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 * Given a vdev type, return the appropriate ops vector.
162 vdev_getops(const char *type)
164 vdev_ops_t *ops, **opspp;
166 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
167 if (strcmp(ops->vdev_op_type, type) == 0)
174 * Default asize function: return the MAX of psize with the asize of
175 * all children. This is what's used by anything other than RAID-Z.
178 vdev_default_asize(vdev_t *vd, uint64_t psize)
180 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
183 for (int c = 0; c < vd->vdev_children; c++) {
184 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
185 asize = MAX(asize, csize);
192 * Get the minimum allocatable size. We define the allocatable size as
193 * the vdev's asize rounded to the nearest metaslab. This allows us to
194 * replace or attach devices which don't have the same physical size but
195 * can still satisfy the same number of allocations.
198 vdev_get_min_asize(vdev_t *vd)
200 vdev_t *pvd = vd->vdev_parent;
203 * If our parent is NULL (inactive spare or cache) or is the root,
204 * just return our own asize.
207 return (vd->vdev_asize);
210 * The top-level vdev just returns the allocatable size rounded
211 * to the nearest metaslab.
213 if (vd == vd->vdev_top)
214 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
217 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
218 * so each child must provide at least 1/Nth of its asize.
220 if (pvd->vdev_ops == &vdev_raidz_ops)
221 return (pvd->vdev_min_asize / pvd->vdev_children);
223 return (pvd->vdev_min_asize);
227 vdev_set_min_asize(vdev_t *vd)
229 vd->vdev_min_asize = vdev_get_min_asize(vd);
231 for (int c = 0; c < vd->vdev_children; c++)
232 vdev_set_min_asize(vd->vdev_child[c]);
236 vdev_lookup_top(spa_t *spa, uint64_t vdev)
238 vdev_t *rvd = spa->spa_root_vdev;
240 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
242 if (vdev < rvd->vdev_children) {
243 ASSERT(rvd->vdev_child[vdev] != NULL);
244 return (rvd->vdev_child[vdev]);
251 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
255 if (vd->vdev_guid == guid)
258 for (int c = 0; c < vd->vdev_children; c++)
259 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
267 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
269 size_t oldsize, newsize;
270 uint64_t id = cvd->vdev_id;
273 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
274 ASSERT(cvd->vdev_parent == NULL);
276 cvd->vdev_parent = pvd;
281 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
283 oldsize = pvd->vdev_children * sizeof (vdev_t *);
284 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
285 newsize = pvd->vdev_children * sizeof (vdev_t *);
287 newchild = kmem_zalloc(newsize, KM_SLEEP);
288 if (pvd->vdev_child != NULL) {
289 bcopy(pvd->vdev_child, newchild, oldsize);
290 kmem_free(pvd->vdev_child, oldsize);
293 pvd->vdev_child = newchild;
294 pvd->vdev_child[id] = cvd;
296 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
297 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
300 * Walk up all ancestors to update guid sum.
302 for (; pvd != NULL; pvd = pvd->vdev_parent)
303 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
307 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
310 uint_t id = cvd->vdev_id;
312 ASSERT(cvd->vdev_parent == pvd);
317 ASSERT(id < pvd->vdev_children);
318 ASSERT(pvd->vdev_child[id] == cvd);
320 pvd->vdev_child[id] = NULL;
321 cvd->vdev_parent = NULL;
323 for (c = 0; c < pvd->vdev_children; c++)
324 if (pvd->vdev_child[c])
327 if (c == pvd->vdev_children) {
328 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
329 pvd->vdev_child = NULL;
330 pvd->vdev_children = 0;
334 * Walk up all ancestors to update guid sum.
336 for (; pvd != NULL; pvd = pvd->vdev_parent)
337 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
341 * Remove any holes in the child array.
344 vdev_compact_children(vdev_t *pvd)
346 vdev_t **newchild, *cvd;
347 int oldc = pvd->vdev_children;
350 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
352 for (int c = newc = 0; c < oldc; c++)
353 if (pvd->vdev_child[c])
356 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
358 for (int c = newc = 0; c < oldc; c++) {
359 if ((cvd = pvd->vdev_child[c]) != NULL) {
360 newchild[newc] = cvd;
361 cvd->vdev_id = newc++;
365 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
366 pvd->vdev_child = newchild;
367 pvd->vdev_children = newc;
371 * Allocate and minimally initialize a vdev_t.
374 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
378 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
380 if (spa->spa_root_vdev == NULL) {
381 ASSERT(ops == &vdev_root_ops);
382 spa->spa_root_vdev = vd;
383 spa->spa_load_guid = spa_generate_guid(NULL);
386 if (guid == 0 && ops != &vdev_hole_ops) {
387 if (spa->spa_root_vdev == vd) {
389 * The root vdev's guid will also be the pool guid,
390 * which must be unique among all pools.
392 guid = spa_generate_guid(NULL);
395 * Any other vdev's guid must be unique within the pool.
397 guid = spa_generate_guid(spa);
399 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
404 vd->vdev_guid = guid;
405 vd->vdev_guid_sum = guid;
407 vd->vdev_state = VDEV_STATE_CLOSED;
408 vd->vdev_ishole = (ops == &vdev_hole_ops);
410 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
411 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
412 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
413 for (int t = 0; t < DTL_TYPES; t++) {
414 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
417 txg_list_create(&vd->vdev_ms_list,
418 offsetof(struct metaslab, ms_txg_node));
419 txg_list_create(&vd->vdev_dtl_list,
420 offsetof(struct vdev, vdev_dtl_node));
421 vd->vdev_stat.vs_timestamp = gethrtime();
429 * Allocate a new vdev. The 'alloctype' is used to control whether we are
430 * creating a new vdev or loading an existing one - the behavior is slightly
431 * different for each case.
434 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
439 uint64_t guid = 0, islog, nparity;
442 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
445 return (SET_ERROR(EINVAL));
447 if ((ops = vdev_getops(type)) == NULL)
448 return (SET_ERROR(EINVAL));
451 * If this is a load, get the vdev guid from the nvlist.
452 * Otherwise, vdev_alloc_common() will generate one for us.
454 if (alloctype == VDEV_ALLOC_LOAD) {
457 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
459 return (SET_ERROR(EINVAL));
461 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
462 return (SET_ERROR(EINVAL));
463 } else if (alloctype == VDEV_ALLOC_SPARE) {
464 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
465 return (SET_ERROR(EINVAL));
466 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
467 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
468 return (SET_ERROR(EINVAL));
469 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
470 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
471 return (SET_ERROR(EINVAL));
475 * The first allocated vdev must be of type 'root'.
477 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
478 return (SET_ERROR(EINVAL));
481 * Determine whether we're a log vdev.
484 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
485 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
486 return (SET_ERROR(ENOTSUP));
488 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
489 return (SET_ERROR(ENOTSUP));
492 * Set the nparity property for RAID-Z vdevs.
495 if (ops == &vdev_raidz_ops) {
496 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
498 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
499 return (SET_ERROR(EINVAL));
501 * Previous versions could only support 1 or 2 parity
505 spa_version(spa) < SPA_VERSION_RAIDZ2)
506 return (SET_ERROR(ENOTSUP));
508 spa_version(spa) < SPA_VERSION_RAIDZ3)
509 return (SET_ERROR(ENOTSUP));
512 * We require the parity to be specified for SPAs that
513 * support multiple parity levels.
515 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
516 return (SET_ERROR(EINVAL));
518 * Otherwise, we default to 1 parity device for RAID-Z.
525 ASSERT(nparity != -1ULL);
527 vd = vdev_alloc_common(spa, id, guid, ops);
529 vd->vdev_islog = islog;
530 vd->vdev_nparity = nparity;
532 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
533 vd->vdev_path = spa_strdup(vd->vdev_path);
534 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
535 vd->vdev_devid = spa_strdup(vd->vdev_devid);
536 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
537 &vd->vdev_physpath) == 0)
538 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
539 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
540 vd->vdev_fru = spa_strdup(vd->vdev_fru);
543 * Set the whole_disk property. If it's not specified, leave the value
546 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
547 &vd->vdev_wholedisk) != 0)
548 vd->vdev_wholedisk = -1ULL;
551 * Look for the 'not present' flag. This will only be set if the device
552 * was not present at the time of import.
554 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
555 &vd->vdev_not_present);
558 * Get the alignment requirement.
560 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
563 * Retrieve the vdev creation time.
565 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
569 * If we're a top-level vdev, try to load the allocation parameters.
571 if (parent && !parent->vdev_parent &&
572 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
573 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
575 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
577 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
579 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
583 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
584 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
585 alloctype == VDEV_ALLOC_ADD ||
586 alloctype == VDEV_ALLOC_SPLIT ||
587 alloctype == VDEV_ALLOC_ROOTPOOL);
588 vd->vdev_mg = metaslab_group_create(islog ?
589 spa_log_class(spa) : spa_normal_class(spa), vd);
593 * If we're a leaf vdev, try to load the DTL object and other state.
595 if (vd->vdev_ops->vdev_op_leaf &&
596 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
597 alloctype == VDEV_ALLOC_ROOTPOOL)) {
598 if (alloctype == VDEV_ALLOC_LOAD) {
599 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
600 &vd->vdev_dtl_object);
601 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
605 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
608 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
609 &spare) == 0 && spare)
613 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
616 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
617 &vd->vdev_resilver_txg);
620 * When importing a pool, we want to ignore the persistent fault
621 * state, as the diagnosis made on another system may not be
622 * valid in the current context. Local vdevs will
623 * remain in the faulted state.
625 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
626 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
628 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
630 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
633 if (vd->vdev_faulted || vd->vdev_degraded) {
637 VDEV_AUX_ERR_EXCEEDED;
638 if (nvlist_lookup_string(nv,
639 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
640 strcmp(aux, "external") == 0)
641 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
647 * Add ourselves to the parent's list of children.
