4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa_impl.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
41 #include <sys/fs/zfs.h>
44 #include <sys/dsl_scan.h>
46 SYSCTL_DECL(_vfs_zfs);
47 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
50 * Virtual device management.
53 static vdev_ops_t *vdev_ops_table[] = {
72 * Given a vdev type, return the appropriate ops vector.
75 vdev_getops(const char *type)
77 vdev_ops_t *ops, **opspp;
79 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
80 if (strcmp(ops->vdev_op_type, type) == 0)
87 * Default asize function: return the MAX of psize with the asize of
88 * all children. This is what's used by anything other than RAID-Z.
91 vdev_default_asize(vdev_t *vd, uint64_t psize)
93 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
96 for (int c = 0; c < vd->vdev_children; c++) {
97 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
98 asize = MAX(asize, csize);
105 * Get the minimum allocatable size. We define the allocatable size as
106 * the vdev's asize rounded to the nearest metaslab. This allows us to
107 * replace or attach devices which don't have the same physical size but
108 * can still satisfy the same number of allocations.
111 vdev_get_min_asize(vdev_t *vd)
113 vdev_t *pvd = vd->vdev_parent;
116 * If our parent is NULL (inactive spare or cache) or is the root,
117 * just return our own asize.
120 return (vd->vdev_asize);
123 * The top-level vdev just returns the allocatable size rounded
124 * to the nearest metaslab.
126 if (vd == vd->vdev_top)
127 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
130 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
131 * so each child must provide at least 1/Nth of its asize.
133 if (pvd->vdev_ops == &vdev_raidz_ops)
134 return (pvd->vdev_min_asize / pvd->vdev_children);
136 return (pvd->vdev_min_asize);
140 vdev_set_min_asize(vdev_t *vd)
142 vd->vdev_min_asize = vdev_get_min_asize(vd);
144 for (int c = 0; c < vd->vdev_children; c++)
145 vdev_set_min_asize(vd->vdev_child[c]);
149 vdev_lookup_top(spa_t *spa, uint64_t vdev)
151 vdev_t *rvd = spa->spa_root_vdev;
153 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
155 if (vdev < rvd->vdev_children) {
156 ASSERT(rvd->vdev_child[vdev] != NULL);
157 return (rvd->vdev_child[vdev]);
164 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
168 if (vd->vdev_guid == guid)
171 for (int c = 0; c < vd->vdev_children; c++)
172 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
180 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
182 size_t oldsize, newsize;
183 uint64_t id = cvd->vdev_id;
186 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
187 ASSERT(cvd->vdev_parent == NULL);
189 cvd->vdev_parent = pvd;
194 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
196 oldsize = pvd->vdev_children * sizeof (vdev_t *);
197 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
198 newsize = pvd->vdev_children * sizeof (vdev_t *);
200 newchild = kmem_zalloc(newsize, KM_SLEEP);
201 if (pvd->vdev_child != NULL) {
202 bcopy(pvd->vdev_child, newchild, oldsize);
203 kmem_free(pvd->vdev_child, oldsize);
206 pvd->vdev_child = newchild;
207 pvd->vdev_child[id] = cvd;
209 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
210 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
213 * Walk up all ancestors to update guid sum.
215 for (; pvd != NULL; pvd = pvd->vdev_parent)
216 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
220 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
223 uint_t id = cvd->vdev_id;
225 ASSERT(cvd->vdev_parent == pvd);
230 ASSERT(id < pvd->vdev_children);
231 ASSERT(pvd->vdev_child[id] == cvd);
233 pvd->vdev_child[id] = NULL;
234 cvd->vdev_parent = NULL;
236 for (c = 0; c < pvd->vdev_children; c++)
237 if (pvd->vdev_child[c])
240 if (c == pvd->vdev_children) {
241 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
242 pvd->vdev_child = NULL;
243 pvd->vdev_children = 0;
247 * Walk up all ancestors to update guid sum.
249 for (; pvd != NULL; pvd = pvd->vdev_parent)
250 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
254 * Remove any holes in the child array.
257 vdev_compact_children(vdev_t *pvd)
259 vdev_t **newchild, *cvd;
260 int oldc = pvd->vdev_children;
263 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
265 for (int c = newc = 0; c < oldc; c++)
266 if (pvd->vdev_child[c])
269 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
271 for (int c = newc = 0; c < oldc; c++) {
272 if ((cvd = pvd->vdev_child[c]) != NULL) {
273 newchild[newc] = cvd;
274 cvd->vdev_id = newc++;
278 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
279 pvd->vdev_child = newchild;
280 pvd->vdev_children = newc;
284 * Allocate and minimally initialize a vdev_t.
287 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
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 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
324 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
325 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
326 for (int t = 0; t < DTL_TYPES; t++) {
327 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
330 txg_list_create(&vd->vdev_ms_list,
331 offsetof(struct metaslab, ms_txg_node));
332 txg_list_create(&vd->vdev_dtl_list,
333 offsetof(struct vdev, vdev_dtl_node));
334 vd->vdev_stat.vs_timestamp = gethrtime();
342 * Allocate a new vdev. The 'alloctype' is used to control whether we are
343 * creating a new vdev or loading an existing one - the behavior is slightly
344 * different for each case.
347 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
352 uint64_t guid = 0, islog, nparity;
355 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
357 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
360 if ((ops = vdev_getops(type)) == NULL)
364 * If this is a load, get the vdev guid from the nvlist.
365 * Otherwise, vdev_alloc_common() will generate one for us.
367 if (alloctype == VDEV_ALLOC_LOAD) {
370 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
374 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
376 } else if (alloctype == VDEV_ALLOC_SPARE) {
377 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
379 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
380 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
382 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
383 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
388 * The first allocated vdev must be of type 'root'.
390 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
394 * Determine whether we're a log vdev.
397 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
398 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
401 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
405 * Set the nparity property for RAID-Z vdevs.
408 if (ops == &vdev_raidz_ops) {
409 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
411 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
414 * Previous versions could only support 1 or 2 parity
418 spa_version(spa) < SPA_VERSION_RAIDZ2)
421 spa_version(spa) < SPA_VERSION_RAIDZ3)
425 * We require the parity to be specified for SPAs that
426 * support multiple parity levels.
428 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
431 * Otherwise, we default to 1 parity device for RAID-Z.
438 ASSERT(nparity != -1ULL);
440 vd = vdev_alloc_common(spa, id, guid, ops);
442 vd->vdev_islog = islog;
443 vd->vdev_nparity = nparity;
445 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
446 vd->vdev_path = spa_strdup(vd->vdev_path);
447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
448 vd->vdev_devid = spa_strdup(vd->vdev_devid);
449 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
450 &vd->vdev_physpath) == 0)
451 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
452 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
453 vd->vdev_fru = spa_strdup(vd->vdev_fru);
456 * Set the whole_disk property. If it's not specified, leave the value
459 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
460 &vd->vdev_wholedisk) != 0)
461 vd->vdev_wholedisk = -1ULL;
464 * Look for the 'not present' flag. This will only be set if the device
465 * was not present at the time of import.
467 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
468 &vd->vdev_not_present);
471 * Get the alignment requirement.
473 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
476 * Retrieve the vdev creation time.
478 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
482 * If we're a top-level vdev, try to load the allocation parameters.
484 if (parent && !parent->vdev_parent &&
485 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
486 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
488 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
490 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
496 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
497 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
498 alloctype == VDEV_ALLOC_ADD ||
499 alloctype == VDEV_ALLOC_SPLIT ||
500 alloctype == VDEV_ALLOC_ROOTPOOL);
501 vd->vdev_mg = metaslab_group_create(islog ?
502 spa_log_class(spa) : spa_normal_class(spa), vd);
506 * If we're a leaf vdev, try to load the DTL object and other state.
508 if (vd->vdev_ops->vdev_op_leaf &&
509 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
510 alloctype == VDEV_ALLOC_ROOTPOOL)) {
511 if (alloctype == VDEV_ALLOC_LOAD) {
512 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
513 &vd->vdev_dtl_smo.smo_object);
514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
518 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
521 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
522 &spare) == 0 && spare)
526 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
529 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
530 &vd->vdev_resilvering);
533 * When importing a pool, we want to ignore the persistent fault
534 * state, as the diagnosis made on another system may not be
535 * valid in the current context. Local vdevs will
536 * remain in the faulted state.
538 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
539 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
541 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
543 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
546 if (vd->vdev_faulted || vd->vdev_degraded) {
550 VDEV_AUX_ERR_EXCEEDED;
551 if (nvlist_lookup_string(nv,
552 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
553 strcmp(aux, "external") == 0)
554 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
560 * Add ourselves to the parent's list of children.
562 vdev_add_child(parent, vd);
570 vdev_free(vdev_t *vd)
572 spa_t *spa = vd->vdev_spa;
575 * vdev_free() implies closing the vdev first. This is simpler than
576 * trying to ensure complicated semantics for all callers.
