim_map_create(vd);
1401 for (int c = 0; c < vd->vdev_children; c++) {
1402 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1403 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1409 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1410 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1412 if (vd->vdev_children == 0) {
1413 if (osize < SPA_MINDEVSIZE) {
1414 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1415 VDEV_AUX_TOO_SMALL);
1416 return (SET_ERROR(EOVERFLOW));
1419 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1420 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1421 VDEV_LABEL_END_SIZE);
1423 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1424 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1425 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1426 VDEV_AUX_TOO_SMALL);
1427 return (SET_ERROR(EOVERFLOW));
1431 max_asize = max_osize;
1434 vd->vdev_psize = psize;
1437 * Make sure the allocatable size hasn't shrunk too much.
1439 if (asize < vd->vdev_min_asize) {
1440 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1441 VDEV_AUX_BAD_LABEL);
1442 return (SET_ERROR(EINVAL));
1445 vd->vdev_physical_ashift =
1446 MAX(physical_ashift, vd->vdev_physical_ashift);
1447 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1448 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1450 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1451 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1452 VDEV_AUX_ASHIFT_TOO_BIG);
1456 if (vd->vdev_asize == 0) {
1458 * This is the first-ever open, so use the computed values.
1459 * For testing purposes, a higher ashift can be requested.
1461 vd->vdev_asize = asize;
1462 vd->vdev_max_asize = max_asize;
1465 * Make sure the alignment requirement hasn't increased.
1467 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1468 vd->vdev_ops->vdev_op_leaf) {
1469 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1470 VDEV_AUX_BAD_LABEL);
1473 vd->vdev_max_asize = max_asize;
1477 * If all children are healthy we update asize if either:
1478 * The asize has increased, due to a device expansion caused by dynamic
1479 * LUN growth or vdev replacement, and automatic expansion is enabled;
1480 * making the additional space available.
1482 * The asize has decreased, due to a device shrink usually caused by a
1483 * vdev replace with a smaller device. This ensures that calculations
1484 * based of max_asize and asize e.g. esize are always valid. It's safe
1485 * to do this as we've already validated that asize is greater than
1488 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1489 ((asize > vd->vdev_asize &&
1490 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1491 (asize < vd->vdev_asize)))
1492 vd->vdev_asize = asize;
1494 vdev_set_min_asize(vd);
1497 * Ensure we can issue some IO before declaring the
1498 * vdev open for business.
1500 if (vd->vdev_ops->vdev_op_leaf &&
1501 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1502 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1503 VDEV_AUX_ERR_EXCEEDED);
1507 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1508 !vd->vdev_isl2cache && !vd->vdev_islog) {
1509 if (vd->vdev_ashift > spa->spa_max_ashift)
1510 spa->spa_max_ashift = vd->vdev_ashift;
1511 if (vd->vdev_ashift < spa->spa_min_ashift)
1512 spa->spa_min_ashift = vd->vdev_ashift;
1516 * Track the min and max ashift values for normal data devices.
1518 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1519 !vd->vdev_islog && vd->vdev_aux == NULL) {
1520 if (vd->vdev_ashift > spa->spa_max_ashift)
1521 spa->spa_max_ashift = vd->vdev_ashift;
1522 if (vd->vdev_ashift < spa->spa_min_ashift)
1523 spa->spa_min_ashift = vd->vdev_ashift;
1527 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1528 * resilver. But don't do this if we are doing a reopen for a scrub,
1529 * since this would just restart the scrub we are already doing.
1531 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1532 vdev_resilver_needed(vd, NULL, NULL))
1533 spa_async_request(spa, SPA_ASYNC_RESILVER);
1539 * Called once the vdevs are all opened, this routine validates the label
1540 * contents. This needs to be done before vdev_load() so that we don't
1541 * inadvertently do repair I/Os to the wrong device.
1543 * If 'strict' is false ignore the spa guid check. This is necessary because
1544 * if the machine crashed during a re-guid the new guid might have been written
1545 * to all of the vdev labels, but not the cached config. The strict check
1546 * will be performed when the pool is opened again using the mos config.
1548 * This function will only return failure if one of the vdevs indicates that it
1549 * has since been destroyed or exported. This is only possible if
1550 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1551 * will be updated but the function will return 0.
1554 vdev_validate(vdev_t *vd, boolean_t strict)
1556 spa_t *spa = vd->vdev_spa;
1558 uint64_t guid = 0, top_guid;
1561 for (int c = 0; c < vd->vdev_children; c++)
1562 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1563 return (SET_ERROR(EBADF));
1566 * If the device has already failed, or was marked offline, don't do
1567 * any further validation. Otherwise, label I/O will fail and we will
1568 * overwrite the previous state.
1570 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1571 uint64_t aux_guid = 0;
1573 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1574 spa_last_synced_txg(spa) : -1ULL;
1576 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1577 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1578 VDEV_AUX_BAD_LABEL);
1583 * Determine if this vdev has been split off into another
1584 * pool. If so, then refuse to open it.
1586 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1587 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1588 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1589 VDEV_AUX_SPLIT_POOL);
1594 if (strict && (nvlist_lookup_uint64(label,
1595 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1596 guid != spa_guid(spa))) {
1597 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1598 VDEV_AUX_CORRUPT_DATA);
1603 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1604 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1609 * If this vdev just became a top-level vdev because its
1610 * sibling was detached, it will have adopted the parent's
1611 * vdev guid -- but the label may or may not be on disk yet.
1612 * Fortunately, either version of the label will have the
1613 * same top guid, so if we're a top-level vdev, we can
1614 * safely compare to that instead.
1616 * If we split this vdev off instead, then we also check the
1617 * original pool's guid. We don't want to consider the vdev
1618 * corrupt if it is partway through a split operation.
1620 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1622 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1624 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1625 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1626 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1627 VDEV_AUX_CORRUPT_DATA);
1632 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1634 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1635 VDEV_AUX_CORRUPT_DATA);
1643 * If this is a verbatim import, no need to check the
1644 * state of the pool.
1646 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1647 spa_load_state(spa) == SPA_LOAD_OPEN &&
1648 state != POOL_STATE_ACTIVE)
1649 return (SET_ERROR(EBADF));
1652 * If we were able to open and validate a vdev that was
1653 * previously marked permanently unavailable, clear that state
1656 if (vd->vdev_not_present)
1657 vd->vdev_not_present = 0;
1664 * Close a virtual device.
1667 vdev_close(vdev_t *vd)
1669 spa_t *spa = vd->vdev_spa;
1670 vdev_t *pvd = vd->vdev_parent;
1672 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1675 * If our parent is reopening, then we are as well, unless we are
1678 if (pvd != NULL && pvd->vdev_reopening)
1679 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1681 vd->vdev_ops->vdev_op_close(vd);
1683 vdev_cache_purge(vd);
1685 if (vd->vdev_ops->vdev_op_leaf)
1686 trim_map_destroy(vd);
1689 * We record the previous state before we close it, so that if we are
1690 * doing a reopen(), we don't generate FMA ereports if we notice that
1691 * it's still faulted.
