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, 2020 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/uberblock_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/space_map.h>
47 #include <sys/space_reftree.h>
50 #include <sys/fs/zfs.h>
53 #include <sys/dsl_scan.h>
55 #include <sys/vdev_initialize.h>
56 #include <sys/vdev_trim.h>
58 #include <sys/zfs_ratelimit.h>
60 /* default target for number of metaslabs per top-level vdev */
61 int zfs_vdev_default_ms_count = 200;
63 /* minimum number of metaslabs per top-level vdev */
64 int zfs_vdev_min_ms_count = 16;
66 /* practical upper limit of total metaslabs per top-level vdev */
67 int zfs_vdev_ms_count_limit = 1ULL << 17;
69 /* lower limit for metaslab size (512M) */
70 int zfs_vdev_default_ms_shift = 29;
72 /* upper limit for metaslab size (16G) */
73 int zfs_vdev_max_ms_shift = 34;
75 int vdev_validate_skip = B_FALSE;
78 * Since the DTL space map of a vdev is not expected to have a lot of
79 * entries, we default its block size to 4K.
81 int zfs_vdev_dtl_sm_blksz = (1 << 12);
84 * Rate limit slow IO (delay) events to this many per second.
86 unsigned int zfs_slow_io_events_per_second = 20;
89 * Rate limit checksum events after this many checksum errors per second.
91 unsigned int zfs_checksum_events_per_second = 20;
94 * Ignore errors during scrub/resilver. Allows to work around resilver
95 * upon import when there are pool errors.
97 int zfs_scan_ignore_errors = 0;
100 * vdev-wide space maps that have lots of entries written to them at
101 * the end of each transaction can benefit from a higher I/O bandwidth
102 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
104 int zfs_vdev_standard_sm_blksz = (1 << 17);
107 * Tunable parameter for debugging or performance analysis. Setting this
108 * will cause pool corruption on power loss if a volatile out-of-order
109 * write cache is enabled.
111 int zfs_nocacheflush = 0;
113 uint64_t zfs_vdev_max_auto_ashift = ASHIFT_MAX;
114 uint64_t zfs_vdev_min_auto_ashift = ASHIFT_MIN;
118 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
124 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
127 if (vd->vdev_path != NULL) {
128 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
131 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
132 vd->vdev_ops->vdev_op_type,
133 (u_longlong_t)vd->vdev_id,
134 (u_longlong_t)vd->vdev_guid, buf);
139 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
143 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
144 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
145 vd->vdev_ops->vdev_op_type);
149 switch (vd->vdev_state) {
150 case VDEV_STATE_UNKNOWN:
151 (void) snprintf(state, sizeof (state), "unknown");
153 case VDEV_STATE_CLOSED:
154 (void) snprintf(state, sizeof (state), "closed");
156 case VDEV_STATE_OFFLINE:
157 (void) snprintf(state, sizeof (state), "offline");
159 case VDEV_STATE_REMOVED:
160 (void) snprintf(state, sizeof (state), "removed");
162 case VDEV_STATE_CANT_OPEN:
163 (void) snprintf(state, sizeof (state), "can't open");
165 case VDEV_STATE_FAULTED:
166 (void) snprintf(state, sizeof (state), "faulted");
168 case VDEV_STATE_DEGRADED:
169 (void) snprintf(state, sizeof (state), "degraded");
171 case VDEV_STATE_HEALTHY:
172 (void) snprintf(state, sizeof (state), "healthy");
175 (void) snprintf(state, sizeof (state), "<state %u>",
176 (uint_t)vd->vdev_state);
179 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
180 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
181 vd->vdev_islog ? " (log)" : "",
182 (u_longlong_t)vd->vdev_guid,
183 vd->vdev_path ? vd->vdev_path : "N/A", state);
185 for (uint64_t i = 0; i < vd->vdev_children; i++)
186 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
190 * Virtual device management.
193 static vdev_ops_t *vdev_ops_table[] = {
208 * Given a vdev type, return the appropriate ops vector.
211 vdev_getops(const char *type)
213 vdev_ops_t *ops, **opspp;
215 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
216 if (strcmp(ops->vdev_op_type, type) == 0)
224 vdev_default_xlate(vdev_t *vd, const range_seg64_t *in, range_seg64_t *res)
226 res->rs_start = in->rs_start;
227 res->rs_end = in->rs_end;
231 * Derive the enumerated allocation bias from string input.
232 * String origin is either the per-vdev zap or zpool(1M).
234 static vdev_alloc_bias_t
235 vdev_derive_alloc_bias(const char *bias)
237 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
239 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
240 alloc_bias = VDEV_BIAS_LOG;
241 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
242 alloc_bias = VDEV_BIAS_SPECIAL;
243 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
244 alloc_bias = VDEV_BIAS_DEDUP;
250 * Default asize function: return the MAX of psize with the asize of
251 * all children. This is what's used by anything other than RAID-Z.
254 vdev_default_asize(vdev_t *vd, uint64_t psize)
256 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
259 for (int c = 0; c < vd->vdev_children; c++) {
260 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
261 asize = MAX(asize, csize);
268 * Get the minimum allocatable size. We define the allocatable size as
269 * the vdev's asize rounded to the nearest metaslab. This allows us to
270 * replace or attach devices which don't have the same physical size but
271 * can still satisfy the same number of allocations.
274 vdev_get_min_asize(vdev_t *vd)
276 vdev_t *pvd = vd->vdev_parent;
279 * If our parent is NULL (inactive spare or cache) or is the root,
280 * just return our own asize.
283 return (vd->vdev_asize);
286 * The top-level vdev just returns the allocatable size rounded
287 * to the nearest metaslab.
289 if (vd == vd->vdev_top)
290 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
293 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
294 * so each child must provide at least 1/Nth of its asize.
296 if (pvd->vdev_ops == &vdev_raidz_ops)
297 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
300 return (pvd->vdev_min_asize);
304 vdev_set_min_asize(vdev_t *vd)
306 vd->vdev_min_asize = vdev_get_min_asize(vd);
308 for (int c = 0; c < vd->vdev_children; c++)
309 vdev_set_min_asize(vd->vdev_child[c]);
313 vdev_lookup_top(spa_t *spa, uint64_t vdev)
315 vdev_t *rvd = spa->spa_root_vdev;
317 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
319 if (vdev < rvd->vdev_children) {
320 ASSERT(rvd->vdev_child[vdev] != NULL);
321 return (rvd->vdev_child[vdev]);
328 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
332 if (vd->vdev_guid == guid)
335 for (int c = 0; c < vd->vdev_children; c++)
336 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
344 vdev_count_leaves_impl(vdev_t *vd)
348 if (vd->vdev_ops->vdev_op_leaf)
351 for (int c = 0; c < vd->vdev_children; c++)
352 n += vdev_count_leaves_impl(vd->vdev_child[c]);
358 vdev_count_leaves(spa_t *spa)
362 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
363 rc = vdev_count_leaves_impl(spa->spa_root_vdev);
364 spa_config_exit(spa, SCL_VDEV, FTAG);
370 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
372 size_t oldsize, newsize;
373 uint64_t id = cvd->vdev_id;
376 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
377 ASSERT(cvd->vdev_parent == NULL);
379 cvd->vdev_parent = pvd;
384 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
386 oldsize = pvd->vdev_children * sizeof (vdev_t *);
387 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
388 newsize = pvd->vdev_children * sizeof (vdev_t *);
390 newchild = kmem_alloc(newsize, KM_SLEEP);
391 if (pvd->vdev_child != NULL) {
392 bcopy(pvd->vdev_child, newchild, oldsize);
393 kmem_free(pvd->vdev_child, oldsize);
396 pvd->vdev_child = newchild;
397 pvd->vdev_child[id] = cvd;
399 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
400 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
403 * Walk up all ancestors to update guid sum.
405 for (; pvd != NULL; pvd = pvd->vdev_parent)
406 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
408 if (cvd->vdev_ops->vdev_op_leaf) {
409 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd);
410 cvd->vdev_spa->spa_leaf_list_gen++;
415 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
418 uint_t id = cvd->vdev_id;
420 ASSERT(cvd->vdev_parent == pvd);
425 ASSERT(id < pvd->vdev_children);
426 ASSERT(pvd->vdev_child[id] == cvd);
428 pvd->vdev_child[id] = NULL;
429 cvd->vdev_parent = NULL;
431 for (c = 0; c < pvd->vdev_children; c++)
432 if (pvd->vdev_child[c])
435 if (c == pvd->vdev_children) {
436 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
437 pvd->vdev_child = NULL;
438 pvd->vdev_children = 0;
441 if (cvd->vdev_ops->vdev_op_leaf) {
442 spa_t *spa = cvd->vdev_spa;
443 list_remove(&spa->spa_leaf_list, cvd);
444 spa->spa_leaf_list_gen++;
448 * Walk up all ancestors to update guid sum.
450 for (; pvd != NULL; pvd = pvd->vdev_parent)
451 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
455 * Remove any holes in the child array.
458 vdev_compact_children(vdev_t *pvd)
460 vdev_t **newchild, *cvd;
461 int oldc = pvd->vdev_children;
464 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
469 for (int c = newc = 0; c < oldc; c++)
470 if (pvd->vdev_child[c])
474 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP);
476 for (int c = newc = 0; c < oldc; c++) {
477 if ((cvd = pvd->vdev_child[c]) != NULL) {
478 newchild[newc] = cvd;
479 cvd->vdev_id = newc++;
486 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
487 pvd->vdev_child = newchild;
488 pvd->vdev_children = newc;
492 * Allocate and minimally initialize a vdev_t.
495 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
498 vdev_indirect_config_t *vic;
500 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
501 vic = &vd->vdev_indirect_config;
503 if (spa->spa_root_vdev == NULL) {
504 ASSERT(ops == &vdev_root_ops);
505 spa->spa_root_vdev = vd;
506 spa->spa_load_guid = spa_generate_guid(NULL);
509 if (guid == 0 && ops != &vdev_hole_ops) {
510 if (spa->spa_root_vdev == vd) {
512 * The root vdev's guid will also be the pool guid,
513 * which must be unique among all pools.
515 guid = spa_generate_guid(NULL);
518 * Any other vdev's guid must be unique within the pool.
520 guid = spa_generate_guid(spa);
522 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
527 vd->vdev_guid = guid;
528 vd->vdev_guid_sum = guid;
530 vd->vdev_state = VDEV_STATE_CLOSED;
531 vd->vdev_ishole = (ops == &vdev_hole_ops);
532 vic->vic_prev_indirect_vdev = UINT64_MAX;
534 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
535 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
536 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL,
540 * Initialize rate limit structs for events. We rate limit ZIO delay
541 * and checksum events so that we don't overwhelm ZED with thousands
542 * of events when a disk is acting up.
544 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second,
546 zfs_ratelimit_init(&vd->vdev_checksum_rl,
547 &zfs_checksum_events_per_second, 1);
549 list_link_init(&vd->vdev_config_dirty_node);
550 list_link_init(&vd->vdev_state_dirty_node);
551 list_link_init(&vd->vdev_initialize_node);
552 list_link_init(&vd->vdev_leaf_node);
553 list_link_init(&vd->vdev_trim_node);
554 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL);
555 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
556 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
557 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
559 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
560 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
561 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
562 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
564 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL);
565 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL);
566 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL);
567 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL);
568 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL);
569 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL);
571 mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL);
572 mutex_init(&vd->vdev_rebuild_io_lock, NULL, MUTEX_DEFAULT, NULL);
573 cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL);
574 cv_init(&vd->vdev_rebuild_io_cv, NULL, CV_DEFAULT, NULL);
576 for (int t = 0; t < DTL_TYPES; t++) {
577 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
581 txg_list_create(&vd->vdev_ms_list, spa,
582 offsetof(struct metaslab, ms_txg_node));
583 txg_list_create(&vd->vdev_dtl_list, spa,
584 offsetof(struct vdev, vdev_dtl_node));
585 vd->vdev_stat.vs_timestamp = gethrtime();
593 * Allocate a new vdev. The 'alloctype' is used to control whether we are
594 * creating a new vdev or loading an existing one - the behavior is slightly
595 * different for each case.
598 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
603 uint64_t guid = 0, islog, nparity;
605 vdev_indirect_config_t *vic;
608 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
609 boolean_t top_level = (parent && !parent->vdev_parent);
611 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
613 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
614 return (SET_ERROR(EINVAL));
616 if ((ops = vdev_getops(type)) == NULL)
617 return (SET_ERROR(EINVAL));
620 * If this is a load, get the vdev guid from the nvlist.
621 * Otherwise, vdev_alloc_common() will generate one for us.
623 if (alloctype == VDEV_ALLOC_LOAD) {
626 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
628 return (SET_ERROR(EINVAL));
630 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
631 return (SET_ERROR(EINVAL));
632 } else if (alloctype == VDEV_ALLOC_SPARE) {
633 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
634 return (SET_ERROR(EINVAL));
635 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
636 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
637 return (SET_ERROR(EINVAL));
638 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
639 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
640 return (SET_ERROR(EINVAL));
644 * The first allocated vdev must be of type 'root'.
646 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
647 return (SET_ERROR(EINVAL));
650 * Determine whether we're a log vdev.
653 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
654 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
655 return (SET_ERROR(ENOTSUP));
657 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
658 return (SET_ERROR(ENOTSUP));
661 * Set the nparity property for RAID-Z vdevs.
664 if (ops == &vdev_raidz_ops) {
665 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
667 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
668 return (SET_ERROR(EINVAL));
670 * Previous versions could only support 1 or 2 parity
674 spa_version(spa) < SPA_VERSION_RAIDZ2)
675 return (SET_ERROR(ENOTSUP));
677 spa_version(spa) < SPA_VERSION_RAIDZ3)
678 return (SET_ERROR(ENOTSUP));
681 * We require the parity to be specified for SPAs that
682 * support multiple parity levels.
684 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
685 return (SET_ERROR(EINVAL));
687 * Otherwise, we default to 1 parity device for RAID-Z.
694 ASSERT(nparity != -1ULL);
697 * If creating a top-level vdev, check for allocation classes input
699 if (top_level && alloctype == VDEV_ALLOC_ADD) {
702 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
704 alloc_bias = vdev_derive_alloc_bias(bias);
706 /* spa_vdev_add() expects feature to be enabled */
707 if (spa->spa_load_state != SPA_LOAD_CREATE &&
708 !spa_feature_is_enabled(spa,
709 SPA_FEATURE_ALLOCATION_CLASSES)) {
710 return (SET_ERROR(ENOTSUP));
715 vd = vdev_alloc_common(spa, id, guid, ops);
716 vic = &vd->vdev_indirect_config;
718 vd->vdev_islog = islog;
719 vd->vdev_nparity = nparity;
720 if (top_level && alloc_bias != VDEV_BIAS_NONE)
721 vd->vdev_alloc_bias = alloc_bias;
723 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
724 vd->vdev_path = spa_strdup(vd->vdev_path);
727 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
728 * fault on a vdev and want it to persist across imports (like with
731 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp);
732 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) {
733 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
734 vd->vdev_faulted = 1;
735 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
738 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
739 vd->vdev_devid = spa_strdup(vd->vdev_devid);
740 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
741 &vd->vdev_physpath) == 0)
742 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
744 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
745 &vd->vdev_enc_sysfs_path) == 0)
746 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path);
748 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
749 vd->vdev_fru = spa_strdup(vd->vdev_fru);
752 * Set the whole_disk property. If it's not specified, leave the value
755 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
756 &vd->vdev_wholedisk) != 0)
757 vd->vdev_wholedisk = -1ULL;
759 ASSERT0(vic->vic_mapping_object);
760 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
761 &vic->vic_mapping_object);
762 ASSERT0(vic->vic_births_object);
763 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
764 &vic->vic_births_object);
765 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
766 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
767 &vic->vic_prev_indirect_vdev);
770 * Look for the 'not present' flag. This will only be set if the device
771 * was not present at the time of import.
