]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
771cb77c88ccd5493ec3b094f2bae438be4324c8
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 content
= tst
->ss
->container_content(tst
, NULL
);
413 goto loop
; /* empty container */
415 if (content
->next
== NULL
)
418 if (!ident_matches(ident
, content
, tst
,
420 report_missmatch
? devname
: NULL
)) {
421 content
= content
->next
;
423 } else if (is_member_busy(content
->text_version
)) {
424 if (report_missmatch
)
425 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
426 content
->text_version
,
429 content
= content
->next
;
432 st
= tst
; tst
= NULL
;
433 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
434 fprintf(stderr
, Name
": %s is a container, but is not "
435 "only device given: confused and aborting\n",
437 st
->ss
->free_super(st
);
438 dev_policy_free(pol
);
442 fprintf(stderr
, Name
": found match on member %s in %s\n",
443 content
->text_version
, devname
);
447 if (!ident_matches(ident
, content
, tst
,
449 report_missmatch
? devname
: NULL
))
454 if (st
->minor_version
== -1)
455 st
->minor_version
= tst
->minor_version
;
456 if (st
->ss
!= tst
->ss
||
457 st
->minor_version
!= tst
->minor_version
||
458 st
->ss
->compare_super(st
, tst
) != 0) {
459 /* Some mismatch. If exactly one array matches this host,
460 * we can resolve on that one.
461 * Or, if we are auto assembling, we just ignore the second
467 int first
= st
->ss
->match_home(st
, homehost
);
468 int last
= tst
->ss
->match_home(tst
, homehost
);
470 (first
== 1 || last
== 1)) {
471 /* We can do something */
472 if (first
) {/* just ignore this one */
473 if (report_missmatch
)
474 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
477 } else { /* reject all those sofar */
478 struct mddev_dev
*td
;
479 if (report_missmatch
)
480 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
482 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
490 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
492 tst
->ss
->free_super(tst
);
493 st
->ss
->free_super(st
);
494 dev_policy_free(pol
);
501 dev_policy_free(pol
);
504 tst
->ss
->free_super(tst
);
507 if (!st
|| !st
->sb
|| !content
)
510 /* Now need to open the array device. Use create_mddev */
511 if (content
== &info
)
512 st
->ss
->getinfo_super(st
, content
, NULL
);
514 trustworthy
= FOREIGN
;
515 name
= content
->name
;
516 switch (st
->ss
->match_home(st
, homehost
)
517 ?: st
->ss
->match_home(st
, "any")) {
520 name
= strchr(content
->name
, ':');
524 name
= content
->name
;
528 /* If the array is listed in mdadm.conf or on
529 * command line, then we trust the name
530 * even if the array doesn't look local
535 content
->array
.level
== LEVEL_CONTAINER
) {
536 name
= content
->text_version
;
537 trustworthy
= METADATA
;
540 if (name
[0] && trustworthy
!= LOCAL
&&
541 ! require_homehost
&&
542 conf_name_is_free(name
))
545 if (trustworthy
== LOCAL
&&
547 /* Ignore 'host:' prefix of name */
548 name
= strchr(name
, ':')+1;
550 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
553 st
->ss
->free_super(st
);
559 vers
= md_get_version(mdfd
);
561 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
562 " Upgrade your kernel or try --build\n");
566 if (mddev_busy(fd2devnum(mdfd
))) {
567 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
569 for (tmpdev
= devlist
;
570 tmpdev
&& tmpdev
->used
!= 1;
571 tmpdev
= tmpdev
->next
)
573 if (tmpdev
&& auto_assem
)
574 fprintf(stderr
, Name
": %s needed for %s...\n",
575 mddev
, tmpdev
->devname
);
578 st
->ss
->free_super(st
);
583 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
586 if (content
!= &info
) {
587 /* This is a member of a container. Try starting the array. */
588 return assemble_container_content(st
, mdfd
, content
, runstop
,
589 chosen_name
, verbose
);
592 /* Ok, no bad inconsistancy, we can try updating etc */
594 content
->update_private
= NULL
;
595 devices
= malloc(num_devs
* sizeof(*devices
));
596 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
597 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
598 char *devname
= tmpdev