1693 vd->vdev_prevstate = vd->vdev_state;
1695 if (vd->vdev_offline)
1696 vd->vdev_state = VDEV_STATE_OFFLINE;
1698 vd->vdev_state = VDEV_STATE_CLOSED;
1699 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1703 vdev_hold(vdev_t *vd)
1705 spa_t *spa = vd->vdev_spa;
1707 ASSERT(spa_is_root(spa));
1708 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1711 for (int c = 0; c < vd->vdev_children; c++)
1712 vdev_hold(vd->vdev_child[c]);
1714 if (vd->vdev_ops->vdev_op_leaf)
1715 vd->vdev_ops->vdev_op_hold(vd);
1719 vdev_rele(vdev_t *vd)
1721 spa_t *spa = vd->vdev_spa;
1723 ASSERT(spa_is_root(spa));
1724 for (int c = 0; c < vd->vdev_children; c++)
1725 vdev_rele(vd->vdev_child[c]);
1727 if (vd->vdev_ops->vdev_op_leaf)
1728 vd->vdev_ops->vdev_op_rele(vd);
1732 * Reopen all interior vdevs and any unopened leaves. We don't actually
1733 * reopen leaf vdevs which had previously been opened as they might deadlock
1734 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1735 * If the leaf has never been opened then open it, as usual.
1738 vdev_reopen(vdev_t *vd)
1740 spa_t *spa = vd->vdev_spa;
1742 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1744 /* set the reopening flag unless we're taking the vdev offline */
1745 vd->vdev_reopening = !vd->vdev_offline;
1747 (void) vdev_open(vd);
1750 * Call vdev_validate() here to make sure we have the same device.
1751 * Otherwise, a device with an invalid label could be successfully
1752 * opened in response to vdev_reopen().
1755 (void) vdev_validate_aux(vd);
1756 if (vdev_readable(vd) && vdev_writeable(vd) &&
1757 vd->vdev_aux == &spa->spa_l2cache &&
1758 !l2arc_vdev_present(vd))
1759 l2arc_add_vdev(spa, vd);
1761 (void) vdev_validate(vd, B_TRUE);
1765 * Reassess parent vdev's health.
1767 vdev_propagate_state(vd);
1771 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1776 * Normally, partial opens (e.g. of a mirror) are allowed.
1777 * For a create, however, we want to fail the request if
1778 * there are any components we can't open.
1780 error = vdev_open(vd);
1782 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1784 return (error ? error : ENXIO);
1788 * Recursively load DTLs and initialize all labels.
1790 if ((error = vdev_dtl_load(vd)) != 0 ||
1791 (error = vdev_label_init(vd, txg, isreplacing ?
1792 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1801 vdev_metaslab_set_size(vdev_t *vd)
1804 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1806 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1807 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1811 * Maximize performance by inflating the configured ashift for top level
1812 * vdevs to be as close to the physical ashift as possible while maintaining
1813 * administrator defined limits and ensuring it doesn't go below the
1817 vdev_ashift_optimize(vdev_t *vd)
1819 if (vd == vd->vdev_top) {
1820 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1821 vd->vdev_ashift = MIN(
1822 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1823 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1826 * Unusual case where logical ashift > physical ashift
1827 * so we can't cap the calculated ashift based on max
1828 * ashift as that would cause failures.
1829 * We still check if we need to increase it to match
1832 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1839 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1841 ASSERT(vd == vd->vdev_top);
1842 /* indirect vdevs don't have metaslabs or dtls */
1843 ASSERT(vdev_is_concrete(vd) || flags == 0);
1844 ASSERT(ISP2(flags));
1845 ASSERT(spa_writeable(vd->vdev_spa));
1847 if (flags & VDD_METASLAB)
1848 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1850 if (flags & VDD_DTL)
1851 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1853 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1857 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1859 for (int c = 0; c < vd->vdev_children; c++)
1860 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1862 if (vd->vdev_ops->vdev_op_leaf)
1863 vdev_dirty(vd->vdev_top, flags, vd, txg);
1869 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1870 * the vdev has less than perfect replication. There are four kinds of DTL:
1872 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1874 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1876 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1877 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1878 * txgs that was scrubbed.
1880 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1881 * persistent errors or just some device being offline.
1882 * Unlike the other three, the DTL_OUTAGE map is not generally
1883 * maintained; it's only computed when needed, typically to
1884 * determine whether a device can be detached.
1886 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1887 * either has the data or it doesn't.
1889 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1890 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1891 * if any child is less than fully replicated, then so is its parent.
1892 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1893 * comprising only those txgs which appear in 'maxfaults' or more children;
1894 * those are the txgs we don't have enough replication to read. For example,
1895 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1896 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1897 * two child DTL_MISSING maps.
1899 * It should be clear from the above that to compute the DTLs and outage maps
1900 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1901 * Therefore, that is all we keep on disk. When loading the pool, or after
1902 * a configuration change, we generate all other DTLs from first principles.
1905 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1907 range_tree_t *rt = vd->vdev_dtl[t];
1909 ASSERT(t < DTL_TYPES);
1910 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1911 ASSERT(spa_writeable(vd->vdev_spa));
1913 mutex_enter(&vd->vdev_dtl_lock);
1914 if (!range_tree_contains(rt, txg, size))
1915 range_tree_add(rt, txg, size);
1916 mutex_exit(&vd->vdev_dtl_lock);
1920 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1922 range_tree_t *rt = vd->vdev_dtl[t];
1923 boolean_t dirty = B_FALSE;
1925 ASSERT(t < DTL_TYPES);
1926 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1929 * While we are loading the pool, the DTLs have not been loaded yet.
1930 * Ignore the DTLs and try all devices. This avoids a recursive
1931 * mutex enter on the vdev_dtl_lock, and also makes us try hard
1932 * when loading the pool (relying on the checksum to ensure that
1933 * we get the right data -- note that we while loading, we are
1934 * only reading the MOS, which is always checksummed).
1936 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
1939 mutex_enter(&vd->vdev_dtl_lock);
1940 if (range_tree_space(rt) != 0)
1941 dirty = range_tree_contains(rt, txg, size);
1942 mutex_exit(&vd->vdev_dtl_lock);
1948 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1950 range_tree_t *rt = vd->vdev_dtl[t];
1953 mutex_enter(&vd->vdev_dtl_lock);
1954 empty = (range_tree_space(rt) == 0);
1955 mutex_exit(&vd->vdev_dtl_lock);
1961 * Returns the lowest txg in the DTL range.
1964 vdev_dtl_min(vdev_t *vd)
1968 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1969 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1970 ASSERT0(vd->vdev_children);
1972 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1973 return (rs->rs_start - 1);
1977 * Returns the highest txg in the DTL.
1980 vdev_dtl_max(vdev_t *vd)
1984 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1985 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1986 ASSERT0(vd->vdev_children);
1988 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1989 return (rs->rs_end);
1993 * Determine if a resilvering vdev should remove any DTL entries from
1994 * its range. If the vdev was resilvering for the entire duration of the
1995 * scan then it should excise that range from its DTLs. Otherwise, this
1996 * vdev is considered partially resilvered and should leave its DTL
1997 * entries intact. The comment in vdev_dtl_reassess() describes how we
2001 vdev_dtl_should_excise(vdev_t *vd)
2003 spa_t *spa = vd->vdev_spa;
2004 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2006 ASSERT0(scn->scn_phys.scn_errors);
2007 ASSERT0(vd->vdev_children);
2009 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2012 if (vd->vdev_resilver_txg == 0 ||
2013 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
2017 * When a resilver is initiated the scan will assign the scn_max_txg
2018 * value to the highest txg value that exists in all DTLs. If this
2019 * device's max DTL is not part of this scan (i.e. it is not in
2020 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2023 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2024 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2025 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2026 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2033 * Reassess DTLs after a config change or scrub completion.
