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.
24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 #include <sys/zfs_context.h>
31 #include <sys/fm/fs/zfs.h>
33 #include <sys/spa_impl.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/uberblock_impl.h>
38 #include <sys/metaslab.h>
39 #include <sys/metaslab_impl.h>
40 #include <sys/space_map.h>
41 #include <sys/space_reftree.h>
44 #include <sys/fs/zfs.h>
47 #include <sys/dsl_scan.h>
50 #include <sys/zfs_ratelimit.h>
53 * When a vdev is added, it will be divided into approximately (but no
54 * more than) this number of metaslabs.
56 int metaslabs_per_vdev = 200;
59 * Virtual device management.
62 static vdev_ops_t *vdev_ops_table[] = {
76 * Given a vdev type, return the appropriate ops vector.
79 vdev_getops(const char *type)
81 vdev_ops_t *ops, **opspp;
83 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
84 if (strcmp(ops->vdev_op_type, type) == 0)
91 * Default asize function: return the MAX of psize with the asize of
92 * all children. This is what's used by anything other than RAID-Z.
95 vdev_default_asize(vdev_t *vd, uint64_t psize)
97 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
101 for (c = 0; c < vd->vdev_children; c++) {
102 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
103 asize = MAX(asize, csize);
110 * Get the minimum allocatable size. We define the allocatable size as
111 * the vdev's asize rounded to the nearest metaslab. This allows us to
112 * replace or attach devices which don't have the same physical size but
113 * can still satisfy the same number of allocations.
116 vdev_get_min_asize(vdev_t *vd)
118 vdev_t *pvd = vd->vdev_parent;
121 * If our parent is NULL (inactive spare or cache) or is the root,
122 * just return our own asize.
125 return (vd->vdev_asize);
128 * The top-level vdev just returns the allocatable size rounded
129 * to the nearest metaslab.
131 if (vd == vd->vdev_top)
132 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
135 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
136 * so each child must provide at least 1/Nth of its asize.
138 if (pvd->vdev_ops == &vdev_raidz_ops)
139 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
142 return (pvd->vdev_min_asize);
146 vdev_set_min_asize(vdev_t *vd)
149 vd->vdev_min_asize = vdev_get_min_asize(vd);
151 for (c = 0; c < vd->vdev_children; c++)
152 vdev_set_min_asize(vd->vdev_child[c]);
156 vdev_lookup_top(spa_t *spa, uint64_t vdev)
158 vdev_t *rvd = spa->spa_root_vdev;
160 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
162 if (vdev < rvd->vdev_children) {
163 ASSERT(rvd->vdev_child[vdev] != NULL);
164 return (rvd->vdev_child[vdev]);
171 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
176 if (vd->vdev_guid == guid)
179 for (c = 0; c < vd->vdev_children; c++)
180 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
188 vdev_count_leaves_impl(vdev_t *vd)
193 if (vd->vdev_ops->vdev_op_leaf)
196 for (c = 0; c < vd->vdev_children; c++)
197 n += vdev_count_leaves_impl(vd->vdev_child[c]);
203 vdev_count_leaves(spa_t *spa)
205 return (vdev_count_leaves_impl(spa->spa_root_vdev));
209 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
211 size_t oldsize, newsize;
212 uint64_t id = cvd->vdev_id;
215 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
216 ASSERT(cvd->vdev_parent == NULL);
218 cvd->vdev_parent = pvd;
223 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
225 oldsize = pvd->vdev_children * sizeof (vdev_t *);
226 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
227 newsize = pvd->vdev_children * sizeof (vdev_t *);
229 newchild = kmem_alloc(newsize, KM_SLEEP);
230 if (pvd->vdev_child != NULL) {
231 bcopy(pvd->vdev_child, newchild, oldsize);
232 kmem_free(pvd->vdev_child, oldsize);
235 pvd->vdev_child = newchild;
236 pvd->vdev_child[id] = cvd;
238 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
239 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
242 * Walk up all ancestors to update guid sum.
244 for (; pvd != NULL; pvd = pvd->vdev_parent)
245 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
249 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
252 uint_t id = cvd->vdev_id;
254 ASSERT(cvd->vdev_parent == pvd);
259 ASSERT(id < pvd->vdev_children);
260 ASSERT(pvd->vdev_child[id] == cvd);
262 pvd->vdev_child[id] = NULL;
263 cvd->vdev_parent = NULL;
265 for (c = 0; c < pvd->vdev_children; c++)
266 if (pvd->vdev_child[c])
269 if (c == pvd->vdev_children) {
270 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
271 pvd->vdev_child = NULL;
272 pvd->vdev_children = 0;
276 * Walk up all ancestors to update guid sum.
278 for (; pvd != NULL; pvd = pvd->vdev_parent)
279 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
283 * Remove any holes in the child array.
286 vdev_compact_children(vdev_t *pvd)
288 vdev_t **newchild, *cvd;
289 int oldc = pvd->vdev_children;
293 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
295 for (c = newc = 0; c < oldc; c++)
296 if (pvd->vdev_child[c])
299 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP);
301 for (c = newc = 0; c < oldc; c++) {
302 if ((cvd = pvd->vdev_child[c]) != NULL) {
303 newchild[newc] = cvd;
304 cvd->vdev_id = newc++;
308 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
309 pvd->vdev_child = newchild;
310 pvd->vdev_children = newc;
314 * Allocate and minimally initialize a vdev_t.
317 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
322 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
324 if (spa->spa_root_vdev == NULL) {
325 ASSERT(ops == &vdev_root_ops);
326 spa->spa_root_vdev = vd;
327 spa->spa_load_guid = spa_generate_guid(NULL);
330 if (guid == 0 && ops != &vdev_hole_ops) {
331 if (spa->spa_root_vdev == vd) {
333 * The root vdev's guid will also be the pool guid,
334 * which must be unique among all pools.
336 guid = spa_generate_guid(NULL);
339 * Any other vdev's guid must be unique within the pool.
341 guid = spa_generate_guid(spa);
343 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
348 vd->vdev_guid = guid;
349 vd->vdev_guid_sum = guid;
351 vd->vdev_state = VDEV_STATE_CLOSED;
352 vd->vdev_ishole = (ops == &vdev_hole_ops);
355 * Initialize rate limit structs for events. We rate limit ZIO delay
356 * and checksum events so that we don't overwhelm ZED with thousands
357 * of events when a disk is acting up.
359 zfs_ratelimit_init(&vd->vdev_delay_rl, DELAYS_PER_SECOND, 1);
360 zfs_ratelimit_init(&vd->vdev_checksum_rl, CHECKSUMS_PER_SECOND, 1);
362 list_link_init(&vd->vdev_config_dirty_node);
363 list_link_init(&vd->vdev_state_dirty_node);
364 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL);
365 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
366 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
367 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
369 for (t = 0; t < DTL_TYPES; t++) {
370 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
373 txg_list_create(&vd->vdev_ms_list,
374 offsetof(struct metaslab, ms_txg_node));
375 txg_list_create(&vd->vdev_dtl_list,
376 offsetof(struct vdev, vdev_dtl_node));
377 vd->vdev_stat.vs_timestamp = gethrtime();
385 * Allocate a new vdev. The 'alloctype' is used to control whether we are
386 * creating a new vdev or loading an existing one - the behavior is slightly
387 * different for each case.
390 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
395 uint64_t guid = 0, islog, nparity;
398 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
400 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
401 return (SET_ERROR(EINVAL));
403 if ((ops = vdev_getops(type)) == NULL)
404 return (SET_ERROR(EINVAL));
407 * If this is a load, get the vdev guid from the nvlist.
408 * Otherwise, vdev_alloc_common() will generate one for us.
410 if (alloctype == VDEV_ALLOC_LOAD) {
413 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
415 return (SET_ERROR(EINVAL));
417 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
418 return (SET_ERROR(EINVAL));
419 } else if (alloctype == VDEV_ALLOC_SPARE) {
420 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
421 return (SET_ERROR(EINVAL));
422 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
423 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
424 return (SET_ERROR(EINVAL));
425 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
426 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
427 return (SET_ERROR(EINVAL));
431 * The first allocated vdev must be of type 'root'.
433 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
434 return (SET_ERROR(EINVAL));
437 * Determine whether we're a log vdev.
440 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
441 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
442 return (SET_ERROR(ENOTSUP));
444 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
445 return (SET_ERROR(ENOTSUP));
448 * Set the nparity property for RAID-Z vdevs.
451 if (ops == &vdev_raidz_ops) {
452 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
454 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
455 return (SET_ERROR(EINVAL));
457 * Previous versions could only support 1 or 2 parity
461 spa_version(spa) < SPA_VERSION_RAIDZ2)
462 return (SET_ERROR(ENOTSUP));
464 spa_version(spa) < SPA_VERSION_RAIDZ3)
465 return (SET_ERROR(ENOTSUP));
468 * We require the parity to be specified for SPAs that
469 * support multiple parity levels.
471 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
472 return (SET_ERROR(EINVAL));
474 * Otherwise, we default to 1 parity device for RAID-Z.
481 ASSERT(nparity != -1ULL);
483 vd = vdev_alloc_common(spa, id, guid, ops);
485 vd->vdev_islog = islog;
486 vd->vdev_nparity = nparity;
488 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
489 vd->vdev_path = spa_strdup(vd->vdev_path);
490 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
491 vd->vdev_devid = spa_strdup(vd->vdev_devid);
492 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
493 &vd->vdev_physpath) == 0)
494 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
496 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
497 &vd->vdev_enc_sysfs_path) == 0)
498 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path);
500 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
501 vd->vdev_fru = spa_strdup(vd->vdev_fru);
504 * Set the whole_disk property. If it's not specified, leave the value
507 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
508 &vd->vdev_wholedisk) != 0)
509 vd->vdev_wholedisk = -1ULL;
512 * Look for the 'not present' flag. This will only be set if the device
513 * was not present at the time of import.
515 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
516 &vd->vdev_not_present);
519 * Get the alignment requirement.
521 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
524 * Retrieve the vdev creation time.
526 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
530 * If we're a top-level vdev, try to load the allocation parameters.
532 if (parent && !parent->vdev_parent &&
533 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
534 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
536 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
538 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
540 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
542 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
545 ASSERT0(vd->vdev_top_zap);
548 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
549 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
550 alloctype == VDEV_ALLOC_ADD ||
551 alloctype == VDEV_ALLOC_SPLIT ||
552 alloctype == VDEV_ALLOC_ROOTPOOL);
553 vd->vdev_mg = metaslab_group_create(islog ?
554 spa_log_class(spa) : spa_normal_class(spa), vd);
557 if (vd->vdev_ops->vdev_op_leaf &&
558 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
559 (void) nvlist_lookup_uint64(nv,
560 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
562 ASSERT0(vd->vdev_leaf_zap);
566 * If we're a leaf vdev, try to load the DTL object and other state.
