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
34 int Assemble(char *mddev
, int mdfd
,
35 mddev_ident_t ident
, char *conffile
,
37 int readonly
, int runstop
,
39 int verbose
, int force
)
42 * The task of Assemble is to find a collection of
43 * devices that should (according to their superblocks)
44 * form an array, and to give this collection to the MD driver.
45 * In Linux-2.4 and later, this involves submitting a
46 * SET_ARRAY_INFO ioctl with no arg - to prepare
47 * the array - and then submit a number of
48 * ADD_NEW_DISK ioctls to add disks into
49 * the array. Finally RUN_ARRAY might
50 * be submitted to start the array.
52 * Much of the work of Assemble is in finding and/or
53 * checking the disks to make sure they look right.
55 * If mddev is not set, then scan must be and we
56 * read through the config file for dev+uuid mapping
57 * We recurse, setting mddev, for each device that
59 * - has a valid uuid (or any uuid if !uuidset
61 * If mddev is set, we try to determine state of md.
62 * check version - must be at least 0.90.0
63 * check kernel version. must be at least 2.4.
64 * If not, we can possibly fall back on START_ARRAY
65 * Try to GET_ARRAY_INFO.
66 * If possible, give up
67 * If not, try to STOP_ARRAY just to make sure
69 * If !uuidset and scan, look in conf-file for uuid
70 * If not found, give up
71 * If !devlist and scan and uuidset, get list of devs from conf-file
74 * Check superblock - discard if bad
75 * Check uuid (set if we don't have one) - discard if no match
76 * Check superblock similarity if we have a superblock - discard if different
77 * Record events, devicenum, utime
78 * This should give us a list of devices for the array
79 * We should collect the most recent event and utime numbers
81 * Count disks with recent enough event count
82 * While force && !enough disks
83 * Choose newest rejected disks, update event count
84 * mark clean and rewrite superblock
87 * foreach device with recent events : ADD_NEW_DISK
88 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
90 * Check the device numbers in superblock are right
91 * update superblock if any changes
97 mdu_array_info_t array
;
98 mdp_super_t first_super
, super
;
102 int oldmajor
, oldminor
;
109 int *best
= NULL
; /* indexed by raid_disk */
111 int devcnt
= 0, okcnt
, sparecnt
;
117 int start_partial_ok
= force
|| devlist
==NULL
;
121 vers
= md_get_version(mdfd
);
123 fprintf(stderr
, Name
": %s appears not to be an md device.\n", mddev
);
127 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
128 " Upgrade your kernel or try --build\n");
131 if (get_linux_version() < 2004000)
134 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
)>=0) {
135 fprintf(stderr
, Name
": device %s already active - cannot assemble it\n",
139 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
142 * If any subdevs are listed, then any that don't
143 * match ident are discarded. Remainder must all match and
145 * If no subdevs, then we scan all devices in the config file, but
146 * there must be something in the identity
150 ident
->uuid_set
== 0 &&
151 ident
->super_minor
< 0 &&
152 ident
->devices
== NULL
) {
153 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
158 devlist
= conf_get_devs(conffile
);
161 tmpdev
= devlist
; num_devs
= 0;
164 tmpdev
= tmpdev
->next
;
166 best
= malloc(num_devs
* sizeof(*best
));
167 devices
= malloc(num_devs
* sizeof(*devices
));
169 first_super
.md_magic
= 0;
170 for (i
=0; i
<num_devs
; i
++)
174 fprintf(stderr
, Name
": looking for devices for %s\n",
184 devname
= devlist
->devname
;
185 devlist
= devlist
->next
;
187 if (ident
->devices
&&
188 !match_oneof(ident
->devices
, devname
)) {
189 if (inargv
|| verbose
)
190 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
194 dfd
= open(devname
, O_RDONLY
, 0);
196 if (inargv
|| verbose
)
197 fprintf(stderr
, Name
": cannot open device %s: %s\n",
198 devname
, strerror(errno
