2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
28 static int name_matches(char *found
, char *required
, char *homehost
, int require_homehost
)
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
36 if (strcmp(found
, required
)==0)
38 sep
= strchr(found
, ':');
42 if (strncmp(found
, "any:", 4) == 0 ||
43 (homehost
&& strcmp(homehost
, "any") == 0) ||
45 (homehost
&& strlen(homehost
) == l
&&
46 strncmp(found
, homehost
, l
) == 0)) {
47 /* matching homehost */
48 if (strcmp(sep
+1, required
) == 0)
54 static int is_member_busy(char *metadata_version
)
56 /* check if the given member array is active */
57 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
58 struct mdstat_ent
*ent
;
61 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
62 if (ent
->metadata_version
== NULL
)
64 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
66 if (!is_subarray(&ent
->metadata_version
[9]))
68 /* Skip first char - it can be '/' or '-' */
69 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
79 static int ident_matches(struct mddev_ident
*ident
,
80 struct mdinfo
*content
,
81 struct supertype
*tst
,
82 char *homehost
, int require_homehost
,
83 char *update
, char *devname
)
86 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
87 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0 &&
88 memcmp(content
->uuid
, uuid_zero
, sizeof(int[4])) != 0) {
90 pr_err("%s has wrong uuid.\n", devname
);
93 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
94 name_matches(content
->name
, ident
->name
, homehost
, require_homehost
)==0) {
96 pr_err("%s has wrong name.\n", devname
);
99 if (ident
->super_minor
!= UnSet
&&
100 ident
->super_minor
!= content
->array
.md_minor
) {
102 pr_err("%s has wrong super-minor.\n",
106 if (ident
->level
!= UnSet
&&
107 ident
->level
!= content
->array
.level
) {
109 pr_err("%s has wrong raid level.\n",
113 if (ident
->raid_disks
!= UnSet
&&
114 content
->array
.raid_disks
!= 0 && /* metadata doesn't know how many to expect */
115 ident
->raid_disks
!= content
->array
.raid_disks
) {
117 pr_err("%s requires wrong number of drives.\n",
121 if (ident
->member
&& ident
->member
[0]) {
122 /* content->text_version must match */
123 char *s
= strchr(content
->text_version
+1, '/');
126 pr_err("%s is not a container and one is required.\n",
129 } else if (strcmp(ident
->member
, s
+1) != 0) {
131 pr_err("skipping wrong member %s is %s\n",
132 content
->text_version
, devname
);
139 static int select_devices(struct mddev_dev
*devlist
,
140 struct mddev_ident
*ident
,
141 struct supertype
**stp
,
142 struct mdinfo
**contentp
,
144 int inargv
, int auto_assem
)
146 struct mddev_dev
*tmpdev
;
148 struct supertype
*st
= *stp
;
149 struct mdinfo
*content
= NULL
;
150 int report_mismatch
= ((inargv
&& c
->verbose
>= 0) || c
->verbose
> 0);
151 struct domainlist
*domains
= NULL
;
153 tmpdev
= devlist
; num_devs
= 0;
159 tmpdev
->disposition
= 0;
160 tmpdev
= tmpdev
->next
;
163 /* first walk the list of devices to find a consistent set
164 * that match the criterea, if that is possible.
165 * We flag the ones we like with 'used'.
167 for (tmpdev
= devlist
;
169 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
170 char *devname
= tmpdev
->devname
;
173 struct supertype
*tst
;
174 struct dev_policy
*pol
= NULL
;
175 int found_container
= 0;
177 if (tmpdev
->used
> 1)
180 if (ident
->container
) {
181 if (ident
->container
[0] == '/' &&
182 !same_dev(ident
->container
, devname
)) {
184 pr_err("%s is not the container required (%s)\n",
185 devname
, ident
->container
);
188 } else if (ident
->devices
&&
189 !match_oneof(ident
->devices
, devname
)) {
190 /* Note that we ignore the "device=" identifier if a
191 * "container=" is given. Checking both is unnecessarily
195 pr_err("%s is not one of %s\n", devname
, ident
->devices
);
201 dfd
= dev_open(devname
, O_RDONLY
);
204 pr_err("cannot open device %s: %s\n",
205 devname
, strerror(errno
));
207 } else if (fstat(dfd
, &stb
)< 0) {
209 pr_err("fstat failed for %s: %s\n",
210 devname
, strerror(errno
));
212 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
213 pr_err("%s is not a block device.\n",
216 } else if (must_be_container(dfd
)) {
218 /* already found some components, this cannot
222 pr_err("%s is a container, but we are looking for components\n",
225 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
227 pr_err("not a recognisable container: %s\n",
230 } else if (!tst
->ss
->load_container
231 || tst
->ss
->load_container(tst
, dfd
, NULL
)) {
233 pr_err("no correct container type: %s\n",
236 } else if (auto_assem
&&
237 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
238 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
240 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
241 devname
, tst
->ss
->name
);
246 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
248 pr_err("no recogniseable superblock on %s\n",
251 } else if ((tst
->ignore_hw_compat
= 0),
252 tst
->ss
->load_super(tst
, dfd
,
253 report_mismatch
? devname
: NULL
)) {
255 pr_err("no RAID superblock on %s\n",
258 } else if (tst
->ss
->compare_super
== NULL
) {
260 pr_err("Cannot assemble %s metadata on %s\n",
261 tst
->ss
->name
, devname
);
263 } else if (auto_assem
&& st
== NULL
&&
264 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
265 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
267 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
268 devname
, tst
->ss
->name
);
272 if (dfd
>= 0) close(dfd
);
273 if (tmpdev
->used
== 2) {
274 if (auto_assem
|| !inargv
)
275 /* Ignore unrecognised devices during auto-assembly */
277 if (ident
->uuid_set
|| ident
->name
[0] ||
278 ident
->super_minor
!= UnSet
)
279 /* Ignore unrecognised device if looking for
283 pr_err("%s has no superblock - assembly aborted\n",
286 st
->ss
->free_super(st
);
287 dev_policy_free(pol
);
288 domain_free(domains
);
292 if (found_container
) {
293 /* tmpdev is a container. We need to be either
294 * looking for a member, or auto-assembling
296 /* should be safe to try an exclusive open now, we
297 * have rejected anything that some other mdadm might
300 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
303 pr_err("%s is busy - skipping\n", devname
);
308 if (ident
->container
&& ident
->container
[0] != '/') {
313 tst
->ss
->getinfo_super(tst
, content
, NULL
);
315 if (!parse_uuid(ident
->container
, uuid
) ||
316 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
318 pr_err("%s has wrong UUID to be required container\n",
323 /* It is worth looking inside this container.