->devname
;
600 /* looks like a good enough match to update the super block if needed */
604 /* prepare useful information in info structures */
606 struct supertype
*tst
;
610 if (strcmp(update
, "uuid")==0 &&
613 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
614 read(rfd
, ident
->uuid
, 16) != 16) {
615 *(__u32
*)(ident
->uuid
) = random();
616 *(__u32
*)(ident
->uuid
+1) = random();
617 *(__u32
*)(ident
->uuid
+2) = random();
618 *(__u32
*)(ident
->uuid
+3) = random();
620 if (rfd
>= 0) close(rfd
);
622 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
624 remove_partitions(dfd
);
627 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
628 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
637 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
639 memcpy(content
->uuid
, ident
->uuid
, 16);
640 strcpy(content
->name
, ident
->name
);
641 content
->array
.md_minor
= minor(stb2
.st_rdev
);
643 if (strcmp(update
, "byteorder") == 0)
646 err
= tst
->ss
->update_super(tst
, content
, update
,
652 Name
": --update=%s not understood"
653 " for %s metadata\n",
654 update
, tst
->ss
->name
);
655 tst
->ss
->free_super(tst
);
663 if (strcmp(update
, "uuid")==0 &&
666 memcpy(ident
->uuid
, content
->uuid
, 16);
668 if (tst
->ss
->store_super(tst
, dfd
))
669 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
673 if (strcmp(update
, "uuid")==0 &&
674 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
675 if (bitmap_update_uuid(ident
->bitmap_fd
,
677 tst
->ss
->swapuuid
) != 0)
678 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
682 tst
->ss
->free_super(tst
);
686 struct supertype
*tst
= dup_super(st
);
688 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
690 remove_partitions(dfd
);
692 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
693 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
702 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
703 tst
->ss
->free_super(tst
);
710 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
711 devname
, mddev
, content
->disk
.raid_disk
);
712 devices
[devcnt
].devname
= devname
;
713 devices
[devcnt
].uptodate
= 0;
714 devices
[devcnt
].i
= *content
;
715 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
716 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
717 if (most_recent
< devcnt
) {
718 if (devices
[devcnt
].i
.events
719 > devices
[most_recent
].i
.events
)
720 most_recent
= devcnt
;
722 if (content
->array
.level
== LEVEL_MULTIPATH
)
723 /* with multipath, the raid_disk from the superblock is meaningless */
726 i
= devices
[devcnt
].i
.disk
.raid_disk
;
728 if (nextspare
< content
->array
.raid_disks
)
729 nextspare
= content
->array
.raid_disks
;
732 if (i
>= content
->array
.raid_disks
&&
738 int newbestcnt
= i
+10;
739 int *newbest
= malloc(sizeof(int)*newbestcnt
);
741 for (c
=0; c
< newbestcnt
; c
++)
743 newbest
[c
] = best
[c
];
748 bestcnt
= newbestcnt
;
751 devices
[best
[i
]].i
.events
752 == devices
[devcnt
].i
.events
753 && (devices
[best
[i
]].i
.disk
.minor
754 != devices
[devcnt
].i
.disk
.minor
)
756 && content
->array
.level
!= LEVEL_MULTIPATH
) {
757 /* two different devices with identical superblock.
758 * Could be a mis-detection caused by overlapping
759 * partitions. fail-safe.
761 fprintf(stderr
, Name
": WARNING %s and %s appear"
762 " to have very similar superblocks.\n"
763 " If they are really different, "
764 "please --zero the superblock on one\n"
765 " If they are the same or overlap,"
766 " please remove one from %s.\n",
767 devices
[best
[i
]].devname
, devname
,
768 inargv
? "the list" :
769 "the\n DEVICE list in mdadm.conf"
777 || (devices
[best
[i
]].i
.events
778 < devices
[devcnt
].i
.events
))
783 free(content
->update_private
);
784 content
->update_private
= NULL
;
787 fprintf(stderr
, Name
": no devices found for %s\n",
790 st
->ss
->free_super(st
);
797 if (update
&& strcmp(update
, "byteorder")==0)
798 st
->minor_version
= 90;
800 st
->ss
->getinfo_super(st
, content
, NULL
);
801 clean
= content
->array
.state
& 1;
803 /* now we have some devices that might be suitable.
