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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2011 by Delphix. All rights reserved.
28 * Pool import support functions.
30 * To import a pool, we rely on reading the configuration information from the
31 * ZFS label of each device. If we successfully read the label, then we
32 * organize the configuration information in the following hierarchy:
34 * pool guid -> toplevel vdev guid -> label txg
36 * Duplicate entries matching this same tuple will be discarded. Once we have
37 * examined every device, we pick the best label txg config for each toplevel
38 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
39 * update any paths that have changed. Finally, we attempt to import the pool
40 * using our derived config, and record the results.
55 #include <sys/dktp/fdisk.h>
56 #include <sys/efi_partition.h>
57 #include <thread_pool.h>
59 #include <sys/vdev_impl.h>
62 #include "libzfs_impl.h"
65 * Intermediate structures used to gather configuration information.
67 typedef struct config_entry
{
70 struct config_entry
*ce_next
;
73 typedef struct vdev_entry
{
75 config_entry_t
*ve_configs
;
76 struct vdev_entry
*ve_next
;
79 typedef struct pool_entry
{
81 vdev_entry_t
*pe_vdevs
;
82 struct pool_entry
*pe_next
;
85 typedef struct name_entry
{
88 struct name_entry
*ne_next
;
91 typedef struct pool_list
{
97 get_devid(const char *path
)
103 if ((fd
= open(path
, O_RDONLY
)) < 0)
108 if (devid_get(fd
, &devid
) == 0) {
109 if (devid_get_minor_name(fd
, &minor
) == 0)
110 ret
= devid_str_encode(devid
, minor
);
112 devid_str_free(minor
);
122 * Go through and fix up any path and/or devid information for the given vdev
126 fix_paths(nvlist_t
*nv
, name_entry_t
*names
)
131 name_entry_t
*ne
, *best
;
135 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
136 &child
, &children
) == 0) {
137 for (c
= 0; c
< children
; c
++)
138 if (fix_paths(child
[c
], names
) != 0)
144 * This is a leaf (file or disk) vdev. In either case, go through
145 * the name list and see if we find a matching guid. If so, replace
146 * the path and see if we can calculate a new devid.
148 * There may be multiple names associated with a particular guid, in
149 * which case we have overlapping slices or multiple paths to the same
150 * disk. If this is the case, then we want to pick the path that is
151 * the most similar to the original, where "most similar" is the number
152 * of matching characters starting from the end of the path. This will
153 * preserve slice numbers even if the disks have been reorganized, and
154 * will also catch preferred disk names if multiple paths exist.
156 verify(nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) == 0);
157 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &path
) != 0)
162 for (ne
= names
; ne
!= NULL
; ne
= ne
->ne_next
) {
163 if (ne
->ne_guid
== guid
) {
164 const char *src
, *dst
;
172 src
= ne
->ne_name
+ strlen(ne
->ne_name
) - 1;
173 dst
= path
+ strlen(path
) - 1;
174 for (count
= 0; src
>= ne
->ne_name
&& dst
>= path
;
175 src
--, dst
--, count
++)
180 * At this point, 'count' is the number of characters
181 * matched from the end.
183 if (count
> matched
|| best
== NULL
) {
193 if (nvlist_add_string(nv
, ZPOOL_CONFIG_PATH
, best
->ne_name
) != 0)
196 if ((devid
= get_devid(best
->ne_name
)) == NULL
) {
197 (void) nvlist_remove_all(nv
, ZPOOL_CONFIG_DEVID
);
199 if (nvlist_add_string(nv
, ZPOOL_CONFIG_DEVID
, devid
) != 0)
201 devid_str_free(devid
);
208 * Add the given configuration to the list of known devices.
211 add_config(libzfs_handle_t
*hdl
, pool_list_t
*pl
, const char *path
,
214 uint64_t pool_guid
, vdev_guid
, top_guid
, txg
, state
;
221 * If this is a hot spare not currently in use or level 2 cache
222 * device, add it to the list of names to translate, but don't do
225 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
227 (state
== POOL_STATE_SPARE
|| state
== POOL_STATE_L2CACHE
) &&
228 nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, &vdev_guid
) == 0) {
229 if ((ne
= zfs_alloc(hdl
, sizeof (name_entry_t
))) == NULL
)
232 if ((ne
->ne_name
= zfs_strdup(hdl
, path
)) == NULL
) {
236 ne
->ne_guid
= vdev_guid
;
237 ne
->ne_next
= pl
->names
;
243 * If we have a valid config but cannot read any of these fields, then
244 * it means we have a half-initialized label. In vdev_label_init()
245 * we write a label with txg == 0 so that we can identify the device
246 * in case the user refers to the same disk later on. If we fail to
247 * create the pool, we'll be left with a label in this state
248 * which should not be considered part of a valid pool.
250 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
252 nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
,
254 nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
256 nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
257 &txg
) != 0 || txg
== 0) {
263 * First, see if we know about this pool. If not, then add it to the
264 * list of known pools.