569 if (vd->vdev_ops->vdev_op_leaf &&
570 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
571 alloctype == VDEV_ALLOC_ROOTPOOL)) {
572 if (alloctype == VDEV_ALLOC_LOAD) {
573 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
574 &vd->vdev_dtl_object);
575 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
579 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
582 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
583 &spare) == 0 && spare)
587 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
590 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
591 &vd->vdev_resilver_txg);
594 * When importing a pool, we want to ignore the persistent fault
595 * state, as the diagnosis made on another system may not be
596 * valid in the current context. Local vdevs will
597 * remain in the faulted state.
599 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
600 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
602 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
604 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
607 if (vd->vdev_faulted || vd->vdev_degraded) {
611 VDEV_AUX_ERR_EXCEEDED;
612 if (nvlist_lookup_string(nv,
613 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
614 strcmp(aux, "external") == 0)
615 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
621 * Add ourselves to the parent's list of children.
623 vdev_add_child(parent, vd);
631 vdev_free(vdev_t *vd)
634 spa_t *spa = vd->vdev_spa;
637 * vdev_free() implies closing the vdev first. This is simpler than
638 * trying to ensure complicated semantics for all callers.
642 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
643 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
648 for (c = 0; c < vd->vdev_children; c++)
649 vdev_free(vd->vdev_child[c]);
651 ASSERT(vd->vdev_child == NULL);
652 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
655 * Discard allocation state.
657 if (vd->vdev_mg != NULL) {
658 vdev_metaslab_fini(vd);
659 metaslab_group_destroy(vd->vdev_mg);
662 ASSERT0(vd->vdev_stat.vs_space);
663 ASSERT0(vd->vdev_stat.vs_dspace);
664 ASSERT0(vd->vdev_stat.vs_alloc);
667 * Remove this vdev from its parent's child list.
669 vdev_remove_child(vd->vdev_parent, vd);
671 ASSERT(vd->vdev_parent == NULL);
674 * Clean up vdev structure.
680 spa_strfree(vd->vdev_path);
682 spa_strfree(vd->vdev_devid);
683 if (vd->vdev_physpath)
684 spa_strfree(vd->vdev_physpath);
686 if (vd->vdev_enc_sysfs_path)
687 spa_strfree(vd->vdev_enc_sysfs_path);
690 spa_strfree(vd->vdev_fru);
692 if (vd->vdev_isspare)
693 spa_spare_remove(vd);
694 if (vd->vdev_isl2cache)
695 spa_l2cache_remove(vd);
697 txg_list_destroy(&vd->vdev_ms_list);
698 txg_list_destroy(&vd->vdev_dtl_list);
700 mutex_enter(&vd->vdev_dtl_lock);
701 space_map_close(vd->vdev_dtl_sm);
702 for (t = 0; t < DTL_TYPES; t++) {
703 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
704 range_tree_destroy(vd->vdev_dtl[t]);
706 mutex_exit(&vd->vdev_dtl_lock);
708 mutex_destroy(&vd->vdev_queue_lock);
709 mutex_destroy(&vd->vdev_dtl_lock);
710 mutex_destroy(&vd->vdev_stat_lock);
711 mutex_destroy(&vd->vdev_probe_lock);
713 zfs_ratelimit_fini(&vd->vdev_delay_rl);
714 zfs_ratelimit_fini(&vd->vdev_checksum_rl);
716 if (vd == spa->spa_root_vdev)
717 spa->spa_root_vdev = NULL;
719 kmem_free(vd, sizeof (vdev_t));
723 * Transfer top-level vdev state from svd to tvd.
726 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
728 spa_t *spa = svd->vdev_spa;
733 ASSERT(tvd == tvd->vdev_top);
735 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite;
736 tvd->vdev_ms_array = svd->vdev_ms_array;
737 tvd->vdev_ms_shift = svd->vdev_ms_shift;
738 tvd->vdev_ms_count = svd->vdev_ms_count;
739 tvd->vdev_top_zap = svd->vdev_top_zap;
741 svd->vdev_ms_array = 0;
742 svd->vdev_ms_shift = 0;
743 svd->vdev_ms_count = 0;
744 svd->vdev_top_zap = 0;
747 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
748 tvd->vdev_mg = svd->vdev_mg;
749 tvd->vdev_ms = svd->vdev_ms;
754 if (tvd->vdev_mg != NULL)
755 tvd->vdev_mg->mg_vd = tvd;
757 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
758 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
759 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
761 svd->vdev_stat.vs_alloc = 0;
762 svd->vdev_stat.vs_space = 0;
763 svd->vdev_stat.vs_dspace = 0;
765 for (t = 0; t < TXG_SIZE; t++) {
766 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
767 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
768 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
769 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
770 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
771 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
774 if (list_link_active(&svd->vdev_config_dirty_node)) {
775 vdev_config_clean(svd);
776 vdev_config_dirty(tvd);
779 if (list_link_active(&svd->vdev_state_dirty_node)) {
780 vdev_state_clean(svd);
781 vdev_state_dirty(tvd);
784 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
785 svd->vdev_deflate_ratio = 0;
787 tvd->vdev_islog = svd->vdev_islog;
792 vdev_top_update(vdev_t *tvd, vdev_t *vd)
801 for (c = 0; c < vd->vdev_children; c++)
802 vdev_top_update(tvd, vd->vdev_child[c]);
806 * Add a mirror/replacing vdev above an existing vdev.
809 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
811 spa_t *spa = cvd->vdev_spa;
812 vdev_t *pvd = cvd->vdev_parent;
815 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
817 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
819 mvd->vdev_asize = cvd->vdev_asize;
820 mvd->vdev_min_asize = cvd->vdev_min_asize;
821 mvd->vdev_max_asize = cvd->vdev_max_asize;
822 mvd->vdev_ashift = cvd->vdev_ashift;
823 mvd->vdev_state = cvd->vdev_state;
824 mvd->vdev_crtxg = cvd->vdev_crtxg;
826 vdev_remove_child(pvd, cvd);
827 vdev_add_child(pvd, mvd);
828 cvd->vdev_id = mvd->vdev_children;
829 vdev_add_child(mvd, cvd);
830 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
832 if (mvd == mvd->vdev_top)
833 vdev_top_transfer(cvd, mvd);
839 * Remove a 1-way mirror/replacing vdev from the tree.
842 vdev_remove_parent(vdev_t *cvd)
844 vdev_t *mvd = cvd->vdev_parent;
845 vdev_t *pvd = mvd->vdev_parent;
847 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
849 ASSERT(mvd->vdev_children == 1);
850 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
851 mvd->vdev_ops == &vdev_replacing_ops ||
852 mvd->vdev_ops == &vdev_spare_ops);
853 cvd->vdev_ashift = mvd->vdev_ashift;
855 vdev_remove_child(mvd, cvd);
856 vdev_remove_child(pvd, mvd);
859 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
860 * Otherwise, we could have detached an offline device, and when we
861 * go to import the pool we'll think we have two top-level vdevs,
862 * instead of a different version of the same top-level vdev.
864 if (mvd->vdev_top == mvd) {
865 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
866 cvd->vdev_orig_guid = cvd->vdev_guid;
867 cvd->vdev_guid += guid_delta;
868 cvd->vdev_guid_sum += guid_delta;
871 * If pool not set for autoexpand, we need to also preserve
872 * mvd's asize to prevent automatic expansion of cvd.
873 * Otherwise if we are adjusting the mirror by attaching and
874 * detaching children of non-uniform sizes, the mirror could
875 * autoexpand, unexpectedly requiring larger devices to
876 * re-establish the mirror.
878 if (!cvd->vdev_spa->spa_autoexpand)
879 cvd->vdev_asize = mvd->vdev_asize;
881 cvd->vdev_id = mvd->vdev_id;
882 vdev_add_child(pvd, cvd);
883 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
885 if (cvd == cvd->vdev_top)
886 vdev_top_transfer(mvd, cvd);
888 ASSERT(mvd->vdev_children == 0);
893 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
895 spa_t *spa = vd->vdev_spa;
896 objset_t *mos = spa->spa_meta_objset;
898 uint64_t oldc = vd->vdev_ms_count;
899 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
903 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
906 * This vdev is not being allocated from yet or is a hole.
908 if (vd->vdev_ms_shift == 0)
911 ASSERT(!vd->vdev_ishole);
914 * Compute the raidz-deflation ratio. Note, we hard-code
915 * in 128k (1 << 17) because it is the "typical" blocksize.
916 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
917 * otherwise it would inconsistently account for existing bp's.
919 vd->vdev_deflate_ratio = (1 << 17) /
920 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
922 ASSERT(oldc <= newc);
924 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
927 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
928 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
932 vd->vdev_ms_count = newc;
934 for (m = oldc; m < newc; m++) {
938 error = dmu_read(mos, vd->vdev_ms_array,
939 m * sizeof (uint64_t), sizeof (uint64_t), &object,
945 error = metaslab_init(vd->vdev_mg, m, object, txg,
952 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
955 * If the vdev is being removed we don't activate
956 * the metaslabs since we want to ensure that no new
957 * allocations are performed on this device.
959 if (oldc == 0 && !vd->vdev_removing)
960 metaslab_group_activate(vd->vdev_mg);
963 spa_config_exit(spa, SCL_ALLOC, FTAG);
969 vdev_metaslab_fini(vdev_t *vd)
972 uint64_t count = vd->vdev_ms_count;
974 if (vd->vdev_ms != NULL) {
975 metaslab_group_passivate(vd->vdev_mg);
976 for (m = 0; m < count; m++) {
977 metaslab_t *msp = vd->vdev_ms[m];
982 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
986 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
989 typedef struct vdev_probe_stats {
990 boolean_t vps_readable;
991 boolean_t vps_writeable;
993 } vdev_probe_stats_t;
996 vdev_probe_done(zio_t *zio)
998 spa_t *spa = zio->io_spa;
999 vdev_t *vd = zio->io_vd;
1000 vdev_probe_stats_t *vps = zio->io_private;
1002 ASSERT(vd->vdev_probe_zio != NULL);
1004 if (zio->io_type == ZIO_TYPE_READ) {
1005 if (zio->io_error == 0)
1006 vps->vps_readable = 1;
1007 if (zio->io_error == 0 && spa_writeable(spa)) {
1008 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1009 zio->io_offset, zio->io_size, zio->io_abd,
1010 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1011 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1013 abd_free(zio->io_abd);
1015 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1016 if (zio->io_error == 0)
1017 vps->vps_writeable = 1;
1018 abd_free(zio->io_abd);
1019 } else if (zio->io_type == ZIO_TYPE_NULL) {
1023 vd->vdev_cant_read |= !vps->vps_readable;
1024 vd->vdev_cant_write |= !vps->vps_writeable;
1026 if (vdev_readable(vd) &&
1027 (vdev_writeable(vd) || !spa_writeable(spa))) {
1030 ASSERT(zio->io_error != 0);
1031 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1032 spa, vd, NULL, 0, 0);
1033 zio->io_error = SET_ERROR(ENXIO);
1036 mutex_enter(&vd->vdev_probe_lock);
1037 ASSERT(vd->vdev_probe_zio == zio);
1038 vd->vdev_probe_zio = NULL;
1039 mutex_exit(&vd->vdev_probe_lock);
1042 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1043 if (!vdev_accessible(vd, pio))
1044 pio->io_error = SET_ERROR(ENXIO);
1046 kmem_free(vps, sizeof (*vps));
1051 * Determine whether this device is accessible.
