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;
481 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
484 uint_t id = cvd->vdev_id;
486 ASSERT(cvd->vdev_parent == pvd);
491 ASSERT(id < pvd->vdev_children);
492 ASSERT(pvd->vdev_child[id] == cvd);
494 pvd->vdev_child[id] = NULL;
495 cvd->vdev_parent = NULL;
497 for (c = 0; c < pvd->vdev_children; c++)
498 if (pvd->vdev_child[c])
501 if (c == pvd->vdev_children) {
502 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
503 pvd->vdev_child = NULL;
504 pvd->vdev_children = 0;
508 * Walk up all ancestors to update guid sum.
510 for (; pvd != NULL; pvd = pvd->vdev_parent)
511 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
515 * Remove any holes in the child array.
518 vdev_compact_children(vdev_t *pvd)
520 vdev_t **newchild, *cvd;
521 int oldc = pvd->vdev_children;
524 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
529 for (int c = newc = 0; c < oldc; c++)
530 if (pvd->vdev_child[c])
534 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
536 for (int c = newc = 0; c < oldc; c++) {
537 if ((cvd = pvd->vdev_child[c]) != NULL) {
538 newchild[newc] = cvd;
539 cvd->vdev_id = newc++;
546 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
547 pvd->vdev_child = newchild;
548 pvd->vdev_children = newc;
552 * Allocate and minimally initialize a vdev_t.
555 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
558 vdev_indirect_config_t *vic;
560 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
561 vic = &vd->vdev_indirect_config;
563 if (spa->spa_root_vdev == NULL) {
564 ASSERT(ops == &vdev_root_ops);
565 spa->spa_root_vdev = vd;
566 spa->spa_load_guid = spa_generate_guid(NULL);
569 if (guid == 0 && ops != &vdev_hole_ops) {
570 if (spa->spa_root_vdev == vd) {
572 * The root vdev's guid will also be the pool guid,
573 * which must be unique among all pools.
575 guid = spa_generate_guid(NULL);
578 * Any other vdev's guid must be unique within the pool.
580 guid = spa_generate_guid(spa);
582 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
587 vd->vdev_guid = guid;
588 vd->vdev_guid_sum = guid;
590 vd->vdev_state = VDEV_STATE_CLOSED;
591 vd->vdev_ishole = (ops == &vdev_hole_ops);
592 vic->vic_prev_indirect_vdev = UINT64_MAX;
594 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
595 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
596 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
598 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
599 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
600 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
601 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
602 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
603 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
604 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
605 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
606 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
608 for (int t = 0; t < DTL_TYPES; t++) {
609 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
611 txg_list_create(&vd->vdev_ms_list, spa,
612 offsetof(struct metaslab, ms_txg_node));
613 txg_list_create(&vd->vdev_dtl_list, spa,
614 offsetof(struct vdev, vdev_dtl_node));
615 vd->vdev_stat.vs_timestamp = gethrtime();
623 * Allocate a new vdev. The 'alloctype' is used to control whether we are
624 * creating a new vdev or loading an existing one - the behavior is slightly
625 * different for each case.
628 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
633 uint64_t guid = 0, islog, nparity;
635 vdev_indirect_config_t *vic;
637 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
639 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
640 return (SET_ERROR(EINVAL));
642 if ((ops = vdev_getops(type)) == NULL)
643 return (SET_ERROR(EINVAL));
646 * If this is a load, get the vdev guid from the nvlist.
647 * Otherwise, vdev_alloc_common() will generate one for us.
649 if (alloctype == VDEV_ALLOC_LOAD) {
652 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
654 return (SET_ERROR(EINVAL));
656 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
657 return (SET_ERROR(EINVAL));
658 } else if (alloctype == VDEV_ALLOC_SPARE) {
659 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
660 return (SET_ERROR(EINVAL));
661 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
662 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
663 return (SET_ERROR(EINVAL));
664 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
665 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
666 return (SET_ERROR(EINVAL));
670 * The first allocated vdev must be of type 'root'.
672 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
673 return (SET_ERROR(EINVAL));
676 * Determine whether we're a log vdev.
679 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
680 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
681 return (SET_ERROR(ENOTSUP));
683 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
684 return (SET_ERROR(ENOTSUP));
687 * Set the nparity property for RAID-Z vdevs.
690 if (ops == &vdev_raidz_ops) {
691 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
693 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
694 return (SET_ERROR(EINVAL));
696 * Previous versions could only support 1 or 2 parity
700 spa_version(spa) < SPA_VERSION_RAIDZ2)
701 return (SET_ERROR(ENOTSUP));
703 spa_version(spa) < SPA_VERSION_RAIDZ3)
704 return (SET_ERROR(ENOTSUP));
707 * We require the parity to be specified for SPAs that
708 * support multiple parity levels.
710 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
711 return (SET_ERROR(EINVAL));
713 * Otherwise, we default to 1 parity device for RAID-Z.
720 ASSERT(nparity != -1ULL);
722 vd = vdev_alloc_common(spa, id, guid, ops);
723 vic = &vd->vdev_indirect_config;
725 vd->vdev_islog = islog;
726 vd->vdev_nparity = nparity;
728 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
729 vd->vdev_path = spa_strdup(vd->vdev_path);
730 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
731 vd->vdev_devid = spa_strdup(vd->vdev_devid);
732 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
733 &vd->vdev_physpath) == 0)
734 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
735 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
736 vd->vdev_fru = spa_strdup(vd->vdev_fru);
739 * Set the whole_disk property. If it's not specified, leave the value
742 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
743 &vd->vdev_wholedisk) != 0)
744 vd->vdev_wholedisk = -1ULL;
746 ASSERT0(vic->vic_mapping_object);
747 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
748 &vic->vic_mapping_object);
749 ASSERT0(vic->vic_births_object);
750 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
751 &vic->vic_births_object);
752 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
753 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
754 &vic->vic_prev_indirect_vdev);
757 * Look for the 'not present' flag. This will only be set if the device
758 * was not present at the time of import.
760 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
761 &vd->vdev_not_present);
764 * Get the alignment requirement.
766 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
769 * Retrieve the vdev creation time.
771 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
775 * If we're a top-level vdev, try to load the allocation parameters.
777 if (parent && !parent->vdev_parent &&
778 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
779 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
781 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
783 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
785 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
787 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
790 ASSERT0(vd->vdev_top_zap);
793 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
794 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
795 alloctype == VDEV_ALLOC_ADD ||
796 alloctype == VDEV_ALLOC_SPLIT ||
797 alloctype == VDEV_ALLOC_ROOTPOOL);
798 vd->vdev_mg = metaslab_group_create(islog ?
799 spa_log_class(spa) : spa_normal_class(spa), vd,
800 spa->spa_alloc_count);
803 if (vd->vdev_ops->vdev_op_leaf &&
804 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
805 (void) nvlist_lookup_uint64(nv,
806 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
808 ASSERT0(vd->vdev_leaf_zap);
812 * If we're a leaf vdev, try to load the DTL object and other state.
815 if (vd->vdev_ops->vdev_op_leaf &&
816 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
817 alloctype == VDEV_ALLOC_ROOTPOOL)) {
818 if (alloctype == VDEV_ALLOC_LOAD) {
819 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
820 &vd->vdev_dtl_object);
821 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
825 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
828 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
829 &spare) == 0 && spare)
833 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
836 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
837 &vd->vdev_resilver_txg);
840 * When importing a pool, we want to ignore the persistent fault
841 * state, as the diagnosis made on another system may not be
842 * valid in the current context. Local vdevs will
843 * remain in the faulted state.
845 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
846 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
848 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
850 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
853 if (vd->vdev_faulted || vd->vdev_degraded) {
857 VDEV_AUX_ERR_EXCEEDED;
858 if (nvlist_lookup_string(nv,
859 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
860 strcmp(aux, "external") == 0)
861 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
867 * Add ourselves to the parent's list of children.
869 vdev_add_child(parent, vd);
877 vdev_free(vdev_t *vd)
879 spa_t *spa = vd->vdev_spa;
880 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
883 * Scan queues are normally destroyed at the end of a scan. If the
884 * queue exists here, that implies the vdev is being removed while
885 * the scan is still running.
887 if (vd->vdev_scan_io_queue != NULL) {
888 mutex_enter(&vd->vdev_scan_io_queue_lock);
889 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
890 vd->vdev_scan_io_queue = NULL;
891 mutex_exit(&vd->vdev_scan_io_queue_lock);
895 * vdev_free() implies closing the vdev first. This is simpler than
896 * trying to ensure complicated semantics for all callers.
900 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
901 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
906 for (int c = 0; c < vd->vdev_children; c++)
907 vdev_free(vd->vdev_child[c]);
909 ASSERT(vd->vdev_child == NULL);
910 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
911 ASSERT(vd->vdev_initialize_thread == NULL);
914 * Discard allocation state.
916 if (vd->vdev_mg != NULL) {
917 vdev_metaslab_fini(vd);
918 metaslab_group_destroy(vd->vdev_mg);
921 ASSERT0(vd->vdev_stat.vs_space);
922 ASSERT0(vd->vdev_stat.vs_dspace);
923 ASSERT0(vd->vdev_stat.vs_alloc);
926 * Remove this vdev from its parent's child list.
928 vdev_remove_child(vd->vdev_parent, vd);
930 ASSERT(vd->vdev_parent == NULL);
933 * Clean up vdev structure.
939 spa_strfree(vd->vdev_path);
941 spa_strfree(vd->vdev_devid);
942 if (vd->vdev_physpath)
943 spa_strfree(vd->vdev_physpath);
945 spa_strfree(vd->vdev_fru);
947 if (vd->vdev_isspare)
948 spa_spare_remove(vd);
949 if (vd->vdev_isl2cache)
950 spa_l2cache_remove(vd);
952 txg_list_destroy(&vd->vdev_ms_list);
953 txg_list_destroy(&vd->vdev_dtl_list);
955 mutex_enter(&vd->vdev_dtl_lock);
956 space_map_close(vd->vdev_dtl_sm);
957 for (int t = 0; t < DTL_TYPES; t++) {
958 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
959 range_tree_destroy(vd->vdev_dtl[t]);
961 mutex_exit(&vd->vdev_dtl_lock);
963 EQUIV(vd->vdev_indirect_births != NULL,
964 vd->vdev_indirect_mapping != NULL);
965 if (vd->vdev_indirect_births != NULL) {
966 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
967 vdev_indirect_births_close(vd->vdev_indirect_births);
970 if (vd->vdev_obsolete_sm != NULL) {
971 ASSERT(vd->vdev_removing ||
972 vd->vdev_ops == &vdev_indirect_ops);
973 space_map_close(vd->vdev_obsolete_sm);
974 vd->vdev_obsolete_sm = NULL;
976 range_tree_destroy(vd->vdev_obsolete_segments);
977 rw_destroy(&vd->vdev_indirect_rwlock);
978 mutex_destroy(&vd->vdev_obsolete_lock);
980 mutex_destroy(&vd->vdev_queue_lock);
981 mutex_destroy(&vd->vdev_dtl_lock);
982 mutex_destroy(&vd->vdev_stat_lock);
983 mutex_destroy(&vd->vdev_probe_lock);
984 mutex_destroy(&vd->vdev_scan_io_queue_lock);
985 mutex_destroy(&vd->vdev_initialize_lock);
986 mutex_destroy(&vd->vdev_initialize_io_lock);
987 cv_destroy(&vd->vdev_initialize_io_cv);
988 cv_destroy(&vd->vdev_initialize_cv);
990 if (vd == spa->spa_root_vdev)
991 spa->spa_root_vdev = NULL;
993 kmem_free(vd, sizeof (vdev_t));
997 * Transfer top-level vdev state from svd to tvd.
