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) 2012 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>
48 * Virtual device management.
51 static vdev_ops_t *vdev_ops_table[] = {
64 /* maximum scrub/resilver I/O queue per leaf vdev */
65 int zfs_scrub_limit = 10;
68 * Given a vdev type, return the appropriate ops vector.
71 vdev_getops(const char *type)
73 vdev_ops_t *ops, **opspp;
75 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
76 if (strcmp(ops->vdev_op_type, type) == 0)
83 * Default asize function: return the MAX of psize with the asize of
84 * all children. This is what's used by anything other than RAID-Z.
87 vdev_default_asize(vdev_t *vd, uint64_t psize)
89 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
93 for (c = 0; c < vd->vdev_children; c++) {
94 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
95 asize = MAX(asize, csize);
102 * Get the minimum allocatable size. We define the allocatable size as
103 * the vdev's asize rounded to the nearest metaslab. This allows us to
104 * replace or attach devices which don't have the same physical size but
105 * can still satisfy the same number of allocations.
108 vdev_get_min_asize(vdev_t *vd)
110 vdev_t *pvd = vd->vdev_parent;
113 * If our parent is NULL (inactive spare or cache) or is the root,
114 * just return our own asize.
117 return (vd->vdev_asize);
120 * The top-level vdev just returns the allocatable size rounded
121 * to the nearest metaslab.
123 if (vd == vd->vdev_top)
124 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
127 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
128 * so each child must provide at least 1/Nth of its asize.
130 if (pvd->vdev_ops == &vdev_raidz_ops)
131 return (pvd->vdev_min_asize / pvd->vdev_children);
133 return (pvd->vdev_min_asize);
137 vdev_set_min_asize(vdev_t *vd)
140 vd->vdev_min_asize = vdev_get_min_asize(vd);
142 for (c = 0; c < vd->vdev_children; c++)
143 vdev_set_min_asize(vd->vdev_child[c]);
147 vdev_lookup_top(spa_t *spa, uint64_t vdev)
149 vdev_t *rvd = spa->spa_root_vdev;
151 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
153 if (vdev < rvd->vdev_children) {
154 ASSERT(rvd->vdev_child[vdev] != NULL);
155 return (rvd->vdev_child[vdev]);
162 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
167 if (vd->vdev_guid == guid)
170 for (c = 0; c < vd->vdev_children; c++)
171 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
179 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
181 size_t oldsize, newsize;
182 uint64_t id = cvd->vdev_id;
185 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
186 ASSERT(cvd->vdev_parent == NULL);
188 cvd->vdev_parent = pvd;
193 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
195 oldsize = pvd->vdev_children * sizeof (vdev_t *);
196 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
197 newsize = pvd->vdev_children * sizeof (vdev_t *);
199 newchild = kmem_zalloc(newsize, KM_PUSHPAGE);
200 if (pvd->vdev_child != NULL) {
201 bcopy(pvd->vdev_child, newchild, oldsize);
202 kmem_free(pvd->vdev_child, oldsize);
205 pvd->vdev_child = newchild;
206 pvd->vdev_child[id] = cvd;
208 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
209 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
212 * Walk up all ancestors to update guid sum.
214 for (; pvd != NULL; pvd = pvd->vdev_parent)
215 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
219 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
222 uint_t id = cvd->vdev_id;
224 ASSERT(cvd->vdev_parent == pvd);
229 ASSERT(id < pvd->vdev_children);
230 ASSERT(pvd->vdev_child[id] == cvd);
232 pvd->vdev_child[id] = NULL;
233 cvd->vdev_parent = NULL;
235 for (c = 0; c < pvd->vdev_children; c++)
236 if (pvd->vdev_child[c])
239 if (c == pvd->vdev_children) {
240 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
241 pvd->vdev_child = NULL;
242 pvd->vdev_children = 0;
246 * Walk up all ancestors to update guid sum.
248 for (; pvd != NULL; pvd = pvd->vdev_parent)
249 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
253 * Remove any holes in the child array.
256 vdev_compact_children(vdev_t *pvd)
258 vdev_t **newchild, *cvd;
259 int oldc = pvd->vdev_children;
263 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
265 for (c = newc = 0; c < oldc; c++)
266 if (pvd->vdev_child[c])
269 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_PUSHPAGE);
271 for (c = newc = 0; c < oldc; c++) {
272 if ((cvd = pvd->vdev_child[c]) != NULL) {
273 newchild[newc] = cvd;
274 cvd->vdev_id = newc++;
278 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
279 pvd->vdev_child = newchild;
280 pvd->vdev_children = newc;
284 * Allocate and minimally initialize a vdev_t.
287 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
292 vd = kmem_zalloc(sizeof (vdev_t), KM_PUSHPAGE);
294 if (spa->spa_root_vdev == NULL) {
295 ASSERT(ops == &vdev_root_ops);
296 spa->spa_root_vdev = vd;
297 spa->spa_load_guid = spa_generate_guid(NULL);
300 if (guid == 0 && ops != &vdev_hole_ops) {
301 if (spa->spa_root_vdev == vd) {
303 * The root vdev's guid will also be the pool guid,
304 * which must be unique among all pools.
306 guid = spa_generate_guid(NULL);
309 * Any other vdev's guid must be unique within the pool.
311 guid = spa_generate_guid(spa);
313 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
318 vd->vdev_guid = guid;
319 vd->vdev_guid_sum = guid;
321 vd->vdev_state = VDEV_STATE_CLOSED;
322 vd->vdev_ishole = (ops == &vdev_hole_ops);
324 list_link_init(&vd->vdev_config_dirty_node);
325 list_link_init(&vd->vdev_state_dirty_node);
326 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
327 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
328 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
329 for (t = 0; t < DTL_TYPES; t++) {
330 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
333 txg_list_create(&vd->vdev_ms_list,
334 offsetof(struct metaslab, ms_txg_node));
335 txg_list_create(&vd->vdev_dtl_list,
336 offsetof(struct vdev, vdev_dtl_node));
337 vd->vdev_stat.vs_timestamp = gethrtime();
345 * Allocate a new vdev. The 'alloctype' is used to control whether we are
346 * creating a new vdev or loading an existing one - the behavior is slightly
347 * different for each case.
350 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
355 uint64_t guid = 0, islog, nparity;
358 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
360 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
363 if ((ops = vdev_getops(type)) == NULL)
367 * If this is a load, get the vdev guid from the nvlist.
368 * Otherwise, vdev_alloc_common() will generate one for us.
370 if (alloctype == VDEV_ALLOC_LOAD) {
373 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
377 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
379 } else if (alloctype == VDEV_ALLOC_SPARE) {
380 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
382 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
383 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
385 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
386 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
391 * The first allocated vdev must be of type 'root'.
393 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
397 * Determine whether we're a log vdev.
400 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
401 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
404 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
408 * Set the nparity property for RAID-Z vdevs.
411 if (ops == &vdev_raidz_ops) {
412 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
414 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
417 * Previous versions could only support 1 or 2 parity
421 spa_version(spa) < SPA_VERSION_RAIDZ2)
424 spa_version(spa) < SPA_VERSION_RAIDZ3)
428 * We require the parity to be specified for SPAs that
429 * support multiple parity levels.
431 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
434 * Otherwise, we default to 1 parity device for RAID-Z.
441 ASSERT(nparity != -1ULL);
443 vd = vdev_alloc_common(spa, id, guid, ops);
445 vd->vdev_islog = islog;
446 vd->vdev_nparity = nparity;
448 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
449 vd->vdev_path = spa_strdup(vd->vdev_path);
450 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
451 vd->vdev_devid = spa_strdup(vd->vdev_devid);
452 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
453 &vd->vdev_physpath) == 0)
454 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
455 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
456 vd->vdev_fru = spa_strdup(vd->vdev_fru);
459 * Set the whole_disk property. If it's not specified, leave the value
462 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
463 &vd->vdev_wholedisk) != 0)
464 vd->vdev_wholedisk = -1ULL;
467 * Look for the 'not present' flag. This will only be set if the device
468 * was not present at the time of import.
470 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
471 &vd->vdev_not_present);
474 * Get the alignment requirement.
476 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
479 * Retrieve the vdev creation time.
481 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
485 * If we're a top-level vdev, try to load the allocation parameters.
487 if (parent && !parent->vdev_parent &&
488 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
489 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
491 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
493 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
495 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
499 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
500 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
501 alloctype == VDEV_ALLOC_ADD ||
502 alloctype == VDEV_ALLOC_SPLIT ||
503 alloctype == VDEV_ALLOC_ROOTPOOL);
504 vd->vdev_mg = metaslab_group_create(islog ?
505 spa_log_class(spa) : spa_normal_class(spa), vd);
509 * If we're a leaf vdev, try to load the DTL object and other state.
511 if (vd->vdev_ops->vdev_op_leaf &&
512 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
513 alloctype == VDEV_ALLOC_ROOTPOOL)) {
514 if (alloctype == VDEV_ALLOC_LOAD) {
515 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
516 &vd->vdev_dtl_smo.smo_object);
517 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
521 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
524 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
525 &spare) == 0 && spare)
529 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
532 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
533 &vd->vdev_resilvering);
536 * When importing a pool, we want to ignore the persistent fault
537 * state, as the diagnosis made on another system may not be
538 * valid in the current context. Local vdevs will
539 * remain in the faulted state.
