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 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/fm/fs/zfs.h>
32 #include <sys/spa_impl.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/vdev_impl.h>
36 #include <sys/uberblock_impl.h>
37 #include <sys/metaslab.h>
38 #include <sys/metaslab_impl.h>
39 #include <sys/space_map.h>
40 #include <sys/space_reftree.h>
43 #include <sys/fs/zfs.h>
46 #include <sys/dsl_scan.h>
47 #include <sys/trim_map.h>
49 SYSCTL_DECL(_vfs_zfs);
50 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
53 * Virtual device management.
57 * The limit for ZFS to automatically increase a top-level vdev's ashift
58 * from logical ashift to physical ashift.
60 * Example: one or more 512B emulation child vdevs
61 * child->vdev_ashift = 9 (512 bytes)
62 * child->vdev_physical_ashift = 12 (4096 bytes)
63 * zfs_max_auto_ashift = 11 (2048 bytes)
64 * zfs_min_auto_ashift = 9 (512 bytes)
66 * On pool creation or the addition of a new top-level vdev, ZFS will
67 * increase the ashift of the top-level vdev to 2048 as limited by
68 * zfs_max_auto_ashift.
70 * Example: one or more 512B emulation child vdevs
71 * child->vdev_ashift = 9 (512 bytes)
72 * child->vdev_physical_ashift = 12 (4096 bytes)
73 * zfs_max_auto_ashift = 13 (8192 bytes)
74 * zfs_min_auto_ashift = 9 (512 bytes)
76 * On pool creation or the addition of a new top-level vdev, ZFS will
77 * increase the ashift of the top-level vdev to 4096 to match the
78 * max vdev_physical_ashift.
80 * Example: one or more 512B emulation child vdevs
81 * child->vdev_ashift = 9 (512 bytes)
82 * child->vdev_physical_ashift = 9 (512 bytes)
83 * zfs_max_auto_ashift = 13 (8192 bytes)
84 * zfs_min_auto_ashift = 12 (4096 bytes)
86 * On pool creation or the addition of a new top-level vdev, ZFS will
87 * increase the ashift of the top-level vdev to 4096 to match the
88 * zfs_min_auto_ashift.
90 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
91 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
94 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
99 val = zfs_max_auto_ashift;
100 err = sysctl_handle_64(oidp, &val, 0, req);
101 if (err != 0 || req->newptr == NULL)
104 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
107 zfs_max_auto_ashift = val;
111 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
112 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
113 sysctl_vfs_zfs_max_auto_ashift, "QU",
114 "Max ashift used when optimising for logical -> physical sectors size on "
115 "new top-level vdevs.");
118 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
123 val = zfs_min_auto_ashift;
124 err = sysctl_handle_64(oidp, &val, 0, req);
125 if (err != 0 || req->newptr == NULL)
128 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
131 zfs_min_auto_ashift = val;
135 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
136 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
137 sysctl_vfs_zfs_min_auto_ashift, "QU",
138 "Min ashift used when creating new top-level vdevs.");
140 static vdev_ops_t *vdev_ops_table[] = {
159 * When a vdev is added, it will be divided into approximately (but no
160 * more than) this number of metaslabs.
162 int metaslabs_per_vdev = 200;
163 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
164 &metaslabs_per_vdev, 0,
165 "When a vdev is added, how many metaslabs the vdev should be divided into");
168 * Given a vdev type, return the appropriate ops vector.
171 vdev_getops(const char *type)
173 vdev_ops_t *ops, **opspp;
175 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
176 if (strcmp(ops->vdev_op_type, type) == 0)
183 * Default asize function: return the MAX of psize with the asize of
184 * all children. This is what's used by anything other than RAID-Z.
187 vdev_default_asize(vdev_t *vd, uint64_t psize)
189 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
192 for (int c = 0; c < vd->vdev_children; c++) {
193 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
194 asize = MAX(asize, csize);
201 * Get the minimum allocatable size. We define the allocatable size as
202 * the vdev's asize rounded to the nearest metaslab. This allows us to
203 * replace or attach devices which don't have the same physical size but
204 * can still satisfy the same number of allocations.
207 vdev_get_min_asize(vdev_t *vd)
209 vdev_t *pvd = vd->vdev_parent;
212 * If our parent is NULL (inactive spare or cache) or is the root,
213 * just return our own asize.
216 return (vd->vdev_asize);
219 * The top-level vdev just returns the allocatable size rounded
220 * to the nearest metaslab.
222 if (vd == vd->vdev_top)
223 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
226 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
227 * so each child must provide at least 1/Nth of its asize.
229 if (pvd->vdev_ops == &vdev_raidz_ops)
230 return (pvd->vdev_min_asize / pvd->vdev_children);
232 return (pvd->vdev_min_asize);
236 vdev_set_min_asize(vdev_t *vd)
238 vd->vdev_min_asize = vdev_get_min_asize(vd);
240 for (int c = 0; c < vd->vdev_children; c++)
241 vdev_set_min_asize(vd->vdev_child[c]);
245 vdev_lookup_top(spa_t *spa, uint64_t vdev)
247 vdev_t *rvd = spa->spa_root_vdev;
249 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
251 if (vdev < rvd->vdev_children) {
252 ASSERT(rvd->vdev_child[vdev] != NULL);
253 return (rvd->vdev_child[vdev]);
260 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
264 if (vd->vdev_guid == guid)
267 for (int c = 0; c < vd->vdev_children; c++)
268 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
276 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
278 size_t oldsize, newsize;
279 uint64_t id = cvd->vdev_id;
282 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
283 ASSERT(cvd->vdev_parent == NULL);
285 cvd->vdev_parent = pvd;
290 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
292 oldsize = pvd->vdev_children * sizeof (vdev_t *);
293 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
294 newsize = pvd->vdev_children * sizeof (vdev_t *);
296 newchild = kmem_zalloc(newsize, KM_SLEEP);
297 if (pvd->vdev_child != NULL) {
298 bcopy(pvd->vdev_child, newchild, oldsize);
299 kmem_free(pvd->vdev_child, oldsize);
302 pvd->vdev_child = newchild;
303 pvd->vdev_child[id] = cvd;
305 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
306 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
309 * Walk up all ancestors to update guid sum.
311 for (; pvd != NULL; pvd = pvd->vdev_parent)
312 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
316 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
319 uint_t id = cvd->vdev_id;
321 ASSERT(cvd->vdev_parent == pvd);
326 ASSERT(id < pvd->vdev_children);
327 ASSERT(pvd->vdev_child[id] == cvd);
329 pvd->vdev_child[id] = NULL;
330 cvd->vdev_parent = NULL;
332 for (c = 0; c < pvd->vdev_children; c++)
333 if (pvd->vdev_child[c])
336 if (c == pvd->vdev_children) {
337 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
338 pvd->vdev_child = NULL;
339 pvd->vdev_children = 0;
343 * Walk up all ancestors to update guid sum.
345 for (; pvd != NULL; pvd = pvd->vdev_parent)
346 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
350 * Remove any holes in the child array.
353 vdev_compact_children(vdev_t *pvd)
355 vdev_t **newchild, *cvd;
356 int oldc = pvd->vdev_children;
359 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
361 for (int c = newc = 0; c < oldc; c++)
362 if (pvd->vdev_child[c])
365 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
367 for (int c = newc = 0; c < oldc; c++) {
368 if ((cvd = pvd->vdev_child[c]) != NULL) {
369 newchild[newc] = cvd;
370 cvd->vdev_id = newc++;
374 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
375 pvd->vdev_child = newchild;
376 pvd->vdev_children = newc;
380 * Allocate and minimally initialize a vdev_t.
383 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
387 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
389 if (spa->spa_root_vdev == NULL) {
390 ASSERT(ops == &vdev_root_ops);
391 spa->spa_root_vdev = vd;
392 spa->spa_load_guid = spa_generate_guid(NULL);
395 if (guid == 0 && ops != &vdev_hole_ops) {
396 if (spa->spa_root_vdev == vd) {
398 * The root vdev's guid will also be the pool guid,
399 * which must be unique among all pools.
401 guid = spa_generate_guid(NULL);
404 * Any other vdev's guid must be unique within the pool.
406 guid = spa_generate_guid(spa);
408 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
413 vd->vdev_guid = guid;
414 vd->vdev_guid_sum = guid;
416 vd->vdev_state = VDEV_STATE_CLOSED;
417 vd->vdev_ishole = (ops == &vdev_hole_ops);
419 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
420 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
421 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
422 for (int t = 0; t < DTL_TYPES; t++) {
423 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
426 txg_list_create(&vd->vdev_ms_list,
427 offsetof(struct metaslab, ms_txg_node));
428 txg_list_create(&vd->vdev_dtl_list,
429 offsetof(struct vdev, vdev_dtl_node));
430 vd->vdev_stat.vs_timestamp = gethrtime();
438 * Allocate a new vdev. The 'alloctype' is used to control whether we are
439 * creating a new vdev or loading an existing one - the behavior is slightly
440 * different for each case.
443 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
448 uint64_t guid = 0, islog, nparity;
451 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
453 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
454 return (SET_ERROR(EINVAL));
456 if ((ops = vdev_getops(type)) == NULL)
457 return (SET_ERROR(EINVAL));
460 * If this is a load, get the vdev guid from the nvlist.
461 * Otherwise, vdev_alloc_common() will generate one for us.
463 if (alloctype == VDEV_ALLOC_LOAD) {
466 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
468 return (SET_ERROR(EINVAL));
470 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
471 return (SET_ERROR(EINVAL));
472 } else if (alloctype == VDEV_ALLOC_SPARE) {
473 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
474 return (SET_ERROR(EINVAL));
475 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
476 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
477 return (SET_ERROR(EINVAL));
478 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
479 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
480 return (SET_ERROR(EINVAL));
484 * The first allocated vdev must be of type 'root'.
486 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
487 return (SET_ERROR(EINVAL));
490 * Determine whether we're a log vdev.
493 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
494 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
495 return (SET_ERROR(ENOTSUP));
497 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
498 return (SET_ERROR(ENOTSUP));
501 * Set the nparity property for RAID-Z vdevs.
