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 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011 by Delphix. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa_impl.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
41 #include <sys/fs/zfs.h>
44 #include <sys/dsl_scan.h>
46 SYSCTL_DECL(_vfs_zfs);
47 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
50 * Virtual device management.
53 static vdev_ops_t *vdev_ops_table[] = {
70 /* maximum scrub/resilver I/O queue per leaf vdev */
71 int zfs_scrub_limit = 10;
73 TUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit);
74 SYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0,
75 "Maximum scrub/resilver I/O queue");
78 * Given a vdev type, return the appropriate ops vector.
81 vdev_getops(const char *type)
83 vdev_ops_t *ops, **opspp;
85 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
86 if (strcmp(ops->vdev_op_type, type) == 0)
93 * Default asize function: return the MAX of psize with the asize of
94 * all children. This is what's used by anything other than RAID-Z.
97 vdev_default_asize(vdev_t *vd, uint64_t psize)
99 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
102 for (int c = 0; c < vd->vdev_children; c++) {
103 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
104 asize = MAX(asize, csize);
111 * Get the minimum allocatable size. We define the allocatable size as
112 * the vdev's asize rounded to the nearest metaslab. This allows us to
113 * replace or attach devices which don't have the same physical size but
114 * can still satisfy the same number of allocations.
117 vdev_get_min_asize(vdev_t *vd)
119 vdev_t *pvd = vd->vdev_parent;
122 * The our parent is NULL (inactive spare or cache) or is the root,
123 * just return our own asize.
126 return (vd->vdev_asize);
129 * The top-level vdev just returns the allocatable size rounded
130 * to the nearest metaslab.
132 if (vd == vd->vdev_top)
133 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
136 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
137 * so each child must provide at least 1/Nth of its asize.
139 if (pvd->vdev_ops == &vdev_raidz_ops)
140 return (pvd->vdev_min_asize / pvd->vdev_children);
142 return (pvd->vdev_min_asize);
146 vdev_set_min_asize(vdev_t *vd)
148 vd->vdev_min_asize = vdev_get_min_asize(vd);
150 for (int c = 0; c < vd->vdev_children; c++)
151 vdev_set_min_asize(vd->vdev_child[c]);
155 vdev_lookup_top(spa_t *spa, uint64_t vdev)
157 vdev_t *rvd = spa->spa_root_vdev;
159 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
161 if (vdev < rvd->vdev_children) {
162 ASSERT(rvd->vdev_child[vdev] != NULL);
163 return (rvd->vdev_child[vdev]);
170 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
174 if (vd->vdev_guid == guid)
177 for (int c = 0; c < vd->vdev_children; c++)
178 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
186 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
188 size_t oldsize, newsize;
189 uint64_t id = cvd->vdev_id;
192 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
193 ASSERT(cvd->vdev_parent == NULL);
195 cvd->vdev_parent = pvd;
200 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
202 oldsize = pvd->vdev_children * sizeof (vdev_t *);
203 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
204 newsize = pvd->vdev_children * sizeof (vdev_t *);
206 newchild = kmem_zalloc(newsize, KM_SLEEP);
207 if (pvd->vdev_child != NULL) {
208 bcopy(pvd->vdev_child, newchild, oldsize);
209 kmem_free(pvd->vdev_child, oldsize);
212 pvd->vdev_child = newchild;
213 pvd->vdev_child[id] = cvd;
215 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
216 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
219 * Walk up all ancestors to update guid sum.
221 for (; pvd != NULL; pvd = pvd->vdev_parent)
222 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
226 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
229 uint_t id = cvd->vdev_id;
231 ASSERT(cvd->vdev_parent == pvd);
236 ASSERT(id < pvd->vdev_children);
237 ASSERT(pvd->vdev_child[id] == cvd);
239 pvd->vdev_child[id] = NULL;
240 cvd->vdev_parent = NULL;
242 for (c = 0; c < pvd->vdev_children; c++)
243 if (pvd->vdev_child[c])
246 if (c == pvd->vdev_children) {
247 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
248 pvd->vdev_child = NULL;
249 pvd->vdev_children = 0;
253 * Walk up all ancestors to update guid sum.
255 for (; pvd != NULL; pvd = pvd->vdev_parent)
256 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
260 * Remove any holes in the child array.
263 vdev_compact_children(vdev_t *pvd)
265 vdev_t **newchild, *cvd;
266 int oldc = pvd->vdev_children;
269 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
271 for (int c = newc = 0; c < oldc; c++)
272 if (pvd->vdev_child[c])
275 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
277 for (int c = newc = 0; c < oldc; c++) {
278 if ((cvd = pvd->vdev_child[c]) != NULL) {
279 newchild[newc] = cvd;
280 cvd->vdev_id = newc++;
284 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
285 pvd->vdev_child = newchild;
286 pvd->vdev_children = newc;
290 * Allocate and minimally initialize a vdev_t.
293 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
297 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
299 if (spa->spa_root_vdev == NULL) {
300 ASSERT(ops == &vdev_root_ops);
301 spa->spa_root_vdev = vd;
302 spa->spa_load_guid = spa_generate_guid(NULL);
305 if (guid == 0 && ops != &vdev_hole_ops) {
306 if (spa->spa_root_vdev == vd) {
308 * The root vdev's guid will also be the pool guid,
309 * which must be unique among all pools.
311 guid = spa_generate_guid(NULL);
314 * Any other vdev's guid must be unique within the pool.
316 guid = spa_generate_guid(spa);
318 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
323 vd->vdev_guid = guid;
324 vd->vdev_guid_sum = guid;
326 vd->vdev_state = VDEV_STATE_CLOSED;
327 vd->vdev_ishole = (ops == &vdev_hole_ops);
329 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
330 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
331 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
332 for (int t = 0; t < DTL_TYPES; t++) {
333 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
336 txg_list_create(&vd->vdev_ms_list,
337 offsetof(struct metaslab, ms_txg_node));
338 txg_list_create(&vd->vdev_dtl_list,
339 offsetof(struct vdev, vdev_dtl_node));
340 vd->vdev_stat.vs_timestamp = gethrtime();
348 * Allocate a new vdev. The 'alloctype' is used to control whether we are
349 * creating a new vdev or loading an existing one - the behavior is slightly
350 * different for each case.
353 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
358 uint64_t guid = 0, islog, nparity;
361 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
363 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
366 if ((ops = vdev_getops(type)) == NULL)
370 * If this is a load, get the vdev guid from the nvlist.
371 * Otherwise, vdev_alloc_common() will generate one for us.
373 if (alloctype == VDEV_ALLOC_LOAD) {
376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
380 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
382 } else if (alloctype == VDEV_ALLOC_SPARE) {
383 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
385 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
386 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
388 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
389 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
394 * The first allocated vdev must be of type 'root'.
396 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
400 * Determine whether we're a log vdev.
403 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
404 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
407 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
411 * Set the nparity property for RAID-Z vdevs.
414 if (ops == &vdev_raidz_ops) {
415 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
417 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
420 * Previous versions could only support 1 or 2 parity
424 spa_version(spa) < SPA_VERSION_RAIDZ2)
427 spa_version(spa) < SPA_VERSION_RAIDZ3)
431 * We require the parity to be specified for SPAs that
432 * support multiple parity levels.
434 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
437 * Otherwise, we default to 1 parity device for RAID-Z.
444 ASSERT(nparity != -1ULL);
446 vd = vdev_alloc_common(spa, id, guid, ops);
448 vd->vdev_islog = islog;
449 vd->vdev_nparity = nparity;
451 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
452 vd->vdev_path = spa_strdup(vd->vdev_path);
453 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
454 vd->vdev_devid = spa_strdup(vd->vdev_devid);
455 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
456 &vd->vdev_physpath) == 0)
457 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
458 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
459 vd->vdev_fru = spa_strdup(vd->vdev_fru);
462 * Set the whole_disk property. If it's not specified, leave the value
465 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
466 &vd->vdev_wholedisk) != 0)
467 vd->vdev_wholedisk = -1ULL;
470 * Look for the 'not present' flag. This will only be set if the device
471 * was not present at the time of import.
473 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
474 &vd->vdev_not_present);
477 * Get the alignment requirement.
479 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
482 * Retrieve the vdev creation time.
484 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
488 * If we're a top-level vdev, try to load the allocation parameters.
490 if (parent && !parent->vdev_parent &&
491 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
494 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
496 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
498 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
502 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
503 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
504 alloctype == VDEV_ALLOC_ADD ||
505 alloctype == VDEV_ALLOC_SPLIT ||
506 alloctype == VDEV_ALLOC_ROOTPOOL);
507 vd->vdev_mg = metaslab_group_create(islog ?
508 spa_log_class(spa) : spa_normal_class(spa), vd);
512 * If we're a leaf vdev, try to load the DTL object and other state.
