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,
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_notrim = B_FALSE;
1227 vd->vdev_min_asize = vdev_get_min_asize(vd);
1230 * If this vdev is not removed, check its fault status. If it's
1231 * faulted, bail out of the open.
1233 if (!vd->vdev_removed && vd->vdev_faulted) {
1234 ASSERT(vd->vdev_children == 0);
1235 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1236 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1237 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1238 vd->vdev_label_aux);
1239 return (SET_ERROR(ENXIO));
1240 } else if (vd->vdev_offline) {
1241 ASSERT(vd->vdev_children == 0);
1242 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1243 return (SET_ERROR(ENXIO));
1246 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1247 &logical_ashift, &physical_ashift);
1250 * Reset the vdev_reopening flag so that we actually close
1251 * the vdev on error.
1253 vd->vdev_reopening = B_FALSE;
1254 if (zio_injection_enabled && error == 0)
1255 error = zio_handle_device_injection(vd, NULL, ENXIO);
1258 if (vd->vdev_removed &&
1259 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1260 vd->vdev_removed = B_FALSE;
1262 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1263 vd->vdev_stat.vs_aux);
1267 vd->vdev_removed = B_FALSE;
1270 * Recheck the faulted flag now that we have confirmed that
1271 * the vdev is accessible. If we're faulted, bail.
1273 if (vd->vdev_faulted) {
1274 ASSERT(vd->vdev_children == 0);
1275 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1276 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1277 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1278 vd->vdev_label_aux);
1279 return (SET_ERROR(ENXIO));
1282 if (vd->vdev_degraded) {
1283 ASSERT(vd->vdev_children == 0);
1284 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1285 VDEV_AUX_ERR_EXCEEDED);
1287 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1291 * For hole or missing vdevs we just return success.
1293 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1296 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1297 trim_map_create(vd);
1299 for (int c = 0; c < vd->vdev_children; c++) {
1300 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1301 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1307 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1308 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1310 if (vd->vdev_children == 0) {
1311 if (osize < SPA_MINDEVSIZE) {
1312 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1313 VDEV_AUX_TOO_SMALL);
1314 return (SET_ERROR(EOVERFLOW));
1317 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1318 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1319 VDEV_LABEL_END_SIZE);
1321 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1322 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1323 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1324 VDEV_AUX_TOO_SMALL);
1325 return (SET_ERROR(EOVERFLOW));
1329 max_asize = max_osize;
1332 vd->vdev_psize = psize;
1335 * Make sure the allocatable size hasn't shrunk.
1337 if (asize < vd->vdev_min_asize) {
1338 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1339 VDEV_AUX_BAD_LABEL);
1340 return (SET_ERROR(EINVAL));
1343 vd->vdev_physical_ashift =
1344 MAX(physical_ashift, vd->vdev_physical_ashift);
1345 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1346 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1348 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1349 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1350 VDEV_AUX_ASHIFT_TOO_BIG);
1354 if (vd->vdev_asize == 0) {
1356 * This is the first-ever open, so use the computed values.
1357 * For testing purposes, a higher ashift can be requested.
1359 vd->vdev_asize = asize;
1360 vd->vdev_max_asize = max_asize;
1363 * Make sure the alignment requirement hasn't increased.
1365 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1366 vd->vdev_ops->vdev_op_leaf) {
1367 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1368 VDEV_AUX_BAD_LABEL);
1371 vd->vdev_max_asize = max_asize;
1375 * If all children are healthy and the asize has increased,
1376 * then we've experienced dynamic LUN growth. If automatic
1377 * expansion is enabled then use the additional space.
1379 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1380 (vd->vdev_expanding || spa->spa_autoexpand))
1381 vd->vdev_asize = asize;
1383 vdev_set_min_asize(vd);
1386 * Ensure we can issue some IO before declaring the
1387 * vdev open for business.
1389 if (vd->vdev_ops->vdev_op_leaf &&
1390 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1391 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1392 VDEV_AUX_ERR_EXCEEDED);
1397 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1398 * resilver. But don't do this if we are doing a reopen for a scrub,
1399 * since this would just restart the scrub we are already doing.
1401 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1402 vdev_resilver_needed(vd, NULL, NULL))
1403 spa_async_request(spa, SPA_ASYNC_RESILVER);
1409 * Called once the vdevs are all opened, this routine validates the label
1410 * contents. This needs to be done before vdev_load() so that we don't
1411 * inadvertently do repair I/Os to the wrong device.
1413 * If 'strict' is false ignore the spa guid check. This is necessary because
1414 * if the machine crashed during a re-guid the new guid might have been written
1415 * to all of the vdev labels, but not the cached config. The strict check
1416 * will be performed when the pool is opened again using the mos config.
1418 * This function will only return failure if one of the vdevs indicates that it
1419 * has since been destroyed or exported. This is only possible if
1420 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1421 * will be updated but the function will return 0.
1424 vdev_validate(vdev_t *vd, boolean_t strict)
1426 spa_t *spa = vd->vdev_spa;
1428 uint64_t guid = 0, top_guid;
1431 for (int c = 0; c < vd->vdev_children; c++)
1432 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1433 return (SET_ERROR(EBADF));
1436 * If the device has already failed, or was marked offline, don't do
1437 * any further validation. Otherwise, label I/O will fail and we will
1438 * overwrite the previous state.
1440 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1441 uint64_t aux_guid = 0;
1443 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1444 spa_last_synced_txg(spa) : -1ULL;
1446 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1447 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1448 VDEV_AUX_BAD_LABEL);
1453 * Determine if this vdev has been split off into another
1454 * pool. If so, then refuse to open it.
1456 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1457 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1458 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1459 VDEV_AUX_SPLIT_POOL);
1464 if (strict && (nvlist_lookup_uint64(label,
1465 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1466 guid != spa_guid(spa))) {
1467 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1468 VDEV_AUX_CORRUPT_DATA);
1473 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1474 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1479 * If this vdev just became a top-level vdev because its
1480 * sibling was detached, it will have adopted the parent's
1481 * vdev guid -- but the label may or may not be on disk yet.
1482 * Fortunately, either version of the label will have the
1483 * same top guid, so if we're a top-level vdev, we can
1484 * safely compare to that instead.
1486 * If we split this vdev off instead, then we also check the
1487 * original pool's guid. We don't want to consider the vdev
1488 * corrupt if it is partway through a split operation.
1490 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1492 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1494 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1495 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1496 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1497 VDEV_AUX_CORRUPT_DATA);
1502 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1504 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1505 VDEV_AUX_CORRUPT_DATA);
1513 * If this is a verbatim import, no need to check the
1514 * state of the pool.
1516 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1517 spa_load_state(spa) == SPA_LOAD_OPEN &&
1518 state != POOL_STATE_ACTIVE)
1519 return (SET_ERROR(EBADF));
1522 * If we were able to open and validate a vdev that was
1523 * previously marked permanently unavailable, clear that state
1526 if (vd->vdev_not_present)
1527 vd->vdev_not_present = 0;
1534 * Close a virtual device.
