]>
Commit | Line | Data |
---|---|---|
64c4757e | 1 | /* |
9a9dab36 | 2 | * mdadm - manage Linux "md" devices aka RAID arrays. |
64c4757e | 3 | * |
e736b623 | 4 | * Copyright (C) 2001-2009 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 | |
e736b623 | 22 | * Email: <neilb@suse.de> |
64c4757e NB |
23 | */ |
24 | ||
9a9dab36 | 25 | #include "mdadm.h" |
41a3b72a | 26 | #include <ctype.h> |
64c4757e | 27 | |
624920bb NB |
28 | static int name_matches(char *found, char *required, char *homehost) |
29 | { | |
30 | /* See if the name found matches the required name, possibly | |
31 | * prefixed with 'homehost' | |
32 | */ | |
33 | char fnd[33]; | |
34 | ||
35 | strncpy(fnd, found, 32); | |
36 | fnd[32] = 0; | |
37 | if (strcmp(found, required)==0) | |
38 | return 1; | |
39 | if (homehost) { | |
40 | int l = strlen(homehost); | |
41 | if (l < 32 && fnd[l] == ':' && | |
42 | strcmp(fnd+l+1, required)==0) | |
43 | return 1; | |
44 | } | |
45 | return 0; | |
46 | } | |
47 | ||
9008ed1c | 48 | static int is_member_busy(char *metadata_version) |
2de8884f DW |
49 | { |
50 | /* check if the given member array is active */ | |
51 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
52 | struct mdstat_ent *ent; | |
53 | int busy = 0; | |
54 | ||
55 | for (ent = mdstat; ent; ent = ent->next) { | |
56 | if (ent->metadata_version == NULL) | |
57 | continue; | |
58 | if (strncmp(ent->metadata_version, "external:", 9) != 0) | |
59 | continue; | |
60 | if (!is_subarray(&ent->metadata_version[9])) | |
61 | continue; | |
62 | /* Skip first char - it can be '/' or '-' */ | |
63 | if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) { | |
64 | busy = 1; | |
65 | break; | |
66 | } | |
67 | } | |
68 | free_mdstat(mdstat); | |
69 | ||
70 | return busy; | |
71 | } | |
72 | ||
7f91af49 | 73 | int Assemble(struct supertype *st, char *mddev, |
8aec876d | 74 | mddev_ident_t ident, |
06b0d786 | 75 | mddev_dev_t devlist, char *backup_file, |
64c4757e | 76 | int readonly, int runstop, |
0ac91628 | 77 | char *update, char *homehost, int require_homehost, |
64c4757e NB |
78 | int verbose, int force) |
79 | { | |
80 | /* | |
52826846 NB |
81 | * The task of Assemble is to find a collection of |
82 | * devices that should (according to their superblocks) | |
83 | * form an array, and to give this collection to the MD driver. | |
84 | * In Linux-2.4 and later, this involves submitting a | |
64c4757e NB |
85 | * SET_ARRAY_INFO ioctl with no arg - to prepare |
86 | * the array - and then submit a number of | |
87 | * ADD_NEW_DISK ioctls to add disks into | |
88 | * the array. Finally RUN_ARRAY might | |
89 | * be submitted to start the array. | |
90 | * | |
91 | * Much of the work of Assemble is in finding and/or | |
92 | * checking the disks to make sure they look right. | |
93 | * | |
4b1ac34b | 94 | * If mddev is not set, then scan must be set and we |
64c4757e NB |
95 | * read through the config file for dev+uuid mapping |
96 | * We recurse, setting mddev, for each device that | |
97 | * - isn't running | |
4b1ac34b | 98 | * - has a valid uuid (or any uuid if !uuidset) |
64c4757e NB |
99 | * |
100 | * If mddev is set, we try to determine state of md. | |
101 | * check version - must be at least 0.90.0 | |
102 | * check kernel version. must be at least 2.4. | |
103 | * If not, we can possibly fall back on START_ARRAY | |
104 | * Try to GET_ARRAY_INFO. | |
105 | * If possible, give up | |
106 | * If not, try to STOP_ARRAY just to make sure | |
107 | * | |
108 | * If !uuidset and scan, look in conf-file for uuid | |
109 | * If not found, give up | |
b8a8ccf9 | 110 | * If !devlist and scan and uuidset, get list of devs from conf-file |
64c4757e NB |
111 | * |
112 | * For each device: | |
113 | * Check superblock - discard if bad | |
114 | * Check uuid (set if we don't have one) - discard if no match | |
5787fa49 | 115 | * Check superblock similarity if we have a superblock - discard if different |