773 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
774 &vd->vdev_not_present);
777 * Get the alignment requirement.
779 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
782 * Retrieve the vdev creation time.
784 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
788 * If we're a top-level vdev, try to load the allocation parameters.
791 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
792 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
794 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
796 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
798 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
800 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
803 ASSERT0(vd->vdev_top_zap);
806 if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
807 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
808 alloctype == VDEV_ALLOC_ADD ||
809 alloctype == VDEV_ALLOC_SPLIT ||
810 alloctype == VDEV_ALLOC_ROOTPOOL);
811 /* Note: metaslab_group_create() is now deferred */
814 if (vd->vdev_ops->vdev_op_leaf &&
815 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
816 (void) nvlist_lookup_uint64(nv,
817 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
819 ASSERT0(vd->vdev_leaf_zap);
823 * If we're a leaf vdev, try to load the DTL object and other state.
826 if (vd->vdev_ops->vdev_op_leaf &&
827 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
828 alloctype == VDEV_ALLOC_ROOTPOOL)) {
829 if (alloctype == VDEV_ALLOC_LOAD) {
830 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
831 &vd->vdev_dtl_object);
832 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
836 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
839 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
840 &spare) == 0 && spare)
844 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
847 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
848 &vd->vdev_resilver_txg);
850 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
851 &vd->vdev_rebuild_txg);
853 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER))
854 vdev_defer_resilver(vd);
857 * In general, when importing a pool we want to ignore the
858 * persistent fault state, as the diagnosis made on another
859 * system may not be valid in the current context. The only
860 * exception is if we forced a vdev to a persistently faulted
861 * state with 'zpool offline -f'. The persistent fault will
862 * remain across imports until cleared.
864 * Local vdevs will remain in the faulted state.
866 if (spa_load_state(spa) == SPA_LOAD_OPEN ||
867 spa_load_state(spa) == SPA_LOAD_IMPORT) {
868 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
870 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
872 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
875 if (vd->vdev_faulted || vd->vdev_degraded) {
879 VDEV_AUX_ERR_EXCEEDED;
880 if (nvlist_lookup_string(nv,
881 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
882 strcmp(aux, "external") == 0)
883 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
885 vd->vdev_faulted = 0ULL;
891 * Add ourselves to the parent's list of children.
893 vdev_add_child(parent, vd);
901 vdev_free(vdev_t *vd)
903 spa_t *spa = vd->vdev_spa;
905 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
906 ASSERT3P(vd->vdev_trim_thread, ==, NULL);
907 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL);
908 ASSERT3P(vd->vdev_rebuild_thread, ==, NULL);
911 * Scan queues are normally destroyed at the end of a scan. If the
912 * queue exists here, that implies the vdev is being removed while
913 * the scan is still running.
915 if (vd->vdev_scan_io_queue != NULL) {
916 mutex_enter(&vd->vdev_scan_io_queue_lock);
917 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
918 vd->vdev_scan_io_queue = NULL;
919 mutex_exit(&vd->vdev_scan_io_queue_lock);
923 * vdev_free() implies closing the vdev first. This is simpler than
924 * trying to ensure complicated semantics for all callers.
928 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
929 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
934 for (int c = 0; c < vd->vdev_children; c++)
935 vdev_free(vd->vdev_child[c]);
937 ASSERT(vd->vdev_child == NULL);
938 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
941 * Discard allocation state.
943 if (vd->vdev_mg != NULL) {
944 vdev_metaslab_fini(vd);
945 metaslab_group_destroy(vd->vdev_mg);
949 ASSERT0(vd->vdev_stat.vs_space);
950 ASSERT0(vd->vdev_stat.vs_dspace);
951 ASSERT0(vd->vdev_stat.vs_alloc);
954 * Remove this vdev from its parent's child list.
956 vdev_remove_child(vd->vdev_parent, vd);
958 ASSERT(vd->vdev_parent == NULL);
959 ASSERT(!list_link_active(&vd->vdev_leaf_node));
962 * Clean up vdev structure.
968 spa_strfree(vd->vdev_path);
970 spa_strfree(vd->vdev_devid);
971 if (vd->vdev_physpath)
972 spa_strfree(vd->vdev_physpath);
974 if (vd->vdev_enc_sysfs_path)
975 spa_strfree(vd->vdev_enc_sysfs_path);
978 spa_strfree(vd->vdev_fru);
980 if (vd->vdev_isspare)
981 spa_spare_remove(vd);
982 if (vd->vdev_isl2cache)
983 spa_l2cache_remove(vd);
985 txg_list_destroy(&vd->vdev_ms_list);
986 txg_list_destroy(&vd->vdev_dtl_list);
988 mutex_enter(&vd->vdev_dtl_lock);
989 space_map_close(vd->vdev_dtl_sm);
990 for (int t = 0; t < DTL_TYPES; t++) {
991 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
992 range_tree_destroy(vd->vdev_dtl[t]);
994 mutex_exit(&vd->vdev_dtl_lock);
996 EQUIV(vd->vdev_indirect_births != NULL,
997 vd->vdev_indirect_mapping != NULL);
998 if (vd->vdev_indirect_births != NULL) {
999 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
1000 vdev_indirect_births_close(vd->vdev_indirect_births);
1003 if (vd->vdev_obsolete_sm != NULL) {
1004 ASSERT(vd->vdev_removing ||
1005 vd->vdev_ops == &vdev_indirect_ops);
1006 space_map_close(vd->vdev_obsolete_sm);
1007 vd->vdev_obsolete_sm = NULL;
1009 range_tree_destroy(vd->vdev_obsolete_segments);
1010 rw_destroy(&vd->vdev_indirect_rwlock);
1011 mutex_destroy(&vd->vdev_obsolete_lock);
1013 mutex_destroy(&vd->vdev_dtl_lock);
1014 mutex_destroy(&vd->vdev_stat_lock);
1015 mutex_destroy(&vd->vdev_probe_lock);
1016 mutex_destroy(&vd->vdev_scan_io_queue_lock);
1018 mutex_destroy(&vd->vdev_initialize_lock);
1019 mutex_destroy(&vd->vdev_initialize_io_lock);
1020 cv_destroy(&vd->vdev_initialize_io_cv);
1021 cv_destroy(&vd->vdev_initialize_cv);
1023 mutex_destroy(&vd->vdev_trim_lock);
1024 mutex_destroy(&vd->vdev_autotrim_lock);
1025 mutex_destroy(&vd->vdev_trim_io_lock);
1026 cv_destroy(&vd->vdev_trim_cv);
1027 cv_destroy(&vd->vdev_autotrim_cv);
1028 cv_destroy(&vd->vdev_trim_io_cv);
1030 mutex_destroy(&vd->vdev_rebuild_lock);
1031 mutex_destroy(&vd->vdev_rebuild_io_lock);
1032 cv_destroy(&vd->vdev_rebuild_cv);
1033 cv_destroy(&vd->vdev_rebuild_io_cv);
1035 zfs_ratelimit_fini(&vd->vdev_delay_rl);
1036 zfs_ratelimit_fini(&vd->vdev_checksum_rl);
1038 if (vd == spa->spa_root_vdev)
1039 spa->spa_root_vdev = NULL;
1041 kmem_free(vd, sizeof (vdev_t));
1045 * Transfer top-level vdev state from svd to tvd.
1048 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
1050 spa_t *spa = svd->vdev_spa;
1055 ASSERT(tvd == tvd->vdev_top);
1057 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite;
1058 tvd->vdev_ms_array = svd->vdev_ms_array;
1059 tvd->vdev_ms_shift = svd->vdev_ms_shift;
1060 tvd->vdev_ms_count = svd->vdev_ms_count;
1061 tvd->vdev_top_zap = svd->vdev_top_zap;
1063 svd->vdev_ms_array = 0;
1064 svd->vdev_ms_shift = 0;
1065 svd->vdev_ms_count = 0;
1066 svd->vdev_top_zap = 0;
1069 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1070 tvd->vdev_mg = svd->vdev_mg;
1071 tvd->vdev_ms = svd->vdev_ms;
1073 svd->vdev_mg = NULL;
1074 svd->vdev_ms = NULL;
1076 if (tvd->vdev_mg != NULL)
1077 tvd->vdev_mg->mg_vd = tvd;
1079 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1080 svd->vdev_checkpoint_sm = NULL;
1082 tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
1083 svd->vdev_alloc_bias = VDEV_BIAS_NONE;
1085 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1086 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1087 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1089 svd->vdev_stat.vs_alloc = 0;
1090 svd->vdev_stat.vs_space = 0;
1091 svd->vdev_stat.vs_dspace = 0;
1094 * State which may be set on a top-level vdev that's in the
1095 * process of being removed.
1097 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1098 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1099 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1100 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1101 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1102 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1103 ASSERT0(tvd->vdev_removing);
1104 ASSERT0(tvd->vdev_rebuilding);
1105 tvd->vdev_removing = svd->vdev_removing;
1106 tvd->vdev_rebuilding = svd->vdev_rebuilding;
1107 tvd->vdev_rebuild_config = svd->vdev_rebuild_config;
1108 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1109 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1110 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1111 range_tree_swap(&svd->vdev_obsolete_segments,
1112 &tvd->vdev_obsolete_segments);
1113 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1114 svd->vdev_indirect_config.vic_mapping_object = 0;
1115 svd->vdev_indirect_config.vic_births_object = 0;
1116 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1117 svd->vdev_indirect_mapping = NULL;
1118 svd->vdev_indirect_births = NULL;
1119 svd->vdev_obsolete_sm = NULL;
1120 svd->vdev_removing = 0;
1121 svd->vdev_rebuilding = 0;
1123 for (t = 0; t < TXG_SIZE; t++) {
1124 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1125 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1126 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1127 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1128 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1129 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1132 if (list_link_active(&svd->vdev_config_dirty_node)) {
1133 vdev_config_clean(svd);
1134 vdev_config_dirty(tvd);
1137 if (list_link_active(&svd->vdev_state_dirty_node)) {
1138 vdev_state_clean(svd);
1139 vdev_state_dirty(tvd);
1142 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1143 svd->vdev_deflate_ratio = 0;
1145 tvd->vdev_islog = svd->vdev_islog;
1146 svd->vdev_islog = 0;
1148 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1152 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1159 for (int c = 0; c < vd->vdev_children; c++)
1160 vdev_top_update(tvd, vd->vdev_child[c]);
1164 * Add a mirror/replacing vdev above an existing vdev.
1167 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1169 spa_t *spa = cvd->vdev_spa;
1170 vdev_t *pvd = cvd->vdev_parent;
1173 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1175 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1177 mvd->vdev_asize = cvd->vdev_asize;
1178 mvd->vdev_min_asize = cvd->vdev_min_asize;
1179 mvd->vdev_max_asize = cvd->vdev_max_asize;
1180 mvd->vdev_psize = cvd->vdev_psize;
1181 mvd->vdev_ashift = cvd->vdev_ashift;
1182 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1183 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1184 mvd->vdev_state = cvd->vdev_state;
1185 mvd->vdev_crtxg = cvd->vdev_crtxg;
1187 vdev_remove_child(pvd, cvd);
1188 vdev_add_child(pvd, mvd);
1189 cvd->vdev_id = mvd->vdev_children;
1190 vdev_add_child(mvd, cvd);
1191 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1193 if (mvd == mvd->vdev_top)
1194 vdev_top_transfer(cvd, mvd);
1200 * Remove a 1-way mirror/replacing vdev from the tree.
1203 vdev_remove_parent(vdev_t *cvd)
1205 vdev_t *mvd = cvd->vdev_parent;
1206 vdev_t *pvd = mvd->vdev_parent;
1208 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1210 ASSERT(mvd->vdev_children == 1);
1211 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1212 mvd->vdev_ops == &vdev_replacing_ops ||
1213 mvd->vdev_ops == &vdev_spare_ops);
1214 cvd->vdev_ashift = mvd->vdev_ashift;
1215 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1216 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1217 vdev_remove_child(mvd, cvd);
1218 vdev_remove_child(pvd, mvd);
1221 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1222 * Otherwise, we could have detached an offline device, and when we
1223 * go to import the pool we'll think we have two top-level vdevs,
1224 * instead of a different version of the same top-level vdev.
1226 if (mvd->vdev_top == mvd) {
1227 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1228 cvd->vdev_orig_guid = cvd->vdev_guid;
1229 cvd->vdev_guid += guid_delta;
1230 cvd->vdev_guid_sum += guid_delta;
1233 * If pool not set for autoexpand, we need to also preserve
1234 * mvd's asize to prevent automatic expansion of cvd.
1235 * Otherwise if we are adjusting the mirror by attaching and
1236 * detaching children of non-uniform sizes, the mirror could
1237 * autoexpand, unexpectedly requiring larger devices to
1238 * re-establish the mirror.
1240 if (!cvd->vdev_spa->spa_autoexpand)
1241 cvd->vdev_asize = mvd->vdev_asize;
1243 cvd->vdev_id = mvd->vdev_id;
1244 vdev_add_child(pvd, cvd);
1245 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1247 if (cvd == cvd->vdev_top)
1248 vdev_top_transfer(mvd, cvd);
1250 ASSERT(mvd->vdev_children == 0);
1255 vdev_metaslab_group_create(vdev_t *vd)
1257 spa_t *spa = vd->vdev_spa;
1260 * metaslab_group_create was delayed until allocation bias was available
1262 if (vd->vdev_mg == NULL) {
1263 metaslab_class_t *mc;
1265 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
1266 vd->vdev_alloc_bias = VDEV_BIAS_LOG;
1268 ASSERT3U(vd->vdev_islog, ==,
1269 (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
1271 switch (vd->vdev_alloc_bias) {
1273 mc = spa_log_class(spa);
1275 case VDEV_BIAS_SPECIAL:
1276 mc = spa_special_class(spa);
1278 case VDEV_BIAS_DEDUP:
1279 mc = spa_dedup_class(spa);
1282 mc = spa_normal_class(spa);
1285 vd->vdev_mg = metaslab_group_create(mc, vd,
1286 spa->spa_alloc_count);
1289 * The spa ashift values currently only reflect the
1290 * general vdev classes. Class destination is late
1291 * binding so ashift checking had to wait until now
1293 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1294 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
1295 if (vd->vdev_ashift > spa->spa_max_ashift)
1296 spa->spa_max_ashift = vd->vdev_ashift;
1297 if (vd->vdev_ashift < spa->spa_min_ashift)
1298 spa->spa_min_ashift = vd->vdev_ashift;
1304 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1306 spa_t *spa = vd->vdev_spa;
1307 objset_t *mos = spa->spa_meta_objset;
1309 uint64_t oldc = vd->vdev_ms_count;
1310 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1313 boolean_t expanding = (oldc != 0);
1315 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1318 * This vdev is not being allocated from yet or is a hole.
1320 if (vd->vdev_ms_shift == 0)
1323 ASSERT(!vd->vdev_ishole);
1325 ASSERT(oldc <= newc);
1327 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1330 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1331 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1335 vd->vdev_ms_count = newc;
1336 for (m = oldc; m < newc; m++) {
1337 uint64_t object = 0;
1340 * vdev_ms_array may be 0 if we are creating the "fake"
1341 * metaslabs for an indirect vdev for zdb's leak detection.
1342 * See zdb_leak_init().
