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, 2018 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.
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/bpobj.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/uberblock_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/metaslab_impl.h>
44 #include <sys/space_map.h>
45 #include <sys/space_reftree.h>
48 #include <sys/fs/zfs.h>
51 #include <sys/dsl_scan.h>
54 #include <sys/zfs_ratelimit.h>
56 /* target number of metaslabs per top-level vdev */
57 int vdev_max_ms_count = 200;
59 /* minimum number of metaslabs per top-level vdev */
60 int vdev_min_ms_count = 16;
62 /* practical upper limit of total metaslabs per top-level vdev */
63 int vdev_ms_count_limit = 1ULL << 17;
65 /* lower limit for metaslab size (512M) */
66 int vdev_default_ms_shift = 29;
68 /* upper limit for metaslab size (256G) */
69 int vdev_max_ms_shift = 38;
71 int vdev_validate_skip = B_FALSE;
74 * Since the DTL space map of a vdev is not expected to have a lot of
75 * entries, we default its block size to 4K.
77 int vdev_dtl_sm_blksz = (1 << 12);
80 * Rate limit slow IO (delay) events to this many per second.
82 unsigned int zfs_slow_io_events_per_second = 20;
85 * Rate limit checksum events after this many checksum errors per second.
87 unsigned int zfs_checksum_events_per_second = 20;
90 * Ignore errors during scrub/resilver. Allows to work around resilver
91 * upon import when there are pool errors.
93 int zfs_scan_ignore_errors = 0;
96 * vdev-wide space maps that have lots of entries written to them at
97 * the end of each transaction can benefit from a higher I/O bandwidth
98 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
100 int vdev_standard_sm_blksz = (1 << 17);
103 * Tunable parameter for debugging or performance analysis. Setting this
104 * will cause pool corruption on power loss if a volatile out-of-order
105 * write cache is enabled.
107 int zfs_nocacheflush = 0;
111 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
117 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
120 if (vd->vdev_path != NULL) {
121 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
124 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
125 vd->vdev_ops->vdev_op_type,
126 (u_longlong_t)vd->vdev_id,
127 (u_longlong_t)vd->vdev_guid, buf);
132 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
136 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
137 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
138 vd->vdev_ops->vdev_op_type);
142 switch (vd->vdev_state) {
143 case VDEV_STATE_UNKNOWN:
144 (void) snprintf(state, sizeof (state), "unknown");
146 case VDEV_STATE_CLOSED:
147 (void) snprintf(state, sizeof (state), "closed");
149 case VDEV_STATE_OFFLINE:
150 (void) snprintf(state, sizeof (state), "offline");
152 case VDEV_STATE_REMOVED:
153 (void) snprintf(state, sizeof (state), "removed");
155 case VDEV_STATE_CANT_OPEN:
156 (void) snprintf(state, sizeof (state), "can't open");
158 case VDEV_STATE_FAULTED:
159 (void) snprintf(state, sizeof (state), "faulted");
161 case VDEV_STATE_DEGRADED:
162 (void) snprintf(state, sizeof (state), "degraded");
164 case VDEV_STATE_HEALTHY:
165 (void) snprintf(state, sizeof (state), "healthy");
168 (void) snprintf(state, sizeof (state), "<state %u>",
169 (uint_t)vd->vdev_state);
172 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
173 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
174 vd->vdev_islog ? " (log)" : "",
175 (u_longlong_t)vd->vdev_guid,
176 vd->vdev_path ? vd->vdev_path : "N/A", state);
178 for (uint64_t i = 0; i < vd->vdev_children; i++)
179 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
183 * Virtual device management.
186 static vdev_ops_t *vdev_ops_table[] = {
201 * Given a vdev type, return the appropriate ops vector.
204 vdev_getops(const char *type)
206 vdev_ops_t *ops, **opspp;
208 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
209 if (strcmp(ops->vdev_op_type, type) == 0)
216 * Derive the enumerated alloction bias from string input.
217 * String origin is either the per-vdev zap or zpool(1M).
219 static vdev_alloc_bias_t
220 vdev_derive_alloc_bias(const char *bias)
222 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
224 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
225 alloc_bias = VDEV_BIAS_LOG;
226 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
227 alloc_bias = VDEV_BIAS_SPECIAL;
228 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
229 alloc_bias = VDEV_BIAS_DEDUP;
235 * Default asize function: return the MAX of psize with the asize of
236 * all children. This is what's used by anything other than RAID-Z.
239 vdev_default_asize(vdev_t *vd, uint64_t psize)
241 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
244 for (int c = 0; c < vd->vdev_children; c++) {
245 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
246 asize = MAX(asize, csize);
253 * Get the minimum allocatable size. We define the allocatable size as
254 * the vdev's asize rounded to the nearest metaslab. This allows us to
255 * replace or attach devices which don't have the same physical size but
256 * can still satisfy the same number of allocations.
259 vdev_get_min_asize(vdev_t *vd)
261 vdev_t *pvd = vd->vdev_parent;
264 * If our parent is NULL (inactive spare or cache) or is the root,
265 * just return our own asize.
268 return (vd->vdev_asize);
271 * The top-level vdev just returns the allocatable size rounded
272 * to the nearest metaslab.
274 if (vd == vd->vdev_top)
275 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
278 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
279 * so each child must provide at least 1/Nth of its asize.
281 if (pvd->vdev_ops == &vdev_raidz_ops)
282 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
285 return (pvd->vdev_min_asize);
289 vdev_set_min_asize(vdev_t *vd)
291 vd->vdev_min_asize = vdev_get_min_asize(vd);
293 for (int c = 0; c < vd->vdev_children; c++)
294 vdev_set_min_asize(vd->vdev_child[c]);
298 vdev_lookup_top(spa_t *spa, uint64_t vdev)
300 vdev_t *rvd = spa->spa_root_vdev;
302 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
304 if (vdev < rvd->vdev_children) {
305 ASSERT(rvd->vdev_child[vdev] != NULL);
306 return (rvd->vdev_child[vdev]);
313 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
317 if (vd->vdev_guid == guid)
320 for (int c = 0; c < vd->vdev_children; c++)
321 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
329 vdev_count_leaves_impl(vdev_t *vd)
333 if (vd->vdev_ops->vdev_op_leaf)
336 for (int c = 0; c < vd->vdev_children; c++)
337 n += vdev_count_leaves_impl(vd->vdev_child[c]);
343 vdev_count_leaves(spa_t *spa)
347 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
348 rc = vdev_count_leaves_impl(spa->spa_root_vdev);
349 spa_config_exit(spa, SCL_VDEV, FTAG);
355 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
357 size_t oldsize, newsize;
358 uint64_t id = cvd->vdev_id;
361 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
362 ASSERT(cvd->vdev_parent == NULL);
364 cvd->vdev_parent = pvd;
369 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
371 oldsize = pvd->vdev_children * sizeof (vdev_t *);
372 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
373 newsize = pvd->vdev_children * sizeof (vdev_t *);
375 newchild = kmem_alloc(newsize, KM_SLEEP);
376 if (pvd->vdev_child != NULL) {
377 bcopy(pvd->vdev_child, newchild, oldsize);
378 kmem_free(pvd->vdev_child, oldsize);
381 pvd->vdev_child = newchild;
382 pvd->vdev_child[id] = cvd;
384 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
385 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
388 * Walk up all ancestors to update guid sum.
390 for (; pvd != NULL; pvd = pvd->vdev_parent)
391 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
395 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
398 uint_t id = cvd->vdev_id;
400 ASSERT(cvd->vdev_parent == pvd);
405 ASSERT(id < pvd->vdev_children);
406 ASSERT(pvd->vdev_child[id] == cvd);
408 pvd->vdev_child[id] = NULL;
409 cvd->vdev_parent = NULL;
411 for (c = 0; c < pvd->vdev_children; c++)
412 if (pvd->vdev_child[c])
415 if (c == pvd->vdev_children) {
416 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
417 pvd->vdev_child = NULL;
418 pvd->vdev_children = 0;
422 * Walk up all ancestors to update guid sum.
424 for (; pvd != NULL; pvd = pvd->vdev_parent)
425 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
429 * Remove any holes in the child array.
432 vdev_compact_children(vdev_t *pvd)
434 vdev_t **newchild, *cvd;
435 int oldc = pvd->vdev_children;
438 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
443 for (int c = newc = 0; c < oldc; c++)
444 if (pvd->vdev_child[c])
448 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP);
450 for (int c = newc = 0; c < oldc; c++) {
451 if ((cvd = pvd->vdev_child[c]) != NULL) {
452 newchild[newc] = cvd;
453 cvd->vdev_id = newc++;
460 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
461 pvd->vdev_child = newchild;
462 pvd->vdev_children = newc;
466 * Allocate and minimally initialize a vdev_t.
469 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
472 vdev_indirect_config_t *vic;
474 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
475 vic = &vd->vdev_indirect_config;
477 if (spa->spa_root_vdev == NULL) {
478 ASSERT(ops == &vdev_root_ops);
479 spa->spa_root_vdev = vd;
480 spa->spa_load_guid = spa_generate_guid(NULL);
483 if (guid == 0 && ops != &vdev_hole_ops) {
484 if (spa->spa_root_vdev == vd) {
486 * The root vdev's guid will also be the pool guid,
487 * which must be unique among all pools.
489 guid = spa_generate_guid(NULL);
492 * Any other vdev's guid must be unique within the pool.
494 guid = spa_generate_guid(spa);
496 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
501 vd->vdev_guid = guid;
502 vd->vdev_guid_sum = guid;
504 vd->vdev_state = VDEV_STATE_CLOSED;
505 vd->vdev_ishole = (ops == &vdev_hole_ops);
506 vic->vic_prev_indirect_vdev = UINT64_MAX;
508 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
509 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
510 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
513 * Initialize rate limit structs for events. We rate limit ZIO delay
514 * and checksum events so that we don't overwhelm ZED with thousands
515 * of events when a disk is acting up.
517 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second,
519 zfs_ratelimit_init(&vd->vdev_checksum_rl,
520 &zfs_checksum_events_per_second, 1);
522 list_link_init(&vd->vdev_config_dirty_node);
523 list_link_init(&vd->vdev_state_dirty_node);
524 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL);
525 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
526 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
527 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
528 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
530 for (int t = 0; t < DTL_TYPES; t++) {
531 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
533 txg_list_create(&vd->vdev_ms_list, spa,
534 offsetof(struct metaslab, ms_txg_node));
535 txg_list_create(&vd->vdev_dtl_list, spa,
536 offsetof(struct vdev, vdev_dtl_node));
537 vd->vdev_stat.vs_timestamp = gethrtime();
545 * Allocate a new vdev. The 'alloctype' is used to control whether we are
546 * creating a new vdev or loading an existing one - the behavior is slightly
547 * different for each case.
550 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
555 uint64_t guid = 0, islog, nparity;
557 vdev_indirect_config_t *vic;
560 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
561 boolean_t top_level = (parent && !parent->vdev_parent);
563 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
565 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
566 return (SET_ERROR(EINVAL));
568 if ((ops = vdev_getops(type)) == NULL)
569 return (SET_ERROR(EINVAL));
572 * If this is a load, get the vdev guid from the nvlist.
573 * Otherwise, vdev_alloc_common() will generate one for us.
575 if (alloctype == VDEV_ALLOC_LOAD) {
578 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
580 return (SET_ERROR(EINVAL));
582 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
583 return (SET_ERROR(EINVAL));
584 } else if (alloctype == VDEV_ALLOC_SPARE) {
585 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
586 return (SET_ERROR(EINVAL));
587 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
588 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
589 return (SET_ERROR(EINVAL));
590 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
591 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
592 return (SET_ERROR(EINVAL));
596 * The first allocated vdev must be of type 'root'.
598 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
599 return (SET_ERROR(EINVAL));
602 * Determine whether we're a log vdev.
605 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
606 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
607 return (SET_ERROR(ENOTSUP));
609 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
610 return (SET_ERROR(ENOTSUP));
613 * Set the nparity property for RAID-Z vdevs.
616 if (ops == &vdev_raidz_ops) {
617 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
619 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
620 return (SET_ERROR(EINVAL));
622 * Previous versions could only support 1 or 2 parity
626 spa_version(spa) < SPA_VERSION_RAIDZ2)
627 return (SET_ERROR(ENOTSUP));
629 spa_version(spa) < SPA_VERSION_RAIDZ3)
630 return (SET_ERROR(ENOTSUP));
633 * We require the parity to be specified for SPAs that
634 * support multiple parity levels.
636 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
637 return (SET_ERROR(EINVAL));
639 * Otherwise, we default to 1 parity device for RAID-Z.
646 ASSERT(nparity != -1ULL);
649 * If creating a top-level vdev, check for allocation classes input
651 if (top_level && alloctype == VDEV_ALLOC_ADD) {
654 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
656 alloc_bias = vdev_derive_alloc_bias(bias);
658 /* spa_vdev_add() expects feature to be enabled */
659 if (spa->spa_load_state != SPA_LOAD_CREATE &&
660 !spa_feature_is_enabled(spa,
661 SPA_FEATURE_ALLOCATION_CLASSES)) {
662 return (SET_ERROR(ENOTSUP));
667 vd = vdev_alloc_common(spa, id, guid, ops);
668 vic = &vd->vdev_indirect_config;
670 vd->vdev_islog = islog;
671 vd->vdev_nparity = nparity;
672 if (top_level && alloc_bias != VDEV_BIAS_NONE)
673 vd->vdev_alloc_bias = alloc_bias;
675 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
676 vd->vdev_path = spa_strdup(vd->vdev_path);
679 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
680 * fault on a vdev and want it to persist across imports (like with
683 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp);
684 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) {
685 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
686 vd->vdev_faulted = 1;
687 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
690 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
691 vd->vdev_devid = spa_strdup(vd->vdev_devid);
692 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
693 &vd->vdev_physpath) == 0)
694 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
696 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
697 &vd->vdev_enc_sysfs_path) == 0)
698 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path);
700 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
701 vd->vdev_fru = spa_strdup(vd->vdev_fru);
704 * Set the whole_disk property. If it's not specified, leave the value
707 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
708 &vd->vdev_wholedisk) != 0)
709 vd->vdev_wholedisk = -1ULL;
711 ASSERT0(vic->vic_mapping_object);
712 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
713 &vic->vic_mapping_object);
714 ASSERT0(vic->vic_births_object);
715 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
716 &vic->vic_births_object);
717 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
718 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
719 &vic->vic_prev_indirect_vdev);
722 * Look for the 'not present' flag. This will only be set if the device
723 * was not present at the time of import.
