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, 2018 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright 2016 Toomas Soome <tsoome@me.com>
29 * Copyright 2017 Joyent, Inc.
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/bpobj.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/uberblock_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/metaslab_impl.h>
44 #include <sys/space_map.h>
45 #include <sys/space_reftree.h>
48 #include <sys/fs/zfs.h>
51 #include <sys/dsl_scan.h>
53 #include <sys/trim_map.h>
54 #include <sys/vdev_initialize.h>
56 SYSCTL_DECL(_vfs_zfs);
57 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
60 * Virtual device management.
64 * The limit for ZFS to automatically increase a top-level vdev's ashift
65 * from logical ashift to physical ashift.
67 * Example: one or more 512B emulation child vdevs
68 * child->vdev_ashift = 9 (512 bytes)
69 * child->vdev_physical_ashift = 12 (4096 bytes)
70 * zfs_max_auto_ashift = 11 (2048 bytes)
71 * zfs_min_auto_ashift = 9 (512 bytes)
73 * On pool creation or the addition of a new top-level vdev, ZFS will
74 * increase the ashift of the top-level vdev to 2048 as limited by
75 * zfs_max_auto_ashift.
77 * Example: one or more 512B emulation child vdevs
78 * child->vdev_ashift = 9 (512 bytes)
79 * child->vdev_physical_ashift = 12 (4096 bytes)
80 * zfs_max_auto_ashift = 13 (8192 bytes)
81 * zfs_min_auto_ashift = 9 (512 bytes)
83 * On pool creation or the addition of a new top-level vdev, ZFS will
84 * increase the ashift of the top-level vdev to 4096 to match the
85 * max vdev_physical_ashift.
87 * Example: one or more 512B emulation child vdevs
88 * child->vdev_ashift = 9 (512 bytes)
89 * child->vdev_physical_ashift = 9 (512 bytes)
90 * zfs_max_auto_ashift = 13 (8192 bytes)
91 * zfs_min_auto_ashift = 12 (4096 bytes)
93 * On pool creation or the addition of a new top-level vdev, ZFS will
94 * increase the ashift of the top-level vdev to 4096 to match the
95 * zfs_min_auto_ashift.
97 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
98 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
101 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
106 val = zfs_max_auto_ashift;
107 err = sysctl_handle_64(oidp, &val, 0, req);
108 if (err != 0 || req->newptr == NULL)
111 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
114 zfs_max_auto_ashift = val;
118 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
119 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
120 sysctl_vfs_zfs_max_auto_ashift, "QU",
121 "Max ashift used when optimising for logical -> physical sectors size on "
122 "new top-level vdevs.");
125 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
130 val = zfs_min_auto_ashift;
131 err = sysctl_handle_64(oidp, &val, 0, req);
132 if (err != 0 || req->newptr == NULL)
135 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
138 zfs_min_auto_ashift = val;
142 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
143 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
144 sysctl_vfs_zfs_min_auto_ashift, "QU",
145 "Min ashift used when creating new top-level vdevs.");
147 static vdev_ops_t *vdev_ops_table[] = {
166 /* target number of metaslabs per top-level vdev */
167 int vdev_max_ms_count = 200;
168 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, max_ms_count, CTLFLAG_RDTUN,
169 &vdev_max_ms_count, 0,
170 "Maximum number of metaslabs per top-level vdev");
172 /* minimum number of metaslabs per top-level vdev */
173 int vdev_min_ms_count = 16;
174 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, min_ms_count, CTLFLAG_RDTUN,
175 &vdev_min_ms_count, 0,
176 "Minimum number of metaslabs per top-level vdev");
178 /* practical upper limit of total metaslabs per top-level vdev */
179 int vdev_ms_count_limit = 1ULL << 17;
181 /* lower limit for metaslab size (512M) */
182 int vdev_default_ms_shift = 29;
183 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, default_ms_shift, CTLFLAG_RDTUN,
184 &vdev_default_ms_shift, 0,
185 "Shift between vdev size and number of metaslabs");
187 /* upper limit for metaslab size (256G) */
188 int vdev_max_ms_shift = 38;
190 boolean_t vdev_validate_skip = B_FALSE;
193 * Since the DTL space map of a vdev is not expected to have a lot of
194 * entries, we default its block size to 4K.
196 int vdev_dtl_sm_blksz = (1 << 12);
197 SYSCTL_INT(_vfs_zfs, OID_AUTO, dtl_sm_blksz, CTLFLAG_RDTUN,
198 &vdev_dtl_sm_blksz, 0,
199 "Block size for DTL space map. Power of 2 and greater than 4096.");
202 * vdev-wide space maps that have lots of entries written to them at
203 * the end of each transaction can benefit from a higher I/O bandwidth
204 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
206 int vdev_standard_sm_blksz = (1 << 17);
207 SYSCTL_INT(_vfs_zfs, OID_AUTO, standard_sm_blksz, CTLFLAG_RDTUN,
208 &vdev_standard_sm_blksz, 0,
209 "Block size for standard space map. Power of 2 and greater than 4096.");
213 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
219 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
222 if (vd->vdev_path != NULL) {
223 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
226 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
227 vd->vdev_ops->vdev_op_type,
228 (u_longlong_t)vd->vdev_id,
229 (u_longlong_t)vd->vdev_guid, buf);
234 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
238 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
239 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
240 vd->vdev_ops->vdev_op_type);
244 switch (vd->vdev_state) {
245 case VDEV_STATE_UNKNOWN:
246 (void) snprintf(state, sizeof (state), "unknown");
248 case VDEV_STATE_CLOSED:
249 (void) snprintf(state, sizeof (state), "closed");
251 case VDEV_STATE_OFFLINE:
252 (void) snprintf(state, sizeof (state), "offline");
254 case VDEV_STATE_REMOVED:
255 (void) snprintf(state, sizeof (state), "removed");
257 case VDEV_STATE_CANT_OPEN:
258 (void) snprintf(state, sizeof (state), "can't open");
260 case VDEV_STATE_FAULTED:
261 (void) snprintf(state, sizeof (state), "faulted");
263 case VDEV_STATE_DEGRADED:
264 (void) snprintf(state, sizeof (state), "degraded");
266 case VDEV_STATE_HEALTHY:
267 (void) snprintf(state, sizeof (state), "healthy");
270 (void) snprintf(state, sizeof (state), "<state %u>",
271 (uint_t)vd->vdev_state);
274 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
275 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
276 vd->vdev_islog ? " (log)" : "",
277 (u_longlong_t)vd->vdev_guid,
278 vd->vdev_path ? vd->vdev_path : "N/A", state);
280 for (uint64_t i = 0; i < vd->vdev_children; i++)
281 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
285 * Given a vdev type, return the appropriate ops vector.
288 vdev_getops(const char *type)
290 vdev_ops_t *ops, **opspp;
292 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
293 if (strcmp(ops->vdev_op_type, type) == 0)
301 vdev_default_xlate(vdev_t *vd, const range_seg_t *in, range_seg_t *res)
303 res->rs_start = in->rs_start;
304 res->rs_end = in->rs_end;
308 * Default asize function: return the MAX of psize with the asize of
309 * all children. This is what's used by anything other than RAID-Z.
312 vdev_default_asize(vdev_t *vd, uint64_t psize)
314 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
317 for (int c = 0; c < vd->vdev_children; c++) {
318 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
319 asize = MAX(asize, csize);
326 * Get the minimum allocatable size. We define the allocatable size as
327 * the vdev's asize rounded to the nearest metaslab. This allows us to
328 * replace or attach devices which don't have the same physical size but
329 * can still satisfy the same number of allocations.
332 vdev_get_min_asize(vdev_t *vd)
334 vdev_t *pvd = vd->vdev_parent;
337 * If our parent is NULL (inactive spare or cache) or is the root,
338 * just return our own asize.
341 return (vd->vdev_asize);
344 * The top-level vdev just returns the allocatable size rounded
345 * to the nearest metaslab.
347 if (vd == vd->vdev_top)
348 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
351 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
352 * so each child must provide at least 1/Nth of its asize.
354 if (pvd->vdev_ops == &vdev_raidz_ops)
355 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
358 return (pvd->vdev_min_asize);
362 vdev_set_min_asize(vdev_t *vd)
364 vd->vdev_min_asize = vdev_get_min_asize(vd);
366 for (int c = 0; c < vd->vdev_children; c++)
367 vdev_set_min_asize(vd->vdev_child[c]);
371 vdev_lookup_top(spa_t *spa, uint64_t vdev)
373 vdev_t *rvd = spa->spa_root_vdev;
375 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
377 if (vdev < rvd->vdev_children) {
378 ASSERT(rvd->vdev_child[vdev] != NULL);
379 return (rvd->vdev_child[vdev]);
386 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
390 if (vd->vdev_guid == guid)
393 for (int c = 0; c < vd->vdev_children; c++)
394 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
402 vdev_count_leaves_impl(vdev_t *vd)
406 if (vd->vdev_ops->vdev_op_leaf)
409 for (int c = 0; c < vd->vdev_children; c++)
410 n += vdev_count_leaves_impl(vd->vdev_child[c]);
416 vdev_count_leaves(spa_t *spa)
418 return (vdev_count_leaves_impl(spa->spa_root_vdev));
422 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
424 size_t oldsize, newsize;
425 uint64_t id = cvd->vdev_id;
427 spa_t *spa = cvd->vdev_spa;
429 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
430 ASSERT(cvd->vdev_parent == NULL);
432 cvd->vdev_parent = pvd;
437 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
439 oldsize = pvd->vdev_children * sizeof (vdev_t *);
440 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
441 newsize = pvd->vdev_children * sizeof (vdev_t *);
443 newchild = kmem_zalloc(newsize, KM_SLEEP);
444 if (pvd->vdev_child != NULL) {
445 bcopy(pvd->vdev_child, newchild, oldsize);
446 kmem_free(pvd->vdev_child, oldsize);
449 pvd->vdev_child = newchild;
450 pvd->vdev_child[id] = cvd;
452 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
453 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
456 * Walk up all ancestors to update guid sum.
458 for (; pvd != NULL; pvd = pvd->vdev_parent)
459 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
463 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
466 uint_t id = cvd->vdev_id;
468 ASSERT(cvd->vdev_parent == pvd);
473 ASSERT(id < pvd->vdev_children);
474 ASSERT(pvd->vdev_child[id] == cvd);
476 pvd->vdev_child[id] = NULL;
477 cvd->vdev_parent = NULL;
479 for (c = 0; c < pvd->vdev_children; c++)
480 if (pvd->vdev_child[c])
483 if (c == pvd->vdev_children) {
484 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
485 pvd->vdev_child = NULL;
486 pvd->vdev_children = 0;
490 * Walk up all ancestors to update guid sum.
492 for (; pvd != NULL; pvd = pvd->vdev_parent)
493 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
497 * Remove any holes in the child array.
500 vdev_compact_children(vdev_t *pvd)
502 vdev_t **newchild, *cvd;
503 int oldc = pvd->vdev_children;
506 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
511 for (int c = newc = 0; c < oldc; c++)
512 if (pvd->vdev_child[c])
516 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
518 for (int c = newc = 0; c < oldc; c++) {
519 if ((cvd = pvd->vdev_child[c]) != NULL) {
520 newchild[newc] = cvd;
521 cvd->vdev_id = newc++;
528 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
529 pvd->vdev_child = newchild;
530 pvd->vdev_children = newc;
534 * Allocate and minimally initialize a vdev_t.
537 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
540 vdev_indirect_config_t *vic;
542 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
543 vic = &vd->vdev_indirect_config;
545 if (spa->spa_root_vdev == NULL) {
546 ASSERT(ops == &vdev_root_ops);
547 spa->spa_root_vdev = vd;
548 spa->spa_load_guid = spa_generate_guid(NULL);
551 if (guid == 0 && ops != &vdev_hole_ops) {
552 if (spa->spa_root_vdev == vd) {
554 * The root vdev's guid will also be the pool guid,
555 * which must be unique among all pools.
557 guid = spa_generate_guid(NULL);
560 * Any other vdev's guid must be unique within the pool.
562 guid = spa_generate_guid(spa);
564 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
569 vd->vdev_guid = guid;
570 vd->vdev_guid_sum = guid;
572 vd->vdev_state = VDEV_STATE_CLOSED;
573 vd->vdev_ishole = (ops == &vdev_hole_ops);
574 vic->vic_prev_indirect_vdev = UINT64_MAX;
576 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
577 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
578 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
580 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
581 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
582 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
583 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
584 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
585 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
586 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
587 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
588 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
590 for (int t = 0; t < DTL_TYPES; t++) {
591 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
593 txg_list_create(&vd->vdev_ms_list, spa,
594 offsetof(struct metaslab, ms_txg_node));
595 txg_list_create(&vd->vdev_dtl_list, spa,
596 offsetof(struct vdev, vdev_dtl_node));
597 vd->vdev_stat.vs_timestamp = gethrtime();
605 * Allocate a new vdev. The 'alloctype' is used to control whether we are
606 * creating a new vdev or loading an existing one - the behavior is slightly
607 * different for each case.
