4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright 2016 Toomas Soome <tsoome@me.com>
29 * Copyright 2017 Joyent, Inc.
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/bpobj.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/uberblock_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/metaslab_impl.h>
44 #include <sys/space_map.h>
45 #include <sys/space_reftree.h>
48 #include <sys/fs/zfs.h>
51 #include <sys/dsl_scan.h>
53 #include <sys/trim_map.h>
55 SYSCTL_DECL(_vfs_zfs);
56 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
59 * Virtual device management.
63 * The limit for ZFS to automatically increase a top-level vdev's ashift
64 * from logical ashift to physical ashift.
66 * Example: one or more 512B emulation child vdevs
67 * child->vdev_ashift = 9 (512 bytes)
68 * child->vdev_physical_ashift = 12 (4096 bytes)
69 * zfs_max_auto_ashift = 11 (2048 bytes)
70 * zfs_min_auto_ashift = 9 (512 bytes)
72 * On pool creation or the addition of a new top-level vdev, ZFS will
73 * increase the ashift of the top-level vdev to 2048 as limited by
74 * zfs_max_auto_ashift.
76 * Example: one or more 512B emulation child vdevs
77 * child->vdev_ashift = 9 (512 bytes)
78 * child->vdev_physical_ashift = 12 (4096 bytes)
79 * zfs_max_auto_ashift = 13 (8192 bytes)
80 * zfs_min_auto_ashift = 9 (512 bytes)
82 * On pool creation or the addition of a new top-level vdev, ZFS will
83 * increase the ashift of the top-level vdev to 4096 to match the
84 * max vdev_physical_ashift.
86 * Example: one or more 512B emulation child vdevs
87 * child->vdev_ashift = 9 (512 bytes)
88 * child->vdev_physical_ashift = 9 (512 bytes)
89 * zfs_max_auto_ashift = 13 (8192 bytes)
90 * zfs_min_auto_ashift = 12 (4096 bytes)
92 * On pool creation or the addition of a new top-level vdev, ZFS will
93 * increase the ashift of the top-level vdev to 4096 to match the
94 * zfs_min_auto_ashift.
96 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
97 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
100 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
105 val = zfs_max_auto_ashift;
106 err = sysctl_handle_64(oidp, &val, 0, req);
107 if (err != 0 || req->newptr == NULL)
110 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
113 zfs_max_auto_ashift = val;
117 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
118 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
119 sysctl_vfs_zfs_max_auto_ashift, "QU",
120 "Max ashift used when optimising for logical -> physical sectors size on "
121 "new top-level vdevs.");
124 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
129 val = zfs_min_auto_ashift;
130 err = sysctl_handle_64(oidp, &val, 0, req);
131 if (err != 0 || req->newptr == NULL)
134 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
137 zfs_min_auto_ashift = val;
141 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
142 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
143 sysctl_vfs_zfs_min_auto_ashift, "QU",
144 "Min ashift used when creating new top-level vdevs.");
146 static vdev_ops_t *vdev_ops_table[] = {
166 * When a vdev is added, it will be divided into approximately (but no
167 * more than) this number of metaslabs.
169 int metaslabs_per_vdev = 200;
170 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
171 &metaslabs_per_vdev, 0,
172 "When a vdev is added, how many metaslabs the vdev should be divided into");
176 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
182 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
185 if (vd->vdev_path != NULL) {
186 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
189 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
190 vd->vdev_ops->vdev_op_type,
191 (u_longlong_t)vd->vdev_id,
192 (u_longlong_t)vd->vdev_guid, buf);
197 * Given a vdev type, return the appropriate ops vector.
200 vdev_getops(const char *type)
202 vdev_ops_t *ops, **opspp;
204 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
205 if (strcmp(ops->vdev_op_type, type) == 0)
212 * Default asize function: return the MAX of psize with the asize of
213 * all children. This is what's used by anything other than RAID-Z.
216 vdev_default_asize(vdev_t *vd, uint64_t psize)
218 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
221 for (int c = 0; c < vd->vdev_children; c++) {
222 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
223 asize = MAX(asize, csize);
230 * Get the minimum allocatable size. We define the allocatable size as
231 * the vdev's asize rounded to the nearest metaslab. This allows us to
232 * replace or attach devices which don't have the same physical size but
233 * can still satisfy the same number of allocations.
236 vdev_get_min_asize(vdev_t *vd)
238 vdev_t *pvd = vd->vdev_parent;
241 * If our parent is NULL (inactive spare or cache) or is the root,
242 * just return our own asize.
245 return (vd->vdev_asize);
248 * The top-level vdev just returns the allocatable size rounded
249 * to the nearest metaslab.
251 if (vd == vd->vdev_top)
252 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
255 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
256 * so each child must provide at least 1/Nth of its asize.
258 if (pvd->vdev_ops == &vdev_raidz_ops)
259 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
262 return (pvd->vdev_min_asize);
266 vdev_set_min_asize(vdev_t *vd)
268 vd->vdev_min_asize = vdev_get_min_asize(vd);
270 for (int c = 0; c < vd->vdev_children; c++)
271 vdev_set_min_asize(vd->vdev_child[c]);
275 vdev_lookup_top(spa_t *spa, uint64_t vdev)
277 vdev_t *rvd = spa->spa_root_vdev;
279 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
281 if (vdev < rvd->vdev_children) {
282 ASSERT(rvd->vdev_child[vdev] != NULL);
283 return (rvd->vdev_child[vdev]);
290 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
294 if (vd->vdev_guid == guid)
297 for (int c = 0; c < vd->vdev_children; c++)
298 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
306 vdev_count_leaves_impl(vdev_t *vd)
310 if (vd->vdev_ops->vdev_op_leaf)
313 for (int c = 0; c < vd->vdev_children; c++)
314 n += vdev_count_leaves_impl(vd->vdev_child[c]);
320 vdev_count_leaves(spa_t *spa)
322 return (vdev_count_leaves_impl(spa->spa_root_vdev));
326 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
328 size_t oldsize, newsize;
329 uint64_t id = cvd->vdev_id;
331 spa_t *spa = cvd->vdev_spa;
333 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
334 ASSERT(cvd->vdev_parent == NULL);
336 cvd->vdev_parent = pvd;
341 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
343 oldsize = pvd->vdev_children * sizeof (vdev_t *);
344 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
345 newsize = pvd->vdev_children * sizeof (vdev_t *);
347 newchild = kmem_zalloc(newsize, KM_SLEEP);
348 if (pvd->vdev_child != NULL) {
349 bcopy(pvd->vdev_child, newchild, oldsize);
350 kmem_free(pvd->vdev_child, oldsize);
353 pvd->vdev_child = newchild;
354 pvd->vdev_child[id] = cvd;
356 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
357 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
360 * Walk up all ancestors to update guid sum.
362 for (; pvd != NULL; pvd = pvd->vdev_parent)
363 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
367 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
370 uint_t id = cvd->vdev_id;
372 ASSERT(cvd->vdev_parent == pvd);
377 ASSERT(id < pvd->vdev_children);
378 ASSERT(pvd->vdev_child[id] == cvd);
380 pvd->vdev_child[id] = NULL;
381 cvd->vdev_parent = NULL;
383 for (c = 0; c < pvd->vdev_children; c++)
384 if (pvd->vdev_child[c])
387 if (c == pvd->vdev_children) {
388 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
389 pvd->vdev_child = NULL;
390 pvd->vdev_children = 0;
394 * Walk up all ancestors to update guid sum.
396 for (; pvd != NULL; pvd = pvd->vdev_parent)
397 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
401 * Remove any holes in the child array.
404 vdev_compact_children(vdev_t *pvd)
406 vdev_t **newchild, *cvd;
407 int oldc = pvd->vdev_children;
410 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
412 for (int c = newc = 0; c < oldc; c++)
413 if (pvd->vdev_child[c])
416 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
418 for (int c = newc = 0; c < oldc; c++) {
419 if ((cvd = pvd->vdev_child[c]) != NULL) {
420 newchild[newc] = cvd;
421 cvd->vdev_id = newc++;
425 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
426 pvd->vdev_child = newchild;
427 pvd->vdev_children = newc;
431 * Allocate and minimally initialize a vdev_t.
434 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
437 vdev_indirect_config_t *vic;
439 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
440 vic = &vd->vdev_indirect_config;
442 if (spa->spa_root_vdev == NULL) {
443 ASSERT(ops == &vdev_root_ops);
444 spa->spa_root_vdev = vd;
445 spa->spa_load_guid = spa_generate_guid(NULL);
448 if (guid == 0 && ops != &vdev_hole_ops) {
449 if (spa->spa_root_vdev == vd) {
451 * The root vdev's guid will also be the pool guid,
452 * which must be unique among all pools.
454 guid = spa_generate_guid(NULL);
457 * Any other vdev's guid must be unique within the pool.
459 guid = spa_generate_guid(spa);
461 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
466 vd->vdev_guid = guid;
467 vd->vdev_guid_sum = guid;
469 vd->vdev_state = VDEV_STATE_CLOSED;
470 vd->vdev_ishole = (ops == &vdev_hole_ops);
471 vic->vic_prev_indirect_vdev = UINT64_MAX;
473 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
474 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
475 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
477 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
478 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
479 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
480 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
481 for (int t = 0; t < DTL_TYPES; t++) {
482 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
484 txg_list_create(&vd->vdev_ms_list, spa,
485 offsetof(struct metaslab, ms_txg_node));
486 txg_list_create(&vd->vdev_dtl_list, spa,
487 offsetof(struct vdev, vdev_dtl_node));
488 vd->vdev_stat.vs_timestamp = gethrtime();
496 * Allocate a new vdev. The 'alloctype' is used to control whether we are
497 * creating a new vdev or loading an existing one - the behavior is slightly
498 * different for each case.
501 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
506 uint64_t guid = 0, islog, nparity;
508 vdev_indirect_config_t *vic;
510 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
512 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
513 return (SET_ERROR(EINVAL));
515 if ((ops = vdev_getops(type)) == NULL)
516 return (SET_ERROR(EINVAL));
519 * If this is a load, get the vdev guid from the nvlist.
520 * Otherwise, vdev_alloc_common() will generate one for us.
522 if (alloctype == VDEV_ALLOC_LOAD) {
525 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
527 return (SET_ERROR(EINVAL));
529 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
530 return (SET_ERROR(EINVAL));
531 } else if (alloctype == VDEV_ALLOC_SPARE) {
532 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
533 return (SET_ERROR(EINVAL));
534 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
535 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
536 return (SET_ERROR(EINVAL));
537 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
538 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
539 return (SET_ERROR(EINVAL));
543 * The first allocated vdev must be of type 'root'.
545 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
546 return (SET_ERROR(EINVAL));
549 * Determine whether we're a log vdev.
552 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
553 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
554 return (SET_ERROR(ENOTSUP));
556 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
557 return (SET_ERROR(ENOTSUP));
560 * Set the nparity property for RAID-Z vdevs.
563 if (ops == &vdev_raidz_ops) {
564 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
566 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
567 return (SET_ERROR(EINVAL));
569 * Previous versions could only support 1 or 2 parity
573 spa_version(spa) < SPA_VERSION_RAIDZ2)
574 return (SET_ERROR(ENOTSUP));
576 spa_version(spa) < SPA_VERSION_RAIDZ3)
577 return (SET_ERROR(ENOTSUP));
580 * We require the parity to be specified for SPAs that
581 * support multiple parity levels.
