]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
dbf78f9420a9ec89b4f5d37d9ae15d4d13e0f503
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 static int ident_matches(struct mddev_ident
*ident
,
74 struct mdinfo
*content
,
75 struct supertype
*tst
,
77 char *update
, char *devname
)
80 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
81 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0) {
83 fprintf(stderr
, Name
": %s has wrong uuid.\n",
87 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
88 name_matches(content
->name
, ident
->name
, homehost
)==0) {
90 fprintf(stderr
, Name
": %s has wrong name.\n",
94 if (ident
->super_minor
!= UnSet
&&
95 ident
->super_minor
!= content
->array
.md_minor
) {
97 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
101 if (ident
->level
!= UnSet
&&
102 ident
->level
!= content
->array
.level
) {
104 fprintf(stderr
, Name
": %s has wrong raid level.\n",
108 if (ident
->raid_disks
!= UnSet
&&
109 ident
->raid_disks
!= content
->array
.raid_disks
) {
111 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
115 if (ident
->member
&& ident
->member
[0]) {
116 /* content->text_version must match */
117 char *s
= strchr(content
->text_version
+1, '/');
120 fprintf(stderr
, Name
": %s is not a container and one is required.\n",
123 } else if (strcmp(ident
->member
, s
+1) != 0) {
125 fprintf(stderr
, Name
": skipping wrong member %s is %s\n",
126 content
->text_version
, devname
);
134 int Assemble(struct supertype
*st
, char *mddev
,
135 struct mddev_ident
*ident
,
136 struct mddev_dev
*devlist
, char *backup_file
,
137 int readonly
, int runstop
,
138 char *update
, char *homehost
, int require_homehost
,
139 int verbose
, int force
)
142 * The task of Assemble is to find a collection of
143 * devices that should (according to their superblocks)
144 * form an array, and to give this collection to the MD driver.
145 * In Linux-2.4 and later, this involves submitting a
146 * SET_ARRAY_INFO ioctl with no arg - to prepare
147 * the array - and then submit a number of
148 * ADD_NEW_DISK ioctls to add disks into
149 * the array. Finally RUN_ARRAY might
150 * be submitted to start the array.
152 * Much of the work of Assemble is in finding and/or
153 * checking the disks to make sure they look right.
155 * If mddev is not set, then scan must be set and we
156 * read through the config file for dev+uuid mapping
157 * We recurse, setting mddev, for each device that
159 * - has a valid uuid (or any uuid if !uuidset)
161 * If mddev is set, we try to determine state of md.
162 * check version - must be at least 0.90.0
163 * check kernel version. must be at least 2.4.
164 * If not, we can possibly fall back on START_ARRAY
165 * Try to GET_ARRAY_INFO.
166 * If possible, give up
167 * If not, try to STOP_ARRAY just to make sure
169 * If !uuidset and scan, look in conf-file for uuid
170 * If not found, give up
171 * If !devlist and scan and uuidset, get list of devs from conf-file
174 * Check superblock - discard if bad
175 * Check uuid (set if we don't have one) - discard if no match
176 * Check superblock similarity if we have a superblock - discard if different
177 * Record events, devicenum
178 * This should give us a list of devices for the array
179 * We should collect the most recent event number
181 * Count disks with recent enough event count
182 * While force && !enough disks
183 * Choose newest rejected disks, update event count
184 * mark clean and rewrite superblock
187 * foreach device with recent events : ADD_NEW_DISK
188 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
190 * Check the device numbers in superblock are right
191 * update superblock if any changes
197 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
198 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
199 && (ident
->container
== NULL
|| ident
->member
== NULL
));
201 int vers
= vers
; /* Keep gcc quite - it really is initialised */
204 int uptodate
; /* set once we decide that this device is as
205 * recent as everything else in the array.