504 if (ops == &vdev_raidz_ops) {
505 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
507 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
508 return (SET_ERROR(EINVAL));
510 * Previous versions could only support 1 or 2 parity
514 spa_version(spa) < SPA_VERSION_RAIDZ2)
515 return (SET_ERROR(ENOTSUP));
517 spa_version(spa) < SPA_VERSION_RAIDZ3)
518 return (SET_ERROR(ENOTSUP));
521 * We require the parity to be specified for SPAs that
522 * support multiple parity levels.
524 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
525 return (SET_ERROR(EINVAL));
527 * Otherwise, we default to 1 parity device for RAID-Z.
534 ASSERT(nparity != -1ULL);
536 vd = vdev_alloc_common(spa, id, guid, ops);
538 vd->vdev_islog = islog;
539 vd->vdev_nparity = nparity;
541 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
542 vd->vdev_path = spa_strdup(vd->vdev_path);
543 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
544 vd->vdev_devid = spa_strdup(vd->vdev_devid);
545 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
546 &vd->vdev_physpath) == 0)
547 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
548 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
549 vd->vdev_fru = spa_strdup(vd->vdev_fru);
552 * Set the whole_disk property. If it's not specified, leave the value
555 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
556 &vd->vdev_wholedisk) != 0)
557 vd->vdev_wholedisk = -1ULL;
560 * Look for the 'not present' flag. This will only be set if the device
561 * was not present at the time of import.
563 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
564 &vd->vdev_not_present);
567 * Get the alignment requirement.
569 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
572 * Retrieve the vdev creation time.
574 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
578 * If we're a top-level vdev, try to load the allocation parameters.
580 if (parent && !parent->vdev_parent &&
581 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
582 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
584 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
586 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
588 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
592 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
593 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
594 alloctype == VDEV_ALLOC_ADD ||
595 alloctype == VDEV_ALLOC_SPLIT ||
596 alloctype == VDEV_ALLOC_ROOTPOOL);
597 vd->vdev_mg = metaslab_group_create(islog ?
598 spa_log_class(spa) : spa_normal_class(spa), vd);
602 * If we're a leaf vdev, try to load the DTL object and other state.
604 if (vd->vdev_ops->vdev_op_leaf &&
605 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
606 alloctype == VDEV_ALLOC_ROOTPOOL)) {
607 if (alloctype == VDEV_ALLOC_LOAD) {
608 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
609 &vd->vdev_dtl_object);
610 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
614 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
617 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
618 &spare) == 0 && spare)
622 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
625 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
626 &vd->vdev_resilver_txg);
629 * When importing a pool, we want to ignore the persistent fault
630 * state, as the diagnosis made on another system may not be
631 * valid in the current context. Local vdevs will
632 * remain in the faulted state.
634 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
635 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
637 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
639 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
642 if (vd->vdev_faulted || vd->vdev_degraded) {
646 VDEV_AUX_ERR_EXCEEDED;
647 if (nvlist_lookup_string(nv,
648 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
649 strcmp(aux, "external") == 0)
650 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
656 * Add ourselves to the parent's list of children.
658 vdev_add_child(parent, vd);
666 vdev_free(vdev_t *vd)
668 spa_t *spa = vd->vdev_spa;
671 * vdev_free() implies closing the vdev first. This is simpler than
672 * trying to ensure complicated semantics for all callers.
676 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
677 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
682 for (int c = 0; c < vd->vdev_children; c++)
683 vdev_free(vd->vdev_child[c]);
685 ASSERT(vd->vdev_child == NULL);
686 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
689 * Discard allocation state.
691 if (vd->vdev_mg != NULL) {
692 vdev_metaslab_fini(vd);
693 metaslab_group_destroy(vd->vdev_mg);
696 ASSERT0(vd->vdev_stat.vs_space);
697 ASSERT0(vd->vdev_stat.vs_dspace);
698 ASSERT0(vd->vdev_stat.vs_alloc);
701 * Remove this vdev from its parent's child list.
703 vdev_remove_child(vd->vdev_parent, vd);
705 ASSERT(vd->vdev_parent == NULL);
708 * Clean up vdev structure.
714 spa_strfree(vd->vdev_path);
716 spa_strfree(vd->vdev_devid);
717 if (vd->vdev_physpath)
718 spa_strfree(vd->vdev_physpath);
720 spa_strfree(vd->vdev_fru);
722 if (vd->vdev_isspare)
723 spa_spare_remove(vd);
724 if (vd->vdev_isl2cache)
725 spa_l2cache_remove(vd);
727 txg_list_destroy(&vd->vdev_ms_list);
728 txg_list_destroy(&vd->vdev_dtl_list);
730 mutex_enter(&vd->vdev_dtl_lock);
731 space_map_close(vd->vdev_dtl_sm);
732 for (int t = 0; t < DTL_TYPES; t++) {
733 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
734 range_tree_destroy(vd->vdev_dtl[t]);
736 mutex_exit(&vd->vdev_dtl_lock);
738 mutex_destroy(&vd->vdev_dtl_lock);
739 mutex_destroy(&vd->vdev_stat_lock);
740 mutex_destroy(&vd->vdev_probe_lock);
742 if (vd == spa->spa_root_vdev)
743 spa->spa_root_vdev = NULL;
745 kmem_free(vd, sizeof (vdev_t));
749 * Transfer top-level vdev state from svd to tvd.
752 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
754 spa_t *spa = svd->vdev_spa;
759 ASSERT(tvd == tvd->vdev_top);
761 tvd->vdev_ms_array = svd->vdev_ms_array;
762 tvd->vdev_ms_shift = svd->vdev_ms_shift;
763 tvd->vdev_ms_count = svd->vdev_ms_count;
765 svd->vdev_ms_array = 0;
766 svd->vdev_ms_shift = 0;
767 svd->vdev_ms_count = 0;
770 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
771 tvd->vdev_mg = svd->vdev_mg;
772 tvd->vdev_ms = svd->vdev_ms;
777 if (tvd->vdev_mg != NULL)
778 tvd->vdev_mg->mg_vd = tvd;
780 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
781 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
782 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
784 svd->vdev_stat.vs_alloc = 0;
785 svd->vdev_stat.vs_space = 0;
786 svd->vdev_stat.vs_dspace = 0;
788 for (t = 0; t < TXG_SIZE; t++) {
789 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
790 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
791 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
792 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
793 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
794 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
797 if (list_link_active(&svd->vdev_config_dirty_node)) {
798 vdev_config_clean(svd);
799 vdev_config_dirty(tvd);
802 if (list_link_active(&svd->vdev_state_dirty_node)) {
803 vdev_state_clean(svd);
804 vdev_state_dirty(tvd);
807 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
808 svd->vdev_deflate_ratio = 0;
810 tvd->vdev_islog = svd->vdev_islog;
815 vdev_top_update(vdev_t *tvd, vdev_t *vd)
822 for (int c = 0; c < vd->vdev_children; c++)
823 vdev_top_update(tvd, vd->vdev_child[c]);
827 * Add a mirror/replacing vdev above an existing vdev.
830 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
832 spa_t *spa = cvd->vdev_spa;
833 vdev_t *pvd = cvd->vdev_parent;
836 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
838 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
840 mvd->vdev_asize = cvd->vdev_asize;
841 mvd->vdev_min_asize = cvd->vdev_min_asize;
842 mvd->vdev_max_asize = cvd->vdev_max_asize;
843 mvd->vdev_ashift = cvd->vdev_ashift;
844 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
845 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
846 mvd->vdev_state = cvd->vdev_state;
847 mvd->vdev_crtxg = cvd->vdev_crtxg;
849 vdev_remove_child(pvd, cvd);
850 vdev_add_child(pvd, mvd);
851 cvd->vdev_id = mvd->vdev_children;
852 vdev_add_child(mvd, cvd);
853 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
855 if (mvd == mvd->vdev_top)
856 vdev_top_transfer(cvd, mvd);
862 * Remove a 1-way mirror/replacing vdev from the tree.
865 vdev_remove_parent(vdev_t *cvd)
867 vdev_t *mvd = cvd->vdev_parent;
868 vdev_t *pvd = mvd->vdev_parent;
870 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
872 ASSERT(mvd->vdev_children == 1);
873 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
874 mvd->vdev_ops == &vdev_replacing_ops ||
875 mvd->vdev_ops == &vdev_spare_ops);
876 cvd->vdev_ashift = mvd->vdev_ashift;
877 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
878 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
880 vdev_remove_child(mvd, cvd);
881 vdev_remove_child(pvd, mvd);
884 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
885 * Otherwise, we could have detached an offline device, and when we
886 * go to import the pool we'll think we have two top-level vdevs,
887 * instead of a different version of the same top-level vdev.
889 if (mvd->vdev_top == mvd) {
890 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
891 cvd->vdev_orig_guid = cvd->vdev_guid;
892 cvd->vdev_guid += guid_delta;
893 cvd->vdev_guid_sum += guid_delta;
895 cvd->vdev_id = mvd->vdev_id;
896 vdev_add_child(pvd, cvd);
897 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
899 if (cvd == cvd->vdev_top)
900 vdev_top_transfer(mvd, cvd);
902 ASSERT(mvd->vdev_children == 0);
907 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
909 spa_t *spa = vd->vdev_spa;
910 objset_t *mos = spa->spa_meta_objset;
912 uint64_t oldc = vd->vdev_ms_count;
913 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
917 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
920 * This vdev is not being allocated from yet or is a hole.
922 if (vd->vdev_ms_shift == 0)
925 ASSERT(!vd->vdev_ishole);
928 * Compute the raidz-deflation ratio. Note, we hard-code
929 * in 128k (1 << 17) because it is the current "typical" blocksize.
930 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
931 * or we will inconsistently account for existing bp's.
933 vd->vdev_deflate_ratio = (1 << 17) /
934 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
936 ASSERT(oldc <= newc);
938 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
941 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
942 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
946 vd->vdev_ms_count = newc;
948 for (m = oldc; m < newc; m++) {
952 error = dmu_read(mos, vd->vdev_ms_array,
953 m * sizeof (uint64_t), sizeof (uint64_t), &object,
958 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg);
962 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
965 * If the vdev is being removed we don't activate
966 * the metaslabs since we want to ensure that no new
967 * allocations are performed on this device.
