]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
32 static int name_matches(char *found
, char *required
, char *homehost
)
34 /* See if the name found matches the required name, possibly
35 * prefixed with 'homehost'
39 strncpy(fnd
, found
, 32);
41 if (strcmp(found
, required
)==0)
44 int l
= strlen(homehost
);
45 if (l
< 32 && fnd
[l
] == ':' &&
46 strcmp(fnd
+l
+1, required
)==0)
52 int Assemble(struct supertype
*st
, char *mddev
, int mdfd
,
53 mddev_ident_t ident
, char *conffile
,
54 mddev_dev_t devlist
, char *backup_file
,
55 int readonly
, int runstop
,
56 char *update
, char *homehost
,
57 int verbose
, int force
)
60 * The task of Assemble is to find a collection of
61 * devices that should (according to their superblocks)
62 * form an array, and to give this collection to the MD driver.
63 * In Linux-2.4 and later, this involves submitting a
64 * SET_ARRAY_INFO ioctl with no arg - to prepare
65 * the array - and then submit a number of
66 * ADD_NEW_DISK ioctls to add disks into
67 * the array. Finally RUN_ARRAY might
68 * be submitted to start the array.
70 * Much of the work of Assemble is in finding and/or
71 * checking the disks to make sure they look right.
73 * If mddev is not set, then scan must be set and we
74 * read through the config file for dev+uuid mapping
75 * We recurse, setting mddev, for each device that
77 * - has a valid uuid (or any uuid if !uuidset)
79 * If mddev is set, we try to determine state of md.
80 * check version - must be at least 0.90.0
81 * check kernel version. must be at least 2.4.
82 * If not, we can possibly fall back on START_ARRAY
83 * Try to GET_ARRAY_INFO.
84 * If possible, give up
85 * If not, try to STOP_ARRAY just to make sure
87 * If !uuidset and scan, look in conf-file for uuid
88 * If not found, give up
89 * If !devlist and scan and uuidset, get list of devs from conf-file
92 * Check superblock - discard if bad
93 * Check uuid (set if we don't have one) - discard if no match
94 * Check superblock similarity if we have a superblock - discard if different
95 * Record events, devicenum
96 * This should give us a list of devices for the array
97 * We should collect the most recent event number
99 * Count disks with recent enough event count
100 * While force && !enough disks
101 * Choose newest rejected disks, update event count
102 * mark clean and rewrite superblock
105 * foreach device with recent events : ADD_NEW_DISK
106 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
108 * Check the device numbers in superblock are right
109 * update superblock if any changes
115 void *first_super
= NULL
, *super
= NULL
;
118 unsigned int major
, minor
;
119 unsigned int oldmajor
, oldminor
;
126 int *best
= NULL
; /* indexed by raid_disk */
127 unsigned int bestcnt
= 0;
129 unsigned int okcnt
, sparecnt
;
130 unsigned int req_cnt
;
136 int start_partial_ok
= (runstop
>= 0) && (force
|| devlist
==NULL
);
137 unsigned int num_devs
;
143 vers
= md_get_version(mdfd
);
145 fprintf(stderr
, Name
": %s appears not to be an md device.\n", mddev
);
149 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
150 " Upgrade your kernel or try --build\n");
153 if (get_linux_version() < 2004000)
156 if (ioctl(mdfd
, GET_ARRAY_INFO
, &info
.array
)>=0) {
157 fprintf(stderr
, Name
": device %s already active - cannot assemble it\n",
161 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
164 * If any subdevs are listed, then any that don't
165 * match ident are discarded. Remainder must all match and
167 * If no subdevs, then we scan all devices in the config file, but
168 * there must be something in the identity
172 ident
->uuid_set
== 0 &&
173 ident
->super_minor
< 0 &&
174 ident
->devices
== NULL
) {
175 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
180 devlist
= conf_get_devs(conffile
);
183 tmpdev
= devlist
; num_devs
= 0;
186 tmpdev
= tmpdev
->next
;
188 devices
= malloc(num_devs
* sizeof(*devices
));
190 if (!st
&& ident
->st
) st
= ident
->st
;
193 fprintf(stderr
, Name
": looking for devices for %s\n",
196 /* first walk the list of devices to find a consistent set
197 * that match the criterea, if that is possible.
198 * We flag the one we like with 'used'.
