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, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
30 #include <sys/zfs_context.h>
31 #include <sys/fm/fs/zfs.h>
33 #include <sys/spa_impl.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/uberblock_impl.h>
38 #include <sys/metaslab.h>
39 #include <sys/metaslab_impl.h>
40 #include <sys/space_map.h>
41 #include <sys/space_reftree.h>
44 #include <sys/fs/zfs.h>
47 #include <sys/dsl_scan.h>
48 #include <sys/trim_map.h>
50 SYSCTL_DECL(_vfs_zfs);
51 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
54 * Virtual device management.
58 * The limit for ZFS to automatically increase a top-level vdev's ashift
59 * from logical ashift to physical ashift.
61 * Example: one or more 512B emulation child vdevs
62 * child->vdev_ashift = 9 (512 bytes)
63 * child->vdev_physical_ashift = 12 (4096 bytes)
64 * zfs_max_auto_ashift = 11 (2048 bytes)
65 * zfs_min_auto_ashift = 9 (512 bytes)
67 * On pool creation or the addition of a new top-level vdev, ZFS will
68 * increase the ashift of the top-level vdev to 2048 as limited by
69 * zfs_max_auto_ashift.
71 * Example: one or more 512B emulation child vdevs
72 * child->vdev_ashift = 9 (512 bytes)
73 * child->vdev_physical_ashift = 12 (4096 bytes)
74 * zfs_max_auto_ashift = 13 (8192 bytes)
75 * zfs_min_auto_ashift = 9 (512 bytes)
77 * On pool creation or the addition of a new top-level vdev, ZFS will
78 * increase the ashift of the top-level vdev to 4096 to match the
79 * max vdev_physical_ashift.
81 * Example: one or more 512B emulation child vdevs
82 * child->vdev_ashift = 9 (512 bytes)
83 * child->vdev_physical_ashift = 9 (512 bytes)
84 * zfs_max_auto_ashift = 13 (8192 bytes)
85 * zfs_min_auto_ashift = 12 (4096 bytes)
87 * On pool creation or the addition of a new top-level vdev, ZFS will
88 * increase the ashift of the top-level vdev to 4096 to match the
89 * zfs_min_auto_ashift.
91 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
92 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
95 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
100 val = zfs_max_auto_ashift;
101 err = sysctl_handle_64(oidp, &val, 0, req);
102 if (err != 0 || req->newptr == NULL)
105 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
108 zfs_max_auto_ashift = val;
112 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
113 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
114 sysctl_vfs_zfs_max_auto_ashift, "QU",
115 "Max ashift used when optimising for logical -> physical sectors size on "
116 "new top-level vdevs.");
119 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
124 val = zfs_min_auto_ashift;
125 err = sysctl_handle_64(oidp, &val, 0, req);
126 if (err != 0 || req->newptr == NULL)
129 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
132 zfs_min_auto_ashift = val;
136 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
137 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
138 sysctl_vfs_zfs_min_auto_ashift, "QU",
139 "Min ashift used when creating new top-level vdevs.");
141 static vdev_ops_t *vdev_ops_table[] = {
160 * When a vdev is added, it will be divided into approximately (but no
161 * more than) this number of metaslabs.
163 int metaslabs_per_vdev = 200;
164 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
165 &metaslabs_per_vdev, 0,
166 "When a vdev is added, how many metaslabs the vdev should be divided into");
169 * Given a vdev type, return the appropriate ops vector.
172 vdev_getops(const char *type)
174 vdev_ops_t *ops, **opspp;
176 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
177 if (strcmp(ops->vdev_op_type, type) == 0)
184 * Default asize function: return the MAX of psize with the asize of
185 * all children. This is what's used by anything other than RAID-Z.
188 vdev_default_asize(vdev_t *vd, uint64_t psize)
190 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
193 for (int c = 0; c < vd->vdev_children; c++) {
194 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
195 asize = MAX(asize, csize);
202 * Get the minimum allocatable size. We define the allocatable size as
203 * the vdev's asize rounded to the nearest metaslab. This allows us to
204 * replace or attach devices which don't have the same physical size but
205 * can still satisfy the same number of allocations.
208 vdev_get_min_asize(vdev_t *vd)
210 vdev_t *pvd = vd->vdev_parent;
213 * If our parent is NULL (inactive spare or cache) or is the root,
214 * just return our own asize.
217 return (vd->vdev_asize);
220 * The top-level vdev just returns the allocatable size rounded
221 * to the nearest metaslab.
223 if (vd == vd->vdev_top)
224 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
227 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
228 * so each child must provide at least 1/Nth of its asize.
230 if (pvd->vdev_ops == &vdev_raidz_ops)
231 return (pvd->vdev_min_asize / pvd->vdev_children);
233 return (pvd->vdev_min_asize);
237 vdev_set_min_asize(vdev_t *vd)
239 vd->vdev_min_asize = vdev_get_min_asize(vd);
241 for (int c = 0; c < vd->vdev_children; c++)
242 vdev_set_min_asize(vd->vdev_child[c]);
246 vdev_lookup_top(spa_t *spa, uint64_t vdev)
248 vdev_t *rvd = spa->spa_root_vdev;
250 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
252 if (vdev < rvd->vdev_children) {
253 ASSERT(rvd->vdev_child[vdev] != NULL);
254 return (rvd->vdev_child[vdev]);
261 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
265 if (vd->vdev_guid == guid)
268 for (int c = 0; c < vd->vdev_children; c++)
269 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
277 vdev_count_leaves_impl(vdev_t *vd)
281 if (vd->vdev_ops->vdev_op_leaf)
284 for (int c = 0; c < vd->vdev_children; c++)
285 n += vdev_count_leaves_impl(vd->vdev_child[c]);
291 vdev_count_leaves(spa_t *spa)
293 return (vdev_count_leaves_impl(spa->spa_root_vdev));
297 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
299 size_t oldsize, newsize;
300 uint64_t id = cvd->vdev_id;
302 spa_t *spa = cvd->vdev_spa;
304 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
305 ASSERT(cvd->vdev_parent == NULL);
307 cvd->vdev_parent = pvd;
312 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
314 oldsize = pvd->vdev_children * sizeof (vdev_t *);
315 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
316 newsize = pvd->vdev_children * sizeof (vdev_t *);
318 newchild = kmem_zalloc(newsize, KM_SLEEP);
319 if (pvd->vdev_child != NULL) {
320 bcopy(pvd->vdev_child, newchild, oldsize);
321 kmem_free(pvd->vdev_child, oldsize);
324 pvd->vdev_child = newchild;
325 pvd->vdev_child[id] = cvd;
327 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
328 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
331 * Walk up all ancestors to update guid sum.
333 for (; pvd != NULL; pvd = pvd->vdev_parent)
334 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
338 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
341 uint_t id = cvd->vdev_id;
343 ASSERT(cvd->vdev_parent == pvd);
348 ASSERT(id < pvd->vdev_children);
349 ASSERT(pvd->vdev_child[id] == cvd);
351 pvd->vdev_child[id] = NULL;
352 cvd->vdev_parent = NULL;
354 for (c = 0; c < pvd->vdev_children; c++)
355 if (pvd->vdev_child[c])
358 if (c == pvd->vdev_children) {
359 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
360 pvd->vdev_child = NULL;
361 pvd->vdev_children = 0;
365 * Walk up all ancestors to update guid sum.
367 for (; pvd != NULL; pvd = pvd->vdev_parent)
368 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
372 * Remove any holes in the child array.
375 vdev_compact_children(vdev_t *pvd)
377 vdev_t **newchild, *cvd;
378 int oldc = pvd->vdev_children;
381 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
383 for (int c = newc = 0; c < oldc; c++)
384 if (pvd->vdev_child[c])
387 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
389 for (int c = newc = 0; c < oldc; c++) {
390 if ((cvd = pvd->vdev_child[c]) != NULL) {
391 newchild[newc] = cvd;
392 cvd->vdev_id = newc++;
396 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
397 pvd->vdev_child = newchild;
398 pvd->vdev_children = newc;
402 * Allocate and minimally initialize a vdev_t.
405 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
409 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
411 if (spa->spa_root_vdev == NULL) {
412 ASSERT(ops == &vdev_root_ops);
413 spa->spa_root_vdev = vd;
414 spa->spa_load_guid = spa_generate_guid(NULL);
417 if (guid == 0 && ops != &vdev_hole_ops) {
418 if (spa->spa_root_vdev == vd) {
420 * The root vdev's guid will also be the pool guid,
421 * which must be unique among all pools.
423 guid = spa_generate_guid(NULL);
426 * Any other vdev's guid must be unique within the pool.
428 guid = spa_generate_guid(spa);
430 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
435 vd->vdev_guid = guid;
436 vd->vdev_guid_sum = guid;
438 vd->vdev_state = VDEV_STATE_CLOSED;
439 vd->vdev_ishole = (ops == &vdev_hole_ops);
441 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
442 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
443 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
445 for (int t = 0; t < DTL_TYPES; t++) {
446 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
449 txg_list_create(&vd->vdev_ms_list,
450 offsetof(struct metaslab, ms_txg_node));
451 txg_list_create(&vd->vdev_dtl_list,
452 offsetof(struct vdev, vdev_dtl_node));
453 vd->vdev_stat.vs_timestamp = gethrtime();
461 * Allocate a new vdev. The 'alloctype' is used to control whether we are
462 * creating a new vdev or loading an existing one - the behavior is slightly
463 * different for each case.
466 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
471 uint64_t guid = 0, islog, nparity;
474 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
476 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
477 return (SET_ERROR(EINVAL));
479 if ((ops = vdev_getops(type)) == NULL)
480 return (SET_ERROR(EINVAL));
483 * If this is a load, get the vdev guid from the nvlist.
484 * Otherwise, vdev_alloc_common() will generate one for us.
486 if (alloctype == VDEV_ALLOC_LOAD) {
489 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
491 return (SET_ERROR(EINVAL));
493 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
494 return (SET_ERROR(EINVAL));
495 } else if (alloctype == VDEV_ALLOC_SPARE) {
496 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
497 return (SET_ERROR(EINVAL));
498 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
499 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
500 return (SET_ERROR(EINVAL));
501 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
502 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
503 return (SET_ERROR(EINVAL));
507 * The first allocated vdev must be of type 'root'.
509 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
510 return (SET_ERROR(EINVAL));
513 * Determine whether we're a log vdev.