1053 * Read and write to several known locations: the pad regions of each
1054 * vdev label but the first, which we leave alone in case it contains
1058 vdev_probe(vdev_t *vd, zio_t *zio)
1060 spa_t *spa = vd->vdev_spa;
1061 vdev_probe_stats_t *vps = NULL;
1065 ASSERT(vd->vdev_ops->vdev_op_leaf);
1068 * Don't probe the probe.
1070 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1074 * To prevent 'probe storms' when a device fails, we create
1075 * just one probe i/o at a time. All zios that want to probe
1076 * this vdev will become parents of the probe io.
1078 mutex_enter(&vd->vdev_probe_lock);
1080 if ((pio = vd->vdev_probe_zio) == NULL) {
1081 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1083 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1084 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1087 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1089 * vdev_cant_read and vdev_cant_write can only
1090 * transition from TRUE to FALSE when we have the
1091 * SCL_ZIO lock as writer; otherwise they can only
1092 * transition from FALSE to TRUE. This ensures that
1093 * any zio looking at these values can assume that
1094 * failures persist for the life of the I/O. That's
1095 * important because when a device has intermittent
1096 * connectivity problems, we want to ensure that
1097 * they're ascribed to the device (ENXIO) and not
1100 * Since we hold SCL_ZIO as writer here, clear both
1101 * values so the probe can reevaluate from first
1104 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1105 vd->vdev_cant_read = B_FALSE;
1106 vd->vdev_cant_write = B_FALSE;
1109 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1110 vdev_probe_done, vps,
1111 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1114 * We can't change the vdev state in this context, so we
1115 * kick off an async task to do it on our behalf.
1118 vd->vdev_probe_wanted = B_TRUE;
1119 spa_async_request(spa, SPA_ASYNC_PROBE);
1124 zio_add_child(zio, pio);
1126 mutex_exit(&vd->vdev_probe_lock);
1129 ASSERT(zio != NULL);
1133 for (l = 1; l < VDEV_LABELS; l++) {
1134 zio_nowait(zio_read_phys(pio, vd,
1135 vdev_label_offset(vd->vdev_psize, l,
1136 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1137 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1138 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1139 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1150 vdev_open_child(void *arg)
1154 vd->vdev_open_thread = curthread;
1155 vd->vdev_open_error = vdev_open(vd);
1156 vd->vdev_open_thread = NULL;
1160 vdev_uses_zvols(vdev_t *vd)
1165 if (zvol_is_zvol(vd->vdev_path))
1169 for (c = 0; c < vd->vdev_children; c++)
1170 if (vdev_uses_zvols(vd->vdev_child[c]))
1177 vdev_open_children(vdev_t *vd)
1180 int children = vd->vdev_children;
1184 * in order to handle pools on top of zvols, do the opens
1185 * in a single thread so that the same thread holds the
1186 * spa_namespace_lock
1188 if (vdev_uses_zvols(vd)) {
1190 for (c = 0; c < children; c++)
1191 vd->vdev_child[c]->vdev_open_error =
1192 vdev_open(vd->vdev_child[c]);
1194 tq = taskq_create("vdev_open", children, minclsyspri,
1195 children, children, TASKQ_PREPOPULATE);
1199 for (c = 0; c < children; c++)
1200 VERIFY(taskq_dispatch(tq, vdev_open_child,
1201 vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID);
1206 vd->vdev_nonrot = B_TRUE;
1208 for (c = 0; c < children; c++)
1209 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1213 * Prepare a virtual device for access.
1216 vdev_open(vdev_t *vd)
1218 spa_t *spa = vd->vdev_spa;
1221 uint64_t max_osize = 0;
1222 uint64_t asize, max_asize, psize;
1223 uint64_t ashift = 0;
1226 ASSERT(vd->vdev_open_thread == curthread ||
1227 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1228 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1229 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1230 vd->vdev_state == VDEV_STATE_OFFLINE);
1232 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1233 vd->vdev_cant_read = B_FALSE;
1234 vd->vdev_cant_write = B_FALSE;
1235 vd->vdev_min_asize = vdev_get_min_asize(vd);
1238 * If this vdev is not removed, check its fault status. If it's
1239 * faulted, bail out of the open.
1241 if (!vd->vdev_removed && vd->vdev_faulted) {
1242 ASSERT(vd->vdev_children == 0);
1243 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1244 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1245 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1246 vd->vdev_label_aux);
1247 return (SET_ERROR(ENXIO));
1248 } else if (vd->vdev_offline) {
1249 ASSERT(vd->vdev_children == 0);
1250 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1251 return (SET_ERROR(ENXIO));
1254 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1257 * Reset the vdev_reopening flag so that we actually close
1258 * the vdev on error.
1260 vd->vdev_reopening = B_FALSE;
1261 if (zio_injection_enabled && error == 0)
1262 error = zio_handle_device_injection(vd, NULL, ENXIO);
1265 if (vd->vdev_removed &&
1266 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1267 vd->vdev_removed = B_FALSE;
1269 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1270 vd->vdev_stat.vs_aux);
1274 vd->vdev_removed = B_FALSE;
1277 * Recheck the faulted flag now that we have confirmed that
1278 * the vdev is accessible. If we're faulted, bail.
1280 if (vd->vdev_faulted) {
1281 ASSERT(vd->vdev_children == 0);
1282 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1283 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1284 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1285 vd->vdev_label_aux);
1286 return (SET_ERROR(ENXIO));
1289 if (vd->vdev_degraded) {
1290 ASSERT(vd->vdev_children == 0);
1291 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1292 VDEV_AUX_ERR_EXCEEDED);
1294 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1298 * For hole or missing vdevs we just return success.
1300 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1303 for (c = 0; c < vd->vdev_children; c++) {
1304 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1305 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1311 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1312 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1314 if (vd->vdev_children == 0) {
1315 if (osize < SPA_MINDEVSIZE) {
1316 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1317 VDEV_AUX_TOO_SMALL);
1318 return (SET_ERROR(EOVERFLOW));
1321 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1322 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1323 VDEV_LABEL_END_SIZE);
1325 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1326 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1327 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1328 VDEV_AUX_TOO_SMALL);
1329 return (SET_ERROR(EOVERFLOW));
1333 max_asize = max_osize;
1337 * If the vdev was expanded, record this so that we can re-create the
1338 * uberblock rings in labels {2,3}, during the next sync.
1340 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0))
1341 vd->vdev_copy_uberblocks = B_TRUE;
1343 vd->vdev_psize = psize;
1346 * Make sure the allocatable size hasn't shrunk too much.
1348 if (asize < vd->vdev_min_asize) {
1349 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1350 VDEV_AUX_BAD_LABEL);
1351 return (SET_ERROR(EINVAL));
1354 if (vd->vdev_asize == 0) {
1356 * This is the first-ever open, so use the computed values.
1357 * For compatibility, a different ashift can be requested.
1359 vd->vdev_asize = asize;
1360 vd->vdev_max_asize = max_asize;
1361 if (vd->vdev_ashift == 0) {
1362 vd->vdev_ashift = ashift; /* use detected value */
1364 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
1365 vd->vdev_ashift > ASHIFT_MAX)) {
1366 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1367 VDEV_AUX_BAD_ASHIFT);
1368 return (SET_ERROR(EDOM));
1372 * Detect if the alignment requirement has increased.
1373 * We don't want to make the pool unavailable, just
1374 * post an event instead.
1376 if (ashift > vd->vdev_top->vdev_ashift &&
1377 vd->vdev_ops->vdev_op_leaf) {
1378 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
1379 spa, vd, NULL, 0, 0);
1382 vd->vdev_max_asize = max_asize;
1386 * If all children are healthy we update asize if either:
1387 * The asize has increased, due to a device expansion caused by dynamic
1388 * LUN growth or vdev replacement, and automatic expansion is enabled;
1389 * making the additional space available.
1391 * The asize has decreased, due to a device shrink usually caused by a
1392 * vdev replace with a smaller device. This ensures that calculations
1393 * based of max_asize and asize e.g. esize are always valid. It's safe
1394 * to do this as we've already validated that asize is greater than
1397 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1398 ((asize > vd->vdev_asize &&
1399 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1400 (asize < vd->vdev_asize)))
1401 vd->vdev_asize = asize;
1403 vdev_set_min_asize(vd);
1406 * Ensure we can issue some IO before declaring the
1407 * vdev open for business.
1409 if (vd->vdev_ops->vdev_op_leaf &&
1410 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1411 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1412 VDEV_AUX_ERR_EXCEEDED);
1417 * Track the min and max ashift values for normal data devices.
1419 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1420 !vd->vdev_islog && vd->vdev_aux == NULL) {
1421 if (vd->vdev_ashift > spa->spa_max_ashift)
1422 spa->spa_max_ashift = vd->vdev_ashift;
1423 if (vd->vdev_ashift < spa->spa_min_ashift)
1424 spa->spa_min_ashift = vd->vdev_ashift;
1428 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1429 * resilver. But don't do this if we are doing a reopen for a scrub,
1430 * since this would just restart the scrub we are already doing.
1432 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1433 vdev_resilver_needed(vd, NULL, NULL))
1434 spa_async_request(spa, SPA_ASYNC_RESILVER);
1440 * Called once the vdevs are all opened, this routine validates the label
1441 * contents. This needs to be done before vdev_load() so that we don't
1442 * inadvertently do repair I/Os to the wrong device.
1444 * If 'strict' is false ignore the spa guid check. This is necessary because
1445 * if the machine crashed during a re-guid the new guid might have been written
1446 * to all of the vdev labels, but not the cached config. The strict check
1447 * will be performed when the pool is opened again using the mos config.
1449 * This function will only return failure if one of the vdevs indicates that it
1450 * has since been destroyed or exported. This is only possible if
1451 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1452 * will be updated but the function will return 0.
1455 vdev_validate(vdev_t *vd, boolean_t strict)
1457 spa_t *spa = vd->vdev_spa;
1459 uint64_t guid = 0, top_guid;
1463 for (c = 0; c < vd->vdev_children; c++)
1464 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1465 return (SET_ERROR(EBADF));
1468 * If the device has already failed, or was marked offline, don't do
1469 * any further validation. Otherwise, label I/O will fail and we will
1470 * overwrite the previous state.
