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 3 parity devices.
410 if (nparity == 0 || nparity > 3)
413 * Previous versions could only support 1 or 2 parity
417 spa_version(spa) < SPA_VERSION_RAIDZ2)
420 spa_version(spa) < SPA_VERSION_RAIDZ3)
424 * We require the parity to be specified for SPAs that
425 * support multiple parity levels.
427 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
430 * Otherwise, we default to 1 parity device for RAID-Z.
437 ASSERT(nparity != -1ULL);
439 vd = vdev_alloc_common(spa, id, guid, ops);
441 vd->vdev_islog = islog;
442 vd->vdev_nparity = nparity;
444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
445 vd->vdev_path = spa_strdup(vd->vdev_path);
446 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
447 vd->vdev_devid = spa_strdup(vd->vdev_devid);
448 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
449 &vd->vdev_physpath) == 0)
450 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
451 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
452 vd->vdev_fru = spa_strdup(vd->vdev_fru);
455 * Set the whole_disk property. If it's not specified, leave the value
458 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
459 &vd->vdev_wholedisk) != 0)
460 vd->vdev_wholedisk = -1ULL;
463 * Look for the 'not present' flag. This will only be set if the device
464 * was not present at the time of import.
466 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
467 &vd->vdev_not_present);
470 * Get the alignment requirement.
472 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
475 * If we're a top-level vdev, try to load the allocation parameters.
477 if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) {
478 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
480 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
482 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
487 * If we're a leaf vdev, try to load the DTL object and other state.
489 if (vd->vdev_ops->vdev_op_leaf &&
490 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
491 alloctype == VDEV_ALLOC_ROOTPOOL)) {
492 if (alloctype == VDEV_ALLOC_LOAD) {
493 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
494 &vd->vdev_dtl_smo.smo_object);
495 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
499 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
502 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
503 &spare) == 0 && spare)
507 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
511 * When importing a pool, we want to ignore the persistent fault
512 * state, as the diagnosis made on another system may not be
513 * valid in the current context.
515 if (spa->spa_load_state == SPA_LOAD_OPEN) {
516 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
518 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
520 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
526 * Add ourselves to the parent's list of children.
528 vdev_add_child(parent, vd);
536 vdev_free(vdev_t *vd)
538 spa_t *spa = vd->vdev_spa;
541 * vdev_free() implies closing the vdev first. This is simpler than
542 * trying to ensure complicated semantics for all callers.
546 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
551 for (int c = 0; c < vd->vdev_children; c++)
552 vdev_free(vd->vdev_child[c]);
554 ASSERT(vd->vdev_child == NULL);
555 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
558 * Discard allocation state.
560 if (vd == vd->vdev_top)
561 vdev_metaslab_fini(vd);
563 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
564 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
565 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
568 * Remove this vdev from its parent's child list.
570 vdev_remove_child(vd->vdev_parent, vd);
572 ASSERT(vd->vdev_parent == NULL);
575 * Clean up vdev structure.
581 spa_strfree(vd->vdev_path);
583 spa_strfree(vd->vdev_devid);
584 if (vd->vdev_physpath)
585 spa_strfree(vd->vdev_physpath);
587 spa_strfree(vd->vdev_fru);
589 if (vd->vdev_isspare)
590 spa_spare_remove(vd);
591 if (vd->vdev_isl2cache)
592 spa_l2cache_remove(vd);
594 txg_list_destroy(&vd->vdev_ms_list);
595 txg_list_destroy(&vd->vdev_dtl_list);
597 mutex_enter(&vd->vdev_dtl_lock);
598 for (int t = 0; t < DTL_TYPES; t++) {
599 space_map_unload(&vd->vdev_dtl[t]);
600 space_map_destroy(&vd->vdev_dtl[t]);
602 mutex_exit(&vd->vdev_dtl_lock);
604 mutex_destroy(&vd->vdev_dtl_lock);
605 mutex_destroy(&vd->vdev_stat_lock);
606 mutex_destroy(&vd->vdev_probe_lock);
608 if (vd == spa->spa_root_vdev)
609 spa->spa_root_vdev = NULL;
611 kmem_free(vd, sizeof (vdev_t));
615 * Transfer top-level vdev state from svd to tvd.
618 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
620 spa_t *spa = svd->vdev_spa;
625 ASSERT(tvd == tvd->vdev_top);
627 tvd->vdev_ms_array = svd->vdev_ms_array;
628 tvd->vdev_ms_shift = svd->vdev_ms_shift;
629 tvd->vdev_ms_count = svd->vdev_ms_count;
631 svd->vdev_ms_array = 0;
632 svd->vdev_ms_shift = 0;
633 svd->vdev_ms_count = 0;
635 tvd->vdev_mg = svd->vdev_mg;
636 tvd->vdev_ms = svd->vdev_ms;
641 if (tvd->vdev_mg != NULL)
642 tvd->vdev_mg->mg_vd = tvd;
644 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
645 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
646 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
648 svd->vdev_stat.vs_alloc = 0;
649 svd->vdev_stat.vs_space = 0;
650 svd->vdev_stat.vs_dspace = 0;
652 for (t = 0; t < TXG_SIZE; t++) {
653 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
654 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
655 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
656 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
657 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
658 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
661 if (list_link_active(&svd->vdev_config_dirty_node)) {
662 vdev_config_clean(svd);
663 vdev_config_dirty(tvd);
666 if (list_link_active(&svd->vdev_state_dirty_node)) {
667 vdev_state_clean(svd);
668 vdev_state_dirty(tvd);
671 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
672 svd->vdev_deflate_ratio = 0;
674 tvd->vdev_islog = svd->vdev_islog;
679 vdev_top_update(vdev_t *tvd, vdev_t *vd)
686 for (int c = 0; c < vd->vdev_children; c++)
687 vdev_top_update(tvd, vd->vdev_child[c]);
691 * Add a mirror/replacing vdev above an existing vdev.
694 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
696 spa_t *spa = cvd->vdev_spa;
697 vdev_t *pvd = cvd->vdev_parent;
700 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
702 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
704 mvd->vdev_asize = cvd->vdev_asize;
705 mvd->vdev_min_asize = cvd->vdev_min_asize;
706 mvd->vdev_ashift = cvd->vdev_ashift;
707 mvd->vdev_state = cvd->vdev_state;
709 vdev_remove_child(pvd, cvd);
710 vdev_add_child(pvd, mvd);
711 cvd->vdev_id = mvd->vdev_children;
712 vdev_add_child(mvd, cvd);
713 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
715 if (mvd == mvd->vdev_top)
716 vdev_top_transfer(cvd, mvd);
722 * Remove a 1-way mirror/replacing vdev from the tree.
725 vdev_remove_parent(vdev_t *cvd)
727 vdev_t *mvd = cvd->vdev_parent;
728 vdev_t *pvd = mvd->vdev_parent;
730 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
732 ASSERT(mvd->vdev_children == 1);
733 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
734 mvd->vdev_ops == &vdev_replacing_ops ||
735 mvd->vdev_ops == &vdev_spare_ops);
736 cvd->vdev_ashift = mvd->vdev_ashift;
738 vdev_remove_child(mvd, cvd);
739 vdev_remove_child(pvd, mvd);
742 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
743 * Otherwise, we could have detached an offline device, and when we
744 * go to import the pool we'll think we have two top-level vdevs,
745 * instead of a different version of the same top-level vdev.