1000 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
1002 spa_t *spa = svd->vdev_spa;
1007 ASSERT(tvd == tvd->vdev_top);
1009 tvd->vdev_ms_array = svd->vdev_ms_array;
1010 tvd->vdev_ms_shift = svd->vdev_ms_shift;
1011 tvd->vdev_ms_count = svd->vdev_ms_count;
1012 tvd->vdev_top_zap = svd->vdev_top_zap;
1014 svd->vdev_ms_array = 0;
1015 svd->vdev_ms_shift = 0;
1016 svd->vdev_ms_count = 0;
1017 svd->vdev_top_zap = 0;
1020 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1021 tvd->vdev_mg = svd->vdev_mg;
1022 tvd->vdev_ms = svd->vdev_ms;
1024 svd->vdev_mg = NULL;
1025 svd->vdev_ms = NULL;
1027 if (tvd->vdev_mg != NULL)
1028 tvd->vdev_mg->mg_vd = tvd;
1030 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1031 svd->vdev_checkpoint_sm = NULL;
1033 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1034 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1035 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1037 svd->vdev_stat.vs_alloc = 0;
1038 svd->vdev_stat.vs_space = 0;
1039 svd->vdev_stat.vs_dspace = 0;
1042 * State which may be set on a top-level vdev that's in the
1043 * process of being removed.
1045 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1046 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1047 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1048 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1049 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1050 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1051 ASSERT0(tvd->vdev_removing);
1052 tvd->vdev_removing = svd->vdev_removing;
1053 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1054 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1055 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1056 range_tree_swap(&svd->vdev_obsolete_segments,
1057 &tvd->vdev_obsolete_segments);
1058 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1059 svd->vdev_indirect_config.vic_mapping_object = 0;
1060 svd->vdev_indirect_config.vic_births_object = 0;
1061 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1062 svd->vdev_indirect_mapping = NULL;
1063 svd->vdev_indirect_births = NULL;
1064 svd->vdev_obsolete_sm = NULL;
1065 svd->vdev_removing = 0;
1067 for (t = 0; t < TXG_SIZE; t++) {
1068 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1069 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1070 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1071 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1072 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1073 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1076 if (list_link_active(&svd->vdev_config_dirty_node)) {
1077 vdev_config_clean(svd);
1078 vdev_config_dirty(tvd);
1081 if (list_link_active(&svd->vdev_state_dirty_node)) {
1082 vdev_state_clean(svd);
1083 vdev_state_dirty(tvd);
1086 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1087 svd->vdev_deflate_ratio = 0;
1089 tvd->vdev_islog = svd->vdev_islog;
1090 svd->vdev_islog = 0;
1092 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1096 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1103 for (int c = 0; c < vd->vdev_children; c++)
1104 vdev_top_update(tvd, vd->vdev_child[c]);
1108 * Add a mirror/replacing vdev above an existing vdev.
1111 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1113 spa_t *spa = cvd->vdev_spa;
1114 vdev_t *pvd = cvd->vdev_parent;
1117 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1119 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1121 mvd->vdev_asize = cvd->vdev_asize;
1122 mvd->vdev_min_asize = cvd->vdev_min_asize;
1123 mvd->vdev_max_asize = cvd->vdev_max_asize;
1124 mvd->vdev_psize = cvd->vdev_psize;
1125 mvd->vdev_ashift = cvd->vdev_ashift;
1126 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1127 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1128 mvd->vdev_state = cvd->vdev_state;
1129 mvd->vdev_crtxg = cvd->vdev_crtxg;
1131 vdev_remove_child(pvd, cvd);
1132 vdev_add_child(pvd, mvd);
1133 cvd->vdev_id = mvd->vdev_children;
1134 vdev_add_child(mvd, cvd);
1135 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1137 if (mvd == mvd->vdev_top)
1138 vdev_top_transfer(cvd, mvd);
1144 * Remove a 1-way mirror/replacing vdev from the tree.
1147 vdev_remove_parent(vdev_t *cvd)
1149 vdev_t *mvd = cvd->vdev_parent;
1150 vdev_t *pvd = mvd->vdev_parent;
1152 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1154 ASSERT(mvd->vdev_children == 1);
1155 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1156 mvd->vdev_ops == &vdev_replacing_ops ||
1157 mvd->vdev_ops == &vdev_spare_ops);
1158 cvd->vdev_ashift = mvd->vdev_ashift;
1159 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1160 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1162 vdev_remove_child(mvd, cvd);
1163 vdev_remove_child(pvd, mvd);
1166 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1167 * Otherwise, we could have detached an offline device, and when we
1168 * go to import the pool we'll think we have two top-level vdevs,
1169 * instead of a different version of the same top-level vdev.
1171 if (mvd->vdev_top == mvd) {
1172 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1173 cvd->vdev_orig_guid = cvd->vdev_guid;
1174 cvd->vdev_guid += guid_delta;
1175 cvd->vdev_guid_sum += guid_delta;
1177 cvd->vdev_id = mvd->vdev_id;
1178 vdev_add_child(pvd, cvd);
1179 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1181 if (cvd == cvd->vdev_top)
1182 vdev_top_transfer(mvd, cvd);
1184 ASSERT(mvd->vdev_children == 0);
1189 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1191 spa_t *spa = vd->vdev_spa;
1192 objset_t *mos = spa->spa_meta_objset;
1194 uint64_t oldc = vd->vdev_ms_count;
1195 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1199 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1202 * This vdev is not being allocated from yet or is a hole.
1204 if (vd->vdev_ms_shift == 0)
1207 ASSERT(!vd->vdev_ishole);
1209 ASSERT(oldc <= newc);
1211 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1214 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1215 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1219 vd->vdev_ms_count = newc;
1220 for (m = oldc; m < newc; m++) {
1221 uint64_t object = 0;
1224 * vdev_ms_array may be 0 if we are creating the "fake"
1225 * metaslabs for an indirect vdev for zdb's leak detection.
1226 * See zdb_leak_init().
1228 if (txg == 0 && vd->vdev_ms_array != 0) {
1229 error = dmu_read(mos, vd->vdev_ms_array,
1230 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1233 vdev_dbgmsg(vd, "unable to read the metaslab "
1234 "array [error=%d]", error);
1239 error = metaslab_init(vd->vdev_mg, m, object, txg,
1242 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1249 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1252 * If the vdev is being removed we don't activate
1253 * the metaslabs since we want to ensure that no new
1254 * allocations are performed on this device.
1256 if (oldc == 0 && !vd->vdev_removing)
1257 metaslab_group_activate(vd->vdev_mg);
1260 spa_config_exit(spa, SCL_ALLOC, FTAG);
1266 vdev_metaslab_fini(vdev_t *vd)
1268 if (vd->vdev_checkpoint_sm != NULL) {
1269 ASSERT(spa_feature_is_active(vd->vdev_spa,
1270 SPA_FEATURE_POOL_CHECKPOINT));
1271 space_map_close(vd->vdev_checkpoint_sm);
1273 * Even though we close the space map, we need to set its
1274 * pointer to NULL. The reason is that vdev_metaslab_fini()
1275 * may be called multiple times for certain operations
1276 * (i.e. when destroying a pool) so we need to ensure that
1277 * this clause never executes twice. This logic is similar
1278 * to the one used for the vdev_ms clause below.
1280 vd->vdev_checkpoint_sm = NULL;
1283 if (vd->vdev_ms != NULL) {
1284 uint64_t count = vd->vdev_ms_count;
1286 metaslab_group_passivate(vd->vdev_mg);
1287 for (uint64_t m = 0; m < count; m++) {
1288 metaslab_t *msp = vd->vdev_ms[m];
1293 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1296 vd->vdev_ms_count = 0;
1298 ASSERT0(vd->vdev_ms_count);
1301 typedef struct vdev_probe_stats {
1302 boolean_t vps_readable;
1303 boolean_t vps_writeable;
1305 } vdev_probe_stats_t;
1308 vdev_probe_done(zio_t *zio)
1310 spa_t *spa = zio->io_spa;
1311 vdev_t *vd = zio->io_vd;
1312 vdev_probe_stats_t *vps = zio->io_private;
1314 ASSERT(vd->vdev_probe_zio != NULL);
1316 if (zio->io_type == ZIO_TYPE_READ) {
1317 if (zio->io_error == 0)
1318 vps->vps_readable = 1;
1319 if (zio->io_error == 0 && spa_writeable(spa)) {
1320 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1321 zio->io_offset, zio->io_size, zio->io_abd,
1322 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1323 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1325 abd_free(zio->io_abd);
1327 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1328 if (zio->io_error == 0)
1329 vps->vps_writeable = 1;
1330 abd_free(zio->io_abd);
1331 } else if (zio->io_type == ZIO_TYPE_NULL) {
1334 vd->vdev_cant_read |= !vps->vps_readable;
1335 vd->vdev_cant_write |= !vps->vps_writeable;
1337 if (vdev_readable(vd) &&
1338 (vdev_writeable(vd) || !spa_writeable(spa))) {
1341 ASSERT(zio->io_error != 0);
1342 vdev_dbgmsg(vd, "failed probe");
1343 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1344 spa, vd, NULL, 0, 0);
1345 zio->io_error = SET_ERROR(ENXIO);
1348 mutex_enter(&vd->vdev_probe_lock);
1349 ASSERT(vd->vdev_probe_zio == zio);
1350 vd->vdev_probe_zio = NULL;
1351 mutex_exit(&vd->vdev_probe_lock);
1353 zio_link_t *zl = NULL;
1354 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1355 if (!vdev_accessible(vd, pio))
1356 pio->io_error = SET_ERROR(ENXIO);
1358 kmem_free(vps, sizeof (*vps));
1363 * Determine whether this device is accessible.
1365 * Read and write to several known locations: the pad regions of each
1366 * vdev label but the first, which we leave alone in case it contains
1370 vdev_probe(vdev_t *vd, zio_t *zio)
1372 spa_t *spa = vd->vdev_spa;
1373 vdev_probe_stats_t *vps = NULL;
1376 ASSERT(vd->vdev_ops->vdev_op_leaf);
1379 * Don't probe the probe.
1381 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1385 * To prevent 'probe storms' when a device fails, we create
1386 * just one probe i/o at a time. All zios that want to probe
1387 * this vdev will become parents of the probe io.
1389 mutex_enter(&vd->vdev_probe_lock);
1391 if ((pio = vd->vdev_probe_zio) == NULL) {
1392 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1394 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1395 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1398 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1400 * vdev_cant_read and vdev_cant_write can only
1401 * transition from TRUE to FALSE when we have the
1402 * SCL_ZIO lock as writer; otherwise they can only
1403 * transition from FALSE to TRUE. This ensures that
1404 * any zio looking at these values can assume that
1405 * failures persist for the life of the I/O. That's
1406 * important because when a device has intermittent
1407 * connectivity problems, we want to ensure that
1408 * they're ascribed to the device (ENXIO) and not
1411 * Since we hold SCL_ZIO as writer here, clear both
1412 * values so the probe can reevaluate from first
1415 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1416 vd->vdev_cant_read = B_FALSE;
1417 vd->vdev_cant_write = B_FALSE;
1420 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1421 vdev_probe_done, vps,
1422 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1425 * We can't change the vdev state in this context, so we
1426 * kick off an async task to do it on our behalf.
