]>
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" |
41a3b72a | 31 | #include <ctype.h> |
64c4757e | 32 | |
624920bb NB |
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 | ||
9008ed1c | 53 | static int is_member_busy(char *metadata_version) |
2de8884f DW |
54 | { |
55 | /* check if the given member array is active */ | |
56 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
57 | struct mdstat_ent *ent; | |
58 | int busy = 0; | |
59 | ||
60 | for (ent = mdstat; ent; ent = ent->next) { | |
61 | if (ent->metadata_version == NULL) | |
62 | continue; | |
63 | if (strncmp(ent->metadata_version, "external:", 9) != 0) | |
64 | continue; | |
65 | if (!is_subarray(&ent->metadata_version[9])) | |
66 | continue; | |
67 | /* Skip first char - it can be '/' or '-' */ | |
68 | if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) { | |
69 | busy = 1; | |
70 | break; | |
71 | } | |
72 | } | |
73 | free_mdstat(mdstat); | |
74 | ||
75 | return busy; | |
76 | } | |
77 | ||
7f91af49 | 78 | int Assemble(struct supertype *st, char *mddev, |
8aec876d | 79 | mddev_ident_t ident, |
06b0d786 | 80 | mddev_dev_t devlist, char *backup_file, |
64c4757e | 81 | int readonly, int runstop, |
e5eac01f | 82 | char *update, char *homehost, |
64c4757e NB |
83 | int verbose, int force) |
84 | { | |
85 | /* | |
52826846 NB |
86 | * The task of Assemble is to find a collection of |
87 | * devices that should (according to their superblocks) | |
88 | * form an array, and to give this collection to the MD driver. | |
89 | * In Linux-2.4 and later, this involves submitting a | |
64c4757e NB |
90 | * SET_ARRAY_INFO ioctl with no arg - to prepare |
91 | * the array - and then submit a number of | |
92 | * ADD_NEW_DISK ioctls to add disks into | |
93 | * the array. Finally RUN_ARRAY might | |
94 | * be submitted to start the array. | |
95 | * | |
96 | * Much of the work of Assemble is in finding and/or | |
97 | * checking the disks to make sure they look right. | |
98 | * | |
4b1ac34b | 99 | * If mddev is not set, then scan must be set and we |
64c4757e NB |
100 | * read through the config file for dev+uuid mapping |
101 | * We recurse, setting mddev, for each device that | |
102 | * - isn't running | |
4b1ac34b | 103 | * - has a valid uuid (or any uuid if !uuidset) |
64c4757e NB |
104 | * |
105 | * If mddev is set, we try to determine state of md. | |
106 | * check version - must be at least 0.90.0 | |
107 | * check kernel version. must be at least 2.4. | |
108 | * If not, we can possibly fall back on START_ARRAY | |
109 | * Try to GET_ARRAY_INFO. | |
110 | * If possible, give up | |
111 | * If not, try to STOP_ARRAY just to make sure | |
112 | * | |
113 | * If !uuidset and scan, look in conf-file for uuid | |
114 | * If not found, give up | |
b8a8ccf9 | 115 | * If !devlist and scan and uuidset, get list of devs from conf-file |
64c4757e NB |
116 | * |
117 | * For each device: | |
118 | * Check superblock - discard if bad | |
119 | * Check uuid (set if we don't have one) - discard if no match | |
5787fa49 | 120 | * Check superblock similarity if we have a superblock - discard if different |
4b1ac34b | 121 | * Record events, devicenum |
64c4757e | 122 | * This should give us a list of devices for the array |
4b1ac34b | 123 | * We should collect the most recent event number |
64c4757e NB |
124 | * |
125 | * Count disks with recent enough event count | |
126 | * While force && !enough disks | |
127 | * Choose newest rejected disks, update event count | |
128 | * mark clean and rewrite superblock | |
129 | * If recent kernel: | |
130 | * SET_ARRAY_INFO | |
131 | * foreach device with recent events : ADD_NEW_DISK | |
132 | * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY | |
133 | * If old kernel: | |
134 | * Check the device numbers in superblock are right | |
135 | * update superblock if any changes | |
136 | * START_ARRAY | |
137 | * | |
138 | */ | |
c30e5369 N |
139 | int mdfd; |
140 | int clean; | |
f05641cf N |
141 | int auto_assem = (mddev == NULL && !ident->uuid_set && |
142 | ident->super_minor == UnSet && ident->name[0] == 0 | |
aa7c284c | 143 | && (ident->container == NULL || ident->member == NULL)); |
64c4757e | 144 | int old_linux = 0; |
c30e5369 | 145 | int vers = vers; /* Keep gcc quite - it really is initialised */ |
64c4757e NB |
146 | struct { |
147 | char *devname; | |
213ee40b NB |
148 | int uptodate; /* set once we decide that this device is as |
149 | * recent as everything else in the array. | |
150 | */ | |
151 | struct mdinfo i; | |
5787fa49 | 152 | } *devices; |
56eedc1a | 153 | int *best = NULL; /* indexed by raid_disk */ |
98c6faba NB |
154 | unsigned int bestcnt = 0; |
155 | int devcnt = 0; | |
156 | unsigned int okcnt, sparecnt; | |
157 | unsigned int req_cnt; | |
158 | unsigned int i; | |
64c4757e | 159 | int most_recent = 0; |
cd29a5c8 | 160 | int chosen_drive; |
52826846 | 161 | int change = 0; |
cd29a5c8 | 162 | int inargv = 0; |
52437b4f | 163 | int report_missmatch; |
bf4fb153 | 164 | int bitmap_done; |
7f91af49 N |
165 | int start_partial_ok = (runstop >= 0) && |
166 | (force || devlist==NULL || auto_assem); | |
98c6faba | 167 | unsigned int num_devs; |
5787fa49 | 168 | mddev_dev_t tmpdev; |
4b1ac34b | 169 | struct mdinfo info; |
98dbd966 | 170 | struct mdinfo *content = NULL; |
265e0f17 | 171 | char *avail; |
308e1801 | 172 | int nextspare = 0; |
197e3eb6 | 173 | char *name = NULL; |
69207ff6 | 174 | int trustworthy; |
a04d5763 | 175 | char chosen_name[1024]; |
8a46fe84 | 176 | |
682c7051 | 177 | if (get_linux_version() < 2004000) |
64c4757e NB |
178 | old_linux = 1; |
179 | ||
64c4757e | 180 | /* |
52826846 NB |
181 | * If any subdevs are listed, then any that don't |
182 | * match ident are discarded. Remainder must all match and | |
183 | * become the array. | |
184 | * If no subdevs, then we scan all devices in the config file, but | |
185 | * there must be something in the identity | |
64c4757e | 186 | */ |
64c4757e | 187 | |
cd29a5c8 | 188 | if (!devlist && |
52826846 NB |
189 | ident->uuid_set == 0 && |
190 | ident->super_minor < 0 && | |
191 | ident->devices == NULL) { | |
192 | fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n", | |
8a46fe84 | 193 | mddev ? mddev : "further assembly"); |
52826846 | 194 | return 1; |
64c4757e | 195 | } |
