]>
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 | |
624920bb NB |
32 | static int name_matches(char *found, char *required, char *homehost) |
33 | { | |
34 | /* See if the name found matches the required name, possibly | |
35 | * prefixed with 'homehost' | |
36 | */ | |
37 | char fnd[33]; | |
38 | ||
39 | strncpy(fnd, found, 32); | |
40 | fnd[32] = 0; | |
41 | if (strcmp(found, required)==0) | |
42 | return 1; | |
43 | if (homehost) { | |
44 | int l = strlen(homehost); | |
45 | if (l < 32 && fnd[l] == ':' && | |
46 | strcmp(fnd+l+1, required)==0) | |
47 | return 1; | |
48 | } | |
49 | return 0; | |
50 | } | |
51 | ||
82d9eba6 | 52 | int Assemble(struct supertype *st, char *mddev, int mdfd, |
52826846 | 53 | mddev_ident_t ident, char *conffile, |
06b0d786 | 54 | mddev_dev_t devlist, char *backup_file, |
64c4757e | 55 | int readonly, int runstop, |
e5eac01f | 56 | char *update, char *homehost, |
64c4757e NB |
57 | int verbose, int force) |
58 | { | |
59 | /* | |
52826846 NB |
60 | * The task of Assemble is to find a collection of |
61 | * devices that should (according to their superblocks) | |
62 | * form an array, and to give this collection to the MD driver. | |
63 | * In Linux-2.4 and later, this involves submitting a | |
64c4757e NB |
64 | * SET_ARRAY_INFO ioctl with no arg - to prepare |
65 | * the array - and then submit a number of | |
66 | * ADD_NEW_DISK ioctls to add disks into | |
67 | * the array. Finally RUN_ARRAY might | |
68 | * be submitted to start the array. | |
69 | * | |
70 | * Much of the work of Assemble is in finding and/or | |
71 | * checking the disks to make sure they look right. | |
72 | * | |
4b1ac34b | 73 | * If mddev is not set, then scan must be set and we |
64c4757e NB |
74 | * read through the config file for dev+uuid mapping |
75 | * We recurse, setting mddev, for each device that | |
76 | * - isn't running | |
4b1ac34b | 77 | * - has a valid uuid (or any uuid if !uuidset) |
64c4757e NB |
78 | * |
79 | * If mddev is set, we try to determine state of md. | |
80 | * check version - must be at least 0.90.0 | |
81 | * check kernel version. must be at least 2.4. | |
82 | * If not, we can possibly fall back on START_ARRAY | |
83 | * Try to GET_ARRAY_INFO. | |
84 | * If possible, give up | |
85 | * If not, try to STOP_ARRAY just to make sure | |
86 | * | |
87 | * If !uuidset and scan, look in conf-file for uuid | |
88 | * If not found, give up | |
b8a8ccf9 | 89 | * If !devlist and scan and uuidset, get list of devs from conf-file |
64c4757e NB |
90 | * |
91 | * For each device: | |
92 | * Check superblock - discard if bad | |
93 | * Check uuid (set if we don't have one) - discard if no match | |
5787fa49 | 94 | * Check superblock similarity if we have a superblock - discard if different |
4b1ac34b | 95 | * Record events, devicenum |
64c4757e | 96 | * This should give us a list of devices for the array |
4b1ac34b | 97 | * We should collect the most recent event number |
64c4757e NB |
98 | * |
99 | * Count disks with recent enough event count | |
100 | * While force && !enough disks | |
101 | * Choose newest rejected disks, update event count | |
102 | * mark clean and rewrite superblock | |
103 | * If recent kernel: | |
104 | * SET_ARRAY_INFO | |
105 | * foreach device with recent events : ADD_NEW_DISK | |
106 | * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY | |
107 | * If old kernel: | |
108 | * Check the device numbers in superblock are right | |
109 | * update superblock if any changes | |
110 | * START_ARRAY | |
111 | * | |
112 | */ | |
113 | int old_linux = 0; | |
114 | int vers; | |
4b1ac34b | 115 | void *first_super = NULL, *super = NULL; |
64c4757e NB |
116 | struct { |
117 | char *devname; | |
98c6faba NB |
118 | unsigned int major, minor; |
119 | unsigned int oldmajor, oldminor; | |
64c4757e | 120 | long long events; |
64c4757e | 121 | int uptodate; |
d013a55e | 122 | int state; |
52826846 | 123 | int raid_disk; |
6a41304b | 124 | int disk_nr; |
5787fa49 | 125 | } *devices; |
