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 "typical" blocksize.
930 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
931 * otherwise it would 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,
959 error = metaslab_init(vd->vdev_mg, m, object, txg,
966 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
969 * If the vdev is being removed we don't activate
970 * the metaslabs since we want to ensure that no new
971 * allocations are performed on this device.
973 if (oldc == 0 && !vd->vdev_removing)
974 metaslab_group_activate(vd->vdev_mg);
977 spa_config_exit(spa, SCL_ALLOC, FTAG);
983 vdev_metaslab_fini(vdev_t *vd)
986 uint64_t count = vd->vdev_ms_count;
988 if (vd->vdev_ms != NULL) {
989 metaslab_group_passivate(vd->vdev_mg);
990 for (m = 0; m < count; m++) {
991 metaslab_t *msp = vd->vdev_ms[m];
996 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1001 typedef struct vdev_probe_stats {
1002 boolean_t vps_readable;
1003 boolean_t vps_writeable;
1005 } vdev_probe_stats_t;
1008 vdev_probe_done(zio_t *zio)
1010 spa_t *spa = zio->io_spa;
1011 vdev_t *vd = zio->io_vd;
1012 vdev_probe_stats_t *vps = zio->io_private;
1014 ASSERT(vd->vdev_probe_zio != NULL);
1016 if (zio->io_type == ZIO_TYPE_READ) {
1017 if (zio->io_error == 0)
1018 vps->vps_readable = 1;
1019 if (zio->io_error == 0 && spa_writeable(spa)) {
1020 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1021 zio->io_offset, zio->io_size, zio->io_data,
1022 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1023 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1025 zio_buf_free(zio->io_data, zio->io_size);
1027 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1028 if (zio->io_error == 0)
1029 vps->vps_writeable = 1;
1030 zio_buf_free(zio->io_data, zio->io_size);
1031 } else if (zio->io_type == ZIO_TYPE_NULL) {
1034 vd->vdev_cant_read |= !vps->vps_readable;
1035 vd->vdev_cant_write |= !vps->vps_writeable;
1037 if (vdev_readable(vd) &&
1038 (vdev_writeable(vd) || !spa_writeable(spa))) {
1041 ASSERT(zio->io_error != 0);
1042 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1043 spa, vd, NULL, 0, 0);
1044 zio->io_error = SET_ERROR(ENXIO);
1047 mutex_enter(&vd->vdev_probe_lock);
1048 ASSERT(vd->vdev_probe_zio == zio);
1049 vd->vdev_probe_zio = NULL;
1050 mutex_exit(&vd->vdev_probe_lock);
1052 while ((pio = zio_walk_parents(zio)) != NULL)
1053 if (!vdev_accessible(vd, pio))
1054 pio->io_error = SET_ERROR(ENXIO);
1056 kmem_free(vps, sizeof (*vps));
1061 * Determine whether this device is accessible.
1063 * Read and write to several known locations: the pad regions of each
1064 * vdev label but the first, which we leave alone in case it contains
1068 vdev_probe(vdev_t *vd, zio_t *zio)
1070 spa_t *spa = vd->vdev_spa;
1071 vdev_probe_stats_t *vps = NULL;
1074 ASSERT(vd->vdev_ops->vdev_op_leaf);
1077 * Don't probe the probe.
1079 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1083 * To prevent 'probe storms' when a device fails, we create
1084 * just one probe i/o at a time. All zios that want to probe
1085 * this vdev will become parents of the probe io.
1087 mutex_enter(&vd->vdev_probe_lock);
1089 if ((pio = vd->vdev_probe_zio) == NULL) {
1090 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1092 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1093 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1096 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1098 * vdev_cant_read and vdev_cant_write can only
1099 * transition from TRUE to FALSE when we have the
1100 * SCL_ZIO lock as writer; otherwise they can only
1101 * transition from FALSE to TRUE. This ensures that
1102 * any zio looking at these values can assume that
1103 * failures persist for the life of the I/O. That's
1104 * important because when a device has intermittent
1105 * connectivity problems, we want to ensure that
1106 * they're ascribed to the device (ENXIO) and not
1109 * Since we hold SCL_ZIO as writer here, clear both
1110 * values so the probe can reevaluate from first
1113 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1114 vd->vdev_cant_read = B_FALSE;
1115 vd->vdev_cant_write = B_FALSE;
1118 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1119 vdev_probe_done, vps,
1120 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1123 * We can't change the vdev state in this context, so we
1124 * kick off an async task to do it on our behalf.
1127 vd->vdev_probe_wanted = B_TRUE;
1128 spa_async_request(spa, SPA_ASYNC_PROBE);
1133 zio_add_child(zio, pio);
1135 mutex_exit(&vd->vdev_probe_lock);
1138 ASSERT(zio != NULL);
1142 for (int l = 1; l < VDEV_LABELS; l++) {
1143 zio_nowait(zio_read_phys(pio, vd,
1144 vdev_label_offset(vd->vdev_psize, l,
1145 offsetof(vdev_label_t, vl_pad2)),
1146 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1147 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1148 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1159 vdev_open_child(void *arg)
1163 vd->vdev_open_thread = curthread;
1164 vd->vdev_open_error = vdev_open(vd);
1165 vd->vdev_open_thread = NULL;
1169 vdev_uses_zvols(vdev_t *vd)
1171 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1172 strlen(ZVOL_DIR)) == 0)
1174 for (int c = 0; c < vd->vdev_children; c++)
1175 if (vdev_uses_zvols(vd->vdev_child[c]))
1181 vdev_open_children(vdev_t *vd)
1184 int children = vd->vdev_children;
1187 * in order to handle pools on top of zvols, do the opens
1188 * in a single thread so that the same thread holds the
1189 * spa_namespace_lock
1191 if (B_TRUE || vdev_uses_zvols(vd)) {
1192 for (int c = 0; c < children; c++)
1193 vd->vdev_child[c]->vdev_open_error =
1194 vdev_open(vd->vdev_child[c]);
1197 tq = taskq_create("vdev_open", children, minclsyspri,
1198 children, children, TASKQ_PREPOPULATE);
1200 for (int c = 0; c < children; c++)
1201 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1208 * Prepare a virtual device for access.
1211 vdev_open(vdev_t *vd)
1213 spa_t *spa = vd->vdev_spa;
1216 uint64_t max_osize = 0;
1217 uint64_t asize, max_asize, psize;
1218 uint64_t logical_ashift = 0;
1219 uint64_t physical_ashift = 0;
1221 ASSERT(vd->vdev_open_thread == curthread ||
1222 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1223 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1224 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1225 vd->vdev_state == VDEV_STATE_OFFLINE);
1227 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1228 vd->vdev_cant_read = B_FALSE;
1229 vd->vdev_cant_write = B_FALSE;
1230 vd->vdev_notrim = B_FALSE;
1231 vd->vdev_min_asize = vdev_get_min_asize(vd);
1234 * If this vdev is not removed, check its fault status. If it's
1235 * faulted, bail out of the open.
1237 if (!vd->vdev_removed && vd->vdev_faulted) {
1238 ASSERT(vd->vdev_children == 0);
1239 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1240 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1241 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1242 vd->vdev_label_aux);
1243 return (SET_ERROR(ENXIO));
1244 } else if (vd->vdev_offline) {
1245 ASSERT(vd->vdev_children == 0);
1246 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1247 return (SET_ERROR(ENXIO));
1250 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1251 &logical_ashift, &physical_ashift);
1254 * Reset the vdev_reopening flag so that we actually close
1255 * the vdev on error.