514 if (vd->vdev_ops->vdev_op_leaf &&
515 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
516 alloctype == VDEV_ALLOC_ROOTPOOL)) {
517 if (alloctype == VDEV_ALLOC_LOAD) {
518 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
519 &vd->vdev_dtl_smo.smo_object);
520 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
524 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
527 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
528 &spare) == 0 && spare)
532 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
535 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
536 &vd->vdev_resilvering);
539 * When importing a pool, we want to ignore the persistent fault
540 * state, as the diagnosis made on another system may not be
541 * valid in the current context. Local vdevs will
542 * remain in the faulted state.
544 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
545 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
547 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
549 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
552 if (vd->vdev_faulted || vd->vdev_degraded) {
556 VDEV_AUX_ERR_EXCEEDED;
557 if (nvlist_lookup_string(nv,
558 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
559 strcmp(aux, "external") == 0)
560 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
566 * Add ourselves to the parent's list of children.
568 vdev_add_child(parent, vd);
576 vdev_free(vdev_t *vd)
578 spa_t *spa = vd->vdev_spa;
581 * vdev_free() implies closing the vdev first. This is simpler than
582 * trying to ensure complicated semantics for all callers.
586 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
587 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
592 for (int c = 0; c < vd->vdev_children; c++)
593 vdev_free(vd->vdev_child[c]);
595 ASSERT(vd->vdev_child == NULL);
596 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
599 * Discard allocation state.
601 if (vd->vdev_mg != NULL) {
602 vdev_metaslab_fini(vd);
603 metaslab_group_destroy(vd->vdev_mg);
606 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
607 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
608 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
611 * Remove this vdev from its parent's child list.
613 vdev_remove_child(vd->vdev_parent, vd);
615 ASSERT(vd->vdev_parent == NULL);
618 * Clean up vdev structure.
624 spa_strfree(vd->vdev_path);
626 spa_strfree(vd->vdev_devid);
627 if (vd->vdev_physpath)
628 spa_strfree(vd->vdev_physpath);
630 spa_strfree(vd->vdev_fru);
632 if (vd->vdev_isspare)
633 spa_spare_remove(vd);
634 if (vd->vdev_isl2cache)
635 spa_l2cache_remove(vd);
637 txg_list_destroy(&vd->vdev_ms_list);
638 txg_list_destroy(&vd->vdev_dtl_list);
640 mutex_enter(&vd->vdev_dtl_lock);
641 for (int t = 0; t < DTL_TYPES; t++) {
642 space_map_unload(&vd->vdev_dtl[t]);
643 space_map_destroy(&vd->vdev_dtl[t]);
645 mutex_exit(&vd->vdev_dtl_lock);
647 mutex_destroy(&vd->vdev_dtl_lock);
648 mutex_destroy(&vd->vdev_stat_lock);
649 mutex_destroy(&vd->vdev_probe_lock);
651 if (vd == spa->spa_root_vdev)
652 spa->spa_root_vdev = NULL;
654 kmem_free(vd, sizeof (vdev_t));
658 * Transfer top-level vdev state from svd to tvd.
661 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
663 spa_t *spa = svd->vdev_spa;
668 ASSERT(tvd == tvd->vdev_top);
670 tvd->vdev_ms_array = svd->vdev_ms_array;
671 tvd->vdev_ms_shift = svd->vdev_ms_shift;
672 tvd->vdev_ms_count = svd->vdev_ms_count;
674 svd->vdev_ms_array = 0;
675 svd->vdev_ms_shift = 0;
676 svd->vdev_ms_count = 0;
679 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
680 tvd->vdev_mg = svd->vdev_mg;
681 tvd->vdev_ms = svd->vdev_ms;
686 if (tvd->vdev_mg != NULL)
687 tvd->vdev_mg->mg_vd = tvd;
689 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
690 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
691 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
693 svd->vdev_stat.vs_alloc = 0;
694 svd->vdev_stat.vs_space = 0;
695 svd->vdev_stat.vs_dspace = 0;
697 for (t = 0; t < TXG_SIZE; t++) {
698 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
699 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
700 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
701 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
702 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
703 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
706 if (list_link_active(&svd->vdev_config_dirty_node)) {
707 vdev_config_clean(svd);
708 vdev_config_dirty(tvd);
711 if (list_link_active(&svd->vdev_state_dirty_node)) {
712 vdev_state_clean(svd);
713 vdev_state_dirty(tvd);
716 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
717 svd->vdev_deflate_ratio = 0;
719 tvd->vdev_islog = svd->vdev_islog;
724 vdev_top_update(vdev_t *tvd, vdev_t *vd)
731 for (int c = 0; c < vd->vdev_children; c++)
732 vdev_top_update(tvd, vd->vdev_child[c]);
736 * Add a mirror/replacing vdev above an existing vdev.
739 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
741 spa_t *spa = cvd->vdev_spa;
742 vdev_t *pvd = cvd->vdev_parent;
745 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
747 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
749 mvd->vdev_asize = cvd->vdev_asize;
750 mvd->vdev_min_asize = cvd->vdev_min_asize;
751 mvd->vdev_ashift = cvd->vdev_ashift;
752 mvd->vdev_state = cvd->vdev_state;
753 mvd->vdev_crtxg = cvd->vdev_crtxg;
755 vdev_remove_child(pvd, cvd);
756 vdev_add_child(pvd, mvd);
757 cvd->vdev_id = mvd->vdev_children;
758 vdev_add_child(mvd, cvd);
759 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
761 if (mvd == mvd->vdev_top)
762 vdev_top_transfer(cvd, mvd);
768 * Remove a 1-way mirror/replacing vdev from the tree.
771 vdev_remove_parent(vdev_t *cvd)
773 vdev_t *mvd = cvd->vdev_parent;
774 vdev_t *pvd = mvd->vdev_parent;
776 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
778 ASSERT(mvd->vdev_children == 1);
779 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
780 mvd->vdev_ops == &vdev_replacing_ops ||
781 mvd->vdev_ops == &vdev_spare_ops);
782 cvd->vdev_ashift = mvd->vdev_ashift;
784 vdev_remove_child(mvd, cvd);
785 vdev_remove_child(pvd, mvd);
788 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
789 * Otherwise, we could have detached an offline device, and when we
790 * go to import the pool we'll think we have two top-level vdevs,
791 * instead of a different version of the same top-level vdev.
793 if (mvd->vdev_top == mvd) {
794 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
795 cvd->vdev_orig_guid = cvd->vdev_guid;
796 cvd->vdev_guid += guid_delta;
797 cvd->vdev_guid_sum += guid_delta;
799 cvd->vdev_id = mvd->vdev_id;
800 vdev_add_child(pvd, cvd);
801 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
803 if (cvd == cvd->vdev_top)
804 vdev_top_transfer(mvd, cvd);
806 ASSERT(mvd->vdev_children == 0);
811 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
813 spa_t *spa = vd->vdev_spa;
814 objset_t *mos = spa->spa_meta_objset;
816 uint64_t oldc = vd->vdev_ms_count;
817 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
821 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
824 * This vdev is not being allocated from yet or is a hole.
826 if (vd->vdev_ms_shift == 0)
829 ASSERT(!vd->vdev_ishole);
832 * Compute the raidz-deflation ratio. Note, we hard-code
833 * in 128k (1 << 17) because it is the current "typical" blocksize.
834 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
835 * or we will inconsistently account for existing bp's.
837 vd->vdev_deflate_ratio = (1 << 17) /
838 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
840 ASSERT(oldc <= newc);
842 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
845 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
846 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
850 vd->vdev_ms_count = newc;
852 for (m = oldc; m < newc; m++) {
853 space_map_obj_t smo = { 0, 0, 0 };
856 error = dmu_read(mos, vd->vdev_ms_array,
857 m * sizeof (uint64_t), sizeof (uint64_t), &object,
863 error = dmu_bonus_hold(mos, object, FTAG, &db);
866 ASSERT3U(db->db_size, >=, sizeof (smo));
867 bcopy(db->db_data, &smo, sizeof (smo));
868 ASSERT3U(smo.smo_object, ==, object);
869 dmu_buf_rele(db, FTAG);
872 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
873 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
877 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
880 * If the vdev is being removed we don't activate
881 * the metaslabs since we want to ensure that no new
882 * allocations are performed on this device.