580 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
581 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
586 for (int c = 0; c < vd->vdev_children; c++)
587 vdev_free(vd->vdev_child[c]);
589 ASSERT(vd->vdev_child == NULL);
590 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
593 * Discard allocation state.
595 if (vd->vdev_mg != NULL) {
596 vdev_metaslab_fini(vd);
597 metaslab_group_destroy(vd->vdev_mg);
600 ASSERT0(vd->vdev_stat.vs_space);
601 ASSERT0(vd->vdev_stat.vs_dspace);
602 ASSERT0(vd->vdev_stat.vs_alloc);
605 * Remove this vdev from its parent's child list.
607 vdev_remove_child(vd->vdev_parent, vd);
609 ASSERT(vd->vdev_parent == NULL);
612 * Clean up vdev structure.
618 spa_strfree(vd->vdev_path);
620 spa_strfree(vd->vdev_devid);
621 if (vd->vdev_physpath)
622 spa_strfree(vd->vdev_physpath);
624 spa_strfree(vd->vdev_fru);
626 if (vd->vdev_isspare)
627 spa_spare_remove(vd);
628 if (vd->vdev_isl2cache)
629 spa_l2cache_remove(vd);
631 txg_list_destroy(&vd->vdev_ms_list);
632 txg_list_destroy(&vd->vdev_dtl_list);
634 mutex_enter(&vd->vdev_dtl_lock);
635 for (int t = 0; t < DTL_TYPES; t++) {
636 space_map_unload(&vd->vdev_dtl[t]);
637 space_map_destroy(&vd->vdev_dtl[t]);
639 mutex_exit(&vd->vdev_dtl_lock);
641 mutex_destroy(&vd->vdev_dtl_lock);
642 mutex_destroy(&vd->vdev_stat_lock);
643 mutex_destroy(&vd->vdev_probe_lock);
645 if (vd == spa->spa_root_vdev)
646 spa->spa_root_vdev = NULL;
648 kmem_free(vd, sizeof (vdev_t));
652 * Transfer top-level vdev state from svd to tvd.
655 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
657 spa_t *spa = svd->vdev_spa;
662 ASSERT(tvd == tvd->vdev_top);
664 tvd->vdev_ms_array = svd->vdev_ms_array;
665 tvd->vdev_ms_shift = svd->vdev_ms_shift;
666 tvd->vdev_ms_count = svd->vdev_ms_count;
668 svd->vdev_ms_array = 0;
669 svd->vdev_ms_shift = 0;
670 svd->vdev_ms_count = 0;
673 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
674 tvd->vdev_mg = svd->vdev_mg;
675 tvd->vdev_ms = svd->vdev_ms;
680 if (tvd->vdev_mg != NULL)
681 tvd->vdev_mg->mg_vd = tvd;
683 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
684 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
685 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
687 svd->vdev_stat.vs_alloc = 0;
688 svd->vdev_stat.vs_space = 0;
689 svd->vdev_stat.vs_dspace = 0;
691 for (t = 0; t < TXG_SIZE; t++) {
692 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
693 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
694 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
695 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
696 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
697 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
700 if (list_link_active(&svd->vdev_config_dirty_node)) {
701 vdev_config_clean(svd);
702 vdev_config_dirty(tvd);
705 if (list_link_active(&svd->vdev_state_dirty_node)) {
706 vdev_state_clean(svd);
707 vdev_state_dirty(tvd);
710 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
711 svd->vdev_deflate_ratio = 0;
713 tvd->vdev_islog = svd->vdev_islog;
718 vdev_top_update(vdev_t *tvd, vdev_t *vd)
725 for (int c = 0; c < vd->vdev_children; c++)
726 vdev_top_update(tvd, vd->vdev_child[c]);
730 * Add a mirror/replacing vdev above an existing vdev.
733 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
735 spa_t *spa = cvd->vdev_spa;
736 vdev_t *pvd = cvd->vdev_parent;
739 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
741 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
743 mvd->vdev_asize = cvd->vdev_asize;
744 mvd->vdev_min_asize = cvd->vdev_min_asize;
745 mvd->vdev_max_asize = cvd->vdev_max_asize;
746 mvd->vdev_ashift = cvd->vdev_ashift;
747 mvd->vdev_state = cvd->vdev_state;
748 mvd->vdev_crtxg = cvd->vdev_crtxg;
750 vdev_remove_child(pvd, cvd);
751 vdev_add_child(pvd, mvd);
752 cvd->vdev_id = mvd->vdev_children;
753 vdev_add_child(mvd, cvd);
754 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
756 if (mvd == mvd->vdev_top)
757 vdev_top_transfer(cvd, mvd);
763 * Remove a 1-way mirror/replacing vdev from the tree.
766 vdev_remove_parent(vdev_t *cvd)
768 vdev_t *mvd = cvd->vdev_parent;
769 vdev_t *pvd = mvd->vdev_parent;
771 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
773 ASSERT(mvd->vdev_children == 1);
774 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
775 mvd->vdev_ops == &vdev_replacing_ops ||
776 mvd->vdev_ops == &vdev_spare_ops);
777 cvd->vdev_ashift = mvd->vdev_ashift;
779 vdev_remove_child(mvd, cvd);
780 vdev_remove_child(pvd, mvd);
783 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
784 * Otherwise, we could have detached an offline device, and when we
785 * go to import the pool we'll think we have two top-level vdevs,
786 * instead of a different version of the same top-level vdev.
788 if (mvd->vdev_top == mvd) {
789 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
790 cvd->vdev_orig_guid = cvd->vdev_guid;
791 cvd->vdev_guid += guid_delta;
792 cvd->vdev_guid_sum += guid_delta;
794 cvd->vdev_id = mvd->vdev_id;
795 vdev_add_child(pvd, cvd);
796 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
798 if (cvd == cvd->vdev_top)
799 vdev_top_transfer(mvd, cvd);
801 ASSERT(mvd->vdev_children == 0);
806 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
808 spa_t *spa = vd->vdev_spa;
809 objset_t *mos = spa->spa_meta_objset;
811 uint64_t oldc = vd->vdev_ms_count;
812 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
816 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
819 * This vdev is not being allocated from yet or is a hole.
821 if (vd->vdev_ms_shift == 0)
824 ASSERT(!vd->vdev_ishole);
827 * Compute the raidz-deflation ratio. Note, we hard-code
828 * in 128k (1 << 17) because it is the current "typical" blocksize.
829 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
830 * or we will inconsistently account for existing bp's.
832 vd->vdev_deflate_ratio = (1 << 17) /
833 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
835 ASSERT(oldc <= newc);
837 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
840 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
841 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
845 vd->vdev_ms_count = newc;
847 for (m = oldc; m < newc; m++) {
848 space_map_obj_t smo = { 0, 0, 0 };
851 error = dmu_read(mos, vd->vdev_ms_array,
852 m * sizeof (uint64_t), sizeof (uint64_t), &object,
858 error = dmu_bonus_hold(mos, object, FTAG, &db);
861 ASSERT3U(db->db_size, >=, sizeof (smo));
862 bcopy(db->db_data, &smo, sizeof (smo));
863 ASSERT3U(smo.smo_object, ==, object);
864 dmu_buf_rele(db, FTAG);
867 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
868 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
872 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
875 * If the vdev is being removed we don't activate
876 * the metaslabs since we want to ensure that no new
877 * allocations are performed on this device.
879 if (oldc == 0 && !vd->vdev_removing)
880 metaslab_group_activate(vd->vdev_mg);
883 spa_config_exit(spa, SCL_ALLOC, FTAG);
889 vdev_metaslab_fini(vdev_t *vd)
892 uint64_t count = vd->vdev_ms_count;
894 if (vd->vdev_ms != NULL) {
895 metaslab_group_passivate(vd->vdev_mg);
896 for (m = 0; m < count; m++)
897 if (vd->vdev_ms[m] != NULL)
898 metaslab_fini(vd->vdev_ms[m]);
899 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
904 typedef struct vdev_probe_stats {
905 boolean_t vps_readable;
906 boolean_t vps_writeable;
908 } vdev_probe_stats_t;
911 vdev_probe_done(zio_t *zio)
913 spa_t *spa = zio->io_spa;
914 vdev_t *vd = zio->io_vd;
915 vdev_probe_stats_t *vps = zio->io_private;
917 ASSERT(vd->vdev_probe_zio != NULL);
919 if (zio->io_type == ZIO_TYPE_READ) {
920 if (zio->io_error == 0)
921 vps->vps_readable = 1;
922 if (zio->io_error == 0 && spa_writeable(spa)) {
923 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
924 zio->io_offset, zio->io_size, zio->io_data,
925 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
926 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
928 zio_buf_free(zio->io_data, zio->io_size);
930 } else if (zio->io_type == ZIO_TYPE_WRITE) {
931 if (zio->io_error == 0)
932 vps->vps_writeable = 1;
933 zio_buf_free(zio->io_data, zio->io_size);
934 } else if (zio->io_type == ZIO_TYPE_NULL) {
937 vd->vdev_cant_read |= !vps->vps_readable;
938 vd->vdev_cant_write |= !vps->vps_writeable;
940 if (vdev_readable(vd) &&
941 (vdev_writeable(vd) || !spa_writeable(spa))) {
944 ASSERT(zio->io_error != 0);
945 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
946 spa, vd, NULL, 0, 0);
947 zio->io_error = ENXIO;
950 mutex_enter(&vd->vdev_probe_lock);
951 ASSERT(vd->vdev_probe_zio == zio);
952 vd->vdev_probe_zio = NULL;
953 mutex_exit(&vd->vdev_probe_lock);
955 while ((pio = zio_walk_parents(zio)) != NULL)
956 if (!vdev_accessible(vd, pio))
957 pio->io_error = ENXIO;
959 kmem_free(vps, sizeof (*vps));
964 * Determine whether this device is accessible by reading and writing
965 * to several known locations: the pad regions of each vdev label
966 * but the first (which we leave alone in case it contains a VTOC).
