]>
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 | */ | |
d55e3aef | 113 | int must_close = 0; |
64c4757e NB |
114 | int old_linux = 0; |
115 | int vers; | |
4b1ac34b | 116 | void *first_super = NULL, *super = NULL; |
64c4757e NB |
117 | struct { |
118 | char *devname; | |
98c6faba NB |
119 | unsigned int major, minor; |
120 | unsigned int oldmajor, oldminor; | |
64c4757e | 121 | long long events; |
64c4757e | 122 | int uptodate; |
d013a55e | 123 | int state; |
52826846 | 124 | int raid_disk; |
6a41304b | 125 | int disk_nr; |
5787fa49 | 126 | } *devices; |
56eedc1a | 127 | int *best = NULL; /* indexed by raid_disk */ |
98c6faba NB |
128 | unsigned int bestcnt = 0; |
129 | int devcnt = 0; | |
130 | unsigned int okcnt, sparecnt; | |
131 | unsigned int req_cnt; | |
132 | unsigned int i; | |
64c4757e | 133 | int most_recent = 0; |
cd29a5c8 | 134 | int chosen_drive; |
52826846 | 135 | int change = 0; |
cd29a5c8 | 136 | int inargv = 0; |
b8a8ccf9 | 137 | int start_partial_ok = (runstop >= 0) && (force || devlist==NULL); |
98c6faba | 138 | unsigned int num_devs; |
5787fa49 | 139 | mddev_dev_t tmpdev; |
4b1ac34b | 140 | struct mdinfo info; |
265e0f17 | 141 | char *avail; |
308e1801 | 142 | int nextspare = 0; |
8a46fe84 | 143 | |
682c7051 | 144 | if (get_linux_version() < 2004000) |
64c4757e NB |
145 | old_linux = 1; |
146 | ||
8a46fe84 NB |
147 | if (mdfd >= 0) { |
148 | vers = md_get_version(mdfd); | |
149 | if (vers <= 0) { | |
150 | fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev); | |
151 | return 1; | |
152 | } | |
153 | if (vers < 9000) { | |
154 | fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n" | |
155 | " Upgrade your kernel or try --build\n"); | |
156 | return 1; | |
157 | } | |
64c4757e | 158 | |
8a46fe84 NB |
159 | if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) { |
160 | fprintf(stderr, Name ": device %s already active - cannot assemble it\n", | |
161 | mddev); | |
162 | return 1; | |
163 | } | |
164 | ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */ | |
165 | } | |
64c4757e | 166 | /* |
52826846 NB |
167 | * If any subdevs are listed, then any that don't |
168 | * match ident are discarded. Remainder must all match and | |
169 | * become the array. | |
170 | * If no subdevs, then we scan all devices in the config file, but | |
171 | * there must be something in the identity | |
64c4757e | 172 | */ |
64c4757e | 173 | |
cd29a5c8 | 174 | if (!devlist && |
52826846 NB |
175 | ident->uuid_set == 0 && |
176 | ident->super_minor < 0 && | |
177 | ident->devices == NULL) { | |
178 | fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n", | |
8a46fe84 | 179 | mddev ? mddev : "further assembly"); |
52826846 | 180 | return 1; |
64c4757e | 181 | } |
cd29a5c8 | 182 | if (devlist == NULL) |
64c4757e | 183 | devlist = conf_get_devs(conffile); |
da6b5ca9 NB |
184 | else if (mdfd >= 0) |
185 | inargv = 1; | |
64c4757e | 186 | |
5787fa49 NB |
187 | tmpdev = devlist; num_devs = 0; |
188 | while (tmpdev) { | |
da6b5ca9 NB |
189 | if (tmpdev->used) |
190 | tmpdev->used = 2; | |
191 | else | |
192 | num_devs++; | |
5787fa49 NB |
193 | tmpdev = tmpdev->next; |
194 | } | |
5787fa49 NB |
195 | devices = malloc(num_devs * sizeof(*devices)); |
196 | ||
82d9eba6 | 197 | if (!st && ident->st) st = ident->st; |
64c4757e | 198 | |
dab6685f | 199 | if (verbose>0) |
82b27616 | 200 | fprintf(stderr, Name ": looking for devices for %s\n", |
8a46fe84 | 201 | mddev ? mddev : "further assembly"); |
82b27616 | 202 | |
811e6cbe NB |
203 | /* first walk the list of devices to find a consistent set |
204 | * that match the criterea, if that is possible. | |
205 | * We flag the one we like with 'used'. | |
206 | */ | |
207 | for (tmpdev = devlist; | |
208 | tmpdev; | |
209 | tmpdev = tmpdev->next) { | |
210 | char *devname = tmpdev->devname; | |
64c4757e NB |
211 | int dfd; |
212 | struct stat stb; | |
82d9eba6 | 213 | struct supertype *tst = st; |
