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_PUSHPAGE);
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_PUSHPAGE);
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_PUSHPAGE);
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_PUSHPAGE | 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_PUSHPAGE);
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 * If 'strict' is false ignore the spa guid check. This is necessary because
1314 * if the machine crashed during a re-guid the new guid might have been written
1315 * to all of the vdev labels, but not the cached config. The strict check
1316 * will be performed when the pool is opened again using the mos config.
1318 * This function will only return failure if one of the vdevs indicates that it
1319 * has since been destroyed or exported. This is only possible if
1320 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1321 * will be updated but the function will return 0.
1324 vdev_validate(vdev_t *vd, boolean_t strict)
1326 spa_t *spa = vd->vdev_spa;
1328 uint64_t guid = 0, top_guid;
1332 for (c = 0; c < vd->vdev_children; c++)
1333 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1337 * If the device has already failed, or was marked offline, don't do
1338 * any further validation. Otherwise, label I/O will fail and we will
1339 * overwrite the previous state.
1341 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1342 uint64_t aux_guid = 0;
1345 if ((label = vdev_label_read_config(vd)) == NULL) {
1346 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1347 VDEV_AUX_BAD_LABEL);
1352 * Determine if this vdev has been split off into another
1353 * pool. If so, then refuse to open it.
1355 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1356 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1357 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1358 VDEV_AUX_SPLIT_POOL);
1363 if (strict && (nvlist_lookup_uint64(label,
1364 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1365 guid != spa_guid(spa))) {
1366 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1367 VDEV_AUX_CORRUPT_DATA);
1372 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1373 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1378 * If this vdev just became a top-level vdev because its
1379 * sibling was detached, it will have adopted the parent's
1380 * vdev guid -- but the label may or may not be on disk yet.
1381 * Fortunately, either version of the label will have the
1382 * same top guid, so if we're a top-level vdev, we can
1383 * safely compare to that instead.
1385 * If we split this vdev off instead, then we also check the
1386 * original pool's guid. We don't want to consider the vdev
1387 * corrupt if it is partway through a split operation.
1389 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1391 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1393 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1394 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1395 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1396 VDEV_AUX_CORRUPT_DATA);
1401 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1403 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1404 VDEV_AUX_CORRUPT_DATA);
1412 * If this is a verbatim import, no need to check the
1413 * state of the pool.
1415 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1416 spa_load_state(spa) == SPA_LOAD_OPEN &&
1417 state != POOL_STATE_ACTIVE)
1421 * If we were able to open and validate a vdev that was
1422 * previously marked permanently unavailable, clear that state
1425 if (vd->vdev_not_present)
1426 vd->vdev_not_present = 0;
1433 * Close a virtual device.
1436 vdev_close(vdev_t *vd)
1438 vdev_t *pvd = vd->vdev_parent;
1439 ASSERTV(spa_t *spa = vd->vdev_spa);
1441 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1444 * If our parent is reopening, then we are as well, unless we are
1447 if (pvd != NULL && pvd->vdev_reopening)
1448 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1450 vd->vdev_ops->vdev_op_close(vd);
1452 vdev_cache_purge(vd);
1455 * We record the previous state before we close it, so that if we are
1456 * doing a reopen(), we don't generate FMA ereports if we notice that
1457 * it's still faulted.
1459 vd->vdev_prevstate = vd->vdev_state;
1461 if (vd->vdev_offline)
1462 vd->vdev_state = VDEV_STATE_OFFLINE;
1464 vd->vdev_state = VDEV_STATE_CLOSED;
1465 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1469 vdev_hold(vdev_t *vd)
1471 spa_t *spa = vd->vdev_spa;
1474 ASSERT(spa_is_root(spa));
1475 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1478 for (c = 0; c < vd->vdev_children; c++)
1479 vdev_hold(vd->vdev_child[c]);
1481 if (vd->vdev_ops->vdev_op_leaf)
1482 vd->vdev_ops->vdev_op_hold(vd);
1486 vdev_rele(vdev_t *vd)
1490 ASSERT(spa_is_root(vd->vdev_spa));
1491 for (c = 0; c < vd->vdev_children; c++)
1492 vdev_rele(vd->vdev_child[c]);
1494 if (vd->vdev_ops->vdev_op_leaf)
1495 vd->vdev_ops->vdev_op_rele(vd);
1499 * Reopen all interior vdevs and any unopened leaves. We don't actually
1500 * reopen leaf vdevs which had previously been opened as they might deadlock
1501 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1502 * If the leaf has never been opened then open it, as usual.
1505 vdev_reopen(vdev_t *vd)
1507 spa_t *spa = vd->vdev_spa;
1509 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1511 /* set the reopening flag unless we're taking the vdev offline */
1512 vd->vdev_reopening = !vd->vdev_offline;
1514 (void) vdev_open(vd);
1517 * Call vdev_validate() here to make sure we have the same device.
1518 * Otherwise, a device with an invalid label could be successfully
1519 * opened in response to vdev_reopen().
1522 (void) vdev_validate_aux(vd);
1523 if (vdev_readable(vd) && vdev_writeable(vd) &&
1524 vd->vdev_aux == &spa->spa_l2cache &&
1525 !l2arc_vdev_present(vd))
1526 l2arc_add_vdev(spa, vd);
1528 (void) vdev_validate(vd, B_TRUE);
1532 * Reassess parent vdev's health.
1534 vdev_propagate_state(vd);
1538 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1543 * Normally, partial opens (e.g. of a mirror) are allowed.
1544 * For a create, however, we want to fail the request if
1545 * there are any components we can't open.
1547 error = vdev_open(vd);
1549 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1551 return (error ? error : ENXIO);
1555 * Recursively initialize all labels.
1557 if ((error = vdev_label_init(vd, txg, isreplacing ?