210 int *best
= NULL
; /* indexed by raid_disk */
213 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
214 unsigned int req_cnt
;
220 int report_missmatch
;
222 int start_partial_ok
= (runstop
>= 0) &&
223 (force
|| devlist
==NULL
|| auto_assem
);
224 unsigned int num_devs
;
225 struct mddev_dev
*tmpdev
;
227 struct mdinfo
*content
= NULL
;
232 char chosen_name
[1024];
234 if (get_linux_version() < 2004000)
238 * If any subdevs are listed, then any that don't
239 * match ident are discarded. Remainder must all match and
241 * If no subdevs, then we scan all devices in the config file, but
242 * there must be something in the identity
246 ident
->uuid_set
== 0 &&
247 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
248 ident
->name
[0] == 0 &&
249 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
250 ident
->devices
== NULL
) {
251 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
252 mddev
? mddev
: "further assembly");
257 devlist
= conf_get_devs();
261 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
263 /* We come back here when doing auto-assembly and attempting some
264 * set of devices failed. Those are now marked as ->used==2 and
265 * we ignore them and try again
268 tmpdev
= devlist
; num_devs
= 0;
274 tmpdev
= tmpdev
->next
;
277 if (!st
&& ident
->st
) st
= ident
->st
;
280 fprintf(stderr
, Name
": looking for devices for %s\n",
281 mddev
? mddev
: "further assembly");
283 /* first walk the list of devices to find a consistent set
284 * that match the criterea, if that is possible.
285 * We flag the ones we like with 'used'.
287 for (tmpdev
= devlist
;
289 tmpdev
= tmpdev
->next
) {
290 char *devname
= tmpdev
->devname
;
293 struct supertype
*tst
= dup_super(st
);
294 struct dev_policy
*pol
= NULL
;
296 if (tmpdev
->used
> 1) continue;
298 if (ident
->devices
&&
299 !match_oneof(ident
->devices
, devname
)) {
300 if (report_missmatch
)
301 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
305 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
307 if (report_missmatch
)
308 fprintf(stderr
, Name
": cannot open device %s: %s\n",
309 devname
, strerror(errno
));
311 } else if (fstat(dfd
, &stb
)< 0) {
313 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
314 devname
, strerror(errno
));
316 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
317 fprintf(stderr
, Name
": %s is not a block device.\n",
320 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
321 if (report_missmatch
)
322 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
325 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
326 if (report_missmatch
)
327 fprintf(stderr
, Name
": no RAID superblock on %s\n",
330 } else if (tst
->ss
->compare_super
== NULL
) {
331 if (report_missmatch
)
332 fprintf(stderr
, Name
": Cannot assemble %s metadata on %s\n",
333 tst
->ss
->name
, devname
);
335 } else if (auto_assem
&& st
== NULL
&&
336 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
337 tst
->ss
->match_home(tst
, homehost
) == 1)) {
338 if (report_missmatch
)
339 fprintf(stderr
, Name
": %s has metadata type %s for which "
340 "auto-assembly is disabled\n",
341 devname
, tst
->ss
->name
);
345 memset(content
, 0, sizeof(*content
));
346 tst
->ss
->getinfo_super(tst
, content
, NULL
);
348 if (dfd
>= 0) close(dfd
);
349 if (tmpdev
->used
== 2) {
351 /* Ignore unrecognised devices during auto-assembly */
353 if (ident
->uuid_set
|| ident
->name
[0] ||
354 ident
->super_minor
!= UnSet
)
355 /* Ignore unrecognised device if looking for
360 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
363 st
->ss
->free_super(st
);
364 dev_policy_free(pol
);
368 if (tst
->ss
->container_content
369 && tst
->loaded_container
) {
370 /* tmpdev is a container. We need to be either
371 * looking for a member, or auto-assembling
374 /* already found some components, this cannot
377 if (report_missmatch
)
378 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
383 if (ident
->container
) {
384 if (ident
->container
[0] == '/' &&
385 !same_dev(ident
->container
, devname
)) {
386 if (report_missmatch
)
387 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
388 devname
, ident
->container
);
391 if (ident
->container
[0] != '/') {
394 if (!parse_uuid(ident
->container
, uuid
) ||
395 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
396 if (report_missmatch
)
397 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
403 /* It is worth looking inside this container.
