4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
26 #include <sys/zfs_context.h>
27 #include <sys/fm/fs/zfs.h>
29 #include <sys/spa_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/vdev_impl.h>
33 #include <sys/uberblock_impl.h>
34 #include <sys/metaslab.h>
35 #include <sys/metaslab_impl.h>
36 #include <sys/space_map.h>
39 #include <sys/fs/zfs.h>
42 #include <sys/dsl_scan.h>
45 * Virtual device management.
48 static vdev_ops_t *vdev_ops_table[] = {
61 /* maximum scrub/resilver I/O queue per leaf vdev */
62 int zfs_scrub_limit = 10;
65 * Given a vdev type, return the appropriate ops vector.
68 vdev_getops(const char *type)
70 vdev_ops_t *ops, **opspp;
72 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
73 if (strcmp(ops->vdev_op_type, type) == 0)
80 * Default asize function: return the MAX of psize with the asize of
81 * all children. This is what's used by anything other than RAID-Z.
84 vdev_default_asize(vdev_t *vd, uint64_t psize)
86 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
90 for (c = 0; c < vd->vdev_children; c++) {
91 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
92 asize = MAX(asize, csize);
99 * Get the minimum allocatable size. We define the allocatable size as
100 * the vdev's asize rounded to the nearest metaslab. This allows us to
101 * replace or attach devices which don't have the same physical size but
102 * can still satisfy the same number of allocations.
105 vdev_get_min_asize(vdev_t *vd)
107 vdev_t *pvd = vd->vdev_parent;
110 * The our parent is NULL (inactive spare or cache) or is the root,
111 * just return our own asize.
114 return (vd->vdev_asize);
117 * The top-level vdev just returns the allocatable size rounded
118 * to the nearest metaslab.
120 if (vd == vd->vdev_top)
121 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
124 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
125 * so each child must provide at least 1/Nth of its asize.
127 if (pvd->vdev_ops == &vdev_raidz_ops)
128 return (pvd->vdev_min_asize / pvd->vdev_children);
130 return (pvd->vdev_min_asize);
134 vdev_set_min_asize(vdev_t *vd)
137 vd->vdev_min_asize = vdev_get_min_asize(vd);
139 for (c = 0; c < vd->vdev_children; c++)
140 vdev_set_min_asize(vd->vdev_child[c]);
144 vdev_lookup_top(spa_t *spa, uint64_t vdev)
146 vdev_t *rvd = spa->spa_root_vdev;
148 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
150 if (vdev < rvd->vdev_children) {
151 ASSERT(rvd->vdev_child[vdev] != NULL);
152 return (rvd->vdev_child[vdev]);
159 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
164 if (vd->vdev_guid == guid)
167 for (c = 0; c < vd->vdev_children; c++)
168 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
176 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
178 size_t oldsize, newsize;
179 uint64_t id = cvd->vdev_id;
182 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
183 ASSERT(cvd->vdev_parent == NULL);
185 cvd->vdev_parent = pvd;
190 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
192 oldsize = pvd->vdev_children * sizeof (vdev_t *);
193 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
194 newsize = pvd->vdev_children * sizeof (vdev_t *);
196 newchild = kmem_zalloc(newsize, KM_SLEEP);
197 if (pvd->vdev_child != NULL) {
198 bcopy(pvd->vdev_child, newchild, oldsize);
199 kmem_free(pvd->vdev_child, oldsize);
202 pvd->vdev_child = newchild;
203 pvd->vdev_child[id] = cvd;
205 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
206 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
209 * Walk up all ancestors to update guid sum.
211 for (; pvd != NULL; pvd = pvd->vdev_parent)
212 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
216 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
219 uint_t id = cvd->vdev_id;
221 ASSERT(cvd->vdev_parent == pvd);
226 ASSERT(id < pvd->vdev_children);
227 ASSERT(pvd->vdev_child[id] == cvd);
229 pvd->vdev_child[id] = NULL;
230 cvd->vdev_parent = NULL;
232 for (c = 0; c < pvd->vdev_children; c++)
233 if (pvd->vdev_child[c])
236 if (c == pvd->vdev_children) {
237 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
238 pvd->vdev_child = NULL;
239 pvd->vdev_children = 0;
243 * Walk up all ancestors to update guid sum.
245 for (; pvd != NULL; pvd = pvd->vdev_parent)
246 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
250 * Remove any holes in the child array.
253 vdev_compact_children(vdev_t *pvd)
255 vdev_t **newchild, *cvd;
256 int oldc = pvd->vdev_children;
260 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
262 for (c = newc = 0; c < oldc; c++)
263 if (pvd->vdev_child[c])
266 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
268 for (c = newc = 0; c < oldc; c++) {
269 if ((cvd = pvd->vdev_child[c]) != NULL) {
270 newchild[newc] = cvd;
271 cvd->vdev_id = newc++;
275 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
276 pvd->vdev_child = newchild;
277 pvd->vdev_children = newc;
281 * Allocate and minimally initialize a vdev_t.
284 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;
296 if (guid == 0 && ops != &vdev_hole_ops) {
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 guid = spa_generate_guid(NULL);
305 * Any other vdev's guid must be unique within the pool.
307 guid = spa_generate_guid(spa);
309 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
314 vd->vdev_guid = guid;
315 vd->vdev_guid_sum = guid;
317 vd->vdev_state = VDEV_STATE_CLOSED;
318 vd->vdev_ishole = (ops == &vdev_hole_ops);
320 list_link_init(&vd->vdev_config_dirty_node);
321 list_link_init(&vd->vdev_state_dirty_node);
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 (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)
400 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
404 * Set the nparity property for RAID-Z vdevs.
407 if (ops == &vdev_raidz_ops) {
408 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
410 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
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 * Retrieve the vdev creation time.
477 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
481 * If we're a top-level vdev, try to load the allocation parameters.
483 if (parent && !parent->vdev_parent &&
484 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
485 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
487 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
489 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
491 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
495 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
496 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
497 alloctype == VDEV_ALLOC_ADD ||
498 alloctype == VDEV_ALLOC_SPLIT ||
499 alloctype == VDEV_ALLOC_ROOTPOOL);
500 vd->vdev_mg = metaslab_group_create(islog ?
501 spa_log_class(spa) : spa_normal_class(spa), vd);
505 * If we're a leaf vdev, try to load the DTL object and other state.
507 if (vd->vdev_ops->vdev_op_leaf &&
508 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
509 alloctype == VDEV_ALLOC_ROOTPOOL)) {
510 if (alloctype == VDEV_ALLOC_LOAD) {
511 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
512 &vd->vdev_dtl_smo.smo_object);
513 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
517 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
520 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
521 &spare) == 0 && spare)
525 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
528 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
529 &vd->vdev_resilvering);
532 * When importing a pool, we want to ignore the persistent fault
533 * state, as the diagnosis made on another system may not be
534 * valid in the current context. Local vdevs will
535 * remain in the faulted state.
537 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
538 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
540 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
542 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
545 if (vd->vdev_faulted || vd->vdev_degraded) {
549 VDEV_AUX_ERR_EXCEEDED;
550 if (nvlist_lookup_string(nv,
551 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
552 strcmp(aux, "external") == 0)
553 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
559 * Add ourselves to the parent's list of children.
561 vdev_add_child(parent, vd);
569 vdev_free(vdev_t *vd)
572 spa_t *spa = vd->vdev_spa;
575 * vdev_free() implies closing the vdev first. This is simpler than
576 * trying to ensure complicated semantics for all callers.
580 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
581 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
586 for (c = 0; c < vd->vdev_children; c++)
587 vdev_free(vd->vdev_child[c]);
589 ASSERT(vd->vdev_child == NULL);
590 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
593 * Discard allocation state.
595 if (vd->vdev_mg != NULL) {
596 vdev_metaslab_fini(vd);
597 metaslab_group_destroy(vd->vdev_mg);
600 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
601 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
602 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
605 * Remove this vdev from its parent's child list.
607 vdev_remove_child(vd->vdev_parent, vd);
609 ASSERT(vd->vdev_parent == NULL);
612 * Clean up vdev structure.
618 spa_strfree(vd->vdev_path);
620 spa_strfree(vd->vdev_devid);
621 if (vd->vdev_physpath)
622 spa_strfree(vd->vdev_physpath);
624 spa_strfree(vd->vdev_fru);
626 if (vd->vdev_isspare)
627 spa_spare_remove(vd);
628 if (vd->vdev_isl2cache)
629 spa_l2cache_remove(vd);
631 txg_list_destroy(&vd->vdev_ms_list);
632 txg_list_destroy(&vd->vdev_dtl_list);
634 mutex_enter(&vd->vdev_dtl_lock);
635 for (t = 0; t < DTL_TYPES; t++) {
636 space_map_unload(&vd->vdev_dtl[t]);
637 space_map_destroy(&vd->vdev_dtl[t]);
639 mutex_exit(&vd->vdev_dtl_lock);
641 mutex_destroy(&vd->vdev_dtl_lock);
642 mutex_destroy(&vd->vdev_stat_lock);
643 mutex_destroy(&vd->vdev_probe_lock);
645 if (vd == spa->spa_root_vdev)
646 spa->spa_root_vdev = NULL;
648 kmem_free(vd, sizeof (vdev_t));
652 * Transfer top-level vdev state from svd to tvd.
