]>
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.
148 int *best
= NULL
; /* indexed by raid_disk */
151 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
152 unsigned int req_cnt
;
158 int report_missmatch
;
160 int start_partial_ok
= (runstop
>= 0) &&
161 (force
|| devlist
==NULL
|| auto_assem
);
162 unsigned int num_devs
;
165 struct mdinfo
*content
= NULL
;
170 char chosen_name
[1024];
172 if (get_linux_version() < 2004000)
176 * If any subdevs are listed, then any that don't
177 * match ident are discarded. Remainder must all match and
179 * If no subdevs, then we scan all devices in the config file, but
180 * there must be something in the identity
184 ident
->uuid_set
== 0 &&
185 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
186 ident
->name
[0] == 0 &&
187 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
188 ident
->devices
== NULL
) {
189 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
190 mddev
? mddev
: "further assembly");
195 devlist
= conf_get_devs();
199 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
201 /* We come back here when doing auto-assembly and attempting some
202 * set of devices failed. Those are now marked as ->used==2 and
203 * we ignore them and try again
206 tmpdev
= devlist
; num_devs
= 0;
212 tmpdev
= tmpdev
->next
;
214 devices
= malloc(num_devs
* sizeof(*devices
));
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
);
234 if (tmpdev
->used
> 1) continue;
236 if (ident
->devices
&&
237 !match_oneof(ident
->devices
, devname
)) {
238 if (report_missmatch
)
239 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
243 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
245 if (report_missmatch
)
246 fprintf(stderr
, Name
": cannot open device %s: %s\n",
247 devname
, strerror(errno
));
249 } else if (fstat(dfd
, &stb
)< 0) {
251 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
252 devname
, strerror(errno
));
254 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
255 fprintf(stderr
, Name
": %s is not a block device.\n",
258 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
259 if (report_missmatch
)
260 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
263 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
264 if (report_missmatch
)
265 fprintf( stderr
, Name
": no RAID superblock on %s\n",
267 } else if (auto_assem
&& st
== NULL
&&
268 !conf_test_metadata(tst
->ss
->name
,
269 tst
->ss
->match_home(tst
, homehost
) == 1)) {
270 if (report_missmatch
)
271 fprintf(stderr
, Name
": %s has metadata type %s for which "
272 "auto-assembly is disabled\n",
273 devname
, tst
->ss
->name
);
274 tst
->ss
->free_super(tst
);
278 memset(content
, 0, sizeof(*content
));
279 tst
->ss
->getinfo_super(tst
, content
);
281 if (dfd
>= 0) close(dfd
);
283 if (tst
&& tst
->sb
&& tst
->ss
->container_content
284 && tst
->loaded_container
) {
285 /* tmpdev is a container. We need to be either
286 * looking for a member, or auto-assembling
289 /* already found some components, this cannot
292 if (report_missmatch
)
293 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
298 if (ident
->container
) {
299 if (ident
->container
[0] == '/' &&
300 !same_dev(ident
->container
, devname
)) {
301 if (report_missmatch
)
302 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
303 devname
, ident
->container
);
306 if (ident
->container
[0] != '/') {
309 if (!parse_uuid(ident
->container
, uuid
) ||
310 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
311 if (report_missmatch
)
312 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
318 /* It is worth looking inside this container.
