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>
54 #include <sys/vdev_initialize.h>
56 SYSCTL_DECL(_vfs_zfs);
57 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
60 * Virtual device management.
64 * The limit for ZFS to automatically increase a top-level vdev's ashift
65 * from logical ashift to physical ashift.
67 * Example: one or more 512B emulation child vdevs
68 * child->vdev_ashift = 9 (512 bytes)
69 * child->vdev_physical_ashift = 12 (4096 bytes)
70 * zfs_max_auto_ashift = 11 (2048 bytes)
71 * zfs_min_auto_ashift = 9 (512 bytes)
73 * On pool creation or the addition of a new top-level vdev, ZFS will
74 * increase the ashift of the top-level vdev to 2048 as limited by
75 * zfs_max_auto_ashift.
77 * Example: one or more 512B emulation child vdevs
78 * child->vdev_ashift = 9 (512 bytes)
79 * child->vdev_physical_ashift = 12 (4096 bytes)
80 * zfs_max_auto_ashift = 13 (8192 bytes)
81 * zfs_min_auto_ashift = 9 (512 bytes)
83 * On pool creation or the addition of a new top-level vdev, ZFS will
84 * increase the ashift of the top-level vdev to 4096 to match the
85 * max vdev_physical_ashift.
87 * Example: one or more 512B emulation child vdevs
88 * child->vdev_ashift = 9 (512 bytes)
89 * child->vdev_physical_ashift = 9 (512 bytes)
90 * zfs_max_auto_ashift = 13 (8192 bytes)
91 * zfs_min_auto_ashift = 12 (4096 bytes)
93 * On pool creation or the addition of a new top-level vdev, ZFS will
94 * increase the ashift of the top-level vdev to 4096 to match the
95 * zfs_min_auto_ashift.
97 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
98 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
101 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
106 val = zfs_max_auto_ashift;
107 err = sysctl_handle_64(oidp, &val, 0, req);
108 if (err != 0 || req->newptr == NULL)
111 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
114 zfs_max_auto_ashift = val;
118 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
119 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
120 sysctl_vfs_zfs_max_auto_ashift, "QU",
121 "Max ashift used when optimising for logical -> physical sectors size on "
122 "new top-level vdevs.");
125 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
130 val = zfs_min_auto_ashift;
131 err = sysctl_handle_64(oidp, &val, 0, req);
132 if (err != 0 || req->newptr == NULL)
135 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
138 zfs_min_auto_ashift = val;
142 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
143 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
144 sysctl_vfs_zfs_min_auto_ashift, "QU",
145 "Min ashift used when creating new top-level vdevs.");
147 static vdev_ops_t *vdev_ops_table[] = {
166 /* target number of metaslabs per top-level vdev */
167 int vdev_max_ms_count = 200;
168 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, max_ms_count, CTLFLAG_RDTUN,
169 &vdev_max_ms_count, 0,
170 "Maximum number of metaslabs per top-level vdev");
172 /* minimum number of metaslabs per top-level vdev */
173 int vdev_min_ms_count = 16;
174 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, min_ms_count, CTLFLAG_RDTUN,
175 &vdev_min_ms_count, 0,
176 "Minimum number of metaslabs per top-level vdev");
178 /* practical upper limit of total metaslabs per top-level vdev */
179 int vdev_ms_count_limit = 1ULL << 17;
181 /* lower limit for metaslab size (512M) */
182 int vdev_default_ms_shift = 29;
183 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, default_ms_shift, CTLFLAG_RDTUN,
184 &vdev_default_ms_shift, 0,
185 "Shift between vdev size and number of metaslabs");
187 /* upper limit for metaslab size (256G) */
188 int vdev_max_ms_shift = 38;
190 boolean_t vdev_validate_skip = B_FALSE;
193 * Since the DTL space map of a vdev is not expected to have a lot of
194 * entries, we default its block size to 4K.
196 int vdev_dtl_sm_blksz = (1 << 12);
197 SYSCTL_INT(_vfs_zfs, OID_AUTO, dtl_sm_blksz, CTLFLAG_RDTUN,
198 &vdev_dtl_sm_blksz, 0,
199 "Block size for DTL space map. Power of 2 and greater than 4096.");
202 * vdev-wide space maps that have lots of entries written to them at
203 * the end of each transaction can benefit from a higher I/O bandwidth
204 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
206 int vdev_standard_sm_blksz = (1 << 17);
207 SYSCTL_INT(_vfs_zfs, OID_AUTO, standard_sm_blksz, CTLFLAG_RDTUN,
208 &vdev_standard_sm_blksz, 0,
209 "Block size for standard space map. Power of 2 and greater than 4096.");
213 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
219 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
222 if (vd->vdev_path != NULL) {
223 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
226 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
227 vd->vdev_ops->vdev_op_type,
228 (u_longlong_t)vd->vdev_id,
229 (u_longlong_t)vd->vdev_guid, buf);
234 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
238 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
239 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
240 vd->vdev_ops->vdev_op_type);
244 switch (vd->vdev_state) {
245 case VDEV_STATE_UNKNOWN:
246 (void) snprintf(state, sizeof (state), "unknown");
248 case VDEV_STATE_CLOSED:
249 (void) snprintf(state, sizeof (state), "closed");
251 case VDEV_STATE_OFFLINE:
252 (void) snprintf(state, sizeof (state), "offline");
254 case VDEV_STATE_REMOVED:
255 (void) snprintf(state, sizeof (state), "removed");
257 case VDEV_STATE_CANT_OPEN:
258 (void) snprintf(state, sizeof (state), "can't open");
260 case VDEV_STATE_FAULTED:
261 (void) snprintf(state, sizeof (state), "faulted");
263 case VDEV_STATE_DEGRADED:
264 (void) snprintf(state, sizeof (state), "degraded");
266 case VDEV_STATE_HEALTHY:
267 (void) snprintf(state, sizeof (state), "healthy");
270 (void) snprintf(state, sizeof (state), "<state %u>",
271 (uint_t)vd->vdev_state);
274 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
275 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
276 vd->vdev_islog ? " (log)" : "",
277 (u_longlong_t)vd->vdev_guid,
278 vd->vdev_path ? vd->vdev_path : "N/A", state);
280 for (uint64_t i = 0; i < vd->vdev_children; i++)
281 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
285 * Given a vdev type, return the appropriate ops vector.
288 vdev_getops(const char *type)
290 vdev_ops_t *ops, **opspp;
292 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
293 if (strcmp(ops->vdev_op_type, type) == 0)
301 vdev_default_xlate(vdev_t *vd, const range_seg_t *in, range_seg_t *res)
303 res->rs_start = in->rs_start;
304 res->rs_end = in->rs_end;
308 * Default asize function: return the MAX of psize with the asize of
309 * all children. This is what's used by anything other than RAID-Z.
312 vdev_default_asize(vdev_t *vd, uint64_t psize)
314 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
317 for (int c = 0; c < vd->vdev_children; c++) {
318 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
319 asize = MAX(asize, csize);
326 * Get the minimum allocatable size. We define the allocatable size as
327 * the vdev's asize rounded to the nearest metaslab. This allows us to
328 * replace or attach devices which don't have the same physical size but
329 * can still satisfy the same number of allocations.
332 vdev_get_min_asize(vdev_t *vd)
334 vdev_t *pvd = vd->vdev_parent;
337 * If our parent is NULL (inactive spare or cache) or is the root,
338 * just return our own asize.
341 return (vd->vdev_asize);
344 * The top-level vdev just returns the allocatable size rounded
345 * to the nearest metaslab.
347 if (vd == vd->vdev_top)
348 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
351 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
352 * so each child must provide at least 1/Nth of its asize.
354 if (pvd->vdev_ops == &vdev_raidz_ops)
355 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
358 return (pvd->vdev_min_asize);
362 vdev_set_min_asize(vdev_t *vd)
364 vd->vdev_min_asize = vdev_get_min_asize(vd);
366 for (int c = 0; c < vd->vdev_children; c++)
367 vdev_set_min_asize(vd->vdev_child[c]);
371 vdev_lookup_top(spa_t *spa, uint64_t vdev)
373 vdev_t *rvd = spa->spa_root_vdev;
375 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
377 if (vdev < rvd->vdev_children) {
378 ASSERT(rvd->vdev_child[vdev] != NULL);
379 return (rvd->vdev_child[vdev]);
386 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
390 if (vd->vdev_guid == guid)
393 for (int c = 0; c < vd->vdev_children; c++)
394 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
402 vdev_count_leaves_impl(vdev_t *vd)
406 if (vd->vdev_ops->vdev_op_leaf)
409 for (int c = 0; c < vd->vdev_children; c++)
410 n += vdev_count_leaves_impl(vd->vdev_child[c]);
416 vdev_count_leaves(spa_t *spa)
418 return (vdev_count_leaves_impl(spa->spa_root_vdev));
422 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
424 size_t oldsize, newsize;
425 uint64_t id = cvd->vdev_id;
427 spa_t *spa = cvd->vdev_spa;
429 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
430 ASSERT(cvd->vdev_parent == NULL);
432 cvd->vdev_parent = pvd;
437 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
439 oldsize = pvd->vdev_children * sizeof (vdev_t *);
440 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
441 newsize = pvd->vdev_children * sizeof (vdev_t *);
443 newchild = kmem_zalloc(newsize, KM_SLEEP);
444 if (pvd->vdev_child != NULL) {
445 bcopy(pvd->vdev_child, newchild, oldsize);
446 kmem_free(pvd->vdev_child, oldsize);
449 pvd->vdev_child = newchild;
450 pvd->vdev_child[id] = cvd;
452 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
453 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
456 * Walk up all ancestors to update guid sum.
458 for (; pvd != NULL; pvd = pvd->vdev_parent)
459 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
463 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
466 uint_t id = cvd->vdev_id;
468 ASSERT(cvd->vdev_parent == pvd);
473 ASSERT(id < pvd->vdev_children);
474 ASSERT(pvd->vdev_child[id] == cvd);
476 pvd->vdev_child[id] = NULL;
477 cvd->vdev_parent = NULL;
479 for (c = 0; c < pvd->vdev_children; c++)
480 if (pvd->vdev_child[c])
483 if (c == pvd->vdev_children) {
484 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
485 pvd->vdev_child = NULL;
486 pvd->vdev_children = 0;
490 * Walk up all ancestors to update guid sum.
492 for (; pvd != NULL; pvd = pvd->vdev_parent)
493 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
497 * Remove any holes in the child array.
500 vdev_compact_children(vdev_t *pvd)
502 vdev_t **newchild, *cvd;
503 int oldc = pvd->vdev_children;
506 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
508 for (int c = newc = 0; c < oldc; c++)
509 if (pvd->vdev_child[c])
512 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
514 for (int c = newc = 0; c < oldc; c++) {
515 if ((cvd = pvd->vdev_child[c]) != NULL) {
516 newchild[newc] = cvd;
517 cvd->vdev_id = newc++;
521 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
522 pvd->vdev_child = newchild;
523 pvd->vdev_children = newc;
527 * Allocate and minimally initialize a vdev_t.
530 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
533 vdev_indirect_config_t *vic;
535 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
536 vic = &vd->vdev_indirect_config;
538 if (spa->spa_root_vdev == NULL) {
539 ASSERT(ops == &vdev_root_ops);
540 spa->spa_root_vdev = vd;
541 spa->spa_load_guid = spa_generate_guid(NULL);
544 if (guid == 0 && ops != &vdev_hole_ops) {
545 if (spa->spa_root_vdev == vd) {
547 * The root vdev's guid will also be the pool guid,
548 * which must be unique among all pools.
550 guid = spa_generate_guid(NULL);
553 * Any other vdev's guid must be unique within the pool.
555 guid = spa_generate_guid(spa);
557 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
562 vd->vdev_guid = guid;
563 vd->vdev_guid_sum = guid;
565 vd->vdev_state = VDEV_STATE_CLOSED;
566 vd->vdev_ishole = (ops == &vdev_hole_ops);
567 vic->vic_prev_indirect_vdev = UINT64_MAX;
569 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
570 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
571 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
573 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
574 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
575 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
576 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
577 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
578 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
579 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
580 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
581 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
583 for (int t = 0; t < DTL_TYPES; t++) {
584 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
586 txg_list_create(&vd->vdev_ms_list, spa,
587 offsetof(struct metaslab, ms_txg_node));
588 txg_list_create(&vd->vdev_dtl_list, spa,
589 offsetof(struct vdev, vdev_dtl_node));
590 vd->vdev_stat.vs_timestamp = gethrtime();
598 * Allocate a new vdev. The 'alloctype' is used to control whether we are
599 * creating a new vdev or loading an existing one - the behavior is slightly
600 * different for each case.
603 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
608 uint64_t guid = 0, islog, nparity;
610 vdev_indirect_config_t *vic;
612 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
614 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
615 return (SET_ERROR(EINVAL));
617 if ((ops = vdev_getops(type)) == NULL)
618 return (SET_ERROR(EINVAL));
621 * If this is a load, get the vdev guid from the nvlist.