52826846 | 214 | |
da6b5ca9 NB |
215 | if (tmpdev->used > 1) continue; |
216 | ||
52826846 | 217 | if (ident->devices && |
56eedc1a | 218 | !match_oneof(ident->devices, devname)) { |
6a41304b | 219 | if ((inargv && verbose>=0) || verbose > 0) |
56eedc1a | 220 | fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices); |
52826846 | 221 | continue; |
56eedc1a | 222 | } |
4b1ac34b NB |
223 | |
224 | if (super) { | |
225 | free(super); | |
226 | super = NULL; | |
227 | } | |
52826846 | 228 | |
8b0dabea | 229 | dfd = dev_open(devname, O_RDONLY|O_EXCL); |
64c4757e | 230 | if (dfd < 0) { |
6a41304b | 231 | if ((inargv && verbose >= 0) || verbose > 0) |
682c7051 | 232 | fprintf(stderr, Name ": cannot open device %s: %s\n", |
64c4757e | 233 | devname, strerror(errno)); |
da6b5ca9 | 234 | tmpdev->used = 2; |
52826846 NB |
235 | } else if (fstat(dfd, &stb)< 0) { |
236 | /* Impossible! */ | |
237 | fprintf(stderr, Name ": fstat failed for %s: %s\n", | |
238 | devname, strerror(errno)); | |
da6b5ca9 | 239 | tmpdev->used = 2; |
cd29a5c8 NB |
240 | } else if ((stb.st_mode & S_IFMT) != S_IFBLK) { |
241 | fprintf(stderr, Name ": %s is not a block device.\n", | |
52826846 | 242 | devname); |
da6b5ca9 | 243 | tmpdev->used = 2; |
82d9eba6 | 244 | } else if (!tst && (tst = guess_super(dfd)) == NULL) { |
6a41304b | 245 | if ((inargv && verbose >= 0) || verbose > 0) |
da6b5ca9 NB |
246 | fprintf(stderr, Name ": no recogniseable superblock on %s\n", |
247 | devname); | |
248 | tmpdev->used = 2; | |
82d9eba6 | 249 | } else if (tst->ss->load_super(tst,dfd, &super, NULL)) { |
6a41304b | 250 | if ((inargv && verbose >= 0) || verbose > 0) |
682c7051 | 251 | fprintf( stderr, Name ": no RAID superblock on %s\n", |
64c4757e | 252 | devname); |
52826846 | 253 | } else { |
31317663 | 254 | tst->ss->getinfo_super(&info, super); |
64c4757e | 255 | } |
c06487ce | 256 | if (dfd >= 0) close(dfd); |
52826846 | 257 | |
838acbc2 | 258 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && |
f277ce36 | 259 | (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) { |
6a41304b | 260 | if ((inargv && verbose >= 0) || verbose > 0) |
52826846 NB |
261 | fprintf(stderr, Name ": %s has wrong uuid.\n", |
262 | devname); | |
263 | continue; | |
82b27616 | 264 | } |
c4f12c13 | 265 | if (ident->name[0] && (!update || strcmp(update, "name")!= 0) && |
624920bb | 266 | (!super || name_matches(info.name, ident->name, homehost)==0)) { |
6a41304b | 267 | if ((inargv && verbose >= 0) || verbose > 0) |
947fd4dd NB |
268 | fprintf(stderr, Name ": %s has wrong name.\n", |
269 | devname); | |
270 | continue; | |
271 | } | |
98c6faba | 272 | if (ident->super_minor != UnSet && |
4b1ac34b | 273 | (!super || ident->super_minor != info.array.md_minor)) { |
6a41304b | 274 | if ((inargv && verbose >= 0) || verbose > 0) |
52826846 | 275 | fprintf(stderr, Name ": %s has wrong super-minor.\n", |
64c4757e NB |
276 | devname); |
277 | continue; | |
278 | } | |
98c6faba | 279 | if (ident->level != UnSet && |
4b1ac34b | 280 | (!super|| ident->level != info.array.level)) { |
6a41304b | 281 | if ((inargv && verbose >= 0) || verbose > 0) |
cd29a5c8 NB |
282 | fprintf(stderr, Name ": %s has wrong raid level.\n", |
283 | devname); | |
284 | continue; | |
285 | } | |
98c6faba | 286 | if (ident->raid_disks != UnSet && |
4b1ac34b | 287 | (!super || ident->raid_disks!= info.array.raid_disks)) { |
6a41304b | 288 | if ((inargv && verbose >= 0) || verbose > 0) |
cd29a5c8 NB |
289 | fprintf(stderr, Name ": %s requires wrong number of drives.\n", |
290 | devname); | |
291 | continue; | |
292 | } | |
da6b5ca9 NB |
293 | if (mdfd < 0) { |
294 | if (tst == NULL || super == NULL) | |
295 | continue; | |
296 | if (tst->ss->match_home(super, homehost)==0) { | |
297 | if ((inargv && verbose >= 0) || verbose > 0) | |
298 | fprintf(stderr, Name ": %s is not built for host %s.\n", | |
299 | devname, homehost); | |
300 | /* Auto-assemble, and this is not a usable host */ | |