541 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
542 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
544 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
546 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
549 if (vd->vdev_faulted || vd->vdev_degraded) {
553 VDEV_AUX_ERR_EXCEEDED;
554 if (nvlist_lookup_string(nv,
555 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
556 strcmp(aux, "external") == 0)
557 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
563 * Add ourselves to the parent's list of children.
565 vdev_add_child(parent, vd);
573 vdev_free(vdev_t *vd)
576 spa_t *spa = vd->vdev_spa;
579 * vdev_free() implies closing the vdev first. This is simpler than
580 * trying to ensure complicated semantics for all callers.
584 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
585 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
590 for (c = 0; c < vd->vdev_children; c++)
591 vdev_free(vd->vdev_child[c]);
593 ASSERT(vd->vdev_child == NULL);
594 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
597 * Discard allocation state.
599 if (vd->vdev_mg != NULL) {
600 vdev_metaslab_fini(vd);
601 metaslab_group_destroy(vd->vdev_mg);
604 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
605 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
606 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
609 * Remove this vdev from its parent's child list.
611 vdev_remove_child(vd->vdev_parent, vd);
613 ASSERT(vd->vdev_parent == NULL);
616 * Clean up vdev structure.
622 spa_strfree(vd->vdev_path);
624 spa_strfree(vd->vdev_devid);
625 if (vd->vdev_physpath)
626 spa_strfree(vd->vdev_physpath);
628 spa_strfree(vd->vdev_fru);
630 if (vd->vdev_isspare)
631 spa_spare_remove(vd);
632 if (vd->vdev_isl2cache)
633 spa_l2cache_remove(vd);
635 txg_list_destroy(&vd->vdev_ms_list);
636 txg_list_destroy(&vd->vdev_dtl_list);
638 mutex_enter(&vd->vdev_dtl_lock);
639 for (t = 0; t < DTL_TYPES; t++) {
640 space_map_unload(&vd->vdev_dtl[t]);
641 space_map_destroy(&vd->vdev_dtl[t]);
643 mutex_exit(&vd->vdev_dtl_lock);
645 mutex_destroy(&vd->vdev_dtl_lock);
646 mutex_destroy(&vd->vdev_stat_lock);
647 mutex_destroy(&vd->vdev_probe_lock);
649 if (vd == spa->spa_root_vdev)
650 spa->spa_root_vdev = NULL;
652 kmem_free(vd, sizeof (vdev_t));
656 * Transfer top-level vdev state from svd to tvd.
659 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
661 spa_t *spa = svd->vdev_spa;
666 ASSERT(tvd == tvd->vdev_top);
668 tvd->vdev_ms_array = svd->vdev_ms_array;
669 tvd->vdev_ms_shift = svd->vdev_ms_shift;
670 tvd->vdev_ms_count = svd->vdev_ms_count;
672 svd->vdev_ms_array = 0;
673 svd->vdev_ms_shift = 0;
674 svd->vdev_ms_count = 0;
677 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
678 tvd->vdev_mg = svd->vdev_mg;
679 tvd->vdev_ms = svd->vdev_ms;
684 if (tvd->vdev_mg != NULL)
685 tvd->vdev_mg->mg_vd = tvd;
687 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
688 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
689 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
691 svd->vdev_stat.vs_alloc = 0;
692 svd->vdev_stat.vs_space = 0;
693 svd->vdev_stat.vs_dspace = 0;
695 for (t = 0; t < TXG_SIZE; t++) {
696 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
697 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
698 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
699 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
700 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
701 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
704 if (list_link_active(&svd->vdev_config_dirty_node)) {
705 vdev_config_clean(svd);
706 vdev_config_dirty(tvd);
709 if (list_link_active(&svd->vdev_state_dirty_node)) {
710 vdev_state_clean(svd);
711 vdev_state_dirty(tvd);
714 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
715 svd->vdev_deflate_ratio = 0;
717 tvd->vdev_islog = svd->vdev_islog;
722 vdev_top_update(vdev_t *tvd, vdev_t *vd)
731 for (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_max_asize = cvd->vdev_max_asize;
752 mvd->vdev_ashift = cvd->vdev_ashift;
753 mvd->vdev_state = cvd->vdev_state;
754 mvd->vdev_crtxg = cvd->vdev_crtxg;
756 vdev_remove_child(pvd, cvd);
757 vdev_add_child(pvd, mvd);
758 cvd->vdev_id = mvd->vdev_children;
759 vdev_add_child(mvd, cvd);
760 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
762 if (mvd == mvd->vdev_top)
763 vdev_top_transfer(cvd, mvd);
769 * Remove a 1-way mirror/replacing vdev from the tree.
772 vdev_remove_parent(vdev_t *cvd)
774 vdev_t *mvd = cvd->vdev_parent;
775 vdev_t *pvd = mvd->vdev_parent;
777 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
779 ASSERT(mvd->vdev_children == 1);
780 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
781 mvd->vdev_ops == &vdev_replacing_ops ||
782 mvd->vdev_ops == &vdev_spare_ops);
783 cvd->vdev_ashift = mvd->vdev_ashift;
785 vdev_remove_child(mvd, cvd);
786 vdev_remove_child(pvd, mvd);
789 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
790 * Otherwise, we could have detached an offline device, and when we
791 * go to import the pool we'll think we have two top-level vdevs,
792 * instead of a different version of the same top-level vdev.
794 if (mvd->vdev_top == mvd) {
795 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
796 cvd->vdev_orig_guid = cvd->vdev_guid;
797 cvd->vdev_guid += guid_delta;
798 cvd->vdev_guid_sum += guid_delta;
800 cvd->vdev_id = mvd->vdev_id;
801 vdev_add_child(pvd, cvd);
802 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
804 if (cvd == cvd->vdev_top)
805 vdev_top_transfer(mvd, cvd);
807 ASSERT(mvd->vdev_children == 0);
812 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
814 spa_t *spa = vd->vdev_spa;
815 objset_t *mos = spa->spa_meta_objset;
817 uint64_t oldc = vd->vdev_ms_count;
818 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
822 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
825 * This vdev is not being allocated from yet or is a hole.
827 if (vd->vdev_ms_shift == 0)
830 ASSERT(!vd->vdev_ishole);
833 * Compute the raidz-deflation ratio. Note, we hard-code
834 * in 128k (1 << 17) because it is the current "typical" blocksize.
835 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
836 * or we will inconsistently account for existing bp's.
838 vd->vdev_deflate_ratio = (1 << 17) /
839 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
841 ASSERT(oldc <= newc);
843 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_PUSHPAGE | KM_NODEBUG);
846 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
847 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
851 vd->vdev_ms_count = newc;
853 for (m = oldc; m < newc; m++) {
854 space_map_obj_t smo = { 0, 0, 0 };
857 error = dmu_read(mos, vd->vdev_ms_array,
858 m * sizeof (uint64_t), sizeof (uint64_t), &object,
864 error = dmu_bonus_hold(mos, object, FTAG, &db);
867 ASSERT3U(db->db_size, >=, sizeof (smo));
868 bcopy(db->db_data, &smo, sizeof (smo));
869 ASSERT3U(smo.smo_object, ==, object);
870 dmu_buf_rele(db, FTAG);
873 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
874 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
878 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
881 * If the vdev is being removed we don't activate
882 * the metaslabs since we want to ensure that no new
883 * allocations are performed on this device.
885 if (oldc == 0 && !vd->vdev_removing)
886 metaslab_group_activate(vd->vdev_mg);
889 spa_config_exit(spa, SCL_ALLOC, FTAG);
895 vdev_metaslab_fini(vdev_t *vd)
898 uint64_t count = vd->vdev_ms_count;
900 if (vd->vdev_ms != NULL) {
901 metaslab_group_passivate(vd->vdev_mg);
902 for (m = 0; m < count; m++)
903 if (vd->vdev_ms[m] != NULL)
904 metaslab_fini(vd->vdev_ms[m]);
905 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
909 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
912 typedef struct vdev_probe_stats {
913 boolean_t vps_readable;
914 boolean_t vps_writeable;
916 } vdev_probe_stats_t;
919 vdev_probe_done(zio_t *zio)
921 spa_t *spa = zio->io_spa;
922 vdev_t *vd = zio->io_vd;
923 vdev_probe_stats_t *vps = zio->io_private;
925 ASSERT(vd->vdev_probe_zio != NULL);
927 if (zio->io_type == ZIO_TYPE_READ) {
928 if (zio->io_error == 0)
929 vps->vps_readable = 1;
930 if (zio->io_error == 0 && spa_writeable(spa)) {
931 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
932 zio->io_offset, zio->io_size, zio->io_data,
933 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
934 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
936 zio_buf_free(zio->io_data, zio->io_size);
938 } else if (zio->io_type == ZIO_TYPE_WRITE) {
939 if (zio->io_error == 0)
940 vps->vps_writeable = 1;
941 zio_buf_free(zio->io_data, zio->io_size);
942 } else if (zio->io_type == ZIO_TYPE_NULL) {
945 vd->vdev_cant_read |= !vps->vps_readable;
946 vd->vdev_cant_write |= !vps->vps_writeable;
948 if (vdev_readable(vd) &&
949 (vdev_writeable(vd) || !spa_writeable(spa))) {
952 ASSERT(zio->io_error != 0);
953 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
954 spa, vd, NULL, 0, 0);
955 zio->io_error = ENXIO;
958 mutex_enter(&vd->vdev_probe_lock);
959 ASSERT(vd->vdev_probe_zio == zio);
960 vd->vdev_probe_zio = NULL;
961 mutex_exit(&vd->vdev_probe_lock);
963 while ((pio = zio_walk_parents(zio)) != NULL)
964 if (!vdev_accessible(vd, pio))
965 pio->io_error = ENXIO;
967 kmem_free(vps, sizeof (*vps));
972 * Determine whether this device is accessible by reading and writing
973 * to several known locations: the pad regions of each vdev label
974 * but the first (which we leave alone in case it contains a VTOC).
