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 /* default target for number of metaslabs per top-level vdev */
167 int zfs_vdev_default_ms_count = 200;
168 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, default_ms_count, CTLFLAG_RWTUN,
169 &zfs_vdev_default_ms_count, 0,
170 "Target number of metaslabs per top-level vdev");
172 /* minimum number of metaslabs per top-level vdev */
173 int zfs_vdev_min_ms_count = 16;
174 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, min_ms_count, CTLFLAG_RWTUN,
175 &zfs_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 zfs_vdev_ms_count_limit = 1ULL << 17;
180 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, max_ms_count_limit, CTLFLAG_RWTUN,
181 &zfs_vdev_ms_count_limit, 0,
182 "Maximum number of metaslabs per top-level vdev");
184 /* lower limit for metaslab size (512M) */
185 int zfs_vdev_default_ms_shift = 29;
186 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, default_ms_shift, CTLFLAG_RWTUN,
187 &zfs_vdev_default_ms_shift, 0,
188 "Default shift between vdev size and number of metaslabs");
190 /* upper limit for metaslab size (16G) */
191 int zfs_vdev_max_ms_shift = 34;
192 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, max_ms_shift, CTLFLAG_RWTUN,
193 &zfs_vdev_max_ms_shift, 0,
194 "Maximum shift between vdev size and number of metaslabs");
196 boolean_t vdev_validate_skip = B_FALSE;
197 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, validate_skip, CTLFLAG_RWTUN,
198 &vdev_validate_skip, 0,
199 "Bypass vdev validation");
202 * Since the DTL space map of a vdev is not expected to have a lot of
203 * entries, we default its block size to 4K.
205 int vdev_dtl_sm_blksz = (1 << 12);
206 SYSCTL_INT(_vfs_zfs, OID_AUTO, dtl_sm_blksz, CTLFLAG_RDTUN,
207 &vdev_dtl_sm_blksz, 0,
208 "Block size for DTL space map. Power of 2 and greater than 4096.");
211 * vdev-wide space maps that have lots of entries written to them at
212 * the end of each transaction can benefit from a higher I/O bandwidth
213 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
215 int vdev_standard_sm_blksz = (1 << 17);
216 SYSCTL_INT(_vfs_zfs, OID_AUTO, standard_sm_blksz, CTLFLAG_RDTUN,
217 &vdev_standard_sm_blksz, 0,
218 "Block size for standard space map. Power of 2 and greater than 4096.");
221 * Tunable parameter for debugging or performance analysis. Setting this
222 * will cause pool corruption on power loss if a volatile out-of-order
223 * write cache is enabled.
225 boolean_t zfs_nocacheflush = B_FALSE;
226 SYSCTL_INT(_vfs_zfs, OID_AUTO, cache_flush_disable, CTLFLAG_RWTUN,
227 &zfs_nocacheflush, 0, "Disable cache flush");
231 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
237 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
240 if (vd->vdev_path != NULL) {
241 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
244 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
245 vd->vdev_ops->vdev_op_type,
246 (u_longlong_t)vd->vdev_id,
247 (u_longlong_t)vd->vdev_guid, buf);
252 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
256 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
257 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
258 vd->vdev_ops->vdev_op_type);
262 switch (vd->vdev_state) {
263 case VDEV_STATE_UNKNOWN:
264 (void) snprintf(state, sizeof (state), "unknown");
266 case VDEV_STATE_CLOSED:
267 (void) snprintf(state, sizeof (state), "closed");
269 case VDEV_STATE_OFFLINE:
270 (void) snprintf(state, sizeof (state), "offline");
272 case VDEV_STATE_REMOVED:
273 (void) snprintf(state, sizeof (state), "removed");
275 case VDEV_STATE_CANT_OPEN:
276 (void) snprintf(state, sizeof (state), "can't open");
278 case VDEV_STATE_FAULTED:
279 (void) snprintf(state, sizeof (state), "faulted");
281 case VDEV_STATE_DEGRADED:
282 (void) snprintf(state, sizeof (state), "degraded");
284 case VDEV_STATE_HEALTHY:
285 (void) snprintf(state, sizeof (state), "healthy");
288 (void) snprintf(state, sizeof (state), "<state %u>",
289 (uint_t)vd->vdev_state);
292 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
293 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
294 vd->vdev_islog ? " (log)" : "",
295 (u_longlong_t)vd->vdev_guid,
296 vd->vdev_path ? vd->vdev_path : "N/A", state);
298 for (uint64_t i = 0; i < vd->vdev_children; i++)
299 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
303 * Given a vdev type, return the appropriate ops vector.
306 vdev_getops(const char *type)
308 vdev_ops_t *ops, **opspp;
310 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
311 if (strcmp(ops->vdev_op_type, type) == 0)
319 vdev_default_xlate(vdev_t *vd, const range_seg_t *in, range_seg_t *res)
321 res->rs_start = in->rs_start;
322 res->rs_end = in->rs_end;
326 * Default asize function: return the MAX of psize with the asize of
327 * all children. This is what's used by anything other than RAID-Z.
330 vdev_default_asize(vdev_t *vd, uint64_t psize)
332 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
335 for (int c = 0; c < vd->vdev_children; c++) {
336 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
337 asize = MAX(asize, csize);
344 * Get the minimum allocatable size. We define the allocatable size as
345 * the vdev's asize rounded to the nearest metaslab. This allows us to
346 * replace or attach devices which don't have the same physical size but
347 * can still satisfy the same number of allocations.
350 vdev_get_min_asize(vdev_t *vd)
352 vdev_t *pvd = vd->vdev_parent;
355 * If our parent is NULL (inactive spare or cache) or is the root,
356 * just return our own asize.
359 return (vd->vdev_asize);
362 * The top-level vdev just returns the allocatable size rounded
363 * to the nearest metaslab.
365 if (vd == vd->vdev_top)
366 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
369 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
370 * so each child must provide at least 1/Nth of its asize.
372 if (pvd->vdev_ops == &vdev_raidz_ops)
373 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
376 return (pvd->vdev_min_asize);
380 vdev_set_min_asize(vdev_t *vd)
382 vd->vdev_min_asize = vdev_get_min_asize(vd);
384 for (int c = 0; c < vd->vdev_children; c++)
385 vdev_set_min_asize(vd->vdev_child[c]);
389 vdev_lookup_top(spa_t *spa, uint64_t vdev)
391 vdev_t *rvd = spa->spa_root_vdev;
393 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
395 if (vdev < rvd->vdev_children) {
396 ASSERT(rvd->vdev_child[vdev] != NULL);
397 return (rvd->vdev_child[vdev]);
404 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
408 if (vd->vdev_guid == guid)
411 for (int c = 0; c < vd->vdev_children; c++)
412 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
420 vdev_count_leaves_impl(vdev_t *vd)
424 if (vd->vdev_ops->vdev_op_leaf)
427 for (int c = 0; c < vd->vdev_children; c++)
428 n += vdev_count_leaves_impl(vd->vdev_child[c]);
434 vdev_count_leaves(spa_t *spa)
436 return (vdev_count_leaves_impl(spa->spa_root_vdev));
440 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
442 size_t oldsize, newsize;
443 uint64_t id = cvd->vdev_id;
445 spa_t *spa = cvd->vdev_spa;
447 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
448 ASSERT(cvd->vdev_parent == NULL);
450 cvd->vdev_parent = pvd;
455 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
457 oldsize = pvd->vdev_children * sizeof (vdev_t *);
458 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
459 newsize = pvd->vdev_children * sizeof (vdev_t *);
461 newchild = kmem_zalloc(newsize, KM_SLEEP);
462 if (pvd->vdev_child != NULL) {
463 bcopy(pvd->vdev_child, newchild, oldsize);
464 kmem_free(pvd->vdev_child, oldsize);
467 pvd->vdev_child = newchild;
468 pvd->vdev_child[id] = cvd;
470 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
471 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
474 * Walk up all ancestors to update guid sum.
476 for (; pvd != NULL; pvd = pvd->vdev_parent)
477 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
479 if (cvd->vdev_ops->vdev_op_leaf) {
480 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd);
481 cvd->vdev_spa->spa_leaf_list_gen++;
486 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
489 uint_t id = cvd->vdev_id;
491 ASSERT(cvd->vdev_parent == pvd);
496 ASSERT(id < pvd->vdev_children);
497 ASSERT(pvd->vdev_child[id] == cvd);
499 pvd->vdev_child[id] = NULL;
500 cvd->vdev_parent = NULL;
502 for (c = 0; c < pvd->vdev_children; c++)
503 if (pvd->vdev_child[c])
506 if (c == pvd->vdev_children) {
507 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
508 pvd->vdev_child = NULL;
509 pvd->vdev_children = 0;
512 if (cvd->vdev_ops->vdev_op_leaf) {
513 spa_t *spa = cvd->vdev_spa;
514 list_remove(&spa->spa_leaf_list, cvd);
515 spa->spa_leaf_list_gen++;
519 * Walk up all ancestors to update guid sum.
521 for (; pvd != NULL; pvd = pvd->vdev_parent)
522 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
526 * Remove any holes in the child array.
529 vdev_compact_children(vdev_t *pvd)
531 vdev_t **newchild, *cvd;
532 int oldc = pvd->vdev_children;
535 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
540 for (int c = newc = 0; c < oldc; c++)
541 if (pvd->vdev_child[c])
545 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
547 for (int c = newc = 0; c < oldc; c++) {
548 if ((cvd = pvd->vdev_child[c]) != NULL) {
549 newchild[newc] = cvd;
550 cvd->vdev_id = newc++;
557 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
558 pvd->vdev_child = newchild;
559 pvd->vdev_children = newc;
563 * Allocate and minimally initialize a vdev_t.
566 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
569 vdev_indirect_config_t *vic;
571 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
572 vic = &vd->vdev_indirect_config;
574 if (spa->spa_root_vdev == NULL) {
575 ASSERT(ops == &vdev_root_ops);
576 spa->spa_root_vdev = vd;
577 spa->spa_load_guid = spa_generate_guid(NULL);
580 if (guid == 0 && ops != &vdev_hole_ops) {
581 if (spa->spa_root_vdev == vd) {
583 * The root vdev's guid will also be the pool guid,
584 * which must be unique among all pools.
586 guid = spa_generate_guid(NULL);
589 * Any other vdev's guid must be unique within the pool.
591 guid = spa_generate_guid(spa);
593 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
598 vd->vdev_guid = guid;
599 vd->vdev_guid_sum = guid;
601 vd->vdev_state = VDEV_STATE_CLOSED;
602 vd->vdev_ishole = (ops == &vdev_hole_ops);
603 vic->vic_prev_indirect_vdev = UINT64_MAX;
605 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
606 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
607 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
609 list_link_init(&vd->vdev_leaf_node);
610 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
611 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
612 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
613 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
614 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
615 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
616 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
617 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
618 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
620 for (int t = 0; t < DTL_TYPES; t++) {
621 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
623 txg_list_create(&vd->vdev_ms_list, spa,
624 offsetof(struct metaslab, ms_txg_node));
625 txg_list_create(&vd->vdev_dtl_list, spa,
626 offsetof(struct vdev, vdev_dtl_node));
627 vd->vdev_stat.vs_timestamp = gethrtime();
635 * Allocate a new vdev. The 'alloctype' is used to control whether we are
636 * creating a new vdev or loading an existing one - the behavior is slightly
637 * different for each case.
640 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
645 uint64_t guid = 0, islog, nparity;
647 vdev_indirect_config_t *vic;
649 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
651 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
652 return (SET_ERROR(EINVAL));
654 if ((ops = vdev_getops(type)) == NULL)
655 return (SET_ERROR(EINVAL));
658 * If this is a load, get the vdev guid from the nvlist.
659 * Otherwise, vdev_alloc_common() will generate one for us.
661 if (alloctype == VDEV_ALLOC_LOAD) {
664 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
666 return (SET_ERROR(EINVAL));
668 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
669 return (SET_ERROR(EINVAL));
670 } else if (alloctype == VDEV_ALLOC_SPARE) {
671 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
672 return (SET_ERROR(EINVAL));
673 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
674 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
675 return (SET_ERROR(EINVAL));
676 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
677 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
678 return (SET_ERROR(EINVAL));
682 * The first allocated vdev must be of type 'root'.
684 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
685 return (SET_ERROR(EINVAL));
688 * Determine whether we're a log vdev.
691 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
692 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
693 return (SET_ERROR(ENOTSUP));
695 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
696 return (SET_ERROR(ENOTSUP));
699 * Set the nparity property for RAID-Z vdevs.
702 if (ops == &vdev_raidz_ops) {
703 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
705 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
706 return (SET_ERROR(EINVAL));
708 * Previous versions could only support 1 or 2 parity
712 spa_version(spa) < SPA_VERSION_RAIDZ2)
713 return (SET_ERROR(ENOTSUP));
715 spa_version(spa) < SPA_VERSION_RAIDZ3)
716 return (SET_ERROR(ENOTSUP));
719 * We require the parity to be specified for SPAs that
720 * support multiple parity levels.
