]>
Commit | Line | Data |
---|---|---|
64c4757e | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
64c4757e | 3 | * |
4f589ad0 | 4 | * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de> |
64c4757e NB |
5 | * |
6 | * | |
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. | |
11 | * | |
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. | |
16 | * | |
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 | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neilb@cse.unsw.edu.au> | |
23 | * Paper: Neil Brown | |
24 | * School of Computer Science and Engineering | |
25 | * The University of New South Wales | |
26 | * Sydney, 2052 | |
27 | * Australia | |
28 | */ | |
29 | ||
9a9dab36 | 30 | #include "mdadm.h" |
64c4757e | 31 | |
82d9eba6 | 32 | int Assemble(struct supertype *st, char *mddev, int mdfd, |
52826846 | 33 | mddev_ident_t ident, char *conffile, |
06b0d786 | 34 | mddev_dev_t devlist, char *backup_file, |
64c4757e | 35 | int readonly, int runstop, |
e5eac01f | 36 | char *update, char *homehost, |
64c4757e NB |
37 | int verbose, int force) |
38 | { | |
39 | /* | |
52826846 NB |
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 | |
64c4757e NB |
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. | |
49 | * | |
50 | * Much of the work of Assemble is in finding and/or | |
51 | * checking the disks to make sure they look right. | |
52 | * | |
4b1ac34b | 53 | * If mddev is not set, then scan must be set and we |
64c4757e NB |
54 | * read through the config file for dev+uuid mapping |
55 | * We recurse, setting mddev, for each device that | |
56 | * - isn't running | |
4b1ac34b | 57 | * - has a valid uuid (or any uuid if !uuidset) |
64c4757e NB |
58 | * |
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 | |
66 | * | |
67 | * If !uuidset and scan, look in conf-file for uuid | |
68 | * If not found, give up | |
b8a8ccf9 | 69 | * If !devlist and scan and uuidset, get list of devs from conf-file |
64c4757e NB |
70 | * |
71 | * For each device: | |
72 | * Check superblock - discard if bad | |
73 | * Check uuid (set if we don't have one) - discard if no match | |
5787fa49 | 74 | * Check superblock similarity if we have a superblock - discard if different |
4b1ac34b | 75 | * Record events, devicenum |
64c4757e | 76 | * This should give us a list of devices for the array |
4b1ac34b | 77 | * We should collect the most recent event number |
64c4757e NB |
78 | * |
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 | |
83 | * If recent kernel: | |
84 | * SET_ARRAY_INFO | |
85 | * foreach device with recent events : ADD_NEW_DISK | |
86 | * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY | |
87 | * If old kernel: | |
88 | * Check the device numbers in superblock are right | |
89 | * update superblock if any changes | |
90 | * START_ARRAY | |
91 | * | |
92 | */ | |
93 | int old_linux = 0; | |
94 | int vers; | |
4b1ac34b | 95 | void *first_super = NULL, *super = NULL; |
64c4757e NB |
96 | struct { |
97 | char *devname; | |
98c6faba NB |
98 | unsigned int major, minor; |
99 | unsigned int oldmajor, oldminor; | |
64c4757e | 100 | long long events; |
64c4757e | 101 | int uptodate; |
d013a55e | 102 | int state; |
52826846 | 103 | int raid_disk; |
6a41304b | 104 | int disk_nr; |
5787fa49 | 105 | } *devices; |
56eedc1a | 106 | int *best = NULL; /* indexed by raid_disk */ |
98c6faba NB |
107 | unsigned int bestcnt = 0; |
108 | int devcnt = 0; | |
109 | unsigned int okcnt, sparecnt; | |
110 | unsigned int req_cnt; | |
111 | unsigned int i; | |
64c4757e | 112 | int most_recent = 0; |
cd29a5c8 | 113 | int chosen_drive; |