622 * Otherwise, vdev_alloc_common() will generate one for us.
624 if (alloctype == VDEV_ALLOC_LOAD) {
627 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
629 return (SET_ERROR(EINVAL));
631 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
632 return (SET_ERROR(EINVAL));
633 } else if (alloctype == VDEV_ALLOC_SPARE) {
634 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
635 return (SET_ERROR(EINVAL));
636 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
637 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
638 return (SET_ERROR(EINVAL));
639 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
640 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
641 return (SET_ERROR(EINVAL));
645 * The first allocated vdev must be of type 'root'.
647 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
648 return (SET_ERROR(EINVAL));
651 * Determine whether we're a log vdev.
654 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
655 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
656 return (SET_ERROR(ENOTSUP));
658 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
659 return (SET_ERROR(ENOTSUP));
662 * Set the nparity property for RAID-Z vdevs.
665 if (ops == &vdev_raidz_ops) {
666 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
668 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
669 return (SET_ERROR(EINVAL));
671 * Previous versions could only support 1 or 2 parity
675 spa_version(spa) < SPA_VERSION_RAIDZ2)
676 return (SET_ERROR(ENOTSUP));
678 spa_version(spa) < SPA_VERSION_RAIDZ3)
679 return (SET_ERROR(ENOTSUP));
682 * We require the parity to be specified for SPAs that
683 * support multiple parity levels.
685 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
686 return (SET_ERROR(EINVAL));
688 * Otherwise, we default to 1 parity device for RAID-Z.
695 ASSERT(nparity != -1ULL);
697 vd = vdev_alloc_common(spa, id, guid, ops);
698 vic = &vd->vdev_indirect_config;
700 vd->vdev_islog = islog;
701 vd->vdev_nparity = nparity;
703 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
704 vd->vdev_path = spa_strdup(vd->vdev_path);
705 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
706 vd->vdev_devid = spa_strdup(vd->vdev_devid);
707 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
708 &vd->vdev_physpath) == 0)
709 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
710 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
711 vd->vdev_fru = spa_strdup(vd->vdev_fru);
714 * Set the whole_disk property. If it's not specified, leave the value
717 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
718 &vd->vdev_wholedisk) != 0)
719 vd->vdev_wholedisk = -1ULL;
721 ASSERT0(vic->vic_mapping_object);
722 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
723 &vic->vic_mapping_object);
724 ASSERT0(vic->vic_births_object);
725 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
726 &vic->vic_births_object);
727 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
728 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
729 &vic->vic_prev_indirect_vdev);
732 * Look for the 'not present' flag. This will only be set if the device
733 * was not present at the time of import.
735 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
736 &vd->vdev_not_present);
739 * Get the alignment requirement.
741 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
744 * Retrieve the vdev creation time.
746 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
750 * If we're a top-level vdev, try to load the allocation parameters.
752 if (parent && !parent->vdev_parent &&
753 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
754 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
756 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
758 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
760 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
762 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
765 ASSERT0(vd->vdev_top_zap);
768 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
769 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
770 alloctype == VDEV_ALLOC_ADD ||
771 alloctype == VDEV_ALLOC_SPLIT ||
772 alloctype == VDEV_ALLOC_ROOTPOOL);
773 vd->vdev_mg = metaslab_group_create(islog ?
774 spa_log_class(spa) : spa_normal_class(spa), vd,
775 spa->spa_alloc_count);
778 if (vd->vdev_ops->vdev_op_leaf &&
779 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
780 (void) nvlist_lookup_uint64(nv,
781 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
783 ASSERT0(vd->vdev_leaf_zap);
787 * If we're a leaf vdev, try to load the DTL object and other state.
790 if (vd->vdev_ops->vdev_op_leaf &&
791 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
792 alloctype == VDEV_ALLOC_ROOTPOOL)) {
793 if (alloctype == VDEV_ALLOC_LOAD) {
794 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
795 &vd->vdev_dtl_object);
796 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
800 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
803 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
804 &spare) == 0 && spare)
808 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
811 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
812 &vd->vdev_resilver_txg);
815 * When importing a pool, we want to ignore the persistent fault
816 * state, as the diagnosis made on another system may not be
817 * valid in the current context. Local vdevs will
818 * remain in the faulted state.
820 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
821 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
823 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
825 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
828 if (vd->vdev_faulted || vd->vdev_degraded) {
832 VDEV_AUX_ERR_EXCEEDED;
833 if (nvlist_lookup_string(nv,
834 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
835 strcmp(aux, "external") == 0)
836 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
842 * Add ourselves to the parent's list of children.
844 vdev_add_child(parent, vd);
852 vdev_free(vdev_t *vd)
854 spa_t *spa = vd->vdev_spa;
855 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
858 * Scan queues are normally destroyed at the end of a scan. If the
859 * queue exists here, that implies the vdev is being removed while
860 * the scan is still running.
862 if (vd->vdev_scan_io_queue != NULL) {
863 mutex_enter(&vd->vdev_scan_io_queue_lock);
864 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
865 vd->vdev_scan_io_queue = NULL;
866 mutex_exit(&vd->vdev_scan_io_queue_lock);
870 * vdev_free() implies closing the vdev first. This is simpler than
871 * trying to ensure complicated semantics for all callers.
875 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
876 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
881 for (int c = 0; c < vd->vdev_children; c++)
882 vdev_free(vd->vdev_child[c]);
884 ASSERT(vd->vdev_child == NULL);
885 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
886 ASSERT(vd->vdev_initialize_thread == NULL);
889 * Discard allocation state.
891 if (vd->vdev_mg != NULL) {
892 vdev_metaslab_fini(vd);
893 metaslab_group_destroy(vd->vdev_mg);
896 ASSERT0(vd->vdev_stat.vs_space);
897 ASSERT0(vd->vdev_stat.vs_dspace);
898 ASSERT0(vd->vdev_stat.vs_alloc);
901 * Remove this vdev from its parent's child list.
903 vdev_remove_child(vd->vdev_parent, vd);
905 ASSERT(vd->vdev_parent == NULL);
908 * Clean up vdev structure.
914 spa_strfree(vd->vdev_path);
916 spa_strfree(vd->vdev_devid);
917 if (vd->vdev_physpath)
918 spa_strfree(vd->vdev_physpath);
920 spa_strfree(vd->vdev_fru);
922 if (vd->vdev_isspare)
923 spa_spare_remove(vd);
924 if (vd->vdev_isl2cache)
925 spa_l2cache_remove(vd);
927 txg_list_destroy(&vd->vdev_ms_list);
928 txg_list_destroy(&vd->vdev_dtl_list);
930 mutex_enter(&vd->vdev_dtl_lock);
931 space_map_close(vd->vdev_dtl_sm);
932 for (int t = 0; t < DTL_TYPES; t++) {
933 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
934 range_tree_destroy(vd->vdev_dtl[t]);
936 mutex_exit(&vd->vdev_dtl_lock);
938 EQUIV(vd->vdev_indirect_births != NULL,
939 vd->vdev_indirect_mapping != NULL);
940 if (vd->vdev_indirect_births != NULL) {
941 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
942 vdev_indirect_births_close(vd->vdev_indirect_births);
945 if (vd->vdev_obsolete_sm != NULL) {
946 ASSERT(vd->vdev_removing ||
947 vd->vdev_ops == &vdev_indirect_ops);
948 space_map_close(vd->vdev_obsolete_sm);
949 vd->vdev_obsolete_sm = NULL;
951 range_tree_destroy(vd->vdev_obsolete_segments);
952 rw_destroy(&vd->vdev_indirect_rwlock);
953 mutex_destroy(&vd->vdev_obsolete_lock);
955 mutex_destroy(&vd->vdev_queue_lock);
956 mutex_destroy(&vd->vdev_dtl_lock);
957 mutex_destroy(&vd->vdev_stat_lock);
958 mutex_destroy(&vd->vdev_probe_lock);
959 mutex_destroy(&vd->vdev_scan_io_queue_lock);
960 mutex_destroy(&vd->vdev_initialize_lock);
961 mutex_destroy(&vd->vdev_initialize_io_lock);
962 cv_destroy(&vd->vdev_initialize_io_cv);
963 cv_destroy(&vd->vdev_initialize_cv);
965 if (vd == spa->spa_root_vdev)
966 spa->spa_root_vdev = NULL;
968 kmem_free(vd, sizeof (vdev_t));
972 * Transfer top-level vdev state from svd to tvd.
975 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
977 spa_t *spa = svd->vdev_spa;
982 ASSERT(tvd == tvd->vdev_top);
984 tvd->vdev_ms_array = svd->vdev_ms_array;
985 tvd->vdev_ms_shift = svd->vdev_ms_shift;
986 tvd->vdev_ms_count = svd->vdev_ms_count;
987 tvd->vdev_top_zap = svd->vdev_top_zap;
989 svd->vdev_ms_array = 0;
990 svd->vdev_ms_shift = 0;
991 svd->vdev_ms_count = 0;
992 svd->vdev_top_zap = 0;
995 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
996 tvd->vdev_mg = svd->vdev_mg;
997 tvd->vdev_ms = svd->vdev_ms;
1000 svd->vdev_ms = NULL;
1002 if (tvd->vdev_mg != NULL)
1003 tvd->vdev_mg->mg_vd = tvd;
1005 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1006 svd->vdev_checkpoint_sm = NULL;
1008 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1009 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1010 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1012 svd->vdev_stat.vs_alloc = 0;
1013 svd->vdev_stat.vs_space = 0;
1014 svd->vdev_stat.vs_dspace = 0;
1017 * State which may be set on a top-level vdev that's in the
1018 * process of being removed.
1020 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1021 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1022 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1023 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1024 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1025 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1026 ASSERT0(tvd->vdev_removing);
1027 tvd->vdev_removing = svd->vdev_removing;
1028 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1029 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1030 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1031 range_tree_swap(&svd->vdev_obsolete_segments,
1032 &tvd->vdev_obsolete_segments);
1033 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1034 svd->vdev_indirect_config.vic_mapping_object = 0;
1035 svd->vdev_indirect_config.vic_births_object = 0;
1036 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1037 svd->vdev_indirect_mapping = NULL;
1038 svd->vdev_indirect_births = NULL;
1039 svd->vdev_obsolete_sm = NULL;
1040 svd->vdev_removing = 0;
1042 for (t = 0; t < TXG_SIZE; t++) {
1043 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1044 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1045 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1046 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1047 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1048 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1051 if (list_link_active(&svd->vdev_config_dirty_node)) {
1052 vdev_config_clean(svd);
1053 vdev_config_dirty(tvd);
1056 if (list_link_active(&svd->vdev_state_dirty_node)) {
1057 vdev_state_clean(svd);
1058 vdev_state_dirty(tvd);
1061 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1062 svd->vdev_deflate_ratio = 0;
1064 tvd->vdev_islog = svd->vdev_islog;
1065 svd->vdev_islog = 0;
1067 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1071 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1078 for (int c = 0; c < vd->vdev_children; c++)
1079 vdev_top_update(tvd, vd->vdev_child[c]);
1083 * Add a mirror/replacing vdev above an existing vdev.
1086 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1088 spa_t *spa = cvd->vdev_spa;
1089 vdev_t *pvd = cvd->vdev_parent;
1092 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1094 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1096 mvd->vdev_asize = cvd->vdev_asize;
1097 mvd->vdev_min_asize = cvd->vdev_min_asize;
1098 mvd->vdev_max_asize = cvd->vdev_max_asize;
1099 mvd->vdev_psize = cvd->vdev_psize;
1100 mvd->vdev_ashift = cvd->vdev_ashift;
1101 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1102 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1103 mvd->vdev_state = cvd->vdev_state;
1104 mvd->vdev_crtxg = cvd->vdev_crtxg;
1106 vdev_remove_child(pvd, cvd);
1107 vdev_add_child(pvd, mvd);
1108 cvd->vdev_id = mvd->vdev_children;
1109 vdev_add_child(mvd, cvd);
1110 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1112 if (mvd == mvd->vdev_top)
1113 vdev_top_transfer(cvd, mvd);
1119 * Remove a 1-way mirror/replacing vdev from the tree.
1122 vdev_remove_parent(vdev_t *cvd)
1124 vdev_t *mvd = cvd->vdev_parent;
1125 vdev_t *pvd = mvd->vdev_parent;
1127 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1129 ASSERT(mvd->vdev_children == 1);
1130 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1131 mvd->vdev_ops == &vdev_replacing_ops ||
1132 mvd->vdev_ops == &vdev_spare_ops);
1133 cvd->vdev_ashift = mvd->vdev_ashift;
1134 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1135 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1137 vdev_remove_child(mvd, cvd);
1138 vdev_remove_child(pvd, mvd);
1141 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1142 * Otherwise, we could have detached an offline device, and when we
1143 * go to import the pool we'll think we have two top-level vdevs,
1144 * instead of a different version of the same top-level vdev.
