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>
45 * Virtual device management.
48 static vdev_ops_t *vdev_ops_table[] = {
60 /* maximum scrub/resilver I/O queue per leaf vdev */
61 int zfs_scrub_limit = 10;
64 * Given a vdev type, return the appropriate ops vector.
67 vdev_getops(const char *type)
69 vdev_ops_t *ops, **opspp;
71 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
72 if (strcmp(ops->vdev_op_type, type) == 0)
79 * Default asize function: return the MAX of psize with the asize of
80 * all children. This is what's used by anything other than RAID-Z.
83 vdev_default_asize(vdev_t *vd, uint64_t psize)
85 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
88 for (int c = 0; c < vd->vdev_children; c++) {
89 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
90 asize = MAX(asize, csize);
97 * Get the minimum allocatable size. We define the allocatable size as
98 * the vdev's asize rounded to the nearest metaslab. This allows us to
99 * replace or attach devices which don't have the same physical size but
100 * can still satisfy the same number of allocations.
103 vdev_get_min_asize(vdev_t *vd)
105 vdev_t *pvd = vd->vdev_parent;
108 * The our parent is NULL (inactive spare or cache) or is the root,
109 * just return our own asize.
112 return (vd->vdev_asize);
115 * The top-level vdev just returns the allocatable size rounded
116 * to the nearest metaslab.
118 if (vd == vd->vdev_top)
119 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
122 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
123 * so each child must provide at least 1/Nth of its asize.
125 if (pvd->vdev_ops == &vdev_raidz_ops)
126 return (pvd->vdev_min_asize / pvd->vdev_children);
128 return (pvd->vdev_min_asize);
132 vdev_set_min_asize(vdev_t *vd)
134 vd->vdev_min_asize = vdev_get_min_asize(vd);
136 for (int c = 0; c < vd->vdev_children; c++)
137 vdev_set_min_asize(vd->vdev_child[c]);
141 vdev_lookup_top(spa_t *spa, uint64_t vdev)
143 vdev_t *rvd = spa->spa_root_vdev;
145 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
147 if (vdev < rvd->vdev_children) {
148 ASSERT(rvd->vdev_child[vdev] != NULL);
149 return (rvd->vdev_child[vdev]);
156 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
160 if (vd->vdev_guid == guid)
163 for (int c = 0; c < vd->vdev_children; c++)
164 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
172 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
174 size_t oldsize, newsize;
175 uint64_t id = cvd->vdev_id;
178 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
179 ASSERT(cvd->vdev_parent == NULL);
181 cvd->vdev_parent = pvd;
186 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
188 oldsize = pvd->vdev_children * sizeof (vdev_t *);
189 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
190 newsize = pvd->vdev_children * sizeof (vdev_t *);
192 newchild = kmem_zalloc(newsize, KM_SLEEP);
193 if (pvd->vdev_child != NULL) {
194 bcopy(pvd->vdev_child, newchild, oldsize);
195 kmem_free(pvd->vdev_child, oldsize);
198 pvd->vdev_child = newchild;
199 pvd->vdev_child[id] = cvd;
201 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
202 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
205 * Walk up all ancestors to update guid sum.
207 for (; pvd != NULL; pvd = pvd->vdev_parent)
208 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
210 if (cvd->vdev_ops->vdev_op_leaf)
211 cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit;
215 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
218 uint_t id = cvd->vdev_id;
220 ASSERT(cvd->vdev_parent == pvd);
225 ASSERT(id < pvd->vdev_children);
226 ASSERT(pvd->vdev_child[id] == cvd);
228 pvd->vdev_child[id] = NULL;
229 cvd->vdev_parent = NULL;
231 for (c = 0; c < pvd->vdev_children; c++)
232 if (pvd->vdev_child[c])
235 if (c == pvd->vdev_children) {
236 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
237 pvd->vdev_child = NULL;
238 pvd->vdev_children = 0;
242 * Walk up all ancestors to update guid sum.
244 for (; pvd != NULL; pvd = pvd->vdev_parent)
245 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
247 if (cvd->vdev_ops->vdev_op_leaf)
248 cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit;
252 * Remove any holes in the child array.
255 vdev_compact_children(vdev_t *pvd)
257 vdev_t **newchild, *cvd;
258 int oldc = pvd->vdev_children;
261 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
263 for (int c = newc = 0; c < oldc; c++)
264 if (pvd->vdev_child[c])
267 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
269 for (int c = newc = 0; c < oldc; c++) {
270 if ((cvd = pvd->vdev_child[c]) != NULL) {
271 newchild[newc] = cvd;
272 cvd->vdev_id = newc++;
276 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
277 pvd->vdev_child = newchild;
278 pvd->vdev_children = newc;
282 * Allocate and minimally initialize a vdev_t.
285 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
289 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
291 if (spa->spa_root_vdev == NULL) {
292 ASSERT(ops == &vdev_root_ops);
293 spa->spa_root_vdev = vd;
297 if (spa->spa_root_vdev == vd) {
299 * The root vdev's guid will also be the pool guid,
300 * which must be unique among all pools.
302 while (guid == 0 || spa_guid_exists(guid, 0))
303 guid = spa_get_random(-1ULL);
306 * Any other vdev's guid must be unique within the pool.
309 spa_guid_exists(spa_guid(spa), guid))
310 guid = spa_get_random(-1ULL);
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;
322 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
323 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
324 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
325 for (int t = 0; t < DTL_TYPES; t++) {
326 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
329 txg_list_create(&vd->vdev_ms_list,
330 offsetof(struct metaslab, ms_txg_node));
331 txg_list_create(&vd->vdev_dtl_list,
332 offsetof(struct vdev, vdev_dtl_node));
333 vd->vdev_stat.vs_timestamp = gethrtime();
341 * Allocate a new vdev. The 'alloctype' is used to control whether we are
342 * creating a new vdev or loading an existing one - the behavior is slightly
343 * different for each case.
346 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
351 uint64_t guid = 0, islog, nparity;
354 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
356 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
359 if ((ops = vdev_getops(type)) == NULL)
363 * If this is a load, get the vdev guid from the nvlist.
364 * Otherwise, vdev_alloc_common() will generate one for us.
366 if (alloctype == VDEV_ALLOC_LOAD) {
369 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
373 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
375 } else if (alloctype == VDEV_ALLOC_SPARE) {
376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
378 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
379 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
381 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
382 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
387 * The first allocated vdev must be of type 'root'.
389 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
393 * Determine whether we're a log vdev.
396 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
397 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
401 * Set the nparity property for RAID-Z vdevs.
404 if (ops == &vdev_raidz_ops) {
405 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
408 * Currently, we can only support 2 parity devices.
410 if (nparity == 0 || nparity > 2)
413 * Older versions can only support 1 parity device.