884 if (oldc == 0 && !vd->vdev_removing)
885 metaslab_group_activate(vd->vdev_mg);
888 spa_config_exit(spa, SCL_ALLOC, FTAG);
894 vdev_metaslab_fini(vdev_t *vd)
897 uint64_t count = vd->vdev_ms_count;
899 if (vd->vdev_ms != NULL) {
900 metaslab_group_passivate(vd->vdev_mg);
901 for (m = 0; m < count; m++)
902 if (vd->vdev_ms[m] != NULL)
903 metaslab_fini(vd->vdev_ms[m]);
904 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
909 typedef struct vdev_probe_stats {
910 boolean_t vps_readable;
911 boolean_t vps_writeable;
913 } vdev_probe_stats_t;
916 vdev_probe_done(zio_t *zio)
918 spa_t *spa = zio->io_spa;
919 vdev_t *vd = zio->io_vd;
920 vdev_probe_stats_t *vps = zio->io_private;
922 ASSERT(vd->vdev_probe_zio != NULL);
924 if (zio->io_type == ZIO_TYPE_READ) {
925 if (zio->io_error == 0)
926 vps->vps_readable = 1;
927 if (zio->io_error == 0 && spa_writeable(spa)) {
928 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
929 zio->io_offset, zio->io_size, zio->io_data,
930 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
931 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
933 zio_buf_free(zio->io_data, zio->io_size);
935 } else if (zio->io_type == ZIO_TYPE_WRITE) {
936 if (zio->io_error == 0)
937 vps->vps_writeable = 1;
938 zio_buf_free(zio->io_data, zio->io_size);
939 } else if (zio->io_type == ZIO_TYPE_NULL) {
942 vd->vdev_cant_read |= !vps->vps_readable;
943 vd->vdev_cant_write |= !vps->vps_writeable;
945 if (vdev_readable(vd) &&
946 (vdev_writeable(vd) || !spa_writeable(spa))) {
949 ASSERT(zio->io_error != 0);
950 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
951 spa, vd, NULL, 0, 0);
952 zio->io_error = ENXIO;
955 mutex_enter(&vd->vdev_probe_lock);
956 ASSERT(vd->vdev_probe_zio == zio);
957 vd->vdev_probe_zio = NULL;
958 mutex_exit(&vd->vdev_probe_lock);
960 while ((pio = zio_walk_parents(zio)) != NULL)
961 if (!vdev_accessible(vd, pio))
962 pio->io_error = ENXIO;
964 kmem_free(vps, sizeof (*vps));
969 * Determine whether this device is accessible by reading and writing
970 * to several known locations: the pad regions of each vdev label
971 * but the first (which we leave alone in case it contains a VTOC).
974 vdev_probe(vdev_t *vd, zio_t *zio)
976 spa_t *spa = vd->vdev_spa;
977 vdev_probe_stats_t *vps = NULL;
980 ASSERT(vd->vdev_ops->vdev_op_leaf);
983 * Don't probe the probe.
985 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
989 * To prevent 'probe storms' when a device fails, we create
990 * just one probe i/o at a time. All zios that want to probe
991 * this vdev will become parents of the probe io.
993 mutex_enter(&vd->vdev_probe_lock);
995 if ((pio = vd->vdev_probe_zio) == NULL) {
996 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
998 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
999 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1002 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1004 * vdev_cant_read and vdev_cant_write can only
1005 * transition from TRUE to FALSE when we have the
1006 * SCL_ZIO lock as writer; otherwise they can only
1007 * transition from FALSE to TRUE. This ensures that
1008 * any zio looking at these values can assume that
1009 * failures persist for the life of the I/O. That's
1010 * important because when a device has intermittent
1011 * connectivity problems, we want to ensure that
1012 * they're ascribed to the device (ENXIO) and not
1015 * Since we hold SCL_ZIO as writer here, clear both
1016 * values so the probe can reevaluate from first
1019 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1020 vd->vdev_cant_read = B_FALSE;
1021 vd->vdev_cant_write = B_FALSE;
1024 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1025 vdev_probe_done, vps,
1026 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1029 * We can't change the vdev state in this context, so we
1030 * kick off an async task to do it on our behalf.
1033 vd->vdev_probe_wanted = B_TRUE;
1034 spa_async_request(spa, SPA_ASYNC_PROBE);
1039 zio_add_child(zio, pio);
1041 mutex_exit(&vd->vdev_probe_lock);
1044 ASSERT(zio != NULL);
1048 for (int l = 1; l < VDEV_LABELS; l++) {
1049 zio_nowait(zio_read_phys(pio, vd,
1050 vdev_label_offset(vd->vdev_psize, l,
1051 offsetof(vdev_label_t, vl_pad2)),
1052 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1053 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1054 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1065 vdev_open_child(void *arg)
1069 vd->vdev_open_thread = curthread;
1070 vd->vdev_open_error = vdev_open(vd);
1071 vd->vdev_open_thread = NULL;
1075 vdev_uses_zvols(vdev_t *vd)
1077 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1078 strlen(ZVOL_DIR)) == 0)
1080 for (int c = 0; c < vd->vdev_children; c++)
1081 if (vdev_uses_zvols(vd->vdev_child[c]))
1087 vdev_open_children(vdev_t *vd)
1090 int children = vd->vdev_children;
1093 * in order to handle pools on top of zvols, do the opens
1094 * in a single thread so that the same thread holds the
1095 * spa_namespace_lock
1097 if (B_TRUE || vdev_uses_zvols(vd)) {
1098 for (int c = 0; c < children; c++)
1099 vd->vdev_child[c]->vdev_open_error =
1100 vdev_open(vd->vdev_child[c]);
1103 tq = taskq_create("vdev_open", children, minclsyspri,
1104 children, children, TASKQ_PREPOPULATE);
1106 for (int c = 0; c < children; c++)
1107 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1114 * Prepare a virtual device for access.
1117 vdev_open(vdev_t *vd)
1119 spa_t *spa = vd->vdev_spa;
1122 uint64_t asize, psize;
1123 uint64_t ashift = 0;
1125 ASSERT(vd->vdev_open_thread == curthread ||
1126 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1127 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1128 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1129 vd->vdev_state == VDEV_STATE_OFFLINE);
1131 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1132 vd->vdev_cant_read = B_FALSE;
1133 vd->vdev_cant_write = B_FALSE;
1134 vd->vdev_min_asize = vdev_get_min_asize(vd);
1137 * If this vdev is not removed, check its fault status. If it's
1138 * faulted, bail out of the open.
1140 if (!vd->vdev_removed && vd->vdev_faulted) {
1141 ASSERT(vd->vdev_children == 0);
1142 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1143 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1144 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1145 vd->vdev_label_aux);
1147 } else if (vd->vdev_offline) {
1148 ASSERT(vd->vdev_children == 0);
1149 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1153 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1156 * Reset the vdev_reopening flag so that we actually close
1157 * the vdev on error.
1159 vd->vdev_reopening = B_FALSE;
1160 if (zio_injection_enabled && error == 0)
1161 error = zio_handle_device_injection(vd, NULL, ENXIO);
1164 if (vd->vdev_removed &&
1165 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1166 vd->vdev_removed = B_FALSE;
1168 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1169 vd->vdev_stat.vs_aux);
1173 vd->vdev_removed = B_FALSE;
1176 * Recheck the faulted flag now that we have confirmed that
1177 * the vdev is accessible. If we're faulted, bail.
1179 if (vd->vdev_faulted) {
1180 ASSERT(vd->vdev_children == 0);
1181 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1182 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1183 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1184 vd->vdev_label_aux);
1188 if (vd->vdev_degraded) {
1189 ASSERT(vd->vdev_children == 0);
1190 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1191 VDEV_AUX_ERR_EXCEEDED);
1193 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1197 * For hole or missing vdevs we just return success.
1199 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1202 for (int c = 0; c < vd->vdev_children; c++) {
1203 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1204 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1210 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1212 if (vd->vdev_children == 0) {
1213 if (osize < SPA_MINDEVSIZE) {
1214 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1215 VDEV_AUX_TOO_SMALL);
1219 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1221 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1222 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1223 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1224 VDEV_AUX_TOO_SMALL);
1231 vd->vdev_psize = psize;
1234 * Make sure the allocatable size hasn't shrunk.
1236 if (asize < vd->vdev_min_asize) {
1237 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1238 VDEV_AUX_BAD_LABEL);
1242 if (vd->vdev_asize == 0) {
1244 * This is the first-ever open, so use the computed values.
1245 * For testing purposes, a higher ashift can be requested.
1247 vd->vdev_asize = asize;
1248 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1251 * Make sure the alignment requirement hasn't increased.
1253 if (ashift > vd->vdev_top->vdev_ashift) {
1254 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1255 VDEV_AUX_BAD_LABEL);
1261 * If all children are healthy and the asize has increased,
1262 * then we've experienced dynamic LUN growth. If automatic
1263 * expansion is enabled then use the additional space.
1265 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1266 (vd->vdev_expanding || spa->spa_autoexpand))
1267 vd->vdev_asize = asize;
1269 vdev_set_min_asize(vd);
1272 * Ensure we can issue some IO before declaring the
1273 * vdev open for business.
1275 if (vd->vdev_ops->vdev_op_leaf &&
1276 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1277 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1278 VDEV_AUX_ERR_EXCEEDED);
1283 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1284 * resilver. But don't do this if we are doing a reopen for a scrub,
1285 * since this would just restart the scrub we are already doing.
1287 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1288 vdev_resilver_needed(vd, NULL, NULL))
1289 spa_async_request(spa, SPA_ASYNC_RESILVER);
1295 * Called once the vdevs are all opened, this routine validates the label
1296 * contents. This needs to be done before vdev_load() so that we don't
1297 * inadvertently do repair I/Os to the wrong device.
1299 * If 'strict' is false ignore the spa guid check. This is necessary because
1300 * if the machine crashed during a re-guid the new guid might have been written
1301 * to all of the vdev labels, but not the cached config. The strict check
1302 * will be performed when the pool is opened again using the mos config.