1344 if (txg == 0 && vd->vdev_ms_array != 0) {
1345 error = dmu_read(mos, vd->vdev_ms_array,
1346 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1349 vdev_dbgmsg(vd, "unable to read the metaslab "
1350 "array [error=%d]", error);
1357 * To accommodate zdb_leak_init() fake indirect
1358 * metaslabs, we allocate a metaslab group for
1359 * indirect vdevs which normally don't have one.
1361 if (vd->vdev_mg == NULL) {
1362 ASSERT0(vdev_is_concrete(vd));
1363 vdev_metaslab_group_create(vd);
1366 error = metaslab_init(vd->vdev_mg, m, object, txg,
1369 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1376 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1379 * If the vdev is being removed we don't activate
1380 * the metaslabs since we want to ensure that no new
1381 * allocations are performed on this device.
1383 if (!expanding && !vd->vdev_removing) {
1384 metaslab_group_activate(vd->vdev_mg);
1388 spa_config_exit(spa, SCL_ALLOC, FTAG);
1391 * Regardless whether this vdev was just added or it is being
1392 * expanded, the metaslab count has changed. Recalculate the
1395 spa_log_sm_set_blocklimit(spa);
1401 vdev_metaslab_fini(vdev_t *vd)
1403 if (vd->vdev_checkpoint_sm != NULL) {
1404 ASSERT(spa_feature_is_active(vd->vdev_spa,
1405 SPA_FEATURE_POOL_CHECKPOINT));
1406 space_map_close(vd->vdev_checkpoint_sm);
1408 * Even though we close the space map, we need to set its
1409 * pointer to NULL. The reason is that vdev_metaslab_fini()
1410 * may be called multiple times for certain operations
1411 * (i.e. when destroying a pool) so we need to ensure that
1412 * this clause never executes twice. This logic is similar
1413 * to the one used for the vdev_ms clause below.
1415 vd->vdev_checkpoint_sm = NULL;
1418 if (vd->vdev_ms != NULL) {
1419 metaslab_group_t *mg = vd->vdev_mg;
1420 metaslab_group_passivate(mg);
1422 uint64_t count = vd->vdev_ms_count;
1423 for (uint64_t m = 0; m < count; m++) {
1424 metaslab_t *msp = vd->vdev_ms[m];
1428 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1431 vd->vdev_ms_count = 0;
1433 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
1434 ASSERT0(mg->mg_histogram[i]);
1436 ASSERT0(vd->vdev_ms_count);
1437 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
1440 typedef struct vdev_probe_stats {
1441 boolean_t vps_readable;
1442 boolean_t vps_writeable;
1444 } vdev_probe_stats_t;
1447 vdev_probe_done(zio_t *zio)
1449 spa_t *spa = zio->io_spa;
1450 vdev_t *vd = zio->io_vd;
1451 vdev_probe_stats_t *vps = zio->io_private;
1453 ASSERT(vd->vdev_probe_zio != NULL);
1455 if (zio->io_type == ZIO_TYPE_READ) {
1456 if (zio->io_error == 0)
1457 vps->vps_readable = 1;
1458 if (zio->io_error == 0 && spa_writeable(spa)) {
1459 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1460 zio->io_offset, zio->io_size, zio->io_abd,
1461 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1462 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1464 abd_free(zio->io_abd);
1466 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1467 if (zio->io_error == 0)
1468 vps->vps_writeable = 1;
1469 abd_free(zio->io_abd);
1470 } else if (zio->io_type == ZIO_TYPE_NULL) {
1474 vd->vdev_cant_read |= !vps->vps_readable;
1475 vd->vdev_cant_write |= !vps->vps_writeable;
1477 if (vdev_readable(vd) &&
1478 (vdev_writeable(vd) || !spa_writeable(spa))) {
1481 ASSERT(zio->io_error != 0);
1482 vdev_dbgmsg(vd, "failed probe");
1483 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1484 spa, vd, NULL, NULL, 0);
1485 zio->io_error = SET_ERROR(ENXIO);
1488 mutex_enter(&vd->vdev_probe_lock);
1489 ASSERT(vd->vdev_probe_zio == zio);
1490 vd->vdev_probe_zio = NULL;
1491 mutex_exit(&vd->vdev_probe_lock);
1494 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1495 if (!vdev_accessible(vd, pio))
1496 pio->io_error = SET_ERROR(ENXIO);
1498 kmem_free(vps, sizeof (*vps));
1503 * Determine whether this device is accessible.
1505 * Read and write to several known locations: the pad regions of each
1506 * vdev label but the first, which we leave alone in case it contains
1510 vdev_probe(vdev_t *vd, zio_t *zio)
1512 spa_t *spa = vd->vdev_spa;
1513 vdev_probe_stats_t *vps = NULL;
1516 ASSERT(vd->vdev_ops->vdev_op_leaf);
1519 * Don't probe the probe.
1521 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1525 * To prevent 'probe storms' when a device fails, we create
1526 * just one probe i/o at a time. All zios that want to probe
1527 * this vdev will become parents of the probe io.
1529 mutex_enter(&vd->vdev_probe_lock);
1531 if ((pio = vd->vdev_probe_zio) == NULL) {
1532 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1534 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1535 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1538 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1540 * vdev_cant_read and vdev_cant_write can only
1541 * transition from TRUE to FALSE when we have the
1542 * SCL_ZIO lock as writer; otherwise they can only
1543 * transition from FALSE to TRUE. This ensures that
1544 * any zio looking at these values can assume that
1545 * failures persist for the life of the I/O. That's
1546 * important because when a device has intermittent
1547 * connectivity problems, we want to ensure that
1548 * they're ascribed to the device (ENXIO) and not
1551 * Since we hold SCL_ZIO as writer here, clear both
1552 * values so the probe can reevaluate from first
1555 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1556 vd->vdev_cant_read = B_FALSE;
1557 vd->vdev_cant_write = B_FALSE;
1560 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1561 vdev_probe_done, vps,
1562 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1565 * We can't change the vdev state in this context, so we
1566 * kick off an async task to do it on our behalf.
1569 vd->vdev_probe_wanted = B_TRUE;
1570 spa_async_request(spa, SPA_ASYNC_PROBE);
1575 zio_add_child(zio, pio);
1577 mutex_exit(&vd->vdev_probe_lock);
1580 ASSERT(zio != NULL);
1584 for (int l = 1; l < VDEV_LABELS; l++) {
1585 zio_nowait(zio_read_phys(pio, vd,
1586 vdev_label_offset(vd->vdev_psize, l,
1587 offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE,
1588 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1589 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1590 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1601 vdev_open_child(void *arg)
1605 vd->vdev_open_thread = curthread;
1606 vd->vdev_open_error = vdev_open(vd);
1607 vd->vdev_open_thread = NULL;
1611 vdev_uses_zvols(vdev_t *vd)
1614 if (zvol_is_zvol(vd->vdev_path))
1618 for (int c = 0; c < vd->vdev_children; c++)
1619 if (vdev_uses_zvols(vd->vdev_child[c]))
1626 vdev_open_children(vdev_t *vd)
1629 int children = vd->vdev_children;
1632 * in order to handle pools on top of zvols, do the opens
1633 * in a single thread so that the same thread holds the
1634 * spa_namespace_lock
1636 if (vdev_uses_zvols(vd)) {
1638 for (int c = 0; c < children; c++)
1639 vd->vdev_child[c]->vdev_open_error =
1640 vdev_open(vd->vdev_child[c]);
1642 tq = taskq_create("vdev_open", children, minclsyspri,
1643 children, children, TASKQ_PREPOPULATE);
1647 for (int c = 0; c < children; c++)
1648 VERIFY(taskq_dispatch(tq, vdev_open_child,
1649 vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID);
1654 vd->vdev_nonrot = B_TRUE;
1656 for (int c = 0; c < children; c++)
1657 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1661 * Compute the raidz-deflation ratio. Note, we hard-code
1662 * in 128k (1 << 17) because it is the "typical" blocksize.
1663 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1664 * otherwise it would inconsistently account for existing bp's.
1667 vdev_set_deflate_ratio(vdev_t *vd)
1669 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1670 vd->vdev_deflate_ratio = (1 << 17) /
1671 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1676 * Maximize performance by inflating the configured ashift for top level
1677 * vdevs to be as close to the physical ashift as possible while maintaining
1678 * administrator defined limits and ensuring it doesn't go below the
1682 vdev_ashift_optimize(vdev_t *vd)
1684 ASSERT(vd == vd->vdev_top);
1686 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1687 vd->vdev_ashift = MIN(
1688 MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift),
1689 MAX(zfs_vdev_min_auto_ashift,
1690 vd->vdev_physical_ashift));
1693 * If the logical and physical ashifts are the same, then
1694 * we ensure that the top-level vdev's ashift is not smaller
1695 * than our minimum ashift value. For the unusual case
1696 * where logical ashift > physical ashift, we can't cap
1697 * the calculated ashift based on max ashift as that
1698 * would cause failures.
1699 * We still check if we need to increase it to match
1702 vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift,
1708 * Prepare a virtual device for access.
1711 vdev_open(vdev_t *vd)
1713 spa_t *spa = vd->vdev_spa;
1716 uint64_t max_osize = 0;
1717 uint64_t asize, max_asize, psize;
1718 uint64_t logical_ashift = 0;
1719 uint64_t physical_ashift = 0;
1721 ASSERT(vd->vdev_open_thread == curthread ||
1722 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1723 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1724 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1725 vd->vdev_state == VDEV_STATE_OFFLINE);
1727 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1728 vd->vdev_cant_read = B_FALSE;
1729 vd->vdev_cant_write = B_FALSE;
1730 vd->vdev_min_asize = vdev_get_min_asize(vd);
1733 * If this vdev is not removed, check its fault status. If it's
1734 * faulted, bail out of the open.
1736 if (!vd->vdev_removed && vd->vdev_faulted) {
1737 ASSERT(vd->vdev_children == 0);
1738 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1739 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1740 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1741 vd->vdev_label_aux);
1742 return (SET_ERROR(ENXIO));
1743 } else if (vd->vdev_offline) {
1744 ASSERT(vd->vdev_children == 0);
1745 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1746 return (SET_ERROR(ENXIO));
1749 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1750 &logical_ashift, &physical_ashift);
1752 * Physical volume size should never be larger than its max size, unless
1753 * the disk has shrunk while we were reading it or the device is buggy
1754 * or damaged: either way it's not safe for use, bail out of the open.
1756 if (osize > max_osize) {
1757 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1758 VDEV_AUX_OPEN_FAILED);
1759 return (SET_ERROR(ENXIO));
1763 * Reset the vdev_reopening flag so that we actually close
1764 * the vdev on error.
1766 vd->vdev_reopening = B_FALSE;
1767 if (zio_injection_enabled && error == 0)
1768 error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO));
1771 if (vd->vdev_removed &&
1772 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1773 vd->vdev_removed = B_FALSE;
1775 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1776 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1777 vd->vdev_stat.vs_aux);
1779 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1780 vd->vdev_stat.vs_aux);
1785 vd->vdev_removed = B_FALSE;
1788 * Recheck the faulted flag now that we have confirmed that
1789 * the vdev is accessible. If we're faulted, bail.
1791 if (vd->vdev_faulted) {
1792 ASSERT(vd->vdev_children == 0);
1793 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1794 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1795 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1796 vd->vdev_label_aux);
1797 return (SET_ERROR(ENXIO));
1800 if (vd->vdev_degraded) {
1801 ASSERT(vd->vdev_children == 0);
1802 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1803 VDEV_AUX_ERR_EXCEEDED);
1805 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1809 * For hole or missing vdevs we just return success.
1811 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1814 for (int c = 0; c < vd->vdev_children; c++) {
1815 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1816 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1822 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1823 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1825 if (vd->vdev_children == 0) {
1826 if (osize < SPA_MINDEVSIZE) {
1827 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1828 VDEV_AUX_TOO_SMALL);
1829 return (SET_ERROR(EOVERFLOW));
1832 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1833 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1834 VDEV_LABEL_END_SIZE);
1836 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1837 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1838 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1839 VDEV_AUX_TOO_SMALL);
1840 return (SET_ERROR(EOVERFLOW));
1844 max_asize = max_osize;
1848 * If the vdev was expanded, record this so that we can re-create the
1849 * uberblock rings in labels {2,3}, during the next sync.
1851 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0))
1852 vd->vdev_copy_uberblocks = B_TRUE;
1854 vd->vdev_psize = psize;
1857 * Make sure the allocatable size hasn't shrunk too much.
1859 if (asize < vd->vdev_min_asize) {
1860 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1861 VDEV_AUX_BAD_LABEL);
1862 return (SET_ERROR(EINVAL));
1866 * We can always set the logical/physical ashift members since
1867 * their values are only used to calculate the vdev_ashift when
1868 * the device is first added to the config. These values should
1869 * not be used for anything else since they may change whenever
1870 * the device is reopened and we don't store them in the label.
1872 vd->vdev_physical_ashift =
1873 MAX(physical_ashift, vd->vdev_physical_ashift);
1874 vd->vdev_logical_ashift = MAX(logical_ashift,
1875 vd->vdev_logical_ashift);
1877 if (vd->vdev_asize == 0) {
1879 * This is the first-ever open, so use the computed values.
1880 * For compatibility, a different ashift can be requested.
1882 vd->vdev_asize = asize;
1883 vd->vdev_max_asize = max_asize;
1886 * If the vdev_ashift was not overriden at creation time,
1887 * then set it the logical ashift and optimize the ashift.
1889 if (vd->vdev_ashift == 0) {
1890 vd->vdev_ashift = vd->vdev_logical_ashift;
1892 if (vd->vdev_logical_ashift > ASHIFT_MAX) {
1893 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1894 VDEV_AUX_ASHIFT_TOO_BIG);
1895 return (SET_ERROR(EDOM));
1898 if (vd->vdev_top == vd) {
1899 vdev_ashift_optimize(vd);
1902 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
1903 vd->vdev_ashift > ASHIFT_MAX)) {
1904 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1905 VDEV_AUX_BAD_ASHIFT);
1906 return (SET_ERROR(EDOM));
1910 * Make sure the alignment required hasn't increased.
1912 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1913 vd->vdev_ops->vdev_op_leaf) {
1914 (void) zfs_ereport_post(
1915 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
1916 spa, vd, NULL, NULL, 0);
1917 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1918 VDEV_AUX_BAD_LABEL);
1919 return (SET_ERROR(EDOM));
1921 vd->vdev_max_asize = max_asize;
1925 * If all children are healthy we update asize if either:
1926 * The asize has increased, due to a device expansion caused by dynamic
1927 * LUN growth or vdev replacement, and automatic expansion is enabled;
1928 * making the additional space available.
1930 * The asize has decreased, due to a device shrink usually caused by a
1931 * vdev replace with a smaller device. This ensures that calculations
1932 * based of max_asize and asize e.g. esize are always valid. It's safe
1933 * to do this as we've already validated that asize is greater than
1936 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1937 ((asize > vd->vdev_asize &&
1938 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1939 (asize < vd->vdev_asize)))
1940 vd->vdev_asize = asize;
1942 vdev_set_min_asize(vd);
1945 * Ensure we can issue some IO before declaring the
1946 * vdev open for business.
1948 if (vd->vdev_ops->vdev_op_leaf &&
1949 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1950 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1951 VDEV_AUX_ERR_EXCEEDED);
1956 * Track the min and max ashift values for normal data devices.