610 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
615 uint64_t guid = 0, islog, nparity;
617 vdev_indirect_config_t *vic;
619 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
621 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
622 return (SET_ERROR(EINVAL));
624 if ((ops = vdev_getops(type)) == NULL)
625 return (SET_ERROR(EINVAL));
628 * If this is a load, get the vdev guid from the nvlist.
629 * Otherwise, vdev_alloc_common() will generate one for us.
631 if (alloctype == VDEV_ALLOC_LOAD) {
634 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
636 return (SET_ERROR(EINVAL));
638 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
639 return (SET_ERROR(EINVAL));
640 } else if (alloctype == VDEV_ALLOC_SPARE) {
641 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
642 return (SET_ERROR(EINVAL));
643 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
644 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
645 return (SET_ERROR(EINVAL));
646 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
647 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
648 return (SET_ERROR(EINVAL));
652 * The first allocated vdev must be of type 'root'.
654 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
655 return (SET_ERROR(EINVAL));
658 * Determine whether we're a log vdev.
661 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
662 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
663 return (SET_ERROR(ENOTSUP));
665 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
666 return (SET_ERROR(ENOTSUP));
669 * Set the nparity property for RAID-Z vdevs.
672 if (ops == &vdev_raidz_ops) {
673 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
675 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
676 return (SET_ERROR(EINVAL));
678 * Previous versions could only support 1 or 2 parity
682 spa_version(spa) < SPA_VERSION_RAIDZ2)
683 return (SET_ERROR(ENOTSUP));
685 spa_version(spa) < SPA_VERSION_RAIDZ3)
686 return (SET_ERROR(ENOTSUP));
689 * We require the parity to be specified for SPAs that
690 * support multiple parity levels.
692 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
693 return (SET_ERROR(EINVAL));
695 * Otherwise, we default to 1 parity device for RAID-Z.
702 ASSERT(nparity != -1ULL);
704 vd = vdev_alloc_common(spa, id, guid, ops);
705 vic = &vd->vdev_indirect_config;
707 vd->vdev_islog = islog;
708 vd->vdev_nparity = nparity;
710 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
711 vd->vdev_path = spa_strdup(vd->vdev_path);
712 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
713 vd->vdev_devid = spa_strdup(vd->vdev_devid);
714 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
715 &vd->vdev_physpath) == 0)
716 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
717 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
718 vd->vdev_fru = spa_strdup(vd->vdev_fru);
721 * Set the whole_disk property. If it's not specified, leave the value
724 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
725 &vd->vdev_wholedisk) != 0)
726 vd->vdev_wholedisk = -1ULL;
728 ASSERT0(vic->vic_mapping_object);
729 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
730 &vic->vic_mapping_object);
731 ASSERT0(vic->vic_births_object);
732 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
733 &vic->vic_births_object);
734 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
735 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
736 &vic->vic_prev_indirect_vdev);
739 * Look for the 'not present' flag. This will only be set if the device
740 * was not present at the time of import.
742 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
743 &vd->vdev_not_present);
746 * Get the alignment requirement.
748 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
751 * Retrieve the vdev creation time.
753 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
757 * If we're a top-level vdev, try to load the allocation parameters.
759 if (parent && !parent->vdev_parent &&
760 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
761 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
763 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
765 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
767 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
769 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
772 ASSERT0(vd->vdev_top_zap);
775 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
776 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
777 alloctype == VDEV_ALLOC_ADD ||
778 alloctype == VDEV_ALLOC_SPLIT ||
779 alloctype == VDEV_ALLOC_ROOTPOOL);
780 vd->vdev_mg = metaslab_group_create(islog ?
781 spa_log_class(spa) : spa_normal_class(spa), vd,
782 spa->spa_alloc_count);
785 if (vd->vdev_ops->vdev_op_leaf &&
786 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
787 (void) nvlist_lookup_uint64(nv,
788 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
790 ASSERT0(vd->vdev_leaf_zap);
794 * If we're a leaf vdev, try to load the DTL object and other state.
797 if (vd->vdev_ops->vdev_op_leaf &&
798 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
799 alloctype == VDEV_ALLOC_ROOTPOOL)) {
800 if (alloctype == VDEV_ALLOC_LOAD) {
801 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
802 &vd->vdev_dtl_object);
803 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
807 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
810 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
811 &spare) == 0 && spare)
815 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
818 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
819 &vd->vdev_resilver_txg);
822 * When importing a pool, we want to ignore the persistent fault
823 * state, as the diagnosis made on another system may not be
824 * valid in the current context. Local vdevs will
825 * remain in the faulted state.
827 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
828 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
830 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
832 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
835 if (vd->vdev_faulted || vd->vdev_degraded) {
839 VDEV_AUX_ERR_EXCEEDED;
840 if (nvlist_lookup_string(nv,
841 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
842 strcmp(aux, "external") == 0)
843 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
849 * Add ourselves to the parent's list of children.
851 vdev_add_child(parent, vd);
859 vdev_free(vdev_t *vd)
861 spa_t *spa = vd->vdev_spa;
862 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
865 * Scan queues are normally destroyed at the end of a scan. If the
866 * queue exists here, that implies the vdev is being removed while
867 * the scan is still running.
869 if (vd->vdev_scan_io_queue != NULL) {
870 mutex_enter(&vd->vdev_scan_io_queue_lock);
871 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
872 vd->vdev_scan_io_queue = NULL;
873 mutex_exit(&vd->vdev_scan_io_queue_lock);
877 * vdev_free() implies closing the vdev first. This is simpler than
878 * trying to ensure complicated semantics for all callers.
882 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
883 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
888 for (int c = 0; c < vd->vdev_children; c++)
889 vdev_free(vd->vdev_child[c]);
891 ASSERT(vd->vdev_child == NULL);
892 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
893 ASSERT(vd->vdev_initialize_thread == NULL);
896 * Discard allocation state.
898 if (vd->vdev_mg != NULL) {
899 vdev_metaslab_fini(vd);
900 metaslab_group_destroy(vd->vdev_mg);
903 ASSERT0(vd->vdev_stat.vs_space);
904 ASSERT0(vd->vdev_stat.vs_dspace);
905 ASSERT0(vd->vdev_stat.vs_alloc);
908 * Remove this vdev from its parent's child list.
910 vdev_remove_child(vd->vdev_parent, vd);
912 ASSERT(vd->vdev_parent == NULL);
915 * Clean up vdev structure.
921 spa_strfree(vd->vdev_path);
923 spa_strfree(vd->vdev_devid);
924 if (vd->vdev_physpath)
925 spa_strfree(vd->vdev_physpath);
927 spa_strfree(vd->vdev_fru);
929 if (vd->vdev_isspare)
930 spa_spare_remove(vd);
931 if (vd->vdev_isl2cache)
932 spa_l2cache_remove(vd);
934 txg_list_destroy(&vd->vdev_ms_list);
935 txg_list_destroy(&vd->vdev_dtl_list);
937 mutex_enter(&vd->vdev_dtl_lock);
938 space_map_close(vd->vdev_dtl_sm);
939 for (int t = 0; t < DTL_TYPES; t++) {
940 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
941 range_tree_destroy(vd->vdev_dtl[t]);
943 mutex_exit(&vd->vdev_dtl_lock);
945 EQUIV(vd->vdev_indirect_births != NULL,
946 vd->vdev_indirect_mapping != NULL);
947 if (vd->vdev_indirect_births != NULL) {
948 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
949 vdev_indirect_births_close(vd->vdev_indirect_births);
952 if (vd->vdev_obsolete_sm != NULL) {
953 ASSERT(vd->vdev_removing ||
954 vd->vdev_ops == &vdev_indirect_ops);
955 space_map_close(vd->vdev_obsolete_sm);
956 vd->vdev_obsolete_sm = NULL;
958 range_tree_destroy(vd->vdev_obsolete_segments);
959 rw_destroy(&vd->vdev_indirect_rwlock);
960 mutex_destroy(&vd->vdev_obsolete_lock);
962 mutex_destroy(&vd->vdev_queue_lock);
963 mutex_destroy(&vd->vdev_dtl_lock);
964 mutex_destroy(&vd->vdev_stat_lock);
965 mutex_destroy(&vd->vdev_probe_lock);
966 mutex_destroy(&vd->vdev_scan_io_queue_lock);
967 mutex_destroy(&vd->vdev_initialize_lock);
968 mutex_destroy(&vd->vdev_initialize_io_lock);
969 cv_destroy(&vd->vdev_initialize_io_cv);
970 cv_destroy(&vd->vdev_initialize_cv);
972 if (vd == spa->spa_root_vdev)
973 spa->spa_root_vdev = NULL;
975 kmem_free(vd, sizeof (vdev_t));
979 * Transfer top-level vdev state from svd to tvd.
982 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
984 spa_t *spa = svd->vdev_spa;
989 ASSERT(tvd == tvd->vdev_top);
991 tvd->vdev_ms_array = svd->vdev_ms_array;
992 tvd->vdev_ms_shift = svd->vdev_ms_shift;
993 tvd->vdev_ms_count = svd->vdev_ms_count;
994 tvd->vdev_top_zap = svd->vdev_top_zap;
996 svd->vdev_ms_array = 0;
997 svd->vdev_ms_shift = 0;
998 svd->vdev_ms_count = 0;
999 svd->vdev_top_zap = 0;
1002 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1003 tvd->vdev_mg = svd->vdev_mg;
1004 tvd->vdev_ms = svd->vdev_ms;
1006 svd->vdev_mg = NULL;
1007 svd->vdev_ms = NULL;
1009 if (tvd->vdev_mg != NULL)
1010 tvd->vdev_mg->mg_vd = tvd;
1012 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1013 svd->vdev_checkpoint_sm = NULL;
1015 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1016 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1017 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1019 svd->vdev_stat.vs_alloc = 0;
1020 svd->vdev_stat.vs_space = 0;
1021 svd->vdev_stat.vs_dspace = 0;
1024 * State which may be set on a top-level vdev that's in the
1025 * process of being removed.
1027 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1028 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1029 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1030 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1031 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1032 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1033 ASSERT0(tvd->vdev_removing);
1034 tvd->vdev_removing = svd->vdev_removing;
1035 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1036 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1037 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1038 range_tree_swap(&svd->vdev_obsolete_segments,
1039 &tvd->vdev_obsolete_segments);
1040 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1041 svd->vdev_indirect_config.vic_mapping_object = 0;
1042 svd->vdev_indirect_config.vic_births_object = 0;
1043 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1044 svd->vdev_indirect_mapping = NULL;
1045 svd->vdev_indirect_births = NULL;
1046 svd->vdev_obsolete_sm = NULL;
1047 svd->vdev_removing = 0;
1049 for (t = 0; t < TXG_SIZE; t++) {
1050 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1051 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1052 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1053 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1054 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1055 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1058 if (list_link_active(&svd->vdev_config_dirty_node)) {
1059 vdev_config_clean(svd);
1060 vdev_config_dirty(tvd);
1063 if (list_link_active(&svd->vdev_state_dirty_node)) {
1064 vdev_state_clean(svd);
1065 vdev_state_dirty(tvd);
1068 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1069 svd->vdev_deflate_ratio = 0;
1071 tvd->vdev_islog = svd->vdev_islog;
1072 svd->vdev_islog = 0;
1074 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1078 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1085 for (int c = 0; c < vd->vdev_children; c++)
1086 vdev_top_update(tvd, vd->vdev_child[c]);
1090 * Add a mirror/replacing vdev above an existing vdev.
1093 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1095 spa_t *spa = cvd->vdev_spa;
1096 vdev_t *pvd = cvd->vdev_parent;
1099 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1101 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1103 mvd->vdev_asize = cvd->vdev_asize;
1104 mvd->vdev_min_asize = cvd->vdev_min_asize;
1105 mvd->vdev_max_asize = cvd->vdev_max_asize;
1106 mvd->vdev_psize = cvd->vdev_psize;
1107 mvd->vdev_ashift = cvd->vdev_ashift;
1108 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1109 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1110 mvd->vdev_state = cvd->vdev_state;
1111 mvd->vdev_crtxg = cvd->vdev_crtxg;
1113 vdev_remove_child(pvd, cvd);
1114 vdev_add_child(pvd, mvd);
1115 cvd->vdev_id = mvd->vdev_children;
1116 vdev_add_child(mvd, cvd);
1117 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1119 if (mvd == mvd->vdev_top)
1120 vdev_top_transfer(cvd, mvd);
1126 * Remove a 1-way mirror/replacing vdev from the tree.
1129 vdev_remove_parent(vdev_t *cvd)
1131 vdev_t *mvd = cvd->vdev_parent;
1132 vdev_t *pvd = mvd->vdev_parent;
1134 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1136 ASSERT(mvd->vdev_children == 1);
1137 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1138 mvd->vdev_ops == &vdev_replacing_ops ||
1139 mvd->vdev_ops == &vdev_spare_ops);
1140 cvd->vdev_ashift = mvd->vdev_ashift;
1141 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1142 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1144 vdev_remove_child(mvd, cvd);
1145 vdev_remove_child(pvd, mvd);
1148 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1149 * Otherwise, we could have detached an offline device, and when we
1150 * go to import the pool we'll think we have two top-level vdevs,
1151 * instead of a different version of the same top-level vdev.