1304 * This function will only return failure if one of the vdevs indicates that it
1305 * has since been destroyed or exported. This is only possible if
1306 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1307 * will be updated but the function will return 0.
1310 vdev_validate(vdev_t *vd, boolean_t strict)
1312 spa_t *spa = vd->vdev_spa;
1314 uint64_t guid = 0, top_guid;
1317 for (int c = 0; c < vd->vdev_children; c++)
1318 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1322 * If the device has already failed, or was marked offline, don't do
1323 * any further validation. Otherwise, label I/O will fail and we will
1324 * overwrite the previous state.
1326 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1327 uint64_t aux_guid = 0;
1330 if ((label = vdev_label_read_config(vd)) == NULL) {
1331 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1332 VDEV_AUX_BAD_LABEL);
1337 * Determine if this vdev has been split off into another
1338 * pool. If so, then refuse to open it.
1340 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1341 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1342 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1343 VDEV_AUX_SPLIT_POOL);
1348 if (strict && (nvlist_lookup_uint64(label,
1349 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1350 guid != spa_guid(spa))) {
1351 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1352 VDEV_AUX_CORRUPT_DATA);
1357 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1358 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1363 * If this vdev just became a top-level vdev because its
1364 * sibling was detached, it will have adopted the parent's
1365 * vdev guid -- but the label may or may not be on disk yet.
1366 * Fortunately, either version of the label will have the
1367 * same top guid, so if we're a top-level vdev, we can
1368 * safely compare to that instead.
1370 * If we split this vdev off instead, then we also check the
1371 * original pool's guid. We don't want to consider the vdev
1372 * corrupt if it is partway through a split operation.
1374 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1376 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1378 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1379 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1380 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1381 VDEV_AUX_CORRUPT_DATA);
1386 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1388 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1389 VDEV_AUX_CORRUPT_DATA);
1397 * If this is a verbatim import, no need to check the
1398 * state of the pool.
1400 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1401 spa_load_state(spa) == SPA_LOAD_OPEN &&
1402 state != POOL_STATE_ACTIVE)
1406 * If we were able to open and validate a vdev that was
1407 * previously marked permanently unavailable, clear that state
1410 if (vd->vdev_not_present)
1411 vd->vdev_not_present = 0;
1418 * Close a virtual device.
1421 vdev_close(vdev_t *vd)
1423 spa_t *spa = vd->vdev_spa;
1424 vdev_t *pvd = vd->vdev_parent;
1426 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1429 * If our parent is reopening, then we are as well, unless we are
1432 if (pvd != NULL && pvd->vdev_reopening)
1433 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1435 vd->vdev_ops->vdev_op_close(vd);
1437 vdev_cache_purge(vd);
1440 * We record the previous state before we close it, so that if we are
1441 * doing a reopen(), we don't generate FMA ereports if we notice that
1442 * it's still faulted.
1444 vd->vdev_prevstate = vd->vdev_state;
1446 if (vd->vdev_offline)
1447 vd->vdev_state = VDEV_STATE_OFFLINE;
1449 vd->vdev_state = VDEV_STATE_CLOSED;
1450 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1454 vdev_hold(vdev_t *vd)
1456 spa_t *spa = vd->vdev_spa;
1458 ASSERT(spa_is_root(spa));
1459 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1462 for (int c = 0; c < vd->vdev_children; c++)
1463 vdev_hold(vd->vdev_child[c]);
1465 if (vd->vdev_ops->vdev_op_leaf)
1466 vd->vdev_ops->vdev_op_hold(vd);
1470 vdev_rele(vdev_t *vd)
1472 spa_t *spa = vd->vdev_spa;
1474 ASSERT(spa_is_root(spa));
1475 for (int c = 0; c < vd->vdev_children; c++)
1476 vdev_rele(vd->vdev_child[c]);
1478 if (vd->vdev_ops->vdev_op_leaf)
1479 vd->vdev_ops->vdev_op_rele(vd);
1483 * Reopen all interior vdevs and any unopened leaves. We don't actually
1484 * reopen leaf vdevs which had previously been opened as they might deadlock
1485 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1486 * If the leaf has never been opened then open it, as usual.
1489 vdev_reopen(vdev_t *vd)
1491 spa_t *spa = vd->vdev_spa;
1493 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1495 /* set the reopening flag unless we're taking the vdev offline */
1496 vd->vdev_reopening = !vd->vdev_offline;
1498 (void) vdev_open(vd);
1501 * Call vdev_validate() here to make sure we have the same device.
1502 * Otherwise, a device with an invalid label could be successfully
1503 * opened in response to vdev_reopen().
1506 (void) vdev_validate_aux(vd);
1507 if (vdev_readable(vd) && vdev_writeable(vd) &&
1508 vd->vdev_aux == &spa->spa_l2cache &&
1509 !l2arc_vdev_present(vd))
1510 l2arc_add_vdev(spa, vd);
1512 (void) vdev_validate(vd, B_TRUE);
1516 * Reassess parent vdev's health.
1518 vdev_propagate_state(vd);
1522 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1527 * Normally, partial opens (e.g. of a mirror) are allowed.
1528 * For a create, however, we want to fail the request if
1529 * there are any components we can't open.
1531 error = vdev_open(vd);
1533 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1535 return (error ? error : ENXIO);
1539 * Recursively initialize all labels.
1541 if ((error = vdev_label_init(vd, txg, isreplacing ?
1542 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1551 vdev_metaslab_set_size(vdev_t *vd)
1554 * Aim for roughly 200 metaslabs per vdev.
1556 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1557 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1561 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1563 ASSERT(vd == vd->vdev_top);
1564 ASSERT(!vd->vdev_ishole);
1565 ASSERT(ISP2(flags));
1566 ASSERT(spa_writeable(vd->vdev_spa));
1568 if (flags & VDD_METASLAB)
1569 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1571 if (flags & VDD_DTL)
1572 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1574 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1580 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1581 * the vdev has less than perfect replication. There are four kinds of DTL:
1583 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1585 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1587 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1588 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1589 * txgs that was scrubbed.
1591 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1592 * persistent errors or just some device being offline.
1593 * Unlike the other three, the DTL_OUTAGE map is not generally
1594 * maintained; it's only computed when needed, typically to
1595 * determine whether a device can be detached.
1597 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1598 * either has the data or it doesn't.
1600 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1601 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1602 * if any child is less than fully replicated, then so is its parent.
1603 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1604 * comprising only those txgs which appear in 'maxfaults' or more children;
1605 * those are the txgs we don't have enough replication to read. For example,
1606 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1607 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1608 * two child DTL_MISSING maps.
1610 * It should be clear from the above that to compute the DTLs and outage maps
1611 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1612 * Therefore, that is all we keep on disk. When loading the pool, or after
1613 * a configuration change, we generate all other DTLs from first principles.
1616 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1618 space_map_t *sm = &vd->vdev_dtl[t];
1620 ASSERT(t < DTL_TYPES);
1621 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1622 ASSERT(spa_writeable(vd->vdev_spa));
1624 mutex_enter(sm->sm_lock);
1625 if (!space_map_contains(sm, txg, size))
1626 space_map_add(sm, txg, size);
1627 mutex_exit(sm->sm_lock);
1631 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1633 space_map_t *sm = &vd->vdev_dtl[t];
1634 boolean_t dirty = B_FALSE;
1636 ASSERT(t < DTL_TYPES);
1637 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1639 mutex_enter(sm->sm_lock);
1640 if (sm->sm_space != 0)
1641 dirty = space_map_contains(sm, txg, size);
1642 mutex_exit(sm->sm_lock);
1648 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1650 space_map_t *sm = &vd->vdev_dtl[t];
1653 mutex_enter(sm->sm_lock);
1654 empty = (sm->sm_space == 0);
1655 mutex_exit(sm->sm_lock);
1661 * Reassess DTLs after a config change or scrub completion.
1664 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1666 spa_t *spa = vd->vdev_spa;
1670 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1672 for (int c = 0; c < vd->vdev_children; c++)
1673 vdev_dtl_reassess(vd->vdev_child[c], txg,
1674 scrub_txg, scrub_done);
1676 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1679 if (vd->vdev_ops->vdev_op_leaf) {
1680 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1682 mutex_enter(&vd->vdev_dtl_lock);
1683 if (scrub_txg != 0 &&
1684 (spa->spa_scrub_started ||
1685 (scn && scn->scn_phys.scn_errors == 0))) {
1687 * We completed a scrub up to scrub_txg. If we
1688 * did it without rebooting, then the scrub dtl
1689 * will be valid, so excise the old region and
1690 * fold in the scrub dtl. Otherwise, leave the
1691 * dtl as-is if there was an error.
1693 * There's little trick here: to excise the beginning
1694 * of the DTL_MISSING map, we put it into a reference
1695 * tree and then add a segment with refcnt -1 that
1696 * covers the range [0, scrub_txg). This means
1697 * that each txg in that range has refcnt -1 or 0.
