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 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright 2016 Toomas Soome <tsoome@me.com>
29 * Copyright 2017 Joyent, Inc.
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>
53 #include <sys/trim_map.h>
55 SYSCTL_DECL(_vfs_zfs);
56 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
59 * Virtual device management.
63 * The limit for ZFS to automatically increase a top-level vdev's ashift
64 * from logical ashift to physical ashift.
66 * Example: one or more 512B emulation child vdevs
67 * child->vdev_ashift = 9 (512 bytes)
68 * child->vdev_physical_ashift = 12 (4096 bytes)
69 * zfs_max_auto_ashift = 11 (2048 bytes)
70 * zfs_min_auto_ashift = 9 (512 bytes)
72 * On pool creation or the addition of a new top-level vdev, ZFS will
73 * increase the ashift of the top-level vdev to 2048 as limited by
74 * zfs_max_auto_ashift.
76 * Example: one or more 512B emulation child vdevs
77 * child->vdev_ashift = 9 (512 bytes)
78 * child->vdev_physical_ashift = 12 (4096 bytes)
79 * zfs_max_auto_ashift = 13 (8192 bytes)
80 * zfs_min_auto_ashift = 9 (512 bytes)
82 * On pool creation or the addition of a new top-level vdev, ZFS will
83 * increase the ashift of the top-level vdev to 4096 to match the
84 * max vdev_physical_ashift.
86 * Example: one or more 512B emulation child vdevs
87 * child->vdev_ashift = 9 (512 bytes)
88 * child->vdev_physical_ashift = 9 (512 bytes)
89 * zfs_max_auto_ashift = 13 (8192 bytes)
90 * zfs_min_auto_ashift = 12 (4096 bytes)
92 * On pool creation or the addition of a new top-level vdev, ZFS will
93 * increase the ashift of the top-level vdev to 4096 to match the
94 * zfs_min_auto_ashift.
96 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
97 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
100 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
105 val = zfs_max_auto_ashift;
106 err = sysctl_handle_64(oidp, &val, 0, req);
107 if (err != 0 || req->newptr == NULL)
110 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
113 zfs_max_auto_ashift = val;
117 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
118 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
119 sysctl_vfs_zfs_max_auto_ashift, "QU",
120 "Max ashift used when optimising for logical -> physical sectors size on "
121 "new top-level vdevs.");
124 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
129 val = zfs_min_auto_ashift;
130 err = sysctl_handle_64(oidp, &val, 0, req);
131 if (err != 0 || req->newptr == NULL)
134 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
137 zfs_min_auto_ashift = val;
141 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
142 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
143 sysctl_vfs_zfs_min_auto_ashift, "QU",
144 "Min ashift used when creating new top-level vdevs.");
146 static vdev_ops_t *vdev_ops_table[] = {
165 /* maximum number of metaslabs per top-level vdev */
166 int vdev_max_ms_count = 200;
167 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, max_ms_count, CTLFLAG_RDTUN,
168 &vdev_max_ms_count, 0,
169 "Maximum number of metaslabs per top-level vdev");
171 /* minimum amount of metaslabs per top-level vdev */
172 int vdev_min_ms_count = 16;
173 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, min_ms_count, CTLFLAG_RDTUN,
174 &vdev_min_ms_count, 0,
175 "Minimum number of metaslabs per top-level vdev");
177 /* see comment in vdev_metaslab_set_size() */
178 int vdev_default_ms_shift = 29;
179 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, default_ms_shift, CTLFLAG_RDTUN,
180 &vdev_default_ms_shift, 0,
181 "Shift between vdev size and number of metaslabs");
183 boolean_t vdev_validate_skip = B_FALSE;
186 * Since the DTL space map of a vdev is not expected to have a lot of
187 * entries, we default its block size to 4K.
189 int vdev_dtl_sm_blksz = (1 << 12);
190 SYSCTL_INT(_vfs_zfs, OID_AUTO, dtl_sm_blksz, CTLFLAG_RDTUN,
191 &vdev_dtl_sm_blksz, 0,
192 "Block size for DTL space map. Power of 2 and greater than 4096.");
195 * vdev-wide space maps that have lots of entries written to them at
196 * the end of each transaction can benefit from a higher I/O bandwidth
197 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
199 int vdev_standard_sm_blksz = (1 << 17);
200 SYSCTL_INT(_vfs_zfs, OID_AUTO, standard_sm_blksz, CTLFLAG_RDTUN,
201 &vdev_standard_sm_blksz, 0,
202 "Block size for standard space map. Power of 2 and greater than 4096.");
206 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
212 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
215 if (vd->vdev_path != NULL) {
216 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
219 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
220 vd->vdev_ops->vdev_op_type,
221 (u_longlong_t)vd->vdev_id,
222 (u_longlong_t)vd->vdev_guid, buf);
227 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
231 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
232 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
233 vd->vdev_ops->vdev_op_type);
237 switch (vd->vdev_state) {
238 case VDEV_STATE_UNKNOWN:
239 (void) snprintf(state, sizeof (state), "unknown");
241 case VDEV_STATE_CLOSED:
242 (void) snprintf(state, sizeof (state), "closed");
244 case VDEV_STATE_OFFLINE:
245 (void) snprintf(state, sizeof (state), "offline");
247 case VDEV_STATE_REMOVED:
248 (void) snprintf(state, sizeof (state), "removed");
250 case VDEV_STATE_CANT_OPEN:
251 (void) snprintf(state, sizeof (state), "can't open");
253 case VDEV_STATE_FAULTED:
254 (void) snprintf(state, sizeof (state), "faulted");
256 case VDEV_STATE_DEGRADED:
257 (void) snprintf(state, sizeof (state), "degraded");
259 case VDEV_STATE_HEALTHY:
260 (void) snprintf(state, sizeof (state), "healthy");
263 (void) snprintf(state, sizeof (state), "<state %u>",
264 (uint_t)vd->vdev_state);
267 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
268 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
269 vd->vdev_islog ? " (log)" : "",
270 (u_longlong_t)vd->vdev_guid,
271 vd->vdev_path ? vd->vdev_path : "N/A", state);
273 for (uint64_t i = 0; i < vd->vdev_children; i++)
274 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
278 * Given a vdev type, return the appropriate ops vector.
281 vdev_getops(const char *type)
283 vdev_ops_t *ops, **opspp;
285 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
286 if (strcmp(ops->vdev_op_type, type) == 0)
293 * Default asize function: return the MAX of psize with the asize of
294 * all children. This is what's used by anything other than RAID-Z.
297 vdev_default_asize(vdev_t *vd, uint64_t psize)
299 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
302 for (int c = 0; c < vd->vdev_children; c++) {
303 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
304 asize = MAX(asize, csize);
311 * Get the minimum allocatable size. We define the allocatable size as
312 * the vdev's asize rounded to the nearest metaslab. This allows us to
313 * replace or attach devices which don't have the same physical size but
314 * can still satisfy the same number of allocations.
317 vdev_get_min_asize(vdev_t *vd)
319 vdev_t *pvd = vd->vdev_parent;
322 * If our parent is NULL (inactive spare or cache) or is the root,
323 * just return our own asize.
326 return (vd->vdev_asize);
329 * The top-level vdev just returns the allocatable size rounded
330 * to the nearest metaslab.
332 if (vd == vd->vdev_top)
333 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
336 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
337 * so each child must provide at least 1/Nth of its asize.
339 if (pvd->vdev_ops == &vdev_raidz_ops)
340 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
343 return (pvd->vdev_min_asize);
347 vdev_set_min_asize(vdev_t *vd)
349 vd->vdev_min_asize = vdev_get_min_asize(vd);
351 for (int c = 0; c < vd->vdev_children; c++)
352 vdev_set_min_asize(vd->vdev_child[c]);
356 vdev_lookup_top(spa_t *spa, uint64_t vdev)
358 vdev_t *rvd = spa->spa_root_vdev;
360 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
362 if (vdev < rvd->vdev_children) {
363 ASSERT(rvd->vdev_child[vdev] != NULL);
364 return (rvd->vdev_child[vdev]);
371 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
375 if (vd->vdev_guid == guid)
378 for (int c = 0; c < vd->vdev_children; c++)
379 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
387 vdev_count_leaves_impl(vdev_t *vd)
391 if (vd->vdev_ops->vdev_op_leaf)
394 for (int c = 0; c < vd->vdev_children; c++)
395 n += vdev_count_leaves_impl(vd->vdev_child[c]);
401 vdev_count_leaves(spa_t *spa)
403 return (vdev_count_leaves_impl(spa->spa_root_vdev));
407 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
409 size_t oldsize, newsize;
410 uint64_t id = cvd->vdev_id;
412 spa_t *spa = cvd->vdev_spa;
414 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
415 ASSERT(cvd->vdev_parent == NULL);
417 cvd->vdev_parent = pvd;
422 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
424 oldsize = pvd->vdev_children * sizeof (vdev_t *);
425 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
426 newsize = pvd->vdev_children * sizeof (vdev_t *);
428 newchild = kmem_zalloc(newsize, KM_SLEEP);
429 if (pvd->vdev_child != NULL) {
430 bcopy(pvd->vdev_child, newchild, oldsize);
431 kmem_free(pvd->vdev_child, oldsize);
434 pvd->vdev_child = newchild;
435 pvd->vdev_child[id] = cvd;
437 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
438 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
441 * Walk up all ancestors to update guid sum.
443 for (; pvd != NULL; pvd = pvd->vdev_parent)
444 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
448 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
451 uint_t id = cvd->vdev_id;
453 ASSERT(cvd->vdev_parent == pvd);
458 ASSERT(id < pvd->vdev_children);
459 ASSERT(pvd->vdev_child[id] == cvd);
461 pvd->vdev_child[id] = NULL;
462 cvd->vdev_parent = NULL;
464 for (c = 0; c < pvd->vdev_children; c++)
465 if (pvd->vdev_child[c])
468 if (c == pvd->vdev_children) {
469 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
470 pvd->vdev_child = NULL;
471 pvd->vdev_children = 0;
475 * Walk up all ancestors to update guid sum.
477 for (; pvd != NULL; pvd = pvd->vdev_parent)
478 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
482 * Remove any holes in the child array.
485 vdev_compact_children(vdev_t *pvd)
487 vdev_t **newchild, *cvd;
488 int oldc = pvd->vdev_children;
491 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
493 for (int c = newc = 0; c < oldc; c++)
494 if (pvd->vdev_child[c])
497 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
499 for (int c = newc = 0; c < oldc; c++) {
500 if ((cvd = pvd->vdev_child[c]) != NULL) {
501 newchild[newc] = cvd;
502 cvd->vdev_id = newc++;
506 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
507 pvd->vdev_child = newchild;
508 pvd->vdev_children = newc;
512 * Allocate and minimally initialize a vdev_t.
515 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
518 vdev_indirect_config_t *vic;
520 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
521 vic = &vd->vdev_indirect_config;
523 if (spa->spa_root_vdev == NULL) {
524 ASSERT(ops == &vdev_root_ops);
525 spa->spa_root_vdev = vd;
526 spa->spa_load_guid = spa_generate_guid(NULL);
529 if (guid == 0 && ops != &vdev_hole_ops) {
530 if (spa->spa_root_vdev == vd) {
532 * The root vdev's guid will also be the pool guid,
533 * which must be unique among all pools.
535 guid = spa_generate_guid(NULL);
538 * Any other vdev's guid must be unique within the pool.
540 guid = spa_generate_guid(spa);
542 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
547 vd->vdev_guid = guid;
548 vd->vdev_guid_sum = guid;
550 vd->vdev_state = VDEV_STATE_CLOSED;
551 vd->vdev_ishole = (ops == &vdev_hole_ops);
552 vic->vic_prev_indirect_vdev = UINT64_MAX;
554 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
555 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
556 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
558 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
559 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
560 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
561 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
562 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
564 for (int t = 0; t < DTL_TYPES; t++) {
565 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
567 txg_list_create(&vd->vdev_ms_list, spa,
568 offsetof(struct metaslab, ms_txg_node));
569 txg_list_create(&vd->vdev_dtl_list, spa,
570 offsetof(struct vdev, vdev_dtl_node));
571 vd->vdev_stat.vs_timestamp = gethrtime();
579 * Allocate a new vdev. The 'alloctype' is used to control whether we are
580 * creating a new vdev or loading an existing one - the behavior is slightly
581 * different for each case.
584 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
589 uint64_t guid = 0, islog, nparity;
591 vdev_indirect_config_t *vic;
593 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
595 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
596 return (SET_ERROR(EINVAL));
598 if ((ops = vdev_getops(type)) == NULL)
599 return (SET_ERROR(EINVAL));
602 * If this is a load, get the vdev guid from the nvlist.