416 spa_version(spa) < SPA_VERSION_RAID6)
420 * We require the parity to be specified for SPAs that
421 * support multiple parity levels.
423 if (spa_version(spa) >= SPA_VERSION_RAID6)
426 * Otherwise, we default to 1 parity device for RAID-Z.
433 ASSERT(nparity != -1ULL);
435 vd = vdev_alloc_common(spa, id, guid, ops);
437 vd->vdev_islog = islog;
438 vd->vdev_nparity = nparity;
440 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
441 vd->vdev_path = spa_strdup(vd->vdev_path);
442 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
443 vd->vdev_devid = spa_strdup(vd->vdev_devid);
444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
445 &vd->vdev_physpath) == 0)
446 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
448 vd->vdev_fru = spa_strdup(vd->vdev_fru);
451 * Set the whole_disk property. If it's not specified, leave the value
454 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
455 &vd->vdev_wholedisk) != 0)
456 vd->vdev_wholedisk = -1ULL;
459 * Look for the 'not present' flag. This will only be set if the device
460 * was not present at the time of import.
462 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
463 &vd->vdev_not_present);
466 * Get the alignment requirement.
468 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
471 * If we're a top-level vdev, try to load the allocation parameters.
473 if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) {
474 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
476 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
478 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
483 * If we're a leaf vdev, try to load the DTL object and other state.
485 if (vd->vdev_ops->vdev_op_leaf &&
486 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
487 alloctype == VDEV_ALLOC_ROOTPOOL)) {
488 if (alloctype == VDEV_ALLOC_LOAD) {
489 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
490 &vd->vdev_dtl_smo.smo_object);
491 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
495 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
498 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
499 &spare) == 0 && spare)
503 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
507 * When importing a pool, we want to ignore the persistent fault
508 * state, as the diagnosis made on another system may not be
509 * valid in the current context.
511 if (spa->spa_load_state == SPA_LOAD_OPEN) {
512 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
516 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
522 * Add ourselves to the parent's list of children.
524 vdev_add_child(parent, vd);
532 vdev_free(vdev_t *vd)
534 spa_t *spa = vd->vdev_spa;
537 * vdev_free() implies closing the vdev first. This is simpler than
538 * trying to ensure complicated semantics for all callers.
542 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
547 for (int c = 0; c < vd->vdev_children; c++)
548 vdev_free(vd->vdev_child[c]);
550 ASSERT(vd->vdev_child == NULL);
551 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
554 * Discard allocation state.
556 if (vd == vd->vdev_top)
557 vdev_metaslab_fini(vd);
559 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
560 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
561 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
564 * Remove this vdev from its parent's child list.
566 vdev_remove_child(vd->vdev_parent, vd);
568 ASSERT(vd->vdev_parent == NULL);
571 * Clean up vdev structure.
577 spa_strfree(vd->vdev_path);
579 spa_strfree(vd->vdev_devid);
580 if (vd->vdev_physpath)
581 spa_strfree(vd->vdev_physpath);
583 spa_strfree(vd->vdev_fru);
585 if (vd->vdev_isspare)
586 spa_spare_remove(vd);
587 if (vd->vdev_isl2cache)
588 spa_l2cache_remove(vd);
590 txg_list_destroy(&vd->vdev_ms_list);
591 txg_list_destroy(&vd->vdev_dtl_list);
593 mutex_enter(&vd->vdev_dtl_lock);
594 for (int t = 0; t < DTL_TYPES; t++) {
595 space_map_unload(&vd->vdev_dtl[t]);
596 space_map_destroy(&vd->vdev_dtl[t]);
598 mutex_exit(&vd->vdev_dtl_lock);
600 mutex_destroy(&vd->vdev_dtl_lock);
601 mutex_destroy(&vd->vdev_stat_lock);
602 mutex_destroy(&vd->vdev_probe_lock);
604 if (vd == spa->spa_root_vdev)
605 spa->spa_root_vdev = NULL;
607 kmem_free(vd, sizeof (vdev_t));
611 * Transfer top-level vdev state from svd to tvd.
614 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
616 spa_t *spa = svd->vdev_spa;
621 ASSERT(tvd == tvd->vdev_top);
623 tvd->vdev_ms_array = svd->vdev_ms_array;
624 tvd->vdev_ms_shift = svd->vdev_ms_shift;
625 tvd->vdev_ms_count = svd->vdev_ms_count;
627 svd->vdev_ms_array = 0;
628 svd->vdev_ms_shift = 0;
629 svd->vdev_ms_count = 0;
631 tvd->vdev_mg = svd->vdev_mg;
632 tvd->vdev_ms = svd->vdev_ms;
637 if (tvd->vdev_mg != NULL)
638 tvd->vdev_mg->mg_vd = tvd;
640 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
641 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
642 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
644 svd->vdev_stat.vs_alloc = 0;
645 svd->vdev_stat.vs_space = 0;
646 svd->vdev_stat.vs_dspace = 0;
648 for (t = 0; t < TXG_SIZE; t++) {
649 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
650 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
651 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
652 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
653 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
654 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
657 if (list_link_active(&svd->vdev_config_dirty_node)) {
658 vdev_config_clean(svd);
659 vdev_config_dirty(tvd);
662 if (list_link_active(&svd->vdev_state_dirty_node)) {
663 vdev_state_clean(svd);
664 vdev_state_dirty(tvd);
667 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
668 svd->vdev_deflate_ratio = 0;
670 tvd->vdev_islog = svd->vdev_islog;
675 vdev_top_update(vdev_t *tvd, vdev_t *vd)
682 for (int c = 0; c < vd->vdev_children; c++)
683 vdev_top_update(tvd, vd->vdev_child[c]);
687 * Add a mirror/replacing vdev above an existing vdev.
690 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
692 spa_t *spa = cvd->vdev_spa;
693 vdev_t *pvd = cvd->vdev_parent;
696 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
698 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
700 mvd->vdev_asize = cvd->vdev_asize;
701 mvd->vdev_min_asize = cvd->vdev_min_asize;
702 mvd->vdev_ashift = cvd->vdev_ashift;
703 mvd->vdev_state = cvd->vdev_state;
705 vdev_remove_child(pvd, cvd);
706 vdev_add_child(pvd, mvd);
707 cvd->vdev_id = mvd->vdev_children;
708 vdev_add_child(mvd, cvd);
709 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
711 if (mvd == mvd->vdev_top)
712 vdev_top_transfer(cvd, mvd);
718 * Remove a 1-way mirror/replacing vdev from the tree.