56eedc1a | 126 | int *best = NULL; /* indexed by raid_disk */ |
98c6faba NB |
127 | unsigned int bestcnt = 0; |
128 | int devcnt = 0; | |
129 | unsigned int okcnt, sparecnt; | |
130 | unsigned int req_cnt; | |
131 | unsigned int i; | |
64c4757e | 132 | int most_recent = 0; |
cd29a5c8 | 133 | int chosen_drive; |
52826846 | 134 | int change = 0; |
cd29a5c8 | 135 | int inargv = 0; |
b8a8ccf9 | 136 | int start_partial_ok = (runstop >= 0) && (force || devlist==NULL); |
98c6faba | 137 | unsigned int num_devs; |
5787fa49 | 138 | mddev_dev_t tmpdev; |
4b1ac34b | 139 | struct mdinfo info; |
265e0f17 | 140 | char *avail; |
308e1801 | 141 | int nextspare = 0; |
8a46fe84 | 142 | |
682c7051 | 143 | if (get_linux_version() < 2004000) |
64c4757e NB |
144 | old_linux = 1; |
145 | ||
8a46fe84 NB |
146 | if (mdfd >= 0) { |
147 | vers = md_get_version(mdfd); | |
148 | if (vers <= 0) { | |
149 | fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev); | |
150 | return 1; | |
151 | } | |
152 | if (vers < 9000) { | |
153 | fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n" | |
154 | " Upgrade your kernel or try --build\n"); | |
155 | return 1; | |
156 | } | |
64c4757e | 157 | |
8a46fe84 NB |
158 | if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) { |
159 | fprintf(stderr, Name ": device %s already active - cannot assemble it\n", | |
160 | mddev); | |
161 | return 1; | |
162 | } | |
163 | ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */ | |
164 | } | |
64c4757e | 165 | /* |
52826846 NB |
166 | * If any subdevs are listed, then any that don't |
167 | * match ident are discarded. Remainder must all match and | |
168 | * become the array. | |
169 | * If no subdevs, then we scan all devices in the config file, but | |
170 | * there must be something in the identity | |
64c4757e | 171 | */ |
64c4757e | 172 | |
cd29a5c8 | 173 | if (!devlist && |
52826846 NB |
174 | ident->uuid_set == 0 && |
175 | ident->super_minor < 0 && | |
176 | ident->devices == NULL) { | |
177 | fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n", | |
8a46fe84 | 178 | mddev ? mddev : "further assembly"); |
52826846 | 179 | return 1; |
64c4757e | 180 | } |
cd29a5c8 | 181 | if (devlist == NULL) |
64c4757e | 182 | devlist = conf_get_devs(conffile); |
cd29a5c8 | 183 | else inargv = 1; |
64c4757e | 184 | |
5787fa49 NB |
185 | tmpdev = devlist; num_devs = 0; |
186 | while (tmpdev) { | |
187 | num_devs++; | |
188 | tmpdev = tmpdev->next; | |
189 | } | |
5787fa49 NB |
190 | devices = malloc(num_devs * sizeof(*devices)); |
191 | ||
82d9eba6 | 192 | if (!st && ident->st) st = ident->st; |
64c4757e | 193 | |
dab6685f | 194 | if (verbose>0) |
82b27616 | 195 | fprintf(stderr, Name ": looking for devices for %s\n", |
8a46fe84 | 196 | mddev ? mddev : "further assembly"); |
82b27616 | 197 | |
811e6cbe NB |
198 | /* first walk the list of devices to find a consistent set |
199 | * that match the criterea, if that is possible. | |
200 | * We flag the one we like with 'used'. | |
201 | */ | |
202 | for (tmpdev = devlist; | |
203 | tmpdev; | |
204 | tmpdev = tmpdev->next) { | |
205 | char *devname = tmpdev->devname; | |
64c4757e NB |
206 | int dfd; |
207 | struct stat stb; | |
82d9eba6 | 208 | struct supertype *tst = st; |
52826846 | 209 | |
52826846 | 210 | if (ident->devices && |
56eedc1a | 211 | !match_oneof(ident->devices, devname)) { |
6a41304b | 212 | if ((inargv && verbose>=0) || verbose > 0) |
56eedc1a | 213 | fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices); |
52826846 | 214 | continue; |
56eedc1a | 215 | } |
4b1ac34b NB |
216 | |
217 | if (super) { | |
218 | free(super); | |
219 | super = NULL; | |
220 | } | |
52826846 | 221 | |
8b0dabea | 222 | dfd = dev_open(devname, O_RDONLY|O_EXCL); |
64c4757e | 223 | if (dfd < 0) { |
6a41304b | 224 | if ((inargv && verbose >= 0) || verbose > 0) |
682c7051 | 225 | fprintf(stderr, Name ": cannot open device %s: %s\n", |
64c4757e | 226 | devname, strerror(errno)); |
52826846 NB |
227 | } else if (fstat(dfd, &stb)< 0) { |
228 | /* Impossible! */ | |
