]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de> | |
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 | ||
30 | #include "mdadm.h" | |
31 | #include <ctype.h> | |
32 | ||
33 | static int name_matches(char *found, char *required, char *homehost) | |
34 | { | |
35 | /* See if the name found matches the required name, possibly | |
36 | * prefixed with 'homehost' | |
37 | */ | |
38 | char fnd[33]; | |
39 | ||
40 | strncpy(fnd, found, 32); | |
41 | fnd[32] = 0; | |
42 | if (strcmp(found, required)==0) | |
43 | return 1; | |
44 | if (homehost) { | |
45 | int l = strlen(homehost); | |
46 | if (l < 32 && fnd[l] == ':' && | |
47 | strcmp(fnd+l+1, required)==0) | |
48 | return 1; | |
49 | } | |
50 | return 0; | |
51 | } | |
52 | ||
53 | int Assemble(struct supertype *st, char *mddev, int mdfd, | |
54 | mddev_ident_t ident, | |
55 | mddev_dev_t devlist, char *backup_file, | |
56 | int readonly, int runstop, | |
57 | char *update, char *homehost, | |
58 | int verbose, int force) | |
59 | { | |
60 | /* | |
61 | * The task of Assemble is to find a collection of | |
62 | * devices that should (according to their superblocks) | |
63 | * form an array, and to give this collection to the MD driver. | |
64 | * In Linux-2.4 and later, this involves submitting a | |
65 | * SET_ARRAY_INFO ioctl with no arg - to prepare | |
66 | * the array - and then submit a number of | |
67 | * ADD_NEW_DISK ioctls to add disks into | |
68 | * the array. Finally RUN_ARRAY might | |
69 | * be submitted to start the array. | |
70 | * | |
71 | * Much of the work of Assemble is in finding and/or | |
72 | * checking the disks to make sure they look right. | |
73 | * | |
74 | * If mddev is not set, then scan must be set and we | |
75 | * read through the config file for dev+uuid mapping | |
76 | * We recurse, setting mddev, for each device that | |
77 | * - isn't running | |
78 | * - has a valid uuid (or any uuid if !uuidset) | |
79 | * | |
80 | * If mddev is set, we try to determine state of md. | |
81 | * check version - must be at least 0.90.0 | |
82 | * check kernel version. must be at least 2.4. | |
83 | * If not, we can possibly fall back on START_ARRAY | |
84 | * Try to GET_ARRAY_INFO. | |
85 | * If possible, give up | |
86 | * If not, try to STOP_ARRAY just to make sure | |
87 | * | |
88 | * If !uuidset and scan, look in conf-file for uuid | |
89 | * If not found, give up | |
90 | * If !devlist and scan and uuidset, get list of devs from conf-file | |
91 | * | |
92 | * For each device: | |
93 | * Check superblock - discard if bad | |
94 | * Check uuid (set if we don't have one) - discard if no match | |
95 | * Check superblock similarity if we have a superblock - discard if different | |
96 | * Record events, devicenum | |
97 | * This should give us a list of devices for the array | |
98 | * We should collect the most recent event number | |
99 | * | |
100 | * Count disks with recent enough event count | |
101 | * While force && !enough disks | |
102 | * Choose newest rejected disks, update event count | |
103 | * mark clean and rewrite superblock | |
104 | * If recent kernel: | |
105 | * SET_ARRAY_INFO | |
106 | * foreach device with recent events : ADD_NEW_DISK | |
107 | * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY | |
108 | * If old kernel: | |
109 | * Check the device numbers in superblock are right | |
110 | * update superblock if any changes | |
111 | * START_ARRAY | |
112 | * | |
113 | */ | |
114 | int clean = 0; | |
115 | int must_close = 0; | |
116 | int old_linux = 0; | |
117 | int vers = 0; /* Keep gcc quite - it really is initialised */ | |
118 | struct { | |
119 | char *devname; | |
120 | int uptodate; /* set once we decide that this device is as | |
121 | * recent as everything else in the array. | |
122 | */ | |
123 | struct mdinfo i; | |
124 | } *devices; | |
125 | int *best = NULL; /* indexed by raid_disk */ | |
126 | unsigned int bestcnt = 0; | |
127 | int devcnt = 0; | |
128 | unsigned int okcnt, sparecnt; | |
129 | unsigned int req_cnt; | |
130 | unsigned int i; | |
131 | int most_recent = 0; | |
132 | int chosen_drive; | |
133 | int change = 0; | |
134 | int inargv = 0; | |
135 | int bitmap_done; | |
136 | int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0); | |
137 | unsigned int num_devs; | |
138 | mddev_dev_t tmpdev; | |
139 | struct mdinfo info; | |
140 | char *avail; | |
141 | int nextspare = 0; | |
142 | int uuid_for_name = 0; | |
143 | ||
144 | memset(&info, 0, sizeof(info)); | |
145 | ||
146 | if (get_linux_version() < 2004000) | |
147 | old_linux = 1; | |
148 | ||