71d60c48 | 196 | |
cd29a5c8 | 197 | if (devlist == NULL) |
8aec876d | 198 | devlist = conf_get_devs(); |
7f91af49 | 199 | else if (mddev) |
da6b5ca9 | 200 | inargv = 1; |
64c4757e | 201 | |
52437b4f | 202 | report_missmatch = ((inargv && verbose >= 0) || verbose > 0); |
60e1bc1a | 203 | try_again: |
c30e5369 N |
204 | /* We come back here when doing auto-assembly and attempting some |
205 | * set of devices failed. Those are now marked as ->used==2 and | |
206 | * we ignore them and try again | |
207 | */ | |
60e1bc1a | 208 | |
5787fa49 NB |
209 | tmpdev = devlist; num_devs = 0; |
210 | while (tmpdev) { | |
da6b5ca9 NB |
211 | if (tmpdev->used) |
212 | tmpdev->used = 2; | |
213 | else | |
214 | num_devs++; | |
5787fa49 NB |
215 | tmpdev = tmpdev->next; |
216 | } | |
5787fa49 NB |
217 | devices = malloc(num_devs * sizeof(*devices)); |
218 | ||
82d9eba6 | 219 | if (!st && ident->st) st = ident->st; |
64c4757e | 220 | |
dab6685f | 221 | if (verbose>0) |
82b27616 | 222 | fprintf(stderr, Name ": looking for devices for %s\n", |
8a46fe84 | 223 | mddev ? mddev : "further assembly"); |
82b27616 | 224 | |
811e6cbe NB |
225 | /* first walk the list of devices to find a consistent set |
226 | * that match the criterea, if that is possible. | |
c30e5369 | 227 | * We flag the ones we like with 'used'. |
811e6cbe NB |
228 | */ |
229 | for (tmpdev = devlist; | |
230 | tmpdev; | |
231 | tmpdev = tmpdev->next) { | |
232 | char *devname = tmpdev->devname; | |
64c4757e NB |
233 | int dfd; |
234 | struct stat stb; | |
3da92f27 | 235 | struct supertype *tst = dup_super(st); |
52826846 | 236 | |
da6b5ca9 NB |
237 | if (tmpdev->used > 1) continue; |
238 | ||
52826846 | 239 | if (ident->devices && |
56eedc1a | 240 | !match_oneof(ident->devices, devname)) { |
52437b4f | 241 | if (report_missmatch) |
56eedc1a | 242 | fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices); |
52826846 | 243 | continue; |
56eedc1a | 244 | } |
4b1ac34b | 245 | |
8b0dabea | 246 | dfd = dev_open(devname, O_RDONLY|O_EXCL); |
64c4757e | 247 | if (dfd < 0) { |
52437b4f | 248 | if (report_missmatch) |
682c7051 | 249 | fprintf(stderr, Name ": cannot open device %s: %s\n", |
64c4757e | 250 | devname, strerror(errno)); |
da6b5ca9 | 251 | tmpdev->used = 2; |
52826846 NB |
252 | } else if (fstat(dfd, &stb)< 0) { |
253 | /* Impossible! */ | |
254 | fprintf(stderr, Name ": fstat failed for %s: %s\n", | |
255 | devname, strerror(errno)); | |
da6b5ca9 | 256 | tmpdev->used = 2; |
cd29a5c8 NB |
257 | } else if ((stb.st_mode & S_IFMT) != S_IFBLK) { |
258 | fprintf(stderr, Name ": %s is not a block device.\n", | |
52826846 | 259 | devname); |
da6b5ca9 | 260 | tmpdev->used = 2; |
82d9eba6 | 261 | } else if (!tst && (tst = guess_super(dfd)) == NULL) { |
52437b4f | 262 | if (report_missmatch) |
da6b5ca9 NB |
263 | fprintf(stderr, Name ": no recogniseable superblock on %s\n", |
264 | devname); | |
265 | tmpdev->used = 2; | |
31015d57 N |
266 | } else if (auto_assem && st == NULL && |
267 | !conf_test_metadata(tst->ss->name)) { | |
268 | if (report_missmatch) | |
269 | fprintf(stderr, Name ": %s has metadata type %s for which " | |
270 | "auto-assembly is disabled\n", | |
271 | devname, tst->ss->name); | |
272 | tmpdev->used = 2; | |
3da92f27 | 273 | } else if (tst->ss->load_super(tst,dfd, NULL)) { |
52437b4f | 274 | if (report_missmatch) |
682c7051 | 275 | fprintf( stderr, Name ": no RAID superblock on %s\n", |
64c4757e | 276 | devname); |
52826846 | 277 | } else { |
98dbd966 DW |
278 | content = &info; |
279 | memset(content, 0, sizeof(*content)); | |
280 | tst->ss->getinfo_super(tst, content); | |
64c4757e | 281 | } |
c06487ce | 282 | if (dfd >= 0) close(dfd); |
52826846 | 283 | |
9008ed1c N |
284 | if (tst && tst->sb && tst->ss->container_content |
285 | && tst->loaded_container) { | |
286 | /* tmpdev is a container. We need to be either | |
287 | * looking for a member, or auto-assembling | |
288 | */ | |
289 | if (st) { | |
290 | /* already found some components, this cannot | |
291 | * be another one. | |
292 | */ | |
293 | if (report_missmatch) | |
294 | fprintf(stderr, Name ": %s is a container, but we are looking for components\n", | |
295 | devname); | |
296 | goto loop; | |
297 | } | |
298 | ||
299 | if (ident->container) { | |
300 | if (ident->container[0] == '/' && | |
301 | !same_dev(ident->container, devname)) { | |
302 | if (report_missmatch) | |
303 | fprintf(stderr, Name ": %s is not the container required (%s)\n", | |
304 | devname, ident->container); | |
305 | goto loop; | |
306 | } | |
307 | if (ident->container[0] != '/') { | |
308 | /* we have a uuid */ | |
309 | int uuid[4]; | |
310 | if (!parse_uuid(ident->container, uuid) || | |
311 | !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) { | |
312 | if (report_missmatch) | |
313 | fprintf(stderr, Name ": %s has wrong UUID to be required container\n", | |
314 | devname); | |
315 | goto loop; | |
316 | } | |
317 | } | |
318 | } | |
319 | /* It is worth looking inside this container. | |
320 | */ | |
321 | next_member: | |
322 | if (tmpdev->content) | |
323 | content = tmpdev->content; | |
324 | else | |
325 | content = tst->ss->container_content(tst); | |
326 | ||
327 | tmpdev->content = content->next; | |
328 | if (tmpdev->content == NULL) | |
f7ad3ccc | 329 | tmpdev->used = 2; |
9008ed1c N |
330 | |
331 | } else if (ident->container || ident->member) { | |
332 | /* No chance of this matching if we don't have | |
333 | * a container */ | |
334 | if (report_missmatch) | |
335 | fprintf(stderr, Name "%s is not a container, and one is required.\n", | |
336 | devname); | |
337 | goto loop; | |
338 | } | |
339 | ||
838acbc2 | 340 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && |
3da92f27 | 341 | (!tst || !tst->sb || |
98dbd966 | 342 | same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) { |
52437b4f | 343 | if (report_missmatch) |
52826846 NB |
344 | fprintf(stderr, Name ": %s has wrong uuid.\n", |
345 | devname); | |
df37ffc0 | 346 | goto loop; |
82b27616 | 347 | } |
c4f12c13 | 348 | if (ident->name[0] && (!update || strcmp(update, "name")!= 0) && |
3da92f27 | 349 | (!tst || !tst->sb || |
98dbd966 | 350 | name_matches(content->name, ident->name, homehost)==0)) { |
52437b4f | 351 | if (report_missmatch) |
947fd4dd NB |