229 | fprintf(stderr, Name ": fstat failed for %s: %s\n", | |
230 | devname, strerror(errno)); | |
cd29a5c8 NB |
231 | } else if ((stb.st_mode & S_IFMT) != S_IFBLK) { |
232 | fprintf(stderr, Name ": %s is not a block device.\n", | |
52826846 | 233 | devname); |
82d9eba6 | 234 | } else if (!tst && (tst = guess_super(dfd)) == NULL) { |
6a41304b | 235 | if ((inargv && verbose >= 0) || verbose > 0) |
f9ce90ba | 236 | fprintf(stderr, Name ": no recogniseable superblock\n"); |
82d9eba6 | 237 | } else if (tst->ss->load_super(tst,dfd, &super, NULL)) { |
6a41304b | 238 | if ((inargv && verbose >= 0) || verbose > 0) |
682c7051 | 239 | fprintf( stderr, Name ": no RAID superblock on %s\n", |
64c4757e | 240 | devname); |
52826846 | 241 | } else { |
31317663 | 242 | tst->ss->getinfo_super(&info, super); |
64c4757e | 243 | } |
c06487ce | 244 | if (dfd >= 0) close(dfd); |
52826846 | 245 | |
838acbc2 | 246 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && |
f277ce36 | 247 | (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) { |
6a41304b | 248 | if ((inargv && verbose >= 0) || verbose > 0) |
52826846 NB |
249 | fprintf(stderr, Name ": %s has wrong uuid.\n", |
250 | devname); | |
251 | continue; | |
82b27616 | 252 | } |
c4f12c13 | 253 | if (ident->name[0] && (!update || strcmp(update, "name")!= 0) && |
624920bb | 254 | (!super || name_matches(info.name, ident->name, homehost)==0)) { |
6a41304b | 255 | if ((inargv && verbose >= 0) || verbose > 0) |
947fd4dd NB |
256 | fprintf(stderr, Name ": %s has wrong name.\n", |
257 | devname); | |
258 | continue; | |
259 | } | |
98c6faba | 260 | if (ident->super_minor != UnSet && |
4b1ac34b | 261 | (!super || ident->super_minor != info.array.md_minor)) { |
6a41304b | 262 | if ((inargv && verbose >= 0) || verbose > 0) |
52826846 | 263 | fprintf(stderr, Name ": %s has wrong super-minor.\n", |
64c4757e NB |
264 | devname); |
265 | continue; | |
266 | } | |
98c6faba | 267 | if (ident->level != UnSet && |
4b1ac34b | 268 | (!super|| ident->level != info.array.level)) { |
6a41304b | 269 | if ((inargv && verbose >= 0) || verbose > 0) |
cd29a5c8 NB |
270 | fprintf(stderr, Name ": %s has wrong raid level.\n", |
271 | devname); | |
272 | continue; | |
273 | } | |
98c6faba | 274 | if (ident->raid_disks != UnSet && |
4b1ac34b | 275 | (!super || ident->raid_disks!= info.array.raid_disks)) { |
6a41304b | 276 | if ((inargv && verbose >= 0) || verbose > 0) |
cd29a5c8 NB |
277 | fprintf(stderr, Name ": %s requires wrong number of drives.\n", |
278 | devname); | |
279 | continue; | |
280 | } | |
52826846 | 281 | |
83b6208e | 282 | /* If we are this far, then we are nearly commited to this device. |
52826846 | 283 | * If the super_block doesn't exist, or doesn't match others, |
83b6208e NB |
284 | * then we probably cannot continue |
285 | * However if one of the arrays is for the homehost, and | |
286 | * the other isn't that can disambiguate. | |
52826846 | 287 | */ |
52826846 | 288 | |
4b1ac34b | 289 | if (!super) { |
52826846 NB |
290 | fprintf(stderr, Name ": %s has no superblock - assembly aborted\n", |
291 | devname); | |
4b1ac34b | 292 | free(first_super); |
52826846 NB |
293 | return 1; |
294 | } | |
838acbc2 | 295 | |
83b6208e NB |
296 | if (st == NULL) |
297 | st = tst; | |
298 | if (st->ss != tst->ss || | |
299 | st->minor_version != tst->minor_version || | |
300 | st->ss->compare_super(&first_super, super) != 0) { | |
301 | /* Some mismatch. If exactly one array matches this host, | |
302 | * we can resolve on that one | |
303 | */ | |
304 | if (homehost) { | |
305 | int first = st->ss->match_home(first_super, homehost); | |
306 | int last = tst->ss->match_home(super, homehost); | |
307 | if (first+last == 1) { | |
308 | /* We can do something */ | |
309 | if (first) {/* just ignore this one */ | |
310 | if ((inargv && verbose >= 0) || verbose > 0) | |
311 | fprintf(stderr, Name ": %s misses out due to wrong homehost\n", | |
312 | devname); | |
313 | continue; | |
314 | } else { /* reject all those sofar */ | |