1537 vdev_close(vdev_t *vd)
1539 spa_t *spa = vd->vdev_spa;
1540 vdev_t *pvd = vd->vdev_parent;
1542 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1545 * If our parent is reopening, then we are as well, unless we are
1548 if (pvd != NULL && pvd->vdev_reopening)
1549 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1551 vd->vdev_ops->vdev_op_close(vd);
1553 vdev_cache_purge(vd);
1555 if (vd->vdev_ops->vdev_op_leaf)
1556 trim_map_destroy(vd);
1559 * We record the previous state before we close it, so that if we are
1560 * doing a reopen(), we don't generate FMA ereports if we notice that
1561 * it's still faulted.
1563 vd->vdev_prevstate = vd->vdev_state;
1565 if (vd->vdev_offline)
1566 vd->vdev_state = VDEV_STATE_OFFLINE;
1568 vd->vdev_state = VDEV_STATE_CLOSED;
1569 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1573 vdev_hold(vdev_t *vd)
1575 spa_t *spa = vd->vdev_spa;
1577 ASSERT(spa_is_root(spa));
1578 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1581 for (int c = 0; c < vd->vdev_children; c++)
1582 vdev_hold(vd->vdev_child[c]);
1584 if (vd->vdev_ops->vdev_op_leaf)
1585 vd->vdev_ops->vdev_op_hold(vd);
1589 vdev_rele(vdev_t *vd)
1591 spa_t *spa = vd->vdev_spa;
1593 ASSERT(spa_is_root(spa));
1594 for (int c = 0; c < vd->vdev_children; c++)
1595 vdev_rele(vd->vdev_child[c]);
1597 if (vd->vdev_ops->vdev_op_leaf)
1598 vd->vdev_ops->vdev_op_rele(vd);
1602 * Reopen all interior vdevs and any unopened leaves. We don't actually
1603 * reopen leaf vdevs which had previously been opened as they might deadlock
1604 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1605 * If the leaf has never been opened then open it, as usual.
1608 vdev_reopen(vdev_t *vd)
1610 spa_t *spa = vd->vdev_spa;
1612 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1614 /* set the reopening flag unless we're taking the vdev offline */
1615 vd->vdev_reopening = !vd->vdev_offline;
1617 (void) vdev_open(vd);
1620 * Call vdev_validate() here to make sure we have the same device.
1621 * Otherwise, a device with an invalid label could be successfully
1622 * opened in response to vdev_reopen().
1625 (void) vdev_validate_aux(vd);
1626 if (vdev_readable(vd) && vdev_writeable(vd) &&
1627 vd->vdev_aux == &spa->spa_l2cache &&
1628 !l2arc_vdev_present(vd))
1629 l2arc_add_vdev(spa, vd);
1631 (void) vdev_validate(vd, B_TRUE);
1635 * Reassess parent vdev's health.
1637 vdev_propagate_state(vd);
1641 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1646 * Normally, partial opens (e.g. of a mirror) are allowed.
1647 * For a create, however, we want to fail the request if
1648 * there are any components we can't open.
1650 error = vdev_open(vd);
1652 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1654 return (error ? error : ENXIO);
1658 * Recursively load DTLs and initialize all labels.
1660 if ((error = vdev_dtl_load(vd)) != 0 ||
1661 (error = vdev_label_init(vd, txg, isreplacing ?
1662 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1671 vdev_metaslab_set_size(vdev_t *vd)
1674 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1676 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1677 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1681 * Maximize performance by inflating the configured ashift for top level
1682 * vdevs to be as close to the physical ashift as possible while maintaining
1683 * administrator defined limits and ensuring it doesn't go below the
1687 vdev_ashift_optimize(vdev_t *vd)
1689 if (vd == vd->vdev_top) {
1690 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1691 vd->vdev_ashift = MIN(
1692 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1693 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1696 * Unusual case where logical ashift > physical ashift
1697 * so we can't cap the calculated ashift based on max
1698 * ashift as that would cause failures.
1699 * We still check if we need to increase it to match
1702 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1709 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1711 ASSERT(vd == vd->vdev_top);
1712 ASSERT(!vd->vdev_ishole);
1713 ASSERT(ISP2(flags));
1714 ASSERT(spa_writeable(vd->vdev_spa));
1716 if (flags & VDD_METASLAB)
1717 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1719 if (flags & VDD_DTL)
1720 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1722 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1726 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1728 for (int c = 0; c < vd->vdev_children; c++)
1729 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1731 if (vd->vdev_ops->vdev_op_leaf)
1732 vdev_dirty(vd->vdev_top, flags, vd, txg);
1738 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1739 * the vdev has less than perfect replication. There are four kinds of DTL:
1741 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1743 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1745 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1746 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1747 * txgs that was scrubbed.
1749 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1750 * persistent errors or just some device being offline.
1751 * Unlike the other three, the DTL_OUTAGE map is not generally
1752 * maintained; it's only computed when needed, typically to
1753 * determine whether a device can be detached.
1755 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1756 * either has the data or it doesn't.
1758 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1759 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1760 * if any child is less than fully replicated, then so is its parent.
1761 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1762 * comprising only those txgs which appear in 'maxfaults' or more children;
1763 * those are the txgs we don't have enough replication to read. For example,
1764 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1765 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1766 * two child DTL_MISSING maps.
1768 * It should be clear from the above that to compute the DTLs and outage maps
1769 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1770 * Therefore, that is all we keep on disk. When loading the pool, or after
1771 * a configuration change, we generate all other DTLs from first principles.
1774 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1776 range_tree_t *rt = vd->vdev_dtl[t];
1778 ASSERT(t < DTL_TYPES);
1779 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1780 ASSERT(spa_writeable(vd->vdev_spa));
1782 mutex_enter(rt->rt_lock);
1783 if (!range_tree_contains(rt, txg, size))
1784 range_tree_add(rt, txg, size);
1785 mutex_exit(rt->rt_lock);
1789 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1791 range_tree_t *rt = vd->vdev_dtl[t];
1792 boolean_t dirty = B_FALSE;
1794 ASSERT(t < DTL_TYPES);
1795 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1797 mutex_enter(rt->rt_lock);
1798 if (range_tree_space(rt) != 0)
1799 dirty = range_tree_contains(rt, txg, size);
1800 mutex_exit(rt->rt_lock);
1806 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1808 range_tree_t *rt = vd->vdev_dtl[t];
1811 mutex_enter(rt->rt_lock);
1812 empty = (range_tree_space(rt) == 0);
1813 mutex_exit(rt->rt_lock);
1819 * Returns the lowest txg in the DTL range.
1822 vdev_dtl_min(vdev_t *vd)
1826 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1827 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1828 ASSERT0(vd->vdev_children);
1830 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1831 return (rs->rs_start - 1);
1835 * Returns the highest txg in the DTL.