722 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
723 return (SET_ERROR(EINVAL));
725 * Otherwise, we default to 1 parity device for RAID-Z.
732 ASSERT(nparity != -1ULL);
734 vd = vdev_alloc_common(spa, id, guid, ops);
735 vic = &vd->vdev_indirect_config;
737 vd->vdev_islog = islog;
738 vd->vdev_nparity = nparity;
740 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
741 vd->vdev_path = spa_strdup(vd->vdev_path);
742 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
743 vd->vdev_devid = spa_strdup(vd->vdev_devid);
744 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
745 &vd->vdev_physpath) == 0)
746 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
747 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
748 vd->vdev_fru = spa_strdup(vd->vdev_fru);
751 * Set the whole_disk property. If it's not specified, leave the value
754 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
755 &vd->vdev_wholedisk) != 0)
756 vd->vdev_wholedisk = -1ULL;
758 ASSERT0(vic->vic_mapping_object);
759 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
760 &vic->vic_mapping_object);
761 ASSERT0(vic->vic_births_object);
762 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
763 &vic->vic_births_object);
764 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
765 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
766 &vic->vic_prev_indirect_vdev);
769 * Look for the 'not present' flag. This will only be set if the device
770 * was not present at the time of import.
772 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
773 &vd->vdev_not_present);
776 * Get the alignment requirement.
778 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
781 * Retrieve the vdev creation time.
783 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
787 * If we're a top-level vdev, try to load the allocation parameters.
789 if (parent && !parent->vdev_parent &&
790 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
791 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
793 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
795 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
797 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
799 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
802 ASSERT0(vd->vdev_top_zap);
805 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
806 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
807 alloctype == VDEV_ALLOC_ADD ||
808 alloctype == VDEV_ALLOC_SPLIT ||
809 alloctype == VDEV_ALLOC_ROOTPOOL);
810 vd->vdev_mg = metaslab_group_create(islog ?
811 spa_log_class(spa) : spa_normal_class(spa), vd,
812 spa->spa_alloc_count);
815 if (vd->vdev_ops->vdev_op_leaf &&
816 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
817 (void) nvlist_lookup_uint64(nv,
818 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
820 ASSERT0(vd->vdev_leaf_zap);
824 * If we're a leaf vdev, try to load the DTL object and other state.
827 if (vd->vdev_ops->vdev_op_leaf &&
828 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
829 alloctype == VDEV_ALLOC_ROOTPOOL)) {
830 if (alloctype == VDEV_ALLOC_LOAD) {
831 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
832 &vd->vdev_dtl_object);
833 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
837 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
840 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
841 &spare) == 0 && spare)
845 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
848 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
849 &vd->vdev_resilver_txg);
852 * When importing a pool, we want to ignore the persistent fault
853 * state, as the diagnosis made on another system may not be
854 * valid in the current context. Local vdevs will
855 * remain in the faulted state.
857 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
858 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
860 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
862 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
865 if (vd->vdev_faulted || vd->vdev_degraded) {
869 VDEV_AUX_ERR_EXCEEDED;
870 if (nvlist_lookup_string(nv,
871 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
872 strcmp(aux, "external") == 0)
873 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
879 * Add ourselves to the parent's list of children.
881 vdev_add_child(parent, vd);
889 vdev_free(vdev_t *vd)
891 spa_t *spa = vd->vdev_spa;
892 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
895 * Scan queues are normally destroyed at the end of a scan. If the
896 * queue exists here, that implies the vdev is being removed while
897 * the scan is still running.
899 if (vd->vdev_scan_io_queue != NULL) {
900 mutex_enter(&vd->vdev_scan_io_queue_lock);
901 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
902 vd->vdev_scan_io_queue = NULL;
903 mutex_exit(&vd->vdev_scan_io_queue_lock);
907 * vdev_free() implies closing the vdev first. This is simpler than
908 * trying to ensure complicated semantics for all callers.
912 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
913 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
918 for (int c = 0; c < vd->vdev_children; c++)
919 vdev_free(vd->vdev_child[c]);
921 ASSERT(vd->vdev_child == NULL);
922 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
923 ASSERT(vd->vdev_initialize_thread == NULL);
926 * Discard allocation state.
928 if (vd->vdev_mg != NULL) {
929 vdev_metaslab_fini(vd);
930 metaslab_group_destroy(vd->vdev_mg);
933 ASSERT0(vd->vdev_stat.vs_space);
934 ASSERT0(vd->vdev_stat.vs_dspace);
935 ASSERT0(vd->vdev_stat.vs_alloc);
938 * Remove this vdev from its parent's child list.
940 vdev_remove_child(vd->vdev_parent, vd);
942 ASSERT(vd->vdev_parent == NULL);
943 ASSERT(!list_link_active(&vd->vdev_leaf_node));
946 * Clean up vdev structure.
952 spa_strfree(vd->vdev_path);
954 spa_strfree(vd->vdev_devid);
955 if (vd->vdev_physpath)
956 spa_strfree(vd->vdev_physpath);
958 spa_strfree(vd->vdev_fru);
960 if (vd->vdev_isspare)
961 spa_spare_remove(vd);
962 if (vd->vdev_isl2cache)
963 spa_l2cache_remove(vd);
965 txg_list_destroy(&vd->vdev_ms_list);
966 txg_list_destroy(&vd->vdev_dtl_list);
968 mutex_enter(&vd->vdev_dtl_lock);
969 space_map_close(vd->vdev_dtl_sm);
970 for (int t = 0; t < DTL_TYPES; t++) {
971 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
972 range_tree_destroy(vd->vdev_dtl[t]);
974 mutex_exit(&vd->vdev_dtl_lock);
976 EQUIV(vd->vdev_indirect_births != NULL,
977 vd->vdev_indirect_mapping != NULL);
978 if (vd->vdev_indirect_births != NULL) {
979 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
980 vdev_indirect_births_close(vd->vdev_indirect_births);
983 if (vd->vdev_obsolete_sm != NULL) {
984 ASSERT(vd->vdev_removing ||
985 vd->vdev_ops == &vdev_indirect_ops);
986 space_map_close(vd->vdev_obsolete_sm);
987 vd->vdev_obsolete_sm = NULL;
989 range_tree_destroy(vd->vdev_obsolete_segments);
990 rw_destroy(&vd->vdev_indirect_rwlock);
991 mutex_destroy(&vd->vdev_obsolete_lock);
993 mutex_destroy(&vd->vdev_queue_lock);
994 mutex_destroy(&vd->vdev_dtl_lock);
995 mutex_destroy(&vd->vdev_stat_lock);
996 mutex_destroy(&vd->vdev_probe_lock);
997 mutex_destroy(&vd->vdev_scan_io_queue_lock);
998 mutex_destroy(&vd->vdev_initialize_lock);
999 mutex_destroy(&vd->vdev_initialize_io_lock);
1000 cv_destroy(&vd->vdev_initialize_io_cv);
1001 cv_destroy(&vd->vdev_initialize_cv);
1003 if (vd == spa->spa_root_vdev)
1004 spa->spa_root_vdev = NULL;
1006 kmem_free(vd, sizeof (vdev_t));
1010 * Transfer top-level vdev state from svd to tvd.
1013 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
1015 spa_t *spa = svd->vdev_spa;
1020 ASSERT(tvd == tvd->vdev_top);
1022 tvd->vdev_ms_array = svd->vdev_ms_array;
1023 tvd->vdev_ms_shift = svd->vdev_ms_shift;
1024 tvd->vdev_ms_count = svd->vdev_ms_count;
1025 tvd->vdev_top_zap = svd->vdev_top_zap;
1027 svd->vdev_ms_array = 0;
1028 svd->vdev_ms_shift = 0;
1029 svd->vdev_ms_count = 0;
1030 svd->vdev_top_zap = 0;
1033 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1034 tvd->vdev_mg = svd->vdev_mg;
1035 tvd->vdev_ms = svd->vdev_ms;
1037 svd->vdev_mg = NULL;
1038 svd->vdev_ms = NULL;
1040 if (tvd->vdev_mg != NULL)
1041 tvd->vdev_mg->mg_vd = tvd;
1043 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1044 svd->vdev_checkpoint_sm = NULL;
1046 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1047 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1048 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1050 svd->vdev_stat.vs_alloc = 0;
1051 svd->vdev_stat.vs_space = 0;
1052 svd->vdev_stat.vs_dspace = 0;
1055 * State which may be set on a top-level vdev that's in the
1056 * process of being removed.
1058 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1059 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1060 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1061 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1062 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1063 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1064 ASSERT0(tvd->vdev_removing);
1065 tvd->vdev_removing = svd->vdev_removing;
1066 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1067 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1068 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1069 range_tree_swap(&svd->vdev_obsolete_segments,
1070 &tvd->vdev_obsolete_segments);
1071 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1072 svd->vdev_indirect_config.vic_mapping_object = 0;
1073 svd->vdev_indirect_config.vic_births_object = 0;
1074 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1075 svd->vdev_indirect_mapping = NULL;
1076 svd->vdev_indirect_births = NULL;
1077 svd->vdev_obsolete_sm = NULL;
1078 svd->vdev_removing = 0;
1080 for (t = 0; t < TXG_SIZE; t++) {
1081 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1082 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1083 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1084 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1085 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1086 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1089 if (list_link_active(&svd->vdev_config_dirty_node)) {
1090 vdev_config_clean(svd);
1091 vdev_config_dirty(tvd);
1094 if (list_link_active(&svd->vdev_state_dirty_node)) {
1095 vdev_state_clean(svd);
1096 vdev_state_dirty(tvd);
1099 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1100 svd->vdev_deflate_ratio = 0;
1102 tvd->vdev_islog = svd->vdev_islog;
1103 svd->vdev_islog = 0;
1105 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1109 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1116 for (int c = 0; c < vd->vdev_children; c++)
1117 vdev_top_update(tvd, vd->vdev_child[c]);
1121 * Add a mirror/replacing vdev above an existing vdev.
1124 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1126 spa_t *spa = cvd->vdev_spa;
1127 vdev_t *pvd = cvd->vdev_parent;
1130 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1132 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1134 mvd->vdev_asize = cvd->vdev_asize;
1135 mvd->vdev_min_asize = cvd->vdev_min_asize;
1136 mvd->vdev_max_asize = cvd->vdev_max_asize;
1137 mvd->vdev_psize = cvd->vdev_psize;
1138 mvd->vdev_ashift = cvd->vdev_ashift;
1139 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1140 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1141 mvd->vdev_state = cvd->vdev_state;
1142 mvd->vdev_crtxg = cvd->vdev_crtxg;
1144 vdev_remove_child(pvd, cvd);
1145 vdev_add_child(pvd, mvd);
1146 cvd->vdev_id = mvd->vdev_children;
1147 vdev_add_child(mvd, cvd);
1148 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1150 if (mvd == mvd->vdev_top)
1151 vdev_top_transfer(cvd, mvd);
1157 * Remove a 1-way mirror/replacing vdev from the tree.
1160 vdev_remove_parent(vdev_t *cvd)
1162 vdev_t *mvd = cvd->vdev_parent;
1163 vdev_t *pvd = mvd->vdev_parent;
1165 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1167 ASSERT(mvd->vdev_children == 1);
1168 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1169 mvd->vdev_ops == &vdev_replacing_ops ||
1170 mvd->vdev_ops == &vdev_spare_ops);
1171 cvd->vdev_ashift = mvd->vdev_ashift;
1172 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1173 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1175 vdev_remove_child(mvd, cvd);
1176 vdev_remove_child(pvd, mvd);
1179 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1180 * Otherwise, we could have detached an offline device, and when we
1181 * go to import the pool we'll think we have two top-level vdevs,
1182 * instead of a different version of the same top-level vdev.
1184 if (mvd->vdev_top == mvd) {
1185 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1186 cvd->vdev_orig_guid = cvd->vdev_guid;
1187 cvd->vdev_guid += guid_delta;
1188 cvd->vdev_guid_sum += guid_delta;
1190 cvd->vdev_id = mvd->vdev_id;
1191 vdev_add_child(pvd, cvd);
1192 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1194 if (cvd == cvd->vdev_top)
1195 vdev_top_transfer(mvd, cvd);
1197 ASSERT(mvd->vdev_children == 0);
1202 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1204 spa_t *spa = vd->vdev_spa;
1205 objset_t *mos = spa->spa_meta_objset;
1207 uint64_t oldc = vd->vdev_ms_count;
1208 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1212 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1215 * This vdev is not being allocated from yet or is a hole.
