]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
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
)
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
35 strncpy(fnd
, found
, 32);
37 if (strcmp(found
, required
)==0)
40 int l
= strlen(homehost
);
41 if (l
< 32 && fnd
[l
] == ':' &&
42 strcmp(fnd
+l
+1, required
)==0)
48 static int is_member_busy(char *metadata_version
)
50 /* check if the given member array is active */
51 struct mdstat_ent
*mdstat
= mdstat_read(1, 0);
52 struct mdstat_ent
*ent
;
55 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
56 if (ent
->metadata_version
== NULL
)
58 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
60 if (!is_subarray(&ent
->metadata_version
[9]))
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
73 int Assemble(struct supertype
*st
, char *mddev
,
75 mddev_dev_t devlist
, char *backup_file
,
76 int readonly
, int runstop
,
77 char *update
, char *homehost
, int require_homehost
,
78 int verbose
, int force
)
81 * The task of Assemble is to find a collection of
82 * devices that should (according to their superblocks)
83 * form an array, and to give this collection to the MD driver.
84 * In Linux-2.4 and later, this involves submitting a
85 * SET_ARRAY_INFO ioctl with no arg - to prepare
86 * the array - and then submit a number of
87 * ADD_NEW_DISK ioctls to add disks into
88 * the array. Finally RUN_ARRAY might
89 * be submitted to start the array.
91 * Much of the work of Assemble is in finding and/or
92 * checking the disks to make sure they look right.
94 * If mddev is not set, then scan must be set and we
95 * read through the config file for dev+uuid mapping
96 * We recurse, setting mddev, for each device that
98 * - has a valid uuid (or any uuid if !uuidset)
100 * If mddev is set, we try to determine state of md.
101 * check version - must be at least 0.90.0
102 * check kernel version. must be at least 2.4.
103 * If not, we can possibly fall back on START_ARRAY
104 * Try to GET_ARRAY_INFO.
105 * If possible, give up
106 * If not, try to STOP_ARRAY just to make sure
108 * If !uuidset and scan, look in conf-file for uuid
109 * If not found, give up
110 * If !devlist and scan and uuidset, get list of devs from conf-file
113 * Check superblock - discard if bad
114 * Check uuid (set if we don't have one) - discard if no match
115 * Check superblock similarity if we have a superblock - discard if different
116 * Record events, devicenum
117 * This should give us a list of devices for the array
118 * We should collect the most recent event number
120 * Count disks with recent enough event count
121 * While force && !enough disks
122 * Choose newest rejected disks, update event count
123 * mark clean and rewrite superblock
126 * foreach device with recent events : ADD_NEW_DISK
127 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
129 * Check the device numbers in superblock are right
130 * update superblock if any changes
136 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
137 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
138 && (ident
->container
== NULL
|| ident
->member
== NULL
));
140 int vers
= vers
; /* Keep gcc quite - it really is initialised */
143 int uptodate
; /* set once we decide that this device is as
144 * recent as everything else in the array.
149 int *best
= NULL
; /* indexed by raid_disk */
152 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
153 unsigned int req_cnt
;
159 int report_missmatch
;
161 int start_partial_ok
= (runstop
>= 0) &&
162 (force
|| devlist
==NULL
|| auto_assem
);
163 unsigned int num_devs
;
166 struct mdinfo
*content
= NULL
;
171 char chosen_name
[1024];
173 if (get_linux_version() < 2004000)
177 * If any subdevs are listed, then any that don't
178 * match ident are discarded. Remainder must all match and
180 * If no subdevs, then we scan all devices in the config file, but
181 * there must be something in the identity
185 ident
->uuid_set
== 0 &&
186 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
187 ident
->name
[0] == 0 &&
188 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
189 ident
->devices
== NULL
) {
190 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
191 mddev
? mddev
: "further assembly");
196 devlist
= conf_get_devs();
200 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
202 /* We come back here when doing auto-assembly and attempting some
203 * set of devices failed. Those are now marked as ->used==2 and
204 * we ignore them and try again
207 tmpdev
= devlist
; num_devs
= 0;
213 tmpdev
= tmpdev
->next
;
216 if (!st
&& ident
->st
) st
= ident
->st
;
219 fprintf(stderr
, Name
": looking for devices for %s\n",
220 mddev
? mddev
: "further assembly");
222 /* first walk the list of devices to find a consistent set
223 * that match the criterea, if that is possible.
