]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
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
,
35 int readonly
, int runstop
,
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
= force
|| devlist
==NULL
;
117 unsigned int num_devs
;
120 struct mddev_ident_s ident2
;
124 vers
= md_get_version(mdfd
);
126 fprintf(stderr
, Name
": %s appears not to be an md device.\n", mddev
);
130 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
131 " Upgrade your kernel or try --build\n");
134 if (get_linux_version() < 2004000)
137 if (ioctl(mdfd
, GET_ARRAY_INFO
, &info
.array
)>=0) {
138 fprintf(stderr
, Name
": device %s already active - cannot assemble it\n",
142 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
145 * If any subdevs are listed, then any that don't
146 * match ident are discarded. Remainder must all match and
148 * If no subdevs, then we scan all devices in the config file, but
149 * there must be something in the identity
153 ident
->uuid_set
== 0 &&
154 ident
->super_minor
< 0 &&
155 ident
->devices
== NULL
) {
156 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
161 devlist
= conf_get_devs(conffile
);
164 tmpdev
= devlist
; num_devs
= 0;
167 tmpdev
= tmpdev
->next
;
169 devices
= malloc(num_devs
* sizeof(*devices
));
171 if (!st
&& ident
->st
) st
= ident
->st
;
174 fprintf(stderr
, Name
": looking for devices for %s\n",
181 struct supertype
*tst
= st
;
183 devname
= devlist
->devname
;
184 devlist
= devlist
->next
;
186 if (ident
->devices
&&
187 !match_oneof(ident
->devices
, devname
)) {
188 if ((inargv
&& verbose
>=0) || verbose
> 0)
189 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
198 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
200 if ((inargv
&& verbose
>= 0) || verbose
> 0)
201 fprintf(stderr
, Name
": cannot open device %s: %s\n",
202 devname
, strerror(errno
));
203 } else if (fstat(dfd
, &stb
)< 0) {
205 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
206 devname
, strerror(errno
));
207 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
208 fprintf(stderr
, Name
": %s is not a block device.\n",
210 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
211 if ((inargv
&& verbose
>= 0) || verbose
> 0)
212 fprintf(stderr
, Name
": no recogniseable superblock\n");
213 } else if (tst
->ss
->load_super(tst
,dfd
, &super
, NULL
)) {
214 if ((inargv
&& verbose
>= 0) || verbose
> 0)
215 fprintf( stderr
, Name
": no RAID superblock on %s\n",
218 tst
->ss
->getinfo_super(&info
, &ident2
, super
);
220 if (dfd
>= 0) close(dfd
);
222 if (ident
->uuid_set
&& (!update
&& strcmp(update
, "uuid")!= 0) &&
223 (!super
|| same_uuid(info
.uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0)) {
224 if ((inargv
&& verbose
>= 0) || verbose
> 0)
225 fprintf(stderr
, Name
": %s has wrong uuid.\n",
229 if (ident
->name
[0] &&
230 (!super
|| strncmp(ident2
.name
, ident
->name
, 32)!=0)) {
231 if ((inargv
&& verbose
>= 0) || verbose
> 0)
232 fprintf(stderr
, Name
": %s has wrong name.\n",
236 if (ident
->super_minor
!= UnSet
&&
237 (!super
|| ident
->super_minor
!= info
.array
.md_minor
)) {
238 if ((inargv
&& verbose
>= 0) || verbose
> 0)
239 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
243 if (ident
->level
!= UnSet
&&
244 (!super
|| ident
->level
!= info
.array
.level
)) {
245 if ((inargv
&& verbose
>= 0) || verbose
> 0)
246 fprintf(stderr
, Name
": %s has wrong raid level.\n",
250 if (ident
->raid_disks
!= UnSet
&&
251 (!super
|| ident
->raid_disks
!= info
.array
.raid_disks
)) {
252 if ((inargv
&& verbose
>= 0) || verbose
> 0)
253 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
258 /* If we are this far, then we are commited to this device.