149 | if (mdfd >= 0) { | |
150 | vers = md_get_version(mdfd); | |
151 | if (vers <= 0) { | |
152 | fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev); | |
153 | return 1; | |
154 | } | |
155 | if (vers < 9000) { | |
156 | fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n" | |
157 | " Upgrade your kernel or try --build\n"); | |
158 | return 1; | |
159 | } | |
160 | ||
161 | if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) { | |
162 | fprintf(stderr, Name ": device %s already active - cannot assemble it\n", | |
163 | mddev); | |
164 | return 1; | |
165 | } | |
166 | ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */ | |
167 | } | |
168 | /* | |
169 | * If any subdevs are listed, then any that don't | |
170 | * match ident are discarded. Remainder must all match and | |
171 | * become the array. | |
172 | * If no subdevs, then we scan all devices in the config file, but | |
173 | * there must be something in the identity | |
174 | */ | |
175 | ||
176 | if (!devlist && | |
177 | ident->uuid_set == 0 && | |
178 | ident->super_minor < 0 && | |
179 | ident->devices == NULL) { | |
180 | fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n", | |
181 | mddev ? mddev : "further assembly"); | |
182 | return 1; | |
183 | } | |
184 | if (devlist == NULL) | |
185 | devlist = conf_get_devs(); | |
186 | else if (mdfd >= 0) | |
187 | inargv = 1; | |
188 | ||
189 | try_again: | |
190 | ||
191 | tmpdev = devlist; num_devs = 0; | |
192 | while (tmpdev) { | |
193 | if (tmpdev->used) | |
194 | tmpdev->used = 2; | |
195 | else | |
196 | num_devs++; | |
197 | tmpdev = tmpdev->next; | |
198 | } | |
199 | devices = malloc(num_devs * sizeof(*devices)); | |
200 | ||
201 | if (!st && ident->st) st = ident->st; | |
202 | ||
203 | if (verbose>0) | |
204 | fprintf(stderr, Name ": looking for devices for %s\n", | |
205 | mddev ? mddev : "further assembly"); | |
206 | ||
207 | /* first walk the list of devices to find a consistent set | |
208 | * that match the criterea, if that is possible. | |
209 | * We flag the one we like with 'used'. | |
210 | */ | |
211 | for (tmpdev = devlist; | |
212 | tmpdev; | |
213 | tmpdev = tmpdev->next) { | |
214 | char *devname = tmpdev->devname; | |
215 | int dfd; | |
216 | struct stat stb; | |
217 | struct supertype *tst = dup_super(st); | |
218 | ||
219 | if (tmpdev->used > 1) continue; | |
220 | ||
221 | if (ident->devices && | |
222 | !match_oneof(ident->devices, devname)) { | |
223 | if ((inargv && verbose>=0) || verbose > 0) | |
224 | fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices); | |
225 | continue; | |
226 | } | |
227 | ||
228 | dfd = dev_open(devname, O_RDONLY|O_EXCL); | |
229 | if (dfd < 0) { | |
230 | if ((inargv && verbose >= 0) || verbose > 0) | |
231 | fprintf(stderr, Name ": cannot open device %s: %s\n", | |
232 | devname, strerror(errno)); | |
233 | tmpdev->used = 2; | |
234 | } else if (fstat(dfd, &stb)< 0) { | |
235 | /* Impossible! */ | |
236 | fprintf(stderr, Name ": fstat failed for %s: %s\n", | |
237 | devname, strerror(errno)); | |
238 | tmpdev->used = 2; | |
239 | } else if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
240 | fprintf(stderr, Name ": %s is not a block device.\n", | |
241 | devname); | |
242 | tmpdev->used = 2; | |
243 | } else if (!tst && (tst = guess_super(dfd)) == NULL) { | |
244 | if ((inargv && verbose >= 0) || verbose > 0) | |
245 | fprintf(stderr, Name ": no recogniseable superblock on %s\n", | |
246 | devname); | |
247 | tmpdev->used = 2; | |
248 | } else if (tst->ss->load_super(tst,dfd, NULL)) { | |
249 | if ((inargv && verbose >= 0) || verbose > 0) | |
250 | fprintf( stderr, Name ": no RAID superblock on %s\n", | |
251 | devname); | |
252 | } else { | |
253 | tst->ss->getinfo_super(tst, &info); | |
254 | } | |
255 | if (dfd >= 0) close(dfd); | |
256 | ||
257 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && | |
258 | (!tst || !tst->sb || | |
259 | same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) { | |
260 | if ((inargv && verbose >= 0) || verbose > 0) | |
261 | fprintf(stderr, Name ": %s has wrong uuid.\n", | |
262 | devname); | |
263 | goto loop; | |
264 | } | |
265 | if (ident->name[0] && (!update || strcmp(update, "name")!= 0) && | |
266 | (!tst || !tst->sb || | |
267 | name_matches(info.name, ident->name, homehost)==0)) { | |
268 | if ((inargv && verbose >= 0) || verbose > 0) | |
269 | fprintf(stderr, Name ": %s has wrong name.\n", | |
270 | devname); | |
271 | goto loop; | |
272 | } | |
273 | if (ident->super_minor != UnSet && | |
274 | (!tst || !tst->sb || | |
275 | ident->super_minor != info.array.md_minor)) { | |