224 * We flag the ones we like with 'used'.
226 for (tmpdev
= devlist
;
228 tmpdev
= tmpdev
->next
) {
229 char *devname
= tmpdev
->devname
;
232 struct supertype
*tst
= dup_super(st
);
233 struct dev_policy
*pol
= NULL
;
235 if (tmpdev
->used
> 1) continue;
237 if (ident
->devices
&&
238 !match_oneof(ident
->devices
, devname
)) {
239 if (report_missmatch
)
240 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
244 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
246 if (report_missmatch
)
247 fprintf(stderr
, Name
": cannot open device %s: %s\n",
248 devname
, strerror(errno
));
250 } else if (fstat(dfd
, &stb
)< 0) {
252 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
253 devname
, strerror(errno
));
255 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
256 fprintf(stderr
, Name
": %s is not a block device.\n",
259 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
260 if (report_missmatch
)
261 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
264 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
265 if (report_missmatch
)
266 fprintf(stderr
, Name
": no RAID superblock on %s\n",
269 } else if (tst
->ss
->compare_super
== NULL
) {
270 if (report_missmatch
)
271 fprintf(stderr
, Name
": Cannot assemble %s metadata on %s\n",
272 tst
->ss
->name
, devname
);
274 } else if (auto_assem
&& st
== NULL
&&
275 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
276 tst
->ss
->match_home(tst
, homehost
) == 1)) {
277 if (report_missmatch
)
278 fprintf(stderr
, Name
": %s has metadata type %s for which "
279 "auto-assembly is disabled\n",
280 devname
, tst
->ss
->name
);
284 memset(content
, 0, sizeof(*content
));
285 tst
->ss
->getinfo_super(tst
, content
, NULL
);
287 if (dfd
>= 0) close(dfd
);
288 if (tmpdev
->used
== 2) {
290 /* Ignore unrecognised devices during auto-assembly */
292 if (ident
->uuid_set
|| ident
->name
[0] ||
293 ident
->super_minor
!= UnSet
)
294 /* Ignore unrecognised device if looking for
299 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
302 st
->ss
->free_super(st
);
303 dev_policy_free(pol
);
307 if (tst
->ss
->container_content
308 && tst
->loaded_container
) {
309 /* tmpdev is a container. We need to be either
310 * looking for a member, or auto-assembling
313 /* already found some components, this cannot
316 if (report_missmatch
)
317 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
322 if (ident
->container
) {
323 if (ident
->container
[0] == '/' &&
324 !same_dev(ident
->container
, devname
)) {
325 if (report_missmatch
)
326 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
327 devname
, ident
->container
);
330 if (ident
->container
[0] != '/') {
333 if (!parse_uuid(ident
->container
, uuid
) ||
334 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
335 if (report_missmatch
)
336 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
342 /* It is worth looking inside this container.
345 fprintf(stderr
, Name
": looking in container %s\n",
349 content
= tmpdev
->content
;
351 content
= tst
->ss
->container_content(tst
, NULL
);
353 goto loop
; /* empty container */
355 tmpdev
->content
= content
->next
;
356 if (tmpdev
->content
== NULL
)
359 } else if (ident
->container
|| ident
->member
) {
360 /* No chance of this matching if we don't have
362 if (report_missmatch
)
363 fprintf(stderr
, Name
"%s is not a container, and one is required.\n",
368 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
369 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0) {
370 if (report_missmatch
)
371 fprintf(stderr
, Name
": %s has wrong uuid.\n",
375 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
376 name_matches(content
->name
, ident
->name
, homehost
)==0) {
377 if (report_missmatch
)
378 fprintf(stderr
, Name
": %s has wrong name.\n",
382 if (ident
->super_minor
!= UnSet
&&
383 ident
->super_minor
!= content
->array
.md_minor
) {
384 if (report_missmatch
)
385 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
389 if (ident
->level
!= UnSet
&&
390 ident
->level
!= content
->array
.level
) {
391 if (report_missmatch
)
392 fprintf(stderr
, Name
": %s has wrong raid level.\n",
396 if (ident
->raid_disks
!= UnSet
&&
397 ident
->raid_disks
!= content
->array
.raid_disks
) {
398 if (report_missmatch
)
399 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
404 if (tst
->ss
->container_content
405 && tst
->loaded_container
) {
406 /* we have the one container we need, don't keep
407 * looking. If the chosen member is active, skip.