301 | continue; | |
302 | } | |
303 | } | |
83b6208e | 304 | /* If we are this far, then we are nearly commited to this device. |
52826846 | 305 | * If the super_block doesn't exist, or doesn't match others, |
83b6208e NB |
306 | * then we probably cannot continue |
307 | * However if one of the arrays is for the homehost, and | |
308 | * the other isn't that can disambiguate. | |
52826846 | 309 | */ |
52826846 | 310 | |
4b1ac34b | 311 | if (!super) { |
52826846 NB |
312 | fprintf(stderr, Name ": %s has no superblock - assembly aborted\n", |
313 | devname); | |
4b1ac34b | 314 | free(first_super); |
52826846 NB |
315 | return 1; |
316 | } | |
838acbc2 | 317 | |
83b6208e NB |
318 | if (st == NULL) |
319 | st = tst; | |
320 | if (st->ss != tst->ss || | |
321 | st->minor_version != tst->minor_version || | |
322 | st->ss->compare_super(&first_super, super) != 0) { | |
323 | /* Some mismatch. If exactly one array matches this host, | |
da6b5ca9 NB |
324 | * we can resolve on that one. |
325 | * Or, if we are auto assembling, we just ignore the second | |
326 | * for now. | |
83b6208e | 327 | */ |
da6b5ca9 NB |
328 | if (mdfd < 0) |
329 | continue; | |
83b6208e NB |
330 | if (homehost) { |
331 | int first = st->ss->match_home(first_super, homehost); | |
332 | int last = tst->ss->match_home(super, homehost); | |
333 | if (first+last == 1) { | |
334 | /* We can do something */ | |
335 | if (first) {/* just ignore this one */ | |
336 | if ((inargv && verbose >= 0) || verbose > 0) | |
337 | fprintf(stderr, Name ": %s misses out due to wrong homehost\n", | |
338 | devname); | |
339 | continue; | |
340 | } else { /* reject all those sofar */ | |
341 | mddev_dev_t td; | |
342 | if ((inargv && verbose >= 0) || verbose > 0) | |
343 | fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n", | |
344 | devname); | |
345 | for (td=devlist; td != tmpdev; td=td->next) | |
346 | if (td->used == 1) | |
347 | td->used = 0; | |
348 | tmpdev->used = 1; | |
349 | continue; | |
350 | } | |
351 | } | |
352 | } | |
52826846 NB |
353 | fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n", |
354 | devname); | |
4b1ac34b NB |
355 | free(super); |
356 | free(first_super); | |
52826846 | 357 | return 1; |
64c4757e NB |
358 | } |
359 | ||
811e6cbe NB |
360 | tmpdev->used = 1; |
361 | } | |
362 | ||
8a46fe84 NB |
363 | if (mdfd < 0) { |
364 | /* So... it is up to me to open the device. | |
365 | * We create a name '/dev/md/XXX' based on the info in the | |
366 | * superblock, and call open_mddev on that | |
367 | */ | |
da6b5ca9 NB |
368 | mdu_array_info_t inf; |
369 | char *c; | |
370 | if (!first_super) { | |
d55e3aef | 371 | return 2; |
da6b5ca9 NB |
372 | } |
373 | st->ss->getinfo_super(&info, first_super); | |
374 | c = strchr(info.name, ':'); | |
375 | if (c) c++; else c= info.name; | |
376 | asprintf(&mddev, "/dev/md/%s", c); | |
377 | mdfd = open_mddev(mddev, ident->autof); | |
8a46fe84 NB |
378 | if (mdfd < 0) |
379 | return mdfd; | |
da6b5ca9 NB |
380 | vers = md_get_version(mdfd); |
381 | if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) { | |
382 | fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev); | |
383 | close(mdfd); | |
384 | free(first_super); | |
385 | return 1; | |
386 | } | |
d55e3aef | 387 | must_close = 1; |
8a46fe84 NB |
388 | } |
389 | ||
811e6cbe | 390 | /* Ok, no bad inconsistancy, we can try updating etc */ |
da6b5ca9 | 391 | for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) { |
811e6cbe NB |
392 | char *devname = tmpdev->devname; |
393 | struct stat stb; | |
5787fa49 NB |
394 | /* looks like a good enough match to update the super block if needed */ |
395 | if (update) { | |
811e6cbe | 396 | int dfd; |
4b1ac34b NB |
397 | /* prepare useful information in info structures */ |
398 | struct stat stb2; | |
399 | fstat(mdfd, &stb2); | |
7d99579f NB |
400 | |
401 | if (strcmp(update, "uuid")==0 && | |
402 | !ident->uuid_set) { | |
