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) {
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;
672 tvd->vdev_mg = svd->vdev_mg;
673 tvd->vdev_ms = svd->vdev_ms;
678 if (tvd->vdev_mg != NULL)
679 tvd->vdev_mg->mg_vd = tvd;
681 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
682 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
683 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
685 svd->vdev_stat.vs_alloc = 0;
686 svd->vdev_stat.vs_space = 0;
687 svd->vdev_stat.vs_dspace = 0;
689 for (t = 0; t < TXG_SIZE; t++) {
690 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
691 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
692 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
693 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
694 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
695 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
698 if (list_link_active(&svd->vdev_config_dirty_node)) {
699 vdev_config_clean(svd);
700 vdev_config_dirty(tvd);
703 if (list_link_active(&svd->vdev_state_dirty_node)) {
704 vdev_state_clean(svd);
705 vdev_state_dirty(tvd);
708 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
709 svd->vdev_deflate_ratio = 0;
711 tvd->vdev_islog = svd->vdev_islog;
716 vdev_top_update(vdev_t *tvd, vdev_t *vd)
725 for (c = 0; c < vd->vdev_children; c++)
726 vdev_top_update(tvd, vd->vdev_child[c]);
730 * Add a mirror/replacing vdev above an existing vdev.
733 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
735 spa_t *spa = cvd->vdev_spa;
736 vdev_t *pvd = cvd->vdev_parent;
739 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
741 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
743 mvd->vdev_asize = cvd->vdev_asize;
744 mvd->vdev_min_asize = cvd->vdev_min_asize;
745 mvd->vdev_ashift = cvd->vdev_ashift;
746 mvd->vdev_state = cvd->vdev_state;
747 mvd->vdev_crtxg = cvd->vdev_crtxg;
749 vdev_remove_child(pvd, cvd);
750 vdev_add_child(pvd, mvd);
751 cvd->vdev_id = mvd->vdev_children;
752 vdev_add_child(mvd, cvd);
753 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
755 if (mvd == mvd->vdev_top)
756 vdev_top_transfer(cvd, mvd);
762 * Remove a 1-way mirror/replacing vdev from the tree.
765 vdev_remove_parent(vdev_t *cvd)
767 vdev_t *mvd = cvd->vdev_parent;
768 vdev_t *pvd = mvd->vdev_parent;
770 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
772 ASSERT(mvd->vdev_children == 1);
773 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
774 mvd->vdev_ops == &vdev_replacing_ops ||
775 mvd->vdev_ops == &vdev_spare_ops);
776 cvd->vdev_ashift = mvd->vdev_ashift;
778 vdev_remove_child(mvd, cvd);
779 vdev_remove_child(pvd, mvd);
782 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
783 * Otherwise, we could have detached an offline device, and when we
784 * go to import the pool we'll think we have two top-level vdevs,
785 * instead of a different version of the same top-level vdev.
787 if (mvd->vdev_top == mvd) {
788 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
789 cvd->vdev_orig_guid = cvd->vdev_guid;
790 cvd->vdev_guid += guid_delta;
791 cvd->vdev_guid_sum += guid_delta;
793 cvd->vdev_id = mvd->vdev_id;
794 vdev_add_child(pvd, cvd);
795 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
797 if (cvd == cvd->vdev_top)
798 vdev_top_transfer(mvd, cvd);
800 ASSERT(mvd->vdev_children == 0);
805 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
807 spa_t *spa = vd->vdev_spa;
808 objset_t *mos = spa->spa_meta_objset;
810 uint64_t oldc = vd->vdev_ms_count;
811 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
815 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
818 * This vdev is not being allocated from yet or is a hole.
820 if (vd->vdev_ms_shift == 0)
823 ASSERT(!vd->vdev_ishole);
826 * Compute the raidz-deflation ratio. Note, we hard-code
827 * in 128k (1 << 17) because it is the current "typical" blocksize.
828 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
829 * or we will inconsistently account for existing bp's.
831 vd->vdev_deflate_ratio = (1 << 17) /
832 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
834 ASSERT(oldc <= newc);
836 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
839 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
840 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
844 vd->vdev_ms_count = newc;
846 for (m = oldc; m < newc; m++) {
847 space_map_obj_t smo = { 0, 0, 0 };
850 error = dmu_read(mos, vd->vdev_ms_array,
851 m * sizeof (uint64_t), sizeof (uint64_t), &object,
857 error = dmu_bonus_hold(mos, object, FTAG, &db);
860 ASSERT3U(db->db_size, >=, sizeof (smo));
861 bcopy(db->db_data, &smo, sizeof (smo));
862 ASSERT3U(smo.smo_object, ==, object);
863 dmu_buf_rele(db, FTAG);
866 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
867 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
871 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
874 * If the vdev is being removed we don't activate
875 * the metaslabs since we want to ensure that no new
876 * allocations are performed on this device.
878 if (oldc == 0 && !vd->vdev_removing)
879 metaslab_group_activate(vd->vdev_mg);
882 spa_config_exit(spa, SCL_ALLOC, FTAG);
888 vdev_metaslab_fini(vdev_t *vd)
891 uint64_t count = vd->vdev_ms_count;
893 if (vd->vdev_ms != NULL) {
894 metaslab_group_passivate(vd->vdev_mg);
895 for (m = 0; m < count; m++)
896 if (vd->vdev_ms[m] != NULL)
897 metaslab_fini(vd->vdev_ms[m]);
898 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
903 typedef struct vdev_probe_stats {
904 boolean_t vps_readable;
905 boolean_t vps_writeable;
907 } vdev_probe_stats_t;
910 vdev_probe_done(zio_t *zio)
912 spa_t *spa = zio->io_spa;
913 vdev_t *vd = zio->io_vd;
914 vdev_probe_stats_t *vps = zio->io_private;
916 ASSERT(vd->vdev_probe_zio != NULL);
918 if (zio->io_type == ZIO_TYPE_READ) {
919 if (zio->io_error == 0)
920 vps->vps_readable = 1;
921 if (zio->io_error == 0 && spa_writeable(spa)) {
922 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
923 zio->io_offset, zio->io_size, zio->io_data,
924 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
925 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
927 zio_buf_free(zio->io_data, zio->io_size);
929 } else if (zio->io_type == ZIO_TYPE_WRITE) {
930 if (zio->io_error == 0)
931 vps->vps_writeable = 1;
932 zio_buf_free(zio->io_data, zio->io_size);
933 } else if (zio->io_type == ZIO_TYPE_NULL) {
936 vd->vdev_cant_read |= !vps->vps_readable;
937 vd->vdev_cant_write |= !vps->vps_writeable;
939 if (vdev_readable(vd) &&
940 (vdev_writeable(vd) || !spa_writeable(spa))) {
943 ASSERT(zio->io_error != 0);
944 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
945 spa, vd, NULL, 0, 0);
946 zio->io_error = ENXIO;
949 mutex_enter(&vd->vdev_probe_lock);
950 ASSERT(vd->vdev_probe_zio == zio);
951 vd->vdev_probe_zio = NULL;
952 mutex_exit(&vd->vdev_probe_lock);
954 while ((pio = zio_walk_parents(zio)) != NULL)
955 if (!vdev_accessible(vd, pio))
956 pio->io_error = ENXIO;
958 kmem_free(vps, sizeof (*vps));
963 * Determine whether this device is accessible by reading and writing
964 * to several known locations: the pad regions of each vdev label
965 * but the first (which we leave alone in case it contains a VTOC).
