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]
28 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 #include <sys/zfs_context.h>
32 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/vdev_impl.h>
38 #include <sys/uberblock_impl.h>
39 #include <sys/metaslab.h>
40 #include <sys/metaslab_impl.h>
41 #include <sys/space_map.h>
42 #include <sys/space_reftree.h>
45 #include <sys/fs/zfs.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/trim_map.h>
51 SYSCTL_DECL(_vfs_zfs);
52 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
55 * Virtual device management.
59 * The limit for ZFS to automatically increase a top-level vdev's ashift
60 * from logical ashift to physical ashift.
62 * Example: one or more 512B emulation child vdevs
63 * child->vdev_ashift = 9 (512 bytes)
64 * child->vdev_physical_ashift = 12 (4096 bytes)
65 * zfs_max_auto_ashift = 11 (2048 bytes)
66 * zfs_min_auto_ashift = 9 (512 bytes)
68 * On pool creation or the addition of a new top-level vdev, ZFS will
69 * increase the ashift of the top-level vdev to 2048 as limited by
70 * zfs_max_auto_ashift.
72 * Example: one or more 512B emulation child vdevs
73 * child->vdev_ashift = 9 (512 bytes)
74 * child->vdev_physical_ashift = 12 (4096 bytes)
75 * zfs_max_auto_ashift = 13 (8192 bytes)
76 * zfs_min_auto_ashift = 9 (512 bytes)
78 * On pool creation or the addition of a new top-level vdev, ZFS will
79 * increase the ashift of the top-level vdev to 4096 to match the
80 * max vdev_physical_ashift.
82 * Example: one or more 512B emulation child vdevs
83 * child->vdev_ashift = 9 (512 bytes)
84 * child->vdev_physical_ashift = 9 (512 bytes)
85 * zfs_max_auto_ashift = 13 (8192 bytes)
86 * zfs_min_auto_ashift = 12 (4096 bytes)
88 * On pool creation or the addition of a new top-level vdev, ZFS will
89 * increase the ashift of the top-level vdev to 4096 to match the
90 * zfs_min_auto_ashift.
92 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
93 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
96 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
101 val = zfs_max_auto_ashift;
102 err = sysctl_handle_64(oidp, &val, 0, req);
103 if (err != 0 || req->newptr == NULL)
106 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
109 zfs_max_auto_ashift = val;
113 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
114 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
115 sysctl_vfs_zfs_max_auto_ashift, "QU",
116 "Max ashift used when optimising for logical -> physical sectors size on "
117 "new top-level vdevs.");
120 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
125 val = zfs_min_auto_ashift;
126 err = sysctl_handle_64(oidp, &val, 0, req);
127 if (err != 0 || req->newptr == NULL)
130 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
133 zfs_min_auto_ashift = val;
137 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
138 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
139 sysctl_vfs_zfs_min_auto_ashift, "QU",
140 "Min ashift used when creating new top-level vdevs.");
142 static vdev_ops_t *vdev_ops_table[] = {
161 * When a vdev is added, it will be divided into approximately (but no
162 * more than) this number of metaslabs.
164 int metaslabs_per_vdev = 200;
165 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
166 &metaslabs_per_vdev, 0,
167 "When a vdev is added, how many metaslabs the vdev should be divided into");
170 * Given a vdev type, return the appropriate ops vector.
173 vdev_getops(const char *type)
175 vdev_ops_t *ops, **opspp;
177 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
178 if (strcmp(ops->vdev_op_type, type) == 0)
185 * Default asize function: return the MAX of psize with the asize of
186 * all children. This is what's used by anything other than RAID-Z.
189 vdev_default_asize(vdev_t *vd, uint64_t psize)
191 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
194 for (int c = 0; c < vd->vdev_children; c++) {
195 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
196 asize = MAX(asize, csize);
203 * Get the minimum allocatable size. We define the allocatable size as
204 * the vdev's asize rounded to the nearest metaslab. This allows us to
205 * replace or attach devices which don't have the same physical size but
206 * can still satisfy the same number of allocations.
209 vdev_get_min_asize(vdev_t *vd)
211 vdev_t *pvd = vd->vdev_parent;
214 * If our parent is NULL (inactive spare or cache) or is the root,
215 * just return our own asize.
218 return (vd->vdev_asize);
221 * The top-level vdev just returns the allocatable size rounded
222 * to the nearest metaslab.
224 if (vd == vd->vdev_top)
225 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
228 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
229 * so each child must provide at least 1/Nth of its asize.
231 if (pvd->vdev_ops == &vdev_raidz_ops)
232 return (pvd->vdev_min_asize / pvd->vdev_children);
234 return (pvd->vdev_min_asize);
238 vdev_set_min_asize(vdev_t *vd)
240 vd->vdev_min_asize = vdev_get_min_asize(vd);
242 for (int c = 0; c < vd->vdev_children; c++)
243 vdev_set_min_asize(vd->vdev_child[c]);
247 vdev_lookup_top(spa_t *spa, uint64_t vdev)
249 vdev_t *rvd = spa->spa_root_vdev;
251 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
253 if (vdev < rvd->vdev_children) {
254 ASSERT(rvd->vdev_child[vdev] != NULL);
255 return (rvd->vdev_child[vdev]);
262 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
266 if (vd->vdev_guid == guid)
269 for (int c = 0; c < vd->vdev_children; c++)
270 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
278 vdev_count_leaves_impl(vdev_t *vd)
282 if (vd->vdev_ops->vdev_op_leaf)
285 for (int c = 0; c < vd->vdev_children; c++)
286 n += vdev_count_leaves_impl(vd->vdev_child[c]);
292 vdev_count_leaves(spa_t *spa)
294 return (vdev_count_leaves_impl(spa->spa_root_vdev));
298 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
300 size_t oldsize, newsize;
301 uint64_t id = cvd->vdev_id;
303 spa_t *spa = cvd->vdev_spa;
305 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
306 ASSERT(cvd->vdev_parent == NULL);
308 cvd->vdev_parent = pvd;
313 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
315 oldsize = pvd->vdev_children * sizeof (vdev_t *);
316 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
317 newsize = pvd->vdev_children * sizeof (vdev_t *);
319 newchild = kmem_zalloc(newsize, KM_SLEEP);
320 if (pvd->vdev_child != NULL) {
321 bcopy(pvd->vdev_child, newchild, oldsize);
322 kmem_free(pvd->vdev_child, oldsize);
325 pvd->vdev_child = newchild;
326 pvd->vdev_child[id] = cvd;
328 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
329 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
332 * Walk up all ancestors to update guid sum.
334 for (; pvd != NULL; pvd = pvd->vdev_parent)
335 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
339 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
342 uint_t id = cvd->vdev_id;
344 ASSERT(cvd->vdev_parent == pvd);
349 ASSERT(id < pvd->vdev_children);
350 ASSERT(pvd->vdev_child[id] == cvd);
352 pvd->vdev_child[id] = NULL;
353 cvd->vdev_parent = NULL;
355 for (c = 0; c < pvd->vdev_children; c++)
356 if (pvd->vdev_child[c])
359 if (c == pvd->vdev_children) {
360 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
361 pvd->vdev_child = NULL;
362 pvd->vdev_children = 0;
366 * Walk up all ancestors to update guid sum.
368 for (; pvd != NULL; pvd = pvd->vdev_parent)
369 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
373 * Remove any holes in the child array.
376 vdev_compact_children(vdev_t *pvd)
378 vdev_t **newchild, *cvd;
379 int oldc = pvd->vdev_children;
382 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
384 for (int c = newc = 0; c < oldc; c++)
385 if (pvd->vdev_child[c])
388 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
390 for (int c = newc = 0; c < oldc; c++) {
391 if ((cvd = pvd->vdev_child[c]) != NULL) {
392 newchild[newc] = cvd;
393 cvd->vdev_id = newc++;
397 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
398 pvd->vdev_child = newchild;
399 pvd->vdev_children = newc;
403 * Allocate and minimally initialize a vdev_t.
406 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
410 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
412 if (spa->spa_root_vdev == NULL) {
413 ASSERT(ops == &vdev_root_ops);
414 spa->spa_root_vdev = vd;
415 spa->spa_load_guid = spa_generate_guid(NULL);
418 if (guid == 0 && ops != &vdev_hole_ops) {
419 if (spa->spa_root_vdev == vd) {
421 * The root vdev's guid will also be the pool guid,
422 * which must be unique among all pools.
424 guid = spa_generate_guid(NULL);
427 * Any other vdev's guid must be unique within the pool.
429 guid = spa_generate_guid(spa);
431 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
436 vd->vdev_guid = guid;
437 vd->vdev_guid_sum = guid;
439 vd->vdev_state = VDEV_STATE_CLOSED;
440 vd->vdev_ishole = (ops == &vdev_hole_ops);
442 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
443 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
445 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
446 for (int t = 0; t < DTL_TYPES; t++) {
447 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
450 txg_list_create(&vd->vdev_ms_list,
451 offsetof(struct metaslab, ms_txg_node));
452 txg_list_create(&vd->vdev_dtl_list,
453 offsetof(struct vdev, vdev_dtl_node));
454 vd->vdev_stat.vs_timestamp = gethrtime();
462 * Allocate a new vdev. The 'alloctype' is used to control whether we are
463 * creating a new vdev or loading an existing one - the behavior is slightly
464 * different for each case.
467 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
472 uint64_t guid = 0, islog, nparity;
475 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
477 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
478 return (SET_ERROR(EINVAL));
480 if ((ops = vdev_getops(type)) == NULL)
481 return (SET_ERROR(EINVAL));
484 * If this is a load, get the vdev guid from the nvlist.
485 * Otherwise, vdev_alloc_common() will generate one for us.
487 if (alloctype == VDEV_ALLOC_LOAD) {
490 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
492 return (SET_ERROR(EINVAL));
494 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
495 return (SET_ERROR(EINVAL));
496 } else if (alloctype == VDEV_ALLOC_SPARE) {
497 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
498 return (SET_ERROR(EINVAL));
499 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
500 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
501 return (SET_ERROR(EINVAL));
502 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
503 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
504 return (SET_ERROR(EINVAL));
508 * The first allocated vdev must be of type 'root'.
510 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
511 return (SET_ERROR(EINVAL));
514 * Determine whether we're a log vdev.
517 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
518 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
519 return (SET_ERROR(ENOTSUP));
521 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
522 return (SET_ERROR(ENOTSUP));
525 * Set the nparity property for RAID-Z vdevs.