276 | if ((inargv && verbose >= 0) || verbose > 0) | |
277 | fprintf(stderr, Name ": %s has wrong super-minor.\n", | |
278 | devname); | |
279 | goto loop; | |
280 | } | |
281 | if (ident->level != UnSet && | |
282 | (!tst || !tst->sb || | |
283 | ident->level != info.array.level)) { | |
284 | if ((inargv && verbose >= 0) || verbose > 0) | |
285 | fprintf(stderr, Name ": %s has wrong raid level.\n", | |
286 | devname); | |
287 | goto loop; | |
288 | } | |
289 | if (ident->raid_disks != UnSet && | |
290 | (!tst || !tst->sb || | |
291 | ident->raid_disks!= info.array.raid_disks)) { | |
292 | if ((inargv && verbose >= 0) || verbose > 0) | |
293 | fprintf(stderr, Name ": %s requires wrong number of drives.\n", | |
294 | devname); | |
295 | goto loop; | |
296 | } | |
297 | if (mdfd < 0) { | |
298 | if (tst == NULL || tst->sb == NULL) | |
299 | continue; | |
300 | switch(tst->ss->match_home(tst, homehost)) | |
301 | { | |
302 | case 1: /* happy with match. */ | |
303 | break; | |
304 | case -1: /* cannot match */ | |
305 | uuid_for_name = 1; | |
306 | break; | |
307 | case 0: /* Doesn't match */ | |
308 | if (update) | |
309 | /* We are changing the name*/ | |
310 | break; | |
311 | if ((inargv && verbose >= 0) || verbose > 0) | |
312 | fprintf(stderr, Name ": %s is not built for " | |
313 | "host %s - using UUID for " | |
314 | "device name.\n", | |
315 | devname, homehost); | |
316 | ||
317 | /* Auto-assemble, and this is not a usable host */ | |
318 | /* if update != NULL, we are updating the host | |
319 | * name... */ | |
320 | uuid_for_name = 1; | |
321 | break; | |
322 | } | |
323 | } | |
324 | /* If we are this far, then we are nearly commited to this device. | |
325 | * If the super_block doesn't exist, or doesn't match others, | |
326 | * then we probably cannot continue | |
327 | * However if one of the arrays is for the homehost, and | |
328 | * the other isn't that can disambiguate. | |
329 | */ | |
330 | ||
331 | if (!tst || !tst->sb) { | |
332 | fprintf(stderr, Name ": %s has no superblock - assembly aborted\n", | |
333 | devname); | |
334 | if (st) | |
335 | st->ss->free_super(st); | |
336 | return 1; | |
337 | } | |
338 | ||
339 | if (st == NULL) | |
340 | st = dup_super(tst); | |
341 | if (st->minor_version == -1) | |
342 | st->minor_version = tst->minor_version; | |
343 | if (st->ss != tst->ss || | |
344 | st->minor_version != tst->minor_version || | |
345 | st->ss->compare_super(st, tst) != 0) { | |
346 | /* Some mismatch. If exactly one array matches this host, | |
347 | * we can resolve on that one. | |
348 | * Or, if we are auto assembling, we just ignore the second | |
349 | * for now. | |
350 | */ | |
351 | if (mdfd < 0) | |
352 | goto loop; | |
353 | if (homehost) { | |
354 | int first = st->ss->match_home(st, homehost); | |
355 | int last = tst->ss->match_home(tst, homehost); | |
356 | if (first != last && | |
357 | (first == 1 || last == 1)) { | |
358 | /* We can do something */ | |
359 | if (first) {/* just ignore this one */ | |
360 | if ((inargv && verbose >= 0) || verbose > 0) | |
361 | fprintf(stderr, Name ": %s misses out due to wrong homehost\n", | |
362 | devname); | |
363 | goto loop; | |
364 | } else { /* reject all those sofar */ | |
365 | mddev_dev_t td; | |
366 | if ((inargv && verbose >= 0) || verbose > 0) | |
367 | fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n", | |
368 | devname); | |
369 | for (td=devlist; td != tmpdev; td=td->next) | |
370 | if (td->used == 1) | |
371 | td->used = 0; | |
372 | tmpdev->used = 1; | |
373 | goto loop; | |
374 | } | |
375 | } | |
376 | } | |
377 | fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n", | |
378 | devname); | |
379 | tst->ss->free_super(tst); | |
380 | st->ss->free_super(st); | |
381 | return 1; | |
382 | } | |
383 | ||
384 | tmpdev->used = 1; | |
385 | ||
386 | loop: | |
387 | if (tst) | |
388 | tst->ss->free_super(tst); | |
389 | } | |
390 | ||
391 | if (mdfd < 0) { | |
392 | /* So... it is up to me to open the device. | |
393 | * We create a name '/dev/md/XXX' based on the info in the | |
394 | * superblock, and call open_mddev on that | |
395 | */ | |
396 | mdu_array_info_t inf; | |
397 | char *c; | |
398 | char nbuf[64]; | |
399 | int rc; | |
400 | ||
401 | if (!st || !st->sb) { | |
402 | return 2; | |
403 | } | |
404 | st->ss->getinfo_super(st, &info); | |
405 | if (uuid_for_name) | |
406 | c = fname_from_uuid(st, &info, nbuf, '-'); | |
407 | else { | |
408 | c = strchr(info.name, ':'); | |
409 | if (c) c++; else c= info.name; | |
410 | } | |
411 | if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6)) | |