1257 vd->vdev_reopening = B_FALSE;
1258 if (zio_injection_enabled && error == 0)
1259 error = zio_handle_device_injection(vd, NULL, ENXIO);
1262 if (vd->vdev_removed &&
1263 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1264 vd->vdev_removed = B_FALSE;
1266 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1267 vd->vdev_stat.vs_aux);
1271 vd->vdev_removed = B_FALSE;
1274 * Recheck the faulted flag now that we have confirmed that
1275 * the vdev is accessible. If we're faulted, bail.
1277 if (vd->vdev_faulted) {
1278 ASSERT(vd->vdev_children == 0);
1279 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1280 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1281 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1282 vd->vdev_label_aux);
1283 return (SET_ERROR(ENXIO));
1286 if (vd->vdev_degraded) {
1287 ASSERT(vd->vdev_children == 0);
1288 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1289 VDEV_AUX_ERR_EXCEEDED);
1291 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1295 * For hole or missing vdevs we just return success.
1297 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1300 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1301 trim_map_create(vd);
1303 for (int 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;
1336 vd->vdev_psize = psize;
1339 * Make sure the allocatable size hasn't shrunk.
1341 if (asize < vd->vdev_min_asize) {
1342 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1343 VDEV_AUX_BAD_LABEL);
1344 return (SET_ERROR(EINVAL));
1347 vd->vdev_physical_ashift =
1348 MAX(physical_ashift, vd->vdev_physical_ashift);
1349 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1350 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1352 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1353 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1354 VDEV_AUX_ASHIFT_TOO_BIG);
1358 if (vd->vdev_asize == 0) {
1360 * This is the first-ever open, so use the computed values.
1361 * For testing purposes, a higher ashift can be requested.
1363 vd->vdev_asize = asize;
1364 vd->vdev_max_asize = max_asize;
1367 * Make sure the alignment requirement hasn't increased.
1369 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1370 vd->vdev_ops->vdev_op_leaf) {
1371 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1372 VDEV_AUX_BAD_LABEL);
1375 vd->vdev_max_asize = max_asize;
1379 * If all children are healthy and the asize has increased,
1380 * then we've experienced dynamic LUN growth. If automatic
1381 * expansion is enabled then use the additional space.
1383 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1384 (vd->vdev_expanding || spa->spa_autoexpand))
1385 vd->vdev_asize = asize;
1387 vdev_set_min_asize(vd);
1390 * Ensure we can issue some IO before declaring the
1391 * vdev open for business.
1393 if (vd->vdev_ops->vdev_op_leaf &&
1394 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1395 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1396 VDEV_AUX_ERR_EXCEEDED);
1401 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1402 * resilver. But don't do this if we are doing a reopen for a scrub,
1403 * since this would just restart the scrub we are already doing.
1405 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1406 vdev_resilver_needed(vd, NULL, NULL))
1407 spa_async_request(spa, SPA_ASYNC_RESILVER);
1413 * Called once the vdevs are all opened, this routine validates the label
1414 * contents. This needs to be done before vdev_load() so that we don't
1415 * inadvertently do repair I/Os to the wrong device.
1417 * If 'strict' is false ignore the spa guid check. This is necessary because
1418 * if the machine crashed during a re-guid the new guid might have been written
1419 * to all of the vdev labels, but not the cached config. The strict check
1420 * will be performed when the pool is opened again using the mos config.
1422 * This function will only return failure if one of the vdevs indicates that it
1423 * has since been destroyed or exported. This is only possible if
1424 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1425 * will be updated but the function will return 0.
1428 vdev_validate(vdev_t *vd, boolean_t strict)
1430 spa_t *spa = vd->vdev_spa;
1432 uint64_t guid = 0, top_guid;
1435 for (int c = 0; c < vd->vdev_children; c++)
1436 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1437 return (SET_ERROR(EBADF));
1440 * If the device has already failed, or was marked offline, don't do
1441 * any further validation. Otherwise, label I/O will fail and we will
1442 * overwrite the previous state.
1444 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1445 uint64_t aux_guid = 0;
1447 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1448 spa_last_synced_txg(spa) : -1ULL;
1450 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1451 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1452 VDEV_AUX_BAD_LABEL);
1457 * Determine if this vdev has been split off into another
1458 * pool. If so, then refuse to open it.
1460 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1461 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1462 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1463 VDEV_AUX_SPLIT_POOL);
1468 if (strict && (nvlist_lookup_uint64(label,
1469 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1470 guid != spa_guid(spa))) {
1471 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1472 VDEV_AUX_CORRUPT_DATA);
1477 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1478 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1483 * If this vdev just became a top-level vdev because its
1484 * sibling was detached, it will have adopted the parent's
1485 * vdev guid -- but the label may or may not be on disk yet.
1486 * Fortunately, either version of the label will have the
1487 * same top guid, so if we're a top-level vdev, we can
1488 * safely compare to that instead.
1490 * If we split this vdev off instead, then we also check the
1491 * original pool's guid. We don't want to consider the vdev
1492 * corrupt if it is partway through a split operation.
1494 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1496 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1498 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1499 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1500 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1501 VDEV_AUX_CORRUPT_DATA);
1506 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1508 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1509 VDEV_AUX_CORRUPT_DATA);
1517 * If this is a verbatim import, no need to check the
1518 * state of the pool.
1520 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1521 spa_load_state(spa) == SPA_LOAD_OPEN &&
1522 state != POOL_STATE_ACTIVE)
1523 return (SET_ERROR(EBADF));
1526 * If we were able to open and validate a vdev that was
1527 * previously marked permanently unavailable, clear that state
1530 if (vd->vdev_not_present)
1531 vd->vdev_not_present = 0;
1538 * Close a virtual device.
1541 vdev_close(vdev_t *vd)
1543 spa_t *spa = vd->vdev_spa;
1544 vdev_t *pvd = vd->vdev_parent;
1546 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1549 * If our parent is reopening, then we are as well, unless we are
1552 if (pvd != NULL && pvd->vdev_reopening)
1553 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1555 vd->vdev_ops->vdev_op_close(vd);
1557 vdev_cache_purge(vd);
1559 if (vd->vdev_ops->vdev_op_leaf)
1560 trim_map_destroy(vd);
1563 * We record the previous state before we close it, so that if we are
1564 * doing a reopen(), we don't generate FMA ereports if we notice that
1565 * it's still faulted.
1567 vd->vdev_prevstate = vd->vdev_state;
1569 if (vd->vdev_offline)
1570 vd->vdev_state = VDEV_STATE_OFFLINE;
1572 vd->vdev_state = VDEV_STATE_CLOSED;
1573 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1577 vdev_hold(vdev_t *vd)
1579 spa_t *spa = vd->vdev_spa;
1581 ASSERT(spa_is_root(spa));
1582 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1585 for (int c = 0; c < vd->vdev_children; c++)
1586 vdev_hold(vd->vdev_child[c]);
1588 if (vd->vdev_ops->vdev_op_leaf)
1589 vd->vdev_ops->vdev_op_hold(vd);
1593 vdev_rele(vdev_t *vd)
1595 spa_t *spa = vd->vdev_spa;
1597 ASSERT(spa_is_root(spa));
1598 for (int c = 0; c < vd->vdev_children; c++)
1599 vdev_rele(vd->vdev_child[c]);
1601 if (vd->vdev_ops->vdev_op_leaf)
1602 vd->vdev_ops->vdev_op_rele(vd);
1606 * Reopen all interior vdevs and any unopened leaves. We don't actually
1607 * reopen leaf vdevs which had previously been opened as they might deadlock
1608 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1609 * If the leaf has never been opened then open it, as usual.