2036 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2038 spa_t *spa = vd->vdev_spa;
2042 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2044 for (int c = 0; c < vd->vdev_children; c++)
2045 vdev_dtl_reassess(vd->vdev_child[c], txg,
2046 scrub_txg, scrub_done);
2048 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2051 if (vd->vdev_ops->vdev_op_leaf) {
2052 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2054 mutex_enter(&vd->vdev_dtl_lock);
2057 * If we've completed a scan cleanly then determine
2058 * if this vdev should remove any DTLs. We only want to
2059 * excise regions on vdevs that were available during
2060 * the entire duration of this scan.
2062 if (scrub_txg != 0 &&
2063 (spa->spa_scrub_started ||
2064 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2065 vdev_dtl_should_excise(vd)) {
2067 * We completed a scrub up to scrub_txg. If we
2068 * did it without rebooting, then the scrub dtl
2069 * will be valid, so excise the old region and
2070 * fold in the scrub dtl. Otherwise, leave the
2071 * dtl as-is if there was an error.
2073 * There's little trick here: to excise the beginning
2074 * of the DTL_MISSING map, we put it into a reference
2075 * tree and then add a segment with refcnt -1 that
2076 * covers the range [0, scrub_txg). This means
2077 * that each txg in that range has refcnt -1 or 0.
2078 * We then add DTL_SCRUB with a refcnt of 2, so that
2079 * entries in the range [0, scrub_txg) will have a
2080 * positive refcnt -- either 1 or 2. We then convert
2081 * the reference tree into the new DTL_MISSING map.
2083 space_reftree_create(&reftree);
2084 space_reftree_add_map(&reftree,
2085 vd->vdev_dtl[DTL_MISSING], 1);
2086 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2087 space_reftree_add_map(&reftree,
2088 vd->vdev_dtl[DTL_SCRUB], 2);
2089 space_reftree_generate_map(&reftree,
2090 vd->vdev_dtl[DTL_MISSING], 1);
2091 space_reftree_destroy(&reftree);
2093 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2094 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2095 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2097 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2098 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2099 if (!vdev_readable(vd))
2100 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2102 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2103 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2106 * If the vdev was resilvering and no longer has any
2107 * DTLs then reset its resilvering flag and dirty
2108 * the top level so that we persist the change.
2110 if (vd->vdev_resilver_txg != 0 &&
2111 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
2112 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
2113 vd->vdev_resilver_txg = 0;
2114 vdev_config_dirty(vd->vdev_top);
2117 mutex_exit(&vd->vdev_dtl_lock);
2120 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2124 mutex_enter(&vd->vdev_dtl_lock);
2125 for (int t = 0; t < DTL_TYPES; t++) {
2126 /* account for child's outage in parent's missing map */
2127 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2129 continue; /* leaf vdevs only */
2130 if (t == DTL_PARTIAL)
2131 minref = 1; /* i.e. non-zero */
2132 else if (vd->vdev_nparity != 0)
2133 minref = vd->vdev_nparity + 1; /* RAID-Z */
2135 minref = vd->vdev_children; /* any kind of mirror */
2136 space_reftree_create(&reftree);
2137 for (int c = 0; c < vd->vdev_children; c++) {
2138 vdev_t *cvd = vd->vdev_child[c];
2139 mutex_enter(&cvd->vdev_dtl_lock);
2140 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2141 mutex_exit(&cvd->vdev_dtl_lock);
2143 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2144 space_reftree_destroy(&reftree);
2146 mutex_exit(&vd->vdev_dtl_lock);
2150 vdev_dtl_load(vdev_t *vd)
2152 spa_t *spa = vd->vdev_spa;
2153 objset_t *mos = spa->spa_meta_objset;
2156 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2157 ASSERT(vdev_is_concrete(vd));
2159 error = space_map_open(&vd->vdev_dtl_sm, mos,
2160 vd->vdev_dtl_object, 0, -1ULL, 0);
2163 ASSERT(vd->vdev_dtl_sm != NULL);
2165 mutex_enter(&vd->vdev_dtl_lock);
2168 * Now that we've opened the space_map we need to update
2171 space_map_update(vd->vdev_dtl_sm);
2173 error = space_map_load(vd->vdev_dtl_sm,
2174 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2175 mutex_exit(&vd->vdev_dtl_lock);
2180 for (int c = 0; c < vd->vdev_children; c++) {
2181 error = vdev_dtl_load(vd->vdev_child[c]);
2190 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2192 spa_t *spa = vd->vdev_spa;
2194 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2195 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2200 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2202 spa_t *spa = vd->vdev_spa;
2203 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2204 DMU_OT_NONE, 0, tx);
2207 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2214 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2216 if (vd->vdev_ops != &vdev_hole_ops &&
2217 vd->vdev_ops != &vdev_missing_ops &&
2218 vd->vdev_ops != &vdev_root_ops &&
2219 !vd->vdev_top->vdev_removing) {
2220 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2221 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2223 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2224 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2227 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2228 vdev_construct_zaps(vd->vdev_child[i], tx);
2233 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2235 spa_t *spa = vd->vdev_spa;
2236 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2237 objset_t *mos = spa->spa_meta_objset;
2238 range_tree_t *rtsync;
2240 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2242 ASSERT(vdev_is_concrete(vd));
2243 ASSERT(vd->vdev_ops->vdev_op_leaf);
2245 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2247 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2248 mutex_enter(&vd->vdev_dtl_lock);
2249 space_map_free(vd->vdev_dtl_sm, tx);
2250 space_map_close(vd->vdev_dtl_sm);
2251 vd->vdev_dtl_sm = NULL;
2252 mutex_exit(&vd->vdev_dtl_lock);
2255 * We only destroy the leaf ZAP for detached leaves or for
2256 * removed log devices. Removed data devices handle leaf ZAP
2257 * cleanup later, once cancellation is no longer possible.
2259 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2260 vd->vdev_top->vdev_islog)) {
2261 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2262 vd->vdev_leaf_zap = 0;
2269 if (vd->vdev_dtl_sm == NULL) {
2270 uint64_t new_object;
2272 new_object = space_map_alloc(mos, tx);
2273 VERIFY3U(new_object, !=, 0);
2275 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2277 ASSERT(vd->vdev_dtl_sm != NULL);
2280 rtsync = range_tree_create(NULL, NULL);
2282 mutex_enter(&vd->vdev_dtl_lock);
2283 range_tree_walk(rt, range_tree_add, rtsync);
2284 mutex_exit(&vd->vdev_dtl_lock);
2286 space_map_truncate(vd->vdev_dtl_sm, tx);
2287 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2288 range_tree_vacate(rtsync, NULL, NULL);
2290 range_tree_destroy(rtsync);
2293 * If the object for the space map has changed then dirty
2294 * the top level so that we update the config.
2296 if (object != space_map_object(vd->vdev_dtl_sm)) {
2297 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2298 "new object %llu", txg, spa_name(spa), object,
2299 space_map_object(vd->vdev_dtl_sm));
2300 vdev_config_dirty(vd->vdev_top);
2305 mutex_enter(&vd->vdev_dtl_lock);
2306 space_map_update(vd->vdev_dtl_sm);
2307 mutex_exit(&vd->vdev_dtl_lock);
2311 * Determine whether the specified vdev can be offlined/detached/removed
2312 * without losing data.