969 if (oldc == 0 && !vd->vdev_removing)
970 metaslab_group_activate(vd->vdev_mg);
973 spa_config_exit(spa, SCL_ALLOC, FTAG);
979 vdev_metaslab_fini(vdev_t *vd)
982 uint64_t count = vd->vdev_ms_count;
984 if (vd->vdev_ms != NULL) {
985 metaslab_group_passivate(vd->vdev_mg);
986 for (m = 0; m < count; m++) {
987 metaslab_t *msp = vd->vdev_ms[m];
992 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
997 typedef struct vdev_probe_stats {
998 boolean_t vps_readable;
999 boolean_t vps_writeable;
1001 } vdev_probe_stats_t;
1004 vdev_probe_done(zio_t *zio)
1006 spa_t *spa = zio->io_spa;
1007 vdev_t *vd = zio->io_vd;
1008 vdev_probe_stats_t *vps = zio->io_private;
1010 ASSERT(vd->vdev_probe_zio != NULL);
1012 if (zio->io_type == ZIO_TYPE_READ) {
1013 if (zio->io_error == 0)
1014 vps->vps_readable = 1;
1015 if (zio->io_error == 0 && spa_writeable(spa)) {
1016 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1017 zio->io_offset, zio->io_size, zio->io_data,
1018 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1019 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1021 zio_buf_free(zio->io_data, zio->io_size);
1023 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1024 if (zio->io_error == 0)
1025 vps->vps_writeable = 1;
1026 zio_buf_free(zio->io_data, zio->io_size);
1027 } else if (zio->io_type == ZIO_TYPE_NULL) {
1030 vd->vdev_cant_read |= !vps->vps_readable;
1031 vd->vdev_cant_write |= !vps->vps_writeable;
1033 if (vdev_readable(vd) &&
1034 (vdev_writeable(vd) || !spa_writeable(spa))) {
1037 ASSERT(zio->io_error != 0);
1038 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1039 spa, vd, NULL, 0, 0);
1040 zio->io_error = SET_ERROR(ENXIO);
1043 mutex_enter(&vd->vdev_probe_lock);
1044 ASSERT(vd->vdev_probe_zio == zio);
1045 vd->vdev_probe_zio = NULL;
1046 mutex_exit(&vd->vdev_probe_lock);
1048 while ((pio = zio_walk_parents(zio)) != NULL)
1049 if (!vdev_accessible(vd, pio))
1050 pio->io_error = SET_ERROR(ENXIO);
1052 kmem_free(vps, sizeof (*vps));
1057 * Determine whether this device is accessible.
1059 * Read and write to several known locations: the pad regions of each
1060 * vdev label but the first, which we leave alone in case it contains
1064 vdev_probe(vdev_t *vd, zio_t *zio)
1066 spa_t *spa = vd->vdev_spa;
1067 vdev_probe_stats_t *vps = NULL;
1070 ASSERT(vd->vdev_ops->vdev_op_leaf);
1073 * Don't probe the probe.
1075 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1079 * To prevent 'probe storms' when a device fails, we create
1080 * just one probe i/o at a time. All zios that want to probe
1081 * this vdev will become parents of the probe io.
1083 mutex_enter(&vd->vdev_probe_lock);
1085 if ((pio = vd->vdev_probe_zio) == NULL) {
1086 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1088 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1089 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1092 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1094 * vdev_cant_read and vdev_cant_write can only
1095 * transition from TRUE to FALSE when we have the
1096 * SCL_ZIO lock as writer; otherwise they can only
1097 * transition from FALSE to TRUE. This ensures that
1098 * any zio looking at these values can assume that
1099 * failures persist for the life of the I/O. That's
1100 * important because when a device has intermittent
1101 * connectivity problems, we want to ensure that
1102 * they're ascribed to the device (ENXIO) and not
1105 * Since we hold SCL_ZIO as writer here, clear both
1106 * values so the probe can reevaluate from first
1109 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1110 vd->vdev_cant_read = B_FALSE;
1111 vd->vdev_cant_write = B_FALSE;
1114 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1115 vdev_probe_done, vps,
1116 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1119 * We can't change the vdev state in this context, so we
1120 * kick off an async task to do it on our behalf.
1123 vd->vdev_probe_wanted = B_TRUE;
1124 spa_async_request(spa, SPA_ASYNC_PROBE);
1129 zio_add_child(zio, pio);
1131 mutex_exit(&vd->vdev_probe_lock);
1134 ASSERT(zio != NULL);
1138 for (int l = 1; l < VDEV_LABELS; l++) {
1139 zio_nowait(zio_read_phys(pio, vd,
1140 vdev_label_offset(vd->vdev_psize, l,
1141 offsetof(vdev_label_t, vl_pad2)),
1142 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1143 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1144 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1155 vdev_open_child(void *arg)
1159 vd->vdev_open_thread = curthread;
1160 vd->vdev_open_error = vdev_open(vd);
1161 vd->vdev_open_thread = NULL;
1165 vdev_uses_zvols(vdev_t *vd)
1167 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1168 strlen(ZVOL_DIR)) == 0)
1170 for (int c = 0; c < vd->vdev_children; c++)
1171 if (vdev_uses_zvols(vd->vdev_child[c]))
1177 vdev_open_children(vdev_t *vd)
1180 int children = vd->vdev_children;
1183 * in order to handle pools on top of zvols, do the opens
1184 * in a single thread so that the same thread holds the
1185 * spa_namespace_lock
1187 if (B_TRUE || vdev_uses_zvols(vd)) {
1188 for (int c = 0; c < children; c++)
1189 vd->vdev_child[c]->vdev_open_error =
1190 vdev_open(vd->vdev_child[c]);
1193 tq = taskq_create("vdev_open", children, minclsyspri,
1194 children, children, TASKQ_PREPOPULATE);
1196 for (int c = 0; c < children; c++)
1197 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1204 * Prepare a virtual device for access.
1207 vdev_open(vdev_t *vd)
1209 spa_t *spa = vd->vdev_spa;
1212 uint64_t max_osize = 0;
1213 uint64_t asize, max_asize, psize;
1214 uint64_t logical_ashift = 0;
1215 uint64_t physical_ashift = 0;
1217 ASSERT(vd->vdev_open_thread == curthread ||
1218 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1219 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1220 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1221 vd->vdev_state == VDEV_STATE_OFFLINE);
1223 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1224 vd->vdev_cant_read = B_FALSE;
1225 vd->vdev_cant_write = B_FALSE;
1226 vd->vdev_min_asize = vdev_get_min_asize(vd);
1229 * If this vdev is not removed, check its fault status. If it's
1230 * faulted, bail out of the open.
1232 if (!vd->vdev_removed && vd->vdev_faulted) {
1233 ASSERT(vd->vdev_children == 0);
1234 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1235 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1236 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1237 vd->vdev_label_aux);
1238 return (SET_ERROR(ENXIO));
1239 } else if (vd->vdev_offline) {
1240 ASSERT(vd->vdev_children == 0);
1241 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1242 return (SET_ERROR(ENXIO));
1245 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1246 &logical_ashift, &physical_ashift);
1249 * Reset the vdev_reopening flag so that we actually close
1250 * the vdev on error.
1252 vd->vdev_reopening = B_FALSE;
1253 if (zio_injection_enabled && error == 0)
1254 error = zio_handle_device_injection(vd, NULL, ENXIO);
1257 if (vd->vdev_removed &&
1258 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1259 vd->vdev_removed = B_FALSE;
1261 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1262 vd->vdev_stat.vs_aux);
1266 vd->vdev_removed = B_FALSE;
1269 * Recheck the faulted flag now that we have confirmed that
1270 * the vdev is accessible. If we're faulted, bail.
1272 if (vd->vdev_faulted) {
1273 ASSERT(vd->vdev_children == 0);
1274 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1275 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1276 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1277 vd->vdev_label_aux);
1278 return (SET_ERROR(ENXIO));
1281 if (vd->vdev_degraded) {
1282 ASSERT(vd->vdev_children == 0);
1283 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1284 VDEV_AUX_ERR_EXCEEDED);
1286 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1290 * For hole or missing vdevs we just return success.
1292 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1295 if (vd->vdev_ops->vdev_op_leaf) {
1296 vd->vdev_notrim = B_FALSE;
1297 trim_map_create(vd);
1300 for (int c = 0; c < vd->vdev_children; c++) {
1301 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1302 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1308 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1309 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1311 if (vd->vdev_children == 0) {
1312 if (osize < SPA_MINDEVSIZE) {
1313 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1314 VDEV_AUX_TOO_SMALL);
1315 return (SET_ERROR(EOVERFLOW));
1318 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1319 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1320 VDEV_LABEL_END_SIZE);
1322 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1323 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1324 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1325 VDEV_AUX_TOO_SMALL);
1326 return (SET_ERROR(EOVERFLOW));
1330 max_asize = max_osize;
1333 vd->vdev_psize = psize;
1336 * Make sure the allocatable size hasn't shrunk.
1338 if (asize < vd->vdev_min_asize) {
1339 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1340 VDEV_AUX_BAD_LABEL);
1341 return (SET_ERROR(EINVAL));
1344 vd->vdev_physical_ashift =
1345 MAX(physical_ashift, vd->vdev_physical_ashift);
1346 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1347 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1349 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1350 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1351 VDEV_AUX_ASHIFT_TOO_BIG);
1355 if (vd->vdev_asize == 0) {
1357 * This is the first-ever open, so use the computed values.
1358 * For testing purposes, a higher ashift can be requested.
1360 vd->vdev_asize = asize;
1361 vd->vdev_max_asize = max_asize;
1364 * Make sure the alignment requirement hasn't increased.
1366 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1367 vd->vdev_ops->vdev_op_leaf) {
1368 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1369 VDEV_AUX_BAD_LABEL);
1372 vd->vdev_max_asize = max_asize;
1376 * If all children are healthy and the asize has increased,
1377 * then we've experienced dynamic LUN growth. If automatic
1378 * expansion is enabled then use the additional space.
1380 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1381 (vd->vdev_expanding || spa->spa_autoexpand))
1382 vd->vdev_asize = asize;
1384 vdev_set_min_asize(vd);
1387 * Ensure we can issue some IO before declaring the
1388 * vdev open for business.
1390 if (vd->vdev_ops->vdev_op_leaf &&
1391 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1392 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1393 VDEV_AUX_ERR_EXCEEDED);
1398 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1399 * resilver. But don't do this if we are doing a reopen for a scrub,
1400 * since this would just restart the scrub we are already doing.