725 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
726 &vd->vdev_not_present);
729 * Get the alignment requirement.
731 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
734 * Retrieve the vdev creation time.
736 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
740 * If we're a top-level vdev, try to load the allocation parameters.
743 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
744 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
746 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
748 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
750 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
752 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
755 ASSERT0(vd->vdev_top_zap);
758 if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
759 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
760 alloctype == VDEV_ALLOC_ADD ||
761 alloctype == VDEV_ALLOC_SPLIT ||
762 alloctype == VDEV_ALLOC_ROOTPOOL);
763 /* Note: metaslab_group_create() is now deferred */
766 if (vd->vdev_ops->vdev_op_leaf &&
767 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
768 (void) nvlist_lookup_uint64(nv,
769 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
771 ASSERT0(vd->vdev_leaf_zap);
775 * If we're a leaf vdev, try to load the DTL object and other state.
778 if (vd->vdev_ops->vdev_op_leaf &&
779 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
780 alloctype == VDEV_ALLOC_ROOTPOOL)) {
781 if (alloctype == VDEV_ALLOC_LOAD) {
782 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
783 &vd->vdev_dtl_object);
784 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
788 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
791 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
792 &spare) == 0 && spare)
796 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
799 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
800 &vd->vdev_resilver_txg);
802 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER))
803 vdev_set_deferred_resilver(spa, vd);
806 * In general, when importing a pool we want to ignore the
807 * persistent fault state, as the diagnosis made on another
808 * system may not be valid in the current context. The only
809 * exception is if we forced a vdev to a persistently faulted
810 * state with 'zpool offline -f'. The persistent fault will
811 * remain across imports until cleared.
813 * Local vdevs will remain in the faulted state.
815 if (spa_load_state(spa) == SPA_LOAD_OPEN ||
816 spa_load_state(spa) == SPA_LOAD_IMPORT) {
817 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
819 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
821 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
824 if (vd->vdev_faulted || vd->vdev_degraded) {
828 VDEV_AUX_ERR_EXCEEDED;
829 if (nvlist_lookup_string(nv,
830 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
831 strcmp(aux, "external") == 0)
832 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
834 vd->vdev_faulted = 0ULL;
840 * Add ourselves to the parent's list of children.
842 vdev_add_child(parent, vd);
850 vdev_free(vdev_t *vd)
852 spa_t *spa = vd->vdev_spa;
855 * Scan queues are normally destroyed at the end of a scan. If the
856 * queue exists here, that implies the vdev is being removed while
857 * the scan is still running.
859 if (vd->vdev_scan_io_queue != NULL) {
860 mutex_enter(&vd->vdev_scan_io_queue_lock);
861 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
862 vd->vdev_scan_io_queue = NULL;
863 mutex_exit(&vd->vdev_scan_io_queue_lock);
867 * vdev_free() implies closing the vdev first. This is simpler than
868 * trying to ensure complicated semantics for all callers.
872 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
873 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
878 for (int c = 0; c < vd->vdev_children; c++)
879 vdev_free(vd->vdev_child[c]);
881 ASSERT(vd->vdev_child == NULL);
882 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
885 * Discard allocation state.
887 if (vd->vdev_mg != NULL) {
888 vdev_metaslab_fini(vd);
889 metaslab_group_destroy(vd->vdev_mg);
892 ASSERT0(vd->vdev_stat.vs_space);
893 ASSERT0(vd->vdev_stat.vs_dspace);
894 ASSERT0(vd->vdev_stat.vs_alloc);
897 * Remove this vdev from its parent's child list.
899 vdev_remove_child(vd->vdev_parent, vd);
901 ASSERT(vd->vdev_parent == NULL);
904 * Clean up vdev structure.
910 spa_strfree(vd->vdev_path);
912 spa_strfree(vd->vdev_devid);
913 if (vd->vdev_physpath)
914 spa_strfree(vd->vdev_physpath);
916 if (vd->vdev_enc_sysfs_path)
917 spa_strfree(vd->vdev_enc_sysfs_path);
920 spa_strfree(vd->vdev_fru);
922 if (vd->vdev_isspare)
923 spa_spare_remove(vd);
924 if (vd->vdev_isl2cache)
925 spa_l2cache_remove(vd);
927 txg_list_destroy(&vd->vdev_ms_list);
928 txg_list_destroy(&vd->vdev_dtl_list);
930 mutex_enter(&vd->vdev_dtl_lock);
931 space_map_close(vd->vdev_dtl_sm);
932 for (int t = 0; t < DTL_TYPES; t++) {
933 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
934 range_tree_destroy(vd->vdev_dtl[t]);
936 mutex_exit(&vd->vdev_dtl_lock);
938 EQUIV(vd->vdev_indirect_births != NULL,
939 vd->vdev_indirect_mapping != NULL);
940 if (vd->vdev_indirect_births != NULL) {
941 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
942 vdev_indirect_births_close(vd->vdev_indirect_births);
945 if (vd->vdev_obsolete_sm != NULL) {
946 ASSERT(vd->vdev_removing ||
947 vd->vdev_ops == &vdev_indirect_ops);
948 space_map_close(vd->vdev_obsolete_sm);
949 vd->vdev_obsolete_sm = NULL;
951 range_tree_destroy(vd->vdev_obsolete_segments);
952 rw_destroy(&vd->vdev_indirect_rwlock);
953 mutex_destroy(&vd->vdev_obsolete_lock);
955 mutex_destroy(&vd->vdev_queue_lock);
956 mutex_destroy(&vd->vdev_dtl_lock);
957 mutex_destroy(&vd->vdev_stat_lock);
958 mutex_destroy(&vd->vdev_probe_lock);
959 mutex_destroy(&vd->vdev_scan_io_queue_lock);
961 zfs_ratelimit_fini(&vd->vdev_delay_rl);
962 zfs_ratelimit_fini(&vd->vdev_checksum_rl);
964 if (vd == spa->spa_root_vdev)
965 spa->spa_root_vdev = NULL;
967 kmem_free(vd, sizeof (vdev_t));
971 * Transfer top-level vdev state from svd to tvd.
974 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
976 spa_t *spa = svd->vdev_spa;
981 ASSERT(tvd == tvd->vdev_top);
983 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite;
984 tvd->vdev_ms_array = svd->vdev_ms_array;
985 tvd->vdev_ms_shift = svd->vdev_ms_shift;
986 tvd->vdev_ms_count = svd->vdev_ms_count;
987 tvd->vdev_top_zap = svd->vdev_top_zap;
989 svd->vdev_ms_array = 0;
990 svd->vdev_ms_shift = 0;
991 svd->vdev_ms_count = 0;
992 svd->vdev_top_zap = 0;
995 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
996 tvd->vdev_mg = svd->vdev_mg;
997 tvd->vdev_ms = svd->vdev_ms;
1000 svd->vdev_ms = NULL;
1002 if (tvd->vdev_mg != NULL)
1003 tvd->vdev_mg->mg_vd = tvd;
1005 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1006 svd->vdev_checkpoint_sm = NULL;
1008 tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
1009 svd->vdev_alloc_bias = VDEV_BIAS_NONE;
1011 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1012 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1013 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1015 svd->vdev_stat.vs_alloc = 0;
1016 svd->vdev_stat.vs_space = 0;
1017 svd->vdev_stat.vs_dspace = 0;
1020 * State which may be set on a top-level vdev that's in the
1021 * process of being removed.
1023 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1024 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1025 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1026 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1027 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1028 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1029 ASSERT0(tvd->vdev_removing);
1030 tvd->vdev_removing = svd->vdev_removing;
1031 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1032 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1033 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1034 range_tree_swap(&svd->vdev_obsolete_segments,
1035 &tvd->vdev_obsolete_segments);
1036 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1037 svd->vdev_indirect_config.vic_mapping_object = 0;
1038 svd->vdev_indirect_config.vic_births_object = 0;
1039 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1040 svd->vdev_indirect_mapping = NULL;
1041 svd->vdev_indirect_births = NULL;
1042 svd->vdev_obsolete_sm = NULL;
1043 svd->vdev_removing = 0;
1045 for (t = 0; t < TXG_SIZE; t++) {
1046 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1047 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1048 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1049 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1050 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1051 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1054 if (list_link_active(&svd->vdev_config_dirty_node)) {
1055 vdev_config_clean(svd);
1056 vdev_config_dirty(tvd);
1059 if (list_link_active(&svd->vdev_state_dirty_node)) {
1060 vdev_state_clean(svd);
1061 vdev_state_dirty(tvd);
1064 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1065 svd->vdev_deflate_ratio = 0;
1067 tvd->vdev_islog = svd->vdev_islog;
1068 svd->vdev_islog = 0;
1070 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1074 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1081 for (int c = 0; c < vd->vdev_children; c++)
1082 vdev_top_update(tvd, vd->vdev_child[c]);
1086 * Add a mirror/replacing vdev above an existing vdev.
1089 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1091 spa_t *spa = cvd->vdev_spa;
1092 vdev_t *pvd = cvd->vdev_parent;
1095 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1097 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1099 mvd->vdev_asize = cvd->vdev_asize;
1100 mvd->vdev_min_asize = cvd->vdev_min_asize;
1101 mvd->vdev_max_asize = cvd->vdev_max_asize;
1102 mvd->vdev_psize = cvd->vdev_psize;
1103 mvd->vdev_ashift = cvd->vdev_ashift;
1104 mvd->vdev_state = cvd->vdev_state;
1105 mvd->vdev_crtxg = cvd->vdev_crtxg;
1107 vdev_remove_child(pvd, cvd);
1108 vdev_add_child(pvd, mvd);
1109 cvd->vdev_id = mvd->vdev_children;
1110 vdev_add_child(mvd, cvd);
1111 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1113 if (mvd == mvd->vdev_top)
1114 vdev_top_transfer(cvd, mvd);
1120 * Remove a 1-way mirror/replacing vdev from the tree.
1123 vdev_remove_parent(vdev_t *cvd)
1125 vdev_t *mvd = cvd->vdev_parent;
1126 vdev_t *pvd = mvd->vdev_parent;
1128 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1130 ASSERT(mvd->vdev_children == 1);
1131 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1132 mvd->vdev_ops == &vdev_replacing_ops ||
1133 mvd->vdev_ops == &vdev_spare_ops);
1134 cvd->vdev_ashift = mvd->vdev_ashift;
1136 vdev_remove_child(mvd, cvd);
1137 vdev_remove_child(pvd, mvd);
1140 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1141 * Otherwise, we could have detached an offline device, and when we
1142 * go to import the pool we'll think we have two top-level vdevs,
1143 * instead of a different version of the same top-level vdev.
1145 if (mvd->vdev_top == mvd) {
1146 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1147 cvd->vdev_orig_guid = cvd->vdev_guid;
1148 cvd->vdev_guid += guid_delta;
1149 cvd->vdev_guid_sum += guid_delta;
1152 * If pool not set for autoexpand, we need to also preserve
1153 * mvd's asize to prevent automatic expansion of cvd.
1154 * Otherwise if we are adjusting the mirror by attaching and
1155 * detaching children of non-uniform sizes, the mirror could
1156 * autoexpand, unexpectedly requiring larger devices to
1157 * re-establish the mirror.
1159 if (!cvd->vdev_spa->spa_autoexpand)
1160 cvd->vdev_asize = mvd->vdev_asize;
1162 cvd->vdev_id = mvd->vdev_id;
1163 vdev_add_child(pvd, cvd);
1164 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1166 if (cvd == cvd->vdev_top)
1167 vdev_top_transfer(mvd, cvd);
1169 ASSERT(mvd->vdev_children == 0);
1174 vdev_metaslab_group_create(vdev_t *vd)
1176 spa_t *spa = vd->vdev_spa;
1179 * metaslab_group_create was delayed until allocation bias was available
1181 if (vd->vdev_mg == NULL) {
1182 metaslab_class_t *mc;
1184 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
1185 vd->vdev_alloc_bias = VDEV_BIAS_LOG;
1187 ASSERT3U(vd->vdev_islog, ==,
1188 (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
1190 switch (vd->vdev_alloc_bias) {
1192 mc = spa_log_class(spa);
1194 case VDEV_BIAS_SPECIAL:
1195 mc = spa_special_class(spa);
1197 case VDEV_BIAS_DEDUP:
1198 mc = spa_dedup_class(spa);
1201 mc = spa_normal_class(spa);
1204 vd->vdev_mg = metaslab_group_create(mc, vd,
1205 spa->spa_alloc_count);
1208 * The spa ashift values currently only reflect the
1209 * general vdev classes. Class destination is late
1210 * binding so ashift checking had to wait until now
1212 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1213 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
1214 if (vd->vdev_ashift > spa->spa_max_ashift)
1215 spa->spa_max_ashift = vd->vdev_ashift;
1216 if (vd->vdev_ashift < spa->spa_min_ashift)
1217 spa->spa_min_ashift = vd->vdev_ashift;
1223 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1225 spa_t *spa = vd->vdev_spa;
1226 objset_t *mos = spa->spa_meta_objset;
1228 uint64_t oldc = vd->vdev_ms_count;
1229 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1232 boolean_t expanding = (oldc != 0);
1234 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1237 * This vdev is not being allocated from yet or is a hole.