352 | fprintf(stderr, Name ": %s has wrong name.\n", |
353 | devname); | |
df37ffc0 | 354 | goto loop; |
947fd4dd | 355 | } |
98c6faba | 356 | if (ident->super_minor != UnSet && |
3da92f27 | 357 | (!tst || !tst->sb || |
98dbd966 | 358 | ident->super_minor != content->array.md_minor)) { |
52437b4f | 359 | if (report_missmatch) |
52826846 | 360 | fprintf(stderr, Name ": %s has wrong super-minor.\n", |
64c4757e | 361 | devname); |
df37ffc0 | 362 | goto loop; |
64c4757e | 363 | } |
98c6faba | 364 | if (ident->level != UnSet && |
3da92f27 | 365 | (!tst || !tst->sb || |
98dbd966 | 366 | ident->level != content->array.level)) { |
52437b4f | 367 | if (report_missmatch) |
cd29a5c8 NB |
368 | fprintf(stderr, Name ": %s has wrong raid level.\n", |
369 | devname); | |
df37ffc0 | 370 | goto loop; |
cd29a5c8 | 371 | } |
98c6faba | 372 | if (ident->raid_disks != UnSet && |
3da92f27 | 373 | (!tst || !tst->sb || |
98dbd966 | 374 | ident->raid_disks!= content->array.raid_disks)) { |
52437b4f | 375 | if (report_missmatch) |
cd29a5c8 NB |
376 | fprintf(stderr, Name ": %s requires wrong number of drives.\n", |
377 | devname); | |
df37ffc0 | 378 | goto loop; |
cd29a5c8 | 379 | } |
c30e5369 | 380 | if (auto_assem) { |
3da92f27 | 381 | if (tst == NULL || tst->sb == NULL) |
da6b5ca9 | 382 | continue; |
da6b5ca9 | 383 | } |
83b6208e | 384 | /* If we are this far, then we are nearly commited to this device. |
52826846 | 385 | * If the super_block doesn't exist, or doesn't match others, |
83b6208e NB |
386 | * then we probably cannot continue |
387 | * However if one of the arrays is for the homehost, and | |
388 | * the other isn't that can disambiguate. | |
52826846 | 389 | */ |
52826846 | 390 | |
3da92f27 | 391 | if (!tst || !tst->sb) { |
52826846 NB |
392 | fprintf(stderr, Name ": %s has no superblock - assembly aborted\n", |
393 | devname); | |
ddd1a492 NB |
394 | if (st) |
395 | st->ss->free_super(st); | |
52826846 NB |
396 | return 1; |
397 | } | |
838acbc2 | 398 | |
9008ed1c N |
399 | if (tst && tst->sb && tst->ss->container_content |
400 | && tst->loaded_container) { | |
401 | /* we have the one container we need, don't keep | |
402 | * looking. If the chosen member is active, skip. | |
403 | */ | |
404 | if (is_member_busy(content->text_version)) { | |
f7ad3ccc N |
405 | if (report_missmatch) |
406 | fprintf(stderr, Name ": member %s in %s is already assembled\n", | |
407 | content->text_version, | |
408 | devname); | |
409 | tst->ss->free_super(tst); | |
410 | tst = NULL; | |
411 | content = NULL; | |
9008ed1c N |
412 | if (auto_assem) |
413 | goto loop; | |
9008ed1c N |
414 | return 1; |
415 | } | |
416 | st = tst; tst = NULL; | |
417 | if (!auto_assem && tmpdev->next != NULL) { | |
418 | fprintf(stderr, Name ": %s is a container, but is not " | |
419 | "only device given: confused and aborting\n", | |
420 | devname); | |
421 | st->ss->free_super(st); | |
422 | return 1; | |
423 | } | |
424 | break; | |
425 | } | |
83b6208e | 426 | if (st == NULL) |
3da92f27 NB |
427 | st = dup_super(tst); |
428 | if (st->minor_version == -1) | |
429 | st->minor_version = tst->minor_version; | |
83b6208e NB |
430 | if (st->ss != tst->ss || |
431 | st->minor_version != tst->minor_version || | |
64557c33 | 432 | st->ss->compare_super(st, tst) != 0) { |
83b6208e | 433 | /* Some mismatch. If exactly one array matches this host, |
da6b5ca9 NB |
434 | * we can resolve on that one. |
435 | * Or, if we are auto assembling, we just ignore the second | |
436 | * for now. | |
83b6208e | 437 | */ |
c30e5369 | 438 | if (auto_assem) |
df37ffc0 | 439 | goto loop; |
83b6208e | 440 | if (homehost) { |
3da92f27 NB |
441 | int first = st->ss->match_home(st, homehost); |
442 | int last = tst->ss->match_home(tst, homehost); | |
9362c1c8 N |
443 | if (first != last && |
444 | (first == 1 || last == 1)) { | |
83b6208e NB |
445 | /* We can do something */ |
446 | if (first) {/* just ignore this one */ | |
52437b4f | 447 | if (report_missmatch) |
83b6208e NB |
448 | fprintf(stderr, Name ": %s misses out due to wrong homehost\n", |
449 | devname); | |
df37ffc0 | 450 | goto loop; |
83b6208e NB |
451 | } else { /* reject all those sofar */ |
452 | mddev_dev_t td; | |
52437b4f | 453 | if (report_missmatch) |
83b6208e NB |
454 | fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n", |
455 | devname); | |
456 | for (td=devlist; td != tmpdev; td=td->next) | |
457 | if (td->used == 1) | |
458 | td->used = 0; | |
459 | tmpdev->used = 1; | |
df37ffc0 | 460 | goto loop; |
83b6208e NB |
461 | } |
462 | } | |
463 | } | |
52826846 NB |
464 | fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n", |
465 | devname); | |
3da92f27 NB |
466 | tst->ss->free_super(tst); |
467 | st->ss->free_super(st); | |
52826846 | 468 | return 1; |
64c4757e NB |
469 | } |
470 | ||
811e6cbe | 471 | tmpdev->used = 1; |
df37ffc0 NB |
472 | |
473 | loop: | |
9008ed1c N |
474 | if (tmpdev->content) |
475 | goto next_member; | |
3d2b16e7 NB |
476 | if (tst) |
477 | tst->ss->free_super(tst); | |
811e6cbe NB |
478 | } |
479 | ||
98dbd966 | 480 | if (!st || !st->sb || !content) |
c30e5369 N |
481 | return 2; |
482 | ||
69207ff6 | 483 | /* Now need to open array the device. Use create_mddev */ |
98dbd966 DW |
484 | if (content == &info) |
485 | st->ss->getinfo_super(st, content); | |
3d2c4fc7 | 486 | |
69207ff6 | 487 | trustworthy = FOREIGN; |
745f72f6 N |
488 | switch (st->ss->match_home(st, homehost) |
489 | ?: st->ss->match_home(st, "any")) { | |
69207ff6 N |
490 | case 0: |
491 | trustworthy = FOREIGN; | |
98dbd966 | 492 | name = content->name; |
69207ff6 N |
493 | break; |
494 | case 1: | |
495 | trustworthy = LOCAL; | |
98dbd966 | 496 | name = strchr(content->name, ':'); |
69207ff6 N |
497 | if (name) |
498 | name++; | |
3d2c4fc7 | 499 | else |
98dbd966 | 500 | name = content->name; |
69207ff6 N |
501 | break; |
502 | case -1: | |
a4bc1720 | 503 | trustworthy = FOREIGN; |
69207ff6 | 504 | break; |
c30e5369 | 505 | } |
ac2ecf55 | 506 | if (!auto_assem && trustworthy == FOREIGN) |
aa7c284c | 507 | /* If the array is listed in mdadm.conf or on |
ac2ecf55 N |
508 | * command line, then we trust the name |
509 | * even if the array doesn't look local | |
510 | */ | |
511 | trustworthy = LOCAL; | |
512 | ||
98dbd966 DW |
513 | if (content->name[0] == 0 && |
514 | content->array.level == LEVEL_CONTAINER) { | |