1402 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1403 vdev_resilver_needed(vd, NULL, NULL))
1404 spa_async_request(spa, SPA_ASYNC_RESILVER);
1410 * Called once the vdevs are all opened, this routine validates the label
1411 * contents. This needs to be done before vdev_load() so that we don't
1412 * inadvertently do repair I/Os to the wrong device.
1414 * If 'strict' is false ignore the spa guid check. This is necessary because
1415 * if the machine crashed during a re-guid the new guid might have been written
1416 * to all of the vdev labels, but not the cached config. The strict check
1417 * will be performed when the pool is opened again using the mos config.
1419 * This function will only return failure if one of the vdevs indicates that it
1420 * has since been destroyed or exported. This is only possible if
1421 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1422 * will be updated but the function will return 0.
1425 vdev_validate(vdev_t *vd, boolean_t strict)
1427 spa_t *spa = vd->vdev_spa;
1429 uint64_t guid = 0, top_guid;
1432 for (int c = 0; c < vd->vdev_children; c++)
1433 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1434 return (SET_ERROR(EBADF));
1437 * If the device has already failed, or was marked offline, don't do
1438 * any further validation. Otherwise, label I/O will fail and we will
1439 * overwrite the previous state.
1441 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1442 uint64_t aux_guid = 0;
1444 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1445 spa_last_synced_txg(spa) : -1ULL;
1447 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1448 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1449 VDEV_AUX_BAD_LABEL);
1454 * Determine if this vdev has been split off into another
1455 * pool. If so, then refuse to open it.
1457 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1458 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1459 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1460 VDEV_AUX_SPLIT_POOL);
1465 if (strict && (nvlist_lookup_uint64(label,
1466 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1467 guid != spa_guid(spa))) {
1468 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1469 VDEV_AUX_CORRUPT_DATA);
1474 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1475 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1480 * If this vdev just became a top-level vdev because its
1481 * sibling was detached, it will have adopted the parent's
1482 * vdev guid -- but the label may or may not be on disk yet.
1483 * Fortunately, either version of the label will have the
1484 * same top guid, so if we're a top-level vdev, we can
1485 * safely compare to that instead.
1487 * If we split this vdev off instead, then we also check the
1488 * original pool's guid. We don't want to consider the vdev
1489 * corrupt if it is partway through a split operation.
1491 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1493 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1495 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1496 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1497 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1498 VDEV_AUX_CORRUPT_DATA);
1503 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1505 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1506 VDEV_AUX_CORRUPT_DATA);
1514 * If this is a verbatim import, no need to check the
1515 * state of the pool.
1517 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1518 spa_load_state(spa) == SPA_LOAD_OPEN &&
1519 state != POOL_STATE_ACTIVE)
1520 return (SET_ERROR(EBADF));
1523 * If we were able to open and validate a vdev that was
1524 * previously marked permanently unavailable, clear that state
1527 if (vd->vdev_not_present)
1528 vd->vdev_not_present = 0;
1535 * Close a virtual device.
1538 vdev_close(vdev_t *vd)
1540 spa_t *spa = vd->vdev_spa;
1541 vdev_t *pvd = vd->vdev_parent;
1543 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1546 * If our parent is reopening, then we are as well, unless we are
1549 if (pvd != NULL && pvd->vdev_reopening)
1550 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1552 vd->vdev_ops->vdev_op_close(vd);
1554 vdev_cache_purge(vd);
1556 if (vd->vdev_ops->vdev_op_leaf)
1557 trim_map_destroy(vd);
1560 * We record the previous state before we close it, so that if we are
1561 * doing a reopen(), we don't generate FMA ereports if we notice that
1562 * it's still faulted.
1564 vd->vdev_prevstate = vd->vdev_state;
1566 if (vd->vdev_offline)
1567 vd->vdev_state = VDEV_STATE_OFFLINE;
1569 vd->vdev_state = VDEV_STATE_CLOSED;
1570 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1574 vdev_hold(vdev_t *vd)
1576 spa_t *spa = vd->vdev_spa;
1578 ASSERT(spa_is_root(spa));
1579 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1582 for (int c = 0; c < vd->vdev_children; c++)
1583 vdev_hold(vd->vdev_child[c]);
1585 if (vd->vdev_ops->vdev_op_leaf)
1586 vd->vdev_ops->vdev_op_hold(vd);
1590 vdev_rele(vdev_t *vd)
1592 spa_t *spa = vd->vdev_spa;
1594 ASSERT(spa_is_root(spa));
1595 for (int c = 0; c < vd->vdev_children; c++)
1596 vdev_rele(vd->vdev_child[c]);
1598 if (vd->vdev_ops->vdev_op_leaf)
1599 vd->vdev_ops->vdev_op_rele(vd);
1603 * Reopen all interior vdevs and any unopened leaves. We don't actually
1604 * reopen leaf vdevs which had previously been opened as they might deadlock
1605 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1606 * If the leaf has never been opened then open it, as usual.
1609 vdev_reopen(vdev_t *vd)
1611 spa_t *spa = vd->vdev_spa;
1613 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1615 /* set the reopening flag unless we're taking the vdev offline */
1616 vd->vdev_reopening = !vd->vdev_offline;
1618 (void) vdev_open(vd);
1621 * Call vdev_validate() here to make sure we have the same device.
1622 * Otherwise, a device with an invalid label could be successfully
1623 * opened in response to vdev_reopen().
1626 (void) vdev_validate_aux(vd);
1627 if (vdev_readable(vd) && vdev_writeable(vd) &&
1628 vd->vdev_aux == &spa->spa_l2cache &&
1629 !l2arc_vdev_present(vd))
1630 l2arc_add_vdev(spa, vd);
1632 (void) vdev_validate(vd, B_TRUE);
1636 * Reassess parent vdev's health.
1638 vdev_propagate_state(vd);
1642 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1647 * Normally, partial opens (e.g. of a mirror) are allowed.
1648 * For a create, however, we want to fail the request if
1649 * there are any components we can't open.
1651 error = vdev_open(vd);
1653 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1655 return (error ? error : ENXIO);
1659 * Recursively load DTLs and initialize all labels.
1661 if ((error = vdev_dtl_load(vd)) != 0 ||
1662 (error = vdev_label_init(vd, txg, isreplacing ?
1663 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1672 vdev_metaslab_set_size(vdev_t *vd)
1675 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1677 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1678 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1682 * Maximize performance by inflating the configured ashift for top level
1683 * vdevs to be as close to the physical ashift as possible while maintaining
1684 * administrator defined limits and ensuring it doesn't go below the
1688 vdev_ashift_optimize(vdev_t *vd)
1690 if (vd == vd->vdev_top) {
1691 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1692 vd->vdev_ashift = MIN(
1693 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1694 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1697 * Unusual case where logical ashift > physical ashift
1698 * so we can't cap the calculated ashift based on max
1699 * ashift as that would cause failures.
1700 * We still check if we need to increase it to match
1703 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1710 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1712 ASSERT(vd == vd->vdev_top);
1713 ASSERT(!vd->vdev_ishole);
1714 ASSERT(ISP2(flags));
1715 ASSERT(spa_writeable(vd->vdev_spa));
1717 if (flags & VDD_METASLAB)
1718 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1720 if (flags & VDD_DTL)
1721 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1723 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1727 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1729 for (int c = 0; c < vd->vdev_children; c++)
1730 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1732 if (vd->vdev_ops->vdev_op_leaf)
1733 vdev_dirty(vd->vdev_top, flags, vd, txg);
1739 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1740 * the vdev has less than perfect replication. There are four kinds of DTL:
1742 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1744 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1746 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1747 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1748 * txgs that was scrubbed.
1750 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1751 * persistent errors or just some device being offline.
1752 * Unlike the other three, the DTL_OUTAGE map is not generally
1753 * maintained; it's only computed when needed, typically to
1754 * determine whether a device can be detached.
1756 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1757 * either has the data or it doesn't.
1759 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1760 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1761 * if any child is less than fully replicated, then so is its parent.
1762 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1763 * comprising only those txgs which appear in 'maxfaults' or more children;
1764 * those are the txgs we don't have enough replication to read. For example,
1765 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1766 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1767 * two child DTL_MISSING maps.
1769 * It should be clear from the above that to compute the DTLs and outage maps
1770 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1771 * Therefore, that is all we keep on disk. When loading the pool, or after
1772 * a configuration change, we generate all other DTLs from first principles.
1775 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1777 range_tree_t *rt = vd->vdev_dtl[t];
1779 ASSERT(t < DTL_TYPES);
1780 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1781 ASSERT(spa_writeable(vd->vdev_spa));
1783 mutex_enter(rt->rt_lock);
1784 if (!range_tree_contains(rt, txg, size))
1785 range_tree_add(rt, txg, size);
1786 mutex_exit(rt->rt_lock);
1790 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1792 range_tree_t *rt = vd->vdev_dtl[t];
1793 boolean_t dirty = B_FALSE;
1795 ASSERT(t < DTL_TYPES);
1796 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1798 mutex_enter(rt->rt_lock);
1799 if (range_tree_space(rt) != 0)
1800 dirty = range_tree_contains(rt, txg, size);
1801 mutex_exit(rt->rt_lock);
1807 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1809 range_tree_t *rt = vd->vdev_dtl[t];
1812 mutex_enter(rt->rt_lock);
1813 empty = (range_tree_space(rt) == 0);
1814 mutex_exit(rt->rt_lock);
1820 * Returns the lowest txg in the DTL range.
1823 vdev_dtl_min(vdev_t *vd)
1827 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1828 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1829 ASSERT0(vd->vdev_children);
1831 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1832 return (rs->rs_start - 1);
1836 * Returns the highest txg in the DTL.