412 | /* /dev/md/d0 style for partitionable */ | |
413 | rc = asprintf(&mddev, "/dev/md/d%s", c); | |
414 | else | |
415 | rc = asprintf(&mddev, "/dev/md/%s", c); | |
416 | if (rc < 0) | |
417 | mdfd = -1; | |
418 | else | |
419 | mdfd = open_mddev(mddev, ident->autof); | |
420 | if (mdfd < 0) { | |
421 | st->ss->free_super(st); | |
422 | free(devices); | |
423 | goto try_again; | |
424 | } | |
425 | vers = md_get_version(mdfd); | |
426 | if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) { | |
427 | for (tmpdev = devlist ; | |
428 | tmpdev && tmpdev->used != 1; | |
429 | tmpdev = tmpdev->next) | |
430 | ; | |
431 | fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev); | |
432 | if (tmpdev) | |
433 | fprintf(stderr, Name ": %s needed for %s...\n", | |
434 | mddev, tmpdev->devname); | |
435 | close(mdfd); | |
436 | mdfd = -1; | |
437 | st->ss->free_super(st); | |
438 | free(devices); | |
439 | goto try_again; | |
440 | } | |
441 | must_close = 1; | |
442 | } | |
443 | ||
444 | /* Ok, no bad inconsistancy, we can try updating etc */ | |
445 | bitmap_done = 0; | |
446 | for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) { | |
447 | char *devname = tmpdev->devname; | |
448 | struct stat stb; | |
449 | /* looks like a good enough match to update the super block if needed */ | |
450 | #ifndef MDASSEMBLE | |
451 | if (update) { | |
452 | int dfd; | |
453 | /* prepare useful information in info structures */ | |
454 | struct stat stb2; | |
455 | struct supertype *tst; | |
456 | fstat(mdfd, &stb2); | |
457 | ||
458 | if (strcmp(update, "uuid")==0 && | |
459 | !ident->uuid_set) { | |
460 | int rfd; | |
461 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
462 | read(rfd, ident->uuid, 16) != 16) { | |
463 | *(__u32*)(ident->uuid) = random(); | |
464 | *(__u32*)(ident->uuid+1) = random(); | |
465 | *(__u32*)(ident->uuid+2) = random(); | |
466 | *(__u32*)(ident->uuid+3) = random(); | |
467 | } | |
468 | if (rfd >= 0) close(rfd); | |
469 | } | |
470 | dfd = dev_open(devname, O_RDWR|O_EXCL); | |
471 | ||
472 | remove_partitions(dfd); | |
473 | ||
474 | tst = dup_super(st); | |
475 | tst->ss->load_super(tst, dfd, NULL); | |
476 | tst->ss->getinfo_super(tst, &info); | |
477 | ||
478 | memcpy(info.uuid, ident->uuid, 16); | |
479 | strcpy(info.name, ident->name); | |
480 | info.array.md_minor = minor(stb2.st_rdev); | |
481 | ||
482 | tst->ss->update_super(tst, &info, update, | |
483 | devname, verbose, | |
484 | ident->uuid_set, homehost); | |
485 | if (strcmp(update, "uuid")==0 && | |
486 | !ident->uuid_set) { | |
487 | ident->uuid_set = 1; | |
488 | memcpy(ident->uuid, info.uuid, 16); | |
489 | } | |
490 | if (dfd < 0) | |
491 | fprintf(stderr, Name ": Cannot open %s for superblock update\n", | |
492 | devname); | |
493 | else if (tst->ss->store_super(tst, dfd)) | |
494 | fprintf(stderr, Name ": Could not re-write superblock on %s.\n", | |
495 | devname); | |
496 | if (dfd >= 0) | |
497 | close(dfd); | |
498 | ||
499 | if (strcmp(update, "uuid")==0 && | |
500 | ident->bitmap_fd >= 0 && !bitmap_done) { | |
501 | if (bitmap_update_uuid(ident->bitmap_fd, | |
502 | info.uuid, | |
503 | tst->ss->swapuuid) != 0) | |
504 | fprintf(stderr, Name ": Could not update uuid on external bitmap.\n"); | |
505 | else | |
506 | bitmap_done = 1; | |
507 | } | |
508 | tst->ss->free_super(tst); | |
509 | } else | |
510 | #endif | |
511 | { | |
512 | struct supertype *tst = dup_super(st); | |
513 | int dfd; | |
514 | dfd = dev_open(devname, O_RDWR|O_EXCL); | |
515 | ||
516 | remove_partitions(dfd); | |
517 | ||
518 | tst->ss->load_super(tst, dfd, NULL); | |
519 | tst->ss->getinfo_super(tst, &info); | |
520 | tst->ss->free_super(tst); | |
521 | close(dfd); | |
522 | } | |
523 | ||
524 | stat(devname, &stb); | |
525 | ||
526 | if (verbose > 0) | |
527 | fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n", | |
528 | devname, mddev, info.disk.raid_disk); | |
529 | devices[devcnt].devname = devname; | |
530 | devices[devcnt].uptodate = 0; | |
531 | devices[devcnt].i = info; | |
532 | devices[devcnt].i.disk.major = major(stb.st_rdev); | |
533 | devices[devcnt].i.disk.minor = minor(stb.st_rdev); | |
534 | if (most_recent < devcnt) { | |
535 | if (devices[devcnt].i.events | |
536 | > devices[most_recent].i.events) | |
537 | most_recent = devcnt; | |
538 | } | |
539 | if (info.array.level == -4) | |
540 | /* with multipath, the raid_disk from the superblock is meaningless */ | |
541 | i = devcnt; | |
542 | else | |
543 | i = devices[devcnt].i.disk.raid_disk; | |
544 | if (i+1 == 0) { | |