406 fprintf(stderr
, Name
": looking in container %s\n",
409 for (content
= tst
->ss
->container_content(tst
, NULL
);
411 content
= content
->next
) {
413 if (!ident_matches(ident
, content
, tst
,
415 report_missmatch
? devname
: NULL
))
416 /* message already printed */;
417 else if (is_member_busy(content
->text_version
)) {
418 if (report_missmatch
)
419 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
420 content
->text_version
,
427 goto loop
; /* empty container */
430 st
= tst
; tst
= NULL
;
431 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
432 fprintf(stderr
, Name
": %s is a container, but is not "
433 "only device given: confused and aborting\n",
435 st
->ss
->free_super(st
);
436 dev_policy_free(pol
);
440 fprintf(stderr
, Name
": found match on member %s in %s\n",
441 content
->text_version
, devname
);
445 if (!ident_matches(ident
, content
, tst
,
447 report_missmatch
? devname
: NULL
))
452 if (st
->minor_version
== -1)
453 st
->minor_version
= tst
->minor_version
;
454 if (st
->ss
!= tst
->ss
||
455 st
->minor_version
!= tst
->minor_version
||
456 st
->ss
->compare_super(st
, tst
) != 0) {
457 /* Some mismatch. If exactly one array matches this host,
458 * we can resolve on that one.
459 * Or, if we are auto assembling, we just ignore the second
465 int first
= st
->ss
->match_home(st
, homehost
);
466 int last
= tst
->ss
->match_home(tst
, homehost
);
468 (first
== 1 || last
== 1)) {
469 /* We can do something */
470 if (first
) {/* just ignore this one */
471 if (report_missmatch
)
472 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
475 } else { /* reject all those sofar */
476 struct mddev_dev
*td
;
477 if (report_missmatch
)
478 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
480 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
488 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
490 tst
->ss
->free_super(tst
);
491 st
->ss
->free_super(st
);
492 dev_policy_free(pol
);
499 dev_policy_free(pol
);
502 tst
->ss
->free_super(tst
);
505 if (!st
|| !st
->sb
|| !content
)
508 /* Now need to open the array device. Use create_mddev */
509 if (content
== &info
)
510 st
->ss
->getinfo_super(st
, content
, NULL
);
512 trustworthy
= FOREIGN
;
513 name
= content
->name
;
514 switch (st
->ss
->match_home(st
, homehost
)
515 ?: st
->ss
->match_home(st
, "any")) {
518 name
= strchr(content
->name
, ':');
522 name
= content
->name
;
526 /* If the array is listed in mdadm.conf or on
527 * command line, then we trust the name
528 * even if the array doesn't look local
533 content
->array
.level
== LEVEL_CONTAINER
) {
534 name
= content
->text_version
;
535 trustworthy
= METADATA
;
538 if (name
[0] && trustworthy
!= LOCAL
&&
539 ! require_homehost
&&
540 conf_name_is_free(name
))
543 if (trustworthy
== LOCAL
&&
545 /* Ignore 'host:' prefix of name */
546 name
= strchr(name
, ':')+1;
548 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
551 st
->ss
->free_super(st
);
557 vers
= md_get_version(mdfd
);
559 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
560 " Upgrade your kernel or try --build\n");
564 if (mddev_busy(fd2devnum(mdfd
))) {
565 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
567 for (tmpdev
= devlist
;
568 tmpdev
&& tmpdev
->used
!= 1;
569 tmpdev
= tmpdev
->next
)
571 if (tmpdev
&& auto_assem
)
572 fprintf(stderr
, Name
": %s needed for %s...\n",
573 mddev
, tmpdev
->devname
);
576 st
->ss
->free_super(st
);
581 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
584 if (content
!= &info
) {
585 /* This is a member of a container. Try starting the array. */
586 return assemble_container_content(st
, mdfd
, content
, runstop
,
587 chosen_name
, verbose
);
590 /* Ok, no bad inconsistancy, we can try updating etc */
592 content
->update_private
= NULL
;
593 devices
= malloc(num_devs
* sizeof(*devices
));
594 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
595 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
596 char *devname
= tmpdev
->devname
;
598 /* looks like a good enough match to update the super block if needed */