1839 vdev_dtl_max(vdev_t *vd)
1843 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1844 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1845 ASSERT0(vd->vdev_children);
1847 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1848 return (rs->rs_end);
1852 * Determine if a resilvering vdev should remove any DTL entries from
1853 * its range. If the vdev was resilvering for the entire duration of the
1854 * scan then it should excise that range from its DTLs. Otherwise, this
1855 * vdev is considered partially resilvered and should leave its DTL
1856 * entries intact. The comment in vdev_dtl_reassess() describes how we
1860 vdev_dtl_should_excise(vdev_t *vd)
1862 spa_t *spa = vd->vdev_spa;
1863 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1865 ASSERT0(scn->scn_phys.scn_errors);
1866 ASSERT0(vd->vdev_children);
1868 if (vd->vdev_resilver_txg == 0 ||
1869 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1873 * When a resilver is initiated the scan will assign the scn_max_txg
1874 * value to the highest txg value that exists in all DTLs. If this
1875 * device's max DTL is not part of this scan (i.e. it is not in
1876 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1879 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1880 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1881 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1882 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1889 * Reassess DTLs after a config change or scrub completion.
1892 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1894 spa_t *spa = vd->vdev_spa;
1898 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1900 for (int c = 0; c < vd->vdev_children; c++)
1901 vdev_dtl_reassess(vd->vdev_child[c], txg,
1902 scrub_txg, scrub_done);
1904 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1907 if (vd->vdev_ops->vdev_op_leaf) {
1908 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1910 mutex_enter(&vd->vdev_dtl_lock);
1913 * If we've completed a scan cleanly then determine
1914 * if this vdev should remove any DTLs. We only want to
1915 * excise regions on vdevs that were available during
1916 * the entire duration of this scan.
1918 if (scrub_txg != 0 &&
1919 (spa->spa_scrub_started ||
1920 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1921 vdev_dtl_should_excise(vd)) {
1923 * We completed a scrub up to scrub_txg. If we
1924 * did it without rebooting, then the scrub dtl
1925 * will be valid, so excise the old region and
1926 * fold in the scrub dtl. Otherwise, leave the
1927 * dtl as-is if there was an error.
1929 * There's little trick here: to excise the beginning
1930 * of the DTL_MISSING map, we put it into a reference
1931 * tree and then add a segment with refcnt -1 that
1932 * covers the range [0, scrub_txg). This means
1933 * that each txg in that range has refcnt -1 or 0.
1934 * We then add DTL_SCRUB with a refcnt of 2, so that
1935 * entries in the range [0, scrub_txg) will have a
1936 * positive refcnt -- either 1 or 2. We then convert
1937 * the reference tree into the new DTL_MISSING map.
1939 space_reftree_create(&reftree);
1940 space_reftree_add_map(&reftree,
1941 vd->vdev_dtl[DTL_MISSING], 1);
1942 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1943 space_reftree_add_map(&reftree,
1944 vd->vdev_dtl[DTL_SCRUB], 2);
1945 space_reftree_generate_map(&reftree,
1946 vd->vdev_dtl[DTL_MISSING], 1);
1947 space_reftree_destroy(&reftree);
1949 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1950 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1951 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1953 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1954 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1955 if (!vdev_readable(vd))
1956 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1958 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1959 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1962 * If the vdev was resilvering and no longer has any
1963 * DTLs then reset its resilvering flag and dirty
1964 * the top level so that we persist the change.
1966 if (vd->vdev_resilver_txg != 0 &&
1967 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1968 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
1969 vd->vdev_resilver_txg = 0;
1970 vdev_config_dirty(vd->vdev_top);
1973 mutex_exit(&vd->vdev_dtl_lock);
1976 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1980 mutex_enter(&vd->vdev_dtl_lock);
1981 for (int t = 0; t < DTL_TYPES; t++) {
1982 /* account for child's outage in parent's missing map */
1983 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1985 continue; /* leaf vdevs only */
1986 if (t == DTL_PARTIAL)
1987 minref = 1; /* i.e. non-zero */
1988 else if (vd->vdev_nparity != 0)
1989 minref = vd->vdev_nparity + 1; /* RAID-Z */
1991 minref = vd->vdev_children; /* any kind of mirror */
1992 space_reftree_create(&reftree);
1993 for (int c = 0; c < vd->vdev_children; c++) {
1994 vdev_t *cvd = vd->vdev_child[c];
1995 mutex_enter(&cvd->vdev_dtl_lock);
1996 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1997 mutex_exit(&cvd->vdev_dtl_lock);
1999 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2000 space_reftree_destroy(&reftree);
2002 mutex_exit(&vd->vdev_dtl_lock);
2006 vdev_dtl_load(vdev_t *vd)
2008 spa_t *spa = vd->vdev_spa;
2009 objset_t *mos = spa->spa_meta_objset;
2012 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2013 ASSERT(!vd->vdev_ishole);
2015 error = space_map_open(&vd->vdev_dtl_sm, mos,
2016 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2019 ASSERT(vd->vdev_dtl_sm != NULL);
2021 mutex_enter(&vd->vdev_dtl_lock);
2024 * Now that we've opened the space_map we need to update
2027 space_map_update(vd->vdev_dtl_sm);
2029 error = space_map_load(vd->vdev_dtl_sm,
2030 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2031 mutex_exit(&vd->vdev_dtl_lock);
2036 for (int c = 0; c < vd->vdev_children; c++) {
2037 error = vdev_dtl_load(vd->vdev_child[c]);
2046 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2048 spa_t *spa = vd->vdev_spa;
2049 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2050 objset_t *mos = spa->spa_meta_objset;
2051 range_tree_t *rtsync;
2054 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2056 ASSERT(!vd->vdev_ishole);
2057 ASSERT(vd->vdev_ops->vdev_op_leaf);
2059 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2061 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2062 mutex_enter(&vd->vdev_dtl_lock);
2063 space_map_free(vd->vdev_dtl_sm, tx);
2064 space_map_close(vd->vdev_dtl_sm);
2065 vd->vdev_dtl_sm = NULL;
2066 mutex_exit(&vd->vdev_dtl_lock);
2071 if (vd->vdev_dtl_sm == NULL) {
2072 uint64_t new_object;
2074 new_object = space_map_alloc(mos, tx);
2075 VERIFY3U(new_object, !=, 0);
2077 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2078 0, -1ULL, 0, &vd->vdev_dtl_lock));
2079 ASSERT(vd->vdev_dtl_sm != NULL);
2082 bzero(&rtlock, sizeof(rtlock));
2083 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2085 rtsync = range_tree_create(NULL, NULL, &rtlock);
2087 mutex_enter(&rtlock);
2089 mutex_enter(&vd->vdev_dtl_lock);
2090 range_tree_walk(rt, range_tree_add, rtsync);
2091 mutex_exit(&vd->vdev_dtl_lock);
2093 space_map_truncate(vd->vdev_dtl_sm, tx);
2094 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2095 range_tree_vacate(rtsync, NULL, NULL);
2097 range_tree_destroy(rtsync);
2099 mutex_exit(&rtlock);
2100 mutex_destroy(&rtlock);
2103 * If the object for the space map has changed then dirty
2104 * the top level so that we update the config.
2106 if (object != space_map_object(vd->vdev_dtl_sm)) {
2107 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2108 "new object %llu", txg, spa_name(spa), object,
2109 space_map_object(vd->vdev_dtl_sm));
2110 vdev_config_dirty(vd->vdev_top);
2115 mutex_enter(&vd->vdev_dtl_lock);
2116 space_map_update(vd->vdev_dtl_sm);
2117 mutex_exit(&vd->vdev_dtl_lock);
2121 * Determine whether the specified vdev can be offlined/detached/removed
2122 * without losing data.
2125 vdev_dtl_required(vdev_t *vd)
2127 spa_t *spa = vd->vdev_spa;
2128 vdev_t *tvd = vd->vdev_top;
2129 uint8_t cant_read = vd->vdev_cant_read;
2132 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2134 if (vd == spa->spa_root_vdev || vd == tvd)
2138 * Temporarily mark the device as unreadable, and then determine
2139 * whether this results in any DTL outages in the top-level vdev.
2140 * If not, we can safely offline/detach/remove the device.
2142 vd->vdev_cant_read = B_TRUE;
2143 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2144 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2145 vd->vdev_cant_read = cant_read;
2146 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2148 if (!required && zio_injection_enabled)
2149 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2155 * Determine if resilver is needed, and if so the txg range.
2158 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2160 boolean_t needed = B_FALSE;
2161 uint64_t thismin = UINT64_MAX;
2162 uint64_t thismax = 0;
2164 if (vd->vdev_children == 0) {
2165 mutex_enter(&vd->vdev_dtl_lock);
2166 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2167 vdev_writeable(vd)) {
2169 thismin = vdev_dtl_min(vd);
2170 thismax = vdev_dtl_max(vd);
2173 mutex_exit(&vd->vdev_dtl_lock);
2175 for (int c = 0; c < vd->vdev_children; c++) {
2176 vdev_t *cvd = vd->vdev_child[c];
2177 uint64_t cmin, cmax;
2179 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2180 thismin = MIN(thismin, cmin);
2181 thismax = MAX(thismax, cmax);
2187 if (needed && minp) {
2195 vdev_load(vdev_t *vd)
2198 * Recursively load all children.
2200 for (int c = 0; c < vd->vdev_children; c++)
2201 vdev_load(vd->vdev_child[c]);
2204 * If this is a top-level vdev, initialize its metaslabs.
2206 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2207 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2208 vdev_metaslab_init(vd, 0) != 0))
2209 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2210 VDEV_AUX_CORRUPT_DATA);
2213 * If this is a leaf vdev, load its DTL.
2215 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2216 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2217 VDEV_AUX_CORRUPT_DATA);
2221 * The special vdev case is used for hot spares and l2cache devices. Its
2222 * sole purpose it to set the vdev state for the associated vdev. To do this,
2223 * we make sure that we can open the underlying device, then try to read the
2224 * label, and make sure that the label is sane and that it hasn't been
2225 * repurposed to another pool.
2228 vdev_validate_aux(vdev_t *vd)
2231 uint64_t guid, version;
2234 if (!vdev_readable(vd))
2237 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2238 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2239 VDEV_AUX_CORRUPT_DATA);
2243 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2244 !SPA_VERSION_IS_SUPPORTED(version) ||
2245 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2246 guid != vd->vdev_guid ||
2247 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2248 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2249 VDEV_AUX_CORRUPT_DATA);
2255 * We don't actually check the pool state here. If it's in fact in
2256 * use by another pool, we update this fact on the fly when requested.