655 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
657 spa_t *spa = svd->vdev_spa;
662 ASSERT(tvd == tvd->vdev_top);
664 tvd->vdev_ms_array = svd->vdev_ms_array;
665 tvd->vdev_ms_shift = svd->vdev_ms_shift;
666 tvd->vdev_ms_count = svd->vdev_ms_count;
668 svd->vdev_ms_array = 0;
669 svd->vdev_ms_shift = 0;
670 svd->vdev_ms_count = 0;
673 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
674 tvd->vdev_mg = svd->vdev_mg;
675 tvd->vdev_ms = svd->vdev_ms;
680 if (tvd->vdev_mg != NULL)
681 tvd->vdev_mg->mg_vd = tvd;
683 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
684 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
685 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
687 svd->vdev_stat.vs_alloc = 0;
688 svd->vdev_stat.vs_space = 0;
689 svd->vdev_stat.vs_dspace = 0;
691 for (t = 0; t < TXG_SIZE; t++) {
692 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
693 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
694 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
695 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
696 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
697 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
700 if (list_link_active(&svd->vdev_config_dirty_node)) {
701 vdev_config_clean(svd);
702 vdev_config_dirty(tvd);
705 if (list_link_active(&svd->vdev_state_dirty_node)) {
706 vdev_state_clean(svd);
707 vdev_state_dirty(tvd);
710 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
711 svd->vdev_deflate_ratio = 0;
713 tvd->vdev_islog = svd->vdev_islog;
718 vdev_top_update(vdev_t *tvd, vdev_t *vd)
727 for (c = 0; c < vd->vdev_children; c++)
728 vdev_top_update(tvd, vd->vdev_child[c]);
732 * Add a mirror/replacing vdev above an existing vdev.
735 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
737 spa_t *spa = cvd->vdev_spa;
738 vdev_t *pvd = cvd->vdev_parent;
741 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
743 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
745 mvd->vdev_asize = cvd->vdev_asize;
746 mvd->vdev_min_asize = cvd->vdev_min_asize;
747 mvd->vdev_ashift = cvd->vdev_ashift;
748 mvd->vdev_state = cvd->vdev_state;
749 mvd->vdev_crtxg = cvd->vdev_crtxg;
751 vdev_remove_child(pvd, cvd);
752 vdev_add_child(pvd, mvd);
753 cvd->vdev_id = mvd->vdev_children;
754 vdev_add_child(mvd, cvd);
755 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
757 if (mvd == mvd->vdev_top)
758 vdev_top_transfer(cvd, mvd);
764 * Remove a 1-way mirror/replacing vdev from the tree.
767 vdev_remove_parent(vdev_t *cvd)
769 vdev_t *mvd = cvd->vdev_parent;
770 vdev_t *pvd = mvd->vdev_parent;
772 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
774 ASSERT(mvd->vdev_children == 1);
775 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
776 mvd->vdev_ops == &vdev_replacing_ops ||
777 mvd->vdev_ops == &vdev_spare_ops);
778 cvd->vdev_ashift = mvd->vdev_ashift;
780 vdev_remove_child(mvd, cvd);
781 vdev_remove_child(pvd, mvd);
784 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
785 * Otherwise, we could have detached an offline device, and when we
786 * go to import the pool we'll think we have two top-level vdevs,
787 * instead of a different version of the same top-level vdev.
789 if (mvd->vdev_top == mvd) {
790 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
791 cvd->vdev_orig_guid = cvd->vdev_guid;
792 cvd->vdev_guid += guid_delta;
793 cvd->vdev_guid_sum += guid_delta;
795 cvd->vdev_id = mvd->vdev_id;
796 vdev_add_child(pvd, cvd);
797 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
799 if (cvd == cvd->vdev_top)
800 vdev_top_transfer(mvd, cvd);
802 ASSERT(mvd->vdev_children == 0);
807 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
809 spa_t *spa = vd->vdev_spa;
810 objset_t *mos = spa->spa_meta_objset;
812 uint64_t oldc = vd->vdev_ms_count;
813 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
817 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
820 * This vdev is not being allocated from yet or is a hole.
822 if (vd->vdev_ms_shift == 0)
825 ASSERT(!vd->vdev_ishole);
828 * Compute the raidz-deflation ratio. Note, we hard-code
829 * in 128k (1 << 17) because it is the current "typical" blocksize.
830 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
831 * or we will inconsistently account for existing bp's.
833 vd->vdev_deflate_ratio = (1 << 17) /
834 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
836 ASSERT(oldc <= newc);
838 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP | KM_NODEBUG);
841 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
842 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
846 vd->vdev_ms_count = newc;
848 for (m = oldc; m < newc; m++) {
849 space_map_obj_t smo = { 0, 0, 0 };
852 error = dmu_read(mos, vd->vdev_ms_array,
853 m * sizeof (uint64_t), sizeof (uint64_t), &object,
859 error = dmu_bonus_hold(mos, object, FTAG, &db);
862 ASSERT3U(db->db_size, >=, sizeof (smo));
863 bcopy(db->db_data, &smo, sizeof (smo));
864 ASSERT3U(smo.smo_object, ==, object);
865 dmu_buf_rele(db, FTAG);
868 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
869 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
873 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
876 * If the vdev is being removed we don't activate
877 * the metaslabs since we want to ensure that no new
878 * allocations are performed on this device.
880 if (oldc == 0 && !vd->vdev_removing)
881 metaslab_group_activate(vd->vdev_mg);
884 spa_config_exit(spa, SCL_ALLOC, FTAG);
890 vdev_metaslab_fini(vdev_t *vd)
893 uint64_t count = vd->vdev_ms_count;
895 if (vd->vdev_ms != NULL) {
896 metaslab_group_passivate(vd->vdev_mg);
897 for (m = 0; m < count; m++)
898 if (vd->vdev_ms[m] != NULL)
899 metaslab_fini(vd->vdev_ms[m]);
900 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
905 typedef struct vdev_probe_stats {
906 boolean_t vps_readable;
907 boolean_t vps_writeable;
909 } vdev_probe_stats_t;
912 vdev_probe_done(zio_t *zio)
914 spa_t *spa = zio->io_spa;
915 vdev_t *vd = zio->io_vd;
916 vdev_probe_stats_t *vps = zio->io_private;
918 ASSERT(vd->vdev_probe_zio != NULL);
920 if (zio->io_type == ZIO_TYPE_READ) {
921 if (zio->io_error == 0)
922 vps->vps_readable = 1;
923 if (zio->io_error == 0 && spa_writeable(spa)) {
924 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
925 zio->io_offset, zio->io_size, zio->io_data,
926 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
927 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
929 zio_buf_free(zio->io_data, zio->io_size);
931 } else if (zio->io_type == ZIO_TYPE_WRITE) {
932 if (zio->io_error == 0)
933 vps->vps_writeable = 1;
934 zio_buf_free(zio->io_data, zio->io_size);
935 } else if (zio->io_type == ZIO_TYPE_NULL) {
938 vd->vdev_cant_read |= !vps->vps_readable;
939 vd->vdev_cant_write |= !vps->vps_writeable;
941 if (vdev_readable(vd) &&
942 (vdev_writeable(vd) || !spa_writeable(spa))) {
945 ASSERT(zio->io_error != 0);
946 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
947 spa, vd, NULL, 0, 0);
948 zio->io_error = ENXIO;
951 mutex_enter(&vd->vdev_probe_lock);
952 ASSERT(vd->vdev_probe_zio == zio);
953 vd->vdev_probe_zio = NULL;
954 mutex_exit(&vd->vdev_probe_lock);
956 while ((pio = zio_walk_parents(zio)) != NULL)
957 if (!vdev_accessible(vd, pio))
958 pio->io_error = ENXIO;
960 kmem_free(vps, sizeof (*vps));
965 * Determine whether this device is accessible by reading and writing
966 * to several known locations: the pad regions of each vdev label
967 * but the first (which we leave alone in case it contains a VTOC).
970 vdev_probe(vdev_t *vd, zio_t *zio)
972 spa_t *spa = vd->vdev_spa;
973 vdev_probe_stats_t *vps = NULL;
977 ASSERT(vd->vdev_ops->vdev_op_leaf);
980 * Don't probe the probe.
