]>
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 int Assemble(struct supertype
*st
, char *mddev
, int mdfd
,
33 mddev_ident_t ident
, char *conffile
,
34 mddev_dev_t devlist
, char *backup_file
,
35 int readonly
, int runstop
,
36 char *update
, char *homehost
,
37 int verbose
, int force
)
40 * The task of Assemble is to find a collection of
41 * devices that should (according to their superblocks)
42 * form an array, and to give this collection to the MD driver.
43 * In Linux-2.4 and later, this involves submitting a
44 * SET_ARRAY_INFO ioctl with no arg - to prepare
45 * the array - and then submit a number of
46 * ADD_NEW_DISK ioctls to add disks into
47 * the array. Finally RUN_ARRAY might
48 * be submitted to start the array.
50 * Much of the work of Assemble is in finding and/or
51 * checking the disks to make sure they look right.
53 * If mddev is not set, then scan must be set and we
54 * read through the config file for dev+uuid mapping
55 * We recurse, setting mddev, for each device that
57 * - has a valid uuid (or any uuid if !uuidset)
59 * If mddev is set, we try to determine state of md.
60 * check version - must be at least 0.90.0
61 * check kernel version. must be at least 2.4.
62 * If not, we can possibly fall back on START_ARRAY
63 * Try to GET_ARRAY_INFO.
64 * If possible, give up
65 * If not, try to STOP_ARRAY just to make sure
67 * If !uuidset and scan, look in conf-file for uuid
68 * If not found, give up
69 * If !devlist and scan and uuidset, get list of devs from conf-file
72 * Check superblock - discard if bad
73 * Check uuid (set if we don't have one) - discard if no match
74 * Check superblock similarity if we have a superblock - discard if different
75 * Record events, devicenum
76 * This should give us a list of devices for the array
77 * We should collect the most recent event number
79 * Count disks with recent enough event count
80 * While force && !enough disks
81 * Choose newest rejected disks, update event count
82 * mark clean and rewrite superblock
85 * foreach device with recent events : ADD_NEW_DISK
86 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
88 * Check the device numbers in superblock are right
89 * update superblock if any changes
95 void *first_super
= NULL
, *super
= NULL
;
98 unsigned int major
, minor
;
99 unsigned int oldmajor
, oldminor
;
106 int *best
= NULL
; /* indexed by raid_disk */
107 unsigned int bestcnt
= 0;
109 unsigned int okcnt
, sparecnt
;
110 unsigned int req_cnt
;
116 int start_partial_ok
= (runstop
>= 0) && (force
|| devlist
==NULL
);
117 unsigned int num_devs
;
123 vers
= md_get_version(mdfd
);
125 fprintf(stderr
, Name
": %s appears not to be an md device.\n", mddev
);
129 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
130 " Upgrade your kernel or try --build\n");
133 if (get_linux_version() < 2004000)
136 if (ioctl(mdfd
, GET_ARRAY_INFO
, &info
.array
)>=0) {
137 fprintf(stderr
, Name
": device %s already active - cannot assemble it\n",
141 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
144 * If any subdevs are listed, then any that don't
145 * match ident are discarded. Remainder must all match and
147 * If no subdevs, then we scan all devices in the config file, but
148 * there must be something in the identity
152 ident
->uuid_set
== 0 &&
153 ident
->super_minor
< 0 &&
154 ident
->devices
== NULL
) {
155 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
160 devlist
= conf_get_devs(conffile
);
163 tmpdev
= devlist
; num_devs
= 0;
166 tmpdev
= tmpdev
->next
;
168 devices
= malloc(num_devs
* sizeof(*devices
));
170 if (!st
&& ident
->st
) st
= ident
->st
;
173 fprintf(stderr
, Name
": looking for devices for %s\n",
180 struct supertype
*tst
= st
;
182 devname
= devlist
->devname
;
183 devlist
= devlist
->next
;
185 if (ident
->devices
&&
186 !match_oneof(ident
->devices
, devname
)) {
187 if ((inargv
&& verbose
>=0) || verbose
> 0)
188 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
197 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
199 if ((inargv
&& verbose
>= 0) || verbose
> 0)
200 fprintf(stderr
, Name
": cannot open device %s: %s\n",
201 devname
, strerror(errno
));
202 } else if (fstat(dfd
, &stb
)< 0) {
204 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
205 devname
, strerror(errno
));
206 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
207 fprintf(stderr
, Name
": %s is not a block device.\n",
209 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