266 for (pe
= pl
->pools
; pe
!= NULL
; pe
= pe
->pe_next
) {
267 if (pe
->pe_guid
== pool_guid
)
272 if ((pe
= zfs_alloc(hdl
, sizeof (pool_entry_t
))) == NULL
) {
276 pe
->pe_guid
= pool_guid
;
277 pe
->pe_next
= pl
->pools
;
282 * Second, see if we know about this toplevel vdev. Add it if its
285 for (ve
= pe
->pe_vdevs
; ve
!= NULL
; ve
= ve
->ve_next
) {
286 if (ve
->ve_guid
== top_guid
)
291 if ((ve
= zfs_alloc(hdl
, sizeof (vdev_entry_t
))) == NULL
) {
295 ve
->ve_guid
= top_guid
;
296 ve
->ve_next
= pe
->pe_vdevs
;
301 * Third, see if we have a config with a matching transaction group. If
302 * so, then we do nothing. Otherwise, add it to the list of known
305 for (ce
= ve
->ve_configs
; ce
!= NULL
; ce
= ce
->ce_next
) {
306 if (ce
->ce_txg
== txg
)
311 if ((ce
= zfs_alloc(hdl
, sizeof (config_entry_t
))) == NULL
) {
316 ce
->ce_config
= config
;
317 ce
->ce_next
= ve
->ve_configs
;
324 * At this point we've successfully added our config to the list of
325 * known configs. The last thing to do is add the vdev guid -> path
326 * mappings so that we can fix up the configuration as necessary before
329 if ((ne
= zfs_alloc(hdl
, sizeof (name_entry_t
))) == NULL
)
332 if ((ne
->ne_name
= zfs_strdup(hdl
, path
)) == NULL
) {
337 ne
->ne_guid
= vdev_guid
;
338 ne
->ne_next
= pl
->names
;
345 * Returns true if the named pool matches the given GUID.
348 pool_active(libzfs_handle_t
*hdl
, const char *name
, uint64_t guid
,
354 if (zpool_open_silent(hdl
, name
, &zhp
) != 0)
362 verify(nvlist_lookup_uint64(zhp
->zpool_config
, ZPOOL_CONFIG_POOL_GUID
,
367 *isactive
= (theguid
== guid
);
372 refresh_config(libzfs_handle_t
*hdl
, nvlist_t
*config
)
375 zfs_cmd_t zc
= { 0 };
378 if (zcmd_write_conf_nvlist(hdl
, &zc
, config
) != 0)
381 if (zcmd_alloc_dst_nvlist(hdl
, &zc
,
382 zc
.zc_nvlist_conf_size
* 2) != 0) {
383 zcmd_free_nvlists(&zc
);
387 while ((err
= ioctl(hdl
->libzfs_fd
, ZFS_IOC_POOL_TRYIMPORT
,
388 &zc
)) != 0 && errno
== ENOMEM
) {
389 if (zcmd_expand_dst_nvlist(hdl
, &zc
) != 0) {
390 zcmd_free_nvlists(&zc
);
396 zcmd_free_nvlists(&zc
);
400 if (zcmd_read_dst_nvlist(hdl
, &zc
, &nvl
) != 0) {
401 zcmd_free_nvlists(&zc
);
405 zcmd_free_nvlists(&zc
);
410 * Determine if the vdev id is a hole in the namespace.
413 vdev_is_hole(uint64_t *hole_array
, uint_t holes
, uint_t id
)
415 for (int c
= 0; c
< holes
; c
++) {
417 /* Top-level is a hole */
418 if (hole_array
[c
] == id
)
425 * Convert our list of pools into the definitive set of configurations. We
426 * start by picking the best config for each toplevel vdev. Once that's done,
427 * we assemble the toplevel vdevs into a full config for the pool. We make a
428 * pass to fix up any incorrect paths, and then add it to the main list to
429 * return to the user.
432 get_configs(libzfs_handle_t
*hdl
, pool_list_t
*pl
, boolean_t active_ok
)
437 nvlist_t
*ret
= NULL
, *config
= NULL
, *tmp
, *nvtop
, *nvroot
;
438 nvlist_t
**spares
, **l2cache
;
439 uint_t i
, nspares
, nl2cache
;
440 boolean_t config_seen
;
442 char *name
, *hostname
, *comment
;
443 uint64_t version
, guid
;
445 nvlist_t
**child
= NULL
;
447 uint64_t *hole_array
, max_id
;
452 boolean_t found_one
= B_FALSE
;
453 boolean_t valid_top_config
= B_FALSE
;
455 if (nvlist_alloc(&ret
, 0, 0) != 0)
458 for (pe
= pl
->pools
; pe
!= NULL
; pe
= pe
->pe_next
) {
459 uint64_t id
, max_txg
= 0;
461 if (nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) != 0)
463 config_seen
= B_FALSE
;
466 * Iterate over all toplevel vdevs. Grab the pool configuration
467 * from the first one we find, and then go through the rest and
468 * add them as necessary to the 'vdevs' member of the config.
470 for (ve
= pe
->pe_vdevs
; ve
!= NULL
; ve
= ve
->ve_next
) {
473 * Determine the best configuration for this vdev by
474 * selecting the config with the latest transaction
478 for (ce
= ve
->ve_configs
; ce
!= NULL
;
481 if (ce
->ce_txg
> best_txg
) {
483 best_txg
= ce
->ce_txg
;
488 * We rely on the fact that the max txg for the
489 * pool will contain the most up-to-date information
490 * about the valid top-levels in the vdev namespace.
