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, 2021 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Toomas Soome <tsoome@me.com>
28 * Copyright 2017 Joyent, Inc.
29 * Copyright (c) 2017, Intel Corporation.
30 * Copyright (c) 2019, Datto Inc. All rights reserved.
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
37 #include <sys/bpobj.h>
39 #include <sys/dmu_tx.h>
40 #include <sys/dsl_dir.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/vdev_rebuild.h>
43 #include <sys/vdev_draid.h>
44 #include <sys/uberblock_impl.h>
45 #include <sys/metaslab.h>
46 #include <sys/metaslab_impl.h>
47 #include <sys/space_map.h>
48 #include <sys/space_reftree.h>
51 #include <sys/fs/zfs.h>
54 #include <sys/dsl_scan.h>
55 #include <sys/vdev_raidz.h>
57 #include <sys/vdev_initialize.h>
58 #include <sys/vdev_trim.h>
60 #include <sys/zfs_ratelimit.h>
63 * One metaslab from each (normal-class) vdev is used by the ZIL. These are
64 * called "embedded slog metaslabs", are referenced by vdev_log_mg, and are
65 * part of the spa_embedded_log_class. The metaslab with the most free space
66 * in each vdev is selected for this purpose when the pool is opened (or a
67 * vdev is added). See vdev_metaslab_init().
69 * Log blocks can be allocated from the following locations. Each one is tried
70 * in order until the allocation succeeds:
71 * 1. dedicated log vdevs, aka "slog" (spa_log_class)
72 * 2. embedded slog metaslabs (spa_embedded_log_class)
73 * 3. other metaslabs in normal vdevs (spa_normal_class)
75 * zfs_embedded_slog_min_ms disables the embedded slog if there are fewer
76 * than this number of metaslabs in the vdev. This ensures that we don't set
77 * aside an unreasonable amount of space for the ZIL. If set to less than
78 * 1 << (spa_slop_shift + 1), on small pools the usable space may be reduced
79 * (by more than 1<<spa_slop_shift) due to the embedded slog metaslab.
81 int zfs_embedded_slog_min_ms = 64;
83 /* default target for number of metaslabs per top-level vdev */
84 int zfs_vdev_default_ms_count = 200;
86 /* minimum number of metaslabs per top-level vdev */
87 int zfs_vdev_min_ms_count = 16;
89 /* practical upper limit of total metaslabs per top-level vdev */
90 int zfs_vdev_ms_count_limit = 1ULL << 17;
92 /* lower limit for metaslab size (512M) */
93 int zfs_vdev_default_ms_shift = 29;
95 /* upper limit for metaslab size (16G) */
96 int zfs_vdev_max_ms_shift = 34;
98 int vdev_validate_skip = B_FALSE;
101 * Since the DTL space map of a vdev is not expected to have a lot of
102 * entries, we default its block size to 4K.
104 int zfs_vdev_dtl_sm_blksz = (1 << 12);
107 * Rate limit slow IO (delay) events to this many per second.
109 unsigned int zfs_slow_io_events_per_second = 20;
112 * Rate limit checksum events after this many checksum errors per second.
114 unsigned int zfs_checksum_events_per_second = 20;
117 * Ignore errors during scrub/resilver. Allows to work around resilver
118 * upon import when there are pool errors.
120 int zfs_scan_ignore_errors = 0;
123 * vdev-wide space maps that have lots of entries written to them at
124 * the end of each transaction can benefit from a higher I/O bandwidth
125 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
127 int zfs_vdev_standard_sm_blksz = (1 << 17);
130 * Tunable parameter for debugging or performance analysis. Setting this
131 * will cause pool corruption on power loss if a volatile out-of-order
132 * write cache is enabled.
134 int zfs_nocacheflush = 0;
136 uint64_t zfs_vdev_max_auto_ashift = ASHIFT_MAX;
137 uint64_t zfs_vdev_min_auto_ashift = ASHIFT_MIN;
141 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
147 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
150 if (vd->vdev_path != NULL) {
151 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
154 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
155 vd->vdev_ops->vdev_op_type,
156 (u_longlong_t)vd->vdev_id,
157 (u_longlong_t)vd->vdev_guid, buf);
162 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
166 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
167 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
168 vd->vdev_ops->vdev_op_type);
172 switch (vd->vdev_state) {
173 case VDEV_STATE_UNKNOWN:
174 (void) snprintf(state, sizeof (state), "unknown");
176 case VDEV_STATE_CLOSED:
177 (void) snprintf(state, sizeof (state), "closed");
179 case VDEV_STATE_OFFLINE:
180 (void) snprintf(state, sizeof (state), "offline");
182 case VDEV_STATE_REMOVED:
183 (void) snprintf(state, sizeof (state), "removed");
185 case VDEV_STATE_CANT_OPEN:
186 (void) snprintf(state, sizeof (state), "can't open");
188 case VDEV_STATE_FAULTED:
189 (void) snprintf(state, sizeof (state), "faulted");
191 case VDEV_STATE_DEGRADED:
192 (void) snprintf(state, sizeof (state), "degraded");
194 case VDEV_STATE_HEALTHY:
195 (void) snprintf(state, sizeof (state), "healthy");
198 (void) snprintf(state, sizeof (state), "<state %u>",
199 (uint_t)vd->vdev_state);
202 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
203 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
204 vd->vdev_islog ? " (log)" : "",
205 (u_longlong_t)vd->vdev_guid,
206 vd->vdev_path ? vd->vdev_path : "N/A", state);
208 for (uint64_t i = 0; i < vd->vdev_children; i++)
209 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
213 * Virtual device management.
216 static vdev_ops_t *vdev_ops_table[] = {
220 &vdev_draid_spare_ops,
233 * Given a vdev type, return the appropriate ops vector.
236 vdev_getops(const char *type)
238 vdev_ops_t *ops, **opspp;
240 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
241 if (strcmp(ops->vdev_op_type, type) == 0)
248 * Given a vdev and a metaslab class, find which metaslab group we're
249 * interested in. All vdevs may belong to two different metaslab classes.
250 * Dedicated slog devices use only the primary metaslab group, rather than a
251 * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL.
254 vdev_get_mg(vdev_t *vd, metaslab_class_t *mc)
256 if (mc == spa_embedded_log_class(vd->vdev_spa) &&
257 vd->vdev_log_mg != NULL)
258 return (vd->vdev_log_mg);
260 return (vd->vdev_mg);
265 vdev_default_xlate(vdev_t *vd, const range_seg64_t *logical_rs,
266 range_seg64_t *physical_rs, range_seg64_t *remain_rs)
268 physical_rs->rs_start = logical_rs->rs_start;
269 physical_rs->rs_end = logical_rs->rs_end;
273 * Derive the enumerated allocation bias from string input.
274 * String origin is either the per-vdev zap or zpool(8).
276 static vdev_alloc_bias_t
277 vdev_derive_alloc_bias(const char *bias)
279 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
281 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
282 alloc_bias = VDEV_BIAS_LOG;
283 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
284 alloc_bias = VDEV_BIAS_SPECIAL;
285 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
286 alloc_bias = VDEV_BIAS_DEDUP;
292 * Default asize function: return the MAX of psize with the asize of
293 * all children. This is what's used by anything other than RAID-Z.
296 vdev_default_asize(vdev_t *vd, uint64_t psize)
298 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
301 for (int c = 0; c < vd->vdev_children; c++) {
302 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
303 asize = MAX(asize, csize);
310 vdev_default_min_asize(vdev_t *vd)
312 return (vd->vdev_min_asize);
316 * Get the minimum allocatable size. We define the allocatable size as
317 * the vdev's asize rounded to the nearest metaslab. This allows us to
318 * replace or attach devices which don't have the same physical size but
319 * can still satisfy the same number of allocations.
322 vdev_get_min_asize(vdev_t *vd)
324 vdev_t *pvd = vd->vdev_parent;
327 * If our parent is NULL (inactive spare or cache) or is the root,
328 * just return our own asize.
331 return (vd->vdev_asize);
334 * The top-level vdev just returns the allocatable size rounded
335 * to the nearest metaslab.
337 if (vd == vd->vdev_top)
338 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
340 return (pvd->vdev_ops->vdev_op_min_asize(pvd));
344 vdev_set_min_asize(vdev_t *vd)
346 vd->vdev_min_asize = vdev_get_min_asize(vd);
348 for (int c = 0; c < vd->vdev_children; c++)
349 vdev_set_min_asize(vd->vdev_child[c]);
353 * Get the minimal allocation size for the top-level vdev.
356 vdev_get_min_alloc(vdev_t *vd)
358 uint64_t min_alloc = 1ULL << vd->vdev_ashift;
360 if (vd->vdev_ops->vdev_op_min_alloc != NULL)
361 min_alloc = vd->vdev_ops->vdev_op_min_alloc(vd);
367 * Get the parity level for a top-level vdev.
370 vdev_get_nparity(vdev_t *vd)
372 uint64_t nparity = 0;
374 if (vd->vdev_ops->vdev_op_nparity != NULL)
375 nparity = vd->vdev_ops->vdev_op_nparity(vd);
381 * Get the number of data disks for a top-level vdev.
384 vdev_get_ndisks(vdev_t *vd)
388 if (vd->vdev_ops->vdev_op_ndisks != NULL)
389 ndisks = vd->vdev_ops->vdev_op_ndisks(vd);
395 vdev_lookup_top(spa_t *spa, uint64_t vdev)
397 vdev_t *rvd = spa->spa_root_vdev;
399 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
401 if (vdev < rvd->vdev_children) {
402 ASSERT(rvd->vdev_child[vdev] != NULL);
403 return (rvd->vdev_child[vdev]);
410 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
414 if (vd->vdev_guid == guid)
417 for (int c = 0; c < vd->vdev_children; c++)
418 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
426 vdev_count_leaves_impl(vdev_t *vd)
430 if (vd->vdev_ops->vdev_op_leaf)
433 for (int c = 0; c < vd->vdev_children; c++)
434 n += vdev_count_leaves_impl(vd->vdev_child[c]);
440 vdev_count_leaves(spa_t *spa)
444 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
445 rc = vdev_count_leaves_impl(spa->spa_root_vdev);
446 spa_config_exit(spa, SCL_VDEV, FTAG);
452 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
454 size_t oldsize, newsize;
455 uint64_t id = cvd->vdev_id;
458 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
459 ASSERT(cvd->vdev_parent == NULL);
461 cvd->vdev_parent = pvd;
466 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
468 oldsize = pvd->vdev_children * sizeof (vdev_t *);
469 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
470 newsize = pvd->vdev_children * sizeof (vdev_t *);
472 newchild = kmem_alloc(newsize, KM_SLEEP);
473 if (pvd->vdev_child != NULL) {
474 bcopy(pvd->vdev_child, newchild, oldsize);
475 kmem_free(pvd->vdev_child, oldsize);
478 pvd->vdev_child = newchild;
479 pvd->vdev_child[id] = cvd;
481 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
482 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
485 * Walk up all ancestors to update guid sum.
487 for (; pvd != NULL; pvd = pvd->vdev_parent)
488 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
490 if (cvd->vdev_ops->vdev_op_leaf) {
491 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd);
492 cvd->vdev_spa->spa_leaf_list_gen++;
497 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
500 uint_t id = cvd->vdev_id;
502 ASSERT(cvd->vdev_parent == pvd);
507 ASSERT(id < pvd->vdev_children);
508 ASSERT(pvd->vdev_child[id] == cvd);
510 pvd->vdev_child[id] = NULL;
511 cvd->vdev_parent = NULL;
513 for (c = 0; c < pvd->vdev_children; c++)
514 if (pvd->vdev_child[c])
517 if (c == pvd->vdev_children) {
518 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
519 pvd->vdev_child = NULL;
520 pvd->vdev_children = 0;
523 if (cvd->vdev_ops->vdev_op_leaf) {
524 spa_t *spa = cvd->vdev_spa;
525 list_remove(&spa->spa_leaf_list, cvd);
526 spa->spa_leaf_list_gen++;
530 * Walk up all ancestors to update guid sum.
532 for (; pvd != NULL; pvd = pvd->vdev_parent)
533 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
537 * Remove any holes in the child array.
540 vdev_compact_children(vdev_t *pvd)
542 vdev_t **newchild, *cvd;
543 int oldc = pvd->vdev_children;
546 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
551 for (int c = newc = 0; c < oldc; c++)
552 if (pvd->vdev_child[c])
556 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP);
558 for (int c = newc = 0; c < oldc; c++) {
559 if ((cvd = pvd->vdev_child[c]) != NULL) {
560 newchild[newc] = cvd;
561 cvd->vdev_id = newc++;
568 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
569 pvd->vdev_child = newchild;
570 pvd->vdev_children = newc;
574 * Allocate and minimally initialize a vdev_t.
577 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
580 vdev_indirect_config_t *vic;
582 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
583 vic = &vd->vdev_indirect_config;
585 if (spa->spa_root_vdev == NULL) {
586 ASSERT(ops == &vdev_root_ops);
587 spa->spa_root_vdev = vd;
588 spa->spa_load_guid = spa_generate_guid(NULL);
591 if (guid == 0 && ops != &vdev_hole_ops) {
592 if (spa->spa_root_vdev == vd) {
594 * The root vdev's guid will also be the pool guid,
595 * which must be unique among all pools.
597 guid = spa_generate_guid(NULL);
600 * Any other vdev's guid must be unique within the pool.
602 guid = spa_generate_guid(spa);
604 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
609 vd->vdev_guid = guid;
610 vd->vdev_guid_sum = guid;
612 vd->vdev_state = VDEV_STATE_CLOSED;
613 vd->vdev_ishole = (ops == &vdev_hole_ops);
614 vic->vic_prev_indirect_vdev = UINT64_MAX;
616 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
617 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
618 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL,
622 * Initialize rate limit structs for events. We rate limit ZIO delay
623 * and checksum events so that we don't overwhelm ZED with thousands
624 * of events when a disk is acting up.
626 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second,
628 zfs_ratelimit_init(&vd->vdev_checksum_rl,
629 &zfs_checksum_events_per_second, 1);
631 list_link_init(&vd->vdev_config_dirty_node);
632 list_link_init(&vd->vdev_state_dirty_node);
633 list_link_init(&vd->vdev_initialize_node);
634 list_link_init(&vd->vdev_leaf_node);
635 list_link_init(&vd->vdev_trim_node);
637 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL);
638 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
639 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
640 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
642 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
643 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
644 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
645 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
647 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL);
648 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL);
649 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL);
650 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL);
651 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL);
652 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL);
654 mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL);
655 cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL);
657 for (int t = 0; t < DTL_TYPES; t++) {
658 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
662 txg_list_create(&vd->vdev_ms_list, spa,
663 offsetof(struct metaslab, ms_txg_node));
664 txg_list_create(&vd->vdev_dtl_list, spa,
665 offsetof(struct vdev, vdev_dtl_node));
666 vd->vdev_stat.vs_timestamp = gethrtime();
674 * Allocate a new vdev. The 'alloctype' is used to control whether we are
675 * creating a new vdev or loading an existing one - the behavior is slightly
676 * different for each case.
679 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
684 uint64_t guid = 0, islog;
686 vdev_indirect_config_t *vic;
689 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
690 boolean_t top_level = (parent && !parent->vdev_parent);
692 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
694 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
695 return (SET_ERROR(EINVAL));
697 if ((ops = vdev_getops(type)) == NULL)
698 return (SET_ERROR(EINVAL));
701 * If this is a load, get the vdev guid from the nvlist.
702 * Otherwise, vdev_alloc_common() will generate one for us.
704 if (alloctype == VDEV_ALLOC_LOAD) {
707 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
709 return (SET_ERROR(EINVAL));
711 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
712 return (SET_ERROR(EINVAL));
713 } else if (alloctype == VDEV_ALLOC_SPARE) {
714 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
715 return (SET_ERROR(EINVAL));
716 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
717 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
718 return (SET_ERROR(EINVAL));
719 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
720 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
721 return (SET_ERROR(EINVAL));
725 * The first allocated vdev must be of type 'root'.
727 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
728 return (SET_ERROR(EINVAL));
731 * Determine whether we're a log vdev.
734 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
735 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
736 return (SET_ERROR(ENOTSUP));
738 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
739 return (SET_ERROR(ENOTSUP));
741 if (top_level && alloctype == VDEV_ALLOC_ADD) {
745 * If creating a top-level vdev, check for allocation
748 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
750 alloc_bias = vdev_derive_alloc_bias(bias);
752 /* spa_vdev_add() expects feature to be enabled */
753 if (spa->spa_load_state != SPA_LOAD_CREATE &&
754 !spa_feature_is_enabled(spa,
755 SPA_FEATURE_ALLOCATION_CLASSES)) {
756 return (SET_ERROR(ENOTSUP));
760 /* spa_vdev_add() expects feature to be enabled */
761 if (ops == &vdev_draid_ops &&
762 spa->spa_load_state != SPA_LOAD_CREATE &&
763 !spa_feature_is_enabled(spa, SPA_FEATURE_DRAID)) {
764 return (SET_ERROR(ENOTSUP));
769 * Initialize the vdev specific data. This is done before calling
770 * vdev_alloc_common() since it may fail and this simplifies the
771 * error reporting and cleanup code paths.
