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 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
27 #include <sys/zfs_context.h>
28 #include <sys/fm/fs/zfs.h>
30 #include <sys/spa_impl.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/vdev_impl.h>
34 #include <sys/uberblock_impl.h>
35 #include <sys/metaslab.h>
36 #include <sys/metaslab_impl.h>
37 #include <sys/space_map.h>
40 #include <sys/fs/zfs.h>
44 SYSCTL_DECL(_vfs_zfs);
45 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
48 * Virtual device management.
51 static vdev_ops_t *vdev_ops_table[] = {
67 /* maximum scrub/resilver I/O queue per leaf vdev */
68 int zfs_scrub_limit = 10;
70 TUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit);
71 SYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0,
72 "Maximum scrub/resilver I/O queue");
75 * Given a vdev type, return the appropriate ops vector.
78 vdev_getops(const char *type)
80 vdev_ops_t *ops, **opspp;
82 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
83 if (strcmp(ops->vdev_op_type, type) == 0)
90 * Default asize function: return the MAX of psize with the asize of
91 * all children. This is what's used by anything other than RAID-Z.
94 vdev_default_asize(vdev_t *vd, uint64_t psize)
96 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
100 for (c = 0; c < vd->vdev_children; c++) {
101 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
102 asize = MAX(asize, csize);
109 * Get the replaceable or attachable device size.
110 * If the parent is a mirror or raidz, the replaceable size is the minimum
111 * psize of all its children. For the rest, just return our own psize.
122 vdev_get_rsize(vdev_t *vd)
127 pvd = vd->vdev_parent;
130 * If our parent is NULL or the root, just return our own psize.
132 if (pvd == NULL || pvd->vdev_parent == NULL)
133 return (vd->vdev_psize);
137 for (c = 0; c < pvd->vdev_children; c++) {
138 cvd = pvd->vdev_child[c];
139 rsize = MIN(rsize - 1, cvd->vdev_psize - 1) + 1;
146 vdev_lookup_top(spa_t *spa, uint64_t vdev)
148 vdev_t *rvd = spa->spa_root_vdev;
150 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
152 if (vdev < rvd->vdev_children) {
153 ASSERT(rvd->vdev_child[vdev] != NULL);
154 return (rvd->vdev_child[vdev]);
161 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
166 if (vd->vdev_guid == guid)
169 for (c = 0; c < vd->vdev_children; c++)
170 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
178 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
180 size_t oldsize, newsize;
181 uint64_t id = cvd->vdev_id;
184 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
185 ASSERT(cvd->vdev_parent == NULL);
187 cvd->vdev_parent = pvd;
192 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
194 oldsize = pvd->vdev_children * sizeof (vdev_t *);
195 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
196 newsize = pvd->vdev_children * sizeof (vdev_t *);
198 newchild = kmem_zalloc(newsize, KM_SLEEP);
199 if (pvd->vdev_child != NULL) {
200 bcopy(pvd->vdev_child, newchild, oldsize);
201 kmem_free(pvd->vdev_child, oldsize);
204 pvd->vdev_child = newchild;
205 pvd->vdev_child[id] = cvd;
207 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
208 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
211 * Walk up all ancestors to update guid sum.
213 for (; pvd != NULL; pvd = pvd->vdev_parent)
214 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
216 if (cvd->vdev_ops->vdev_op_leaf)
217 cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit;
221 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
224 uint_t id = cvd->vdev_id;
226 ASSERT(cvd->vdev_parent == pvd);
231 ASSERT(id < pvd->vdev_children);
232 ASSERT(pvd->vdev_child[id] == cvd);
234 pvd->vdev_child[id] = NULL;
235 cvd->vdev_parent = NULL;
237 for (c = 0; c < pvd->vdev_children; c++)
238 if (pvd->vdev_child[c])
241 if (c == pvd->vdev_children) {
242 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
243 pvd->vdev_child = NULL;
244 pvd->vdev_children = 0;
248 * Walk up all ancestors to update guid sum.
250 for (; pvd != NULL; pvd = pvd->vdev_parent)
251 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
253 if (cvd->vdev_ops->vdev_op_leaf)
254 cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit;
258 * Remove any holes in the child array.
261 vdev_compact_children(vdev_t *pvd)
263 vdev_t **newchild, *cvd;
264 int oldc = pvd->vdev_children;
267 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
269 for (c = newc = 0; c < oldc; c++)
270 if (pvd->vdev_child[c])
273 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
275 for (c = newc = 0; c < oldc; c++) {
276 if ((cvd = pvd->vdev_child[c]) != NULL) {
277 newchild[newc] = cvd;
278 cvd->vdev_id = newc++;
282 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
283 pvd->vdev_child = newchild;
284 pvd->vdev_children = newc;
288 * Allocate and minimally initialize a vdev_t.
291 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
295 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
297 if (spa->spa_root_vdev == NULL) {
298 ASSERT(ops == &vdev_root_ops);
299 spa->spa_root_vdev = vd;
303 if (spa->spa_root_vdev == vd) {
305 * The root vdev's guid will also be the pool guid,
306 * which must be unique among all pools.
308 while (guid == 0 || spa_guid_exists(guid, 0))
309 guid = spa_get_random(-1ULL);
312 * Any other vdev's guid must be unique within the pool.
315 spa_guid_exists(spa_guid(spa), guid))
316 guid = spa_get_random(-1ULL);
318 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
323 vd->vdev_guid = guid;
324 vd->vdev_guid_sum = guid;
326 vd->vdev_state = VDEV_STATE_CLOSED;
328 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
329 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
330 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
331 for (int t = 0; t < DTL_TYPES; t++) {
332 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
335 txg_list_create(&vd->vdev_ms_list,
336 offsetof(struct metaslab, ms_txg_node));
337 txg_list_create(&vd->vdev_dtl_list,
338 offsetof(struct vdev, vdev_dtl_node));
339 vd->vdev_stat.vs_timestamp = gethrtime();
347 * Allocate a new vdev. The 'alloctype' is used to control whether we are
348 * creating a new vdev or loading an existing one - the behavior is slightly
349 * different for each case.
352 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
357 uint64_t guid = 0, islog, nparity;
360 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
362 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
365 if ((ops = vdev_getops(type)) == NULL)
369 * If this is a load, get the vdev guid from the nvlist.
370 * Otherwise, vdev_alloc_common() will generate one for us.
372 if (alloctype == VDEV_ALLOC_LOAD) {
375 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
379 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
381 } else if (alloctype == VDEV_ALLOC_SPARE) {
382 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
384 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
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)
404 * Set the nparity property for RAID-Z vdevs.
407 if (ops == &vdev_raidz_ops) {
408 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
411 * Currently, we can only support 2 parity devices.
413 if (nparity == 0 || nparity > 2)
416 * Older versions can only support 1 parity device.
419 spa_version(spa) < SPA_VERSION_RAID6)
423 * We require the parity to be specified for SPAs that
424 * support multiple parity levels.
426 if (spa_version(spa) >= SPA_VERSION_RAID6)
429 * Otherwise, we default to 1 parity device for RAID-Z.
436 ASSERT(nparity != -1ULL);
438 vd = vdev_alloc_common(spa, id, guid, ops);
440 vd->vdev_islog = islog;
441 vd->vdev_nparity = nparity;
443 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
444 vd->vdev_path = spa_strdup(vd->vdev_path);
445 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
446 vd->vdev_devid = spa_strdup(vd->vdev_devid);
447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
448 &vd->vdev_physpath) == 0)
449 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
450 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
451 vd->vdev_fru = spa_strdup(vd->vdev_fru);
454 * Set the whole_disk property. If it's not specified, leave the value
457 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
458 &vd->vdev_wholedisk) != 0)
459 vd->vdev_wholedisk = -1ULL;
462 * Look for the 'not present' flag. This will only be set if the device
463 * was not present at the time of import.
