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;
115 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
121 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
124 if (vd->vdev_path != NULL) {
125 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
128 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
129 vd->vdev_ops->vdev_op_type,
130 (u_longlong_t)vd->vdev_id,
131 (u_longlong_t)vd->vdev_guid, buf);
136 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
140 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
141 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
142 vd->vdev_ops->vdev_op_type);
146 switch (vd->vdev_state) {
147 case VDEV_STATE_UNKNOWN:
148 (void) snprintf(state, sizeof (state), "unknown");
150 case VDEV_STATE_CLOSED:
151 (void) snprintf(state, sizeof (state), "closed");
153 case VDEV_STATE_OFFLINE:
154 (void) snprintf(state, sizeof (state), "offline");
156 case VDEV_STATE_REMOVED:
157 (void) snprintf(state, sizeof (state), "removed");
159 case VDEV_STATE_CANT_OPEN:
160 (void) snprintf(state, sizeof (state), "can't open");
162 case VDEV_STATE_FAULTED:
163 (void) snprintf(state, sizeof (state), "faulted");
165 case VDEV_STATE_DEGRADED:
166 (void) snprintf(state, sizeof (state), "degraded");
168 case VDEV_STATE_HEALTHY:
169 (void) snprintf(state, sizeof (state), "healthy");
172 (void) snprintf(state, sizeof (state), "<state %u>",
173 (uint_t)vd->vdev_state);
176 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
177 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
178 vd->vdev_islog ? " (log)" : "",
179 (u_longlong_t)vd->vdev_guid,
180 vd->vdev_path ? vd->vdev_path : "N/A", state);
182 for (uint64_t i = 0; i < vd->vdev_children; i++)
183 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
187 * Virtual device management.
190 static vdev_ops_t *vdev_ops_table[] = {
205 * Given a vdev type, return the appropriate ops vector.
208 vdev_getops(const char *type)
210 vdev_ops_t *ops, **opspp;
212 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
213 if (strcmp(ops->vdev_op_type, type) == 0)
221 vdev_default_xlate(vdev_t *vd, const range_seg64_t *in, range_seg64_t *res)
223 res->rs_start = in->rs_start;
224 res->rs_end = in->rs_end;
228 * Derive the enumerated allocation bias from string input.
229 * String origin is either the per-vdev zap or zpool(1M).
231 static vdev_alloc_bias_t
232 vdev_derive_alloc_bias(const char *bias)
234 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
236 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
237 alloc_bias = VDEV_BIAS_LOG;
238 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
239 alloc_bias = VDEV_BIAS_SPECIAL;
240 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
241 alloc_bias = VDEV_BIAS_DEDUP;
247 * Default asize function: return the MAX of psize with the asize of
248 * all children. This is what's used by anything other than RAID-Z.
251 vdev_default_asize(vdev_t *vd, uint64_t psize)
253 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
256 for (int c = 0; c < vd->vdev_children; c++) {
257 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
258 asize = MAX(asize, csize);
265 * Get the minimum allocatable size. We define the allocatable size as
266 * the vdev's asize rounded to the nearest metaslab. This allows us to
267 * replace or attach devices which don't have the same physical size but
268 * can still satisfy the same number of allocations.
271 vdev_get_min_asize(vdev_t *vd)
273 vdev_t *pvd = vd->vdev_parent;
276 * If our parent is NULL (inactive spare or cache) or is the root,
277 * just return our own asize.
280 return (vd->vdev_asize);
283 * The top-level vdev just returns the allocatable size rounded
284 * to the nearest metaslab.
286 if (vd == vd->vdev_top)
287 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
290 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
291 * so each child must provide at least 1/Nth of its asize.
293 if (pvd->vdev_ops == &vdev_raidz_ops)
294 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
297 return (pvd->vdev_min_asize);
301 vdev_set_min_asize(vdev_t *vd)
303 vd->vdev_min_asize = vdev_get_min_asize(vd);
305 for (int c = 0; c < vd->vdev_children; c++)
306 vdev_set_min_asize(vd->vdev_child[c]);
310 vdev_lookup_top(spa_t *spa, uint64_t vdev)
312 vdev_t *rvd = spa->spa_root_vdev;
314 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
316 if (vdev < rvd->vdev_children) {
317 ASSERT(rvd->vdev_child[vdev] != NULL);
318 return (rvd->vdev_child[vdev]);
325 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
329 if (vd->vdev_guid == guid)
332 for (int c = 0; c < vd->vdev_children; c++)
333 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
341 vdev_count_leaves_impl(vdev_t *vd)
345 if (vd->vdev_ops->vdev_op_leaf)
348 for (int c = 0; c < vd->vdev_children; c++)
349 n += vdev_count_leaves_impl(vd->vdev_child[c]);
355 vdev_count_leaves(spa_t *spa)
359 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
360 rc = vdev_count_leaves_impl(spa->spa_root_vdev);
361 spa_config_exit(spa, SCL_VDEV, FTAG);
367 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
369 size_t oldsize, newsize;
370 uint64_t id = cvd->vdev_id;
373 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
374 ASSERT(cvd->vdev_parent == NULL);
376 cvd->vdev_parent = pvd;
381 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
383 oldsize = pvd->vdev_children * sizeof (vdev_t *);
384 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
385 newsize = pvd->vdev_children * sizeof (vdev_t *);
387 newchild = kmem_alloc(newsize, KM_SLEEP);
388 if (pvd->vdev_child != NULL) {
389 bcopy(pvd->vdev_child, newchild, oldsize);
390 kmem_free(pvd->vdev_child, oldsize);
393 pvd->vdev_child = newchild;
394 pvd->vdev_child[id] = cvd;
396 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
397 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
400 * Walk up all ancestors to update guid sum.
402 for (; pvd != NULL; pvd = pvd->vdev_parent)
403 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
405 if (cvd->vdev_ops->vdev_op_leaf) {
406 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd);
407 cvd->vdev_spa->spa_leaf_list_gen++;
412 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
415 uint_t id = cvd->vdev_id;
417 ASSERT(cvd->vdev_parent == pvd);
422 ASSERT(id < pvd->vdev_children);
423 ASSERT(pvd->vdev_child[id] == cvd);
425 pvd->vdev_child[id] = NULL;
426 cvd->vdev_parent = NULL;
428 for (c = 0; c < pvd->vdev_children; c++)
429 if (pvd->vdev_child[c])
432 if (c == pvd->vdev_children) {
433 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
434 pvd->vdev_child = NULL;
435 pvd->vdev_children = 0;
438 if (cvd->vdev_ops->vdev_op_leaf) {
439 spa_t *spa = cvd->vdev_spa;
440 list_remove(&spa->spa_leaf_list, cvd);
441 spa->spa_leaf_list_gen++;
445 * Walk up all ancestors to update guid sum.
447 for (; pvd != NULL; pvd = pvd->vdev_parent)
448 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
452 * Remove any holes in the child array.
455 vdev_compact_children(vdev_t *pvd)
457 vdev_t **newchild, *cvd;
458 int oldc = pvd->vdev_children;
461 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
466 for (int c = newc = 0; c < oldc; c++)
467 if (pvd->vdev_child[c])
471 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP);
473 for (int c = newc = 0; c < oldc; c++) {
474 if ((cvd = pvd->vdev_child[c]) != NULL) {
475 newchild[newc] = cvd;
476 cvd->vdev_id = newc++;
483 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
484 pvd->vdev_child = newchild;
485 pvd->vdev_children = newc;
489 * Allocate and minimally initialize a vdev_t.
492 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
495 vdev_indirect_config_t *vic;
497 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
498 vic = &vd->vdev_indirect_config;
500 if (spa->spa_root_vdev == NULL) {
501 ASSERT(ops == &vdev_root_ops);
502 spa->spa_root_vdev = vd;
503 spa->spa_load_guid = spa_generate_guid(NULL);
506 if (guid == 0 && ops != &vdev_hole_ops) {
507 if (spa->spa_root_vdev == vd) {
509 * The root vdev's guid will also be the pool guid,
510 * which must be unique among all pools.
512 guid = spa_generate_guid(NULL);
515 * Any other vdev's guid must be unique within the pool.
517 guid = spa_generate_guid(spa);
519 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
524 vd->vdev_guid = guid;
525 vd->vdev_guid_sum = guid;
527 vd->vdev_state = VDEV_STATE_CLOSED;
528 vd->vdev_ishole = (ops == &vdev_hole_ops);
529 vic->vic_prev_indirect_vdev = UINT64_MAX;
531 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
532 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
533 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL,
537 * Initialize rate limit structs for events. We rate limit ZIO delay
538 * and checksum events so that we don't overwhelm ZED with thousands
539 * of events when a disk is acting up.
541 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second,
543 zfs_ratelimit_init(&vd->vdev_checksum_rl,
544 &zfs_checksum_events_per_second, 1);
546 list_link_init(&vd->vdev_config_dirty_node);
547 list_link_init(&vd->vdev_state_dirty_node);
548 list_link_init(&vd->vdev_initialize_node);
549 list_link_init(&vd->vdev_leaf_node);
550 list_link_init(&vd->vdev_trim_node);
551 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL);
552 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
553 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
554 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
556 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
557 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
558 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
559 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
561 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL);
562 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL);
563 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL);
564 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL);
565 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL);
566 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL);
568 mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL);
569 mutex_init(&vd->vdev_rebuild_io_lock, NULL, MUTEX_DEFAULT, NULL);
570 cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL);
571 cv_init(&vd->vdev_rebuild_io_cv, NULL, CV_DEFAULT, NULL);
573 for (int t = 0; t < DTL_TYPES; t++) {
574 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
578 txg_list_create(&vd->vdev_ms_list, spa,
579 offsetof(struct metaslab, ms_txg_node));
580 txg_list_create(&vd->vdev_dtl_list, spa,
581 offsetof(struct vdev, vdev_dtl_node));
582 vd->vdev_stat.vs_timestamp = gethrtime();
590 * Allocate a new vdev. The 'alloctype' is used to control whether we are
591 * creating a new vdev or loading an existing one - the behavior is slightly
592 * different for each case.
595 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
600 uint64_t guid = 0, islog, nparity;
602 vdev_indirect_config_t *vic;
605 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
606 boolean_t top_level = (parent && !parent->vdev_parent);
608 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
610 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
611 return (SET_ERROR(EINVAL));
613 if ((ops = vdev_getops(type)) == NULL)
614 return (SET_ERROR(EINVAL));
617 * If this is a load, get the vdev guid from the nvlist.
618 * Otherwise, vdev_alloc_common() will generate one for us.
620 if (alloctype == VDEV_ALLOC_LOAD) {
623 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
625 return (SET_ERROR(EINVAL));
627 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
628 return (SET_ERROR(EINVAL));
629 } else if (alloctype == VDEV_ALLOC_SPARE) {
630 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
631 return (SET_ERROR(EINVAL));
632 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
633 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
634 return (SET_ERROR(EINVAL));
635 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
636 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
637 return (SET_ERROR(EINVAL));
641 * The first allocated vdev must be of type 'root'.
643 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
644 return (SET_ERROR(EINVAL));
647 * Determine whether we're a log vdev.
650 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
651 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
652 return (SET_ERROR(ENOTSUP));
654 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
655 return (SET_ERROR(ENOTSUP));
658 * Set the nparity property for RAID-Z vdevs.
661 if (ops == &vdev_raidz_ops) {
662 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
664 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
665 return (SET_ERROR(EINVAL));
667 * Previous versions could only support 1 or 2 parity
671 spa_version(spa) < SPA_VERSION_RAIDZ2)
672 return (SET_ERROR(ENOTSUP));
674 spa_version(spa) < SPA_VERSION_RAIDZ3)
675 return (SET_ERROR(ENOTSUP));
678 * We require the parity to be specified for SPAs that
679 * support multiple parity levels.
681 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
682 return (SET_ERROR(EINVAL));
684 * Otherwise, we default to 1 parity device for RAID-Z.
691 ASSERT(nparity != -1ULL);
694 * If creating a top-level vdev, check for allocation classes input
696 if (top_level && alloctype == VDEV_ALLOC_ADD) {
699 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
701 alloc_bias = vdev_derive_alloc_bias(bias);
703 /* spa_vdev_add() expects feature to be enabled */
704 if (spa->spa_load_state != SPA_LOAD_CREATE &&
705 !spa_feature_is_enabled(spa,
706 SPA_FEATURE_ALLOCATION_CLASSES)) {
707 return (SET_ERROR(ENOTSUP));
712 vd = vdev_alloc_common(spa, id, guid, ops);
713 vic = &vd->vdev_indirect_config;
715 vd->vdev_islog = islog;
716 vd->vdev_nparity = nparity;
717 if (top_level && alloc_bias != VDEV_BIAS_NONE)
718 vd->vdev_alloc_bias = alloc_bias;
720 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
721 vd->vdev_path = spa_strdup(vd->vdev_path);
724 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
725 * fault on a vdev and want it to persist across imports (like with
728 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp);
729 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) {
730 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
731 vd->vdev_faulted = 1;
732 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
735 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
736 vd->vdev_devid = spa_strdup(vd->vdev_devid);
737 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
738 &vd->vdev_physpath) == 0)
739 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
741 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
742 &vd->vdev_enc_sysfs_path) == 0)
743 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path);
745 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
746 vd->vdev_fru = spa_strdup(vd->vdev_fru);
749 * Set the whole_disk property. If it's not specified, leave the value
752 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
753 &vd->vdev_wholedisk) != 0)
754 vd->vdev_wholedisk = -1ULL;
756 ASSERT0(vic->vic_mapping_object);
757 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
758 &vic->vic_mapping_object);
759 ASSERT0(vic->vic_births_object);
760 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
761 &vic->vic_births_object);
762 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
763 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
764 &vic->vic_prev_indirect_vdev);
767 * Look for the 'not present' flag. This will only be set if the device
768 * was not present at the time of import.
770 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
771 &vd->vdev_not_present);
774 * Get the alignment requirement.
776 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
779 * Retrieve the vdev creation time.
781 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
785 * If we're a top-level vdev, try to load the allocation parameters.
788 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
789 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
791 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
793 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
795 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
797 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
800 ASSERT0(vd->vdev_top_zap);
803 if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
804 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
805 alloctype == VDEV_ALLOC_ADD ||
806 alloctype == VDEV_ALLOC_SPLIT ||
807 alloctype == VDEV_ALLOC_ROOTPOOL);
808 /* Note: metaslab_group_create() is now deferred */
811 if (vd->vdev_ops->vdev_op_leaf &&
812 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
813 (void) nvlist_lookup_uint64(nv,
814 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
816 ASSERT0(vd->vdev_leaf_zap);
820 * If we're a leaf vdev, try to load the DTL object and other state.
