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
; /* indexed by raid_disk */
110 int devcnt
= 0, okcnt
, sparecnt
;
116 int start_partial_ok
= force
|| devlist
==NULL
;
120 vers
= md_get_version(mdfd
);
122 fprintf(stderr
, Name
": %s appears not to be an md device.\n", mddev
);
126 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
127 " Upgrade your kernel or try --build\n");
130 if (get_linux_version() < 2004000)
133 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
)>=0) {
134 fprintf(stderr
, Name
": device %s already active - cannot assemble it\n",
138 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
141 * If any subdevs are listed, then any that don't
142 * match ident are discarded. Remainder must all match and
144 * If no subdevs, then we scan all devices in the config file, but
145 * there must be something in the identity
149 ident
->uuid_set
== 0 &&
150 ident
->super_minor
< 0 &&
151 ident
->devices
== NULL
) {
152 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
157 devlist
= conf_get_devs(conffile
);
160 tmpdev
= devlist
; num_devs
= 0;
163 tmpdev
= tmpdev
->next
;
165 best
= malloc(num_devs
* sizeof(*best
));
166 devices
= malloc(num_devs
* sizeof(*devices
));
168 first_super
.md_magic
= 0;
169 for (i
=0; i
<num_devs
; i
++)
173 fprintf(stderr
, Name
": looking for devices for %s\n",
183 devname
= devlist
->devname
;
184 devlist
= devlist
->next
;
186 if (ident
->devices
&&
187 !match_oneof(ident
->devices
, devname
))
190 dfd
= open(devname
, O_RDONLY
, 0);
192 if (inargv
|| verbose
)
193 fprintf(stderr
, Name
": cannot open device %s: %s\n",
194 devname
, strerror(errno
));
195 } else if (fstat(dfd
, &stb
)< 0) {
197 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
198 devname
, strerror(errno
));
200 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
201 fprintf(stderr
, Name
": %s is not a block device.\n",
204 } else if (load_super(dfd
, &super
)) {
205 if (inargv
|| verbose
)
206 fprintf( stderr
, Name
": no RAID superblock on %s\n",
211 uuid_from_super(this_uuid
, &super
);
215 if (ident
->uuid_set
&&
216 (!havesuper
|| same_uuid(this_uuid
, ident
->uuid
)==0)) {
217 if (inargv
|| verbose
)
218 fprintf(stderr
, Name
": %s has wrong uuid.\n",
222 if (ident
->super_minor
>= 0 &&
223 (!havesuper
|| ident
->super_minor
!= super
.md_minor
)) {
224 if (inargv
|| verbose
)
225 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
229 if (ident
->level
!= -10 &&
230 (!havesuper
|| ident
->level
!= super
.level
)) {
231 if (inargv
|| verbose
)
232 fprintf(stderr
, Name
": %s has wrong raid level.\n",
236 if (ident
->raid_disks
!= -1 &&
237 (!havesuper
|| ident
->raid_disks
!= super
.raid_disks
)) {
238 if (inargv
|| verbose
)
239 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
244 /* If we are this far, then we are commited to this device.
245 * If the super_block doesn't exist, or doesn't match others,
246 * then we cannot continue
250 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
254 if (compare_super(&first_super
, &super
)) {
255 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
261 /* this is needed until we get a more relaxed super block format */
262 if (devcnt
>= MD_SB_DISKS
) {
263 fprintf(stderr
, Name
": ouch - too many devices appear to be in this array. Ignoring %s\n",
268 /* looks like a good enough match to update the super block if needed */
270 if (strcmp(update
, "sparc2.2")==0 ) {
271 /* 2.2 sparc put the events in the wrong place
272 * So we copy the tail of the superblock
273 * up 4 bytes before continuing
275 __u32
*sb32
= (__u32
*)&super
;
276 memcpy(sb32
+MD_SB_GENERIC_CONSTANT_WORDS
+7,
277 sb32
+MD_SB_GENERIC_CONSTANT_WORDS
+7+1,
278 (MD_SB_WORDS
- (MD_SB_GENERIC_CONSTANT_WORDS
+7+1))*4);
279 fprintf (stderr
, Name