492 if (best_txg
> max_txg
) {
493 (void) nvlist_remove(config
,
494 ZPOOL_CONFIG_VDEV_CHILDREN
,
496 (void) nvlist_remove(config
,
497 ZPOOL_CONFIG_HOLE_ARRAY
,
498 DATA_TYPE_UINT64_ARRAY
);
504 valid_top_config
= B_FALSE
;
506 if (nvlist_lookup_uint64(tmp
,
507 ZPOOL_CONFIG_VDEV_CHILDREN
, &max_id
) == 0) {
508 verify(nvlist_add_uint64(config
,
509 ZPOOL_CONFIG_VDEV_CHILDREN
,
511 valid_top_config
= B_TRUE
;
514 if (nvlist_lookup_uint64_array(tmp
,
515 ZPOOL_CONFIG_HOLE_ARRAY
, &hole_array
,
517 verify(nvlist_add_uint64_array(config
,
518 ZPOOL_CONFIG_HOLE_ARRAY
,
519 hole_array
, holes
) == 0);
525 * Copy the relevant pieces of data to the pool
531 * comment (if available)
533 * hostid (if available)
534 * hostname (if available)
538 verify(nvlist_lookup_uint64(tmp
,
539 ZPOOL_CONFIG_VERSION
, &version
) == 0);
540 if (nvlist_add_uint64(config
,
541 ZPOOL_CONFIG_VERSION
, version
) != 0)
543 verify(nvlist_lookup_uint64(tmp
,
544 ZPOOL_CONFIG_POOL_GUID
, &guid
) == 0);
545 if (nvlist_add_uint64(config
,
546 ZPOOL_CONFIG_POOL_GUID
, guid
) != 0)
548 verify(nvlist_lookup_string(tmp
,
549 ZPOOL_CONFIG_POOL_NAME
, &name
) == 0);
550 if (nvlist_add_string(config
,
551 ZPOOL_CONFIG_POOL_NAME
, name
) != 0)
555 * COMMENT is optional, don't bail if it's not
556 * there, instead, set it to NULL.
558 if (nvlist_lookup_string(tmp
,
559 ZPOOL_CONFIG_COMMENT
, &comment
) != 0)
561 else if (nvlist_add_string(config
,
562 ZPOOL_CONFIG_COMMENT
, comment
) != 0)
565 verify(nvlist_lookup_uint64(tmp
,
566 ZPOOL_CONFIG_POOL_STATE
, &state
) == 0);
567 if (nvlist_add_uint64(config
,
568 ZPOOL_CONFIG_POOL_STATE
, state
) != 0)
572 if (nvlist_lookup_uint64(tmp
,
573 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
574 if (nvlist_add_uint64(config
,
575 ZPOOL_CONFIG_HOSTID
, hostid
) != 0)
577 verify(nvlist_lookup_string(tmp
,
578 ZPOOL_CONFIG_HOSTNAME
,
580 if (nvlist_add_string(config
,
581 ZPOOL_CONFIG_HOSTNAME
,
586 config_seen
= B_TRUE
;
590 * Add this top-level vdev to the child array.
592 verify(nvlist_lookup_nvlist(tmp
,
593 ZPOOL_CONFIG_VDEV_TREE
, &nvtop
) == 0);
594 verify(nvlist_lookup_uint64(nvtop
, ZPOOL_CONFIG_ID
,
597 if (id
>= children
) {
600 newchild
= zfs_alloc(hdl
, (id
+ 1) *
601 sizeof (nvlist_t
*));
602 if (newchild
== NULL
)
605 for (c
= 0; c
< children
; c
++)
606 newchild
[c
] = child
[c
];
612 if (nvlist_dup(nvtop
, &child
[id
], 0) != 0)
618 * If we have information about all the top-levels then
619 * clean up the nvlist which we've constructed. This
620 * means removing any extraneous devices that are
621 * beyond the valid range or adding devices to the end
622 * of our array which appear to be missing.
624 if (valid_top_config
) {
625 if (max_id
< children
) {
626 for (c
= max_id
; c
< children
; c
++)
627 nvlist_free(child
[c
]);
629 } else if (max_id
> children
) {
632 newchild
= zfs_alloc(hdl
, (max_id
) *
633 sizeof (nvlist_t
*));
634 if (newchild
== NULL
)
637 for (c
= 0; c
< children
; c
++)
638 newchild
[c
] = child
[c
];
646 verify(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
650 * The vdev namespace may contain holes as a result of
651 * device removal. We must add them back into the vdev
652 * tree before we process any missing devices.
655 ASSERT(valid_top_config
);
657 for (c
= 0; c
< children
; c
++) {
660 if (child
[c
] != NULL
||
661 !vdev_is_hole(hole_array
, holes
, c
))
664 if (nvlist_alloc(&holey
, NV_UNIQUE_NAME
,
669 * Holes in the namespace are treated as
670 * "hole" top-level vdevs and have a
671 * special flag set on them.
