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>
47 * Virtual device management.
50 static vdev_ops_t *vdev_ops_table[] = {
63 /* maximum scrub/resilver I/O queue per leaf vdev */
64 int zfs_scrub_limit = 10;
67 * Given a vdev type, return the appropriate ops vector.
70 vdev_getops(const char *type)
72 vdev_ops_t *ops, **opspp;
74 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
75 if (strcmp(ops->vdev_op_type, type) == 0)
82 * Default asize function: return the MAX of psize with the asize of
83 * all children. This is what's used by anything other than RAID-Z.
86 vdev_default_asize(vdev_t *vd, uint64_t psize)
88 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
92 for (c = 0; c < vd->vdev_children; c++) {
93 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
94 asize = MAX(asize, csize);
101 * Get the minimum allocatable size. We define the allocatable size as
102 * the vdev's asize rounded to the nearest metaslab. This allows us to
103 * replace or attach devices which don't have the same physical size but
104 * can still satisfy the same number of allocations.
107 vdev_get_min_asize(vdev_t *vd)
109 vdev_t *pvd = vd->vdev_parent;
112 * The our parent is NULL (inactive spare or cache) or is the root,
113 * just return our own asize.
116 return (vd->vdev_asize);
119 * The top-level vdev just returns the allocatable size rounded
120 * to the nearest metaslab.
122 if (vd == vd->vdev_top)
123 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
126 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
127 * so each child must provide at least 1/Nth of its asize.
129 if (pvd->vdev_ops == &vdev_raidz_ops)
130 return (pvd->vdev_min_asize / pvd->vdev_children);
132 return (pvd->vdev_min_asize);
136 vdev_set_min_asize(vdev_t *vd)
139 vd->vdev_min_asize = vdev_get_min_asize(vd);
141 for (c = 0; c < vd->vdev_children; c++)
142 vdev_set_min_asize(vd->vdev_child[c]);
146 vdev_lookup_top(spa_t *spa, uint64_t vdev)
148 vdev_t *rvd = spa->spa_root_vdev;
150 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
152 if (vdev < rvd->vdev_children) {
153 ASSERT(rvd->vdev_child[vdev] != NULL);
154 return (rvd->vdev_child[vdev]);
161 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
166 if (vd->vdev_guid == guid)
169 for (c = 0; c < vd->vdev_children; c++)
170 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
178 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
180 size_t oldsize, newsize;
181 uint64_t id = cvd->vdev_id;
184 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
185 ASSERT(cvd->vdev_parent == NULL);
187 cvd->vdev_parent = pvd;
192 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
194 oldsize = pvd->vdev_children * sizeof (vdev_t *);
195 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
196 newsize = pvd->vdev_children * sizeof (vdev_t *);
198 newchild = kmem_zalloc(newsize, KM_SLEEP);
199 if (pvd->vdev_child != NULL) {
200 bcopy(pvd->vdev_child, newchild, oldsize);
201 kmem_free(pvd->vdev_child, oldsize);
204 pvd->vdev_child = newchild;
205 pvd->vdev_child[id] = cvd;
207 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
208 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
211 * Walk up all ancestors to update guid sum.
213 for (; pvd != NULL; pvd = pvd->vdev_parent)
214 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
218 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
221 uint_t id = cvd->vdev_id;
223 ASSERT(cvd->vdev_parent == pvd);
228 ASSERT(id < pvd->vdev_children);
229 ASSERT(pvd->vdev_child[id] == cvd);
231 pvd->vdev_child[id] = NULL;
232 cvd->vdev_parent = NULL;
234 for (c = 0; c < pvd->vdev_children; c++)
235 if (pvd->vdev_child[c])
238 if (c == pvd->vdev_children) {
239 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
240 pvd->vdev_child = NULL;
241 pvd->vdev_children = 0;
245 * Walk up all ancestors to update guid sum.
247 for (; pvd != NULL; pvd = pvd->vdev_parent)
248 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
252 * Remove any holes in the child array.
255 vdev_compact_children(vdev_t *pvd)
257 vdev_t **newchild, *cvd;
258 int oldc = pvd->vdev_children;
262 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
264 for (c = newc = 0; c < oldc; c++)
265 if (pvd->vdev_child[c])
268 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
270 for (c = newc = 0; c < oldc; c++) {
271 if ((cvd = pvd->vdev_child[c]) != NULL) {
272 newchild[newc] = cvd;
273 cvd->vdev_id = newc++;
277 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
278 pvd->vdev_child = newchild;
279 pvd->vdev_children = newc;
283 * Allocate and minimally initialize a vdev_t.
286 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
291 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
293 if (spa->spa_root_vdev == NULL) {
294 ASSERT(ops == &vdev_root_ops);
295 spa->spa_root_vdev = vd;
296 spa->spa_load_guid = spa_generate_guid(NULL);
299 if (guid == 0 && ops != &vdev_hole_ops) {
300 if (spa->spa_root_vdev == vd) {
302 * The root vdev's guid will also be the pool guid,
303 * which must be unique among all pools.
305 guid = spa_generate_guid(NULL);
308 * Any other vdev's guid must be unique within the pool.
310 guid = spa_generate_guid(spa);
312 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
317 vd->vdev_guid = guid;
318 vd->vdev_guid_sum = guid;
320 vd->vdev_state = VDEV_STATE_CLOSED;
321 vd->vdev_ishole = (ops == &vdev_hole_ops);
323 list_link_init(&vd->vdev_config_dirty_node);
324 list_link_init(&vd->vdev_state_dirty_node);
325 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
326 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
327 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
328 for (t = 0; t < DTL_TYPES; t++) {
329 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
332 txg_list_create(&vd->vdev_ms_list,
333 offsetof(struct metaslab, ms_txg_node));
334 txg_list_create(&vd->vdev_dtl_list,
335 offsetof(struct vdev, vdev_dtl_node));
336 vd->vdev_stat.vs_timestamp = gethrtime();
344 * Allocate a new vdev. The 'alloctype' is used to control whether we are
345 * creating a new vdev or loading an existing one - the behavior is slightly
346 * different for each case.
349 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
354 uint64_t guid = 0, islog, nparity;
357 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
359 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
362 if ((ops = vdev_getops(type)) == NULL)
366 * If this is a load, get the vdev guid from the nvlist.
367 * Otherwise, vdev_alloc_common() will generate one for us.
369 if (alloctype == VDEV_ALLOC_LOAD) {
372 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
378 } else if (alloctype == VDEV_ALLOC_SPARE) {
379 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
381 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
382 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
384 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
385 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
390 * The first allocated vdev must be of type 'root'.
392 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
396 * Determine whether we're a log vdev.
399 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
400 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
403 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
407 * Set the nparity property for RAID-Z vdevs.
410 if (ops == &vdev_raidz_ops) {
411 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
413 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
416 * Previous versions could only support 1 or 2 parity
420 spa_version(spa) < SPA_VERSION_RAIDZ2)
423 spa_version(spa) < SPA_VERSION_RAIDZ3)
427 * We require the parity to be specified for SPAs that
428 * support multiple parity levels.
430 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
433 * Otherwise, we default to 1 parity device for RAID-Z.
440 ASSERT(nparity != -1ULL);
442 vd = vdev_alloc_common(spa, id, guid, ops);
444 vd->vdev_islog = islog;
445 vd->vdev_nparity = nparity;
447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
448 vd->vdev_path = spa_strdup(vd->vdev_path);
449 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
450 vd->vdev_devid = spa_strdup(vd->vdev_devid);
451 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
452 &vd->vdev_physpath) == 0)
453 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
454 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
455 vd->vdev_fru = spa_strdup(vd->vdev_fru);
458 * Set the whole_disk property. If it's not specified, leave the value
461 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
462 &vd->vdev_wholedisk) != 0)
463 vd->vdev_wholedisk = -1ULL;
466 * Look for the 'not present' flag. This will only be set if the device
467 * was not present at the time of import.
469 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
470 &vd->vdev_not_present);
473 * Get the alignment requirement.
475 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
478 * Retrieve the vdev creation time.
480 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
484 * If we're a top-level vdev, try to load the allocation parameters.
486 if (parent && !parent->vdev_parent &&
487 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
488 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
490 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
494 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
498 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
499 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
500 alloctype == VDEV_ALLOC_ADD ||
501 alloctype == VDEV_ALLOC_SPLIT ||
502 alloctype == VDEV_ALLOC_ROOTPOOL);
503 vd->vdev_mg = metaslab_group_create(islog ?
504 spa_log_class(spa) : spa_normal_class(spa), vd);
508 * If we're a leaf vdev, try to load the DTL object and other state.