1217 if (vd->vdev_ms_shift == 0)
1220 ASSERT(!vd->vdev_ishole);
1222 ASSERT(oldc <= newc);
1224 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1227 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1228 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1232 vd->vdev_ms_count = newc;
1233 for (m = oldc; m < newc; m++) {
1234 uint64_t object = 0;
1237 * vdev_ms_array may be 0 if we are creating the "fake"
1238 * metaslabs for an indirect vdev for zdb's leak detection.
1239 * See zdb_leak_init().
1241 if (txg == 0 && vd->vdev_ms_array != 0) {
1242 error = dmu_read(mos, vd->vdev_ms_array,
1243 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1246 vdev_dbgmsg(vd, "unable to read the metaslab "
1247 "array [error=%d]", error);
1252 error = metaslab_init(vd->vdev_mg, m, object, txg,
1255 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1262 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1265 * If the vdev is being removed we don't activate
1266 * the metaslabs since we want to ensure that no new
1267 * allocations are performed on this device.
1269 if (oldc == 0 && !vd->vdev_removing)
1270 metaslab_group_activate(vd->vdev_mg);
1273 spa_config_exit(spa, SCL_ALLOC, FTAG);
1279 vdev_metaslab_fini(vdev_t *vd)
1281 if (vd->vdev_checkpoint_sm != NULL) {
1282 ASSERT(spa_feature_is_active(vd->vdev_spa,
1283 SPA_FEATURE_POOL_CHECKPOINT));
1284 space_map_close(vd->vdev_checkpoint_sm);
1286 * Even though we close the space map, we need to set its
1287 * pointer to NULL. The reason is that vdev_metaslab_fini()
1288 * may be called multiple times for certain operations
1289 * (i.e. when destroying a pool) so we need to ensure that
1290 * this clause never executes twice. This logic is similar
1291 * to the one used for the vdev_ms clause below.
1293 vd->vdev_checkpoint_sm = NULL;
1296 if (vd->vdev_ms != NULL) {
1297 uint64_t count = vd->vdev_ms_count;
1299 metaslab_group_passivate(vd->vdev_mg);
1300 for (uint64_t m = 0; m < count; m++) {
1301 metaslab_t *msp = vd->vdev_ms[m];
1306 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1309 vd->vdev_ms_count = 0;
1311 ASSERT0(vd->vdev_ms_count);
1314 typedef struct vdev_probe_stats {
1315 boolean_t vps_readable;
1316 boolean_t vps_writeable;
1318 } vdev_probe_stats_t;
1321 vdev_probe_done(zio_t *zio)
1323 spa_t *spa = zio->io_spa;
1324 vdev_t *vd = zio->io_vd;
1325 vdev_probe_stats_t *vps = zio->io_private;
1327 ASSERT(vd->vdev_probe_zio != NULL);
1329 if (zio->io_type == ZIO_TYPE_READ) {
1330 if (zio->io_error == 0)
1331 vps->vps_readable = 1;
1332 if (zio->io_error == 0 && spa_writeable(spa)) {
1333 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1334 zio->io_offset, zio->io_size, zio->io_abd,
1335 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1336 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1338 abd_free(zio->io_abd);
1340 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1341 if (zio->io_error == 0)
1342 vps->vps_writeable = 1;
1343 abd_free(zio->io_abd);
1344 } else if (zio->io_type == ZIO_TYPE_NULL) {
1347 vd->vdev_cant_read |= !vps->vps_readable;
1348 vd->vdev_cant_write |= !vps->vps_writeable;
1350 if (vdev_readable(vd) &&
1351 (vdev_writeable(vd) || !spa_writeable(spa))) {
1354 ASSERT(zio->io_error != 0);
1355 vdev_dbgmsg(vd, "failed probe");
1356 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1357 spa, vd, NULL, 0, 0);
1358 zio->io_error = SET_ERROR(ENXIO);
1361 mutex_enter(&vd->vdev_probe_lock);
1362 ASSERT(vd->vdev_probe_zio == zio);
1363 vd->vdev_probe_zio = NULL;
1364 mutex_exit(&vd->vdev_probe_lock);
1366 zio_link_t *zl = NULL;
1367 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1368 if (!vdev_accessible(vd, pio))
1369 pio->io_error = SET_ERROR(ENXIO);
1371 kmem_free(vps, sizeof (*vps));
1376 * Determine whether this device is accessible.
1378 * Read and write to several known locations: the pad regions of each
1379 * vdev label but the first, which we leave alone in case it contains
1383 vdev_probe(vdev_t *vd, zio_t *zio)
1385 spa_t *spa = vd->vdev_spa;
1386 vdev_probe_stats_t *vps = NULL;
1389 ASSERT(vd->vdev_ops->vdev_op_leaf);
1392 * Don't probe the probe.
1394 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1398 * To prevent 'probe storms' when a device fails, we create
1399 * just one probe i/o at a time. All zios that want to probe
1400 * this vdev will become parents of the probe io.
1402 mutex_enter(&vd->vdev_probe_lock);
1404 if ((pio = vd->vdev_probe_zio) == NULL) {
1405 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1407 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1408 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1411 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1413 * vdev_cant_read and vdev_cant_write can only
1414 * transition from TRUE to FALSE when we have the
1415 * SCL_ZIO lock as writer; otherwise they can only
1416 * transition from FALSE to TRUE. This ensures that
1417 * any zio looking at these values can assume that
1418 * failures persist for the life of the I/O. That's
1419 * important because when a device has intermittent
1420 * connectivity problems, we want to ensure that
1421 * they're ascribed to the device (ENXIO) and not
1424 * Since we hold SCL_ZIO as writer here, clear both
1425 * values so the probe can reevaluate from first
1428 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1429 vd->vdev_cant_read = B_FALSE;
1430 vd->vdev_cant_write = B_FALSE;
1433 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1434 vdev_probe_done, vps,
1435 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1438 * We can't change the vdev state in this context, so we
1439 * kick off an async task to do it on our behalf.
1442 vd->vdev_probe_wanted = B_TRUE;
1443 spa_async_request(spa, SPA_ASYNC_PROBE);
1448 zio_add_child(zio, pio);
1450 mutex_exit(&vd->vdev_probe_lock);
1453 ASSERT(zio != NULL);
1457 for (int l = 1; l < VDEV_LABELS; l++) {
1458 zio_nowait(zio_read_phys(pio, vd,
1459 vdev_label_offset(vd->vdev_psize, l,
1460 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1461 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1462 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1463 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1474 vdev_open_child(void *arg)
1478 vd->vdev_open_thread = curthread;
1479 vd->vdev_open_error = vdev_open(vd);
1480 vd->vdev_open_thread = NULL;
1484 vdev_uses_zvols(vdev_t *vd)
1486 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1487 strlen(ZVOL_DIR)) == 0)
1489 for (int c = 0; c < vd->vdev_children; c++)
1490 if (vdev_uses_zvols(vd->vdev_child[c]))
1496 vdev_open_children(vdev_t *vd)
1499 int children = vd->vdev_children;
1501 vd->vdev_nonrot = B_TRUE;
1504 * in order to handle pools on top of zvols, do the opens
1505 * in a single thread so that the same thread holds the
1506 * spa_namespace_lock
1508 if (B_TRUE || vdev_uses_zvols(vd)) {
1509 for (int c = 0; c < children; c++) {
1510 vd->vdev_child[c]->vdev_open_error =
1511 vdev_open(vd->vdev_child[c]);
1512 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1516 tq = taskq_create("vdev_open", children, minclsyspri,
1517 children, children, TASKQ_PREPOPULATE);
1519 for (int c = 0; c < children; c++)
1520 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1525 for (int c = 0; c < children; c++)
1526 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1530 * Compute the raidz-deflation ratio. Note, we hard-code
1531 * in 128k (1 << 17) because it is the "typical" blocksize.
1532 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1533 * otherwise it would inconsistently account for existing bp's.
1536 vdev_set_deflate_ratio(vdev_t *vd)
1538 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1539 vd->vdev_deflate_ratio = (1 << 17) /
1540 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1545 * Prepare a virtual device for access.
1548 vdev_open(vdev_t *vd)
1550 spa_t *spa = vd->vdev_spa;
1553 uint64_t max_osize = 0;
1554 uint64_t asize, max_asize, psize;
1555 uint64_t logical_ashift = 0;
1556 uint64_t physical_ashift = 0;
1558 ASSERT(vd->vdev_open_thread == curthread ||
1559 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1560 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1561 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1562 vd->vdev_state == VDEV_STATE_OFFLINE);
1564 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1565 vd->vdev_cant_read = B_FALSE;
1566 vd->vdev_cant_write = B_FALSE;
1567 vd->vdev_notrim = B_FALSE;
1568 vd->vdev_min_asize = vdev_get_min_asize(vd);
1571 * If this vdev is not removed, check its fault status. If it's
1572 * faulted, bail out of the open.
1574 if (!vd->vdev_removed && vd->vdev_faulted) {
1575 ASSERT(vd->vdev_children == 0);
1576 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1577 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1578 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1579 vd->vdev_label_aux);
1580 return (SET_ERROR(ENXIO));
1581 } else if (vd->vdev_offline) {
1582 ASSERT(vd->vdev_children == 0);
1583 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1584 return (SET_ERROR(ENXIO));
1587 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1588 &logical_ashift, &physical_ashift);
1591 * Reset the vdev_reopening flag so that we actually close
1592 * the vdev on error.
1594 vd->vdev_reopening = B_FALSE;
1595 if (zio_injection_enabled && error == 0)
1596 error = zio_handle_device_injection(vd, NULL, ENXIO);
1599 if (vd->vdev_removed &&
1600 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1601 vd->vdev_removed = B_FALSE;
1603 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1604 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1605 vd->vdev_stat.vs_aux);
1607 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1608 vd->vdev_stat.vs_aux);
1613 vd->vdev_removed = B_FALSE;
1616 * Recheck the faulted flag now that we have confirmed that
1617 * the vdev is accessible. If we're faulted, bail.
1619 if (vd->vdev_faulted) {
1620 ASSERT(vd->vdev_children == 0);
1621 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1622 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1623 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1624 vd->vdev_label_aux);
1625 return (SET_ERROR(ENXIO));
1628 if (vd->vdev_degraded) {
1629 ASSERT(vd->vdev_children == 0);
1630 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1631 VDEV_AUX_ERR_EXCEEDED);
1633 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1637 * For hole or missing vdevs we just return success.
1639 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1642 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1643 trim_map_create(vd);
1645 for (int c = 0; c < vd->vdev_children; c++) {
1646 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1647 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1653 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1654 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1656 if (vd->vdev_children == 0) {
1657 if (osize < SPA_MINDEVSIZE) {
1658 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1659 VDEV_AUX_TOO_SMALL);
1660 return (SET_ERROR(EOVERFLOW));
1663 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1664 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1665 VDEV_LABEL_END_SIZE);
1667 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1668 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1669 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1670 VDEV_AUX_TOO_SMALL);
1671 return (SET_ERROR(EOVERFLOW));
1675 max_asize = max_osize;
1678 vd->vdev_psize = psize;
1681 * Make sure the allocatable size hasn't shrunk too much.
1683 if (asize < vd->vdev_min_asize) {
1684 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1685 VDEV_AUX_BAD_LABEL);
1686 return (SET_ERROR(EINVAL));
1689 vd->vdev_physical_ashift =
1690 MAX(physical_ashift, vd->vdev_physical_ashift);
1691 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1692 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1694 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1695 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1696 VDEV_AUX_ASHIFT_TOO_BIG);
1700 if (vd->vdev_asize == 0) {
1702 * This is the first-ever open, so use the computed values.
1703 * For testing purposes, a higher ashift can be requested.
1705 vd->vdev_asize = asize;
1706 vd->vdev_max_asize = max_asize;
1709 * Make sure the alignment requirement hasn't increased.
1711 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1712 vd->vdev_ops->vdev_op_leaf) {
1713 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1714 VDEV_AUX_BAD_LABEL);
1717 vd->vdev_max_asize = max_asize;
1721 * If all children are healthy we update asize if either:
1722 * The asize has increased, due to a device expansion caused by dynamic
1723 * LUN growth or vdev replacement, and automatic expansion is enabled;
1724 * making the additional space available.
1726 * The asize has decreased, due to a device shrink usually caused by a
1727 * vdev replace with a smaller device. This ensures that calculations
1728 * based of max_asize and asize e.g. esize are always valid. It's safe
1729 * to do this as we've already validated that asize is greater than
1732 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1733 ((asize > vd->vdev_asize &&
1734 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1735 (asize < vd->vdev_asize)))
1736 vd->vdev_asize = asize;
1738 vdev_set_min_asize(vd);
1741 * Ensure we can issue some IO before declaring the
1742 * vdev open for business.
1744 if (vd->vdev_ops->vdev_op_leaf &&
1745 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1746 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1747 VDEV_AUX_ERR_EXCEEDED);
1752 * Track the min and max ashift values for normal data devices.