515 | name = content->text_version; | |
a4bc1720 N |
516 | trustworthy = METADATA; |
517 | } | |
a04d5763 N |
518 | mdfd = create_mddev(mddev, name, ident->autof, trustworthy, |
519 | chosen_name); | |
c30e5369 N |
520 | if (mdfd < 0) { |
521 | st->ss->free_super(st); | |
522 | free(devices); | |
523 | if (auto_assem) | |
60e1bc1a | 524 | goto try_again; |
c30e5369 N |
525 | return 1; |
526 | } | |
a04d5763 | 527 | mddev = chosen_name; |
c30e5369 N |
528 | vers = md_get_version(mdfd); |
529 | if (vers < 9000) { | |
530 | fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n" | |
531 | " Upgrade your kernel or try --build\n"); | |
532 | close(mdfd); | |
533 | return 1; | |
534 | } | |
66afdfa9 | 535 | if (mddev_busy(fd2devnum(mdfd))) { |
c30e5369 N |
536 | fprintf(stderr, Name ": %s already active, cannot restart it!\n", |
537 | mddev); | |
538 | for (tmpdev = devlist ; | |
539 | tmpdev && tmpdev->used != 1; | |
540 | tmpdev = tmpdev->next) | |
541 | ; | |
542 | if (tmpdev && auto_assem) | |
543 | fprintf(stderr, Name ": %s needed for %s...\n", | |
544 | mddev, tmpdev->devname); | |
545 | close(mdfd); | |
546 | mdfd = -3; | |
547 | st->ss->free_super(st); | |
548 | free(devices); | |
549 | if (auto_assem) | |
550 | goto try_again; | |
551 | return 1; | |
8a46fe84 | 552 | } |
c30e5369 | 553 | ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */ |
8a46fe84 | 554 | |
9008ed1c N |
555 | #ifndef MDASSEMBLE |
556 | if (content != &info) { | |
557 | /* This is a member of a container. Try starting the array. */ | |
558 | return assemble_container_content(st, mdfd, content, runstop, | |
559 | chosen_name, verbose); | |
560 | } | |
561 | #endif | |
811e6cbe | 562 | /* Ok, no bad inconsistancy, we can try updating etc */ |
bf4fb153 | 563 | bitmap_done = 0; |
da6b5ca9 | 564 | for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) { |
811e6cbe NB |
565 | char *devname = tmpdev->devname; |
566 | struct stat stb; | |
5787fa49 | 567 | /* looks like a good enough match to update the super block if needed */ |
d1f1011b | 568 | #ifndef MDASSEMBLE |
5787fa49 | 569 | if (update) { |
811e6cbe | 570 | int dfd; |
4b1ac34b NB |
571 | /* prepare useful information in info structures */ |
572 | struct stat stb2; | |
3da92f27 | 573 | struct supertype *tst; |
4b1ac34b | 574 | fstat(mdfd, &stb2); |
7d99579f NB |
575 | |
576 | if (strcmp(update, "uuid")==0 && | |
577 | !ident->uuid_set) { | |
578 | int rfd; | |
579 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
580 | read(rfd, ident->uuid, 16) != 16) { | |
581 | *(__u32*)(ident->uuid) = random(); | |
582 | *(__u32*)(ident->uuid+1) = random(); | |
583 | *(__u32*)(ident->uuid+2) = random(); | |
584 | *(__u32*)(ident->uuid+3) = random(); | |
585 | } | |
586 | if (rfd >= 0) close(rfd); | |
7d99579f | 587 | } |
811e6cbe NB |
588 | dfd = dev_open(devname, O_RDWR|O_EXCL); |
589 | ||
0430ed48 NB |
590 | remove_partitions(dfd); |
591 | ||
3da92f27 NB |
592 | tst = dup_super(st); |
593 | tst->ss->load_super(tst, dfd, NULL); | |
98dbd966 | 594 | tst->ss->getinfo_super(tst, content); |
811e6cbe | 595 | |
98dbd966 DW |
596 | memcpy(content->uuid, ident->uuid, 16); |
597 | strcpy(content->name, ident->name); | |
598 | content->array.md_minor = minor(stb2.st_rdev); | |
811e6cbe | 599 | |
98dbd966 | 600 | tst->ss->update_super(tst, content, update, |
3da92f27 NB |
601 | devname, verbose, |
602 | ident->uuid_set, homehost); | |
e5eac01f NB |
603 | if (strcmp(update, "uuid")==0 && |
604 | !ident->uuid_set) { | |
605 | ident->uuid_set = 1; | |
98dbd966 | 606 | memcpy(ident->uuid, content->uuid, 16); |
e5eac01f | 607 | } |
b8a8ccf9 | 608 | if (dfd < 0) |
5787fa49 NB |
609 | fprintf(stderr, Name ": Cannot open %s for superblock update\n", |
610 | devname); | |
3da92f27 | 611 | else if (tst->ss->store_super(tst, dfd)) |
5787fa49 NB |
612 | fprintf(stderr, Name ": Could not re-write superblock on %s.\n", |
613 | devname); | |
614 | if (dfd >= 0) | |
615 | close(dfd); | |
8131b493 NB |
616 | |
617 | if (strcmp(update, "uuid")==0 && | |
bf4fb153 | 618 | ident->bitmap_fd >= 0 && !bitmap_done) { |
3da92f27 | 619 | if (bitmap_update_uuid(ident->bitmap_fd, |
98dbd966 | 620 | content->uuid, |
3da92f27 | 621 | tst->ss->swapuuid) != 0) |
bf4fb153 NB |
622 | fprintf(stderr, Name ": Could not update uuid on external bitmap.\n"); |
623 | else | |
624 | bitmap_done = 1; | |
625 | } | |
3da92f27 | 626 | tst->ss->free_super(tst); |
d1f1011b NB |
627 | } else |
628 | #endif | |
629 | { | |
30e1b9a5 | 630 | struct supertype *tst = dup_super(st); |
da6b5ca9 NB |
631 | int dfd; |
632 | dfd = dev_open(devname, O_RDWR|O_EXCL); | |
633 | ||
0430ed48 NB |
634 | remove_partitions(dfd); |
635 | ||
3da92f27 | 636 | tst->ss->load_super(tst, dfd, NULL); |
98dbd966 | 637 | tst->ss->getinfo_super(tst, content); |
3da92f27 | 638 | tst->ss->free_super(tst); |
da6b5ca9 | 639 | close(dfd); |
5787fa49 NB |
640 | } |
641 | ||
811e6cbe NB |
642 | stat(devname, &stb); |
643 | ||
dab6685f | 644 | if (verbose > 0) |
cd29a5c8 | 645 | fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n", |
98dbd966 | 646 | devname, mddev, content->disk.raid_disk); |
64c4757e | 647 | devices[devcnt].devname = devname; |
64c4757e | 648 | devices[devcnt].uptodate = 0; |
98dbd966 | 649 | devices[devcnt].i = *content; |
213ee40b NB |
650 | devices[devcnt].i.disk.major = major(stb.st_rdev); |
651 | devices[devcnt].i.disk.minor = minor(stb.st_rdev); | |
64c4757e | 652 | if (most_recent < devcnt) { |
213ee40b NB |
653 | if (devices[devcnt].i.events |
654 | > devices[most_recent].i.events) | |
64c4757e NB |
655 | most_recent = devcnt; |
656 | } | |
98dbd966 | 657 | if (content->array.level == -4) |
5787fa49 NB |
658 | /* with multipath, the raid_disk from the superblock is meaningless */ |
659 | i = devcnt; | |
660 | else | |
213ee40b | 661 | i = devices[devcnt].i.disk.raid_disk; |
308e1801 | 662 | if (i+1 == 0) { |
98dbd966 DW |
663 | if (nextspare < content->array.raid_disks) |
664 | nextspare = content->array.raid_disks; | |
308e1801 | 665 | i = nextspare++; |
08110d41 | 666 | } else { |
98dbd966 | 667 | if (i >= content->array.raid_disks && |
08110d41 NB |
668 | i >= nextspare) |
669 | nextspare = i+1; | |
308e1801 | 670 | } |
98c6faba | 671 | if (i < 10000) { |
56eedc1a | 672 | if (i >= bestcnt) { |