403 | int rfd; | |
404 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
405 | read(rfd, ident->uuid, 16) != 16) { | |
406 | *(__u32*)(ident->uuid) = random(); | |
407 | *(__u32*)(ident->uuid+1) = random(); | |
408 | *(__u32*)(ident->uuid+2) = random(); | |
409 | *(__u32*)(ident->uuid+3) = random(); | |
410 | } | |
411 | if (rfd >= 0) close(rfd); | |
7d99579f | 412 | } |
811e6cbe NB |
413 | dfd = dev_open(devname, O_RDWR|O_EXCL); |
414 | ||
415 | if (super) { | |
416 | free(super); | |
417 | super = NULL; | |
418 | } | |
419 | ||
420 | st->ss->load_super(st, dfd, &super, NULL); | |
421 | st->ss->getinfo_super(&info, super); | |
422 | ||
7d99579f | 423 | memcpy(info.uuid, ident->uuid, 16); |
e5eac01f | 424 | strcpy(info.name, ident->name); |
811e6cbe NB |
425 | info.array.md_minor = minor(stb2.st_rdev); |
426 | ||
e5eac01f NB |
427 | st->ss->update_super(&info, super, update, devname, verbose, |
428 | ident->uuid_set, homehost); | |
429 | if (strcmp(update, "uuid")==0 && | |
430 | !ident->uuid_set) { | |
431 | ident->uuid_set = 1; | |
432 | memcpy(ident->uuid, info.uuid, 16); | |
433 | } | |
b8a8ccf9 | 434 | if (dfd < 0) |
5787fa49 NB |
435 | fprintf(stderr, Name ": Cannot open %s for superblock update\n", |
436 | devname); | |
96395475 | 437 | else if (st->ss->store_super(st, dfd, super)) |
5787fa49 NB |
438 | fprintf(stderr, Name ": Could not re-write superblock on %s.\n", |
439 | devname); | |
440 | if (dfd >= 0) | |
441 | close(dfd); | |
8131b493 NB |
442 | |
443 | if (strcmp(update, "uuid")==0 && | |
444 | ident->bitmap_fd) | |
445 | bitmap_update_uuid(ident->bitmap_fd, info.uuid); | |
da6b5ca9 NB |
446 | } else { |
447 | int dfd; | |
448 | dfd = dev_open(devname, O_RDWR|O_EXCL); | |
449 | ||
450 | if (super) { | |
451 | free(super); | |
452 | super = NULL; | |
453 | } | |
454 | ||
455 | st->ss->load_super(st, dfd, &super, NULL); | |
456 | st->ss->getinfo_super(&info, super); | |
457 | close(dfd); | |
5787fa49 NB |
458 | } |
459 | ||
811e6cbe NB |
460 | stat(devname, &stb); |
461 | ||
dab6685f | 462 | if (verbose > 0) |
cd29a5c8 | 463 | fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n", |
4b1ac34b | 464 | devname, mddev, info.disk.raid_disk); |
64c4757e | 465 | devices[devcnt].devname = devname; |
0df46c2a NB |
466 | devices[devcnt].major = major(stb.st_rdev); |
467 | devices[devcnt].minor = minor(stb.st_rdev); | |
4b1ac34b NB |
468 | devices[devcnt].oldmajor = info.disk.major; |
469 | devices[devcnt].oldminor = info.disk.minor; | |
470 | devices[devcnt].events = info.events; | |
471 | devices[devcnt].raid_disk = info.disk.raid_disk; | |
6a41304b | 472 | devices[devcnt].disk_nr = info.disk.number; |
64c4757e | 473 | devices[devcnt].uptodate = 0; |
4b1ac34b | 474 | devices[devcnt].state = info.disk.state; |
64c4757e NB |
475 | if (most_recent < devcnt) { |
476 | if (devices[devcnt].events | |
477 | > devices[most_recent].events) | |
478 | most_recent = devcnt; | |
479 | } | |
b8a8ccf9 | 480 | if (info.array.level == -4) |
5787fa49 NB |
481 | /* with multipath, the raid_disk from the superblock is meaningless */ |
482 | i = devcnt; | |
483 | else | |
484 | i = devices[devcnt].raid_disk; | |
308e1801 NB |
485 | if (i+1 == 0) { |
486 | if (nextspare < info.array.raid_disks) | |
487 | nextspare = info.array.raid_disks; | |
488 | i = nextspare++; | |
489 | } | |
98c6faba | 490 | if (i < 10000) { |
56eedc1a | 491 | if (i >= bestcnt) { |
98c6faba | 492 | unsigned int newbestcnt = i+10; |
56eedc1a | 493 | int *newbest = malloc(sizeof(int)*newbestcnt); |
98c6faba | 494 | unsigned int c; |
56eedc1a NB |
495 | for (c=0; c < newbestcnt; c++) |
496 | if (c < bestcnt) | |
497 | newbest[c] = best[c]; | |
498 | else | |
499 | newbest[c] = -1; | |
500 | if (best)free(best); | |
501 | best = newbest; | |
502 | bestcnt = newbestcnt; | |
503 | } | |
52826846 NB |
504 | if (best[i] == -1 |
505 | || devices[best[i]].events < devices[devcnt].events) | |