1612 vdev_reopen(vdev_t *vd)
1614 spa_t *spa = vd->vdev_spa;
1616 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1618 /* set the reopening flag unless we're taking the vdev offline */
1619 vd->vdev_reopening = !vd->vdev_offline;
1621 (void) vdev_open(vd);
1624 * Call vdev_validate() here to make sure we have the same device.
1625 * Otherwise, a device with an invalid label could be successfully
1626 * opened in response to vdev_reopen().
1629 (void) vdev_validate_aux(vd);
1630 if (vdev_readable(vd) && vdev_writeable(vd) &&
1631 vd->vdev_aux == &spa->spa_l2cache &&
1632 !l2arc_vdev_present(vd))
1633 l2arc_add_vdev(spa, vd);
1635 (void) vdev_validate(vd, B_TRUE);
1639 * Reassess parent vdev's health.
1641 vdev_propagate_state(vd);
1645 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1650 * Normally, partial opens (e.g. of a mirror) are allowed.
1651 * For a create, however, we want to fail the request if
1652 * there are any components we can't open.
1654 error = vdev_open(vd);
1656 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1658 return (error ? error : ENXIO);
1662 * Recursively load DTLs and initialize all labels.
1664 if ((error = vdev_dtl_load(vd)) != 0 ||
1665 (error = vdev_label_init(vd, txg, isreplacing ?
1666 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1675 vdev_metaslab_set_size(vdev_t *vd)
1678 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1680 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1681 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1685 * Maximize performance by inflating the configured ashift for top level
1686 * vdevs to be as close to the physical ashift as possible while maintaining
1687 * administrator defined limits and ensuring it doesn't go below the
1691 vdev_ashift_optimize(vdev_t *vd)
1693 if (vd == vd->vdev_top) {
1694 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1695 vd->vdev_ashift = MIN(
1696 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1697 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1700 * Unusual case where logical ashift > physical ashift
1701 * so we can't cap the calculated ashift based on max
1702 * ashift as that would cause failures.
1703 * We still check if we need to increase it to match
1706 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1713 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1715 ASSERT(vd == vd->vdev_top);
1716 ASSERT(!vd->vdev_ishole);
1717 ASSERT(ISP2(flags));
1718 ASSERT(spa_writeable(vd->vdev_spa));
1720 if (flags & VDD_METASLAB)
1721 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1723 if (flags & VDD_DTL)
1724 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1726 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1730 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1732 for (int 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_resilver_txg == 0 ||
1872 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1876 * When a resilver is initiated the scan will assign the scn_max_txg
1877 * value to the highest txg value that exists in all DTLs. If this
1878 * device's max DTL is not part of this scan (i.e. it is not in
1879 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1882 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1883 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1884 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1885 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1892 * Reassess DTLs after a config change or scrub completion.
1895 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1897 spa_t *spa = vd->vdev_spa;
1901 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1903 for (int c = 0; c < vd->vdev_children; c++)
1904 vdev_dtl_reassess(vd->vdev_child[c], txg,
1905 scrub_txg, scrub_done);
1907 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1910 if (vd->vdev_ops->vdev_op_leaf) {
1911 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1913 mutex_enter(&vd->vdev_dtl_lock);
1916 * If we've completed a scan cleanly then determine
1917 * if this vdev should remove any DTLs. We only want to
1918 * excise regions on vdevs that were available during
1919 * the entire duration of this scan.
1921 if (scrub_txg != 0 &&
1922 (spa->spa_scrub_started ||
1923 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1924 vdev_dtl_should_excise(vd)) {
1926 * We completed a scrub up to scrub_txg. If we
1927 * did it without rebooting, then the scrub dtl
1928 * will be valid, so excise the old region and
1929 * fold in the scrub dtl. Otherwise, leave the
1930 * dtl as-is if there was an error.
1932 * There's little trick here: to excise the beginning
1933 * of the DTL_MISSING map, we put it into a reference
1934 * tree and then add a segment with refcnt -1 that
1935 * covers the range [0, scrub_txg). This means
1936 * that each txg in that range has refcnt -1 or 0.
1937 * We then add DTL_SCRUB with a refcnt of 2, so that
1938 * entries in the range [0, scrub_txg) will have a
1939 * positive refcnt -- either 1 or 2. We then convert
1940 * the reference tree into the new DTL_MISSING map.
1942 space_reftree_create(&reftree);
1943 space_reftree_add_map(&reftree,
1944 vd->vdev_dtl[DTL_MISSING], 1);
1945 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1946 space_reftree_add_map(&reftree,
1947 vd->vdev_dtl[DTL_SCRUB], 2);
1948 space_reftree_generate_map(&reftree,
1949 vd->vdev_dtl[DTL_MISSING], 1);
1950 space_reftree_destroy(&reftree);
1952 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1953 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1954 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1956 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1957 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1958 if (!vdev_readable(vd))
1959 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1961 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1962 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1965 * If the vdev was resilvering and no longer has any
1966 * DTLs then reset its resilvering flag and dirty
1967 * the top level so that we persist the change.