1698 * We then add DTL_SCRUB with a refcnt of 2, so that
1699 * entries in the range [0, scrub_txg) will have a
1700 * positive refcnt -- either 1 or 2. We then convert
1701 * the reference tree into the new DTL_MISSING map.
1703 space_map_ref_create(&reftree);
1704 space_map_ref_add_map(&reftree,
1705 &vd->vdev_dtl[DTL_MISSING], 1);
1706 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1707 space_map_ref_add_map(&reftree,
1708 &vd->vdev_dtl[DTL_SCRUB], 2);
1709 space_map_ref_generate_map(&reftree,
1710 &vd->vdev_dtl[DTL_MISSING], 1);
1711 space_map_ref_destroy(&reftree);
1713 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1714 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1715 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1717 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1718 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1719 if (!vdev_readable(vd))
1720 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1722 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1723 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1724 mutex_exit(&vd->vdev_dtl_lock);
1727 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1731 mutex_enter(&vd->vdev_dtl_lock);
1732 for (int t = 0; t < DTL_TYPES; t++) {
1733 /* account for child's outage in parent's missing map */
1734 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1736 continue; /* leaf vdevs only */
1737 if (t == DTL_PARTIAL)
1738 minref = 1; /* i.e. non-zero */
1739 else if (vd->vdev_nparity != 0)
1740 minref = vd->vdev_nparity + 1; /* RAID-Z */
1742 minref = vd->vdev_children; /* any kind of mirror */
1743 space_map_ref_create(&reftree);
1744 for (int c = 0; c < vd->vdev_children; c++) {
1745 vdev_t *cvd = vd->vdev_child[c];
1746 mutex_enter(&cvd->vdev_dtl_lock);
1747 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1748 mutex_exit(&cvd->vdev_dtl_lock);
1750 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1751 space_map_ref_destroy(&reftree);
1753 mutex_exit(&vd->vdev_dtl_lock);
1757 vdev_dtl_load(vdev_t *vd)
1759 spa_t *spa = vd->vdev_spa;
1760 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1761 objset_t *mos = spa->spa_meta_objset;
1765 ASSERT(vd->vdev_children == 0);
1767 if (smo->smo_object == 0)
1770 ASSERT(!vd->vdev_ishole);
1772 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1775 ASSERT3U(db->db_size, >=, sizeof (*smo));
1776 bcopy(db->db_data, smo, sizeof (*smo));
1777 dmu_buf_rele(db, FTAG);
1779 mutex_enter(&vd->vdev_dtl_lock);
1780 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1781 NULL, SM_ALLOC, smo, mos);
1782 mutex_exit(&vd->vdev_dtl_lock);
1788 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1790 spa_t *spa = vd->vdev_spa;
1791 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1792 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1793 objset_t *mos = spa->spa_meta_objset;
1799 ASSERT(!vd->vdev_ishole);
1801 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1803 if (vd->vdev_detached) {
1804 if (smo->smo_object != 0) {
1805 int err = dmu_object_free(mos, smo->smo_object, tx);
1806 ASSERT3U(err, ==, 0);
1807 smo->smo_object = 0;
1813 if (smo->smo_object == 0) {
1814 ASSERT(smo->smo_objsize == 0);
1815 ASSERT(smo->smo_alloc == 0);
1816 smo->smo_object = dmu_object_alloc(mos,
1817 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1818 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1819 ASSERT(smo->smo_object != 0);
1820 vdev_config_dirty(vd->vdev_top);
1823 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1825 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1828 mutex_enter(&smlock);
1830 mutex_enter(&vd->vdev_dtl_lock);
1831 space_map_walk(sm, space_map_add, &smsync);
1832 mutex_exit(&vd->vdev_dtl_lock);
1834 space_map_truncate(smo, mos, tx);
1835 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1837 space_map_destroy(&smsync);
1839 mutex_exit(&smlock);
1840 mutex_destroy(&smlock);
1842 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1843 dmu_buf_will_dirty(db, tx);
1844 ASSERT3U(db->db_size, >=, sizeof (*smo));
1845 bcopy(smo, db->db_data, sizeof (*smo));
1846 dmu_buf_rele(db, FTAG);
1852 * Determine whether the specified vdev can be offlined/detached/removed
1853 * without losing data.
1856 vdev_dtl_required(vdev_t *vd)
1858 spa_t *spa = vd->vdev_spa;
1859 vdev_t *tvd = vd->vdev_top;
1860 uint8_t cant_read = vd->vdev_cant_read;
1863 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1865 if (vd == spa->spa_root_vdev || vd == tvd)
1869 * Temporarily mark the device as unreadable, and then determine
1870 * whether this results in any DTL outages in the top-level vdev.
1871 * If not, we can safely offline/detach/remove the device.
1873 vd->vdev_cant_read = B_TRUE;
1874 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1875 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1876 vd->vdev_cant_read = cant_read;
1877 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1879 if (!required && zio_injection_enabled)
1880 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1886 * Determine if resilver is needed, and if so the txg range.
1889 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1891 boolean_t needed = B_FALSE;
1892 uint64_t thismin = UINT64_MAX;
1893 uint64_t thismax = 0;
1895 if (vd->vdev_children == 0) {
1896 mutex_enter(&vd->vdev_dtl_lock);
1897 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1898 vdev_writeable(vd)) {
1901 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1902 thismin = ss->ss_start - 1;
1903 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1904 thismax = ss->ss_end;
1907 mutex_exit(&vd->vdev_dtl_lock);
1909 for (int c = 0; c < vd->vdev_children; c++) {
1910 vdev_t *cvd = vd->vdev_child[c];
1911 uint64_t cmin, cmax;
1913 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1914 thismin = MIN(thismin, cmin);
1915 thismax = MAX(thismax, cmax);
1921 if (needed && minp) {
1929 vdev_load(vdev_t *vd)
1932 * Recursively load all children.
1934 for (int c = 0; c < vd->vdev_children; c++)
1935 vdev_load(vd->vdev_child[c]);
1938 * If this is a top-level vdev, initialize its metaslabs.
1940 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1941 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1942 vdev_metaslab_init(vd, 0) != 0))
1943 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1944 VDEV_AUX_CORRUPT_DATA);
1947 * If this is a leaf vdev, load its DTL.
1949 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1950 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1951 VDEV_AUX_CORRUPT_DATA);
1955 * The special vdev case is used for hot spares and l2cache devices. Its
1956 * sole purpose it to set the vdev state for the associated vdev. To do this,
1957 * we make sure that we can open the underlying device, then try to read the
1958 * label, and make sure that the label is sane and that it hasn't been
1959 * repurposed to another pool.
1962 vdev_validate_aux(vdev_t *vd)
1965 uint64_t guid, version;
1968 if (!vdev_readable(vd))
1971 if ((label = vdev_label_read_config(vd)) == NULL) {
1972 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1973 VDEV_AUX_CORRUPT_DATA);
1977 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1978 version > SPA_VERSION ||
1979 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1980 guid != vd->vdev_guid ||
1981 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1982 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1983 VDEV_AUX_CORRUPT_DATA);
1989 * We don't actually check the pool state here. If it's in fact in
1990 * use by another pool, we update this fact on the fly when requested.
1997 vdev_remove(vdev_t *vd, uint64_t txg)
1999 spa_t *spa = vd->vdev_spa;
2000 objset_t *mos = spa->spa_meta_objset;
2003 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2005 if (vd->vdev_dtl_smo.smo_object) {
2006 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
2007 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2008 vd->vdev_dtl_smo.smo_object = 0;
2011 if (vd->vdev_ms != NULL) {
2012 for (int m = 0; m < vd->vdev_ms_count; m++) {
2013 metaslab_t *msp = vd->vdev_ms[m];
2015 if (msp == NULL || msp->ms_smo.smo_object == 0)
2018 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2019 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2020 msp->ms_smo.smo_object = 0;
2024 if (vd->vdev_ms_array) {
2025 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2026 vd->vdev_ms_array = 0;
2027 vd->vdev_ms_shift = 0;
2033 vdev_sync_done(vdev_t *vd, uint64_t txg)
2036 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2038 ASSERT(!vd->vdev_ishole);
2040 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2041 metaslab_sync_done(msp, txg);
2044 metaslab_sync_reassess(vd->vdev_mg);
2048 vdev_sync(vdev_t *vd, uint64_t txg)
2050 spa_t *spa = vd->vdev_spa;
2055 ASSERT(!vd->vdev_ishole);
2057 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2058 ASSERT(vd == vd->vdev_top);
2059 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2060 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2061 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2062 ASSERT(vd->vdev_ms_array != 0);
2063 vdev_config_dirty(vd);
2068 * Remove the metadata associated with this vdev once it's empty.
2070 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2071 vdev_remove(vd, txg);
2073 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2074 metaslab_sync(msp, txg);
2075 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2078 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2079 vdev_dtl_sync(lvd, txg);
2081 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2085 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2087 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2091 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2092 * not be opened, and no I/O is attempted.