200 for (tmpdev
= devlist
;
202 tmpdev
= tmpdev
->next
) {
203 char *devname
= tmpdev
->devname
;
206 struct supertype
*tst
= st
;
208 if (ident
->devices
&&
209 !match_oneof(ident
->devices
, devname
)) {
210 if ((inargv
&& verbose
>=0) || verbose
> 0)
211 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
220 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
222 if ((inargv
&& verbose
>= 0) || verbose
> 0)
223 fprintf(stderr
, Name
": cannot open device %s: %s\n",
224 devname
, strerror(errno
));
225 } else if (fstat(dfd
, &stb
)< 0) {
227 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
228 devname
, strerror(errno
));
229 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
230 fprintf(stderr
, Name
": %s is not a block device.\n",
232 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
233 if ((inargv
&& verbose
>= 0) || verbose
> 0)
234 fprintf(stderr
, Name
": no recogniseable superblock\n");
235 } else if (tst
->ss
->load_super(tst
,dfd
, &super
, NULL
)) {
236 if ((inargv
&& verbose
>= 0) || verbose
> 0)
237 fprintf( stderr
, Name
": no RAID superblock on %s\n",
240 tst
->ss
->getinfo_super(&info
, super
);
242 if (dfd
>= 0) close(dfd
);
244 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
245 (!super
|| same_uuid(info
.uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0)) {
246 if ((inargv
&& verbose
>= 0) || verbose
> 0)
247 fprintf(stderr
, Name
": %s has wrong uuid.\n",
251 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
252 (!super
|| name_matches(info
.name
, ident
->name
, homehost
)==0)) {
253 if ((inargv
&& verbose
>= 0) || verbose
> 0)
254 fprintf(stderr
, Name
": %s has wrong name.\n",
258 if (ident
->super_minor
!= UnSet
&&
259 (!super
|| ident
->super_minor
!= info
.array
.md_minor
)) {
260 if ((inargv
&& verbose
>= 0) || verbose
> 0)
261 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
265 if (ident
->level
!= UnSet
&&
266 (!super
|| ident
->level
!= info
.array
.level
)) {
267 if ((inargv
&& verbose
>= 0) || verbose
> 0)
268 fprintf(stderr
, Name
": %s has wrong raid level.\n",
272 if (ident
->raid_disks
!= UnSet
&&
273 (!super
|| ident
->raid_disks
!= info
.array
.raid_disks
)) {
274 if ((inargv
&& verbose
>= 0) || verbose
> 0)
275 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
280 /* If we are this far, then we are commited to this device.
281 * If the super_block doesn't exist, or doesn't match others,
282 * then we cannot continue
286 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
293 st
= tst
; /* commit to this format, if haven't already */
294 if (st
->ss
->compare_super(&first_super
, super
)) {
295 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
305 /* Ok, no bad inconsistancy, we can try updating etc */
306 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
) {
307 char *devname
= tmpdev
->devname
;
309 /* looks like a good enough match to update the super block if needed */
312 /* prepare useful information in info structures */
316 if (strcmp(update
, "uuid")==0 &&
319 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
320 read(rfd
, ident
->uuid
, 16) != 16) {
321 *(__u32
*)(ident
->uuid
) = random();
322 *(__u32
*)(ident
->uuid
+1) = random();
323 *(__u32
*)(ident
->uuid
+2) = random();
324 *(__u32
*)(ident
->uuid
+3) = random();
326 if (rfd
>= 0) close(rfd
);
328 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
335 st
->ss
->load_super(st
, dfd
, &super
, NULL
);
336 st
->ss
->getinfo_super(&info
, super
);
338 memcpy(info
.uuid
, ident
->uuid
, 16);
339 strcpy(info
.name
, ident
->name
);
340 info
.array
.md_minor
= minor(stb2
.st_rdev
);
342 st
->ss
->update_super(&info
, super
, update
, devname
, verbose
,
343 ident
->uuid_set
, homehost
);
344 if (strcmp(update
, "uuid")==0 &&
347 memcpy(ident
->uuid
, info
.uuid
, 16);
350 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
352 else if (st
->ss
->store_super(st
, dfd
, super
))
353 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
358 if (strcmp(update
, "uuid")==0 &&
360 bitmap_update_uuid(ident
->bitmap_fd
, info
.uuid
);
366 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
367 devname
, mddev
, info
.disk
.raid_disk
);
368 devices
[devcnt
].devname
= devname
;
369 devices
[devcnt
].major
= major(stb
.st_rdev
);
370 devices
[devcnt
].minor
= minor(stb
.st_rdev
);
371 devices
[devcnt
].oldmajor
= info
.disk
.major
;
372 devices
[devcnt
].oldminor
= info
.disk
.minor
;
373 devices
[devcnt
].events
= info
.events
;
374 devices
[devcnt
].raid_disk
= info
.disk
.raid_disk
;
375 devices
[devcnt
].disk_nr
= info
.disk
.number
;
376 devices
[devcnt
].uptodate
= 0;
377 devices
[devcnt
].state
= info
.disk
.state
;
378 if (most_recent
< devcnt
) {
379 if (devices
[devcnt
].events
380 > devices
[most_recent
].events
)
381 most_recent
= devcnt
;
383 if (info
.array
.level
== -4)
384 /* with multipath, the raid_disk from the superblock is meaningless */
387 i
= devices
[devcnt
].raid_disk
;
389 if (nextspare
< info
.array
.raid_disks
)
390 nextspare
= info
.array
.raid_disks
;
395 unsigned int newbestcnt
= i
+10;
396 int *newbest
= malloc(sizeof(int)*newbestcnt
);
398 for (c
=0; c
< newbestcnt
; c
++)
400 newbest
[c
] = best
[c
];
405 bestcnt
= newbestcnt
;
408 || devices
[best
[i
]].events
< devices
[devcnt
].events
)
418 if (update
&& strcmp(update
, "byteorder")==0)
419 st
->minor_version
= 90;
422 fprintf(stderr
, Name
": no devices found for %s\n",
428 st
->ss
->getinfo_super(&info
, first_super
);
430 /* now we have some devices that might be suitable.