603 * Otherwise, vdev_alloc_common() will generate one for us.
605 if (alloctype == VDEV_ALLOC_LOAD) {
608 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
610 return (SET_ERROR(EINVAL));
612 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
613 return (SET_ERROR(EINVAL));
614 } else if (alloctype == VDEV_ALLOC_SPARE) {
615 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
616 return (SET_ERROR(EINVAL));
617 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
618 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
619 return (SET_ERROR(EINVAL));
620 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
621 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
622 return (SET_ERROR(EINVAL));
626 * The first allocated vdev must be of type 'root'.
628 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
629 return (SET_ERROR(EINVAL));
632 * Determine whether we're a log vdev.
635 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
636 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
637 return (SET_ERROR(ENOTSUP));
639 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
640 return (SET_ERROR(ENOTSUP));
643 * Set the nparity property for RAID-Z vdevs.
646 if (ops == &vdev_raidz_ops) {
647 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
649 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
650 return (SET_ERROR(EINVAL));
652 * Previous versions could only support 1 or 2 parity
656 spa_version(spa) < SPA_VERSION_RAIDZ2)
657 return (SET_ERROR(ENOTSUP));
659 spa_version(spa) < SPA_VERSION_RAIDZ3)
660 return (SET_ERROR(ENOTSUP));
663 * We require the parity to be specified for SPAs that
664 * support multiple parity levels.
666 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
667 return (SET_ERROR(EINVAL));
669 * Otherwise, we default to 1 parity device for RAID-Z.
676 ASSERT(nparity != -1ULL);
678 vd = vdev_alloc_common(spa, id, guid, ops);
679 vic = &vd->vdev_indirect_config;
681 vd->vdev_islog = islog;
682 vd->vdev_nparity = nparity;
684 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
685 vd->vdev_path = spa_strdup(vd->vdev_path);
686 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
687 vd->vdev_devid = spa_strdup(vd->vdev_devid);
688 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
689 &vd->vdev_physpath) == 0)
690 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
691 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
692 vd->vdev_fru = spa_strdup(vd->vdev_fru);
695 * Set the whole_disk property. If it's not specified, leave the value
698 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
699 &vd->vdev_wholedisk) != 0)
700 vd->vdev_wholedisk = -1ULL;
702 ASSERT0(vic->vic_mapping_object);
703 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
704 &vic->vic_mapping_object);
705 ASSERT0(vic->vic_births_object);
706 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
707 &vic->vic_births_object);
708 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
709 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
710 &vic->vic_prev_indirect_vdev);
713 * Look for the 'not present' flag. This will only be set if the device
714 * was not present at the time of import.
716 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
717 &vd->vdev_not_present);
720 * Get the alignment requirement.
722 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
725 * Retrieve the vdev creation time.
727 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
731 * If we're a top-level vdev, try to load the allocation parameters.
733 if (parent && !parent->vdev_parent &&
734 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
735 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
737 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
739 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
741 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
743 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
746 ASSERT0(vd->vdev_top_zap);
749 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
750 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
751 alloctype == VDEV_ALLOC_ADD ||
752 alloctype == VDEV_ALLOC_SPLIT ||
753 alloctype == VDEV_ALLOC_ROOTPOOL);
754 vd->vdev_mg = metaslab_group_create(islog ?
755 spa_log_class(spa) : spa_normal_class(spa), vd);
758 if (vd->vdev_ops->vdev_op_leaf &&
759 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
760 (void) nvlist_lookup_uint64(nv,
761 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
763 ASSERT0(vd->vdev_leaf_zap);
767 * If we're a leaf vdev, try to load the DTL object and other state.
770 if (vd->vdev_ops->vdev_op_leaf &&
771 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
772 alloctype == VDEV_ALLOC_ROOTPOOL)) {
773 if (alloctype == VDEV_ALLOC_LOAD) {
774 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
775 &vd->vdev_dtl_object);
776 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
780 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
783 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
784 &spare) == 0 && spare)
788 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
791 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
792 &vd->vdev_resilver_txg);
795 * When importing a pool, we want to ignore the persistent fault
796 * state, as the diagnosis made on another system may not be
797 * valid in the current context. Local vdevs will
798 * remain in the faulted state.
800 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
801 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
803 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
805 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
808 if (vd->vdev_faulted || vd->vdev_degraded) {
812 VDEV_AUX_ERR_EXCEEDED;
813 if (nvlist_lookup_string(nv,
814 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
815 strcmp(aux, "external") == 0)
816 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
822 * Add ourselves to the parent's list of children.
824 vdev_add_child(parent, vd);
832 vdev_free(vdev_t *vd)
834 spa_t *spa = vd->vdev_spa;
837 * Scan queues are normally destroyed at the end of a scan. If the
838 * queue exists here, that implies the vdev is being removed while
839 * the scan is still running.
841 if (vd->vdev_scan_io_queue != NULL) {
842 mutex_enter(&vd->vdev_scan_io_queue_lock);
843 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
844 vd->vdev_scan_io_queue = NULL;
845 mutex_exit(&vd->vdev_scan_io_queue_lock);
849 * vdev_free() implies closing the vdev first. This is simpler than
850 * trying to ensure complicated semantics for all callers.
854 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
855 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
860 for (int c = 0; c < vd->vdev_children; c++)
861 vdev_free(vd->vdev_child[c]);
863 ASSERT(vd->vdev_child == NULL);
864 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
867 * Discard allocation state.
869 if (vd->vdev_mg != NULL) {
870 vdev_metaslab_fini(vd);
871 metaslab_group_destroy(vd->vdev_mg);
874 ASSERT0(vd->vdev_stat.vs_space);
875 ASSERT0(vd->vdev_stat.vs_dspace);
876 ASSERT0(vd->vdev_stat.vs_alloc);
879 * Remove this vdev from its parent's child list.
881 vdev_remove_child(vd->vdev_parent, vd);
883 ASSERT(vd->vdev_parent == NULL);
886 * Clean up vdev structure.
892 spa_strfree(vd->vdev_path);
894 spa_strfree(vd->vdev_devid);
895 if (vd->vdev_physpath)
896 spa_strfree(vd->vdev_physpath);
898 spa_strfree(vd->vdev_fru);
900 if (vd->vdev_isspare)
901 spa_spare_remove(vd);
902 if (vd->vdev_isl2cache)
903 spa_l2cache_remove(vd);
905 txg_list_destroy(&vd->vdev_ms_list);
906 txg_list_destroy(&vd->vdev_dtl_list);
908 mutex_enter(&vd->vdev_dtl_lock);
909 space_map_close(vd->vdev_dtl_sm);
910 for (int t = 0; t < DTL_TYPES; t++) {
911 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
912 range_tree_destroy(vd->vdev_dtl[t]);
914 mutex_exit(&vd->vdev_dtl_lock);
916 EQUIV(vd->vdev_indirect_births != NULL,
917 vd->vdev_indirect_mapping != NULL);
918 if (vd->vdev_indirect_births != NULL) {
919 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
920 vdev_indirect_births_close(vd->vdev_indirect_births);
923 if (vd->vdev_obsolete_sm != NULL) {
924 ASSERT(vd->vdev_removing ||
925 vd->vdev_ops == &vdev_indirect_ops);
926 space_map_close(vd->vdev_obsolete_sm);
927 vd->vdev_obsolete_sm = NULL;
929 range_tree_destroy(vd->vdev_obsolete_segments);
930 rw_destroy(&vd->vdev_indirect_rwlock);
931 mutex_destroy(&vd->vdev_obsolete_lock);
933 mutex_destroy(&vd->vdev_queue_lock);
934 mutex_destroy(&vd->vdev_dtl_lock);
935 mutex_destroy(&vd->vdev_stat_lock);
936 mutex_destroy(&vd->vdev_probe_lock);
937 mutex_destroy(&vd->vdev_scan_io_queue_lock);
939 if (vd == spa->spa_root_vdev)
940 spa->spa_root_vdev = NULL;
942 kmem_free(vd, sizeof (vdev_t));
946 * Transfer top-level vdev state from svd to tvd.
949 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
951 spa_t *spa = svd->vdev_spa;
956 ASSERT(tvd == tvd->vdev_top);
958 tvd->vdev_ms_array = svd->vdev_ms_array;
959 tvd->vdev_ms_shift = svd->vdev_ms_shift;
960 tvd->vdev_ms_count = svd->vdev_ms_count;
961 tvd->vdev_top_zap = svd->vdev_top_zap;
963 svd->vdev_ms_array = 0;
964 svd->vdev_ms_shift = 0;
965 svd->vdev_ms_count = 0;
966 svd->vdev_top_zap = 0;
969 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
970 tvd->vdev_mg = svd->vdev_mg;
971 tvd->vdev_ms = svd->vdev_ms;
976 if (tvd->vdev_mg != NULL)
977 tvd->vdev_mg->mg_vd = tvd;
979 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
980 svd->vdev_checkpoint_sm = NULL;
982 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
983 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
984 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
986 svd->vdev_stat.vs_alloc = 0;
987 svd->vdev_stat.vs_space = 0;
988 svd->vdev_stat.vs_dspace = 0;
990 for (t = 0; t < TXG_SIZE; t++) {
991 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
992 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
993 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
994 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
995 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
996 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
999 if (list_link_active(&svd->vdev_config_dirty_node)) {
1000 vdev_config_clean(svd);
1001 vdev_config_dirty(tvd);
1004 if (list_link_active(&svd->vdev_state_dirty_node)) {
1005 vdev_state_clean(svd);
1006 vdev_state_dirty(tvd);
1009 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1010 svd->vdev_deflate_ratio = 0;
1012 tvd->vdev_islog = svd->vdev_islog;
1013 svd->vdev_islog = 0;
1015 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1019 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1026 for (int c = 0; c < vd->vdev_children; c++)
1027 vdev_top_update(tvd, vd->vdev_child[c]);
1031 * Add a mirror/replacing vdev above an existing vdev.
1034 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1036 spa_t *spa = cvd->vdev_spa;
1037 vdev_t *pvd = cvd->vdev_parent;
1040 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1042 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1044 mvd->vdev_asize = cvd->vdev_asize;
1045 mvd->vdev_min_asize = cvd->vdev_min_asize;
1046 mvd->vdev_max_asize = cvd->vdev_max_asize;
1047 mvd->vdev_psize = cvd->vdev_psize;
1048 mvd->vdev_ashift = cvd->vdev_ashift;
1049 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1050 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1051 mvd->vdev_state = cvd->vdev_state;
1052 mvd->vdev_crtxg = cvd->vdev_crtxg;
1054 vdev_remove_child(pvd, cvd);
1055 vdev_add_child(pvd, mvd);
1056 cvd->vdev_id = mvd->vdev_children;
1057 vdev_add_child(mvd, cvd);
1058 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1060 if (mvd == mvd->vdev_top)
1061 vdev_top_transfer(cvd, mvd);
1067 * Remove a 1-way mirror/replacing vdev from the tree.
1070 vdev_remove_parent(vdev_t *cvd)
1072 vdev_t *mvd = cvd->vdev_parent;
1073 vdev_t *pvd = mvd->vdev_parent;
1075 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1077 ASSERT(mvd->vdev_children == 1);
1078 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1079 mvd->vdev_ops == &vdev_replacing_ops ||
1080 mvd->vdev_ops == &vdev_spare_ops);
1081 cvd->vdev_ashift = mvd->vdev_ashift;
1082 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1083 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1085 vdev_remove_child(mvd, cvd);
1086 vdev_remove_child(pvd, mvd);
1089 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1090 * Otherwise, we could have detached an offline device, and when we
1091 * go to import the pool we'll think we have two top-level vdevs,
1092 * instead of a different version of the same top-level vdev.
1094 if (mvd->vdev_top == mvd) {
1095 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1096 cvd->vdev_orig_guid = cvd->vdev_guid;
1097 cvd->vdev_guid += guid_delta;
1098 cvd->vdev_guid_sum += guid_delta;
1100 cvd->vdev_id = mvd->vdev_id;
1101 vdev_add_child(pvd, cvd);
1102 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1104 if (cvd == cvd->vdev_top)
1105 vdev_top_transfer(mvd, cvd);
1107 ASSERT(mvd->vdev_children == 0);
1112 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1114 spa_t *spa = vd->vdev_spa;
1115 objset_t *mos = spa->spa_meta_objset;
1117 uint64_t oldc = vd->vdev_ms_count;
1118 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1122 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1125 * This vdev is not being allocated from yet or is a hole.