465 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
466 &vd->vdev_not_present);
469 * Get the alignment requirement.
471 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
474 * If we're a top-level vdev, try to load the allocation parameters.
476 if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) {
477 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
479 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
481 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
486 * If we're a leaf vdev, try to load the DTL object and other state.
488 if (vd->vdev_ops->vdev_op_leaf &&
489 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE)) {
490 if (alloctype == VDEV_ALLOC_LOAD) {
491 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
492 &vd->vdev_dtl_smo.smo_object);
493 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
496 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
500 * When importing a pool, we want to ignore the persistent fault
501 * state, as the diagnosis made on another system may not be
502 * valid in the current context.
504 if (spa->spa_load_state == SPA_LOAD_OPEN) {
505 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
507 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
509 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
515 * Add ourselves to the parent's list of children.
517 vdev_add_child(parent, vd);
525 vdev_free(vdev_t *vd)
528 spa_t *spa = vd->vdev_spa;
531 * vdev_free() implies closing the vdev first. This is simpler than
532 * trying to ensure complicated semantics for all callers.
536 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
541 for (c = 0; c < vd->vdev_children; c++)
542 vdev_free(vd->vdev_child[c]);
544 ASSERT(vd->vdev_child == NULL);
545 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
548 * Discard allocation state.
550 if (vd == vd->vdev_top)
551 vdev_metaslab_fini(vd);
553 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
554 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
555 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
558 * Remove this vdev from its parent's child list.
560 vdev_remove_child(vd->vdev_parent, vd);
562 ASSERT(vd->vdev_parent == NULL);
565 * Clean up vdev structure.
571 spa_strfree(vd->vdev_path);
573 spa_strfree(vd->vdev_devid);
574 if (vd->vdev_physpath)
575 spa_strfree(vd->vdev_physpath);
577 spa_strfree(vd->vdev_fru);
579 if (vd->vdev_isspare)
580 spa_spare_remove(vd);
581 if (vd->vdev_isl2cache)
582 spa_l2cache_remove(vd);
584 txg_list_destroy(&vd->vdev_ms_list);
585 txg_list_destroy(&vd->vdev_dtl_list);
587 mutex_enter(&vd->vdev_dtl_lock);
588 for (int t = 0; t < DTL_TYPES; t++) {
589 space_map_unload(&vd->vdev_dtl[t]);
590 space_map_destroy(&vd->vdev_dtl[t]);
592 mutex_exit(&vd->vdev_dtl_lock);
594 mutex_destroy(&vd->vdev_dtl_lock);
595 mutex_destroy(&vd->vdev_stat_lock);
596 mutex_destroy(&vd->vdev_probe_lock);
598 if (vd == spa->spa_root_vdev)
599 spa->spa_root_vdev = NULL;
601 kmem_free(vd, sizeof (vdev_t));
605 * Transfer top-level vdev state from svd to tvd.
608 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
610 spa_t *spa = svd->vdev_spa;
615 ASSERT(tvd == tvd->vdev_top);
617 tvd->vdev_ms_array = svd->vdev_ms_array;
618 tvd->vdev_ms_shift = svd->vdev_ms_shift;
619 tvd->vdev_ms_count = svd->vdev_ms_count;
621 svd->vdev_ms_array = 0;
622 svd->vdev_ms_shift = 0;
623 svd->vdev_ms_count = 0;
625 tvd->vdev_mg = svd->vdev_mg;
626 tvd->vdev_ms = svd->vdev_ms;
631 if (tvd->vdev_mg != NULL)
632 tvd->vdev_mg->mg_vd = tvd;
634 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
635 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
636 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
638 svd->vdev_stat.vs_alloc = 0;
639 svd->vdev_stat.vs_space = 0;
640 svd->vdev_stat.vs_dspace = 0;
642 for (t = 0; t < TXG_SIZE; t++) {
643 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
644 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
645 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
646 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
647 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
648 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
651 if (list_link_active(&svd->vdev_config_dirty_node)) {
652 vdev_config_clean(svd);
653 vdev_config_dirty(tvd);
656 if (list_link_active(&svd->vdev_state_dirty_node)) {
657 vdev_state_clean(svd);
658 vdev_state_dirty(tvd);
661 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
662 svd->vdev_deflate_ratio = 0;
664 tvd->vdev_islog = svd->vdev_islog;
669 vdev_top_update(vdev_t *tvd, vdev_t *vd)
678 for (c = 0; c < vd->vdev_children; c++)
679 vdev_top_update(tvd, vd->vdev_child[c]);
683 * Add a mirror/replacing vdev above an existing vdev.
686 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
688 spa_t *spa = cvd->vdev_spa;
689 vdev_t *pvd = cvd->vdev_parent;
692 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
694 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
696 mvd->vdev_asize = cvd->vdev_asize;
697 mvd->vdev_ashift = cvd->vdev_ashift;
698 mvd->vdev_state = cvd->vdev_state;
700 vdev_remove_child(pvd, cvd);
701 vdev_add_child(pvd, mvd);
702 cvd->vdev_id = mvd->vdev_children;
703 vdev_add_child(mvd, cvd);
704 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
706 if (mvd == mvd->vdev_top)
707 vdev_top_transfer(cvd, mvd);
713 * Remove a 1-way mirror/replacing vdev from the tree.
716 vdev_remove_parent(vdev_t *cvd)
718 vdev_t *mvd = cvd->vdev_parent;
719 vdev_t *pvd = mvd->vdev_parent;
721 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
723 ASSERT(mvd->vdev_children == 1);
724 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
725 mvd->vdev_ops == &vdev_replacing_ops ||
726 mvd->vdev_ops == &vdev_spare_ops);
727 cvd->vdev_ashift = mvd->vdev_ashift;
729 vdev_remove_child(mvd, cvd);
730 vdev_remove_child(pvd, mvd);
733 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
734 * Otherwise, we could have detached an offline device, and when we
735 * go to import the pool we'll think we have two top-level vdevs,
736 * instead of a different version of the same top-level vdev.
738 if (mvd->vdev_top == mvd) {
739 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
740 cvd->vdev_guid += guid_delta;
741 cvd->vdev_guid_sum += guid_delta;
743 cvd->vdev_id = mvd->vdev_id;
744 vdev_add_child(pvd, cvd);
745 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
747 if (cvd == cvd->vdev_top)
748 vdev_top_transfer(mvd, cvd);
750 ASSERT(mvd->vdev_children == 0);
755 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
757 spa_t *spa = vd->vdev_spa;
758 objset_t *mos = spa->spa_meta_objset;
759 metaslab_class_t *mc;
761 uint64_t oldc = vd->vdev_ms_count;
762 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
766 if (vd->vdev_ms_shift == 0) /* not being allocated from yet */
770 * Compute the raidz-deflation ratio. Note, we hard-code
771 * in 128k (1 << 17) because it is the current "typical" blocksize.
772 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
773 * or we will inconsistently account for existing bp's.