2263 vdev_remove(vdev_t *vd, uint64_t txg)
2265 spa_t *spa = vd->vdev_spa;
2266 objset_t *mos = spa->spa_meta_objset;
2269 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2271 if (vd->vdev_ms != NULL) {
2272 metaslab_group_t *mg = vd->vdev_mg;
2274 metaslab_group_histogram_verify(mg);
2275 metaslab_class_histogram_verify(mg->mg_class);
2277 for (int m = 0; m < vd->vdev_ms_count; m++) {
2278 metaslab_t *msp = vd->vdev_ms[m];
2280 if (msp == NULL || msp->ms_sm == NULL)
2283 mutex_enter(&msp->ms_lock);
2285 * If the metaslab was not loaded when the vdev
2286 * was removed then the histogram accounting may
2287 * not be accurate. Update the histogram information
2288 * here so that we ensure that the metaslab group
2289 * and metaslab class are up-to-date.
2291 metaslab_group_histogram_remove(mg, msp);
2293 VERIFY0(space_map_allocated(msp->ms_sm));
2294 space_map_free(msp->ms_sm, tx);
2295 space_map_close(msp->ms_sm);
2297 mutex_exit(&msp->ms_lock);
2300 metaslab_group_histogram_verify(mg);
2301 metaslab_class_histogram_verify(mg->mg_class);
2302 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2303 ASSERT0(mg->mg_histogram[i]);
2307 if (vd->vdev_ms_array) {
2308 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2309 vd->vdev_ms_array = 0;
2315 vdev_sync_done(vdev_t *vd, uint64_t txg)
2318 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2320 ASSERT(!vd->vdev_ishole);
2322 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2323 metaslab_sync_done(msp, txg);
2326 metaslab_sync_reassess(vd->vdev_mg);
2330 vdev_sync(vdev_t *vd, uint64_t txg)
2332 spa_t *spa = vd->vdev_spa;
2337 ASSERT(!vd->vdev_ishole);
2339 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2340 ASSERT(vd == vd->vdev_top);
2341 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2342 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2343 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2344 ASSERT(vd->vdev_ms_array != 0);
2345 vdev_config_dirty(vd);
2350 * Remove the metadata associated with this vdev once it's empty.
2352 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2353 vdev_remove(vd, txg);
2355 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2356 metaslab_sync(msp, txg);
2357 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2360 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2361 vdev_dtl_sync(lvd, txg);
2363 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2367 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2369 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2373 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2374 * not be opened, and no I/O is attempted.
2377 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2381 spa_vdev_state_enter(spa, SCL_NONE);
2383 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2384 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2386 if (!vd->vdev_ops->vdev_op_leaf)
2387 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2392 * We don't directly use the aux state here, but if we do a
2393 * vdev_reopen(), we need this value to be present to remember why we
2396 vd->vdev_label_aux = aux;
2399 * Faulted state takes precedence over degraded.
2401 vd->vdev_delayed_close = B_FALSE;
2402 vd->vdev_faulted = 1ULL;
2403 vd->vdev_degraded = 0ULL;
2404 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2407 * If this device has the only valid copy of the data, then
2408 * back off and simply mark the vdev as degraded instead.
2410 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2411 vd->vdev_degraded = 1ULL;
2412 vd->vdev_faulted = 0ULL;
2415 * If we reopen the device and it's not dead, only then do we
2420 if (vdev_readable(vd))
2421 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2424 return (spa_vdev_state_exit(spa, vd, 0));
2428 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2429 * user that something is wrong. The vdev continues to operate as normal as far
2430 * as I/O is concerned.
2433 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2437 spa_vdev_state_enter(spa, SCL_NONE);
2439 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2440 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2442 if (!vd->vdev_ops->vdev_op_leaf)
2443 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2446 * If the vdev is already faulted, then don't do anything.
2448 if (vd->vdev_faulted || vd->vdev_degraded)
2449 return (spa_vdev_state_exit(spa, NULL, 0));
2451 vd->vdev_degraded = 1ULL;
2452 if (!vdev_is_dead(vd))
2453 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2456 return (spa_vdev_state_exit(spa, vd, 0));
2460 * Online the given vdev.
2462 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2463 * spare device should be detached when the device finishes resilvering.
2464 * Second, the online should be treated like a 'test' online case, so no FMA
2465 * events are generated if the device fails to open.
2468 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2470 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2472 spa_vdev_state_enter(spa, SCL_NONE);
2474 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2475 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2477 if (!vd->vdev_ops->vdev_op_leaf)
2478 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2481 vd->vdev_offline = B_FALSE;
2482 vd->vdev_tmpoffline = B_FALSE;
2483 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2484 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2486 /* XXX - L2ARC 1.0 does not support expansion */
2487 if (!vd->vdev_aux) {
2488 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2489 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2493 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2495 if (!vd->vdev_aux) {
2496 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2497 pvd->vdev_expanding = B_FALSE;
2501 *newstate = vd->vdev_state;
2502 if ((flags & ZFS_ONLINE_UNSPARE) &&
2503 !vdev_is_dead(vd) && vd->vdev_parent &&
2504 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2505 vd->vdev_parent->vdev_child[0] == vd)
2506 vd->vdev_unspare = B_TRUE;
2508 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2510 /* XXX - L2ARC 1.0 does not support expansion */
2512 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2513 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2515 return (spa_vdev_state_exit(spa, vd, 0));
2519 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2523 uint64_t generation;
2524 metaslab_group_t *mg;
2527 spa_vdev_state_enter(spa, SCL_ALLOC);
2529 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2530 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2532 if (!vd->vdev_ops->vdev_op_leaf)
2533 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2537 generation = spa->spa_config_generation + 1;
2540 * If the device isn't already offline, try to offline it.
2542 if (!vd->vdev_offline) {
2544 * If this device has the only valid copy of some data,
2545 * don't allow it to be offlined. Log devices are always
2548 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2549 vdev_dtl_required(vd))
2550 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2553 * If the top-level is a slog and it has had allocations
2554 * then proceed. We check that the vdev's metaslab group
2555 * is not NULL since it's possible that we may have just
2556 * added this vdev but not yet initialized its metaslabs.
2558 if (tvd->vdev_islog && mg != NULL) {
2560 * Prevent any future allocations.
2562 metaslab_group_passivate(mg);
2563 (void) spa_vdev_state_exit(spa, vd, 0);
2565 error = spa_offline_log(spa);
2567 spa_vdev_state_enter(spa, SCL_ALLOC);
2570 * Check to see if the config has changed.
2572 if (error || generation != spa->spa_config_generation) {
2573 metaslab_group_activate(mg);
2575 return (spa_vdev_state_exit(spa,
2577 (void) spa_vdev_state_exit(spa, vd, 0);
2580 ASSERT0(tvd->vdev_stat.vs_alloc);
2584 * Offline this device and reopen its top-level vdev.
2585 * If the top-level vdev is a log device then just offline
2586 * it. Otherwise, if this action results in the top-level
2587 * vdev becoming unusable, undo it and fail the request.
2589 vd->vdev_offline = B_TRUE;
2592 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2593 vdev_is_dead(tvd)) {
2594 vd->vdev_offline = B_FALSE;
2596 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2600 * Add the device back into the metaslab rotor so that
2601 * once we online the device it's open for business.
2603 if (tvd->vdev_islog && mg != NULL)
2604 metaslab_group_activate(mg);
2607 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2609 return (spa_vdev_state_exit(spa, vd, 0));
2613 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2617 mutex_enter(&spa->spa_vdev_top_lock);
2618 error = vdev_offline_locked(spa, guid, flags);
2619 mutex_exit(&spa->spa_vdev_top_lock);
2625 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2626 * vdev_offline(), we assume the spa config is locked. We also clear all
2627 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2630 vdev_clear(spa_t *spa, vdev_t *vd)
2632 vdev_t *rvd = spa->spa_root_vdev;
2634 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2639 vd->vdev_stat.vs_read_errors = 0;
2640 vd->vdev_stat.vs_write_errors = 0;
2641 vd->vdev_stat.vs_checksum_errors = 0;
2643 for (int c = 0; c < vd->vdev_children; c++)
2644 vdev_clear(spa, vd->vdev_child[c]);
2647 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2648 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2650 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2651 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2655 * If we're in the FAULTED state or have experienced failed I/O, then
2656 * clear the persistent state and attempt to reopen the device. We
2657 * also mark the vdev config dirty, so that the new faulted state is
2658 * written out to disk.
2660 if (vd->vdev_faulted || vd->vdev_degraded ||
2661 !vdev_readable(vd) || !vdev_writeable(vd)) {
2664 * When reopening in reponse to a clear event, it may be due to
2665 * a fmadm repair request. In this case, if the device is
2666 * still broken, we want to still post the ereport again.
2668 vd->vdev_forcefault = B_TRUE;
2670 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2671 vd->vdev_cant_read = B_FALSE;
2672 vd->vdev_cant_write = B_FALSE;
2674 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2676 vd->vdev_forcefault = B_FALSE;
2678 if (vd != rvd && vdev_writeable(vd->vdev_top))
2679 vdev_state_dirty(vd->vdev_top);
2681 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2682 spa_async_request(spa, SPA_ASYNC_RESILVER);
2684 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2688 * When clearing a FMA-diagnosed fault, we always want to
2689 * unspare the device, as we assume that the original spare was
2690 * done in response to the FMA fault.
2692 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2693 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2694 vd->vdev_parent->vdev_child[0] == vd)
2695 vd->vdev_unspare = B_TRUE;
2699 vdev_is_dead(vdev_t *vd)
2702 * Holes and missing devices are always considered "dead".
2703 * This simplifies the code since we don't have to check for
2704 * these types of devices in the various code paths.
2705 * Instead we rely on the fact that we skip over dead devices
2706 * before issuing I/O to them.
2708 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2709 vd->vdev_ops == &vdev_missing_ops);
2713 vdev_readable(vdev_t *vd)
2715 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2719 vdev_writeable(vdev_t *vd)
2721 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2725 vdev_allocatable(vdev_t *vd)
2727 uint64_t state = vd->vdev_state;
2730 * We currently allow allocations from vdevs which may be in the
2731 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2732 * fails to reopen then we'll catch it later when we're holding
2733 * the proper locks. Note that we have to get the vdev state
2734 * in a local variable because although it changes atomically,
2735 * we're asking two separate questions about it.