969 vdev_probe(vdev_t *vd, zio_t *zio)
971 spa_t *spa = vd->vdev_spa;
972 vdev_probe_stats_t *vps = NULL;
975 ASSERT(vd->vdev_ops->vdev_op_leaf);
978 * Don't probe the probe.
980 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
984 * To prevent 'probe storms' when a device fails, we create
985 * just one probe i/o at a time. All zios that want to probe
986 * this vdev will become parents of the probe io.
988 mutex_enter(&vd->vdev_probe_lock);
990 if ((pio = vd->vdev_probe_zio) == NULL) {
991 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
993 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
994 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
997 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
999 * vdev_cant_read and vdev_cant_write can only
1000 * transition from TRUE to FALSE when we have the
1001 * SCL_ZIO lock as writer; otherwise they can only
1002 * transition from FALSE to TRUE. This ensures that
1003 * any zio looking at these values can assume that
1004 * failures persist for the life of the I/O. That's
1005 * important because when a device has intermittent
1006 * connectivity problems, we want to ensure that
1007 * they're ascribed to the device (ENXIO) and not
1010 * Since we hold SCL_ZIO as writer here, clear both
1011 * values so the probe can reevaluate from first
1014 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1015 vd->vdev_cant_read = B_FALSE;
1016 vd->vdev_cant_write = B_FALSE;
1019 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1020 vdev_probe_done, vps,
1021 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1024 * We can't change the vdev state in this context, so we
1025 * kick off an async task to do it on our behalf.
1028 vd->vdev_probe_wanted = B_TRUE;
1029 spa_async_request(spa, SPA_ASYNC_PROBE);
1034 zio_add_child(zio, pio);
1036 mutex_exit(&vd->vdev_probe_lock);
1039 ASSERT(zio != NULL);
1043 for (int l = 1; l < VDEV_LABELS; l++) {
1044 zio_nowait(zio_read_phys(pio, vd,
1045 vdev_label_offset(vd->vdev_psize, l,
1046 offsetof(vdev_label_t, vl_pad2)),
1047 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1048 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1049 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1060 vdev_open_child(void *arg)
1064 vd->vdev_open_thread = curthread;
1065 vd->vdev_open_error = vdev_open(vd);
1066 vd->vdev_open_thread = NULL;
1070 vdev_uses_zvols(vdev_t *vd)
1072 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1073 strlen(ZVOL_DIR)) == 0)
1075 for (int c = 0; c < vd->vdev_children; c++)
1076 if (vdev_uses_zvols(vd->vdev_child[c]))
1082 vdev_open_children(vdev_t *vd)
1085 int children = vd->vdev_children;
1088 * in order to handle pools on top of zvols, do the opens
1089 * in a single thread so that the same thread holds the
1090 * spa_namespace_lock
1092 if (B_TRUE || vdev_uses_zvols(vd)) {
1093 for (int c = 0; c < children; c++)
1094 vd->vdev_child[c]->vdev_open_error =
1095 vdev_open(vd->vdev_child[c]);
1098 tq = taskq_create("vdev_open", children, minclsyspri,
1099 children, children, TASKQ_PREPOPULATE);
1101 for (int c = 0; c < children; c++)
1102 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1109 * Prepare a virtual device for access.
1112 vdev_open(vdev_t *vd)
1114 spa_t *spa = vd->vdev_spa;
1117 uint64_t max_osize = 0;
1118 uint64_t asize, max_asize, psize;
1119 uint64_t ashift = 0;
1121 ASSERT(vd->vdev_open_thread == curthread ||
1122 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1123 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1124 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1125 vd->vdev_state == VDEV_STATE_OFFLINE);
1127 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1128 vd->vdev_cant_read = B_FALSE;
1129 vd->vdev_cant_write = B_FALSE;
1130 vd->vdev_min_asize = vdev_get_min_asize(vd);
1133 * If this vdev is not removed, check its fault status. If it's
1134 * faulted, bail out of the open.
1136 if (!vd->vdev_removed && vd->vdev_faulted) {
1137 ASSERT(vd->vdev_children == 0);
1138 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1139 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1140 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1141 vd->vdev_label_aux);
1143 } else if (vd->vdev_offline) {
1144 ASSERT(vd->vdev_children == 0);
1145 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1149 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1152 * Reset the vdev_reopening flag so that we actually close
1153 * the vdev on error.
1155 vd->vdev_reopening = B_FALSE;
1156 if (zio_injection_enabled && error == 0)
1157 error = zio_handle_device_injection(vd, NULL, ENXIO);
1160 if (vd->vdev_removed &&
1161 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1162 vd->vdev_removed = B_FALSE;
1164 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1165 vd->vdev_stat.vs_aux);
1169 vd->vdev_removed = B_FALSE;
1172 * Recheck the faulted flag now that we have confirmed that
1173 * the vdev is accessible. If we're faulted, bail.
1175 if (vd->vdev_faulted) {
1176 ASSERT(vd->vdev_children == 0);
1177 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1178 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1179 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1180 vd->vdev_label_aux);
1184 if (vd->vdev_degraded) {
1185 ASSERT(vd->vdev_children == 0);
1186 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1187 VDEV_AUX_ERR_EXCEEDED);
1189 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1193 * For hole or missing vdevs we just return success.
1195 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1198 for (int c = 0; c < vd->vdev_children; c++) {
1199 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1200 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1206 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1207 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1209 if (vd->vdev_children == 0) {
1210 if (osize < SPA_MINDEVSIZE) {
1211 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1212 VDEV_AUX_TOO_SMALL);
1216 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1217 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1218 VDEV_LABEL_END_SIZE);
1220 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1221 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1222 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1223 VDEV_AUX_TOO_SMALL);
1228 max_asize = max_osize;
1231 vd->vdev_psize = psize;
1234 * Make sure the allocatable size hasn't shrunk.
1236 if (asize < vd->vdev_min_asize) {
1237 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1238 VDEV_AUX_BAD_LABEL);
1242 if (vd->vdev_asize == 0) {
1244 * This is the first-ever open, so use the computed values.
1245 * For testing purposes, a higher ashift can be requested.
1247 vd->vdev_asize = asize;
1248 vd->vdev_max_asize = max_asize;
1249 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1252 * Make sure the alignment requirement hasn't increased.
1254 if (ashift > vd->vdev_top->vdev_ashift) {
1255 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1256 VDEV_AUX_BAD_LABEL);
1259 vd->vdev_max_asize = max_asize;
1263 * If all children are healthy and the asize has increased,
1264 * then we've experienced dynamic LUN growth. If automatic
1265 * expansion is enabled then use the additional space.
1267 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1268 (vd->vdev_expanding || spa->spa_autoexpand))
1269 vd->vdev_asize = asize;
1271 vdev_set_min_asize(vd);
1274 * Ensure we can issue some IO before declaring the
1275 * vdev open for business.
1277 if (vd->vdev_ops->vdev_op_leaf &&
1278 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1279 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1280 VDEV_AUX_ERR_EXCEEDED);
1285 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1286 * resilver. But don't do this if we are doing a reopen for a scrub,
1287 * since this would just restart the scrub we are already doing.
1289 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1290 vdev_resilver_needed(vd, NULL, NULL))
1291 spa_async_request(spa, SPA_ASYNC_RESILVER);
1297 * Called once the vdevs are all opened, this routine validates the label
1298 * contents. This needs to be done before vdev_load() so that we don't
1299 * inadvertently do repair I/Os to the wrong device.
1301 * If 'strict' is false ignore the spa guid check. This is necessary because
1302 * if the machine crashed during a re-guid the new guid might have been written
1303 * to all of the vdev labels, but not the cached config. The strict check
1304 * will be performed when the pool is opened again using the mos config.
1306 * This function will only return failure if one of the vdevs indicates that it
1307 * has since been destroyed or exported. This is only possible if
1308 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1309 * will be updated but the function will return 0.