321 fprintf(stderr
, Name
": looking in container %s\n",
325 content
= tmpdev
->content
;
327 content
= tst
->ss
->container_content(tst
);
329 goto loop
; /* empty container */
331 tmpdev
->content
= content
->next
;
332 if (tmpdev
->content
== NULL
)
335 } else if (ident
->container
|| ident
->member
) {
336 /* No chance of this matching if we don't have
338 if (report_missmatch
)
339 fprintf(stderr
, Name
"%s is not a container, and one is required.\n",
344 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
346 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0)) {
347 if (report_missmatch
)
348 fprintf(stderr
, Name
": %s has wrong uuid.\n",
352 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
354 name_matches(content
->name
, ident
->name
, homehost
)==0)) {
355 if (report_missmatch
)
356 fprintf(stderr
, Name
": %s has wrong name.\n",
360 if (ident
->super_minor
!= UnSet
&&
362 ident
->super_minor
!= content
->array
.md_minor
)) {
363 if (report_missmatch
)
364 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
368 if (ident
->level
!= UnSet
&&
370 ident
->level
!= content
->array
.level
)) {
371 if (report_missmatch
)
372 fprintf(stderr
, Name
": %s has wrong raid level.\n",
376 if (ident
->raid_disks
!= UnSet
&&
378 ident
->raid_disks
!= content
->array
.raid_disks
)) {
379 if (report_missmatch
)
380 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
385 if (tst
== NULL
|| tst
->sb
== NULL
)
388 /* If we are this far, then we are nearly commited to this device.
389 * If the super_block doesn't exist, or doesn't match others,
390 * then we probably cannot continue
391 * However if one of the arrays is for the homehost, and
392 * the other isn't that can disambiguate.
395 if (!tst
|| !tst
->sb
) {
396 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
399 st
->ss
->free_super(st
);
403 if (tst
&& tst
->sb
&& tst
->ss
->container_content
404 && tst
->loaded_container
) {
405 /* we have the one container we need, don't keep
406 * looking. If the chosen member is active, skip.
408 if (is_member_busy(content
->text_version
)) {
409 if (report_missmatch
)
410 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
411 content
->text_version
,
416 tst
->ss
->free_super(tst
);
423 if (ident
->member
&& ident
->member
[0]) {
424 char *s
= strchr(content
->text_version
+1, '/');
426 fprintf(stderr
, Name
": badly formatted version: %s\n",
427 content
->text_version
);
430 if (strcmp(ident
->member
, s
+1) != 0) {
431 if (report_missmatch
)
433 Name
": skipping wrong member %s\n",
434 content
->text_version
);
438 st
= tst
; tst
= NULL
;
439 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
440 fprintf(stderr
, Name
": %s is a container, but is not "
441 "only device given: confused and aborting\n",
443 st
->ss
->free_super(st
);
447 fprintf(stderr
, Name
": found match on member %s in %s\n",
448 content
->text_version
, devname
);
453 if (st
->minor_version
== -1)
454 st
->minor_version
= tst
->minor_version
;
455 if (st
->ss
!= tst
->ss
||
456 st
->minor_version
!= tst
->minor_version
||
457 st
->ss
->compare_super(st
, tst
) != 0) {
458 /* Some mismatch. If exactly one array matches this host,
459 * we can resolve on that one.
460 * Or, if we are auto assembling, we just ignore the second
466 int first
= st
->ss
->match_home(st
, homehost
);
467 int last
= tst
->ss
->match_home(tst
, homehost
);
469 (first
== 1 || last
== 1)) {
470 /* We can do something */
471 if (first
) {/* just ignore this one */
472 if (report_missmatch
)
473 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
476 } else { /* reject all those sofar */
478 if (report_missmatch
)
479 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
481 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
489 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
491 tst
->ss
->free_super(tst
);
492 st
->ss
->free_super(st
);
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
);
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
);
558 vers
= md_get_version(mdfd
);
560 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
561 " Upgrade your kernel or try --build\n");
565 if (mddev_busy(fd2devnum(mdfd
