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, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/fm/fs/zfs.h>
32 #include <sys/spa_impl.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/vdev_impl.h>
36 #include <sys/uberblock_impl.h>
37 #include <sys/metaslab.h>
38 #include <sys/metaslab_impl.h>
39 #include <sys/space_map.h>
40 #include <sys/space_reftree.h>
43 #include <sys/fs/zfs.h>
46 #include <sys/dsl_scan.h>
47 #include <sys/trim_map.h>
49 SYSCTL_DECL(_vfs_zfs);
50 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
53 * Virtual device management.
57 * The limit for ZFS to automatically increase a top-level vdev's ashift
58 * from logical ashift to physical ashift.
60 * Example: one or more 512B emulation child vdevs
61 * child->vdev_ashift = 9 (512 bytes)
62 * child->vdev_physical_ashift = 12 (4096 bytes)
63 * zfs_max_auto_ashift = 11 (2048 bytes)
64 * zfs_min_auto_ashift = 9 (512 bytes)
66 * On pool creation or the addition of a new top-level vdev, ZFS will
67 * increase the ashift of the top-level vdev to 2048 as limited by
68 * zfs_max_auto_ashift.
70 * Example: one or more 512B emulation child vdevs
71 * child->vdev_ashift = 9 (512 bytes)
72 * child->vdev_physical_ashift = 12 (4096 bytes)
73 * zfs_max_auto_ashift = 13 (8192 bytes)
74 * zfs_min_auto_ashift = 9 (512 bytes)
76 * On pool creation or the addition of a new top-level vdev, ZFS will
77 * increase the ashift of the top-level vdev to 4096 to match the
78 * max vdev_physical_ashift.
80 * Example: one or more 512B emulation child vdevs
81 * child->vdev_ashift = 9 (512 bytes)
82 * child->vdev_physical_ashift = 9 (512 bytes)
83 * zfs_max_auto_ashift = 13 (8192 bytes)
84 * zfs_min_auto_ashift = 12 (4096 bytes)
86 * On pool creation or the addition of a new top-level vdev, ZFS will
87 * increase the ashift of the top-level vdev to 4096 to match the
88 * zfs_min_auto_ashift.
90 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
91 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
94 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
99 val = zfs_max_auto_ashift;
100 err = sysctl_handle_64(oidp, &val, 0, req);
101 if (err != 0 || req->newptr == NULL)
104 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
107 zfs_max_auto_ashift = val;
111 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
112 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
113 sysctl_vfs_zfs_max_auto_ashift, "QU",
114 "Max ashift used when optimising for logical -> physical sectors size on "
115 "new top-level vdevs.");
118 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
123 val = zfs_min_auto_ashift;
124 err = sysctl_handle_64(oidp, &val, 0, req);
125 if (err != 0 || req->newptr == NULL)
128 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
131 zfs_min_auto_ashift = val;
135 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
136 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
137 sysctl_vfs_zfs_min_auto_ashift, "QU",
138 "Min ashift used when creating new top-level vdevs.");
140 static vdev_ops_t *vdev_ops_table[] = {
159 * When a vdev is added, it will be divided into approximately (but no
160 * more than) this number of metaslabs.
162 int metaslabs_per_vdev = 200;
163 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
164 &metaslabs_per_vdev, 0,
165 "When a vdev is added, how many metaslabs the vdev should be divided into");
168 * Given a vdev type, return the appropriate ops vector.
171 vdev_getops(const char *type)
173 vdev_ops_t *ops, **opspp;
175 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
176 if (strcmp(ops->vdev_op_type, type) == 0)
183 * Default asize function: return the MAX of psize with the asize of
184 * all children. This is what's used by anything other than RAID-Z.
187 vdev_default_asize(vdev_t *vd, uint64_t psize)
189 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
192 for (int c = 0; c < vd->vdev_children; c++) {
193 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
194 asize = MAX(asize, csize);
201 * Get the minimum allocatable size. We define the allocatable size as
202 * the vdev's asize rounded to the nearest metaslab. This allows us to
203 * replace or attach devices which don't have the same physical size but
204 * can still satisfy the same number of allocations.
207 vdev_get_min_asize(vdev_t *vd)
209 vdev_t *pvd = vd->vdev_parent;
212 * If our parent is NULL (inactive spare or cache) or is the root,
213 * just return our own asize.
216 return (vd->vdev_asize);
219 * The top-level vdev just returns the allocatable size rounded
220 * to the nearest metaslab.
222 if (vd == vd->vdev_top)
223 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
226 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
227 * so each child must provide at least 1/Nth of its asize.
229 if (pvd->vdev_ops == &vdev_raidz_ops)
230 return (pvd->vdev_min_asize / pvd->vdev_children);
232 return (pvd->vdev_min_asize);
236 vdev_set_min_asize(vdev_t *vd)
238 vd->vdev_min_asize = vdev_get_min_asize(vd);
240 for (int c = 0; c < vd->vdev_children; c++)
241 vdev_set_min_asize(vd->vdev_child[c]);
245 vdev_lookup_top(spa_t *spa, uint64_t vdev)
247 vdev_t *rvd = spa->spa_root_vdev;
249 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
251 if (vdev < rvd->vdev_children) {
252 ASSERT(rvd->vdev_child[vdev] != NULL);
253 return (rvd->vdev_child[vdev]);
260 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
264 if (vd->vdev_guid == guid)
267 for (int c = 0; c < vd->vdev_children; c++)
268 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
276 vdev_count_leaves_impl(vdev_t *vd)
280 if (vd->vdev_ops->vdev_op_leaf)
283 for (int c = 0; c < vd->vdev_children; c++)
284 n += vdev_count_leaves_impl(vd->vdev_child[c]);
290 vdev_count_leaves(spa_t *spa)
292 return (vdev_count_leaves_impl(spa->spa_root_vdev));
296 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
298 size_t oldsize, newsize;
299 uint64_t id = cvd->vdev_id;
301 spa_t *spa = cvd->vdev_spa;
303 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
304 ASSERT(cvd->vdev_parent == NULL);
306 cvd->vdev_parent = pvd;
311 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
313 oldsize = pvd->vdev_children * sizeof (vdev_t *);
314 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
315 newsize = pvd->vdev_children * sizeof (vdev_t *);
317 newchild = kmem_zalloc(newsize, KM_SLEEP);
318 if (pvd->vdev_child != NULL) {
319 bcopy(pvd->vdev_child, newchild, oldsize);
320 kmem_free(pvd->vdev_child, oldsize);
323 pvd->vdev_child = newchild;
324 pvd->vdev_child[id] = cvd;
326 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
327 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
330 * Walk up all ancestors to update guid sum.
332 for (; pvd != NULL; pvd = pvd->vdev_parent)
333 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
337 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
340 uint_t id = cvd->vdev_id;
342 ASSERT(cvd->vdev_parent == pvd);
347 ASSERT(id < pvd->vdev_children);
348 ASSERT(pvd->vdev_child[id] == cvd);
350 pvd->vdev_child[id] = NULL;
351 cvd->vdev_parent = NULL;
353 for (c = 0; c < pvd->vdev_children; c++)
354 if (pvd->vdev_child[c])
357 if (c == pvd->vdev_children) {
358 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
359 pvd->vdev_child = NULL;
360 pvd->vdev_children = 0;
364 * Walk up all ancestors to update guid sum.
366 for (; pvd != NULL; pvd = pvd->vdev_parent)
367 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
371 * Remove any holes in the child array.
374 vdev_compact_children(vdev_t *pvd)
376 vdev_t **newchild, *cvd;
377 int oldc = pvd->vdev_children;
380 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
382 for (int c = newc = 0; c < oldc; c++)
383 if (pvd->vdev_child[c])
386 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
388 for (int c = newc = 0; c < oldc; c++) {
389 if ((cvd = pvd->vdev_child[c]) != NULL) {
390 newchild[newc] = cvd;
391 cvd->vdev_id = newc++;
395 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
396 pvd->vdev_child = newchild;
397 pvd->vdev_children = newc;
401 * Allocate and minimally initialize a vdev_t.
404 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
408 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
410 if (spa->spa_root_vdev == NULL) {
411 ASSERT(ops == &vdev_root_ops);
412 spa->spa_root_vdev = vd;
413 spa->spa_load_guid = spa_generate_guid(NULL);
416 if (guid == 0 && ops != &vdev_hole_ops) {
417 if (spa->spa_root_vdev == vd) {
419 * The root vdev's guid will also be the pool guid,
420 * which must be unique among all pools.
422 guid = spa_generate_guid(NULL);
425 * Any other vdev's guid must be unique within the pool.
427 guid = spa_generate_guid(spa);
429 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
434 vd->vdev_guid = guid;
435 vd->vdev_guid_sum = guid;
437 vd->vdev_state = VDEV_STATE_CLOSED;
438 vd->vdev_ishole = (ops == &vdev_hole_ops);
440 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
441 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
442 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
443 for (int t = 0; t < DTL_TYPES; t++) {
444 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
447 txg_list_create(&vd->vdev_ms_list,
448 offsetof(struct metaslab, ms_txg_node));
449 txg_list_create(&vd->vdev_dtl_list,
450 offsetof(struct vdev, vdev_dtl_node));
451 vd->vdev_stat.vs_timestamp = gethrtime();
459 * Allocate a new vdev. The 'alloctype' is used to control whether we are
460 * creating a new vdev or loading an existing one - the behavior is slightly
461 * different for each case.
464 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
469 uint64_t guid = 0, islog, nparity;
472 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
474 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
475 return (SET_ERROR(EINVAL));
477 if ((ops = vdev_getops(type)) == NULL)
478 return (SET_ERROR(EINVAL));
481 * If this is a load, get the vdev guid from the nvlist.
482 * Otherwise, vdev_alloc_common() will generate one for us.
484 if (alloctype == VDEV_ALLOC_LOAD) {
487 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
489 return (SET_ERROR(EINVAL));
491 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
492 return (SET_ERROR(EINVAL));
493 } else if (alloctype == VDEV_ALLOC_SPARE) {
494 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
495 return (SET_ERROR(EINVAL));
496 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
497 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
498 return (SET_ERROR(EINVAL));
499 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
500 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
501 return (SET_ERROR(EINVAL));
505 * The first allocated vdev must be of type 'root'.
507 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
508 return (SET_ERROR(EINVAL));
511 * Determine whether we're a log vdev.