1312 vdev_validate(vdev_t *vd, boolean_t strict)
1314 spa_t *spa = vd->vdev_spa;
1316 uint64_t guid = 0, top_guid;
1319 for (int c = 0; c < vd->vdev_children; c++)
1320 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1324 * If the device has already failed, or was marked offline, don't do
1325 * any further validation. Otherwise, label I/O will fail and we will
1326 * overwrite the previous state.
1328 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1329 uint64_t aux_guid = 0;
1331 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1332 spa_last_synced_txg(spa) : -1ULL;
1334 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1335 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1336 VDEV_AUX_BAD_LABEL);
1341 * Determine if this vdev has been split off into another
1342 * pool. If so, then refuse to open it.
1344 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1345 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1346 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1347 VDEV_AUX_SPLIT_POOL);
1352 if (strict && (nvlist_lookup_uint64(label,
1353 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1354 guid != spa_guid(spa))) {
1355 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1356 VDEV_AUX_CORRUPT_DATA);
1361 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1362 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1367 * If this vdev just became a top-level vdev because its
1368 * sibling was detached, it will have adopted the parent's
1369 * vdev guid -- but the label may or may not be on disk yet.
1370 * Fortunately, either version of the label will have the
1371 * same top guid, so if we're a top-level vdev, we can
1372 * safely compare to that instead.
1374 * If we split this vdev off instead, then we also check the
1375 * original pool's guid. We don't want to consider the vdev
1376 * corrupt if it is partway through a split operation.
1378 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1380 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1382 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1383 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1384 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1385 VDEV_AUX_CORRUPT_DATA);
1390 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1392 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1393 VDEV_AUX_CORRUPT_DATA);
1401 * If this is a verbatim import, no need to check the
1402 * state of the pool.
1404 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1405 spa_load_state(spa) == SPA_LOAD_OPEN &&
1406 state != POOL_STATE_ACTIVE)
1410 * If we were able to open and validate a vdev that was
1411 * previously marked permanently unavailable, clear that state
1414 if (vd->vdev_not_present)
1415 vd->vdev_not_present = 0;
1422 * Close a virtual device.
1425 vdev_close(vdev_t *vd)
1427 spa_t *spa = vd->vdev_spa;
1428 vdev_t *pvd = vd->vdev_parent;
1430 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1433 * If our parent is reopening, then we are as well, unless we are
1436 if (pvd != NULL && pvd->vdev_reopening)
1437 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1439 vd->vdev_ops->vdev_op_close(vd);
1441 vdev_cache_purge(vd);
1444 * We record the previous state before we close it, so that if we are
1445 * doing a reopen(), we don't generate FMA ereports if we notice that
1446 * it's still faulted.
1448 vd->vdev_prevstate = vd->vdev_state;
1450 if (vd->vdev_offline)
1451 vd->vdev_state = VDEV_STATE_OFFLINE;
1453 vd->vdev_state = VDEV_STATE_CLOSED;
1454 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1458 vdev_hold(vdev_t *vd)
1460 spa_t *spa = vd->vdev_spa;
1462 ASSERT(spa_is_root(spa));
1463 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1466 for (int c = 0; c < vd->vdev_children; c++)
1467 vdev_hold(vd->vdev_child[c]);
1469 if (vd->vdev_ops->vdev_op_leaf)
1470 vd->vdev_ops->vdev_op_hold(vd);
1474 vdev_rele(vdev_t *vd)
1476 spa_t *spa = vd->vdev_spa;
1478 ASSERT(spa_is_root(spa));
1479 for (int c = 0; c < vd->vdev_children; c++)
1480 vdev_rele(vd->vdev_child[c]);
1482 if (vd->vdev_ops->vdev_op_leaf)
1483 vd->vdev_ops->vdev_op_rele(vd);
1487 * Reopen all interior vdevs and any unopened leaves. We don't actually
1488 * reopen leaf vdevs which had previously been opened as they might deadlock
1489 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1490 * If the leaf has never been opened then open it, as usual.
1493 vdev_reopen(vdev_t *vd)
1495 spa_t *spa = vd->vdev_spa;
1497 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1499 /* set the reopening flag unless we're taking the vdev offline */
1500 vd->vdev_reopening = !vd->vdev_offline;
1502 (void) vdev_open(vd);
1505 * Call vdev_validate() here to make sure we have the same device.
1506 * Otherwise, a device with an invalid label could be successfully
1507 * opened in response to vdev_reopen().
1510 (void) vdev_validate_aux(vd);
1511 if (vdev_readable(vd) && vdev_writeable(vd) &&
1512 vd->vdev_aux == &spa->spa_l2cache &&
1513 !l2arc_vdev_present(vd))
1514 l2arc_add_vdev(spa, vd);
1516 (void) vdev_validate(vd, B_TRUE);
1520 * Reassess parent vdev's health.
1522 vdev_propagate_state(vd);
1526 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1531 * Normally, partial opens (e.g. of a mirror) are allowed.
1532 * For a create, however, we want to fail the request if
1533 * there are any components we can't open.
1535 error = vdev_open(vd);
1537 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1539 return (error ? error : ENXIO);
1543 * Recursively initialize all labels.
1545 if ((error = vdev_label_init(vd, txg, isreplacing ?
1546 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1555 vdev_metaslab_set_size(vdev_t *vd)
1558 * Aim for roughly 200 metaslabs per vdev.
1560 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1561 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1565 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1567 ASSERT(vd == vd->vdev_top);
1568 ASSERT(!vd->vdev_ishole);
1569 ASSERT(ISP2(flags));
1570 ASSERT(spa_writeable(vd->vdev_spa));
1572 if (flags & VDD_METASLAB)
1573 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1575 if (flags & VDD_DTL)
1576 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1578 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1584 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1585 * the vdev has less than perfect replication. There are four kinds of DTL:
1587 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1589 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1591 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1592 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1593 * txgs that was scrubbed.
1595 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1596 * persistent errors or just some device being offline.
1597 * Unlike the other three, the DTL_OUTAGE map is not generally
1598 * maintained; it's only computed when needed, typically to
1599 * determine whether a device can be detached.
1601 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1602 * either has the data or it doesn't.
1604 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1605 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1606 * if any child is less than fully replicated, then so is its parent.
1607 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1608 * comprising only those txgs which appear in 'maxfaults' or more children;
1609 * those are the txgs we don't have enough replication to read. For example,
1610 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1611 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1612 * two child DTL_MISSING maps.
1614 * It should be clear from the above that to compute the DTLs and outage maps
1615 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1616 * Therefore, that is all we keep on disk. When loading the pool, or after
1617 * a configuration change, we generate all other DTLs from first principles.
1620 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1622 space_map_t *sm = &vd->vdev_dtl[t];
1624 ASSERT(t < DTL_TYPES);
1625 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1626 ASSERT(spa_writeable(vd->vdev_spa));
1628 mutex_enter(sm->sm_lock);
1629 if (!space_map_contains(sm, txg, size))
1630 space_map_add(sm, txg, size);
1631 mutex_exit(sm->sm_lock);
1635 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1637 space_map_t *sm = &vd->vdev_dtl[t];
1638 boolean_t dirty = B_FALSE;
1640 ASSERT(t < DTL_TYPES);
1641 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1643 mutex_enter(sm->sm_lock);
1644 if (sm->sm_space != 0)
1645 dirty = space_map_contains(sm, txg, size);
1646 mutex_exit(sm->sm_lock);
1652 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1654 space_map_t *sm = &vd->vdev_dtl[t];
1657 mutex_enter(sm->sm_lock);
1658 empty = (sm->sm_space == 0);
1659 mutex_exit(sm->sm_lock);
1665 * Reassess DTLs after a config change or scrub completion.
1668 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1670 spa_t *spa = vd->vdev_spa;
1674 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1676 for (int c = 0; c < vd->vdev_children; c++)
1677 vdev_dtl_reassess(vd->vdev_child[c], txg,
1678 scrub_txg, scrub_done);
1680 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1683 if (vd->vdev_ops->vdev_op_leaf) {
1684 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1686 mutex_enter(&vd->vdev_dtl_lock);
1687 if (scrub_txg != 0 &&
1688 (spa->spa_scrub_started ||
1689 (scn && scn->scn_phys.scn_errors == 0))) {
1691 * We completed a scrub up to scrub_txg. If we
1692 * did it without rebooting, then the scrub dtl
1693 * will be valid, so excise the old region and
1694 * fold in the scrub dtl. Otherwise, leave the
1695 * dtl as-is if there was an error.
1697 * There's little trick here: to excise the beginning
1698 * of the DTL_MISSING map, we put it into a reference
1699 * tree and then add a segment with refcnt -1 that
1700 * covers the range [0, scrub_txg). This means
1701 * that each txg in that range has refcnt -1 or 0.
1702 * We then add DTL_SCRUB with a refcnt of 2, so that
1703 * entries in the range [0, scrub_txg) will have a
1704 * positive refcnt -- either 1 or 2. We then convert
1705 * the reference tree into the new DTL_MISSING map.