210 if ((inargv
&& verbose
>= 0) || verbose
> 0)
211 fprintf(stderr
, Name
": no recogniseable superblock\n");
212 } else if (tst
->ss
->load_super(tst
,dfd
, &super
, NULL
)) {
213 if ((inargv
&& verbose
>= 0) || verbose
> 0)
214 fprintf( stderr
, Name
": no RAID superblock on %s\n",
217 tst
->ss
->getinfo_super(&info
, super
);
219 if (dfd
>= 0) close(dfd
);
221 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
222 (!super
|| same_uuid(info
.uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0)) {
223 if ((inargv
&& verbose
>= 0) || verbose
> 0)
224 fprintf(stderr
, Name
": %s has wrong uuid.\n",
228 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
229 (!super
|| strncmp(info
.name
, ident
->name
, 32)!=0)) {
230 if ((inargv
&& verbose
>= 0) || verbose
> 0)
231 fprintf(stderr
, Name
": %s has wrong name.\n",
235 if (ident
->super_minor
!= UnSet
&&
236 (!super
|| ident
->super_minor
!= info
.array
.md_minor
)) {
237 if ((inargv
&& verbose
>= 0) || verbose
> 0)
238 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
242 if (ident
->level
!= UnSet
&&
243 (!super
|| ident
->level
!= info
.array
.level
)) {
244 if ((inargv
&& verbose
>= 0) || verbose
> 0)
245 fprintf(stderr
, Name
": %s has wrong raid level.\n",
249 if (ident
->raid_disks
!= UnSet
&&
250 (!super
|| ident
->raid_disks
!= info
.array
.raid_disks
)) {
251 if ((inargv
&& verbose
>= 0) || verbose
> 0)
252 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
257 /* If we are this far, then we are commited to this device.
258 * If the super_block doesn't exist, or doesn't match others,
259 * then we cannot continue
263 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
270 st
= tst
; /* commit to this format, if haven't already */
271 if (st
->ss
->compare_super(&first_super
, super
)) {
272 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
279 /* looks like a good enough match to update the super block if needed */
281 /* prepare useful information in info structures */
284 info
.array
.md_minor
= minor(stb2
.st_rdev
);
286 if (strcmp(update
, "uuid")==0 &&
289 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
290 read(rfd
, ident
->uuid
, 16) != 16) {
291 *(__u32
*)(ident
->uuid
) = random();
292 *(__u32
*)(ident
->uuid
+1) = random();
293 *(__u32
*)(ident
->uuid
+2) = random();
294 *(__u32
*)(ident
->uuid
+3) = random();
296 if (rfd
>= 0) close(rfd
);
298 memcpy(info
.uuid
, ident
->uuid
, 16);
299 strcpy(info
.name
, ident
->name
);
300 st
->ss
->update_super(&info
, super
, update
, devname
, verbose
,
301 ident
->uuid_set
, homehost
);
302 if (strcmp(update
, "uuid")==0 &&
305 memcpy(ident
->uuid
, info
.uuid
, 16);
307 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
309 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
311 else if (st
->ss
->store_super(st
, dfd
, super
))
312 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
317 if (strcmp(update
, "uuid")==0 &&
319 bitmap_update_uuid(ident
->bitmap_fd
, info
.uuid
);
323 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
324 devname
, mddev
, info
.disk
.raid_disk
);
325 devices
[devcnt
].devname
= devname
;
326 devices
[devcnt
].major
= major(stb
.st_rdev
);
327 devices
[devcnt
].minor
= minor(stb
.st_rdev
);
328 devices
[devcnt
].oldmajor
= info
.disk
.major
;
329 devices
[devcnt
].oldminor
= info
.disk
.minor
;
330 devices
[devcnt
].events
= info
.events
;
331 devices
[devcnt
].raid_disk
= info
.disk
.raid_disk
;
332 devices
[devcnt
].disk_nr
= info
.disk
.number
;
333 devices
[devcnt
].uptodate
= 0;
334 devices
[devcnt
].state
= info
.disk
.state
;
335 if (most_recent
< devcnt
) {
336 if (devices
[devcnt
].events
337 > devices
[most_recent
].events
)
338 most_recent
= devcnt
;
340 if (info
.array
.level
== -4)
341 /* with multipath, the raid_disk from the superblock is meaningless */
344 i
= devices
[devcnt
].raid_disk
;
346 if (nextspare
< info
.array
.raid_disks
)
347 nextspare
= info
.array
.raid_disks
;
352 unsigned int newbestcnt
= i
+10;
353 int *newbest
= malloc(sizeof(int)*newbestcnt
);
355 for (c
=0; c
< newbestcnt
; c
++)
357 newbest
[c
] = best
[c
];
362 bestcnt
= newbestcnt
;
365 || devices
[best
[i
]].events
< devices
[devcnt
].events
)
375 if (update
&& strcmp(update
, "byteorder")==0)
376 st
->minor_version
= 90;
379 fprintf(stderr
, Name
": no devices found for %s\n",
385 st
->ss
->getinfo_super(&info
, first_super
);
387 /* now we have some devices that might be suitable.