514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
515 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
516 return (SET_ERROR(ENOTSUP));
518 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
519 return (SET_ERROR(ENOTSUP));
522 * Set the nparity property for RAID-Z vdevs.
525 if (ops == &vdev_raidz_ops) {
526 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
528 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
529 return (SET_ERROR(EINVAL));
531 * Previous versions could only support 1 or 2 parity
535 spa_version(spa) < SPA_VERSION_RAIDZ2)
536 return (SET_ERROR(ENOTSUP));
538 spa_version(spa) < SPA_VERSION_RAIDZ3)
539 return (SET_ERROR(ENOTSUP));
542 * We require the parity to be specified for SPAs that
543 * support multiple parity levels.
545 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
546 return (SET_ERROR(EINVAL));
548 * Otherwise, we default to 1 parity device for RAID-Z.
555 ASSERT(nparity != -1ULL);
557 vd = vdev_alloc_common(spa, id, guid, ops);
559 vd->vdev_islog = islog;
560 vd->vdev_nparity = nparity;
562 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
563 vd->vdev_path = spa_strdup(vd->vdev_path);
564 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
565 vd->vdev_devid = spa_strdup(vd->vdev_devid);
566 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
567 &vd->vdev_physpath) == 0)
568 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
569 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
570 vd->vdev_fru = spa_strdup(vd->vdev_fru);
573 * Set the whole_disk property. If it's not specified, leave the value
576 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
577 &vd->vdev_wholedisk) != 0)
578 vd->vdev_wholedisk = -1ULL;
581 * Look for the 'not present' flag. This will only be set if the device
582 * was not present at the time of import.
584 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
585 &vd->vdev_not_present);
588 * Get the alignment requirement.
590 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
593 * Retrieve the vdev creation time.
595 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
599 * If we're a top-level vdev, try to load the allocation parameters.
601 if (parent && !parent->vdev_parent &&
602 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
603 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
605 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
607 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
609 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
613 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
614 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
615 alloctype == VDEV_ALLOC_ADD ||
616 alloctype == VDEV_ALLOC_SPLIT ||
617 alloctype == VDEV_ALLOC_ROOTPOOL);
618 vd->vdev_mg = metaslab_group_create(islog ?
619 spa_log_class(spa) : spa_normal_class(spa), vd);
623 * If we're a leaf vdev, try to load the DTL object and other state.
625 if (vd->vdev_ops->vdev_op_leaf &&
626 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
627 alloctype == VDEV_ALLOC_ROOTPOOL)) {
628 if (alloctype == VDEV_ALLOC_LOAD) {
629 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
630 &vd->vdev_dtl_object);
631 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
635 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
638 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
639 &spare) == 0 && spare)
643 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
646 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
647 &vd->vdev_resilver_txg);
650 * When importing a pool, we want to ignore the persistent fault
651 * state, as the diagnosis made on another system may not be
652 * valid in the current context. Local vdevs will
653 * remain in the faulted state.
655 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
656 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
658 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
660 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
663 if (vd->vdev_faulted || vd->vdev_degraded) {
667 VDEV_AUX_ERR_EXCEEDED;
668 if (nvlist_lookup_string(nv,
669 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
670 strcmp(aux, "external") == 0)
671 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
677 * Add ourselves to the parent's list of children.
679 vdev_add_child(parent, vd);
687 vdev_free(vdev_t *vd)
689 spa_t *spa = vd->vdev_spa;
692 * vdev_free() implies closing the vdev first. This is simpler than
693 * trying to ensure complicated semantics for all callers.
697 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
698 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
703 for (int c = 0; c < vd->vdev_children; c++)
704 vdev_free(vd->vdev_child[c]);
706 ASSERT(vd->vdev_child == NULL);
707 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
710 * Discard allocation state.
712 if (vd->vdev_mg != NULL) {
713 vdev_metaslab_fini(vd);
714 metaslab_group_destroy(vd->vdev_mg);
717 ASSERT0(vd->vdev_stat.vs_space);
718 ASSERT0(vd->vdev_stat.vs_dspace);
719 ASSERT0(vd->vdev_stat.vs_alloc);
722 * Remove this vdev from its parent's child list.
724 vdev_remove_child(vd->vdev_parent, vd);
726 ASSERT(vd->vdev_parent == NULL);
729 * Clean up vdev structure.
735 spa_strfree(vd->vdev_path);
737 spa_strfree(vd->vdev_devid);
738 if (vd->vdev_physpath)
739 spa_strfree(vd->vdev_physpath);
741 spa_strfree(vd->vdev_fru);
743 if (vd->vdev_isspare)
744 spa_spare_remove(vd);
745 if (vd->vdev_isl2cache)
746 spa_l2cache_remove(vd);
748 txg_list_destroy(&vd->vdev_ms_list);
749 txg_list_destroy(&vd->vdev_dtl_list);
751 mutex_enter(&vd->vdev_dtl_lock);
752 space_map_close(vd->vdev_dtl_sm);
753 for (int t = 0; t < DTL_TYPES; t++) {
754 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
755 range_tree_destroy(vd->vdev_dtl[t]);
757 mutex_exit(&vd->vdev_dtl_lock);
759 mutex_destroy(&vd->vdev_dtl_lock);
760 mutex_destroy(&vd->vdev_stat_lock);
761 mutex_destroy(&vd->vdev_probe_lock);
763 if (vd == spa->spa_root_vdev)
764 spa->spa_root_vdev = NULL;
766 kmem_free(vd, sizeof (vdev_t));
770 * Transfer top-level vdev state from svd to tvd.
773 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
775 spa_t *spa = svd->vdev_spa;
780 ASSERT(tvd == tvd->vdev_top);
782 tvd->vdev_ms_array = svd->vdev_ms_array;
783 tvd->vdev_ms_shift = svd->vdev_ms_shift;
784 tvd->vdev_ms_count = svd->vdev_ms_count;
786 svd->vdev_ms_array = 0;
787 svd->vdev_ms_shift = 0;
788 svd->vdev_ms_count = 0;
791 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
792 tvd->vdev_mg = svd->vdev_mg;
793 tvd->vdev_ms = svd->vdev_ms;
798 if (tvd->vdev_mg != NULL)
799 tvd->vdev_mg->mg_vd = tvd;
801 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
802 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
803 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
805 svd->vdev_stat.vs_alloc = 0;
806 svd->vdev_stat.vs_space = 0;
807 svd->vdev_stat.vs_dspace = 0;
809 for (t = 0; t < TXG_SIZE; t++) {
810 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
811 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
812 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
813 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
814 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
815 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
818 if (list_link_active(&svd->vdev_config_dirty_node)) {
819 vdev_config_clean(svd);
820 vdev_config_dirty(tvd);
823 if (list_link_active(&svd->vdev_state_dirty_node)) {
824 vdev_state_clean(svd);
825 vdev_state_dirty(tvd);
828 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
829 svd->vdev_deflate_ratio = 0;
831 tvd->vdev_islog = svd->vdev_islog;
836 vdev_top_update(vdev_t *tvd, vdev_t *vd)
843 for (int c = 0; c < vd->vdev_children; c++)
844 vdev_top_update(tvd, vd->vdev_child[c]);
848 * Add a mirror/replacing vdev above an existing vdev.
851 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
853 spa_t *spa = cvd->vdev_spa;
854 vdev_t *pvd = cvd->vdev_parent;
857 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
859 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
861 mvd->vdev_asize = cvd->vdev_asize;
862 mvd->vdev_min_asize = cvd->vdev_min_asize;
863 mvd->vdev_max_asize = cvd->vdev_max_asize;
864 mvd->vdev_ashift = cvd->vdev_ashift;
865 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
866 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
867 mvd->vdev_state = cvd->vdev_state;
868 mvd->vdev_crtxg = cvd->vdev_crtxg;
870 vdev_remove_child(pvd, cvd);
871 vdev_add_child(pvd, mvd);
872 cvd->vdev_id = mvd->vdev_children;
873 vdev_add_child(mvd, cvd);
874 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
876 if (mvd == mvd->vdev_top)
877 vdev_top_transfer(cvd, mvd);
883 * Remove a 1-way mirror/replacing vdev from the tree.
886 vdev_remove_parent(vdev_t *cvd)
888 vdev_t *mvd = cvd->vdev_parent;
889 vdev_t *pvd = mvd->vdev_parent;
891 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
893 ASSERT(mvd->vdev_children == 1);
894 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
895 mvd->vdev_ops == &vdev_replacing_ops ||
896 mvd->vdev_ops == &vdev_spare_ops);
897 cvd->vdev_ashift = mvd->vdev_ashift;
898 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
899 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
901 vdev_remove_child(mvd, cvd);
902 vdev_remove_child(pvd, mvd);
905 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
906 * Otherwise, we could have detached an offline device, and when we
907 * go to import the pool we'll think we have two top-level vdevs,
908 * instead of a different version of the same top-level vdev.
910 if (mvd->vdev_top == mvd) {
911 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
912 cvd->vdev_orig_guid = cvd->vdev_guid;
913 cvd->vdev_guid += guid_delta;
914 cvd->vdev_guid_sum += guid_delta;
916 cvd->vdev_id = mvd->vdev_id;
917 vdev_add_child(pvd, cvd);
918 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
920 if (cvd == cvd->vdev_top)
921 vdev_top_transfer(mvd, cvd);
923 ASSERT(mvd->vdev_children == 0);
928 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
930 spa_t *spa = vd->vdev_spa;
931 objset_t *mos = spa->spa_meta_objset;
933 uint64_t oldc = vd->vdev_ms_count;
934 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
938 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
941 * This vdev is not being allocated from yet or is a hole.
943 if (vd->vdev_ms_shift == 0)
946 ASSERT(!vd->vdev_ishole);
949 * Compute the raidz-deflation ratio. Note, we hard-code
950 * in 128k (1 << 17) because it is the "typical" blocksize.
951 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
952 * otherwise it would inconsistently account for existing bp's.
954 vd->vdev_deflate_ratio = (1 << 17) /
955 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
957 ASSERT(oldc <= newc);
959 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
962 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
963 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
967 vd->vdev_ms_count = newc;
969 for (m = oldc; m < newc; m++) {
973 error = dmu_read(mos, vd->vdev_ms_array,
974 m * sizeof (uint64_t), sizeof (uint64_t), &object,
980 error = metaslab_init(vd->vdev_mg, m, object, txg,
987 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
990 * If the vdev is being removed we don't activate
991 * the metaslabs since we want to ensure that no new
992 * allocations are performed on this device.