1153 if (mvd->vdev_top == mvd) {
1154 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1155 cvd->vdev_orig_guid = cvd->vdev_guid;
1156 cvd->vdev_guid += guid_delta;
1157 cvd->vdev_guid_sum += guid_delta;
1159 cvd->vdev_id = mvd->vdev_id;
1160 vdev_add_child(pvd, cvd);
1161 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1163 if (cvd == cvd->vdev_top)
1164 vdev_top_transfer(mvd, cvd);
1166 ASSERT(mvd->vdev_children == 0);
1171 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1173 spa_t *spa = vd->vdev_spa;
1174 objset_t *mos = spa->spa_meta_objset;
1176 uint64_t oldc = vd->vdev_ms_count;
1177 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1181 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1184 * This vdev is not being allocated from yet or is a hole.
1186 if (vd->vdev_ms_shift == 0)
1189 ASSERT(!vd->vdev_ishole);
1191 ASSERT(oldc <= newc);
1193 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1196 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1197 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1201 vd->vdev_ms_count = newc;
1202 for (m = oldc; m < newc; m++) {
1203 uint64_t object = 0;
1206 * vdev_ms_array may be 0 if we are creating the "fake"
1207 * metaslabs for an indirect vdev for zdb's leak detection.
1208 * See zdb_leak_init().
1210 if (txg == 0 && vd->vdev_ms_array != 0) {
1211 error = dmu_read(mos, vd->vdev_ms_array,
1212 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1215 vdev_dbgmsg(vd, "unable to read the metaslab "
1216 "array [error=%d]", error);
1221 error = metaslab_init(vd->vdev_mg, m, object, txg,
1224 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1231 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1234 * If the vdev is being removed we don't activate
1235 * the metaslabs since we want to ensure that no new
1236 * allocations are performed on this device.
1238 if (oldc == 0 && !vd->vdev_removing)
1239 metaslab_group_activate(vd->vdev_mg);
1242 spa_config_exit(spa, SCL_ALLOC, FTAG);
1248 vdev_metaslab_fini(vdev_t *vd)
1250 if (vd->vdev_checkpoint_sm != NULL) {
1251 ASSERT(spa_feature_is_active(vd->vdev_spa,
1252 SPA_FEATURE_POOL_CHECKPOINT));
1253 space_map_close(vd->vdev_checkpoint_sm);
1255 * Even though we close the space map, we need to set its
1256 * pointer to NULL. The reason is that vdev_metaslab_fini()
1257 * may be called multiple times for certain operations
1258 * (i.e. when destroying a pool) so we need to ensure that
1259 * this clause never executes twice. This logic is similar
1260 * to the one used for the vdev_ms clause below.
1262 vd->vdev_checkpoint_sm = NULL;
1265 if (vd->vdev_ms != NULL) {
1266 uint64_t count = vd->vdev_ms_count;
1268 metaslab_group_passivate(vd->vdev_mg);
1269 for (uint64_t m = 0; m < count; m++) {
1270 metaslab_t *msp = vd->vdev_ms[m];
1275 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1278 vd->vdev_ms_count = 0;
1280 ASSERT0(vd->vdev_ms_count);
1283 typedef struct vdev_probe_stats {
1284 boolean_t vps_readable;
1285 boolean_t vps_writeable;
1287 } vdev_probe_stats_t;
1290 vdev_probe_done(zio_t *zio)
1292 spa_t *spa = zio->io_spa;
1293 vdev_t *vd = zio->io_vd;
1294 vdev_probe_stats_t *vps = zio->io_private;
1296 ASSERT(vd->vdev_probe_zio != NULL);
1298 if (zio->io_type == ZIO_TYPE_READ) {
1299 if (zio->io_error == 0)
1300 vps->vps_readable = 1;
1301 if (zio->io_error == 0 && spa_writeable(spa)) {
1302 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1303 zio->io_offset, zio->io_size, zio->io_abd,
1304 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1305 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1307 abd_free(zio->io_abd);
1309 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1310 if (zio->io_error == 0)
1311 vps->vps_writeable = 1;
1312 abd_free(zio->io_abd);
1313 } else if (zio->io_type == ZIO_TYPE_NULL) {
1316 vd->vdev_cant_read |= !vps->vps_readable;
1317 vd->vdev_cant_write |= !vps->vps_writeable;
1319 if (vdev_readable(vd) &&
1320 (vdev_writeable(vd) || !spa_writeable(spa))) {
1323 ASSERT(zio->io_error != 0);
1324 vdev_dbgmsg(vd, "failed probe");
1325 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1326 spa, vd, NULL, 0, 0);
1327 zio->io_error = SET_ERROR(ENXIO);
1330 mutex_enter(&vd->vdev_probe_lock);
1331 ASSERT(vd->vdev_probe_zio == zio);
1332 vd->vdev_probe_zio = NULL;
1333 mutex_exit(&vd->vdev_probe_lock);
1335 zio_link_t *zl = NULL;
1336 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1337 if (!vdev_accessible(vd, pio))
1338 pio->io_error = SET_ERROR(ENXIO);
1340 kmem_free(vps, sizeof (*vps));
1345 * Determine whether this device is accessible.
1347 * Read and write to several known locations: the pad regions of each
1348 * vdev label but the first, which we leave alone in case it contains
1352 vdev_probe(vdev_t *vd, zio_t *zio)
1354 spa_t *spa = vd->vdev_spa;
1355 vdev_probe_stats_t *vps = NULL;
1358 ASSERT(vd->vdev_ops->vdev_op_leaf);
1361 * Don't probe the probe.
1363 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1367 * To prevent 'probe storms' when a device fails, we create
1368 * just one probe i/o at a time. All zios that want to probe
1369 * this vdev will become parents of the probe io.
1371 mutex_enter(&vd->vdev_probe_lock);
1373 if ((pio = vd->vdev_probe_zio) == NULL) {
1374 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1376 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1377 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1380 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1382 * vdev_cant_read and vdev_cant_write can only
1383 * transition from TRUE to FALSE when we have the
1384 * SCL_ZIO lock as writer; otherwise they can only
1385 * transition from FALSE to TRUE. This ensures that
1386 * any zio looking at these values can assume that
1387 * failures persist for the life of the I/O. That's
1388 * important because when a device has intermittent
1389 * connectivity problems, we want to ensure that
1390 * they're ascribed to the device (ENXIO) and not
1393 * Since we hold SCL_ZIO as writer here, clear both
1394 * values so the probe can reevaluate from first
1397 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1398 vd->vdev_cant_read = B_FALSE;
1399 vd->vdev_cant_write = B_FALSE;
1402 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1403 vdev_probe_done, vps,
1404 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1407 * We can't change the vdev state in this context, so we
1408 * kick off an async task to do it on our behalf.
1411 vd->vdev_probe_wanted = B_TRUE;
1412 spa_async_request(spa, SPA_ASYNC_PROBE);
1417 zio_add_child(zio, pio);
1419 mutex_exit(&vd->vdev_probe_lock);
1422 ASSERT(zio != NULL);
1426 for (int l = 1; l < VDEV_LABELS; l++) {
1427 zio_nowait(zio_read_phys(pio, vd,
1428 vdev_label_offset(vd->vdev_psize, l,
1429 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1430 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1431 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1432 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1443 vdev_open_child(void *arg)
1447 vd->vdev_open_thread = curthread;
1448 vd->vdev_open_error = vdev_open(vd);
1449 vd->vdev_open_thread = NULL;
1453 vdev_uses_zvols(vdev_t *vd)
1455 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1456 strlen(ZVOL_DIR)) == 0)
1458 for (int c = 0; c < vd->vdev_children; c++)
1459 if (vdev_uses_zvols(vd->vdev_child[c]))
1465 vdev_open_children(vdev_t *vd)
1468 int children = vd->vdev_children;
1471 * in order to handle pools on top of zvols, do the opens
1472 * in a single thread so that the same thread holds the
1473 * spa_namespace_lock
1475 if (B_TRUE || vdev_uses_zvols(vd)) {
1476 for (int c = 0; c < children; c++)
1477 vd->vdev_child[c]->vdev_open_error =
1478 vdev_open(vd->vdev_child[c]);
1481 tq = taskq_create("vdev_open", children, minclsyspri,
1482 children, children, TASKQ_PREPOPULATE);
1484 for (int c = 0; c < children; c++)
1485 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1492 * Compute the raidz-deflation ratio. Note, we hard-code
1493 * in 128k (1 << 17) because it is the "typical" blocksize.
1494 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1495 * otherwise it would inconsistently account for existing bp's.
1498 vdev_set_deflate_ratio(vdev_t *vd)
1500 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1501 vd->vdev_deflate_ratio = (1 << 17) /
1502 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1507 * Prepare a virtual device for access.
1510 vdev_open(vdev_t *vd)
1512 spa_t *spa = vd->vdev_spa;
1515 uint64_t max_osize = 0;
1516 uint64_t asize, max_asize, psize;
1517 uint64_t logical_ashift = 0;
1518 uint64_t physical_ashift = 0;
1520 ASSERT(vd->vdev_open_thread == curthread ||
1521 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1522 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1523 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1524 vd->vdev_state == VDEV_STATE_OFFLINE);
1526 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1527 vd->vdev_cant_read = B_FALSE;
1528 vd->vdev_cant_write = B_FALSE;
1529 vd->vdev_notrim = B_FALSE;
1530 vd->vdev_min_asize = vdev_get_min_asize(vd);
1533 * If this vdev is not removed, check its fault status. If it's
1534 * faulted, bail out of the open.
1536 if (!vd->vdev_removed && vd->vdev_faulted) {
1537 ASSERT(vd->vdev_children == 0);
1538 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1539 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1540 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1541 vd->vdev_label_aux);
1542 return (SET_ERROR(ENXIO));
1543 } else if (vd->vdev_offline) {
1544 ASSERT(vd->vdev_children == 0);
1545 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1546 return (SET_ERROR(ENXIO));
1549 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1550 &logical_ashift, &physical_ashift);
1553 * Reset the vdev_reopening flag so that we actually close
1554 * the vdev on error.
1556 vd->vdev_reopening = B_FALSE;
1557 if (zio_injection_enabled && error == 0)
1558 error = zio_handle_device_injection(vd, NULL, ENXIO);
1561 if (vd->vdev_removed &&
1562 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1563 vd->vdev_removed = B_FALSE;
1565 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1566 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1567 vd->vdev_stat.vs_aux);
1569 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1570 vd->vdev_stat.vs_aux);
1575 vd->vdev_removed = B_FALSE;
1578 * Recheck the faulted flag now that we have confirmed that
1579 * the vdev is accessible. If we're faulted, bail.
1581 if (vd->vdev_faulted) {
1582 ASSERT(vd->vdev_children == 0);
1583 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1584 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1585 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1586 vd->vdev_label_aux);
1587 return (SET_ERROR(ENXIO));
1590 if (vd->vdev_degraded) {
1591 ASSERT(vd->vdev_children == 0);
1592 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1593 VDEV_AUX_ERR_EXCEEDED);
1595 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1599 * For hole or missing vdevs we just return success.
1601 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1604 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1605 trim_map_create(vd);
1607 for (int c = 0; c < vd->vdev_children; c++) {
1608 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1609 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1615 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1616 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1618 if (vd->vdev_children == 0) {
1619 if (osize < SPA_MINDEVSIZE) {
1620 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1621 VDEV_AUX_TOO_SMALL);
1622 return (SET_ERROR(EOVERFLOW));
1625 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1626 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1627 VDEV_LABEL_END_SIZE);
1629 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1630 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1631 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1632 VDEV_AUX_TOO_SMALL);
1633 return (SET_ERROR(EOVERFLOW));
1637 max_asize = max_osize;
1640 vd->vdev_psize = psize;
1643 * Make sure the allocatable size hasn't shrunk too much.
1645 if (asize < vd->vdev_min_asize) {
1646 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1647 VDEV_AUX_BAD_LABEL);
1648 return (SET_ERROR(EINVAL));
1651 vd->vdev_physical_ashift =
1652 MAX(physical_ashift, vd->vdev_physical_ashift);
1653 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1654 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1656 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1657 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1658 VDEV_AUX_ASHIFT_TOO_BIG);
1662 if (vd->vdev_asize == 0) {
1664 * This is the first-ever open, so use the computed values.
1665 * For testing purposes, a higher ashift can be requested.
1667 vd->vdev_asize = asize;
1668 vd->vdev_max_asize = max_asize;
1671 * Make sure the alignment requirement hasn't increased.
1673 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1674 vd->vdev_ops->vdev_op_leaf) {
1675 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1676 VDEV_AUX_BAD_LABEL);
1679 vd->vdev_max_asize = max_asize;
1683 * If all children are healthy we update asize if either:
1684 * The asize has increased, due to a device expansion caused by dynamic
1685 * LUN growth or vdev replacement, and automatic expansion is enabled;
1686 * making the additional space available.