315 | mddev_dev_t td; | |
316 | if ((inargv && verbose >= 0) || verbose > 0) | |
317 | fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n", | |
318 | devname); | |
319 | for (td=devlist; td != tmpdev; td=td->next) | |
320 | if (td->used == 1) | |
321 | td->used = 0; | |
322 | tmpdev->used = 1; | |
323 | continue; | |
324 | } | |
325 | } | |
326 | } | |
52826846 NB |
327 | fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n", |
328 | devname); | |
4b1ac34b NB |
329 | free(super); |
330 | free(first_super); | |
52826846 | 331 | return 1; |
64c4757e NB |
332 | } |
333 | ||
811e6cbe NB |
334 | tmpdev->used = 1; |
335 | } | |
336 | ||
8a46fe84 NB |
337 | if (mdfd < 0) { |
338 | /* So... it is up to me to open the device. | |
339 | * We create a name '/dev/md/XXX' based on the info in the | |
340 | * superblock, and call open_mddev on that | |
341 | */ | |
342 | asprintf(&mddev, "/dev/md/%s", info.name); | |
343 | mdfd = open_mddev(mddev, 0); | |
344 | if (mdfd < 0) | |
345 | return mdfd; | |
346 | } | |
347 | ||
811e6cbe NB |
348 | /* Ok, no bad inconsistancy, we can try updating etc */ |
349 | for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used) { | |
350 | char *devname = tmpdev->devname; | |
351 | struct stat stb; | |
5787fa49 NB |
352 | /* looks like a good enough match to update the super block if needed */ |
353 | if (update) { | |
811e6cbe | 354 | int dfd; |
4b1ac34b NB |
355 | /* prepare useful information in info structures */ |
356 | struct stat stb2; | |
357 | fstat(mdfd, &stb2); | |
7d99579f NB |
358 | |
359 | if (strcmp(update, "uuid")==0 && | |
360 | !ident->uuid_set) { | |
361 | int rfd; | |
362 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
363 | read(rfd, ident->uuid, 16) != 16) { | |
364 | *(__u32*)(ident->uuid) = random(); | |
365 | *(__u32*)(ident->uuid+1) = random(); | |
366 | *(__u32*)(ident->uuid+2) = random(); | |
367 | *(__u32*)(ident->uuid+3) = random(); | |
368 | } | |
369 | if (rfd >= 0) close(rfd); | |
7d99579f | 370 | } |
811e6cbe NB |
371 | dfd = dev_open(devname, O_RDWR|O_EXCL); |
372 | ||
373 | if (super) { | |
374 | free(super); | |
375 | super = NULL; | |
376 | } | |
377 | ||
378 | st->ss->load_super(st, dfd, &super, NULL); | |
379 | st->ss->getinfo_super(&info, super); | |
380 | ||
7d99579f | 381 | memcpy(info.uuid, ident->uuid, 16); |
e5eac01f | 382 | strcpy(info.name, ident->name); |
811e6cbe NB |
383 | info.array.md_minor = minor(stb2.st_rdev); |
384 | ||
e5eac01f NB |
385 | st->ss->update_super(&info, super, update, devname, verbose, |
386 | ident->uuid_set, homehost); | |
387 | if (strcmp(update, "uuid")==0 && | |
388 | !ident->uuid_set) { | |
389 | ident->uuid_set = 1; | |
390 | memcpy(ident->uuid, info.uuid, 16); | |
391 | } | |
b8a8ccf9 | 392 | if (dfd < 0) |
5787fa49 NB |
393 | fprintf(stderr, Name ": Cannot open %s for superblock update\n", |
394 | devname); | |
96395475 | 395 | else if (st->ss->store_super(st, dfd, super)) |
5787fa49 NB |
396 | fprintf(stderr, Name ": Could not re-write superblock on %s.\n", |
397 | devname); | |
398 | if (dfd >= 0) | |
399 | close(dfd); | |
8131b493 NB |
400 | |
401 | if (strcmp(update, "uuid")==0 && | |
402 | ident->bitmap_fd) | |
403 | bitmap_update_uuid(ident->bitmap_fd, info.uuid); | |
5787fa49 NB |
404 | } |
405 | ||
811e6cbe NB |
406 | stat(devname, &stb); |
407 | ||
dab6685f | 408 | if (verbose > 0) |
cd29a5c8 | 409 | fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n", |
4b1ac34b | 410 | devname, mddev, info.disk.raid_disk); |
64c4757e | 411 | devices[devcnt].devname = devname; |
0df46c2a NB |
412 | devices[devcnt].major = major(stb.st_rdev); |
413 | devices[devcnt].minor = minor(stb.st_rdev); | |
4b1ac34b NB |
414 | devices[devcnt].oldmajor = info.disk.major; |
415 | devices[devcnt].oldminor = info.disk.minor; | |
416 | devices[devcnt].events = info.events; | |
417 | devices[devcnt].raid_disk = info.disk.raid_disk; | |
6a41304b | 418 | devices[devcnt].disk_nr = info.disk.number; |
64c4757e | 419 | devices[devcnt].uptodate = 0; |