775 vd->vdev_deflate_ratio = (1 << 17) /
776 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
778 ASSERT(oldc <= newc);
781 mc = spa->spa_log_class;
783 mc = spa->spa_normal_class;
785 if (vd->vdev_mg == NULL)
786 vd->vdev_mg = metaslab_group_create(mc, vd);
788 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
791 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
792 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
796 vd->vdev_ms_count = newc;
798 for (m = oldc; m < newc; m++) {
799 space_map_obj_t smo = { 0, 0, 0 };
802 error = dmu_read(mos, vd->vdev_ms_array,
803 m * sizeof (uint64_t), sizeof (uint64_t), &object,
809 error = dmu_bonus_hold(mos, object, FTAG, &db);
812 ASSERT3U(db->db_size, >=, sizeof (smo));
813 bcopy(db->db_data, &smo, sizeof (smo));
814 ASSERT3U(smo.smo_object, ==, object);
815 dmu_buf_rele(db, FTAG);
818 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
819 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
826 vdev_metaslab_fini(vdev_t *vd)
829 uint64_t count = vd->vdev_ms_count;
831 if (vd->vdev_ms != NULL) {
832 for (m = 0; m < count; m++)
833 if (vd->vdev_ms[m] != NULL)
834 metaslab_fini(vd->vdev_ms[m]);
835 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
840 typedef struct vdev_probe_stats {
841 boolean_t vps_readable;
842 boolean_t vps_writeable;
844 } vdev_probe_stats_t;
847 vdev_probe_done(zio_t *zio)
849 spa_t *spa = zio->io_spa;
850 vdev_t *vd = zio->io_vd;
851 vdev_probe_stats_t *vps = zio->io_private;
853 ASSERT(vd->vdev_probe_zio != NULL);
855 if (zio->io_type == ZIO_TYPE_READ) {
856 if (zio->io_error == 0)
857 vps->vps_readable = 1;
858 if (zio->io_error == 0 && spa_writeable(spa)) {
859 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
860 zio->io_offset, zio->io_size, zio->io_data,
861 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
862 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
864 zio_buf_free(zio->io_data, zio->io_size);
866 } else if (zio->io_type == ZIO_TYPE_WRITE) {
867 if (zio->io_error == 0)
868 vps->vps_writeable = 1;
869 zio_buf_free(zio->io_data, zio->io_size);
870 } else if (zio->io_type == ZIO_TYPE_NULL) {
873 vd->vdev_cant_read |= !vps->vps_readable;
874 vd->vdev_cant_write |= !vps->vps_writeable;
876 if (vdev_readable(vd) &&
877 (vdev_writeable(vd) || !spa_writeable(spa))) {
880 ASSERT(zio->io_error != 0);
881 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
882 spa, vd, NULL, 0, 0);
883 zio->io_error = ENXIO;
886 mutex_enter(&vd->vdev_probe_lock);
887 ASSERT(vd->vdev_probe_zio == zio);
888 vd->vdev_probe_zio = NULL;
889 mutex_exit(&vd->vdev_probe_lock);
891 while ((pio = zio_walk_parents(zio)) != NULL)
892 if (!vdev_accessible(vd, pio))
893 pio->io_error = ENXIO;
895 kmem_free(vps, sizeof (*vps));
900 * Determine whether this device is accessible by reading and writing
901 * to several known locations: the pad regions of each vdev label
902 * but the first (which we leave alone in case it contains a VTOC).
905 vdev_probe(vdev_t *vd, zio_t *zio)
907 spa_t *spa = vd->vdev_spa;
908 vdev_probe_stats_t *vps = NULL;
911 ASSERT(vd->vdev_ops->vdev_op_leaf);
914 * Don't probe the probe.
916 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
920 * To prevent 'probe storms' when a device fails, we create
921 * just one probe i/o at a time. All zios that want to probe
922 * this vdev will become parents of the probe io.
924 mutex_enter(&vd->vdev_probe_lock);
926 if ((pio = vd->vdev_probe_zio) == NULL) {
927 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
929 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
930 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
933 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
935 * vdev_cant_read and vdev_cant_write can only
936 * transition from TRUE to FALSE when we have the
937 * SCL_ZIO lock as writer; otherwise they can only
938 * transition from FALSE to TRUE. This ensures that
939 * any zio looking at these values can assume that
940 * failures persist for the life of the I/O. That's
941 * important because when a device has intermittent
942 * connectivity problems, we want to ensure that
943 * they're ascribed to the device (ENXIO) and not
946 * Since we hold SCL_ZIO as writer here, clear both
947 * values so the probe can reevaluate from first
950 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
951 vd->vdev_cant_read = B_FALSE;
952 vd->vdev_cant_write = B_FALSE;
955 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
956 vdev_probe_done, vps,
957 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
960 vd->vdev_probe_wanted = B_TRUE;
961 spa_async_request(spa, SPA_ASYNC_PROBE);
966 zio_add_child(zio, pio);
968 mutex_exit(&vd->vdev_probe_lock);
975 for (int l = 1; l < VDEV_LABELS; l++) {
976 zio_nowait(zio_read_phys(pio, vd,
977 vdev_label_offset(vd->vdev_psize, l,
978 offsetof(vdev_label_t, vl_pad2)),
979 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
980 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
981 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
992 * Prepare a virtual device for access.
995 vdev_open(vdev_t *vd)
997 spa_t *spa = vd->vdev_spa;
1001 uint64_t asize, psize;
1002 uint64_t ashift = 0;
1004 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1006 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1007 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1008 vd->vdev_state == VDEV_STATE_OFFLINE);
1010 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1011 vd->vdev_cant_read = B_FALSE;
1012 vd->vdev_cant_write = B_FALSE;
1014 if (!vd->vdev_removed && vd->vdev_faulted) {
1015 ASSERT(vd->vdev_children == 0);
1016 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1017 VDEV_AUX_ERR_EXCEEDED);
1019 } else if (vd->vdev_offline) {
1020 ASSERT(vd->vdev_children == 0);
1021 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1025 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1027 if (zio_injection_enabled && error == 0)
1028 error = zio_handle_device_injection(vd, NULL, ENXIO);
1031 if (vd->vdev_removed &&
1032 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1033 vd->vdev_removed = B_FALSE;
1035 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1036 vd->vdev_stat.vs_aux);
1040 vd->vdev_removed = B_FALSE;
1042 if (vd->vdev_degraded) {
1043 ASSERT(vd->vdev_children == 0);
1044 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1045 VDEV_AUX_ERR_EXCEEDED);
1047 vd->vdev_state = VDEV_STATE_HEALTHY;
1050 for (c = 0; c < vd->vdev_children; c++)
1051 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1052 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1057 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1059 if (vd->vdev_children == 0) {
1060 if (osize < SPA_MINDEVSIZE) {
1061 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1062 VDEV_AUX_TOO_SMALL);
1066 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1068 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1069 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1070 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1071 VDEV_AUX_TOO_SMALL);
1078 vd->vdev_psize = psize;
1080 if (vd->vdev_asize == 0) {
1082 * This is the first-ever open, so use the computed values.
1083 * For testing purposes, a higher ashift can be requested.
1085 vd->vdev_asize = asize;
1086 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1089 * Make sure the alignment requirement hasn't increased.
1091 if (ashift > vd->vdev_top->vdev_ashift) {
1092 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1093 VDEV_AUX_BAD_LABEL);
1098 * Make sure the device hasn't shrunk.
1100 if (asize < vd->vdev_asize) {
1101 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1102 VDEV_AUX_BAD_LABEL);
1107 * If all children are healthy and the asize has increased,
1108 * then we've experienced dynamic LUN growth.
1110 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1111 asize > vd->vdev_asize) {
1112 vd->vdev_asize = asize;
1117 * Ensure we can issue some IO before declaring the
1118 * vdev open for business.
1120 if (vd->vdev_ops->vdev_op_leaf &&
1121 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1122 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1123 VDEV_AUX_IO_FAILURE);
1128 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1129 * resilver. But don't do this if we are doing a reopen for a scrub,
1130 * since this would just restart the scrub we are already doing.
1132 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1133 vdev_resilver_needed(vd, NULL, NULL))
1134 spa_async_request(spa, SPA_ASYNC_RESILVER);
1140 * Called once the vdevs are all opened, this routine validates the label
1141 * contents. This needs to be done before vdev_load() so that we don't
1142 * inadvertently do repair I/Os to the wrong device.
1144 * This function will only return failure if one of the vdevs indicates that it
1145 * has since been destroyed or exported. This is only possible if
1146 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1147 * will be updated but the function will return 0.