1688 * The asize has decreased, due to a device shrink usually caused by a
1689 * vdev replace with a smaller device. This ensures that calculations
1690 * based of max_asize and asize e.g. esize are always valid. It's safe
1691 * to do this as we've already validated that asize is greater than
1694 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1695 ((asize > vd->vdev_asize &&
1696 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1697 (asize < vd->vdev_asize)))
1698 vd->vdev_asize = asize;
1700 vdev_set_min_asize(vd);
1703 * Ensure we can issue some IO before declaring the
1704 * vdev open for business.
1706 if (vd->vdev_ops->vdev_op_leaf &&
1707 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1708 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1709 VDEV_AUX_ERR_EXCEEDED);
1714 * Track the min and max ashift values for normal data devices.
1716 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1717 !vd->vdev_islog && vd->vdev_aux == NULL) {
1718 if (vd->vdev_ashift > spa->spa_max_ashift)
1719 spa->spa_max_ashift = vd->vdev_ashift;
1720 if (vd->vdev_ashift < spa->spa_min_ashift)
1721 spa->spa_min_ashift = vd->vdev_ashift;
1725 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1726 * resilver. But don't do this if we are doing a reopen for a scrub,
1727 * since this would just restart the scrub we are already doing.
1729 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1730 vdev_resilver_needed(vd, NULL, NULL))
1731 spa_async_request(spa, SPA_ASYNC_RESILVER);
1737 * Called once the vdevs are all opened, this routine validates the label
1738 * contents. This needs to be done before vdev_load() so that we don't
1739 * inadvertently do repair I/Os to the wrong device.
1741 * This function will only return failure if one of the vdevs indicates that it
1742 * has since been destroyed or exported. This is only possible if
1743 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1744 * will be updated but the function will return 0.
1747 vdev_validate(vdev_t *vd)
1749 spa_t *spa = vd->vdev_spa;
1751 uint64_t guid = 0, aux_guid = 0, top_guid;
1756 if (vdev_validate_skip)
1759 for (uint64_t c = 0; c < vd->vdev_children; c++)
1760 if (vdev_validate(vd->vdev_child[c]) != 0)
1761 return (SET_ERROR(EBADF));
1764 * If the device has already failed, or was marked offline, don't do
1765 * any further validation. Otherwise, label I/O will fail and we will
1766 * overwrite the previous state.
1768 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1772 * If we are performing an extreme rewind, we allow for a label that
1773 * was modified at a point after the current txg.
1774 * If config lock is not held do not check for the txg. spa_sync could
1775 * be updating the vdev's label before updating spa_last_synced_txg.
1777 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1778 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1781 txg = spa_last_synced_txg(spa);
1783 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1784 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1785 VDEV_AUX_BAD_LABEL);
1786 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
1787 "txg %llu", (u_longlong_t)txg);
1792 * Determine if this vdev has been split off into another
1793 * pool. If so, then refuse to open it.
1795 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1796 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1797 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1798 VDEV_AUX_SPLIT_POOL);
1800 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1804 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1805 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1806 VDEV_AUX_CORRUPT_DATA);
1808 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1809 ZPOOL_CONFIG_POOL_GUID);
1814 * If config is not trusted then ignore the spa guid check. This is
1815 * necessary because if the machine crashed during a re-guid the new
1816 * guid might have been written to all of the vdev labels, but not the
1817 * cached config. The check will be performed again once we have the
1818 * trusted config from the MOS.
1820 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1821 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1822 VDEV_AUX_CORRUPT_DATA);
1824 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
1825 "match config (%llu != %llu)", (u_longlong_t)guid,
1826 (u_longlong_t)spa_guid(spa));
1830 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1831 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1835 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
1836 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1837 VDEV_AUX_CORRUPT_DATA);
1839 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1844 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
1846 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1847 VDEV_AUX_CORRUPT_DATA);
1849 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1850 ZPOOL_CONFIG_TOP_GUID);
1855 * If this vdev just became a top-level vdev because its sibling was
1856 * detached, it will have adopted the parent's vdev guid -- but the
1857 * label may or may not be on disk yet. Fortunately, either version
1858 * of the label will have the same top guid, so if we're a top-level
1859 * vdev, we can safely compare to that instead.
1860 * However, if the config comes from a cachefile that failed to update
1861 * after the detach, a top-level vdev will appear as a non top-level
1862 * vdev in the config. Also relax the constraints if we perform an
1865 * If we split this vdev off instead, then we also check the
1866 * original pool's guid. We don't want to consider the vdev
1867 * corrupt if it is partway through a split operation.
1869 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
1870 boolean_t mismatch = B_FALSE;
1871 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
1872 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
1875 if (vd->vdev_guid != top_guid &&
1876 vd->vdev_top->vdev_guid != guid)
1881 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1882 VDEV_AUX_CORRUPT_DATA);
1884 vdev_dbgmsg(vd, "vdev_validate: config guid "
1885 "doesn't match label guid");
1886 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
1887 (u_longlong_t)vd->vdev_guid,
1888 (u_longlong_t)vd->vdev_top->vdev_guid);
1889 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
1890 "aux_guid %llu", (u_longlong_t)guid,
1891 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
1896 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1898 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1899 VDEV_AUX_CORRUPT_DATA);
1901 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1902 ZPOOL_CONFIG_POOL_STATE);
1909 * If this is a verbatim import, no need to check the
1910 * state of the pool.
1912 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1913 spa_load_state(spa) == SPA_LOAD_OPEN &&
1914 state != POOL_STATE_ACTIVE) {
1915 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
1916 "for spa %s", (u_longlong_t)state, spa->spa_name);
1917 return (SET_ERROR(EBADF));
1921 * If we were able to open and validate a vdev that was
1922 * previously marked permanently unavailable, clear that state
1925 if (vd->vdev_not_present)
1926 vd->vdev_not_present = 0;
1932 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
1934 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
1935 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
1936 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1937 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
1938 dvd->vdev_path, svd->vdev_path);
1939 spa_strfree(dvd->vdev_path);
1940 dvd->vdev_path = spa_strdup(svd->vdev_path);
1942 } else if (svd->vdev_path != NULL) {
1943 dvd->vdev_path = spa_strdup(svd->vdev_path);
1944 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1945 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
1950 * Recursively copy vdev paths from one vdev to another. Source and destination
1951 * vdev trees must have same geometry otherwise return error. Intended to copy
1952 * paths from userland config into MOS config.
1955 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
1957 if ((svd->vdev_ops == &vdev_missing_ops) ||
1958 (svd->vdev_ishole && dvd->vdev_ishole) ||
1959 (dvd->vdev_ops == &vdev_indirect_ops))
1962 if (svd->vdev_ops != dvd->vdev_ops) {
1963 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
1964 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
1965 return (SET_ERROR(EINVAL));
1968 if (svd->vdev_guid != dvd->vdev_guid) {
1969 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
1970 "%llu)", (u_longlong_t)svd->vdev_guid,
1971 (u_longlong_t)dvd->vdev_guid);
1972 return (SET_ERROR(EINVAL));
1975 if (svd->vdev_children != dvd->vdev_children) {
1976 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
1977 "%llu != %llu", (u_longlong_t)svd->vdev_children,
1978 (u_longlong_t)dvd->vdev_children);
1979 return (SET_ERROR(EINVAL));
1982 for (uint64_t i = 0; i < svd->vdev_children; i++) {
1983 int error = vdev_copy_path_strict(svd->vdev_child[i],
1984 dvd->vdev_child[i]);
1989 if (svd->vdev_ops->vdev_op_leaf)
1990 vdev_copy_path_impl(svd, dvd);
1996 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
1998 ASSERT(stvd->vdev_top == stvd);
1999 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2001 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2002 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2005 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2009 * The idea here is that while a vdev can shift positions within
2010 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2011 * step outside of it.
2013 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2015 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2018 ASSERT(vd->vdev_ops->vdev_op_leaf);
2020 vdev_copy_path_impl(vd, dvd);
2024 * Recursively copy vdev paths from one root vdev to another. Source and
2025 * destination vdev trees may differ in geometry. For each destination leaf
2026 * vdev, search a vdev with the same guid and top vdev id in the source.
2027 * Intended to copy paths from userland config into MOS config.
2030 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2032 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2033 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2034 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2036 for (uint64_t i = 0; i < children; i++) {
2037 vdev_copy_path_search(srvd->vdev_child[i],
2038 drvd->vdev_child[i]);
2043 * Close a virtual device.
2046 vdev_close(vdev_t *vd)
2048 spa_t *spa = vd->vdev_spa;
2049 vdev_t *pvd = vd->vdev_parent;
2051 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2054 * If our parent is reopening, then we are as well, unless we are
2057 if (pvd != NULL && pvd->vdev_reopening)
2058 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2060 vd->vdev_ops->vdev_op_close(vd);
2062 vdev_cache_purge(vd);
2064 if (vd->vdev_ops->vdev_op_leaf)
2065 trim_map_destroy(vd);
2068 * We record the previous state before we close it, so that if we are
2069 * doing a reopen(), we don't generate FMA ereports if we notice that
2070 * it's still faulted.
2072 vd->vdev_prevstate = vd->vdev_state;
2074 if (vd->vdev_offline)
2075 vd->vdev_state = VDEV_STATE_OFFLINE;
2077 vd->vdev_state = VDEV_STATE_CLOSED;
2078 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2082 vdev_hold(vdev_t *vd)
2084 spa_t *spa = vd->vdev_spa;
2086 ASSERT(spa_is_root(spa));
2087 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2090 for (int c = 0; c < vd->vdev_children; c++)
2091 vdev_hold(vd->vdev_child[c]);
2093 if (vd->vdev_ops->vdev_op_leaf)
2094 vd->vdev_ops->vdev_op_hold(vd);
2098 vdev_rele(vdev_t *vd)
2100 spa_t *spa = vd->vdev_spa;
2102 ASSERT(spa_is_root(spa));
2103 for (int c = 0; c < vd->vdev_children; c++)
2104 vdev_rele(vd->vdev_child[c]);
2106 if (vd->vdev_ops->vdev_op_leaf)
2107 vd->vdev_ops->vdev_op_rele(vd);
2111 * Reopen all interior vdevs and any unopened leaves. We don't actually
2112 * reopen leaf vdevs which had previously been opened as they might deadlock
2113 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2114 * If the leaf has never been opened then open it, as usual.
2117 vdev_reopen(vdev_t *vd)
2119 spa_t *spa = vd->vdev_spa;
2121 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2123 /* set the reopening flag unless we're taking the vdev offline */
2124 vd->vdev_reopening = !vd->vdev_offline;
2126 (void) vdev_open(vd);
2129 * Call vdev_validate() here to make sure we have the same device.
2130 * Otherwise, a device with an invalid label could be successfully
2131 * opened in response to vdev_reopen().
2134 (void) vdev_validate_aux(vd);
2135 if (vdev_readable(vd) && vdev_writeable(vd) &&
2136 vd->vdev_aux == &spa->spa_l2cache &&
2137 !l2arc_vdev_present(vd))
2138 l2arc_add_vdev(spa, vd);
2140 (void) vdev_validate(vd);
2144 * Reassess parent vdev's health.
2146 vdev_propagate_state(vd);
2150 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2155 * Normally, partial opens (e.g. of a mirror) are allowed.
2156 * For a create, however, we want to fail the request if
2157 * there are any components we can't open.
2159 error = vdev_open(vd);
2161 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2163 return (error ? error : ENXIO);
2167 * Recursively load DTLs and initialize all labels.
2169 if ((error = vdev_dtl_load(vd)) != 0 ||
2170 (error = vdev_label_init(vd, txg, isreplacing ?
2171 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2180 vdev_metaslab_set_size(vdev_t *vd)
2182 uint64_t asize = vd->vdev_asize;
2183 uint64_t ms_count = asize >> vdev_default_ms_shift;
2187 * There are two dimensions to the metaslab sizing calculation:
2188 * the size of the metaslab and the count of metaslabs per vdev.
2189 * In general, we aim for vdev_max_ms_count (200) metaslabs. The
2190 * range of the dimensions are as follows:
2192 * 2^29 <= ms_size <= 2^38
2193 * 16 <= ms_count <= 131,072
2195 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2196 * at least 512MB (2^29) to minimize fragmentation effects when
2197 * testing with smaller devices. However, the count constraint
2198 * of at least 16 metaslabs will override this minimum size goal.
2200 * On the upper end of vdev sizes, we aim for a maximum metaslab
2201 * size of 256GB. However, we will cap the total count to 2^17
2202 * metaslabs to keep our memory footprint in check.
2204 * The net effect of applying above constrains is summarized below.
