4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. 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;
281 spa_t *spa = cvd->vdev_spa;
283 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
284 ASSERT(cvd->vdev_parent == NULL);
286 cvd->vdev_parent = pvd;
291 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
293 oldsize = pvd->vdev_children * sizeof (vdev_t *);
294 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
295 newsize = pvd->vdev_children * sizeof (vdev_t *);
297 newchild = kmem_zalloc(newsize, KM_SLEEP);
298 if (pvd->vdev_child != NULL) {
299 bcopy(pvd->vdev_child, newchild, oldsize);
300 kmem_free(pvd->vdev_child, oldsize);
303 pvd->vdev_child = newchild;
304 pvd->vdev_child[id] = cvd;
306 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
307 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
310 * Walk up all ancestors to update guid sum.
312 for (; pvd != NULL; pvd = pvd->vdev_parent)
313 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
317 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
320 uint_t id = cvd->vdev_id;
322 ASSERT(cvd->vdev_parent == pvd);
327 ASSERT(id < pvd->vdev_children);
328 ASSERT(pvd->vdev_child[id] == cvd);
330 pvd->vdev_child[id] = NULL;
331 cvd->vdev_parent = NULL;
333 for (c = 0; c < pvd->vdev_children; c++)
334 if (pvd->vdev_child[c])
337 if (c == pvd->vdev_children) {
338 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
339 pvd->vdev_child = NULL;
340 pvd->vdev_children = 0;
344 * Walk up all ancestors to update guid sum.
346 for (; pvd != NULL; pvd = pvd->vdev_parent)
347 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
351 * Remove any holes in the child array.
354 vdev_compact_children(vdev_t *pvd)
356 vdev_t **newchild, *cvd;
357 int oldc = pvd->vdev_children;
360 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
362 for (int c = newc = 0; c < oldc; c++)
363 if (pvd->vdev_child[c])
366 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
368 for (int c = newc = 0; c < oldc; c++) {
369 if ((cvd = pvd->vdev_child[c]) != NULL) {
370 newchild[newc] = cvd;
371 cvd->vdev_id = newc++;
375 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
376 pvd->vdev_child = newchild;
377 pvd->vdev_children = newc;
381 * Allocate and minimally initialize a vdev_t.
384 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
388 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
390 if (spa->spa_root_vdev == NULL) {
391 ASSERT(ops == &vdev_root_ops);
392 spa->spa_root_vdev = vd;
393 spa->spa_load_guid = spa_generate_guid(NULL);
396 if (guid == 0 && ops != &vdev_hole_ops) {
397 if (spa->spa_root_vdev == vd) {
399 * The root vdev's guid will also be the pool guid,
400 * which must be unique among all pools.
402 guid = spa_generate_guid(NULL);
405 * Any other vdev's guid must be unique within the pool.
407 guid = spa_generate_guid(spa);
409 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
414 vd->vdev_guid = guid;
415 vd->vdev_guid_sum = guid;
417 vd->vdev_state = VDEV_STATE_CLOSED;
418 vd->vdev_ishole = (ops == &vdev_hole_ops);
420 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
421 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
422 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
423 for (int t = 0; t < DTL_TYPES; t++) {
424 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
427 txg_list_create(&vd->vdev_ms_list,
428 offsetof(struct metaslab, ms_txg_node));
429 txg_list_create(&vd->vdev_dtl_list,
430 offsetof(struct vdev, vdev_dtl_node));
431 vd->vdev_stat.vs_timestamp = gethrtime();
439 * Allocate a new vdev. The 'alloctype' is used to control whether we are
440 * creating a new vdev or loading an existing one - the behavior is slightly
441 * different for each case.
444 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
449 uint64_t guid = 0, islog, nparity;
452 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
454 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
455 return (SET_ERROR(EINVAL));
457 if ((ops = vdev_getops(type)) == NULL)
458 return (SET_ERROR(EINVAL));
461 * If this is a load, get the vdev guid from the nvlist.
462 * Otherwise, vdev_alloc_common() will generate one for us.
464 if (alloctype == VDEV_ALLOC_LOAD) {
467 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
469 return (SET_ERROR(EINVAL));
471 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
472 return (SET_ERROR(EINVAL));
473 } else if (alloctype == VDEV_ALLOC_SPARE) {
474 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
475 return (SET_ERROR(EINVAL));
476 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
477 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
478 return (SET_ERROR(EINVAL));
479 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
480 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
481 return (SET_ERROR(EINVAL));
485 * The first allocated vdev must be of type 'root'.
487 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
488 return (SET_ERROR(EINVAL));
491 * Determine whether we're a log vdev.
494 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
495 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
496 return (SET_ERROR(ENOTSUP));
498 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
499 return (SET_ERROR(ENOTSUP));
502 * Set the nparity property for RAID-Z vdevs.
505 if (ops == &vdev_raidz_ops) {
506 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
508 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
509 return (SET_ERROR(EINVAL));
511 * Previous versions could only support 1 or 2 parity
515 spa_version(spa) < SPA_VERSION_RAIDZ2)
516 return (SET_ERROR(ENOTSUP));
518 spa_version(spa) < SPA_VERSION_RAIDZ3)
519 return (SET_ERROR(ENOTSUP));
522 * We require the parity to be specified for SPAs that
523 * support multiple parity levels.
525 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
526 return (SET_ERROR(EINVAL));
528 * Otherwise, we default to 1 parity device for RAID-Z.
535 ASSERT(nparity != -1ULL);
537 vd = vdev_alloc_common(spa, id, guid, ops);
539 vd->vdev_islog = islog;
540 vd->vdev_nparity = nparity;
542 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
543 vd->vdev_path = spa_strdup(vd->vdev_path);
544 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
545 vd->vdev_devid = spa_strdup(vd->vdev_devid);
546 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
547 &vd->vdev_physpath) == 0)
548 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
549 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
550 vd->vdev_fru = spa_strdup(vd->vdev_fru);
553 * Set the whole_disk property. If it's not specified, leave the value
556 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
557 &vd->vdev_wholedisk) != 0)
558 vd->vdev_wholedisk = -1ULL;
561 * Look for the 'not present' flag. This will only be set if the device
562 * was not present at the time of import.
564 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
565 &vd->vdev_not_present);
568 * Get the alignment requirement.
570 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
573 * Retrieve the vdev creation time.
575 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
579 * If we're a top-level vdev, try to load the allocation parameters.
581 if (parent && !parent->vdev_parent &&
582 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
583 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
585 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
587 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
589 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
593 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
594 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
595 alloctype == VDEV_ALLOC_ADD ||
596 alloctype == VDEV_ALLOC_SPLIT ||
597 alloctype == VDEV_ALLOC_ROOTPOOL);
598 vd->vdev_mg = metaslab_group_create(islog ?
599 spa_log_class(spa) : spa_normal_class(spa), vd);
603 * If we're a leaf vdev, try to load the DTL object and other state.
605 if (vd->vdev_ops->vdev_op_leaf &&
606 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
607 alloctype == VDEV_ALLOC_ROOTPOOL)) {
608 if (alloctype == VDEV_ALLOC_LOAD) {
609 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
610 &vd->vdev_dtl_object);
611 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
615 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
618 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
619 &spare) == 0 && spare)
623 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
626 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
627 &vd->vdev_resilver_txg);
630 * When importing a pool, we want to ignore the persistent fault
631 * state, as the diagnosis made on another system may not be
632 * valid in the current context. Local vdevs will
633 * remain in the faulted state.
635 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
636 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
638 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
640 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
643 if (vd->vdev_faulted || vd->vdev_degraded) {
647 VDEV_AUX_ERR_EXCEEDED;
648 if (nvlist_lookup_string(nv,
649 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
650 strcmp(aux, "external") == 0)
651 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
657 * Add ourselves to the parent's list of children.
659 vdev_add_child(parent, vd);
667 vdev_free(vdev_t *vd)
669 spa_t *spa = vd->vdev_spa;
672 * vdev_free() implies closing the vdev first. This is simpler than
673 * trying to ensure complicated semantics for all callers.
677 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
678 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
683 for (int c = 0; c < vd->vdev_children; c++)
684 vdev_free(vd->vdev_child[c]);
686 ASSERT(vd->vdev_child == NULL);
687 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
690 * Discard allocation state.
692 if (vd->vdev_mg != NULL) {
693 vdev_metaslab_fini(vd);
694 metaslab_group_destroy(vd->vdev_mg);
697 ASSERT0(vd->vdev_stat.vs_space);
698 ASSERT0(vd->vdev_stat.vs_dspace);
699 ASSERT0(vd->vdev_stat.vs_alloc);
702 * Remove this vdev from its parent's child list.
704 vdev_remove_child(vd->vdev_parent, vd);
706 ASSERT(vd->vdev_parent == NULL);
709 * Clean up vdev structure.
715 spa_strfree(vd->vdev_path);
717 spa_strfree(vd->vdev_devid);
718 if (vd->vdev_physpath)
719 spa_strfree(vd->vdev_physpath);
721 spa_strfree(vd->vdev_fru);
723 if (vd->vdev_isspare)
724 spa_spare_remove(vd);
725 if (vd->vdev_isl2cache)
726 spa_l2cache_remove(vd);
728 txg_list_destroy(&vd->vdev_ms_list);
729 txg_list_destroy(&vd->vdev_dtl_list);
731 mutex_enter(&vd->vdev_dtl_lock);
732 space_map_close(vd->vdev_dtl_sm);
733 for (int t = 0; t < DTL_TYPES; t++) {
734 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
735 range_tree_destroy(vd->vdev_dtl[t]);
737 mutex_exit(&vd->vdev_dtl_lock);
739 mutex_destroy(&vd->vdev_dtl_lock);
740 mutex_destroy(&vd->vdev_stat_lock);
741 mutex_destroy(&vd->vdev_probe_lock);
743 if (vd == spa->spa_root_vdev)
744 spa->spa_root_vdev = NULL;
746 kmem_free(vd, sizeof (vdev_t));
750 * Transfer top-level vdev state from svd to tvd.
