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