721 vdev_remove_parent(vdev_t *cvd)
723 vdev_t *mvd = cvd->vdev_parent;
724 vdev_t *pvd = mvd->vdev_parent;
726 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
728 ASSERT(mvd->vdev_children == 1);
729 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
730 mvd->vdev_ops == &vdev_replacing_ops ||
731 mvd->vdev_ops == &vdev_spare_ops);
732 cvd->vdev_ashift = mvd->vdev_ashift;
734 vdev_remove_child(mvd, cvd);
735 vdev_remove_child(pvd, mvd);
738 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
739 * Otherwise, we could have detached an offline device, and when we
740 * go to import the pool we'll think we have two top-level vdevs,
741 * instead of a different version of the same top-level vdev.
743 if (mvd->vdev_top == mvd) {
744 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
745 cvd->vdev_guid += guid_delta;
746 cvd->vdev_guid_sum += guid_delta;
748 cvd->vdev_id = mvd->vdev_id;
749 vdev_add_child(pvd, cvd);
750 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
752 if (cvd == cvd->vdev_top)
753 vdev_top_transfer(mvd, cvd);
755 ASSERT(mvd->vdev_children == 0);
760 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
762 spa_t *spa = vd->vdev_spa;
763 objset_t *mos = spa->spa_meta_objset;
764 metaslab_class_t *mc;
766 uint64_t oldc = vd->vdev_ms_count;
767 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
771 if (vd->vdev_ms_shift == 0) /* not being allocated from yet */
775 * Compute the raidz-deflation ratio. Note, we hard-code
776 * in 128k (1 << 17) because it is the current "typical" blocksize.
777 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
778 * or we will inconsistently account for existing bp's.
780 vd->vdev_deflate_ratio = (1 << 17) /
781 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
783 ASSERT(oldc <= newc);
786 mc = spa->spa_log_class;
788 mc = spa->spa_normal_class;
790 if (vd->vdev_mg == NULL)
791 vd->vdev_mg = metaslab_group_create(mc, vd);
793 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
796 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
797 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
801 vd->vdev_ms_count = newc;
803 for (m = oldc; m < newc; m++) {
804 space_map_obj_t smo = { 0, 0, 0 };
807 error = dmu_read(mos, vd->vdev_ms_array,
808 m * sizeof (uint64_t), sizeof (uint64_t), &object,
814 error = dmu_bonus_hold(mos, object, FTAG, &db);
817 ASSERT3U(db->db_size, >=, sizeof (smo));
818 bcopy(db->db_data, &smo, sizeof (smo));
819 ASSERT3U(smo.smo_object, ==, object);
820 dmu_buf_rele(db, FTAG);
823 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
824 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
831 vdev_metaslab_fini(vdev_t *vd)
834 uint64_t count = vd->vdev_ms_count;
836 if (vd->vdev_ms != NULL) {
837 for (m = 0; m < count; m++)
838 if (vd->vdev_ms[m] != NULL)
839 metaslab_fini(vd->vdev_ms[m]);
840 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
845 typedef struct vdev_probe_stats {
846 boolean_t vps_readable;
847 boolean_t vps_writeable;
849 } vdev_probe_stats_t;
852 vdev_probe_done(zio_t *zio)
854 spa_t *spa = zio->io_spa;
855 vdev_t *vd = zio->io_vd;
856 vdev_probe_stats_t *vps = zio->io_private;
858 ASSERT(vd->vdev_probe_zio != NULL);
860 if (zio->io_type == ZIO_TYPE_READ) {
861 if (zio->io_error == 0)
862 vps->vps_readable = 1;
863 if (zio->io_error == 0 && spa_writeable(spa)) {
864 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
865 zio->io_offset, zio->io_size, zio->io_data,
866 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
867 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
869 zio_buf_free(zio->io_data, zio->io_size);
871 } else if (zio->io_type == ZIO_TYPE_WRITE) {
872 if (zio->io_error == 0)
873 vps->vps_writeable = 1;
874 zio_buf_free(zio->io_data, zio->io_size);
875 } else if (zio->io_type == ZIO_TYPE_NULL) {
878 vd->vdev_cant_read |= !vps->vps_readable;
879 vd->vdev_cant_write |= !vps->vps_writeable;
881 if (vdev_readable(vd) &&
882 (vdev_writeable(vd) || !spa_writeable(spa))) {
885 ASSERT(zio->io_error != 0);
886 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
887 spa, vd, NULL, 0, 0);
888 zio->io_error = ENXIO;
891 mutex_enter(&vd->vdev_probe_lock);
892 ASSERT(vd->vdev_probe_zio == zio);
893 vd->vdev_probe_zio = NULL;
894 mutex_exit(&vd->vdev_probe_lock);
896 while ((pio = zio_walk_parents(zio)) != NULL)
897 if (!vdev_accessible(vd, pio))
898 pio->io_error = ENXIO;
900 kmem_free(vps, sizeof (*vps));
905 * Determine whether this device is accessible by reading and writing
906 * to several known locations: the pad regions of each vdev label
907 * but the first (which we leave alone in case it contains a VTOC).
910 vdev_probe(vdev_t *vd, zio_t *zio)
912 spa_t *spa = vd->vdev_spa;
913 vdev_probe_stats_t *vps = NULL;
916 ASSERT(vd->vdev_ops->vdev_op_leaf);
919 * Don't probe the probe.
921 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
925 * To prevent 'probe storms' when a device fails, we create
926 * just one probe i/o at a time. All zios that want to probe
927 * this vdev will become parents of the probe io.
929 mutex_enter(&vd->vdev_probe_lock);
931 if ((pio = vd->vdev_probe_zio) == NULL) {
932 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
934 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
935 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
938 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
940 * vdev_cant_read and vdev_cant_write can only
941 * transition from TRUE to FALSE when we have the
942 * SCL_ZIO lock as writer; otherwise they can only
943 * transition from FALSE to TRUE. This ensures that
944 * any zio looking at these values can assume that
945 * failures persist for the life of the I/O. That's
946 * important because when a device has intermittent
947 * connectivity problems, we want to ensure that
948 * they're ascribed to the device (ENXIO) and not
951 * Since we hold SCL_ZIO as writer here, clear both
952 * values so the probe can reevaluate from first
955 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
956 vd->vdev_cant_read = B_FALSE;
957 vd->vdev_cant_write = B_FALSE;
960 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
961 vdev_probe_done, vps,
962 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
965 vd->vdev_probe_wanted = B_TRUE;
966 spa_async_request(spa, SPA_ASYNC_PROBE);
971 zio_add_child(zio, pio);
973 mutex_exit(&vd->vdev_probe_lock);
980 for (int l = 1; l < VDEV_LABELS; l++) {
981 zio_nowait(zio_read_phys(pio, vd,
982 vdev_label_offset(vd->vdev_psize, l,
983 offsetof(vdev_label_t, vl_pad2)),
984 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
985 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
986 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
997 * Prepare a virtual device for access.