753 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
755 spa_t *spa = svd->vdev_spa;
760 ASSERT(tvd == tvd->vdev_top);
762 tvd->vdev_ms_array = svd->vdev_ms_array;
763 tvd->vdev_ms_shift = svd->vdev_ms_shift;
764 tvd->vdev_ms_count = svd->vdev_ms_count;
766 svd->vdev_ms_array = 0;
767 svd->vdev_ms_shift = 0;
768 svd->vdev_ms_count = 0;
771 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
772 tvd->vdev_mg = svd->vdev_mg;
773 tvd->vdev_ms = svd->vdev_ms;
778 if (tvd->vdev_mg != NULL)
779 tvd->vdev_mg->mg_vd = tvd;
781 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
782 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
783 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
785 svd->vdev_stat.vs_alloc = 0;
786 svd->vdev_stat.vs_space = 0;
787 svd->vdev_stat.vs_dspace = 0;
789 for (t = 0; t < TXG_SIZE; t++) {
790 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
791 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
792 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
793 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
794 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
795 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
798 if (list_link_active(&svd->vdev_config_dirty_node)) {
799 vdev_config_clean(svd);
800 vdev_config_dirty(tvd);
803 if (list_link_active(&svd->vdev_state_dirty_node)) {
804 vdev_state_clean(svd);
805 vdev_state_dirty(tvd);
808 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
809 svd->vdev_deflate_ratio = 0;
811 tvd->vdev_islog = svd->vdev_islog;
816 vdev_top_update(vdev_t *tvd, vdev_t *vd)
823 for (int c = 0; c < vd->vdev_children; c++)
824 vdev_top_update(tvd, vd->vdev_child[c]);
828 * Add a mirror/replacing vdev above an existing vdev.
831 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
833 spa_t *spa = cvd->vdev_spa;
834 vdev_t *pvd = cvd->vdev_parent;
837 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
839 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
841 mvd->vdev_asize = cvd->vdev_asize;
842 mvd->vdev_min_asize = cvd->vdev_min_asize;
843 mvd->vdev_max_asize = cvd->vdev_max_asize;
844 mvd->vdev_ashift = cvd->vdev_ashift;
845 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
846 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
847 mvd->vdev_state = cvd->vdev_state;
848 mvd->vdev_crtxg = cvd->vdev_crtxg;
850 vdev_remove_child(pvd, cvd);
851 vdev_add_child(pvd, mvd);
852 cvd->vdev_id = mvd->vdev_children;
853 vdev_add_child(mvd, cvd);
854 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
856 if (mvd == mvd->vdev_top)
857 vdev_top_transfer(cvd, mvd);
863 * Remove a 1-way mirror/replacing vdev from the tree.
866 vdev_remove_parent(vdev_t *cvd)
868 vdev_t *mvd = cvd->vdev_parent;
869 vdev_t *pvd = mvd->vdev_parent;
871 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
873 ASSERT(mvd->vdev_children == 1);
874 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
875 mvd->vdev_ops == &vdev_replacing_ops ||
876 mvd->vdev_ops == &vdev_spare_ops);
877 cvd->vdev_ashift = mvd->vdev_ashift;
878 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
879 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
881 vdev_remove_child(mvd, cvd);
882 vdev_remove_child(pvd, mvd);
885 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
886 * Otherwise, we could have detached an offline device, and when we
887 * go to import the pool we'll think we have two top-level vdevs,
888 * instead of a different version of the same top-level vdev.
890 if (mvd->vdev_top == mvd) {
891 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
892 cvd->vdev_orig_guid = cvd->vdev_guid;
893 cvd->vdev_guid += guid_delta;
894 cvd->vdev_guid_sum += guid_delta;
896 cvd->vdev_id = mvd->vdev_id;
897 vdev_add_child(pvd, cvd);
898 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
900 if (cvd == cvd->vdev_top)
901 vdev_top_transfer(mvd, cvd);
903 ASSERT(mvd->vdev_children == 0);
908 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
910 spa_t *spa = vd->vdev_spa;
911 objset_t *mos = spa->spa_meta_objset;
913 uint64_t oldc = vd->vdev_ms_count;
914 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
918 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
921 * This vdev is not being allocated from yet or is a hole.
923 if (vd->vdev_ms_shift == 0)
926 ASSERT(!vd->vdev_ishole);
929 * Compute the raidz-deflation ratio. Note, we hard-code
930 * in 128k (1 << 17) because it is the "typical" blocksize.
931 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
932 * otherwise it would inconsistently account for existing bp's.
934 vd->vdev_deflate_ratio = (1 << 17) /
935 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
937 ASSERT(oldc <= newc);
939 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
942 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
943 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
947 vd->vdev_ms_count = newc;
949 for (m = oldc; m < newc; m++) {
953 error = dmu_read(mos, vd->vdev_ms_array,
954 m * sizeof (uint64_t), sizeof (uint64_t), &object,
960 error = metaslab_init(vd->vdev_mg, m, object, txg,
967 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
970 * If the vdev is being removed we don't activate
971 * the metaslabs since we want to ensure that no new
972 * allocations are performed on this device.
974 if (oldc == 0 && !vd->vdev_removing)
975 metaslab_group_activate(vd->vdev_mg);
978 spa_config_exit(spa, SCL_ALLOC, FTAG);
984 vdev_metaslab_fini(vdev_t *vd)
987 uint64_t count = vd->vdev_ms_count;
989 if (vd->vdev_ms != NULL) {
990 metaslab_group_passivate(vd->vdev_mg);
991 for (m = 0; m < count; m++) {
992 metaslab_t *msp = vd->vdev_ms[m];
997 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1002 typedef struct vdev_probe_stats {
1003 boolean_t vps_readable;
1004 boolean_t vps_writeable;
1006 } vdev_probe_stats_t;
1009 vdev_probe_done(zio_t *zio)
1011 spa_t *spa = zio->io_spa;
1012 vdev_t *vd = zio->io_vd;
1013 vdev_probe_stats_t *vps = zio->io_private;
1015 ASSERT(vd->vdev_probe_zio != NULL);
1017 if (zio->io_type == ZIO_TYPE_READ) {
1018 if (zio->io_error == 0)
1019 vps->vps_readable = 1;
1020 if (zio->io_error == 0 && spa_writeable(spa)) {
1021 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1022 zio->io_offset, zio->io_size, zio->io_data,
1023 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1024 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1026 zio_buf_free(zio->io_data, zio->io_size);
1028 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1029 if (zio->io_error == 0)
1030 vps->vps_writeable = 1;
1031 zio_buf_free(zio->io_data, zio->io_size);
1032 } else if (zio->io_type == ZIO_TYPE_NULL) {
1035 vd->vdev_cant_read |= !vps->vps_readable;
1036 vd->vdev_cant_write |= !vps->vps_writeable;
1038 if (vdev_readable(vd) &&
1039 (vdev_writeable(vd) || !spa_writeable(spa))) {
1042 ASSERT(zio->io_error != 0);
1043 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1044 spa, vd, NULL, 0, 0);
1045 zio->io_error = SET_ERROR(ENXIO);
1048 mutex_enter(&vd->vdev_probe_lock);
1049 ASSERT(vd->vdev_probe_zio == zio);
1050 vd->vdev_probe_zio = NULL;
1051 mutex_exit(&vd->vdev_probe_lock);
1053 while ((pio = zio_walk_parents(zio)) != NULL)
1054 if (!vdev_accessible(vd, pio))
1055 pio->io_error = SET_ERROR(ENXIO);
1057 kmem_free(vps, sizeof (*vps));
1062 * Determine whether this device is accessible.
1064 * Read and write to several known locations: the pad regions of each
1065 * vdev label but the first, which we leave alone in case it contains
1069 vdev_probe(vdev_t *vd, zio_t *zio)
1071 spa_t *spa = vd->vdev_spa;
1072 vdev_probe_stats_t *vps = NULL;
1075 ASSERT(vd->vdev_ops->vdev_op_leaf);
1078 * Don't probe the probe.
1080 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1084 * To prevent 'probe storms' when a device fails, we create
1085 * just one probe i/o at a time. All zios that want to probe
1086 * this vdev will become parents of the probe io.
1088 mutex_enter(&vd->vdev_probe_lock);
1090 if ((pio = vd->vdev_probe_zio) == NULL) {
1091 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1093 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1094 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1097 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1099 * vdev_cant_read and vdev_cant_write can only
1100 * transition from TRUE to FALSE when we have the
1101 * SCL_ZIO lock as writer; otherwise they can only
1102 * transition from FALSE to TRUE. This ensures that
1103 * any zio looking at these values can assume that
1104 * failures persist for the life of the I/O. That's
1105 * important because when a device has intermittent
1106 * connectivity problems, we want to ensure that
1107 * they're ascribed to the device (ENXIO) and not
1110 * Since we hold SCL_ZIO as writer here, clear both
1111 * values so the probe can reevaluate from first
1114 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1115 vd->vdev_cant_read = B_FALSE;
1116 vd->vdev_cant_write = B_FALSE;
1119 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1120 vdev_probe_done, vps,
1121 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1124 * We can't change the vdev state in this context, so we
1125 * kick off an async task to do it on our behalf.