673 if (nvlist_add_string(holey
,
675 VDEV_TYPE_HOLE
) != 0 ||
676 nvlist_add_uint64(holey
,
677 ZPOOL_CONFIG_ID
, c
) != 0 ||
678 nvlist_add_uint64(holey
,
679 ZPOOL_CONFIG_GUID
, 0ULL) != 0)
686 * Look for any missing top-level vdevs. If this is the case,
687 * create a faked up 'missing' vdev as a placeholder. We cannot
688 * simply compress the child array, because the kernel performs
689 * certain checks to make sure the vdev IDs match their location
690 * in the configuration.
692 for (c
= 0; c
< children
; c
++) {
693 if (child
[c
] == NULL
) {
695 if (nvlist_alloc(&missing
, NV_UNIQUE_NAME
,
698 if (nvlist_add_string(missing
,
700 VDEV_TYPE_MISSING
) != 0 ||
701 nvlist_add_uint64(missing
,
702 ZPOOL_CONFIG_ID
, c
) != 0 ||
703 nvlist_add_uint64(missing
,
704 ZPOOL_CONFIG_GUID
, 0ULL) != 0) {
705 nvlist_free(missing
);
713 * Put all of this pool's top-level vdevs into a root vdev.
715 if (nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, 0) != 0)
717 if (nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
718 VDEV_TYPE_ROOT
) != 0 ||
719 nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) != 0 ||
720 nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, guid
) != 0 ||
721 nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
722 child
, children
) != 0) {
727 for (c
= 0; c
< children
; c
++)
728 nvlist_free(child
[c
]);
734 * Go through and fix up any paths and/or devids based on our
735 * known list of vdev GUID -> path mappings.
737 if (fix_paths(nvroot
, pl
->names
) != 0) {
743 * Add the root vdev to this pool's configuration.
745 if (nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
753 * zdb uses this path to report on active pools that were
754 * imported or created using -R.
760 * Determine if this pool is currently active, in which case we
761 * can't actually import it.
763 verify(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
765 verify(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
768 if (pool_active(hdl
, name
, guid
, &isactive
) != 0)
777 if ((nvl
= refresh_config(hdl
, config
)) == NULL
) {
787 * Go through and update the paths for spares, now that we have
790 verify(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
792 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
793 &spares
, &nspares
) == 0) {
794 for (i
= 0; i
< nspares
; i
++) {
795 if (fix_paths(spares
[i
], pl
->names
) != 0)
801 * Update the paths for l2cache devices.
803 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
804 &l2cache
, &nl2cache
) == 0) {
805 for (i
= 0; i
< nl2cache
; i
++) {
806 if (fix_paths(l2cache
[i
], pl
->names
) != 0)
812 * Restore the original information read from the actual label.
814 (void) nvlist_remove(config
, ZPOOL_CONFIG_HOSTID
,
816 (void) nvlist_remove(config
, ZPOOL_CONFIG_HOSTNAME
,
819 verify(nvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
,
821 verify(nvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
,
827 * Add this pool to the list of configs.
829 verify(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
831 if (nvlist_add_nvlist(ret
, name
, config
) != 0)
847 (void) no_memory(hdl
);
851 for (c
= 0; c
< children
; c
++)
852 nvlist_free(child
[c
]);
859 * Return the offset of the given label.
862 label_offset(uint64_t size
, int l
)
864 ASSERT(P2PHASE_TYPED(size
, sizeof (vdev_label_t
), uint64_t) == 0);
865 return (l
* sizeof (vdev_label_t
) + (l
< VDEV_LABELS
/ 2 ?
866 0 : size
- VDEV_LABELS
* sizeof (vdev_label_t
)));
870 * Given a file descriptor, read the label information and return an nvlist
871 * describing the configuration, if there is one.
874 zpool_read_label(int fd
, nvlist_t
**config
)
876 struct stat64 statbuf
;
879 uint64_t state
, txg
, size
;
883 if (fstat64(fd
, &statbuf
) == -1)
885 size
= P2ALIGN_TYPED(statbuf
.st_size
, sizeof (vdev_label_t
), uint64_t);
887 if ((label
= malloc(sizeof (vdev_label_t
))) == NULL
)
890 for (l
= 0; l
< VDEV_LABELS
; l
++) {
891 if (pread64(fd
, label
, sizeof (vdev_label_t
),
892 label_offset(size
, l
)) != sizeof (vdev_label_t
))
895 if (nvlist_unpack(label
->vl_vdev_phys
.vp_nvlist
,
896 sizeof (label
->vl_vdev_phys
.vp_nvlist
), config
, 0) != 0)
899 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_STATE
,
900 &state
) != 0 || state
> POOL_STATE_L2CACHE
) {
901 nvlist_free(*config
);
905 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
906 (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_TXG
,
907 &txg
) != 0 || txg
== 0)) {