583 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
584 return (SET_ERROR(EINVAL));
586 * Otherwise, we default to 1 parity device for RAID-Z.
593 ASSERT(nparity != -1ULL);
595 vd = vdev_alloc_common(spa, id, guid, ops);
596 vic = &vd->vdev_indirect_config;
598 vd->vdev_islog = islog;
599 vd->vdev_nparity = nparity;
601 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
602 vd->vdev_path = spa_strdup(vd->vdev_path);
603 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
604 vd->vdev_devid = spa_strdup(vd->vdev_devid);
605 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
606 &vd->vdev_physpath) == 0)
607 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
608 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
609 vd->vdev_fru = spa_strdup(vd->vdev_fru);
612 * Set the whole_disk property. If it's not specified, leave the value
615 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
616 &vd->vdev_wholedisk) != 0)
617 vd->vdev_wholedisk = -1ULL;
619 ASSERT0(vic->vic_mapping_object);
620 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
621 &vic->vic_mapping_object);
622 ASSERT0(vic->vic_births_object);
623 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
624 &vic->vic_births_object);
625 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
626 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
627 &vic->vic_prev_indirect_vdev);
630 * Look for the 'not present' flag. This will only be set if the device
631 * was not present at the time of import.
633 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
634 &vd->vdev_not_present);
637 * Get the alignment requirement.
639 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
642 * Retrieve the vdev creation time.
644 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
648 * If we're a top-level vdev, try to load the allocation parameters.
650 if (parent && !parent->vdev_parent &&
651 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
652 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
654 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
656 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
658 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
660 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
663 ASSERT0(vd->vdev_top_zap);
666 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
667 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
668 alloctype == VDEV_ALLOC_ADD ||
669 alloctype == VDEV_ALLOC_SPLIT ||
670 alloctype == VDEV_ALLOC_ROOTPOOL);
671 vd->vdev_mg = metaslab_group_create(islog ?
672 spa_log_class(spa) : spa_normal_class(spa), vd);
675 if (vd->vdev_ops->vdev_op_leaf &&
676 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
677 (void) nvlist_lookup_uint64(nv,
678 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
680 ASSERT0(vd->vdev_leaf_zap);
684 * If we're a leaf vdev, try to load the DTL object and other state.
687 if (vd->vdev_ops->vdev_op_leaf &&
688 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
689 alloctype == VDEV_ALLOC_ROOTPOOL)) {
690 if (alloctype == VDEV_ALLOC_LOAD) {
691 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
692 &vd->vdev_dtl_object);
693 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
697 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
700 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
701 &spare) == 0 && spare)
705 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
708 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
709 &vd->vdev_resilver_txg);
712 * When importing a pool, we want to ignore the persistent fault
713 * state, as the diagnosis made on another system may not be
714 * valid in the current context. Local vdevs will
715 * remain in the faulted state.
717 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
718 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
720 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
722 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
725 if (vd->vdev_faulted || vd->vdev_degraded) {
729 VDEV_AUX_ERR_EXCEEDED;
730 if (nvlist_lookup_string(nv,
731 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
732 strcmp(aux, "external") == 0)
733 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
739 * Add ourselves to the parent's list of children.
741 vdev_add_child(parent, vd);
749 vdev_free(vdev_t *vd)
751 spa_t *spa = vd->vdev_spa;
754 * vdev_free() implies closing the vdev first. This is simpler than
755 * trying to ensure complicated semantics for all callers.
759 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
760 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
765 for (int c = 0; c < vd->vdev_children; c++)
766 vdev_free(vd->vdev_child[c]);
768 ASSERT(vd->vdev_child == NULL);
769 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
772 * Discard allocation state.
774 if (vd->vdev_mg != NULL) {
775 vdev_metaslab_fini(vd);
776 metaslab_group_destroy(vd->vdev_mg);
779 ASSERT0(vd->vdev_stat.vs_space);
780 ASSERT0(vd->vdev_stat.vs_dspace);
781 ASSERT0(vd->vdev_stat.vs_alloc);
784 * Remove this vdev from its parent's child list.
786 vdev_remove_child(vd->vdev_parent, vd);
788 ASSERT(vd->vdev_parent == NULL);
791 * Clean up vdev structure.
797 spa_strfree(vd->vdev_path);
799 spa_strfree(vd->vdev_devid);
800 if (vd->vdev_physpath)
801 spa_strfree(vd->vdev_physpath);
803 spa_strfree(vd->vdev_fru);
805 if (vd->vdev_isspare)
806 spa_spare_remove(vd);
807 if (vd->vdev_isl2cache)
808 spa_l2cache_remove(vd);
810 txg_list_destroy(&vd->vdev_ms_list);
811 txg_list_destroy(&vd->vdev_dtl_list);
813 mutex_enter(&vd->vdev_dtl_lock);
814 space_map_close(vd->vdev_dtl_sm);
815 for (int t = 0; t < DTL_TYPES; t++) {
816 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
817 range_tree_destroy(vd->vdev_dtl[t]);
819 mutex_exit(&vd->vdev_dtl_lock);
821 EQUIV(vd->vdev_indirect_births != NULL,
822 vd->vdev_indirect_mapping != NULL);
823 if (vd->vdev_indirect_births != NULL) {
824 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
825 vdev_indirect_births_close(vd->vdev_indirect_births);
828 if (vd->vdev_obsolete_sm != NULL) {
829 ASSERT(vd->vdev_removing ||
830 vd->vdev_ops == &vdev_indirect_ops);
831 space_map_close(vd->vdev_obsolete_sm);
832 vd->vdev_obsolete_sm = NULL;
834 range_tree_destroy(vd->vdev_obsolete_segments);
835 rw_destroy(&vd->vdev_indirect_rwlock);
836 mutex_destroy(&vd->vdev_obsolete_lock);
838 mutex_destroy(&vd->vdev_queue_lock);
839 mutex_destroy(&vd->vdev_dtl_lock);
840 mutex_destroy(&vd->vdev_stat_lock);
841 mutex_destroy(&vd->vdev_probe_lock);
843 if (vd == spa->spa_root_vdev)
844 spa->spa_root_vdev = NULL;
846 kmem_free(vd, sizeof (vdev_t));
850 * Transfer top-level vdev state from svd to tvd.
853 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
855 spa_t *spa = svd->vdev_spa;
860 ASSERT(tvd == tvd->vdev_top);
862 tvd->vdev_ms_array = svd->vdev_ms_array;
863 tvd->vdev_ms_shift = svd->vdev_ms_shift;
864 tvd->vdev_ms_count = svd->vdev_ms_count;
865 tvd->vdev_top_zap = svd->vdev_top_zap;
867 svd->vdev_ms_array = 0;
868 svd->vdev_ms_shift = 0;
869 svd->vdev_ms_count = 0;
870 svd->vdev_top_zap = 0;
873 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
874 tvd->vdev_mg = svd->vdev_mg;
875 tvd->vdev_ms = svd->vdev_ms;
880 if (tvd->vdev_mg != NULL)
881 tvd->vdev_mg->mg_vd = tvd;
883 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
884 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
885 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
887 svd->vdev_stat.vs_alloc = 0;
888 svd->vdev_stat.vs_space = 0;
889 svd->vdev_stat.vs_dspace = 0;
891 for (t = 0; t < TXG_SIZE; t++) {
892 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
893 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
894 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
895 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
896 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
897 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
900 if (list_link_active(&svd->vdev_config_dirty_node)) {
901 vdev_config_clean(svd);
902 vdev_config_dirty(tvd);
905 if (list_link_active(&svd->vdev_state_dirty_node)) {
906 vdev_state_clean(svd);
907 vdev_state_dirty(tvd);
910 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
911 svd->vdev_deflate_ratio = 0;
913 tvd->vdev_islog = svd->vdev_islog;
918 vdev_top_update(vdev_t *tvd, vdev_t *vd)
925 for (int c = 0; c < vd->vdev_children; c++)
926 vdev_top_update(tvd, vd->vdev_child[c]);
930 * Add a mirror/replacing vdev above an existing vdev.
933 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
935 spa_t *spa = cvd->vdev_spa;
936 vdev_t *pvd = cvd->vdev_parent;
939 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
941 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
943 mvd->vdev_asize = cvd->vdev_asize;
944 mvd->vdev_min_asize = cvd->vdev_min_asize;
945 mvd->vdev_max_asize = cvd->vdev_max_asize;
946 mvd->vdev_psize = cvd->vdev_psize;
947 mvd->vdev_ashift = cvd->vdev_ashift;
948 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
949 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
950 mvd->vdev_state = cvd->vdev_state;
951 mvd->vdev_crtxg = cvd->vdev_crtxg;
953 vdev_remove_child(pvd, cvd);
954 vdev_add_child(pvd, mvd);
955 cvd->vdev_id = mvd->vdev_children;
956 vdev_add_child(mvd, cvd);
957 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
959 if (mvd == mvd->vdev_top)
960 vdev_top_transfer(cvd, mvd);
966 * Remove a 1-way mirror/replacing vdev from the tree.
969 vdev_remove_parent(vdev_t *cvd)
971 vdev_t *mvd = cvd->vdev_parent;
972 vdev_t *pvd = mvd->vdev_parent;
974 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
976 ASSERT(mvd->vdev_children == 1);
977 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
978 mvd->vdev_ops == &vdev_replacing_ops ||
979 mvd->vdev_ops == &vdev_spare_ops);
980 cvd->vdev_ashift = mvd->vdev_ashift;
981 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
982 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
984 vdev_remove_child(mvd, cvd);
985 vdev_remove_child(pvd, mvd);
988 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
989 * Otherwise, we could have detached an offline device, and when we
990 * go to import the pool we'll think we have two top-level vdevs,
991 * instead of a different version of the same top-level vdev.
993 if (mvd->vdev_top == mvd) {
994 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
995 cvd->vdev_orig_guid = cvd->vdev_guid;
996 cvd->vdev_guid += guid_delta;
997 cvd->vdev_guid_sum += guid_delta;
999 cvd->vdev_id = mvd->vdev_id;
1000 vdev_add_child(pvd, cvd);
1001 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1003 if (cvd == cvd->vdev_top)
1004 vdev_top_transfer(mvd, cvd);
1006 ASSERT(mvd->vdev_children == 0);
1011 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1013 spa_t *spa = vd->vdev_spa;
1014 objset_t *mos = spa->spa_meta_objset;
1016 uint64_t oldc = vd->vdev_ms_count;
1017 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1021 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1024 * This vdev is not being allocated from yet or is a hole.
1026 if (vd->vdev_ms_shift == 0)
1029 ASSERT(!vd->vdev_ishole);
1031 ASSERT(oldc <= newc);
1033 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1036 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1037 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1041 vd->vdev_ms_count = newc;
1043 for (m = oldc; m < newc; m++) {
1044 uint64_t object = 0;
1047 * vdev_ms_array may be 0 if we are creating the "fake"
1048 * metaslabs for an indirect vdev for zdb's leak detection.
1049 * See zdb_leak_init().
1051 if (txg == 0 && vd->vdev_ms_array != 0) {
1052 error = dmu_read(mos, vd->vdev_ms_array,
1053 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1056 vdev_dbgmsg(vd, "unable to read the metaslab "
1057 "array [error=%d]", error);
1062 error = metaslab_init(vd->vdev_mg, m, object, txg,
1065 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1072 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1075 * If the vdev is being removed we don't activate
1076 * the metaslabs since we want to ensure that no new
1077 * allocations are performed on this device.
