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;
101
unsigned int major, minor;
102
unsigned int oldmajor, oldminor;
109
int *best = NULL; /* indexed by raid_disk */
110
unsigned int bestcnt = 0;
112
unsigned int okcnt, sparecnt;
113
unsigned int req_cnt;
119
int start_partial_ok = force || devlist==NULL;
120
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, &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
best = malloc(num_devs * sizeof(*best));
169
devices = malloc(num_devs * sizeof(*devices));
171
first_super.md_magic = 0;
172
for (i=0; i<num_devs; i++)
176
fprintf(stderr, Name ": looking for devices for %s\n",
186
devname = devlist->devname;
187
devlist = devlist->next;
189
if (ident->devices &&
190
!match_oneof(ident->devices, devname)) {
191
if (inargv || verbose)
192
fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
196
dfd = open(devname, O_RDONLY, 0);
198
if (inargv || verbose)
199
fprintf(stderr, Name ": cannot open device %s: %s\n",
200
devname, strerror(errno));
201
} else if (fstat(dfd, &stb)< 0) {
203
fprintf(stderr, Name ": fstat failed for %s: %s\n",
204
devname, strerror(errno));
206
} else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
207
fprintf(stderr, Name ": %s is not a block device.\n",
210
} else if (load_super(dfd, &super)) {
211
if (inargv || verbose)
212
fprintf( stderr, Name ": no RAID superblock on %s\n",
217
uuid_from_super(this_uuid, &super);
221
if (ident->uuid_set &&
222
(!havesuper || same_uuid(this_uuid, ident->uuid)==0)) {
223
if (inargv || verbose)
224
fprintf(stderr, Name ": %s has wrong uuid.\n",
228
if (ident->super_minor != UnSet &&
229
(!havesuper || ident->super_minor != super.md_minor)) {
230
if (inargv || verbose)
231
fprintf(stderr, Name ": %s has wrong super-minor.\n",
235
if (ident->level != UnSet &&
236
(!havesuper|| ident->level != (int)super.level)) {
237
if (inargv || verbose)
238
fprintf(stderr, Name ": %s has wrong raid level.\n",
242
if (ident->raid_disks != UnSet &&
243
(!havesuper || ident->raid_disks!= super.raid_disks)) {
244
if (inargv || verbose)
245
fprintf(stderr, Name ": %s requires wrong number of drives.\n",
250
/* If we are this far, then we are commited to this device.
251
* If the super_block doesn't exist, or doesn't match others,
252
* then we cannot continue
256
fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
260
if (compare_super(&first_super, &super)) {
261
fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
267
/* this is needed until we get a more relaxed super block format */
268
if (devcnt >= MD_SB_DISKS) {
269
fprintf(stderr, Name ": ouch - too many devices appear to be in this array. Ignoring %s\n",
274
/* looks like a good enough match to update the super block if needed */
276
if (strcmp(update, "sparc2.2")==0 ) {
277
/* 2.2 sparc put the events in the wrong place
278
* So we copy the tail of the superblock
279
* up 4 bytes before continuing
281
__u32 *sb32 = (__u32*)&super;
282
memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
283
sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
284
(MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
285
fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
288
if (strcmp(update, "super-minor") ==0) {
291
super.md_minor = MINOR(stb2.st_rdev);
293
fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
294
devname, super.md_minor);
296
if (strcmp(update, "summaries") == 0) {
297
/* set nr_disks, active_disks, working_disks,
298
* failed_disks, spare_disks based on disks[]
299
* array in superblock.
