]>
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",
268 } else if (tst
->ss
->compare_super
== NULL
) {
269 if (report_missmatch
)
270 fprintf(stderr
, Name
": Cannot assemble %s metadata on %s\n",
271 tst
->ss
->name
, devname
);
272 tst
->ss
->free_super(tst
);
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
);
281 tst
->ss
->free_super(tst
);
285 memset(content
, 0, sizeof(*content
));
286 tst
->ss
->getinfo_super(tst
, content
, NULL
);
288 if (dfd
>= 0) close(dfd
);
290 if (tst
&& tst
->sb
&& tst
->ss
->container_content
291 && tst
->loaded_container
) {
292 /* tmpdev is a container. We need to be either
293 * looking for a member, or auto-assembling
296 /* already found some components, this cannot
299 if (report_missmatch
)
300 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
305 if (ident
->container
) {
306 if (ident
->container
[0] == '/' &&
307 !same_dev(ident
->container
, devname
)) {
308 if (report_missmatch
)
309 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
310 devname
, ident
->container
);
313 if (ident
->container
[0] != '/') {
316 if (!parse_uuid(ident
->container
, uuid
) ||
317 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
318 if (report_missmatch
)
319 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
325 /* It is worth looking inside this container.
328 fprintf(stderr
, Name
": looking in container %s\n",
332 content
= tmpdev
->content
;
334 content
= tst
->ss
->container_content(tst
);
336 goto loop
; /* empty container */
338 tmpdev
->content
= content
->next
;
339 if (tmpdev
->content
== NULL
)
342 } else if (ident
->container
|| ident
->member
) {
343 /* No chance of this matching if we don't have
345 if (report_missmatch
)
346 fprintf(stderr
, Name
"%s is not a container, and one is required.\n",
351 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
353 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0)) {
354 if (report_missmatch
)
355 fprintf(stderr
, Name
": %s has wrong uuid.\n",
359 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
361 name_matches(content
->name
, ident
->name
, homehost
)==0)) {
362 if (report_missmatch
)
363 fprintf(stderr
, Name
": %s has wrong name.\n",
367 if (ident
->super_minor
!= UnSet
&&
369 ident
->super_minor
!= content
->array
.md_minor
)) {
370 if (report_missmatch
)
371 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
375 if (ident
->level
!= UnSet
&&
377 ident
->level
!= content
->array
.level
)) {
378 if (report_missmatch
)
379 fprintf(stderr
, Name
": %s has wrong raid level.\n",
383 if (ident
->raid_disks
!= UnSet
&&
385 ident
->raid_disks
!= content
->array
.raid_disks
)) {
386 if (report_missmatch
)
387 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
392 if (tst
== NULL
|| tst
->sb
== NULL
)
395 /* If we are this far, then we are nearly commited to this device.
396 * If the super_block doesn't exist, or doesn't match others,
397 * then we probably cannot continue
398 * However if one of the arrays is for the homehost, and
399 * the other isn't that can disambiguate.
402 if (!tst
|| !tst
->sb
) {
403 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
406 st
->ss
->free_super(st
);
407 dev_policy_free(pol
);
411 if (tst
&& tst
->sb
&& tst
->ss
->container_content
412 && tst
->loaded_container
) {
413 /* we have the one container we need, don't keep
414 * looking. If the chosen member is active, skip.
416 if (is_member_busy(content
->text_version
)) {
417 if (report_missmatch
)
418 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
419 content
->text_version
,
424 tst
->ss
->free_super(tst
);
429 dev_policy_free(pol
);
432 if (ident
->member
&& ident
->member
[0]) {
433 char *s
= strchr(content
->text_version
+1, '/');
435 fprintf(stderr
, Name
": badly formatted version: %s\n",
436 content
->text_version
);
439 if (strcmp(ident
->member
, s
+1) != 0) {
440 if (report_missmatch
)
442 Name
": skipping wrong member %s\n",
443 content
->text_version
);
447 st
= tst
; tst
= NULL
;
448 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
449 fprintf(stderr
, Name
": %s is a container, but is not "
450 "only device given: confused and aborting\n",
452 st
->ss
->free_super(st
);
453 dev_policy_free(pol
);
457 fprintf(stderr
, Name
": found match on member %s in %s\n",
458 content
->text_version
, devname
);
463 if (st
->minor_version
== -1)
464 st
->minor_version
= tst
->minor_version
;
465 if (st
->ss
!= tst
->ss
||
466 st
->minor_version
!= tst
->minor_version
||
467 st
->ss
->compare_super(st
, tst
) != 0) {
468 /* Some mismatch. If exactly one array matches this host,
469 * we can resolve on that one.