326 pr_err("looking in container %s\n",
329 for (content
= tst
->ss
->container_content(tst
, NULL
);
331 content
= content
->next
) {
333 if (!ident_matches(ident
, content
, tst
,
334 c
->homehost
, c
->require_homehost
,
336 report_mismatch
? devname
: NULL
))
337 /* message already printed */;
338 else if (is_member_busy(content
->text_version
)) {
340 pr_err("member %s in %s is already assembled\n",
341 content
->text_version
,
343 } else if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
344 /* do not assemble arrays with unsupported configurations */
345 pr_err("Cannot activate member %s in %s.\n",
346 content
->text_version
,
353 goto loop
; /* empty container */
356 st
= tst
; tst
= NULL
;
357 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
358 pr_err("%s is a container, but is not only device given: confused and aborting\n",
360 st
->ss
->free_super(st
);
361 dev_policy_free(pol
);
362 domain_free(domains
);
366 pr_err("found match on member %s in %s\n",
367 content
->text_version
, devname
);
369 /* make sure we finished the loop */
374 tst
->ss
->getinfo_super(tst
, content
, NULL
);
376 if (!ident_matches(ident
, content
, tst
,
377 c
->homehost
, c
->require_homehost
,
379 report_mismatch
? devname
: NULL
))
383 /* Never auto-assemble things that conflict
384 * with mdadm.conf in some way
386 struct mddev_ident
*match
;
389 match
= conf_match(tst
, content
, devname
,
390 report_mismatch
? c
->verbose
: -1,
392 if (!match
&& rv
== 2)
394 if (match
&& match
->devname
&&
395 strcasecmp(match
->devname
, "<ignore>") == 0) {
397 pr_err("%s is a member of an explicitly ignored array\n",
401 if (match
&& !ident_matches(match
, content
, tst
,
402 c
->homehost
, c
->require_homehost
,
404 report_mismatch
? devname
: NULL
))
405 /* Array exists in mdadm.conf but some
406 * details don't match, so reject it
411 /* should be safe to try an exclusive open now, we
412 * have rejected anything that some other mdadm might
415 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
418 pr_err("%s is busy - skipping\n", devname
);
425 if (st
->minor_version
== -1)
426 st
->minor_version
= tst
->minor_version
;
428 if (memcmp(content
->uuid
, uuid_zero
,
429 sizeof(int[4])) == 0) {
430 /* this is a floating spare. It cannot define
431 * an array unless there are no more arrays of
432 * this type to be found. It can be included
433 * in an array of this type though.
439 if (st
->ss
!= tst
->ss
||
440 st
->minor_version
!= tst
->minor_version
||
441 st
->ss
->compare_super(st
, tst
) != 0) {
442 /* Some mismatch. If exactly one array matches this host,
443 * we can resolve on that one.
444 * Or, if we are auto assembling, we just ignore the second
450 int first
= st
->ss
->match_home(st
, c
->homehost
);
451 int last
= tst
->ss
->match_home(tst
, c
->homehost
);
453 (first
== 1 || last
== 1)) {
454 /* We can do something */
455 if (first
) {/* just ignore this one */
457 pr_err("%s misses out due to wrong homehost\n",
460 } else { /* reject all those sofar */
461 struct mddev_dev
*td
;
463 pr_err("%s overrides previous devices due to good homehost\n",
465 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
473 pr_err("superblock on %s doesn't match others - assembly aborted\n",
475 tst
->ss
->free_super(tst
);
476 st
->ss
->free_super(st
);
477 dev_policy_free(pol
);
478 domain_free(domains
);
484 /* Collect domain information from members only */
485 if (tmpdev
&& tmpdev
->used
== 1) {
487 pol
= devid_policy(stb
.st_rdev
);
488 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
490 dev_policy_free(pol
);
493 tst
->ss
->free_super(tst
);
496 /* Check if we found some imsm spares but no members */
499 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
501 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
502 if (tmpdev
->used
!= 3)
508 /* we need sb from one of the spares */
509 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
511 st
->ss
->load_super(st
, dfd
, NULL
))
518 /* Now reject spares that don't match domains of identified members */
519 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
521 if (tmpdev
->used
!= 3)
523 if (stat(tmpdev
->devname
, &stb
)< 0) {
524 pr_err("fstat failed for %s: %s\n",
525 tmpdev
->devname
, strerror(errno
));
528 struct dev_policy
*pol
= devid_policy(stb
.st_rdev
);
529 int dt
= domain_test(domains
, pol
, NULL
);
530 if (inargv
&& dt
!= 0)
531 /* take this spare as domains match
532 * if there are any */
534 else if (!inargv
&& dt
== 1)
535 /* device wasn't explicitly listed, so need
536 * explicit domain match - which we have */
539 /* if domains don't match mark as unused */
541 dev_policy_free(pol
);
544 domain_free(domains
);
546 if (st
&& st
->sb
&& content
== *contentp
)
547 st
->ss
->getinfo_super(st
, content
, NULL
);
555 int uptodate
; /* set once we decide that this device is as
556 * recent as everything else in the array.