994 if (oldc == 0 && !vd->vdev_removing)
995 metaslab_group_activate(vd->vdev_mg);
998 spa_config_exit(spa, SCL_ALLOC, FTAG);
1004 vdev_metaslab_fini(vdev_t *vd)
1007 uint64_t count = vd->vdev_ms_count;
1009 if (vd->vdev_ms != NULL) {
1010 metaslab_group_passivate(vd->vdev_mg);
1011 for (m = 0; m < count; m++) {
1012 metaslab_t *msp = vd->vdev_ms[m];
1017 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1022 typedef struct vdev_probe_stats {
1023 boolean_t vps_readable;
1024 boolean_t vps_writeable;
1026 } vdev_probe_stats_t;
1029 vdev_probe_done(zio_t *zio)
1031 spa_t *spa = zio->io_spa;
1032 vdev_t *vd = zio->io_vd;
1033 vdev_probe_stats_t *vps = zio->io_private;
1035 ASSERT(vd->vdev_probe_zio != NULL);
1037 if (zio->io_type == ZIO_TYPE_READ) {
1038 if (zio->io_error == 0)
1039 vps->vps_readable = 1;
1040 if (zio->io_error == 0 && spa_writeable(spa)) {
1041 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1042 zio->io_offset, zio->io_size, zio->io_data,
1043 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1044 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1046 zio_buf_free(zio->io_data, zio->io_size);
1048 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1049 if (zio->io_error == 0)
1050 vps->vps_writeable = 1;
1051 zio_buf_free(zio->io_data, zio->io_size);
1052 } else if (zio->io_type == ZIO_TYPE_NULL) {
1055 vd->vdev_cant_read |= !vps->vps_readable;
1056 vd->vdev_cant_write |= !vps->vps_writeable;
1058 if (vdev_readable(vd) &&
1059 (vdev_writeable(vd) || !spa_writeable(spa))) {
1062 ASSERT(zio->io_error != 0);
1063 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1064 spa, vd, NULL, 0, 0);
1065 zio->io_error = SET_ERROR(ENXIO);
1068 mutex_enter(&vd->vdev_probe_lock);
1069 ASSERT(vd->vdev_probe_zio == zio);
1070 vd->vdev_probe_zio = NULL;
1071 mutex_exit(&vd->vdev_probe_lock);
1073 while ((pio = zio_walk_parents(zio)) != NULL)
1074 if (!vdev_accessible(vd, pio))
1075 pio->io_error = SET_ERROR(ENXIO);
1077 kmem_free(vps, sizeof (*vps));
1082 * Determine whether this device is accessible.
1084 * Read and write to several known locations: the pad regions of each
1085 * vdev label but the first, which we leave alone in case it contains
1089 vdev_probe(vdev_t *vd, zio_t *zio)
1091 spa_t *spa = vd->vdev_spa;
1092 vdev_probe_stats_t *vps = NULL;
1095 ASSERT(vd->vdev_ops->vdev_op_leaf);
1098 * Don't probe the probe.
1100 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1104 * To prevent 'probe storms' when a device fails, we create
1105 * just one probe i/o at a time. All zios that want to probe
1106 * this vdev will become parents of the probe io.
1108 mutex_enter(&vd->vdev_probe_lock);
1110 if ((pio = vd->vdev_probe_zio) == NULL) {
1111 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1113 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1114 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1117 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1119 * vdev_cant_read and vdev_cant_write can only
1120 * transition from TRUE to FALSE when we have the
1121 * SCL_ZIO lock as writer; otherwise they can only
1122 * transition from FALSE to TRUE. This ensures that
1123 * any zio looking at these values can assume that
1124 * failures persist for the life of the I/O. That's
1125 * important because when a device has intermittent
1126 * connectivity problems, we want to ensure that
1127 * they're ascribed to the device (ENXIO) and not
1130 * Since we hold SCL_ZIO as writer here, clear both
1131 * values so the probe can reevaluate from first
1134 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1135 vd->vdev_cant_read = B_FALSE;
1136 vd->vdev_cant_write = B_FALSE;
1139 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1140 vdev_probe_done, vps,
1141 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1144 * We can't change the vdev state in this context, so we
1145 * kick off an async task to do it on our behalf.
1148 vd->vdev_probe_wanted = B_TRUE;
1149 spa_async_request(spa, SPA_ASYNC_PROBE);
1154 zio_add_child(zio, pio);
1156 mutex_exit(&vd->vdev_probe_lock);
1159 ASSERT(zio != NULL);
1163 for (int l = 1; l < VDEV_LABELS; l++) {
1164 zio_nowait(zio_read_phys(pio, vd,
1165 vdev_label_offset(vd->vdev_psize, l,
1166 offsetof(vdev_label_t, vl_pad2)),
1167 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1168 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1169 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1180 vdev_open_child(void *arg)
1184 vd->vdev_open_thread = curthread;
1185 vd->vdev_open_error = vdev_open(vd);
1186 vd->vdev_open_thread = NULL;
1190 vdev_uses_zvols(vdev_t *vd)
1192 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1193 strlen(ZVOL_DIR)) == 0)
1195 for (int c = 0; c < vd->vdev_children; c++)
1196 if (vdev_uses_zvols(vd->vdev_child[c]))
1202 vdev_open_children(vdev_t *vd)
1205 int children = vd->vdev_children;
1208 * in order to handle pools on top of zvols, do the opens
1209 * in a single thread so that the same thread holds the
1210 * spa_namespace_lock
1212 if (B_TRUE || vdev_uses_zvols(vd)) {
1213 for (int c = 0; c < children; c++)
1214 vd->vdev_child[c]->vdev_open_error =
1215 vdev_open(vd->vdev_child[c]);
1218 tq = taskq_create("vdev_open", children, minclsyspri,
1219 children, children, TASKQ_PREPOPULATE);
1221 for (int c = 0; c < children; c++)
1222 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1229 * Prepare a virtual device for access.
1232 vdev_open(vdev_t *vd)
1234 spa_t *spa = vd->vdev_spa;
1237 uint64_t max_osize = 0;
1238 uint64_t asize, max_asize, psize;
1239 uint64_t logical_ashift = 0;
1240 uint64_t physical_ashift = 0;
1242 ASSERT(vd->vdev_open_thread == curthread ||
1243 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1244 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1245 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1246 vd->vdev_state == VDEV_STATE_OFFLINE);
1248 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1249 vd->vdev_cant_read = B_FALSE;
1250 vd->vdev_cant_write = B_FALSE;
1251 vd->vdev_notrim = B_FALSE;
1252 vd->vdev_min_asize = vdev_get_min_asize(vd);
1255 * If this vdev is not removed, check its fault status. If it's
1256 * faulted, bail out of the open.
1258 if (!vd->vdev_removed && vd->vdev_faulted) {
1259 ASSERT(vd->vdev_children == 0);
1260 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1261 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1262 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1263 vd->vdev_label_aux);
1264 return (SET_ERROR(ENXIO));
1265 } else if (vd->vdev_offline) {
1266 ASSERT(vd->vdev_children == 0);
1267 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1268 return (SET_ERROR(ENXIO));
1271 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1272 &logical_ashift, &physical_ashift);
1275 * Reset the vdev_reopening flag so that we actually close
1276 * the vdev on error.
1278 vd->vdev_reopening = B_FALSE;
1279 if (zio_injection_enabled && error == 0)
1280 error = zio_handle_device_injection(vd, NULL, ENXIO);
1283 if (vd->vdev_removed &&
1284 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1285 vd->vdev_removed = B_FALSE;
1287 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1288 vd->vdev_stat.vs_aux);
1292 vd->vdev_removed = B_FALSE;
1295 * Recheck the faulted flag now that we have confirmed that
1296 * the vdev is accessible. If we're faulted, bail.
1298 if (vd->vdev_faulted) {
1299 ASSERT(vd->vdev_children == 0);
1300 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1301 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1302 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1303 vd->vdev_label_aux);
1304 return (SET_ERROR(ENXIO));
1307 if (vd->vdev_degraded) {
1308 ASSERT(vd->vdev_children == 0);
1309 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1310 VDEV_AUX_ERR_EXCEEDED);
1312 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1316 * For hole or missing vdevs we just return success.
1318 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1321 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1322 trim_map_create(vd);
1324 for (int c = 0; c < vd->vdev_children; c++) {
1325 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1326 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1332 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1333 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1335 if (vd->vdev_children == 0) {
1336 if (osize < SPA_MINDEVSIZE) {
1337 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1338 VDEV_AUX_TOO_SMALL);
1339 return (SET_ERROR(EOVERFLOW));
1342 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1343 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1344 VDEV_LABEL_END_SIZE);
1346 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1347 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1348 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1349 VDEV_AUX_TOO_SMALL);
1350 return (SET_ERROR(EOVERFLOW));
1354 max_asize = max_osize;
1357 vd->vdev_psize = psize;
1360 * Make sure the allocatable size hasn't shrunk.
1362 if (asize < vd->vdev_min_asize) {
1363 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1364 VDEV_AUX_BAD_LABEL);
1365 return (SET_ERROR(EINVAL));
1368 vd->vdev_physical_ashift =
1369 MAX(physical_ashift, vd->vdev_physical_ashift);
1370 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1371 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1373 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1374 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1375 VDEV_AUX_ASHIFT_TOO_BIG);
1379 if (vd->vdev_asize == 0) {
1381 * This is the first-ever open, so use the computed values.
1382 * For testing purposes, a higher ashift can be requested.
1384 vd->vdev_asize = asize;
1385 vd->vdev_max_asize = max_asize;
1388 * Make sure the alignment requirement hasn't increased.
1390 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1391 vd->vdev_ops->vdev_op_leaf) {
1392 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1393 VDEV_AUX_BAD_LABEL);
1396 vd->vdev_max_asize = max_asize;
1400 * If all children are healthy and the asize has increased,
1401 * then we've experienced dynamic LUN growth. If automatic
1402 * expansion is enabled then use the additional space.
1404 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1405 (vd->vdev_expanding || spa->spa_autoexpand))
1406 vd->vdev_asize = asize;
1408 vdev_set_min_asize(vd);
1411 * Ensure we can issue some IO before declaring the
1412 * vdev open for business.
1414 if (vd->vdev_ops->vdev_op_leaf &&
1415 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1416 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1417 VDEV_AUX_ERR_EXCEEDED);
1422 * Track the min and max ashift values for normal data devices.