1239 if (vd->vdev_ms_shift == 0)
1242 ASSERT(!vd->vdev_ishole);
1244 ASSERT(oldc <= newc);
1246 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1249 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1250 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1254 vd->vdev_ms_count = newc;
1255 for (m = oldc; m < newc; m++) {
1256 uint64_t object = 0;
1259 * vdev_ms_array may be 0 if we are creating the "fake"
1260 * metaslabs for an indirect vdev for zdb's leak detection.
1261 * See zdb_leak_init().
1263 if (txg == 0 && vd->vdev_ms_array != 0) {
1264 error = dmu_read(mos, vd->vdev_ms_array,
1265 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1268 vdev_dbgmsg(vd, "unable to read the metaslab "
1269 "array [error=%d]", error);
1276 * To accomodate zdb_leak_init() fake indirect
1277 * metaslabs, we allocate a metaslab group for
1278 * indirect vdevs which normally don't have one.
1280 if (vd->vdev_mg == NULL) {
1281 ASSERT0(vdev_is_concrete(vd));
1282 vdev_metaslab_group_create(vd);
1285 error = metaslab_init(vd->vdev_mg, m, object, txg,
1288 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1295 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1298 * If the vdev is being removed we don't activate
1299 * the metaslabs since we want to ensure that no new
1300 * allocations are performed on this device.
1302 if (!expanding && !vd->vdev_removing) {
1303 metaslab_group_activate(vd->vdev_mg);
1307 spa_config_exit(spa, SCL_ALLOC, FTAG);
1313 vdev_metaslab_fini(vdev_t *vd)
1315 if (vd->vdev_checkpoint_sm != NULL) {
1316 ASSERT(spa_feature_is_active(vd->vdev_spa,
1317 SPA_FEATURE_POOL_CHECKPOINT));
1318 space_map_close(vd->vdev_checkpoint_sm);
1320 * Even though we close the space map, we need to set its
1321 * pointer to NULL. The reason is that vdev_metaslab_fini()
1322 * may be called multiple times for certain operations
1323 * (i.e. when destroying a pool) so we need to ensure that
1324 * this clause never executes twice. This logic is similar
1325 * to the one used for the vdev_ms clause below.
1327 vd->vdev_checkpoint_sm = NULL;
1330 if (vd->vdev_ms != NULL) {
1331 uint64_t count = vd->vdev_ms_count;
1333 metaslab_group_passivate(vd->vdev_mg);
1334 for (uint64_t m = 0; m < count; m++) {
1335 metaslab_t *msp = vd->vdev_ms[m];
1340 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1343 vd->vdev_ms_count = 0;
1345 ASSERT0(vd->vdev_ms_count);
1346 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
1349 typedef struct vdev_probe_stats {
1350 boolean_t vps_readable;
1351 boolean_t vps_writeable;
1353 } vdev_probe_stats_t;
1356 vdev_probe_done(zio_t *zio)
1358 spa_t *spa = zio->io_spa;
1359 vdev_t *vd = zio->io_vd;
1360 vdev_probe_stats_t *vps = zio->io_private;
1362 ASSERT(vd->vdev_probe_zio != NULL);
1364 if (zio->io_type == ZIO_TYPE_READ) {
1365 if (zio->io_error == 0)
1366 vps->vps_readable = 1;
1367 if (zio->io_error == 0 && spa_writeable(spa)) {
1368 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1369 zio->io_offset, zio->io_size, zio->io_abd,
1370 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1371 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1373 abd_free(zio->io_abd);
1375 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1376 if (zio->io_error == 0)
1377 vps->vps_writeable = 1;
1378 abd_free(zio->io_abd);
1379 } else if (zio->io_type == ZIO_TYPE_NULL) {
1383 vd->vdev_cant_read |= !vps->vps_readable;
1384 vd->vdev_cant_write |= !vps->vps_writeable;
1386 if (vdev_readable(vd) &&
1387 (vdev_writeable(vd) || !spa_writeable(spa))) {
1390 ASSERT(zio->io_error != 0);
1391 vdev_dbgmsg(vd, "failed probe");
1392 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1393 spa, vd, NULL, NULL, 0, 0);
1394 zio->io_error = SET_ERROR(ENXIO);
1397 mutex_enter(&vd->vdev_probe_lock);
1398 ASSERT(vd->vdev_probe_zio == zio);
1399 vd->vdev_probe_zio = NULL;
1400 mutex_exit(&vd->vdev_probe_lock);
1403 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1404 if (!vdev_accessible(vd, pio))
1405 pio->io_error = SET_ERROR(ENXIO);
1407 kmem_free(vps, sizeof (*vps));
1412 * Determine whether this device is accessible.
1414 * Read and write to several known locations: the pad regions of each
1415 * vdev label but the first, which we leave alone in case it contains
1419 vdev_probe(vdev_t *vd, zio_t *zio)
1421 spa_t *spa = vd->vdev_spa;
1422 vdev_probe_stats_t *vps = NULL;
1425 ASSERT(vd->vdev_ops->vdev_op_leaf);
1428 * Don't probe the probe.
1430 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1434 * To prevent 'probe storms' when a device fails, we create
1435 * just one probe i/o at a time. All zios that want to probe
1436 * this vdev will become parents of the probe io.
1438 mutex_enter(&vd->vdev_probe_lock);
1440 if ((pio = vd->vdev_probe_zio) == NULL) {
1441 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1443 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1444 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1447 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1449 * vdev_cant_read and vdev_cant_write can only
1450 * transition from TRUE to FALSE when we have the
1451 * SCL_ZIO lock as writer; otherwise they can only
1452 * transition from FALSE to TRUE. This ensures that
1453 * any zio looking at these values can assume that
1454 * failures persist for the life of the I/O. That's
1455 * important because when a device has intermittent
1456 * connectivity problems, we want to ensure that
1457 * they're ascribed to the device (ENXIO) and not
1460 * Since we hold SCL_ZIO as writer here, clear both
1461 * values so the probe can reevaluate from first
1464 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1465 vd->vdev_cant_read = B_FALSE;
1466 vd->vdev_cant_write = B_FALSE;
1469 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1470 vdev_probe_done, vps,
1471 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1474 * We can't change the vdev state in this context, so we
1475 * kick off an async task to do it on our behalf.
1478 vd->vdev_probe_wanted = B_TRUE;
1479 spa_async_request(spa, SPA_ASYNC_PROBE);
1484 zio_add_child(zio, pio);
1486 mutex_exit(&vd->vdev_probe_lock);
1489 ASSERT(zio != NULL);
1493 for (int l = 1; l < VDEV_LABELS; l++) {
1494 zio_nowait(zio_read_phys(pio, vd,
1495 vdev_label_offset(vd->vdev_psize, l,
1496 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1497 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1498 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1499 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1510 vdev_open_child(void *arg)
1514 vd->vdev_open_thread = curthread;
1515 vd->vdev_open_error = vdev_open(vd);
1516 vd->vdev_open_thread = NULL;
1520 vdev_uses_zvols(vdev_t *vd)
1523 if (zvol_is_zvol(vd->vdev_path))
1527 for (int c = 0; c < vd->vdev_children; c++)
1528 if (vdev_uses_zvols(vd->vdev_child[c]))
1535 vdev_open_children(vdev_t *vd)
1538 int children = vd->vdev_children;
1541 * in order to handle pools on top of zvols, do the opens
1542 * in a single thread so that the same thread holds the
1543 * spa_namespace_lock
1545 if (vdev_uses_zvols(vd)) {
1547 for (int c = 0; c < children; c++)
1548 vd->vdev_child[c]->vdev_open_error =
1549 vdev_open(vd->vdev_child[c]);
1551 tq = taskq_create("vdev_open", children, minclsyspri,
1552 children, children, TASKQ_PREPOPULATE);
1556 for (int c = 0; c < children; c++)
1557 VERIFY(taskq_dispatch(tq, vdev_open_child,
1558 vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID);
1563 vd->vdev_nonrot = B_TRUE;
1565 for (int c = 0; c < children; c++)
1566 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1570 * Compute the raidz-deflation ratio. Note, we hard-code
1571 * in 128k (1 << 17) because it is the "typical" blocksize.
1572 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1573 * otherwise it would inconsistently account for existing bp's.
1576 vdev_set_deflate_ratio(vdev_t *vd)
1578 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1579 vd->vdev_deflate_ratio = (1 << 17) /
1580 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1585 * Prepare a virtual device for access.
1588 vdev_open(vdev_t *vd)
1590 spa_t *spa = vd->vdev_spa;
1593 uint64_t max_osize = 0;
1594 uint64_t asize, max_asize, psize;
1595 uint64_t ashift = 0;
1597 ASSERT(vd->vdev_open_thread == curthread ||
1598 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1599 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1600 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1601 vd->vdev_state == VDEV_STATE_OFFLINE);
1603 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1604 vd->vdev_cant_read = B_FALSE;
1605 vd->vdev_cant_write = B_FALSE;
1606 vd->vdev_min_asize = vdev_get_min_asize(vd);
1609 * If this vdev is not removed, check its fault status. If it's
1610 * faulted, bail out of the open.
1612 if (!vd->vdev_removed && vd->vdev_faulted) {
1613 ASSERT(vd->vdev_children == 0);
1614 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1615 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1616 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1617 vd->vdev_label_aux);
1618 return (SET_ERROR(ENXIO));
1619 } else if (vd->vdev_offline) {
1620 ASSERT(vd->vdev_children == 0);
1621 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1622 return (SET_ERROR(ENXIO));
1625 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1628 * Reset the vdev_reopening flag so that we actually close
1629 * the vdev on error.
1631 vd->vdev_reopening = B_FALSE;
1632 if (zio_injection_enabled && error == 0)
1633 error = zio_handle_device_injection(vd, NULL, ENXIO);
1636 if (vd->vdev_removed &&
1637 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1638 vd->vdev_removed = B_FALSE;
1640 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1641 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1642 vd->vdev_stat.vs_aux);
1644 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1645 vd->vdev_stat.vs_aux);
1650 vd->vdev_removed = B_FALSE;
1653 * Recheck the faulted flag now that we have confirmed that
1654 * the vdev is accessible. If we're faulted, bail.
1656 if (vd->vdev_faulted) {
1657 ASSERT(vd->vdev_children == 0);
1658 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1659 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1660 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1661 vd->vdev_label_aux);
1662 return (SET_ERROR(ENXIO));
1665 if (vd->vdev_degraded) {
1666 ASSERT(vd->vdev_children == 0);
1667 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1668 VDEV_AUX_ERR_EXCEEDED);
1670 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1674 * For hole or missing vdevs we just return success.
1676 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1679 for (int c = 0; c < vd->vdev_children; c++) {
1680 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1681 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1687 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1688 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1690 if (vd->vdev_children == 0) {
1691 if (osize < SPA_MINDEVSIZE) {
1692 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1693 VDEV_AUX_TOO_SMALL);
1694 return (SET_ERROR(EOVERFLOW));
1697 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1698 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1699 VDEV_LABEL_END_SIZE);
1701 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1702 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1703 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1704 VDEV_AUX_TOO_SMALL);
1705 return (SET_ERROR(EOVERFLOW));
1709 max_asize = max_osize;
1713 * If the vdev was expanded, record this so that we can re-create the
1714 * uberblock rings in labels {2,3}, during the next sync.
1716 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0))
1717 vd->vdev_copy_uberblocks = B_TRUE;
1719 vd->vdev_psize = psize;
1722 * Make sure the allocatable size hasn't shrunk too much.
1724 if (asize < vd->vdev_min_asize) {
1725 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1726 VDEV_AUX_BAD_LABEL);
1727 return (SET_ERROR(EINVAL));
1730 if (vd->vdev_asize == 0) {
1732 * This is the first-ever open, so use the computed values.
1733 * For compatibility, a different ashift can be requested.
1735 vd->vdev_asize = asize;
1736 vd->vdev_max_asize = max_asize;
1737 if (vd->vdev_ashift == 0) {
1738 vd->vdev_ashift = ashift; /* use detected value */
1740 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
1741 vd->vdev_ashift > ASHIFT_MAX)) {
1742 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1743 VDEV_AUX_BAD_ASHIFT);
1744 return (SET_ERROR(EDOM));
1748 * Detect if the alignment requirement has increased.
1749 * We don't want to make the pool unavailable, just
1750 * post an event instead.
1752 if (ashift > vd->vdev_top->vdev_ashift &&
1753 vd->vdev_ops->vdev_op_leaf) {
1754 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
1755 spa, vd, NULL, NULL, 0, 0);
1758 vd->vdev_max_asize = max_asize;
1762 * If all children are healthy we update asize if either:
1763 * The asize has increased, due to a device expansion caused by dynamic
1764 * LUN growth or vdev replacement, and automatic expansion is enabled;
1765 * making the additional space available.
1767 * The asize has decreased, due to a device shrink usually caused by a
1768 * vdev replace with a smaller device. This ensures that calculations
1769 * based of max_asize and asize e.g. esize are always valid. It's safe
1770 * to do this as we've already validated that asize is greater than
1773 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1774 ((asize > vd->vdev_asize &&
1775 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1776 (asize < vd->vdev_asize)))
1777 vd->vdev_asize = asize;
1779 vdev_set_min_asize(vd);
1782 * Ensure we can issue some IO before declaring the
1783 * vdev open for business.
1785 if (vd->vdev_ops->vdev_op_leaf &&
1786 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1787 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1788 VDEV_AUX_ERR_EXCEEDED);
1793 * Track the min and max ashift values for normal data devices.
1795 * DJB - TBD these should perhaps be tracked per allocation class
1796 * (e.g. spa_min_ashift is used to round up post compression buffers)
1798 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1799 vd->vdev_alloc_bias == VDEV_BIAS_NONE &&
1800 vd->vdev_aux == NULL) {
1801 if (vd->vdev_ashift > spa->spa_max_ashift)
1802 spa->spa_max_ashift = vd->vdev_ashift;
1803 if (vd->vdev_ashift < spa->spa_min_ashift)
1804 spa->spa_min_ashift = vd->vdev_ashift;
1808 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1809 * resilver. But don't do this if we are doing a reopen for a scrub,
1810 * since this would just restart the scrub we are already doing.