409 if (is_member_busy(content
->text_version
)) {
410 if (report_missmatch
)
411 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
412 content
->text_version
,
417 tst
->ss
->free_super(tst
);
422 dev_policy_free(pol
);
425 if (ident
->member
&& ident
->member
[0]) {
426 char *s
= strchr(content
->text_version
+1, '/');
428 fprintf(stderr
, Name
": badly formatted version: %s\n",
429 content
->text_version
);
432 if (strcmp(ident
->member
, s
+1) != 0) {
433 if (report_missmatch
)
435 Name
": skipping wrong member %s\n",
436 content
->text_version
);
440 st
= tst
; tst
= NULL
;
441 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
442 fprintf(stderr
, Name
": %s is a container, but is not "
443 "only device given: confused and aborting\n",
445 st
->ss
->free_super(st
);
446 dev_policy_free(pol
);
450 fprintf(stderr
, Name
": found match on member %s in %s\n",
451 content
->text_version
, devname
);
456 if (st
->minor_version
== -1)
457 st
->minor_version
= tst
->minor_version
;
458 if (st
->ss
!= tst
->ss
||
459 st
->minor_version
!= tst
->minor_version
||
460 st
->ss
->compare_super(st
, tst
) != 0) {
461 /* Some mismatch. If exactly one array matches this host,
462 * we can resolve on that one.
463 * Or, if we are auto assembling, we just ignore the second
469 int first
= st
->ss
->match_home(st
, homehost
);
470 int last
= tst
->ss
->match_home(tst
, homehost
);
472 (first
== 1 || last
== 1)) {
473 /* We can do something */
474 if (first
) {/* just ignore this one */
475 if (report_missmatch
)
476 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
479 } else { /* reject all those sofar */
481 if (report_missmatch
)
482 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
484 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
492 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
494 tst
->ss
->free_super(tst
);
495 st
->ss
->free_super(st
);
496 dev_policy_free(pol
);
503 dev_policy_free(pol
);
508 tst
->ss
->free_super(tst
);
511 if (!st
|| !st
->sb
|| !content
)
514 /* Now need to open the array device. Use create_mddev */
515 if (content
== &info
)
516 st
->ss
->getinfo_super(st
, content
, NULL
);
518 trustworthy
= FOREIGN
;
519 name
= content
->name
;
520 switch (st
->ss
->match_home(st
, homehost
)
521 ?: st
->ss
->match_home(st
, "any")) {
524 name
= strchr(content
->name
, ':');
528 name
= content
->name
;
532 /* If the array is listed in mdadm.conf or on
533 * command line, then we trust the name
534 * even if the array doesn't look local
539 content
->array
.level
== LEVEL_CONTAINER
) {
540 name
= content
->text_version
;
541 trustworthy
= METADATA
;
544 if (name
[0] && trustworthy
!= LOCAL
&&
545 ! require_homehost
&&
546 conf_name_is_free(name
))
549 if (trustworthy
== LOCAL
&&
551 /* Ignore 'host:' prefix of name */
552 name
= strchr(name
, ':')+1;
554 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
557 st
->ss
->free_super(st
);
563 vers
= md_get_version(mdfd
);
565 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
566 " Upgrade your kernel or try --build\n");
570 if (mddev_busy(fd2devnum(mdfd
))) {
571 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
573 for (tmpdev
= devlist
;
574 tmpdev
&& tmpdev
->used
!= 1;
575 tmpdev
= tmpdev
->next
)
577 if (tmpdev
&& auto_assem
)
578 fprintf(stderr
, Name
": %s needed for %s...\n",
579 mddev
, tmpdev
->devname
);
582 st
->ss
->free_super(st
);
587 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
590 if (content
!= &info
) {
591 /* This is a member of a container. Try starting the array. */
592 return assemble_container_content(st
, mdfd
, content
, runstop
,
593 chosen_name
, verbose
);
596 /* Ok, no bad inconsistancy, we can try updating etc */
598 content
->update_private
= NULL
;
599 devices
= malloc(num_devs
* sizeof(*devices
));
600 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
601 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
602 char *devname
= tmpdev
->devname
;
604 /* looks like a good enough match to update the super block if needed */