1969 if (vd->vdev_resilver_txg != 0 &&
1970 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1971 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
1972 vd->vdev_resilver_txg = 0;
1973 vdev_config_dirty(vd->vdev_top);
1976 mutex_exit(&vd->vdev_dtl_lock);
1979 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1983 mutex_enter(&vd->vdev_dtl_lock);
1984 for (int t = 0; t < DTL_TYPES; t++) {
1985 /* account for child's outage in parent's missing map */
1986 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1988 continue; /* leaf vdevs only */
1989 if (t == DTL_PARTIAL)
1990 minref = 1; /* i.e. non-zero */
1991 else if (vd->vdev_nparity != 0)
1992 minref = vd->vdev_nparity + 1; /* RAID-Z */
1994 minref = vd->vdev_children; /* any kind of mirror */
1995 space_reftree_create(&reftree);
1996 for (int c = 0; c < vd->vdev_children; c++) {
1997 vdev_t *cvd = vd->vdev_child[c];
1998 mutex_enter(&cvd->vdev_dtl_lock);
1999 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2000 mutex_exit(&cvd->vdev_dtl_lock);
2002 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2003 space_reftree_destroy(&reftree);
2005 mutex_exit(&vd->vdev_dtl_lock);
2009 vdev_dtl_load(vdev_t *vd)
2011 spa_t *spa = vd->vdev_spa;
2012 objset_t *mos = spa->spa_meta_objset;
2015 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2016 ASSERT(!vd->vdev_ishole);
2018 error = space_map_open(&vd->vdev_dtl_sm, mos,
2019 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2022 ASSERT(vd->vdev_dtl_sm != NULL);
2024 mutex_enter(&vd->vdev_dtl_lock);
2027 * Now that we've opened the space_map we need to update
2030 space_map_update(vd->vdev_dtl_sm);
2032 error = space_map_load(vd->vdev_dtl_sm,
2033 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2034 mutex_exit(&vd->vdev_dtl_lock);
2039 for (int c = 0; c < vd->vdev_children; c++) {
2040 error = vdev_dtl_load(vd->vdev_child[c]);
2049 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2051 spa_t *spa = vd->vdev_spa;
2052 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2053 objset_t *mos = spa->spa_meta_objset;
2054 range_tree_t *rtsync;
2057 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2059 ASSERT(!vd->vdev_ishole);
2060 ASSERT(vd->vdev_ops->vdev_op_leaf);
2062 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2064 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2065 mutex_enter(&vd->vdev_dtl_lock);
2066 space_map_free(vd->vdev_dtl_sm, tx);
2067 space_map_close(vd->vdev_dtl_sm);
2068 vd->vdev_dtl_sm = NULL;
2069 mutex_exit(&vd->vdev_dtl_lock);
2074 if (vd->vdev_dtl_sm == NULL) {
2075 uint64_t new_object;
2077 new_object = space_map_alloc(mos, tx);
2078 VERIFY3U(new_object, !=, 0);
2080 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2081 0, -1ULL, 0, &vd->vdev_dtl_lock));
2082 ASSERT(vd->vdev_dtl_sm != NULL);
2085 bzero(&rtlock, sizeof(rtlock));
2086 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2088 rtsync = range_tree_create(NULL, NULL, &rtlock);
2090 mutex_enter(&rtlock);
2092 mutex_enter(&vd->vdev_dtl_lock);
2093 range_tree_walk(rt, range_tree_add, rtsync);
2094 mutex_exit(&vd->vdev_dtl_lock);
2096 space_map_truncate(vd->vdev_dtl_sm, tx);
2097 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2098 range_tree_vacate(rtsync, NULL, NULL);
2100 range_tree_destroy(rtsync);
2102 mutex_exit(&rtlock);
2103 mutex_destroy(&rtlock);
2106 * If the object for the space map has changed then dirty
2107 * the top level so that we update the config.
2109 if (object != space_map_object(vd->vdev_dtl_sm)) {
2110 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2111 "new object %llu", txg, spa_name(spa), object,
2112 space_map_object(vd->vdev_dtl_sm));
2113 vdev_config_dirty(vd->vdev_top);
2118 mutex_enter(&vd->vdev_dtl_lock);
2119 space_map_update(vd->vdev_dtl_sm);
2120 mutex_exit(&vd->vdev_dtl_lock);
2124 * Determine whether the specified vdev can be offlined/detached/removed
2125 * without losing data.
2128 vdev_dtl_required(vdev_t *vd)
2130 spa_t *spa = vd->vdev_spa;
2131 vdev_t *tvd = vd->vdev_top;
2132 uint8_t cant_read = vd->vdev_cant_read;
2135 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2137 if (vd == spa->spa_root_vdev || vd == tvd)
2141 * Temporarily mark the device as unreadable, and then determine
2142 * whether this results in any DTL outages in the top-level vdev.
2143 * If not, we can safely offline/detach/remove the device.
2145 vd->vdev_cant_read = B_TRUE;
2146 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2147 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2148 vd->vdev_cant_read = cant_read;
2149 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2151 if (!required && zio_injection_enabled)
2152 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2158 * Determine if resilver is needed, and if so the txg range.
2161 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2163 boolean_t needed = B_FALSE;
2164 uint64_t thismin = UINT64_MAX;
2165 uint64_t thismax = 0;
2167 if (vd->vdev_children == 0) {
2168 mutex_enter(&vd->vdev_dtl_lock);
2169 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2170 vdev_writeable(vd)) {
2172 thismin = vdev_dtl_min(vd);
2173 thismax = vdev_dtl_max(vd);
2176 mutex_exit(&vd->vdev_dtl_lock);
2178 for (int c = 0; c < vd->vdev_children; c++) {
2179 vdev_t *cvd = vd->vdev_child[c];
2180 uint64_t cmin, cmax;
2182 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2183 thismin = MIN(thismin, cmin);
2184 thismax = MAX(thismax, cmax);
2190 if (needed && minp) {
2198 vdev_load(vdev_t *vd)
2201 * Recursively load all children.
2203 for (int c = 0; c < vd->vdev_children; c++)
2204 vdev_load(vd->vdev_child[c]);
2207 * If this is a top-level vdev, initialize its metaslabs.
2209 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2210 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2211 vdev_metaslab_init(vd, 0) != 0))
2212 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2213 VDEV_AUX_CORRUPT_DATA);
2216 * If this is a leaf vdev, load its DTL.
2218 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2219 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2220 VDEV_AUX_CORRUPT_DATA);
2224 * The special vdev case is used for hot spares and l2cache devices. Its
2225 * sole purpose it to set the vdev state for the associated vdev. To do this,
2226 * we make sure that we can open the underlying device, then try to read the
2227 * label, and make sure that the label is sane and that it hasn't been
2228 * repurposed to another pool.
2231 vdev_validate_aux(vdev_t *vd)
2234 uint64_t guid, version;
2237 if (!vdev_readable(vd))
2240 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2241 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2242 VDEV_AUX_CORRUPT_DATA);
2246 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2247 !SPA_VERSION_IS_SUPPORTED(version) ||
2248 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2249 guid != vd->vdev_guid ||
2250 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2251 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2252 VDEV_AUX_CORRUPT_DATA);
2258 * We don't actually check the pool state here. If it's in fact in
2259 * use by another pool, we update this fact on the fly when requested.
2266 vdev_remove(vdev_t *vd, uint64_t txg)
2268 spa_t *spa = vd->vdev_spa;
2269 objset_t *mos = spa->spa_meta_objset;
2272 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2274 if (vd->vdev_ms != NULL) {
2275 metaslab_group_t *mg = vd->vdev_mg;
2277 metaslab_group_histogram_verify(mg);
2278 metaslab_class_histogram_verify(mg->mg_class);
2280 for (int m = 0; m < vd->vdev_ms_count; m++) {
2281 metaslab_t *msp = vd->vdev_ms[m];
2283 if (msp == NULL || msp->ms_sm == NULL)
2286 mutex_enter(&msp->ms_lock);
2288 * If the metaslab was not loaded when the vdev
2289 * was removed then the histogram accounting may
2290 * not be accurate. Update the histogram information
2291 * here so that we ensure that the metaslab group
2292 * and metaslab class are up-to-date.
2294 metaslab_group_histogram_remove(mg, msp);
2296 VERIFY0(space_map_allocated(msp->ms_sm));
2297 space_map_free(msp->ms_sm, tx);
2298 space_map_close(msp->ms_sm);
2300 mutex_exit(&msp->ms_lock);
2303 metaslab_group_histogram_verify(mg);
2304 metaslab_class_histogram_verify(mg->mg_class);
2305 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2306 ASSERT0(mg->mg_histogram[i]);
2310 if (vd->vdev_ms_array) {
2311 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2312 vd->vdev_ms_array = 0;
2318 vdev_sync_done(vdev_t *vd, uint64_t txg)
2321 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2323 ASSERT(!vd->vdev_ishole);
2325 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2326 metaslab_sync_done(msp, txg);
2329 metaslab_sync_reassess(vd->vdev_mg);
2333 vdev_sync(vdev_t *vd, uint64_t txg)
2335 spa_t *spa = vd->vdev_spa;
2340 ASSERT(!vd->vdev_ishole);
2342 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2343 ASSERT(vd == vd->vdev_top);
2344 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2345 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2346 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2347 ASSERT(vd->vdev_ms_array != 0);
2348 vdev_config_dirty(vd);
2353 * Remove the metadata associated with this vdev once it's empty.