52826846 | 114 | int change = 0; |
cd29a5c8 | 115 | int inargv = 0; |
b8a8ccf9 | 116 | int start_partial_ok = (runstop >= 0) && (force || devlist==NULL); |
98c6faba | 117 | unsigned int num_devs; |
5787fa49 | 118 | mddev_dev_t tmpdev; |
4b1ac34b | 119 | struct mdinfo info; |
265e0f17 | 120 | char *avail; |
308e1801 | 121 | int nextspare = 0; |
64c4757e | 122 | |
64c4757e NB |
123 | vers = md_get_version(mdfd); |
124 | if (vers <= 0) { | |
e0d19036 | 125 | fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev); |
64c4757e NB |
126 | return 1; |
127 | } | |
682c7051 NB |
128 | if (vers < 9000) { |
129 | fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n" | |
52826846 | 130 | " Upgrade your kernel or try --build\n"); |
64c4757e NB |
131 | return 1; |
132 | } | |
682c7051 | 133 | if (get_linux_version() < 2004000) |
64c4757e NB |
134 | old_linux = 1; |
135 | ||
4b1ac34b | 136 | if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) { |
682c7051 | 137 | fprintf(stderr, Name ": device %s already active - cannot assemble it\n", |
64c4757e NB |
138 | mddev); |
139 | return 1; | |
140 | } | |
141 | ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */ | |
142 | ||
143 | /* | |
52826846 NB |
144 | * If any subdevs are listed, then any that don't |
145 | * match ident are discarded. Remainder must all match and | |
146 | * become the array. | |
147 | * If no subdevs, then we scan all devices in the config file, but | |
148 | * there must be something in the identity | |
64c4757e | 149 | */ |
64c4757e | 150 | |
cd29a5c8 | 151 | if (!devlist && |
52826846 NB |
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", | |
156 | mddev); | |
157 | return 1; | |
64c4757e | 158 | } |
cd29a5c8 | 159 | if (devlist == NULL) |
64c4757e | 160 | devlist = conf_get_devs(conffile); |
cd29a5c8 | 161 | else inargv = 1; |
64c4757e | 162 | |
5787fa49 NB |
163 | tmpdev = devlist; num_devs = 0; |
164 | while (tmpdev) { | |
165 | num_devs++; | |
166 | tmpdev = tmpdev->next; | |
167 | } | |
5787fa49 NB |
168 | devices = malloc(num_devs * sizeof(*devices)); |
169 | ||
82d9eba6 | 170 | if (!st && ident->st) st = ident->st; |
64c4757e | 171 | |
dab6685f | 172 | if (verbose>0) |
82b27616 NB |
173 | fprintf(stderr, Name ": looking for devices for %s\n", |
174 | mddev); | |
175 | ||
cd29a5c8 | 176 | while ( devlist) { |
64c4757e | 177 | char *devname; |
64c4757e NB |
178 | int dfd; |
179 | struct stat stb; | |
82d9eba6 | 180 | struct supertype *tst = st; |
52826846 | 181 | |
cd29a5c8 NB |
182 | devname = devlist->devname; |
183 | devlist = devlist->next; | |
64c4757e | 184 | |
52826846 | 185 | if (ident->devices && |
56eedc1a | 186 | !match_oneof(ident->devices, devname)) { |
6a41304b | 187 | if ((inargv && verbose>=0) || verbose > 0) |
56eedc1a | 188 | fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices); |
52826846 | 189 | continue; |
56eedc1a | 190 | } |
4b1ac34b NB |
191 | |
192 | if (super) { | |
193 | free(super); | |
194 | super = NULL; | |
195 | } | |
52826846 | 196 | |
8b0dabea | 197 | dfd = dev_open(devname, O_RDONLY|O_EXCL); |
64c4757e | 198 | if (dfd < 0) { |
6a41304b | 199 | if ((inargv && verbose >= 0) || verbose > 0) |
682c7051 | 200 | fprintf(stderr, Name ": cannot open device %s: %s\n", |
64c4757e | 201 | devname, strerror(errno)); |
52826846 NB |
202 | } else if (fstat(dfd, &stb)< 0) { |
203 | /* Impossible! */ | |
204 | fprintf(stderr, Name ": fstat failed for %s: %s\n", | |
205 | devname, strerror(errno)); | |
cd29a5c8 NB |
206 | } else if ((stb.st_mode & S_IFMT) != S_IFBLK) { |
207 | fprintf(stderr, Name ": %s is not a block device.\n", | |