1150 vdev_validate(vdev_t *vd)
1152 spa_t *spa = vd->vdev_spa;
1155 uint64_t guid, top_guid;
1158 for (c = 0; c < vd->vdev_children; c++)
1159 if (vdev_validate(vd->vdev_child[c]) != 0)
1163 * If the device has already failed, or was marked offline, don't do
1164 * any further validation. Otherwise, label I/O will fail and we will
1165 * overwrite the previous state.
1167 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1169 if ((label = vdev_label_read_config(vd)) == NULL) {
1170 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1171 VDEV_AUX_BAD_LABEL);
1175 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1176 &guid) != 0 || guid != spa_guid(spa)) {
1177 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1178 VDEV_AUX_CORRUPT_DATA);
1184 * If this vdev just became a top-level vdev because its
1185 * sibling was detached, it will have adopted the parent's
1186 * vdev guid -- but the label may or may not be on disk yet.
1187 * Fortunately, either version of the label will have the
1188 * same top guid, so if we're a top-level vdev, we can
1189 * safely compare to that instead.
1191 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1193 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1195 (vd->vdev_guid != guid &&
1196 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1197 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1198 VDEV_AUX_CORRUPT_DATA);
1203 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1205 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1206 VDEV_AUX_CORRUPT_DATA);
1214 * If spa->spa_load_verbatim is true, no need to check the
1215 * state of the pool.
1217 if (!spa->spa_load_verbatim &&
1218 spa->spa_load_state == SPA_LOAD_OPEN &&
1219 state != POOL_STATE_ACTIVE)
1223 * If we were able to open and validate a vdev that was
1224 * previously marked permanently unavailable, clear that state
1227 if (vd->vdev_not_present)
1228 vd->vdev_not_present = 0;
1235 * Close a virtual device.
1238 vdev_close(vdev_t *vd)
1240 spa_t *spa = vd->vdev_spa;
1242 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1244 vd->vdev_ops->vdev_op_close(vd);
1246 vdev_cache_purge(vd);
1249 * We record the previous state before we close it, so that if we are
1250 * doing a reopen(), we don't generate FMA ereports if we notice that
1251 * it's still faulted.
1253 vd->vdev_prevstate = vd->vdev_state;
1255 if (vd->vdev_offline)
1256 vd->vdev_state = VDEV_STATE_OFFLINE;
1258 vd->vdev_state = VDEV_STATE_CLOSED;
1259 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1263 vdev_reopen(vdev_t *vd)
1265 spa_t *spa = vd->vdev_spa;
1267 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1270 (void) vdev_open(vd);
1273 * Call vdev_validate() here to make sure we have the same device.
1274 * Otherwise, a device with an invalid label could be successfully
1275 * opened in response to vdev_reopen().
1278 (void) vdev_validate_aux(vd);
1279 if (vdev_readable(vd) && vdev_writeable(vd) &&
1280 vd->vdev_aux == &spa->spa_l2cache &&
1281 !l2arc_vdev_present(vd)) {
1282 uint64_t size = vdev_get_rsize(vd);
1283 l2arc_add_vdev(spa, vd,
1284 VDEV_LABEL_START_SIZE,
1285 size - VDEV_LABEL_START_SIZE);
1288 (void) vdev_validate(vd);
1292 * Reassess parent vdev's health.
1294 vdev_propagate_state(vd);
1298 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1303 * Normally, partial opens (e.g. of a mirror) are allowed.
1304 * For a create, however, we want to fail the request if
1305 * there are any components we can't open.
1307 error = vdev_open(vd);
1309 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1311 return (error ? error : ENXIO);
1315 * Recursively initialize all labels.
1317 if ((error = vdev_label_init(vd, txg, isreplacing ?
1318 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1327 * The is the latter half of vdev_create(). It is distinct because it
1328 * involves initiating transactions in order to do metaslab creation.
1329 * For creation, we want to try to create all vdevs at once and then undo it
1330 * if anything fails; this is much harder if we have pending transactions.
1333 vdev_init(vdev_t *vd, uint64_t txg)
1336 * Aim for roughly 200 metaslabs per vdev.
1338 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1339 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1342 * Initialize the vdev's metaslabs. This can't fail because
1343 * there's nothing to read when creating all new metaslabs.
1345 VERIFY(vdev_metaslab_init(vd, txg) == 0);
1349 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1351 ASSERT(vd == vd->vdev_top);
1352 ASSERT(ISP2(flags));
1354 if (flags & VDD_METASLAB)
1355 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1357 if (flags & VDD_DTL)
1358 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1360 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1366 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1367 * the vdev has less than perfect replication. There are three kinds of DTL:
1369 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1371 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1373 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1374 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1375 * txgs that was scrubbed.
1377 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1378 * persistent errors or just some device being offline.
1379 * Unlike the other three, the DTL_OUTAGE map is not generally
1380 * maintained; it's only computed when needed, typically to
1381 * determine whether a device can be detached.
1383 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1384 * either has the data or it doesn't.
1386 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1387 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1388 * if any child is less than fully replicated, then so is its parent.
1389 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1390 * comprising only those txgs which appear in 'maxfaults' or more children;
1391 * those are the txgs we don't have enough replication to read. For example,
1392 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1393 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1394 * two child DTL_MISSING maps.
1396 * It should be clear from the above that to compute the DTLs and outage maps
1397 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1398 * Therefore, that is all we keep on disk. When loading the pool, or after
1399 * a configuration change, we generate all other DTLs from first principles.
1402 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1404 space_map_t *sm = &vd->vdev_dtl[t];
1406 ASSERT(t < DTL_TYPES);
1407 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1409 mutex_enter(sm->sm_lock);
1410 if (!space_map_contains(sm, txg, size))
1411 space_map_add(sm, txg, size);
1412 mutex_exit(sm->sm_lock);
1416 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1418 space_map_t *sm = &vd->vdev_dtl[t];
1419 boolean_t dirty = B_FALSE;
1421 ASSERT(t < DTL_TYPES);
1422 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1424 mutex_enter(sm->sm_lock);
1425 if (sm->sm_space != 0)
1426 dirty = space_map_contains(sm, txg, size);
1427 mutex_exit(sm->sm_lock);
1433 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1435 space_map_t *sm = &vd->vdev_dtl[t];
1438 mutex_enter(sm->sm_lock);
1439 empty = (sm->sm_space == 0);
1440 mutex_exit(sm->sm_lock);
1446 * Reassess DTLs after a config change or scrub completion.
1449 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1451 spa_t *spa = vd->vdev_spa;
1455 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1457 for (int c = 0; c < vd->vdev_children; c++)
1458 vdev_dtl_reassess(vd->vdev_child[c], txg,
1459 scrub_txg, scrub_done);
1461 if (vd == spa->spa_root_vdev)
1464 if (vd->vdev_ops->vdev_op_leaf) {
1465 mutex_enter(&vd->vdev_dtl_lock);
1466 if (scrub_txg != 0 &&
1467 (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) {
1468 /* XXX should check scrub_done? */
1470 * We completed a scrub up to scrub_txg. If we
1471 * did it without rebooting, then the scrub dtl
1472 * will be valid, so excise the old region and
1473 * fold in the scrub dtl. Otherwise, leave the
1474 * dtl as-is if there was an error.
1476 * There's little trick here: to excise the beginning
1477 * of the DTL_MISSING map, we put it into a reference
1478 * tree and then add a segment with refcnt -1 that
1479 * covers the range [0, scrub_txg). This means
1480 * that each txg in that range has refcnt -1 or 0.
1481 * We then add DTL_SCRUB with a refcnt of 2, so that
1482 * entries in the range [0, scrub_txg) will have a
1483 * positive refcnt -- either 1 or 2. We then convert
1484 * the reference tree into the new DTL_MISSING map.