968 vdev_probe(vdev_t *vd, zio_t *zio)
970 spa_t *spa = vd->vdev_spa;
971 vdev_probe_stats_t *vps = NULL;
975 ASSERT(vd->vdev_ops->vdev_op_leaf);
978 * Don't probe the probe.
980 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
984 * To prevent 'probe storms' when a device fails, we create
985 * just one probe i/o at a time. All zios that want to probe
986 * this vdev will become parents of the probe io.
988 mutex_enter(&vd->vdev_probe_lock);
990 if ((pio = vd->vdev_probe_zio) == NULL) {
991 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
993 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
994 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
997 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
999 * vdev_cant_read and vdev_cant_write can only
1000 * transition from TRUE to FALSE when we have the
1001 * SCL_ZIO lock as writer; otherwise they can only
1002 * transition from FALSE to TRUE. This ensures that
1003 * any zio looking at these values can assume that
1004 * failures persist for the life of the I/O. That's
1005 * important because when a device has intermittent
1006 * connectivity problems, we want to ensure that
1007 * they're ascribed to the device (ENXIO) and not
1010 * Since we hold SCL_ZIO as writer here, clear both
1011 * values so the probe can reevaluate from first
1014 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1015 vd->vdev_cant_read = B_FALSE;
1016 vd->vdev_cant_write = B_FALSE;
1019 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1020 vdev_probe_done, vps,
1021 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1024 * We can't change the vdev state in this context, so we
1025 * kick off an async task to do it on our behalf.
1028 vd->vdev_probe_wanted = B_TRUE;
1029 spa_async_request(spa, SPA_ASYNC_PROBE);
1034 zio_add_child(zio, pio);
1036 mutex_exit(&vd->vdev_probe_lock);
1039 ASSERT(zio != NULL);
1043 for (l = 1; l < VDEV_LABELS; l++) {
1044 zio_nowait(zio_read_phys(pio, vd,
1045 vdev_label_offset(vd->vdev_psize, l,
1046 offsetof(vdev_label_t, vl_pad2)),
1047 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1048 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1049 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1060 vdev_open_child(void *arg)
1064 vd->vdev_open_thread = curthread;
1065 vd->vdev_open_error = vdev_open(vd);
1066 vd->vdev_open_thread = NULL;
1070 vdev_uses_zvols(vdev_t *vd)
1074 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1075 strlen(ZVOL_DIR)) == 0)
1077 for (c = 0; c < vd->vdev_children; c++)
1078 if (vdev_uses_zvols(vd->vdev_child[c]))
1084 vdev_open_children(vdev_t *vd)
1087 int children = vd->vdev_children;
1091 * in order to handle pools on top of zvols, do the opens
1092 * in a single thread so that the same thread holds the
1093 * spa_namespace_lock
1095 if (vdev_uses_zvols(vd)) {
1096 for (c = 0; c < children; c++)
1097 vd->vdev_child[c]->vdev_open_error =
1098 vdev_open(vd->vdev_child[c]);
1101 tq = taskq_create("vdev_open", children, minclsyspri,
1102 children, children, TASKQ_PREPOPULATE);
1104 for (c = 0; c < children; c++)
1105 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1112 * Prepare a virtual device for access.
1115 vdev_open(vdev_t *vd)
1117 spa_t *spa = vd->vdev_spa;
1120 uint64_t asize, psize;
1121 uint64_t ashift = 0;
1124 ASSERT(vd->vdev_open_thread == curthread ||
1125 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1126 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1127 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1128 vd->vdev_state == VDEV_STATE_OFFLINE);
1130 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1131 vd->vdev_cant_read = B_FALSE;
1132 vd->vdev_cant_write = B_FALSE;
1133 vd->vdev_min_asize = vdev_get_min_asize(vd);
1136 * If this vdev is not removed, check its fault status. If it's
1137 * faulted, bail out of the open.
1139 if (!vd->vdev_removed && vd->vdev_faulted) {
1140 ASSERT(vd->vdev_children == 0);
1141 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1142 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1143 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1144 vd->vdev_label_aux);
1146 } else if (vd->vdev_offline) {
1147 ASSERT(vd->vdev_children == 0);
1148 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1152 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1155 * Reset the vdev_reopening flag so that we actually close
1156 * the vdev on error.
1158 vd->vdev_reopening = B_FALSE;
1159 if (zio_injection_enabled && error == 0)
1160 error = zio_handle_device_injection(vd, NULL, ENXIO);
1163 if (vd->vdev_removed &&
1164 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1165 vd->vdev_removed = B_FALSE;
1167 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1168 vd->vdev_stat.vs_aux);
1172 vd->vdev_removed = B_FALSE;
1175 * Recheck the faulted flag now that we have confirmed that
1176 * the vdev is accessible. If we're faulted, bail.
1178 if (vd->vdev_faulted) {
1179 ASSERT(vd->vdev_children == 0);
1180 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1181 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1182 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1183 vd->vdev_label_aux);
1187 if (vd->vdev_degraded) {
1188 ASSERT(vd->vdev_children == 0);
1189 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1190 VDEV_AUX_ERR_EXCEEDED);
1192 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1196 * For hole or missing vdevs we just return success.
1198 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1201 for (c = 0; c < vd->vdev_children; c++) {
1202 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1203 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1209 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1211 if (vd->vdev_children == 0) {
1212 if (osize < SPA_MINDEVSIZE) {
1213 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1214 VDEV_AUX_TOO_SMALL);
1218 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1220 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1221 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1222 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1223 VDEV_AUX_TOO_SMALL);
1230 vd->vdev_psize = psize;
1233 * Make sure the allocatable size hasn't shrunk.
1235 if (asize < vd->vdev_min_asize) {
1236 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1237 VDEV_AUX_BAD_LABEL);
1241 if (vd->vdev_asize == 0) {
1243 * This is the first-ever open, so use the computed values.
1244 * For testing purposes, a higher ashift can be requested.
1246 vd->vdev_asize = asize;
1247 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1250 * Make sure the alignment requirement hasn't increased.
1252 if (ashift > vd->vdev_top->vdev_ashift) {
1253 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1254 VDEV_AUX_BAD_LABEL);
1260 * If all children are healthy and the asize has increased,
1261 * then we've experienced dynamic LUN growth. If automatic
1262 * expansion is enabled then use the additional space.
1264 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1265 (vd->vdev_expanding || spa->spa_autoexpand))
1266 vd->vdev_asize = asize;
1268 vdev_set_min_asize(vd);
1271 * Ensure we can issue some IO before declaring the
1272 * vdev open for business.
1274 if (vd->vdev_ops->vdev_op_leaf &&
1275 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1276 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1277 VDEV_AUX_ERR_EXCEEDED);
1282 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1283 * resilver. But don't do this if we are doing a reopen for a scrub,
1284 * since this would just restart the scrub we are already doing.
1286 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1287 vdev_resilver_needed(vd, NULL, NULL))
1288 spa_async_request(spa, SPA_ASYNC_RESILVER);
1294 * Called once the vdevs are all opened, this routine validates the label
1295 * contents. This needs to be done before vdev_load() so that we don't
1296 * inadvertently do repair I/Os to the wrong device.
1298 * This function will only return failure if one of the vdevs indicates that it
1299 * has since been destroyed or exported. This is only possible if
1300 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1301 * will be updated but the function will return 0.