528 if (ops == &vdev_raidz_ops) {
529 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
531 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
532 return (SET_ERROR(EINVAL));
534 * Previous versions could only support 1 or 2 parity
538 spa_version(spa) < SPA_VERSION_RAIDZ2)
539 return (SET_ERROR(ENOTSUP));
541 spa_version(spa) < SPA_VERSION_RAIDZ3)
542 return (SET_ERROR(ENOTSUP));
545 * We require the parity to be specified for SPAs that
546 * support multiple parity levels.
548 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
549 return (SET_ERROR(EINVAL));
551 * Otherwise, we default to 1 parity device for RAID-Z.
558 ASSERT(nparity != -1ULL);
560 vd = vdev_alloc_common(spa, id, guid, ops);
562 vd->vdev_islog = islog;
563 vd->vdev_nparity = nparity;
565 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
566 vd->vdev_path = spa_strdup(vd->vdev_path);
567 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
568 vd->vdev_devid = spa_strdup(vd->vdev_devid);
569 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
570 &vd->vdev_physpath) == 0)
571 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
572 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
573 vd->vdev_fru = spa_strdup(vd->vdev_fru);
576 * Set the whole_disk property. If it's not specified, leave the value
579 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
580 &vd->vdev_wholedisk) != 0)
581 vd->vdev_wholedisk = -1ULL;
584 * Look for the 'not present' flag. This will only be set if the device
585 * was not present at the time of import.
587 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
588 &vd->vdev_not_present);
591 * Get the alignment requirement.
593 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
596 * Retrieve the vdev creation time.
598 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
602 * If we're a top-level vdev, try to load the allocation parameters.
604 if (parent && !parent->vdev_parent &&
605 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
606 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
608 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
610 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
612 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
614 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
617 ASSERT0(vd->vdev_top_zap);
620 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
621 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
622 alloctype == VDEV_ALLOC_ADD ||
623 alloctype == VDEV_ALLOC_SPLIT ||
624 alloctype == VDEV_ALLOC_ROOTPOOL);
625 vd->vdev_mg = metaslab_group_create(islog ?
626 spa_log_class(spa) : spa_normal_class(spa), vd);
629 if (vd->vdev_ops->vdev_op_leaf &&
630 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
631 (void) nvlist_lookup_uint64(nv,
632 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
634 ASSERT0(vd->vdev_leaf_zap);
638 * If we're a leaf vdev, try to load the DTL object and other state.
641 if (vd->vdev_ops->vdev_op_leaf &&
642 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
643 alloctype == VDEV_ALLOC_ROOTPOOL)) {
644 if (alloctype == VDEV_ALLOC_LOAD) {
645 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
646 &vd->vdev_dtl_object);
647 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
651 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
654 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
655 &spare) == 0 && spare)
659 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
662 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
663 &vd->vdev_resilver_txg);
666 * When importing a pool, we want to ignore the persistent fault
667 * state, as the diagnosis made on another system may not be
668 * valid in the current context. Local vdevs will
669 * remain in the faulted state.
671 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
672 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
674 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
676 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
679 if (vd->vdev_faulted || vd->vdev_degraded) {
683 VDEV_AUX_ERR_EXCEEDED;
684 if (nvlist_lookup_string(nv,
685 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
686 strcmp(aux, "external") == 0)
687 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
693 * Add ourselves to the parent's list of children.
695 vdev_add_child(parent, vd);
703 vdev_free(vdev_t *vd)
705 spa_t *spa = vd->vdev_spa;
708 * vdev_free() implies closing the vdev first. This is simpler than
709 * trying to ensure complicated semantics for all callers.
713 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
714 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
719 for (int c = 0; c < vd->vdev_children; c++)
720 vdev_free(vd->vdev_child[c]);
722 ASSERT(vd->vdev_child == NULL);
723 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
726 * Discard allocation state.
728 if (vd->vdev_mg != NULL) {
729 vdev_metaslab_fini(vd);
730 metaslab_group_destroy(vd->vdev_mg);
733 ASSERT0(vd->vdev_stat.vs_space);
734 ASSERT0(vd->vdev_stat.vs_dspace);
735 ASSERT0(vd->vdev_stat.vs_alloc);
738 * Remove this vdev from its parent's child list.
740 vdev_remove_child(vd->vdev_parent, vd);
742 ASSERT(vd->vdev_parent == NULL);
745 * Clean up vdev structure.
751 spa_strfree(vd->vdev_path);
753 spa_strfree(vd->vdev_devid);
754 if (vd->vdev_physpath)
755 spa_strfree(vd->vdev_physpath);
757 spa_strfree(vd->vdev_fru);
759 if (vd->vdev_isspare)
760 spa_spare_remove(vd);
761 if (vd->vdev_isl2cache)
762 spa_l2cache_remove(vd);
764 txg_list_destroy(&vd->vdev_ms_list);
765 txg_list_destroy(&vd->vdev_dtl_list);
767 mutex_enter(&vd->vdev_dtl_lock);
768 space_map_close(vd->vdev_dtl_sm);
769 for (int t = 0; t < DTL_TYPES; t++) {
770 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
771 range_tree_destroy(vd->vdev_dtl[t]);
773 mutex_exit(&vd->vdev_dtl_lock);
775 mutex_destroy(&vd->vdev_queue_lock);
776 mutex_destroy(&vd->vdev_dtl_lock);
777 mutex_destroy(&vd->vdev_stat_lock);
778 mutex_destroy(&vd->vdev_probe_lock);
780 if (vd == spa->spa_root_vdev)
781 spa->spa_root_vdev = NULL;
783 kmem_free(vd, sizeof (vdev_t));
787 * Transfer top-level vdev state from svd to tvd.
790 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
792 spa_t *spa = svd->vdev_spa;
797 ASSERT(tvd == tvd->vdev_top);
799 tvd->vdev_ms_array = svd->vdev_ms_array;
800 tvd->vdev_ms_shift = svd->vdev_ms_shift;
801 tvd->vdev_ms_count = svd->vdev_ms_count;
802 tvd->vdev_top_zap = svd->vdev_top_zap;
804 svd->vdev_ms_array = 0;
805 svd->vdev_ms_shift = 0;
806 svd->vdev_ms_count = 0;
807 svd->vdev_top_zap = 0;
810 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
811 tvd->vdev_mg = svd->vdev_mg;
812 tvd->vdev_ms = svd->vdev_ms;
817 if (tvd->vdev_mg != NULL)
818 tvd->vdev_mg->mg_vd = tvd;
820 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
821 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
822 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
824 svd->vdev_stat.vs_alloc = 0;
825 svd->vdev_stat.vs_space = 0;
826 svd->vdev_stat.vs_dspace = 0;
828 for (t = 0; t < TXG_SIZE; t++) {
829 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
830 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
831 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
832 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
833 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
834 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
837 if (list_link_active(&svd->vdev_config_dirty_node)) {
838 vdev_config_clean(svd);
839 vdev_config_dirty(tvd);
842 if (list_link_active(&svd->vdev_state_dirty_node)) {
843 vdev_state_clean(svd);
844 vdev_state_dirty(tvd);
847 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
848 svd->vdev_deflate_ratio = 0;
850 tvd->vdev_islog = svd->vdev_islog;
855 vdev_top_update(vdev_t *tvd, vdev_t *vd)
862 for (int c = 0; c < vd->vdev_children; c++)
863 vdev_top_update(tvd, vd->vdev_child[c]);
867 * Add a mirror/replacing vdev above an existing vdev.
870 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
872 spa_t *spa = cvd->vdev_spa;
873 vdev_t *pvd = cvd->vdev_parent;
876 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
878 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
880 mvd->vdev_asize = cvd->vdev_asize;
881 mvd->vdev_min_asize = cvd->vdev_min_asize;
882 mvd->vdev_max_asize = cvd->vdev_max_asize;
883 mvd->vdev_ashift = cvd->vdev_ashift;
884 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
885 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
886 mvd->vdev_state = cvd->vdev_state;
887 mvd->vdev_crtxg = cvd->vdev_crtxg;
889 vdev_remove_child(pvd, cvd);
890 vdev_add_child(pvd, mvd);
891 cvd->vdev_id = mvd->vdev_children;
892 vdev_add_child(mvd, cvd);
893 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
895 if (mvd == mvd->vdev_top)
896 vdev_top_transfer(cvd, mvd);
902 * Remove a 1-way mirror/replacing vdev from the tree.
905 vdev_remove_parent(vdev_t *cvd)
907 vdev_t *mvd = cvd->vdev_parent;
908 vdev_t *pvd = mvd->vdev_parent;
910 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
912 ASSERT(mvd->vdev_children == 1);
913 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
914 mvd->vdev_ops == &vdev_replacing_ops ||
915 mvd->vdev_ops == &vdev_spare_ops);
916 cvd->vdev_ashift = mvd->vdev_ashift;
917 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
918 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
920 vdev_remove_child(mvd, cvd);
921 vdev_remove_child(pvd, mvd);
924 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
925 * Otherwise, we could have detached an offline device, and when we
926 * go to import the pool we'll think we have two top-level vdevs,
927 * instead of a different version of the same top-level vdev.
929 if (mvd->vdev_top == mvd) {
930 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
931 cvd->vdev_orig_guid = cvd->vdev_guid;
932 cvd->vdev_guid += guid_delta;
933 cvd->vdev_guid_sum += guid_delta;
935 cvd->vdev_id = mvd->vdev_id;
936 vdev_add_child(pvd, cvd);
937 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
939 if (cvd == cvd->vdev_top)
940 vdev_top_transfer(mvd, cvd);
942 ASSERT(mvd->vdev_children == 0);
947 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
949 spa_t *spa = vd->vdev_spa;
950 objset_t *mos = spa->spa_meta_objset;
952 uint64_t oldc = vd->vdev_ms_count;
953 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
957 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
960 * This vdev is not being allocated from yet or is a hole.
962 if (vd->vdev_ms_shift == 0)
965 ASSERT(!vd->vdev_ishole);
968 * Compute the raidz-deflation ratio. Note, we hard-code
969 * in 128k (1 << 17) because it is the "typical" blocksize.
970 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
971 * otherwise it would inconsistently account for existing bp's.
973 vd->vdev_deflate_ratio = (1 << 17) /
974 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
976 ASSERT(oldc <= newc);
978 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
981 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
982 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
986 vd->vdev_ms_count = newc;
988 for (m = oldc; m < newc; m++) {
992 error = dmu_read(mos, vd->vdev_ms_array,
993 m * sizeof (uint64_t), sizeof (uint64_t), &object,
999 error = metaslab_init(vd->vdev_mg, m, object, txg,
1006 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1009 * If the vdev is being removed we don't activate
1010 * the metaslabs since we want to ensure that no new
1011 * allocations are performed on this device.