1812 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1813 vdev_resilver_needed(vd, NULL, NULL)) {
1814 if (dsl_scan_resilvering(spa->spa_dsl_pool) &&
1815 spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER))
1816 vdev_set_deferred_resilver(spa, vd);
1818 spa_async_request(spa, SPA_ASYNC_RESILVER);
1825 * Called once the vdevs are all opened, this routine validates the label
1826 * contents. This needs to be done before vdev_load() so that we don't
1827 * inadvertently do repair I/Os to the wrong device.
1829 * This function will only return failure if one of the vdevs indicates that it
1830 * has since been destroyed or exported. This is only possible if
1831 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1832 * will be updated but the function will return 0.
1835 vdev_validate(vdev_t *vd)
1837 spa_t *spa = vd->vdev_spa;
1839 uint64_t guid = 0, aux_guid = 0, top_guid;
1844 if (vdev_validate_skip)
1847 for (uint64_t c = 0; c < vd->vdev_children; c++)
1848 if (vdev_validate(vd->vdev_child[c]) != 0)
1849 return (SET_ERROR(EBADF));
1852 * If the device has already failed, or was marked offline, don't do
1853 * any further validation. Otherwise, label I/O will fail and we will
1854 * overwrite the previous state.
1856 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1860 * If we are performing an extreme rewind, we allow for a label that
1861 * was modified at a point after the current txg.
1862 * If config lock is not held do not check for the txg. spa_sync could
1863 * be updating the vdev's label before updating spa_last_synced_txg.
1865 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1866 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1869 txg = spa_last_synced_txg(spa);
1871 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1872 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1873 VDEV_AUX_BAD_LABEL);
1874 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
1875 "txg %llu", (u_longlong_t)txg);
1880 * Determine if this vdev has been split off into another
1881 * pool. If so, then refuse to open it.
1883 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1884 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1885 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1886 VDEV_AUX_SPLIT_POOL);
1888 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1892 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1893 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1894 VDEV_AUX_CORRUPT_DATA);
1896 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1897 ZPOOL_CONFIG_POOL_GUID);
1902 * If config is not trusted then ignore the spa guid check. This is
1903 * necessary because if the machine crashed during a re-guid the new
1904 * guid might have been written to all of the vdev labels, but not the
1905 * cached config. The check will be performed again once we have the
1906 * trusted config from the MOS.
1908 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1909 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1910 VDEV_AUX_CORRUPT_DATA);
1912 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
1913 "match config (%llu != %llu)", (u_longlong_t)guid,
1914 (u_longlong_t)spa_guid(spa));
1918 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1919 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1923 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
1924 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1925 VDEV_AUX_CORRUPT_DATA);
1927 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1932 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
1934 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1935 VDEV_AUX_CORRUPT_DATA);
1937 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1938 ZPOOL_CONFIG_TOP_GUID);
1943 * If this vdev just became a top-level vdev because its sibling was
1944 * detached, it will have adopted the parent's vdev guid -- but the
1945 * label may or may not be on disk yet. Fortunately, either version
1946 * of the label will have the same top guid, so if we're a top-level
1947 * vdev, we can safely compare to that instead.
1948 * However, if the config comes from a cachefile that failed to update
1949 * after the detach, a top-level vdev will appear as a non top-level
1950 * vdev in the config. Also relax the constraints if we perform an
1953 * If we split this vdev off instead, then we also check the
1954 * original pool's guid. We don't want to consider the vdev
1955 * corrupt if it is partway through a split operation.
1957 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
1958 boolean_t mismatch = B_FALSE;
1959 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
1960 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
1963 if (vd->vdev_guid != top_guid &&
1964 vd->vdev_top->vdev_guid != guid)
1969 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1970 VDEV_AUX_CORRUPT_DATA);
1972 vdev_dbgmsg(vd, "vdev_validate: config guid "
1973 "doesn't match label guid");
1974 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
1975 (u_longlong_t)vd->vdev_guid,
1976 (u_longlong_t)vd->vdev_top->vdev_guid);
1977 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
1978 "aux_guid %llu", (u_longlong_t)guid,
1979 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
1984 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1986 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1987 VDEV_AUX_CORRUPT_DATA);
1989 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1990 ZPOOL_CONFIG_POOL_STATE);
1997 * If this is a verbatim import, no need to check the
1998 * state of the pool.
2000 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
2001 spa_load_state(spa) == SPA_LOAD_OPEN &&
2002 state != POOL_STATE_ACTIVE) {
2003 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
2004 "for spa %s", (u_longlong_t)state, spa->spa_name);
2005 return (SET_ERROR(EBADF));
2009 * If we were able to open and validate a vdev that was
2010 * previously marked permanently unavailable, clear that state
2013 if (vd->vdev_not_present)
2014 vd->vdev_not_present = 0;
2020 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
2022 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
2023 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
2024 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2025 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
2026 dvd->vdev_path, svd->vdev_path);
2027 spa_strfree(dvd->vdev_path);
2028 dvd->vdev_path = spa_strdup(svd->vdev_path);
2030 } else if (svd->vdev_path != NULL) {
2031 dvd->vdev_path = spa_strdup(svd->vdev_path);
2032 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2033 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
2038 * Recursively copy vdev paths from one vdev to another. Source and destination
2039 * vdev trees must have same geometry otherwise return error. Intended to copy
2040 * paths from userland config into MOS config.
2043 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
2045 if ((svd->vdev_ops == &vdev_missing_ops) ||
2046 (svd->vdev_ishole && dvd->vdev_ishole) ||
2047 (dvd->vdev_ops == &vdev_indirect_ops))
2050 if (svd->vdev_ops != dvd->vdev_ops) {
2051 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
2052 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
2053 return (SET_ERROR(EINVAL));
2056 if (svd->vdev_guid != dvd->vdev_guid) {
2057 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
2058 "%llu)", (u_longlong_t)svd->vdev_guid,
2059 (u_longlong_t)dvd->vdev_guid);
2060 return (SET_ERROR(EINVAL));
2063 if (svd->vdev_children != dvd->vdev_children) {
2064 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2065 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2066 (u_longlong_t)dvd->vdev_children);
2067 return (SET_ERROR(EINVAL));
2070 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2071 int error = vdev_copy_path_strict(svd->vdev_child[i],
2072 dvd->vdev_child[i]);
2077 if (svd->vdev_ops->vdev_op_leaf)
2078 vdev_copy_path_impl(svd, dvd);
2084 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2086 ASSERT(stvd->vdev_top == stvd);
2087 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2089 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2090 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2093 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2097 * The idea here is that while a vdev can shift positions within
2098 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2099 * step outside of it.
2101 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2103 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2106 ASSERT(vd->vdev_ops->vdev_op_leaf);
2108 vdev_copy_path_impl(vd, dvd);
2112 * Recursively copy vdev paths from one root vdev to another. Source and
2113 * destination vdev trees may differ in geometry. For each destination leaf
2114 * vdev, search a vdev with the same guid and top vdev id in the source.
2115 * Intended to copy paths from userland config into MOS config.
2118 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2120 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2121 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2122 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2124 for (uint64_t i = 0; i < children; i++) {
2125 vdev_copy_path_search(srvd->vdev_child[i],
2126 drvd->vdev_child[i]);
2131 * Close a virtual device.
2134 vdev_close(vdev_t *vd)
2136 vdev_t *pvd = vd->vdev_parent;
2137 ASSERTV(spa_t *spa = vd->vdev_spa);
2139 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2142 * If our parent is reopening, then we are as well, unless we are
2145 if (pvd != NULL && pvd->vdev_reopening)
2146 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2148 vd->vdev_ops->vdev_op_close(vd);
2150 vdev_cache_purge(vd);
2153 * We record the previous state before we close it, so that if we are
2154 * doing a reopen(), we don't generate FMA ereports if we notice that
2155 * it's still faulted.
2157 vd->vdev_prevstate = vd->vdev_state;
2159 if (vd->vdev_offline)
2160 vd->vdev_state = VDEV_STATE_OFFLINE;
2162 vd->vdev_state = VDEV_STATE_CLOSED;
2163 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2167 vdev_hold(vdev_t *vd)
2169 spa_t *spa = vd->vdev_spa;
2171 ASSERT(spa_is_root(spa));
2172 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2175 for (int c = 0; c < vd->vdev_children; c++)
2176 vdev_hold(vd->vdev_child[c]);
2178 if (vd->vdev_ops->vdev_op_leaf)
2179 vd->vdev_ops->vdev_op_hold(vd);
2183 vdev_rele(vdev_t *vd)
2185 ASSERT(spa_is_root(vd->vdev_spa));
2186 for (int c = 0; c < vd->vdev_children; c++)
2187 vdev_rele(vd->vdev_child[c]);
2189 if (vd->vdev_ops->vdev_op_leaf)
2190 vd->vdev_ops->vdev_op_rele(vd);
2194 * Reopen all interior vdevs and any unopened leaves. We don't actually
2195 * reopen leaf vdevs which had previously been opened as they might deadlock
2196 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2197 * If the leaf has never been opened then open it, as usual.
2200 vdev_reopen(vdev_t *vd)
2202 spa_t *spa = vd->vdev_spa;
2204 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2206 /* set the reopening flag unless we're taking the vdev offline */
2207 vd->vdev_reopening = !vd->vdev_offline;
2209 (void) vdev_open(vd);
2212 * Call vdev_validate() here to make sure we have the same device.
2213 * Otherwise, a device with an invalid label could be successfully
2214 * opened in response to vdev_reopen().
2217 (void) vdev_validate_aux(vd);
2218 if (vdev_readable(vd) && vdev_writeable(vd) &&
2219 vd->vdev_aux == &spa->spa_l2cache &&
2220 !l2arc_vdev_present(vd))
2221 l2arc_add_vdev(spa, vd);
2223 (void) vdev_validate(vd);
2227 * Reassess parent vdev's health.
2229 vdev_propagate_state(vd);
2233 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2238 * Normally, partial opens (e.g. of a mirror) are allowed.
2239 * For a create, however, we want to fail the request if
2240 * there are any components we can't open.
2242 error = vdev_open(vd);
2244 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2246 return (error ? error : ENXIO);
2250 * Recursively load DTLs and initialize all labels.
2252 if ((error = vdev_dtl_load(vd)) != 0 ||
2253 (error = vdev_label_init(vd, txg, isreplacing ?
2254 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2263 vdev_metaslab_set_size(vdev_t *vd)
2265 uint64_t asize = vd->vdev_asize;
2266 uint64_t ms_count = asize >> vdev_default_ms_shift;
2270 * There are two dimensions to the metaslab sizing calculation:
2271 * the size of the metaslab and the count of metaslabs per vdev.
2272 * In general, we aim for vdev_max_ms_count (200) metaslabs. The
2273 * range of the dimensions are as follows:
2275 * 2^29 <= ms_size <= 2^38
2276 * 16 <= ms_count <= 131,072
2278 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2279 * at least 512MB (2^29) to minimize fragmentation effects when
2280 * testing with smaller devices. However, the count constraint
2281 * of at least 16 metaslabs will override this minimum size goal.
2283 * On the upper end of vdev sizes, we aim for a maximum metaslab
2284 * size of 256GB. However, we will cap the total count to 2^17
2285 * metaslabs to keep our memory footprint in check.
2287 * The net effect of applying above constrains is summarized below.
2289 * vdev size metaslab count
2290 * -------------|-----------------
2292 * 8GB - 100GB one per 512MB
2294 * 50TB - 32PB one per 256GB
2296 * -------------------------------
2299 if (ms_count < vdev_min_ms_count)
2300 ms_shift = highbit64(asize / vdev_min_ms_count);
2301 else if (ms_count > vdev_max_ms_count)
2302 ms_shift = highbit64(asize / vdev_max_ms_count);
2304 ms_shift = vdev_default_ms_shift;
2306 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2307 ms_shift = SPA_MAXBLOCKSHIFT;
2308 } else if (ms_shift > vdev_max_ms_shift) {
2309 ms_shift = vdev_max_ms_shift;
2310 /* cap the total count to constrain memory footprint */
2311 if ((asize >> ms_shift) > vdev_ms_count_limit)
2312 ms_shift = highbit64(asize / vdev_ms_count_limit);
2315 vd->vdev_ms_shift = ms_shift;
2316 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2320 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2322 ASSERT(vd == vd->vdev_top);
2323 /* indirect vdevs don't have metaslabs or dtls */
2324 ASSERT(vdev_is_concrete(vd) || flags == 0);
2325 ASSERT(ISP2(flags));
2326 ASSERT(spa_writeable(vd->vdev_spa));
2328 if (flags & VDD_METASLAB)
2329 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2331 if (flags & VDD_DTL)
2332 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2334 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2338 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2340 for (int c = 0; c < vd->vdev_children; c++)
2341 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2343 if (vd->vdev_ops->vdev_op_leaf)
2344 vdev_dirty(vd->vdev_top, flags, vd, txg);
2350 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2351 * the vdev has less than perfect replication. There are four kinds of DTL:
2353 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2355 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2357 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2358 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2359 * txgs that was scrubbed.
2361 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2362 * persistent errors or just some device being offline.
2363 * Unlike the other three, the DTL_OUTAGE map is not generally
2364 * maintained; it's only computed when needed, typically to
2365 * determine whether a device can be detached.
2367 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2368 * either has the data or it doesn't.
2370 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2371 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2372 * if any child is less than fully replicated, then so is its parent.
2373 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2374 * comprising only those txgs which appear in 'maxfaults' or more children;
2375 * those are the txgs we don't have enough replication to read. For example,
2376 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2377 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2378 * two child DTL_MISSING maps.
2380 * It should be clear from the above that to compute the DTLs and outage maps
2381 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2382 * Therefore, that is all we keep on disk. When loading the pool, or after
2383 * a configuration change, we generate all other DTLs from first principles.