1304 vdev_validate(vdev_t *vd)
1306 spa_t *spa = vd->vdev_spa;
1308 uint64_t guid = 0, top_guid;
1312 for (c = 0; c < vd->vdev_children; c++)
1313 if (vdev_validate(vd->vdev_child[c]) != 0)
1317 * If the device has already failed, or was marked offline, don't do
1318 * any further validation. Otherwise, label I/O will fail and we will
1319 * overwrite the previous state.
1321 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1322 uint64_t aux_guid = 0;
1325 if ((label = vdev_label_read_config(vd)) == NULL) {
1326 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1327 VDEV_AUX_BAD_LABEL);
1332 * Determine if this vdev has been split off into another
1333 * pool. If so, then refuse to open it.
1335 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1336 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1337 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1338 VDEV_AUX_SPLIT_POOL);
1343 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1344 &guid) != 0 || guid != spa_guid(spa)) {
1345 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1346 VDEV_AUX_CORRUPT_DATA);
1351 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1352 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1357 * If this vdev just became a top-level vdev because its
1358 * sibling was detached, it will have adopted the parent's
1359 * vdev guid -- but the label may or may not be on disk yet.
1360 * Fortunately, either version of the label will have the
1361 * same top guid, so if we're a top-level vdev, we can
1362 * safely compare to that instead.
1364 * If we split this vdev off instead, then we also check the
1365 * original pool's guid. We don't want to consider the vdev
1366 * corrupt if it is partway through a split operation.
1368 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1370 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1372 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1373 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1374 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1375 VDEV_AUX_CORRUPT_DATA);
1380 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1382 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1383 VDEV_AUX_CORRUPT_DATA);
1391 * If this is a verbatim import, no need to check the
1392 * state of the pool.
1394 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1395 spa_load_state(spa) == SPA_LOAD_OPEN &&
1396 state != POOL_STATE_ACTIVE)
1400 * If we were able to open and validate a vdev that was
1401 * previously marked permanently unavailable, clear that state
1404 if (vd->vdev_not_present)
1405 vd->vdev_not_present = 0;
1412 * Close a virtual device.
1415 vdev_close(vdev_t *vd)
1417 vdev_t *pvd = vd->vdev_parent;
1418 ASSERTV(spa_t *spa = vd->vdev_spa);
1420 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1423 * If our parent is reopening, then we are as well, unless we are
1426 if (pvd != NULL && pvd->vdev_reopening)
1427 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1429 vd->vdev_ops->vdev_op_close(vd);
1431 vdev_cache_purge(vd);
1434 * We record the previous state before we close it, so that if we are
1435 * doing a reopen(), we don't generate FMA ereports if we notice that
1436 * it's still faulted.
1438 vd->vdev_prevstate = vd->vdev_state;
1440 if (vd->vdev_offline)
1441 vd->vdev_state = VDEV_STATE_OFFLINE;
1443 vd->vdev_state = VDEV_STATE_CLOSED;
1444 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1448 vdev_hold(vdev_t *vd)
1450 spa_t *spa = vd->vdev_spa;
1453 ASSERT(spa_is_root(spa));
1454 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1457 for (c = 0; c < vd->vdev_children; c++)
1458 vdev_hold(vd->vdev_child[c]);
1460 if (vd->vdev_ops->vdev_op_leaf)
1461 vd->vdev_ops->vdev_op_hold(vd);
1465 vdev_rele(vdev_t *vd)
1469 ASSERT(spa_is_root(vd->vdev_spa));
1470 for (c = 0; c < vd->vdev_children; c++)
1471 vdev_rele(vd->vdev_child[c]);
1473 if (vd->vdev_ops->vdev_op_leaf)
1474 vd->vdev_ops->vdev_op_rele(vd);
1478 * Reopen all interior vdevs and any unopened leaves. We don't actually
1479 * reopen leaf vdevs which had previously been opened as they might deadlock
1480 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1481 * If the leaf has never been opened then open it, as usual.
1484 vdev_reopen(vdev_t *vd)
1486 spa_t *spa = vd->vdev_spa;
1488 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1490 /* set the reopening flag unless we're taking the vdev offline */
1491 vd->vdev_reopening = !vd->vdev_offline;
1493 (void) vdev_open(vd);
1496 * Call vdev_validate() here to make sure we have the same device.
1497 * Otherwise, a device with an invalid label could be successfully
1498 * opened in response to vdev_reopen().
1501 (void) vdev_validate_aux(vd);
1502 if (vdev_readable(vd) && vdev_writeable(vd) &&
1503 vd->vdev_aux == &spa->spa_l2cache &&
1504 !l2arc_vdev_present(vd))
1505 l2arc_add_vdev(spa, vd);
1507 (void) vdev_validate(vd);
1511 * Reassess parent vdev's health.
1513 vdev_propagate_state(vd);
1517 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1522 * Normally, partial opens (e.g. of a mirror) are allowed.
1523 * For a create, however, we want to fail the request if
1524 * there are any components we can't open.
1526 error = vdev_open(vd);
1528 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1530 return (error ? error : ENXIO);
1534 * Recursively initialize all labels.
1536 if ((error = vdev_label_init(vd, txg, isreplacing ?
1537 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1546 vdev_metaslab_set_size(vdev_t *vd)
1549 * Aim for roughly 200 metaslabs per vdev.
1551 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1552 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1556 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1558 ASSERT(vd == vd->vdev_top);
1559 ASSERT(!vd->vdev_ishole);
1560 ASSERT(ISP2(flags));
1561 ASSERT(spa_writeable(vd->vdev_spa));
1563 if (flags & VDD_METASLAB)
1564 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1566 if (flags & VDD_DTL)
1567 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1569 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1575 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1576 * the vdev has less than perfect replication. There are four kinds of DTL:
1578 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1580 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1582 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1583 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1584 * txgs that was scrubbed.
1586 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1587 * persistent errors or just some device being offline.
1588 * Unlike the other three, the DTL_OUTAGE map is not generally
1589 * maintained; it's only computed when needed, typically to
1590 * determine whether a device can be detached.
1592 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1593 * either has the data or it doesn't.
1595 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1596 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1597 * if any child is less than fully replicated, then so is its parent.
1598 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1599 * comprising only those txgs which appear in 'maxfaults' or more children;
1600 * those are the txgs we don't have enough replication to read. For example,
1601 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1602 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1603 * two child DTL_MISSING maps.
1605 * It should be clear from the above that to compute the DTLs and outage maps
1606 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1607 * Therefore, that is all we keep on disk. When loading the pool, or after
1608 * a configuration change, we generate all other DTLs from first principles.
1611 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1613 space_map_t *sm = &vd->vdev_dtl[t];
1615 ASSERT(t < DTL_TYPES);
1616 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1617 ASSERT(spa_writeable(vd->vdev_spa));
1619 mutex_enter(sm->sm_lock);
1620 if (!space_map_contains(sm, txg, size))
1621 space_map_add(sm, txg, size);
1622 mutex_exit(sm->sm_lock);
1626 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1628 space_map_t *sm = &vd->vdev_dtl[t];
1629 boolean_t dirty = B_FALSE;
1631 ASSERT(t < DTL_TYPES);
1632 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1634 mutex_enter(sm->sm_lock);
1635 if (sm->sm_space != 0)
1636 dirty = space_map_contains(sm, txg, size);
1637 mutex_exit(sm->sm_lock);
1643 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1645 space_map_t *sm = &vd->vdev_dtl[t];
1648 mutex_enter(sm->sm_lock);
1649 empty = (sm->sm_space == 0);
1650 mutex_exit(sm->sm_lock);
1656 * Reassess DTLs after a config change or scrub completion.