1128 vd->vdev_probe_wanted = B_TRUE;
1129 spa_async_request(spa, SPA_ASYNC_PROBE);
1134 zio_add_child(zio, pio);
1136 mutex_exit(&vd->vdev_probe_lock);
1139 ASSERT(zio != NULL);
1143 for (int l = 1; l < VDEV_LABELS; l++) {
1144 zio_nowait(zio_read_phys(pio, vd,
1145 vdev_label_offset(vd->vdev_psize, l,
1146 offsetof(vdev_label_t, vl_pad2)),
1147 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1148 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1149 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1160 vdev_open_child(void *arg)
1164 vd->vdev_open_thread = curthread;
1165 vd->vdev_open_error = vdev_open(vd);
1166 vd->vdev_open_thread = NULL;
1170 vdev_uses_zvols(vdev_t *vd)
1172 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1173 strlen(ZVOL_DIR)) == 0)
1175 for (int c = 0; c < vd->vdev_children; c++)
1176 if (vdev_uses_zvols(vd->vdev_child[c]))
1182 vdev_open_children(vdev_t *vd)
1185 int children = vd->vdev_children;
1188 * in order to handle pools on top of zvols, do the opens
1189 * in a single thread so that the same thread holds the
1190 * spa_namespace_lock
1192 if (B_TRUE || vdev_uses_zvols(vd)) {
1193 for (int c = 0; c < children; c++)
1194 vd->vdev_child[c]->vdev_open_error =
1195 vdev_open(vd->vdev_child[c]);
1198 tq = taskq_create("vdev_open", children, minclsyspri,
1199 children, children, TASKQ_PREPOPULATE);
1201 for (int c = 0; c < children; c++)
1202 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1209 * Prepare a virtual device for access.
1212 vdev_open(vdev_t *vd)
1214 spa_t *spa = vd->vdev_spa;
1217 uint64_t max_osize = 0;
1218 uint64_t asize, max_asize, psize;
1219 uint64_t logical_ashift = 0;
1220 uint64_t physical_ashift = 0;
1222 ASSERT(vd->vdev_open_thread == curthread ||
1223 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1224 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1225 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1226 vd->vdev_state == VDEV_STATE_OFFLINE);
1228 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1229 vd->vdev_cant_read = B_FALSE;
1230 vd->vdev_cant_write = B_FALSE;
1231 vd->vdev_notrim = B_FALSE;
1232 vd->vdev_min_asize = vdev_get_min_asize(vd);
1235 * If this vdev is not removed, check its fault status. If it's
1236 * faulted, bail out of the open.
1238 if (!vd->vdev_removed && vd->vdev_faulted) {
1239 ASSERT(vd->vdev_children == 0);
1240 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1241 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1242 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1243 vd->vdev_label_aux);
1244 return (SET_ERROR(ENXIO));
1245 } else if (vd->vdev_offline) {
1246 ASSERT(vd->vdev_children == 0);
1247 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1248 return (SET_ERROR(ENXIO));
1251 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1252 &logical_ashift, &physical_ashift);
1255 * Reset the vdev_reopening flag so that we actually close
1256 * the vdev on error.
1258 vd->vdev_reopening = B_FALSE;
1259 if (zio_injection_enabled && error == 0)
1260 error = zio_handle_device_injection(vd, NULL, ENXIO);
1263 if (vd->vdev_removed &&
1264 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1265 vd->vdev_removed = B_FALSE;
1267 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1268 vd->vdev_stat.vs_aux);
1272 vd->vdev_removed = B_FALSE;
1275 * Recheck the faulted flag now that we have confirmed that
1276 * the vdev is accessible. If we're faulted, bail.
1278 if (vd->vdev_faulted) {
1279 ASSERT(vd->vdev_children == 0);
1280 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1281 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1282 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1283 vd->vdev_label_aux);
1284 return (SET_ERROR(ENXIO));
1287 if (vd->vdev_degraded) {
1288 ASSERT(vd->vdev_children == 0);
1289 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1290 VDEV_AUX_ERR_EXCEEDED);
1292 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1296 * For hole or missing vdevs we just return success.
1298 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1301 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1302 trim_map_create(vd);
1304 for (int c = 0; c < vd->vdev_children; c++) {
1305 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1306 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1312 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1313 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1315 if (vd->vdev_children == 0) {
1316 if (osize < SPA_MINDEVSIZE) {
1317 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1318 VDEV_AUX_TOO_SMALL);
1319 return (SET_ERROR(EOVERFLOW));
1322 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1323 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1324 VDEV_LABEL_END_SIZE);
1326 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1327 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1328 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1329 VDEV_AUX_TOO_SMALL);
1330 return (SET_ERROR(EOVERFLOW));
1334 max_asize = max_osize;
1337 vd->vdev_psize = psize;
1340 * Make sure the allocatable size hasn't shrunk.
1342 if (asize < vd->vdev_min_asize) {
1343 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1344 VDEV_AUX_BAD_LABEL);
1345 return (SET_ERROR(EINVAL));
1348 vd->vdev_physical_ashift =
1349 MAX(physical_ashift, vd->vdev_physical_ashift);
1350 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1351 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1353 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1354 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1355 VDEV_AUX_ASHIFT_TOO_BIG);
1359 if (vd->vdev_asize == 0) {
1361 * This is the first-ever open, so use the computed values.
1362 * For testing purposes, a higher ashift can be requested.
1364 vd->vdev_asize = asize;
1365 vd->vdev_max_asize = max_asize;
1368 * Make sure the alignment requirement hasn't increased.
1370 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1371 vd->vdev_ops->vdev_op_leaf) {
1372 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1373 VDEV_AUX_BAD_LABEL);
1376 vd->vdev_max_asize = max_asize;
1380 * If all children are healthy and the asize has increased,
1381 * then we've experienced dynamic LUN growth. If automatic
1382 * expansion is enabled then use the additional space.
1384 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1385 (vd->vdev_expanding || spa->spa_autoexpand))
1386 vd->vdev_asize = asize;
1388 vdev_set_min_asize(vd);
1391 * Ensure we can issue some IO before declaring the
1392 * vdev open for business.
1394 if (vd->vdev_ops->vdev_op_leaf &&
1395 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1396 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1397 VDEV_AUX_ERR_EXCEEDED);
1402 * Track the min and max ashift values for normal data devices.
1404 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1405 !vd->vdev_islog && vd->vdev_aux == NULL) {
1406 if (vd->vdev_ashift > spa->spa_max_ashift)
1407 spa->spa_max_ashift = vd->vdev_ashift;
1408 if (vd->vdev_ashift < spa->spa_min_ashift)
1409 spa->spa_min_ashift = vd->vdev_ashift;
1413 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1414 * resilver. But don't do this if we are doing a reopen for a scrub,
1415 * since this would just restart the scrub we are already doing.
1417 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1418 vdev_resilver_needed(vd, NULL, NULL))
1419 spa_async_request(spa, SPA_ASYNC_RESILVER);
1425 * Called once the vdevs are all opened, this routine validates the label
1426 * contents. This needs to be done before vdev_load() so that we don't
1427 * inadvertently do repair I/Os to the wrong device.
1429 * If 'strict' is false ignore the spa guid check. This is necessary because
1430 * if the machine crashed during a re-guid the new guid might have been written
1431 * to all of the vdev labels, but not the cached config. The strict check
1432 * will be performed when the pool is opened again using the mos config.
1434 * This function will only return failure if one of the vdevs indicates that it
1435 * has since been destroyed or exported. This is only possible if
1436 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1437 * will be updated but the function will return 0.
1440 vdev_validate(vdev_t *vd, boolean_t strict)
1442 spa_t *spa = vd->vdev_spa;
1444 uint64_t guid = 0, top_guid;
1447 for (int c = 0; c < vd->vdev_children; c++)
1448 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1449 return (SET_ERROR(EBADF));
1452 * If the device has already failed, or was marked offline, don't do
1453 * any further validation. Otherwise, label I/O will fail and we will
1454 * overwrite the previous state.
1456 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1457 uint64_t aux_guid = 0;
1459 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1460 spa_last_synced_txg(spa) : -1ULL;
1462 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1463 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1464 VDEV_AUX_BAD_LABEL);
1469 * Determine if this vdev has been split off into another
1470 * pool. If so, then refuse to open it.
1472 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1473 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1474 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1475 VDEV_AUX_SPLIT_POOL);
1480 if (strict && (nvlist_lookup_uint64(label,
1481 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1482 guid != spa_guid(spa))) {
1483 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1484 VDEV_AUX_CORRUPT_DATA);
1489 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1490 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1495 * If this vdev just became a top-level vdev because its
1496 * sibling was detached, it will have adopted the parent's
1497 * vdev guid -- but the label may or may not be on disk yet.
1498 * Fortunately, either version of the label will have the
1499 * same top guid, so if we're a top-level vdev, we can
1500 * safely compare to that instead.
1502 * If we split this vdev off instead, then we also check the
1503 * original pool's guid. We don't want to consider the vdev
1504 * corrupt if it is partway through a split operation.
1506 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1508 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1510 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1511 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1512 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1513 VDEV_AUX_CORRUPT_DATA);
1518 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1520 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1521 VDEV_AUX_CORRUPT_DATA);
1529 * If this is a verbatim import, no need to check the
1530 * state of the pool.
1532 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1533 spa_load_state(spa) == SPA_LOAD_OPEN &&
1534 state != POOL_STATE_ACTIVE)
1535 return (SET_ERROR(EBADF));
1538 * If we were able to open and validate a vdev that was
1539 * previously marked permanently unavailable, clear that state
1542 if (vd->vdev_not_present)
1543 vd->vdev_not_present = 0;
1550 * Close a virtual device.