1146 if (mvd->vdev_top == mvd) {
1147 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1148 cvd->vdev_orig_guid = cvd->vdev_guid;
1149 cvd->vdev_guid += guid_delta;
1150 cvd->vdev_guid_sum += guid_delta;
1152 cvd->vdev_id = mvd->vdev_id;
1153 vdev_add_child(pvd, cvd);
1154 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1156 if (cvd == cvd->vdev_top)
1157 vdev_top_transfer(mvd, cvd);
1159 ASSERT(mvd->vdev_children == 0);
1164 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1166 spa_t *spa = vd->vdev_spa;
1167 objset_t *mos = spa->spa_meta_objset;
1169 uint64_t oldc = vd->vdev_ms_count;
1170 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1174 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1177 * This vdev is not being allocated from yet or is a hole.
1179 if (vd->vdev_ms_shift == 0)
1182 ASSERT(!vd->vdev_ishole);
1184 ASSERT(oldc <= newc);
1186 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1189 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1190 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1194 vd->vdev_ms_count = newc;
1195 for (m = oldc; m < newc; m++) {
1196 uint64_t object = 0;
1199 * vdev_ms_array may be 0 if we are creating the "fake"
1200 * metaslabs for an indirect vdev for zdb's leak detection.
1201 * See zdb_leak_init().
1203 if (txg == 0 && vd->vdev_ms_array != 0) {
1204 error = dmu_read(mos, vd->vdev_ms_array,
1205 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1208 vdev_dbgmsg(vd, "unable to read the metaslab "
1209 "array [error=%d]", error);
1214 error = metaslab_init(vd->vdev_mg, m, object, txg,
1217 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1224 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1227 * If the vdev is being removed we don't activate
1228 * the metaslabs since we want to ensure that no new
1229 * allocations are performed on this device.
1231 if (oldc == 0 && !vd->vdev_removing)
1232 metaslab_group_activate(vd->vdev_mg);
1235 spa_config_exit(spa, SCL_ALLOC, FTAG);
1241 vdev_metaslab_fini(vdev_t *vd)
1243 if (vd->vdev_checkpoint_sm != NULL) {
1244 ASSERT(spa_feature_is_active(vd->vdev_spa,
1245 SPA_FEATURE_POOL_CHECKPOINT));
1246 space_map_close(vd->vdev_checkpoint_sm);
1248 * Even though we close the space map, we need to set its
1249 * pointer to NULL. The reason is that vdev_metaslab_fini()
1250 * may be called multiple times for certain operations
1251 * (i.e. when destroying a pool) so we need to ensure that
1252 * this clause never executes twice. This logic is similar
1253 * to the one used for the vdev_ms clause below.
1255 vd->vdev_checkpoint_sm = NULL;
1258 if (vd->vdev_ms != NULL) {
1259 uint64_t count = vd->vdev_ms_count;
1261 metaslab_group_passivate(vd->vdev_mg);
1262 for (uint64_t m = 0; m < count; m++) {
1263 metaslab_t *msp = vd->vdev_ms[m];
1268 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1271 vd->vdev_ms_count = 0;
1273 ASSERT0(vd->vdev_ms_count);
1276 typedef struct vdev_probe_stats {
1277 boolean_t vps_readable;
1278 boolean_t vps_writeable;
1280 } vdev_probe_stats_t;
1283 vdev_probe_done(zio_t *zio)
1285 spa_t *spa = zio->io_spa;
1286 vdev_t *vd = zio->io_vd;
1287 vdev_probe_stats_t *vps = zio->io_private;
1289 ASSERT(vd->vdev_probe_zio != NULL);
1291 if (zio->io_type == ZIO_TYPE_READ) {
1292 if (zio->io_error == 0)
1293 vps->vps_readable = 1;
1294 if (zio->io_error == 0 && spa_writeable(spa)) {
1295 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1296 zio->io_offset, zio->io_size, zio->io_abd,
1297 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1298 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1300 abd_free(zio->io_abd);
1302 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1303 if (zio->io_error == 0)
1304 vps->vps_writeable = 1;
1305 abd_free(zio->io_abd);
1306 } else if (zio->io_type == ZIO_TYPE_NULL) {
1309 vd->vdev_cant_read |= !vps->vps_readable;
1310 vd->vdev_cant_write |= !vps->vps_writeable;
1312 if (vdev_readable(vd) &&
1313 (vdev_writeable(vd) || !spa_writeable(spa))) {
1316 ASSERT(zio->io_error != 0);
1317 vdev_dbgmsg(vd, "failed probe");
1318 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1319 spa, vd, NULL, 0, 0);
1320 zio->io_error = SET_ERROR(ENXIO);
1323 mutex_enter(&vd->vdev_probe_lock);
1324 ASSERT(vd->vdev_probe_zio == zio);
1325 vd->vdev_probe_zio = NULL;
1326 mutex_exit(&vd->vdev_probe_lock);
1328 zio_link_t *zl = NULL;
1329 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1330 if (!vdev_accessible(vd, pio))
1331 pio->io_error = SET_ERROR(ENXIO);
1333 kmem_free(vps, sizeof (*vps));
1338 * Determine whether this device is accessible.
1340 * Read and write to several known locations: the pad regions of each
1341 * vdev label but the first, which we leave alone in case it contains
1345 vdev_probe(vdev_t *vd, zio_t *zio)
1347 spa_t *spa = vd->vdev_spa;
1348 vdev_probe_stats_t *vps = NULL;
1351 ASSERT(vd->vdev_ops->vdev_op_leaf);
1354 * Don't probe the probe.
1356 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1360 * To prevent 'probe storms' when a device fails, we create
1361 * just one probe i/o at a time. All zios that want to probe
1362 * this vdev will become parents of the probe io.
1364 mutex_enter(&vd->vdev_probe_lock);
1366 if ((pio = vd->vdev_probe_zio) == NULL) {
1367 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1369 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1370 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1373 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1375 * vdev_cant_read and vdev_cant_write can only
1376 * transition from TRUE to FALSE when we have the
1377 * SCL_ZIO lock as writer; otherwise they can only
1378 * transition from FALSE to TRUE. This ensures that
1379 * any zio looking at these values can assume that
1380 * failures persist for the life of the I/O. That's
1381 * important because when a device has intermittent
1382 * connectivity problems, we want to ensure that
1383 * they're ascribed to the device (ENXIO) and not
1386 * Since we hold SCL_ZIO as writer here, clear both
1387 * values so the probe can reevaluate from first
1390 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1391 vd->vdev_cant_read = B_FALSE;
1392 vd->vdev_cant_write = B_FALSE;
1395 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1396 vdev_probe_done, vps,
1397 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1400 * We can't change the vdev state in this context, so we
1401 * kick off an async task to do it on our behalf.
1404 vd->vdev_probe_wanted = B_TRUE;
1405 spa_async_request(spa, SPA_ASYNC_PROBE);
1410 zio_add_child(zio, pio);
1412 mutex_exit(&vd->vdev_probe_lock);
1415 ASSERT(zio != NULL);
1419 for (int l = 1; l < VDEV_LABELS; l++) {
1420 zio_nowait(zio_read_phys(pio, vd,
1421 vdev_label_offset(vd->vdev_psize, l,
1422 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1423 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1424 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1425 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1436 vdev_open_child(void *arg)
1440 vd->vdev_open_thread = curthread;
1441 vd->vdev_open_error = vdev_open(vd);
1442 vd->vdev_open_thread = NULL;
1446 vdev_uses_zvols(vdev_t *vd)
1448 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1449 strlen(ZVOL_DIR)) == 0)
1451 for (int c = 0; c < vd->vdev_children; c++)
1452 if (vdev_uses_zvols(vd->vdev_child[c]))
1458 vdev_open_children(vdev_t *vd)
1461 int children = vd->vdev_children;
1464 * in order to handle pools on top of zvols, do the opens
1465 * in a single thread so that the same thread holds the
1466 * spa_namespace_lock
1468 if (B_TRUE || vdev_uses_zvols(vd)) {
1469 for (int c = 0; c < children; c++)
1470 vd->vdev_child[c]->vdev_open_error =
1471 vdev_open(vd->vdev_child[c]);
1474 tq = taskq_create("vdev_open", children, minclsyspri,
1475 children, children, TASKQ_PREPOPULATE);
1477 for (int c = 0; c < children; c++)
1478 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1485 * Compute the raidz-deflation ratio. Note, we hard-code
1486 * in 128k (1 << 17) because it is the "typical" blocksize.
1487 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1488 * otherwise it would inconsistently account for existing bp's.
1491 vdev_set_deflate_ratio(vdev_t *vd)
1493 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1494 vd->vdev_deflate_ratio = (1 << 17) /
1495 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1500 * Prepare a virtual device for access.
1503 vdev_open(vdev_t *vd)
1505 spa_t *spa = vd->vdev_spa;
1508 uint64_t max_osize = 0;
1509 uint64_t asize, max_asize, psize;
1510 uint64_t logical_ashift = 0;
1511 uint64_t physical_ashift = 0;
1513 ASSERT(vd->vdev_open_thread == curthread ||
1514 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1515 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1516 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1517 vd->vdev_state == VDEV_STATE_OFFLINE);
1519 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1520 vd->vdev_cant_read = B_FALSE;
1521 vd->vdev_cant_write = B_FALSE;
1522 vd->vdev_notrim = B_FALSE;
1523 vd->vdev_min_asize = vdev_get_min_asize(vd);
1526 * If this vdev is not removed, check its fault status. If it's
1527 * faulted, bail out of the open.
1529 if (!vd->vdev_removed && vd->vdev_faulted) {
1530 ASSERT(vd->vdev_children == 0);
1531 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1532 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1533 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1534 vd->vdev_label_aux);
1535 return (SET_ERROR(ENXIO));
1536 } else if (vd->vdev_offline) {
1537 ASSERT(vd->vdev_children == 0);
1538 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1539 return (SET_ERROR(ENXIO));
1542 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1543 &logical_ashift, &physical_ashift);
1546 * Reset the vdev_reopening flag so that we actually close
1547 * the vdev on error.
1549 vd->vdev_reopening = B_FALSE;
1550 if (zio_injection_enabled && error == 0)
1551 error = zio_handle_device_injection(vd, NULL, ENXIO);
1554 if (vd->vdev_removed &&
1555 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1556 vd->vdev_removed = B_FALSE;
1558 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1559 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1560 vd->vdev_stat.vs_aux);
1562 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1563 vd->vdev_stat.vs_aux);
1568 vd->vdev_removed = B_FALSE;
1571 * Recheck the faulted flag now that we have confirmed that
1572 * the vdev is accessible. If we're faulted, bail.
1574 if (vd->vdev_faulted) {
1575 ASSERT(vd->vdev_children == 0);
1576 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1577 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1578 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1579 vd->vdev_label_aux);
1580 return (SET_ERROR(ENXIO));
1583 if (vd->vdev_degraded) {
1584 ASSERT(vd->vdev_children == 0);
1585 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1586 VDEV_AUX_ERR_EXCEEDED);
1588 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1592 * For hole or missing vdevs we just return success.
1594 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1597 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1598 trim_map_create(vd);
1600 for (int c = 0; c < vd->vdev_children; c++) {
1601 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1602 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1608 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1609 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1611 if (vd->vdev_children == 0) {
1612 if (osize < SPA_MINDEVSIZE) {
1613 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1614 VDEV_AUX_TOO_SMALL);
1615 return (SET_ERROR(EOVERFLOW));
1618 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1619 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1620 VDEV_LABEL_END_SIZE);
1622 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1623 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1624 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1625 VDEV_AUX_TOO_SMALL);
1626 return (SET_ERROR(EOVERFLOW));
1630 max_asize = max_osize;
1633 vd->vdev_psize = psize;
1636 * Make sure the allocatable size hasn't shrunk too much.
1638 if (asize < vd->vdev_min_asize) {
1639 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1640 VDEV_AUX_BAD_LABEL);
1641 return (SET_ERROR(EINVAL));
1644 vd->vdev_physical_ashift =
1645 MAX(physical_ashift, vd->vdev_physical_ashift);
1646 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1647 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1649 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1650 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1651 VDEV_AUX_ASHIFT_TOO_BIG);
1655 if (vd->vdev_asize == 0) {
1657 * This is the first-ever open, so use the computed values.
1658 * For testing purposes, a higher ashift can be requested.
1660 vd->vdev_asize = asize;
1661 vd->vdev_max_asize = max_asize;
1664 * Make sure the alignment requirement hasn't increased.
1666 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1667 vd->vdev_ops->vdev_op_leaf) {
1668 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1669 VDEV_AUX_BAD_LABEL);
1672 vd->vdev_max_asize = max_asize;
1676 * If all children are healthy we update asize if either:
1677 * The asize has increased, due to a device expansion caused by dynamic
1678 * LUN growth or vdev replacement, and automatic expansion is enabled;
1679 * making the additional space available.