977 vdev_probe(vdev_t *vd, zio_t *zio)
979 spa_t *spa = vd->vdev_spa;
980 vdev_probe_stats_t *vps = NULL;
984 ASSERT(vd->vdev_ops->vdev_op_leaf);
987 * Don't probe the probe.
989 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
993 * To prevent 'probe storms' when a device fails, we create
994 * just one probe i/o at a time. All zios that want to probe
995 * this vdev will become parents of the probe io.
997 mutex_enter(&vd->vdev_probe_lock);
999 if ((pio = vd->vdev_probe_zio) == NULL) {
1000 vps = kmem_zalloc(sizeof (*vps), KM_PUSHPAGE);
1002 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1003 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1006 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1008 * vdev_cant_read and vdev_cant_write can only
1009 * transition from TRUE to FALSE when we have the
1010 * SCL_ZIO lock as writer; otherwise they can only
1011 * transition from FALSE to TRUE. This ensures that
1012 * any zio looking at these values can assume that
1013 * failures persist for the life of the I/O. That's
1014 * important because when a device has intermittent
1015 * connectivity problems, we want to ensure that
1016 * they're ascribed to the device (ENXIO) and not
1019 * Since we hold SCL_ZIO as writer here, clear both
1020 * values so the probe can reevaluate from first
1023 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1024 vd->vdev_cant_read = B_FALSE;
1025 vd->vdev_cant_write = B_FALSE;
1028 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1029 vdev_probe_done, vps,
1030 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1033 * We can't change the vdev state in this context, so we
1034 * kick off an async task to do it on our behalf.
1037 vd->vdev_probe_wanted = B_TRUE;
1038 spa_async_request(spa, SPA_ASYNC_PROBE);
1043 zio_add_child(zio, pio);
1045 mutex_exit(&vd->vdev_probe_lock);
1048 ASSERT(zio != NULL);
1052 for (l = 1; l < VDEV_LABELS; l++) {
1053 zio_nowait(zio_read_phys(pio, vd,
1054 vdev_label_offset(vd->vdev_psize, l,
1055 offsetof(vdev_label_t, vl_pad2)),
1056 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1057 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1058 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1069 vdev_open_child(void *arg)
1073 vd->vdev_open_thread = curthread;
1074 vd->vdev_open_error = vdev_open(vd);
1075 vd->vdev_open_thread = NULL;
1079 vdev_uses_zvols(vdev_t *vd)
1084 if (zvol_is_zvol(vd->vdev_path))
1088 for (c = 0; c < vd->vdev_children; c++)
1089 if (vdev_uses_zvols(vd->vdev_child[c]))
1096 vdev_open_children(vdev_t *vd)
1099 int children = vd->vdev_children;
1103 * in order to handle pools on top of zvols, do the opens
1104 * in a single thread so that the same thread holds the
1105 * spa_namespace_lock
1107 if (vdev_uses_zvols(vd)) {
1108 for (c = 0; c < children; c++)
1109 vd->vdev_child[c]->vdev_open_error =
1110 vdev_open(vd->vdev_child[c]);
1113 tq = taskq_create("vdev_open", children, minclsyspri,
1114 children, children, TASKQ_PREPOPULATE);
1116 for (c = 0; c < children; c++)
1117 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1124 * Prepare a virtual device for access.
1127 vdev_open(vdev_t *vd)
1129 spa_t *spa = vd->vdev_spa;
1132 uint64_t max_osize = 0;
1133 uint64_t asize, max_asize, psize;
1134 uint64_t ashift = 0;
1137 ASSERT(vd->vdev_open_thread == curthread ||
1138 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1139 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1140 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1141 vd->vdev_state == VDEV_STATE_OFFLINE);
1143 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1144 vd->vdev_cant_read = B_FALSE;
1145 vd->vdev_cant_write = B_FALSE;
1146 vd->vdev_min_asize = vdev_get_min_asize(vd);
1149 * If this vdev is not removed, check its fault status. If it's
1150 * faulted, bail out of the open.
1152 if (!vd->vdev_removed && vd->vdev_faulted) {
1153 ASSERT(vd->vdev_children == 0);
1154 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1155 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1156 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1157 vd->vdev_label_aux);
1159 } else if (vd->vdev_offline) {
1160 ASSERT(vd->vdev_children == 0);
1161 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1165 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1168 * Reset the vdev_reopening flag so that we actually close
1169 * the vdev on error.
1171 vd->vdev_reopening = B_FALSE;
1172 if (zio_injection_enabled && error == 0)
1173 error = zio_handle_device_injection(vd, NULL, ENXIO);
1176 if (vd->vdev_removed &&
1177 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1178 vd->vdev_removed = B_FALSE;
1180 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1181 vd->vdev_stat.vs_aux);
1185 vd->vdev_removed = B_FALSE;
1188 * Recheck the faulted flag now that we have confirmed that
1189 * the vdev is accessible. If we're faulted, bail.
1191 if (vd->vdev_faulted) {
1192 ASSERT(vd->vdev_children == 0);
1193 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1194 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1195 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1196 vd->vdev_label_aux);
1200 if (vd->vdev_degraded) {
1201 ASSERT(vd->vdev_children == 0);
1202 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1203 VDEV_AUX_ERR_EXCEEDED);
1205 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1209 * For hole or missing vdevs we just return success.
1211 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1214 for (c = 0; c < vd->vdev_children; c++) {
1215 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1216 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1222 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1223 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1225 if (vd->vdev_children == 0) {
1226 if (osize < SPA_MINDEVSIZE) {
1227 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1228 VDEV_AUX_TOO_SMALL);
1232 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1233 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1234 VDEV_LABEL_END_SIZE);
1236 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1237 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1238 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1239 VDEV_AUX_TOO_SMALL);
1244 max_asize = max_osize;
1247 vd->vdev_psize = psize;
1250 * Make sure the allocatable size hasn't shrunk.
1252 if (asize < vd->vdev_min_asize) {
1253 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1254 VDEV_AUX_BAD_LABEL);
1258 if (vd->vdev_asize == 0) {
1260 * This is the first-ever open, so use the computed values.
1261 * For testing purposes, a higher ashift can be requested.
1263 vd->vdev_asize = asize;
1264 vd->vdev_max_asize = max_asize;
1265 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1268 * Detect if the alignment requirement has increased.
1269 * We don't want to make the pool unavailable, just
1270 * post an event instead.
1272 if (ashift > vd->vdev_top->vdev_ashift &&
1273 vd->vdev_ops->vdev_op_leaf) {
1274 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
1275 spa, vd, NULL, 0, 0);
1278 vd->vdev_max_asize = max_asize;
1282 * If all children are healthy and the asize has increased,
1283 * then we've experienced dynamic LUN growth. If automatic
1284 * expansion is enabled then use the additional space.
1286 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1287 (vd->vdev_expanding || spa->spa_autoexpand))
1288 vd->vdev_asize = asize;
1290 vdev_set_min_asize(vd);
1293 * Ensure we can issue some IO before declaring the
1294 * vdev open for business.
1296 if (vd->vdev_ops->vdev_op_leaf &&
1297 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1298 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1299 VDEV_AUX_ERR_EXCEEDED);
1304 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1305 * resilver. But don't do this if we are doing a reopen for a scrub,
1306 * since this would just restart the scrub we are already doing.
1308 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1309 vdev_resilver_needed(vd, NULL, NULL))
1310 spa_async_request(spa, SPA_ASYNC_RESILVER);
1316 * Called once the vdevs are all opened, this routine validates the label
1317 * contents. This needs to be done before vdev_load() so that we don't
1318 * inadvertently do repair I/Os to the wrong device.
1320 * If 'strict' is false ignore the spa guid check. This is necessary because
1321 * if the machine crashed during a re-guid the new guid might have been written
1322 * to all of the vdev labels, but not the cached config. The strict check
1323 * will be performed when the pool is opened again using the mos config.