1000 vdev_open(vdev_t *vd)
1002 spa_t *spa = vd->vdev_spa;
1005 uint64_t asize, psize;
1006 uint64_t ashift = 0;
1008 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1010 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1011 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1012 vd->vdev_state == VDEV_STATE_OFFLINE);
1014 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1015 vd->vdev_cant_read = B_FALSE;
1016 vd->vdev_cant_write = B_FALSE;
1017 vd->vdev_min_asize = vdev_get_min_asize(vd);
1019 if (!vd->vdev_removed && vd->vdev_faulted) {
1020 ASSERT(vd->vdev_children == 0);
1021 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1022 VDEV_AUX_ERR_EXCEEDED);
1024 } else if (vd->vdev_offline) {
1025 ASSERT(vd->vdev_children == 0);
1026 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1030 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1032 if (zio_injection_enabled && error == 0)
1033 error = zio_handle_device_injection(vd, NULL, ENXIO);
1036 if (vd->vdev_removed &&
1037 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1038 vd->vdev_removed = B_FALSE;
1040 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1041 vd->vdev_stat.vs_aux);
1045 vd->vdev_removed = B_FALSE;
1047 if (vd->vdev_degraded) {
1048 ASSERT(vd->vdev_children == 0);
1049 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1050 VDEV_AUX_ERR_EXCEEDED);
1052 vd->vdev_state = VDEV_STATE_HEALTHY;
1055 for (int c = 0; c < vd->vdev_children; c++) {
1056 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1057 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1063 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1065 if (vd->vdev_children == 0) {
1066 if (osize < SPA_MINDEVSIZE) {
1067 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1068 VDEV_AUX_TOO_SMALL);
1072 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1074 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1075 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1076 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1077 VDEV_AUX_TOO_SMALL);
1084 vd->vdev_psize = psize;
1087 * Make sure the allocatable size hasn't shrunk.
1089 if (asize < vd->vdev_min_asize) {
1090 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1091 VDEV_AUX_BAD_LABEL);
1095 if (vd->vdev_asize == 0) {
1097 * This is the first-ever open, so use the computed values.
1098 * For testing purposes, a higher ashift can be requested.
1100 vd->vdev_asize = asize;
1101 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1104 * Make sure the alignment requirement hasn't increased.
1106 if (ashift > vd->vdev_top->vdev_ashift) {
1107 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1108 VDEV_AUX_BAD_LABEL);
1114 * If all children are healthy and the asize has increased,
1115 * then we've experienced dynamic LUN growth. If automatic
1116 * expansion is enabled then use the additional space.
1118 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1119 (vd->vdev_expanding || spa->spa_autoexpand))
1120 vd->vdev_asize = asize;
1122 vdev_set_min_asize(vd);
1125 * Ensure we can issue some IO before declaring the
1126 * vdev open for business.
1128 if (vd->vdev_ops->vdev_op_leaf &&
1129 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1130 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1131 VDEV_AUX_IO_FAILURE);
1136 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1137 * resilver. But don't do this if we are doing a reopen for a scrub,
1138 * since this would just restart the scrub we are already doing.
1140 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1141 vdev_resilver_needed(vd, NULL, NULL))
1142 spa_async_request(spa, SPA_ASYNC_RESILVER);
1148 * Called once the vdevs are all opened, this routine validates the label
1149 * contents. This needs to be done before vdev_load() so that we don't
1150 * inadvertently do repair I/Os to the wrong device.
1152 * This function will only return failure if one of the vdevs indicates that it
1153 * has since been destroyed or exported. This is only possible if
1154 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1155 * will be updated but the function will return 0.
1158 vdev_validate(vdev_t *vd)
1160 spa_t *spa = vd->vdev_spa;
1162 uint64_t guid, top_guid;
1165 for (int c = 0; c < vd->vdev_children; c++)
1166 if (vdev_validate(vd->vdev_child[c]) != 0)
1170 * If the device has already failed, or was marked offline, don't do
1171 * any further validation. Otherwise, label I/O will fail and we will
1172 * overwrite the previous state.
1174 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1176 if ((label = vdev_label_read_config(vd)) == NULL) {
1177 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1178 VDEV_AUX_BAD_LABEL);
1182 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1183 &guid) != 0 || guid != spa_guid(spa)) {
1184 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1185 VDEV_AUX_CORRUPT_DATA);
1191 * If this vdev just became a top-level vdev because its
1192 * sibling was detached, it will have adopted the parent's
1193 * vdev guid -- but the label may or may not be on disk yet.
1194 * Fortunately, either version of the label will have the
1195 * same top guid, so if we're a top-level vdev, we can
1196 * safely compare to that instead.
1198 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1200 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1202 (vd->vdev_guid != guid &&
1203 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1204 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1205 VDEV_AUX_CORRUPT_DATA);
1210 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1212 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1213 VDEV_AUX_CORRUPT_DATA);
1220 if (spa->spa_load_state == SPA_LOAD_OPEN &&
1221 state != POOL_STATE_ACTIVE)
1225 * If we were able to open and validate a vdev that was
1226 * previously marked permanently unavailable, clear that state
1229 if (vd->vdev_not_present)
1230 vd->vdev_not_present = 0;
1237 * Close a virtual device.
1240 vdev_close(vdev_t *vd)
1242 spa_t *spa = vd->vdev_spa;
1244 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1246 vd->vdev_ops->vdev_op_close(vd);
1248 vdev_cache_purge(vd);
1251 * We record the previous state before we close it, so that if we are
1252 * doing a reopen(), we don't generate FMA ereports if we notice that
1253 * it's still faulted.
1255 vd->vdev_prevstate = vd->vdev_state;
1257 if (vd->vdev_offline)
1258 vd->vdev_state = VDEV_STATE_OFFLINE;
1260 vd->vdev_state = VDEV_STATE_CLOSED;
1261 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1265 vdev_reopen(vdev_t *vd)
1267 spa_t *spa = vd->vdev_spa;
1269 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1272 (void) vdev_open(vd);
1275 * Call vdev_validate() here to make sure we have the same device.
1276 * Otherwise, a device with an invalid label could be successfully
1277 * opened in response to vdev_reopen().
1280 (void) vdev_validate_aux(vd);
1281 if (vdev_readable(vd) && vdev_writeable(vd) &&
1282 vd->vdev_aux == &spa->spa_l2cache &&
1283 !l2arc_vdev_present(vd))
1284 l2arc_add_vdev(spa, vd);
1286 (void) vdev_validate(vd);
1290 * Reassess parent vdev's health.
1292 vdev_propagate_state(vd);
1296 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1301 * Normally, partial opens (e.g. of a mirror) are allowed.
1302 * For a create, however, we want to fail the request if
1303 * there are any components we can't open.