2355 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2356 vdev_remove(vd, txg);
2358 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2359 metaslab_sync(msp, txg);
2360 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2363 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2364 vdev_dtl_sync(lvd, txg);
2366 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2370 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2372 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2376 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2377 * not be opened, and no I/O is attempted.
2380 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2384 spa_vdev_state_enter(spa, SCL_NONE);
2386 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2387 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2389 if (!vd->vdev_ops->vdev_op_leaf)
2390 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2395 * We don't directly use the aux state here, but if we do a
2396 * vdev_reopen(), we need this value to be present to remember why we
2399 vd->vdev_label_aux = aux;
2402 * Faulted state takes precedence over degraded.
2404 vd->vdev_delayed_close = B_FALSE;
2405 vd->vdev_faulted = 1ULL;
2406 vd->vdev_degraded = 0ULL;
2407 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2410 * If this device has the only valid copy of the data, then
2411 * back off and simply mark the vdev as degraded instead.
2413 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2414 vd->vdev_degraded = 1ULL;
2415 vd->vdev_faulted = 0ULL;
2418 * If we reopen the device and it's not dead, only then do we
2423 if (vdev_readable(vd))
2424 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2427 return (spa_vdev_state_exit(spa, vd, 0));
2431 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2432 * user that something is wrong. The vdev continues to operate as normal as far
2433 * as I/O is concerned.
2436 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2440 spa_vdev_state_enter(spa, SCL_NONE);
2442 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2443 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2445 if (!vd->vdev_ops->vdev_op_leaf)
2446 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2449 * If the vdev is already faulted, then don't do anything.
2451 if (vd->vdev_faulted || vd->vdev_degraded)
2452 return (spa_vdev_state_exit(spa, NULL, 0));
2454 vd->vdev_degraded = 1ULL;
2455 if (!vdev_is_dead(vd))
2456 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2459 return (spa_vdev_state_exit(spa, vd, 0));
2463 * Online the given vdev.
2465 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2466 * spare device should be detached when the device finishes resilvering.
2467 * Second, the online should be treated like a 'test' online case, so no FMA
2468 * events are generated if the device fails to open.
2471 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2473 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2475 spa_vdev_state_enter(spa, SCL_NONE);
2477 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2478 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2480 if (!vd->vdev_ops->vdev_op_leaf)
2481 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2484 vd->vdev_offline = B_FALSE;
2485 vd->vdev_tmpoffline = B_FALSE;
2486 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2487 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2489 /* XXX - L2ARC 1.0 does not support expansion */
2490 if (!vd->vdev_aux) {
2491 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2492 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2496 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2498 if (!vd->vdev_aux) {
2499 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2500 pvd->vdev_expanding = B_FALSE;
2504 *newstate = vd->vdev_state;
2505 if ((flags & ZFS_ONLINE_UNSPARE) &&
2506 !vdev_is_dead(vd) && vd->vdev_parent &&
2507 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2508 vd->vdev_parent->vdev_child[0] == vd)
2509 vd->vdev_unspare = B_TRUE;
2511 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2513 /* XXX - L2ARC 1.0 does not support expansion */
2515 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2516 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2518 return (spa_vdev_state_exit(spa, vd, 0));
2522 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2526 uint64_t generation;
2527 metaslab_group_t *mg;
2530 spa_vdev_state_enter(spa, SCL_ALLOC);
2532 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2533 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2535 if (!vd->vdev_ops->vdev_op_leaf)
2536 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2540 generation = spa->spa_config_generation + 1;
2543 * If the device isn't already offline, try to offline it.
2545 if (!vd->vdev_offline) {
2547 * If this device has the only valid copy of some data,
2548 * don't allow it to be offlined. Log devices are always
2551 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2552 vdev_dtl_required(vd))
2553 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2556 * If the top-level is a slog and it has had allocations
2557 * then proceed. We check that the vdev's metaslab group
2558 * is not NULL since it's possible that we may have just
2559 * added this vdev but not yet initialized its metaslabs.
2561 if (tvd->vdev_islog && mg != NULL) {
2563 * Prevent any future allocations.
2565 metaslab_group_passivate(mg);
2566 (void) spa_vdev_state_exit(spa, vd, 0);
2568 error = spa_offline_log(spa);
2570 spa_vdev_state_enter(spa, SCL_ALLOC);
2573 * Check to see if the config has changed.
2575 if (error || generation != spa->spa_config_generation) {
2576 metaslab_group_activate(mg);
2578 return (spa_vdev_state_exit(spa,
2580 (void) spa_vdev_state_exit(spa, vd, 0);
2583 ASSERT0(tvd->vdev_stat.vs_alloc);
2587 * Offline this device and reopen its top-level vdev.
2588 * If the top-level vdev is a log device then just offline
2589 * it. Otherwise, if this action results in the top-level
2590 * vdev becoming unusable, undo it and fail the request.
2592 vd->vdev_offline = B_TRUE;
2595 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2596 vdev_is_dead(tvd)) {
2597 vd->vdev_offline = B_FALSE;
2599 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2603 * Add the device back into the metaslab rotor so that
2604 * once we online the device it's open for business.
2606 if (tvd->vdev_islog && mg != NULL)
2607 metaslab_group_activate(mg);
2610 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2612 return (spa_vdev_state_exit(spa, vd, 0));
2616 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2620 mutex_enter(&spa->spa_vdev_top_lock);
2621 error = vdev_offline_locked(spa, guid, flags);
2622 mutex_exit(&spa->spa_vdev_top_lock);
2628 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2629 * vdev_offline(), we assume the spa config is locked. We also clear all
2630 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2633 vdev_clear(spa_t *spa, vdev_t *vd)
2635 vdev_t *rvd = spa->spa_root_vdev;
2637 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2642 vd->vdev_stat.vs_read_errors = 0;
2643 vd->vdev_stat.vs_write_errors = 0;
2644 vd->vdev_stat.vs_checksum_errors = 0;
2646 for (int c = 0; c < vd->vdev_children; c++)
2647 vdev_clear(spa, vd->vdev_child[c]);
2650 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2651 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2653 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2654 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2658 * If we're in the FAULTED state or have experienced failed I/O, then
2659 * clear the persistent state and attempt to reopen the device. We
2660 * also mark the vdev config dirty, so that the new faulted state is
2661 * written out to disk.
2663 if (vd->vdev_faulted || vd->vdev_degraded ||
2664 !vdev_readable(vd) || !vdev_writeable(vd)) {
2667 * When reopening in reponse to a clear event, it may be due to
2668 * a fmadm repair request. In this case, if the device is
2669 * still broken, we want to still post the ereport again.
2671 vd->vdev_forcefault = B_TRUE;
2673 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2674 vd->vdev_cant_read = B_FALSE;
2675 vd->vdev_cant_write = B_FALSE;
2677 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2679 vd->vdev_forcefault = B_FALSE;
2681 if (vd != rvd && vdev_writeable(vd->vdev_top))
2682 vdev_state_dirty(vd->vdev_top);
2684 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2685 spa_async_request(spa, SPA_ASYNC_RESILVER);
2687 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2691 * When clearing a FMA-diagnosed fault, we always want to
2692 * unspare the device, as we assume that the original spare was
2693 * done in response to the FMA fault.