470 * Or, if we are auto assembling, we just ignore the second
476 int first
= st
->ss
->match_home(st
, homehost
);
477 int last
= tst
->ss
->match_home(tst
, homehost
);
479 (first
== 1 || last
== 1)) {
480 /* We can do something */
481 if (first
) {/* just ignore this one */
482 if (report_missmatch
)
483 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
486 } else { /* reject all those sofar */
488 if (report_missmatch
)
489 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
491 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
499 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
501 tst
->ss
->free_super(tst
);
502 st
->ss
->free_super(st
);
503 dev_policy_free(pol
);
510 dev_policy_free(pol
);
515 tst
->ss
->free_super(tst
);
518 if (!st
|| !st
->sb
|| !content
)
521 /* Now need to open the array device. Use create_mddev */
522 if (content
== &info
)
523 st
->ss
->getinfo_super(st
, content
, NULL
);
525 trustworthy
= FOREIGN
;
526 name
= content
->name
;
527 switch (st
->ss
->match_home(st
, homehost
)
528 ?: st
->ss
->match_home(st
, "any")) {
531 name
= strchr(content
->name
, ':');
535 name
= content
->name
;
539 /* If the array is listed in mdadm.conf or on
540 * command line, then we trust the name
541 * even if the array doesn't look local
546 content
->array
.level
== LEVEL_CONTAINER
) {
547 name
= content
->text_version
;
548 trustworthy
= METADATA
;
551 if (name
[0] && trustworthy
!= LOCAL
&&
552 ! require_homehost
&&
553 conf_name_is_free(name
))
556 if (trustworthy
== LOCAL
&&
558 /* Ignore 'host:' prefix of name */
559 name
= strchr(name
, ':')+1;
561 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
564 st
->ss
->free_super(st
);
570 vers
= md_get_version(mdfd
);
572 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
573 " Upgrade your kernel or try --build\n");
577 if (mddev_busy(fd2devnum(mdfd
))) {
578 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
580 for (tmpdev
= devlist
;
581 tmpdev
&& tmpdev
->used
!= 1;
582 tmpdev
= tmpdev
->next
)
584 if (tmpdev
&& auto_assem
)
585 fprintf(stderr
, Name
": %s needed for %s...\n",
586 mddev
, tmpdev
->devname
);
589 st
->ss
->free_super(st
);
594 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
597 if (content
!= &info
) {
598 /* This is a member of a container. Try starting the array. */
599 return assemble_container_content(st
, mdfd
, content
, runstop
,
600 chosen_name
, verbose
);
603 /* Ok, no bad inconsistancy, we can try updating etc */
605 content
->update_private
= NULL
;
606 devices
= malloc(num_devs
* sizeof(*devices
));
607 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
608 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
609 char *devname
= tmpdev
->devname
;
611 /* looks like a good enough match to update the super block if needed */
615 /* prepare useful information in info structures */
617 struct supertype
*tst
;
621 if (strcmp(update
, "uuid")==0 &&
624 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
625 read(rfd
, ident
->uuid
, 16) != 16) {
626 *(__u32
*)(ident
->uuid
) = random();
627 *(__u32
*)(ident
->uuid
+1) = random();
628 *(__u32
*)(ident
->uuid
+2) = random();
629 *(__u32
*)(ident
->uuid
+3) = random();
631 if (rfd
>= 0) close(rfd
);
633 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
635 remove_partitions(dfd
);
638 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
639 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
648 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
650 memcpy(content
->uuid
, ident
->uuid
, 16);
651 strcpy(content
->name
, ident
->name
);
652 content
->array
.md_minor
= minor(stb2
.st_rdev
);
654 if (strcmp(update
, "byteorder") == 0)
657 err
= tst