649 vdev_add_child(parent, vd);
657 vdev_free(vdev_t *vd)
659 spa_t *spa = vd->vdev_spa;
662 * vdev_free() implies closing the vdev first. This is simpler than
663 * trying to ensure complicated semantics for all callers.
667 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
668 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
673 for (int c = 0; c < vd->vdev_children; c++)
674 vdev_free(vd->vdev_child[c]);
676 ASSERT(vd->vdev_child == NULL);
677 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
680 * Discard allocation state.
682 if (vd->vdev_mg != NULL) {
683 vdev_metaslab_fini(vd);
684 metaslab_group_destroy(vd->vdev_mg);
687 ASSERT0(vd->vdev_stat.vs_space);
688 ASSERT0(vd->vdev_stat.vs_dspace);
689 ASSERT0(vd->vdev_stat.vs_alloc);
692 * Remove this vdev from its parent's child list.
694 vdev_remove_child(vd->vdev_parent, vd);
696 ASSERT(vd->vdev_parent == NULL);
699 * Clean up vdev structure.
705 spa_strfree(vd->vdev_path);
707 spa_strfree(vd->vdev_devid);
708 if (vd->vdev_physpath)
709 spa_strfree(vd->vdev_physpath);
711 spa_strfree(vd->vdev_fru);
713 if (vd->vdev_isspare)
714 spa_spare_remove(vd);
715 if (vd->vdev_isl2cache)
716 spa_l2cache_remove(vd);
718 txg_list_destroy(&vd->vdev_ms_list);
719 txg_list_destroy(&vd->vdev_dtl_list);
721 mutex_enter(&vd->vdev_dtl_lock);
722 space_map_close(vd->vdev_dtl_sm);
723 for (int t = 0; t < DTL_TYPES; t++) {
724 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
725 range_tree_destroy(vd->vdev_dtl[t]);
727 mutex_exit(&vd->vdev_dtl_lock);
729 mutex_destroy(&vd->vdev_dtl_lock);
730 mutex_destroy(&vd->vdev_stat_lock);
731 mutex_destroy(&vd->vdev_probe_lock);
733 if (vd == spa->spa_root_vdev)
734 spa->spa_root_vdev = NULL;
736 kmem_free(vd, sizeof (vdev_t));
740 * Transfer top-level vdev state from svd to tvd.
743 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
745 spa_t *spa = svd->vdev_spa;
750 ASSERT(tvd == tvd->vdev_top);
752 tvd->vdev_ms_array = svd->vdev_ms_array;
753 tvd->vdev_ms_shift = svd->vdev_ms_shift;
754 tvd->vdev_ms_count = svd->vdev_ms_count;
756 svd->vdev_ms_array = 0;
757 svd->vdev_ms_shift = 0;
758 svd->vdev_ms_count = 0;
761 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
762 tvd->vdev_mg = svd->vdev_mg;
763 tvd->vdev_ms = svd->vdev_ms;
768 if (tvd->vdev_mg != NULL)
769 tvd->vdev_mg->mg_vd = tvd;
771 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
772 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
773 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
775 svd->vdev_stat.vs_alloc = 0;
776 svd->vdev_stat.vs_space = 0;
777 svd->vdev_stat.vs_dspace = 0;
779 for (t = 0; t < TXG_SIZE; t++) {
780 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
781 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
782 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
783 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
784 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
785 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
788 if (list_link_active(&svd->vdev_config_dirty_node)) {
789 vdev_config_clean(svd);
790 vdev_config_dirty(tvd);
793 if (list_link_active(&svd->vdev_state_dirty_node)) {
794 vdev_state_clean(svd);
795 vdev_state_dirty(tvd);
798 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
799 svd->vdev_deflate_ratio = 0;
801 tvd->vdev_islog = svd->vdev_islog;
806 vdev_top_update(vdev_t *tvd, vdev_t *vd)
813 for (int c = 0; c < vd->vdev_children; c++)
814 vdev_top_update(tvd, vd->vdev_child[c]);
818 * Add a mirror/replacing vdev above an existing vdev.
821 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
823 spa_t *spa = cvd->vdev_spa;
824 vdev_t *pvd = cvd->vdev_parent;
827 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
829 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
831 mvd->vdev_asize = cvd->vdev_asize;
832 mvd->vdev_min_asize = cvd->vdev_min_asize;
833 mvd->vdev_max_asize = cvd->vdev_max_asize;
834 mvd->vdev_ashift = cvd->vdev_ashift;
835 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
836 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
837 mvd->vdev_state = cvd->vdev_state;
838 mvd->vdev_crtxg = cvd->vdev_crtxg;
840 vdev_remove_child(pvd, cvd);
841 vdev_add_child(pvd, mvd);
842 cvd->vdev_id = mvd->vdev_children;
843 vdev_add_child(mvd, cvd);
844 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
846 if (mvd == mvd->vdev_top)
847 vdev_top_transfer(cvd, mvd);
853 * Remove a 1-way mirror/replacing vdev from the tree.
856 vdev_remove_parent(vdev_t *cvd)
858 vdev_t *mvd = cvd->vdev_parent;
859 vdev_t *pvd = mvd->vdev_parent;
861 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
863 ASSERT(mvd->vdev_children == 1);
864 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
865 mvd->vdev_ops == &vdev_replacing_ops ||
866 mvd->vdev_ops == &vdev_spare_ops);
867 cvd->vdev_ashift = mvd->vdev_ashift;
868 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
869 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
871 vdev_remove_child(mvd, cvd);
872 vdev_remove_child(pvd, mvd);
875 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
876 * Otherwise, we could have detached an offline device, and when we
877 * go to import the pool we'll think we have two top-level vdevs,
878 * instead of a different version of the same top-level vdev.
880 if (mvd->vdev_top == mvd) {
881 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
882 cvd->vdev_orig_guid = cvd->vdev_guid;
883 cvd->vdev_guid += guid_delta;
884 cvd->vdev_guid_sum += guid_delta;
886 cvd->vdev_id = mvd->vdev_id;
887 vdev_add_child(pvd, cvd);
888 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
890 if (cvd == cvd->vdev_top)
891 vdev_top_transfer(mvd, cvd);
893 ASSERT(mvd->vdev_children == 0);
898 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
900 spa_t *spa = vd->vdev_spa;
901 objset_t *mos = spa->spa_meta_objset;
903 uint64_t oldc = vd->vdev_ms_count;
904 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
908 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
911 * This vdev is not being allocated from yet or is a hole.
913 if (vd->vdev_ms_shift == 0)
916 ASSERT(!vd->vdev_ishole);
919 * Compute the raidz-deflation ratio. Note, we hard-code
920 * in 128k (1 << 17) because it is the current "typical" blocksize.
921 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
922 * or we will inconsistently account for existing bp's.
924 vd->vdev_deflate_ratio = (1 << 17) /
925 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
927 ASSERT(oldc <= newc);
929 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
932 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
933 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
937 vd->vdev_ms_count = newc;
939 for (m = oldc; m < newc; m++) {
943 error = dmu_read(mos, vd->vdev_ms_array,
944 m * sizeof (uint64_t), sizeof (uint64_t), &object,
949 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg);
953 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
956 * If the vdev is being removed we don't activate
957 * the metaslabs since we want to ensure that no new
958 * allocations are performed on this device.
960 if (oldc == 0 && !vd->vdev_removing)
961 metaslab_group_activate(vd->vdev_mg);
964 spa_config_exit(spa, SCL_ALLOC, FTAG);
970 vdev_metaslab_fini(vdev_t *vd)
973 uint64_t count = vd->vdev_ms_count;
975 if (vd->vdev_ms != NULL) {
976 metaslab_group_passivate(vd->vdev_mg);
977 for (m = 0; m < count; m++) {
978 metaslab_t *msp = vd->vdev_ms[m];
983 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
988 typedef struct vdev_probe_stats {
989 boolean_t vps_readable;
990 boolean_t vps_writeable;
992 } vdev_probe_stats_t;
995 vdev_probe_done(zio_t *zio)
997 spa_t *spa = zio->io_spa;
998 vdev_t *vd = zio->io_vd;
999 vdev_probe_stats_t *vps = zio->io_private;
1001 ASSERT(vd->vdev_probe_zio != NULL);
1003 if (zio->io_type == ZIO_TYPE_READ) {
1004 if (zio->io_error == 0)
1005 vps->vps_readable = 1;
1006 if (zio->io_error == 0 && spa_writeable(spa)) {
1007 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1008 zio->io_offset, zio->io_size, zio->io_data,
1009 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1010 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1012 zio_buf_free(zio->io_data, zio->io_size);
1014 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1015 if (zio->io_error == 0)
1016 vps->vps_writeable = 1;
1017 zio_buf_free(zio->io_data, zio->io_size);
1018 } else if (zio->io_type == ZIO_TYPE_NULL) {
1021 vd->vdev_cant_read |= !vps->vps_readable;
1022 vd->vdev_cant_write |= !vps->vps_writeable;
1024 if (vdev_readable(vd) &&
1025 (vdev_writeable(vd) || !spa_writeable(spa))) {
1028 ASSERT(zio->io_error != 0);
1029 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1030 spa, vd, NULL, 0, 0);
1031 zio->io_error = SET_ERROR(ENXIO);
1034 mutex_enter(&vd->vdev_probe_lock);
1035 ASSERT(vd->vdev_probe_zio == zio);
1036 vd->vdev_probe_zio = NULL;
1037 mutex_exit(&vd->vdev_probe_lock);
1039 while ((pio = zio_walk_parents(zio)) != NULL)
1040 if (!vdev_accessible(vd, pio))
1041 pio->io_error = SET_ERROR(ENXIO);
1043 kmem_free(vps, sizeof (*vps));
1048 * Determine whether this device is accessible.