774 if (ops->vdev_op_init != NULL) {
775 rc = ops->vdev_op_init(spa, nv, &tsd);
781 vd = vdev_alloc_common(spa, id, guid, ops);
783 vd->vdev_islog = islog;
785 if (top_level && alloc_bias != VDEV_BIAS_NONE)
786 vd->vdev_alloc_bias = alloc_bias;
788 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
789 vd->vdev_path = spa_strdup(vd->vdev_path);
792 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
793 * fault on a vdev and want it to persist across imports (like with
796 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp);
797 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) {
798 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
799 vd->vdev_faulted = 1;
800 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
803 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
804 vd->vdev_devid = spa_strdup(vd->vdev_devid);
805 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
806 &vd->vdev_physpath) == 0)
807 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
809 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
810 &vd->vdev_enc_sysfs_path) == 0)
811 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path);
813 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
814 vd->vdev_fru = spa_strdup(vd->vdev_fru);
817 * Set the whole_disk property. If it's not specified, leave the value
820 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
821 &vd->vdev_wholedisk) != 0)
822 vd->vdev_wholedisk = -1ULL;
824 vic = &vd->vdev_indirect_config;
826 ASSERT0(vic->vic_mapping_object);
827 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
828 &vic->vic_mapping_object);
829 ASSERT0(vic->vic_births_object);
830 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
831 &vic->vic_births_object);
832 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
833 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
834 &vic->vic_prev_indirect_vdev);
837 * Look for the 'not present' flag. This will only be set if the device
838 * was not present at the time of import.
840 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
841 &vd->vdev_not_present);
844 * Get the alignment requirement.
846 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
849 * Retrieve the vdev creation time.
851 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
855 * If we're a top-level vdev, try to load the allocation parameters.
858 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
859 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
861 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
863 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
865 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
867 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
870 ASSERT0(vd->vdev_top_zap);
873 if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
874 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
875 alloctype == VDEV_ALLOC_ADD ||
876 alloctype == VDEV_ALLOC_SPLIT ||
877 alloctype == VDEV_ALLOC_ROOTPOOL);
878 /* Note: metaslab_group_create() is now deferred */
881 if (vd->vdev_ops->vdev_op_leaf &&
882 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
883 (void) nvlist_lookup_uint64(nv,
884 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
886 ASSERT0(vd->vdev_leaf_zap);
890 * If we're a leaf vdev, try to load the DTL object and other state.
893 if (vd->vdev_ops->vdev_op_leaf &&
894 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
895 alloctype == VDEV_ALLOC_ROOTPOOL)) {
896 if (alloctype == VDEV_ALLOC_LOAD) {
897 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
898 &vd->vdev_dtl_object);
899 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
903 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
906 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
907 &spare) == 0 && spare)
911 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
914 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
915 &vd->vdev_resilver_txg);
917 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
918 &vd->vdev_rebuild_txg);
920 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER))
921 vdev_defer_resilver(vd);
924 * In general, when importing a pool we want to ignore the
925 * persistent fault state, as the diagnosis made on another
926 * system may not be valid in the current context. The only
927 * exception is if we forced a vdev to a persistently faulted
928 * state with 'zpool offline -f'. The persistent fault will
929 * remain across imports until cleared.
931 * Local vdevs will remain in the faulted state.
933 if (spa_load_state(spa) == SPA_LOAD_OPEN ||
934 spa_load_state(spa) == SPA_LOAD_IMPORT) {
935 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
937 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
939 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
942 if (vd->vdev_faulted || vd->vdev_degraded) {
946 VDEV_AUX_ERR_EXCEEDED;
947 if (nvlist_lookup_string(nv,
948 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
949 strcmp(aux, "external") == 0)
950 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
952 vd->vdev_faulted = 0ULL;
958 * Add ourselves to the parent's list of children.
960 vdev_add_child(parent, vd);
968 vdev_free(vdev_t *vd)
970 spa_t *spa = vd->vdev_spa;
972 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
973 ASSERT3P(vd->vdev_trim_thread, ==, NULL);
974 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL);
975 ASSERT3P(vd->vdev_rebuild_thread, ==, NULL);
978 * Scan queues are normally destroyed at the end of a scan. If the
979 * queue exists here, that implies the vdev is being removed while
980 * the scan is still running.
982 if (vd->vdev_scan_io_queue != NULL) {
983 mutex_enter(&vd->vdev_scan_io_queue_lock);
984 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
985 vd->vdev_scan_io_queue = NULL;
986 mutex_exit(&vd->vdev_scan_io_queue_lock);
990 * vdev_free() implies closing the vdev first. This is simpler than
991 * trying to ensure complicated semantics for all callers.
995 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
996 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
1001 for (int c = 0; c < vd->vdev_children; c++)
1002 vdev_free(vd->vdev_child[c]);
1004 ASSERT(vd->vdev_child == NULL);
1005 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
1007 if (vd->vdev_ops->vdev_op_fini != NULL)
1008 vd->vdev_ops->vdev_op_fini(vd);
1011 * Discard allocation state.
1013 if (vd->vdev_mg != NULL) {
1014 vdev_metaslab_fini(vd);
1015 metaslab_group_destroy(vd->vdev_mg);
1018 if (vd->vdev_log_mg != NULL) {
1019 ASSERT0(vd->vdev_ms_count);
1020 metaslab_group_destroy(vd->vdev_log_mg);
1021 vd->vdev_log_mg = NULL;
1024 ASSERT0(vd->vdev_stat.vs_space);
1025 ASSERT0(vd->vdev_stat.vs_dspace);
1026 ASSERT0(vd->vdev_stat.vs_alloc);
1029 * Remove this vdev from its parent's child list.
1031 vdev_remove_child(vd->vdev_parent, vd);
1033 ASSERT(vd->vdev_parent == NULL);
1034 ASSERT(!list_link_active(&vd->vdev_leaf_node));
1037 * Clean up vdev structure.
1039 vdev_queue_fini(vd);
1040 vdev_cache_fini(vd);
1043 spa_strfree(vd->vdev_path);
1045 spa_strfree(vd->vdev_devid);
1046 if (vd->vdev_physpath)
1047 spa_strfree(vd->vdev_physpath);
1049 if (vd->vdev_enc_sysfs_path)
1050 spa_strfree(vd->vdev_enc_sysfs_path);
1053 spa_strfree(vd->vdev_fru);
1055 if (vd->vdev_isspare)
1056 spa_spare_remove(vd);
1057 if (vd->vdev_isl2cache)
1058 spa_l2cache_remove(vd);
1060 txg_list_destroy(&vd->vdev_ms_list);
1061 txg_list_destroy(&vd->vdev_dtl_list);
1063 mutex_enter(&vd->vdev_dtl_lock);
1064 space_map_close(vd->vdev_dtl_sm);
1065 for (int t = 0; t < DTL_TYPES; t++) {
1066 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
1067 range_tree_destroy(vd->vdev_dtl[t]);
1069 mutex_exit(&vd->vdev_dtl_lock);
1071 EQUIV(vd->vdev_indirect_births != NULL,
1072 vd->vdev_indirect_mapping != NULL);
1073 if (vd->vdev_indirect_births != NULL) {
1074 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
1075 vdev_indirect_births_close(vd->vdev_indirect_births);
1078 if (vd->vdev_obsolete_sm != NULL) {
1079 ASSERT(vd->vdev_removing ||
1080 vd->vdev_ops == &vdev_indirect_ops);
1081 space_map_close(vd->vdev_obsolete_sm);
1082 vd->vdev_obsolete_sm = NULL;
1084 range_tree_destroy(vd->vdev_obsolete_segments);
1085 rw_destroy(&vd->vdev_indirect_rwlock);
1086 mutex_destroy(&vd->vdev_obsolete_lock);
1088 mutex_destroy(&vd->vdev_dtl_lock);
1089 mutex_destroy(&vd->vdev_stat_lock);
1090 mutex_destroy(&vd->vdev_probe_lock);
1091 mutex_destroy(&vd->vdev_scan_io_queue_lock);
1093 mutex_destroy(&vd->vdev_initialize_lock);
1094 mutex_destroy(&vd->vdev_initialize_io_lock);
1095 cv_destroy(&vd->vdev_initialize_io_cv);
1096 cv_destroy(&vd->vdev_initialize_cv);
1098 mutex_destroy(&vd->vdev_trim_lock);
1099 mutex_destroy(&vd->vdev_autotrim_lock);
1100 mutex_destroy(&vd->vdev_trim_io_lock);
1101 cv_destroy(&vd->vdev_trim_cv);
1102 cv_destroy(&vd->vdev_autotrim_cv);
1103 cv_destroy(&vd->vdev_trim_io_cv);
1105 mutex_destroy(&vd->vdev_rebuild_lock);
1106 cv_destroy(&vd->vdev_rebuild_cv);
1108 zfs_ratelimit_fini(&vd->vdev_delay_rl);
1109 zfs_ratelimit_fini(&vd->vdev_checksum_rl);
1111 if (vd == spa->spa_root_vdev)
1112 spa->spa_root_vdev = NULL;
1114 kmem_free(vd, sizeof (vdev_t));
1118 * Transfer top-level vdev state from svd to tvd.
1121 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
1123 spa_t *spa = svd->vdev_spa;
1128 ASSERT(tvd == tvd->vdev_top);
1130 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite;
1131 tvd->vdev_ms_array = svd->vdev_ms_array;
1132 tvd->vdev_ms_shift = svd->vdev_ms_shift;
1133 tvd->vdev_ms_count = svd->vdev_ms_count;
1134 tvd->vdev_top_zap = svd->vdev_top_zap;
1136 svd->vdev_ms_array = 0;
1137 svd->vdev_ms_shift = 0;
1138 svd->vdev_ms_count = 0;
1139 svd->vdev_top_zap = 0;
1142 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1143 if (tvd->vdev_log_mg)
1144 ASSERT3P(tvd->vdev_log_mg, ==, svd->vdev_log_mg);
1145 tvd->vdev_mg = svd->vdev_mg;
1146 tvd->vdev_log_mg = svd->vdev_log_mg;
1147 tvd->vdev_ms = svd->vdev_ms;
1149 svd->vdev_mg = NULL;
1150 svd->vdev_log_mg = NULL;
1151 svd->vdev_ms = NULL;
1153 if (tvd->vdev_mg != NULL)
1154 tvd->vdev_mg->mg_vd = tvd;
1155 if (tvd->vdev_log_mg != NULL)
1156 tvd->vdev_log_mg->mg_vd = tvd;
1158 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1159 svd->vdev_checkpoint_sm = NULL;
1161 tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
1162 svd->vdev_alloc_bias = VDEV_BIAS_NONE;
1164 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1165 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1166 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1168 svd->vdev_stat.vs_alloc = 0;
1169 svd->vdev_stat.vs_space = 0;
1170 svd->vdev_stat.vs_dspace = 0;
1173 * State which may be set on a top-level vdev that's in the
1174 * process of being removed.
1176 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1177 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1178 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1179 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1180 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1181 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1182 ASSERT0(tvd->vdev_removing);
1183 ASSERT0(tvd->vdev_rebuilding);
1184 tvd->vdev_removing = svd->vdev_removing;
1185 tvd->vdev_rebuilding = svd->vdev_rebuilding;
1186 tvd->vdev_rebuild_config = svd->vdev_rebuild_config;
1187 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1188 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1189 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1190 range_tree_swap(&svd->vdev_obsolete_segments,
1191 &tvd->vdev_obsolete_segments);
1192 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1193 svd->vdev_indirect_config.vic_mapping_object = 0;
1194 svd->vdev_indirect_config.vic_births_object = 0;
1195 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1196 svd->vdev_indirect_mapping = NULL;
1197 svd->vdev_indirect_births = NULL;
1198 svd->vdev_obsolete_sm = NULL;
1199 svd->vdev_removing = 0;
1200 svd->vdev_rebuilding = 0;
1202 for (t = 0; t < TXG_SIZE; t++) {
1203 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1204 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1205 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1206 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1207 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1208 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1211 if (list_link_active(&svd->vdev_config_dirty_node)) {
1212 vdev_config_clean(svd);
1213 vdev_config_dirty(tvd);
1216 if (list_link_active(&svd->vdev_state_dirty_node)) {
1217 vdev_state_clean(svd);
1218 vdev_state_dirty(tvd);
1221 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1222 svd->vdev_deflate_ratio = 0;
1224 tvd->vdev_islog = svd->vdev_islog;
1225 svd->vdev_islog = 0;
1227 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1231 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1238 for (int c = 0; c < vd->vdev_children; c++)
1239 vdev_top_update(tvd, vd->vdev_child[c]);
1243 * Add a mirror/replacing vdev above an existing vdev. There is no need to
1244 * call .vdev_op_init() since mirror/replacing vdevs do not have private state.
1247 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1249 spa_t *spa = cvd->vdev_spa;
1250 vdev_t *pvd = cvd->vdev_parent;
1253 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1255 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1257 mvd->vdev_asize = cvd->vdev_asize;
1258 mvd->vdev_min_asize = cvd->vdev_min_asize;
1259 mvd->vdev_max_asize = cvd->vdev_max_asize;
1260 mvd->vdev_psize = cvd->vdev_psize;
1261 mvd->vdev_ashift = cvd->vdev_ashift;
1262 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1263 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1264 mvd->vdev_state = cvd->vdev_state;
1265 mvd->vdev_crtxg = cvd->vdev_crtxg;
1267 vdev_remove_child(pvd, cvd);
1268 vdev_add_child(pvd, mvd);
1269 cvd->vdev_id = mvd->vdev_children;
1270 vdev_add_child(mvd, cvd);
1271 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1273 if (mvd == mvd->vdev_top)
1274 vdev_top_transfer(cvd, mvd);
1280 * Remove a 1-way mirror/replacing vdev from the tree.
1283 vdev_remove_parent(vdev_t *cvd)
1285 vdev_t *mvd = cvd->vdev_parent;
1286 vdev_t *pvd = mvd->vdev_parent;
1288 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1290 ASSERT(mvd->vdev_children == 1);
1291 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1292 mvd->vdev_ops == &vdev_replacing_ops ||
1293 mvd->vdev_ops == &vdev_spare_ops);
1294 cvd->vdev_ashift = mvd->vdev_ashift;
1295 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1296 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1297 vdev_remove_child(mvd, cvd);
1298 vdev_remove_child(pvd, mvd);
1301 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1302 * Otherwise, we could have detached an offline device, and when we
1303 * go to import the pool we'll think we have two top-level vdevs,
1304 * instead of a different version of the same top-level vdev.
1306 if (mvd->vdev_top == mvd) {
1307 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1308 cvd->vdev_orig_guid = cvd->vdev_guid;
1309 cvd->vdev_guid += guid_delta;
1310 cvd->vdev_guid_sum += guid_delta;
1313 * If pool not set for autoexpand, we need to also preserve
1314 * mvd's asize to prevent automatic expansion of cvd.
1315 * Otherwise if we are adjusting the mirror by attaching and
1316 * detaching children of non-uniform sizes, the mirror could
1317 * autoexpand, unexpectedly requiring larger devices to
1318 * re-establish the mirror.
1320 if (!cvd->vdev_spa->spa_autoexpand)
1321 cvd->vdev_asize = mvd->vdev_asize;
1323 cvd->vdev_id = mvd->vdev_id;
1324 vdev_add_child(pvd, cvd);
1325 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1327 if (cvd == cvd->vdev_top)
1328 vdev_top_transfer(mvd, cvd);
1330 ASSERT(mvd->vdev_children == 0);
1335 vdev_metaslab_group_create(vdev_t *vd)
1337 spa_t *spa = vd->vdev_spa;
1340 * metaslab_group_create was delayed until allocation bias was available
1342 if (vd->vdev_mg == NULL) {
1343 metaslab_class_t *mc;
1345 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
1346 vd->vdev_alloc_bias = VDEV_BIAS_LOG;
1348 ASSERT3U(vd->vdev_islog, ==,
1349 (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
1351 switch (vd->vdev_alloc_bias) {
1353 mc = spa_log_class(spa);
1355 case VDEV_BIAS_SPECIAL:
1356 mc = spa_special_class(spa);
1358 case VDEV_BIAS_DEDUP:
1359 mc = spa_dedup_class(spa);
1362 mc = spa_normal_class(spa);
1365 vd->vdev_mg = metaslab_group_create(mc, vd,
1366 spa->spa_alloc_count);
1368 if (!vd->vdev_islog) {
1369 vd->vdev_log_mg = metaslab_group_create(
1370 spa_embedded_log_class(spa), vd, 1);
1374 * The spa ashift min/max only apply for the normal metaslab
1375 * class. Class destination is late binding so ashift boundry
1376 * setting had to wait until now.