506 | best[i] = devcnt; | |
56eedc1a | 507 | } |
64c4757e NB |
508 | devcnt++; |
509 | } | |
510 | ||
4b1ac34b NB |
511 | if (super) |
512 | free(super); | |
513 | super = NULL; | |
514 | ||
586ed405 NB |
515 | if (update && strcmp(update, "byteorder")==0) |
516 | st->minor_version = 90; | |
517 | ||
64c4757e | 518 | if (devcnt == 0) { |
682c7051 | 519 | fprintf(stderr, Name ": no devices found for %s\n", |
64c4757e | 520 | mddev); |
4b1ac34b | 521 | free(first_super); |
d55e3aef | 522 | if (must_close) close(mdfd); |
64c4757e NB |
523 | return 1; |
524 | } | |
4b1ac34b | 525 | |
31317663 | 526 | st->ss->getinfo_super(&info, first_super); |
4b1ac34b | 527 | |
64c4757e NB |
528 | /* now we have some devices that might be suitable. |
529 | * I wonder how many | |
530 | */ | |
265e0f17 NB |
531 | avail = malloc(info.array.raid_disks); |
532 | memset(avail, 0, info.array.raid_disks); | |
64c4757e | 533 | okcnt = 0; |
52826846 | 534 | sparecnt=0; |
56eedc1a | 535 | for (i=0; i< bestcnt ;i++) { |
64c4757e | 536 | int j = best[i]; |
ee04451c NB |
537 | int event_margin = 1; /* always allow a difference of '1' |
538 | * like the kernel does | |
539 | */ | |
64c4757e | 540 | if (j < 0) continue; |
d013a55e NB |
541 | /* note: we ignore error flags in multipath arrays |
542 | * as they don't make sense | |
543 | */ | |
4b1ac34b | 544 | if (info.array.level != -4) |
aa88f531 NB |
545 | if (!(devices[j].state & (1<<MD_DISK_SYNC))) { |
546 | if (!(devices[j].state & (1<<MD_DISK_FAULTY))) | |
547 | sparecnt++; | |
d013a55e | 548 | continue; |
aa88f531 | 549 | } |
cd29a5c8 | 550 | if (devices[j].events+event_margin >= |
64c4757e NB |
551 | devices[most_recent].events) { |
552 | devices[j].uptodate = 1; | |
265e0f17 | 553 | if (i < info.array.raid_disks) { |
52826846 | 554 | okcnt++; |
265e0f17 NB |
555 | avail[i]=1; |
556 | } else | |
52826846 | 557 | sparecnt++; |
64c4757e NB |
558 | } |
559 | } | |
265e0f17 NB |
560 | while (force && !enough(info.array.level, info.array.raid_disks, |
561 | info.array.layout, | |
562 | avail, okcnt)) { | |
64c4757e NB |
563 | /* Choose the newest best drive which is |
564 | * not up-to-date, update the superblock | |
565 | * and add it. | |
566 | */ | |
52826846 | 567 | int fd; |
cd29a5c8 | 568 | chosen_drive = -1; |
4b1ac34b | 569 | for (i=0; i<info.array.raid_disks && i < bestcnt; i++) { |
52826846 NB |
570 | int j = best[i]; |
571 | if (j>=0 && | |
572 | !devices[j].uptodate && | |
573 | devices[j].events > 0 && | |
574 | (chosen_drive < 0 || | |
575 | devices[j].events > devices[chosen_drive].events)) | |
576 | chosen_drive = j; | |
577 | } | |
578 | if (chosen_drive < 0) | |
579 | break; | |
dab6685f NB |
580 | if (verbose >= 0) |
581 | fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n", | |
582 | devices[chosen_drive].devname, devices[chosen_drive].raid_disk, | |
583 | (int)(devices[chosen_drive].events), | |
584 | (int)(devices[most_recent].events)); | |
8b0dabea | 585 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 NB |
586 | if (fd < 0) { |
587 | fprintf(stderr, Name ": Couldn't open %s for write - not updating\n", | |
588 | devices[chosen_drive].devname); | |
589 | devices[chosen_drive].events = 0; | |
590 | continue; | |
591 | } | |
82d9eba6 | 592 | if (st->ss->load_super(st,fd, &super, NULL)) { |
52826846 NB |
593 | close(fd); |
594 | fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n", | |
595 | devices[chosen_drive].devname); | |
596 | devices[chosen_drive].events = 0; | |
597 | continue; | |
598 | } | |
4b1ac34b | 599 | info.events = devices[most_recent].events; |
e5eac01f | 600 | st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL); |
4b1ac34b | 601 | |
96395475 | 602 | if (st->ss->store_super(st, fd, super)) { |
52826846 NB |
603 | close(fd); |
604 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
605 | devices[chosen_drive].devname); | |
606 | devices[chosen_drive].events = 0; | |