));
199 } else if (fstat(dfd
, &stb
)< 0) {
201 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
202 devname
, strerror(errno
));
204 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
205 fprintf(stderr
, Name
": %s is not a block device.\n",
208 } else if (load_super(dfd
, &super
)) {
209 if (inargv
|| verbose
)
210 fprintf( stderr
, Name
": no RAID superblock on %s\n",
215 uuid_from_super(this_uuid
, &super
);
219 if (ident
->uuid_set
&&
220 (!havesuper
|| same_uuid(this_uuid
, ident
->uuid
)==0)) {
221 if (inargv
|| verbose
)
222 fprintf(stderr
, Name
": %s has wrong uuid.\n",
226 if (ident
->super_minor
>= 0 &&
227 (!havesuper
|| ident
->super_minor
!= super
.md_minor
)) {
228 if (inargv
|| verbose
)
229 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
233 if (ident
->level
!= -10 &&
234 (!havesuper
|| ident
->level
!= super
.level
)) {
235 if (inargv
|| verbose
)
236 fprintf(stderr
, Name
": %s has wrong raid level.\n",
240 if (ident
->raid_disks
!= -1 &&
241 (!havesuper
|| ident
->raid_disks
!= super
.raid_disks
)) {
242 if (inargv
|| verbose
)
243 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
248 /* If we are this far, then we are commited to this device.
249 * If the super_block doesn't exist, or doesn't match others,
250 * then we cannot continue
254 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
258 if (compare_super(&first_super
, &super
)) {
259 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
265 /* this is needed until we get a more relaxed super block format */
266 if (devcnt
>= MD_SB_DISKS
) {
267 fprintf(stderr
, Name
": ouch - too many devices appear to be in this array. Ignoring %s\n",
272 /* looks like a good enough match to update the super block if needed */
274 if (strcmp(update
, "sparc2.2")==0 ) {
275 /* 2.2 sparc put the events in the wrong place
276 * So we copy the tail of the superblock
277 * up 4 bytes before continuing
279 __u32
*sb32
= (__u32
*)&super
;
280 memcpy(sb32
+MD_SB_GENERIC_CONSTANT_WORDS
+7,
281 sb32
+MD_SB_GENERIC_CONSTANT_WORDS
+7+1,
282 (MD_SB_WORDS
- (MD_SB_GENERIC_CONSTANT_WORDS
+7+1))*4);
283 fprintf (stderr
, Name
": adjusting superblock of %s for 2.2/sparc compatability.\n",
286 if (strcmp(update
, "super-minor") ==0) {
289 super
.md_minor
= MINOR(stb2
.st_rdev
);
291 fprintf(stderr
, Name
": updating superblock of %s with minor number %d\n",
292 devname
, super
.md_minor
);
294 super
.sb_csum
= calc_sb_csum(&super
);
295 dfd
= open(devname
, O_RDWR
, 0);
297 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
299 else if (store_super(dfd
, &super
))
300 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
307 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
308 devname
, mddev
, super
.this_disk
.raid_disk
);
309 devices
[devcnt
].devname
= devname
;
310 devices
[devcnt
].major
= MAJOR(stb
.st_rdev
);
311 devices
[devcnt
].minor
= MINOR(stb
.st_rdev
);
312 devices
[devcnt
].oldmajor
= super
.this_disk
.major
;
313 devices
[devcnt
].oldminor
= super
.this_disk
.minor
;
314 devices
[devcnt
].events
= md_event(&super
);
315 devices
[devcnt
].utime
= super
.utime
;
316 devices
[devcnt
].raid_disk
= super
.this_disk
.raid_disk
;
317 devices
[devcnt
].uptodate
= 0;
318 devices
[devcnt
].state
= super
.this_disk
.state
;
319 if (most_recent
< devcnt
) {
320 if (devices
[devcnt
].events
321 > devices
[most_recent
].events
)
322 most_recent
= devcnt
;
324 if (super
.level
== -4)
325 /* with multipath, the raid_disk from the superblock is meaningless */
328 i
= devices
[devcnt
].raid_disk
;
329 if (i
>=0 && i
< 10000) {
331 int newbestcnt
= i
+10;
332 int *newbest
= malloc(sizeof(int)*newbestcnt
);
334 for (c
=0; c
< newbestcnt
; c
++)
336 newbest
[c
] = best
[c
];
341 bestcnt
= newbestcnt
;
344 || devices
[best
[i
]].events
< devices
[devcnt
].events
)
351 fprintf(stderr
, Name
": no devices found for %s\n",
355 /* now we have some devices that might be suitable.