1127 if (vd->vdev_ms_shift == 0)
1130 ASSERT(!vd->vdev_ishole);
1132 ASSERT(oldc <= newc);
1134 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1137 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1138 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1142 vd->vdev_ms_count = newc;
1144 for (m = oldc; m < newc; m++) {
1145 uint64_t object = 0;
1148 * vdev_ms_array may be 0 if we are creating the "fake"
1149 * metaslabs for an indirect vdev for zdb's leak detection.
1150 * See zdb_leak_init().
1152 if (txg == 0 && vd->vdev_ms_array != 0) {
1153 error = dmu_read(mos, vd->vdev_ms_array,
1154 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1157 vdev_dbgmsg(vd, "unable to read the metaslab "
1158 "array [error=%d]", error);
1163 error = metaslab_init(vd->vdev_mg, m, object, txg,
1166 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1173 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1176 * If the vdev is being removed we don't activate
1177 * the metaslabs since we want to ensure that no new
1178 * allocations are performed on this device.
1180 if (oldc == 0 && !vd->vdev_removing)
1181 metaslab_group_activate(vd->vdev_mg);
1184 spa_config_exit(spa, SCL_ALLOC, FTAG);
1190 vdev_metaslab_fini(vdev_t *vd)
1192 if (vd->vdev_checkpoint_sm != NULL) {
1193 ASSERT(spa_feature_is_active(vd->vdev_spa,
1194 SPA_FEATURE_POOL_CHECKPOINT));
1195 space_map_close(vd->vdev_checkpoint_sm);
1197 * Even though we close the space map, we need to set its
1198 * pointer to NULL. The reason is that vdev_metaslab_fini()
1199 * may be called multiple times for certain operations
1200 * (i.e. when destroying a pool) so we need to ensure that
1201 * this clause never executes twice. This logic is similar
1202 * to the one used for the vdev_ms clause below.
1204 vd->vdev_checkpoint_sm = NULL;
1207 if (vd->vdev_ms != NULL) {
1208 uint64_t count = vd->vdev_ms_count;
1210 metaslab_group_passivate(vd->vdev_mg);
1211 for (uint64_t m = 0; m < count; m++) {
1212 metaslab_t *msp = vd->vdev_ms[m];
1217 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1220 vd->vdev_ms_count = 0;
1222 ASSERT0(vd->vdev_ms_count);
1225 typedef struct vdev_probe_stats {
1226 boolean_t vps_readable;
1227 boolean_t vps_writeable;
1229 } vdev_probe_stats_t;
1232 vdev_probe_done(zio_t *zio)
1234 spa_t *spa = zio->io_spa;
1235 vdev_t *vd = zio->io_vd;
1236 vdev_probe_stats_t *vps = zio->io_private;
1238 ASSERT(vd->vdev_probe_zio != NULL);
1240 if (zio->io_type == ZIO_TYPE_READ) {
1241 if (zio->io_error == 0)
1242 vps->vps_readable = 1;
1243 if (zio->io_error == 0 && spa_writeable(spa)) {
1244 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1245 zio->io_offset, zio->io_size, zio->io_abd,
1246 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1247 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1249 abd_free(zio->io_abd);
1251 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1252 if (zio->io_error == 0)
1253 vps->vps_writeable = 1;
1254 abd_free(zio->io_abd);
1255 } else if (zio->io_type == ZIO_TYPE_NULL) {
1258 vd->vdev_cant_read |= !vps->vps_readable;
1259 vd->vdev_cant_write |= !vps->vps_writeable;
1261 if (vdev_readable(vd) &&
1262 (vdev_writeable(vd) || !spa_writeable(spa))) {
1265 ASSERT(zio->io_error != 0);
1266 vdev_dbgmsg(vd, "failed probe");
1267 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1268 spa, vd, NULL, 0, 0);
1269 zio->io_error = SET_ERROR(ENXIO);
1272 mutex_enter(&vd->vdev_probe_lock);
1273 ASSERT(vd->vdev_probe_zio == zio);
1274 vd->vdev_probe_zio = NULL;
1275 mutex_exit(&vd->vdev_probe_lock);
1277 zio_link_t *zl = NULL;
1278 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1279 if (!vdev_accessible(vd, pio))
1280 pio->io_error = SET_ERROR(ENXIO);
1282 kmem_free(vps, sizeof (*vps));
1287 * Determine whether this device is accessible.
1289 * Read and write to several known locations: the pad regions of each
1290 * vdev label but the first, which we leave alone in case it contains
1294 vdev_probe(vdev_t *vd, zio_t *zio)
1296 spa_t *spa = vd->vdev_spa;
1297 vdev_probe_stats_t *vps = NULL;
1300 ASSERT(vd->vdev_ops->vdev_op_leaf);
1303 * Don't probe the probe.
1305 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1309 * To prevent 'probe storms' when a device fails, we create
1310 * just one probe i/o at a time. All zios that want to probe
1311 * this vdev will become parents of the probe io.
1313 mutex_enter(&vd->vdev_probe_lock);
1315 if ((pio = vd->vdev_probe_zio) == NULL) {
1316 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1318 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1319 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1322 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1324 * vdev_cant_read and vdev_cant_write can only
1325 * transition from TRUE to FALSE when we have the
1326 * SCL_ZIO lock as writer; otherwise they can only
1327 * transition from FALSE to TRUE. This ensures that
1328 * any zio looking at these values can assume that
1329 * failures persist for the life of the I/O. That's
1330 * important because when a device has intermittent
1331 * connectivity problems, we want to ensure that
1332 * they're ascribed to the device (ENXIO) and not
1335 * Since we hold SCL_ZIO as writer here, clear both
1336 * values so the probe can reevaluate from first
1339 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1340 vd->vdev_cant_read = B_FALSE;
1341 vd->vdev_cant_write = B_FALSE;
1344 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1345 vdev_probe_done, vps,
1346 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1349 * We can't change the vdev state in this context, so we
1350 * kick off an async task to do it on our behalf.
1353 vd->vdev_probe_wanted = B_TRUE;
1354 spa_async_request(spa, SPA_ASYNC_PROBE);
1359 zio_add_child(zio, pio);
1361 mutex_exit(&vd->vdev_probe_lock);
1364 ASSERT(zio != NULL);
1368 for (int l = 1; l < VDEV_LABELS; l++) {
1369 zio_nowait(zio_read_phys(pio, vd,
1370 vdev_label_offset(vd->vdev_psize, l,
1371 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1372 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1373 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1374 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1385 vdev_open_child(void *arg)
1389 vd->vdev_open_thread = curthread;
1390 vd->vdev_open_error = vdev_open(vd);
1391 vd->vdev_open_thread = NULL;
1395 vdev_uses_zvols(vdev_t *vd)
1397 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1398 strlen(ZVOL_DIR)) == 0)
1400 for (int c = 0; c < vd->vdev_children; c++)
1401 if (vdev_uses_zvols(vd->vdev_child[c]))
1407 vdev_open_children(vdev_t *vd)
1410 int children = vd->vdev_children;
1413 * in order to handle pools on top of zvols, do the opens
1414 * in a single thread so that the same thread holds the
1415 * spa_namespace_lock
1417 if (B_TRUE || vdev_uses_zvols(vd)) {
1418 for (int c = 0; c < children; c++)
1419 vd->vdev_child[c]->vdev_open_error =
1420 vdev_open(vd->vdev_child[c]);
1423 tq = taskq_create("vdev_open", children, minclsyspri,
1424 children, children, TASKQ_PREPOPULATE);
1426 for (int c = 0; c < children; c++)
1427 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1434 * Compute the raidz-deflation ratio. Note, we hard-code
1435 * in 128k (1 << 17) because it is the "typical" blocksize.
1436 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1437 * otherwise it would inconsistently account for existing bp's.
1440 vdev_set_deflate_ratio(vdev_t *vd)
1442 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1443 vd->vdev_deflate_ratio = (1 << 17) /
1444 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1449 * Prepare a virtual device for access.
1452 vdev_open(vdev_t *vd)
1454 spa_t *spa = vd->vdev_spa;
1457 uint64_t max_osize = 0;
1458 uint64_t asize, max_asize, psize;
1459 uint64_t logical_ashift = 0;
1460 uint64_t physical_ashift = 0;
1462 ASSERT(vd->vdev_open_thread == curthread ||
1463 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1464 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1465 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1466 vd->vdev_state == VDEV_STATE_OFFLINE);
1468 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1469 vd->vdev_cant_read = B_FALSE;
1470 vd->vdev_cant_write = B_FALSE;
1471 vd->vdev_notrim = B_FALSE;
1472 vd->vdev_min_asize = vdev_get_min_asize(vd);
1475 * If this vdev is not removed, check its fault status. If it's
1476 * faulted, bail out of the open.
1478 if (!vd->vdev_removed && vd->vdev_faulted) {
1479 ASSERT(vd->vdev_children == 0);
1480 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1481 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1482 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1483 vd->vdev_label_aux);
1484 return (SET_ERROR(ENXIO));
1485 } else if (vd->vdev_offline) {
1486 ASSERT(vd->vdev_children == 0);
1487 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1488 return (SET_ERROR(ENXIO));
1491 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1492 &logical_ashift, &physical_ashift);
1495 * Reset the vdev_reopening flag so that we actually close
1496 * the vdev on error.
1498 vd->vdev_reopening = B_FALSE;
1499 if (zio_injection_enabled && error == 0)
1500 error = zio_handle_device_injection(vd, NULL, ENXIO);
1503 if (vd->vdev_removed &&
1504 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1505 vd->vdev_removed = B_FALSE;
1507 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1508 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1509 vd->vdev_stat.vs_aux);
1511 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1512 vd->vdev_stat.vs_aux);
1517 vd->vdev_removed = B_FALSE;
1520 * Recheck the faulted flag now that we have confirmed that
1521 * the vdev is accessible. If we're faulted, bail.
1523 if (vd->vdev_faulted) {
1524 ASSERT(vd->vdev_children == 0);
1525 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1526 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1527 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1528 vd->vdev_label_aux);
1529 return (SET_ERROR(ENXIO));
1532 if (vd->vdev_degraded) {
1533 ASSERT(vd->vdev_children == 0);
1534 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1535 VDEV_AUX_ERR_EXCEEDED);
1537 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1541 * For hole or missing vdevs we just return success.
1543 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1546 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1547 trim_map_create(vd);
1549 for (int c = 0; c < vd->vdev_children; c++) {
1550 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1551 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1557 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1558 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1560 if (vd->vdev_children == 0) {
1561 if (osize < SPA_MINDEVSIZE) {
1562 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1563 VDEV_AUX_TOO_SMALL);
1564 return (SET_ERROR(EOVERFLOW));
1567 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1568 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1569 VDEV_LABEL_END_SIZE);
1571 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1572 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1573 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1574 VDEV_AUX_TOO_SMALL);
1575 return (SET_ERROR(EOVERFLOW));
1579 max_asize = max_osize;
1582 vd->vdev_psize = psize;
1585 * Make sure the allocatable size hasn't shrunk too much.
1587 if (asize < vd->vdev_min_asize) {
1588 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1589 VDEV_AUX_BAD_LABEL);
1590 return (SET_ERROR(EINVAL));
1593 vd->vdev_physical_ashift =
1594 MAX(physical_ashift, vd->vdev_physical_ashift);
1595 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1596 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1598 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1599 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1600 VDEV_AUX_ASHIFT_TOO_BIG);
1604 if (vd->vdev_asize == 0) {
1606 * This is the first-ever open, so use the computed values.
1607 * For testing purposes, a higher ashift can be requested.
1609 vd->vdev_asize = asize;
1610 vd->vdev_max_asize = max_asize;
1613 * Make sure the alignment requirement hasn't increased.
1615 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1616 vd->vdev_ops->vdev_op_leaf) {
1617 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1618 VDEV_AUX_BAD_LABEL);
1621 vd->vdev_max_asize = max_asize;
1625 * If all children are healthy we update asize if either:
1626 * The asize has increased, due to a device expansion caused by dynamic
1627 * LUN growth or vdev replacement, and automatic expansion is enabled;
1628 * making the additional space available.
1630 * The asize has decreased, due to a device shrink usually caused by a
1631 * vdev replace with a smaller device. This ensures that calculations
1632 * based of max_asize and asize e.g. esize are always valid. It's safe
1633 * to do this as we've already validated that asize is greater than
1636 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1637 ((asize > vd->vdev_asize &&
1638 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1639 (asize < vd->vdev_asize)))
1640 vd->vdev_asize = asize;
1642 vdev_set_min_asize(vd);
1645 * Ensure we can issue some IO before declaring the
1646 * vdev open for business.
1648 if (vd->vdev_ops->vdev_op_leaf &&
1649 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1650 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1651 VDEV_AUX_ERR_EXCEEDED);
1656 * Track the min and max ashift values for normal data devices.