1558 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1567 vdev_metaslab_set_size(vdev_t *vd)
1570 * Aim for roughly 200 metaslabs per vdev.
1572 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1573 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1577 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1579 ASSERT(vd == vd->vdev_top);
1580 ASSERT(!vd->vdev_ishole);
1581 ASSERT(ISP2(flags));
1582 ASSERT(spa_writeable(vd->vdev_spa));
1584 if (flags & VDD_METASLAB)
1585 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1587 if (flags & VDD_DTL)
1588 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1590 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1596 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1597 * the vdev has less than perfect replication. There are four kinds of DTL:
1599 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1601 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1603 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1604 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1605 * txgs that was scrubbed.
1607 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1608 * persistent errors or just some device being offline.
1609 * Unlike the other three, the DTL_OUTAGE map is not generally
1610 * maintained; it's only computed when needed, typically to
1611 * determine whether a device can be detached.
1613 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1614 * either has the data or it doesn't.
1616 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1617 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1618 * if any child is less than fully replicated, then so is its parent.
1619 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1620 * comprising only those txgs which appear in 'maxfaults' or more children;
1621 * those are the txgs we don't have enough replication to read. For example,
1622 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1623 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1624 * two child DTL_MISSING maps.
1626 * It should be clear from the above that to compute the DTLs and outage maps
1627 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1628 * Therefore, that is all we keep on disk. When loading the pool, or after
1629 * a configuration change, we generate all other DTLs from first principles.
1632 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1634 space_map_t *sm = &vd->vdev_dtl[t];
1636 ASSERT(t < DTL_TYPES);
1637 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1638 ASSERT(spa_writeable(vd->vdev_spa));
1640 mutex_enter(sm->sm_lock);
1641 if (!space_map_contains(sm, txg, size))
1642 space_map_add(sm, txg, size);
1643 mutex_exit(sm->sm_lock);
1647 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1649 space_map_t *sm = &vd->vdev_dtl[t];
1650 boolean_t dirty = B_FALSE;
1652 ASSERT(t < DTL_TYPES);
1653 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1655 mutex_enter(sm->sm_lock);
1656 if (sm->sm_space != 0)
1657 dirty = space_map_contains(sm, txg, size);
1658 mutex_exit(sm->sm_lock);
1664 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1666 space_map_t *sm = &vd->vdev_dtl[t];
1669 mutex_enter(sm->sm_lock);
1670 empty = (sm->sm_space == 0);
1671 mutex_exit(sm->sm_lock);
1677 * Reassess DTLs after a config change or scrub completion.
1680 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1682 spa_t *spa = vd->vdev_spa;
1686 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1688 for (c = 0; c < vd->vdev_children; c++)
1689 vdev_dtl_reassess(vd->vdev_child[c], txg,
1690 scrub_txg, scrub_done);
1692 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1695 if (vd->vdev_ops->vdev_op_leaf) {
1696 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1698 mutex_enter(&vd->vdev_dtl_lock);
1699 if (scrub_txg != 0 &&
1700 (spa->spa_scrub_started ||
1701 (scn && scn->scn_phys.scn_errors == 0))) {
1703 * We completed a scrub up to scrub_txg. If we
1704 * did it without rebooting, then the scrub dtl
1705 * will be valid, so excise the old region and
1706 * fold in the scrub dtl. Otherwise, leave the
1707 * dtl as-is if there was an error.
1709 * There's little trick here: to excise the beginning
1710 * of the DTL_MISSING map, we put it into a reference
1711 * tree and then add a segment with refcnt -1 that
1712 * covers the range [0, scrub_txg). This means
1713 * that each txg in that range has refcnt -1 or 0.
1714 * We then add DTL_SCRUB with a refcnt of 2, so that
1715 * entries in the range [0, scrub_txg) will have a
1716 * positive refcnt -- either 1 or 2. We then convert
1717 * the reference tree into the new DTL_MISSING map.
1719 space_map_ref_create(&reftree);
1720 space_map_ref_add_map(&reftree,
1721 &vd->vdev_dtl[DTL_MISSING], 1);
1722 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1723 space_map_ref_add_map(&reftree,
1724 &vd->vdev_dtl[DTL_SCRUB], 2);
1725 space_map_ref_generate_map(&reftree,
1726 &vd->vdev_dtl[DTL_MISSING], 1);
1727 space_map_ref_destroy(&reftree);
1729 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1730 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1731 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1733 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1734 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1735 if (!vdev_readable(vd))
1736 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1738 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1739 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1740 mutex_exit(&vd->vdev_dtl_lock);
1743 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1747 mutex_enter(&vd->vdev_dtl_lock);
1748 for (t = 0; t < DTL_TYPES; t++) {
1749 /* account for child's outage in parent's missing map */
1750 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1752 continue; /* leaf vdevs only */
1753 if (t == DTL_PARTIAL)
1754 minref = 1; /* i.e. non-zero */
1755 else if (vd->vdev_nparity != 0)
1756 minref = vd->vdev_nparity + 1; /* RAID-Z */
1758 minref = vd->vdev_children; /* any kind of mirror */
1759 space_map_ref_create(&reftree);
1760 for (c = 0; c < vd->vdev_children; c++) {
1761 vdev_t *cvd = vd->vdev_child[c];
1762 mutex_enter(&cvd->vdev_dtl_lock);
1763 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1764 mutex_exit(&cvd->vdev_dtl_lock);
1766 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1767 space_map_ref_destroy(&reftree);
1769 mutex_exit(&vd->vdev_dtl_lock);
1773 vdev_dtl_load(vdev_t *vd)
1775 spa_t *spa = vd->vdev_spa;
1776 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1777 objset_t *mos = spa->spa_meta_objset;
1781 ASSERT(vd->vdev_children == 0);
1783 if (smo->smo_object == 0)
1786 ASSERT(!vd->vdev_ishole);
1788 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1791 ASSERT3U(db->db_size, >=, sizeof (*smo));