1013 if (oldc == 0 && !vd->vdev_removing)
1014 metaslab_group_activate(vd->vdev_mg);
1017 spa_config_exit(spa, SCL_ALLOC, FTAG);
1023 vdev_metaslab_fini(vdev_t *vd)
1026 uint64_t count = vd->vdev_ms_count;
1028 if (vd->vdev_ms != NULL) {
1029 metaslab_group_passivate(vd->vdev_mg);
1030 for (m = 0; m < count; m++) {
1031 metaslab_t *msp = vd->vdev_ms[m];
1036 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1041 typedef struct vdev_probe_stats {
1042 boolean_t vps_readable;
1043 boolean_t vps_writeable;
1045 } vdev_probe_stats_t;
1048 vdev_probe_done(zio_t *zio)
1050 spa_t *spa = zio->io_spa;
1051 vdev_t *vd = zio->io_vd;
1052 vdev_probe_stats_t *vps = zio->io_private;
1054 ASSERT(vd->vdev_probe_zio != NULL);
1056 if (zio->io_type == ZIO_TYPE_READ) {
1057 if (zio->io_error == 0)
1058 vps->vps_readable = 1;
1059 if (zio->io_error == 0 && spa_writeable(spa)) {
1060 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1061 zio->io_offset, zio->io_size, zio->io_data,
1062 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1063 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1065 zio_buf_free(zio->io_data, zio->io_size);
1067 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1068 if (zio->io_error == 0)
1069 vps->vps_writeable = 1;
1070 zio_buf_free(zio->io_data, zio->io_size);
1071 } else if (zio->io_type == ZIO_TYPE_NULL) {
1074 vd->vdev_cant_read |= !vps->vps_readable;
1075 vd->vdev_cant_write |= !vps->vps_writeable;
1077 if (vdev_readable(vd) &&
1078 (vdev_writeable(vd) || !spa_writeable(spa))) {
1081 ASSERT(zio->io_error != 0);
1082 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1083 spa, vd, NULL, 0, 0);
1084 zio->io_error = SET_ERROR(ENXIO);
1087 mutex_enter(&vd->vdev_probe_lock);
1088 ASSERT(vd->vdev_probe_zio == zio);
1089 vd->vdev_probe_zio = NULL;
1090 mutex_exit(&vd->vdev_probe_lock);
1092 zio_link_t *zl = NULL;
1093 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1094 if (!vdev_accessible(vd, pio))
1095 pio->io_error = SET_ERROR(ENXIO);
1097 kmem_free(vps, sizeof (*vps));
1102 * Determine whether this device is accessible.
1104 * Read and write to several known locations: the pad regions of each
1105 * vdev label but the first, which we leave alone in case it contains
1109 vdev_probe(vdev_t *vd, zio_t *zio)
1111 spa_t *spa = vd->vdev_spa;
1112 vdev_probe_stats_t *vps = NULL;
1115 ASSERT(vd->vdev_ops->vdev_op_leaf);
1118 * Don't probe the probe.
1120 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1124 * To prevent 'probe storms' when a device fails, we create
1125 * just one probe i/o at a time. All zios that want to probe
1126 * this vdev will become parents of the probe io.
1128 mutex_enter(&vd->vdev_probe_lock);
1130 if ((pio = vd->vdev_probe_zio) == NULL) {
1131 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1133 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1134 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1137 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1139 * vdev_cant_read and vdev_cant_write can only
1140 * transition from TRUE to FALSE when we have the
1141 * SCL_ZIO lock as writer; otherwise they can only
1142 * transition from FALSE to TRUE. This ensures that
1143 * any zio looking at these values can assume that
1144 * failures persist for the life of the I/O. That's
1145 * important because when a device has intermittent
1146 * connectivity problems, we want to ensure that
1147 * they're ascribed to the device (ENXIO) and not
1150 * Since we hold SCL_ZIO as writer here, clear both
1151 * values so the probe can reevaluate from first
1154 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1155 vd->vdev_cant_read = B_FALSE;
1156 vd->vdev_cant_write = B_FALSE;
1159 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1160 vdev_probe_done, vps,
1161 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1164 * We can't change the vdev state in this context, so we
1165 * kick off an async task to do it on our behalf.
1168 vd->vdev_probe_wanted = B_TRUE;
1169 spa_async_request(spa, SPA_ASYNC_PROBE);
1174 zio_add_child(zio, pio);
1176 mutex_exit(&vd->vdev_probe_lock);
1179 ASSERT(zio != NULL);
1183 for (int l = 1; l < VDEV_LABELS; l++) {
1184 zio_nowait(zio_read_phys(pio, vd,
1185 vdev_label_offset(vd->vdev_psize, l,
1186 offsetof(vdev_label_t, vl_pad2)),
1187 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1188 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1189 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1200 vdev_open_child(void *arg)
1204 vd->vdev_open_thread = curthread;
1205 vd->vdev_open_error = vdev_open(vd);
1206 vd->vdev_open_thread = NULL;
1210 vdev_uses_zvols(vdev_t *vd)
1212 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1213 strlen(ZVOL_DIR)) == 0)
1215 for (int c = 0; c < vd->vdev_children; c++)
1216 if (vdev_uses_zvols(vd->vdev_child[c]))
1222 vdev_open_children(vdev_t *vd)
1225 int children = vd->vdev_children;
1228 * in order to handle pools on top of zvols, do the opens
1229 * in a single thread so that the same thread holds the
1230 * spa_namespace_lock
1232 if (B_TRUE || vdev_uses_zvols(vd)) {
1233 for (int c = 0; c < children; c++)
1234 vd->vdev_child[c]->vdev_open_error =
1235 vdev_open(vd->vdev_child[c]);
1238 tq = taskq_create("vdev_open", children, minclsyspri,
1239 children, children, TASKQ_PREPOPULATE);
1241 for (int c = 0; c < children; c++)
1242 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1249 * Prepare a virtual device for access.
1252 vdev_open(vdev_t *vd)
1254 spa_t *spa = vd->vdev_spa;
1257 uint64_t max_osize = 0;
1258 uint64_t asize, max_asize, psize;
1259 uint64_t logical_ashift = 0;
1260 uint64_t physical_ashift = 0;
1262 ASSERT(vd->vdev_open_thread == curthread ||
1263 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1264 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1265 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1266 vd->vdev_state == VDEV_STATE_OFFLINE);
1268 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1269 vd->vdev_cant_read = B_FALSE;
1270 vd->vdev_cant_write = B_FALSE;
1271 vd->vdev_notrim = B_FALSE;
1272 vd->vdev_min_asize = vdev_get_min_asize(vd);
1275 * If this vdev is not removed, check its fault status. If it's
1276 * faulted, bail out of the open.
1278 if (!vd->vdev_removed && vd->vdev_faulted) {
1279 ASSERT(vd->vdev_children == 0);
1280 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1281 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1282 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1283 vd->vdev_label_aux);
1284 return (SET_ERROR(ENXIO));
1285 } else if (vd->vdev_offline) {
1286 ASSERT(vd->vdev_children == 0);
1287 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1288 return (SET_ERROR(ENXIO));
1291 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1292 &logical_ashift, &physical_ashift);
1295 * Reset the vdev_reopening flag so that we actually close
1296 * the vdev on error.
1298 vd->vdev_reopening = B_FALSE;
1299 if (zio_injection_enabled && error == 0)
1300 error = zio_handle_device_injection(vd, NULL, ENXIO);
1303 if (vd->vdev_removed &&
1304 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1305 vd->vdev_removed = B_FALSE;
1307 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1308 vd->vdev_stat.vs_aux);
1312 vd->vdev_removed = B_FALSE;
1315 * Recheck the faulted flag now that we have confirmed that
1316 * the vdev is accessible. If we're faulted, bail.
1318 if (vd->vdev_faulted) {
1319 ASSERT(vd->vdev_children == 0);
1320 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1321 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1322 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1323 vd->vdev_label_aux);
1324 return (SET_ERROR(ENXIO));
1327 if (vd->vdev_degraded) {
1328 ASSERT(vd->vdev_children == 0);
1329 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1330 VDEV_AUX_ERR_EXCEEDED);
1332 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1336 * For hole or missing vdevs we just return success.
1338 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1341 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1342 trim_map_create(vd);
1344 for (int c = 0; c < vd->vdev_children; c++) {
1345 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1346 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1352 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1353 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1355 if (vd->vdev_children == 0) {
1356 if (osize < SPA_MINDEVSIZE) {
1357 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1358 VDEV_AUX_TOO_SMALL);
1359 return (SET_ERROR(EOVERFLOW));
1362 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1363 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1364 VDEV_LABEL_END_SIZE);
1366 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1367 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1368 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1369 VDEV_AUX_TOO_SMALL);
1370 return (SET_ERROR(EOVERFLOW));
1374 max_asize = max_osize;
1377 vd->vdev_psize = psize;
1380 * Make sure the allocatable size hasn't shrunk.
1382 if (asize < vd->vdev_min_asize) {
1383 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1384 VDEV_AUX_BAD_LABEL);
1385 return (SET_ERROR(EINVAL));
1388 vd->vdev_physical_ashift =
1389 MAX(physical_ashift, vd->vdev_physical_ashift);
1390 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1391 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1393 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1394 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1395 VDEV_AUX_ASHIFT_TOO_BIG);
1399 if (vd->vdev_asize == 0) {
1401 * This is the first-ever open, so use the computed values.
1402 * For testing purposes, a higher ashift can be requested.
1404 vd->vdev_asize = asize;
1405 vd->vdev_max_asize = max_asize;
1408 * Make sure the alignment requirement hasn't increased.
1410 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1411 vd->vdev_ops->vdev_op_leaf) {
1412 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1413 VDEV_AUX_BAD_LABEL);
1416 vd->vdev_max_asize = max_asize;
1420 * If all children are healthy and the asize has increased,
1421 * then we've experienced dynamic LUN growth. If automatic
1422 * expansion is enabled then use the additional space.
1424 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1425 (vd->vdev_expanding || spa->spa_autoexpand))
1426 vd->vdev_asize = asize;
1428 vdev_set_min_asize(vd);
1431 * Ensure we can issue some IO before declaring the
1432 * vdev open for business.
1434 if (vd->vdev_ops->vdev_op_leaf &&
1435 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1436 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1437 VDEV_AUX_ERR_EXCEEDED);
1442 * Track the min and max ashift values for normal data devices.