2695 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2696 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2697 vd->vdev_parent->vdev_child[0] == vd)
2698 vd->vdev_unspare = B_TRUE;
2702 vdev_is_dead(vdev_t *vd)
2705 * Holes and missing devices are always considered "dead".
2706 * This simplifies the code since we don't have to check for
2707 * these types of devices in the various code paths.
2708 * Instead we rely on the fact that we skip over dead devices
2709 * before issuing I/O to them.
2711 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2712 vd->vdev_ops == &vdev_missing_ops);
2716 vdev_readable(vdev_t *vd)
2718 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2722 vdev_writeable(vdev_t *vd)
2724 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2728 vdev_allocatable(vdev_t *vd)
2730 uint64_t state = vd->vdev_state;
2733 * We currently allow allocations from vdevs which may be in the
2734 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2735 * fails to reopen then we'll catch it later when we're holding
2736 * the proper locks. Note that we have to get the vdev state
2737 * in a local variable because although it changes atomically,
2738 * we're asking two separate questions about it.
2740 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2741 !vd->vdev_cant_write && !vd->vdev_ishole);
2745 vdev_accessible(vdev_t *vd, zio_t *zio)
2747 ASSERT(zio->io_vd == vd);
2749 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2752 if (zio->io_type == ZIO_TYPE_READ)
2753 return (!vd->vdev_cant_read);
2755 if (zio->io_type == ZIO_TYPE_WRITE)
2756 return (!vd->vdev_cant_write);
2762 * Get statistics for the given vdev.
2765 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2767 spa_t *spa = vd->vdev_spa;
2768 vdev_t *rvd = spa->spa_root_vdev;
2770 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2772 mutex_enter(&vd->vdev_stat_lock);
2773 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2774 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2775 vs->vs_state = vd->vdev_state;
2776 vs->vs_rsize = vdev_get_min_asize(vd);
2777 if (vd->vdev_ops->vdev_op_leaf)
2778 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2779 if (vd->vdev_max_asize != 0)
2780 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2781 vs->vs_configured_ashift = vd->vdev_top != NULL
2782 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2783 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2784 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2785 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2786 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2790 * If we're getting stats on the root vdev, aggregate the I/O counts
2791 * over all top-level vdevs (i.e. the direct children of the root).
2794 for (int c = 0; c < rvd->vdev_children; c++) {
2795 vdev_t *cvd = rvd->vdev_child[c];
2796 vdev_stat_t *cvs = &cvd->vdev_stat;
2798 for (int t = 0; t < ZIO_TYPES; t++) {
2799 vs->vs_ops[t] += cvs->vs_ops[t];
2800 vs->vs_bytes[t] += cvs->vs_bytes[t];
2802 cvs->vs_scan_removing = cvd->vdev_removing;
2805 mutex_exit(&vd->vdev_stat_lock);
2809 vdev_clear_stats(vdev_t *vd)
2811 mutex_enter(&vd->vdev_stat_lock);
2812 vd->vdev_stat.vs_space = 0;
2813 vd->vdev_stat.vs_dspace = 0;
2814 vd->vdev_stat.vs_alloc = 0;
2815 mutex_exit(&vd->vdev_stat_lock);
2819 vdev_scan_stat_init(vdev_t *vd)
2821 vdev_stat_t *vs = &vd->vdev_stat;
2823 for (int c = 0; c < vd->vdev_children; c++)
2824 vdev_scan_stat_init(vd->vdev_child[c]);
2826 mutex_enter(&vd->vdev_stat_lock);
2827 vs->vs_scan_processed = 0;
2828 mutex_exit(&vd->vdev_stat_lock);
2832 vdev_stat_update(zio_t *zio, uint64_t psize)
2834 spa_t *spa = zio->io_spa;
2835 vdev_t *rvd = spa->spa_root_vdev;
2836 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2838 uint64_t txg = zio->io_txg;
2839 vdev_stat_t *vs = &vd->vdev_stat;
2840 zio_type_t type = zio->io_type;
2841 int flags = zio->io_flags;
2844 * If this i/o is a gang leader, it didn't do any actual work.
2846 if (zio->io_gang_tree)
2849 if (zio->io_error == 0) {
2851 * If this is a root i/o, don't count it -- we've already
2852 * counted the top-level vdevs, and vdev_get_stats() will
2853 * aggregate them when asked. This reduces contention on
2854 * the root vdev_stat_lock and implicitly handles blocks
2855 * that compress away to holes, for which there is no i/o.
2856 * (Holes never create vdev children, so all the counters
2857 * remain zero, which is what we want.)
2859 * Note: this only applies to successful i/o (io_error == 0)
2860 * because unlike i/o counts, errors are not additive.
2861 * When reading a ditto block, for example, failure of
2862 * one top-level vdev does not imply a root-level error.
2867 ASSERT(vd == zio->io_vd);
2869 if (flags & ZIO_FLAG_IO_BYPASS)
2872 mutex_enter(&vd->vdev_stat_lock);
2874 if (flags & ZIO_FLAG_IO_REPAIR) {
2875 if (flags & ZIO_FLAG_SCAN_THREAD) {
2876 dsl_scan_phys_t *scn_phys =
2877 &spa->spa_dsl_pool->dp_scan->scn_phys;
2878 uint64_t *processed = &scn_phys->scn_processed;
2881 if (vd->vdev_ops->vdev_op_leaf)
2882 atomic_add_64(processed, psize);
2883 vs->vs_scan_processed += psize;
2886 if (flags & ZIO_FLAG_SELF_HEAL)
2887 vs->vs_self_healed += psize;
2891 vs->vs_bytes[type] += psize;
2893 mutex_exit(&vd->vdev_stat_lock);
2897 if (flags & ZIO_FLAG_SPECULATIVE)
2901 * If this is an I/O error that is going to be retried, then ignore the
2902 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2903 * hard errors, when in reality they can happen for any number of
2904 * innocuous reasons (bus resets, MPxIO link failure, etc).
2906 if (zio->io_error == EIO &&
2907 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2911 * Intent logs writes won't propagate their error to the root
2912 * I/O so don't mark these types of failures as pool-level
2915 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2918 mutex_enter(&vd->vdev_stat_lock);
2919 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2920 if (zio->io_error == ECKSUM)
2921 vs->vs_checksum_errors++;
2923 vs->vs_read_errors++;
2925 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2926 vs->vs_write_errors++;
2927 mutex_exit(&vd->vdev_stat_lock);
2929 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2930 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2931 (flags & ZIO_FLAG_SCAN_THREAD) ||
2932 spa->spa_claiming)) {
2934 * This is either a normal write (not a repair), or it's
2935 * a repair induced by the scrub thread, or it's a repair
2936 * made by zil_claim() during spa_load() in the first txg.
2937 * In the normal case, we commit the DTL change in the same
2938 * txg as the block was born. In the scrub-induced repair
2939 * case, we know that scrubs run in first-pass syncing context,
2940 * so we commit the DTL change in spa_syncing_txg(spa).
2941 * In the zil_claim() case, we commit in spa_first_txg(spa).