1486 space_map_ref_create(&reftree);
1487 space_map_ref_add_map(&reftree,
1488 &vd->vdev_dtl[DTL_MISSING], 1);
1489 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1490 space_map_ref_add_map(&reftree,
1491 &vd->vdev_dtl[DTL_SCRUB], 2);
1492 space_map_ref_generate_map(&reftree,
1493 &vd->vdev_dtl[DTL_MISSING], 1);
1494 space_map_ref_destroy(&reftree);
1496 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1497 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1498 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1500 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1501 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1502 if (!vdev_readable(vd))
1503 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1505 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1506 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1507 mutex_exit(&vd->vdev_dtl_lock);
1510 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1514 mutex_enter(&vd->vdev_dtl_lock);
1515 for (int t = 0; t < DTL_TYPES; t++) {
1516 /* account for child's outage in parent's missing map */
1517 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1519 continue; /* leaf vdevs only */
1520 if (t == DTL_PARTIAL)
1521 minref = 1; /* i.e. non-zero */
1522 else if (vd->vdev_nparity != 0)
1523 minref = vd->vdev_nparity + 1; /* RAID-Z */
1525 minref = vd->vdev_children; /* any kind of mirror */
1526 space_map_ref_create(&reftree);
1527 for (int c = 0; c < vd->vdev_children; c++) {
1528 vdev_t *cvd = vd->vdev_child[c];
1529 mutex_enter(&cvd->vdev_dtl_lock);
1530 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1531 mutex_exit(&cvd->vdev_dtl_lock);
1533 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1534 space_map_ref_destroy(&reftree);
1536 mutex_exit(&vd->vdev_dtl_lock);
1540 vdev_dtl_load(vdev_t *vd)
1542 spa_t *spa = vd->vdev_spa;
1543 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1544 objset_t *mos = spa->spa_meta_objset;
1548 ASSERT(vd->vdev_children == 0);
1550 if (smo->smo_object == 0)
1553 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1556 ASSERT3U(db->db_size, >=, sizeof (*smo));
1557 bcopy(db->db_data, smo, sizeof (*smo));
1558 dmu_buf_rele(db, FTAG);
1560 mutex_enter(&vd->vdev_dtl_lock);
1561 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1562 NULL, SM_ALLOC, smo, mos);
1563 mutex_exit(&vd->vdev_dtl_lock);
1569 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1571 spa_t *spa = vd->vdev_spa;
1572 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1573 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1574 objset_t *mos = spa->spa_meta_objset;
1580 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1582 if (vd->vdev_detached) {
1583 if (smo->smo_object != 0) {
1584 int err = dmu_object_free(mos, smo->smo_object, tx);
1585 ASSERT3U(err, ==, 0);
1586 smo->smo_object = 0;
1592 if (smo->smo_object == 0) {
1593 ASSERT(smo->smo_objsize == 0);
1594 ASSERT(smo->smo_alloc == 0);
1595 smo->smo_object = dmu_object_alloc(mos,
1596 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1597 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1598 ASSERT(smo->smo_object != 0);
1599 vdev_config_dirty(vd->vdev_top);
1602 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1604 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1607 mutex_enter(&smlock);
1609 mutex_enter(&vd->vdev_dtl_lock);
1610 space_map_walk(sm, space_map_add, &smsync);
1611 mutex_exit(&vd->vdev_dtl_lock);
1613 space_map_truncate(smo, mos, tx);
1614 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1616 space_map_destroy(&smsync);
1618 mutex_exit(&smlock);
1619 mutex_destroy(&smlock);
1621 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1622 dmu_buf_will_dirty(db, tx);
1623 ASSERT3U(db->db_size, >=, sizeof (*smo));
1624 bcopy(smo, db->db_data, sizeof (*smo));
1625 dmu_buf_rele(db, FTAG);
1631 * Determine whether the specified vdev can be offlined/detached/removed
1632 * without losing data.
1635 vdev_dtl_required(vdev_t *vd)
1637 spa_t *spa = vd->vdev_spa;
1638 vdev_t *tvd = vd->vdev_top;
1639 uint8_t cant_read = vd->vdev_cant_read;
1642 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1644 if (vd == spa->spa_root_vdev || vd == tvd)
1648 * Temporarily mark the device as unreadable, and then determine
1649 * whether this results in any DTL outages in the top-level vdev.
1650 * If not, we can safely offline/detach/remove the device.
1652 vd->vdev_cant_read = B_TRUE;
1653 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1654 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1655 vd->vdev_cant_read = cant_read;
1656 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1662 * Determine if resilver is needed, and if so the txg range.
1665 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1667 boolean_t needed = B_FALSE;
1668 uint64_t thismin = UINT64_MAX;
1669 uint64_t thismax = 0;
1671 if (vd->vdev_children == 0) {
1672 mutex_enter(&vd->vdev_dtl_lock);
1673 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1674 vdev_writeable(vd)) {
1677 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1678 thismin = ss->ss_start - 1;
1679 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1680 thismax = ss->ss_end;
1683 mutex_exit(&vd->vdev_dtl_lock);
1685 for (int c = 0; c < vd->vdev_children; c++) {
1686 vdev_t *cvd = vd->vdev_child[c];
1687 uint64_t cmin, cmax;
1689 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1690 thismin = MIN(thismin, cmin);
1691 thismax = MAX(thismax, cmax);
1697 if (needed && minp) {
1705 vdev_load(vdev_t *vd)
1708 * Recursively load all children.
1710 for (int c = 0; c < vd->vdev_children; c++)
1711 vdev_load(vd->vdev_child[c]);
1714 * If this is a top-level vdev, initialize its metaslabs.
1716 if (vd == vd->vdev_top &&
1717 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1718 vdev_metaslab_init(vd, 0) != 0))
1719 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1720 VDEV_AUX_CORRUPT_DATA);
1723 * If this is a leaf vdev, load its DTL.
1725 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1726 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1727 VDEV_AUX_CORRUPT_DATA);
1731 * The special vdev case is used for hot spares and l2cache devices. Its
1732 * sole purpose it to set the vdev state for the associated vdev. To do this,
1733 * we make sure that we can open the underlying device, then try to read the
1734 * label, and make sure that the label is sane and that it hasn't been
1735 * repurposed to another pool.
1738 vdev_validate_aux(vdev_t *vd)
1741 uint64_t guid, version;
1744 if (!vdev_readable(vd))
1747 if ((label = vdev_label_read_config(vd)) == NULL) {
1748 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1749 VDEV_AUX_CORRUPT_DATA);
1753 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1754 version > SPA_VERSION ||
1755 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1756 guid != vd->vdev_guid ||
1757 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1758 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1759 VDEV_AUX_CORRUPT_DATA);
1765 * We don't actually check the pool state here. If it's in fact in
1766 * use by another pool, we update this fact on the fly when requested.
1773 vdev_sync_done(vdev_t *vd, uint64_t txg)
1776 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
1778 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
1779 metaslab_sync_done(msp, txg);
1782 metaslab_sync_reassess(vd->vdev_mg);
1786 vdev_sync(vdev_t *vd, uint64_t txg)
1788 spa_t *spa = vd->vdev_spa;
1793 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
1794 ASSERT(vd == vd->vdev_top);
1795 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1796 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
1797 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
1798 ASSERT(vd->vdev_ms_array != 0);
1799 vdev_config_dirty(vd);
1803 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
1804 metaslab_sync(msp, txg);
1805 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
1808 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
1809 vdev_dtl_sync(lvd, txg);
1811 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
1815 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
1817 return (vd->vdev_ops->vdev_op_asize(vd, psize));
1821 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
1822 * not be opened, and no I/O is attempted.
1825 vdev_fault(spa_t *spa, uint64_t guid)
1829 spa_vdev_state_enter(spa);
1831 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1832 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1834 if (!vd->vdev_ops->vdev_op_leaf)
1835 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1838 * Faulted state takes precedence over degraded.