))) {
566 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
568 for (tmpdev
= devlist
;
569 tmpdev
&& tmpdev
->used
!= 1;
570 tmpdev
= tmpdev
->next
)
572 if (tmpdev
&& auto_assem
)
573 fprintf(stderr
, Name
": %s needed for %s...\n",
574 mddev
, tmpdev
->devname
);
577 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 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
;
607 if (strcmp(update
, "uuid")==0 &&
610 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
611 read(rfd
, ident
->uuid
, 16) != 16) {
612 *(__u32
*)(ident
->uuid
) = random();
613 *(__u32
*)(ident
->uuid
+1) = random();
614 *(__u32
*)(ident
->uuid
+2) = random();
615 *(__u32
*)(ident
->uuid
+3) = random();
617 if (rfd
>= 0) close(rfd
);
619 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
622 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
623 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
630 tst
->ss
->getinfo_super(tst
, content
);
632 memcpy(content
->uuid
, ident
->uuid
, 16);
633 strcpy(content
->name
, ident
->name
);
634 content
->array
.md_minor
= minor(stb2
.st_rdev
);
636 tst
->ss
->update_super(tst
, content
, update
,
638 ident
->uuid_set
, homehost
);
639 if (strcmp(update
, "uuid")==0 &&
642 memcpy(ident
->uuid
, content
->uuid
, 16);
645 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
647 else if (tst
->ss
->store_super(tst
, dfd
))
648 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
653 if (strcmp(update
, "uuid")==0 &&
654 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
655 if (bitmap_update_uuid(ident
->bitmap_fd
,
657 tst
->ss
->swapuuid
) != 0)
658 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
662 tst
->ss
->free_super(tst
);
666 struct supertype
*tst
= dup_super(st
);
668 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
670 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
671 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
678 tst
->ss
->getinfo_super(tst
, content
);
679 tst
->ss
->free_super(tst
);
686 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
687 devname
, mddev
, content
->disk
.raid_disk
);
688 devices
[devcnt
].devname
= devname
;
689 devices
[devcnt
].uptodate
= 0;
690 devices
[devcnt
].i
= *content
;
691 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
692 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
693 if (most_recent
< devcnt
) {
694 if (devices
[devcnt
].i
.events
695 > devices
[most_recent
].i
.events
)
696 most_recent
= devcnt
;
698 if (content
->array
.level
== LEVEL_MULTIPATH
)
699 /* with multipath, the raid_disk from the superblock is meaningless */
702 i
= devices
[devcnt
].i
.disk
.raid_disk
;
704 if (nextspare
< content
->array
.raid_disks
)
705 nextspare
= content
->array
.raid_disks
;
708 if (i
>= content
->array
.raid_disks
&&
714 int newbestcnt
= i
+10;
715 int *newbest
= malloc(sizeof(int)*newbestcnt
);
717 for (c
=0; c
< newbestcnt
; c
++)
719 newbest
[c
] = best
[c
];
724 bestcnt
= newbestcnt
;
727 devices
[best
[i
]].i
.events
728 == devices
[devcnt
].i
.events
729 && (devices
[best
[i
]].i
.disk
.minor
730 != devices
[devcnt
].i
.disk
.minor
)
732 && content
->array
.level
!= LEVEL_MULTIPATH
) {
733 /* two different devices with identical superblock.
734 * Could be a mis-detection caused by overlapping
735 * partitions. fail-safe.
737 fprintf(stderr
, Name
": WARNING %s and %s appear"
738 " to have very similar superblocks.\n"
739 " If they are really different, "
740 "please --zero the superblock on one\n"
741 " If they are the same or overlap,"
742 " please remove one from %s.\n",
743 devices
[best
[i
]].devname
, devname
,
744 inargv
? "the list" :
745 "the\n DEVICE list in mdadm.conf"
751 || (devices
[best
[i
]].i
.events
752 < devices
[devcnt
].i
.events
))
757 free(content
->update_private
);
758 content
->update_private
= NULL
;
761 fprintf(stderr
, Name
": no devices found for %s\n",
764 st
->ss
->free_super(st
);
769 if (update
&& strcmp(update
, "byteorder")==0)
770 st
->minor_version
= 90;
772 st
->ss
->getinfo_super(st
, content
);
773 clean
= content
->array
.state
& 1;
775 /* now we have some devices that might be suitable.