516 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
517 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
518 return (SET_ERROR(ENOTSUP));
520 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
521 return (SET_ERROR(ENOTSUP));
524 * Set the nparity property for RAID-Z vdevs.
527 if (ops == &vdev_raidz_ops) {
528 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
530 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
531 return (SET_ERROR(EINVAL));
533 * Previous versions could only support 1 or 2 parity
537 spa_version(spa) < SPA_VERSION_RAIDZ2)
538 return (SET_ERROR(ENOTSUP));
540 spa_version(spa) < SPA_VERSION_RAIDZ3)
541 return (SET_ERROR(ENOTSUP));
544 * We require the parity to be specified for SPAs that
545 * support multiple parity levels.
547 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
548 return (SET_ERROR(EINVAL));
550 * Otherwise, we default to 1 parity device for RAID-Z.
557 ASSERT(nparity != -1ULL);
559 vd = vdev_alloc_common(spa, id, guid, ops);
561 vd->vdev_islog = islog;
562 vd->vdev_nparity = nparity;
564 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
565 vd->vdev_path = spa_strdup(vd->vdev_path);
566 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
567 vd->vdev_devid = spa_strdup(vd->vdev_devid);
568 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
569 &vd->vdev_physpath) == 0)
570 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
571 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
572 vd->vdev_fru = spa_strdup(vd->vdev_fru);
575 * Set the whole_disk property. If it's not specified, leave the value
578 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
579 &vd->vdev_wholedisk) != 0)
580 vd->vdev_wholedisk = -1ULL;
583 * Look for the 'not present' flag. This will only be set if the device
584 * was not present at the time of import.
586 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
587 &vd->vdev_not_present);
590 * Get the alignment requirement.
592 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
595 * Retrieve the vdev creation time.
597 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
601 * If we're a top-level vdev, try to load the allocation parameters.
603 if (parent && !parent->vdev_parent &&
604 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
605 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
607 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
609 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
611 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
615 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
616 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
617 alloctype == VDEV_ALLOC_ADD ||
618 alloctype == VDEV_ALLOC_SPLIT ||
619 alloctype == VDEV_ALLOC_ROOTPOOL);
620 vd->vdev_mg = metaslab_group_create(islog ?
621 spa_log_class(spa) : spa_normal_class(spa), vd);
625 * If we're a leaf vdev, try to load the DTL object and other state.
627 if (vd->vdev_ops->vdev_op_leaf &&
628 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
629 alloctype == VDEV_ALLOC_ROOTPOOL)) {
630 if (alloctype == VDEV_ALLOC_LOAD) {
631 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
632 &vd->vdev_dtl_object);
633 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
637 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
640 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
641 &spare) == 0 && spare)
645 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
648 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
649 &vd->vdev_resilver_txg);
652 * When importing a pool, we want to ignore the persistent fault
653 * state, as the diagnosis made on another system may not be
654 * valid in the current context. Local vdevs will
655 * remain in the faulted state.
657 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
658 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
660 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
662 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
665 if (vd->vdev_faulted || vd->vdev_degraded) {
669 VDEV_AUX_ERR_EXCEEDED;
670 if (nvlist_lookup_string(nv,
671 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
672 strcmp(aux, "external") == 0)
673 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
679 * Add ourselves to the parent's list of children.
681 vdev_add_child(parent, vd);
689 vdev_free(vdev_t *vd)
691 spa_t *spa = vd->vdev_spa;
694 * vdev_free() implies closing the vdev first. This is simpler than
695 * trying to ensure complicated semantics for all callers.
699 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
700 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
705 for (int c = 0; c < vd->vdev_children; c++)
706 vdev_free(vd->vdev_child[c]);
708 ASSERT(vd->vdev_child == NULL);
709 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
712 * Discard allocation state.
714 if (vd->vdev_mg != NULL) {
715 vdev_metaslab_fini(vd);
716 metaslab_group_destroy(vd->vdev_mg);
719 ASSERT0(vd->vdev_stat.vs_space);
720 ASSERT0(vd->vdev_stat.vs_dspace);
721 ASSERT0(vd->vdev_stat.vs_alloc);
724 * Remove this vdev from its parent's child list.
726 vdev_remove_child(vd->vdev_parent, vd);
728 ASSERT(vd->vdev_parent == NULL);
731 * Clean up vdev structure.
737 spa_strfree(vd->vdev_path);
739 spa_strfree(vd->vdev_devid);
740 if (vd->vdev_physpath)
741 spa_strfree(vd->vdev_physpath);
743 spa_strfree(vd->vdev_fru);
745 if (vd->vdev_isspare)
746 spa_spare_remove(vd);
747 if (vd->vdev_isl2cache)
748 spa_l2cache_remove(vd);
750 txg_list_destroy(&vd->vdev_ms_list);
751 txg_list_destroy(&vd->vdev_dtl_list);
753 mutex_enter(&vd->vdev_dtl_lock);
754 space_map_close(vd->vdev_dtl_sm);
755 for (int t = 0; t < DTL_TYPES; t++) {
756 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
757 range_tree_destroy(vd->vdev_dtl[t]);
759 mutex_exit(&vd->vdev_dtl_lock);
761 mutex_destroy(&vd->vdev_queue_lock);
762 mutex_destroy(&vd->vdev_dtl_lock);
763 mutex_destroy(&vd->vdev_stat_lock);
764 mutex_destroy(&vd->vdev_probe_lock);
766 if (vd == spa->spa_root_vdev)
767 spa->spa_root_vdev = NULL;
769 kmem_free(vd, sizeof (vdev_t));
773 * Transfer top-level vdev state from svd to tvd.
776 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
778 spa_t *spa = svd->vdev_spa;
783 ASSERT(tvd == tvd->vdev_top);
785 tvd->vdev_ms_array = svd->vdev_ms_array;
786 tvd->vdev_ms_shift = svd->vdev_ms_shift;
787 tvd->vdev_ms_count = svd->vdev_ms_count;
789 svd->vdev_ms_array = 0;
790 svd->vdev_ms_shift = 0;
791 svd->vdev_ms_count = 0;
794 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
795 tvd->vdev_mg = svd->vdev_mg;
796 tvd->vdev_ms = svd->vdev_ms;
801 if (tvd->vdev_mg != NULL)
802 tvd->vdev_mg->mg_vd = tvd;
804 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
805 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
806 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
808 svd->vdev_stat.vs_alloc = 0;
809 svd->vdev_stat.vs_space = 0;
810 svd->vdev_stat.vs_dspace = 0;
812 for (t = 0; t < TXG_SIZE; t++) {
813 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
814 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
815 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
816 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
817 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
818 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
821 if (list_link_active(&svd->vdev_config_dirty_node)) {
822 vdev_config_clean(svd);
823 vdev_config_dirty(tvd);
826 if (list_link_active(&svd->vdev_state_dirty_node)) {
827 vdev_state_clean(svd);
828 vdev_state_dirty(tvd);
831 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
832 svd->vdev_deflate_ratio = 0;
834 tvd->vdev_islog = svd->vdev_islog;
839 vdev_top_update(vdev_t *tvd, vdev_t *vd)
846 for (int c = 0; c < vd->vdev_children; c++)
847 vdev_top_update(tvd, vd->vdev_child[c]);
851 * Add a mirror/replacing vdev above an existing vdev.
854 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
856 spa_t *spa = cvd->vdev_spa;
857 vdev_t *pvd = cvd->vdev_parent;
860 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
862 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
864 mvd->vdev_asize = cvd->vdev_asize;
865 mvd->vdev_min_asize = cvd->vdev_min_asize;
866 mvd->vdev_max_asize = cvd->vdev_max_asize;
867 mvd->vdev_ashift = cvd->vdev_ashift;
868 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
869 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
870 mvd->vdev_state = cvd->vdev_state;
871 mvd->vdev_crtxg = cvd->vdev_crtxg;
873 vdev_remove_child(pvd, cvd);
874 vdev_add_child(pvd, mvd);
875 cvd->vdev_id = mvd->vdev_children;
876 vdev_add_child(mvd, cvd);
877 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
879 if (mvd == mvd->vdev_top)
880 vdev_top_transfer(cvd, mvd);
886 * Remove a 1-way mirror/replacing vdev from the tree.
889 vdev_remove_parent(vdev_t *cvd)
891 vdev_t *mvd = cvd->vdev_parent;
892 vdev_t *pvd = mvd->vdev_parent;
894 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
896 ASSERT(mvd->vdev_children == 1);
897 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
898 mvd->vdev_ops == &vdev_replacing_ops ||
899 mvd->vdev_ops == &vdev_spare_ops);
900 cvd->vdev_ashift = mvd->vdev_ashift;
901 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
902 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
904 vdev_remove_child(mvd, cvd);
905 vdev_remove_child(pvd, mvd);
908 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
909 * Otherwise, we could have detached an offline device, and when we
910 * go to import the pool we'll think we have two top-level vdevs,
911 * instead of a different version of the same top-level vdev.
913 if (mvd->vdev_top == mvd) {
914 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
915 cvd->vdev_orig_guid = cvd->vdev_guid;
916 cvd->vdev_guid += guid_delta;
917 cvd->vdev_guid_sum += guid_delta;
919 cvd->vdev_id = mvd->vdev_id;
920 vdev_add_child(pvd, cvd);
921 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
923 if (cvd == cvd->vdev_top)
924 vdev_top_transfer(mvd, cvd);
926 ASSERT(mvd->vdev_children == 0);
931 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
933 spa_t *spa = vd->vdev_spa;
934 objset_t *mos = spa->spa_meta_objset;
936 uint64_t oldc = vd->vdev_ms_count;
937 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
941 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
944 * This vdev is not being allocated from yet or is a hole.
946 if (vd->vdev_ms_shift == 0)
949 ASSERT(!vd->vdev_ishole);
952 * Compute the raidz-deflation ratio. Note, we hard-code
953 * in 128k (1 << 17) because it is the "typical" blocksize.
954 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
955 * otherwise it would inconsistently account for existing bp's.
957 vd->vdev_deflate_ratio = (1 << 17) /
958 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
960 ASSERT(oldc <= newc);
962 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
965 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
966 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
970 vd->vdev_ms_count = newc;
972 for (m = oldc; m < newc; m++) {
976 error = dmu_read(mos, vd->vdev_ms_array,
977 m * sizeof (uint64_t), sizeof (uint64_t), &object,
983 error = metaslab_init(vd->vdev_mg, m, object, txg,
990 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
993 * If the vdev is being removed we don't activate
994 * the metaslabs since we want to ensure that no new
995 * allocations are performed on this device.