2095 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2099 spa_vdev_state_enter(spa, SCL_NONE);
2101 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2102 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2104 if (!vd->vdev_ops->vdev_op_leaf)
2105 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2110 * We don't directly use the aux state here, but if we do a
2111 * vdev_reopen(), we need this value to be present to remember why we
2114 vd->vdev_label_aux = aux;
2117 * Faulted state takes precedence over degraded.
2119 vd->vdev_delayed_close = B_FALSE;
2120 vd->vdev_faulted = 1ULL;
2121 vd->vdev_degraded = 0ULL;
2122 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2125 * If this device has the only valid copy of the data, then
2126 * back off and simply mark the vdev as degraded instead.
2128 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2129 vd->vdev_degraded = 1ULL;
2130 vd->vdev_faulted = 0ULL;
2133 * If we reopen the device and it's not dead, only then do we
2138 if (vdev_readable(vd))
2139 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2142 return (spa_vdev_state_exit(spa, vd, 0));
2146 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2147 * user that something is wrong. The vdev continues to operate as normal as far
2148 * as I/O is concerned.
2151 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2155 spa_vdev_state_enter(spa, SCL_NONE);
2157 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2158 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2160 if (!vd->vdev_ops->vdev_op_leaf)
2161 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2164 * If the vdev is already faulted, then don't do anything.
2166 if (vd->vdev_faulted || vd->vdev_degraded)
2167 return (spa_vdev_state_exit(spa, NULL, 0));
2169 vd->vdev_degraded = 1ULL;
2170 if (!vdev_is_dead(vd))
2171 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2174 return (spa_vdev_state_exit(spa, vd, 0));
2178 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2179 * any attached spare device should be detached when the device finishes
2180 * resilvering. Second, the online should be treated like a 'test' online case,
2181 * so no FMA events are generated if the device fails to open.
2184 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2186 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2188 spa_vdev_state_enter(spa, SCL_NONE);
2190 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2191 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2193 if (!vd->vdev_ops->vdev_op_leaf)
2194 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2197 vd->vdev_offline = B_FALSE;
2198 vd->vdev_tmpoffline = B_FALSE;
2199 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2200 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2202 /* XXX - L2ARC 1.0 does not support expansion */
2203 if (!vd->vdev_aux) {
2204 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2205 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2209 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2211 if (!vd->vdev_aux) {
2212 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2213 pvd->vdev_expanding = B_FALSE;
2217 *newstate = vd->vdev_state;
2218 if ((flags & ZFS_ONLINE_UNSPARE) &&
2219 !vdev_is_dead(vd) && vd->vdev_parent &&
2220 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2221 vd->vdev_parent->vdev_child[0] == vd)
2222 vd->vdev_unspare = B_TRUE;
2224 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2226 /* XXX - L2ARC 1.0 does not support expansion */
2228 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2229 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2231 return (spa_vdev_state_exit(spa, vd, 0));
2235 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2239 uint64_t generation;
2240 metaslab_group_t *mg;
2243 spa_vdev_state_enter(spa, SCL_ALLOC);
2245 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2246 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2248 if (!vd->vdev_ops->vdev_op_leaf)
2249 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2253 generation = spa->spa_config_generation + 1;
2256 * If the device isn't already offline, try to offline it.
2258 if (!vd->vdev_offline) {
2260 * If this device has the only valid copy of some data,
2261 * don't allow it to be offlined. Log devices are always
2264 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2265 vdev_dtl_required(vd))
2266 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2269 * If the top-level is a slog and it has had allocations
2270 * then proceed. We check that the vdev's metaslab group
2271 * is not NULL since it's possible that we may have just
2272 * added this vdev but not yet initialized its metaslabs.
2274 if (tvd->vdev_islog && mg != NULL) {
2276 * Prevent any future allocations.
2278 metaslab_group_passivate(mg);
2279 (void) spa_vdev_state_exit(spa, vd, 0);
2281 error = spa_offline_log(spa);
2283 spa_vdev_state_enter(spa, SCL_ALLOC);
2286 * Check to see if the config has changed.
2288 if (error || generation != spa->spa_config_generation) {
2289 metaslab_group_activate(mg);
2291 return (spa_vdev_state_exit(spa,
2293 (void) spa_vdev_state_exit(spa, vd, 0);
2296 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2300 * Offline this device and reopen its top-level vdev.
2301 * If the top-level vdev is a log device then just offline
2302 * it. Otherwise, if this action results in the top-level
2303 * vdev becoming unusable, undo it and fail the request.
2305 vd->vdev_offline = B_TRUE;
2308 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2309 vdev_is_dead(tvd)) {
2310 vd->vdev_offline = B_FALSE;
2312 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2316 * Add the device back into the metaslab rotor so that
2317 * once we online the device it's open for business.
2319 if (tvd->vdev_islog && mg != NULL)
2320 metaslab_group_activate(mg);
2323 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2325 return (spa_vdev_state_exit(spa, vd, 0));
2329 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2333 mutex_enter(&spa->spa_vdev_top_lock);
2334 error = vdev_offline_locked(spa, guid, flags);
2335 mutex_exit(&spa->spa_vdev_top_lock);
2341 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2342 * vdev_offline(), we assume the spa config is locked. We also clear all
2343 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2346 vdev_clear(spa_t *spa, vdev_t *vd)
2348 vdev_t *rvd = spa->spa_root_vdev;
2350 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2355 vd->vdev_stat.vs_read_errors = 0;
2356 vd->vdev_stat.vs_write_errors = 0;
2357 vd->vdev_stat.vs_checksum_errors = 0;
2359 for (int c = 0; c < vd->vdev_children; c++)
2360 vdev_clear(spa, vd->vdev_child[c]);
2363 * If we're in the FAULTED state or have experienced failed I/O, then
2364 * clear the persistent state and attempt to reopen the device. We
2365 * also mark the vdev config dirty, so that the new faulted state is
2366 * written out to disk.
2368 if (vd->vdev_faulted || vd->vdev_degraded ||
2369 !vdev_readable(vd) || !vdev_writeable(vd)) {
2372 * When reopening in reponse to a clear event, it may be due to
2373 * a fmadm repair request. In this case, if the device is
2374 * still broken, we want to still post the ereport again.
2376 vd->vdev_forcefault = B_TRUE;
2378 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2379 vd->vdev_cant_read = B_FALSE;
2380 vd->vdev_cant_write = B_FALSE;
2382 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2384 vd->vdev_forcefault = B_FALSE;
2386 if (vd != rvd && vdev_writeable(vd->vdev_top))
2387 vdev_state_dirty(vd->vdev_top);
2389 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2390 spa_async_request(spa, SPA_ASYNC_RESILVER);
2392 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2396 * When clearing a FMA-diagnosed fault, we always want to
2397 * unspare the device, as we assume that the original spare was
2398 * done in response to the FMA fault.
2400 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2401 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2402 vd->vdev_parent->vdev_child[0] == vd)
2403 vd->vdev_unspare = B_TRUE;
2407 vdev_is_dead(vdev_t *vd)
2410 * Holes and missing devices are always considered "dead".
2411 * This simplifies the code since we don't have to check for
2412 * these types of devices in the various code paths.
2413 * Instead we rely on the fact that we skip over dead devices
2414 * before issuing I/O to them.
2416 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2417 vd->vdev_ops == &vdev_missing_ops);
2421 vdev_readable(vdev_t *vd)
2423 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2427 vdev_writeable(vdev_t *vd)
2429 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2433 vdev_allocatable(vdev_t *vd)
2435 uint64_t state = vd->vdev_state;
2438 * We currently allow allocations from vdevs which may be in the
2439 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2440 * fails to reopen then we'll catch it later when we're holding
2441 * the proper locks. Note that we have to get the vdev state
2442 * in a local variable because although it changes atomically,
2443 * we're asking two separate questions about it.
2445 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2446 !vd->vdev_cant_write && !vd->vdev_ishole);
2450 vdev_accessible(vdev_t *vd, zio_t *zio)
2452 ASSERT(zio->io_vd == vd);
2454 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2457 if (zio->io_type == ZIO_TYPE_READ)
2458 return (!vd->vdev_cant_read);
2460 if (zio->io_type == ZIO_TYPE_WRITE)
2461 return (!vd->vdev_cant_write);
2467 * Get statistics for the given vdev.
2470 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2472 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2474 mutex_enter(&vd->vdev_stat_lock);
2475 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2476 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2477 vs->vs_state = vd->vdev_state;
2478 vs->vs_rsize = vdev_get_min_asize(vd);
2479 if (vd->vdev_ops->vdev_op_leaf)
2480 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2481 mutex_exit(&vd->vdev_stat_lock);
2484 * If we're getting stats on the root vdev, aggregate the I/O counts
2485 * over all top-level vdevs (i.e. the direct children of the root).