2943 * We currently do not make DTL entries for failed spontaneous
2944 * self-healing writes triggered by normal (non-scrubbing)
2945 * reads, because we have no transactional context in which to
2946 * do so -- and it's not clear that it'd be desirable anyway.
2948 if (vd->vdev_ops->vdev_op_leaf) {
2949 uint64_t commit_txg = txg;
2950 if (flags & ZIO_FLAG_SCAN_THREAD) {
2951 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2952 ASSERT(spa_sync_pass(spa) == 1);
2953 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2954 commit_txg = spa_syncing_txg(spa);
2955 } else if (spa->spa_claiming) {
2956 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2957 commit_txg = spa_first_txg(spa);
2959 ASSERT(commit_txg >= spa_syncing_txg(spa));
2960 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2962 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2963 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2964 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2967 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2972 * Update the in-core space usage stats for this vdev, its metaslab class,
2973 * and the root vdev.
2976 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2977 int64_t space_delta)
2979 int64_t dspace_delta = space_delta;
2980 spa_t *spa = vd->vdev_spa;
2981 vdev_t *rvd = spa->spa_root_vdev;
2982 metaslab_group_t *mg = vd->vdev_mg;
2983 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2985 ASSERT(vd == vd->vdev_top);
2988 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2989 * factor. We must calculate this here and not at the root vdev
2990 * because the root vdev's psize-to-asize is simply the max of its
2991 * childrens', thus not accurate enough for us.
2993 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2994 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2995 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2996 vd->vdev_deflate_ratio;
2998 mutex_enter(&vd->vdev_stat_lock);
2999 vd->vdev_stat.vs_alloc += alloc_delta;
3000 vd->vdev_stat.vs_space += space_delta;
3001 vd->vdev_stat.vs_dspace += dspace_delta;
3002 mutex_exit(&vd->vdev_stat_lock);
3004 if (mc == spa_normal_class(spa)) {
3005 mutex_enter(&rvd->vdev_stat_lock);
3006 rvd->vdev_stat.vs_alloc += alloc_delta;
3007 rvd->vdev_stat.vs_space += space_delta;
3008 rvd->vdev_stat.vs_dspace += dspace_delta;
3009 mutex_exit(&rvd->vdev_stat_lock);
3013 ASSERT(rvd == vd->vdev_parent);
3014 ASSERT(vd->vdev_ms_count != 0);
3016 metaslab_class_space_update(mc,
3017 alloc_delta, defer_delta, space_delta, dspace_delta);
3022 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3023 * so that it will be written out next time the vdev configuration is synced.
3024 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3027 vdev_config_dirty(vdev_t *vd)
3029 spa_t *spa = vd->vdev_spa;
3030 vdev_t *rvd = spa->spa_root_vdev;
3033 ASSERT(spa_writeable(spa));
3036 * If this is an aux vdev (as with l2cache and spare devices), then we
3037 * update the vdev config manually and set the sync flag.
3039 if (vd->vdev_aux != NULL) {
3040 spa_aux_vdev_t *sav = vd->vdev_aux;
3044 for (c = 0; c < sav->sav_count; c++) {
3045 if (sav->sav_vdevs[c] == vd)
3049 if (c == sav->sav_count) {
3051 * We're being removed. There's nothing more to do.
3053 ASSERT(sav->sav_sync == B_TRUE);
3057 sav->sav_sync = B_TRUE;
3059 if (nvlist_lookup_nvlist_array(sav->sav_config,
3060 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3061 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3062 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3068 * Setting the nvlist in the middle if the array is a little
3069 * sketchy, but it will work.
3071 nvlist_free(aux[c]);
3072 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3078 * The dirty list is protected by the SCL_CONFIG lock. The caller
3079 * must either hold SCL_CONFIG as writer, or must be the sync thread
3080 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3081 * so this is sufficient to ensure mutual exclusion.
3083 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3084 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3085 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3088 for (c = 0; c < rvd->vdev_children; c++)
3089 vdev_config_dirty(rvd->vdev_child[c]);
3091 ASSERT(vd == vd->vdev_top);
3093 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3095 list_insert_head(&spa->spa_config_dirty_list, vd);
3100 vdev_config_clean(vdev_t *vd)
3102 spa_t *spa = vd->vdev_spa;
3104 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3105 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3106 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3108 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3109 list_remove(&spa->spa_config_dirty_list, vd);
3113 * Mark a top-level vdev's state as dirty, so that the next pass of
3114 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3115 * the state changes from larger config changes because they require
3116 * much less locking, and are often needed for administrative actions.
3119 vdev_state_dirty(vdev_t *vd)
3121 spa_t *spa = vd->vdev_spa;
3123 ASSERT(spa_writeable(spa));
3124 ASSERT(vd == vd->vdev_top);
3127 * The state list is protected by the SCL_STATE lock. The caller
3128 * must either hold SCL_STATE as writer, or must be the sync thread
3129 * (which holds SCL_STATE as reader). There's only one sync thread,
3130 * so this is sufficient to ensure mutual exclusion.
3132 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3133 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3134 spa_config_held(spa, SCL_STATE, RW_READER)));
3136 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3137 list_insert_head(&spa->spa_state_dirty_list, vd);
3141 vdev_state_clean(vdev_t *vd)
3143 spa_t *spa = vd->vdev_spa;
3145 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3146 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3147 spa_config_held(spa, SCL_STATE, RW_READER)));
3149 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3150 list_remove(&spa->spa_state_dirty_list, vd);
3154 * Propagate vdev state up from children to parent.
3157 vdev_propagate_state(vdev_t *vd)
3159 spa_t *spa = vd->vdev_spa;
3160 vdev_t *rvd = spa->spa_root_vdev;
3161 int degraded = 0, faulted = 0;
3165 if (vd->vdev_children > 0) {
3166 for (int c = 0; c < vd->vdev_children; c++) {
3167 child = vd->vdev_child[c];
3170 * Don't factor holes into the decision.
3172 if (child->vdev_ishole)
3175 if (!vdev_readable(child) ||
3176 (!vdev_writeable(child) && spa_writeable(spa))) {
3178 * Root special: if there is a top-level log
3179 * device, treat the root vdev as if it were
3182 if (child->vdev_islog && vd == rvd)
3186 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3190 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3194 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3197 * Root special: if there is a top-level vdev that cannot be
3198 * opened due to corrupted metadata, then propagate the root
3199 * vdev's aux state as 'corrupt' rather than 'insufficient
3202 if (corrupted && vd == rvd &&
3203 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3204 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3205 VDEV_AUX_CORRUPT_DATA);
3208 if (vd->vdev_parent)
3209 vdev_propagate_state(vd->vdev_parent);
3213 * Set a vdev's state. If this is during an open, we don't update the parent
3214 * state, because we're in the process of opening children depth-first.
3215 * Otherwise, we propagate the change to the parent.
3217 * If this routine places a device in a faulted state, an appropriate ereport is
3221 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3223 uint64_t save_state;
3224 spa_t *spa = vd->vdev_spa;
3226 if (state == vd->vdev_state) {
3227 vd->vdev_stat.vs_aux = aux;
3231 save_state = vd->vdev_state;
3233 vd->vdev_state = state;
3234 vd->vdev_stat.vs_aux = aux;
3237 * If we are setting the vdev state to anything but an open state, then
3238 * always close the underlying device unless the device has requested
3239 * a delayed close (i.e. we're about to remove or fault the device).
3240 * Otherwise, we keep accessible but invalid devices open forever.