823 if (vd->vdev_ops->vdev_op_leaf &&
824 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
825 alloctype == VDEV_ALLOC_ROOTPOOL)) {
826 if (alloctype == VDEV_ALLOC_LOAD) {
827 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
828 &vd->vdev_dtl_object);
829 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
833 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
836 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
837 &spare) == 0 && spare)
841 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
844 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
845 &vd->vdev_resilver_txg);
847 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
848 &vd->vdev_rebuild_txg);
850 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER))
851 vdev_defer_resilver(vd);
854 * In general, when importing a pool we want to ignore the
855 * persistent fault state, as the diagnosis made on another
856 * system may not be valid in the current context. The only
857 * exception is if we forced a vdev to a persistently faulted
858 * state with 'zpool offline -f'. The persistent fault will
859 * remain across imports until cleared.
861 * Local vdevs will remain in the faulted state.
863 if (spa_load_state(spa) == SPA_LOAD_OPEN ||
864 spa_load_state(spa) == SPA_LOAD_IMPORT) {
865 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
867 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
869 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
872 if (vd->vdev_faulted || vd->vdev_degraded) {
876 VDEV_AUX_ERR_EXCEEDED;
877 if (nvlist_lookup_string(nv,
878 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
879 strcmp(aux, "external") == 0)
880 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
882 vd->vdev_faulted = 0ULL;
888 * Add ourselves to the parent's list of children.
890 vdev_add_child(parent, vd);
898 vdev_free(vdev_t *vd)
900 spa_t *spa = vd->vdev_spa;
902 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
903 ASSERT3P(vd->vdev_trim_thread, ==, NULL);
904 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL);
905 ASSERT3P(vd->vdev_rebuild_thread, ==, NULL);
908 * Scan queues are normally destroyed at the end of a scan. If the
909 * queue exists here, that implies the vdev is being removed while
910 * the scan is still running.
912 if (vd->vdev_scan_io_queue != NULL) {
913 mutex_enter(&vd->vdev_scan_io_queue_lock);
914 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
915 vd->vdev_scan_io_queue = NULL;
916 mutex_exit(&vd->vdev_scan_io_queue_lock);
920 * vdev_free() implies closing the vdev first. This is simpler than
921 * trying to ensure complicated semantics for all callers.
925 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
926 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
931 for (int c = 0; c < vd->vdev_children; c++)
932 vdev_free(vd->vdev_child[c]);
934 ASSERT(vd->vdev_child == NULL);
935 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
938 * Discard allocation state.
940 if (vd->vdev_mg != NULL) {
941 vdev_metaslab_fini(vd);
942 metaslab_group_destroy(vd->vdev_mg);
946 ASSERT0(vd->vdev_stat.vs_space);
947 ASSERT0(vd->vdev_stat.vs_dspace);
948 ASSERT0(vd->vdev_stat.vs_alloc);
951 * Remove this vdev from its parent's child list.
953 vdev_remove_child(vd->vdev_parent, vd);
955 ASSERT(vd->vdev_parent == NULL);
956 ASSERT(!list_link_active(&vd->vdev_leaf_node));
959 * Clean up vdev structure.
965 spa_strfree(vd->vdev_path);
967 spa_strfree(vd->vdev_devid);
968 if (vd->vdev_physpath)
969 spa_strfree(vd->vdev_physpath);
971 if (vd->vdev_enc_sysfs_path)
972 spa_strfree(vd->vdev_enc_sysfs_path);
975 spa_strfree(vd->vdev_fru);
977 if (vd->vdev_isspare)
978 spa_spare_remove(vd);
979 if (vd->vdev_isl2cache)
980 spa_l2cache_remove(vd);
982 txg_list_destroy(&vd->vdev_ms_list);
983 txg_list_destroy(&vd->vdev_dtl_list);
985 mutex_enter(&vd->vdev_dtl_lock);
986 space_map_close(vd->vdev_dtl_sm);
987 for (int t = 0; t < DTL_TYPES; t++) {
988 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
989 range_tree_destroy(vd->vdev_dtl[t]);
991 mutex_exit(&vd->vdev_dtl_lock);
993 EQUIV(vd->vdev_indirect_births != NULL,
994 vd->vdev_indirect_mapping != NULL);
995 if (vd->vdev_indirect_births != NULL) {
996 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
997 vdev_indirect_births_close(vd->vdev_indirect_births);
1000 if (vd->vdev_obsolete_sm != NULL) {
1001 ASSERT(vd->vdev_removing ||
1002 vd->vdev_ops == &vdev_indirect_ops);
1003 space_map_close(vd->vdev_obsolete_sm);
1004 vd->vdev_obsolete_sm = NULL;
1006 range_tree_destroy(vd->vdev_obsolete_segments);
1007 rw_destroy(&vd->vdev_indirect_rwlock);
1008 mutex_destroy(&vd->vdev_obsolete_lock);
1010 mutex_destroy(&vd->vdev_dtl_lock);
1011 mutex_destroy(&vd->vdev_stat_lock);
1012 mutex_destroy(&vd->vdev_probe_lock);
1013 mutex_destroy(&vd->vdev_scan_io_queue_lock);
1015 mutex_destroy(&vd->vdev_initialize_lock);
1016 mutex_destroy(&vd->vdev_initialize_io_lock);
1017 cv_destroy(&vd->vdev_initialize_io_cv);
1018 cv_destroy(&vd->vdev_initialize_cv);
1020 mutex_destroy(&vd->vdev_trim_lock);
1021 mutex_destroy(&vd->vdev_autotrim_lock);
1022 mutex_destroy(&vd->vdev_trim_io_lock);
1023 cv_destroy(&vd->vdev_trim_cv);
1024 cv_destroy(&vd->vdev_autotrim_cv);
1025 cv_destroy(&vd->vdev_trim_io_cv);
1027 mutex_destroy(&vd->vdev_rebuild_lock);
1028 mutex_destroy(&vd->vdev_rebuild_io_lock);
1029 cv_destroy(&vd->vdev_rebuild_cv);
1030 cv_destroy(&vd->vdev_rebuild_io_cv);
1032 zfs_ratelimit_fini(&vd->vdev_delay_rl);
1033 zfs_ratelimit_fini(&vd->vdev_checksum_rl);
1035 if (vd == spa->spa_root_vdev)
1036 spa->spa_root_vdev = NULL;
1038 kmem_free(vd, sizeof (vdev_t));
1042 * Transfer top-level vdev state from svd to tvd.
1045 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
1047 spa_t *spa = svd->vdev_spa;
1052 ASSERT(tvd == tvd->vdev_top);
1054 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite;
1055 tvd->vdev_ms_array = svd->vdev_ms_array;
1056 tvd->vdev_ms_shift = svd->vdev_ms_shift;
1057 tvd->vdev_ms_count = svd->vdev_ms_count;
1058 tvd->vdev_top_zap = svd->vdev_top_zap;
1060 svd->vdev_ms_array = 0;
1061 svd->vdev_ms_shift = 0;
1062 svd->vdev_ms_count = 0;
1063 svd->vdev_top_zap = 0;
1066 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1067 tvd->vdev_mg = svd->vdev_mg;
1068 tvd->vdev_ms = svd->vdev_ms;
1070 svd->vdev_mg = NULL;
1071 svd->vdev_ms = NULL;
1073 if (tvd->vdev_mg != NULL)
1074 tvd->vdev_mg->mg_vd = tvd;
1076 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1077 svd->vdev_checkpoint_sm = NULL;
1079 tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
1080 svd->vdev_alloc_bias = VDEV_BIAS_NONE;
1082 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1083 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1084 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1086 svd->vdev_stat.vs_alloc = 0;
1087 svd->vdev_stat.vs_space = 0;
1088 svd->vdev_stat.vs_dspace = 0;
1091 * State which may be set on a top-level vdev that's in the
1092 * process of being removed.
1094 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1095 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1096 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1097 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1098 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1099 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1100 ASSERT0(tvd->vdev_removing);
1101 ASSERT0(tvd->vdev_rebuilding);
1102 tvd->vdev_removing = svd->vdev_removing;
1103 tvd->vdev_rebuilding = svd->vdev_rebuilding;
1104 tvd->vdev_rebuild_config = svd->vdev_rebuild_config;
1105 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1106 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1107 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1108 range_tree_swap(&svd->vdev_obsolete_segments,
1109 &tvd->vdev_obsolete_segments);
1110 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1111 svd->vdev_indirect_config.vic_mapping_object = 0;
1112 svd->vdev_indirect_config.vic_births_object = 0;
1113 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1114 svd->vdev_indirect_mapping = NULL;
1115 svd->vdev_indirect_births = NULL;
1116 svd->vdev_obsolete_sm = NULL;
1117 svd->vdev_removing = 0;
1118 svd->vdev_rebuilding = 0;
1120 for (t = 0; t < TXG_SIZE; t++) {
1121 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1122 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1123 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1124 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1125 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1126 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1129 if (list_link_active(&svd->vdev_config_dirty_node)) {
1130 vdev_config_clean(svd);
1131 vdev_config_dirty(tvd);
1134 if (list_link_active(&svd->vdev_state_dirty_node)) {
1135 vdev_state_clean(svd);
1136 vdev_state_dirty(tvd);
1139 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1140 svd->vdev_deflate_ratio = 0;
1142 tvd->vdev_islog = svd->vdev_islog;
1143 svd->vdev_islog = 0;
1145 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1149 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1156 for (int c = 0; c < vd->vdev_children; c++)
1157 vdev_top_update(tvd, vd->vdev_child[c]);
1161 * Add a mirror/replacing vdev above an existing vdev.
1164 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1166 spa_t *spa = cvd->vdev_spa;
1167 vdev_t *pvd = cvd->vdev_parent;
1170 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1172 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1174 mvd->vdev_asize = cvd->vdev_asize;
1175 mvd->vdev_min_asize = cvd->vdev_min_asize;
1176 mvd->vdev_max_asize = cvd->vdev_max_asize;
1177 mvd->vdev_psize = cvd->vdev_psize;
1178 mvd->vdev_ashift = cvd->vdev_ashift;
1179 mvd->vdev_state = cvd->vdev_state;
1180 mvd->vdev_crtxg = cvd->vdev_crtxg;
1182 vdev_remove_child(pvd, cvd);
1183 vdev_add_child(pvd, mvd);
1184 cvd->vdev_id = mvd->vdev_children;
1185 vdev_add_child(mvd, cvd);
1186 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1188 if (mvd == mvd->vdev_top)
1189 vdev_top_transfer(cvd, mvd);
1195 * Remove a 1-way mirror/replacing vdev from the tree.
1198 vdev_remove_parent(vdev_t *cvd)
1200 vdev_t *mvd = cvd->vdev_parent;
1201 vdev_t *pvd = mvd->vdev_parent;
1203 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1205 ASSERT(mvd->vdev_children == 1);
1206 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1207 mvd->vdev_ops == &vdev_replacing_ops ||
1208 mvd->vdev_ops == &vdev_spare_ops);
1209 cvd->vdev_ashift = mvd->vdev_ashift;
1211 vdev_remove_child(mvd, cvd);
1212 vdev_remove_child(pvd, mvd);
1215 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1216 * Otherwise, we could have detached an offline device, and when we
1217 * go to import the pool we'll think we have two top-level vdevs,
1218 * instead of a different version of the same top-level vdev.
1220 if (mvd->vdev_top == mvd) {
1221 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1222 cvd->vdev_orig_guid = cvd->vdev_guid;
1223 cvd->vdev_guid += guid_delta;
1224 cvd->vdev_guid_sum += guid_delta;
1227 * If pool not set for autoexpand, we need to also preserve
1228 * mvd's asize to prevent automatic expansion of cvd.
1229 * Otherwise if we are adjusting the mirror by attaching and
1230 * detaching children of non-uniform sizes, the mirror could
1231 * autoexpand, unexpectedly requiring larger devices to
1232 * re-establish the mirror.
1234 if (!cvd->vdev_spa->spa_autoexpand)
1235 cvd->vdev_asize = mvd->vdev_asize;
1237 cvd->vdev_id = mvd->vdev_id;
1238 vdev_add_child(pvd, cvd);
1239 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1241 if (cvd == cvd->vdev_top)
1242 vdev_top_transfer(mvd, cvd);
1244 ASSERT(mvd->vdev_children == 0);
1249 vdev_metaslab_group_create(vdev_t *vd)
1251 spa_t *spa = vd->vdev_spa;
1254 * metaslab_group_create was delayed until allocation bias was available
1256 if (vd->vdev_mg == NULL) {
1257 metaslab_class_t *mc;
1259 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
1260 vd->vdev_alloc_bias = VDEV_BIAS_LOG;
1262 ASSERT3U(vd->vdev_islog, ==,
1263 (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
1265 switch (vd->vdev_alloc_bias) {
1267 mc = spa_log_class(spa);
1269 case VDEV_BIAS_SPECIAL:
1270 mc = spa_special_class(spa);
1272 case VDEV_BIAS_DEDUP:
1273 mc = spa_dedup_class(spa);
1276 mc = spa_normal_class(spa);
1279 vd->vdev_mg = metaslab_group_create(mc, vd,
1280 spa->spa_alloc_count);
1283 * The spa ashift values currently only reflect the
1284 * general vdev classes. Class destination is late
1285 * binding so ashift checking had to wait until now
1287 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1288 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
1289 if (vd->vdev_ashift > spa->spa_max_ashift)
1290 spa->spa_max_ashift = vd->vdev_ashift;
1291 if (vd->vdev_ashift < spa->spa_min_ashift)
1292 spa->spa_min_ashift = vd->vdev_ashift;
1298 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1300 spa_t *spa = vd->vdev_spa;
1301 objset_t *mos = spa->spa_meta_objset;
1303 uint64_t oldc = vd->vdev_ms_count;
1304 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1307 boolean_t expanding = (oldc != 0);
1309 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1312 * This vdev is not being allocated from yet or is a hole.
1314 if (vd->vdev_ms_shift == 0)
1317 ASSERT(!vd->vdev_ishole);
1319 ASSERT(oldc <= newc);
1321 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1324 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1325 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1329 vd->vdev_ms_count = newc;
1330 for (m = oldc; m < newc; m++) {
1331 uint64_t object = 0;
1334 * vdev_ms_array may be 0 if we are creating the "fake"
1335 * metaslabs for an indirect vdev for zdb's leak detection.
1336 * See zdb_leak_init().
1338 if (txg == 0 && vd->vdev_ms_array != 0) {
1339 error = dmu_read(mos, vd->vdev_ms_array,
1340 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1343 vdev_dbgmsg(vd, "unable to read the metaslab "
1344 "array [error=%d]", error);
1351 * To accommodate zdb_leak_init() fake indirect
1352 * metaslabs, we allocate a metaslab group for
1353 * indirect vdevs which normally don't have one.
1355 if (vd->vdev_mg == NULL) {
1356 ASSERT0(vdev_is_concrete(vd));
1357 vdev_metaslab_group_create(vd);
1360 error = metaslab_init(vd->vdev_mg, m, object, txg,
1363 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1370 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1373 * If the vdev is being removed we don't activate
1374 * the metaslabs since we want to ensure that no new
1375 * allocations are performed on this device.