2737 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2738 !vd->vdev_cant_write && !vd->vdev_ishole);
2742 vdev_accessible(vdev_t *vd, zio_t *zio)
2744 ASSERT(zio->io_vd == vd);
2746 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2749 if (zio->io_type == ZIO_TYPE_READ)
2750 return (!vd->vdev_cant_read);
2752 if (zio->io_type == ZIO_TYPE_WRITE)
2753 return (!vd->vdev_cant_write);
2759 * Get statistics for the given vdev.
2762 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2764 spa_t *spa = vd->vdev_spa;
2765 vdev_t *rvd = spa->spa_root_vdev;
2767 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2769 mutex_enter(&vd->vdev_stat_lock);
2770 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2771 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2772 vs->vs_state = vd->vdev_state;
2773 vs->vs_rsize = vdev_get_min_asize(vd);
2774 if (vd->vdev_ops->vdev_op_leaf)
2775 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2776 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2777 vs->vs_configured_ashift = vd->vdev_top != NULL
2778 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2779 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2780 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2781 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2782 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2786 * If we're getting stats on the root vdev, aggregate the I/O counts
2787 * over all top-level vdevs (i.e. the direct children of the root).
2790 for (int c = 0; c < rvd->vdev_children; c++) {
2791 vdev_t *cvd = rvd->vdev_child[c];
2792 vdev_stat_t *cvs = &cvd->vdev_stat;
2794 for (int t = 0; t < ZIO_TYPES; t++) {
2795 vs->vs_ops[t] += cvs->vs_ops[t];
2796 vs->vs_bytes[t] += cvs->vs_bytes[t];
2798 cvs->vs_scan_removing = cvd->vdev_removing;
2801 mutex_exit(&vd->vdev_stat_lock);
2805 vdev_clear_stats(vdev_t *vd)
2807 mutex_enter(&vd->vdev_stat_lock);
2808 vd->vdev_stat.vs_space = 0;
2809 vd->vdev_stat.vs_dspace = 0;
2810 vd->vdev_stat.vs_alloc = 0;
2811 mutex_exit(&vd->vdev_stat_lock);
2815 vdev_scan_stat_init(vdev_t *vd)
2817 vdev_stat_t *vs = &vd->vdev_stat;
2819 for (int c = 0; c < vd->vdev_children; c++)
2820 vdev_scan_stat_init(vd->vdev_child[c]);
2822 mutex_enter(&vd->vdev_stat_lock);
2823 vs->vs_scan_processed = 0;
2824 mutex_exit(&vd->vdev_stat_lock);
2828 vdev_stat_update(zio_t *zio, uint64_t psize)
2830 spa_t *spa = zio->io_spa;
2831 vdev_t *rvd = spa->spa_root_vdev;
2832 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2834 uint64_t txg = zio->io_txg;
2835 vdev_stat_t *vs = &vd->vdev_stat;
2836 zio_type_t type = zio->io_type;
2837 int flags = zio->io_flags;
2840 * If this i/o is a gang leader, it didn't do any actual work.
2842 if (zio->io_gang_tree)
2845 if (zio->io_error == 0) {
2847 * If this is a root i/o, don't count it -- we've already
2848 * counted the top-level vdevs, and vdev_get_stats() will
2849 * aggregate them when asked. This reduces contention on
2850 * the root vdev_stat_lock and implicitly handles blocks
2851 * that compress away to holes, for which there is no i/o.
2852 * (Holes never create vdev children, so all the counters
2853 * remain zero, which is what we want.)
2855 * Note: this only applies to successful i/o (io_error == 0)
2856 * because unlike i/o counts, errors are not additive.
2857 * When reading a ditto block, for example, failure of
2858 * one top-level vdev does not imply a root-level error.
2863 ASSERT(vd == zio->io_vd);
2865 if (flags & ZIO_FLAG_IO_BYPASS)
2868 mutex_enter(&vd->vdev_stat_lock);
2870 if (flags & ZIO_FLAG_IO_REPAIR) {
2871 if (flags & ZIO_FLAG_SCAN_THREAD) {
2872 dsl_scan_phys_t *scn_phys =
2873 &spa->spa_dsl_pool->dp_scan->scn_phys;
2874 uint64_t *processed = &scn_phys->scn_processed;
2877 if (vd->vdev_ops->vdev_op_leaf)
2878 atomic_add_64(processed, psize);
2879 vs->vs_scan_processed += psize;
2882 if (flags & ZIO_FLAG_SELF_HEAL)
2883 vs->vs_self_healed += psize;
2887 vs->vs_bytes[type] += psize;
2889 mutex_exit(&vd->vdev_stat_lock);
2893 if (flags & ZIO_FLAG_SPECULATIVE)
2897 * If this is an I/O error that is going to be retried, then ignore the
2898 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2899 * hard errors, when in reality they can happen for any number of
2900 * innocuous reasons (bus resets, MPxIO link failure, etc).
2902 if (zio->io_error == EIO &&
2903 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2907 * Intent logs writes won't propagate their error to the root
2908 * I/O so don't mark these types of failures as pool-level
2911 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2914 mutex_enter(&vd->vdev_stat_lock);
2915 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2916 if (zio->io_error == ECKSUM)
2917 vs->vs_checksum_errors++;
2919 vs->vs_read_errors++;
2921 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2922 vs->vs_write_errors++;
2923 mutex_exit(&vd->vdev_stat_lock);
2925 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2926 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2927 (flags & ZIO_FLAG_SCAN_THREAD) ||
2928 spa->spa_claiming)) {
2930 * This is either a normal write (not a repair), or it's
2931 * a repair induced by the scrub thread, or it's a repair
2932 * made by zil_claim() during spa_load() in the first txg.
2933 * In the normal case, we commit the DTL change in the same
2934 * txg as the block was born. In the scrub-induced repair
2935 * case, we know that scrubs run in first-pass syncing context,
2936 * so we commit the DTL change in spa_syncing_txg(spa).
2937 * In the zil_claim() case, we commit in spa_first_txg(spa).
2939 * We currently do not make DTL entries for failed spontaneous
2940 * self-healing writes triggered by normal (non-scrubbing)
2941 * reads, because we have no transactional context in which to
2942 * do so -- and it's not clear that it'd be desirable anyway.
2944 if (vd->vdev_ops->vdev_op_leaf) {
2945 uint64_t commit_txg = txg;
2946 if (flags & ZIO_FLAG_SCAN_THREAD) {
2947 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2948 ASSERT(spa_sync_pass(spa) == 1);
2949 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2950 commit_txg = spa_syncing_txg(spa);
2951 } else if (spa->spa_claiming) {
2952 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2953 commit_txg = spa_first_txg(spa);
2955 ASSERT(commit_txg >= spa_syncing_txg(spa));
2956 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2958 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2959 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2960 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2963 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2968 * Update the in-core space usage stats for this vdev, its metaslab class,
2969 * and the root vdev.
2972 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2973 int64_t space_delta)
2975 int64_t dspace_delta = space_delta;
2976 spa_t *spa = vd->vdev_spa;
2977 vdev_t *rvd = spa->spa_root_vdev;
2978 metaslab_group_t *mg = vd->vdev_mg;
2979 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2981 ASSERT(vd == vd->vdev_top);
2984 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2985 * factor. We must calculate this here and not at the root vdev
2986 * because the root vdev's psize-to-asize is simply the max of its
2987 * childrens', thus not accurate enough for us.
2989 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2990 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2991 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2992 vd->vdev_deflate_ratio;
2994 mutex_enter(&vd->vdev_stat_lock);
2995 vd->vdev_stat.vs_alloc += alloc_delta;
2996 vd->vdev_stat.vs_space += space_delta;
2997 vd->vdev_stat.vs_dspace += dspace_delta;
2998 mutex_exit(&vd->vdev_stat_lock);
3000 if (mc == spa_normal_class(spa)) {
3001 mutex_enter(&rvd->vdev_stat_lock);
3002 rvd->vdev_stat.vs_alloc += alloc_delta;
3003 rvd->vdev_stat.vs_space += space_delta;
3004 rvd->vdev_stat.vs_dspace += dspace_delta;
3005 mutex_exit(&rvd->vdev_stat_lock);
3009 ASSERT(rvd == vd->vdev_parent);
3010 ASSERT(vd->vdev_ms_count != 0);
3012 metaslab_class_space_update(mc,
3013 alloc_delta, defer_delta, space_delta, dspace_delta);
3018 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3019 * so that it will be written out next time the vdev configuration is synced.
3020 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3023 vdev_config_dirty(vdev_t *vd)
3025 spa_t *spa = vd->vdev_spa;
3026 vdev_t *rvd = spa->spa_root_vdev;
3029 ASSERT(spa_writeable(spa));
3032 * If this is an aux vdev (as with l2cache and spare devices), then we
3033 * update the vdev config manually and set the sync flag.
3035 if (vd->vdev_aux != NULL) {
3036 spa_aux_vdev_t *sav = vd->vdev_aux;
3040 for (c = 0; c < sav->sav_count; c++) {
3041 if (sav->sav_vdevs[c] == vd)
3045 if (c == sav->sav_count) {
3047 * We're being removed. There's nothing more to do.
3049 ASSERT(sav->sav_sync == B_TRUE);
3053 sav->sav_sync = B_TRUE;
3055 if (nvlist_lookup_nvlist_array(sav->sav_config,
3056 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3057 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3058 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3064 * Setting the nvlist in the middle if the array is a little
3065 * sketchy, but it will work.
3067 nvlist_free(aux[c]);
3068 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3074 * The dirty list is protected by the SCL_CONFIG lock. The caller
3075 * must either hold SCL_CONFIG as writer, or must be the sync thread
3076 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3077 * so this is sufficient to ensure mutual exclusion.
3079 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3080 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3081 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3084 for (c = 0; c < rvd->vdev_children; c++)
3085 vdev_config_dirty(rvd->vdev_child[c]);
3087 ASSERT(vd == vd->vdev_top);
3089 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3091 list_insert_head(&spa->spa_config_dirty_list, vd);
3096 vdev_config_clean(vdev_t *vd)
3098 spa_t *spa = vd->vdev_spa;
3100 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3101 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3102 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3104 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3105 list_remove(&spa->spa_config_dirty_list, vd);
3109 * Mark a top-level vdev's state as dirty, so that the next pass of
3110 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3111 * the state changes from larger config changes because they require
3112 * much less locking, and are often needed for administrative actions.