1754 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1755 !vd->vdev_islog && vd->vdev_aux == NULL) {
1756 if (vd->vdev_ashift > spa->spa_max_ashift)
1757 spa->spa_max_ashift = vd->vdev_ashift;
1758 if (vd->vdev_ashift < spa->spa_min_ashift)
1759 spa->spa_min_ashift = vd->vdev_ashift;
1763 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1764 * resilver. But don't do this if we are doing a reopen for a scrub,
1765 * since this would just restart the scrub we are already doing.
1767 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1768 vdev_resilver_needed(vd, NULL, NULL))
1769 spa_async_request(spa, SPA_ASYNC_RESILVER);
1775 * Called once the vdevs are all opened, this routine validates the label
1776 * contents. This needs to be done before vdev_load() so that we don't
1777 * inadvertently do repair I/Os to the wrong device.
1779 * This function will only return failure if one of the vdevs indicates that it
1780 * has since been destroyed or exported. This is only possible if
1781 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1782 * will be updated but the function will return 0.
1785 vdev_validate(vdev_t *vd)
1787 spa_t *spa = vd->vdev_spa;
1789 uint64_t guid = 0, aux_guid = 0, top_guid;
1794 if (vdev_validate_skip)
1797 for (uint64_t c = 0; c < vd->vdev_children; c++)
1798 if (vdev_validate(vd->vdev_child[c]) != 0)
1799 return (SET_ERROR(EBADF));
1802 * If the device has already failed, or was marked offline, don't do
1803 * any further validation. Otherwise, label I/O will fail and we will
1804 * overwrite the previous state.
1806 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1810 * If we are performing an extreme rewind, we allow for a label that
1811 * was modified at a point after the current txg.
1812 * If config lock is not held do not check for the txg. spa_sync could
1813 * be updating the vdev's label before updating spa_last_synced_txg.
1815 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1816 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1819 txg = spa_last_synced_txg(spa);
1821 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1822 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1823 VDEV_AUX_BAD_LABEL);
1824 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
1825 "txg %llu", (u_longlong_t)txg);
1830 * Determine if this vdev has been split off into another
1831 * pool. If so, then refuse to open it.
1833 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1834 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1835 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1836 VDEV_AUX_SPLIT_POOL);
1838 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1842 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1843 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1844 VDEV_AUX_CORRUPT_DATA);
1846 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1847 ZPOOL_CONFIG_POOL_GUID);
1852 * If config is not trusted then ignore the spa guid check. This is
1853 * necessary because if the machine crashed during a re-guid the new
1854 * guid might have been written to all of the vdev labels, but not the
1855 * cached config. The check will be performed again once we have the
1856 * trusted config from the MOS.
1858 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1859 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1860 VDEV_AUX_CORRUPT_DATA);
1862 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
1863 "match config (%llu != %llu)", (u_longlong_t)guid,
1864 (u_longlong_t)spa_guid(spa));
1868 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1869 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1873 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
1874 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1875 VDEV_AUX_CORRUPT_DATA);
1877 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1882 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
1884 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1885 VDEV_AUX_CORRUPT_DATA);
1887 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1888 ZPOOL_CONFIG_TOP_GUID);
1893 * If this vdev just became a top-level vdev because its sibling was
1894 * detached, it will have adopted the parent's vdev guid -- but the
1895 * label may or may not be on disk yet. Fortunately, either version
1896 * of the label will have the same top guid, so if we're a top-level
1897 * vdev, we can safely compare to that instead.
1898 * However, if the config comes from a cachefile that failed to update
1899 * after the detach, a top-level vdev will appear as a non top-level
1900 * vdev in the config. Also relax the constraints if we perform an
1903 * If we split this vdev off instead, then we also check the
1904 * original pool's guid. We don't want to consider the vdev
1905 * corrupt if it is partway through a split operation.
1907 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
1908 boolean_t mismatch = B_FALSE;
1909 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
1910 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
1913 if (vd->vdev_guid != top_guid &&
1914 vd->vdev_top->vdev_guid != guid)
1919 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1920 VDEV_AUX_CORRUPT_DATA);
1922 vdev_dbgmsg(vd, "vdev_validate: config guid "
1923 "doesn't match label guid");
1924 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
1925 (u_longlong_t)vd->vdev_guid,
1926 (u_longlong_t)vd->vdev_top->vdev_guid);
1927 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
1928 "aux_guid %llu", (u_longlong_t)guid,
1929 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
1934 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1936 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1937 VDEV_AUX_CORRUPT_DATA);
1939 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1940 ZPOOL_CONFIG_POOL_STATE);
1947 * If this is a verbatim import, no need to check the
1948 * state of the pool.
1950 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1951 spa_load_state(spa) == SPA_LOAD_OPEN &&
1952 state != POOL_STATE_ACTIVE) {
1953 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
1954 "for spa %s", (u_longlong_t)state, spa->spa_name);
1955 return (SET_ERROR(EBADF));
1959 * If we were able to open and validate a vdev that was
1960 * previously marked permanently unavailable, clear that state
1963 if (vd->vdev_not_present)
1964 vd->vdev_not_present = 0;
1970 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
1972 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
1973 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
1974 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1975 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
1976 dvd->vdev_path, svd->vdev_path);
1977 spa_strfree(dvd->vdev_path);
1978 dvd->vdev_path = spa_strdup(svd->vdev_path);
1980 } else if (svd->vdev_path != NULL) {
1981 dvd->vdev_path = spa_strdup(svd->vdev_path);
1982 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1983 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
1988 * Recursively copy vdev paths from one vdev to another. Source and destination
1989 * vdev trees must have same geometry otherwise return error. Intended to copy
1990 * paths from userland config into MOS config.
1993 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
1995 if ((svd->vdev_ops == &vdev_missing_ops) ||
1996 (svd->vdev_ishole && dvd->vdev_ishole) ||
1997 (dvd->vdev_ops == &vdev_indirect_ops))
2000 if (svd->vdev_ops != dvd->vdev_ops) {
2001 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
2002 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
2003 return (SET_ERROR(EINVAL));
2006 if (svd->vdev_guid != dvd->vdev_guid) {
2007 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
2008 "%llu)", (u_longlong_t)svd->vdev_guid,
2009 (u_longlong_t)dvd->vdev_guid);
2010 return (SET_ERROR(EINVAL));
2013 if (svd->vdev_children != dvd->vdev_children) {
2014 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2015 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2016 (u_longlong_t)dvd->vdev_children);
2017 return (SET_ERROR(EINVAL));
2020 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2021 int error = vdev_copy_path_strict(svd->vdev_child[i],
2022 dvd->vdev_child[i]);
2027 if (svd->vdev_ops->vdev_op_leaf)
2028 vdev_copy_path_impl(svd, dvd);
2034 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2036 ASSERT(stvd->vdev_top == stvd);
2037 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2039 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2040 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2043 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2047 * The idea here is that while a vdev can shift positions within
2048 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2049 * step outside of it.
2051 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2053 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2056 ASSERT(vd->vdev_ops->vdev_op_leaf);
2058 vdev_copy_path_impl(vd, dvd);
2062 * Recursively copy vdev paths from one root vdev to another. Source and
2063 * destination vdev trees may differ in geometry. For each destination leaf
2064 * vdev, search a vdev with the same guid and top vdev id in the source.
2065 * Intended to copy paths from userland config into MOS config.
2068 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2070 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2071 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2072 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2074 for (uint64_t i = 0; i < children; i++) {
2075 vdev_copy_path_search(srvd->vdev_child[i],
2076 drvd->vdev_child[i]);
2081 * Close a virtual device.
2084 vdev_close(vdev_t *vd)
2086 spa_t *spa = vd->vdev_spa;
2087 vdev_t *pvd = vd->vdev_parent;
2089 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2092 * If our parent is reopening, then we are as well, unless we are
2095 if (pvd != NULL && pvd->vdev_reopening)
2096 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2098 vd->vdev_ops->vdev_op_close(vd);
2100 vdev_cache_purge(vd);
2102 if (vd->vdev_ops->vdev_op_leaf)
2103 trim_map_destroy(vd);
2106 * We record the previous state before we close it, so that if we are
2107 * doing a reopen(), we don't generate FMA ereports if we notice that
2108 * it's still faulted.
2110 vd->vdev_prevstate = vd->vdev_state;
2112 if (vd->vdev_offline)
2113 vd->vdev_state = VDEV_STATE_OFFLINE;
2115 vd->vdev_state = VDEV_STATE_CLOSED;
2116 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2120 vdev_hold(vdev_t *vd)
2122 spa_t *spa = vd->vdev_spa;
2124 ASSERT(spa_is_root(spa));
2125 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2128 for (int c = 0; c < vd->vdev_children; c++)
2129 vdev_hold(vd->vdev_child[c]);
2131 if (vd->vdev_ops->vdev_op_leaf)
2132 vd->vdev_ops->vdev_op_hold(vd);
2136 vdev_rele(vdev_t *vd)
2138 spa_t *spa = vd->vdev_spa;
2140 ASSERT(spa_is_root(spa));
2141 for (int c = 0; c < vd->vdev_children; c++)
2142 vdev_rele(vd->vdev_child[c]);
2144 if (vd->vdev_ops->vdev_op_leaf)
2145 vd->vdev_ops->vdev_op_rele(vd);
2149 * Reopen all interior vdevs and any unopened leaves. We don't actually
2150 * reopen leaf vdevs which had previously been opened as they might deadlock
2151 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2152 * If the leaf has never been opened then open it, as usual.
2155 vdev_reopen(vdev_t *vd)
2157 spa_t *spa = vd->vdev_spa;
2159 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2161 /* set the reopening flag unless we're taking the vdev offline */
2162 vd->vdev_reopening = !vd->vdev_offline;
2164 (void) vdev_open(vd);
2167 * Call vdev_validate() here to make sure we have the same device.
2168 * Otherwise, a device with an invalid label could be successfully
2169 * opened in response to vdev_reopen().
2172 (void) vdev_validate_aux(vd);
2173 if (vdev_readable(vd) && vdev_writeable(vd) &&
2174 vd->vdev_aux == &spa->spa_l2cache &&
2175 !l2arc_vdev_present(vd))
2176 l2arc_add_vdev(spa, vd);
2178 (void) vdev_validate(vd);
2182 * Reassess parent vdev's health.
2184 vdev_propagate_state(vd);
2188 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2193 * Normally, partial opens (e.g. of a mirror) are allowed.
2194 * For a create, however, we want to fail the request if
2195 * there are any components we can't open.
2197 error = vdev_open(vd);
2199 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2201 return (error ? error : ENXIO);
2205 * Recursively load DTLs and initialize all labels.
2207 if ((error = vdev_dtl_load(vd)) != 0 ||
2208 (error = vdev_label_init(vd, txg, isreplacing ?
2209 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2218 vdev_metaslab_set_size(vdev_t *vd)
2220 uint64_t asize = vd->vdev_asize;
2221 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2225 * There are two dimensions to the metaslab sizing calculation:
2226 * the size of the metaslab and the count of metaslabs per vdev.
2228 * The default values used below are a good balance between memory
2229 * usage (larger metaslab size means more memory needed for loaded
2230 * metaslabs; more metaslabs means more memory needed for the
2231 * metaslab_t structs), metaslab load time (larger metaslabs take
2232 * longer to load), and metaslab sync time (more metaslabs means
2233 * more time spent syncing all of them).
2235 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2236 * The range of the dimensions are as follows:
2238 * 2^29 <= ms_size <= 2^34
2239 * 16 <= ms_count <= 131,072
2241 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2242 * at least 512MB (2^29) to minimize fragmentation effects when
2243 * testing with smaller devices. However, the count constraint
2244 * of at least 16 metaslabs will override this minimum size goal.
2246 * On the upper end of vdev sizes, we aim for a maximum metaslab
2247 * size of 16GB. However, we will cap the total count to 2^17
2248 * metaslabs to keep our memory footprint in check and let the
2249 * metaslab size grow from there if that limit is hit.
2251 * The net effect of applying above constrains is summarized below.
2253 * vdev size metaslab count
2254 * --------------|-----------------
2256 * 8GB - 100GB one per 512MB
2258 * 3TB - 2PB one per 16GB
2260 * --------------------------------
2262 * Finally, note that all of the above calculate the initial
2263 * number of metaslabs. Expanding a top-level vdev will result
2264 * in additional metaslabs being allocated making it possible
2265 * to exceed the zfs_vdev_ms_count_limit.