908 nvlist_free(*config
);
921 typedef struct rdsk_node
{
924 libzfs_handle_t
*rn_hdl
;
928 boolean_t rn_nozpool
;
932 slice_cache_compare(const void *arg1
, const void *arg2
)
934 const char *nm1
= ((rdsk_node_t
*)arg1
)->rn_name
;
935 const char *nm2
= ((rdsk_node_t
*)arg2
)->rn_name
;
936 char *nm1slice
, *nm2slice
;
940 * slices zero and two are the most likely to provide results,
943 nm1slice
= strstr(nm1
, "s0");
944 nm2slice
= strstr(nm2
, "s0");
945 if (nm1slice
&& !nm2slice
) {
948 if (!nm1slice
&& nm2slice
) {
951 nm1slice
= strstr(nm1
, "s2");
952 nm2slice
= strstr(nm2
, "s2");
953 if (nm1slice
&& !nm2slice
) {
956 if (!nm1slice
&& nm2slice
) {
960 rv
= strcmp(nm1
, nm2
);
963 return (rv
> 0 ? 1 : -1);
967 check_one_slice(avl_tree_t
*r
, char *diskname
, uint_t partno
,
968 diskaddr_t size
, uint_t blksz
)
972 char sname
[MAXNAMELEN
];
974 tmpnode
.rn_name
= &sname
[0];
975 (void) snprintf(tmpnode
.rn_name
, MAXNAMELEN
, "%s%u",
978 * protect against division by zero for disk labels that
979 * contain a bogus sector size
983 /* too small to contain a zpool? */
984 if ((size
< (SPA_MINDEVSIZE
/ blksz
)) &&
985 (node
= avl_find(r
, &tmpnode
, NULL
)))
986 node
->rn_nozpool
= B_TRUE
;
990 nozpool_all_slices(avl_tree_t
*r
, const char *sname
)
992 char diskname
[MAXNAMELEN
];
996 (void) strncpy(diskname
, sname
, MAXNAMELEN
);
997 if (((ptr
= strrchr(diskname
, 's')) == NULL
) &&
998 ((ptr
= strrchr(diskname
, 'p')) == NULL
))
1002 for (i
= 0; i
< NDKMAP
; i
++)
1003 check_one_slice(r
, diskname
, i
, 0, 1);
1005 for (i
= 0; i
<= FD_NUMPART
; i
++)
1006 check_one_slice(r
, diskname
, i
, 0, 1);
1010 check_slices(avl_tree_t
*r
, int fd
, const char *sname
)
1012 struct extvtoc vtoc
;
1014 char diskname
[MAXNAMELEN
];
1018 (void) strncpy(diskname
, sname
, MAXNAMELEN
);
1019 if ((ptr
= strrchr(diskname
, 's')) == NULL
|| !isdigit(ptr
[1]))
1023 if (read_extvtoc(fd
, &vtoc
) >= 0) {
1024 for (i
= 0; i
< NDKMAP
; i
++)
1025 check_one_slice(r
, diskname
, i
,
1026 vtoc
.v_part
[i
].p_size
, vtoc
.v_sectorsz
);
1027 } else if (efi_alloc_and_read(fd
, &gpt
) >= 0) {
1029 * on x86 we'll still have leftover links that point
1030 * to slices s[9-15], so use NDKMAP instead
1032 for (i
= 0; i
< NDKMAP
; i
++)
1033 check_one_slice(r
, diskname
, i
,
1034 gpt
->efi_parts
[i
].p_size
, gpt
->efi_lbasize
);
1035 /* nodes p[1-4] are never used with EFI labels */
1037 for (i
= 1; i
<= FD_NUMPART
; i
++)
1038 check_one_slice(r
, diskname
, i
, 0, 1);
1044 zpool_open_func(void *arg
)
1046 rdsk_node_t
*rn
= arg
;
1047 struct stat64 statbuf
;
1053 if ((fd
= openat64(rn
->rn_dfd
, rn
->rn_name
, O_RDONLY
)) < 0) {
1054 /* symlink to a device that's no longer there */
1055 if (errno
== ENOENT
)
1056 nozpool_all_slices(rn
->rn_avl
, rn
->rn_name
);
1060 * Ignore failed stats. We only want regular
1061 * files, character devs and block devs.
1063 if (fstat64(fd
, &statbuf
) != 0 ||
1064 (!S_ISREG(statbuf
.st_mode
) &&
1065 !S_ISCHR(statbuf
.st_mode
) &&
1066 !S_ISBLK(statbuf
.st_mode
))) {
1070 /* this file is too small to hold a zpool */
1071 if (S_ISREG(statbuf
.st_mode
) &&
1072 statbuf
.st_size
< SPA_MINDEVSIZE
) {
1075 } else if (!S_ISREG(statbuf
.st_mode
)) {
1077 * Try to read the disk label first so we don't have to
1078 * open a bunch of minor nodes that can't have a zpool.
1080 check_slices(rn
->rn_avl
, fd
, rn
->rn_name
);
1083 if ((zpool_read_label(fd
, &config
)) != 0) {
1085 (void) no_memory(rn
->rn_hdl
);
1091 rn
->rn_config
= config
;
1092 if (config
!= NULL
) {
1093 assert(rn
->rn_nozpool
== B_FALSE
);
1098 * Given a file descriptor, clear (zero) the label information. This function
1099 * is currently only used in the appliance stack as part of the ZFS sysevent
1103 zpool_clear_label(int fd
)
1105 struct stat64 statbuf
;
1107 vdev_label_t
*label
;
1110 if (fstat64(fd
, &statbuf
) == -1)
1112 size
= P2ALIGN_TYPED(statbuf
.st_size
, sizeof (vdev_label_t
), uint64_t);
1114 if ((label
= calloc(sizeof (vdev_label_t
), 1)) == NULL
)
1117 for (l
= 0; l
< VDEV_LABELS
; l
++) {
1118 if (pwrite64(fd
, label
, sizeof (vdev_label_t
),
1119 label_offset(size
, l
)) != sizeof (vdev_label_t
))
1128 * Given a list of directories to search, find all pools stored on disk. This
1129 * includes partial pools which are not available to import. If no args are
1130 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1131 * poolname or guid (but not both) are provided by the caller when trying
1132 * to import a specific pool.