1325 * This function will only return failure if one of the vdevs indicates that it
1326 * has since been destroyed or exported. This is only possible if
1327 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1328 * will be updated but the function will return 0.
1331 vdev_validate(vdev_t *vd, boolean_t strict)
1333 spa_t *spa = vd->vdev_spa;
1335 uint64_t guid = 0, top_guid;
1339 for (c = 0; c < vd->vdev_children; c++)
1340 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1344 * If the device has already failed, or was marked offline, don't do
1345 * any further validation. Otherwise, label I/O will fail and we will
1346 * overwrite the previous state.
1348 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1349 uint64_t aux_guid = 0;
1352 if ((label = vdev_label_read_config(vd)) == NULL) {
1353 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1354 VDEV_AUX_BAD_LABEL);
1359 * Determine if this vdev has been split off into another
1360 * pool. If so, then refuse to open it.
1362 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1363 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1364 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1365 VDEV_AUX_SPLIT_POOL);
1370 if (strict && (nvlist_lookup_uint64(label,
1371 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1372 guid != spa_guid(spa))) {
1373 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1374 VDEV_AUX_CORRUPT_DATA);
1379 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1380 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1385 * If this vdev just became a top-level vdev because its
1386 * sibling was detached, it will have adopted the parent's
1387 * vdev guid -- but the label may or may not be on disk yet.
1388 * Fortunately, either version of the label will have the
1389 * same top guid, so if we're a top-level vdev, we can
1390 * safely compare to that instead.
1392 * If we split this vdev off instead, then we also check the
1393 * original pool's guid. We don't want to consider the vdev
1394 * corrupt if it is partway through a split operation.
1396 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1398 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1400 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1401 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1402 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1403 VDEV_AUX_CORRUPT_DATA);
1408 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1410 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1411 VDEV_AUX_CORRUPT_DATA);
1419 * If this is a verbatim import, no need to check the
1420 * state of the pool.
1422 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1423 spa_load_state(spa) == SPA_LOAD_OPEN &&
1424 state != POOL_STATE_ACTIVE)
1428 * If we were able to open and validate a vdev that was
1429 * previously marked permanently unavailable, clear that state
1432 if (vd->vdev_not_present)
1433 vd->vdev_not_present = 0;
1440 * Close a virtual device.
1443 vdev_close(vdev_t *vd)
1445 vdev_t *pvd = vd->vdev_parent;
1446 ASSERTV(spa_t *spa = vd->vdev_spa);
1448 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1451 * If our parent is reopening, then we are as well, unless we are
1454 if (pvd != NULL && pvd->vdev_reopening)
1455 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1457 vd->vdev_ops->vdev_op_close(vd);
1459 vdev_cache_purge(vd);
1462 * We record the previous state before we close it, so that if we are
1463 * doing a reopen(), we don't generate FMA ereports if we notice that
1464 * it's still faulted.
1466 vd->vdev_prevstate = vd->vdev_state;
1468 if (vd->vdev_offline)
1469 vd->vdev_state = VDEV_STATE_OFFLINE;
1471 vd->vdev_state = VDEV_STATE_CLOSED;
1472 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1476 vdev_hold(vdev_t *vd)
1478 spa_t *spa = vd->vdev_spa;
1481 ASSERT(spa_is_root(spa));
1482 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1485 for (c = 0; c < vd->vdev_children; c++)
1486 vdev_hold(vd->vdev_child[c]);
1488 if (vd->vdev_ops->vdev_op_leaf)
1489 vd->vdev_ops->vdev_op_hold(vd);
1493 vdev_rele(vdev_t *vd)
1497 ASSERT(spa_is_root(vd->vdev_spa));
1498 for (c = 0; c < vd->vdev_children; c++)
1499 vdev_rele(vd->vdev_child[c]);
1501 if (vd->vdev_ops->vdev_op_leaf)
1502 vd->vdev_ops->vdev_op_rele(vd);
1506 * Reopen all interior vdevs and any unopened leaves. We don't actually
1507 * reopen leaf vdevs which had previously been opened as they might deadlock
1508 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1509 * If the leaf has never been opened then open it, as usual.
1512 vdev_reopen(vdev_t *vd)
1514 spa_t *spa = vd->vdev_spa;
1516 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1518 /* set the reopening flag unless we're taking the vdev offline */
1519 vd->vdev_reopening = !vd->vdev_offline;
1521 (void) vdev_open(vd);
1524 * Call vdev_validate() here to make sure we have the same device.
1525 * Otherwise, a device with an invalid label could be successfully
1526 * opened in response to vdev_reopen().
1529 (void) vdev_validate_aux(vd);
1530 if (vdev_readable(vd) && vdev_writeable(vd) &&
1531 vd->vdev_aux == &spa->spa_l2cache &&
1532 !l2arc_vdev_present(vd))
1533 l2arc_add_vdev(spa, vd);
1535 (void) vdev_validate(vd, B_TRUE);
1539 * Reassess parent vdev's health.
1541 vdev_propagate_state(vd);
1545 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1550 * Normally, partial opens (e.g. of a mirror) are allowed.
1551 * For a create, however, we want to fail the request if
1552 * there are any components we can't open.
1554 error = vdev_open(vd);
1556 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1558 return (error ? error : ENXIO);
1562 * Recursively initialize all labels.
1564 if ((error = vdev_label_init(vd, txg, isreplacing ?
1565 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1574 vdev_metaslab_set_size(vdev_t *vd)
1577 * Aim for roughly 200 metaslabs per vdev.
1579 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1580 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1584 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1586 ASSERT(vd == vd->vdev_top);
1587 ASSERT(!vd->vdev_ishole);
1588 ASSERT(ISP2(flags));
1589 ASSERT(spa_writeable(vd->vdev_spa));
1591 if (flags & VDD_METASLAB)
1592 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1594 if (flags & VDD_DTL)
1595 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1597 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1603 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1604 * the vdev has less than perfect replication. There are four kinds of DTL:
1606 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1608 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1610 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1611 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1612 * txgs that was scrubbed.
1614 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1615 * persistent errors or just some device being offline.
1616 * Unlike the other three, the DTL_OUTAGE map is not generally
1617 * maintained; it's only computed when needed, typically to
1618 * determine whether a device can be detached.
1620 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1621 * either has the data or it doesn't.
1623 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1624 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1625 * if any child is less than fully replicated, then so is its parent.
1626 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1627 * comprising only those txgs which appear in 'maxfaults' or more children;
1628 * those are the txgs we don't have enough replication to read. For example,
1629 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1630 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1631 * two child DTL_MISSING maps.
1633 * It should be clear from the above that to compute the DTLs and outage maps
1634 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1635 * Therefore, that is all we keep on disk. When loading the pool, or after
1636 * a configuration change, we generate all other DTLs from first principles.
1639 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1641 space_map_t *sm = &vd->vdev_dtl[t];
1643 ASSERT(t < DTL_TYPES);
1644 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1645 ASSERT(spa_writeable(vd->vdev_spa));
1647 mutex_enter(sm->sm_lock);
1648 if (!space_map_contains(sm, txg, size))
1649 space_map_add(sm, txg, size);
1650 mutex_exit(sm->sm_lock);
1654 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1656 space_map_t *sm = &vd->vdev_dtl[t];
1657 boolean_t dirty = B_FALSE;
1659 ASSERT(t < DTL_TYPES);
1660 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1662 mutex_enter(sm->sm_lock);
1663 if (sm->sm_space != 0)
1664 dirty = space_map_contains(sm, txg, size);
1665 mutex_exit(sm->sm_lock);
1671 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1673 space_map_t *sm = &vd->vdev_dtl[t];
1676 mutex_enter(sm->sm_lock);
1677 empty = (sm->sm_space == 0);
1678 mutex_exit(sm->sm_lock);
1684 * Reassess DTLs after a config change or scrub completion.
1687 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1689 spa_t *spa = vd->vdev_spa;
1693 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1695 for (c = 0; c < vd->vdev_children; c++)
1696 vdev_dtl_reassess(vd->vdev_child[c], txg,
1697 scrub_txg, scrub_done);
1699 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1702 if (vd->vdev_ops->vdev_op_leaf) {
1703 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1705 mutex_enter(&vd->vdev_dtl_lock);
1706 if (scrub_txg != 0 &&
1707 (spa->spa_scrub_started ||
1708 (scn && scn->scn_phys.scn_errors == 0))) {
1710 * We completed a scrub up to scrub_txg. If we
1711 * did it without rebooting, then the scrub dtl
1712 * will be valid, so excise the old region and
1713 * fold in the scrub dtl. Otherwise, leave the
1714 * dtl as-is if there was an error.
1716 * There's little trick here: to excise the beginning
1717 * of the DTL_MISSING map, we put it into a reference
1718 * tree and then add a segment with refcnt -1 that
1719 * covers the range [0, scrub_txg). This means
1720 * that each txg in that range has refcnt -1 or 0.
1721 * We then add DTL_SCRUB with a refcnt of 2, so that
1722 * entries in the range [0, scrub_txg) will have a
1723 * positive refcnt -- either 1 or 2. We then convert
1724 * the reference tree into the new DTL_MISSING map.