98c6faba | 673 | unsigned int newbestcnt = i+10; |
56eedc1a | 674 | int *newbest = malloc(sizeof(int)*newbestcnt); |
98c6faba | 675 | unsigned int c; |
56eedc1a NB |
676 | for (c=0; c < newbestcnt; c++) |
677 | if (c < bestcnt) | |
678 | newbest[c] = best[c]; | |
679 | else | |
680 | newbest[c] = -1; | |
681 | if (best)free(best); | |
682 | best = newbest; | |
683 | bestcnt = newbestcnt; | |
684 | } | |
6a255b69 | 685 | if (best[i] >=0 && |
213ee40b NB |
686 | devices[best[i]].i.events |
687 | == devices[devcnt].i.events | |
688 | && (devices[best[i]].i.disk.minor | |
689 | != devices[devcnt].i.disk.minor) | |
b8ac1967 | 690 | && st->ss == &super0 |
98dbd966 | 691 | && content->array.level != LEVEL_MULTIPATH) { |
6a255b69 NB |
692 | /* two different devices with identical superblock. |
693 | * Could be a mis-detection caused by overlapping | |
694 | * partitions. fail-safe. | |
695 | */ | |
696 | fprintf(stderr, Name ": WARNING %s and %s appear" | |
697 | " to have very similar superblocks.\n" | |
698 | " If they are really different, " | |
699 | "please --zero the superblock on one\n" | |
d2df86e0 NB |
700 | " If they are the same or overlap," |
701 | " please remove one from %s.\n", | |
702 | devices[best[i]].devname, devname, | |
703 | inargv ? "the list" : | |
704 | "the\n DEVICE list in mdadm.conf" | |
705 | ); | |
7f91af49 | 706 | close(mdfd); |
6a255b69 NB |
707 | return 1; |
708 | } | |
52826846 | 709 | if (best[i] == -1 |
213ee40b NB |
710 | || (devices[best[i]].i.events |
711 | < devices[devcnt].i.events)) | |
52826846 | 712 | best[i] = devcnt; |
56eedc1a | 713 | } |
64c4757e | 714 | devcnt++; |
df37ffc0 | 715 | } |
4b1ac34b | 716 | |
64c4757e | 717 | if (devcnt == 0) { |
682c7051 | 718 | fprintf(stderr, Name ": no devices found for %s\n", |
64c4757e | 719 | mddev); |
3d2b16e7 NB |
720 | if (st) |
721 | st->ss->free_super(st); | |
7f91af49 | 722 | close(mdfd); |
64c4757e NB |
723 | return 1; |
724 | } | |
4b1ac34b | 725 | |
3d2b16e7 NB |
726 | if (update && strcmp(update, "byteorder")==0) |
727 | st->minor_version = 90; | |
728 | ||
98dbd966 DW |
729 | st->ss->getinfo_super(st, content); |
730 | clean = content->array.state & 1; | |
4b1ac34b | 731 | |
64c4757e NB |
732 | /* now we have some devices that might be suitable. |
733 | * I wonder how many | |
734 | */ | |
98dbd966 DW |
735 | avail = malloc(content->array.raid_disks); |
736 | memset(avail, 0, content->array.raid_disks); | |
64c4757e | 737 | okcnt = 0; |
52826846 | 738 | sparecnt=0; |
56eedc1a | 739 | for (i=0; i< bestcnt ;i++) { |
64c4757e | 740 | int j = best[i]; |
ee04451c NB |
741 | int event_margin = 1; /* always allow a difference of '1' |
742 | * like the kernel does | |
743 | */ | |
64c4757e | 744 | if (j < 0) continue; |
d013a55e NB |
745 | /* note: we ignore error flags in multipath arrays |
746 | * as they don't make sense | |
747 | */ | |
98dbd966 | 748 | if (content->array.level != -4) |
213ee40b NB |
749 | if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) { |
750 | if (!(devices[j].i.disk.state | |
751 | & (1<<MD_DISK_FAULTY))) | |
aa88f531 | 752 | sparecnt++; |
d013a55e | 753 | continue; |
aa88f531 | 754 | } |
213ee40b NB |
755 | if (devices[j].i.events+event_margin >= |
756 | devices[most_recent].i.events) { | |
64c4757e | 757 | devices[j].uptodate = 1; |
98dbd966 | 758 | if (i < content->array.raid_disks) { |
52826846 | 759 | okcnt++; |
265e0f17 NB |
760 | avail[i]=1; |
761 | } else | |
52826846 | 762 | sparecnt++; |
64c4757e NB |
763 | } |
764 | } | |
98dbd966 DW |
765 | while (force && !enough(content->array.level, content->array.raid_disks, |
766 | content->array.layout, 1, | |
265e0f17 | 767 | avail, okcnt)) { |
64c4757e NB |
768 | /* Choose the newest best drive which is |
769 | * not up-to-date, update the superblock | |
770 | * and add it. | |
771 | */ | |
52826846 | 772 | int fd; |
3da92f27 | 773 | struct supertype *tst; |
c5a6f9a6 | 774 | long long current_events; |
cd29a5c8 | 775 | chosen_drive = -1; |
98dbd966 | 776 | for (i=0; i<content->array.raid_disks && i < bestcnt; i++) { |
52826846 NB |
777 | int j = best[i]; |
778 | if (j>=0 && | |
779 | !devices[j].uptodate && | |
52826846 | 780 | (chosen_drive < 0 || |
213ee40b NB |
781 | devices[j].i.events |
782 | > devices[chosen_drive].i.events)) | |
52826846 NB |
783 | chosen_drive = j; |
784 | } | |
785 | if (chosen_drive < 0) | |
786 | break; | |
213ee40b | 787 | current_events = devices[chosen_drive].i.events; |
c5a6f9a6 | 788 | add_another: |
dab6685f NB |
789 | if (verbose >= 0) |
790 | fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n", | |
213ee40b NB |
791 | devices[chosen_drive].devname, |
792 | devices[chosen_drive].i.disk.raid_disk, | |
793 | (int)(devices[chosen_drive].i.events), | |
794 | (int)(devices[most_recent].i.events)); | |
8b0dabea | 795 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 NB |
796 | if (fd < 0) { |
797 | fprintf(stderr, Name ": Couldn't open %s for write - not updating\n", | |
798 | devices[chosen_drive].devname); | |
213ee40b | 799 | devices[chosen_drive].i.events = 0; |
52826846 NB |
800 | continue; |
801 | } | |
3da92f27 | 802 | tst = dup_super(st); |
60b435db | 803 | if (tst->ss->load_super(tst,fd, NULL)) { |
52826846 NB |
804 | close(fd); |
805 | fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n", | |
806 | devices[chosen_drive].devname); | |
213ee40b | 807 | devices[chosen_drive].i.events = 0; |
52826846 NB |
808 | continue; |
809 | } | |
98dbd966 DW |
810 | content->events = devices[most_recent].i.events; |
811 | tst->ss->update_super(tst, content, "force-one", | |
67a8c82d NB |
812 | devices[chosen_drive].devname, verbose, |
813 | 0, NULL); | |
4b1ac34b | 814 | |
3da92f27 | 815 | if (tst->ss->store_super(tst, fd)) { |
52826846 NB |
816 | close(fd); |
817 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
818 | devices[chosen_drive].devname); | |
213ee40b | 819 | devices[chosen_drive].i.events = 0; |
3da92f27 | 820 | tst->ss->free_super(tst); |
52826846 NB |
821 | continue; |
822 | } | |
823 | close(fd); | |
213ee40b | 824 | devices[chosen_drive].i.events = devices[most_recent].i.events; |
52826846 | 825 | devices[chosen_drive].uptodate = 1; |
265e0f17 | 826 | avail[chosen_drive] = 1; |
52826846 | 827 | okcnt++; |
3da92f27 | 828 | tst->ss->free_super(tst); |