2315 vdev_dtl_required(vdev_t *vd)
2317 spa_t *spa = vd->vdev_spa;
2318 vdev_t *tvd = vd->vdev_top;
2319 uint8_t cant_read = vd->vdev_cant_read;
2322 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2324 if (vd == spa->spa_root_vdev || vd == tvd)
2328 * Temporarily mark the device as unreadable, and then determine
2329 * whether this results in any DTL outages in the top-level vdev.
2330 * If not, we can safely offline/detach/remove the device.
2332 vd->vdev_cant_read = B_TRUE;
2333 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2334 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2335 vd->vdev_cant_read = cant_read;
2336 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2338 if (!required && zio_injection_enabled)
2339 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2345 * Determine if resilver is needed, and if so the txg range.
2348 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2350 boolean_t needed = B_FALSE;
2351 uint64_t thismin = UINT64_MAX;
2352 uint64_t thismax = 0;
2354 if (vd->vdev_children == 0) {
2355 mutex_enter(&vd->vdev_dtl_lock);
2356 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2357 vdev_writeable(vd)) {
2359 thismin = vdev_dtl_min(vd);
2360 thismax = vdev_dtl_max(vd);
2363 mutex_exit(&vd->vdev_dtl_lock);
2365 for (int c = 0; c < vd->vdev_children; c++) {
2366 vdev_t *cvd = vd->vdev_child[c];
2367 uint64_t cmin, cmax;
2369 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2370 thismin = MIN(thismin, cmin);
2371 thismax = MAX(thismax, cmax);
2377 if (needed && minp) {
2385 vdev_load(vdev_t *vd)
2389 * Recursively load all children.
2391 for (int c = 0; c < vd->vdev_children; c++) {
2392 error = vdev_load(vd->vdev_child[c]);
2398 vdev_set_deflate_ratio(vd);
2401 * If this is a top-level vdev, initialize its metaslabs.
2403 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2404 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2405 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2406 VDEV_AUX_CORRUPT_DATA);
2407 return (SET_ERROR(ENXIO));
2408 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2409 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2410 VDEV_AUX_CORRUPT_DATA);
2416 * If this is a leaf vdev, load its DTL.
2418 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2419 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2420 VDEV_AUX_CORRUPT_DATA);
2424 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2425 if (obsolete_sm_object != 0) {
2426 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2427 ASSERT(vd->vdev_asize != 0);
2428 ASSERT(vd->vdev_obsolete_sm == NULL);
2430 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2431 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2432 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2433 VDEV_AUX_CORRUPT_DATA);
2436 space_map_update(vd->vdev_obsolete_sm);
2443 * The special vdev case is used for hot spares and l2cache devices. Its
2444 * sole purpose it to set the vdev state for the associated vdev. To do this,
2445 * we make sure that we can open the underlying device, then try to read the
2446 * label, and make sure that the label is sane and that it hasn't been
2447 * repurposed to another pool.
2450 vdev_validate_aux(vdev_t *vd)
2453 uint64_t guid, version;
2456 if (!vdev_readable(vd))
2459 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2460 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2461 VDEV_AUX_CORRUPT_DATA);
2465 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2466 !SPA_VERSION_IS_SUPPORTED(version) ||
2467 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2468 guid != vd->vdev_guid ||
2469 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2470 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2471 VDEV_AUX_CORRUPT_DATA);
2477 * We don't actually check the pool state here. If it's in fact in
2478 * use by another pool, we update this fact on the fly when requested.
2485 * Free the objects used to store this vdev's spacemaps, and the array
2486 * that points to them.
2489 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
2491 if (vd->vdev_ms_array == 0)
2494 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2495 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
2496 size_t array_bytes = array_count * sizeof (uint64_t);
2497 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
2498 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
2499 array_bytes, smobj_array, 0));
2501 for (uint64_t i = 0; i < array_count; i++) {
2502 uint64_t smobj = smobj_array[i];
2506 space_map_free_obj(mos, smobj, tx);
2509 kmem_free(smobj_array, array_bytes);
2510 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
2511 vd->vdev_ms_array = 0;
2515 vdev_remove_empty(vdev_t *vd, uint64_t txg)
2517 spa_t *spa = vd->vdev_spa;
2520 ASSERT(vd == vd->vdev_top);
2521 ASSERT3U(txg, ==, spa_syncing_txg(spa));
2523 if (vd->vdev_ms != NULL) {
2524 metaslab_group_t *mg = vd->vdev_mg;
2526 metaslab_group_histogram_verify(mg);
2527 metaslab_class_histogram_verify(mg->mg_class);
2529 for (int m = 0; m < vd->vdev_ms_count; m++) {
2530 metaslab_t *msp = vd->vdev_ms[m];
2532 if (msp == NULL || msp->ms_sm == NULL)
2535 mutex_enter(&msp->ms_lock);
2537 * If the metaslab was not loaded when the vdev
2538 * was removed then the histogram accounting may
2539 * not be accurate. Update the histogram information
2540 * here so that we ensure that the metaslab group
2541 * and metaslab class are up-to-date.
2543 metaslab_group_histogram_remove(mg, msp);
2545 VERIFY0(space_map_allocated(msp->ms_sm));
2546 space_map_close(msp->ms_sm);
2548 mutex_exit(&msp->ms_lock);
2551 metaslab_group_histogram_verify(mg);
2552 metaslab_class_histogram_verify(mg->mg_class);
2553 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2554 ASSERT0(mg->mg_histogram[i]);
2557 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2558 vdev_destroy_spacemaps(vd, tx);
2560 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
2561 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
2562 vd->vdev_top_zap = 0;
2568 vdev_sync_done(vdev_t *vd, uint64_t txg)
2571 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2573 ASSERT(vdev_is_concrete(vd));
2575 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2576 metaslab_sync_done(msp, txg);
2579 metaslab_sync_reassess(vd->vdev_mg);
2583 vdev_sync(vdev_t *vd, uint64_t txg)
2585 spa_t *spa = vd->vdev_spa;
2590 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
2593 ASSERT(vd->vdev_removing ||
2594 vd->vdev_ops == &vdev_indirect_ops);
2596 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2597 vdev_indirect_sync_obsolete(vd, tx);
2601 * If the vdev is indirect, it can't have dirty
2602 * metaslabs or DTLs.
2604 if (vd->vdev_ops == &vdev_indirect_ops) {
2605 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
2606 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
2611 ASSERT(vdev_is_concrete(vd));
2613 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
2614 !vd->vdev_removing) {
2615 ASSERT(vd == vd->vdev_top);
2616 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
2617 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2618 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2619 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2620 ASSERT(vd->vdev_ms_array != 0);
2621 vdev_config_dirty(vd);
2625 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2626 metaslab_sync(msp, txg);
2627 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2630 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2631 vdev_dtl_sync(lvd, txg);
2634 * Remove the metadata associated with this vdev once it's empty.
2635 * Note that this is typically used for log/cache device removal;
2636 * we don't empty toplevel vdevs when removing them. But if
2637 * a toplevel happens to be emptied, this is not harmful.
2639 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) {
2640 vdev_remove_empty(vd, txg);
2643 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2647 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2649 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2653 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2654 * not be opened, and no I/O is attempted.
2657 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2661 spa_vdev_state_enter(spa, SCL_NONE);
2663 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2664 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2666 if (!vd->vdev_ops->vdev_op_leaf)
2667 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2672 * We don't directly use the aux state here, but if we do a
2673 * vdev_reopen(), we need this value to be present to remember why we
2676 vd->vdev_label_aux = aux;
2679 * Faulted state takes precedence over degraded.