1958 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1959 vd->vdev_alloc_bias == VDEV_BIAS_NONE &&
1960 vd->vdev_islog == 0 && vd->vdev_aux == NULL) {
1961 if (vd->vdev_ashift > spa->spa_max_ashift)
1962 spa->spa_max_ashift = vd->vdev_ashift;
1963 if (vd->vdev_ashift < spa->spa_min_ashift)
1964 spa->spa_min_ashift = vd->vdev_ashift;
1968 * If this is a leaf vdev, assess whether a resilver is needed.
1969 * But don't do this if we are doing a reopen for a scrub, since
1970 * this would just restart the scrub we are already doing.
1972 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen)
1973 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd);
1979 * Called once the vdevs are all opened, this routine validates the label
1980 * contents. This needs to be done before vdev_load() so that we don't
1981 * inadvertently do repair I/Os to the wrong device.
1983 * This function will only return failure if one of the vdevs indicates that it
1984 * has since been destroyed or exported. This is only possible if
1985 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1986 * will be updated but the function will return 0.
1989 vdev_validate(vdev_t *vd)
1991 spa_t *spa = vd->vdev_spa;
1993 uint64_t guid = 0, aux_guid = 0, top_guid;
1998 if (vdev_validate_skip)
2001 for (uint64_t c = 0; c < vd->vdev_children; c++)
2002 if (vdev_validate(vd->vdev_child[c]) != 0)
2003 return (SET_ERROR(EBADF));
2006 * If the device has already failed, or was marked offline, don't do
2007 * any further validation. Otherwise, label I/O will fail and we will
2008 * overwrite the previous state.
2010 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
2014 * If we are performing an extreme rewind, we allow for a label that
2015 * was modified at a point after the current txg.
2016 * If config lock is not held do not check for the txg. spa_sync could
2017 * be updating the vdev's label before updating spa_last_synced_txg.
2019 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
2020 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
2023 txg = spa_last_synced_txg(spa);
2025 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
2026 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2027 VDEV_AUX_BAD_LABEL);
2028 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
2029 "txg %llu", (u_longlong_t)txg);
2034 * Determine if this vdev has been split off into another
2035 * pool. If so, then refuse to open it.
2037 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
2038 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
2039 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2040 VDEV_AUX_SPLIT_POOL);
2042 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
2046 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
2047 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2048 VDEV_AUX_CORRUPT_DATA);
2050 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2051 ZPOOL_CONFIG_POOL_GUID);
2056 * If config is not trusted then ignore the spa guid check. This is
2057 * necessary because if the machine crashed during a re-guid the new
2058 * guid might have been written to all of the vdev labels, but not the
2059 * cached config. The check will be performed again once we have the
2060 * trusted config from the MOS.
2062 if (spa->spa_trust_config && guid != spa_guid(spa)) {
2063 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2064 VDEV_AUX_CORRUPT_DATA);
2066 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
2067 "match config (%llu != %llu)", (u_longlong_t)guid,
2068 (u_longlong_t)spa_guid(spa));
2072 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
2073 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
2077 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
2078 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2079 VDEV_AUX_CORRUPT_DATA);
2081 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2086 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
2088 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2089 VDEV_AUX_CORRUPT_DATA);
2091 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2092 ZPOOL_CONFIG_TOP_GUID);
2097 * If this vdev just became a top-level vdev because its sibling was
2098 * detached, it will have adopted the parent's vdev guid -- but the
2099 * label may or may not be on disk yet. Fortunately, either version
2100 * of the label will have the same top guid, so if we're a top-level
2101 * vdev, we can safely compare to that instead.
2102 * However, if the config comes from a cachefile that failed to update
2103 * after the detach, a top-level vdev will appear as a non top-level
2104 * vdev in the config. Also relax the constraints if we perform an
2107 * If we split this vdev off instead, then we also check the
2108 * original pool's guid. We don't want to consider the vdev
2109 * corrupt if it is partway through a split operation.
2111 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
2112 boolean_t mismatch = B_FALSE;
2113 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
2114 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
2117 if (vd->vdev_guid != top_guid &&
2118 vd->vdev_top->vdev_guid != guid)
2123 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2124 VDEV_AUX_CORRUPT_DATA);
2126 vdev_dbgmsg(vd, "vdev_validate: config guid "
2127 "doesn't match label guid");
2128 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
2129 (u_longlong_t)vd->vdev_guid,
2130 (u_longlong_t)vd->vdev_top->vdev_guid);
2131 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
2132 "aux_guid %llu", (u_longlong_t)guid,
2133 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
2138 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
2140 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2141 VDEV_AUX_CORRUPT_DATA);
2143 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2144 ZPOOL_CONFIG_POOL_STATE);
2151 * If this is a verbatim import, no need to check the
2152 * state of the pool.
2154 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
2155 spa_load_state(spa) == SPA_LOAD_OPEN &&
2156 state != POOL_STATE_ACTIVE) {
2157 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
2158 "for spa %s", (u_longlong_t)state, spa->spa_name);
2159 return (SET_ERROR(EBADF));
2163 * If we were able to open and validate a vdev that was
2164 * previously marked permanently unavailable, clear that state
2167 if (vd->vdev_not_present)
2168 vd->vdev_not_present = 0;
2174 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
2176 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
2177 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
2178 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2179 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
2180 dvd->vdev_path, svd->vdev_path);
2181 spa_strfree(dvd->vdev_path);
2182 dvd->vdev_path = spa_strdup(svd->vdev_path);
2184 } else if (svd->vdev_path != NULL) {
2185 dvd->vdev_path = spa_strdup(svd->vdev_path);
2186 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2187 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
2192 * Recursively copy vdev paths from one vdev to another. Source and destination
2193 * vdev trees must have same geometry otherwise return error. Intended to copy
2194 * paths from userland config into MOS config.
2197 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
2199 if ((svd->vdev_ops == &vdev_missing_ops) ||
2200 (svd->vdev_ishole && dvd->vdev_ishole) ||
2201 (dvd->vdev_ops == &vdev_indirect_ops))
2204 if (svd->vdev_ops != dvd->vdev_ops) {
2205 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
2206 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
2207 return (SET_ERROR(EINVAL));
2210 if (svd->vdev_guid != dvd->vdev_guid) {
2211 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
2212 "%llu)", (u_longlong_t)svd->vdev_guid,
2213 (u_longlong_t)dvd->vdev_guid);
2214 return (SET_ERROR(EINVAL));
2217 if (svd->vdev_children != dvd->vdev_children) {
2218 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2219 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2220 (u_longlong_t)dvd->vdev_children);
2221 return (SET_ERROR(EINVAL));
2224 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2225 int error = vdev_copy_path_strict(svd->vdev_child[i],
2226 dvd->vdev_child[i]);
2231 if (svd->vdev_ops->vdev_op_leaf)
2232 vdev_copy_path_impl(svd, dvd);
2238 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2240 ASSERT(stvd->vdev_top == stvd);
2241 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2243 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2244 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2247 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2251 * The idea here is that while a vdev can shift positions within
2252 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2253 * step outside of it.
2255 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2257 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2260 ASSERT(vd->vdev_ops->vdev_op_leaf);
2262 vdev_copy_path_impl(vd, dvd);
2266 * Recursively copy vdev paths from one root vdev to another. Source and
2267 * destination vdev trees may differ in geometry. For each destination leaf
2268 * vdev, search a vdev with the same guid and top vdev id in the source.
2269 * Intended to copy paths from userland config into MOS config.
2272 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2274 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2275 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2276 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2278 for (uint64_t i = 0; i < children; i++) {
2279 vdev_copy_path_search(srvd->vdev_child[i],
2280 drvd->vdev_child[i]);
2285 * Close a virtual device.
2288 vdev_close(vdev_t *vd)
2290 vdev_t *pvd = vd->vdev_parent;
2291 spa_t *spa __maybe_unused = vd->vdev_spa;
2293 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2296 * If our parent is reopening, then we are as well, unless we are
2299 if (pvd != NULL && pvd->vdev_reopening)
2300 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2302 vd->vdev_ops->vdev_op_close(vd);
2304 vdev_cache_purge(vd);
2307 * We record the previous state before we close it, so that if we are
2308 * doing a reopen(), we don't generate FMA ereports if we notice that
2309 * it's still faulted.
2311 vd->vdev_prevstate = vd->vdev_state;
2313 if (vd->vdev_offline)
2314 vd->vdev_state = VDEV_STATE_OFFLINE;
2316 vd->vdev_state = VDEV_STATE_CLOSED;
2317 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2321 vdev_hold(vdev_t *vd)
2323 spa_t *spa = vd->vdev_spa;
2325 ASSERT(spa_is_root(spa));
2326 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2329 for (int c = 0; c < vd->vdev_children; c++)
2330 vdev_hold(vd->vdev_child[c]);
2332 if (vd->vdev_ops->vdev_op_leaf)
2333 vd->vdev_ops->vdev_op_hold(vd);
2337 vdev_rele(vdev_t *vd)
2339 ASSERT(spa_is_root(vd->vdev_spa));
2340 for (int c = 0; c < vd->vdev_children; c++)
2341 vdev_rele(vd->vdev_child[c]);
2343 if (vd->vdev_ops->vdev_op_leaf)
2344 vd->vdev_ops->vdev_op_rele(vd);
2348 * Reopen all interior vdevs and any unopened leaves. We don't actually
2349 * reopen leaf vdevs which had previously been opened as they might deadlock
2350 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2351 * If the leaf has never been opened then open it, as usual.
2354 vdev_reopen(vdev_t *vd)
2356 spa_t *spa = vd->vdev_spa;
2358 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2360 /* set the reopening flag unless we're taking the vdev offline */
2361 vd->vdev_reopening = !vd->vdev_offline;
2363 (void) vdev_open(vd);
2366 * Call vdev_validate() here to make sure we have the same device.
2367 * Otherwise, a device with an invalid label could be successfully
2368 * opened in response to vdev_reopen().
2371 (void) vdev_validate_aux(vd);
2372 if (vdev_readable(vd) && vdev_writeable(vd) &&
2373 vd->vdev_aux == &spa->spa_l2cache) {
2375 * In case the vdev is present we should evict all ARC
2376 * buffers and pointers to log blocks and reclaim their
2377 * space before restoring its contents to L2ARC.
2379 if (l2arc_vdev_present(vd)) {
2380 l2arc_rebuild_vdev(vd, B_TRUE);
2382 l2arc_add_vdev(spa, vd);
2384 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
2385 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
2388 (void) vdev_validate(vd);
2392 * Reassess parent vdev's health.
2394 vdev_propagate_state(vd);
2398 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2403 * Normally, partial opens (e.g. of a mirror) are allowed.
2404 * For a create, however, we want to fail the request if
2405 * there are any components we can't open.
2407 error = vdev_open(vd);
2409 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2411 return (error ? error : SET_ERROR(ENXIO));
2415 * Recursively load DTLs and initialize all labels.
2417 if ((error = vdev_dtl_load(vd)) != 0 ||
2418 (error = vdev_label_init(vd, txg, isreplacing ?
2419 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2428 vdev_metaslab_set_size(vdev_t *vd)
2430 uint64_t asize = vd->vdev_asize;
2431 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2435 * There are two dimensions to the metaslab sizing calculation:
2436 * the size of the metaslab and the count of metaslabs per vdev.
2438 * The default values used below are a good balance between memory
2439 * usage (larger metaslab size means more memory needed for loaded
2440 * metaslabs; more metaslabs means more memory needed for the
2441 * metaslab_t structs), metaslab load time (larger metaslabs take
2442 * longer to load), and metaslab sync time (more metaslabs means
2443 * more time spent syncing all of them).
2445 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2446 * The range of the dimensions are as follows:
2448 * 2^29 <= ms_size <= 2^34
2449 * 16 <= ms_count <= 131,072
2451 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2452 * at least 512MB (2^29) to minimize fragmentation effects when
2453 * testing with smaller devices. However, the count constraint
2454 * of at least 16 metaslabs will override this minimum size goal.
2456 * On the upper end of vdev sizes, we aim for a maximum metaslab
2457 * size of 16GB. However, we will cap the total count to 2^17
2458 * metaslabs to keep our memory footprint in check and let the
2459 * metaslab size grow from there if that limit is hit.
2461 * The net effect of applying above constrains is summarized below.
2463 * vdev size metaslab count
2464 * --------------|-----------------
2466 * 8GB - 100GB one per 512MB
2468 * 3TB - 2PB one per 16GB
2470 * --------------------------------
2472 * Finally, note that all of the above calculate the initial
2473 * number of metaslabs. Expanding a top-level vdev will result
2474 * in additional metaslabs being allocated making it possible
2475 * to exceed the zfs_vdev_ms_count_limit.
2478 if (ms_count < zfs_vdev_min_ms_count)
2479 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2480 else if (ms_count > zfs_vdev_default_ms_count)
2481 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2483 ms_shift = zfs_vdev_default_ms_shift;
2485 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2486 ms_shift = SPA_MAXBLOCKSHIFT;
2487 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2488 ms_shift = zfs_vdev_max_ms_shift;
2489 /* cap the total count to constrain memory footprint */
2490 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2491 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2494 vd->vdev_ms_shift = ms_shift;
2495 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2499 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2501 ASSERT(vd == vd->vdev_top);
2502 /* indirect vdevs don't have metaslabs or dtls */
2503 ASSERT(vdev_is_concrete(vd) || flags == 0);
2504 ASSERT(ISP2(flags));
2505 ASSERT(spa_writeable(vd->vdev_spa));
2507 if (flags & VDD_METASLAB)
2508 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2510 if (flags & VDD_DTL)
2511 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2513 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2517 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2519 for (int c = 0; c < vd->vdev_children; c++)
2520 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2522 if (vd->vdev_ops->vdev_op_leaf)
2523 vdev_dirty(vd->vdev_top, flags, vd, txg);
2529 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2530 * the vdev has less than perfect replication. There are four kinds of DTL:
2532 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2534 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2536 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2537 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2538 * txgs that was scrubbed.
2540 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2541 * persistent errors or just some device being offline.
2542 * Unlike the other three, the DTL_OUTAGE map is not generally
2543 * maintained; it's only computed when needed, typically to
2544 * determine whether a device can be detached.
2546 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2547 * either has the data or it doesn't.
2549 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2550 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2551 * if any child is less than fully replicated, then so is its parent.
2552 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2553 * comprising only those txgs which appear in 'maxfaults' or more children;
2554 * those are the txgs we don't have enough replication to read. For example,
2555 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2556 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2557 * two child DTL_MISSING maps.
2559 * It should be clear from the above that to compute the DTLs and outage maps
2560 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2561 * Therefore, that is all we keep on disk. When loading the pool, or after
2562 * a configuration change, we generate all other DTLs from first principles.
2565 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2567 range_tree_t *rt = vd->vdev_dtl[t];
2569 ASSERT(t < DTL_TYPES);
2570 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2571 ASSERT(spa_writeable(vd->vdev_spa));
2573 mutex_enter(&vd->vdev_dtl_lock);
2574 if (!range_tree_contains(rt, txg, size))
2575 range_tree_add(rt, txg, size);
2576 mutex_exit(&vd->vdev_dtl_lock);
2580 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2582 range_tree_t *rt = vd->vdev_dtl[t];
2583 boolean_t dirty = B_FALSE;
2585 ASSERT(t < DTL_TYPES);
2586 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2589 * While we are loading the pool, the DTLs have not been loaded yet.
2590 * Ignore the DTLs and try all devices. This avoids a recursive
2591 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2592 * when loading the pool (relying on the checksum to ensure that
2593 * we get the right data -- note that we while loading, we are
2594 * only reading the MOS, which is always checksummed).