1135 zpool_find_import_impl(libzfs_handle_t
*hdl
, importargs_t
*iarg
)
1137 int i
, dirs
= iarg
->paths
;
1139 struct dirent64
*dp
;
1140 char path
[MAXPATHLEN
];
1141 char *end
, **dir
= iarg
->path
;
1143 nvlist_t
*ret
= NULL
;
1144 static char *default_dir
= "/dev/dsk";
1145 pool_list_t pools
= { 0 };
1146 pool_entry_t
*pe
, *penext
;
1147 vdev_entry_t
*ve
, *venext
;
1148 config_entry_t
*ce
, *cenext
;
1149 name_entry_t
*ne
, *nenext
;
1150 avl_tree_t slice_cache
;
1160 * Go through and read the label configuration information from every
1161 * possible device, organizing the information according to pool GUID
1162 * and toplevel GUID.
1164 for (i
= 0; i
< dirs
; i
++) {
1169 /* use realpath to normalize the path */
1170 if (realpath(dir
[i
], path
) == 0) {
1171 (void) zfs_error_fmt(hdl
, EZFS_BADPATH
,
1172 dgettext(TEXT_DOMAIN
, "cannot open '%s'"), dir
[i
]);
1175 end
= &path
[strlen(path
)];
1178 pathleft
= &path
[sizeof (path
)] - end
;
1181 * Using raw devices instead of block devices when we're
1182 * reading the labels skips a bunch of slow operations during
1183 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1185 if (strcmp(path
, "/dev/dsk/") == 0)
1186 rdsk
= "/dev/rdsk/";
1190 if ((dfd
= open64(rdsk
, O_RDONLY
)) < 0 ||
1191 (dirp
= fdopendir(dfd
)) == NULL
) {
1192 zfs_error_aux(hdl
, strerror(errno
));
1193 (void) zfs_error_fmt(hdl
, EZFS_BADPATH
,
1194 dgettext(TEXT_DOMAIN
, "cannot open '%s'"),
1199 avl_create(&slice_cache
, slice_cache_compare
,
1200 sizeof (rdsk_node_t
), offsetof(rdsk_node_t
, rn_node
));
1202 * This is not MT-safe, but we have no MT consumers of libzfs
1204 while ((dp
= readdir64(dirp
)) != NULL
) {
1205 const char *name
= dp
->d_name
;
1206 if (name
[0] == '.' &&
1207 (name
[1] == 0 || (name
[1] == '.' && name
[2] == 0)))
1210 slice
= zfs_alloc(hdl
, sizeof (rdsk_node_t
));
1211 slice
->rn_name
= zfs_strdup(hdl
, name
);
1212 slice
->rn_avl
= &slice_cache
;
1213 slice
->rn_dfd
= dfd
;
1214 slice
->rn_hdl
= hdl
;
1215 slice
->rn_nozpool
= B_FALSE
;
1216 avl_add(&slice_cache
, slice
);
1219 * create a thread pool to do all of this in parallel;
1220 * rn_nozpool is not protected, so this is racy in that
1221 * multiple tasks could decide that the same slice can
1222 * not hold a zpool, which is benign. Also choose
1223 * double the number of processors; we hold a lot of
1224 * locks in the kernel, so going beyond this doesn't
1227 t
= tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN
),
1229 for (slice
= avl_first(&slice_cache
); slice
;
1230 (slice
= avl_walk(&slice_cache
, slice
,
1232 (void) tpool_dispatch(t
, zpool_open_func
, slice
);
1237 while ((slice
= avl_destroy_nodes(&slice_cache
,
1238 &cookie
)) != NULL
) {
1239 if (slice
->rn_config
!= NULL
) {
1240 nvlist_t
*config
= slice
->rn_config
;
1241 boolean_t matched
= B_TRUE
;
1243 if (iarg
->poolname
!= NULL
) {
1246 matched
= nvlist_lookup_string(config
,
1247 ZPOOL_CONFIG_POOL_NAME
,
1249 strcmp(iarg
->poolname
, pname
) == 0;
1250 } else if (iarg
->guid
!= 0) {
1253 matched
= nvlist_lookup_uint64(config
,
1254 ZPOOL_CONFIG_POOL_GUID
,
1256 iarg
->guid
== this_guid
;
1259 nvlist_free(config
);
1263 /* use the non-raw path for the config */
1264 (void) strlcpy(end
, slice
->rn_name
, pathleft
);
1265 if (add_config(hdl
, &pools
, path
, config
) != 0)
1268 free(slice
->rn_name
);
1271 avl_destroy(&slice_cache
);
1273 (void) closedir(dirp
);
1277 ret
= get_configs(hdl
, &pools
, iarg
->can_be_active
);
1280 for (pe
= pools
.pools
; pe
!= NULL
; pe
= penext
) {
1281 penext
= pe
->pe_next
;
1282 for (ve
= pe
->pe_vdevs
; ve
!= NULL
; ve
= venext
) {
1283 venext
= ve
->ve_next
;
1284 for (ce
= ve
->ve_configs
; ce
!= NULL
; ce
= cenext
) {
1285 cenext
= ce
->ce_next
;
1287 nvlist_free(ce
->ce_config
);
1295 for (ne
= pools
.names
; ne
!= NULL
; ne
= nenext
) {
1296 nenext
= ne
->ne_next
;
1303 (void) closedir(dirp
);
1309 zpool_find_import(libzfs_handle_t
*hdl
, int argc
, char **argv
)
1311 importargs_t iarg
= { 0 };
1316 return (zpool_find_import_impl(hdl
, &iarg
));
1320 * Given a cache file, return the contents as a list of importable pools.