2681 vd->vdev_delayed_close = B_FALSE;
2682 vd->vdev_faulted = 1ULL;
2683 vd->vdev_degraded = 0ULL;
2684 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2687 * If this device has the only valid copy of the data, then
2688 * back off and simply mark the vdev as degraded instead.
2690 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2691 vd->vdev_degraded = 1ULL;
2692 vd->vdev_faulted = 0ULL;
2695 * If we reopen the device and it's not dead, only then do we
2700 if (vdev_readable(vd))
2701 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2704 return (spa_vdev_state_exit(spa, vd, 0));
2708 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2709 * user that something is wrong. The vdev continues to operate as normal as far
2710 * as I/O is concerned.
2713 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2717 spa_vdev_state_enter(spa, SCL_NONE);
2719 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2720 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2722 if (!vd->vdev_ops->vdev_op_leaf)
2723 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2726 * If the vdev is already faulted, then don't do anything.
2728 if (vd->vdev_faulted || vd->vdev_degraded)
2729 return (spa_vdev_state_exit(spa, NULL, 0));
2731 vd->vdev_degraded = 1ULL;
2732 if (!vdev_is_dead(vd))
2733 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2736 return (spa_vdev_state_exit(spa, vd, 0));
2740 * Online the given vdev.
2742 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2743 * spare device should be detached when the device finishes resilvering.
2744 * Second, the online should be treated like a 'test' online case, so no FMA
2745 * events are generated if the device fails to open.
2748 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2750 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2751 boolean_t wasoffline;
2752 vdev_state_t oldstate;
2754 spa_vdev_state_enter(spa, SCL_NONE);
2756 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2757 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2759 if (!vd->vdev_ops->vdev_op_leaf)
2760 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2762 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
2763 oldstate = vd->vdev_state;
2766 vd->vdev_offline = B_FALSE;
2767 vd->vdev_tmpoffline = B_FALSE;
2768 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2769 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2771 /* XXX - L2ARC 1.0 does not support expansion */
2772 if (!vd->vdev_aux) {
2773 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2774 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2778 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2780 if (!vd->vdev_aux) {
2781 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2782 pvd->vdev_expanding = B_FALSE;
2786 *newstate = vd->vdev_state;
2787 if ((flags & ZFS_ONLINE_UNSPARE) &&
2788 !vdev_is_dead(vd) && vd->vdev_parent &&
2789 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2790 vd->vdev_parent->vdev_child[0] == vd)
2791 vd->vdev_unspare = B_TRUE;
2793 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2795 /* XXX - L2ARC 1.0 does not support expansion */
2797 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2798 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2802 (oldstate < VDEV_STATE_DEGRADED &&
2803 vd->vdev_state >= VDEV_STATE_DEGRADED))
2804 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
2806 return (spa_vdev_state_exit(spa, vd, 0));
2810 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2814 uint64_t generation;
2815 metaslab_group_t *mg;
2818 spa_vdev_state_enter(spa, SCL_ALLOC);
2820 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2821 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2823 if (!vd->vdev_ops->vdev_op_leaf)
2824 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2828 generation = spa->spa_config_generation + 1;
2831 * If the device isn't already offline, try to offline it.
2833 if (!vd->vdev_offline) {
2835 * If this device has the only valid copy of some data,
2836 * don't allow it to be offlined. Log devices are always
2839 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2840 vdev_dtl_required(vd))
2841 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2844 * If the top-level is a slog and it has had allocations
2845 * then proceed. We check that the vdev's metaslab group
2846 * is not NULL since it's possible that we may have just
2847 * added this vdev but not yet initialized its metaslabs.
2849 if (tvd->vdev_islog && mg != NULL) {
2851 * Prevent any future allocations.
2853 metaslab_group_passivate(mg);
2854 (void) spa_vdev_state_exit(spa, vd, 0);
2856 error = spa_reset_logs(spa);
2858 spa_vdev_state_enter(spa, SCL_ALLOC);
2861 * Check to see if the config has changed.
2863 if (error || generation != spa->spa_config_generation) {
2864 metaslab_group_activate(mg);
2866 return (spa_vdev_state_exit(spa,
2868 (void) spa_vdev_state_exit(spa, vd, 0);
2871 ASSERT0(tvd->vdev_stat.vs_alloc);
2875 * Offline this device and reopen its top-level vdev.
2876 * If the top-level vdev is a log device then just offline
2877 * it. Otherwise, if this action results in the top-level
2878 * vdev becoming unusable, undo it and fail the request.
2880 vd->vdev_offline = B_TRUE;
2883 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2884 vdev_is_dead(tvd)) {
2885 vd->vdev_offline = B_FALSE;
2887 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2891 * Add the device back into the metaslab rotor so that
2892 * once we online the device it's open for business.
2894 if (tvd->vdev_islog && mg != NULL)
2895 metaslab_group_activate(mg);
2898 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2900 return (spa_vdev_state_exit(spa, vd, 0));
2904 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2908 mutex_enter(&spa->spa_vdev_top_lock);
2909 error = vdev_offline_locked(spa, guid, flags);
2910 mutex_exit(&spa->spa_vdev_top_lock);
2916 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2917 * vdev_offline(), we assume the spa config is locked. We also clear all
2918 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2921 vdev_clear(spa_t *spa, vdev_t *vd)
2923 vdev_t *rvd = spa->spa_root_vdev;
2925 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2930 vd->vdev_stat.vs_read_errors = 0;
2931 vd->vdev_stat.vs_write_errors = 0;
2932 vd->vdev_stat.vs_checksum_errors = 0;
2934 for (int c = 0; c < vd->vdev_children; c++)
2935 vdev_clear(spa, vd->vdev_child[c]);
2938 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2939 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2941 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2942 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2946 * It makes no sense to "clear" an indirect vdev.
2948 if (!vdev_is_concrete(vd))
2952 * If we're in the FAULTED state or have experienced failed I/O, then
2953 * clear the persistent state and attempt to reopen the device. We
2954 * also mark the vdev config dirty, so that the new faulted state is
2955 * written out to disk.
2957 if (vd->vdev_faulted || vd->vdev_degraded ||
2958 !vdev_readable(vd) || !vdev_writeable(vd)) {
2961 * When reopening in reponse to a clear event, it may be due to
2962 * a fmadm repair request. In this case, if the device is
2963 * still broken, we want to still post the ereport again.
2965 vd->vdev_forcefault = B_TRUE;
2967 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2968 vd->vdev_cant_read = B_FALSE;
2969 vd->vdev_cant_write = B_FALSE;
2971 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2973 vd->vdev_forcefault = B_FALSE;
2975 if (vd != rvd && vdev_writeable(vd->vdev_top))
2976 vdev_state_dirty(vd->vdev_top);
2978 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2979 spa_async_request(spa, SPA_ASYNC_RESILVER);
2981 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
2985 * When clearing a FMA-diagnosed fault, we always want to
2986 * unspare the device, as we assume that the original spare was
2987 * done in response to the FMA fault.
2989 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2990 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2991 vd->vdev_parent->vdev_child[0] == vd)
2992 vd->vdev_unspare = B_TRUE;
2996 vdev_is_dead(vdev_t *vd)
2999 * Holes and missing devices are always considered "dead".
3000 * This simplifies the code since we don't have to check for
3001 * these types of devices in the various code paths.