982 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
986 * To prevent 'probe storms' when a device fails, we create
987 * just one probe i/o at a time. All zios that want to probe
988 * this vdev will become parents of the probe io.
990 mutex_enter(&vd->vdev_probe_lock);
992 if ((pio = vd->vdev_probe_zio) == NULL) {
993 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
995 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
996 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
999 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1001 * vdev_cant_read and vdev_cant_write can only
1002 * transition from TRUE to FALSE when we have the
1003 * SCL_ZIO lock as writer; otherwise they can only
1004 * transition from FALSE to TRUE. This ensures that
1005 * any zio looking at these values can assume that
1006 * failures persist for the life of the I/O. That's
1007 * important because when a device has intermittent
1008 * connectivity problems, we want to ensure that
1009 * they're ascribed to the device (ENXIO) and not
1012 * Since we hold SCL_ZIO as writer here, clear both
1013 * values so the probe can reevaluate from first
1016 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1017 vd->vdev_cant_read = B_FALSE;
1018 vd->vdev_cant_write = B_FALSE;
1021 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1022 vdev_probe_done, vps,
1023 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1026 * We can't change the vdev state in this context, so we
1027 * kick off an async task to do it on our behalf.
1030 vd->vdev_probe_wanted = B_TRUE;
1031 spa_async_request(spa, SPA_ASYNC_PROBE);
1036 zio_add_child(zio, pio);
1038 mutex_exit(&vd->vdev_probe_lock);
1041 ASSERT(zio != NULL);
1045 for (l = 1; l < VDEV_LABELS; l++) {
1046 zio_nowait(zio_read_phys(pio, vd,
1047 vdev_label_offset(vd->vdev_psize, l,
1048 offsetof(vdev_label_t, vl_pad2)),
1049 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1050 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1051 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1062 vdev_open_child(void *arg)
1066 vd->vdev_open_thread = curthread;
1067 vd->vdev_open_error = vdev_open(vd);
1068 vd->vdev_open_thread = NULL;
1072 vdev_uses_zvols(vdev_t *vd)
1075 * Stacking zpools on top of zvols is unsupported until we implement a method
1076 * for determining if an arbitrary block device is a zvol without using the
1077 * path. Solaris would check the 'zvol' path component but this does not
1078 * exist in the Linux port, so we really should do something like stat the
1079 * file and check the major number. This is complicated by the fact that
1080 * we need to do this portably in user or kernel space.
1085 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1086 strlen(ZVOL_DIR)) == 0)
1088 for (c = 0; c < vd->vdev_children; c++)
1089 if (vdev_uses_zvols(vd->vdev_child[c]))
1096 vdev_open_children(vdev_t *vd)
1099 int children = vd->vdev_children;
1103 * in order to handle pools on top of zvols, do the opens
1104 * in a single thread so that the same thread holds the
1105 * spa_namespace_lock
1107 if (vdev_uses_zvols(vd)) {
1108 for (c = 0; c < children; c++)
1109 vd->vdev_child[c]->vdev_open_error =
1110 vdev_open(vd->vdev_child[c]);
1113 tq = taskq_create("vdev_open", children, minclsyspri,
1114 children, children, TASKQ_PREPOPULATE);
1116 for (c = 0; c < children; c++)
1117 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1124 * Prepare a virtual device for access.
1127 vdev_open(vdev_t *vd)
1129 spa_t *spa = vd->vdev_spa;
1132 uint64_t asize, psize;
1133 uint64_t ashift = 0;
1136 ASSERT(vd->vdev_open_thread == curthread ||
1137 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1138 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1139 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1140 vd->vdev_state == VDEV_STATE_OFFLINE);
1142 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1143 vd->vdev_cant_read = B_FALSE;
1144 vd->vdev_cant_write = B_FALSE;
1145 vd->vdev_min_asize = vdev_get_min_asize(vd);
1148 * If this vdev is not removed, check its fault status. If it's
1149 * faulted, bail out of the open.
1151 if (!vd->vdev_removed && vd->vdev_faulted) {
1152 ASSERT(vd->vdev_children == 0);
1153 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1154 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1155 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1156 vd->vdev_label_aux);
1158 } else if (vd->vdev_offline) {
1159 ASSERT(vd->vdev_children == 0);
1160 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1164 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1167 * Reset the vdev_reopening flag so that we actually close
1168 * the vdev on error.
1170 vd->vdev_reopening = B_FALSE;
1171 if (zio_injection_enabled && error == 0)
1172 error = zio_handle_device_injection(vd, NULL, ENXIO);
1175 if (vd->vdev_removed &&
1176 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1177 vd->vdev_removed = B_FALSE;
1179 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1180 vd->vdev_stat.vs_aux);
1184 vd->vdev_removed = B_FALSE;
1187 * Recheck the faulted flag now that we have confirmed that
1188 * the vdev is accessible. If we're faulted, bail.
1190 if (vd->vdev_faulted) {
1191 ASSERT(vd->vdev_children == 0);
1192 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1193 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1194 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1195 vd->vdev_label_aux);
1199 if (vd->vdev_degraded) {
1200 ASSERT(vd->vdev_children == 0);
1201 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1202 VDEV_AUX_ERR_EXCEEDED);
1204 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1208 * For hole or missing vdevs we just return success.
1210 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1213 for (c = 0; c < vd->vdev_children; c++) {
1214 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1215 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1221 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1223 if (vd->vdev_children == 0) {
1224 if (osize < SPA_MINDEVSIZE) {
1225 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1226 VDEV_AUX_TOO_SMALL);
1230 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1232 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1233 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1234 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1235 VDEV_AUX_TOO_SMALL);
1242 vd->vdev_psize = psize;
1245 * Make sure the allocatable size hasn't shrunk.
1247 if (asize < vd->vdev_min_asize) {
1248 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1249 VDEV_AUX_BAD_LABEL);
1253 if (vd->vdev_asize == 0) {
1255 * This is the first-ever open, so use the computed values.
1256 * For testing purposes, a higher ashift can be requested.
1258 vd->vdev_asize = asize;
1259 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1262 * Make sure the alignment requirement hasn't increased.
1264 if (ashift > vd->vdev_top->vdev_ashift) {
1265 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1266 VDEV_AUX_BAD_LABEL);
1272 * If all children are healthy and the asize has increased,
1273 * then we've experienced dynamic LUN growth. If automatic
1274 * expansion is enabled then use the additional space.
1276 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1277 (vd->vdev_expanding || spa->spa_autoexpand))
1278 vd->vdev_asize = asize;
1280 vdev_set_min_asize(vd);
1283 * Ensure we can issue some IO before declaring the
1284 * vdev open for business.
1286 if (vd->vdev_ops->vdev_op_leaf &&
1287 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1288 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1289 VDEV_AUX_ERR_EXCEEDED);
1294 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1295 * resilver. But don't do this if we are doing a reopen for a scrub,
1296 * since this would just restart the scrub we are already doing.
1298 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1299 vdev_resilver_needed(vd, NULL, NULL))
1300 spa_async_request(spa, SPA_ASYNC_RESILVER);
1306 * Called once the vdevs are all opened, this routine validates the label
1307 * contents. This needs to be done before vdev_load() so that we don't
1308 * inadvertently do repair I/Os to the wrong device.
1310 * This function will only return failure if one of the vdevs indicates that it
1311 * has since been destroyed or exported. This is only possible if
1312 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1313 * will be updated but the function will return 0.
1316 vdev_validate(vdev_t *vd)
1318 spa_t *spa = vd->vdev_spa;
1320 uint64_t guid = 0, top_guid;
1324 for (c = 0; c < vd->vdev_children; c++)
1325 if (vdev_validate(vd->vdev_child[c]) != 0)
1329 * If the device has already failed, or was marked offline, don't do
1330 * any further validation. Otherwise, label I/O will fail and we will
1331 * overwrite the previous state.
1333 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1334 uint64_t aux_guid = 0;
1337 if ((label = vdev_label_read_config(vd)) == NULL) {
1338 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1339 VDEV_AUX_BAD_LABEL);
1344 * Determine if this vdev has been split off into another
1345 * pool. If so, then refuse to open it.
1347 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1348 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1349 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1350 VDEV_AUX_SPLIT_POOL);
1355 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1356 &guid) != 0 || guid != spa_guid(spa)) {
1357 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1358 VDEV_AUX_CORRUPT_DATA);
1363 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1364 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1369 * If this vdev just became a top-level vdev because its
1370 * sibling was detached, it will have adopted the parent's
1371 * vdev guid -- but the label may or may not be on disk yet.
1372 * Fortunately, either version of the label will have the
1373 * same top guid, so if we're a top-level vdev, we can
1374 * safely compare to that instead.
1376 * If we split this vdev off instead, then we also check the
1377 * original pool's guid. We don't want to consider the vdev
1378 * corrupt if it is partway through a split operation.