1444 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1445 !vd->vdev_islog && vd->vdev_aux == NULL) {
1446 if (vd->vdev_ashift > spa->spa_max_ashift)
1447 spa->spa_max_ashift = vd->vdev_ashift;
1448 if (vd->vdev_ashift < spa->spa_min_ashift)
1449 spa->spa_min_ashift = vd->vdev_ashift;
1453 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1454 * resilver. But don't do this if we are doing a reopen for a scrub,
1455 * since this would just restart the scrub we are already doing.
1457 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1458 vdev_resilver_needed(vd, NULL, NULL))
1459 spa_async_request(spa, SPA_ASYNC_RESILVER);
1465 * Called once the vdevs are all opened, this routine validates the label
1466 * contents. This needs to be done before vdev_load() so that we don't
1467 * inadvertently do repair I/Os to the wrong device.
1469 * If 'strict' is false ignore the spa guid check. This is necessary because
1470 * if the machine crashed during a re-guid the new guid might have been written
1471 * to all of the vdev labels, but not the cached config. The strict check
1472 * will be performed when the pool is opened again using the mos config.
1474 * This function will only return failure if one of the vdevs indicates that it
1475 * has since been destroyed or exported. This is only possible if
1476 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1477 * will be updated but the function will return 0.
1480 vdev_validate(vdev_t *vd, boolean_t strict)
1482 spa_t *spa = vd->vdev_spa;
1484 uint64_t guid = 0, top_guid;
1487 for (int c = 0; c < vd->vdev_children; c++)
1488 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1489 return (SET_ERROR(EBADF));
1492 * If the device has already failed, or was marked offline, don't do
1493 * any further validation. Otherwise, label I/O will fail and we will
1494 * overwrite the previous state.
1496 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1497 uint64_t aux_guid = 0;
1499 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1500 spa_last_synced_txg(spa) : -1ULL;
1502 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1503 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1504 VDEV_AUX_BAD_LABEL);
1509 * Determine if this vdev has been split off into another
1510 * pool. If so, then refuse to open it.
1512 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1513 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1514 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1515 VDEV_AUX_SPLIT_POOL);
1520 if (strict && (nvlist_lookup_uint64(label,
1521 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1522 guid != spa_guid(spa))) {
1523 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1524 VDEV_AUX_CORRUPT_DATA);
1529 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1530 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1535 * If this vdev just became a top-level vdev because its
1536 * sibling was detached, it will have adopted the parent's
1537 * vdev guid -- but the label may or may not be on disk yet.
1538 * Fortunately, either version of the label will have the
1539 * same top guid, so if we're a top-level vdev, we can
1540 * safely compare to that instead.
1542 * If we split this vdev off instead, then we also check the
1543 * original pool's guid. We don't want to consider the vdev
1544 * corrupt if it is partway through a split operation.
1546 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1548 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1550 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1551 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1552 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1553 VDEV_AUX_CORRUPT_DATA);
1558 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1560 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1561 VDEV_AUX_CORRUPT_DATA);
1569 * If this is a verbatim import, no need to check the
1570 * state of the pool.
1572 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1573 spa_load_state(spa) == SPA_LOAD_OPEN &&
1574 state != POOL_STATE_ACTIVE)
1575 return (SET_ERROR(EBADF));
1578 * If we were able to open and validate a vdev that was
1579 * previously marked permanently unavailable, clear that state
1582 if (vd->vdev_not_present)
1583 vd->vdev_not_present = 0;
1590 * Close a virtual device.
1593 vdev_close(vdev_t *vd)
1595 spa_t *spa = vd->vdev_spa;
1596 vdev_t *pvd = vd->vdev_parent;
1598 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1601 * If our parent is reopening, then we are as well, unless we are
1604 if (pvd != NULL && pvd->vdev_reopening)
1605 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1607 vd->vdev_ops->vdev_op_close(vd);
1609 vdev_cache_purge(vd);
1611 if (vd->vdev_ops->vdev_op_leaf)
1612 trim_map_destroy(vd);
1615 * We record the previous state before we close it, so that if we are
1616 * doing a reopen(), we don't generate FMA ereports if we notice that
1617 * it's still faulted.
1619 vd->vdev_prevstate = vd->vdev_state;
1621 if (vd->vdev_offline)
1622 vd->vdev_state = VDEV_STATE_OFFLINE;
1624 vd->vdev_state = VDEV_STATE_CLOSED;
1625 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1629 vdev_hold(vdev_t *vd)
1631 spa_t *spa = vd->vdev_spa;
1633 ASSERT(spa_is_root(spa));
1634 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1637 for (int c = 0; c < vd->vdev_children; c++)
1638 vdev_hold(vd->vdev_child[c]);
1640 if (vd->vdev_ops->vdev_op_leaf)
1641 vd->vdev_ops->vdev_op_hold(vd);
1645 vdev_rele(vdev_t *vd)
1647 spa_t *spa = vd->vdev_spa;
1649 ASSERT(spa_is_root(spa));
1650 for (int c = 0; c < vd->vdev_children; c++)
1651 vdev_rele(vd->vdev_child[c]);
1653 if (vd->vdev_ops->vdev_op_leaf)
1654 vd->vdev_ops->vdev_op_rele(vd);
1658 * Reopen all interior vdevs and any unopened leaves. We don't actually
1659 * reopen leaf vdevs which had previously been opened as they might deadlock
1660 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1661 * If the leaf has never been opened then open it, as usual.
1664 vdev_reopen(vdev_t *vd)
1666 spa_t *spa = vd->vdev_spa;
1668 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1670 /* set the reopening flag unless we're taking the vdev offline */
1671 vd->vdev_reopening = !vd->vdev_offline;
1673 (void) vdev_open(vd);
1676 * Call vdev_validate() here to make sure we have the same device.
1677 * Otherwise, a device with an invalid label could be successfully
1678 * opened in response to vdev_reopen().
1681 (void) vdev_validate_aux(vd);
1682 if (vdev_readable(vd) && vdev_writeable(vd) &&
1683 vd->vdev_aux == &spa->spa_l2cache &&
1684 !l2arc_vdev_present(vd))
1685 l2arc_add_vdev(spa, vd);
1687 (void) vdev_validate(vd, B_TRUE);
1691 * Reassess parent vdev's health.
1693 vdev_propagate_state(vd);
1697 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1702 * Normally, partial opens (e.g. of a mirror) are allowed.
1703 * For a create, however, we want to fail the request if
1704 * there are any components we can't open.
1706 error = vdev_open(vd);
1708 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1710 return (error ? error : ENXIO);
1714 * Recursively load DTLs and initialize all labels.
1716 if ((error = vdev_dtl_load(vd)) != 0 ||
1717 (error = vdev_label_init(vd, txg, isreplacing ?
1718 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1727 vdev_metaslab_set_size(vdev_t *vd)
1730 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1732 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1733 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1737 * Maximize performance by inflating the configured ashift for top level
1738 * vdevs to be as close to the physical ashift as possible while maintaining
1739 * administrator defined limits and ensuring it doesn't go below the
1743 vdev_ashift_optimize(vdev_t *vd)
1745 if (vd == vd->vdev_top) {
1746 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1747 vd->vdev_ashift = MIN(
1748 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1749 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1752 * Unusual case where logical ashift > physical ashift
1753 * so we can't cap the calculated ashift based on max
1754 * ashift as that would cause failures.
1755 * We still check if we need to increase it to match
1758 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1765 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1767 ASSERT(vd == vd->vdev_top);
1768 ASSERT(!vd->vdev_ishole);
1769 ASSERT(ISP2(flags));
1770 ASSERT(spa_writeable(vd->vdev_spa));
1772 if (flags & VDD_METASLAB)
1773 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1775 if (flags & VDD_DTL)
1776 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1778 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1782 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1784 for (int c = 0; c < vd->vdev_children; c++)
1785 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1787 if (vd->vdev_ops->vdev_op_leaf)
1788 vdev_dirty(vd->vdev_top, flags, vd, txg);
1794 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1795 * the vdev has less than perfect replication. There are four kinds of DTL:
1797 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1799 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1801 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1802 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1803 * txgs that was scrubbed.
1805 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1806 * persistent errors or just some device being offline.
1807 * Unlike the other three, the DTL_OUTAGE map is not generally
1808 * maintained; it's only computed when needed, typically to
1809 * determine whether a device can be detached.
1811 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1812 * either has the data or it doesn't.
1814 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1815 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1816 * if any child is less than fully replicated, then so is its parent.
1817 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1818 * comprising only those txgs which appear in 'maxfaults' or more children;
1819 * those are the txgs we don't have enough replication to read. For example,
1820 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1821 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1822 * two child DTL_MISSING maps.
1824 * It should be clear from the above that to compute the DTLs and outage maps
1825 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1826 * Therefore, that is all we keep on disk. When loading the pool, or after
1827 * a configuration change, we generate all other DTLs from first principles.
1830 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1832 range_tree_t *rt = vd->vdev_dtl[t];
1834 ASSERT(t < DTL_TYPES);
1835 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1836 ASSERT(spa_writeable(vd->vdev_spa));
1838 mutex_enter(rt->rt_lock);
1839 if (!range_tree_contains(rt, txg, size))
1840 range_tree_add(rt, txg, size);
1841 mutex_exit(rt->rt_lock);
1845 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1847 range_tree_t *rt = vd->vdev_dtl[t];
1848 boolean_t dirty = B_FALSE;
1850 ASSERT(t < DTL_TYPES);
1851 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1853 mutex_enter(rt->rt_lock);
1854 if (range_tree_space(rt) != 0)
1855 dirty = range_tree_contains(rt, txg, size);
1856 mutex_exit(rt->rt_lock);
1862 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1864 range_tree_t *rt = vd->vdev_dtl[t];
1867 mutex_enter(rt->rt_lock);
1868 empty = (range_tree_space(rt) == 0);
1869 mutex_exit(rt->rt_lock);
1875 * Returns the lowest txg in the DTL range.
1878 vdev_dtl_min(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_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1887 return (rs->rs_start - 1);
1891 * Returns the highest txg in the DTL.
1894 vdev_dtl_max(vdev_t *vd)
1898 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1899 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1900 ASSERT0(vd->vdev_children);
1902 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1903 return (rs->rs_end);
1907 * Determine if a resilvering vdev should remove any DTL entries from
1908 * its range. If the vdev was resilvering for the entire duration of the
1909 * scan then it should excise that range from its DTLs. Otherwise, this
1910 * vdev is considered partially resilvered and should leave its DTL
1911 * entries intact. The comment in vdev_dtl_reassess() describes how we
1915 vdev_dtl_should_excise(vdev_t *vd)
1917 spa_t *spa = vd->vdev_spa;
1918 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1920 ASSERT0(scn->scn_phys.scn_errors);
1921 ASSERT0(vd->vdev_children);
1923 if (vd->vdev_resilver_txg == 0 ||
1924 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1928 * When a resilver is initiated the scan will assign the scn_max_txg
1929 * value to the highest txg value that exists in all DTLs. If this
1930 * device's max DTL is not part of this scan (i.e. it is not in
1931 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1934 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1935 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1936 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1937 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1944 * Reassess DTLs after a config change or scrub completion.