747 if (mvd->vdev_top == mvd) {
748 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
749 cvd->vdev_guid += guid_delta;
750 cvd->vdev_guid_sum += guid_delta;
752 cvd->vdev_id = mvd->vdev_id;
753 vdev_add_child(pvd, cvd);
754 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
756 if (cvd == cvd->vdev_top)
757 vdev_top_transfer(mvd, cvd);
759 ASSERT(mvd->vdev_children == 0);
764 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
766 spa_t *spa = vd->vdev_spa;
767 objset_t *mos = spa->spa_meta_objset;
768 metaslab_class_t *mc;
770 uint64_t oldc = vd->vdev_ms_count;
771 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
775 if (vd->vdev_ms_shift == 0) /* not being allocated from yet */
779 * Compute the raidz-deflation ratio. Note, we hard-code
780 * in 128k (1 << 17) because it is the current "typical" blocksize.
781 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
782 * or we will inconsistently account for existing bp's.
784 vd->vdev_deflate_ratio = (1 << 17) /
785 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
787 ASSERT(oldc <= newc);
790 mc = spa->spa_log_class;
792 mc = spa->spa_normal_class;
794 if (vd->vdev_mg == NULL)
795 vd->vdev_mg = metaslab_group_create(mc, vd);
797 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
800 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
801 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
805 vd->vdev_ms_count = newc;
807 for (m = oldc; m < newc; m++) {
808 space_map_obj_t smo = { 0, 0, 0 };
811 error = dmu_read(mos, vd->vdev_ms_array,
812 m * sizeof (uint64_t), sizeof (uint64_t), &object,
818 error = dmu_bonus_hold(mos, object, FTAG, &db);
821 ASSERT3U(db->db_size, >=, sizeof (smo));
822 bcopy(db->db_data, &smo, sizeof (smo));
823 ASSERT3U(smo.smo_object, ==, object);
824 dmu_buf_rele(db, FTAG);
827 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
828 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
835 vdev_metaslab_fini(vdev_t *vd)
838 uint64_t count = vd->vdev_ms_count;
840 if (vd->vdev_ms != NULL) {
841 for (m = 0; m < count; m++)
842 if (vd->vdev_ms[m] != NULL)
843 metaslab_fini(vd->vdev_ms[m]);
844 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
849 typedef struct vdev_probe_stats {
850 boolean_t vps_readable;
851 boolean_t vps_writeable;
853 } vdev_probe_stats_t;
856 vdev_probe_done(zio_t *zio)
858 spa_t *spa = zio->io_spa;
859 vdev_t *vd = zio->io_vd;
860 vdev_probe_stats_t *vps = zio->io_private;
862 ASSERT(vd->vdev_probe_zio != NULL);
864 if (zio->io_type == ZIO_TYPE_READ) {
865 if (zio->io_error == 0)
866 vps->vps_readable = 1;
867 if (zio->io_error == 0 && spa_writeable(spa)) {
868 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
869 zio->io_offset, zio->io_size, zio->io_data,
870 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
871 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
873 zio_buf_free(zio->io_data, zio->io_size);
875 } else if (zio->io_type == ZIO_TYPE_WRITE) {
876 if (zio->io_error == 0)
877 vps->vps_writeable = 1;
878 zio_buf_free(zio->io_data, zio->io_size);
879 } else if (zio->io_type == ZIO_TYPE_NULL) {
882 vd->vdev_cant_read |= !vps->vps_readable;
883 vd->vdev_cant_write |= !vps->vps_writeable;
885 if (vdev_readable(vd) &&
886 (vdev_writeable(vd) || !spa_writeable(spa))) {
889 ASSERT(zio->io_error != 0);
890 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
891 spa, vd, NULL, 0, 0);
892 zio->io_error = ENXIO;
895 mutex_enter(&vd->vdev_probe_lock);
896 ASSERT(vd->vdev_probe_zio == zio);
897 vd->vdev_probe_zio = NULL;
898 mutex_exit(&vd->vdev_probe_lock);
900 while ((pio = zio_walk_parents(zio)) != NULL)
901 if (!vdev_accessible(vd, pio))
902 pio->io_error = ENXIO;
904 kmem_free(vps, sizeof (*vps));
909 * Determine whether this device is accessible by reading and writing
910 * to several known locations: the pad regions of each vdev label
911 * but the first (which we leave alone in case it contains a VTOC).
914 vdev_probe(vdev_t *vd, zio_t *zio)
916 spa_t *spa = vd->vdev_spa;
917 vdev_probe_stats_t *vps = NULL;
920 ASSERT(vd->vdev_ops->vdev_op_leaf);
923 * Don't probe the probe.
925 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
929 * To prevent 'probe storms' when a device fails, we create
930 * just one probe i/o at a time. All zios that want to probe
931 * this vdev will become parents of the probe io.
933 mutex_enter(&vd->vdev_probe_lock);
935 if ((pio = vd->vdev_probe_zio) == NULL) {
936 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
938 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
939 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
942 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
944 * vdev_cant_read and vdev_cant_write can only
945 * transition from TRUE to FALSE when we have the
946 * SCL_ZIO lock as writer; otherwise they can only
947 * transition from FALSE to TRUE. This ensures that
948 * any zio looking at these values can assume that
949 * failures persist for the life of the I/O. That's
950 * important because when a device has intermittent
951 * connectivity problems, we want to ensure that
952 * they're ascribed to the device (ENXIO) and not
955 * Since we hold SCL_ZIO as writer here, clear both
956 * values so the probe can reevaluate from first
959 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
960 vd->vdev_cant_read = B_FALSE;
961 vd->vdev_cant_write = B_FALSE;
964 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
965 vdev_probe_done, vps,
966 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
969 vd->vdev_probe_wanted = B_TRUE;
970 spa_async_request(spa, SPA_ASYNC_PROBE);
975 zio_add_child(zio, pio);
977 mutex_exit(&vd->vdev_probe_lock);
984 for (int l = 1; l < VDEV_LABELS; l++) {
985 zio_nowait(zio_read_phys(pio, vd,
986 vdev_label_offset(vd->vdev_psize, l,
987 offsetof(vdev_label_t, vl_pad2)),
988 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
989 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
990 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1001 vdev_open_child(void *arg)
1005 vd->vdev_open_thread = curthread;
1006 vd->vdev_open_error = vdev_open(vd);
1007 vd->vdev_open_thread = NULL;
1011 vdev_open_children(vdev_t *vd)
1014 int children = vd->vdev_children;
1016 tq = taskq_create("vdev_open", children, minclsyspri,
1017 children, children, TASKQ_PREPOPULATE);
1019 for (int c = 0; c < children; c++)
1020 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1027 * Prepare a virtual device for access.