2386 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2388 range_tree_t *rt = vd->vdev_dtl[t];
2390 ASSERT(t < DTL_TYPES);
2391 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2392 ASSERT(spa_writeable(vd->vdev_spa));
2394 mutex_enter(&vd->vdev_dtl_lock);
2395 if (!range_tree_contains(rt, txg, size))
2396 range_tree_add(rt, txg, size);
2397 mutex_exit(&vd->vdev_dtl_lock);
2401 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2403 range_tree_t *rt = vd->vdev_dtl[t];
2404 boolean_t dirty = B_FALSE;
2406 ASSERT(t < DTL_TYPES);
2407 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2410 * While we are loading the pool, the DTLs have not been loaded yet.
2411 * Ignore the DTLs and try all devices. This avoids a recursive
2412 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2413 * when loading the pool (relying on the checksum to ensure that
2414 * we get the right data -- note that we while loading, we are
2415 * only reading the MOS, which is always checksummed).
2417 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2420 mutex_enter(&vd->vdev_dtl_lock);
2421 if (!range_tree_is_empty(rt))
2422 dirty = range_tree_contains(rt, txg, size);
2423 mutex_exit(&vd->vdev_dtl_lock);
2429 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2431 range_tree_t *rt = vd->vdev_dtl[t];
2434 mutex_enter(&vd->vdev_dtl_lock);
2435 empty = range_tree_is_empty(rt);
2436 mutex_exit(&vd->vdev_dtl_lock);
2442 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2445 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2447 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2449 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2450 vd->vdev_ops->vdev_op_leaf)
2453 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2457 * Returns the lowest txg in the DTL range.
2460 vdev_dtl_min(vdev_t *vd)
2464 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2465 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2466 ASSERT0(vd->vdev_children);
2468 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2469 return (rs->rs_start - 1);
2473 * Returns the highest txg in the DTL.
2476 vdev_dtl_max(vdev_t *vd)
2480 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2481 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2482 ASSERT0(vd->vdev_children);
2484 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2485 return (rs->rs_end);
2489 * Determine if a resilvering vdev should remove any DTL entries from
2490 * its range. If the vdev was resilvering for the entire duration of the
2491 * scan then it should excise that range from its DTLs. Otherwise, this
2492 * vdev is considered partially resilvered and should leave its DTL
2493 * entries intact. The comment in vdev_dtl_reassess() describes how we
2497 vdev_dtl_should_excise(vdev_t *vd)
2499 spa_t *spa = vd->vdev_spa;
2500 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2502 ASSERT0(scn->scn_phys.scn_errors);
2503 ASSERT0(vd->vdev_children);
2505 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2508 if (vd->vdev_resilver_deferred)
2511 if (vd->vdev_resilver_txg == 0 ||
2512 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2516 * When a resilver is initiated the scan will assign the scn_max_txg
2517 * value to the highest txg value that exists in all DTLs. If this
2518 * device's max DTL is not part of this scan (i.e. it is not in
2519 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2522 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2523 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2524 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2525 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2532 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2533 * write operations will be issued to the pool.
2536 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2538 spa_t *spa = vd->vdev_spa;
2542 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2544 for (int c = 0; c < vd->vdev_children; c++)
2545 vdev_dtl_reassess(vd->vdev_child[c], txg,
2546 scrub_txg, scrub_done);
2548 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2551 if (vd->vdev_ops->vdev_op_leaf) {
2552 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2554 mutex_enter(&vd->vdev_dtl_lock);
2557 * If requested, pretend the scan completed cleanly.
2559 if (zfs_scan_ignore_errors && scn)
2560 scn->scn_phys.scn_errors = 0;
2563 * If we've completed a scan cleanly then determine
2564 * if this vdev should remove any DTLs. We only want to
2565 * excise regions on vdevs that were available during
2566 * the entire duration of this scan.
2568 if (scrub_txg != 0 &&
2569 (spa->spa_scrub_started ||
2570 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2571 vdev_dtl_should_excise(vd)) {
2573 * We completed a scrub up to scrub_txg. If we
2574 * did it without rebooting, then the scrub dtl
2575 * will be valid, so excise the old region and
2576 * fold in the scrub dtl. Otherwise, leave the
2577 * dtl as-is if there was an error.
2579 * There's little trick here: to excise the beginning
2580 * of the DTL_MISSING map, we put it into a reference
2581 * tree and then add a segment with refcnt -1 that
2582 * covers the range [0, scrub_txg). This means
2583 * that each txg in that range has refcnt -1 or 0.
2584 * We then add DTL_SCRUB with a refcnt of 2, so that
2585 * entries in the range [0, scrub_txg) will have a
2586 * positive refcnt -- either 1 or 2. We then convert
2587 * the reference tree into the new DTL_MISSING map.
2589 space_reftree_create(&reftree);
2590 space_reftree_add_map(&reftree,
2591 vd->vdev_dtl[DTL_MISSING], 1);
2592 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2593 space_reftree_add_map(&reftree,
2594 vd->vdev_dtl[DTL_SCRUB], 2);
2595 space_reftree_generate_map(&reftree,
2596 vd->vdev_dtl[DTL_MISSING], 1);
2597 space_reftree_destroy(&reftree);
2599 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2600 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2601 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2603 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2604 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2605 if (!vdev_readable(vd))
2606 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2608 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2609 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2612 * If the vdev was resilvering and no longer has any
2613 * DTLs then reset its resilvering flag and dirty
2614 * the top level so that we persist the change.
2616 if (txg != 0 && vd->vdev_resilver_txg != 0 &&
2617 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2618 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2619 vd->vdev_resilver_txg = 0;
2620 vdev_config_dirty(vd->vdev_top);
2623 mutex_exit(&vd->vdev_dtl_lock);
2626 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2630 mutex_enter(&vd->vdev_dtl_lock);
2631 for (int t = 0; t < DTL_TYPES; t++) {
2632 /* account for child's outage in parent's missing map */
2633 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2635 continue; /* leaf vdevs only */
2636 if (t == DTL_PARTIAL)
2637 minref = 1; /* i.e. non-zero */
2638 else if (vd->vdev_nparity != 0)
2639 minref = vd->vdev_nparity + 1; /* RAID-Z */
2641 minref = vd->vdev_children; /* any kind of mirror */
2642 space_reftree_create(&reftree);
2643 for (int c = 0; c < vd->vdev_children; c++) {
2644 vdev_t *cvd = vd->vdev_child[c];
2645 mutex_enter(&cvd->vdev_dtl_lock);
2646 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2647 mutex_exit(&cvd->vdev_dtl_lock);
2649 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2650 space_reftree_destroy(&reftree);
2652 mutex_exit(&vd->vdev_dtl_lock);
2656 vdev_dtl_load(vdev_t *vd)
2658 spa_t *spa = vd->vdev_spa;
2659 objset_t *mos = spa->spa_meta_objset;
2662 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2663 ASSERT(vdev_is_concrete(vd));
2665 error = space_map_open(&vd->vdev_dtl_sm, mos,
2666 vd->vdev_dtl_object, 0, -1ULL, 0);
2669 ASSERT(vd->vdev_dtl_sm != NULL);
2671 mutex_enter(&vd->vdev_dtl_lock);
2674 * Now that we've opened the space_map we need to update
2677 space_map_update(vd->vdev_dtl_sm);
2679 error = space_map_load(vd->vdev_dtl_sm,
2680 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2681 mutex_exit(&vd->vdev_dtl_lock);
2686 for (int c = 0; c < vd->vdev_children; c++) {
2687 error = vdev_dtl_load(vd->vdev_child[c]);
2696 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
2698 spa_t *spa = vd->vdev_spa;
2699 objset_t *mos = spa->spa_meta_objset;
2700 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
2703 ASSERT(alloc_bias != VDEV_BIAS_NONE);
2706 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
2707 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
2708 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
2710 ASSERT(string != NULL);
2711 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
2712 1, strlen(string) + 1, string, tx));
2714 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
2715 spa_activate_allocation_classes(spa, tx);
2720 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2722 spa_t *spa = vd->vdev_spa;
2724 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2725 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2730 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2732 spa_t *spa = vd->vdev_spa;
2733 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2734 DMU_OT_NONE, 0, tx);
2737 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2744 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2746 if (vd->vdev_ops != &vdev_hole_ops &&
2747 vd->vdev_ops != &vdev_missing_ops &&
2748 vd->vdev_ops != &vdev_root_ops &&
2749 !vd->vdev_top->vdev_removing) {
2750 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2751 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2753 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2754 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2755 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
2756 vdev_zap_allocation_data(vd, tx);
2760 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2761 vdev_construct_zaps(vd->vdev_child[i], tx);
2766 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2768 spa_t *spa = vd->vdev_spa;
2769 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2770 objset_t *mos = spa->spa_meta_objset;
2771 range_tree_t *rtsync;
2773 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2775 ASSERT(vdev_is_concrete(vd));
2776 ASSERT(vd->vdev_ops->vdev_op_leaf);
2778 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2780 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2781 mutex_enter(&vd->vdev_dtl_lock);
2782 space_map_free(vd->vdev_dtl_sm, tx);
2783 space_map_close(vd->vdev_dtl_sm);
2784 vd->vdev_dtl_sm = NULL;
2785 mutex_exit(&vd->vdev_dtl_lock);
2788 * We only destroy the leaf ZAP for detached leaves or for
2789 * removed log devices. Removed data devices handle leaf ZAP
2790 * cleanup later, once cancellation is no longer possible.
2792 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2793 vd->vdev_top->vdev_islog)) {
2794 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2795 vd->vdev_leaf_zap = 0;
2802 if (vd->vdev_dtl_sm == NULL) {
2803 uint64_t new_object;
2805 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx);
2806 VERIFY3U(new_object, !=, 0);
2808 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2810 ASSERT(vd->vdev_dtl_sm != NULL);
2813 rtsync = range_tree_create(NULL, NULL);
2815 mutex_enter(&vd->vdev_dtl_lock);
2816 range_tree_walk(rt, range_tree_add, rtsync);
2817 mutex_exit(&vd->vdev_dtl_lock);
2819 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx);
2820 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
2821 range_tree_vacate(rtsync, NULL, NULL);
2823 range_tree_destroy(rtsync);
2826 * If the object for the space map has changed then dirty
2827 * the top level so that we update the config.
2829 if (object != space_map_object(vd->vdev_dtl_sm)) {
2830 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2831 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2832 (u_longlong_t)object,
2833 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2834 vdev_config_dirty(vd->vdev_top);
2839 mutex_enter(&vd->vdev_dtl_lock);
2840 space_map_update(vd->vdev_dtl_sm);
2841 mutex_exit(&vd->vdev_dtl_lock);
2845 * Determine whether the specified vdev can be offlined/detached/removed
2846 * without losing data.
2849 vdev_dtl_required(vdev_t *vd)
2851 spa_t *spa = vd->vdev_spa;
2852 vdev_t *tvd = vd->vdev_top;
2853 uint8_t cant_read = vd->vdev_cant_read;
2856 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2858 if (vd == spa->spa_root_vdev || vd == tvd)
2862 * Temporarily mark the device as unreadable, and then determine
2863 * whether this results in any DTL outages in the top-level vdev.
2864 * If not, we can safely offline/detach/remove the device.
2866 vd->vdev_cant_read = B_TRUE;
2867 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2868 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2869 vd->vdev_cant_read = cant_read;
2870 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2872 if (!required && zio_injection_enabled)
2873 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2879 * Determine if resilver is needed, and if so the txg range.
2882 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2884 boolean_t needed = B_FALSE;
2885 uint64_t thismin = UINT64_MAX;
2886 uint64_t thismax = 0;
2888 if (vd->vdev_children == 0) {
2889 mutex_enter(&vd->vdev_dtl_lock);
2890 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2891 vdev_writeable(vd)) {
2893 thismin = vdev_dtl_min(vd);
2894 thismax = vdev_dtl_max(vd);
2897 mutex_exit(&vd->vdev_dtl_lock);
2899 for (int c = 0; c < vd->vdev_children; c++) {
2900 vdev_t *cvd = vd->vdev_child[c];
2901 uint64_t cmin, cmax;
2903 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2904 thismin = MIN(thismin, cmin);
2905 thismax = MAX(thismax, cmax);
2911 if (needed && minp) {
2919 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
2920 * will contain either the checkpoint spacemap object or zero if none exists.
2921 * All other errors are returned to the caller.
2924 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj)
2926 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
2928 if (vd->vdev_top_zap == 0) {
2933 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
2934 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj);
2935 if (error == ENOENT) {
2944 vdev_load(vdev_t *vd)
2949 * Recursively load all children.
2951 for (int c = 0; c < vd->vdev_children; c++) {
2952 error = vdev_load(vd->vdev_child[c]);
2958 vdev_set_deflate_ratio(vd);
2961 * On spa_load path, grab the allocation bias from our zap
2963 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
2964 spa_t *spa = vd->vdev_spa;
2967 if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
2968 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
2970 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
2971 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
2976 * If this is a top-level vdev, initialize its metaslabs.
2978 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2979 vdev_metaslab_group_create(vd);
2981 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2982 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2983 VDEV_AUX_CORRUPT_DATA);
2984 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2985 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2986 (u_longlong_t)vd->vdev_asize);
2987 return (SET_ERROR(ENXIO));
2988 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2989 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2990 "[error=%d]", error);
2991 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2992 VDEV_AUX_CORRUPT_DATA);
2996 uint64_t checkpoint_sm_obj;
2997 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj);
2998 if (error == 0 && checkpoint_sm_obj != 0) {
2999 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3000 ASSERT(vd->vdev_asize != 0);
3001 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
3003 if ((error = space_map_open(&vd->vdev_checkpoint_sm,
3004 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
3005 vd->vdev_ashift))) {
3006 vdev_dbgmsg(vd, "vdev_load: space_map_open "
3007 "failed for checkpoint spacemap (obj %llu) "
3009 (u_longlong_t)checkpoint_sm_obj, error);
3012 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
3013 space_map_update(vd->vdev_checkpoint_sm);
3016 * Since the checkpoint_sm contains free entries
3017 * exclusively we can use sm_alloc to indicate the
3018 * cumulative checkpointed space that has been freed.