1424 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1425 !vd->vdev_islog && vd->vdev_aux == NULL) {
1426 if (vd->vdev_ashift > spa->spa_max_ashift)
1427 spa->spa_max_ashift = vd->vdev_ashift;
1428 if (vd->vdev_ashift < spa->spa_min_ashift)
1429 spa->spa_min_ashift = vd->vdev_ashift;
1433 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1434 * resilver. But don't do this if we are doing a reopen for a scrub,
1435 * since this would just restart the scrub we are already doing.
1437 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1438 vdev_resilver_needed(vd, NULL, NULL))
1439 spa_async_request(spa, SPA_ASYNC_RESILVER);
1445 * Called once the vdevs are all opened, this routine validates the label
1446 * contents. This needs to be done before vdev_load() so that we don't
1447 * inadvertently do repair I/Os to the wrong device.
1449 * If 'strict' is false ignore the spa guid check. This is necessary because
1450 * if the machine crashed during a re-guid the new guid might have been written
1451 * to all of the vdev labels, but not the cached config. The strict check
1452 * will be performed when the pool is opened again using the mos config.
1454 * This function will only return failure if one of the vdevs indicates that it
1455 * has since been destroyed or exported. This is only possible if
1456 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1457 * will be updated but the function will return 0.
1460 vdev_validate(vdev_t *vd, boolean_t strict)
1462 spa_t *spa = vd->vdev_spa;
1464 uint64_t guid = 0, top_guid;
1467 for (int c = 0; c < vd->vdev_children; c++)
1468 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1469 return (SET_ERROR(EBADF));
1472 * If the device has already failed, or was marked offline, don't do
1473 * any further validation. Otherwise, label I/O will fail and we will
1474 * overwrite the previous state.
1476 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1477 uint64_t aux_guid = 0;
1479 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1480 spa_last_synced_txg(spa) : -1ULL;
1482 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1483 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1484 VDEV_AUX_BAD_LABEL);
1489 * Determine if this vdev has been split off into another
1490 * pool. If so, then refuse to open it.
1492 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1493 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1494 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1495 VDEV_AUX_SPLIT_POOL);
1500 if (strict && (nvlist_lookup_uint64(label,
1501 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1502 guid != spa_guid(spa))) {
1503 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1504 VDEV_AUX_CORRUPT_DATA);
1509 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1510 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1515 * If this vdev just became a top-level vdev because its
1516 * sibling was detached, it will have adopted the parent's
1517 * vdev guid -- but the label may or may not be on disk yet.
1518 * Fortunately, either version of the label will have the
1519 * same top guid, so if we're a top-level vdev, we can
1520 * safely compare to that instead.
1522 * If we split this vdev off instead, then we also check the
1523 * original pool's guid. We don't want to consider the vdev
1524 * corrupt if it is partway through a split operation.
1526 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1528 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1530 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1531 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1532 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1533 VDEV_AUX_CORRUPT_DATA);
1538 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1540 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1541 VDEV_AUX_CORRUPT_DATA);
1549 * If this is a verbatim import, no need to check the
1550 * state of the pool.
1552 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1553 spa_load_state(spa) == SPA_LOAD_OPEN &&
1554 state != POOL_STATE_ACTIVE)
1555 return (SET_ERROR(EBADF));
1558 * If we were able to open and validate a vdev that was
1559 * previously marked permanently unavailable, clear that state
1562 if (vd->vdev_not_present)
1563 vd->vdev_not_present = 0;
1570 * Close a virtual device.
1573 vdev_close(vdev_t *vd)
1575 spa_t *spa = vd->vdev_spa;
1576 vdev_t *pvd = vd->vdev_parent;
1578 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1581 * If our parent is reopening, then we are as well, unless we are
1584 if (pvd != NULL && pvd->vdev_reopening)
1585 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1587 vd->vdev_ops->vdev_op_close(vd);
1589 vdev_cache_purge(vd);
1591 if (vd->vdev_ops->vdev_op_leaf)
1592 trim_map_destroy(vd);
1595 * We record the previous state before we close it, so that if we are
1596 * doing a reopen(), we don't generate FMA ereports if we notice that
1597 * it's still faulted.
1599 vd->vdev_prevstate = vd->vdev_state;
1601 if (vd->vdev_offline)
1602 vd->vdev_state = VDEV_STATE_OFFLINE;
1604 vd->vdev_state = VDEV_STATE_CLOSED;
1605 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1609 vdev_hold(vdev_t *vd)
1611 spa_t *spa = vd->vdev_spa;
1613 ASSERT(spa_is_root(spa));
1614 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1617 for (int c = 0; c < vd->vdev_children; c++)
1618 vdev_hold(vd->vdev_child[c]);
1620 if (vd->vdev_ops->vdev_op_leaf)
1621 vd->vdev_ops->vdev_op_hold(vd);
1625 vdev_rele(vdev_t *vd)
1627 spa_t *spa = vd->vdev_spa;
1629 ASSERT(spa_is_root(spa));
1630 for (int c = 0; c < vd->vdev_children; c++)
1631 vdev_rele(vd->vdev_child[c]);
1633 if (vd->vdev_ops->vdev_op_leaf)
1634 vd->vdev_ops->vdev_op_rele(vd);
1638 * Reopen all interior vdevs and any unopened leaves. We don't actually
1639 * reopen leaf vdevs which had previously been opened as they might deadlock
1640 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1641 * If the leaf has never been opened then open it, as usual.
1644 vdev_reopen(vdev_t *vd)
1646 spa_t *spa = vd->vdev_spa;
1648 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1650 /* set the reopening flag unless we're taking the vdev offline */
1651 vd->vdev_reopening = !vd->vdev_offline;
1653 (void) vdev_open(vd);
1656 * Call vdev_validate() here to make sure we have the same device.
1657 * Otherwise, a device with an invalid label could be successfully
1658 * opened in response to vdev_reopen().
1661 (void) vdev_validate_aux(vd);
1662 if (vdev_readable(vd) && vdev_writeable(vd) &&
1663 vd->vdev_aux == &spa->spa_l2cache &&
1664 !l2arc_vdev_present(vd))
1665 l2arc_add_vdev(spa, vd);
1667 (void) vdev_validate(vd, B_TRUE);
1671 * Reassess parent vdev's health.
1673 vdev_propagate_state(vd);
1677 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1682 * Normally, partial opens (e.g. of a mirror) are allowed.
1683 * For a create, however, we want to fail the request if
1684 * there are any components we can't open.
1686 error = vdev_open(vd);
1688 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1690 return (error ? error : ENXIO);
1694 * Recursively load DTLs and initialize all labels.
1696 if ((error = vdev_dtl_load(vd)) != 0 ||
1697 (error = vdev_label_init(vd, txg, isreplacing ?
1698 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1707 vdev_metaslab_set_size(vdev_t *vd)
1710 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1712 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1713 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1717 * Maximize performance by inflating the configured ashift for top level
1718 * vdevs to be as close to the physical ashift as possible while maintaining
1719 * administrator defined limits and ensuring it doesn't go below the
1723 vdev_ashift_optimize(vdev_t *vd)
1725 if (vd == vd->vdev_top) {
1726 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1727 vd->vdev_ashift = MIN(
1728 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1729 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1732 * Unusual case where logical ashift > physical ashift
1733 * so we can't cap the calculated ashift based on max
1734 * ashift as that would cause failures.
1735 * We still check if we need to increase it to match
1738 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1745 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1747 ASSERT(vd == vd->vdev_top);
1748 ASSERT(!vd->vdev_ishole);
1749 ASSERT(ISP2(flags));
1750 ASSERT(spa_writeable(vd->vdev_spa));
1752 if (flags & VDD_METASLAB)
1753 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1755 if (flags & VDD_DTL)
1756 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1758 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1762 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1764 for (int c = 0; c < vd->vdev_children; c++)
1765 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1767 if (vd->vdev_ops->vdev_op_leaf)
1768 vdev_dirty(vd->vdev_top, flags, vd, txg);
1774 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1775 * the vdev has less than perfect replication. There are four kinds of DTL:
1777 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1779 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1781 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1782 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1783 * txgs that was scrubbed.
1785 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1786 * persistent errors or just some device being offline.
1787 * Unlike the other three, the DTL_OUTAGE map is not generally
1788 * maintained; it's only computed when needed, typically to
1789 * determine whether a device can be detached.
1791 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1792 * either has the data or it doesn't.
1794 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1795 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1796 * if any child is less than fully replicated, then so is its parent.
1797 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1798 * comprising only those txgs which appear in 'maxfaults' or more children;
1799 * those are the txgs we don't have enough replication to read. For example,
1800 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1801 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1802 * two child DTL_MISSING maps.
1804 * It should be clear from the above that to compute the DTLs and outage maps
1805 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1806 * Therefore, that is all we keep on disk. When loading the pool, or after
1807 * a configuration change, we generate all other DTLs from first principles.
1810 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1812 range_tree_t *rt = vd->vdev_dtl[t];
1814 ASSERT(t < DTL_TYPES);
1815 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1816 ASSERT(spa_writeable(vd->vdev_spa));
1818 mutex_enter(rt->rt_lock);
1819 if (!range_tree_contains(rt, txg, size))
1820 range_tree_add(rt, txg, size);
1821 mutex_exit(rt->rt_lock);
1825 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1827 range_tree_t *rt = vd->vdev_dtl[t];
1828 boolean_t dirty = B_FALSE;
1830 ASSERT(t < DTL_TYPES);
1831 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1833 mutex_enter(rt->rt_lock);
1834 if (range_tree_space(rt) != 0)
1835 dirty = range_tree_contains(rt, txg, size);
1836 mutex_exit(rt->rt_lock);
1842 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1844 range_tree_t *rt = vd->vdev_dtl[t];
1847 mutex_enter(rt->rt_lock);
1848 empty = (range_tree_space(rt) == 0);
1849 mutex_exit(rt->rt_lock);
1855 * Returns the lowest txg in the DTL range.
1858 vdev_dtl_min(vdev_t *vd)
1862 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1863 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1864 ASSERT0(vd->vdev_children);
1866 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1867 return (rs->rs_start - 1);
1871 * Returns the highest txg in the DTL.