1030 vdev_open(vdev_t *vd)
1032 spa_t *spa = vd->vdev_spa;
1035 uint64_t asize, psize;
1036 uint64_t ashift = 0;
1038 ASSERT(vd->vdev_open_thread == curthread ||
1039 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1040 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1041 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1042 vd->vdev_state == VDEV_STATE_OFFLINE);
1044 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1045 vd->vdev_cant_read = B_FALSE;
1046 vd->vdev_cant_write = B_FALSE;
1047 vd->vdev_min_asize = vdev_get_min_asize(vd);
1049 if (!vd->vdev_removed && vd->vdev_faulted) {
1050 ASSERT(vd->vdev_children == 0);
1051 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1052 VDEV_AUX_ERR_EXCEEDED);
1054 } else if (vd->vdev_offline) {
1055 ASSERT(vd->vdev_children == 0);
1056 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1060 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1062 if (zio_injection_enabled && error == 0)
1063 error = zio_handle_device_injection(vd, NULL, ENXIO);
1066 if (vd->vdev_removed &&
1067 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1068 vd->vdev_removed = B_FALSE;
1070 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1071 vd->vdev_stat.vs_aux);
1075 vd->vdev_removed = B_FALSE;
1077 if (vd->vdev_degraded) {
1078 ASSERT(vd->vdev_children == 0);
1079 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1080 VDEV_AUX_ERR_EXCEEDED);
1082 vd->vdev_state = VDEV_STATE_HEALTHY;
1085 for (int c = 0; c < vd->vdev_children; c++) {
1086 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1087 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1093 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1095 if (vd->vdev_children == 0) {
1096 if (osize < SPA_MINDEVSIZE) {
1097 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1098 VDEV_AUX_TOO_SMALL);
1102 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1104 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1105 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1106 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1107 VDEV_AUX_TOO_SMALL);
1114 vd->vdev_psize = psize;
1117 * Make sure the allocatable size hasn't shrunk.
1119 if (asize < vd->vdev_min_asize) {
1120 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1121 VDEV_AUX_BAD_LABEL);
1125 if (vd->vdev_asize == 0) {
1127 * This is the first-ever open, so use the computed values.
1128 * For testing purposes, a higher ashift can be requested.
1130 vd->vdev_asize = asize;
1131 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1134 * Make sure the alignment requirement hasn't increased.
1136 if (ashift > vd->vdev_top->vdev_ashift) {
1137 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1138 VDEV_AUX_BAD_LABEL);
1144 * If all children are healthy and the asize has increased,
1145 * then we've experienced dynamic LUN growth. If automatic
1146 * expansion is enabled then use the additional space.
1148 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1149 (vd->vdev_expanding || spa->spa_autoexpand))
1150 vd->vdev_asize = asize;
1152 vdev_set_min_asize(vd);
1155 * Ensure we can issue some IO before declaring the
1156 * vdev open for business.
1158 if (vd->vdev_ops->vdev_op_leaf &&
1159 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1160 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1161 VDEV_AUX_IO_FAILURE);
1166 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1167 * resilver. But don't do this if we are doing a reopen for a scrub,
1168 * since this would just restart the scrub we are already doing.
1170 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1171 vdev_resilver_needed(vd, NULL, NULL))
1172 spa_async_request(spa, SPA_ASYNC_RESILVER);
1178 * Called once the vdevs are all opened, this routine validates the label
1179 * contents. This needs to be done before vdev_load() so that we don't
1180 * inadvertently do repair I/Os to the wrong device.
1182 * This function will only return failure if one of the vdevs indicates that it
1183 * has since been destroyed or exported. This is only possible if
1184 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1185 * will be updated but the function will return 0.
1188 vdev_validate(vdev_t *vd)
1190 spa_t *spa = vd->vdev_spa;
1192 uint64_t guid, top_guid;
1195 for (int c = 0; c < vd->vdev_children; c++)
1196 if (vdev_validate(vd->vdev_child[c]) != 0)
1200 * If the device has already failed, or was marked offline, don't do
1201 * any further validation. Otherwise, label I/O will fail and we will
1202 * overwrite the previous state.
1204 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1206 if ((label = vdev_label_read_config(vd)) == NULL) {
1207 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1208 VDEV_AUX_BAD_LABEL);
1212 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1213 &guid) != 0 || guid != spa_guid(spa)) {
1214 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1215 VDEV_AUX_CORRUPT_DATA);
1221 * If this vdev just became a top-level vdev because its
1222 * sibling was detached, it will have adopted the parent's
1223 * vdev guid -- but the label may or may not be on disk yet.
1224 * Fortunately, either version of the label will have the
1225 * same top guid, so if we're a top-level vdev, we can
1226 * safely compare to that instead.
1228 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1230 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1232 (vd->vdev_guid != guid &&
1233 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1234 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1235 VDEV_AUX_CORRUPT_DATA);
1240 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1242 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1243 VDEV_AUX_CORRUPT_DATA);
1251 * If spa->spa_load_verbatim is true, no need to check the
1252 * state of the pool.
1254 if (!spa->spa_load_verbatim &&
1255 spa->spa_load_state == SPA_LOAD_OPEN &&
1256 state != POOL_STATE_ACTIVE)
1260 * If we were able to open and validate a vdev that was
1261 * previously marked permanently unavailable, clear that state
1264 if (vd->vdev_not_present)
1265 vd->vdev_not_present = 0;
1272 * Close a virtual device.
1275 vdev_close(vdev_t *vd)
1277 spa_t *spa = vd->vdev_spa;
1279 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1281 vd->vdev_ops->vdev_op_close(vd);
1283 vdev_cache_purge(vd);
1286 * We record the previous state before we close it, so that if we are
1287 * doing a reopen(), we don't generate FMA ereports if we notice that
1288 * it's still faulted.
1290 vd->vdev_prevstate = vd->vdev_state;
1292 if (vd->vdev_offline)
1293 vd->vdev_state = VDEV_STATE_OFFLINE;
1295 vd->vdev_state = VDEV_STATE_CLOSED;
1296 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1300 vdev_reopen(vdev_t *vd)
1302 spa_t *spa = vd->vdev_spa;
1304 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1307 (void) vdev_open(vd);
1310 * Call vdev_validate() here to make sure we have the same device.
1311 * Otherwise, a device with an invalid label could be successfully
1312 * opened in response to vdev_reopen().
1315 (void) vdev_validate_aux(vd);
1316 if (vdev_readable(vd) && vdev_writeable(vd) &&
1317 vd->vdev_aux == &spa->spa_l2cache &&
1318 !l2arc_vdev_present(vd))
1319 l2arc_add_vdev(spa, vd);
1321 (void) vdev_validate(vd);
1325 * Reassess parent vdev's health.
1327 vdev_propagate_state(vd);
1331 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1336 * Normally, partial opens (e.g. of a mirror) are allowed.
1337 * For a create, however, we want to fail the request if
1338 * there are any components we can't open.
1340 error = vdev_open(vd);
1342 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1344 return (error ? error : ENXIO);
1348 * Recursively initialize all labels.
1350 if ((error = vdev_label_init(vd, txg, isreplacing ?