778 avail
= malloc(content
->array
.raid_disks
);
779 memset(avail
, 0, content
->array
.raid_disks
);
783 for (i
=0; i
< bestcnt
; i
++) {
785 int event_margin
= 1; /* always allow a difference of '1'
786 * like the kernel does
789 /* note: we ignore error flags in multipath arrays
790 * as they don't make sense
792 if (content
->array
.level
!= LEVEL_MULTIPATH
)
793 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
794 if (!(devices
[j
].i
.disk
.state
795 & (1<<MD_DISK_FAULTY
)))
799 if (devices
[j
].i
.events
+event_margin
>=
800 devices
[most_recent
].i
.events
) {
801 devices
[j
].uptodate
= 1;
802 if (i
< content
->array
.raid_disks
) {
803 if (devices
[j
].i
.recovery_start
== MaxSector
) {
812 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
813 content
->array
.layout
, 1,
815 /* Choose the newest best drive which is
816 * not up-to-date, update the superblock
820 struct supertype
*tst
;
821 unsigned long long current_events
;
823 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
826 !devices
[j
].uptodate
&&
827 devices
[j
].i
.recovery_start
== MaxSector
&&
830 > devices
[chosen_drive
].i
.events
))
833 if (chosen_drive
< 0)
835 current_events
= devices
[chosen_drive
].i
.events
;
838 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
839 devices
[chosen_drive
].devname
,
840 devices
[chosen_drive
].i
.disk
.raid_disk
,
841 (int)(devices
[chosen_drive
].i
.events
),
842 (int)(devices
[most_recent
].i
.events
));
843 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
845 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
846 devices
[chosen_drive
].devname
);
847 devices
[chosen_drive
].i
.events
= 0;
851 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
853 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
854 devices
[chosen_drive
].devname
);
855 devices
[chosen_drive
].i
.events
= 0;
858 content
->events
= devices
[most_recent
].i
.events
;
859 tst
->ss
->update_super(tst
, content
, "force-one",
860 devices
[chosen_drive
].devname
, verbose
,
863 if (tst
->ss
->store_super(tst
, fd
)) {
865 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
866 devices
[chosen_drive
].devname
);
867 devices
[chosen_drive
].i
.events
= 0;
868 tst
->ss
->free_super(tst
);
872 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
873 devices
[chosen_drive
].uptodate
= 1;
874 avail
[chosen_drive
] = 1;
876 tst
->ss
->free_super(tst
);
878 /* If there are any other drives of the same vintage,
879 * add them in as well. We can't lose and we might gain
881 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
884 !devices
[j
].uptodate
&&
885 devices
[j
].i
.events
== current_events
) {
892 /* Now we want to look at the superblock which the kernel will base things on
893 * and compare the devices that we think are working with the devices that the
894 * superblock thinks are working.
895 * If there are differences and --force is given, then update this chosen
899 st
->ss
->free_super(st
);
900 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
906 if (!devices
[j
].uptodate
)
909 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
910 fprintf(stderr
, Name
": Cannot open %s: %s\n",
911 devices
[j
].devname
, strerror(errno
));
915 if (st
->ss
->load_super(st
,fd
, NULL
)) {
917 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
924 if (st
->sb
== NULL
) {
925 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
929 st
->ss
->getinfo_super(st
, content
);
931 sysfs_init(content
, mdfd
, 0);
933 for (i
=0; i
<bestcnt
; i
++) {
935 unsigned int desired_state
;
937 if (i
< content
->array
.raid_disks
)
938 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
944 if (!devices
[j
].uptodate
)
947 devices
[j
].i
.disk
.state
= desired_state
;
948 if (!(devices
[j
].i
.array
.state
& 1))
951 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
955 fprintf(stderr
, Name
": "
956 "clearing FAULTY flag for device %d in %s for %s\n",
957 j
, mddev
, devices
[j
].devname
);
961 fprintf(stderr
, Name
": "
962 "device %d in %s has wrong state in superblock, but %s seems ok\n",
963 i
, mddev
, devices
[j
].devname
);
967 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
968 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
973 if (force
&& !clean
&&
974 !enough(content
->array
.level
, content
->array
.raid_disks
,
975 content
->array
.layout
, clean
,
977 change
+= st
->ss
->update_super(st
, content
, "force-array",
978 devices
[chosen_drive
].devname
, verbose
,
985 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
987 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
988 devices
[chosen_drive
].devname
);
992 if (st
->ss
->store_super(st
, fd
)) {
994 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
995 devices
[chosen_drive
].devname
);
1002 /* If we are in the middle of a reshape we may need to restore saved data
1003 * that was moved aside due to the reshape overwriting live data
1004 * The code of doing this lives in Grow.c
1007 if (content
->reshape_active
) {
1009 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1011 fprintf(stderr
, Name
":%s has an active reshape - checking "
1012 "if critical section needs to be restored\n",
1014 for (i
=0; i
<bestcnt
; i
++) {
1017 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1018 if (fdlist
[i
] < 0) {
1019 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1020 devices
[j
].devname
);
1028 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1031 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1034 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1035 if (backup_file
== NULL
)
1036 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1042 /* count number of in-sync devices according to the superblock.