1079 if (oldc == 0 && !vd->vdev_removing)
1080 metaslab_group_activate(vd->vdev_mg);
1083 spa_config_exit(spa, SCL_ALLOC, FTAG);
1089 vdev_metaslab_fini(vdev_t *vd)
1091 if (vd->vdev_ms != NULL) {
1092 uint64_t count = vd->vdev_ms_count;
1094 metaslab_group_passivate(vd->vdev_mg);
1095 for (uint64_t m = 0; m < count; m++) {
1096 metaslab_t *msp = vd->vdev_ms[m];
1101 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1104 vd->vdev_ms_count = 0;
1106 ASSERT0(vd->vdev_ms_count);
1109 typedef struct vdev_probe_stats {
1110 boolean_t vps_readable;
1111 boolean_t vps_writeable;
1113 } vdev_probe_stats_t;
1116 vdev_probe_done(zio_t *zio)
1118 spa_t *spa = zio->io_spa;
1119 vdev_t *vd = zio->io_vd;
1120 vdev_probe_stats_t *vps = zio->io_private;
1122 ASSERT(vd->vdev_probe_zio != NULL);
1124 if (zio->io_type == ZIO_TYPE_READ) {
1125 if (zio->io_error == 0)
1126 vps->vps_readable = 1;
1127 if (zio->io_error == 0 && spa_writeable(spa)) {
1128 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1129 zio->io_offset, zio->io_size, zio->io_abd,
1130 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1131 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1133 abd_free(zio->io_abd);
1135 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1136 if (zio->io_error == 0)
1137 vps->vps_writeable = 1;
1138 abd_free(zio->io_abd);
1139 } else if (zio->io_type == ZIO_TYPE_NULL) {
1142 vd->vdev_cant_read |= !vps->vps_readable;
1143 vd->vdev_cant_write |= !vps->vps_writeable;
1145 if (vdev_readable(vd) &&
1146 (vdev_writeable(vd) || !spa_writeable(spa))) {
1149 ASSERT(zio->io_error != 0);
1150 vdev_dbgmsg(vd, "failed probe");
1151 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1152 spa, vd, NULL, 0, 0);
1153 zio->io_error = SET_ERROR(ENXIO);
1156 mutex_enter(&vd->vdev_probe_lock);
1157 ASSERT(vd->vdev_probe_zio == zio);
1158 vd->vdev_probe_zio = NULL;
1159 mutex_exit(&vd->vdev_probe_lock);
1161 zio_link_t *zl = NULL;
1162 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1163 if (!vdev_accessible(vd, pio))
1164 pio->io_error = SET_ERROR(ENXIO);
1166 kmem_free(vps, sizeof (*vps));
1171 * Determine whether this device is accessible.
1173 * Read and write to several known locations: the pad regions of each
1174 * vdev label but the first, which we leave alone in case it contains
1178 vdev_probe(vdev_t *vd, zio_t *zio)
1180 spa_t *spa = vd->vdev_spa;
1181 vdev_probe_stats_t *vps = NULL;
1184 ASSERT(vd->vdev_ops->vdev_op_leaf);
1187 * Don't probe the probe.
1189 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1193 * To prevent 'probe storms' when a device fails, we create
1194 * just one probe i/o at a time. All zios that want to probe
1195 * this vdev will become parents of the probe io.
1197 mutex_enter(&vd->vdev_probe_lock);
1199 if ((pio = vd->vdev_probe_zio) == NULL) {
1200 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1202 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1203 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1206 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1208 * vdev_cant_read and vdev_cant_write can only
1209 * transition from TRUE to FALSE when we have the
1210 * SCL_ZIO lock as writer; otherwise they can only
1211 * transition from FALSE to TRUE. This ensures that
1212 * any zio looking at these values can assume that
1213 * failures persist for the life of the I/O. That's
1214 * important because when a device has intermittent
1215 * connectivity problems, we want to ensure that
1216 * they're ascribed to the device (ENXIO) and not
1219 * Since we hold SCL_ZIO as writer here, clear both
1220 * values so the probe can reevaluate from first
1223 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1224 vd->vdev_cant_read = B_FALSE;
1225 vd->vdev_cant_write = B_FALSE;
1228 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1229 vdev_probe_done, vps,
1230 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1233 * We can't change the vdev state in this context, so we
1234 * kick off an async task to do it on our behalf.
1237 vd->vdev_probe_wanted = B_TRUE;
1238 spa_async_request(spa, SPA_ASYNC_PROBE);
1243 zio_add_child(zio, pio);
1245 mutex_exit(&vd->vdev_probe_lock);
1248 ASSERT(zio != NULL);
1252 for (int l = 1; l < VDEV_LABELS; l++) {
1253 zio_nowait(zio_read_phys(pio, vd,
1254 vdev_label_offset(vd->vdev_psize, l,
1255 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1256 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1257 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1258 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1269 vdev_open_child(void *arg)
1273 vd->vdev_open_thread = curthread;
1274 vd->vdev_open_error = vdev_open(vd);
1275 vd->vdev_open_thread = NULL;
1279 vdev_uses_zvols(vdev_t *vd)
1281 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1282 strlen(ZVOL_DIR)) == 0)
1284 for (int c = 0; c < vd->vdev_children; c++)
1285 if (vdev_uses_zvols(vd->vdev_child[c]))
1291 vdev_open_children(vdev_t *vd)
1294 int children = vd->vdev_children;
1297 * in order to handle pools on top of zvols, do the opens
1298 * in a single thread so that the same thread holds the
1299 * spa_namespace_lock
1301 if (B_TRUE || vdev_uses_zvols(vd)) {
1302 for (int c = 0; c < children; c++)
1303 vd->vdev_child[c]->vdev_open_error =
1304 vdev_open(vd->vdev_child[c]);
1307 tq = taskq_create("vdev_open", children, minclsyspri,
1308 children, children, TASKQ_PREPOPULATE);
1310 for (int c = 0; c < children; c++)
1311 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1318 * Compute the raidz-deflation ratio. Note, we hard-code
1319 * in 128k (1 << 17) because it is the "typical" blocksize.
1320 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1321 * otherwise it would inconsistently account for existing bp's.
1324 vdev_set_deflate_ratio(vdev_t *vd)
1326 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1327 vd->vdev_deflate_ratio = (1 << 17) /
1328 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1333 * Prepare a virtual device for access.
1336 vdev_open(vdev_t *vd)
1338 spa_t *spa = vd->vdev_spa;
1341 uint64_t max_osize = 0;
1342 uint64_t asize, max_asize, psize;
1343 uint64_t logical_ashift = 0;
1344 uint64_t physical_ashift = 0;
1346 ASSERT(vd->vdev_open_thread == curthread ||
1347 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1348 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1349 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1350 vd->vdev_state == VDEV_STATE_OFFLINE);
1352 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1353 vd->vdev_cant_read = B_FALSE;
1354 vd->vdev_cant_write = B_FALSE;
1355 vd->vdev_notrim = B_FALSE;
1356 vd->vdev_min_asize = vdev_get_min_asize(vd);
1359 * If this vdev is not removed, check its fault status. If it's
1360 * faulted, bail out of the open.
1362 if (!vd->vdev_removed && vd->vdev_faulted) {
1363 ASSERT(vd->vdev_children == 0);
1364 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1365 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1366 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1367 vd->vdev_label_aux);
1368 return (SET_ERROR(ENXIO));
1369 } else if (vd->vdev_offline) {
1370 ASSERT(vd->vdev_children == 0);
1371 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1372 return (SET_ERROR(ENXIO));
1375 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1376 &logical_ashift, &physical_ashift);
1379 * Reset the vdev_reopening flag so that we actually close
1380 * the vdev on error.
1382 vd->vdev_reopening = B_FALSE;
1383 if (zio_injection_enabled && error == 0)
1384 error = zio_handle_device_injection(vd, NULL, ENXIO);
1387 if (vd->vdev_removed &&
1388 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1389 vd->vdev_removed = B_FALSE;
1391 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1392 vd->vdev_stat.vs_aux);
1396 vd->vdev_removed = B_FALSE;
1399 * Recheck the faulted flag now that we have confirmed that
1400 * the vdev is accessible. If we're faulted, bail.
1402 if (vd->vdev_faulted) {
1403 ASSERT(vd->vdev_children == 0);
1404 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1405 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1406 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1407 vd->vdev_label_aux);
1408 return (SET_ERROR(ENXIO));
1411 if (vd->vdev_degraded) {
1412 ASSERT(vd->vdev_children == 0);
1413 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1414 VDEV_AUX_ERR_EXCEEDED);
1416 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1420 * For hole or missing vdevs we just return success.
1422 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1425 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1426 trim_map_create(vd);
1428 for (int c = 0; c < vd->vdev_children; c++) {
1429 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1430 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1436 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1437 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1439 if (vd->vdev_children == 0) {
1440 if (osize < SPA_MINDEVSIZE) {
1441 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1442 VDEV_AUX_TOO_SMALL);
1443 return (SET_ERROR(EOVERFLOW));
1446 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1447 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1448 VDEV_LABEL_END_SIZE);
1450 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1451 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1452 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1453 VDEV_AUX_TOO_SMALL);
1454 return (SET_ERROR(EOVERFLOW));
1458 max_asize = max_osize;
1461 vd->vdev_psize = psize;
1464 * Make sure the allocatable size hasn't shrunk too much.
1466 if (asize < vd->vdev_min_asize) {
1467 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1468 VDEV_AUX_BAD_LABEL);
1469 return (SET_ERROR(EINVAL));
1472 vd->vdev_physical_ashift =
1473 MAX(physical_ashift, vd->vdev_physical_ashift);
1474 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1475 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1477 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1478 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1479 VDEV_AUX_ASHIFT_TOO_BIG);
1483 if (vd->vdev_asize == 0) {
1485 * This is the first-ever open, so use the computed values.
1486 * For testing purposes, a higher ashift can be requested.
1488 vd->vdev_asize = asize;
1489 vd->vdev_max_asize = max_asize;
1492 * Make sure the alignment requirement hasn't increased.
1494 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1495 vd->vdev_ops->vdev_op_leaf) {
1496 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1497 VDEV_AUX_BAD_LABEL);
1500 vd->vdev_max_asize = max_asize;
1504 * If all children are healthy we update asize if either:
1505 * The asize has increased, due to a device expansion caused by dynamic
1506 * LUN growth or vdev replacement, and automatic expansion is enabled;
1507 * making the additional space available.
1509 * The asize has decreased, due to a device shrink usually caused by a
1510 * vdev replace with a smaller device. This ensures that calculations
1511 * based of max_asize and asize e.g. esize are always valid. It's safe
1512 * to do this as we've already validated that asize is greater than
1515 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1516 ((asize > vd->vdev_asize &&
1517 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1518 (asize < vd->vdev_asize)))
1519 vd->vdev_asize = asize;
1521 vdev_set_min_asize(vd);
1524 * Ensure we can issue some IO before declaring the
1525 * vdev open for business.
1527 if (vd->vdev_ops->vdev_op_leaf &&
1528 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1529 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1530 VDEV_AUX_ERR_EXCEEDED);
1534 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1535 !vd->vdev_isl2cache && !vd->vdev_islog) {
1536 if (vd->vdev_ashift > spa->spa_max_ashift)
1537 spa->spa_max_ashift = vd->vdev_ashift;
1538 if (vd->vdev_ashift < spa->spa_min_ashift)
1539 spa->spa_min_ashift = vd->vdev_ashift;
1543 * Track the min and max ashift values for normal data devices.