1305 error = vdev_open(vd);
1307 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1309 return (error ? error : ENXIO);
1313 * Recursively initialize all labels.
1315 if ((error = vdev_label_init(vd, txg, isreplacing ?
1316 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1325 vdev_metaslab_set_size(vdev_t *vd)
1328 * Aim for roughly 200 metaslabs per vdev.
1330 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1331 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1335 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1337 ASSERT(vd == vd->vdev_top);
1338 ASSERT(ISP2(flags));
1340 if (flags & VDD_METASLAB)
1341 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1343 if (flags & VDD_DTL)
1344 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1346 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1352 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1353 * the vdev has less than perfect replication. There are three kinds of DTL:
1355 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1357 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1359 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1360 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1361 * txgs that was scrubbed.
1363 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1364 * persistent errors or just some device being offline.
1365 * Unlike the other three, the DTL_OUTAGE map is not generally
1366 * maintained; it's only computed when needed, typically to
1367 * determine whether a device can be detached.
1369 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1370 * either has the data or it doesn't.
1372 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1373 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1374 * if any child is less than fully replicated, then so is its parent.
1375 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1376 * comprising only those txgs which appear in 'maxfaults' or more children;
1377 * those are the txgs we don't have enough replication to read. For example,
1378 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1379 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1380 * two child DTL_MISSING maps.
1382 * It should be clear from the above that to compute the DTLs and outage maps
1383 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1384 * Therefore, that is all we keep on disk. When loading the pool, or after
1385 * a configuration change, we generate all other DTLs from first principles.
1388 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1390 space_map_t *sm = &vd->vdev_dtl[t];
1392 ASSERT(t < DTL_TYPES);
1393 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1395 mutex_enter(sm->sm_lock);
1396 if (!space_map_contains(sm, txg, size))
1397 space_map_add(sm, txg, size);
1398 mutex_exit(sm->sm_lock);
1402 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1404 space_map_t *sm = &vd->vdev_dtl[t];
1405 boolean_t dirty = B_FALSE;
1407 ASSERT(t < DTL_TYPES);
1408 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1410 mutex_enter(sm->sm_lock);
1411 if (sm->sm_space != 0)
1412 dirty = space_map_contains(sm, txg, size);
1413 mutex_exit(sm->sm_lock);
1419 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1421 space_map_t *sm = &vd->vdev_dtl[t];
1424 mutex_enter(sm->sm_lock);
1425 empty = (sm->sm_space == 0);
1426 mutex_exit(sm->sm_lock);
1432 * Reassess DTLs after a config change or scrub completion.
1435 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1437 spa_t *spa = vd->vdev_spa;
1441 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1443 for (int c = 0; c < vd->vdev_children; c++)
1444 vdev_dtl_reassess(vd->vdev_child[c], txg,
1445 scrub_txg, scrub_done);
1447 if (vd == spa->spa_root_vdev)
1450 if (vd->vdev_ops->vdev_op_leaf) {
1451 mutex_enter(&vd->vdev_dtl_lock);
1452 if (scrub_txg != 0 &&
1453 (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) {
1454 /* XXX should check scrub_done? */
1456 * We completed a scrub up to scrub_txg. If we
1457 * did it without rebooting, then the scrub dtl
1458 * will be valid, so excise the old region and
1459 * fold in the scrub dtl. Otherwise, leave the
1460 * dtl as-is if there was an error.
1462 * There's little trick here: to excise the beginning
1463 * of the DTL_MISSING map, we put it into a reference
1464 * tree and then add a segment with refcnt -1 that
1465 * covers the range [0, scrub_txg). This means
1466 * that each txg in that range has refcnt -1 or 0.
1467 * We then add DTL_SCRUB with a refcnt of 2, so that
1468 * entries in the range [0, scrub_txg) will have a
1469 * positive refcnt -- either 1 or 2. We then convert
1470 * the reference tree into the new DTL_MISSING map.
1472 space_map_ref_create(&reftree);
1473 space_map_ref_add_map(&reftree,
1474 &vd->vdev_dtl[DTL_MISSING], 1);
1475 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1476 space_map_ref_add_map(&reftree,
1477 &vd->vdev_dtl[DTL_SCRUB], 2);
1478 space_map_ref_generate_map(&reftree,
1479 &vd->vdev_dtl[DTL_MISSING], 1);
1480 space_map_ref_destroy(&reftree);
1482 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1483 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1484 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1486 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1487 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1488 if (!vdev_readable(vd))
1489 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1491 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1492 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1493 mutex_exit(&vd->vdev_dtl_lock);
1496 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1500 mutex_enter(&vd->vdev_dtl_lock);
1501 for (int t = 0; t < DTL_TYPES; t++) {
1503 continue; /* leaf vdevs only */
1504 if (t == DTL_PARTIAL)
1505 minref = 1; /* i.e. non-zero */
1506 else if (vd->vdev_nparity != 0)
1507 minref = vd->vdev_nparity + 1; /* RAID-Z */
1509 minref = vd->vdev_children; /* any kind of mirror */
1510 space_map_ref_create(&reftree);
1511 for (int c = 0; c < vd->vdev_children; c++) {
1512 vdev_t *cvd = vd->vdev_child[c];
1513 mutex_enter(&cvd->vdev_dtl_lock);
1514 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[t], 1);
1515 mutex_exit(&cvd->vdev_dtl_lock);
1517 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1518 space_map_ref_destroy(&reftree);
1520 mutex_exit(&vd->vdev_dtl_lock);
1524 vdev_dtl_load(vdev_t *vd)
1526 spa_t *spa = vd->vdev_spa;
1527 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1528 objset_t *mos = spa->spa_meta_objset;
1532 ASSERT(vd->vdev_children == 0);
1534 if (smo->smo_object == 0)
1537 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1540 ASSERT3U(db->db_size, >=, sizeof (*smo));
1541 bcopy(db->db_data, smo, sizeof (*smo));
1542 dmu_buf_rele(db, FTAG);
1544 mutex_enter(&vd->vdev_dtl_lock);
1545 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1546 NULL, SM_ALLOC, smo, mos);
1547 mutex_exit(&vd->vdev_dtl_lock);
1553 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1555 spa_t *spa = vd->vdev_spa;
1556 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1557 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1558 objset_t *mos = spa->spa_meta_objset;
1564 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1566 if (vd->vdev_detached) {
1567 if (smo->smo_object != 0) {
1568 int err = dmu_object_free(mos, smo->smo_object, tx);
1569 ASSERT3U(err, ==, 0);
1570 smo->smo_object = 0;
1576 if (smo->smo_object == 0) {
1577 ASSERT(smo->smo_objsize == 0);
1578 ASSERT(smo->smo_alloc == 0);
1579 smo->smo_object = dmu_object_alloc(mos,
1580 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1581 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1582 ASSERT(smo->smo_object != 0);
1583 vdev_config_dirty(vd->vdev_top);
1586 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1588 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1591 mutex_enter(&smlock);
1593 mutex_enter(&vd->vdev_dtl_lock);
1594 space_map_walk(sm, space_map_add, &smsync);
1595 mutex_exit(&vd->vdev_dtl_lock);
1597 space_map_truncate(smo, mos, tx);
1598 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1600 space_map_destroy(&smsync);
1602 mutex_exit(&smlock);
1603 mutex_destroy(&smlock);
1605 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1606 dmu_buf_will_dirty(db, tx);
1607 ASSERT3U(db->db_size, >=, sizeof (*smo));
1608 bcopy(smo, db->db_data, sizeof (*smo));
1609 dmu_buf_rele(db, FTAG);
1615 * Determine whether the specified vdev can be offlined/detached/removed
1616 * without losing data.