558 int included
; /* set if the device is already in the array
559 * due to a previous '-I'
564 static int load_devices(struct devs
*devices
, char *devmap
,
565 struct mddev_ident
*ident
, struct supertype
**stp
,
566 struct mddev_dev
*devlist
, struct context
*c
,
567 struct mdinfo
*content
,
568 int mdfd
, char *mddev
,
569 int *most_recentp
, int *bestcntp
, int **bestp
,
572 struct mddev_dev
*tmpdev
;
576 int most_recent
= -1;
579 struct supertype
*st
= *stp
;
581 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
582 char *devname
= tmpdev
->devname
;
584 struct supertype
*tst
;
588 if (tmpdev
->used
!= 1)
590 /* looks like a good enough match to update the super block if needed */
592 /* prepare useful information in info structures */
597 if (strcmp(c
->update
, "uuid") == 0 && !ident
->uuid_set
)
598 random_uuid((__u8
*)ident
->uuid
);
600 if (strcmp(c
->update
, "ppl") == 0 &&
601 ident
->bitmap_fd
>= 0) {
602 pr_err("PPL is not compatible with bitmap\n");
606 dfd
= dev_open(devname
,
607 tmpdev
->disposition
== 'I'
608 ? O_RDWR
: (O_RDWR
|O_EXCL
));
611 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
612 pr_err("cannot re-read metadata from %s - aborting\n",
619 tst
->ss
->free_super(tst
);
624 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
626 memcpy(content
->uuid
, ident
->uuid
, 16);
627 strcpy(content
->name
, ident
->name
);
628 content
->array
.md_minor
= minor(stb2
.st_rdev
);
630 if (strcmp(c
->update
, "byteorder") == 0)
632 else if (strcmp(c
->update
, "home-cluster") == 0) {
633 tst
->cluster_name
= c
->homecluster
;
634 err
= tst
->ss
->write_bitmap(tst
, dfd
, NameUpdate
);
635 } else if (strcmp(c
->update
, "nodes") == 0) {
636 tst
->nodes
= c
->nodes
;
637 err
= tst
->ss
->write_bitmap(tst
, dfd
, NodeNumUpdate
);
638 } else if (strcmp(c
->update
, "revert-reshape") == 0 &&
640 err
= tst
->ss
->update_super(tst
, content
,
641 "revert-reshape-nobackup",
646 err
= tst
->ss
->update_super(tst
, content
, c
->update
,
652 pr_err("--update=%s not understood for %s metadata\n",
653 c
->update
, tst
->ss
->name
);
654 tst
->ss
->free_super(tst
);
663 if (strcmp(c
->update
, "uuid")==0 &&
666 memcpy(ident
->uuid
, content
->uuid
, 16);
668 if (tst
->ss
->store_super(tst
, dfd
))
669 pr_err("Could not re-write superblock on %s.\n",
672 if (strcmp(c
->update
, "uuid")==0 &&
673 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
674 if (bitmap_update_uuid(ident
->bitmap_fd
,
676 tst
->ss
->swapuuid
) != 0)
677 pr_err("Could not update uuid on external bitmap.\n");
682 dfd
= dev_open(devname
,
683 tmpdev
->disposition
== 'I'
684 ? O_RDWR
: (O_RDWR
|O_EXCL
));
687 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
688 pr_err("cannot re-read metadata from %s - aborting\n",
695 tst
->ss
->free_super(tst
);
700 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
707 pr_err("%s is identified as a member of %s, slot %d%s.\n",
708 devname
, mddev
, content
->disk
.raid_disk
,
709 (content
->disk
.state
& (1<<MD_DISK_REPLACEMENT
)) ? " replacement":"");
710 devices
[devcnt
].devname
= devname
;
711 devices
[devcnt
].uptodate
= 0;
712 devices
[devcnt
].included
= (tmpdev
->disposition
== 'I');
713 devices
[devcnt
].i
= *content
;
714 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
715 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
717 if (devices
[devcnt
].i
.disk
.state
== 6) {
718 if (most_recent
< 0 ||
719 devices
[devcnt
].i
.events
720 > devices
[most_recent
].i
.events
) {
721 struct supertype
*tmp
= tst
;
724 most_recent
= devcnt
;
727 tst
->ss
->free_super(tst
);
730 if (content
->array
.level
== LEVEL_MULTIPATH
)
731 /* with multipath, the raid_disk from the superblock is meaningless */
734 i
= devices
[devcnt
].i
.disk
.raid_disk
;
735 if (i
+1 == 0 || i
== MD_DISK_ROLE_JOURNAL
) {
736 if (nextspare
< content
->array
.raid_disks
*2)
737 nextspare
= content
->array
.raid_disks
*2;
740 /* i is raid_disk - double it so there is room for
743 if (devices
[devcnt
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))
745 if (i
>= content
->array
.raid_disks
*2 &&
751 int newbestcnt
= i
+10;
752 int *newbest
= xmalloc(sizeof(int)*newbestcnt
);
754 for (c
=0; c
< newbestcnt
; c
++)
756 newbest
[c
] = best
[c
];
761 bestcnt
= newbestcnt
;
764 devices
[best
[i
]].i
.events
765 == devices
[devcnt
].i
.events
766 && (devices
[best
[i
]].i
.disk
.minor
767 != devices
[devcnt
].i
.disk
.minor
)
769 && content
->array
.level
!= LEVEL_MULTIPATH
) {
770 /* two different devices with identical superblock.
771 * Could be a mis-detection caused by overlapping
772 * partitions. fail-safe.
774 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
775 " If they are really different, please --zero the superblock on one\n"
776 " If they are the same or overlap, please remove one from %s.\n",
777 devices
[best
[i
]].devname
, devname
,
778 inargv
? "the list" :
779 "the\n DEVICE list in mdadm.conf"
788 || (devices
[best
[i
]].i
.events
789 < devices
[devcnt
].i
.events
))
794 if (most_recent
>= 0)
795 *most_recentp
= most_recent
;
802 static int force_array(struct mdinfo
*content
,
803 struct devs
*devices
,
804 int *best
, int bestcnt
, char *avail
,
806 struct supertype
*st
,
810 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
811 content
->array
.layout
, 1,
814 (content
->reshape_active
&& content
->delta_disks
> 0 &&
815 !enough(content
->array
.level
, (content
->array
.raid_disks
816 - content
->delta_disks
),
817 content
->new_layout
, 1,
820 /* Choose the newest best drive which is
821 * not up-to-date, update the superblock
825 struct supertype
*tst
;
826 unsigned long long current_events
;
827 int chosen_drive
= -1;
831 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
836 if (devices
[j
].uptodate
)
838 if (devices
[j
].i
.recovery_start
!= MaxSector
) {
840 if (!devices
[j
].i
.reshape_active
||
841 devices
[j
].i
.delta_disks
<= 0)
843 /* When increasing number of devices, an
844 * added device also appears to be
845 * recovering. It is safe to include it
846 * as long as it won't be a source of
848 * For now, just allow for last data
849 * devices in RAID4 or last devices in RAID4/5/6.