3020 vd->vdev_stat.vs_checkpoint_space =
3021 -vd->vdev_checkpoint_sm->sm_alloc;
3022 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
3023 vd->vdev_stat.vs_checkpoint_space;
3024 } else if (error != 0) {
3025 vdev_dbgmsg(vd, "vdev_load: failed to retrieve "
3026 "checkpoint space map object from vdev ZAP "
3027 "[error=%d]", error);
3033 * If this is a leaf vdev, load its DTL.
3035 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
3036 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3037 VDEV_AUX_CORRUPT_DATA);
3038 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
3039 "[error=%d]", error);
3043 uint64_t obsolete_sm_object;
3044 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object);
3045 if (error == 0 && obsolete_sm_object != 0) {
3046 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3047 ASSERT(vd->vdev_asize != 0);
3048 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
3050 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
3051 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
3052 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3053 VDEV_AUX_CORRUPT_DATA);
3054 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
3055 "obsolete spacemap (obj %llu) [error=%d]",
3056 (u_longlong_t)obsolete_sm_object, error);
3059 space_map_update(vd->vdev_obsolete_sm);
3060 } else if (error != 0) {
3061 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete "
3062 "space map object from vdev ZAP [error=%d]", error);
3070 * The special vdev case is used for hot spares and l2cache devices. Its
3071 * sole purpose it to set the vdev state for the associated vdev. To do this,
3072 * we make sure that we can open the underlying device, then try to read the
3073 * label, and make sure that the label is sane and that it hasn't been
3074 * repurposed to another pool.
3077 vdev_validate_aux(vdev_t *vd)
3080 uint64_t guid, version;
3083 if (!vdev_readable(vd))
3086 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3087 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3088 VDEV_AUX_CORRUPT_DATA);
3092 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3093 !SPA_VERSION_IS_SUPPORTED(version) ||
3094 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3095 guid != vd->vdev_guid ||
3096 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3097 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3098 VDEV_AUX_CORRUPT_DATA);
3104 * We don't actually check the pool state here. If it's in fact in
3105 * use by another pool, we update this fact on the fly when requested.
3112 * Free the objects used to store this vdev's spacemaps, and the array
3113 * that points to them.
3116 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3118 if (vd->vdev_ms_array == 0)
3121 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3122 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3123 size_t array_bytes = array_count * sizeof (uint64_t);
3124 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3125 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3126 array_bytes, smobj_array, 0));
3128 for (uint64_t i = 0; i < array_count; i++) {
3129 uint64_t smobj = smobj_array[i];
3133 space_map_free_obj(mos, smobj, tx);
3136 kmem_free(smobj_array, array_bytes);
3137 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3138 vd->vdev_ms_array = 0;
3142 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3144 spa_t *spa = vd->vdev_spa;
3146 ASSERT(vd->vdev_islog);
3147 ASSERT(vd == vd->vdev_top);
3148 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3150 if (vd->vdev_ms != NULL) {
3151 metaslab_group_t *mg = vd->vdev_mg;
3153 metaslab_group_histogram_verify(mg);
3154 metaslab_class_histogram_verify(mg->mg_class);
3156 for (int m = 0; m < vd->vdev_ms_count; m++) {
3157 metaslab_t *msp = vd->vdev_ms[m];
3159 if (msp == NULL || msp->ms_sm == NULL)
3162 mutex_enter(&msp->ms_lock);
3164 * If the metaslab was not loaded when the vdev
3165 * was removed then the histogram accounting may
3166 * not be accurate. Update the histogram information
3167 * here so that we ensure that the metaslab group
3168 * and metaslab class are up-to-date.
3170 metaslab_group_histogram_remove(mg, msp);
3172 VERIFY0(space_map_allocated(msp->ms_sm));
3173 space_map_close(msp->ms_sm);
3175 mutex_exit(&msp->ms_lock);
3178 if (vd->vdev_checkpoint_sm != NULL) {
3179 ASSERT(spa_has_checkpoint(spa));
3180 space_map_close(vd->vdev_checkpoint_sm);
3181 vd->vdev_checkpoint_sm = NULL;
3184 metaslab_group_histogram_verify(mg);
3185 metaslab_class_histogram_verify(mg->mg_class);
3187 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
3188 ASSERT0(mg->mg_histogram[i]);
3191 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3193 vdev_destroy_spacemaps(vd, tx);
3194 if (vd->vdev_top_zap != 0) {
3195 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3196 vd->vdev_top_zap = 0;
3203 vdev_sync_done(vdev_t *vd, uint64_t txg)
3206 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3208 ASSERT(vdev_is_concrete(vd));
3210 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
3211 metaslab_sync_done(msp, txg);
3214 metaslab_sync_reassess(vd->vdev_mg);
3218 vdev_sync(vdev_t *vd, uint64_t txg)
3220 spa_t *spa = vd->vdev_spa;
3225 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3228 ASSERT(vd->vdev_removing ||
3229 vd->vdev_ops == &vdev_indirect_ops);
3231 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3232 vdev_indirect_sync_obsolete(vd, tx);
3236 * If the vdev is indirect, it can't have dirty
3237 * metaslabs or DTLs.
3239 if (vd->vdev_ops == &vdev_indirect_ops) {
3240 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3241 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3246 ASSERT(vdev_is_concrete(vd));
3248 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3249 !vd->vdev_removing) {
3250 ASSERT(vd == vd->vdev_top);
3251 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3252 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3253 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3254 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3255 ASSERT(vd->vdev_ms_array != 0);
3256 vdev_config_dirty(vd);
3260 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3261 metaslab_sync(msp, txg);
3262 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3265 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3266 vdev_dtl_sync(lvd, txg);
3269 * If this is an empty log device being removed, destroy the
3270 * metadata associated with it.
3272 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3273 vdev_remove_empty_log(vd, txg);
3275 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3279 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3281 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3285 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3286 * not be opened, and no I/O is attempted.
3289 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3293 spa_vdev_state_enter(spa, SCL_NONE);
3295 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3296 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3298 if (!vd->vdev_ops->vdev_op_leaf)
3299 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3304 * If user did a 'zpool offline -f' then make the fault persist across
3307 if (aux == VDEV_AUX_EXTERNAL_PERSIST) {
3309 * There are two kinds of forced faults: temporary and
3310 * persistent. Temporary faults go away at pool import, while
3311 * persistent faults stay set. Both types of faults can be
3312 * cleared with a zpool clear.
3314 * We tell if a vdev is persistently faulted by looking at the
3315 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3316 * import then it's a persistent fault. Otherwise, it's
3317 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3318 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3319 * tells vdev_config_generate() (which gets run later) to set
3320 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3322 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
3323 vd->vdev_tmpoffline = B_FALSE;
3324 aux = VDEV_AUX_EXTERNAL;
3326 vd->vdev_tmpoffline = B_TRUE;
3330 * We don't directly use the aux state here, but if we do a
3331 * vdev_reopen(), we need this value to be present to remember why we
3334 vd->vdev_label_aux = aux;
3337 * Faulted state takes precedence over degraded.
3339 vd->vdev_delayed_close = B_FALSE;
3340 vd->vdev_faulted = 1ULL;
3341 vd->vdev_degraded = 0ULL;
3342 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3345 * If this device has the only valid copy of the data, then
3346 * back off and simply mark the vdev as degraded instead.
3348 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3349 vd->vdev_degraded = 1ULL;
3350 vd->vdev_faulted = 0ULL;
3353 * If we reopen the device and it's not dead, only then do we
3358 if (vdev_readable(vd))
3359 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3362 return (spa_vdev_state_exit(spa, vd, 0));
3366 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3367 * user that something is wrong. The vdev continues to operate as normal as far
3368 * as I/O is concerned.
3371 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3375 spa_vdev_state_enter(spa, SCL_NONE);
3377 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3378 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3380 if (!vd->vdev_ops->vdev_op_leaf)
3381 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3384 * If the vdev is already faulted, then don't do anything.
3386 if (vd->vdev_faulted || vd->vdev_degraded)
3387 return (spa_vdev_state_exit(spa, NULL, 0));
3389 vd->vdev_degraded = 1ULL;
3390 if (!vdev_is_dead(vd))
3391 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3394 return (spa_vdev_state_exit(spa, vd, 0));
3398 * Online the given vdev.
3400 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3401 * spare device should be detached when the device finishes resilvering.
3402 * Second, the online should be treated like a 'test' online case, so no FMA
3403 * events are generated if the device fails to open.
3406 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3408 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3409 boolean_t wasoffline;
3410 vdev_state_t oldstate;
3412 spa_vdev_state_enter(spa, SCL_NONE);
3414 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3415 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3417 if (!vd->vdev_ops->vdev_op_leaf)
3418 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3420 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3421 oldstate = vd->vdev_state;
3424 vd->vdev_offline = B_FALSE;
3425 vd->vdev_tmpoffline = B_FALSE;
3426 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3427 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3429 /* XXX - L2ARC 1.0 does not support expansion */
3430 if (!vd->vdev_aux) {
3431 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3432 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
3433 spa->spa_autoexpand);
3434 vd->vdev_expansion_time = gethrestime_sec();
3438 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3440 if (!vd->vdev_aux) {
3441 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3442 pvd->vdev_expanding = B_FALSE;
3446 *newstate = vd->vdev_state;
3447 if ((flags & ZFS_ONLINE_UNSPARE) &&
3448 !vdev_is_dead(vd) && vd->vdev_parent &&
3449 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3450 vd->vdev_parent->vdev_child[0] == vd)
3451 vd->vdev_unspare = B_TRUE;
3453 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3455 /* XXX - L2ARC 1.0 does not support expansion */
3457 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3458 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3462 (oldstate < VDEV_STATE_DEGRADED &&
3463 vd->vdev_state >= VDEV_STATE_DEGRADED))
3464 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3466 return (spa_vdev_state_exit(spa, vd, 0));
3470 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3474 uint64_t generation;
3475 metaslab_group_t *mg;
3478 spa_vdev_state_enter(spa, SCL_ALLOC);
3480 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3481 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3483 if (!vd->vdev_ops->vdev_op_leaf)
3484 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3488 generation = spa->spa_config_generation + 1;
3491 * If the device isn't already offline, try to offline it.
3493 if (!vd->vdev_offline) {
3495 * If this device has the only valid copy of some data,
3496 * don't allow it to be offlined. Log devices are always
3499 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3500 vdev_dtl_required(vd))
3501 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3504 * If the top-level is a slog and it has had allocations
3505 * then proceed. We check that the vdev's metaslab group
3506 * is not NULL since it's possible that we may have just
3507 * added this vdev but not yet initialized its metaslabs.
3509 if (tvd->vdev_islog && mg != NULL) {
3511 * Prevent any future allocations.
3513 metaslab_group_passivate(mg);
3514 (void) spa_vdev_state_exit(spa, vd, 0);
3516 error = spa_reset_logs(spa);
3519 * If the log device was successfully reset but has
3520 * checkpointed data, do not offline it.
3523 tvd->vdev_checkpoint_sm != NULL) {
3524 ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc,
3526 error = ZFS_ERR_CHECKPOINT_EXISTS;
3529 spa_vdev_state_enter(spa, SCL_ALLOC);
3532 * Check to see if the config has changed.
3534 if (error || generation != spa->spa_config_generation) {
3535 metaslab_group_activate(mg);
3537 return (spa_vdev_state_exit(spa,
3539 (void) spa_vdev_state_exit(spa, vd, 0);
3542 ASSERT0(tvd->vdev_stat.vs_alloc);
3546 * Offline this device and reopen its top-level vdev.
3547 * If the top-level vdev is a log device then just offline
3548 * it. Otherwise, if this action results in the top-level
3549 * vdev becoming unusable, undo it and fail the request.
3551 vd->vdev_offline = B_TRUE;
3554 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3555 vdev_is_dead(tvd)) {
3556 vd->vdev_offline = B_FALSE;
3558 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3562 * Add the device back into the metaslab rotor so that
3563 * once we online the device it's open for business.
3565 if (tvd->vdev_islog && mg != NULL)
3566 metaslab_group_activate(mg);
3569 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3571 return (spa_vdev_state_exit(spa, vd, 0));
3575 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3579 mutex_enter(&spa->spa_vdev_top_lock);
3580 error = vdev_offline_locked(spa, guid, flags);
3581 mutex_exit(&spa->spa_vdev_top_lock);
3587 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3588 * vdev_offline(), we assume the spa config is locked. We also clear all
3589 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3592 vdev_clear(spa_t *spa, vdev_t *vd)
3594 vdev_t *rvd = spa->spa_root_vdev;
3596 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3601 vd->vdev_stat.vs_read_errors = 0;
3602 vd->vdev_stat.vs_write_errors = 0;
3603 vd->vdev_stat.vs_checksum_errors = 0;
3604 vd->vdev_stat.vs_slow_ios = 0;
3606 for (int c = 0; c < vd->vdev_children; c++)
3607 vdev_clear(spa, vd->vdev_child[c]);
3610 * It makes no sense to "clear" an indirect vdev.
3612 if (!vdev_is_concrete(vd))
3616 * If we're in the FAULTED state or have experienced failed I/O, then
3617 * clear the persistent state and attempt to reopen the device. We
3618 * also mark the vdev config dirty, so that the new faulted state is
3619 * written out to disk.
3621 if (vd->vdev_faulted || vd->vdev_degraded ||
3622 !vdev_readable(vd) || !vdev_writeable(vd)) {
3624 * When reopening in response to a clear event, it may be due to
3625 * a fmadm repair request. In this case, if the device is
3626 * still broken, we want to still post the ereport again.