259 * If the super_block doesn't exist, or doesn't match others,
260 * then we cannot continue
264 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
269 st
= tst
; /* commit to this format, if haven't already */
270 if (st
->ss
->compare_super(&first_super
, super
)) {
271 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
278 /* looks like a good enough match to update the super block if needed */
280 /* prepare useful information in info structures */
283 info
.array
.md_minor
= minor(stb2
.st_rdev
);
285 if (strcmp(update
, "uuid")==0 &&
288 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
289 read(rfd
, ident
->uuid
, 16) != 16) {
290 *(__u32
*)(ident
->uuid
) = random();
291 *(__u32
*)(ident
->uuid
+1) = random();
292 *(__u32
*)(ident
->uuid
+2) = random();
293 *(__u32
*)(ident
->uuid
+3) = random();
295 if (rfd
>= 0) close(rfd
);
298 memcpy(info
.uuid
, ident
->uuid
, 16);
299 st
->ss
->update_super(&info
, super
, update
, devname
, verbose
);
301 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
303 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
305 else if (st
->ss
->store_super(st
, dfd
, super
))
306 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
313 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
314 devname
, mddev
, info
.disk
.raid_disk
);
315 devices
[devcnt
].devname
= devname
;
316 devices
[devcnt
].major
= major(stb
.st_rdev
);
317 devices
[devcnt
].minor
= minor(stb
.st_rdev
);
318 devices
[devcnt
].oldmajor
= info
.disk
.major
;
319 devices
[devcnt
].oldminor
= info
.disk
.minor
;
320 devices
[devcnt
].events
= info
.events
;
321 devices
[devcnt
].raid_disk
= info
.disk
.raid_disk
;
322 devices
[devcnt
].disk_nr
= info
.disk
.number
;
323 devices
[devcnt
].uptodate
= 0;
324 devices
[devcnt
].state
= info
.disk
.state
;
325 if (most_recent
< devcnt
) {
326 if (devices
[devcnt
].events
327 > devices
[most_recent
].events
)
328 most_recent
= devcnt
;
330 if (info
.array
.level
== -4)
331 /* with multipath, the raid_disk from the superblock is meaningless */
334 i
= devices
[devcnt
].raid_disk
;
336 if (nextspare
< info
.array
.raid_disks
)
337 nextspare
= info
.array
.raid_disks
;
342 unsigned int newbestcnt
= i
+10;
343 int *newbest
= malloc(sizeof(int)*newbestcnt
);
345 for (c
=0; c
< newbestcnt
; c
++)
347 newbest
[c
] = best
[c
];
352 bestcnt
= newbestcnt
;
355 || devices
[best
[i
]].events
< devices
[devcnt
].events
)
365 if (update
&& strcmp(update
, "byteorder")==0)
366 st
->minor_version
= 90;
369 fprintf(stderr
, Name
": no devices found for %s\n",
375 st
->ss
->getinfo_super(&info
, &ident2
, first_super
);
377 /* now we have some devices that might be suitable.
380 avail
= malloc(info
.array
.raid_disks
);
381 memset(avail
, 0, info
.array
.raid_disks
);
384 for (i
=0; i
< bestcnt
;i
++) {
386 int event_margin
= 1; /* always allow a difference of '1'
387 * like the kernel does
390 /* note: we ignore error flags in multipath arrays
391 * as they don't make sense
393 if (info
.array
.level
!= -4)
394 if (!(devices
[j
].state
& (1<<MD_DISK_SYNC
))) {
395 if (!(devices
[j
].state
& (1<<MD_DISK_FAULTY
)))
399 if (devices
[j
].events
+event_margin
>=
400 devices
[most_recent
].events
) {
401 devices
[j
].uptodate
= 1;
402 if (i
< info
.array
.raid_disks
) {
409 while (force
&& !enough(info
.array
.level
, info
.array
.raid_disks
,
412 /* Choose the newest best drive which is
413 * not up-to-date, update the superblock
418 for (i
=0; i
<info
.array
.raid_disks
&& i
< bestcnt
; i
++) {
421 !devices
[j
].uptodate
&&
422 devices
[j
].events
> 0 &&
424 devices
[j
].events
> devices
[chosen_drive
].events
))
427 if (chosen_drive
< 0)
430 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
431 devices
[chosen_drive
].devname
, devices
[chosen_drive
].raid_disk
,
432 (int)(devices
[chosen_drive
].events
),
433 (int)(devices
[most_recent
].events
));
434 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
436 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
437 devices
[chosen_drive
].devname
);
438 devices
[chosen_drive
].events
= 0;
441 if (st
->ss
->load_super(st
,fd
, &super
, NULL
)) {
443 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
444 devices
[chosen_drive
].devname
);
445 devices
[chosen_drive
].events
= 0;
448 info
.events
= devices
[most_recent
].events
;
449 st
->ss
->update_super(&info
, super
, "force", devices
[chosen_drive
].devname
, verbose
);
451 if (st
->ss
->store_super(st
, fd
, super
)) {
453 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
454 devices
[chosen_drive
].devname
);
455 devices
[chosen_drive
].events
= 0;
460 devices
[chosen_drive
].events
= devices
[most_recent
].events
;
461 devices
[chosen_drive
].uptodate
= 1;
462 avail
[chosen_drive
] = 1;
467 /* Now we want to look at the superblock which the kernel will base things on
468 * and compare the devices that we think are working with the devices that the
469 * superblock thinks are working.