545 | if (nextspare < info.array.raid_disks) | |
546 | nextspare = info.array.raid_disks; | |
547 | i = nextspare++; | |
548 | } else { | |
549 | if (i >= info.array.raid_disks && | |
550 | i >= nextspare) | |
551 | nextspare = i+1; | |
552 | } | |
553 | if (i < 10000) { | |
554 | if (i >= bestcnt) { | |
555 | unsigned int newbestcnt = i+10; | |
556 | int *newbest = malloc(sizeof(int)*newbestcnt); | |
557 | unsigned int c; | |
558 | for (c=0; c < newbestcnt; c++) | |
559 | if (c < bestcnt) | |
560 | newbest[c] = best[c]; | |
561 | else | |
562 | newbest[c] = -1; | |
563 | if (best)free(best); | |
564 | best = newbest; | |
565 | bestcnt = newbestcnt; | |
566 | } | |
567 | if (best[i] >=0 && | |
568 | devices[best[i]].i.events | |
569 | == devices[devcnt].i.events | |
570 | && (devices[best[i]].i.disk.minor | |
571 | != devices[devcnt].i.disk.minor) | |
572 | && st->ss == &super0 | |
573 | && info.array.level != LEVEL_MULTIPATH) { | |
574 | /* two different devices with identical superblock. | |
575 | * Could be a mis-detection caused by overlapping | |
576 | * partitions. fail-safe. | |
577 | */ | |
578 | fprintf(stderr, Name ": WARNING %s and %s appear" | |
579 | " to have very similar superblocks.\n" | |
580 | " If they are really different, " | |
581 | "please --zero the superblock on one\n" | |
582 | " If they are the same or overlap," | |
583 | " please remove one from %s.\n", | |
584 | devices[best[i]].devname, devname, | |
585 | inargv ? "the list" : | |
586 | "the\n DEVICE list in mdadm.conf" | |
587 | ); | |
588 | if (must_close) close(mdfd); | |
589 | return 1; | |
590 | } | |
591 | if (best[i] == -1 | |
592 | || (devices[best[i]].i.events | |
593 | < devices[devcnt].i.events)) | |
594 | best[i] = devcnt; | |
595 | } | |
596 | devcnt++; | |
597 | } | |
598 | ||
599 | if (devcnt == 0) { | |
600 | fprintf(stderr, Name ": no devices found for %s\n", | |
601 | mddev); | |
602 | if (st) | |
603 | st->ss->free_super(st); | |
604 | if (must_close) close(mdfd); | |
605 | return 1; | |
606 | } | |
607 | ||
608 | if (update && strcmp(update, "byteorder")==0) | |
609 | st->minor_version = 90; | |
610 | ||
611 | st->ss->getinfo_super(st, &info); | |
612 | clean = info.array.state & 1; | |
613 | ||
614 | /* now we have some devices that might be suitable. | |
615 | * I wonder how many | |
616 | */ | |
617 | avail = malloc(info.array.raid_disks); | |
618 | memset(avail, 0, info.array.raid_disks); | |
619 | okcnt = 0; | |
620 | sparecnt=0; | |
621 | for (i=0; i< bestcnt ;i++) { | |
622 | int j = best[i]; | |
623 | int event_margin = 1; /* always allow a difference of '1' | |
624 | * like the kernel does | |
625 | */ | |
626 | if (j < 0) continue; | |
627 | /* note: we ignore error flags in multipath arrays | |
628 | * as they don't make sense | |
629 | */ | |
630 | if (info.array.level != -4) | |
631 | if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) { | |
632 | if (!(devices[j].i.disk.state | |
633 | & (1<<MD_DISK_FAULTY))) | |
634 | sparecnt++; | |
635 | continue; | |
636 | } | |
637 | if (devices[j].i.events+event_margin >= | |
638 | devices[most_recent].i.events) { | |
639 | devices[j].uptodate = 1; | |
640 | if (i < info.array.raid_disks) { | |
641 | okcnt++; | |
642 | avail[i]=1; | |
643 | } else | |
644 | sparecnt++; | |
645 | } | |
646 | } | |
647 | while (force && !enough(info.array.level, info.array.raid_disks, | |
648 | info.array.layout, 1, | |
649 | avail, okcnt)) { | |
650 | /* Choose the newest best drive which is | |
651 | * not up-to-date, update the superblock | |
652 | * and add it. | |
653 | */ | |
654 | int fd; | |
655 | struct supertype *tst; | |
656 | long long current_events; | |
657 | chosen_drive = -1; | |
658 | for (i=0; i<info.array.raid_disks && i < bestcnt; i++) { | |
659 | int j = best[i]; | |
660 | if (j>=0 && | |
661 | !devices[j].uptodate && | |
662 | devices[j].i.events > 0 && | |
663 | (chosen_drive < 0 || | |
664 | devices[j].i.events | |
665 | > devices[chosen_drive].i.events)) | |
666 | chosen_drive = j; | |
667 | } | |
668 | if (chosen_drive < 0) | |
669 | break; | |
670 | current_events = devices[chosen_drive].i.events; | |
671 | add_another: | |
672 | if (verbose >= 0) | |
673 | fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n", | |
674 | devices[chosen_drive].devname, | |
675 | devices[chosen_drive].i.disk.raid_disk, | |
676 | (int)(devices[chosen_drive].i.events), | |
677 | (int)(devices[most_recent].i.events)); | |
678 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); | |
679 | if (fd < 0) { | |
680 | fprintf(stderr, Name ": Couldn't open %s for write - not updating\n", | |