510 if (vd->vdev_ops->vdev_op_leaf &&
511 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
512 alloctype == VDEV_ALLOC_ROOTPOOL)) {
513 if (alloctype == VDEV_ALLOC_LOAD) {
514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
515 &vd->vdev_dtl_smo.smo_object);
516 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
520 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
523 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
524 &spare) == 0 && spare)
528 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
531 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
532 &vd->vdev_resilvering);
535 * When importing a pool, we want to ignore the persistent fault
536 * state, as the diagnosis made on another system may not be
537 * valid in the current context. Local vdevs will
538 * remain in the faulted state.
540 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
541 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
543 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
545 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
548 if (vd->vdev_faulted || vd->vdev_degraded) {
552 VDEV_AUX_ERR_EXCEEDED;
553 if (nvlist_lookup_string(nv,
554 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
555 strcmp(aux, "external") == 0)
556 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
562 * Add ourselves to the parent's list of children.
564 vdev_add_child(parent, vd);
572 vdev_free(vdev_t *vd)
575 spa_t *spa = vd->vdev_spa;
578 * vdev_free() implies closing the vdev first. This is simpler than
579 * trying to ensure complicated semantics for all callers.
583 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
584 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
589 for (c = 0; c < vd->vdev_children; c++)
590 vdev_free(vd->vdev_child[c]);
592 ASSERT(vd->vdev_child == NULL);
593 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
596 * Discard allocation state.
598 if (vd->vdev_mg != NULL) {
599 vdev_metaslab_fini(vd);
600 metaslab_group_destroy(vd->vdev_mg);
603 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
604 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
605 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
608 * Remove this vdev from its parent's child list.
610 vdev_remove_child(vd->vdev_parent, vd);
612 ASSERT(vd->vdev_parent == NULL);
615 * Clean up vdev structure.
621 spa_strfree(vd->vdev_path);
623 spa_strfree(vd->vdev_devid);
624 if (vd->vdev_physpath)
625 spa_strfree(vd->vdev_physpath);
627 spa_strfree(vd->vdev_fru);
629 if (vd->vdev_isspare)
630 spa_spare_remove(vd);
631 if (vd->vdev_isl2cache)
632 spa_l2cache_remove(vd);
634 txg_list_destroy(&vd->vdev_ms_list);
635 txg_list_destroy(&vd->vdev_dtl_list);
637 mutex_enter(&vd->vdev_dtl_lock);
638 for (t = 0; t < DTL_TYPES; t++) {
639 space_map_unload(&vd->vdev_dtl[t]);
640 space_map_destroy(&vd->vdev_dtl[t]);
642 mutex_exit(&vd->vdev_dtl_lock);
644 mutex_destroy(&vd->vdev_dtl_lock);
645 mutex_destroy(&vd->vdev_stat_lock);
646 mutex_destroy(&vd->vdev_probe_lock);
648 if (vd == spa->spa_root_vdev)
649 spa->spa_root_vdev = NULL;
651 kmem_free(vd, sizeof (vdev_t));
655 * Transfer top-level vdev state from svd to tvd.
658 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
660 spa_t *spa = svd->vdev_spa;
665 ASSERT(tvd == tvd->vdev_top);
667 tvd->vdev_ms_array = svd->vdev_ms_array;
668 tvd->vdev_ms_shift = svd->vdev_ms_shift;
669 tvd->vdev_ms_count = svd->vdev_ms_count;
671 svd->vdev_ms_array = 0;
672 svd->vdev_ms_shift = 0;
673 svd->vdev_ms_count = 0;
676 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
677 tvd->vdev_mg = svd->vdev_mg;
678 tvd->vdev_ms = svd->vdev_ms;
683 if (tvd->vdev_mg != NULL)
684 tvd->vdev_mg->mg_vd = tvd;
686 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
687 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
688 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
690 svd->vdev_stat.vs_alloc = 0;
691 svd->vdev_stat.vs_space = 0;
692 svd->vdev_stat.vs_dspace = 0;
694 for (t = 0; t < TXG_SIZE; t++) {
695 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
696 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
697 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
698 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
699 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
700 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
703 if (list_link_active(&svd->vdev_config_dirty_node)) {
704 vdev_config_clean(svd);
705 vdev_config_dirty(tvd);
708 if (list_link_active(&svd->vdev_state_dirty_node)) {
709 vdev_state_clean(svd);
710 vdev_state_dirty(tvd);
713 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
714 svd->vdev_deflate_ratio = 0;
716 tvd->vdev_islog = svd->vdev_islog;
721 vdev_top_update(vdev_t *tvd, vdev_t *vd)
730 for (c = 0; c < vd->vdev_children; c++)
731 vdev_top_update(tvd, vd->vdev_child[c]);
735 * Add a mirror/replacing vdev above an existing vdev.
738 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
740 spa_t *spa = cvd->vdev_spa;
741 vdev_t *pvd = cvd->vdev_parent;
744 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
746 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
748 mvd->vdev_asize = cvd->vdev_asize;
749 mvd->vdev_min_asize = cvd->vdev_min_asize;
750 mvd->vdev_ashift = cvd->vdev_ashift;
751 mvd->vdev_state = cvd->vdev_state;
752 mvd->vdev_crtxg = cvd->vdev_crtxg;
754 vdev_remove_child(pvd, cvd);
755 vdev_add_child(pvd, mvd);
756 cvd->vdev_id = mvd->vdev_children;
757 vdev_add_child(mvd, cvd);
758 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
760 if (mvd == mvd->vdev_top)
761 vdev_top_transfer(cvd, mvd);
767 * Remove a 1-way mirror/replacing vdev from the tree.
770 vdev_remove_parent(vdev_t *cvd)
772 vdev_t *mvd = cvd->vdev_parent;
773 vdev_t *pvd = mvd->vdev_parent;
775 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
777 ASSERT(mvd->vdev_children == 1);
778 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
779 mvd->vdev_ops == &vdev_replacing_ops ||
780 mvd->vdev_ops == &vdev_spare_ops);
781 cvd->vdev_ashift = mvd->vdev_ashift;
783 vdev_remove_child(mvd, cvd);
784 vdev_remove_child(pvd, mvd);
787 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
788 * Otherwise, we could have detached an offline device, and when we
789 * go to import the pool we'll think we have two top-level vdevs,
790 * instead of a different version of the same top-level vdev.
792 if (mvd->vdev_top == mvd) {
793 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
794 cvd->vdev_orig_guid = cvd->vdev_guid;
795 cvd->vdev_guid += guid_delta;
796 cvd->vdev_guid_sum += guid_delta;
798 cvd->vdev_id = mvd->vdev_id;
799 vdev_add_child(pvd, cvd);
800 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
802 if (cvd == cvd->vdev_top)
803 vdev_top_transfer(mvd, cvd);
805 ASSERT(mvd->vdev_children == 0);
810 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
812 spa_t *spa = vd->vdev_spa;
813 objset_t *mos = spa->spa_meta_objset;
815 uint64_t oldc = vd->vdev_ms_count;
816 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
820 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
823 * This vdev is not being allocated from yet or is a hole.
825 if (vd->vdev_ms_shift == 0)
828 ASSERT(!vd->vdev_ishole);
831 * Compute the raidz-deflation ratio. Note, we hard-code
832 * in 128k (1 << 17) because it is the current "typical" blocksize.
833 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
834 * or we will inconsistently account for existing bp's.
836 vd->vdev_deflate_ratio = (1 << 17) /
837 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
839 ASSERT(oldc <= newc);
841 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP | KM_NODEBUG);
844 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
845 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
849 vd->vdev_ms_count = newc;
851 for (m = oldc; m < newc; m++) {
852 space_map_obj_t smo = { 0, 0, 0 };
855 error = dmu_read(mos, vd->vdev_ms_array,
856 m * sizeof (uint64_t), sizeof (uint64_t), &object,
862 error = dmu_bonus_hold(mos, object, FTAG, &db);
865 ASSERT3U(db->db_size, >=, sizeof (smo));
866 bcopy(db->db_data, &smo, sizeof (smo));
867 ASSERT3U(smo.smo_object, ==, object);
868 dmu_buf_rele(db, FTAG);
871 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
872 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
876 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
879 * If the vdev is being removed we don't activate
880 * the metaslabs since we want to ensure that no new
881 * allocations are performed on this device.