1838 vdev_dtl_max(vdev_t *vd)
1842 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1843 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1844 ASSERT0(vd->vdev_children);
1846 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1847 return (rs->rs_end);
1851 * Determine if a resilvering vdev should remove any DTL entries from
1852 * its range. If the vdev was resilvering for the entire duration of the
1853 * scan then it should excise that range from its DTLs. Otherwise, this
1854 * vdev is considered partially resilvered and should leave its DTL
1855 * entries intact. The comment in vdev_dtl_reassess() describes how we
1859 vdev_dtl_should_excise(vdev_t *vd)
1861 spa_t *spa = vd->vdev_spa;
1862 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1864 ASSERT0(scn->scn_phys.scn_errors);
1865 ASSERT0(vd->vdev_children);
1867 if (vd->vdev_resilver_txg == 0 ||
1868 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1872 * When a resilver is initiated the scan will assign the scn_max_txg
1873 * value to the highest txg value that exists in all DTLs. If this
1874 * device's max DTL is not part of this scan (i.e. it is not in
1875 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1878 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1879 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1880 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1881 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1888 * Reassess DTLs after a config change or scrub completion.
1891 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1893 spa_t *spa = vd->vdev_spa;
1897 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1899 for (int c = 0; c < vd->vdev_children; c++)
1900 vdev_dtl_reassess(vd->vdev_child[c], txg,
1901 scrub_txg, scrub_done);
1903 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1906 if (vd->vdev_ops->vdev_op_leaf) {
1907 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1909 mutex_enter(&vd->vdev_dtl_lock);
1912 * If we've completed a scan cleanly then determine
1913 * if this vdev should remove any DTLs. We only want to
1914 * excise regions on vdevs that were available during
1915 * the entire duration of this scan.
1917 if (scrub_txg != 0 &&
1918 (spa->spa_scrub_started ||
1919 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1920 vdev_dtl_should_excise(vd)) {
1922 * We completed a scrub up to scrub_txg. If we
1923 * did it without rebooting, then the scrub dtl
1924 * will be valid, so excise the old region and
1925 * fold in the scrub dtl. Otherwise, leave the
1926 * dtl as-is if there was an error.
1928 * There's little trick here: to excise the beginning
1929 * of the DTL_MISSING map, we put it into a reference
1930 * tree and then add a segment with refcnt -1 that
1931 * covers the range [0, scrub_txg). This means
1932 * that each txg in that range has refcnt -1 or 0.
1933 * We then add DTL_SCRUB with a refcnt of 2, so that
1934 * entries in the range [0, scrub_txg) will have a
1935 * positive refcnt -- either 1 or 2. We then convert
1936 * the reference tree into the new DTL_MISSING map.
1938 space_reftree_create(&reftree);
1939 space_reftree_add_map(&reftree,
1940 vd->vdev_dtl[DTL_MISSING], 1);
1941 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1942 space_reftree_add_map(&reftree,
1943 vd->vdev_dtl[DTL_SCRUB], 2);
1944 space_reftree_generate_map(&reftree,
1945 vd->vdev_dtl[DTL_MISSING], 1);
1946 space_reftree_destroy(&reftree);
1948 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1949 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1950 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1952 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1953 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1954 if (!vdev_readable(vd))
1955 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1957 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1958 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1961 * If the vdev was resilvering and no longer has any
1962 * DTLs then reset its resilvering flag and dirty
1963 * the top level so that we persist the change.
1965 if (vd->vdev_resilver_txg != 0 &&
1966 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1967 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
1968 vd->vdev_resilver_txg = 0;
1969 vdev_config_dirty(vd->vdev_top);
1972 mutex_exit(&vd->vdev_dtl_lock);
1975 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1979 mutex_enter(&vd->vdev_dtl_lock);
1980 for (int t = 0; t < DTL_TYPES; t++) {
1981 /* account for child's outage in parent's missing map */
1982 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1984 continue; /* leaf vdevs only */
1985 if (t == DTL_PARTIAL)
1986 minref = 1; /* i.e. non-zero */
1987 else if (vd->vdev_nparity != 0)
1988 minref = vd->vdev_nparity + 1; /* RAID-Z */
1990 minref = vd->vdev_children; /* any kind of mirror */
1991 space_reftree_create(&reftree);
1992 for (int c = 0; c < vd->vdev_children; c++) {
1993 vdev_t *cvd = vd->vdev_child[c];
1994 mutex_enter(&cvd->vdev_dtl_lock);
1995 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1996 mutex_exit(&cvd->vdev_dtl_lock);
1998 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
1999 space_reftree_destroy(&reftree);
2001 mutex_exit(&vd->vdev_dtl_lock);
2005 vdev_dtl_load(vdev_t *vd)
2007 spa_t *spa = vd->vdev_spa;
2008 objset_t *mos = spa->spa_meta_objset;
2011 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2012 ASSERT(!vd->vdev_ishole);
2014 error = space_map_open(&vd->vdev_dtl_sm, mos,
2015 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2018 ASSERT(vd->vdev_dtl_sm != NULL);
2020 mutex_enter(&vd->vdev_dtl_lock);
2023 * Now that we've opened the space_map we need to update
2026 space_map_update(vd->vdev_dtl_sm);
2028 error = space_map_load(vd->vdev_dtl_sm,
2029 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2030 mutex_exit(&vd->vdev_dtl_lock);
2035 for (int c = 0; c < vd->vdev_children; c++) {
2036 error = vdev_dtl_load(vd->vdev_child[c]);
2045 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2047 spa_t *spa = vd->vdev_spa;
2048 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2049 objset_t *mos = spa->spa_meta_objset;
2050 range_tree_t *rtsync;
2053 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2055 ASSERT(!vd->vdev_ishole);
2056 ASSERT(vd->vdev_ops->vdev_op_leaf);
2058 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2060 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2061 mutex_enter(&vd->vdev_dtl_lock);
2062 space_map_free(vd->vdev_dtl_sm, tx);
2063 space_map_close(vd->vdev_dtl_sm);
2064 vd->vdev_dtl_sm = NULL;
2065 mutex_exit(&vd->vdev_dtl_lock);
2070 if (vd->vdev_dtl_sm == NULL) {
2071 uint64_t new_object;
2073 new_object = space_map_alloc(mos, tx);
2074 VERIFY3U(new_object, !=, 0);
2076 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2077 0, -1ULL, 0, &vd->vdev_dtl_lock));
2078 ASSERT(vd->vdev_dtl_sm != NULL);
2081 bzero(&rtlock, sizeof(rtlock));
2082 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2084 rtsync = range_tree_create(NULL, NULL, &rtlock);
2086 mutex_enter(&rtlock);
2088 mutex_enter(&vd->vdev_dtl_lock);
2089 range_tree_walk(rt, range_tree_add, rtsync);
2090 mutex_exit(&vd->vdev_dtl_lock);
2092 space_map_truncate(vd->vdev_dtl_sm, tx);
2093 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2094 range_tree_vacate(rtsync, NULL, NULL);
2096 range_tree_destroy(rtsync);
2098 mutex_exit(&rtlock);
2099 mutex_destroy(&rtlock);
2102 * If the object for the space map has changed then dirty
2103 * the top level so that we update the config.