4b1ac34b | 607 | free(super); |
52826846 NB |
608 | continue; |
609 | } | |
610 | close(fd); | |
611 | devices[chosen_drive].events = devices[most_recent].events; | |
612 | devices[chosen_drive].uptodate = 1; | |
265e0f17 | 613 | avail[chosen_drive] = 1; |
52826846 | 614 | okcnt++; |
4b1ac34b | 615 | free(super); |
64c4757e | 616 | } |
52826846 NB |
617 | |
618 | /* Now we want to look at the superblock which the kernel will base things on | |
619 | * and compare the devices that we think are working with the devices that the | |
620 | * superblock thinks are working. | |
621 | * If there are differences and --force is given, then update this chosen | |
622 | * superblock. | |
623 | */ | |
cd29a5c8 | 624 | chosen_drive = -1; |
4b1ac34b | 625 | super = NULL; |
56eedc1a | 626 | for (i=0; chosen_drive < 0 && i<bestcnt; i++) { |
52826846 NB |
627 | int j = best[i]; |
628 | int fd; | |
4b1ac34b | 629 | |
52826846 NB |
630 | if (j<0) |
631 | continue; | |
632 | if (!devices[j].uptodate) | |
633 | continue; | |
634 | chosen_drive = j; | |
8b0dabea | 635 | if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) { |
52826846 NB |
636 | fprintf(stderr, Name ": Cannot open %s: %s\n", |
637 | devices[j].devname, strerror(errno)); | |
d55e3aef | 638 | if (must_close) close(mdfd); |
52826846 NB |
639 | return 1; |
640 | } | |
82d9eba6 | 641 | if (st->ss->load_super(st,fd, &super, NULL)) { |
52826846 NB |
642 | close(fd); |
643 | fprintf(stderr, Name ": RAID superblock has disappeared from %s\n", | |
644 | devices[j].devname); | |
d55e3aef | 645 | if (must_close) close(mdfd); |
52826846 NB |
646 | return 1; |
647 | } | |
648 | close(fd); | |
649 | } | |
4b1ac34b NB |
650 | if (super == NULL) { |
651 | fprintf(stderr, Name ": No suitable drives found for %s\n", mddev); | |
d55e3aef | 652 | if (must_close) close(mdfd); |
4b1ac34b NB |
653 | return 1; |
654 | } | |
31317663 | 655 | st->ss->getinfo_super(&info, super); |
56eedc1a | 656 | for (i=0; i<bestcnt; i++) { |
52826846 | 657 | int j = best[i]; |
98c6faba | 658 | unsigned int desired_state; |
11a3e71d | 659 | |
4b1ac34b | 660 | if (i < info.array.raid_disks) |
11a3e71d NB |
661 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); |
662 | else | |
663 | desired_state = 0; | |
664 | ||
52826846 NB |
665 | if (j<0) |
666 | continue; | |
667 | if (!devices[j].uptodate) | |
668 | continue; | |
6a41304b | 669 | info.disk.number = devices[j].disk_nr; |
fbf8a0b7 | 670 | info.disk.raid_disk = i; |
4b1ac34b NB |
671 | info.disk.state = desired_state; |
672 | ||
52826846 | 673 | if (devices[j].uptodate && |
e5eac01f | 674 | st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) { |
52826846 | 675 | if (force) { |
dab6685f NB |
676 | if (verbose >= 0) |
677 | fprintf(stderr, Name ": " | |
678 | "clearing FAULTY flag for device %d in %s for %s\n", | |
679 | j, mddev, devices[j].devname); | |
4b1ac34b | 680 | change = 1; |
52826846 | 681 | } else { |
dab6685f NB |
682 | if (verbose >= -1) |
683 | fprintf(stderr, Name ": " | |
684 | "device %d in %s has wrong state in superblock, but %s seems ok\n", | |
685 | i, mddev, devices[j].devname); | |
52826846 NB |
686 | } |
687 | } | |
4b1ac34b | 688 | #if 0 |
52826846 NB |
689 | if (!devices[j].uptodate && |
690 | !(super.disks[i].state & (1 << MD_DISK_FAULTY))) { | |
691 | fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", | |
692 | i, mddev); | |
693 | } | |
4b1ac34b | 694 | #endif |
52826846 | 695 | } |
4b1ac34b NB |
696 | if (force && okcnt == info.array.raid_disks-1) { |
697 | /* FIXME check event count */ | |
b8a8ccf9 | 698 | change += st->ss->update_super(&info, super, "force", |
e5eac01f | 699 | devices[chosen_drive].devname, verbose, 0, NULL); |
d013a55e | 700 | } |
52826846 | 701 | |
4b1ac34b | 702 | if (change) { |
52826846 | 703 | int fd; |
8b0dabea | 704 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 | 705 | if (fd < 0) { |
353632d9 | 706 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", |