851 delta
= devices
[j
].i
.delta_disks
;
852 if (devices
[j
].i
.array
.level
>= 4 &&
853 devices
[j
].i
.array
.level
<= 6 &&
854 i
/2 >= content
->array
.raid_disks
- delta
)
856 else if (devices
[j
].i
.array
.level
== 4 &&
857 i
/2 >= content
->array
.raid_disks
- delta
- 1)
862 if (chosen_drive
< 0 ||
864 > devices
[chosen_drive
].i
.events
)
867 if (chosen_drive
< 0)
869 current_events
= devices
[chosen_drive
].i
.events
;
872 pr_err("forcing event count in %s(%d) from %d upto %d\n",
873 devices
[chosen_drive
].devname
,
874 devices
[chosen_drive
].i
.disk
.raid_disk
,
875 (int)(devices
[chosen_drive
].i
.events
),
876 (int)(devices
[most_recent
].i
.events
));
877 fd
= dev_open(devices
[chosen_drive
].devname
,
878 devices
[chosen_drive
].included
? O_RDWR
881 pr_err("Couldn't open %s for write - not updating\n",
882 devices
[chosen_drive
].devname
);
883 devices
[chosen_drive
].i
.events
= 0;
887 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
889 pr_err("RAID superblock disappeared from %s - not updating.\n",
890 devices
[chosen_drive
].devname
);
891 devices
[chosen_drive
].i
.events
= 0;
894 content
->events
= devices
[most_recent
].i
.events
;
895 tst
->ss
->update_super(tst
, content
, "force-one",
896 devices
[chosen_drive
].devname
, c
->verbose
,
899 if (tst
->ss
->store_super(tst
, fd
)) {
901 pr_err("Could not re-write superblock on %s\n",
902 devices
[chosen_drive
].devname
);
903 devices
[chosen_drive
].i
.events
= 0;
904 tst
->ss
->free_super(tst
);
908 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
909 devices
[chosen_drive
].uptodate
= 1;
910 avail
[chosen_drive
] = 1;
912 tst
->ss
->free_super(tst
);
913 /* If there are any other drives of the same vintage,
914 * add them in as well. We can't lose and we might gain
917 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
921 !devices
[j
].uptodate
&&
922 devices
[j
].i
.recovery_start
== MaxSector
&&
923 devices
[j
].i
.events
== current_events
) {
932 static int start_array(int mdfd
,
934 struct mdinfo
*content
,
935 struct supertype
*st
,
936 struct mddev_ident
*ident
,
937 int *best
, int bestcnt
,
939 struct devs
*devices
,
941 unsigned int sparecnt
,
942 unsigned int rebuilding_cnt
,
943 unsigned int journalcnt
,
945 int clean
, char *avail
,
946 int start_partial_ok
,
953 unsigned int req_cnt
;
955 if (content
->journal_device_required
&& (content
->journal_clean
== 0)) {
957 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
960 pr_err("Journal is missing or stale, starting array read only.\n");
964 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
967 rv
= set_array_info(mdfd
, st
, content
);
969 pr_err("failed to set array info for %s: %s\n",
970 mddev
, strerror(errno
));
973 if (ident
->bitmap_fd
>= 0) {
974 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
975 pr_err("SET_BITMAP_FILE failed.\n");
978 } else if (ident
->bitmap_file
) {
979 /* From config file */
980 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
982 pr_err("Could not open bitmap file %s\n",
986 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
987 pr_err("Failed to set bitmapfile for %s\n", mddev
);
994 /* First, add the raid disks, but add the chosen one last */
995 for (i
=0; i
<= bestcnt
; i
++) {
999 if (j
== chosen_drive
)
1004 if (j
>= 0 && !devices
[j
].included
) {
1005 int dfd
= dev_open(devices
[j
].devname
,
1008 remove_partitions(dfd
);
1011 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1014 pr_err("failed to add %s to %s: %s\n",
1018 if (i
< content
->array
.raid_disks
* 2
1023 } else if (c
->verbose
> 0)
1024 pr_err("added %s to %s as %d%s%s\n",
1025 devices
[j
].devname
, mddev
,
1026 devices
[j
].i
.disk
.raid_disk
,
1027 devices
[j
].uptodate
?"":
1028 " (possibly out of date)",
1029 (devices
[j
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))?" replacement":"");
1030 } else if (j
>= 0) {
1032 pr_err("%s is already in %s as %d\n",
1033 devices
[j
].devname
, mddev
,
1034 devices
[j
].i
.disk
.raid_disk
);
1035 } else if (c
->verbose
> 0 && i
< content
->array
.raid_disks
*2
1037 pr_err("no uptodate device for slot %d of %s\n",
1041 if (content
->array
.level
== LEVEL_CONTAINER
) {
1042 if (c
->verbose
>= 0) {
1043 pr_err("Container %s has been assembled with %d drive%s",
1044 mddev
, okcnt
+sparecnt
+journalcnt
,
1045 okcnt
+sparecnt
+journalcnt
==1?"":"s");
1046 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1047 fprintf(stderr
, " (out of %d)",
1048 content
->array
.raid_disks
);
1049 fprintf(stderr
, "\n");
1052 if (st
->ss
->validate_container
) {
1053 struct mdinfo
*devices_list
;
1054 struct mdinfo
*info_devices
= xmalloc(sizeof(struct mdinfo
)*(okcnt
+sparecnt
));
1056 devices_list
= NULL
;
1057 for (count
= 0; count
< okcnt
+sparecnt
; count
++) {
1058 info_devices
[count
] = devices
[count
].i
;
1059 info_devices
[count
].next
= devices_list
;
1060 devices_list
= &info_devices
[count
];
1062 if (st
->ss
->validate_container(devices_list
))
1063 pr_err("Mismatch detected!\n");
1067 st
->ss
->free_super(st
);
1068 sysfs_uevent(content
, "change");
1069 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1070 /* Was partial, is still partial, so signal an error
1071 * to ensure we don't retry */
1076 /* Get number of in-sync devices according to the superblock.
1077 * We must have this number to start the array without -s or -R
1079 req_cnt
= content
->array
.working_disks
;
1081 if (c
->runstop
== 1 ||
1083 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1084 content
->array
.layout
, clean
, avail
) &&
1085 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1087 /* This array is good-to-go.
1088 * If a reshape is in progress then we might need to
1089 * continue monitoring it. In that case we start
1090 * it read-only and let the grow code make it writable.