c5a6f9a6 NB |
829 | |
830 | /* If there are any other drives of the same vintage, | |
831 | * add them in as well. We can't lose and we might gain | |
832 | */ | |
98dbd966 | 833 | for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) { |
c5a6f9a6 NB |
834 | int j = best[i]; |
835 | if (j >= 0 && | |
836 | !devices[j].uptodate && | |
213ee40b | 837 | devices[j].i.events == current_events) { |
c5a6f9a6 NB |
838 | chosen_drive = j; |
839 | goto add_another; | |
840 | } | |
841 | } | |
64c4757e | 842 | } |
52826846 NB |
843 | |
844 | /* Now we want to look at the superblock which the kernel will base things on | |
845 | * and compare the devices that we think are working with the devices that the | |
846 | * superblock thinks are working. | |
847 | * If there are differences and --force is given, then update this chosen | |
848 | * superblock. | |
849 | */ | |
cd29a5c8 | 850 | chosen_drive = -1; |
3da92f27 | 851 | st->ss->free_super(st); |
56eedc1a | 852 | for (i=0; chosen_drive < 0 && i<bestcnt; i++) { |
52826846 NB |
853 | int j = best[i]; |
854 | int fd; | |
4b1ac34b | 855 | |
52826846 NB |
856 | if (j<0) |
857 | continue; | |
858 | if (!devices[j].uptodate) | |
859 | continue; | |
860 | chosen_drive = j; | |
8b0dabea | 861 | if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) { |
52826846 NB |
862 | fprintf(stderr, Name ": Cannot open %s: %s\n", |
863 | devices[j].devname, strerror(errno)); | |
7f91af49 | 864 | close(mdfd); |
52826846 NB |
865 | return 1; |
866 | } | |
3da92f27 | 867 | if (st->ss->load_super(st,fd, NULL)) { |
52826846 NB |
868 | close(fd); |
869 | fprintf(stderr, Name ": RAID superblock has disappeared from %s\n", | |
870 | devices[j].devname); | |
7f91af49 | 871 | close(mdfd); |
52826846 NB |
872 | return 1; |
873 | } | |
874 | close(fd); | |
875 | } | |
3da92f27 | 876 | if (st->sb == NULL) { |
4b1ac34b | 877 | fprintf(stderr, Name ": No suitable drives found for %s\n", mddev); |
7f91af49 | 878 | close(mdfd); |
4b1ac34b NB |
879 | return 1; |
880 | } | |
98dbd966 | 881 | st->ss->getinfo_super(st, content); |
f35f2525 | 882 | #ifndef MDASSEMBLE |
98dbd966 | 883 | sysfs_init(content, mdfd, 0); |
f35f2525 | 884 | #endif |
56eedc1a | 885 | for (i=0; i<bestcnt; i++) { |
52826846 | 886 | int j = best[i]; |
98c6faba | 887 | unsigned int desired_state; |
11a3e71d | 888 | |
98dbd966 | 889 | if (i < content->array.raid_disks) |
11a3e71d NB |
890 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); |
891 | else | |
892 | desired_state = 0; | |
893 | ||
52826846 NB |
894 | if (j<0) |
895 | continue; | |
896 | if (!devices[j].uptodate) | |
897 | continue; | |
4b1ac34b | 898 | |
213ee40b | 899 | devices[j].i.disk.state = desired_state; |
4e9a6ff7 N |
900 | if (!(devices[j].i.array.state & 1)) |
901 | clean = 0; | |
213ee40b NB |
902 | |
903 | if (st->ss->update_super(st, &devices[j].i, "assemble", NULL, | |
68c7d6d7 | 904 | verbose, 0, NULL)) { |
52826846 | 905 | if (force) { |
dab6685f NB |
906 | if (verbose >= 0) |
907 | fprintf(stderr, Name ": " | |
908 | "clearing FAULTY flag for device %d in %s for %s\n", | |
909 | j, mddev, devices[j].devname); | |
4b1ac34b | 910 | change = 1; |
52826846 | 911 | } else { |
dab6685f NB |
912 | if (verbose >= -1) |
913 | fprintf(stderr, Name ": " | |
914 | "device %d in %s has wrong state in superblock, but %s seems ok\n", | |
915 | i, mddev, devices[j].devname); | |
52826846 NB |
916 | } |
917 | } | |
4b1ac34b | 918 | #if 0 |
213ee40b | 919 | if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) { |
52826846 NB |
920 | fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", |
921 | i, mddev); | |
922 | } | |
4b1ac34b | 923 | #endif |
52826846 | 924 | } |
c5a6f9a6 | 925 | if (force && !clean && |
98dbd966 DW |
926 | !enough(content->array.level, content->array.raid_disks, |
927 | content->array.layout, clean, | |
c5a6f9a6 | 928 | avail, okcnt)) { |
98dbd966 | 929 | change += st->ss->update_super(st, content, "force-array", |
67a8c82d NB |
930 | devices[chosen_drive].devname, verbose, |
931 | 0, NULL); | |
583315d9 | 932 | clean = 1; |
d013a55e | 933 | } |
52826846 | 934 | |
4b1ac34b | 935 | if (change) { |
52826846 | 936 | int fd; |
8b0dabea | 937 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 | 938 | if (fd < 0) { |
353632d9 | 939 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", |
52826846 | 940 | devices[chosen_drive].devname); |
7f91af49 | 941 | close(mdfd); |
52826846 NB |
942 | return 1; |
943 | } | |
3da92f27 | 944 | if (st->ss->store_super(st, fd)) { |
52826846 NB |
945 | close(fd); |
946 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
947 | devices[chosen_drive].devname); | |
7f91af49 | 948 | close(mdfd); |
52826846 NB |
949 | return 1; |
950 | } | |
951 | close(fd); | |
52826846 NB |
952 | } |
953 | ||
353632d9 NB |
954 | /* If we are in the middle of a reshape we may need to restore saved data |
955 | * that was moved aside due to the reshape overwriting live data | |
956 | * The code of doing this lives in Grow.c | |
957 | */ | |
39c5a909 | 958 | #ifndef MDASSEMBLE |
98dbd966 | 959 | if (content->reshape_active) { |
353632d9 NB |
960 | int err = 0; |
961 | int *fdlist = malloc(sizeof(int)* bestcnt); | |
962 | for (i=0; i<bestcnt; i++) { | |
963 | int j = best[i]; | |
964 | if (j >= 0) { | |
965 | fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
966 | if (fdlist[i] < 0) { | |
967 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", | |
968 | devices[j].devname); | |
969 | err = 1; | |
970 | break; | |
971 | } | |
972 | } else | |
973 | fdlist[i] = -1; | |
974 | } | |
975 | if (!err) | |
98dbd966 | 976 | err = Grow_restart(st, content, fdlist, bestcnt, backup_file); |
353632d9 NB |
977 | while (i>0) { |
978 | i--; | |
979 | if (fdlist[i]>=0) close(fdlist[i]); | |
980 | } | |
981 | if (err) { | |
982 | fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n"); | |
7f91af49 | 983 | close(mdfd); |
353632d9 NB |
984 | return err; |
985 | } | |
986 | } | |
39c5a909 | 987 | #endif |
aa88f531 NB |
988 | /* count number of in-sync devices according to the superblock. |
989 | * We must have this number to start the array without -s or -R | |
990 | */ | |
98dbd966 | 991 | req_cnt = content->array.working_disks; |
aa88f531 | 992 | |
64c4757e NB |