52826846 | 208 | devname); |
82d9eba6 | 209 | } else if (!tst && (tst = guess_super(dfd)) == NULL) { |
6a41304b | 210 | if ((inargv && verbose >= 0) || verbose > 0) |
f9ce90ba | 211 | fprintf(stderr, Name ": no recogniseable superblock\n"); |
82d9eba6 | 212 | } else if (tst->ss->load_super(tst,dfd, &super, NULL)) { |
6a41304b | 213 | if ((inargv && verbose >= 0) || verbose > 0) |
682c7051 | 214 | fprintf( stderr, Name ": no RAID superblock on %s\n", |
64c4757e | 215 | devname); |
52826846 | 216 | } else { |
31317663 | 217 | tst->ss->getinfo_super(&info, super); |
64c4757e | 218 | } |
c06487ce | 219 | if (dfd >= 0) close(dfd); |
52826846 | 220 | |
838acbc2 | 221 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && |
f277ce36 | 222 | (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) { |
6a41304b | 223 | if ((inargv && verbose >= 0) || verbose > 0) |
52826846 NB |
224 | fprintf(stderr, Name ": %s has wrong uuid.\n", |
225 | devname); | |
226 | continue; | |
82b27616 | 227 | } |
947fd4dd | 228 | if (ident->name[0] && |
31317663 | 229 | (!super || strncmp(info.name, ident->name, 32)!=0)) { |
6a41304b | 230 | if ((inargv && verbose >= 0) || verbose > 0) |
947fd4dd NB |
231 | fprintf(stderr, Name ": %s has wrong name.\n", |
232 | devname); | |
233 | continue; | |
234 | } | |
98c6faba | 235 | if (ident->super_minor != UnSet && |
4b1ac34b | 236 | (!super || ident->super_minor != info.array.md_minor)) { |
6a41304b | 237 | if ((inargv && verbose >= 0) || verbose > 0) |
52826846 | 238 | fprintf(stderr, Name ": %s has wrong super-minor.\n", |
64c4757e NB |
239 | devname); |
240 | continue; | |
241 | } | |
98c6faba | 242 | if (ident->level != UnSet && |
4b1ac34b | 243 | (!super|| ident->level != info.array.level)) { |
6a41304b | 244 | if ((inargv && verbose >= 0) || verbose > 0) |
cd29a5c8 NB |
245 | fprintf(stderr, Name ": %s has wrong raid level.\n", |
246 | devname); | |
247 | continue; | |
248 | } | |
98c6faba | 249 | if (ident->raid_disks != UnSet && |
4b1ac34b | 250 | (!super || ident->raid_disks!= info.array.raid_disks)) { |
6a41304b | 251 | if ((inargv && verbose >= 0) || verbose > 0) |
cd29a5c8 NB |
252 | fprintf(stderr, Name ": %s requires wrong number of drives.\n", |
253 | devname); | |
254 | continue; | |
255 | } | |
52826846 NB |
256 | |
257 | /* If we are this far, then we are commited to this device. | |
258 | * If the super_block doesn't exist, or doesn't match others, | |
259 | * then we cannot continue | |
260 | */ | |
52826846 | 261 | |
4b1ac34b | 262 | if (!super) { |
52826846 NB |
263 | fprintf(stderr, Name ": %s has no superblock - assembly aborted\n", |
264 | devname); | |
4b1ac34b | 265 | free(first_super); |
52826846 NB |
266 | return 1; |
267 | } | |
838acbc2 NB |
268 | |
269 | ||
82d9eba6 NB |
270 | st = tst; /* commit to this format, if haven't already */ |
271 | if (st->ss->compare_super(&first_super, super)) { | |
52826846 NB |
272 | fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n", |
273 | devname); | |
4b1ac34b NB |
274 | free(super); |
275 | free(first_super); | |
52826846 | 276 | return 1; |
64c4757e NB |
277 | } |
278 | ||
5787fa49 NB |
279 | /* looks like a good enough match to update the super block if needed */ |
280 | if (update) { | |
4b1ac34b NB |
281 | /* prepare useful information in info structures */ |
282 | struct stat stb2; | |
283 | fstat(mdfd, &stb2); | |
284 | info.array.md_minor = minor(stb2.st_rdev); | |
7d99579f NB |
285 | |
286 | if (strcmp(update, "uuid")==0 && | |
287 | !ident->uuid_set) { | |
288 | int rfd; | |
289 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