602 /* prepare useful information in info structures */
604 struct supertype
*tst
;
608 if (strcmp(update
, "uuid")==0 &&
611 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
612 read(rfd
, ident
->uuid
, 16) != 16) {
613 *(__u32
*)(ident
->uuid
) = random();
614 *(__u32
*)(ident
->uuid
+1) = random();
615 *(__u32
*)(ident
->uuid
+2) = random();
616 *(__u32
*)(ident
->uuid
+3) = random();
618 if (rfd
>= 0) close(rfd
);
620 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
622 remove_partitions(dfd
);
625 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
626 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
635 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
637 memcpy(content
->uuid
, ident
->uuid
, 16);
638 strcpy(content
->name
, ident
->name
);
639 content
->array
.md_minor
= minor(stb2
.st_rdev
);
641 if (strcmp(update
, "byteorder") == 0)
644 err
= tst
->ss
->update_super(tst
, content
, update
,
650 Name
": --update=%s not understood"
651 " for %s metadata\n",
652 update
, tst
->ss
->name
);
653 tst
->ss
->free_super(tst
);
661 if (strcmp(update
, "uuid")==0 &&
664 memcpy(ident
->uuid
, content
->uuid
, 16);
666 if (tst
->ss
->store_super(tst
, dfd
))
667 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
671 if (strcmp(update
, "uuid")==0 &&
672 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
673 if (bitmap_update_uuid(ident
->bitmap_fd
,
675 tst
->ss
->swapuuid
) != 0)
676 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
680 tst
->ss
->free_super(tst
);
684 struct supertype
*tst
= dup_super(st
);
686 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
688 remove_partitions(dfd
);
690 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
691 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
700 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
701 tst
->ss
->free_super(tst
);
708 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
709 devname
, mddev
, content
->disk
.raid_disk
);
710 devices
[devcnt
].devname
= devname
;
711 devices
[devcnt
].uptodate
= 0;
712 devices
[devcnt
].i
= *content
;
713 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
714 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
715 if (most_recent
< devcnt
) {
716 if (devices
[devcnt
].i
.events
717 > devices
[most_recent
].i
.events
)
718 most_recent
= devcnt
;
720 if (content
->array
.level
== LEVEL_MULTIPATH
)
721 /* with multipath, the raid_disk from the superblock is meaningless */
724 i
= devices
[devcnt
].i
.disk
.raid_disk
;
726 if (nextspare
< content
->array
.raid_disks
)
727 nextspare
= content
->array
.raid_disks
;
730 if (i
>= content
->array
.raid_disks
&&
736 int newbestcnt
= i
+10;
737 int *newbest
= malloc(sizeof(int)*newbestcnt
);
739 for (c
=0; c
< newbestcnt
; c
++)
741 newbest
[c
] = best
[c
];
746 bestcnt
= newbestcnt
;
749 devices
[best
[i
]].i
.events
750 == devices
[devcnt
].i
.events
751 && (devices
[best
[i
]].i
.disk
.minor
752 != devices
[devcnt
].i
.disk
.minor
)
754 && content
->array
.level
!= LEVEL_MULTIPATH
) {
755 /* two different devices with identical superblock.
756 * Could be a mis-detection caused by overlapping
757 * partitions. fail-safe.
759 fprintf(stderr
, Name
": WARNING %s and %s appear"
760 " to have very similar superblocks.\n"
761 " If they are really different, "
762 "please --zero the superblock on one\n"
763 " If they are the same or overlap,"
764 " please remove one from %s.\n",
765 devices
[best
[i
]].devname
, devname
,
766 inargv
? "the list" :
767 "the\n DEVICE list in mdadm.conf"
775 || (devices
[best
[i
]].i
.events
776 < devices
[devcnt
].i
.events
))
781 free(content
->update_private
);
782 content
->update_private
= NULL
;
785 fprintf(stderr
, Name
": no devices found for %s\n",
788 st
->ss
->free_super(st
);
795 if (update
&& strcmp(update
, "byteorder")==0)
796 st
->minor_version
= 90;
798 st
->ss
->getinfo_super(st
, content
, NULL
);
799 clean
= content
->array
.state
& 1;
801 /* now we have some devices that might be suitable.