1377 if (!expanding && !vd->vdev_removing) {
1378 metaslab_group_activate(vd->vdev_mg);
1382 spa_config_exit(spa, SCL_ALLOC, FTAG);
1385 * Regardless whether this vdev was just added or it is being
1386 * expanded, the metaslab count has changed. Recalculate the
1389 spa_log_sm_set_blocklimit(spa);
1395 vdev_metaslab_fini(vdev_t *vd)
1397 if (vd->vdev_checkpoint_sm != NULL) {
1398 ASSERT(spa_feature_is_active(vd->vdev_spa,
1399 SPA_FEATURE_POOL_CHECKPOINT));
1400 space_map_close(vd->vdev_checkpoint_sm);
1402 * Even though we close the space map, we need to set its
1403 * pointer to NULL. The reason is that vdev_metaslab_fini()
1404 * may be called multiple times for certain operations
1405 * (i.e. when destroying a pool) so we need to ensure that
1406 * this clause never executes twice. This logic is similar
1407 * to the one used for the vdev_ms clause below.
1409 vd->vdev_checkpoint_sm = NULL;
1412 if (vd->vdev_ms != NULL) {
1413 metaslab_group_t *mg = vd->vdev_mg;
1414 metaslab_group_passivate(mg);
1416 uint64_t count = vd->vdev_ms_count;
1417 for (uint64_t m = 0; m < count; m++) {
1418 metaslab_t *msp = vd->vdev_ms[m];
1422 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1425 vd->vdev_ms_count = 0;
1427 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
1428 ASSERT0(mg->mg_histogram[i]);
1430 ASSERT0(vd->vdev_ms_count);
1431 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
1434 typedef struct vdev_probe_stats {
1435 boolean_t vps_readable;
1436 boolean_t vps_writeable;
1438 } vdev_probe_stats_t;
1441 vdev_probe_done(zio_t *zio)
1443 spa_t *spa = zio->io_spa;
1444 vdev_t *vd = zio->io_vd;
1445 vdev_probe_stats_t *vps = zio->io_private;
1447 ASSERT(vd->vdev_probe_zio != NULL);
1449 if (zio->io_type == ZIO_TYPE_READ) {
1450 if (zio->io_error == 0)
1451 vps->vps_readable = 1;
1452 if (zio->io_error == 0 && spa_writeable(spa)) {
1453 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1454 zio->io_offset, zio->io_size, zio->io_abd,
1455 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1456 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1458 abd_free(zio->io_abd);
1460 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1461 if (zio->io_error == 0)
1462 vps->vps_writeable = 1;
1463 abd_free(zio->io_abd);
1464 } else if (zio->io_type == ZIO_TYPE_NULL) {
1468 vd->vdev_cant_read |= !vps->vps_readable;
1469 vd->vdev_cant_write |= !vps->vps_writeable;
1471 if (vdev_readable(vd) &&
1472 (vdev_writeable(vd) || !spa_writeable(spa))) {
1475 ASSERT(zio->io_error != 0);
1476 vdev_dbgmsg(vd, "failed probe");
1477 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1478 spa, vd, NULL, NULL, 0, 0);
1479 zio->io_error = SET_ERROR(ENXIO);
1482 mutex_enter(&vd->vdev_probe_lock);
1483 ASSERT(vd->vdev_probe_zio == zio);
1484 vd->vdev_probe_zio = NULL;
1485 mutex_exit(&vd->vdev_probe_lock);
1488 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1489 if (!vdev_accessible(vd, pio))
1490 pio->io_error = SET_ERROR(ENXIO);
1492 kmem_free(vps, sizeof (*vps));
1497 * Determine whether this device is accessible.
1499 * Read and write to several known locations: the pad regions of each
1500 * vdev label but the first, which we leave alone in case it contains
1504 vdev_probe(vdev_t *vd, zio_t *zio)
1506 spa_t *spa = vd->vdev_spa;
1507 vdev_probe_stats_t *vps = NULL;
1510 ASSERT(vd->vdev_ops->vdev_op_leaf);
1513 * Don't probe the probe.
1515 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1519 * To prevent 'probe storms' when a device fails, we create
1520 * just one probe i/o at a time. All zios that want to probe
1521 * this vdev will become parents of the probe io.
1523 mutex_enter(&vd->vdev_probe_lock);
1525 if ((pio = vd->vdev_probe_zio) == NULL) {
1526 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1528 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1529 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1532 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1534 * vdev_cant_read and vdev_cant_write can only
1535 * transition from TRUE to FALSE when we have the
1536 * SCL_ZIO lock as writer; otherwise they can only
1537 * transition from FALSE to TRUE. This ensures that
1538 * any zio looking at these values can assume that
1539 * failures persist for the life of the I/O. That's
1540 * important because when a device has intermittent
1541 * connectivity problems, we want to ensure that
1542 * they're ascribed to the device (ENXIO) and not
1545 * Since we hold SCL_ZIO as writer here, clear both
1546 * values so the probe can reevaluate from first
1549 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1550 vd->vdev_cant_read = B_FALSE;
1551 vd->vdev_cant_write = B_FALSE;
1554 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1555 vdev_probe_done, vps,
1556 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1559 * We can't change the vdev state in this context, so we
1560 * kick off an async task to do it on our behalf.
1563 vd->vdev_probe_wanted = B_TRUE;
1564 spa_async_request(spa, SPA_ASYNC_PROBE);
1569 zio_add_child(zio, pio);
1571 mutex_exit(&vd->vdev_probe_lock);
1574 ASSERT(zio != NULL);
1578 for (int l = 1; l < VDEV_LABELS; l++) {
1579 zio_nowait(zio_read_phys(pio, vd,
1580 vdev_label_offset(vd->vdev_psize, l,
1581 offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE,
1582 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1583 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1584 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1595 vdev_open_child(void *arg)
1599 vd->vdev_open_thread = curthread;
1600 vd->vdev_open_error = vdev_open(vd);
1601 vd->vdev_open_thread = NULL;
1605 vdev_uses_zvols(vdev_t *vd)
1608 if (zvol_is_zvol(vd->vdev_path))
1612 for (int c = 0; c < vd->vdev_children; c++)
1613 if (vdev_uses_zvols(vd->vdev_child[c]))
1620 vdev_open_children(vdev_t *vd)
1623 int children = vd->vdev_children;
1626 * in order to handle pools on top of zvols, do the opens
1627 * in a single thread so that the same thread holds the
1628 * spa_namespace_lock
1630 if (vdev_uses_zvols(vd)) {
1632 for (int c = 0; c < children; c++)
1633 vd->vdev_child[c]->vdev_open_error =
1634 vdev_open(vd->vdev_child[c]);
1636 tq = taskq_create("vdev_open", children, minclsyspri,
1637 children, children, TASKQ_PREPOPULATE);
1641 for (int c = 0; c < children; c++)
1642 VERIFY(taskq_dispatch(tq, vdev_open_child,
1643 vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID);
1648 vd->vdev_nonrot = B_TRUE;
1650 for (int c = 0; c < children; c++)
1651 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1655 * Compute the raidz-deflation ratio. Note, we hard-code
1656 * in 128k (1 << 17) because it is the "typical" blocksize.
1657 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1658 * otherwise it would inconsistently account for existing bp's.
1661 vdev_set_deflate_ratio(vdev_t *vd)
1663 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1664 vd->vdev_deflate_ratio = (1 << 17) /
1665 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1670 * Prepare a virtual device for access.
1673 vdev_open(vdev_t *vd)
1675 spa_t *spa = vd->vdev_spa;
1678 uint64_t max_osize = 0;
1679 uint64_t asize, max_asize, psize;
1680 uint64_t ashift = 0;
1682 ASSERT(vd->vdev_open_thread == curthread ||
1683 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1684 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1685 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1686 vd->vdev_state == VDEV_STATE_OFFLINE);
1688 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1689 vd->vdev_cant_read = B_FALSE;
1690 vd->vdev_cant_write = B_FALSE;
1691 vd->vdev_min_asize = vdev_get_min_asize(vd);
1694 * If this vdev is not removed, check its fault status. If it's
1695 * faulted, bail out of the open.
1697 if (!vd->vdev_removed && vd->vdev_faulted) {
1698 ASSERT(vd->vdev_children == 0);
1699 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1700 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1701 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1702 vd->vdev_label_aux);
1703 return (SET_ERROR(ENXIO));
1704 } else if (vd->vdev_offline) {
1705 ASSERT(vd->vdev_children == 0);
1706 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1707 return (SET_ERROR(ENXIO));
1710 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1713 * Physical volume size should never be larger than its max size, unless
1714 * the disk has shrunk while we were reading it or the device is buggy
1715 * or damaged: either way it's not safe for use, bail out of the open.
1717 if (osize > max_osize) {
1718 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1719 VDEV_AUX_OPEN_FAILED);
1720 return (SET_ERROR(ENXIO));
1724 * Reset the vdev_reopening flag so that we actually close
1725 * the vdev on error.
1727 vd->vdev_reopening = B_FALSE;
1728 if (zio_injection_enabled && error == 0)
1729 error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO));
1732 if (vd->vdev_removed &&
1733 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1734 vd->vdev_removed = B_FALSE;
1736 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1737 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1738 vd->vdev_stat.vs_aux);
1740 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1741 vd->vdev_stat.vs_aux);
1746 vd->vdev_removed = B_FALSE;
1749 * Recheck the faulted flag now that we have confirmed that
1750 * the vdev is accessible. If we're faulted, bail.
1752 if (vd->vdev_faulted) {
1753 ASSERT(vd->vdev_children == 0);
1754 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1755 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1756 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1757 vd->vdev_label_aux);
1758 return (SET_ERROR(ENXIO));
1761 if (vd->vdev_degraded) {
1762 ASSERT(vd->vdev_children == 0);
1763 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1764 VDEV_AUX_ERR_EXCEEDED);
1766 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1770 * For hole or missing vdevs we just return success.
1772 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1775 for (int c = 0; c < vd->vdev_children; c++) {
1776 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1777 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1783 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1784 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1786 if (vd->vdev_children == 0) {
1787 if (osize < SPA_MINDEVSIZE) {
1788 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1789 VDEV_AUX_TOO_SMALL);
1790 return (SET_ERROR(EOVERFLOW));
1793 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1794 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1795 VDEV_LABEL_END_SIZE);
1797 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1798 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1799 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1800 VDEV_AUX_TOO_SMALL);
1801 return (SET_ERROR(EOVERFLOW));
1805 max_asize = max_osize;
1809 * If the vdev was expanded, record this so that we can re-create the
1810 * uberblock rings in labels {2,3}, during the next sync.
1812 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0))
1813 vd->vdev_copy_uberblocks = B_TRUE;
1815 vd->vdev_psize = psize;
1818 * Make sure the allocatable size hasn't shrunk too much.
1820 if (asize < vd->vdev_min_asize) {
1821 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1822 VDEV_AUX_BAD_LABEL);
1823 return (SET_ERROR(EINVAL));
1826 if (vd->vdev_asize == 0) {
1828 * This is the first-ever open, so use the computed values.
1829 * For compatibility, a different ashift can be requested.
1831 vd->vdev_asize = asize;
1832 vd->vdev_max_asize = max_asize;
1833 if (vd->vdev_ashift == 0) {
1834 vd->vdev_ashift = ashift; /* use detected value */
1836 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
1837 vd->vdev_ashift > ASHIFT_MAX)) {
1838 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1839 VDEV_AUX_BAD_ASHIFT);
1840 return (SET_ERROR(EDOM));
1844 * Detect if the alignment requirement has increased.
1845 * We don't want to make the pool unavailable, just
1846 * post an event instead.
1848 if (ashift > vd->vdev_top->vdev_ashift &&
1849 vd->vdev_ops->vdev_op_leaf) {
1850 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
1851 spa, vd, NULL, NULL, 0, 0);
1854 vd->vdev_max_asize = max_asize;
1858 * If all children are healthy we update asize if either:
1859 * The asize has increased, due to a device expansion caused by dynamic
1860 * LUN growth or vdev replacement, and automatic expansion is enabled;
1861 * making the additional space available.
1863 * The asize has decreased, due to a device shrink usually caused by a
1864 * vdev replace with a smaller device. This ensures that calculations
1865 * based of max_asize and asize e.g. esize are always valid. It's safe
1866 * to do this as we've already validated that asize is greater than
1869 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1870 ((asize > vd->vdev_asize &&
1871 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1872 (asize < vd->vdev_asize)))
1873 vd->vdev_asize = asize;
1875 vdev_set_min_asize(vd);
1878 * Ensure we can issue some IO before declaring the
1879 * vdev open for business.
1881 if (vd->vdev_ops->vdev_op_leaf &&
1882 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1883 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1884 VDEV_AUX_ERR_EXCEEDED);
1889 * Track the min and max ashift values for normal data devices.
1891 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1892 vd->vdev_alloc_bias == VDEV_BIAS_NONE &&
1893 vd->vdev_islog == 0 && vd->vdev_aux == NULL) {
1894 if (vd->vdev_ashift > spa->spa_max_ashift)
1895 spa->spa_max_ashift = vd->vdev_ashift;
1896 if (vd->vdev_ashift < spa->spa_min_ashift)
1897 spa->spa_min_ashift = vd->vdev_ashift;
1901 * If this is a leaf vdev, assess whether a resilver is needed.
1902 * But don't do this if we are doing a reopen for a scrub, since
1903 * this would just restart the scrub we are already doing.
1905 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen)
1906 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd);
1912 * Called once the vdevs are all opened, this routine validates the label
1913 * contents. This needs to be done before vdev_load() so that we don't
1914 * inadvertently do repair I/Os to the wrong device.
1916 * This function will only return failure if one of the vdevs indicates that it
1917 * has since been destroyed or exported. This is only possible if
1918 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1919 * will be updated but the function will return 0.
1922 vdev_validate(vdev_t *vd)
1924 spa_t *spa = vd->vdev_spa;
1926 uint64_t guid = 0, aux_guid = 0, top_guid;
1931 if (vdev_validate_skip)
1934 for (uint64_t c = 0; c < vd->vdev_children; c++)
1935 if (vdev_validate(vd->vdev_child[c]) != 0)
1936 return (SET_ERROR(EBADF));
1939 * If the device has already failed, or was marked offline, don't do
1940 * any further validation. Otherwise, label I/O will fail and we will
1941 * overwrite the previous state.
1943 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1947 * If we are performing an extreme rewind, we allow for a label that
1948 * was modified at a point after the current txg.
1949 * If config lock is not held do not check for the txg. spa_sync could
1950 * be updating the vdev's label before updating spa_last_synced_txg.
1952 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1953 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1956 txg = spa_last_synced_txg(spa);
1958 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1959 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1960 VDEV_AUX_BAD_LABEL);
1961 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
1962 "txg %llu", (u_longlong_t)txg);
1967 * Determine if this vdev has been split off into another
1968 * pool. If so, then refuse to open it.
1970 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1971 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1972 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1973 VDEV_AUX_SPLIT_POOL);
1975 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1979 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1980 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1981 VDEV_AUX_CORRUPT_DATA);
1983 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1984 ZPOOL_CONFIG_POOL_GUID);
1989 * If config is not trusted then ignore the spa guid check. This is
1990 * necessary because if the machine crashed during a re-guid the new
1991 * guid might have been written to all of the vdev labels, but not the
1992 * cached config. The check will be performed again once we have the
1993 * trusted config from the MOS.