997 if (oldc == 0 && !vd->vdev_removing)
998 metaslab_group_activate(vd->vdev_mg);
1001 spa_config_exit(spa, SCL_ALLOC, FTAG);
1007 vdev_metaslab_fini(vdev_t *vd)
1010 uint64_t count = vd->vdev_ms_count;
1012 if (vd->vdev_ms != NULL) {
1013 metaslab_group_passivate(vd->vdev_mg);
1014 for (m = 0; m < count; m++) {
1015 metaslab_t *msp = vd->vdev_ms[m];
1020 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1025 typedef struct vdev_probe_stats {
1026 boolean_t vps_readable;
1027 boolean_t vps_writeable;
1029 } vdev_probe_stats_t;
1032 vdev_probe_done(zio_t *zio)
1034 spa_t *spa = zio->io_spa;
1035 vdev_t *vd = zio->io_vd;
1036 vdev_probe_stats_t *vps = zio->io_private;
1038 ASSERT(vd->vdev_probe_zio != NULL);
1040 if (zio->io_type == ZIO_TYPE_READ) {
1041 if (zio->io_error == 0)
1042 vps->vps_readable = 1;
1043 if (zio->io_error == 0 && spa_writeable(spa)) {
1044 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1045 zio->io_offset, zio->io_size, zio->io_data,
1046 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1047 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1049 zio_buf_free(zio->io_data, zio->io_size);
1051 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1052 if (zio->io_error == 0)
1053 vps->vps_writeable = 1;
1054 zio_buf_free(zio->io_data, zio->io_size);
1055 } else if (zio->io_type == ZIO_TYPE_NULL) {
1058 vd->vdev_cant_read |= !vps->vps_readable;
1059 vd->vdev_cant_write |= !vps->vps_writeable;
1061 if (vdev_readable(vd) &&
1062 (vdev_writeable(vd) || !spa_writeable(spa))) {
1065 ASSERT(zio->io_error != 0);
1066 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1067 spa, vd, NULL, 0, 0);
1068 zio->io_error = SET_ERROR(ENXIO);
1071 mutex_enter(&vd->vdev_probe_lock);
1072 ASSERT(vd->vdev_probe_zio == zio);
1073 vd->vdev_probe_zio = NULL;
1074 mutex_exit(&vd->vdev_probe_lock);
1076 zio_link_t *zl = NULL;
1077 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1078 if (!vdev_accessible(vd, pio))
1079 pio->io_error = SET_ERROR(ENXIO);
1081 kmem_free(vps, sizeof (*vps));
1086 * Determine whether this device is accessible.
1088 * Read and write to several known locations: the pad regions of each
1089 * vdev label but the first, which we leave alone in case it contains
1093 vdev_probe(vdev_t *vd, zio_t *zio)
1095 spa_t *spa = vd->vdev_spa;
1096 vdev_probe_stats_t *vps = NULL;
1099 ASSERT(vd->vdev_ops->vdev_op_leaf);
1102 * Don't probe the probe.
1104 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1108 * To prevent 'probe storms' when a device fails, we create
1109 * just one probe i/o at a time. All zios that want to probe
1110 * this vdev will become parents of the probe io.
1112 mutex_enter(&vd->vdev_probe_lock);
1114 if ((pio = vd->vdev_probe_zio) == NULL) {
1115 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1117 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1118 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1121 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1123 * vdev_cant_read and vdev_cant_write can only
1124 * transition from TRUE to FALSE when we have the
1125 * SCL_ZIO lock as writer; otherwise they can only
1126 * transition from FALSE to TRUE. This ensures that
1127 * any zio looking at these values can assume that
1128 * failures persist for the life of the I/O. That's
1129 * important because when a device has intermittent
1130 * connectivity problems, we want to ensure that
1131 * they're ascribed to the device (ENXIO) and not
1134 * Since we hold SCL_ZIO as writer here, clear both
1135 * values so the probe can reevaluate from first
1138 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1139 vd->vdev_cant_read = B_FALSE;
1140 vd->vdev_cant_write = B_FALSE;
1143 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1144 vdev_probe_done, vps,
1145 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1148 * We can't change the vdev state in this context, so we
1149 * kick off an async task to do it on our behalf.
1152 vd->vdev_probe_wanted = B_TRUE;
1153 spa_async_request(spa, SPA_ASYNC_PROBE);
1158 zio_add_child(zio, pio);
1160 mutex_exit(&vd->vdev_probe_lock);
1163 ASSERT(zio != NULL);
1167 for (int l = 1; l < VDEV_LABELS; l++) {
1168 zio_nowait(zio_read_phys(pio, vd,
1169 vdev_label_offset(vd->vdev_psize, l,
1170 offsetof(vdev_label_t, vl_pad2)),
1171 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1172 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1173 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1184 vdev_open_child(void *arg)
1188 vd->vdev_open_thread = curthread;
1189 vd->vdev_open_error = vdev_open(vd);
1190 vd->vdev_open_thread = NULL;
1194 vdev_uses_zvols(vdev_t *vd)
1196 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1197 strlen(ZVOL_DIR)) == 0)
1199 for (int c = 0; c < vd->vdev_children; c++)
1200 if (vdev_uses_zvols(vd->vdev_child[c]))
1206 vdev_open_children(vdev_t *vd)
1209 int children = vd->vdev_children;
1212 * in order to handle pools on top of zvols, do the opens
1213 * in a single thread so that the same thread holds the
1214 * spa_namespace_lock
1216 if (B_TRUE || vdev_uses_zvols(vd)) {
1217 for (int c = 0; c < children; c++)
1218 vd->vdev_child[c]->vdev_open_error =
1219 vdev_open(vd->vdev_child[c]);
1222 tq = taskq_create("vdev_open", children, minclsyspri,
1223 children, children, TASKQ_PREPOPULATE);
1225 for (int c = 0; c < children; c++)
1226 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1233 * Prepare a virtual device for access.
1236 vdev_open(vdev_t *vd)
1238 spa_t *spa = vd->vdev_spa;
1241 uint64_t max_osize = 0;
1242 uint64_t asize, max_asize, psize;
1243 uint64_t logical_ashift = 0;
1244 uint64_t physical_ashift = 0;
1246 ASSERT(vd->vdev_open_thread == curthread ||
1247 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1248 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1249 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1250 vd->vdev_state == VDEV_STATE_OFFLINE);
1252 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1253 vd->vdev_cant_read = B_FALSE;
1254 vd->vdev_cant_write = B_FALSE;
1255 vd->vdev_notrim = B_FALSE;
1256 vd->vdev_min_asize = vdev_get_min_asize(vd);
1259 * If this vdev is not removed, check its fault status. If it's
1260 * faulted, bail out of the open.
1262 if (!vd->vdev_removed && vd->vdev_faulted) {
1263 ASSERT(vd->vdev_children == 0);
1264 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1265 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1266 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1267 vd->vdev_label_aux);
1268 return (SET_ERROR(ENXIO));
1269 } else if (vd->vdev_offline) {
1270 ASSERT(vd->vdev_children == 0);
1271 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1272 return (SET_ERROR(ENXIO));
1275 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1276 &logical_ashift, &physical_ashift);
1279 * Reset the vdev_reopening flag so that we actually close
1280 * the vdev on error.
1282 vd->vdev_reopening = B_FALSE;
1283 if (zio_injection_enabled && error == 0)
1284 error = zio_handle_device_injection(vd, NULL, ENXIO);
1287 if (vd->vdev_removed &&
1288 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1289 vd->vdev_removed = B_FALSE;
1291 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1292 vd->vdev_stat.vs_aux);
1296 vd->vdev_removed = B_FALSE;
1299 * Recheck the faulted flag now that we have confirmed that
1300 * the vdev is accessible. If we're faulted, bail.
1302 if (vd->vdev_faulted) {
1303 ASSERT(vd->vdev_children == 0);
1304 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1305 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1306 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1307 vd->vdev_label_aux);
1308 return (SET_ERROR(ENXIO));
1311 if (vd->vdev_degraded) {
1312 ASSERT(vd->vdev_children == 0);
1313 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1314 VDEV_AUX_ERR_EXCEEDED);
1316 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1320 * For hole or missing vdevs we just return success.
1322 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1325 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1326 trim_map_create(vd);
1328 for (int c = 0; c < vd->vdev_children; c++) {
1329 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1330 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1336 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1337 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1339 if (vd->vdev_children == 0) {
1340 if (osize < SPA_MINDEVSIZE) {
1341 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1342 VDEV_AUX_TOO_SMALL);
1343 return (SET_ERROR(EOVERFLOW));
1346 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1347 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1348 VDEV_LABEL_END_SIZE);
1350 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1351 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1352 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1353 VDEV_AUX_TOO_SMALL);
1354 return (SET_ERROR(EOVERFLOW));
1358 max_asize = max_osize;
1361 vd->vdev_psize = psize;
1364 * Make sure the allocatable size hasn't shrunk.
1366 if (asize < vd->vdev_min_asize) {
1367 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1368 VDEV_AUX_BAD_LABEL);
1369 return (SET_ERROR(EINVAL));
1372 vd->vdev_physical_ashift =
1373 MAX(physical_ashift, vd->vdev_physical_ashift);
1374 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1375 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1377 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1378 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1379 VDEV_AUX_ASHIFT_TOO_BIG);
1383 if (vd->vdev_asize == 0) {
1385 * This is the first-ever open, so use the computed values.
1386 * For testing purposes, a higher ashift can be requested.
1388 vd->vdev_asize = asize;
1389 vd->vdev_max_asize = max_asize;
1392 * Make sure the alignment requirement hasn't increased.
1394 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1395 vd->vdev_ops->vdev_op_leaf) {
1396 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1397 VDEV_AUX_BAD_LABEL);
1400 vd->vdev_max_asize = max_asize;
1404 * If all children are healthy and the asize has increased,
1405 * then we've experienced dynamic LUN growth. If automatic
1406 * expansion is enabled then use the additional space.
1408 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1409 (vd->vdev_expanding || spa->spa_autoexpand))
1410 vd->vdev_asize = asize;
1412 vdev_set_min_asize(vd);
1415 * Ensure we can issue some IO before declaring the
1416 * vdev open for business.
1418 if (vd->vdev_ops->vdev_op_leaf &&
1419 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1420 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1421 VDEV_AUX_ERR_EXCEEDED);
1426 * Track the min and max ashift values for normal data devices.