1093 if (content
->reshape_active
&&
1094 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1095 content
->delta_disks
<= 0) {
1096 if (!c
->backup_file
) {
1097 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1099 pr_err("Please provided one with \"--backup-file=...\"\n");
1101 strcmp(c
->update
, "revert-reshape") == 0)
1102 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1105 rv
= sysfs_set_str(content
, NULL
,
1106 "array_state", "readonly");
1108 rv
= Grow_continue(mdfd
, st
, content
,
1111 } else if (c
->readonly
&&
1112 sysfs_attribute_available(
1113 content
, NULL
, "array_state")) {
1114 rv
= sysfs_set_str(content
, NULL
,
1115 "array_state", "readonly");
1117 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1118 reopen_mddev(mdfd
); /* drop O_EXCL */
1120 if (c
->verbose
>= 0) {
1121 pr_err("%s has been started with %d drive%s",
1122 mddev
, okcnt
, okcnt
==1?"":"s");
1123 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1124 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1126 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1128 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1129 if (content
->journal_clean
)
1130 fprintf(stderr
, " and %d journal", journalcnt
);
1131 fprintf(stderr
, ".\n");
1133 if (content
->reshape_active
&&
1134 content
->array
.level
>= 4 &&
1135 content
->array
.level
<= 6) {
1136 /* might need to increase the size
1137 * of the stripe cache - default is 256
1139 int chunk_size
= content
->array
.chunk_size
;
1140 if (content
->reshape_active
&&
1141 content
->new_chunk
> chunk_size
)
1142 chunk_size
= content
->new_chunk
;
1143 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1144 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1146 sysfs_set_num(sra
, NULL
,
1147 "stripe_cache_size",
1148 (4 * chunk_size
/ 4096) + 1);
1152 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1153 /* If any devices did not get added
1154 * because the kernel rejected them based
1155 * on event count, try adding them
1156 * again providing the action policy is
1157 * 're-add' or greater. The bitmap
1158 * might allow them to be included, or
1159 * they will become spares.
1161 for (i
= 0; i
< bestcnt
; i
++) {
1163 if (j
>= 0 && !devices
[j
].uptodate
) {
1164 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1166 rv
= add_disk(mdfd
, st
, content
,
1168 if (rv
== 0 && c
->verbose
>= 0)
1169 pr_err("%s has been re-added.\n",
1170 devices
[j
].devname
);
1174 if (content
->array
.level
== 6 &&
1175 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1177 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1179 sysfs_set_str(sra
, NULL
,
1180 "sync_action", "repair");
1185 pr_err("failed to RUN_ARRAY %s: %s\n",
1186 mddev
, strerror(errno
));
1188 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1189 content
->array
.layout
, 1, avail
))
1190 pr_err("Not enough devices to start the array.\n");
1191 else if (!enough(content
->array
.level
,
1192 content
->array
.raid_disks
,
1193 content
->array
.layout
, clean
,
1195 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1199 if (c
->runstop
== -1) {
1200 pr_err("%s assembled from %d drive%s",
1201 mddev
, okcnt
, okcnt
==1?"":"s");
1202 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1203 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1204 fprintf(stderr
, ", but not started.\n");
1207 if (c
->verbose
>= -1) {
1208 pr_err("%s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1210 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1212 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1213 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1214 content
->array
.layout
, 1, avail
))
1215 fprintf(stderr
, " - not enough to start the array.\n");
1216 else if (!enough(content
->array
.level
,
1217 content
->array
.raid_disks
,
1218 content
->array
.layout
, clean
,
1220 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1222 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1223 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1225 fprintf(stderr
, " - need %d to start", req_cnt
);
1226 fprintf(stderr
, " (use --run to insist).\n");
1232 int Assemble(struct supertype
*st
, char *mddev
,
1233 struct mddev_ident
*ident
,
1234 struct mddev_dev
*devlist
,
1238 * The task of Assemble is to find a collection of
1239 * devices that should (according to their superblocks)
1240 * form an array, and to give this collection to the MD driver.
1241 * In Linux-2.4 and later, this involves submitting a
1242 * SET_ARRAY_INFO ioctl with no arg - to prepare
1243 * the array - and then submit a number of
1244 * ADD_NEW_DISK ioctls to add disks into
1245 * the array. Finally RUN_ARRAY might
1246 * be submitted to start the array.
1248 * Much of the work of Assemble is in finding and/or
1249 * checking the disks to make sure they look right.
1251 * If mddev is not set, then scan must be set and we
1252 * read through the config file for dev+uuid mapping
1253 * We recurse, setting mddev, for each device that
1255 * - has a valid uuid (or any uuid if !uuidset)
1257 * If mddev is set, we try to determine state of md.
1258 * check version - must be at least 0.90.0
1259 * check kernel version. must be at least 2.4.
1260 * If not, we can possibly fall back on START_ARRAY
1261 * Try to GET_ARRAY_INFO.
1262 * If possible, give up
1263 * If not, try to STOP_ARRAY just to make sure
1265 * If !uuidset and scan, look in conf-file for uuid
1266 * If not found, give up
1267 * If !devlist and scan and uuidset, get list of devs from conf-file
1270 * Check superblock - discard if bad
1271 * Check uuid (set if we don't have one) - discard if no match
1272 * Check superblock similarity if we have a superblock - discard if different
1273 * Record events, devicenum
1274 * This should give us a list of devices for the array
1275 * We should collect the most recent event number
1277 * Count disks with recent enough event count
1278 * While force && !enough disks
1279 * Choose newest rejected disks, update event count
1280 * mark clean and rewrite superblock
1283 * foreach device with recent events : ADD_NEW_DISK
1284 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1286 * Check the device numbers in superblock are right
1287 * update superblock if any changes
1294 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1295 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
1296 && (ident
->container
== NULL
|| ident
->member
== NULL
));
1297 struct devs
*devices
;
1299 int *best
= NULL
; /* indexed by raid_disk */
1302 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
, journalcnt
;
1303 int journal_clean
= 0;
1306 int most_recent
= 0;
1310 int start_partial_ok
= (c
->runstop
>= 0) &&
1311 (c
->force
|| devlist
==NULL
|| auto_assem
);
1313 struct mddev_dev
*tmpdev
;
1315 struct mdinfo
*content
= NULL
;
1316 struct mdinfo
*pre_exist
= NULL
;
1319 char chosen_name
[1024];
1320 struct map_ent
*map
= NULL
;
1324 * If any subdevs are listed, then any that don't
1325 * match ident are discarded. Remainder must all match and
1327 * If no subdevs, then we scan all devices in the config file, but
1328 * there must be something in the identity
1332 ident
->uuid_set
== 0 &&
1333 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1334 ident
->name
[0] == 0 &&
1335 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1336 ident
->devices
== NULL
) {
1337 pr_err("No identity information available for %s - cannot assemble.\n",
1338 mddev
? mddev
: "further assembly");
1342 if (devlist
== NULL
)
1343 devlist
= conf_get_devs();
1348 /* We come back here when doing auto-assembly and attempting some
1349 * set of devices failed. Those are now marked as ->used==2 and
1350 * we ignore them and try again
1352 if (!st
&& ident
->st
)
1355 pr_err("looking for devices for %s\n",
1356 mddev
? mddev
: "further assembly");
1360 st
->ignore_hw_compat
= 1;
1361 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1362 inargv
, auto_assem
);
1366 if (!st
|| !st
->sb
|| !content
)
1369 /* We have a full set of devices - we now need to find the
1371 * However there is a risk that we are racing with "mdadm -I"
1372 * and the array is already partially assembled - we will have
1373 * rejected any devices already in this address.