1681 * The asize has decreased, due to a device shrink usually caused by a
1682 * vdev replace with a smaller device. This ensures that calculations
1683 * based of max_asize and asize e.g. esize are always valid. It's safe
1684 * to do this as we've already validated that asize is greater than
1687 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1688 ((asize > vd->vdev_asize &&
1689 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1690 (asize < vd->vdev_asize)))
1691 vd->vdev_asize = asize;
1693 vdev_set_min_asize(vd);
1696 * Ensure we can issue some IO before declaring the
1697 * vdev open for business.
1699 if (vd->vdev_ops->vdev_op_leaf &&
1700 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1701 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1702 VDEV_AUX_ERR_EXCEEDED);
1707 * Track the min and max ashift values for normal data devices.
1709 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1710 !vd->vdev_islog && vd->vdev_aux == NULL) {
1711 if (vd->vdev_ashift > spa->spa_max_ashift)
1712 spa->spa_max_ashift = vd->vdev_ashift;
1713 if (vd->vdev_ashift < spa->spa_min_ashift)
1714 spa->spa_min_ashift = vd->vdev_ashift;
1718 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1719 * resilver. But don't do this if we are doing a reopen for a scrub,
1720 * since this would just restart the scrub we are already doing.
1722 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1723 vdev_resilver_needed(vd, NULL, NULL))
1724 spa_async_request(spa, SPA_ASYNC_RESILVER);
1730 * Called once the vdevs are all opened, this routine validates the label
1731 * contents. This needs to be done before vdev_load() so that we don't
1732 * inadvertently do repair I/Os to the wrong device.
1734 * This function will only return failure if one of the vdevs indicates that it
1735 * has since been destroyed or exported. This is only possible if
1736 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1737 * will be updated but the function will return 0.
1740 vdev_validate(vdev_t *vd)
1742 spa_t *spa = vd->vdev_spa;
1744 uint64_t guid = 0, aux_guid = 0, top_guid;
1749 if (vdev_validate_skip)
1752 for (uint64_t c = 0; c < vd->vdev_children; c++)
1753 if (vdev_validate(vd->vdev_child[c]) != 0)
1754 return (SET_ERROR(EBADF));
1757 * If the device has already failed, or was marked offline, don't do
1758 * any further validation. Otherwise, label I/O will fail and we will
1759 * overwrite the previous state.
1761 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1765 * If we are performing an extreme rewind, we allow for a label that
1766 * was modified at a point after the current txg.
1767 * If config lock is not held do not check for the txg. spa_sync could
1768 * be updating the vdev's label before updating spa_last_synced_txg.
1770 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1771 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1774 txg = spa_last_synced_txg(spa);
1776 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1777 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1778 VDEV_AUX_BAD_LABEL);
1779 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
1780 "txg %llu", (u_longlong_t)txg);
1785 * Determine if this vdev has been split off into another
1786 * pool. If so, then refuse to open it.
1788 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1789 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1790 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1791 VDEV_AUX_SPLIT_POOL);
1793 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1797 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1798 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1799 VDEV_AUX_CORRUPT_DATA);
1801 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1802 ZPOOL_CONFIG_POOL_GUID);
1807 * If config is not trusted then ignore the spa guid check. This is
1808 * necessary because if the machine crashed during a re-guid the new
1809 * guid might have been written to all of the vdev labels, but not the
1810 * cached config. The check will be performed again once we have the
1811 * trusted config from the MOS.
1813 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1814 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1815 VDEV_AUX_CORRUPT_DATA);
1817 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
1818 "match config (%llu != %llu)", (u_longlong_t)guid,
1819 (u_longlong_t)spa_guid(spa));
1823 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1824 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1828 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
1829 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1830 VDEV_AUX_CORRUPT_DATA);
1832 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1837 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
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_TOP_GUID);
1848 * If this vdev just became a top-level vdev because its sibling was
1849 * detached, it will have adopted the parent's vdev guid -- but the
1850 * label may or may not be on disk yet. Fortunately, either version
1851 * of the label will have the same top guid, so if we're a top-level
1852 * vdev, we can safely compare to that instead.
1853 * However, if the config comes from a cachefile that failed to update
1854 * after the detach, a top-level vdev will appear as a non top-level
1855 * vdev in the config. Also relax the constraints if we perform an
1858 * If we split this vdev off instead, then we also check the
1859 * original pool's guid. We don't want to consider the vdev
1860 * corrupt if it is partway through a split operation.
1862 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
1863 boolean_t mismatch = B_FALSE;
1864 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
1865 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
1868 if (vd->vdev_guid != top_guid &&
1869 vd->vdev_top->vdev_guid != guid)
1874 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1875 VDEV_AUX_CORRUPT_DATA);
1877 vdev_dbgmsg(vd, "vdev_validate: config guid "
1878 "doesn't match label guid");
1879 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
1880 (u_longlong_t)vd->vdev_guid,
1881 (u_longlong_t)vd->vdev_top->vdev_guid);
1882 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
1883 "aux_guid %llu", (u_longlong_t)guid,
1884 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
1889 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1891 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1892 VDEV_AUX_CORRUPT_DATA);
1894 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1895 ZPOOL_CONFIG_POOL_STATE);
1902 * If this is a verbatim import, no need to check the
1903 * state of the pool.
1905 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1906 spa_load_state(spa) == SPA_LOAD_OPEN &&
1907 state != POOL_STATE_ACTIVE) {
1908 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
1909 "for spa %s", (u_longlong_t)state, spa->spa_name);
1910 return (SET_ERROR(EBADF));
1914 * If we were able to open and validate a vdev that was
1915 * previously marked permanently unavailable, clear that state
1918 if (vd->vdev_not_present)
1919 vd->vdev_not_present = 0;
1925 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
1927 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
1928 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
1929 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1930 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
1931 dvd->vdev_path, svd->vdev_path);
1932 spa_strfree(dvd->vdev_path);
1933 dvd->vdev_path = spa_strdup(svd->vdev_path);
1935 } else if (svd->vdev_path != NULL) {
1936 dvd->vdev_path = spa_strdup(svd->vdev_path);
1937 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1938 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
1943 * Recursively copy vdev paths from one vdev to another. Source and destination
1944 * vdev trees must have same geometry otherwise return error. Intended to copy
1945 * paths from userland config into MOS config.
1948 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
1950 if ((svd->vdev_ops == &vdev_missing_ops) ||
1951 (svd->vdev_ishole && dvd->vdev_ishole) ||
1952 (dvd->vdev_ops == &vdev_indirect_ops))
1955 if (svd->vdev_ops != dvd->vdev_ops) {
1956 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
1957 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
1958 return (SET_ERROR(EINVAL));
1961 if (svd->vdev_guid != dvd->vdev_guid) {
1962 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
1963 "%llu)", (u_longlong_t)svd->vdev_guid,
1964 (u_longlong_t)dvd->vdev_guid);
1965 return (SET_ERROR(EINVAL));
1968 if (svd->vdev_children != dvd->vdev_children) {
1969 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
1970 "%llu != %llu", (u_longlong_t)svd->vdev_children,
1971 (u_longlong_t)dvd->vdev_children);
1972 return (SET_ERROR(EINVAL));
1975 for (uint64_t i = 0; i < svd->vdev_children; i++) {
1976 int error = vdev_copy_path_strict(svd->vdev_child[i],
1977 dvd->vdev_child[i]);
1982 if (svd->vdev_ops->vdev_op_leaf)
1983 vdev_copy_path_impl(svd, dvd);
1989 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
1991 ASSERT(stvd->vdev_top == stvd);
1992 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
1994 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
1995 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
1998 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2002 * The idea here is that while a vdev can shift positions within
2003 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2004 * step outside of it.
2006 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2008 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2011 ASSERT(vd->vdev_ops->vdev_op_leaf);
2013 vdev_copy_path_impl(vd, dvd);
2017 * Recursively copy vdev paths from one root vdev to another. Source and
2018 * destination vdev trees may differ in geometry. For each destination leaf
2019 * vdev, search a vdev with the same guid and top vdev id in the source.
2020 * Intended to copy paths from userland config into MOS config.
2023 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2025 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2026 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2027 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2029 for (uint64_t i = 0; i < children; i++) {
2030 vdev_copy_path_search(srvd->vdev_child[i],
2031 drvd->vdev_child[i]);
2036 * Close a virtual device.
2039 vdev_close(vdev_t *vd)
2041 spa_t *spa = vd->vdev_spa;
2042 vdev_t *pvd = vd->vdev_parent;
2044 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2047 * If our parent is reopening, then we are as well, unless we are
2050 if (pvd != NULL && pvd->vdev_reopening)
2051 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2053 vd->vdev_ops->vdev_op_close(vd);
2055 vdev_cache_purge(vd);
2057 if (vd->vdev_ops->vdev_op_leaf)
2058 trim_map_destroy(vd);
2061 * We record the previous state before we close it, so that if we are
2062 * doing a reopen(), we don't generate FMA ereports if we notice that
2063 * it's still faulted.
2065 vd->vdev_prevstate = vd->vdev_state;
2067 if (vd->vdev_offline)
2068 vd->vdev_state = VDEV_STATE_OFFLINE;
2070 vd->vdev_state = VDEV_STATE_CLOSED;
2071 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2075 vdev_hold(vdev_t *vd)
2077 spa_t *spa = vd->vdev_spa;
2079 ASSERT(spa_is_root(spa));
2080 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2083 for (int c = 0; c < vd->vdev_children; c++)
2084 vdev_hold(vd->vdev_child[c]);
2086 if (vd->vdev_ops->vdev_op_leaf)
2087 vd->vdev_ops->vdev_op_hold(vd);
2091 vdev_rele(vdev_t *vd)
2093 spa_t *spa = vd->vdev_spa;
2095 ASSERT(spa_is_root(spa));
2096 for (int c = 0; c < vd->vdev_children; c++)
2097 vdev_rele(vd->vdev_child[c]);
2099 if (vd->vdev_ops->vdev_op_leaf)
2100 vd->vdev_ops->vdev_op_rele(vd);
2104 * Reopen all interior vdevs and any unopened leaves. We don't actually
2105 * reopen leaf vdevs which had previously been opened as they might deadlock
2106 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2107 * If the leaf has never been opened then open it, as usual.
2110 vdev_reopen(vdev_t *vd)
2112 spa_t *spa = vd->vdev_spa;
2114 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2116 /* set the reopening flag unless we're taking the vdev offline */
2117 vd->vdev_reopening = !vd->vdev_offline;
2119 (void) vdev_open(vd);
2122 * Call vdev_validate() here to make sure we have the same device.
2123 * Otherwise, a device with an invalid label could be successfully
2124 * opened in response to vdev_reopen().
2127 (void) vdev_validate_aux(vd);
2128 if (vdev_readable(vd) && vdev_writeable(vd) &&
2129 vd->vdev_aux == &spa->spa_l2cache &&
2130 !l2arc_vdev_present(vd))
2131 l2arc_add_vdev(spa, vd);
2133 (void) vdev_validate(vd);
2137 * Reassess parent vdev's health.
2139 vdev_propagate_state(vd);
2143 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2148 * Normally, partial opens (e.g. of a mirror) are allowed.
2149 * For a create, however, we want to fail the request if
2150 * there are any components we can't open.
2152 error = vdev_open(vd);
2154 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2156 return (error ? error : ENXIO);
2160 * Recursively load DTLs and initialize all labels.
2162 if ((error = vdev_dtl_load(vd)) != 0 ||
2163 (error = vdev_label_init(vd, txg, isreplacing ?
2164 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2173 vdev_metaslab_set_size(vdev_t *vd)
2175 uint64_t asize = vd->vdev_asize;
2176 uint64_t ms_count = asize >> vdev_default_ms_shift;
2180 * There are two dimensions to the metaslab sizing calculation:
2181 * the size of the metaslab and the count of metaslabs per vdev.
2182 * In general, we aim for vdev_max_ms_count (200) metaslabs. The
2183 * range of the dimensions are as follows:
2185 * 2^29 <= ms_size <= 2^38
2186 * 16 <= ms_count <= 131,072
2188 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2189 * at least 512MB (2^29) to minimize fragmentation effects when
2190 * testing with smaller devices. However, the count constraint
2191 * of at least 16 metaslabs will override this minimum size goal.
2193 * On the upper end of vdev sizes, we aim for a maximum metaslab
2194 * size of 256GB. However, we will cap the total count to 2^17
2195 * metaslabs to keep our memory footprint in check.
2197 * The net effect of applying above constrains is summarized below.