290 | read(rfd, ident->uuid, 16) != 16) { | |
291 | *(__u32*)(ident->uuid) = random(); | |
292 | *(__u32*)(ident->uuid+1) = random(); | |
293 | *(__u32*)(ident->uuid+2) = random(); | |
294 | *(__u32*)(ident->uuid+3) = random(); | |
295 | } | |
296 | if (rfd >= 0) close(rfd); | |
7d99579f NB |
297 | } |
298 | memcpy(info.uuid, ident->uuid, 16); | |
e5eac01f NB |
299 | strcpy(info.name, ident->name); |
300 | st->ss->update_super(&info, super, update, devname, verbose, | |
301 | ident->uuid_set, homehost); | |
302 | if (strcmp(update, "uuid")==0 && | |
303 | !ident->uuid_set) { | |
304 | ident->uuid_set = 1; | |
305 | memcpy(ident->uuid, info.uuid, 16); | |
306 | } | |
8b0dabea | 307 | dfd = dev_open(devname, O_RDWR|O_EXCL); |
b8a8ccf9 | 308 | if (dfd < 0) |
5787fa49 NB |
309 | fprintf(stderr, Name ": Cannot open %s for superblock update\n", |
310 | devname); | |
96395475 | 311 | else if (st->ss->store_super(st, dfd, super)) |
5787fa49 NB |
312 | fprintf(stderr, Name ": Could not re-write superblock on %s.\n", |
313 | devname); | |
314 | if (dfd >= 0) | |
315 | close(dfd); | |
8131b493 NB |
316 | |
317 | if (strcmp(update, "uuid")==0 && | |
318 | ident->bitmap_fd) | |
319 | bitmap_update_uuid(ident->bitmap_fd, info.uuid); | |
5787fa49 NB |
320 | } |
321 | ||
dab6685f | 322 | if (verbose > 0) |
cd29a5c8 | 323 | fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n", |
4b1ac34b | 324 | devname, mddev, info.disk.raid_disk); |
64c4757e | 325 | devices[devcnt].devname = devname; |
0df46c2a NB |
326 | devices[devcnt].major = major(stb.st_rdev); |
327 | devices[devcnt].minor = minor(stb.st_rdev); | |
4b1ac34b NB |
328 | devices[devcnt].oldmajor = info.disk.major; |
329 | devices[devcnt].oldminor = info.disk.minor; | |
330 | devices[devcnt].events = info.events; | |
331 | devices[devcnt].raid_disk = info.disk.raid_disk; | |
6a41304b | 332 | devices[devcnt].disk_nr = info.disk.number; |
64c4757e | 333 | devices[devcnt].uptodate = 0; |
4b1ac34b | 334 | devices[devcnt].state = info.disk.state; |
64c4757e NB |
335 | if (most_recent < devcnt) { |
336 | if (devices[devcnt].events | |
337 | > devices[most_recent].events) | |
338 | most_recent = devcnt; | |
339 | } | |
b8a8ccf9 | 340 | if (info.array.level == -4) |
5787fa49 NB |
341 | /* with multipath, the raid_disk from the superblock is meaningless */ |
342 | i = devcnt; | |
343 | else | |
344 | i = devices[devcnt].raid_disk; | |
308e1801 NB |
345 | if (i+1 == 0) { |
346 | if (nextspare < info.array.raid_disks) | |
347 | nextspare = info.array.raid_disks; | |
348 | i = nextspare++; | |
349 | } | |
98c6faba | 350 | if (i < 10000) { |
56eedc1a | 351 | if (i >= bestcnt) { |
98c6faba | 352 | unsigned int newbestcnt = i+10; |
56eedc1a | 353 | int *newbest = malloc(sizeof(int)*newbestcnt); |
98c6faba | 354 | unsigned int c; |
56eedc1a NB |
355 | for (c=0; c < newbestcnt; c++) |
356 | if (c < bestcnt) | |
357 | newbest[c] = best[c]; | |
358 | else | |
359 | newbest[c] = -1; | |
360 | if (best)free(best); | |
361 | best = newbest; | |
362 | bestcnt = newbestcnt; | |
363 | } | |
52826846 NB |
364 | if (best[i] == -1 |
365 | || devices[best[i]].events < devices[devcnt].events) | |
366 | best[i] = devcnt; | |
56eedc1a | 367 | } |
64c4757e NB |
368 | devcnt++; |
369 | } | |
370 | ||
4b1ac34b NB |
371 | if (super) |
372 | free(super); | |
373 | super = NULL; | |
374 | ||
586ed405 NB |
375 | if (update && strcmp(update, "byteorder")==0) |
376 | st->minor_version = 90; | |
377 | ||
64c4757e | 378 | if (devcnt == 0) { |
682c7051 | 379 | fprintf(stderr, Name ": no devices found for %s\n", |
64c4757e | 380 | mddev); |