804 avail
= malloc(content
->array
.raid_disks
);
805 memset(avail
, 0, content
->array
.raid_disks
);
809 for (i
=0; i
< bestcnt
; i
++) {
811 int event_margin
= 1; /* always allow a difference of '1'
812 * like the kernel does
815 /* note: we ignore error flags in multipath arrays
816 * as they don't make sense
818 if (content
->array
.level
!= LEVEL_MULTIPATH
)
819 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
820 if (!(devices
[j
].i
.disk
.state
821 & (1<<MD_DISK_FAULTY
)))
825 /* If this devices thinks that 'most_recent' has failed, then
826 * we must reject this device.
828 if (j
!= most_recent
&&
829 content
->array
.raid_disks
> 0 &&
830 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
831 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
833 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
834 devices
[j
].devname
, devices
[most_recent
].devname
);
838 if (devices
[j
].i
.events
+event_margin
>=
839 devices
[most_recent
].i
.events
) {
840 devices
[j
].uptodate
= 1;
841 if (i
< content
->array
.raid_disks
) {
842 if (devices
[j
].i
.recovery_start
== MaxSector
) {
852 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
853 content
->array
.layout
, 1,
855 /* Choose the newest best drive which is
856 * not up-to-date, update the superblock
860 struct supertype
*tst
;
861 unsigned long long current_events
;
863 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
866 !devices
[j
].uptodate
&&
867 devices
[j
].i
.recovery_start
== MaxSector
&&
870 > devices
[chosen_drive
].i
.events
))
873 if (chosen_drive
< 0)
875 current_events
= devices
[chosen_drive
].i
.events
;
878 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
879 devices
[chosen_drive
].devname
,
880 devices
[chosen_drive
].i
.disk
.raid_disk
,
881 (int)(devices
[chosen_drive
].i
.events
),
882 (int)(devices
[most_recent
].i
.events
));
883 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
885 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
886 devices
[chosen_drive
].devname
);
887 devices
[chosen_drive
].i
.events
= 0;
891 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
893 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
894 devices
[chosen_drive
].devname
);
895 devices
[chosen_drive
].i
.events
= 0;
898 content
->events
= devices
[most_recent
].i
.events
;
899 tst
->ss
->update_super(tst
, content
, "force-one",
900 devices
[chosen_drive
].devname
, verbose
,
903 if (tst
->ss
->store_super(tst
, fd
)) {
905 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
906 devices
[chosen_drive
].devname
);
907 devices
[chosen_drive
].i
.events
= 0;
908 tst
->ss
->free_super(tst
);
912 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
913 devices
[chosen_drive
].uptodate
= 1;
914 avail
[chosen_drive
] = 1;
916 tst
->ss
->free_super(tst
);
918 /* If there are any other drives of the same vintage,
919 * add them in as well. We can't lose and we might gain
921 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
924 !devices
[j
].uptodate
&&
925 devices
[j
].i
.events
== current_events
) {
932 /* Now we want to look at the superblock which the kernel will base things on
933 * and compare the devices that we think are working with the devices that the
934 * superblock thinks are working.
935 * If there are differences and --force is given, then update this chosen
939 st
->ss
->free_super(st
);
940 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
946 if (!devices
[j
].uptodate
)
949 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
950 fprintf(stderr
, Name
": Cannot open %s: %s\n",
951 devices
[j
].devname
, strerror(errno
));
956 if (st
->ss
->load_super(st
,fd
, NULL
)) {
958 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
966 if (st
->sb
== NULL
) {
967 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
972 st
->ss
->getinfo_super(st
, content
, NULL
);
974 sysfs_init(content
, mdfd
, 0);
976 for (i
=0; i
<bestcnt
; i
++) {
978 unsigned int desired_state
;
980 if (i
< content
->array
.raid_disks
)
981 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
987 if (!devices
[j
].uptodate
)
990 devices
[j
].i
.disk
.state
= desired_state
;
991 if (!(devices
[j
].i
.array
.state
& 1))
994 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
998 fprintf(stderr
, Name
": "
999 "clearing FAULTY flag for device %d in %s for %s\n",
1000 j
, mddev
, devices
[j
].devname
);
1004 fprintf(stderr
, Name
": "
1005 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1006 i
, mddev
, devices
[j
].devname
);
1010 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1011 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1016 if (force
&& !clean
&&
1017 !enough(content
->array
.level
, content
->array
.raid_disks
,
1018 content
->array
.layout
, clean
,
1020 change
+= st
->ss
->update_super(st
, content
, "force-array",
1021 devices
[chosen_drive
].devname
, verbose
,
1028 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1030 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1031 devices
[chosen_drive
].devname
);
1036 if (st
->ss
->store_super(st
, fd
)) {
1038 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1039 devices
[chosen_drive
].devname
);
1047 /* If we are in the middle of a reshape we may need to restore saved data
1048 * that was moved aside due to the reshape overwriting live data
1049 * The code of doing this lives in Grow.c
1052 if (content
->reshape_active
) {
1054 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1056 fprintf(stderr
, Name
":%s has an active reshape - checking "
1057 "if critical section needs to be restored\n",
1059 for (i
=0; i
<bestcnt
; i
++) {
1062 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1063 if (fdlist
[i
] < 0) {
1064 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1065 devices
[j
].devname
);
1073 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1076 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1079 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1080 if (backup_file
== NULL
)
1081 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1088 /* count number of in-sync devices according to the superblock.