2268 if (ms_count < zfs_vdev_min_ms_count)
2269 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2270 else if (ms_count > zfs_vdev_default_ms_count)
2271 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2273 ms_shift = zfs_vdev_default_ms_shift;
2275 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2276 ms_shift = SPA_MAXBLOCKSHIFT;
2277 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2278 ms_shift = zfs_vdev_max_ms_shift;
2279 /* cap the total count to constrain memory footprint */
2280 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2281 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2284 vd->vdev_ms_shift = ms_shift;
2285 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2289 * Maximize performance by inflating the configured ashift for top level
2290 * vdevs to be as close to the physical ashift as possible while maintaining
2291 * administrator defined limits and ensuring it doesn't go below the
2295 vdev_ashift_optimize(vdev_t *vd)
2297 if (vd == vd->vdev_top) {
2298 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
2299 vd->vdev_ashift = MIN(
2300 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
2301 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
2304 * Unusual case where logical ashift > physical ashift
2305 * so we can't cap the calculated ashift based on max
2306 * ashift as that would cause failures.
2307 * We still check if we need to increase it to match
2310 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
2317 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2319 ASSERT(vd == vd->vdev_top);
2320 /* indirect vdevs don't have metaslabs or dtls */
2321 ASSERT(vdev_is_concrete(vd) || flags == 0);
2322 ASSERT(ISP2(flags));
2323 ASSERT(spa_writeable(vd->vdev_spa));
2325 if (flags & VDD_METASLAB)
2326 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2328 if (flags & VDD_DTL)
2329 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2331 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2335 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2337 for (int c = 0; c < vd->vdev_children; c++)
2338 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2340 if (vd->vdev_ops->vdev_op_leaf)
2341 vdev_dirty(vd->vdev_top, flags, vd, txg);
2347 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2348 * the vdev has less than perfect replication. There are four kinds of DTL:
2350 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2352 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2354 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2355 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2356 * txgs that was scrubbed.
2358 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2359 * persistent errors or just some device being offline.
2360 * Unlike the other three, the DTL_OUTAGE map is not generally
2361 * maintained; it's only computed when needed, typically to
2362 * determine whether a device can be detached.
2364 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2365 * either has the data or it doesn't.
2367 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2368 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2369 * if any child is less than fully replicated, then so is its parent.
2370 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2371 * comprising only those txgs which appear in 'maxfaults' or more children;
2372 * those are the txgs we don't have enough replication to read. For example,
2373 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2374 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2375 * two child DTL_MISSING maps.
2377 * It should be clear from the above that to compute the DTLs and outage maps
2378 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2379 * Therefore, that is all we keep on disk. When loading the pool, or after
2380 * a configuration change, we generate all other DTLs from first principles.
2383 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2385 range_tree_t *rt = vd->vdev_dtl[t];
2387 ASSERT(t < DTL_TYPES);
2388 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2389 ASSERT(spa_writeable(vd->vdev_spa));
2391 mutex_enter(&vd->vdev_dtl_lock);
2392 if (!range_tree_contains(rt, txg, size))
2393 range_tree_add(rt, txg, size);
2394 mutex_exit(&vd->vdev_dtl_lock);
2398 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2400 range_tree_t *rt = vd->vdev_dtl[t];
2401 boolean_t dirty = B_FALSE;
2403 ASSERT(t < DTL_TYPES);
2404 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2407 * While we are loading the pool, the DTLs have not been loaded yet.
2408 * Ignore the DTLs and try all devices. This avoids a recursive
2409 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2410 * when loading the pool (relying on the checksum to ensure that
2411 * we get the right data -- note that we while loading, we are
2412 * only reading the MOS, which is always checksummed).
2414 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2417 mutex_enter(&vd->vdev_dtl_lock);
2418 if (!range_tree_is_empty(rt))
2419 dirty = range_tree_contains(rt, txg, size);
2420 mutex_exit(&vd->vdev_dtl_lock);
2426 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2428 range_tree_t *rt = vd->vdev_dtl[t];
2431 mutex_enter(&vd->vdev_dtl_lock);
2432 empty = range_tree_is_empty(rt);
2433 mutex_exit(&vd->vdev_dtl_lock);
2439 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2442 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2444 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2446 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2447 vd->vdev_ops->vdev_op_leaf)
2450 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2454 * Returns the lowest txg in the DTL range.
2457 vdev_dtl_min(vdev_t *vd)
2461 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2462 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2463 ASSERT0(vd->vdev_children);
2465 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2466 return (rs->rs_start - 1);
2470 * Returns the highest txg in the DTL.
2473 vdev_dtl_max(vdev_t *vd)
2477 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2478 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2479 ASSERT0(vd->vdev_children);
2481 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2482 return (rs->rs_end);
2486 * Determine if a resilvering vdev should remove any DTL entries from
2487 * its range. If the vdev was resilvering for the entire duration of the
2488 * scan then it should excise that range from its DTLs. Otherwise, this
2489 * vdev is considered partially resilvered and should leave its DTL
2490 * entries intact. The comment in vdev_dtl_reassess() describes how we
2494 vdev_dtl_should_excise(vdev_t *vd)
2496 spa_t *spa = vd->vdev_spa;
2497 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2499 ASSERT0(scn->scn_phys.scn_errors);
2500 ASSERT0(vd->vdev_children);
2502 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2505 if (vd->vdev_resilver_txg == 0 ||
2506 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2510 * When a resilver is initiated the scan will assign the scn_max_txg
2511 * value to the highest txg value that exists in all DTLs. If this
2512 * device's max DTL is not part of this scan (i.e. it is not in
2513 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2516 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2517 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2518 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2519 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2526 * Reassess DTLs after a config change or scrub completion.
2529 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2531 spa_t *spa = vd->vdev_spa;
2535 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2537 for (int c = 0; c < vd->vdev_children; c++)
2538 vdev_dtl_reassess(vd->vdev_child[c], txg,
2539 scrub_txg, scrub_done);
2541 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2544 if (vd->vdev_ops->vdev_op_leaf) {
2545 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2547 mutex_enter(&vd->vdev_dtl_lock);
2550 * If we've completed a scan cleanly then determine
2551 * if this vdev should remove any DTLs. We only want to
2552 * excise regions on vdevs that were available during
2553 * the entire duration of this scan.
2555 if (scrub_txg != 0 &&
2556 (spa->spa_scrub_started ||
2557 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2558 vdev_dtl_should_excise(vd)) {
2560 * We completed a scrub up to scrub_txg. If we
2561 * did it without rebooting, then the scrub dtl
2562 * will be valid, so excise the old region and
2563 * fold in the scrub dtl. Otherwise, leave the
2564 * dtl as-is if there was an error.
2566 * There's little trick here: to excise the beginning
2567 * of the DTL_MISSING map, we put it into a reference
2568 * tree and then add a segment with refcnt -1 that
2569 * covers the range [0, scrub_txg). This means
2570 * that each txg in that range has refcnt -1 or 0.
2571 * We then add DTL_SCRUB with a refcnt of 2, so that
2572 * entries in the range [0, scrub_txg) will have a
2573 * positive refcnt -- either 1 or 2. We then convert
2574 * the reference tree into the new DTL_MISSING map.
2576 space_reftree_create(&reftree);
2577 space_reftree_add_map(&reftree,
2578 vd->vdev_dtl[DTL_MISSING], 1);
2579 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2580 space_reftree_add_map(&reftree,
2581 vd->vdev_dtl[DTL_SCRUB], 2);
2582 space_reftree_generate_map(&reftree,
2583 vd->vdev_dtl[DTL_MISSING], 1);
2584 space_reftree_destroy(&reftree);
2586 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2587 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2588 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2590 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2591 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2592 if (!vdev_readable(vd))
2593 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2595 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2596 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2599 * If the vdev was resilvering and no longer has any
2600 * DTLs then reset its resilvering flag and dirty
2601 * the top level so that we persist the change.
2603 if (vd->vdev_resilver_txg != 0 &&
2604 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2605 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2606 vd->vdev_resilver_txg = 0;
2607 vdev_config_dirty(vd->vdev_top);
2610 mutex_exit(&vd->vdev_dtl_lock);
2613 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2617 mutex_enter(&vd->vdev_dtl_lock);
2618 for (int t = 0; t < DTL_TYPES; t++) {
2619 /* account for child's outage in parent's missing map */
2620 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2622 continue; /* leaf vdevs only */
2623 if (t == DTL_PARTIAL)
2624 minref = 1; /* i.e. non-zero */
2625 else if (vd->vdev_nparity != 0)
2626 minref = vd->vdev_nparity + 1; /* RAID-Z */
2628 minref = vd->vdev_children; /* any kind of mirror */
2629 space_reftree_create(&reftree);
2630 for (int c = 0; c < vd->vdev_children; c++) {
2631 vdev_t *cvd = vd->vdev_child[c];
2632 mutex_enter(&cvd->vdev_dtl_lock);
2633 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2634 mutex_exit(&cvd->vdev_dtl_lock);
2636 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2637 space_reftree_destroy(&reftree);
2639 mutex_exit(&vd->vdev_dtl_lock);
2643 vdev_dtl_load(vdev_t *vd)
2645 spa_t *spa = vd->vdev_spa;
2646 objset_t *mos = spa->spa_meta_objset;
2649 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2650 ASSERT(vdev_is_concrete(vd));
2652 error = space_map_open(&vd->vdev_dtl_sm, mos,
2653 vd->vdev_dtl_object, 0, -1ULL, 0);
2656 ASSERT(vd->vdev_dtl_sm != NULL);
2658 mutex_enter(&vd->vdev_dtl_lock);
2661 * Now that we've opened the space_map we need to update
2664 space_map_update(vd->vdev_dtl_sm);
2666 error = space_map_load(vd->vdev_dtl_sm,
2667 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2668 mutex_exit(&vd->vdev_dtl_lock);
2673 for (int c = 0; c < vd->vdev_children; c++) {
2674 error = vdev_dtl_load(vd->vdev_child[c]);
2683 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2685 spa_t *spa = vd->vdev_spa;
2687 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2688 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2693 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2695 spa_t *spa = vd->vdev_spa;
2696 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2697 DMU_OT_NONE, 0, tx);
2700 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2707 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2709 if (vd->vdev_ops != &vdev_hole_ops &&
2710 vd->vdev_ops != &vdev_missing_ops &&
2711 vd->vdev_ops != &vdev_root_ops &&
2712 !vd->vdev_top->vdev_removing) {
2713 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2714 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2716 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2717 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2720 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2721 vdev_construct_zaps(vd->vdev_child[i], tx);
2726 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2728 spa_t *spa = vd->vdev_spa;
2729 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2730 objset_t *mos = spa->spa_meta_objset;
2731 range_tree_t *rtsync;
2733 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2735 ASSERT(vdev_is_concrete(vd));
2736 ASSERT(vd->vdev_ops->vdev_op_leaf);
2738 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2740 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2741 mutex_enter(&vd->vdev_dtl_lock);
2742 space_map_free(vd->vdev_dtl_sm, tx);
2743 space_map_close(vd->vdev_dtl_sm);
2744 vd->vdev_dtl_sm = NULL;
2745 mutex_exit(&vd->vdev_dtl_lock);
2748 * We only destroy the leaf ZAP for detached leaves or for
2749 * removed log devices. Removed data devices handle leaf ZAP
2750 * cleanup later, once cancellation is no longer possible.
2752 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2753 vd->vdev_top->vdev_islog)) {
2754 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2755 vd->vdev_leaf_zap = 0;
2762 if (vd->vdev_dtl_sm == NULL) {
2763 uint64_t new_object;
2765 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx);
2766 VERIFY3U(new_object, !=, 0);
2768 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2770 ASSERT(vd->vdev_dtl_sm != NULL);
2773 rtsync = range_tree_create(NULL, NULL);
2775 mutex_enter(&vd->vdev_dtl_lock);
2776 range_tree_walk(rt, range_tree_add, rtsync);
2777 mutex_exit(&vd->vdev_dtl_lock);
2779 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx);
2780 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
2781 range_tree_vacate(rtsync, NULL, NULL);
2783 range_tree_destroy(rtsync);
2786 * If the object for the space map has changed then dirty
2787 * the top level so that we update the config.
2789 if (object != space_map_object(vd->vdev_dtl_sm)) {
2790 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2791 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2792 (u_longlong_t)object,
2793 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2794 vdev_config_dirty(vd->vdev_top);
2799 mutex_enter(&vd->vdev_dtl_lock);
2800 space_map_update(vd->vdev_dtl_sm);
2801 mutex_exit(&vd->vdev_dtl_lock);
2805 * Determine whether the specified vdev can be offlined/detached/removed
2806 * without losing data.
2809 vdev_dtl_required(vdev_t *vd)
2811 spa_t *spa = vd->vdev_spa;
2812 vdev_t *tvd = vd->vdev_top;
2813 uint8_t cant_read = vd->vdev_cant_read;
2816 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2818 if (vd == spa->spa_root_vdev || vd == tvd)
2822 * Temporarily mark the device as unreadable, and then determine
2823 * whether this results in any DTL outages in the top-level vdev.
2824 * If not, we can safely offline/detach/remove the device.