1374 * So we take a lock on the map file - to prevent further races -
1375 * and look for the uuid in there. If found and the array is
1376 * active, we abort. If found and the array is not active
1377 * we commit to that md device and add all the contained devices
1378 * to our list. We flag them so that we don't try to re-add,
1379 * but can remove if they turn out to not be wanted.
1382 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1383 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1386 mp
= map_by_uuid(&map
, content
->uuid
);
1389 /* array already exists. */
1390 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1391 if (pre_exist
->array
.level
!= UnSet
) {
1392 pr_err("Found some drive for an array that is already active: %s\n",
1394 pr_err("giving up.\n");
1397 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1398 /* We want to add this device to our list,
1399 * but it could already be there if "mdadm -I"
1400 * started *after* we checked for O_EXCL.
1401 * If we add it to the top of the list
1402 * it will be preferred over later copies.
1404 struct mddev_dev
*newdev
;
1405 char *devname
= map_dev(dv
->disk
.major
,
1410 newdev
= xmalloc(sizeof(*newdev
));
1411 newdev
->devname
= devname
;
1412 newdev
->disposition
= 'I';
1414 newdev
->next
= devlist
;
1418 strcpy(chosen_name
, mp
->path
);
1419 if (c
->verbose
> 0 || mddev
== NULL
||
1420 strcmp(mddev
, chosen_name
) != 0)
1421 pr_err("Merging with already-assembled %s\n",
1423 mdfd
= open_dev_excl(mp
->devnm
);
1425 int trustworthy
= FOREIGN
;
1426 name
= content
->name
;
1427 switch (st
->ss
->match_home(st
, c
->homehost
)
1428 ?: st
->ss
->match_home(st
, "any")) {
1430 trustworthy
= LOCAL
;
1431 name
= strchr(content
->name
, ':');
1435 name
= content
->name
;
1439 /* If the array is listed in mdadm.conf or on
1440 * command line, then we trust the name
1441 * even if the array doesn't look local
1443 trustworthy
= LOCAL
;
1446 content
->array
.level
== LEVEL_CONTAINER
) {
1447 name
= content
->text_version
;
1448 trustworthy
= METADATA
;
1451 if (name
[0] && trustworthy
!= LOCAL
&&
1452 ! c
->require_homehost
&&
1453 conf_name_is_free(name
))
1454 trustworthy
= LOCAL
;
1456 if (trustworthy
== LOCAL
&&
1458 /* Ignore 'host:' prefix of name */
1459 name
= strchr(name
, ':')+1;
1461 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1465 st
->ss
->free_super(st
);
1470 mddev
= chosen_name
;
1471 if (get_linux_version() < 2004000) {
1472 pr_err("Assemble requires Linux 2.4 or later, and\n"
1473 " md driver version 0.90.0 or later.\n"
1474 " Upgrade your kernel or try --build\n");
1478 if (pre_exist
== NULL
) {
1479 if (mddev_busy(fd2devnm(mdfd
))) {
1480 pr_err("%s already active, cannot restart it!\n",
1482 for (tmpdev
= devlist
;
1483 tmpdev
&& tmpdev
->used
!= 1;
1484 tmpdev
= tmpdev
->next
)
1486 if (tmpdev
&& auto_assem
)
1487 pr_err("%s needed for %s...\n",
1488 mddev
, tmpdev
->devname
);
1491 st
->ss
->free_super(st
);
1496 /* just incase it was started but has no content */
1497 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1500 if (content
!= &info
) {
1501 /* This is a member of a container. Try starting the array. */
1503 err
= assemble_container_content(st
, mdfd
, content
, c
,
1509 /* Ok, no bad inconsistancy, we can try updating etc */
1510 devices
= xcalloc(num_devs
, sizeof(*devices
));
1511 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1512 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1513 c
, content
, mdfd
, mddev
,
1514 &most_recent
, &bestcnt
, &best
, inargv
);
1519 pr_err("no devices found for %s\n",
1522 st
->ss
->free_super(st
);
1529 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1530 st
->minor_version
= 90;
1532 st
->ss
->getinfo_super(st
, content
, NULL
);
1533 clean
= content
->array
.state
& 1;
1535 /* now we have some devices that might be suitable.
1538 avail
= xcalloc(content
->array
.raid_disks
, 1);
1544 for (i
=0; i
< bestcnt
; i
++) {
1546 int event_margin
= 1; /* always allow a difference of '1'
1547 * like the kernel does
1549 if (j
< 0) continue;
1550 /* note: we ignore error flags in multipath arrays
1551 * as they don't make sense
1553 if (content
->array
.level
!= LEVEL_MULTIPATH
) {
1554 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
)) {
1555 if (content
->journal_device_required
)
1557 else /* unexpected journal, mark as faulty */
1558 devices
[j
].i
.disk
.state
|= (1<<MD_DISK_FAULTY
);
1559 } else if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1560 if (!(devices
[j
].i
.disk
.state
1561 & (1<<MD_DISK_FAULTY
))) {
1562 devices
[j
].uptodate
= 1;
1568 /* If this device thinks that 'most_recent' has failed, then
1569 * we must reject this device.