1472 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1473 uint64_t aux_guid = 0;
1475 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1476 spa_last_synced_txg(spa) : -1ULL;
1478 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1479 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1480 VDEV_AUX_BAD_LABEL);
1485 * Determine if this vdev has been split off into another
1486 * pool. If so, then refuse to open it.
1488 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1489 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1490 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1491 VDEV_AUX_SPLIT_POOL);
1496 if (strict && (nvlist_lookup_uint64(label,
1497 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1498 guid != spa_guid(spa))) {
1499 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1500 VDEV_AUX_CORRUPT_DATA);
1505 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1506 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1511 * If this vdev just became a top-level vdev because its
1512 * sibling was detached, it will have adopted the parent's
1513 * vdev guid -- but the label may or may not be on disk yet.
1514 * Fortunately, either version of the label will have the
1515 * same top guid, so if we're a top-level vdev, we can
1516 * safely compare to that instead.
1518 * If we split this vdev off instead, then we also check the
1519 * original pool's guid. We don't want to consider the vdev
1520 * corrupt if it is partway through a split operation.
1522 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1524 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1526 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1527 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1528 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1529 VDEV_AUX_CORRUPT_DATA);
1534 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1536 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1537 VDEV_AUX_CORRUPT_DATA);
1545 * If this is a verbatim import, no need to check the
1546 * state of the pool.
1548 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1549 spa_load_state(spa) == SPA_LOAD_OPEN &&
1550 state != POOL_STATE_ACTIVE)
1551 return (SET_ERROR(EBADF));
1554 * If we were able to open and validate a vdev that was
1555 * previously marked permanently unavailable, clear that state
1558 if (vd->vdev_not_present)
1559 vd->vdev_not_present = 0;
1566 * Close a virtual device.
1569 vdev_close(vdev_t *vd)
1571 vdev_t *pvd = vd->vdev_parent;
1572 ASSERTV(spa_t *spa = vd->vdev_spa);
1574 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1577 * If our parent is reopening, then we are as well, unless we are
1580 if (pvd != NULL && pvd->vdev_reopening)
1581 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1583 vd->vdev_ops->vdev_op_close(vd);
1585 vdev_cache_purge(vd);
1588 * We record the previous state before we close it, so that if we are
1589 * doing a reopen(), we don't generate FMA ereports if we notice that
1590 * it's still faulted.
1592 vd->vdev_prevstate = vd->vdev_state;
1594 if (vd->vdev_offline)
1595 vd->vdev_state = VDEV_STATE_OFFLINE;
1597 vd->vdev_state = VDEV_STATE_CLOSED;
1598 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1602 vdev_hold(vdev_t *vd)
1604 spa_t *spa = vd->vdev_spa;
1607 ASSERT(spa_is_root(spa));
1608 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1611 for (c = 0; c < vd->vdev_children; c++)
1612 vdev_hold(vd->vdev_child[c]);
1614 if (vd->vdev_ops->vdev_op_leaf)
1615 vd->vdev_ops->vdev_op_hold(vd);
1619 vdev_rele(vdev_t *vd)
1623 ASSERT(spa_is_root(vd->vdev_spa));
1624 for (c = 0; c < vd->vdev_children; c++)
1625 vdev_rele(vd->vdev_child[c]);
1627 if (vd->vdev_ops->vdev_op_leaf)
1628 vd->vdev_ops->vdev_op_rele(vd);
1632 * Reopen all interior vdevs and any unopened leaves. We don't actually
1633 * reopen leaf vdevs which had previously been opened as they might deadlock
1634 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1635 * If the leaf has never been opened then open it, as usual.
1638 vdev_reopen(vdev_t *vd)
1640 spa_t *spa = vd->vdev_spa;
1642 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1644 /* set the reopening flag unless we're taking the vdev offline */
1645 vd->vdev_reopening = !vd->vdev_offline;
1647 (void) vdev_open(vd);
1650 * Call vdev_validate() here to make sure we have the same device.
1651 * Otherwise, a device with an invalid label could be successfully
1652 * opened in response to vdev_reopen().
1655 (void) vdev_validate_aux(vd);
1656 if (vdev_readable(vd) && vdev_writeable(vd) &&
1657 vd->vdev_aux == &spa->spa_l2cache &&
1658 !l2arc_vdev_present(vd))
1659 l2arc_add_vdev(spa, vd);
1661 (void) vdev_validate(vd, B_TRUE);
1665 * Reassess parent vdev's health.
1667 vdev_propagate_state(vd);
1671 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1676 * Normally, partial opens (e.g. of a mirror) are allowed.
1677 * For a create, however, we want to fail the request if
1678 * there are any components we can't open.
1680 error = vdev_open(vd);
1682 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1684 return (error ? error : ENXIO);
1688 * Recursively load DTLs and initialize all labels.
1690 if ((error = vdev_dtl_load(vd)) != 0 ||
1691 (error = vdev_label_init(vd, txg, isreplacing ?
1692 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1701 vdev_metaslab_set_size(vdev_t *vd)
1704 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1706 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1707 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1711 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1713 ASSERT(vd == vd->vdev_top);
1714 ASSERT(!vd->vdev_ishole);
1715 ASSERT(ISP2(flags));
1716 ASSERT(spa_writeable(vd->vdev_spa));
1718 if (flags & VDD_METASLAB)
1719 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1721 if (flags & VDD_DTL)
1722 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1724 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1728 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1732 for (c = 0; c < vd->vdev_children; c++)
1733 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1735 if (vd->vdev_ops->vdev_op_leaf)
1736 vdev_dirty(vd->vdev_top, flags, vd, txg);
1742 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1743 * the vdev has less than perfect replication. There are four kinds of DTL:
1745 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1747 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1749 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1750 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1751 * txgs that was scrubbed.
1753 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1754 * persistent errors or just some device being offline.
1755 * Unlike the other three, the DTL_OUTAGE map is not generally
1756 * maintained; it's only computed when needed, typically to
1757 * determine whether a device can be detached.
1759 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1760 * either has the data or it doesn't.
1762 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1763 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1764 * if any child is less than fully replicated, then so is its parent.
1765 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1766 * comprising only those txgs which appear in 'maxfaults' or more children;
1767 * those are the txgs we don't have enough replication to read. For example,
1768 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1769 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1770 * two child DTL_MISSING maps.
1772 * It should be clear from the above that to compute the DTLs and outage maps
1773 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1774 * Therefore, that is all we keep on disk. When loading the pool, or after
1775 * a configuration change, we generate all other DTLs from first principles.
1778 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1780 range_tree_t *rt = vd->vdev_dtl[t];
1782 ASSERT(t < DTL_TYPES);
1783 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1784 ASSERT(spa_writeable(vd->vdev_spa));
1786 mutex_enter(rt->rt_lock);
1787 if (!range_tree_contains(rt, txg, size))
1788 range_tree_add(rt, txg, size);
1789 mutex_exit(rt->rt_lock);
1793 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1795 range_tree_t *rt = vd->vdev_dtl[t];
1796 boolean_t dirty = B_FALSE;
1798 ASSERT(t < DTL_TYPES);
1799 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1801 mutex_enter(rt->rt_lock);
1802 if (range_tree_space(rt) != 0)
1803 dirty = range_tree_contains(rt, txg, size);
1804 mutex_exit(rt->rt_lock);
1810 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1812 range_tree_t *rt = vd->vdev_dtl[t];
1815 mutex_enter(rt->rt_lock);
1816 empty = (range_tree_space(rt) == 0);
1817 mutex_exit(rt->rt_lock);
1823 * Returns the lowest txg in the DTL range.
1826 vdev_dtl_min(vdev_t *vd)
1830 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1831 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1832 ASSERT0(vd->vdev_children);
1834 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1835 return (rs->rs_start - 1);
1839 * Returns the highest txg in the DTL.
1842 vdev_dtl_max(vdev_t *vd)
1846 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1847 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1848 ASSERT0(vd->vdev_children);
1850 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1851 return (rs->rs_end);
1855 * Determine if a resilvering vdev should remove any DTL entries from
1856 * its range. If the vdev was resilvering for the entire duration of the
1857 * scan then it should excise that range from its DTLs. Otherwise, this
1858 * vdev is considered partially resilvered and should leave its DTL
1859 * entries intact. The comment in vdev_dtl_reassess() describes how we
1863 vdev_dtl_should_excise(vdev_t *vd)
1865 spa_t *spa = vd->vdev_spa;
1866 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1868 ASSERT0(scn->scn_phys.scn_errors);
1869 ASSERT0(vd->vdev_children);
1871 if (vd->vdev_state < VDEV_STATE_DEGRADED)
1874 if (vd->vdev_resilver_txg == 0 ||
1875 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1879 * When a resilver is initiated the scan will assign the scn_max_txg
1880 * value to the highest txg value that exists in all DTLs. If this
1881 * device's max DTL is not part of this scan (i.e. it is not in
1882 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1885 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1886 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1887 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1888 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1895 * Reassess DTLs after a config change or scrub completion.
1898 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1900 spa_t *spa = vd->vdev_spa;
1904 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1906 for (c = 0; c < vd->vdev_children; c++)
1907 vdev_dtl_reassess(vd->vdev_child[c], txg,
1908 scrub_txg, scrub_done);
1910 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1913 if (vd->vdev_ops->vdev_op_leaf) {
1914 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1916 mutex_enter(&vd->vdev_dtl_lock);
1919 * If we've completed a scan cleanly then determine
1920 * if this vdev should remove any DTLs. We only want to
1921 * excise regions on vdevs that were available during
1922 * the entire duration of this scan.
1924 if (scrub_txg != 0 &&
1925 (spa->spa_scrub_started ||
1926 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1927 vdev_dtl_should_excise(vd)) {
1929 * We completed a scrub up to scrub_txg. If we
1930 * did it without rebooting, then the scrub dtl
1931 * will be valid, so excise the old region and
1932 * fold in the scrub dtl. Otherwise, leave the
1933 * dtl as-is if there was an error.
1935 * There's little trick here: to excise the beginning
1936 * of the DTL_MISSING map, we put it into a reference
1937 * tree and then add a segment with refcnt -1 that
1938 * covers the range [0, scrub_txg). This means
1939 * that each txg in that range has refcnt -1 or 0.
1940 * We then add DTL_SCRUB with a refcnt of 2, so that
1941 * entries in the range [0, scrub_txg) will have a
1942 * positive refcnt -- either 1 or 2. We then convert
1943 * the reference tree into the new DTL_MISSING map.
1945 space_reftree_create(&reftree);
1946 space_reftree_add_map(&reftree,
1947 vd->vdev_dtl[DTL_MISSING], 1);
1948 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1949 space_reftree_add_map(&reftree,
1950 vd->vdev_dtl[DTL_SCRUB], 2);
1951 space_reftree_generate_map(&reftree,
1952 vd->vdev_dtl[DTL_MISSING], 1);
1953 space_reftree_destroy(&reftree);
1955 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1956 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1957 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1959 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1960 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1961 if (!vdev_readable(vd))
1962 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1964 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1965 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1968 * If the vdev was resilvering and no longer has any
1969 * DTLs then reset its resilvering flag and dirty
1970 * the top level so that we persist the change.