1658 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1659 !vd->vdev_islog && vd->vdev_aux == NULL) {
1660 if (vd->vdev_ashift > spa->spa_max_ashift)
1661 spa->spa_max_ashift = vd->vdev_ashift;
1662 if (vd->vdev_ashift < spa->spa_min_ashift)
1663 spa->spa_min_ashift = vd->vdev_ashift;
1667 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1668 * resilver. But don't do this if we are doing a reopen for a scrub,
1669 * since this would just restart the scrub we are already doing.
1671 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1672 vdev_resilver_needed(vd, NULL, NULL))
1673 spa_async_request(spa, SPA_ASYNC_RESILVER);
1679 * Called once the vdevs are all opened, this routine validates the label
1680 * contents. This needs to be done before vdev_load() so that we don't
1681 * inadvertently do repair I/Os to the wrong device.
1683 * This function will only return failure if one of the vdevs indicates that it
1684 * has since been destroyed or exported. This is only possible if
1685 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1686 * will be updated but the function will return 0.
1689 vdev_validate(vdev_t *vd)
1691 spa_t *spa = vd->vdev_spa;
1693 uint64_t guid = 0, aux_guid = 0, top_guid;
1698 if (vdev_validate_skip)
1701 for (uint64_t c = 0; c < vd->vdev_children; c++)
1702 if (vdev_validate(vd->vdev_child[c]) != 0)
1703 return (SET_ERROR(EBADF));
1706 * If the device has already failed, or was marked offline, don't do
1707 * any further validation. Otherwise, label I/O will fail and we will
1708 * overwrite the previous state.
1710 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1714 * If we are performing an extreme rewind, we allow for a label that
1715 * was modified at a point after the current txg.
1716 * If config lock is not held do not check for the txg. spa_sync could
1717 * be updating the vdev's label before updating spa_last_synced_txg.
1719 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1720 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1723 txg = spa_last_synced_txg(spa);
1725 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1726 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1727 VDEV_AUX_BAD_LABEL);
1728 vdev_dbgmsg(vd, "vdev_validate: failed reading config");
1733 * Determine if this vdev has been split off into another
1734 * pool. If so, then refuse to open it.
1736 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1737 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1738 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1739 VDEV_AUX_SPLIT_POOL);
1741 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1745 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1746 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1747 VDEV_AUX_CORRUPT_DATA);
1749 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1750 ZPOOL_CONFIG_POOL_GUID);
1755 * If config is not trusted then ignore the spa guid check. This is
1756 * necessary because if the machine crashed during a re-guid the new
1757 * guid might have been written to all of the vdev labels, but not the
1758 * cached config. The check will be performed again once we have the
1759 * trusted config from the MOS.
1761 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1762 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1763 VDEV_AUX_CORRUPT_DATA);
1765 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
1766 "match config (%llu != %llu)", (u_longlong_t)guid,
1767 (u_longlong_t)spa_guid(spa));
1771 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1772 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1776 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
1777 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1778 VDEV_AUX_CORRUPT_DATA);
1780 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1785 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
1787 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1788 VDEV_AUX_CORRUPT_DATA);
1790 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1791 ZPOOL_CONFIG_TOP_GUID);
1796 * If this vdev just became a top-level vdev because its sibling was
1797 * detached, it will have adopted the parent's vdev guid -- but the
1798 * label may or may not be on disk yet. Fortunately, either version
1799 * of the label will have the same top guid, so if we're a top-level
1800 * vdev, we can safely compare to that instead.
1801 * However, if the config comes from a cachefile that failed to update
1802 * after the detach, a top-level vdev will appear as a non top-level
1803 * vdev in the config. Also relax the constraints if we perform an
1806 * If we split this vdev off instead, then we also check the
1807 * original pool's guid. We don't want to consider the vdev
1808 * corrupt if it is partway through a split operation.
1810 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
1811 boolean_t mismatch = B_FALSE;
1812 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
1813 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
1816 if (vd->vdev_guid != top_guid &&
1817 vd->vdev_top->vdev_guid != guid)
1822 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1823 VDEV_AUX_CORRUPT_DATA);
1825 vdev_dbgmsg(vd, "vdev_validate: config guid "
1826 "doesn't match label guid");
1827 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
1828 (u_longlong_t)vd->vdev_guid,
1829 (u_longlong_t)vd->vdev_top->vdev_guid);
1830 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
1831 "aux_guid %llu", (u_longlong_t)guid,
1832 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
1837 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1839 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1840 VDEV_AUX_CORRUPT_DATA);
1842 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1843 ZPOOL_CONFIG_POOL_STATE);
1850 * If this is a verbatim import, no need to check the
1851 * state of the pool.
1853 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1854 spa_load_state(spa) == SPA_LOAD_OPEN &&
1855 state != POOL_STATE_ACTIVE) {
1856 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
1857 "for spa %s", (u_longlong_t)state, spa->spa_name);
1858 return (SET_ERROR(EBADF));
1862 * If we were able to open and validate a vdev that was
1863 * previously marked permanently unavailable, clear that state
1866 if (vd->vdev_not_present)
1867 vd->vdev_not_present = 0;
1873 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
1875 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
1876 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
1877 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1878 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
1879 dvd->vdev_path, svd->vdev_path);
1880 spa_strfree(dvd->vdev_path);
1881 dvd->vdev_path = spa_strdup(svd->vdev_path);
1883 } else if (svd->vdev_path != NULL) {
1884 dvd->vdev_path = spa_strdup(svd->vdev_path);
1885 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1886 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
1891 * Recursively copy vdev paths from one vdev to another. Source and destination
1892 * vdev trees must have same geometry otherwise return error. Intended to copy
1893 * paths from userland config into MOS config.
1896 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
1898 if ((svd->vdev_ops == &vdev_missing_ops) ||
1899 (svd->vdev_ishole && dvd->vdev_ishole) ||
1900 (dvd->vdev_ops == &vdev_indirect_ops))
1903 if (svd->vdev_ops != dvd->vdev_ops) {
1904 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
1905 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
1906 return (SET_ERROR(EINVAL));
1909 if (svd->vdev_guid != dvd->vdev_guid) {
1910 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
1911 "%llu)", (u_longlong_t)svd->vdev_guid,
1912 (u_longlong_t)dvd->vdev_guid);
1913 return (SET_ERROR(EINVAL));
1916 if (svd->vdev_children != dvd->vdev_children) {
1917 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
1918 "%llu != %llu", (u_longlong_t)svd->vdev_children,
1919 (u_longlong_t)dvd->vdev_children);
1920 return (SET_ERROR(EINVAL));
1923 for (uint64_t i = 0; i < svd->vdev_children; i++) {
1924 int error = vdev_copy_path_strict(svd->vdev_child[i],
1925 dvd->vdev_child[i]);
1930 if (svd->vdev_ops->vdev_op_leaf)
1931 vdev_copy_path_impl(svd, dvd);
1937 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
1939 ASSERT(stvd->vdev_top == stvd);
1940 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
1942 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
1943 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
1946 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
1950 * The idea here is that while a vdev can shift positions within
1951 * a top vdev (when replacing, attaching mirror, etc.) it cannot
1952 * step outside of it.
1954 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
1956 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
1959 ASSERT(vd->vdev_ops->vdev_op_leaf);
1961 vdev_copy_path_impl(vd, dvd);
1965 * Recursively copy vdev paths from one root vdev to another. Source and
1966 * destination vdev trees may differ in geometry. For each destination leaf
1967 * vdev, search a vdev with the same guid and top vdev id in the source.
1968 * Intended to copy paths from userland config into MOS config.
1971 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
1973 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
1974 ASSERT(srvd->vdev_ops == &vdev_root_ops);
1975 ASSERT(drvd->vdev_ops == &vdev_root_ops);
1977 for (uint64_t i = 0; i < children; i++) {
1978 vdev_copy_path_search(srvd->vdev_child[i],
1979 drvd->vdev_child[i]);
1984 * Close a virtual device.
1987 vdev_close(vdev_t *vd)
1989 spa_t *spa = vd->vdev_spa;
1990 vdev_t *pvd = vd->vdev_parent;
1992 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1995 * If our parent is reopening, then we are as well, unless we are
1998 if (pvd != NULL && pvd->vdev_reopening)
1999 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2001 vd->vdev_ops->vdev_op_close(vd);
2003 vdev_cache_purge(vd);
2005 if (vd->vdev_ops->vdev_op_leaf)
2006 trim_map_destroy(vd);
2009 * We record the previous state before we close it, so that if we are
2010 * doing a reopen(), we don't generate FMA ereports if we notice that
2011 * it's still faulted.
2013 vd->vdev_prevstate = vd->vdev_state;
2015 if (vd->vdev_offline)
2016 vd->vdev_state = VDEV_STATE_OFFLINE;
2018 vd->vdev_state = VDEV_STATE_CLOSED;
2019 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2023 vdev_hold(vdev_t *vd)
2025 spa_t *spa = vd->vdev_spa;
2027 ASSERT(spa_is_root(spa));
2028 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2031 for (int c = 0; c < vd->vdev_children; c++)
2032 vdev_hold(vd->vdev_child[c]);
2034 if (vd->vdev_ops->vdev_op_leaf)
2035 vd->vdev_ops->vdev_op_hold(vd);
2039 vdev_rele(vdev_t *vd)
2041 spa_t *spa = vd->vdev_spa;
2043 ASSERT(spa_is_root(spa));
2044 for (int c = 0; c < vd->vdev_children; c++)
2045 vdev_rele(vd->vdev_child[c]);
2047 if (vd->vdev_ops->vdev_op_leaf)
2048 vd->vdev_ops->vdev_op_rele(vd);
2052 * Reopen all interior vdevs and any unopened leaves. We don't actually
2053 * reopen leaf vdevs which had previously been opened as they might deadlock
2054 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2055 * If the leaf has never been opened then open it, as usual.
2058 vdev_reopen(vdev_t *vd)
2060 spa_t *spa = vd->vdev_spa;
2062 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2064 /* set the reopening flag unless we're taking the vdev offline */
2065 vd->vdev_reopening = !vd->vdev_offline;
2067 (void) vdev_open(vd);
2070 * Call vdev_validate() here to make sure we have the same device.
2071 * Otherwise, a device with an invalid label could be successfully
2072 * opened in response to vdev_reopen().
2075 (void) vdev_validate_aux(vd);
2076 if (vdev_readable(vd) && vdev_writeable(vd) &&
2077 vd->vdev_aux == &spa->spa_l2cache &&
2078 !l2arc_vdev_present(vd))
2079 l2arc_add_vdev(spa, vd);
2081 (void) vdev_validate(vd);
2085 * Reassess parent vdev's health.
2087 vdev_propagate_state(vd);
2091 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2096 * Normally, partial opens (e.g. of a mirror) are allowed.
2097 * For a create, however, we want to fail the request if
2098 * there are any components we can't open.
2100 error = vdev_open(vd);
2102 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2104 return (error ? error : ENXIO);
2108 * Recursively load DTLs and initialize all labels.
2110 if ((error = vdev_dtl_load(vd)) != 0 ||
2111 (error = vdev_label_init(vd, txg, isreplacing ?
2112 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2121 vdev_metaslab_set_size(vdev_t *vd)
2123 uint64_t asize = vd->vdev_asize;
2124 uint64_t ms_shift = 0;
2127 * For vdevs that are bigger than 8G the metaslab size varies in
2128 * a way that the number of metaslabs increases in powers of two,
2129 * linearly in terms of vdev_asize, starting from 16 metaslabs.
2130 * So for vdev_asize of 8G we get 16 metaslabs, for 16G, we get 32,
2131 * and so on, until we hit the maximum metaslab count limit
2132 * [vdev_max_ms_count] from which point the metaslab count stays
2135 ms_shift = vdev_default_ms_shift;
2137 if ((asize >> ms_shift) < vdev_min_ms_count) {
2139 * For devices that are less than 8G we want to have
2140 * exactly 16 metaslabs. We don't want less as integer
2141 * division rounds down, so less metaslabs mean more
2142 * wasted space. We don't want more as these vdevs are
2143 * small and in the likely event that we are running
2144 * out of space, the SPA will have a hard time finding
2145 * space due to fragmentation.
2147 ms_shift = highbit64(asize / vdev_min_ms_count);
2148 ms_shift = MAX(ms_shift, SPA_MAXBLOCKSHIFT);
2150 } else if ((asize >> ms_shift) > vdev_max_ms_count) {
2151 ms_shift = highbit64(asize / vdev_max_ms_count);
2154 vd->vdev_ms_shift = ms_shift;
2155 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2159 * Maximize performance by inflating the configured ashift for top level
2160 * vdevs to be as close to the physical ashift as possible while maintaining
2161 * administrator defined limits and ensuring it doesn't go below the
2165 vdev_ashift_optimize(vdev_t *vd)
2167 if (vd == vd->vdev_top) {
2168 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
2169 vd->vdev_ashift = MIN(
2170 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
2171 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
2174 * Unusual case where logical ashift > physical ashift
2175 * so we can't cap the calculated ashift based on max
2176 * ashift as that would cause failures.