4b1ac34b | 420 | devices[devcnt].state = info.disk.state; |
64c4757e NB |
421 | if (most_recent < devcnt) { |
422 | if (devices[devcnt].events | |
423 | > devices[most_recent].events) | |
424 | most_recent = devcnt; | |
425 | } | |
b8a8ccf9 | 426 | if (info.array.level == -4) |
5787fa49 NB |
427 | /* with multipath, the raid_disk from the superblock is meaningless */ |
428 | i = devcnt; | |
429 | else | |
430 | i = devices[devcnt].raid_disk; | |
308e1801 NB |
431 | if (i+1 == 0) { |
432 | if (nextspare < info.array.raid_disks) | |
433 | nextspare = info.array.raid_disks; | |
434 | i = nextspare++; | |
435 | } | |
98c6faba | 436 | if (i < 10000) { |
56eedc1a | 437 | if (i >= bestcnt) { |
98c6faba | 438 | unsigned int newbestcnt = i+10; |
56eedc1a | 439 | int *newbest = malloc(sizeof(int)*newbestcnt); |
98c6faba | 440 | unsigned int c; |
56eedc1a NB |
441 | for (c=0; c < newbestcnt; c++) |
442 | if (c < bestcnt) | |
443 | newbest[c] = best[c]; | |
444 | else | |
445 | newbest[c] = -1; | |
446 | if (best)free(best); | |
447 | best = newbest; | |
448 | bestcnt = newbestcnt; | |
449 | } | |
52826846 NB |
450 | if (best[i] == -1 |
451 | || devices[best[i]].events < devices[devcnt].events) | |
452 | best[i] = devcnt; | |
56eedc1a | 453 | } |
64c4757e NB |
454 | devcnt++; |
455 | } | |
456 | ||
4b1ac34b NB |
457 | if (super) |
458 | free(super); | |
459 | super = NULL; | |
460 | ||
586ed405 NB |
461 | if (update && strcmp(update, "byteorder")==0) |
462 | st->minor_version = 90; | |
463 | ||
64c4757e | 464 | if (devcnt == 0) { |
682c7051 | 465 | fprintf(stderr, Name ": no devices found for %s\n", |
64c4757e | 466 | mddev); |
4b1ac34b | 467 | free(first_super); |
64c4757e NB |
468 | return 1; |
469 | } | |
4b1ac34b | 470 | |
31317663 | 471 | st->ss->getinfo_super(&info, first_super); |
4b1ac34b | 472 | |
64c4757e NB |
473 | /* now we have some devices that might be suitable. |
474 | * I wonder how many | |
475 | */ | |
265e0f17 NB |
476 | avail = malloc(info.array.raid_disks); |
477 | memset(avail, 0, info.array.raid_disks); | |
64c4757e | 478 | okcnt = 0; |
52826846 | 479 | sparecnt=0; |
56eedc1a | 480 | for (i=0; i< bestcnt ;i++) { |
64c4757e | 481 | int j = best[i]; |
ee04451c NB |
482 | int event_margin = 1; /* always allow a difference of '1' |
483 | * like the kernel does | |
484 | */ | |
64c4757e | 485 | if (j < 0) continue; |
d013a55e NB |
486 | /* note: we ignore error flags in multipath arrays |
487 | * as they don't make sense | |
488 | */ | |
4b1ac34b | 489 | if (info.array.level != -4) |
aa88f531 NB |
490 | if (!(devices[j].state & (1<<MD_DISK_SYNC))) { |
491 | if (!(devices[j].state & (1<<MD_DISK_FAULTY))) | |
492 | sparecnt++; | |
d013a55e | 493 | continue; |
aa88f531 | 494 | } |
cd29a5c8 | 495 | if (devices[j].events+event_margin >= |
64c4757e NB |
496 | devices[most_recent].events) { |
497 | devices[j].uptodate = 1; | |
265e0f17 | 498 | if (i < info.array.raid_disks) { |
52826846 | 499 | okcnt++; |
265e0f17 NB |
500 | avail[i]=1; |
501 | } else | |
52826846 | 502 | sparecnt++; |
64c4757e NB |
503 | } |
504 | } | |
265e0f17 NB |
505 | while (force && !enough(info.array.level, info.array.raid_disks, |
506 | info.array.layout, | |
507 | avail, okcnt)) { | |
64c4757e NB |
508 | /* Choose the newest best drive which is |
509 | * not up-to-date, update the superblock | |
510 | * and add it. | |
511 | */ | |
52826846 | 512 | int fd; |
cd29a5c8 | 513 | chosen_drive = -1; |
4b1ac34b | 514 | for (i=0; i<info.array.raid_disks && i < bestcnt; i++) { |
52826846 NB |
515 | int j = best[i]; |
516 | if (j>=0 && | |
517 | !devices[j].uptodate && | |
518 | devices[j].events > 0 && | |
519 | (chosen_drive < 0 || | |
520 | devices[j].events > devices[chosen_drive].events)) | |
521 | chosen_drive = j; | |
522 | } | |
523 | if (chosen_drive < 0) | |
524 | break; | |
dab6685f NB |
525 | if (verbose >= 0) |
526 | fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n", | |
527 | devices[chosen_drive].devname, devices[chosen_drive].raid_disk, | |