": adjusting superblock of %s for 2.2/sparc compatability.\n",
282 if (strcmp(update
, "super-minor") ==0) {
285 super
.md_minor
= MINOR(stb2
.st_rdev
);
287 fprintf(stderr
, Name
": updating superblock of %s with minor number %d\n",
288 devname
, super
.md_minor
);
290 super
.sb_csum
= calc_sb_csum(&super
);
291 dfd
= open(devname
, O_RDWR
, 0);
293 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
295 else if (store_super(dfd
, &super
))
296 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
303 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
304 devname
, mddev
, super
.this_disk
.raid_disk
);
305 devices
[devcnt
].devname
= devname
;
306 devices
[devcnt
].major
= MAJOR(stb
.st_rdev
);
307 devices
[devcnt
].minor
= MINOR(stb
.st_rdev
);
308 devices
[devcnt
].oldmajor
= super
.this_disk
.major
;
309 devices
[devcnt
].oldminor
= super
.this_disk
.minor
;
310 devices
[devcnt
].events
= md_event(&super
);
311 devices
[devcnt
].utime
= super
.utime
;
312 devices
[devcnt
].raid_disk
= super
.this_disk
.raid_disk
;
313 devices
[devcnt
].uptodate
= 0;
314 devices
[devcnt
].state
= super
.this_disk
.state
;
315 if (most_recent
< devcnt
) {
316 if (devices
[devcnt
].events
317 > devices
[most_recent
].events
)
318 most_recent
= devcnt
;
320 if (super
.level
== -4)
321 /* with multipath, the raid_disk from the superblock is meaningless */
324 i
= devices
[devcnt
].raid_disk
;
325 if (i
>=0 && i
< num_devs
)
327 || devices
[best
[i
]].events
< devices
[devcnt
].events
)
334 fprintf(stderr
, Name
": no devices found for %s\n",
338 /* now we have some devices that might be suitable.
343 for (i
=0; i
< num_devs
;i
++) {
345 int event_margin
= !force
;
347 /* note: we ignore error flags in multipath arrays
348 * as they don't make sense
350 if (first_super
.level
!= -4)
351 if (!(devices
[j
].state
& (1<<MD_DISK_SYNC
)))
353 if (devices
[j
].events
+event_margin
>=
354 devices
[most_recent
].events
) {
355 devices
[j
].uptodate
= 1;
356 if (i
< first_super
.raid_disks
)
362 while (force
&& !enough(first_super
.level
, first_super
.raid_disks
, okcnt
)) {
363 /* Choose the newest best drive which is
364 * not up-to-date, update the superblock
369 for (i
=0; i
<first_super
.raid_disks
; i
++) {
372 !devices
[j
].uptodate
&&
373 devices
[j
].events
> 0 &&
375 devices
[j
].events
> devices
[chosen_drive
].events
))
378 if (chosen_drive
< 0)
380 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
381 devices
[chosen_drive
].devname
, devices
[chosen_drive
].raid_disk
,
382 (int)(devices
[chosen_drive
].events
),
383 (int)(devices
[most_recent
].events
));
384 fd
= open(devices
[chosen_drive
].devname
, O_RDWR
);
386 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
387 devices
[chosen_drive
].devname
);
388 devices
[chosen_drive
].events
= 0;
391 if (load_super(fd
, &super
)) {
393 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
394 devices
[chosen_drive
].devname
);
395 devices
[chosen_drive
].events
= 0;
398 super
.events_hi
= (devices
[most_recent
].events
>>32)&0xFFFFFFFF;
399 super
.events_lo
= (devices
[most_recent
].events
)&0xFFFFFFFF;
400 if (super
.level
== 5 || super
.level
== 4) {
401 /* need to force clean */
402 super
.state
= (1<<MD_SB_CLEAN
);
404 super
.sb_csum
= calc_sb_csum(&super
);
405 /*DRYRUN*/ if (store_super(fd
, &super
)) {
407 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
408 devices
[chosen_drive
].devname
);
409 devices
[chosen_drive
].events
= 0;
413 devices
[chosen_drive
].events
= devices
[most_recent
].events
;
414 devices
[chosen_drive
].uptodate
= 1;
418 /* Now we want to look at the superblock which the kernel will base things on
419 * and compare the devices that we think are working with the devices that the
420 * superblock thinks are working.