883 if (oldc == 0 && !vd->vdev_removing)
884 metaslab_group_activate(vd->vdev_mg);
887 spa_config_exit(spa, SCL_ALLOC, FTAG);
893 vdev_metaslab_fini(vdev_t *vd)
896 uint64_t count = vd->vdev_ms_count;
898 if (vd->vdev_ms != NULL) {
899 metaslab_group_passivate(vd->vdev_mg);
900 for (m = 0; m < count; m++)
901 if (vd->vdev_ms[m] != NULL)
902 metaslab_fini(vd->vdev_ms[m]);
903 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
908 typedef struct vdev_probe_stats {
909 boolean_t vps_readable;
910 boolean_t vps_writeable;
912 } vdev_probe_stats_t;
915 vdev_probe_done(zio_t *zio)
917 spa_t *spa = zio->io_spa;
918 vdev_t *vd = zio->io_vd;
919 vdev_probe_stats_t *vps = zio->io_private;
921 ASSERT(vd->vdev_probe_zio != NULL);
923 if (zio->io_type == ZIO_TYPE_READ) {
924 if (zio->io_error == 0)
925 vps->vps_readable = 1;
926 if (zio->io_error == 0 && spa_writeable(spa)) {
927 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
928 zio->io_offset, zio->io_size, zio->io_data,
929 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
930 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
932 zio_buf_free(zio->io_data, zio->io_size);
934 } else if (zio->io_type == ZIO_TYPE_WRITE) {
935 if (zio->io_error == 0)
936 vps->vps_writeable = 1;
937 zio_buf_free(zio->io_data, zio->io_size);
938 } else if (zio->io_type == ZIO_TYPE_NULL) {
941 vd->vdev_cant_read |= !vps->vps_readable;
942 vd->vdev_cant_write |= !vps->vps_writeable;
944 if (vdev_readable(vd) &&
945 (vdev_writeable(vd) || !spa_writeable(spa))) {
948 ASSERT(zio->io_error != 0);
949 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
950 spa, vd, NULL, 0, 0);
951 zio->io_error = ENXIO;
954 mutex_enter(&vd->vdev_probe_lock);
955 ASSERT(vd->vdev_probe_zio == zio);
956 vd->vdev_probe_zio = NULL;
957 mutex_exit(&vd->vdev_probe_lock);
959 while ((pio = zio_walk_parents(zio)) != NULL)
960 if (!vdev_accessible(vd, pio))
961 pio->io_error = ENXIO;
963 kmem_free(vps, sizeof (*vps));
968 * Determine whether this device is accessible by reading and writing
969 * to several known locations: the pad regions of each vdev label
970 * but the first (which we leave alone in case it contains a VTOC).
973 vdev_probe(vdev_t *vd, zio_t *zio)
975 spa_t *spa = vd->vdev_spa;
976 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 (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)
1078 * Stacking zpools on top of zvols is unsupported until we implement a method
1079 * for determining if an arbitrary block device is a zvol without using the
1080 * path. Solaris would check the 'zvol' path component but this does not
1081 * exist in the Linux port, so we really should do something like stat the
1082 * file and check the major number. This is complicated by the fact that
1083 * we need to do this portably in user or kernel space.
1088 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1089 strlen(ZVOL_DIR)) == 0)
1091 for (c = 0; c < vd->vdev_children; c++)
1092 if (vdev_uses_zvols(vd->vdev_child[c]))
1099 vdev_open_children(vdev_t *vd)
1102 int children = vd->vdev_children;
1106 * in order to handle pools on top of zvols, do the opens
1107 * in a single thread so that the same thread holds the
1108 * spa_namespace_lock
1110 if (vdev_uses_zvols(vd)) {
1111 for (c = 0; c < children; c++)
1112 vd->vdev_child[c]->vdev_open_error =
1113 vdev_open(vd->vdev_child[c]);
1116 tq = taskq_create("vdev_open", children, minclsyspri,
1117 children, children, TASKQ_PREPOPULATE);
1119 for (c = 0; c < children; c++)
1120 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1127 * Prepare a virtual device for access.
1130 vdev_open(vdev_t *vd)
1132 spa_t *spa = vd->vdev_spa;
1135 uint64_t asize, psize;
1136 uint64_t ashift = 0;
1139 ASSERT(vd->vdev_open_thread == curthread ||
1140 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1141 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1142 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1143 vd->vdev_state == VDEV_STATE_OFFLINE);
1145 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1146 vd->vdev_cant_read = B_FALSE;
1147 vd->vdev_cant_write = B_FALSE;
1148 vd->vdev_min_asize = vdev_get_min_asize(vd);
1151 * If this vdev is not removed, check its fault status. If it's
1152 * faulted, bail out of the open.
1154 if (!vd->vdev_removed && vd->vdev_faulted) {
1155 ASSERT(vd->vdev_children == 0);
1156 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1157 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1158 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1159 vd->vdev_label_aux);
1161 } else if (vd->vdev_offline) {
1162 ASSERT(vd->vdev_children == 0);
1163 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1167 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1170 * Reset the vdev_reopening flag so that we actually close
1171 * the vdev on error.
1173 vd->vdev_reopening = B_FALSE;
1174 if (zio_injection_enabled && error == 0)
1175 error = zio_handle_device_injection(vd, NULL, ENXIO);
1178 if (vd->vdev_removed &&
1179 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1180 vd->vdev_removed = B_FALSE;
1182 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1183 vd->vdev_stat.vs_aux);
1187 vd->vdev_removed = B_FALSE;
1190 * Recheck the faulted flag now that we have confirmed that
1191 * the vdev is accessible. If we're faulted, bail.
1193 if (vd->vdev_faulted) {
1194 ASSERT(vd->vdev_children == 0);
1195 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1196 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1197 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1198 vd->vdev_label_aux);
1202 if (vd->vdev_degraded) {
1203 ASSERT(vd->vdev_children == 0);
1204 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1205 VDEV_AUX_ERR_EXCEEDED);
1207 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1211 * For hole or missing vdevs we just return success.
1213 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1216 for (c = 0; c < vd->vdev_children; c++) {
1217 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1218 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1224 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1226 if (vd->vdev_children == 0) {
1227 if (osize < SPA_MINDEVSIZE) {
1228 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1229 VDEV_AUX_TOO_SMALL);
1233 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1235 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1236 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1237 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1238 VDEV_AUX_TOO_SMALL);
1245 vd->vdev_psize = psize;
1248 * Make sure the allocatable size hasn't shrunk.
1250 if (asize < vd->vdev_min_asize) {
1251 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1252 VDEV_AUX_BAD_LABEL);
1256 if (vd->vdev_asize == 0) {
1258 * This is the first-ever open, so use the computed values.
1259 * For testing purposes, a higher ashift can be requested.
1261 vd->vdev_asize = asize;
1262 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1265 * Make sure the alignment requirement hasn't increased.
1267 if (ashift > vd->vdev_top->vdev_ashift) {
1268 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1269 VDEV_AUX_BAD_LABEL);
1275 * If all children are healthy and the asize has increased,
1276 * then we've experienced dynamic LUN growth. If automatic
1277 * expansion is enabled then use the additional space.
1279 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1280 (vd->vdev_expanding || spa->spa_autoexpand))
1281 vd->vdev_asize = asize;
1283 vdev_set_min_asize(vd);
1286 * Ensure we can issue some IO before declaring the
1287 * vdev open for business.
1289 if (vd->vdev_ops->vdev_op_leaf &&
1290 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1291 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1292 VDEV_AUX_ERR_EXCEEDED);
1297 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1298 * resilver. But don't do this if we are doing a reopen for a scrub,
1299 * since this would just restart the scrub we are already doing.
1301 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1302 vdev_resilver_needed(vd, NULL, NULL))
1303 spa_async_request(spa, SPA_ASYNC_RESILVER);
1309 * Called once the vdevs are all opened, this routine validates the label
1310 * contents. This needs to be done before vdev_load() so that we don't
1311 * inadvertently do repair I/Os to the wrong device.
1313 * This function will only return failure if one of the vdevs indicates that it
1314 * has since been destroyed or exported. This is only possible if
1315 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1316 * will be updated but the function will return 0.
1319 vdev_validate(vdev_t *vd)
1321 spa_t *spa = vd->vdev_spa;
1323 uint64_t guid = 0, top_guid;
1327 for (c = 0; c < vd->vdev_children; c++)
1328 if (vdev_validate(vd->vdev_child[c]) != 0)
1332 * If the device has already failed, or was marked offline, don't do
1333 * any further validation. Otherwise, label I/O will fail and we will
1334 * overwrite the previous state.
1336 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1337 uint64_t aux_guid = 0;
1340 if ((label = vdev_label_read_config(vd)) == NULL) {
1341 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1342 VDEV_AUX_BAD_LABEL);
1347 * Determine if this vdev has been split off into another
1348 * pool. If so, then refuse to open it.
1350 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1351 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1352 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1353 VDEV_AUX_SPLIT_POOL);
1358 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1359 &guid) != 0 || guid != spa_guid(spa)) {
1360 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1361 VDEV_AUX_CORRUPT_DATA);
1366 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1367 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1372 * If this vdev just became a top-level vdev because its
1373 * sibling was detached, it will have adopted the parent's
1374 * vdev guid -- but the label may or may not be on disk yet.
1375 * Fortunately, either version of the label will have the
1376 * same top guid, so if we're a top-level vdev, we can
1377 * safely compare to that instead.