1840 vd->vdev_faulted = 1ULL;
1841 vd->vdev_degraded = 0ULL;
1842 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED);
1845 * If marking the vdev as faulted cause the top-level vdev to become
1846 * unavailable, then back off and simply mark the vdev as degraded
1849 if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) {
1850 vd->vdev_degraded = 1ULL;
1851 vd->vdev_faulted = 0ULL;
1854 * If we reopen the device and it's not dead, only then do we
1859 if (vdev_readable(vd)) {
1860 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1861 VDEV_AUX_ERR_EXCEEDED);
1865 return (spa_vdev_state_exit(spa, vd, 0));
1869 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
1870 * user that something is wrong. The vdev continues to operate as normal as far
1871 * as I/O is concerned.
1874 vdev_degrade(spa_t *spa, uint64_t guid)
1878 spa_vdev_state_enter(spa);
1880 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1881 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1883 if (!vd->vdev_ops->vdev_op_leaf)
1884 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1887 * If the vdev is already faulted, then don't do anything.
1889 if (vd->vdev_faulted || vd->vdev_degraded)
1890 return (spa_vdev_state_exit(spa, NULL, 0));
1892 vd->vdev_degraded = 1ULL;
1893 if (!vdev_is_dead(vd))
1894 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1895 VDEV_AUX_ERR_EXCEEDED);
1897 return (spa_vdev_state_exit(spa, vd, 0));
1901 * Online the given vdev. If 'unspare' is set, it implies two things. First,
1902 * any attached spare device should be detached when the device finishes
1903 * resilvering. Second, the online should be treated like a 'test' online case,
1904 * so no FMA events are generated if the device fails to open.
1907 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
1911 spa_vdev_state_enter(spa);
1913 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1914 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1916 if (!vd->vdev_ops->vdev_op_leaf)
1917 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1919 vd->vdev_offline = B_FALSE;
1920 vd->vdev_tmpoffline = B_FALSE;
1921 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
1922 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
1923 vdev_reopen(vd->vdev_top);
1924 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
1927 *newstate = vd->vdev_state;
1928 if ((flags & ZFS_ONLINE_UNSPARE) &&
1929 !vdev_is_dead(vd) && vd->vdev_parent &&
1930 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
1931 vd->vdev_parent->vdev_child[0] == vd)
1932 vd->vdev_unspare = B_TRUE;
1934 return (spa_vdev_state_exit(spa, vd, 0));
1938 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
1943 spa_vdev_state_enter(spa);
1945 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1946 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1948 if (!vd->vdev_ops->vdev_op_leaf)
1949 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1954 * If the device isn't already offline, try to offline it.
1956 if (!vd->vdev_offline) {
1958 * If this device has the only valid copy of some data,
1959 * don't allow it to be offlined. Log devices are always
1962 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
1963 vdev_dtl_required(vd))
1964 return (spa_vdev_state_exit(spa, NULL, EBUSY));
1967 * Offline this device and reopen its top-level vdev.
1968 * If the top-level vdev is a log device then just offline
1969 * it. Otherwise, if this action results in the top-level
1970 * vdev becoming unusable, undo it and fail the request.
1972 vd->vdev_offline = B_TRUE;
1975 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
1976 vdev_is_dead(tvd)) {
1977 vd->vdev_offline = B_FALSE;
1979 return (spa_vdev_state_exit(spa, NULL, EBUSY));
1983 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
1985 if (!tvd->vdev_islog || !vdev_is_dead(tvd))
1986 return (spa_vdev_state_exit(spa, vd, 0));
1988 (void) spa_vdev_state_exit(spa, vd, 0);
1990 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1991 NULL, DS_FIND_CHILDREN);
1993 (void) vdev_online(spa, guid, 0, NULL);
1997 * If we successfully offlined the log device then we need to
1998 * sync out the current txg so that the "stubby" block can be
1999 * removed by zil_sync().
2001 txg_wait_synced(spa->spa_dsl_pool, 0);
2006 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2007 * vdev_offline(), we assume the spa config is locked. We also clear all
2008 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2011 vdev_clear(spa_t *spa, vdev_t *vd)
2013 vdev_t *rvd = spa->spa_root_vdev;
2015 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2020 vd->vdev_stat.vs_read_errors = 0;
2021 vd->vdev_stat.vs_write_errors = 0;
2022 vd->vdev_stat.vs_checksum_errors = 0;
2024 for (int c = 0; c < vd->vdev_children; c++)
2025 vdev_clear(spa, vd->vdev_child[c]);
2028 * If we're in the FAULTED state or have experienced failed I/O, then
2029 * clear the persistent state and attempt to reopen the device. We
2030 * also mark the vdev config dirty, so that the new faulted state is
2031 * written out to disk.
2033 if (vd->vdev_faulted || vd->vdev_degraded ||
2034 !vdev_readable(vd) || !vdev_writeable(vd)) {
2036 vd->vdev_faulted = vd->vdev_degraded = 0;
2037 vd->vdev_cant_read = B_FALSE;
2038 vd->vdev_cant_write = B_FALSE;
2043 vdev_state_dirty(vd->vdev_top);
2045 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2046 spa_async_request(spa, SPA_ASYNC_RESILVER);
2048 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2053 vdev_is_dead(vdev_t *vd)
2055 return (vd->vdev_state < VDEV_STATE_DEGRADED);
2059 vdev_readable(vdev_t *vd)
2061 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2065 vdev_writeable(vdev_t *vd)
2067 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2071 vdev_allocatable(vdev_t *vd)
2073 uint64_t state = vd->vdev_state;
2076 * We currently allow allocations from vdevs which may be in the
2077 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2078 * fails to reopen then we'll catch it later when we're holding
2079 * the proper locks. Note that we have to get the vdev state
2080 * in a local variable because although it changes atomically,
2081 * we're asking two separate questions about it.
2083 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2084 !vd->vdev_cant_write);
2088 vdev_accessible(vdev_t *vd, zio_t *zio)
2090 ASSERT(zio->io_vd == vd);
2092 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2095 if (zio->io_type == ZIO_TYPE_READ)
2096 return (!vd->vdev_cant_read);
2098 if (zio->io_type == ZIO_TYPE_WRITE)
2099 return (!vd->vdev_cant_write);
2105 * Get statistics for the given vdev.
2108 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2110 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2112 mutex_enter(&vd->vdev_stat_lock);
2113 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2114 vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors;
2115 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2116 vs->vs_state = vd->vdev_state;
2117 vs->vs_rsize = vdev_get_rsize(vd);
2118 mutex_exit(&vd->vdev_stat_lock);
2121 * If we're getting stats on the root vdev, aggregate the I/O counts
2122 * over all top-level vdevs (i.e. the direct children of the root).
2125 for (int c = 0; c < rvd->vdev_children; c++) {
2126 vdev_t *cvd = rvd->vdev_child[c];
2127 vdev_stat_t *cvs = &cvd->vdev_stat;
2129 mutex_enter(&vd->vdev_stat_lock);
2130 for (int t = 0; t < ZIO_TYPES; t++) {
2131 vs->vs_ops[t] += cvs->vs_ops[t];
2132 vs->vs_bytes[t] += cvs->vs_bytes[t];
2134 vs->vs_scrub_examined += cvs->vs_scrub_examined;
2135 mutex_exit(&vd->vdev_stat_lock);
2141 vdev_clear_stats(vdev_t *vd)
2143 mutex_enter(&vd->vdev_stat_lock);
2144 vd->vdev_stat.vs_space = 0;
2145 vd->vdev_stat.vs_dspace = 0;
2146 vd->vdev_stat.vs_alloc = 0;
2147 mutex_exit(&vd->vdev_stat_lock);
2151 vdev_stat_update(zio_t *zio, uint64_t psize)
2153 spa_t *spa = zio->io_spa;
2154 vdev_t *rvd = spa->spa_root_vdev;
2155 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2157 uint64_t txg = zio->io_txg;
2158 vdev_stat_t *vs = &vd->vdev_stat;
2159 zio_type_t type = zio->io_type;
2160 int flags = zio->io_flags;
2163 * If this i/o is a gang leader, it didn't do any actual work.