1972 if (vd->vdev_resilver_txg != 0 &&
1973 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1974 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
1975 vd->vdev_resilver_txg = 0;
1976 vdev_config_dirty(vd->vdev_top);
1979 mutex_exit(&vd->vdev_dtl_lock);
1982 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1986 mutex_enter(&vd->vdev_dtl_lock);
1987 for (t = 0; t < DTL_TYPES; t++) {
1990 /* account for child's outage in parent's missing map */
1991 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1993 continue; /* leaf vdevs only */
1994 if (t == DTL_PARTIAL)
1995 minref = 1; /* i.e. non-zero */
1996 else if (vd->vdev_nparity != 0)
1997 minref = vd->vdev_nparity + 1; /* RAID-Z */
1999 minref = vd->vdev_children; /* any kind of mirror */
2000 space_reftree_create(&reftree);
2001 for (c = 0; c < vd->vdev_children; c++) {
2002 vdev_t *cvd = vd->vdev_child[c];
2003 mutex_enter(&cvd->vdev_dtl_lock);
2004 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2005 mutex_exit(&cvd->vdev_dtl_lock);
2007 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2008 space_reftree_destroy(&reftree);
2010 mutex_exit(&vd->vdev_dtl_lock);
2014 vdev_dtl_load(vdev_t *vd)
2016 spa_t *spa = vd->vdev_spa;
2017 objset_t *mos = spa->spa_meta_objset;
2021 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2022 ASSERT(!vd->vdev_ishole);
2024 error = space_map_open(&vd->vdev_dtl_sm, mos,
2025 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2028 ASSERT(vd->vdev_dtl_sm != NULL);
2030 mutex_enter(&vd->vdev_dtl_lock);
2033 * Now that we've opened the space_map we need to update
2036 space_map_update(vd->vdev_dtl_sm);
2038 error = space_map_load(vd->vdev_dtl_sm,
2039 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2040 mutex_exit(&vd->vdev_dtl_lock);
2045 for (c = 0; c < vd->vdev_children; c++) {
2046 error = vdev_dtl_load(vd->vdev_child[c]);
2055 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2057 spa_t *spa = vd->vdev_spa;
2059 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2060 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2065 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2067 spa_t *spa = vd->vdev_spa;
2068 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2069 DMU_OT_NONE, 0, tx);
2072 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2079 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2083 if (vd->vdev_ops != &vdev_hole_ops &&
2084 vd->vdev_ops != &vdev_missing_ops &&
2085 vd->vdev_ops != &vdev_root_ops &&
2086 !vd->vdev_top->vdev_removing) {
2087 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2088 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2090 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2091 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2094 for (i = 0; i < vd->vdev_children; i++) {
2095 vdev_construct_zaps(vd->vdev_child[i], tx);
2100 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2102 spa_t *spa = vd->vdev_spa;
2103 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2104 objset_t *mos = spa->spa_meta_objset;
2105 range_tree_t *rtsync;
2108 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2110 ASSERT(!vd->vdev_ishole);
2111 ASSERT(vd->vdev_ops->vdev_op_leaf);
2113 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2115 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2116 mutex_enter(&vd->vdev_dtl_lock);
2117 space_map_free(vd->vdev_dtl_sm, tx);
2118 space_map_close(vd->vdev_dtl_sm);
2119 vd->vdev_dtl_sm = NULL;
2120 mutex_exit(&vd->vdev_dtl_lock);
2123 * We only destroy the leaf ZAP for detached leaves or for
2124 * removed log devices. Removed data devices handle leaf ZAP
2125 * cleanup later, once cancellation is no longer possible.
2127 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2128 vd->vdev_top->vdev_islog)) {
2129 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2130 vd->vdev_leaf_zap = 0;
2137 if (vd->vdev_dtl_sm == NULL) {
2138 uint64_t new_object;
2140 new_object = space_map_alloc(mos, tx);
2141 VERIFY3U(new_object, !=, 0);
2143 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2144 0, -1ULL, 0, &vd->vdev_dtl_lock));
2145 ASSERT(vd->vdev_dtl_sm != NULL);
2148 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2150 rtsync = range_tree_create(NULL, NULL, &rtlock);
2152 mutex_enter(&rtlock);
2154 mutex_enter(&vd->vdev_dtl_lock);
2155 range_tree_walk(rt, range_tree_add, rtsync);
2156 mutex_exit(&vd->vdev_dtl_lock);
2158 space_map_truncate(vd->vdev_dtl_sm, tx);
2159 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2160 range_tree_vacate(rtsync, NULL, NULL);
2162 range_tree_destroy(rtsync);
2164 mutex_exit(&rtlock);
2165 mutex_destroy(&rtlock);
2168 * If the object for the space map has changed then dirty
2169 * the top level so that we update the config.
2171 if (object != space_map_object(vd->vdev_dtl_sm)) {
2172 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2173 "new object %llu", txg, spa_name(spa), object,
2174 space_map_object(vd->vdev_dtl_sm));
2175 vdev_config_dirty(vd->vdev_top);
2180 mutex_enter(&vd->vdev_dtl_lock);
2181 space_map_update(vd->vdev_dtl_sm);
2182 mutex_exit(&vd->vdev_dtl_lock);
2186 * Determine whether the specified vdev can be offlined/detached/removed
2187 * without losing data.
2190 vdev_dtl_required(vdev_t *vd)
2192 spa_t *spa = vd->vdev_spa;
2193 vdev_t *tvd = vd->vdev_top;
2194 uint8_t cant_read = vd->vdev_cant_read;
2197 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2199 if (vd == spa->spa_root_vdev || vd == tvd)
2203 * Temporarily mark the device as unreadable, and then determine
2204 * whether this results in any DTL outages in the top-level vdev.
2205 * If not, we can safely offline/detach/remove the device.
2207 vd->vdev_cant_read = B_TRUE;
2208 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2209 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2210 vd->vdev_cant_read = cant_read;
2211 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2213 if (!required && zio_injection_enabled)
2214 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2220 * Determine if resilver is needed, and if so the txg range.
2223 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2225 boolean_t needed = B_FALSE;
2226 uint64_t thismin = UINT64_MAX;
2227 uint64_t thismax = 0;
2230 if (vd->vdev_children == 0) {
2231 mutex_enter(&vd->vdev_dtl_lock);
2232 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2233 vdev_writeable(vd)) {
2235 thismin = vdev_dtl_min(vd);
2236 thismax = vdev_dtl_max(vd);
2239 mutex_exit(&vd->vdev_dtl_lock);
2241 for (c = 0; c < vd->vdev_children; c++) {
2242 vdev_t *cvd = vd->vdev_child[c];
2243 uint64_t cmin, cmax;
2245 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2246 thismin = MIN(thismin, cmin);
2247 thismax = MAX(thismax, cmax);
2253 if (needed && minp) {
2261 vdev_load(vdev_t *vd)
2266 * Recursively load all children.
2268 for (c = 0; c < vd->vdev_children; c++)
2269 vdev_load(vd->vdev_child[c]);
2272 * If this is a top-level vdev, initialize its metaslabs.
2274 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2275 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2276 vdev_metaslab_init(vd, 0) != 0))
2277 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2278 VDEV_AUX_CORRUPT_DATA);
2280 * If this is a leaf vdev, load its DTL.
2282 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2283 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2284 VDEV_AUX_CORRUPT_DATA);
2288 * The special vdev case is used for hot spares and l2cache devices. Its
2289 * sole purpose it to set the vdev state for the associated vdev. To do this,
2290 * we make sure that we can open the underlying device, then try to read the
2291 * label, and make sure that the label is sane and that it hasn't been
2292 * repurposed to another pool.
2295 vdev_validate_aux(vdev_t *vd)
2298 uint64_t guid, version;
2301 if (!vdev_readable(vd))
2304 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2305 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2306 VDEV_AUX_CORRUPT_DATA);
2310 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2311 !SPA_VERSION_IS_SUPPORTED(version) ||
2312 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2313 guid != vd->vdev_guid ||
2314 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2315 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2316 VDEV_AUX_CORRUPT_DATA);
2322 * We don't actually check the pool state here. If it's in fact in
2323 * use by another pool, we update this fact on the fly when requested.
2330 vdev_remove(vdev_t *vd, uint64_t txg)
2332 spa_t *spa = vd->vdev_spa;
2333 objset_t *mos = spa->spa_meta_objset;
2337 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2338 ASSERT(vd == vd->vdev_top);
2339 ASSERT3U(txg, ==, spa_syncing_txg(spa));
2341 if (vd->vdev_ms != NULL) {
2342 metaslab_group_t *mg = vd->vdev_mg;
2344 metaslab_group_histogram_verify(mg);
2345 metaslab_class_histogram_verify(mg->mg_class);
2347 for (m = 0; m < vd->vdev_ms_count; m++) {
2348 metaslab_t *msp = vd->vdev_ms[m];
2350 if (msp == NULL || msp->ms_sm == NULL)
2353 mutex_enter(&msp->ms_lock);
2355 * If the metaslab was not loaded when the vdev
2356 * was removed then the histogram accounting may
2357 * not be accurate. Update the histogram information
2358 * here so that we ensure that the metaslab group
2359 * and metaslab class are up-to-date.
2361 metaslab_group_histogram_remove(mg, msp);
2363 VERIFY0(space_map_allocated(msp->ms_sm));
2364 space_map_free(msp->ms_sm, tx);
2365 space_map_close(msp->ms_sm);
2367 mutex_exit(&msp->ms_lock);
2370 metaslab_group_histogram_verify(mg);
2371 metaslab_class_histogram_verify(mg->mg_class);
2372 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2373 ASSERT0(mg->mg_histogram[i]);
2377 if (vd->vdev_ms_array) {
2378 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2379 vd->vdev_ms_array = 0;
2382 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
2383 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
2384 vd->vdev_top_zap = 0;
2390 vdev_sync_done(vdev_t *vd, uint64_t txg)
2393 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2395 ASSERT(!vd->vdev_ishole);
2397 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
2398 metaslab_sync_done(msp, txg);
2401 metaslab_sync_reassess(vd->vdev_mg);
2405 vdev_sync(vdev_t *vd, uint64_t txg)
2407 spa_t *spa = vd->vdev_spa;
2412 ASSERT(!vd->vdev_ishole);
2414 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2415 ASSERT(vd == vd->vdev_top);
2416 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2417 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2418 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2419 ASSERT(vd->vdev_ms_array != 0);
2420 vdev_config_dirty(vd);
2425 * Remove the metadata associated with this vdev once it's empty.