1380 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1382 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1384 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1385 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1386 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1387 VDEV_AUX_CORRUPT_DATA);
1392 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1394 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1395 VDEV_AUX_CORRUPT_DATA);
1403 * If this is a verbatim import, no need to check the
1404 * state of the pool.
1406 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1407 spa_load_state(spa) == SPA_LOAD_OPEN &&
1408 state != POOL_STATE_ACTIVE)
1412 * If we were able to open and validate a vdev that was
1413 * previously marked permanently unavailable, clear that state
1416 if (vd->vdev_not_present)
1417 vd->vdev_not_present = 0;
1424 * Close a virtual device.
1427 vdev_close(vdev_t *vd)
1429 vdev_t *pvd = vd->vdev_parent;
1430 ASSERTV(spa_t *spa = vd->vdev_spa);
1432 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1435 * If our parent is reopening, then we are as well, unless we are
1438 if (pvd != NULL && pvd->vdev_reopening)
1439 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1441 vd->vdev_ops->vdev_op_close(vd);
1443 vdev_cache_purge(vd);
1446 * We record the previous state before we close it, so that if we are
1447 * doing a reopen(), we don't generate FMA ereports if we notice that
1448 * it's still faulted.
1450 vd->vdev_prevstate = vd->vdev_state;
1452 if (vd->vdev_offline)
1453 vd->vdev_state = VDEV_STATE_OFFLINE;
1455 vd->vdev_state = VDEV_STATE_CLOSED;
1456 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1460 vdev_hold(vdev_t *vd)
1462 spa_t *spa = vd->vdev_spa;
1465 ASSERT(spa_is_root(spa));
1466 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1469 for (c = 0; c < vd->vdev_children; c++)
1470 vdev_hold(vd->vdev_child[c]);
1472 if (vd->vdev_ops->vdev_op_leaf)
1473 vd->vdev_ops->vdev_op_hold(vd);
1477 vdev_rele(vdev_t *vd)
1481 ASSERT(spa_is_root(vd->vdev_spa));
1482 for (c = 0; c < vd->vdev_children; c++)
1483 vdev_rele(vd->vdev_child[c]);
1485 if (vd->vdev_ops->vdev_op_leaf)
1486 vd->vdev_ops->vdev_op_rele(vd);
1490 * Reopen all interior vdevs and any unopened leaves. We don't actually
1491 * reopen leaf vdevs which had previously been opened as they might deadlock
1492 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1493 * If the leaf has never been opened then open it, as usual.
1496 vdev_reopen(vdev_t *vd)
1498 spa_t *spa = vd->vdev_spa;
1500 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1502 /* set the reopening flag unless we're taking the vdev offline */
1503 vd->vdev_reopening = !vd->vdev_offline;
1505 (void) vdev_open(vd);
1508 * Call vdev_validate() here to make sure we have the same device.
1509 * Otherwise, a device with an invalid label could be successfully
1510 * opened in response to vdev_reopen().
1513 (void) vdev_validate_aux(vd);
1514 if (vdev_readable(vd) && vdev_writeable(vd) &&
1515 vd->vdev_aux == &spa->spa_l2cache &&
1516 !l2arc_vdev_present(vd))
1517 l2arc_add_vdev(spa, vd);
1519 (void) vdev_validate(vd);
1523 * Reassess parent vdev's health.
1525 vdev_propagate_state(vd);
1529 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1534 * Normally, partial opens (e.g. of a mirror) are allowed.
1535 * For a create, however, we want to fail the request if
1536 * there are any components we can't open.
1538 error = vdev_open(vd);
1540 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1542 return (error ? error : ENXIO);
1546 * Recursively initialize all labels.
1548 if ((error = vdev_label_init(vd, txg, isreplacing ?
1549 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1558 vdev_metaslab_set_size(vdev_t *vd)
1561 * Aim for roughly 200 metaslabs per vdev.
1563 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1564 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1568 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1570 ASSERT(vd == vd->vdev_top);
1571 ASSERT(!vd->vdev_ishole);
1572 ASSERT(ISP2(flags));
1573 ASSERT(spa_writeable(vd->vdev_spa));
1575 if (flags & VDD_METASLAB)
1576 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1578 if (flags & VDD_DTL)
1579 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1581 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1587 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1588 * the vdev has less than perfect replication. There are four kinds of DTL:
1590 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1592 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1594 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1595 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1596 * txgs that was scrubbed.
1598 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1599 * persistent errors or just some device being offline.
1600 * Unlike the other three, the DTL_OUTAGE map is not generally
1601 * maintained; it's only computed when needed, typically to
1602 * determine whether a device can be detached.
1604 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1605 * either has the data or it doesn't.
1607 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1608 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1609 * if any child is less than fully replicated, then so is its parent.
1610 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1611 * comprising only those txgs which appear in 'maxfaults' or more children;
1612 * those are the txgs we don't have enough replication to read. For example,
1613 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1614 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1615 * two child DTL_MISSING maps.
1617 * It should be clear from the above that to compute the DTLs and outage maps
1618 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1619 * Therefore, that is all we keep on disk. When loading the pool, or after
1620 * a configuration change, we generate all other DTLs from first principles.
1623 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1625 space_map_t *sm = &vd->vdev_dtl[t];
1627 ASSERT(t < DTL_TYPES);
1628 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1629 ASSERT(spa_writeable(vd->vdev_spa));
1631 mutex_enter(sm->sm_lock);
1632 if (!space_map_contains(sm, txg, size))
1633 space_map_add(sm, txg, size);
1634 mutex_exit(sm->sm_lock);
1638 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1640 space_map_t *sm = &vd->vdev_dtl[t];
1641 boolean_t dirty = B_FALSE;
1643 ASSERT(t < DTL_TYPES);
1644 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1646 mutex_enter(sm->sm_lock);
1647 if (sm->sm_space != 0)
1648 dirty = space_map_contains(sm, txg, size);
1649 mutex_exit(sm->sm_lock);
1655 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1657 space_map_t *sm = &vd->vdev_dtl[t];
1660 mutex_enter(sm->sm_lock);
1661 empty = (sm->sm_space == 0);
1662 mutex_exit(sm->sm_lock);
1668 * Reassess DTLs after a config change or scrub completion.
1671 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1673 spa_t *spa = vd->vdev_spa;
1677 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1679 for (c = 0; c < vd->vdev_children; c++)
1680 vdev_dtl_reassess(vd->vdev_child[c], txg,
1681 scrub_txg, scrub_done);
1683 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1686 if (vd->vdev_ops->vdev_op_leaf) {
1687 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1689 mutex_enter(&vd->vdev_dtl_lock);
1690 if (scrub_txg != 0 &&
1691 (spa->spa_scrub_started ||
1692 (scn && scn->scn_phys.scn_errors == 0))) {
1694 * We completed a scrub up to scrub_txg. If we
1695 * did it without rebooting, then the scrub dtl
1696 * will be valid, so excise the old region and
1697 * fold in the scrub dtl. Otherwise, leave the
1698 * dtl as-is if there was an error.
1700 * There's little trick here: to excise the beginning
1701 * of the DTL_MISSING map, we put it into a reference
1702 * tree and then add a segment with refcnt -1 that
1703 * covers the range [0, scrub_txg). This means
1704 * that each txg in that range has refcnt -1 or 0.
1705 * We then add DTL_SCRUB with a refcnt of 2, so that
1706 * entries in the range [0, scrub_txg) will have a
1707 * positive refcnt -- either 1 or 2. We then convert
1708 * the reference tree into the new DTL_MISSING map.