2206 * vdev size metaslab count
2207 * -------------|-----------------
2209 * 8GB - 100GB one per 512MB
2211 * 50TB - 32PB one per 256GB
2213 * -------------------------------
2216 if (ms_count < vdev_min_ms_count)
2217 ms_shift = highbit64(asize / vdev_min_ms_count);
2218 else if (ms_count > vdev_max_ms_count)
2219 ms_shift = highbit64(asize / vdev_max_ms_count);
2221 ms_shift = vdev_default_ms_shift;
2223 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2224 ms_shift = SPA_MAXBLOCKSHIFT;
2225 } else if (ms_shift > vdev_max_ms_shift) {
2226 ms_shift = vdev_max_ms_shift;
2227 /* cap the total count to constrain memory footprint */
2228 if ((asize >> ms_shift) > vdev_ms_count_limit)
2229 ms_shift = highbit64(asize / vdev_ms_count_limit);
2232 vd->vdev_ms_shift = ms_shift;
2233 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2237 * Maximize performance by inflating the configured ashift for top level
2238 * vdevs to be as close to the physical ashift as possible while maintaining
2239 * administrator defined limits and ensuring it doesn't go below the
2243 vdev_ashift_optimize(vdev_t *vd)
2245 if (vd == vd->vdev_top) {
2246 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
2247 vd->vdev_ashift = MIN(
2248 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
2249 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
2252 * Unusual case where logical ashift > physical ashift
2253 * so we can't cap the calculated ashift based on max
2254 * ashift as that would cause failures.
2255 * We still check if we need to increase it to match
2258 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
2265 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2267 ASSERT(vd == vd->vdev_top);
2268 /* indirect vdevs don't have metaslabs or dtls */
2269 ASSERT(vdev_is_concrete(vd) || flags == 0);
2270 ASSERT(ISP2(flags));
2271 ASSERT(spa_writeable(vd->vdev_spa));
2273 if (flags & VDD_METASLAB)
2274 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2276 if (flags & VDD_DTL)
2277 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2279 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2283 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2285 for (int c = 0; c < vd->vdev_children; c++)
2286 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2288 if (vd->vdev_ops->vdev_op_leaf)
2289 vdev_dirty(vd->vdev_top, flags, vd, txg);
2295 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2296 * the vdev has less than perfect replication. There are four kinds of DTL:
2298 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2300 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2302 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2303 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2304 * txgs that was scrubbed.
2306 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2307 * persistent errors or just some device being offline.
2308 * Unlike the other three, the DTL_OUTAGE map is not generally
2309 * maintained; it's only computed when needed, typically to
2310 * determine whether a device can be detached.
2312 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2313 * either has the data or it doesn't.
2315 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2316 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2317 * if any child is less than fully replicated, then so is its parent.
2318 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2319 * comprising only those txgs which appear in 'maxfaults' or more children;
2320 * those are the txgs we don't have enough replication to read. For example,
2321 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2322 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2323 * two child DTL_MISSING maps.
2325 * It should be clear from the above that to compute the DTLs and outage maps
2326 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2327 * Therefore, that is all we keep on disk. When loading the pool, or after
2328 * a configuration change, we generate all other DTLs from first principles.
2331 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2333 range_tree_t *rt = vd->vdev_dtl[t];
2335 ASSERT(t < DTL_TYPES);
2336 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2337 ASSERT(spa_writeable(vd->vdev_spa));
2339 mutex_enter(&vd->vdev_dtl_lock);
2340 if (!range_tree_contains(rt, txg, size))
2341 range_tree_add(rt, txg, size);
2342 mutex_exit(&vd->vdev_dtl_lock);
2346 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2348 range_tree_t *rt = vd->vdev_dtl[t];
2349 boolean_t dirty = B_FALSE;
2351 ASSERT(t < DTL_TYPES);
2352 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2355 * While we are loading the pool, the DTLs have not been loaded yet.
2356 * Ignore the DTLs and try all devices. This avoids a recursive
2357 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2358 * when loading the pool (relying on the checksum to ensure that
2359 * we get the right data -- note that we while loading, we are
2360 * only reading the MOS, which is always checksummed).
2362 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2365 mutex_enter(&vd->vdev_dtl_lock);
2366 if (!range_tree_is_empty(rt))
2367 dirty = range_tree_contains(rt, txg, size);
2368 mutex_exit(&vd->vdev_dtl_lock);
2374 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2376 range_tree_t *rt = vd->vdev_dtl[t];
2379 mutex_enter(&vd->vdev_dtl_lock);
2380 empty = range_tree_is_empty(rt);
2381 mutex_exit(&vd->vdev_dtl_lock);
2387 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2390 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2392 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2394 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2395 vd->vdev_ops->vdev_op_leaf)
2398 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2402 * Returns the lowest txg in the DTL range.
2405 vdev_dtl_min(vdev_t *vd)
2409 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2410 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2411 ASSERT0(vd->vdev_children);
2413 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2414 return (rs->rs_start - 1);
2418 * Returns the highest txg in the DTL.
2421 vdev_dtl_max(vdev_t *vd)
2425 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2426 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2427 ASSERT0(vd->vdev_children);
2429 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2430 return (rs->rs_end);
2434 * Determine if a resilvering vdev should remove any DTL entries from
2435 * its range. If the vdev was resilvering for the entire duration of the
2436 * scan then it should excise that range from its DTLs. Otherwise, this
2437 * vdev is considered partially resilvered and should leave its DTL
2438 * entries intact. The comment in vdev_dtl_reassess() describes how we
2442 vdev_dtl_should_excise(vdev_t *vd)
2444 spa_t *spa = vd->vdev_spa;
2445 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2447 ASSERT0(scn->scn_phys.scn_errors);
2448 ASSERT0(vd->vdev_children);
2450 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2453 if (vd->vdev_resilver_txg == 0 ||
2454 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2458 * When a resilver is initiated the scan will assign the scn_max_txg
2459 * value to the highest txg value that exists in all DTLs. If this
2460 * device's max DTL is not part of this scan (i.e. it is not in
2461 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2464 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2465 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2466 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2467 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2474 * Reassess DTLs after a config change or scrub completion.
2477 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2479 spa_t *spa = vd->vdev_spa;
2483 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2485 for (int c = 0; c < vd->vdev_children; c++)
2486 vdev_dtl_reassess(vd->vdev_child[c], txg,
2487 scrub_txg, scrub_done);
2489 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2492 if (vd->vdev_ops->vdev_op_leaf) {
2493 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2495 mutex_enter(&vd->vdev_dtl_lock);
2498 * If we've completed a scan cleanly then determine
2499 * if this vdev should remove any DTLs. We only want to
2500 * excise regions on vdevs that were available during
2501 * the entire duration of this scan.
2503 if (scrub_txg != 0 &&
2504 (spa->spa_scrub_started ||
2505 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2506 vdev_dtl_should_excise(vd)) {
2508 * We completed a scrub up to scrub_txg. If we
2509 * did it without rebooting, then the scrub dtl
2510 * will be valid, so excise the old region and
2511 * fold in the scrub dtl. Otherwise, leave the
2512 * dtl as-is if there was an error.
2514 * There's little trick here: to excise the beginning
2515 * of the DTL_MISSING map, we put it into a reference
2516 * tree and then add a segment with refcnt -1 that
2517 * covers the range [0, scrub_txg). This means
2518 * that each txg in that range has refcnt -1 or 0.
2519 * We then add DTL_SCRUB with a refcnt of 2, so that
2520 * entries in the range [0, scrub_txg) will have a
2521 * positive refcnt -- either 1 or 2. We then convert
2522 * the reference tree into the new DTL_MISSING map.
2524 space_reftree_create(&reftree);
2525 space_reftree_add_map(&reftree,
2526 vd->vdev_dtl[DTL_MISSING], 1);
2527 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2528 space_reftree_add_map(&reftree,
2529 vd->vdev_dtl[DTL_SCRUB], 2);
2530 space_reftree_generate_map(&reftree,
2531 vd->vdev_dtl[DTL_MISSING], 1);
2532 space_reftree_destroy(&reftree);
2534 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2535 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2536 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2538 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2539 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2540 if (!vdev_readable(vd))
2541 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2543 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2544 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2547 * If the vdev was resilvering and no longer has any
2548 * DTLs then reset its resilvering flag and dirty
2549 * the top level so that we persist the change.
2551 if (vd->vdev_resilver_txg != 0 &&
2552 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2553 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2554 vd->vdev_resilver_txg = 0;
2555 vdev_config_dirty(vd->vdev_top);
2558 mutex_exit(&vd->vdev_dtl_lock);
2561 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2565 mutex_enter(&vd->vdev_dtl_lock);
2566 for (int t = 0; t < DTL_TYPES; t++) {
2567 /* account for child's outage in parent's missing map */
2568 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2570 continue; /* leaf vdevs only */
2571 if (t == DTL_PARTIAL)
2572 minref = 1; /* i.e. non-zero */
2573 else if (vd->vdev_nparity != 0)
2574 minref = vd->vdev_nparity + 1; /* RAID-Z */
2576 minref = vd->vdev_children; /* any kind of mirror */
2577 space_reftree_create(&reftree);
2578 for (int c = 0; c < vd->vdev_children; c++) {
2579 vdev_t *cvd = vd->vdev_child[c];
2580 mutex_enter(&cvd->vdev_dtl_lock);
2581 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2582 mutex_exit(&cvd->vdev_dtl_lock);
2584 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2585 space_reftree_destroy(&reftree);
2587 mutex_exit(&vd->vdev_dtl_lock);
2591 vdev_dtl_load(vdev_t *vd)
2593 spa_t *spa = vd->vdev_spa;
2594 objset_t *mos = spa->spa_meta_objset;
2597 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2598 ASSERT(vdev_is_concrete(vd));
2600 error = space_map_open(&vd->vdev_dtl_sm, mos,
2601 vd->vdev_dtl_object, 0, -1ULL, 0);
2604 ASSERT(vd->vdev_dtl_sm != NULL);
2606 mutex_enter(&vd->vdev_dtl_lock);
2609 * Now that we've opened the space_map we need to update
2612 space_map_update(vd->vdev_dtl_sm);
2614 error = space_map_load(vd->vdev_dtl_sm,
2615 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2616 mutex_exit(&vd->vdev_dtl_lock);
2621 for (int c = 0; c < vd->vdev_children; c++) {
2622 error = vdev_dtl_load(vd->vdev_child[c]);
2631 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2633 spa_t *spa = vd->vdev_spa;
2635 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2636 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2641 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2643 spa_t *spa = vd->vdev_spa;
2644 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2645 DMU_OT_NONE, 0, tx);
2648 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2655 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2657 if (vd->vdev_ops != &vdev_hole_ops &&
2658 vd->vdev_ops != &vdev_missing_ops &&
2659 vd->vdev_ops != &vdev_root_ops &&
2660 !vd->vdev_top->vdev_removing) {
2661 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2662 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2664 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2665 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2668 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2669 vdev_construct_zaps(vd->vdev_child[i], tx);
2674 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2676 spa_t *spa = vd->vdev_spa;
2677 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2678 objset_t *mos = spa->spa_meta_objset;
2679 range_tree_t *rtsync;
2681 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2683 ASSERT(vdev_is_concrete(vd));
2684 ASSERT(vd->vdev_ops->vdev_op_leaf);
2686 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2688 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2689 mutex_enter(&vd->vdev_dtl_lock);
2690 space_map_free(vd->vdev_dtl_sm, tx);
2691 space_map_close(vd->vdev_dtl_sm);
2692 vd->vdev_dtl_sm = NULL;
2693 mutex_exit(&vd->vdev_dtl_lock);
2696 * We only destroy the leaf ZAP for detached leaves or for
2697 * removed log devices. Removed data devices handle leaf ZAP
2698 * cleanup later, once cancellation is no longer possible.
2700 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2701 vd->vdev_top->vdev_islog)) {
2702 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2703 vd->vdev_leaf_zap = 0;
2710 if (vd->vdev_dtl_sm == NULL) {
2711 uint64_t new_object;
2713 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx);
2714 VERIFY3U(new_object, !=, 0);
2716 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2718 ASSERT(vd->vdev_dtl_sm != NULL);
2721 rtsync = range_tree_create(NULL, NULL);
2723 mutex_enter(&vd->vdev_dtl_lock);
2724 range_tree_walk(rt, range_tree_add, rtsync);
2725 mutex_exit(&vd->vdev_dtl_lock);
2727 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx);
2728 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
2729 range_tree_vacate(rtsync, NULL, NULL);
2731 range_tree_destroy(rtsync);
2734 * If the object for the space map has changed then dirty
2735 * the top level so that we update the config.
2737 if (object != space_map_object(vd->vdev_dtl_sm)) {
2738 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2739 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2740 (u_longlong_t)object,
2741 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2742 vdev_config_dirty(vd->vdev_top);
2747 mutex_enter(&vd->vdev_dtl_lock);
2748 space_map_update(vd->vdev_dtl_sm);
2749 mutex_exit(&vd->vdev_dtl_lock);
2753 * Determine whether the specified vdev can be offlined/detached/removed
2754 * without losing data.