1995 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1996 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1997 VDEV_AUX_CORRUPT_DATA);
1999 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
2000 "match config (%llu != %llu)", (u_longlong_t)guid,
2001 (u_longlong_t)spa_guid(spa));
2005 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
2006 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
2010 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
2011 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2012 VDEV_AUX_CORRUPT_DATA);
2014 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2019 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
2021 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2022 VDEV_AUX_CORRUPT_DATA);
2024 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2025 ZPOOL_CONFIG_TOP_GUID);
2030 * If this vdev just became a top-level vdev because its sibling was
2031 * detached, it will have adopted the parent's vdev guid -- but the
2032 * label may or may not be on disk yet. Fortunately, either version
2033 * of the label will have the same top guid, so if we're a top-level
2034 * vdev, we can safely compare to that instead.
2035 * However, if the config comes from a cachefile that failed to update
2036 * after the detach, a top-level vdev will appear as a non top-level
2037 * vdev in the config. Also relax the constraints if we perform an
2040 * If we split this vdev off instead, then we also check the
2041 * original pool's guid. We don't want to consider the vdev
2042 * corrupt if it is partway through a split operation.
2044 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
2045 boolean_t mismatch = B_FALSE;
2046 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
2047 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
2050 if (vd->vdev_guid != top_guid &&
2051 vd->vdev_top->vdev_guid != guid)
2056 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2057 VDEV_AUX_CORRUPT_DATA);
2059 vdev_dbgmsg(vd, "vdev_validate: config guid "
2060 "doesn't match label guid");
2061 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
2062 (u_longlong_t)vd->vdev_guid,
2063 (u_longlong_t)vd->vdev_top->vdev_guid);
2064 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
2065 "aux_guid %llu", (u_longlong_t)guid,
2066 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
2071 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
2073 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2074 VDEV_AUX_CORRUPT_DATA);
2076 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2077 ZPOOL_CONFIG_POOL_STATE);
2084 * If this is a verbatim import, no need to check the
2085 * state of the pool.
2087 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
2088 spa_load_state(spa) == SPA_LOAD_OPEN &&
2089 state != POOL_STATE_ACTIVE) {
2090 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
2091 "for spa %s", (u_longlong_t)state, spa->spa_name);
2092 return (SET_ERROR(EBADF));
2096 * If we were able to open and validate a vdev that was
2097 * previously marked permanently unavailable, clear that state
2100 if (vd->vdev_not_present)
2101 vd->vdev_not_present = 0;
2107 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
2109 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
2110 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
2111 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2112 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
2113 dvd->vdev_path, svd->vdev_path);
2114 spa_strfree(dvd->vdev_path);
2115 dvd->vdev_path = spa_strdup(svd->vdev_path);
2117 } else if (svd->vdev_path != NULL) {
2118 dvd->vdev_path = spa_strdup(svd->vdev_path);
2119 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2120 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
2125 * Recursively copy vdev paths from one vdev to another. Source and destination
2126 * vdev trees must have same geometry otherwise return error. Intended to copy
2127 * paths from userland config into MOS config.
2130 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
2132 if ((svd->vdev_ops == &vdev_missing_ops) ||
2133 (svd->vdev_ishole && dvd->vdev_ishole) ||
2134 (dvd->vdev_ops == &vdev_indirect_ops))
2137 if (svd->vdev_ops != dvd->vdev_ops) {
2138 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
2139 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
2140 return (SET_ERROR(EINVAL));
2143 if (svd->vdev_guid != dvd->vdev_guid) {
2144 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
2145 "%llu)", (u_longlong_t)svd->vdev_guid,
2146 (u_longlong_t)dvd->vdev_guid);
2147 return (SET_ERROR(EINVAL));
2150 if (svd->vdev_children != dvd->vdev_children) {
2151 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2152 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2153 (u_longlong_t)dvd->vdev_children);
2154 return (SET_ERROR(EINVAL));
2157 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2158 int error = vdev_copy_path_strict(svd->vdev_child[i],
2159 dvd->vdev_child[i]);
2164 if (svd->vdev_ops->vdev_op_leaf)
2165 vdev_copy_path_impl(svd, dvd);
2171 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2173 ASSERT(stvd->vdev_top == stvd);
2174 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2176 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2177 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2180 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2184 * The idea here is that while a vdev can shift positions within
2185 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2186 * step outside of it.
2188 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2190 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2193 ASSERT(vd->vdev_ops->vdev_op_leaf);
2195 vdev_copy_path_impl(vd, dvd);
2199 * Recursively copy vdev paths from one root vdev to another. Source and
2200 * destination vdev trees may differ in geometry. For each destination leaf
2201 * vdev, search a vdev with the same guid and top vdev id in the source.
2202 * Intended to copy paths from userland config into MOS config.
2205 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2207 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2208 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2209 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2211 for (uint64_t i = 0; i < children; i++) {
2212 vdev_copy_path_search(srvd->vdev_child[i],
2213 drvd->vdev_child[i]);
2218 * Close a virtual device.
2221 vdev_close(vdev_t *vd)
2223 vdev_t *pvd = vd->vdev_parent;
2224 spa_t *spa __maybe_unused = vd->vdev_spa;
2226 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2229 * If our parent is reopening, then we are as well, unless we are
2232 if (pvd != NULL && pvd->vdev_reopening)
2233 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2235 vd->vdev_ops->vdev_op_close(vd);
2237 vdev_cache_purge(vd);
2240 * We record the previous state before we close it, so that if we are
2241 * doing a reopen(), we don't generate FMA ereports if we notice that
2242 * it's still faulted.
2244 vd->vdev_prevstate = vd->vdev_state;
2246 if (vd->vdev_offline)
2247 vd->vdev_state = VDEV_STATE_OFFLINE;
2249 vd->vdev_state = VDEV_STATE_CLOSED;
2250 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2254 vdev_hold(vdev_t *vd)
2256 spa_t *spa = vd->vdev_spa;
2258 ASSERT(spa_is_root(spa));
2259 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2262 for (int c = 0; c < vd->vdev_children; c++)
2263 vdev_hold(vd->vdev_child[c]);
2265 if (vd->vdev_ops->vdev_op_leaf)
2266 vd->vdev_ops->vdev_op_hold(vd);
2270 vdev_rele(vdev_t *vd)
2272 ASSERT(spa_is_root(vd->vdev_spa));
2273 for (int c = 0; c < vd->vdev_children; c++)
2274 vdev_rele(vd->vdev_child[c]);
2276 if (vd->vdev_ops->vdev_op_leaf)
2277 vd->vdev_ops->vdev_op_rele(vd);
2281 * Reopen all interior vdevs and any unopened leaves. We don't actually
2282 * reopen leaf vdevs which had previously been opened as they might deadlock
2283 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2284 * If the leaf has never been opened then open it, as usual.
2287 vdev_reopen(vdev_t *vd)
2289 spa_t *spa = vd->vdev_spa;
2291 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2293 /* set the reopening flag unless we're taking the vdev offline */
2294 vd->vdev_reopening = !vd->vdev_offline;
2296 (void) vdev_open(vd);
2299 * Call vdev_validate() here to make sure we have the same device.
2300 * Otherwise, a device with an invalid label could be successfully
2301 * opened in response to vdev_reopen().
2304 (void) vdev_validate_aux(vd);
2305 if (vdev_readable(vd) && vdev_writeable(vd) &&
2306 vd->vdev_aux == &spa->spa_l2cache) {
2308 * In case the vdev is present we should evict all ARC
2309 * buffers and pointers to log blocks and reclaim their
2310 * space before restoring its contents to L2ARC.
2312 if (l2arc_vdev_present(vd)) {
2313 l2arc_rebuild_vdev(vd, B_TRUE);
2315 l2arc_add_vdev(spa, vd);
2317 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
2318 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
2321 (void) vdev_validate(vd);
2325 * Reassess parent vdev's health.
2327 vdev_propagate_state(vd);
2331 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2336 * Normally, partial opens (e.g. of a mirror) are allowed.
2337 * For a create, however, we want to fail the request if
2338 * there are any components we can't open.
2340 error = vdev_open(vd);
2342 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2344 return (error ? error : SET_ERROR(ENXIO));
2348 * Recursively load DTLs and initialize all labels.
2350 if ((error = vdev_dtl_load(vd)) != 0 ||
2351 (error = vdev_label_init(vd, txg, isreplacing ?
2352 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2361 vdev_metaslab_set_size(vdev_t *vd)
2363 uint64_t asize = vd->vdev_asize;
2364 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2368 * There are two dimensions to the metaslab sizing calculation:
2369 * the size of the metaslab and the count of metaslabs per vdev.
2371 * The default values used below are a good balance between memory
2372 * usage (larger metaslab size means more memory needed for loaded
2373 * metaslabs; more metaslabs means more memory needed for the
2374 * metaslab_t structs), metaslab load time (larger metaslabs take
2375 * longer to load), and metaslab sync time (more metaslabs means
2376 * more time spent syncing all of them).
2378 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2379 * The range of the dimensions are as follows:
2381 * 2^29 <= ms_size <= 2^34
2382 * 16 <= ms_count <= 131,072
2384 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2385 * at least 512MB (2^29) to minimize fragmentation effects when
2386 * testing with smaller devices. However, the count constraint
2387 * of at least 16 metaslabs will override this minimum size goal.
2389 * On the upper end of vdev sizes, we aim for a maximum metaslab
2390 * size of 16GB. However, we will cap the total count to 2^17
2391 * metaslabs to keep our memory footprint in check and let the
2392 * metaslab size grow from there if that limit is hit.
2394 * The net effect of applying above constrains is summarized below.
2396 * vdev size metaslab count
2397 * --------------|-----------------
2399 * 8GB - 100GB one per 512MB
2401 * 3TB - 2PB one per 16GB
2403 * --------------------------------
2405 * Finally, note that all of the above calculate the initial
2406 * number of metaslabs. Expanding a top-level vdev will result
2407 * in additional metaslabs being allocated making it possible
2408 * to exceed the zfs_vdev_ms_count_limit.
2411 if (ms_count < zfs_vdev_min_ms_count)
2412 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2413 else if (ms_count > zfs_vdev_default_ms_count)
2414 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2416 ms_shift = zfs_vdev_default_ms_shift;
2418 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2419 ms_shift = SPA_MAXBLOCKSHIFT;
2420 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2421 ms_shift = zfs_vdev_max_ms_shift;
2422 /* cap the total count to constrain memory footprint */
2423 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2424 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2427 vd->vdev_ms_shift = ms_shift;
2428 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2432 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2434 ASSERT(vd == vd->vdev_top);
2435 /* indirect vdevs don't have metaslabs or dtls */
2436 ASSERT(vdev_is_concrete(vd) || flags == 0);
2437 ASSERT(ISP2(flags));
2438 ASSERT(spa_writeable(vd->vdev_spa));
2440 if (flags & VDD_METASLAB)
2441 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2443 if (flags & VDD_DTL)
2444 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2446 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2450 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2452 for (int c = 0; c < vd->vdev_children; c++)
2453 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2455 if (vd->vdev_ops->vdev_op_leaf)
2456 vdev_dirty(vd->vdev_top, flags, vd, txg);
2462 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2463 * the vdev has less than perfect replication. There are four kinds of DTL:
2465 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2467 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2469 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2470 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2471 * txgs that was scrubbed.
2473 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2474 * persistent errors or just some device being offline.
2475 * Unlike the other three, the DTL_OUTAGE map is not generally
2476 * maintained; it's only computed when needed, typically to
2477 * determine whether a device can be detached.
2479 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2480 * either has the data or it doesn't.
2482 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2483 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2484 * if any child is less than fully replicated, then so is its parent.
2485 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2486 * comprising only those txgs which appear in 'maxfaults' or more children;
2487 * those are the txgs we don't have enough replication to read. For example,
2488 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2489 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2490 * two child DTL_MISSING maps.
2492 * It should be clear from the above that to compute the DTLs and outage maps
2493 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2494 * Therefore, that is all we keep on disk. When loading the pool, or after
2495 * a configuration change, we generate all other DTLs from first principles.
2498 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2500 range_tree_t *rt = vd->vdev_dtl[t];
2502 ASSERT(t < DTL_TYPES);
2503 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2504 ASSERT(spa_writeable(vd->vdev_spa));
2506 mutex_enter(&vd->vdev_dtl_lock);
2507 if (!range_tree_contains(rt, txg, size))
2508 range_tree_add(rt, txg, size);
2509 mutex_exit(&vd->vdev_dtl_lock);
2513 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2515 range_tree_t *rt = vd->vdev_dtl[t];
2516 boolean_t dirty = B_FALSE;
2518 ASSERT(t < DTL_TYPES);
2519 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2522 * While we are loading the pool, the DTLs have not been loaded yet.
2523 * Ignore the DTLs and try all devices. This avoids a recursive
2524 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2525 * when loading the pool (relying on the checksum to ensure that
2526 * we get the right data -- note that we while loading, we are
2527 * only reading the MOS, which is always checksummed).
2529 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2532 mutex_enter(&vd->vdev_dtl_lock);
2533 if (!range_tree_is_empty(rt))
2534 dirty = range_tree_contains(rt, txg, size);
2535 mutex_exit(&vd->vdev_dtl_lock);
2541 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2543 range_tree_t *rt = vd->vdev_dtl[t];
2546 mutex_enter(&vd->vdev_dtl_lock);
2547 empty = range_tree_is_empty(rt);
2548 mutex_exit(&vd->vdev_dtl_lock);
2554 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2557 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2559 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2561 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2562 vd->vdev_ops->vdev_op_leaf)
2565 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2569 * Returns the lowest txg in the DTL range.
2572 vdev_dtl_min(vdev_t *vd)
2574 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2575 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2576 ASSERT0(vd->vdev_children);
2578 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1);
2582 * Returns the highest txg in the DTL.
2585 vdev_dtl_max(vdev_t *vd)
2587 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2588 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2589 ASSERT0(vd->vdev_children);
2591 return (range_tree_max(vd->vdev_dtl[DTL_MISSING]));
2595 * Determine if a resilvering vdev should remove any DTL entries from
2596 * its range. If the vdev was resilvering for the entire duration of the
2597 * scan then it should excise that range from its DTLs. Otherwise, this
2598 * vdev is considered partially resilvered and should leave its DTL
2599 * entries intact. The comment in vdev_dtl_reassess() describes how we
2603 vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done)
2605 ASSERT0(vd->vdev_children);
2607 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2610 if (vd->vdev_resilver_deferred)
2613 if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2617 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2618 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
2620 /* Rebuild not initiated by attach */
2621 if (vd->vdev_rebuild_txg == 0)
2625 * When a rebuild completes without error then all missing data
2626 * up to the rebuild max txg has been reconstructed and the DTL
2627 * is eligible for excision.