993 | /* Almost ready to actually *do* something */ |
994 | if (!old_linux) { | |
fbf8a0b7 | 995 | int rv; |
a19c88b8 | 996 | |
a04d5763 N |
997 | /* First, fill in the map, so that udev can find our name |
998 | * as soon as we become active. | |
999 | */ | |
98dbd966 DW |
1000 | map_update(NULL, fd2devnum(mdfd), content->text_version, |
1001 | content->uuid, chosen_name); | |
a04d5763 | 1002 | |
98dbd966 | 1003 | rv = set_array_info(mdfd, st, content); |
fbf8a0b7 | 1004 | if (rv) { |
f35f2525 | 1005 | fprintf(stderr, Name ": failed to set array info for %s: %s\n", |
64c4757e | 1006 | mddev, strerror(errno)); |
7f91af49 | 1007 | close(mdfd); |
64c4757e NB |
1008 | return 1; |
1009 | } | |
11484a63 | 1010 | if (ident->bitmap_fd >= 0) { |
c82f047c NB |
1011 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { |
1012 | fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n"); | |
7f91af49 | 1013 | close(mdfd); |
c82f047c NB |
1014 | return 1; |
1015 | } | |
7ef02d01 NB |
1016 | } else if (ident->bitmap_file) { |
1017 | /* From config file */ | |
1018 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
1019 | if (bmfd < 0) { | |
1020 | fprintf(stderr, Name ": Could not open bitmap file %s\n", | |
1021 | ident->bitmap_file); | |
7f91af49 | 1022 | close(mdfd); |
7ef02d01 NB |
1023 | return 1; |
1024 | } | |
1025 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
1026 | fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev); | |
1027 | close(bmfd); | |
7f91af49 | 1028 | close(mdfd); |
7ef02d01 NB |
1029 | return 1; |
1030 | } | |
1031 | close(bmfd); | |
c82f047c | 1032 | } |
7ef02d01 | 1033 | |
52826846 | 1034 | /* First, add the raid disks, but add the chosen one last */ |
56eedc1a | 1035 | for (i=0; i<= bestcnt; i++) { |
52826846 | 1036 | int j; |
56eedc1a | 1037 | if (i < bestcnt) { |
52826846 NB |
1038 | j = best[i]; |
1039 | if (j == chosen_drive) | |
1040 | continue; | |
1041 | } else | |
1042 | j = chosen_drive; | |
1043 | ||
56eedc1a | 1044 | if (j >= 0 /* && devices[j].uptodate */) { |
98dbd966 | 1045 | rv = add_disk(mdfd, st, content, &devices[j].i); |
7801ac20 | 1046 | |
a19c88b8 | 1047 | if (rv) { |
213ee40b NB |
1048 | fprintf(stderr, Name ": failed to add " |
1049 | "%s to %s: %s\n", | |
64c4757e NB |
1050 | devices[j].devname, |
1051 | mddev, | |
1052 | strerror(errno)); | |
98dbd966 | 1053 | if (i < content->array.raid_disks |
213ee40b | 1054 | || i == bestcnt) |
52826846 NB |
1055 | okcnt--; |
1056 | else | |
1057 | sparecnt--; | |
dab6685f | 1058 | } else if (verbose > 0) |
213ee40b NB |
1059 | fprintf(stderr, Name ": added %s " |
1060 | "to %s as %d\n", | |
1061 | devices[j].devname, mddev, | |
1062 | devices[j].i.disk.raid_disk); | |
98dbd966 | 1063 | } else if (verbose > 0 && i < content->array.raid_disks) |
213ee40b NB |
1064 | fprintf(stderr, Name ": no uptodate device for " |
1065 | "slot %d of %s\n", | |
64c4757e NB |
1066 | i, mddev); |
1067 | } | |
aba69144 | 1068 | |
98dbd966 | 1069 | if (content->array.level == LEVEL_CONTAINER) { |
598f0d58 NB |
1070 | if (verbose >= 0) { |
1071 | fprintf(stderr, Name ": Container %s has been " | |
1072 | "assembled with %d drive%s", | |
57ed8c91 | 1073 | mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s"); |
98dbd966 | 1074 | if (okcnt < content->array.raid_disks) |
598f0d58 | 1075 | fprintf(stderr, " (out of %d)", |
98dbd966 | 1076 | content->array.raid_disks); |
598f0d58 NB |
1077 | fprintf(stderr, "\n"); |
1078 | } | |
98dbd966 | 1079 | sysfs_uevent(content, "change"); |
a7c6e3fb | 1080 | wait_for(chosen_name, mdfd); |
7f91af49 | 1081 | close(mdfd); |
598f0d58 NB |
1082 | return 0; |
1083 | } | |
1084 | ||
64c4757e | 1085 | if (runstop == 1 || |
b8a8ccf9 | 1086 | (runstop <= 0 && |
98dbd966 DW |
1087 | ( enough(content->array.level, content->array.raid_disks, |
1088 | content->array.layout, clean, avail, okcnt) && | |
aa88f531 NB |
1089 | (okcnt >= req_cnt || start_partial_ok) |
1090 | ))) { | |
82b27616 | 1091 | if (ioctl(mdfd, RUN_ARRAY, NULL)==0) { |
dab6685f NB |
1092 | if (verbose >= 0) { |
1093 | fprintf(stderr, Name ": %s has been started with %d drive%s", | |
1094 | mddev, okcnt, okcnt==1?"":"s"); | |
98dbd966 DW |
1095 | if (okcnt < content->array.raid_disks) |
1096 | fprintf(stderr, " (out of %d)", content->array.raid_disks); | |
dab6685f NB |
1097 | if (sparecnt) |
1098 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
1099 | fprintf(stderr, ".\n"); | |
1100 | } | |
8a659c33 N |
1101 | if (content->reshape_active && |
1102 | content->array.level >= 4 && | |
1103 | content->array.level <= 6) { | |
acee8e89 N |
1104 | /* might need to increase the size |
1105 | * of the stripe cache - default is 256 | |
1106 | */ | |
8a659c33 | 1107 | if (256 < 4 * (content->array.chunk_size/4096)) { |
acee8e89 N |
1108 | struct mdinfo *sra = sysfs_read(mdfd, 0, 0); |
1109 | if (sra) | |
1110 | sysfs_set_num(sra, NULL, | |
1111 | "stripe_cache_size", | |
8a659c33 | 1112 | (4 * content->array.chunk_size / 4096) + 1); |
acee8e89 N |
1113 | } |
1114 | } | |
a7c6e3fb | 1115 | wait_for(mddev, mdfd); |
7f91af49 N |
1116 | close(mdfd); |
1117 | if (auto_assem) { | |
589395d6 | 1118 | int usecs = 1; |
589395d6 NB |
1119 | /* There is a nasty race with 'mdadm --monitor'. |
1120 | * If it opens this device before we close it, | |
1121 | * it gets an incomplete open on which IO | |
598f0d58 NB |
1122 | * doesn't work and the capacity is |
1123 | * wrong. | |
589395d6 NB |
1124 | * If we reopen (to check for layered devices) |
1125 | * before --monitor closes, we loose. | |
1126 | * | |
1127 | * So: wait upto 1 second for there to be | |
1128 | * a non-zero capacity. | |
1129 | */ | |
1130 | while (usecs < 1000) { | |
1131 | mdfd = open(mddev, O_RDONLY); | |
1132 | if (mdfd >= 0) { | |
beae1dfe NB |
1133 | unsigned long long size; |
1134 | if (get_dev_size(mdfd, NULL, &size) && | |
589395d6 NB |
1135 | size > 0) |
1136 | break; | |
1137 | close(mdfd); | |
1138 | } | |
1139 | usleep(usecs); | |
1140 | usecs <<= 1; | |
1141 | } | |
1142 | } | |
64c4757e | 1143 | return 0; |
82b27616 | 1144 | } |
682c7051 | 1145 | fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n", |
64c4757e | 1146 | mddev, strerror(errno)); |
583315d9 | 1147 | |
98dbd966 DW |
1148 | if (!enough(content->array.level, content->array.raid_disks, |