2826 vd->vdev_cant_read = B_TRUE;
2827 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2828 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2829 vd->vdev_cant_read = cant_read;
2830 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2832 if (!required && zio_injection_enabled)
2833 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2839 * Determine if resilver is needed, and if so the txg range.
2842 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2844 boolean_t needed = B_FALSE;
2845 uint64_t thismin = UINT64_MAX;
2846 uint64_t thismax = 0;
2848 if (vd->vdev_children == 0) {
2849 mutex_enter(&vd->vdev_dtl_lock);
2850 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2851 vdev_writeable(vd)) {
2853 thismin = vdev_dtl_min(vd);
2854 thismax = vdev_dtl_max(vd);
2857 mutex_exit(&vd->vdev_dtl_lock);
2859 for (int c = 0; c < vd->vdev_children; c++) {
2860 vdev_t *cvd = vd->vdev_child[c];
2861 uint64_t cmin, cmax;
2863 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2864 thismin = MIN(thismin, cmin);
2865 thismax = MAX(thismax, cmax);
2871 if (needed && minp) {
2879 * Gets the checkpoint space map object from the vdev's ZAP.
2880 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2881 * or the ZAP doesn't exist yet.
2884 vdev_checkpoint_sm_object(vdev_t *vd)
2886 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
2887 if (vd->vdev_top_zap == 0) {
2891 uint64_t sm_obj = 0;
2892 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
2893 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj);
2895 ASSERT(err == 0 || err == ENOENT);
2901 vdev_load(vdev_t *vd)
2905 * Recursively load all children.
2907 for (int c = 0; c < vd->vdev_children; c++) {
2908 error = vdev_load(vd->vdev_child[c]);
2914 vdev_set_deflate_ratio(vd);
2917 * If this is a top-level vdev, initialize its metaslabs.
2919 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2920 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2921 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2922 VDEV_AUX_CORRUPT_DATA);
2923 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2924 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2925 (u_longlong_t)vd->vdev_asize);
2926 return (SET_ERROR(ENXIO));
2927 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2928 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2929 "[error=%d]", error);
2930 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2931 VDEV_AUX_CORRUPT_DATA);
2935 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd);
2936 if (checkpoint_sm_obj != 0) {
2937 objset_t *mos = spa_meta_objset(vd->vdev_spa);
2938 ASSERT(vd->vdev_asize != 0);
2939 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
2941 if ((error = space_map_open(&vd->vdev_checkpoint_sm,
2942 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
2943 vd->vdev_ashift))) {
2944 vdev_dbgmsg(vd, "vdev_load: space_map_open "
2945 "failed for checkpoint spacemap (obj %llu) "
2947 (u_longlong_t)checkpoint_sm_obj, error);
2950 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
2951 space_map_update(vd->vdev_checkpoint_sm);
2954 * Since the checkpoint_sm contains free entries
2955 * exclusively we can use sm_alloc to indicate the
2956 * culmulative checkpointed space that has been freed.
2958 vd->vdev_stat.vs_checkpoint_space =
2959 -vd->vdev_checkpoint_sm->sm_alloc;
2960 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
2961 vd->vdev_stat.vs_checkpoint_space;
2966 * If this is a leaf vdev, load its DTL.
2968 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2969 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2970 VDEV_AUX_CORRUPT_DATA);
2971 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
2972 "[error=%d]", error);
2976 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2977 if (obsolete_sm_object != 0) {
2978 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2979 ASSERT(vd->vdev_asize != 0);
2980 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
2982 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2983 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2984 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2985 VDEV_AUX_CORRUPT_DATA);
2986 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
2987 "obsolete spacemap (obj %llu) [error=%d]",
2988 (u_longlong_t)obsolete_sm_object, error);
2991 space_map_update(vd->vdev_obsolete_sm);
2998 * The special vdev case is used for hot spares and l2cache devices. Its
2999 * sole purpose it to set the vdev state for the associated vdev. To do this,
3000 * we make sure that we can open the underlying device, then try to read the
3001 * label, and make sure that the label is sane and that it hasn't been
3002 * repurposed to another pool.
3005 vdev_validate_aux(vdev_t *vd)
3008 uint64_t guid, version;
3011 if (!vdev_readable(vd))
3014 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3015 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3016 VDEV_AUX_CORRUPT_DATA);
3020 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3021 !SPA_VERSION_IS_SUPPORTED(version) ||
3022 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3023 guid != vd->vdev_guid ||
3024 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3025 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3026 VDEV_AUX_CORRUPT_DATA);
3032 * We don't actually check the pool state here. If it's in fact in
3033 * use by another pool, we update this fact on the fly when requested.
3040 * Free the objects used to store this vdev's spacemaps, and the array
3041 * that points to them.
3044 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3046 if (vd->vdev_ms_array == 0)
3049 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3050 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3051 size_t array_bytes = array_count * sizeof (uint64_t);
3052 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3053 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3054 array_bytes, smobj_array, 0));
3056 for (uint64_t i = 0; i < array_count; i++) {
3057 uint64_t smobj = smobj_array[i];
3061 space_map_free_obj(mos, smobj, tx);
3064 kmem_free(smobj_array, array_bytes);
3065 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3066 vd->vdev_ms_array = 0;
3070 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3072 spa_t *spa = vd->vdev_spa;
3074 ASSERT(vd->vdev_islog);
3075 ASSERT(vd == vd->vdev_top);
3076 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3078 if (vd->vdev_ms != NULL) {
3079 metaslab_group_t *mg = vd->vdev_mg;
3081 metaslab_group_histogram_verify(mg);
3082 metaslab_class_histogram_verify(mg->mg_class);
3084 for (int m = 0; m < vd->vdev_ms_count; m++) {
3085 metaslab_t *msp = vd->vdev_ms[m];
3087 if (msp == NULL || msp->ms_sm == NULL)
3090 mutex_enter(&msp->ms_lock);
3092 * If the metaslab was not loaded when the vdev
3093 * was removed then the histogram accounting may
3094 * not be accurate. Update the histogram information
3095 * here so that we ensure that the metaslab group
3096 * and metaslab class are up-to-date.
3098 metaslab_group_histogram_remove(mg, msp);
3100 VERIFY0(space_map_allocated(msp->ms_sm));
3101 space_map_close(msp->ms_sm);
3103 mutex_exit(&msp->ms_lock);
3106 if (vd->vdev_checkpoint_sm != NULL) {
3107 ASSERT(spa_has_checkpoint(spa));
3108 space_map_close(vd->vdev_checkpoint_sm);
3109 vd->vdev_checkpoint_sm = NULL;
3112 metaslab_group_histogram_verify(mg);
3113 metaslab_class_histogram_verify(mg->mg_class);
3114 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
3115 ASSERT0(mg->mg_histogram[i]);
3118 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3120 vdev_destroy_spacemaps(vd, tx);
3121 if (vd->vdev_top_zap != 0) {
3122 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3123 vd->vdev_top_zap = 0;
3130 vdev_sync_done(vdev_t *vd, uint64_t txg)
3133 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3135 ASSERT(vdev_is_concrete(vd));
3137 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3139 metaslab_sync_done(msp, txg);
3142 metaslab_sync_reassess(vd->vdev_mg);
3146 vdev_sync(vdev_t *vd, uint64_t txg)
3148 spa_t *spa = vd->vdev_spa;
3153 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3156 ASSERT(vd->vdev_removing ||
3157 vd->vdev_ops == &vdev_indirect_ops);
3159 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3160 vdev_indirect_sync_obsolete(vd, tx);
3164 * If the vdev is indirect, it can't have dirty
3165 * metaslabs or DTLs.
3167 if (vd->vdev_ops == &vdev_indirect_ops) {
3168 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3169 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3174 ASSERT(vdev_is_concrete(vd));
3176 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3177 !vd->vdev_removing) {
3178 ASSERT(vd == vd->vdev_top);
3179 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3180 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3181 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3182 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3183 ASSERT(vd->vdev_ms_array != 0);
3184 vdev_config_dirty(vd);
3188 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3189 metaslab_sync(msp, txg);
3190 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3193 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3194 vdev_dtl_sync(lvd, txg);
3197 * If this is an empty log device being removed, destroy the
3198 * metadata associated with it.
3200 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3201 vdev_remove_empty_log(vd, txg);
3203 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3207 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3209 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3213 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3214 * not be opened, and no I/O is attempted.
3217 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3221 spa_vdev_state_enter(spa, SCL_NONE);
3223 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3224 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3226 if (!vd->vdev_ops->vdev_op_leaf)
3227 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3232 * We don't directly use the aux state here, but if we do a
3233 * vdev_reopen(), we need this value to be present to remember why we
3236 vd->vdev_label_aux = aux;
3239 * Faulted state takes precedence over degraded.
3241 vd->vdev_delayed_close = B_FALSE;
3242 vd->vdev_faulted = 1ULL;
3243 vd->vdev_degraded = 0ULL;
3244 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3247 * If this device has the only valid copy of the data, then
3248 * back off and simply mark the vdev as degraded instead.
3250 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3251 vd->vdev_degraded = 1ULL;
3252 vd->vdev_faulted = 0ULL;
3255 * If we reopen the device and it's not dead, only then do we
3260 if (vdev_readable(vd))
3261 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3264 return (spa_vdev_state_exit(spa, vd, 0));
3268 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3269 * user that something is wrong. The vdev continues to operate as normal as far
3270 * as I/O is concerned.
3273 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3277 spa_vdev_state_enter(spa, SCL_NONE);
3279 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3280 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3282 if (!vd->vdev_ops->vdev_op_leaf)
3283 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3286 * If the vdev is already faulted, then don't do anything.
3288 if (vd->vdev_faulted || vd->vdev_degraded)
3289 return (spa_vdev_state_exit(spa, NULL, 0));
3291 vd->vdev_degraded = 1ULL;
3292 if (!vdev_is_dead(vd))
3293 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3296 return (spa_vdev_state_exit(spa, vd, 0));
3300 * Online the given vdev.
3302 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3303 * spare device should be detached when the device finishes resilvering.
3304 * Second, the online should be treated like a 'test' online case, so no FMA
3305 * events are generated if the device fails to open.
3308 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3310 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3311 boolean_t wasoffline;
3312 vdev_state_t oldstate;
3314 spa_vdev_state_enter(spa, SCL_NONE);
3316 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3317 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3319 if (!vd->vdev_ops->vdev_op_leaf)
3320 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3322 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3323 oldstate = vd->vdev_state;
3326 vd->vdev_offline = B_FALSE;
3327 vd->vdev_tmpoffline = B_FALSE;
3328 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3329 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3331 /* XXX - L2ARC 1.0 does not support expansion */
3332 if (!vd->vdev_aux) {
3333 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3334 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
3338 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3340 if (!vd->vdev_aux) {
3341 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3342 pvd->vdev_expanding = B_FALSE;
3346 *newstate = vd->vdev_state;
3347 if ((flags & ZFS_ONLINE_UNSPARE) &&
3348 !vdev_is_dead(vd) && vd->vdev_parent &&
3349 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3350 vd->vdev_parent->vdev_child[0] == vd)
3351 vd->vdev_unspare = B_TRUE;
3353 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3355 /* XXX - L2ARC 1.0 does not support expansion */
3357 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3358 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3361 /* Restart initializing if necessary */
3362 mutex_enter(&vd->vdev_initialize_lock);
3363 if (vdev_writeable(vd) &&
3364 vd->vdev_initialize_thread == NULL &&
3365 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3366 (void) vdev_initialize(vd);
3368 mutex_exit(&vd->vdev_initialize_lock);
3371 (oldstate < VDEV_STATE_DEGRADED &&
3372 vd->vdev_state >= VDEV_STATE_DEGRADED))
3373 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3375 return (spa_vdev_state_exit(spa, vd, 0));
3379 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3383 uint64_t generation;
3384 metaslab_group_t *mg;
3387 spa_vdev_state_enter(spa, SCL_ALLOC);
3389 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3390 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3392 if (!vd->vdev_ops->vdev_op_leaf)
3393 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3397 generation = spa->spa_config_generation + 1;
3400 * If the device isn't already offline, try to offline it.
3402 if (!vd->vdev_offline) {
3404 * If this device has the only valid copy of some data,
3405 * don't allow it to be offlined. Log devices are always
3408 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3409 vdev_dtl_required(vd))
3410 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3413 * If the top-level is a slog and it has had allocations
3414 * then proceed. We check that the vdev's metaslab group
3415 * is not NULL since it's possible that we may have just
3416 * added this vdev but not yet initialized its metaslabs.
3418 if (tvd->vdev_islog && mg != NULL) {
3420 * Prevent any future allocations.
3422 metaslab_group_passivate(mg);
3423 (void) spa_vdev_state_exit(spa, vd, 0);
3425 error = spa_reset_logs(spa);
3428 * If the log device was successfully reset but has
3429 * checkpointed data, do not offline it.