2596 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2599 mutex_enter(&vd->vdev_dtl_lock);
2600 if (!range_tree_is_empty(rt))
2601 dirty = range_tree_contains(rt, txg, size);
2602 mutex_exit(&vd->vdev_dtl_lock);
2608 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2610 range_tree_t *rt = vd->vdev_dtl[t];
2613 mutex_enter(&vd->vdev_dtl_lock);
2614 empty = range_tree_is_empty(rt);
2615 mutex_exit(&vd->vdev_dtl_lock);
2621 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2624 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2626 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2628 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2629 vd->vdev_ops->vdev_op_leaf)
2632 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2636 * Returns the lowest txg in the DTL range.
2639 vdev_dtl_min(vdev_t *vd)
2641 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2642 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2643 ASSERT0(vd->vdev_children);
2645 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1);
2649 * Returns the highest txg in the DTL.
2652 vdev_dtl_max(vdev_t *vd)
2654 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2655 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2656 ASSERT0(vd->vdev_children);
2658 return (range_tree_max(vd->vdev_dtl[DTL_MISSING]));
2662 * Determine if a resilvering vdev should remove any DTL entries from
2663 * its range. If the vdev was resilvering for the entire duration of the
2664 * scan then it should excise that range from its DTLs. Otherwise, this
2665 * vdev is considered partially resilvered and should leave its DTL
2666 * entries intact. The comment in vdev_dtl_reassess() describes how we
2670 vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done)
2672 ASSERT0(vd->vdev_children);
2674 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2677 if (vd->vdev_resilver_deferred)
2680 if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2684 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2685 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
2687 /* Rebuild not initiated by attach */
2688 if (vd->vdev_rebuild_txg == 0)
2692 * When a rebuild completes without error then all missing data
2693 * up to the rebuild max txg has been reconstructed and the DTL
2694 * is eligible for excision.
2696 if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE &&
2697 vdev_dtl_max(vd) <= vrp->vrp_max_txg) {
2698 ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd));
2699 ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg);
2700 ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg);
2704 dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan;
2705 dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys;
2707 /* Resilver not initiated by attach */
2708 if (vd->vdev_resilver_txg == 0)
2712 * When a resilver is initiated the scan will assign the
2713 * scn_max_txg value to the highest txg value that exists
2714 * in all DTLs. If this device's max DTL is not part of this
2715 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
2716 * then it is not eligible for excision.
2718 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2719 ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd));
2720 ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg);
2721 ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg);
2730 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2731 * write operations will be issued to the pool.
2734 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
2735 boolean_t scrub_done, boolean_t rebuild_done)
2737 spa_t *spa = vd->vdev_spa;
2741 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2743 for (int c = 0; c < vd->vdev_children; c++)
2744 vdev_dtl_reassess(vd->vdev_child[c], txg,
2745 scrub_txg, scrub_done, rebuild_done);
2747 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2750 if (vd->vdev_ops->vdev_op_leaf) {
2751 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2752 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2753 boolean_t check_excise = B_FALSE;
2754 boolean_t wasempty = B_TRUE;
2756 mutex_enter(&vd->vdev_dtl_lock);
2759 * If requested, pretend the scan or rebuild completed cleanly.
2761 if (zfs_scan_ignore_errors) {
2763 scn->scn_phys.scn_errors = 0;
2765 vr->vr_rebuild_phys.vrp_errors = 0;
2768 if (scrub_txg != 0 &&
2769 !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
2771 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
2772 "dtl:%llu/%llu errors:%llu",
2773 (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg,
2774 (u_longlong_t)scrub_txg, spa->spa_scrub_started,
2775 (u_longlong_t)vdev_dtl_min(vd),
2776 (u_longlong_t)vdev_dtl_max(vd),
2777 (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0));
2781 * If we've completed a scrub/resilver or a rebuild cleanly
2782 * then determine if this vdev should remove any DTLs. We
2783 * only want to excise regions on vdevs that were available
2784 * during the entire duration of this scan.
2787 vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) {
2788 check_excise = B_TRUE;
2790 if (spa->spa_scrub_started ||
2791 (scn != NULL && scn->scn_phys.scn_errors == 0)) {
2792 check_excise = B_TRUE;
2796 if (scrub_txg && check_excise &&
2797 vdev_dtl_should_excise(vd, rebuild_done)) {
2799 * We completed a scrub, resilver or rebuild up to
2800 * scrub_txg. If we did it without rebooting, then
2801 * the scrub dtl will be valid, so excise the old
2802 * region and fold in the scrub dtl. Otherwise,
2803 * leave the dtl as-is if there was an error.
2805 * There's little trick here: to excise the beginning
2806 * of the DTL_MISSING map, we put it into a reference
2807 * tree and then add a segment with refcnt -1 that
2808 * covers the range [0, scrub_txg). This means
2809 * that each txg in that range has refcnt -1 or 0.
2810 * We then add DTL_SCRUB with a refcnt of 2, so that
2811 * entries in the range [0, scrub_txg) will have a
2812 * positive refcnt -- either 1 or 2. We then convert
2813 * the reference tree into the new DTL_MISSING map.
2815 space_reftree_create(&reftree);
2816 space_reftree_add_map(&reftree,
2817 vd->vdev_dtl[DTL_MISSING], 1);
2818 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2819 space_reftree_add_map(&reftree,
2820 vd->vdev_dtl[DTL_SCRUB], 2);
2821 space_reftree_generate_map(&reftree,
2822 vd->vdev_dtl[DTL_MISSING], 1);
2823 space_reftree_destroy(&reftree);
2825 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
2826 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
2827 (u_longlong_t)vdev_dtl_min(vd),
2828 (u_longlong_t)vdev_dtl_max(vd));
2829 } else if (!wasempty) {
2830 zfs_dbgmsg("DTL_MISSING is now empty");
2833 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2834 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2835 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2837 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2838 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2839 if (!vdev_readable(vd))
2840 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2842 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2843 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2846 * If the vdev was resilvering or rebuilding and no longer
2847 * has any DTLs then reset the appropriate flag and dirty
2848 * the top level so that we persist the change.
2851 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2852 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2853 if (vd->vdev_rebuild_txg != 0) {
2854 vd->vdev_rebuild_txg = 0;
2855 vdev_config_dirty(vd->vdev_top);
2856 } else if (vd->vdev_resilver_txg != 0) {
2857 vd->vdev_resilver_txg = 0;
2858 vdev_config_dirty(vd->vdev_top);
2862 mutex_exit(&vd->vdev_dtl_lock);
2865 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2869 mutex_enter(&vd->vdev_dtl_lock);
2870 for (int t = 0; t < DTL_TYPES; t++) {
2871 /* account for child's outage in parent's missing map */
2872 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2874 continue; /* leaf vdevs only */
2875 if (t == DTL_PARTIAL)
2876 minref = 1; /* i.e. non-zero */
2877 else if (vd->vdev_nparity != 0)
2878 minref = vd->vdev_nparity + 1; /* RAID-Z */
2880 minref = vd->vdev_children; /* any kind of mirror */
2881 space_reftree_create(&reftree);
2882 for (int c = 0; c < vd->vdev_children; c++) {
2883 vdev_t *cvd = vd->vdev_child[c];
2884 mutex_enter(&cvd->vdev_dtl_lock);
2885 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2886 mutex_exit(&cvd->vdev_dtl_lock);
2888 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2889 space_reftree_destroy(&reftree);
2891 mutex_exit(&vd->vdev_dtl_lock);
2895 vdev_dtl_load(vdev_t *vd)
2897 spa_t *spa = vd->vdev_spa;
2898 objset_t *mos = spa->spa_meta_objset;
2901 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2902 ASSERT(vdev_is_concrete(vd));
2904 error = space_map_open(&vd->vdev_dtl_sm, mos,
2905 vd->vdev_dtl_object, 0, -1ULL, 0);
2908 ASSERT(vd->vdev_dtl_sm != NULL);
2910 mutex_enter(&vd->vdev_dtl_lock);
2911 error = space_map_load(vd->vdev_dtl_sm,
2912 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2913 mutex_exit(&vd->vdev_dtl_lock);
2918 for (int c = 0; c < vd->vdev_children; c++) {
2919 error = vdev_dtl_load(vd->vdev_child[c]);
2928 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
2930 spa_t *spa = vd->vdev_spa;
2931 objset_t *mos = spa->spa_meta_objset;
2932 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
2935 ASSERT(alloc_bias != VDEV_BIAS_NONE);
2938 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
2939 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
2940 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
2942 ASSERT(string != NULL);
2943 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
2944 1, strlen(string) + 1, string, tx));
2946 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
2947 spa_activate_allocation_classes(spa, tx);
2952 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2954 spa_t *spa = vd->vdev_spa;
2956 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2957 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2962 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2964 spa_t *spa = vd->vdev_spa;
2965 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2966 DMU_OT_NONE, 0, tx);
2969 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2976 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2978 if (vd->vdev_ops != &vdev_hole_ops &&
2979 vd->vdev_ops != &vdev_missing_ops &&
2980 vd->vdev_ops != &vdev_root_ops &&
2981 !vd->vdev_top->vdev_removing) {
2982 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2983 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2985 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2986 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2987 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
2988 vdev_zap_allocation_data(vd, tx);
2992 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2993 vdev_construct_zaps(vd->vdev_child[i], tx);
2998 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
3000 spa_t *spa = vd->vdev_spa;
3001 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
3002 objset_t *mos = spa->spa_meta_objset;
3003 range_tree_t *rtsync;
3005 uint64_t object = space_map_object(vd->vdev_dtl_sm);
3007 ASSERT(vdev_is_concrete(vd));
3008 ASSERT(vd->vdev_ops->vdev_op_leaf);
3010 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3012 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
3013 mutex_enter(&vd->vdev_dtl_lock);
3014 space_map_free(vd->vdev_dtl_sm, tx);
3015 space_map_close(vd->vdev_dtl_sm);
3016 vd->vdev_dtl_sm = NULL;
3017 mutex_exit(&vd->vdev_dtl_lock);
3020 * We only destroy the leaf ZAP for detached leaves or for
3021 * removed log devices. Removed data devices handle leaf ZAP
3022 * cleanup later, once cancellation is no longer possible.
3024 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
3025 vd->vdev_top->vdev_islog)) {
3026 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
3027 vd->vdev_leaf_zap = 0;
3034 if (vd->vdev_dtl_sm == NULL) {
3035 uint64_t new_object;
3037 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx);
3038 VERIFY3U(new_object, !=, 0);
3040 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
3042 ASSERT(vd->vdev_dtl_sm != NULL);
3045 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
3047 mutex_enter(&vd->vdev_dtl_lock);
3048 range_tree_walk(rt, range_tree_add, rtsync);
3049 mutex_exit(&vd->vdev_dtl_lock);
3051 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx);
3052 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
3053 range_tree_vacate(rtsync, NULL, NULL);
3055 range_tree_destroy(rtsync);
3058 * If the object for the space map has changed then dirty
3059 * the top level so that we update the config.
3061 if (object != space_map_object(vd->vdev_dtl_sm)) {
3062 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
3063 "new object %llu", (u_longlong_t)txg, spa_name(spa),
3064 (u_longlong_t)object,
3065 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
3066 vdev_config_dirty(vd->vdev_top);
3073 * Determine whether the specified vdev can be offlined/detached/removed
3074 * without losing data.
3077 vdev_dtl_required(vdev_t *vd)
3079 spa_t *spa = vd->vdev_spa;
3080 vdev_t *tvd = vd->vdev_top;
3081 uint8_t cant_read = vd->vdev_cant_read;
3084 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3086 if (vd == spa->spa_root_vdev || vd == tvd)
3090 * Temporarily mark the device as unreadable, and then determine
3091 * whether this results in any DTL outages in the top-level vdev.
3092 * If not, we can safely offline/detach/remove the device.
3094 vd->vdev_cant_read = B_TRUE;
3095 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3096 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
3097 vd->vdev_cant_read = cant_read;
3098 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3100 if (!required && zio_injection_enabled) {
3101 required = !!zio_handle_device_injection(vd, NULL,
3109 * Determine if resilver is needed, and if so the txg range.
3112 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
3114 boolean_t needed = B_FALSE;
3115 uint64_t thismin = UINT64_MAX;
3116 uint64_t thismax = 0;
3118 if (vd->vdev_children == 0) {
3119 mutex_enter(&vd->vdev_dtl_lock);
3120 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
3121 vdev_writeable(vd)) {
3123 thismin = vdev_dtl_min(vd);
3124 thismax = vdev_dtl_max(vd);
3127 mutex_exit(&vd->vdev_dtl_lock);
3129 for (int c = 0; c < vd->vdev_children; c++) {
3130 vdev_t *cvd = vd->vdev_child[c];
3131 uint64_t cmin, cmax;
3133 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
3134 thismin = MIN(thismin, cmin);
3135 thismax = MAX(thismax, cmax);
3141 if (needed && minp) {
3149 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3150 * will contain either the checkpoint spacemap object or zero if none exists.
3151 * All other errors are returned to the caller.
3154 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj)
3156 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
3158 if (vd->vdev_top_zap == 0) {
3163 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
3164 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj);
3165 if (error == ENOENT) {
3174 vdev_load(vdev_t *vd)
3179 * Recursively load all children.
3181 for (int c = 0; c < vd->vdev_children; c++) {
3182 error = vdev_load(vd->vdev_child[c]);
3188 vdev_set_deflate_ratio(vd);
3191 * On spa_load path, grab the allocation bias from our zap
3193 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3194 spa_t *spa = vd->vdev_spa;
3197 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
3198 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
3201 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
3202 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
3203 } else if (error != ENOENT) {
3204 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3205 VDEV_AUX_CORRUPT_DATA);
3206 vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) "
3207 "failed [error=%d]", vd->vdev_top_zap, error);
3213 * Load any rebuild state from the top-level vdev zap.
3215 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3216 error = vdev_rebuild_load(vd);
3217 if (error && error != ENOTSUP) {
3218 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3219 VDEV_AUX_CORRUPT_DATA);
3220 vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load "
3221 "failed [error=%d]", error);
3227 * If this is a top-level vdev, initialize its metaslabs.
3229 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
3230 vdev_metaslab_group_create(vd);
3232 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
3233 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3234 VDEV_AUX_CORRUPT_DATA);
3235 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
3236 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
3237 (u_longlong_t)vd->vdev_asize);
3238 return (SET_ERROR(ENXIO));
3241 error = vdev_metaslab_init(vd, 0);
3243 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
3244 "[error=%d]", error);
3245 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3246 VDEV_AUX_CORRUPT_DATA);
3250 uint64_t checkpoint_sm_obj;
3251 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj);
3252 if (error == 0 && checkpoint_sm_obj != 0) {
3253 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3254 ASSERT(vd->vdev_asize != 0);
3255 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
3257 error = space_map_open(&vd->vdev_checkpoint_sm,
3258 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
3261 vdev_dbgmsg(vd, "vdev_load: space_map_open "
3262 "failed for checkpoint spacemap (obj %llu) "
3264 (u_longlong_t)checkpoint_sm_obj, error);
3267 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
3270 * Since the checkpoint_sm contains free entries
3271 * exclusively we can use space_map_allocated() to
3272 * indicate the cumulative checkpointed space that
3275 vd->vdev_stat.vs_checkpoint_space =
3276 -space_map_allocated(vd->vdev_checkpoint_sm);
3277 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
3278 vd->vdev_stat.vs_checkpoint_space;
3279 } else if (error != 0) {
3280 vdev_dbgmsg(vd, "vdev_load: failed to retrieve "
3281 "checkpoint space map object from vdev ZAP "
3282 "[error=%d]", error);
3288 * If this is a leaf vdev, load its DTL.