4b1ac34b | 381 | free(first_super); |
64c4757e NB |
382 | return 1; |
383 | } | |
4b1ac34b | 384 | |
31317663 | 385 | st->ss->getinfo_super(&info, first_super); |
4b1ac34b | 386 | |
64c4757e NB |
387 | /* now we have some devices that might be suitable. |
388 | * I wonder how many | |
389 | */ | |
265e0f17 NB |
390 | avail = malloc(info.array.raid_disks); |
391 | memset(avail, 0, info.array.raid_disks); | |
64c4757e | 392 | okcnt = 0; |
52826846 | 393 | sparecnt=0; |
56eedc1a | 394 | for (i=0; i< bestcnt ;i++) { |
64c4757e | 395 | int j = best[i]; |
ee04451c NB |
396 | int event_margin = 1; /* always allow a difference of '1' |
397 | * like the kernel does | |
398 | */ | |
64c4757e | 399 | if (j < 0) continue; |
d013a55e NB |
400 | /* note: we ignore error flags in multipath arrays |
401 | * as they don't make sense | |
402 | */ | |
4b1ac34b | 403 | if (info.array.level != -4) |
aa88f531 NB |
404 | if (!(devices[j].state & (1<<MD_DISK_SYNC))) { |
405 | if (!(devices[j].state & (1<<MD_DISK_FAULTY))) | |
406 | sparecnt++; | |
d013a55e | 407 | continue; |
aa88f531 | 408 | } |
cd29a5c8 | 409 | if (devices[j].events+event_margin >= |
64c4757e NB |
410 | devices[most_recent].events) { |
411 | devices[j].uptodate = 1; | |
265e0f17 | 412 | if (i < info.array.raid_disks) { |
52826846 | 413 | okcnt++; |
265e0f17 NB |
414 | avail[i]=1; |
415 | } else | |
52826846 | 416 | sparecnt++; |
64c4757e NB |
417 | } |
418 | } | |
265e0f17 NB |
419 | while (force && !enough(info.array.level, info.array.raid_disks, |
420 | info.array.layout, | |
421 | avail, okcnt)) { | |
64c4757e NB |
422 | /* Choose the newest best drive which is |
423 | * not up-to-date, update the superblock | |
424 | * and add it. | |
425 | */ | |
52826846 | 426 | int fd; |
cd29a5c8 | 427 | chosen_drive = -1; |
4b1ac34b | 428 | for (i=0; i<info.array.raid_disks && i < bestcnt; i++) { |
52826846 NB |
429 | int j = best[i]; |
430 | if (j>=0 && | |
431 | !devices[j].uptodate && | |
432 | devices[j].events > 0 && | |
433 | (chosen_drive < 0 || | |
434 | devices[j].events > devices[chosen_drive].events)) | |
435 | chosen_drive = j; | |
436 | } | |
437 | if (chosen_drive < 0) | |
438 | break; | |
dab6685f NB |
439 | if (verbose >= 0) |
440 | fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n", | |
441 | devices[chosen_drive].devname, devices[chosen_drive].raid_disk, | |
442 | (int)(devices[chosen_drive].events), | |
443 | (int)(devices[most_recent].events)); | |
8b0dabea | 444 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 NB |
445 | if (fd < 0) { |
446 | fprintf(stderr, Name ": Couldn't open %s for write - not updating\n", | |
447 | devices[chosen_drive].devname); | |
448 | devices[chosen_drive].events = 0; | |
449 | continue; | |
450 | } | |
82d9eba6 | 451 | if (st->ss->load_super(st,fd, &super, NULL)) { |
52826846 NB |
452 | close(fd); |
453 | fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n", | |
454 | devices[chosen_drive].devname); | |
455 | devices[chosen_drive].events = 0; | |
456 | continue; | |
457 | } | |
4b1ac34b | 458 | info.events = devices[most_recent].events; |
e5eac01f | 459 | st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL); |
4b1ac34b | 460 | |
96395475 | 461 | if (st->ss->store_super(st, fd, super)) { |
52826846 NB |
462 | close(fd); |
463 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
464 | devices[chosen_drive].devname); | |
465 | devices[chosen_drive].events = 0; | |
4b1ac34b | 466 | free(super); |
52826846 NB |
467 | continue; |
468 | } | |
469 | close(fd); | |
470 | devices[chosen_drive].events = devices[most_recent].events; | |