2105 if (object != space_map_object(vd->vdev_dtl_sm)) {
2106 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2107 "new object %llu", txg, spa_name(spa), object,
2108 space_map_object(vd->vdev_dtl_sm));
2109 vdev_config_dirty(vd->vdev_top);
2114 mutex_enter(&vd->vdev_dtl_lock);
2115 space_map_update(vd->vdev_dtl_sm);
2116 mutex_exit(&vd->vdev_dtl_lock);
2120 * Determine whether the specified vdev can be offlined/detached/removed
2121 * without losing data.
2124 vdev_dtl_required(vdev_t *vd)
2126 spa_t *spa = vd->vdev_spa;
2127 vdev_t *tvd = vd->vdev_top;
2128 uint8_t cant_read = vd->vdev_cant_read;
2131 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2133 if (vd == spa->spa_root_vdev || vd == tvd)
2137 * Temporarily mark the device as unreadable, and then determine
2138 * whether this results in any DTL outages in the top-level vdev.
2139 * If not, we can safely offline/detach/remove the device.
2141 vd->vdev_cant_read = B_TRUE;
2142 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2143 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2144 vd->vdev_cant_read = cant_read;
2145 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2147 if (!required && zio_injection_enabled)
2148 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2154 * Determine if resilver is needed, and if so the txg range.
2157 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2159 boolean_t needed = B_FALSE;
2160 uint64_t thismin = UINT64_MAX;
2161 uint64_t thismax = 0;
2163 if (vd->vdev_children == 0) {
2164 mutex_enter(&vd->vdev_dtl_lock);
2165 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2166 vdev_writeable(vd)) {
2168 thismin = vdev_dtl_min(vd);
2169 thismax = vdev_dtl_max(vd);
2172 mutex_exit(&vd->vdev_dtl_lock);
2174 for (int c = 0; c < vd->vdev_children; c++) {
2175 vdev_t *cvd = vd->vdev_child[c];
2176 uint64_t cmin, cmax;
2178 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2179 thismin = MIN(thismin, cmin);
2180 thismax = MAX(thismax, cmax);
2186 if (needed && minp) {
2194 vdev_load(vdev_t *vd)
2197 * Recursively load all children.
2199 for (int c = 0; c < vd->vdev_children; c++)
2200 vdev_load(vd->vdev_child[c]);
2203 * If this is a top-level vdev, initialize its metaslabs.
2205 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2206 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2207 vdev_metaslab_init(vd, 0) != 0))
2208 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2209 VDEV_AUX_CORRUPT_DATA);
2212 * If this is a leaf vdev, load its DTL.
2214 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2215 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2216 VDEV_AUX_CORRUPT_DATA);
2220 * The special vdev case is used for hot spares and l2cache devices. Its
2221 * sole purpose it to set the vdev state for the associated vdev. To do this,
2222 * we make sure that we can open the underlying device, then try to read the
2223 * label, and make sure that the label is sane and that it hasn't been
2224 * repurposed to another pool.
2227 vdev_validate_aux(vdev_t *vd)
2230 uint64_t guid, version;
2233 if (!vdev_readable(vd))
2236 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2237 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2238 VDEV_AUX_CORRUPT_DATA);
2242 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2243 !SPA_VERSION_IS_SUPPORTED(version) ||
2244 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2245 guid != vd->vdev_guid ||
2246 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2247 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2248 VDEV_AUX_CORRUPT_DATA);
2254 * We don't actually check the pool state here. If it's in fact in
2255 * use by another pool, we update this fact on the fly when requested.
2262 vdev_remove(vdev_t *vd, uint64_t txg)
2264 spa_t *spa = vd->vdev_spa;
2265 objset_t *mos = spa->spa_meta_objset;
2268 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2270 if (vd->vdev_ms != NULL) {
2271 metaslab_group_t *mg = vd->vdev_mg;
2273 metaslab_group_histogram_verify(mg);
2274 metaslab_class_histogram_verify(mg->mg_class);
2276 for (int m = 0; m < vd->vdev_ms_count; m++) {
2277 metaslab_t *msp = vd->vdev_ms[m];
2279 if (msp == NULL || msp->ms_sm == NULL)
2282 mutex_enter(&msp->ms_lock);
2284 * If the metaslab was not loaded when the vdev
2285 * was removed then the histogram accounting may
2286 * not be accurate. Update the histogram information
2287 * here so that we ensure that the metaslab group
2288 * and metaslab class are up-to-date.
2290 metaslab_group_histogram_remove(mg, msp);
2292 VERIFY0(space_map_allocated(msp->ms_sm));
2293 space_map_free(msp->ms_sm, tx);
2294 space_map_close(msp->ms_sm);
2296 mutex_exit(&msp->ms_lock);
2299 metaslab_group_histogram_verify(mg);
2300 metaslab_class_histogram_verify(mg->mg_class);
2301 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2302 ASSERT0(mg->mg_histogram[i]);
2306 if (vd->vdev_ms_array) {
2307 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2308 vd->vdev_ms_array = 0;
2314 vdev_sync_done(vdev_t *vd, uint64_t txg)
2317 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2319 ASSERT(!vd->vdev_ishole);
2321 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2322 metaslab_sync_done(msp, txg);
2325 metaslab_sync_reassess(vd->vdev_mg);
2329 vdev_sync(vdev_t *vd, uint64_t txg)
2331 spa_t *spa = vd->vdev_spa;
2336 ASSERT(!vd->vdev_ishole);
2338 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2339 ASSERT(vd == vd->vdev_top);
2340 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2341 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2342 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2343 ASSERT(vd->vdev_ms_array != 0);
2344 vdev_config_dirty(vd);
2349 * Remove the metadata associated with this vdev once it's empty.
2351 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2352 vdev_remove(vd, txg);
2354 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2355 metaslab_sync(msp, txg);
2356 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2359 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2360 vdev_dtl_sync(lvd, txg);
2362 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2366 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2368 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2372 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2373 * not be opened, and no I/O is attempted.
2376 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2380 spa_vdev_state_enter(spa, SCL_NONE);
2382 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2383 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2385 if (!vd->vdev_ops->vdev_op_leaf)
2386 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2391 * We don't directly use the aux state here, but if we do a
2392 * vdev_reopen(), we need this value to be present to remember why we
2395 vd->vdev_label_aux = aux;
2398 * Faulted state takes precedence over degraded.
2400 vd->vdev_delayed_close = B_FALSE;
2401 vd->vdev_faulted = 1ULL;
2402 vd->vdev_degraded = 0ULL;
2403 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2406 * If this device has the only valid copy of the data, then
2407 * back off and simply mark the vdev as degraded instead.
2409 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2410 vd->vdev_degraded = 1ULL;
2411 vd->vdev_faulted = 0ULL;
2414 * If we reopen the device and it's not dead, only then do we
2419 if (vdev_readable(vd))
2420 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2423 return (spa_vdev_state_exit(spa, vd, 0));
2427 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2428 * user that something is wrong. The vdev continues to operate as normal as far
2429 * as I/O is concerned.