1089 * We must have this number to start the array without -s or -R
1091 req_cnt
= content
->array
.working_disks
;
1093 /* Almost ready to actually *do* something */
1097 /* First, fill in the map, so that udev can find our name
1098 * as soon as we become active.
1100 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1101 content
->uuid
, chosen_name
);
1103 rv
= set_array_info(mdfd
, st
, content
);
1105 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1106 mddev
, strerror(errno
));
1107 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1112 if (ident
->bitmap_fd
>= 0) {
1113 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1114 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1115 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1120 } else if (ident
->bitmap_file
) {
1121 /* From config file */
1122 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1124 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1125 ident
->bitmap_file
);
1126 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1131 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1132 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1134 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1142 /* First, add the raid disks, but add the chosen one last */
1143 for (i
=0; i
<= bestcnt
; i
++) {
1147 if (j
== chosen_drive
)
1152 if (j
>= 0 /* && devices[j].uptodate */) {
1153 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1156 fprintf(stderr
, Name
": failed to add "
1161 if (i
< content
->array
.raid_disks
1166 } else if (verbose
> 0)
1167 fprintf(stderr
, Name
": added %s "
1169 devices
[j
].devname
, mddev
,
1170 devices
[j
].i
.disk
.raid_disk
);
1171 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1172 fprintf(stderr
, Name
": no uptodate device for "
1177 if (content
->array
.level
== LEVEL_CONTAINER
) {
1179 fprintf(stderr
, Name
": Container %s has been "
1180 "assembled with %d drive%s",
1181 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1182 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1183 fprintf(stderr
, " (out of %d)",
1184 content
->array
.raid_disks
);
1185 fprintf(stderr
, "\n");
1187 sysfs_uevent(content
, "change");
1188 wait_for(chosen_name
, mdfd
);
1196 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1197 content
->array
.layout
, clean
, avail
, okcnt
) &&
1198 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1200 /* This array is good-to-go.
1201 * If a reshape is in progress then we might need to
1202 * continue monitoring it. In that case we start
1203 * it read-only and let the grow code make it writable.
1207 if (content
->reshape_active
&&
1208 content
->delta_disks
<= 0)
1209 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1212 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1215 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1216 mddev
, okcnt
, okcnt
==1?"":"s");
1217 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1218 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1220 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1222 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1223 fprintf(stderr
, ".\n");
1225 if (content
->reshape_active
&&
1226 content
->array
.level
>= 4 &&
1227 content
->array
.level
<= 6) {
1228 /* might need to increase the size
1229 * of the stripe cache - default is 256
1231 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1232 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1234 sysfs_set_num(sra
, NULL
,
1235 "stripe_cache_size",
1236 (4 * content
->array
.chunk_size
/ 4096) + 1);
1239 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1240 /* If any devices did not get added
1241 * because the kernel rejected them based
1242 * on event count, try adding them
1243 * again providing the action policy is
1244 * 're-add' or greater. The bitmap
1245 * might allow them to be included, or
1246 * they will become spares.