360 for (i
=0; i
< bestcnt
;i
++) {
362 int event_margin
= !force
;
364 /* note: we ignore error flags in multipath arrays
365 * as they don't make sense
367 if (first_super
.level
!= -4)
368 if (!(devices
[j
].state
& (1<<MD_DISK_SYNC
)))
370 if (devices
[j
].events
+event_margin
>=
371 devices
[most_recent
].events
) {
372 devices
[j
].uptodate
= 1;
373 if (i
< first_super
.raid_disks
)
379 while (force
&& !enough(first_super
.level
, first_super
.raid_disks
, okcnt
)) {
380 /* Choose the newest best drive which is
381 * not up-to-date, update the superblock
386 for (i
=0; i
<first_super
.raid_disks
&& i
< bestcnt
; i
++) {
389 !devices
[j
].uptodate
&&
390 devices
[j
].events
> 0 &&
392 devices
[j
].events
> devices
[chosen_drive
].events
))
395 if (chosen_drive
< 0)
397 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
398 devices
[chosen_drive
].devname
, devices
[chosen_drive
].raid_disk
,
399 (int)(devices
[chosen_drive
].events
),
400 (int)(devices
[most_recent
].events
));
401 fd
= open(devices
[chosen_drive
].devname
, O_RDWR
);
403 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
404 devices
[chosen_drive
].devname
);
405 devices
[chosen_drive
].events
= 0;
408 if (load_super(fd
, &super
)) {
410 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
411 devices
[chosen_drive
].devname
);
412 devices
[chosen_drive
].events
= 0;
415 super
.events_hi
= (devices
[most_recent
].events
>>32)&0xFFFFFFFF;
416 super
.events_lo
= (devices
[most_recent
].events
)&0xFFFFFFFF;
417 if (super
.level
== 5 || super
.level
== 4) {
418 /* need to force clean */
419 super
.state
= (1<<MD_SB_CLEAN
);
421 super
.sb_csum
= calc_sb_csum(&super
);
422 /*DRYRUN*/ if (store_super(fd
, &super
)) {
424 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
425 devices
[chosen_drive
].devname
);
426 devices
[chosen_drive
].events
= 0;
430 devices
[chosen_drive
].events
= devices
[most_recent
].events
;
431 devices
[chosen_drive
].uptodate
= 1;
435 /* Now we want to look at the superblock which the kernel will base things on
436 * and compare the devices that we think are working with the devices that the
437 * superblock thinks are working.