2199 * vdev size metaslab count
2200 * -------------|-----------------
2202 * 8GB - 100GB one per 512MB
2204 * 50TB - 32PB one per 256GB
2206 * -------------------------------
2209 if (ms_count < vdev_min_ms_count)
2210 ms_shift = highbit64(asize / vdev_min_ms_count);
2211 else if (ms_count > vdev_max_ms_count)
2212 ms_shift = highbit64(asize / vdev_max_ms_count);
2214 ms_shift = vdev_default_ms_shift;
2216 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2217 ms_shift = SPA_MAXBLOCKSHIFT;
2218 } else if (ms_shift > vdev_max_ms_shift) {
2219 ms_shift = vdev_max_ms_shift;
2220 /* cap the total count to constrain memory footprint */
2221 if ((asize >> ms_shift) > vdev_ms_count_limit)
2222 ms_shift = highbit64(asize / vdev_ms_count_limit);
2225 vd->vdev_ms_shift = ms_shift;
2226 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2230 * Maximize performance by inflating the configured ashift for top level
2231 * vdevs to be as close to the physical ashift as possible while maintaining
2232 * administrator defined limits and ensuring it doesn't go below the
2236 vdev_ashift_optimize(vdev_t *vd)
2238 if (vd == vd->vdev_top) {
2239 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
2240 vd->vdev_ashift = MIN(
2241 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
2242 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
2245 * Unusual case where logical ashift > physical ashift
2246 * so we can't cap the calculated ashift based on max
2247 * ashift as that would cause failures.
2248 * We still check if we need to increase it to match
2251 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
2258 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2260 ASSERT(vd == vd->vdev_top);
2261 /* indirect vdevs don't have metaslabs or dtls */
2262 ASSERT(vdev_is_concrete(vd) || flags == 0);
2263 ASSERT(ISP2(flags));
2264 ASSERT(spa_writeable(vd->vdev_spa));
2266 if (flags & VDD_METASLAB)
2267 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2269 if (flags & VDD_DTL)
2270 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2272 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2276 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2278 for (int c = 0; c < vd->vdev_children; c++)
2279 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2281 if (vd->vdev_ops->vdev_op_leaf)
2282 vdev_dirty(vd->vdev_top, flags, vd, txg);
2288 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2289 * the vdev has less than perfect replication. There are four kinds of DTL:
2291 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2293 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2295 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2296 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2297 * txgs that was scrubbed.
2299 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2300 * persistent errors or just some device being offline.
2301 * Unlike the other three, the DTL_OUTAGE map is not generally
2302 * maintained; it's only computed when needed, typically to
2303 * determine whether a device can be detached.
2305 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2306 * either has the data or it doesn't.
2308 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2309 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2310 * if any child is less than fully replicated, then so is its parent.
2311 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2312 * comprising only those txgs which appear in 'maxfaults' or more children;
2313 * those are the txgs we don't have enough replication to read. For example,
2314 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2315 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2316 * two child DTL_MISSING maps.
2318 * It should be clear from the above that to compute the DTLs and outage maps
2319 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2320 * Therefore, that is all we keep on disk. When loading the pool, or after
2321 * a configuration change, we generate all other DTLs from first principles.
2324 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2326 range_tree_t *rt = vd->vdev_dtl[t];
2328 ASSERT(t < DTL_TYPES);
2329 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2330 ASSERT(spa_writeable(vd->vdev_spa));
2332 mutex_enter(&vd->vdev_dtl_lock);
2333 if (!range_tree_contains(rt, txg, size))
2334 range_tree_add(rt, txg, size);
2335 mutex_exit(&vd->vdev_dtl_lock);
2339 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2341 range_tree_t *rt = vd->vdev_dtl[t];
2342 boolean_t dirty = B_FALSE;
2344 ASSERT(t < DTL_TYPES);
2345 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2348 * While we are loading the pool, the DTLs have not been loaded yet.
2349 * Ignore the DTLs and try all devices. This avoids a recursive
2350 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2351 * when loading the pool (relying on the checksum to ensure that
2352 * we get the right data -- note that we while loading, we are
2353 * only reading the MOS, which is always checksummed).
2355 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2358 mutex_enter(&vd->vdev_dtl_lock);
2359 if (!range_tree_is_empty(rt))
2360 dirty = range_tree_contains(rt, txg, size);
2361 mutex_exit(&vd->vdev_dtl_lock);
2367 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2369 range_tree_t *rt = vd->vdev_dtl[t];
2372 mutex_enter(&vd->vdev_dtl_lock);
2373 empty = range_tree_is_empty(rt);
2374 mutex_exit(&vd->vdev_dtl_lock);
2380 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2383 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2385 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2387 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2388 vd->vdev_ops->vdev_op_leaf)
2391 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2395 * Returns the lowest txg in the DTL range.
2398 vdev_dtl_min(vdev_t *vd)
2402 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2403 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2404 ASSERT0(vd->vdev_children);
2406 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2407 return (rs->rs_start - 1);
2411 * Returns the highest txg in the DTL.
2414 vdev_dtl_max(vdev_t *vd)
2418 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2419 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2420 ASSERT0(vd->vdev_children);
2422 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2423 return (rs->rs_end);
2427 * Determine if a resilvering vdev should remove any DTL entries from
2428 * its range. If the vdev was resilvering for the entire duration of the
2429 * scan then it should excise that range from its DTLs. Otherwise, this
2430 * vdev is considered partially resilvered and should leave its DTL
2431 * entries intact. The comment in vdev_dtl_reassess() describes how we
2435 vdev_dtl_should_excise(vdev_t *vd)
2437 spa_t *spa = vd->vdev_spa;
2438 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2440 ASSERT0(scn->scn_phys.scn_errors);
2441 ASSERT0(vd->vdev_children);
2443 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2446 if (vd->vdev_resilver_txg == 0 ||
2447 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2451 * When a resilver is initiated the scan will assign the scn_max_txg
2452 * value to the highest txg value that exists in all DTLs. If this
2453 * device's max DTL is not part of this scan (i.e. it is not in
2454 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2457 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2458 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2459 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2460 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2467 * Reassess DTLs after a config change or scrub completion.
2470 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2472 spa_t *spa = vd->vdev_spa;
2476 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2478 for (int c = 0; c < vd->vdev_children; c++)
2479 vdev_dtl_reassess(vd->vdev_child[c], txg,
2480 scrub_txg, scrub_done);
2482 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2485 if (vd->vdev_ops->vdev_op_leaf) {
2486 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2488 mutex_enter(&vd->vdev_dtl_lock);
2491 * If we've completed a scan cleanly then determine
2492 * if this vdev should remove any DTLs. We only want to
2493 * excise regions on vdevs that were available during
2494 * the entire duration of this scan.
2496 if (scrub_txg != 0 &&
2497 (spa->spa_scrub_started ||
2498 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2499 vdev_dtl_should_excise(vd)) {
2501 * We completed a scrub up to scrub_txg. If we
2502 * did it without rebooting, then the scrub dtl
2503 * will be valid, so excise the old region and
2504 * fold in the scrub dtl. Otherwise, leave the
2505 * dtl as-is if there was an error.
2507 * There's little trick here: to excise the beginning
2508 * of the DTL_MISSING map, we put it into a reference
2509 * tree and then add a segment with refcnt -1 that
2510 * covers the range [0, scrub_txg). This means
2511 * that each txg in that range has refcnt -1 or 0.
2512 * We then add DTL_SCRUB with a refcnt of 2, so that
2513 * entries in the range [0, scrub_txg) will have a
2514 * positive refcnt -- either 1 or 2. We then convert
2515 * the reference tree into the new DTL_MISSING map.
2517 space_reftree_create(&reftree);
2518 space_reftree_add_map(&reftree,
2519 vd->vdev_dtl[DTL_MISSING], 1);
2520 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2521 space_reftree_add_map(&reftree,
2522 vd->vdev_dtl[DTL_SCRUB], 2);
2523 space_reftree_generate_map(&reftree,
2524 vd->vdev_dtl[DTL_MISSING], 1);
2525 space_reftree_destroy(&reftree);
2527 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2528 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2529 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2531 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2532 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2533 if (!vdev_readable(vd))
2534 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2536 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2537 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2540 * If the vdev was resilvering and no longer has any
2541 * DTLs then reset its resilvering flag and dirty
2542 * the top level so that we persist the change.
2544 if (vd->vdev_resilver_txg != 0 &&
2545 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2546 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2547 vd->vdev_resilver_txg = 0;
2548 vdev_config_dirty(vd->vdev_top);
2551 mutex_exit(&vd->vdev_dtl_lock);
2554 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2558 mutex_enter(&vd->vdev_dtl_lock);
2559 for (int t = 0; t < DTL_TYPES; t++) {
2560 /* account for child's outage in parent's missing map */
2561 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2563 continue; /* leaf vdevs only */
2564 if (t == DTL_PARTIAL)
2565 minref = 1; /* i.e. non-zero */
2566 else if (vd->vdev_nparity != 0)
2567 minref = vd->vdev_nparity + 1; /* RAID-Z */
2569 minref = vd->vdev_children; /* any kind of mirror */
2570 space_reftree_create(&reftree);
2571 for (int c = 0; c < vd->vdev_children; c++) {
2572 vdev_t *cvd = vd->vdev_child[c];
2573 mutex_enter(&cvd->vdev_dtl_lock);
2574 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2575 mutex_exit(&cvd->vdev_dtl_lock);
2577 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2578 space_reftree_destroy(&reftree);
2580 mutex_exit(&vd->vdev_dtl_lock);
2584 vdev_dtl_load(vdev_t *vd)
2586 spa_t *spa = vd->vdev_spa;
2587 objset_t *mos = spa->spa_meta_objset;
2590 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2591 ASSERT(vdev_is_concrete(vd));
2593 error = space_map_open(&vd->vdev_dtl_sm, mos,
2594 vd->vdev_dtl_object, 0, -1ULL, 0);
2597 ASSERT(vd->vdev_dtl_sm != NULL);
2599 mutex_enter(&vd->vdev_dtl_lock);
2602 * Now that we've opened the space_map we need to update
2605 space_map_update(vd->vdev_dtl_sm);
2607 error = space_map_load(vd->vdev_dtl_sm,
2608 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2609 mutex_exit(&vd->vdev_dtl_lock);
2614 for (int c = 0; c < vd->vdev_children; c++) {
2615 error = vdev_dtl_load(vd->vdev_child[c]);
2624 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2626 spa_t *spa = vd->vdev_spa;
2628 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2629 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2634 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2636 spa_t *spa = vd->vdev_spa;
2637 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2638 DMU_OT_NONE, 0, tx);
2641 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2648 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2650 if (vd->vdev_ops != &vdev_hole_ops &&
2651 vd->vdev_ops != &vdev_missing_ops &&
2652 vd->vdev_ops != &vdev_root_ops &&
2653 !vd->vdev_top->vdev_removing) {
2654 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2655 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2657 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2658 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2661 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2662 vdev_construct_zaps(vd->vdev_child[i], tx);
2667 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2669 spa_t *spa = vd->vdev_spa;
2670 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2671 objset_t *mos = spa->spa_meta_objset;
2672 range_tree_t *rtsync;
2674 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2676 ASSERT(vdev_is_concrete(vd));
2677 ASSERT(vd->vdev_ops->vdev_op_leaf);
2679 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2681 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2682 mutex_enter(&vd->vdev_dtl_lock);
2683 space_map_free(vd->vdev_dtl_sm, tx);
2684 space_map_close(vd->vdev_dtl_sm);
2685 vd->vdev_dtl_sm = NULL;
2686 mutex_exit(&vd->vdev_dtl_lock);
2689 * We only destroy the leaf ZAP for detached leaves or for
2690 * removed log devices. Removed data devices handle leaf ZAP
2691 * cleanup later, once cancellation is no longer possible.
2693 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2694 vd->vdev_top->vdev_islog)) {
2695 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2696 vd->vdev_leaf_zap = 0;
2703 if (vd->vdev_dtl_sm == NULL) {
2704 uint64_t new_object;
2706 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx);
2707 VERIFY3U(new_object, !=, 0);
2709 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2711 ASSERT(vd->vdev_dtl_sm != NULL);
2714 rtsync = range_tree_create(NULL, NULL);
2716 mutex_enter(&vd->vdev_dtl_lock);
2717 range_tree_walk(rt, range_tree_add, rtsync);
2718 mutex_exit(&vd->vdev_dtl_lock);
2720 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx);
2721 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
2722 range_tree_vacate(rtsync, NULL, NULL);
2724 range_tree_destroy(rtsync);
2727 * If the object for the space map has changed then dirty
2728 * the top level so that we update the config.
2730 if (object != space_map_object(vd->vdev_dtl_sm)) {
2731 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2732 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2733 (u_longlong_t)object,
2734 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2735 vdev_config_dirty(vd->vdev_top);
2740 mutex_enter(&vd->vdev_dtl_lock);
2741 space_map_update(vd->vdev_dtl_sm);
2742 mutex_exit(&vd->vdev_dtl_lock);
2746 * Determine whether the specified vdev can be offlined/detached/removed
2747 * without losing data.
2750 vdev_dtl_required(vdev_t *vd)
2752 spa_t *spa = vd->vdev_spa;
2753 vdev_t *tvd = vd->vdev_top;
2754 uint8_t cant_read = vd->vdev_cant_read;
2757 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2759 if (vd == spa->spa_root_vdev || vd == tvd)
2763 * Temporarily mark the device as unreadable, and then determine
2764 * whether this results in any DTL outages in the top-level vdev.