608 /* prepare useful information in info structures */
610 struct supertype
*tst
;
614 if (strcmp(update
, "uuid")==0 &&
617 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
618 read(rfd
, ident
->uuid
, 16) != 16) {
619 *(__u32
*)(ident
->uuid
) = random();
620 *(__u32
*)(ident
->uuid
+1) = random();
621 *(__u32
*)(ident
->uuid
+2) = random();
622 *(__u32
*)(ident
->uuid
+3) = random();
624 if (rfd
>= 0) close(rfd
);
626 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
628 remove_partitions(dfd
);
631 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
632 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
641 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
643 memcpy(content
->uuid
, ident
->uuid
, 16);
644 strcpy(content
->name
, ident
->name
);
645 content
->array
.md_minor
= minor(stb2
.st_rdev
);
647 if (strcmp(update
, "byteorder") == 0)
650 err
= tst
->ss
->update_super(tst
, content
, update
,
656 Name
": --update=%s not understood"
657 " for %s metadata\n",
658 update
, tst
->ss
->name
);
659 tst
->ss
->free_super(tst
);
667 if (strcmp(update
, "uuid")==0 &&
670 memcpy(ident
->uuid
, content
->uuid
, 16);
672 if (tst
->ss
->store_super(tst
, dfd
))
673 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
677 if (strcmp(update
, "uuid")==0 &&
678 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
679 if (bitmap_update_uuid(ident
->bitmap_fd
,
681 tst
->ss
->swapuuid
) != 0)
682 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
686 tst
->ss
->free_super(tst
);
690 struct supertype
*tst
= dup_super(st
);
692 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
694 remove_partitions(dfd
);
696 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
697 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
706 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
707 tst
->ss
->free_super(tst
);
714 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
715 devname
, mddev
, content
->disk
.raid_disk
);
716 devices
[devcnt
].devname
= devname
;
717 devices
[devcnt
].uptodate
= 0;
718 devices
[devcnt
].i
= *content
;
719 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
720 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
721 if (most_recent
< devcnt
) {
722 if (devices
[devcnt
].i
.events
723 > devices
[most_recent
].i
.events
)
724 most_recent
= devcnt
;
726 if (content
->array
.level
== LEVEL_MULTIPATH
)
727 /* with multipath, the raid_disk from the superblock is meaningless */
730 i
= devices
[devcnt
].i
.disk
.raid_disk
;
732 if (nextspare
< content
->array
.raid_disks
)
733 nextspare
= content
->array
.raid_disks
;
736 if (i
>= content
->array
.raid_disks
&&
742 int newbestcnt
= i
+10;
743 int *newbest
= malloc(sizeof(int)*newbestcnt
);
745 for (c
=0; c
< newbestcnt
; c
++)
747 newbest
[c
] = best
[c
];
752 bestcnt
= newbestcnt
;
755 devices
[best
[i
]].i
.events
756 == devices
[devcnt
].i
.events
757 && (devices
[best
[i
]].i
.disk
.minor
758 != devices
[devcnt
].i
.disk
.minor
)
760 && content
->array
.level
!= LEVEL_MULTIPATH
) {
761 /* two different devices with identical superblock.
762 * Could be a mis-detection caused by overlapping
763 * partitions. fail-safe.
765 fprintf(stderr
, Name
": WARNING %s and %s appear"
766 " to have very similar superblocks.\n"
767 " If they are really different, "
768 "please --zero the superblock on one\n"
769 " If they are the same or overlap,"
770 " please remove one from %s.\n",
771 devices
[best
[i
]].devname
, devname
,
772 inargv
? "the list" :
773 "the\n DEVICE list in mdadm.conf"
781 || (devices
[best
[i
]].i
.events
782 < devices
[devcnt
].i
.events
))
787 free(content
->update_private
);
788 content
->update_private
= NULL
;
791 fprintf(stderr
, Name
": no devices found for %s\n",
794 st
->ss
->free_super(st
);
801 if (update
&& strcmp(update
, "byteorder")==0)
802 st
->minor_version
= 90;
804 st
->ss
->getinfo_super(st
, content
, NULL
);
805 clean
= content
->array
.state
& 1;
807 /* now we have some devices that might be suitable.