528 | (int)(devices[chosen_drive].events), | |
529 | (int)(devices[most_recent].events)); | |
8b0dabea | 530 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 NB |
531 | if (fd < 0) { |
532 | fprintf(stderr, Name ": Couldn't open %s for write - not updating\n", | |
533 | devices[chosen_drive].devname); | |
534 | devices[chosen_drive].events = 0; | |
535 | continue; | |
536 | } | |
82d9eba6 | 537 | if (st->ss->load_super(st,fd, &super, NULL)) { |
52826846 NB |
538 | close(fd); |
539 | fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n", | |
540 | devices[chosen_drive].devname); | |
541 | devices[chosen_drive].events = 0; | |
542 | continue; | |
543 | } | |
4b1ac34b | 544 | info.events = devices[most_recent].events; |
e5eac01f | 545 | st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL); |
4b1ac34b | 546 | |
96395475 | 547 | if (st->ss->store_super(st, fd, super)) { |
52826846 NB |
548 | close(fd); |
549 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
550 | devices[chosen_drive].devname); | |
551 | devices[chosen_drive].events = 0; | |
4b1ac34b | 552 | free(super); |
52826846 NB |
553 | continue; |
554 | } | |
555 | close(fd); | |
556 | devices[chosen_drive].events = devices[most_recent].events; | |
557 | devices[chosen_drive].uptodate = 1; | |
265e0f17 | 558 | avail[chosen_drive] = 1; |
52826846 | 559 | okcnt++; |
4b1ac34b | 560 | free(super); |
64c4757e | 561 | } |
52826846 NB |
562 | |
563 | /* Now we want to look at the superblock which the kernel will base things on | |
564 | * and compare the devices that we think are working with the devices that the | |
565 | * superblock thinks are working. | |
566 | * If there are differences and --force is given, then update this chosen | |
567 | * superblock. | |
568 | */ | |
cd29a5c8 | 569 | chosen_drive = -1; |
4b1ac34b | 570 | super = NULL; |
56eedc1a | 571 | for (i=0; chosen_drive < 0 && i<bestcnt; i++) { |
52826846 NB |
572 | int j = best[i]; |
573 | int fd; | |
4b1ac34b | 574 | |
52826846 NB |
575 | if (j<0) |
576 | continue; | |
577 | if (!devices[j].uptodate) | |
578 | continue; | |
579 | chosen_drive = j; | |
8b0dabea | 580 | if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) { |
52826846 NB |
581 | fprintf(stderr, Name ": Cannot open %s: %s\n", |
582 | devices[j].devname, strerror(errno)); | |
583 | return 1; | |
584 | } | |
82d9eba6 | 585 | if (st->ss->load_super(st,fd, &super, NULL)) { |
52826846 NB |
586 | close(fd); |
587 | fprintf(stderr, Name ": RAID superblock has disappeared from %s\n", | |
588 | devices[j].devname); | |
589 | return 1; | |
590 | } | |
591 | close(fd); | |
592 | } | |
4b1ac34b NB |
593 | if (super == NULL) { |
594 | fprintf(stderr, Name ": No suitable drives found for %s\n", mddev); | |
595 | return 1; | |
596 | } | |
31317663 | 597 | st->ss->getinfo_super(&info, super); |
56eedc1a | 598 | for (i=0; i<bestcnt; i++) { |
52826846 | 599 | int j = best[i]; |
98c6faba | 600 | unsigned int desired_state; |
11a3e71d | 601 | |
4b1ac34b | 602 | if (i < info.array.raid_disks) |
11a3e71d NB |
603 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); |
604 | else | |
605 | desired_state = 0; | |
606 | ||
52826846 NB |
607 | if (j<0) |
608 | continue; | |
609 | if (!devices[j].uptodate) | |
610 | continue; | |
6a41304b | 611 | info.disk.number = devices[j].disk_nr; |
fbf8a0b7 | 612 | info.disk.raid_disk = i; |
4b1ac34b NB |
613 | info.disk.state = desired_state; |
614 | ||
52826846 | 615 | if (devices[j].uptodate && |
e5eac01f | 616 | st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) { |
52826846 | 617 | if (force) { |
dab6685f NB |
618 | if (verbose >= 0) |
619 | fprintf(stderr, Name ": " | |
620 | "clearing FAULTY flag for device %d in %s for %s\n", | |
621 | j, mddev, devices[j].devname); | |
4b1ac34b | 622 | change = 1; |
52826846 | 623 | } else { |
dab6685f NB |
624 | if (verbose >= -1) |
625 | fprintf(stderr, Name ": " | |
626 | "device %d in %s has wrong state in superblock, but %s seems ok\n", | |
627 | i, mddev, devices[j].devname); | |