3628 vd->vdev_forcefault = B_TRUE;
3630 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3631 vd->vdev_cant_read = B_FALSE;
3632 vd->vdev_cant_write = B_FALSE;
3633 vd->vdev_stat.vs_aux = 0;
3635 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3637 vd->vdev_forcefault = B_FALSE;
3639 if (vd != rvd && vdev_writeable(vd->vdev_top))
3640 vdev_state_dirty(vd->vdev_top);
3642 if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) {
3643 if (dsl_scan_resilvering(spa->spa_dsl_pool) &&
3644 spa_feature_is_enabled(spa,
3645 SPA_FEATURE_RESILVER_DEFER))
3646 vdev_set_deferred_resilver(spa, vd);
3648 spa_async_request(spa, SPA_ASYNC_RESILVER);
3651 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3655 * When clearing a FMA-diagnosed fault, we always want to
3656 * unspare the device, as we assume that the original spare was
3657 * done in response to the FMA fault.
3659 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3660 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3661 vd->vdev_parent->vdev_child[0] == vd)
3662 vd->vdev_unspare = B_TRUE;
3666 vdev_is_dead(vdev_t *vd)
3669 * Holes and missing devices are always considered "dead".
3670 * This simplifies the code since we don't have to check for
3671 * these types of devices in the various code paths.
3672 * Instead we rely on the fact that we skip over dead devices
3673 * before issuing I/O to them.
3675 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3676 vd->vdev_ops == &vdev_hole_ops ||
3677 vd->vdev_ops == &vdev_missing_ops);
3681 vdev_readable(vdev_t *vd)
3683 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3687 vdev_writeable(vdev_t *vd)
3689 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3690 vdev_is_concrete(vd));
3694 vdev_allocatable(vdev_t *vd)
3696 uint64_t state = vd->vdev_state;
3699 * We currently allow allocations from vdevs which may be in the
3700 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3701 * fails to reopen then we'll catch it later when we're holding
3702 * the proper locks. Note that we have to get the vdev state
3703 * in a local variable because although it changes atomically,
3704 * we're asking two separate questions about it.
3706 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3707 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3708 vd->vdev_mg->mg_initialized);
3712 vdev_accessible(vdev_t *vd, zio_t *zio)
3714 ASSERT(zio->io_vd == vd);
3716 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3719 if (zio->io_type == ZIO_TYPE_READ)
3720 return (!vd->vdev_cant_read);
3722 if (zio->io_type == ZIO_TYPE_WRITE)
3723 return (!vd->vdev_cant_write);
3729 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs)
3732 for (t = 0; t < ZIO_TYPES; t++) {
3733 vs->vs_ops[t] += cvs->vs_ops[t];
3734 vs->vs_bytes[t] += cvs->vs_bytes[t];
3737 cvs->vs_scan_removing = cvd->vdev_removing;
3741 * Get extended stats
3744 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx)
3747 for (t = 0; t < ZIO_TYPES; t++) {
3748 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++)
3749 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b];
3751 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) {
3752 vsx->vsx_total_histo[t][b] +=
3753 cvsx->vsx_total_histo[t][b];
3757 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
3758 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) {
3759 vsx->vsx_queue_histo[t][b] +=
3760 cvsx->vsx_queue_histo[t][b];
3762 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t];
3763 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t];
3765 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++)
3766 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b];
3768 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++)
3769 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b];
3775 vdev_is_spacemap_addressable(vdev_t *vd)
3778 * Assuming 47 bits of the space map entry dedicated for the entry's
3779 * offset (see description in space_map.h), we calculate the maximum
3780 * address that can be described by a space map entry for the given
3783 uint64_t shift = vd->vdev_ashift + 47;
3785 if (shift >= 63) /* detect potential overflow */
3788 return (vd->vdev_asize < (1ULL << shift));
3792 * Get statistics for the given vdev.
3795 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
3799 * If we're getting stats on the root vdev, aggregate the I/O counts
3800 * over all top-level vdevs (i.e. the direct children of the root).
3802 if (!vd->vdev_ops->vdev_op_leaf) {
3804 memset(vs->vs_ops, 0, sizeof (vs->vs_ops));
3805 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes));
3808 memset(vsx, 0, sizeof (*vsx));
3810 for (int c = 0; c < vd->vdev_children; c++) {
3811 vdev_t *cvd = vd->vdev_child[c];
3812 vdev_stat_t *cvs = &cvd->vdev_stat;
3813 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex;
3815 vdev_get_stats_ex_impl(cvd, cvs, cvsx);
3817 vdev_get_child_stat(cvd, vs, cvs);
3819 vdev_get_child_stat_ex(cvd, vsx, cvsx);
3824 * We're a leaf. Just copy our ZIO active queue stats in. The
3825 * other leaf stats are updated in vdev_stat_update().
3830 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex));
3832 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) {
3833 vsx->vsx_active_queue[t] =
3834 vd->vdev_queue.vq_class[t].vqc_active;
3835 vsx->vsx_pend_queue[t] = avl_numnodes(
3836 &vd->vdev_queue.vq_class[t].vqc_queued_tree);
3842 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
3844 vdev_t *tvd = vd->vdev_top;
3845 mutex_enter(&vd->vdev_stat_lock);
3847 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3848 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3849 vs->vs_state = vd->vdev_state;
3850 vs->vs_rsize = vdev_get_min_asize(vd);
3851 if (vd->vdev_ops->vdev_op_leaf)
3852 vs->vs_rsize += VDEV_LABEL_START_SIZE +
3853 VDEV_LABEL_END_SIZE;
3855 * Report expandable space on top-level, non-auxillary devices
3856 * only. The expandable space is reported in terms of metaslab
3857 * sized units since that determines how much space the pool
3860 if (vd->vdev_aux == NULL && tvd != NULL) {
3861 vs->vs_esize = P2ALIGN(
3862 vd->vdev_max_asize - vd->vdev_asize,
3863 1ULL << tvd->vdev_ms_shift);
3865 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3866 vdev_is_concrete(vd)) {
3867 vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
3868 vd->vdev_mg->mg_fragmentation : 0;
3870 if (vd->vdev_ops->vdev_op_leaf)
3871 vs->vs_resilver_deferred = vd->vdev_resilver_deferred;
3874 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_READER) != 0);
3875 vdev_get_stats_ex_impl(vd, vs, vsx);
3876 mutex_exit(&vd->vdev_stat_lock);
3880 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3882 return (vdev_get_stats_ex(vd, vs, NULL));
3886 vdev_clear_stats(vdev_t *vd)
3888 mutex_enter(&vd->vdev_stat_lock);
3889 vd->vdev_stat.vs_space = 0;
3890 vd->vdev_stat.vs_dspace = 0;
3891 vd->vdev_stat.vs_alloc = 0;
3892 mutex_exit(&vd->vdev_stat_lock);
3896 vdev_scan_stat_init(vdev_t *vd)
3898 vdev_stat_t *vs = &vd->vdev_stat;
3900 for (int c = 0; c < vd->vdev_children; c++)
3901 vdev_scan_stat_init(vd->vdev_child[c]);
3903 mutex_enter(&vd->vdev_stat_lock);
3904 vs->vs_scan_processed = 0;
3905 mutex_exit(&vd->vdev_stat_lock);
3909 vdev_stat_update(zio_t *zio, uint64_t psize)
3911 spa_t *spa = zio->io_spa;
3912 vdev_t *rvd = spa->spa_root_vdev;
3913 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3915 uint64_t txg = zio->io_txg;
3916 vdev_stat_t *vs = &vd->vdev_stat;
3917 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex;
3918 zio_type_t type = zio->io_type;
3919 int flags = zio->io_flags;
3922 * If this i/o is a gang leader, it didn't do any actual work.
3924 if (zio->io_gang_tree)
3927 if (zio->io_error == 0) {
3929 * If this is a root i/o, don't count it -- we've already
3930 * counted the top-level vdevs, and vdev_get_stats() will
3931 * aggregate them when asked. This reduces contention on
3932 * the root vdev_stat_lock and implicitly handles blocks
3933 * that compress away to holes, for which there is no i/o.
3934 * (Holes never create vdev children, so all the counters
3935 * remain zero, which is what we want.)
3937 * Note: this only applies to successful i/o (io_error == 0)
3938 * because unlike i/o counts, errors are not additive.
3939 * When reading a ditto block, for example, failure of
3940 * one top-level vdev does not imply a root-level error.
3945 ASSERT(vd == zio->io_vd);
3947 if (flags & ZIO_FLAG_IO_BYPASS)
3950 mutex_enter(&vd->vdev_stat_lock);
3952 if (flags & ZIO_FLAG_IO_REPAIR) {
3953 if (flags & ZIO_FLAG_SCAN_THREAD) {
3954 dsl_scan_phys_t *scn_phys =
3955 &spa->spa_dsl_pool->dp_scan->scn_phys;
3956 uint64_t *processed = &scn_phys->scn_processed;
3959 if (vd->vdev_ops->vdev_op_leaf)
3960 atomic_add_64(processed, psize);
3961 vs->vs_scan_processed += psize;
3964 if (flags & ZIO_FLAG_SELF_HEAL)
3965 vs->vs_self_healed += psize;
3969 * The bytes/ops/histograms are recorded at the leaf level and
3970 * aggregated into the higher level vdevs in vdev_get_stats().
3972 if (vd->vdev_ops->vdev_op_leaf &&
3973 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) {
3976 vs->vs_bytes[type] += psize;
3978 if (flags & ZIO_FLAG_DELEGATED) {
3979 vsx->vsx_agg_histo[zio->io_priority]
3980 [RQ_HISTO(zio->io_size)]++;
3982 vsx->vsx_ind_histo[zio->io_priority]
3983 [RQ_HISTO(zio->io_size)]++;
3986 if (zio->io_delta && zio->io_delay) {
3987 vsx->vsx_queue_histo[zio->io_priority]
3988 [L_HISTO(zio->io_delta - zio->io_delay)]++;
3989 vsx->vsx_disk_histo[type]
3990 [L_HISTO(zio->io_delay)]++;
3991 vsx->vsx_total_histo[type]
3992 [L_HISTO(zio->io_delta)]++;
3996 mutex_exit(&vd->vdev_stat_lock);
4000 if (flags & ZIO_FLAG_SPECULATIVE)
4004 * If this is an I/O error that is going to be retried, then ignore the
4005 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4006 * hard errors, when in reality they can happen for any number of
4007 * innocuous reasons (bus resets, MPxIO link failure, etc).
4009 if (zio->io_error == EIO &&
4010 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
4014 * Intent logs writes won't propagate their error to the root
4015 * I/O so don't mark these types of failures as pool-level
4018 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
4021 mutex_enter(&vd->vdev_stat_lock);
4022 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
4023 if (zio->io_error == ECKSUM)
4024 vs->vs_checksum_errors++;
4026 vs->vs_read_errors++;
4028 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
4029 vs->vs_write_errors++;
4030 mutex_exit(&vd->vdev_stat_lock);
4032 if (spa->spa_load_state == SPA_LOAD_NONE &&
4033 type == ZIO_TYPE_WRITE && txg != 0 &&
4034 (!(flags & ZIO_FLAG_IO_REPAIR) ||
4035 (flags & ZIO_FLAG_SCAN_THREAD) ||
4036 spa->spa_claiming)) {
4038 * This is either a normal write (not a repair), or it's
4039 * a repair induced by the scrub thread, or it's a repair
4040 * made by zil_claim() during spa_load() in the first txg.
4041 * In the normal case, we commit the DTL change in the same
4042 * txg as the block was born. In the scrub-induced repair
4043 * case, we know that scrubs run in first-pass syncing context,
4044 * so we commit the DTL change in spa_syncing_txg(spa).
4045 * In the zil_claim() case, we commit in spa_first_txg(spa).
4047 * We currently do not make DTL entries for failed spontaneous
4048 * self-healing writes triggered by normal (non-scrubbing)
4049 * reads, because we have no transactional context in which to
4050 * do so -- and it's not clear that it'd be desirable anyway.
4052 if (vd->vdev_ops->vdev_op_leaf) {
4053 uint64_t commit_txg = txg;
4054 if (flags & ZIO_FLAG_SCAN_THREAD) {
4055 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4056 ASSERT(spa_sync_pass(spa) == 1);
4057 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
4058 commit_txg = spa_syncing_txg(spa);
4059 } else if (spa->spa_claiming) {
4060 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4061 commit_txg = spa_first_txg(spa);
4063 ASSERT(commit_txg >= spa_syncing_txg(spa));
4064 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
4066 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
4067 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
4068 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
4071 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
4076 vdev_deflated_space(vdev_t *vd, int64_t space)
4078 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
4079 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
4081 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
4085 * Update the in-core space usage stats for this vdev and the root vdev.
4088 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
4089 int64_t space_delta)
4091 int64_t dspace_delta;
4092 spa_t *spa = vd->vdev_spa;
4093 vdev_t *rvd = spa->spa_root_vdev;
4095 ASSERT(vd == vd->vdev_top);
4098 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4099 * factor. We must calculate this here and not at the root vdev
4100 * because the root vdev's psize-to-asize is simply the max of its
4101 * childrens', thus not accurate enough for us.
4103 dspace_delta = vdev_deflated_space(vd, space_delta);
4105 mutex_enter(&vd->vdev_stat_lock);
4106 vd->vdev_stat.vs_alloc += alloc_delta;
4107 vd->vdev_stat.vs_space += space_delta;
4108 vd->vdev_stat.vs_dspace += dspace_delta;
4109 mutex_exit(&vd->vdev_stat_lock);
4111 /* every class but log contributes to root space stats */
4112 if (vd->vdev_mg != NULL && !vd->vdev_islog) {
4113 mutex_enter(&rvd->vdev_stat_lock);
4114 rvd->vdev_stat.vs_alloc += alloc_delta;
4115 rvd->vdev_stat.vs_space += space_delta;
4116 rvd->vdev_stat.vs_dspace += dspace_delta;
4117 mutex_exit(&rvd->vdev_stat_lock);
4119 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4123 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4124 * so that it will be written out next time the vdev configuration is synced.
4125 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4128 vdev_config_dirty(vdev_t *vd)
4130 spa_t *spa = vd->vdev_spa;
4131 vdev_t *rvd = spa->spa_root_vdev;
4134 ASSERT(spa_writeable(spa));
4137 * If this is an aux vdev (as with l2cache and spare devices), then we
4138 * update the vdev config manually and set the sync flag.