3432 tvd->vdev_checkpoint_sm != NULL) {
3433 ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc,
3435 error = ZFS_ERR_CHECKPOINT_EXISTS;
3438 spa_vdev_state_enter(spa, SCL_ALLOC);
3441 * Check to see if the config has changed.
3443 if (error || generation != spa->spa_config_generation) {
3444 metaslab_group_activate(mg);
3446 return (spa_vdev_state_exit(spa,
3448 (void) spa_vdev_state_exit(spa, vd, 0);
3451 ASSERT0(tvd->vdev_stat.vs_alloc);
3455 * Offline this device and reopen its top-level vdev.
3456 * If the top-level vdev is a log device then just offline
3457 * it. Otherwise, if this action results in the top-level
3458 * vdev becoming unusable, undo it and fail the request.
3460 vd->vdev_offline = B_TRUE;
3463 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3464 vdev_is_dead(tvd)) {
3465 vd->vdev_offline = B_FALSE;
3467 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3471 * Add the device back into the metaslab rotor so that
3472 * once we online the device it's open for business.
3474 if (tvd->vdev_islog && mg != NULL)
3475 metaslab_group_activate(mg);
3478 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3480 return (spa_vdev_state_exit(spa, vd, 0));
3484 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3488 mutex_enter(&spa->spa_vdev_top_lock);
3489 error = vdev_offline_locked(spa, guid, flags);
3490 mutex_exit(&spa->spa_vdev_top_lock);
3496 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3497 * vdev_offline(), we assume the spa config is locked. We also clear all
3498 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3501 vdev_clear(spa_t *spa, vdev_t *vd)
3503 vdev_t *rvd = spa->spa_root_vdev;
3505 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3510 vd->vdev_stat.vs_read_errors = 0;
3511 vd->vdev_stat.vs_write_errors = 0;
3512 vd->vdev_stat.vs_checksum_errors = 0;
3514 for (int c = 0; c < vd->vdev_children; c++)
3515 vdev_clear(spa, vd->vdev_child[c]);
3518 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
3519 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
3521 for (int c = 0; c < spa->spa_spares.sav_count; c++)
3522 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
3526 * It makes no sense to "clear" an indirect vdev.
3528 if (!vdev_is_concrete(vd))
3532 * If we're in the FAULTED state or have experienced failed I/O, then
3533 * clear the persistent state and attempt to reopen the device. We
3534 * also mark the vdev config dirty, so that the new faulted state is
3535 * written out to disk.
3537 if (vd->vdev_faulted || vd->vdev_degraded ||
3538 !vdev_readable(vd) || !vdev_writeable(vd)) {
3541 * When reopening in reponse to a clear event, it may be due to
3542 * a fmadm repair request. In this case, if the device is
3543 * still broken, we want to still post the ereport again.
3545 vd->vdev_forcefault = B_TRUE;
3547 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3548 vd->vdev_cant_read = B_FALSE;
3549 vd->vdev_cant_write = B_FALSE;
3551 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3553 vd->vdev_forcefault = B_FALSE;
3555 if (vd != rvd && vdev_writeable(vd->vdev_top))
3556 vdev_state_dirty(vd->vdev_top);
3558 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
3559 spa_async_request(spa, SPA_ASYNC_RESILVER);
3561 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3565 * When clearing a FMA-diagnosed fault, we always want to
3566 * unspare the device, as we assume that the original spare was
3567 * done in response to the FMA fault.
3569 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3570 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3571 vd->vdev_parent->vdev_child[0] == vd)
3572 vd->vdev_unspare = B_TRUE;
3576 vdev_is_dead(vdev_t *vd)
3579 * Holes and missing devices are always considered "dead".
3580 * This simplifies the code since we don't have to check for
3581 * these types of devices in the various code paths.
3582 * Instead we rely on the fact that we skip over dead devices
3583 * before issuing I/O to them.
3585 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3586 vd->vdev_ops == &vdev_hole_ops ||
3587 vd->vdev_ops == &vdev_missing_ops);
3591 vdev_readable(vdev_t *vd)
3593 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3597 vdev_writeable(vdev_t *vd)
3599 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3600 vdev_is_concrete(vd));
3604 vdev_allocatable(vdev_t *vd)
3606 uint64_t state = vd->vdev_state;
3609 * We currently allow allocations from vdevs which may be in the
3610 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3611 * fails to reopen then we'll catch it later when we're holding
3612 * the proper locks. Note that we have to get the vdev state
3613 * in a local variable because although it changes atomically,
3614 * we're asking two separate questions about it.
3616 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3617 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3618 vd->vdev_mg->mg_initialized);
3622 vdev_accessible(vdev_t *vd, zio_t *zio)
3624 ASSERT(zio->io_vd == vd);
3626 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3629 if (zio->io_type == ZIO_TYPE_READ)
3630 return (!vd->vdev_cant_read);
3632 if (zio->io_type == ZIO_TYPE_WRITE)
3633 return (!vd->vdev_cant_write);
3639 vdev_is_spacemap_addressable(vdev_t *vd)
3641 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
3645 * If double-word space map entries are not enabled we assume
3646 * 47 bits of the space map entry are dedicated to the entry's
3647 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
3648 * to calculate the maximum address that can be described by a
3649 * space map entry for the given device.
3651 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
3653 if (shift >= 63) /* detect potential overflow */
3656 return (vd->vdev_asize < (1ULL << shift));
3660 * Get statistics for the given vdev.
3663 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3665 spa_t *spa = vd->vdev_spa;
3666 vdev_t *rvd = spa->spa_root_vdev;
3667 vdev_t *tvd = vd->vdev_top;
3669 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3671 mutex_enter(&vd->vdev_stat_lock);
3672 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3673 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3674 vs->vs_state = vd->vdev_state;
3675 vs->vs_rsize = vdev_get_min_asize(vd);
3676 if (vd->vdev_ops->vdev_op_leaf) {
3677 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3679 * Report intializing progress. Since we don't have the
3680 * initializing locks held, this is only an estimate (although a
3681 * fairly accurate one).
3683 vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done;
3684 vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est;
3685 vs->vs_initialize_state = vd->vdev_initialize_state;
3686 vs->vs_initialize_action_time = vd->vdev_initialize_action_time;
3689 * Report expandable space on top-level, non-auxillary devices only.
3690 * The expandable space is reported in terms of metaslab sized units
3691 * since that determines how much space the pool can expand.
3693 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
3694 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3695 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3697 vs->vs_configured_ashift = vd->vdev_top != NULL
3698 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
3699 vs->vs_logical_ashift = vd->vdev_logical_ashift;
3700 vs->vs_physical_ashift = vd->vdev_physical_ashift;
3701 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3702 vdev_is_concrete(vd)) {
3703 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
3707 * If we're getting stats on the root vdev, aggregate the I/O counts
3708 * over all top-level vdevs (i.e. the direct children of the root).
3711 for (int c = 0; c < rvd->vdev_children; c++) {
3712 vdev_t *cvd = rvd->vdev_child[c];
3713 vdev_stat_t *cvs = &cvd->vdev_stat;
3715 for (int t = 0; t < ZIO_TYPES; t++) {
3716 vs->vs_ops[t] += cvs->vs_ops[t];
3717 vs->vs_bytes[t] += cvs->vs_bytes[t];
3719 cvs->vs_scan_removing = cvd->vdev_removing;
3722 mutex_exit(&vd->vdev_stat_lock);
3726 vdev_clear_stats(vdev_t *vd)
3728 mutex_enter(&vd->vdev_stat_lock);
3729 vd->vdev_stat.vs_space = 0;
3730 vd->vdev_stat.vs_dspace = 0;
3731 vd->vdev_stat.vs_alloc = 0;
3732 mutex_exit(&vd->vdev_stat_lock);
3736 vdev_scan_stat_init(vdev_t *vd)
3738 vdev_stat_t *vs = &vd->vdev_stat;
3740 for (int c = 0; c < vd->vdev_children; c++)
3741 vdev_scan_stat_init(vd->vdev_child[c]);
3743 mutex_enter(&vd->vdev_stat_lock);
3744 vs->vs_scan_processed = 0;
3745 mutex_exit(&vd->vdev_stat_lock);
3749 vdev_stat_update(zio_t *zio, uint64_t psize)
3751 spa_t *spa = zio->io_spa;
3752 vdev_t *rvd = spa->spa_root_vdev;
3753 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3755 uint64_t txg = zio->io_txg;
3756 vdev_stat_t *vs = &vd->vdev_stat;
3757 zio_type_t type = zio->io_type;
3758 int flags = zio->io_flags;
3761 * If this i/o is a gang leader, it didn't do any actual work.
3763 if (zio->io_gang_tree)
3766 if (zio->io_error == 0) {
3768 * If this is a root i/o, don't count it -- we've already
3769 * counted the top-level vdevs, and vdev_get_stats() will
3770 * aggregate them when asked. This reduces contention on
3771 * the root vdev_stat_lock and implicitly handles blocks
3772 * that compress away to holes, for which there is no i/o.
3773 * (Holes never create vdev children, so all the counters
3774 * remain zero, which is what we want.)
3776 * Note: this only applies to successful i/o (io_error == 0)
3777 * because unlike i/o counts, errors are not additive.
3778 * When reading a ditto block, for example, failure of
3779 * one top-level vdev does not imply a root-level error.
3784 ASSERT(vd == zio->io_vd);
3786 if (flags & ZIO_FLAG_IO_BYPASS)
3789 mutex_enter(&vd->vdev_stat_lock);
3791 if (flags & ZIO_FLAG_IO_REPAIR) {
3792 if (flags & ZIO_FLAG_SCAN_THREAD) {
3793 dsl_scan_phys_t *scn_phys =
3794 &spa->spa_dsl_pool->dp_scan->scn_phys;
3795 uint64_t *processed = &scn_phys->scn_processed;
3798 if (vd->vdev_ops->vdev_op_leaf)
3799 atomic_add_64(processed, psize);
3800 vs->vs_scan_processed += psize;
3803 if (flags & ZIO_FLAG_SELF_HEAL)
3804 vs->vs_self_healed += psize;
3808 vs->vs_bytes[type] += psize;
3810 mutex_exit(&vd->vdev_stat_lock);
3814 if (flags & ZIO_FLAG_SPECULATIVE)
3818 * If this is an I/O error that is going to be retried, then ignore the
3819 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3820 * hard errors, when in reality they can happen for any number of
3821 * innocuous reasons (bus resets, MPxIO link failure, etc).
3823 if (zio->io_error == EIO &&
3824 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3828 * Intent logs writes won't propagate their error to the root
3829 * I/O so don't mark these types of failures as pool-level
3832 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3835 mutex_enter(&vd->vdev_stat_lock);
3836 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3837 if (zio->io_error == ECKSUM)
3838 vs->vs_checksum_errors++;
3840 vs->vs_read_errors++;
3842 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3843 vs->vs_write_errors++;
3844 mutex_exit(&vd->vdev_stat_lock);
3846 if (spa->spa_load_state == SPA_LOAD_NONE &&
3847 type == ZIO_TYPE_WRITE && txg != 0 &&
3848 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3849 (flags & ZIO_FLAG_SCAN_THREAD) ||
3850 spa->spa_claiming)) {
3852 * This is either a normal write (not a repair), or it's
3853 * a repair induced by the scrub thread, or it's a repair
3854 * made by zil_claim() during spa_load() in the first txg.
3855 * In the normal case, we commit the DTL change in the same
3856 * txg as the block was born. In the scrub-induced repair
3857 * case, we know that scrubs run in first-pass syncing context,
3858 * so we commit the DTL change in spa_syncing_txg(spa).
3859 * In the zil_claim() case, we commit in spa_first_txg(spa).
3861 * We currently do not make DTL entries for failed spontaneous
3862 * self-healing writes triggered by normal (non-scrubbing)
3863 * reads, because we have no transactional context in which to
3864 * do so -- and it's not clear that it'd be desirable anyway.
3866 if (vd->vdev_ops->vdev_op_leaf) {
3867 uint64_t commit_txg = txg;
3868 if (flags & ZIO_FLAG_SCAN_THREAD) {
3869 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3870 ASSERT(spa_sync_pass(spa) == 1);
3871 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3872 commit_txg = spa_syncing_txg(spa);
3873 } else if (spa->spa_claiming) {
3874 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3875 commit_txg = spa_first_txg(spa);
3877 ASSERT(commit_txg >= spa_syncing_txg(spa));
3878 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3880 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3881 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3882 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3885 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3890 * Update the in-core space usage stats for this vdev, its metaslab class,
3891 * and the root vdev.
3894 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3895 int64_t space_delta)
3897 int64_t dspace_delta = space_delta;
3898 spa_t *spa = vd->vdev_spa;
3899 vdev_t *rvd = spa->spa_root_vdev;
3900 metaslab_group_t *mg = vd->vdev_mg;
3901 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3903 ASSERT(vd == vd->vdev_top);
3906 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3907 * factor. We must calculate this here and not at the root vdev
3908 * because the root vdev's psize-to-asize is simply the max of its
3909 * childrens', thus not accurate enough for us.