1428 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1429 !vd->vdev_islog && vd->vdev_aux == NULL) {
1430 if (vd->vdev_ashift > spa->spa_max_ashift)
1431 spa->spa_max_ashift = vd->vdev_ashift;
1432 if (vd->vdev_ashift < spa->spa_min_ashift)
1433 spa->spa_min_ashift = vd->vdev_ashift;
1437 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1438 * resilver. But don't do this if we are doing a reopen for a scrub,
1439 * since this would just restart the scrub we are already doing.
1441 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1442 vdev_resilver_needed(vd, NULL, NULL))
1443 spa_async_request(spa, SPA_ASYNC_RESILVER);
1449 * Called once the vdevs are all opened, this routine validates the label
1450 * contents. This needs to be done before vdev_load() so that we don't
1451 * inadvertently do repair I/Os to the wrong device.
1453 * If 'strict' is false ignore the spa guid check. This is necessary because
1454 * if the machine crashed during a re-guid the new guid might have been written
1455 * to all of the vdev labels, but not the cached config. The strict check
1456 * will be performed when the pool is opened again using the mos config.
1458 * This function will only return failure if one of the vdevs indicates that it
1459 * has since been destroyed or exported. This is only possible if
1460 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1461 * will be updated but the function will return 0.
1464 vdev_validate(vdev_t *vd, boolean_t strict)
1466 spa_t *spa = vd->vdev_spa;
1468 uint64_t guid = 0, top_guid;
1471 for (int c = 0; c < vd->vdev_children; c++)
1472 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1473 return (SET_ERROR(EBADF));
1476 * If the device has already failed, or was marked offline, don't do
1477 * any further validation. Otherwise, label I/O will fail and we will
1478 * overwrite the previous state.
1480 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1481 uint64_t aux_guid = 0;
1483 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1484 spa_last_synced_txg(spa) : -1ULL;
1486 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1487 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1488 VDEV_AUX_BAD_LABEL);
1493 * Determine if this vdev has been split off into another
1494 * pool. If so, then refuse to open it.
1496 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1497 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1498 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1499 VDEV_AUX_SPLIT_POOL);
1504 if (strict && (nvlist_lookup_uint64(label,
1505 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1506 guid != spa_guid(spa))) {
1507 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1508 VDEV_AUX_CORRUPT_DATA);
1513 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1514 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1519 * If this vdev just became a top-level vdev because its
1520 * sibling was detached, it will have adopted the parent's
1521 * vdev guid -- but the label may or may not be on disk yet.
1522 * Fortunately, either version of the label will have the
1523 * same top guid, so if we're a top-level vdev, we can
1524 * safely compare to that instead.
1526 * If we split this vdev off instead, then we also check the
1527 * original pool's guid. We don't want to consider the vdev
1528 * corrupt if it is partway through a split operation.
1530 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1532 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1534 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1535 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1536 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1537 VDEV_AUX_CORRUPT_DATA);
1542 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1544 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1545 VDEV_AUX_CORRUPT_DATA);
1553 * If this is a verbatim import, no need to check the
1554 * state of the pool.
1556 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1557 spa_load_state(spa) == SPA_LOAD_OPEN &&
1558 state != POOL_STATE_ACTIVE)
1559 return (SET_ERROR(EBADF));
1562 * If we were able to open and validate a vdev that was
1563 * previously marked permanently unavailable, clear that state
1566 if (vd->vdev_not_present)
1567 vd->vdev_not_present = 0;
1574 * Close a virtual device.
1577 vdev_close(vdev_t *vd)
1579 spa_t *spa = vd->vdev_spa;
1580 vdev_t *pvd = vd->vdev_parent;
1582 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1585 * If our parent is reopening, then we are as well, unless we are
1588 if (pvd != NULL && pvd->vdev_reopening)
1589 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1591 vd->vdev_ops->vdev_op_close(vd);
1593 vdev_cache_purge(vd);
1595 if (vd->vdev_ops->vdev_op_leaf)
1596 trim_map_destroy(vd);
1599 * We record the previous state before we close it, so that if we are
1600 * doing a reopen(), we don't generate FMA ereports if we notice that
1601 * it's still faulted.
1603 vd->vdev_prevstate = vd->vdev_state;
1605 if (vd->vdev_offline)
1606 vd->vdev_state = VDEV_STATE_OFFLINE;
1608 vd->vdev_state = VDEV_STATE_CLOSED;
1609 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1613 vdev_hold(vdev_t *vd)
1615 spa_t *spa = vd->vdev_spa;
1617 ASSERT(spa_is_root(spa));
1618 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1621 for (int c = 0; c < vd->vdev_children; c++)
1622 vdev_hold(vd->vdev_child[c]);
1624 if (vd->vdev_ops->vdev_op_leaf)
1625 vd->vdev_ops->vdev_op_hold(vd);
1629 vdev_rele(vdev_t *vd)
1631 spa_t *spa = vd->vdev_spa;
1633 ASSERT(spa_is_root(spa));
1634 for (int c = 0; c < vd->vdev_children; c++)
1635 vdev_rele(vd->vdev_child[c]);
1637 if (vd->vdev_ops->vdev_op_leaf)
1638 vd->vdev_ops->vdev_op_rele(vd);
1642 * Reopen all interior vdevs and any unopened leaves. We don't actually
1643 * reopen leaf vdevs which had previously been opened as they might deadlock
1644 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1645 * If the leaf has never been opened then open it, as usual.
1648 vdev_reopen(vdev_t *vd)
1650 spa_t *spa = vd->vdev_spa;
1652 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1654 /* set the reopening flag unless we're taking the vdev offline */
1655 vd->vdev_reopening = !vd->vdev_offline;
1657 (void) vdev_open(vd);
1660 * Call vdev_validate() here to make sure we have the same device.
1661 * Otherwise, a device with an invalid label could be successfully
1662 * opened in response to vdev_reopen().
1665 (void) vdev_validate_aux(vd);
1666 if (vdev_readable(vd) && vdev_writeable(vd) &&
1667 vd->vdev_aux == &spa->spa_l2cache &&
1668 !l2arc_vdev_present(vd))
1669 l2arc_add_vdev(spa, vd);
1671 (void) vdev_validate(vd, B_TRUE);
1675 * Reassess parent vdev's health.
1677 vdev_propagate_state(vd);
1681 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1686 * Normally, partial opens (e.g. of a mirror) are allowed.
1687 * For a create, however, we want to fail the request if
1688 * there are any components we can't open.
1690 error = vdev_open(vd);
1692 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1694 return (error ? error : ENXIO);
1698 * Recursively load DTLs and initialize all labels.
1700 if ((error = vdev_dtl_load(vd)) != 0 ||
1701 (error = vdev_label_init(vd, txg, isreplacing ?
1702 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1711 vdev_metaslab_set_size(vdev_t *vd)
1714 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1716 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1717 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1721 * Maximize performance by inflating the configured ashift for top level
1722 * vdevs to be as close to the physical ashift as possible while maintaining
1723 * administrator defined limits and ensuring it doesn't go below the
1727 vdev_ashift_optimize(vdev_t *vd)
1729 if (vd == vd->vdev_top) {
1730 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1731 vd->vdev_ashift = MIN(
1732 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1733 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1736 * Unusual case where logical ashift > physical ashift
1737 * so we can't cap the calculated ashift based on max
1738 * ashift as that would cause failures.
1739 * We still check if we need to increase it to match
1742 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1749 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1751 ASSERT(vd == vd->vdev_top);
1752 ASSERT(!vd->vdev_ishole);
1753 ASSERT(ISP2(flags));
1754 ASSERT(spa_writeable(vd->vdev_spa));
1756 if (flags & VDD_METASLAB)
1757 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1759 if (flags & VDD_DTL)
1760 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1762 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1766 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1768 for (int c = 0; c < vd->vdev_children; c++)
1769 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1771 if (vd->vdev_ops->vdev_op_leaf)
1772 vdev_dirty(vd->vdev_top, flags, vd, txg);
1778 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1779 * the vdev has less than perfect replication. There are four kinds of DTL:
1781 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1783 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1785 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1786 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1787 * txgs that was scrubbed.
1789 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1790 * persistent errors or just some device being offline.
1791 * Unlike the other three, the DTL_OUTAGE map is not generally
1792 * maintained; it's only computed when needed, typically to
1793 * determine whether a device can be detached.
1795 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1796 * either has the data or it doesn't.
1798 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1799 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1800 * if any child is less than fully replicated, then so is its parent.
1801 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1802 * comprising only those txgs which appear in 'maxfaults' or more children;
1803 * those are the txgs we don't have enough replication to read. For example,
1804 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1805 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1806 * two child DTL_MISSING maps.
1808 * It should be clear from the above that to compute the DTLs and outage maps
1809 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1810 * Therefore, that is all we keep on disk. When loading the pool, or after
1811 * a configuration change, we generate all other DTLs from first principles.
1814 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1816 range_tree_t *rt = vd->vdev_dtl[t];
1818 ASSERT(t < DTL_TYPES);
1819 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1820 ASSERT(spa_writeable(vd->vdev_spa));
1822 mutex_enter(rt->rt_lock);
1823 if (!range_tree_contains(rt, txg, size))
1824 range_tree_add(rt, txg, size);
1825 mutex_exit(rt->rt_lock);
1829 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1831 range_tree_t *rt = vd->vdev_dtl[t];
1832 boolean_t dirty = B_FALSE;
1834 ASSERT(t < DTL_TYPES);
1835 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1837 mutex_enter(rt->rt_lock);
1838 if (range_tree_space(rt) != 0)
1839 dirty = range_tree_contains(rt, txg, size);
1840 mutex_exit(rt->rt_lock);
1846 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1848 range_tree_t *rt = vd->vdev_dtl[t];
1851 mutex_enter(rt->rt_lock);
1852 empty = (range_tree_space(rt) == 0);
1853 mutex_exit(rt->rt_lock);
1859 * Returns the lowest txg in the DTL range.
1862 vdev_dtl_min(vdev_t *vd)
1866 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1867 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1868 ASSERT0(vd->vdev_children);
1870 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1871 return (rs->rs_start - 1);
1875 * Returns the highest txg in the DTL.