1378 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1379 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
1380 if (vd->vdev_ashift > spa->spa_max_ashift)
1381 spa->spa_max_ashift = vd->vdev_ashift;
1382 if (vd->vdev_ashift < spa->spa_min_ashift)
1383 spa->spa_min_ashift = vd->vdev_ashift;
1385 uint64_t min_alloc = vdev_get_min_alloc(vd);
1386 if (min_alloc < spa->spa_min_alloc)
1387 spa->spa_min_alloc = min_alloc;
1393 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1395 spa_t *spa = vd->vdev_spa;
1396 uint64_t oldc = vd->vdev_ms_count;
1397 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1400 boolean_t expanding = (oldc != 0);
1402 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1405 * This vdev is not being allocated from yet or is a hole.
1407 if (vd->vdev_ms_shift == 0)
1410 ASSERT(!vd->vdev_ishole);
1412 ASSERT(oldc <= newc);
1414 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1417 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1418 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1422 vd->vdev_ms_count = newc;
1424 for (uint64_t m = oldc; m < newc; m++) {
1425 uint64_t object = 0;
1427 * vdev_ms_array may be 0 if we are creating the "fake"
1428 * metaslabs for an indirect vdev for zdb's leak detection.
1429 * See zdb_leak_init().
1431 if (txg == 0 && vd->vdev_ms_array != 0) {
1432 error = dmu_read(spa->spa_meta_objset,
1434 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1437 vdev_dbgmsg(vd, "unable to read the metaslab "
1438 "array [error=%d]", error);
1443 error = metaslab_init(vd->vdev_mg, m, object, txg,
1446 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1453 * Find the emptiest metaslab on the vdev and mark it for use for
1454 * embedded slog by moving it from the regular to the log metaslab
1457 if (vd->vdev_mg->mg_class == spa_normal_class(spa) &&
1458 vd->vdev_ms_count > zfs_embedded_slog_min_ms &&
1459 avl_is_empty(&vd->vdev_log_mg->mg_metaslab_tree)) {
1460 uint64_t slog_msid = 0;
1461 uint64_t smallest = UINT64_MAX;
1464 * Note, we only search the new metaslabs, because the old
1465 * (pre-existing) ones may be active (e.g. have non-empty
1466 * range_tree's), and we don't move them to the new
1469 for (uint64_t m = oldc; m < newc; m++) {
1471 space_map_allocated(vd->vdev_ms[m]->ms_sm);
1472 if (alloc < smallest) {
1477 metaslab_t *slog_ms = vd->vdev_ms[slog_msid];
1479 * The metaslab was marked as dirty at the end of
1480 * metaslab_init(). Remove it from the dirty list so that we
1481 * can uninitialize and reinitialize it to the new class.
1484 (void) txg_list_remove_this(&vd->vdev_ms_list,
1487 uint64_t sm_obj = space_map_object(slog_ms->ms_sm);
1488 metaslab_fini(slog_ms);
1489 VERIFY0(metaslab_init(vd->vdev_log_mg, slog_msid, sm_obj, txg,
1490 &vd->vdev_ms[slog_msid]));
1494 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1497 * If the vdev is being removed we don't activate
1498 * the metaslabs since we want to ensure that no new
1499 * allocations are performed on this device.
1501 if (!expanding && !vd->vdev_removing) {
1502 metaslab_group_activate(vd->vdev_mg);
1503 if (vd->vdev_log_mg != NULL)
1504 metaslab_group_activate(vd->vdev_log_mg);
1508 spa_config_exit(spa, SCL_ALLOC, FTAG);
1511 * Regardless whether this vdev was just added or it is being
1512 * expanded, the metaslab count has changed. Recalculate the
1515 spa_log_sm_set_blocklimit(spa);
1521 vdev_metaslab_fini(vdev_t *vd)
1523 if (vd->vdev_checkpoint_sm != NULL) {
1524 ASSERT(spa_feature_is_active(vd->vdev_spa,
1525 SPA_FEATURE_POOL_CHECKPOINT));
1526 space_map_close(vd->vdev_checkpoint_sm);
1528 * Even though we close the space map, we need to set its
1529 * pointer to NULL. The reason is that vdev_metaslab_fini()
1530 * may be called multiple times for certain operations
1531 * (i.e. when destroying a pool) so we need to ensure that
1532 * this clause never executes twice. This logic is similar
1533 * to the one used for the vdev_ms clause below.
1535 vd->vdev_checkpoint_sm = NULL;
1538 if (vd->vdev_ms != NULL) {
1539 metaslab_group_t *mg = vd->vdev_mg;
1541 metaslab_group_passivate(mg);
1542 if (vd->vdev_log_mg != NULL) {
1543 ASSERT(!vd->vdev_islog);
1544 metaslab_group_passivate(vd->vdev_log_mg);
1547 uint64_t count = vd->vdev_ms_count;
1548 for (uint64_t m = 0; m < count; m++) {
1549 metaslab_t *msp = vd->vdev_ms[m];
1553 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1555 vd->vdev_ms_count = 0;
1557 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
1558 ASSERT0(mg->mg_histogram[i]);
1559 if (vd->vdev_log_mg != NULL)
1560 ASSERT0(vd->vdev_log_mg->mg_histogram[i]);
1563 ASSERT0(vd->vdev_ms_count);
1564 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
1567 typedef struct vdev_probe_stats {
1568 boolean_t vps_readable;
1569 boolean_t vps_writeable;
1571 } vdev_probe_stats_t;
1574 vdev_probe_done(zio_t *zio)
1576 spa_t *spa = zio->io_spa;
1577 vdev_t *vd = zio->io_vd;
1578 vdev_probe_stats_t *vps = zio->io_private;
1580 ASSERT(vd->vdev_probe_zio != NULL);
1582 if (zio->io_type == ZIO_TYPE_READ) {
1583 if (zio->io_error == 0)
1584 vps->vps_readable = 1;
1585 if (zio->io_error == 0 && spa_writeable(spa)) {
1586 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1587 zio->io_offset, zio->io_size, zio->io_abd,
1588 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1589 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1591 abd_free(zio->io_abd);
1593 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1594 if (zio->io_error == 0)
1595 vps->vps_writeable = 1;
1596 abd_free(zio->io_abd);
1597 } else if (zio->io_type == ZIO_TYPE_NULL) {
1601 vd->vdev_cant_read |= !vps->vps_readable;
1602 vd->vdev_cant_write |= !vps->vps_writeable;
1604 if (vdev_readable(vd) &&
1605 (vdev_writeable(vd) || !spa_writeable(spa))) {
1608 ASSERT(zio->io_error != 0);
1609 vdev_dbgmsg(vd, "failed probe");
1610 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1611 spa, vd, NULL, NULL, 0);
1612 zio->io_error = SET_ERROR(ENXIO);
1615 mutex_enter(&vd->vdev_probe_lock);
1616 ASSERT(vd->vdev_probe_zio == zio);
1617 vd->vdev_probe_zio = NULL;
1618 mutex_exit(&vd->vdev_probe_lock);
1621 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1622 if (!vdev_accessible(vd, pio))
1623 pio->io_error = SET_ERROR(ENXIO);
1625 kmem_free(vps, sizeof (*vps));
1630 * Determine whether this device is accessible.
1632 * Read and write to several known locations: the pad regions of each
1633 * vdev label but the first, which we leave alone in case it contains
1637 vdev_probe(vdev_t *vd, zio_t *zio)
1639 spa_t *spa = vd->vdev_spa;
1640 vdev_probe_stats_t *vps = NULL;
1643 ASSERT(vd->vdev_ops->vdev_op_leaf);
1646 * Don't probe the probe.
1648 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1652 * To prevent 'probe storms' when a device fails, we create
1653 * just one probe i/o at a time. All zios that want to probe
1654 * this vdev will become parents of the probe io.
1656 mutex_enter(&vd->vdev_probe_lock);
1658 if ((pio = vd->vdev_probe_zio) == NULL) {
1659 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1661 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1662 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1665 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1667 * vdev_cant_read and vdev_cant_write can only
1668 * transition from TRUE to FALSE when we have the
1669 * SCL_ZIO lock as writer; otherwise they can only
1670 * transition from FALSE to TRUE. This ensures that
1671 * any zio looking at these values can assume that
1672 * failures persist for the life of the I/O. That's
1673 * important because when a device has intermittent
1674 * connectivity problems, we want to ensure that
1675 * they're ascribed to the device (ENXIO) and not
1678 * Since we hold SCL_ZIO as writer here, clear both
1679 * values so the probe can reevaluate from first
1682 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1683 vd->vdev_cant_read = B_FALSE;
1684 vd->vdev_cant_write = B_FALSE;
1687 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1688 vdev_probe_done, vps,
1689 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1692 * We can't change the vdev state in this context, so we
1693 * kick off an async task to do it on our behalf.
1696 vd->vdev_probe_wanted = B_TRUE;
1697 spa_async_request(spa, SPA_ASYNC_PROBE);
1702 zio_add_child(zio, pio);
1704 mutex_exit(&vd->vdev_probe_lock);
1707 ASSERT(zio != NULL);
1711 for (int l = 1; l < VDEV_LABELS; l++) {
1712 zio_nowait(zio_read_phys(pio, vd,
1713 vdev_label_offset(vd->vdev_psize, l,
1714 offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE,
1715 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1716 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1717 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1728 vdev_load_child(void *arg)
1732 vd->vdev_load_error = vdev_load(vd);
1736 vdev_open_child(void *arg)
1740 vd->vdev_open_thread = curthread;
1741 vd->vdev_open_error = vdev_open(vd);
1742 vd->vdev_open_thread = NULL;
1746 vdev_uses_zvols(vdev_t *vd)
1749 if (zvol_is_zvol(vd->vdev_path))
1753 for (int c = 0; c < vd->vdev_children; c++)
1754 if (vdev_uses_zvols(vd->vdev_child[c]))
1761 * Returns B_TRUE if the passed child should be opened.
1764 vdev_default_open_children_func(vdev_t *vd)
1770 * Open the requested child vdevs. If any of the leaf vdevs are using
1771 * a ZFS volume then do the opens in a single thread. This avoids a
1772 * deadlock when the current thread is holding the spa_namespace_lock.
1775 vdev_open_children_impl(vdev_t *vd, vdev_open_children_func_t *open_func)
1777 int children = vd->vdev_children;
1779 taskq_t *tq = taskq_create("vdev_open", children, minclsyspri,
1780 children, children, TASKQ_PREPOPULATE);
1781 vd->vdev_nonrot = B_TRUE;
1783 for (int c = 0; c < children; c++) {
1784 vdev_t *cvd = vd->vdev_child[c];
1786 if (open_func(cvd) == B_FALSE)
1789 if (tq == NULL || vdev_uses_zvols(vd)) {
1790 cvd->vdev_open_error = vdev_open(cvd);
1792 VERIFY(taskq_dispatch(tq, vdev_open_child,
1793 cvd, TQ_SLEEP) != TASKQID_INVALID);
1796 vd->vdev_nonrot &= cvd->vdev_nonrot;
1806 * Open all child vdevs.
1809 vdev_open_children(vdev_t *vd)
1811 vdev_open_children_impl(vd, vdev_default_open_children_func);
1815 * Conditionally open a subset of child vdevs.
1818 vdev_open_children_subset(vdev_t *vd, vdev_open_children_func_t *open_func)
1820 vdev_open_children_impl(vd, open_func);
1824 * Compute the raidz-deflation ratio. Note, we hard-code
1825 * in 128k (1 << 17) because it is the "typical" blocksize.
1826 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1827 * otherwise it would inconsistently account for existing bp's.
1830 vdev_set_deflate_ratio(vdev_t *vd)
1832 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1833 vd->vdev_deflate_ratio = (1 << 17) /
1834 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1839 * Maximize performance by inflating the configured ashift for top level
1840 * vdevs to be as close to the physical ashift as possible while maintaining
1841 * administrator defined limits and ensuring it doesn't go below the
1845 vdev_ashift_optimize(vdev_t *vd)
1847 ASSERT(vd == vd->vdev_top);
1849 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1850 vd->vdev_ashift = MIN(
1851 MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift),
1852 MAX(zfs_vdev_min_auto_ashift,
1853 vd->vdev_physical_ashift));
1856 * If the logical and physical ashifts are the same, then
1857 * we ensure that the top-level vdev's ashift is not smaller
1858 * than our minimum ashift value. For the unusual case
1859 * where logical ashift > physical ashift, we can't cap
1860 * the calculated ashift based on max ashift as that
1861 * would cause failures.
1862 * We still check if we need to increase it to match
1865 vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift,
1871 * Prepare a virtual device for access.
1874 vdev_open(vdev_t *vd)
1876 spa_t *spa = vd->vdev_spa;
1879 uint64_t max_osize = 0;
1880 uint64_t asize, max_asize, psize;
1881 uint64_t logical_ashift = 0;
1882 uint64_t physical_ashift = 0;
1884 ASSERT(vd->vdev_open_thread == curthread ||
1885 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1886 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1887 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1888 vd->vdev_state == VDEV_STATE_OFFLINE);
1890 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1891 vd->vdev_cant_read = B_FALSE;
1892 vd->vdev_cant_write = B_FALSE;
1893 vd->vdev_min_asize = vdev_get_min_asize(vd);
1896 * If this vdev is not removed, check its fault status. If it's
1897 * faulted, bail out of the open.
1899 if (!vd->vdev_removed && vd->vdev_faulted) {
1900 ASSERT(vd->vdev_children == 0);
1901 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1902 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1903 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1904 vd->vdev_label_aux);
1905 return (SET_ERROR(ENXIO));
1906 } else if (vd->vdev_offline) {
1907 ASSERT(vd->vdev_children == 0);
1908 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1909 return (SET_ERROR(ENXIO));
1912 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1913 &logical_ashift, &physical_ashift);
1915 * Physical volume size should never be larger than its max size, unless
1916 * the disk has shrunk while we were reading it or the device is buggy
1917 * or damaged: either way it's not safe for use, bail out of the open.
1919 if (osize > max_osize) {
1920 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1921 VDEV_AUX_OPEN_FAILED);
1922 return (SET_ERROR(ENXIO));
1926 * Reset the vdev_reopening flag so that we actually close
1927 * the vdev on error.
1929 vd->vdev_reopening = B_FALSE;
1930 if (zio_injection_enabled && error == 0)
1931 error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO));
1934 if (vd->vdev_removed &&
1935 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1936 vd->vdev_removed = B_FALSE;
1938 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1939 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1940 vd->vdev_stat.vs_aux);
1942 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1943 vd->vdev_stat.vs_aux);
1948 vd->vdev_removed = B_FALSE;
1951 * Recheck the faulted flag now that we have confirmed that
1952 * the vdev is accessible. If we're faulted, bail.
1954 if (vd->vdev_faulted) {
1955 ASSERT(vd->vdev_children == 0);
1956 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1957 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1958 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1959 vd->vdev_label_aux);
1960 return (SET_ERROR(ENXIO));
1963 if (vd->vdev_degraded) {
1964 ASSERT(vd->vdev_children == 0);
1965 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1966 VDEV_AUX_ERR_EXCEEDED);
1968 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1972 * For hole or missing vdevs we just return success.
1974 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1977 for (int c = 0; c < vd->vdev_children; c++) {
1978 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1979 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1985 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1986 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1988 if (vd->vdev_children == 0) {
1989 if (osize < SPA_MINDEVSIZE) {
1990 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1991 VDEV_AUX_TOO_SMALL);
1992 return (SET_ERROR(EOVERFLOW));
1995 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1996 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1997 VDEV_LABEL_END_SIZE);
1999 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
2000 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
2001 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2002 VDEV_AUX_TOO_SMALL);
2003 return (SET_ERROR(EOVERFLOW));
2007 max_asize = max_osize;
2011 * If the vdev was expanded, record this so that we can re-create the
2012 * uberblock rings in labels {2,3}, during the next sync.
2014 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0))
2015 vd->vdev_copy_uberblocks = B_TRUE;
2017 vd->vdev_psize = psize;
2020 * Make sure the allocatable size hasn't shrunk too much.
2022 if (asize < vd->vdev_min_asize) {
2023 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2024 VDEV_AUX_BAD_LABEL);
2025 return (SET_ERROR(EINVAL));
2029 * We can always set the logical/physical ashift members since
2030 * their values are only used to calculate the vdev_ashift when
2031 * the device is first added to the config. These values should
2032 * not be used for anything else since they may change whenever
2033 * the device is reopened and we don't store them in the label.
2035 vd->vdev_physical_ashift =
2036 MAX(physical_ashift, vd->vdev_physical_ashift);
2037 vd->vdev_logical_ashift = MAX(logical_ashift,
2038 vd->vdev_logical_ashift);
2040 if (vd->vdev_asize == 0) {
2042 * This is the first-ever open, so use the computed values.
2043 * For compatibility, a different ashift can be requested.
2045 vd->vdev_asize = asize;
2046 vd->vdev_max_asize = max_asize;
2049 * If the vdev_ashift was not overriden at creation time,
2050 * then set it the logical ashift and optimize the ashift.