1553 vdev_close(vdev_t *vd)
1555 spa_t *spa = vd->vdev_spa;
1556 vdev_t *pvd = vd->vdev_parent;
1558 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1561 * If our parent is reopening, then we are as well, unless we are
1564 if (pvd != NULL && pvd->vdev_reopening)
1565 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1567 vd->vdev_ops->vdev_op_close(vd);
1569 vdev_cache_purge(vd);
1571 if (vd->vdev_ops->vdev_op_leaf)
1572 trim_map_destroy(vd);
1575 * We record the previous state before we close it, so that if we are
1576 * doing a reopen(), we don't generate FMA ereports if we notice that
1577 * it's still faulted.
1579 vd->vdev_prevstate = vd->vdev_state;
1581 if (vd->vdev_offline)
1582 vd->vdev_state = VDEV_STATE_OFFLINE;
1584 vd->vdev_state = VDEV_STATE_CLOSED;
1585 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1589 vdev_hold(vdev_t *vd)
1591 spa_t *spa = vd->vdev_spa;
1593 ASSERT(spa_is_root(spa));
1594 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1597 for (int c = 0; c < vd->vdev_children; c++)
1598 vdev_hold(vd->vdev_child[c]);
1600 if (vd->vdev_ops->vdev_op_leaf)
1601 vd->vdev_ops->vdev_op_hold(vd);
1605 vdev_rele(vdev_t *vd)
1607 spa_t *spa = vd->vdev_spa;
1609 ASSERT(spa_is_root(spa));
1610 for (int c = 0; c < vd->vdev_children; c++)
1611 vdev_rele(vd->vdev_child[c]);
1613 if (vd->vdev_ops->vdev_op_leaf)
1614 vd->vdev_ops->vdev_op_rele(vd);
1618 * Reopen all interior vdevs and any unopened leaves. We don't actually
1619 * reopen leaf vdevs which had previously been opened as they might deadlock
1620 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1621 * If the leaf has never been opened then open it, as usual.
1624 vdev_reopen(vdev_t *vd)
1626 spa_t *spa = vd->vdev_spa;
1628 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1630 /* set the reopening flag unless we're taking the vdev offline */
1631 vd->vdev_reopening = !vd->vdev_offline;
1633 (void) vdev_open(vd);
1636 * Call vdev_validate() here to make sure we have the same device.
1637 * Otherwise, a device with an invalid label could be successfully
1638 * opened in response to vdev_reopen().
1641 (void) vdev_validate_aux(vd);
1642 if (vdev_readable(vd) && vdev_writeable(vd) &&
1643 vd->vdev_aux == &spa->spa_l2cache &&
1644 !l2arc_vdev_present(vd))
1645 l2arc_add_vdev(spa, vd);
1647 (void) vdev_validate(vd, B_TRUE);
1651 * Reassess parent vdev's health.
1653 vdev_propagate_state(vd);
1657 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1662 * Normally, partial opens (e.g. of a mirror) are allowed.
1663 * For a create, however, we want to fail the request if
1664 * there are any components we can't open.
1666 error = vdev_open(vd);
1668 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1670 return (error ? error : ENXIO);
1674 * Recursively load DTLs and initialize all labels.
1676 if ((error = vdev_dtl_load(vd)) != 0 ||
1677 (error = vdev_label_init(vd, txg, isreplacing ?
1678 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1687 vdev_metaslab_set_size(vdev_t *vd)
1690 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1692 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1693 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1697 * Maximize performance by inflating the configured ashift for top level
1698 * vdevs to be as close to the physical ashift as possible while maintaining
1699 * administrator defined limits and ensuring it doesn't go below the
1703 vdev_ashift_optimize(vdev_t *vd)
1705 if (vd == vd->vdev_top) {
1706 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1707 vd->vdev_ashift = MIN(
1708 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1709 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1712 * Unusual case where logical ashift > physical ashift
1713 * so we can't cap the calculated ashift based on max
1714 * ashift as that would cause failures.
1715 * We still check if we need to increase it to match
1718 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1725 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1727 ASSERT(vd == vd->vdev_top);
1728 ASSERT(!vd->vdev_ishole);
1729 ASSERT(ISP2(flags));
1730 ASSERT(spa_writeable(vd->vdev_spa));
1732 if (flags & VDD_METASLAB)
1733 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1735 if (flags & VDD_DTL)
1736 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1738 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1742 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1744 for (int c = 0; c < vd->vdev_children; c++)
1745 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1747 if (vd->vdev_ops->vdev_op_leaf)
1748 vdev_dirty(vd->vdev_top, flags, vd, txg);
1754 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1755 * the vdev has less than perfect replication. There are four kinds of DTL:
1757 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1759 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1761 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1762 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1763 * txgs that was scrubbed.
1765 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1766 * persistent errors or just some device being offline.
1767 * Unlike the other three, the DTL_OUTAGE map is not generally
1768 * maintained; it's only computed when needed, typically to
1769 * determine whether a device can be detached.
1771 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1772 * either has the data or it doesn't.
1774 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1775 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1776 * if any child is less than fully replicated, then so is its parent.
1777 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1778 * comprising only those txgs which appear in 'maxfaults' or more children;
1779 * those are the txgs we don't have enough replication to read. For example,
1780 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1781 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1782 * two child DTL_MISSING maps.
1784 * It should be clear from the above that to compute the DTLs and outage maps
1785 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1786 * Therefore, that is all we keep on disk. When loading the pool, or after
1787 * a configuration change, we generate all other DTLs from first principles.
1790 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1792 range_tree_t *rt = vd->vdev_dtl[t];
1794 ASSERT(t < DTL_TYPES);
1795 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1796 ASSERT(spa_writeable(vd->vdev_spa));
1798 mutex_enter(rt->rt_lock);
1799 if (!range_tree_contains(rt, txg, size))
1800 range_tree_add(rt, txg, size);
1801 mutex_exit(rt->rt_lock);
1805 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1807 range_tree_t *rt = vd->vdev_dtl[t];
1808 boolean_t dirty = B_FALSE;
1810 ASSERT(t < DTL_TYPES);
1811 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1813 mutex_enter(rt->rt_lock);
1814 if (range_tree_space(rt) != 0)
1815 dirty = range_tree_contains(rt, txg, size);
1816 mutex_exit(rt->rt_lock);
1822 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1824 range_tree_t *rt = vd->vdev_dtl[t];
1827 mutex_enter(rt->rt_lock);
1828 empty = (range_tree_space(rt) == 0);
1829 mutex_exit(rt->rt_lock);
1835 * Returns the lowest txg in the DTL range.
1838 vdev_dtl_min(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_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1847 return (rs->rs_start - 1);
1851 * Returns the highest txg in the DTL.
1854 vdev_dtl_max(vdev_t *vd)
1858 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1859 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1860 ASSERT0(vd->vdev_children);
1862 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1863 return (rs->rs_end);
1867 * Determine if a resilvering vdev should remove any DTL entries from
1868 * its range. If the vdev was resilvering for the entire duration of the
1869 * scan then it should excise that range from its DTLs. Otherwise, this
1870 * vdev is considered partially resilvered and should leave its DTL
1871 * entries intact. The comment in vdev_dtl_reassess() describes how we
1875 vdev_dtl_should_excise(vdev_t *vd)
1877 spa_t *spa = vd->vdev_spa;
1878 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1880 ASSERT0(scn->scn_phys.scn_errors);
1881 ASSERT0(vd->vdev_children);
1883 if (vd->vdev_resilver_txg == 0 ||
1884 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1888 * When a resilver is initiated the scan will assign the scn_max_txg
1889 * value to the highest txg value that exists in all DTLs. If this
1890 * device's max DTL is not part of this scan (i.e. it is not in
1891 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1894 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1895 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1896 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1897 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1904 * Reassess DTLs after a config change or scrub completion.
1907 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1909 spa_t *spa = vd->vdev_spa;
1913 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1915 for (int c = 0; c < vd->vdev_children; c++)
1916 vdev_dtl_reassess(vd->vdev_child[c], txg,
1917 scrub_txg, scrub_done);
1919 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1922 if (vd->vdev_ops->vdev_op_leaf) {
1923 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1925 mutex_enter(&vd->vdev_dtl_lock);
1928 * If we've completed a scan cleanly then determine
1929 * if this vdev should remove any DTLs. We only want to
1930 * excise regions on vdevs that were available during
1931 * the entire duration of this scan.
1933 if (scrub_txg != 0 &&
1934 (spa->spa_scrub_started ||
1935 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1936 vdev_dtl_should_excise(vd)) {
1938 * We completed a scrub up to scrub_txg. If we
1939 * did it without rebooting, then the scrub dtl
1940 * will be valid, so excise the old region and
1941 * fold in the scrub dtl. Otherwise, leave the
1942 * dtl as-is if there was an error.
1944 * There's little trick here: to excise the beginning
1945 * of the DTL_MISSING map, we put it into a reference
1946 * tree and then add a segment with refcnt -1 that
1947 * covers the range [0, scrub_txg). This means
1948 * that each txg in that range has refcnt -1 or 0.
1949 * We then add DTL_SCRUB with a refcnt of 2, so that
1950 * entries in the range [0, scrub_txg) will have a
1951 * positive refcnt -- either 1 or 2. We then convert
1952 * the reference tree into the new DTL_MISSING map.
1954 space_reftree_create(&reftree);
1955 space_reftree_add_map(&reftree,
1956 vd->vdev_dtl[DTL_MISSING], 1);
1957 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1958 space_reftree_add_map(&reftree,
1959 vd->vdev_dtl[DTL_SCRUB], 2);
1960 space_reftree_generate_map(&reftree,
1961 vd->vdev_dtl[DTL_MISSING], 1);
1962 space_reftree_destroy(&reftree);
1964 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1965 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1966 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1968 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1969 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1970 if (!vdev_readable(vd))
1971 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1973 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1974 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1977 * If the vdev was resilvering and no longer has any
1978 * DTLs then reset its resilvering flag and dirty
1979 * the top level so that we persist the change.