2177 * We still check if we need to increase it to match
2180 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
2187 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2189 ASSERT(vd == vd->vdev_top);
2190 /* indirect vdevs don't have metaslabs or dtls */
2191 ASSERT(vdev_is_concrete(vd) || flags == 0);
2192 ASSERT(ISP2(flags));
2193 ASSERT(spa_writeable(vd->vdev_spa));
2195 if (flags & VDD_METASLAB)
2196 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2198 if (flags & VDD_DTL)
2199 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2201 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2205 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2207 for (int c = 0; c < vd->vdev_children; c++)
2208 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2210 if (vd->vdev_ops->vdev_op_leaf)
2211 vdev_dirty(vd->vdev_top, flags, vd, txg);
2217 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2218 * the vdev has less than perfect replication. There are four kinds of DTL:
2220 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2222 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2224 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2225 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2226 * txgs that was scrubbed.
2228 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2229 * persistent errors or just some device being offline.
2230 * Unlike the other three, the DTL_OUTAGE map is not generally
2231 * maintained; it's only computed when needed, typically to
2232 * determine whether a device can be detached.
2234 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2235 * either has the data or it doesn't.
2237 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2238 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2239 * if any child is less than fully replicated, then so is its parent.
2240 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2241 * comprising only those txgs which appear in 'maxfaults' or more children;
2242 * those are the txgs we don't have enough replication to read. For example,
2243 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2244 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2245 * two child DTL_MISSING maps.
2247 * It should be clear from the above that to compute the DTLs and outage maps
2248 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2249 * Therefore, that is all we keep on disk. When loading the pool, or after
2250 * a configuration change, we generate all other DTLs from first principles.
2253 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2255 range_tree_t *rt = vd->vdev_dtl[t];
2257 ASSERT(t < DTL_TYPES);
2258 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2259 ASSERT(spa_writeable(vd->vdev_spa));
2261 mutex_enter(&vd->vdev_dtl_lock);
2262 if (!range_tree_contains(rt, txg, size))
2263 range_tree_add(rt, txg, size);
2264 mutex_exit(&vd->vdev_dtl_lock);
2268 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2270 range_tree_t *rt = vd->vdev_dtl[t];
2271 boolean_t dirty = B_FALSE;
2273 ASSERT(t < DTL_TYPES);
2274 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2277 * While we are loading the pool, the DTLs have not been loaded yet.
2278 * Ignore the DTLs and try all devices. This avoids a recursive
2279 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2280 * when loading the pool (relying on the checksum to ensure that
2281 * we get the right data -- note that we while loading, we are
2282 * only reading the MOS, which is always checksummed).
2284 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2287 mutex_enter(&vd->vdev_dtl_lock);
2288 if (!range_tree_is_empty(rt))
2289 dirty = range_tree_contains(rt, txg, size);
2290 mutex_exit(&vd->vdev_dtl_lock);
2296 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2298 range_tree_t *rt = vd->vdev_dtl[t];
2301 mutex_enter(&vd->vdev_dtl_lock);
2302 empty = range_tree_is_empty(rt);
2303 mutex_exit(&vd->vdev_dtl_lock);
2309 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2312 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2314 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2316 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2317 vd->vdev_ops->vdev_op_leaf)
2320 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2324 * Returns the lowest txg in the DTL range.
2327 vdev_dtl_min(vdev_t *vd)
2331 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2332 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2333 ASSERT0(vd->vdev_children);
2335 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2336 return (rs->rs_start - 1);
2340 * Returns the highest txg in the DTL.
2343 vdev_dtl_max(vdev_t *vd)
2347 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2348 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2349 ASSERT0(vd->vdev_children);
2351 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2352 return (rs->rs_end);
2356 * Determine if a resilvering vdev should remove any DTL entries from
2357 * its range. If the vdev was resilvering for the entire duration of the
2358 * scan then it should excise that range from its DTLs. Otherwise, this
2359 * vdev is considered partially resilvered and should leave its DTL
2360 * entries intact. The comment in vdev_dtl_reassess() describes how we
2364 vdev_dtl_should_excise(vdev_t *vd)
2366 spa_t *spa = vd->vdev_spa;
2367 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2369 ASSERT0(scn->scn_phys.scn_errors);
2370 ASSERT0(vd->vdev_children);
2372 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2375 if (vd->vdev_resilver_txg == 0 ||
2376 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2380 * When a resilver is initiated the scan will assign the scn_max_txg
2381 * value to the highest txg value that exists in all DTLs. If this
2382 * device's max DTL is not part of this scan (i.e. it is not in
2383 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2386 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2387 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2388 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2389 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2396 * Reassess DTLs after a config change or scrub completion.
2399 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2401 spa_t *spa = vd->vdev_spa;
2405 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2407 for (int c = 0; c < vd->vdev_children; c++)
2408 vdev_dtl_reassess(vd->vdev_child[c], txg,
2409 scrub_txg, scrub_done);
2411 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2414 if (vd->vdev_ops->vdev_op_leaf) {
2415 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2417 mutex_enter(&vd->vdev_dtl_lock);
2420 * If we've completed a scan cleanly then determine
2421 * if this vdev should remove any DTLs. We only want to
2422 * excise regions on vdevs that were available during
2423 * the entire duration of this scan.
2425 if (scrub_txg != 0 &&
2426 (spa->spa_scrub_started ||
2427 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2428 vdev_dtl_should_excise(vd)) {
2430 * We completed a scrub up to scrub_txg. If we
2431 * did it without rebooting, then the scrub dtl
2432 * will be valid, so excise the old region and
2433 * fold in the scrub dtl. Otherwise, leave the
2434 * dtl as-is if there was an error.
2436 * There's little trick here: to excise the beginning
2437 * of the DTL_MISSING map, we put it into a reference
2438 * tree and then add a segment with refcnt -1 that
2439 * covers the range [0, scrub_txg). This means
2440 * that each txg in that range has refcnt -1 or 0.
2441 * We then add DTL_SCRUB with a refcnt of 2, so that
2442 * entries in the range [0, scrub_txg) will have a
2443 * positive refcnt -- either 1 or 2. We then convert
2444 * the reference tree into the new DTL_MISSING map.
2446 space_reftree_create(&reftree);
2447 space_reftree_add_map(&reftree,
2448 vd->vdev_dtl[DTL_MISSING], 1);
2449 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2450 space_reftree_add_map(&reftree,
2451 vd->vdev_dtl[DTL_SCRUB], 2);
2452 space_reftree_generate_map(&reftree,
2453 vd->vdev_dtl[DTL_MISSING], 1);
2454 space_reftree_destroy(&reftree);
2456 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2457 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2458 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2460 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2461 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2462 if (!vdev_readable(vd))
2463 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2465 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2466 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2469 * If the vdev was resilvering and no longer has any
2470 * DTLs then reset its resilvering flag and dirty
2471 * the top level so that we persist the change.
2473 if (vd->vdev_resilver_txg != 0 &&
2474 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2475 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2476 vd->vdev_resilver_txg = 0;
2477 vdev_config_dirty(vd->vdev_top);
2480 mutex_exit(&vd->vdev_dtl_lock);
2483 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2487 mutex_enter(&vd->vdev_dtl_lock);
2488 for (int t = 0; t < DTL_TYPES; t++) {
2489 /* account for child's outage in parent's missing map */
2490 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2492 continue; /* leaf vdevs only */
2493 if (t == DTL_PARTIAL)
2494 minref = 1; /* i.e. non-zero */
2495 else if (vd->vdev_nparity != 0)
2496 minref = vd->vdev_nparity + 1; /* RAID-Z */
2498 minref = vd->vdev_children; /* any kind of mirror */
2499 space_reftree_create(&reftree);
2500 for (int c = 0; c < vd->vdev_children; c++) {
2501 vdev_t *cvd = vd->vdev_child[c];
2502 mutex_enter(&cvd->vdev_dtl_lock);
2503 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2504 mutex_exit(&cvd->vdev_dtl_lock);
2506 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2507 space_reftree_destroy(&reftree);
2509 mutex_exit(&vd->vdev_dtl_lock);
2513 vdev_dtl_load(vdev_t *vd)
2515 spa_t *spa = vd->vdev_spa;
2516 objset_t *mos = spa->spa_meta_objset;
2519 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2520 ASSERT(vdev_is_concrete(vd));
2522 error = space_map_open(&vd->vdev_dtl_sm, mos,
2523 vd->vdev_dtl_object, 0, -1ULL, 0);
2526 ASSERT(vd->vdev_dtl_sm != NULL);
2528 mutex_enter(&vd->vdev_dtl_lock);
2531 * Now that we've opened the space_map we need to update
2534 space_map_update(vd->vdev_dtl_sm);
2536 error = space_map_load(vd->vdev_dtl_sm,
2537 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2538 mutex_exit(&vd->vdev_dtl_lock);
2543 for (int c = 0; c < vd->vdev_children; c++) {
2544 error = vdev_dtl_load(vd->vdev_child[c]);
2553 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2555 spa_t *spa = vd->vdev_spa;
2557 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2558 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2563 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2565 spa_t *spa = vd->vdev_spa;
2566 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2567 DMU_OT_NONE, 0, tx);
2570 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2577 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2579 if (vd->vdev_ops != &vdev_hole_ops &&
2580 vd->vdev_ops != &vdev_missing_ops &&
2581 vd->vdev_ops != &vdev_root_ops &&
2582 !vd->vdev_top->vdev_removing) {
2583 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2584 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2586 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2587 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2590 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2591 vdev_construct_zaps(vd->vdev_child[i], tx);
2596 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2598 spa_t *spa = vd->vdev_spa;
2599 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2600 objset_t *mos = spa->spa_meta_objset;
2601 range_tree_t *rtsync;
2603 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2605 ASSERT(vdev_is_concrete(vd));
2606 ASSERT(vd->vdev_ops->vdev_op_leaf);
2608 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2610 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2611 mutex_enter(&vd->vdev_dtl_lock);
2612 space_map_free(vd->vdev_dtl_sm, tx);
2613 space_map_close(vd->vdev_dtl_sm);
2614 vd->vdev_dtl_sm = NULL;
2615 mutex_exit(&vd->vdev_dtl_lock);
2618 * We only destroy the leaf ZAP for detached leaves or for
2619 * removed log devices. Removed data devices handle leaf ZAP
2620 * cleanup later, once cancellation is no longer possible.
2622 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2623 vd->vdev_top->vdev_islog)) {
2624 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2625 vd->vdev_leaf_zap = 0;
2632 if (vd->vdev_dtl_sm == NULL) {
2633 uint64_t new_object;
2635 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx);
2636 VERIFY3U(new_object, !=, 0);
2638 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2640 ASSERT(vd->vdev_dtl_sm != NULL);
2643 rtsync = range_tree_create(NULL, NULL);
2645 mutex_enter(&vd->vdev_dtl_lock);
2646 range_tree_walk(rt, range_tree_add, rtsync);
2647 mutex_exit(&vd->vdev_dtl_lock);
2649 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx);
2650 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2651 range_tree_vacate(rtsync, NULL, NULL);
2653 range_tree_destroy(rtsync);
2656 * If the object for the space map has changed then dirty
2657 * the top level so that we update the config.
2659 if (object != space_map_object(vd->vdev_dtl_sm)) {
2660 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2661 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2662 (u_longlong_t)object,
2663 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2664 vdev_config_dirty(vd->vdev_top);
2669 mutex_enter(&vd->vdev_dtl_lock);
2670 space_map_update(vd->vdev_dtl_sm);
2671 mutex_exit(&vd->vdev_dtl_lock);
2675 * Determine whether the specified vdev can be offlined/detached/removed
2676 * without losing data.
2679 vdev_dtl_required(vdev_t *vd)
2681 spa_t *spa = vd->vdev_spa;
2682 vdev_t *tvd = vd->vdev_top;
2683 uint8_t cant_read = vd->vdev_cant_read;
2686 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2688 if (vd == spa->spa_root_vdev || vd == tvd)
2692 * Temporarily mark the device as unreadable, and then determine
2693 * whether this results in any DTL outages in the top-level vdev.
2694 * If not, we can safely offline/detach/remove the device.
2696 vd->vdev_cant_read = B_TRUE;
2697 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2698 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2699 vd->vdev_cant_read = cant_read;
2700 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2702 if (!required && zio_injection_enabled)
2703 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2709 * Determine if resilver is needed, and if so the txg range.