433 avail
= malloc(info
.array
.raid_disks
);
434 memset(avail
, 0, info
.array
.raid_disks
);
437 for (i
=0; i
< bestcnt
;i
++) {
439 int event_margin
= 1; /* always allow a difference of '1'
440 * like the kernel does
443 /* note: we ignore error flags in multipath arrays
444 * as they don't make sense
446 if (info
.array
.level
!= -4)
447 if (!(devices
[j
].state
& (1<<MD_DISK_SYNC
))) {
448 if (!(devices
[j
].state
& (1<<MD_DISK_FAULTY
)))
452 if (devices
[j
].events
+event_margin
>=
453 devices
[most_recent
].events
) {
454 devices
[j
].uptodate
= 1;
455 if (i
< info
.array
.raid_disks
) {
462 while (force
&& !enough(info
.array
.level
, info
.array
.raid_disks
,
465 /* Choose the newest best drive which is
466 * not up-to-date, update the superblock
471 for (i
=0; i
<info
.array
.raid_disks
&& i
< bestcnt
; i
++) {
474 !devices
[j
].uptodate
&&
475 devices
[j
].events
> 0 &&
477 devices
[j
].events
> devices
[chosen_drive
].events
))
480 if (chosen_drive
< 0)
483 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
484 devices
[chosen_drive
].devname
, devices
[chosen_drive
].raid_disk
,
485 (int)(devices
[chosen_drive
].events
),
486 (int)(devices
[most_recent
].events
));
487 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
489 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
490 devices
[chosen_drive
].devname
);
491 devices
[chosen_drive
].events
= 0;
494 if (st
->ss
->load_super(st
,fd
, &super
, NULL
)) {
496 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
497 devices
[chosen_drive
].devname
);
498 devices
[chosen_drive
].events
= 0;
501 info
.events
= devices
[most_recent
].events
;
502 st
->ss
->update_super(&info
, super
, "force", devices
[chosen_drive
].devname
, verbose
, 0, NULL
);
504 if (st
->ss
->store_super(st
, fd
, super
)) {
506 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
507 devices
[chosen_drive
].devname
);
508 devices
[chosen_drive
].events
= 0;
513 devices
[chosen_drive
].events
= devices
[most_recent
].events
;
514 devices
[chosen_drive
].uptodate
= 1;
515 avail
[chosen_drive
] = 1;
520 /* Now we want to look at the superblock which the kernel will base things on
521 * and compare the devices that we think are working with the devices that the
522 * superblock thinks are working.
523 * If there are differences and --force is given, then update this chosen
528 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
534 if (!devices
[j
].uptodate
)
537 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
538 fprintf(stderr
, Name
": Cannot open %s: %s\n",
539 devices
[j
].devname
, strerror(errno
));
542 if (st
->ss
->load_super(st
,fd
, &super
, NULL
)) {
544 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
551 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
554 st
->ss
->getinfo_super(&info
, super
);
555 for (i
=0; i
<bestcnt
; i
++) {
557 unsigned int desired_state
;
559 if (i
< info
.array
.raid_disks
)
560 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
566 if (!devices
[j
].uptodate
)
568 info
.disk
.number
= devices
[j
].disk_nr
;
569 info
.disk
.raid_disk
= i
;
570 info
.disk
.state
= desired_state
;
572 if (devices
[j
].uptodate
&&
573 st
->ss
->update_super(&info
, super
, "assemble", NULL
, verbose
, 0, NULL
)) {
576 fprintf(stderr
, Name
": "
577 "clearing FAULTY flag for device %d in %s for %s\n",
578 j
, mddev
, devices
[j
].devname
);
582 fprintf(stderr
, Name
": "
583 "device %d in %s has wrong state in superblock, but %s seems ok\n",
584 i
, mddev
, devices
[j
].devname
);
588 if (!devices
[j
].uptodate
&&
589 !(super
.disks
[i
].state
& (1 << MD_DISK_FAULTY
))) {
590 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
595 if (force
&& okcnt
== info
.array
.raid_disks
-1) {
596 /* FIXME check event count */
597 change
+= st
->ss
->update_super(&info
, super
, "force",
598 devices
[chosen_drive
].devname
, verbose
, 0, NULL
);
603 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
605 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
606 devices
[chosen_drive
].devname
);
609 if (st
->ss
->store_super(st
, fd
, super
)) {
611 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
612 devices
[chosen_drive
].devname
);
618 /* If we are in the middle of a reshape we may need to restore saved data
619 * that was moved aside due to the reshape overwriting live data
620 * The code of doing this lives in Grow.c
623 if (info
.reshape_active
) {
625 int *fdlist
= malloc(sizeof(int)* bestcnt
);
626 for (i
=0; i
<bestcnt
; i
++) {
629 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
631 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
640 err
= Grow_restart(st
, &info
, fdlist
, bestcnt
, backup_file
);
643 if (fdlist
[i
]>=0) close(fdlist
[i
]);
646 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
651 /* count number of in-sync devices according to the superblock.