3002 * Instead we rely on the fact that we skip over dead devices
3003 * before issuing I/O to them.
3005 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3006 vd->vdev_ops == &vdev_hole_ops ||
3007 vd->vdev_ops == &vdev_missing_ops);
3011 vdev_readable(vdev_t *vd)
3013 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3017 vdev_writeable(vdev_t *vd)
3019 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3020 vdev_is_concrete(vd));
3024 vdev_allocatable(vdev_t *vd)
3026 uint64_t state = vd->vdev_state;
3029 * We currently allow allocations from vdevs which may be in the
3030 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3031 * fails to reopen then we'll catch it later when we're holding
3032 * the proper locks. Note that we have to get the vdev state
3033 * in a local variable because although it changes atomically,
3034 * we're asking two separate questions about it.
3036 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3037 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3038 vd->vdev_mg->mg_initialized);
3042 vdev_accessible(vdev_t *vd, zio_t *zio)
3044 ASSERT(zio->io_vd == vd);
3046 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3049 if (zio->io_type == ZIO_TYPE_READ)
3050 return (!vd->vdev_cant_read);
3052 if (zio->io_type == ZIO_TYPE_WRITE)
3053 return (!vd->vdev_cant_write);
3059 * Get statistics for the given vdev.
3062 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3064 spa_t *spa = vd->vdev_spa;
3065 vdev_t *rvd = spa->spa_root_vdev;
3066 vdev_t *tvd = vd->vdev_top;
3068 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3070 mutex_enter(&vd->vdev_stat_lock);
3071 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3072 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3073 vs->vs_state = vd->vdev_state;
3074 vs->vs_rsize = vdev_get_min_asize(vd);
3075 if (vd->vdev_ops->vdev_op_leaf)
3076 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3078 * Report expandable space on top-level, non-auxillary devices only.
3079 * The expandable space is reported in terms of metaslab sized units
3080 * since that determines how much space the pool can expand.
3082 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
3083 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3084 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3086 vs->vs_configured_ashift = vd->vdev_top != NULL
3087 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
3088 vs->vs_logical_ashift = vd->vdev_logical_ashift;
3089 vs->vs_physical_ashift = vd->vdev_physical_ashift;
3090 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3091 vdev_is_concrete(vd)) {
3092 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
3096 * If we're getting stats on the root vdev, aggregate the I/O counts
3097 * over all top-level vdevs (i.e. the direct children of the root).
3100 for (int c = 0; c < rvd->vdev_children; c++) {
3101 vdev_t *cvd = rvd->vdev_child[c];
3102 vdev_stat_t *cvs = &cvd->vdev_stat;
3104 for (int t = 0; t < ZIO_TYPES; t++) {
3105 vs->vs_ops[t] += cvs->vs_ops[t];
3106 vs->vs_bytes[t] += cvs->vs_bytes[t];
3108 cvs->vs_scan_removing = cvd->vdev_removing;
3111 mutex_exit(&vd->vdev_stat_lock);
3115 vdev_clear_stats(vdev_t *vd)
3117 mutex_enter(&vd->vdev_stat_lock);
3118 vd->vdev_stat.vs_space = 0;
3119 vd->vdev_stat.vs_dspace = 0;
3120 vd->vdev_stat.vs_alloc = 0;
3121 mutex_exit(&vd->vdev_stat_lock);
3125 vdev_scan_stat_init(vdev_t *vd)
3127 vdev_stat_t *vs = &vd->vdev_stat;
3129 for (int c = 0; c < vd->vdev_children; c++)
3130 vdev_scan_stat_init(vd->vdev_child[c]);
3132 mutex_enter(&vd->vdev_stat_lock);
3133 vs->vs_scan_processed = 0;
3134 mutex_exit(&vd->vdev_stat_lock);
3138 vdev_stat_update(zio_t *zio, uint64_t psize)
3140 spa_t *spa = zio->io_spa;
3141 vdev_t *rvd = spa->spa_root_vdev;
3142 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3144 uint64_t txg = zio->io_txg;
3145 vdev_stat_t *vs = &vd->vdev_stat;
3146 zio_type_t type = zio->io_type;
3147 int flags = zio->io_flags;
3150 * If this i/o is a gang leader, it didn't do any actual work.
3152 if (zio->io_gang_tree)
3155 if (zio->io_error == 0) {
3157 * If this is a root i/o, don't count it -- we've already
3158 * counted the top-level vdevs, and vdev_get_stats() will
3159 * aggregate them when asked. This reduces contention on
3160 * the root vdev_stat_lock and implicitly handles blocks
3161 * that compress away to holes, for which there is no i/o.
3162 * (Holes never create vdev children, so all the counters
3163 * remain zero, which is what we want.)
3165 * Note: this only applies to successful i/o (io_error == 0)
3166 * because unlike i/o counts, errors are not additive.
3167 * When reading a ditto block, for example, failure of
3168 * one top-level vdev does not imply a root-level error.
3173 ASSERT(vd == zio->io_vd);
3175 if (flags & ZIO_FLAG_IO_BYPASS)
3178 mutex_enter(&vd->vdev_stat_lock);
3180 if (flags & ZIO_FLAG_IO_REPAIR) {
3181 if (flags & ZIO_FLAG_SCAN_THREAD) {
3182 dsl_scan_phys_t *scn_phys =
3183 &spa->spa_dsl_pool->dp_scan->scn_phys;
3184 uint64_t *processed = &scn_phys->scn_processed;
3187 if (vd->vdev_ops->vdev_op_leaf)
3188 atomic_add_64(processed, psize);
3189 vs->vs_scan_processed += psize;
3192 if (flags & ZIO_FLAG_SELF_HEAL)
3193 vs->vs_self_healed += psize;
3197 vs->vs_bytes[type] += psize;
3199 mutex_exit(&vd->vdev_stat_lock);
3203 if (flags & ZIO_FLAG_SPECULATIVE)
3207 * If this is an I/O error that is going to be retried, then ignore the
3208 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3209 * hard errors, when in reality they can happen for any number of
3210 * innocuous reasons (bus resets, MPxIO link failure, etc).
3212 if (zio->io_error == EIO &&
3213 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3217 * Intent logs writes won't propagate their error to the root
3218 * I/O so don't mark these types of failures as pool-level
3221 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3224 mutex_enter(&vd->vdev_stat_lock);
3225 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3226 if (zio->io_error == ECKSUM)
3227 vs->vs_checksum_errors++;
3229 vs->vs_read_errors++;
3231 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3232 vs->vs_write_errors++;
3233 mutex_exit(&vd->vdev_stat_lock);
3235 if (spa->spa_load_state == SPA_LOAD_NONE &&
3236 type == ZIO_TYPE_WRITE && txg != 0 &&
3237 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3238 (flags & ZIO_FLAG_SCAN_THREAD) ||
3239 spa->spa_claiming)) {
3241 * This is either a normal write (not a repair), or it's
3242 * a repair induced by the scrub thread, or it's a repair
3243 * made by zil_claim() during spa_load() in the first txg.
3244 * In the normal case, we commit the DTL change in the same
3245 * txg as the block was born. In the scrub-induced repair
3246 * case, we know that scrubs run in first-pass syncing context,
3247 * so we commit the DTL change in spa_syncing_txg(spa).
3248 * In the zil_claim() case, we commit in spa_first_txg(spa).