1878 vdev_dtl_max(vdev_t *vd)
1882 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1883 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1884 ASSERT0(vd->vdev_children);
1886 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1887 return (rs->rs_end);
1891 * Determine if a resilvering vdev should remove any DTL entries from
1892 * its range. If the vdev was resilvering for the entire duration of the
1893 * scan then it should excise that range from its DTLs. Otherwise, this
1894 * vdev is considered partially resilvered and should leave its DTL
1895 * entries intact. The comment in vdev_dtl_reassess() describes how we
1899 vdev_dtl_should_excise(vdev_t *vd)
1901 spa_t *spa = vd->vdev_spa;
1902 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1904 ASSERT0(scn->scn_phys.scn_errors);
1905 ASSERT0(vd->vdev_children);
1907 if (vd->vdev_resilver_txg == 0 ||
1908 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1912 * When a resilver is initiated the scan will assign the scn_max_txg
1913 * value to the highest txg value that exists in all DTLs. If this
1914 * device's max DTL is not part of this scan (i.e. it is not in
1915 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1918 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1919 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1920 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1921 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1928 * Reassess DTLs after a config change or scrub completion.
1931 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1933 spa_t *spa = vd->vdev_spa;
1937 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1939 for (int c = 0; c < vd->vdev_children; c++)
1940 vdev_dtl_reassess(vd->vdev_child[c], txg,
1941 scrub_txg, scrub_done);
1943 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1946 if (vd->vdev_ops->vdev_op_leaf) {
1947 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1949 mutex_enter(&vd->vdev_dtl_lock);
1952 * If we've completed a scan cleanly then determine
1953 * if this vdev should remove any DTLs. We only want to
1954 * excise regions on vdevs that were available during
1955 * the entire duration of this scan.
1957 if (scrub_txg != 0 &&
1958 (spa->spa_scrub_started ||
1959 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1960 vdev_dtl_should_excise(vd)) {
1962 * We completed a scrub up to scrub_txg. If we
1963 * did it without rebooting, then the scrub dtl
1964 * will be valid, so excise the old region and
1965 * fold in the scrub dtl. Otherwise, leave the
1966 * dtl as-is if there was an error.
1968 * There's little trick here: to excise the beginning
1969 * of the DTL_MISSING map, we put it into a reference
1970 * tree and then add a segment with refcnt -1 that
1971 * covers the range [0, scrub_txg). This means
1972 * that each txg in that range has refcnt -1 or 0.
1973 * We then add DTL_SCRUB with a refcnt of 2, so that
1974 * entries in the range [0, scrub_txg) will have a
1975 * positive refcnt -- either 1 or 2. We then convert
1976 * the reference tree into the new DTL_MISSING map.
1978 space_reftree_create(&reftree);
1979 space_reftree_add_map(&reftree,
1980 vd->vdev_dtl[DTL_MISSING], 1);
1981 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1982 space_reftree_add_map(&reftree,
1983 vd->vdev_dtl[DTL_SCRUB], 2);
1984 space_reftree_generate_map(&reftree,
1985 vd->vdev_dtl[DTL_MISSING], 1);
1986 space_reftree_destroy(&reftree);
1988 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1989 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1990 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1992 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1993 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1994 if (!vdev_readable(vd))
1995 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1997 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1998 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2001 * If the vdev was resilvering and no longer has any
2002 * DTLs then reset its resilvering flag and dirty
2003 * the top level so that we persist the change.
2005 if (vd->vdev_resilver_txg != 0 &&
2006 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
2007 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
2008 vd->vdev_resilver_txg = 0;
2009 vdev_config_dirty(vd->vdev_top);
2012 mutex_exit(&vd->vdev_dtl_lock);
2015 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2019 mutex_enter(&vd->vdev_dtl_lock);
2020 for (int t = 0; t < DTL_TYPES; t++) {
2021 /* account for child's outage in parent's missing map */
2022 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2024 continue; /* leaf vdevs only */
2025 if (t == DTL_PARTIAL)
2026 minref = 1; /* i.e. non-zero */
2027 else if (vd->vdev_nparity != 0)
2028 minref = vd->vdev_nparity + 1; /* RAID-Z */
2030 minref = vd->vdev_children; /* any kind of mirror */
2031 space_reftree_create(&reftree);
2032 for (int c = 0; c < vd->vdev_children; c++) {
2033 vdev_t *cvd = vd->vdev_child[c];
2034 mutex_enter(&cvd->vdev_dtl_lock);
2035 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2036 mutex_exit(&cvd->vdev_dtl_lock);
2038 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2039 space_reftree_destroy(&reftree);
2041 mutex_exit(&vd->vdev_dtl_lock);
2045 vdev_dtl_load(vdev_t *vd)
2047 spa_t *spa = vd->vdev_spa;
2048 objset_t *mos = spa->spa_meta_objset;
2051 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2052 ASSERT(!vd->vdev_ishole);
2054 error = space_map_open(&vd->vdev_dtl_sm, mos,
2055 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2058 ASSERT(vd->vdev_dtl_sm != NULL);
2060 mutex_enter(&vd->vdev_dtl_lock);
2063 * Now that we've opened the space_map we need to update
2066 space_map_update(vd->vdev_dtl_sm);
2068 error = space_map_load(vd->vdev_dtl_sm,
2069 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2070 mutex_exit(&vd->vdev_dtl_lock);
2075 for (int c = 0; c < vd->vdev_children; c++) {
2076 error = vdev_dtl_load(vd->vdev_child[c]);
2085 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2087 spa_t *spa = vd->vdev_spa;
2088 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2089 objset_t *mos = spa->spa_meta_objset;
2090 range_tree_t *rtsync;
2093 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2095 ASSERT(!vd->vdev_ishole);
2096 ASSERT(vd->vdev_ops->vdev_op_leaf);
2098 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2100 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2101 mutex_enter(&vd->vdev_dtl_lock);
2102 space_map_free(vd->vdev_dtl_sm, tx);
2103 space_map_close(vd->vdev_dtl_sm);
2104 vd->vdev_dtl_sm = NULL;
2105 mutex_exit(&vd->vdev_dtl_lock);
2110 if (vd->vdev_dtl_sm == NULL) {
2111 uint64_t new_object;
2113 new_object = space_map_alloc(mos, tx);
2114 VERIFY3U(new_object, !=, 0);
2116 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2117 0, -1ULL, 0, &vd->vdev_dtl_lock));
2118 ASSERT(vd->vdev_dtl_sm != NULL);
2121 bzero(&rtlock, sizeof(rtlock));
2122 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2124 rtsync = range_tree_create(NULL, NULL, &rtlock);
2126 mutex_enter(&rtlock);
2128 mutex_enter(&vd->vdev_dtl_lock);
2129 range_tree_walk(rt, range_tree_add, rtsync);
2130 mutex_exit(&vd->vdev_dtl_lock);
2132 space_map_truncate(vd->vdev_dtl_sm, tx);
2133 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2134 range_tree_vacate(rtsync, NULL, NULL);
2136 range_tree_destroy(rtsync);
2138 mutex_exit(&rtlock);
2139 mutex_destroy(&rtlock);
2142 * If the object for the space map has changed then dirty
2143 * the top level so that we update the config.
2145 if (object != space_map_object(vd->vdev_dtl_sm)) {
2146 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2147 "new object %llu", txg, spa_name(spa), object,
2148 space_map_object(vd->vdev_dtl_sm));
2149 vdev_config_dirty(vd->vdev_top);
2154 mutex_enter(&vd->vdev_dtl_lock);
2155 space_map_update(vd->vdev_dtl_sm);
2156 mutex_exit(&vd->vdev_dtl_lock);
2160 * Determine whether the specified vdev can be offlined/detached/removed
2161 * without losing data.
2164 vdev_dtl_required(vdev_t *vd)
2166 spa_t *spa = vd->vdev_spa;
2167 vdev_t *tvd = vd->vdev_top;
2168 uint8_t cant_read = vd->vdev_cant_read;
2171 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2173 if (vd == spa->spa_root_vdev || vd == tvd)
2177 * Temporarily mark the device as unreadable, and then determine
2178 * whether this results in any DTL outages in the top-level vdev.
2179 * If not, we can safely offline/detach/remove the device.
2181 vd->vdev_cant_read = B_TRUE;
2182 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2183 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2184 vd->vdev_cant_read = cant_read;
2185 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2187 if (!required && zio_injection_enabled)
2188 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2194 * Determine if resilver is needed, and if so the txg range.
2197 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2199 boolean_t needed = B_FALSE;
2200 uint64_t thismin = UINT64_MAX;
2201 uint64_t thismax = 0;
2203 if (vd->vdev_children == 0) {
2204 mutex_enter(&vd->vdev_dtl_lock);
2205 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2206 vdev_writeable(vd)) {
2208 thismin = vdev_dtl_min(vd);
2209 thismax = vdev_dtl_max(vd);
2212 mutex_exit(&vd->vdev_dtl_lock);
2214 for (int c = 0; c < vd->vdev_children; c++) {
2215 vdev_t *cvd = vd->vdev_child[c];
2216 uint64_t cmin, cmax;
2218 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2219 thismin = MIN(thismin, cmin);
2220 thismax = MAX(thismax, cmax);
2226 if (needed && minp) {
2234 vdev_load(vdev_t *vd)
2237 * Recursively load all children.
2239 for (int c = 0; c < vd->vdev_children; c++)
2240 vdev_load(vd->vdev_child[c]);
2243 * If this is a top-level vdev, initialize its metaslabs.
2245 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2246 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2247 vdev_metaslab_init(vd, 0) != 0))
2248 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2249 VDEV_AUX_CORRUPT_DATA);
2252 * If this is a leaf vdev, load its DTL.
2254 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2255 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2256 VDEV_AUX_CORRUPT_DATA);
2260 * The special vdev case is used for hot spares and l2cache devices. Its
2261 * sole purpose it to set the vdev state for the associated vdev. To do this,
2262 * we make sure that we can open the underlying device, then try to read the
2263 * label, and make sure that the label is sane and that it hasn't been
2264 * repurposed to another pool.
2267 vdev_validate_aux(vdev_t *vd)
2270 uint64_t guid, version;
2273 if (!vdev_readable(vd))
2276 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2277 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2278 VDEV_AUX_CORRUPT_DATA);
2282 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2283 !SPA_VERSION_IS_SUPPORTED(version) ||
2284 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2285 guid != vd->vdev_guid ||
2286 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2287 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2288 VDEV_AUX_CORRUPT_DATA);
2294 * We don't actually check the pool state here. If it's in fact in
2295 * use by another pool, we update this fact on the fly when requested.