1050 * Read and write to several known locations: the pad regions of each
1051 * vdev label but the first, which we leave alone in case it contains
1055 vdev_probe(vdev_t *vd, zio_t *zio)
1057 spa_t *spa = vd->vdev_spa;
1058 vdev_probe_stats_t *vps = NULL;
1061 ASSERT(vd->vdev_ops->vdev_op_leaf);
1064 * Don't probe the probe.
1066 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1070 * To prevent 'probe storms' when a device fails, we create
1071 * just one probe i/o at a time. All zios that want to probe
1072 * this vdev will become parents of the probe io.
1074 mutex_enter(&vd->vdev_probe_lock);
1076 if ((pio = vd->vdev_probe_zio) == NULL) {
1077 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1079 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1080 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1083 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1085 * vdev_cant_read and vdev_cant_write can only
1086 * transition from TRUE to FALSE when we have the
1087 * SCL_ZIO lock as writer; otherwise they can only
1088 * transition from FALSE to TRUE. This ensures that
1089 * any zio looking at these values can assume that
1090 * failures persist for the life of the I/O. That's
1091 * important because when a device has intermittent
1092 * connectivity problems, we want to ensure that
1093 * they're ascribed to the device (ENXIO) and not
1096 * Since we hold SCL_ZIO as writer here, clear both
1097 * values so the probe can reevaluate from first
1100 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1101 vd->vdev_cant_read = B_FALSE;
1102 vd->vdev_cant_write = B_FALSE;
1105 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1106 vdev_probe_done, vps,
1107 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1110 * We can't change the vdev state in this context, so we
1111 * kick off an async task to do it on our behalf.
1114 vd->vdev_probe_wanted = B_TRUE;
1115 spa_async_request(spa, SPA_ASYNC_PROBE);
1120 zio_add_child(zio, pio);
1122 mutex_exit(&vd->vdev_probe_lock);
1125 ASSERT(zio != NULL);
1129 for (int l = 1; l < VDEV_LABELS; l++) {
1130 zio_nowait(zio_read_phys(pio, vd,
1131 vdev_label_offset(vd->vdev_psize, l,
1132 offsetof(vdev_label_t, vl_pad2)),
1133 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1134 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1135 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1146 vdev_open_child(void *arg)
1150 vd->vdev_open_thread = curthread;
1151 vd->vdev_open_error = vdev_open(vd);
1152 vd->vdev_open_thread = NULL;
1156 vdev_uses_zvols(vdev_t *vd)
1158 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1159 strlen(ZVOL_DIR)) == 0)
1161 for (int c = 0; c < vd->vdev_children; c++)
1162 if (vdev_uses_zvols(vd->vdev_child[c]))
1168 vdev_open_children(vdev_t *vd)
1171 int children = vd->vdev_children;
1174 * in order to handle pools on top of zvols, do the opens
1175 * in a single thread so that the same thread holds the
1176 * spa_namespace_lock
1178 if (B_TRUE || vdev_uses_zvols(vd)) {
1179 for (int c = 0; c < children; c++)
1180 vd->vdev_child[c]->vdev_open_error =
1181 vdev_open(vd->vdev_child[c]);
1184 tq = taskq_create("vdev_open", children, minclsyspri,
1185 children, children, TASKQ_PREPOPULATE);
1187 for (int c = 0; c < children; c++)
1188 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1195 * Prepare a virtual device for access.
1198 vdev_open(vdev_t *vd)
1200 spa_t *spa = vd->vdev_spa;
1203 uint64_t max_osize = 0;
1204 uint64_t asize, max_asize, psize;
1205 uint64_t logical_ashift = 0;
1206 uint64_t physical_ashift = 0;
1208 ASSERT(vd->vdev_open_thread == curthread ||
1209 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1210 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1211 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1212 vd->vdev_state == VDEV_STATE_OFFLINE);
1214 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1215 vd->vdev_cant_read = B_FALSE;
1216 vd->vdev_cant_write = B_FALSE;
1217 vd->vdev_notrim = B_FALSE;
1218 vd->vdev_min_asize = vdev_get_min_asize(vd);
1221 * If this vdev is not removed, check its fault status. If it's
1222 * faulted, bail out of the open.
1224 if (!vd->vdev_removed && vd->vdev_faulted) {
1225 ASSERT(vd->vdev_children == 0);
1226 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1227 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1228 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1229 vd->vdev_label_aux);
1230 return (SET_ERROR(ENXIO));
1231 } else if (vd->vdev_offline) {
1232 ASSERT(vd->vdev_children == 0);
1233 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1234 return (SET_ERROR(ENXIO));
1237 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1238 &logical_ashift, &physical_ashift);
1241 * Reset the vdev_reopening flag so that we actually close
1242 * the vdev on error.
1244 vd->vdev_reopening = B_FALSE;
1245 if (zio_injection_enabled && error == 0)
1246 error = zio_handle_device_injection(vd, NULL, ENXIO);
1249 if (vd->vdev_removed &&
1250 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1251 vd->vdev_removed = B_FALSE;
1253 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1254 vd->vdev_stat.vs_aux);
1258 vd->vdev_removed = B_FALSE;
1261 * Recheck the faulted flag now that we have confirmed that
1262 * the vdev is accessible. If we're faulted, bail.
1264 if (vd->vdev_faulted) {
1265 ASSERT(vd->vdev_children == 0);
1266 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1267 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1268 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1269 vd->vdev_label_aux);
1270 return (SET_ERROR(ENXIO));
1273 if (vd->vdev_degraded) {
1274 ASSERT(vd->vdev_children == 0);
1275 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1276 VDEV_AUX_ERR_EXCEEDED);
1278 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1282 * For hole or missing vdevs we just return success.
1284 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1287 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1288 trim_map_create(vd);
1290 for (int c = 0; c < vd->vdev_children; c++) {
1291 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1292 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1298 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1299 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1301 if (vd->vdev_children == 0) {
1302 if (osize < SPA_MINDEVSIZE) {
1303 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1304 VDEV_AUX_TOO_SMALL);
1305 return (SET_ERROR(EOVERFLOW));
1308 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1309 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1310 VDEV_LABEL_END_SIZE);
1312 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1313 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1314 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1315 VDEV_AUX_TOO_SMALL);
1316 return (SET_ERROR(EOVERFLOW));
1320 max_asize = max_osize;
1323 vd->vdev_psize = psize;
1326 * Make sure the allocatable size hasn't shrunk.
1328 if (asize < vd->vdev_min_asize) {
1329 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1330 VDEV_AUX_BAD_LABEL);
1331 return (SET_ERROR(EINVAL));
1334 vd->vdev_physical_ashift =
1335 MAX(physical_ashift, vd->vdev_physical_ashift);
1336 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1337 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1339 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1340 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1341 VDEV_AUX_ASHIFT_TOO_BIG);
1345 if (vd->vdev_asize == 0) {
1347 * This is the first-ever open, so use the computed values.
1348 * For testing purposes, a higher ashift can be requested.
1350 vd->vdev_asize = asize;
1351 vd->vdev_max_asize = max_asize;
1354 * Make sure the alignment requirement hasn't increased.
1356 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1357 vd->vdev_ops->vdev_op_leaf) {
1358 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1359 VDEV_AUX_BAD_LABEL);
1362 vd->vdev_max_asize = max_asize;
1366 * If all children are healthy and the asize has increased,
1367 * then we've experienced dynamic LUN growth. If automatic
1368 * expansion is enabled then use the additional space.
1370 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1371 (vd->vdev_expanding || spa->spa_autoexpand))
1372 vd->vdev_asize = asize;
1374 vdev_set_min_asize(vd);
1377 * Ensure we can issue some IO before declaring the
1378 * vdev open for business.
1380 if (vd->vdev_ops->vdev_op_leaf &&
1381 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1382 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1383 VDEV_AUX_ERR_EXCEEDED);
1388 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1389 * resilver. But don't do this if we are doing a reopen for a scrub,
1390 * since this would just restart the scrub we are already doing.
1392 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1393 vdev_resilver_needed(vd, NULL, NULL))
1394 spa_async_request(spa, SPA_ASYNC_RESILVER);
1400 * Called once the vdevs are all opened, this routine validates the label
1401 * contents. This needs to be done before vdev_load() so that we don't
1402 * inadvertently do repair I/Os to the wrong device.
1404 * If 'strict' is false ignore the spa guid check. This is necessary because
1405 * if the machine crashed during a re-guid the new guid might have been written
1406 * to all of the vdev labels, but not the cached config. The strict check
1407 * will be performed when the pool is opened again using the mos config.
1409 * This function will only return failure if one of the vdevs indicates that it
1410 * has since been destroyed or exported. This is only possible if
1411 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1412 * will be updated but the function will return 0.
1415 vdev_validate(vdev_t *vd, boolean_t strict)
1417 spa_t *spa = vd->vdev_spa;
1419 uint64_t guid = 0, top_guid;
1422 for (int c = 0; c < vd->vdev_children; c++)
1423 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1424 return (SET_ERROR(EBADF));
1427 * If the device has already failed, or was marked offline, don't do
1428 * any further validation. Otherwise, label I/O will fail and we will
1429 * overwrite the previous state.
1431 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1432 uint64_t aux_guid = 0;
1434 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1435 spa_last_synced_txg(spa) : -1ULL;
1437 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1438 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1439 VDEV_AUX_BAD_LABEL);
1444 * Determine if this vdev has been split off into another
1445 * pool. If so, then refuse to open it.
1447 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1448 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1449 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1450 VDEV_AUX_SPLIT_POOL);
1455 if (strict && (nvlist_lookup_uint64(label,
1456 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1457 guid != spa_guid(spa))) {
1458 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1459 VDEV_AUX_CORRUPT_DATA);
1464 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1465 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1470 * If this vdev just became a top-level vdev because its
1471 * sibling was detached, it will have adopted the parent's
1472 * vdev guid -- but the label may or may not be on disk yet.
1473 * Fortunately, either version of the label will have the
1474 * same top guid, so if we're a top-level vdev, we can
1475 * safely compare to that instead.