1545 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1546 !vd->vdev_islog && vd->vdev_aux == NULL) {
1547 if (vd->vdev_ashift > spa->spa_max_ashift)
1548 spa->spa_max_ashift = vd->vdev_ashift;
1549 if (vd->vdev_ashift < spa->spa_min_ashift)
1550 spa->spa_min_ashift = vd->vdev_ashift;
1554 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1555 * resilver. But don't do this if we are doing a reopen for a scrub,
1556 * since this would just restart the scrub we are already doing.
1558 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1559 vdev_resilver_needed(vd, NULL, NULL))
1560 spa_async_request(spa, SPA_ASYNC_RESILVER);
1566 * Called once the vdevs are all opened, this routine validates the label
1567 * contents. This needs to be done before vdev_load() so that we don't
1568 * inadvertently do repair I/Os to the wrong device.
1570 * If 'strict' is false ignore the spa guid check. This is necessary because
1571 * if the machine crashed during a re-guid the new guid might have been written
1572 * to all of the vdev labels, but not the cached config. The strict check
1573 * will be performed when the pool is opened again using the mos config.
1575 * This function will only return failure if one of the vdevs indicates that it
1576 * has since been destroyed or exported. This is only possible if
1577 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1578 * will be updated but the function will return 0.
1581 vdev_validate(vdev_t *vd, boolean_t strict)
1583 spa_t *spa = vd->vdev_spa;
1585 uint64_t guid = 0, top_guid;
1588 for (int c = 0; c < vd->vdev_children; c++)
1589 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1590 return (SET_ERROR(EBADF));
1593 * If the device has already failed, or was marked offline, don't do
1594 * any further validation. Otherwise, label I/O will fail and we will
1595 * overwrite the previous state.
1597 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1598 uint64_t aux_guid = 0;
1600 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1601 spa_last_synced_txg(spa) : -1ULL;
1603 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1604 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1605 VDEV_AUX_BAD_LABEL);
1606 vdev_dbgmsg(vd, "vdev_validate: failed reading config");
1611 * Determine if this vdev has been split off into another
1612 * pool. If so, then refuse to open it.
1614 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1615 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1616 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1617 VDEV_AUX_SPLIT_POOL);
1619 vdev_dbgmsg(vd, "vdev_validate: vdev split into other "
1624 if (strict && (nvlist_lookup_uint64(label,
1625 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1626 guid != spa_guid(spa))) {
1627 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1628 VDEV_AUX_CORRUPT_DATA);
1630 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid "
1631 "doesn't match config (%llu != %llu)",
1633 (u_longlong_t)spa_guid(spa));
1637 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1638 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1643 * If this vdev just became a top-level vdev because its
1644 * sibling was detached, it will have adopted the parent's
1645 * vdev guid -- but the label may or may not be on disk yet.
1646 * Fortunately, either version of the label will have the
1647 * same top guid, so if we're a top-level vdev, we can
1648 * safely compare to that instead.
1650 * If we split this vdev off instead, then we also check the
1651 * original pool's guid. We don't want to consider the vdev
1652 * corrupt if it is partway through a split operation.
1654 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1656 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1658 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1659 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1660 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1661 VDEV_AUX_CORRUPT_DATA);
1663 vdev_dbgmsg(vd, "vdev_validate: config guid doesn't "
1664 "match label guid (%llu != %llu)",
1665 (u_longlong_t)vd->vdev_guid, (u_longlong_t)guid);
1669 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1671 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1672 VDEV_AUX_CORRUPT_DATA);
1674 vdev_dbgmsg(vd, "vdev_validate: '%s' missing",
1675 ZPOOL_CONFIG_POOL_STATE);
1682 * If this is a verbatim import, no need to check the
1683 * state of the pool.
1685 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1686 spa_load_state(spa) == SPA_LOAD_OPEN &&
1687 state != POOL_STATE_ACTIVE) {
1688 vdev_dbgmsg(vd, "vdev_validate: invalid pool state "
1689 "(%llu) for spa %s", (u_longlong_t)state,
1691 return (SET_ERROR(EBADF));
1695 * If we were able to open and validate a vdev that was
1696 * previously marked permanently unavailable, clear that state
1699 if (vd->vdev_not_present)
1700 vd->vdev_not_present = 0;
1707 * Close a virtual device.
1710 vdev_close(vdev_t *vd)
1712 spa_t *spa = vd->vdev_spa;
1713 vdev_t *pvd = vd->vdev_parent;
1715 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1718 * If our parent is reopening, then we are as well, unless we are
1721 if (pvd != NULL && pvd->vdev_reopening)
1722 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1724 vd->vdev_ops->vdev_op_close(vd);
1726 vdev_cache_purge(vd);
1728 if (vd->vdev_ops->vdev_op_leaf)
1729 trim_map_destroy(vd);
1732 * We record the previous state before we close it, so that if we are
1733 * doing a reopen(), we don't generate FMA ereports if we notice that
1734 * it's still faulted.
1736 vd->vdev_prevstate = vd->vdev_state;
1738 if (vd->vdev_offline)
1739 vd->vdev_state = VDEV_STATE_OFFLINE;
1741 vd->vdev_state = VDEV_STATE_CLOSED;
1742 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1746 vdev_hold(vdev_t *vd)
1748 spa_t *spa = vd->vdev_spa;
1750 ASSERT(spa_is_root(spa));
1751 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1754 for (int c = 0; c < vd->vdev_children; c++)
1755 vdev_hold(vd->vdev_child[c]);
1757 if (vd->vdev_ops->vdev_op_leaf)
1758 vd->vdev_ops->vdev_op_hold(vd);
1762 vdev_rele(vdev_t *vd)
1764 spa_t *spa = vd->vdev_spa;
1766 ASSERT(spa_is_root(spa));
1767 for (int c = 0; c < vd->vdev_children; c++)
1768 vdev_rele(vd->vdev_child[c]);
1770 if (vd->vdev_ops->vdev_op_leaf)
1771 vd->vdev_ops->vdev_op_rele(vd);
1775 * Reopen all interior vdevs and any unopened leaves. We don't actually
1776 * reopen leaf vdevs which had previously been opened as they might deadlock
1777 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1778 * If the leaf has never been opened then open it, as usual.
1781 vdev_reopen(vdev_t *vd)
1783 spa_t *spa = vd->vdev_spa;
1785 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1787 /* set the reopening flag unless we're taking the vdev offline */
1788 vd->vdev_reopening = !vd->vdev_offline;
1790 (void) vdev_open(vd);
1793 * Call vdev_validate() here to make sure we have the same device.
1794 * Otherwise, a device with an invalid label could be successfully
1795 * opened in response to vdev_reopen().
1798 (void) vdev_validate_aux(vd);
1799 if (vdev_readable(vd) && vdev_writeable(vd) &&
1800 vd->vdev_aux == &spa->spa_l2cache &&
1801 !l2arc_vdev_present(vd))
1802 l2arc_add_vdev(spa, vd);
1804 (void) vdev_validate(vd, B_TRUE);
1808 * Reassess parent vdev's health.
1810 vdev_propagate_state(vd);
1814 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1819 * Normally, partial opens (e.g. of a mirror) are allowed.
1820 * For a create, however, we want to fail the request if
1821 * there are any components we can't open.
1823 error = vdev_open(vd);
1825 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1827 return (error ? error : ENXIO);
1831 * Recursively load DTLs and initialize all labels.
1833 if ((error = vdev_dtl_load(vd)) != 0 ||
1834 (error = vdev_label_init(vd, txg, isreplacing ?
1835 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1844 vdev_metaslab_set_size(vdev_t *vd)
1847 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1849 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1850 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1854 * Maximize performance by inflating the configured ashift for top level
1855 * vdevs to be as close to the physical ashift as possible while maintaining
1856 * administrator defined limits and ensuring it doesn't go below the
1860 vdev_ashift_optimize(vdev_t *vd)
1862 if (vd == vd->vdev_top) {
1863 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1864 vd->vdev_ashift = MIN(
1865 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1866 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1869 * Unusual case where logical ashift > physical ashift
1870 * so we can't cap the calculated ashift based on max
1871 * ashift as that would cause failures.
1872 * We still check if we need to increase it to match
1875 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1882 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1884 ASSERT(vd == vd->vdev_top);
1885 /* indirect vdevs don't have metaslabs or dtls */
1886 ASSERT(vdev_is_concrete(vd) || flags == 0);
1887 ASSERT(ISP2(flags));
1888 ASSERT(spa_writeable(vd->vdev_spa));
1890 if (flags & VDD_METASLAB)
1891 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1893 if (flags & VDD_DTL)
1894 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1896 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1900 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1902 for (int c = 0; c < vd->vdev_children; c++)
1903 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1905 if (vd->vdev_ops->vdev_op_leaf)
1906 vdev_dirty(vd->vdev_top, flags, vd, txg);
1912 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1913 * the vdev has less than perfect replication. There are four kinds of DTL:
1915 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1917 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1919 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1920 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1921 * txgs that was scrubbed.
1923 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1924 * persistent errors or just some device being offline.
1925 * Unlike the other three, the DTL_OUTAGE map is not generally
1926 * maintained; it's only computed when needed, typically to
1927 * determine whether a device can be detached.
1929 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1930 * either has the data or it doesn't.
1932 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1933 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1934 * if any child is less than fully replicated, then so is its parent.
1935 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1936 * comprising only those txgs which appear in 'maxfaults' or more children;
1937 * those are the txgs we don't have enough replication to read. For example,
1938 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1939 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1940 * two child DTL_MISSING maps.
1942 * It should be clear from the above that to compute the DTLs and outage maps
1943 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1944 * Therefore, that is all we keep on disk. When loading the pool, or after
1945 * a configuration change, we generate all other DTLs from first principles.
1948 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1950 range_tree_t *rt = vd->vdev_dtl[t];
1952 ASSERT(t < DTL_TYPES);
1953 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1954 ASSERT(spa_writeable(vd->vdev_spa));
1956 mutex_enter(&vd->vdev_dtl_lock);
1957 if (!range_tree_contains(rt, txg, size))
1958 range_tree_add(rt, txg, size);
1959 mutex_exit(&vd->vdev_dtl_lock);
1963 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1965 range_tree_t *rt = vd->vdev_dtl[t];
1966 boolean_t dirty = B_FALSE;
1968 ASSERT(t < DTL_TYPES);
1969 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1972 * While we are loading the pool, the DTLs have not been loaded yet.
1973 * Ignore the DTLs and try all devices. This avoids a recursive
1974 * mutex enter on the vdev_dtl_lock, and also makes us try hard
1975 * when loading the pool (relying on the checksum to ensure that
1976 * we get the right data -- note that we while loading, we are
1977 * only reading the MOS, which is always checksummed).
1979 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
1982 mutex_enter(&vd->vdev_dtl_lock);
1983 if (range_tree_space(rt) != 0)
1984 dirty = range_tree_contains(rt, txg, size);
1985 mutex_exit(&vd->vdev_dtl_lock);
1991 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1993 range_tree_t *rt = vd->vdev_dtl[t];
1996 mutex_enter(&vd->vdev_dtl_lock);
1997 empty = (range_tree_space(rt) == 0);
1998 mutex_exit(&vd->vdev_dtl_lock);
2004 * Returns the lowest txg in the DTL range.