2052 if (vd->vdev_ashift == 0) {
2053 vd->vdev_ashift = vd->vdev_logical_ashift;
2055 if (vd->vdev_logical_ashift > ASHIFT_MAX) {
2056 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2057 VDEV_AUX_ASHIFT_TOO_BIG);
2058 return (SET_ERROR(EDOM));
2061 if (vd->vdev_top == vd) {
2062 vdev_ashift_optimize(vd);
2065 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
2066 vd->vdev_ashift > ASHIFT_MAX)) {
2067 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2068 VDEV_AUX_BAD_ASHIFT);
2069 return (SET_ERROR(EDOM));
2073 * Make sure the alignment required hasn't increased.
2075 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
2076 vd->vdev_ops->vdev_op_leaf) {
2077 (void) zfs_ereport_post(
2078 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
2079 spa, vd, NULL, NULL, 0);
2080 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2081 VDEV_AUX_BAD_LABEL);
2082 return (SET_ERROR(EDOM));
2084 vd->vdev_max_asize = max_asize;
2088 * If all children are healthy we update asize if either:
2089 * The asize has increased, due to a device expansion caused by dynamic
2090 * LUN growth or vdev replacement, and automatic expansion is enabled;
2091 * making the additional space available.
2093 * The asize has decreased, due to a device shrink usually caused by a
2094 * vdev replace with a smaller device. This ensures that calculations
2095 * based of max_asize and asize e.g. esize are always valid. It's safe
2096 * to do this as we've already validated that asize is greater than
2099 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
2100 ((asize > vd->vdev_asize &&
2101 (vd->vdev_expanding || spa->spa_autoexpand)) ||
2102 (asize < vd->vdev_asize)))
2103 vd->vdev_asize = asize;
2105 vdev_set_min_asize(vd);
2108 * Ensure we can issue some IO before declaring the
2109 * vdev open for business.
2111 if (vd->vdev_ops->vdev_op_leaf &&
2112 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
2113 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
2114 VDEV_AUX_ERR_EXCEEDED);
2119 * Track the the minimum allocation size.
2121 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
2122 vd->vdev_islog == 0 && vd->vdev_aux == NULL) {
2123 uint64_t min_alloc = vdev_get_min_alloc(vd);
2124 if (min_alloc < spa->spa_min_alloc)
2125 spa->spa_min_alloc = min_alloc;
2129 * If this is a leaf vdev, assess whether a resilver is needed.
2130 * But don't do this if we are doing a reopen for a scrub, since
2131 * this would just restart the scrub we are already doing.
2133 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen)
2134 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd);
2140 vdev_validate_child(void *arg)
2144 vd->vdev_validate_thread = curthread;
2145 vd->vdev_validate_error = vdev_validate(vd);
2146 vd->vdev_validate_thread = NULL;
2150 * Called once the vdevs are all opened, this routine validates the label
2151 * contents. This needs to be done before vdev_load() so that we don't
2152 * inadvertently do repair I/Os to the wrong device.
2154 * This function will only return failure if one of the vdevs indicates that it
2155 * has since been destroyed or exported. This is only possible if
2156 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
2157 * will be updated but the function will return 0.
2160 vdev_validate(vdev_t *vd)
2162 spa_t *spa = vd->vdev_spa;
2165 uint64_t guid = 0, aux_guid = 0, top_guid;
2169 int children = vd->vdev_children;
2171 if (vdev_validate_skip)
2175 tq = taskq_create("vdev_validate", children, minclsyspri,
2176 children, children, TASKQ_PREPOPULATE);
2179 for (uint64_t c = 0; c < children; c++) {
2180 vdev_t *cvd = vd->vdev_child[c];
2182 if (tq == NULL || vdev_uses_zvols(cvd)) {
2183 vdev_validate_child(cvd);
2185 VERIFY(taskq_dispatch(tq, vdev_validate_child, cvd,
2186 TQ_SLEEP) != TASKQID_INVALID);
2193 for (int c = 0; c < children; c++) {
2194 int error = vd->vdev_child[c]->vdev_validate_error;
2197 return (SET_ERROR(EBADF));
2202 * If the device has already failed, or was marked offline, don't do
2203 * any further validation. Otherwise, label I/O will fail and we will
2204 * overwrite the previous state.
2206 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
2210 * If we are performing an extreme rewind, we allow for a label that
2211 * was modified at a point after the current txg.
2212 * If config lock is not held do not check for the txg. spa_sync could
2213 * be updating the vdev's label before updating spa_last_synced_txg.
2215 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
2216 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
2219 txg = spa_last_synced_txg(spa);
2221 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
2222 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2223 VDEV_AUX_BAD_LABEL);
2224 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
2225 "txg %llu", (u_longlong_t)txg);
2230 * Determine if this vdev has been split off into another
2231 * pool. If so, then refuse to open it.
2233 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
2234 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
2235 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2236 VDEV_AUX_SPLIT_POOL);
2238 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
2242 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
2243 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2244 VDEV_AUX_CORRUPT_DATA);
2246 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2247 ZPOOL_CONFIG_POOL_GUID);
2252 * If config is not trusted then ignore the spa guid check. This is
2253 * necessary because if the machine crashed during a re-guid the new
2254 * guid might have been written to all of the vdev labels, but not the
2255 * cached config. The check will be performed again once we have the
2256 * trusted config from the MOS.
2258 if (spa->spa_trust_config && guid != spa_guid(spa)) {
2259 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2260 VDEV_AUX_CORRUPT_DATA);
2262 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
2263 "match config (%llu != %llu)", (u_longlong_t)guid,
2264 (u_longlong_t)spa_guid(spa));
2268 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
2269 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
2273 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
2274 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2275 VDEV_AUX_CORRUPT_DATA);
2277 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2282 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
2284 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2285 VDEV_AUX_CORRUPT_DATA);
2287 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2288 ZPOOL_CONFIG_TOP_GUID);
2293 * If this vdev just became a top-level vdev because its sibling was
2294 * detached, it will have adopted the parent's vdev guid -- but the
2295 * label may or may not be on disk yet. Fortunately, either version
2296 * of the label will have the same top guid, so if we're a top-level
2297 * vdev, we can safely compare to that instead.
2298 * However, if the config comes from a cachefile that failed to update
2299 * after the detach, a top-level vdev will appear as a non top-level
2300 * vdev in the config. Also relax the constraints if we perform an
2303 * If we split this vdev off instead, then we also check the
2304 * original pool's guid. We don't want to consider the vdev
2305 * corrupt if it is partway through a split operation.
2307 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
2308 boolean_t mismatch = B_FALSE;
2309 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
2310 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
2313 if (vd->vdev_guid != top_guid &&
2314 vd->vdev_top->vdev_guid != guid)
2319 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2320 VDEV_AUX_CORRUPT_DATA);
2322 vdev_dbgmsg(vd, "vdev_validate: config guid "
2323 "doesn't match label guid");
2324 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
2325 (u_longlong_t)vd->vdev_guid,
2326 (u_longlong_t)vd->vdev_top->vdev_guid);
2327 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
2328 "aux_guid %llu", (u_longlong_t)guid,
2329 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
2334 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
2336 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2337 VDEV_AUX_CORRUPT_DATA);
2339 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2340 ZPOOL_CONFIG_POOL_STATE);
2347 * If this is a verbatim import, no need to check the
2348 * state of the pool.
2350 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
2351 spa_load_state(spa) == SPA_LOAD_OPEN &&
2352 state != POOL_STATE_ACTIVE) {
2353 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
2354 "for spa %s", (u_longlong_t)state, spa->spa_name);
2355 return (SET_ERROR(EBADF));
2359 * If we were able to open and validate a vdev that was
2360 * previously marked permanently unavailable, clear that state
2363 if (vd->vdev_not_present)
2364 vd->vdev_not_present = 0;
2370 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
2372 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
2373 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
2374 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2375 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
2376 dvd->vdev_path, svd->vdev_path);
2377 spa_strfree(dvd->vdev_path);
2378 dvd->vdev_path = spa_strdup(svd->vdev_path);
2380 } else if (svd->vdev_path != NULL) {
2381 dvd->vdev_path = spa_strdup(svd->vdev_path);
2382 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2383 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
2388 * Recursively copy vdev paths from one vdev to another. Source and destination
2389 * vdev trees must have same geometry otherwise return error. Intended to copy
2390 * paths from userland config into MOS config.
2393 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
2395 if ((svd->vdev_ops == &vdev_missing_ops) ||
2396 (svd->vdev_ishole && dvd->vdev_ishole) ||
2397 (dvd->vdev_ops == &vdev_indirect_ops))
2400 if (svd->vdev_ops != dvd->vdev_ops) {
2401 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
2402 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
2403 return (SET_ERROR(EINVAL));
2406 if (svd->vdev_guid != dvd->vdev_guid) {
2407 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
2408 "%llu)", (u_longlong_t)svd->vdev_guid,
2409 (u_longlong_t)dvd->vdev_guid);
2410 return (SET_ERROR(EINVAL));
2413 if (svd->vdev_children != dvd->vdev_children) {
2414 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2415 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2416 (u_longlong_t)dvd->vdev_children);
2417 return (SET_ERROR(EINVAL));
2420 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2421 int error = vdev_copy_path_strict(svd->vdev_child[i],
2422 dvd->vdev_child[i]);
2427 if (svd->vdev_ops->vdev_op_leaf)
2428 vdev_copy_path_impl(svd, dvd);
2434 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2436 ASSERT(stvd->vdev_top == stvd);
2437 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2439 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2440 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2443 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2447 * The idea here is that while a vdev can shift positions within
2448 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2449 * step outside of it.
2451 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2453 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2456 ASSERT(vd->vdev_ops->vdev_op_leaf);
2458 vdev_copy_path_impl(vd, dvd);
2462 * Recursively copy vdev paths from one root vdev to another. Source and
2463 * destination vdev trees may differ in geometry. For each destination leaf
2464 * vdev, search a vdev with the same guid and top vdev id in the source.
2465 * Intended to copy paths from userland config into MOS config.
2468 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2470 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2471 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2472 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2474 for (uint64_t i = 0; i < children; i++) {
2475 vdev_copy_path_search(srvd->vdev_child[i],
2476 drvd->vdev_child[i]);
2481 * Close a virtual device.
2484 vdev_close(vdev_t *vd)
2486 vdev_t *pvd = vd->vdev_parent;
2487 spa_t *spa __maybe_unused = vd->vdev_spa;
2490 ASSERT(vd->vdev_open_thread == curthread ||
2491 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2494 * If our parent is reopening, then we are as well, unless we are
2497 if (pvd != NULL && pvd->vdev_reopening)
2498 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2500 vd->vdev_ops->vdev_op_close(vd);
2502 vdev_cache_purge(vd);
2505 * We record the previous state before we close it, so that if we are
2506 * doing a reopen(), we don't generate FMA ereports if we notice that
2507 * it's still faulted.
2509 vd->vdev_prevstate = vd->vdev_state;
2511 if (vd->vdev_offline)
2512 vd->vdev_state = VDEV_STATE_OFFLINE;
2514 vd->vdev_state = VDEV_STATE_CLOSED;
2515 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2519 vdev_hold(vdev_t *vd)
2521 spa_t *spa = vd->vdev_spa;
2523 ASSERT(spa_is_root(spa));
2524 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2527 for (int c = 0; c < vd->vdev_children; c++)
2528 vdev_hold(vd->vdev_child[c]);
2530 if (vd->vdev_ops->vdev_op_leaf)
2531 vd->vdev_ops->vdev_op_hold(vd);
2535 vdev_rele(vdev_t *vd)
2537 ASSERT(spa_is_root(vd->vdev_spa));
2538 for (int c = 0; c < vd->vdev_children; c++)
2539 vdev_rele(vd->vdev_child[c]);
2541 if (vd->vdev_ops->vdev_op_leaf)
2542 vd->vdev_ops->vdev_op_rele(vd);
2546 * Reopen all interior vdevs and any unopened leaves. We don't actually
2547 * reopen leaf vdevs which had previously been opened as they might deadlock
2548 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2549 * If the leaf has never been opened then open it, as usual.
2552 vdev_reopen(vdev_t *vd)
2554 spa_t *spa = vd->vdev_spa;
2556 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2558 /* set the reopening flag unless we're taking the vdev offline */
2559 vd->vdev_reopening = !vd->vdev_offline;
2561 (void) vdev_open(vd);
2564 * Call vdev_validate() here to make sure we have the same device.
2565 * Otherwise, a device with an invalid label could be successfully
2566 * opened in response to vdev_reopen().
2569 (void) vdev_validate_aux(vd);
2570 if (vdev_readable(vd) && vdev_writeable(vd) &&
2571 vd->vdev_aux == &spa->spa_l2cache) {
2573 * In case the vdev is present we should evict all ARC
2574 * buffers and pointers to log blocks and reclaim their
2575 * space before restoring its contents to L2ARC.
2577 if (l2arc_vdev_present(vd)) {
2578 l2arc_rebuild_vdev(vd, B_TRUE);
2580 l2arc_add_vdev(spa, vd);
2582 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
2583 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
2586 (void) vdev_validate(vd);
2590 * Reassess parent vdev's health.
2592 vdev_propagate_state(vd);
2596 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2601 * Normally, partial opens (e.g. of a mirror) are allowed.
2602 * For a create, however, we want to fail the request if
2603 * there are any components we can't open.
2605 error = vdev_open(vd);
2607 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2609 return (error ? error : SET_ERROR(ENXIO));
2613 * Recursively load DTLs and initialize all labels.
2615 if ((error = vdev_dtl_load(vd)) != 0 ||
2616 (error = vdev_label_init(vd, txg, isreplacing ?
2617 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2626 vdev_metaslab_set_size(vdev_t *vd)
2628 uint64_t asize = vd->vdev_asize;
2629 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2633 * There are two dimensions to the metaslab sizing calculation:
2634 * the size of the metaslab and the count of metaslabs per vdev.
2636 * The default values used below are a good balance between memory
2637 * usage (larger metaslab size means more memory needed for loaded
2638 * metaslabs; more metaslabs means more memory needed for the
2639 * metaslab_t structs), metaslab load time (larger metaslabs take
2640 * longer to load), and metaslab sync time (more metaslabs means
2641 * more time spent syncing all of them).
2643 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2644 * The range of the dimensions are as follows:
2646 * 2^29 <= ms_size <= 2^34
2647 * 16 <= ms_count <= 131,072
2649 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2650 * at least 512MB (2^29) to minimize fragmentation effects when
2651 * testing with smaller devices. However, the count constraint
2652 * of at least 16 metaslabs will override this minimum size goal.
2654 * On the upper end of vdev sizes, we aim for a maximum metaslab
2655 * size of 16GB. However, we will cap the total count to 2^17
2656 * metaslabs to keep our memory footprint in check and let the
2657 * metaslab size grow from there if that limit is hit.
2659 * The net effect of applying above constrains is summarized below.
2661 * vdev size metaslab count
2662 * --------------|-----------------
2664 * 8GB - 100GB one per 512MB
2666 * 3TB - 2PB one per 16GB
2668 * --------------------------------
2670 * Finally, note that all of the above calculate the initial
2671 * number of metaslabs. Expanding a top-level vdev will result
2672 * in additional metaslabs being allocated making it possible
2673 * to exceed the zfs_vdev_ms_count_limit.
2676 if (ms_count < zfs_vdev_min_ms_count)
2677 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2678 else if (ms_count > zfs_vdev_default_ms_count)
2679 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2681 ms_shift = zfs_vdev_default_ms_shift;
2683 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2684 ms_shift = SPA_MAXBLOCKSHIFT;
2685 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2686 ms_shift = zfs_vdev_max_ms_shift;
2687 /* cap the total count to constrain memory footprint */
2688 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2689 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2692 vd->vdev_ms_shift = ms_shift;
2693 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2697 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2699 ASSERT(vd == vd->vdev_top);
2700 /* indirect vdevs don't have metaslabs or dtls */
2701 ASSERT(vdev_is_concrete(vd) || flags == 0);
2702 ASSERT(ISP2(flags));
2703 ASSERT(spa_writeable(vd->vdev_spa));
2705 if (flags & VDD_METASLAB)
2706 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2708 if (flags & VDD_DTL)
2709 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2711 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2715 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2717 for (int c = 0; c < vd->vdev_children; c++)
2718 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2720 if (vd->vdev_ops->vdev_op_leaf)
2721 vdev_dirty(vd->vdev_top, flags, vd, txg);
2727 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2728 * the vdev has less than perfect replication. There are four kinds of DTL:
2730 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2732 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2734 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2735 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2736 * txgs that was scrubbed.
2738 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2739 * persistent errors or just some device being offline.
2740 * Unlike the other three, the DTL_OUTAGE map is not generally
2741 * maintained; it's only computed when needed, typically to
2742 * determine whether a device can be detached.
2744 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2745 * either has the data or it doesn't.
2747 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2748 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2749 * if any child is less than fully replicated, then so is its parent.
2750 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2751 * comprising only those txgs which appear in 'maxfaults' or more children;
2752 * those are the txgs we don't have enough replication to read. For example,
2753 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2754 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2755 * two child DTL_MISSING maps.