1947 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1949 spa_t *spa = vd->vdev_spa;
1953 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1955 for (int c = 0; c < vd->vdev_children; c++)
1956 vdev_dtl_reassess(vd->vdev_child[c], txg,
1957 scrub_txg, scrub_done);
1959 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1962 if (vd->vdev_ops->vdev_op_leaf) {
1963 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1965 mutex_enter(&vd->vdev_dtl_lock);
1968 * If we've completed a scan cleanly then determine
1969 * if this vdev should remove any DTLs. We only want to
1970 * excise regions on vdevs that were available during
1971 * the entire duration of this scan.
1973 if (scrub_txg != 0 &&
1974 (spa->spa_scrub_started ||
1975 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1976 vdev_dtl_should_excise(vd)) {
1978 * We completed a scrub up to scrub_txg. If we
1979 * did it without rebooting, then the scrub dtl
1980 * will be valid, so excise the old region and
1981 * fold in the scrub dtl. Otherwise, leave the
1982 * dtl as-is if there was an error.
1984 * There's little trick here: to excise the beginning
1985 * of the DTL_MISSING map, we put it into a reference
1986 * tree and then add a segment with refcnt -1 that
1987 * covers the range [0, scrub_txg). This means
1988 * that each txg in that range has refcnt -1 or 0.
1989 * We then add DTL_SCRUB with a refcnt of 2, so that
1990 * entries in the range [0, scrub_txg) will have a
1991 * positive refcnt -- either 1 or 2. We then convert
1992 * the reference tree into the new DTL_MISSING map.
1994 space_reftree_create(&reftree);
1995 space_reftree_add_map(&reftree,
1996 vd->vdev_dtl[DTL_MISSING], 1);
1997 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1998 space_reftree_add_map(&reftree,
1999 vd->vdev_dtl[DTL_SCRUB], 2);
2000 space_reftree_generate_map(&reftree,
2001 vd->vdev_dtl[DTL_MISSING], 1);
2002 space_reftree_destroy(&reftree);
2004 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2005 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2006 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2008 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2009 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2010 if (!vdev_readable(vd))
2011 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2013 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2014 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2017 * If the vdev was resilvering and no longer has any
2018 * DTLs then reset its resilvering flag and dirty
2019 * the top level so that we persist the change.
2021 if (vd->vdev_resilver_txg != 0 &&
2022 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
2023 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
2024 vd->vdev_resilver_txg = 0;
2025 vdev_config_dirty(vd->vdev_top);
2028 mutex_exit(&vd->vdev_dtl_lock);
2031 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2035 mutex_enter(&vd->vdev_dtl_lock);
2036 for (int t = 0; t < DTL_TYPES; t++) {
2037 /* account for child's outage in parent's missing map */
2038 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2040 continue; /* leaf vdevs only */
2041 if (t == DTL_PARTIAL)
2042 minref = 1; /* i.e. non-zero */
2043 else if (vd->vdev_nparity != 0)
2044 minref = vd->vdev_nparity + 1; /* RAID-Z */
2046 minref = vd->vdev_children; /* any kind of mirror */
2047 space_reftree_create(&reftree);
2048 for (int c = 0; c < vd->vdev_children; c++) {
2049 vdev_t *cvd = vd->vdev_child[c];
2050 mutex_enter(&cvd->vdev_dtl_lock);
2051 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2052 mutex_exit(&cvd->vdev_dtl_lock);
2054 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2055 space_reftree_destroy(&reftree);
2057 mutex_exit(&vd->vdev_dtl_lock);
2061 vdev_dtl_load(vdev_t *vd)
2063 spa_t *spa = vd->vdev_spa;
2064 objset_t *mos = spa->spa_meta_objset;
2067 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2068 ASSERT(!vd->vdev_ishole);
2070 error = space_map_open(&vd->vdev_dtl_sm, mos,
2071 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2074 ASSERT(vd->vdev_dtl_sm != NULL);
2076 mutex_enter(&vd->vdev_dtl_lock);
2079 * Now that we've opened the space_map we need to update
2082 space_map_update(vd->vdev_dtl_sm);
2084 error = space_map_load(vd->vdev_dtl_sm,
2085 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2086 mutex_exit(&vd->vdev_dtl_lock);
2091 for (int c = 0; c < vd->vdev_children; c++) {
2092 error = vdev_dtl_load(vd->vdev_child[c]);
2101 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2103 spa_t *spa = vd->vdev_spa;
2105 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2106 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2111 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2113 spa_t *spa = vd->vdev_spa;
2114 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2115 DMU_OT_NONE, 0, tx);
2118 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2125 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2127 if (vd->vdev_ops != &vdev_hole_ops &&
2128 vd->vdev_ops != &vdev_missing_ops &&
2129 vd->vdev_ops != &vdev_root_ops &&
2130 !vd->vdev_top->vdev_removing) {
2131 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2132 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2134 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2135 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2138 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2139 vdev_construct_zaps(vd->vdev_child[i], tx);
2144 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2146 spa_t *spa = vd->vdev_spa;
2147 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2148 objset_t *mos = spa->spa_meta_objset;
2149 range_tree_t *rtsync;
2152 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2154 ASSERT(!vd->vdev_ishole);
2155 ASSERT(vd->vdev_ops->vdev_op_leaf);
2157 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2159 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2160 mutex_enter(&vd->vdev_dtl_lock);
2161 space_map_free(vd->vdev_dtl_sm, tx);
2162 space_map_close(vd->vdev_dtl_sm);
2163 vd->vdev_dtl_sm = NULL;
2164 mutex_exit(&vd->vdev_dtl_lock);
2167 * We only destroy the leaf ZAP for detached leaves or for
2168 * removed log devices. Removed data devices handle leaf ZAP
2169 * cleanup later, once cancellation is no longer possible.
2171 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2172 vd->vdev_top->vdev_islog)) {
2173 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2174 vd->vdev_leaf_zap = 0;
2181 if (vd->vdev_dtl_sm == NULL) {
2182 uint64_t new_object;
2184 new_object = space_map_alloc(mos, tx);
2185 VERIFY3U(new_object, !=, 0);
2187 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2188 0, -1ULL, 0, &vd->vdev_dtl_lock));
2189 ASSERT(vd->vdev_dtl_sm != NULL);
2192 bzero(&rtlock, sizeof(rtlock));
2193 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2195 rtsync = range_tree_create(NULL, NULL, &rtlock);
2197 mutex_enter(&rtlock);
2199 mutex_enter(&vd->vdev_dtl_lock);
2200 range_tree_walk(rt, range_tree_add, rtsync);
2201 mutex_exit(&vd->vdev_dtl_lock);
2203 space_map_truncate(vd->vdev_dtl_sm, tx);
2204 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2205 range_tree_vacate(rtsync, NULL, NULL);
2207 range_tree_destroy(rtsync);
2209 mutex_exit(&rtlock);
2210 mutex_destroy(&rtlock);
2213 * If the object for the space map has changed then dirty
2214 * the top level so that we update the config.
2216 if (object != space_map_object(vd->vdev_dtl_sm)) {
2217 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2218 "new object %llu", txg, spa_name(spa), object,
2219 space_map_object(vd->vdev_dtl_sm));
2220 vdev_config_dirty(vd->vdev_top);
2225 mutex_enter(&vd->vdev_dtl_lock);
2226 space_map_update(vd->vdev_dtl_sm);
2227 mutex_exit(&vd->vdev_dtl_lock);
2231 * Determine whether the specified vdev can be offlined/detached/removed
2232 * without losing data.
2235 vdev_dtl_required(vdev_t *vd)
2237 spa_t *spa = vd->vdev_spa;
2238 vdev_t *tvd = vd->vdev_top;
2239 uint8_t cant_read = vd->vdev_cant_read;
2242 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2244 if (vd == spa->spa_root_vdev || vd == tvd)
2248 * Temporarily mark the device as unreadable, and then determine
2249 * whether this results in any DTL outages in the top-level vdev.
2250 * If not, we can safely offline/detach/remove the device.
2252 vd->vdev_cant_read = B_TRUE;
2253 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2254 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2255 vd->vdev_cant_read = cant_read;
2256 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2258 if (!required && zio_injection_enabled)
2259 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2265 * Determine if resilver is needed, and if so the txg range.
2268 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2270 boolean_t needed = B_FALSE;
2271 uint64_t thismin = UINT64_MAX;
2272 uint64_t thismax = 0;
2274 if (vd->vdev_children == 0) {
2275 mutex_enter(&vd->vdev_dtl_lock);
2276 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2277 vdev_writeable(vd)) {
2279 thismin = vdev_dtl_min(vd);
2280 thismax = vdev_dtl_max(vd);
2283 mutex_exit(&vd->vdev_dtl_lock);
2285 for (int c = 0; c < vd->vdev_children; c++) {
2286 vdev_t *cvd = vd->vdev_child[c];
2287 uint64_t cmin, cmax;
2289 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2290 thismin = MIN(thismin, cmin);
2291 thismax = MAX(thismax, cmax);
2297 if (needed && minp) {
2305 vdev_load(vdev_t *vd)
2308 * Recursively load all children.
2310 for (int c = 0; c < vd->vdev_children; c++)
2311 vdev_load(vd->vdev_child[c]);
2314 * If this is a top-level vdev, initialize its metaslabs.
2316 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2317 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2318 vdev_metaslab_init(vd, 0) != 0))
2319 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2320 VDEV_AUX_CORRUPT_DATA);
2323 * If this is a leaf vdev, load its DTL.
2325 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2326 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2327 VDEV_AUX_CORRUPT_DATA);
2331 * The special vdev case is used for hot spares and l2cache devices. Its
2332 * sole purpose it to set the vdev state for the associated vdev. To do this,
2333 * we make sure that we can open the underlying device, then try to read the
2334 * label, and make sure that the label is sane and that it hasn't been
2335 * repurposed to another pool.