1429 vd->vdev_probe_wanted = B_TRUE;
1430 spa_async_request(spa, SPA_ASYNC_PROBE);
1435 zio_add_child(zio, pio);
1437 mutex_exit(&vd->vdev_probe_lock);
1440 ASSERT(zio != NULL);
1444 for (int l = 1; l < VDEV_LABELS; l++) {
1445 zio_nowait(zio_read_phys(pio, vd,
1446 vdev_label_offset(vd->vdev_psize, l,
1447 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1448 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1449 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1450 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1461 vdev_open_child(void *arg)
1465 vd->vdev_open_thread = curthread;
1466 vd->vdev_open_error = vdev_open(vd);
1467 vd->vdev_open_thread = NULL;
1471 vdev_uses_zvols(vdev_t *vd)
1473 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1474 strlen(ZVOL_DIR)) == 0)
1476 for (int c = 0; c < vd->vdev_children; c++)
1477 if (vdev_uses_zvols(vd->vdev_child[c]))
1483 vdev_open_children(vdev_t *vd)
1486 int children = vd->vdev_children;
1488 vd->vdev_nonrot = B_TRUE;
1491 * in order to handle pools on top of zvols, do the opens
1492 * in a single thread so that the same thread holds the
1493 * spa_namespace_lock
1495 if (B_TRUE || vdev_uses_zvols(vd)) {
1496 for (int c = 0; c < children; c++) {
1497 vd->vdev_child[c]->vdev_open_error =
1498 vdev_open(vd->vdev_child[c]);
1499 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1503 tq = taskq_create("vdev_open", children, minclsyspri,
1504 children, children, TASKQ_PREPOPULATE);
1506 for (int c = 0; c < children; c++)
1507 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1512 for (int c = 0; c < children; c++)
1513 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1517 * Compute the raidz-deflation ratio. Note, we hard-code
1518 * in 128k (1 << 17) because it is the "typical" blocksize.
1519 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1520 * otherwise it would inconsistently account for existing bp's.
1523 vdev_set_deflate_ratio(vdev_t *vd)
1525 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1526 vd->vdev_deflate_ratio = (1 << 17) /
1527 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1532 * Prepare a virtual device for access.
1535 vdev_open(vdev_t *vd)
1537 spa_t *spa = vd->vdev_spa;
1540 uint64_t max_osize = 0;
1541 uint64_t asize, max_asize, psize;
1542 uint64_t logical_ashift = 0;
1543 uint64_t physical_ashift = 0;
1545 ASSERT(vd->vdev_open_thread == curthread ||
1546 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1547 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1548 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1549 vd->vdev_state == VDEV_STATE_OFFLINE);
1551 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1552 vd->vdev_cant_read = B_FALSE;
1553 vd->vdev_cant_write = B_FALSE;
1554 vd->vdev_notrim = B_FALSE;
1555 vd->vdev_min_asize = vdev_get_min_asize(vd);
1558 * If this vdev is not removed, check its fault status. If it's
1559 * faulted, bail out of the open.
1561 if (!vd->vdev_removed && vd->vdev_faulted) {
1562 ASSERT(vd->vdev_children == 0);
1563 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1564 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1565 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1566 vd->vdev_label_aux);
1567 return (SET_ERROR(ENXIO));
1568 } else if (vd->vdev_offline) {
1569 ASSERT(vd->vdev_children == 0);
1570 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1571 return (SET_ERROR(ENXIO));
1574 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1575 &logical_ashift, &physical_ashift);
1578 * Reset the vdev_reopening flag so that we actually close
1579 * the vdev on error.
1581 vd->vdev_reopening = B_FALSE;
1582 if (zio_injection_enabled && error == 0)
1583 error = zio_handle_device_injection(vd, NULL, ENXIO);
1586 if (vd->vdev_removed &&
1587 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1588 vd->vdev_removed = B_FALSE;
1590 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1591 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1592 vd->vdev_stat.vs_aux);
1594 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1595 vd->vdev_stat.vs_aux);
1600 vd->vdev_removed = B_FALSE;
1603 * Recheck the faulted flag now that we have confirmed that
1604 * the vdev is accessible. If we're faulted, bail.
1606 if (vd->vdev_faulted) {
1607 ASSERT(vd->vdev_children == 0);
1608 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1609 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1610 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1611 vd->vdev_label_aux);
1612 return (SET_ERROR(ENXIO));
1615 if (vd->vdev_degraded) {
1616 ASSERT(vd->vdev_children == 0);
1617 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1618 VDEV_AUX_ERR_EXCEEDED);
1620 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1624 * For hole or missing vdevs we just return success.
1626 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1629 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1630 trim_map_create(vd);
1632 for (int c = 0; c < vd->vdev_children; c++) {
1633 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1634 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1640 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1641 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1643 if (vd->vdev_children == 0) {
1644 if (osize < SPA_MINDEVSIZE) {
1645 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1646 VDEV_AUX_TOO_SMALL);
1647 return (SET_ERROR(EOVERFLOW));
1650 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1651 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1652 VDEV_LABEL_END_SIZE);
1654 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1655 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1656 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1657 VDEV_AUX_TOO_SMALL);
1658 return (SET_ERROR(EOVERFLOW));
1662 max_asize = max_osize;
1665 vd->vdev_psize = psize;
1668 * Make sure the allocatable size hasn't shrunk too much.
1670 if (asize < vd->vdev_min_asize) {
1671 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1672 VDEV_AUX_BAD_LABEL);
1673 return (SET_ERROR(EINVAL));
1676 vd->vdev_physical_ashift =
1677 MAX(physical_ashift, vd->vdev_physical_ashift);
1678 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1679 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1681 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1682 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1683 VDEV_AUX_ASHIFT_TOO_BIG);
1687 if (vd->vdev_asize == 0) {
1689 * This is the first-ever open, so use the computed values.
1690 * For testing purposes, a higher ashift can be requested.
1692 vd->vdev_asize = asize;
1693 vd->vdev_max_asize = max_asize;
1696 * Make sure the alignment requirement hasn't increased.
1698 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1699 vd->vdev_ops->vdev_op_leaf) {
1700 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1701 VDEV_AUX_BAD_LABEL);
1704 vd->vdev_max_asize = max_asize;
1708 * If all children are healthy we update asize if either:
1709 * The asize has increased, due to a device expansion caused by dynamic
1710 * LUN growth or vdev replacement, and automatic expansion is enabled;
1711 * making the additional space available.
1713 * The asize has decreased, due to a device shrink usually caused by a
1714 * vdev replace with a smaller device. This ensures that calculations
1715 * based of max_asize and asize e.g. esize are always valid. It's safe
1716 * to do this as we've already validated that asize is greater than
1719 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1720 ((asize > vd->vdev_asize &&
1721 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1722 (asize < vd->vdev_asize)))
1723 vd->vdev_asize = asize;
1725 vdev_set_min_asize(vd);
1728 * Ensure we can issue some IO before declaring the
1729 * vdev open for business.
1731 if (vd->vdev_ops->vdev_op_leaf &&
1732 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1733 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1734 VDEV_AUX_ERR_EXCEEDED);
1739 * Track the min and max ashift values for normal data devices.
1741 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1742 !vd->vdev_islog && vd->vdev_aux == NULL) {
1743 if (vd->vdev_ashift > spa->spa_max_ashift)
1744 spa->spa_max_ashift = vd->vdev_ashift;
1745 if (vd->vdev_ashift < spa->spa_min_ashift)
1746 spa->spa_min_ashift = vd->vdev_ashift;
1750 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1751 * resilver. But don't do this if we are doing a reopen for a scrub,
1752 * since this would just restart the scrub we are already doing.
1754 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1755 vdev_resilver_needed(vd, NULL, NULL))
1756 spa_async_request(spa, SPA_ASYNC_RESILVER);
1762 * Called once the vdevs are all opened, this routine validates the label
1763 * contents. This needs to be done before vdev_load() so that we don't
1764 * inadvertently do repair I/Os to the wrong device.
1766 * This function will only return failure if one of the vdevs indicates that it
1767 * has since been destroyed or exported. This is only possible if
1768 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1769 * will be updated but the function will return 0.
1772 vdev_validate(vdev_t *vd)
1774 spa_t *spa = vd->vdev_spa;
1776 uint64_t guid = 0, aux_guid = 0, top_guid;
1781 if (vdev_validate_skip)
1784 for (uint64_t c = 0; c < vd->vdev_children; c++)
1785 if (vdev_validate(vd->vdev_child[c]) != 0)
1786 return (SET_ERROR(EBADF));
1789 * If the device has already failed, or was marked offline, don't do
1790 * any further validation. Otherwise, label I/O will fail and we will
1791 * overwrite the previous state.
1793 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1797 * If we are performing an extreme rewind, we allow for a label that
1798 * was modified at a point after the current txg.
1799 * If config lock is not held do not check for the txg. spa_sync could
1800 * be updating the vdev's label before updating spa_last_synced_txg.
1802 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1803 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1806 txg = spa_last_synced_txg(spa);
1808 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1809 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1810 VDEV_AUX_BAD_LABEL);
1811 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
1812 "txg %llu", (u_longlong_t)txg);
1817 * Determine if this vdev has been split off into another
1818 * pool. If so, then refuse to open it.
1820 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1821 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1822 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1823 VDEV_AUX_SPLIT_POOL);
1825 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1829 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1830 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1831 VDEV_AUX_CORRUPT_DATA);
1833 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1834 ZPOOL_CONFIG_POOL_GUID);
1839 * If config is not trusted then ignore the spa guid check. This is
1840 * necessary because if the machine crashed during a re-guid the new
1841 * guid might have been written to all of the vdev labels, but not the
1842 * cached config. The check will be performed again once we have the
1843 * trusted config from the MOS.
1845 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1846 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1847 VDEV_AUX_CORRUPT_DATA);
1849 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
1850 "match config (%llu != %llu)", (u_longlong_t)guid,
1851 (u_longlong_t)spa_guid(spa));
1855 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1856 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1860 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
1861 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1862 VDEV_AUX_CORRUPT_DATA);
1864 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1869 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
1871 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1872 VDEV_AUX_CORRUPT_DATA);
1874 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1875 ZPOOL_CONFIG_TOP_GUID);
1880 * If this vdev just became a top-level vdev because its sibling was
1881 * detached, it will have adopted the parent's vdev guid -- but the
1882 * label may or may not be on disk yet. Fortunately, either version
1883 * of the label will have the same top guid, so if we're a top-level
1884 * vdev, we can safely compare to that instead.
1885 * However, if the config comes from a cachefile that failed to update
1886 * after the detach, a top-level vdev will appear as a non top-level
1887 * vdev in the config. Also relax the constraints if we perform an
1890 * If we split this vdev off instead, then we also check the
1891 * original pool's guid. We don't want to consider the vdev
1892 * corrupt if it is partway through a split operation.
1894 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
1895 boolean_t mismatch = B_FALSE;
1896 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
1897 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
1900 if (vd->vdev_guid != top_guid &&
1901 vd->vdev_top->vdev_guid != guid)
1906 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1907 VDEV_AUX_CORRUPT_DATA);
1909 vdev_dbgmsg(vd, "vdev_validate: config guid "
1910 "doesn't match label guid");
1911 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
1912 (u_longlong_t)vd->vdev_guid,
1913 (u_longlong_t)vd->vdev_top->vdev_guid);
1914 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
1915 "aux_guid %llu", (u_longlong_t)guid,
1916 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
1921 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1923 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1924 VDEV_AUX_CORRUPT_DATA);
1926 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1927 ZPOOL_CONFIG_POOL_STATE);
1934 * If this is a verbatim import, no need to check the
1935 * state of the pool.
1937 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1938 spa_load_state(spa) == SPA_LOAD_OPEN &&
1939 state != POOL_STATE_ACTIVE) {
1940 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
1941 "for spa %s", (u_longlong_t)state, spa->spa_name);
1942 return (SET_ERROR(EBADF));
1946 * If we were able to open and validate a vdev that was
1947 * previously marked permanently unavailable, clear that state
1950 if (vd->vdev_not_present)
1951 vd->vdev_not_present = 0;
1957 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
1959 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
1960 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
1961 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1962 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
1963 dvd->vdev_path, svd->vdev_path);
1964 spa_strfree(dvd->vdev_path);
1965 dvd->vdev_path = spa_strdup(svd->vdev_path);
1967 } else if (svd->vdev_path != NULL) {
1968 dvd->vdev_path = spa_strdup(svd->vdev_path);
1969 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1970 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
1975 * Recursively copy vdev paths from one vdev to another. Source and destination
1976 * vdev trees must have same geometry otherwise return error. Intended to copy
1977 * paths from userland config into MOS config.