1792 bcopy(db->db_data, smo, sizeof (*smo));
1793 dmu_buf_rele(db, FTAG);
1795 mutex_enter(&vd->vdev_dtl_lock);
1796 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1797 NULL, SM_ALLOC, smo, mos);
1798 mutex_exit(&vd->vdev_dtl_lock);
1804 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1806 spa_t *spa = vd->vdev_spa;
1807 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1808 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1809 objset_t *mos = spa->spa_meta_objset;
1815 ASSERT(!vd->vdev_ishole);
1817 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1819 if (vd->vdev_detached) {
1820 if (smo->smo_object != 0) {
1821 VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx));
1822 smo->smo_object = 0;
1828 if (smo->smo_object == 0) {
1829 ASSERT(smo->smo_objsize == 0);
1830 ASSERT(smo->smo_alloc == 0);
1831 smo->smo_object = dmu_object_alloc(mos,
1832 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1833 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1834 ASSERT(smo->smo_object != 0);
1835 vdev_config_dirty(vd->vdev_top);
1838 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1840 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1843 mutex_enter(&smlock);
1845 mutex_enter(&vd->vdev_dtl_lock);
1846 space_map_walk(sm, space_map_add, &smsync);
1847 mutex_exit(&vd->vdev_dtl_lock);
1849 space_map_truncate(smo, mos, tx);
1850 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1852 space_map_destroy(&smsync);
1854 mutex_exit(&smlock);
1855 mutex_destroy(&smlock);
1857 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1858 dmu_buf_will_dirty(db, tx);
1859 ASSERT3U(db->db_size, >=, sizeof (*smo));
1860 bcopy(smo, db->db_data, sizeof (*smo));
1861 dmu_buf_rele(db, FTAG);
1867 * Determine whether the specified vdev can be offlined/detached/removed
1868 * without losing data.
1871 vdev_dtl_required(vdev_t *vd)
1873 spa_t *spa = vd->vdev_spa;
1874 vdev_t *tvd = vd->vdev_top;
1875 uint8_t cant_read = vd->vdev_cant_read;
1878 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1880 if (vd == spa->spa_root_vdev || vd == tvd)
1884 * Temporarily mark the device as unreadable, and then determine
1885 * whether this results in any DTL outages in the top-level vdev.
1886 * If not, we can safely offline/detach/remove the device.
1888 vd->vdev_cant_read = B_TRUE;
1889 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1890 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1891 vd->vdev_cant_read = cant_read;
1892 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1894 if (!required && zio_injection_enabled)
1895 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1901 * Determine if resilver is needed, and if so the txg range.
1904 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1906 boolean_t needed = B_FALSE;
1907 uint64_t thismin = UINT64_MAX;
1908 uint64_t thismax = 0;
1911 if (vd->vdev_children == 0) {
1912 mutex_enter(&vd->vdev_dtl_lock);
1913 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1914 vdev_writeable(vd)) {
1917 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1918 thismin = ss->ss_start - 1;
1919 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1920 thismax = ss->ss_end;
1923 mutex_exit(&vd->vdev_dtl_lock);
1925 for (c = 0; c < vd->vdev_children; c++) {
1926 vdev_t *cvd = vd->vdev_child[c];
1927 uint64_t cmin, cmax;
1929 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1930 thismin = MIN(thismin, cmin);
1931 thismax = MAX(thismax, cmax);
1937 if (needed && minp) {
1945 vdev_load(vdev_t *vd)
1950 * Recursively load all children.
1952 for (c = 0; c < vd->vdev_children; c++)
1953 vdev_load(vd->vdev_child[c]);
1956 * If this is a top-level vdev, initialize its metaslabs.
1958 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1959 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1960 vdev_metaslab_init(vd, 0) != 0))
1961 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1962 VDEV_AUX_CORRUPT_DATA);
1965 * If this is a leaf vdev, load its DTL.
1967 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1968 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1969 VDEV_AUX_CORRUPT_DATA);
1973 * The special vdev case is used for hot spares and l2cache devices. Its
1974 * sole purpose it to set the vdev state for the associated vdev. To do this,
1975 * we make sure that we can open the underlying device, then try to read the
1976 * label, and make sure that the label is sane and that it hasn't been
1977 * repurposed to another pool.
1980 vdev_validate_aux(vdev_t *vd)
1983 uint64_t guid, version;
1986 if (!vdev_readable(vd))
1989 if ((label = vdev_label_read_config(vd)) == NULL) {
1990 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1991 VDEV_AUX_CORRUPT_DATA);
1995 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1996 version > SPA_VERSION ||
1997 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1998 guid != vd->vdev_guid ||
1999 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2000 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2001 VDEV_AUX_CORRUPT_DATA);
2007 * We don't actually check the pool state here. If it's in fact in
2008 * use by another pool, we update this fact on the fly when requested.
2015 vdev_remove(vdev_t *vd, uint64_t txg)
2017 spa_t *spa = vd->vdev_spa;
2018 objset_t *mos = spa->spa_meta_objset;
2022 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2024 if (vd->vdev_dtl_smo.smo_object) {
2025 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
2026 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2027 vd->vdev_dtl_smo.smo_object = 0;
2030 if (vd->vdev_ms != NULL) {
2031 for (m = 0; m < vd->vdev_ms_count; m++) {
2032 metaslab_t *msp = vd->vdev_ms[m];
2034 if (msp == NULL || msp->ms_smo.smo_object == 0)
2037 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2038 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2039 msp->ms_smo.smo_object = 0;
2043 if (vd->vdev_ms_array) {
2044 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2045 vd->vdev_ms_array = 0;
2046 vd->vdev_ms_shift = 0;
2052 vdev_sync_done(vdev_t *vd, uint64_t txg)
2055 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2057 ASSERT(!vd->vdev_ishole);
2059 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
2060 metaslab_sync_done(msp, txg);
2063 metaslab_sync_reassess(vd->vdev_mg);
2067 vdev_sync(vdev_t *vd, uint64_t txg)
2069 spa_t *spa = vd->vdev_spa;
2074 ASSERT(!vd->vdev_ishole);
2076 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2077 ASSERT(vd == vd->vdev_top);
2078 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2079 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2080 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2081 ASSERT(vd->vdev_ms_array != 0);
2082 vdev_config_dirty(vd);
2087 * Remove the metadata associated with this vdev once it's empty.