1726 space_map_ref_create(&reftree);
1727 space_map_ref_add_map(&reftree,
1728 &vd->vdev_dtl[DTL_MISSING], 1);
1729 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1730 space_map_ref_add_map(&reftree,
1731 &vd->vdev_dtl[DTL_SCRUB], 2);
1732 space_map_ref_generate_map(&reftree,
1733 &vd->vdev_dtl[DTL_MISSING], 1);
1734 space_map_ref_destroy(&reftree);
1736 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1737 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1738 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1740 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1741 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1742 if (!vdev_readable(vd))
1743 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1745 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1746 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1747 mutex_exit(&vd->vdev_dtl_lock);
1750 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1754 mutex_enter(&vd->vdev_dtl_lock);
1755 for (t = 0; t < DTL_TYPES; t++) {
1756 /* account for child's outage in parent's missing map */
1757 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1759 continue; /* leaf vdevs only */
1760 if (t == DTL_PARTIAL)
1761 minref = 1; /* i.e. non-zero */
1762 else if (vd->vdev_nparity != 0)
1763 minref = vd->vdev_nparity + 1; /* RAID-Z */
1765 minref = vd->vdev_children; /* any kind of mirror */
1766 space_map_ref_create(&reftree);
1767 for (c = 0; c < vd->vdev_children; c++) {
1768 vdev_t *cvd = vd->vdev_child[c];
1769 mutex_enter(&cvd->vdev_dtl_lock);
1770 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1771 mutex_exit(&cvd->vdev_dtl_lock);
1773 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1774 space_map_ref_destroy(&reftree);
1776 mutex_exit(&vd->vdev_dtl_lock);
1780 vdev_dtl_load(vdev_t *vd)
1782 spa_t *spa = vd->vdev_spa;
1783 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1784 objset_t *mos = spa->spa_meta_objset;
1788 ASSERT(vd->vdev_children == 0);
1790 if (smo->smo_object == 0)
1793 ASSERT(!vd->vdev_ishole);
1795 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1798 ASSERT3U(db->db_size, >=, sizeof (*smo));
1799 bcopy(db->db_data, smo, sizeof (*smo));
1800 dmu_buf_rele(db, FTAG);
1802 mutex_enter(&vd->vdev_dtl_lock);
1803 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1804 NULL, SM_ALLOC, smo, mos);
1805 mutex_exit(&vd->vdev_dtl_lock);
1811 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1813 spa_t *spa = vd->vdev_spa;
1814 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1815 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1816 objset_t *mos = spa->spa_meta_objset;
1822 ASSERT(!vd->vdev_ishole);
1824 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1826 if (vd->vdev_detached) {
1827 if (smo->smo_object != 0) {
1828 VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx));
1829 smo->smo_object = 0;
1835 if (smo->smo_object == 0) {
1836 ASSERT(smo->smo_objsize == 0);
1837 ASSERT(smo->smo_alloc == 0);
1838 smo->smo_object = dmu_object_alloc(mos,
1839 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1840 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1841 ASSERT(smo->smo_object != 0);
1842 vdev_config_dirty(vd->vdev_top);
1845 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1847 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1850 mutex_enter(&smlock);
1852 mutex_enter(&vd->vdev_dtl_lock);
1853 space_map_walk(sm, space_map_add, &smsync);
1854 mutex_exit(&vd->vdev_dtl_lock);
1856 space_map_truncate(smo, mos, tx);
1857 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1859 space_map_destroy(&smsync);
1861 mutex_exit(&smlock);
1862 mutex_destroy(&smlock);
1864 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1865 dmu_buf_will_dirty(db, tx);
1866 ASSERT3U(db->db_size, >=, sizeof (*smo));
1867 bcopy(smo, db->db_data, sizeof (*smo));
1868 dmu_buf_rele(db, FTAG);
1874 * Determine whether the specified vdev can be offlined/detached/removed
1875 * without losing data.
1878 vdev_dtl_required(vdev_t *vd)
1880 spa_t *spa = vd->vdev_spa;
1881 vdev_t *tvd = vd->vdev_top;
1882 uint8_t cant_read = vd->vdev_cant_read;
1885 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1887 if (vd == spa->spa_root_vdev || vd == tvd)
1891 * Temporarily mark the device as unreadable, and then determine
1892 * whether this results in any DTL outages in the top-level vdev.
1893 * If not, we can safely offline/detach/remove the device.
1895 vd->vdev_cant_read = B_TRUE;
1896 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1897 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1898 vd->vdev_cant_read = cant_read;
1899 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1901 if (!required && zio_injection_enabled)
1902 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1908 * Determine if resilver is needed, and if so the txg range.
1911 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1913 boolean_t needed = B_FALSE;
1914 uint64_t thismin = UINT64_MAX;
1915 uint64_t thismax = 0;
1918 if (vd->vdev_children == 0) {
1919 mutex_enter(&vd->vdev_dtl_lock);
1920 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1921 vdev_writeable(vd)) {
1924 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1925 thismin = ss->ss_start - 1;
1926 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1927 thismax = ss->ss_end;
1930 mutex_exit(&vd->vdev_dtl_lock);
1932 for (c = 0; c < vd->vdev_children; c++) {
1933 vdev_t *cvd = vd->vdev_child[c];
1934 uint64_t cmin, cmax;
1936 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1937 thismin = MIN(thismin, cmin);
1938 thismax = MAX(thismax, cmax);
1944 if (needed && minp) {
1952 vdev_load(vdev_t *vd)
1957 * Recursively load all children.
1959 for (c = 0; c < vd->vdev_children; c++)
1960 vdev_load(vd->vdev_child[c]);
1963 * If this is a top-level vdev, initialize its metaslabs.
1965 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1966 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1967 vdev_metaslab_init(vd, 0) != 0))
1968 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1969 VDEV_AUX_CORRUPT_DATA);
1972 * If this is a leaf vdev, load its DTL.
1974 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1975 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1976 VDEV_AUX_CORRUPT_DATA);
1980 * The special vdev case is used for hot spares and l2cache devices. Its
1981 * sole purpose it to set the vdev state for the associated vdev. To do this,
1982 * we make sure that we can open the underlying device, then try to read the
1983 * label, and make sure that the label is sane and that it hasn't been
1984 * repurposed to another pool.
1987 vdev_validate_aux(vdev_t *vd)
1990 uint64_t guid, version;
1993 if (!vdev_readable(vd))
1996 if ((label = vdev_label_read_config(vd)) == NULL) {
1997 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1998 VDEV_AUX_CORRUPT_DATA);
2002 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2003 version > SPA_VERSION ||
2004 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2005 guid != vd->vdev_guid ||
2006 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2007 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2008 VDEV_AUX_CORRUPT_DATA);
2014 * We don't actually check the pool state here. If it's in fact in
2015 * use by another pool, we update this fact on the fly when requested.
2022 vdev_remove(vdev_t *vd, uint64_t txg)
2024 spa_t *spa = vd->vdev_spa;
2025 objset_t *mos = spa->spa_meta_objset;
2029 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2031 if (vd->vdev_dtl_smo.smo_object) {
2032 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
2033 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2034 vd->vdev_dtl_smo.smo_object = 0;
2037 if (vd->vdev_ms != NULL) {
2038 for (m = 0; m < vd->vdev_ms_count; m++) {
2039 metaslab_t *msp = vd->vdev_ms[m];
2041 if (msp == NULL || msp->ms_smo.smo_object == 0)
2044 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2045 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2046 msp->ms_smo.smo_object = 0;
2050 if (vd->vdev_ms_array) {
2051 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2052 vd->vdev_ms_array = 0;
2053 vd->vdev_ms_shift = 0;
2059 vdev_sync_done(vdev_t *vd, uint64_t txg)
2062 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2064 ASSERT(!vd->vdev_ishole);
2066 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
2067 metaslab_sync_done(msp, txg);
2070 metaslab_sync_reassess(vd->vdev_mg);
2074 vdev_sync(vdev_t *vd, uint64_t txg)
2076 spa_t *spa = vd->vdev_spa;
2081 ASSERT(!vd->vdev_ishole);
2083 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2084 ASSERT(vd == vd->vdev_top);
2085 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2086 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2087 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2088 ASSERT(vd->vdev_ms_array != 0);
2089 vdev_config_dirty(vd);
2094 * Remove the metadata associated with this vdev once it's empty.
2096 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2097 vdev_remove(vd, txg);
2099 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2100 metaslab_sync(msp, txg);
2101 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2104 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2105 vdev_dtl_sync(lvd, txg);
2107 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2111 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2113 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2117 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2118 * not be opened, and no I/O is attempted.
2121 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2125 spa_vdev_state_enter(spa, SCL_NONE);
2127 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2128 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2130 if (!vd->vdev_ops->vdev_op_leaf)
2131 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2136 * We don't directly use the aux state here, but if we do a
2137 * vdev_reopen(), we need this value to be present to remember why we
2140 vd->vdev_label_aux = aux;
2143 * Faulted state takes precedence over degraded.