1379 * If we split this vdev off instead, then we also check the
1380 * original pool's guid. We don't want to consider the vdev
1381 * corrupt if it is partway through a split operation.
1383 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1385 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1387 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1388 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1389 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1390 VDEV_AUX_CORRUPT_DATA);
1395 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1397 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1398 VDEV_AUX_CORRUPT_DATA);
1406 * If this is a verbatim import, no need to check the
1407 * state of the pool.
1409 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1410 spa_load_state(spa) == SPA_LOAD_OPEN &&
1411 state != POOL_STATE_ACTIVE)
1415 * If we were able to open and validate a vdev that was
1416 * previously marked permanently unavailable, clear that state
1419 if (vd->vdev_not_present)
1420 vd->vdev_not_present = 0;
1427 * Close a virtual device.
1430 vdev_close(vdev_t *vd)
1432 vdev_t *pvd = vd->vdev_parent;
1433 ASSERTV(spa_t *spa = vd->vdev_spa);
1435 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1438 * If our parent is reopening, then we are as well, unless we are
1441 if (pvd != NULL && pvd->vdev_reopening)
1442 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1444 vd->vdev_ops->vdev_op_close(vd);
1446 vdev_cache_purge(vd);
1449 * We record the previous state before we close it, so that if we are
1450 * doing a reopen(), we don't generate FMA ereports if we notice that
1451 * it's still faulted.
1453 vd->vdev_prevstate = vd->vdev_state;
1455 if (vd->vdev_offline)
1456 vd->vdev_state = VDEV_STATE_OFFLINE;
1458 vd->vdev_state = VDEV_STATE_CLOSED;
1459 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1463 vdev_hold(vdev_t *vd)
1465 spa_t *spa = vd->vdev_spa;
1468 ASSERT(spa_is_root(spa));
1469 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1472 for (c = 0; c < vd->vdev_children; c++)
1473 vdev_hold(vd->vdev_child[c]);
1475 if (vd->vdev_ops->vdev_op_leaf)
1476 vd->vdev_ops->vdev_op_hold(vd);
1480 vdev_rele(vdev_t *vd)
1484 ASSERT(spa_is_root(vd->vdev_spa));
1485 for (c = 0; c < vd->vdev_children; c++)
1486 vdev_rele(vd->vdev_child[c]);
1488 if (vd->vdev_ops->vdev_op_leaf)
1489 vd->vdev_ops->vdev_op_rele(vd);
1493 * Reopen all interior vdevs and any unopened leaves. We don't actually
1494 * reopen leaf vdevs which had previously been opened as they might deadlock
1495 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1496 * If the leaf has never been opened then open it, as usual.
1499 vdev_reopen(vdev_t *vd)
1501 spa_t *spa = vd->vdev_spa;
1503 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1505 /* set the reopening flag unless we're taking the vdev offline */
1506 vd->vdev_reopening = !vd->vdev_offline;
1508 (void) vdev_open(vd);
1511 * Call vdev_validate() here to make sure we have the same device.
1512 * Otherwise, a device with an invalid label could be successfully
1513 * opened in response to vdev_reopen().
1516 (void) vdev_validate_aux(vd);
1517 if (vdev_readable(vd) && vdev_writeable(vd) &&
1518 vd->vdev_aux == &spa->spa_l2cache &&
1519 !l2arc_vdev_present(vd))
1520 l2arc_add_vdev(spa, vd);
1522 (void) vdev_validate(vd);
1526 * Reassess parent vdev's health.
1528 vdev_propagate_state(vd);
1532 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1537 * Normally, partial opens (e.g. of a mirror) are allowed.
1538 * For a create, however, we want to fail the request if
1539 * there are any components we can't open.
1541 error = vdev_open(vd);
1543 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1545 return (error ? error : ENXIO);
1549 * Recursively initialize all labels.
1551 if ((error = vdev_label_init(vd, txg, isreplacing ?
1552 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1561 vdev_metaslab_set_size(vdev_t *vd)
1564 * Aim for roughly 200 metaslabs per vdev.
1566 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1567 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1571 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1573 ASSERT(vd == vd->vdev_top);
1574 ASSERT(!vd->vdev_ishole);
1575 ASSERT(ISP2(flags));
1576 ASSERT(spa_writeable(vd->vdev_spa));
1578 if (flags & VDD_METASLAB)
1579 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1581 if (flags & VDD_DTL)
1582 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1584 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1590 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1591 * the vdev has less than perfect replication. There are four kinds of DTL:
1593 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1595 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1597 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1598 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1599 * txgs that was scrubbed.
1601 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1602 * persistent errors or just some device being offline.
1603 * Unlike the other three, the DTL_OUTAGE map is not generally
1604 * maintained; it's only computed when needed, typically to
1605 * determine whether a device can be detached.
1607 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1608 * either has the data or it doesn't.
1610 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1611 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1612 * if any child is less than fully replicated, then so is its parent.
1613 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1614 * comprising only those txgs which appear in 'maxfaults' or more children;
1615 * those are the txgs we don't have enough replication to read. For example,
1616 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1617 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1618 * two child DTL_MISSING maps.
1620 * It should be clear from the above that to compute the DTLs and outage maps
1621 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1622 * Therefore, that is all we keep on disk. When loading the pool, or after
1623 * a configuration change, we generate all other DTLs from first principles.
1626 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1628 space_map_t *sm = &vd->vdev_dtl[t];
1630 ASSERT(t < DTL_TYPES);
1631 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1632 ASSERT(spa_writeable(vd->vdev_spa));
1634 mutex_enter(sm->sm_lock);
1635 if (!space_map_contains(sm, txg, size))
1636 space_map_add(sm, txg, size);
1637 mutex_exit(sm->sm_lock);
1641 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1643 space_map_t *sm = &vd->vdev_dtl[t];
1644 boolean_t dirty = B_FALSE;
1646 ASSERT(t < DTL_TYPES);
1647 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1649 mutex_enter(sm->sm_lock);
1650 if (sm->sm_space != 0)
1651 dirty = space_map_contains(sm, txg, size);
1652 mutex_exit(sm->sm_lock);
1658 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1660 space_map_t *sm = &vd->vdev_dtl[t];
1663 mutex_enter(sm->sm_lock);
1664 empty = (sm->sm_space == 0);
1665 mutex_exit(sm->sm_lock);
1671 * Reassess DTLs after a config change or scrub completion.
1674 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1676 spa_t *spa = vd->vdev_spa;
1680 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1682 for (c = 0; c < vd->vdev_children; c++)
1683 vdev_dtl_reassess(vd->vdev_child[c], txg,
1684 scrub_txg, scrub_done);
1686 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1689 if (vd->vdev_ops->vdev_op_leaf) {
1690 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1692 mutex_enter(&vd->vdev_dtl_lock);
1693 if (scrub_txg != 0 &&
1694 (spa->spa_scrub_started ||
1695 (scn && scn->scn_phys.scn_errors == 0))) {
1697 * We completed a scrub up to scrub_txg. If we
1698 * did it without rebooting, then the scrub dtl
1699 * will be valid, so excise the old region and
1700 * fold in the scrub dtl. Otherwise, leave the
1701 * dtl as-is if there was an error.
1703 * There's little trick here: to excise the beginning
1704 * of the DTL_MISSING map, we put it into a reference
1705 * tree and then add a segment with refcnt -1 that
1706 * covers the range [0, scrub_txg). This means
1707 * that each txg in that range has refcnt -1 or 0.
1708 * We then add DTL_SCRUB with a refcnt of 2, so that
1709 * entries in the range [0, scrub_txg) will have a
1710 * positive refcnt -- either 1 or 2. We then convert
1711 * the reference tree into the new DTL_MISSING map.
