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
;
118 int start_partial_ok
= force
|| devlist
==NULL
;
122 vers
= md_get_version(mdfd
);
124 fprintf(stderr
, Name
": %s appears not to be an md device.\n", mddev
);
128 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
129 " Upgrade your kernel or try --build\n");
132 if (get_linux_version() < 2004000)
135 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
)>=0) {
136 fprintf(stderr
, Name
": device %s already active - cannot assemble it\n",
140 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
143 * If any subdevs are listed, then any that don't
144 * match ident are discarded. Remainder must all match and
146 * If no subdevs, then we scan all devices in the config file, but
147 * there must be something in the identity
151 ident
->uuid_set
== 0 &&
152 ident
->super_minor
< 0 &&
153 ident
->devices
== NULL
) {
154 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
159 devlist
= conf_get_devs(conffile
);
162 tmpdev
= devlist
; num_devs
= 0;
165 tmpdev
= tmpdev
->next
;
167 best
= malloc(num_devs
* sizeof(*best
));
168 devices
= malloc(num_devs
* sizeof(*devices
));
170 first_super
.md_magic
= 0;
171 for (i
=0; i
<num_devs
; i
++)
175 fprintf(stderr
, Name
": looking for devices for %s\n",
185 devname
= devlist
->devname
;
186 devlist
= devlist
->next
;
188 if (ident
->devices
&&
189 !match_oneof(ident
->devices
, devname
)) {
190 if (inargv
|| verbose
)
191 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
195 dfd
= open(devname
, O_RDONLY
, 0);
197 if (inargv
|| verbose
)
198 fprintf(stderr
, Name
": cannot open device %s: %s\n",
199 devname
, strerror(errno
));
200 } else if (fstat(dfd
, &stb
)< 0) {
202 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
203 devname
, strerror(errno
));
205 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
206 fprintf(stderr
, Name
": %s is not a block device.\n",
209 } else if (load_super(dfd
, &super
)) {
210 if (inargv
|| verbose
)
211 fprintf( stderr
, Name
": no RAID superblock on %s\n",
216 uuid_from_super(this_uuid
, &super
);
220 if (ident
->uuid_set
&&
221 (!havesuper
|| same_uuid(this_uuid
, ident
->uuid
)==0)) {
222 if (inargv
|| verbose
)
223 fprintf(stderr
, Name
": %s has wrong uuid.\n",
227 if (ident
->super_minor
>= 0 &&
228 (!havesuper
|| ident
->super_minor
!= super
.md_minor
)) {
229 if (inargv
|| verbose
)
230 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
234 if (ident
->level
!= -10 &&
235 (!havesuper
|| ident
->level
!= super
.level
)) {
236 if (inargv
|| verbose
)
237 fprintf(stderr
, Name
": %s has wrong raid level.\n",
241 if (ident
->raid_disks
!= -1 &&
242 (!havesuper
|| ident
->raid_disks
!= super
.raid_disks
)) {
243 if (inargv
|| verbose
)
244 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
249 /* If we are this far, then we are commited to this device.
250 * If the super_block doesn't exist, or doesn't match others,
251 * then we cannot continue
255 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
259 if (compare_super(&first_super
, &super
)) {
260 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
266 /* this is needed until we get a more relaxed super block format */
267 if (devcnt
>= MD_SB_DISKS
) {
268 fprintf(stderr
, Name
": ouch - too many devices appear to be in this array. Ignoring %s\n",
273 /* looks like a good enough match to update the super block if needed */
275 if (strcmp(update
, "sparc2.2")==0 ) {
276 /* 2.2 sparc put the events in the wrong place
277 * So we copy the tail of the superblock
278 * up 4 bytes before continuing
280 __u32
*sb32
= (__u32
*)&super
;
281 memcpy(sb32
+MD_SB_GENERIC_CONSTANT_WORDS
+7,
282 sb32
+MD_SB_GENERIC_CONSTANT_WORDS
+7+1,
283 (MD_SB_WORDS
- (MD_SB_GENERIC_CONSTANT_WORDS
+7+1))*4);
284 fprintf (stderr
, Name
": adjusting superblock of %s for 2.2/sparc compatability.\n",
287 if (strcmp(update
, "super-minor") ==0) {
290 super
.md_minor
= MINOR(stb2
.st_rdev
);
292 fprintf(stderr
, Name
": updating superblock of %s with minor number %d\n",
293 devname
, super
.md_minor
);
295 super
.sb_csum
= calc_sb_csum(&super
);
296 dfd
= open(devname
, O_RDWR
, 0);
298 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
300 else if (store_super(dfd
, &super
))
301 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
308 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
309 devname
, mddev
, super
.this_disk
.raid_disk
);
310 devices
[devcnt
].devname
= devname
;
311 devices
[devcnt
].major
= MAJOR(stb
.st_rdev
);
312 devices
[devcnt
].minor
= MINOR(stb
.st_rdev
);
313 devices
[devcnt
].oldmajor
= super
.this_disk
.major
;
314 devices
[devcnt
].oldminor
= super
.this_disk
.minor
;
315 devices
[devcnt
].events
= md_event(&super
);
316 devices
[devcnt
].utime
= super
.utime
;
317 devices
[devcnt
].raid_disk
= super
.this_disk
.raid_disk
;
318 devices
[devcnt
].uptodate
= 0;
319 devices
[devcnt
].state
= super
.this_disk
.state
;
320 if (most_recent
< devcnt
) {
321 if (devices
[devcnt
].events
322 > devices
[most_recent
].events
)
323 most_recent
= devcnt
;
325 if (super
.level
== -4)
326 /* with multipath, the raid_disk from the superblock is meaningless */
329 i
= devices
[devcnt
].raid_disk
;
330 if (i
>=0 && i
< 10000) {
332 int newbestcnt
= i
+10;
333 int *newbest
= malloc(sizeof(int)*newbestcnt
);
335 for (c
=0; c
< newbestcnt
; c
++)
337 newbest
[c
] = best
[c
];
342 bestcnt
= newbestcnt
;
345 || devices
[best
[i
]].events
< devices
[devcnt
].events
)
352 fprintf(stderr
, Name
": no devices found for %s\n",
356 /* now we have some devices that might be suitable.