471 | devices[chosen_drive].uptodate = 1; | |
265e0f17 | 472 | avail[chosen_drive] = 1; |
52826846 | 473 | okcnt++; |
4b1ac34b | 474 | free(super); |
64c4757e | 475 | } |
52826846 NB |
476 | |
477 | /* Now we want to look at the superblock which the kernel will base things on | |
478 | * and compare the devices that we think are working with the devices that the | |
479 | * superblock thinks are working. | |
480 | * If there are differences and --force is given, then update this chosen | |
481 | * superblock. | |
482 | */ | |
cd29a5c8 | 483 | chosen_drive = -1; |
4b1ac34b | 484 | super = NULL; |
56eedc1a | 485 | for (i=0; chosen_drive < 0 && i<bestcnt; i++) { |
52826846 NB |
486 | int j = best[i]; |
487 | int fd; | |
4b1ac34b | 488 | |
52826846 NB |
489 | if (j<0) |
490 | continue; | |
491 | if (!devices[j].uptodate) | |
492 | continue; | |
493 | chosen_drive = j; | |
8b0dabea | 494 | if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) { |
52826846 NB |
495 | fprintf(stderr, Name ": Cannot open %s: %s\n", |
496 | devices[j].devname, strerror(errno)); | |
497 | return 1; | |
498 | } | |
82d9eba6 | 499 | if (st->ss->load_super(st,fd, &super, NULL)) { |
52826846 NB |
500 | close(fd); |
501 | fprintf(stderr, Name ": RAID superblock has disappeared from %s\n", | |
502 | devices[j].devname); | |
503 | return 1; | |
504 | } | |
505 | close(fd); | |
506 | } | |
4b1ac34b NB |
507 | if (super == NULL) { |
508 | fprintf(stderr, Name ": No suitable drives found for %s\n", mddev); | |
509 | return 1; | |
510 | } | |
31317663 | 511 | st->ss->getinfo_super(&info, super); |
56eedc1a | 512 | for (i=0; i<bestcnt; i++) { |
52826846 | 513 | int j = best[i]; |
98c6faba | 514 | unsigned int desired_state; |
11a3e71d | 515 | |
4b1ac34b | 516 | if (i < info.array.raid_disks) |
11a3e71d NB |
517 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); |
518 | else | |
519 | desired_state = 0; | |
520 | ||
52826846 NB |
521 | if (j<0) |
522 | continue; | |
523 | if (!devices[j].uptodate) | |
524 | continue; | |
6a41304b | 525 | info.disk.number = devices[j].disk_nr; |
fbf8a0b7 | 526 | info.disk.raid_disk = i; |
4b1ac34b NB |
527 | info.disk.state = desired_state; |
528 | ||
52826846 | 529 | if (devices[j].uptodate && |
e5eac01f | 530 | st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) { |
52826846 | 531 | if (force) { |
dab6685f NB |
532 | if (verbose >= 0) |
533 | fprintf(stderr, Name ": " | |
534 | "clearing FAULTY flag for device %d in %s for %s\n", | |
535 | j, mddev, devices[j].devname); | |
4b1ac34b | 536 | change = 1; |
52826846 | 537 | } else { |
dab6685f NB |
538 | if (verbose >= -1) |
539 | fprintf(stderr, Name ": " | |
540 | "device %d in %s has wrong state in superblock, but %s seems ok\n", | |
541 | i, mddev, devices[j].devname); | |
52826846 NB |
542 | } |
543 | } | |
4b1ac34b | 544 | #if 0 |
52826846 NB |
545 | if (!devices[j].uptodate && |
546 | !(super.disks[i].state & (1 << MD_DISK_FAULTY))) { | |
547 | fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", | |
548 | i, mddev); | |
549 | } | |
4b1ac34b | 550 | #endif |
52826846 | 551 | } |
4b1ac34b NB |
552 | if (force && okcnt == info.array.raid_disks-1) { |
553 | /* FIXME check event count */ | |
b8a8ccf9 | 554 | change += st->ss->update_super(&info, super, "force", |
e5eac01f | 555 | devices[chosen_drive].devname, verbose, 0, NULL); |
d013a55e | 556 | } |
52826846 | 557 | |
4b1ac34b | 558 | if (change) { |
52826846 | 559 | int fd; |
8b0dabea | 560 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 | 561 | if (fd < 0) { |
353632d9 | 562 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", |