1351 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1360 vdev_metaslab_set_size(vdev_t *vd)
1363 * Aim for roughly 200 metaslabs per vdev.
1365 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1366 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1370 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1372 ASSERT(vd == vd->vdev_top);
1373 ASSERT(ISP2(flags));
1375 if (flags & VDD_METASLAB)
1376 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1378 if (flags & VDD_DTL)
1379 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1381 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1387 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1388 * the vdev has less than perfect replication. There are three kinds of DTL:
1390 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1392 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1394 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1395 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1396 * txgs that was scrubbed.
1398 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1399 * persistent errors or just some device being offline.
1400 * Unlike the other three, the DTL_OUTAGE map is not generally
1401 * maintained; it's only computed when needed, typically to
1402 * determine whether a device can be detached.
1404 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1405 * either has the data or it doesn't.
1407 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1408 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1409 * if any child is less than fully replicated, then so is its parent.
1410 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1411 * comprising only those txgs which appear in 'maxfaults' or more children;
1412 * those are the txgs we don't have enough replication to read. For example,
1413 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1414 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1415 * two child DTL_MISSING maps.
1417 * It should be clear from the above that to compute the DTLs and outage maps
1418 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1419 * Therefore, that is all we keep on disk. When loading the pool, or after
1420 * a configuration change, we generate all other DTLs from first principles.
1423 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1425 space_map_t *sm = &vd->vdev_dtl[t];
1427 ASSERT(t < DTL_TYPES);
1428 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1430 mutex_enter(sm->sm_lock);
1431 if (!space_map_contains(sm, txg, size))
1432 space_map_add(sm, txg, size);
1433 mutex_exit(sm->sm_lock);
1437 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1439 space_map_t *sm = &vd->vdev_dtl[t];
1440 boolean_t dirty = B_FALSE;
1442 ASSERT(t < DTL_TYPES);
1443 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1445 mutex_enter(sm->sm_lock);
1446 if (sm->sm_space != 0)
1447 dirty = space_map_contains(sm, txg, size);
1448 mutex_exit(sm->sm_lock);
1454 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1456 space_map_t *sm = &vd->vdev_dtl[t];
1459 mutex_enter(sm->sm_lock);
1460 empty = (sm->sm_space == 0);
1461 mutex_exit(sm->sm_lock);
1467 * Reassess DTLs after a config change or scrub completion.
1470 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1472 spa_t *spa = vd->vdev_spa;
1476 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1478 for (int c = 0; c < vd->vdev_children; c++)
1479 vdev_dtl_reassess(vd->vdev_child[c], txg,
1480 scrub_txg, scrub_done);
1482 if (vd == spa->spa_root_vdev)
1485 if (vd->vdev_ops->vdev_op_leaf) {
1486 mutex_enter(&vd->vdev_dtl_lock);
1487 if (scrub_txg != 0 &&
1488 (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) {
1489 /* XXX should check scrub_done? */
1491 * We completed a scrub up to scrub_txg. If we
1492 * did it without rebooting, then the scrub dtl
1493 * will be valid, so excise the old region and
1494 * fold in the scrub dtl. Otherwise, leave the
1495 * dtl as-is if there was an error.
1497 * There's little trick here: to excise the beginning
1498 * of the DTL_MISSING map, we put it into a reference
1499 * tree and then add a segment with refcnt -1 that
1500 * covers the range [0, scrub_txg). This means
1501 * that each txg in that range has refcnt -1 or 0.
1502 * We then add DTL_SCRUB with a refcnt of 2, so that
1503 * entries in the range [0, scrub_txg) will have a
1504 * positive refcnt -- either 1 or 2. We then convert
1505 * the reference tree into the new DTL_MISSING map.
1507 space_map_ref_create(&reftree);
1508 space_map_ref_add_map(&reftree,
1509 &vd->vdev_dtl[DTL_MISSING], 1);
1510 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1511 space_map_ref_add_map(&reftree,
1512 &vd->vdev_dtl[DTL_SCRUB], 2);
1513 space_map_ref_generate_map(&reftree,
1514 &vd->vdev_dtl[DTL_MISSING], 1);
1515 space_map_ref_destroy(&reftree);
1517 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1518 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1519 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1521 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1522 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1523 if (!vdev_readable(vd))
1524 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1526 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1527 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1528 mutex_exit(&vd->vdev_dtl_lock);
1531 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1535 mutex_enter(&vd->vdev_dtl_lock);
1536 for (int t = 0; t < DTL_TYPES; t++) {
1538 continue; /* leaf vdevs only */
1539 if (t == DTL_PARTIAL)
1540 minref = 1; /* i.e. non-zero */
1541 else if (vd->vdev_nparity != 0)
1542 minref = vd->vdev_nparity + 1; /* RAID-Z */
1544 minref = vd->vdev_children; /* any kind of mirror */
1545 space_map_ref_create(&reftree);
1546 for (int c = 0; c < vd->vdev_children; c++) {
1547 vdev_t *cvd = vd->vdev_child[c];
1548 mutex_enter(&cvd->vdev_dtl_lock);
1549 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[t], 1);
1550 mutex_exit(&cvd->vdev_dtl_lock);
1552 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1553 space_map_ref_destroy(&reftree);
1555 mutex_exit(&vd->vdev_dtl_lock);
1559 vdev_dtl_load(vdev_t *vd)
1561 spa_t *spa = vd->vdev_spa;
1562 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1563 objset_t *mos = spa->spa_meta_objset;
1567 ASSERT(vd->vdev_children == 0);
1569 if (smo->smo_object == 0)
1572 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1575 ASSERT3U(db->db_size, >=, sizeof (*smo));
1576 bcopy(db->db_data, smo, sizeof (*smo));
1577 dmu_buf_rele(db, FTAG);
1579 mutex_enter(&vd->vdev_dtl_lock);
1580 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1581 NULL, SM_ALLOC, smo, mos);
1582 mutex_exit(&vd->vdev_dtl_lock);
1588 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1590 spa_t *spa = vd->vdev_spa;
1591 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1592 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1593 objset_t *mos = spa->spa_meta_objset;
1599 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1601 if (vd->vdev_detached) {
1602 if (smo->smo_object != 0) {
1603 int err = dmu_object_free(mos, smo->smo_object, tx);
1604 ASSERT3U(err, ==, 0);
1605 smo->smo_object = 0;
1611 if (smo->smo_object == 0) {
1612 ASSERT(smo->smo_objsize == 0);
1613 ASSERT(smo->smo_alloc == 0);
1614 smo->smo_object = dmu_object_alloc(mos,
1615 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1616 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1617 ASSERT(smo->smo_object != 0);
1618 vdev_config_dirty(vd->vdev_top);
1621 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1623 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1626 mutex_enter(&smlock);
1628 mutex_enter(&vd->vdev_dtl_lock);
1629 space_map_walk(sm, space_map_add, &smsync);
1630 mutex_exit(&vd->vdev_dtl_lock);
1632 space_map_truncate(smo, mos, tx);
1633 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1635 space_map_destroy(&smsync);
1637 mutex_exit(&smlock);
1638 mutex_destroy(&smlock);
1640 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1641 dmu_buf_will_dirty(db, tx);
1642 ASSERT3U(db->db_size, >=, sizeof (*smo));
1643 bcopy(smo, db->db_data, sizeof (*smo));
1644 dmu_buf_rele(db, FTAG);
1650 * Determine whether the specified vdev can be offlined/detached/removed
1651 * without losing data.