1874 vdev_dtl_max(vdev_t *vd)
1878 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1879 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1880 ASSERT0(vd->vdev_children);
1882 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1883 return (rs->rs_end);
1887 * Determine if a resilvering vdev should remove any DTL entries from
1888 * its range. If the vdev was resilvering for the entire duration of the
1889 * scan then it should excise that range from its DTLs. Otherwise, this
1890 * vdev is considered partially resilvered and should leave its DTL
1891 * entries intact. The comment in vdev_dtl_reassess() describes how we
1895 vdev_dtl_should_excise(vdev_t *vd)
1897 spa_t *spa = vd->vdev_spa;
1898 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1900 ASSERT0(scn->scn_phys.scn_errors);
1901 ASSERT0(vd->vdev_children);
1903 if (vd->vdev_resilver_txg == 0 ||
1904 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1908 * When a resilver is initiated the scan will assign the scn_max_txg
1909 * value to the highest txg value that exists in all DTLs. If this
1910 * device's max DTL is not part of this scan (i.e. it is not in
1911 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1914 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1915 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1916 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1917 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1924 * Reassess DTLs after a config change or scrub completion.
1927 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1929 spa_t *spa = vd->vdev_spa;
1933 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1935 for (int c = 0; c < vd->vdev_children; c++)
1936 vdev_dtl_reassess(vd->vdev_child[c], txg,
1937 scrub_txg, scrub_done);
1939 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1942 if (vd->vdev_ops->vdev_op_leaf) {
1943 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1945 mutex_enter(&vd->vdev_dtl_lock);
1948 * If we've completed a scan cleanly then determine
1949 * if this vdev should remove any DTLs. We only want to
1950 * excise regions on vdevs that were available during
1951 * the entire duration of this scan.
1953 if (scrub_txg != 0 &&
1954 (spa->spa_scrub_started ||
1955 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1956 vdev_dtl_should_excise(vd)) {
1958 * We completed a scrub up to scrub_txg. If we
1959 * did it without rebooting, then the scrub dtl
1960 * will be valid, so excise the old region and
1961 * fold in the scrub dtl. Otherwise, leave the
1962 * dtl as-is if there was an error.
1964 * There's little trick here: to excise the beginning
1965 * of the DTL_MISSING map, we put it into a reference
1966 * tree and then add a segment with refcnt -1 that
1967 * covers the range [0, scrub_txg). This means
1968 * that each txg in that range has refcnt -1 or 0.
1969 * We then add DTL_SCRUB with a refcnt of 2, so that
1970 * entries in the range [0, scrub_txg) will have a
1971 * positive refcnt -- either 1 or 2. We then convert
1972 * the reference tree into the new DTL_MISSING map.
1974 space_reftree_create(&reftree);
1975 space_reftree_add_map(&reftree,
1976 vd->vdev_dtl[DTL_MISSING], 1);
1977 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1978 space_reftree_add_map(&reftree,
1979 vd->vdev_dtl[DTL_SCRUB], 2);
1980 space_reftree_generate_map(&reftree,
1981 vd->vdev_dtl[DTL_MISSING], 1);
1982 space_reftree_destroy(&reftree);
1984 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1985 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1986 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1988 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1989 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1990 if (!vdev_readable(vd))
1991 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1993 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1994 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1997 * If the vdev was resilvering and no longer has any
1998 * DTLs then reset its resilvering flag and dirty
1999 * the top level so that we persist the change.
2001 if (vd->vdev_resilver_txg != 0 &&
2002 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
2003 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
2004 vd->vdev_resilver_txg = 0;
2005 vdev_config_dirty(vd->vdev_top);
2008 mutex_exit(&vd->vdev_dtl_lock);
2011 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2015 mutex_enter(&vd->vdev_dtl_lock);
2016 for (int t = 0; t < DTL_TYPES; t++) {
2017 /* account for child's outage in parent's missing map */
2018 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2020 continue; /* leaf vdevs only */
2021 if (t == DTL_PARTIAL)
2022 minref = 1; /* i.e. non-zero */
2023 else if (vd->vdev_nparity != 0)
2024 minref = vd->vdev_nparity + 1; /* RAID-Z */
2026 minref = vd->vdev_children; /* any kind of mirror */
2027 space_reftree_create(&reftree);
2028 for (int c = 0; c < vd->vdev_children; c++) {
2029 vdev_t *cvd = vd->vdev_child[c];
2030 mutex_enter(&cvd->vdev_dtl_lock);
2031 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2032 mutex_exit(&cvd->vdev_dtl_lock);
2034 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2035 space_reftree_destroy(&reftree);
2037 mutex_exit(&vd->vdev_dtl_lock);
2041 vdev_dtl_load(vdev_t *vd)
2043 spa_t *spa = vd->vdev_spa;
2044 objset_t *mos = spa->spa_meta_objset;
2047 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2048 ASSERT(!vd->vdev_ishole);
2050 error = space_map_open(&vd->vdev_dtl_sm, mos,
2051 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2054 ASSERT(vd->vdev_dtl_sm != NULL);
2056 mutex_enter(&vd->vdev_dtl_lock);
2059 * Now that we've opened the space_map we need to update
2062 space_map_update(vd->vdev_dtl_sm);
2064 error = space_map_load(vd->vdev_dtl_sm,
2065 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2066 mutex_exit(&vd->vdev_dtl_lock);
2071 for (int c = 0; c < vd->vdev_children; c++) {
2072 error = vdev_dtl_load(vd->vdev_child[c]);
2081 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2083 spa_t *spa = vd->vdev_spa;
2084 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2085 objset_t *mos = spa->spa_meta_objset;
2086 range_tree_t *rtsync;
2089 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2091 ASSERT(!vd->vdev_ishole);
2092 ASSERT(vd->vdev_ops->vdev_op_leaf);
2094 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2096 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2097 mutex_enter(&vd->vdev_dtl_lock);
2098 space_map_free(vd->vdev_dtl_sm, tx);
2099 space_map_close(vd->vdev_dtl_sm);
2100 vd->vdev_dtl_sm = NULL;
2101 mutex_exit(&vd->vdev_dtl_lock);
2106 if (vd->vdev_dtl_sm == NULL) {
2107 uint64_t new_object;
2109 new_object = space_map_alloc(mos, tx);
2110 VERIFY3U(new_object, !=, 0);
2112 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2113 0, -1ULL, 0, &vd->vdev_dtl_lock));
2114 ASSERT(vd->vdev_dtl_sm != NULL);
2117 bzero(&rtlock, sizeof(rtlock));
2118 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2120 rtsync = range_tree_create(NULL, NULL, &rtlock);
2122 mutex_enter(&rtlock);
2124 mutex_enter(&vd->vdev_dtl_lock);
2125 range_tree_walk(rt, range_tree_add, rtsync);
2126 mutex_exit(&vd->vdev_dtl_lock);
2128 space_map_truncate(vd->vdev_dtl_sm, tx);
2129 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2130 range_tree_vacate(rtsync, NULL, NULL);
2132 range_tree_destroy(rtsync);
2134 mutex_exit(&rtlock);
2135 mutex_destroy(&rtlock);
2138 * If the object for the space map has changed then dirty
2139 * the top level so that we update the config.
2141 if (object != space_map_object(vd->vdev_dtl_sm)) {
2142 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2143 "new object %llu", txg, spa_name(spa), object,
2144 space_map_object(vd->vdev_dtl_sm));
2145 vdev_config_dirty(vd->vdev_top);
2150 mutex_enter(&vd->vdev_dtl_lock);
2151 space_map_update(vd->vdev_dtl_sm);
2152 mutex_exit(&vd->vdev_dtl_lock);
2156 * Determine whether the specified vdev can be offlined/detached/removed
2157 * without losing data.
2160 vdev_dtl_required(vdev_t *vd)
2162 spa_t *spa = vd->vdev_spa;
2163 vdev_t *tvd = vd->vdev_top;
2164 uint8_t cant_read = vd->vdev_cant_read;
2167 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2169 if (vd == spa->spa_root_vdev || vd == tvd)
2173 * Temporarily mark the device as unreadable, and then determine
2174 * whether this results in any DTL outages in the top-level vdev.
2175 * If not, we can safely offline/detach/remove the device.
2177 vd->vdev_cant_read = B_TRUE;
2178 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2179 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2180 vd->vdev_cant_read = cant_read;
2181 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2183 if (!required && zio_injection_enabled)
2184 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2190 * Determine if resilver is needed, and if so the txg range.
2193 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2195 boolean_t needed = B_FALSE;
2196 uint64_t thismin = UINT64_MAX;
2197 uint64_t thismax = 0;
2199 if (vd->vdev_children == 0) {
2200 mutex_enter(&vd->vdev_dtl_lock);
2201 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2202 vdev_writeable(vd)) {
2204 thismin = vdev_dtl_min(vd);
2205 thismax = vdev_dtl_max(vd);
2208 mutex_exit(&vd->vdev_dtl_lock);
2210 for (int c = 0; c < vd->vdev_children; c++) {
2211 vdev_t *cvd = vd->vdev_child[c];
2212 uint64_t cmin, cmax;
2214 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2215 thismin = MIN(thismin, cmin);
2216 thismax = MAX(thismax, cmax);
2222 if (needed && minp) {
2230 vdev_load(vdev_t *vd)
2233 * Recursively load all children.
2235 for (int c = 0; c < vd->vdev_children; c++)
2236 vdev_load(vd->vdev_child[c]);
2239 * If this is a top-level vdev, initialize its metaslabs.
2241 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2242 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2243 vdev_metaslab_init(vd, 0) != 0))
2244 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2245 VDEV_AUX_CORRUPT_DATA);
2248 * If this is a leaf vdev, load its DTL.
2250 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2251 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2252 VDEV_AUX_CORRUPT_DATA);
2256 * The special vdev case is used for hot spares and l2cache devices. Its
2257 * sole purpose it to set the vdev state for the associated vdev. To do this,
2258 * we make sure that we can open the underlying device, then try to read the
2259 * label, and make sure that the label is sane and that it hasn't been
2260 * repurposed to another pool.
2263 vdev_validate_aux(vdev_t *vd)
2266 uint64_t guid, version;
2269 if (!vdev_readable(vd))
2272 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2273 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2274 VDEV_AUX_CORRUPT_DATA);
2278 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2279 !SPA_VERSION_IS_SUPPORTED(version) ||
2280 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2281 guid != vd->vdev_guid ||
2282 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2283 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2284 VDEV_AUX_CORRUPT_DATA);
2290 * We don't actually check the pool state here. If it's in fact in
2291 * use by another pool, we update this fact on the fly when requested.