390 avail
= malloc(info
.array
.raid_disks
);
391 memset(avail
, 0, info
.array
.raid_disks
);
394 for (i
=0; i
< bestcnt
;i
++) {
396 int event_margin
= 1; /* always allow a difference of '1'
397 * like the kernel does
400 /* note: we ignore error flags in multipath arrays
401 * as they don't make sense
403 if (info
.array
.level
!= -4)
404 if (!(devices
[j
].state
& (1<<MD_DISK_SYNC
))) {
405 if (!(devices
[j
].state
& (1<<MD_DISK_FAULTY
)))
409 if (devices
[j
].events
+event_margin
>=
410 devices
[most_recent
].events
) {
411 devices
[j
].uptodate
= 1;
412 if (i
< info
.array
.raid_disks
) {
419 while (force
&& !enough(info
.array
.level
, info
.array
.raid_disks
,
422 /* Choose the newest best drive which is
423 * not up-to-date, update the superblock
428 for (i
=0; i
<info
.array
.raid_disks
&& i
< bestcnt
; i
++) {
431 !devices
[j
].uptodate
&&
432 devices
[j
].events
> 0 &&
434 devices
[j
].events
> devices
[chosen_drive
].events
))
437 if (chosen_drive
< 0)
440 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
441 devices
[chosen_drive
].devname
, devices
[chosen_drive
].raid_disk
,
442 (int)(devices
[chosen_drive
].events
),
443 (int)(devices
[most_recent
].events
));
444 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
446 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
447 devices
[chosen_drive
].devname
);
448 devices
[chosen_drive
].events
= 0;
451 if (st
->ss
->load_super(st
,fd
, &super
, NULL
)) {
453 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
454 devices
[chosen_drive
].devname
);
455 devices
[chosen_drive
].events
= 0;
458 info
.events
= devices
[most_recent
].events
;
459 st
->ss
->update_super(&info
, super
, "force", devices
[chosen_drive
].devname
, verbose
, 0, NULL
);
461 if (st
->ss
->store_super(st
, fd
, super
)) {
463 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
464 devices
[chosen_drive
].devname
);
465 devices
[chosen_drive
].events
= 0;
470 devices
[chosen_drive
].events
= devices
[most_recent
].events
;
471 devices
[chosen_drive
].uptodate
= 1;
472 avail
[chosen_drive
] = 1;
477 /* Now we want to look at the superblock which the kernel will base things on
478 * and compare the devices that we think are working with the devices that the
479 * superblock thinks are working.
480 * If there are differences and --force is given, then update this chosen
485 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
491 if (!devices
[j
].uptodate
)
494 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
495 fprintf(stderr
, Name
": Cannot open %s: %s\n",
496 devices
[j
].devname
, strerror(errno
));
499 if (st
->ss
->load_super(st
,fd
, &super
, NULL
)) {
501 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
508 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
511 st
->ss
->getinfo_super(&info
, super
);
512 for (i
=0; i
<bestcnt
; i
++) {
514 unsigned int desired_state
;
516 if (i
< info
.array
.raid_disks
)
517 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
523 if (!devices
[j
].uptodate
)
525 info
.disk
.number
= devices
[j
].disk_nr
;
526 info
.disk
.raid_disk
= i
;
527 info
.disk
.state
= desired_state
;
529 if (devices
[j
].uptodate
&&
530 st
->ss
->update_super(&info
, super
, "assemble", NULL
, verbose
, 0, NULL
)) {
533 fprintf(stderr
, Name
": "
534 "clearing FAULTY flag for device %d in %s for %s\n",
535 j
, mddev
, devices
[j
].devname
);
539 fprintf(stderr
, Name
": "
540 "device %d in %s has wrong state in superblock, but %s seems ok\n",
541 i
, mddev
, devices
[j
].devname
);
545 if (!devices
[j
].uptodate
&&
546 !(super
.disks
[i
].state
& (1 << MD_DISK_FAULTY
))) {
547 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
552 if (force
&& okcnt
== info
.array
.raid_disks
-1) {
553 /* FIXME check event count */
554 change
+= st
->ss
->update_super(&info
, super
, "force",
555 devices
[chosen_drive
].devname
, verbose
, 0, NULL
);
560 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
562 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
563 devices
[chosen_drive
].devname
);
566 if (st
->ss
->store_super(st
, fd
, super
)) {
568 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
569 devices
[chosen_drive
].devname
);
575 /* If we are in the middle of a reshape we may need to restore saved data
576 * that was moved aside due to the reshape overwriting live data
577 * The code of doing this lives in Grow.c
580 if (info
.reshape_active
) {
582 int *fdlist
= malloc(sizeof(int)* bestcnt
);
583 for (i
=0; i
<bestcnt
; i
++) {
586 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
588 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
597 err
= Grow_restart(st
, &info
, fdlist
, bestcnt
, backup_file
);
600 if (fdlist
[i
]>=0) close(fdlist
[i
]);
603 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
608 /* count number of in-sync devices according to the superblock.