1654 vdev_dtl_required(vdev_t *vd)
1656 spa_t *spa = vd->vdev_spa;
1657 vdev_t *tvd = vd->vdev_top;
1658 uint8_t cant_read = vd->vdev_cant_read;
1661 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1663 if (vd == spa->spa_root_vdev || vd == tvd)
1667 * Temporarily mark the device as unreadable, and then determine
1668 * whether this results in any DTL outages in the top-level vdev.
1669 * If not, we can safely offline/detach/remove the device.
1671 vd->vdev_cant_read = B_TRUE;
1672 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1673 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1674 vd->vdev_cant_read = cant_read;
1675 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1681 * Determine if resilver is needed, and if so the txg range.
1684 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1686 boolean_t needed = B_FALSE;
1687 uint64_t thismin = UINT64_MAX;
1688 uint64_t thismax = 0;
1690 if (vd->vdev_children == 0) {
1691 mutex_enter(&vd->vdev_dtl_lock);
1692 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1693 vdev_writeable(vd)) {
1696 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1697 thismin = ss->ss_start - 1;
1698 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1699 thismax = ss->ss_end;
1702 mutex_exit(&vd->vdev_dtl_lock);
1704 for (int c = 0; c < vd->vdev_children; c++) {
1705 vdev_t *cvd = vd->vdev_child[c];
1706 uint64_t cmin, cmax;
1708 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1709 thismin = MIN(thismin, cmin);
1710 thismax = MAX(thismax, cmax);
1716 if (needed && minp) {
1724 vdev_load(vdev_t *vd)
1727 * Recursively load all children.
1729 for (int c = 0; c < vd->vdev_children; c++)
1730 vdev_load(vd->vdev_child[c]);
1733 * If this is a top-level vdev, initialize its metaslabs.
1735 if (vd == vd->vdev_top &&
1736 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1737 vdev_metaslab_init(vd, 0) != 0))
1738 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1739 VDEV_AUX_CORRUPT_DATA);
1742 * If this is a leaf vdev, load its DTL.
1744 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1745 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1746 VDEV_AUX_CORRUPT_DATA);
1750 * The special vdev case is used for hot spares and l2cache devices. Its
1751 * sole purpose it to set the vdev state for the associated vdev. To do this,
1752 * we make sure that we can open the underlying device, then try to read the
1753 * label, and make sure that the label is sane and that it hasn't been
1754 * repurposed to another pool.
1757 vdev_validate_aux(vdev_t *vd)
1760 uint64_t guid, version;
1763 if (!vdev_readable(vd))
1766 if ((label = vdev_label_read_config(vd)) == NULL) {
1767 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1768 VDEV_AUX_CORRUPT_DATA);
1772 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1773 version > SPA_VERSION ||
1774 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1775 guid != vd->vdev_guid ||
1776 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1777 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1778 VDEV_AUX_CORRUPT_DATA);
1784 * We don't actually check the pool state here. If it's in fact in
1785 * use by another pool, we update this fact on the fly when requested.
1792 vdev_sync_done(vdev_t *vd, uint64_t txg)
1796 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
1797 metaslab_sync_done(msp, txg);
1801 vdev_sync(vdev_t *vd, uint64_t txg)
1803 spa_t *spa = vd->vdev_spa;
1808 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
1809 ASSERT(vd == vd->vdev_top);
1810 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1811 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
1812 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
1813 ASSERT(vd->vdev_ms_array != 0);
1814 vdev_config_dirty(vd);
1818 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
1819 metaslab_sync(msp, txg);
1820 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
1823 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
1824 vdev_dtl_sync(lvd, txg);
1826 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
1830 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
1832 return (vd->vdev_ops->vdev_op_asize(vd, psize));
1836 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
1837 * not be opened, and no I/O is attempted.
1840 vdev_fault(spa_t *spa, uint64_t guid)
1844 spa_vdev_state_enter(spa);
1846 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1847 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1849 if (!vd->vdev_ops->vdev_op_leaf)
1850 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1853 * Faulted state takes precedence over degraded.
1855 vd->vdev_faulted = 1ULL;
1856 vd->vdev_degraded = 0ULL;
1857 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED);
1860 * If marking the vdev as faulted cause the top-level vdev to become
1861 * unavailable, then back off and simply mark the vdev as degraded
1864 if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) {
1865 vd->vdev_degraded = 1ULL;
1866 vd->vdev_faulted = 0ULL;
1869 * If we reopen the device and it's not dead, only then do we
1874 if (vdev_readable(vd)) {
1875 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1876 VDEV_AUX_ERR_EXCEEDED);
1880 return (spa_vdev_state_exit(spa, vd, 0));
1884 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
1885 * user that something is wrong. The vdev continues to operate as normal as far
1886 * as I/O is concerned.
1889 vdev_degrade(spa_t *spa, uint64_t guid)
1893 spa_vdev_state_enter(spa);
1895 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1896 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1898 if (!vd->vdev_ops->vdev_op_leaf)
1899 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1902 * If the vdev is already faulted, then don't do anything.
1904 if (vd->vdev_faulted || vd->vdev_degraded)
1905 return (spa_vdev_state_exit(spa, NULL, 0));
1907 vd->vdev_degraded = 1ULL;
1908 if (!vdev_is_dead(vd))
1909 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1910 VDEV_AUX_ERR_EXCEEDED);
1912 return (spa_vdev_state_exit(spa, vd, 0));
1916 * Online the given vdev. If 'unspare' is set, it implies two things. First,
1917 * any attached spare device should be detached when the device finishes
1918 * resilvering. Second, the online should be treated like a 'test' online case,
1919 * so no FMA events are generated if the device fails to open.
1922 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
1924 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
1926 spa_vdev_state_enter(spa);
1928 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1929 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1931 if (!vd->vdev_ops->vdev_op_leaf)
1932 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1935 vd->vdev_offline = B_FALSE;
1936 vd->vdev_tmpoffline = B_FALSE;
1937 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
1938 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
1940 /* XXX - L2ARC 1.0 does not support expansion */
1941 if (!vd->vdev_aux) {
1942 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
1943 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
1947 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
1949 if (!vd->vdev_aux) {
1950 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
1951 pvd->vdev_expanding = B_FALSE;
1955 *newstate = vd->vdev_state;
1956 if ((flags & ZFS_ONLINE_UNSPARE) &&
1957 !vdev_is_dead(vd) && vd->vdev_parent &&
1958 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
1959 vd->vdev_parent->vdev_child[0] == vd)
1960 vd->vdev_unspare = B_TRUE;
1962 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
1964 /* XXX - L2ARC 1.0 does not support expansion */
1966 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
1967 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1969 return (spa_vdev_state_exit(spa, vd, 0));
1973 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
1978 spa_vdev_state_enter(spa);
1980 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1981 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1983 if (!vd->vdev_ops->vdev_op_leaf)
1984 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1989 * If the device isn't already offline, try to offline it.