1980 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
1982 if ((svd->vdev_ops == &vdev_missing_ops) ||
1983 (svd->vdev_ishole && dvd->vdev_ishole) ||
1984 (dvd->vdev_ops == &vdev_indirect_ops))
1987 if (svd->vdev_ops != dvd->vdev_ops) {
1988 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
1989 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
1990 return (SET_ERROR(EINVAL));
1993 if (svd->vdev_guid != dvd->vdev_guid) {
1994 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
1995 "%llu)", (u_longlong_t)svd->vdev_guid,
1996 (u_longlong_t)dvd->vdev_guid);
1997 return (SET_ERROR(EINVAL));
2000 if (svd->vdev_children != dvd->vdev_children) {
2001 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2002 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2003 (u_longlong_t)dvd->vdev_children);
2004 return (SET_ERROR(EINVAL));
2007 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2008 int error = vdev_copy_path_strict(svd->vdev_child[i],
2009 dvd->vdev_child[i]);
2014 if (svd->vdev_ops->vdev_op_leaf)
2015 vdev_copy_path_impl(svd, dvd);
2021 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2023 ASSERT(stvd->vdev_top == stvd);
2024 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2026 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2027 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2030 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2034 * The idea here is that while a vdev can shift positions within
2035 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2036 * step outside of it.
2038 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2040 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2043 ASSERT(vd->vdev_ops->vdev_op_leaf);
2045 vdev_copy_path_impl(vd, dvd);
2049 * Recursively copy vdev paths from one root vdev to another. Source and
2050 * destination vdev trees may differ in geometry. For each destination leaf
2051 * vdev, search a vdev with the same guid and top vdev id in the source.
2052 * Intended to copy paths from userland config into MOS config.
2055 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2057 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2058 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2059 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2061 for (uint64_t i = 0; i < children; i++) {
2062 vdev_copy_path_search(srvd->vdev_child[i],
2063 drvd->vdev_child[i]);
2068 * Close a virtual device.
2071 vdev_close(vdev_t *vd)
2073 spa_t *spa = vd->vdev_spa;
2074 vdev_t *pvd = vd->vdev_parent;
2076 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2079 * If our parent is reopening, then we are as well, unless we are
2082 if (pvd != NULL && pvd->vdev_reopening)
2083 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2085 vd->vdev_ops->vdev_op_close(vd);
2087 vdev_cache_purge(vd);
2089 if (vd->vdev_ops->vdev_op_leaf)
2090 trim_map_destroy(vd);
2093 * We record the previous state before we close it, so that if we are
2094 * doing a reopen(), we don't generate FMA ereports if we notice that
2095 * it's still faulted.
2097 vd->vdev_prevstate = vd->vdev_state;
2099 if (vd->vdev_offline)
2100 vd->vdev_state = VDEV_STATE_OFFLINE;
2102 vd->vdev_state = VDEV_STATE_CLOSED;
2103 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2107 vdev_hold(vdev_t *vd)
2109 spa_t *spa = vd->vdev_spa;
2111 ASSERT(spa_is_root(spa));
2112 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2115 for (int c = 0; c < vd->vdev_children; c++)
2116 vdev_hold(vd->vdev_child[c]);
2118 if (vd->vdev_ops->vdev_op_leaf)
2119 vd->vdev_ops->vdev_op_hold(vd);
2123 vdev_rele(vdev_t *vd)
2125 spa_t *spa = vd->vdev_spa;
2127 ASSERT(spa_is_root(spa));
2128 for (int c = 0; c < vd->vdev_children; c++)
2129 vdev_rele(vd->vdev_child[c]);
2131 if (vd->vdev_ops->vdev_op_leaf)
2132 vd->vdev_ops->vdev_op_rele(vd);
2136 * Reopen all interior vdevs and any unopened leaves. We don't actually
2137 * reopen leaf vdevs which had previously been opened as they might deadlock
2138 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2139 * If the leaf has never been opened then open it, as usual.
2142 vdev_reopen(vdev_t *vd)
2144 spa_t *spa = vd->vdev_spa;
2146 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2148 /* set the reopening flag unless we're taking the vdev offline */
2149 vd->vdev_reopening = !vd->vdev_offline;
2151 (void) vdev_open(vd);
2154 * Call vdev_validate() here to make sure we have the same device.
2155 * Otherwise, a device with an invalid label could be successfully
2156 * opened in response to vdev_reopen().
2159 (void) vdev_validate_aux(vd);
2160 if (vdev_readable(vd) && vdev_writeable(vd) &&
2161 vd->vdev_aux == &spa->spa_l2cache &&
2162 !l2arc_vdev_present(vd))
2163 l2arc_add_vdev(spa, vd);
2165 (void) vdev_validate(vd);
2169 * Reassess parent vdev's health.
2171 vdev_propagate_state(vd);
2175 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2180 * Normally, partial opens (e.g. of a mirror) are allowed.
2181 * For a create, however, we want to fail the request if
2182 * there are any components we can't open.
2184 error = vdev_open(vd);
2186 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2188 return (error ? error : ENXIO);
2192 * Recursively load DTLs and initialize all labels.
2194 if ((error = vdev_dtl_load(vd)) != 0 ||
2195 (error = vdev_label_init(vd, txg, isreplacing ?
2196 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2205 vdev_metaslab_set_size(vdev_t *vd)
2207 uint64_t asize = vd->vdev_asize;
2208 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2212 * There are two dimensions to the metaslab sizing calculation:
2213 * the size of the metaslab and the count of metaslabs per vdev.
2215 * The default values used below are a good balance between memory
2216 * usage (larger metaslab size means more memory needed for loaded
2217 * metaslabs; more metaslabs means more memory needed for the
2218 * metaslab_t structs), metaslab load time (larger metaslabs take
2219 * longer to load), and metaslab sync time (more metaslabs means
2220 * more time spent syncing all of them).
2222 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2223 * The range of the dimensions are as follows:
2225 * 2^29 <= ms_size <= 2^34
2226 * 16 <= ms_count <= 131,072
2228 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2229 * at least 512MB (2^29) to minimize fragmentation effects when
2230 * testing with smaller devices. However, the count constraint
2231 * of at least 16 metaslabs will override this minimum size goal.
2233 * On the upper end of vdev sizes, we aim for a maximum metaslab
2234 * size of 16GB. However, we will cap the total count to 2^17
2235 * metaslabs to keep our memory footprint in check and let the
2236 * metaslab size grow from there if that limit is hit.
2238 * The net effect of applying above constrains is summarized below.
2240 * vdev size metaslab count
2241 * --------------|-----------------
2243 * 8GB - 100GB one per 512MB
2245 * 3TB - 2PB one per 16GB
2247 * --------------------------------
2249 * Finally, note that all of the above calculate the initial
2250 * number of metaslabs. Expanding a top-level vdev will result
2251 * in additional metaslabs being allocated making it possible
2252 * to exceed the zfs_vdev_ms_count_limit.
2255 if (ms_count < zfs_vdev_min_ms_count)
2256 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2257 else if (ms_count > zfs_vdev_default_ms_count)
2258 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2260 ms_shift = zfs_vdev_default_ms_shift;
2262 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2263 ms_shift = SPA_MAXBLOCKSHIFT;
2264 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2265 ms_shift = zfs_vdev_max_ms_shift;
2266 /* cap the total count to constrain memory footprint */
2267 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2268 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2271 vd->vdev_ms_shift = ms_shift;
2272 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2276 * Maximize performance by inflating the configured ashift for top level
2277 * vdevs to be as close to the physical ashift as possible while maintaining
2278 * administrator defined limits and ensuring it doesn't go below the
2282 vdev_ashift_optimize(vdev_t *vd)
2284 if (vd == vd->vdev_top) {
2285 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
2286 vd->vdev_ashift = MIN(
2287 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
2288 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
2291 * Unusual case where logical ashift > physical ashift
2292 * so we can't cap the calculated ashift based on max
2293 * ashift as that would cause failures.
2294 * We still check if we need to increase it to match
2297 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
2304 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2306 ASSERT(vd == vd->vdev_top);
2307 /* indirect vdevs don't have metaslabs or dtls */
2308 ASSERT(vdev_is_concrete(vd) || flags == 0);
2309 ASSERT(ISP2(flags));
2310 ASSERT(spa_writeable(vd->vdev_spa));
2312 if (flags & VDD_METASLAB)
2313 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2315 if (flags & VDD_DTL)
2316 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2318 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2322 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2324 for (int c = 0; c < vd->vdev_children; c++)
2325 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2327 if (vd->vdev_ops->vdev_op_leaf)
2328 vdev_dirty(vd->vdev_top, flags, vd, txg);
2334 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2335 * the vdev has less than perfect replication. There are four kinds of DTL:
2337 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2339 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2341 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2342 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2343 * txgs that was scrubbed.
2345 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2346 * persistent errors or just some device being offline.
2347 * Unlike the other three, the DTL_OUTAGE map is not generally
2348 * maintained; it's only computed when needed, typically to
2349 * determine whether a device can be detached.
2351 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2352 * either has the data or it doesn't.
2354 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2355 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2356 * if any child is less than fully replicated, then so is its parent.
2357 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2358 * comprising only those txgs which appear in 'maxfaults' or more children;
2359 * those are the txgs we don't have enough replication to read. For example,
2360 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2361 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2362 * two child DTL_MISSING maps.
2364 * It should be clear from the above that to compute the DTLs and outage maps
2365 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2366 * Therefore, that is all we keep on disk. When loading the pool, or after
2367 * a configuration change, we generate all other DTLs from first principles.
2370 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2372 range_tree_t *rt = vd->vdev_dtl[t];
2374 ASSERT(t < DTL_TYPES);
2375 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2376 ASSERT(spa_writeable(vd->vdev_spa));
2378 mutex_enter(&vd->vdev_dtl_lock);
2379 if (!range_tree_contains(rt, txg, size))
2380 range_tree_add(rt, txg, size);
2381 mutex_exit(&vd->vdev_dtl_lock);
2385 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2387 range_tree_t *rt = vd->vdev_dtl[t];
2388 boolean_t dirty = B_FALSE;
2390 ASSERT(t < DTL_TYPES);
2391 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2394 * While we are loading the pool, the DTLs have not been loaded yet.
2395 * Ignore the DTLs and try all devices. This avoids a recursive
2396 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2397 * when loading the pool (relying on the checksum to ensure that
2398 * we get the right data -- note that we while loading, we are
2399 * only reading the MOS, which is always checksummed).
2401 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2404 mutex_enter(&vd->vdev_dtl_lock);
2405 if (!range_tree_is_empty(rt))
2406 dirty = range_tree_contains(rt, txg, size);
2407 mutex_exit(&vd->vdev_dtl_lock);
2413 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2415 range_tree_t *rt = vd->vdev_dtl[t];
2418 mutex_enter(&vd->vdev_dtl_lock);
2419 empty = range_tree_is_empty(rt);
2420 mutex_exit(&vd->vdev_dtl_lock);
2426 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2429 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2431 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2433 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2434 vd->vdev_ops->vdev_op_leaf)
2437 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2441 * Returns the lowest txg in the DTL range.
2444 vdev_dtl_min(vdev_t *vd)
2448 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2449 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2450 ASSERT0(vd->vdev_children);
2452 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2453 return (rs->rs_start - 1);
2457 * Returns the highest txg in the DTL.