2629 if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE &&
2630 vdev_dtl_max(vd) <= vrp->vrp_max_txg) {
2631 ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd));
2632 ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg);
2633 ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg);
2637 dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan;
2638 dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys;
2640 /* Resilver not initiated by attach */
2641 if (vd->vdev_resilver_txg == 0)
2645 * When a resilver is initiated the scan will assign the
2646 * scn_max_txg value to the highest txg value that exists
2647 * in all DTLs. If this device's max DTL is not part of this
2648 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
2649 * then it is not eligible for excision.
2651 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2652 ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd));
2653 ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg);
2654 ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg);
2663 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2664 * write operations will be issued to the pool.
2667 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
2668 boolean_t scrub_done, boolean_t rebuild_done)
2670 spa_t *spa = vd->vdev_spa;
2674 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2676 for (int c = 0; c < vd->vdev_children; c++)
2677 vdev_dtl_reassess(vd->vdev_child[c], txg,
2678 scrub_txg, scrub_done, rebuild_done);
2680 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2683 if (vd->vdev_ops->vdev_op_leaf) {
2684 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2685 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2686 boolean_t check_excise = B_FALSE;
2687 boolean_t wasempty = B_TRUE;
2689 mutex_enter(&vd->vdev_dtl_lock);
2692 * If requested, pretend the scan or rebuild completed cleanly.
2694 if (zfs_scan_ignore_errors) {
2696 scn->scn_phys.scn_errors = 0;
2698 vr->vr_rebuild_phys.vrp_errors = 0;
2701 if (scrub_txg != 0 &&
2702 !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
2704 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
2705 "dtl:%llu/%llu errors:%llu",
2706 (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg,
2707 (u_longlong_t)scrub_txg, spa->spa_scrub_started,
2708 (u_longlong_t)vdev_dtl_min(vd),
2709 (u_longlong_t)vdev_dtl_max(vd),
2710 (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0));
2714 * If we've completed a scrub/resilver or a rebuild cleanly
2715 * then determine if this vdev should remove any DTLs. We
2716 * only want to excise regions on vdevs that were available
2717 * during the entire duration of this scan.
2720 vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) {
2721 check_excise = B_TRUE;
2723 if (spa->spa_scrub_started ||
2724 (scn != NULL && scn->scn_phys.scn_errors == 0)) {
2725 check_excise = B_TRUE;
2729 if (scrub_txg && check_excise &&
2730 vdev_dtl_should_excise(vd, rebuild_done)) {
2732 * We completed a scrub, resilver or rebuild up to
2733 * scrub_txg. If we did it without rebooting, then
2734 * the scrub dtl will be valid, so excise the old
2735 * region and fold in the scrub dtl. Otherwise,
2736 * leave the dtl as-is if there was an error.
2738 * There's little trick here: to excise the beginning
2739 * of the DTL_MISSING map, we put it into a reference
2740 * tree and then add a segment with refcnt -1 that
2741 * covers the range [0, scrub_txg). This means
2742 * that each txg in that range has refcnt -1 or 0.
2743 * We then add DTL_SCRUB with a refcnt of 2, so that
2744 * entries in the range [0, scrub_txg) will have a
2745 * positive refcnt -- either 1 or 2. We then convert
2746 * the reference tree into the new DTL_MISSING map.
2748 space_reftree_create(&reftree);
2749 space_reftree_add_map(&reftree,
2750 vd->vdev_dtl[DTL_MISSING], 1);
2751 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2752 space_reftree_add_map(&reftree,
2753 vd->vdev_dtl[DTL_SCRUB], 2);
2754 space_reftree_generate_map(&reftree,
2755 vd->vdev_dtl[DTL_MISSING], 1);
2756 space_reftree_destroy(&reftree);
2758 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
2759 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
2760 (u_longlong_t)vdev_dtl_min(vd),
2761 (u_longlong_t)vdev_dtl_max(vd));
2762 } else if (!wasempty) {
2763 zfs_dbgmsg("DTL_MISSING is now empty");
2766 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2767 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2768 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2770 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2771 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2772 if (!vdev_readable(vd))
2773 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2775 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2776 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2779 * If the vdev was resilvering or rebuilding and no longer
2780 * has any DTLs then reset the appropriate flag and dirty
2781 * the top level so that we persist the change.
2784 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2785 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2786 if (vd->vdev_rebuild_txg != 0) {
2787 vd->vdev_rebuild_txg = 0;
2788 vdev_config_dirty(vd->vdev_top);
2789 } else if (vd->vdev_resilver_txg != 0) {
2790 vd->vdev_resilver_txg = 0;
2791 vdev_config_dirty(vd->vdev_top);
2795 mutex_exit(&vd->vdev_dtl_lock);
2798 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2802 mutex_enter(&vd->vdev_dtl_lock);
2803 for (int t = 0; t < DTL_TYPES; t++) {
2804 /* account for child's outage in parent's missing map */
2805 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2807 continue; /* leaf vdevs only */
2808 if (t == DTL_PARTIAL)
2809 minref = 1; /* i.e. non-zero */
2810 else if (vd->vdev_nparity != 0)
2811 minref = vd->vdev_nparity + 1; /* RAID-Z */
2813 minref = vd->vdev_children; /* any kind of mirror */
2814 space_reftree_create(&reftree);
2815 for (int c = 0; c < vd->vdev_children; c++) {
2816 vdev_t *cvd = vd->vdev_child[c];
2817 mutex_enter(&cvd->vdev_dtl_lock);
2818 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2819 mutex_exit(&cvd->vdev_dtl_lock);
2821 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2822 space_reftree_destroy(&reftree);
2824 mutex_exit(&vd->vdev_dtl_lock);
2828 vdev_dtl_load(vdev_t *vd)
2830 spa_t *spa = vd->vdev_spa;
2831 objset_t *mos = spa->spa_meta_objset;
2834 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2835 ASSERT(vdev_is_concrete(vd));
2837 error = space_map_open(&vd->vdev_dtl_sm, mos,
2838 vd->vdev_dtl_object, 0, -1ULL, 0);
2841 ASSERT(vd->vdev_dtl_sm != NULL);
2843 mutex_enter(&vd->vdev_dtl_lock);
2844 error = space_map_load(vd->vdev_dtl_sm,
2845 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2846 mutex_exit(&vd->vdev_dtl_lock);
2851 for (int c = 0; c < vd->vdev_children; c++) {
2852 error = vdev_dtl_load(vd->vdev_child[c]);
2861 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
2863 spa_t *spa = vd->vdev_spa;
2864 objset_t *mos = spa->spa_meta_objset;
2865 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
2868 ASSERT(alloc_bias != VDEV_BIAS_NONE);
2871 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
2872 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
2873 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
2875 ASSERT(string != NULL);
2876 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
2877 1, strlen(string) + 1, string, tx));
2879 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
2880 spa_activate_allocation_classes(spa, tx);
2885 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2887 spa_t *spa = vd->vdev_spa;
2889 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2890 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2895 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2897 spa_t *spa = vd->vdev_spa;
2898 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2899 DMU_OT_NONE, 0, tx);
2902 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2909 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2911 if (vd->vdev_ops != &vdev_hole_ops &&
2912 vd->vdev_ops != &vdev_missing_ops &&
2913 vd->vdev_ops != &vdev_root_ops &&
2914 !vd->vdev_top->vdev_removing) {
2915 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2916 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2918 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2919 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2920 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
2921 vdev_zap_allocation_data(vd, tx);
2925 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2926 vdev_construct_zaps(vd->vdev_child[i], tx);
2931 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2933 spa_t *spa = vd->vdev_spa;
2934 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2935 objset_t *mos = spa->spa_meta_objset;
2936 range_tree_t *rtsync;
2938 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2940 ASSERT(vdev_is_concrete(vd));
2941 ASSERT(vd->vdev_ops->vdev_op_leaf);
2943 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2945 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2946 mutex_enter(&vd->vdev_dtl_lock);
2947 space_map_free(vd->vdev_dtl_sm, tx);
2948 space_map_close(vd->vdev_dtl_sm);
2949 vd->vdev_dtl_sm = NULL;
2950 mutex_exit(&vd->vdev_dtl_lock);
2953 * We only destroy the leaf ZAP for detached leaves or for
2954 * removed log devices. Removed data devices handle leaf ZAP
2955 * cleanup later, once cancellation is no longer possible.
2957 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2958 vd->vdev_top->vdev_islog)) {
2959 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2960 vd->vdev_leaf_zap = 0;
2967 if (vd->vdev_dtl_sm == NULL) {
2968 uint64_t new_object;
2970 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx);
2971 VERIFY3U(new_object, !=, 0);
2973 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2975 ASSERT(vd->vdev_dtl_sm != NULL);
2978 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
2980 mutex_enter(&vd->vdev_dtl_lock);
2981 range_tree_walk(rt, range_tree_add, rtsync);
2982 mutex_exit(&vd->vdev_dtl_lock);
2984 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx);
2985 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
2986 range_tree_vacate(rtsync, NULL, NULL);
2988 range_tree_destroy(rtsync);
2991 * If the object for the space map has changed then dirty
2992 * the top level so that we update the config.
2994 if (object != space_map_object(vd->vdev_dtl_sm)) {
2995 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2996 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2997 (u_longlong_t)object,
2998 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2999 vdev_config_dirty(vd->vdev_top);
3006 * Determine whether the specified vdev can be offlined/detached/removed
3007 * without losing data.
3010 vdev_dtl_required(vdev_t *vd)
3012 spa_t *spa = vd->vdev_spa;
3013 vdev_t *tvd = vd->vdev_top;
3014 uint8_t cant_read = vd->vdev_cant_read;
3017 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3019 if (vd == spa->spa_root_vdev || vd == tvd)
3023 * Temporarily mark the device as unreadable, and then determine
3024 * whether this results in any DTL outages in the top-level vdev.
3025 * If not, we can safely offline/detach/remove the device.
3027 vd->vdev_cant_read = B_TRUE;
3028 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3029 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
3030 vd->vdev_cant_read = cant_read;
3031 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3033 if (!required && zio_injection_enabled) {
3034 required = !!zio_handle_device_injection(vd, NULL,
3042 * Determine if resilver is needed, and if so the txg range.
3045 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
3047 boolean_t needed = B_FALSE;
3048 uint64_t thismin = UINT64_MAX;
3049 uint64_t thismax = 0;
3051 if (vd->vdev_children == 0) {
3052 mutex_enter(&vd->vdev_dtl_lock);
3053 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
3054 vdev_writeable(vd)) {
3056 thismin = vdev_dtl_min(vd);
3057 thismax = vdev_dtl_max(vd);
3060 mutex_exit(&vd->vdev_dtl_lock);
3062 for (int c = 0; c < vd->vdev_children; c++) {
3063 vdev_t *cvd = vd->vdev_child[c];
3064 uint64_t cmin, cmax;
3066 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
3067 thismin = MIN(thismin, cmin);
3068 thismax = MAX(thismax, cmax);
3074 if (needed && minp) {
3082 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3083 * will contain either the checkpoint spacemap object or zero if none exists.
3084 * All other errors are returned to the caller.
3087 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj)
3089 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
3091 if (vd->vdev_top_zap == 0) {
3096 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
3097 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj);
3098 if (error == ENOENT) {
3107 vdev_load(vdev_t *vd)
3112 * Recursively load all children.
3114 for (int c = 0; c < vd->vdev_children; c++) {
3115 error = vdev_load(vd->vdev_child[c]);
3121 vdev_set_deflate_ratio(vd);
3124 * On spa_load path, grab the allocation bias from our zap
3126 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3127 spa_t *spa = vd->vdev_spa;
3130 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
3131 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
3134 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
3135 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
3136 } else if (error != ENOENT) {
3137 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3138 VDEV_AUX_CORRUPT_DATA);
3139 vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) "
3140 "failed [error=%d]", vd->vdev_top_zap, error);
3146 * Load any rebuild state from the top-level vdev zap.
3148 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3149 error = vdev_rebuild_load(vd);
3150 if (error && error != ENOTSUP) {
3151 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3152 VDEV_AUX_CORRUPT_DATA);
3153 vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load "
3154 "failed [error=%d]", error);
3160 * If this is a top-level vdev, initialize its metaslabs.
3162 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
3163 vdev_metaslab_group_create(vd);
3165 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
3166 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3167 VDEV_AUX_CORRUPT_DATA);
3168 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
3169 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
3170 (u_longlong_t)vd->vdev_asize);
3171 return (SET_ERROR(ENXIO));
3174 error = vdev_metaslab_init(vd, 0);
3176 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
3177 "[error=%d]", error);
3178 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3179 VDEV_AUX_CORRUPT_DATA);
3183 uint64_t checkpoint_sm_obj;
3184 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj);
3185 if (error == 0 && checkpoint_sm_obj != 0) {
3186 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3187 ASSERT(vd->vdev_asize != 0);
3188 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
3190 error = space_map_open(&vd->vdev_checkpoint_sm,
3191 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
3194 vdev_dbgmsg(vd, "vdev_load: space_map_open "
3195 "failed for checkpoint spacemap (obj %llu) "
3197 (u_longlong_t)checkpoint_sm_obj, error);
3200 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
3203 * Since the checkpoint_sm contains free entries
3204 * exclusively we can use space_map_allocated() to
3205 * indicate the cumulative checkpointed space that
3208 vd->vdev_stat.vs_checkpoint_space =
3209 -space_map_allocated(vd->vdev_checkpoint_sm);
3210 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
3211 vd->vdev_stat.vs_checkpoint_space;
3212 } else if (error != 0) {
3213 vdev_dbgmsg(vd, "vdev_load: failed to retrieve "
3214 "checkpoint space map object from vdev ZAP "
3215 "[error=%d]", error);
3221 * If this is a leaf vdev, load its DTL.
3223 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
3224 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3225 VDEV_AUX_CORRUPT_DATA);
3226 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
3227 "[error=%d]", error);
3231 uint64_t obsolete_sm_object;
3232 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object);
3233 if (error == 0 && obsolete_sm_object != 0) {
3234 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3235 ASSERT(vd->vdev_asize != 0);
3236 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
3238 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
3239 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
3240 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3241 VDEV_AUX_CORRUPT_DATA);
3242 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
3243 "obsolete spacemap (obj %llu) [error=%d]",
3244 (u_longlong_t)obsolete_sm_object, error);
3247 } else if (error != 0) {
3248 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete "
3249 "space map object from vdev ZAP [error=%d]", error);
3257 * The special vdev case is used for hot spares and l2cache devices. Its
3258 * sole purpose it to set the vdev state for the associated vdev. To do this,
3259 * we make sure that we can open the underlying device, then try to read the
3260 * label, and make sure that the label is sane and that it hasn't been
3261 * repurposed to another pool.
3264 vdev_validate_aux(vdev_t *vd)
3267 uint64_t guid, version;
3270 if (!vdev_readable(vd))
3273 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3274 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3275 VDEV_AUX_CORRUPT_DATA);
3279 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3280 !SPA_VERSION_IS_SUPPORTED(version) ||
3281 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3282 guid != vd->vdev_guid ||
3283 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3284 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3285 VDEV_AUX_CORRUPT_DATA);
3291 * We don't actually check the pool state here. If it's in fact in
3292 * use by another pool, we update this fact on the fly when requested.