2757 vdev_dtl_required(vdev_t *vd)
2759 spa_t *spa = vd->vdev_spa;
2760 vdev_t *tvd = vd->vdev_top;
2761 uint8_t cant_read = vd->vdev_cant_read;
2764 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2766 if (vd == spa->spa_root_vdev || vd == tvd)
2770 * Temporarily mark the device as unreadable, and then determine
2771 * whether this results in any DTL outages in the top-level vdev.
2772 * If not, we can safely offline/detach/remove the device.
2774 vd->vdev_cant_read = B_TRUE;
2775 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2776 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2777 vd->vdev_cant_read = cant_read;
2778 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2780 if (!required && zio_injection_enabled)
2781 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2787 * Determine if resilver is needed, and if so the txg range.
2790 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2792 boolean_t needed = B_FALSE;
2793 uint64_t thismin = UINT64_MAX;
2794 uint64_t thismax = 0;
2796 if (vd->vdev_children == 0) {
2797 mutex_enter(&vd->vdev_dtl_lock);
2798 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2799 vdev_writeable(vd)) {
2801 thismin = vdev_dtl_min(vd);
2802 thismax = vdev_dtl_max(vd);
2805 mutex_exit(&vd->vdev_dtl_lock);
2807 for (int c = 0; c < vd->vdev_children; c++) {
2808 vdev_t *cvd = vd->vdev_child[c];
2809 uint64_t cmin, cmax;
2811 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2812 thismin = MIN(thismin, cmin);
2813 thismax = MAX(thismax, cmax);
2819 if (needed && minp) {
2827 * Gets the checkpoint space map object from the vdev's ZAP.
2828 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2829 * or the ZAP doesn't exist yet.
2832 vdev_checkpoint_sm_object(vdev_t *vd)
2834 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
2835 if (vd->vdev_top_zap == 0) {
2839 uint64_t sm_obj = 0;
2840 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
2841 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj);
2843 ASSERT(err == 0 || err == ENOENT);
2849 vdev_load(vdev_t *vd)
2853 * Recursively load all children.
2855 for (int c = 0; c < vd->vdev_children; c++) {
2856 error = vdev_load(vd->vdev_child[c]);
2862 vdev_set_deflate_ratio(vd);
2865 * If this is a top-level vdev, initialize its metaslabs.
2867 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2868 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2869 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2870 VDEV_AUX_CORRUPT_DATA);
2871 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2872 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2873 (u_longlong_t)vd->vdev_asize);
2874 return (SET_ERROR(ENXIO));
2875 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2876 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2877 "[error=%d]", error);
2878 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2879 VDEV_AUX_CORRUPT_DATA);
2883 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd);
2884 if (checkpoint_sm_obj != 0) {
2885 objset_t *mos = spa_meta_objset(vd->vdev_spa);
2886 ASSERT(vd->vdev_asize != 0);
2887 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
2889 if ((error = space_map_open(&vd->vdev_checkpoint_sm,
2890 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
2891 vd->vdev_ashift))) {
2892 vdev_dbgmsg(vd, "vdev_load: space_map_open "
2893 "failed for checkpoint spacemap (obj %llu) "
2895 (u_longlong_t)checkpoint_sm_obj, error);
2898 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
2899 space_map_update(vd->vdev_checkpoint_sm);
2902 * Since the checkpoint_sm contains free entries
2903 * exclusively we can use sm_alloc to indicate the
2904 * culmulative checkpointed space that has been freed.
2906 vd->vdev_stat.vs_checkpoint_space =
2907 -vd->vdev_checkpoint_sm->sm_alloc;
2908 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
2909 vd->vdev_stat.vs_checkpoint_space;
2914 * If this is a leaf vdev, load its DTL.
2916 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2917 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2918 VDEV_AUX_CORRUPT_DATA);
2919 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
2920 "[error=%d]", error);
2924 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2925 if (obsolete_sm_object != 0) {
2926 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2927 ASSERT(vd->vdev_asize != 0);
2928 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
2930 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2931 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2932 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2933 VDEV_AUX_CORRUPT_DATA);
2934 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
2935 "obsolete spacemap (obj %llu) [error=%d]",
2936 (u_longlong_t)obsolete_sm_object, error);
2939 space_map_update(vd->vdev_obsolete_sm);
2946 * The special vdev case is used for hot spares and l2cache devices. Its
2947 * sole purpose it to set the vdev state for the associated vdev. To do this,
2948 * we make sure that we can open the underlying device, then try to read the
2949 * label, and make sure that the label is sane and that it hasn't been
2950 * repurposed to another pool.
2953 vdev_validate_aux(vdev_t *vd)
2956 uint64_t guid, version;
2959 if (!vdev_readable(vd))
2962 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2963 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2964 VDEV_AUX_CORRUPT_DATA);
2968 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2969 !SPA_VERSION_IS_SUPPORTED(version) ||
2970 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2971 guid != vd->vdev_guid ||
2972 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2973 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2974 VDEV_AUX_CORRUPT_DATA);
2980 * We don't actually check the pool state here. If it's in fact in
2981 * use by another pool, we update this fact on the fly when requested.
2988 * Free the objects used to store this vdev's spacemaps, and the array
2989 * that points to them.
2992 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
2994 if (vd->vdev_ms_array == 0)
2997 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2998 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
2999 size_t array_bytes = array_count * sizeof (uint64_t);
3000 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3001 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3002 array_bytes, smobj_array, 0));
3004 for (uint64_t i = 0; i < array_count; i++) {
3005 uint64_t smobj = smobj_array[i];
3009 space_map_free_obj(mos, smobj, tx);
3012 kmem_free(smobj_array, array_bytes);
3013 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3014 vd->vdev_ms_array = 0;
3018 vdev_remove_empty(vdev_t *vd, uint64_t txg)
3020 spa_t *spa = vd->vdev_spa;
3023 ASSERT(vd == vd->vdev_top);
3024 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3026 if (vd->vdev_ms != NULL) {
3027 metaslab_group_t *mg = vd->vdev_mg;
3029 metaslab_group_histogram_verify(mg);
3030 metaslab_class_histogram_verify(mg->mg_class);
3032 for (int m = 0; m < vd->vdev_ms_count; m++) {
3033 metaslab_t *msp = vd->vdev_ms[m];
3035 if (msp == NULL || msp->ms_sm == NULL)
3038 mutex_enter(&msp->ms_lock);
3040 * If the metaslab was not loaded when the vdev
3041 * was removed then the histogram accounting may
3042 * not be accurate. Update the histogram information
3043 * here so that we ensure that the metaslab group
3044 * and metaslab class are up-to-date.
3046 metaslab_group_histogram_remove(mg, msp);
3048 VERIFY0(space_map_allocated(msp->ms_sm));
3049 space_map_close(msp->ms_sm);
3051 mutex_exit(&msp->ms_lock);
3054 if (vd->vdev_checkpoint_sm != NULL) {
3055 ASSERT(spa_has_checkpoint(spa));
3056 space_map_close(vd->vdev_checkpoint_sm);
3057 vd->vdev_checkpoint_sm = NULL;
3060 metaslab_group_histogram_verify(mg);
3061 metaslab_class_histogram_verify(mg->mg_class);
3062 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
3063 ASSERT0(mg->mg_histogram[i]);
3066 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3067 vdev_destroy_spacemaps(vd, tx);
3069 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
3070 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3071 vd->vdev_top_zap = 0;
3077 vdev_sync_done(vdev_t *vd, uint64_t txg)
3080 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3082 ASSERT(vdev_is_concrete(vd));
3084 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3086 metaslab_sync_done(msp, txg);
3089 metaslab_sync_reassess(vd->vdev_mg);
3093 vdev_sync(vdev_t *vd, uint64_t txg)
3095 spa_t *spa = vd->vdev_spa;
3100 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3103 ASSERT(vd->vdev_removing ||
3104 vd->vdev_ops == &vdev_indirect_ops);
3106 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3107 vdev_indirect_sync_obsolete(vd, tx);
3111 * If the vdev is indirect, it can't have dirty
3112 * metaslabs or DTLs.
3114 if (vd->vdev_ops == &vdev_indirect_ops) {
3115 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3116 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3121 ASSERT(vdev_is_concrete(vd));
3123 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3124 !vd->vdev_removing) {
3125 ASSERT(vd == vd->vdev_top);
3126 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3127 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3128 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3129 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3130 ASSERT(vd->vdev_ms_array != 0);
3131 vdev_config_dirty(vd);
3135 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3136 metaslab_sync(msp, txg);
3137 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3140 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3141 vdev_dtl_sync(lvd, txg);
3144 * Remove the metadata associated with this vdev once it's empty.
3145 * Note that this is typically used for log/cache device removal;
3146 * we don't empty toplevel vdevs when removing them. But if
3147 * a toplevel happens to be emptied, this is not harmful.
3149 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) {
3150 vdev_remove_empty(vd, txg);
3153 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3157 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3159 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3163 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3164 * not be opened, and no I/O is attempted.
3167 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3171 spa_vdev_state_enter(spa, SCL_NONE);
3173 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3174 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3176 if (!vd->vdev_ops->vdev_op_leaf)
3177 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3182 * We don't directly use the aux state here, but if we do a
3183 * vdev_reopen(), we need this value to be present to remember why we
3186 vd->vdev_label_aux = aux;
3189 * Faulted state takes precedence over degraded.
3191 vd->vdev_delayed_close = B_FALSE;
3192 vd->vdev_faulted = 1ULL;
3193 vd->vdev_degraded = 0ULL;
3194 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3197 * If this device has the only valid copy of the data, then
3198 * back off and simply mark the vdev as degraded instead.
3200 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3201 vd->vdev_degraded = 1ULL;
3202 vd->vdev_faulted = 0ULL;
3205 * If we reopen the device and it's not dead, only then do we
3210 if (vdev_readable(vd))
3211 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3214 return (spa_vdev_state_exit(spa, vd, 0));
3218 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3219 * user that something is wrong. The vdev continues to operate as normal as far
3220 * as I/O is concerned.
3223 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3227 spa_vdev_state_enter(spa, SCL_NONE);
3229 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3230 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3232 if (!vd->vdev_ops->vdev_op_leaf)
3233 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3236 * If the vdev is already faulted, then don't do anything.
3238 if (vd->vdev_faulted || vd->vdev_degraded)
3239 return (spa_vdev_state_exit(spa, NULL, 0));
3241 vd->vdev_degraded = 1ULL;
3242 if (!vdev_is_dead(vd))
3243 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3246 return (spa_vdev_state_exit(spa, vd, 0));
3250 * Online the given vdev.
3252 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3253 * spare device should be detached when the device finishes resilvering.
3254 * Second, the online should be treated like a 'test' online case, so no FMA
3255 * events are generated if the device fails to open.
3258 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3260 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3261 boolean_t wasoffline;
3262 vdev_state_t oldstate;
3264 spa_vdev_state_enter(spa, SCL_NONE);
3266 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3267 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3269 if (!vd->vdev_ops->vdev_op_leaf)
3270 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3272 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3273 oldstate = vd->vdev_state;
3276 vd->vdev_offline = B_FALSE;
3277 vd->vdev_tmpoffline = B_FALSE;
3278 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3279 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3281 /* XXX - L2ARC 1.0 does not support expansion */
3282 if (!vd->vdev_aux) {
3283 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3284 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
3288 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3290 if (!vd->vdev_aux) {
3291 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3292 pvd->vdev_expanding = B_FALSE;
3296 *newstate = vd->vdev_state;
3297 if ((flags & ZFS_ONLINE_UNSPARE) &&
3298 !vdev_is_dead(vd) && vd->vdev_parent &&
3299 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3300 vd->vdev_parent->vdev_child[0] == vd)
3301 vd->vdev_unspare = B_TRUE;
3303 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3305 /* XXX - L2ARC 1.0 does not support expansion */
3307 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3308 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3311 /* Restart initializing if necessary */
3312 mutex_enter(&vd->vdev_initialize_lock);
3313 if (vdev_writeable(vd) &&
3314 vd->vdev_initialize_thread == NULL &&
3315 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3316 (void) vdev_initialize(vd);
3318 mutex_exit(&vd->vdev_initialize_lock);
3321 (oldstate < VDEV_STATE_DEGRADED &&
3322 vd->vdev_state >= VDEV_STATE_DEGRADED))
3323 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3325 return (spa_vdev_state_exit(spa, vd, 0));
3329 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3333 uint64_t generation;
3334 metaslab_group_t *mg;
3337 spa_vdev_state_enter(spa, SCL_ALLOC);
3339 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3340 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3342 if (!vd->vdev_ops->vdev_op_leaf)
3343 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3347 generation = spa->spa_config_generation + 1;
3350 * If the device isn't already offline, try to offline it.
3352 if (!vd->vdev_offline) {
3354 * If this device has the only valid copy of some data,
3355 * don't allow it to be offlined. Log devices are always
3358 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3359 vdev_dtl_required(vd))
3360 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3363 * If the top-level is a slog and it has had allocations
3364 * then proceed. We check that the vdev's metaslab group
3365 * is not NULL since it's possible that we may have just
3366 * added this vdev but not yet initialized its metaslabs.