1659 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1661 spa_t *spa = vd->vdev_spa;
1665 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1667 for (c = 0; c < vd->vdev_children; c++)
1668 vdev_dtl_reassess(vd->vdev_child[c], txg,
1669 scrub_txg, scrub_done);
1671 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1674 if (vd->vdev_ops->vdev_op_leaf) {
1675 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1677 mutex_enter(&vd->vdev_dtl_lock);
1678 if (scrub_txg != 0 &&
1679 (spa->spa_scrub_started ||
1680 (scn && scn->scn_phys.scn_errors == 0))) {
1682 * We completed a scrub up to scrub_txg. If we
1683 * did it without rebooting, then the scrub dtl
1684 * will be valid, so excise the old region and
1685 * fold in the scrub dtl. Otherwise, leave the
1686 * dtl as-is if there was an error.
1688 * There's little trick here: to excise the beginning
1689 * of the DTL_MISSING map, we put it into a reference
1690 * tree and then add a segment with refcnt -1 that
1691 * covers the range [0, scrub_txg). This means
1692 * that each txg in that range has refcnt -1 or 0.
1693 * We then add DTL_SCRUB with a refcnt of 2, so that
1694 * entries in the range [0, scrub_txg) will have a
1695 * positive refcnt -- either 1 or 2. We then convert
1696 * the reference tree into the new DTL_MISSING map.
1698 space_map_ref_create(&reftree);
1699 space_map_ref_add_map(&reftree,
1700 &vd->vdev_dtl[DTL_MISSING], 1);
1701 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1702 space_map_ref_add_map(&reftree,
1703 &vd->vdev_dtl[DTL_SCRUB], 2);
1704 space_map_ref_generate_map(&reftree,
1705 &vd->vdev_dtl[DTL_MISSING], 1);
1706 space_map_ref_destroy(&reftree);
1708 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1709 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1710 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1712 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1713 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1714 if (!vdev_readable(vd))
1715 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1717 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1718 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1719 mutex_exit(&vd->vdev_dtl_lock);
1722 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1726 mutex_enter(&vd->vdev_dtl_lock);
1727 for (t = 0; t < DTL_TYPES; t++) {
1728 /* account for child's outage in parent's missing map */
1729 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1731 continue; /* leaf vdevs only */
1732 if (t == DTL_PARTIAL)
1733 minref = 1; /* i.e. non-zero */
1734 else if (vd->vdev_nparity != 0)
1735 minref = vd->vdev_nparity + 1; /* RAID-Z */
1737 minref = vd->vdev_children; /* any kind of mirror */
1738 space_map_ref_create(&reftree);
1739 for (c = 0; c < vd->vdev_children; c++) {
1740 vdev_t *cvd = vd->vdev_child[c];
1741 mutex_enter(&cvd->vdev_dtl_lock);
1742 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1743 mutex_exit(&cvd->vdev_dtl_lock);
1745 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1746 space_map_ref_destroy(&reftree);
1748 mutex_exit(&vd->vdev_dtl_lock);
1752 vdev_dtl_load(vdev_t *vd)
1754 spa_t *spa = vd->vdev_spa;
1755 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1756 objset_t *mos = spa->spa_meta_objset;
1760 ASSERT(vd->vdev_children == 0);
1762 if (smo->smo_object == 0)
1765 ASSERT(!vd->vdev_ishole);
1767 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1770 ASSERT3U(db->db_size, >=, sizeof (*smo));
1771 bcopy(db->db_data, smo, sizeof (*smo));
1772 dmu_buf_rele(db, FTAG);
1774 mutex_enter(&vd->vdev_dtl_lock);
1775 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1776 NULL, SM_ALLOC, smo, mos);
1777 mutex_exit(&vd->vdev_dtl_lock);
1783 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1785 spa_t *spa = vd->vdev_spa;
1786 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1787 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1788 objset_t *mos = spa->spa_meta_objset;
1794 ASSERT(!vd->vdev_ishole);
1796 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1798 if (vd->vdev_detached) {
1799 if (smo->smo_object != 0) {
1800 VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx));
1801 smo->smo_object = 0;
1807 if (smo->smo_object == 0) {
1808 ASSERT(smo->smo_objsize == 0);
1809 ASSERT(smo->smo_alloc == 0);
1810 smo->smo_object = dmu_object_alloc(mos,
1811 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1812 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1813 ASSERT(smo->smo_object != 0);
1814 vdev_config_dirty(vd->vdev_top);
1817 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1819 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1822 mutex_enter(&smlock);
1824 mutex_enter(&vd->vdev_dtl_lock);
1825 space_map_walk(sm, space_map_add, &smsync);
1826 mutex_exit(&vd->vdev_dtl_lock);
1828 space_map_truncate(smo, mos, tx);
1829 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1831 space_map_destroy(&smsync);
1833 mutex_exit(&smlock);
1834 mutex_destroy(&smlock);
1836 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1837 dmu_buf_will_dirty(db, tx);
1838 ASSERT3U(db->db_size, >=, sizeof (*smo));
1839 bcopy(smo, db->db_data, sizeof (*smo));
1840 dmu_buf_rele(db, FTAG);
1846 * Determine whether the specified vdev can be offlined/detached/removed
1847 * without losing data.
1850 vdev_dtl_required(vdev_t *vd)
1852 spa_t *spa = vd->vdev_spa;
1853 vdev_t *tvd = vd->vdev_top;
1854 uint8_t cant_read = vd->vdev_cant_read;
1857 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1859 if (vd == spa->spa_root_vdev || vd == tvd)
1863 * Temporarily mark the device as unreadable, and then determine
1864 * whether this results in any DTL outages in the top-level vdev.
1865 * If not, we can safely offline/detach/remove the device.
1867 vd->vdev_cant_read = B_TRUE;
1868 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1869 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1870 vd->vdev_cant_read = cant_read;
1871 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1873 if (!required && zio_injection_enabled)
1874 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1880 * Determine if resilver is needed, and if so the txg range.
1883 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1885 boolean_t needed = B_FALSE;
1886 uint64_t thismin = UINT64_MAX;
1887 uint64_t thismax = 0;
1890 if (vd->vdev_children == 0) {
1891 mutex_enter(&vd->vdev_dtl_lock);
1892 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1893 vdev_writeable(vd)) {
1896 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1897 thismin = ss->ss_start - 1;
1898 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1899 thismax = ss->ss_end;
1902 mutex_exit(&vd->vdev_dtl_lock);
1904 for (c = 0; c < vd->vdev_children; c++) {
1905 vdev_t *cvd = vd->vdev_child[c];
1906 uint64_t cmin, cmax;
1908 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1909 thismin = MIN(thismin, cmin);
1910 thismax = MAX(thismax, cmax);
1916 if (needed && minp) {
1924 vdev_load(vdev_t *vd)
1929 * Recursively load all children.
1931 for (c = 0; c < vd->vdev_children; c++)
1932 vdev_load(vd->vdev_child[c]);
1935 * If this is a top-level vdev, initialize its metaslabs.
1937 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1938 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1939 vdev_metaslab_init(vd, 0) != 0))
1940 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1941 VDEV_AUX_CORRUPT_DATA);
1944 * If this is a leaf vdev, load its DTL.