2007 vdev_dtl_min(vdev_t *vd)
2011 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2012 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2013 ASSERT0(vd->vdev_children);
2015 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2016 return (rs->rs_start - 1);
2020 * Returns the highest txg in the DTL.
2023 vdev_dtl_max(vdev_t *vd)
2027 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2028 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2029 ASSERT0(vd->vdev_children);
2031 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2032 return (rs->rs_end);
2036 * Determine if a resilvering vdev should remove any DTL entries from
2037 * its range. If the vdev was resilvering for the entire duration of the
2038 * scan then it should excise that range from its DTLs. Otherwise, this
2039 * vdev is considered partially resilvered and should leave its DTL
2040 * entries intact. The comment in vdev_dtl_reassess() describes how we
2044 vdev_dtl_should_excise(vdev_t *vd)
2046 spa_t *spa = vd->vdev_spa;
2047 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2049 ASSERT0(scn->scn_phys.scn_errors);
2050 ASSERT0(vd->vdev_children);
2052 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2055 if (vd->vdev_resilver_txg == 0 ||
2056 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
2060 * When a resilver is initiated the scan will assign the scn_max_txg
2061 * value to the highest txg value that exists in all DTLs. If this
2062 * device's max DTL is not part of this scan (i.e. it is not in
2063 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2066 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2067 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2068 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2069 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2076 * Reassess DTLs after a config change or scrub completion.
2079 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2081 spa_t *spa = vd->vdev_spa;
2085 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2087 for (int c = 0; c < vd->vdev_children; c++)
2088 vdev_dtl_reassess(vd->vdev_child[c], txg,
2089 scrub_txg, scrub_done);
2091 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2094 if (vd->vdev_ops->vdev_op_leaf) {
2095 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2097 mutex_enter(&vd->vdev_dtl_lock);
2100 * If we've completed a scan cleanly then determine
2101 * if this vdev should remove any DTLs. We only want to
2102 * excise regions on vdevs that were available during
2103 * the entire duration of this scan.
2105 if (scrub_txg != 0 &&
2106 (spa->spa_scrub_started ||
2107 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2108 vdev_dtl_should_excise(vd)) {
2110 * We completed a scrub up to scrub_txg. If we
2111 * did it without rebooting, then the scrub dtl
2112 * will be valid, so excise the old region and
2113 * fold in the scrub dtl. Otherwise, leave the
2114 * dtl as-is if there was an error.
2116 * There's little trick here: to excise the beginning
2117 * of the DTL_MISSING map, we put it into a reference
2118 * tree and then add a segment with refcnt -1 that
2119 * covers the range [0, scrub_txg). This means
2120 * that each txg in that range has refcnt -1 or 0.
2121 * We then add DTL_SCRUB with a refcnt of 2, so that
2122 * entries in the range [0, scrub_txg) will have a
2123 * positive refcnt -- either 1 or 2. We then convert
2124 * the reference tree into the new DTL_MISSING map.
2126 space_reftree_create(&reftree);
2127 space_reftree_add_map(&reftree,
2128 vd->vdev_dtl[DTL_MISSING], 1);
2129 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2130 space_reftree_add_map(&reftree,
2131 vd->vdev_dtl[DTL_SCRUB], 2);
2132 space_reftree_generate_map(&reftree,
2133 vd->vdev_dtl[DTL_MISSING], 1);
2134 space_reftree_destroy(&reftree);
2136 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2137 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2138 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2140 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2141 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2142 if (!vdev_readable(vd))
2143 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2145 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2146 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2149 * If the vdev was resilvering and no longer has any
2150 * DTLs then reset its resilvering flag and dirty
2151 * the top level so that we persist the change.
2153 if (vd->vdev_resilver_txg != 0 &&
2154 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
2155 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
2156 vd->vdev_resilver_txg = 0;
2157 vdev_config_dirty(vd->vdev_top);
2160 mutex_exit(&vd->vdev_dtl_lock);
2163 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2167 mutex_enter(&vd->vdev_dtl_lock);
2168 for (int t = 0; t < DTL_TYPES; t++) {
2169 /* account for child's outage in parent's missing map */
2170 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2172 continue; /* leaf vdevs only */
2173 if (t == DTL_PARTIAL)
2174 minref = 1; /* i.e. non-zero */
2175 else if (vd->vdev_nparity != 0)
2176 minref = vd->vdev_nparity + 1; /* RAID-Z */
2178 minref = vd->vdev_children; /* any kind of mirror */
2179 space_reftree_create(&reftree);
2180 for (int c = 0; c < vd->vdev_children; c++) {
2181 vdev_t *cvd = vd->vdev_child[c];
2182 mutex_enter(&cvd->vdev_dtl_lock);
2183 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2184 mutex_exit(&cvd->vdev_dtl_lock);
2186 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2187 space_reftree_destroy(&reftree);
2189 mutex_exit(&vd->vdev_dtl_lock);
2193 vdev_dtl_load(vdev_t *vd)
2195 spa_t *spa = vd->vdev_spa;
2196 objset_t *mos = spa->spa_meta_objset;
2199 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2200 ASSERT(vdev_is_concrete(vd));
2202 error = space_map_open(&vd->vdev_dtl_sm, mos,
2203 vd->vdev_dtl_object, 0, -1ULL, 0);
2206 ASSERT(vd->vdev_dtl_sm != NULL);
2208 mutex_enter(&vd->vdev_dtl_lock);
2211 * Now that we've opened the space_map we need to update
2214 space_map_update(vd->vdev_dtl_sm);
2216 error = space_map_load(vd->vdev_dtl_sm,
2217 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2218 mutex_exit(&vd->vdev_dtl_lock);
2223 for (int c = 0; c < vd->vdev_children; c++) {
2224 error = vdev_dtl_load(vd->vdev_child[c]);
2233 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2235 spa_t *spa = vd->vdev_spa;
2237 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2238 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2243 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2245 spa_t *spa = vd->vdev_spa;
2246 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2247 DMU_OT_NONE, 0, tx);
2250 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2257 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2259 if (vd->vdev_ops != &vdev_hole_ops &&
2260 vd->vdev_ops != &vdev_missing_ops &&
2261 vd->vdev_ops != &vdev_root_ops &&
2262 !vd->vdev_top->vdev_removing) {
2263 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2264 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2266 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2267 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2270 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2271 vdev_construct_zaps(vd->vdev_child[i], tx);
2276 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2278 spa_t *spa = vd->vdev_spa;
2279 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2280 objset_t *mos = spa->spa_meta_objset;
2281 range_tree_t *rtsync;
2283 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2285 ASSERT(vdev_is_concrete(vd));
2286 ASSERT(vd->vdev_ops->vdev_op_leaf);
2288 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2290 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2291 mutex_enter(&vd->vdev_dtl_lock);
2292 space_map_free(vd->vdev_dtl_sm, tx);
2293 space_map_close(vd->vdev_dtl_sm);
2294 vd->vdev_dtl_sm = NULL;
2295 mutex_exit(&vd->vdev_dtl_lock);
2298 * We only destroy the leaf ZAP for detached leaves or for
2299 * removed log devices. Removed data devices handle leaf ZAP
2300 * cleanup later, once cancellation is no longer possible.
2302 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2303 vd->vdev_top->vdev_islog)) {
2304 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2305 vd->vdev_leaf_zap = 0;
2312 if (vd->vdev_dtl_sm == NULL) {
2313 uint64_t new_object;
2315 new_object = space_map_alloc(mos, tx);
2316 VERIFY3U(new_object, !=, 0);
2318 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2320 ASSERT(vd->vdev_dtl_sm != NULL);
2323 rtsync = range_tree_create(NULL, NULL);
2325 mutex_enter(&vd->vdev_dtl_lock);
2326 range_tree_walk(rt, range_tree_add, rtsync);
2327 mutex_exit(&vd->vdev_dtl_lock);
2329 space_map_truncate(vd->vdev_dtl_sm, tx);
2330 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2331 range_tree_vacate(rtsync, NULL, NULL);
2333 range_tree_destroy(rtsync);
2336 * If the object for the space map has changed then dirty
2337 * the top level so that we update the config.
2339 if (object != space_map_object(vd->vdev_dtl_sm)) {
2340 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2341 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2342 (u_longlong_t)object,
2343 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2344 vdev_config_dirty(vd->vdev_top);
2349 mutex_enter(&vd->vdev_dtl_lock);
2350 space_map_update(vd->vdev_dtl_sm);
2351 mutex_exit(&vd->vdev_dtl_lock);
2355 * Determine whether the specified vdev can be offlined/detached/removed
2356 * without losing data.
2359 vdev_dtl_required(vdev_t *vd)
2361 spa_t *spa = vd->vdev_spa;
2362 vdev_t *tvd = vd->vdev_top;
2363 uint8_t cant_read = vd->vdev_cant_read;
2366 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2368 if (vd == spa->spa_root_vdev || vd == tvd)
2372 * Temporarily mark the device as unreadable, and then determine
2373 * whether this results in any DTL outages in the top-level vdev.
2374 * If not, we can safely offline/detach/remove the device.
2376 vd->vdev_cant_read = B_TRUE;
2377 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2378 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2379 vd->vdev_cant_read = cant_read;
2380 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2382 if (!required && zio_injection_enabled)
2383 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2389 * Determine if resilver is needed, and if so the txg range.
2392 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2394 boolean_t needed = B_FALSE;
2395 uint64_t thismin = UINT64_MAX;
2396 uint64_t thismax = 0;
2398 if (vd->vdev_children == 0) {
2399 mutex_enter(&vd->vdev_dtl_lock);
2400 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2401 vdev_writeable(vd)) {
2403 thismin = vdev_dtl_min(vd);
2404 thismax = vdev_dtl_max(vd);
2407 mutex_exit(&vd->vdev_dtl_lock);
2409 for (int c = 0; c < vd->vdev_children; c++) {
2410 vdev_t *cvd = vd->vdev_child[c];
2411 uint64_t cmin, cmax;
2413 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2414 thismin = MIN(thismin, cmin);
2415 thismax = MAX(thismax, cmax);
2421 if (needed && minp) {
2429 vdev_load(vdev_t *vd)
2433 * Recursively load all children.
2435 for (int c = 0; c < vd->vdev_children; c++) {
2436 error = vdev_load(vd->vdev_child[c]);
2442 vdev_set_deflate_ratio(vd);
2445 * If this is a top-level vdev, initialize its metaslabs.
2447 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2448 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2449 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2450 VDEV_AUX_CORRUPT_DATA);
2451 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2452 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2453 (u_longlong_t)vd->vdev_asize);
2454 return (SET_ERROR(ENXIO));
2455 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2456 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2457 "[error=%d]", error);
2458 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2459 VDEV_AUX_CORRUPT_DATA);
2465 * If this is a leaf vdev, load its DTL.
2467 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2468 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2469 VDEV_AUX_CORRUPT_DATA);
2470 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
2471 "[error=%d]", error);
2475 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2476 if (obsolete_sm_object != 0) {
2477 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2478 ASSERT(vd->vdev_asize != 0);
2479 ASSERT(vd->vdev_obsolete_sm == NULL);
2481 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2482 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2483 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2484 VDEV_AUX_CORRUPT_DATA);
2485 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
2486 "obsolete spacemap (obj %llu) [error=%d]",
2487 (u_longlong_t)obsolete_sm_object, error);
2490 space_map_update(vd->vdev_obsolete_sm);
2497 * The special vdev case is used for hot spares and l2cache devices. Its
2498 * sole purpose it to set the vdev state for the associated vdev. To do this,
2499 * we make sure that we can open the underlying device, then try to read the
2500 * label, and make sure that the label is sane and that it hasn't been
2501 * repurposed to another pool.