1991 if (!vd->vdev_offline) {
1993 * If this device has the only valid copy of some data,
1994 * don't allow it to be offlined. Log devices are always
1997 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
1998 vdev_dtl_required(vd))
1999 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2002 * Offline this device and reopen its top-level vdev.
2003 * If the top-level vdev is a log device then just offline
2004 * it. Otherwise, if this action results in the top-level
2005 * vdev becoming unusable, undo it and fail the request.
2007 vd->vdev_offline = B_TRUE;
2010 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2011 vdev_is_dead(tvd)) {
2012 vd->vdev_offline = B_FALSE;
2014 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2018 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2020 if (!tvd->vdev_islog || !vdev_is_dead(tvd))
2021 return (spa_vdev_state_exit(spa, vd, 0));
2023 (void) spa_vdev_state_exit(spa, vd, 0);
2025 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
2026 NULL, DS_FIND_CHILDREN);
2028 (void) vdev_online(spa, guid, 0, NULL);
2032 * If we successfully offlined the log device then we need to
2033 * sync out the current txg so that the "stubby" block can be
2034 * removed by zil_sync().
2036 txg_wait_synced(spa->spa_dsl_pool, 0);
2041 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2042 * vdev_offline(), we assume the spa config is locked. We also clear all
2043 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2046 vdev_clear(spa_t *spa, vdev_t *vd)
2048 vdev_t *rvd = spa->spa_root_vdev;
2050 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2055 vd->vdev_stat.vs_read_errors = 0;
2056 vd->vdev_stat.vs_write_errors = 0;
2057 vd->vdev_stat.vs_checksum_errors = 0;
2059 for (int c = 0; c < vd->vdev_children; c++)
2060 vdev_clear(spa, vd->vdev_child[c]);
2063 * If we're in the FAULTED state or have experienced failed I/O, then
2064 * clear the persistent state and attempt to reopen the device. We
2065 * also mark the vdev config dirty, so that the new faulted state is
2066 * written out to disk.
2068 if (vd->vdev_faulted || vd->vdev_degraded ||
2069 !vdev_readable(vd) || !vdev_writeable(vd)) {
2071 vd->vdev_faulted = vd->vdev_degraded = 0;
2072 vd->vdev_cant_read = B_FALSE;
2073 vd->vdev_cant_write = B_FALSE;
2078 vdev_state_dirty(vd->vdev_top);
2080 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2081 spa_async_request(spa, SPA_ASYNC_RESILVER);
2083 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2088 vdev_is_dead(vdev_t *vd)
2090 return (vd->vdev_state < VDEV_STATE_DEGRADED);
2094 vdev_readable(vdev_t *vd)
2096 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2100 vdev_writeable(vdev_t *vd)
2102 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2106 vdev_allocatable(vdev_t *vd)
2108 uint64_t state = vd->vdev_state;
2111 * We currently allow allocations from vdevs which may be in the
2112 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2113 * fails to reopen then we'll catch it later when we're holding
2114 * the proper locks. Note that we have to get the vdev state
2115 * in a local variable because although it changes atomically,
2116 * we're asking two separate questions about it.
2118 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2119 !vd->vdev_cant_write);
2123 vdev_accessible(vdev_t *vd, zio_t *zio)
2125 ASSERT(zio->io_vd == vd);
2127 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2130 if (zio->io_type == ZIO_TYPE_READ)
2131 return (!vd->vdev_cant_read);
2133 if (zio->io_type == ZIO_TYPE_WRITE)
2134 return (!vd->vdev_cant_write);
2140 * Get statistics for the given vdev.
2143 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2145 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2147 mutex_enter(&vd->vdev_stat_lock);
2148 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2149 vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors;
2150 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2151 vs->vs_state = vd->vdev_state;
2152 vs->vs_rsize = vdev_get_min_asize(vd);
2153 if (vd->vdev_ops->vdev_op_leaf)
2154 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2155 mutex_exit(&vd->vdev_stat_lock);
2158 * If we're getting stats on the root vdev, aggregate the I/O counts
2159 * over all top-level vdevs (i.e. the direct children of the root).
2162 for (int c = 0; c < rvd->vdev_children; c++) {
2163 vdev_t *cvd = rvd->vdev_child[c];
2164 vdev_stat_t *cvs = &cvd->vdev_stat;
2166 mutex_enter(&vd->vdev_stat_lock);
2167 for (int t = 0; t < ZIO_TYPES; t++) {
2168 vs->vs_ops[t] += cvs->vs_ops[t];
2169 vs->vs_bytes[t] += cvs->vs_bytes[t];
2171 vs->vs_scrub_examined += cvs->vs_scrub_examined;
2172 mutex_exit(&vd->vdev_stat_lock);
2178 vdev_clear_stats(vdev_t *vd)
2180 mutex_enter(&vd->vdev_stat_lock);
2181 vd->vdev_stat.vs_space = 0;
2182 vd->vdev_stat.vs_dspace = 0;
2183 vd->vdev_stat.vs_alloc = 0;
2184 mutex_exit(&vd->vdev_stat_lock);
2188 vdev_stat_update(zio_t *zio, uint64_t psize)
2190 spa_t *spa = zio->io_spa;
2191 vdev_t *rvd = spa->spa_root_vdev;
2192 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2194 uint64_t txg = zio->io_txg;
2195 vdev_stat_t *vs = &vd->vdev_stat;
2196 zio_type_t type = zio->io_type;
2197 int flags = zio->io_flags;
2200 * If this i/o is a gang leader, it didn't do any actual work.
2202 if (zio->io_gang_tree)
2205 if (zio->io_error == 0) {
2207 * If this is a root i/o, don't count it -- we've already
2208 * counted the top-level vdevs, and vdev_get_stats() will
2209 * aggregate them when asked. This reduces contention on
2210 * the root vdev_stat_lock and implicitly handles blocks
2211 * that compress away to holes, for which there is no i/o.
2212 * (Holes never create vdev children, so all the counters
2213 * remain zero, which is what we want.)
2215 * Note: this only applies to successful i/o (io_error == 0)
2216 * because unlike i/o counts, errors are not additive.
2217 * When reading a ditto block, for example, failure of
2218 * one top-level vdev does not imply a root-level error.
2223 ASSERT(vd == zio->io_vd);
2225 if (flags & ZIO_FLAG_IO_BYPASS)
2228 mutex_enter(&vd->vdev_stat_lock);
2230 if (flags & ZIO_FLAG_IO_REPAIR) {
2231 if (flags & ZIO_FLAG_SCRUB_THREAD)
2232 vs->vs_scrub_repaired += psize;
2233 if (flags & ZIO_FLAG_SELF_HEAL)
2234 vs->vs_self_healed += psize;
2238 vs->vs_bytes[type] += psize;
2240 mutex_exit(&vd->vdev_stat_lock);
2244 if (flags & ZIO_FLAG_SPECULATIVE)
2248 * If this is an I/O error that is going to be retried, then ignore the
2249 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2250 * hard errors, when in reality they can happen for any number of
2251 * innocuous reasons (bus resets, MPxIO link failure, etc).