1248 for (i
= 0; i
<= bestcnt
; i
++) {
1250 if (j
>= 0 && !devices
[j
].uptodate
) {
1251 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1253 rv
= add_disk(mdfd
, st
, content
,
1255 if (rv
== 0 && verbose
>= 0)
1257 Name
": %s has been re-added.\n",
1258 devices
[j
].devname
);
1262 wait_for(mddev
, mdfd
);
1266 /* There is a nasty race with 'mdadm --monitor'.
1267 * If it opens this device before we close it,
1268 * it gets an incomplete open on which IO
1269 * doesn't work and the capacity is
1271 * If we reopen (to check for layered devices)
1272 * before --monitor closes, we loose.
1274 * So: wait upto 1 second for there to be
1275 * a non-zero capacity.
1277 while (usecs
< 1000) {
1278 mdfd
= open(mddev
, O_RDONLY
);
1280 unsigned long long size
;
1281 if (get_dev_size(mdfd
, NULL
, &size
) &&
1293 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1294 mddev
, strerror(errno
));
1296 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1297 content
->array
.layout
, 1, avail
, okcnt
))
1298 fprintf(stderr
, Name
": Not enough devices to "
1299 "start the array.\n");
1300 else if (!enough(content
->array
.level
,
1301 content
->array
.raid_disks
,
1302 content
->array
.layout
, clean
,
1304 fprintf(stderr
, Name
": Not enough devices to "
1305 "start the array while not clean "
1306 "- consider --force.\n");
1309 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1314 if (runstop
== -1) {
1315 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1316 mddev
, okcnt
, okcnt
==1?"":"s");
1317 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1318 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1319 fprintf(stderr
, ", but not started.\n");
1324 if (verbose
>= -1) {
1325 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1327 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1329 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1330 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1331 content
->array
.layout
, 1, avail
, okcnt
))
1332 fprintf(stderr
, " - not enough to start the array.\n");
1333 else if (!enough(content
->array
.level
,
1334 content
->array
.raid_disks
,
1335 content
->array
.layout
, clean
,
1337 fprintf(stderr
, " - not enough to start the "
1338 "array while not clean - consider "
1341 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1342 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1344 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1345 fprintf(stderr
, " (use --run to insist).\n");
1349 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1354 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1355 * been updated to point to the current locations of devices.
1356 * so we can just start the array
1359 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1360 devices
[chosen_drive
].i
.disk
.minor
);
1361 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1362 fprintf(stderr
, Name
": Cannot start array: %s\n",
1373 int assemble_container_content(struct supertype
*st
, int mdfd
,
1374 struct mdinfo
*content
, int runstop
,
1375 char *chosen_name
, int verbose
)
1377 struct mdinfo
*dev
, *sra
;
1378 int working
= 0, preexist
= 0;
1379 struct map_ent
*map
= NULL
;
1381 sysfs_init(content
, mdfd
, 0);
1383 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1384 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1385 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1392 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1393 if (sysfs_add_disk(content
, dev
, 1) == 0)
1395 else if (errno
== EEXIST
)
1399 return 1;/* Nothing new, don't try to start */
1402 map_update(&map
, fd2devnum(mdfd
),
1403 content
->text_version
,
1404 content
->uuid
, chosen_name
);
1407 (working
+ preexist
) >= content
->array
.working_disks
) {
1410 switch(content
->array
.level
) {
1412 case LEVEL_MULTIPATH
:
1414 err
= sysfs_set_str(content
, NULL
, "array_state",
1418 err
= sysfs_set_str(content
, NULL
, "array_state",
1420 /* start mdmon if needed. */
1422 if (!mdmon_running(st
->container_dev
))
1423 start_mdmon(st
->container_dev
);
1424 ping_monitor(devnum2devname(st
->container_dev
));
1429 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1432 fprintf(stderr
, Name
1433 ": array %s now has %d devices",
1434 chosen_name
, working
+ preexist
);
1436 fprintf(stderr
, Name
1437 ": Started %s with %d devices",
1438 chosen_name
, working
+ preexist
);
1440 fprintf(stderr
, " (%d new)", working
);
1441 fprintf(stderr
, "\n");
1444 wait_for(chosen_name
, mdfd
);
1447 /* FIXME should have an O_EXCL and wait for read-auto */
1450 fprintf(stderr
, Name
1451 ": %s assembled with %d devices but "
1453 chosen_name
, working
);