3299 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx)
3301 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3303 if (vd->vdev_top_zap == 0)
3306 uint64_t object = 0;
3307 int err = zap_lookup(mos, vd->vdev_top_zap,
3308 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object);
3313 VERIFY0(dmu_object_free(mos, object, tx));
3314 VERIFY0(zap_remove(mos, vd->vdev_top_zap,
3315 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx));
3319 * Free the objects used to store this vdev's spacemaps, and the array
3320 * that points to them.
3323 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3325 if (vd->vdev_ms_array == 0)
3328 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3329 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3330 size_t array_bytes = array_count * sizeof (uint64_t);
3331 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3332 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3333 array_bytes, smobj_array, 0));
3335 for (uint64_t i = 0; i < array_count; i++) {
3336 uint64_t smobj = smobj_array[i];
3340 space_map_free_obj(mos, smobj, tx);
3343 kmem_free(smobj_array, array_bytes);
3344 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3345 vdev_destroy_ms_flush_data(vd, tx);
3346 vd->vdev_ms_array = 0;
3350 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3352 spa_t *spa = vd->vdev_spa;
3354 ASSERT(vd->vdev_islog);
3355 ASSERT(vd == vd->vdev_top);
3356 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3358 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3360 vdev_destroy_spacemaps(vd, tx);
3361 if (vd->vdev_top_zap != 0) {
3362 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3363 vd->vdev_top_zap = 0;
3370 vdev_sync_done(vdev_t *vd, uint64_t txg)
3373 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3375 ASSERT(vdev_is_concrete(vd));
3377 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3379 metaslab_sync_done(msp, txg);
3382 metaslab_sync_reassess(vd->vdev_mg);
3386 vdev_sync(vdev_t *vd, uint64_t txg)
3388 spa_t *spa = vd->vdev_spa;
3392 ASSERT3U(txg, ==, spa->spa_syncing_txg);
3393 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3394 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3395 ASSERT(vd->vdev_removing ||
3396 vd->vdev_ops == &vdev_indirect_ops);
3398 vdev_indirect_sync_obsolete(vd, tx);
3401 * If the vdev is indirect, it can't have dirty
3402 * metaslabs or DTLs.
3404 if (vd->vdev_ops == &vdev_indirect_ops) {
3405 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3406 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3412 ASSERT(vdev_is_concrete(vd));
3414 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3415 !vd->vdev_removing) {
3416 ASSERT(vd == vd->vdev_top);
3417 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3418 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3419 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3420 ASSERT(vd->vdev_ms_array != 0);
3421 vdev_config_dirty(vd);
3424 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3425 metaslab_sync(msp, txg);
3426 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3429 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3430 vdev_dtl_sync(lvd, txg);
3433 * If this is an empty log device being removed, destroy the
3434 * metadata associated with it.
3436 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3437 vdev_remove_empty_log(vd, txg);
3439 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3444 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3446 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3450 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3451 * not be opened, and no I/O is attempted.
3454 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3458 spa_vdev_state_enter(spa, SCL_NONE);
3460 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3461 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3463 if (!vd->vdev_ops->vdev_op_leaf)
3464 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3469 * If user did a 'zpool offline -f' then make the fault persist across
3472 if (aux == VDEV_AUX_EXTERNAL_PERSIST) {
3474 * There are two kinds of forced faults: temporary and
3475 * persistent. Temporary faults go away at pool import, while
3476 * persistent faults stay set. Both types of faults can be
3477 * cleared with a zpool clear.
3479 * We tell if a vdev is persistently faulted by looking at the
3480 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3481 * import then it's a persistent fault. Otherwise, it's
3482 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3483 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3484 * tells vdev_config_generate() (which gets run later) to set
3485 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3487 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
3488 vd->vdev_tmpoffline = B_FALSE;
3489 aux = VDEV_AUX_EXTERNAL;
3491 vd->vdev_tmpoffline = B_TRUE;
3495 * We don't directly use the aux state here, but if we do a
3496 * vdev_reopen(), we need this value to be present to remember why we
3499 vd->vdev_label_aux = aux;
3502 * Faulted state takes precedence over degraded.
3504 vd->vdev_delayed_close = B_FALSE;
3505 vd->vdev_faulted = 1ULL;
3506 vd->vdev_degraded = 0ULL;
3507 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3510 * If this device has the only valid copy of the data, then
3511 * back off and simply mark the vdev as degraded instead.
3513 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3514 vd->vdev_degraded = 1ULL;
3515 vd->vdev_faulted = 0ULL;
3518 * If we reopen the device and it's not dead, only then do we
3523 if (vdev_readable(vd))
3524 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3527 return (spa_vdev_state_exit(spa, vd, 0));
3531 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3532 * user that something is wrong. The vdev continues to operate as normal as far
3533 * as I/O is concerned.
3536 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3540 spa_vdev_state_enter(spa, SCL_NONE);
3542 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3543 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3545 if (!vd->vdev_ops->vdev_op_leaf)
3546 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3549 * If the vdev is already faulted, then don't do anything.
3551 if (vd->vdev_faulted || vd->vdev_degraded)
3552 return (spa_vdev_state_exit(spa, NULL, 0));
3554 vd->vdev_degraded = 1ULL;
3555 if (!vdev_is_dead(vd))
3556 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3559 return (spa_vdev_state_exit(spa, vd, 0));
3563 * Online the given vdev.
3565 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3566 * spare device should be detached when the device finishes resilvering.
3567 * Second, the online should be treated like a 'test' online case, so no FMA
3568 * events are generated if the device fails to open.
3571 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3573 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3574 boolean_t wasoffline;
3575 vdev_state_t oldstate;
3577 spa_vdev_state_enter(spa, SCL_NONE);
3579 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3580 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3582 if (!vd->vdev_ops->vdev_op_leaf)
3583 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3585 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3586 oldstate = vd->vdev_state;
3589 vd->vdev_offline = B_FALSE;
3590 vd->vdev_tmpoffline = B_FALSE;
3591 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3592 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3594 /* XXX - L2ARC 1.0 does not support expansion */
3595 if (!vd->vdev_aux) {
3596 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3597 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
3598 spa->spa_autoexpand);
3599 vd->vdev_expansion_time = gethrestime_sec();
3603 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3605 if (!vd->vdev_aux) {
3606 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3607 pvd->vdev_expanding = B_FALSE;
3611 *newstate = vd->vdev_state;
3612 if ((flags & ZFS_ONLINE_UNSPARE) &&
3613 !vdev_is_dead(vd) && vd->vdev_parent &&
3614 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3615 vd->vdev_parent->vdev_child[0] == vd)
3616 vd->vdev_unspare = B_TRUE;
3618 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3620 /* XXX - L2ARC 1.0 does not support expansion */
3622 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3623 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3626 /* Restart initializing if necessary */
3627 mutex_enter(&vd->vdev_initialize_lock);
3628 if (vdev_writeable(vd) &&
3629 vd->vdev_initialize_thread == NULL &&
3630 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3631 (void) vdev_initialize(vd);
3633 mutex_exit(&vd->vdev_initialize_lock);
3636 * Restart trimming if necessary. We do not restart trimming for cache
3637 * devices here. This is triggered by l2arc_rebuild_vdev()
3638 * asynchronously for the whole device or in l2arc_evict() as it evicts
3639 * space for upcoming writes.
3641 mutex_enter(&vd->vdev_trim_lock);
3642 if (vdev_writeable(vd) && !vd->vdev_isl2cache &&
3643 vd->vdev_trim_thread == NULL &&
3644 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) {
3645 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial,
3646 vd->vdev_trim_secure);
3648 mutex_exit(&vd->vdev_trim_lock);
3651 (oldstate < VDEV_STATE_DEGRADED &&
3652 vd->vdev_state >= VDEV_STATE_DEGRADED))
3653 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3655 return (spa_vdev_state_exit(spa, vd, 0));
3659 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3663 uint64_t generation;
3664 metaslab_group_t *mg;
3667 spa_vdev_state_enter(spa, SCL_ALLOC);
3669 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3670 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3672 if (!vd->vdev_ops->vdev_op_leaf)
3673 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3677 generation = spa->spa_config_generation + 1;
3680 * If the device isn't already offline, try to offline it.
3682 if (!vd->vdev_offline) {
3684 * If this device has the only valid copy of some data,
3685 * don't allow it to be offlined. Log devices are always
3688 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3689 vdev_dtl_required(vd))
3690 return (spa_vdev_state_exit(spa, NULL,
3694 * If the top-level is a slog and it has had allocations
3695 * then proceed. We check that the vdev's metaslab group
3696 * is not NULL since it's possible that we may have just
3697 * added this vdev but not yet initialized its metaslabs.
3699 if (tvd->vdev_islog && mg != NULL) {
3701 * Prevent any future allocations.
3703 metaslab_group_passivate(mg);
3704 (void) spa_vdev_state_exit(spa, vd, 0);
3706 error = spa_reset_logs(spa);
3709 * If the log device was successfully reset but has
3710 * checkpointed data, do not offline it.
3713 tvd->vdev_checkpoint_sm != NULL) {
3714 ASSERT3U(space_map_allocated(
3715 tvd->vdev_checkpoint_sm), !=, 0);
3716 error = ZFS_ERR_CHECKPOINT_EXISTS;
3719 spa_vdev_state_enter(spa, SCL_ALLOC);
3722 * Check to see if the config has changed.
3724 if (error || generation != spa->spa_config_generation) {
3725 metaslab_group_activate(mg);
3727 return (spa_vdev_state_exit(spa,
3729 (void) spa_vdev_state_exit(spa, vd, 0);
3732 ASSERT0(tvd->vdev_stat.vs_alloc);
3736 * Offline this device and reopen its top-level vdev.
3737 * If the top-level vdev is a log device then just offline
3738 * it. Otherwise, if this action results in the top-level
3739 * vdev becoming unusable, undo it and fail the request.
3741 vd->vdev_offline = B_TRUE;
3744 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3745 vdev_is_dead(tvd)) {
3746 vd->vdev_offline = B_FALSE;
3748 return (spa_vdev_state_exit(spa, NULL,
3753 * Add the device back into the metaslab rotor so that
3754 * once we online the device it's open for business.
3756 if (tvd->vdev_islog && mg != NULL)
3757 metaslab_group_activate(mg);
3760 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3762 return (spa_vdev_state_exit(spa, vd, 0));
3766 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3770 mutex_enter(&spa->spa_vdev_top_lock);
3771 error = vdev_offline_locked(spa, guid, flags);
3772 mutex_exit(&spa->spa_vdev_top_lock);
3778 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3779 * vdev_offline(), we assume the spa config is locked. We also clear all
3780 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3783 vdev_clear(spa_t *spa, vdev_t *vd)
3785 vdev_t *rvd = spa->spa_root_vdev;
3787 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3792 vd->vdev_stat.vs_read_errors = 0;
3793 vd->vdev_stat.vs_write_errors = 0;
3794 vd->vdev_stat.vs_checksum_errors = 0;
3795 vd->vdev_stat.vs_slow_ios = 0;
3797 for (int c = 0; c < vd->vdev_children; c++)
3798 vdev_clear(spa, vd->vdev_child[c]);
3801 * It makes no sense to "clear" an indirect vdev.
3803 if (!vdev_is_concrete(vd))
3807 * If we're in the FAULTED state or have experienced failed I/O, then
3808 * clear the persistent state and attempt to reopen the device. We
3809 * also mark the vdev config dirty, so that the new faulted state is
3810 * written out to disk.
3812 if (vd->vdev_faulted || vd->vdev_degraded ||
3813 !vdev_readable(vd) || !vdev_writeable(vd)) {
3815 * When reopening in response to a clear event, it may be due to
3816 * a fmadm repair request. In this case, if the device is
3817 * still broken, we want to still post the ereport again.
3819 vd->vdev_forcefault = B_TRUE;
3821 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3822 vd->vdev_cant_read = B_FALSE;
3823 vd->vdev_cant_write = B_FALSE;
3824 vd->vdev_stat.vs_aux = 0;
3826 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3828 vd->vdev_forcefault = B_FALSE;
3830 if (vd != rvd && vdev_writeable(vd->vdev_top))
3831 vdev_state_dirty(vd->vdev_top);
3833 /* If a resilver isn't required, check if vdevs can be culled */
3834 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) &&
3835 !dsl_scan_resilvering(spa->spa_dsl_pool) &&
3836 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool))
3837 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3839 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3843 * When clearing a FMA-diagnosed fault, we always want to
3844 * unspare the device, as we assume that the original spare was
3845 * done in response to the FMA fault.
3847 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3848 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3849 vd->vdev_parent->vdev_child[0] == vd)
3850 vd->vdev_unspare = B_TRUE;
3854 vdev_is_dead(vdev_t *vd)
3857 * Holes and missing devices are always considered "dead".
3858 * This simplifies the code since we don't have to check for
3859 * these types of devices in the various code paths.
3860 * Instead we rely on the fact that we skip over dead devices
3861 * before issuing I/O to them.
3863 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3864 vd->vdev_ops == &vdev_hole_ops ||
3865 vd->vdev_ops == &vdev_missing_ops);
3869 vdev_readable(vdev_t *vd)
3871 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3875 vdev_writeable(vdev_t *vd)
3877 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3878 vdev_is_concrete(vd));
3882 vdev_allocatable(vdev_t *vd)
3884 uint64_t state = vd->vdev_state;
3887 * We currently allow allocations from vdevs which may be in the
3888 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3889 * fails to reopen then we'll catch it later when we're holding
3890 * the proper locks. Note that we have to get the vdev state
3891 * in a local variable because although it changes atomically,
3892 * we're asking two separate questions about it.
3894 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3895 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3896 vd->vdev_mg->mg_initialized);
3900 vdev_accessible(vdev_t *vd, zio_t *zio)
3902 ASSERT(zio->io_vd == vd);
3904 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3907 if (zio->io_type == ZIO_TYPE_READ)
3908 return (!vd->vdev_cant_read);
3910 if (zio->io_type == ZIO_TYPE_WRITE)
3911 return (!vd->vdev_cant_write);
3917 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs)
3919 for (int t = 0; t < VS_ZIO_TYPES; t++) {
3920 vs->vs_ops[t] += cvs->vs_ops[t];
3921 vs->vs_bytes[t] += cvs->vs_bytes[t];
3924 cvs->vs_scan_removing = cvd->vdev_removing;
3928 * Get extended stats
3931 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx)
3934 for (t = 0; t < ZIO_TYPES; t++) {
3935 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++)
3936 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b];
3938 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) {
3939 vsx->vsx_total_histo[t][b] +=
3940 cvsx->vsx_total_histo[t][b];
3944 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
3945 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) {
3946 vsx->vsx_queue_histo[t][b] +=
3947 cvsx->vsx_queue_histo[t][b];
3949 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t];
3950 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t];
3952 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++)
3953 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b];
3955 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++)
3956 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b];
3962 vdev_is_spacemap_addressable(vdev_t *vd)
3964 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
3968 * If double-word space map entries are not enabled we assume
3969 * 47 bits of the space map entry are dedicated to the entry's
3970 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
3971 * to calculate the maximum address that can be described by a
3972 * space map entry for the given device.