2253 if (zio->io_error == EIO &&
2254 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2257 mutex_enter(&vd->vdev_stat_lock);
2258 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2259 if (zio->io_error == ECKSUM)
2260 vs->vs_checksum_errors++;
2262 vs->vs_read_errors++;
2264 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2265 vs->vs_write_errors++;
2266 mutex_exit(&vd->vdev_stat_lock);
2268 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2269 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2270 (flags & ZIO_FLAG_SCRUB_THREAD))) {
2272 * This is either a normal write (not a repair), or it's a
2273 * repair induced by the scrub thread. In the normal case,
2274 * we commit the DTL change in the same txg as the block
2275 * was born. In the scrub-induced repair case, we know that
2276 * scrubs run in first-pass syncing context, so we commit
2277 * the DTL change in spa->spa_syncing_txg.
2279 * We currently do not make DTL entries for failed spontaneous
2280 * self-healing writes triggered by normal (non-scrubbing)
2281 * reads, because we have no transactional context in which to
2282 * do so -- and it's not clear that it'd be desirable anyway.
2284 if (vd->vdev_ops->vdev_op_leaf) {
2285 uint64_t commit_txg = txg;
2286 if (flags & ZIO_FLAG_SCRUB_THREAD) {
2287 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2288 ASSERT(spa_sync_pass(spa) == 1);
2289 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2290 commit_txg = spa->spa_syncing_txg;
2292 ASSERT(commit_txg >= spa->spa_syncing_txg);
2293 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2295 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2296 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2297 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2300 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2305 vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete)
2307 vdev_stat_t *vs = &vd->vdev_stat;
2309 for (int c = 0; c < vd->vdev_children; c++)
2310 vdev_scrub_stat_update(vd->vdev_child[c], type, complete);
2312 mutex_enter(&vd->vdev_stat_lock);
2314 if (type == POOL_SCRUB_NONE) {
2316 * Update completion and end time. Leave everything else alone
2317 * so we can report what happened during the previous scrub.
2319 vs->vs_scrub_complete = complete;
2320 vs->vs_scrub_end = gethrestime_sec();
2322 vs->vs_scrub_type = type;
2323 vs->vs_scrub_complete = 0;
2324 vs->vs_scrub_examined = 0;
2325 vs->vs_scrub_repaired = 0;
2326 vs->vs_scrub_start = gethrestime_sec();
2327 vs->vs_scrub_end = 0;
2330 mutex_exit(&vd->vdev_stat_lock);
2334 * Update the in-core space usage stats for this vdev and the root vdev.
2337 vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta,
2338 boolean_t update_root)
2340 int64_t dspace_delta = space_delta;
2341 spa_t *spa = vd->vdev_spa;
2342 vdev_t *rvd = spa->spa_root_vdev;
2344 ASSERT(vd == vd->vdev_top);
2347 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2348 * factor. We must calculate this here and not at the root vdev
2349 * because the root vdev's psize-to-asize is simply the max of its
2350 * childrens', thus not accurate enough for us.
2352 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2353 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2354 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2355 vd->vdev_deflate_ratio;
2357 mutex_enter(&vd->vdev_stat_lock);
2358 vd->vdev_stat.vs_space += space_delta;
2359 vd->vdev_stat.vs_alloc += alloc_delta;
2360 vd->vdev_stat.vs_dspace += dspace_delta;
2361 mutex_exit(&vd->vdev_stat_lock);
2364 ASSERT(rvd == vd->vdev_parent);
2365 ASSERT(vd->vdev_ms_count != 0);
2368 * Don't count non-normal (e.g. intent log) space as part of
2369 * the pool's capacity.
2371 if (vd->vdev_mg->mg_class != spa->spa_normal_class)
2374 mutex_enter(&rvd->vdev_stat_lock);
2375 rvd->vdev_stat.vs_space += space_delta;
2376 rvd->vdev_stat.vs_alloc += alloc_delta;
2377 rvd->vdev_stat.vs_dspace += dspace_delta;
2378 mutex_exit(&rvd->vdev_stat_lock);
2383 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2384 * so that it will be written out next time the vdev configuration is synced.
2385 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2388 vdev_config_dirty(vdev_t *vd)
2390 spa_t *spa = vd->vdev_spa;
2391 vdev_t *rvd = spa->spa_root_vdev;
2395 * If this is an aux vdev (as with l2cache and spare devices), then we
2396 * update the vdev config manually and set the sync flag.
2398 if (vd->vdev_aux != NULL) {
2399 spa_aux_vdev_t *sav = vd->vdev_aux;
2403 for (c = 0; c < sav->sav_count; c++) {
2404 if (sav->sav_vdevs[c] == vd)
2408 if (c == sav->sav_count) {
2410 * We're being removed. There's nothing more to do.
2412 ASSERT(sav->sav_sync == B_TRUE);
2416 sav->sav_sync = B_TRUE;
2418 if (nvlist_lookup_nvlist_array(sav->sav_config,
2419 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2420 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2421 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2427 * Setting the nvlist in the middle if the array is a little
2428 * sketchy, but it will work.
2430 nvlist_free(aux[c]);
2431 aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE);
2437 * The dirty list is protected by the SCL_CONFIG lock. The caller
2438 * must either hold SCL_CONFIG as writer, or must be the sync thread
2439 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2440 * so this is sufficient to ensure mutual exclusion.
2442 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2443 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2444 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2447 for (c = 0; c < rvd->vdev_children; c++)
2448 vdev_config_dirty(rvd->vdev_child[c]);
2450 ASSERT(vd == vd->vdev_top);
2452 if (!list_link_active(&vd->vdev_config_dirty_node))
2453 list_insert_head(&spa->spa_config_dirty_list, vd);
2458 vdev_config_clean(vdev_t *vd)
2460 spa_t *spa = vd->vdev_spa;
2462 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2463 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2464 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2466 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2467 list_remove(&spa->spa_config_dirty_list, vd);
2471 * Mark a top-level vdev's state as dirty, so that the next pass of
2472 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2473 * the state changes from larger config changes because they require
2474 * much less locking, and are often needed for administrative actions.
2477 vdev_state_dirty(vdev_t *vd)
2479 spa_t *spa = vd->vdev_spa;
2481 ASSERT(vd == vd->vdev_top);
2484 * The state list is protected by the SCL_STATE lock. The caller
2485 * must either hold SCL_STATE as writer, or must be the sync thread
2486 * (which holds SCL_STATE as reader). There's only one sync thread,
2487 * so this is sufficient to ensure mutual exclusion.
2489 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2490 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2491 spa_config_held(spa, SCL_STATE, RW_READER)));
2493 if (!list_link_active(&vd->vdev_state_dirty_node))
2494 list_insert_head(&spa->spa_state_dirty_list, vd);
2498 vdev_state_clean(vdev_t *vd)
2500 spa_t *spa = vd->vdev_spa;
2502 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2503 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2504 spa_config_held(spa, SCL_STATE, RW_READER)));
2506 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2507 list_remove(&spa->spa_state_dirty_list, vd);
2511 * Propagate vdev state up from children to parent.