2432 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2436 spa_vdev_state_enter(spa, SCL_NONE);
2438 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2439 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2441 if (!vd->vdev_ops->vdev_op_leaf)
2442 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2445 * If the vdev is already faulted, then don't do anything.
2447 if (vd->vdev_faulted || vd->vdev_degraded)
2448 return (spa_vdev_state_exit(spa, NULL, 0));
2450 vd->vdev_degraded = 1ULL;
2451 if (!vdev_is_dead(vd))
2452 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2455 return (spa_vdev_state_exit(spa, vd, 0));
2459 * Online the given vdev.
2461 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2462 * spare device should be detached when the device finishes resilvering.
2463 * Second, the online should be treated like a 'test' online case, so no FMA
2464 * events are generated if the device fails to open.
2467 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2469 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2471 spa_vdev_state_enter(spa, SCL_NONE);
2473 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2474 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2476 if (!vd->vdev_ops->vdev_op_leaf)
2477 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2480 vd->vdev_offline = B_FALSE;
2481 vd->vdev_tmpoffline = B_FALSE;
2482 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2483 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2485 /* XXX - L2ARC 1.0 does not support expansion */
2486 if (!vd->vdev_aux) {
2487 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2488 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2492 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2494 if (!vd->vdev_aux) {
2495 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2496 pvd->vdev_expanding = B_FALSE;
2500 *newstate = vd->vdev_state;
2501 if ((flags & ZFS_ONLINE_UNSPARE) &&
2502 !vdev_is_dead(vd) && vd->vdev_parent &&
2503 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2504 vd->vdev_parent->vdev_child[0] == vd)
2505 vd->vdev_unspare = B_TRUE;
2507 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2509 /* XXX - L2ARC 1.0 does not support expansion */
2511 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2512 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2514 return (spa_vdev_state_exit(spa, vd, 0));
2518 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2522 uint64_t generation;
2523 metaslab_group_t *mg;
2526 spa_vdev_state_enter(spa, SCL_ALLOC);
2528 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2529 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2531 if (!vd->vdev_ops->vdev_op_leaf)
2532 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2536 generation = spa->spa_config_generation + 1;
2539 * If the device isn't already offline, try to offline it.
2541 if (!vd->vdev_offline) {
2543 * If this device has the only valid copy of some data,
2544 * don't allow it to be offlined. Log devices are always
2547 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2548 vdev_dtl_required(vd))
2549 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2552 * If the top-level is a slog and it has had allocations
2553 * then proceed. We check that the vdev's metaslab group
2554 * is not NULL since it's possible that we may have just
2555 * added this vdev but not yet initialized its metaslabs.
2557 if (tvd->vdev_islog && mg != NULL) {
2559 * Prevent any future allocations.
2561 metaslab_group_passivate(mg);
2562 (void) spa_vdev_state_exit(spa, vd, 0);
2564 error = spa_offline_log(spa);
2566 spa_vdev_state_enter(spa, SCL_ALLOC);
2569 * Check to see if the config has changed.
2571 if (error || generation != spa->spa_config_generation) {
2572 metaslab_group_activate(mg);
2574 return (spa_vdev_state_exit(spa,
2576 (void) spa_vdev_state_exit(spa, vd, 0);
2579 ASSERT0(tvd->vdev_stat.vs_alloc);
2583 * Offline this device and reopen its top-level vdev.
2584 * If the top-level vdev is a log device then just offline
2585 * it. Otherwise, if this action results in the top-level
2586 * vdev becoming unusable, undo it and fail the request.
2588 vd->vdev_offline = B_TRUE;
2591 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2592 vdev_is_dead(tvd)) {
2593 vd->vdev_offline = B_FALSE;
2595 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2599 * Add the device back into the metaslab rotor so that
2600 * once we online the device it's open for business.
2602 if (tvd->vdev_islog && mg != NULL)
2603 metaslab_group_activate(mg);
2606 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2608 return (spa_vdev_state_exit(spa, vd, 0));
2612 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2616 mutex_enter(&spa->spa_vdev_top_lock);
2617 error = vdev_offline_locked(spa, guid, flags);
2618 mutex_exit(&spa->spa_vdev_top_lock);
2624 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2625 * vdev_offline(), we assume the spa config is locked. We also clear all
2626 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2629 vdev_clear(spa_t *spa, vdev_t *vd)
2631 vdev_t *rvd = spa->spa_root_vdev;
2633 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2638 vd->vdev_stat.vs_read_errors = 0;
2639 vd->vdev_stat.vs_write_errors = 0;
2640 vd->vdev_stat.vs_checksum_errors = 0;
2642 for (int c = 0; c < vd->vdev_children; c++)
2643 vdev_clear(spa, vd->vdev_child[c]);
2646 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2647 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2649 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2650 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2654 * If we're in the FAULTED state or have experienced failed I/O, then
2655 * clear the persistent state and attempt to reopen the device. We
2656 * also mark the vdev config dirty, so that the new faulted state is
2657 * written out to disk.
2659 if (vd->vdev_faulted || vd->vdev_degraded ||
2660 !vdev_readable(vd) || !vdev_writeable(vd)) {
2663 * When reopening in reponse to a clear event, it may be due to
2664 * a fmadm repair request. In this case, if the device is
2665 * still broken, we want to still post the ereport again.
2667 vd->vdev_forcefault = B_TRUE;
2669 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2670 vd->vdev_cant_read = B_FALSE;
2671 vd->vdev_cant_write = B_FALSE;
2673 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2675 vd->vdev_forcefault = B_FALSE;
2677 if (vd != rvd && vdev_writeable(vd->vdev_top))
2678 vdev_state_dirty(vd->vdev_top);
2680 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2681 spa_async_request(spa, SPA_ASYNC_RESILVER);
2683 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2687 * When clearing a FMA-diagnosed fault, we always want to
2688 * unspare the device, as we assume that the original spare was
2689 * done in response to the FMA fault.
2691 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2692 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2693 vd->vdev_parent->vdev_child[0] == vd)
2694 vd->vdev_unspare = B_TRUE;
2698 vdev_is_dead(vdev_t *vd)
2701 * Holes and missing devices are always considered "dead".
2702 * This simplifies the code since we don't have to check for
2703 * these types of devices in the various code paths.
2704 * Instead we rely on the fact that we skip over dead devices
2705 * before issuing I/O to them.
2707 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2708 vd->vdev_ops == &vdev_missing_ops);
2712 vdev_readable(vdev_t *vd)
2714 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2718 vdev_writeable(vdev_t *vd)
2720 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2724 vdev_allocatable(vdev_t *vd)
2726 uint64_t state = vd->vdev_state;
2729 * We currently allow allocations from vdevs which may be in the
2730 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2731 * fails to reopen then we'll catch it later when we're holding
2732 * the proper locks. Note that we have to get the vdev state
2733 * in a local variable because although it changes atomically,
2734 * we're asking two separate questions about it.