1946 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1947 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1948 VDEV_AUX_CORRUPT_DATA);
1952 * The special vdev case is used for hot spares and l2cache devices. Its
1953 * sole purpose it to set the vdev state for the associated vdev. To do this,
1954 * we make sure that we can open the underlying device, then try to read the
1955 * label, and make sure that the label is sane and that it hasn't been
1956 * repurposed to another pool.
1959 vdev_validate_aux(vdev_t *vd)
1962 uint64_t guid, version;
1965 if (!vdev_readable(vd))
1968 if ((label = vdev_label_read_config(vd)) == NULL) {
1969 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1970 VDEV_AUX_CORRUPT_DATA);
1974 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1975 version > SPA_VERSION ||
1976 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1977 guid != vd->vdev_guid ||
1978 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1979 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1980 VDEV_AUX_CORRUPT_DATA);
1986 * We don't actually check the pool state here. If it's in fact in
1987 * use by another pool, we update this fact on the fly when requested.
1994 vdev_remove(vdev_t *vd, uint64_t txg)
1996 spa_t *spa = vd->vdev_spa;
1997 objset_t *mos = spa->spa_meta_objset;
2001 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2003 if (vd->vdev_dtl_smo.smo_object) {
2004 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
2005 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2006 vd->vdev_dtl_smo.smo_object = 0;
2009 if (vd->vdev_ms != NULL) {
2010 for (m = 0; m < vd->vdev_ms_count; m++) {
2011 metaslab_t *msp = vd->vdev_ms[m];
2013 if (msp == NULL || msp->ms_smo.smo_object == 0)
2016 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2017 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2018 msp->ms_smo.smo_object = 0;
2022 if (vd->vdev_ms_array) {
2023 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2024 vd->vdev_ms_array = 0;
2025 vd->vdev_ms_shift = 0;
2031 vdev_sync_done(vdev_t *vd, uint64_t txg)
2034 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2036 ASSERT(!vd->vdev_ishole);
2038 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
2039 metaslab_sync_done(msp, txg);
2042 metaslab_sync_reassess(vd->vdev_mg);
2046 vdev_sync(vdev_t *vd, uint64_t txg)
2048 spa_t *spa = vd->vdev_spa;
2053 ASSERT(!vd->vdev_ishole);
2055 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2056 ASSERT(vd == vd->vdev_top);
2057 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2058 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2059 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2060 ASSERT(vd->vdev_ms_array != 0);
2061 vdev_config_dirty(vd);
2066 * Remove the metadata associated with this vdev once it's empty.
2068 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2069 vdev_remove(vd, txg);
2071 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2072 metaslab_sync(msp, txg);
2073 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2076 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2077 vdev_dtl_sync(lvd, txg);
2079 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2083 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2085 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2089 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2090 * not be opened, and no I/O is attempted.
2093 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2097 spa_vdev_state_enter(spa, SCL_NONE);
2099 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2100 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2102 if (!vd->vdev_ops->vdev_op_leaf)
2103 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2108 * We don't directly use the aux state here, but if we do a
2109 * vdev_reopen(), we need this value to be present to remember why we
2112 vd->vdev_label_aux = aux;
2115 * Faulted state takes precedence over degraded.
2117 vd->vdev_delayed_close = B_FALSE;
2118 vd->vdev_faulted = 1ULL;
2119 vd->vdev_degraded = 0ULL;
2120 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2123 * If this device has the only valid copy of the data, then
2124 * back off and simply mark the vdev as degraded instead.
2126 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2127 vd->vdev_degraded = 1ULL;
2128 vd->vdev_faulted = 0ULL;
2131 * If we reopen the device and it's not dead, only then do we
2136 if (vdev_readable(vd))
2137 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2140 return (spa_vdev_state_exit(spa, vd, 0));
2144 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2145 * user that something is wrong. The vdev continues to operate as normal as far
2146 * as I/O is concerned.
2149 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2153 spa_vdev_state_enter(spa, SCL_NONE);
2155 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2156 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2158 if (!vd->vdev_ops->vdev_op_leaf)
2159 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2162 * If the vdev is already faulted, then don't do anything.
2164 if (vd->vdev_faulted || vd->vdev_degraded)
2165 return (spa_vdev_state_exit(spa, NULL, 0));
2167 vd->vdev_degraded = 1ULL;
2168 if (!vdev_is_dead(vd))
2169 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2172 return (spa_vdev_state_exit(spa, vd, 0));
2176 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2177 * any attached spare device should be detached when the device finishes
2178 * resilvering. Second, the online should be treated like a 'test' online case,
2179 * so no FMA events are generated if the device fails to open.
2182 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2184 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2186 spa_vdev_state_enter(spa, SCL_NONE);
2188 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2189 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2191 if (!vd->vdev_ops->vdev_op_leaf)
2192 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2195 vd->vdev_offline = B_FALSE;
2196 vd->vdev_tmpoffline = B_FALSE;
2197 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2198 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2200 /* XXX - L2ARC 1.0 does not support expansion */
2201 if (!vd->vdev_aux) {
2202 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2203 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2207 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2209 if (!vd->vdev_aux) {
2210 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2211 pvd->vdev_expanding = B_FALSE;
2215 *newstate = vd->vdev_state;
2216 if ((flags & ZFS_ONLINE_UNSPARE) &&
2217 !vdev_is_dead(vd) && vd->vdev_parent &&
2218 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2219 vd->vdev_parent->vdev_child[0] == vd)
2220 vd->vdev_unspare = B_TRUE;
2222 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2224 /* XXX - L2ARC 1.0 does not support expansion */
2226 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2227 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2229 return (spa_vdev_state_exit(spa, vd, 0));
2233 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2237 uint64_t generation;
2238 metaslab_group_t *mg;
2241 spa_vdev_state_enter(spa, SCL_ALLOC);
2243 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2244 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2246 if (!vd->vdev_ops->vdev_op_leaf)
2247 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2251 generation = spa->spa_config_generation + 1;
2254 * If the device isn't already offline, try to offline it.
2256 if (!vd->vdev_offline) {
2258 * If this device has the only valid copy of some data,
2259 * don't allow it to be offlined. Log devices are always
2262 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2263 vdev_dtl_required(vd))
2264 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2267 * If the top-level is a slog and it has had allocations
2268 * then proceed. We check that the vdev's metaslab group
2269 * is not NULL since it's possible that we may have just
2270 * added this vdev but not yet initialized its metaslabs.
2272 if (tvd->vdev_islog && mg != NULL) {
2274 * Prevent any future allocations.
2276 metaslab_group_passivate(mg);
2277 (void) spa_vdev_state_exit(spa, vd, 0);
2279 error = spa_offline_log(spa);
2281 spa_vdev_state_enter(spa, SCL_ALLOC);
2284 * Check to see if the config has changed.
2286 if (error || generation != spa->spa_config_generation) {
2287 metaslab_group_activate(mg);
2289 return (spa_vdev_state_exit(spa,
2291 (void) spa_vdev_state_exit(spa, vd, 0);
2294 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2298 * Offline this device and reopen its top-level vdev.
2299 * If the top-level vdev is a log device then just offline
2300 * it. Otherwise, if this action results in the top-level
2301 * vdev becoming unusable, undo it and fail the request.
2303 vd->vdev_offline = B_TRUE;
2306 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2307 vdev_is_dead(tvd)) {
2308 vd->vdev_offline = B_FALSE;
2310 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2314 * Add the device back into the metaslab rotor so that
2315 * once we online the device it's open for business.