1981 if (vd->vdev_resilver_txg != 0 &&
1982 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1983 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
1984 vd->vdev_resilver_txg = 0;
1985 vdev_config_dirty(vd->vdev_top);
1988 mutex_exit(&vd->vdev_dtl_lock);
1991 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1995 mutex_enter(&vd->vdev_dtl_lock);
1996 for (int t = 0; t < DTL_TYPES; t++) {
1997 /* account for child's outage in parent's missing map */
1998 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2000 continue; /* leaf vdevs only */
2001 if (t == DTL_PARTIAL)
2002 minref = 1; /* i.e. non-zero */
2003 else if (vd->vdev_nparity != 0)
2004 minref = vd->vdev_nparity + 1; /* RAID-Z */
2006 minref = vd->vdev_children; /* any kind of mirror */
2007 space_reftree_create(&reftree);
2008 for (int c = 0; c < vd->vdev_children; c++) {
2009 vdev_t *cvd = vd->vdev_child[c];
2010 mutex_enter(&cvd->vdev_dtl_lock);
2011 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2012 mutex_exit(&cvd->vdev_dtl_lock);
2014 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2015 space_reftree_destroy(&reftree);
2017 mutex_exit(&vd->vdev_dtl_lock);
2021 vdev_dtl_load(vdev_t *vd)
2023 spa_t *spa = vd->vdev_spa;
2024 objset_t *mos = spa->spa_meta_objset;
2027 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2028 ASSERT(!vd->vdev_ishole);
2030 error = space_map_open(&vd->vdev_dtl_sm, mos,
2031 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2034 ASSERT(vd->vdev_dtl_sm != NULL);
2036 mutex_enter(&vd->vdev_dtl_lock);
2039 * Now that we've opened the space_map we need to update
2042 space_map_update(vd->vdev_dtl_sm);
2044 error = space_map_load(vd->vdev_dtl_sm,
2045 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2046 mutex_exit(&vd->vdev_dtl_lock);
2051 for (int c = 0; c < vd->vdev_children; c++) {
2052 error = vdev_dtl_load(vd->vdev_child[c]);
2061 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2063 spa_t *spa = vd->vdev_spa;
2064 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2065 objset_t *mos = spa->spa_meta_objset;
2066 range_tree_t *rtsync;
2069 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2071 ASSERT(!vd->vdev_ishole);
2072 ASSERT(vd->vdev_ops->vdev_op_leaf);
2074 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2076 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2077 mutex_enter(&vd->vdev_dtl_lock);
2078 space_map_free(vd->vdev_dtl_sm, tx);
2079 space_map_close(vd->vdev_dtl_sm);
2080 vd->vdev_dtl_sm = NULL;
2081 mutex_exit(&vd->vdev_dtl_lock);
2086 if (vd->vdev_dtl_sm == NULL) {
2087 uint64_t new_object;
2089 new_object = space_map_alloc(mos, tx);
2090 VERIFY3U(new_object, !=, 0);
2092 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2093 0, -1ULL, 0, &vd->vdev_dtl_lock));
2094 ASSERT(vd->vdev_dtl_sm != NULL);
2097 bzero(&rtlock, sizeof(rtlock));
2098 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2100 rtsync = range_tree_create(NULL, NULL, &rtlock);
2102 mutex_enter(&rtlock);
2104 mutex_enter(&vd->vdev_dtl_lock);
2105 range_tree_walk(rt, range_tree_add, rtsync);
2106 mutex_exit(&vd->vdev_dtl_lock);
2108 space_map_truncate(vd->vdev_dtl_sm, tx);
2109 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2110 range_tree_vacate(rtsync, NULL, NULL);
2112 range_tree_destroy(rtsync);
2114 mutex_exit(&rtlock);
2115 mutex_destroy(&rtlock);
2118 * If the object for the space map has changed then dirty
2119 * the top level so that we update the config.
2121 if (object != space_map_object(vd->vdev_dtl_sm)) {
2122 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2123 "new object %llu", txg, spa_name(spa), object,
2124 space_map_object(vd->vdev_dtl_sm));
2125 vdev_config_dirty(vd->vdev_top);
2130 mutex_enter(&vd->vdev_dtl_lock);
2131 space_map_update(vd->vdev_dtl_sm);
2132 mutex_exit(&vd->vdev_dtl_lock);
2136 * Determine whether the specified vdev can be offlined/detached/removed
2137 * without losing data.
2140 vdev_dtl_required(vdev_t *vd)
2142 spa_t *spa = vd->vdev_spa;
2143 vdev_t *tvd = vd->vdev_top;
2144 uint8_t cant_read = vd->vdev_cant_read;
2147 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2149 if (vd == spa->spa_root_vdev || vd == tvd)
2153 * Temporarily mark the device as unreadable, and then determine
2154 * whether this results in any DTL outages in the top-level vdev.
2155 * If not, we can safely offline/detach/remove the device.
2157 vd->vdev_cant_read = B_TRUE;
2158 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2159 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2160 vd->vdev_cant_read = cant_read;
2161 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2163 if (!required && zio_injection_enabled)
2164 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2170 * Determine if resilver is needed, and if so the txg range.
2173 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2175 boolean_t needed = B_FALSE;
2176 uint64_t thismin = UINT64_MAX;
2177 uint64_t thismax = 0;
2179 if (vd->vdev_children == 0) {
2180 mutex_enter(&vd->vdev_dtl_lock);
2181 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2182 vdev_writeable(vd)) {
2184 thismin = vdev_dtl_min(vd);
2185 thismax = vdev_dtl_max(vd);
2188 mutex_exit(&vd->vdev_dtl_lock);
2190 for (int c = 0; c < vd->vdev_children; c++) {
2191 vdev_t *cvd = vd->vdev_child[c];
2192 uint64_t cmin, cmax;
2194 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2195 thismin = MIN(thismin, cmin);
2196 thismax = MAX(thismax, cmax);
2202 if (needed && minp) {
2210 vdev_load(vdev_t *vd)
2213 * Recursively load all children.
2215 for (int c = 0; c < vd->vdev_children; c++)
2216 vdev_load(vd->vdev_child[c]);
2219 * If this is a top-level vdev, initialize its metaslabs.
2221 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2222 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2223 vdev_metaslab_init(vd, 0) != 0))
2224 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2225 VDEV_AUX_CORRUPT_DATA);
2228 * If this is a leaf vdev, load its DTL.
2230 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2231 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2232 VDEV_AUX_CORRUPT_DATA);
2236 * The special vdev case is used for hot spares and l2cache devices. Its
2237 * sole purpose it to set the vdev state for the associated vdev. To do this,
2238 * we make sure that we can open the underlying device, then try to read the
2239 * label, and make sure that the label is sane and that it hasn't been
2240 * repurposed to another pool.
2243 vdev_validate_aux(vdev_t *vd)
2246 uint64_t guid, version;
2249 if (!vdev_readable(vd))
2252 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2253 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2254 VDEV_AUX_CORRUPT_DATA);
2258 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2259 !SPA_VERSION_IS_SUPPORTED(version) ||
2260 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2261 guid != vd->vdev_guid ||
2262 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2263 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2264 VDEV_AUX_CORRUPT_DATA);
2270 * We don't actually check the pool state here. If it's in fact in
2271 * use by another pool, we update this fact on the fly when requested.
2278 vdev_remove(vdev_t *vd, uint64_t txg)
2280 spa_t *spa = vd->vdev_spa;
2281 objset_t *mos = spa->spa_meta_objset;
2284 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2286 if (vd->vdev_ms != NULL) {
2287 metaslab_group_t *mg = vd->vdev_mg;
2289 metaslab_group_histogram_verify(mg);
2290 metaslab_class_histogram_verify(mg->mg_class);
2292 for (int m = 0; m < vd->vdev_ms_count; m++) {
2293 metaslab_t *msp = vd->vdev_ms[m];
2295 if (msp == NULL || msp->ms_sm == NULL)
2298 mutex_enter(&msp->ms_lock);
2300 * If the metaslab was not loaded when the vdev
2301 * was removed then the histogram accounting may
2302 * not be accurate. Update the histogram information
2303 * here so that we ensure that the metaslab group
2304 * and metaslab class are up-to-date.
2306 metaslab_group_histogram_remove(mg, msp);
2308 VERIFY0(space_map_allocated(msp->ms_sm));
2309 space_map_free(msp->ms_sm, tx);
2310 space_map_close(msp->ms_sm);
2312 mutex_exit(&msp->ms_lock);
2315 metaslab_group_histogram_verify(mg);
2316 metaslab_class_histogram_verify(mg->mg_class);
2317 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2318 ASSERT0(mg->mg_histogram[i]);
2322 if (vd->vdev_ms_array) {
2323 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2324 vd->vdev_ms_array = 0;
2330 vdev_sync_done(vdev_t *vd, uint64_t txg)
2333 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2335 ASSERT(!vd->vdev_ishole);
2337 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2338 metaslab_sync_done(msp, txg);
2341 metaslab_sync_reassess(vd->vdev_mg);
2345 vdev_sync(vdev_t *vd, uint64_t txg)
2347 spa_t *spa = vd->vdev_spa;
2352 ASSERT(!vd->vdev_ishole);
2354 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2355 ASSERT(vd == vd->vdev_top);
2356 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2357 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2358 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2359 ASSERT(vd->vdev_ms_array != 0);
2360 vdev_config_dirty(vd);
2365 * Remove the metadata associated with this vdev once it's empty.