681 | devices[chosen_drive].devname); | |
682 | devices[chosen_drive].i.events = 0; | |
683 | continue; | |
684 | } | |
685 | tst = dup_super(st); | |
686 | if (tst->ss->load_super(tst,fd, NULL)) { | |
687 | close(fd); | |
688 | fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n", | |
689 | devices[chosen_drive].devname); | |
690 | devices[chosen_drive].i.events = 0; | |
691 | continue; | |
692 | } | |
693 | info.events = devices[most_recent].i.events; | |
694 | tst->ss->update_super(tst, &info, "force-one", | |
695 | devices[chosen_drive].devname, verbose, | |
696 | 0, NULL); | |
697 | ||
698 | if (tst->ss->store_super(tst, fd)) { | |
699 | close(fd); | |
700 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
701 | devices[chosen_drive].devname); | |
702 | devices[chosen_drive].i.events = 0; | |
703 | tst->ss->free_super(tst); | |
704 | continue; | |
705 | } | |
706 | close(fd); | |
707 | devices[chosen_drive].i.events = devices[most_recent].i.events; | |
708 | devices[chosen_drive].uptodate = 1; | |
709 | avail[chosen_drive] = 1; | |
710 | okcnt++; | |
711 | tst->ss->free_super(tst); | |
712 | ||
713 | /* If there are any other drives of the same vintage, | |
714 | * add them in as well. We can't lose and we might gain | |
715 | */ | |
716 | for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) { | |
717 | int j = best[i]; | |
718 | if (j >= 0 && | |
719 | !devices[j].uptodate && | |
720 | devices[j].i.events > 0 && | |
721 | devices[j].i.events == current_events) { | |
722 | chosen_drive = j; | |
723 | goto add_another; | |
724 | } | |
725 | } | |
726 | } | |
727 | ||
728 | /* Now we want to look at the superblock which the kernel will base things on | |
729 | * and compare the devices that we think are working with the devices that the | |
730 | * superblock thinks are working. | |
731 | * If there are differences and --force is given, then update this chosen | |
732 | * superblock. | |
733 | */ | |
734 | chosen_drive = -1; | |
735 | st->ss->free_super(st); | |
736 | for (i=0; chosen_drive < 0 && i<bestcnt; i++) { | |
737 | int j = best[i]; | |
738 | int fd; | |
739 | ||
740 | if (j<0) | |
741 | continue; | |
742 | if (!devices[j].uptodate) | |
743 | continue; | |
744 | chosen_drive = j; | |
745 | if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) { | |
746 | fprintf(stderr, Name ": Cannot open %s: %s\n", | |
747 | devices[j].devname, strerror(errno)); | |
748 | if (must_close) close(mdfd); | |
749 | return 1; | |
750 | } | |
751 | if (st->ss->load_super(st,fd, NULL)) { | |
752 | close(fd); | |
753 | fprintf(stderr, Name ": RAID superblock has disappeared from %s\n", | |
754 | devices[j].devname); | |
755 | if (must_close) close(mdfd); | |
756 | return 1; | |
757 | } | |
758 | close(fd); | |
759 | } | |
760 | if (st->sb == NULL) { | |
761 | fprintf(stderr, Name ": No suitable drives found for %s\n", mddev); | |
762 | if (must_close) close(mdfd); | |
763 | return 1; | |
764 | } | |
765 | st->ss->getinfo_super(st, &info); | |
766 | #ifndef MDASSEMBLE | |
767 | sysfs_init(&info, mdfd, 0); | |
768 | #endif | |
769 | for (i=0; i<bestcnt; i++) { | |
770 | int j = best[i]; | |
771 | unsigned int desired_state; | |
772 | ||
773 | if (i < info.array.raid_disks) | |
774 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); | |
775 | else | |
776 | desired_state = 0; | |
777 | ||
778 | if (j<0) | |
779 | continue; | |
780 | if (!devices[j].uptodate) | |
781 | continue; | |
782 | ||
783 | devices[j].i.disk.state = desired_state; | |
784 | ||
785 | if (st->ss->update_super(st, &devices[j].i, "assemble", NULL, | |
786 | verbose, 0, NULL)) { | |
787 | if (force) { | |
788 | if (verbose >= 0) | |
789 | fprintf(stderr, Name ": " | |
790 | "clearing FAULTY flag for device %d in %s for %s\n", | |
791 | j, mddev, devices[j].devname); | |
792 | change = 1; | |
793 | } else { | |
794 | if (verbose >= -1) | |
795 | fprintf(stderr, Name ": " | |
796 | "device %d in %s has wrong state in superblock, but %s seems ok\n", | |
797 | i, mddev, devices[j].devname); | |
798 | } | |
799 | } | |
800 | #if 0 | |
801 | if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) { | |
802 | fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", | |
803 | i, mddev); | |
804 | } | |
805 | #endif | |
806 | } | |
807 | if (force && !clean && | |
808 | !enough(info.array.level, info.array.raid_disks, | |
809 | info.array.layout, clean, | |
810 | avail, okcnt)) { | |
811 | change += st->ss->update_super(st, &info, "force-array", | |
812 | devices[chosen_drive].devname, verbose, | |