2317 if (tvd->vdev_islog && mg != NULL)
2318 metaslab_group_activate(mg);
2321 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2323 return (spa_vdev_state_exit(spa, vd, 0));
2327 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2331 mutex_enter(&spa->spa_vdev_top_lock);
2332 error = vdev_offline_locked(spa, guid, flags);
2333 mutex_exit(&spa->spa_vdev_top_lock);
2339 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2340 * vdev_offline(), we assume the spa config is locked. We also clear all
2341 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2344 vdev_clear(spa_t *spa, vdev_t *vd)
2346 vdev_t *rvd = spa->spa_root_vdev;
2349 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2354 vd->vdev_stat.vs_read_errors = 0;
2355 vd->vdev_stat.vs_write_errors = 0;
2356 vd->vdev_stat.vs_checksum_errors = 0;
2358 for (c = 0; c < vd->vdev_children; c++)
2359 vdev_clear(spa, vd->vdev_child[c]);
2362 * If we're in the FAULTED state or have experienced failed I/O, then
2363 * clear the persistent state and attempt to reopen the device. We
2364 * also mark the vdev config dirty, so that the new faulted state is
2365 * written out to disk.
2367 if (vd->vdev_faulted || vd->vdev_degraded ||
2368 !vdev_readable(vd) || !vdev_writeable(vd)) {
2371 * When reopening in reponse to a clear event, it may be due to
2372 * a fmadm repair request. In this case, if the device is
2373 * still broken, we want to still post the ereport again.
2375 vd->vdev_forcefault = B_TRUE;
2377 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2378 vd->vdev_cant_read = B_FALSE;
2379 vd->vdev_cant_write = B_FALSE;
2381 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2383 vd->vdev_forcefault = B_FALSE;
2385 if (vd != rvd && vdev_writeable(vd->vdev_top))
2386 vdev_state_dirty(vd->vdev_top);
2388 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2389 spa_async_request(spa, SPA_ASYNC_RESILVER);
2391 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2395 * When clearing a FMA-diagnosed fault, we always want to
2396 * unspare the device, as we assume that the original spare was
2397 * done in response to the FMA fault.
2399 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2400 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2401 vd->vdev_parent->vdev_child[0] == vd)
2402 vd->vdev_unspare = B_TRUE;
2406 vdev_is_dead(vdev_t *vd)
2409 * Holes and missing devices are always considered "dead".
2410 * This simplifies the code since we don't have to check for
2411 * these types of devices in the various code paths.
2412 * Instead we rely on the fact that we skip over dead devices
2413 * before issuing I/O to them.
2415 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2416 vd->vdev_ops == &vdev_missing_ops);
2420 vdev_readable(vdev_t *vd)
2422 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2426 vdev_writeable(vdev_t *vd)
2428 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2432 vdev_allocatable(vdev_t *vd)
2434 uint64_t state = vd->vdev_state;
2437 * We currently allow allocations from vdevs which may be in the
2438 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2439 * fails to reopen then we'll catch it later when we're holding
2440 * the proper locks. Note that we have to get the vdev state
2441 * in a local variable because although it changes atomically,
2442 * we're asking two separate questions about it.
2444 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2445 !vd->vdev_cant_write && !vd->vdev_ishole);
2449 vdev_accessible(vdev_t *vd, zio_t *zio)
2451 ASSERT(zio->io_vd == vd);
2453 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2456 if (zio->io_type == ZIO_TYPE_READ)
2457 return (!vd->vdev_cant_read);
2459 if (zio->io_type == ZIO_TYPE_WRITE)
2460 return (!vd->vdev_cant_write);
2466 * Get statistics for the given vdev.
2469 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2471 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2474 mutex_enter(&vd->vdev_stat_lock);
2475 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2476 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2477 vs->vs_state = vd->vdev_state;
2478 vs->vs_rsize = vdev_get_min_asize(vd);
2479 if (vd->vdev_ops->vdev_op_leaf)
2480 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2481 mutex_exit(&vd->vdev_stat_lock);
2484 * If we're getting stats on the root vdev, aggregate the I/O counts
2485 * over all top-level vdevs (i.e. the direct children of the root).
2488 for (c = 0; c < rvd->vdev_children; c++) {
2489 vdev_t *cvd = rvd->vdev_child[c];
2490 vdev_stat_t *cvs = &cvd->vdev_stat;
2492 mutex_enter(&vd->vdev_stat_lock);
2493 for (t = 0; t < ZIO_TYPES; t++) {
2494 vs->vs_ops[t] += cvs->vs_ops[t];
2495 vs->vs_bytes[t] += cvs->vs_bytes[t];
2497 cvs->vs_scan_removing = cvd->vdev_removing;
2498 mutex_exit(&vd->vdev_stat_lock);
2504 vdev_clear_stats(vdev_t *vd)
2506 mutex_enter(&vd->vdev_stat_lock);
2507 vd->vdev_stat.vs_space = 0;
2508 vd->vdev_stat.vs_dspace = 0;
2509 vd->vdev_stat.vs_alloc = 0;
2510 mutex_exit(&vd->vdev_stat_lock);
2514 vdev_scan_stat_init(vdev_t *vd)
2516 vdev_stat_t *vs = &vd->vdev_stat;
2519 for (c = 0; c < vd->vdev_children; c++)
2520 vdev_scan_stat_init(vd->vdev_child[c]);
2522 mutex_enter(&vd->vdev_stat_lock);
2523 vs->vs_scan_processed = 0;
2524 mutex_exit(&vd->vdev_stat_lock);
2528 vdev_stat_update(zio_t *zio, uint64_t psize)
2530 spa_t *spa = zio->io_spa;
2531 vdev_t *rvd = spa->spa_root_vdev;
2532 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2534 uint64_t txg = zio->io_txg;
2535 vdev_stat_t *vs = &vd->vdev_stat;
2536 zio_type_t type = zio->io_type;
2537 int flags = zio->io_flags;
2540 * If this i/o is a gang leader, it didn't do any actual work.
2542 if (zio->io_gang_tree)
2545 if (zio->io_error == 0) {
2547 * If this is a root i/o, don't count it -- we've already
2548 * counted the top-level vdevs, and vdev_get_stats() will
2549 * aggregate them when asked. This reduces contention on
2550 * the root vdev_stat_lock and implicitly handles blocks
2551 * that compress away to holes, for which there is no i/o.
2552 * (Holes never create vdev children, so all the counters
2553 * remain zero, which is what we want.)
2555 * Note: this only applies to successful i/o (io_error == 0)
2556 * because unlike i/o counts, errors are not additive.
2557 * When reading a ditto block, for example, failure of
2558 * one top-level vdev does not imply a root-level error.
2563 ASSERT(vd == zio->io_vd);
2565 if (flags & ZIO_FLAG_IO_BYPASS)
2568 mutex_enter(&vd->vdev_stat_lock);
2570 if (flags & ZIO_FLAG_IO_REPAIR) {
2571 if (flags & ZIO_FLAG_SCAN_THREAD) {
2572 dsl_scan_phys_t *scn_phys =
2573 &spa->spa_dsl_pool->dp_scan->scn_phys;
2574 uint64_t *processed = &scn_phys->scn_processed;
2577 if (vd->vdev_ops->vdev_op_leaf)
2578 atomic_add_64(processed, psize);
2579 vs->vs_scan_processed += psize;
2582 if (flags & ZIO_FLAG_SELF_HEAL)
2583 vs->vs_self_healed += psize;
2587 vs->vs_bytes[type] += psize;
2589 mutex_exit(&vd->vdev_stat_lock);
2593 if (flags & ZIO_FLAG_SPECULATIVE)
2597 * If this is an I/O error that is going to be retried, then ignore the
2598 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2599 * hard errors, when in reality they can happen for any number of
2600 * innocuous reasons (bus resets, MPxIO link failure, etc).