2757 * It should be clear from the above that to compute the DTLs and outage maps
2758 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2759 * Therefore, that is all we keep on disk. When loading the pool, or after
2760 * a configuration change, we generate all other DTLs from first principles.
2763 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2765 range_tree_t *rt = vd->vdev_dtl[t];
2767 ASSERT(t < DTL_TYPES);
2768 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2769 ASSERT(spa_writeable(vd->vdev_spa));
2771 mutex_enter(&vd->vdev_dtl_lock);
2772 if (!range_tree_contains(rt, txg, size))
2773 range_tree_add(rt, txg, size);
2774 mutex_exit(&vd->vdev_dtl_lock);
2778 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2780 range_tree_t *rt = vd->vdev_dtl[t];
2781 boolean_t dirty = B_FALSE;
2783 ASSERT(t < DTL_TYPES);
2784 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2787 * While we are loading the pool, the DTLs have not been loaded yet.
2788 * This isn't a problem but it can result in devices being tried
2789 * which are known to not have the data. In which case, the import
2790 * is relying on the checksum to ensure that we get the right data.
2791 * Note that while importing we are only reading the MOS, which is
2792 * always checksummed.
2794 mutex_enter(&vd->vdev_dtl_lock);
2795 if (!range_tree_is_empty(rt))
2796 dirty = range_tree_contains(rt, txg, size);
2797 mutex_exit(&vd->vdev_dtl_lock);
2803 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2805 range_tree_t *rt = vd->vdev_dtl[t];
2808 mutex_enter(&vd->vdev_dtl_lock);
2809 empty = range_tree_is_empty(rt);
2810 mutex_exit(&vd->vdev_dtl_lock);
2816 * Check if the txg falls within the range which must be
2817 * resilvered. DVAs outside this range can always be skipped.
2820 vdev_default_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize,
2821 uint64_t phys_birth)
2823 /* Set by sequential resilver. */
2824 if (phys_birth == TXG_UNKNOWN)
2827 return (vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1));
2831 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered.
2834 vdev_dtl_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize,
2835 uint64_t phys_birth)
2837 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2839 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2840 vd->vdev_ops->vdev_op_leaf)
2843 return (vd->vdev_ops->vdev_op_need_resilver(vd, dva, psize,
2848 * Returns the lowest txg in the DTL range.
2851 vdev_dtl_min(vdev_t *vd)
2853 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2854 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2855 ASSERT0(vd->vdev_children);
2857 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1);
2861 * Returns the highest txg in the DTL.
2864 vdev_dtl_max(vdev_t *vd)
2866 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2867 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2868 ASSERT0(vd->vdev_children);
2870 return (range_tree_max(vd->vdev_dtl[DTL_MISSING]));
2874 * Determine if a resilvering vdev should remove any DTL entries from
2875 * its range. If the vdev was resilvering for the entire duration of the
2876 * scan then it should excise that range from its DTLs. Otherwise, this
2877 * vdev is considered partially resilvered and should leave its DTL
2878 * entries intact. The comment in vdev_dtl_reassess() describes how we
2882 vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done)
2884 ASSERT0(vd->vdev_children);
2886 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2889 if (vd->vdev_resilver_deferred)
2892 if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2896 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2897 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
2899 /* Rebuild not initiated by attach */
2900 if (vd->vdev_rebuild_txg == 0)
2904 * When a rebuild completes without error then all missing data
2905 * up to the rebuild max txg has been reconstructed and the DTL
2906 * is eligible for excision.
2908 if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE &&
2909 vdev_dtl_max(vd) <= vrp->vrp_max_txg) {
2910 ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd));
2911 ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg);
2912 ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg);
2916 dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan;
2917 dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys;
2919 /* Resilver not initiated by attach */
2920 if (vd->vdev_resilver_txg == 0)
2924 * When a resilver is initiated the scan will assign the
2925 * scn_max_txg value to the highest txg value that exists
2926 * in all DTLs. If this device's max DTL is not part of this
2927 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
2928 * then it is not eligible for excision.
2930 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2931 ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd));
2932 ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg);
2933 ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg);
2942 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2943 * write operations will be issued to the pool.
2946 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
2947 boolean_t scrub_done, boolean_t rebuild_done)
2949 spa_t *spa = vd->vdev_spa;
2953 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2955 for (int c = 0; c < vd->vdev_children; c++)
2956 vdev_dtl_reassess(vd->vdev_child[c], txg,
2957 scrub_txg, scrub_done, rebuild_done);
2959 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2962 if (vd->vdev_ops->vdev_op_leaf) {
2963 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2964 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2965 boolean_t check_excise = B_FALSE;
2966 boolean_t wasempty = B_TRUE;
2968 mutex_enter(&vd->vdev_dtl_lock);
2971 * If requested, pretend the scan or rebuild completed cleanly.
2973 if (zfs_scan_ignore_errors) {
2975 scn->scn_phys.scn_errors = 0;
2977 vr->vr_rebuild_phys.vrp_errors = 0;
2980 if (scrub_txg != 0 &&
2981 !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
2983 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
2984 "dtl:%llu/%llu errors:%llu",
2985 (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg,
2986 (u_longlong_t)scrub_txg, spa->spa_scrub_started,
2987 (u_longlong_t)vdev_dtl_min(vd),
2988 (u_longlong_t)vdev_dtl_max(vd),
2989 (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0));
2993 * If we've completed a scrub/resilver or a rebuild cleanly
2994 * then determine if this vdev should remove any DTLs. We
2995 * only want to excise regions on vdevs that were available
2996 * during the entire duration of this scan.
2999 vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) {
3000 check_excise = B_TRUE;
3002 if (spa->spa_scrub_started ||
3003 (scn != NULL && scn->scn_phys.scn_errors == 0)) {
3004 check_excise = B_TRUE;
3008 if (scrub_txg && check_excise &&
3009 vdev_dtl_should_excise(vd, rebuild_done)) {
3011 * We completed a scrub, resilver or rebuild up to
3012 * scrub_txg. If we did it without rebooting, then
3013 * the scrub dtl will be valid, so excise the old
3014 * region and fold in the scrub dtl. Otherwise,
3015 * leave the dtl as-is if there was an error.
3017 * There's little trick here: to excise the beginning
3018 * of the DTL_MISSING map, we put it into a reference
3019 * tree and then add a segment with refcnt -1 that
3020 * covers the range [0, scrub_txg). This means
3021 * that each txg in that range has refcnt -1 or 0.
3022 * We then add DTL_SCRUB with a refcnt of 2, so that
3023 * entries in the range [0, scrub_txg) will have a
3024 * positive refcnt -- either 1 or 2. We then convert
3025 * the reference tree into the new DTL_MISSING map.
3027 space_reftree_create(&reftree);
3028 space_reftree_add_map(&reftree,
3029 vd->vdev_dtl[DTL_MISSING], 1);
3030 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
3031 space_reftree_add_map(&reftree,
3032 vd->vdev_dtl[DTL_SCRUB], 2);
3033 space_reftree_generate_map(&reftree,
3034 vd->vdev_dtl[DTL_MISSING], 1);
3035 space_reftree_destroy(&reftree);
3037 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
3038 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
3039 (u_longlong_t)vdev_dtl_min(vd),
3040 (u_longlong_t)vdev_dtl_max(vd));
3041 } else if (!wasempty) {
3042 zfs_dbgmsg("DTL_MISSING is now empty");
3045 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
3046 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
3047 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
3049 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
3050 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
3051 if (!vdev_readable(vd))
3052 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
3054 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
3055 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
3058 * If the vdev was resilvering or rebuilding and no longer
3059 * has any DTLs then reset the appropriate flag and dirty
3060 * the top level so that we persist the change.
3063 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
3064 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
3065 if (vd->vdev_rebuild_txg != 0) {
3066 vd->vdev_rebuild_txg = 0;
3067 vdev_config_dirty(vd->vdev_top);
3068 } else if (vd->vdev_resilver_txg != 0) {
3069 vd->vdev_resilver_txg = 0;
3070 vdev_config_dirty(vd->vdev_top);
3074 mutex_exit(&vd->vdev_dtl_lock);
3077 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
3081 mutex_enter(&vd->vdev_dtl_lock);
3082 for (int t = 0; t < DTL_TYPES; t++) {
3083 /* account for child's outage in parent's missing map */
3084 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
3086 continue; /* leaf vdevs only */
3087 if (t == DTL_PARTIAL)
3088 minref = 1; /* i.e. non-zero */
3089 else if (vdev_get_nparity(vd) != 0)
3090 minref = vdev_get_nparity(vd) + 1; /* RAID-Z, dRAID */
3092 minref = vd->vdev_children; /* any kind of mirror */
3093 space_reftree_create(&reftree);
3094 for (int c = 0; c < vd->vdev_children; c++) {
3095 vdev_t *cvd = vd->vdev_child[c];
3096 mutex_enter(&cvd->vdev_dtl_lock);
3097 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
3098 mutex_exit(&cvd->vdev_dtl_lock);
3100 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
3101 space_reftree_destroy(&reftree);
3103 mutex_exit(&vd->vdev_dtl_lock);
3107 vdev_dtl_load(vdev_t *vd)
3109 spa_t *spa = vd->vdev_spa;
3110 objset_t *mos = spa->spa_meta_objset;
3114 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
3115 ASSERT(vdev_is_concrete(vd));
3117 error = space_map_open(&vd->vdev_dtl_sm, mos,
3118 vd->vdev_dtl_object, 0, -1ULL, 0);
3121 ASSERT(vd->vdev_dtl_sm != NULL);
3123 rt = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
3124 error = space_map_load(vd->vdev_dtl_sm, rt, SM_ALLOC);
3126 mutex_enter(&vd->vdev_dtl_lock);
3127 range_tree_walk(rt, range_tree_add,
3128 vd->vdev_dtl[DTL_MISSING]);
3129 mutex_exit(&vd->vdev_dtl_lock);
3132 range_tree_vacate(rt, NULL, NULL);
3133 range_tree_destroy(rt);
3138 for (int c = 0; c < vd->vdev_children; c++) {
3139 error = vdev_dtl_load(vd->vdev_child[c]);
3148 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
3150 spa_t *spa = vd->vdev_spa;
3151 objset_t *mos = spa->spa_meta_objset;
3152 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
3155 ASSERT(alloc_bias != VDEV_BIAS_NONE);
3158 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
3159 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
3160 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
3162 ASSERT(string != NULL);
3163 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
3164 1, strlen(string) + 1, string, tx));
3166 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
3167 spa_activate_allocation_classes(spa, tx);
3172 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
3174 spa_t *spa = vd->vdev_spa;
3176 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
3177 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
3182 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
3184 spa_t *spa = vd->vdev_spa;
3185 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
3186 DMU_OT_NONE, 0, tx);
3189 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
3196 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
3198 if (vd->vdev_ops != &vdev_hole_ops &&
3199 vd->vdev_ops != &vdev_missing_ops &&
3200 vd->vdev_ops != &vdev_root_ops &&
3201 !vd->vdev_top->vdev_removing) {
3202 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
3203 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
3205 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
3206 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
3207 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
3208 vdev_zap_allocation_data(vd, tx);
3212 for (uint64_t i = 0; i < vd->vdev_children; i++) {
3213 vdev_construct_zaps(vd->vdev_child[i], tx);
3218 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
3220 spa_t *spa = vd->vdev_spa;
3221 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
3222 objset_t *mos = spa->spa_meta_objset;
3223 range_tree_t *rtsync;
3225 uint64_t object = space_map_object(vd->vdev_dtl_sm);
3227 ASSERT(vdev_is_concrete(vd));
3228 ASSERT(vd->vdev_ops->vdev_op_leaf);
3230 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3232 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
3233 mutex_enter(&vd->vdev_dtl_lock);
3234 space_map_free(vd->vdev_dtl_sm, tx);
3235 space_map_close(vd->vdev_dtl_sm);
3236 vd->vdev_dtl_sm = NULL;
3237 mutex_exit(&vd->vdev_dtl_lock);
3240 * We only destroy the leaf ZAP for detached leaves or for
3241 * removed log devices. Removed data devices handle leaf ZAP
3242 * cleanup later, once cancellation is no longer possible.
3244 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
3245 vd->vdev_top->vdev_islog)) {
3246 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
3247 vd->vdev_leaf_zap = 0;
3254 if (vd->vdev_dtl_sm == NULL) {
3255 uint64_t new_object;
3257 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx);
3258 VERIFY3U(new_object, !=, 0);
3260 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
3262 ASSERT(vd->vdev_dtl_sm != NULL);
3265 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
3267 mutex_enter(&vd->vdev_dtl_lock);
3268 range_tree_walk(rt, range_tree_add, rtsync);
3269 mutex_exit(&vd->vdev_dtl_lock);
3271 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx);
3272 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
3273 range_tree_vacate(rtsync, NULL, NULL);
3275 range_tree_destroy(rtsync);
3278 * If the object for the space map has changed then dirty
3279 * the top level so that we update the config.
3281 if (object != space_map_object(vd->vdev_dtl_sm)) {
3282 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
3283 "new object %llu", (u_longlong_t)txg, spa_name(spa),
3284 (u_longlong_t)object,
3285 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
3286 vdev_config_dirty(vd->vdev_top);
3293 * Determine whether the specified vdev can be offlined/detached/removed
3294 * without losing data.
3297 vdev_dtl_required(vdev_t *vd)
3299 spa_t *spa = vd->vdev_spa;
3300 vdev_t *tvd = vd->vdev_top;
3301 uint8_t cant_read = vd->vdev_cant_read;
3304 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3306 if (vd == spa->spa_root_vdev || vd == tvd)
3310 * Temporarily mark the device as unreadable, and then determine
3311 * whether this results in any DTL outages in the top-level vdev.
3312 * If not, we can safely offline/detach/remove the device.
3314 vd->vdev_cant_read = B_TRUE;
3315 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3316 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
3317 vd->vdev_cant_read = cant_read;
3318 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3320 if (!required && zio_injection_enabled) {
3321 required = !!zio_handle_device_injection(vd, NULL,
3329 * Determine if resilver is needed, and if so the txg range.
3332 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
3334 boolean_t needed = B_FALSE;
3335 uint64_t thismin = UINT64_MAX;
3336 uint64_t thismax = 0;
3338 if (vd->vdev_children == 0) {
3339 mutex_enter(&vd->vdev_dtl_lock);
3340 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
3341 vdev_writeable(vd)) {
3343 thismin = vdev_dtl_min(vd);
3344 thismax = vdev_dtl_max(vd);
3347 mutex_exit(&vd->vdev_dtl_lock);
3349 for (int c = 0; c < vd->vdev_children; c++) {
3350 vdev_t *cvd = vd->vdev_child[c];
3351 uint64_t cmin, cmax;
3353 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
3354 thismin = MIN(thismin, cmin);
3355 thismax = MAX(thismax, cmax);
3361 if (needed && minp) {
3369 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3370 * will contain either the checkpoint spacemap object or zero if none exists.
3371 * All other errors are returned to the caller.
3374 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj)
3376 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
3378 if (vd->vdev_top_zap == 0) {
3383 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
3384 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj);
3385 if (error == ENOENT) {
3394 vdev_load(vdev_t *vd)
3396 int children = vd->vdev_children;
3401 * It's only worthwhile to use the taskq for the root vdev, because the
3402 * slow part is metaslab_init, and that only happens for top-level
3405 if (vd->vdev_ops == &vdev_root_ops && vd->vdev_children > 0) {
3406 tq = taskq_create("vdev_load", children, minclsyspri,
3407 children, children, TASKQ_PREPOPULATE);
3411 * Recursively load all children.
3413 for (int c = 0; c < vd->vdev_children; c++) {
3414 vdev_t *cvd = vd->vdev_child[c];
3416 if (tq == NULL || vdev_uses_zvols(cvd)) {
3417 cvd->vdev_load_error = vdev_load(cvd);
3419 VERIFY(taskq_dispatch(tq, vdev_load_child,
3420 cvd, TQ_SLEEP) != TASKQID_INVALID);
3429 for (int c = 0; c < vd->vdev_children; c++) {
3430 int error = vd->vdev_child[c]->vdev_load_error;
3436 vdev_set_deflate_ratio(vd);
3439 * On spa_load path, grab the allocation bias from our zap
3441 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3442 spa_t *spa = vd->vdev_spa;
3445 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
3446 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
3449 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
3450 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
3451 } else if (error != ENOENT) {
3452 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3453 VDEV_AUX_CORRUPT_DATA);
3454 vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) "
3455 "failed [error=%d]", vd->vdev_top_zap, error);
3461 * Load any rebuild state from the top-level vdev zap.
3463 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3464 error = vdev_rebuild_load(vd);
3465 if (error && error != ENOTSUP) {
3466 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3467 VDEV_AUX_CORRUPT_DATA);
3468 vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load "
3469 "failed [error=%d]", error);
3475 * If this is a top-level vdev, initialize its metaslabs.