2736 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2737 !vd->vdev_cant_write && !vd->vdev_ishole);
2741 vdev_accessible(vdev_t *vd, zio_t *zio)
2743 ASSERT(zio->io_vd == vd);
2745 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2748 if (zio->io_type == ZIO_TYPE_READ)
2749 return (!vd->vdev_cant_read);
2751 if (zio->io_type == ZIO_TYPE_WRITE)
2752 return (!vd->vdev_cant_write);
2758 * Get statistics for the given vdev.
2761 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2763 spa_t *spa = vd->vdev_spa;
2764 vdev_t *rvd = spa->spa_root_vdev;
2766 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2768 mutex_enter(&vd->vdev_stat_lock);
2769 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2770 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2771 vs->vs_state = vd->vdev_state;
2772 vs->vs_rsize = vdev_get_min_asize(vd);
2773 if (vd->vdev_ops->vdev_op_leaf)
2774 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2775 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2776 vs->vs_configured_ashift = vd->vdev_top != NULL
2777 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2778 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2779 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2780 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2781 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2785 * If we're getting stats on the root vdev, aggregate the I/O counts
2786 * over all top-level vdevs (i.e. the direct children of the root).
2789 for (int c = 0; c < rvd->vdev_children; c++) {
2790 vdev_t *cvd = rvd->vdev_child[c];
2791 vdev_stat_t *cvs = &cvd->vdev_stat;
2793 for (int t = 0; t < ZIO_TYPES; t++) {
2794 vs->vs_ops[t] += cvs->vs_ops[t];
2795 vs->vs_bytes[t] += cvs->vs_bytes[t];
2797 cvs->vs_scan_removing = cvd->vdev_removing;
2800 mutex_exit(&vd->vdev_stat_lock);
2804 vdev_clear_stats(vdev_t *vd)
2806 mutex_enter(&vd->vdev_stat_lock);
2807 vd->vdev_stat.vs_space = 0;
2808 vd->vdev_stat.vs_dspace = 0;
2809 vd->vdev_stat.vs_alloc = 0;
2810 mutex_exit(&vd->vdev_stat_lock);
2814 vdev_scan_stat_init(vdev_t *vd)
2816 vdev_stat_t *vs = &vd->vdev_stat;
2818 for (int c = 0; c < vd->vdev_children; c++)
2819 vdev_scan_stat_init(vd->vdev_child[c]);
2821 mutex_enter(&vd->vdev_stat_lock);
2822 vs->vs_scan_processed = 0;
2823 mutex_exit(&vd->vdev_stat_lock);
2827 vdev_stat_update(zio_t *zio, uint64_t psize)
2829 spa_t *spa = zio->io_spa;
2830 vdev_t *rvd = spa->spa_root_vdev;
2831 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2833 uint64_t txg = zio->io_txg;
2834 vdev_stat_t *vs = &vd->vdev_stat;
2835 zio_type_t type = zio->io_type;
2836 int flags = zio->io_flags;
2839 * If this i/o is a gang leader, it didn't do any actual work.
2841 if (zio->io_gang_tree)
2844 if (zio->io_error == 0) {
2846 * If this is a root i/o, don't count it -- we've already
2847 * counted the top-level vdevs, and vdev_get_stats() will
2848 * aggregate them when asked. This reduces contention on
2849 * the root vdev_stat_lock and implicitly handles blocks
2850 * that compress away to holes, for which there is no i/o.
2851 * (Holes never create vdev children, so all the counters
2852 * remain zero, which is what we want.)
2854 * Note: this only applies to successful i/o (io_error == 0)
2855 * because unlike i/o counts, errors are not additive.
2856 * When reading a ditto block, for example, failure of
2857 * one top-level vdev does not imply a root-level error.
2862 ASSERT(vd == zio->io_vd);
2864 if (flags & ZIO_FLAG_IO_BYPASS)
2867 mutex_enter(&vd->vdev_stat_lock);
2869 if (flags & ZIO_FLAG_IO_REPAIR) {
2870 if (flags & ZIO_FLAG_SCAN_THREAD) {
2871 dsl_scan_phys_t *scn_phys =
2872 &spa->spa_dsl_pool->dp_scan->scn_phys;
2873 uint64_t *processed = &scn_phys->scn_processed;
2876 if (vd->vdev_ops->vdev_op_leaf)
2877 atomic_add_64(processed, psize);
2878 vs->vs_scan_processed += psize;
2881 if (flags & ZIO_FLAG_SELF_HEAL)
2882 vs->vs_self_healed += psize;
2886 vs->vs_bytes[type] += psize;
2888 mutex_exit(&vd->vdev_stat_lock);
2892 if (flags & ZIO_FLAG_SPECULATIVE)
2896 * If this is an I/O error that is going to be retried, then ignore the
2897 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2898 * hard errors, when in reality they can happen for any number of
2899 * innocuous reasons (bus resets, MPxIO link failure, etc).
2901 if (zio->io_error == EIO &&
2902 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2906 * Intent logs writes won't propagate their error to the root
2907 * I/O so don't mark these types of failures as pool-level
2910 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2913 mutex_enter(&vd->vdev_stat_lock);
2914 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2915 if (zio->io_error == ECKSUM)
2916 vs->vs_checksum_errors++;
2918 vs->vs_read_errors++;
2920 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2921 vs->vs_write_errors++;
2922 mutex_exit(&vd->vdev_stat_lock);
2924 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2925 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2926 (flags & ZIO_FLAG_SCAN_THREAD) ||
2927 spa->spa_claiming)) {
2929 * This is either a normal write (not a repair), or it's
2930 * a repair induced by the scrub thread, or it's a repair
2931 * made by zil_claim() during spa_load() in the first txg.
2932 * In the normal case, we commit the DTL change in the same
2933 * txg as the block was born. In the scrub-induced repair
2934 * case, we know that scrubs run in first-pass syncing context,
2935 * so we commit the DTL change in spa_syncing_txg(spa).
2936 * In the zil_claim() case, we commit in spa_first_txg(spa).
2938 * We currently do not make DTL entries for failed spontaneous
2939 * self-healing writes triggered by normal (non-scrubbing)
2940 * reads, because we have no transactional context in which to
2941 * do so -- and it's not clear that it'd be desirable anyway.
2943 if (vd->vdev_ops->vdev_op_leaf) {
2944 uint64_t commit_txg = txg;
2945 if (flags & ZIO_FLAG_SCAN_THREAD) {
2946 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2947 ASSERT(spa_sync_pass(spa) == 1);
2948 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2949 commit_txg = spa_syncing_txg(spa);
2950 } else if (spa->spa_claiming) {
2951 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2952 commit_txg = spa_first_txg(spa);
2954 ASSERT(commit_txg >= spa_syncing_txg(spa));
2955 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2957 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2958 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2959 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2962 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2967 * Update the in-core space usage stats for this vdev, its metaslab class,
2968 * and the root vdev.