1619 vdev_dtl_required(vdev_t *vd)
1621 spa_t *spa = vd->vdev_spa;
1622 vdev_t *tvd = vd->vdev_top;
1623 uint8_t cant_read = vd->vdev_cant_read;
1626 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1628 if (vd == spa->spa_root_vdev || vd == tvd)
1632 * Temporarily mark the device as unreadable, and then determine
1633 * whether this results in any DTL outages in the top-level vdev.
1634 * If not, we can safely offline/detach/remove the device.
1636 vd->vdev_cant_read = B_TRUE;
1637 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1638 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1639 vd->vdev_cant_read = cant_read;
1640 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1646 * Determine if resilver is needed, and if so the txg range.
1649 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1651 boolean_t needed = B_FALSE;
1652 uint64_t thismin = UINT64_MAX;
1653 uint64_t thismax = 0;
1655 if (vd->vdev_children == 0) {
1656 mutex_enter(&vd->vdev_dtl_lock);
1657 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1658 vdev_writeable(vd)) {
1661 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1662 thismin = ss->ss_start - 1;
1663 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1664 thismax = ss->ss_end;
1667 mutex_exit(&vd->vdev_dtl_lock);
1669 for (int c = 0; c < vd->vdev_children; c++) {
1670 vdev_t *cvd = vd->vdev_child[c];
1671 uint64_t cmin, cmax;
1673 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1674 thismin = MIN(thismin, cmin);
1675 thismax = MAX(thismax, cmax);
1681 if (needed && minp) {
1689 vdev_load(vdev_t *vd)
1692 * Recursively load all children.
1694 for (int c = 0; c < vd->vdev_children; c++)
1695 vdev_load(vd->vdev_child[c]);
1698 * If this is a top-level vdev, initialize its metaslabs.
1700 if (vd == vd->vdev_top &&
1701 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1702 vdev_metaslab_init(vd, 0) != 0))
1703 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1704 VDEV_AUX_CORRUPT_DATA);
1707 * If this is a leaf vdev, load its DTL.
1709 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1710 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1711 VDEV_AUX_CORRUPT_DATA);
1715 * The special vdev case is used for hot spares and l2cache devices. Its
1716 * sole purpose it to set the vdev state for the associated vdev. To do this,
1717 * we make sure that we can open the underlying device, then try to read the
1718 * label, and make sure that the label is sane and that it hasn't been
1719 * repurposed to another pool.
1722 vdev_validate_aux(vdev_t *vd)
1725 uint64_t guid, version;
1728 if (!vdev_readable(vd))
1731 if ((label = vdev_label_read_config(vd)) == NULL) {
1732 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1733 VDEV_AUX_CORRUPT_DATA);
1737 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1738 version > SPA_VERSION ||
1739 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1740 guid != vd->vdev_guid ||
1741 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1742 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1743 VDEV_AUX_CORRUPT_DATA);
1749 * We don't actually check the pool state here. If it's in fact in
1750 * use by another pool, we update this fact on the fly when requested.
1757 vdev_sync_done(vdev_t *vd, uint64_t txg)
1761 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
1762 metaslab_sync_done(msp, txg);
1766 vdev_sync(vdev_t *vd, uint64_t txg)
1768 spa_t *spa = vd->vdev_spa;
1773 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
1774 ASSERT(vd == vd->vdev_top);
1775 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1776 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
1777 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
1778 ASSERT(vd->vdev_ms_array != 0);
1779 vdev_config_dirty(vd);
1783 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
1784 metaslab_sync(msp, txg);
1785 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
1788 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
1789 vdev_dtl_sync(lvd, txg);
1791 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
1795 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
1797 return (vd->vdev_ops->vdev_op_asize(vd, psize));
1801 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
1802 * not be opened, and no I/O is attempted.
1805 vdev_fault(spa_t *spa, uint64_t guid)
1809 spa_vdev_state_enter(spa);
1811 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1812 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1814 if (!vd->vdev_ops->vdev_op_leaf)
1815 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1818 * Faulted state takes precedence over degraded.
1820 vd->vdev_faulted = 1ULL;
1821 vd->vdev_degraded = 0ULL;
1822 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED);
1825 * If marking the vdev as faulted cause the top-level vdev to become
1826 * unavailable, then back off and simply mark the vdev as degraded
1829 if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) {
1830 vd->vdev_degraded = 1ULL;
1831 vd->vdev_faulted = 0ULL;
1834 * If we reopen the device and it's not dead, only then do we
1839 if (vdev_readable(vd)) {
1840 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1841 VDEV_AUX_ERR_EXCEEDED);
1845 return (spa_vdev_state_exit(spa, vd, 0));
1849 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
1850 * user that something is wrong. The vdev continues to operate as normal as far
1851 * as I/O is concerned.
1854 vdev_degrade(spa_t *spa, uint64_t guid)
1858 spa_vdev_state_enter(spa);
1860 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1861 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1863 if (!vd->vdev_ops->vdev_op_leaf)
1864 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1867 * If the vdev is already faulted, then don't do anything.
1869 if (vd->vdev_faulted || vd->vdev_degraded)
1870 return (spa_vdev_state_exit(spa, NULL, 0));
1872 vd->vdev_degraded = 1ULL;
1873 if (!vdev_is_dead(vd))
1874 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1875 VDEV_AUX_ERR_EXCEEDED);
1877 return (spa_vdev_state_exit(spa, vd, 0));
1881 * Online the given vdev. If 'unspare' is set, it implies two things. First,
1882 * any attached spare device should be detached when the device finishes
1883 * resilvering. Second, the online should be treated like a 'test' online case,
1884 * so no FMA events are generated if the device fails to open.