2602 if (zio->io_error == EIO &&
2603 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2607 * Intent logs writes won't propagate their error to the root
2608 * I/O so don't mark these types of failures as pool-level
2611 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2614 mutex_enter(&vd->vdev_stat_lock);
2615 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2616 if (zio->io_error == ECKSUM)
2617 vs->vs_checksum_errors++;
2619 vs->vs_read_errors++;
2621 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2622 vs->vs_write_errors++;
2623 mutex_exit(&vd->vdev_stat_lock);
2625 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2626 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2627 (flags & ZIO_FLAG_SCAN_THREAD) ||
2628 spa->spa_claiming)) {
2630 * This is either a normal write (not a repair), or it's
2631 * a repair induced by the scrub thread, or it's a repair
2632 * made by zil_claim() during spa_load() in the first txg.
2633 * In the normal case, we commit the DTL change in the same
2634 * txg as the block was born. In the scrub-induced repair
2635 * case, we know that scrubs run in first-pass syncing context,
2636 * so we commit the DTL change in spa_syncing_txg(spa).
2637 * In the zil_claim() case, we commit in spa_first_txg(spa).
2639 * We currently do not make DTL entries for failed spontaneous
2640 * self-healing writes triggered by normal (non-scrubbing)
2641 * reads, because we have no transactional context in which to
2642 * do so -- and it's not clear that it'd be desirable anyway.
2644 if (vd->vdev_ops->vdev_op_leaf) {
2645 uint64_t commit_txg = txg;
2646 if (flags & ZIO_FLAG_SCAN_THREAD) {
2647 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2648 ASSERT(spa_sync_pass(spa) == 1);
2649 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2650 commit_txg = spa_syncing_txg(spa);
2651 } else if (spa->spa_claiming) {
2652 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2653 commit_txg = spa_first_txg(spa);
2655 ASSERT(commit_txg >= spa_syncing_txg(spa));
2656 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2658 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2659 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2660 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2663 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2668 * Update the in-core space usage stats for this vdev, its metaslab class,
2669 * and the root vdev.
2672 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2673 int64_t space_delta)
2675 int64_t dspace_delta = space_delta;
2676 spa_t *spa = vd->vdev_spa;
2677 vdev_t *rvd = spa->spa_root_vdev;
2678 metaslab_group_t *mg = vd->vdev_mg;
2679 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2681 ASSERT(vd == vd->vdev_top);
2684 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2685 * factor. We must calculate this here and not at the root vdev
2686 * because the root vdev's psize-to-asize is simply the max of its
2687 * childrens', thus not accurate enough for us.
2689 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2690 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2691 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2692 vd->vdev_deflate_ratio;
2694 mutex_enter(&vd->vdev_stat_lock);
2695 vd->vdev_stat.vs_alloc += alloc_delta;
2696 vd->vdev_stat.vs_space += space_delta;
2697 vd->vdev_stat.vs_dspace += dspace_delta;
2698 mutex_exit(&vd->vdev_stat_lock);
2700 if (mc == spa_normal_class(spa)) {
2701 mutex_enter(&rvd->vdev_stat_lock);
2702 rvd->vdev_stat.vs_alloc += alloc_delta;
2703 rvd->vdev_stat.vs_space += space_delta;
2704 rvd->vdev_stat.vs_dspace += dspace_delta;
2705 mutex_exit(&rvd->vdev_stat_lock);
2709 ASSERT(rvd == vd->vdev_parent);
2710 ASSERT(vd->vdev_ms_count != 0);
2712 metaslab_class_space_update(mc,
2713 alloc_delta, defer_delta, space_delta, dspace_delta);
2718 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2719 * so that it will be written out next time the vdev configuration is synced.
2720 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2723 vdev_config_dirty(vdev_t *vd)
2725 spa_t *spa = vd->vdev_spa;
2726 vdev_t *rvd = spa->spa_root_vdev;
2729 ASSERT(spa_writeable(spa));
2732 * If this is an aux vdev (as with l2cache and spare devices), then we
2733 * update the vdev config manually and set the sync flag.
2735 if (vd->vdev_aux != NULL) {
2736 spa_aux_vdev_t *sav = vd->vdev_aux;
2740 for (c = 0; c < sav->sav_count; c++) {
2741 if (sav->sav_vdevs[c] == vd)
2745 if (c == sav->sav_count) {
2747 * We're being removed. There's nothing more to do.
2749 ASSERT(sav->sav_sync == B_TRUE);
2753 sav->sav_sync = B_TRUE;
2755 if (nvlist_lookup_nvlist_array(sav->sav_config,
2756 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2757 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2758 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2764 * Setting the nvlist in the middle if the array is a little
2765 * sketchy, but it will work.
2767 nvlist_free(aux[c]);
2768 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2774 * The dirty list is protected by the SCL_CONFIG lock. The caller
2775 * must either hold SCL_CONFIG as writer, or must be the sync thread
2776 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2777 * so this is sufficient to ensure mutual exclusion.
2779 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2780 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2781 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2784 for (c = 0; c < rvd->vdev_children; c++)
2785 vdev_config_dirty(rvd->vdev_child[c]);
2787 ASSERT(vd == vd->vdev_top);
2789 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2791 list_insert_head(&spa->spa_config_dirty_list, vd);
2796 vdev_config_clean(vdev_t *vd)
2798 spa_t *spa = vd->vdev_spa;
2800 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2801 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2802 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2804 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2805 list_remove(&spa->spa_config_dirty_list, vd);
2809 * Mark a top-level vdev's state as dirty, so that the next pass of
2810 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2811 * the state changes from larger config changes because they require
2812 * much less locking, and are often needed for administrative actions.
2815 vdev_state_dirty(vdev_t *vd)
2817 spa_t *spa = vd->vdev_spa;
2819 ASSERT(spa_writeable(spa));
2820 ASSERT(vd == vd->vdev_top);
2823 * The state list is protected by the SCL_STATE lock. The caller
2824 * must either hold SCL_STATE as writer, or must be the sync thread
2825 * (which holds SCL_STATE as reader). There's only one sync thread,
2826 * so this is sufficient to ensure mutual exclusion.
2828 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2829 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2830 spa_config_held(spa, SCL_STATE, RW_READER)));
2832 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2833 list_insert_head(&spa->spa_state_dirty_list, vd);
2837 vdev_state_clean(vdev_t *vd)
2839 spa_t *spa = vd->vdev_spa;
2841 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2842 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2843 spa_config_held(spa, SCL_STATE, RW_READER)));
2845 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2846 list_remove(&spa->spa_state_dirty_list, vd);
2850 * Propagate vdev state up from children to parent.
2853 vdev_propagate_state(vdev_t *vd)
2855 spa_t *spa = vd->vdev_spa;
2856 vdev_t *rvd = spa->spa_root_vdev;
2857 int degraded = 0, faulted = 0;
2862 if (vd->vdev_children > 0) {
2863 for (c = 0; c < vd->vdev_children; c++) {
2864 child = vd->vdev_child[c];
2867 * Don't factor holes into the decision.