1321 * poolname or guid (but not both) are provided by the caller when trying
1322 * to import a specific pool.
1325 zpool_find_import_cached(libzfs_handle_t
*hdl
, const char *cachefile
,
1326 char *poolname
, uint64_t guid
)
1330 struct stat64 statbuf
;
1331 nvlist_t
*raw
, *src
, *dst
;
1338 verify(poolname
== NULL
|| guid
== 0);
1340 if ((fd
= open(cachefile
, O_RDONLY
)) < 0) {
1341 zfs_error_aux(hdl
, "%s", strerror(errno
));
1342 (void) zfs_error(hdl
, EZFS_BADCACHE
,
1343 dgettext(TEXT_DOMAIN
, "failed to open cache file"));
1347 if (fstat64(fd
, &statbuf
) != 0) {
1348 zfs_error_aux(hdl
, "%s", strerror(errno
));
1350 (void) zfs_error(hdl
, EZFS_BADCACHE
,
1351 dgettext(TEXT_DOMAIN
, "failed to get size of cache file"));
1355 if ((buf
= zfs_alloc(hdl
, statbuf
.st_size
)) == NULL
) {
1360 if (read(fd
, buf
, statbuf
.st_size
) != statbuf
.st_size
) {
1363 (void) zfs_error(hdl
, EZFS_BADCACHE
,
1364 dgettext(TEXT_DOMAIN
,
1365 "failed to read cache file contents"));
1371 if (nvlist_unpack(buf
, statbuf
.st_size
, &raw
, 0) != 0) {
1373 (void) zfs_error(hdl
, EZFS_BADCACHE
,
1374 dgettext(TEXT_DOMAIN
,
1375 "invalid or corrupt cache file contents"));
1382 * Go through and get the current state of the pools and refresh their
1385 if (nvlist_alloc(&pools
, 0, 0) != 0) {
1386 (void) no_memory(hdl
);
1392 while ((elem
= nvlist_next_nvpair(raw
, elem
)) != NULL
) {
1393 verify(nvpair_value_nvlist(elem
, &src
) == 0);
1395 verify(nvlist_lookup_string(src
, ZPOOL_CONFIG_POOL_NAME
,
1397 if (poolname
!= NULL
&& strcmp(poolname
, name
) != 0)
1400 verify(nvlist_lookup_uint64(src
, ZPOOL_CONFIG_POOL_GUID
,
1403 verify(nvlist_lookup_uint64(src
, ZPOOL_CONFIG_POOL_GUID
,
1405 if (guid
!= this_guid
)
1409 if (pool_active(hdl
, name
, this_guid
, &active
) != 0) {
1418 if ((dst
= refresh_config(hdl
, src
)) == NULL
) {
1424 if (nvlist_add_nvlist(pools
, nvpair_name(elem
), dst
) != 0) {
1425 (void) no_memory(hdl
);
1439 name_or_guid_exists(zpool_handle_t
*zhp
, void *data
)
1441 importargs_t
*import
= data
;
1444 if (import
->poolname
!= NULL
) {
1447 verify(nvlist_lookup_string(zhp
->zpool_config
,
1448 ZPOOL_CONFIG_POOL_NAME
, &pool_name
) == 0);
1449 if (strcmp(pool_name
, import
->poolname
) == 0)
1454 verify(nvlist_lookup_uint64(zhp
->zpool_config
,
1455 ZPOOL_CONFIG_POOL_GUID
, &pool_guid
) == 0);
1456 if (pool_guid
== import
->guid
)
1465 zpool_search_import(libzfs_handle_t
*hdl
, importargs_t
*import
)
1467 verify(import
->poolname
== NULL
|| import
->guid
== 0);
1470 import
->exists
= zpool_iter(hdl
, name_or_guid_exists
, import
);
1472 if (import
->cachefile
!= NULL
)
1473 return (zpool_find_import_cached(hdl
, import
->cachefile
,
1474 import
->poolname
, import
->guid
));
1476 return (zpool_find_import_impl(hdl
, import
));
1480 find_guid(nvlist_t
*nv
, uint64_t guid
)
1486 verify(nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &tmp
) == 0);
1490 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1491 &child
, &children
) == 0) {
1492 for (c
= 0; c
< children
; c
++)
1493 if (find_guid(child
[c
], guid
))
1500 typedef struct aux_cbdata
{
1501 const char *cb_type
;
1503 zpool_handle_t
*cb_zhp
;
1507 find_aux(zpool_handle_t
*zhp
, void *data
)
1509 aux_cbdata_t
*cbp
= data
;
1515 verify(nvlist_lookup_nvlist(zhp
->zpool_config
, ZPOOL_CONFIG_VDEV_TREE
,
1518 if (nvlist_lookup_nvlist_array(nvroot
, cbp
->cb_type
,
1519 &list
, &count
) == 0) {
1520 for (i
= 0; i
< count
; i
++) {
1521 verify(nvlist_lookup_uint64(list
[i
],
1522 ZPOOL_CONFIG_GUID
, &guid
) == 0);
1523 if (guid
== cbp
->cb_guid
) {
1535 * Determines if the pool is in use. If so, it returns true and the state of
1536 * the pool as well as the name of the pool. Both strings are allocated and
1537 * must be freed by the caller.