3290 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
3291 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3292 VDEV_AUX_CORRUPT_DATA);
3293 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
3294 "[error=%d]", error);
3298 uint64_t obsolete_sm_object;
3299 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object);
3300 if (error == 0 && obsolete_sm_object != 0) {
3301 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3302 ASSERT(vd->vdev_asize != 0);
3303 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
3305 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
3306 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
3307 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3308 VDEV_AUX_CORRUPT_DATA);
3309 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
3310 "obsolete spacemap (obj %llu) [error=%d]",
3311 (u_longlong_t)obsolete_sm_object, error);
3314 } else if (error != 0) {
3315 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete "
3316 "space map object from vdev ZAP [error=%d]", error);
3324 * The special vdev case is used for hot spares and l2cache devices. Its
3325 * sole purpose it to set the vdev state for the associated vdev. To do this,
3326 * we make sure that we can open the underlying device, then try to read the
3327 * label, and make sure that the label is sane and that it hasn't been
3328 * repurposed to another pool.
3331 vdev_validate_aux(vdev_t *vd)
3334 uint64_t guid, version;
3337 if (!vdev_readable(vd))
3340 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3341 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3342 VDEV_AUX_CORRUPT_DATA);
3346 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3347 !SPA_VERSION_IS_SUPPORTED(version) ||
3348 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3349 guid != vd->vdev_guid ||
3350 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3351 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3352 VDEV_AUX_CORRUPT_DATA);
3358 * We don't actually check the pool state here. If it's in fact in
3359 * use by another pool, we update this fact on the fly when requested.
3366 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx)
3368 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3370 if (vd->vdev_top_zap == 0)
3373 uint64_t object = 0;
3374 int err = zap_lookup(mos, vd->vdev_top_zap,
3375 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object);
3380 VERIFY0(dmu_object_free(mos, object, tx));
3381 VERIFY0(zap_remove(mos, vd->vdev_top_zap,
3382 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx));
3386 * Free the objects used to store this vdev's spacemaps, and the array
3387 * that points to them.
3390 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3392 if (vd->vdev_ms_array == 0)
3395 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3396 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3397 size_t array_bytes = array_count * sizeof (uint64_t);
3398 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3399 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3400 array_bytes, smobj_array, 0));
3402 for (uint64_t i = 0; i < array_count; i++) {
3403 uint64_t smobj = smobj_array[i];
3407 space_map_free_obj(mos, smobj, tx);
3410 kmem_free(smobj_array, array_bytes);
3411 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3412 vdev_destroy_ms_flush_data(vd, tx);
3413 vd->vdev_ms_array = 0;
3417 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3419 spa_t *spa = vd->vdev_spa;
3421 ASSERT(vd->vdev_islog);
3422 ASSERT(vd == vd->vdev_top);
3423 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3425 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3427 vdev_destroy_spacemaps(vd, tx);
3428 if (vd->vdev_top_zap != 0) {
3429 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3430 vd->vdev_top_zap = 0;
3437 vdev_sync_done(vdev_t *vd, uint64_t txg)
3440 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3442 ASSERT(vdev_is_concrete(vd));
3444 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3446 metaslab_sync_done(msp, txg);
3449 metaslab_sync_reassess(vd->vdev_mg);
3453 vdev_sync(vdev_t *vd, uint64_t txg)
3455 spa_t *spa = vd->vdev_spa;
3459 ASSERT3U(txg, ==, spa->spa_syncing_txg);
3460 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3461 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3462 ASSERT(vd->vdev_removing ||
3463 vd->vdev_ops == &vdev_indirect_ops);
3465 vdev_indirect_sync_obsolete(vd, tx);
3468 * If the vdev is indirect, it can't have dirty
3469 * metaslabs or DTLs.
3471 if (vd->vdev_ops == &vdev_indirect_ops) {
3472 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3473 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3479 ASSERT(vdev_is_concrete(vd));
3481 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3482 !vd->vdev_removing) {
3483 ASSERT(vd == vd->vdev_top);
3484 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3485 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3486 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3487 ASSERT(vd->vdev_ms_array != 0);
3488 vdev_config_dirty(vd);
3491 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3492 metaslab_sync(msp, txg);
3493 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3496 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3497 vdev_dtl_sync(lvd, txg);
3500 * If this is an empty log device being removed, destroy the
3501 * metadata associated with it.
3503 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3504 vdev_remove_empty_log(vd, txg);
3506 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3511 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3513 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3517 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3518 * not be opened, and no I/O is attempted.
3521 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3525 spa_vdev_state_enter(spa, SCL_NONE);
3527 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3528 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3530 if (!vd->vdev_ops->vdev_op_leaf)
3531 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3536 * If user did a 'zpool offline -f' then make the fault persist across
3539 if (aux == VDEV_AUX_EXTERNAL_PERSIST) {
3541 * There are two kinds of forced faults: temporary and
3542 * persistent. Temporary faults go away at pool import, while
3543 * persistent faults stay set. Both types of faults can be
3544 * cleared with a zpool clear.
3546 * We tell if a vdev is persistently faulted by looking at the
3547 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3548 * import then it's a persistent fault. Otherwise, it's
3549 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3550 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3551 * tells vdev_config_generate() (which gets run later) to set
3552 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3554 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
3555 vd->vdev_tmpoffline = B_FALSE;
3556 aux = VDEV_AUX_EXTERNAL;
3558 vd->vdev_tmpoffline = B_TRUE;
3562 * We don't directly use the aux state here, but if we do a
3563 * vdev_reopen(), we need this value to be present to remember why we
3566 vd->vdev_label_aux = aux;
3569 * Faulted state takes precedence over degraded.
3571 vd->vdev_delayed_close = B_FALSE;
3572 vd->vdev_faulted = 1ULL;
3573 vd->vdev_degraded = 0ULL;
3574 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3577 * If this device has the only valid copy of the data, then
3578 * back off and simply mark the vdev as degraded instead.
3580 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3581 vd->vdev_degraded = 1ULL;
3582 vd->vdev_faulted = 0ULL;
3585 * If we reopen the device and it's not dead, only then do we
3590 if (vdev_readable(vd))
3591 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3594 return (spa_vdev_state_exit(spa, vd, 0));
3598 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3599 * user that something is wrong. The vdev continues to operate as normal as far
3600 * as I/O is concerned.
3603 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3607 spa_vdev_state_enter(spa, SCL_NONE);
3609 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3610 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3612 if (!vd->vdev_ops->vdev_op_leaf)
3613 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3616 * If the vdev is already faulted, then don't do anything.
3618 if (vd->vdev_faulted || vd->vdev_degraded)
3619 return (spa_vdev_state_exit(spa, NULL, 0));
3621 vd->vdev_degraded = 1ULL;
3622 if (!vdev_is_dead(vd))
3623 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3626 return (spa_vdev_state_exit(spa, vd, 0));
3630 * Online the given vdev.
3632 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3633 * spare device should be detached when the device finishes resilvering.
3634 * Second, the online should be treated like a 'test' online case, so no FMA
3635 * events are generated if the device fails to open.
3638 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3640 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3641 boolean_t wasoffline;
3642 vdev_state_t oldstate;
3644 spa_vdev_state_enter(spa, SCL_NONE);
3646 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3647 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3649 if (!vd->vdev_ops->vdev_op_leaf)
3650 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3652 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3653 oldstate = vd->vdev_state;
3656 vd->vdev_offline = B_FALSE;
3657 vd->vdev_tmpoffline = B_FALSE;
3658 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3659 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3661 /* XXX - L2ARC 1.0 does not support expansion */
3662 if (!vd->vdev_aux) {
3663 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3664 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
3665 spa->spa_autoexpand);
3666 vd->vdev_expansion_time = gethrestime_sec();
3670 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3672 if (!vd->vdev_aux) {
3673 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3674 pvd->vdev_expanding = B_FALSE;
3678 *newstate = vd->vdev_state;
3679 if ((flags & ZFS_ONLINE_UNSPARE) &&
3680 !vdev_is_dead(vd) && vd->vdev_parent &&
3681 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3682 vd->vdev_parent->vdev_child[0] == vd)
3683 vd->vdev_unspare = B_TRUE;
3685 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3687 /* XXX - L2ARC 1.0 does not support expansion */
3689 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3690 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3693 /* Restart initializing if necessary */
3694 mutex_enter(&vd->vdev_initialize_lock);
3695 if (vdev_writeable(vd) &&
3696 vd->vdev_initialize_thread == NULL &&
3697 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3698 (void) vdev_initialize(vd);
3700 mutex_exit(&vd->vdev_initialize_lock);
3703 * Restart trimming if necessary. We do not restart trimming for cache
3704 * devices here. This is triggered by l2arc_rebuild_vdev()
3705 * asynchronously for the whole device or in l2arc_evict() as it evicts
3706 * space for upcoming writes.
3708 mutex_enter(&vd->vdev_trim_lock);
3709 if (vdev_writeable(vd) && !vd->vdev_isl2cache &&
3710 vd->vdev_trim_thread == NULL &&
3711 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) {
3712 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial,
3713 vd->vdev_trim_secure);
3715 mutex_exit(&vd->vdev_trim_lock);
3718 (oldstate < VDEV_STATE_DEGRADED &&
3719 vd->vdev_state >= VDEV_STATE_DEGRADED))
3720 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3722 return (spa_vdev_state_exit(spa, vd, 0));
3726 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3730 uint64_t generation;
3731 metaslab_group_t *mg;
3734 spa_vdev_state_enter(spa, SCL_ALLOC);
3736 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3737 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3739 if (!vd->vdev_ops->vdev_op_leaf)
3740 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3744 generation = spa->spa_config_generation + 1;
3747 * If the device isn't already offline, try to offline it.
3749 if (!vd->vdev_offline) {
3751 * If this device has the only valid copy of some data,
3752 * don't allow it to be offlined. Log devices are always
3755 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3756 vdev_dtl_required(vd))
3757 return (spa_vdev_state_exit(spa, NULL,
3761 * If the top-level is a slog and it has had allocations
3762 * then proceed. We check that the vdev's metaslab group
3763 * is not NULL since it's possible that we may have just
3764 * added this vdev but not yet initialized its metaslabs.
3766 if (tvd->vdev_islog && mg != NULL) {
3768 * Prevent any future allocations.
3770 metaslab_group_passivate(mg);
3771 (void) spa_vdev_state_exit(spa, vd, 0);
3773 error = spa_reset_logs(spa);
3776 * If the log device was successfully reset but has
3777 * checkpointed data, do not offline it.
3780 tvd->vdev_checkpoint_sm != NULL) {
3781 ASSERT3U(space_map_allocated(
3782 tvd->vdev_checkpoint_sm), !=, 0);
3783 error = ZFS_ERR_CHECKPOINT_EXISTS;
3786 spa_vdev_state_enter(spa, SCL_ALLOC);
3789 * Check to see if the config has changed.
3791 if (error || generation != spa->spa_config_generation) {
3792 metaslab_group_activate(mg);
3794 return (spa_vdev_state_exit(spa,
3796 (void) spa_vdev_state_exit(spa, vd, 0);
3799 ASSERT0(tvd->vdev_stat.vs_alloc);
3803 * Offline this device and reopen its top-level vdev.
3804 * If the top-level vdev is a log device then just offline
3805 * it. Otherwise, if this action results in the top-level
3806 * vdev becoming unusable, undo it and fail the request.
3808 vd->vdev_offline = B_TRUE;
3811 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3812 vdev_is_dead(tvd)) {
3813 vd->vdev_offline = B_FALSE;
3815 return (spa_vdev_state_exit(spa, NULL,
3820 * Add the device back into the metaslab rotor so that
3821 * once we online the device it's open for business.
3823 if (tvd->vdev_islog && mg != NULL)
3824 metaslab_group_activate(mg);
3827 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3829 return (spa_vdev_state_exit(spa, vd, 0));
3833 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3837 mutex_enter(&spa->spa_vdev_top_lock);
3838 error = vdev_offline_locked(spa, guid, flags);
3839 mutex_exit(&spa->spa_vdev_top_lock);
3845 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3846 * vdev_offline(), we assume the spa config is locked. We also clear all
3847 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3850 vdev_clear(spa_t *spa, vdev_t *vd)
3852 vdev_t *rvd = spa->spa_root_vdev;
3854 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3859 vd->vdev_stat.vs_read_errors = 0;
3860 vd->vdev_stat.vs_write_errors = 0;
3861 vd->vdev_stat.vs_checksum_errors = 0;
3862 vd->vdev_stat.vs_slow_ios = 0;
3864 for (int c = 0; c < vd->vdev_children; c++)
3865 vdev_clear(spa, vd->vdev_child[c]);
3868 * It makes no sense to "clear" an indirect vdev.
3870 if (!vdev_is_concrete(vd))
3874 * If we're in the FAULTED state or have experienced failed I/O, then
3875 * clear the persistent state and attempt to reopen the device. We
3876 * also mark the vdev config dirty, so that the new faulted state is
3877 * written out to disk.
3879 if (vd->vdev_faulted || vd->vdev_degraded ||
3880 !vdev_readable(vd) || !vdev_writeable(vd)) {
3882 * When reopening in response to a clear event, it may be due to
3883 * a fmadm repair request. In this case, if the device is
3884 * still broken, we want to still post the ereport again.
3886 vd->vdev_forcefault = B_TRUE;
3888 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3889 vd->vdev_cant_read = B_FALSE;
3890 vd->vdev_cant_write = B_FALSE;
3891 vd->vdev_stat.vs_aux = 0;
3893 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3895 vd->vdev_forcefault = B_FALSE;
3897 if (vd != rvd && vdev_writeable(vd->vdev_top))
3898 vdev_state_dirty(vd->vdev_top);
3900 /* If a resilver isn't required, check if vdevs can be culled */
3901 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) &&
3902 !dsl_scan_resilvering(spa->spa_dsl_pool) &&
3903 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool))
3904 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3906 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3910 * When clearing a FMA-diagnosed fault, we always want to
3911 * unspare the device, as we assume that the original spare was
3912 * done in response to the FMA fault.
3914 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3915 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3916 vd->vdev_parent->vdev_child[0] == vd)
3917 vd->vdev_unspare = B_TRUE;
3921 vdev_is_dead(vdev_t *vd)
3924 * Holes and missing devices are always considered "dead".
3925 * This simplifies the code since we don't have to check for
3926 * these types of devices in the various code paths.
3927 * Instead we rely on the fact that we skip over dead devices
3928 * before issuing I/O to them.
3930 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3931 vd->vdev_ops == &vdev_hole_ops ||
3932 vd->vdev_ops == &vdev_missing_ops);
3936 vdev_readable(vdev_t *vd)
3938 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3942 vdev_writeable(vdev_t *vd)
3944 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3945 vdev_is_concrete(vd));
3949 vdev_allocatable(vdev_t *vd)
3951 uint64_t state = vd->vdev_state;
3954 * We currently allow allocations from vdevs which may be in the
3955 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3956 * fails to reopen then we'll catch it later when we're holding
3957 * the proper locks. Note that we have to get the vdev state
3958 * in a local variable because although it changes atomically,
3959 * we're asking two separate questions about it.