300
* Also make sure extra slots aren't 'failed'
302
super.nr_disks = super.active_disks =
303
super.working_disks = super.failed_disks =
304
super.spare_disks = 0;
305
for (i=0; i < MD_SB_DISKS ; i++)
306
if (super.disks[i].major ||
307
super.disks[i].minor) {
308
int state = super.disks[i].state;
309
if (state & (1<<MD_DISK_REMOVED))
312
if (state & (1<<MD_DISK_ACTIVE))
313
super.active_disks++;
314
if (state & (1<<MD_DISK_FAULTY))
315
super.failed_disks++;
317
super.working_disks++;
320
} else if (i >= super.raid_disks && super.disks[i].number == 0)
321
super.disks[i].state = 0;
323
super.sb_csum = calc_sb_csum(&super);
324
dfd = open(devname, O_RDWR, 0);
326
fprintf(stderr, Name ": Cannot open %s for superblock update\n",
328
else if (store_super(dfd, &super))
329
fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
336
fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
337
devname, mddev, super.this_disk.raid_disk);
338
devices[devcnt].devname = devname;
339
devices[devcnt].major = MAJOR(stb.st_rdev);
340
devices[devcnt].minor = MINOR(stb.st_rdev);
341
devices[devcnt].oldmajor = super.this_disk.major;
342
devices[devcnt].oldminor = super.this_disk.minor;
343
devices[devcnt].events = md_event(&super);
344
devices[devcnt].utime = super.utime;
345
devices[devcnt].raid_disk = super.this_disk.raid_disk;
346
devices[devcnt].uptodate = 0;
347
devices[devcnt].state = super.this_disk.state;
348
if (most_recent < devcnt) {
349
if (devices[devcnt].events
350
> devices[most_recent].events)
351
most_recent = devcnt;
353
if ((int)super.level == -4)
354
/* with multipath, the raid_disk from the superblock is meaningless */
357
i = devices[devcnt].raid_disk;
360
unsigned int newbestcnt = i+10;
361
int *newbest = malloc(sizeof(int)*newbestcnt);
363
for (c=0; c < newbestcnt; c++)
365
newbest[c] = best[c];
370
bestcnt = newbestcnt;
373
|| devices[best[i]].events < devices[devcnt].events)
380
fprintf(stderr, Name ": no devices found for %s\n",
384
/* now we have some devices that might be suitable.
389
for (i=0; i< bestcnt ;i++) {
391
int event_margin = !force;
393
/* note: we ignore error flags in multipath arrays
394
* as they don't make sense
396
if ((int)first_super.level != -4)
397
if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
398
if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
402
if (devices[j].events+event_margin >=
403
devices[most_recent].events) {
404
devices[j].uptodate = 1;
405
if (i < first_super.raid_disks)
411
while (force && !enough(first_super.level, first_super.raid_disks, okcnt)) {
412
/* Choose the newest best drive which is
413
* not up-to-date, update the superblock
418
for (i=0; i<first_super.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)
429
fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
430
devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
431
(int)(devices[chosen_drive].events),
432
(int)(devices[most_recent].events));
433
fd = open(devices[chosen_drive].devname, O_RDWR);
435
fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
436
devices[chosen_drive].devname);
437
devices[chosen_drive].events = 0;
440
if (load_super(fd, &super)) {
442
fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
443
devices[chosen_drive].devname);
444
devices[chosen_drive].events = 0;
447
super.events_hi = (devices[most_recent].events>>32)&0xFFFFFFFF;
448
super.events_lo = (devices[most_recent].events)&0xFFFFFFFF;
449
if (super.level == 5 || super.level == 4) {
450
/* need to force clean */
451
super.state = (1<<MD_SB_CLEAN);
453
super.sb_csum = calc_sb_csum(&super);
454
/*DRYRUN*/ if (store_super(fd, &super)) {
456
fprintf(stderr, Name ": Could not re-write superblock on %s\n",
457
devices[chosen_drive].devname);
458
devices[chosen_drive].events = 0;
462
devices[chosen_drive].events = devices[most_recent].events;
463
devices[chosen_drive].uptodate = 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
474
for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
479
if (!devices[j].uptodate)
482
if ((fd=open(devices[j].devname, O_RDONLY))< 0) {
483
fprintf(stderr, Name ": Cannot open %s: %s\n",
484
devices[j].devname, strerror(errno));
487
if (load_super(fd, &super)) {
489
fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
496
for (i=0; i<bestcnt; i++) {
498