806 avail
= malloc(content
->array
.raid_disks
);
807 memset(avail
, 0, content
->array
.raid_disks
);
811 for (i
=0; i
< bestcnt
; i
++) {
813 int event_margin
= 1; /* always allow a difference of '1'
814 * like the kernel does
817 /* note: we ignore error flags in multipath arrays
818 * as they don't make sense
820 if (content
->array
.level
!= LEVEL_MULTIPATH
)
821 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
822 if (!(devices
[j
].i
.disk
.state
823 & (1<<MD_DISK_FAULTY
)))
827 /* If this devices thinks that 'most_recent' has failed, then
828 * we must reject this device.
830 if (j
!= most_recent
&&
831 content
->array
.raid_disks
> 0 &&
832 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
833 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
835 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
836 devices
[j
].devname
, devices
[most_recent
].devname
);
840 if (devices
[j
].i
.events
+event_margin
>=
841 devices
[most_recent
].i
.events
) {
842 devices
[j
].uptodate
= 1;
843 if (i
< content
->array
.raid_disks
) {
844 if (devices
[j
].i
.recovery_start
== MaxSector
) {
854 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
855 content
->array
.layout
, 1,
857 /* Choose the newest best drive which is
858 * not up-to-date, update the superblock
862 struct supertype
*tst
;
863 unsigned long long current_events
;
865 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
868 !devices
[j
].uptodate
&&
869 devices
[j
].i
.recovery_start
== MaxSector
&&
872 > devices
[chosen_drive
].i
.events
))
875 if (chosen_drive
< 0)
877 current_events
= devices
[chosen_drive
].i
.events
;
880 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
881 devices
[chosen_drive
].devname
,
882 devices
[chosen_drive
].i
.disk
.raid_disk
,
883 (int)(devices
[chosen_drive
].i
.events
),
884 (int)(devices
[most_recent
].i
.events
));
885 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
887 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
888 devices
[chosen_drive
].devname
);
889 devices
[chosen_drive
].i
.events
= 0;
893 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
895 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
896 devices
[chosen_drive
].devname
);
897 devices
[chosen_drive
].i
.events
= 0;
900 content
->events
= devices
[most_recent
].i
.events
;
901 tst
->ss
->update_super(tst
, content
, "force-one",
902 devices
[chosen_drive
].devname
, verbose
,
905 if (tst
->ss
->store_super(tst
, fd
)) {
907 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
908 devices
[chosen_drive
].devname
);
909 devices
[chosen_drive
].i
.events
= 0;
910 tst
->ss
->free_super(tst
);
914 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
915 devices
[chosen_drive
].uptodate
= 1;
916 avail
[chosen_drive
] = 1;
918 tst
->ss
->free_super(tst
);
920 /* If there are any other drives of the same vintage,
921 * add them in as well. We can't lose and we might gain
923 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
926 !devices
[j
].uptodate
&&
927 devices
[j
].i
.events
== current_events
) {
934 /* Now we want to look at the superblock which the kernel will base things on
935 * and compare the devices that we think are working with the devices that the
936 * superblock thinks are working.
937 * If there are differences and --force is given, then update this chosen
941 st
->ss
->free_super(st
);
942 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
948 if (!devices
[j
].uptodate
)
951 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
952 fprintf(stderr
, Name
": Cannot open %s: %s\n",
953 devices
[j
].devname
, strerror(errno
));
958 if (st
->ss
->load_super(st
,fd
, NULL
)) {
960 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
968 if (st
->sb
== NULL
) {
969 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
974 st
->ss
->getinfo_super(st
, content
, NULL
);
976 sysfs_init(content
, mdfd
, 0);
978 for (i
=0; i
<bestcnt
; i
++) {
980 unsigned int desired_state
;
982 if (i
< content
->array
.raid_disks
)
983 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
989 if (!devices
[j
].uptodate
)
992 devices
[j
].i
.disk
.state
= desired_state
;
993 if (!(devices
[j
].i
.array
.state
& 1))
996 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1000 fprintf(stderr
, Name
": "
1001 "clearing FAULTY flag for device %d in %s for %s\n",
1002 j
, mddev
, devices
[j
].devname
);
1006 fprintf(stderr
, Name
": "
1007 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1008 i
, mddev
, devices
[j
].devname
);
1012 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1013 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1018 if (force
&& !clean
&&
1019 !enough(content
->array
.level
, content
->array
.raid_disks
,
1020 content
->array
.layout
, clean
,
1022 change
+= st
->ss
->update_super(st
, content
, "force-array",
1023 devices
[chosen_drive
].devname
, verbose
,
1030 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1032 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1033 devices
[chosen_drive
].devname
);
1038 if (st
->ss
->store_super(st
, fd
)) {
1040 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1041 devices
[chosen_drive
].devname
);
1049 /* If we are in the middle of a reshape we may need to restore saved data
1050 * that was moved aside due to the reshape overwriting live data
1051 * The code of doing this lives in Grow.c
1054 if (content
->reshape_active
) {
1056 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1058 fprintf(stderr
, Name
":%s has an active reshape - checking "
1059 "if critical section needs to be restored\n",
1061 for (i
=0; i
<bestcnt
; i
++) {
1064 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1065 if (fdlist
[i
] < 0) {
1066 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1067 devices
[j
].devname
);
1075 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1078 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1081 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1082 if (backup_file
== NULL
)
1083 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1090 /* count number of in-sync devices according to the superblock.