2514 vdev_propagate_state(vdev_t *vd)
2516 spa_t *spa = vd->vdev_spa;
2517 vdev_t *rvd = spa->spa_root_vdev;
2518 int degraded = 0, faulted = 0;
2522 if (vd->vdev_children > 0) {
2523 for (int c = 0; c < vd->vdev_children; c++) {
2524 child = vd->vdev_child[c];
2526 if (!vdev_readable(child) ||
2527 (!vdev_writeable(child) && spa_writeable(spa))) {
2529 * Root special: if there is a top-level log
2530 * device, treat the root vdev as if it were
2533 if (child->vdev_islog && vd == rvd)
2537 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2541 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2545 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2548 * Root special: if there is a top-level vdev that cannot be
2549 * opened due to corrupted metadata, then propagate the root
2550 * vdev's aux state as 'corrupt' rather than 'insufficient
2553 if (corrupted && vd == rvd &&
2554 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2555 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2556 VDEV_AUX_CORRUPT_DATA);
2559 if (vd->vdev_parent)
2560 vdev_propagate_state(vd->vdev_parent);
2564 * Set a vdev's state. If this is during an open, we don't update the parent
2565 * state, because we're in the process of opening children depth-first.
2566 * Otherwise, we propagate the change to the parent.
2568 * If this routine places a device in a faulted state, an appropriate ereport is
2572 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2574 uint64_t save_state;
2575 spa_t *spa = vd->vdev_spa;
2577 if (state == vd->vdev_state) {
2578 vd->vdev_stat.vs_aux = aux;
2582 save_state = vd->vdev_state;
2584 vd->vdev_state = state;
2585 vd->vdev_stat.vs_aux = aux;
2588 * If we are setting the vdev state to anything but an open state, then
2589 * always close the underlying device. Otherwise, we keep accessible
2590 * but invalid devices open forever. We don't call vdev_close() itself,
2591 * because that implies some extra checks (offline, etc) that we don't
2592 * want here. This is limited to leaf devices, because otherwise
2593 * closing the device will affect other children.
2595 if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf)
2596 vd->vdev_ops->vdev_op_close(vd);
2598 if (vd->vdev_removed &&
2599 state == VDEV_STATE_CANT_OPEN &&
2600 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2602 * If the previous state is set to VDEV_STATE_REMOVED, then this
2603 * device was previously marked removed and someone attempted to
2604 * reopen it. If this failed due to a nonexistent device, then
2605 * keep the device in the REMOVED state. We also let this be if
2606 * it is one of our special test online cases, which is only
2607 * attempting to online the device and shouldn't generate an FMA
2610 vd->vdev_state = VDEV_STATE_REMOVED;
2611 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2612 } else if (state == VDEV_STATE_REMOVED) {
2614 * Indicate to the ZFS DE that this device has been removed, and
2615 * any recent errors should be ignored.
2617 zfs_post_remove(spa, vd);
2618 vd->vdev_removed = B_TRUE;
2619 } else if (state == VDEV_STATE_CANT_OPEN) {
2621 * If we fail to open a vdev during an import, we mark it as
2622 * "not available", which signifies that it was never there to
2623 * begin with. Failure to open such a device is not considered
2626 if (spa->spa_load_state == SPA_LOAD_IMPORT &&
2627 vd->vdev_ops->vdev_op_leaf)
2628 vd->vdev_not_present = 1;
2631 * Post the appropriate ereport. If the 'prevstate' field is
2632 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2633 * that this is part of a vdev_reopen(). In this case, we don't
2634 * want to post the ereport if the device was already in the
2635 * CANT_OPEN state beforehand.
2637 * If the 'checkremove' flag is set, then this is an attempt to
2638 * online the device in response to an insertion event. If we
2639 * hit this case, then we have detected an insertion event for a
2640 * faulted or offline device that wasn't in the removed state.
2641 * In this scenario, we don't post an ereport because we are
2642 * about to replace the device, or attempt an online with
2643 * vdev_forcefault, which will generate the fault for us.
2645 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
2646 !vd->vdev_not_present && !vd->vdev_checkremove &&
2647 vd != spa->spa_root_vdev) {
2651 case VDEV_AUX_OPEN_FAILED:
2652 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
2654 case VDEV_AUX_CORRUPT_DATA:
2655 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
2657 case VDEV_AUX_NO_REPLICAS:
2658 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
2660 case VDEV_AUX_BAD_GUID_SUM:
2661 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
2663 case VDEV_AUX_TOO_SMALL:
2664 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
2666 case VDEV_AUX_BAD_LABEL:
2667 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
2669 case VDEV_AUX_IO_FAILURE:
2670 class = FM_EREPORT_ZFS_IO_FAILURE;
2673 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
2676 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
2679 /* Erase any notion of persistent removed state */
2680 vd->vdev_removed = B_FALSE;
2682 vd->vdev_removed = B_FALSE;
2685 if (!isopen && vd->vdev_parent)
2686 vdev_propagate_state(vd->vdev_parent);
2690 * Check the vdev configuration to ensure that it's capable of supporting
2691 * a root pool. Currently, we do not support RAID-Z or partial configuration.
2692 * In addition, only a single top-level vdev is allowed and none of the leaves
2693 * can be wholedisks.
2696 vdev_is_bootable(vdev_t *vd)
2698 if (!vd->vdev_ops->vdev_op_leaf) {
2699 char *vdev_type = vd->vdev_ops->vdev_op_type;
2701 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
2702 vd->vdev_children > 1) {
2704 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
2705 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
2708 } else if (vd->vdev_wholedisk == 1) {
2712 for (int c = 0; c < vd->vdev_children; c++) {
2713 if (!vdev_is_bootable(vd->vdev_child[c]))
2720 vdev_load_log_state(vdev_t *vd, nvlist_t *nv)
2725 spa_t *spa = vd->vdev_spa;
2727 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2728 &child, &children) == 0) {
2729 for (int c = 0; c < children; c++)
2730 vdev_load_log_state(vd->vdev_child[c], child[c]);
2733 if (vd->vdev_ops->vdev_op_leaf && nvlist_lookup_uint64(nv,
2734 ZPOOL_CONFIG_OFFLINE, &val) == 0 && val) {
2737 * It would be nice to call vdev_offline()
2738 * directly but the pool isn't fully loaded and
2739 * the txg threads have not been started yet.
2741 spa_config_enter(spa, SCL_STATE_ALL, FTAG, RW_WRITER);
2742 vd->vdev_offline = val;
2743 vdev_reopen(vd->vdev_top);
2744 spa_config_exit(spa, SCL_STATE_ALL, FTAG);
2749 * Expand a vdev if possible.
2752 vdev_expand(vdev_t *vd, uint64_t txg)
2754 ASSERT(vd->vdev_top == vd);
2755 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2757 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
2758 VERIFY(vdev_metaslab_init(vd, txg) == 0);
2759 vdev_config_dirty(vd);