2504 vdev_validate_aux(vdev_t *vd)
2507 uint64_t guid, version;
2510 if (!vdev_readable(vd))
2513 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2514 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2515 VDEV_AUX_CORRUPT_DATA);
2519 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2520 !SPA_VERSION_IS_SUPPORTED(version) ||
2521 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2522 guid != vd->vdev_guid ||
2523 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2524 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2525 VDEV_AUX_CORRUPT_DATA);
2531 * We don't actually check the pool state here. If it's in fact in
2532 * use by another pool, we update this fact on the fly when requested.
2539 * Free the objects used to store this vdev's spacemaps, and the array
2540 * that points to them.
2543 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
2545 if (vd->vdev_ms_array == 0)
2548 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2549 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
2550 size_t array_bytes = array_count * sizeof (uint64_t);
2551 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
2552 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
2553 array_bytes, smobj_array, 0));
2555 for (uint64_t i = 0; i < array_count; i++) {
2556 uint64_t smobj = smobj_array[i];
2560 space_map_free_obj(mos, smobj, tx);
2563 kmem_free(smobj_array, array_bytes);
2564 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
2565 vd->vdev_ms_array = 0;
2569 vdev_remove_empty(vdev_t *vd, uint64_t txg)
2571 spa_t *spa = vd->vdev_spa;
2574 ASSERT(vd == vd->vdev_top);
2575 ASSERT3U(txg, ==, spa_syncing_txg(spa));
2577 if (vd->vdev_ms != NULL) {
2578 metaslab_group_t *mg = vd->vdev_mg;
2580 metaslab_group_histogram_verify(mg);
2581 metaslab_class_histogram_verify(mg->mg_class);
2583 for (int m = 0; m < vd->vdev_ms_count; m++) {
2584 metaslab_t *msp = vd->vdev_ms[m];
2586 if (msp == NULL || msp->ms_sm == NULL)
2589 mutex_enter(&msp->ms_lock);
2591 * If the metaslab was not loaded when the vdev
2592 * was removed then the histogram accounting may
2593 * not be accurate. Update the histogram information
2594 * here so that we ensure that the metaslab group
2595 * and metaslab class are up-to-date.
2597 metaslab_group_histogram_remove(mg, msp);
2599 VERIFY0(space_map_allocated(msp->ms_sm));
2600 space_map_close(msp->ms_sm);
2602 mutex_exit(&msp->ms_lock);
2605 metaslab_group_histogram_verify(mg);
2606 metaslab_class_histogram_verify(mg->mg_class);
2607 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2608 ASSERT0(mg->mg_histogram[i]);
2611 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2612 vdev_destroy_spacemaps(vd, tx);
2614 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
2615 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
2616 vd->vdev_top_zap = 0;
2622 vdev_sync_done(vdev_t *vd, uint64_t txg)
2625 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2627 ASSERT(vdev_is_concrete(vd));
2629 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2630 metaslab_sync_done(msp, txg);
2633 metaslab_sync_reassess(vd->vdev_mg);
2637 vdev_sync(vdev_t *vd, uint64_t txg)
2639 spa_t *spa = vd->vdev_spa;
2644 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
2647 ASSERT(vd->vdev_removing ||
2648 vd->vdev_ops == &vdev_indirect_ops);
2650 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2651 vdev_indirect_sync_obsolete(vd, tx);
2655 * If the vdev is indirect, it can't have dirty
2656 * metaslabs or DTLs.
2658 if (vd->vdev_ops == &vdev_indirect_ops) {
2659 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
2660 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
2665 ASSERT(vdev_is_concrete(vd));
2667 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
2668 !vd->vdev_removing) {
2669 ASSERT(vd == vd->vdev_top);
2670 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
2671 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2672 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2673 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2674 ASSERT(vd->vdev_ms_array != 0);
2675 vdev_config_dirty(vd);
2679 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2680 metaslab_sync(msp, txg);
2681 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2684 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2685 vdev_dtl_sync(lvd, txg);
2688 * Remove the metadata associated with this vdev once it's empty.
2689 * Note that this is typically used for log/cache device removal;
2690 * we don't empty toplevel vdevs when removing them. But if
2691 * a toplevel happens to be emptied, this is not harmful.
2693 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) {
2694 vdev_remove_empty(vd, txg);
2697 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2701 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2703 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2707 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2708 * not be opened, and no I/O is attempted.
2711 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2715 spa_vdev_state_enter(spa, SCL_NONE);
2717 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2718 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2720 if (!vd->vdev_ops->vdev_op_leaf)
2721 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2726 * We don't directly use the aux state here, but if we do a
2727 * vdev_reopen(), we need this value to be present to remember why we
2730 vd->vdev_label_aux = aux;
2733 * Faulted state takes precedence over degraded.
2735 vd->vdev_delayed_close = B_FALSE;
2736 vd->vdev_faulted = 1ULL;
2737 vd->vdev_degraded = 0ULL;
2738 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2741 * If this device has the only valid copy of the data, then
2742 * back off and simply mark the vdev as degraded instead.
2744 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2745 vd->vdev_degraded = 1ULL;
2746 vd->vdev_faulted = 0ULL;
2749 * If we reopen the device and it's not dead, only then do we
2754 if (vdev_readable(vd))
2755 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2758 return (spa_vdev_state_exit(spa, vd, 0));
2762 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2763 * user that something is wrong. The vdev continues to operate as normal as far
2764 * as I/O is concerned.
2767 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2771 spa_vdev_state_enter(spa, SCL_NONE);
2773 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2774 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2776 if (!vd->vdev_ops->vdev_op_leaf)
2777 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2780 * If the vdev is already faulted, then don't do anything.
2782 if (vd->vdev_faulted || vd->vdev_degraded)
2783 return (spa_vdev_state_exit(spa, NULL, 0));
2785 vd->vdev_degraded = 1ULL;
2786 if (!vdev_is_dead(vd))
2787 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2790 return (spa_vdev_state_exit(spa, vd, 0));
2794 * Online the given vdev.
2796 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2797 * spare device should be detached when the device finishes resilvering.
2798 * Second, the online should be treated like a 'test' online case, so no FMA
2799 * events are generated if the device fails to open.
2802 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2804 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2805 boolean_t wasoffline;
2806 vdev_state_t oldstate;
2808 spa_vdev_state_enter(spa, SCL_NONE);
2810 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2811 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2813 if (!vd->vdev_ops->vdev_op_leaf)
2814 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2816 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
2817 oldstate = vd->vdev_state;
2820 vd->vdev_offline = B_FALSE;
2821 vd->vdev_tmpoffline = B_FALSE;
2822 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2823 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2825 /* XXX - L2ARC 1.0 does not support expansion */
2826 if (!vd->vdev_aux) {
2827 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2828 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2832 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2834 if (!vd->vdev_aux) {
2835 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2836 pvd->vdev_expanding = B_FALSE;
2840 *newstate = vd->vdev_state;
2841 if ((flags & ZFS_ONLINE_UNSPARE) &&
2842 !vdev_is_dead(vd) && vd->vdev_parent &&
2843 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2844 vd->vdev_parent->vdev_child[0] == vd)
2845 vd->vdev_unspare = B_TRUE;
2847 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2849 /* XXX - L2ARC 1.0 does not support expansion */
2851 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2852 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2856 (oldstate < VDEV_STATE_DEGRADED &&
2857 vd->vdev_state >= VDEV_STATE_DEGRADED))
2858 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
2860 return (spa_vdev_state_exit(spa, vd, 0));
2864 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2868 uint64_t generation;
2869 metaslab_group_t *mg;
2872 spa_vdev_state_enter(spa, SCL_ALLOC);
2874 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2875 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2877 if (!vd->vdev_ops->vdev_op_leaf)
2878 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2882 generation = spa->spa_config_generation + 1;
2885 * If the device isn't already offline, try to offline it.
2887 if (!vd->vdev_offline) {
2889 * If this device has the only valid copy of some data,
2890 * don't allow it to be offlined. Log devices are always
2893 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2894 vdev_dtl_required(vd))
2895 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2898 * If the top-level is a slog and it has had allocations
2899 * then proceed. We check that the vdev's metaslab group
2900 * is not NULL since it's possible that we may have just
2901 * added this vdev but not yet initialized its metaslabs.
2903 if (tvd->vdev_islog && mg != NULL) {
2905 * Prevent any future allocations.
2907 metaslab_group_passivate(mg);
2908 (void) spa_vdev_state_exit(spa, vd, 0);
2910 error = spa_reset_logs(spa);
2912 spa_vdev_state_enter(spa, SCL_ALLOC);
2915 * Check to see if the config has changed.
2917 if (error || generation != spa->spa_config_generation) {
2918 metaslab_group_activate(mg);
2920 return (spa_vdev_state_exit(spa,
2922 (void) spa_vdev_state_exit(spa, vd, 0);
2925 ASSERT0(tvd->vdev_stat.vs_alloc);
2929 * Offline this device and reopen its top-level vdev.
2930 * If the top-level vdev is a log device then just offline
2931 * it. Otherwise, if this action results in the top-level
2932 * vdev becoming unusable, undo it and fail the request.
2934 vd->vdev_offline = B_TRUE;
2937 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2938 vdev_is_dead(tvd)) {
2939 vd->vdev_offline = B_FALSE;
2941 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2945 * Add the device back into the metaslab rotor so that
2946 * once we online the device it's open for business.
2948 if (tvd->vdev_islog && mg != NULL)
2949 metaslab_group_activate(mg);
2952 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2954 return (spa_vdev_state_exit(spa, vd, 0));
2958 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2962 mutex_enter(&spa->spa_vdev_top_lock);
2963 error = vdev_offline_locked(spa, guid, flags);
2964 mutex_exit(&spa->spa_vdev_top_lock);
2970 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2971 * vdev_offline(), we assume the spa config is locked. We also clear all
2972 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2975 vdev_clear(spa_t *spa, vdev_t *vd)
2977 vdev_t *rvd = spa->spa_root_vdev;
2979 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2984 vd->vdev_stat.vs_read_errors = 0;
2985 vd->vdev_stat.vs_write_errors = 0;
2986 vd->vdev_stat.vs_checksum_errors = 0;
2988 for (int c = 0; c < vd->vdev_children; c++)
2989 vdev_clear(spa, vd->vdev_child[c]);
2992 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2993 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2995 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2996 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
3000 * It makes no sense to "clear" an indirect vdev.
3002 if (!vdev_is_concrete(vd))
3006 * If we're in the FAULTED state or have experienced failed I/O, then
3007 * clear the persistent state and attempt to reopen the device. We
3008 * also mark the vdev config dirty, so that the new faulted state is
3009 * written out to disk.
3011 if (vd->vdev_faulted || vd->vdev_degraded ||
3012 !vdev_readable(vd) || !vdev_writeable(vd)) {
3015 * When reopening in reponse to a clear event, it may be due to
3016 * a fmadm repair request. In this case, if the device is
3017 * still broken, we want to still post the ereport again.