3477 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
3478 vdev_metaslab_group_create(vd);
3480 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
3481 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3482 VDEV_AUX_CORRUPT_DATA);
3483 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
3484 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
3485 (u_longlong_t)vd->vdev_asize);
3486 return (SET_ERROR(ENXIO));
3489 error = vdev_metaslab_init(vd, 0);
3491 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
3492 "[error=%d]", error);
3493 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3494 VDEV_AUX_CORRUPT_DATA);
3498 uint64_t checkpoint_sm_obj;
3499 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj);
3500 if (error == 0 && checkpoint_sm_obj != 0) {
3501 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3502 ASSERT(vd->vdev_asize != 0);
3503 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
3505 error = space_map_open(&vd->vdev_checkpoint_sm,
3506 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
3509 vdev_dbgmsg(vd, "vdev_load: space_map_open "
3510 "failed for checkpoint spacemap (obj %llu) "
3512 (u_longlong_t)checkpoint_sm_obj, error);
3515 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
3518 * Since the checkpoint_sm contains free entries
3519 * exclusively we can use space_map_allocated() to
3520 * indicate the cumulative checkpointed space that
3523 vd->vdev_stat.vs_checkpoint_space =
3524 -space_map_allocated(vd->vdev_checkpoint_sm);
3525 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
3526 vd->vdev_stat.vs_checkpoint_space;
3527 } else if (error != 0) {
3528 vdev_dbgmsg(vd, "vdev_load: failed to retrieve "
3529 "checkpoint space map object from vdev ZAP "
3530 "[error=%d]", error);
3536 * If this is a leaf vdev, load its DTL.
3538 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
3539 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3540 VDEV_AUX_CORRUPT_DATA);
3541 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
3542 "[error=%d]", error);
3546 uint64_t obsolete_sm_object;
3547 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object);
3548 if (error == 0 && obsolete_sm_object != 0) {
3549 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3550 ASSERT(vd->vdev_asize != 0);
3551 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
3553 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
3554 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
3555 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3556 VDEV_AUX_CORRUPT_DATA);
3557 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
3558 "obsolete spacemap (obj %llu) [error=%d]",
3559 (u_longlong_t)obsolete_sm_object, error);
3562 } else if (error != 0) {
3563 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete "
3564 "space map object from vdev ZAP [error=%d]", error);
3572 * The special vdev case is used for hot spares and l2cache devices. Its
3573 * sole purpose it to set the vdev state for the associated vdev. To do this,
3574 * we make sure that we can open the underlying device, then try to read the
3575 * label, and make sure that the label is sane and that it hasn't been
3576 * repurposed to another pool.
3579 vdev_validate_aux(vdev_t *vd)
3582 uint64_t guid, version;
3585 if (!vdev_readable(vd))
3588 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3589 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3590 VDEV_AUX_CORRUPT_DATA);
3594 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3595 !SPA_VERSION_IS_SUPPORTED(version) ||
3596 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3597 guid != vd->vdev_guid ||
3598 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3599 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3600 VDEV_AUX_CORRUPT_DATA);
3606 * We don't actually check the pool state here. If it's in fact in
3607 * use by another pool, we update this fact on the fly when requested.
3614 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx)
3616 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3618 if (vd->vdev_top_zap == 0)
3621 uint64_t object = 0;
3622 int err = zap_lookup(mos, vd->vdev_top_zap,
3623 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object);
3628 VERIFY0(dmu_object_free(mos, object, tx));
3629 VERIFY0(zap_remove(mos, vd->vdev_top_zap,
3630 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx));
3634 * Free the objects used to store this vdev's spacemaps, and the array
3635 * that points to them.
3638 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3640 if (vd->vdev_ms_array == 0)
3643 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3644 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3645 size_t array_bytes = array_count * sizeof (uint64_t);
3646 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3647 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3648 array_bytes, smobj_array, 0));
3650 for (uint64_t i = 0; i < array_count; i++) {
3651 uint64_t smobj = smobj_array[i];
3655 space_map_free_obj(mos, smobj, tx);
3658 kmem_free(smobj_array, array_bytes);
3659 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3660 vdev_destroy_ms_flush_data(vd, tx);
3661 vd->vdev_ms_array = 0;
3665 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3667 spa_t *spa = vd->vdev_spa;
3669 ASSERT(vd->vdev_islog);
3670 ASSERT(vd == vd->vdev_top);
3671 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3673 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3675 vdev_destroy_spacemaps(vd, tx);
3676 if (vd->vdev_top_zap != 0) {
3677 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3678 vd->vdev_top_zap = 0;
3685 vdev_sync_done(vdev_t *vd, uint64_t txg)
3688 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3690 ASSERT(vdev_is_concrete(vd));
3692 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3694 metaslab_sync_done(msp, txg);
3697 metaslab_sync_reassess(vd->vdev_mg);
3698 if (vd->vdev_log_mg != NULL)
3699 metaslab_sync_reassess(vd->vdev_log_mg);
3704 vdev_sync(vdev_t *vd, uint64_t txg)
3706 spa_t *spa = vd->vdev_spa;
3710 ASSERT3U(txg, ==, spa->spa_syncing_txg);
3711 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3712 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3713 ASSERT(vd->vdev_removing ||
3714 vd->vdev_ops == &vdev_indirect_ops);
3716 vdev_indirect_sync_obsolete(vd, tx);
3719 * If the vdev is indirect, it can't have dirty
3720 * metaslabs or DTLs.
3722 if (vd->vdev_ops == &vdev_indirect_ops) {
3723 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3724 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3730 ASSERT(vdev_is_concrete(vd));
3732 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3733 !vd->vdev_removing) {
3734 ASSERT(vd == vd->vdev_top);
3735 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3736 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3737 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3738 ASSERT(vd->vdev_ms_array != 0);
3739 vdev_config_dirty(vd);
3742 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3743 metaslab_sync(msp, txg);
3744 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3747 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3748 vdev_dtl_sync(lvd, txg);
3751 * If this is an empty log device being removed, destroy the
3752 * metadata associated with it.
3754 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3755 vdev_remove_empty_log(vd, txg);
3757 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3762 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3764 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3768 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3769 * not be opened, and no I/O is attempted.
3772 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3776 spa_vdev_state_enter(spa, SCL_NONE);
3778 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3779 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3781 if (!vd->vdev_ops->vdev_op_leaf)
3782 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3787 * If user did a 'zpool offline -f' then make the fault persist across
3790 if (aux == VDEV_AUX_EXTERNAL_PERSIST) {
3792 * There are two kinds of forced faults: temporary and
3793 * persistent. Temporary faults go away at pool import, while
3794 * persistent faults stay set. Both types of faults can be
3795 * cleared with a zpool clear.
3797 * We tell if a vdev is persistently faulted by looking at the
3798 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3799 * import then it's a persistent fault. Otherwise, it's
3800 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3801 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3802 * tells vdev_config_generate() (which gets run later) to set
3803 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3805 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
3806 vd->vdev_tmpoffline = B_FALSE;
3807 aux = VDEV_AUX_EXTERNAL;
3809 vd->vdev_tmpoffline = B_TRUE;
3813 * We don't directly use the aux state here, but if we do a
3814 * vdev_reopen(), we need this value to be present to remember why we
3817 vd->vdev_label_aux = aux;
3820 * Faulted state takes precedence over degraded.
3822 vd->vdev_delayed_close = B_FALSE;
3823 vd->vdev_faulted = 1ULL;
3824 vd->vdev_degraded = 0ULL;
3825 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3828 * If this device has the only valid copy of the data, then
3829 * back off and simply mark the vdev as degraded instead.
3831 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3832 vd->vdev_degraded = 1ULL;
3833 vd->vdev_faulted = 0ULL;
3836 * If we reopen the device and it's not dead, only then do we
3841 if (vdev_readable(vd))
3842 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3845 return (spa_vdev_state_exit(spa, vd, 0));
3849 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3850 * user that something is wrong. The vdev continues to operate as normal as far
3851 * as I/O is concerned.
3854 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3858 spa_vdev_state_enter(spa, SCL_NONE);
3860 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3861 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3863 if (!vd->vdev_ops->vdev_op_leaf)
3864 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3867 * If the vdev is already faulted, then don't do anything.
3869 if (vd->vdev_faulted || vd->vdev_degraded)
3870 return (spa_vdev_state_exit(spa, NULL, 0));
3872 vd->vdev_degraded = 1ULL;
3873 if (!vdev_is_dead(vd))
3874 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3877 return (spa_vdev_state_exit(spa, vd, 0));
3881 * Online the given vdev.
3883 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3884 * spare device should be detached when the device finishes resilvering.
3885 * Second, the online should be treated like a 'test' online case, so no FMA
3886 * events are generated if the device fails to open.
3889 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3891 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3892 boolean_t wasoffline;
3893 vdev_state_t oldstate;
3895 spa_vdev_state_enter(spa, SCL_NONE);
3897 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3898 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3900 if (!vd->vdev_ops->vdev_op_leaf)
3901 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3903 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3904 oldstate = vd->vdev_state;
3907 vd->vdev_offline = B_FALSE;
3908 vd->vdev_tmpoffline = B_FALSE;
3909 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3910 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3912 /* XXX - L2ARC 1.0 does not support expansion */
3913 if (!vd->vdev_aux) {
3914 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3915 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
3916 spa->spa_autoexpand);
3917 vd->vdev_expansion_time = gethrestime_sec();
3921 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3923 if (!vd->vdev_aux) {
3924 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3925 pvd->vdev_expanding = B_FALSE;
3929 *newstate = vd->vdev_state;
3930 if ((flags & ZFS_ONLINE_UNSPARE) &&
3931 !vdev_is_dead(vd) && vd->vdev_parent &&
3932 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3933 vd->vdev_parent->vdev_child[0] == vd)
3934 vd->vdev_unspare = B_TRUE;
3936 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3938 /* XXX - L2ARC 1.0 does not support expansion */
3940 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3941 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3944 /* Restart initializing if necessary */
3945 mutex_enter(&vd->vdev_initialize_lock);
3946 if (vdev_writeable(vd) &&
3947 vd->vdev_initialize_thread == NULL &&
3948 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3949 (void) vdev_initialize(vd);
3951 mutex_exit(&vd->vdev_initialize_lock);
3954 * Restart trimming if necessary. We do not restart trimming for cache
3955 * devices here. This is triggered by l2arc_rebuild_vdev()
3956 * asynchronously for the whole device or in l2arc_evict() as it evicts
3957 * space for upcoming writes.
3959 mutex_enter(&vd->vdev_trim_lock);
3960 if (vdev_writeable(vd) && !vd->vdev_isl2cache &&
3961 vd->vdev_trim_thread == NULL &&
3962 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) {
3963 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial,
3964 vd->vdev_trim_secure);
3966 mutex_exit(&vd->vdev_trim_lock);
3969 (oldstate < VDEV_STATE_DEGRADED &&
3970 vd->vdev_state >= VDEV_STATE_DEGRADED))
3971 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3973 return (spa_vdev_state_exit(spa, vd, 0));
3977 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3981 uint64_t generation;
3982 metaslab_group_t *mg;
3985 spa_vdev_state_enter(spa, SCL_ALLOC);
3987 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3988 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3990 if (!vd->vdev_ops->vdev_op_leaf)
3991 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3993 if (vd->vdev_ops == &vdev_draid_spare_ops)
3994 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3998 generation = spa->spa_config_generation + 1;
4001 * If the device isn't already offline, try to offline it.
4003 if (!vd->vdev_offline) {
4005 * If this device has the only valid copy of some data,
4006 * don't allow it to be offlined. Log devices are always
4009 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
4010 vdev_dtl_required(vd))
4011 return (spa_vdev_state_exit(spa, NULL,
4015 * If the top-level is a slog and it has had allocations
4016 * then proceed. We check that the vdev's metaslab group
4017 * is not NULL since it's possible that we may have just
4018 * added this vdev but not yet initialized its metaslabs.
4020 if (tvd->vdev_islog && mg != NULL) {
4022 * Prevent any future allocations.
4024 ASSERT3P(tvd->vdev_log_mg, ==, NULL);
4025 metaslab_group_passivate(mg);
4026 (void) spa_vdev_state_exit(spa, vd, 0);
4028 error = spa_reset_logs(spa);
4031 * If the log device was successfully reset but has
4032 * checkpointed data, do not offline it.
4035 tvd->vdev_checkpoint_sm != NULL) {
4036 ASSERT3U(space_map_allocated(
4037 tvd->vdev_checkpoint_sm), !=, 0);
4038 error = ZFS_ERR_CHECKPOINT_EXISTS;
4041 spa_vdev_state_enter(spa, SCL_ALLOC);
4044 * Check to see if the config has changed.
4046 if (error || generation != spa->spa_config_generation) {
4047 metaslab_group_activate(mg);
4049 return (spa_vdev_state_exit(spa,
4051 (void) spa_vdev_state_exit(spa, vd, 0);
4054 ASSERT0(tvd->vdev_stat.vs_alloc);
4058 * Offline this device and reopen its top-level vdev.
4059 * If the top-level vdev is a log device then just offline
4060 * it. Otherwise, if this action results in the top-level
4061 * vdev becoming unusable, undo it and fail the request.
4063 vd->vdev_offline = B_TRUE;
4066 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
4067 vdev_is_dead(tvd)) {
4068 vd->vdev_offline = B_FALSE;
4070 return (spa_vdev_state_exit(spa, NULL,
4075 * Add the device back into the metaslab rotor so that
4076 * once we online the device it's open for business.
4078 if (tvd->vdev_islog && mg != NULL)
4079 metaslab_group_activate(mg);
4082 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
4084 return (spa_vdev_state_exit(spa, vd, 0));
4088 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
4092 mutex_enter(&spa->spa_vdev_top_lock);
4093 error = vdev_offline_locked(spa, guid, flags);
4094 mutex_exit(&spa->spa_vdev_top_lock);
4100 * Clear the error counts associated with this vdev. Unlike vdev_online() and
4101 * vdev_offline(), we assume the spa config is locked. We also clear all
4102 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
4105 vdev_clear(spa_t *spa, vdev_t *vd)
4107 vdev_t *rvd = spa->spa_root_vdev;
4109 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
4114 vd->vdev_stat.vs_read_errors = 0;
4115 vd->vdev_stat.vs_write_errors = 0;
4116 vd->vdev_stat.vs_checksum_errors = 0;
4117 vd->vdev_stat.vs_slow_ios = 0;
4119 for (int c = 0; c < vd->vdev_children; c++)
4120 vdev_clear(spa, vd->vdev_child[c]);
4123 * It makes no sense to "clear" an indirect vdev.
4125 if (!vdev_is_concrete(vd))
4129 * If we're in the FAULTED state or have experienced failed I/O, then
4130 * clear the persistent state and attempt to reopen the device. We
4131 * also mark the vdev config dirty, so that the new faulted state is
4132 * written out to disk.
4134 if (vd->vdev_faulted || vd->vdev_degraded ||
4135 !vdev_readable(vd) || !vdev_writeable(vd)) {
4137 * When reopening in response to a clear event, it may be due to
4138 * a fmadm repair request. In this case, if the device is
4139 * still broken, we want to still post the ereport again.
4141 vd->vdev_forcefault = B_TRUE;
4143 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
4144 vd->vdev_cant_read = B_FALSE;
4145 vd->vdev_cant_write = B_FALSE;
4146 vd->vdev_stat.vs_aux = 0;
4148 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
4150 vd->vdev_forcefault = B_FALSE;
4152 if (vd != rvd && vdev_writeable(vd->vdev_top))
4153 vdev_state_dirty(vd->vdev_top);
4155 /* If a resilver isn't required, check if vdevs can be culled */
4156 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) &&
4157 !dsl_scan_resilvering(spa->spa_dsl_pool) &&
4158 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool))
4159 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4161 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
4165 * When clearing a FMA-diagnosed fault, we always want to
4166 * unspare the device, as we assume that the original spare was
4167 * done in response to the FMA fault.
4169 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
4170 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4171 vd->vdev_parent->vdev_child[0] == vd)
4172 vd->vdev_unspare = B_TRUE;
4174 /* Clear recent error events cache (i.e. duplicate events tracking) */
4175 zfs_ereport_clear(spa, vd);
4179 vdev_is_dead(vdev_t *vd)
4182 * Holes and missing devices are always considered "dead".
4183 * This simplifies the code since we don't have to check for
4184 * these types of devices in the various code paths.
4185 * Instead we rely on the fact that we skip over dead devices
4186 * before issuing I/O to them.
4188 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
4189 vd->vdev_ops == &vdev_hole_ops ||
4190 vd->vdev_ops == &vdev_missing_ops);
4194 vdev_readable(vdev_t *vd)
4196 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
4200 vdev_writeable(vdev_t *vd)
4202 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
4203 vdev_is_concrete(vd));
4207 vdev_allocatable(vdev_t *vd)
4209 uint64_t state = vd->vdev_state;
4212 * We currently allow allocations from vdevs which may be in the
4213 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
4214 * fails to reopen then we'll catch it later when we're holding
4215 * the proper locks. Note that we have to get the vdev state
4216 * in a local variable because although it changes atomically,
4217 * we're asking two separate questions about it.