->ss
->update_super(tst
, content
, update
,
663 Name
": --update=%s not understood"
664 " for %s metadata\n",
665 update
, tst
->ss
->name
);
666 tst
->ss
->free_super(tst
);
674 if (strcmp(update
, "uuid")==0 &&
677 memcpy(ident
->uuid
, content
->uuid
, 16);
679 if (tst
->ss
->store_super(tst
, dfd
))
680 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
684 if (strcmp(update
, "uuid")==0 &&
685 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
686 if (bitmap_update_uuid(ident
->bitmap_fd
,
688 tst
->ss
->swapuuid
) != 0)
689 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
693 tst
->ss
->free_super(tst
);
697 struct supertype
*tst
= dup_super(st
);
699 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
701 remove_partitions(dfd
);
703 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
704 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
713 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
714 tst
->ss
->free_super(tst
);
721 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
722 devname
, mddev
, content
->disk
.raid_disk
);
723 devices
[devcnt
].devname
= devname
;
724 devices
[devcnt
].uptodate
= 0;
725 devices
[devcnt
].i
= *content
;
726 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
727 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
728 if (most_recent
< devcnt
) {
729 if (devices
[devcnt
].i
.events
730 > devices
[most_recent
].i
.events
)
731 most_recent
= devcnt
;
733 if (content
->array
.level
== LEVEL_MULTIPATH
)
734 /* with multipath, the raid_disk from the superblock is meaningless */
737 i
= devices
[devcnt
].i
.disk
.raid_disk
;
739 if (nextspare
< content
->array
.raid_disks
)
740 nextspare
= content
->array
.raid_disks
;
743 if (i
>= content
->array
.raid_disks
&&
749 int newbestcnt
= i
+10;
750 int *newbest
= malloc(sizeof(int)*newbestcnt
);
752 for (c
=0; c
< newbestcnt
; c
++)
754 newbest
[c
] = best
[c
];
759 bestcnt
= newbestcnt
;
762 devices
[best
[i
]].i
.events
763 == devices
[devcnt
].i
.events
764 && (devices
[best
[i
]].i
.disk
.minor
765 != devices
[devcnt
].i
.disk
.minor
)
767 && content
->array
.level
!= LEVEL_MULTIPATH
) {
768 /* two different devices with identical superblock.
769 * Could be a mis-detection caused by overlapping
770 * partitions. fail-safe.
772 fprintf(stderr
, Name
": WARNING %s and %s appear"
773 " to have very similar superblocks.\n"
774 " If they are really different, "
775 "please --zero the superblock on one\n"
776 " If they are the same or overlap,"
777 " please remove one from %s.\n",
778 devices
[best
[i
]].devname
, devname
,
779 inargv
? "the list" :
780 "the\n DEVICE list in mdadm.conf"
788 || (devices
[best
[i
]].i
.events
789 < devices
[devcnt
].i
.events
))
794 free(content
->update_private
);
795 content
->update_private
= NULL
;
798 fprintf(stderr
, Name
": no devices found for %s\n",
801 st
->ss
->free_super(st
);
808 if (update
&& strcmp(update
, "byteorder")==0)
809 st
->minor_version
= 90;
811 st
->ss
->getinfo_super(st
, content
, NULL
);
812 clean
= content
->array
.state
& 1;
814 /* now we have some devices that might be suitable.
817 avail
= malloc(content
->array
.raid_disks
);
818 memset(avail
, 0, content
->array
.raid_disks
);
822 for (i
=0; i
< bestcnt
; i
++) {
824 int event_margin
= 1; /* always allow a difference of '1'
825 * like the kernel does
828 /* note: we ignore error flags in multipath arrays
829 * as they don't make sense
831 if (content
->array
.level
!= LEVEL_MULTIPATH
)
832 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
833 if (!(devices
[j
].i
.disk
.state
834 & (1<<MD_DISK_FAULTY
)))
838 /* If this devices thinks that 'most_recent' has failed, then
839 * we must reject this device.