2460 vdev_dtl_max(vdev_t *vd)
2464 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2465 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2466 ASSERT0(vd->vdev_children);
2468 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2469 return (rs->rs_end);
2473 * Determine if a resilvering vdev should remove any DTL entries from
2474 * its range. If the vdev was resilvering for the entire duration of the
2475 * scan then it should excise that range from its DTLs. Otherwise, this
2476 * vdev is considered partially resilvered and should leave its DTL
2477 * entries intact. The comment in vdev_dtl_reassess() describes how we
2481 vdev_dtl_should_excise(vdev_t *vd)
2483 spa_t *spa = vd->vdev_spa;
2484 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2486 ASSERT0(scn->scn_phys.scn_errors);
2487 ASSERT0(vd->vdev_children);
2489 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2492 if (vd->vdev_resilver_txg == 0 ||
2493 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2497 * When a resilver is initiated the scan will assign the scn_max_txg
2498 * value to the highest txg value that exists in all DTLs. If this
2499 * device's max DTL is not part of this scan (i.e. it is not in
2500 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2503 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2504 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2505 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2506 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2513 * Reassess DTLs after a config change or scrub completion.
2516 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2518 spa_t *spa = vd->vdev_spa;
2522 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2524 for (int c = 0; c < vd->vdev_children; c++)
2525 vdev_dtl_reassess(vd->vdev_child[c], txg,
2526 scrub_txg, scrub_done);
2528 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2531 if (vd->vdev_ops->vdev_op_leaf) {
2532 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2534 mutex_enter(&vd->vdev_dtl_lock);
2537 * If we've completed a scan cleanly then determine
2538 * if this vdev should remove any DTLs. We only want to
2539 * excise regions on vdevs that were available during
2540 * the entire duration of this scan.
2542 if (scrub_txg != 0 &&
2543 (spa->spa_scrub_started ||
2544 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2545 vdev_dtl_should_excise(vd)) {
2547 * We completed a scrub up to scrub_txg. If we
2548 * did it without rebooting, then the scrub dtl
2549 * will be valid, so excise the old region and
2550 * fold in the scrub dtl. Otherwise, leave the
2551 * dtl as-is if there was an error.
2553 * There's little trick here: to excise the beginning
2554 * of the DTL_MISSING map, we put it into a reference
2555 * tree and then add a segment with refcnt -1 that
2556 * covers the range [0, scrub_txg). This means
2557 * that each txg in that range has refcnt -1 or 0.
2558 * We then add DTL_SCRUB with a refcnt of 2, so that
2559 * entries in the range [0, scrub_txg) will have a
2560 * positive refcnt -- either 1 or 2. We then convert
2561 * the reference tree into the new DTL_MISSING map.
2563 space_reftree_create(&reftree);
2564 space_reftree_add_map(&reftree,
2565 vd->vdev_dtl[DTL_MISSING], 1);
2566 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2567 space_reftree_add_map(&reftree,
2568 vd->vdev_dtl[DTL_SCRUB], 2);
2569 space_reftree_generate_map(&reftree,
2570 vd->vdev_dtl[DTL_MISSING], 1);
2571 space_reftree_destroy(&reftree);
2573 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2574 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2575 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2577 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2578 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2579 if (!vdev_readable(vd))
2580 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2582 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2583 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2586 * If the vdev was resilvering and no longer has any
2587 * DTLs then reset its resilvering flag and dirty
2588 * the top level so that we persist the change.
2590 if (vd->vdev_resilver_txg != 0 &&
2591 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2592 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2593 vd->vdev_resilver_txg = 0;
2594 vdev_config_dirty(vd->vdev_top);
2597 mutex_exit(&vd->vdev_dtl_lock);
2600 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2604 mutex_enter(&vd->vdev_dtl_lock);
2605 for (int t = 0; t < DTL_TYPES; t++) {
2606 /* account for child's outage in parent's missing map */
2607 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2609 continue; /* leaf vdevs only */
2610 if (t == DTL_PARTIAL)
2611 minref = 1; /* i.e. non-zero */
2612 else if (vd->vdev_nparity != 0)
2613 minref = vd->vdev_nparity + 1; /* RAID-Z */
2615 minref = vd->vdev_children; /* any kind of mirror */
2616 space_reftree_create(&reftree);
2617 for (int c = 0; c < vd->vdev_children; c++) {
2618 vdev_t *cvd = vd->vdev_child[c];
2619 mutex_enter(&cvd->vdev_dtl_lock);
2620 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2621 mutex_exit(&cvd->vdev_dtl_lock);
2623 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2624 space_reftree_destroy(&reftree);
2626 mutex_exit(&vd->vdev_dtl_lock);
2630 vdev_dtl_load(vdev_t *vd)
2632 spa_t *spa = vd->vdev_spa;
2633 objset_t *mos = spa->spa_meta_objset;
2636 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2637 ASSERT(vdev_is_concrete(vd));
2639 error = space_map_open(&vd->vdev_dtl_sm, mos,
2640 vd->vdev_dtl_object, 0, -1ULL, 0);
2643 ASSERT(vd->vdev_dtl_sm != NULL);
2645 mutex_enter(&vd->vdev_dtl_lock);
2648 * Now that we've opened the space_map we need to update
2651 space_map_update(vd->vdev_dtl_sm);
2653 error = space_map_load(vd->vdev_dtl_sm,
2654 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2655 mutex_exit(&vd->vdev_dtl_lock);
2660 for (int c = 0; c < vd->vdev_children; c++) {
2661 error = vdev_dtl_load(vd->vdev_child[c]);
2670 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2672 spa_t *spa = vd->vdev_spa;
2674 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2675 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2680 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2682 spa_t *spa = vd->vdev_spa;
2683 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2684 DMU_OT_NONE, 0, tx);
2687 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2694 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2696 if (vd->vdev_ops != &vdev_hole_ops &&
2697 vd->vdev_ops != &vdev_missing_ops &&
2698 vd->vdev_ops != &vdev_root_ops &&
2699 !vd->vdev_top->vdev_removing) {
2700 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2701 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2703 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2704 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2707 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2708 vdev_construct_zaps(vd->vdev_child[i], tx);
2713 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2715 spa_t *spa = vd->vdev_spa;
2716 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2717 objset_t *mos = spa->spa_meta_objset;
2718 range_tree_t *rtsync;
2720 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2722 ASSERT(vdev_is_concrete(vd));
2723 ASSERT(vd->vdev_ops->vdev_op_leaf);
2725 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2727 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2728 mutex_enter(&vd->vdev_dtl_lock);
2729 space_map_free(vd->vdev_dtl_sm, tx);
2730 space_map_close(vd->vdev_dtl_sm);
2731 vd->vdev_dtl_sm = NULL;
2732 mutex_exit(&vd->vdev_dtl_lock);
2735 * We only destroy the leaf ZAP for detached leaves or for
2736 * removed log devices. Removed data devices handle leaf ZAP
2737 * cleanup later, once cancellation is no longer possible.
2739 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2740 vd->vdev_top->vdev_islog)) {
2741 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2742 vd->vdev_leaf_zap = 0;
2749 if (vd->vdev_dtl_sm == NULL) {
2750 uint64_t new_object;
2752 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx);
2753 VERIFY3U(new_object, !=, 0);
2755 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2757 ASSERT(vd->vdev_dtl_sm != NULL);
2760 rtsync = range_tree_create(NULL, NULL);
2762 mutex_enter(&vd->vdev_dtl_lock);
2763 range_tree_walk(rt, range_tree_add, rtsync);
2764 mutex_exit(&vd->vdev_dtl_lock);
2766 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx);
2767 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
2768 range_tree_vacate(rtsync, NULL, NULL);
2770 range_tree_destroy(rtsync);
2773 * If the object for the space map has changed then dirty
2774 * the top level so that we update the config.
2776 if (object != space_map_object(vd->vdev_dtl_sm)) {
2777 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2778 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2779 (u_longlong_t)object,
2780 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2781 vdev_config_dirty(vd->vdev_top);
2786 mutex_enter(&vd->vdev_dtl_lock);
2787 space_map_update(vd->vdev_dtl_sm);
2788 mutex_exit(&vd->vdev_dtl_lock);
2792 * Determine whether the specified vdev can be offlined/detached/removed
2793 * without losing data.
2796 vdev_dtl_required(vdev_t *vd)
2798 spa_t *spa = vd->vdev_spa;
2799 vdev_t *tvd = vd->vdev_top;
2800 uint8_t cant_read = vd->vdev_cant_read;
2803 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2805 if (vd == spa->spa_root_vdev || vd == tvd)
2809 * Temporarily mark the device as unreadable, and then determine
2810 * whether this results in any DTL outages in the top-level vdev.
2811 * If not, we can safely offline/detach/remove the device.
2813 vd->vdev_cant_read = B_TRUE;
2814 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2815 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2816 vd->vdev_cant_read = cant_read;
2817 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2819 if (!required && zio_injection_enabled)
2820 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2826 * Determine if resilver is needed, and if so the txg range.
2829 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2831 boolean_t needed = B_FALSE;
2832 uint64_t thismin = UINT64_MAX;
2833 uint64_t thismax = 0;
2835 if (vd->vdev_children == 0) {
2836 mutex_enter(&vd->vdev_dtl_lock);
2837 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2838 vdev_writeable(vd)) {
2840 thismin = vdev_dtl_min(vd);
2841 thismax = vdev_dtl_max(vd);
2844 mutex_exit(&vd->vdev_dtl_lock);
2846 for (int c = 0; c < vd->vdev_children; c++) {
2847 vdev_t *cvd = vd->vdev_child[c];
2848 uint64_t cmin, cmax;
2850 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2851 thismin = MIN(thismin, cmin);
2852 thismax = MAX(thismax, cmax);
2858 if (needed && minp) {
2866 * Gets the checkpoint space map object from the vdev's ZAP.
2867 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2868 * or the ZAP doesn't exist yet.
2871 vdev_checkpoint_sm_object(vdev_t *vd)
2873 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
2874 if (vd->vdev_top_zap == 0) {
2878 uint64_t sm_obj = 0;
2879 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
2880 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj);
2882 ASSERT(err == 0 || err == ENOENT);
2888 vdev_load(vdev_t *vd)
2892 * Recursively load all children.
2894 for (int c = 0; c < vd->vdev_children; c++) {
2895 error = vdev_load(vd->vdev_child[c]);
2901 vdev_set_deflate_ratio(vd);
2904 * If this is a top-level vdev, initialize its metaslabs.
2906 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2907 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2908 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2909 VDEV_AUX_CORRUPT_DATA);
2910 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2911 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2912 (u_longlong_t)vd->vdev_asize);
2913 return (SET_ERROR(ENXIO));
2914 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2915 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2916 "[error=%d]", error);
2917 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2918 VDEV_AUX_CORRUPT_DATA);
2922 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd);
2923 if (checkpoint_sm_obj != 0) {
2924 objset_t *mos = spa_meta_objset(vd->vdev_spa);
2925 ASSERT(vd->vdev_asize != 0);
2926 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
2928 if ((error = space_map_open(&vd->vdev_checkpoint_sm,
2929 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
2930 vd->vdev_ashift))) {
2931 vdev_dbgmsg(vd, "vdev_load: space_map_open "
2932 "failed for checkpoint spacemap (obj %llu) "
2934 (u_longlong_t)checkpoint_sm_obj, error);
2937 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
2938 space_map_update(vd->vdev_checkpoint_sm);
2941 * Since the checkpoint_sm contains free entries
2942 * exclusively we can use sm_alloc to indicate the
2943 * culmulative checkpointed space that has been freed.
2945 vd->vdev_stat.vs_checkpoint_space =
2946 -vd->vdev_checkpoint_sm->sm_alloc;
2947 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
2948 vd->vdev_stat.vs_checkpoint_space;
2953 * If this is a leaf vdev, load its DTL.