2338 vdev_validate_aux(vdev_t *vd)
2341 uint64_t guid, version;
2344 if (!vdev_readable(vd))
2347 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2348 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2349 VDEV_AUX_CORRUPT_DATA);
2353 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2354 !SPA_VERSION_IS_SUPPORTED(version) ||
2355 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2356 guid != vd->vdev_guid ||
2357 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2358 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2359 VDEV_AUX_CORRUPT_DATA);
2365 * We don't actually check the pool state here. If it's in fact in
2366 * use by another pool, we update this fact on the fly when requested.
2373 vdev_remove(vdev_t *vd, uint64_t txg)
2375 spa_t *spa = vd->vdev_spa;
2376 objset_t *mos = spa->spa_meta_objset;
2379 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2380 ASSERT(vd == vd->vdev_top);
2381 ASSERT3U(txg, ==, spa_syncing_txg(spa));
2383 if (vd->vdev_ms != NULL) {
2384 metaslab_group_t *mg = vd->vdev_mg;
2386 metaslab_group_histogram_verify(mg);
2387 metaslab_class_histogram_verify(mg->mg_class);
2389 for (int m = 0; m < vd->vdev_ms_count; m++) {
2390 metaslab_t *msp = vd->vdev_ms[m];
2392 if (msp == NULL || msp->ms_sm == NULL)
2395 mutex_enter(&msp->ms_lock);
2397 * If the metaslab was not loaded when the vdev
2398 * was removed then the histogram accounting may
2399 * not be accurate. Update the histogram information
2400 * here so that we ensure that the metaslab group
2401 * and metaslab class are up-to-date.
2403 metaslab_group_histogram_remove(mg, msp);
2405 VERIFY0(space_map_allocated(msp->ms_sm));
2406 space_map_free(msp->ms_sm, tx);
2407 space_map_close(msp->ms_sm);
2409 mutex_exit(&msp->ms_lock);
2412 metaslab_group_histogram_verify(mg);
2413 metaslab_class_histogram_verify(mg->mg_class);
2414 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2415 ASSERT0(mg->mg_histogram[i]);
2419 if (vd->vdev_ms_array) {
2420 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2421 vd->vdev_ms_array = 0;
2424 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
2425 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
2426 vd->vdev_top_zap = 0;
2432 vdev_sync_done(vdev_t *vd, uint64_t txg)
2435 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2437 ASSERT(!vd->vdev_ishole);
2439 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2440 metaslab_sync_done(msp, txg);
2443 metaslab_sync_reassess(vd->vdev_mg);
2447 vdev_sync(vdev_t *vd, uint64_t txg)
2449 spa_t *spa = vd->vdev_spa;
2454 ASSERT(!vd->vdev_ishole);
2456 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2457 ASSERT(vd == vd->vdev_top);
2458 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2459 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2460 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2461 ASSERT(vd->vdev_ms_array != 0);
2462 vdev_config_dirty(vd);
2467 * Remove the metadata associated with this vdev once it's empty.
2469 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2470 vdev_remove(vd, txg);
2472 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2473 metaslab_sync(msp, txg);
2474 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2477 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2478 vdev_dtl_sync(lvd, txg);
2480 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2484 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2486 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2490 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2491 * not be opened, and no I/O is attempted.
2494 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2498 spa_vdev_state_enter(spa, SCL_NONE);
2500 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2501 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2503 if (!vd->vdev_ops->vdev_op_leaf)
2504 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2509 * We don't directly use the aux state here, but if we do a
2510 * vdev_reopen(), we need this value to be present to remember why we
2513 vd->vdev_label_aux = aux;
2516 * Faulted state takes precedence over degraded.
2518 vd->vdev_delayed_close = B_FALSE;
2519 vd->vdev_faulted = 1ULL;
2520 vd->vdev_degraded = 0ULL;
2521 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2524 * If this device has the only valid copy of the data, then
2525 * back off and simply mark the vdev as degraded instead.
2527 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2528 vd->vdev_degraded = 1ULL;
2529 vd->vdev_faulted = 0ULL;
2532 * If we reopen the device and it's not dead, only then do we
2537 if (vdev_readable(vd))
2538 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2541 return (spa_vdev_state_exit(spa, vd, 0));
2545 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2546 * user that something is wrong. The vdev continues to operate as normal as far
2547 * as I/O is concerned.
2550 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2554 spa_vdev_state_enter(spa, SCL_NONE);
2556 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2557 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2559 if (!vd->vdev_ops->vdev_op_leaf)
2560 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2563 * If the vdev is already faulted, then don't do anything.
2565 if (vd->vdev_faulted || vd->vdev_degraded)
2566 return (spa_vdev_state_exit(spa, NULL, 0));
2568 vd->vdev_degraded = 1ULL;
2569 if (!vdev_is_dead(vd))
2570 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2573 return (spa_vdev_state_exit(spa, vd, 0));
2577 * Online the given vdev.
2579 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2580 * spare device should be detached when the device finishes resilvering.
2581 * Second, the online should be treated like a 'test' online case, so no FMA
2582 * events are generated if the device fails to open.
2585 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2587 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2588 boolean_t postevent = B_FALSE;
2590 spa_vdev_state_enter(spa, SCL_NONE);
2592 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2593 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2595 if (!vd->vdev_ops->vdev_op_leaf)
2596 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2599 (vd->vdev_offline == B_TRUE || vd->vdev_tmpoffline == B_TRUE) ?
2603 vd->vdev_offline = B_FALSE;
2604 vd->vdev_tmpoffline = B_FALSE;
2605 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2606 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2608 /* XXX - L2ARC 1.0 does not support expansion */
2609 if (!vd->vdev_aux) {
2610 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2611 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2615 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2617 if (!vd->vdev_aux) {
2618 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2619 pvd->vdev_expanding = B_FALSE;
2623 *newstate = vd->vdev_state;
2624 if ((flags & ZFS_ONLINE_UNSPARE) &&
2625 !vdev_is_dead(vd) && vd->vdev_parent &&
2626 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2627 vd->vdev_parent->vdev_child[0] == vd)
2628 vd->vdev_unspare = B_TRUE;
2630 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2632 /* XXX - L2ARC 1.0 does not support expansion */
2634 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2635 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2639 spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE);
2641 return (spa_vdev_state_exit(spa, vd, 0));
2645 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2649 uint64_t generation;
2650 metaslab_group_t *mg;
2653 spa_vdev_state_enter(spa, SCL_ALLOC);
2655 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2656 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2658 if (!vd->vdev_ops->vdev_op_leaf)
2659 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2663 generation = spa->spa_config_generation + 1;
2666 * If the device isn't already offline, try to offline it.
2668 if (!vd->vdev_offline) {
2670 * If this device has the only valid copy of some data,
2671 * don't allow it to be offlined. Log devices are always
2674 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2675 vdev_dtl_required(vd))
2676 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2679 * If the top-level is a slog and it has had allocations
2680 * then proceed. We check that the vdev's metaslab group
2681 * is not NULL since it's possible that we may have just
2682 * added this vdev but not yet initialized its metaslabs.
2684 if (tvd->vdev_islog && mg != NULL) {
2686 * Prevent any future allocations.
2688 metaslab_group_passivate(mg);
2689 (void) spa_vdev_state_exit(spa, vd, 0);
2691 error = spa_offline_log(spa);
2693 spa_vdev_state_enter(spa, SCL_ALLOC);
2696 * Check to see if the config has changed.
2698 if (error || generation != spa->spa_config_generation) {
2699 metaslab_group_activate(mg);
2701 return (spa_vdev_state_exit(spa,
2703 (void) spa_vdev_state_exit(spa, vd, 0);
2706 ASSERT0(tvd->vdev_stat.vs_alloc);
2710 * Offline this device and reopen its top-level vdev.
2711 * If the top-level vdev is a log device then just offline
2712 * it. Otherwise, if this action results in the top-level
2713 * vdev becoming unusable, undo it and fail the request.
2715 vd->vdev_offline = B_TRUE;
2718 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2719 vdev_is_dead(tvd)) {
2720 vd->vdev_offline = B_FALSE;
2722 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2726 * Add the device back into the metaslab rotor so that
2727 * once we online the device it's open for business.
2729 if (tvd->vdev_islog && mg != NULL)
2730 metaslab_group_activate(mg);
2733 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2735 return (spa_vdev_state_exit(spa, vd, 0));
2739 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2743 mutex_enter(&spa->spa_vdev_top_lock);
2744 error = vdev_offline_locked(spa, guid, flags);
2745 mutex_exit(&spa->spa_vdev_top_lock);
2751 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2752 * vdev_offline(), we assume the spa config is locked. We also clear all
2753 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2756 vdev_clear(spa_t *spa, vdev_t *vd)
2758 vdev_t *rvd = spa->spa_root_vdev;
2760 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2765 vd->vdev_stat.vs_read_errors = 0;
2766 vd->vdev_stat.vs_write_errors = 0;
2767 vd->vdev_stat.vs_checksum_errors = 0;
2769 for (int c = 0; c < vd->vdev_children; c++)
2770 vdev_clear(spa, vd->vdev_child[c]);
2773 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2774 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2776 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2777 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2781 * If we're in the FAULTED state or have experienced failed I/O, then
2782 * clear the persistent state and attempt to reopen the device. We
2783 * also mark the vdev config dirty, so that the new faulted state is
2784 * written out to disk.
2786 if (vd->vdev_faulted || vd->vdev_degraded ||
2787 !vdev_readable(vd) || !vdev_writeable(vd)) {
2790 * When reopening in reponse to a clear event, it may be due to
2791 * a fmadm repair request. In this case, if the device is
2792 * still broken, we want to still post the ereport again.
2794 vd->vdev_forcefault = B_TRUE;
2796 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2797 vd->vdev_cant_read = B_FALSE;
2798 vd->vdev_cant_write = B_FALSE;
2800 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2802 vd->vdev_forcefault = B_FALSE;
2804 if (vd != rvd && vdev_writeable(vd->vdev_top))
2805 vdev_state_dirty(vd->vdev_top);
2807 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2808 spa_async_request(spa, SPA_ASYNC_RESILVER);
2810 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2814 * When clearing a FMA-diagnosed fault, we always want to
2815 * unspare the device, as we assume that the original spare was
2816 * done in response to the FMA fault.
2818 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2819 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2820 vd->vdev_parent->vdev_child[0] == vd)
2821 vd->vdev_unspare = B_TRUE;
2825 vdev_is_dead(vdev_t *vd)
2828 * Holes and missing devices are always considered "dead".
2829 * This simplifies the code since we don't have to check for
2830 * these types of devices in the various code paths.
2831 * Instead we rely on the fact that we skip over dead devices
2832 * before issuing I/O to them.