2089 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2090 vdev_remove(vd, txg);
2092 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2093 metaslab_sync(msp, txg);
2094 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2097 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2098 vdev_dtl_sync(lvd, txg);
2100 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2104 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2106 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2110 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2111 * not be opened, and no I/O is attempted.
2114 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2118 spa_vdev_state_enter(spa, SCL_NONE);
2120 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2121 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2123 if (!vd->vdev_ops->vdev_op_leaf)
2124 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2129 * We don't directly use the aux state here, but if we do a
2130 * vdev_reopen(), we need this value to be present to remember why we
2133 vd->vdev_label_aux = aux;
2136 * Faulted state takes precedence over degraded.
2138 vd->vdev_delayed_close = B_FALSE;
2139 vd->vdev_faulted = 1ULL;
2140 vd->vdev_degraded = 0ULL;
2141 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2144 * If this device has the only valid copy of the data, then
2145 * back off and simply mark the vdev as degraded instead.
2147 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2148 vd->vdev_degraded = 1ULL;
2149 vd->vdev_faulted = 0ULL;
2152 * If we reopen the device and it's not dead, only then do we
2157 if (vdev_readable(vd))
2158 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2161 return (spa_vdev_state_exit(spa, vd, 0));
2165 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2166 * user that something is wrong. The vdev continues to operate as normal as far
2167 * as I/O is concerned.
2170 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2174 spa_vdev_state_enter(spa, SCL_NONE);
2176 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2177 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2179 if (!vd->vdev_ops->vdev_op_leaf)
2180 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2183 * If the vdev is already faulted, then don't do anything.
2185 if (vd->vdev_faulted || vd->vdev_degraded)
2186 return (spa_vdev_state_exit(spa, NULL, 0));
2188 vd->vdev_degraded = 1ULL;
2189 if (!vdev_is_dead(vd))
2190 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2193 return (spa_vdev_state_exit(spa, vd, 0));
2197 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2198 * any attached spare device should be detached when the device finishes
2199 * resilvering. Second, the online should be treated like a 'test' online case,
2200 * so no FMA events are generated if the device fails to open.
2203 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2205 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2207 spa_vdev_state_enter(spa, SCL_NONE);
2209 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2210 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2212 if (!vd->vdev_ops->vdev_op_leaf)
2213 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2216 vd->vdev_offline = B_FALSE;
2217 vd->vdev_tmpoffline = B_FALSE;
2218 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2219 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2221 /* XXX - L2ARC 1.0 does not support expansion */
2222 if (!vd->vdev_aux) {
2223 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2224 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2228 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2230 if (!vd->vdev_aux) {
2231 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2232 pvd->vdev_expanding = B_FALSE;
2236 *newstate = vd->vdev_state;
2237 if ((flags & ZFS_ONLINE_UNSPARE) &&
2238 !vdev_is_dead(vd) && vd->vdev_parent &&
2239 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2240 vd->vdev_parent->vdev_child[0] == vd)
2241 vd->vdev_unspare = B_TRUE;
2243 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2245 /* XXX - L2ARC 1.0 does not support expansion */
2247 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2248 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2250 return (spa_vdev_state_exit(spa, vd, 0));
2254 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2258 uint64_t generation;
2259 metaslab_group_t *mg;
2262 spa_vdev_state_enter(spa, SCL_ALLOC);
2264 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2265 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2267 if (!vd->vdev_ops->vdev_op_leaf)
2268 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2272 generation = spa->spa_config_generation + 1;
2275 * If the device isn't already offline, try to offline it.
2277 if (!vd->vdev_offline) {
2279 * If this device has the only valid copy of some data,
2280 * don't allow it to be offlined. Log devices are always
2283 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2284 vdev_dtl_required(vd))
2285 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2288 * If the top-level is a slog and it has had allocations
2289 * then proceed. We check that the vdev's metaslab group
2290 * is not NULL since it's possible that we may have just
2291 * added this vdev but not yet initialized its metaslabs.
2293 if (tvd->vdev_islog && mg != NULL) {
2295 * Prevent any future allocations.
2297 metaslab_group_passivate(mg);
2298 (void) spa_vdev_state_exit(spa, vd, 0);
2300 error = spa_offline_log(spa);
2302 spa_vdev_state_enter(spa, SCL_ALLOC);
2305 * Check to see if the config has changed.
2307 if (error || generation != spa->spa_config_generation) {
2308 metaslab_group_activate(mg);
2310 return (spa_vdev_state_exit(spa,
2312 (void) spa_vdev_state_exit(spa, vd, 0);
2315 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2319 * Offline this device and reopen its top-level vdev.
2320 * If the top-level vdev is a log device then just offline
2321 * it. Otherwise, if this action results in the top-level
2322 * vdev becoming unusable, undo it and fail the request.
2324 vd->vdev_offline = B_TRUE;
2327 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2328 vdev_is_dead(tvd)) {
2329 vd->vdev_offline = B_FALSE;
2331 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2335 * Add the device back into the metaslab rotor so that
2336 * once we online the device it's open for business.