2298 vdev_remove(vdev_t *vd, uint64_t txg)
2300 spa_t *spa = vd->vdev_spa;
2301 objset_t *mos = spa->spa_meta_objset;
2304 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2306 if (vd->vdev_ms != NULL) {
2307 metaslab_group_t *mg = vd->vdev_mg;
2309 metaslab_group_histogram_verify(mg);
2310 metaslab_class_histogram_verify(mg->mg_class);
2312 for (int m = 0; m < vd->vdev_ms_count; m++) {
2313 metaslab_t *msp = vd->vdev_ms[m];
2315 if (msp == NULL || msp->ms_sm == NULL)
2318 mutex_enter(&msp->ms_lock);
2320 * If the metaslab was not loaded when the vdev
2321 * was removed then the histogram accounting may
2322 * not be accurate. Update the histogram information
2323 * here so that we ensure that the metaslab group
2324 * and metaslab class are up-to-date.
2326 metaslab_group_histogram_remove(mg, msp);
2328 VERIFY0(space_map_allocated(msp->ms_sm));
2329 space_map_free(msp->ms_sm, tx);
2330 space_map_close(msp->ms_sm);
2332 mutex_exit(&msp->ms_lock);
2335 metaslab_group_histogram_verify(mg);
2336 metaslab_class_histogram_verify(mg->mg_class);
2337 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2338 ASSERT0(mg->mg_histogram[i]);
2342 if (vd->vdev_ms_array) {
2343 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2344 vd->vdev_ms_array = 0;
2350 vdev_sync_done(vdev_t *vd, uint64_t txg)
2353 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2355 ASSERT(!vd->vdev_ishole);
2357 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2358 metaslab_sync_done(msp, txg);
2361 metaslab_sync_reassess(vd->vdev_mg);
2365 vdev_sync(vdev_t *vd, uint64_t txg)
2367 spa_t *spa = vd->vdev_spa;
2372 ASSERT(!vd->vdev_ishole);
2374 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2375 ASSERT(vd == vd->vdev_top);
2376 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2377 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2378 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2379 ASSERT(vd->vdev_ms_array != 0);
2380 vdev_config_dirty(vd);
2385 * Remove the metadata associated with this vdev once it's empty.
2387 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2388 vdev_remove(vd, txg);
2390 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2391 metaslab_sync(msp, txg);
2392 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2395 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2396 vdev_dtl_sync(lvd, txg);
2398 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2402 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2404 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2408 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2409 * not be opened, and no I/O is attempted.
2412 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2416 spa_vdev_state_enter(spa, SCL_NONE);
2418 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2419 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2421 if (!vd->vdev_ops->vdev_op_leaf)
2422 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2427 * We don't directly use the aux state here, but if we do a
2428 * vdev_reopen(), we need this value to be present to remember why we
2431 vd->vdev_label_aux = aux;
2434 * Faulted state takes precedence over degraded.
2436 vd->vdev_delayed_close = B_FALSE;
2437 vd->vdev_faulted = 1ULL;
2438 vd->vdev_degraded = 0ULL;
2439 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2442 * If this device has the only valid copy of the data, then
2443 * back off and simply mark the vdev as degraded instead.
2445 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2446 vd->vdev_degraded = 1ULL;
2447 vd->vdev_faulted = 0ULL;
2450 * If we reopen the device and it's not dead, only then do we
2455 if (vdev_readable(vd))
2456 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2459 return (spa_vdev_state_exit(spa, vd, 0));
2463 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2464 * user that something is wrong. The vdev continues to operate as normal as far
2465 * as I/O is concerned.
2468 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2472 spa_vdev_state_enter(spa, SCL_NONE);
2474 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2475 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2477 if (!vd->vdev_ops->vdev_op_leaf)
2478 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2481 * If the vdev is already faulted, then don't do anything.
2483 if (vd->vdev_faulted || vd->vdev_degraded)
2484 return (spa_vdev_state_exit(spa, NULL, 0));
2486 vd->vdev_degraded = 1ULL;
2487 if (!vdev_is_dead(vd))
2488 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2491 return (spa_vdev_state_exit(spa, vd, 0));
2495 * Online the given vdev.
2497 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2498 * spare device should be detached when the device finishes resilvering.
2499 * Second, the online should be treated like a 'test' online case, so no FMA
2500 * events are generated if the device fails to open.
2503 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2505 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2506 boolean_t postevent = B_FALSE;
2508 spa_vdev_state_enter(spa, SCL_NONE);
2510 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2511 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2513 if (!vd->vdev_ops->vdev_op_leaf)
2514 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2517 (vd->vdev_offline == B_TRUE || vd->vdev_tmpoffline == B_TRUE) ?
2521 vd->vdev_offline = B_FALSE;
2522 vd->vdev_tmpoffline = B_FALSE;
2523 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2524 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2526 /* XXX - L2ARC 1.0 does not support expansion */
2527 if (!vd->vdev_aux) {
2528 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2529 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2533 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2535 if (!vd->vdev_aux) {
2536 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2537 pvd->vdev_expanding = B_FALSE;
2541 *newstate = vd->vdev_state;
2542 if ((flags & ZFS_ONLINE_UNSPARE) &&
2543 !vdev_is_dead(vd) && vd->vdev_parent &&
2544 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2545 vd->vdev_parent->vdev_child[0] == vd)
2546 vd->vdev_unspare = B_TRUE;
2548 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2550 /* XXX - L2ARC 1.0 does not support expansion */
2552 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2553 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2557 spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE);
2559 return (spa_vdev_state_exit(spa, vd, 0));
2563 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2567 uint64_t generation;
2568 metaslab_group_t *mg;
2571 spa_vdev_state_enter(spa, SCL_ALLOC);
2573 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2574 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2576 if (!vd->vdev_ops->vdev_op_leaf)
2577 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2581 generation = spa->spa_config_generation + 1;
2584 * If the device isn't already offline, try to offline it.
2586 if (!vd->vdev_offline) {
2588 * If this device has the only valid copy of some data,
2589 * don't allow it to be offlined. Log devices are always
2592 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2593 vdev_dtl_required(vd))
2594 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2597 * If the top-level is a slog and it has had allocations
2598 * then proceed. We check that the vdev's metaslab group
2599 * is not NULL since it's possible that we may have just
2600 * added this vdev but not yet initialized its metaslabs.
2602 if (tvd->vdev_islog && mg != NULL) {
2604 * Prevent any future allocations.
2606 metaslab_group_passivate(mg);
2607 (void) spa_vdev_state_exit(spa, vd, 0);
2609 error = spa_offline_log(spa);
2611 spa_vdev_state_enter(spa, SCL_ALLOC);
2614 * Check to see if the config has changed.
2616 if (error || generation != spa->spa_config_generation) {
2617 metaslab_group_activate(mg);
2619 return (spa_vdev_state_exit(spa,
2621 (void) spa_vdev_state_exit(spa, vd, 0);
2624 ASSERT0(tvd->vdev_stat.vs_alloc);
2628 * Offline this device and reopen its top-level vdev.
2629 * If the top-level vdev is a log device then just offline
2630 * it. Otherwise, if this action results in the top-level
2631 * vdev becoming unusable, undo it and fail the request.
2633 vd->vdev_offline = B_TRUE;
2636 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2637 vdev_is_dead(tvd)) {
2638 vd->vdev_offline = B_FALSE;
2640 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2644 * Add the device back into the metaslab rotor so that
2645 * once we online the device it's open for business.
2647 if (tvd->vdev_islog && mg != NULL)
2648 metaslab_group_activate(mg);
2651 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2653 return (spa_vdev_state_exit(spa, vd, 0));
2657 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2661 mutex_enter(&spa->spa_vdev_top_lock);
2662 error = vdev_offline_locked(spa, guid, flags);
2663 mutex_exit(&spa->spa_vdev_top_lock);
2669 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2670 * vdev_offline(), we assume the spa config is locked. We also clear all
2671 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2674 vdev_clear(spa_t *spa, vdev_t *vd)
2676 vdev_t *rvd = spa->spa_root_vdev;
2678 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2683 vd->vdev_stat.vs_read_errors = 0;
2684 vd->vdev_stat.vs_write_errors = 0;
2685 vd->vdev_stat.vs_checksum_errors = 0;
2687 for (int c = 0; c < vd->vdev_children; c++)
2688 vdev_clear(spa, vd->vdev_child[c]);
2691 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2692 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2694 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2695 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2699 * If we're in the FAULTED state or have experienced failed I/O, then
2700 * clear the persistent state and attempt to reopen the device. We
2701 * also mark the vdev config dirty, so that the new faulted state is
2702 * written out to disk.
2704 if (vd->vdev_faulted || vd->vdev_degraded ||
2705 !vdev_readable(vd) || !vdev_writeable(vd)) {
2708 * When reopening in reponse to a clear event, it may be due to
2709 * a fmadm repair request. In this case, if the device is
2710 * still broken, we want to still post the ereport again.
2712 vd->vdev_forcefault = B_TRUE;
2714 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2715 vd->vdev_cant_read = B_FALSE;
2716 vd->vdev_cant_write = B_FALSE;
2718 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2720 vd->vdev_forcefault = B_FALSE;
2722 if (vd != rvd && vdev_writeable(vd->vdev_top))
2723 vdev_state_dirty(vd->vdev_top);
2725 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2726 spa_async_request(spa, SPA_ASYNC_RESILVER);
2728 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2732 * When clearing a FMA-diagnosed fault, we always want to
2733 * unspare the device, as we assume that the original spare was
2734 * done in response to the FMA fault.
2736 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2737 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2738 vd->vdev_parent->vdev_child[0] == vd)
2739 vd->vdev_unspare = B_TRUE;
2743 vdev_is_dead(vdev_t *vd)
2746 * Holes and missing devices are always considered "dead".
2747 * This simplifies the code since we don't have to check for
2748 * these types of devices in the various code paths.
2749 * Instead we rely on the fact that we skip over dead devices
2750 * before issuing I/O to them.
2752 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2753 vd->vdev_ops == &vdev_missing_ops);
2757 vdev_readable(vdev_t *vd)
2759 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2763 vdev_writeable(vdev_t *vd)
2765 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2769 vdev_allocatable(vdev_t *vd)
2771 uint64_t state = vd->vdev_state;
2774 * We currently allow allocations from vdevs which may be in the
2775 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2776 * fails to reopen then we'll catch it later when we're holding
2777 * the proper locks. Note that we have to get the vdev state
2778 * in a local variable because although it changes atomically,
2779 * we're asking two separate questions about it.