3115 vdev_state_dirty(vdev_t *vd)
3117 spa_t *spa = vd->vdev_spa;
3119 ASSERT(spa_writeable(spa));
3120 ASSERT(vd == vd->vdev_top);
3123 * The state list is protected by the SCL_STATE lock. The caller
3124 * must either hold SCL_STATE as writer, or must be the sync thread
3125 * (which holds SCL_STATE as reader). There's only one sync thread,
3126 * so this is sufficient to ensure mutual exclusion.
3128 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3129 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3130 spa_config_held(spa, SCL_STATE, RW_READER)));
3132 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3133 list_insert_head(&spa->spa_state_dirty_list, vd);
3137 vdev_state_clean(vdev_t *vd)
3139 spa_t *spa = vd->vdev_spa;
3141 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3142 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3143 spa_config_held(spa, SCL_STATE, RW_READER)));
3145 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3146 list_remove(&spa->spa_state_dirty_list, vd);
3150 * Propagate vdev state up from children to parent.
3153 vdev_propagate_state(vdev_t *vd)
3155 spa_t *spa = vd->vdev_spa;
3156 vdev_t *rvd = spa->spa_root_vdev;
3157 int degraded = 0, faulted = 0;
3161 if (vd->vdev_children > 0) {
3162 for (int c = 0; c < vd->vdev_children; c++) {
3163 child = vd->vdev_child[c];
3166 * Don't factor holes into the decision.
3168 if (child->vdev_ishole)
3171 if (!vdev_readable(child) ||
3172 (!vdev_writeable(child) && spa_writeable(spa))) {
3174 * Root special: if there is a top-level log
3175 * device, treat the root vdev as if it were
3178 if (child->vdev_islog && vd == rvd)
3182 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3186 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3190 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3193 * Root special: if there is a top-level vdev that cannot be
3194 * opened due to corrupted metadata, then propagate the root
3195 * vdev's aux state as 'corrupt' rather than 'insufficient
3198 if (corrupted && vd == rvd &&
3199 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3200 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3201 VDEV_AUX_CORRUPT_DATA);
3204 if (vd->vdev_parent)
3205 vdev_propagate_state(vd->vdev_parent);
3209 * Set a vdev's state. If this is during an open, we don't update the parent
3210 * state, because we're in the process of opening children depth-first.
3211 * Otherwise, we propagate the change to the parent.
3213 * If this routine places a device in a faulted state, an appropriate ereport is
3217 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3219 uint64_t save_state;
3220 spa_t *spa = vd->vdev_spa;
3222 if (state == vd->vdev_state) {
3223 vd->vdev_stat.vs_aux = aux;
3227 save_state = vd->vdev_state;
3229 vd->vdev_state = state;
3230 vd->vdev_stat.vs_aux = aux;
3233 * If we are setting the vdev state to anything but an open state, then
3234 * always close the underlying device unless the device has requested
3235 * a delayed close (i.e. we're about to remove or fault the device).
3236 * Otherwise, we keep accessible but invalid devices open forever.
3237 * We don't call vdev_close() itself, because that implies some extra
3238 * checks (offline, etc) that we don't want here. This is limited to
3239 * leaf devices, because otherwise closing the device will affect other
3242 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3243 vd->vdev_ops->vdev_op_leaf)
3244 vd->vdev_ops->vdev_op_close(vd);
3247 * If we have brought this vdev back into service, we need
3248 * to notify fmd so that it can gracefully repair any outstanding
3249 * cases due to a missing device. We do this in all cases, even those
3250 * that probably don't correlate to a repaired fault. This is sure to
3251 * catch all cases, and we let the zfs-retire agent sort it out. If
3252 * this is a transient state it's OK, as the retire agent will
3253 * double-check the state of the vdev before repairing it.
3255 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3256 vd->vdev_prevstate != state)
3257 zfs_post_state_change(spa, vd);
3259 if (vd->vdev_removed &&
3260 state == VDEV_STATE_CANT_OPEN &&
3261 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3263 * If the previous state is set to VDEV_STATE_REMOVED, then this
3264 * device was previously marked removed and someone attempted to
3265 * reopen it. If this failed due to a nonexistent device, then
3266 * keep the device in the REMOVED state. We also let this be if
3267 * it is one of our special test online cases, which is only
3268 * attempting to online the device and shouldn't generate an FMA
3271 vd->vdev_state = VDEV_STATE_REMOVED;
3272 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3273 } else if (state == VDEV_STATE_REMOVED) {
3274 vd->vdev_removed = B_TRUE;
3275 } else if (state == VDEV_STATE_CANT_OPEN) {
3277 * If we fail to open a vdev during an import or recovery, we
3278 * mark it as "not available", which signifies that it was
3279 * never there to begin with. Failure to open such a device
3280 * is not considered an error.
3282 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3283 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3284 vd->vdev_ops->vdev_op_leaf)
3285 vd->vdev_not_present = 1;
3288 * Post the appropriate ereport. If the 'prevstate' field is
3289 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3290 * that this is part of a vdev_reopen(). In this case, we don't
3291 * want to post the ereport if the device was already in the
3292 * CANT_OPEN state beforehand.
3294 * If the 'checkremove' flag is set, then this is an attempt to
3295 * online the device in response to an insertion event. If we
3296 * hit this case, then we have detected an insertion event for a
3297 * faulted or offline device that wasn't in the removed state.
3298 * In this scenario, we don't post an ereport because we are
3299 * about to replace the device, or attempt an online with
3300 * vdev_forcefault, which will generate the fault for us.
3302 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3303 !vd->vdev_not_present && !vd->vdev_checkremove &&
3304 vd != spa->spa_root_vdev) {
3308 case VDEV_AUX_OPEN_FAILED:
3309 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3311 case VDEV_AUX_CORRUPT_DATA:
3312 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3314 case VDEV_AUX_NO_REPLICAS:
3315 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3317 case VDEV_AUX_BAD_GUID_SUM:
3318 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3320 case VDEV_AUX_TOO_SMALL:
3321 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3323 case VDEV_AUX_BAD_LABEL:
3324 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3327 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3330 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3333 /* Erase any notion of persistent removed state */
3334 vd->vdev_removed = B_FALSE;
3336 vd->vdev_removed = B_FALSE;
3339 if (!isopen && vd->vdev_parent)
3340 vdev_propagate_state(vd->vdev_parent);
3344 * Check the vdev configuration to ensure that it's capable of supporting
3347 * On Solaris, we do not support RAID-Z or partial configuration. In
3348 * addition, only a single top-level vdev is allowed and none of the
3349 * leaves can be wholedisks.
3351 * For FreeBSD, we can boot from any configuration. There is a
3352 * limitation that the boot filesystem must be either uncompressed or
3353 * compresses with lzjb compression but I'm not sure how to enforce
3357 vdev_is_bootable(vdev_t *vd)
3360 if (!vd->vdev_ops->vdev_op_leaf) {
3361 char *vdev_type = vd->vdev_ops->vdev_op_type;
3363 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3364 vd->vdev_children > 1) {
3366 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3367 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3370 } else if (vd->vdev_wholedisk == 1) {
3374 for (int c = 0; c < vd->vdev_children; c++) {
3375 if (!vdev_is_bootable(vd->vdev_child[c]))
3383 * Load the state from the original vdev tree (ovd) which
3384 * we've retrieved from the MOS config object. If the original
3385 * vdev was offline or faulted then we transfer that state to the
3386 * device in the current vdev tree (nvd).
3389 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3391 spa_t *spa = nvd->vdev_spa;
3393 ASSERT(nvd->vdev_top->vdev_islog);
3394 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3395 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3397 for (int c = 0; c < nvd->vdev_children; c++)
3398 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3400 if (nvd->vdev_ops->vdev_op_leaf) {
3402 * Restore the persistent vdev state
3404 nvd->vdev_offline = ovd->vdev_offline;
3405 nvd->vdev_faulted = ovd->vdev_faulted;
3406 nvd->vdev_degraded = ovd->vdev_degraded;
3407 nvd->vdev_removed = ovd->vdev_removed;
3412 * Determine if a log device has valid content. If the vdev was
3413 * removed or faulted in the MOS config then we know that
3414 * the content on the log device has already been written to the pool.
3417 vdev_log_state_valid(vdev_t *vd)
3419 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3423 for (int c = 0; c < vd->vdev_children; c++)
3424 if (vdev_log_state_valid(vd->vdev_child[c]))
3431 * Expand a vdev if possible.
3434 vdev_expand(vdev_t *vd, uint64_t txg)
3436 ASSERT(vd->vdev_top == vd);
3437 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3439 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3440 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3441 vdev_config_dirty(vd);
3449 vdev_split(vdev_t *vd)
3451 vdev_t *cvd, *pvd = vd->vdev_parent;
3453 vdev_remove_child(pvd, vd);
3454 vdev_compact_children(pvd);
3456 cvd = pvd->vdev_child[0];
3457 if (pvd->vdev_children == 1) {
3458 vdev_remove_parent(cvd);
3459 cvd->vdev_splitting = B_TRUE;
3461 vdev_propagate_state(cvd);
3465 vdev_deadman(vdev_t *vd)
3467 for (int c = 0; c < vd->vdev_children; c++) {
3468 vdev_t *cvd = vd->vdev_child[c];
3473 if (vd->vdev_ops->vdev_op_leaf) {
3474 vdev_queue_t *vq = &vd->vdev_queue;
3476 mutex_enter(&vq->vq_lock);
3477 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3478 spa_t *spa = vd->vdev_spa;
3483 * Look at the head of all the pending queues,
3484 * if any I/O has been outstanding for longer than
3485 * the spa_deadman_synctime we panic the system.
3487 fio = avl_first(&vq->vq_active_tree);
3488 delta = gethrtime() - fio->io_timestamp;
3489 if (delta > spa_deadman_synctime(spa)) {
3490 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3491 "delta %lluns, last io %lluns",
3492 fio->io_timestamp, delta,
3493 vq->vq_io_complete_ts);
3494 fm_panic("I/O to pool '%s' appears to be "
3495 "hung on vdev guid %llu at '%s'.",
3497 (long long unsigned int) vd->vdev_guid,
3501 mutex_exit(&vq->vq_lock);