2338 if (tvd->vdev_islog && mg != NULL)
2339 metaslab_group_activate(mg);
2342 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2344 return (spa_vdev_state_exit(spa, vd, 0));
2348 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2352 mutex_enter(&spa->spa_vdev_top_lock);
2353 error = vdev_offline_locked(spa, guid, flags);
2354 mutex_exit(&spa->spa_vdev_top_lock);
2360 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2361 * vdev_offline(), we assume the spa config is locked. We also clear all
2362 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2365 vdev_clear(spa_t *spa, vdev_t *vd)
2367 vdev_t *rvd = spa->spa_root_vdev;
2370 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2375 vd->vdev_stat.vs_read_errors = 0;
2376 vd->vdev_stat.vs_write_errors = 0;
2377 vd->vdev_stat.vs_checksum_errors = 0;
2379 for (c = 0; c < vd->vdev_children; c++)
2380 vdev_clear(spa, vd->vdev_child[c]);
2383 * If we're in the FAULTED state or have experienced failed I/O, then
2384 * clear the persistent state and attempt to reopen the device. We
2385 * also mark the vdev config dirty, so that the new faulted state is
2386 * written out to disk.
2388 if (vd->vdev_faulted || vd->vdev_degraded ||
2389 !vdev_readable(vd) || !vdev_writeable(vd)) {
2392 * When reopening in reponse to a clear event, it may be due to
2393 * a fmadm repair request. In this case, if the device is
2394 * still broken, we want to still post the ereport again.
2396 vd->vdev_forcefault = B_TRUE;
2398 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2399 vd->vdev_cant_read = B_FALSE;
2400 vd->vdev_cant_write = B_FALSE;
2402 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2404 vd->vdev_forcefault = B_FALSE;
2406 if (vd != rvd && vdev_writeable(vd->vdev_top))
2407 vdev_state_dirty(vd->vdev_top);
2409 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2410 spa_async_request(spa, SPA_ASYNC_RESILVER);
2412 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR);
2416 * When clearing a FMA-diagnosed fault, we always want to
2417 * unspare the device, as we assume that the original spare was
2418 * done in response to the FMA fault.
2420 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2421 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2422 vd->vdev_parent->vdev_child[0] == vd)
2423 vd->vdev_unspare = B_TRUE;
2427 vdev_is_dead(vdev_t *vd)
2430 * Holes and missing devices are always considered "dead".
2431 * This simplifies the code since we don't have to check for
2432 * these types of devices in the various code paths.
2433 * Instead we rely on the fact that we skip over dead devices
2434 * before issuing I/O to them.
2436 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2437 vd->vdev_ops == &vdev_missing_ops);
2441 vdev_readable(vdev_t *vd)
2443 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2447 vdev_writeable(vdev_t *vd)
2449 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2453 vdev_allocatable(vdev_t *vd)
2455 uint64_t state = vd->vdev_state;
2458 * We currently allow allocations from vdevs which may be in the
2459 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2460 * fails to reopen then we'll catch it later when we're holding
2461 * the proper locks. Note that we have to get the vdev state
2462 * in a local variable because although it changes atomically,
2463 * we're asking two separate questions about it.
2465 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2466 !vd->vdev_cant_write && !vd->vdev_ishole);
2470 vdev_accessible(vdev_t *vd, zio_t *zio)
2472 ASSERT(zio->io_vd == vd);
2474 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2477 if (zio->io_type == ZIO_TYPE_READ)
2478 return (!vd->vdev_cant_read);
2480 if (zio->io_type == ZIO_TYPE_WRITE)
2481 return (!vd->vdev_cant_write);
2487 * Get statistics for the given vdev.
2490 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2492 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2495 mutex_enter(&vd->vdev_stat_lock);
2496 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2497 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2498 vs->vs_state = vd->vdev_state;
2499 vs->vs_rsize = vdev_get_min_asize(vd);
2500 if (vd->vdev_ops->vdev_op_leaf)
2501 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2502 mutex_exit(&vd->vdev_stat_lock);
2505 * If we're getting stats on the root vdev, aggregate the I/O counts
2506 * over all top-level vdevs (i.e. the direct children of the root).
2509 for (c = 0; c < rvd->vdev_children; c++) {
2510 vdev_t *cvd = rvd->vdev_child[c];
2511 vdev_stat_t *cvs = &cvd->vdev_stat;
2513 mutex_enter(&vd->vdev_stat_lock);
2514 for (t = 0; t < ZIO_TYPES; t++) {
2515 vs->vs_ops[t] += cvs->vs_ops[t];
2516 vs->vs_bytes[t] += cvs->vs_bytes[t];
2518 cvs->vs_scan_removing = cvd->vdev_removing;
2519 mutex_exit(&vd->vdev_stat_lock);
2525 vdev_clear_stats(vdev_t *vd)
2527 mutex_enter(&vd->vdev_stat_lock);
2528 vd->vdev_stat.vs_space = 0;
2529 vd->vdev_stat.vs_dspace = 0;
2530 vd->vdev_stat.vs_alloc = 0;
2531 mutex_exit(&vd->vdev_stat_lock);
2535 vdev_scan_stat_init(vdev_t *vd)
2537 vdev_stat_t *vs = &vd->vdev_stat;
2540 for (c = 0; c < vd->vdev_children; c++)
2541 vdev_scan_stat_init(vd->vdev_child[c]);
2543 mutex_enter(&vd->vdev_stat_lock);
2544 vs->vs_scan_processed = 0;
2545 mutex_exit(&vd->vdev_stat_lock);
2549 vdev_stat_update(zio_t *zio, uint64_t psize)
2551 spa_t *spa = zio->io_spa;
2552 vdev_t *rvd = spa->spa_root_vdev;
2553 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2555 uint64_t txg = zio->io_txg;
2556 vdev_stat_t *vs = &vd->vdev_stat;
2557 zio_type_t type = zio->io_type;
2558 int flags = zio->io_flags;
2561 * If this i/o is a gang leader, it didn't do any actual work.