4219 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
4220 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
4221 vd->vdev_mg->mg_initialized);
4225 vdev_accessible(vdev_t *vd, zio_t *zio)
4227 ASSERT(zio->io_vd == vd);
4229 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
4232 if (zio->io_type == ZIO_TYPE_READ)
4233 return (!vd->vdev_cant_read);
4235 if (zio->io_type == ZIO_TYPE_WRITE)
4236 return (!vd->vdev_cant_write);
4242 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs)
4245 * Exclude the dRAID spare when aggregating to avoid double counting
4246 * the ops and bytes. These IOs are counted by the physical leaves.
4248 if (cvd->vdev_ops == &vdev_draid_spare_ops)
4251 for (int t = 0; t < VS_ZIO_TYPES; t++) {
4252 vs->vs_ops[t] += cvs->vs_ops[t];
4253 vs->vs_bytes[t] += cvs->vs_bytes[t];
4256 cvs->vs_scan_removing = cvd->vdev_removing;
4260 * Get extended stats
4263 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx)
4266 for (t = 0; t < ZIO_TYPES; t++) {
4267 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++)
4268 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b];
4270 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) {
4271 vsx->vsx_total_histo[t][b] +=
4272 cvsx->vsx_total_histo[t][b];
4276 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
4277 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) {
4278 vsx->vsx_queue_histo[t][b] +=
4279 cvsx->vsx_queue_histo[t][b];
4281 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t];
4282 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t];
4284 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++)
4285 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b];
4287 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++)
4288 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b];
4294 vdev_is_spacemap_addressable(vdev_t *vd)
4296 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
4300 * If double-word space map entries are not enabled we assume
4301 * 47 bits of the space map entry are dedicated to the entry's
4302 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4303 * to calculate the maximum address that can be described by a
4304 * space map entry for the given device.
4306 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
4308 if (shift >= 63) /* detect potential overflow */
4311 return (vd->vdev_asize < (1ULL << shift));
4315 * Get statistics for the given vdev.
4318 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4322 * If we're getting stats on the root vdev, aggregate the I/O counts
4323 * over all top-level vdevs (i.e. the direct children of the root).
4325 if (!vd->vdev_ops->vdev_op_leaf) {
4327 memset(vs->vs_ops, 0, sizeof (vs->vs_ops));
4328 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes));
4331 memset(vsx, 0, sizeof (*vsx));
4333 for (int c = 0; c < vd->vdev_children; c++) {
4334 vdev_t *cvd = vd->vdev_child[c];
4335 vdev_stat_t *cvs = &cvd->vdev_stat;
4336 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex;
4338 vdev_get_stats_ex_impl(cvd, cvs, cvsx);
4340 vdev_get_child_stat(cvd, vs, cvs);
4342 vdev_get_child_stat_ex(cvd, vsx, cvsx);
4346 * We're a leaf. Just copy our ZIO active queue stats in. The
4347 * other leaf stats are updated in vdev_stat_update().
4352 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex));
4354 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) {
4355 vsx->vsx_active_queue[t] =
4356 vd->vdev_queue.vq_class[t].vqc_active;
4357 vsx->vsx_pend_queue[t] = avl_numnodes(
4358 &vd->vdev_queue.vq_class[t].vqc_queued_tree);
4364 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4366 vdev_t *tvd = vd->vdev_top;
4367 mutex_enter(&vd->vdev_stat_lock);
4369 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
4370 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
4371 vs->vs_state = vd->vdev_state;
4372 vs->vs_rsize = vdev_get_min_asize(vd);
4374 if (vd->vdev_ops->vdev_op_leaf) {
4375 vs->vs_rsize += VDEV_LABEL_START_SIZE +
4376 VDEV_LABEL_END_SIZE;
4378 * Report initializing progress. Since we don't
4379 * have the initializing locks held, this is only
4380 * an estimate (although a fairly accurate one).
4382 vs->vs_initialize_bytes_done =
4383 vd->vdev_initialize_bytes_done;
4384 vs->vs_initialize_bytes_est =
4385 vd->vdev_initialize_bytes_est;
4386 vs->vs_initialize_state = vd->vdev_initialize_state;
4387 vs->vs_initialize_action_time =
4388 vd->vdev_initialize_action_time;
4391 * Report manual TRIM progress. Since we don't have
4392 * the manual TRIM locks held, this is only an
4393 * estimate (although fairly accurate one).
4395 vs->vs_trim_notsup = !vd->vdev_has_trim;
4396 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done;
4397 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est;
4398 vs->vs_trim_state = vd->vdev_trim_state;
4399 vs->vs_trim_action_time = vd->vdev_trim_action_time;
4401 /* Set when there is a deferred resilver. */
4402 vs->vs_resilver_deferred = vd->vdev_resilver_deferred;
4406 * Report expandable space on top-level, non-auxiliary devices
4407 * only. The expandable space is reported in terms of metaslab
4408 * sized units since that determines how much space the pool
4411 if (vd->vdev_aux == NULL && tvd != NULL) {
4412 vs->vs_esize = P2ALIGN(
4413 vd->vdev_max_asize - vd->vdev_asize,
4414 1ULL << tvd->vdev_ms_shift);
4417 vs->vs_configured_ashift = vd->vdev_top != NULL
4418 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
4419 vs->vs_logical_ashift = vd->vdev_logical_ashift;
4420 vs->vs_physical_ashift = vd->vdev_physical_ashift;
4423 * Report fragmentation and rebuild progress for top-level,
4424 * non-auxiliary, concrete devices.
4426 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
4427 vdev_is_concrete(vd)) {
4429 * The vdev fragmentation rating doesn't take into
4430 * account the embedded slog metaslab (vdev_log_mg).
4431 * Since it's only one metaslab, it would have a tiny
4432 * impact on the overall fragmentation.
4434 vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
4435 vd->vdev_mg->mg_fragmentation : 0;
4439 vdev_get_stats_ex_impl(vd, vs, vsx);
4440 mutex_exit(&vd->vdev_stat_lock);
4444 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
4446 return (vdev_get_stats_ex(vd, vs, NULL));
4450 vdev_clear_stats(vdev_t *vd)
4452 mutex_enter(&vd->vdev_stat_lock);
4453 vd->vdev_stat.vs_space = 0;
4454 vd->vdev_stat.vs_dspace = 0;
4455 vd->vdev_stat.vs_alloc = 0;
4456 mutex_exit(&vd->vdev_stat_lock);
4460 vdev_scan_stat_init(vdev_t *vd)
4462 vdev_stat_t *vs = &vd->vdev_stat;
4464 for (int c = 0; c < vd->vdev_children; c++)
4465 vdev_scan_stat_init(vd->vdev_child[c]);
4467 mutex_enter(&vd->vdev_stat_lock);
4468 vs->vs_scan_processed = 0;
4469 mutex_exit(&vd->vdev_stat_lock);
4473 vdev_stat_update(zio_t *zio, uint64_t psize)
4475 spa_t *spa = zio->io_spa;
4476 vdev_t *rvd = spa->spa_root_vdev;
4477 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
4479 uint64_t txg = zio->io_txg;
4480 vdev_stat_t *vs = &vd->vdev_stat;
4481 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex;
4482 zio_type_t type = zio->io_type;
4483 int flags = zio->io_flags;
4486 * If this i/o is a gang leader, it didn't do any actual work.
4488 if (zio->io_gang_tree)
4491 if (zio->io_error == 0) {
4493 * If this is a root i/o, don't count it -- we've already
4494 * counted the top-level vdevs, and vdev_get_stats() will
4495 * aggregate them when asked. This reduces contention on
4496 * the root vdev_stat_lock and implicitly handles blocks
4497 * that compress away to holes, for which there is no i/o.
4498 * (Holes never create vdev children, so all the counters
4499 * remain zero, which is what we want.)
4501 * Note: this only applies to successful i/o (io_error == 0)
4502 * because unlike i/o counts, errors are not additive.
4503 * When reading a ditto block, for example, failure of
4504 * one top-level vdev does not imply a root-level error.
4509 ASSERT(vd == zio->io_vd);
4511 if (flags & ZIO_FLAG_IO_BYPASS)
4514 mutex_enter(&vd->vdev_stat_lock);
4516 if (flags & ZIO_FLAG_IO_REPAIR) {
4518 * Repair is the result of a resilver issued by the
4519 * scan thread (spa_sync).
4521 if (flags & ZIO_FLAG_SCAN_THREAD) {
4522 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
4523 dsl_scan_phys_t *scn_phys = &scn->scn_phys;
4524 uint64_t *processed = &scn_phys->scn_processed;
4526 if (vd->vdev_ops->vdev_op_leaf)
4527 atomic_add_64(processed, psize);
4528 vs->vs_scan_processed += psize;
4532 * Repair is the result of a rebuild issued by the
4533 * rebuild thread (vdev_rebuild_thread). To avoid
4534 * double counting repaired bytes the virtual dRAID
4535 * spare vdev is excluded from the processed bytes.
4537 if (zio->io_priority == ZIO_PRIORITY_REBUILD) {
4538 vdev_t *tvd = vd->vdev_top;
4539 vdev_rebuild_t *vr = &tvd->vdev_rebuild_config;
4540 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
4541 uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt;
4543 if (vd->vdev_ops->vdev_op_leaf &&
4544 vd->vdev_ops != &vdev_draid_spare_ops) {
4545 atomic_add_64(rebuilt, psize);
4547 vs->vs_rebuild_processed += psize;
4550 if (flags & ZIO_FLAG_SELF_HEAL)
4551 vs->vs_self_healed += psize;
4555 * The bytes/ops/histograms are recorded at the leaf level and
4556 * aggregated into the higher level vdevs in vdev_get_stats().
4558 if (vd->vdev_ops->vdev_op_leaf &&
4559 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) {
4560 zio_type_t vs_type = type;
4561 zio_priority_t priority = zio->io_priority;
4564 * TRIM ops and bytes are reported to user space as
4565 * ZIO_TYPE_IOCTL. This is done to preserve the
4566 * vdev_stat_t structure layout for user space.
4568 if (type == ZIO_TYPE_TRIM)
4569 vs_type = ZIO_TYPE_IOCTL;
4572 * Solely for the purposes of 'zpool iostat -lqrw'
4573 * reporting use the priority to catagorize the IO.
4574 * Only the following are reported to user space:
4576 * ZIO_PRIORITY_SYNC_READ,
4577 * ZIO_PRIORITY_SYNC_WRITE,
4578 * ZIO_PRIORITY_ASYNC_READ,
4579 * ZIO_PRIORITY_ASYNC_WRITE,
4580 * ZIO_PRIORITY_SCRUB,
4581 * ZIO_PRIORITY_TRIM.
4583 if (priority == ZIO_PRIORITY_REBUILD) {
4584 priority = ((type == ZIO_TYPE_WRITE) ?
4585 ZIO_PRIORITY_ASYNC_WRITE :
4586 ZIO_PRIORITY_SCRUB);
4587 } else if (priority == ZIO_PRIORITY_INITIALIZING) {
4588 ASSERT3U(type, ==, ZIO_TYPE_WRITE);
4589 priority = ZIO_PRIORITY_ASYNC_WRITE;
4590 } else if (priority == ZIO_PRIORITY_REMOVAL) {
4591 priority = ((type == ZIO_TYPE_WRITE) ?
4592 ZIO_PRIORITY_ASYNC_WRITE :
4593 ZIO_PRIORITY_ASYNC_READ);
4596 vs->vs_ops[vs_type]++;
4597 vs->vs_bytes[vs_type] += psize;
4599 if (flags & ZIO_FLAG_DELEGATED) {
4600 vsx->vsx_agg_histo[priority]
4601 [RQ_HISTO(zio->io_size)]++;
4603 vsx->vsx_ind_histo[priority]
4604 [RQ_HISTO(zio->io_size)]++;
4607 if (zio->io_delta && zio->io_delay) {
4608 vsx->vsx_queue_histo[priority]
4609 [L_HISTO(zio->io_delta - zio->io_delay)]++;
4610 vsx->vsx_disk_histo[type]
4611 [L_HISTO(zio->io_delay)]++;
4612 vsx->vsx_total_histo[type]
4613 [L_HISTO(zio->io_delta)]++;
4617 mutex_exit(&vd->vdev_stat_lock);
4621 if (flags & ZIO_FLAG_SPECULATIVE)
4625 * If this is an I/O error that is going to be retried, then ignore the
4626 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4627 * hard errors, when in reality they can happen for any number of
4628 * innocuous reasons (bus resets, MPxIO link failure, etc).
4630 if (zio->io_error == EIO &&
4631 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
4635 * Intent logs writes won't propagate their error to the root
4636 * I/O so don't mark these types of failures as pool-level
4639 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
4642 if (type == ZIO_TYPE_WRITE && txg != 0 &&
4643 (!(flags & ZIO_FLAG_IO_REPAIR) ||
4644 (flags & ZIO_FLAG_SCAN_THREAD) ||
4645 spa->spa_claiming)) {
4647 * This is either a normal write (not a repair), or it's
4648 * a repair induced by the scrub thread, or it's a repair
4649 * made by zil_claim() during spa_load() in the first txg.
4650 * In the normal case, we commit the DTL change in the same
4651 * txg as the block was born. In the scrub-induced repair
4652 * case, we know that scrubs run in first-pass syncing context,
4653 * so we commit the DTL change in spa_syncing_txg(spa).
4654 * In the zil_claim() case, we commit in spa_first_txg(spa).
4656 * We currently do not make DTL entries for failed spontaneous
4657 * self-healing writes triggered by normal (non-scrubbing)
4658 * reads, because we have no transactional context in which to
4659 * do so -- and it's not clear that it'd be desirable anyway.
4661 if (vd->vdev_ops->vdev_op_leaf) {
4662 uint64_t commit_txg = txg;
4663 if (flags & ZIO_FLAG_SCAN_THREAD) {
4664 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4665 ASSERT(spa_sync_pass(spa) == 1);
4666 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
4667 commit_txg = spa_syncing_txg(spa);
4668 } else if (spa->spa_claiming) {
4669 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4670 commit_txg = spa_first_txg(spa);
4672 ASSERT(commit_txg >= spa_syncing_txg(spa));
4673 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
4675 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
4676 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
4677 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
4680 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
4685 vdev_deflated_space(vdev_t *vd, int64_t space)
4687 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
4688 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
4690 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
4694 * Update the in-core space usage stats for this vdev, its metaslab class,
4695 * and the root vdev.
4698 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
4699 int64_t space_delta)
4701 int64_t dspace_delta;
4702 spa_t *spa = vd->vdev_spa;
4703 vdev_t *rvd = spa->spa_root_vdev;
4705 ASSERT(vd == vd->vdev_top);
4708 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4709 * factor. We must calculate this here and not at the root vdev
4710 * because the root vdev's psize-to-asize is simply the max of its
4711 * children's, thus not accurate enough for us.
4713 dspace_delta = vdev_deflated_space(vd, space_delta);
4715 mutex_enter(&vd->vdev_stat_lock);
4716 /* ensure we won't underflow */
4717 if (alloc_delta < 0) {
4718 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta);
4721 vd->vdev_stat.vs_alloc += alloc_delta;
4722 vd->vdev_stat.vs_space += space_delta;
4723 vd->vdev_stat.vs_dspace += dspace_delta;
4724 mutex_exit(&vd->vdev_stat_lock);
4726 /* every class but log contributes to root space stats */
4727 if (vd->vdev_mg != NULL && !vd->vdev_islog) {
4728 ASSERT(!vd->vdev_isl2cache);
4729 mutex_enter(&rvd->vdev_stat_lock);
4730 rvd->vdev_stat.vs_alloc += alloc_delta;
4731 rvd->vdev_stat.vs_space += space_delta;
4732 rvd->vdev_stat.vs_dspace += dspace_delta;
4733 mutex_exit(&rvd->vdev_stat_lock);
4735 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4739 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4740 * so that it will be written out next time the vdev configuration is synced.
4741 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4744 vdev_config_dirty(vdev_t *vd)
4746 spa_t *spa = vd->vdev_spa;
4747 vdev_t *rvd = spa->spa_root_vdev;
4750 ASSERT(spa_writeable(spa));
4753 * If this is an aux vdev (as with l2cache and spare devices), then we
4754 * update the vdev config manually and set the sync flag.
4756 if (vd->vdev_aux != NULL) {
4757 spa_aux_vdev_t *sav = vd->vdev_aux;
4761 for (c = 0; c < sav->sav_count; c++) {
4762 if (sav->sav_vdevs[c] == vd)
4766 if (c == sav->sav_count) {
4768 * We're being removed. There's nothing more to do.
4770 ASSERT(sav->sav_sync == B_TRUE);
4774 sav->sav_sync = B_TRUE;
4776 if (nvlist_lookup_nvlist_array(sav->sav_config,
4777 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
4778 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
4779 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
4785 * Setting the nvlist in the middle if the array is a little
4786 * sketchy, but it will work.