810 avail
= malloc(content
->array
.raid_disks
);
811 memset(avail
, 0, content
->array
.raid_disks
);
815 for (i
=0; i
< bestcnt
; i
++) {
817 int event_margin
= 1; /* always allow a difference of '1'
818 * like the kernel does
821 /* note: we ignore error flags in multipath arrays
822 * as they don't make sense
824 if (content
->array
.level
!= LEVEL_MULTIPATH
)
825 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
826 if (!(devices
[j
].i
.disk
.state
827 & (1<<MD_DISK_FAULTY
)))
831 /* If this devices thinks that 'most_recent' has failed, then
832 * we must reject this device.
834 if (j
!= most_recent
&&
835 content
->array
.raid_disks
> 0 &&
836 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
837 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
839 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
840 devices
[j
].devname
, devices
[most_recent
].devname
);
844 if (devices
[j
].i
.events
+event_margin
>=
845 devices
[most_recent
].i
.events
) {
846 devices
[j
].uptodate
= 1;
847 if (i
< content
->array
.raid_disks
) {
848 if (devices
[j
].i
.recovery_start
== MaxSector
) {
858 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
859 content
->array
.layout
, 1,
861 /* Choose the newest best drive which is
862 * not up-to-date, update the superblock
866 struct supertype
*tst
;
867 unsigned long long current_events
;
869 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
872 !devices
[j
].uptodate
&&
873 devices
[j
].i
.recovery_start
== MaxSector
&&
876 > devices
[chosen_drive
].i
.events
))
879 if (chosen_drive
< 0)
881 current_events
= devices
[chosen_drive
].i
.events
;
884 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
885 devices
[chosen_drive
].devname
,
886 devices
[chosen_drive
].i
.disk
.raid_disk
,
887 (int)(devices
[chosen_drive
].i
.events
),
888 (int)(devices
[most_recent
].i
.events
));
889 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
891 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
892 devices
[chosen_drive
].devname
);
893 devices
[chosen_drive
].i
.events
= 0;
897 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
899 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
900 devices
[chosen_drive
].devname
);
901 devices
[chosen_drive
].i
.events
= 0;
904 content
->events
= devices
[most_recent
].i
.events
;
905 tst
->ss
->update_super(tst
, content
, "force-one",
906 devices
[chosen_drive
].devname
, verbose
,
909 if (tst
->ss
->store_super(tst
, fd
)) {
911 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
912 devices
[chosen_drive
].devname
);
913 devices
[chosen_drive
].i
.events
= 0;
914 tst
->ss
->free_super(tst
);
918 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
919 devices
[chosen_drive
].uptodate
= 1;
920 avail
[chosen_drive
] = 1;
922 tst
->ss
->free_super(tst
);
924 /* If there are any other drives of the same vintage,
925 * add them in as well. We can't lose and we might gain
927 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
930 !devices
[j
].uptodate
&&
931 devices
[j
].i
.events
== current_events
) {
938 /* Now we want to look at the superblock which the kernel will base things on
939 * and compare the devices that we think are working with the devices that the
940 * superblock thinks are working.
941 * If there are differences and --force is given, then update this chosen
945 st
->ss
->free_super(st
);
946 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
952 if (!devices
[j
].uptodate
)
955 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
956 fprintf(stderr
, Name
": Cannot open %s: %s\n",
957 devices
[j
].devname
, strerror(errno
));
962 if (st
->ss
->load_super(st
,fd
, NULL
)) {
964 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
972 if (st
->sb
== NULL
) {
973 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
978 st
->ss
->getinfo_super(st
, content
, NULL
);
980 sysfs_init(content
, mdfd
, 0);
982 for (i
=0; i
<bestcnt
; i
++) {
984 unsigned int desired_state
;
986 if (i
< content
->array
.raid_disks
)
987 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
993 if (!devices
[j
].uptodate
)
996 devices
[j
].i
.disk
.state
= desired_state
;
997 if (!(devices
[j
].i
.array
.state
& 1))
1000 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1001 verbose
, 0, NULL
)) {
1004 fprintf(stderr
, Name
": "
1005 "clearing FAULTY flag for device %d in %s for %s\n",
1006 j
, mddev
, devices
[j
].devname
);
1010 fprintf(stderr
, Name
": "
1011 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1012 i
, mddev
, devices
[j
].devname
);
1016 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1017 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1022 if (force
&& !clean
&&
1023 !enough(content
->array
.level
, content
->array
.raid_disks
,
1024 content
->array
.layout
, clean
,
1026 change
+= st
->ss
->update_super(st
, content
, "force-array",
1027 devices
[chosen_drive
].devname
, verbose
,
1034 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1036 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1037 devices
[chosen_drive
].devname
);
1042 if (st
->ss
->store_super(st
, fd
)) {
1044 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1045 devices
[chosen_drive
].devname
);
1053 /* If we are in the middle of a reshape we may need to restore saved data
1054 * that was moved aside due to the reshape overwriting live data
1055 * The code of doing this lives in Grow.c
1058 if (content
->reshape_active
) {
1060 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1062 fprintf(stderr
, Name
":%s has an active reshape - checking "
1063 "if critical section needs to be restored\n",
1065 for (i
=0; i
<bestcnt
; i
++) {
1068 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1069 if (fdlist
[i
] < 0) {
1070 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1071 devices
[j
].devname
);
1079 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1082 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1085 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1086 if (backup_file
== NULL
)
1087 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1094 /* count number of in-sync devices according to the superblock.