2955 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2956 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2957 VDEV_AUX_CORRUPT_DATA);
2958 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
2959 "[error=%d]", error);
2963 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2964 if (obsolete_sm_object != 0) {
2965 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2966 ASSERT(vd->vdev_asize != 0);
2967 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
2969 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2970 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2971 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2972 VDEV_AUX_CORRUPT_DATA);
2973 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
2974 "obsolete spacemap (obj %llu) [error=%d]",
2975 (u_longlong_t)obsolete_sm_object, error);
2978 space_map_update(vd->vdev_obsolete_sm);
2985 * The special vdev case is used for hot spares and l2cache devices. Its
2986 * sole purpose it to set the vdev state for the associated vdev. To do this,
2987 * we make sure that we can open the underlying device, then try to read the
2988 * label, and make sure that the label is sane and that it hasn't been
2989 * repurposed to another pool.
2992 vdev_validate_aux(vdev_t *vd)
2995 uint64_t guid, version;
2998 if (!vdev_readable(vd))
3001 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3002 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3003 VDEV_AUX_CORRUPT_DATA);
3007 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3008 !SPA_VERSION_IS_SUPPORTED(version) ||
3009 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3010 guid != vd->vdev_guid ||
3011 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3012 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3013 VDEV_AUX_CORRUPT_DATA);
3019 * We don't actually check the pool state here. If it's in fact in
3020 * use by another pool, we update this fact on the fly when requested.
3027 * Free the objects used to store this vdev's spacemaps, and the array
3028 * that points to them.
3031 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3033 if (vd->vdev_ms_array == 0)
3036 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3037 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3038 size_t array_bytes = array_count * sizeof (uint64_t);
3039 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3040 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3041 array_bytes, smobj_array, 0));
3043 for (uint64_t i = 0; i < array_count; i++) {
3044 uint64_t smobj = smobj_array[i];
3048 space_map_free_obj(mos, smobj, tx);
3051 kmem_free(smobj_array, array_bytes);
3052 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3053 vd->vdev_ms_array = 0;
3057 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3059 spa_t *spa = vd->vdev_spa;
3061 ASSERT(vd->vdev_islog);
3062 ASSERT(vd == vd->vdev_top);
3063 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3065 if (vd->vdev_ms != NULL) {
3066 metaslab_group_t *mg = vd->vdev_mg;
3068 metaslab_group_histogram_verify(mg);
3069 metaslab_class_histogram_verify(mg->mg_class);
3071 for (int m = 0; m < vd->vdev_ms_count; m++) {
3072 metaslab_t *msp = vd->vdev_ms[m];
3074 if (msp == NULL || msp->ms_sm == NULL)
3077 mutex_enter(&msp->ms_lock);
3079 * If the metaslab was not loaded when the vdev
3080 * was removed then the histogram accounting may
3081 * not be accurate. Update the histogram information
3082 * here so that we ensure that the metaslab group
3083 * and metaslab class are up-to-date.
3085 metaslab_group_histogram_remove(mg, msp);
3087 VERIFY0(space_map_allocated(msp->ms_sm));
3088 space_map_close(msp->ms_sm);
3090 mutex_exit(&msp->ms_lock);
3093 if (vd->vdev_checkpoint_sm != NULL) {
3094 ASSERT(spa_has_checkpoint(spa));
3095 space_map_close(vd->vdev_checkpoint_sm);
3096 vd->vdev_checkpoint_sm = NULL;
3099 metaslab_group_histogram_verify(mg);
3100 metaslab_class_histogram_verify(mg->mg_class);
3101 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
3102 ASSERT0(mg->mg_histogram[i]);
3105 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3107 vdev_destroy_spacemaps(vd, tx);
3108 if (vd->vdev_top_zap != 0) {
3109 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3110 vd->vdev_top_zap = 0;
3117 vdev_sync_done(vdev_t *vd, uint64_t txg)
3120 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3122 ASSERT(vdev_is_concrete(vd));
3124 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3126 metaslab_sync_done(msp, txg);
3129 metaslab_sync_reassess(vd->vdev_mg);
3133 vdev_sync(vdev_t *vd, uint64_t txg)
3135 spa_t *spa = vd->vdev_spa;
3140 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3143 ASSERT(vd->vdev_removing ||
3144 vd->vdev_ops == &vdev_indirect_ops);
3146 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3147 vdev_indirect_sync_obsolete(vd, tx);
3151 * If the vdev is indirect, it can't have dirty
3152 * metaslabs or DTLs.
3154 if (vd->vdev_ops == &vdev_indirect_ops) {
3155 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3156 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3161 ASSERT(vdev_is_concrete(vd));
3163 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3164 !vd->vdev_removing) {
3165 ASSERT(vd == vd->vdev_top);
3166 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3167 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3168 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3169 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3170 ASSERT(vd->vdev_ms_array != 0);
3171 vdev_config_dirty(vd);
3175 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3176 metaslab_sync(msp, txg);
3177 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3180 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3181 vdev_dtl_sync(lvd, txg);
3184 * If this is an empty log device being removed, destroy the
3185 * metadata associated with it.
3187 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3188 vdev_remove_empty_log(vd, txg);
3190 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3194 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3196 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3200 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3201 * not be opened, and no I/O is attempted.
3204 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3208 spa_vdev_state_enter(spa, SCL_NONE);
3210 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3211 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3213 if (!vd->vdev_ops->vdev_op_leaf)
3214 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3219 * We don't directly use the aux state here, but if we do a
3220 * vdev_reopen(), we need this value to be present to remember why we
3223 vd->vdev_label_aux = aux;
3226 * Faulted state takes precedence over degraded.
3228 vd->vdev_delayed_close = B_FALSE;
3229 vd->vdev_faulted = 1ULL;
3230 vd->vdev_degraded = 0ULL;
3231 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3234 * If this device has the only valid copy of the data, then
3235 * back off and simply mark the vdev as degraded instead.
3237 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3238 vd->vdev_degraded = 1ULL;
3239 vd->vdev_faulted = 0ULL;
3242 * If we reopen the device and it's not dead, only then do we
3247 if (vdev_readable(vd))
3248 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3251 return (spa_vdev_state_exit(spa, vd, 0));
3255 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3256 * user that something is wrong. The vdev continues to operate as normal as far
3257 * as I/O is concerned.
3260 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3264 spa_vdev_state_enter(spa, SCL_NONE);
3266 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3267 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3269 if (!vd->vdev_ops->vdev_op_leaf)
3270 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3273 * If the vdev is already faulted, then don't do anything.
3275 if (vd->vdev_faulted || vd->vdev_degraded)
3276 return (spa_vdev_state_exit(spa, NULL, 0));
3278 vd->vdev_degraded = 1ULL;
3279 if (!vdev_is_dead(vd))
3280 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3283 return (spa_vdev_state_exit(spa, vd, 0));
3287 * Online the given vdev.
3289 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3290 * spare device should be detached when the device finishes resilvering.
3291 * Second, the online should be treated like a 'test' online case, so no FMA
3292 * events are generated if the device fails to open.
3295 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3297 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3298 boolean_t wasoffline;
3299 vdev_state_t oldstate;
3301 spa_vdev_state_enter(spa, SCL_NONE);
3303 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3304 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3306 if (!vd->vdev_ops->vdev_op_leaf)
3307 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3309 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3310 oldstate = vd->vdev_state;
3313 vd->vdev_offline = B_FALSE;
3314 vd->vdev_tmpoffline = B_FALSE;
3315 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3316 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3318 /* XXX - L2ARC 1.0 does not support expansion */
3319 if (!vd->vdev_aux) {
3320 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3321 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
3325 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3327 if (!vd->vdev_aux) {
3328 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3329 pvd->vdev_expanding = B_FALSE;
3333 *newstate = vd->vdev_state;
3334 if ((flags & ZFS_ONLINE_UNSPARE) &&
3335 !vdev_is_dead(vd) && vd->vdev_parent &&
3336 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3337 vd->vdev_parent->vdev_child[0] == vd)
3338 vd->vdev_unspare = B_TRUE;
3340 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3342 /* XXX - L2ARC 1.0 does not support expansion */
3344 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3345 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3348 /* Restart initializing if necessary */
3349 mutex_enter(&vd->vdev_initialize_lock);
3350 if (vdev_writeable(vd) &&
3351 vd->vdev_initialize_thread == NULL &&
3352 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3353 (void) vdev_initialize(vd);
3355 mutex_exit(&vd->vdev_initialize_lock);
3358 (oldstate < VDEV_STATE_DEGRADED &&
3359 vd->vdev_state >= VDEV_STATE_DEGRADED))
3360 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3362 return (spa_vdev_state_exit(spa, vd, 0));
3366 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3370 uint64_t generation;
3371 metaslab_group_t *mg;
3374 spa_vdev_state_enter(spa, SCL_ALLOC);
3376 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3377 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3379 if (!vd->vdev_ops->vdev_op_leaf)
3380 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3384 generation = spa->spa_config_generation + 1;
3387 * If the device isn't already offline, try to offline it.
3389 if (!vd->vdev_offline) {
3391 * If this device has the only valid copy of some data,
3392 * don't allow it to be offlined. Log devices are always
3395 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3396 vdev_dtl_required(vd))
3397 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3400 * If the top-level is a slog and it has had allocations
3401 * then proceed. We check that the vdev's metaslab group
3402 * is not NULL since it's possible that we may have just
3403 * added this vdev but not yet initialized its metaslabs.
3405 if (tvd->vdev_islog && mg != NULL) {
3407 * Prevent any future allocations.
3409 metaslab_group_passivate(mg);
3410 (void) spa_vdev_state_exit(spa, vd, 0);
3412 error = spa_reset_logs(spa);
3415 * If the log device was successfully reset but has
3416 * checkpointed data, do not offline it.
3419 tvd->vdev_checkpoint_sm != NULL) {
3420 ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc,
3422 error = ZFS_ERR_CHECKPOINT_EXISTS;
3425 spa_vdev_state_enter(spa, SCL_ALLOC);
3428 * Check to see if the config has changed.
3430 if (error || generation != spa->spa_config_generation) {
3431 metaslab_group_activate(mg);
3433 return (spa_vdev_state_exit(spa,
3435 (void) spa_vdev_state_exit(spa, vd, 0);
3438 ASSERT0(tvd->vdev_stat.vs_alloc);
3442 * Offline this device and reopen its top-level vdev.
3443 * If the top-level vdev is a log device then just offline
3444 * it. Otherwise, if this action results in the top-level
3445 * vdev becoming unusable, undo it and fail the request.
3447 vd->vdev_offline = B_TRUE;
3450 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3451 vdev_is_dead(tvd)) {
3452 vd->vdev_offline = B_FALSE;
3454 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3458 * Add the device back into the metaslab rotor so that
3459 * once we online the device it's open for business.
3461 if (tvd->vdev_islog && mg != NULL)
3462 metaslab_group_activate(mg);
3465 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3467 return (spa_vdev_state_exit(spa, vd, 0));
3471 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3475 mutex_enter(&spa->spa_vdev_top_lock);
3476 error = vdev_offline_locked(spa, guid, flags);
3477 mutex_exit(&spa->spa_vdev_top_lock);
3483 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3484 * vdev_offline(), we assume the spa config is locked. We also clear all
3485 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3488 vdev_clear(spa_t *spa, vdev_t *vd)
3490 vdev_t *rvd = spa->spa_root_vdev;
3492 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3497 vd->vdev_stat.vs_read_errors = 0;
3498 vd->vdev_stat.vs_write_errors = 0;
3499 vd->vdev_stat.vs_checksum_errors = 0;
3501 for (int c = 0; c < vd->vdev_children; c++)
3502 vdev_clear(spa, vd->vdev_child[c]);
3505 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
3506 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
3508 for (int c = 0; c < spa->spa_spares.sav_count; c++)
3509 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
3513 * It makes no sense to "clear" an indirect vdev.