421 * If there are differences and --force is given, then update this chosen
425 for (i
=0; chosen_drive
< 0 && i
<num_devs
; i
++) {
430 if (!devices
[j
].uptodate
)
433 if ((fd
=open(devices
[j
].devname
, O_RDONLY
))< 0) {
434 fprintf(stderr
, Name
": Cannot open %s: %s\n",
435 devices
[j
].devname
, strerror(errno
));
438 if (load_super(fd
, &super
)) {
440 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
447 for (i
=0; i
<num_devs
; i
++) {
451 if (i
< super
.raid_disks
)
452 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
458 if (!devices
[j
].uptodate
)
462 if (devices
[j
].major
!= super
.disks
[i
].major
||
463 devices
[j
].minor
!= super
.disks
[i
].minor
) {
465 super
.disks
[i
].major
= devices
[j
].major
;
466 super
.disks
[i
].minor
= devices
[j
].minor
;
469 if (devices
[j
].oldmajor
!= super
.disks
[i
].major
||
470 devices
[j
].oldminor
!= super
.disks
[i
].minor
) {
472 super
.disks
[i
].major
= devices
[j
].oldmajor
;
473 super
.disks
[i
].minor
= devices
[j
].oldminor
;
475 if (devices
[j
].uptodate
&&
476 (super
.disks
[i
].state
!= desired_state
)) {
478 fprintf(stderr
, Name
": "
479 "clearing FAULTY flag for device %d in %s for %s\n",
480 j
, mddev
, devices
[j
].devname
);
481 super
.disks
[i
].state
= desired_state
;
484 fprintf(stderr
, Name
": "
485 "device %d in %s has wrong state in superblock, but %s seems ok\n",
486 i
, mddev
, devices
[j
].devname
);
489 if (!devices
[j
].uptodate
&&
490 !(super
.disks
[i
].state
& (1 << MD_DISK_FAULTY
))) {
491 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
495 if (force
&& (super
.level
== 4 || super
.level
== 5) &&
496 okcnt
== super
.raid_disks
-1) {
497 super
.state
= (1<< MD_SB_CLEAN
);
501 if ((force
&& (change
& 2))
502 || (old_linux
&& (change
& 1))) {
504 super
.sb_csum
= calc_sb_csum(&super
);
505 fd
= open(devices
[chosen_drive
].devname
, O_RDWR
);
507 fprintf(stderr
, Name
": Could open %s for write - cannot Assemble array.\n",
508 devices
[chosen_drive
].devname
);
511 if (store_super(fd
, &super
)) {
513 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
514 devices
[chosen_drive
].devname
);
521 /* Almost ready to actually *do* something */
523 if (ioctl(mdfd
, SET_ARRAY_INFO
, NULL
) != 0) {
524 fprintf(stderr
, Name
": SET_ARRAY_INFO failed for %s: %s\n",
525 mddev
, strerror(errno
));
528 /* First, add the raid disks, but add the chosen one last */
529 for (i
=0; i
<= num_devs
; i
++) {
533 if (j
== chosen_drive
)
538 if (j
>= 0 && devices
[j
].uptodate
) {
539 mdu_disk_info_t disk
;
540 memset(&disk
, 0, sizeof(disk
));
541 disk
.major
= devices
[j
].major
;
542 disk
.minor
= devices
[j
].minor
;
543 if (ioctl(mdfd
, ADD_NEW_DISK
, &disk
)!=0) {
544 fprintf(stderr
, Name
": failed to add %s to %s: %s\n",
548 if (i
< first_super
.raid_disks
)
553 fprintf(stderr
, Name
": added %s to %s as %d\n",
554 devices
[j
].devname
, mddev
, devices
[j
].raid_disk
);
555 } else if (verbose
&& i
< first_super
.raid_disks
)
556 fprintf(stderr
, Name
": no uptodate device for slot %d of %s\n",
562 ( first_super
.raid_disks
== okcnt
563 || (start_partial_ok
&& enough(first_super
.level
, first_super
.raid_disks
, okcnt
)))
565 if (ioctl(mdfd
, RUN_ARRAY
, NULL
)==0) {
566 fprintf(stderr
, Name
": %s has been started with %d drive%s",
567 mddev
, okcnt
, okcnt
==1?"":"s");
569 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
570 fprintf(stderr
, ".\n");
573 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
574 mddev
, strerror(errno
));
578 fprintf(stderr
, Name
": %s assembled from %d drive%s, but not started.\n",
579 mddev
, okcnt
, okcnt
==1?"":"s");
582 fprintf(stderr
, Name
": %s assembled from %d drive%s - not enough to start it (use --run to insist).\n",
583 mddev
, okcnt
, okcnt
==1?"":"s");
586 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
587 * been updated to point to the current locations of devices.
588 * so we can just start the array
591 dev
= MKDEV(devices
[chosen_drive
].major
,
592 devices
[chosen_drive
].minor
);
593 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
594 fprintf(stderr
, Name
": Cannot start array: %s\n",