2165 if (zio->io_gang_tree)
2168 if (zio->io_error == 0) {
2170 * If this is a root i/o, don't count it -- we've already
2171 * counted the top-level vdevs, and vdev_get_stats() will
2172 * aggregate them when asked. This reduces contention on
2173 * the root vdev_stat_lock and implicitly handles blocks
2174 * that compress away to holes, for which there is no i/o.
2175 * (Holes never create vdev children, so all the counters
2176 * remain zero, which is what we want.)
2178 * Note: this only applies to successful i/o (io_error == 0)
2179 * because unlike i/o counts, errors are not additive.
2180 * When reading a ditto block, for example, failure of
2181 * one top-level vdev does not imply a root-level error.
2186 ASSERT(vd == zio->io_vd);
2188 if (flags & ZIO_FLAG_IO_BYPASS)
2191 mutex_enter(&vd->vdev_stat_lock);
2193 if (flags & ZIO_FLAG_IO_REPAIR) {
2194 if (flags & ZIO_FLAG_SCRUB_THREAD)
2195 vs->vs_scrub_repaired += psize;
2196 if (flags & ZIO_FLAG_SELF_HEAL)
2197 vs->vs_self_healed += psize;
2201 vs->vs_bytes[type] += psize;
2203 mutex_exit(&vd->vdev_stat_lock);
2207 if (flags & ZIO_FLAG_SPECULATIVE)
2211 * If this is an I/O error that is going to be retried, then ignore the
2212 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2213 * hard errors, when in reality they can happen for any number of
2214 * innocuous reasons (bus resets, MPxIO link failure, etc).
2216 if (zio->io_error == EIO &&
2217 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2220 mutex_enter(&vd->vdev_stat_lock);
2221 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2222 if (zio->io_error == ECKSUM)
2223 vs->vs_checksum_errors++;
2225 vs->vs_read_errors++;
2227 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2228 vs->vs_write_errors++;
2229 mutex_exit(&vd->vdev_stat_lock);
2231 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2232 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2233 (flags & ZIO_FLAG_SCRUB_THREAD))) {
2235 * This is either a normal write (not a repair), or it's a
2236 * repair induced by the scrub thread. In the normal case,
2237 * we commit the DTL change in the same txg as the block
2238 * was born. In the scrub-induced repair case, we know that
2239 * scrubs run in first-pass syncing context, so we commit
2240 * the DTL change in spa->spa_syncing_txg.
2242 * We currently do not make DTL entries for failed spontaneous
2243 * self-healing writes triggered by normal (non-scrubbing)
2244 * reads, because we have no transactional context in which to
2245 * do so -- and it's not clear that it'd be desirable anyway.
2247 if (vd->vdev_ops->vdev_op_leaf) {
2248 uint64_t commit_txg = txg;
2249 if (flags & ZIO_FLAG_SCRUB_THREAD) {
2250 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2251 ASSERT(spa_sync_pass(spa) == 1);
2252 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2253 commit_txg = spa->spa_syncing_txg;
2255 ASSERT(commit_txg >= spa->spa_syncing_txg);
2256 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2258 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2259 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2260 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2263 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2268 vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete)
2271 vdev_stat_t *vs = &vd->vdev_stat;
2273 for (c = 0; c < vd->vdev_children; c++)
2274 vdev_scrub_stat_update(vd->vdev_child[c], type, complete);
2276 mutex_enter(&vd->vdev_stat_lock);
2278 if (type == POOL_SCRUB_NONE) {
2280 * Update completion and end time. Leave everything else alone
2281 * so we can report what happened during the previous scrub.
2283 vs->vs_scrub_complete = complete;
2284 vs->vs_scrub_end = gethrestime_sec();
2286 vs->vs_scrub_type = type;
2287 vs->vs_scrub_complete = 0;
2288 vs->vs_scrub_examined = 0;
2289 vs->vs_scrub_repaired = 0;
2290 vs->vs_scrub_start = gethrestime_sec();
2291 vs->vs_scrub_end = 0;
2294 mutex_exit(&vd->vdev_stat_lock);
2298 * Update the in-core space usage stats for this vdev and the root vdev.
2301 vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta,
2302 boolean_t update_root)
2304 int64_t dspace_delta = space_delta;
2305 spa_t *spa = vd->vdev_spa;
2306 vdev_t *rvd = spa->spa_root_vdev;
2308 ASSERT(vd == vd->vdev_top);
2311 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2312 * factor. We must calculate this here and not at the root vdev
2313 * because the root vdev's psize-to-asize is simply the max of its
2314 * childrens', thus not accurate enough for us.
2316 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2317 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2318 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2319 vd->vdev_deflate_ratio;
2321 mutex_enter(&vd->vdev_stat_lock);
2322 vd->vdev_stat.vs_space += space_delta;
2323 vd->vdev_stat.vs_alloc += alloc_delta;
2324 vd->vdev_stat.vs_dspace += dspace_delta;
2325 mutex_exit(&vd->vdev_stat_lock);
2328 ASSERT(rvd == vd->vdev_parent);
2329 ASSERT(vd->vdev_ms_count != 0);
2332 * Don't count non-normal (e.g. intent log) space as part of
2333 * the pool's capacity.
2335 if (vd->vdev_mg->mg_class != spa->spa_normal_class)
2338 mutex_enter(&rvd->vdev_stat_lock);
2339 rvd->vdev_stat.vs_space += space_delta;
2340 rvd->vdev_stat.vs_alloc += alloc_delta;
2341 rvd->vdev_stat.vs_dspace += dspace_delta;
2342 mutex_exit(&rvd->vdev_stat_lock);
2347 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2348 * so that it will be written out next time the vdev configuration is synced.
2349 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2352 vdev_config_dirty(vdev_t *vd)
2354 spa_t *spa = vd->vdev_spa;
2355 vdev_t *rvd = spa->spa_root_vdev;
2359 * If this is an aux vdev (as with l2cache and spare devices), then we
2360 * update the vdev config manually and set the sync flag.
2362 if (vd->vdev_aux != NULL) {
2363 spa_aux_vdev_t *sav = vd->vdev_aux;
2367 for (c = 0; c < sav->sav_count; c++) {
2368 if (sav->sav_vdevs[c] == vd)
2372 if (c == sav->sav_count) {
2374 * We're being removed. There's nothing more to do.
2376 ASSERT(sav->sav_sync == B_TRUE);
2380 sav->sav_sync = B_TRUE;
2382 if (nvlist_lookup_nvlist_array(sav->sav_config,
2383 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2384 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2385 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2391 * Setting the nvlist in the middle if the array is a little
2392 * sketchy, but it will work.
2394 nvlist_free(aux[c]);
2395 aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE);
2401 * The dirty list is protected by the SCL_CONFIG lock. The caller
2402 * must either hold SCL_CONFIG as writer, or must be the sync thread
2403 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2404 * so this is sufficient to ensure mutual exclusion.
2406 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2407 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2408 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2411 for (c = 0; c < rvd->vdev_children; c++)
2412 vdev_config_dirty(rvd->vdev_child[c]);
2414 ASSERT(vd == vd->vdev_top);
2416 if (!list_link_active(&vd->vdev_config_dirty_node))
2417 list_insert_head(&spa->spa_config_dirty_list, vd);
2422 vdev_config_clean(vdev_t *vd)
2424 spa_t *spa = vd->vdev_spa;
2426 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2427 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2428 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2430 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2431 list_remove(&spa->spa_config_dirty_list, vd);
2435 * Mark a top-level vdev's state as dirty, so that the next pass of
2436 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2437 * the state changes from larger config changes because they require
2438 * much less locking, and are often needed for administrative actions.