652 * We must have this number to start the array without -s or -R
654 req_cnt
= info
.array
.working_disks
;
656 /* Almost ready to actually *do* something */
659 if ((vers
% 100) >= 1) { /* can use different versions */
660 mdu_array_info_t inf
;
661 memset(&inf
, 0, sizeof(inf
));
662 inf
.major_version
= st
->ss
->major
;
663 inf
.minor_version
= st
->minor_version
;
664 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
666 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
669 fprintf(stderr
, Name
": SET_ARRAY_INFO failed for %s: %s\n",
670 mddev
, strerror(errno
));
673 if (ident
->bitmap_fd
>= 0) {
674 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
675 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
678 } else if (ident
->bitmap_file
) {
679 /* From config file */
680 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
682 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
686 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
687 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
694 /* First, add the raid disks, but add the chosen one last */
695 for (i
=0; i
<= bestcnt
; i
++) {
699 if (j
== chosen_drive
)
704 if (j
>= 0 /* && devices[j].uptodate */) {
705 mdu_disk_info_t disk
;
706 memset(&disk
, 0, sizeof(disk
));
707 disk
.major
= devices
[j
].major
;
708 disk
.minor
= devices
[j
].minor
;
709 if (ioctl(mdfd
, ADD_NEW_DISK
, &disk
)!=0) {
710 fprintf(stderr
, Name
": failed to add %s to %s: %s\n",
714 if (i
< info
.array
.raid_disks
|| i
== bestcnt
)
718 } else if (verbose
> 0)
719 fprintf(stderr
, Name
": added %s to %s as %d\n",
720 devices
[j
].devname
, mddev
, devices
[j
].raid_disk
);
721 } else if (verbose
> 0 && i
< info
.array
.raid_disks
)
722 fprintf(stderr
, Name
": no uptodate device for slot %d of %s\n",
728 ( enough(info
.array
.level
, info
.array
.raid_disks
, info
.array
.layout
, avail
, okcnt
) &&
729 (okcnt
>= req_cnt
|| start_partial_ok
)
731 if (ioctl(mdfd
, RUN_ARRAY
, NULL
)==0) {
733 fprintf(stderr
, Name
": %s has been started with %d drive%s",
734 mddev
, okcnt
, okcnt
==1?"":"s");
735 if (okcnt
< info
.array
.raid_disks
)
736 fprintf(stderr
, " (out of %d)", info
.array
.raid_disks
);
738 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
739 fprintf(stderr
, ".\n");
743 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
744 mddev
, strerror(errno
));
748 fprintf(stderr
, Name
": %s assembled from %d drive%s",
749 mddev
, okcnt
, okcnt
==1?"":"s");
750 if (okcnt
!= info
.array
.raid_disks
)
751 fprintf(stderr
, " (out of %d)", info
.array
.raid_disks
);
752 fprintf(stderr
, ", but not started.\n");
756 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
758 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
759 if (!enough(info
.array
.level
, info
.array
.raid_disks
, info
.array
.layout
, avail
, okcnt
))
760 fprintf(stderr
, " - not enough to start the array.\n");
762 if (req_cnt
== info
.array
.raid_disks
)
763 fprintf(stderr
, " - need all %d to start it", req_cnt
);
765 fprintf(stderr
, " - need %d of %d to start", req_cnt
, info
.array
.raid_disks
);
766 fprintf(stderr
, " (use --run to insist).\n");
771 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
772 * been updated to point to the current locations of devices.
773 * so we can just start the array
776 dev
= makedev(devices
[chosen_drive
].major
,
777 devices
[chosen_drive
].minor
);
778 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
779 fprintf(stderr
, Name
": Cannot start array: %s\n",