1043 * We must have this number to start the array without -s or -R
1045 req_cnt
= content
->array
.working_disks
;
1047 /* Almost ready to actually *do* something */
1051 /* First, fill in the map, so that udev can find our name
1052 * as soon as we become active.
1054 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1055 content
->uuid
, chosen_name
);
1057 rv
= set_array_info(mdfd
, st
, content
);
1059 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1060 mddev
, strerror(errno
));
1061 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1065 if (ident
->bitmap_fd
>= 0) {
1066 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1067 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1068 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1072 } else if (ident
->bitmap_file
) {
1073 /* From config file */
1074 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1076 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1077 ident
->bitmap_file
);
1078 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1082 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1083 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1085 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1092 /* First, add the raid disks, but add the chosen one last */
1093 for (i
=0; i
<= bestcnt
; i
++) {
1097 if (j
== chosen_drive
)
1102 if (j
>= 0 /* && devices[j].uptodate */) {
1103 int dfd
= dev_open(devices
[j
].devname
,
1106 remove_partitions(dfd
);
1109 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1112 fprintf(stderr
, Name
": failed to add "
1117 if (i
< content
->array
.raid_disks
1122 } else if (verbose
> 0)
1123 fprintf(stderr
, Name
": added %s "
1125 devices
[j
].devname
, mddev
,
1126 devices
[j
].i
.disk
.raid_disk
);
1127 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1128 fprintf(stderr
, Name
": no uptodate device for "
1133 if (content
->array
.level
== LEVEL_CONTAINER
) {
1135 fprintf(stderr
, Name
": Container %s has been "
1136 "assembled with %d drive%s",
1137 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1138 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1139 fprintf(stderr
, " (out of %d)",
1140 content
->array
.raid_disks
);
1141 fprintf(stderr
, "\n");
1143 sysfs_uevent(content
, "change");
1144 wait_for(chosen_name
, mdfd
);
1151 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1152 content
->array
.layout
, clean
, avail
, okcnt
) &&
1153 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1155 /* This array is good-to-go.
1156 * If a reshape is in progress then we might need to
1157 * continue monitoring it. In that case we start
1158 * it read-only and let the grow code make it writable.
1162 if (content
->reshape_active
&&
1163 content
->delta_disks
<= 0)
1164 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1167 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1170 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1171 mddev
, okcnt
, okcnt
==1?"":"s");
1172 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1173 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1175 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1177 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1178 fprintf(stderr
, ".\n");
1180 if (content
->reshape_active
&&
1181 content
->array
.level
>= 4 &&
1182 content
->array
.level
<= 6) {
1183 /* might need to increase the size
1184 * of the stripe cache - default is 256
1186 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1187 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1189 sysfs_set_num(sra
, NULL
,
1190 "stripe_cache_size",
1191 (4 * content
->array
.chunk_size
/ 4096) + 1);
1194 wait_for(mddev
, mdfd
);
1198 /* There is a nasty race with 'mdadm --monitor'.