52826846 NB |
628 | } |
629 | } | |
4b1ac34b | 630 | #if 0 |
52826846 NB |
631 | if (!devices[j].uptodate && |
632 | !(super.disks[i].state & (1 << MD_DISK_FAULTY))) { | |
633 | fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", | |
634 | i, mddev); | |
635 | } | |
4b1ac34b | 636 | #endif |
52826846 | 637 | } |
4b1ac34b NB |
638 | if (force && okcnt == info.array.raid_disks-1) { |
639 | /* FIXME check event count */ | |
b8a8ccf9 | 640 | change += st->ss->update_super(&info, super, "force", |
e5eac01f | 641 | devices[chosen_drive].devname, verbose, 0, NULL); |
d013a55e | 642 | } |
52826846 | 643 | |
4b1ac34b | 644 | if (change) { |
52826846 | 645 | int fd; |
8b0dabea | 646 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 | 647 | if (fd < 0) { |
353632d9 | 648 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", |
52826846 NB |
649 | devices[chosen_drive].devname); |
650 | return 1; | |
651 | } | |
96395475 | 652 | if (st->ss->store_super(st, fd, super)) { |
52826846 NB |
653 | close(fd); |
654 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
655 | devices[chosen_drive].devname); | |
656 | return 1; | |
657 | } | |
658 | close(fd); | |
52826846 NB |
659 | } |
660 | ||
353632d9 NB |
661 | /* If we are in the middle of a reshape we may need to restore saved data |
662 | * that was moved aside due to the reshape overwriting live data | |
663 | * The code of doing this lives in Grow.c | |
664 | */ | |
39c5a909 | 665 | #ifndef MDASSEMBLE |
353632d9 NB |
666 | if (info.reshape_active) { |
667 | int err = 0; | |
668 | int *fdlist = malloc(sizeof(int)* bestcnt); | |
669 | for (i=0; i<bestcnt; i++) { | |
670 | int j = best[i]; | |
671 | if (j >= 0) { | |
672 | fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
673 | if (fdlist[i] < 0) { | |
674 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", | |
675 | devices[j].devname); | |
676 | err = 1; | |
677 | break; | |
678 | } | |
679 | } else | |
680 | fdlist[i] = -1; | |
681 | } | |
682 | if (!err) | |
06b0d786 | 683 | err = Grow_restart(st, &info, fdlist, bestcnt, backup_file); |
353632d9 NB |
684 | while (i>0) { |
685 | i--; | |
686 | if (fdlist[i]>=0) close(fdlist[i]); | |
687 | } | |
688 | if (err) { | |
689 | fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n"); | |
690 | return err; | |
691 | } | |
692 | } | |
39c5a909 | 693 | #endif |
aa88f531 NB |
694 | /* count number of in-sync devices according to the superblock. |
695 | * We must have this number to start the array without -s or -R | |
696 | */ | |
4b1ac34b | 697 | req_cnt = info.array.working_disks; |
aa88f531 | 698 | |
64c4757e NB |
699 | /* Almost ready to actually *do* something */ |
700 | if (!old_linux) { | |
fbf8a0b7 NB |
701 | int rv; |
702 | if ((vers % 100) >= 1) { /* can use different versions */ | |
703 | mdu_array_info_t inf; | |
704 | memset(&inf, 0, sizeof(inf)); | |
705 | inf.major_version = st->ss->major; | |
706 | inf.minor_version = st->minor_version; | |
707 | rv = ioctl(mdfd, SET_ARRAY_INFO, &inf); | |
b8a8ccf9 | 708 | } else |
fbf8a0b7 NB |
709 | rv = ioctl(mdfd, SET_ARRAY_INFO, NULL); |
710 | ||
711 | if (rv) { | |
682c7051 | 712 | fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n", |
64c4757e NB |
713 | mddev, strerror(errno)); |
714 | return 1; | |
715 | } | |
11484a63 | 716 | if (ident->bitmap_fd >= 0) { |
c82f047c NB |
717 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { |
718 | fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n"); | |
719 | return 1; | |
720 | } | |
7ef02d01 NB |
721 | } else if (ident->bitmap_file) { |
722 | /* From config file */ | |
723 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
724 | if (bmfd < 0) { | |
725 | fprintf(stderr, Name ": Could not open bitmap file %s\n", | |
726 | ident->bitmap_file); | |
727 | return 1; | |
728 | } | |
729 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
730 | fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev); | |
731 | close(bmfd); | |
732 | return 1; | |
733 | } | |