3250 * We currently do not make DTL entries for failed spontaneous
3251 * self-healing writes triggered by normal (non-scrubbing)
3252 * reads, because we have no transactional context in which to
3253 * do so -- and it's not clear that it'd be desirable anyway.
3255 if (vd->vdev_ops->vdev_op_leaf) {
3256 uint64_t commit_txg = txg;
3257 if (flags & ZIO_FLAG_SCAN_THREAD) {
3258 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3259 ASSERT(spa_sync_pass(spa) == 1);
3260 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3261 commit_txg = spa_syncing_txg(spa);
3262 } else if (spa->spa_claiming) {
3263 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3264 commit_txg = spa_first_txg(spa);
3266 ASSERT(commit_txg >= spa_syncing_txg(spa));
3267 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3269 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3270 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3271 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3274 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3279 * Update the in-core space usage stats for this vdev, its metaslab class,
3280 * and the root vdev.
3283 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3284 int64_t space_delta)
3286 int64_t dspace_delta = space_delta;
3287 spa_t *spa = vd->vdev_spa;
3288 vdev_t *rvd = spa->spa_root_vdev;
3289 metaslab_group_t *mg = vd->vdev_mg;
3290 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3292 ASSERT(vd == vd->vdev_top);
3295 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3296 * factor. We must calculate this here and not at the root vdev
3297 * because the root vdev's psize-to-asize is simply the max of its
3298 * childrens', thus not accurate enough for us.
3300 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3301 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3302 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3303 vd->vdev_deflate_ratio;
3305 mutex_enter(&vd->vdev_stat_lock);
3306 vd->vdev_stat.vs_alloc += alloc_delta;
3307 vd->vdev_stat.vs_space += space_delta;
3308 vd->vdev_stat.vs_dspace += dspace_delta;
3309 mutex_exit(&vd->vdev_stat_lock);
3311 if (mc == spa_normal_class(spa)) {
3312 mutex_enter(&rvd->vdev_stat_lock);
3313 rvd->vdev_stat.vs_alloc += alloc_delta;
3314 rvd->vdev_stat.vs_space += space_delta;
3315 rvd->vdev_stat.vs_dspace += dspace_delta;
3316 mutex_exit(&rvd->vdev_stat_lock);
3320 ASSERT(rvd == vd->vdev_parent);
3321 ASSERT(vd->vdev_ms_count != 0);
3323 metaslab_class_space_update(mc,
3324 alloc_delta, defer_delta, space_delta, dspace_delta);
3329 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3330 * so that it will be written out next time the vdev configuration is synced.
3331 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3334 vdev_config_dirty(vdev_t *vd)
3336 spa_t *spa = vd->vdev_spa;
3337 vdev_t *rvd = spa->spa_root_vdev;
3340 ASSERT(spa_writeable(spa));
3343 * If this is an aux vdev (as with l2cache and spare devices), then we
3344 * update the vdev config manually and set the sync flag.
3346 if (vd->vdev_aux != NULL) {
3347 spa_aux_vdev_t *sav = vd->vdev_aux;
3351 for (c = 0; c < sav->sav_count; c++) {
3352 if (sav->sav_vdevs[c] == vd)
3356 if (c == sav->sav_count) {
3358 * We're being removed. There's nothing more to do.
3360 ASSERT(sav->sav_sync == B_TRUE);
3364 sav->sav_sync = B_TRUE;
3366 if (nvlist_lookup_nvlist_array(sav->sav_config,
3367 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3368 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3369 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3375 * Setting the nvlist in the middle if the array is a little
3376 * sketchy, but it will work.
3378 nvlist_free(aux[c]);
3379 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3385 * The dirty list is protected by the SCL_CONFIG lock. The caller
3386 * must either hold SCL_CONFIG as writer, or must be the sync thread
3387 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3388 * so this is sufficient to ensure mutual exclusion.
3390 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3391 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3392 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3395 for (c = 0; c < rvd->vdev_children; c++)
3396 vdev_config_dirty(rvd->vdev_child[c]);
3398 ASSERT(vd == vd->vdev_top);
3400 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3401 vdev_is_concrete(vd)) {
3402 list_insert_head(&spa->spa_config_dirty_list, vd);
3408 vdev_config_clean(vdev_t *vd)
3410 spa_t *spa = vd->vdev_spa;
3412 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3413 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3414 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3416 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3417 list_remove(&spa->spa_config_dirty_list, vd);
3421 * Mark a top-level vdev's state as dirty, so that the next pass of
3422 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3423 * the state changes from larger config changes because they require
3424 * much less locking, and are often needed for administrative actions.
3427 vdev_state_dirty(vdev_t *vd)
3429 spa_t *spa = vd->vdev_spa;
3431 ASSERT(spa_writeable(spa));
3432 ASSERT(vd == vd->vdev_top);
3435 * The state list is protected by the SCL_STATE lock. The caller
3436 * must either hold SCL_STATE as writer, or must be the sync thread
3437 * (which holds SCL_STATE as reader). There's only one sync thread,
3438 * so this is sufficient to ensure mutual exclusion.
3440 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3441 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3442 spa_config_held(spa, SCL_STATE, RW_READER)));
3444 if (!list_link_active(&vd->vdev_state_dirty_node) &&
3445 vdev_is_concrete(vd))
3446 list_insert_head(&spa->spa_state_dirty_list, vd);
3450 vdev_state_clean(vdev_t *vd)
3452 spa_t *spa = vd->vdev_spa;
3454 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3455 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3456 spa_config_held(spa, SCL_STATE, RW_READER)));
3458 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3459 list_remove(&spa->spa_state_dirty_list, vd);
3463 * Propagate vdev state up from children to parent.
3466 vdev_propagate_state(vdev_t *vd)
3468 spa_t *spa = vd->vdev_spa;
3469 vdev_t *rvd = spa->spa_root_vdev;
3470 int degraded = 0, faulted = 0;
3474 if (vd->vdev_children > 0) {
3475 for (int c = 0; c < vd->vdev_children; c++) {
3476 child = vd->vdev_child[c];
3479 * Don't factor holes or indirect vdevs into the
3482 if (!vdev_is_concrete(child))
3485 if (!vdev_readable(child) ||
3486 (!vdev_writeable(child) && spa_writeable(spa))) {
3488 * Root special: if there is a top-level log
3489 * device, treat the root vdev as if it were
3492 if (child->vdev_islog && vd == rvd)
3496 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3500 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3504 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3507 * Root special: if there is a top-level vdev that cannot be
3508 * opened due to corrupted metadata, then propagate the root
3509 * vdev's aux state as 'corrupt' rather than 'insufficient
3512 if (corrupted && vd == rvd &&
3513 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3514 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3515 VDEV_AUX_CORRUPT_DATA);
3518 if (vd->vdev_parent)
3519 vdev_propagate_state(vd->vdev_parent);
3523 * Set a vdev's state. If this is during an open, we don't update the parent
3524 * state, because we're in the process of opening children depth-first.
3525 * Otherwise, we propagate the change to the parent.
3527 * If this routine places a device in a faulted state, an appropriate ereport is
3531 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3533 uint64_t save_state;
3534 spa_t *spa = vd->vdev_spa;
3536 if (state == vd->vdev_state) {
3537 vd->vdev_stat.vs_aux = aux;
3541 save_state = vd->vdev_state;
3543 vd->vdev_state = state;
3544 vd->vdev_stat.vs_aux = aux;
3547 * If we are setting the vdev state to anything but an open state, then
3548 * always close the underlying device unless the device has requested
3549 * a delayed close (i.e. we're about to remove or fault the device).