52826846 NB |
563 | devices[chosen_drive].devname); |
564 | return 1; | |
565 | } | |
96395475 | 566 | if (st->ss->store_super(st, fd, super)) { |
52826846 NB |
567 | close(fd); |
568 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
569 | devices[chosen_drive].devname); | |
570 | return 1; | |
571 | } | |
572 | close(fd); | |
52826846 NB |
573 | } |
574 | ||
353632d9 NB |
575 | /* If we are in the middle of a reshape we may need to restore saved data |
576 | * that was moved aside due to the reshape overwriting live data | |
577 | * The code of doing this lives in Grow.c | |
578 | */ | |
39c5a909 | 579 | #ifndef MDASSEMBLE |
353632d9 NB |
580 | if (info.reshape_active) { |
581 | int err = 0; | |
582 | int *fdlist = malloc(sizeof(int)* bestcnt); | |
583 | for (i=0; i<bestcnt; i++) { | |
584 | int j = best[i]; | |
585 | if (j >= 0) { | |
586 | fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
587 | if (fdlist[i] < 0) { | |
588 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", | |
589 | devices[j].devname); | |
590 | err = 1; | |
591 | break; | |
592 | } | |
593 | } else | |
594 | fdlist[i] = -1; | |
595 | } | |
596 | if (!err) | |
06b0d786 | 597 | err = Grow_restart(st, &info, fdlist, bestcnt, backup_file); |
353632d9 NB |
598 | while (i>0) { |
599 | i--; | |
600 | if (fdlist[i]>=0) close(fdlist[i]); | |
601 | } | |
602 | if (err) { | |
603 | fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n"); | |
604 | return err; | |
605 | } | |
606 | } | |
39c5a909 | 607 | #endif |
aa88f531 NB |
608 | /* count number of in-sync devices according to the superblock. |
609 | * We must have this number to start the array without -s or -R | |
610 | */ | |
4b1ac34b | 611 | req_cnt = info.array.working_disks; |
aa88f531 | 612 | |
64c4757e NB |
613 | /* Almost ready to actually *do* something */ |
614 | if (!old_linux) { | |
fbf8a0b7 NB |
615 | int rv; |
616 | if ((vers % 100) >= 1) { /* can use different versions */ | |
617 | mdu_array_info_t inf; | |
618 | memset(&inf, 0, sizeof(inf)); | |
619 | inf.major_version = st->ss->major; | |
620 | inf.minor_version = st->minor_version; | |
621 | rv = ioctl(mdfd, SET_ARRAY_INFO, &inf); | |
b8a8ccf9 | 622 | } else |
fbf8a0b7 NB |
623 | rv = ioctl(mdfd, SET_ARRAY_INFO, NULL); |
624 | ||
625 | if (rv) { | |
682c7051 | 626 | fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n", |
64c4757e NB |
627 | mddev, strerror(errno)); |
628 | return 1; | |
629 | } | |
11484a63 | 630 | if (ident->bitmap_fd >= 0) { |
c82f047c NB |
631 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { |
632 | fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n"); | |
633 | return 1; | |
634 | } | |
7ef02d01 NB |
635 | } else if (ident->bitmap_file) { |
636 | /* From config file */ | |
637 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
638 | if (bmfd < 0) { | |
639 | fprintf(stderr, Name ": Could not open bitmap file %s\n", | |
640 | ident->bitmap_file); | |
641 | return 1; | |
642 | } | |
643 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
644 | fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev); | |
645 | close(bmfd); | |
646 | return 1; | |
647 | } | |
648 | close(bmfd); | |
c82f047c | 649 | } |
7ef02d01 | 650 | |
52826846 | 651 | /* First, add the raid disks, but add the chosen one last */ |
56eedc1a | 652 | for (i=0; i<= bestcnt; i++) { |
52826846 | 653 | int j; |
56eedc1a | 654 | if (i < bestcnt) { |
52826846 NB |
655 | j = best[i]; |
656 | if (j == chosen_drive) | |
657 | continue; | |
658 | } else | |
659 | j = chosen_drive; | |
660 | ||