1713 space_map_ref_create(&reftree);
1714 space_map_ref_add_map(&reftree,
1715 &vd->vdev_dtl[DTL_MISSING], 1);
1716 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1717 space_map_ref_add_map(&reftree,
1718 &vd->vdev_dtl[DTL_SCRUB], 2);
1719 space_map_ref_generate_map(&reftree,
1720 &vd->vdev_dtl[DTL_MISSING], 1);
1721 space_map_ref_destroy(&reftree);
1723 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1724 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1725 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1727 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1728 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1729 if (!vdev_readable(vd))
1730 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1732 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1733 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1734 mutex_exit(&vd->vdev_dtl_lock);
1737 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1741 mutex_enter(&vd->vdev_dtl_lock);
1742 for (t = 0; t < DTL_TYPES; t++) {
1743 /* account for child's outage in parent's missing map */
1744 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1746 continue; /* leaf vdevs only */
1747 if (t == DTL_PARTIAL)
1748 minref = 1; /* i.e. non-zero */
1749 else if (vd->vdev_nparity != 0)
1750 minref = vd->vdev_nparity + 1; /* RAID-Z */
1752 minref = vd->vdev_children; /* any kind of mirror */
1753 space_map_ref_create(&reftree);
1754 for (c = 0; c < vd->vdev_children; c++) {
1755 vdev_t *cvd = vd->vdev_child[c];
1756 mutex_enter(&cvd->vdev_dtl_lock);
1757 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1758 mutex_exit(&cvd->vdev_dtl_lock);
1760 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1761 space_map_ref_destroy(&reftree);
1763 mutex_exit(&vd->vdev_dtl_lock);
1767 vdev_dtl_load(vdev_t *vd)
1769 spa_t *spa = vd->vdev_spa;
1770 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1771 objset_t *mos = spa->spa_meta_objset;
1775 ASSERT(vd->vdev_children == 0);
1777 if (smo->smo_object == 0)
1780 ASSERT(!vd->vdev_ishole);
1782 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1785 ASSERT3U(db->db_size, >=, sizeof (*smo));
1786 bcopy(db->db_data, smo, sizeof (*smo));
1787 dmu_buf_rele(db, FTAG);
1789 mutex_enter(&vd->vdev_dtl_lock);
1790 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1791 NULL, SM_ALLOC, smo, mos);
1792 mutex_exit(&vd->vdev_dtl_lock);
1798 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1800 spa_t *spa = vd->vdev_spa;
1801 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1802 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1803 objset_t *mos = spa->spa_meta_objset;
1809 ASSERT(!vd->vdev_ishole);
1811 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1813 if (vd->vdev_detached) {
1814 if (smo->smo_object != 0) {
1815 VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx));
1816 smo->smo_object = 0;
1822 if (smo->smo_object == 0) {
1823 ASSERT(smo->smo_objsize == 0);
1824 ASSERT(smo->smo_alloc == 0);
1825 smo->smo_object = dmu_object_alloc(mos,
1826 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1827 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1828 ASSERT(smo->smo_object != 0);
1829 vdev_config_dirty(vd->vdev_top);
1832 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1834 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1837 mutex_enter(&smlock);
1839 mutex_enter(&vd->vdev_dtl_lock);
1840 space_map_walk(sm, space_map_add, &smsync);
1841 mutex_exit(&vd->vdev_dtl_lock);
1843 space_map_truncate(smo, mos, tx);
1844 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1846 space_map_destroy(&smsync);
1848 mutex_exit(&smlock);
1849 mutex_destroy(&smlock);
1851 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1852 dmu_buf_will_dirty(db, tx);
1853 ASSERT3U(db->db_size, >=, sizeof (*smo));
1854 bcopy(smo, db->db_data, sizeof (*smo));
1855 dmu_buf_rele(db, FTAG);
1861 * Determine whether the specified vdev can be offlined/detached/removed
1862 * without losing data.
1865 vdev_dtl_required(vdev_t *vd)
1867 spa_t *spa = vd->vdev_spa;
1868 vdev_t *tvd = vd->vdev_top;
1869 uint8_t cant_read = vd->vdev_cant_read;
1872 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1874 if (vd == spa->spa_root_vdev || vd == tvd)
1878 * Temporarily mark the device as unreadable, and then determine
1879 * whether this results in any DTL outages in the top-level vdev.
1880 * If not, we can safely offline/detach/remove the device.
1882 vd->vdev_cant_read = B_TRUE;
1883 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1884 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1885 vd->vdev_cant_read = cant_read;
1886 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1888 if (!required && zio_injection_enabled)
1889 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1895 * Determine if resilver is needed, and if so the txg range.
1898 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1900 boolean_t needed = B_FALSE;
1901 uint64_t thismin = UINT64_MAX;
1902 uint64_t thismax = 0;
1905 if (vd->vdev_children == 0) {
1906 mutex_enter(&vd->vdev_dtl_lock);
1907 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1908 vdev_writeable(vd)) {
1911 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1912 thismin = ss->ss_start - 1;
1913 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1914 thismax = ss->ss_end;
1917 mutex_exit(&vd->vdev_dtl_lock);
1919 for (c = 0; c < vd->vdev_children; c++) {
1920 vdev_t *cvd = vd->vdev_child[c];
1921 uint64_t cmin, cmax;
1923 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1924 thismin = MIN(thismin, cmin);
1925 thismax = MAX(thismax, cmax);
1931 if (needed && minp) {
1939 vdev_load(vdev_t *vd)
1944 * Recursively load all children.
1946 for (c = 0; c < vd->vdev_children; c++)
1947 vdev_load(vd->vdev_child[c]);
1950 * If this is a top-level vdev, initialize its metaslabs.
1952 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1953 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1954 vdev_metaslab_init(vd, 0) != 0))
1955 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1956 VDEV_AUX_CORRUPT_DATA);
1959 * If this is a leaf vdev, load its DTL.
1961 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1962 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1963 VDEV_AUX_CORRUPT_DATA);
1967 * The special vdev case is used for hot spares and l2cache devices. Its
1968 * sole purpose it to set the vdev state for the associated vdev. To do this,
1969 * we make sure that we can open the underlying device, then try to read the
1970 * label, and make sure that the label is sane and that it hasn't been
1971 * repurposed to another pool.
1974 vdev_validate_aux(vdev_t *vd)
1977 uint64_t guid, version;
1980 if (!vdev_readable(vd))
1983 if ((label = vdev_label_read_config(vd)) == NULL) {
1984 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1985 VDEV_AUX_CORRUPT_DATA);
1989 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1990 version > SPA_VERSION ||
1991 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1992 guid != vd->vdev_guid ||
1993 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1994 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1995 VDEV_AUX_CORRUPT_DATA);
2001 * We don't actually check the pool state here. If it's in fact in
2002 * use by another pool, we update this fact on the fly when requested.
2009 vdev_remove(vdev_t *vd, uint64_t txg)
2011 spa_t *spa = vd->vdev_spa;
2012 objset_t *mos = spa->spa_meta_objset;
2016 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2018 if (vd->vdev_dtl_smo.smo_object) {
2019 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
2020 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2021 vd->vdev_dtl_smo.smo_object = 0;
2024 if (vd->vdev_ms != NULL) {
2025 for (m = 0; m < vd->vdev_ms_count; m++) {
2026 metaslab_t *msp = vd->vdev_ms[m];
2028 if (msp == NULL || msp->ms_smo.smo_object == 0)
2031 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2032 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2033 msp->ms_smo.smo_object = 0;
2037 if (vd->vdev_ms_array) {
2038 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2039 vd->vdev_ms_array = 0;
2040 vd->vdev_ms_shift = 0;
2046 vdev_sync_done(vdev_t *vd, uint64_t txg)
2049 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2051 ASSERT(!vd->vdev_ishole);
2053 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
2054 metaslab_sync_done(msp, txg);
2057 metaslab_sync_reassess(vd->vdev_mg);
2061 vdev_sync(vdev_t *vd, uint64_t txg)
2063 spa_t *spa = vd->vdev_spa;
2068 ASSERT(!vd->vdev_ishole);
2070 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2071 ASSERT(vd == vd->vdev_top);
2072 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2073 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2074 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2075 ASSERT(vd->vdev_ms_array != 0);
2076 vdev_config_dirty(vd);
2081 * Remove the metadata associated with this vdev once it's empty.
2083 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2084 vdev_remove(vd, txg);
2086 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2087 metaslab_sync(msp, txg);
2088 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2091 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2092 vdev_dtl_sync(lvd, txg);
2094 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2098 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2100 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2104 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2105 * not be opened, and no I/O is attempted.
2108 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2112 spa_vdev_state_enter(spa, SCL_NONE);
2114 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2115 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2117 if (!vd->vdev_ops->vdev_op_leaf)
2118 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2123 * We don't directly use the aux state here, but if we do a
2124 * vdev_reopen(), we need this value to be present to remember why we
2127 vd->vdev_label_aux = aux;
2130 * Faulted state takes precedence over degraded.
2132 vd->vdev_delayed_close = B_FALSE;
2133 vd->vdev_faulted = 1ULL;
2134 vd->vdev_degraded = 0ULL;
2135 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2138 * If this device has the only valid copy of the data, then
2139 * back off and simply mark the vdev as degraded instead.
2141 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2142 vd->vdev_degraded = 1ULL;
2143 vd->vdev_faulted = 0ULL;
2146 * If we reopen the device and it's not dead, only then do we
2151 if (vdev_readable(vd))
2152 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2155 return (spa_vdev_state_exit(spa, vd, 0));
2159 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2160 * user that something is wrong. The vdev continues to operate as normal as far
2161 * as I/O is concerned.
2164 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2168 spa_vdev_state_enter(spa, SCL_NONE);
2170 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2171 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2173 if (!vd->vdev_ops->vdev_op_leaf)
2174 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2177 * If the vdev is already faulted, then don't do anything.