2712 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2714 boolean_t needed = B_FALSE;
2715 uint64_t thismin = UINT64_MAX;
2716 uint64_t thismax = 0;
2718 if (vd->vdev_children == 0) {
2719 mutex_enter(&vd->vdev_dtl_lock);
2720 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2721 vdev_writeable(vd)) {
2723 thismin = vdev_dtl_min(vd);
2724 thismax = vdev_dtl_max(vd);
2727 mutex_exit(&vd->vdev_dtl_lock);
2729 for (int c = 0; c < vd->vdev_children; c++) {
2730 vdev_t *cvd = vd->vdev_child[c];
2731 uint64_t cmin, cmax;
2733 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2734 thismin = MIN(thismin, cmin);
2735 thismax = MAX(thismax, cmax);
2741 if (needed && minp) {
2749 * Gets the checkpoint space map object from the vdev's ZAP.
2750 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2751 * or the ZAP doesn't exist yet.
2754 vdev_checkpoint_sm_object(vdev_t *vd)
2756 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
2757 if (vd->vdev_top_zap == 0) {
2761 uint64_t sm_obj = 0;
2762 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
2763 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj);
2765 ASSERT(err == 0 || err == ENOENT);
2771 vdev_load(vdev_t *vd)
2775 * Recursively load all children.
2777 for (int c = 0; c < vd->vdev_children; c++) {
2778 error = vdev_load(vd->vdev_child[c]);
2784 vdev_set_deflate_ratio(vd);
2787 * If this is a top-level vdev, initialize its metaslabs.
2789 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2790 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2791 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2792 VDEV_AUX_CORRUPT_DATA);
2793 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2794 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2795 (u_longlong_t)vd->vdev_asize);
2796 return (SET_ERROR(ENXIO));
2797 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2798 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2799 "[error=%d]", error);
2800 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2801 VDEV_AUX_CORRUPT_DATA);
2805 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd);
2806 if (checkpoint_sm_obj != 0) {
2807 objset_t *mos = spa_meta_objset(vd->vdev_spa);
2808 ASSERT(vd->vdev_asize != 0);
2809 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
2811 if ((error = space_map_open(&vd->vdev_checkpoint_sm,
2812 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
2813 vd->vdev_ashift))) {
2814 vdev_dbgmsg(vd, "vdev_load: space_map_open "
2815 "failed for checkpoint spacemap (obj %llu) "
2817 (u_longlong_t)checkpoint_sm_obj, error);
2820 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
2821 space_map_update(vd->vdev_checkpoint_sm);
2824 * Since the checkpoint_sm contains free entries
2825 * exclusively we can use sm_alloc to indicate the
2826 * culmulative checkpointed space that has been freed.
2828 vd->vdev_stat.vs_checkpoint_space =
2829 -vd->vdev_checkpoint_sm->sm_alloc;
2830 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
2831 vd->vdev_stat.vs_checkpoint_space;
2836 * If this is a leaf vdev, load its DTL.
2838 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2839 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2840 VDEV_AUX_CORRUPT_DATA);
2841 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
2842 "[error=%d]", error);
2846 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2847 if (obsolete_sm_object != 0) {
2848 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2849 ASSERT(vd->vdev_asize != 0);
2850 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
2852 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2853 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2854 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2855 VDEV_AUX_CORRUPT_DATA);
2856 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
2857 "obsolete spacemap (obj %llu) [error=%d]",
2858 (u_longlong_t)obsolete_sm_object, error);
2861 space_map_update(vd->vdev_obsolete_sm);
2868 * The special vdev case is used for hot spares and l2cache devices. Its
2869 * sole purpose it to set the vdev state for the associated vdev. To do this,
2870 * we make sure that we can open the underlying device, then try to read the
2871 * label, and make sure that the label is sane and that it hasn't been
2872 * repurposed to another pool.
2875 vdev_validate_aux(vdev_t *vd)
2878 uint64_t guid, version;
2881 if (!vdev_readable(vd))
2884 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2885 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2886 VDEV_AUX_CORRUPT_DATA);
2890 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2891 !SPA_VERSION_IS_SUPPORTED(version) ||
2892 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2893 guid != vd->vdev_guid ||
2894 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2895 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2896 VDEV_AUX_CORRUPT_DATA);
2902 * We don't actually check the pool state here. If it's in fact in
2903 * use by another pool, we update this fact on the fly when requested.
2910 * Free the objects used to store this vdev's spacemaps, and the array
2911 * that points to them.
2914 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
2916 if (vd->vdev_ms_array == 0)
2919 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2920 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
2921 size_t array_bytes = array_count * sizeof (uint64_t);
2922 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
2923 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
2924 array_bytes, smobj_array, 0));
2926 for (uint64_t i = 0; i < array_count; i++) {
2927 uint64_t smobj = smobj_array[i];
2931 space_map_free_obj(mos, smobj, tx);
2934 kmem_free(smobj_array, array_bytes);
2935 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
2936 vd->vdev_ms_array = 0;
2940 vdev_remove_empty(vdev_t *vd, uint64_t txg)
2942 spa_t *spa = vd->vdev_spa;
2945 ASSERT(vd == vd->vdev_top);
2946 ASSERT3U(txg, ==, spa_syncing_txg(spa));
2948 if (vd->vdev_ms != NULL) {
2949 metaslab_group_t *mg = vd->vdev_mg;
2951 metaslab_group_histogram_verify(mg);
2952 metaslab_class_histogram_verify(mg->mg_class);
2954 for (int m = 0; m < vd->vdev_ms_count; m++) {
2955 metaslab_t *msp = vd->vdev_ms[m];
2957 if (msp == NULL || msp->ms_sm == NULL)
2960 mutex_enter(&msp->ms_lock);
2962 * If the metaslab was not loaded when the vdev
2963 * was removed then the histogram accounting may
2964 * not be accurate. Update the histogram information
2965 * here so that we ensure that the metaslab group
2966 * and metaslab class are up-to-date.
2968 metaslab_group_histogram_remove(mg, msp);
2970 VERIFY0(space_map_allocated(msp->ms_sm));
2971 space_map_close(msp->ms_sm);
2973 mutex_exit(&msp->ms_lock);
2976 if (vd->vdev_checkpoint_sm != NULL) {
2977 ASSERT(spa_has_checkpoint(spa));
2978 space_map_close(vd->vdev_checkpoint_sm);
2979 vd->vdev_checkpoint_sm = NULL;
2982 metaslab_group_histogram_verify(mg);
2983 metaslab_class_histogram_verify(mg->mg_class);
2984 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2985 ASSERT0(mg->mg_histogram[i]);
2988 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2989 vdev_destroy_spacemaps(vd, tx);
2991 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
2992 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
2993 vd->vdev_top_zap = 0;
2999 vdev_sync_done(vdev_t *vd, uint64_t txg)
3002 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3004 ASSERT(vdev_is_concrete(vd));
3006 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3007 metaslab_sync_done(msp, txg);
3010 metaslab_sync_reassess(vd->vdev_mg);
3014 vdev_sync(vdev_t *vd, uint64_t txg)
3016 spa_t *spa = vd->vdev_spa;
3021 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3024 ASSERT(vd->vdev_removing ||
3025 vd->vdev_ops == &vdev_indirect_ops);
3027 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3028 vdev_indirect_sync_obsolete(vd, tx);
3032 * If the vdev is indirect, it can't have dirty
3033 * metaslabs or DTLs.
3035 if (vd->vdev_ops == &vdev_indirect_ops) {
3036 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3037 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3042 ASSERT(vdev_is_concrete(vd));
3044 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3045 !vd->vdev_removing) {
3046 ASSERT(vd == vd->vdev_top);
3047 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3048 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3049 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3050 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3051 ASSERT(vd->vdev_ms_array != 0);
3052 vdev_config_dirty(vd);
3056 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3057 metaslab_sync(msp, txg);
3058 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3061 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3062 vdev_dtl_sync(lvd, txg);
3065 * Remove the metadata associated with this vdev once it's empty.
3066 * Note that this is typically used for log/cache device removal;
3067 * we don't empty toplevel vdevs when removing them. But if
3068 * a toplevel happens to be emptied, this is not harmful.
3070 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) {
3071 vdev_remove_empty(vd, txg);
3074 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3078 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3080 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3084 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3085 * not be opened, and no I/O is attempted.
3088 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3092 spa_vdev_state_enter(spa, SCL_NONE);
3094 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3095 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3097 if (!vd->vdev_ops->vdev_op_leaf)
3098 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3103 * We don't directly use the aux state here, but if we do a
3104 * vdev_reopen(), we need this value to be present to remember why we
3107 vd->vdev_label_aux = aux;
3110 * Faulted state takes precedence over degraded.
3112 vd->vdev_delayed_close = B_FALSE;
3113 vd->vdev_faulted = 1ULL;
3114 vd->vdev_degraded = 0ULL;
3115 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3118 * If this device has the only valid copy of the data, then
3119 * back off and simply mark the vdev as degraded instead.
3121 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3122 vd->vdev_degraded = 1ULL;
3123 vd->vdev_faulted = 0ULL;
3126 * If we reopen the device and it's not dead, only then do we
3131 if (vdev_readable(vd))
3132 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3135 return (spa_vdev_state_exit(spa, vd, 0));
3139 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3140 * user that something is wrong. The vdev continues to operate as normal as far
3141 * as I/O is concerned.
3144 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3148 spa_vdev_state_enter(spa, SCL_NONE);
3150 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3151 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3153 if (!vd->vdev_ops->vdev_op_leaf)
3154 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3157 * If the vdev is already faulted, then don't do anything.
3159 if (vd->vdev_faulted || vd->vdev_degraded)
3160 return (spa_vdev_state_exit(spa, NULL, 0));
3162 vd->vdev_degraded = 1ULL;
3163 if (!vdev_is_dead(vd))
3164 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3167 return (spa_vdev_state_exit(spa, vd, 0));
3171 * Online the given vdev.
3173 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3174 * spare device should be detached when the device finishes resilvering.
3175 * Second, the online should be treated like a 'test' online case, so no FMA
3176 * events are generated if the device fails to open.
3179 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3181 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3182 boolean_t wasoffline;
3183 vdev_state_t oldstate;
3185 spa_vdev_state_enter(spa, SCL_NONE);
3187 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3188 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3190 if (!vd->vdev_ops->vdev_op_leaf)
3191 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3193 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3194 oldstate = vd->vdev_state;
3197 vd->vdev_offline = B_FALSE;
3198 vd->vdev_tmpoffline = B_FALSE;
3199 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3200 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3202 /* XXX - L2ARC 1.0 does not support expansion */
3203 if (!vd->vdev_aux) {
3204 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3205 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
3209 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3211 if (!vd->vdev_aux) {
3212 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3213 pvd->vdev_expanding = B_FALSE;
3217 *newstate = vd->vdev_state;
3218 if ((flags & ZFS_ONLINE_UNSPARE) &&
3219 !vdev_is_dead(vd) && vd->vdev_parent &&
3220 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3221 vd->vdev_parent->vdev_child[0] == vd)
3222 vd->vdev_unspare = B_TRUE;
3224 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3226 /* XXX - L2ARC 1.0 does not support expansion */
3228 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3229 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3233 (oldstate < VDEV_STATE_DEGRADED &&
3234 vd->vdev_state >= VDEV_STATE_DEGRADED))
3235 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3237 return (spa_vdev_state_exit(spa, vd, 0));
3241 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3245 uint64_t generation;
3246 metaslab_group_t *mg;
3249 spa_vdev_state_enter(spa, SCL_ALLOC);
3251 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3252 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3254 if (!vd->vdev_ops->vdev_op_leaf)
3255 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3259 generation = spa->spa_config_generation + 1;
3262 * If the device isn't already offline, try to offline it.
3264 if (!vd->vdev_offline) {
3266 * If this device has the only valid copy of some data,
3267 * don't allow it to be offlined. Log devices are always
3270 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3271 vdev_dtl_required(vd))
3272 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3275 * If the top-level is a slog and it has had allocations
3276 * then proceed. We check that the vdev's metaslab group
3277 * is not NULL since it's possible that we may have just
3278 * added this vdev but not yet initialized its metaslabs.
3280 if (tvd->vdev_islog && mg != NULL) {
3282 * Prevent any future allocations.
3284 metaslab_group_passivate(mg);
3285 (void) spa_vdev_state_exit(spa, vd, 0);
3287 error = spa_reset_logs(spa);
3290 * If the log device was successfully reset but has
3291 * checkpointed data, do not offline it.