1707 space_map_ref_create(&reftree);
1708 space_map_ref_add_map(&reftree,
1709 &vd->vdev_dtl[DTL_MISSING], 1);
1710 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1711 space_map_ref_add_map(&reftree,
1712 &vd->vdev_dtl[DTL_SCRUB], 2);
1713 space_map_ref_generate_map(&reftree,
1714 &vd->vdev_dtl[DTL_MISSING], 1);
1715 space_map_ref_destroy(&reftree);
1717 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1718 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1719 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1721 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1722 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1723 if (!vdev_readable(vd))
1724 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1726 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1727 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1728 mutex_exit(&vd->vdev_dtl_lock);
1731 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1735 mutex_enter(&vd->vdev_dtl_lock);
1736 for (int t = 0; t < DTL_TYPES; t++) {
1737 /* account for child's outage in parent's missing map */
1738 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1740 continue; /* leaf vdevs only */
1741 if (t == DTL_PARTIAL)
1742 minref = 1; /* i.e. non-zero */
1743 else if (vd->vdev_nparity != 0)
1744 minref = vd->vdev_nparity + 1; /* RAID-Z */
1746 minref = vd->vdev_children; /* any kind of mirror */
1747 space_map_ref_create(&reftree);
1748 for (int c = 0; c < vd->vdev_children; c++) {
1749 vdev_t *cvd = vd->vdev_child[c];
1750 mutex_enter(&cvd->vdev_dtl_lock);
1751 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1752 mutex_exit(&cvd->vdev_dtl_lock);
1754 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1755 space_map_ref_destroy(&reftree);
1757 mutex_exit(&vd->vdev_dtl_lock);
1761 vdev_dtl_load(vdev_t *vd)
1763 spa_t *spa = vd->vdev_spa;
1764 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1765 objset_t *mos = spa->spa_meta_objset;
1769 ASSERT(vd->vdev_children == 0);
1771 if (smo->smo_object == 0)
1774 ASSERT(!vd->vdev_ishole);
1776 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1779 ASSERT3U(db->db_size, >=, sizeof (*smo));
1780 bcopy(db->db_data, smo, sizeof (*smo));
1781 dmu_buf_rele(db, FTAG);
1783 mutex_enter(&vd->vdev_dtl_lock);
1784 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1785 NULL, SM_ALLOC, smo, mos);
1786 mutex_exit(&vd->vdev_dtl_lock);
1792 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1794 spa_t *spa = vd->vdev_spa;
1795 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1796 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1797 objset_t *mos = spa->spa_meta_objset;
1803 ASSERT(!vd->vdev_ishole);
1805 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1807 if (vd->vdev_detached) {
1808 if (smo->smo_object != 0) {
1809 int err = dmu_object_free(mos, smo->smo_object, tx);
1811 smo->smo_object = 0;
1817 if (smo->smo_object == 0) {
1818 ASSERT(smo->smo_objsize == 0);
1819 ASSERT(smo->smo_alloc == 0);
1820 smo->smo_object = dmu_object_alloc(mos,
1821 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1822 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1823 ASSERT(smo->smo_object != 0);
1824 vdev_config_dirty(vd->vdev_top);
1827 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1829 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1832 mutex_enter(&smlock);
1834 mutex_enter(&vd->vdev_dtl_lock);
1835 space_map_walk(sm, space_map_add, &smsync);
1836 mutex_exit(&vd->vdev_dtl_lock);
1838 space_map_truncate(smo, mos, tx);
1839 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1841 space_map_destroy(&smsync);
1843 mutex_exit(&smlock);
1844 mutex_destroy(&smlock);
1846 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1847 dmu_buf_will_dirty(db, tx);
1848 ASSERT3U(db->db_size, >=, sizeof (*smo));
1849 bcopy(smo, db->db_data, sizeof (*smo));
1850 dmu_buf_rele(db, FTAG);
1856 * Determine whether the specified vdev can be offlined/detached/removed
1857 * without losing data.
1860 vdev_dtl_required(vdev_t *vd)
1862 spa_t *spa = vd->vdev_spa;
1863 vdev_t *tvd = vd->vdev_top;
1864 uint8_t cant_read = vd->vdev_cant_read;
1867 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1869 if (vd == spa->spa_root_vdev || vd == tvd)
1873 * Temporarily mark the device as unreadable, and then determine
1874 * whether this results in any DTL outages in the top-level vdev.
1875 * If not, we can safely offline/detach/remove the device.
1877 vd->vdev_cant_read = B_TRUE;
1878 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1879 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1880 vd->vdev_cant_read = cant_read;
1881 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1883 if (!required && zio_injection_enabled)
1884 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1890 * Determine if resilver is needed, and if so the txg range.
1893 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1895 boolean_t needed = B_FALSE;
1896 uint64_t thismin = UINT64_MAX;
1897 uint64_t thismax = 0;
1899 if (vd->vdev_children == 0) {
1900 mutex_enter(&vd->vdev_dtl_lock);
1901 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1902 vdev_writeable(vd)) {
1905 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1906 thismin = ss->ss_start - 1;
1907 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1908 thismax = ss->ss_end;
1911 mutex_exit(&vd->vdev_dtl_lock);
1913 for (int c = 0; c < vd->vdev_children; c++) {
1914 vdev_t *cvd = vd->vdev_child[c];
1915 uint64_t cmin, cmax;
1917 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1918 thismin = MIN(thismin, cmin);
1919 thismax = MAX(thismax, cmax);
1925 if (needed && minp) {
1933 vdev_load(vdev_t *vd)
1936 * Recursively load all children.
1938 for (int c = 0; c < vd->vdev_children; c++)
1939 vdev_load(vd->vdev_child[c]);
1942 * If this is a top-level vdev, initialize its metaslabs.
1944 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1945 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1946 vdev_metaslab_init(vd, 0) != 0))
1947 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1948 VDEV_AUX_CORRUPT_DATA);
1951 * If this is a leaf vdev, load its DTL.
1953 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1954 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1955 VDEV_AUX_CORRUPT_DATA);
1959 * The special vdev case is used for hot spares and l2cache devices. Its
1960 * sole purpose it to set the vdev state for the associated vdev. To do this,
1961 * we make sure that we can open the underlying device, then try to read the
1962 * label, and make sure that the label is sane and that it hasn't been
1963 * repurposed to another pool.
1966 vdev_validate_aux(vdev_t *vd)
1969 uint64_t guid, version;
1972 if (!vdev_readable(vd))
1975 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
1976 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1977 VDEV_AUX_CORRUPT_DATA);
1981 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1982 !SPA_VERSION_IS_SUPPORTED(version) ||
1983 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1984 guid != vd->vdev_guid ||
1985 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1986 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1987 VDEV_AUX_CORRUPT_DATA);
1993 * We don't actually check the pool state here. If it's in fact in
1994 * use by another pool, we update this fact on the fly when requested.
2001 vdev_remove(vdev_t *vd, uint64_t txg)
2003 spa_t *spa = vd->vdev_spa;
2004 objset_t *mos = spa->spa_meta_objset;
2007 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2009 if (vd->vdev_dtl_smo.smo_object) {
2010 ASSERT0(vd->vdev_dtl_smo.smo_alloc);
2011 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2012 vd->vdev_dtl_smo.smo_object = 0;
2015 if (vd->vdev_ms != NULL) {
2016 for (int m = 0; m < vd->vdev_ms_count; m++) {
2017 metaslab_t *msp = vd->vdev_ms[m];
2019 if (msp == NULL || msp->ms_smo.smo_object == 0)
2022 ASSERT0(msp->ms_smo.smo_alloc);
2023 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2024 msp->ms_smo.smo_object = 0;
2028 if (vd->vdev_ms_array) {
2029 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2030 vd->vdev_ms_array = 0;
2031 vd->vdev_ms_shift = 0;
2037 vdev_sync_done(vdev_t *vd, uint64_t txg)
2040 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2042 ASSERT(!vd->vdev_ishole);
2044 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2045 metaslab_sync_done(msp, txg);
2048 metaslab_sync_reassess(vd->vdev_mg);
2052 vdev_sync(vdev_t *vd, uint64_t txg)
2054 spa_t *spa = vd->vdev_spa;
2059 ASSERT(!vd->vdev_ishole);
2061 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2062 ASSERT(vd == vd->vdev_top);
2063 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2064 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2065 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2066 ASSERT(vd->vdev_ms_array != 0);
2067 vdev_config_dirty(vd);
2072 * Remove the metadata associated with this vdev once it's empty.
2074 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2075 vdev_remove(vd, txg);
2077 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2078 metaslab_sync(msp, txg);
2079 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2082 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2083 vdev_dtl_sync(lvd, txg);
2085 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2089 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2091 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2095 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2096 * not be opened, and no I/O is attempted.
2099 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2103 spa_vdev_state_enter(spa, SCL_NONE);
2105 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2106 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2108 if (!vd->vdev_ops->vdev_op_leaf)
2109 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2114 * We don't directly use the aux state here, but if we do a
2115 * vdev_reopen(), we need this value to be present to remember why we
2118 vd->vdev_label_aux = aux;
2121 * Faulted state takes precedence over degraded.