2302 vdev_remove(vdev_t *vd, uint64_t txg)
2304 spa_t *spa = vd->vdev_spa;
2305 objset_t *mos = spa->spa_meta_objset;
2308 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2310 if (vd->vdev_ms != NULL) {
2311 metaslab_group_t *mg = vd->vdev_mg;
2313 metaslab_group_histogram_verify(mg);
2314 metaslab_class_histogram_verify(mg->mg_class);
2316 for (int m = 0; m < vd->vdev_ms_count; m++) {
2317 metaslab_t *msp = vd->vdev_ms[m];
2319 if (msp == NULL || msp->ms_sm == NULL)
2322 mutex_enter(&msp->ms_lock);
2324 * If the metaslab was not loaded when the vdev
2325 * was removed then the histogram accounting may
2326 * not be accurate. Update the histogram information
2327 * here so that we ensure that the metaslab group
2328 * and metaslab class are up-to-date.
2330 metaslab_group_histogram_remove(mg, msp);
2332 VERIFY0(space_map_allocated(msp->ms_sm));
2333 space_map_free(msp->ms_sm, tx);
2334 space_map_close(msp->ms_sm);
2336 mutex_exit(&msp->ms_lock);
2339 metaslab_group_histogram_verify(mg);
2340 metaslab_class_histogram_verify(mg->mg_class);
2341 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2342 ASSERT0(mg->mg_histogram[i]);
2346 if (vd->vdev_ms_array) {
2347 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2348 vd->vdev_ms_array = 0;
2354 vdev_sync_done(vdev_t *vd, uint64_t txg)
2357 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2359 ASSERT(!vd->vdev_ishole);
2361 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2362 metaslab_sync_done(msp, txg);
2365 metaslab_sync_reassess(vd->vdev_mg);
2369 vdev_sync(vdev_t *vd, uint64_t txg)
2371 spa_t *spa = vd->vdev_spa;
2376 ASSERT(!vd->vdev_ishole);
2378 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2379 ASSERT(vd == vd->vdev_top);
2380 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2381 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2382 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2383 ASSERT(vd->vdev_ms_array != 0);
2384 vdev_config_dirty(vd);
2389 * Remove the metadata associated with this vdev once it's empty.
2391 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2392 vdev_remove(vd, txg);
2394 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2395 metaslab_sync(msp, txg);
2396 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2399 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2400 vdev_dtl_sync(lvd, txg);
2402 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2406 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2408 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2412 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2413 * not be opened, and no I/O is attempted.
2416 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2420 spa_vdev_state_enter(spa, SCL_NONE);
2422 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2423 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2425 if (!vd->vdev_ops->vdev_op_leaf)
2426 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2431 * We don't directly use the aux state here, but if we do a
2432 * vdev_reopen(), we need this value to be present to remember why we
2435 vd->vdev_label_aux = aux;
2438 * Faulted state takes precedence over degraded.
2440 vd->vdev_delayed_close = B_FALSE;
2441 vd->vdev_faulted = 1ULL;
2442 vd->vdev_degraded = 0ULL;
2443 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2446 * If this device has the only valid copy of the data, then
2447 * back off and simply mark the vdev as degraded instead.
2449 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2450 vd->vdev_degraded = 1ULL;
2451 vd->vdev_faulted = 0ULL;
2454 * If we reopen the device and it's not dead, only then do we
2459 if (vdev_readable(vd))
2460 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2463 return (spa_vdev_state_exit(spa, vd, 0));
2467 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2468 * user that something is wrong. The vdev continues to operate as normal as far
2469 * as I/O is concerned.
2472 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2476 spa_vdev_state_enter(spa, SCL_NONE);
2478 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2479 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2481 if (!vd->vdev_ops->vdev_op_leaf)
2482 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2485 * If the vdev is already faulted, then don't do anything.
2487 if (vd->vdev_faulted || vd->vdev_degraded)
2488 return (spa_vdev_state_exit(spa, NULL, 0));
2490 vd->vdev_degraded = 1ULL;
2491 if (!vdev_is_dead(vd))
2492 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2495 return (spa_vdev_state_exit(spa, vd, 0));
2499 * Online the given vdev.
2501 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2502 * spare device should be detached when the device finishes resilvering.
2503 * Second, the online should be treated like a 'test' online case, so no FMA
2504 * events are generated if the device fails to open.
2507 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2509 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2510 boolean_t postevent = B_FALSE;
2512 spa_vdev_state_enter(spa, SCL_NONE);
2514 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2515 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2517 if (!vd->vdev_ops->vdev_op_leaf)
2518 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2521 (vd->vdev_offline == B_TRUE || vd->vdev_tmpoffline == B_TRUE) ?
2525 vd->vdev_offline = B_FALSE;
2526 vd->vdev_tmpoffline = B_FALSE;
2527 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2528 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2530 /* XXX - L2ARC 1.0 does not support expansion */
2531 if (!vd->vdev_aux) {
2532 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2533 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2537 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2539 if (!vd->vdev_aux) {
2540 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2541 pvd->vdev_expanding = B_FALSE;
2545 *newstate = vd->vdev_state;
2546 if ((flags & ZFS_ONLINE_UNSPARE) &&
2547 !vdev_is_dead(vd) && vd->vdev_parent &&
2548 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2549 vd->vdev_parent->vdev_child[0] == vd)
2550 vd->vdev_unspare = B_TRUE;
2552 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2554 /* XXX - L2ARC 1.0 does not support expansion */
2556 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2557 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2561 spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE);
2563 return (spa_vdev_state_exit(spa, vd, 0));
2567 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2571 uint64_t generation;
2572 metaslab_group_t *mg;
2575 spa_vdev_state_enter(spa, SCL_ALLOC);
2577 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2578 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2580 if (!vd->vdev_ops->vdev_op_leaf)
2581 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2585 generation = spa->spa_config_generation + 1;
2588 * If the device isn't already offline, try to offline it.
2590 if (!vd->vdev_offline) {
2592 * If this device has the only valid copy of some data,
2593 * don't allow it to be offlined. Log devices are always
2596 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2597 vdev_dtl_required(vd))
2598 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2601 * If the top-level is a slog and it has had allocations
2602 * then proceed. We check that the vdev's metaslab group
2603 * is not NULL since it's possible that we may have just
2604 * added this vdev but not yet initialized its metaslabs.
2606 if (tvd->vdev_islog && mg != NULL) {
2608 * Prevent any future allocations.
2610 metaslab_group_passivate(mg);
2611 (void) spa_vdev_state_exit(spa, vd, 0);
2613 error = spa_offline_log(spa);
2615 spa_vdev_state_enter(spa, SCL_ALLOC);
2618 * Check to see if the config has changed.
2620 if (error || generation != spa->spa_config_generation) {
2621 metaslab_group_activate(mg);
2623 return (spa_vdev_state_exit(spa,
2625 (void) spa_vdev_state_exit(spa, vd, 0);
2628 ASSERT0(tvd->vdev_stat.vs_alloc);
2632 * Offline this device and reopen its top-level vdev.
2633 * If the top-level vdev is a log device then just offline
2634 * it. Otherwise, if this action results in the top-level
2635 * vdev becoming unusable, undo it and fail the request.
2637 vd->vdev_offline = B_TRUE;
2640 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2641 vdev_is_dead(tvd)) {
2642 vd->vdev_offline = B_FALSE;
2644 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2648 * Add the device back into the metaslab rotor so that
2649 * once we online the device it's open for business.
2651 if (tvd->vdev_islog && mg != NULL)
2652 metaslab_group_activate(mg);
2655 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2657 return (spa_vdev_state_exit(spa, vd, 0));
2661 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2665 mutex_enter(&spa->spa_vdev_top_lock);
2666 error = vdev_offline_locked(spa, guid, flags);
2667 mutex_exit(&spa->spa_vdev_top_lock);
2673 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2674 * vdev_offline(), we assume the spa config is locked. We also clear all
2675 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2678 vdev_clear(spa_t *spa, vdev_t *vd)
2680 vdev_t *rvd = spa->spa_root_vdev;
2682 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2687 vd->vdev_stat.vs_read_errors = 0;
2688 vd->vdev_stat.vs_write_errors = 0;
2689 vd->vdev_stat.vs_checksum_errors = 0;
2691 for (int c = 0; c < vd->vdev_children; c++)
2692 vdev_clear(spa, vd->vdev_child[c]);
2695 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2696 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2698 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2699 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2703 * If we're in the FAULTED state or have experienced failed I/O, then
2704 * clear the persistent state and attempt to reopen the device. We
2705 * also mark the vdev config dirty, so that the new faulted state is
2706 * written out to disk.
2708 if (vd->vdev_faulted || vd->vdev_degraded ||
2709 !vdev_readable(vd) || !vdev_writeable(vd)) {
2712 * When reopening in reponse to a clear event, it may be due to
2713 * a fmadm repair request. In this case, if the device is
2714 * still broken, we want to still post the ereport again.
2716 vd->vdev_forcefault = B_TRUE;
2718 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2719 vd->vdev_cant_read = B_FALSE;
2720 vd->vdev_cant_write = B_FALSE;
2722 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2724 vd->vdev_forcefault = B_FALSE;
2726 if (vd != rvd && vdev_writeable(vd->vdev_top))
2727 vdev_state_dirty(vd->vdev_top);
2729 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2730 spa_async_request(spa, SPA_ASYNC_RESILVER);
2732 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2736 * When clearing a FMA-diagnosed fault, we always want to
2737 * unspare the device, as we assume that the original spare was
2738 * done in response to the FMA fault.
2740 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2741 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2742 vd->vdev_parent->vdev_child[0] == vd)
2743 vd->vdev_unspare = B_TRUE;
2747 vdev_is_dead(vdev_t *vd)
2750 * Holes and missing devices are always considered "dead".
2751 * This simplifies the code since we don't have to check for
2752 * these types of devices in the various code paths.
2753 * Instead we rely on the fact that we skip over dead devices
2754 * before issuing I/O to them.
2756 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2757 vd->vdev_ops == &vdev_missing_ops);
2761 vdev_readable(vdev_t *vd)
2763 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2767 vdev_writeable(vdev_t *vd)
2769 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2773 vdev_allocatable(vdev_t *vd)
2775 uint64_t state = vd->vdev_state;
2778 * We currently allow allocations from vdevs which may be in the
2779 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2780 * fails to reopen then we'll catch it later when we're holding
2781 * the proper locks. Note that we have to get the vdev state
2782 * in a local variable because although it changes atomically,
2783 * we're asking two separate questions about it.