2179 if (vd->vdev_faulted || vd->vdev_degraded)
2180 return (spa_vdev_state_exit(spa, NULL, 0));
2182 vd->vdev_degraded = 1ULL;
2183 if (!vdev_is_dead(vd))
2184 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2187 return (spa_vdev_state_exit(spa, vd, 0));
2191 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2192 * any attached spare device should be detached when the device finishes
2193 * resilvering. Second, the online should be treated like a 'test' online case,
2194 * so no FMA events are generated if the device fails to open.
2197 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2199 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2201 spa_vdev_state_enter(spa, SCL_NONE);
2203 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2204 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2206 if (!vd->vdev_ops->vdev_op_leaf)
2207 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2210 vd->vdev_offline = B_FALSE;
2211 vd->vdev_tmpoffline = B_FALSE;
2212 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2213 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2215 /* XXX - L2ARC 1.0 does not support expansion */
2216 if (!vd->vdev_aux) {
2217 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2218 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2222 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2224 if (!vd->vdev_aux) {
2225 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2226 pvd->vdev_expanding = B_FALSE;
2230 *newstate = vd->vdev_state;
2231 if ((flags & ZFS_ONLINE_UNSPARE) &&
2232 !vdev_is_dead(vd) && vd->vdev_parent &&
2233 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2234 vd->vdev_parent->vdev_child[0] == vd)
2235 vd->vdev_unspare = B_TRUE;
2237 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2239 /* XXX - L2ARC 1.0 does not support expansion */
2241 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2242 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2244 return (spa_vdev_state_exit(spa, vd, 0));
2248 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2252 uint64_t generation;
2253 metaslab_group_t *mg;
2256 spa_vdev_state_enter(spa, SCL_ALLOC);
2258 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2259 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2261 if (!vd->vdev_ops->vdev_op_leaf)
2262 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2266 generation = spa->spa_config_generation + 1;
2269 * If the device isn't already offline, try to offline it.
2271 if (!vd->vdev_offline) {
2273 * If this device has the only valid copy of some data,
2274 * don't allow it to be offlined. Log devices are always
2277 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2278 vdev_dtl_required(vd))
2279 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2282 * If the top-level is a slog and it has had allocations
2283 * then proceed. We check that the vdev's metaslab group
2284 * is not NULL since it's possible that we may have just
2285 * added this vdev but not yet initialized its metaslabs.
2287 if (tvd->vdev_islog && mg != NULL) {
2289 * Prevent any future allocations.
2291 metaslab_group_passivate(mg);
2292 (void) spa_vdev_state_exit(spa, vd, 0);
2294 error = spa_offline_log(spa);
2296 spa_vdev_state_enter(spa, SCL_ALLOC);
2299 * Check to see if the config has changed.
2301 if (error || generation != spa->spa_config_generation) {
2302 metaslab_group_activate(mg);
2304 return (spa_vdev_state_exit(spa,
2306 (void) spa_vdev_state_exit(spa, vd, 0);
2309 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2313 * Offline this device and reopen its top-level vdev.
2314 * If the top-level vdev is a log device then just offline
2315 * it. Otherwise, if this action results in the top-level
2316 * vdev becoming unusable, undo it and fail the request.
2318 vd->vdev_offline = B_TRUE;
2321 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2322 vdev_is_dead(tvd)) {
2323 vd->vdev_offline = B_FALSE;
2325 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2329 * Add the device back into the metaslab rotor so that
2330 * once we online the device it's open for business.
2332 if (tvd->vdev_islog && mg != NULL)
2333 metaslab_group_activate(mg);
2336 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2338 return (spa_vdev_state_exit(spa, vd, 0));
2342 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2346 mutex_enter(&spa->spa_vdev_top_lock);
2347 error = vdev_offline_locked(spa, guid, flags);
2348 mutex_exit(&spa->spa_vdev_top_lock);
2354 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2355 * vdev_offline(), we assume the spa config is locked. We also clear all
2356 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2359 vdev_clear(spa_t *spa, vdev_t *vd)
2361 vdev_t *rvd = spa->spa_root_vdev;
2364 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2369 vd->vdev_stat.vs_read_errors = 0;
2370 vd->vdev_stat.vs_write_errors = 0;
2371 vd->vdev_stat.vs_checksum_errors = 0;
2373 for (c = 0; c < vd->vdev_children; c++)
2374 vdev_clear(spa, vd->vdev_child[c]);
2377 * If we're in the FAULTED state or have experienced failed I/O, then
2378 * clear the persistent state and attempt to reopen the device. We
2379 * also mark the vdev config dirty, so that the new faulted state is
2380 * written out to disk.
2382 if (vd->vdev_faulted || vd->vdev_degraded ||
2383 !vdev_readable(vd) || !vdev_writeable(vd)) {
2386 * When reopening in reponse to a clear event, it may be due to
2387 * a fmadm repair request. In this case, if the device is
2388 * still broken, we want to still post the ereport again.
2390 vd->vdev_forcefault = B_TRUE;
2392 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2393 vd->vdev_cant_read = B_FALSE;
2394 vd->vdev_cant_write = B_FALSE;
2396 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2398 vd->vdev_forcefault = B_FALSE;
2400 if (vd != rvd && vdev_writeable(vd->vdev_top))
2401 vdev_state_dirty(vd->vdev_top);
2403 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2404 spa_async_request(spa, SPA_ASYNC_RESILVER);
2406 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR);
2410 * When clearing a FMA-diagnosed fault, we always want to
2411 * unspare the device, as we assume that the original spare was
2412 * done in response to the FMA fault.
2414 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2415 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2416 vd->vdev_parent->vdev_child[0] == vd)
2417 vd->vdev_unspare = B_TRUE;
2421 vdev_is_dead(vdev_t *vd)
2424 * Holes and missing devices are always considered "dead".
2425 * This simplifies the code since we don't have to check for
2426 * these types of devices in the various code paths.
2427 * Instead we rely on the fact that we skip over dead devices
2428 * before issuing I/O to them.
2430 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2431 vd->vdev_ops == &vdev_missing_ops);
2435 vdev_readable(vdev_t *vd)
2437 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2441 vdev_writeable(vdev_t *vd)
2443 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2447 vdev_allocatable(vdev_t *vd)
2449 uint64_t state = vd->vdev_state;
2452 * We currently allow allocations from vdevs which may be in the
2453 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2454 * fails to reopen then we'll catch it later when we're holding
2455 * the proper locks. Note that we have to get the vdev state
2456 * in a local variable because although it changes atomically,
2457 * we're asking two separate questions about it.
2459 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2460 !vd->vdev_cant_write && !vd->vdev_ishole);
2464 vdev_accessible(vdev_t *vd, zio_t *zio)
2466 ASSERT(zio->io_vd == vd);
2468 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2471 if (zio->io_type == ZIO_TYPE_READ)
2472 return (!vd->vdev_cant_read);
2474 if (zio->io_type == ZIO_TYPE_WRITE)
2475 return (!vd->vdev_cant_write);
2481 * Get statistics for the given vdev.
2484 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2486 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2489 mutex_enter(&vd->vdev_stat_lock);
2490 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2491 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2492 vs->vs_state = vd->vdev_state;
2493 vs->vs_rsize = vdev_get_min_asize(vd);
2494 if (vd->vdev_ops->vdev_op_leaf)
2495 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2496 mutex_exit(&vd->vdev_stat_lock);
2499 * If we're getting stats on the root vdev, aggregate the I/O counts
2500 * over all top-level vdevs (i.e. the direct children of the root).
2503 for (c = 0; c < rvd->vdev_children; c++) {
2504 vdev_t *cvd = rvd->vdev_child[c];
2505 vdev_stat_t *cvs = &cvd->vdev_stat;
2507 mutex_enter(&vd->vdev_stat_lock);
2508 for (t = 0; t < ZIO_TYPES; t++) {
2509 vs->vs_ops[t] += cvs->vs_ops[t];
2510 vs->vs_bytes[t] += cvs->vs_bytes[t];
2512 cvs->vs_scan_removing = cvd->vdev_removing;
2513 mutex_exit(&vd->vdev_stat_lock);
2519 vdev_clear_stats(vdev_t *vd)
2521 mutex_enter(&vd->vdev_stat_lock);
2522 vd->vdev_stat.vs_space = 0;
2523 vd->vdev_stat.vs_dspace = 0;
2524 vd->vdev_stat.vs_alloc = 0;
2525 mutex_exit(&vd->vdev_stat_lock);
2529 vdev_scan_stat_init(vdev_t *vd)
2531 vdev_stat_t *vs = &vd->vdev_stat;
2534 for (c = 0; c < vd->vdev_children; c++)
2535 vdev_scan_stat_init(vd->vdev_child[c]);
2537 mutex_enter(&vd->vdev_stat_lock);
2538 vs->vs_scan_processed = 0;
2539 mutex_exit(&vd->vdev_stat_lock);
2543 vdev_stat_update(zio_t *zio, uint64_t psize)
2545 spa_t *spa = zio->io_spa;
2546 vdev_t *rvd = spa->spa_root_vdev;
2547 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2549 uint64_t txg = zio->io_txg;
2550 vdev_stat_t *vs = &vd->vdev_stat;
2551 zio_type_t type = zio->io_type;
2552 int flags = zio->io_flags;
2555 * If this i/o is a gang leader, it didn't do any actual work.