3368 if (tvd->vdev_islog && mg != NULL) {
3370 * Prevent any future allocations.
3372 metaslab_group_passivate(mg);
3373 (void) spa_vdev_state_exit(spa, vd, 0);
3375 error = spa_reset_logs(spa);
3378 * If the log device was successfully reset but has
3379 * checkpointed data, do not offline it.
3382 tvd->vdev_checkpoint_sm != NULL) {
3383 ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc,
3385 error = ZFS_ERR_CHECKPOINT_EXISTS;
3388 spa_vdev_state_enter(spa, SCL_ALLOC);
3391 * Check to see if the config has changed.
3393 if (error || generation != spa->spa_config_generation) {
3394 metaslab_group_activate(mg);
3396 return (spa_vdev_state_exit(spa,
3398 (void) spa_vdev_state_exit(spa, vd, 0);
3401 ASSERT0(tvd->vdev_stat.vs_alloc);
3405 * Offline this device and reopen its top-level vdev.
3406 * If the top-level vdev is a log device then just offline
3407 * it. Otherwise, if this action results in the top-level
3408 * vdev becoming unusable, undo it and fail the request.
3410 vd->vdev_offline = B_TRUE;
3413 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3414 vdev_is_dead(tvd)) {
3415 vd->vdev_offline = B_FALSE;
3417 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3421 * Add the device back into the metaslab rotor so that
3422 * once we online the device it's open for business.
3424 if (tvd->vdev_islog && mg != NULL)
3425 metaslab_group_activate(mg);
3428 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3430 return (spa_vdev_state_exit(spa, vd, 0));
3434 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3438 mutex_enter(&spa->spa_vdev_top_lock);
3439 error = vdev_offline_locked(spa, guid, flags);
3440 mutex_exit(&spa->spa_vdev_top_lock);
3446 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3447 * vdev_offline(), we assume the spa config is locked. We also clear all
3448 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3451 vdev_clear(spa_t *spa, vdev_t *vd)
3453 vdev_t *rvd = spa->spa_root_vdev;
3455 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3460 vd->vdev_stat.vs_read_errors = 0;
3461 vd->vdev_stat.vs_write_errors = 0;
3462 vd->vdev_stat.vs_checksum_errors = 0;
3464 for (int c = 0; c < vd->vdev_children; c++)
3465 vdev_clear(spa, vd->vdev_child[c]);
3468 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
3469 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
3471 for (int c = 0; c < spa->spa_spares.sav_count; c++)
3472 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
3476 * It makes no sense to "clear" an indirect vdev.
3478 if (!vdev_is_concrete(vd))
3482 * If we're in the FAULTED state or have experienced failed I/O, then
3483 * clear the persistent state and attempt to reopen the device. We
3484 * also mark the vdev config dirty, so that the new faulted state is
3485 * written out to disk.
3487 if (vd->vdev_faulted || vd->vdev_degraded ||
3488 !vdev_readable(vd) || !vdev_writeable(vd)) {
3491 * When reopening in reponse to a clear event, it may be due to
3492 * a fmadm repair request. In this case, if the device is
3493 * still broken, we want to still post the ereport again.
3495 vd->vdev_forcefault = B_TRUE;
3497 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3498 vd->vdev_cant_read = B_FALSE;
3499 vd->vdev_cant_write = B_FALSE;
3501 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3503 vd->vdev_forcefault = B_FALSE;
3505 if (vd != rvd && vdev_writeable(vd->vdev_top))
3506 vdev_state_dirty(vd->vdev_top);
3508 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
3509 spa_async_request(spa, SPA_ASYNC_RESILVER);
3511 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3515 * When clearing a FMA-diagnosed fault, we always want to
3516 * unspare the device, as we assume that the original spare was
3517 * done in response to the FMA fault.
3519 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3520 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3521 vd->vdev_parent->vdev_child[0] == vd)
3522 vd->vdev_unspare = B_TRUE;
3526 vdev_is_dead(vdev_t *vd)
3529 * Holes and missing devices are always considered "dead".
3530 * This simplifies the code since we don't have to check for
3531 * these types of devices in the various code paths.
3532 * Instead we rely on the fact that we skip over dead devices
3533 * before issuing I/O to them.
3535 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3536 vd->vdev_ops == &vdev_hole_ops ||
3537 vd->vdev_ops == &vdev_missing_ops);
3541 vdev_readable(vdev_t *vd)
3543 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3547 vdev_writeable(vdev_t *vd)
3549 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3550 vdev_is_concrete(vd));
3554 vdev_allocatable(vdev_t *vd)
3556 uint64_t state = vd->vdev_state;
3559 * We currently allow allocations from vdevs which may be in the
3560 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3561 * fails to reopen then we'll catch it later when we're holding
3562 * the proper locks. Note that we have to get the vdev state
3563 * in a local variable because although it changes atomically,
3564 * we're asking two separate questions about it.
3566 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3567 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3568 vd->vdev_mg->mg_initialized);
3572 vdev_accessible(vdev_t *vd, zio_t *zio)
3574 ASSERT(zio->io_vd == vd);
3576 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3579 if (zio->io_type == ZIO_TYPE_READ)
3580 return (!vd->vdev_cant_read);
3582 if (zio->io_type == ZIO_TYPE_WRITE)
3583 return (!vd->vdev_cant_write);
3589 vdev_is_spacemap_addressable(vdev_t *vd)
3592 * Assuming 47 bits of the space map entry dedicated for the entry's
3593 * offset (see description in space_map.h), we calculate the maximum
3594 * address that can be described by a space map entry for the given
3597 uint64_t shift = vd->vdev_ashift + 47;
3599 if (shift >= 63) /* detect potential overflow */
3602 return (vd->vdev_asize < (1ULL << shift));
3606 * Get statistics for the given vdev.
3609 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3611 spa_t *spa = vd->vdev_spa;
3612 vdev_t *rvd = spa->spa_root_vdev;
3613 vdev_t *tvd = vd->vdev_top;
3615 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3617 mutex_enter(&vd->vdev_stat_lock);
3618 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3619 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3620 vs->vs_state = vd->vdev_state;
3621 vs->vs_rsize = vdev_get_min_asize(vd);
3622 if (vd->vdev_ops->vdev_op_leaf) {
3623 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3625 * Report intializing progress. Since we don't have the
3626 * initializing locks held, this is only an estimate (although a
3627 * fairly accurate one).
3629 vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done;
3630 vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est;
3631 vs->vs_initialize_state = vd->vdev_initialize_state;
3632 vs->vs_initialize_action_time = vd->vdev_initialize_action_time;
3635 * Report expandable space on top-level, non-auxillary devices only.
3636 * The expandable space is reported in terms of metaslab sized units
3637 * since that determines how much space the pool can expand.
3639 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
3640 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3641 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3643 vs->vs_configured_ashift = vd->vdev_top != NULL
3644 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
3645 vs->vs_logical_ashift = vd->vdev_logical_ashift;
3646 vs->vs_physical_ashift = vd->vdev_physical_ashift;
3647 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3648 vdev_is_concrete(vd)) {
3649 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
3653 * If we're getting stats on the root vdev, aggregate the I/O counts
3654 * over all top-level vdevs (i.e. the direct children of the root).
3657 for (int c = 0; c < rvd->vdev_children; c++) {
3658 vdev_t *cvd = rvd->vdev_child[c];
3659 vdev_stat_t *cvs = &cvd->vdev_stat;
3661 for (int t = 0; t < ZIO_TYPES; t++) {
3662 vs->vs_ops[t] += cvs->vs_ops[t];
3663 vs->vs_bytes[t] += cvs->vs_bytes[t];
3665 cvs->vs_scan_removing = cvd->vdev_removing;
3668 mutex_exit(&vd->vdev_stat_lock);
3672 vdev_clear_stats(vdev_t *vd)
3674 mutex_enter(&vd->vdev_stat_lock);
3675 vd->vdev_stat.vs_space = 0;
3676 vd->vdev_stat.vs_dspace = 0;
3677 vd->vdev_stat.vs_alloc = 0;
3678 mutex_exit(&vd->vdev_stat_lock);
3682 vdev_scan_stat_init(vdev_t *vd)
3684 vdev_stat_t *vs = &vd->vdev_stat;
3686 for (int c = 0; c < vd->vdev_children; c++)
3687 vdev_scan_stat_init(vd->vdev_child[c]);
3689 mutex_enter(&vd->vdev_stat_lock);
3690 vs->vs_scan_processed = 0;
3691 mutex_exit(&vd->vdev_stat_lock);
3695 vdev_stat_update(zio_t *zio, uint64_t psize)
3697 spa_t *spa = zio->io_spa;
3698 vdev_t *rvd = spa->spa_root_vdev;
3699 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3701 uint64_t txg = zio->io_txg;
3702 vdev_stat_t *vs = &vd->vdev_stat;
3703 zio_type_t type = zio->io_type;
3704 int flags = zio->io_flags;
3707 * If this i/o is a gang leader, it didn't do any actual work.
3709 if (zio->io_gang_tree)
3712 if (zio->io_error == 0) {
3714 * If this is a root i/o, don't count it -- we've already
3715 * counted the top-level vdevs, and vdev_get_stats() will
3716 * aggregate them when asked. This reduces contention on
3717 * the root vdev_stat_lock and implicitly handles blocks
3718 * that compress away to holes, for which there is no i/o.
3719 * (Holes never create vdev children, so all the counters
3720 * remain zero, which is what we want.)
3722 * Note: this only applies to successful i/o (io_error == 0)
3723 * because unlike i/o counts, errors are not additive.
3724 * When reading a ditto block, for example, failure of
3725 * one top-level vdev does not imply a root-level error.
3730 ASSERT(vd == zio->io_vd);
3732 if (flags & ZIO_FLAG_IO_BYPASS)
3735 mutex_enter(&vd->vdev_stat_lock);
3737 if (flags & ZIO_FLAG_IO_REPAIR) {
3738 if (flags & ZIO_FLAG_SCAN_THREAD) {
3739 dsl_scan_phys_t *scn_phys =
3740 &spa->spa_dsl_pool->dp_scan->scn_phys;
3741 uint64_t *processed = &scn_phys->scn_processed;
3744 if (vd->vdev_ops->vdev_op_leaf)
3745 atomic_add_64(processed, psize);
3746 vs->vs_scan_processed += psize;
3749 if (flags & ZIO_FLAG_SELF_HEAL)
3750 vs->vs_self_healed += psize;
3754 vs->vs_bytes[type] += psize;
3756 mutex_exit(&vd->vdev_stat_lock);
3760 if (flags & ZIO_FLAG_SPECULATIVE)
3764 * If this is an I/O error that is going to be retried, then ignore the
3765 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3766 * hard errors, when in reality they can happen for any number of
3767 * innocuous reasons (bus resets, MPxIO link failure, etc).
3769 if (zio->io_error == EIO &&
3770 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3774 * Intent logs writes won't propagate their error to the root
3775 * I/O so don't mark these types of failures as pool-level
3778 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3781 mutex_enter(&vd->vdev_stat_lock);
3782 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3783 if (zio->io_error == ECKSUM)
3784 vs->vs_checksum_errors++;
3786 vs->vs_read_errors++;
3788 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3789 vs->vs_write_errors++;
3790 mutex_exit(&vd->vdev_stat_lock);
3792 if (spa->spa_load_state == SPA_LOAD_NONE &&
3793 type == ZIO_TYPE_WRITE && txg != 0 &&
3794 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3795 (flags & ZIO_FLAG_SCAN_THREAD) ||
3796 spa->spa_claiming)) {
3798 * This is either a normal write (not a repair), or it's
3799 * a repair induced by the scrub thread, or it's a repair
3800 * made by zil_claim() during spa_load() in the first txg.
3801 * In the normal case, we commit the DTL change in the same
3802 * txg as the block was born. In the scrub-induced repair
3803 * case, we know that scrubs run in first-pass syncing context,
3804 * so we commit the DTL change in spa_syncing_txg(spa).
3805 * In the zil_claim() case, we commit in spa_first_txg(spa).
3807 * We currently do not make DTL entries for failed spontaneous
3808 * self-healing writes triggered by normal (non-scrubbing)
3809 * reads, because we have no transactional context in which to
3810 * do so -- and it's not clear that it'd be desirable anyway.
3812 if (vd->vdev_ops->vdev_op_leaf) {
3813 uint64_t commit_txg = txg;
3814 if (flags & ZIO_FLAG_SCAN_THREAD) {
3815 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3816 ASSERT(spa_sync_pass(spa) == 1);
3817 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3818 commit_txg = spa_syncing_txg(spa);
3819 } else if (spa->spa_claiming) {
3820 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3821 commit_txg = spa_first_txg(spa);
3823 ASSERT(commit_txg >= spa_syncing_txg(spa));
3824 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3826 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3827 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3828 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3831 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3836 * Update the in-core space usage stats for this vdev, its metaslab class,
3837 * and the root vdev.