1477 * If we split this vdev off instead, then we also check the
1478 * original pool's guid. We don't want to consider the vdev
1479 * corrupt if it is partway through a split operation.
1481 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1483 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1485 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1486 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1487 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1488 VDEV_AUX_CORRUPT_DATA);
1493 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1495 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1496 VDEV_AUX_CORRUPT_DATA);
1504 * If this is a verbatim import, no need to check the
1505 * state of the pool.
1507 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1508 spa_load_state(spa) == SPA_LOAD_OPEN &&
1509 state != POOL_STATE_ACTIVE)
1510 return (SET_ERROR(EBADF));
1513 * If we were able to open and validate a vdev that was
1514 * previously marked permanently unavailable, clear that state
1517 if (vd->vdev_not_present)
1518 vd->vdev_not_present = 0;
1525 * Close a virtual device.
1528 vdev_close(vdev_t *vd)
1530 spa_t *spa = vd->vdev_spa;
1531 vdev_t *pvd = vd->vdev_parent;
1533 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1536 * If our parent is reopening, then we are as well, unless we are
1539 if (pvd != NULL && pvd->vdev_reopening)
1540 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1542 vd->vdev_ops->vdev_op_close(vd);
1544 vdev_cache_purge(vd);
1546 if (vd->vdev_ops->vdev_op_leaf)
1547 trim_map_destroy(vd);
1550 * We record the previous state before we close it, so that if we are
1551 * doing a reopen(), we don't generate FMA ereports if we notice that
1552 * it's still faulted.
1554 vd->vdev_prevstate = vd->vdev_state;
1556 if (vd->vdev_offline)
1557 vd->vdev_state = VDEV_STATE_OFFLINE;
1559 vd->vdev_state = VDEV_STATE_CLOSED;
1560 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1564 vdev_hold(vdev_t *vd)
1566 spa_t *spa = vd->vdev_spa;
1568 ASSERT(spa_is_root(spa));
1569 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1572 for (int c = 0; c < vd->vdev_children; c++)
1573 vdev_hold(vd->vdev_child[c]);
1575 if (vd->vdev_ops->vdev_op_leaf)
1576 vd->vdev_ops->vdev_op_hold(vd);
1580 vdev_rele(vdev_t *vd)
1582 spa_t *spa = vd->vdev_spa;
1584 ASSERT(spa_is_root(spa));
1585 for (int c = 0; c < vd->vdev_children; c++)
1586 vdev_rele(vd->vdev_child[c]);
1588 if (vd->vdev_ops->vdev_op_leaf)
1589 vd->vdev_ops->vdev_op_rele(vd);
1593 * Reopen all interior vdevs and any unopened leaves. We don't actually
1594 * reopen leaf vdevs which had previously been opened as they might deadlock
1595 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1596 * If the leaf has never been opened then open it, as usual.
1599 vdev_reopen(vdev_t *vd)
1601 spa_t *spa = vd->vdev_spa;
1603 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1605 /* set the reopening flag unless we're taking the vdev offline */
1606 vd->vdev_reopening = !vd->vdev_offline;
1608 (void) vdev_open(vd);
1611 * Call vdev_validate() here to make sure we have the same device.
1612 * Otherwise, a device with an invalid label could be successfully
1613 * opened in response to vdev_reopen().
1616 (void) vdev_validate_aux(vd);
1617 if (vdev_readable(vd) && vdev_writeable(vd) &&
1618 vd->vdev_aux == &spa->spa_l2cache &&
1619 !l2arc_vdev_present(vd))
1620 l2arc_add_vdev(spa, vd);
1622 (void) vdev_validate(vd, B_TRUE);
1626 * Reassess parent vdev's health.
1628 vdev_propagate_state(vd);
1632 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1637 * Normally, partial opens (e.g. of a mirror) are allowed.
1638 * For a create, however, we want to fail the request if
1639 * there are any components we can't open.
1641 error = vdev_open(vd);
1643 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1645 return (error ? error : ENXIO);
1649 * Recursively load DTLs and initialize all labels.
1651 if ((error = vdev_dtl_load(vd)) != 0 ||
1652 (error = vdev_label_init(vd, txg, isreplacing ?
1653 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1662 vdev_metaslab_set_size(vdev_t *vd)
1665 * Aim for roughly 200 metaslabs per vdev.
1667 vd->vdev_ms_shift = highbit64(vd->vdev_asize / 200);
1668 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1672 * Maximize performance by inflating the configured ashift for top level
1673 * vdevs to be as close to the physical ashift as possible while maintaining
1674 * administrator defined limits and ensuring it doesn't go below the
1678 vdev_ashift_optimize(vdev_t *vd)
1680 if (vd == vd->vdev_top) {
1681 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1682 vd->vdev_ashift = MIN(
1683 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1684 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1687 * Unusual case where logical ashift > physical ashift
1688 * so we can't cap the calculated ashift based on max
1689 * ashift as that would cause failures.
1690 * We still check if we need to increase it to match
1693 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1700 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1702 ASSERT(vd == vd->vdev_top);
1703 ASSERT(!vd->vdev_ishole);
1704 ASSERT(ISP2(flags));
1705 ASSERT(spa_writeable(vd->vdev_spa));
1707 if (flags & VDD_METASLAB)
1708 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1710 if (flags & VDD_DTL)
1711 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1713 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1717 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1719 for (int c = 0; c < vd->vdev_children; c++)
1720 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1722 if (vd->vdev_ops->vdev_op_leaf)
1723 vdev_dirty(vd->vdev_top, flags, vd, txg);
1729 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1730 * the vdev has less than perfect replication. There are four kinds of DTL:
1732 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1734 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1736 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1737 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1738 * txgs that was scrubbed.
1740 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1741 * persistent errors or just some device being offline.
1742 * Unlike the other three, the DTL_OUTAGE map is not generally
1743 * maintained; it's only computed when needed, typically to
1744 * determine whether a device can be detached.
1746 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1747 * either has the data or it doesn't.
1749 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1750 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1751 * if any child is less than fully replicated, then so is its parent.
1752 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1753 * comprising only those txgs which appear in 'maxfaults' or more children;
1754 * those are the txgs we don't have enough replication to read. For example,
1755 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1756 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1757 * two child DTL_MISSING maps.
1759 * It should be clear from the above that to compute the DTLs and outage maps
1760 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1761 * Therefore, that is all we keep on disk. When loading the pool, or after
1762 * a configuration change, we generate all other DTLs from first principles.
1765 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1767 range_tree_t *rt = vd->vdev_dtl[t];
1769 ASSERT(t < DTL_TYPES);
1770 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1771 ASSERT(spa_writeable(vd->vdev_spa));
1773 mutex_enter(rt->rt_lock);
1774 if (!range_tree_contains(rt, txg, size))
1775 range_tree_add(rt, txg, size);
1776 mutex_exit(rt->rt_lock);
1780 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1782 range_tree_t *rt = vd->vdev_dtl[t];
1783 boolean_t dirty = B_FALSE;
1785 ASSERT(t < DTL_TYPES);
1786 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1788 mutex_enter(rt->rt_lock);
1789 if (range_tree_space(rt) != 0)
1790 dirty = range_tree_contains(rt, txg, size);
1791 mutex_exit(rt->rt_lock);
1797 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1799 range_tree_t *rt = vd->vdev_dtl[t];
1802 mutex_enter(rt->rt_lock);
1803 empty = (range_tree_space(rt) == 0);
1804 mutex_exit(rt->rt_lock);
1810 * Returns the lowest txg in the DTL range.
1813 vdev_dtl_min(vdev_t *vd)
1817 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1818 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1819 ASSERT0(vd->vdev_children);
1821 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1822 return (rs->rs_start - 1);
1826 * Returns the highest txg in the DTL.
1829 vdev_dtl_max(vdev_t *vd)
1833 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1834 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1835 ASSERT0(vd->vdev_children);
1837 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1838 return (rs->rs_end);
1842 * Determine if a resilvering vdev should remove any DTL entries from
1843 * its range. If the vdev was resilvering for the entire duration of the
1844 * scan then it should excise that range from its DTLs. Otherwise, this
1845 * vdev is considered partially resilvered and should leave its DTL
1846 * entries intact. The comment in vdev_dtl_reassess() describes how we
1850 vdev_dtl_should_excise(vdev_t *vd)
1852 spa_t *spa = vd->vdev_spa;
1853 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1855 ASSERT0(scn->scn_phys.scn_errors);
1856 ASSERT0(vd->vdev_children);
1858 if (vd->vdev_resilver_txg == 0 ||
1859 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1863 * When a resilver is initiated the scan will assign the scn_max_txg
1864 * value to the highest txg value that exists in all DTLs. If this
1865 * device's max DTL is not part of this scan (i.e. it is not in
1866 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1869 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1870 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1871 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1872 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1879 * Reassess DTLs after a config change or scrub completion.
1882 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1884 spa_t *spa = vd->vdev_spa;
1888 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1890 for (int c = 0; c < vd->vdev_children; c++)
1891 vdev_dtl_reassess(vd->vdev_child[c], txg,
1892 scrub_txg, scrub_done);
1894 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1897 if (vd->vdev_ops->vdev_op_leaf) {
1898 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1900 mutex_enter(&vd->vdev_dtl_lock);
1903 * If we've completed a scan cleanly then determine
1904 * if this vdev should remove any DTLs. We only want to
1905 * excise regions on vdevs that were available during
1906 * the entire duration of this scan.
1908 if (scrub_txg != 0 &&
1909 (spa->spa_scrub_started ||
1910 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1911 vdev_dtl_should_excise(vd)) {
1913 * We completed a scrub up to scrub_txg. If we
1914 * did it without rebooting, then the scrub dtl
1915 * will be valid, so excise the old region and
1916 * fold in the scrub dtl. Otherwise, leave the
1917 * dtl as-is if there was an error.