3550 * Otherwise, we keep accessible but invalid devices open forever.
3551 * We don't call vdev_close() itself, because that implies some extra
3552 * checks (offline, etc) that we don't want here. This is limited to
3553 * leaf devices, because otherwise closing the device will affect other
3556 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3557 vd->vdev_ops->vdev_op_leaf)
3558 vd->vdev_ops->vdev_op_close(vd);
3560 if (vd->vdev_removed &&
3561 state == VDEV_STATE_CANT_OPEN &&
3562 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3564 * If the previous state is set to VDEV_STATE_REMOVED, then this
3565 * device was previously marked removed and someone attempted to
3566 * reopen it. If this failed due to a nonexistent device, then
3567 * keep the device in the REMOVED state. We also let this be if
3568 * it is one of our special test online cases, which is only
3569 * attempting to online the device and shouldn't generate an FMA
3572 vd->vdev_state = VDEV_STATE_REMOVED;
3573 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3574 } else if (state == VDEV_STATE_REMOVED) {
3575 vd->vdev_removed = B_TRUE;
3576 } else if (state == VDEV_STATE_CANT_OPEN) {
3578 * If we fail to open a vdev during an import or recovery, we
3579 * mark it as "not available", which signifies that it was
3580 * never there to begin with. Failure to open such a device
3581 * is not considered an error.
3583 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3584 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3585 vd->vdev_ops->vdev_op_leaf)
3586 vd->vdev_not_present = 1;
3589 * Post the appropriate ereport. If the 'prevstate' field is
3590 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3591 * that this is part of a vdev_reopen(). In this case, we don't
3592 * want to post the ereport if the device was already in the
3593 * CANT_OPEN state beforehand.
3595 * If the 'checkremove' flag is set, then this is an attempt to
3596 * online the device in response to an insertion event. If we
3597 * hit this case, then we have detected an insertion event for a
3598 * faulted or offline device that wasn't in the removed state.
3599 * In this scenario, we don't post an ereport because we are
3600 * about to replace the device, or attempt an online with
3601 * vdev_forcefault, which will generate the fault for us.
3603 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3604 !vd->vdev_not_present && !vd->vdev_checkremove &&
3605 vd != spa->spa_root_vdev) {
3609 case VDEV_AUX_OPEN_FAILED:
3610 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3612 case VDEV_AUX_CORRUPT_DATA:
3613 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3615 case VDEV_AUX_NO_REPLICAS:
3616 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3618 case VDEV_AUX_BAD_GUID_SUM:
3619 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3621 case VDEV_AUX_TOO_SMALL:
3622 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3624 case VDEV_AUX_BAD_LABEL:
3625 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3628 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3631 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3634 /* Erase any notion of persistent removed state */
3635 vd->vdev_removed = B_FALSE;
3637 vd->vdev_removed = B_FALSE;
3641 * Notify the fmd of the state change. Be verbose and post
3642 * notifications even for stuff that's not important; the fmd agent can
3643 * sort it out. Don't emit state change events for non-leaf vdevs since
3644 * they can't change state on their own. The FMD can check their state
3645 * if it wants to when it sees that a leaf vdev had a state change.
3647 if (vd->vdev_ops->vdev_op_leaf)
3648 zfs_post_state_change(spa, vd);
3650 if (!isopen && vd->vdev_parent)
3651 vdev_propagate_state(vd->vdev_parent);
3655 * Check the vdev configuration to ensure that it's capable of supporting
3656 * a root pool. We do not support partial configuration.
3657 * In addition, only a single top-level vdev is allowed.
3659 * FreeBSD does not have above limitations.
3662 vdev_is_bootable(vdev_t *vd)
3665 if (!vd->vdev_ops->vdev_op_leaf) {
3666 char *vdev_type = vd->vdev_ops->vdev_op_type;
3668 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3669 vd->vdev_children > 1) {
3671 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
3672 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
3677 for (int c = 0; c < vd->vdev_children; c++) {
3678 if (!vdev_is_bootable(vd->vdev_child[c]))
3681 #endif /* illumos */
3686 vdev_is_concrete(vdev_t *vd)
3688 vdev_ops_t *ops = vd->vdev_ops;
3689 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
3690 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
3698 * Load the state from the original vdev tree (ovd) which
3699 * we've retrieved from the MOS config object. If the original
3700 * vdev was offline or faulted then we transfer that state to the
3701 * device in the current vdev tree (nvd).
3704 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3706 spa_t *spa = nvd->vdev_spa;
3708 ASSERT(nvd->vdev_top->vdev_islog);
3709 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3710 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3712 for (int c = 0; c < nvd->vdev_children; c++)
3713 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3715 if (nvd->vdev_ops->vdev_op_leaf) {
3717 * Restore the persistent vdev state
3719 nvd->vdev_offline = ovd->vdev_offline;
3720 nvd->vdev_faulted = ovd->vdev_faulted;
3721 nvd->vdev_degraded = ovd->vdev_degraded;
3722 nvd->vdev_removed = ovd->vdev_removed;
3727 * Determine if a log device has valid content. If the vdev was
3728 * removed or faulted in the MOS config then we know that
3729 * the content on the log device has already been written to the pool.
3732 vdev_log_state_valid(vdev_t *vd)
3734 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3738 for (int c = 0; c < vd->vdev_children; c++)
3739 if (vdev_log_state_valid(vd->vdev_child[c]))
3746 * Expand a vdev if possible.
3749 vdev_expand(vdev_t *vd, uint64_t txg)
3751 ASSERT(vd->vdev_top == vd);
3752 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3754 vdev_set_deflate_ratio(vd);
3756 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
3757 vdev_is_concrete(vd)) {
3758 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3759 vdev_config_dirty(vd);
3767 vdev_split(vdev_t *vd)
3769 vdev_t *cvd, *pvd = vd->vdev_parent;
3771 vdev_remove_child(pvd, vd);
3772 vdev_compact_children(pvd);
3774 cvd = pvd->vdev_child[0];
3775 if (pvd->vdev_children == 1) {
3776 vdev_remove_parent(cvd);
3777 cvd->vdev_splitting = B_TRUE;
3779 vdev_propagate_state(cvd);
3783 vdev_deadman(vdev_t *vd)
3785 for (int c = 0; c < vd->vdev_children; c++) {
3786 vdev_t *cvd = vd->vdev_child[c];
3791 if (vd->vdev_ops->vdev_op_leaf) {
3792 vdev_queue_t *vq = &vd->vdev_queue;
3794 mutex_enter(&vq->vq_lock);
3795 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3796 spa_t *spa = vd->vdev_spa;
3801 * Look at the head of all the pending queues,
3802 * if any I/O has been outstanding for longer than
3803 * the spa_deadman_synctime we panic the system.
3805 fio = avl_first(&vq->vq_active_tree);
3806 delta = gethrtime() - fio->io_timestamp;
3807 if (delta > spa_deadman_synctime(spa)) {
3808 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3809 "delta %lluns, last io %lluns",
3810 fio->io_timestamp, delta,
3811 vq->vq_io_complete_ts);
3812 fm_panic("I/O to pool '%s' appears to be "
3813 "hung on vdev guid %llu at '%s'.",
3815 (long long unsigned int) vd->vdev_guid,
3819 mutex_exit(&vq->vq_lock);