2765 * If not, we can safely offline/detach/remove the device.
2767 vd->vdev_cant_read = B_TRUE;
2768 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2769 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2770 vd->vdev_cant_read = cant_read;
2771 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2773 if (!required && zio_injection_enabled)
2774 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2780 * Determine if resilver is needed, and if so the txg range.
2783 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2785 boolean_t needed = B_FALSE;
2786 uint64_t thismin = UINT64_MAX;
2787 uint64_t thismax = 0;
2789 if (vd->vdev_children == 0) {
2790 mutex_enter(&vd->vdev_dtl_lock);
2791 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2792 vdev_writeable(vd)) {
2794 thismin = vdev_dtl_min(vd);
2795 thismax = vdev_dtl_max(vd);
2798 mutex_exit(&vd->vdev_dtl_lock);
2800 for (int c = 0; c < vd->vdev_children; c++) {
2801 vdev_t *cvd = vd->vdev_child[c];
2802 uint64_t cmin, cmax;
2804 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2805 thismin = MIN(thismin, cmin);
2806 thismax = MAX(thismax, cmax);
2812 if (needed && minp) {
2820 * Gets the checkpoint space map object from the vdev's ZAP.
2821 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2822 * or the ZAP doesn't exist yet.
2825 vdev_checkpoint_sm_object(vdev_t *vd)
2827 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
2828 if (vd->vdev_top_zap == 0) {
2832 uint64_t sm_obj = 0;
2833 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
2834 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj);
2836 ASSERT(err == 0 || err == ENOENT);
2842 vdev_load(vdev_t *vd)
2846 * Recursively load all children.
2848 for (int c = 0; c < vd->vdev_children; c++) {
2849 error = vdev_load(vd->vdev_child[c]);
2855 vdev_set_deflate_ratio(vd);
2858 * If this is a top-level vdev, initialize its metaslabs.
2860 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2861 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2862 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2863 VDEV_AUX_CORRUPT_DATA);
2864 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2865 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2866 (u_longlong_t)vd->vdev_asize);
2867 return (SET_ERROR(ENXIO));
2868 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2869 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2870 "[error=%d]", error);
2871 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2872 VDEV_AUX_CORRUPT_DATA);
2876 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd);
2877 if (checkpoint_sm_obj != 0) {
2878 objset_t *mos = spa_meta_objset(vd->vdev_spa);
2879 ASSERT(vd->vdev_asize != 0);
2880 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
2882 if ((error = space_map_open(&vd->vdev_checkpoint_sm,
2883 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
2884 vd->vdev_ashift))) {
2885 vdev_dbgmsg(vd, "vdev_load: space_map_open "
2886 "failed for checkpoint spacemap (obj %llu) "
2888 (u_longlong_t)checkpoint_sm_obj, error);
2891 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
2892 space_map_update(vd->vdev_checkpoint_sm);
2895 * Since the checkpoint_sm contains free entries
2896 * exclusively we can use sm_alloc to indicate the
2897 * culmulative checkpointed space that has been freed.
2899 vd->vdev_stat.vs_checkpoint_space =
2900 -vd->vdev_checkpoint_sm->sm_alloc;
2901 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
2902 vd->vdev_stat.vs_checkpoint_space;
2907 * If this is a leaf vdev, load its DTL.
2909 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2910 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2911 VDEV_AUX_CORRUPT_DATA);
2912 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
2913 "[error=%d]", error);
2917 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2918 if (obsolete_sm_object != 0) {
2919 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2920 ASSERT(vd->vdev_asize != 0);
2921 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
2923 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2924 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2925 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2926 VDEV_AUX_CORRUPT_DATA);
2927 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
2928 "obsolete spacemap (obj %llu) [error=%d]",
2929 (u_longlong_t)obsolete_sm_object, error);
2932 space_map_update(vd->vdev_obsolete_sm);
2939 * The special vdev case is used for hot spares and l2cache devices. Its
2940 * sole purpose it to set the vdev state for the associated vdev. To do this,
2941 * we make sure that we can open the underlying device, then try to read the
2942 * label, and make sure that the label is sane and that it hasn't been
2943 * repurposed to another pool.
2946 vdev_validate_aux(vdev_t *vd)
2949 uint64_t guid, version;
2952 if (!vdev_readable(vd))
2955 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2956 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2957 VDEV_AUX_CORRUPT_DATA);
2961 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2962 !SPA_VERSION_IS_SUPPORTED(version) ||
2963 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2964 guid != vd->vdev_guid ||
2965 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2966 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2967 VDEV_AUX_CORRUPT_DATA);
2973 * We don't actually check the pool state here. If it's in fact in
2974 * use by another pool, we update this fact on the fly when requested.
2981 * Free the objects used to store this vdev's spacemaps, and the array
2982 * that points to them.
2985 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
2987 if (vd->vdev_ms_array == 0)
2990 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2991 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
2992 size_t array_bytes = array_count * sizeof (uint64_t);
2993 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
2994 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
2995 array_bytes, smobj_array, 0));
2997 for (uint64_t i = 0; i < array_count; i++) {
2998 uint64_t smobj = smobj_array[i];
3002 space_map_free_obj(mos, smobj, tx);
3005 kmem_free(smobj_array, array_bytes);
3006 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3007 vd->vdev_ms_array = 0;
3011 vdev_remove_empty(vdev_t *vd, uint64_t txg)
3013 spa_t *spa = vd->vdev_spa;
3016 ASSERT(vd == vd->vdev_top);
3017 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3019 if (vd->vdev_ms != NULL) {
3020 metaslab_group_t *mg = vd->vdev_mg;
3022 metaslab_group_histogram_verify(mg);
3023 metaslab_class_histogram_verify(mg->mg_class);
3025 for (int m = 0; m < vd->vdev_ms_count; m++) {
3026 metaslab_t *msp = vd->vdev_ms[m];
3028 if (msp == NULL || msp->ms_sm == NULL)
3031 mutex_enter(&msp->ms_lock);
3033 * If the metaslab was not loaded when the vdev
3034 * was removed then the histogram accounting may
3035 * not be accurate. Update the histogram information
3036 * here so that we ensure that the metaslab group
3037 * and metaslab class are up-to-date.
3039 metaslab_group_histogram_remove(mg, msp);
3041 VERIFY0(space_map_allocated(msp->ms_sm));
3042 space_map_close(msp->ms_sm);
3044 mutex_exit(&msp->ms_lock);
3047 if (vd->vdev_checkpoint_sm != NULL) {
3048 ASSERT(spa_has_checkpoint(spa));
3049 space_map_close(vd->vdev_checkpoint_sm);
3050 vd->vdev_checkpoint_sm = NULL;
3053 metaslab_group_histogram_verify(mg);
3054 metaslab_class_histogram_verify(mg->mg_class);
3055 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
3056 ASSERT0(mg->mg_histogram[i]);
3059 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3060 vdev_destroy_spacemaps(vd, tx);
3062 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
3063 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3064 vd->vdev_top_zap = 0;
3070 vdev_sync_done(vdev_t *vd, uint64_t txg)
3073 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3075 ASSERT(vdev_is_concrete(vd));
3077 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3079 metaslab_sync_done(msp, txg);
3082 metaslab_sync_reassess(vd->vdev_mg);
3086 vdev_sync(vdev_t *vd, uint64_t txg)
3088 spa_t *spa = vd->vdev_spa;
3093 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3096 ASSERT(vd->vdev_removing ||
3097 vd->vdev_ops == &vdev_indirect_ops);
3099 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3100 vdev_indirect_sync_obsolete(vd, tx);
3104 * If the vdev is indirect, it can't have dirty
3105 * metaslabs or DTLs.
3107 if (vd->vdev_ops == &vdev_indirect_ops) {
3108 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3109 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3114 ASSERT(vdev_is_concrete(vd));
3116 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3117 !vd->vdev_removing) {
3118 ASSERT(vd == vd->vdev_top);
3119 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3120 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3121 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3122 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3123 ASSERT(vd->vdev_ms_array != 0);
3124 vdev_config_dirty(vd);
3128 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3129 metaslab_sync(msp, txg);
3130 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3133 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3134 vdev_dtl_sync(lvd, txg);
3137 * Remove the metadata associated with this vdev once it's empty.
3138 * Note that this is typically used for log/cache device removal;
3139 * we don't empty toplevel vdevs when removing them. But if
3140 * a toplevel happens to be emptied, this is not harmful.
3142 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) {
3143 vdev_remove_empty(vd, txg);
3146 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3150 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3152 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3156 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3157 * not be opened, and no I/O is attempted.
3160 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3164 spa_vdev_state_enter(spa, SCL_NONE);
3166 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3167 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3169 if (!vd->vdev_ops->vdev_op_leaf)
3170 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3175 * We don't directly use the aux state here, but if we do a
3176 * vdev_reopen(), we need this value to be present to remember why we
3179 vd->vdev_label_aux = aux;
3182 * Faulted state takes precedence over degraded.
3184 vd->vdev_delayed_close = B_FALSE;
3185 vd->vdev_faulted = 1ULL;
3186 vd->vdev_degraded = 0ULL;
3187 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3190 * If this device has the only valid copy of the data, then
3191 * back off and simply mark the vdev as degraded instead.
3193 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3194 vd->vdev_degraded = 1ULL;
3195 vd->vdev_faulted = 0ULL;
3198 * If we reopen the device and it's not dead, only then do we
3203 if (vdev_readable(vd))
3204 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3207 return (spa_vdev_state_exit(spa, vd, 0));
3211 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3212 * user that something is wrong. The vdev continues to operate as normal as far
3213 * as I/O is concerned.
3216 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3220 spa_vdev_state_enter(spa, SCL_NONE);
3222 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3223 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3225 if (!vd->vdev_ops->vdev_op_leaf)
3226 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3229 * If the vdev is already faulted, then don't do anything.
3231 if (vd->vdev_faulted || vd->vdev_degraded)
3232 return (spa_vdev_state_exit(spa, NULL, 0));
3234 vd->vdev_degraded = 1ULL;
3235 if (!vdev_is_dead(vd))
3236 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3239 return (spa_vdev_state_exit(spa, vd, 0));
3243 * Online the given vdev.
3245 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3246 * spare device should be detached when the device finishes resilvering.
3247 * Second, the online should be treated like a 'test' online case, so no FMA
3248 * events are generated if the device fails to open.
3251 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3253 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3254 boolean_t wasoffline;
3255 vdev_state_t oldstate;
3257 spa_vdev_state_enter(spa, SCL_NONE);
3259 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3260 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3262 if (!vd->vdev_ops->vdev_op_leaf)
3263 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3265 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3266 oldstate = vd->vdev_state;
3269 vd->vdev_offline = B_FALSE;
3270 vd->vdev_tmpoffline = B_FALSE;
3271 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3272 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3274 /* XXX - L2ARC 1.0 does not support expansion */
3275 if (!vd->vdev_aux) {
3276 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3277 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
3281 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3283 if (!vd->vdev_aux) {
3284 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3285 pvd->vdev_expanding = B_FALSE;
3289 *newstate = vd->vdev_state;
3290 if ((flags & ZFS_ONLINE_UNSPARE) &&
3291 !vdev_is_dead(vd) && vd->vdev_parent &&
3292 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3293 vd->vdev_parent->vdev_child[0] == vd)
3294 vd->vdev_unspare = B_TRUE;
3296 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3298 /* XXX - L2ARC 1.0 does not support expansion */
3300 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3301 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3304 /* Restart initializing if necessary */
3305 mutex_enter(&vd->vdev_initialize_lock);
3306 if (vdev_writeable(vd) &&
3307 vd->vdev_initialize_thread == NULL &&
3308 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3309 (void) vdev_initialize(vd);
3311 mutex_exit(&vd->vdev_initialize_lock);
3314 (oldstate < VDEV_STATE_DEGRADED &&
3315 vd->vdev_state >= VDEV_STATE_DEGRADED))
3316 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3318 return (spa_vdev_state_exit(spa, vd, 0));
3322 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3326 uint64_t generation;
3327 metaslab_group_t *mg;
3330 spa_vdev_state_enter(spa, SCL_ALLOC);
3332 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3333 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3335 if (!vd->vdev_ops->vdev_op_leaf)
3336 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3340 generation = spa->spa_config_generation + 1;
3343 * If the device isn't already offline, try to offline it.
3345 if (!vd->vdev_offline) {
3347 * If this device has the only valid copy of some data,
3348 * don't allow it to be offlined. Log devices are always
3351 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3352 vdev_dtl_required(vd))
3353 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3356 * If the top-level is a slog and it has had allocations
3357 * then proceed. We check that the vdev's metaslab group
3358 * is not NULL since it's possible that we may have just
3359 * added this vdev but not yet initialized its metaslabs.
3361 if (tvd->vdev_islog && mg != NULL) {
3363 * Prevent any future allocations.