734 | close(bmfd); | |
c82f047c | 735 | } |
7ef02d01 | 736 | |
52826846 | 737 | /* First, add the raid disks, but add the chosen one last */ |
56eedc1a | 738 | for (i=0; i<= bestcnt; i++) { |
52826846 | 739 | int j; |
56eedc1a | 740 | if (i < bestcnt) { |
52826846 NB |
741 | j = best[i]; |
742 | if (j == chosen_drive) | |
743 | continue; | |
744 | } else | |
745 | j = chosen_drive; | |
746 | ||
56eedc1a | 747 | if (j >= 0 /* && devices[j].uptodate */) { |
64c4757e NB |
748 | mdu_disk_info_t disk; |
749 | memset(&disk, 0, sizeof(disk)); | |
750 | disk.major = devices[j].major; | |
751 | disk.minor = devices[j].minor; | |
752 | if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) { | |
682c7051 | 753 | fprintf(stderr, Name ": failed to add %s to %s: %s\n", |
64c4757e NB |
754 | devices[j].devname, |
755 | mddev, | |
756 | strerror(errno)); | |
4b1ac34b | 757 | if (i < info.array.raid_disks || i == bestcnt) |
52826846 NB |
758 | okcnt--; |
759 | else | |
760 | sparecnt--; | |
dab6685f | 761 | } else if (verbose > 0) |
52826846 NB |
762 | fprintf(stderr, Name ": added %s to %s as %d\n", |
763 | devices[j].devname, mddev, devices[j].raid_disk); | |
dab6685f | 764 | } else if (verbose > 0 && i < info.array.raid_disks) |
682c7051 | 765 | fprintf(stderr, Name ": no uptodate device for slot %d of %s\n", |
64c4757e NB |
766 | i, mddev); |
767 | } | |
52826846 | 768 | |
64c4757e | 769 | if (runstop == 1 || |
b8a8ccf9 | 770 | (runstop <= 0 && |
265e0f17 | 771 | ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) && |
aa88f531 NB |
772 | (okcnt >= req_cnt || start_partial_ok) |
773 | ))) { | |
82b27616 | 774 | if (ioctl(mdfd, RUN_ARRAY, NULL)==0) { |
dab6685f NB |
775 | if (verbose >= 0) { |
776 | fprintf(stderr, Name ": %s has been started with %d drive%s", | |
777 | mddev, okcnt, okcnt==1?"":"s"); | |
b8a8ccf9 | 778 | if (okcnt < info.array.raid_disks) |
dab6685f NB |
779 | fprintf(stderr, " (out of %d)", info.array.raid_disks); |
780 | if (sparecnt) | |
781 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
782 | fprintf(stderr, ".\n"); | |
783 | } | |
64c4757e | 784 | return 0; |
82b27616 | 785 | } |
682c7051 | 786 | fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n", |
64c4757e NB |
787 | mddev, strerror(errno)); |
788 | return 1; | |
789 | } | |
82b27616 | 790 | if (runstop == -1) { |
b8a8ccf9 | 791 | fprintf(stderr, Name ": %s assembled from %d drive%s", |
52826846 | 792 | mddev, okcnt, okcnt==1?"":"s"); |
b8a8ccf9 NB |
793 | if (okcnt != info.array.raid_disks) |
794 | fprintf(stderr, " (out of %d)", info.array.raid_disks); | |
795 | fprintf(stderr, ", but not started.\n"); | |
64c4757e | 796 | return 0; |
82b27616 | 797 | } |
dab6685f NB |
798 | if (verbose >= 0) { |
799 | fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s"); | |
800 | if (sparecnt) | |
801 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
265e0f17 | 802 | if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt)) |
dab6685f NB |
803 | fprintf(stderr, " - not enough to start the array.\n"); |
804 | else { | |
805 | if (req_cnt == info.array.raid_disks) | |
806 | fprintf(stderr, " - need all %d to start it", req_cnt); | |
807 | else | |
808 | fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks); | |
809 | fprintf(stderr, " (use --run to insist).\n"); | |
810 | } | |
aa88f531 | 811 | } |
64c4757e | 812 | return 1; |
82b27616 | 813 | } else { |
52826846 NB |
814 | /* The "chosen_drive" is a good choice, and if necessary, the superblock has |
815 | * been updated to point to the current locations of devices. | |
816 | * so we can just start the array | |
82b27616 | 817 | */ |
cd29a5c8 | 818 | unsigned long dev; |
0df46c2a | 819 | dev = makedev(devices[chosen_drive].major, |
82b27616 NB |
820 | devices[chosen_drive].minor); |
821 | if (ioctl(mdfd, START_ARRAY, dev)) { | |
822 | fprintf(stderr, Name ": Cannot start array: %s\n", | |
823 | strerror(errno)); | |
824 | } | |
825 | ||
64c4757e | 826 | } |
e0d19036 | 827 | return 0; |
64c4757e | 828 | } |