2441 vdev_state_dirty(vdev_t *vd)
2443 spa_t *spa = vd->vdev_spa;
2445 ASSERT(vd == vd->vdev_top);
2448 * The state list is protected by the SCL_STATE lock. The caller
2449 * must either hold SCL_STATE as writer, or must be the sync thread
2450 * (which holds SCL_STATE as reader). There's only one sync thread,
2451 * so this is sufficient to ensure mutual exclusion.
2453 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2454 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2455 spa_config_held(spa, SCL_STATE, RW_READER)));
2457 if (!list_link_active(&vd->vdev_state_dirty_node))
2458 list_insert_head(&spa->spa_state_dirty_list, vd);
2462 vdev_state_clean(vdev_t *vd)
2464 spa_t *spa = vd->vdev_spa;
2466 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2467 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2468 spa_config_held(spa, SCL_STATE, RW_READER)));
2470 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2471 list_remove(&spa->spa_state_dirty_list, vd);
2475 * Propagate vdev state up from children to parent.
2478 vdev_propagate_state(vdev_t *vd)
2480 spa_t *spa = vd->vdev_spa;
2481 vdev_t *rvd = spa->spa_root_vdev;
2482 int degraded = 0, faulted = 0;
2487 if (vd->vdev_children > 0) {
2488 for (c = 0; c < vd->vdev_children; c++) {
2489 child = vd->vdev_child[c];
2491 if (!vdev_readable(child) ||
2492 (!vdev_writeable(child) && spa_writeable(spa))) {
2494 * Root special: if there is a top-level log
2495 * device, treat the root vdev as if it were
2498 if (child->vdev_islog && vd == rvd)
2502 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2506 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2510 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2513 * Root special: if there is a top-level vdev that cannot be
2514 * opened due to corrupted metadata, then propagate the root
2515 * vdev's aux state as 'corrupt' rather than 'insufficient
2518 if (corrupted && vd == rvd &&
2519 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2520 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2521 VDEV_AUX_CORRUPT_DATA);
2524 if (vd->vdev_parent)
2525 vdev_propagate_state(vd->vdev_parent);
2529 * Set a vdev's state. If this is during an open, we don't update the parent
2530 * state, because we're in the process of opening children depth-first.
2531 * Otherwise, we propagate the change to the parent.
2533 * If this routine places a device in a faulted state, an appropriate ereport is
2537 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2539 uint64_t save_state;
2540 spa_t *spa = vd->vdev_spa;
2542 if (state == vd->vdev_state) {
2543 vd->vdev_stat.vs_aux = aux;
2547 save_state = vd->vdev_state;
2549 vd->vdev_state = state;
2550 vd->vdev_stat.vs_aux = aux;
2553 * If we are setting the vdev state to anything but an open state, then
2554 * always close the underlying device. Otherwise, we keep accessible
2555 * but invalid devices open forever. We don't call vdev_close() itself,
2556 * because that implies some extra checks (offline, etc) that we don't
2557 * want here. This is limited to leaf devices, because otherwise
2558 * closing the device will affect other children.
2560 if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf)
2561 vd->vdev_ops->vdev_op_close(vd);
2563 if (vd->vdev_removed &&
2564 state == VDEV_STATE_CANT_OPEN &&
2565 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2567 * If the previous state is set to VDEV_STATE_REMOVED, then this
2568 * device was previously marked removed and someone attempted to
2569 * reopen it. If this failed due to a nonexistent device, then
2570 * keep the device in the REMOVED state. We also let this be if
2571 * it is one of our special test online cases, which is only
2572 * attempting to online the device and shouldn't generate an FMA
2575 vd->vdev_state = VDEV_STATE_REMOVED;
2576 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2577 } else if (state == VDEV_STATE_REMOVED) {
2579 * Indicate to the ZFS DE that this device has been removed, and
2580 * any recent errors should be ignored.
2582 zfs_post_remove(spa, vd);
2583 vd->vdev_removed = B_TRUE;
2584 } else if (state == VDEV_STATE_CANT_OPEN) {
2586 * If we fail to open a vdev during an import, we mark it as
2587 * "not available", which signifies that it was never there to
2588 * begin with. Failure to open such a device is not considered
2591 if (spa->spa_load_state == SPA_LOAD_IMPORT &&
2592 vd->vdev_ops->vdev_op_leaf)
2593 vd->vdev_not_present = 1;
2596 * Post the appropriate ereport. If the 'prevstate' field is
2597 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2598 * that this is part of a vdev_reopen(). In this case, we don't
2599 * want to post the ereport if the device was already in the
2600 * CANT_OPEN state beforehand.
2602 * If the 'checkremove' flag is set, then this is an attempt to
2603 * online the device in response to an insertion event. If we
2604 * hit this case, then we have detected an insertion event for a
2605 * faulted or offline device that wasn't in the removed state.
2606 * In this scenario, we don't post an ereport because we are
2607 * about to replace the device, or attempt an online with
2608 * vdev_forcefault, which will generate the fault for us.
2610 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
2611 !vd->vdev_not_present && !vd->vdev_checkremove &&
2612 vd != spa->spa_root_vdev) {
2616 case VDEV_AUX_OPEN_FAILED:
2617 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
2619 case VDEV_AUX_CORRUPT_DATA:
2620 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
2622 case VDEV_AUX_NO_REPLICAS:
2623 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
2625 case VDEV_AUX_BAD_GUID_SUM:
2626 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
2628 case VDEV_AUX_TOO_SMALL:
2629 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
2631 case VDEV_AUX_BAD_LABEL:
2632 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
2634 case VDEV_AUX_IO_FAILURE:
2635 class = FM_EREPORT_ZFS_IO_FAILURE;
2638 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
2641 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
2644 /* Erase any notion of persistent removed state */
2645 vd->vdev_removed = B_FALSE;
2647 vd->vdev_removed = B_FALSE;
2650 if (!isopen && vd->vdev_parent)
2651 vdev_propagate_state(vd->vdev_parent);
2655 * Check the vdev configuration to ensure that it's capable of supporting
2658 * On Solaris, we do not support RAID-Z or partial configuration. In
2659 * addition, only a single top-level vdev is allowed and none of the
2660 * leaves can be wholedisks.
2662 * For FreeBSD, we can boot from any configuration. There is a
2663 * limitation that the boot filesystem must be either uncompressed or
2664 * compresses with lzjb compression but I'm not sure how to enforce
2668 vdev_is_bootable(vdev_t *vd)
2671 if (!vd->vdev_ops->vdev_op_leaf) {
2672 char *vdev_type = vd->vdev_ops->vdev_op_type;
2674 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
2675 vd->vdev_children > 1) {
2677 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
2678 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
2681 } else if (vd->vdev_wholedisk == 1) {
2685 for (c = 0; c < vd->vdev_children; c++) {
2686 if (!vdev_is_bootable(vd->vdev_child[c]))
2694 vdev_load_log_state(vdev_t *vd, nvlist_t *nv)
2699 spa_t *spa = vd->vdev_spa;
2701 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2702 &child, &children) == 0) {
2703 for (c = 0; c < children; c++)
2704 vdev_load_log_state(vd->vdev_child[c], child[c]);
2707 if (vd->vdev_ops->vdev_op_leaf && nvlist_lookup_uint64(nv,
2708 ZPOOL_CONFIG_OFFLINE, &val) == 0 && val) {
2711 * It would be nice to call vdev_offline()
2712 * directly but the pool isn't fully loaded and
2713 * the txg threads have not been started yet.
2715 spa_config_enter(spa, SCL_STATE_ALL, FTAG, RW_WRITER);
2716 vd->vdev_offline = val;
2717 vdev_reopen(vd->vdev_top);
2718 spa_config_exit(spa, SCL_STATE_ALL, FTAG);