2145 vd->vdev_delayed_close = B_FALSE;
2146 vd->vdev_faulted = 1ULL;
2147 vd->vdev_degraded = 0ULL;
2148 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2151 * If this device has the only valid copy of the data, then
2152 * back off and simply mark the vdev as degraded instead.
2154 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2155 vd->vdev_degraded = 1ULL;
2156 vd->vdev_faulted = 0ULL;
2159 * If we reopen the device and it's not dead, only then do we
2164 if (vdev_readable(vd))
2165 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2168 return (spa_vdev_state_exit(spa, vd, 0));
2172 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2173 * user that something is wrong. The vdev continues to operate as normal as far
2174 * as I/O is concerned.
2177 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2181 spa_vdev_state_enter(spa, SCL_NONE);
2183 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2184 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2186 if (!vd->vdev_ops->vdev_op_leaf)
2187 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2190 * If the vdev is already faulted, then don't do anything.
2192 if (vd->vdev_faulted || vd->vdev_degraded)
2193 return (spa_vdev_state_exit(spa, NULL, 0));
2195 vd->vdev_degraded = 1ULL;
2196 if (!vdev_is_dead(vd))
2197 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2200 return (spa_vdev_state_exit(spa, vd, 0));
2204 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2205 * any attached spare device should be detached when the device finishes
2206 * resilvering. Second, the online should be treated like a 'test' online case,
2207 * so no FMA events are generated if the device fails to open.
2210 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2212 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2214 spa_vdev_state_enter(spa, SCL_NONE);
2216 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2217 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2219 if (!vd->vdev_ops->vdev_op_leaf)
2220 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2223 vd->vdev_offline = B_FALSE;
2224 vd->vdev_tmpoffline = B_FALSE;
2225 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2226 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2228 /* XXX - L2ARC 1.0 does not support expansion */
2229 if (!vd->vdev_aux) {
2230 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2231 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2235 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2237 if (!vd->vdev_aux) {
2238 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2239 pvd->vdev_expanding = B_FALSE;
2243 *newstate = vd->vdev_state;
2244 if ((flags & ZFS_ONLINE_UNSPARE) &&
2245 !vdev_is_dead(vd) && vd->vdev_parent &&
2246 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2247 vd->vdev_parent->vdev_child[0] == vd)
2248 vd->vdev_unspare = B_TRUE;
2250 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2252 /* XXX - L2ARC 1.0 does not support expansion */
2254 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2255 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2257 return (spa_vdev_state_exit(spa, vd, 0));
2261 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2265 uint64_t generation;
2266 metaslab_group_t *mg;
2269 spa_vdev_state_enter(spa, SCL_ALLOC);
2271 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2272 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2274 if (!vd->vdev_ops->vdev_op_leaf)
2275 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2279 generation = spa->spa_config_generation + 1;
2282 * If the device isn't already offline, try to offline it.
2284 if (!vd->vdev_offline) {
2286 * If this device has the only valid copy of some data,
2287 * don't allow it to be offlined. Log devices are always
2290 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2291 vdev_dtl_required(vd))
2292 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2295 * If the top-level is a slog and it has had allocations
2296 * then proceed. We check that the vdev's metaslab group
2297 * is not NULL since it's possible that we may have just
2298 * added this vdev but not yet initialized its metaslabs.
2300 if (tvd->vdev_islog && mg != NULL) {
2302 * Prevent any future allocations.
2304 metaslab_group_passivate(mg);
2305 (void) spa_vdev_state_exit(spa, vd, 0);
2307 error = spa_offline_log(spa);
2309 spa_vdev_state_enter(spa, SCL_ALLOC);
2312 * Check to see if the config has changed.
2314 if (error || generation != spa->spa_config_generation) {
2315 metaslab_group_activate(mg);
2317 return (spa_vdev_state_exit(spa,
2319 (void) spa_vdev_state_exit(spa, vd, 0);
2322 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2326 * Offline this device and reopen its top-level vdev.
2327 * If the top-level vdev is a log device then just offline
2328 * it. Otherwise, if this action results in the top-level
2329 * vdev becoming unusable, undo it and fail the request.
2331 vd->vdev_offline = B_TRUE;
2334 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2335 vdev_is_dead(tvd)) {
2336 vd->vdev_offline = B_FALSE;
2338 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2342 * Add the device back into the metaslab rotor so that
2343 * once we online the device it's open for business.
2345 if (tvd->vdev_islog && mg != NULL)
2346 metaslab_group_activate(mg);
2349 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2351 return (spa_vdev_state_exit(spa, vd, 0));
2355 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2359 mutex_enter(&spa->spa_vdev_top_lock);
2360 error = vdev_offline_locked(spa, guid, flags);
2361 mutex_exit(&spa->spa_vdev_top_lock);
2367 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2368 * vdev_offline(), we assume the spa config is locked. We also clear all
2369 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2372 vdev_clear(spa_t *spa, vdev_t *vd)
2374 vdev_t *rvd = spa->spa_root_vdev;
2377 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2382 vd->vdev_stat.vs_read_errors = 0;
2383 vd->vdev_stat.vs_write_errors = 0;
2384 vd->vdev_stat.vs_checksum_errors = 0;
2386 for (c = 0; c < vd->vdev_children; c++)
2387 vdev_clear(spa, vd->vdev_child[c]);
2390 * If we're in the FAULTED state or have experienced failed I/O, then
2391 * clear the persistent state and attempt to reopen the device. We
2392 * also mark the vdev config dirty, so that the new faulted state is
2393 * written out to disk.
2395 if (vd->vdev_faulted || vd->vdev_degraded ||
2396 !vdev_readable(vd) || !vdev_writeable(vd)) {
2399 * When reopening in reponse to a clear event, it may be due to
2400 * a fmadm repair request. In this case, if the device is
2401 * still broken, we want to still post the ereport again.
2403 vd->vdev_forcefault = B_TRUE;
2405 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2406 vd->vdev_cant_read = B_FALSE;
2407 vd->vdev_cant_write = B_FALSE;
2409 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2411 vd->vdev_forcefault = B_FALSE;
2413 if (vd != rvd && vdev_writeable(vd->vdev_top))
2414 vdev_state_dirty(vd->vdev_top);
2416 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2417 spa_async_request(spa, SPA_ASYNC_RESILVER);
2419 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR);
2423 * When clearing a FMA-diagnosed fault, we always want to
2424 * unspare the device, as we assume that the original spare was
2425 * done in response to the FMA fault.
2427 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2428 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2429 vd->vdev_parent->vdev_child[0] == vd)
2430 vd->vdev_unspare = B_TRUE;
2434 vdev_is_dead(vdev_t *vd)
2437 * Holes and missing devices are always considered "dead".
2438 * This simplifies the code since we don't have to check for
2439 * these types of devices in the various code paths.
2440 * Instead we rely on the fact that we skip over dead devices
2441 * before issuing I/O to them.
2443 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2444 vd->vdev_ops == &vdev_missing_ops);
2448 vdev_readable(vdev_t *vd)
2450 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2454 vdev_writeable(vdev_t *vd)
2456 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2460 vdev_allocatable(vdev_t *vd)
2462 uint64_t state = vd->vdev_state;
2465 * We currently allow allocations from vdevs which may be in the
2466 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2467 * fails to reopen then we'll catch it later when we're holding
2468 * the proper locks. Note that we have to get the vdev state
2469 * in a local variable because although it changes atomically,
2470 * we're asking two separate questions about it.
2472 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2473 !vd->vdev_cant_write && !vd->vdev_ishole);
2477 vdev_accessible(vdev_t *vd, zio_t *zio)
2479 ASSERT(zio->io_vd == vd);
2481 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2484 if (zio->io_type == ZIO_TYPE_READ)
2485 return (!vd->vdev_cant_read);
2487 if (zio->io_type == ZIO_TYPE_WRITE)
2488 return (!vd->vdev_cant_write);
2494 * Get statistics for the given vdev.
2497 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2499 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2502 mutex_enter(&vd->vdev_stat_lock);
2503 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2504 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2505 vs->vs_state = vd->vdev_state;
2506 vs->vs_rsize = vdev_get_min_asize(vd);
2507 if (vd->vdev_ops->vdev_op_leaf)
2508 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2509 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2510 mutex_exit(&vd->vdev_stat_lock);
2513 * If we're getting stats on the root vdev, aggregate the I/O counts
2514 * over all top-level vdevs (i.e. the direct children of the root).