2971 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2972 int64_t space_delta)
2974 int64_t dspace_delta = space_delta;
2975 spa_t *spa = vd->vdev_spa;
2976 vdev_t *rvd = spa->spa_root_vdev;
2977 metaslab_group_t *mg = vd->vdev_mg;
2978 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2980 ASSERT(vd == vd->vdev_top);
2983 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2984 * factor. We must calculate this here and not at the root vdev
2985 * because the root vdev's psize-to-asize is simply the max of its
2986 * childrens', thus not accurate enough for us.
2988 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2989 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2990 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2991 vd->vdev_deflate_ratio;
2993 mutex_enter(&vd->vdev_stat_lock);
2994 vd->vdev_stat.vs_alloc += alloc_delta;
2995 vd->vdev_stat.vs_space += space_delta;
2996 vd->vdev_stat.vs_dspace += dspace_delta;
2997 mutex_exit(&vd->vdev_stat_lock);
2999 if (mc == spa_normal_class(spa)) {
3000 mutex_enter(&rvd->vdev_stat_lock);
3001 rvd->vdev_stat.vs_alloc += alloc_delta;
3002 rvd->vdev_stat.vs_space += space_delta;
3003 rvd->vdev_stat.vs_dspace += dspace_delta;
3004 mutex_exit(&rvd->vdev_stat_lock);
3008 ASSERT(rvd == vd->vdev_parent);
3009 ASSERT(vd->vdev_ms_count != 0);
3011 metaslab_class_space_update(mc,
3012 alloc_delta, defer_delta, space_delta, dspace_delta);
3017 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3018 * so that it will be written out next time the vdev configuration is synced.
3019 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3022 vdev_config_dirty(vdev_t *vd)
3024 spa_t *spa = vd->vdev_spa;
3025 vdev_t *rvd = spa->spa_root_vdev;
3028 ASSERT(spa_writeable(spa));
3031 * If this is an aux vdev (as with l2cache and spare devices), then we
3032 * update the vdev config manually and set the sync flag.
3034 if (vd->vdev_aux != NULL) {
3035 spa_aux_vdev_t *sav = vd->vdev_aux;
3039 for (c = 0; c < sav->sav_count; c++) {
3040 if (sav->sav_vdevs[c] == vd)
3044 if (c == sav->sav_count) {
3046 * We're being removed. There's nothing more to do.
3048 ASSERT(sav->sav_sync == B_TRUE);
3052 sav->sav_sync = B_TRUE;
3054 if (nvlist_lookup_nvlist_array(sav->sav_config,
3055 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3056 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3057 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3063 * Setting the nvlist in the middle if the array is a little
3064 * sketchy, but it will work.
3066 nvlist_free(aux[c]);
3067 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3073 * The dirty list is protected by the SCL_CONFIG lock. The caller
3074 * must either hold SCL_CONFIG as writer, or must be the sync thread
3075 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3076 * so this is sufficient to ensure mutual exclusion.
3078 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3079 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3080 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3083 for (c = 0; c < rvd->vdev_children; c++)
3084 vdev_config_dirty(rvd->vdev_child[c]);
3086 ASSERT(vd == vd->vdev_top);
3088 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3090 list_insert_head(&spa->spa_config_dirty_list, vd);
3095 vdev_config_clean(vdev_t *vd)
3097 spa_t *spa = vd->vdev_spa;
3099 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3100 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3101 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3103 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3104 list_remove(&spa->spa_config_dirty_list, vd);
3108 * Mark a top-level vdev's state as dirty, so that the next pass of
3109 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3110 * the state changes from larger config changes because they require
3111 * much less locking, and are often needed for administrative actions.
3114 vdev_state_dirty(vdev_t *vd)
3116 spa_t *spa = vd->vdev_spa;
3118 ASSERT(spa_writeable(spa));
3119 ASSERT(vd == vd->vdev_top);
3122 * The state list is protected by the SCL_STATE lock. The caller
3123 * must either hold SCL_STATE as writer, or must be the sync thread
3124 * (which holds SCL_STATE as reader). There's only one sync thread,
3125 * so this is sufficient to ensure mutual exclusion.
3127 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3128 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3129 spa_config_held(spa, SCL_STATE, RW_READER)));
3131 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3132 list_insert_head(&spa->spa_state_dirty_list, vd);
3136 vdev_state_clean(vdev_t *vd)
3138 spa_t *spa = vd->vdev_spa;
3140 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3141 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3142 spa_config_held(spa, SCL_STATE, RW_READER)));
3144 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3145 list_remove(&spa->spa_state_dirty_list, vd);
3149 * Propagate vdev state up from children to parent.
3152 vdev_propagate_state(vdev_t *vd)
3154 spa_t *spa = vd->vdev_spa;
3155 vdev_t *rvd = spa->spa_root_vdev;
3156 int degraded = 0, faulted = 0;
3160 if (vd->vdev_children > 0) {
3161 for (int c = 0; c < vd->vdev_children; c++) {
3162 child = vd->vdev_child[c];
3165 * Don't factor holes into the decision.
3167 if (child->vdev_ishole)
3170 if (!vdev_readable(child) ||
3171 (!vdev_writeable(child) && spa_writeable(spa))) {
3173 * Root special: if there is a top-level log
3174 * device, treat the root vdev as if it were
3177 if (child->vdev_islog && vd == rvd)
3181 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3185 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3189 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3192 * Root special: if there is a top-level vdev that cannot be
3193 * opened due to corrupted metadata, then propagate the root
3194 * vdev's aux state as 'corrupt' rather than 'insufficient
3197 if (corrupted && vd == rvd &&
3198 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3199 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3200 VDEV_AUX_CORRUPT_DATA);
3203 if (vd->vdev_parent)
3204 vdev_propagate_state(vd->vdev_parent);
3208 * Set a vdev's state. If this is during an open, we don't update the parent
3209 * state, because we're in the process of opening children depth-first.
3210 * Otherwise, we propagate the change to the parent.
3212 * If this routine places a device in a faulted state, an appropriate ereport is
3216 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3218 uint64_t save_state;
3219 spa_t *spa = vd->vdev_spa;
3221 if (state == vd->vdev_state) {
3222 vd->vdev_stat.vs_aux = aux;
3226 save_state = vd->vdev_state;
3228 vd->vdev_state = state;
3229 vd->vdev_stat.vs_aux = aux;
3232 * If we are setting the vdev state to anything but an open state, then
3233 * always close the underlying device unless the device has requested
3234 * a delayed close (i.e. we're about to remove or fault the device).
3235 * Otherwise, we keep accessible but invalid devices open forever.