2557 if (zio->io_gang_tree)
2560 if (zio->io_error == 0) {
2562 * If this is a root i/o, don't count it -- we've already
2563 * counted the top-level vdevs, and vdev_get_stats() will
2564 * aggregate them when asked. This reduces contention on
2565 * the root vdev_stat_lock and implicitly handles blocks
2566 * that compress away to holes, for which there is no i/o.
2567 * (Holes never create vdev children, so all the counters
2568 * remain zero, which is what we want.)
2570 * Note: this only applies to successful i/o (io_error == 0)
2571 * because unlike i/o counts, errors are not additive.
2572 * When reading a ditto block, for example, failure of
2573 * one top-level vdev does not imply a root-level error.
2578 ASSERT(vd == zio->io_vd);
2580 if (flags & ZIO_FLAG_IO_BYPASS)
2583 mutex_enter(&vd->vdev_stat_lock);
2585 if (flags & ZIO_FLAG_IO_REPAIR) {
2586 if (flags & ZIO_FLAG_SCAN_THREAD) {
2587 dsl_scan_phys_t *scn_phys =
2588 &spa->spa_dsl_pool->dp_scan->scn_phys;
2589 uint64_t *processed = &scn_phys->scn_processed;
2592 if (vd->vdev_ops->vdev_op_leaf)
2593 atomic_add_64(processed, psize);
2594 vs->vs_scan_processed += psize;
2597 if (flags & ZIO_FLAG_SELF_HEAL)
2598 vs->vs_self_healed += psize;
2602 vs->vs_bytes[type] += psize;
2604 mutex_exit(&vd->vdev_stat_lock);
2608 if (flags & ZIO_FLAG_SPECULATIVE)
2612 * If this is an I/O error that is going to be retried, then ignore the
2613 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2614 * hard errors, when in reality they can happen for any number of
2615 * innocuous reasons (bus resets, MPxIO link failure, etc).
2617 if (zio->io_error == EIO &&
2618 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2622 * Intent logs writes won't propagate their error to the root
2623 * I/O so don't mark these types of failures as pool-level
2626 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2629 mutex_enter(&vd->vdev_stat_lock);
2630 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2631 if (zio->io_error == ECKSUM)
2632 vs->vs_checksum_errors++;
2634 vs->vs_read_errors++;
2636 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2637 vs->vs_write_errors++;
2638 mutex_exit(&vd->vdev_stat_lock);
2640 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2641 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2642 (flags & ZIO_FLAG_SCAN_THREAD) ||
2643 spa->spa_claiming)) {
2645 * This is either a normal write (not a repair), or it's
2646 * a repair induced by the scrub thread, or it's a repair
2647 * made by zil_claim() during spa_load() in the first txg.
2648 * In the normal case, we commit the DTL change in the same
2649 * txg as the block was born. In the scrub-induced repair
2650 * case, we know that scrubs run in first-pass syncing context,
2651 * so we commit the DTL change in spa_syncing_txg(spa).
2652 * In the zil_claim() case, we commit in spa_first_txg(spa).
2654 * We currently do not make DTL entries for failed spontaneous
2655 * self-healing writes triggered by normal (non-scrubbing)
2656 * reads, because we have no transactional context in which to
2657 * do so -- and it's not clear that it'd be desirable anyway.
2659 if (vd->vdev_ops->vdev_op_leaf) {
2660 uint64_t commit_txg = txg;
2661 if (flags & ZIO_FLAG_SCAN_THREAD) {
2662 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2663 ASSERT(spa_sync_pass(spa) == 1);
2664 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2665 commit_txg = spa_syncing_txg(spa);
2666 } else if (spa->spa_claiming) {
2667 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2668 commit_txg = spa_first_txg(spa);
2670 ASSERT(commit_txg >= spa_syncing_txg(spa));
2671 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2673 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2674 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2675 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2678 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2683 * Update the in-core space usage stats for this vdev, its metaslab class,
2684 * and the root vdev.
2687 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2688 int64_t space_delta)
2690 int64_t dspace_delta = space_delta;
2691 spa_t *spa = vd->vdev_spa;
2692 vdev_t *rvd = spa->spa_root_vdev;
2693 metaslab_group_t *mg = vd->vdev_mg;
2694 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2696 ASSERT(vd == vd->vdev_top);
2699 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2700 * factor. We must calculate this here and not at the root vdev
2701 * because the root vdev's psize-to-asize is simply the max of its
2702 * childrens', thus not accurate enough for us.
2704 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2705 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2706 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2707 vd->vdev_deflate_ratio;
2709 mutex_enter(&vd->vdev_stat_lock);
2710 vd->vdev_stat.vs_alloc += alloc_delta;
2711 vd->vdev_stat.vs_space += space_delta;
2712 vd->vdev_stat.vs_dspace += dspace_delta;
2713 mutex_exit(&vd->vdev_stat_lock);
2715 if (mc == spa_normal_class(spa)) {
2716 mutex_enter(&rvd->vdev_stat_lock);
2717 rvd->vdev_stat.vs_alloc += alloc_delta;
2718 rvd->vdev_stat.vs_space += space_delta;
2719 rvd->vdev_stat.vs_dspace += dspace_delta;
2720 mutex_exit(&rvd->vdev_stat_lock);
2724 ASSERT(rvd == vd->vdev_parent);
2725 ASSERT(vd->vdev_ms_count != 0);
2727 metaslab_class_space_update(mc,
2728 alloc_delta, defer_delta, space_delta, dspace_delta);
2733 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2734 * so that it will be written out next time the vdev configuration is synced.
2735 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2738 vdev_config_dirty(vdev_t *vd)
2740 spa_t *spa = vd->vdev_spa;
2741 vdev_t *rvd = spa->spa_root_vdev;
2744 ASSERT(spa_writeable(spa));
2747 * If this is an aux vdev (as with l2cache and spare devices), then we
2748 * update the vdev config manually and set the sync flag.
2750 if (vd->vdev_aux != NULL) {
2751 spa_aux_vdev_t *sav = vd->vdev_aux;
2755 for (c = 0; c < sav->sav_count; c++) {
2756 if (sav->sav_vdevs[c] == vd)
2760 if (c == sav->sav_count) {
2762 * We're being removed. There's nothing more to do.
2764 ASSERT(sav->sav_sync == B_TRUE);
2768 sav->sav_sync = B_TRUE;
2770 if (nvlist_lookup_nvlist_array(sav->sav_config,
2771 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2772 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2773 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2779 * Setting the nvlist in the middle if the array is a little
2780 * sketchy, but it will work.
2782 nvlist_free(aux[c]);
2783 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2789 * The dirty list is protected by the SCL_CONFIG lock. The caller
2790 * must either hold SCL_CONFIG as writer, or must be the sync thread
2791 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2792 * so this is sufficient to ensure mutual exclusion.
2794 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2795 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2796 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2799 for (c = 0; c < rvd->vdev_children; c++)
2800 vdev_config_dirty(rvd->vdev_child[c]);
2802 ASSERT(vd == vd->vdev_top);
2804 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2806 list_insert_head(&spa->spa_config_dirty_list, vd);
2811 vdev_config_clean(vdev_t *vd)
2813 spa_t *spa = vd->vdev_spa;
2815 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2816 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2817 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2819 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2820 list_remove(&spa->spa_config_dirty_list, vd);
2824 * Mark a top-level vdev's state as dirty, so that the next pass of
2825 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2826 * the state changes from larger config changes because they require
2827 * much less locking, and are often needed for administrative actions.
2830 vdev_state_dirty(vdev_t *vd)
2832 spa_t *spa = vd->vdev_spa;
2834 ASSERT(spa_writeable(spa));
2835 ASSERT(vd == vd->vdev_top);
2838 * The state list is protected by the SCL_STATE lock. The caller
2839 * must either hold SCL_STATE as writer, or must be the sync thread
2840 * (which holds SCL_STATE as reader). There's only one sync thread,
2841 * so this is sufficient to ensure mutual exclusion.
2843 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2844 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2845 spa_config_held(spa, SCL_STATE, RW_READER)));
2847 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2848 list_insert_head(&spa->spa_state_dirty_list, vd);
2852 vdev_state_clean(vdev_t *vd)
2854 spa_t *spa = vd->vdev_spa;
2856 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2857 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2858 spa_config_held(spa, SCL_STATE, RW_READER)));
2860 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2861 list_remove(&spa->spa_state_dirty_list, vd);
2865 * Propagate vdev state up from children to parent.