1887 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
1889 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
1891 spa_vdev_state_enter(spa);
1893 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1894 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1896 if (!vd->vdev_ops->vdev_op_leaf)
1897 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1900 vd->vdev_offline = B_FALSE;
1901 vd->vdev_tmpoffline = B_FALSE;
1902 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
1903 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
1905 /* XXX - L2ARC 1.0 does not support expansion */
1906 if (!vd->vdev_aux) {
1907 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
1908 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
1912 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
1914 if (!vd->vdev_aux) {
1915 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
1916 pvd->vdev_expanding = B_FALSE;
1920 *newstate = vd->vdev_state;
1921 if ((flags & ZFS_ONLINE_UNSPARE) &&
1922 !vdev_is_dead(vd) && vd->vdev_parent &&
1923 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
1924 vd->vdev_parent->vdev_child[0] == vd)
1925 vd->vdev_unspare = B_TRUE;
1927 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
1929 /* XXX - L2ARC 1.0 does not support expansion */
1931 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
1932 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
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_min_asize(vd);
2118 if (vd->vdev_ops->vdev_op_leaf)
2119 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2120 mutex_exit(&vd->vdev_stat_lock);
2123 * If we're getting stats on the root vdev, aggregate the I/O counts
2124 * over all top-level vdevs (i.e. the direct children of the root).
2127 for (int c = 0; c < rvd->vdev_children; c++) {
2128 vdev_t *cvd = rvd->vdev_child[c];
2129 vdev_stat_t *cvs = &cvd->vdev_stat;
2131 mutex_enter(&vd->vdev_stat_lock);
2132 for (int t = 0; t < ZIO_TYPES; t++) {
2133 vs->vs_ops[t] += cvs->vs_ops[t];
2134 vs->vs_bytes[t] += cvs->vs_bytes[t];
2136 vs->vs_scrub_examined += cvs->vs_scrub_examined;
2137 mutex_exit(&vd->vdev_stat_lock);
2143 vdev_clear_stats(vdev_t *vd)
2145 mutex_enter(&vd->vdev_stat_lock);
2146 vd->vdev_stat.vs_space = 0;
2147 vd->vdev_stat.vs_dspace = 0;
2148 vd->vdev_stat.vs_alloc = 0;
2149 mutex_exit(&vd->vdev_stat_lock);
2153 vdev_stat_update(zio_t *zio, uint64_t psize)
2155 spa_t *spa = zio->io_spa;
2156 vdev_t *rvd = spa->spa_root_vdev;
2157 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2159 uint64_t txg = zio->io_txg;
2160 vdev_stat_t *vs = &vd->vdev_stat;
2161 zio_type_t type = zio->io_type;
2162 int flags = zio->io_flags;
2165 * If this i/o is a gang leader, it didn't do any actual work.
2167 if (zio->io_gang_tree)
2170 if (zio->io_error == 0) {
2172 * If this is a root i/o, don't count it -- we've already
2173 * counted the top-level vdevs, and vdev_get_stats() will
2174 * aggregate them when asked. This reduces contention on
2175 * the root vdev_stat_lock and implicitly handles blocks
2176 * that compress away to holes, for which there is no i/o.
2177 * (Holes never create vdev children, so all the counters
2178 * remain zero, which is what we want.)
2180 * Note: this only applies to successful i/o (io_error == 0)
2181 * because unlike i/o counts, errors are not additive.
2182 * When reading a ditto block, for example, failure of
2183 * one top-level vdev does not imply a root-level error.
2188 ASSERT(vd == zio->io_vd);
2190 if (flags & ZIO_FLAG_IO_BYPASS)
2193 mutex_enter(&vd->vdev_stat_lock);
2195 if (flags & ZIO_FLAG_IO_REPAIR) {
2196 if (flags & ZIO_FLAG_SCRUB_THREAD)
2197 vs->vs_scrub_repaired += psize;
2198 if (flags & ZIO_FLAG_SELF_HEAL)
2199 vs->vs_self_healed += psize;
2203 vs->vs_bytes[type] += psize;
2205 mutex_exit(&vd->vdev_stat_lock);
2209 if (flags & ZIO_FLAG_SPECULATIVE)
2213 * If this is an I/O error that is going to be retried, then ignore the
2214 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2215 * hard errors, when in reality they can happen for any number of
2216 * innocuous reasons (bus resets, MPxIO link failure, etc).
2218 if (zio->io_error == EIO &&
2219 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2222 mutex_enter(&vd->vdev_stat_lock);
2223 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2224 if (zio->io_error == ECKSUM)
2225 vs->vs_checksum_errors++;
2227 vs->vs_read_errors++;
2229 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2230 vs->vs_write_errors++;
2231 mutex_exit(&vd->vdev_stat_lock);
2233 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2234 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2235 (flags & ZIO_FLAG_SCRUB_THREAD))) {
2237 * This is either a normal write (not a repair), or it's a
2238 * repair induced by the scrub thread. In the normal case,
2239 * we commit the DTL change in the same txg as the block
2240 * was born. In the scrub-induced repair case, we know that
2241 * scrubs run in first-pass syncing context, so we commit
2242 * the DTL change in spa->spa_syncing_txg.
2244 * We currently do not make DTL entries for failed spontaneous
2245 * self-healing writes triggered by normal (non-scrubbing)
2246 * reads, because we have no transactional context in which to
2247 * do so -- and it's not clear that it'd be desirable anyway.
2249 if (vd->vdev_ops->vdev_op_leaf) {
2250 uint64_t commit_txg = txg;
2251 if (flags & ZIO_FLAG_SCRUB_THREAD) {
2252 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2253 ASSERT(spa_sync_pass(spa) == 1);
2254 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2255 commit_txg = spa->spa_syncing_txg;
2257 ASSERT(commit_txg >= spa->spa_syncing_txg);
2258 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2260 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2261 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2262 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2265 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2270 vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete)
2272 vdev_stat_t *vs = &vd->vdev_stat;
2274 for (int c = 0; c < vd->vdev_children; c++)
2275 vdev_scrub_stat_update(vd->vdev_child[c], type, complete);
2277 mutex_enter(&vd->vdev_stat_lock);
2279 if (type == POOL_SCRUB_NONE) {
2281 * Update completion and end time. Leave everything else alone
2282 * so we can report what happened during the previous scrub.
2284 vs->vs_scrub_complete = complete;
2285 vs->vs_scrub_end = gethrestime_sec();
2287 vs->vs_scrub_type = type;
2288 vs->vs_scrub_complete = 0;
2289 vs->vs_scrub_examined = 0;
2290 vs->vs_scrub_repaired = 0;
2291 vs->vs_scrub_start = gethrestime_sec();
2292 vs->vs_scrub_end = 0;
2295 mutex_exit(&vd->vdev_stat_lock);
2299 * Update the in-core space usage stats for this vdev and the root vdev.