1710 space_map_ref_create(&reftree);
1711 space_map_ref_add_map(&reftree,
1712 &vd->vdev_dtl[DTL_MISSING], 1);
1713 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1714 space_map_ref_add_map(&reftree,
1715 &vd->vdev_dtl[DTL_SCRUB], 2);
1716 space_map_ref_generate_map(&reftree,
1717 &vd->vdev_dtl[DTL_MISSING], 1);
1718 space_map_ref_destroy(&reftree);
1720 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1721 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1722 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1724 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1725 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1726 if (!vdev_readable(vd))
1727 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1729 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1730 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1731 mutex_exit(&vd->vdev_dtl_lock);
1734 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1738 mutex_enter(&vd->vdev_dtl_lock);
1739 for (t = 0; t < DTL_TYPES; t++) {
1740 /* account for child's outage in parent's missing map */
1741 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1743 continue; /* leaf vdevs only */
1744 if (t == DTL_PARTIAL)
1745 minref = 1; /* i.e. non-zero */
1746 else if (vd->vdev_nparity != 0)
1747 minref = vd->vdev_nparity + 1; /* RAID-Z */
1749 minref = vd->vdev_children; /* any kind of mirror */
1750 space_map_ref_create(&reftree);
1751 for (c = 0; c < vd->vdev_children; c++) {
1752 vdev_t *cvd = vd->vdev_child[c];
1753 mutex_enter(&cvd->vdev_dtl_lock);
1754 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1755 mutex_exit(&cvd->vdev_dtl_lock);
1757 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1758 space_map_ref_destroy(&reftree);
1760 mutex_exit(&vd->vdev_dtl_lock);
1764 vdev_dtl_load(vdev_t *vd)
1766 spa_t *spa = vd->vdev_spa;
1767 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1768 objset_t *mos = spa->spa_meta_objset;
1772 ASSERT(vd->vdev_children == 0);
1774 if (smo->smo_object == 0)
1777 ASSERT(!vd->vdev_ishole);
1779 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1782 ASSERT3U(db->db_size, >=, sizeof (*smo));
1783 bcopy(db->db_data, smo, sizeof (*smo));
1784 dmu_buf_rele(db, FTAG);
1786 mutex_enter(&vd->vdev_dtl_lock);
1787 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1788 NULL, SM_ALLOC, smo, mos);
1789 mutex_exit(&vd->vdev_dtl_lock);
1795 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1797 spa_t *spa = vd->vdev_spa;
1798 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1799 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1800 objset_t *mos = spa->spa_meta_objset;
1806 ASSERT(!vd->vdev_ishole);
1808 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1810 if (vd->vdev_detached) {
1811 if (smo->smo_object != 0) {
1812 VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx));
1813 smo->smo_object = 0;
1819 if (smo->smo_object == 0) {
1820 ASSERT(smo->smo_objsize == 0);
1821 ASSERT(smo->smo_alloc == 0);
1822 smo->smo_object = dmu_object_alloc(mos,
1823 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1824 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1825 ASSERT(smo->smo_object != 0);
1826 vdev_config_dirty(vd->vdev_top);
1829 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1831 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1834 mutex_enter(&smlock);
1836 mutex_enter(&vd->vdev_dtl_lock);
1837 space_map_walk(sm, space_map_add, &smsync);
1838 mutex_exit(&vd->vdev_dtl_lock);
1840 space_map_truncate(smo, mos, tx);
1841 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1843 space_map_destroy(&smsync);
1845 mutex_exit(&smlock);
1846 mutex_destroy(&smlock);
1848 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1849 dmu_buf_will_dirty(db, tx);
1850 ASSERT3U(db->db_size, >=, sizeof (*smo));
1851 bcopy(smo, db->db_data, sizeof (*smo));
1852 dmu_buf_rele(db, FTAG);
1858 * Determine whether the specified vdev can be offlined/detached/removed
1859 * without losing data.
1862 vdev_dtl_required(vdev_t *vd)
1864 spa_t *spa = vd->vdev_spa;
1865 vdev_t *tvd = vd->vdev_top;
1866 uint8_t cant_read = vd->vdev_cant_read;
1869 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1871 if (vd == spa->spa_root_vdev || vd == tvd)
1875 * Temporarily mark the device as unreadable, and then determine
1876 * whether this results in any DTL outages in the top-level vdev.
1877 * If not, we can safely offline/detach/remove the device.
1879 vd->vdev_cant_read = B_TRUE;
1880 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1881 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1882 vd->vdev_cant_read = cant_read;
1883 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1885 if (!required && zio_injection_enabled)
1886 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1892 * Determine if resilver is needed, and if so the txg range.
1895 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1897 boolean_t needed = B_FALSE;
1898 uint64_t thismin = UINT64_MAX;
1899 uint64_t thismax = 0;
1902 if (vd->vdev_children == 0) {
1903 mutex_enter(&vd->vdev_dtl_lock);
1904 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1905 vdev_writeable(vd)) {
1908 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1909 thismin = ss->ss_start - 1;
1910 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1911 thismax = ss->ss_end;
1914 mutex_exit(&vd->vdev_dtl_lock);
1916 for (c = 0; c < vd->vdev_children; c++) {
1917 vdev_t *cvd = vd->vdev_child[c];
1918 uint64_t cmin, cmax;
1920 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1921 thismin = MIN(thismin, cmin);
1922 thismax = MAX(thismax, cmax);
1928 if (needed && minp) {
1936 vdev_load(vdev_t *vd)
1941 * Recursively load all children.
1943 for (c = 0; c < vd->vdev_children; c++)
1944 vdev_load(vd->vdev_child[c]);
1947 * If this is a top-level vdev, initialize its metaslabs.
1949 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1950 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1951 vdev_metaslab_init(vd, 0) != 0))
1952 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1953 VDEV_AUX_CORRUPT_DATA);
1956 * If this is a leaf vdev, load its DTL.
1958 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1959 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1960 VDEV_AUX_CORRUPT_DATA);
1964 * The special vdev case is used for hot spares and l2cache devices. Its
1965 * sole purpose it to set the vdev state for the associated vdev. To do this,
1966 * we make sure that we can open the underlying device, then try to read the
1967 * label, and make sure that the label is sane and that it hasn't been
1968 * repurposed to another pool.
1971 vdev_validate_aux(vdev_t *vd)
1974 uint64_t guid, version;
1977 if (!vdev_readable(vd))
1980 if ((label = vdev_label_read_config(vd)) == NULL) {
1981 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1982 VDEV_AUX_CORRUPT_DATA);
1986 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1987 version > SPA_VERSION ||
1988 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1989 guid != vd->vdev_guid ||
1990 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1991 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1992 VDEV_AUX_CORRUPT_DATA);
1998 * We don't actually check the pool state here. If it's in fact in
1999 * use by another pool, we update this fact on the fly when requested.
2006 vdev_remove(vdev_t *vd, uint64_t txg)
2008 spa_t *spa = vd->vdev_spa;
2009 objset_t *mos = spa->spa_meta_objset;
2013 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2015 if (vd->vdev_dtl_smo.smo_object) {
2016 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
2017 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2018 vd->vdev_dtl_smo.smo_object = 0;
2021 if (vd->vdev_ms != NULL) {
2022 for (m = 0; m < vd->vdev_ms_count; m++) {
2023 metaslab_t *msp = vd->vdev_ms[m];
2025 if (msp == NULL || msp->ms_smo.smo_object == 0)
2028 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2029 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2030 msp->ms_smo.smo_object = 0;
2034 if (vd->vdev_ms_array) {
2035 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2036 vd->vdev_ms_array = 0;
2037 vd->vdev_ms_shift = 0;
2043 vdev_sync_done(vdev_t *vd, uint64_t txg)
2046 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2048 ASSERT(!vd->vdev_ishole);
2050 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
2051 metaslab_sync_done(msp, txg);
2054 metaslab_sync_reassess(vd->vdev_mg);
2058 vdev_sync(vdev_t *vd, uint64_t txg)
2060 spa_t *spa = vd->vdev_spa;
2065 ASSERT(!vd->vdev_ishole);
2067 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2068 ASSERT(vd == vd->vdev_top);
2069 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2070 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2071 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2072 ASSERT(vd->vdev_ms_array != 0);
2073 vdev_config_dirty(vd);
2078 * Remove the metadata associated with this vdev once it's empty.
2080 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2081 vdev_remove(vd, txg);
2083 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2084 metaslab_sync(msp, txg);
2085 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2088 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2089 vdev_dtl_sync(lvd, txg);
2091 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2095 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2097 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2101 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2102 * not be opened, and no I/O is attempted.
2105 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2109 spa_vdev_state_enter(spa, SCL_NONE);
2111 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2112 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2114 if (!vd->vdev_ops->vdev_op_leaf)
2115 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2120 * We don't directly use the aux state here, but if we do a
2121 * vdev_reopen(), we need this value to be present to remember why we
2124 vd->vdev_label_aux = aux;
2127 * Faulted state takes precedence over degraded.
2129 vd->vdev_delayed_close = B_FALSE;
2130 vd->vdev_faulted = 1ULL;
2131 vd->vdev_degraded = 0ULL;
2132 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2135 * If this device has the only valid copy of the data, then
2136 * back off and simply mark the vdev as degraded instead.
2138 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2139 vd->vdev_degraded = 1ULL;
2140 vd->vdev_faulted = 0ULL;
2143 * If we reopen the device and it's not dead, only then do we
2148 if (vdev_readable(vd))
2149 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2152 return (spa_vdev_state_exit(spa, vd, 0));
2156 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2157 * user that something is wrong. The vdev continues to operate as normal as far
2158 * as I/O is concerned.
2161 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2165 spa_vdev_state_enter(spa, SCL_NONE);
2167 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2168 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2170 if (!vd->vdev_ops->vdev_op_leaf)
2171 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2174 * If the vdev is already faulted, then don't do anything.