609 * We must have this number to start the array without -s or -R
611 req_cnt
= info
.array
.working_disks
;
613 /* Almost ready to actually *do* something */
616 if ((vers
% 100) >= 1) { /* can use different versions */
617 mdu_array_info_t inf
;
618 memset(&inf
, 0, sizeof(inf
));
619 inf
.major_version
= st
->ss
->major
;
620 inf
.minor_version
= st
->minor_version
;
621 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
623 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
626 fprintf(stderr
, Name
": SET_ARRAY_INFO failed for %s: %s\n",
627 mddev
, strerror(errno
));
630 if (ident
->bitmap_fd
>= 0) {
631 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
632 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
635 } else if (ident
->bitmap_file
) {
636 /* From config file */
637 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
639 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
643 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
644 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
651 /* First, add the raid disks, but add the chosen one last */
652 for (i
=0; i
<= bestcnt
; i
++) {
656 if (j
== chosen_drive
)
661 if (j
>= 0 /* && devices[j].uptodate */) {
662 mdu_disk_info_t disk
;
663 memset(&disk
, 0, sizeof(disk
));
664 disk
.major
= devices
[j
].major
;
665 disk
.minor
= devices
[j
].minor
;
666 if (ioctl(mdfd
, ADD_NEW_DISK
, &disk
)!=0) {
667 fprintf(stderr
, Name
": failed to add %s to %s: %s\n",
671 if (i
< info
.array
.raid_disks
|| i
== bestcnt
)
675 } else if (verbose
> 0)
676 fprintf(stderr
, Name
": added %s to %s as %d\n",
677 devices
[j
].devname
, mddev
, devices
[j
].raid_disk
);
678 } else if (verbose
> 0 && i
< info
.array
.raid_disks
)
679 fprintf(stderr
, Name
": no uptodate device for slot %d of %s\n",
685 ( enough(info
.array
.level
, info
.array
.raid_disks
, info
.array
.layout
, avail
, okcnt
) &&
686 (okcnt
>= req_cnt
|| start_partial_ok
)
688 if (ioctl(mdfd
, RUN_ARRAY
, NULL
)==0) {
690 fprintf(stderr
, Name
": %s has been started with %d drive%s",
691 mddev
, okcnt
, okcnt
==1?"":"s");
692 if (okcnt
< info
.array
.raid_disks
)
693 fprintf(stderr
, " (out of %d)", info
.array
.raid_disks
);
695 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
696 fprintf(stderr
, ".\n");
700 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
701 mddev
, strerror(errno
));
705 fprintf(stderr
, Name
": %s assembled from %d drive%s",
706 mddev
, okcnt
, okcnt
==1?"":"s");
707 if (okcnt
!= info
.array
.raid_disks
)
708 fprintf(stderr
, " (out of %d)", info
.array
.raid_disks
);
709 fprintf(stderr
, ", but not started.\n");
713 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
715 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
716 if (!enough(info
.array
.level
, info
.array
.raid_disks
, info
.array
.layout
, avail
, okcnt
))
717 fprintf(stderr
, " - not enough to start the array.\n");
719 if (req_cnt
== info
.array
.raid_disks
)
720 fprintf(stderr
, " - need all %d to start it", req_cnt
);
722 fprintf(stderr
, " - need %d of %d to start", req_cnt
, info
.array
.raid_disks
);
723 fprintf(stderr
, " (use --run to insist).\n");
728 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
729 * been updated to point to the current locations of devices.
730 * so we can just start the array
733 dev
= makedev(devices
[chosen_drive
].major
,
734 devices
[chosen_drive
].minor
);
735 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
736 fprintf(stderr
, Name
": Cannot start array: %s\n",