2367 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2368 vdev_remove(vd, txg);
2370 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2371 metaslab_sync(msp, txg);
2372 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2375 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2376 vdev_dtl_sync(lvd, txg);
2378 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2382 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2384 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2388 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2389 * not be opened, and no I/O is attempted.
2392 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2396 spa_vdev_state_enter(spa, SCL_NONE);
2398 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2399 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2401 if (!vd->vdev_ops->vdev_op_leaf)
2402 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2407 * We don't directly use the aux state here, but if we do a
2408 * vdev_reopen(), we need this value to be present to remember why we
2411 vd->vdev_label_aux = aux;
2414 * Faulted state takes precedence over degraded.
2416 vd->vdev_delayed_close = B_FALSE;
2417 vd->vdev_faulted = 1ULL;
2418 vd->vdev_degraded = 0ULL;
2419 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2422 * If this device has the only valid copy of the data, then
2423 * back off and simply mark the vdev as degraded instead.
2425 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2426 vd->vdev_degraded = 1ULL;
2427 vd->vdev_faulted = 0ULL;
2430 * If we reopen the device and it's not dead, only then do we
2435 if (vdev_readable(vd))
2436 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2439 return (spa_vdev_state_exit(spa, vd, 0));
2443 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2444 * user that something is wrong. The vdev continues to operate as normal as far
2445 * as I/O is concerned.
2448 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2452 spa_vdev_state_enter(spa, SCL_NONE);
2454 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2455 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2457 if (!vd->vdev_ops->vdev_op_leaf)
2458 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2461 * If the vdev is already faulted, then don't do anything.
2463 if (vd->vdev_faulted || vd->vdev_degraded)
2464 return (spa_vdev_state_exit(spa, NULL, 0));
2466 vd->vdev_degraded = 1ULL;
2467 if (!vdev_is_dead(vd))
2468 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2471 return (spa_vdev_state_exit(spa, vd, 0));
2475 * Online the given vdev.
2477 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2478 * spare device should be detached when the device finishes resilvering.
2479 * Second, the online should be treated like a 'test' online case, so no FMA
2480 * events are generated if the device fails to open.
2483 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2485 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2487 spa_vdev_state_enter(spa, SCL_NONE);
2489 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2490 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2492 if (!vd->vdev_ops->vdev_op_leaf)
2493 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2496 vd->vdev_offline = B_FALSE;
2497 vd->vdev_tmpoffline = B_FALSE;
2498 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2499 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2501 /* XXX - L2ARC 1.0 does not support expansion */
2502 if (!vd->vdev_aux) {
2503 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2504 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2508 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2510 if (!vd->vdev_aux) {
2511 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2512 pvd->vdev_expanding = B_FALSE;
2516 *newstate = vd->vdev_state;
2517 if ((flags & ZFS_ONLINE_UNSPARE) &&
2518 !vdev_is_dead(vd) && vd->vdev_parent &&
2519 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2520 vd->vdev_parent->vdev_child[0] == vd)
2521 vd->vdev_unspare = B_TRUE;
2523 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2525 /* XXX - L2ARC 1.0 does not support expansion */
2527 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2528 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2530 return (spa_vdev_state_exit(spa, vd, 0));
2534 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2538 uint64_t generation;
2539 metaslab_group_t *mg;
2542 spa_vdev_state_enter(spa, SCL_ALLOC);
2544 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2545 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2547 if (!vd->vdev_ops->vdev_op_leaf)
2548 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2552 generation = spa->spa_config_generation + 1;
2555 * If the device isn't already offline, try to offline it.
2557 if (!vd->vdev_offline) {
2559 * If this device has the only valid copy of some data,
2560 * don't allow it to be offlined. Log devices are always
2563 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2564 vdev_dtl_required(vd))
2565 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2568 * If the top-level is a slog and it has had allocations
2569 * then proceed. We check that the vdev's metaslab group
2570 * is not NULL since it's possible that we may have just
2571 * added this vdev but not yet initialized its metaslabs.
2573 if (tvd->vdev_islog && mg != NULL) {
2575 * Prevent any future allocations.
2577 metaslab_group_passivate(mg);
2578 (void) spa_vdev_state_exit(spa, vd, 0);
2580 error = spa_offline_log(spa);
2582 spa_vdev_state_enter(spa, SCL_ALLOC);
2585 * Check to see if the config has changed.
2587 if (error || generation != spa->spa_config_generation) {
2588 metaslab_group_activate(mg);
2590 return (spa_vdev_state_exit(spa,
2592 (void) spa_vdev_state_exit(spa, vd, 0);
2595 ASSERT0(tvd->vdev_stat.vs_alloc);
2599 * Offline this device and reopen its top-level vdev.
2600 * If the top-level vdev is a log device then just offline
2601 * it. Otherwise, if this action results in the top-level
2602 * vdev becoming unusable, undo it and fail the request.
2604 vd->vdev_offline = B_TRUE;
2607 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2608 vdev_is_dead(tvd)) {
2609 vd->vdev_offline = B_FALSE;
2611 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2615 * Add the device back into the metaslab rotor so that
2616 * once we online the device it's open for business.
2618 if (tvd->vdev_islog && mg != NULL)
2619 metaslab_group_activate(mg);
2622 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2624 return (spa_vdev_state_exit(spa, vd, 0));
2628 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2632 mutex_enter(&spa->spa_vdev_top_lock);
2633 error = vdev_offline_locked(spa, guid, flags);
2634 mutex_exit(&spa->spa_vdev_top_lock);
2640 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2641 * vdev_offline(), we assume the spa config is locked. We also clear all
2642 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2645 vdev_clear(spa_t *spa, vdev_t *vd)
2647 vdev_t *rvd = spa->spa_root_vdev;
2649 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2654 vd->vdev_stat.vs_read_errors = 0;
2655 vd->vdev_stat.vs_write_errors = 0;
2656 vd->vdev_stat.vs_checksum_errors = 0;
2658 for (int c = 0; c < vd->vdev_children; c++)
2659 vdev_clear(spa, vd->vdev_child[c]);
2662 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2663 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2665 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2666 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2670 * If we're in the FAULTED state or have experienced failed I/O, then
2671 * clear the persistent state and attempt to reopen the device. We
2672 * also mark the vdev config dirty, so that the new faulted state is
2673 * written out to disk.
2675 if (vd->vdev_faulted || vd->vdev_degraded ||
2676 !vdev_readable(vd) || !vdev_writeable(vd)) {
2679 * When reopening in reponse to a clear event, it may be due to
2680 * a fmadm repair request. In this case, if the device is
2681 * still broken, we want to still post the ereport again.
2683 vd->vdev_forcefault = B_TRUE;
2685 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2686 vd->vdev_cant_read = B_FALSE;
2687 vd->vdev_cant_write = B_FALSE;
2689 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2691 vd->vdev_forcefault = B_FALSE;
2693 if (vd != rvd && vdev_writeable(vd->vdev_top))
2694 vdev_state_dirty(vd->vdev_top);
2696 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2697 spa_async_request(spa, SPA_ASYNC_RESILVER);
2699 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2703 * When clearing a FMA-diagnosed fault, we always want to
2704 * unspare the device, as we assume that the original spare was
2705 * done in response to the FMA fault.
2707 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2708 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2709 vd->vdev_parent->vdev_child[0] == vd)
2710 vd->vdev_unspare = B_TRUE;
2714 vdev_is_dead(vdev_t *vd)
2717 * Holes and missing devices are always considered "dead".
2718 * This simplifies the code since we don't have to check for
2719 * these types of devices in the various code paths.
2720 * Instead we rely on the fact that we skip over dead devices
2721 * before issuing I/O to them.
2723 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2724 vd->vdev_ops == &vdev_missing_ops);
2728 vdev_readable(vdev_t *vd)
2730 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2734 vdev_writeable(vdev_t *vd)
2736 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2740 vdev_allocatable(vdev_t *vd)
2742 uint64_t state = vd->vdev_state;
2745 * We currently allow allocations from vdevs which may be in the
2746 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2747 * fails to reopen then we'll catch it later when we're holding
2748 * the proper locks. Note that we have to get the vdev state
2749 * in a local variable because although it changes atomically,
2750 * we're asking two separate questions about it.
2752 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2753 !vd->vdev_cant_write && !vd->vdev_ishole);
2757 vdev_accessible(vdev_t *vd, zio_t *zio)
2759 ASSERT(zio->io_vd == vd);
2761 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2764 if (zio->io_type == ZIO_TYPE_READ)
2765 return (!vd->vdev_cant_read);
2767 if (zio->io_type == ZIO_TYPE_WRITE)
2768 return (!vd->vdev_cant_write);
2774 * Get statistics for the given vdev.
2777 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2779 spa_t *spa = vd->vdev_spa;
2780 vdev_t *rvd = spa->spa_root_vdev;
2782 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2784 mutex_enter(&vd->vdev_stat_lock);
2785 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2786 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2787 vs->vs_state = vd->vdev_state;
2788 vs->vs_rsize = vdev_get_min_asize(vd);
2789 if (vd->vdev_ops->vdev_op_leaf)
2790 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2791 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2792 vs->vs_configured_ashift = vd->vdev_top != NULL
2793 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2794 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2795 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2796 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2797 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2801 * If we're getting stats on the root vdev, aggregate the I/O counts
2802 * over all top-level vdevs (i.e. the direct children of the root).