52826846 | 707 | devices[chosen_drive].devname); |
d55e3aef | 708 | if (must_close) close(mdfd); |
52826846 NB |
709 | return 1; |
710 | } | |
96395475 | 711 | if (st->ss->store_super(st, fd, super)) { |
52826846 NB |
712 | close(fd); |
713 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
714 | devices[chosen_drive].devname); | |
d55e3aef | 715 | if (must_close) close(mdfd); |
52826846 NB |
716 | return 1; |
717 | } | |
718 | close(fd); | |
52826846 NB |
719 | } |
720 | ||
353632d9 NB |
721 | /* If we are in the middle of a reshape we may need to restore saved data |
722 | * that was moved aside due to the reshape overwriting live data | |
723 | * The code of doing this lives in Grow.c | |
724 | */ | |
39c5a909 | 725 | #ifndef MDASSEMBLE |
353632d9 NB |
726 | if (info.reshape_active) { |
727 | int err = 0; | |
728 | int *fdlist = malloc(sizeof(int)* bestcnt); | |
729 | for (i=0; i<bestcnt; i++) { | |
730 | int j = best[i]; | |
731 | if (j >= 0) { | |
732 | fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
733 | if (fdlist[i] < 0) { | |
734 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", | |
735 | devices[j].devname); | |
736 | err = 1; | |
737 | break; | |
738 | } | |
739 | } else | |
740 | fdlist[i] = -1; | |
741 | } | |
742 | if (!err) | |
06b0d786 | 743 | err = Grow_restart(st, &info, fdlist, bestcnt, backup_file); |
353632d9 NB |
744 | while (i>0) { |
745 | i--; | |
746 | if (fdlist[i]>=0) close(fdlist[i]); | |
747 | } | |
748 | if (err) { | |
749 | fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n"); | |
d55e3aef | 750 | if (must_close) close(mdfd); |
353632d9 NB |
751 | return err; |
752 | } | |
753 | } | |
39c5a909 | 754 | #endif |
aa88f531 NB |
755 | /* count number of in-sync devices according to the superblock. |
756 | * We must have this number to start the array without -s or -R | |
757 | */ | |
4b1ac34b | 758 | req_cnt = info.array.working_disks; |
aa88f531 | 759 | |
64c4757e NB |
760 | /* Almost ready to actually *do* something */ |
761 | if (!old_linux) { | |
fbf8a0b7 NB |
762 | int rv; |
763 | if ((vers % 100) >= 1) { /* can use different versions */ | |
764 | mdu_array_info_t inf; | |
765 | memset(&inf, 0, sizeof(inf)); | |
766 | inf.major_version = st->ss->major; | |
767 | inf.minor_version = st->minor_version; | |
768 | rv = ioctl(mdfd, SET_ARRAY_INFO, &inf); | |
b8a8ccf9 | 769 | } else |
fbf8a0b7 NB |
770 | rv = ioctl(mdfd, SET_ARRAY_INFO, NULL); |
771 | ||
772 | if (rv) { | |
682c7051 | 773 | fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n", |
64c4757e | 774 | mddev, strerror(errno)); |
d55e3aef | 775 | if (must_close) close(mdfd); |
64c4757e NB |
776 | return 1; |
777 | } | |
11484a63 | 778 | if (ident->bitmap_fd >= 0) { |
c82f047c NB |
779 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { |
780 | fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n"); | |
d55e3aef | 781 | if (must_close) close(mdfd); |
c82f047c NB |
782 | return 1; |
783 | } | |
7ef02d01 NB |
784 | } else if (ident->bitmap_file) { |
785 | /* From config file */ | |
786 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
787 | if (bmfd < 0) { | |
788 | fprintf(stderr, Name ": Could not open bitmap file %s\n", | |
789 | ident->bitmap_file); | |
d55e3aef | 790 | if (must_close) close(mdfd); |
7ef02d01 NB |
791 | return 1; |
792 | } | |
793 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
794 | fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev); | |
795 | close(bmfd); | |
d55e3aef | 796 | if (must_close) close(mdfd); |
7ef02d01 NB |
797 | return 1; |
798 | } | |
799 | close(bmfd); | |
c82f047c | 800 | } |
7ef02d01 | 801 | |
52826846 | 802 | /* First, add the raid disks, but add the chosen one last */ |
56eedc1a | 803 | for (i=0; i<= bestcnt; i++) { |
52826846 | 804 | int j; |
56eedc1a | 805 | if (i < bestcnt) { |
52826846 NB |
806 | j = best[i]; |
807 | if (j == chosen_drive) | |