2563 if (zio->io_gang_tree)
2566 if (zio->io_error == 0) {
2568 * If this is a root i/o, don't count it -- we've already
2569 * counted the top-level vdevs, and vdev_get_stats() will
2570 * aggregate them when asked. This reduces contention on
2571 * the root vdev_stat_lock and implicitly handles blocks
2572 * that compress away to holes, for which there is no i/o.
2573 * (Holes never create vdev children, so all the counters
2574 * remain zero, which is what we want.)
2576 * Note: this only applies to successful i/o (io_error == 0)
2577 * because unlike i/o counts, errors are not additive.
2578 * When reading a ditto block, for example, failure of
2579 * one top-level vdev does not imply a root-level error.
2584 ASSERT(vd == zio->io_vd);
2586 if (flags & ZIO_FLAG_IO_BYPASS)
2589 mutex_enter(&vd->vdev_stat_lock);
2591 if (flags & ZIO_FLAG_IO_REPAIR) {
2592 if (flags & ZIO_FLAG_SCAN_THREAD) {
2593 dsl_scan_phys_t *scn_phys =
2594 &spa->spa_dsl_pool->dp_scan->scn_phys;
2595 uint64_t *processed = &scn_phys->scn_processed;
2598 if (vd->vdev_ops->vdev_op_leaf)
2599 atomic_add_64(processed, psize);
2600 vs->vs_scan_processed += psize;
2603 if (flags & ZIO_FLAG_SELF_HEAL)
2604 vs->vs_self_healed += psize;
2608 vs->vs_bytes[type] += psize;
2610 mutex_exit(&vd->vdev_stat_lock);
2614 if (flags & ZIO_FLAG_SPECULATIVE)
2618 * If this is an I/O error that is going to be retried, then ignore the
2619 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2620 * hard errors, when in reality they can happen for any number of
2621 * innocuous reasons (bus resets, MPxIO link failure, etc).
2623 if (zio->io_error == EIO &&
2624 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2628 * Intent logs writes won't propagate their error to the root
2629 * I/O so don't mark these types of failures as pool-level
2632 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2635 mutex_enter(&vd->vdev_stat_lock);
2636 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2637 if (zio->io_error == ECKSUM)
2638 vs->vs_checksum_errors++;
2640 vs->vs_read_errors++;
2642 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2643 vs->vs_write_errors++;
2644 mutex_exit(&vd->vdev_stat_lock);
2646 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2647 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2648 (flags & ZIO_FLAG_SCAN_THREAD) ||
2649 spa->spa_claiming)) {
2651 * This is either a normal write (not a repair), or it's
2652 * a repair induced by the scrub thread, or it's a repair
2653 * made by zil_claim() during spa_load() in the first txg.
2654 * In the normal case, we commit the DTL change in the same
2655 * txg as the block was born. In the scrub-induced repair
2656 * case, we know that scrubs run in first-pass syncing context,
2657 * so we commit the DTL change in spa_syncing_txg(spa).
2658 * In the zil_claim() case, we commit in spa_first_txg(spa).
2660 * We currently do not make DTL entries for failed spontaneous
2661 * self-healing writes triggered by normal (non-scrubbing)
2662 * reads, because we have no transactional context in which to
2663 * do so -- and it's not clear that it'd be desirable anyway.
2665 if (vd->vdev_ops->vdev_op_leaf) {
2666 uint64_t commit_txg = txg;
2667 if (flags & ZIO_FLAG_SCAN_THREAD) {
2668 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2669 ASSERT(spa_sync_pass(spa) == 1);
2670 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2671 commit_txg = spa_syncing_txg(spa);
2672 } else if (spa->spa_claiming) {
2673 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2674 commit_txg = spa_first_txg(spa);
2676 ASSERT(commit_txg >= spa_syncing_txg(spa));
2677 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2679 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2680 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2681 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2684 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2689 * Update the in-core space usage stats for this vdev, its metaslab class,
2690 * and the root vdev.
2693 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2694 int64_t space_delta)
2696 int64_t dspace_delta = space_delta;
2697 spa_t *spa = vd->vdev_spa;
2698 vdev_t *rvd = spa->spa_root_vdev;
2699 metaslab_group_t *mg = vd->vdev_mg;
2700 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2702 ASSERT(vd == vd->vdev_top);
2705 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2706 * factor. We must calculate this here and not at the root vdev
2707 * because the root vdev's psize-to-asize is simply the max of its
2708 * childrens', thus not accurate enough for us.
2710 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2711 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2712 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2713 vd->vdev_deflate_ratio;
2715 mutex_enter(&vd->vdev_stat_lock);
2716 vd->vdev_stat.vs_alloc += alloc_delta;
2717 vd->vdev_stat.vs_space += space_delta;
2718 vd->vdev_stat.vs_dspace += dspace_delta;
2719 mutex_exit(&vd->vdev_stat_lock);
2721 if (mc == spa_normal_class(spa)) {
2722 mutex_enter(&rvd->vdev_stat_lock);
2723 rvd->vdev_stat.vs_alloc += alloc_delta;
2724 rvd->vdev_stat.vs_space += space_delta;
2725 rvd->vdev_stat.vs_dspace += dspace_delta;
2726 mutex_exit(&rvd->vdev_stat_lock);
2730 ASSERT(rvd == vd->vdev_parent);
2731 ASSERT(vd->vdev_ms_count != 0);
2733 metaslab_class_space_update(mc,
2734 alloc_delta, defer_delta, space_delta, dspace_delta);
2739 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2740 * so that it will be written out next time the vdev configuration is synced.