2781 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2782 !vd->vdev_cant_write && !vd->vdev_ishole);
2786 vdev_accessible(vdev_t *vd, zio_t *zio)
2788 ASSERT(zio->io_vd == vd);
2790 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2793 if (zio->io_type == ZIO_TYPE_READ)
2794 return (!vd->vdev_cant_read);
2796 if (zio->io_type == ZIO_TYPE_WRITE)
2797 return (!vd->vdev_cant_write);
2803 * Get statistics for the given vdev.
2806 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2808 spa_t *spa = vd->vdev_spa;
2809 vdev_t *rvd = spa->spa_root_vdev;
2811 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2813 mutex_enter(&vd->vdev_stat_lock);
2814 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2815 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2816 vs->vs_state = vd->vdev_state;
2817 vs->vs_rsize = vdev_get_min_asize(vd);
2818 if (vd->vdev_ops->vdev_op_leaf)
2819 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2820 if (vd->vdev_max_asize != 0)
2821 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2822 vs->vs_configured_ashift = vd->vdev_top != NULL
2823 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2824 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2825 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2826 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2827 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2831 * If we're getting stats on the root vdev, aggregate the I/O counts
2832 * over all top-level vdevs (i.e. the direct children of the root).
2835 for (int c = 0; c < rvd->vdev_children; c++) {
2836 vdev_t *cvd = rvd->vdev_child[c];
2837 vdev_stat_t *cvs = &cvd->vdev_stat;
2839 for (int t = 0; t < ZIO_TYPES; t++) {
2840 vs->vs_ops[t] += cvs->vs_ops[t];
2841 vs->vs_bytes[t] += cvs->vs_bytes[t];
2843 cvs->vs_scan_removing = cvd->vdev_removing;
2846 mutex_exit(&vd->vdev_stat_lock);
2850 vdev_clear_stats(vdev_t *vd)
2852 mutex_enter(&vd->vdev_stat_lock);
2853 vd->vdev_stat.vs_space = 0;
2854 vd->vdev_stat.vs_dspace = 0;
2855 vd->vdev_stat.vs_alloc = 0;
2856 mutex_exit(&vd->vdev_stat_lock);
2860 vdev_scan_stat_init(vdev_t *vd)
2862 vdev_stat_t *vs = &vd->vdev_stat;
2864 for (int c = 0; c < vd->vdev_children; c++)
2865 vdev_scan_stat_init(vd->vdev_child[c]);
2867 mutex_enter(&vd->vdev_stat_lock);
2868 vs->vs_scan_processed = 0;
2869 mutex_exit(&vd->vdev_stat_lock);
2873 vdev_stat_update(zio_t *zio, uint64_t psize)
2875 spa_t *spa = zio->io_spa;
2876 vdev_t *rvd = spa->spa_root_vdev;
2877 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2879 uint64_t txg = zio->io_txg;
2880 vdev_stat_t *vs = &vd->vdev_stat;
2881 zio_type_t type = zio->io_type;
2882 int flags = zio->io_flags;
2885 * If this i/o is a gang leader, it didn't do any actual work.
2887 if (zio->io_gang_tree)
2890 if (zio->io_error == 0) {
2892 * If this is a root i/o, don't count it -- we've already
2893 * counted the top-level vdevs, and vdev_get_stats() will
2894 * aggregate them when asked. This reduces contention on
2895 * the root vdev_stat_lock and implicitly handles blocks
2896 * that compress away to holes, for which there is no i/o.
2897 * (Holes never create vdev children, so all the counters
2898 * remain zero, which is what we want.)
2900 * Note: this only applies to successful i/o (io_error == 0)
2901 * because unlike i/o counts, errors are not additive.
2902 * When reading a ditto block, for example, failure of
2903 * one top-level vdev does not imply a root-level error.
2908 ASSERT(vd == zio->io_vd);
2910 if (flags & ZIO_FLAG_IO_BYPASS)
2913 mutex_enter(&vd->vdev_stat_lock);
2915 if (flags & ZIO_FLAG_IO_REPAIR) {
2916 if (flags & ZIO_FLAG_SCAN_THREAD) {
2917 dsl_scan_phys_t *scn_phys =
2918 &spa->spa_dsl_pool->dp_scan->scn_phys;
2919 uint64_t *processed = &scn_phys->scn_processed;
2922 if (vd->vdev_ops->vdev_op_leaf)
2923 atomic_add_64(processed, psize);
2924 vs->vs_scan_processed += psize;
2927 if (flags & ZIO_FLAG_SELF_HEAL)
2928 vs->vs_self_healed += psize;
2932 vs->vs_bytes[type] += psize;
2934 mutex_exit(&vd->vdev_stat_lock);
2938 if (flags & ZIO_FLAG_SPECULATIVE)
2942 * If this is an I/O error that is going to be retried, then ignore the
2943 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2944 * hard errors, when in reality they can happen for any number of
2945 * innocuous reasons (bus resets, MPxIO link failure, etc).
2947 if (zio->io_error == EIO &&
2948 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2952 * Intent logs writes won't propagate their error to the root
2953 * I/O so don't mark these types of failures as pool-level
2956 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2959 mutex_enter(&vd->vdev_stat_lock);
2960 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2961 if (zio->io_error == ECKSUM)
2962 vs->vs_checksum_errors++;
2964 vs->vs_read_errors++;
2966 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2967 vs->vs_write_errors++;
2968 mutex_exit(&vd->vdev_stat_lock);
2970 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2971 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2972 (flags & ZIO_FLAG_SCAN_THREAD) ||
2973 spa->spa_claiming)) {
2975 * This is either a normal write (not a repair), or it's
2976 * a repair induced by the scrub thread, or it's a repair
2977 * made by zil_claim() during spa_load() in the first txg.
2978 * In the normal case, we commit the DTL change in the same
2979 * txg as the block was born. In the scrub-induced repair
2980 * case, we know that scrubs run in first-pass syncing context,
2981 * so we commit the DTL change in spa_syncing_txg(spa).
2982 * In the zil_claim() case, we commit in spa_first_txg(spa).
2984 * We currently do not make DTL entries for failed spontaneous
2985 * self-healing writes triggered by normal (non-scrubbing)
2986 * reads, because we have no transactional context in which to
2987 * do so -- and it's not clear that it'd be desirable anyway.
2989 if (vd->vdev_ops->vdev_op_leaf) {
2990 uint64_t commit_txg = txg;
2991 if (flags & ZIO_FLAG_SCAN_THREAD) {
2992 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2993 ASSERT(spa_sync_pass(spa) == 1);
2994 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2995 commit_txg = spa_syncing_txg(spa);
2996 } else if (spa->spa_claiming) {
2997 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2998 commit_txg = spa_first_txg(spa);
3000 ASSERT(commit_txg >= spa_syncing_txg(spa));
3001 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3003 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3004 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3005 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3008 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3013 * Update the in-core space usage stats for this vdev, its metaslab class,
3014 * and the root vdev.
3017 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3018 int64_t space_delta)
3020 int64_t dspace_delta = space_delta;
3021 spa_t *spa = vd->vdev_spa;
3022 vdev_t *rvd = spa->spa_root_vdev;
3023 metaslab_group_t *mg = vd->vdev_mg;
3024 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3026 ASSERT(vd == vd->vdev_top);
3029 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3030 * factor. We must calculate this here and not at the root vdev
3031 * because the root vdev's psize-to-asize is simply the max of its
3032 * childrens', thus not accurate enough for us.
3034 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3035 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3036 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3037 vd->vdev_deflate_ratio;
3039 mutex_enter(&vd->vdev_stat_lock);
3040 vd->vdev_stat.vs_alloc += alloc_delta;
3041 vd->vdev_stat.vs_space += space_delta;
3042 vd->vdev_stat.vs_dspace += dspace_delta;
3043 mutex_exit(&vd->vdev_stat_lock);
3045 if (mc == spa_normal_class(spa)) {
3046 mutex_enter(&rvd->vdev_stat_lock);
3047 rvd->vdev_stat.vs_alloc += alloc_delta;
3048 rvd->vdev_stat.vs_space += space_delta;
3049 rvd->vdev_stat.vs_dspace += dspace_delta;
3050 mutex_exit(&rvd->vdev_stat_lock);
3054 ASSERT(rvd == vd->vdev_parent);
3055 ASSERT(vd->vdev_ms_count != 0);
3057 metaslab_class_space_update(mc,
3058 alloc_delta, defer_delta, space_delta, dspace_delta);
3063 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3064 * so that it will be written out next time the vdev configuration is synced.
3065 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3068 vdev_config_dirty(vdev_t *vd)
3070 spa_t *spa = vd->vdev_spa;
3071 vdev_t *rvd = spa->spa_root_vdev;
3074 ASSERT(spa_writeable(spa));
3077 * If this is an aux vdev (as with l2cache and spare devices), then we
3078 * update the vdev config manually and set the sync flag.
3080 if (vd->vdev_aux != NULL) {
3081 spa_aux_vdev_t *sav = vd->vdev_aux;
3085 for (c = 0; c < sav->sav_count; c++) {
3086 if (sav->sav_vdevs[c] == vd)
3090 if (c == sav->sav_count) {
3092 * We're being removed. There's nothing more to do.
3094 ASSERT(sav->sav_sync == B_TRUE);
3098 sav->sav_sync = B_TRUE;
3100 if (nvlist_lookup_nvlist_array(sav->sav_config,
3101 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3102 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3103 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3109 * Setting the nvlist in the middle if the array is a little
3110 * sketchy, but it will work.
3112 nvlist_free(aux[c]);
3113 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3119 * The dirty list is protected by the SCL_CONFIG lock. The caller
3120 * must either hold SCL_CONFIG as writer, or must be the sync thread
3121 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3122 * so this is sufficient to ensure mutual exclusion.
3124 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3125 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3126 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3129 for (c = 0; c < rvd->vdev_children; c++)
3130 vdev_config_dirty(rvd->vdev_child[c]);
3132 ASSERT(vd == vd->vdev_top);
3134 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3136 list_insert_head(&spa->spa_config_dirty_list, vd);
3141 vdev_config_clean(vdev_t *vd)
3143 spa_t *spa = vd->vdev_spa;
3145 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3146 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3147 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3149 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3150 list_remove(&spa->spa_config_dirty_list, vd);
3154 * Mark a top-level vdev's state as dirty, so that the next pass of
3155 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3156 * the state changes from larger config changes because they require
3157 * much less locking, and are often needed for administrative actions.