2785 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2786 !vd->vdev_cant_write && !vd->vdev_ishole &&
2787 vd->vdev_mg->mg_initialized);
2791 vdev_accessible(vdev_t *vd, zio_t *zio)
2793 ASSERT(zio->io_vd == vd);
2795 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2798 if (zio->io_type == ZIO_TYPE_READ)
2799 return (!vd->vdev_cant_read);
2801 if (zio->io_type == ZIO_TYPE_WRITE)
2802 return (!vd->vdev_cant_write);
2808 * Get statistics for the given vdev.
2811 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2813 spa_t *spa = vd->vdev_spa;
2814 vdev_t *rvd = spa->spa_root_vdev;
2815 vdev_t *tvd = vd->vdev_top;
2817 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2819 mutex_enter(&vd->vdev_stat_lock);
2820 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2821 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2822 vs->vs_state = vd->vdev_state;
2823 vs->vs_rsize = vdev_get_min_asize(vd);
2824 if (vd->vdev_ops->vdev_op_leaf)
2825 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2827 * Report expandable space on top-level, non-auxillary devices only.
2828 * The expandable space is reported in terms of metaslab sized units
2829 * since that determines how much space the pool can expand.
2831 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
2832 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize,
2833 1ULL << tvd->vdev_ms_shift);
2835 vs->vs_configured_ashift = vd->vdev_top != NULL
2836 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2837 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2838 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2839 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2840 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2844 * If we're getting stats on the root vdev, aggregate the I/O counts
2845 * over all top-level vdevs (i.e. the direct children of the root).
2848 for (int c = 0; c < rvd->vdev_children; c++) {
2849 vdev_t *cvd = rvd->vdev_child[c];
2850 vdev_stat_t *cvs = &cvd->vdev_stat;
2852 for (int t = 0; t < ZIO_TYPES; t++) {
2853 vs->vs_ops[t] += cvs->vs_ops[t];
2854 vs->vs_bytes[t] += cvs->vs_bytes[t];
2856 cvs->vs_scan_removing = cvd->vdev_removing;
2859 mutex_exit(&vd->vdev_stat_lock);
2863 vdev_clear_stats(vdev_t *vd)
2865 mutex_enter(&vd->vdev_stat_lock);
2866 vd->vdev_stat.vs_space = 0;
2867 vd->vdev_stat.vs_dspace = 0;
2868 vd->vdev_stat.vs_alloc = 0;
2869 mutex_exit(&vd->vdev_stat_lock);
2873 vdev_scan_stat_init(vdev_t *vd)
2875 vdev_stat_t *vs = &vd->vdev_stat;
2877 for (int c = 0; c < vd->vdev_children; c++)
2878 vdev_scan_stat_init(vd->vdev_child[c]);
2880 mutex_enter(&vd->vdev_stat_lock);
2881 vs->vs_scan_processed = 0;
2882 mutex_exit(&vd->vdev_stat_lock);
2886 vdev_stat_update(zio_t *zio, uint64_t psize)
2888 spa_t *spa = zio->io_spa;
2889 vdev_t *rvd = spa->spa_root_vdev;
2890 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2892 uint64_t txg = zio->io_txg;
2893 vdev_stat_t *vs = &vd->vdev_stat;
2894 zio_type_t type = zio->io_type;
2895 int flags = zio->io_flags;
2898 * If this i/o is a gang leader, it didn't do any actual work.
2900 if (zio->io_gang_tree)
2903 if (zio->io_error == 0) {
2905 * If this is a root i/o, don't count it -- we've already
2906 * counted the top-level vdevs, and vdev_get_stats() will
2907 * aggregate them when asked. This reduces contention on
2908 * the root vdev_stat_lock and implicitly handles blocks
2909 * that compress away to holes, for which there is no i/o.
2910 * (Holes never create vdev children, so all the counters
2911 * remain zero, which is what we want.)
2913 * Note: this only applies to successful i/o (io_error == 0)
2914 * because unlike i/o counts, errors are not additive.
2915 * When reading a ditto block, for example, failure of
2916 * one top-level vdev does not imply a root-level error.
2921 ASSERT(vd == zio->io_vd);
2923 if (flags & ZIO_FLAG_IO_BYPASS)
2926 mutex_enter(&vd->vdev_stat_lock);
2928 if (flags & ZIO_FLAG_IO_REPAIR) {
2929 if (flags & ZIO_FLAG_SCAN_THREAD) {
2930 dsl_scan_phys_t *scn_phys =
2931 &spa->spa_dsl_pool->dp_scan->scn_phys;
2932 uint64_t *processed = &scn_phys->scn_processed;
2935 if (vd->vdev_ops->vdev_op_leaf)
2936 atomic_add_64(processed, psize);
2937 vs->vs_scan_processed += psize;
2940 if (flags & ZIO_FLAG_SELF_HEAL)
2941 vs->vs_self_healed += psize;
2945 vs->vs_bytes[type] += psize;
2947 mutex_exit(&vd->vdev_stat_lock);
2951 if (flags & ZIO_FLAG_SPECULATIVE)
2955 * If this is an I/O error that is going to be retried, then ignore the
2956 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2957 * hard errors, when in reality they can happen for any number of
2958 * innocuous reasons (bus resets, MPxIO link failure, etc).
2960 if (zio->io_error == EIO &&
2961 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2965 * Intent logs writes won't propagate their error to the root
2966 * I/O so don't mark these types of failures as pool-level
2969 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2972 mutex_enter(&vd->vdev_stat_lock);
2973 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2974 if (zio->io_error == ECKSUM)
2975 vs->vs_checksum_errors++;
2977 vs->vs_read_errors++;
2979 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2980 vs->vs_write_errors++;
2981 mutex_exit(&vd->vdev_stat_lock);
2983 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2984 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2985 (flags & ZIO_FLAG_SCAN_THREAD) ||
2986 spa->spa_claiming)) {
2988 * This is either a normal write (not a repair), or it's
2989 * a repair induced by the scrub thread, or it's a repair
2990 * made by zil_claim() during spa_load() in the first txg.
2991 * In the normal case, we commit the DTL change in the same
2992 * txg as the block was born. In the scrub-induced repair
2993 * case, we know that scrubs run in first-pass syncing context,
2994 * so we commit the DTL change in spa_syncing_txg(spa).
2995 * In the zil_claim() case, we commit in spa_first_txg(spa).
2997 * We currently do not make DTL entries for failed spontaneous
2998 * self-healing writes triggered by normal (non-scrubbing)
2999 * reads, because we have no transactional context in which to
3000 * do so -- and it's not clear that it'd be desirable anyway.
3002 if (vd->vdev_ops->vdev_op_leaf) {
3003 uint64_t commit_txg = txg;
3004 if (flags & ZIO_FLAG_SCAN_THREAD) {
3005 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3006 ASSERT(spa_sync_pass(spa) == 1);
3007 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3008 commit_txg = spa_syncing_txg(spa);
3009 } else if (spa->spa_claiming) {
3010 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3011 commit_txg = spa_first_txg(spa);
3013 ASSERT(commit_txg >= spa_syncing_txg(spa));
3014 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3016 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3017 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3018 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3021 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3026 * Update the in-core space usage stats for this vdev, its metaslab class,
3027 * and the root vdev.
3030 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3031 int64_t space_delta)
3033 int64_t dspace_delta = space_delta;
3034 spa_t *spa = vd->vdev_spa;
3035 vdev_t *rvd = spa->spa_root_vdev;
3036 metaslab_group_t *mg = vd->vdev_mg;
3037 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3039 ASSERT(vd == vd->vdev_top);
3042 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3043 * factor. We must calculate this here and not at the root vdev
3044 * because the root vdev's psize-to-asize is simply the max of its
3045 * childrens', thus not accurate enough for us.
3047 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3048 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3049 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3050 vd->vdev_deflate_ratio;
3052 mutex_enter(&vd->vdev_stat_lock);
3053 vd->vdev_stat.vs_alloc += alloc_delta;
3054 vd->vdev_stat.vs_space += space_delta;
3055 vd->vdev_stat.vs_dspace += dspace_delta;
3056 mutex_exit(&vd->vdev_stat_lock);
3058 if (mc == spa_normal_class(spa)) {
3059 mutex_enter(&rvd->vdev_stat_lock);
3060 rvd->vdev_stat.vs_alloc += alloc_delta;
3061 rvd->vdev_stat.vs_space += space_delta;
3062 rvd->vdev_stat.vs_dspace += dspace_delta;
3063 mutex_exit(&rvd->vdev_stat_lock);
3067 ASSERT(rvd == vd->vdev_parent);
3068 ASSERT(vd->vdev_ms_count != 0);
3070 metaslab_class_space_update(mc,
3071 alloc_delta, defer_delta, space_delta, dspace_delta);
3076 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3077 * so that it will be written out next time the vdev configuration is synced.
3078 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3081 vdev_config_dirty(vdev_t *vd)
3083 spa_t *spa = vd->vdev_spa;
3084 vdev_t *rvd = spa->spa_root_vdev;
3087 ASSERT(spa_writeable(spa));
3090 * If this is an aux vdev (as with l2cache and spare devices), then we
3091 * update the vdev config manually and set the sync flag.
3093 if (vd->vdev_aux != NULL) {
3094 spa_aux_vdev_t *sav = vd->vdev_aux;
3098 for (c = 0; c < sav->sav_count; c++) {
3099 if (sav->sav_vdevs[c] == vd)
3103 if (c == sav->sav_count) {
3105 * We're being removed. There's nothing more to do.
3107 ASSERT(sav->sav_sync == B_TRUE);
3111 sav->sav_sync = B_TRUE;
3113 if (nvlist_lookup_nvlist_array(sav->sav_config,
3114 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3115 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3116 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3122 * Setting the nvlist in the middle if the array is a little
3123 * sketchy, but it will work.
3125 nvlist_free(aux[c]);
3126 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3132 * The dirty list is protected by the SCL_CONFIG lock. The caller
3133 * must either hold SCL_CONFIG as writer, or must be the sync thread
3134 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3135 * so this is sufficient to ensure mutual exclusion.
3137 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3138 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3139 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3142 for (c = 0; c < rvd->vdev_children; c++)
3143 vdev_config_dirty(rvd->vdev_child[c]);
3145 ASSERT(vd == vd->vdev_top);
3147 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3149 list_insert_head(&spa->spa_config_dirty_list, vd);
3154 vdev_config_clean(vdev_t *vd)
3156 spa_t *spa = vd->vdev_spa;
3158 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3159 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3160 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3162 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3163 list_remove(&spa->spa_config_dirty_list, vd);
3167 * Mark a top-level vdev's state as dirty, so that the next pass of
3168 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3169 * the state changes from larger config changes because they require
3170 * much less locking, and are often needed for administrative actions.