841 if (j
!= most_recent
&&
842 content
->array
.raid_disks
> 0 &&
843 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
844 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
846 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
847 devices
[j
].devname
, devices
[most_recent
].devname
);
851 if (devices
[j
].i
.events
+event_margin
>=
852 devices
[most_recent
].i
.events
) {
853 devices
[j
].uptodate
= 1;
854 if (i
< content
->array
.raid_disks
) {
855 if (devices
[j
].i
.recovery_start
== MaxSector
) {
865 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
866 content
->array
.layout
, 1,
868 /* Choose the newest best drive which is
869 * not up-to-date, update the superblock
873 struct supertype
*tst
;
874 unsigned long long current_events
;
876 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
879 !devices
[j
].uptodate
&&
880 devices
[j
].i
.recovery_start
== MaxSector
&&
883 > devices
[chosen_drive
].i
.events
))
886 if (chosen_drive
< 0)
888 current_events
= devices
[chosen_drive
].i
.events
;
891 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
892 devices
[chosen_drive
].devname
,
893 devices
[chosen_drive
].i
.disk
.raid_disk
,
894 (int)(devices
[chosen_drive
].i
.events
),
895 (int)(devices
[most_recent
].i
.events
));
896 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
898 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
899 devices
[chosen_drive
].devname
);
900 devices
[chosen_drive
].i
.events
= 0;
904 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
906 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
907 devices
[chosen_drive
].devname
);
908 devices
[chosen_drive
].i
.events
= 0;
911 content
->events
= devices
[most_recent
].i
.events
;
912 tst
->ss
->update_super(tst
, content
, "force-one",
913 devices
[chosen_drive
].devname
, verbose
,
916 if (tst
->ss
->store_super(tst
, fd
)) {
918 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
919 devices
[chosen_drive
].devname
);
920 devices
[chosen_drive
].i
.events
= 0;
921 tst
->ss
->free_super(tst
);
925 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
926 devices
[chosen_drive
].uptodate
= 1;
927 avail
[chosen_drive
] = 1;
929 tst
->ss
->free_super(tst
);
931 /* If there are any other drives of the same vintage,
932 * add them in as well. We can't lose and we might gain
934 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
937 !devices
[j
].uptodate
&&
938 devices
[j
].i
.events
== current_events
) {
945 /* Now we want to look at the superblock which the kernel will base things on
946 * and compare the devices that we think are working with the devices that the
947 * superblock thinks are working.
948 * If there are differences and --force is given, then update this chosen
952 st
->ss
->free_super(st
);
953 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
959 if (!devices
[j
].uptodate
)
962 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
963 fprintf(stderr
, Name
": Cannot open %s: %s\n",
964 devices
[j
].devname
, strerror(errno
));
969 if (st
->ss
->load_super(st
,fd
, NULL
)) {
971 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
979 if (st
->sb
== NULL
) {
980 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
985 st
->ss
->getinfo_super(st
, content
, NULL
);
987 sysfs_init(content
, mdfd
, 0);
989 for (i
=0; i
<bestcnt
; i
++) {
991 unsigned int desired_state
;
993 if (i
< content
->array
.raid_disks
)
994 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1000 if (!devices
[j
].uptodate
)
1003 devices
[j
].i
.disk
.state
= desired_state
;
1004 if (!(devices
[j
].i
.array
.state
& 1))
1007 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1008 verbose
, 0, NULL
)) {
1011 fprintf(stderr
, Name
": "
1012 "clearing FAULTY flag for device %d in %s for %s\n",
1013 j
, mddev
, devices
[j
].devname
);
1017 fprintf(stderr
, Name
": "
1018 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1019 i
, mddev
, devices
[j
].devname
);
1023 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1024 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1029 if (force
&& !clean
&&
1030 !enough(content
->array
.level
, content
->array
.raid_disks
,
1031 content
->array
.layout
, clean
,
1033 change
+= st
->ss
->update_super(st
, content
, "force-array",
1034 devices
[chosen_drive
].devname
, verbose
,
1041 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1043 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1044 devices
[chosen_drive
].devname
);
1049 if (st
->ss
->store_super(st
, fd
)) {
1051 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1052 devices
[chosen_drive
].devname
);
1060 /* If we are in the middle of a reshape we may need to restore saved data
1061 * that was moved aside due to the reshape overwriting live data
1062 * The code of doing this lives in Grow.c
1065 if (content
->reshape_active
) {
1067 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1069 fprintf(stderr
, Name
":%s has an active reshape - checking "
1070 "if critical section needs to be restored\n",
1072 for (i
=0; i
<bestcnt
; i
++) {
1075 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1076 if (fdlist
[i
] < 0) {
1077 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1078 devices
[j
].devname
);
1086 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1089 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1092 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1093 if (backup_file
== NULL
)
1094 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1101 /* count number of in-sync devices according to the superblock.