361 for (i
=0; i
< bestcnt
;i
++) {
363 int event_margin
= !force
;
365 /* note: we ignore error flags in multipath arrays
366 * as they don't make sense
368 if (first_super
.level
!= -4)
369 if (!(devices
[j
].state
& (1<<MD_DISK_SYNC
))) {
370 if (!(devices
[j
].state
& (1<<MD_DISK_FAULTY
)))
374 if (devices
[j
].events
+event_margin
>=
375 devices
[most_recent
].events
) {
376 devices
[j
].uptodate
= 1;
377 if (i
< first_super
.raid_disks
)
383 while (force
&& !enough(first_super
.level
, first_super
.raid_disks
, okcnt
)) {
384 /* Choose the newest best drive which is
385 * not up-to-date, update the superblock
390 for (i
=0; i
<first_super
.raid_disks
&& i
< bestcnt
; i
++) {
393 !devices
[j
].uptodate
&&
394 devices
[j
].events
> 0 &&
396 devices
[j
].events
> devices
[chosen_drive
].events
))
399 if (chosen_drive
< 0)
401 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
402 devices
[chosen_drive
].devname
, devices
[chosen_drive
].raid_disk
,
403 (int)(devices
[chosen_drive
].events
),
404 (int)(devices
[most_recent
].events
));
405 fd
= open(devices
[chosen_drive
].devname
, O_RDWR
);
407 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
408 devices
[chosen_drive
].devname
);
409 devices
[chosen_drive
].events
= 0;
412 if (load_super(fd
, &super
)) {
414 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
415 devices
[chosen_drive
].devname
);
416 devices
[chosen_drive
].events
= 0;
419 super
.events_hi
= (devices
[most_recent
].events
>>32)&0xFFFFFFFF;
420 super
.events_lo
= (devices
[most_recent
].events
)&0xFFFFFFFF;
421 if (super
.level
== 5 || super
.level
== 4) {
422 /* need to force clean */
423 super
.state
= (1<<MD_SB_CLEAN
);
425 super
.sb_csum
= calc_sb_csum(&super
);
426 /*DRYRUN*/ if (store_super(fd
, &super
)) {
428 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
429 devices
[chosen_drive
].devname
);
430 devices
[chosen_drive
].events
= 0;
434 devices
[chosen_drive
].events
= devices
[most_recent
].events
;
435 devices
[chosen_drive
].uptodate
= 1;
439 /* Now we want to look at the superblock which the kernel will base things on
440 * and compare the devices that we think are working with the devices that the
441 * superblock thinks are working.