unsigned int desired_state;
500
if (i < super.raid_disks)
501
desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
507
if (!devices[j].uptodate)
511
if (devices[j].major != super.disks[i].major ||
512
devices[j].minor != super.disks[i].minor) {
514
super.disks[i].major = devices[j].major;
515
super.disks[i].minor = devices[j].minor;
518
if (devices[j].oldmajor != super.disks[i].major ||
519
devices[j].oldminor != super.disks[i].minor) {
521
super.disks[i].major = devices[j].oldmajor;
522
super.disks[i].minor = devices[j].oldminor;
524
if (devices[j].uptodate &&
525
(super.disks[i].state != desired_state)) {
527
fprintf(stderr, Name ": "
528
"clearing FAULTY flag for device %d in %s for %s\n",
529
j, mddev, devices[j].devname);
530
super.disks[i].state = desired_state;
533
fprintf(stderr, Name ": "
534
"device %d in %s has wrong state in superblock, but %s seems ok\n",
535
i, mddev, devices[j].devname);
538
if (!devices[j].uptodate &&
539
!(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
540
fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
544
if (force && (super.level == 4 || super.level == 5) &&
545
okcnt == super.raid_disks-1) {
546
super.state = (1<< MD_SB_CLEAN);
550
if ((force && (change & 2))
551
|| (old_linux && (change & 1))) {
553
super.sb_csum = calc_sb_csum(&super);
554
fd = open(devices[chosen_drive].devname, O_RDWR);
556
fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
557
devices[chosen_drive].devname);
560
if (store_super(fd, &super)) {
562
fprintf(stderr, Name ": Could not re-write superblock on %s\n",
563
devices[chosen_drive].devname);
570
/* count number of in-sync devices according to the superblock.
571
* We must have this number to start the array without -s or -R
574
for (i=0; i<MD_SB_DISKS; i++)
575
if ((first_super.disks[i].state & (1<<MD_DISK_SYNC)) &&
576
(first_super.disks[i].state & (1<<MD_DISK_ACTIVE)) &&
577
!(first_super.disks[i].state & (1<<MD_DISK_FAULTY)))
581
/* Almost ready to actually *do* something */
583
if (ioctl(mdfd, SET_ARRAY_INFO, NULL) != 0) {
584
fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
585
mddev, strerror(errno));
588
/* First, add the raid disks, but add the chosen one last */
589
for (i=0; i<= bestcnt; i++) {
593
if (j == chosen_drive)
598
if (j >= 0 /* && devices[j].uptodate */) {
599
mdu_disk_info_t disk;
600
memset(&disk, 0, sizeof(disk));
601
disk.major = devices[j].major;
602
disk.minor = devices[j].minor;
603
if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
604
fprintf(stderr, Name ": failed to add %s to %s: %s\n",
608
if (i < first_super.raid_disks)
613
fprintf(stderr, Name ": added %s to %s as %d\n",
614
devices[j].devname, mddev, devices[j].raid_disk);
615
} else if (verbose && i < first_super.raid_disks)
616
fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
622
( enough(first_super.level, first_super.raid_disks, okcnt) &&
623
(okcnt >= req_cnt || start_partial_ok)
625
if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
626
fprintf(stderr, Name ": %s has been started with %d drive%s",
627
mddev, okcnt, okcnt==1?"":"s");
628
if (okcnt < first_super.raid_disks)
629
fprintf(stderr, " (out of %d)", first_super.raid_disks);
631
fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
632
fprintf(stderr, ".\n");
635
fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
636
mddev, strerror(errno));
640
fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
641
mddev, okcnt, okcnt==1?"":"s");
644
fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
646
fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
647
if (!enough(first_super.level, first_super.raid_disks, okcnt))
648
fprintf(stderr, " - not enough to start the array.\n");
650
if (req_cnt == first_super.raid_disks)
651
fprintf(stderr, " - need all %d to start it", req_cnt);
653
fprintf(stderr, " - need %d of %d to start", req_cnt, first_super.raid_disks);
654
fprintf(stderr, " (use --run to insist).\n");
658
/* The "chosen_drive" is a good choice, and if necessary, the superblock has
659
* been updated to point to the current locations of devices.
660
* so we can just start the array
663
dev = MKDEV(devices[chosen_drive].major,
664
devices[chosen_drive].minor);
665
if (ioctl(mdfd, START_ARRAY, dev)) {
666
fprintf(stderr, Name ": Cannot start array: %s\n",