2741 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2744 vdev_config_dirty(vdev_t *vd)
2746 spa_t *spa = vd->vdev_spa;
2747 vdev_t *rvd = spa->spa_root_vdev;
2750 ASSERT(spa_writeable(spa));
2753 * If this is an aux vdev (as with l2cache and spare devices), then we
2754 * update the vdev config manually and set the sync flag.
2756 if (vd->vdev_aux != NULL) {
2757 spa_aux_vdev_t *sav = vd->vdev_aux;
2761 for (c = 0; c < sav->sav_count; c++) {
2762 if (sav->sav_vdevs[c] == vd)
2766 if (c == sav->sav_count) {
2768 * We're being removed. There's nothing more to do.
2770 ASSERT(sav->sav_sync == B_TRUE);
2774 sav->sav_sync = B_TRUE;
2776 if (nvlist_lookup_nvlist_array(sav->sav_config,
2777 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2778 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2779 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2785 * Setting the nvlist in the middle if the array is a little
2786 * sketchy, but it will work.
2788 nvlist_free(aux[c]);
2789 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2795 * The dirty list is protected by the SCL_CONFIG lock. The caller
2796 * must either hold SCL_CONFIG as writer, or must be the sync thread
2797 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2798 * so this is sufficient to ensure mutual exclusion.
2800 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2801 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2802 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2805 for (c = 0; c < rvd->vdev_children; c++)
2806 vdev_config_dirty(rvd->vdev_child[c]);
2808 ASSERT(vd == vd->vdev_top);
2810 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2812 list_insert_head(&spa->spa_config_dirty_list, vd);
2817 vdev_config_clean(vdev_t *vd)
2819 spa_t *spa = vd->vdev_spa;
2821 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2822 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2823 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2825 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2826 list_remove(&spa->spa_config_dirty_list, vd);
2830 * Mark a top-level vdev's state as dirty, so that the next pass of
2831 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2832 * the state changes from larger config changes because they require
2833 * much less locking, and are often needed for administrative actions.
2836 vdev_state_dirty(vdev_t *vd)
2838 spa_t *spa = vd->vdev_spa;
2840 ASSERT(spa_writeable(spa));
2841 ASSERT(vd == vd->vdev_top);
2844 * The state list is protected by the SCL_STATE lock. The caller
2845 * must either hold SCL_STATE as writer, or must be the sync thread
2846 * (which holds SCL_STATE as reader). There's only one sync thread,
2847 * so this is sufficient to ensure mutual exclusion.
2849 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2850 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2851 spa_config_held(spa, SCL_STATE, RW_READER)));
2853 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2854 list_insert_head(&spa->spa_state_dirty_list, vd);
2858 vdev_state_clean(vdev_t *vd)
2860 spa_t *spa = vd->vdev_spa;
2862 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2863 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2864 spa_config_held(spa, SCL_STATE, RW_READER)));
2866 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2867 list_remove(&spa->spa_state_dirty_list, vd);
2871 * Propagate vdev state up from children to parent.
2874 vdev_propagate_state(vdev_t *vd)
2876 spa_t *spa = vd->vdev_spa;
2877 vdev_t *rvd = spa->spa_root_vdev;
2878 int degraded = 0, faulted = 0;
2883 if (vd->vdev_children > 0) {
2884 for (c = 0; c < vd->vdev_children; c++) {
2885 child = vd->vdev_child[c];
2888 * Don't factor holes into the decision.
2890 if (child->vdev_ishole)
2893 if (!vdev_readable(child) ||
2894 (!vdev_writeable(child) && spa_writeable(spa))) {
2896 * Root special: if there is a top-level log
2897 * device, treat the root vdev as if it were
2900 if (child->vdev_islog && vd == rvd)
2904 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2908 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2912 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2915 * Root special: if there is a top-level vdev that cannot be
2916 * opened due to corrupted metadata, then propagate the root
2917 * vdev's aux state as 'corrupt' rather than 'insufficient
2920 if (corrupted && vd == rvd &&
2921 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2922 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2923 VDEV_AUX_CORRUPT_DATA);
2926 if (vd->vdev_parent)
2927 vdev_propagate_state(vd->vdev_parent);
2931 * Set a vdev's state. If this is during an open, we don't update the parent
2932 * state, because we're in the process of opening children depth-first.
2933 * Otherwise, we propagate the change to the parent.
2935 * If this routine places a device in a faulted state, an appropriate ereport is
2939 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2941 uint64_t save_state;
2942 spa_t *spa = vd->vdev_spa;
2944 if (state == vd->vdev_state) {
2945 vd->vdev_stat.vs_aux = aux;
2949 save_state = vd->vdev_state;
2951 vd->vdev_state = state;
2952 vd->vdev_stat.vs_aux = aux;
2955 * If we are setting the vdev state to anything but an open state, then
2956 * always close the underlying device unless the device has requested
2957 * a delayed close (i.e. we're about to remove or fault the device).
2958 * Otherwise, we keep accessible but invalid devices open forever.
2959 * We don't call vdev_close() itself, because that implies some extra
2960 * checks (offline, etc) that we don't want here. This is limited to
2961 * leaf devices, because otherwise closing the device will affect other
2964 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2965 vd->vdev_ops->vdev_op_leaf)
2966 vd->vdev_ops->vdev_op_close(vd);
2969 * If we have brought this vdev back into service, we need
2970 * to notify fmd so that it can gracefully repair any outstanding
2971 * cases due to a missing device. We do this in all cases, even those
2972 * that probably don't correlate to a repaired fault. This is sure to
2973 * catch all cases, and we let the zfs-retire agent sort it out. If
2974 * this is a transient state it's OK, as the retire agent will
2975 * double-check the state of the vdev before repairing it.