2123 vd->vdev_delayed_close = B_FALSE;
2124 vd->vdev_faulted = 1ULL;
2125 vd->vdev_degraded = 0ULL;
2126 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2129 * If this device has the only valid copy of the data, then
2130 * back off and simply mark the vdev as degraded instead.
2132 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2133 vd->vdev_degraded = 1ULL;
2134 vd->vdev_faulted = 0ULL;
2137 * If we reopen the device and it's not dead, only then do we
2142 if (vdev_readable(vd))
2143 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2146 return (spa_vdev_state_exit(spa, vd, 0));
2150 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2151 * user that something is wrong. The vdev continues to operate as normal as far
2152 * as I/O is concerned.
2155 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2159 spa_vdev_state_enter(spa, SCL_NONE);
2161 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2162 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2164 if (!vd->vdev_ops->vdev_op_leaf)
2165 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2168 * If the vdev is already faulted, then don't do anything.
2170 if (vd->vdev_faulted || vd->vdev_degraded)
2171 return (spa_vdev_state_exit(spa, NULL, 0));
2173 vd->vdev_degraded = 1ULL;
2174 if (!vdev_is_dead(vd))
2175 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2178 return (spa_vdev_state_exit(spa, vd, 0));
2182 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2183 * any attached spare device should be detached when the device finishes
2184 * resilvering. Second, the online should be treated like a 'test' online case,
2185 * so no FMA events are generated if the device fails to open.
2188 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2190 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2192 spa_vdev_state_enter(spa, SCL_NONE);
2194 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2195 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2197 if (!vd->vdev_ops->vdev_op_leaf)
2198 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2201 vd->vdev_offline = B_FALSE;
2202 vd->vdev_tmpoffline = B_FALSE;
2203 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2204 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2206 /* XXX - L2ARC 1.0 does not support expansion */
2207 if (!vd->vdev_aux) {
2208 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2209 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2213 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2215 if (!vd->vdev_aux) {
2216 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2217 pvd->vdev_expanding = B_FALSE;
2221 *newstate = vd->vdev_state;
2222 if ((flags & ZFS_ONLINE_UNSPARE) &&
2223 !vdev_is_dead(vd) && vd->vdev_parent &&
2224 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2225 vd->vdev_parent->vdev_child[0] == vd)
2226 vd->vdev_unspare = B_TRUE;
2228 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2230 /* XXX - L2ARC 1.0 does not support expansion */
2232 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2233 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2235 return (spa_vdev_state_exit(spa, vd, 0));
2239 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2243 uint64_t generation;
2244 metaslab_group_t *mg;
2247 spa_vdev_state_enter(spa, SCL_ALLOC);
2249 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2250 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2252 if (!vd->vdev_ops->vdev_op_leaf)
2253 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2257 generation = spa->spa_config_generation + 1;
2260 * If the device isn't already offline, try to offline it.
2262 if (!vd->vdev_offline) {
2264 * If this device has the only valid copy of some data,
2265 * don't allow it to be offlined. Log devices are always
2268 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2269 vdev_dtl_required(vd))
2270 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2273 * If the top-level is a slog and it has had allocations
2274 * then proceed. We check that the vdev's metaslab group
2275 * is not NULL since it's possible that we may have just
2276 * added this vdev but not yet initialized its metaslabs.
2278 if (tvd->vdev_islog && mg != NULL) {
2280 * Prevent any future allocations.
2282 metaslab_group_passivate(mg);
2283 (void) spa_vdev_state_exit(spa, vd, 0);
2285 error = spa_offline_log(spa);
2287 spa_vdev_state_enter(spa, SCL_ALLOC);
2290 * Check to see if the config has changed.
2292 if (error || generation != spa->spa_config_generation) {
2293 metaslab_group_activate(mg);
2295 return (spa_vdev_state_exit(spa,
2297 (void) spa_vdev_state_exit(spa, vd, 0);
2300 ASSERT0(tvd->vdev_stat.vs_alloc);
2304 * Offline this device and reopen its top-level vdev.
2305 * If the top-level vdev is a log device then just offline
2306 * it. Otherwise, if this action results in the top-level
2307 * vdev becoming unusable, undo it and fail the request.
2309 vd->vdev_offline = B_TRUE;
2312 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2313 vdev_is_dead(tvd)) {
2314 vd->vdev_offline = B_FALSE;
2316 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2320 * Add the device back into the metaslab rotor so that
2321 * once we online the device it's open for business.
2323 if (tvd->vdev_islog && mg != NULL)
2324 metaslab_group_activate(mg);
2327 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2329 return (spa_vdev_state_exit(spa, vd, 0));
2333 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2337 mutex_enter(&spa->spa_vdev_top_lock);
2338 error = vdev_offline_locked(spa, guid, flags);
2339 mutex_exit(&spa->spa_vdev_top_lock);
2345 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2346 * vdev_offline(), we assume the spa config is locked. We also clear all
2347 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2350 vdev_clear(spa_t *spa, vdev_t *vd)
2352 vdev_t *rvd = spa->spa_root_vdev;
2354 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2359 vd->vdev_stat.vs_read_errors = 0;
2360 vd->vdev_stat.vs_write_errors = 0;
2361 vd->vdev_stat.vs_checksum_errors = 0;
2363 for (int c = 0; c < vd->vdev_children; c++)
2364 vdev_clear(spa, vd->vdev_child[c]);
2367 * If we're in the FAULTED state or have experienced failed I/O, then
2368 * clear the persistent state and attempt to reopen the device. We
2369 * also mark the vdev config dirty, so that the new faulted state is
2370 * written out to disk.
2372 if (vd->vdev_faulted || vd->vdev_degraded ||
2373 !vdev_readable(vd) || !vdev_writeable(vd)) {
2376 * When reopening in reponse to a clear event, it may be due to
2377 * a fmadm repair request. In this case, if the device is
2378 * still broken, we want to still post the ereport again.
2380 vd->vdev_forcefault = B_TRUE;
2382 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2383 vd->vdev_cant_read = B_FALSE;
2384 vd->vdev_cant_write = B_FALSE;
2386 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2388 vd->vdev_forcefault = B_FALSE;
2390 if (vd != rvd && vdev_writeable(vd->vdev_top))
2391 vdev_state_dirty(vd->vdev_top);
2393 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2394 spa_async_request(spa, SPA_ASYNC_RESILVER);
2396 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2400 * When clearing a FMA-diagnosed fault, we always want to
2401 * unspare the device, as we assume that the original spare was
2402 * done in response to the FMA fault.
2404 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2405 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2406 vd->vdev_parent->vdev_child[0] == vd)
2407 vd->vdev_unspare = B_TRUE;
2411 vdev_is_dead(vdev_t *vd)
2414 * Holes and missing devices are always considered "dead".
2415 * This simplifies the code since we don't have to check for
2416 * these types of devices in the various code paths.
2417 * Instead we rely on the fact that we skip over dead devices
2418 * before issuing I/O to them.
2420 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2421 vd->vdev_ops == &vdev_missing_ops);
2425 vdev_readable(vdev_t *vd)
2427 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2431 vdev_writeable(vdev_t *vd)
2433 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2437 vdev_allocatable(vdev_t *vd)
2439 uint64_t state = vd->vdev_state;
2442 * We currently allow allocations from vdevs which may be in the
2443 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2444 * fails to reopen then we'll catch it later when we're holding
2445 * the proper locks. Note that we have to get the vdev state
2446 * in a local variable because although it changes atomically,
2447 * we're asking two separate questions about it.
2449 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2450 !vd->vdev_cant_write && !vd->vdev_ishole);
2454 vdev_accessible(vdev_t *vd, zio_t *zio)
2456 ASSERT(zio->io_vd == vd);
2458 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2461 if (zio->io_type == ZIO_TYPE_READ)
2462 return (!vd->vdev_cant_read);
2464 if (zio->io_type == ZIO_TYPE_WRITE)
2465 return (!vd->vdev_cant_write);
2471 * Get statistics for the given vdev.
2474 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2476 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2478 mutex_enter(&vd->vdev_stat_lock);
2479 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2480 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2481 vs->vs_state = vd->vdev_state;
2482 vs->vs_rsize = vdev_get_min_asize(vd);
2483 if (vd->vdev_ops->vdev_op_leaf)
2484 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2485 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2486 mutex_exit(&vd->vdev_stat_lock);
2489 * If we're getting stats on the root vdev, aggregate the I/O counts
2490 * over all top-level vdevs (i.e. the direct children of the root).