1571 if (j
!= most_recent
&& !c
->force
&&
1572 content
->array
.raid_disks
> 0 &&
1573 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1574 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1575 if (c
->verbose
> -1)
1576 pr_err("ignoring %s as it reports %s as failed\n",
1577 devices
[j
].devname
, devices
[most_recent
].devname
);
1581 /* Require event counter to be same as, or just less than,
1582 * most recent. If it is bigger, it must be a stray spare and
1583 * should be ignored.
1585 if (devices
[j
].i
.events
+event_margin
>=
1586 devices
[most_recent
].i
.events
&&
1587 devices
[j
].i
.events
<=
1588 devices
[most_recent
].i
.events
1590 devices
[j
].uptodate
= 1;
1591 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
))
1593 if (i
< content
->array
.raid_disks
* 2) {
1594 if (devices
[j
].i
.recovery_start
== MaxSector
||
1595 (content
->reshape_active
&&
1596 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1604 } else if (devices
[j
].i
.disk
.raid_disk
!= MD_DISK_ROLE_JOURNAL
)
1610 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1611 avail
, most_recent
, st
, c
);
1616 /* Now we want to look at the superblock which the kernel will base things on
1617 * and compare the devices that we think are working with the devices that the
1618 * superblock thinks are working.
1619 * If there are differences and --force is given, then update this chosen
1623 st
->ss
->free_super(st
);
1624 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1630 if (!devices
[j
].uptodate
)
1632 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1635 if ((fd
=dev_open(devices
[j
].devname
,
1636 devices
[j
].included
? O_RDONLY
1637 : (O_RDONLY
|O_EXCL
)))< 0) {
1638 pr_err("Cannot open %s: %s\n",
1639 devices
[j
].devname
, strerror(errno
));
1644 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1646 pr_err("RAID superblock has disappeared from %s\n",
1647 devices
[j
].devname
);
1654 if (st
->sb
== NULL
) {
1655 pr_err("No suitable drives found for %s\n", mddev
);
1660 st
->ss
->getinfo_super(st
, content
, NULL
);
1661 if (sysfs_init(content
, mdfd
, NULL
)) {
1662 pr_err("Unable to initialize sysfs\n");
1668 /* after reload context, store journal_clean in context */
1669 content
->journal_clean
= journal_clean
;
1670 for (i
=0; i
<bestcnt
; i
++) {
1672 unsigned int desired_state
;
1674 if (devices
[j
].i
.disk
.raid_disk
== MD_DISK_ROLE_JOURNAL
)
1675 desired_state
= (1<<MD_DISK_JOURNAL
);
1676 else if (i
>= content
->array
.raid_disks
* 2)
1679 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1681 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1685 if (!devices
[j
].uptodate
)
1688 devices
[j
].i
.disk
.state
= desired_state
;
1689 if (!(devices
[j
].i
.array
.state
& 1))
1692 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1693 c
->verbose
, 0, NULL
)) {
1695 if (c
->verbose
>= 0)
1696 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1697 j
, mddev
, devices
[j
].devname
);
1700 if (c
->verbose
>= -1)
1701 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1702 i
, mddev
, devices
[j
].devname
);
1706 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1707 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1712 if (c
->force
&& !clean
&&
1713 !enough(content
->array
.level
, content
->array
.raid_disks
,
1714 content
->array
.layout
, clean
,
1716 change
+= st
->ss
->update_super(st
, content
, "force-array",
1717 devices
[chosen_drive
].devname
, c
->verbose
,
1725 fd
= dev_open(devices
[chosen_drive
].devname
,
1726 devices
[chosen_drive
].included
?
1727 O_RDWR
: (O_RDWR
|O_EXCL
));
1729 pr_err("Could not open %s for write - cannot Assemble array.\n",
1730 devices
[chosen_drive
].devname
);
1735 if (st
->ss
->store_super(st
, fd
)) {
1737 pr_err("Could not re-write superblock on %s\n",
1738 devices
[chosen_drive
].devname
);
1743 if (c
->verbose
>= 0)
1744 pr_err("Marking array %s as 'clean'\n",
1749 /* If we are in the middle of a reshape we may need to restore saved data
1750 * that was moved aside due to the reshape overwriting live data
1751 * The code of doing this lives in Grow.c
1753 if (content
->reshape_active
&&
1754 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1756 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1758 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1760 if (!c
->backup_file
)
1761 c
->backup_file
= locate_backup(content
->sys_name
);
1762 enable_fds(bestcnt
/2);
1763 for (i
= 0; i
< bestcnt
/2; i
++) {
1766 fdlist
[i
] = dev_open(devices
[j
].devname
,
1768 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1769 if (fdlist
[i
] < 0) {
1770 pr_err("Could not open %s for write - cannot Assemble array.\n",
1771 devices
[j
].devname
);
1779 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1780 err
= st
->ss
->recover_backup(st
, content
);
1782 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1783 c
->backup_file
, c
->verbose
> 0);
1784 if (err
&& c
->invalid_backup
) {
1786 pr_err("continuing without restoring backup\n");
1792 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1796 pr_err("Failed to restore critical section for reshape, sorry.\n");
1797 if (c
->backup_file
== NULL
)
1798 cont_err("Possibly you needed to specify the --backup-file\n");
1805 /* Almost ready to actually *do* something */
1806 /* First, fill in the map, so that udev can find our name
1807 * as soon as we become active.
1809 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1810 content
->array
.major_version
= 1;
1811 content
->array
.minor_version
= 0;
1812 strcpy(content
->text_version
, "1.0");
1815 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1816 content
->uuid
, chosen_name
);
1818 rv
= start_array(mdfd
, mddev
, content
,
1819 st
, ident
, best
, bestcnt
,
1820 chosen_drive
, devices
, okcnt
, sparecnt
,
1821 rebuilding_cnt
, journalcnt
,
1823 clean
, avail
, start_partial_ok
,
1826 if (rv
== 1 && !pre_exist
)
1827 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1831 wait_for(chosen_name
, mdfd
);
1835 /* There is a nasty race with 'mdadm --monitor'.
1836 * If it opens this device before we close it,
1837 * it gets an incomplete open on which IO
1838 * doesn't work and the capacity is
1840 * If we reopen (to check for layered devices)
1841 * before --monitor closes, we loose.