3911 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3912 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3913 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3914 vd->vdev_deflate_ratio;
3916 mutex_enter(&vd->vdev_stat_lock);
3917 vd->vdev_stat.vs_alloc += alloc_delta;
3918 vd->vdev_stat.vs_space += space_delta;
3919 vd->vdev_stat.vs_dspace += dspace_delta;
3920 mutex_exit(&vd->vdev_stat_lock);
3922 if (mc == spa_normal_class(spa)) {
3923 mutex_enter(&rvd->vdev_stat_lock);
3924 rvd->vdev_stat.vs_alloc += alloc_delta;
3925 rvd->vdev_stat.vs_space += space_delta;
3926 rvd->vdev_stat.vs_dspace += dspace_delta;
3927 mutex_exit(&rvd->vdev_stat_lock);
3931 ASSERT(rvd == vd->vdev_parent);
3932 ASSERT(vd->vdev_ms_count != 0);
3934 metaslab_class_space_update(mc,
3935 alloc_delta, defer_delta, space_delta, dspace_delta);
3940 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3941 * so that it will be written out next time the vdev configuration is synced.
3942 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3945 vdev_config_dirty(vdev_t *vd)
3947 spa_t *spa = vd->vdev_spa;
3948 vdev_t *rvd = spa->spa_root_vdev;
3951 ASSERT(spa_writeable(spa));
3954 * If this is an aux vdev (as with l2cache and spare devices), then we
3955 * update the vdev config manually and set the sync flag.
3957 if (vd->vdev_aux != NULL) {
3958 spa_aux_vdev_t *sav = vd->vdev_aux;
3962 for (c = 0; c < sav->sav_count; c++) {
3963 if (sav->sav_vdevs[c] == vd)
3967 if (c == sav->sav_count) {
3969 * We're being removed. There's nothing more to do.
3971 ASSERT(sav->sav_sync == B_TRUE);
3975 sav->sav_sync = B_TRUE;
3977 if (nvlist_lookup_nvlist_array(sav->sav_config,
3978 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3979 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3980 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3986 * Setting the nvlist in the middle if the array is a little
3987 * sketchy, but it will work.
3989 nvlist_free(aux[c]);
3990 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3996 * The dirty list is protected by the SCL_CONFIG lock. The caller
3997 * must either hold SCL_CONFIG as writer, or must be the sync thread
3998 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3999 * so this is sufficient to ensure mutual exclusion.
4001 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4002 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4003 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4006 for (c = 0; c < rvd->vdev_children; c++)
4007 vdev_config_dirty(rvd->vdev_child[c]);
4009 ASSERT(vd == vd->vdev_top);
4011 if (!list_link_active(&vd->vdev_config_dirty_node) &&
4012 vdev_is_concrete(vd)) {
4013 list_insert_head(&spa->spa_config_dirty_list, vd);
4019 vdev_config_clean(vdev_t *vd)
4021 spa_t *spa = vd->vdev_spa;
4023 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4024 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4025 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4027 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
4028 list_remove(&spa->spa_config_dirty_list, vd);
4032 * Mark a top-level vdev's state as dirty, so that the next pass of
4033 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4034 * the state changes from larger config changes because they require
4035 * much less locking, and are often needed for administrative actions.
4038 vdev_state_dirty(vdev_t *vd)
4040 spa_t *spa = vd->vdev_spa;
4042 ASSERT(spa_writeable(spa));
4043 ASSERT(vd == vd->vdev_top);
4046 * The state list is protected by the SCL_STATE lock. The caller
4047 * must either hold SCL_STATE as writer, or must be the sync thread
4048 * (which holds SCL_STATE as reader). There's only one sync thread,
4049 * so this is sufficient to ensure mutual exclusion.
4051 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4052 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4053 spa_config_held(spa, SCL_STATE, RW_READER)));
4055 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4056 vdev_is_concrete(vd))
4057 list_insert_head(&spa->spa_state_dirty_list, vd);
4061 vdev_state_clean(vdev_t *vd)
4063 spa_t *spa = vd->vdev_spa;
4065 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4066 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4067 spa_config_held(spa, SCL_STATE, RW_READER)));
4069 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4070 list_remove(&spa->spa_state_dirty_list, vd);
4074 * Propagate vdev state up from children to parent.
4077 vdev_propagate_state(vdev_t *vd)
4079 spa_t *spa = vd->vdev_spa;
4080 vdev_t *rvd = spa->spa_root_vdev;
4081 int degraded = 0, faulted = 0;
4085 if (vd->vdev_children > 0) {
4086 for (int c = 0; c < vd->vdev_children; c++) {
4087 child = vd->vdev_child[c];
4090 * Don't factor holes or indirect vdevs into the
4093 if (!vdev_is_concrete(child))
4096 if (!vdev_readable(child) ||
4097 (!vdev_writeable(child) && spa_writeable(spa))) {
4099 * Root special: if there is a top-level log
4100 * device, treat the root vdev as if it were
4103 if (child->vdev_islog && vd == rvd)
4107 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4111 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4115 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4118 * Root special: if there is a top-level vdev that cannot be
4119 * opened due to corrupted metadata, then propagate the root
4120 * vdev's aux state as 'corrupt' rather than 'insufficient
4123 if (corrupted && vd == rvd &&
4124 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4125 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4126 VDEV_AUX_CORRUPT_DATA);
4129 if (vd->vdev_parent)
4130 vdev_propagate_state(vd->vdev_parent);
4134 * Set a vdev's state. If this is during an open, we don't update the parent
4135 * state, because we're in the process of opening children depth-first.
4136 * Otherwise, we propagate the change to the parent.
4138 * If this routine places a device in a faulted state, an appropriate ereport is
4142 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4144 uint64_t save_state;
4145 spa_t *spa = vd->vdev_spa;
4147 if (state == vd->vdev_state) {
4148 vd->vdev_stat.vs_aux = aux;
4152 save_state = vd->vdev_state;
4154 vd->vdev_state = state;
4155 vd->vdev_stat.vs_aux = aux;
4158 * If we are setting the vdev state to anything but an open state, then
4159 * always close the underlying device unless the device has requested
4160 * a delayed close (i.e. we're about to remove or fault the device).
4161 * Otherwise, we keep accessible but invalid devices open forever.
4162 * We don't call vdev_close() itself, because that implies some extra
4163 * checks (offline, etc) that we don't want here. This is limited to
4164 * leaf devices, because otherwise closing the device will affect other
4167 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4168 vd->vdev_ops->vdev_op_leaf)
4169 vd->vdev_ops->vdev_op_close(vd);
4171 if (vd->vdev_removed &&
4172 state == VDEV_STATE_CANT_OPEN &&
4173 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4175 * If the previous state is set to VDEV_STATE_REMOVED, then this
4176 * device was previously marked removed and someone attempted to
4177 * reopen it. If this failed due to a nonexistent device, then
4178 * keep the device in the REMOVED state. We also let this be if
4179 * it is one of our special test online cases, which is only
4180 * attempting to online the device and shouldn't generate an FMA
4183 vd->vdev_state = VDEV_STATE_REMOVED;
4184 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4185 } else if (state == VDEV_STATE_REMOVED) {
4186 vd->vdev_removed = B_TRUE;
4187 } else if (state == VDEV_STATE_CANT_OPEN) {
4189 * If we fail to open a vdev during an import or recovery, we
4190 * mark it as "not available", which signifies that it was
4191 * never there to begin with. Failure to open such a device
4192 * is not considered an error.
4194 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4195 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4196 vd->vdev_ops->vdev_op_leaf)
4197 vd->vdev_not_present = 1;
4200 * Post the appropriate ereport. If the 'prevstate' field is
4201 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4202 * that this is part of a vdev_reopen(). In this case, we don't
4203 * want to post the ereport if the device was already in the
4204 * CANT_OPEN state beforehand.
4206 * If the 'checkremove' flag is set, then this is an attempt to
4207 * online the device in response to an insertion event. If we
4208 * hit this case, then we have detected an insertion event for a
4209 * faulted or offline device that wasn't in the removed state.
4210 * In this scenario, we don't post an ereport because we are
4211 * about to replace the device, or attempt an online with
4212 * vdev_forcefault, which will generate the fault for us.
4214 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4215 !vd->vdev_not_present && !vd->vdev_checkremove &&
4216 vd != spa->spa_root_vdev) {
4220 case VDEV_AUX_OPEN_FAILED:
4221 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4223 case VDEV_AUX_CORRUPT_DATA:
4224 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4226 case VDEV_AUX_NO_REPLICAS:
4227 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4229 case VDEV_AUX_BAD_GUID_SUM:
4230 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4232 case VDEV_AUX_TOO_SMALL:
4233 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4235 case VDEV_AUX_BAD_LABEL:
4236 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4239 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4242 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
4245 /* Erase any notion of persistent removed state */
4246 vd->vdev_removed = B_FALSE;
4248 vd->vdev_removed = B_FALSE;
4252 * Notify the fmd of the state change. Be verbose and post
4253 * notifications even for stuff that's not important; the fmd agent can
4254 * sort it out. Don't emit state change events for non-leaf vdevs since
4255 * they can't change state on their own. The FMD can check their state
4256 * if it wants to when it sees that a leaf vdev had a state change.
4258 if (vd->vdev_ops->vdev_op_leaf)
4259 zfs_post_state_change(spa, vd);
4261 if (!isopen && vd->vdev_parent)
4262 vdev_propagate_state(vd->vdev_parent);
4266 vdev_children_are_offline(vdev_t *vd)
4268 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4270 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4271 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4279 * Check the vdev configuration to ensure that it's capable of supporting
4280 * a root pool. We do not support partial configuration.
4281 * In addition, only a single top-level vdev is allowed.
4283 * FreeBSD does not have above limitations.
4286 vdev_is_bootable(vdev_t *vd)
4289 if (!vd->vdev_ops->vdev_op_leaf) {
4290 char *vdev_type = vd->vdev_ops->vdev_op_type;
4292 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
4293 vd->vdev_children > 1) {
4295 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4296 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4301 for (int c = 0; c < vd->vdev_children; c++) {
4302 if (!vdev_is_bootable(vd->vdev_child[c]))
4305 #endif /* illumos */
4310 vdev_is_concrete(vdev_t *vd)
4312 vdev_ops_t *ops = vd->vdev_ops;
4313 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4314 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4322 * Determine if a log device has valid content. If the vdev was
4323 * removed or faulted in the MOS config then we know that
4324 * the content on the log device has already been written to the pool.
4327 vdev_log_state_valid(vdev_t *vd)
4329 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4333 for (int c = 0; c < vd->vdev_children; c++)
4334 if (vdev_log_state_valid(vd->vdev_child[c]))
4341 * Expand a vdev if possible.
4344 vdev_expand(vdev_t *vd, uint64_t txg)
4346 ASSERT(vd->vdev_top == vd);
4347 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4348 ASSERT(vdev_is_concrete(vd));
4350 vdev_set_deflate_ratio(vd);
4352 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
4353 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4354 vdev_config_dirty(vd);
4362 vdev_split(vdev_t *vd)
4364 vdev_t *cvd, *pvd = vd->vdev_parent;
4366 vdev_remove_child(pvd, vd);
4367 vdev_compact_children(pvd);
4369 cvd = pvd->vdev_child[0];
4370 if (pvd->vdev_children == 1) {
4371 vdev_remove_parent(cvd);
4372 cvd->vdev_splitting = B_TRUE;
4374 vdev_propagate_state(cvd);
4378 vdev_deadman(vdev_t *vd)
4380 for (int c = 0; c < vd->vdev_children; c++) {
4381 vdev_t *cvd = vd->vdev_child[c];
4386 if (vd->vdev_ops->vdev_op_leaf) {
4387 vdev_queue_t *vq = &vd->vdev_queue;
4389 mutex_enter(&vq->vq_lock);
4390 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4391 spa_t *spa = vd->vdev_spa;
4396 * Look at the head of all the pending queues,
4397 * if any I/O has been outstanding for longer than
4398 * the spa_deadman_synctime we panic the system.
4400 fio = avl_first(&vq->vq_active_tree);
4401 delta = gethrtime() - fio->io_timestamp;
4402 if (delta > spa_deadman_synctime(spa)) {
4403 vdev_dbgmsg(vd, "SLOW IO: zio timestamp "
4404 "%lluns, delta %lluns, last io %lluns",
4405 fio->io_timestamp, (u_longlong_t)delta,
4406 vq->vq_io_complete_ts);
4407 fm_panic("I/O to pool '%s' appears to be "
4408 "hung on vdev guid %llu at '%s'.",
4410 (long long unsigned int) vd->vdev_guid,
4414 mutex_exit(&vq->vq_lock);