2427 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2428 vdev_remove(vd, txg);
2430 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2431 metaslab_sync(msp, txg);
2432 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2435 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2436 vdev_dtl_sync(lvd, txg);
2438 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2442 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2444 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2448 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2449 * not be opened, and no I/O is attempted.
2452 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2456 spa_vdev_state_enter(spa, SCL_NONE);
2458 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2459 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2461 if (!vd->vdev_ops->vdev_op_leaf)
2462 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2467 * We don't directly use the aux state here, but if we do a
2468 * vdev_reopen(), we need this value to be present to remember why we
2471 vd->vdev_label_aux = aux;
2474 * Faulted state takes precedence over degraded.
2476 vd->vdev_delayed_close = B_FALSE;
2477 vd->vdev_faulted = 1ULL;
2478 vd->vdev_degraded = 0ULL;
2479 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2482 * If this device has the only valid copy of the data, then
2483 * back off and simply mark the vdev as degraded instead.
2485 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2486 vd->vdev_degraded = 1ULL;
2487 vd->vdev_faulted = 0ULL;
2490 * If we reopen the device and it's not dead, only then do we
2495 if (vdev_readable(vd))
2496 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2499 return (spa_vdev_state_exit(spa, vd, 0));
2503 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2504 * user that something is wrong. The vdev continues to operate as normal as far
2505 * as I/O is concerned.
2508 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2512 spa_vdev_state_enter(spa, SCL_NONE);
2514 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2515 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2517 if (!vd->vdev_ops->vdev_op_leaf)
2518 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2521 * If the vdev is already faulted, then don't do anything.
2523 if (vd->vdev_faulted || vd->vdev_degraded)
2524 return (spa_vdev_state_exit(spa, NULL, 0));
2526 vd->vdev_degraded = 1ULL;
2527 if (!vdev_is_dead(vd))
2528 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2531 return (spa_vdev_state_exit(spa, vd, 0));
2535 * Online the given vdev.
2537 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2538 * spare device should be detached when the device finishes resilvering.
2539 * Second, the online should be treated like a 'test' online case, so no FMA
2540 * events are generated if the device fails to open.
2543 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2545 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2546 boolean_t wasoffline;
2547 vdev_state_t oldstate;
2549 spa_vdev_state_enter(spa, SCL_NONE);
2551 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2552 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2554 if (!vd->vdev_ops->vdev_op_leaf)
2555 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2557 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
2558 oldstate = vd->vdev_state;
2561 vd->vdev_offline = B_FALSE;
2562 vd->vdev_tmpoffline = B_FALSE;
2563 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2564 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2566 /* XXX - L2ARC 1.0 does not support expansion */
2567 if (!vd->vdev_aux) {
2568 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2569 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2573 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2575 if (!vd->vdev_aux) {
2576 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2577 pvd->vdev_expanding = B_FALSE;
2581 *newstate = vd->vdev_state;
2582 if ((flags & ZFS_ONLINE_UNSPARE) &&
2583 !vdev_is_dead(vd) && vd->vdev_parent &&
2584 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2585 vd->vdev_parent->vdev_child[0] == vd)
2586 vd->vdev_unspare = B_TRUE;
2588 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2590 /* XXX - L2ARC 1.0 does not support expansion */
2592 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2593 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2597 (oldstate < VDEV_STATE_DEGRADED &&
2598 vd->vdev_state >= VDEV_STATE_DEGRADED))
2599 spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE);
2601 return (spa_vdev_state_exit(spa, vd, 0));
2605 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2609 uint64_t generation;
2610 metaslab_group_t *mg;
2613 spa_vdev_state_enter(spa, SCL_ALLOC);
2615 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2616 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2618 if (!vd->vdev_ops->vdev_op_leaf)
2619 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2623 generation = spa->spa_config_generation + 1;
2626 * If the device isn't already offline, try to offline it.
2628 if (!vd->vdev_offline) {
2630 * If this device has the only valid copy of some data,
2631 * don't allow it to be offlined. Log devices are always
2634 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2635 vdev_dtl_required(vd))
2636 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2639 * If the top-level is a slog and it has had allocations
2640 * then proceed. We check that the vdev's metaslab group
2641 * is not NULL since it's possible that we may have just
2642 * added this vdev but not yet initialized its metaslabs.
2644 if (tvd->vdev_islog && mg != NULL) {
2646 * Prevent any future allocations.
2648 metaslab_group_passivate(mg);
2649 (void) spa_vdev_state_exit(spa, vd, 0);
2651 error = spa_offline_log(spa);
2653 spa_vdev_state_enter(spa, SCL_ALLOC);
2656 * Check to see if the config has changed.
2658 if (error || generation != spa->spa_config_generation) {
2659 metaslab_group_activate(mg);
2661 return (spa_vdev_state_exit(spa,
2663 (void) spa_vdev_state_exit(spa, vd, 0);
2666 ASSERT0(tvd->vdev_stat.vs_alloc);
2670 * Offline this device and reopen its top-level vdev.
2671 * If the top-level vdev is a log device then just offline
2672 * it. Otherwise, if this action results in the top-level
2673 * vdev becoming unusable, undo it and fail the request.
2675 vd->vdev_offline = B_TRUE;
2678 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2679 vdev_is_dead(tvd)) {
2680 vd->vdev_offline = B_FALSE;
2682 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2686 * Add the device back into the metaslab rotor so that
2687 * once we online the device it's open for business.
2689 if (tvd->vdev_islog && mg != NULL)
2690 metaslab_group_activate(mg);
2693 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2695 return (spa_vdev_state_exit(spa, vd, 0));
2699 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2703 mutex_enter(&spa->spa_vdev_top_lock);
2704 error = vdev_offline_locked(spa, guid, flags);
2705 mutex_exit(&spa->spa_vdev_top_lock);
2711 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2712 * vdev_offline(), we assume the spa config is locked. We also clear all
2713 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2716 vdev_clear(spa_t *spa, vdev_t *vd)
2718 vdev_t *rvd = spa->spa_root_vdev;
2721 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2726 vd->vdev_stat.vs_read_errors = 0;
2727 vd->vdev_stat.vs_write_errors = 0;
2728 vd->vdev_stat.vs_checksum_errors = 0;
2730 for (c = 0; c < vd->vdev_children; c++)
2731 vdev_clear(spa, vd->vdev_child[c]);
2734 * If we're in the FAULTED state or have experienced failed I/O, then
2735 * clear the persistent state and attempt to reopen the device. We
2736 * also mark the vdev config dirty, so that the new faulted state is
2737 * written out to disk.
2739 if (vd->vdev_faulted || vd->vdev_degraded ||
2740 !vdev_readable(vd) || !vdev_writeable(vd)) {
2743 * When reopening in response to a clear event, it may be due to
2744 * a fmadm repair request. In this case, if the device is
2745 * still broken, we want to still post the ereport again.
2747 vd->vdev_forcefault = B_TRUE;
2749 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2750 vd->vdev_cant_read = B_FALSE;
2751 vd->vdev_cant_write = B_FALSE;
2753 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2755 vd->vdev_forcefault = B_FALSE;
2757 if (vd != rvd && vdev_writeable(vd->vdev_top))
2758 vdev_state_dirty(vd->vdev_top);
2760 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2761 spa_async_request(spa, SPA_ASYNC_RESILVER);
2763 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2767 * When clearing a FMA-diagnosed fault, we always want to
2768 * unspare the device, as we assume that the original spare was
2769 * done in response to the FMA fault.
2771 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2772 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2773 vd->vdev_parent->vdev_child[0] == vd)
2774 vd->vdev_unspare = B_TRUE;
2778 vdev_is_dead(vdev_t *vd)
2781 * Holes and missing devices are always considered "dead".
2782 * This simplifies the code since we don't have to check for
2783 * these types of devices in the various code paths.
2784 * Instead we rely on the fact that we skip over dead devices
2785 * before issuing I/O to them.
2787 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2788 vd->vdev_ops == &vdev_missing_ops);
2792 vdev_readable(vdev_t *vd)
2794 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2798 vdev_writeable(vdev_t *vd)
2800 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2804 vdev_allocatable(vdev_t *vd)
2806 uint64_t state = vd->vdev_state;
2809 * We currently allow allocations from vdevs which may be in the
2810 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2811 * fails to reopen then we'll catch it later when we're holding
2812 * the proper locks. Note that we have to get the vdev state
2813 * in a local variable because although it changes atomically,
2814 * we're asking two separate questions about it.
2816 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2817 !vd->vdev_cant_write && !vd->vdev_ishole &&
2818 vd->vdev_mg->mg_initialized);
2822 vdev_accessible(vdev_t *vd, zio_t *zio)
2824 ASSERT(zio->io_vd == vd);
2826 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2829 if (zio->io_type == ZIO_TYPE_READ)
2830 return (!vd->vdev_cant_read);
2832 if (zio->io_type == ZIO_TYPE_WRITE)
2833 return (!vd->vdev_cant_write);
2839 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs)
2842 for (t = 0; t < ZIO_TYPES; t++) {
2843 vs->vs_ops[t] += cvs->vs_ops[t];
2844 vs->vs_bytes[t] += cvs->vs_bytes[t];
2847 cvs->vs_scan_removing = cvd->vdev_removing;
2851 * Get extended stats
2854 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx)
2857 for (t = 0; t < ZIO_TYPES; t++) {
2858 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++)
2859 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b];
2861 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) {
2862 vsx->vsx_total_histo[t][b] +=
2863 cvsx->vsx_total_histo[t][b];
2867 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
2868 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) {
2869 vsx->vsx_queue_histo[t][b] +=
2870 cvsx->vsx_queue_histo[t][b];
2872 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t];
2873 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t];
2875 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++)
2876 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b];
2878 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++)
2879 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b];
2885 * Get statistics for the given vdev.
2888 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
2892 * If we're getting stats on the root vdev, aggregate the I/O counts
2893 * over all top-level vdevs (i.e. the direct children of the root).
2895 if (!vd->vdev_ops->vdev_op_leaf) {
2897 memset(vs->vs_ops, 0, sizeof (vs->vs_ops));
2898 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes));
2901 memset(vsx, 0, sizeof (*vsx));
2903 for (c = 0; c < vd->vdev_children; c++) {
2904 vdev_t *cvd = vd->vdev_child[c];
2905 vdev_stat_t *cvs = &cvd->vdev_stat;
2906 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex;
2908 vdev_get_stats_ex_impl(cvd, cvs, cvsx);
2910 vdev_get_child_stat(cvd, vs, cvs);
2912 vdev_get_child_stat_ex(cvd, vsx, cvsx);
2917 * We're a leaf. Just copy our ZIO active queue stats in. The
2918 * other leaf stats are updated in vdev_stat_update().