813 | 0, NULL); | |
814 | clean = 1; | |
815 | } | |
816 | ||
817 | if (change) { | |
818 | int fd; | |
819 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); | |
820 | if (fd < 0) { | |
821 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", | |
822 | devices[chosen_drive].devname); | |
823 | if (must_close) close(mdfd); | |
824 | return 1; | |
825 | } | |
826 | if (st->ss->store_super(st, fd)) { | |
827 | close(fd); | |
828 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
829 | devices[chosen_drive].devname); | |
830 | if (must_close) close(mdfd); | |
831 | return 1; | |
832 | } | |
833 | close(fd); | |
834 | } | |
835 | ||
836 | /* If we are in the middle of a reshape we may need to restore saved data | |
837 | * that was moved aside due to the reshape overwriting live data | |
838 | * The code of doing this lives in Grow.c | |
839 | */ | |
840 | #ifndef MDASSEMBLE | |
841 | if (info.reshape_active) { | |
842 | int err = 0; | |
843 | int *fdlist = malloc(sizeof(int)* bestcnt); | |
844 | for (i=0; i<bestcnt; i++) { | |
845 | int j = best[i]; | |
846 | if (j >= 0) { | |
847 | fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
848 | if (fdlist[i] < 0) { | |
849 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", | |
850 | devices[j].devname); | |
851 | err = 1; | |
852 | break; | |
853 | } | |
854 | } else | |
855 | fdlist[i] = -1; | |
856 | } | |
857 | if (!err) | |
858 | err = Grow_restart(st, &info, fdlist, bestcnt, backup_file); | |
859 | while (i>0) { | |
860 | i--; | |
861 | if (fdlist[i]>=0) close(fdlist[i]); | |
862 | } | |
863 | if (err) { | |
864 | fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n"); | |
865 | if (must_close) close(mdfd); | |
866 | return err; | |
867 | } | |
868 | } | |
869 | #endif | |
870 | /* count number of in-sync devices according to the superblock. | |
871 | * We must have this number to start the array without -s or -R | |
872 | */ | |
873 | req_cnt = info.array.working_disks; | |
874 | ||
875 | /* Almost ready to actually *do* something */ | |
876 | if (!old_linux) { | |
877 | int rv; | |
878 | ||
879 | rv = set_array_info(mdfd, st, &info); | |
880 | if (rv) { | |
881 | fprintf(stderr, Name ": failed to set array info for %s: %s\n", | |
882 | mddev, strerror(errno)); | |
883 | if (must_close) close(mdfd); | |
884 | return 1; | |
885 | } | |
886 | if (ident->bitmap_fd >= 0) { | |
887 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { | |
888 | fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n"); | |
889 | if (must_close) close(mdfd); | |
890 | return 1; | |
891 | } | |
892 | } else if (ident->bitmap_file) { | |
893 | /* From config file */ | |
894 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
895 | if (bmfd < 0) { | |
896 | fprintf(stderr, Name ": Could not open bitmap file %s\n", | |
897 | ident->bitmap_file); | |
898 | if (must_close) close(mdfd); | |
899 | return 1; | |
900 | } | |
901 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
902 | fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev); | |
903 | close(bmfd); | |
904 | if (must_close) close(mdfd); | |
905 | return 1; | |
906 | } | |
907 | close(bmfd); | |
908 | } | |
909 | ||
910 | /* First, add the raid disks, but add the chosen one last */ | |
911 | for (i=0; i<= bestcnt; i++) { | |
912 | int j; | |
913 | if (i < bestcnt) { | |
914 | j = best[i]; | |
915 | if (j == chosen_drive) | |
916 | continue; | |
917 | } else | |
918 | j = chosen_drive; | |
919 | ||
920 | if (j >= 0 /* && devices[j].uptodate */) { | |
921 | rv = add_disk(mdfd, st, &info, &devices[j].i); | |
922 | ||
923 | if (rv) { | |
924 | fprintf(stderr, Name ": failed to add " | |
925 | "%s to %s: %s\n", | |
926 | devices[j].devname, | |
927 | mddev, | |
928 | strerror(errno)); | |
929 | if (i < info.array.raid_disks | |
930 | || i == bestcnt) | |
931 | okcnt--; | |
932 | else | |
933 | sparecnt--; | |
934 | } else if (verbose > 0) | |
935 | fprintf(stderr, Name ": added %s " | |
936 | "to %s as %d\n", | |
937 | devices[j].devname, mddev, | |
938 | devices[j].i.disk.raid_disk); | |
939 | } else if (verbose > 0 && i < info.array.raid_disks) | |
940 | fprintf(stderr, Name ": no uptodate device for " | |
941 | "slot %d of %s\n", | |
942 | i, mddev); | |
943 | } | |
944 | ||
945 | if (info.array.level == LEVEL_CONTAINER) { | |
946 | if (verbose >= 0) { | |
947 | fprintf(stderr, Name ": Container %s has been " | |
948 | "assembled with %d drive%s", | |
949 | mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s"); | |
950 | if (okcnt < info.array.raid_disks) | |
951 | fprintf(stderr, " (out of %d)", | |