2977 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2978 vd->vdev_prevstate != state)
2979 zfs_post_state_change(spa, vd);
2981 if (vd->vdev_removed &&
2982 state == VDEV_STATE_CANT_OPEN &&
2983 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2985 * If the previous state is set to VDEV_STATE_REMOVED, then this
2986 * device was previously marked removed and someone attempted to
2987 * reopen it. If this failed due to a nonexistent device, then
2988 * keep the device in the REMOVED state. We also let this be if
2989 * it is one of our special test online cases, which is only
2990 * attempting to online the device and shouldn't generate an FMA
2993 vd->vdev_state = VDEV_STATE_REMOVED;
2994 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2995 } else if (state == VDEV_STATE_REMOVED) {
2996 vd->vdev_removed = B_TRUE;
2997 } else if (state == VDEV_STATE_CANT_OPEN) {
2999 * If we fail to open a vdev during an import or recovery, we
3000 * mark it as "not available", which signifies that it was
3001 * never there to begin with. Failure to open such a device
3002 * is not considered an error.
3004 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3005 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3006 vd->vdev_ops->vdev_op_leaf)
3007 vd->vdev_not_present = 1;
3010 * Post the appropriate ereport. If the 'prevstate' field is
3011 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3012 * that this is part of a vdev_reopen(). In this case, we don't
3013 * want to post the ereport if the device was already in the
3014 * CANT_OPEN state beforehand.
3016 * If the 'checkremove' flag is set, then this is an attempt to
3017 * online the device in response to an insertion event. If we
3018 * hit this case, then we have detected an insertion event for a
3019 * faulted or offline device that wasn't in the removed state.
3020 * In this scenario, we don't post an ereport because we are
3021 * about to replace the device, or attempt an online with
3022 * vdev_forcefault, which will generate the fault for us.
3024 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3025 !vd->vdev_not_present && !vd->vdev_checkremove &&
3026 vd != spa->spa_root_vdev) {
3030 case VDEV_AUX_OPEN_FAILED:
3031 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3033 case VDEV_AUX_CORRUPT_DATA:
3034 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3036 case VDEV_AUX_NO_REPLICAS:
3037 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3039 case VDEV_AUX_BAD_GUID_SUM:
3040 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3042 case VDEV_AUX_TOO_SMALL:
3043 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3045 case VDEV_AUX_BAD_LABEL:
3046 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3049 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3052 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3055 /* Erase any notion of persistent removed state */
3056 vd->vdev_removed = B_FALSE;
3058 vd->vdev_removed = B_FALSE;
3061 if (!isopen && vd->vdev_parent)
3062 vdev_propagate_state(vd->vdev_parent);
3066 * Check the vdev configuration to ensure that it's capable of supporting
3070 vdev_is_bootable(vdev_t *vd)
3072 #if defined(__sun__) || defined(__sun)
3074 * Currently, we do not support RAID-Z or partial configuration.
3075 * In addition, only a single top-level vdev is allowed and none of the
3076 * leaves can be wholedisks.
3080 if (!vd->vdev_ops->vdev_op_leaf) {
3081 char *vdev_type = vd->vdev_ops->vdev_op_type;
3083 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3084 vd->vdev_children > 1) {
3086 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3087 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3090 } else if (vd->vdev_wholedisk == 1) {
3094 for (c = 0; c < vd->vdev_children; c++) {
3095 if (!vdev_is_bootable(vd->vdev_child[c]))
3098 #endif /* __sun__ || __sun */
3103 * Load the state from the original vdev tree (ovd) which
3104 * we've retrieved from the MOS config object. If the original
3105 * vdev was offline or faulted then we transfer that state to the
3106 * device in the current vdev tree (nvd).
3109 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3113 ASSERT(nvd->vdev_top->vdev_islog);
3114 ASSERT(spa_config_held(nvd->vdev_spa,
3115 SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3116 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3118 for (c = 0; c < nvd->vdev_children; c++)
3119 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3121 if (nvd->vdev_ops->vdev_op_leaf) {
3123 * Restore the persistent vdev state
3125 nvd->vdev_offline = ovd->vdev_offline;
3126 nvd->vdev_faulted = ovd->vdev_faulted;
3127 nvd->vdev_degraded = ovd->vdev_degraded;
3128 nvd->vdev_removed = ovd->vdev_removed;
3133 * Determine if a log device has valid content. If the vdev was
3134 * removed or faulted in the MOS config then we know that
3135 * the content on the log device has already been written to the pool.
3138 vdev_log_state_valid(vdev_t *vd)
3142 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3146 for (c = 0; c < vd->vdev_children; c++)
3147 if (vdev_log_state_valid(vd->vdev_child[c]))
3154 * Expand a vdev if possible.
3157 vdev_expand(vdev_t *vd, uint64_t txg)
3159 ASSERT(vd->vdev_top == vd);
3160 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3162 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3163 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3164 vdev_config_dirty(vd);
3172 vdev_split(vdev_t *vd)
3174 vdev_t *cvd, *pvd = vd->vdev_parent;
3176 vdev_remove_child(pvd, vd);
3177 vdev_compact_children(pvd);
3179 cvd = pvd->vdev_child[0];
3180 if (pvd->vdev_children == 1) {
3181 vdev_remove_parent(cvd);
3182 cvd->vdev_splitting = B_TRUE;
3184 vdev_propagate_state(cvd);
3187 #if defined(_KERNEL) && defined(HAVE_SPL)
3188 EXPORT_SYMBOL(vdev_fault);
3189 EXPORT_SYMBOL(vdev_degrade);
3190 EXPORT_SYMBOL(vdev_online);
3191 EXPORT_SYMBOL(vdev_offline);
3192 EXPORT_SYMBOL(vdev_clear);
3194 module_param(zfs_scrub_limit, int, 0644);
3195 MODULE_PARM_DESC(zfs_scrub_limit, "Max scrub/resilver I/O per leaf vdev");