2923 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex));
2925 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) {
2926 vsx->vsx_active_queue[t] =
2927 vd->vdev_queue.vq_class[t].vqc_active;
2928 vsx->vsx_pend_queue[t] = avl_numnodes(
2929 &vd->vdev_queue.vq_class[t].vqc_queued_tree);
2935 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
2937 vdev_t *tvd = vd->vdev_top;
2938 mutex_enter(&vd->vdev_stat_lock);
2940 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2941 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2942 vs->vs_state = vd->vdev_state;
2943 vs->vs_rsize = vdev_get_min_asize(vd);
2944 if (vd->vdev_ops->vdev_op_leaf)
2945 vs->vs_rsize += VDEV_LABEL_START_SIZE +
2946 VDEV_LABEL_END_SIZE;
2948 * Report expandable space on top-level, non-auxillary devices
2949 * only. The expandable space is reported in terms of metaslab
2950 * sized units since that determines how much space the pool
2953 if (vd->vdev_aux == NULL && tvd != NULL) {
2954 vs->vs_esize = P2ALIGN(
2955 vd->vdev_max_asize - vd->vdev_asize,
2956 1ULL << tvd->vdev_ms_shift);
2958 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2959 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
2961 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2965 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_READER) != 0);
2966 vdev_get_stats_ex_impl(vd, vs, vsx);
2967 mutex_exit(&vd->vdev_stat_lock);
2971 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2973 return (vdev_get_stats_ex(vd, vs, NULL));
2977 vdev_clear_stats(vdev_t *vd)
2979 mutex_enter(&vd->vdev_stat_lock);
2980 vd->vdev_stat.vs_space = 0;
2981 vd->vdev_stat.vs_dspace = 0;
2982 vd->vdev_stat.vs_alloc = 0;
2983 mutex_exit(&vd->vdev_stat_lock);
2987 vdev_scan_stat_init(vdev_t *vd)
2989 vdev_stat_t *vs = &vd->vdev_stat;
2992 for (c = 0; c < vd->vdev_children; c++)
2993 vdev_scan_stat_init(vd->vdev_child[c]);
2995 mutex_enter(&vd->vdev_stat_lock);
2996 vs->vs_scan_processed = 0;
2997 mutex_exit(&vd->vdev_stat_lock);
3001 vdev_stat_update(zio_t *zio, uint64_t psize)
3003 spa_t *spa = zio->io_spa;
3004 vdev_t *rvd = spa->spa_root_vdev;
3005 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3007 uint64_t txg = zio->io_txg;
3008 vdev_stat_t *vs = &vd->vdev_stat;
3009 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex;
3010 zio_type_t type = zio->io_type;
3011 int flags = zio->io_flags;
3014 * If this i/o is a gang leader, it didn't do any actual work.
3016 if (zio->io_gang_tree)
3019 if (zio->io_error == 0) {
3021 * If this is a root i/o, don't count it -- we've already
3022 * counted the top-level vdevs, and vdev_get_stats() will
3023 * aggregate them when asked. This reduces contention on
3024 * the root vdev_stat_lock and implicitly handles blocks
3025 * that compress away to holes, for which there is no i/o.
3026 * (Holes never create vdev children, so all the counters
3027 * remain zero, which is what we want.)
3029 * Note: this only applies to successful i/o (io_error == 0)
3030 * because unlike i/o counts, errors are not additive.
3031 * When reading a ditto block, for example, failure of
3032 * one top-level vdev does not imply a root-level error.
3037 ASSERT(vd == zio->io_vd);
3039 if (flags & ZIO_FLAG_IO_BYPASS)
3042 mutex_enter(&vd->vdev_stat_lock);
3044 if (flags & ZIO_FLAG_IO_REPAIR) {
3045 if (flags & ZIO_FLAG_SCAN_THREAD) {
3046 dsl_scan_phys_t *scn_phys =
3047 &spa->spa_dsl_pool->dp_scan->scn_phys;
3048 uint64_t *processed = &scn_phys->scn_processed;
3051 if (vd->vdev_ops->vdev_op_leaf)
3052 atomic_add_64(processed, psize);
3053 vs->vs_scan_processed += psize;
3056 if (flags & ZIO_FLAG_SELF_HEAL)
3057 vs->vs_self_healed += psize;
3061 * The bytes/ops/histograms are recorded at the leaf level and
3062 * aggregated into the higher level vdevs in vdev_get_stats().
3064 if (vd->vdev_ops->vdev_op_leaf &&
3065 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) {
3068 vs->vs_bytes[type] += psize;
3070 if (flags & ZIO_FLAG_DELEGATED) {
3071 vsx->vsx_agg_histo[zio->io_priority]
3072 [RQ_HISTO(zio->io_size)]++;
3074 vsx->vsx_ind_histo[zio->io_priority]
3075 [RQ_HISTO(zio->io_size)]++;
3078 if (zio->io_delta && zio->io_delay) {
3079 vsx->vsx_queue_histo[zio->io_priority]
3080 [L_HISTO(zio->io_delta - zio->io_delay)]++;
3081 vsx->vsx_disk_histo[type]
3082 [L_HISTO(zio->io_delay)]++;
3083 vsx->vsx_total_histo[type]
3084 [L_HISTO(zio->io_delta)]++;
3088 mutex_exit(&vd->vdev_stat_lock);
3092 if (flags & ZIO_FLAG_SPECULATIVE)
3096 * If this is an I/O error that is going to be retried, then ignore the
3097 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3098 * hard errors, when in reality they can happen for any number of
3099 * innocuous reasons (bus resets, MPxIO link failure, etc).
3101 if (zio->io_error == EIO &&
3102 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3106 * Intent logs writes won't propagate their error to the root
3107 * I/O so don't mark these types of failures as pool-level
3110 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3113 mutex_enter(&vd->vdev_stat_lock);
3114 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3115 if (zio->io_error == ECKSUM)
3116 vs->vs_checksum_errors++;
3118 vs->vs_read_errors++;
3120 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3121 vs->vs_write_errors++;
3122 mutex_exit(&vd->vdev_stat_lock);
3124 if (type == ZIO_TYPE_WRITE && txg != 0 &&
3125 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3126 (flags & ZIO_FLAG_SCAN_THREAD) ||
3127 spa->spa_claiming)) {
3129 * This is either a normal write (not a repair), or it's
3130 * a repair induced by the scrub thread, or it's a repair
3131 * made by zil_claim() during spa_load() in the first txg.
3132 * In the normal case, we commit the DTL change in the same
3133 * txg as the block was born. In the scrub-induced repair
3134 * case, we know that scrubs run in first-pass syncing context,
3135 * so we commit the DTL change in spa_syncing_txg(spa).
3136 * In the zil_claim() case, we commit in spa_first_txg(spa).
3138 * We currently do not make DTL entries for failed spontaneous
3139 * self-healing writes triggered by normal (non-scrubbing)
3140 * reads, because we have no transactional context in which to
3141 * do so -- and it's not clear that it'd be desirable anyway.
3143 if (vd->vdev_ops->vdev_op_leaf) {
3144 uint64_t commit_txg = txg;
3145 if (flags & ZIO_FLAG_SCAN_THREAD) {
3146 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3147 ASSERT(spa_sync_pass(spa) == 1);
3148 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3149 commit_txg = spa_syncing_txg(spa);
3150 } else if (spa->spa_claiming) {
3151 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3152 commit_txg = spa_first_txg(spa);
3154 ASSERT(commit_txg >= spa_syncing_txg(spa));
3155 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3157 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3158 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3159 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3162 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3167 * Update the in-core space usage stats for this vdev, its metaslab class,
3168 * and the root vdev.
3171 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3172 int64_t space_delta)
3174 int64_t dspace_delta = space_delta;
3175 spa_t *spa = vd->vdev_spa;
3176 vdev_t *rvd = spa->spa_root_vdev;
3177 metaslab_group_t *mg = vd->vdev_mg;
3178 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3180 ASSERT(vd == vd->vdev_top);
3183 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3184 * factor. We must calculate this here and not at the root vdev
3185 * because the root vdev's psize-to-asize is simply the max of its
3186 * childrens', thus not accurate enough for us.
3188 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3189 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3190 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3191 vd->vdev_deflate_ratio;
3193 mutex_enter(&vd->vdev_stat_lock);
3194 vd->vdev_stat.vs_alloc += alloc_delta;
3195 vd->vdev_stat.vs_space += space_delta;
3196 vd->vdev_stat.vs_dspace += dspace_delta;
3197 mutex_exit(&vd->vdev_stat_lock);
3199 if (mc == spa_normal_class(spa)) {
3200 mutex_enter(&rvd->vdev_stat_lock);
3201 rvd->vdev_stat.vs_alloc += alloc_delta;
3202 rvd->vdev_stat.vs_space += space_delta;
3203 rvd->vdev_stat.vs_dspace += dspace_delta;
3204 mutex_exit(&rvd->vdev_stat_lock);
3208 ASSERT(rvd == vd->vdev_parent);
3209 ASSERT(vd->vdev_ms_count != 0);
3211 metaslab_class_space_update(mc,
3212 alloc_delta, defer_delta, space_delta, dspace_delta);
3217 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3218 * so that it will be written out next time the vdev configuration is synced.
3219 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3222 vdev_config_dirty(vdev_t *vd)
3224 spa_t *spa = vd->vdev_spa;
3225 vdev_t *rvd = spa->spa_root_vdev;
3228 ASSERT(spa_writeable(spa));
3231 * If this is an aux vdev (as with l2cache and spare devices), then we
3232 * update the vdev config manually and set the sync flag.
3234 if (vd->vdev_aux != NULL) {
3235 spa_aux_vdev_t *sav = vd->vdev_aux;
3239 for (c = 0; c < sav->sav_count; c++) {
3240 if (sav->sav_vdevs[c] == vd)
3244 if (c == sav->sav_count) {
3246 * We're being removed. There's nothing more to do.
3248 ASSERT(sav->sav_sync == B_TRUE);
3252 sav->sav_sync = B_TRUE;
3254 if (nvlist_lookup_nvlist_array(sav->sav_config,
3255 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3256 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3257 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3263 * Setting the nvlist in the middle if the array is a little
3264 * sketchy, but it will work.
3266 nvlist_free(aux[c]);
3267 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3273 * The dirty list is protected by the SCL_CONFIG lock. The caller
3274 * must either hold SCL_CONFIG as writer, or must be the sync thread
3275 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3276 * so this is sufficient to ensure mutual exclusion.
3278 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3279 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3280 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3283 for (c = 0; c < rvd->vdev_children; c++)
3284 vdev_config_dirty(rvd->vdev_child[c]);
3286 ASSERT(vd == vd->vdev_top);
3288 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3290 list_insert_head(&spa->spa_config_dirty_list, vd);
3295 vdev_config_clean(vdev_t *vd)
3297 spa_t *spa = vd->vdev_spa;
3299 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3300 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3301 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3303 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3304 list_remove(&spa->spa_config_dirty_list, vd);
3308 * Mark a top-level vdev's state as dirty, so that the next pass of
3309 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3310 * the state changes from larger config changes because they require
3311 * much less locking, and are often needed for administrative actions.