3173 vdev_state_dirty(vdev_t *vd)
3175 spa_t *spa = vd->vdev_spa;
3177 ASSERT(spa_writeable(spa));
3178 ASSERT(vd == vd->vdev_top);
3181 * The state list is protected by the SCL_STATE lock. The caller
3182 * must either hold SCL_STATE as writer, or must be the sync thread
3183 * (which holds SCL_STATE as reader). There's only one sync thread,
3184 * so this is sufficient to ensure mutual exclusion.
3186 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3187 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3188 spa_config_held(spa, SCL_STATE, RW_READER)));
3190 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3191 list_insert_head(&spa->spa_state_dirty_list, vd);
3195 vdev_state_clean(vdev_t *vd)
3197 spa_t *spa = vd->vdev_spa;
3199 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3200 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3201 spa_config_held(spa, SCL_STATE, RW_READER)));
3203 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3204 list_remove(&spa->spa_state_dirty_list, vd);
3208 * Propagate vdev state up from children to parent.
3211 vdev_propagate_state(vdev_t *vd)
3213 spa_t *spa = vd->vdev_spa;
3214 vdev_t *rvd = spa->spa_root_vdev;
3215 int degraded = 0, faulted = 0;
3219 if (vd->vdev_children > 0) {
3220 for (int c = 0; c < vd->vdev_children; c++) {
3221 child = vd->vdev_child[c];
3224 * Don't factor holes into the decision.
3226 if (child->vdev_ishole)
3229 if (!vdev_readable(child) ||
3230 (!vdev_writeable(child) && spa_writeable(spa))) {
3232 * Root special: if there is a top-level log
3233 * device, treat the root vdev as if it were
3236 if (child->vdev_islog && vd == rvd)
3240 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3244 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3248 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3251 * Root special: if there is a top-level vdev that cannot be
3252 * opened due to corrupted metadata, then propagate the root
3253 * vdev's aux state as 'corrupt' rather than 'insufficient
3256 if (corrupted && vd == rvd &&
3257 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3258 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3259 VDEV_AUX_CORRUPT_DATA);
3262 if (vd->vdev_parent)
3263 vdev_propagate_state(vd->vdev_parent);
3267 * Set a vdev's state. If this is during an open, we don't update the parent
3268 * state, because we're in the process of opening children depth-first.
3269 * Otherwise, we propagate the change to the parent.
3271 * If this routine places a device in a faulted state, an appropriate ereport is
3275 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3277 uint64_t save_state;
3278 spa_t *spa = vd->vdev_spa;
3280 if (state == vd->vdev_state) {
3281 vd->vdev_stat.vs_aux = aux;
3285 save_state = vd->vdev_state;
3287 vd->vdev_state = state;
3288 vd->vdev_stat.vs_aux = aux;
3291 * If we are setting the vdev state to anything but an open state, then
3292 * always close the underlying device unless the device has requested
3293 * a delayed close (i.e. we're about to remove or fault the device).
3294 * Otherwise, we keep accessible but invalid devices open forever.
3295 * We don't call vdev_close() itself, because that implies some extra
3296 * checks (offline, etc) that we don't want here. This is limited to
3297 * leaf devices, because otherwise closing the device will affect other
3300 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3301 vd->vdev_ops->vdev_op_leaf)
3302 vd->vdev_ops->vdev_op_close(vd);
3305 * If we have brought this vdev back into service, we need
3306 * to notify fmd so that it can gracefully repair any outstanding
3307 * cases due to a missing device. We do this in all cases, even those
3308 * that probably don't correlate to a repaired fault. This is sure to
3309 * catch all cases, and we let the zfs-retire agent sort it out. If
3310 * this is a transient state it's OK, as the retire agent will
3311 * double-check the state of the vdev before repairing it.
3313 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3314 vd->vdev_prevstate != state)
3315 zfs_post_state_change(spa, vd);
3317 if (vd->vdev_removed &&
3318 state == VDEV_STATE_CANT_OPEN &&
3319 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3321 * If the previous state is set to VDEV_STATE_REMOVED, then this
3322 * device was previously marked removed and someone attempted to
3323 * reopen it. If this failed due to a nonexistent device, then
3324 * keep the device in the REMOVED state. We also let this be if
3325 * it is one of our special test online cases, which is only
3326 * attempting to online the device and shouldn't generate an FMA
3329 vd->vdev_state = VDEV_STATE_REMOVED;
3330 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3331 } else if (state == VDEV_STATE_REMOVED) {
3332 vd->vdev_removed = B_TRUE;
3333 } else if (state == VDEV_STATE_CANT_OPEN) {
3335 * If we fail to open a vdev during an import or recovery, we
3336 * mark it as "not available", which signifies that it was
3337 * never there to begin with. Failure to open such a device
3338 * is not considered an error.
3340 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3341 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3342 vd->vdev_ops->vdev_op_leaf)
3343 vd->vdev_not_present = 1;
3346 * Post the appropriate ereport. If the 'prevstate' field is
3347 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3348 * that this is part of a vdev_reopen(). In this case, we don't
3349 * want to post the ereport if the device was already in the
3350 * CANT_OPEN state beforehand.
3352 * If the 'checkremove' flag is set, then this is an attempt to
3353 * online the device in response to an insertion event. If we
3354 * hit this case, then we have detected an insertion event for a
3355 * faulted or offline device that wasn't in the removed state.
3356 * In this scenario, we don't post an ereport because we are
3357 * about to replace the device, or attempt an online with
3358 * vdev_forcefault, which will generate the fault for us.
3360 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3361 !vd->vdev_not_present && !vd->vdev_checkremove &&
3362 vd != spa->spa_root_vdev) {
3366 case VDEV_AUX_OPEN_FAILED:
3367 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3369 case VDEV_AUX_CORRUPT_DATA:
3370 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3372 case VDEV_AUX_NO_REPLICAS:
3373 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3375 case VDEV_AUX_BAD_GUID_SUM:
3376 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3378 case VDEV_AUX_TOO_SMALL:
3379 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3381 case VDEV_AUX_BAD_LABEL:
3382 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3385 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3388 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3391 /* Erase any notion of persistent removed state */
3392 vd->vdev_removed = B_FALSE;
3394 vd->vdev_removed = B_FALSE;
3397 if (!isopen && vd->vdev_parent)
3398 vdev_propagate_state(vd->vdev_parent);
3402 * Check the vdev configuration to ensure that it's capable of supporting
3405 * On Solaris, we do not support RAID-Z or partial configuration. In
3406 * addition, only a single top-level vdev is allowed and none of the
3407 * leaves can be wholedisks.
3409 * For FreeBSD, we can boot from any configuration. There is a
3410 * limitation that the boot filesystem must be either uncompressed or
3411 * compresses with lzjb compression but I'm not sure how to enforce
3415 vdev_is_bootable(vdev_t *vd)
3418 if (!vd->vdev_ops->vdev_op_leaf) {
3419 char *vdev_type = vd->vdev_ops->vdev_op_type;
3421 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3422 vd->vdev_children > 1) {
3424 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3425 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3430 for (int c = 0; c < vd->vdev_children; c++) {
3431 if (!vdev_is_bootable(vd->vdev_child[c]))
3434 #endif /* illumos */
3439 * Load the state from the original vdev tree (ovd) which
3440 * we've retrieved from the MOS config object. If the original
3441 * vdev was offline or faulted then we transfer that state to the
3442 * device in the current vdev tree (nvd).
3445 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3447 spa_t *spa = nvd->vdev_spa;
3449 ASSERT(nvd->vdev_top->vdev_islog);
3450 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3451 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3453 for (int c = 0; c < nvd->vdev_children; c++)
3454 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3456 if (nvd->vdev_ops->vdev_op_leaf) {
3458 * Restore the persistent vdev state
3460 nvd->vdev_offline = ovd->vdev_offline;
3461 nvd->vdev_faulted = ovd->vdev_faulted;
3462 nvd->vdev_degraded = ovd->vdev_degraded;
3463 nvd->vdev_removed = ovd->vdev_removed;
3468 * Determine if a log device has valid content. If the vdev was
3469 * removed or faulted in the MOS config then we know that
3470 * the content on the log device has already been written to the pool.
3473 vdev_log_state_valid(vdev_t *vd)
3475 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3479 for (int c = 0; c < vd->vdev_children; c++)
3480 if (vdev_log_state_valid(vd->vdev_child[c]))
3487 * Expand a vdev if possible.
3490 vdev_expand(vdev_t *vd, uint64_t txg)
3492 ASSERT(vd->vdev_top == vd);
3493 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3495 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3496 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3497 vdev_config_dirty(vd);
3505 vdev_split(vdev_t *vd)
3507 vdev_t *cvd, *pvd = vd->vdev_parent;
3509 vdev_remove_child(pvd, vd);
3510 vdev_compact_children(pvd);
3512 cvd = pvd->vdev_child[0];
3513 if (pvd->vdev_children == 1) {
3514 vdev_remove_parent(cvd);
3515 cvd->vdev_splitting = B_TRUE;
3517 vdev_propagate_state(cvd);
3521 vdev_deadman(vdev_t *vd)
3523 for (int c = 0; c < vd->vdev_children; c++) {
3524 vdev_t *cvd = vd->vdev_child[c];
3529 if (vd->vdev_ops->vdev_op_leaf) {
3530 vdev_queue_t *vq = &vd->vdev_queue;
3532 mutex_enter(&vq->vq_lock);
3533 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3534 spa_t *spa = vd->vdev_spa;
3539 * Look at the head of all the pending queues,
3540 * if any I/O has been outstanding for longer than
3541 * the spa_deadman_synctime we panic the system.
3543 fio = avl_first(&vq->vq_active_tree);
3544 delta = gethrtime() - fio->io_timestamp;
3545 if (delta > spa_deadman_synctime(spa)) {
3546 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3547 "delta %lluns, last io %lluns",
3548 fio->io_timestamp, delta,
3549 vq->vq_io_complete_ts);
3550 fm_panic("I/O to pool '%s' appears to be "
3551 "hung on vdev guid %llu at '%s'.",
3553 (long long unsigned int) vd->vdev_guid,
3557 mutex_exit(&vq->vq_lock);