1199 * If it opens this device before we close it,
1200 * it gets an incomplete open on which IO
1201 * doesn't work and the capacity is
1203 * If we reopen (to check for layered devices)
1204 * before --monitor closes, we loose.
1206 * So: wait upto 1 second for there to be
1207 * a non-zero capacity.
1209 while (usecs
< 1000) {
1210 mdfd
= open(mddev
, O_RDONLY
);
1212 unsigned long long size
;
1213 if (get_dev_size(mdfd
, NULL
, &size
) &&
1224 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1225 mddev
, strerror(errno
));
1227 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1228 content
->array
.layout
, 1, avail
, okcnt
))
1229 fprintf(stderr
, Name
": Not enough devices to "
1230 "start the array.\n");
1231 else if (!enough(content
->array
.level
,
1232 content
->array
.raid_disks
,
1233 content
->array
.layout
, clean
,
1235 fprintf(stderr
, Name
": Not enough devices to "
1236 "start the array while not clean "
1237 "- consider --force.\n");
1240 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1244 if (runstop
== -1) {
1245 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1246 mddev
, okcnt
, okcnt
==1?"":"s");
1247 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1248 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1249 fprintf(stderr
, ", but not started.\n");
1253 if (verbose
>= -1) {
1254 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1256 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1258 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1259 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1260 content
->array
.layout
, 1, avail
, okcnt
))
1261 fprintf(stderr
, " - not enough to start the array.\n");
1262 else if (!enough(content
->array
.level
,
1263 content
->array
.raid_disks
,
1264 content
->array
.layout
, clean
,
1266 fprintf(stderr
, " - not enough to start the "
1267 "array while not clean - consider "
1270 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1271 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1273 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1274 fprintf(stderr
, " (use --run to insist).\n");
1278 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1282 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1283 * been updated to point to the current locations of devices.
1284 * so we can just start the array
1287 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1288 devices
[chosen_drive
].i
.disk
.minor
);
1289 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1290 fprintf(stderr
, Name
": Cannot start array: %s\n",
1300 int assemble_container_content(struct supertype
*st
, int mdfd
,
1301 struct mdinfo
*content
, int runstop
,
1302 char *chosen_name
, int verbose
)
1304 struct mdinfo
*dev
, *sra
;
1305 int working
= 0, preexist
= 0;
1306 struct map_ent
*map
= NULL
;
1308 sysfs_init(content
, mdfd
, 0);
1310 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1311 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1312 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1319 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1320 if (sysfs_add_disk(content
, dev
, 1) == 0)
1322 else if (errno
== EEXIST
)
1326 return 1;/* Nothing new, don't try to start */
1329 map_update(&map
, fd2devnum(mdfd
),
1330 content
->text_version
,
1331 content
->uuid
, chosen_name
);
1334 (working
+ preexist
) >= content
->array
.working_disks
) {
1337 switch(content
->array
.level
) {
1339 case LEVEL_MULTIPATH
:
1341 err
= sysfs_set_str(content
, NULL
, "array_state",
1345 err
= sysfs_set_str(content
, NULL
, "array_state",
1347 /* start mdmon if needed. */
1349 if (!mdmon_running(st
->container_dev
))
1350 start_mdmon(st
->container_dev
);
1351 ping_monitor(devnum2devname(st
->container_dev
));
1356 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1359 fprintf(stderr
, Name
1360 ": array %s now has %d devices",
1361 chosen_name
, working
+ preexist
);
1363 fprintf(stderr
, Name
1364 ": Started %s with %d devices",
1365 chosen_name
, working
+ preexist
);
1367 fprintf(stderr
, " (%d new)", working
);
1368 fprintf(stderr
, "\n");
1371 wait_for(chosen_name
, mdfd
);
1374 /* FIXME should have an O_EXCL and wait for read-auto */
1377 fprintf(stderr
, Name
1378 ": %s assembled with %d devices but "
1380 chosen_name
, working
);