4140 if (vd->vdev_aux != NULL) {
4141 spa_aux_vdev_t *sav = vd->vdev_aux;
4145 for (c = 0; c < sav->sav_count; c++) {
4146 if (sav->sav_vdevs[c] == vd)
4150 if (c == sav->sav_count) {
4152 * We're being removed. There's nothing more to do.
4154 ASSERT(sav->sav_sync == B_TRUE);
4158 sav->sav_sync = B_TRUE;
4160 if (nvlist_lookup_nvlist_array(sav->sav_config,
4161 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
4162 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
4163 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
4169 * Setting the nvlist in the middle if the array is a little
4170 * sketchy, but it will work.
4172 nvlist_free(aux[c]);
4173 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
4179 * The dirty list is protected by the SCL_CONFIG lock. The caller
4180 * must either hold SCL_CONFIG as writer, or must be the sync thread
4181 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4182 * so this is sufficient to ensure mutual exclusion.
4184 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4185 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4186 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4189 for (c = 0; c < rvd->vdev_children; c++)
4190 vdev_config_dirty(rvd->vdev_child[c]);
4192 ASSERT(vd == vd->vdev_top);
4194 if (!list_link_active(&vd->vdev_config_dirty_node) &&
4195 vdev_is_concrete(vd)) {
4196 list_insert_head(&spa->spa_config_dirty_list, vd);
4202 vdev_config_clean(vdev_t *vd)
4204 spa_t *spa = vd->vdev_spa;
4206 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4207 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4208 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4210 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
4211 list_remove(&spa->spa_config_dirty_list, vd);
4215 * Mark a top-level vdev's state as dirty, so that the next pass of
4216 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4217 * the state changes from larger config changes because they require
4218 * much less locking, and are often needed for administrative actions.
4221 vdev_state_dirty(vdev_t *vd)
4223 spa_t *spa = vd->vdev_spa;
4225 ASSERT(spa_writeable(spa));
4226 ASSERT(vd == vd->vdev_top);
4229 * The state list is protected by the SCL_STATE lock. The caller
4230 * must either hold SCL_STATE as writer, or must be the sync thread
4231 * (which holds SCL_STATE as reader). There's only one sync thread,
4232 * so this is sufficient to ensure mutual exclusion.
4234 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4235 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4236 spa_config_held(spa, SCL_STATE, RW_READER)));
4238 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4239 vdev_is_concrete(vd))
4240 list_insert_head(&spa->spa_state_dirty_list, vd);
4244 vdev_state_clean(vdev_t *vd)
4246 spa_t *spa = vd->vdev_spa;
4248 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4249 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4250 spa_config_held(spa, SCL_STATE, RW_READER)));
4252 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4253 list_remove(&spa->spa_state_dirty_list, vd);
4257 * Propagate vdev state up from children to parent.
4260 vdev_propagate_state(vdev_t *vd)
4262 spa_t *spa = vd->vdev_spa;
4263 vdev_t *rvd = spa->spa_root_vdev;
4264 int degraded = 0, faulted = 0;
4268 if (vd->vdev_children > 0) {
4269 for (int c = 0; c < vd->vdev_children; c++) {
4270 child = vd->vdev_child[c];
4273 * Don't factor holes or indirect vdevs into the
4276 if (!vdev_is_concrete(child))
4279 if (!vdev_readable(child) ||
4280 (!vdev_writeable(child) && spa_writeable(spa))) {
4282 * Root special: if there is a top-level log
4283 * device, treat the root vdev as if it were
4286 if (child->vdev_islog && vd == rvd)
4290 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4294 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4298 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4301 * Root special: if there is a top-level vdev that cannot be
4302 * opened due to corrupted metadata, then propagate the root
4303 * vdev's aux state as 'corrupt' rather than 'insufficient
4306 if (corrupted && vd == rvd &&
4307 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4308 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4309 VDEV_AUX_CORRUPT_DATA);
4312 if (vd->vdev_parent)
4313 vdev_propagate_state(vd->vdev_parent);
4317 * Set a vdev's state. If this is during an open, we don't update the parent
4318 * state, because we're in the process of opening children depth-first.
4319 * Otherwise, we propagate the change to the parent.
4321 * If this routine places a device in a faulted state, an appropriate ereport is
4325 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4327 uint64_t save_state;
4328 spa_t *spa = vd->vdev_spa;
4330 if (state == vd->vdev_state) {
4332 * Since vdev_offline() code path is already in an offline
4333 * state we can miss a statechange event to OFFLINE. Check
4334 * the previous state to catch this condition.
4336 if (vd->vdev_ops->vdev_op_leaf &&
4337 (state == VDEV_STATE_OFFLINE) &&
4338 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) {
4339 /* post an offline state change */
4340 zfs_post_state_change(spa, vd, vd->vdev_prevstate);
4342 vd->vdev_stat.vs_aux = aux;
4346 save_state = vd->vdev_state;
4348 vd->vdev_state = state;
4349 vd->vdev_stat.vs_aux = aux;
4352 * If we are setting the vdev state to anything but an open state, then
4353 * always close the underlying device unless the device has requested
4354 * a delayed close (i.e. we're about to remove or fault the device).
4355 * Otherwise, we keep accessible but invalid devices open forever.
4356 * We don't call vdev_close() itself, because that implies some extra
4357 * checks (offline, etc) that we don't want here. This is limited to
4358 * leaf devices, because otherwise closing the device will affect other
4361 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4362 vd->vdev_ops->vdev_op_leaf)
4363 vd->vdev_ops->vdev_op_close(vd);
4365 if (vd->vdev_removed &&
4366 state == VDEV_STATE_CANT_OPEN &&
4367 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4369 * If the previous state is set to VDEV_STATE_REMOVED, then this
4370 * device was previously marked removed and someone attempted to
4371 * reopen it. If this failed due to a nonexistent device, then
4372 * keep the device in the REMOVED state. We also let this be if
4373 * it is one of our special test online cases, which is only
4374 * attempting to online the device and shouldn't generate an FMA
4377 vd->vdev_state = VDEV_STATE_REMOVED;
4378 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4379 } else if (state == VDEV_STATE_REMOVED) {
4380 vd->vdev_removed = B_TRUE;
4381 } else if (state == VDEV_STATE_CANT_OPEN) {
4383 * If we fail to open a vdev during an import or recovery, we
4384 * mark it as "not available", which signifies that it was
4385 * never there to begin with. Failure to open such a device
4386 * is not considered an error.
4388 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4389 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4390 vd->vdev_ops->vdev_op_leaf)
4391 vd->vdev_not_present = 1;
4394 * Post the appropriate ereport. If the 'prevstate' field is
4395 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4396 * that this is part of a vdev_reopen(). In this case, we don't
4397 * want to post the ereport if the device was already in the
4398 * CANT_OPEN state beforehand.
4400 * If the 'checkremove' flag is set, then this is an attempt to
4401 * online the device in response to an insertion event. If we
4402 * hit this case, then we have detected an insertion event for a
4403 * faulted or offline device that wasn't in the removed state.
4404 * In this scenario, we don't post an ereport because we are
4405 * about to replace the device, or attempt an online with
4406 * vdev_forcefault, which will generate the fault for us.
4408 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4409 !vd->vdev_not_present && !vd->vdev_checkremove &&
4410 vd != spa->spa_root_vdev) {
4414 case VDEV_AUX_OPEN_FAILED:
4415 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4417 case VDEV_AUX_CORRUPT_DATA:
4418 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4420 case VDEV_AUX_NO_REPLICAS:
4421 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4423 case VDEV_AUX_BAD_GUID_SUM:
4424 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4426 case VDEV_AUX_TOO_SMALL:
4427 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4429 case VDEV_AUX_BAD_LABEL:
4430 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4432 case VDEV_AUX_BAD_ASHIFT:
4433 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT;
4436 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4439 zfs_ereport_post(class, spa, vd, NULL, NULL,
4443 /* Erase any notion of persistent removed state */
4444 vd->vdev_removed = B_FALSE;
4446 vd->vdev_removed = B_FALSE;
4450 * Notify ZED of any significant state-change on a leaf vdev.
4453 if (vd->vdev_ops->vdev_op_leaf) {
4454 /* preserve original state from a vdev_reopen() */
4455 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) &&
4456 (vd->vdev_prevstate != vd->vdev_state) &&
4457 (save_state <= VDEV_STATE_CLOSED))
4458 save_state = vd->vdev_prevstate;
4460 /* filter out state change due to initial vdev_open */
4461 if (save_state > VDEV_STATE_CLOSED)
4462 zfs_post_state_change(spa, vd, save_state);
4465 if (!isopen && vd->vdev_parent)
4466 vdev_propagate_state(vd->vdev_parent);
4470 vdev_children_are_offline(vdev_t *vd)
4472 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4474 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4475 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4483 * Check the vdev configuration to ensure that it's capable of supporting
4484 * a root pool. We do not support partial configuration.
4487 vdev_is_bootable(vdev_t *vd)
4489 if (!vd->vdev_ops->vdev_op_leaf) {
4490 const char *vdev_type = vd->vdev_ops->vdev_op_type;
4492 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4493 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4498 for (int c = 0; c < vd->vdev_children; c++) {
4499 if (!vdev_is_bootable(vd->vdev_child[c]))
4506 vdev_is_concrete(vdev_t *vd)
4508 vdev_ops_t *ops = vd->vdev_ops;
4509 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4510 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4518 * Determine if a log device has valid content. If the vdev was
4519 * removed or faulted in the MOS config then we know that
4520 * the content on the log device has already been written to the pool.
4523 vdev_log_state_valid(vdev_t *vd)
4525 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4529 for (int c = 0; c < vd->vdev_children; c++)
4530 if (vdev_log_state_valid(vd->vdev_child[c]))
4537 * Expand a vdev if possible.
4540 vdev_expand(vdev_t *vd, uint64_t txg)
4542 ASSERT(vd->vdev_top == vd);
4543 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4544 ASSERT(vdev_is_concrete(vd));
4546 vdev_set_deflate_ratio(vd);
4548 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
4549 vdev_is_concrete(vd)) {
4550 vdev_metaslab_group_create(vd);
4551 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4552 vdev_config_dirty(vd);
4560 vdev_split(vdev_t *vd)
4562 vdev_t *cvd, *pvd = vd->vdev_parent;
4564 vdev_remove_child(pvd, vd);
4565 vdev_compact_children(pvd);
4567 cvd = pvd->vdev_child[0];
4568 if (pvd->vdev_children == 1) {
4569 vdev_remove_parent(cvd);
4570 cvd->vdev_splitting = B_TRUE;
4572 vdev_propagate_state(cvd);
4576 vdev_deadman(vdev_t *vd, char *tag)
4578 for (int c = 0; c < vd->vdev_children; c++) {
4579 vdev_t *cvd = vd->vdev_child[c];
4581 vdev_deadman(cvd, tag);
4584 if (vd->vdev_ops->vdev_op_leaf) {
4585 vdev_queue_t *vq = &vd->vdev_queue;
4587 mutex_enter(&vq->vq_lock);
4588 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4589 spa_t *spa = vd->vdev_spa;
4593 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4594 vd->vdev_path, avl_numnodes(&vq->vq_active_tree));
4597 * Look at the head of all the pending queues,
4598 * if any I/O has been outstanding for longer than
4599 * the spa_deadman_synctime invoke the deadman logic.
4601 fio = avl_first(&vq->vq_active_tree);
4602 delta = gethrtime() - fio->io_timestamp;
4603 if (delta > spa_deadman_synctime(spa))
4604 zio_deadman(fio, tag);
4606 mutex_exit(&vq->vq_lock);
4611 vdev_set_deferred_resilver(spa_t *spa, vdev_t *vd)
4613 for (uint64_t i = 0; i < vd->vdev_children; i++)
4614 vdev_set_deferred_resilver(spa, vd->vdev_child[i]);
4616 if (!vd->vdev_ops->vdev_op_leaf || !vdev_writeable(vd) ||
4617 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
4621 vd->vdev_resilver_deferred = B_TRUE;
4622 spa->spa_resilver_deferred = B_TRUE;
4625 #if defined(_KERNEL)
4626 EXPORT_SYMBOL(vdev_fault);
4627 EXPORT_SYMBOL(vdev_degrade);
4628 EXPORT_SYMBOL(vdev_online);
4629 EXPORT_SYMBOL(vdev_offline);
4630 EXPORT_SYMBOL(vdev_clear);
4632 module_param(vdev_max_ms_count, int, 0644);
4633 MODULE_PARM_DESC(vdev_max_ms_count,
4634 "Target number of metaslabs per top-level vdev");
4636 module_param(vdev_min_ms_count, int, 0644);
4637 MODULE_PARM_DESC(vdev_min_ms_count,
4638 "Minimum number of metaslabs per top-level vdev");
4640 module_param(vdev_ms_count_limit, int, 0644);
4641 MODULE_PARM_DESC(vdev_ms_count_limit,
4642 "Practical upper limit of total metaslabs per top-level vdev");
4644 module_param(zfs_slow_io_events_per_second, uint, 0644);
4645 MODULE_PARM_DESC(zfs_slow_io_events_per_second,
4646 "Rate limit slow IO (delay) events to this many per second");
4648 module_param(zfs_checksum_events_per_second, uint, 0644);
4649 MODULE_PARM_DESC(zfs_checksum_events_per_second, "Rate limit checksum events "
4650 "to this many checksum errors per second (do not set below zed"
4653 module_param(zfs_scan_ignore_errors, int, 0644);
4654 MODULE_PARM_DESC(zfs_scan_ignore_errors,
4655 "Ignore errors during resilver/scrub");
4657 module_param(vdev_validate_skip, int, 0644);
4658 MODULE_PARM_DESC(vdev_validate_skip,
4659 "Bypass vdev_validate()");
4661 module_param(zfs_nocacheflush, int, 0644);
4662 MODULE_PARM_DESC(zfs_nocacheflush, "Disable cache flushes");