3974 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
3976 if (shift >= 63) /* detect potential overflow */
3979 return (vd->vdev_asize < (1ULL << shift));
3983 * Get statistics for the given vdev.
3986 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
3990 * If we're getting stats on the root vdev, aggregate the I/O counts
3991 * over all top-level vdevs (i.e. the direct children of the root).
3993 if (!vd->vdev_ops->vdev_op_leaf) {
3995 memset(vs->vs_ops, 0, sizeof (vs->vs_ops));
3996 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes));
3999 memset(vsx, 0, sizeof (*vsx));
4001 for (int c = 0; c < vd->vdev_children; c++) {
4002 vdev_t *cvd = vd->vdev_child[c];
4003 vdev_stat_t *cvs = &cvd->vdev_stat;
4004 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex;
4006 vdev_get_stats_ex_impl(cvd, cvs, cvsx);
4008 vdev_get_child_stat(cvd, vs, cvs);
4010 vdev_get_child_stat_ex(cvd, vsx, cvsx);
4015 * We're a leaf. Just copy our ZIO active queue stats in. The
4016 * other leaf stats are updated in vdev_stat_update().
4021 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex));
4023 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) {
4024 vsx->vsx_active_queue[t] =
4025 vd->vdev_queue.vq_class[t].vqc_active;
4026 vsx->vsx_pend_queue[t] = avl_numnodes(
4027 &vd->vdev_queue.vq_class[t].vqc_queued_tree);
4033 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4035 vdev_t *tvd = vd->vdev_top;
4036 mutex_enter(&vd->vdev_stat_lock);
4038 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
4039 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
4040 vs->vs_state = vd->vdev_state;
4041 vs->vs_rsize = vdev_get_min_asize(vd);
4043 if (vd->vdev_ops->vdev_op_leaf) {
4044 vs->vs_rsize += VDEV_LABEL_START_SIZE +
4045 VDEV_LABEL_END_SIZE;
4047 * Report initializing progress. Since we don't
4048 * have the initializing locks held, this is only
4049 * an estimate (although a fairly accurate one).
4051 vs->vs_initialize_bytes_done =
4052 vd->vdev_initialize_bytes_done;
4053 vs->vs_initialize_bytes_est =
4054 vd->vdev_initialize_bytes_est;
4055 vs->vs_initialize_state = vd->vdev_initialize_state;
4056 vs->vs_initialize_action_time =
4057 vd->vdev_initialize_action_time;
4060 * Report manual TRIM progress. Since we don't have
4061 * the manual TRIM locks held, this is only an
4062 * estimate (although fairly accurate one).
4064 vs->vs_trim_notsup = !vd->vdev_has_trim;
4065 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done;
4066 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est;
4067 vs->vs_trim_state = vd->vdev_trim_state;
4068 vs->vs_trim_action_time = vd->vdev_trim_action_time;
4070 /* Set when there is a deferred resilver. */
4071 vs->vs_resilver_deferred = vd->vdev_resilver_deferred;
4075 * Report expandable space on top-level, non-auxiliary devices
4076 * only. The expandable space is reported in terms of metaslab
4077 * sized units since that determines how much space the pool
4080 if (vd->vdev_aux == NULL && tvd != NULL) {
4081 vs->vs_esize = P2ALIGN(
4082 vd->vdev_max_asize - vd->vdev_asize,
4083 1ULL << tvd->vdev_ms_shift);
4087 * Report fragmentation and rebuild progress for top-level,
4088 * non-auxiliary, concrete devices.
4090 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
4091 vdev_is_concrete(vd)) {
4092 vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
4093 vd->vdev_mg->mg_fragmentation : 0;
4097 vdev_get_stats_ex_impl(vd, vs, vsx);
4098 mutex_exit(&vd->vdev_stat_lock);
4102 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
4104 return (vdev_get_stats_ex(vd, vs, NULL));
4108 vdev_clear_stats(vdev_t *vd)
4110 mutex_enter(&vd->vdev_stat_lock);
4111 vd->vdev_stat.vs_space = 0;
4112 vd->vdev_stat.vs_dspace = 0;
4113 vd->vdev_stat.vs_alloc = 0;
4114 mutex_exit(&vd->vdev_stat_lock);
4118 vdev_scan_stat_init(vdev_t *vd)
4120 vdev_stat_t *vs = &vd->vdev_stat;
4122 for (int c = 0; c < vd->vdev_children; c++)
4123 vdev_scan_stat_init(vd->vdev_child[c]);
4125 mutex_enter(&vd->vdev_stat_lock);
4126 vs->vs_scan_processed = 0;
4127 mutex_exit(&vd->vdev_stat_lock);
4131 vdev_stat_update(zio_t *zio, uint64_t psize)
4133 spa_t *spa = zio->io_spa;
4134 vdev_t *rvd = spa->spa_root_vdev;
4135 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
4137 uint64_t txg = zio->io_txg;
4138 vdev_stat_t *vs = &vd->vdev_stat;
4139 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex;
4140 zio_type_t type = zio->io_type;
4141 int flags = zio->io_flags;
4144 * If this i/o is a gang leader, it didn't do any actual work.
4146 if (zio->io_gang_tree)
4149 if (zio->io_error == 0) {
4151 * If this is a root i/o, don't count it -- we've already
4152 * counted the top-level vdevs, and vdev_get_stats() will
4153 * aggregate them when asked. This reduces contention on
4154 * the root vdev_stat_lock and implicitly handles blocks
4155 * that compress away to holes, for which there is no i/o.
4156 * (Holes never create vdev children, so all the counters
4157 * remain zero, which is what we want.)
4159 * Note: this only applies to successful i/o (io_error == 0)
4160 * because unlike i/o counts, errors are not additive.
4161 * When reading a ditto block, for example, failure of
4162 * one top-level vdev does not imply a root-level error.
4167 ASSERT(vd == zio->io_vd);
4169 if (flags & ZIO_FLAG_IO_BYPASS)
4172 mutex_enter(&vd->vdev_stat_lock);
4174 if (flags & ZIO_FLAG_IO_REPAIR) {
4176 * Repair is the result of a resilver issued by the
4177 * scan thread (spa_sync).
4179 if (flags & ZIO_FLAG_SCAN_THREAD) {
4180 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
4181 dsl_scan_phys_t *scn_phys = &scn->scn_phys;
4182 uint64_t *processed = &scn_phys->scn_processed;
4184 if (vd->vdev_ops->vdev_op_leaf)
4185 atomic_add_64(processed, psize);
4186 vs->vs_scan_processed += psize;
4190 * Repair is the result of a rebuild issued by the
4191 * rebuild thread (vdev_rebuild_thread).
4193 if (zio->io_priority == ZIO_PRIORITY_REBUILD) {
4194 vdev_t *tvd = vd->vdev_top;
4195 vdev_rebuild_t *vr = &tvd->vdev_rebuild_config;
4196 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
4197 uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt;
4199 if (vd->vdev_ops->vdev_op_leaf)
4200 atomic_add_64(rebuilt, psize);
4201 vs->vs_rebuild_processed += psize;
4204 if (flags & ZIO_FLAG_SELF_HEAL)
4205 vs->vs_self_healed += psize;
4209 * The bytes/ops/histograms are recorded at the leaf level and
4210 * aggregated into the higher level vdevs in vdev_get_stats().
4212 if (vd->vdev_ops->vdev_op_leaf &&
4213 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) {
4214 zio_type_t vs_type = type;
4215 zio_priority_t priority = zio->io_priority;
4218 * TRIM ops and bytes are reported to user space as
4219 * ZIO_TYPE_IOCTL. This is done to preserve the
4220 * vdev_stat_t structure layout for user space.
4222 if (type == ZIO_TYPE_TRIM)
4223 vs_type = ZIO_TYPE_IOCTL;
4226 * Solely for the purposes of 'zpool iostat -lqrw'
4227 * reporting use the priority to catagorize the IO.
4228 * Only the following are reported to user space:
4230 * ZIO_PRIORITY_SYNC_READ,
4231 * ZIO_PRIORITY_SYNC_WRITE,
4232 * ZIO_PRIORITY_ASYNC_READ,
4233 * ZIO_PRIORITY_ASYNC_WRITE,
4234 * ZIO_PRIORITY_SCRUB,
4235 * ZIO_PRIORITY_TRIM.
4237 if (priority == ZIO_PRIORITY_REBUILD) {
4238 priority = ((type == ZIO_TYPE_WRITE) ?
4239 ZIO_PRIORITY_ASYNC_WRITE :
4240 ZIO_PRIORITY_SCRUB);
4241 } else if (priority == ZIO_PRIORITY_INITIALIZING) {
4242 ASSERT3U(type, ==, ZIO_TYPE_WRITE);
4243 priority = ZIO_PRIORITY_ASYNC_WRITE;
4244 } else if (priority == ZIO_PRIORITY_REMOVAL) {
4245 priority = ((type == ZIO_TYPE_WRITE) ?
4246 ZIO_PRIORITY_ASYNC_WRITE :
4247 ZIO_PRIORITY_ASYNC_READ);
4250 vs->vs_ops[vs_type]++;
4251 vs->vs_bytes[vs_type] += psize;
4253 if (flags & ZIO_FLAG_DELEGATED) {
4254 vsx->vsx_agg_histo[priority]
4255 [RQ_HISTO(zio->io_size)]++;
4257 vsx->vsx_ind_histo[priority]
4258 [RQ_HISTO(zio->io_size)]++;
4261 if (zio->io_delta && zio->io_delay) {
4262 vsx->vsx_queue_histo[priority]
4263 [L_HISTO(zio->io_delta - zio->io_delay)]++;
4264 vsx->vsx_disk_histo[type]
4265 [L_HISTO(zio->io_delay)]++;
4266 vsx->vsx_total_histo[type]
4267 [L_HISTO(zio->io_delta)]++;
4271 mutex_exit(&vd->vdev_stat_lock);
4275 if (flags & ZIO_FLAG_SPECULATIVE)
4279 * If this is an I/O error that is going to be retried, then ignore the
4280 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4281 * hard errors, when in reality they can happen for any number of
4282 * innocuous reasons (bus resets, MPxIO link failure, etc).
4284 if (zio->io_error == EIO &&
4285 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
4289 * Intent logs writes won't propagate their error to the root
4290 * I/O so don't mark these types of failures as pool-level
4293 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
4296 if (spa->spa_load_state == SPA_LOAD_NONE &&
4297 type == ZIO_TYPE_WRITE && txg != 0 &&
4298 (!(flags & ZIO_FLAG_IO_REPAIR) ||
4299 (flags & ZIO_FLAG_SCAN_THREAD) ||
4300 spa->spa_claiming)) {
4302 * This is either a normal write (not a repair), or it's
4303 * a repair induced by the scrub thread, or it's a repair
4304 * made by zil_claim() during spa_load() in the first txg.
4305 * In the normal case, we commit the DTL change in the same
4306 * txg as the block was born. In the scrub-induced repair
4307 * case, we know that scrubs run in first-pass syncing context,
4308 * so we commit the DTL change in spa_syncing_txg(spa).
4309 * In the zil_claim() case, we commit in spa_first_txg(spa).
4311 * We currently do not make DTL entries for failed spontaneous
4312 * self-healing writes triggered by normal (non-scrubbing)
4313 * reads, because we have no transactional context in which to
4314 * do so -- and it's not clear that it'd be desirable anyway.
4316 if (vd->vdev_ops->vdev_op_leaf) {
4317 uint64_t commit_txg = txg;
4318 if (flags & ZIO_FLAG_SCAN_THREAD) {
4319 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4320 ASSERT(spa_sync_pass(spa) == 1);
4321 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
4322 commit_txg = spa_syncing_txg(spa);
4323 } else if (spa->spa_claiming) {
4324 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4325 commit_txg = spa_first_txg(spa);
4327 ASSERT(commit_txg >= spa_syncing_txg(spa));
4328 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
4330 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
4331 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
4332 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
4335 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
4340 vdev_deflated_space(vdev_t *vd, int64_t space)
4342 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
4343 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
4345 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
4349 * Update the in-core space usage stats for this vdev, its metaslab class,
4350 * and the root vdev.
4353 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
4354 int64_t space_delta)
4356 int64_t dspace_delta;
4357 spa_t *spa = vd->vdev_spa;
4358 vdev_t *rvd = spa->spa_root_vdev;
4360 ASSERT(vd == vd->vdev_top);
4363 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4364 * factor. We must calculate this here and not at the root vdev
4365 * because the root vdev's psize-to-asize is simply the max of its
4366 * children's, thus not accurate enough for us.
4368 dspace_delta = vdev_deflated_space(vd, space_delta);
4370 mutex_enter(&vd->vdev_stat_lock);
4371 /* ensure we won't underflow */
4372 if (alloc_delta < 0) {
4373 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta);
4376 vd->vdev_stat.vs_alloc += alloc_delta;
4377 vd->vdev_stat.vs_space += space_delta;
4378 vd->vdev_stat.vs_dspace += dspace_delta;
4379 mutex_exit(&vd->vdev_stat_lock);
4381 /* every class but log contributes to root space stats */
4382 if (vd->vdev_mg != NULL && !vd->vdev_islog) {
4383 ASSERT(!vd->vdev_isl2cache);
4384 mutex_enter(&rvd->vdev_stat_lock);
4385 rvd->vdev_stat.vs_alloc += alloc_delta;
4386 rvd->vdev_stat.vs_space += space_delta;
4387 rvd->vdev_stat.vs_dspace += dspace_delta;
4388 mutex_exit(&rvd->vdev_stat_lock);
4390 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4394 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4395 * so that it will be written out next time the vdev configuration is synced.
4396 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4399 vdev_config_dirty(vdev_t *vd)
4401 spa_t *spa = vd->vdev_spa;
4402 vdev_t *rvd = spa->spa_root_vdev;
4405 ASSERT(spa_writeable(spa));
4408 * If this is an aux vdev (as with l2cache and spare devices), then we
4409 * update the vdev config manually and set the sync flag.
4411 if (vd->vdev_aux != NULL) {
4412 spa_aux_vdev_t *sav = vd->vdev_aux;
4416 for (c = 0; c < sav->sav_count; c++) {
4417 if (sav->sav_vdevs[c] == vd)
4421 if (c == sav->sav_count) {
4423 * We're being removed. There's nothing more to do.
4425 ASSERT(sav->sav_sync == B_TRUE);
4429 sav->sav_sync = B_TRUE;
4431 if (nvlist_lookup_nvlist_array(sav->sav_config,
4432 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
4433 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
4434 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
4440 * Setting the nvlist in the middle if the array is a little
4441 * sketchy, but it will work.