1919 * There's little trick here: to excise the beginning
1920 * of the DTL_MISSING map, we put it into a reference
1921 * tree and then add a segment with refcnt -1 that
1922 * covers the range [0, scrub_txg). This means
1923 * that each txg in that range has refcnt -1 or 0.
1924 * We then add DTL_SCRUB with a refcnt of 2, so that
1925 * entries in the range [0, scrub_txg) will have a
1926 * positive refcnt -- either 1 or 2. We then convert
1927 * the reference tree into the new DTL_MISSING map.
1929 space_reftree_create(&reftree);
1930 space_reftree_add_map(&reftree,
1931 vd->vdev_dtl[DTL_MISSING], 1);
1932 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1933 space_reftree_add_map(&reftree,
1934 vd->vdev_dtl[DTL_SCRUB], 2);
1935 space_reftree_generate_map(&reftree,
1936 vd->vdev_dtl[DTL_MISSING], 1);
1937 space_reftree_destroy(&reftree);
1939 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1940 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1941 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1943 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1944 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1945 if (!vdev_readable(vd))
1946 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1948 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1949 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1952 * If the vdev was resilvering and no longer has any
1953 * DTLs then reset its resilvering flag and dirty
1954 * the top level so that we persist the change.
1956 if (vd->vdev_resilver_txg != 0 &&
1957 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1958 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
1959 vd->vdev_resilver_txg = 0;
1960 vdev_config_dirty(vd->vdev_top);
1963 mutex_exit(&vd->vdev_dtl_lock);
1966 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1970 mutex_enter(&vd->vdev_dtl_lock);
1971 for (int t = 0; t < DTL_TYPES; t++) {
1972 /* account for child's outage in parent's missing map */
1973 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1975 continue; /* leaf vdevs only */
1976 if (t == DTL_PARTIAL)
1977 minref = 1; /* i.e. non-zero */
1978 else if (vd->vdev_nparity != 0)
1979 minref = vd->vdev_nparity + 1; /* RAID-Z */
1981 minref = vd->vdev_children; /* any kind of mirror */
1982 space_reftree_create(&reftree);
1983 for (int c = 0; c < vd->vdev_children; c++) {
1984 vdev_t *cvd = vd->vdev_child[c];
1985 mutex_enter(&cvd->vdev_dtl_lock);
1986 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1987 mutex_exit(&cvd->vdev_dtl_lock);
1989 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
1990 space_reftree_destroy(&reftree);
1992 mutex_exit(&vd->vdev_dtl_lock);
1996 vdev_dtl_load(vdev_t *vd)
1998 spa_t *spa = vd->vdev_spa;
1999 objset_t *mos = spa->spa_meta_objset;
2002 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2003 ASSERT(!vd->vdev_ishole);
2005 error = space_map_open(&vd->vdev_dtl_sm, mos,
2006 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2009 ASSERT(vd->vdev_dtl_sm != NULL);
2011 mutex_enter(&vd->vdev_dtl_lock);
2014 * Now that we've opened the space_map we need to update
2017 space_map_update(vd->vdev_dtl_sm);
2019 error = space_map_load(vd->vdev_dtl_sm,
2020 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2021 mutex_exit(&vd->vdev_dtl_lock);
2026 for (int c = 0; c < vd->vdev_children; c++) {
2027 error = vdev_dtl_load(vd->vdev_child[c]);
2036 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2038 spa_t *spa = vd->vdev_spa;
2039 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2040 objset_t *mos = spa->spa_meta_objset;
2041 range_tree_t *rtsync;
2044 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2046 ASSERT(!vd->vdev_ishole);
2047 ASSERT(vd->vdev_ops->vdev_op_leaf);
2049 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2051 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2052 mutex_enter(&vd->vdev_dtl_lock);
2053 space_map_free(vd->vdev_dtl_sm, tx);
2054 space_map_close(vd->vdev_dtl_sm);
2055 vd->vdev_dtl_sm = NULL;
2056 mutex_exit(&vd->vdev_dtl_lock);
2061 if (vd->vdev_dtl_sm == NULL) {
2062 uint64_t new_object;
2064 new_object = space_map_alloc(mos, tx);
2065 VERIFY3U(new_object, !=, 0);
2067 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2068 0, -1ULL, 0, &vd->vdev_dtl_lock));
2069 ASSERT(vd->vdev_dtl_sm != NULL);
2072 bzero(&rtlock, sizeof(rtlock));
2073 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2075 rtsync = range_tree_create(NULL, NULL, &rtlock);
2077 mutex_enter(&rtlock);
2079 mutex_enter(&vd->vdev_dtl_lock);
2080 range_tree_walk(rt, range_tree_add, rtsync);
2081 mutex_exit(&vd->vdev_dtl_lock);
2083 space_map_truncate(vd->vdev_dtl_sm, tx);
2084 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2085 range_tree_vacate(rtsync, NULL, NULL);
2087 range_tree_destroy(rtsync);
2089 mutex_exit(&rtlock);
2090 mutex_destroy(&rtlock);
2093 * If the object for the space map has changed then dirty
2094 * the top level so that we update the config.
2096 if (object != space_map_object(vd->vdev_dtl_sm)) {
2097 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2098 "new object %llu", txg, spa_name(spa), object,
2099 space_map_object(vd->vdev_dtl_sm));
2100 vdev_config_dirty(vd->vdev_top);
2105 mutex_enter(&vd->vdev_dtl_lock);
2106 space_map_update(vd->vdev_dtl_sm);
2107 mutex_exit(&vd->vdev_dtl_lock);
2111 * Determine whether the specified vdev can be offlined/detached/removed
2112 * without losing data.
2115 vdev_dtl_required(vdev_t *vd)
2117 spa_t *spa = vd->vdev_spa;
2118 vdev_t *tvd = vd->vdev_top;
2119 uint8_t cant_read = vd->vdev_cant_read;
2122 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2124 if (vd == spa->spa_root_vdev || vd == tvd)
2128 * Temporarily mark the device as unreadable, and then determine
2129 * whether this results in any DTL outages in the top-level vdev.
2130 * If not, we can safely offline/detach/remove the device.
2132 vd->vdev_cant_read = B_TRUE;
2133 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2134 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2135 vd->vdev_cant_read = cant_read;
2136 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2138 if (!required && zio_injection_enabled)
2139 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2145 * Determine if resilver is needed, and if so the txg range.
2148 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2150 boolean_t needed = B_FALSE;
2151 uint64_t thismin = UINT64_MAX;
2152 uint64_t thismax = 0;
2154 if (vd->vdev_children == 0) {
2155 mutex_enter(&vd->vdev_dtl_lock);
2156 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2157 vdev_writeable(vd)) {
2159 thismin = vdev_dtl_min(vd);
2160 thismax = vdev_dtl_max(vd);
2163 mutex_exit(&vd->vdev_dtl_lock);
2165 for (int c = 0; c < vd->vdev_children; c++) {
2166 vdev_t *cvd = vd->vdev_child[c];
2167 uint64_t cmin, cmax;
2169 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2170 thismin = MIN(thismin, cmin);
2171 thismax = MAX(thismax, cmax);
2177 if (needed && minp) {
2185 vdev_load(vdev_t *vd)
2188 * Recursively load all children.
2190 for (int c = 0; c < vd->vdev_children; c++)
2191 vdev_load(vd->vdev_child[c]);
2194 * If this is a top-level vdev, initialize its metaslabs.
2196 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2197 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2198 vdev_metaslab_init(vd, 0) != 0))
2199 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2200 VDEV_AUX_CORRUPT_DATA);
2203 * If this is a leaf vdev, load its DTL.
2205 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2206 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2207 VDEV_AUX_CORRUPT_DATA);
2211 * The special vdev case is used for hot spares and l2cache devices. Its
2212 * sole purpose it to set the vdev state for the associated vdev. To do this,
2213 * we make sure that we can open the underlying device, then try to read the
2214 * label, and make sure that the label is sane and that it hasn't been
2215 * repurposed to another pool.
2218 vdev_validate_aux(vdev_t *vd)
2221 uint64_t guid, version;
2224 if (!vdev_readable(vd))
2227 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2228 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2229 VDEV_AUX_CORRUPT_DATA);
2233 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2234 !SPA_VERSION_IS_SUPPORTED(version) ||
2235 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2236 guid != vd->vdev_guid ||
2237 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2238 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2239 VDEV_AUX_CORRUPT_DATA);
2245 * We don't actually check the pool state here. If it's in fact in
2246 * use by another pool, we update this fact on the fly when requested.
2253 vdev_remove(vdev_t *vd, uint64_t txg)
2255 spa_t *spa = vd->vdev_spa;
2256 objset_t *mos = spa->spa_meta_objset;
2259 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2261 if (vd->vdev_ms != NULL) {
2262 metaslab_group_t *mg = vd->vdev_mg;
2264 metaslab_group_histogram_verify(mg);
2265 metaslab_class_histogram_verify(mg->mg_class);
2267 for (int m = 0; m < vd->vdev_ms_count; m++) {
2268 metaslab_t *msp = vd->vdev_ms[m];
2270 if (msp == NULL || msp->ms_sm == NULL)
2273 mutex_enter(&msp->ms_lock);
2275 * If the metaslab was not loaded when the vdev
2276 * was removed then the histogram accounting may
2277 * not be accurate. Update the histogram information
2278 * here so that we ensure that the metaslab group
2279 * and metaslab class are up-to-date.