1102 * We must have this number to start the array without -s or -R
1104 req_cnt
= content
->array
.working_disks
;
1106 /* Almost ready to actually *do* something */
1110 /* First, fill in the map, so that udev can find our name
1111 * as soon as we become active.
1113 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1114 content
->uuid
, chosen_name
);
1116 rv
= set_array_info(mdfd
, st
, content
);
1118 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1119 mddev
, strerror(errno
));
1120 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1125 if (ident
->bitmap_fd
>= 0) {
1126 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1127 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1128 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1133 } else if (ident
->bitmap_file
) {
1134 /* From config file */
1135 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1137 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1138 ident
->bitmap_file
);
1139 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1144 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1145 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1147 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1155 /* First, add the raid disks, but add the chosen one last */
1156 for (i
=0; i
<= bestcnt
; i
++) {
1160 if (j
== chosen_drive
)
1165 if (j
>= 0 /* && devices[j].uptodate */) {
1166 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1169 fprintf(stderr
, Name
": failed to add "
1174 if (i
< content
->array
.raid_disks
1179 } else if (verbose
> 0)
1180 fprintf(stderr
, Name
": added %s "
1182 devices
[j
].devname
, mddev
,
1183 devices
[j
].i
.disk
.raid_disk
);
1184 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1185 fprintf(stderr
, Name
": no uptodate device for "
1190 if (content
->array
.level
== LEVEL_CONTAINER
) {
1192 fprintf(stderr
, Name
": Container %s has been "
1193 "assembled with %d drive%s",
1194 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1195 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1196 fprintf(stderr
, " (out of %d)",
1197 content
->array
.raid_disks
);
1198 fprintf(stderr
, "\n");
1200 sysfs_uevent(content
, "change");
1201 wait_for(chosen_name
, mdfd
);
1209 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1210 content
->array
.layout
, clean
, avail
, okcnt
) &&
1211 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1213 /* This array is good-to-go.
1214 * If a reshape is in progress then we might need to
1215 * continue monitoring it. In that case we start
1216 * it read-only and let the grow code make it writable.
1220 if (content
->reshape_active
&&
1221 content
->delta_disks
<= 0)
1222 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1225 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1228 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1229 mddev
, okcnt
, okcnt
==1?"":"s");
1230 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1231 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1233 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1235 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1236 fprintf(stderr
, ".\n");
1238 if (content
->reshape_active
&&
1239 content
->array
.level
>= 4 &&
1240 content
->array
.level
<= 6) {
1241 /* might need to increase the size
1242 * of the stripe cache - default is 256
1244 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1245 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1247 sysfs_set_num(sra
, NULL
,
1248 "stripe_cache_size",
1249 (4 * content
->array
.chunk_size
/ 4096) + 1);
1252 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1253 /* If any devices did not get added
1254 * because the kernel rejected them based
1255 * on event count, try adding them
1256 * again providing the action policy is
1257 * 're-add' or greater. The bitmap
1258 * might allow them to be included, or
1259 * they will become spares.
1261 for (i
= 0; i
<= bestcnt
; i
++) {
1263 if (j
>= 0 && !devices
[j
].uptodate
) {
1264 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1266 rv
= add_disk(mdfd
, st
, content
,
1268 if (rv
== 0 && verbose
>= 0)
1270 Name
": %s has been re-added.\n",
1271 devices
[j
].devname
);
1275 wait_for(mddev
, mdfd
);
1279 /* There is a nasty race with 'mdadm --monitor'.