2302 vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta,
2303 boolean_t update_root)
2305 int64_t dspace_delta = space_delta;
2306 spa_t *spa = vd->vdev_spa;
2307 vdev_t *rvd = spa->spa_root_vdev;
2309 ASSERT(vd == vd->vdev_top);
2312 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2313 * factor. We must calculate this here and not at the root vdev
2314 * because the root vdev's psize-to-asize is simply the max of its
2315 * childrens', thus not accurate enough for us.
2317 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2318 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2319 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2320 vd->vdev_deflate_ratio;
2322 mutex_enter(&vd->vdev_stat_lock);
2323 vd->vdev_stat.vs_space += space_delta;
2324 vd->vdev_stat.vs_alloc += alloc_delta;
2325 vd->vdev_stat.vs_dspace += dspace_delta;
2326 mutex_exit(&vd->vdev_stat_lock);
2329 ASSERT(rvd == vd->vdev_parent);
2330 ASSERT(vd->vdev_ms_count != 0);
2333 * Don't count non-normal (e.g. intent log) space as part of
2334 * the pool's capacity.
2336 if (vd->vdev_mg->mg_class != spa->spa_normal_class)
2339 mutex_enter(&rvd->vdev_stat_lock);
2340 rvd->vdev_stat.vs_space += space_delta;
2341 rvd->vdev_stat.vs_alloc += alloc_delta;
2342 rvd->vdev_stat.vs_dspace += dspace_delta;
2343 mutex_exit(&rvd->vdev_stat_lock);
2348 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2349 * so that it will be written out next time the vdev configuration is synced.
2350 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2353 vdev_config_dirty(vdev_t *vd)
2355 spa_t *spa = vd->vdev_spa;
2356 vdev_t *rvd = spa->spa_root_vdev;
2360 * If this is an aux vdev (as with l2cache and spare devices), then we
2361 * update the vdev config manually and set the sync flag.
2363 if (vd->vdev_aux != NULL) {
2364 spa_aux_vdev_t *sav = vd->vdev_aux;
2368 for (c = 0; c < sav->sav_count; c++) {
2369 if (sav->sav_vdevs[c] == vd)
2373 if (c == sav->sav_count) {
2375 * We're being removed. There's nothing more to do.
2377 ASSERT(sav->sav_sync == B_TRUE);
2381 sav->sav_sync = B_TRUE;
2383 if (nvlist_lookup_nvlist_array(sav->sav_config,
2384 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2385 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2386 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2392 * Setting the nvlist in the middle if the array is a little
2393 * sketchy, but it will work.
2395 nvlist_free(aux[c]);
2396 aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE);
2402 * The dirty list is protected by the SCL_CONFIG lock. The caller
2403 * must either hold SCL_CONFIG as writer, or must be the sync thread
2404 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2405 * so this is sufficient to ensure mutual exclusion.
2407 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2408 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2409 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2412 for (c = 0; c < rvd->vdev_children; c++)
2413 vdev_config_dirty(rvd->vdev_child[c]);
2415 ASSERT(vd == vd->vdev_top);
2417 if (!list_link_active(&vd->vdev_config_dirty_node))
2418 list_insert_head(&spa->spa_config_dirty_list, vd);
2423 vdev_config_clean(vdev_t *vd)
2425 spa_t *spa = vd->vdev_spa;
2427 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2428 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2429 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2431 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2432 list_remove(&spa->spa_config_dirty_list, vd);
2436 * Mark a top-level vdev's state as dirty, so that the next pass of
2437 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2438 * the state changes from larger config changes because they require
2439 * much less locking, and are often needed for administrative actions.
2442 vdev_state_dirty(vdev_t *vd)
2444 spa_t *spa = vd->vdev_spa;
2446 ASSERT(vd == vd->vdev_top);
2449 * The state list is protected by the SCL_STATE lock. The caller
2450 * must either hold SCL_STATE as writer, or must be the sync thread
2451 * (which holds SCL_STATE as reader). There's only one sync thread,
2452 * so this is sufficient to ensure mutual exclusion.
2454 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2455 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2456 spa_config_held(spa, SCL_STATE, RW_READER)));
2458 if (!list_link_active(&vd->vdev_state_dirty_node))
2459 list_insert_head(&spa->spa_state_dirty_list, vd);
2463 vdev_state_clean(vdev_t *vd)
2465 spa_t *spa = vd->vdev_spa;
2467 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2468 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2469 spa_config_held(spa, SCL_STATE, RW_READER)));
2471 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2472 list_remove(&spa->spa_state_dirty_list, vd);
2476 * Propagate vdev state up from children to parent.
2479 vdev_propagate_state(vdev_t *vd)
2481 spa_t *spa = vd->vdev_spa;
2482 vdev_t *rvd = spa->spa_root_vdev;
2483 int degraded = 0, faulted = 0;
2487 if (vd->vdev_children > 0) {
2488 for (int 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
2656 * a root pool. Currently, we do not support RAID-Z or partial configuration.
2657 * In addition, only a single top-level vdev is allowed and none of the leaves
2658 * can be wholedisks.
2661 vdev_is_bootable(vdev_t *vd)
2663 if (!vd->vdev_ops->vdev_op_leaf) {
2664 char *vdev_type = vd->vdev_ops->vdev_op_type;
2666 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
2667 vd->vdev_children > 1) {
2669 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
2670 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
2673 } else if (vd->vdev_wholedisk == 1) {
2677 for (int c = 0; c < vd->vdev_children; c++) {
2678 if (!vdev_is_bootable(vd->vdev_child[c]))
2685 vdev_load_log_state(vdev_t *vd, nvlist_t *nv)
2690 spa_t *spa = vd->vdev_spa;
2692 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2693 &child, &children) == 0) {
2694 for (int c = 0; c < children; c++)
2695 vdev_load_log_state(vd->vdev_child[c], child[c]);
2698 if (vd->vdev_ops->vdev_op_leaf && nvlist_lookup_uint64(nv,
2699 ZPOOL_CONFIG_OFFLINE, &val) == 0 && val) {
2702 * It would be nice to call vdev_offline()
2703 * directly but the pool isn't fully loaded and
2704 * the txg threads have not been started yet.
2706 spa_config_enter(spa, SCL_STATE_ALL, FTAG, RW_WRITER);
2707 vd->vdev_offline = val;
2708 vdev_reopen(vd->vdev_top);
2709 spa_config_exit(spa, SCL_STATE_ALL, FTAG);
2714 * Expand a vdev if possible.
2717 vdev_expand(vdev_t *vd, uint64_t txg)
2719 ASSERT(vd->vdev_top == vd);
2720 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2722 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
2723 VERIFY(vdev_metaslab_init(vd, txg) == 0);
2724 vdev_config_dirty(vd);