3314 vdev_state_dirty(vdev_t *vd)
3316 spa_t *spa = vd->vdev_spa;
3318 ASSERT(spa_writeable(spa));
3319 ASSERT(vd == vd->vdev_top);
3322 * The state list is protected by the SCL_STATE lock. The caller
3323 * must either hold SCL_STATE as writer, or must be the sync thread
3324 * (which holds SCL_STATE as reader). There's only one sync thread,
3325 * so this is sufficient to ensure mutual exclusion.
3327 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3328 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3329 spa_config_held(spa, SCL_STATE, RW_READER)));
3331 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3332 list_insert_head(&spa->spa_state_dirty_list, vd);
3336 vdev_state_clean(vdev_t *vd)
3338 spa_t *spa = vd->vdev_spa;
3340 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3341 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3342 spa_config_held(spa, SCL_STATE, RW_READER)));
3344 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3345 list_remove(&spa->spa_state_dirty_list, vd);
3349 * Propagate vdev state up from children to parent.
3352 vdev_propagate_state(vdev_t *vd)
3354 spa_t *spa = vd->vdev_spa;
3355 vdev_t *rvd = spa->spa_root_vdev;
3356 int degraded = 0, faulted = 0;
3361 if (vd->vdev_children > 0) {
3362 for (c = 0; c < vd->vdev_children; c++) {
3363 child = vd->vdev_child[c];
3366 * Don't factor holes into the decision.
3368 if (child->vdev_ishole)
3371 if (!vdev_readable(child) ||
3372 (!vdev_writeable(child) && spa_writeable(spa))) {
3374 * Root special: if there is a top-level log
3375 * device, treat the root vdev as if it were
3378 if (child->vdev_islog && vd == rvd)
3382 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3386 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3390 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3393 * Root special: if there is a top-level vdev that cannot be
3394 * opened due to corrupted metadata, then propagate the root
3395 * vdev's aux state as 'corrupt' rather than 'insufficient
3398 if (corrupted && vd == rvd &&
3399 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3400 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3401 VDEV_AUX_CORRUPT_DATA);
3404 if (vd->vdev_parent)
3405 vdev_propagate_state(vd->vdev_parent);
3409 * Set a vdev's state. If this is during an open, we don't update the parent
3410 * state, because we're in the process of opening children depth-first.
3411 * Otherwise, we propagate the change to the parent.
3413 * If this routine places a device in a faulted state, an appropriate ereport is
3417 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3419 uint64_t save_state;
3420 spa_t *spa = vd->vdev_spa;
3422 if (state == vd->vdev_state) {
3424 * Since vdev_offline() code path is already in an offline
3425 * state we can miss a statechange event to OFFLINE. Check
3426 * the previous state to catch this condition.
3428 if (vd->vdev_ops->vdev_op_leaf &&
3429 (state == VDEV_STATE_OFFLINE) &&
3430 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) {
3431 /* post an offline state change */
3432 zfs_post_state_change(spa, vd, vd->vdev_prevstate);
3434 vd->vdev_stat.vs_aux = aux;
3438 save_state = vd->vdev_state;
3440 vd->vdev_state = state;
3441 vd->vdev_stat.vs_aux = aux;
3444 * If we are setting the vdev state to anything but an open state, then
3445 * always close the underlying device unless the device has requested
3446 * a delayed close (i.e. we're about to remove or fault the device).
3447 * Otherwise, we keep accessible but invalid devices open forever.
3448 * We don't call vdev_close() itself, because that implies some extra
3449 * checks (offline, etc) that we don't want here. This is limited to
3450 * leaf devices, because otherwise closing the device will affect other
3453 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3454 vd->vdev_ops->vdev_op_leaf)
3455 vd->vdev_ops->vdev_op_close(vd);
3457 if (vd->vdev_removed &&
3458 state == VDEV_STATE_CANT_OPEN &&
3459 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3461 * If the previous state is set to VDEV_STATE_REMOVED, then this
3462 * device was previously marked removed and someone attempted to
3463 * reopen it. If this failed due to a nonexistent device, then
3464 * keep the device in the REMOVED state. We also let this be if
3465 * it is one of our special test online cases, which is only
3466 * attempting to online the device and shouldn't generate an FMA
3469 vd->vdev_state = VDEV_STATE_REMOVED;
3470 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3471 } else if (state == VDEV_STATE_REMOVED) {
3472 vd->vdev_removed = B_TRUE;
3473 } else if (state == VDEV_STATE_CANT_OPEN) {
3475 * If we fail to open a vdev during an import or recovery, we
3476 * mark it as "not available", which signifies that it was
3477 * never there to begin with. Failure to open such a device
3478 * is not considered an error.
3480 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3481 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3482 vd->vdev_ops->vdev_op_leaf)
3483 vd->vdev_not_present = 1;
3486 * Post the appropriate ereport. If the 'prevstate' field is
3487 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3488 * that this is part of a vdev_reopen(). In this case, we don't
3489 * want to post the ereport if the device was already in the
3490 * CANT_OPEN state beforehand.
3492 * If the 'checkremove' flag is set, then this is an attempt to
3493 * online the device in response to an insertion event. If we
3494 * hit this case, then we have detected an insertion event for a
3495 * faulted or offline device that wasn't in the removed state.
3496 * In this scenario, we don't post an ereport because we are
3497 * about to replace the device, or attempt an online with
3498 * vdev_forcefault, which will generate the fault for us.
3500 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3501 !vd->vdev_not_present && !vd->vdev_checkremove &&
3502 vd != spa->spa_root_vdev) {
3506 case VDEV_AUX_OPEN_FAILED:
3507 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3509 case VDEV_AUX_CORRUPT_DATA:
3510 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3512 case VDEV_AUX_NO_REPLICAS:
3513 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3515 case VDEV_AUX_BAD_GUID_SUM:
3516 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3518 case VDEV_AUX_TOO_SMALL:
3519 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3521 case VDEV_AUX_BAD_LABEL:
3522 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3524 case VDEV_AUX_BAD_ASHIFT:
3525 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT;
3528 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3531 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3534 /* Erase any notion of persistent removed state */
3535 vd->vdev_removed = B_FALSE;
3537 vd->vdev_removed = B_FALSE;
3541 * Notify ZED of any significant state-change on a leaf vdev.
3544 if (vd->vdev_ops->vdev_op_leaf) {
3545 /* preserve original state from a vdev_reopen() */
3546 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) &&
3547 (vd->vdev_prevstate != vd->vdev_state) &&
3548 (save_state <= VDEV_STATE_CLOSED))
3549 save_state = vd->vdev_prevstate;
3551 /* filter out state change due to initial vdev_open */
3552 if (save_state > VDEV_STATE_CLOSED)
3553 zfs_post_state_change(spa, vd, save_state);
3556 if (!isopen && vd->vdev_parent)
3557 vdev_propagate_state(vd->vdev_parent);
3561 * Check the vdev configuration to ensure that it's capable of supporting
3562 * a root pool. We do not support partial configuration.
3565 vdev_is_bootable(vdev_t *vd)
3567 if (!vd->vdev_ops->vdev_op_leaf) {
3568 const char *vdev_type = vd->vdev_ops->vdev_op_type;
3570 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0)
3574 for (int c = 0; c < vd->vdev_children; c++) {
3575 if (!vdev_is_bootable(vd->vdev_child[c]))
3582 * Load the state from the original vdev tree (ovd) which
3583 * we've retrieved from the MOS config object. If the original
3584 * vdev was offline or faulted then we transfer that state to the
3585 * device in the current vdev tree (nvd).
3588 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3592 ASSERT(nvd->vdev_top->vdev_islog);
3593 ASSERT(spa_config_held(nvd->vdev_spa,
3594 SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3595 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3597 for (c = 0; c < nvd->vdev_children; c++)
3598 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3600 if (nvd->vdev_ops->vdev_op_leaf) {
3602 * Restore the persistent vdev state
3604 nvd->vdev_offline = ovd->vdev_offline;
3605 nvd->vdev_faulted = ovd->vdev_faulted;
3606 nvd->vdev_degraded = ovd->vdev_degraded;
3607 nvd->vdev_removed = ovd->vdev_removed;
3612 * Determine if a log device has valid content. If the vdev was
3613 * removed or faulted in the MOS config then we know that
3614 * the content on the log device has already been written to the pool.
3617 vdev_log_state_valid(vdev_t *vd)
3621 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3625 for (c = 0; c < vd->vdev_children; c++)
3626 if (vdev_log_state_valid(vd->vdev_child[c]))
3633 * Expand a vdev if possible.
3636 vdev_expand(vdev_t *vd, uint64_t txg)
3638 ASSERT(vd->vdev_top == vd);
3639 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3641 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3642 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3643 vdev_config_dirty(vd);
3651 vdev_split(vdev_t *vd)
3653 vdev_t *cvd, *pvd = vd->vdev_parent;
3655 vdev_remove_child(pvd, vd);
3656 vdev_compact_children(pvd);
3658 cvd = pvd->vdev_child[0];
3659 if (pvd->vdev_children == 1) {
3660 vdev_remove_parent(cvd);
3661 cvd->vdev_splitting = B_TRUE;
3663 vdev_propagate_state(cvd);
3667 vdev_deadman(vdev_t *vd)
3671 for (c = 0; c < vd->vdev_children; c++) {
3672 vdev_t *cvd = vd->vdev_child[c];
3677 if (vd->vdev_ops->vdev_op_leaf) {
3678 vdev_queue_t *vq = &vd->vdev_queue;
3680 mutex_enter(&vq->vq_lock);
3681 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3682 spa_t *spa = vd->vdev_spa;
3687 * Look at the head of all the pending queues,
3688 * if any I/O has been outstanding for longer than
3689 * the spa_deadman_synctime we log a zevent.
3691 fio = avl_first(&vq->vq_active_tree);
3692 delta = gethrtime() - fio->io_timestamp;
3693 if (delta > spa_deadman_synctime(spa)) {
3694 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3695 "delta %lluns, last io %lluns",
3696 fio->io_timestamp, delta,
3697 vq->vq_io_complete_ts);
3698 zfs_ereport_post(FM_EREPORT_ZFS_DELAY,
3699 spa, vd, fio, 0, 0);
3702 mutex_exit(&vq->vq_lock);
3706 #if defined(_KERNEL) && defined(HAVE_SPL)
3707 EXPORT_SYMBOL(vdev_fault);
3708 EXPORT_SYMBOL(vdev_degrade);
3709 EXPORT_SYMBOL(vdev_online);
3710 EXPORT_SYMBOL(vdev_offline);
3711 EXPORT_SYMBOL(vdev_clear);
3713 module_param(metaslabs_per_vdev, int, 0644);
3714 MODULE_PARM_DESC(metaslabs_per_vdev,
3715 "Divide added vdev into approximately (but no more than) this number "