3241 * We don't call vdev_close() itself, because that implies some extra
3242 * checks (offline, etc) that we don't want here. This is limited to
3243 * leaf devices, because otherwise closing the device will affect other
3246 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3247 vd->vdev_ops->vdev_op_leaf)
3248 vd->vdev_ops->vdev_op_close(vd);
3251 * If we have brought this vdev back into service, we need
3252 * to notify fmd so that it can gracefully repair any outstanding
3253 * cases due to a missing device. We do this in all cases, even those
3254 * that probably don't correlate to a repaired fault. This is sure to
3255 * catch all cases, and we let the zfs-retire agent sort it out. If
3256 * this is a transient state it's OK, as the retire agent will
3257 * double-check the state of the vdev before repairing it.
3259 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3260 vd->vdev_prevstate != state)
3261 zfs_post_state_change(spa, vd);
3263 if (vd->vdev_removed &&
3264 state == VDEV_STATE_CANT_OPEN &&
3265 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3267 * If the previous state is set to VDEV_STATE_REMOVED, then this
3268 * device was previously marked removed and someone attempted to
3269 * reopen it. If this failed due to a nonexistent device, then
3270 * keep the device in the REMOVED state. We also let this be if
3271 * it is one of our special test online cases, which is only
3272 * attempting to online the device and shouldn't generate an FMA
3275 vd->vdev_state = VDEV_STATE_REMOVED;
3276 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3277 } else if (state == VDEV_STATE_REMOVED) {
3278 vd->vdev_removed = B_TRUE;
3279 } else if (state == VDEV_STATE_CANT_OPEN) {
3281 * If we fail to open a vdev during an import or recovery, we
3282 * mark it as "not available", which signifies that it was
3283 * never there to begin with. Failure to open such a device
3284 * is not considered an error.
3286 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3287 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3288 vd->vdev_ops->vdev_op_leaf)
3289 vd->vdev_not_present = 1;
3292 * Post the appropriate ereport. If the 'prevstate' field is
3293 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3294 * that this is part of a vdev_reopen(). In this case, we don't
3295 * want to post the ereport if the device was already in the
3296 * CANT_OPEN state beforehand.
3298 * If the 'checkremove' flag is set, then this is an attempt to
3299 * online the device in response to an insertion event. If we
3300 * hit this case, then we have detected an insertion event for a
3301 * faulted or offline device that wasn't in the removed state.
3302 * In this scenario, we don't post an ereport because we are
3303 * about to replace the device, or attempt an online with
3304 * vdev_forcefault, which will generate the fault for us.
3306 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3307 !vd->vdev_not_present && !vd->vdev_checkremove &&
3308 vd != spa->spa_root_vdev) {
3312 case VDEV_AUX_OPEN_FAILED:
3313 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3315 case VDEV_AUX_CORRUPT_DATA:
3316 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3318 case VDEV_AUX_NO_REPLICAS:
3319 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3321 case VDEV_AUX_BAD_GUID_SUM:
3322 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3324 case VDEV_AUX_TOO_SMALL:
3325 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3327 case VDEV_AUX_BAD_LABEL:
3328 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3331 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3334 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3337 /* Erase any notion of persistent removed state */
3338 vd->vdev_removed = B_FALSE;
3340 vd->vdev_removed = B_FALSE;
3343 if (!isopen && vd->vdev_parent)
3344 vdev_propagate_state(vd->vdev_parent);
3348 * Check the vdev configuration to ensure that it's capable of supporting
3351 * On Solaris, we do not support RAID-Z or partial configuration. In
3352 * addition, only a single top-level vdev is allowed and none of the
3353 * leaves can be wholedisks.
3355 * For FreeBSD, we can boot from any configuration. There is a
3356 * limitation that the boot filesystem must be either uncompressed or
3357 * compresses with lzjb compression but I'm not sure how to enforce
3361 vdev_is_bootable(vdev_t *vd)
3364 if (!vd->vdev_ops->vdev_op_leaf) {
3365 char *vdev_type = vd->vdev_ops->vdev_op_type;
3367 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3368 vd->vdev_children > 1) {
3370 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3371 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3374 } else if (vd->vdev_wholedisk == 1) {
3378 for (int c = 0; c < vd->vdev_children; c++) {
3379 if (!vdev_is_bootable(vd->vdev_child[c]))
3382 #endif /* illumos */
3387 * Load the state from the original vdev tree (ovd) which
3388 * we've retrieved from the MOS config object. If the original
3389 * vdev was offline or faulted then we transfer that state to the
3390 * device in the current vdev tree (nvd).
3393 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3395 spa_t *spa = nvd->vdev_spa;
3397 ASSERT(nvd->vdev_top->vdev_islog);
3398 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3399 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3401 for (int c = 0; c < nvd->vdev_children; c++)
3402 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3404 if (nvd->vdev_ops->vdev_op_leaf) {
3406 * Restore the persistent vdev state
3408 nvd->vdev_offline = ovd->vdev_offline;
3409 nvd->vdev_faulted = ovd->vdev_faulted;
3410 nvd->vdev_degraded = ovd->vdev_degraded;
3411 nvd->vdev_removed = ovd->vdev_removed;
3416 * Determine if a log device has valid content. If the vdev was
3417 * removed or faulted in the MOS config then we know that
3418 * the content on the log device has already been written to the pool.
3421 vdev_log_state_valid(vdev_t *vd)
3423 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3427 for (int c = 0; c < vd->vdev_children; c++)
3428 if (vdev_log_state_valid(vd->vdev_child[c]))
3435 * Expand a vdev if possible.
3438 vdev_expand(vdev_t *vd, uint64_t txg)
3440 ASSERT(vd->vdev_top == vd);
3441 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3443 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3444 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3445 vdev_config_dirty(vd);
3453 vdev_split(vdev_t *vd)
3455 vdev_t *cvd, *pvd = vd->vdev_parent;
3457 vdev_remove_child(pvd, vd);
3458 vdev_compact_children(pvd);
3460 cvd = pvd->vdev_child[0];
3461 if (pvd->vdev_children == 1) {
3462 vdev_remove_parent(cvd);
3463 cvd->vdev_splitting = B_TRUE;
3465 vdev_propagate_state(cvd);
3469 vdev_deadman(vdev_t *vd)
3471 for (int c = 0; c < vd->vdev_children; c++) {
3472 vdev_t *cvd = vd->vdev_child[c];
3477 if (vd->vdev_ops->vdev_op_leaf) {
3478 vdev_queue_t *vq = &vd->vdev_queue;
3480 mutex_enter(&vq->vq_lock);
3481 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3482 spa_t *spa = vd->vdev_spa;
3487 * Look at the head of all the pending queues,
3488 * if any I/O has been outstanding for longer than
3489 * the spa_deadman_synctime we panic the system.
3491 fio = avl_first(&vq->vq_active_tree);
3492 delta = gethrtime() - fio->io_timestamp;
3493 if (delta > spa_deadman_synctime(spa)) {
3494 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3495 "delta %lluns, last io %lluns",
3496 fio->io_timestamp, delta,
3497 vq->vq_io_complete_ts);
3498 fm_panic("I/O to pool '%s' appears to be "
3499 "hung on vdev guid %llu at '%s'.",
3501 (long long unsigned int) vd->vdev_guid,
3505 mutex_exit(&vq->vq_lock);