2281 metaslab_group_histogram_remove(mg, msp);
2283 VERIFY0(space_map_allocated(msp->ms_sm));
2284 space_map_free(msp->ms_sm, tx);
2285 space_map_close(msp->ms_sm);
2287 mutex_exit(&msp->ms_lock);
2290 metaslab_group_histogram_verify(mg);
2291 metaslab_class_histogram_verify(mg->mg_class);
2292 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2293 ASSERT0(mg->mg_histogram[i]);
2297 if (vd->vdev_ms_array) {
2298 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2299 vd->vdev_ms_array = 0;
2305 vdev_sync_done(vdev_t *vd, uint64_t txg)
2308 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2310 ASSERT(!vd->vdev_ishole);
2312 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2313 metaslab_sync_done(msp, txg);
2316 metaslab_sync_reassess(vd->vdev_mg);
2320 vdev_sync(vdev_t *vd, uint64_t txg)
2322 spa_t *spa = vd->vdev_spa;
2327 ASSERT(!vd->vdev_ishole);
2329 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2330 ASSERT(vd == vd->vdev_top);
2331 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2332 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2333 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2334 ASSERT(vd->vdev_ms_array != 0);
2335 vdev_config_dirty(vd);
2340 * Remove the metadata associated with this vdev once it's empty.
2342 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2343 vdev_remove(vd, txg);
2345 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2346 metaslab_sync(msp, txg);
2347 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2350 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2351 vdev_dtl_sync(lvd, txg);
2353 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2357 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2359 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2363 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2364 * not be opened, and no I/O is attempted.
2367 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2371 spa_vdev_state_enter(spa, SCL_NONE);
2373 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2374 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2376 if (!vd->vdev_ops->vdev_op_leaf)
2377 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2382 * We don't directly use the aux state here, but if we do a
2383 * vdev_reopen(), we need this value to be present to remember why we
2386 vd->vdev_label_aux = aux;
2389 * Faulted state takes precedence over degraded.
2391 vd->vdev_delayed_close = B_FALSE;
2392 vd->vdev_faulted = 1ULL;
2393 vd->vdev_degraded = 0ULL;
2394 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2397 * If this device has the only valid copy of the data, then
2398 * back off and simply mark the vdev as degraded instead.
2400 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2401 vd->vdev_degraded = 1ULL;
2402 vd->vdev_faulted = 0ULL;
2405 * If we reopen the device and it's not dead, only then do we
2410 if (vdev_readable(vd))
2411 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2414 return (spa_vdev_state_exit(spa, vd, 0));
2418 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2419 * user that something is wrong. The vdev continues to operate as normal as far
2420 * as I/O is concerned.
2423 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2427 spa_vdev_state_enter(spa, SCL_NONE);
2429 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2430 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2432 if (!vd->vdev_ops->vdev_op_leaf)
2433 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2436 * If the vdev is already faulted, then don't do anything.
2438 if (vd->vdev_faulted || vd->vdev_degraded)
2439 return (spa_vdev_state_exit(spa, NULL, 0));
2441 vd->vdev_degraded = 1ULL;
2442 if (!vdev_is_dead(vd))
2443 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2446 return (spa_vdev_state_exit(spa, vd, 0));
2450 * Online the given vdev.
2452 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2453 * spare device should be detached when the device finishes resilvering.
2454 * Second, the online should be treated like a 'test' online case, so no FMA
2455 * events are generated if the device fails to open.
2458 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2460 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2462 spa_vdev_state_enter(spa, SCL_NONE);
2464 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2465 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2467 if (!vd->vdev_ops->vdev_op_leaf)
2468 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2471 vd->vdev_offline = B_FALSE;
2472 vd->vdev_tmpoffline = B_FALSE;
2473 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2474 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2476 /* XXX - L2ARC 1.0 does not support expansion */
2477 if (!vd->vdev_aux) {
2478 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2479 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2483 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2485 if (!vd->vdev_aux) {
2486 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2487 pvd->vdev_expanding = B_FALSE;
2491 *newstate = vd->vdev_state;
2492 if ((flags & ZFS_ONLINE_UNSPARE) &&
2493 !vdev_is_dead(vd) && vd->vdev_parent &&
2494 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2495 vd->vdev_parent->vdev_child[0] == vd)
2496 vd->vdev_unspare = B_TRUE;
2498 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2500 /* XXX - L2ARC 1.0 does not support expansion */
2502 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2503 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2505 return (spa_vdev_state_exit(spa, vd, 0));
2509 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2513 uint64_t generation;
2514 metaslab_group_t *mg;
2517 spa_vdev_state_enter(spa, SCL_ALLOC);
2519 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2520 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2522 if (!vd->vdev_ops->vdev_op_leaf)
2523 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2527 generation = spa->spa_config_generation + 1;
2530 * If the device isn't already offline, try to offline it.
2532 if (!vd->vdev_offline) {
2534 * If this device has the only valid copy of some data,
2535 * don't allow it to be offlined. Log devices are always
2538 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2539 vdev_dtl_required(vd))
2540 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2543 * If the top-level is a slog and it has had allocations
2544 * then proceed. We check that the vdev's metaslab group
2545 * is not NULL since it's possible that we may have just
2546 * added this vdev but not yet initialized its metaslabs.
2548 if (tvd->vdev_islog && mg != NULL) {
2550 * Prevent any future allocations.
2552 metaslab_group_passivate(mg);
2553 (void) spa_vdev_state_exit(spa, vd, 0);
2555 error = spa_offline_log(spa);
2557 spa_vdev_state_enter(spa, SCL_ALLOC);
2560 * Check to see if the config has changed.
2562 if (error || generation != spa->spa_config_generation) {
2563 metaslab_group_activate(mg);
2565 return (spa_vdev_state_exit(spa,
2567 (void) spa_vdev_state_exit(spa, vd, 0);
2570 ASSERT0(tvd->vdev_stat.vs_alloc);
2574 * Offline this device and reopen its top-level vdev.
2575 * If the top-level vdev is a log device then just offline
2576 * it. Otherwise, if this action results in the top-level
2577 * vdev becoming unusable, undo it and fail the request.
2579 vd->vdev_offline = B_TRUE;
2582 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2583 vdev_is_dead(tvd)) {
2584 vd->vdev_offline = B_FALSE;
2586 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2590 * Add the device back into the metaslab rotor so that
2591 * once we online the device it's open for business.
2593 if (tvd->vdev_islog && mg != NULL)
2594 metaslab_group_activate(mg);
2597 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2599 return (spa_vdev_state_exit(spa, vd, 0));
2603 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2607 mutex_enter(&spa->spa_vdev_top_lock);
2608 error = vdev_offline_locked(spa, guid, flags);
2609 mutex_exit(&spa->spa_vdev_top_lock);
2615 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2616 * vdev_offline(), we assume the spa config is locked. We also clear all
2617 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2620 vdev_clear(spa_t *spa, vdev_t *vd)
2622 vdev_t *rvd = spa->spa_root_vdev;
2624 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2629 vd->vdev_stat.vs_read_errors = 0;
2630 vd->vdev_stat.vs_write_errors = 0;
2631 vd->vdev_stat.vs_checksum_errors = 0;
2633 for (int c = 0; c < vd->vdev_children; c++)
2634 vdev_clear(spa, vd->vdev_child[c]);
2637 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2638 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2640 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2641 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2645 * If we're in the FAULTED state or have experienced failed I/O, then
2646 * clear the persistent state and attempt to reopen the device. We
2647 * also mark the vdev config dirty, so that the new faulted state is
2648 * written out to disk.
2650 if (vd->vdev_faulted || vd->vdev_degraded ||
2651 !vdev_readable(vd) || !vdev_writeable(vd)) {
2654 * When reopening in reponse to a clear event, it may be due to
2655 * a fmadm repair request. In this case, if the device is
2656 * still broken, we want to still post the ereport again.
2658 vd->vdev_forcefault = B_TRUE;
2660 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2661 vd->vdev_cant_read = B_FALSE;
2662 vd->vdev_cant_write = B_FALSE;
2664 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2666 vd->vdev_forcefault = B_FALSE;
2668 if (vd != rvd && vdev_writeable(vd->vdev_top))
2669 vdev_state_dirty(vd->vdev_top);
2671 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2672 spa_async_request(spa, SPA_ASYNC_RESILVER);
2674 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2678 * When clearing a FMA-diagnosed fault, we always want to
2679 * unspare the device, as we assume that the original spare was
2680 * done in response to the FMA fault.
2682 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2683 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2684 vd->vdev_parent->vdev_child[0] == vd)
2685 vd->vdev_unspare = B_TRUE;
2689 vdev_is_dead(vdev_t *vd)
2692 * Holes and missing devices are always considered "dead".
2693 * This simplifies the code since we don't have to check for
2694 * these types of devices in the various code paths.
2695 * Instead we rely on the fact that we skip over dead devices
2696 * before issuing I/O to them.
2698 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2699 vd->vdev_ops == &vdev_missing_ops);
2703 vdev_readable(vdev_t *vd)
2705 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2709 vdev_writeable(vdev_t *vd)
2711 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2715 vdev_allocatable(vdev_t *vd)
2717 uint64_t state = vd->vdev_state;
2720 * We currently allow allocations from vdevs which may be in the
2721 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2722 * fails to reopen then we'll catch it later when we're holding
2723 * the proper locks. Note that we have to get the vdev state
2724 * in a local variable because although it changes atomically,
2725 * we're asking two separate questions about it.
2727 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2728 !vd->vdev_cant_write && !vd->vdev_ishole);
2732 vdev_accessible(vdev_t *vd, zio_t *zio)
2734 ASSERT(zio->io_vd == vd);
2736 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2739 if (zio->io_type == ZIO_TYPE_READ)
2740 return (!vd->vdev_cant_read);
2742 if (zio->io_type == ZIO_TYPE_WRITE)
2743 return (!vd->vdev_cant_write);
2749 * Get statistics for the given vdev.
2752 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2754 spa_t *spa = vd->vdev_spa;
2755 vdev_t *rvd = spa->spa_root_vdev;
2757 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2759 mutex_enter(&vd->vdev_stat_lock);
2760 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2761 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2762 vs->vs_state = vd->vdev_state;
2763 vs->vs_rsize = vdev_get_min_asize(vd);
2764 if (vd->vdev_ops->vdev_op_leaf)
2765 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2766 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2767 vs->vs_configured_ashift = vd->vdev_top != NULL
2768 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2769 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2770 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2771 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2772 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2776 * If we're getting stats on the root vdev, aggregate the I/O counts
2777 * over all top-level vdevs (i.e. the direct children of the root).