2869 if (child->vdev_ishole)
2872 if (!vdev_readable(child) ||
2873 (!vdev_writeable(child) && spa_writeable(spa))) {
2875 * Root special: if there is a top-level log
2876 * device, treat the root vdev as if it were
2879 if (child->vdev_islog && vd == rvd)
2883 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2887 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2891 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2894 * Root special: if there is a top-level vdev that cannot be
2895 * opened due to corrupted metadata, then propagate the root
2896 * vdev's aux state as 'corrupt' rather than 'insufficient
2899 if (corrupted && vd == rvd &&
2900 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2901 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2902 VDEV_AUX_CORRUPT_DATA);
2905 if (vd->vdev_parent)
2906 vdev_propagate_state(vd->vdev_parent);
2910 * Set a vdev's state. If this is during an open, we don't update the parent
2911 * state, because we're in the process of opening children depth-first.
2912 * Otherwise, we propagate the change to the parent.
2914 * If this routine places a device in a faulted state, an appropriate ereport is
2918 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2920 uint64_t save_state;
2921 spa_t *spa = vd->vdev_spa;
2923 if (state == vd->vdev_state) {
2924 vd->vdev_stat.vs_aux = aux;
2928 save_state = vd->vdev_state;
2930 vd->vdev_state = state;
2931 vd->vdev_stat.vs_aux = aux;
2934 * If we are setting the vdev state to anything but an open state, then
2935 * always close the underlying device unless the device has requested
2936 * a delayed close (i.e. we're about to remove or fault the device).
2937 * Otherwise, we keep accessible but invalid devices open forever.
2938 * We don't call vdev_close() itself, because that implies some extra
2939 * checks (offline, etc) that we don't want here. This is limited to
2940 * leaf devices, because otherwise closing the device will affect other
2943 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2944 vd->vdev_ops->vdev_op_leaf)
2945 vd->vdev_ops->vdev_op_close(vd);
2948 * If we have brought this vdev back into service, we need
2949 * to notify fmd so that it can gracefully repair any outstanding
2950 * cases due to a missing device. We do this in all cases, even those
2951 * that probably don't correlate to a repaired fault. This is sure to
2952 * catch all cases, and we let the zfs-retire agent sort it out. If
2953 * this is a transient state it's OK, as the retire agent will
2954 * double-check the state of the vdev before repairing it.
2956 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2957 vd->vdev_prevstate != state)
2958 zfs_post_state_change(spa, vd);
2960 if (vd->vdev_removed &&
2961 state == VDEV_STATE_CANT_OPEN &&
2962 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2964 * If the previous state is set to VDEV_STATE_REMOVED, then this
2965 * device was previously marked removed and someone attempted to
2966 * reopen it. If this failed due to a nonexistent device, then
2967 * keep the device in the REMOVED state. We also let this be if
2968 * it is one of our special test online cases, which is only
2969 * attempting to online the device and shouldn't generate an FMA
2972 vd->vdev_state = VDEV_STATE_REMOVED;
2973 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2974 } else if (state == VDEV_STATE_REMOVED) {
2975 vd->vdev_removed = B_TRUE;
2976 } else if (state == VDEV_STATE_CANT_OPEN) {
2978 * If we fail to open a vdev during an import or recovery, we
2979 * mark it as "not available", which signifies that it was
2980 * never there to begin with. Failure to open such a device
2981 * is not considered an error.
2983 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2984 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2985 vd->vdev_ops->vdev_op_leaf)
2986 vd->vdev_not_present = 1;
2989 * Post the appropriate ereport. If the 'prevstate' field is
2990 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2991 * that this is part of a vdev_reopen(). In this case, we don't
2992 * want to post the ereport if the device was already in the
2993 * CANT_OPEN state beforehand.
2995 * If the 'checkremove' flag is set, then this is an attempt to
2996 * online the device in response to an insertion event. If we
2997 * hit this case, then we have detected an insertion event for a
2998 * faulted or offline device that wasn't in the removed state.
2999 * In this scenario, we don't post an ereport because we are
3000 * about to replace the device, or attempt an online with
3001 * vdev_forcefault, which will generate the fault for us.
3003 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3004 !vd->vdev_not_present && !vd->vdev_checkremove &&
3005 vd != spa->spa_root_vdev) {
3009 case VDEV_AUX_OPEN_FAILED:
3010 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3012 case VDEV_AUX_CORRUPT_DATA:
3013 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3015 case VDEV_AUX_NO_REPLICAS:
3016 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3018 case VDEV_AUX_BAD_GUID_SUM:
3019 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3021 case VDEV_AUX_TOO_SMALL:
3022 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3024 case VDEV_AUX_BAD_LABEL:
3025 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3028 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3031 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3034 /* Erase any notion of persistent removed state */
3035 vd->vdev_removed = B_FALSE;
3037 vd->vdev_removed = B_FALSE;
3040 if (!isopen && vd->vdev_parent)
3041 vdev_propagate_state(vd->vdev_parent);
3045 * Check the vdev configuration to ensure that it's capable of supporting
3046 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3047 * In addition, only a single top-level vdev is allowed and none of the leaves
3048 * can be wholedisks.
3051 vdev_is_bootable(vdev_t *vd)
3055 if (!vd->vdev_ops->vdev_op_leaf) {
3056 char *vdev_type = vd->vdev_ops->vdev_op_type;
3058 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3059 vd->vdev_children > 1) {
3061 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3062 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3065 } else if (vd->vdev_wholedisk == 1) {
3069 for (c = 0; c < vd->vdev_children; c++) {
3070 if (!vdev_is_bootable(vd->vdev_child[c]))
3077 * Load the state from the original vdev tree (ovd) which
3078 * we've retrieved from the MOS config object. If the original
3079 * vdev was offline or faulted then we transfer that state to the
3080 * device in the current vdev tree (nvd).
3083 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3087 ASSERT(nvd->vdev_top->vdev_islog);
3088 ASSERT(spa_config_held(nvd->vdev_spa,
3089 SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3090 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3092 for (c = 0; c < nvd->vdev_children; c++)
3093 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3095 if (nvd->vdev_ops->vdev_op_leaf) {
3097 * Restore the persistent vdev state
3099 nvd->vdev_offline = ovd->vdev_offline;
3100 nvd->vdev_faulted = ovd->vdev_faulted;
3101 nvd->vdev_degraded = ovd->vdev_degraded;
3102 nvd->vdev_removed = ovd->vdev_removed;
3107 * Determine if a log device has valid content. If the vdev was
3108 * removed or faulted in the MOS config then we know that
3109 * the content on the log device has already been written to the pool.
3112 vdev_log_state_valid(vdev_t *vd)
3116 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3120 for (c = 0; c < vd->vdev_children; c++)
3121 if (vdev_log_state_valid(vd->vdev_child[c]))
3128 * Expand a vdev if possible.
3131 vdev_expand(vdev_t *vd, uint64_t txg)
3133 ASSERT(vd->vdev_top == vd);
3134 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3136 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3137 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3138 vdev_config_dirty(vd);
3146 vdev_split(vdev_t *vd)
3148 vdev_t *cvd, *pvd = vd->vdev_parent;
3150 vdev_remove_child(pvd, vd);
3151 vdev_compact_children(pvd);
3153 cvd = pvd->vdev_child[0];
3154 if (pvd->vdev_children == 1) {
3155 vdev_remove_parent(cvd);
3156 cvd->vdev_splitting = B_TRUE;
3158 vdev_propagate_state(cvd);