2517 for (c = 0; c < rvd->vdev_children; c++) {
2518 vdev_t *cvd = rvd->vdev_child[c];
2519 vdev_stat_t *cvs = &cvd->vdev_stat;
2521 mutex_enter(&vd->vdev_stat_lock);
2522 for (t = 0; t < ZIO_TYPES; t++) {
2523 vs->vs_ops[t] += cvs->vs_ops[t];
2524 vs->vs_bytes[t] += cvs->vs_bytes[t];
2526 cvs->vs_scan_removing = cvd->vdev_removing;
2527 mutex_exit(&vd->vdev_stat_lock);
2533 vdev_clear_stats(vdev_t *vd)
2535 mutex_enter(&vd->vdev_stat_lock);
2536 vd->vdev_stat.vs_space = 0;
2537 vd->vdev_stat.vs_dspace = 0;
2538 vd->vdev_stat.vs_alloc = 0;
2539 mutex_exit(&vd->vdev_stat_lock);
2543 vdev_scan_stat_init(vdev_t *vd)
2545 vdev_stat_t *vs = &vd->vdev_stat;
2548 for (c = 0; c < vd->vdev_children; c++)
2549 vdev_scan_stat_init(vd->vdev_child[c]);
2551 mutex_enter(&vd->vdev_stat_lock);
2552 vs->vs_scan_processed = 0;
2553 mutex_exit(&vd->vdev_stat_lock);
2557 vdev_stat_update(zio_t *zio, uint64_t psize)
2559 spa_t *spa = zio->io_spa;
2560 vdev_t *rvd = spa->spa_root_vdev;
2561 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2563 uint64_t txg = zio->io_txg;
2564 vdev_stat_t *vs = &vd->vdev_stat;
2565 zio_type_t type = zio->io_type;
2566 int flags = zio->io_flags;
2569 * If this i/o is a gang leader, it didn't do any actual work.
2571 if (zio->io_gang_tree)
2574 if (zio->io_error == 0) {
2576 * If this is a root i/o, don't count it -- we've already
2577 * counted the top-level vdevs, and vdev_get_stats() will
2578 * aggregate them when asked. This reduces contention on
2579 * the root vdev_stat_lock and implicitly handles blocks
2580 * that compress away to holes, for which there is no i/o.
2581 * (Holes never create vdev children, so all the counters
2582 * remain zero, which is what we want.)
2584 * Note: this only applies to successful i/o (io_error == 0)
2585 * because unlike i/o counts, errors are not additive.
2586 * When reading a ditto block, for example, failure of
2587 * one top-level vdev does not imply a root-level error.
2592 ASSERT(vd == zio->io_vd);
2594 if (flags & ZIO_FLAG_IO_BYPASS)
2597 mutex_enter(&vd->vdev_stat_lock);
2599 if (flags & ZIO_FLAG_IO_REPAIR) {
2600 if (flags & ZIO_FLAG_SCAN_THREAD) {
2601 dsl_scan_phys_t *scn_phys =
2602 &spa->spa_dsl_pool->dp_scan->scn_phys;
2603 uint64_t *processed = &scn_phys->scn_processed;
2606 if (vd->vdev_ops->vdev_op_leaf)
2607 atomic_add_64(processed, psize);
2608 vs->vs_scan_processed += psize;
2611 if (flags & ZIO_FLAG_SELF_HEAL)
2612 vs->vs_self_healed += psize;
2616 vs->vs_bytes[type] += psize;
2618 mutex_exit(&vd->vdev_stat_lock);
2622 if (flags & ZIO_FLAG_SPECULATIVE)
2626 * If this is an I/O error that is going to be retried, then ignore the
2627 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2628 * hard errors, when in reality they can happen for any number of
2629 * innocuous reasons (bus resets, MPxIO link failure, etc).
2631 if (zio->io_error == EIO &&
2632 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2636 * Intent logs writes won't propagate their error to the root
2637 * I/O so don't mark these types of failures as pool-level
2640 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2643 mutex_enter(&vd->vdev_stat_lock);
2644 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2645 if (zio->io_error == ECKSUM)
2646 vs->vs_checksum_errors++;
2648 vs->vs_read_errors++;
2650 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2651 vs->vs_write_errors++;
2652 mutex_exit(&vd->vdev_stat_lock);
2654 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2655 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2656 (flags & ZIO_FLAG_SCAN_THREAD) ||
2657 spa->spa_claiming)) {
2659 * This is either a normal write (not a repair), or it's
2660 * a repair induced by the scrub thread, or it's a repair
2661 * made by zil_claim() during spa_load() in the first txg.
2662 * In the normal case, we commit the DTL change in the same
2663 * txg as the block was born. In the scrub-induced repair
2664 * case, we know that scrubs run in first-pass syncing context,
2665 * so we commit the DTL change in spa_syncing_txg(spa).
2666 * In the zil_claim() case, we commit in spa_first_txg(spa).
2668 * We currently do not make DTL entries for failed spontaneous
2669 * self-healing writes triggered by normal (non-scrubbing)
2670 * reads, because we have no transactional context in which to
2671 * do so -- and it's not clear that it'd be desirable anyway.
2673 if (vd->vdev_ops->vdev_op_leaf) {
2674 uint64_t commit_txg = txg;
2675 if (flags & ZIO_FLAG_SCAN_THREAD) {
2676 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2677 ASSERT(spa_sync_pass(spa) == 1);
2678 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2679 commit_txg = spa_syncing_txg(spa);
2680 } else if (spa->spa_claiming) {
2681 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2682 commit_txg = spa_first_txg(spa);
2684 ASSERT(commit_txg >= spa_syncing_txg(spa));
2685 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2687 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2688 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2689 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2692 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2697 * Update the in-core space usage stats for this vdev, its metaslab class,
2698 * and the root vdev.
2701 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2702 int64_t space_delta)
2704 int64_t dspace_delta = space_delta;
2705 spa_t *spa = vd->vdev_spa;
2706 vdev_t *rvd = spa->spa_root_vdev;
2707 metaslab_group_t *mg = vd->vdev_mg;
2708 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2710 ASSERT(vd == vd->vdev_top);
2713 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2714 * factor. We must calculate this here and not at the root vdev
2715 * because the root vdev's psize-to-asize is simply the max of its
2716 * childrens', thus not accurate enough for us.
2718 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2719 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2720 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2721 vd->vdev_deflate_ratio;
2723 mutex_enter(&vd->vdev_stat_lock);
2724 vd->vdev_stat.vs_alloc += alloc_delta;
2725 vd->vdev_stat.vs_space += space_delta;
2726 vd->vdev_stat.vs_dspace += dspace_delta;
2727 mutex_exit(&vd->vdev_stat_lock);
2729 if (mc == spa_normal_class(spa)) {
2730 mutex_enter(&rvd->vdev_stat_lock);
2731 rvd->vdev_stat.vs_alloc += alloc_delta;
2732 rvd->vdev_stat.vs_space += space_delta;
2733 rvd->vdev_stat.vs_dspace += dspace_delta;
2734 mutex_exit(&rvd->vdev_stat_lock);
2738 ASSERT(rvd == vd->vdev_parent);
2739 ASSERT(vd->vdev_ms_count != 0);
2741 metaslab_class_space_update(mc,
2742 alloc_delta, defer_delta, space_delta, dspace_delta);
2747 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2748 * so that it will be written out next time the vdev configuration is synced.
2749 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2752 vdev_config_dirty(vdev_t *vd)
2754 spa_t *spa = vd->vdev_spa;
2755 vdev_t *rvd = spa->spa_root_vdev;
2758 ASSERT(spa_writeable(spa));
2761 * If this is an aux vdev (as with l2cache and spare devices), then we
2762 * update the vdev config manually and set the sync flag.
2764 if (vd->vdev_aux != NULL) {
2765 spa_aux_vdev_t *sav = vd->vdev_aux;
2769 for (c = 0; c < sav->sav_count; c++) {
2770 if (sav->sav_vdevs[c] == vd)
2774 if (c == sav->sav_count) {
2776 * We're being removed. There's nothing more to do.
2778 ASSERT(sav->sav_sync == B_TRUE);
2782 sav->sav_sync = B_TRUE;
2784 if (nvlist_lookup_nvlist_array(sav->sav_config,
2785 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2786 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2787 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2793 * Setting the nvlist in the middle if the array is a little
2794 * sketchy, but it will work.
2796 nvlist_free(aux[c]);
2797 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2803 * The dirty list is protected by the SCL_CONFIG lock. The caller
2804 * must either hold SCL_CONFIG as writer, or must be the sync thread
2805 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2806 * so this is sufficient to ensure mutual exclusion.
2808 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2809 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2810 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2813 for (c = 0; c < rvd->vdev_children; c++)
2814 vdev_config_dirty(rvd->vdev_child[c]);
2816 ASSERT(vd == vd->vdev_top);
2818 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2820 list_insert_head(&spa->spa_config_dirty_list, vd);
2825 vdev_config_clean(vdev_t *vd)
2827 spa_t *spa = vd->vdev_spa;
2829 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2830 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2831 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2833 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2834 list_remove(&spa->spa_config_dirty_list, vd);
2838 * Mark a top-level vdev's state as dirty, so that the next pass of
2839 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2840 * the state changes from larger config changes because they require
2841 * much less locking, and are often needed for administrative actions.
2844 vdev_state_dirty(vdev_t *vd)
2846 spa_t *spa = vd->vdev_spa;
2848 ASSERT(spa_writeable(spa));
2849 ASSERT(vd == vd->vdev_top);
2852 * The state list is protected by the SCL_STATE lock. The caller
2853 * must either hold SCL_STATE as writer, or must be the sync thread
2854 * (which holds SCL_STATE as reader). There's only one sync thread,
2855 * so this is sufficient to ensure mutual exclusion.