2834 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2835 vd->vdev_ops == &vdev_missing_ops);
2839 vdev_readable(vdev_t *vd)
2841 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2845 vdev_writeable(vdev_t *vd)
2847 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2851 vdev_allocatable(vdev_t *vd)
2853 uint64_t state = vd->vdev_state;
2856 * We currently allow allocations from vdevs which may be in the
2857 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2858 * fails to reopen then we'll catch it later when we're holding
2859 * the proper locks. Note that we have to get the vdev state
2860 * in a local variable because although it changes atomically,
2861 * we're asking two separate questions about it.
2863 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2864 !vd->vdev_cant_write && !vd->vdev_ishole &&
2865 vd->vdev_mg->mg_initialized);
2869 vdev_accessible(vdev_t *vd, zio_t *zio)
2871 ASSERT(zio->io_vd == vd);
2873 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2876 if (zio->io_type == ZIO_TYPE_READ)
2877 return (!vd->vdev_cant_read);
2879 if (zio->io_type == ZIO_TYPE_WRITE)
2880 return (!vd->vdev_cant_write);
2886 * Get statistics for the given vdev.
2889 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2891 spa_t *spa = vd->vdev_spa;
2892 vdev_t *rvd = spa->spa_root_vdev;
2893 vdev_t *tvd = vd->vdev_top;
2895 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2897 mutex_enter(&vd->vdev_stat_lock);
2898 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2899 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2900 vs->vs_state = vd->vdev_state;
2901 vs->vs_rsize = vdev_get_min_asize(vd);
2902 if (vd->vdev_ops->vdev_op_leaf)
2903 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2905 * Report expandable space on top-level, non-auxillary devices only.
2906 * The expandable space is reported in terms of metaslab sized units
2907 * since that determines how much space the pool can expand.
2909 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
2910 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize,
2911 1ULL << tvd->vdev_ms_shift);
2913 vs->vs_configured_ashift = vd->vdev_top != NULL
2914 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2915 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2916 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2917 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2918 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2922 * If we're getting stats on the root vdev, aggregate the I/O counts
2923 * over all top-level vdevs (i.e. the direct children of the root).
2926 for (int c = 0; c < rvd->vdev_children; c++) {
2927 vdev_t *cvd = rvd->vdev_child[c];
2928 vdev_stat_t *cvs = &cvd->vdev_stat;
2930 for (int t = 0; t < ZIO_TYPES; t++) {
2931 vs->vs_ops[t] += cvs->vs_ops[t];
2932 vs->vs_bytes[t] += cvs->vs_bytes[t];
2934 cvs->vs_scan_removing = cvd->vdev_removing;
2937 mutex_exit(&vd->vdev_stat_lock);
2941 vdev_clear_stats(vdev_t *vd)
2943 mutex_enter(&vd->vdev_stat_lock);
2944 vd->vdev_stat.vs_space = 0;
2945 vd->vdev_stat.vs_dspace = 0;
2946 vd->vdev_stat.vs_alloc = 0;
2947 mutex_exit(&vd->vdev_stat_lock);
2951 vdev_scan_stat_init(vdev_t *vd)
2953 vdev_stat_t *vs = &vd->vdev_stat;
2955 for (int c = 0; c < vd->vdev_children; c++)
2956 vdev_scan_stat_init(vd->vdev_child[c]);
2958 mutex_enter(&vd->vdev_stat_lock);
2959 vs->vs_scan_processed = 0;
2960 mutex_exit(&vd->vdev_stat_lock);
2964 vdev_stat_update(zio_t *zio, uint64_t psize)
2966 spa_t *spa = zio->io_spa;
2967 vdev_t *rvd = spa->spa_root_vdev;
2968 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2970 uint64_t txg = zio->io_txg;
2971 vdev_stat_t *vs = &vd->vdev_stat;
2972 zio_type_t type = zio->io_type;
2973 int flags = zio->io_flags;
2976 * If this i/o is a gang leader, it didn't do any actual work.
2978 if (zio->io_gang_tree)
2981 if (zio->io_error == 0) {
2983 * If this is a root i/o, don't count it -- we've already
2984 * counted the top-level vdevs, and vdev_get_stats() will
2985 * aggregate them when asked. This reduces contention on
2986 * the root vdev_stat_lock and implicitly handles blocks
2987 * that compress away to holes, for which there is no i/o.
2988 * (Holes never create vdev children, so all the counters
2989 * remain zero, which is what we want.)
2991 * Note: this only applies to successful i/o (io_error == 0)
2992 * because unlike i/o counts, errors are not additive.
2993 * When reading a ditto block, for example, failure of
2994 * one top-level vdev does not imply a root-level error.
2999 ASSERT(vd == zio->io_vd);
3001 if (flags & ZIO_FLAG_IO_BYPASS)
3004 mutex_enter(&vd->vdev_stat_lock);
3006 if (flags & ZIO_FLAG_IO_REPAIR) {
3007 if (flags & ZIO_FLAG_SCAN_THREAD) {
3008 dsl_scan_phys_t *scn_phys =
3009 &spa->spa_dsl_pool->dp_scan->scn_phys;
3010 uint64_t *processed = &scn_phys->scn_processed;
3013 if (vd->vdev_ops->vdev_op_leaf)
3014 atomic_add_64(processed, psize);
3015 vs->vs_scan_processed += psize;
3018 if (flags & ZIO_FLAG_SELF_HEAL)
3019 vs->vs_self_healed += psize;
3023 vs->vs_bytes[type] += psize;
3025 mutex_exit(&vd->vdev_stat_lock);
3029 if (flags & ZIO_FLAG_SPECULATIVE)
3033 * If this is an I/O error that is going to be retried, then ignore the
3034 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3035 * hard errors, when in reality they can happen for any number of
3036 * innocuous reasons (bus resets, MPxIO link failure, etc).
3038 if (zio->io_error == EIO &&
3039 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3043 * Intent logs writes won't propagate their error to the root
3044 * I/O so don't mark these types of failures as pool-level
3047 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3050 mutex_enter(&vd->vdev_stat_lock);
3051 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3052 if (zio->io_error == ECKSUM)
3053 vs->vs_checksum_errors++;
3055 vs->vs_read_errors++;
3057 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3058 vs->vs_write_errors++;
3059 mutex_exit(&vd->vdev_stat_lock);
3061 if (type == ZIO_TYPE_WRITE && txg != 0 &&
3062 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3063 (flags & ZIO_FLAG_SCAN_THREAD) ||
3064 spa->spa_claiming)) {
3066 * This is either a normal write (not a repair), or it's
3067 * a repair induced by the scrub thread, or it's a repair
3068 * made by zil_claim() during spa_load() in the first txg.
3069 * In the normal case, we commit the DTL change in the same
3070 * txg as the block was born. In the scrub-induced repair
3071 * case, we know that scrubs run in first-pass syncing context,
3072 * so we commit the DTL change in spa_syncing_txg(spa).
3073 * In the zil_claim() case, we commit in spa_first_txg(spa).
3075 * We currently do not make DTL entries for failed spontaneous
3076 * self-healing writes triggered by normal (non-scrubbing)
3077 * reads, because we have no transactional context in which to
3078 * do so -- and it's not clear that it'd be desirable anyway.
3080 if (vd->vdev_ops->vdev_op_leaf) {
3081 uint64_t commit_txg = txg;
3082 if (flags & ZIO_FLAG_SCAN_THREAD) {
3083 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3084 ASSERT(spa_sync_pass(spa) == 1);
3085 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3086 commit_txg = spa_syncing_txg(spa);
3087 } else if (spa->spa_claiming) {
3088 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3089 commit_txg = spa_first_txg(spa);
3091 ASSERT(commit_txg >= spa_syncing_txg(spa));
3092 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3094 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3095 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3096 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3099 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3104 * Update the in-core space usage stats for this vdev, its metaslab class,
3105 * and the root vdev.
3108 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3109 int64_t space_delta)
3111 int64_t dspace_delta = space_delta;
3112 spa_t *spa = vd->vdev_spa;
3113 vdev_t *rvd = spa->spa_root_vdev;
3114 metaslab_group_t *mg = vd->vdev_mg;
3115 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3117 ASSERT(vd == vd->vdev_top);
3120 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3121 * factor. We must calculate this here and not at the root vdev
3122 * because the root vdev's psize-to-asize is simply the max of its
3123 * childrens', thus not accurate enough for us.
3125 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3126 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3127 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3128 vd->vdev_deflate_ratio;
3130 mutex_enter(&vd->vdev_stat_lock);
3131 vd->vdev_stat.vs_alloc += alloc_delta;
3132 vd->vdev_stat.vs_space += space_delta;
3133 vd->vdev_stat.vs_dspace += dspace_delta;
3134 mutex_exit(&vd->vdev_stat_lock);
3136 if (mc == spa_normal_class(spa)) {
3137 mutex_enter(&rvd->vdev_stat_lock);
3138 rvd->vdev_stat.vs_alloc += alloc_delta;
3139 rvd->vdev_stat.vs_space += space_delta;
3140 rvd->vdev_stat.vs_dspace += dspace_delta;
3141 mutex_exit(&rvd->vdev_stat_lock);
3145 ASSERT(rvd == vd->vdev_parent);
3146 ASSERT(vd->vdev_ms_count != 0);
3148 metaslab_class_space_update(mc,
3149 alloc_delta, defer_delta, space_delta, dspace_delta);
3154 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3155 * so that it will be written out next time the vdev configuration is synced.
3156 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3159 vdev_config_dirty(vdev_t *vd)
3161 spa_t *spa = vd->vdev_spa;
3162 vdev_t *rvd = spa->spa_root_vdev;
3165 ASSERT(spa_writeable(spa));
3168 * If this is an aux vdev (as with l2cache and spare devices), then we
3169 * update the vdev config manually and set the sync flag.
3171 if (vd->vdev_aux != NULL) {
3172 spa_aux_vdev_t *sav = vd->vdev_aux;
3176 for (c = 0; c < sav->sav_count; c++) {
3177 if (sav->sav_vdevs[c] == vd)
3181 if (c == sav->sav_count) {
3183 * We're being removed. There's nothing more to do.
3185 ASSERT(sav->sav_sync == B_TRUE);
3189 sav->sav_sync = B_TRUE;
3191 if (nvlist_lookup_nvlist_array(sav->sav_config,
3192 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3193 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3194 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3200 * Setting the nvlist in the middle if the array is a little
3201 * sketchy, but it will work.
3203 nvlist_free(aux[c]);
3204 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3210 * The dirty list is protected by the SCL_CONFIG lock. The caller
3211 * must either hold SCL_CONFIG as writer, or must be the sync thread
3212 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3213 * so this is sufficient to ensure mutual exclusion.