2805 for (int c = 0; c < rvd->vdev_children; c++) {
2806 vdev_t *cvd = rvd->vdev_child[c];
2807 vdev_stat_t *cvs = &cvd->vdev_stat;
2809 for (int t = 0; t < ZIO_TYPES; t++) {
2810 vs->vs_ops[t] += cvs->vs_ops[t];
2811 vs->vs_bytes[t] += cvs->vs_bytes[t];
2813 cvs->vs_scan_removing = cvd->vdev_removing;
2816 mutex_exit(&vd->vdev_stat_lock);
2820 vdev_clear_stats(vdev_t *vd)
2822 mutex_enter(&vd->vdev_stat_lock);
2823 vd->vdev_stat.vs_space = 0;
2824 vd->vdev_stat.vs_dspace = 0;
2825 vd->vdev_stat.vs_alloc = 0;
2826 mutex_exit(&vd->vdev_stat_lock);
2830 vdev_scan_stat_init(vdev_t *vd)
2832 vdev_stat_t *vs = &vd->vdev_stat;
2834 for (int c = 0; c < vd->vdev_children; c++)
2835 vdev_scan_stat_init(vd->vdev_child[c]);
2837 mutex_enter(&vd->vdev_stat_lock);
2838 vs->vs_scan_processed = 0;
2839 mutex_exit(&vd->vdev_stat_lock);
2843 vdev_stat_update(zio_t *zio, uint64_t psize)
2845 spa_t *spa = zio->io_spa;
2846 vdev_t *rvd = spa->spa_root_vdev;
2847 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2849 uint64_t txg = zio->io_txg;
2850 vdev_stat_t *vs = &vd->vdev_stat;
2851 zio_type_t type = zio->io_type;
2852 int flags = zio->io_flags;
2855 * If this i/o is a gang leader, it didn't do any actual work.
2857 if (zio->io_gang_tree)
2860 if (zio->io_error == 0) {
2862 * If this is a root i/o, don't count it -- we've already
2863 * counted the top-level vdevs, and vdev_get_stats() will
2864 * aggregate them when asked. This reduces contention on
2865 * the root vdev_stat_lock and implicitly handles blocks
2866 * that compress away to holes, for which there is no i/o.
2867 * (Holes never create vdev children, so all the counters
2868 * remain zero, which is what we want.)
2870 * Note: this only applies to successful i/o (io_error == 0)
2871 * because unlike i/o counts, errors are not additive.
2872 * When reading a ditto block, for example, failure of
2873 * one top-level vdev does not imply a root-level error.
2878 ASSERT(vd == zio->io_vd);
2880 if (flags & ZIO_FLAG_IO_BYPASS)
2883 mutex_enter(&vd->vdev_stat_lock);
2885 if (flags & ZIO_FLAG_IO_REPAIR) {
2886 if (flags & ZIO_FLAG_SCAN_THREAD) {
2887 dsl_scan_phys_t *scn_phys =
2888 &spa->spa_dsl_pool->dp_scan->scn_phys;
2889 uint64_t *processed = &scn_phys->scn_processed;
2892 if (vd->vdev_ops->vdev_op_leaf)
2893 atomic_add_64(processed, psize);
2894 vs->vs_scan_processed += psize;
2897 if (flags & ZIO_FLAG_SELF_HEAL)
2898 vs->vs_self_healed += psize;
2902 vs->vs_bytes[type] += psize;
2904 mutex_exit(&vd->vdev_stat_lock);
2908 if (flags & ZIO_FLAG_SPECULATIVE)
2912 * If this is an I/O error that is going to be retried, then ignore the
2913 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2914 * hard errors, when in reality they can happen for any number of
2915 * innocuous reasons (bus resets, MPxIO link failure, etc).
2917 if (zio->io_error == EIO &&
2918 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2922 * Intent logs writes won't propagate their error to the root
2923 * I/O so don't mark these types of failures as pool-level
2926 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2929 mutex_enter(&vd->vdev_stat_lock);
2930 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2931 if (zio->io_error == ECKSUM)
2932 vs->vs_checksum_errors++;
2934 vs->vs_read_errors++;
2936 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2937 vs->vs_write_errors++;
2938 mutex_exit(&vd->vdev_stat_lock);
2940 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2941 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2942 (flags & ZIO_FLAG_SCAN_THREAD) ||
2943 spa->spa_claiming)) {
2945 * This is either a normal write (not a repair), or it's
2946 * a repair induced by the scrub thread, or it's a repair
2947 * made by zil_claim() during spa_load() in the first txg.
2948 * In the normal case, we commit the DTL change in the same
2949 * txg as the block was born. In the scrub-induced repair
2950 * case, we know that scrubs run in first-pass syncing context,
2951 * so we commit the DTL change in spa_syncing_txg(spa).
2952 * In the zil_claim() case, we commit in spa_first_txg(spa).
2954 * We currently do not make DTL entries for failed spontaneous
2955 * self-healing writes triggered by normal (non-scrubbing)
2956 * reads, because we have no transactional context in which to
2957 * do so -- and it's not clear that it'd be desirable anyway.
2959 if (vd->vdev_ops->vdev_op_leaf) {
2960 uint64_t commit_txg = txg;
2961 if (flags & ZIO_FLAG_SCAN_THREAD) {
2962 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2963 ASSERT(spa_sync_pass(spa) == 1);
2964 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2965 commit_txg = spa_syncing_txg(spa);
2966 } else if (spa->spa_claiming) {
2967 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2968 commit_txg = spa_first_txg(spa);
2970 ASSERT(commit_txg >= spa_syncing_txg(spa));
2971 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2973 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2974 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2975 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2978 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2983 * Update the in-core space usage stats for this vdev, its metaslab class,
2984 * and the root vdev.
2987 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2988 int64_t space_delta)
2990 int64_t dspace_delta = space_delta;
2991 spa_t *spa = vd->vdev_spa;
2992 vdev_t *rvd = spa->spa_root_vdev;
2993 metaslab_group_t *mg = vd->vdev_mg;
2994 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2996 ASSERT(vd == vd->vdev_top);
2999 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3000 * factor. We must calculate this here and not at the root vdev
3001 * because the root vdev's psize-to-asize is simply the max of its
3002 * childrens', thus not accurate enough for us.
3004 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3005 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3006 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3007 vd->vdev_deflate_ratio;
3009 mutex_enter(&vd->vdev_stat_lock);
3010 vd->vdev_stat.vs_alloc += alloc_delta;
3011 vd->vdev_stat.vs_space += space_delta;
3012 vd->vdev_stat.vs_dspace += dspace_delta;
3013 mutex_exit(&vd->vdev_stat_lock);
3015 if (mc == spa_normal_class(spa)) {
3016 mutex_enter(&rvd->vdev_stat_lock);
3017 rvd->vdev_stat.vs_alloc += alloc_delta;
3018 rvd->vdev_stat.vs_space += space_delta;
3019 rvd->vdev_stat.vs_dspace += dspace_delta;
3020 mutex_exit(&rvd->vdev_stat_lock);
3024 ASSERT(rvd == vd->vdev_parent);
3025 ASSERT(vd->vdev_ms_count != 0);
3027 metaslab_class_space_update(mc,
3028 alloc_delta, defer_delta, space_delta, dspace_delta);
3033 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3034 * so that it will be written out next time the vdev configuration is synced.
3035 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3038 vdev_config_dirty(vdev_t *vd)
3040 spa_t *spa = vd->vdev_spa;
3041 vdev_t *rvd = spa->spa_root_vdev;
3044 ASSERT(spa_writeable(spa));
3047 * If this is an aux vdev (as with l2cache and spare devices), then we
3048 * update the vdev config manually and set the sync flag.
3050 if (vd->vdev_aux != NULL) {
3051 spa_aux_vdev_t *sav = vd->vdev_aux;
3055 for (c = 0; c < sav->sav_count; c++) {
3056 if (sav->sav_vdevs[c] == vd)
3060 if (c == sav->sav_count) {
3062 * We're being removed. There's nothing more to do.
3064 ASSERT(sav->sav_sync == B_TRUE);
3068 sav->sav_sync = B_TRUE;
3070 if (nvlist_lookup_nvlist_array(sav->sav_config,
3071 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3072 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3073 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3079 * Setting the nvlist in the middle if the array is a little
3080 * sketchy, but it will work.
3082 nvlist_free(aux[c]);
3083 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3089 * The dirty list is protected by the SCL_CONFIG lock. The caller
3090 * must either hold SCL_CONFIG as writer, or must be the sync thread
3091 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3092 * so this is sufficient to ensure mutual exclusion.
3094 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3095 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3096 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3099 for (c = 0; c < rvd->vdev_children; c++)
3100 vdev_config_dirty(rvd->vdev_child[c]);
3102 ASSERT(vd == vd->vdev_top);
3104 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3106 list_insert_head(&spa->spa_config_dirty_list, vd);
3111 vdev_config_clean(vdev_t *vd)
3113 spa_t *spa = vd->vdev_spa;
3115 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3116 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3117 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3119 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3120 list_remove(&spa->spa_config_dirty_list, vd);
3124 * Mark a top-level vdev's state as dirty, so that the next pass of
3125 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3126 * the state changes from larger config changes because they require
3127 * much less locking, and are often needed for administrative actions.
3130 vdev_state_dirty(vdev_t *vd)
3132 spa_t *spa = vd->vdev_spa;
3134 ASSERT(spa_writeable(spa));
3135 ASSERT(vd == vd->vdev_top);
3138 * The state list is protected by the SCL_STATE lock. The caller
3139 * must either hold SCL_STATE as writer, or must be the sync thread
3140 * (which holds SCL_STATE as reader). There's only one sync thread,
3141 * so this is sufficient to ensure mutual exclusion.