2176 if (vd->vdev_faulted || vd->vdev_degraded)
2177 return (spa_vdev_state_exit(spa, NULL, 0));
2179 vd->vdev_degraded = 1ULL;
2180 if (!vdev_is_dead(vd))
2181 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2184 return (spa_vdev_state_exit(spa, vd, 0));
2188 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2189 * any attached spare device should be detached when the device finishes
2190 * resilvering. Second, the online should be treated like a 'test' online case,
2191 * so no FMA events are generated if the device fails to open.
2194 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2196 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2198 spa_vdev_state_enter(spa, SCL_NONE);
2200 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2201 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2203 if (!vd->vdev_ops->vdev_op_leaf)
2204 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2207 vd->vdev_offline = B_FALSE;
2208 vd->vdev_tmpoffline = B_FALSE;
2209 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2210 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2212 /* XXX - L2ARC 1.0 does not support expansion */
2213 if (!vd->vdev_aux) {
2214 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2215 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2219 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2221 if (!vd->vdev_aux) {
2222 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2223 pvd->vdev_expanding = B_FALSE;
2227 *newstate = vd->vdev_state;
2228 if ((flags & ZFS_ONLINE_UNSPARE) &&
2229 !vdev_is_dead(vd) && vd->vdev_parent &&
2230 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2231 vd->vdev_parent->vdev_child[0] == vd)
2232 vd->vdev_unspare = B_TRUE;
2234 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2236 /* XXX - L2ARC 1.0 does not support expansion */
2238 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2239 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2241 return (spa_vdev_state_exit(spa, vd, 0));
2245 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2249 uint64_t generation;
2250 metaslab_group_t *mg;
2253 spa_vdev_state_enter(spa, SCL_ALLOC);
2255 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2256 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2258 if (!vd->vdev_ops->vdev_op_leaf)
2259 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2263 generation = spa->spa_config_generation + 1;
2266 * If the device isn't already offline, try to offline it.
2268 if (!vd->vdev_offline) {
2270 * If this device has the only valid copy of some data,
2271 * don't allow it to be offlined. Log devices are always
2274 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2275 vdev_dtl_required(vd))
2276 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2279 * If the top-level is a slog and it has had allocations
2280 * then proceed. We check that the vdev's metaslab group
2281 * is not NULL since it's possible that we may have just
2282 * added this vdev but not yet initialized its metaslabs.
2284 if (tvd->vdev_islog && mg != NULL) {
2286 * Prevent any future allocations.
2288 metaslab_group_passivate(mg);
2289 (void) spa_vdev_state_exit(spa, vd, 0);
2291 error = spa_offline_log(spa);
2293 spa_vdev_state_enter(spa, SCL_ALLOC);
2296 * Check to see if the config has changed.
2298 if (error || generation != spa->spa_config_generation) {
2299 metaslab_group_activate(mg);
2301 return (spa_vdev_state_exit(spa,
2303 (void) spa_vdev_state_exit(spa, vd, 0);
2306 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2310 * Offline this device and reopen its top-level vdev.
2311 * If the top-level vdev is a log device then just offline
2312 * it. Otherwise, if this action results in the top-level
2313 * vdev becoming unusable, undo it and fail the request.
2315 vd->vdev_offline = B_TRUE;
2318 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2319 vdev_is_dead(tvd)) {
2320 vd->vdev_offline = B_FALSE;
2322 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2326 * Add the device back into the metaslab rotor so that
2327 * once we online the device it's open for business.
2329 if (tvd->vdev_islog && mg != NULL)
2330 metaslab_group_activate(mg);
2333 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2335 return (spa_vdev_state_exit(spa, vd, 0));
2339 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2343 mutex_enter(&spa->spa_vdev_top_lock);
2344 error = vdev_offline_locked(spa, guid, flags);
2345 mutex_exit(&spa->spa_vdev_top_lock);
2351 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2352 * vdev_offline(), we assume the spa config is locked. We also clear all
2353 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2356 vdev_clear(spa_t *spa, vdev_t *vd)
2358 vdev_t *rvd = spa->spa_root_vdev;
2361 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2366 vd->vdev_stat.vs_read_errors = 0;
2367 vd->vdev_stat.vs_write_errors = 0;
2368 vd->vdev_stat.vs_checksum_errors = 0;
2370 for (c = 0; c < vd->vdev_children; c++)
2371 vdev_clear(spa, vd->vdev_child[c]);
2374 * If we're in the FAULTED state or have experienced failed I/O, then
2375 * clear the persistent state and attempt to reopen the device. We
2376 * also mark the vdev config dirty, so that the new faulted state is
2377 * written out to disk.
2379 if (vd->vdev_faulted || vd->vdev_degraded ||
2380 !vdev_readable(vd) || !vdev_writeable(vd)) {
2383 * When reopening in reponse to a clear event, it may be due to
2384 * a fmadm repair request. In this case, if the device is
2385 * still broken, we want to still post the ereport again.
2387 vd->vdev_forcefault = B_TRUE;
2389 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2390 vd->vdev_cant_read = B_FALSE;
2391 vd->vdev_cant_write = B_FALSE;
2393 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2395 vd->vdev_forcefault = B_FALSE;
2397 if (vd != rvd && vdev_writeable(vd->vdev_top))
2398 vdev_state_dirty(vd->vdev_top);
2400 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2401 spa_async_request(spa, SPA_ASYNC_RESILVER);
2403 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR);
2407 * When clearing a FMA-diagnosed fault, we always want to
2408 * unspare the device, as we assume that the original spare was
2409 * done in response to the FMA fault.
2411 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2412 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2413 vd->vdev_parent->vdev_child[0] == vd)
2414 vd->vdev_unspare = B_TRUE;
2418 vdev_is_dead(vdev_t *vd)
2421 * Holes and missing devices are always considered "dead".
2422 * This simplifies the code since we don't have to check for
2423 * these types of devices in the various code paths.
2424 * Instead we rely on the fact that we skip over dead devices
2425 * before issuing I/O to them.
2427 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2428 vd->vdev_ops == &vdev_missing_ops);
2432 vdev_readable(vdev_t *vd)
2434 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2438 vdev_writeable(vdev_t *vd)
2440 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2444 vdev_allocatable(vdev_t *vd)
2446 uint64_t state = vd->vdev_state;
2449 * We currently allow allocations from vdevs which may be in the
2450 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2451 * fails to reopen then we'll catch it later when we're holding
2452 * the proper locks. Note that we have to get the vdev state
2453 * in a local variable because although it changes atomically,
2454 * we're asking two separate questions about it.
2456 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2457 !vd->vdev_cant_write && !vd->vdev_ishole);
2461 vdev_accessible(vdev_t *vd, zio_t *zio)
2463 ASSERT(zio->io_vd == vd);
2465 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2468 if (zio->io_type == ZIO_TYPE_READ)
2469 return (!vd->vdev_cant_read);
2471 if (zio->io_type == ZIO_TYPE_WRITE)
2472 return (!vd->vdev_cant_write);
2478 * Get statistics for the given vdev.
2481 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2483 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2486 mutex_enter(&vd->vdev_stat_lock);
2487 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2488 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2489 vs->vs_state = vd->vdev_state;
2490 vs->vs_rsize = vdev_get_min_asize(vd);
2491 if (vd->vdev_ops->vdev_op_leaf)
2492 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2493 mutex_exit(&vd->vdev_stat_lock);
2496 * If we're getting stats on the root vdev, aggregate the I/O counts
2497 * over all top-level vdevs (i.e. the direct children of the root).
2500 for (c = 0; c < rvd->vdev_children; c++) {
2501 vdev_t *cvd = rvd->vdev_child[c];
2502 vdev_stat_t *cvs = &cvd->vdev_stat;
2504 mutex_enter(&vd->vdev_stat_lock);
2505 for (t = 0; t < ZIO_TYPES; t++) {
2506 vs->vs_ops[t] += cvs->vs_ops[t];
2507 vs->vs_bytes[t] += cvs->vs_bytes[t];
2509 cvs->vs_scan_removing = cvd->vdev_removing;
2510 mutex_exit(&vd->vdev_stat_lock);
2516 vdev_clear_stats(vdev_t *vd)
2518 mutex_enter(&vd->vdev_stat_lock);
2519 vd->vdev_stat.vs_space = 0;
2520 vd->vdev_stat.vs_dspace = 0;
2521 vd->vdev_stat.vs_alloc = 0;
2522 mutex_exit(&vd->vdev_stat_lock);
2526 vdev_scan_stat_init(vdev_t *vd)
2528 vdev_stat_t *vs = &vd->vdev_stat;
2531 for (c = 0; c < vd->vdev_children; c++)
2532 vdev_scan_stat_init(vd->vdev_child[c]);
2534 mutex_enter(&vd->vdev_stat_lock);
2535 vs->vs_scan_processed = 0;
2536 mutex_exit(&vd->vdev_stat_lock);
2540 vdev_stat_update(zio_t *zio, uint64_t psize)
2542 spa_t *spa = zio->io_spa;
2543 vdev_t *rvd = spa->spa_root_vdev;
2544 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2546 uint64_t txg = zio->io_txg;
2547 vdev_stat_t *vs = &vd->vdev_stat;
2548 zio_type_t type = zio->io_type;
2549 int flags = zio->io_flags;
2552 * If this i/o is a gang leader, it didn't do any actual work.