2493 for (int c = 0; c < rvd->vdev_children; c++) {
2494 vdev_t *cvd = rvd->vdev_child[c];
2495 vdev_stat_t *cvs = &cvd->vdev_stat;
2497 mutex_enter(&vd->vdev_stat_lock);
2498 for (int t = 0; t < ZIO_TYPES; t++) {
2499 vs->vs_ops[t] += cvs->vs_ops[t];
2500 vs->vs_bytes[t] += cvs->vs_bytes[t];
2502 cvs->vs_scan_removing = cvd->vdev_removing;
2503 mutex_exit(&vd->vdev_stat_lock);
2509 vdev_clear_stats(vdev_t *vd)
2511 mutex_enter(&vd->vdev_stat_lock);
2512 vd->vdev_stat.vs_space = 0;
2513 vd->vdev_stat.vs_dspace = 0;
2514 vd->vdev_stat.vs_alloc = 0;
2515 mutex_exit(&vd->vdev_stat_lock);
2519 vdev_scan_stat_init(vdev_t *vd)
2521 vdev_stat_t *vs = &vd->vdev_stat;
2523 for (int c = 0; c < vd->vdev_children; c++)
2524 vdev_scan_stat_init(vd->vdev_child[c]);
2526 mutex_enter(&vd->vdev_stat_lock);
2527 vs->vs_scan_processed = 0;
2528 mutex_exit(&vd->vdev_stat_lock);
2532 vdev_stat_update(zio_t *zio, uint64_t psize)
2534 spa_t *spa = zio->io_spa;
2535 vdev_t *rvd = spa->spa_root_vdev;
2536 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2538 uint64_t txg = zio->io_txg;
2539 vdev_stat_t *vs = &vd->vdev_stat;
2540 zio_type_t type = zio->io_type;
2541 int flags = zio->io_flags;
2544 * If this i/o is a gang leader, it didn't do any actual work.
2546 if (zio->io_gang_tree)
2549 if (zio->io_error == 0) {
2551 * If this is a root i/o, don't count it -- we've already
2552 * counted the top-level vdevs, and vdev_get_stats() will
2553 * aggregate them when asked. This reduces contention on
2554 * the root vdev_stat_lock and implicitly handles blocks
2555 * that compress away to holes, for which there is no i/o.
2556 * (Holes never create vdev children, so all the counters
2557 * remain zero, which is what we want.)
2559 * Note: this only applies to successful i/o (io_error == 0)
2560 * because unlike i/o counts, errors are not additive.
2561 * When reading a ditto block, for example, failure of
2562 * one top-level vdev does not imply a root-level error.
2567 ASSERT(vd == zio->io_vd);
2569 if (flags & ZIO_FLAG_IO_BYPASS)
2572 mutex_enter(&vd->vdev_stat_lock);
2574 if (flags & ZIO_FLAG_IO_REPAIR) {
2575 if (flags & ZIO_FLAG_SCAN_THREAD) {
2576 dsl_scan_phys_t *scn_phys =
2577 &spa->spa_dsl_pool->dp_scan->scn_phys;
2578 uint64_t *processed = &scn_phys->scn_processed;
2581 if (vd->vdev_ops->vdev_op_leaf)
2582 atomic_add_64(processed, psize);
2583 vs->vs_scan_processed += psize;
2586 if (flags & ZIO_FLAG_SELF_HEAL)
2587 vs->vs_self_healed += psize;
2591 vs->vs_bytes[type] += psize;
2593 mutex_exit(&vd->vdev_stat_lock);
2597 if (flags & ZIO_FLAG_SPECULATIVE)
2601 * If this is an I/O error that is going to be retried, then ignore the
2602 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2603 * hard errors, when in reality they can happen for any number of
2604 * innocuous reasons (bus resets, MPxIO link failure, etc).
2606 if (zio->io_error == EIO &&
2607 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2611 * Intent logs writes won't propagate their error to the root
2612 * I/O so don't mark these types of failures as pool-level
2615 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2618 mutex_enter(&vd->vdev_stat_lock);
2619 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2620 if (zio->io_error == ECKSUM)
2621 vs->vs_checksum_errors++;
2623 vs->vs_read_errors++;
2625 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2626 vs->vs_write_errors++;
2627 mutex_exit(&vd->vdev_stat_lock);
2629 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2630 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2631 (flags & ZIO_FLAG_SCAN_THREAD) ||
2632 spa->spa_claiming)) {
2634 * This is either a normal write (not a repair), or it's
2635 * a repair induced by the scrub thread, or it's a repair
2636 * made by zil_claim() during spa_load() in the first txg.
2637 * In the normal case, we commit the DTL change in the same
2638 * txg as the block was born. In the scrub-induced repair
2639 * case, we know that scrubs run in first-pass syncing context,
2640 * so we commit the DTL change in spa_syncing_txg(spa).
2641 * In the zil_claim() case, we commit in spa_first_txg(spa).
2643 * We currently do not make DTL entries for failed spontaneous
2644 * self-healing writes triggered by normal (non-scrubbing)
2645 * reads, because we have no transactional context in which to
2646 * do so -- and it's not clear that it'd be desirable anyway.
2648 if (vd->vdev_ops->vdev_op_leaf) {
2649 uint64_t commit_txg = txg;
2650 if (flags & ZIO_FLAG_SCAN_THREAD) {
2651 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2652 ASSERT(spa_sync_pass(spa) == 1);
2653 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2654 commit_txg = spa_syncing_txg(spa);
2655 } else if (spa->spa_claiming) {
2656 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2657 commit_txg = spa_first_txg(spa);
2659 ASSERT(commit_txg >= spa_syncing_txg(spa));
2660 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2662 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2663 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2664 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2667 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2672 * Update the in-core space usage stats for this vdev, its metaslab class,
2673 * and the root vdev.
2676 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2677 int64_t space_delta)
2679 int64_t dspace_delta = space_delta;
2680 spa_t *spa = vd->vdev_spa;
2681 vdev_t *rvd = spa->spa_root_vdev;
2682 metaslab_group_t *mg = vd->vdev_mg;
2683 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2685 ASSERT(vd == vd->vdev_top);
2688 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2689 * factor. We must calculate this here and not at the root vdev
2690 * because the root vdev's psize-to-asize is simply the max of its
2691 * childrens', thus not accurate enough for us.
2693 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2694 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2695 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2696 vd->vdev_deflate_ratio;
2698 mutex_enter(&vd->vdev_stat_lock);
2699 vd->vdev_stat.vs_alloc += alloc_delta;
2700 vd->vdev_stat.vs_space += space_delta;
2701 vd->vdev_stat.vs_dspace += dspace_delta;
2702 mutex_exit(&vd->vdev_stat_lock);
2704 if (mc == spa_normal_class(spa)) {
2705 mutex_enter(&rvd->vdev_stat_lock);
2706 rvd->vdev_stat.vs_alloc += alloc_delta;
2707 rvd->vdev_stat.vs_space += space_delta;
2708 rvd->vdev_stat.vs_dspace += dspace_delta;
2709 mutex_exit(&rvd->vdev_stat_lock);
2713 ASSERT(rvd == vd->vdev_parent);
2714 ASSERT(vd->vdev_ms_count != 0);
2716 metaslab_class_space_update(mc,
2717 alloc_delta, defer_delta, space_delta, dspace_delta);
2722 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2723 * so that it will be written out next time the vdev configuration is synced.
2724 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2727 vdev_config_dirty(vdev_t *vd)
2729 spa_t *spa = vd->vdev_spa;
2730 vdev_t *rvd = spa->spa_root_vdev;
2733 ASSERT(spa_writeable(spa));
2736 * If this is an aux vdev (as with l2cache and spare devices), then we
2737 * update the vdev config manually and set the sync flag.
2739 if (vd->vdev_aux != NULL) {
2740 spa_aux_vdev_t *sav = vd->vdev_aux;
2744 for (c = 0; c < sav->sav_count; c++) {
2745 if (sav->sav_vdevs[c] == vd)
2749 if (c == sav->sav_count) {
2751 * We're being removed. There's nothing more to do.
2753 ASSERT(sav->sav_sync == B_TRUE);
2757 sav->sav_sync = B_TRUE;
2759 if (nvlist_lookup_nvlist_array(sav->sav_config,
2760 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2761 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2762 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2768 * Setting the nvlist in the middle if the array is a little
2769 * sketchy, but it will work.
2771 nvlist_free(aux[c]);
2772 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2778 * The dirty list is protected by the SCL_CONFIG lock. The caller
2779 * must either hold SCL_CONFIG as writer, or must be the sync thread
2780 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2781 * so this is sufficient to ensure mutual exclusion.
2783 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2784 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2785 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2788 for (c = 0; c < rvd->vdev_children; c++)
2789 vdev_config_dirty(rvd->vdev_child[c]);
2791 ASSERT(vd == vd->vdev_top);
2793 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2795 list_insert_head(&spa->spa_config_dirty_list, vd);
2800 vdev_config_clean(vdev_t *vd)
2802 spa_t *spa = vd->vdev_spa;
2804 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2805 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2806 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2808 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2809 list_remove(&spa->spa_config_dirty_list, vd);
2813 * Mark a top-level vdev's state as dirty, so that the next pass of
2814 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2815 * the state changes from larger config changes because they require
2816 * much less locking, and are often needed for administrative actions.