442 * If there are differences and --force is given, then update this chosen
446 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
451 if (!devices
[j
].uptodate
)
454 if ((fd
=open(devices
[j
].devname
, O_RDONLY
))< 0) {
455 fprintf(stderr
, Name
": Cannot open %s: %s\n",
456 devices
[j
].devname
, strerror(errno
));
459 if (load_super(fd
, &super
)) {
461 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
468 for (i
=0; i
<bestcnt
; i
++) {
472 if (i
< super
.raid_disks
)
473 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
479 if (!devices
[j
].uptodate
)
483 if (devices
[j
].major
!= super
.disks
[i
].major
||
484 devices
[j
].minor
!= super
.disks
[i
].minor
) {
486 super
.disks
[i
].major
= devices
[j
].major
;
487 super
.disks
[i
].minor
= devices
[j
].minor
;
490 if (devices
[j
].oldmajor
!= super
.disks
[i
].major
||
491 devices
[j
].oldminor
!= super
.disks
[i
].minor
) {
493 super
.disks
[i
].major
= devices
[j
].oldmajor
;
494 super
.disks
[i
].minor
= devices
[j
].oldminor
;
496 if (devices
[j
].uptodate
&&
497 (super
.disks
[i
].state
!= desired_state
)) {
499 fprintf(stderr
, Name
": "
500 "clearing FAULTY flag for device %d in %s for %s\n",
501 j
, mddev
, devices
[j
].devname
);
502 super
.disks
[i
].state
= desired_state
;
505 fprintf(stderr
, Name
": "
506 "device %d in %s has wrong state in superblock, but %s seems ok\n",
507 i
, mddev
, devices
[j
].devname
);
510 if (!devices
[j
].uptodate
&&
511 !(super
.disks
[i
].state
& (1 << MD_DISK_FAULTY
))) {
512 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
516 if (force
&& (super
.level
== 4 || super
.level
== 5) &&
517 okcnt
== super
.raid_disks
-1) {
518 super
.state
= (1<< MD_SB_CLEAN
);
522 if ((force
&& (change
& 2))
523 || (old_linux
&& (change
& 1))) {
525 super
.sb_csum
= calc_sb_csum(&super
);
526 fd
= open(devices
[chosen_drive
].devname
, O_RDWR
);
528 fprintf(stderr
, Name
": Could open %s for write - cannot Assemble array.\n",
529 devices
[chosen_drive
].devname
);
532 if (store_super(fd
, &super
)) {
534 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
535 devices
[chosen_drive
].devname
);
542 /* count number of in-sync devices according to the superblock.
543 * We must have this number to start the array without -s or -R
546 for (i
=0; i
<MD_SB_DISKS
; i
++)
547 if ((first_super
.disks
[i
].state
& (1<<MD_DISK_SYNC
)) &&
548 (first_super
.disks
[i
].state
& (1<<MD_DISK_ACTIVE
)) &&
549 !(first_super
.disks
[i
].state
& (1<<MD_DISK_FAULTY
)))
553 /* Almost ready to actually *do* something */
555 if (ioctl(mdfd
, SET_ARRAY_INFO
, NULL
) != 0) {
556 fprintf(stderr
, Name
": SET_ARRAY_INFO failed for %s: %s\n",
557 mddev
, strerror(errno
));
560 /* First, add the raid disks, but add the chosen one last */
561 for (i
=0; i
<= bestcnt
; i
++) {
565 if (j
== chosen_drive
)
570 if (j
>= 0 /* && devices[j].uptodate */) {
571 mdu_disk_info_t disk
;
572 memset(&disk
, 0, sizeof(disk
));
573 disk
.major
= devices
[j
].major
;
574 disk
.minor
= devices
[j
].minor
;
575 if (ioctl(mdfd
, ADD_NEW_DISK
, &disk
)!=0) {
576 fprintf(stderr
, Name
": failed to add %s to %s: %s\n",
580 if (i
< first_super
.raid_disks
)
585 fprintf(stderr
, Name
": added %s to %s as %d\n",
586 devices
[j
].devname
, mddev
, devices
[j
].raid_disk
);
587 } else if (verbose
&& i
< first_super
.raid_disks
)
588 fprintf(stderr
, Name
": no uptodate device for slot %d of %s\n",
594 ( enough(first_super
.level
, first_super
.raid_disks
, okcnt
) &&
595 (okcnt
>= req_cnt
|| start_partial_ok
)
597 if (ioctl(mdfd
, RUN_ARRAY
, NULL
)==0) {
598 fprintf(stderr
, Name
": %s has been started with %d drive%s",
599 mddev
, okcnt
, okcnt
==1?"":"s");
600 if (okcnt
< first_super
.raid_disks
)
601 fprintf(stderr
, " (out of %d)", first_super
.raid_disks
);
603 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
604 fprintf(stderr
, ".\n");
607 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
608 mddev
, strerror(errno
));
612 fprintf(stderr
, Name
": %s assembled from %d drive%s, but not started.\n",
613 mddev
, okcnt
, okcnt
==1?"":"s");
616 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
618 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
619 if (!enough(first_super
.level
, first_super
.raid_disks
, okcnt
))
620 fprintf(stderr
, " - not enough to start the array.\n");
622 if (req_cnt
== first_super
.raid_disks
)
623 fprintf(stderr
, " - need all %d to start it", req_cnt
);
625 fprintf(stderr
, " - need %d of %d to start", req_cnt
, first_super
.raid_disks
);
626 fprintf(stderr
, " (use --run to insist).\n");
630 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
631 * been updated to point to the current locations of devices.
632 * so we can just start the array
635 dev
= MKDEV(devices
[chosen_drive
].major
,
636 devices
[chosen_drive
].minor
);
637 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
638 fprintf(stderr
, Name
": Cannot start array: %s\n",