470 * If there are differences and --force is given, then update this chosen
475 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
481 if (!devices
[j
].uptodate
)
484 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
485 fprintf(stderr
, Name
": Cannot open %s: %s\n",
486 devices
[j
].devname
, strerror(errno
));
489 if (st
->ss
->load_super(st
,fd
, &super
, NULL
)) {
491 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
498 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
501 st
->ss
->getinfo_super(&info
, &ident2
, super
);
502 for (i
=0; i
<bestcnt
; i
++) {
504 unsigned int desired_state
;
506 if (i
< info
.array
.raid_disks
)
507 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
513 if (!devices
[j
].uptodate
)
515 info
.disk
.number
= devices
[j
].disk_nr
;
516 info
.disk
.raid_disk
= i
;
517 info
.disk
.state
= desired_state
;
519 if (devices
[j
].uptodate
&&
520 st
->ss
->update_super(&info
, super
, "assemble", NULL
, verbose
)) {
523 fprintf(stderr
, Name
": "
524 "clearing FAULTY flag for device %d in %s for %s\n",
525 j
, mddev
, devices
[j
].devname
);
529 fprintf(stderr
, Name
": "
530 "device %d in %s has wrong state in superblock, but %s seems ok\n",
531 i
, mddev
, devices
[j
].devname
);
535 if (!devices
[j
].uptodate
&&
536 !(super
.disks
[i
].state
& (1 << MD_DISK_FAULTY
))) {
537 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
542 if (force
&& okcnt
== info
.array
.raid_disks
-1) {
543 /* FIXME check event count */
544 change
+= st
->ss
->update_super(&info
, super
, "force",
545 devices
[chosen_drive
].devname
, verbose
);
550 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
552 fprintf(stderr
, Name
": Could open %s for write - cannot Assemble array.\n",
553 devices
[chosen_drive
].devname
);
556 if (st
->ss
->store_super(st
, fd
, super
)) {
558 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
559 devices
[chosen_drive
].devname
);
565 /* count number of in-sync devices according to the superblock.
566 * We must have this number to start the array without -s or -R
568 req_cnt
= info
.array
.working_disks
;
570 /* Almost ready to actually *do* something */
573 if ((vers
% 100) >= 1) { /* can use different versions */
574 mdu_array_info_t inf
;
575 memset(&inf
, 0, sizeof(inf
));
576 inf
.major_version
= st
->ss
->major
;
577 inf
.minor_version
= st
->minor_version
;
578 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
580 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
583 fprintf(stderr
, Name
": SET_ARRAY_INFO failed for %s: %s\n",
584 mddev
, strerror(errno
));
587 if (ident
->bitmap_fd
>= 0) {
588 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
589 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
594 /* First, add the raid disks, but add the chosen one last */
595 for (i
=0; i
<= bestcnt
; i
++) {
599 if (j
== chosen_drive
)
604 if (j
>= 0 /* && devices[j].uptodate */) {
605 mdu_disk_info_t disk
;
606 memset(&disk
, 0, sizeof(disk
));
607 disk
.major
= devices
[j
].major
;
608 disk
.minor
= devices
[j
].minor
;
609 if (ioctl(mdfd
, ADD_NEW_DISK
, &disk
)!=0) {
610 fprintf(stderr
, Name
": failed to add %s to %s: %s\n",
614 if (i
< info
.array
.raid_disks
|| i
== bestcnt
)
618 } else if (verbose
> 0)
619 fprintf(stderr
, Name
": added %s to %s as %d\n",
620 devices
[j
].devname
, mddev
, devices
[j
].raid_disk
);
621 } else if (verbose
> 0 && i
< info
.array
.raid_disks
)
622 fprintf(stderr
, Name
": no uptodate device for slot %d of %s\n",
628 ( enough(info
.array
.level
, info
.array
.raid_disks
, info
.array
.layout
, avail
, okcnt
) &&
629 (okcnt
>= req_cnt
|| start_partial_ok
)
631 if (ioctl(mdfd
, RUN_ARRAY
, NULL
)==0) {
633 fprintf(stderr
, Name
": %s has been started with %d drive%s",
634 mddev
, okcnt
, okcnt
==1?"":"s");
635 if (okcnt
< info
.array
.raid_disks
)
636 fprintf(stderr
, " (out of %d)", info
.array
.raid_disks
);
638 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
639 fprintf(stderr
, ".\n");
643 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
644 mddev
, strerror(errno
));
648 fprintf(stderr
, Name
": %s assembled from %d drive%s, but not started.\n",
649 mddev
, okcnt
, okcnt
==1?"":"s");
653 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
655 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
656 if (!enough(info
.array
.level
, info
.array
.raid_disks
, info
.array
.layout
, avail
, okcnt
))
657 fprintf(stderr
, " - not enough to start the array.\n");
659 if (req_cnt
== info
.array
.raid_disks
)
660 fprintf(stderr
, " - need all %d to start it", req_cnt
);
662 fprintf(stderr
, " - need %d of %d to start", req_cnt
, info
.array
.raid_disks
);
663 fprintf(stderr
, " (use --run to insist).\n");
668 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
669 * been updated to point to the current locations of devices.
670 * so we can just start the array
673 dev
= makedev(devices
[chosen_drive
].major
,
674 devices
[chosen_drive
].minor
);
675 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
676 fprintf(stderr
, Name
": Cannot start array: %s\n",