2780 for (int c = 0; c < rvd->vdev_children; c++) {
2781 vdev_t *cvd = rvd->vdev_child[c];
2782 vdev_stat_t *cvs = &cvd->vdev_stat;
2784 for (int t = 0; t < ZIO_TYPES; t++) {
2785 vs->vs_ops[t] += cvs->vs_ops[t];
2786 vs->vs_bytes[t] += cvs->vs_bytes[t];
2788 cvs->vs_scan_removing = cvd->vdev_removing;
2791 mutex_exit(&vd->vdev_stat_lock);
2795 vdev_clear_stats(vdev_t *vd)
2797 mutex_enter(&vd->vdev_stat_lock);
2798 vd->vdev_stat.vs_space = 0;
2799 vd->vdev_stat.vs_dspace = 0;
2800 vd->vdev_stat.vs_alloc = 0;
2801 mutex_exit(&vd->vdev_stat_lock);
2805 vdev_scan_stat_init(vdev_t *vd)
2807 vdev_stat_t *vs = &vd->vdev_stat;
2809 for (int c = 0; c < vd->vdev_children; c++)
2810 vdev_scan_stat_init(vd->vdev_child[c]);
2812 mutex_enter(&vd->vdev_stat_lock);
2813 vs->vs_scan_processed = 0;
2814 mutex_exit(&vd->vdev_stat_lock);
2818 vdev_stat_update(zio_t *zio, uint64_t psize)
2820 spa_t *spa = zio->io_spa;
2821 vdev_t *rvd = spa->spa_root_vdev;
2822 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2824 uint64_t txg = zio->io_txg;
2825 vdev_stat_t *vs = &vd->vdev_stat;
2826 zio_type_t type = zio->io_type;
2827 int flags = zio->io_flags;
2830 * If this i/o is a gang leader, it didn't do any actual work.
2832 if (zio->io_gang_tree)
2835 if (zio->io_error == 0) {
2837 * If this is a root i/o, don't count it -- we've already
2838 * counted the top-level vdevs, and vdev_get_stats() will
2839 * aggregate them when asked. This reduces contention on
2840 * the root vdev_stat_lock and implicitly handles blocks
2841 * that compress away to holes, for which there is no i/o.
2842 * (Holes never create vdev children, so all the counters
2843 * remain zero, which is what we want.)
2845 * Note: this only applies to successful i/o (io_error == 0)
2846 * because unlike i/o counts, errors are not additive.
2847 * When reading a ditto block, for example, failure of
2848 * one top-level vdev does not imply a root-level error.
2853 ASSERT(vd == zio->io_vd);
2855 if (flags & ZIO_FLAG_IO_BYPASS)
2858 mutex_enter(&vd->vdev_stat_lock);
2860 if (flags & ZIO_FLAG_IO_REPAIR) {
2861 if (flags & ZIO_FLAG_SCAN_THREAD) {
2862 dsl_scan_phys_t *scn_phys =
2863 &spa->spa_dsl_pool->dp_scan->scn_phys;
2864 uint64_t *processed = &scn_phys->scn_processed;
2867 if (vd->vdev_ops->vdev_op_leaf)
2868 atomic_add_64(processed, psize);
2869 vs->vs_scan_processed += psize;
2872 if (flags & ZIO_FLAG_SELF_HEAL)
2873 vs->vs_self_healed += psize;
2877 vs->vs_bytes[type] += psize;
2879 mutex_exit(&vd->vdev_stat_lock);
2883 if (flags & ZIO_FLAG_SPECULATIVE)
2887 * If this is an I/O error that is going to be retried, then ignore the
2888 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2889 * hard errors, when in reality they can happen for any number of
2890 * innocuous reasons (bus resets, MPxIO link failure, etc).
2892 if (zio->io_error == EIO &&
2893 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2897 * Intent logs writes won't propagate their error to the root
2898 * I/O so don't mark these types of failures as pool-level
2901 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2904 mutex_enter(&vd->vdev_stat_lock);
2905 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2906 if (zio->io_error == ECKSUM)
2907 vs->vs_checksum_errors++;
2909 vs->vs_read_errors++;
2911 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2912 vs->vs_write_errors++;
2913 mutex_exit(&vd->vdev_stat_lock);
2915 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2916 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2917 (flags & ZIO_FLAG_SCAN_THREAD) ||
2918 spa->spa_claiming)) {
2920 * This is either a normal write (not a repair), or it's
2921 * a repair induced by the scrub thread, or it's a repair
2922 * made by zil_claim() during spa_load() in the first txg.
2923 * In the normal case, we commit the DTL change in the same
2924 * txg as the block was born. In the scrub-induced repair
2925 * case, we know that scrubs run in first-pass syncing context,
2926 * so we commit the DTL change in spa_syncing_txg(spa).
2927 * In the zil_claim() case, we commit in spa_first_txg(spa).
2929 * We currently do not make DTL entries for failed spontaneous
2930 * self-healing writes triggered by normal (non-scrubbing)
2931 * reads, because we have no transactional context in which to
2932 * do so -- and it's not clear that it'd be desirable anyway.
2934 if (vd->vdev_ops->vdev_op_leaf) {
2935 uint64_t commit_txg = txg;
2936 if (flags & ZIO_FLAG_SCAN_THREAD) {
2937 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2938 ASSERT(spa_sync_pass(spa) == 1);
2939 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2940 commit_txg = spa_syncing_txg(spa);
2941 } else if (spa->spa_claiming) {
2942 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2943 commit_txg = spa_first_txg(spa);
2945 ASSERT(commit_txg >= spa_syncing_txg(spa));
2946 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2948 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2949 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2950 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2953 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2958 * Update the in-core space usage stats for this vdev, its metaslab class,
2959 * and the root vdev.
2962 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2963 int64_t space_delta)
2965 int64_t dspace_delta = space_delta;
2966 spa_t *spa = vd->vdev_spa;
2967 vdev_t *rvd = spa->spa_root_vdev;
2968 metaslab_group_t *mg = vd->vdev_mg;
2969 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2971 ASSERT(vd == vd->vdev_top);
2974 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2975 * factor. We must calculate this here and not at the root vdev
2976 * because the root vdev's psize-to-asize is simply the max of its
2977 * childrens', thus not accurate enough for us.
2979 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2980 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2981 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2982 vd->vdev_deflate_ratio;
2984 mutex_enter(&vd->vdev_stat_lock);
2985 vd->vdev_stat.vs_alloc += alloc_delta;
2986 vd->vdev_stat.vs_space += space_delta;
2987 vd->vdev_stat.vs_dspace += dspace_delta;
2988 mutex_exit(&vd->vdev_stat_lock);
2990 if (mc == spa_normal_class(spa)) {
2991 mutex_enter(&rvd->vdev_stat_lock);
2992 rvd->vdev_stat.vs_alloc += alloc_delta;
2993 rvd->vdev_stat.vs_space += space_delta;
2994 rvd->vdev_stat.vs_dspace += dspace_delta;
2995 mutex_exit(&rvd->vdev_stat_lock);
2999 ASSERT(rvd == vd->vdev_parent);
3000 ASSERT(vd->vdev_ms_count != 0);
3002 metaslab_class_space_update(mc,
3003 alloc_delta, defer_delta, space_delta, dspace_delta);
3008 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3009 * so that it will be written out next time the vdev configuration is synced.
3010 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3013 vdev_config_dirty(vdev_t *vd)
3015 spa_t *spa = vd->vdev_spa;
3016 vdev_t *rvd = spa->spa_root_vdev;
3019 ASSERT(spa_writeable(spa));
3022 * If this is an aux vdev (as with l2cache and spare devices), then we
3023 * update the vdev config manually and set the sync flag.
3025 if (vd->vdev_aux != NULL) {
3026 spa_aux_vdev_t *sav = vd->vdev_aux;
3030 for (c = 0; c < sav->sav_count; c++) {
3031 if (sav->sav_vdevs[c] == vd)
3035 if (c == sav->sav_count) {
3037 * We're being removed. There's nothing more to do.
3039 ASSERT(sav->sav_sync == B_TRUE);
3043 sav->sav_sync = B_TRUE;
3045 if (nvlist_lookup_nvlist_array(sav->sav_config,
3046 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3047 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3048 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3054 * Setting the nvlist in the middle if the array is a little
3055 * sketchy, but it will work.
3057 nvlist_free(aux[c]);
3058 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3064 * The dirty list is protected by the SCL_CONFIG lock. The caller
3065 * must either hold SCL_CONFIG as writer, or must be the sync thread
3066 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3067 * so this is sufficient to ensure mutual exclusion.
3069 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3070 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3071 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3074 for (c = 0; c < rvd->vdev_children; c++)
3075 vdev_config_dirty(rvd->vdev_child[c]);
3077 ASSERT(vd == vd->vdev_top);
3079 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3081 list_insert_head(&spa->spa_config_dirty_list, vd);
3086 vdev_config_clean(vdev_t *vd)
3088 spa_t *spa = vd->vdev_spa;
3090 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3091 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3092 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3094 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3095 list_remove(&spa->spa_config_dirty_list, vd);
3099 * Mark a top-level vdev's state as dirty, so that the next pass of
3100 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3101 * the state changes from larger config changes because they require
3102 * much less locking, and are often needed for administrative actions.
3105 vdev_state_dirty(vdev_t *vd)
3107 spa_t *spa = vd->vdev_spa;
3109 ASSERT(spa_writeable(spa));
3110 ASSERT(vd == vd->vdev_top);
3113 * The state list is protected by the SCL_STATE lock. The caller
3114 * must either hold SCL_STATE as writer, or must be the sync thread
3115 * (which holds SCL_STATE as reader). There's only one sync thread,
3116 * so this is sufficient to ensure mutual exclusion.