1091 * We must have this number to start the array without -s or -R
1093 req_cnt
= content
->array
.working_disks
;
1095 /* Almost ready to actually *do* something */
1099 /* First, fill in the map, so that udev can find our name
1100 * as soon as we become active.
1102 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1103 content
->uuid
, chosen_name
);
1105 rv
= set_array_info(mdfd
, st
, content
);
1107 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1108 mddev
, strerror(errno
));
1109 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1114 if (ident
->bitmap_fd
>= 0) {
1115 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1116 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1117 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1122 } else if (ident
->bitmap_file
) {
1123 /* From config file */
1124 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1126 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1127 ident
->bitmap_file
);
1128 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1133 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1134 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1136 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1144 /* First, add the raid disks, but add the chosen one last */
1145 for (i
=0; i
<= bestcnt
; i
++) {
1149 if (j
== chosen_drive
)
1154 if (j
>= 0 /* && devices[j].uptodate */) {
1155 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1158 fprintf(stderr
, Name
": failed to add "
1163 if (i
< content
->array
.raid_disks
1168 } else if (verbose
> 0)
1169 fprintf(stderr
, Name
": added %s "
1171 devices
[j
].devname
, mddev
,
1172 devices
[j
].i
.disk
.raid_disk
);
1173 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1174 fprintf(stderr
, Name
": no uptodate device for "
1179 if (content
->array
.level
== LEVEL_CONTAINER
) {
1181 fprintf(stderr
, Name
": Container %s has been "
1182 "assembled with %d drive%s",
1183 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1184 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1185 fprintf(stderr
, " (out of %d)",
1186 content
->array
.raid_disks
);
1187 fprintf(stderr
, "\n");
1189 sysfs_uevent(content
, "change");
1190 wait_for(chosen_name
, mdfd
);
1198 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1199 content
->array
.layout
, clean
, avail
, okcnt
) &&
1200 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1202 /* This array is good-to-go.
1203 * If a reshape is in progress then we might need to
1204 * continue monitoring it. In that case we start
1205 * it read-only and let the grow code make it writable.
1209 if (content
->reshape_active
&&
1210 content
->delta_disks
<= 0)
1211 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1214 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1217 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1218 mddev
, okcnt
, okcnt
==1?"":"s");
1219 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1220 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1222 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1224 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1225 fprintf(stderr
, ".\n");
1227 if (content
->reshape_active
&&
1228 content
->array
.level
>= 4 &&
1229 content
->array
.level
<= 6) {
1230 /* might need to increase the size
1231 * of the stripe cache - default is 256
1233 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1234 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1236 sysfs_set_num(sra
, NULL
,
1237 "stripe_cache_size",
1238 (4 * content
->array
.chunk_size
/ 4096) + 1);
1241 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1242 /* If any devices did not get added
1243 * because the kernel rejected them based
1244 * on event count, try adding them
1245 * again providing the action policy is
1246 * 're-add' or greater. The bitmap
1247 * might allow them to be included, or
1248 * they will become spares.