3961 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3962 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3963 vd->vdev_mg->mg_initialized);
3967 vdev_accessible(vdev_t *vd, zio_t *zio)
3969 ASSERT(zio->io_vd == vd);
3971 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3974 if (zio->io_type == ZIO_TYPE_READ)
3975 return (!vd->vdev_cant_read);
3977 if (zio->io_type == ZIO_TYPE_WRITE)
3978 return (!vd->vdev_cant_write);
3984 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs)
3986 for (int t = 0; t < VS_ZIO_TYPES; t++) {
3987 vs->vs_ops[t] += cvs->vs_ops[t];
3988 vs->vs_bytes[t] += cvs->vs_bytes[t];
3991 cvs->vs_scan_removing = cvd->vdev_removing;
3995 * Get extended stats
3998 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx)
4001 for (t = 0; t < ZIO_TYPES; t++) {
4002 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++)
4003 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b];
4005 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) {
4006 vsx->vsx_total_histo[t][b] +=
4007 cvsx->vsx_total_histo[t][b];
4011 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
4012 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) {
4013 vsx->vsx_queue_histo[t][b] +=
4014 cvsx->vsx_queue_histo[t][b];
4016 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t];
4017 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t];
4019 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++)
4020 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b];
4022 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++)
4023 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b];
4029 vdev_is_spacemap_addressable(vdev_t *vd)
4031 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
4035 * If double-word space map entries are not enabled we assume
4036 * 47 bits of the space map entry are dedicated to the entry's
4037 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4038 * to calculate the maximum address that can be described by a
4039 * space map entry for the given device.
4041 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
4043 if (shift >= 63) /* detect potential overflow */
4046 return (vd->vdev_asize < (1ULL << shift));
4050 * Get statistics for the given vdev.
4053 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4057 * If we're getting stats on the root vdev, aggregate the I/O counts
4058 * over all top-level vdevs (i.e. the direct children of the root).
4060 if (!vd->vdev_ops->vdev_op_leaf) {
4062 memset(vs->vs_ops, 0, sizeof (vs->vs_ops));
4063 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes));
4066 memset(vsx, 0, sizeof (*vsx));
4068 for (int c = 0; c < vd->vdev_children; c++) {
4069 vdev_t *cvd = vd->vdev_child[c];
4070 vdev_stat_t *cvs = &cvd->vdev_stat;
4071 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex;
4073 vdev_get_stats_ex_impl(cvd, cvs, cvsx);
4075 vdev_get_child_stat(cvd, vs, cvs);
4077 vdev_get_child_stat_ex(cvd, vsx, cvsx);
4082 * We're a leaf. Just copy our ZIO active queue stats in. The
4083 * other leaf stats are updated in vdev_stat_update().
4088 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex));
4090 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) {
4091 vsx->vsx_active_queue[t] =
4092 vd->vdev_queue.vq_class[t].vqc_active;
4093 vsx->vsx_pend_queue[t] = avl_numnodes(
4094 &vd->vdev_queue.vq_class[t].vqc_queued_tree);
4100 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4102 vdev_t *tvd = vd->vdev_top;
4103 mutex_enter(&vd->vdev_stat_lock);
4105 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
4106 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
4107 vs->vs_state = vd->vdev_state;
4108 vs->vs_rsize = vdev_get_min_asize(vd);
4110 if (vd->vdev_ops->vdev_op_leaf) {
4111 vs->vs_rsize += VDEV_LABEL_START_SIZE +
4112 VDEV_LABEL_END_SIZE;
4114 * Report initializing progress. Since we don't
4115 * have the initializing locks held, this is only
4116 * an estimate (although a fairly accurate one).
4118 vs->vs_initialize_bytes_done =
4119 vd->vdev_initialize_bytes_done;
4120 vs->vs_initialize_bytes_est =
4121 vd->vdev_initialize_bytes_est;
4122 vs->vs_initialize_state = vd->vdev_initialize_state;
4123 vs->vs_initialize_action_time =
4124 vd->vdev_initialize_action_time;
4127 * Report manual TRIM progress. Since we don't have
4128 * the manual TRIM locks held, this is only an
4129 * estimate (although fairly accurate one).
4131 vs->vs_trim_notsup = !vd->vdev_has_trim;
4132 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done;
4133 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est;
4134 vs->vs_trim_state = vd->vdev_trim_state;
4135 vs->vs_trim_action_time = vd->vdev_trim_action_time;
4137 /* Set when there is a deferred resilver. */
4138 vs->vs_resilver_deferred = vd->vdev_resilver_deferred;
4142 * Report expandable space on top-level, non-auxiliary devices
4143 * only. The expandable space is reported in terms of metaslab
4144 * sized units since that determines how much space the pool
4147 if (vd->vdev_aux == NULL && tvd != NULL) {
4148 vs->vs_esize = P2ALIGN(
4149 vd->vdev_max_asize - vd->vdev_asize,
4150 1ULL << tvd->vdev_ms_shift);
4153 vs->vs_configured_ashift = vd->vdev_top != NULL
4154 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
4155 vs->vs_logical_ashift = vd->vdev_logical_ashift;
4156 vs->vs_physical_ashift = vd->vdev_physical_ashift;
4159 * Report fragmentation and rebuild progress for top-level,
4160 * non-auxiliary, concrete devices.
4162 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
4163 vdev_is_concrete(vd)) {
4164 vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
4165 vd->vdev_mg->mg_fragmentation : 0;
4169 vdev_get_stats_ex_impl(vd, vs, vsx);
4170 mutex_exit(&vd->vdev_stat_lock);
4174 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
4176 return (vdev_get_stats_ex(vd, vs, NULL));
4180 vdev_clear_stats(vdev_t *vd)
4182 mutex_enter(&vd->vdev_stat_lock);
4183 vd->vdev_stat.vs_space = 0;
4184 vd->vdev_stat.vs_dspace = 0;
4185 vd->vdev_stat.vs_alloc = 0;
4186 mutex_exit(&vd->vdev_stat_lock);
4190 vdev_scan_stat_init(vdev_t *vd)
4192 vdev_stat_t *vs = &vd->vdev_stat;
4194 for (int c = 0; c < vd->vdev_children; c++)
4195 vdev_scan_stat_init(vd->vdev_child[c]);
4197 mutex_enter(&vd->vdev_stat_lock);
4198 vs->vs_scan_processed = 0;
4199 mutex_exit(&vd->vdev_stat_lock);
4203 vdev_stat_update(zio_t *zio, uint64_t psize)
4205 spa_t *spa = zio->io_spa;
4206 vdev_t *rvd = spa->spa_root_vdev;
4207 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
4209 uint64_t txg = zio->io_txg;
4210 vdev_stat_t *vs = &vd->vdev_stat;
4211 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex;
4212 zio_type_t type = zio->io_type;
4213 int flags = zio->io_flags;
4216 * If this i/o is a gang leader, it didn't do any actual work.
4218 if (zio->io_gang_tree)
4221 if (zio->io_error == 0) {
4223 * If this is a root i/o, don't count it -- we've already
4224 * counted the top-level vdevs, and vdev_get_stats() will
4225 * aggregate them when asked. This reduces contention on
4226 * the root vdev_stat_lock and implicitly handles blocks
4227 * that compress away to holes, for which there is no i/o.
4228 * (Holes never create vdev children, so all the counters
4229 * remain zero, which is what we want.)
4231 * Note: this only applies to successful i/o (io_error == 0)
4232 * because unlike i/o counts, errors are not additive.
4233 * When reading a ditto block, for example, failure of
4234 * one top-level vdev does not imply a root-level error.
4239 ASSERT(vd == zio->io_vd);
4241 if (flags & ZIO_FLAG_IO_BYPASS)
4244 mutex_enter(&vd->vdev_stat_lock);
4246 if (flags & ZIO_FLAG_IO_REPAIR) {
4248 * Repair is the result of a resilver issued by the
4249 * scan thread (spa_sync).
4251 if (flags & ZIO_FLAG_SCAN_THREAD) {
4252 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
4253 dsl_scan_phys_t *scn_phys = &scn->scn_phys;
4254 uint64_t *processed = &scn_phys->scn_processed;
4256 if (vd->vdev_ops->vdev_op_leaf)
4257 atomic_add_64(processed, psize);
4258 vs->vs_scan_processed += psize;
4262 * Repair is the result of a rebuild issued by the
4263 * rebuild thread (vdev_rebuild_thread).
4265 if (zio->io_priority == ZIO_PRIORITY_REBUILD) {
4266 vdev_t *tvd = vd->vdev_top;
4267 vdev_rebuild_t *vr = &tvd->vdev_rebuild_config;
4268 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
4269 uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt;
4271 if (vd->vdev_ops->vdev_op_leaf)
4272 atomic_add_64(rebuilt, psize);
4273 vs->vs_rebuild_processed += psize;
4276 if (flags & ZIO_FLAG_SELF_HEAL)
4277 vs->vs_self_healed += psize;
4281 * The bytes/ops/histograms are recorded at the leaf level and
4282 * aggregated into the higher level vdevs in vdev_get_stats().
4284 if (vd->vdev_ops->vdev_op_leaf &&
4285 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) {
4286 zio_type_t vs_type = type;
4287 zio_priority_t priority = zio->io_priority;
4290 * TRIM ops and bytes are reported to user space as
4291 * ZIO_TYPE_IOCTL. This is done to preserve the
4292 * vdev_stat_t structure layout for user space.
4294 if (type == ZIO_TYPE_TRIM)
4295 vs_type = ZIO_TYPE_IOCTL;
4298 * Solely for the purposes of 'zpool iostat -lqrw'
4299 * reporting use the priority to catagorize the IO.
4300 * Only the following are reported to user space:
4302 * ZIO_PRIORITY_SYNC_READ,
4303 * ZIO_PRIORITY_SYNC_WRITE,
4304 * ZIO_PRIORITY_ASYNC_READ,
4305 * ZIO_PRIORITY_ASYNC_WRITE,
4306 * ZIO_PRIORITY_SCRUB,
4307 * ZIO_PRIORITY_TRIM.
4309 if (priority == ZIO_PRIORITY_REBUILD) {
4310 priority = ((type == ZIO_TYPE_WRITE) ?
4311 ZIO_PRIORITY_ASYNC_WRITE :
4312 ZIO_PRIORITY_SCRUB);
4313 } else if (priority == ZIO_PRIORITY_INITIALIZING) {
4314 ASSERT3U(type, ==, ZIO_TYPE_WRITE);
4315 priority = ZIO_PRIORITY_ASYNC_WRITE;
4316 } else if (priority == ZIO_PRIORITY_REMOVAL) {
4317 priority = ((type == ZIO_TYPE_WRITE) ?
4318 ZIO_PRIORITY_ASYNC_WRITE :
4319 ZIO_PRIORITY_ASYNC_READ);
4322 vs->vs_ops[vs_type]++;
4323 vs->vs_bytes[vs_type] += psize;
4325 if (flags & ZIO_FLAG_DELEGATED) {
4326 vsx->vsx_agg_histo[priority]
4327 [RQ_HISTO(zio->io_size)]++;
4329 vsx->vsx_ind_histo[priority]
4330 [RQ_HISTO(zio->io_size)]++;
4333 if (zio->io_delta && zio->io_delay) {
4334 vsx->vsx_queue_histo[priority]
4335 [L_HISTO(zio->io_delta - zio->io_delay)]++;
4336 vsx->vsx_disk_histo[type]
4337 [L_HISTO(zio->io_delay)]++;
4338 vsx->vsx_total_histo[type]
4339 [L_HISTO(zio->io_delta)]++;
4343 mutex_exit(&vd->vdev_stat_lock);
4347 if (flags & ZIO_FLAG_SPECULATIVE)
4351 * If this is an I/O error that is going to be retried, then ignore the
4352 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4353 * hard errors, when in reality they can happen for any number of
4354 * innocuous reasons (bus resets, MPxIO link failure, etc).
4356 if (zio->io_error == EIO &&
4357 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
4361 * Intent logs writes won't propagate their error to the root
4362 * I/O so don't mark these types of failures as pool-level
4365 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
4368 if (spa->spa_load_state == SPA_LOAD_NONE &&
4369 type == ZIO_TYPE_WRITE && txg != 0 &&
4370 (!(flags & ZIO_FLAG_IO_REPAIR) ||
4371 (flags & ZIO_FLAG_SCAN_THREAD) ||
4372 spa->spa_claiming)) {
4374 * This is either a normal write (not a repair), or it's
4375 * a repair induced by the scrub thread, or it's a repair
4376 * made by zil_claim() during spa_load() in the first txg.
4377 * In the normal case, we commit the DTL change in the same
4378 * txg as the block was born. In the scrub-induced repair
4379 * case, we know that scrubs run in first-pass syncing context,
4380 * so we commit the DTL change in spa_syncing_txg(spa).
4381 * In the zil_claim() case, we commit in spa_first_txg(spa).
4383 * We currently do not make DTL entries for failed spontaneous
4384 * self-healing writes triggered by normal (non-scrubbing)
4385 * reads, because we have no transactional context in which to
4386 * do so -- and it's not clear that it'd be desirable anyway.
4388 if (vd->vdev_ops->vdev_op_leaf) {
4389 uint64_t commit_txg = txg;
4390 if (flags & ZIO_FLAG_SCAN_THREAD) {
4391 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4392 ASSERT(spa_sync_pass(spa) == 1);
4393 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
4394 commit_txg = spa_syncing_txg(spa);
4395 } else if (spa->spa_claiming) {
4396 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4397 commit_txg = spa_first_txg(spa);
4399 ASSERT(commit_txg >= spa_syncing_txg(spa));
4400 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
4402 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
4403 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
4404 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
4407 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
4412 vdev_deflated_space(vdev_t *vd, int64_t space)
4414 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
4415 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
4417 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
4421 * Update the in-core space usage stats for this vdev, its metaslab class,
4422 * and the root vdev.
4425 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
4426 int64_t space_delta)
4428 int64_t dspace_delta;
4429 spa_t *spa = vd->vdev_spa;
4430 vdev_t *rvd = spa->spa_root_vdev;
4432 ASSERT(vd == vd->vdev_top);
4435 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4436 * factor. We must calculate this here and not at the root vdev
4437 * because the root vdev's psize-to-asize is simply the max of its
4438 * children's, thus not accurate enough for us.
4440 dspace_delta = vdev_deflated_space(vd, space_delta);
4442 mutex_enter(&vd->vdev_stat_lock);
4443 /* ensure we won't underflow */
4444 if (alloc_delta < 0) {
4445 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta);
4448 vd->vdev_stat.vs_alloc += alloc_delta;
4449 vd->vdev_stat.vs_space += space_delta;
4450 vd->vdev_stat.vs_dspace += dspace_delta;
4451 mutex_exit(&vd->vdev_stat_lock);
4453 /* every class but log contributes to root space stats */
4454 if (vd->vdev_mg != NULL && !vd->vdev_islog) {
4455 ASSERT(!vd->vdev_isl2cache);
4456 mutex_enter(&rvd->vdev_stat_lock);
4457 rvd->vdev_stat.vs_alloc += alloc_delta;
4458 rvd->vdev_stat.vs_space += space_delta;
4459 rvd->vdev_stat.vs_dspace += dspace_delta;
4460 mutex_exit(&rvd->vdev_stat_lock);
4462 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4466 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4467 * so that it will be written out next time the vdev configuration is synced.
4468 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4471 vdev_config_dirty(vdev_t *vd)
4473 spa_t *spa = vd->vdev_spa;
4474 vdev_t *rvd = spa->spa_root_vdev;
4477 ASSERT(spa_writeable(spa));
4480 * If this is an aux vdev (as with l2cache and spare devices), then we
4481 * update the vdev config manually and set the sync flag.
4483 if (vd->vdev_aux != NULL) {
4484 spa_aux_vdev_t *sav = vd->vdev_aux;
4488 for (c = 0; c < sav->sav_count; c++) {
4489 if (sav->sav_vdevs[c] == vd)
4493 if (c == sav->sav_count) {
4495 * We're being removed. There's nothing more to do.