3365 metaslab_group_passivate(mg);
3366 (void) spa_vdev_state_exit(spa, vd, 0);
3368 error = spa_reset_logs(spa);
3371 * If the log device was successfully reset but has
3372 * checkpointed data, do not offline it.
3375 tvd->vdev_checkpoint_sm != NULL) {
3376 ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc,
3378 error = ZFS_ERR_CHECKPOINT_EXISTS;
3381 spa_vdev_state_enter(spa, SCL_ALLOC);
3384 * Check to see if the config has changed.
3386 if (error || generation != spa->spa_config_generation) {
3387 metaslab_group_activate(mg);
3389 return (spa_vdev_state_exit(spa,
3391 (void) spa_vdev_state_exit(spa, vd, 0);
3394 ASSERT0(tvd->vdev_stat.vs_alloc);
3398 * Offline this device and reopen its top-level vdev.
3399 * If the top-level vdev is a log device then just offline
3400 * it. Otherwise, if this action results in the top-level
3401 * vdev becoming unusable, undo it and fail the request.
3403 vd->vdev_offline = B_TRUE;
3406 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3407 vdev_is_dead(tvd)) {
3408 vd->vdev_offline = B_FALSE;
3410 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3414 * Add the device back into the metaslab rotor so that
3415 * once we online the device it's open for business.
3417 if (tvd->vdev_islog && mg != NULL)
3418 metaslab_group_activate(mg);
3421 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3423 return (spa_vdev_state_exit(spa, vd, 0));
3427 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3431 mutex_enter(&spa->spa_vdev_top_lock);
3432 error = vdev_offline_locked(spa, guid, flags);
3433 mutex_exit(&spa->spa_vdev_top_lock);
3439 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3440 * vdev_offline(), we assume the spa config is locked. We also clear all
3441 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3444 vdev_clear(spa_t *spa, vdev_t *vd)
3446 vdev_t *rvd = spa->spa_root_vdev;
3448 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3453 vd->vdev_stat.vs_read_errors = 0;
3454 vd->vdev_stat.vs_write_errors = 0;
3455 vd->vdev_stat.vs_checksum_errors = 0;
3457 for (int c = 0; c < vd->vdev_children; c++)
3458 vdev_clear(spa, vd->vdev_child[c]);
3461 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
3462 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
3464 for (int c = 0; c < spa->spa_spares.sav_count; c++)
3465 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
3469 * It makes no sense to "clear" an indirect vdev.
3471 if (!vdev_is_concrete(vd))
3475 * If we're in the FAULTED state or have experienced failed I/O, then
3476 * clear the persistent state and attempt to reopen the device. We
3477 * also mark the vdev config dirty, so that the new faulted state is
3478 * written out to disk.
3480 if (vd->vdev_faulted || vd->vdev_degraded ||
3481 !vdev_readable(vd) || !vdev_writeable(vd)) {
3484 * When reopening in reponse to a clear event, it may be due to
3485 * a fmadm repair request. In this case, if the device is
3486 * still broken, we want to still post the ereport again.
3488 vd->vdev_forcefault = B_TRUE;
3490 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3491 vd->vdev_cant_read = B_FALSE;
3492 vd->vdev_cant_write = B_FALSE;
3494 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3496 vd->vdev_forcefault = B_FALSE;
3498 if (vd != rvd && vdev_writeable(vd->vdev_top))
3499 vdev_state_dirty(vd->vdev_top);
3501 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
3502 spa_async_request(spa, SPA_ASYNC_RESILVER);
3504 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3508 * When clearing a FMA-diagnosed fault, we always want to
3509 * unspare the device, as we assume that the original spare was
3510 * done in response to the FMA fault.
3512 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3513 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3514 vd->vdev_parent->vdev_child[0] == vd)
3515 vd->vdev_unspare = B_TRUE;
3519 vdev_is_dead(vdev_t *vd)
3522 * Holes and missing devices are always considered "dead".
3523 * This simplifies the code since we don't have to check for
3524 * these types of devices in the various code paths.
3525 * Instead we rely on the fact that we skip over dead devices
3526 * before issuing I/O to them.
3528 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3529 vd->vdev_ops == &vdev_hole_ops ||
3530 vd->vdev_ops == &vdev_missing_ops);
3534 vdev_readable(vdev_t *vd)
3536 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3540 vdev_writeable(vdev_t *vd)
3542 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3543 vdev_is_concrete(vd));
3547 vdev_allocatable(vdev_t *vd)
3549 uint64_t state = vd->vdev_state;
3552 * We currently allow allocations from vdevs which may be in the
3553 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3554 * fails to reopen then we'll catch it later when we're holding
3555 * the proper locks. Note that we have to get the vdev state
3556 * in a local variable because although it changes atomically,
3557 * we're asking two separate questions about it.
3559 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3560 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3561 vd->vdev_mg->mg_initialized);
3565 vdev_accessible(vdev_t *vd, zio_t *zio)
3567 ASSERT(zio->io_vd == vd);
3569 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3572 if (zio->io_type == ZIO_TYPE_READ)
3573 return (!vd->vdev_cant_read);
3575 if (zio->io_type == ZIO_TYPE_WRITE)
3576 return (!vd->vdev_cant_write);
3582 vdev_is_spacemap_addressable(vdev_t *vd)
3585 * Assuming 47 bits of the space map entry dedicated for the entry's
3586 * offset (see description in space_map.h), we calculate the maximum
3587 * address that can be described by a space map entry for the given
3590 uint64_t shift = vd->vdev_ashift + 47;
3592 if (shift >= 63) /* detect potential overflow */
3595 return (vd->vdev_asize < (1ULL << shift));
3599 * Get statistics for the given vdev.
3602 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3604 spa_t *spa = vd->vdev_spa;
3605 vdev_t *rvd = spa->spa_root_vdev;
3606 vdev_t *tvd = vd->vdev_top;
3608 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3610 mutex_enter(&vd->vdev_stat_lock);
3611 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3612 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3613 vs->vs_state = vd->vdev_state;
3614 vs->vs_rsize = vdev_get_min_asize(vd);
3615 if (vd->vdev_ops->vdev_op_leaf) {
3616 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3618 * Report intializing progress. Since we don't have the
3619 * initializing locks held, this is only an estimate (although a
3620 * fairly accurate one).
3622 vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done;
3623 vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est;
3624 vs->vs_initialize_state = vd->vdev_initialize_state;
3625 vs->vs_initialize_action_time = vd->vdev_initialize_action_time;
3628 * Report expandable space on top-level, non-auxillary devices only.
3629 * The expandable space is reported in terms of metaslab sized units
3630 * since that determines how much space the pool can expand.
3632 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
3633 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3634 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3636 vs->vs_configured_ashift = vd->vdev_top != NULL
3637 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
3638 vs->vs_logical_ashift = vd->vdev_logical_ashift;
3639 vs->vs_physical_ashift = vd->vdev_physical_ashift;
3640 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3641 vdev_is_concrete(vd)) {
3642 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
3646 * If we're getting stats on the root vdev, aggregate the I/O counts
3647 * over all top-level vdevs (i.e. the direct children of the root).
3650 for (int c = 0; c < rvd->vdev_children; c++) {
3651 vdev_t *cvd = rvd->vdev_child[c];
3652 vdev_stat_t *cvs = &cvd->vdev_stat;
3654 for (int t = 0; t < ZIO_TYPES; t++) {
3655 vs->vs_ops[t] += cvs->vs_ops[t];
3656 vs->vs_bytes[t] += cvs->vs_bytes[t];
3658 cvs->vs_scan_removing = cvd->vdev_removing;
3661 mutex_exit(&vd->vdev_stat_lock);
3665 vdev_clear_stats(vdev_t *vd)
3667 mutex_enter(&vd->vdev_stat_lock);
3668 vd->vdev_stat.vs_space = 0;
3669 vd->vdev_stat.vs_dspace = 0;
3670 vd->vdev_stat.vs_alloc = 0;
3671 mutex_exit(&vd->vdev_stat_lock);
3675 vdev_scan_stat_init(vdev_t *vd)
3677 vdev_stat_t *vs = &vd->vdev_stat;
3679 for (int c = 0; c < vd->vdev_children; c++)
3680 vdev_scan_stat_init(vd->vdev_child[c]);
3682 mutex_enter(&vd->vdev_stat_lock);
3683 vs->vs_scan_processed = 0;
3684 mutex_exit(&vd->vdev_stat_lock);
3688 vdev_stat_update(zio_t *zio, uint64_t psize)
3690 spa_t *spa = zio->io_spa;
3691 vdev_t *rvd = spa->spa_root_vdev;
3692 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3694 uint64_t txg = zio->io_txg;
3695 vdev_stat_t *vs = &vd->vdev_stat;
3696 zio_type_t type = zio->io_type;
3697 int flags = zio->io_flags;
3700 * If this i/o is a gang leader, it didn't do any actual work.
3702 if (zio->io_gang_tree)
3705 if (zio->io_error == 0) {
3707 * If this is a root i/o, don't count it -- we've already
3708 * counted the top-level vdevs, and vdev_get_stats() will
3709 * aggregate them when asked. This reduces contention on
3710 * the root vdev_stat_lock and implicitly handles blocks
3711 * that compress away to holes, for which there is no i/o.
3712 * (Holes never create vdev children, so all the counters
3713 * remain zero, which is what we want.)
3715 * Note: this only applies to successful i/o (io_error == 0)
3716 * because unlike i/o counts, errors are not additive.
3717 * When reading a ditto block, for example, failure of
3718 * one top-level vdev does not imply a root-level error.
3723 ASSERT(vd == zio->io_vd);
3725 if (flags & ZIO_FLAG_IO_BYPASS)
3728 mutex_enter(&vd->vdev_stat_lock);
3730 if (flags & ZIO_FLAG_IO_REPAIR) {
3731 if (flags & ZIO_FLAG_SCAN_THREAD) {
3732 dsl_scan_phys_t *scn_phys =
3733 &spa->spa_dsl_pool->dp_scan->scn_phys;
3734 uint64_t *processed = &scn_phys->scn_processed;
3737 if (vd->vdev_ops->vdev_op_leaf)
3738 atomic_add_64(processed, psize);
3739 vs->vs_scan_processed += psize;
3742 if (flags & ZIO_FLAG_SELF_HEAL)
3743 vs->vs_self_healed += psize;
3747 vs->vs_bytes[type] += psize;
3749 mutex_exit(&vd->vdev_stat_lock);
3753 if (flags & ZIO_FLAG_SPECULATIVE)
3757 * If this is an I/O error that is going to be retried, then ignore the
3758 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3759 * hard errors, when in reality they can happen for any number of
3760 * innocuous reasons (bus resets, MPxIO link failure, etc).
3762 if (zio->io_error == EIO &&
3763 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3767 * Intent logs writes won't propagate their error to the root
3768 * I/O so don't mark these types of failures as pool-level
3771 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3774 mutex_enter(&vd->vdev_stat_lock);
3775 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3776 if (zio->io_error == ECKSUM)
3777 vs->vs_checksum_errors++;
3779 vs->vs_read_errors++;
3781 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3782 vs->vs_write_errors++;
3783 mutex_exit(&vd->vdev_stat_lock);
3785 if (spa->spa_load_state == SPA_LOAD_NONE &&
3786 type == ZIO_TYPE_WRITE && txg != 0 &&
3787 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3788 (flags & ZIO_FLAG_SCAN_THREAD) ||
3789 spa->spa_claiming)) {
3791 * This is either a normal write (not a repair), or it's
3792 * a repair induced by the scrub thread, or it's a repair
3793 * made by zil_claim() during spa_load() in the first txg.
3794 * In the normal case, we commit the DTL change in the same
3795 * txg as the block was born. In the scrub-induced repair
3796 * case, we know that scrubs run in first-pass syncing context,
3797 * so we commit the DTL change in spa_syncing_txg(spa).
3798 * In the zil_claim() case, we commit in spa_first_txg(spa).
3800 * We currently do not make DTL entries for failed spontaneous
3801 * self-healing writes triggered by normal (non-scrubbing)
3802 * reads, because we have no transactional context in which to
3803 * do so -- and it's not clear that it'd be desirable anyway.
3805 if (vd->vdev_ops->vdev_op_leaf) {
3806 uint64_t commit_txg = txg;
3807 if (flags & ZIO_FLAG_SCAN_THREAD) {
3808 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3809 ASSERT(spa_sync_pass(spa) == 1);
3810 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3811 commit_txg = spa_syncing_txg(spa);
3812 } else if (spa->spa_claiming) {
3813 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3814 commit_txg = spa_first_txg(spa);
3816 ASSERT(commit_txg >= spa_syncing_txg(spa));
3817 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3819 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3820 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3821 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3824 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3829 * Update the in-core space usage stats for this vdev, its metaslab class,
3830 * and the root vdev.