56eedc1a | 661 | if (j >= 0 /* && devices[j].uptodate */) { |
64c4757e NB |
662 | mdu_disk_info_t disk; |
663 | memset(&disk, 0, sizeof(disk)); | |
664 | disk.major = devices[j].major; | |
665 | disk.minor = devices[j].minor; | |
666 | if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) { | |
682c7051 | 667 | fprintf(stderr, Name ": failed to add %s to %s: %s\n", |
64c4757e NB |
668 | devices[j].devname, |
669 | mddev, | |
670 | strerror(errno)); | |
4b1ac34b | 671 | if (i < info.array.raid_disks || i == bestcnt) |
52826846 NB |
672 | okcnt--; |
673 | else | |
674 | sparecnt--; | |
dab6685f | 675 | } else if (verbose > 0) |
52826846 NB |
676 | fprintf(stderr, Name ": added %s to %s as %d\n", |
677 | devices[j].devname, mddev, devices[j].raid_disk); | |
dab6685f | 678 | } else if (verbose > 0 && i < info.array.raid_disks) |
682c7051 | 679 | fprintf(stderr, Name ": no uptodate device for slot %d of %s\n", |
64c4757e NB |
680 | i, mddev); |
681 | } | |
52826846 | 682 | |
64c4757e | 683 | if (runstop == 1 || |
b8a8ccf9 | 684 | (runstop <= 0 && |
265e0f17 | 685 | ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) && |
aa88f531 NB |
686 | (okcnt >= req_cnt || start_partial_ok) |
687 | ))) { | |
82b27616 | 688 | if (ioctl(mdfd, RUN_ARRAY, NULL)==0) { |
dab6685f NB |
689 | if (verbose >= 0) { |
690 | fprintf(stderr, Name ": %s has been started with %d drive%s", | |
691 | mddev, okcnt, okcnt==1?"":"s"); | |
b8a8ccf9 | 692 | if (okcnt < info.array.raid_disks) |
dab6685f NB |
693 | fprintf(stderr, " (out of %d)", info.array.raid_disks); |
694 | if (sparecnt) | |
695 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
696 | fprintf(stderr, ".\n"); | |
697 | } | |
64c4757e | 698 | return 0; |
82b27616 | 699 | } |
682c7051 | 700 | fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n", |
64c4757e NB |
701 | mddev, strerror(errno)); |
702 | return 1; | |
703 | } | |
82b27616 | 704 | if (runstop == -1) { |
b8a8ccf9 | 705 | fprintf(stderr, Name ": %s assembled from %d drive%s", |
52826846 | 706 | mddev, okcnt, okcnt==1?"":"s"); |
b8a8ccf9 NB |
707 | if (okcnt != info.array.raid_disks) |
708 | fprintf(stderr, " (out of %d)", info.array.raid_disks); | |
709 | fprintf(stderr, ", but not started.\n"); | |
64c4757e | 710 | return 0; |
82b27616 | 711 | } |
dab6685f NB |
712 | if (verbose >= 0) { |
713 | fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s"); | |
714 | if (sparecnt) | |
715 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
265e0f17 | 716 | if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt)) |
dab6685f NB |
717 | fprintf(stderr, " - not enough to start the array.\n"); |
718 | else { | |
719 | if (req_cnt == info.array.raid_disks) | |
720 | fprintf(stderr, " - need all %d to start it", req_cnt); | |
721 | else | |
722 | fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks); | |
723 | fprintf(stderr, " (use --run to insist).\n"); | |
724 | } | |
aa88f531 | 725 | } |
64c4757e | 726 | return 1; |
82b27616 | 727 | } else { |
52826846 NB |
728 | /* The "chosen_drive" is a good choice, and if necessary, the superblock has |
729 | * been updated to point to the current locations of devices. | |
730 | * so we can just start the array | |
82b27616 | 731 | */ |
cd29a5c8 | 732 | unsigned long dev; |
0df46c2a | 733 | dev = makedev(devices[chosen_drive].major, |
82b27616 NB |
734 | devices[chosen_drive].minor); |
735 | if (ioctl(mdfd, START_ARRAY, dev)) { | |
736 | fprintf(stderr, Name ": Cannot start array: %s\n", | |
737 | strerror(errno)); | |
738 | } | |
739 | ||
64c4757e | 740 | } |
e0d19036 | 741 | return 0; |
64c4757e | 742 | } |