1280 * If it opens this device before we close it,
1281 * it gets an incomplete open on which IO
1282 * doesn't work and the capacity is
1284 * If we reopen (to check for layered devices)
1285 * before --monitor closes, we loose.
1287 * So: wait upto 1 second for there to be
1288 * a non-zero capacity.
1290 while (usecs
< 1000) {
1291 mdfd
= open(mddev
, O_RDONLY
);
1293 unsigned long long size
;
1294 if (get_dev_size(mdfd
, NULL
, &size
) &&
1306 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1307 mddev
, strerror(errno
));
1309 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1310 content
->array
.layout
, 1, avail
, okcnt
))
1311 fprintf(stderr
, Name
": Not enough devices to "
1312 "start the array.\n");
1313 else if (!enough(content
->array
.level
,
1314 content
->array
.raid_disks
,
1315 content
->array
.layout
, clean
,
1317 fprintf(stderr
, Name
": Not enough devices to "
1318 "start the array while not clean "
1319 "- consider --force.\n");
1322 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1327 if (runstop
== -1) {
1328 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1329 mddev
, okcnt
, okcnt
==1?"":"s");
1330 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1331 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1332 fprintf(stderr
, ", but not started.\n");
1337 if (verbose
>= -1) {
1338 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1340 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1342 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1343 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1344 content
->array
.layout
, 1, avail
, okcnt
))
1345 fprintf(stderr
, " - not enough to start the array.\n");
1346 else if (!enough(content
->array
.level
,
1347 content
->array
.raid_disks
,
1348 content
->array
.layout
, clean
,
1350 fprintf(stderr
, " - not enough to start the "
1351 "array while not clean - consider "
1354 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1355 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1357 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1358 fprintf(stderr
, " (use --run to insist).\n");
1362 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1367 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1368 * been updated to point to the current locations of devices.
1369 * so we can just start the array
1372 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1373 devices
[chosen_drive
].i
.disk
.minor
);
1374 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1375 fprintf(stderr
, Name
": Cannot start array: %s\n",
1386 int assemble_container_content(struct supertype
*st
, int mdfd
,
1387 struct mdinfo
*content
, int runstop
,
1388 char *chosen_name
, int verbose
)
1390 struct mdinfo
*dev
, *sra
;
1391 int working
= 0, preexist
= 0;
1392 struct map_ent
*map
= NULL
;
1394 sysfs_init(content
, mdfd
, 0);
1396 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1397 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1398 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1405 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1406 if (sysfs_add_disk(content
, dev
, 1) == 0)
1408 else if (errno
== EEXIST
)
1412 return 1;/* Nothing new, don't try to start */
1415 map_update(&map
, fd2devnum(mdfd
),
1416 content
->text_version
,
1417 content
->uuid
, chosen_name
);
1420 (working
+ preexist
) >= content
->array
.working_disks
) {
1423 switch(content
->array
.level
) {
1425 case LEVEL_MULTIPATH
:
1427 err
= sysfs_set_str(content
, NULL
, "array_state",
1431 err
= sysfs_set_str(content
, NULL
, "array_state",
1433 /* start mdmon if needed. */
1435 if (!mdmon_running(st
->container_dev
))
1436 start_mdmon(st
->container_dev
);
1437 ping_monitor(devnum2devname(st
->container_dev
));
1442 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1445 fprintf(stderr
, Name
1446 ": array %s now has %d devices",
1447 chosen_name
, working
+ preexist
);
1449 fprintf(stderr
, Name
1450 ": Started %s with %d devices",
1451 chosen_name
, working
+ preexist
);
1453 fprintf(stderr
, " (%d new)", working
);
1454 fprintf(stderr
, "\n");
1457 wait_for(chosen_name
, mdfd
);
1460 /* FIXME should have an O_EXCL and wait for read-auto */
1463 fprintf(stderr
, Name
1464 ": %s assembled with %d devices but "
1466 chosen_name
, working
);