952 | info.array.raid_disks); | |
953 | fprintf(stderr, "\n"); | |
954 | } | |
955 | if (must_close) | |
956 | close(mdfd); | |
957 | return 0; | |
958 | } | |
959 | ||
960 | if (runstop == 1 || | |
961 | (runstop <= 0 && | |
962 | ( enough(info.array.level, info.array.raid_disks, | |
963 | info.array.layout, clean, avail, okcnt) && | |
964 | (okcnt >= req_cnt || start_partial_ok) | |
965 | ))) { | |
966 | if (ioctl(mdfd, RUN_ARRAY, NULL)==0) { | |
967 | if (verbose >= 0) { | |
968 | fprintf(stderr, Name ": %s has been started with %d drive%s", | |
969 | mddev, okcnt, okcnt==1?"":"s"); | |
970 | if (okcnt < info.array.raid_disks) | |
971 | fprintf(stderr, " (out of %d)", info.array.raid_disks); | |
972 | if (sparecnt) | |
973 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
974 | fprintf(stderr, ".\n"); | |
975 | } | |
976 | if (must_close) { | |
977 | int usecs = 1; | |
978 | close(mdfd); | |
979 | /* There is a nasty race with 'mdadm --monitor'. | |
980 | * If it opens this device before we close it, | |
981 | * it gets an incomplete open on which IO | |
982 | * doesn't work and the capacity is | |
983 | * wrong. | |
984 | * If we reopen (to check for layered devices) | |
985 | * before --monitor closes, we loose. | |
986 | * | |
987 | * So: wait upto 1 second for there to be | |
988 | * a non-zero capacity. | |
989 | */ | |
990 | while (usecs < 1000) { | |
991 | mdfd = open(mddev, O_RDONLY); | |
992 | if (mdfd >= 0) { | |
993 | unsigned long long size; | |
994 | if (get_dev_size(mdfd, NULL, &size) && | |
995 | size > 0) | |
996 | break; | |
997 | close(mdfd); | |
998 | } | |
999 | usleep(usecs); | |
1000 | usecs <<= 1; | |
1001 | } | |
1002 | } | |
1003 | return 0; | |
1004 | } | |
1005 | fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n", | |
1006 | mddev, strerror(errno)); | |
1007 | ||
1008 | if (!enough(info.array.level, info.array.raid_disks, | |
1009 | info.array.layout, 1, avail, okcnt)) | |
1010 | fprintf(stderr, Name ": Not enough devices to " | |
1011 | "start the array.\n"); | |
1012 | else if (!enough(info.array.level, | |
1013 | info.array.raid_disks, | |
1014 | info.array.layout, clean, | |
1015 | avail, okcnt)) | |
1016 | fprintf(stderr, Name ": Not enough devices to " | |
1017 | "start the array while not clean " | |
1018 | "- consider --force.\n"); | |
1019 | ||
1020 | if (must_close) { | |
1021 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1022 | close(mdfd); | |
1023 | } | |
1024 | return 1; | |
1025 | } | |
1026 | if (runstop == -1) { | |
1027 | fprintf(stderr, Name ": %s assembled from %d drive%s", | |
1028 | mddev, okcnt, okcnt==1?"":"s"); | |
1029 | if (okcnt != info.array.raid_disks) | |
1030 | fprintf(stderr, " (out of %d)", info.array.raid_disks); | |
1031 | fprintf(stderr, ", but not started.\n"); | |
1032 | if (must_close) close(mdfd); | |
1033 | return 0; | |
1034 | } | |
1035 | if (verbose >= -1) { | |
1036 | fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s"); | |
1037 | if (sparecnt) | |
1038 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
1039 | if (!enough(info.array.level, info.array.raid_disks, | |
1040 | info.array.layout, 1, avail, okcnt)) | |
1041 | fprintf(stderr, " - not enough to start the array.\n"); | |
1042 | else if (!enough(info.array.level, | |
1043 | info.array.raid_disks, | |
1044 | info.array.layout, clean, | |
1045 | avail, okcnt)) | |
1046 | fprintf(stderr, " - not enough to start the " | |
1047 | "array while not clean - consider " | |
1048 | "--force.\n"); | |
1049 | else { | |
1050 | if (req_cnt == info.array.raid_disks) | |
1051 | fprintf(stderr, " - need all %d to start it", req_cnt); | |
1052 | else | |
1053 | fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks); | |
1054 | fprintf(stderr, " (use --run to insist).\n"); | |
1055 | } | |
1056 | } | |
1057 | if (must_close) { | |
1058 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1059 | close(mdfd); | |
1060 | } | |
1061 | return 1; | |
1062 | } else { | |
1063 | /* The "chosen_drive" is a good choice, and if necessary, the superblock has | |
1064 | * been updated to point to the current locations of devices. | |
1065 | * so we can just start the array | |
1066 | */ | |
1067 | unsigned long dev; | |
1068 | dev = makedev(devices[chosen_drive].i.disk.major, | |
1069 | devices[chosen_drive].i.disk.minor); | |
1070 | if (ioctl(mdfd, START_ARRAY, dev)) { | |
1071 | fprintf(stderr, Name ": Cannot start array: %s\n", | |
1072 | strerror(errno)); | |
1073 | } | |
1074 | ||
1075 | } | |
1076 | if (must_close) close(mdfd); | |
1077 | return 0; | |
1078 | } |