2857 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2858 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2859 spa_config_held(spa, SCL_STATE, RW_READER)));
2861 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2862 list_insert_head(&spa->spa_state_dirty_list, vd);
2866 vdev_state_clean(vdev_t *vd)
2868 spa_t *spa = vd->vdev_spa;
2870 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2871 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2872 spa_config_held(spa, SCL_STATE, RW_READER)));
2874 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2875 list_remove(&spa->spa_state_dirty_list, vd);
2879 * Propagate vdev state up from children to parent.
2882 vdev_propagate_state(vdev_t *vd)
2884 spa_t *spa = vd->vdev_spa;
2885 vdev_t *rvd = spa->spa_root_vdev;
2886 int degraded = 0, faulted = 0;
2891 if (vd->vdev_children > 0) {
2892 for (c = 0; c < vd->vdev_children; c++) {
2893 child = vd->vdev_child[c];
2896 * Don't factor holes into the decision.
2898 if (child->vdev_ishole)
2901 if (!vdev_readable(child) ||
2902 (!vdev_writeable(child) && spa_writeable(spa))) {
2904 * Root special: if there is a top-level log
2905 * device, treat the root vdev as if it were
2908 if (child->vdev_islog && vd == rvd)
2912 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2916 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2920 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2923 * Root special: if there is a top-level vdev that cannot be
2924 * opened due to corrupted metadata, then propagate the root
2925 * vdev's aux state as 'corrupt' rather than 'insufficient
2928 if (corrupted && vd == rvd &&
2929 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2930 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2931 VDEV_AUX_CORRUPT_DATA);
2934 if (vd->vdev_parent)
2935 vdev_propagate_state(vd->vdev_parent);
2939 * Set a vdev's state. If this is during an open, we don't update the parent
2940 * state, because we're in the process of opening children depth-first.
2941 * Otherwise, we propagate the change to the parent.
2943 * If this routine places a device in a faulted state, an appropriate ereport is
2947 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2949 uint64_t save_state;
2950 spa_t *spa = vd->vdev_spa;
2952 if (state == vd->vdev_state) {
2953 vd->vdev_stat.vs_aux = aux;
2957 save_state = vd->vdev_state;
2959 vd->vdev_state = state;
2960 vd->vdev_stat.vs_aux = aux;
2963 * If we are setting the vdev state to anything but an open state, then
2964 * always close the underlying device unless the device has requested
2965 * a delayed close (i.e. we're about to remove or fault the device).
2966 * Otherwise, we keep accessible but invalid devices open forever.
2967 * We don't call vdev_close() itself, because that implies some extra
2968 * checks (offline, etc) that we don't want here. This is limited to
2969 * leaf devices, because otherwise closing the device will affect other
2972 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2973 vd->vdev_ops->vdev_op_leaf)
2974 vd->vdev_ops->vdev_op_close(vd);
2977 * If we have brought this vdev back into service, we need
2978 * to notify fmd so that it can gracefully repair any outstanding
2979 * cases due to a missing device. We do this in all cases, even those
2980 * that probably don't correlate to a repaired fault. This is sure to
2981 * catch all cases, and we let the zfs-retire agent sort it out. If
2982 * this is a transient state it's OK, as the retire agent will
2983 * double-check the state of the vdev before repairing it.
2985 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2986 vd->vdev_prevstate != state)
2987 zfs_post_state_change(spa, vd);
2989 if (vd->vdev_removed &&
2990 state == VDEV_STATE_CANT_OPEN &&
2991 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2993 * If the previous state is set to VDEV_STATE_REMOVED, then this
2994 * device was previously marked removed and someone attempted to
2995 * reopen it. If this failed due to a nonexistent device, then
2996 * keep the device in the REMOVED state. We also let this be if
2997 * it is one of our special test online cases, which is only
2998 * attempting to online the device and shouldn't generate an FMA
3001 vd->vdev_state = VDEV_STATE_REMOVED;
3002 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3003 } else if (state == VDEV_STATE_REMOVED) {
3004 vd->vdev_removed = B_TRUE;
3005 } else if (state == VDEV_STATE_CANT_OPEN) {
3007 * If we fail to open a vdev during an import or recovery, we
3008 * mark it as "not available", which signifies that it was
3009 * never there to begin with. Failure to open such a device
3010 * is not considered an error.
3012 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3013 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3014 vd->vdev_ops->vdev_op_leaf)
3015 vd->vdev_not_present = 1;
3018 * Post the appropriate ereport. If the 'prevstate' field is
3019 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3020 * that this is part of a vdev_reopen(). In this case, we don't
3021 * want to post the ereport if the device was already in the
3022 * CANT_OPEN state beforehand.
3024 * If the 'checkremove' flag is set, then this is an attempt to
3025 * online the device in response to an insertion event. If we
3026 * hit this case, then we have detected an insertion event for a
3027 * faulted or offline device that wasn't in the removed state.
3028 * In this scenario, we don't post an ereport because we are
3029 * about to replace the device, or attempt an online with
3030 * vdev_forcefault, which will generate the fault for us.
3032 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3033 !vd->vdev_not_present && !vd->vdev_checkremove &&
3034 vd != spa->spa_root_vdev) {
3038 case VDEV_AUX_OPEN_FAILED:
3039 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3041 case VDEV_AUX_CORRUPT_DATA:
3042 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3044 case VDEV_AUX_NO_REPLICAS:
3045 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3047 case VDEV_AUX_BAD_GUID_SUM:
3048 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3050 case VDEV_AUX_TOO_SMALL:
3051 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3053 case VDEV_AUX_BAD_LABEL:
3054 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3057 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3060 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3063 /* Erase any notion of persistent removed state */
3064 vd->vdev_removed = B_FALSE;
3066 vd->vdev_removed = B_FALSE;
3069 if (!isopen && vd->vdev_parent)
3070 vdev_propagate_state(vd->vdev_parent);
3074 * Check the vdev configuration to ensure that it's capable of supporting
3078 vdev_is_bootable(vdev_t *vd)
3080 #if defined(__sun__) || defined(__sun)
3082 * Currently, we do not support RAID-Z or partial configuration.
3083 * In addition, only a single top-level vdev is allowed and none of the
3084 * leaves can be wholedisks.
3088 if (!vd->vdev_ops->vdev_op_leaf) {
3089 char *vdev_type = vd->vdev_ops->vdev_op_type;
3091 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3092 vd->vdev_children > 1) {
3094 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3095 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3098 } else if (vd->vdev_wholedisk == 1) {
3102 for (c = 0; c < vd->vdev_children; c++) {
3103 if (!vdev_is_bootable(vd->vdev_child[c]))
3106 #endif /* __sun__ || __sun */
3111 * Load the state from the original vdev tree (ovd) which
3112 * we've retrieved from the MOS config object. If the original
3113 * vdev was offline or faulted then we transfer that state to the
3114 * device in the current vdev tree (nvd).
3117 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3121 ASSERT(nvd->vdev_top->vdev_islog);
3122 ASSERT(spa_config_held(nvd->vdev_spa,
3123 SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3124 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3126 for (c = 0; c < nvd->vdev_children; c++)
3127 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3129 if (nvd->vdev_ops->vdev_op_leaf) {
3131 * Restore the persistent vdev state
3133 nvd->vdev_offline = ovd->vdev_offline;
3134 nvd->vdev_faulted = ovd->vdev_faulted;
3135 nvd->vdev_degraded = ovd->vdev_degraded;
3136 nvd->vdev_removed = ovd->vdev_removed;
3141 * Determine if a log device has valid content. If the vdev was
3142 * removed or faulted in the MOS config then we know that
3143 * the content on the log device has already been written to the pool.
3146 vdev_log_state_valid(vdev_t *vd)
3150 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3154 for (c = 0; c < vd->vdev_children; c++)
3155 if (vdev_log_state_valid(vd->vdev_child[c]))
3162 * Expand a vdev if possible.
3165 vdev_expand(vdev_t *vd, uint64_t txg)
3167 ASSERT(vd->vdev_top == vd);
3168 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3170 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3171 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3172 vdev_config_dirty(vd);
3180 vdev_split(vdev_t *vd)
3182 vdev_t *cvd, *pvd = vd->vdev_parent;
3184 vdev_remove_child(pvd, vd);
3185 vdev_compact_children(pvd);
3187 cvd = pvd->vdev_child[0];
3188 if (pvd->vdev_children == 1) {
3189 vdev_remove_parent(cvd);
3190 cvd->vdev_splitting = B_TRUE;
3192 vdev_propagate_state(cvd);
3195 #if defined(_KERNEL) && defined(HAVE_SPL)
3196 EXPORT_SYMBOL(vdev_fault);
3197 EXPORT_SYMBOL(vdev_degrade);
3198 EXPORT_SYMBOL(vdev_online);
3199 EXPORT_SYMBOL(vdev_offline);
3200 EXPORT_SYMBOL(vdev_clear);
3202 module_param(zfs_scrub_limit, int, 0644);
3203 MODULE_PARM_DESC(zfs_scrub_limit, "Max scrub/resilver I/O per leaf vdev");