1540 zpool_in_use(libzfs_handle_t
*hdl
, int fd
, pool_state_t
*state
, char **namestr
,
1546 uint64_t guid
, vdev_guid
;
1547 zpool_handle_t
*zhp
;
1548 nvlist_t
*pool_config
;
1549 uint64_t stateval
, isspare
;
1550 aux_cbdata_t cb
= { 0 };
1555 if (zpool_read_label(fd
, &config
) != 0) {
1556 (void) no_memory(hdl
);
1563 verify(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
1565 verify(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
,
1568 if (stateval
!= POOL_STATE_SPARE
&& stateval
!= POOL_STATE_L2CACHE
) {
1569 verify(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
1571 verify(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
1576 case POOL_STATE_EXPORTED
:
1578 * A pool with an exported state may in fact be imported
1579 * read-only, so check the in-core state to see if it's
1580 * active and imported read-only. If it is, set
1581 * its state to active.
1583 if (pool_active(hdl
, name
, guid
, &isactive
) == 0 && isactive
&&
1584 (zhp
= zpool_open_canfail(hdl
, name
)) != NULL
&&
1585 zpool_get_prop_int(zhp
, ZPOOL_PROP_READONLY
, NULL
))
1586 stateval
= POOL_STATE_ACTIVE
;
1591 case POOL_STATE_ACTIVE
:
1593 * For an active pool, we have to determine if it's really part
1594 * of a currently active pool (in which case the pool will exist
1595 * and the guid will be the same), or whether it's part of an
1596 * active pool that was disconnected without being explicitly
1599 if (pool_active(hdl
, name
, guid
, &isactive
) != 0) {
1600 nvlist_free(config
);
1606 * Because the device may have been removed while
1607 * offlined, we only report it as active if the vdev is
1608 * still present in the config. Otherwise, pretend like
1611 if ((zhp
= zpool_open_canfail(hdl
, name
)) != NULL
&&
1612 (pool_config
= zpool_get_config(zhp
, NULL
))
1616 verify(nvlist_lookup_nvlist(pool_config
,
1617 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
1618 ret
= find_guid(nvroot
, vdev_guid
);
1624 * If this is an active spare within another pool, we
1625 * treat it like an unused hot spare. This allows the
1626 * user to create a pool with a hot spare that currently
1627 * in use within another pool. Since we return B_TRUE,
1628 * libdiskmgt will continue to prevent generic consumers
1629 * from using the device.
1631 if (ret
&& nvlist_lookup_uint64(config
,
1632 ZPOOL_CONFIG_IS_SPARE
, &isspare
) == 0 && isspare
)
1633 stateval
= POOL_STATE_SPARE
;
1638 stateval
= POOL_STATE_POTENTIALLY_ACTIVE
;
1643 case POOL_STATE_SPARE
:
1645 * For a hot spare, it can be either definitively in use, or
1646 * potentially active. To determine if it's in use, we iterate
1647 * over all pools in the system and search for one with a spare
1648 * with a matching guid.
1650 * Due to the shared nature of spares, we don't actually report
1651 * the potentially active case as in use. This means the user
1652 * can freely create pools on the hot spares of exported pools,
1653 * but to do otherwise makes the resulting code complicated, and
1654 * we end up having to deal with this case anyway.
1657 cb
.cb_guid
= vdev_guid
;
1658 cb
.cb_type
= ZPOOL_CONFIG_SPARES
;
1659 if (zpool_iter(hdl
, find_aux
, &cb
) == 1) {
1660 name
= (char *)zpool_get_name(cb
.cb_zhp
);
1667 case POOL_STATE_L2CACHE
:
1670 * Check if any pool is currently using this l2cache device.
1673 cb
.cb_guid
= vdev_guid
;
1674 cb
.cb_type
= ZPOOL_CONFIG_L2CACHE
;
1675 if (zpool_iter(hdl
, find_aux
, &cb
) == 1) {
1676 name
= (char *)zpool_get_name(cb
.cb_zhp
);
1689 if ((*namestr
= zfs_strdup(hdl
, name
)) == NULL
) {
1691 zpool_close(cb
.cb_zhp
);
1692 nvlist_free(config
);
1695 *state
= (pool_state_t
)stateval
;
1699 zpool_close(cb
.cb_zhp
);
1701 nvlist_free(config
);