1149 | content->array.layout, 1, avail, okcnt)) | |
583315d9 NB |
1150 | fprintf(stderr, Name ": Not enough devices to " |
1151 | "start the array.\n"); | |
98dbd966 DW |
1152 | else if (!enough(content->array.level, |
1153 | content->array.raid_disks, | |
1154 | content->array.layout, clean, | |
583315d9 NB |
1155 | avail, okcnt)) |
1156 | fprintf(stderr, Name ": Not enough devices to " | |
1157 | "start the array while not clean " | |
1158 | "- consider --force.\n"); | |
1159 | ||
7f91af49 | 1160 | if (auto_assem) |
1c203a4b | 1161 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1162 | close(mdfd); |
64c4757e NB |
1163 | return 1; |
1164 | } | |
82b27616 | 1165 | if (runstop == -1) { |
b8a8ccf9 | 1166 | fprintf(stderr, Name ": %s assembled from %d drive%s", |
52826846 | 1167 | mddev, okcnt, okcnt==1?"":"s"); |
98dbd966 DW |
1168 | if (okcnt != content->array.raid_disks) |
1169 | fprintf(stderr, " (out of %d)", content->array.raid_disks); | |
b8a8ccf9 | 1170 | fprintf(stderr, ", but not started.\n"); |
7f91af49 | 1171 | close(mdfd); |
64c4757e | 1172 | return 0; |
82b27616 | 1173 | } |
4855f95c | 1174 | if (verbose >= -1) { |
dab6685f NB |
1175 | fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s"); |
1176 | if (sparecnt) | |
1177 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
98dbd966 DW |
1178 | if (!enough(content->array.level, content->array.raid_disks, |
1179 | content->array.layout, 1, avail, okcnt)) | |
dab6685f | 1180 | fprintf(stderr, " - not enough to start the array.\n"); |
98dbd966 DW |
1181 | else if (!enough(content->array.level, |
1182 | content->array.raid_disks, | |
1183 | content->array.layout, clean, | |
583315d9 NB |
1184 | avail, okcnt)) |
1185 | fprintf(stderr, " - not enough to start the " | |
1186 | "array while not clean - consider " | |
1187 | "--force.\n"); | |
dab6685f | 1188 | else { |
98dbd966 | 1189 | if (req_cnt == content->array.raid_disks) |
dab6685f NB |
1190 | fprintf(stderr, " - need all %d to start it", req_cnt); |
1191 | else | |
98dbd966 | 1192 | fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks); |
dab6685f NB |
1193 | fprintf(stderr, " (use --run to insist).\n"); |
1194 | } | |
aa88f531 | 1195 | } |
7f91af49 | 1196 | if (auto_assem) |
1c203a4b | 1197 | ioctl(mdfd, STOP_ARRAY, NULL); |
56a8da69 | 1198 | close(mdfd); |
64c4757e | 1199 | return 1; |
82b27616 | 1200 | } else { |
52826846 NB |
1201 | /* The "chosen_drive" is a good choice, and if necessary, the superblock has |
1202 | * been updated to point to the current locations of devices. | |
1203 | * so we can just start the array | |
82b27616 | 1204 | */ |
cd29a5c8 | 1205 | unsigned long dev; |
213ee40b NB |
1206 | dev = makedev(devices[chosen_drive].i.disk.major, |
1207 | devices[chosen_drive].i.disk.minor); | |
82b27616 NB |
1208 | if (ioctl(mdfd, START_ARRAY, dev)) { |
1209 | fprintf(stderr, Name ": Cannot start array: %s\n", | |
1210 | strerror(errno)); | |
1211 | } | |
aba69144 | 1212 | |
64c4757e | 1213 | } |
7f91af49 | 1214 | close(mdfd); |
e0d19036 | 1215 | return 0; |
64c4757e | 1216 | } |
6234c63c DW |
1217 | |
1218 | #ifndef MDASSEMBLE | |
1219 | int assemble_container_content(struct supertype *st, int mdfd, | |
1220 | struct mdinfo *content, int runstop, | |
1221 | char *chosen_name, int verbose) | |
1222 | { | |
1223 | struct mdinfo *dev, *sra; | |
1224 | int working = 0, preexist = 0; | |
1225 | struct map_ent *map = NULL; | |
1226 | ||
1227 | sysfs_init(content, mdfd, 0); | |
1228 | ||
1229 | sra = sysfs_read(mdfd, 0, GET_VERSION); | |
1230 | if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) | |
4408ee76 N |
1231 | if (sysfs_set_array(content, md_get_version(mdfd)) != 0) { |
1232 | close(mdfd); | |
6234c63c | 1233 | return 1; |
4408ee76 | 1234 | } |
6234c63c DW |
1235 | if (sra) |
1236 | sysfs_free(sra); | |
1237 | ||
1238 | for (dev = content->devs; dev; dev = dev->next) | |
462906cd | 1239 | if (sysfs_add_disk(content, dev, 1) == 0) |
6234c63c DW |
1240 | working++; |
1241 | else if (errno == EEXIST) | |
1242 | preexist++; | |
4408ee76 N |
1243 | if (working == 0) { |
1244 | close(mdfd); | |
7cb2aa33 | 1245 | return 1;/* Nothing new, don't try to start */ |
8b4e5ea9 N |
1246 | } |
1247 | ||
1248 | map_update(&map, fd2devnum(mdfd), | |
1249 | content->text_version, | |
1250 | content->uuid, chosen_name); | |
1251 | ||
1252 | if (runstop > 0 || | |
6234c63c | 1253 | (working + preexist) >= content->array.working_disks) { |
bb50e5d3 | 1254 | int err; |
a714580e | 1255 | |
6234c63c DW |
1256 | switch(content->array.level) { |
1257 | case LEVEL_LINEAR: | |
1258 | case LEVEL_MULTIPATH: | |
1259 | case 0: | |
bb50e5d3 N |
1260 | err = sysfs_set_str(content, NULL, "array_state", |
1261 | "active"); | |
6234c63c DW |
1262 | break; |
1263 | default: | |
bb50e5d3 | 1264 | err = sysfs_set_str(content, NULL, "array_state", |
6234c63c DW |
1265 | "readonly"); |
1266 | /* start mdmon if needed. */ | |
bb50e5d3 N |
1267 | if (!err) { |
1268 | if (!mdmon_running(st->container_dev)) | |
1269 | start_mdmon(st->container_dev); | |
1270 | ping_monitor(devnum2devname(st->container_dev)); | |
1271 | } | |
6234c63c DW |
1272 | break; |
1273 | } | |
bb50e5d3 N |
1274 | if (!err) |
1275 | sysfs_set_safemode(content, content->safe_mode_delay); | |
6234c63c | 1276 | if (verbose >= 0) { |
bb50e5d3 N |
1277 | if (err) |
1278 | fprintf(stderr, Name | |
1279 | ": array %s now has %d devices", | |
1280 | chosen_name, working + preexist); | |
1281 | else | |
1282 | fprintf(stderr, Name | |
1283 | ": Started %s with %d devices", | |
1284 | chosen_name, working + preexist); | |
6234c63c DW |
1285 | if (preexist) |
1286 | fprintf(stderr, " (%d new)", working); | |
1287 | fprintf(stderr, "\n"); | |
1288 | } | |
bb50e5d3 | 1289 | if (!err) |
a7c6e3fb | 1290 | wait_for(chosen_name, mdfd); |
4408ee76 | 1291 | close(mdfd); |
7cb2aa33 | 1292 | return 0; |
6234c63c | 1293 | /* FIXME should have an O_EXCL and wait for read-auto */ |
7cb2aa33 | 1294 | } else { |
6234c63c DW |
1295 | if (verbose >= 0) |
1296 | fprintf(stderr, Name | |
1297 | ": %s assembled with %d devices but " | |
1298 | "not started\n", | |
1299 | chosen_name, working); | |
4408ee76 | 1300 | close(mdfd); |
7cb2aa33 N |
1301 | return 1; |
1302 | } | |
6234c63c DW |
1303 | } |
1304 | #endif | |
1305 |