2868 vdev_propagate_state(vdev_t *vd)
2870 spa_t *spa = vd->vdev_spa;
2871 vdev_t *rvd = spa->spa_root_vdev;
2872 int degraded = 0, faulted = 0;
2877 if (vd->vdev_children > 0) {
2878 for (c = 0; c < vd->vdev_children; c++) {
2879 child = vd->vdev_child[c];
2882 * Don't factor holes into the decision.
2884 if (child->vdev_ishole)
2887 if (!vdev_readable(child) ||
2888 (!vdev_writeable(child) && spa_writeable(spa))) {
2890 * Root special: if there is a top-level log
2891 * device, treat the root vdev as if it were
2894 if (child->vdev_islog && vd == rvd)
2898 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2902 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2906 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2909 * Root special: if there is a top-level vdev that cannot be
2910 * opened due to corrupted metadata, then propagate the root
2911 * vdev's aux state as 'corrupt' rather than 'insufficient
2914 if (corrupted && vd == rvd &&
2915 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2916 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2917 VDEV_AUX_CORRUPT_DATA);
2920 if (vd->vdev_parent)
2921 vdev_propagate_state(vd->vdev_parent);
2925 * Set a vdev's state. If this is during an open, we don't update the parent
2926 * state, because we're in the process of opening children depth-first.
2927 * Otherwise, we propagate the change to the parent.
2929 * If this routine places a device in a faulted state, an appropriate ereport is
2933 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2935 uint64_t save_state;
2936 spa_t *spa = vd->vdev_spa;
2938 if (state == vd->vdev_state) {
2939 vd->vdev_stat.vs_aux = aux;
2943 save_state = vd->vdev_state;
2945 vd->vdev_state = state;
2946 vd->vdev_stat.vs_aux = aux;
2949 * If we are setting the vdev state to anything but an open state, then
2950 * always close the underlying device unless the device has requested
2951 * a delayed close (i.e. we're about to remove or fault the device).
2952 * Otherwise, we keep accessible but invalid devices open forever.
2953 * We don't call vdev_close() itself, because that implies some extra
2954 * checks (offline, etc) that we don't want here. This is limited to
2955 * leaf devices, because otherwise closing the device will affect other
2958 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2959 vd->vdev_ops->vdev_op_leaf)
2960 vd->vdev_ops->vdev_op_close(vd);
2963 * If we have brought this vdev back into service, we need
2964 * to notify fmd so that it can gracefully repair any outstanding
2965 * cases due to a missing device. We do this in all cases, even those
2966 * that probably don't correlate to a repaired fault. This is sure to
2967 * catch all cases, and we let the zfs-retire agent sort it out. If
2968 * this is a transient state it's OK, as the retire agent will
2969 * double-check the state of the vdev before repairing it.
2971 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2972 vd->vdev_prevstate != state)
2973 zfs_post_state_change(spa, vd);
2975 if (vd->vdev_removed &&
2976 state == VDEV_STATE_CANT_OPEN &&
2977 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2979 * If the previous state is set to VDEV_STATE_REMOVED, then this
2980 * device was previously marked removed and someone attempted to
2981 * reopen it. If this failed due to a nonexistent device, then
2982 * keep the device in the REMOVED state. We also let this be if
2983 * it is one of our special test online cases, which is only
2984 * attempting to online the device and shouldn't generate an FMA
2987 vd->vdev_state = VDEV_STATE_REMOVED;
2988 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2989 } else if (state == VDEV_STATE_REMOVED) {
2990 vd->vdev_removed = B_TRUE;
2991 } else if (state == VDEV_STATE_CANT_OPEN) {
2993 * If we fail to open a vdev during an import or recovery, we
2994 * mark it as "not available", which signifies that it was
2995 * never there to begin with. Failure to open such a device
2996 * is not considered an error.
2998 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2999 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3000 vd->vdev_ops->vdev_op_leaf)
3001 vd->vdev_not_present = 1;
3004 * Post the appropriate ereport. If the 'prevstate' field is
3005 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3006 * that this is part of a vdev_reopen(). In this case, we don't
3007 * want to post the ereport if the device was already in the
3008 * CANT_OPEN state beforehand.
3010 * If the 'checkremove' flag is set, then this is an attempt to
3011 * online the device in response to an insertion event. If we
3012 * hit this case, then we have detected an insertion event for a
3013 * faulted or offline device that wasn't in the removed state.
3014 * In this scenario, we don't post an ereport because we are
3015 * about to replace the device, or attempt an online with
3016 * vdev_forcefault, which will generate the fault for us.
3018 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3019 !vd->vdev_not_present && !vd->vdev_checkremove &&
3020 vd != spa->spa_root_vdev) {
3024 case VDEV_AUX_OPEN_FAILED:
3025 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3027 case VDEV_AUX_CORRUPT_DATA:
3028 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3030 case VDEV_AUX_NO_REPLICAS:
3031 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3033 case VDEV_AUX_BAD_GUID_SUM:
3034 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3036 case VDEV_AUX_TOO_SMALL:
3037 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3039 case VDEV_AUX_BAD_LABEL:
3040 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3043 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3046 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3049 /* Erase any notion of persistent removed state */
3050 vd->vdev_removed = B_FALSE;
3052 vd->vdev_removed = B_FALSE;
3055 if (!isopen && vd->vdev_parent)
3056 vdev_propagate_state(vd->vdev_parent);
3060 * Check the vdev configuration to ensure that it's capable of supporting
3064 vdev_is_bootable(vdev_t *vd)
3066 #if defined(__sun__) || defined(__sun)
3068 * Currently, we do not support RAID-Z or partial configuration.
3069 * In addition, only a single top-level vdev is allowed and none of the
3070 * leaves can be wholedisks.
3074 if (!vd->vdev_ops->vdev_op_leaf) {
3075 char *vdev_type = vd->vdev_ops->vdev_op_type;
3077 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3078 vd->vdev_children > 1) {
3080 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3081 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3084 } else if (vd->vdev_wholedisk == 1) {
3088 for (c = 0; c < vd->vdev_children; c++) {
3089 if (!vdev_is_bootable(vd->vdev_child[c]))
3092 #endif /* __sun__ || __sun */
3097 * Load the state from the original vdev tree (ovd) which
3098 * we've retrieved from the MOS config object. If the original
3099 * vdev was offline or faulted then we transfer that state to the
3100 * device in the current vdev tree (nvd).
3103 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3107 ASSERT(nvd->vdev_top->vdev_islog);
3108 ASSERT(spa_config_held(nvd->vdev_spa,
3109 SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3110 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3112 for (c = 0; c < nvd->vdev_children; c++)
3113 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3115 if (nvd->vdev_ops->vdev_op_leaf) {
3117 * Restore the persistent vdev state
3119 nvd->vdev_offline = ovd->vdev_offline;
3120 nvd->vdev_faulted = ovd->vdev_faulted;
3121 nvd->vdev_degraded = ovd->vdev_degraded;
3122 nvd->vdev_removed = ovd->vdev_removed;
3127 * Determine if a log device has valid content. If the vdev was
3128 * removed or faulted in the MOS config then we know that
3129 * the content on the log device has already been written to the pool.
3132 vdev_log_state_valid(vdev_t *vd)
3136 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3140 for (c = 0; c < vd->vdev_children; c++)
3141 if (vdev_log_state_valid(vd->vdev_child[c]))
3148 * Expand a vdev if possible.
3151 vdev_expand(vdev_t *vd, uint64_t txg)
3153 ASSERT(vd->vdev_top == vd);
3154 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3156 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3157 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3158 vdev_config_dirty(vd);
3166 vdev_split(vdev_t *vd)
3168 vdev_t *cvd, *pvd = vd->vdev_parent;
3170 vdev_remove_child(pvd, vd);
3171 vdev_compact_children(pvd);
3173 cvd = pvd->vdev_child[0];
3174 if (pvd->vdev_children == 1) {
3175 vdev_remove_parent(cvd);
3176 cvd->vdev_splitting = B_TRUE;
3178 vdev_propagate_state(cvd);
3181 #if defined(_KERNEL) && defined(HAVE_SPL)
3182 EXPORT_SYMBOL(vdev_fault);
3183 EXPORT_SYMBOL(vdev_degrade);
3184 EXPORT_SYMBOL(vdev_online);
3185 EXPORT_SYMBOL(vdev_offline);
3186 EXPORT_SYMBOL(vdev_clear);
3188 module_param(zfs_scrub_limit, int, 0644);
3189 MODULE_PARM_DESC(zfs_scrub_limit, "Max scrub/resilver I/O per leaf vdev");