4788 nvlist_free(aux[c]);
4789 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
4795 * The dirty list is protected by the SCL_CONFIG lock. The caller
4796 * must either hold SCL_CONFIG as writer, or must be the sync thread
4797 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4798 * so this is sufficient to ensure mutual exclusion.
4800 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4801 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4802 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4805 for (c = 0; c < rvd->vdev_children; c++)
4806 vdev_config_dirty(rvd->vdev_child[c]);
4808 ASSERT(vd == vd->vdev_top);
4810 if (!list_link_active(&vd->vdev_config_dirty_node) &&
4811 vdev_is_concrete(vd)) {
4812 list_insert_head(&spa->spa_config_dirty_list, vd);
4818 vdev_config_clean(vdev_t *vd)
4820 spa_t *spa = vd->vdev_spa;
4822 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4823 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4824 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4826 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
4827 list_remove(&spa->spa_config_dirty_list, vd);
4831 * Mark a top-level vdev's state as dirty, so that the next pass of
4832 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4833 * the state changes from larger config changes because they require
4834 * much less locking, and are often needed for administrative actions.
4837 vdev_state_dirty(vdev_t *vd)
4839 spa_t *spa = vd->vdev_spa;
4841 ASSERT(spa_writeable(spa));
4842 ASSERT(vd == vd->vdev_top);
4845 * The state list is protected by the SCL_STATE lock. The caller
4846 * must either hold SCL_STATE as writer, or must be the sync thread
4847 * (which holds SCL_STATE as reader). There's only one sync thread,
4848 * so this is sufficient to ensure mutual exclusion.
4850 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4851 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4852 spa_config_held(spa, SCL_STATE, RW_READER)));
4854 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4855 vdev_is_concrete(vd))
4856 list_insert_head(&spa->spa_state_dirty_list, vd);
4860 vdev_state_clean(vdev_t *vd)
4862 spa_t *spa = vd->vdev_spa;
4864 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4865 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4866 spa_config_held(spa, SCL_STATE, RW_READER)));
4868 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4869 list_remove(&spa->spa_state_dirty_list, vd);
4873 * Propagate vdev state up from children to parent.
4876 vdev_propagate_state(vdev_t *vd)
4878 spa_t *spa = vd->vdev_spa;
4879 vdev_t *rvd = spa->spa_root_vdev;
4880 int degraded = 0, faulted = 0;
4884 if (vd->vdev_children > 0) {
4885 for (int c = 0; c < vd->vdev_children; c++) {
4886 child = vd->vdev_child[c];
4889 * Don't factor holes or indirect vdevs into the
4892 if (!vdev_is_concrete(child))
4895 if (!vdev_readable(child) ||
4896 (!vdev_writeable(child) && spa_writeable(spa))) {
4898 * Root special: if there is a top-level log
4899 * device, treat the root vdev as if it were
4902 if (child->vdev_islog && vd == rvd)
4906 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4910 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4914 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4917 * Root special: if there is a top-level vdev that cannot be
4918 * opened due to corrupted metadata, then propagate the root
4919 * vdev's aux state as 'corrupt' rather than 'insufficient
4922 if (corrupted && vd == rvd &&
4923 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4924 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4925 VDEV_AUX_CORRUPT_DATA);
4928 if (vd->vdev_parent)
4929 vdev_propagate_state(vd->vdev_parent);
4933 * Set a vdev's state. If this is during an open, we don't update the parent
4934 * state, because we're in the process of opening children depth-first.
4935 * Otherwise, we propagate the change to the parent.
4937 * If this routine places a device in a faulted state, an appropriate ereport is
4941 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4943 uint64_t save_state;
4944 spa_t *spa = vd->vdev_spa;
4946 if (state == vd->vdev_state) {
4948 * Since vdev_offline() code path is already in an offline
4949 * state we can miss a statechange event to OFFLINE. Check
4950 * the previous state to catch this condition.
4952 if (vd->vdev_ops->vdev_op_leaf &&
4953 (state == VDEV_STATE_OFFLINE) &&
4954 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) {
4955 /* post an offline state change */
4956 zfs_post_state_change(spa, vd, vd->vdev_prevstate);
4958 vd->vdev_stat.vs_aux = aux;
4962 save_state = vd->vdev_state;
4964 vd->vdev_state = state;
4965 vd->vdev_stat.vs_aux = aux;
4968 * If we are setting the vdev state to anything but an open state, then
4969 * always close the underlying device unless the device has requested
4970 * a delayed close (i.e. we're about to remove or fault the device).
4971 * Otherwise, we keep accessible but invalid devices open forever.
4972 * We don't call vdev_close() itself, because that implies some extra
4973 * checks (offline, etc) that we don't want here. This is limited to
4974 * leaf devices, because otherwise closing the device will affect other
4977 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4978 vd->vdev_ops->vdev_op_leaf)
4979 vd->vdev_ops->vdev_op_close(vd);
4981 if (vd->vdev_removed &&
4982 state == VDEV_STATE_CANT_OPEN &&
4983 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4985 * If the previous state is set to VDEV_STATE_REMOVED, then this
4986 * device was previously marked removed and someone attempted to
4987 * reopen it. If this failed due to a nonexistent device, then
4988 * keep the device in the REMOVED state. We also let this be if
4989 * it is one of our special test online cases, which is only
4990 * attempting to online the device and shouldn't generate an FMA
4993 vd->vdev_state = VDEV_STATE_REMOVED;
4994 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4995 } else if (state == VDEV_STATE_REMOVED) {
4996 vd->vdev_removed = B_TRUE;
4997 } else if (state == VDEV_STATE_CANT_OPEN) {
4999 * If we fail to open a vdev during an import or recovery, we
5000 * mark it as "not available", which signifies that it was
5001 * never there to begin with. Failure to open such a device
5002 * is not considered an error.
5004 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
5005 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
5006 vd->vdev_ops->vdev_op_leaf)
5007 vd->vdev_not_present = 1;
5010 * Post the appropriate ereport. If the 'prevstate' field is
5011 * set to something other than VDEV_STATE_UNKNOWN, it indicates
5012 * that this is part of a vdev_reopen(). In this case, we don't
5013 * want to post the ereport if the device was already in the
5014 * CANT_OPEN state beforehand.
5016 * If the 'checkremove' flag is set, then this is an attempt to
5017 * online the device in response to an insertion event. If we
5018 * hit this case, then we have detected an insertion event for a
5019 * faulted or offline device that wasn't in the removed state.
5020 * In this scenario, we don't post an ereport because we are
5021 * about to replace the device, or attempt an online with
5022 * vdev_forcefault, which will generate the fault for us.
5024 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
5025 !vd->vdev_not_present && !vd->vdev_checkremove &&
5026 vd != spa->spa_root_vdev) {
5030 case VDEV_AUX_OPEN_FAILED:
5031 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
5033 case VDEV_AUX_CORRUPT_DATA:
5034 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
5036 case VDEV_AUX_NO_REPLICAS:
5037 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
5039 case VDEV_AUX_BAD_GUID_SUM:
5040 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
5042 case VDEV_AUX_TOO_SMALL:
5043 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
5045 case VDEV_AUX_BAD_LABEL:
5046 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
5048 case VDEV_AUX_BAD_ASHIFT:
5049 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT;
5052 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
5055 (void) zfs_ereport_post(class, spa, vd, NULL, NULL,
5059 /* Erase any notion of persistent removed state */
5060 vd->vdev_removed = B_FALSE;
5062 vd->vdev_removed = B_FALSE;
5066 * Notify ZED of any significant state-change on a leaf vdev.
5069 if (vd->vdev_ops->vdev_op_leaf) {
5070 /* preserve original state from a vdev_reopen() */
5071 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) &&
5072 (vd->vdev_prevstate != vd->vdev_state) &&
5073 (save_state <= VDEV_STATE_CLOSED))
5074 save_state = vd->vdev_prevstate;
5076 /* filter out state change due to initial vdev_open */
5077 if (save_state > VDEV_STATE_CLOSED)
5078 zfs_post_state_change(spa, vd, save_state);
5081 if (!isopen && vd->vdev_parent)
5082 vdev_propagate_state(vd->vdev_parent);
5086 vdev_children_are_offline(vdev_t *vd)
5088 ASSERT(!vd->vdev_ops->vdev_op_leaf);
5090 for (uint64_t i = 0; i < vd->vdev_children; i++) {
5091 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
5099 * Check the vdev configuration to ensure that it's capable of supporting
5100 * a root pool. We do not support partial configuration.
5103 vdev_is_bootable(vdev_t *vd)
5105 if (!vd->vdev_ops->vdev_op_leaf) {
5106 const char *vdev_type = vd->vdev_ops->vdev_op_type;
5108 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
5109 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
5114 for (int c = 0; c < vd->vdev_children; c++) {
5115 if (!vdev_is_bootable(vd->vdev_child[c]))
5122 vdev_is_concrete(vdev_t *vd)
5124 vdev_ops_t *ops = vd->vdev_ops;
5125 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
5126 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
5134 * Determine if a log device has valid content. If the vdev was
5135 * removed or faulted in the MOS config then we know that
5136 * the content on the log device has already been written to the pool.
5139 vdev_log_state_valid(vdev_t *vd)
5141 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
5145 for (int c = 0; c < vd->vdev_children; c++)
5146 if (vdev_log_state_valid(vd->vdev_child[c]))
5153 * Expand a vdev if possible.
5156 vdev_expand(vdev_t *vd, uint64_t txg)
5158 ASSERT(vd->vdev_top == vd);
5159 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5160 ASSERT(vdev_is_concrete(vd));
5162 vdev_set_deflate_ratio(vd);
5164 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
5165 vdev_is_concrete(vd)) {
5166 vdev_metaslab_group_create(vd);
5167 VERIFY(vdev_metaslab_init(vd, txg) == 0);
5168 vdev_config_dirty(vd);
5176 vdev_split(vdev_t *vd)
5178 vdev_t *cvd, *pvd = vd->vdev_parent;
5180 vdev_remove_child(pvd, vd);
5181 vdev_compact_children(pvd);
5183 cvd = pvd->vdev_child[0];
5184 if (pvd->vdev_children == 1) {
5185 vdev_remove_parent(cvd);
5186 cvd->vdev_splitting = B_TRUE;
5188 vdev_propagate_state(cvd);
5192 vdev_deadman(vdev_t *vd, char *tag)
5194 for (int c = 0; c < vd->vdev_children; c++) {
5195 vdev_t *cvd = vd->vdev_child[c];
5197 vdev_deadman(cvd, tag);
5200 if (vd->vdev_ops->vdev_op_leaf) {
5201 vdev_queue_t *vq = &vd->vdev_queue;
5203 mutex_enter(&vq->vq_lock);
5204 if (avl_numnodes(&vq->vq_active_tree) > 0) {
5205 spa_t *spa = vd->vdev_spa;
5209 zfs_dbgmsg("slow vdev: %s has %d active IOs",
5210 vd->vdev_path, avl_numnodes(&vq->vq_active_tree));
5213 * Look at the head of all the pending queues,
5214 * if any I/O has been outstanding for longer than
5215 * the spa_deadman_synctime invoke the deadman logic.
5217 fio = avl_first(&vq->vq_active_tree);
5218 delta = gethrtime() - fio->io_timestamp;
5219 if (delta > spa_deadman_synctime(spa))
5220 zio_deadman(fio, tag);
5222 mutex_exit(&vq->vq_lock);
5227 vdev_defer_resilver(vdev_t *vd)
5229 ASSERT(vd->vdev_ops->vdev_op_leaf);
5231 vd->vdev_resilver_deferred = B_TRUE;
5232 vd->vdev_spa->spa_resilver_deferred = B_TRUE;
5236 * Clears the resilver deferred flag on all leaf devs under vd. Returns
5237 * B_TRUE if we have devices that need to be resilvered and are available to
5238 * accept resilver I/Os.
5241 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx)
5243 boolean_t resilver_needed = B_FALSE;
5244 spa_t *spa = vd->vdev_spa;
5246 for (int c = 0; c < vd->vdev_children; c++) {
5247 vdev_t *cvd = vd->vdev_child[c];
5248 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx);
5251 if (vd == spa->spa_root_vdev &&
5252 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) {
5253 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx);
5254 vdev_config_dirty(vd);
5255 spa->spa_resilver_deferred = B_FALSE;
5256 return (resilver_needed);
5259 if (!vdev_is_concrete(vd) || vd->vdev_aux ||
5260 !vd->vdev_ops->vdev_op_leaf)
5261 return (resilver_needed);
5263 vd->vdev_resilver_deferred = B_FALSE;
5265 return (!vdev_is_dead(vd) && !vd->vdev_offline &&
5266 vdev_resilver_needed(vd, NULL, NULL));
5270 vdev_xlate_is_empty(range_seg64_t *rs)
5272 return (rs->rs_start == rs->rs_end);
5276 * Translate a logical range to the first contiguous physical range for the
5277 * specified vdev_t. This function is initially called with a leaf vdev and
5278 * will walk each parent vdev until it reaches a top-level vdev. Once the
5279 * top-level is reached the physical range is initialized and the recursive
5280 * function begins to unwind. As it unwinds it calls the parent's vdev
5281 * specific translation function to do the real conversion.
5284 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs,
5285 range_seg64_t *physical_rs, range_seg64_t *remain_rs)
5288 * Walk up the vdev tree
5290 if (vd != vd->vdev_top) {
5291 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs,
5295 * We've reached the top-level vdev, initialize the physical
5296 * range to the logical range and set an empty remaining
5297 * range then start to unwind.
5299 physical_rs->rs_start = logical_rs->rs_start;
5300 physical_rs->rs_end = logical_rs->rs_end;
5302 remain_rs->rs_start = logical_rs->rs_start;
5303 remain_rs->rs_end = logical_rs->rs_start;
5308 vdev_t *pvd = vd->vdev_parent;
5309 ASSERT3P(pvd, !=, NULL);
5310 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL);
5313 * As this recursive function unwinds, translate the logical
5314 * range into its physical and any remaining components by calling
5315 * the vdev specific translate function.
5317 range_seg64_t intermediate = { 0 };
5318 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate, remain_rs);
5320 physical_rs->rs_start = intermediate.rs_start;
5321 physical_rs->rs_end = intermediate.rs_end;
5325 vdev_xlate_walk(vdev_t *vd, const range_seg64_t *logical_rs,
5326 vdev_xlate_func_t *func, void *arg)
5328 range_seg64_t iter_rs = *logical_rs;
5329 range_seg64_t physical_rs;
5330 range_seg64_t remain_rs;
5332 while (!vdev_xlate_is_empty(&iter_rs)) {
5334 vdev_xlate(vd, &iter_rs, &physical_rs, &remain_rs);
5337 * With raidz and dRAID, it's possible that the logical range
5338 * does not live on this leaf vdev. Only when there is a non-
5339 * zero physical size call the provided function.
5341 if (!vdev_xlate_is_empty(&physical_rs))
5342 func(arg, &physical_rs);
5344 iter_rs = remain_rs;
5349 * Look at the vdev tree and determine whether any devices are currently being
5353 vdev_replace_in_progress(vdev_t *vdev)
5355 ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0);
5357 if (vdev->vdev_ops == &vdev_replacing_ops)
5361 * A 'spare' vdev indicates that we have a replace in progress, unless
5362 * it has exactly two children, and the second, the hot spare, has
5363 * finished being resilvered.
5365 if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 ||
5366 !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING)))
5369 for (int i = 0; i < vdev->vdev_children; i++) {
5370 if (vdev_replace_in_progress(vdev->vdev_child[i]))
5377 EXPORT_SYMBOL(vdev_fault);
5378 EXPORT_SYMBOL(vdev_degrade);
5379 EXPORT_SYMBOL(vdev_online);
5380 EXPORT_SYMBOL(vdev_offline);
5381 EXPORT_SYMBOL(vdev_clear);
5384 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, INT, ZMOD_RW,
5385 "Target number of metaslabs per top-level vdev");
5387 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, INT, ZMOD_RW,
5388 "Default limit for metaslab size");
5390 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, INT, ZMOD_RW,
5391 "Minimum number of metaslabs per top-level vdev");
5393 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, INT, ZMOD_RW,
5394 "Practical upper limit of total metaslabs per top-level vdev");
5396 ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW,
5397 "Rate limit slow IO (delay) events to this many per second");
5399 ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW,
5400 "Rate limit checksum events to this many checksum errors per second "
5401 "(do not set below zed threshold).");
5403 ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW,
5404 "Ignore errors during resilver/scrub");
5406 ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW,
5407 "Bypass vdev_validate()");
5409 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW,
5410 "Disable cache flushes");
5412 ZFS_MODULE_PARAM(zfs, zfs_, embedded_slog_min_ms, INT, ZMOD_RW,
5413 "Minimum number of metaslabs required to dedicate one for log blocks");
5415 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift,
5416 param_set_min_auto_ashift, param_get_ulong, ZMOD_RW,
5417 "Minimum ashift used when creating new top-level vdevs");
5419 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift,
5420 param_set_max_auto_ashift, param_get_ulong, ZMOD_RW,
5421 "Maximum ashift used when optimizing for logical -> physical sector "
5422 "size on new top-level vdevs");