3143 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3144 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3145 spa_config_held(spa, SCL_STATE, RW_READER)));
3147 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3148 list_insert_head(&spa->spa_state_dirty_list, vd);
3152 vdev_state_clean(vdev_t *vd)
3154 spa_t *spa = vd->vdev_spa;
3156 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3157 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3158 spa_config_held(spa, SCL_STATE, RW_READER)));
3160 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3161 list_remove(&spa->spa_state_dirty_list, vd);
3165 * Propagate vdev state up from children to parent.
3168 vdev_propagate_state(vdev_t *vd)
3170 spa_t *spa = vd->vdev_spa;
3171 vdev_t *rvd = spa->spa_root_vdev;
3172 int degraded = 0, faulted = 0;
3176 if (vd->vdev_children > 0) {
3177 for (int c = 0; c < vd->vdev_children; c++) {
3178 child = vd->vdev_child[c];
3181 * Don't factor holes into the decision.
3183 if (child->vdev_ishole)
3186 if (!vdev_readable(child) ||
3187 (!vdev_writeable(child) && spa_writeable(spa))) {
3189 * Root special: if there is a top-level log
3190 * device, treat the root vdev as if it were
3193 if (child->vdev_islog && vd == rvd)
3197 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3201 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3205 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3208 * Root special: if there is a top-level vdev that cannot be
3209 * opened due to corrupted metadata, then propagate the root
3210 * vdev's aux state as 'corrupt' rather than 'insufficient
3213 if (corrupted && vd == rvd &&
3214 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3215 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3216 VDEV_AUX_CORRUPT_DATA);
3219 if (vd->vdev_parent)
3220 vdev_propagate_state(vd->vdev_parent);
3224 * Set a vdev's state. If this is during an open, we don't update the parent
3225 * state, because we're in the process of opening children depth-first.
3226 * Otherwise, we propagate the change to the parent.
3228 * If this routine places a device in a faulted state, an appropriate ereport is
3232 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3234 uint64_t save_state;
3235 spa_t *spa = vd->vdev_spa;
3237 if (state == vd->vdev_state) {
3238 vd->vdev_stat.vs_aux = aux;
3242 save_state = vd->vdev_state;
3244 vd->vdev_state = state;
3245 vd->vdev_stat.vs_aux = aux;
3248 * If we are setting the vdev state to anything but an open state, then
3249 * always close the underlying device unless the device has requested
3250 * a delayed close (i.e. we're about to remove or fault the device).
3251 * Otherwise, we keep accessible but invalid devices open forever.
3252 * We don't call vdev_close() itself, because that implies some extra
3253 * checks (offline, etc) that we don't want here. This is limited to
3254 * leaf devices, because otherwise closing the device will affect other
3257 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3258 vd->vdev_ops->vdev_op_leaf)
3259 vd->vdev_ops->vdev_op_close(vd);
3262 * If we have brought this vdev back into service, we need
3263 * to notify fmd so that it can gracefully repair any outstanding
3264 * cases due to a missing device. We do this in all cases, even those
3265 * that probably don't correlate to a repaired fault. This is sure to
3266 * catch all cases, and we let the zfs-retire agent sort it out. If
3267 * this is a transient state it's OK, as the retire agent will
3268 * double-check the state of the vdev before repairing it.
3270 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3271 vd->vdev_prevstate != state)
3272 zfs_post_state_change(spa, vd);
3274 if (vd->vdev_removed &&
3275 state == VDEV_STATE_CANT_OPEN &&
3276 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3278 * If the previous state is set to VDEV_STATE_REMOVED, then this
3279 * device was previously marked removed and someone attempted to
3280 * reopen it. If this failed due to a nonexistent device, then
3281 * keep the device in the REMOVED state. We also let this be if
3282 * it is one of our special test online cases, which is only
3283 * attempting to online the device and shouldn't generate an FMA
3286 vd->vdev_state = VDEV_STATE_REMOVED;
3287 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3288 } else if (state == VDEV_STATE_REMOVED) {
3289 vd->vdev_removed = B_TRUE;
3290 } else if (state == VDEV_STATE_CANT_OPEN) {
3292 * If we fail to open a vdev during an import or recovery, we
3293 * mark it as "not available", which signifies that it was
3294 * never there to begin with. Failure to open such a device
3295 * is not considered an error.
3297 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3298 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3299 vd->vdev_ops->vdev_op_leaf)
3300 vd->vdev_not_present = 1;
3303 * Post the appropriate ereport. If the 'prevstate' field is
3304 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3305 * that this is part of a vdev_reopen(). In this case, we don't
3306 * want to post the ereport if the device was already in the
3307 * CANT_OPEN state beforehand.
3309 * If the 'checkremove' flag is set, then this is an attempt to
3310 * online the device in response to an insertion event. If we
3311 * hit this case, then we have detected an insertion event for a
3312 * faulted or offline device that wasn't in the removed state.
3313 * In this scenario, we don't post an ereport because we are
3314 * about to replace the device, or attempt an online with
3315 * vdev_forcefault, which will generate the fault for us.
3317 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3318 !vd->vdev_not_present && !vd->vdev_checkremove &&
3319 vd != spa->spa_root_vdev) {
3323 case VDEV_AUX_OPEN_FAILED:
3324 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3326 case VDEV_AUX_CORRUPT_DATA:
3327 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3329 case VDEV_AUX_NO_REPLICAS:
3330 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3332 case VDEV_AUX_BAD_GUID_SUM:
3333 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3335 case VDEV_AUX_TOO_SMALL:
3336 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3338 case VDEV_AUX_BAD_LABEL:
3339 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3342 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3345 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3348 /* Erase any notion of persistent removed state */
3349 vd->vdev_removed = B_FALSE;
3351 vd->vdev_removed = B_FALSE;
3354 if (!isopen && vd->vdev_parent)
3355 vdev_propagate_state(vd->vdev_parent);
3359 * Check the vdev configuration to ensure that it's capable of supporting
3362 * On Solaris, we do not support RAID-Z or partial configuration. In
3363 * addition, only a single top-level vdev is allowed and none of the
3364 * leaves can be wholedisks.
3366 * For FreeBSD, we can boot from any configuration. There is a
3367 * limitation that the boot filesystem must be either uncompressed or
3368 * compresses with lzjb compression but I'm not sure how to enforce
3372 vdev_is_bootable(vdev_t *vd)
3375 if (!vd->vdev_ops->vdev_op_leaf) {
3376 char *vdev_type = vd->vdev_ops->vdev_op_type;
3378 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3379 vd->vdev_children > 1) {
3381 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3382 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3385 } else if (vd->vdev_wholedisk == 1) {
3389 for (int c = 0; c < vd->vdev_children; c++) {
3390 if (!vdev_is_bootable(vd->vdev_child[c]))
3398 * Load the state from the original vdev tree (ovd) which
3399 * we've retrieved from the MOS config object. If the original
3400 * vdev was offline or faulted then we transfer that state to the
3401 * device in the current vdev tree (nvd).
3404 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3406 spa_t *spa = nvd->vdev_spa;
3408 ASSERT(nvd->vdev_top->vdev_islog);
3409 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3410 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3412 for (int c = 0; c < nvd->vdev_children; c++)
3413 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3415 if (nvd->vdev_ops->vdev_op_leaf) {
3417 * Restore the persistent vdev state
3419 nvd->vdev_offline = ovd->vdev_offline;
3420 nvd->vdev_faulted = ovd->vdev_faulted;
3421 nvd->vdev_degraded = ovd->vdev_degraded;
3422 nvd->vdev_removed = ovd->vdev_removed;
3427 * Determine if a log device has valid content. If the vdev was
3428 * removed or faulted in the MOS config then we know that
3429 * the content on the log device has already been written to the pool.
3432 vdev_log_state_valid(vdev_t *vd)
3434 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3438 for (int c = 0; c < vd->vdev_children; c++)
3439 if (vdev_log_state_valid(vd->vdev_child[c]))
3446 * Expand a vdev if possible.
3449 vdev_expand(vdev_t *vd, uint64_t txg)
3451 ASSERT(vd->vdev_top == vd);
3452 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3454 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3455 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3456 vdev_config_dirty(vd);
3464 vdev_split(vdev_t *vd)
3466 vdev_t *cvd, *pvd = vd->vdev_parent;
3468 vdev_remove_child(pvd, vd);
3469 vdev_compact_children(pvd);
3471 cvd = pvd->vdev_child[0];
3472 if (pvd->vdev_children == 1) {
3473 vdev_remove_parent(cvd);
3474 cvd->vdev_splitting = B_TRUE;
3476 vdev_propagate_state(cvd);
3480 vdev_deadman(vdev_t *vd)
3482 for (int c = 0; c < vd->vdev_children; c++) {
3483 vdev_t *cvd = vd->vdev_child[c];
3488 if (vd->vdev_ops->vdev_op_leaf) {
3489 vdev_queue_t *vq = &vd->vdev_queue;
3491 mutex_enter(&vq->vq_lock);
3492 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3493 spa_t *spa = vd->vdev_spa;
3498 * Look at the head of all the pending queues,
3499 * if any I/O has been outstanding for longer than
3500 * the spa_deadman_synctime we panic the system.
3502 fio = avl_first(&vq->vq_active_tree);
3503 delta = gethrtime() - fio->io_timestamp;
3504 if (delta > spa_deadman_synctime(spa)) {
3505 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3506 "delta %lluns, last io %lluns",
3507 fio->io_timestamp, delta,
3508 vq->vq_io_complete_ts);
3509 fm_panic("I/O to pool '%s' appears to be "
3510 "hung on vdev guid %llu at '%s'.",
3512 (long long unsigned int) vd->vdev_guid,
3516 mutex_exit(&vq->vq_lock);