2819 vdev_state_dirty(vdev_t *vd)
2821 spa_t *spa = vd->vdev_spa;
2823 ASSERT(spa_writeable(spa));
2824 ASSERT(vd == vd->vdev_top);
2827 * The state list is protected by the SCL_STATE lock. The caller
2828 * must either hold SCL_STATE as writer, or must be the sync thread
2829 * (which holds SCL_STATE as reader). There's only one sync thread,
2830 * so this is sufficient to ensure mutual exclusion.
2832 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2833 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2834 spa_config_held(spa, SCL_STATE, RW_READER)));
2836 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2837 list_insert_head(&spa->spa_state_dirty_list, vd);
2841 vdev_state_clean(vdev_t *vd)
2843 spa_t *spa = vd->vdev_spa;
2845 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2846 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2847 spa_config_held(spa, SCL_STATE, RW_READER)));
2849 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2850 list_remove(&spa->spa_state_dirty_list, vd);
2854 * Propagate vdev state up from children to parent.
2857 vdev_propagate_state(vdev_t *vd)
2859 spa_t *spa = vd->vdev_spa;
2860 vdev_t *rvd = spa->spa_root_vdev;
2861 int degraded = 0, faulted = 0;
2865 if (vd->vdev_children > 0) {
2866 for (int c = 0; c < vd->vdev_children; c++) {
2867 child = vd->vdev_child[c];
2870 * Don't factor holes into the decision.
2872 if (child->vdev_ishole)
2875 if (!vdev_readable(child) ||
2876 (!vdev_writeable(child) && spa_writeable(spa))) {
2878 * Root special: if there is a top-level log
2879 * device, treat the root vdev as if it were
2882 if (child->vdev_islog && vd == rvd)
2886 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2890 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2894 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2897 * Root special: if there is a top-level vdev that cannot be
2898 * opened due to corrupted metadata, then propagate the root
2899 * vdev's aux state as 'corrupt' rather than 'insufficient
2902 if (corrupted && vd == rvd &&
2903 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2904 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2905 VDEV_AUX_CORRUPT_DATA);
2908 if (vd->vdev_parent)
2909 vdev_propagate_state(vd->vdev_parent);
2913 * Set a vdev's state. If this is during an open, we don't update the parent
2914 * state, because we're in the process of opening children depth-first.
2915 * Otherwise, we propagate the change to the parent.
2917 * If this routine places a device in a faulted state, an appropriate ereport is
2921 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2923 uint64_t save_state;
2924 spa_t *spa = vd->vdev_spa;
2926 if (state == vd->vdev_state) {
2927 vd->vdev_stat.vs_aux = aux;
2931 save_state = vd->vdev_state;
2933 vd->vdev_state = state;
2934 vd->vdev_stat.vs_aux = aux;
2937 * If we are setting the vdev state to anything but an open state, then
2938 * always close the underlying device unless the device has requested
2939 * a delayed close (i.e. we're about to remove or fault the device).
2940 * Otherwise, we keep accessible but invalid devices open forever.
2941 * We don't call vdev_close() itself, because that implies some extra
2942 * checks (offline, etc) that we don't want here. This is limited to
2943 * leaf devices, because otherwise closing the device will affect other
2946 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2947 vd->vdev_ops->vdev_op_leaf)
2948 vd->vdev_ops->vdev_op_close(vd);
2951 * If we have brought this vdev back into service, we need
2952 * to notify fmd so that it can gracefully repair any outstanding
2953 * cases due to a missing device. We do this in all cases, even those
2954 * that probably don't correlate to a repaired fault. This is sure to
2955 * catch all cases, and we let the zfs-retire agent sort it out. If
2956 * this is a transient state it's OK, as the retire agent will
2957 * double-check the state of the vdev before repairing it.
2959 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2960 vd->vdev_prevstate != state)
2961 zfs_post_state_change(spa, vd);
2963 if (vd->vdev_removed &&
2964 state == VDEV_STATE_CANT_OPEN &&
2965 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2967 * If the previous state is set to VDEV_STATE_REMOVED, then this
2968 * device was previously marked removed and someone attempted to
2969 * reopen it. If this failed due to a nonexistent device, then
2970 * keep the device in the REMOVED state. We also let this be if
2971 * it is one of our special test online cases, which is only
2972 * attempting to online the device and shouldn't generate an FMA
2975 vd->vdev_state = VDEV_STATE_REMOVED;
2976 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2977 } else if (state == VDEV_STATE_REMOVED) {
2978 vd->vdev_removed = B_TRUE;
2979 } else if (state == VDEV_STATE_CANT_OPEN) {
2981 * If we fail to open a vdev during an import or recovery, we
2982 * mark it as "not available", which signifies that it was
2983 * never there to begin with. Failure to open such a device
2984 * is not considered an error.
2986 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2987 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2988 vd->vdev_ops->vdev_op_leaf)
2989 vd->vdev_not_present = 1;
2992 * Post the appropriate ereport. If the 'prevstate' field is
2993 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2994 * that this is part of a vdev_reopen(). In this case, we don't
2995 * want to post the ereport if the device was already in the
2996 * CANT_OPEN state beforehand.
2998 * If the 'checkremove' flag is set, then this is an attempt to
2999 * online the device in response to an insertion event. If we
3000 * hit this case, then we have detected an insertion event for a
3001 * faulted or offline device that wasn't in the removed state.
3002 * In this scenario, we don't post an ereport because we are
3003 * about to replace the device, or attempt an online with
3004 * vdev_forcefault, which will generate the fault for us.
3006 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3007 !vd->vdev_not_present && !vd->vdev_checkremove &&
3008 vd != spa->spa_root_vdev) {
3012 case VDEV_AUX_OPEN_FAILED:
3013 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3015 case VDEV_AUX_CORRUPT_DATA:
3016 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3018 case VDEV_AUX_NO_REPLICAS:
3019 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3021 case VDEV_AUX_BAD_GUID_SUM:
3022 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3024 case VDEV_AUX_TOO_SMALL:
3025 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3027 case VDEV_AUX_BAD_LABEL:
3028 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3031 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3034 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3037 /* Erase any notion of persistent removed state */
3038 vd->vdev_removed = B_FALSE;
3040 vd->vdev_removed = B_FALSE;
3043 if (!isopen && vd->vdev_parent)
3044 vdev_propagate_state(vd->vdev_parent);
3048 * Check the vdev configuration to ensure that it's capable of supporting
3051 * On Solaris, we do not support RAID-Z or partial configuration. In
3052 * addition, only a single top-level vdev is allowed and none of the
3053 * leaves can be wholedisks.
3055 * For FreeBSD, we can boot from any configuration. There is a
3056 * limitation that the boot filesystem must be either uncompressed or
3057 * compresses with lzjb compression but I'm not sure how to enforce
3061 vdev_is_bootable(vdev_t *vd)
3064 if (!vd->vdev_ops->vdev_op_leaf) {
3065 char *vdev_type = vd->vdev_ops->vdev_op_type;
3067 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3068 vd->vdev_children > 1) {
3070 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3071 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3074 } else if (vd->vdev_wholedisk == 1) {
3078 for (int c = 0; c < vd->vdev_children; c++) {
3079 if (!vdev_is_bootable(vd->vdev_child[c]))
3087 * Load the state from the original vdev tree (ovd) which
3088 * we've retrieved from the MOS config object. If the original
3089 * vdev was offline or faulted then we transfer that state to the
3090 * device in the current vdev tree (nvd).
3093 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3095 spa_t *spa = nvd->vdev_spa;
3097 ASSERT(nvd->vdev_top->vdev_islog);
3098 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3099 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3101 for (int c = 0; c < nvd->vdev_children; c++)
3102 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3104 if (nvd->vdev_ops->vdev_op_leaf) {
3106 * Restore the persistent vdev state
3108 nvd->vdev_offline = ovd->vdev_offline;
3109 nvd->vdev_faulted = ovd->vdev_faulted;
3110 nvd->vdev_degraded = ovd->vdev_degraded;
3111 nvd->vdev_removed = ovd->vdev_removed;
3116 * Determine if a log device has valid content. If the vdev was
3117 * removed or faulted in the MOS config then we know that
3118 * the content on the log device has already been written to the pool.
3121 vdev_log_state_valid(vdev_t *vd)
3123 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3127 for (int c = 0; c < vd->vdev_children; c++)
3128 if (vdev_log_state_valid(vd->vdev_child[c]))
3135 * Expand a vdev if possible.
3138 vdev_expand(vdev_t *vd, uint64_t txg)
3140 ASSERT(vd->vdev_top == vd);
3141 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3143 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3144 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3145 vdev_config_dirty(vd);
3153 vdev_split(vdev_t *vd)
3155 vdev_t *cvd, *pvd = vd->vdev_parent;
3157 vdev_remove_child(pvd, vd);
3158 vdev_compact_children(pvd);
3160 cvd = pvd->vdev_child[0];
3161 if (pvd->vdev_children == 1) {
3162 vdev_remove_parent(cvd);
3163 cvd->vdev_splitting = B_TRUE;
3165 vdev_propagate_state(cvd);