3515 if (!vdev_is_concrete(vd))
3519 * If we're in the FAULTED state or have experienced failed I/O, then
3520 * clear the persistent state and attempt to reopen the device. We
3521 * also mark the vdev config dirty, so that the new faulted state is
3522 * written out to disk.
3524 if (vd->vdev_faulted || vd->vdev_degraded ||
3525 !vdev_readable(vd) || !vdev_writeable(vd)) {
3528 * When reopening in reponse to a clear event, it may be due to
3529 * a fmadm repair request. In this case, if the device is
3530 * still broken, we want to still post the ereport again.
3532 vd->vdev_forcefault = B_TRUE;
3534 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3535 vd->vdev_cant_read = B_FALSE;
3536 vd->vdev_cant_write = B_FALSE;
3538 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3540 vd->vdev_forcefault = B_FALSE;
3542 if (vd != rvd && vdev_writeable(vd->vdev_top))
3543 vdev_state_dirty(vd->vdev_top);
3545 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
3546 spa_async_request(spa, SPA_ASYNC_RESILVER);
3548 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3552 * When clearing a FMA-diagnosed fault, we always want to
3553 * unspare the device, as we assume that the original spare was
3554 * done in response to the FMA fault.
3556 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3557 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3558 vd->vdev_parent->vdev_child[0] == vd)
3559 vd->vdev_unspare = B_TRUE;
3563 vdev_is_dead(vdev_t *vd)
3566 * Holes and missing devices are always considered "dead".
3567 * This simplifies the code since we don't have to check for
3568 * these types of devices in the various code paths.
3569 * Instead we rely on the fact that we skip over dead devices
3570 * before issuing I/O to them.
3572 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3573 vd->vdev_ops == &vdev_hole_ops ||
3574 vd->vdev_ops == &vdev_missing_ops);
3578 vdev_readable(vdev_t *vd)
3580 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3584 vdev_writeable(vdev_t *vd)
3586 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3587 vdev_is_concrete(vd));
3591 vdev_allocatable(vdev_t *vd)
3593 uint64_t state = vd->vdev_state;
3596 * We currently allow allocations from vdevs which may be in the
3597 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3598 * fails to reopen then we'll catch it later when we're holding
3599 * the proper locks. Note that we have to get the vdev state
3600 * in a local variable because although it changes atomically,
3601 * we're asking two separate questions about it.
3603 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3604 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3605 vd->vdev_mg->mg_initialized);
3609 vdev_accessible(vdev_t *vd, zio_t *zio)
3611 ASSERT(zio->io_vd == vd);
3613 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3616 if (zio->io_type == ZIO_TYPE_READ)
3617 return (!vd->vdev_cant_read);
3619 if (zio->io_type == ZIO_TYPE_WRITE)
3620 return (!vd->vdev_cant_write);
3626 vdev_is_spacemap_addressable(vdev_t *vd)
3628 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
3632 * If double-word space map entries are not enabled we assume
3633 * 47 bits of the space map entry are dedicated to the entry's
3634 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
3635 * to calculate the maximum address that can be described by a
3636 * space map entry for the given device.
3638 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
3640 if (shift >= 63) /* detect potential overflow */
3643 return (vd->vdev_asize < (1ULL << shift));
3647 * Get statistics for the given vdev.
3650 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3652 spa_t *spa = vd->vdev_spa;
3653 vdev_t *rvd = spa->spa_root_vdev;
3654 vdev_t *tvd = vd->vdev_top;
3656 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3658 mutex_enter(&vd->vdev_stat_lock);
3659 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3660 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3661 vs->vs_state = vd->vdev_state;
3662 vs->vs_rsize = vdev_get_min_asize(vd);
3663 if (vd->vdev_ops->vdev_op_leaf) {
3664 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3666 * Report intializing progress. Since we don't have the
3667 * initializing locks held, this is only an estimate (although a
3668 * fairly accurate one).
3670 vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done;
3671 vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est;
3672 vs->vs_initialize_state = vd->vdev_initialize_state;
3673 vs->vs_initialize_action_time = vd->vdev_initialize_action_time;
3676 * Report expandable space on top-level, non-auxillary devices only.
3677 * The expandable space is reported in terms of metaslab sized units
3678 * since that determines how much space the pool can expand.
3680 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
3681 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3682 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3684 vs->vs_configured_ashift = vd->vdev_top != NULL
3685 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
3686 vs->vs_logical_ashift = vd->vdev_logical_ashift;
3687 vs->vs_physical_ashift = vd->vdev_physical_ashift;
3688 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3689 vdev_is_concrete(vd)) {
3690 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
3694 * If we're getting stats on the root vdev, aggregate the I/O counts
3695 * over all top-level vdevs (i.e. the direct children of the root).
3698 for (int c = 0; c < rvd->vdev_children; c++) {
3699 vdev_t *cvd = rvd->vdev_child[c];
3700 vdev_stat_t *cvs = &cvd->vdev_stat;
3702 for (int t = 0; t < ZIO_TYPES; t++) {
3703 vs->vs_ops[t] += cvs->vs_ops[t];
3704 vs->vs_bytes[t] += cvs->vs_bytes[t];
3706 cvs->vs_scan_removing = cvd->vdev_removing;
3709 mutex_exit(&vd->vdev_stat_lock);
3713 vdev_clear_stats(vdev_t *vd)
3715 mutex_enter(&vd->vdev_stat_lock);
3716 vd->vdev_stat.vs_space = 0;
3717 vd->vdev_stat.vs_dspace = 0;
3718 vd->vdev_stat.vs_alloc = 0;
3719 mutex_exit(&vd->vdev_stat_lock);
3723 vdev_scan_stat_init(vdev_t *vd)
3725 vdev_stat_t *vs = &vd->vdev_stat;
3727 for (int c = 0; c < vd->vdev_children; c++)
3728 vdev_scan_stat_init(vd->vdev_child[c]);
3730 mutex_enter(&vd->vdev_stat_lock);
3731 vs->vs_scan_processed = 0;
3732 mutex_exit(&vd->vdev_stat_lock);
3736 vdev_stat_update(zio_t *zio, uint64_t psize)
3738 spa_t *spa = zio->io_spa;
3739 vdev_t *rvd = spa->spa_root_vdev;
3740 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3742 uint64_t txg = zio->io_txg;
3743 vdev_stat_t *vs = &vd->vdev_stat;
3744 zio_type_t type = zio->io_type;
3745 int flags = zio->io_flags;
3748 * If this i/o is a gang leader, it didn't do any actual work.
3750 if (zio->io_gang_tree)
3753 if (zio->io_error == 0) {
3755 * If this is a root i/o, don't count it -- we've already
3756 * counted the top-level vdevs, and vdev_get_stats() will
3757 * aggregate them when asked. This reduces contention on
3758 * the root vdev_stat_lock and implicitly handles blocks
3759 * that compress away to holes, for which there is no i/o.
3760 * (Holes never create vdev children, so all the counters
3761 * remain zero, which is what we want.)
3763 * Note: this only applies to successful i/o (io_error == 0)
3764 * because unlike i/o counts, errors are not additive.
3765 * When reading a ditto block, for example, failure of
3766 * one top-level vdev does not imply a root-level error.
3771 ASSERT(vd == zio->io_vd);
3773 if (flags & ZIO_FLAG_IO_BYPASS)
3776 mutex_enter(&vd->vdev_stat_lock);
3778 if (flags & ZIO_FLAG_IO_REPAIR) {
3779 if (flags & ZIO_FLAG_SCAN_THREAD) {
3780 dsl_scan_phys_t *scn_phys =
3781 &spa->spa_dsl_pool->dp_scan->scn_phys;
3782 uint64_t *processed = &scn_phys->scn_processed;
3785 if (vd->vdev_ops->vdev_op_leaf)
3786 atomic_add_64(processed, psize);
3787 vs->vs_scan_processed += psize;
3790 if (flags & ZIO_FLAG_SELF_HEAL)
3791 vs->vs_self_healed += psize;
3795 vs->vs_bytes[type] += psize;
3797 mutex_exit(&vd->vdev_stat_lock);
3801 if (flags & ZIO_FLAG_SPECULATIVE)
3805 * If this is an I/O error that is going to be retried, then ignore the
3806 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3807 * hard errors, when in reality they can happen for any number of
3808 * innocuous reasons (bus resets, MPxIO link failure, etc).
3810 if (zio->io_error == EIO &&
3811 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3815 * Intent logs writes won't propagate their error to the root
3816 * I/O so don't mark these types of failures as pool-level
3819 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3822 mutex_enter(&vd->vdev_stat_lock);
3823 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3824 if (zio->io_error == ECKSUM)
3825 vs->vs_checksum_errors++;
3827 vs->vs_read_errors++;
3829 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3830 vs->vs_write_errors++;
3831 mutex_exit(&vd->vdev_stat_lock);
3833 if (spa->spa_load_state == SPA_LOAD_NONE &&
3834 type == ZIO_TYPE_WRITE && txg != 0 &&
3835 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3836 (flags & ZIO_FLAG_SCAN_THREAD) ||
3837 spa->spa_claiming)) {
3839 * This is either a normal write (not a repair), or it's
3840 * a repair induced by the scrub thread, or it's a repair
3841 * made by zil_claim() during spa_load() in the first txg.
3842 * In the normal case, we commit the DTL change in the same
3843 * txg as the block was born. In the scrub-induced repair
3844 * case, we know that scrubs run in first-pass syncing context,
3845 * so we commit the DTL change in spa_syncing_txg(spa).
3846 * In the zil_claim() case, we commit in spa_first_txg(spa).
3848 * We currently do not make DTL entries for failed spontaneous
3849 * self-healing writes triggered by normal (non-scrubbing)
3850 * reads, because we have no transactional context in which to
3851 * do so -- and it's not clear that it'd be desirable anyway.
3853 if (vd->vdev_ops->vdev_op_leaf) {
3854 uint64_t commit_txg = txg;
3855 if (flags & ZIO_FLAG_SCAN_THREAD) {
3856 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3857 ASSERT(spa_sync_pass(spa) == 1);
3858 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3859 commit_txg = spa_syncing_txg(spa);
3860 } else if (spa->spa_claiming) {
3861 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3862 commit_txg = spa_first_txg(spa);
3864 ASSERT(commit_txg >= spa_syncing_txg(spa));
3865 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3867 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3868 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3869 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3872 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3877 * Update the in-core space usage stats for this vdev, its metaslab class,
3878 * and the root vdev.
3881 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3882 int64_t space_delta)
3884 int64_t dspace_delta = space_delta;
3885 spa_t *spa = vd->vdev_spa;
3886 vdev_t *rvd = spa->spa_root_vdev;
3887 metaslab_group_t *mg = vd->vdev_mg;
3888 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3890 ASSERT(vd == vd->vdev_top);
3893 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3894 * factor. We must calculate this here and not at the root vdev
3895 * because the root vdev's psize-to-asize is simply the max of its
3896 * childrens', thus not accurate enough for us.
3898 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3899 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3900 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3901 vd->vdev_deflate_ratio;
3903 mutex_enter(&vd->vdev_stat_lock);
3904 vd->vdev_stat.vs_alloc += alloc_delta;
3905 vd->vdev_stat.vs_space += space_delta;
3906 vd->vdev_stat.vs_dspace += dspace_delta;
3907 mutex_exit(&vd->vdev_stat_lock);
3909 if (mc == spa_normal_class(spa)) {
3910 mutex_enter(&rvd->vdev_stat_lock);
3911 rvd->vdev_stat.vs_alloc += alloc_delta;
3912 rvd->vdev_stat.vs_space += space_delta;
3913 rvd->vdev_stat.vs_dspace += dspace_delta;
3914 mutex_exit(&rvd->vdev_stat_lock);
3918 ASSERT(rvd == vd->vdev_parent);
3919 ASSERT(vd->vdev_ms_count != 0);
3921 metaslab_class_space_update(mc,
3922 alloc_delta, defer_delta, space_delta, dspace_delta);
3927 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3928 * so that it will be written out next time the vdev configuration is synced.