438 * If there are differences and --force is given, then update this chosen
442 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
447 if (!devices
[j
].uptodate
)
450 if ((fd
=open(devices
[j
].devname
, O_RDONLY
))< 0) {
451 fprintf(stderr
, Name
": Cannot open %s: %s\n",
452 devices
[j
].devname
, strerror(errno
));
455 if (load_super(fd
, &super
)) {
457 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
464 for (i
=0; i
<bestcnt
; i
++) {
468 if (i
< super
.raid_disks
)
469 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
475 if (!devices
[j
].uptodate
)
479 if (devices
[j
].major
!= super
.disks
[i
].major
||
480 devices
[j
].minor
!= super
.disks
[i
].minor
) {
482 super
.disks
[i
].major
= devices
[j
].major
;
483 super
.disks
[i
].minor
= devices
[j
].minor
;
486 if (devices
[j
].oldmajor
!= super
.disks
[i
].major
||
487 devices
[j
].oldminor
!= super
.disks
[i
].minor
) {
489 super
.disks
[i
].major
= devices
[j
].oldmajor
;
490 super
.disks
[i
].minor
= devices
[j
].oldminor
;
492 if (devices
[j
].uptodate
&&
493 (super
.disks
[i
].state
!= desired_state
)) {
495 fprintf(stderr
, Name
": "
496 "clearing FAULTY flag for device %d in %s for %s\n",
497 j
, mddev
, devices
[j
].devname
);
498 super
.disks
[i
].state
= desired_state
;
501 fprintf(stderr
, Name
": "
502 "device %d in %s has wrong state in superblock, but %s seems ok\n",
503 i
, mddev
, devices
[j
].devname
);
506 if (!devices
[j
].uptodate
&&
507 !(super
.disks
[i
].state
& (1 << MD_DISK_FAULTY
))) {
508 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
512 if (force
&& (super
.level
== 4 || super
.level
== 5) &&
513 okcnt
== super
.raid_disks
-1) {
514 super
.state
= (1<< MD_SB_CLEAN
);
518 if ((force
&& (change
& 2))
519 || (old_linux
&& (change
& 1))) {
521 super
.sb_csum
= calc_sb_csum(&super
);
522 fd
= open(devices
[chosen_drive
].devname
, O_RDWR
);
524 fprintf(stderr
, Name
": Could open %s for write - cannot Assemble array.\n",
525 devices
[chosen_drive
].devname
);
528 if (store_super(fd
, &super
)) {
530 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
531 devices
[chosen_drive
].devname
);
538 /* Almost ready to actually *do* something */
540 if (ioctl(mdfd
, SET_ARRAY_INFO
, NULL
) != 0) {
541 fprintf(stderr
, Name
": SET_ARRAY_INFO failed for %s: %s\n",
542 mddev
, strerror(errno
));
545 /* First, add the raid disks, but add the chosen one last */
546 for (i
=0; i
<= bestcnt
; i
++) {
550 if (j
== chosen_drive
)
555 if (j
>= 0 /* && devices[j].uptodate */) {
556 mdu_disk_info_t disk
;
557 memset(&disk
, 0, sizeof(disk
));
558 disk
.major
= devices
[j
].major
;
559 disk
.minor
= devices
[j
].minor
;
560 if (ioctl(mdfd
, ADD_NEW_DISK
, &disk
)!=0) {
561 fprintf(stderr
, Name
": failed to add %s to %s: %s\n",
565 if (i
< first_super
.raid_disks
)
570 fprintf(stderr
, Name
": added %s to %s as %d\n",
571 devices
[j
].devname
, mddev
, devices
[j
].raid_disk
);
572 } else if (verbose
&& i
< first_super
.raid_disks
)
573 fprintf(stderr
, Name
": no uptodate device for slot %d of %s\n",
579 ( first_super
.raid_disks
== okcnt
580 || (start_partial_ok
&& enough(first_super
.level
, first_super
.raid_disks
, okcnt
)))
582 if (ioctl(mdfd
, RUN_ARRAY
, NULL
)==0) {
583 fprintf(stderr
, Name
": %s has been started with %d drive%s",
584 mddev
, okcnt
, okcnt
==1?"":"s");
586 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
587 fprintf(stderr
, ".\n");
590 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
591 mddev
, strerror(errno
));
595 fprintf(stderr
, Name
": %s assembled from %d drive%s, but not started.\n",
596 mddev
, okcnt
, okcnt
==1?"":"s");
599 fprintf(stderr
, Name
": %s assembled from %d drive%s - not enough to start it (use --run to insist).\n",
600 mddev
, okcnt
, okcnt
==1?"":"s");
603 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
604 * been updated to point to the current locations of devices.
605 * so we can just start the array
608 dev
= MKDEV(devices
[chosen_drive
].major
,
609 devices
[chosen_drive
].minor
);
610 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
611 fprintf(stderr
, Name
": Cannot start array: %s\n",