2488 for (int c = 0; c < rvd->vdev_children; c++) {
2489 vdev_t *cvd = rvd->vdev_child[c];
2490 vdev_stat_t *cvs = &cvd->vdev_stat;
2492 mutex_enter(&vd->vdev_stat_lock);
2493 for (int t = 0; t < ZIO_TYPES; t++) {
2494 vs->vs_ops[t] += cvs->vs_ops[t];
2495 vs->vs_bytes[t] += cvs->vs_bytes[t];
2497 cvs->vs_scan_removing = cvd->vdev_removing;
2498 mutex_exit(&vd->vdev_stat_lock);
2504 vdev_clear_stats(vdev_t *vd)
2506 mutex_enter(&vd->vdev_stat_lock);
2507 vd->vdev_stat.vs_space = 0;
2508 vd->vdev_stat.vs_dspace = 0;
2509 vd->vdev_stat.vs_alloc = 0;
2510 mutex_exit(&vd->vdev_stat_lock);
2514 vdev_scan_stat_init(vdev_t *vd)
2516 vdev_stat_t *vs = &vd->vdev_stat;
2518 for (int c = 0; c < vd->vdev_children; c++)
2519 vdev_scan_stat_init(vd->vdev_child[c]);
2521 mutex_enter(&vd->vdev_stat_lock);
2522 vs->vs_scan_processed = 0;
2523 mutex_exit(&vd->vdev_stat_lock);
2527 vdev_stat_update(zio_t *zio, uint64_t psize)
2529 spa_t *spa = zio->io_spa;
2530 vdev_t *rvd = spa->spa_root_vdev;
2531 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2533 uint64_t txg = zio->io_txg;
2534 vdev_stat_t *vs = &vd->vdev_stat;
2535 zio_type_t type = zio->io_type;
2536 int flags = zio->io_flags;
2539 * If this i/o is a gang leader, it didn't do any actual work.
2541 if (zio->io_gang_tree)
2544 if (zio->io_error == 0) {
2546 * If this is a root i/o, don't count it -- we've already
2547 * counted the top-level vdevs, and vdev_get_stats() will
2548 * aggregate them when asked. This reduces contention on
2549 * the root vdev_stat_lock and implicitly handles blocks
2550 * that compress away to holes, for which there is no i/o.
2551 * (Holes never create vdev children, so all the counters
2552 * remain zero, which is what we want.)
2554 * Note: this only applies to successful i/o (io_error == 0)
2555 * because unlike i/o counts, errors are not additive.
2556 * When reading a ditto block, for example, failure of
2557 * one top-level vdev does not imply a root-level error.
2562 ASSERT(vd == zio->io_vd);
2564 if (flags & ZIO_FLAG_IO_BYPASS)
2567 mutex_enter(&vd->vdev_stat_lock);
2569 if (flags & ZIO_FLAG_IO_REPAIR) {
2570 if (flags & ZIO_FLAG_SCAN_THREAD) {
2571 dsl_scan_phys_t *scn_phys =
2572 &spa->spa_dsl_pool->dp_scan->scn_phys;
2573 uint64_t *processed = &scn_phys->scn_processed;
2576 if (vd->vdev_ops->vdev_op_leaf)
2577 atomic_add_64(processed, psize);
2578 vs->vs_scan_processed += psize;
2581 if (flags & ZIO_FLAG_SELF_HEAL)
2582 vs->vs_self_healed += psize;
2586 vs->vs_bytes[type] += psize;
2588 mutex_exit(&vd->vdev_stat_lock);
2592 if (flags & ZIO_FLAG_SPECULATIVE)
2596 * If this is an I/O error that is going to be retried, then ignore the
2597 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2598 * hard errors, when in reality they can happen for any number of
2599 * innocuous reasons (bus resets, MPxIO link failure, etc).
2601 if (zio->io_error == EIO &&
2602 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2606 * Intent logs writes won't propagate their error to the root
2607 * I/O so don't mark these types of failures as pool-level
2610 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2613 mutex_enter(&vd->vdev_stat_lock);
2614 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2615 if (zio->io_error == ECKSUM)
2616 vs->vs_checksum_errors++;
2618 vs->vs_read_errors++;
2620 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2621 vs->vs_write_errors++;
2622 mutex_exit(&vd->vdev_stat_lock);
2624 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2625 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2626 (flags & ZIO_FLAG_SCAN_THREAD) ||
2627 spa->spa_claiming)) {
2629 * This is either a normal write (not a repair), or it's
2630 * a repair induced by the scrub thread, or it's a repair
2631 * made by zil_claim() during spa_load() in the first txg.
2632 * In the normal case, we commit the DTL change in the same
2633 * txg as the block was born. In the scrub-induced repair
2634 * case, we know that scrubs run in first-pass syncing context,
2635 * so we commit the DTL change in spa_syncing_txg(spa).
2636 * In the zil_claim() case, we commit in spa_first_txg(spa).
2638 * We currently do not make DTL entries for failed spontaneous
2639 * self-healing writes triggered by normal (non-scrubbing)
2640 * reads, because we have no transactional context in which to
2641 * do so -- and it's not clear that it'd be desirable anyway.
2643 if (vd->vdev_ops->vdev_op_leaf) {
2644 uint64_t commit_txg = txg;
2645 if (flags & ZIO_FLAG_SCAN_THREAD) {
2646 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2647 ASSERT(spa_sync_pass(spa) == 1);
2648 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2649 commit_txg = spa_syncing_txg(spa);
2650 } else if (spa->spa_claiming) {
2651 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2652 commit_txg = spa_first_txg(spa);
2654 ASSERT(commit_txg >= spa_syncing_txg(spa));
2655 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2657 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2658 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2659 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2662 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2667 * Update the in-core space usage stats for this vdev, its metaslab class,
2668 * and the root vdev.
2671 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2672 int64_t space_delta)
2674 int64_t dspace_delta = space_delta;
2675 spa_t *spa = vd->vdev_spa;
2676 vdev_t *rvd = spa->spa_root_vdev;
2677 metaslab_group_t *mg = vd->vdev_mg;
2678 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2680 ASSERT(vd == vd->vdev_top);
2683 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2684 * factor. We must calculate this here and not at the root vdev
2685 * because the root vdev's psize-to-asize is simply the max of its
2686 * childrens', thus not accurate enough for us.
2688 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2689 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2690 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2691 vd->vdev_deflate_ratio;
2693 mutex_enter(&vd->vdev_stat_lock);
2694 vd->vdev_stat.vs_alloc += alloc_delta;
2695 vd->vdev_stat.vs_space += space_delta;
2696 vd->vdev_stat.vs_dspace += dspace_delta;
2697 mutex_exit(&vd->vdev_stat_lock);
2699 if (mc == spa_normal_class(spa)) {
2700 mutex_enter(&rvd->vdev_stat_lock);
2701 rvd->vdev_stat.vs_alloc += alloc_delta;
2702 rvd->vdev_stat.vs_space += space_delta;
2703 rvd->vdev_stat.vs_dspace += dspace_delta;
2704 mutex_exit(&rvd->vdev_stat_lock);
2708 ASSERT(rvd == vd->vdev_parent);
2709 ASSERT(vd->vdev_ms_count != 0);
2711 metaslab_class_space_update(mc,
2712 alloc_delta, defer_delta, space_delta, dspace_delta);
2717 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2718 * so that it will be written out next time the vdev configuration is synced.
2719 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2722 vdev_config_dirty(vdev_t *vd)
2724 spa_t *spa = vd->vdev_spa;
2725 vdev_t *rvd = spa->spa_root_vdev;
2728 ASSERT(spa_writeable(spa));
2731 * If this is an aux vdev (as with l2cache and spare devices), then we
2732 * update the vdev config manually and set the sync flag.
2734 if (vd->vdev_aux != NULL) {
2735 spa_aux_vdev_t *sav = vd->vdev_aux;
2739 for (c = 0; c < sav->sav_count; c++) {
2740 if (sav->sav_vdevs[c] == vd)
2744 if (c == sav->sav_count) {
2746 * We're being removed. There's nothing more to do.
2748 ASSERT(sav->sav_sync == B_TRUE);
2752 sav->sav_sync = B_TRUE;
2754 if (nvlist_lookup_nvlist_array(sav->sav_config,
2755 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2756 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2757 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2763 * Setting the nvlist in the middle if the array is a little
2764 * sketchy, but it will work.
2766 nvlist_free(aux[c]);
2767 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2773 * The dirty list is protected by the SCL_CONFIG lock. The caller
2774 * must either hold SCL_CONFIG as writer, or must be the sync thread
2775 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2776 * so this is sufficient to ensure mutual exclusion.
2778 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2779 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2780 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2783 for (c = 0; c < rvd->vdev_children; c++)
2784 vdev_config_dirty(rvd->vdev_child[c]);
2786 ASSERT(vd == vd->vdev_top);
2788 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2790 list_insert_head(&spa->spa_config_dirty_list, vd);
2795 vdev_config_clean(vdev_t *vd)
2797 spa_t *spa = vd->vdev_spa;
2799 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2800 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2801 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2803 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2804 list_remove(&spa->spa_config_dirty_list, vd);
2808 * Mark a top-level vdev's state as dirty, so that the next pass of
2809 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2810 * the state changes from larger config changes because they require
2811 * much less locking, and are often needed for administrative actions.