3236 * We don't call vdev_close() itself, because that implies some extra
3237 * checks (offline, etc) that we don't want here. This is limited to
3238 * leaf devices, because otherwise closing the device will affect other
3241 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3242 vd->vdev_ops->vdev_op_leaf)
3243 vd->vdev_ops->vdev_op_close(vd);
3246 * If we have brought this vdev back into service, we need
3247 * to notify fmd so that it can gracefully repair any outstanding
3248 * cases due to a missing device. We do this in all cases, even those
3249 * that probably don't correlate to a repaired fault. This is sure to
3250 * catch all cases, and we let the zfs-retire agent sort it out. If
3251 * this is a transient state it's OK, as the retire agent will
3252 * double-check the state of the vdev before repairing it.
3254 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3255 vd->vdev_prevstate != state)
3256 zfs_post_state_change(spa, vd);
3258 if (vd->vdev_removed &&
3259 state == VDEV_STATE_CANT_OPEN &&
3260 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3262 * If the previous state is set to VDEV_STATE_REMOVED, then this
3263 * device was previously marked removed and someone attempted to
3264 * reopen it. If this failed due to a nonexistent device, then
3265 * keep the device in the REMOVED state. We also let this be if
3266 * it is one of our special test online cases, which is only
3267 * attempting to online the device and shouldn't generate an FMA
3270 vd->vdev_state = VDEV_STATE_REMOVED;
3271 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3272 } else if (state == VDEV_STATE_REMOVED) {
3273 vd->vdev_removed = B_TRUE;
3274 } else if (state == VDEV_STATE_CANT_OPEN) {
3276 * If we fail to open a vdev during an import or recovery, we
3277 * mark it as "not available", which signifies that it was
3278 * never there to begin with. Failure to open such a device
3279 * is not considered an error.
3281 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3282 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3283 vd->vdev_ops->vdev_op_leaf)
3284 vd->vdev_not_present = 1;
3287 * Post the appropriate ereport. If the 'prevstate' field is
3288 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3289 * that this is part of a vdev_reopen(). In this case, we don't
3290 * want to post the ereport if the device was already in the
3291 * CANT_OPEN state beforehand.
3293 * If the 'checkremove' flag is set, then this is an attempt to
3294 * online the device in response to an insertion event. If we
3295 * hit this case, then we have detected an insertion event for a
3296 * faulted or offline device that wasn't in the removed state.
3297 * In this scenario, we don't post an ereport because we are
3298 * about to replace the device, or attempt an online with
3299 * vdev_forcefault, which will generate the fault for us.
3301 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3302 !vd->vdev_not_present && !vd->vdev_checkremove &&
3303 vd != spa->spa_root_vdev) {
3307 case VDEV_AUX_OPEN_FAILED:
3308 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3310 case VDEV_AUX_CORRUPT_DATA:
3311 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3313 case VDEV_AUX_NO_REPLICAS:
3314 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3316 case VDEV_AUX_BAD_GUID_SUM:
3317 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3319 case VDEV_AUX_TOO_SMALL:
3320 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3322 case VDEV_AUX_BAD_LABEL:
3323 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3326 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3329 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3332 /* Erase any notion of persistent removed state */
3333 vd->vdev_removed = B_FALSE;
3335 vd->vdev_removed = B_FALSE;
3338 if (!isopen && vd->vdev_parent)
3339 vdev_propagate_state(vd->vdev_parent);
3343 * Check the vdev configuration to ensure that it's capable of supporting
3346 * On Solaris, we do not support RAID-Z or partial configuration. In
3347 * addition, only a single top-level vdev is allowed and none of the
3348 * leaves can be wholedisks.
3350 * For FreeBSD, we can boot from any configuration. There is a
3351 * limitation that the boot filesystem must be either uncompressed or
3352 * compresses with lzjb compression but I'm not sure how to enforce
3356 vdev_is_bootable(vdev_t *vd)
3359 if (!vd->vdev_ops->vdev_op_leaf) {
3360 char *vdev_type = vd->vdev_ops->vdev_op_type;
3362 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3363 vd->vdev_children > 1) {
3365 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3366 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3369 } else if (vd->vdev_wholedisk == 1) {
3373 for (int c = 0; c < vd->vdev_children; c++) {
3374 if (!vdev_is_bootable(vd->vdev_child[c]))
3382 * Load the state from the original vdev tree (ovd) which
3383 * we've retrieved from the MOS config object. If the original
3384 * vdev was offline or faulted then we transfer that state to the
3385 * device in the current vdev tree (nvd).
3388 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3390 spa_t *spa = nvd->vdev_spa;
3392 ASSERT(nvd->vdev_top->vdev_islog);
3393 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3394 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3396 for (int c = 0; c < nvd->vdev_children; c++)
3397 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3399 if (nvd->vdev_ops->vdev_op_leaf) {
3401 * Restore the persistent vdev state
3403 nvd->vdev_offline = ovd->vdev_offline;
3404 nvd->vdev_faulted = ovd->vdev_faulted;
3405 nvd->vdev_degraded = ovd->vdev_degraded;
3406 nvd->vdev_removed = ovd->vdev_removed;
3411 * Determine if a log device has valid content. If the vdev was
3412 * removed or faulted in the MOS config then we know that
3413 * the content on the log device has already been written to the pool.
3416 vdev_log_state_valid(vdev_t *vd)
3418 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3422 for (int c = 0; c < vd->vdev_children; c++)
3423 if (vdev_log_state_valid(vd->vdev_child[c]))
3430 * Expand a vdev if possible.
3433 vdev_expand(vdev_t *vd, uint64_t txg)
3435 ASSERT(vd->vdev_top == vd);
3436 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3438 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3439 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3440 vdev_config_dirty(vd);
3448 vdev_split(vdev_t *vd)
3450 vdev_t *cvd, *pvd = vd->vdev_parent;
3452 vdev_remove_child(pvd, vd);
3453 vdev_compact_children(pvd);
3455 cvd = pvd->vdev_child[0];
3456 if (pvd->vdev_children == 1) {
3457 vdev_remove_parent(cvd);
3458 cvd->vdev_splitting = B_TRUE;
3460 vdev_propagate_state(cvd);
3464 vdev_deadman(vdev_t *vd)
3466 for (int c = 0; c < vd->vdev_children; c++) {
3467 vdev_t *cvd = vd->vdev_child[c];
3472 if (vd->vdev_ops->vdev_op_leaf) {
3473 vdev_queue_t *vq = &vd->vdev_queue;
3475 mutex_enter(&vq->vq_lock);
3476 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3477 spa_t *spa = vd->vdev_spa;
3482 * Look at the head of all the pending queues,
3483 * if any I/O has been outstanding for longer than
3484 * the spa_deadman_synctime we panic the system.
3486 fio = avl_first(&vq->vq_active_tree);
3487 delta = gethrtime() - fio->io_timestamp;
3488 if (delta > spa_deadman_synctime(spa)) {
3489 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3490 "delta %lluns, last io %lluns",
3491 fio->io_timestamp, delta,
3492 vq->vq_io_complete_ts);
3493 fm_panic("I/O to pool '%s' appears to be "
3494 "hung on vdev guid %llu at '%s'.",
3496 (long long unsigned int) vd->vdev_guid,
3500 mutex_exit(&vq->vq_lock);