1843 * So: wait upto 1 second for there to be
1844 * a non-zero capacity.
1846 while (usecs
< 1000) {
1847 mdfd
= open(mddev
, O_RDONLY
);
1849 unsigned long long size
;
1850 if (get_dev_size(mdfd
, NULL
, &size
) &&
1862 /* '2' means 'OK, but not started yet' */
1863 return rv
== 2 ? 0 : rv
;
1866 int assemble_container_content(struct supertype
*st
, int mdfd
,
1867 struct mdinfo
*content
, struct context
*c
,
1868 char *chosen_name
, int *result
)
1870 struct mdinfo
*dev
, *sra
, *dev2
;
1871 int working
= 0, preexist
= 0;
1878 if (sysfs_init(content
, mdfd
, NULL
)) {
1879 pr_err("Unable to initialize sysfs\n");
1883 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1884 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1885 if (content
->array
.major_version
== -1 &&
1886 content
->array
.minor_version
== -2 &&
1888 content
->text_version
[0] == '/')
1889 content
->text_version
[0] = '-';
1890 if (sysfs_set_array(content
, 9003) != 0) {
1896 /* There are two types of reshape: container wide or sub-array specific
1897 * Check if metadata requests blocking container wide reshapes
1899 start_reshape
= (content
->reshape_active
&&
1900 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1901 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1903 /* Block subarray here if it is under reshape now
1904 * Do not allow for any changes in this array
1906 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1907 block_subarray(content
);
1909 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1910 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1911 if (dev2
->disk
.major
== dev
->disk
.major
&&
1912 dev2
->disk
.minor
== dev
->disk
.minor
)
1916 /* Don't want this one any more */
1917 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1919 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1923 sysfs_set_str(sra
, dev2
, "state", "remove");
1925 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1926 avail
= xcalloc(content
->array
.raid_disks
, 1);
1927 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1928 if (dev
->disk
.raid_disk
>= 0)
1929 avail
[dev
->disk
.raid_disk
] = 1;
1930 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1931 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1932 content
->reshape_active
)
1936 } else if (errno
== EEXIST
)
1940 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1942 return 1;/* Nothing new, don't try to start */
1944 map_update(NULL
, fd2devnm(mdfd
), content
->text_version
,
1945 content
->uuid
, chosen_name
);
1947 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
1948 st
->ss
->validate_ppl
) {
1949 content
->array
.state
|= 1;
1952 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1957 ret
= st
->ss
->validate_ppl(st
, content
, dev
);
1967 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1969 content
->array
.state
&= ~1;
1970 avail
[dev
->disk
.raid_disk
] = 0;
1974 /* have --force - overwrite the invalid ppl */
1975 devpath
= map_dev(dev
->disk
.major
, dev
->disk
.minor
, 0);
1976 dfd
= dev_open(devpath
, O_RDWR
);
1978 pr_err("Failed to open %s\n", devpath
);
1983 err
= st
->ss
->write_init_ppl(st
, content
, dfd
);
1996 if (enough(content
->array
.level
, content
->array
.raid_disks
,
1997 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
1998 if (c
->export
&& result
)
2000 else if (c
->verbose
>= 0) {
2001 pr_err("%s assembled with %d device%s",
2002 chosen_name
, preexist
+ working
,
2003 preexist
+ working
== 1 ? "":"s");
2005 fprintf(stderr
, " (%d new)", working
);
2006 fprintf(stderr
, " but not started\n");
2013 if (c
->runstop
<= 0 &&
2014 (working
+ preexist
+ expansion
) <
2015 content
->array
.working_disks
) {
2016 if (c
->export
&& result
)
2017 *result
|= INCR_UNSAFE
;
2018 else if (c
->verbose
>= 0) {
2019 pr_err("%s assembled with %d device%s",
2020 chosen_name
, preexist
+ working
,
2021 preexist
+ working
== 1 ? "":"s");
2023 fprintf(stderr
, " (%d new)", working
);
2024 fprintf(stderr
, " but not safe to start\n");
2030 if (start_reshape
) {
2031 int spare
= content
->array
.raid_disks
+ expansion
;
2032 if (restore_backup(st
, content
,
2034 spare
, &c
->backup_file
, c
->verbose
) == 1)
2037 err
= sysfs_set_str(content
, NULL
,
2038 "array_state", "readonly");
2042 if (st
->ss
->external
) {
2043 if (!mdmon_running(st
->container_devnm
))
2044 start_mdmon(st
->container_devnm
);
2045 ping_monitor(st
->container_devnm
);
2046 if (mdmon_running(st
->container_devnm
) &&
2047 st
->update_tail
== NULL
)
2048 st
->update_tail
= &st
->updates
;
2051 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
2052 0, c
->freeze_reshape
);
2053 } else switch(content
->array
.level
) {
2055 case LEVEL_MULTIPATH
:
2057 err
= sysfs_set_str(content
, NULL
, "array_state",
2058 c
->readonly
? "readonly" : "active");
2061 err
= sysfs_set_str(content
, NULL
, "array_state",
2063 /* start mdmon if needed. */
2065 if (!mdmon_running(st
->container_devnm
))
2066 start_mdmon(st
->container_devnm
);
2067 ping_monitor(st
->container_devnm
);
2072 sysfs_set_safemode(content
, content
->safe_mode_delay
);
2074 /* Block subarray here if it is not reshaped now
2075 * It has be blocked a little later to allow mdmon to switch in
2078 if (st
->ss
->external
&& content
->recovery_blocked
&&
2080 block_subarray(content
);
2082 if (c
->export
&& result
) {
2086 *result
|= INCR_YES
;
2087 } else if (c
->verbose
>= 0) {
2089 pr_err("array %s now has %d device%s",
2090 chosen_name
, working
+ preexist
,
2091 working
+ preexist
== 1 ? "":"s");
2093 pr_err("Started %s with %d device%s",
2094 chosen_name
, working
+ preexist
,
2095 working
+ preexist
== 1 ? "":"s");
2097 fprintf(stderr
, " (%d new)", working
);
2099 fprintf(stderr
, " ( + %d for expansion)",
2101 fprintf(stderr
, "\n");
2104 wait_for(chosen_name
, mdfd
);
2106 /* FIXME should have an O_EXCL and wait for read-auto */