1095 * We must have this number to start the array without -s or -R
1097 req_cnt
= content
->array
.working_disks
;
1099 /* Almost ready to actually *do* something */
1103 /* First, fill in the map, so that udev can find our name
1104 * as soon as we become active.
1106 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1107 content
->uuid
, chosen_name
);
1109 rv
= set_array_info(mdfd
, st
, content
);
1111 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1112 mddev
, strerror(errno
));
1113 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1118 if (ident
->bitmap_fd
>= 0) {
1119 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1120 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1121 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1126 } else if (ident
->bitmap_file
) {
1127 /* From config file */
1128 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1130 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1131 ident
->bitmap_file
);
1132 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1137 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1138 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1140 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1148 /* First, add the raid disks, but add the chosen one last */
1149 for (i
=0; i
<= bestcnt
; i
++) {
1153 if (j
== chosen_drive
)
1158 if (j
>= 0 /* && devices[j].uptodate */) {
1159 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1162 fprintf(stderr
, Name
": failed to add "
1167 if (i
< content
->array
.raid_disks
1172 } else if (verbose
> 0)
1173 fprintf(stderr
, Name
": added %s "
1175 devices
[j
].devname
, mddev
,
1176 devices
[j
].i
.disk
.raid_disk
);
1177 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1178 fprintf(stderr
, Name
": no uptodate device for "
1183 if (content
->array
.level
== LEVEL_CONTAINER
) {
1185 fprintf(stderr
, Name
": Container %s has been "
1186 "assembled with %d drive%s",
1187 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1188 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1189 fprintf(stderr
, " (out of %d)",
1190 content
->array
.raid_disks
);
1191 fprintf(stderr
, "\n");
1193 sysfs_uevent(content
, "change");
1194 wait_for(chosen_name
, mdfd
);
1202 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1203 content
->array
.layout
, clean
, avail
, okcnt
) &&
1204 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1206 /* This array is good-to-go.
1207 * If a reshape is in progress then we might need to
1208 * continue monitoring it. In that case we start
1209 * it read-only and let the grow code make it writable.
1213 if (content
->reshape_active
&&
1214 content
->delta_disks
<= 0)
1215 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1218 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1221 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1222 mddev
, okcnt
, okcnt
==1?"":"s");
1223 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1224 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1226 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1228 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1229 fprintf(stderr
, ".\n");
1231 if (content
->reshape_active
&&
1232 content
->array
.level
>= 4 &&
1233 content
->array
.level
<= 6) {
1234 /* might need to increase the size
1235 * of the stripe cache - default is 256
1237 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1238 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1240 sysfs_set_num(sra
, NULL
,
1241 "stripe_cache_size",
1242 (4 * content
->array
.chunk_size
/ 4096) + 1);
1245 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1246 /* If any devices did not get added
1247 * because the kernel rejected them based
1248 * on event count, try adding them
1249 * again providing the action policy is
1250 * 're-add' or greater. The bitmap
1251 * might allow them to be included, or
1252 * they will become spares.