4497 ASSERT(sav->sav_sync == B_TRUE);
4501 sav->sav_sync = B_TRUE;
4503 if (nvlist_lookup_nvlist_array(sav->sav_config,
4504 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
4505 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
4506 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
4512 * Setting the nvlist in the middle if the array is a little
4513 * sketchy, but it will work.
4515 nvlist_free(aux[c]);
4516 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
4522 * The dirty list is protected by the SCL_CONFIG lock. The caller
4523 * must either hold SCL_CONFIG as writer, or must be the sync thread
4524 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4525 * so this is sufficient to ensure mutual exclusion.
4527 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4528 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4529 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4532 for (c = 0; c < rvd->vdev_children; c++)
4533 vdev_config_dirty(rvd->vdev_child[c]);
4535 ASSERT(vd == vd->vdev_top);
4537 if (!list_link_active(&vd->vdev_config_dirty_node) &&
4538 vdev_is_concrete(vd)) {
4539 list_insert_head(&spa->spa_config_dirty_list, vd);
4545 vdev_config_clean(vdev_t *vd)
4547 spa_t *spa = vd->vdev_spa;
4549 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4550 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4551 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4553 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
4554 list_remove(&spa->spa_config_dirty_list, vd);
4558 * Mark a top-level vdev's state as dirty, so that the next pass of
4559 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4560 * the state changes from larger config changes because they require
4561 * much less locking, and are often needed for administrative actions.
4564 vdev_state_dirty(vdev_t *vd)
4566 spa_t *spa = vd->vdev_spa;
4568 ASSERT(spa_writeable(spa));
4569 ASSERT(vd == vd->vdev_top);
4572 * The state list is protected by the SCL_STATE lock. The caller
4573 * must either hold SCL_STATE as writer, or must be the sync thread
4574 * (which holds SCL_STATE as reader). There's only one sync thread,
4575 * so this is sufficient to ensure mutual exclusion.
4577 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4578 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4579 spa_config_held(spa, SCL_STATE, RW_READER)));
4581 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4582 vdev_is_concrete(vd))
4583 list_insert_head(&spa->spa_state_dirty_list, vd);
4587 vdev_state_clean(vdev_t *vd)
4589 spa_t *spa = vd->vdev_spa;
4591 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4592 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4593 spa_config_held(spa, SCL_STATE, RW_READER)));
4595 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4596 list_remove(&spa->spa_state_dirty_list, vd);
4600 * Propagate vdev state up from children to parent.
4603 vdev_propagate_state(vdev_t *vd)
4605 spa_t *spa = vd->vdev_spa;
4606 vdev_t *rvd = spa->spa_root_vdev;
4607 int degraded = 0, faulted = 0;
4611 if (vd->vdev_children > 0) {
4612 for (int c = 0; c < vd->vdev_children; c++) {
4613 child = vd->vdev_child[c];
4616 * Don't factor holes or indirect vdevs into the
4619 if (!vdev_is_concrete(child))
4622 if (!vdev_readable(child) ||
4623 (!vdev_writeable(child) && spa_writeable(spa))) {
4625 * Root special: if there is a top-level log
4626 * device, treat the root vdev as if it were
4629 if (child->vdev_islog && vd == rvd)
4633 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4637 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4641 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4644 * Root special: if there is a top-level vdev that cannot be
4645 * opened due to corrupted metadata, then propagate the root
4646 * vdev's aux state as 'corrupt' rather than 'insufficient
4649 if (corrupted && vd == rvd &&
4650 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4651 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4652 VDEV_AUX_CORRUPT_DATA);
4655 if (vd->vdev_parent)
4656 vdev_propagate_state(vd->vdev_parent);
4660 * Set a vdev's state. If this is during an open, we don't update the parent
4661 * state, because we're in the process of opening children depth-first.
4662 * Otherwise, we propagate the change to the parent.
4664 * If this routine places a device in a faulted state, an appropriate ereport is
4668 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4670 uint64_t save_state;
4671 spa_t *spa = vd->vdev_spa;
4673 if (state == vd->vdev_state) {
4675 * Since vdev_offline() code path is already in an offline
4676 * state we can miss a statechange event to OFFLINE. Check
4677 * the previous state to catch this condition.
4679 if (vd->vdev_ops->vdev_op_leaf &&
4680 (state == VDEV_STATE_OFFLINE) &&
4681 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) {
4682 /* post an offline state change */
4683 zfs_post_state_change(spa, vd, vd->vdev_prevstate);
4685 vd->vdev_stat.vs_aux = aux;
4689 save_state = vd->vdev_state;
4691 vd->vdev_state = state;
4692 vd->vdev_stat.vs_aux = aux;
4695 * If we are setting the vdev state to anything but an open state, then
4696 * always close the underlying device unless the device has requested
4697 * a delayed close (i.e. we're about to remove or fault the device).
4698 * Otherwise, we keep accessible but invalid devices open forever.
4699 * We don't call vdev_close() itself, because that implies some extra
4700 * checks (offline, etc) that we don't want here. This is limited to
4701 * leaf devices, because otherwise closing the device will affect other
4704 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4705 vd->vdev_ops->vdev_op_leaf)
4706 vd->vdev_ops->vdev_op_close(vd);
4708 if (vd->vdev_removed &&
4709 state == VDEV_STATE_CANT_OPEN &&
4710 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4712 * If the previous state is set to VDEV_STATE_REMOVED, then this
4713 * device was previously marked removed and someone attempted to
4714 * reopen it. If this failed due to a nonexistent device, then
4715 * keep the device in the REMOVED state. We also let this be if
4716 * it is one of our special test online cases, which is only
4717 * attempting to online the device and shouldn't generate an FMA
4720 vd->vdev_state = VDEV_STATE_REMOVED;
4721 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4722 } else if (state == VDEV_STATE_REMOVED) {
4723 vd->vdev_removed = B_TRUE;
4724 } else if (state == VDEV_STATE_CANT_OPEN) {
4726 * If we fail to open a vdev during an import or recovery, we
4727 * mark it as "not available", which signifies that it was
4728 * never there to begin with. Failure to open such a device
4729 * is not considered an error.
4731 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4732 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4733 vd->vdev_ops->vdev_op_leaf)
4734 vd->vdev_not_present = 1;
4737 * Post the appropriate ereport. If the 'prevstate' field is
4738 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4739 * that this is part of a vdev_reopen(). In this case, we don't
4740 * want to post the ereport if the device was already in the
4741 * CANT_OPEN state beforehand.
4743 * If the 'checkremove' flag is set, then this is an attempt to
4744 * online the device in response to an insertion event. If we
4745 * hit this case, then we have detected an insertion event for a
4746 * faulted or offline device that wasn't in the removed state.
4747 * In this scenario, we don't post an ereport because we are
4748 * about to replace the device, or attempt an online with
4749 * vdev_forcefault, which will generate the fault for us.
4751 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4752 !vd->vdev_not_present && !vd->vdev_checkremove &&
4753 vd != spa->spa_root_vdev) {
4757 case VDEV_AUX_OPEN_FAILED:
4758 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4760 case VDEV_AUX_CORRUPT_DATA:
4761 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4763 case VDEV_AUX_NO_REPLICAS:
4764 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4766 case VDEV_AUX_BAD_GUID_SUM:
4767 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4769 case VDEV_AUX_TOO_SMALL:
4770 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4772 case VDEV_AUX_BAD_LABEL:
4773 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4775 case VDEV_AUX_BAD_ASHIFT:
4776 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT;
4779 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4782 (void) zfs_ereport_post(class, spa, vd, NULL, NULL,
4786 /* Erase any notion of persistent removed state */
4787 vd->vdev_removed = B_FALSE;
4789 vd->vdev_removed = B_FALSE;
4793 * Notify ZED of any significant state-change on a leaf vdev.
4796 if (vd->vdev_ops->vdev_op_leaf) {
4797 /* preserve original state from a vdev_reopen() */
4798 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) &&
4799 (vd->vdev_prevstate != vd->vdev_state) &&
4800 (save_state <= VDEV_STATE_CLOSED))
4801 save_state = vd->vdev_prevstate;
4803 /* filter out state change due to initial vdev_open */
4804 if (save_state > VDEV_STATE_CLOSED)
4805 zfs_post_state_change(spa, vd, save_state);
4808 if (!isopen && vd->vdev_parent)
4809 vdev_propagate_state(vd->vdev_parent);
4813 vdev_children_are_offline(vdev_t *vd)
4815 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4817 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4818 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4826 * Check the vdev configuration to ensure that it's capable of supporting
4827 * a root pool. We do not support partial configuration.
4830 vdev_is_bootable(vdev_t *vd)
4832 if (!vd->vdev_ops->vdev_op_leaf) {
4833 const char *vdev_type = vd->vdev_ops->vdev_op_type;
4835 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4836 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4841 for (int c = 0; c < vd->vdev_children; c++) {
4842 if (!vdev_is_bootable(vd->vdev_child[c]))
4849 vdev_is_concrete(vdev_t *vd)
4851 vdev_ops_t *ops = vd->vdev_ops;
4852 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4853 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4861 * Determine if a log device has valid content. If the vdev was
4862 * removed or faulted in the MOS config then we know that
4863 * the content on the log device has already been written to the pool.
4866 vdev_log_state_valid(vdev_t *vd)
4868 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4872 for (int c = 0; c < vd->vdev_children; c++)
4873 if (vdev_log_state_valid(vd->vdev_child[c]))
4880 * Expand a vdev if possible.
4883 vdev_expand(vdev_t *vd, uint64_t txg)
4885 ASSERT(vd->vdev_top == vd);
4886 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4887 ASSERT(vdev_is_concrete(vd));
4889 vdev_set_deflate_ratio(vd);
4891 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
4892 vdev_is_concrete(vd)) {
4893 vdev_metaslab_group_create(vd);
4894 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4895 vdev_config_dirty(vd);
4903 vdev_split(vdev_t *vd)
4905 vdev_t *cvd, *pvd = vd->vdev_parent;
4907 vdev_remove_child(pvd, vd);
4908 vdev_compact_children(pvd);
4910 cvd = pvd->vdev_child[0];
4911 if (pvd->vdev_children == 1) {
4912 vdev_remove_parent(cvd);
4913 cvd->vdev_splitting = B_TRUE;
4915 vdev_propagate_state(cvd);
4919 vdev_deadman(vdev_t *vd, char *tag)
4921 for (int c = 0; c < vd->vdev_children; c++) {
4922 vdev_t *cvd = vd->vdev_child[c];
4924 vdev_deadman(cvd, tag);
4927 if (vd->vdev_ops->vdev_op_leaf) {
4928 vdev_queue_t *vq = &vd->vdev_queue;
4930 mutex_enter(&vq->vq_lock);
4931 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4932 spa_t *spa = vd->vdev_spa;
4936 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4937 vd->vdev_path, avl_numnodes(&vq->vq_active_tree));
4940 * Look at the head of all the pending queues,
4941 * if any I/O has been outstanding for longer than
4942 * the spa_deadman_synctime invoke the deadman logic.
4944 fio = avl_first(&vq->vq_active_tree);
4945 delta = gethrtime() - fio->io_timestamp;
4946 if (delta > spa_deadman_synctime(spa))
4947 zio_deadman(fio, tag);
4949 mutex_exit(&vq->vq_lock);
4954 vdev_defer_resilver(vdev_t *vd)
4956 ASSERT(vd->vdev_ops->vdev_op_leaf);
4958 vd->vdev_resilver_deferred = B_TRUE;
4959 vd->vdev_spa->spa_resilver_deferred = B_TRUE;
4963 * Clears the resilver deferred flag on all leaf devs under vd. Returns
4964 * B_TRUE if we have devices that need to be resilvered and are available to
4965 * accept resilver I/Os.
4968 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx)
4970 boolean_t resilver_needed = B_FALSE;
4971 spa_t *spa = vd->vdev_spa;
4973 for (int c = 0; c < vd->vdev_children; c++) {
4974 vdev_t *cvd = vd->vdev_child[c];
4975 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx);
4978 if (vd == spa->spa_root_vdev &&
4979 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) {
4980 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx);
4981 vdev_config_dirty(vd);
4982 spa->spa_resilver_deferred = B_FALSE;
4983 return (resilver_needed);
4986 if (!vdev_is_concrete(vd) || vd->vdev_aux ||
4987 !vd->vdev_ops->vdev_op_leaf)
4988 return (resilver_needed);
4990 vd->vdev_resilver_deferred = B_FALSE;
4992 return (!vdev_is_dead(vd) && !vd->vdev_offline &&
4993 vdev_resilver_needed(vd, NULL, NULL));
4997 * Translate a logical range to the physical range for the specified vdev_t.
4998 * This function is initially called with a leaf vdev and will walk each
4999 * parent vdev until it reaches a top-level vdev. Once the top-level is
5000 * reached the physical range is initialized and the recursive function
5001 * begins to unwind. As it unwinds it calls the parent's vdev specific
5002 * translation function to do the real conversion.
5005 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs,
5006 range_seg64_t *physical_rs)
5009 * Walk up the vdev tree
5011 if (vd != vd->vdev_top) {
5012 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs);
5015 * We've reached the top-level vdev, initialize the
5016 * physical range to the logical range and start to
5019 physical_rs->rs_start = logical_rs->rs_start;
5020 physical_rs->rs_end = logical_rs->rs_end;
5024 vdev_t *pvd = vd->vdev_parent;
5025 ASSERT3P(pvd, !=, NULL);
5026 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL);
5029 * As this recursive function unwinds, translate the logical
5030 * range into its physical components by calling the
5031 * vdev specific translate function.
5033 range_seg64_t intermediate = { 0 };
5034 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate);
5036 physical_rs->rs_start = intermediate.rs_start;
5037 physical_rs->rs_end = intermediate.rs_end;
5041 * Look at the vdev tree and determine whether any devices are currently being
5045 vdev_replace_in_progress(vdev_t *vdev)
5047 ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0);
5049 if (vdev->vdev_ops == &vdev_replacing_ops)
5053 * A 'spare' vdev indicates that we have a replace in progress, unless
5054 * it has exactly two children, and the second, the hot spare, has
5055 * finished being resilvered.
5057 if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 ||
5058 !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING)))
5061 for (int i = 0; i < vdev->vdev_children; i++) {
5062 if (vdev_replace_in_progress(vdev->vdev_child[i]))
5069 EXPORT_SYMBOL(vdev_fault);
5070 EXPORT_SYMBOL(vdev_degrade);
5071 EXPORT_SYMBOL(vdev_online);
5072 EXPORT_SYMBOL(vdev_offline);
5073 EXPORT_SYMBOL(vdev_clear);
5076 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, INT, ZMOD_RW,
5077 "Target number of metaslabs per top-level vdev");
5079 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, INT, ZMOD_RW,
5080 "Default limit for metaslab size");
5082 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, INT, ZMOD_RW,
5083 "Minimum number of metaslabs per top-level vdev");
5085 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, INT, ZMOD_RW,
5086 "Practical upper limit of total metaslabs per top-level vdev");
5088 ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW,
5089 "Rate limit slow IO (delay) events to this many per second");
5091 ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW,
5092 "Rate limit checksum events to this many checksum errors per second "
5093 "(do not set below zed threshold).");
5095 ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW,
5096 "Ignore errors during resilver/scrub");
5098 ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW,
5099 "Bypass vdev_validate()");
5101 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW,
5102 "Disable cache flushes");
5104 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift,
5105 param_set_min_auto_ashift, param_get_ulong, ZMOD_RW,
5106 "Minimum ashift used when creating new top-level vdevs");
5108 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift,
5109 param_set_max_auto_ashift, param_get_ulong, ZMOD_RW,
5110 "Maximum ashift used when optimizing for logical -> physical sector "
5111 "size on new top-level vdevs");