2554 if (zio->io_gang_tree)
2557 if (zio->io_error == 0) {
2559 * If this is a root i/o, don't count it -- we've already
2560 * counted the top-level vdevs, and vdev_get_stats() will
2561 * aggregate them when asked. This reduces contention on
2562 * the root vdev_stat_lock and implicitly handles blocks
2563 * that compress away to holes, for which there is no i/o.
2564 * (Holes never create vdev children, so all the counters
2565 * remain zero, which is what we want.)
2567 * Note: this only applies to successful i/o (io_error == 0)
2568 * because unlike i/o counts, errors are not additive.
2569 * When reading a ditto block, for example, failure of
2570 * one top-level vdev does not imply a root-level error.
2575 ASSERT(vd == zio->io_vd);
2577 if (flags & ZIO_FLAG_IO_BYPASS)
2580 mutex_enter(&vd->vdev_stat_lock);
2582 if (flags & ZIO_FLAG_IO_REPAIR) {
2583 if (flags & ZIO_FLAG_SCAN_THREAD) {
2584 dsl_scan_phys_t *scn_phys =
2585 &spa->spa_dsl_pool->dp_scan->scn_phys;
2586 uint64_t *processed = &scn_phys->scn_processed;
2589 if (vd->vdev_ops->vdev_op_leaf)
2590 atomic_add_64(processed, psize);
2591 vs->vs_scan_processed += psize;
2594 if (flags & ZIO_FLAG_SELF_HEAL)
2595 vs->vs_self_healed += psize;
2599 vs->vs_bytes[type] += psize;
2601 mutex_exit(&vd->vdev_stat_lock);
2605 if (flags & ZIO_FLAG_SPECULATIVE)
2609 * If this is an I/O error that is going to be retried, then ignore the
2610 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2611 * hard errors, when in reality they can happen for any number of
2612 * innocuous reasons (bus resets, MPxIO link failure, etc).
2614 if (zio->io_error == EIO &&
2615 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2619 * Intent logs writes won't propagate their error to the root
2620 * I/O so don't mark these types of failures as pool-level
2623 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2626 mutex_enter(&vd->vdev_stat_lock);
2627 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2628 if (zio->io_error == ECKSUM)
2629 vs->vs_checksum_errors++;
2631 vs->vs_read_errors++;
2633 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2634 vs->vs_write_errors++;
2635 mutex_exit(&vd->vdev_stat_lock);
2637 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2638 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2639 (flags & ZIO_FLAG_SCAN_THREAD) ||
2640 spa->spa_claiming)) {
2642 * This is either a normal write (not a repair), or it's
2643 * a repair induced by the scrub thread, or it's a repair
2644 * made by zil_claim() during spa_load() in the first txg.
2645 * In the normal case, we commit the DTL change in the same
2646 * txg as the block was born. In the scrub-induced repair
2647 * case, we know that scrubs run in first-pass syncing context,
2648 * so we commit the DTL change in spa_syncing_txg(spa).
2649 * In the zil_claim() case, we commit in spa_first_txg(spa).
2651 * We currently do not make DTL entries for failed spontaneous
2652 * self-healing writes triggered by normal (non-scrubbing)
2653 * reads, because we have no transactional context in which to
2654 * do so -- and it's not clear that it'd be desirable anyway.
2656 if (vd->vdev_ops->vdev_op_leaf) {
2657 uint64_t commit_txg = txg;
2658 if (flags & ZIO_FLAG_SCAN_THREAD) {
2659 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2660 ASSERT(spa_sync_pass(spa) == 1);
2661 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2662 commit_txg = spa_syncing_txg(spa);
2663 } else if (spa->spa_claiming) {
2664 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2665 commit_txg = spa_first_txg(spa);
2667 ASSERT(commit_txg >= spa_syncing_txg(spa));
2668 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2670 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2671 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2672 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2675 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2680 * Update the in-core space usage stats for this vdev, its metaslab class,
2681 * and the root vdev.
2684 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2685 int64_t space_delta)
2687 int64_t dspace_delta = space_delta;
2688 spa_t *spa = vd->vdev_spa;
2689 vdev_t *rvd = spa->spa_root_vdev;
2690 metaslab_group_t *mg = vd->vdev_mg;
2691 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2693 ASSERT(vd == vd->vdev_top);
2696 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2697 * factor. We must calculate this here and not at the root vdev
2698 * because the root vdev's psize-to-asize is simply the max of its
2699 * childrens', thus not accurate enough for us.
2701 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2702 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2703 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2704 vd->vdev_deflate_ratio;
2706 mutex_enter(&vd->vdev_stat_lock);
2707 vd->vdev_stat.vs_alloc += alloc_delta;
2708 vd->vdev_stat.vs_space += space_delta;
2709 vd->vdev_stat.vs_dspace += dspace_delta;
2710 mutex_exit(&vd->vdev_stat_lock);
2712 if (mc == spa_normal_class(spa)) {
2713 mutex_enter(&rvd->vdev_stat_lock);
2714 rvd->vdev_stat.vs_alloc += alloc_delta;
2715 rvd->vdev_stat.vs_space += space_delta;
2716 rvd->vdev_stat.vs_dspace += dspace_delta;
2717 mutex_exit(&rvd->vdev_stat_lock);
2721 ASSERT(rvd == vd->vdev_parent);
2722 ASSERT(vd->vdev_ms_count != 0);
2724 metaslab_class_space_update(mc,
2725 alloc_delta, defer_delta, space_delta, dspace_delta);
2730 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2731 * so that it will be written out next time the vdev configuration is synced.
2732 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2735 vdev_config_dirty(vdev_t *vd)
2737 spa_t *spa = vd->vdev_spa;
2738 vdev_t *rvd = spa->spa_root_vdev;
2741 ASSERT(spa_writeable(spa));
2744 * If this is an aux vdev (as with l2cache and spare devices), then we
2745 * update the vdev config manually and set the sync flag.
2747 if (vd->vdev_aux != NULL) {
2748 spa_aux_vdev_t *sav = vd->vdev_aux;
2752 for (c = 0; c < sav->sav_count; c++) {
2753 if (sav->sav_vdevs[c] == vd)
2757 if (c == sav->sav_count) {
2759 * We're being removed. There's nothing more to do.
2761 ASSERT(sav->sav_sync == B_TRUE);
2765 sav->sav_sync = B_TRUE;
2767 if (nvlist_lookup_nvlist_array(sav->sav_config,
2768 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2769 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2770 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2776 * Setting the nvlist in the middle if the array is a little
2777 * sketchy, but it will work.
2779 nvlist_free(aux[c]);
2780 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2786 * The dirty list is protected by the SCL_CONFIG lock. The caller
2787 * must either hold SCL_CONFIG as writer, or must be the sync thread
2788 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2789 * so this is sufficient to ensure mutual exclusion.
2791 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2792 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2793 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2796 for (c = 0; c < rvd->vdev_children; c++)
2797 vdev_config_dirty(rvd->vdev_child[c]);
2799 ASSERT(vd == vd->vdev_top);
2801 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2803 list_insert_head(&spa->spa_config_dirty_list, vd);
2808 vdev_config_clean(vdev_t *vd)
2810 spa_t *spa = vd->vdev_spa;
2812 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2813 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2814 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2816 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2817 list_remove(&spa->spa_config_dirty_list, vd);
2821 * Mark a top-level vdev's state as dirty, so that the next pass of
2822 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2823 * the state changes from larger config changes because they require
2824 * much less locking, and are often needed for administrative actions.
2827 vdev_state_dirty(vdev_t *vd)
2829 spa_t *spa = vd->vdev_spa;
2831 ASSERT(spa_writeable(spa));
2832 ASSERT(vd == vd->vdev_top);
2835 * The state list is protected by the SCL_STATE lock. The caller
2836 * must either hold SCL_STATE as writer, or must be the sync thread
2837 * (which holds SCL_STATE as reader). There's only one sync thread,
2838 * so this is sufficient to ensure mutual exclusion.
2840 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2841 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2842 spa_config_held(spa, SCL_STATE, RW_READER)));
2844 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2845 list_insert_head(&spa->spa_state_dirty_list, vd);
2849 vdev_state_clean(vdev_t *vd)
2851 spa_t *spa = vd->vdev_spa;
2853 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2854 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2855 spa_config_held(spa, SCL_STATE, RW_READER)));
2857 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2858 list_remove(&spa->spa_state_dirty_list, vd);
2862 * Propagate vdev state up from children to parent.