808 | continue; | |
809 | } else | |
810 | j = chosen_drive; | |
811 | ||
56eedc1a | 812 | if (j >= 0 /* && devices[j].uptodate */) { |
64c4757e NB |
813 | mdu_disk_info_t disk; |
814 | memset(&disk, 0, sizeof(disk)); | |
815 | disk.major = devices[j].major; | |
816 | disk.minor = devices[j].minor; | |
817 | if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) { | |
682c7051 | 818 | fprintf(stderr, Name ": failed to add %s to %s: %s\n", |
64c4757e NB |
819 | devices[j].devname, |
820 | mddev, | |
821 | strerror(errno)); | |
4b1ac34b | 822 | if (i < info.array.raid_disks || i == bestcnt) |
52826846 NB |
823 | okcnt--; |
824 | else | |
825 | sparecnt--; | |
dab6685f | 826 | } else if (verbose > 0) |
52826846 NB |
827 | fprintf(stderr, Name ": added %s to %s as %d\n", |
828 | devices[j].devname, mddev, devices[j].raid_disk); | |
dab6685f | 829 | } else if (verbose > 0 && i < info.array.raid_disks) |
682c7051 | 830 | fprintf(stderr, Name ": no uptodate device for slot %d of %s\n", |
64c4757e NB |
831 | i, mddev); |
832 | } | |
52826846 | 833 | |
64c4757e | 834 | if (runstop == 1 || |
b8a8ccf9 | 835 | (runstop <= 0 && |
265e0f17 | 836 | ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) && |
aa88f531 NB |
837 | (okcnt >= req_cnt || start_partial_ok) |
838 | ))) { | |
82b27616 | 839 | if (ioctl(mdfd, RUN_ARRAY, NULL)==0) { |
dab6685f NB |
840 | if (verbose >= 0) { |
841 | fprintf(stderr, Name ": %s has been started with %d drive%s", | |
842 | mddev, okcnt, okcnt==1?"":"s"); | |
b8a8ccf9 | 843 | if (okcnt < info.array.raid_disks) |
dab6685f NB |
844 | fprintf(stderr, " (out of %d)", info.array.raid_disks); |
845 | if (sparecnt) | |
846 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
847 | fprintf(stderr, ".\n"); | |
848 | } | |
d55e3aef | 849 | if (must_close) close(mdfd); |
64c4757e | 850 | return 0; |
82b27616 | 851 | } |
682c7051 | 852 | fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n", |
64c4757e | 853 | mddev, strerror(errno)); |
d55e3aef | 854 | if (must_close) close(mdfd); |
64c4757e NB |
855 | return 1; |
856 | } | |
82b27616 | 857 | if (runstop == -1) { |
b8a8ccf9 | 858 | fprintf(stderr, Name ": %s assembled from %d drive%s", |
52826846 | 859 | mddev, okcnt, okcnt==1?"":"s"); |
b8a8ccf9 NB |
860 | if (okcnt != info.array.raid_disks) |
861 | fprintf(stderr, " (out of %d)", info.array.raid_disks); | |
862 | fprintf(stderr, ", but not started.\n"); | |
d55e3aef | 863 | if (must_close) close(mdfd); |
64c4757e | 864 | return 0; |
82b27616 | 865 | } |
dab6685f NB |
866 | if (verbose >= 0) { |
867 | fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s"); | |
868 | if (sparecnt) | |
869 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
265e0f17 | 870 | if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt)) |
dab6685f NB |
871 | fprintf(stderr, " - not enough to start the array.\n"); |
872 | else { | |
873 | if (req_cnt == info.array.raid_disks) | |
874 | fprintf(stderr, " - need all %d to start it", req_cnt); | |
875 | else | |
876 | fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks); | |
877 | fprintf(stderr, " (use --run to insist).\n"); | |
878 | } | |
aa88f531 | 879 | } |
d55e3aef | 880 | if (must_close) close(mdfd); |
64c4757e | 881 | return 1; |
82b27616 | 882 | } else { |
52826846 NB |
883 | /* The "chosen_drive" is a good choice, and if necessary, the superblock has |
884 | * been updated to point to the current locations of devices. | |
885 | * so we can just start the array | |
82b27616 | 886 | */ |
cd29a5c8 | 887 | unsigned long dev; |
0df46c2a | 888 | dev = makedev(devices[chosen_drive].major, |
82b27616 NB |
889 | devices[chosen_drive].minor); |
890 | if (ioctl(mdfd, START_ARRAY, dev)) { | |
891 | fprintf(stderr, Name ": Cannot start array: %s\n", | |
892 | strerror(errno)); | |
893 | } | |
894 | ||
64c4757e | 895 | } |
d55e3aef | 896 | if (must_close) close(mdfd); |
e0d19036 | 897 | return 0; |
64c4757e | 898 | } |