3294 tvd->vdev_checkpoint_sm != NULL) {
3295 ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc,
3297 error = ZFS_ERR_CHECKPOINT_EXISTS;
3300 spa_vdev_state_enter(spa, SCL_ALLOC);
3303 * Check to see if the config has changed.
3305 if (error || generation != spa->spa_config_generation) {
3306 metaslab_group_activate(mg);
3308 return (spa_vdev_state_exit(spa,
3310 (void) spa_vdev_state_exit(spa, vd, 0);
3313 ASSERT0(tvd->vdev_stat.vs_alloc);
3317 * Offline this device and reopen its top-level vdev.
3318 * If the top-level vdev is a log device then just offline
3319 * it. Otherwise, if this action results in the top-level
3320 * vdev becoming unusable, undo it and fail the request.
3322 vd->vdev_offline = B_TRUE;
3325 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3326 vdev_is_dead(tvd)) {
3327 vd->vdev_offline = B_FALSE;
3329 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3333 * Add the device back into the metaslab rotor so that
3334 * once we online the device it's open for business.
3336 if (tvd->vdev_islog && mg != NULL)
3337 metaslab_group_activate(mg);
3340 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3342 return (spa_vdev_state_exit(spa, vd, 0));
3346 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3350 mutex_enter(&spa->spa_vdev_top_lock);
3351 error = vdev_offline_locked(spa, guid, flags);
3352 mutex_exit(&spa->spa_vdev_top_lock);
3358 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3359 * vdev_offline(), we assume the spa config is locked. We also clear all
3360 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3363 vdev_clear(spa_t *spa, vdev_t *vd)
3365 vdev_t *rvd = spa->spa_root_vdev;
3367 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3372 vd->vdev_stat.vs_read_errors = 0;
3373 vd->vdev_stat.vs_write_errors = 0;
3374 vd->vdev_stat.vs_checksum_errors = 0;
3376 for (int c = 0; c < vd->vdev_children; c++)
3377 vdev_clear(spa, vd->vdev_child[c]);
3380 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
3381 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
3383 for (int c = 0; c < spa->spa_spares.sav_count; c++)
3384 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
3388 * It makes no sense to "clear" an indirect vdev.
3390 if (!vdev_is_concrete(vd))
3394 * If we're in the FAULTED state or have experienced failed I/O, then
3395 * clear the persistent state and attempt to reopen the device. We
3396 * also mark the vdev config dirty, so that the new faulted state is
3397 * written out to disk.
3399 if (vd->vdev_faulted || vd->vdev_degraded ||
3400 !vdev_readable(vd) || !vdev_writeable(vd)) {
3403 * When reopening in reponse to a clear event, it may be due to
3404 * a fmadm repair request. In this case, if the device is
3405 * still broken, we want to still post the ereport again.
3407 vd->vdev_forcefault = B_TRUE;
3409 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3410 vd->vdev_cant_read = B_FALSE;
3411 vd->vdev_cant_write = B_FALSE;
3413 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3415 vd->vdev_forcefault = B_FALSE;
3417 if (vd != rvd && vdev_writeable(vd->vdev_top))
3418 vdev_state_dirty(vd->vdev_top);
3420 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
3421 spa_async_request(spa, SPA_ASYNC_RESILVER);
3423 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3427 * When clearing a FMA-diagnosed fault, we always want to
3428 * unspare the device, as we assume that the original spare was
3429 * done in response to the FMA fault.
3431 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3432 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3433 vd->vdev_parent->vdev_child[0] == vd)
3434 vd->vdev_unspare = B_TRUE;
3438 vdev_is_dead(vdev_t *vd)
3441 * Holes and missing devices are always considered "dead".
3442 * This simplifies the code since we don't have to check for
3443 * these types of devices in the various code paths.
3444 * Instead we rely on the fact that we skip over dead devices
3445 * before issuing I/O to them.
3447 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3448 vd->vdev_ops == &vdev_hole_ops ||
3449 vd->vdev_ops == &vdev_missing_ops);
3453 vdev_readable(vdev_t *vd)
3455 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3459 vdev_writeable(vdev_t *vd)
3461 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3462 vdev_is_concrete(vd));
3466 vdev_allocatable(vdev_t *vd)
3468 uint64_t state = vd->vdev_state;
3471 * We currently allow allocations from vdevs which may be in the
3472 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3473 * fails to reopen then we'll catch it later when we're holding
3474 * the proper locks. Note that we have to get the vdev state
3475 * in a local variable because although it changes atomically,
3476 * we're asking two separate questions about it.
3478 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3479 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3480 vd->vdev_mg->mg_initialized);
3484 vdev_accessible(vdev_t *vd, zio_t *zio)
3486 ASSERT(zio->io_vd == vd);
3488 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3491 if (zio->io_type == ZIO_TYPE_READ)
3492 return (!vd->vdev_cant_read);
3494 if (zio->io_type == ZIO_TYPE_WRITE)
3495 return (!vd->vdev_cant_write);
3501 vdev_is_spacemap_addressable(vdev_t *vd)
3504 * Assuming 47 bits of the space map entry dedicated for the entry's
3505 * offset (see description in space_map.h), we calculate the maximum
3506 * address that can be described by a space map entry for the given
3509 uint64_t shift = vd->vdev_ashift + 47;
3511 if (shift >= 63) /* detect potential overflow */
3514 return (vd->vdev_asize < (1ULL << shift));
3518 * Get statistics for the given vdev.
3521 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3523 spa_t *spa = vd->vdev_spa;
3524 vdev_t *rvd = spa->spa_root_vdev;
3525 vdev_t *tvd = vd->vdev_top;
3527 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3529 mutex_enter(&vd->vdev_stat_lock);
3530 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3531 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3532 vs->vs_state = vd->vdev_state;
3533 vs->vs_rsize = vdev_get_min_asize(vd);
3534 if (vd->vdev_ops->vdev_op_leaf)
3535 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3537 * Report expandable space on top-level, non-auxillary devices only.
3538 * The expandable space is reported in terms of metaslab sized units
3539 * since that determines how much space the pool can expand.
3541 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
3542 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3543 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3545 vs->vs_configured_ashift = vd->vdev_top != NULL
3546 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
3547 vs->vs_logical_ashift = vd->vdev_logical_ashift;
3548 vs->vs_physical_ashift = vd->vdev_physical_ashift;
3549 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3550 vdev_is_concrete(vd)) {
3551 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
3555 * If we're getting stats on the root vdev, aggregate the I/O counts
3556 * over all top-level vdevs (i.e. the direct children of the root).
3559 for (int c = 0; c < rvd->vdev_children; c++) {
3560 vdev_t *cvd = rvd->vdev_child[c];
3561 vdev_stat_t *cvs = &cvd->vdev_stat;
3563 for (int t = 0; t < ZIO_TYPES; t++) {
3564 vs->vs_ops[t] += cvs->vs_ops[t];
3565 vs->vs_bytes[t] += cvs->vs_bytes[t];
3567 cvs->vs_scan_removing = cvd->vdev_removing;
3570 mutex_exit(&vd->vdev_stat_lock);
3574 vdev_clear_stats(vdev_t *vd)
3576 mutex_enter(&vd->vdev_stat_lock);
3577 vd->vdev_stat.vs_space = 0;
3578 vd->vdev_stat.vs_dspace = 0;
3579 vd->vdev_stat.vs_alloc = 0;
3580 mutex_exit(&vd->vdev_stat_lock);
3584 vdev_scan_stat_init(vdev_t *vd)
3586 vdev_stat_t *vs = &vd->vdev_stat;
3588 for (int c = 0; c < vd->vdev_children; c++)
3589 vdev_scan_stat_init(vd->vdev_child[c]);
3591 mutex_enter(&vd->vdev_stat_lock);
3592 vs->vs_scan_processed = 0;
3593 mutex_exit(&vd->vdev_stat_lock);
3597 vdev_stat_update(zio_t *zio, uint64_t psize)
3599 spa_t *spa = zio->io_spa;
3600 vdev_t *rvd = spa->spa_root_vdev;
3601 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3603 uint64_t txg = zio->io_txg;
3604 vdev_stat_t *vs = &vd->vdev_stat;
3605 zio_type_t type = zio->io_type;
3606 int flags = zio->io_flags;
3609 * If this i/o is a gang leader, it didn't do any actual work.
3611 if (zio->io_gang_tree)
3614 if (zio->io_error == 0) {
3616 * If this is a root i/o, don't count it -- we've already
3617 * counted the top-level vdevs, and vdev_get_stats() will
3618 * aggregate them when asked. This reduces contention on
3619 * the root vdev_stat_lock and implicitly handles blocks
3620 * that compress away to holes, for which there is no i/o.
3621 * (Holes never create vdev children, so all the counters
3622 * remain zero, which is what we want.)
3624 * Note: this only applies to successful i/o (io_error == 0)
3625 * because unlike i/o counts, errors are not additive.
3626 * When reading a ditto block, for example, failure of
3627 * one top-level vdev does not imply a root-level error.
3632 ASSERT(vd == zio->io_vd);
3634 if (flags & ZIO_FLAG_IO_BYPASS)
3637 mutex_enter(&vd->vdev_stat_lock);
3639 if (flags & ZIO_FLAG_IO_REPAIR) {
3640 if (flags & ZIO_FLAG_SCAN_THREAD) {
3641 dsl_scan_phys_t *scn_phys =
3642 &spa->spa_dsl_pool->dp_scan->scn_phys;
3643 uint64_t *processed = &scn_phys->scn_processed;
3646 if (vd->vdev_ops->vdev_op_leaf)
3647 atomic_add_64(processed, psize);
3648 vs->vs_scan_processed += psize;
3651 if (flags & ZIO_FLAG_SELF_HEAL)
3652 vs->vs_self_healed += psize;
3656 vs->vs_bytes[type] += psize;
3658 mutex_exit(&vd->vdev_stat_lock);
3662 if (flags & ZIO_FLAG_SPECULATIVE)
3666 * If this is an I/O error that is going to be retried, then ignore the
3667 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3668 * hard errors, when in reality they can happen for any number of
3669 * innocuous reasons (bus resets, MPxIO link failure, etc).
3671 if (zio->io_error == EIO &&
3672 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3676 * Intent logs writes won't propagate their error to the root
3677 * I/O so don't mark these types of failures as pool-level
3680 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3683 mutex_enter(&vd->vdev_stat_lock);
3684 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3685 if (zio->io_error == ECKSUM)
3686 vs->vs_checksum_errors++;
3688 vs->vs_read_errors++;
3690 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3691 vs->vs_write_errors++;
3692 mutex_exit(&vd->vdev_stat_lock);
3694 if (spa->spa_load_state == SPA_LOAD_NONE &&
3695 type == ZIO_TYPE_WRITE && txg != 0 &&
3696 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3697 (flags & ZIO_FLAG_SCAN_THREAD) ||
3698 spa->spa_claiming)) {
3700 * This is either a normal write (not a repair), or it's
3701 * a repair induced by the scrub thread, or it's a repair
3702 * made by zil_claim() during spa_load() in the first txg.
3703 * In the normal case, we commit the DTL change in the same
3704 * txg as the block was born. In the scrub-induced repair
3705 * case, we know that scrubs run in first-pass syncing context,
3706 * so we commit the DTL change in spa_syncing_txg(spa).
3707 * In the zil_claim() case, we commit in spa_first_txg(spa).
3709 * We currently do not make DTL entries for failed spontaneous
3710 * self-healing writes triggered by normal (non-scrubbing)
3711 * reads, because we have no transactional context in which to
3712 * do so -- and it's not clear that it'd be desirable anyway.
3714 if (vd->vdev_ops->vdev_op_leaf) {
3715 uint64_t commit_txg = txg;
3716 if (flags & ZIO_FLAG_SCAN_THREAD) {
3717 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3718 ASSERT(spa_sync_pass(spa) == 1);
3719 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3720 commit_txg = spa_syncing_txg(spa);
3721 } else if (spa->spa_claiming) {
3722 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3723 commit_txg = spa_first_txg(spa);
3725 ASSERT(commit_txg >= spa_syncing_txg(spa));
3726 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3728 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3729 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3730 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3733 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3738 * Update the in-core space usage stats for this vdev, its metaslab class,
3739 * and the root vdev.
3742 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3743 int64_t space_delta)
3745 int64_t dspace_delta = space_delta;
3746 spa_t *spa = vd->vdev_spa;
3747 vdev_t *rvd = spa->spa_root_vdev;
3748 metaslab_group_t *mg = vd->vdev_mg;
3749 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3751 ASSERT(vd == vd->vdev_top);
3754 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3755 * factor. We must calculate this here and not at the root vdev
3756 * because the root vdev's psize-to-asize is simply the max of its
3757 * childrens', thus not accurate enough for us.