1250 for (i
= 0; i
<= bestcnt
; i
++) {
1252 if (j
>= 0 && !devices
[j
].uptodate
) {
1253 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1255 rv
= add_disk(mdfd
, st
, content
,
1257 if (rv
== 0 && verbose
>= 0)
1259 Name
": %s has been re-added.\n",
1260 devices
[j
].devname
);
1264 wait_for(mddev
, mdfd
);
1268 /* There is a nasty race with 'mdadm --monitor'.
1269 * If it opens this device before we close it,
1270 * it gets an incomplete open on which IO
1271 * doesn't work and the capacity is
1273 * If we reopen (to check for layered devices)
1274 * before --monitor closes, we loose.
1276 * So: wait upto 1 second for there to be
1277 * a non-zero capacity.
1279 while (usecs
< 1000) {
1280 mdfd
= open(mddev
, O_RDONLY
);
1282 unsigned long long size
;
1283 if (get_dev_size(mdfd
, NULL
, &size
) &&
1295 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1296 mddev
, strerror(errno
));
1298 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1299 content
->array
.layout
, 1, avail
, okcnt
))
1300 fprintf(stderr
, Name
": Not enough devices to "
1301 "start the array.\n");
1302 else if (!enough(content
->array
.level
,
1303 content
->array
.raid_disks
,
1304 content
->array
.layout
, clean
,
1306 fprintf(stderr
, Name
": Not enough devices to "
1307 "start the array while not clean "
1308 "- consider --force.\n");
1311 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1316 if (runstop
== -1) {
1317 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1318 mddev
, okcnt
, okcnt
==1?"":"s");
1319 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1320 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1321 fprintf(stderr
, ", but not started.\n");
1326 if (verbose
>= -1) {
1327 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1329 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1331 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1332 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1333 content
->array
.layout
, 1, avail
, okcnt
))
1334 fprintf(stderr
, " - not enough to start the array.\n");
1335 else if (!enough(content
->array
.level
,
1336 content
->array
.raid_disks
,
1337 content
->array
.layout
, clean
,
1339 fprintf(stderr
, " - not enough to start the "
1340 "array while not clean - consider "
1343 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1344 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1346 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1347 fprintf(stderr
, " (use --run to insist).\n");
1351 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1356 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1357 * been updated to point to the current locations of devices.
1358 * so we can just start the array
1361 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1362 devices
[chosen_drive
].i
.disk
.minor
);
1363 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1364 fprintf(stderr
, Name
": Cannot start array: %s\n",
1375 int assemble_container_content(struct supertype
*st
, int mdfd
,
1376 struct mdinfo
*content
, int runstop
,
1377 char *chosen_name
, int verbose
)
1379 struct mdinfo
*dev
, *sra
;
1380 int working
= 0, preexist
= 0;
1381 struct map_ent
*map
= NULL
;
1383 sysfs_init(content
, mdfd
, 0);
1385 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1386 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1387 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1394 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1395 if (sysfs_add_disk(content
, dev
, 1) == 0)
1397 else if (errno
== EEXIST
)
1401 return 1;/* Nothing new, don't try to start */
1404 map_update(&map
, fd2devnum(mdfd
),
1405 content
->text_version
,
1406 content
->uuid
, chosen_name
);
1409 (working
+ preexist
) >= content
->array
.working_disks
) {
1412 switch(content
->array
.level
) {
1414 case LEVEL_MULTIPATH
:
1416 err
= sysfs_set_str(content
, NULL
, "array_state",
1420 err
= sysfs_set_str(content
, NULL
, "array_state",
1422 /* start mdmon if needed. */
1424 if (!mdmon_running(st
->container_dev
))
1425 start_mdmon(st
->container_dev
);
1426 ping_monitor(devnum2devname(st
->container_dev
));
1431 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1434 fprintf(stderr
, Name
1435 ": array %s now has %d devices",
1436 chosen_name
, working
+ preexist
);
1438 fprintf(stderr
, Name
1439 ": Started %s with %d devices",
1440 chosen_name
, working
+ preexist
);
1442 fprintf(stderr
, " (%d new)", working
);
1443 fprintf(stderr
, "\n");
1446 wait_for(chosen_name
, mdfd
);
1449 /* FIXME should have an O_EXCL and wait for read-auto */
1452 fprintf(stderr
, Name
1453 ": %s assembled with %d devices but "
1455 chosen_name
, working
);