2814 vdev_state_dirty(vdev_t *vd)
2816 spa_t *spa = vd->vdev_spa;
2818 ASSERT(spa_writeable(spa));
2819 ASSERT(vd == vd->vdev_top);
2822 * The state list is protected by the SCL_STATE lock. The caller
2823 * must either hold SCL_STATE as writer, or must be the sync thread
2824 * (which holds SCL_STATE as reader). There's only one sync thread,
2825 * so this is sufficient to ensure mutual exclusion.
2827 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2828 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2829 spa_config_held(spa, SCL_STATE, RW_READER)));
2831 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2832 list_insert_head(&spa->spa_state_dirty_list, vd);
2836 vdev_state_clean(vdev_t *vd)
2838 spa_t *spa = vd->vdev_spa;
2840 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2841 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2842 spa_config_held(spa, SCL_STATE, RW_READER)));
2844 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2845 list_remove(&spa->spa_state_dirty_list, vd);
2849 * Propagate vdev state up from children to parent.
2852 vdev_propagate_state(vdev_t *vd)
2854 spa_t *spa = vd->vdev_spa;
2855 vdev_t *rvd = spa->spa_root_vdev;
2856 int degraded = 0, faulted = 0;
2860 if (vd->vdev_children > 0) {
2861 for (int c = 0; c < vd->vdev_children; c++) {
2862 child = vd->vdev_child[c];
2865 * Don't factor holes into the decision.
2867 if (child->vdev_ishole)
2870 if (!vdev_readable(child) ||
2871 (!vdev_writeable(child) && spa_writeable(spa))) {
2873 * Root special: if there is a top-level log
2874 * device, treat the root vdev as if it were
2877 if (child->vdev_islog && vd == rvd)
2881 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2885 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2889 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2892 * Root special: if there is a top-level vdev that cannot be
2893 * opened due to corrupted metadata, then propagate the root
2894 * vdev's aux state as 'corrupt' rather than 'insufficient
2897 if (corrupted && vd == rvd &&
2898 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2899 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2900 VDEV_AUX_CORRUPT_DATA);
2903 if (vd->vdev_parent)
2904 vdev_propagate_state(vd->vdev_parent);
2908 * Set a vdev's state. If this is during an open, we don't update the parent
2909 * state, because we're in the process of opening children depth-first.
2910 * Otherwise, we propagate the change to the parent.
2912 * If this routine places a device in a faulted state, an appropriate ereport is
2916 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2918 uint64_t save_state;
2919 spa_t *spa = vd->vdev_spa;
2921 if (state == vd->vdev_state) {
2922 vd->vdev_stat.vs_aux = aux;
2926 save_state = vd->vdev_state;
2928 vd->vdev_state = state;
2929 vd->vdev_stat.vs_aux = aux;
2932 * If we are setting the vdev state to anything but an open state, then
2933 * always close the underlying device unless the device has requested
2934 * a delayed close (i.e. we're about to remove or fault the device).
2935 * Otherwise, we keep accessible but invalid devices open forever.
2936 * We don't call vdev_close() itself, because that implies some extra
2937 * checks (offline, etc) that we don't want here. This is limited to
2938 * leaf devices, because otherwise closing the device will affect other
2941 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2942 vd->vdev_ops->vdev_op_leaf)
2943 vd->vdev_ops->vdev_op_close(vd);
2946 * If we have brought this vdev back into service, we need
2947 * to notify fmd so that it can gracefully repair any outstanding
2948 * cases due to a missing device. We do this in all cases, even those
2949 * that probably don't correlate to a repaired fault. This is sure to
2950 * catch all cases, and we let the zfs-retire agent sort it out. If
2951 * this is a transient state it's OK, as the retire agent will
2952 * double-check the state of the vdev before repairing it.
2954 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2955 vd->vdev_prevstate != state)
2956 zfs_post_state_change(spa, vd);
2958 if (vd->vdev_removed &&
2959 state == VDEV_STATE_CANT_OPEN &&
2960 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2962 * If the previous state is set to VDEV_STATE_REMOVED, then this
2963 * device was previously marked removed and someone attempted to
2964 * reopen it. If this failed due to a nonexistent device, then
2965 * keep the device in the REMOVED state. We also let this be if
2966 * it is one of our special test online cases, which is only
2967 * attempting to online the device and shouldn't generate an FMA
2970 vd->vdev_state = VDEV_STATE_REMOVED;
2971 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2972 } else if (state == VDEV_STATE_REMOVED) {
2973 vd->vdev_removed = B_TRUE;
2974 } else if (state == VDEV_STATE_CANT_OPEN) {
2976 * If we fail to open a vdev during an import or recovery, we
2977 * mark it as "not available", which signifies that it was
2978 * never there to begin with. Failure to open such a device
2979 * is not considered an error.
2981 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2982 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2983 vd->vdev_ops->vdev_op_leaf)
2984 vd->vdev_not_present = 1;
2987 * Post the appropriate ereport. If the 'prevstate' field is
2988 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2989 * that this is part of a vdev_reopen(). In this case, we don't
2990 * want to post the ereport if the device was already in the
2991 * CANT_OPEN state beforehand.
2993 * If the 'checkremove' flag is set, then this is an attempt to
2994 * online the device in response to an insertion event. If we
2995 * hit this case, then we have detected an insertion event for a
2996 * faulted or offline device that wasn't in the removed state.
2997 * In this scenario, we don't post an ereport because we are
2998 * about to replace the device, or attempt an online with
2999 * vdev_forcefault, which will generate the fault for us.
3001 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3002 !vd->vdev_not_present && !vd->vdev_checkremove &&
3003 vd != spa->spa_root_vdev) {
3007 case VDEV_AUX_OPEN_FAILED:
3008 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3010 case VDEV_AUX_CORRUPT_DATA:
3011 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3013 case VDEV_AUX_NO_REPLICAS:
3014 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3016 case VDEV_AUX_BAD_GUID_SUM:
3017 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3019 case VDEV_AUX_TOO_SMALL:
3020 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3022 case VDEV_AUX_BAD_LABEL:
3023 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3026 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3029 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3032 /* Erase any notion of persistent removed state */
3033 vd->vdev_removed = B_FALSE;
3035 vd->vdev_removed = B_FALSE;
3038 if (!isopen && vd->vdev_parent)
3039 vdev_propagate_state(vd->vdev_parent);
3043 * Check the vdev configuration to ensure that it's capable of supporting
3046 * On Solaris, we do not support RAID-Z or partial configuration. In
3047 * addition, only a single top-level vdev is allowed and none of the
3048 * leaves can be wholedisks.
3050 * For FreeBSD, we can boot from any configuration. There is a
3051 * limitation that the boot filesystem must be either uncompressed or
3052 * compresses with lzjb compression but I'm not sure how to enforce
3056 vdev_is_bootable(vdev_t *vd)
3059 if (!vd->vdev_ops->vdev_op_leaf) {
3060 char *vdev_type = vd->vdev_ops->vdev_op_type;
3062 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3063 vd->vdev_children > 1) {
3065 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3066 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3069 } else if (vd->vdev_wholedisk == 1) {
3073 for (int c = 0; c < vd->vdev_children; c++) {
3074 if (!vdev_is_bootable(vd->vdev_child[c]))
3082 * Load the state from the original vdev tree (ovd) which
3083 * we've retrieved from the MOS config object. If the original
3084 * vdev was offline or faulted then we transfer that state to the
3085 * device in the current vdev tree (nvd).
3088 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3090 spa_t *spa = nvd->vdev_spa;
3092 ASSERT(nvd->vdev_top->vdev_islog);
3093 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3094 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3096 for (int c = 0; c < nvd->vdev_children; c++)
3097 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3099 if (nvd->vdev_ops->vdev_op_leaf) {
3101 * Restore the persistent vdev state
3103 nvd->vdev_offline = ovd->vdev_offline;
3104 nvd->vdev_faulted = ovd->vdev_faulted;
3105 nvd->vdev_degraded = ovd->vdev_degraded;
3106 nvd->vdev_removed = ovd->vdev_removed;
3111 * Determine if a log device has valid content. If the vdev was
3112 * removed or faulted in the MOS config then we know that
3113 * the content on the log device has already been written to the pool.
3116 vdev_log_state_valid(vdev_t *vd)
3118 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3122 for (int c = 0; c < vd->vdev_children; c++)
3123 if (vdev_log_state_valid(vd->vdev_child[c]))
3130 * Expand a vdev if possible.
3133 vdev_expand(vdev_t *vd, uint64_t txg)
3135 ASSERT(vd->vdev_top == vd);
3136 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3138 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3139 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3140 vdev_config_dirty(vd);
3148 vdev_split(vdev_t *vd)
3150 vdev_t *cvd, *pvd = vd->vdev_parent;
3152 vdev_remove_child(pvd, vd);
3153 vdev_compact_children(pvd);
3155 cvd = pvd->vdev_child[0];
3156 if (pvd->vdev_children == 1) {
3157 vdev_remove_parent(cvd);
3158 cvd->vdev_splitting = B_TRUE;
3160 vdev_propagate_state(cvd);