3118 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3119 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3120 spa_config_held(spa, SCL_STATE, RW_READER)));
3122 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3123 list_insert_head(&spa->spa_state_dirty_list, vd);
3127 vdev_state_clean(vdev_t *vd)
3129 spa_t *spa = vd->vdev_spa;
3131 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3132 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3133 spa_config_held(spa, SCL_STATE, RW_READER)));
3135 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3136 list_remove(&spa->spa_state_dirty_list, vd);
3140 * Propagate vdev state up from children to parent.
3143 vdev_propagate_state(vdev_t *vd)
3145 spa_t *spa = vd->vdev_spa;
3146 vdev_t *rvd = spa->spa_root_vdev;
3147 int degraded = 0, faulted = 0;
3151 if (vd->vdev_children > 0) {
3152 for (int c = 0; c < vd->vdev_children; c++) {
3153 child = vd->vdev_child[c];
3156 * Don't factor holes into the decision.
3158 if (child->vdev_ishole)
3161 if (!vdev_readable(child) ||
3162 (!vdev_writeable(child) && spa_writeable(spa))) {
3164 * Root special: if there is a top-level log
3165 * device, treat the root vdev as if it were
3168 if (child->vdev_islog && vd == rvd)
3172 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3176 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3180 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3183 * Root special: if there is a top-level vdev that cannot be
3184 * opened due to corrupted metadata, then propagate the root
3185 * vdev's aux state as 'corrupt' rather than 'insufficient
3188 if (corrupted && vd == rvd &&
3189 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3190 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3191 VDEV_AUX_CORRUPT_DATA);
3194 if (vd->vdev_parent)
3195 vdev_propagate_state(vd->vdev_parent);
3199 * Set a vdev's state. If this is during an open, we don't update the parent
3200 * state, because we're in the process of opening children depth-first.
3201 * Otherwise, we propagate the change to the parent.
3203 * If this routine places a device in a faulted state, an appropriate ereport is
3207 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3209 uint64_t save_state;
3210 spa_t *spa = vd->vdev_spa;
3212 if (state == vd->vdev_state) {
3213 vd->vdev_stat.vs_aux = aux;
3217 save_state = vd->vdev_state;
3219 vd->vdev_state = state;
3220 vd->vdev_stat.vs_aux = aux;
3223 * If we are setting the vdev state to anything but an open state, then
3224 * always close the underlying device unless the device has requested
3225 * a delayed close (i.e. we're about to remove or fault the device).
3226 * Otherwise, we keep accessible but invalid devices open forever.
3227 * We don't call vdev_close() itself, because that implies some extra
3228 * checks (offline, etc) that we don't want here. This is limited to
3229 * leaf devices, because otherwise closing the device will affect other
3232 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3233 vd->vdev_ops->vdev_op_leaf)
3234 vd->vdev_ops->vdev_op_close(vd);
3237 * If we have brought this vdev back into service, we need
3238 * to notify fmd so that it can gracefully repair any outstanding
3239 * cases due to a missing device. We do this in all cases, even those
3240 * that probably don't correlate to a repaired fault. This is sure to
3241 * catch all cases, and we let the zfs-retire agent sort it out. If
3242 * this is a transient state it's OK, as the retire agent will
3243 * double-check the state of the vdev before repairing it.
3245 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3246 vd->vdev_prevstate != state)
3247 zfs_post_state_change(spa, vd);
3249 if (vd->vdev_removed &&
3250 state == VDEV_STATE_CANT_OPEN &&
3251 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3253 * If the previous state is set to VDEV_STATE_REMOVED, then this
3254 * device was previously marked removed and someone attempted to
3255 * reopen it. If this failed due to a nonexistent device, then
3256 * keep the device in the REMOVED state. We also let this be if
3257 * it is one of our special test online cases, which is only
3258 * attempting to online the device and shouldn't generate an FMA
3261 vd->vdev_state = VDEV_STATE_REMOVED;
3262 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3263 } else if (state == VDEV_STATE_REMOVED) {
3264 vd->vdev_removed = B_TRUE;
3265 } else if (state == VDEV_STATE_CANT_OPEN) {
3267 * If we fail to open a vdev during an import or recovery, we
3268 * mark it as "not available", which signifies that it was
3269 * never there to begin with. Failure to open such a device
3270 * is not considered an error.
3272 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3273 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3274 vd->vdev_ops->vdev_op_leaf)
3275 vd->vdev_not_present = 1;
3278 * Post the appropriate ereport. If the 'prevstate' field is
3279 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3280 * that this is part of a vdev_reopen(). In this case, we don't
3281 * want to post the ereport if the device was already in the
3282 * CANT_OPEN state beforehand.
3284 * If the 'checkremove' flag is set, then this is an attempt to
3285 * online the device in response to an insertion event. If we
3286 * hit this case, then we have detected an insertion event for a
3287 * faulted or offline device that wasn't in the removed state.
3288 * In this scenario, we don't post an ereport because we are
3289 * about to replace the device, or attempt an online with
3290 * vdev_forcefault, which will generate the fault for us.
3292 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3293 !vd->vdev_not_present && !vd->vdev_checkremove &&
3294 vd != spa->spa_root_vdev) {
3298 case VDEV_AUX_OPEN_FAILED:
3299 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3301 case VDEV_AUX_CORRUPT_DATA:
3302 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3304 case VDEV_AUX_NO_REPLICAS:
3305 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3307 case VDEV_AUX_BAD_GUID_SUM:
3308 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3310 case VDEV_AUX_TOO_SMALL:
3311 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3313 case VDEV_AUX_BAD_LABEL:
3314 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3317 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3320 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3323 /* Erase any notion of persistent removed state */
3324 vd->vdev_removed = B_FALSE;
3326 vd->vdev_removed = B_FALSE;
3329 if (!isopen && vd->vdev_parent)
3330 vdev_propagate_state(vd->vdev_parent);
3334 * Check the vdev configuration to ensure that it's capable of supporting
3337 * On Solaris, we do not support RAID-Z or partial configuration. In
3338 * addition, only a single top-level vdev is allowed and none of the
3339 * leaves can be wholedisks.
3341 * For FreeBSD, we can boot from any configuration. There is a
3342 * limitation that the boot filesystem must be either uncompressed or
3343 * compresses with lzjb compression but I'm not sure how to enforce
3347 vdev_is_bootable(vdev_t *vd)
3350 if (!vd->vdev_ops->vdev_op_leaf) {
3351 char *vdev_type = vd->vdev_ops->vdev_op_type;
3353 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3354 vd->vdev_children > 1) {
3356 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3357 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3360 } else if (vd->vdev_wholedisk == 1) {
3364 for (int c = 0; c < vd->vdev_children; c++) {
3365 if (!vdev_is_bootable(vd->vdev_child[c]))
3373 * Load the state from the original vdev tree (ovd) which
3374 * we've retrieved from the MOS config object. If the original
3375 * vdev was offline or faulted then we transfer that state to the
3376 * device in the current vdev tree (nvd).
3379 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3381 spa_t *spa = nvd->vdev_spa;
3383 ASSERT(nvd->vdev_top->vdev_islog);
3384 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3385 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3387 for (int c = 0; c < nvd->vdev_children; c++)
3388 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3390 if (nvd->vdev_ops->vdev_op_leaf) {
3392 * Restore the persistent vdev state
3394 nvd->vdev_offline = ovd->vdev_offline;
3395 nvd->vdev_faulted = ovd->vdev_faulted;
3396 nvd->vdev_degraded = ovd->vdev_degraded;
3397 nvd->vdev_removed = ovd->vdev_removed;
3402 * Determine if a log device has valid content. If the vdev was
3403 * removed or faulted in the MOS config then we know that
3404 * the content on the log device has already been written to the pool.
3407 vdev_log_state_valid(vdev_t *vd)
3409 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3413 for (int c = 0; c < vd->vdev_children; c++)
3414 if (vdev_log_state_valid(vd->vdev_child[c]))
3421 * Expand a vdev if possible.
3424 vdev_expand(vdev_t *vd, uint64_t txg)
3426 ASSERT(vd->vdev_top == vd);
3427 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3429 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3430 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3431 vdev_config_dirty(vd);
3439 vdev_split(vdev_t *vd)
3441 vdev_t *cvd, *pvd = vd->vdev_parent;
3443 vdev_remove_child(pvd, vd);
3444 vdev_compact_children(pvd);
3446 cvd = pvd->vdev_child[0];
3447 if (pvd->vdev_children == 1) {
3448 vdev_remove_parent(cvd);
3449 cvd->vdev_splitting = B_TRUE;
3451 vdev_propagate_state(cvd);
3455 vdev_deadman(vdev_t *vd)
3457 for (int c = 0; c < vd->vdev_children; c++) {
3458 vdev_t *cvd = vd->vdev_child[c];
3463 if (vd->vdev_ops->vdev_op_leaf) {
3464 vdev_queue_t *vq = &vd->vdev_queue;
3466 mutex_enter(&vq->vq_lock);
3467 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3468 spa_t *spa = vd->vdev_spa;
3473 * Look at the head of all the pending queues,
3474 * if any I/O has been outstanding for longer than
3475 * the spa_deadman_synctime we panic the system.
3477 fio = avl_first(&vq->vq_active_tree);
3478 delta = gethrtime() - fio->io_timestamp;
3479 if (delta > spa_deadman_synctime(spa)) {
3480 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3481 "delta %lluns, last io %lluns",
3482 fio->io_timestamp, delta,
3483 vq->vq_io_complete_ts);
3484 fm_panic("I/O to pool '%s' appears to be "
3485 "hung on vdev guid %llu at '%s'.",
3487 (long long unsigned int) vd->vdev_guid,
3491 mutex_exit(&vq->vq_lock);