2865 vdev_propagate_state(vdev_t *vd)
2867 spa_t *spa = vd->vdev_spa;
2868 vdev_t *rvd = spa->spa_root_vdev;
2869 int degraded = 0, faulted = 0;
2874 if (vd->vdev_children > 0) {
2875 for (c = 0; c < vd->vdev_children; c++) {
2876 child = vd->vdev_child[c];
2879 * Don't factor holes into the decision.
2881 if (child->vdev_ishole)
2884 if (!vdev_readable(child) ||
2885 (!vdev_writeable(child) && spa_writeable(spa))) {
2887 * Root special: if there is a top-level log
2888 * device, treat the root vdev as if it were
2891 if (child->vdev_islog && vd == rvd)
2895 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2899 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2903 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2906 * Root special: if there is a top-level vdev that cannot be
2907 * opened due to corrupted metadata, then propagate the root
2908 * vdev's aux state as 'corrupt' rather than 'insufficient
2911 if (corrupted && vd == rvd &&
2912 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2913 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2914 VDEV_AUX_CORRUPT_DATA);
2917 if (vd->vdev_parent)
2918 vdev_propagate_state(vd->vdev_parent);
2922 * Set a vdev's state. If this is during an open, we don't update the parent
2923 * state, because we're in the process of opening children depth-first.
2924 * Otherwise, we propagate the change to the parent.
2926 * If this routine places a device in a faulted state, an appropriate ereport is
2930 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2932 uint64_t save_state;
2933 spa_t *spa = vd->vdev_spa;
2935 if (state == vd->vdev_state) {
2936 vd->vdev_stat.vs_aux = aux;
2940 save_state = vd->vdev_state;
2942 vd->vdev_state = state;
2943 vd->vdev_stat.vs_aux = aux;
2946 * If we are setting the vdev state to anything but an open state, then
2947 * always close the underlying device unless the device has requested
2948 * a delayed close (i.e. we're about to remove or fault the device).
2949 * Otherwise, we keep accessible but invalid devices open forever.
2950 * We don't call vdev_close() itself, because that implies some extra
2951 * checks (offline, etc) that we don't want here. This is limited to
2952 * leaf devices, because otherwise closing the device will affect other
2955 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2956 vd->vdev_ops->vdev_op_leaf)
2957 vd->vdev_ops->vdev_op_close(vd);
2960 * If we have brought this vdev back into service, we need
2961 * to notify fmd so that it can gracefully repair any outstanding
2962 * cases due to a missing device. We do this in all cases, even those
2963 * that probably don't correlate to a repaired fault. This is sure to
2964 * catch all cases, and we let the zfs-retire agent sort it out. If
2965 * this is a transient state it's OK, as the retire agent will
2966 * double-check the state of the vdev before repairing it.
2968 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2969 vd->vdev_prevstate != state)
2970 zfs_post_state_change(spa, vd);
2972 if (vd->vdev_removed &&
2973 state == VDEV_STATE_CANT_OPEN &&
2974 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2976 * If the previous state is set to VDEV_STATE_REMOVED, then this
2977 * device was previously marked removed and someone attempted to
2978 * reopen it. If this failed due to a nonexistent device, then
2979 * keep the device in the REMOVED state. We also let this be if
2980 * it is one of our special test online cases, which is only
2981 * attempting to online the device and shouldn't generate an FMA
2984 vd->vdev_state = VDEV_STATE_REMOVED;
2985 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2986 } else if (state == VDEV_STATE_REMOVED) {
2987 vd->vdev_removed = B_TRUE;
2988 } else if (state == VDEV_STATE_CANT_OPEN) {
2990 * If we fail to open a vdev during an import or recovery, we
2991 * mark it as "not available", which signifies that it was
2992 * never there to begin with. Failure to open such a device
2993 * is not considered an error.
2995 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2996 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2997 vd->vdev_ops->vdev_op_leaf)
2998 vd->vdev_not_present = 1;
3001 * Post the appropriate ereport. If the 'prevstate' field is
3002 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3003 * that this is part of a vdev_reopen(). In this case, we don't
3004 * want to post the ereport if the device was already in the
3005 * CANT_OPEN state beforehand.
3007 * If the 'checkremove' flag is set, then this is an attempt to
3008 * online the device in response to an insertion event. If we
3009 * hit this case, then we have detected an insertion event for a
3010 * faulted or offline device that wasn't in the removed state.
3011 * In this scenario, we don't post an ereport because we are
3012 * about to replace the device, or attempt an online with
3013 * vdev_forcefault, which will generate the fault for us.
3015 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3016 !vd->vdev_not_present && !vd->vdev_checkremove &&
3017 vd != spa->spa_root_vdev) {
3021 case VDEV_AUX_OPEN_FAILED:
3022 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3024 case VDEV_AUX_CORRUPT_DATA:
3025 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3027 case VDEV_AUX_NO_REPLICAS:
3028 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3030 case VDEV_AUX_BAD_GUID_SUM:
3031 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3033 case VDEV_AUX_TOO_SMALL:
3034 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3036 case VDEV_AUX_BAD_LABEL:
3037 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3040 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3043 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3046 /* Erase any notion of persistent removed state */
3047 vd->vdev_removed = B_FALSE;
3049 vd->vdev_removed = B_FALSE;
3052 if (!isopen && vd->vdev_parent)
3053 vdev_propagate_state(vd->vdev_parent);
3057 * Check the vdev configuration to ensure that it's capable of supporting
3061 vdev_is_bootable(vdev_t *vd)
3063 #if defined(__sun__) || defined(__sun)
3065 * Currently, we do not support RAID-Z or partial configuration.
3066 * In addition, only a single top-level vdev is allowed and none of the
3067 * leaves can be wholedisks.
3071 if (!vd->vdev_ops->vdev_op_leaf) {
3072 char *vdev_type = vd->vdev_ops->vdev_op_type;
3074 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3075 vd->vdev_children > 1) {
3077 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3078 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3081 } else if (vd->vdev_wholedisk == 1) {
3085 for (c = 0; c < vd->vdev_children; c++) {
3086 if (!vdev_is_bootable(vd->vdev_child[c]))
3089 #endif /* __sun__ || __sun */
3094 * Load the state from the original vdev tree (ovd) which
3095 * we've retrieved from the MOS config object. If the original
3096 * vdev was offline or faulted then we transfer that state to the
3097 * device in the current vdev tree (nvd).
3100 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3104 ASSERT(nvd->vdev_top->vdev_islog);
3105 ASSERT(spa_config_held(nvd->vdev_spa,
3106 SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3107 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3109 for (c = 0; c < nvd->vdev_children; c++)
3110 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3112 if (nvd->vdev_ops->vdev_op_leaf) {
3114 * Restore the persistent vdev state
3116 nvd->vdev_offline = ovd->vdev_offline;
3117 nvd->vdev_faulted = ovd->vdev_faulted;
3118 nvd->vdev_degraded = ovd->vdev_degraded;
3119 nvd->vdev_removed = ovd->vdev_removed;
3124 * Determine if a log device has valid content. If the vdev was
3125 * removed or faulted in the MOS config then we know that
3126 * the content on the log device has already been written to the pool.
3129 vdev_log_state_valid(vdev_t *vd)
3133 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3137 for (c = 0; c < vd->vdev_children; c++)
3138 if (vdev_log_state_valid(vd->vdev_child[c]))
3145 * Expand a vdev if possible.
3148 vdev_expand(vdev_t *vd, uint64_t txg)
3150 ASSERT(vd->vdev_top == vd);
3151 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3153 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3154 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3155 vdev_config_dirty(vd);
3163 vdev_split(vdev_t *vd)
3165 vdev_t *cvd, *pvd = vd->vdev_parent;
3167 vdev_remove_child(pvd, vd);
3168 vdev_compact_children(pvd);
3170 cvd = pvd->vdev_child[0];
3171 if (pvd->vdev_children == 1) {
3172 vdev_remove_parent(cvd);
3173 cvd->vdev_splitting = B_TRUE;
3175 vdev_propagate_state(cvd);
3178 #if defined(_KERNEL) && defined(HAVE_SPL)
3179 EXPORT_SYMBOL(vdev_fault);
3180 EXPORT_SYMBOL(vdev_degrade);
3181 EXPORT_SYMBOL(vdev_online);
3182 EXPORT_SYMBOL(vdev_offline);
3183 EXPORT_SYMBOL(vdev_clear);
3185 module_param(zfs_scrub_limit, int, 0644);
3186 MODULE_PARM_DESC(zfs_scrub_limit, "Max scrub/resilver I/O per leaf vdev");