3019 vd->vdev_forcefault = B_TRUE;
3021 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3022 vd->vdev_cant_read = B_FALSE;
3023 vd->vdev_cant_write = B_FALSE;
3025 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3027 vd->vdev_forcefault = B_FALSE;
3029 if (vd != rvd && vdev_writeable(vd->vdev_top))
3030 vdev_state_dirty(vd->vdev_top);
3032 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
3033 spa_async_request(spa, SPA_ASYNC_RESILVER);
3035 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3039 * When clearing a FMA-diagnosed fault, we always want to
3040 * unspare the device, as we assume that the original spare was
3041 * done in response to the FMA fault.
3043 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3044 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3045 vd->vdev_parent->vdev_child[0] == vd)
3046 vd->vdev_unspare = B_TRUE;
3050 vdev_is_dead(vdev_t *vd)
3053 * Holes and missing devices are always considered "dead".
3054 * This simplifies the code since we don't have to check for
3055 * these types of devices in the various code paths.
3056 * Instead we rely on the fact that we skip over dead devices
3057 * before issuing I/O to them.
3059 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3060 vd->vdev_ops == &vdev_hole_ops ||
3061 vd->vdev_ops == &vdev_missing_ops);
3065 vdev_readable(vdev_t *vd)
3067 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3071 vdev_writeable(vdev_t *vd)
3073 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3074 vdev_is_concrete(vd));
3078 vdev_allocatable(vdev_t *vd)
3080 uint64_t state = vd->vdev_state;
3083 * We currently allow allocations from vdevs which may be in the
3084 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3085 * fails to reopen then we'll catch it later when we're holding
3086 * the proper locks. Note that we have to get the vdev state
3087 * in a local variable because although it changes atomically,
3088 * we're asking two separate questions about it.
3090 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3091 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3092 vd->vdev_mg->mg_initialized);
3096 vdev_accessible(vdev_t *vd, zio_t *zio)
3098 ASSERT(zio->io_vd == vd);
3100 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3103 if (zio->io_type == ZIO_TYPE_READ)
3104 return (!vd->vdev_cant_read);
3106 if (zio->io_type == ZIO_TYPE_WRITE)
3107 return (!vd->vdev_cant_write);
3113 * Get statistics for the given vdev.
3116 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3118 spa_t *spa = vd->vdev_spa;
3119 vdev_t *rvd = spa->spa_root_vdev;
3120 vdev_t *tvd = vd->vdev_top;
3122 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3124 mutex_enter(&vd->vdev_stat_lock);
3125 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3126 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3127 vs->vs_state = vd->vdev_state;
3128 vs->vs_rsize = vdev_get_min_asize(vd);
3129 if (vd->vdev_ops->vdev_op_leaf)
3130 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3132 * Report expandable space on top-level, non-auxillary devices only.
3133 * The expandable space is reported in terms of metaslab sized units
3134 * since that determines how much space the pool can expand.
3136 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
3137 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3138 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3140 vs->vs_configured_ashift = vd->vdev_top != NULL
3141 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
3142 vs->vs_logical_ashift = vd->vdev_logical_ashift;
3143 vs->vs_physical_ashift = vd->vdev_physical_ashift;
3144 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3145 vdev_is_concrete(vd)) {
3146 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
3150 * If we're getting stats on the root vdev, aggregate the I/O counts
3151 * over all top-level vdevs (i.e. the direct children of the root).
3154 for (int c = 0; c < rvd->vdev_children; c++) {
3155 vdev_t *cvd = rvd->vdev_child[c];
3156 vdev_stat_t *cvs = &cvd->vdev_stat;
3158 for (int t = 0; t < ZIO_TYPES; t++) {
3159 vs->vs_ops[t] += cvs->vs_ops[t];
3160 vs->vs_bytes[t] += cvs->vs_bytes[t];
3162 cvs->vs_scan_removing = cvd->vdev_removing;
3165 mutex_exit(&vd->vdev_stat_lock);
3169 vdev_clear_stats(vdev_t *vd)
3171 mutex_enter(&vd->vdev_stat_lock);
3172 vd->vdev_stat.vs_space = 0;
3173 vd->vdev_stat.vs_dspace = 0;
3174 vd->vdev_stat.vs_alloc = 0;
3175 mutex_exit(&vd->vdev_stat_lock);
3179 vdev_scan_stat_init(vdev_t *vd)
3181 vdev_stat_t *vs = &vd->vdev_stat;
3183 for (int c = 0; c < vd->vdev_children; c++)
3184 vdev_scan_stat_init(vd->vdev_child[c]);
3186 mutex_enter(&vd->vdev_stat_lock);
3187 vs->vs_scan_processed = 0;
3188 mutex_exit(&vd->vdev_stat_lock);
3192 vdev_stat_update(zio_t *zio, uint64_t psize)
3194 spa_t *spa = zio->io_spa;
3195 vdev_t *rvd = spa->spa_root_vdev;
3196 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3198 uint64_t txg = zio->io_txg;
3199 vdev_stat_t *vs = &vd->vdev_stat;
3200 zio_type_t type = zio->io_type;
3201 int flags = zio->io_flags;
3204 * If this i/o is a gang leader, it didn't do any actual work.
3206 if (zio->io_gang_tree)
3209 if (zio->io_error == 0) {
3211 * If this is a root i/o, don't count it -- we've already
3212 * counted the top-level vdevs, and vdev_get_stats() will
3213 * aggregate them when asked. This reduces contention on
3214 * the root vdev_stat_lock and implicitly handles blocks
3215 * that compress away to holes, for which there is no i/o.
3216 * (Holes never create vdev children, so all the counters
3217 * remain zero, which is what we want.)
3219 * Note: this only applies to successful i/o (io_error == 0)
3220 * because unlike i/o counts, errors are not additive.
3221 * When reading a ditto block, for example, failure of
3222 * one top-level vdev does not imply a root-level error.
3227 ASSERT(vd == zio->io_vd);
3229 if (flags & ZIO_FLAG_IO_BYPASS)
3232 mutex_enter(&vd->vdev_stat_lock);
3234 if (flags & ZIO_FLAG_IO_REPAIR) {
3235 if (flags & ZIO_FLAG_SCAN_THREAD) {
3236 dsl_scan_phys_t *scn_phys =
3237 &spa->spa_dsl_pool->dp_scan->scn_phys;
3238 uint64_t *processed = &scn_phys->scn_processed;
3241 if (vd->vdev_ops->vdev_op_leaf)
3242 atomic_add_64(processed, psize);
3243 vs->vs_scan_processed += psize;
3246 if (flags & ZIO_FLAG_SELF_HEAL)
3247 vs->vs_self_healed += psize;
3251 vs->vs_bytes[type] += psize;
3253 mutex_exit(&vd->vdev_stat_lock);
3257 if (flags & ZIO_FLAG_SPECULATIVE)
3261 * If this is an I/O error that is going to be retried, then ignore the
3262 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3263 * hard errors, when in reality they can happen for any number of
3264 * innocuous reasons (bus resets, MPxIO link failure, etc).
3266 if (zio->io_error == EIO &&
3267 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3271 * Intent logs writes won't propagate their error to the root
3272 * I/O so don't mark these types of failures as pool-level
3275 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3278 mutex_enter(&vd->vdev_stat_lock);
3279 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3280 if (zio->io_error == ECKSUM)
3281 vs->vs_checksum_errors++;
3283 vs->vs_read_errors++;
3285 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3286 vs->vs_write_errors++;
3287 mutex_exit(&vd->vdev_stat_lock);
3289 if (spa->spa_load_state == SPA_LOAD_NONE &&
3290 type == ZIO_TYPE_WRITE && txg != 0 &&
3291 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3292 (flags & ZIO_FLAG_SCAN_THREAD) ||
3293 spa->spa_claiming)) {
3295 * This is either a normal write (not a repair), or it's
3296 * a repair induced by the scrub thread, or it's a repair
3297 * made by zil_claim() during spa_load() in the first txg.
3298 * In the normal case, we commit the DTL change in the same
3299 * txg as the block was born. In the scrub-induced repair
3300 * case, we know that scrubs run in first-pass syncing context,
3301 * so we commit the DTL change in spa_syncing_txg(spa).
3302 * In the zil_claim() case, we commit in spa_first_txg(spa).
3304 * We currently do not make DTL entries for failed spontaneous
3305 * self-healing writes triggered by normal (non-scrubbing)
3306 * reads, because we have no transactional context in which to
3307 * do so -- and it's not clear that it'd be desirable anyway.
3309 if (vd->vdev_ops->vdev_op_leaf) {
3310 uint64_t commit_txg = txg;
3311 if (flags & ZIO_FLAG_SCAN_THREAD) {
3312 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3313 ASSERT(spa_sync_pass(spa) == 1);
3314 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3315 commit_txg = spa_syncing_txg(spa);
3316 } else if (spa->spa_claiming) {
3317 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3318 commit_txg = spa_first_txg(spa);
3320 ASSERT(commit_txg >= spa_syncing_txg(spa));
3321 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3323 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3324 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3325 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3328 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3333 * Update the in-core space usage stats for this vdev, its metaslab class,
3334 * and the root vdev.
3337 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3338 int64_t space_delta)
3340 int64_t dspace_delta = space_delta;
3341 spa_t *spa = vd->vdev_spa;
3342 vdev_t *rvd = spa->spa_root_vdev;
3343 metaslab_group_t *mg = vd->vdev_mg;
3344 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3346 ASSERT(vd == vd->vdev_top);
3349 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3350 * factor. We must calculate this here and not at the root vdev
3351 * because the root vdev's psize-to-asize is simply the max of its
3352 * childrens', thus not accurate enough for us.
3354 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3355 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3356 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3357 vd->vdev_deflate_ratio;
3359 mutex_enter(&vd->vdev_stat_lock);
3360 vd->vdev_stat.vs_alloc += alloc_delta;
3361 vd->vdev_stat.vs_space += space_delta;
3362 vd->vdev_stat.vs_dspace += dspace_delta;
3363 mutex_exit(&vd->vdev_stat_lock);
3365 if (mc == spa_normal_class(spa)) {
3366 mutex_enter(&rvd->vdev_stat_lock);
3367 rvd->vdev_stat.vs_alloc += alloc_delta;
3368 rvd->vdev_stat.vs_space += space_delta;
3369 rvd->vdev_stat.vs_dspace += dspace_delta;
3370 mutex_exit(&rvd->vdev_stat_lock);
3374 ASSERT(rvd == vd->vdev_parent);
3375 ASSERT(vd->vdev_ms_count != 0);
3377 metaslab_class_space_update(mc,
3378 alloc_delta, defer_delta, space_delta, dspace_delta);
3383 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3384 * so that it will be written out next time the vdev configuration is synced.
3385 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3388 vdev_config_dirty(vdev_t *vd)
3390 spa_t *spa = vd->vdev_spa;
3391 vdev_t *rvd = spa->spa_root_vdev;
3394 ASSERT(spa_writeable(spa));
3397 * If this is an aux vdev (as with l2cache and spare devices), then we
3398 * update the vdev config manually and set the sync flag.
3400 if (vd->vdev_aux != NULL) {
3401 spa_aux_vdev_t *sav = vd->vdev_aux;
3405 for (c = 0; c < sav->sav_count; c++) {
3406 if (sav->sav_vdevs[c] == vd)
3410 if (c == sav->sav_count) {
3412 * We're being removed. There's nothing more to do.
3414 ASSERT(sav->sav_sync == B_TRUE);
3418 sav->sav_sync = B_TRUE;
3420 if (nvlist_lookup_nvlist_array(sav->sav_config,
3421 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3422 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3423 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3429 * Setting the nvlist in the middle if the array is a little
3430 * sketchy, but it will work.