3160 vdev_state_dirty(vdev_t *vd)
3162 spa_t *spa = vd->vdev_spa;
3164 ASSERT(spa_writeable(spa));
3165 ASSERT(vd == vd->vdev_top);
3168 * The state list is protected by the SCL_STATE lock. The caller
3169 * must either hold SCL_STATE as writer, or must be the sync thread
3170 * (which holds SCL_STATE as reader). There's only one sync thread,
3171 * so this is sufficient to ensure mutual exclusion.
3173 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3174 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3175 spa_config_held(spa, SCL_STATE, RW_READER)));
3177 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3178 list_insert_head(&spa->spa_state_dirty_list, vd);
3182 vdev_state_clean(vdev_t *vd)
3184 spa_t *spa = vd->vdev_spa;
3186 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3187 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3188 spa_config_held(spa, SCL_STATE, RW_READER)));
3190 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3191 list_remove(&spa->spa_state_dirty_list, vd);
3195 * Propagate vdev state up from children to parent.
3198 vdev_propagate_state(vdev_t *vd)
3200 spa_t *spa = vd->vdev_spa;
3201 vdev_t *rvd = spa->spa_root_vdev;
3202 int degraded = 0, faulted = 0;
3206 if (vd->vdev_children > 0) {
3207 for (int c = 0; c < vd->vdev_children; c++) {
3208 child = vd->vdev_child[c];
3211 * Don't factor holes into the decision.
3213 if (child->vdev_ishole)
3216 if (!vdev_readable(child) ||
3217 (!vdev_writeable(child) && spa_writeable(spa))) {
3219 * Root special: if there is a top-level log
3220 * device, treat the root vdev as if it were
3223 if (child->vdev_islog && vd == rvd)
3227 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3231 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3235 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3238 * Root special: if there is a top-level vdev that cannot be
3239 * opened due to corrupted metadata, then propagate the root
3240 * vdev's aux state as 'corrupt' rather than 'insufficient
3243 if (corrupted && vd == rvd &&
3244 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3245 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3246 VDEV_AUX_CORRUPT_DATA);
3249 if (vd->vdev_parent)
3250 vdev_propagate_state(vd->vdev_parent);
3254 * Set a vdev's state. If this is during an open, we don't update the parent
3255 * state, because we're in the process of opening children depth-first.
3256 * Otherwise, we propagate the change to the parent.
3258 * If this routine places a device in a faulted state, an appropriate ereport is
3262 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3264 uint64_t save_state;
3265 spa_t *spa = vd->vdev_spa;
3267 if (state == vd->vdev_state) {
3268 vd->vdev_stat.vs_aux = aux;
3272 save_state = vd->vdev_state;
3274 vd->vdev_state = state;
3275 vd->vdev_stat.vs_aux = aux;
3278 * If we are setting the vdev state to anything but an open state, then
3279 * always close the underlying device unless the device has requested
3280 * a delayed close (i.e. we're about to remove or fault the device).
3281 * Otherwise, we keep accessible but invalid devices open forever.
3282 * We don't call vdev_close() itself, because that implies some extra
3283 * checks (offline, etc) that we don't want here. This is limited to
3284 * leaf devices, because otherwise closing the device will affect other
3287 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3288 vd->vdev_ops->vdev_op_leaf)
3289 vd->vdev_ops->vdev_op_close(vd);
3292 * If we have brought this vdev back into service, we need
3293 * to notify fmd so that it can gracefully repair any outstanding
3294 * cases due to a missing device. We do this in all cases, even those
3295 * that probably don't correlate to a repaired fault. This is sure to
3296 * catch all cases, and we let the zfs-retire agent sort it out. If
3297 * this is a transient state it's OK, as the retire agent will
3298 * double-check the state of the vdev before repairing it.
3300 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3301 vd->vdev_prevstate != state)
3302 zfs_post_state_change(spa, vd);
3304 if (vd->vdev_removed &&
3305 state == VDEV_STATE_CANT_OPEN &&
3306 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3308 * If the previous state is set to VDEV_STATE_REMOVED, then this
3309 * device was previously marked removed and someone attempted to
3310 * reopen it. If this failed due to a nonexistent device, then
3311 * keep the device in the REMOVED state. We also let this be if
3312 * it is one of our special test online cases, which is only
3313 * attempting to online the device and shouldn't generate an FMA
3316 vd->vdev_state = VDEV_STATE_REMOVED;
3317 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3318 } else if (state == VDEV_STATE_REMOVED) {
3319 vd->vdev_removed = B_TRUE;
3320 } else if (state == VDEV_STATE_CANT_OPEN) {
3322 * If we fail to open a vdev during an import or recovery, we
3323 * mark it as "not available", which signifies that it was
3324 * never there to begin with. Failure to open such a device
3325 * is not considered an error.
3327 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3328 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3329 vd->vdev_ops->vdev_op_leaf)
3330 vd->vdev_not_present = 1;
3333 * Post the appropriate ereport. If the 'prevstate' field is
3334 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3335 * that this is part of a vdev_reopen(). In this case, we don't
3336 * want to post the ereport if the device was already in the
3337 * CANT_OPEN state beforehand.
3339 * If the 'checkremove' flag is set, then this is an attempt to
3340 * online the device in response to an insertion event. If we
3341 * hit this case, then we have detected an insertion event for a
3342 * faulted or offline device that wasn't in the removed state.
3343 * In this scenario, we don't post an ereport because we are
3344 * about to replace the device, or attempt an online with
3345 * vdev_forcefault, which will generate the fault for us.
3347 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3348 !vd->vdev_not_present && !vd->vdev_checkremove &&
3349 vd != spa->spa_root_vdev) {
3353 case VDEV_AUX_OPEN_FAILED:
3354 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3356 case VDEV_AUX_CORRUPT_DATA:
3357 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3359 case VDEV_AUX_NO_REPLICAS:
3360 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3362 case VDEV_AUX_BAD_GUID_SUM:
3363 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3365 case VDEV_AUX_TOO_SMALL:
3366 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3368 case VDEV_AUX_BAD_LABEL:
3369 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3372 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3375 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3378 /* Erase any notion of persistent removed state */
3379 vd->vdev_removed = B_FALSE;
3381 vd->vdev_removed = B_FALSE;
3384 if (!isopen && vd->vdev_parent)
3385 vdev_propagate_state(vd->vdev_parent);
3389 * Check the vdev configuration to ensure that it's capable of supporting
3392 * On Solaris, we do not support RAID-Z or partial configuration. In
3393 * addition, only a single top-level vdev is allowed and none of the
3394 * leaves can be wholedisks.
3396 * For FreeBSD, we can boot from any configuration. There is a
3397 * limitation that the boot filesystem must be either uncompressed or
3398 * compresses with lzjb compression but I'm not sure how to enforce
3402 vdev_is_bootable(vdev_t *vd)
3405 if (!vd->vdev_ops->vdev_op_leaf) {
3406 char *vdev_type = vd->vdev_ops->vdev_op_type;
3408 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3409 vd->vdev_children > 1) {
3411 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3412 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3417 for (int c = 0; c < vd->vdev_children; c++) {
3418 if (!vdev_is_bootable(vd->vdev_child[c]))
3421 #endif /* illumos */
3426 * Load the state from the original vdev tree (ovd) which
3427 * we've retrieved from the MOS config object. If the original
3428 * vdev was offline or faulted then we transfer that state to the
3429 * device in the current vdev tree (nvd).
3432 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3434 spa_t *spa = nvd->vdev_spa;
3436 ASSERT(nvd->vdev_top->vdev_islog);
3437 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3438 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3440 for (int c = 0; c < nvd->vdev_children; c++)
3441 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3443 if (nvd->vdev_ops->vdev_op_leaf) {
3445 * Restore the persistent vdev state
3447 nvd->vdev_offline = ovd->vdev_offline;
3448 nvd->vdev_faulted = ovd->vdev_faulted;
3449 nvd->vdev_degraded = ovd->vdev_degraded;
3450 nvd->vdev_removed = ovd->vdev_removed;
3455 * Determine if a log device has valid content. If the vdev was
3456 * removed or faulted in the MOS config then we know that
3457 * the content on the log device has already been written to the pool.
3460 vdev_log_state_valid(vdev_t *vd)
3462 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3466 for (int c = 0; c < vd->vdev_children; c++)
3467 if (vdev_log_state_valid(vd->vdev_child[c]))
3474 * Expand a vdev if possible.
3477 vdev_expand(vdev_t *vd, uint64_t txg)
3479 ASSERT(vd->vdev_top == vd);
3480 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3482 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3483 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3484 vdev_config_dirty(vd);
3492 vdev_split(vdev_t *vd)
3494 vdev_t *cvd, *pvd = vd->vdev_parent;
3496 vdev_remove_child(pvd, vd);
3497 vdev_compact_children(pvd);
3499 cvd = pvd->vdev_child[0];
3500 if (pvd->vdev_children == 1) {
3501 vdev_remove_parent(cvd);
3502 cvd->vdev_splitting = B_TRUE;
3504 vdev_propagate_state(cvd);
3508 vdev_deadman(vdev_t *vd)
3510 for (int c = 0; c < vd->vdev_children; c++) {
3511 vdev_t *cvd = vd->vdev_child[c];
3516 if (vd->vdev_ops->vdev_op_leaf) {
3517 vdev_queue_t *vq = &vd->vdev_queue;
3519 mutex_enter(&vq->vq_lock);
3520 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3521 spa_t *spa = vd->vdev_spa;
3526 * Look at the head of all the pending queues,
3527 * if any I/O has been outstanding for longer than
3528 * the spa_deadman_synctime we panic the system.
3530 fio = avl_first(&vq->vq_active_tree);
3531 delta = gethrtime() - fio->io_timestamp;
3532 if (delta > spa_deadman_synctime(spa)) {
3533 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3534 "delta %lluns, last io %lluns",
3535 fio->io_timestamp, delta,
3536 vq->vq_io_complete_ts);
3537 fm_panic("I/O to pool '%s' appears to be "
3538 "hung on vdev guid %llu at '%s'.",
3540 (long long unsigned int) vd->vdev_guid,
3544 mutex_exit(&vq->vq_lock);