3759 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3760 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3761 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3762 vd->vdev_deflate_ratio;
3764 mutex_enter(&vd->vdev_stat_lock);
3765 vd->vdev_stat.vs_alloc += alloc_delta;
3766 vd->vdev_stat.vs_space += space_delta;
3767 vd->vdev_stat.vs_dspace += dspace_delta;
3768 mutex_exit(&vd->vdev_stat_lock);
3770 if (mc == spa_normal_class(spa)) {
3771 mutex_enter(&rvd->vdev_stat_lock);
3772 rvd->vdev_stat.vs_alloc += alloc_delta;
3773 rvd->vdev_stat.vs_space += space_delta;
3774 rvd->vdev_stat.vs_dspace += dspace_delta;
3775 mutex_exit(&rvd->vdev_stat_lock);
3779 ASSERT(rvd == vd->vdev_parent);
3780 ASSERT(vd->vdev_ms_count != 0);
3782 metaslab_class_space_update(mc,
3783 alloc_delta, defer_delta, space_delta, dspace_delta);
3788 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3789 * so that it will be written out next time the vdev configuration is synced.
3790 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3793 vdev_config_dirty(vdev_t *vd)
3795 spa_t *spa = vd->vdev_spa;
3796 vdev_t *rvd = spa->spa_root_vdev;
3799 ASSERT(spa_writeable(spa));
3802 * If this is an aux vdev (as with l2cache and spare devices), then we
3803 * update the vdev config manually and set the sync flag.
3805 if (vd->vdev_aux != NULL) {
3806 spa_aux_vdev_t *sav = vd->vdev_aux;
3810 for (c = 0; c < sav->sav_count; c++) {
3811 if (sav->sav_vdevs[c] == vd)
3815 if (c == sav->sav_count) {
3817 * We're being removed. There's nothing more to do.
3819 ASSERT(sav->sav_sync == B_TRUE);
3823 sav->sav_sync = B_TRUE;
3825 if (nvlist_lookup_nvlist_array(sav->sav_config,
3826 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3827 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3828 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3834 * Setting the nvlist in the middle if the array is a little
3835 * sketchy, but it will work.
3837 nvlist_free(aux[c]);
3838 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3844 * The dirty list is protected by the SCL_CONFIG lock. The caller
3845 * must either hold SCL_CONFIG as writer, or must be the sync thread
3846 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3847 * so this is sufficient to ensure mutual exclusion.
3849 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3850 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3851 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3854 for (c = 0; c < rvd->vdev_children; c++)
3855 vdev_config_dirty(rvd->vdev_child[c]);
3857 ASSERT(vd == vd->vdev_top);
3859 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3860 vdev_is_concrete(vd)) {
3861 list_insert_head(&spa->spa_config_dirty_list, vd);
3867 vdev_config_clean(vdev_t *vd)
3869 spa_t *spa = vd->vdev_spa;
3871 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3872 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3873 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3875 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3876 list_remove(&spa->spa_config_dirty_list, vd);
3880 * Mark a top-level vdev's state as dirty, so that the next pass of
3881 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3882 * the state changes from larger config changes because they require
3883 * much less locking, and are often needed for administrative actions.
3886 vdev_state_dirty(vdev_t *vd)
3888 spa_t *spa = vd->vdev_spa;
3890 ASSERT(spa_writeable(spa));
3891 ASSERT(vd == vd->vdev_top);
3894 * The state list is protected by the SCL_STATE lock. The caller
3895 * must either hold SCL_STATE as writer, or must be the sync thread
3896 * (which holds SCL_STATE as reader). There's only one sync thread,
3897 * so this is sufficient to ensure mutual exclusion.
3899 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3900 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3901 spa_config_held(spa, SCL_STATE, RW_READER)));
3903 if (!list_link_active(&vd->vdev_state_dirty_node) &&
3904 vdev_is_concrete(vd))
3905 list_insert_head(&spa->spa_state_dirty_list, vd);
3909 vdev_state_clean(vdev_t *vd)
3911 spa_t *spa = vd->vdev_spa;
3913 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3914 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3915 spa_config_held(spa, SCL_STATE, RW_READER)));
3917 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3918 list_remove(&spa->spa_state_dirty_list, vd);
3922 * Propagate vdev state up from children to parent.
3925 vdev_propagate_state(vdev_t *vd)
3927 spa_t *spa = vd->vdev_spa;
3928 vdev_t *rvd = spa->spa_root_vdev;
3929 int degraded = 0, faulted = 0;
3933 if (vd->vdev_children > 0) {
3934 for (int c = 0; c < vd->vdev_children; c++) {
3935 child = vd->vdev_child[c];
3938 * Don't factor holes or indirect vdevs into the
3941 if (!vdev_is_concrete(child))
3944 if (!vdev_readable(child) ||
3945 (!vdev_writeable(child) && spa_writeable(spa))) {
3947 * Root special: if there is a top-level log
3948 * device, treat the root vdev as if it were
3951 if (child->vdev_islog && vd == rvd)
3955 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3959 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3963 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3966 * Root special: if there is a top-level vdev that cannot be
3967 * opened due to corrupted metadata, then propagate the root
3968 * vdev's aux state as 'corrupt' rather than 'insufficient
3971 if (corrupted && vd == rvd &&
3972 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3973 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3974 VDEV_AUX_CORRUPT_DATA);
3977 if (vd->vdev_parent)
3978 vdev_propagate_state(vd->vdev_parent);
3982 * Set a vdev's state. If this is during an open, we don't update the parent
3983 * state, because we're in the process of opening children depth-first.
3984 * Otherwise, we propagate the change to the parent.
3986 * If this routine places a device in a faulted state, an appropriate ereport is
3990 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3992 uint64_t save_state;
3993 spa_t *spa = vd->vdev_spa;
3995 if (state == vd->vdev_state) {
3996 vd->vdev_stat.vs_aux = aux;
4000 save_state = vd->vdev_state;
4002 vd->vdev_state = state;
4003 vd->vdev_stat.vs_aux = aux;
4006 * If we are setting the vdev state to anything but an open state, then
4007 * always close the underlying device unless the device has requested
4008 * a delayed close (i.e. we're about to remove or fault the device).
4009 * Otherwise, we keep accessible but invalid devices open forever.
4010 * We don't call vdev_close() itself, because that implies some extra
4011 * checks (offline, etc) that we don't want here. This is limited to
4012 * leaf devices, because otherwise closing the device will affect other
4015 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4016 vd->vdev_ops->vdev_op_leaf)
4017 vd->vdev_ops->vdev_op_close(vd);
4019 if (vd->vdev_removed &&
4020 state == VDEV_STATE_CANT_OPEN &&
4021 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4023 * If the previous state is set to VDEV_STATE_REMOVED, then this
4024 * device was previously marked removed and someone attempted to
4025 * reopen it. If this failed due to a nonexistent device, then
4026 * keep the device in the REMOVED state. We also let this be if
4027 * it is one of our special test online cases, which is only
4028 * attempting to online the device and shouldn't generate an FMA
4031 vd->vdev_state = VDEV_STATE_REMOVED;
4032 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4033 } else if (state == VDEV_STATE_REMOVED) {
4034 vd->vdev_removed = B_TRUE;
4035 } else if (state == VDEV_STATE_CANT_OPEN) {
4037 * If we fail to open a vdev during an import or recovery, we
4038 * mark it as "not available", which signifies that it was
4039 * never there to begin with. Failure to open such a device
4040 * is not considered an error.
4042 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4043 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4044 vd->vdev_ops->vdev_op_leaf)
4045 vd->vdev_not_present = 1;
4048 * Post the appropriate ereport. If the 'prevstate' field is
4049 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4050 * that this is part of a vdev_reopen(). In this case, we don't
4051 * want to post the ereport if the device was already in the
4052 * CANT_OPEN state beforehand.
4054 * If the 'checkremove' flag is set, then this is an attempt to
4055 * online the device in response to an insertion event. If we
4056 * hit this case, then we have detected an insertion event for a
4057 * faulted or offline device that wasn't in the removed state.
4058 * In this scenario, we don't post an ereport because we are
4059 * about to replace the device, or attempt an online with
4060 * vdev_forcefault, which will generate the fault for us.
4062 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4063 !vd->vdev_not_present && !vd->vdev_checkremove &&
4064 vd != spa->spa_root_vdev) {
4068 case VDEV_AUX_OPEN_FAILED:
4069 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4071 case VDEV_AUX_CORRUPT_DATA:
4072 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4074 case VDEV_AUX_NO_REPLICAS:
4075 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4077 case VDEV_AUX_BAD_GUID_SUM:
4078 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4080 case VDEV_AUX_TOO_SMALL:
4081 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4083 case VDEV_AUX_BAD_LABEL:
4084 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4087 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4090 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
4093 /* Erase any notion of persistent removed state */
4094 vd->vdev_removed = B_FALSE;
4096 vd->vdev_removed = B_FALSE;
4100 * Notify the fmd of the state change. Be verbose and post
4101 * notifications even for stuff that's not important; the fmd agent can
4102 * sort it out. Don't emit state change events for non-leaf vdevs since
4103 * they can't change state on their own. The FMD can check their state
4104 * if it wants to when it sees that a leaf vdev had a state change.
4106 if (vd->vdev_ops->vdev_op_leaf)
4107 zfs_post_state_change(spa, vd);
4109 if (!isopen && vd->vdev_parent)
4110 vdev_propagate_state(vd->vdev_parent);
4114 vdev_children_are_offline(vdev_t *vd)
4116 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4118 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4119 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4127 * Check the vdev configuration to ensure that it's capable of supporting
4128 * a root pool. We do not support partial configuration.
4129 * In addition, only a single top-level vdev is allowed.
4131 * FreeBSD does not have above limitations.
4134 vdev_is_bootable(vdev_t *vd)
4137 if (!vd->vdev_ops->vdev_op_leaf) {
4138 char *vdev_type = vd->vdev_ops->vdev_op_type;
4140 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
4141 vd->vdev_children > 1) {
4143 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4144 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4149 for (int c = 0; c < vd->vdev_children; c++) {
4150 if (!vdev_is_bootable(vd->vdev_child[c]))
4153 #endif /* illumos */
4158 vdev_is_concrete(vdev_t *vd)
4160 vdev_ops_t *ops = vd->vdev_ops;
4161 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4162 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4170 * Determine if a log device has valid content. If the vdev was
4171 * removed or faulted in the MOS config then we know that
4172 * the content on the log device has already been written to the pool.
4175 vdev_log_state_valid(vdev_t *vd)
4177 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4181 for (int c = 0; c < vd->vdev_children; c++)
4182 if (vdev_log_state_valid(vd->vdev_child[c]))
4189 * Expand a vdev if possible.
4192 vdev_expand(vdev_t *vd, uint64_t txg)
4194 ASSERT(vd->vdev_top == vd);
4195 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4197 vdev_set_deflate_ratio(vd);
4199 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
4200 vdev_is_concrete(vd)) {
4201 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4202 vdev_config_dirty(vd);
4210 vdev_split(vdev_t *vd)
4212 vdev_t *cvd, *pvd = vd->vdev_parent;
4214 vdev_remove_child(pvd, vd);
4215 vdev_compact_children(pvd);
4217 cvd = pvd->vdev_child[0];
4218 if (pvd->vdev_children == 1) {
4219 vdev_remove_parent(cvd);
4220 cvd->vdev_splitting = B_TRUE;
4222 vdev_propagate_state(cvd);
4226 vdev_deadman(vdev_t *vd)
4228 for (int c = 0; c < vd->vdev_children; c++) {
4229 vdev_t *cvd = vd->vdev_child[c];
4234 if (vd->vdev_ops->vdev_op_leaf) {
4235 vdev_queue_t *vq = &vd->vdev_queue;
4237 mutex_enter(&vq->vq_lock);
4238 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4239 spa_t *spa = vd->vdev_spa;
4244 * Look at the head of all the pending queues,
4245 * if any I/O has been outstanding for longer than
4246 * the spa_deadman_synctime we panic the system.
4248 fio = avl_first(&vq->vq_active_tree);
4249 delta = gethrtime() - fio->io_timestamp;
4250 if (delta > spa_deadman_synctime(spa)) {
4251 vdev_dbgmsg(vd, "SLOW IO: zio timestamp "
4252 "%lluns, delta %lluns, last io %lluns",
4253 fio->io_timestamp, (u_longlong_t)delta,
4254 vq->vq_io_complete_ts);
4255 fm_panic("I/O to pool '%s' appears to be "
4256 "hung on vdev guid %llu at '%s'.",
4258 (long long unsigned int) vd->vdev_guid,
4262 mutex_exit(&vq->vq_lock);