3432 nvlist_free(aux[c]);
3433 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3439 * The dirty list is protected by the SCL_CONFIG lock. The caller
3440 * must either hold SCL_CONFIG as writer, or must be the sync thread
3441 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3442 * so this is sufficient to ensure mutual exclusion.
3444 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3445 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3446 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3449 for (c = 0; c < rvd->vdev_children; c++)
3450 vdev_config_dirty(rvd->vdev_child[c]);
3452 ASSERT(vd == vd->vdev_top);
3454 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3455 vdev_is_concrete(vd)) {
3456 list_insert_head(&spa->spa_config_dirty_list, vd);
3462 vdev_config_clean(vdev_t *vd)
3464 spa_t *spa = vd->vdev_spa;
3466 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3467 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3468 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3470 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3471 list_remove(&spa->spa_config_dirty_list, vd);
3475 * Mark a top-level vdev's state as dirty, so that the next pass of
3476 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3477 * the state changes from larger config changes because they require
3478 * much less locking, and are often needed for administrative actions.
3481 vdev_state_dirty(vdev_t *vd)
3483 spa_t *spa = vd->vdev_spa;
3485 ASSERT(spa_writeable(spa));
3486 ASSERT(vd == vd->vdev_top);
3489 * The state list is protected by the SCL_STATE lock. The caller
3490 * must either hold SCL_STATE as writer, or must be the sync thread
3491 * (which holds SCL_STATE as reader). There's only one sync thread,
3492 * so this is sufficient to ensure mutual exclusion.
3494 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3495 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3496 spa_config_held(spa, SCL_STATE, RW_READER)));
3498 if (!list_link_active(&vd->vdev_state_dirty_node) &&
3499 vdev_is_concrete(vd))
3500 list_insert_head(&spa->spa_state_dirty_list, vd);
3504 vdev_state_clean(vdev_t *vd)
3506 spa_t *spa = vd->vdev_spa;
3508 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3509 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3510 spa_config_held(spa, SCL_STATE, RW_READER)));
3512 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3513 list_remove(&spa->spa_state_dirty_list, vd);
3517 * Propagate vdev state up from children to parent.
3520 vdev_propagate_state(vdev_t *vd)
3522 spa_t *spa = vd->vdev_spa;
3523 vdev_t *rvd = spa->spa_root_vdev;
3524 int degraded = 0, faulted = 0;
3528 if (vd->vdev_children > 0) {
3529 for (int c = 0; c < vd->vdev_children; c++) {
3530 child = vd->vdev_child[c];
3533 * Don't factor holes or indirect vdevs into the
3536 if (!vdev_is_concrete(child))
3539 if (!vdev_readable(child) ||
3540 (!vdev_writeable(child) && spa_writeable(spa))) {
3542 * Root special: if there is a top-level log
3543 * device, treat the root vdev as if it were
3546 if (child->vdev_islog && vd == rvd)
3550 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3554 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3558 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3561 * Root special: if there is a top-level vdev that cannot be
3562 * opened due to corrupted metadata, then propagate the root
3563 * vdev's aux state as 'corrupt' rather than 'insufficient
3566 if (corrupted && vd == rvd &&
3567 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3568 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3569 VDEV_AUX_CORRUPT_DATA);
3572 if (vd->vdev_parent)
3573 vdev_propagate_state(vd->vdev_parent);
3577 * Set a vdev's state. If this is during an open, we don't update the parent
3578 * state, because we're in the process of opening children depth-first.
3579 * Otherwise, we propagate the change to the parent.
3581 * If this routine places a device in a faulted state, an appropriate ereport is
3585 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3587 uint64_t save_state;
3588 spa_t *spa = vd->vdev_spa;
3590 if (state == vd->vdev_state) {
3591 vd->vdev_stat.vs_aux = aux;
3595 save_state = vd->vdev_state;
3597 vd->vdev_state = state;
3598 vd->vdev_stat.vs_aux = aux;
3601 * If we are setting the vdev state to anything but an open state, then
3602 * always close the underlying device unless the device has requested
3603 * a delayed close (i.e. we're about to remove or fault the device).
3604 * Otherwise, we keep accessible but invalid devices open forever.
3605 * We don't call vdev_close() itself, because that implies some extra
3606 * checks (offline, etc) that we don't want here. This is limited to
3607 * leaf devices, because otherwise closing the device will affect other
3610 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3611 vd->vdev_ops->vdev_op_leaf)
3612 vd->vdev_ops->vdev_op_close(vd);
3614 if (vd->vdev_removed &&
3615 state == VDEV_STATE_CANT_OPEN &&
3616 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3618 * If the previous state is set to VDEV_STATE_REMOVED, then this
3619 * device was previously marked removed and someone attempted to
3620 * reopen it. If this failed due to a nonexistent device, then
3621 * keep the device in the REMOVED state. We also let this be if
3622 * it is one of our special test online cases, which is only
3623 * attempting to online the device and shouldn't generate an FMA
3626 vd->vdev_state = VDEV_STATE_REMOVED;
3627 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3628 } else if (state == VDEV_STATE_REMOVED) {
3629 vd->vdev_removed = B_TRUE;
3630 } else if (state == VDEV_STATE_CANT_OPEN) {
3632 * If we fail to open a vdev during an import or recovery, we
3633 * mark it as "not available", which signifies that it was
3634 * never there to begin with. Failure to open such a device
3635 * is not considered an error.
3637 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3638 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3639 vd->vdev_ops->vdev_op_leaf)
3640 vd->vdev_not_present = 1;
3643 * Post the appropriate ereport. If the 'prevstate' field is
3644 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3645 * that this is part of a vdev_reopen(). In this case, we don't
3646 * want to post the ereport if the device was already in the
3647 * CANT_OPEN state beforehand.
3649 * If the 'checkremove' flag is set, then this is an attempt to
3650 * online the device in response to an insertion event. If we
3651 * hit this case, then we have detected an insertion event for a
3652 * faulted or offline device that wasn't in the removed state.
3653 * In this scenario, we don't post an ereport because we are
3654 * about to replace the device, or attempt an online with
3655 * vdev_forcefault, which will generate the fault for us.
3657 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3658 !vd->vdev_not_present && !vd->vdev_checkremove &&
3659 vd != spa->spa_root_vdev) {
3663 case VDEV_AUX_OPEN_FAILED:
3664 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3666 case VDEV_AUX_CORRUPT_DATA:
3667 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3669 case VDEV_AUX_NO_REPLICAS:
3670 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3672 case VDEV_AUX_BAD_GUID_SUM:
3673 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3675 case VDEV_AUX_TOO_SMALL:
3676 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3678 case VDEV_AUX_BAD_LABEL:
3679 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3682 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3685 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3688 /* Erase any notion of persistent removed state */
3689 vd->vdev_removed = B_FALSE;
3691 vd->vdev_removed = B_FALSE;
3695 * Notify the fmd of the state change. Be verbose and post
3696 * notifications even for stuff that's not important; the fmd agent can
3697 * sort it out. Don't emit state change events for non-leaf vdevs since
3698 * they can't change state on their own. The FMD can check their state
3699 * if it wants to when it sees that a leaf vdev had a state change.
3701 if (vd->vdev_ops->vdev_op_leaf)
3702 zfs_post_state_change(spa, vd);
3704 if (!isopen && vd->vdev_parent)
3705 vdev_propagate_state(vd->vdev_parent);
3709 * Check the vdev configuration to ensure that it's capable of supporting
3710 * a root pool. We do not support partial configuration.
3711 * In addition, only a single top-level vdev is allowed.
3713 * FreeBSD does not have above limitations.
3716 vdev_is_bootable(vdev_t *vd)
3719 if (!vd->vdev_ops->vdev_op_leaf) {
3720 char *vdev_type = vd->vdev_ops->vdev_op_type;
3722 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3723 vd->vdev_children > 1) {
3725 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
3726 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
3731 for (int c = 0; c < vd->vdev_children; c++) {
3732 if (!vdev_is_bootable(vd->vdev_child[c]))
3735 #endif /* illumos */
3740 vdev_is_concrete(vdev_t *vd)
3742 vdev_ops_t *ops = vd->vdev_ops;
3743 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
3744 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
3752 * Load the state from the original vdev tree (ovd) which
3753 * we've retrieved from the MOS config object. If the original
3754 * vdev was offline or faulted then we transfer that state to the
3755 * device in the current vdev tree (nvd).
3758 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3760 spa_t *spa = nvd->vdev_spa;
3762 ASSERT(nvd->vdev_top->vdev_islog);
3763 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3764 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3766 for (int c = 0; c < nvd->vdev_children; c++)
3767 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3769 if (nvd->vdev_ops->vdev_op_leaf) {
3771 * Restore the persistent vdev state
3773 nvd->vdev_offline = ovd->vdev_offline;
3774 nvd->vdev_faulted = ovd->vdev_faulted;
3775 nvd->vdev_degraded = ovd->vdev_degraded;
3776 nvd->vdev_removed = ovd->vdev_removed;
3781 * Determine if a log device has valid content. If the vdev was
3782 * removed or faulted in the MOS config then we know that
3783 * the content on the log device has already been written to the pool.
3786 vdev_log_state_valid(vdev_t *vd)
3788 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3792 for (int c = 0; c < vd->vdev_children; c++)
3793 if (vdev_log_state_valid(vd->vdev_child[c]))
3800 * Expand a vdev if possible.
3803 vdev_expand(vdev_t *vd, uint64_t txg)
3805 ASSERT(vd->vdev_top == vd);
3806 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3808 vdev_set_deflate_ratio(vd);
3810 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
3811 vdev_is_concrete(vd)) {
3812 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3813 vdev_config_dirty(vd);
3821 vdev_split(vdev_t *vd)
3823 vdev_t *cvd, *pvd = vd->vdev_parent;
3825 vdev_remove_child(pvd, vd);
3826 vdev_compact_children(pvd);
3828 cvd = pvd->vdev_child[0];
3829 if (pvd->vdev_children == 1) {
3830 vdev_remove_parent(cvd);
3831 cvd->vdev_splitting = B_TRUE;
3833 vdev_propagate_state(cvd);
3837 vdev_deadman(vdev_t *vd)
3839 for (int c = 0; c < vd->vdev_children; c++) {
3840 vdev_t *cvd = vd->vdev_child[c];
3845 if (vd->vdev_ops->vdev_op_leaf) {
3846 vdev_queue_t *vq = &vd->vdev_queue;
3848 mutex_enter(&vq->vq_lock);
3849 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3850 spa_t *spa = vd->vdev_spa;
3855 * Look at the head of all the pending queues,
3856 * if any I/O has been outstanding for longer than
3857 * the spa_deadman_synctime we panic the system.
3859 fio = avl_first(&vq->vq_active_tree);
3860 delta = gethrtime() - fio->io_timestamp;
3861 if (delta > spa_deadman_synctime(spa)) {
3862 vdev_dbgmsg(vd, "SLOW IO: zio timestamp "
3863 "%lluns, delta %lluns, last io %lluns",
3864 fio->io_timestamp, (u_longlong_t)delta,
3865 vq->vq_io_complete_ts);
3866 fm_panic("I/O to pool '%s' appears to be "
3867 "hung on vdev guid %llu at '%s'.",
3869 (long long unsigned int) vd->vdev_guid,
3873 mutex_exit(&vq->vq_lock);