4443 nvlist_free(aux[c]);
4444 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
4450 * The dirty list is protected by the SCL_CONFIG lock. The caller
4451 * must either hold SCL_CONFIG as writer, or must be the sync thread
4452 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4453 * so this is sufficient to ensure mutual exclusion.
4455 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4456 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4457 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4460 for (c = 0; c < rvd->vdev_children; c++)
4461 vdev_config_dirty(rvd->vdev_child[c]);
4463 ASSERT(vd == vd->vdev_top);
4465 if (!list_link_active(&vd->vdev_config_dirty_node) &&
4466 vdev_is_concrete(vd)) {
4467 list_insert_head(&spa->spa_config_dirty_list, vd);
4473 vdev_config_clean(vdev_t *vd)
4475 spa_t *spa = vd->vdev_spa;
4477 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4478 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4479 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4481 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
4482 list_remove(&spa->spa_config_dirty_list, vd);
4486 * Mark a top-level vdev's state as dirty, so that the next pass of
4487 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4488 * the state changes from larger config changes because they require
4489 * much less locking, and are often needed for administrative actions.
4492 vdev_state_dirty(vdev_t *vd)
4494 spa_t *spa = vd->vdev_spa;
4496 ASSERT(spa_writeable(spa));
4497 ASSERT(vd == vd->vdev_top);
4500 * The state list is protected by the SCL_STATE lock. The caller
4501 * must either hold SCL_STATE as writer, or must be the sync thread
4502 * (which holds SCL_STATE as reader). There's only one sync thread,
4503 * so this is sufficient to ensure mutual exclusion.
4505 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4506 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4507 spa_config_held(spa, SCL_STATE, RW_READER)));
4509 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4510 vdev_is_concrete(vd))
4511 list_insert_head(&spa->spa_state_dirty_list, vd);
4515 vdev_state_clean(vdev_t *vd)
4517 spa_t *spa = vd->vdev_spa;
4519 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4520 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4521 spa_config_held(spa, SCL_STATE, RW_READER)));
4523 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4524 list_remove(&spa->spa_state_dirty_list, vd);
4528 * Propagate vdev state up from children to parent.
4531 vdev_propagate_state(vdev_t *vd)
4533 spa_t *spa = vd->vdev_spa;
4534 vdev_t *rvd = spa->spa_root_vdev;
4535 int degraded = 0, faulted = 0;
4539 if (vd->vdev_children > 0) {
4540 for (int c = 0; c < vd->vdev_children; c++) {
4541 child = vd->vdev_child[c];
4544 * Don't factor holes or indirect vdevs into the
4547 if (!vdev_is_concrete(child))
4550 if (!vdev_readable(child) ||
4551 (!vdev_writeable(child) && spa_writeable(spa))) {
4553 * Root special: if there is a top-level log
4554 * device, treat the root vdev as if it were
4557 if (child->vdev_islog && vd == rvd)
4561 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4565 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4569 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4572 * Root special: if there is a top-level vdev that cannot be
4573 * opened due to corrupted metadata, then propagate the root
4574 * vdev's aux state as 'corrupt' rather than 'insufficient
4577 if (corrupted && vd == rvd &&
4578 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4579 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4580 VDEV_AUX_CORRUPT_DATA);
4583 if (vd->vdev_parent)
4584 vdev_propagate_state(vd->vdev_parent);
4588 * Set a vdev's state. If this is during an open, we don't update the parent
4589 * state, because we're in the process of opening children depth-first.
4590 * Otherwise, we propagate the change to the parent.
4592 * If this routine places a device in a faulted state, an appropriate ereport is
4596 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4598 uint64_t save_state;
4599 spa_t *spa = vd->vdev_spa;
4601 if (state == vd->vdev_state) {
4603 * Since vdev_offline() code path is already in an offline
4604 * state we can miss a statechange event to OFFLINE. Check
4605 * the previous state to catch this condition.
4607 if (vd->vdev_ops->vdev_op_leaf &&
4608 (state == VDEV_STATE_OFFLINE) &&
4609 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) {
4610 /* post an offline state change */
4611 zfs_post_state_change(spa, vd, vd->vdev_prevstate);
4613 vd->vdev_stat.vs_aux = aux;
4617 save_state = vd->vdev_state;
4619 vd->vdev_state = state;
4620 vd->vdev_stat.vs_aux = aux;
4623 * If we are setting the vdev state to anything but an open state, then
4624 * always close the underlying device unless the device has requested
4625 * a delayed close (i.e. we're about to remove or fault the device).
4626 * Otherwise, we keep accessible but invalid devices open forever.
4627 * We don't call vdev_close() itself, because that implies some extra
4628 * checks (offline, etc) that we don't want here. This is limited to
4629 * leaf devices, because otherwise closing the device will affect other
4632 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4633 vd->vdev_ops->vdev_op_leaf)
4634 vd->vdev_ops->vdev_op_close(vd);
4636 if (vd->vdev_removed &&
4637 state == VDEV_STATE_CANT_OPEN &&
4638 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4640 * If the previous state is set to VDEV_STATE_REMOVED, then this
4641 * device was previously marked removed and someone attempted to
4642 * reopen it. If this failed due to a nonexistent device, then
4643 * keep the device in the REMOVED state. We also let this be if
4644 * it is one of our special test online cases, which is only
4645 * attempting to online the device and shouldn't generate an FMA
4648 vd->vdev_state = VDEV_STATE_REMOVED;
4649 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4650 } else if (state == VDEV_STATE_REMOVED) {
4651 vd->vdev_removed = B_TRUE;
4652 } else if (state == VDEV_STATE_CANT_OPEN) {
4654 * If we fail to open a vdev during an import or recovery, we
4655 * mark it as "not available", which signifies that it was
4656 * never there to begin with. Failure to open such a device
4657 * is not considered an error.
4659 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4660 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4661 vd->vdev_ops->vdev_op_leaf)
4662 vd->vdev_not_present = 1;
4665 * Post the appropriate ereport. If the 'prevstate' field is
4666 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4667 * that this is part of a vdev_reopen(). In this case, we don't
4668 * want to post the ereport if the device was already in the
4669 * CANT_OPEN state beforehand.
4671 * If the 'checkremove' flag is set, then this is an attempt to
4672 * online the device in response to an insertion event. If we
4673 * hit this case, then we have detected an insertion event for a
4674 * faulted or offline device that wasn't in the removed state.
4675 * In this scenario, we don't post an ereport because we are
4676 * about to replace the device, or attempt an online with
4677 * vdev_forcefault, which will generate the fault for us.
4679 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4680 !vd->vdev_not_present && !vd->vdev_checkremove &&
4681 vd != spa->spa_root_vdev) {
4685 case VDEV_AUX_OPEN_FAILED:
4686 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4688 case VDEV_AUX_CORRUPT_DATA:
4689 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4691 case VDEV_AUX_NO_REPLICAS:
4692 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4694 case VDEV_AUX_BAD_GUID_SUM:
4695 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4697 case VDEV_AUX_TOO_SMALL:
4698 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4700 case VDEV_AUX_BAD_LABEL:
4701 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4703 case VDEV_AUX_BAD_ASHIFT:
4704 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT;
4707 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4710 zfs_ereport_post(class, spa, vd, NULL, NULL,
4714 /* Erase any notion of persistent removed state */
4715 vd->vdev_removed = B_FALSE;
4717 vd->vdev_removed = B_FALSE;
4721 * Notify ZED of any significant state-change on a leaf vdev.
4724 if (vd->vdev_ops->vdev_op_leaf) {
4725 /* preserve original state from a vdev_reopen() */
4726 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) &&
4727 (vd->vdev_prevstate != vd->vdev_state) &&
4728 (save_state <= VDEV_STATE_CLOSED))
4729 save_state = vd->vdev_prevstate;
4731 /* filter out state change due to initial vdev_open */
4732 if (save_state > VDEV_STATE_CLOSED)
4733 zfs_post_state_change(spa, vd, save_state);
4736 if (!isopen && vd->vdev_parent)
4737 vdev_propagate_state(vd->vdev_parent);
4741 vdev_children_are_offline(vdev_t *vd)
4743 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4745 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4746 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4754 * Check the vdev configuration to ensure that it's capable of supporting
4755 * a root pool. We do not support partial configuration.
4758 vdev_is_bootable(vdev_t *vd)
4760 if (!vd->vdev_ops->vdev_op_leaf) {
4761 const char *vdev_type = vd->vdev_ops->vdev_op_type;
4763 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4764 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4769 for (int c = 0; c < vd->vdev_children; c++) {
4770 if (!vdev_is_bootable(vd->vdev_child[c]))
4777 vdev_is_concrete(vdev_t *vd)
4779 vdev_ops_t *ops = vd->vdev_ops;
4780 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4781 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4789 * Determine if a log device has valid content. If the vdev was
4790 * removed or faulted in the MOS config then we know that
4791 * the content on the log device has already been written to the pool.
4794 vdev_log_state_valid(vdev_t *vd)
4796 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4800 for (int c = 0; c < vd->vdev_children; c++)
4801 if (vdev_log_state_valid(vd->vdev_child[c]))
4808 * Expand a vdev if possible.
4811 vdev_expand(vdev_t *vd, uint64_t txg)
4813 ASSERT(vd->vdev_top == vd);
4814 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4815 ASSERT(vdev_is_concrete(vd));
4817 vdev_set_deflate_ratio(vd);
4819 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
4820 vdev_is_concrete(vd)) {
4821 vdev_metaslab_group_create(vd);
4822 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4823 vdev_config_dirty(vd);
4831 vdev_split(vdev_t *vd)
4833 vdev_t *cvd, *pvd = vd->vdev_parent;
4835 vdev_remove_child(pvd, vd);
4836 vdev_compact_children(pvd);
4838 cvd = pvd->vdev_child[0];
4839 if (pvd->vdev_children == 1) {
4840 vdev_remove_parent(cvd);
4841 cvd->vdev_splitting = B_TRUE;
4843 vdev_propagate_state(cvd);
4847 vdev_deadman(vdev_t *vd, char *tag)
4849 for (int c = 0; c < vd->vdev_children; c++) {
4850 vdev_t *cvd = vd->vdev_child[c];
4852 vdev_deadman(cvd, tag);
4855 if (vd->vdev_ops->vdev_op_leaf) {
4856 vdev_queue_t *vq = &vd->vdev_queue;
4858 mutex_enter(&vq->vq_lock);
4859 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4860 spa_t *spa = vd->vdev_spa;
4864 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4865 vd->vdev_path, avl_numnodes(&vq->vq_active_tree));
4868 * Look at the head of all the pending queues,
4869 * if any I/O has been outstanding for longer than
4870 * the spa_deadman_synctime invoke the deadman logic.
4872 fio = avl_first(&vq->vq_active_tree);
4873 delta = gethrtime() - fio->io_timestamp;
4874 if (delta > spa_deadman_synctime(spa))
4875 zio_deadman(fio, tag);
4877 mutex_exit(&vq->vq_lock);
4882 vdev_defer_resilver(vdev_t *vd)
4884 ASSERT(vd->vdev_ops->vdev_op_leaf);
4886 vd->vdev_resilver_deferred = B_TRUE;
4887 vd->vdev_spa->spa_resilver_deferred = B_TRUE;
4891 * Clears the resilver deferred flag on all leaf devs under vd. Returns
4892 * B_TRUE if we have devices that need to be resilvered and are available to
4893 * accept resilver I/Os.
4896 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx)
4898 boolean_t resilver_needed = B_FALSE;
4899 spa_t *spa = vd->vdev_spa;
4901 for (int c = 0; c < vd->vdev_children; c++) {
4902 vdev_t *cvd = vd->vdev_child[c];
4903 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx);
4906 if (vd == spa->spa_root_vdev &&
4907 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) {
4908 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx);
4909 vdev_config_dirty(vd);
4910 spa->spa_resilver_deferred = B_FALSE;
4911 return (resilver_needed);
4914 if (!vdev_is_concrete(vd) || vd->vdev_aux ||
4915 !vd->vdev_ops->vdev_op_leaf)
4916 return (resilver_needed);
4918 vd->vdev_resilver_deferred = B_FALSE;
4920 return (!vdev_is_dead(vd) && !vd->vdev_offline &&
4921 vdev_resilver_needed(vd, NULL, NULL));
4925 * Translate a logical range to the physical range for the specified vdev_t.
4926 * This function is initially called with a leaf vdev and will walk each
4927 * parent vdev until it reaches a top-level vdev. Once the top-level is
4928 * reached the physical range is initialized and the recursive function
4929 * begins to unwind. As it unwinds it calls the parent's vdev specific
4930 * translation function to do the real conversion.
4933 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs,
4934 range_seg64_t *physical_rs)
4937 * Walk up the vdev tree
4939 if (vd != vd->vdev_top) {
4940 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs);
4943 * We've reached the top-level vdev, initialize the
4944 * physical range to the logical range and start to
4947 physical_rs->rs_start = logical_rs->rs_start;
4948 physical_rs->rs_end = logical_rs->rs_end;
4952 vdev_t *pvd = vd->vdev_parent;
4953 ASSERT3P(pvd, !=, NULL);
4954 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL);
4957 * As this recursive function unwinds, translate the logical
4958 * range into its physical components by calling the
4959 * vdev specific translate function.
4961 range_seg64_t intermediate = { 0 };
4962 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate);
4964 physical_rs->rs_start = intermediate.rs_start;
4965 physical_rs->rs_end = intermediate.rs_end;
4969 * Look at the vdev tree and determine whether any devices are currently being
4973 vdev_replace_in_progress(vdev_t *vdev)
4975 ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0);
4977 if (vdev->vdev_ops == &vdev_replacing_ops)
4981 * A 'spare' vdev indicates that we have a replace in progress, unless
4982 * it has exactly two children, and the second, the hot spare, has
4983 * finished being resilvered.
4985 if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 ||
4986 !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING)))
4989 for (int i = 0; i < vdev->vdev_children; i++) {
4990 if (vdev_replace_in_progress(vdev->vdev_child[i]))
4997 EXPORT_SYMBOL(vdev_fault);
4998 EXPORT_SYMBOL(vdev_degrade);
4999 EXPORT_SYMBOL(vdev_online);
5000 EXPORT_SYMBOL(vdev_offline);
5001 EXPORT_SYMBOL(vdev_clear);
5004 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, INT, ZMOD_RW,
5005 "Target number of metaslabs per top-level vdev");
5007 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, INT, ZMOD_RW,
5008 "Default limit for metaslab size");
5010 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, INT, ZMOD_RW,
5011 "Minimum number of metaslabs per top-level vdev");
5013 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, INT, ZMOD_RW,
5014 "Practical upper limit of total metaslabs per top-level vdev");
5016 ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW,
5017 "Rate limit slow IO (delay) events to this many per second");
5019 ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW,
5020 "Rate limit checksum events to this many checksum errors per second "
5021 "(do not set below zed threshold).");
5023 ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW,
5024 "Ignore errors during resilver/scrub");
5026 ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW,
5027 "Bypass vdev_validate()");
5029 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW,
5030 "Disable cache flushes");