3833 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3834 int64_t space_delta)
3836 int64_t dspace_delta = space_delta;
3837 spa_t *spa = vd->vdev_spa;
3838 vdev_t *rvd = spa->spa_root_vdev;
3839 metaslab_group_t *mg = vd->vdev_mg;
3840 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3842 ASSERT(vd == vd->vdev_top);
3845 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3846 * factor. We must calculate this here and not at the root vdev
3847 * because the root vdev's psize-to-asize is simply the max of its
3848 * childrens', thus not accurate enough for us.
3850 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3851 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3852 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3853 vd->vdev_deflate_ratio;
3855 mutex_enter(&vd->vdev_stat_lock);
3856 vd->vdev_stat.vs_alloc += alloc_delta;
3857 vd->vdev_stat.vs_space += space_delta;
3858 vd->vdev_stat.vs_dspace += dspace_delta;
3859 mutex_exit(&vd->vdev_stat_lock);
3861 if (mc == spa_normal_class(spa)) {
3862 mutex_enter(&rvd->vdev_stat_lock);
3863 rvd->vdev_stat.vs_alloc += alloc_delta;
3864 rvd->vdev_stat.vs_space += space_delta;
3865 rvd->vdev_stat.vs_dspace += dspace_delta;
3866 mutex_exit(&rvd->vdev_stat_lock);
3870 ASSERT(rvd == vd->vdev_parent);
3871 ASSERT(vd->vdev_ms_count != 0);
3873 metaslab_class_space_update(mc,
3874 alloc_delta, defer_delta, space_delta, dspace_delta);
3879 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3880 * so that it will be written out next time the vdev configuration is synced.
3881 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3884 vdev_config_dirty(vdev_t *vd)
3886 spa_t *spa = vd->vdev_spa;
3887 vdev_t *rvd = spa->spa_root_vdev;
3890 ASSERT(spa_writeable(spa));
3893 * If this is an aux vdev (as with l2cache and spare devices), then we
3894 * update the vdev config manually and set the sync flag.
3896 if (vd->vdev_aux != NULL) {
3897 spa_aux_vdev_t *sav = vd->vdev_aux;
3901 for (c = 0; c < sav->sav_count; c++) {
3902 if (sav->sav_vdevs[c] == vd)
3906 if (c == sav->sav_count) {
3908 * We're being removed. There's nothing more to do.
3910 ASSERT(sav->sav_sync == B_TRUE);
3914 sav->sav_sync = B_TRUE;
3916 if (nvlist_lookup_nvlist_array(sav->sav_config,
3917 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3918 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3919 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3925 * Setting the nvlist in the middle if the array is a little
3926 * sketchy, but it will work.
3928 nvlist_free(aux[c]);
3929 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3935 * The dirty list is protected by the SCL_CONFIG lock. The caller
3936 * must either hold SCL_CONFIG as writer, or must be the sync thread
3937 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3938 * so this is sufficient to ensure mutual exclusion.
3940 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3941 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3942 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3945 for (c = 0; c < rvd->vdev_children; c++)
3946 vdev_config_dirty(rvd->vdev_child[c]);
3948 ASSERT(vd == vd->vdev_top);
3950 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3951 vdev_is_concrete(vd)) {
3952 list_insert_head(&spa->spa_config_dirty_list, vd);
3958 vdev_config_clean(vdev_t *vd)
3960 spa_t *spa = vd->vdev_spa;
3962 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3963 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3964 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3966 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3967 list_remove(&spa->spa_config_dirty_list, vd);
3971 * Mark a top-level vdev's state as dirty, so that the next pass of
3972 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3973 * the state changes from larger config changes because they require
3974 * much less locking, and are often needed for administrative actions.
3977 vdev_state_dirty(vdev_t *vd)
3979 spa_t *spa = vd->vdev_spa;
3981 ASSERT(spa_writeable(spa));
3982 ASSERT(vd == vd->vdev_top);
3985 * The state list is protected by the SCL_STATE lock. The caller
3986 * must either hold SCL_STATE as writer, or must be the sync thread
3987 * (which holds SCL_STATE as reader). There's only one sync thread,
3988 * so this is sufficient to ensure mutual exclusion.
3990 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3991 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3992 spa_config_held(spa, SCL_STATE, RW_READER)));
3994 if (!list_link_active(&vd->vdev_state_dirty_node) &&
3995 vdev_is_concrete(vd))
3996 list_insert_head(&spa->spa_state_dirty_list, vd);
4000 vdev_state_clean(vdev_t *vd)
4002 spa_t *spa = vd->vdev_spa;
4004 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4005 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4006 spa_config_held(spa, SCL_STATE, RW_READER)));
4008 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4009 list_remove(&spa->spa_state_dirty_list, vd);
4013 * Propagate vdev state up from children to parent.
4016 vdev_propagate_state(vdev_t *vd)
4018 spa_t *spa = vd->vdev_spa;
4019 vdev_t *rvd = spa->spa_root_vdev;
4020 int degraded = 0, faulted = 0;
4024 if (vd->vdev_children > 0) {
4025 for (int c = 0; c < vd->vdev_children; c++) {
4026 child = vd->vdev_child[c];
4029 * Don't factor holes or indirect vdevs into the
4032 if (!vdev_is_concrete(child))
4035 if (!vdev_readable(child) ||
4036 (!vdev_writeable(child) && spa_writeable(spa))) {
4038 * Root special: if there is a top-level log
4039 * device, treat the root vdev as if it were
4042 if (child->vdev_islog && vd == rvd)
4046 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4050 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4054 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4057 * Root special: if there is a top-level vdev that cannot be
4058 * opened due to corrupted metadata, then propagate the root
4059 * vdev's aux state as 'corrupt' rather than 'insufficient
4062 if (corrupted && vd == rvd &&
4063 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4064 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4065 VDEV_AUX_CORRUPT_DATA);
4068 if (vd->vdev_parent)
4069 vdev_propagate_state(vd->vdev_parent);
4073 * Set a vdev's state. If this is during an open, we don't update the parent
4074 * state, because we're in the process of opening children depth-first.
4075 * Otherwise, we propagate the change to the parent.
4077 * If this routine places a device in a faulted state, an appropriate ereport is
4081 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4083 uint64_t save_state;
4084 spa_t *spa = vd->vdev_spa;
4086 if (state == vd->vdev_state) {
4087 vd->vdev_stat.vs_aux = aux;
4091 save_state = vd->vdev_state;
4093 vd->vdev_state = state;
4094 vd->vdev_stat.vs_aux = aux;
4097 * If we are setting the vdev state to anything but an open state, then
4098 * always close the underlying device unless the device has requested
4099 * a delayed close (i.e. we're about to remove or fault the device).
4100 * Otherwise, we keep accessible but invalid devices open forever.
4101 * We don't call vdev_close() itself, because that implies some extra
4102 * checks (offline, etc) that we don't want here. This is limited to
4103 * leaf devices, because otherwise closing the device will affect other
4106 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4107 vd->vdev_ops->vdev_op_leaf)
4108 vd->vdev_ops->vdev_op_close(vd);
4110 if (vd->vdev_removed &&
4111 state == VDEV_STATE_CANT_OPEN &&
4112 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4114 * If the previous state is set to VDEV_STATE_REMOVED, then this
4115 * device was previously marked removed and someone attempted to
4116 * reopen it. If this failed due to a nonexistent device, then
4117 * keep the device in the REMOVED state. We also let this be if
4118 * it is one of our special test online cases, which is only
4119 * attempting to online the device and shouldn't generate an FMA
4122 vd->vdev_state = VDEV_STATE_REMOVED;
4123 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4124 } else if (state == VDEV_STATE_REMOVED) {
4125 vd->vdev_removed = B_TRUE;
4126 } else if (state == VDEV_STATE_CANT_OPEN) {
4128 * If we fail to open a vdev during an import or recovery, we
4129 * mark it as "not available", which signifies that it was
4130 * never there to begin with. Failure to open such a device
4131 * is not considered an error.
4133 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4134 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4135 vd->vdev_ops->vdev_op_leaf)
4136 vd->vdev_not_present = 1;
4139 * Post the appropriate ereport. If the 'prevstate' field is
4140 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4141 * that this is part of a vdev_reopen(). In this case, we don't
4142 * want to post the ereport if the device was already in the
4143 * CANT_OPEN state beforehand.
4145 * If the 'checkremove' flag is set, then this is an attempt to
4146 * online the device in response to an insertion event. If we
4147 * hit this case, then we have detected an insertion event for a
4148 * faulted or offline device that wasn't in the removed state.
4149 * In this scenario, we don't post an ereport because we are
4150 * about to replace the device, or attempt an online with
4151 * vdev_forcefault, which will generate the fault for us.
4153 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4154 !vd->vdev_not_present && !vd->vdev_checkremove &&
4155 vd != spa->spa_root_vdev) {
4159 case VDEV_AUX_OPEN_FAILED:
4160 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4162 case VDEV_AUX_CORRUPT_DATA:
4163 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4165 case VDEV_AUX_NO_REPLICAS:
4166 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4168 case VDEV_AUX_BAD_GUID_SUM:
4169 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4171 case VDEV_AUX_TOO_SMALL:
4172 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4174 case VDEV_AUX_BAD_LABEL:
4175 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4178 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4181 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
4184 /* Erase any notion of persistent removed state */
4185 vd->vdev_removed = B_FALSE;
4187 vd->vdev_removed = B_FALSE;
4191 * Notify the fmd of the state change. Be verbose and post
4192 * notifications even for stuff that's not important; the fmd agent can
4193 * sort it out. Don't emit state change events for non-leaf vdevs since
4194 * they can't change state on their own. The FMD can check their state
4195 * if it wants to when it sees that a leaf vdev had a state change.
4197 if (vd->vdev_ops->vdev_op_leaf)
4198 zfs_post_state_change(spa, vd);
4200 if (!isopen && vd->vdev_parent)
4201 vdev_propagate_state(vd->vdev_parent);
4205 vdev_children_are_offline(vdev_t *vd)
4207 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4209 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4210 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4218 * Check the vdev configuration to ensure that it's capable of supporting
4219 * a root pool. We do not support partial configuration.
4220 * In addition, only a single top-level vdev is allowed.
4222 * FreeBSD does not have above limitations.
4225 vdev_is_bootable(vdev_t *vd)
4228 if (!vd->vdev_ops->vdev_op_leaf) {
4229 char *vdev_type = vd->vdev_ops->vdev_op_type;
4231 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
4232 vd->vdev_children > 1) {
4234 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4235 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4240 for (int c = 0; c < vd->vdev_children; c++) {
4241 if (!vdev_is_bootable(vd->vdev_child[c]))
4244 #endif /* illumos */
4249 vdev_is_concrete(vdev_t *vd)
4251 vdev_ops_t *ops = vd->vdev_ops;
4252 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4253 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4261 * Determine if a log device has valid content. If the vdev was
4262 * removed or faulted in the MOS config then we know that
4263 * the content on the log device has already been written to the pool.
4266 vdev_log_state_valid(vdev_t *vd)
4268 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4272 for (int c = 0; c < vd->vdev_children; c++)
4273 if (vdev_log_state_valid(vd->vdev_child[c]))
4280 * Expand a vdev if possible.
4283 vdev_expand(vdev_t *vd, uint64_t txg)
4285 ASSERT(vd->vdev_top == vd);
4286 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4287 ASSERT(vdev_is_concrete(vd));
4289 vdev_set_deflate_ratio(vd);
4291 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
4292 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4293 vdev_config_dirty(vd);
4301 vdev_split(vdev_t *vd)
4303 vdev_t *cvd, *pvd = vd->vdev_parent;
4305 vdev_remove_child(pvd, vd);
4306 vdev_compact_children(pvd);
4308 cvd = pvd->vdev_child[0];
4309 if (pvd->vdev_children == 1) {
4310 vdev_remove_parent(cvd);
4311 cvd->vdev_splitting = B_TRUE;
4313 vdev_propagate_state(cvd);
4317 vdev_deadman(vdev_t *vd)
4319 for (int c = 0; c < vd->vdev_children; c++) {
4320 vdev_t *cvd = vd->vdev_child[c];
4325 if (vd->vdev_ops->vdev_op_leaf) {
4326 vdev_queue_t *vq = &vd->vdev_queue;
4328 mutex_enter(&vq->vq_lock);
4329 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4330 spa_t *spa = vd->vdev_spa;
4335 * Look at the head of all the pending queues,
4336 * if any I/O has been outstanding for longer than
4337 * the spa_deadman_synctime we panic the system.
4339 fio = avl_first(&vq->vq_active_tree);
4340 delta = gethrtime() - fio->io_timestamp;
4341 if (delta > spa_deadman_synctime(spa)) {
4342 vdev_dbgmsg(vd, "SLOW IO: zio timestamp "
4343 "%lluns, delta %lluns, last io %lluns",
4344 fio->io_timestamp, (u_longlong_t)delta,
4345 vq->vq_io_complete_ts);
4346 fm_panic("I/O to pool '%s' appears to be "
4347 "hung on vdev guid %llu at '%s'.",
4349 (long long unsigned int) vd->vdev_guid,
4353 mutex_exit(&vq->vq_lock);