3929 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3932 vdev_config_dirty(vdev_t *vd)
3934 spa_t *spa = vd->vdev_spa;
3935 vdev_t *rvd = spa->spa_root_vdev;
3938 ASSERT(spa_writeable(spa));
3941 * If this is an aux vdev (as with l2cache and spare devices), then we
3942 * update the vdev config manually and set the sync flag.
3944 if (vd->vdev_aux != NULL) {
3945 spa_aux_vdev_t *sav = vd->vdev_aux;
3949 for (c = 0; c < sav->sav_count; c++) {
3950 if (sav->sav_vdevs[c] == vd)
3954 if (c == sav->sav_count) {
3956 * We're being removed. There's nothing more to do.
3958 ASSERT(sav->sav_sync == B_TRUE);
3962 sav->sav_sync = B_TRUE;
3964 if (nvlist_lookup_nvlist_array(sav->sav_config,
3965 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3966 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3967 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3973 * Setting the nvlist in the middle if the array is a little
3974 * sketchy, but it will work.
3976 nvlist_free(aux[c]);
3977 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3983 * The dirty list is protected by the SCL_CONFIG lock. The caller
3984 * must either hold SCL_CONFIG as writer, or must be the sync thread
3985 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3986 * so this is sufficient to ensure mutual exclusion.
3988 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3989 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3990 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3993 for (c = 0; c < rvd->vdev_children; c++)
3994 vdev_config_dirty(rvd->vdev_child[c]);
3996 ASSERT(vd == vd->vdev_top);
3998 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3999 vdev_is_concrete(vd)) {
4000 list_insert_head(&spa->spa_config_dirty_list, vd);
4006 vdev_config_clean(vdev_t *vd)
4008 spa_t *spa = vd->vdev_spa;
4010 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4011 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4012 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4014 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
4015 list_remove(&spa->spa_config_dirty_list, vd);
4019 * Mark a top-level vdev's state as dirty, so that the next pass of
4020 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4021 * the state changes from larger config changes because they require
4022 * much less locking, and are often needed for administrative actions.
4025 vdev_state_dirty(vdev_t *vd)
4027 spa_t *spa = vd->vdev_spa;
4029 ASSERT(spa_writeable(spa));
4030 ASSERT(vd == vd->vdev_top);
4033 * The state list is protected by the SCL_STATE lock. The caller
4034 * must either hold SCL_STATE as writer, or must be the sync thread
4035 * (which holds SCL_STATE as reader). There's only one sync thread,
4036 * so this is sufficient to ensure mutual exclusion.
4038 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4039 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4040 spa_config_held(spa, SCL_STATE, RW_READER)));
4042 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4043 vdev_is_concrete(vd))
4044 list_insert_head(&spa->spa_state_dirty_list, vd);
4048 vdev_state_clean(vdev_t *vd)
4050 spa_t *spa = vd->vdev_spa;
4052 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4053 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4054 spa_config_held(spa, SCL_STATE, RW_READER)));
4056 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4057 list_remove(&spa->spa_state_dirty_list, vd);
4061 * Propagate vdev state up from children to parent.
4064 vdev_propagate_state(vdev_t *vd)
4066 spa_t *spa = vd->vdev_spa;
4067 vdev_t *rvd = spa->spa_root_vdev;
4068 int degraded = 0, faulted = 0;
4072 if (vd->vdev_children > 0) {
4073 for (int c = 0; c < vd->vdev_children; c++) {
4074 child = vd->vdev_child[c];
4077 * Don't factor holes or indirect vdevs into the
4080 if (!vdev_is_concrete(child))
4083 if (!vdev_readable(child) ||
4084 (!vdev_writeable(child) && spa_writeable(spa))) {
4086 * Root special: if there is a top-level log
4087 * device, treat the root vdev as if it were
4090 if (child->vdev_islog && vd == rvd)
4094 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4098 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4102 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4105 * Root special: if there is a top-level vdev that cannot be
4106 * opened due to corrupted metadata, then propagate the root
4107 * vdev's aux state as 'corrupt' rather than 'insufficient
4110 if (corrupted && vd == rvd &&
4111 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4112 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4113 VDEV_AUX_CORRUPT_DATA);
4116 if (vd->vdev_parent)
4117 vdev_propagate_state(vd->vdev_parent);
4121 * Set a vdev's state. If this is during an open, we don't update the parent
4122 * state, because we're in the process of opening children depth-first.
4123 * Otherwise, we propagate the change to the parent.
4125 * If this routine places a device in a faulted state, an appropriate ereport is
4129 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4131 uint64_t save_state;
4132 spa_t *spa = vd->vdev_spa;
4134 if (state == vd->vdev_state) {
4135 vd->vdev_stat.vs_aux = aux;
4139 save_state = vd->vdev_state;
4141 vd->vdev_state = state;
4142 vd->vdev_stat.vs_aux = aux;
4145 * If we are setting the vdev state to anything but an open state, then
4146 * always close the underlying device unless the device has requested
4147 * a delayed close (i.e. we're about to remove or fault the device).
4148 * Otherwise, we keep accessible but invalid devices open forever.
4149 * We don't call vdev_close() itself, because that implies some extra
4150 * checks (offline, etc) that we don't want here. This is limited to
4151 * leaf devices, because otherwise closing the device will affect other
4154 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4155 vd->vdev_ops->vdev_op_leaf)
4156 vd->vdev_ops->vdev_op_close(vd);
4158 if (vd->vdev_removed &&
4159 state == VDEV_STATE_CANT_OPEN &&
4160 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4162 * If the previous state is set to VDEV_STATE_REMOVED, then this
4163 * device was previously marked removed and someone attempted to
4164 * reopen it. If this failed due to a nonexistent device, then
4165 * keep the device in the REMOVED state. We also let this be if
4166 * it is one of our special test online cases, which is only
4167 * attempting to online the device and shouldn't generate an FMA
4170 vd->vdev_state = VDEV_STATE_REMOVED;
4171 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4172 } else if (state == VDEV_STATE_REMOVED) {
4173 vd->vdev_removed = B_TRUE;
4174 } else if (state == VDEV_STATE_CANT_OPEN) {
4176 * If we fail to open a vdev during an import or recovery, we
4177 * mark it as "not available", which signifies that it was
4178 * never there to begin with. Failure to open such a device
4179 * is not considered an error.
4181 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4182 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4183 vd->vdev_ops->vdev_op_leaf)
4184 vd->vdev_not_present = 1;
4187 * Post the appropriate ereport. If the 'prevstate' field is
4188 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4189 * that this is part of a vdev_reopen(). In this case, we don't
4190 * want to post the ereport if the device was already in the
4191 * CANT_OPEN state beforehand.
4193 * If the 'checkremove' flag is set, then this is an attempt to
4194 * online the device in response to an insertion event. If we
4195 * hit this case, then we have detected an insertion event for a
4196 * faulted or offline device that wasn't in the removed state.
4197 * In this scenario, we don't post an ereport because we are
4198 * about to replace the device, or attempt an online with
4199 * vdev_forcefault, which will generate the fault for us.
4201 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4202 !vd->vdev_not_present && !vd->vdev_checkremove &&
4203 vd != spa->spa_root_vdev) {
4207 case VDEV_AUX_OPEN_FAILED:
4208 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4210 case VDEV_AUX_CORRUPT_DATA:
4211 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4213 case VDEV_AUX_NO_REPLICAS:
4214 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4216 case VDEV_AUX_BAD_GUID_SUM:
4217 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4219 case VDEV_AUX_TOO_SMALL:
4220 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4222 case VDEV_AUX_BAD_LABEL:
4223 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4226 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4229 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
4232 /* Erase any notion of persistent removed state */
4233 vd->vdev_removed = B_FALSE;
4235 vd->vdev_removed = B_FALSE;
4239 * Notify the fmd of the state change. Be verbose and post
4240 * notifications even for stuff that's not important; the fmd agent can
4241 * sort it out. Don't emit state change events for non-leaf vdevs since
4242 * they can't change state on their own. The FMD can check their state
4243 * if it wants to when it sees that a leaf vdev had a state change.
4245 if (vd->vdev_ops->vdev_op_leaf)
4246 zfs_post_state_change(spa, vd);
4248 if (!isopen && vd->vdev_parent)
4249 vdev_propagate_state(vd->vdev_parent);
4253 vdev_children_are_offline(vdev_t *vd)
4255 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4257 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4258 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4266 * Check the vdev configuration to ensure that it's capable of supporting
4267 * a root pool. We do not support partial configuration.
4268 * In addition, only a single top-level vdev is allowed.
4270 * FreeBSD does not have above limitations.
4273 vdev_is_bootable(vdev_t *vd)
4276 if (!vd->vdev_ops->vdev_op_leaf) {
4277 char *vdev_type = vd->vdev_ops->vdev_op_type;
4279 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
4280 vd->vdev_children > 1) {
4282 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4283 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4288 for (int c = 0; c < vd->vdev_children; c++) {
4289 if (!vdev_is_bootable(vd->vdev_child[c]))
4292 #endif /* illumos */
4297 vdev_is_concrete(vdev_t *vd)
4299 vdev_ops_t *ops = vd->vdev_ops;
4300 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4301 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4309 * Determine if a log device has valid content. If the vdev was
4310 * removed or faulted in the MOS config then we know that
4311 * the content on the log device has already been written to the pool.
4314 vdev_log_state_valid(vdev_t *vd)
4316 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4320 for (int c = 0; c < vd->vdev_children; c++)
4321 if (vdev_log_state_valid(vd->vdev_child[c]))
4328 * Expand a vdev if possible.
4331 vdev_expand(vdev_t *vd, uint64_t txg)
4333 ASSERT(vd->vdev_top == vd);
4334 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4335 ASSERT(vdev_is_concrete(vd));
4337 vdev_set_deflate_ratio(vd);
4339 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
4340 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4341 vdev_config_dirty(vd);
4349 vdev_split(vdev_t *vd)
4351 vdev_t *cvd, *pvd = vd->vdev_parent;
4353 vdev_remove_child(pvd, vd);
4354 vdev_compact_children(pvd);
4356 cvd = pvd->vdev_child[0];
4357 if (pvd->vdev_children == 1) {
4358 vdev_remove_parent(cvd);
4359 cvd->vdev_splitting = B_TRUE;
4361 vdev_propagate_state(cvd);
4365 vdev_deadman(vdev_t *vd)
4367 for (int c = 0; c < vd->vdev_children; c++) {
4368 vdev_t *cvd = vd->vdev_child[c];
4373 if (vd->vdev_ops->vdev_op_leaf) {
4374 vdev_queue_t *vq = &vd->vdev_queue;
4376 mutex_enter(&vq->vq_lock);
4377 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4378 spa_t *spa = vd->vdev_spa;
4383 * Look at the head of all the pending queues,
4384 * if any I/O has been outstanding for longer than
4385 * the spa_deadman_synctime we panic the system.
4387 fio = avl_first(&vq->vq_active_tree);
4388 delta = gethrtime() - fio->io_timestamp;
4389 if (delta > spa_deadman_synctime(spa)) {
4390 vdev_dbgmsg(vd, "SLOW IO: zio timestamp "
4391 "%lluns, delta %lluns, last io %lluns",
4392 fio->io_timestamp, (u_longlong_t)delta,
4393 vq->vq_io_complete_ts);
4394 fm_panic("I/O to pool '%s' appears to be "
4395 "hung on vdev guid %llu at '%s'.",
4397 (long long unsigned int) vd->vdev_guid,
4401 mutex_exit(&vq->vq_lock);