1254 for (i
= 0; i
<= bestcnt
; i
++) {
1256 if (j
>= 0 && !devices
[j
].uptodate
) {
1257 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1259 rv
= add_disk(mdfd
, st
, content
,
1261 if (rv
== 0 && verbose
>= 0)
1263 Name
": %s has been re-added.\n",
1264 devices
[j
].devname
);
1268 wait_for(mddev
, mdfd
);
1272 /* There is a nasty race with 'mdadm --monitor'.
1273 * If it opens this device before we close it,
1274 * it gets an incomplete open on which IO
1275 * doesn't work and the capacity is
1277 * If we reopen (to check for layered devices)
1278 * before --monitor closes, we loose.
1280 * So: wait upto 1 second for there to be
1281 * a non-zero capacity.
1283 while (usecs
< 1000) {
1284 mdfd
= open(mddev
, O_RDONLY
);
1286 unsigned long long size
;
1287 if (get_dev_size(mdfd
, NULL
, &size
) &&
1299 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1300 mddev
, strerror(errno
));
1302 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1303 content
->array
.layout
, 1, avail
, okcnt
))
1304 fprintf(stderr
, Name
": Not enough devices to "
1305 "start the array.\n");
1306 else if (!enough(content
->array
.level
,
1307 content
->array
.raid_disks
,
1308 content
->array
.layout
, clean
,
1310 fprintf(stderr
, Name
": Not enough devices to "
1311 "start the array while not clean "
1312 "- consider --force.\n");
1315 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1320 if (runstop
== -1) {
1321 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1322 mddev
, okcnt
, okcnt
==1?"":"s");
1323 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1324 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1325 fprintf(stderr
, ", but not started.\n");
1330 if (verbose
>= -1) {
1331 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1333 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1335 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1336 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1337 content
->array
.layout
, 1, avail
, okcnt
))
1338 fprintf(stderr
, " - not enough to start the array.\n");
1339 else if (!enough(content
->array
.level
,
1340 content
->array
.raid_disks
,
1341 content
->array
.layout
, clean
,
1343 fprintf(stderr
, " - not enough to start the "
1344 "array while not clean - consider "
1347 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1348 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1350 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1351 fprintf(stderr
, " (use --run to insist).\n");
1355 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1360 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1361 * been updated to point to the current locations of devices.
1362 * so we can just start the array
1365 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1366 devices
[chosen_drive
].i
.disk
.minor
);
1367 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1368 fprintf(stderr
, Name
": Cannot start array: %s\n",
1379 int assemble_container_content(struct supertype
*st
, int mdfd
,
1380 struct mdinfo
*content
, int runstop
,
1381 char *chosen_name
, int verbose
)
1383 struct mdinfo
*dev
, *sra
;
1384 int working
= 0, preexist
= 0;
1385 struct map_ent
*map
= NULL
;
1387 sysfs_init(content
, mdfd
, 0);
1389 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1390 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1391 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1398 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1399 if (sysfs_add_disk(content
, dev
, 1) == 0)
1401 else if (errno
== EEXIST
)
1405 return 1;/* Nothing new, don't try to start */
1408 map_update(&map
, fd2devnum(mdfd
),
1409 content
->text_version
,
1410 content
->uuid
, chosen_name
);
1413 (working
+ preexist
) >= content
->array
.working_disks
) {
1416 switch(content
->array
.level
) {
1418 case LEVEL_MULTIPATH
:
1420 err
= sysfs_set_str(content
, NULL
, "array_state",
1424 err
= sysfs_set_str(content
, NULL
, "array_state",
1426 /* start mdmon if needed. */
1428 if (!mdmon_running(st
->container_dev
))
1429 start_mdmon(st
->container_dev
);
1430 ping_monitor(devnum2devname(st
->container_dev
));
1435 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1438 fprintf(stderr
, Name
1439 ": array %s now has %d devices",
1440 chosen_name
, working
+ preexist
);
1442 fprintf(stderr
, Name
1443 ": Started %s with %d devices",
1444 chosen_name
, working
+ preexist
);
1446 fprintf(stderr
, " (%d new)", working
);
1447 fprintf(stderr
, "\n");
1450 wait_for(chosen_name
, mdfd
);
1453 /* FIXME should have an O_EXCL and wait for read-auto */
1456 fprintf(stderr
, Name
1457 ": %s assembled with %d devices but "
1459 chosen_name
, working
);