3840 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3841 int64_t space_delta)
3843 int64_t dspace_delta = space_delta;
3844 spa_t *spa = vd->vdev_spa;
3845 vdev_t *rvd = spa->spa_root_vdev;
3846 metaslab_group_t *mg = vd->vdev_mg;
3847 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3849 ASSERT(vd == vd->vdev_top);
3852 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3853 * factor. We must calculate this here and not at the root vdev
3854 * because the root vdev's psize-to-asize is simply the max of its
3855 * childrens', thus not accurate enough for us.
3857 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3858 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3859 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3860 vd->vdev_deflate_ratio;
3862 mutex_enter(&vd->vdev_stat_lock);
3863 vd->vdev_stat.vs_alloc += alloc_delta;
3864 vd->vdev_stat.vs_space += space_delta;
3865 vd->vdev_stat.vs_dspace += dspace_delta;
3866 mutex_exit(&vd->vdev_stat_lock);
3868 if (mc == spa_normal_class(spa)) {
3869 mutex_enter(&rvd->vdev_stat_lock);
3870 rvd->vdev_stat.vs_alloc += alloc_delta;
3871 rvd->vdev_stat.vs_space += space_delta;
3872 rvd->vdev_stat.vs_dspace += dspace_delta;
3873 mutex_exit(&rvd->vdev_stat_lock);
3877 ASSERT(rvd == vd->vdev_parent);
3878 ASSERT(vd->vdev_ms_count != 0);
3880 metaslab_class_space_update(mc,
3881 alloc_delta, defer_delta, space_delta, dspace_delta);
3886 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3887 * so that it will be written out next time the vdev configuration is synced.
3888 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3891 vdev_config_dirty(vdev_t *vd)
3893 spa_t *spa = vd->vdev_spa;
3894 vdev_t *rvd = spa->spa_root_vdev;
3897 ASSERT(spa_writeable(spa));
3900 * If this is an aux vdev (as with l2cache and spare devices), then we
3901 * update the vdev config manually and set the sync flag.
3903 if (vd->vdev_aux != NULL) {
3904 spa_aux_vdev_t *sav = vd->vdev_aux;
3908 for (c = 0; c < sav->sav_count; c++) {
3909 if (sav->sav_vdevs[c] == vd)
3913 if (c == sav->sav_count) {
3915 * We're being removed. There's nothing more to do.
3917 ASSERT(sav->sav_sync == B_TRUE);
3921 sav->sav_sync = B_TRUE;
3923 if (nvlist_lookup_nvlist_array(sav->sav_config,
3924 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3925 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3926 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3932 * Setting the nvlist in the middle if the array is a little
3933 * sketchy, but it will work.
3935 nvlist_free(aux[c]);
3936 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3942 * The dirty list is protected by the SCL_CONFIG lock. The caller
3943 * must either hold SCL_CONFIG as writer, or must be the sync thread
3944 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3945 * so this is sufficient to ensure mutual exclusion.
3947 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3948 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3949 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3952 for (c = 0; c < rvd->vdev_children; c++)
3953 vdev_config_dirty(rvd->vdev_child[c]);
3955 ASSERT(vd == vd->vdev_top);
3957 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3958 vdev_is_concrete(vd)) {
3959 list_insert_head(&spa->spa_config_dirty_list, vd);
3965 vdev_config_clean(vdev_t *vd)
3967 spa_t *spa = vd->vdev_spa;
3969 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3970 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3971 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3973 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3974 list_remove(&spa->spa_config_dirty_list, vd);
3978 * Mark a top-level vdev's state as dirty, so that the next pass of
3979 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3980 * the state changes from larger config changes because they require
3981 * much less locking, and are often needed for administrative actions.
3984 vdev_state_dirty(vdev_t *vd)
3986 spa_t *spa = vd->vdev_spa;
3988 ASSERT(spa_writeable(spa));
3989 ASSERT(vd == vd->vdev_top);
3992 * The state list is protected by the SCL_STATE lock. The caller
3993 * must either hold SCL_STATE as writer, or must be the sync thread
3994 * (which holds SCL_STATE as reader). There's only one sync thread,
3995 * so this is sufficient to ensure mutual exclusion.
3997 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3998 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3999 spa_config_held(spa, SCL_STATE, RW_READER)));
4001 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4002 vdev_is_concrete(vd))
4003 list_insert_head(&spa->spa_state_dirty_list, vd);
4007 vdev_state_clean(vdev_t *vd)
4009 spa_t *spa = vd->vdev_spa;
4011 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4012 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4013 spa_config_held(spa, SCL_STATE, RW_READER)));
4015 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4016 list_remove(&spa->spa_state_dirty_list, vd);
4020 * Propagate vdev state up from children to parent.
4023 vdev_propagate_state(vdev_t *vd)
4025 spa_t *spa = vd->vdev_spa;
4026 vdev_t *rvd = spa->spa_root_vdev;
4027 int degraded = 0, faulted = 0;
4031 if (vd->vdev_children > 0) {
4032 for (int c = 0; c < vd->vdev_children; c++) {
4033 child = vd->vdev_child[c];
4036 * Don't factor holes or indirect vdevs into the
4039 if (!vdev_is_concrete(child))
4042 if (!vdev_readable(child) ||
4043 (!vdev_writeable(child) && spa_writeable(spa))) {
4045 * Root special: if there is a top-level log
4046 * device, treat the root vdev as if it were
4049 if (child->vdev_islog && vd == rvd)
4053 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4057 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4061 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4064 * Root special: if there is a top-level vdev that cannot be
4065 * opened due to corrupted metadata, then propagate the root
4066 * vdev's aux state as 'corrupt' rather than 'insufficient
4069 if (corrupted && vd == rvd &&
4070 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4071 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4072 VDEV_AUX_CORRUPT_DATA);
4075 if (vd->vdev_parent)
4076 vdev_propagate_state(vd->vdev_parent);
4080 * Set a vdev's state. If this is during an open, we don't update the parent
4081 * state, because we're in the process of opening children depth-first.
4082 * Otherwise, we propagate the change to the parent.
4084 * If this routine places a device in a faulted state, an appropriate ereport is
4088 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4090 uint64_t save_state;
4091 spa_t *spa = vd->vdev_spa;
4093 if (state == vd->vdev_state) {
4094 vd->vdev_stat.vs_aux = aux;
4098 save_state = vd->vdev_state;
4100 vd->vdev_state = state;
4101 vd->vdev_stat.vs_aux = aux;
4104 * If we are setting the vdev state to anything but an open state, then
4105 * always close the underlying device unless the device has requested
4106 * a delayed close (i.e. we're about to remove or fault the device).
4107 * Otherwise, we keep accessible but invalid devices open forever.
4108 * We don't call vdev_close() itself, because that implies some extra
4109 * checks (offline, etc) that we don't want here. This is limited to
4110 * leaf devices, because otherwise closing the device will affect other
4113 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4114 vd->vdev_ops->vdev_op_leaf)
4115 vd->vdev_ops->vdev_op_close(vd);
4117 if (vd->vdev_removed &&
4118 state == VDEV_STATE_CANT_OPEN &&
4119 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4121 * If the previous state is set to VDEV_STATE_REMOVED, then this
4122 * device was previously marked removed and someone attempted to
4123 * reopen it. If this failed due to a nonexistent device, then
4124 * keep the device in the REMOVED state. We also let this be if
4125 * it is one of our special test online cases, which is only
4126 * attempting to online the device and shouldn't generate an FMA
4129 vd->vdev_state = VDEV_STATE_REMOVED;
4130 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4131 } else if (state == VDEV_STATE_REMOVED) {
4132 vd->vdev_removed = B_TRUE;
4133 } else if (state == VDEV_STATE_CANT_OPEN) {
4135 * If we fail to open a vdev during an import or recovery, we
4136 * mark it as "not available", which signifies that it was
4137 * never there to begin with. Failure to open such a device
4138 * is not considered an error.
4140 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4141 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4142 vd->vdev_ops->vdev_op_leaf)
4143 vd->vdev_not_present = 1;
4146 * Post the appropriate ereport. If the 'prevstate' field is
4147 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4148 * that this is part of a vdev_reopen(). In this case, we don't
4149 * want to post the ereport if the device was already in the
4150 * CANT_OPEN state beforehand.
4152 * If the 'checkremove' flag is set, then this is an attempt to
4153 * online the device in response to an insertion event. If we
4154 * hit this case, then we have detected an insertion event for a
4155 * faulted or offline device that wasn't in the removed state.
4156 * In this scenario, we don't post an ereport because we are
4157 * about to replace the device, or attempt an online with
4158 * vdev_forcefault, which will generate the fault for us.
4160 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4161 !vd->vdev_not_present && !vd->vdev_checkremove &&
4162 vd != spa->spa_root_vdev) {
4166 case VDEV_AUX_OPEN_FAILED:
4167 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4169 case VDEV_AUX_CORRUPT_DATA:
4170 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4172 case VDEV_AUX_NO_REPLICAS:
4173 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4175 case VDEV_AUX_BAD_GUID_SUM:
4176 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4178 case VDEV_AUX_TOO_SMALL:
4179 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4181 case VDEV_AUX_BAD_LABEL:
4182 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4185 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4188 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
4191 /* Erase any notion of persistent removed state */
4192 vd->vdev_removed = B_FALSE;
4194 vd->vdev_removed = B_FALSE;
4198 * Notify the fmd of the state change. Be verbose and post
4199 * notifications even for stuff that's not important; the fmd agent can
4200 * sort it out. Don't emit state change events for non-leaf vdevs since
4201 * they can't change state on their own. The FMD can check their state
4202 * if it wants to when it sees that a leaf vdev had a state change.
4204 if (vd->vdev_ops->vdev_op_leaf)
4205 zfs_post_state_change(spa, vd);
4207 if (!isopen && vd->vdev_parent)
4208 vdev_propagate_state(vd->vdev_parent);
4212 vdev_children_are_offline(vdev_t *vd)
4214 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4216 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4217 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4225 * Check the vdev configuration to ensure that it's capable of supporting
4226 * a root pool. We do not support partial configuration.
4227 * In addition, only a single top-level vdev is allowed.
4229 * FreeBSD does not have above limitations.
4232 vdev_is_bootable(vdev_t *vd)
4235 if (!vd->vdev_ops->vdev_op_leaf) {
4236 char *vdev_type = vd->vdev_ops->vdev_op_type;
4238 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
4239 vd->vdev_children > 1) {
4241 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4242 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4247 for (int c = 0; c < vd->vdev_children; c++) {
4248 if (!vdev_is_bootable(vd->vdev_child[c]))
4251 #endif /* illumos */
4256 vdev_is_concrete(vdev_t *vd)
4258 vdev_ops_t *ops = vd->vdev_ops;
4259 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4260 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4268 * Determine if a log device has valid content. If the vdev was
4269 * removed or faulted in the MOS config then we know that
4270 * the content on the log device has already been written to the pool.
4273 vdev_log_state_valid(vdev_t *vd)
4275 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4279 for (int c = 0; c < vd->vdev_children; c++)
4280 if (vdev_log_state_valid(vd->vdev_child[c]))
4287 * Expand a vdev if possible.
4290 vdev_expand(vdev_t *vd, uint64_t txg)
4292 ASSERT(vd->vdev_top == vd);
4293 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4294 ASSERT(vdev_is_concrete(vd));
4296 vdev_set_deflate_ratio(vd);
4298 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
4299 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4300 vdev_config_dirty(vd);
4308 vdev_split(vdev_t *vd)
4310 vdev_t *cvd, *pvd = vd->vdev_parent;
4312 vdev_remove_child(pvd, vd);
4313 vdev_compact_children(pvd);
4315 cvd = pvd->vdev_child[0];
4316 if (pvd->vdev_children == 1) {
4317 vdev_remove_parent(cvd);
4318 cvd->vdev_splitting = B_TRUE;
4320 vdev_propagate_state(cvd);
4324 vdev_deadman(vdev_t *vd)
4326 for (int c = 0; c < vd->vdev_children; c++) {
4327 vdev_t *cvd = vd->vdev_child[c];
4332 if (vd->vdev_ops->vdev_op_leaf) {
4333 vdev_queue_t *vq = &vd->vdev_queue;
4335 mutex_enter(&vq->vq_lock);
4336 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4337 spa_t *spa = vd->vdev_spa;
4342 * Look at the head of all the pending queues,
4343 * if any I/O has been outstanding for longer than
4344 * the spa_deadman_synctime we panic the system.
4346 fio = avl_first(&vq->vq_active_tree);
4347 delta = gethrtime() - fio->io_timestamp;
4348 if (delta > spa_deadman_synctime(spa)) {
4349 vdev_dbgmsg(vd, "SLOW IO: zio timestamp "
4350 "%lluns, delta %lluns, last io %lluns",
4351 fio->io_timestamp, (u_longlong_t)delta,
4352 vq->vq_io_complete_ts);
4353 fm_panic("I/O to pool '%s' appears to be "
4354 "hung on vdev guid %llu at '%s'.",
4356 (long long unsigned int) vd->vdev_guid,
4360 mutex_exit(&vq->vq_lock);