3215 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3216 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3217 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3220 for (c = 0; c < rvd->vdev_children; c++)
3221 vdev_config_dirty(rvd->vdev_child[c]);
3223 ASSERT(vd == vd->vdev_top);
3225 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3227 list_insert_head(&spa->spa_config_dirty_list, vd);
3232 vdev_config_clean(vdev_t *vd)
3234 spa_t *spa = vd->vdev_spa;
3236 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3237 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3238 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3240 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3241 list_remove(&spa->spa_config_dirty_list, vd);
3245 * Mark a top-level vdev's state as dirty, so that the next pass of
3246 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3247 * the state changes from larger config changes because they require
3248 * much less locking, and are often needed for administrative actions.
3251 vdev_state_dirty(vdev_t *vd)
3253 spa_t *spa = vd->vdev_spa;
3255 ASSERT(spa_writeable(spa));
3256 ASSERT(vd == vd->vdev_top);
3259 * The state list is protected by the SCL_STATE lock. The caller
3260 * must either hold SCL_STATE as writer, or must be the sync thread
3261 * (which holds SCL_STATE as reader). There's only one sync thread,
3262 * so this is sufficient to ensure mutual exclusion.
3264 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3265 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3266 spa_config_held(spa, SCL_STATE, RW_READER)));
3268 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3269 list_insert_head(&spa->spa_state_dirty_list, vd);
3273 vdev_state_clean(vdev_t *vd)
3275 spa_t *spa = vd->vdev_spa;
3277 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3278 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3279 spa_config_held(spa, SCL_STATE, RW_READER)));
3281 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3282 list_remove(&spa->spa_state_dirty_list, vd);
3286 * Propagate vdev state up from children to parent.
3289 vdev_propagate_state(vdev_t *vd)
3291 spa_t *spa = vd->vdev_spa;
3292 vdev_t *rvd = spa->spa_root_vdev;
3293 int degraded = 0, faulted = 0;
3297 if (vd->vdev_children > 0) {
3298 for (int c = 0; c < vd->vdev_children; c++) {
3299 child = vd->vdev_child[c];
3302 * Don't factor holes into the decision.
3304 if (child->vdev_ishole)
3307 if (!vdev_readable(child) ||
3308 (!vdev_writeable(child) && spa_writeable(spa))) {
3310 * Root special: if there is a top-level log
3311 * device, treat the root vdev as if it were
3314 if (child->vdev_islog && vd == rvd)
3318 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3322 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3326 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3329 * Root special: if there is a top-level vdev that cannot be
3330 * opened due to corrupted metadata, then propagate the root
3331 * vdev's aux state as 'corrupt' rather than 'insufficient
3334 if (corrupted && vd == rvd &&
3335 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3336 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3337 VDEV_AUX_CORRUPT_DATA);
3340 if (vd->vdev_parent)
3341 vdev_propagate_state(vd->vdev_parent);
3345 * Set a vdev's state. If this is during an open, we don't update the parent
3346 * state, because we're in the process of opening children depth-first.
3347 * Otherwise, we propagate the change to the parent.
3349 * If this routine places a device in a faulted state, an appropriate ereport is
3353 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3355 uint64_t save_state;
3356 spa_t *spa = vd->vdev_spa;
3358 if (state == vd->vdev_state) {
3359 vd->vdev_stat.vs_aux = aux;
3363 save_state = vd->vdev_state;
3365 vd->vdev_state = state;
3366 vd->vdev_stat.vs_aux = aux;
3369 * If we are setting the vdev state to anything but an open state, then
3370 * always close the underlying device unless the device has requested
3371 * a delayed close (i.e. we're about to remove or fault the device).
3372 * Otherwise, we keep accessible but invalid devices open forever.
3373 * We don't call vdev_close() itself, because that implies some extra
3374 * checks (offline, etc) that we don't want here. This is limited to
3375 * leaf devices, because otherwise closing the device will affect other
3378 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3379 vd->vdev_ops->vdev_op_leaf)
3380 vd->vdev_ops->vdev_op_close(vd);
3382 if (vd->vdev_removed &&
3383 state == VDEV_STATE_CANT_OPEN &&
3384 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3386 * If the previous state is set to VDEV_STATE_REMOVED, then this
3387 * device was previously marked removed and someone attempted to
3388 * reopen it. If this failed due to a nonexistent device, then
3389 * keep the device in the REMOVED state. We also let this be if
3390 * it is one of our special test online cases, which is only
3391 * attempting to online the device and shouldn't generate an FMA
3394 vd->vdev_state = VDEV_STATE_REMOVED;
3395 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3396 } else if (state == VDEV_STATE_REMOVED) {
3397 vd->vdev_removed = B_TRUE;
3398 } else if (state == VDEV_STATE_CANT_OPEN) {
3400 * If we fail to open a vdev during an import or recovery, we
3401 * mark it as "not available", which signifies that it was
3402 * never there to begin with. Failure to open such a device
3403 * is not considered an error.
3405 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3406 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3407 vd->vdev_ops->vdev_op_leaf)
3408 vd->vdev_not_present = 1;
3411 * Post the appropriate ereport. If the 'prevstate' field is
3412 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3413 * that this is part of a vdev_reopen(). In this case, we don't
3414 * want to post the ereport if the device was already in the
3415 * CANT_OPEN state beforehand.
3417 * If the 'checkremove' flag is set, then this is an attempt to
3418 * online the device in response to an insertion event. If we
3419 * hit this case, then we have detected an insertion event for a
3420 * faulted or offline device that wasn't in the removed state.
3421 * In this scenario, we don't post an ereport because we are
3422 * about to replace the device, or attempt an online with
3423 * vdev_forcefault, which will generate the fault for us.
3425 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3426 !vd->vdev_not_present && !vd->vdev_checkremove &&
3427 vd != spa->spa_root_vdev) {
3431 case VDEV_AUX_OPEN_FAILED:
3432 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3434 case VDEV_AUX_CORRUPT_DATA:
3435 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3437 case VDEV_AUX_NO_REPLICAS:
3438 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3440 case VDEV_AUX_BAD_GUID_SUM:
3441 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3443 case VDEV_AUX_TOO_SMALL:
3444 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3446 case VDEV_AUX_BAD_LABEL:
3447 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3450 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3453 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3456 /* Erase any notion of persistent removed state */
3457 vd->vdev_removed = B_FALSE;
3459 vd->vdev_removed = B_FALSE;
3463 * Notify the fmd of the state change. Be verbose and post
3464 * notifications even for stuff that's not important; the fmd agent can
3465 * sort it out. Don't emit state change events for non-leaf vdevs since
3466 * they can't change state on their own. The FMD can check their state
3467 * if it wants to when it sees that a leaf vdev had a state change.
3469 if (vd->vdev_ops->vdev_op_leaf)
3470 zfs_post_state_change(spa, vd);
3472 if (!isopen && vd->vdev_parent)
3473 vdev_propagate_state(vd->vdev_parent);
3477 * Check the vdev configuration to ensure that it's capable of supporting
3478 * a root pool. We do not support partial configuration.
3479 * In addition, only a single top-level vdev is allowed.
3481 * FreeBSD does not have above limitations.
3484 vdev_is_bootable(vdev_t *vd)
3487 if (!vd->vdev_ops->vdev_op_leaf) {
3488 char *vdev_type = vd->vdev_ops->vdev_op_type;
3490 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3491 vd->vdev_children > 1) {
3493 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3498 for (int c = 0; c < vd->vdev_children; c++) {
3499 if (!vdev_is_bootable(vd->vdev_child[c]))
3502 #endif /* illumos */
3507 * Load the state from the original vdev tree (ovd) which
3508 * we've retrieved from the MOS config object. If the original
3509 * vdev was offline or faulted then we transfer that state to the
3510 * device in the current vdev tree (nvd).
3513 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3515 spa_t *spa = nvd->vdev_spa;
3517 ASSERT(nvd->vdev_top->vdev_islog);
3518 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3519 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3521 for (int c = 0; c < nvd->vdev_children; c++)
3522 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3524 if (nvd->vdev_ops->vdev_op_leaf) {
3526 * Restore the persistent vdev state
3528 nvd->vdev_offline = ovd->vdev_offline;
3529 nvd->vdev_faulted = ovd->vdev_faulted;
3530 nvd->vdev_degraded = ovd->vdev_degraded;
3531 nvd->vdev_removed = ovd->vdev_removed;
3536 * Determine if a log device has valid content. If the vdev was
3537 * removed or faulted in the MOS config then we know that
3538 * the content on the log device has already been written to the pool.
3541 vdev_log_state_valid(vdev_t *vd)
3543 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3547 for (int c = 0; c < vd->vdev_children; c++)
3548 if (vdev_log_state_valid(vd->vdev_child[c]))
3555 * Expand a vdev if possible.
3558 vdev_expand(vdev_t *vd, uint64_t txg)
3560 ASSERT(vd->vdev_top == vd);
3561 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3563 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3564 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3565 vdev_config_dirty(vd);
3573 vdev_split(vdev_t *vd)
3575 vdev_t *cvd, *pvd = vd->vdev_parent;
3577 vdev_remove_child(pvd, vd);
3578 vdev_compact_children(pvd);
3580 cvd = pvd->vdev_child[0];
3581 if (pvd->vdev_children == 1) {
3582 vdev_remove_parent(cvd);
3583 cvd->vdev_splitting = B_TRUE;
3585 vdev_propagate_state(cvd);
3589 vdev_deadman(vdev_t *vd)
3591 for (int c = 0; c < vd->vdev_children; c++) {
3592 vdev_t *cvd = vd->vdev_child[c];
3597 if (vd->vdev_ops->vdev_op_leaf) {
3598 vdev_queue_t *vq = &vd->vdev_queue;
3600 mutex_enter(&vq->vq_lock);
3601 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3602 spa_t *spa = vd->vdev_spa;
3607 * Look at the head of all the pending queues,
3608 * if any I/O has been outstanding for longer than
3609 * the spa_deadman_synctime we panic the system.
3611 fio = avl_first(&vq->vq_active_tree);
3612 delta = gethrtime() - fio->io_timestamp;
3613 if (delta > spa_deadman_synctime(spa)) {
3614 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3615 "delta %lluns, last io %lluns",
3616 fio->io_timestamp, delta,
3617 vq->vq_io_complete_ts);
3618 fm_panic("I/O to pool '%s' appears to be "
3619 "hung on vdev guid %llu at '%s'.",
3621 (long long unsigned int) vd->vdev_guid,
3625 mutex_exit(&vq->vq_lock);