4b1ac34b | 116 | * Record events, devicenum |
64c4757e | 117 | * This should give us a list of devices for the array |
4b1ac34b | 118 | * We should collect the most recent event number |
64c4757e NB |
119 | * |
120 | * Count disks with recent enough event count | |
121 | * While force && !enough disks | |
122 | * Choose newest rejected disks, update event count | |
123 | * mark clean and rewrite superblock | |
124 | * If recent kernel: | |
125 | * SET_ARRAY_INFO | |
126 | * foreach device with recent events : ADD_NEW_DISK | |
127 | * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY | |
128 | * If old kernel: | |
129 | * Check the device numbers in superblock are right | |
130 | * update superblock if any changes | |
131 | * START_ARRAY | |
132 | * | |
133 | */ | |
c30e5369 N |
134 | int mdfd; |
135 | int clean; | |
f05641cf N |
136 | int auto_assem = (mddev == NULL && !ident->uuid_set && |
137 | ident->super_minor == UnSet && ident->name[0] == 0 | |
aa7c284c | 138 | && (ident->container == NULL || ident->member == NULL)); |
64c4757e | 139 | int old_linux = 0; |
c30e5369 | 140 | int vers = vers; /* Keep gcc quite - it really is initialised */ |
64c4757e NB |
141 | struct { |
142 | char *devname; | |
213ee40b NB |
143 | int uptodate; /* set once we decide that this device is as |
144 | * recent as everything else in the array. | |
145 | */ | |
146 | struct mdinfo i; | |
5787fa49 | 147 | } *devices; |
56eedc1a | 148 | int *best = NULL; /* indexed by raid_disk */ |
98c6faba NB |
149 | unsigned int bestcnt = 0; |
150 | int devcnt = 0; | |
1ff98339 | 151 | unsigned int okcnt, sparecnt, rebuilding_cnt; |
98c6faba NB |
152 | unsigned int req_cnt; |
153 | unsigned int i; | |
64c4757e | 154 | int most_recent = 0; |
cd29a5c8 | 155 | int chosen_drive; |
52826846 | 156 | int change = 0; |
cd29a5c8 | 157 | int inargv = 0; |
52437b4f | 158 | int report_missmatch; |
bf4fb153 | 159 | int bitmap_done; |
7f91af49 N |
160 | int start_partial_ok = (runstop >= 0) && |
161 | (force || devlist==NULL || auto_assem); | |
98c6faba | 162 | unsigned int num_devs; |
5787fa49 | 163 | mddev_dev_t tmpdev; |
4b1ac34b | 164 | struct mdinfo info; |
98dbd966 | 165 | struct mdinfo *content = NULL; |
265e0f17 | 166 | char *avail; |
308e1801 | 167 | int nextspare = 0; |
197e3eb6 | 168 | char *name = NULL; |
69207ff6 | 169 | int trustworthy; |
a04d5763 | 170 | char chosen_name[1024]; |
8a46fe84 | 171 | |
682c7051 | 172 | if (get_linux_version() < 2004000) |
64c4757e NB |
173 | old_linux = 1; |
174 | ||
64c4757e | 175 | /* |
52826846 NB |
176 | * If any subdevs are listed, then any that don't |
177 | * match ident are discarded. Remainder must all match and | |
178 | * become the array. | |
179 | * If no subdevs, then we scan all devices in the config file, but | |
180 | * there must be something in the identity | |
64c4757e | 181 | */ |
64c4757e | 182 | |
cd29a5c8 | 183 | if (!devlist && |
52826846 NB |
184 | ident->uuid_set == 0 && |
185 | ident->super_minor < 0 && | |
e0fe762a N |
186 | ident->name[0] == 0 && |
187 | (ident->container == NULL || ident->member == NULL) && | |
52826846 NB |
188 | ident->devices == NULL) { |
189 | fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n", | |
8a46fe84 | 190 | mddev ? mddev : "further assembly"); |
52826846 | 191 | return 1; |
64c4757e | 192 | } |
71d60c48 | 193 | |
cd29a5c8 | 194 | if (devlist == NULL) |
8aec876d | 195 | devlist = conf_get_devs(); |
7f91af49 | 196 | else if (mddev) |
da6b5ca9 | 197 | inargv = 1; |
64c4757e | 198 | |
52437b4f | 199 | report_missmatch = ((inargv && verbose >= 0) || verbose > 0); |
60e1bc1a | 200 | try_again: |
c30e5369 N |
201 | /* We come back here when doing auto-assembly and attempting some |
202 | * set of devices failed. Those are now marked as ->used==2 and | |
203 | * we ignore them and try again | |
204 | */ | |
60e1bc1a | 205 | |
5787fa49 NB |
206 | tmpdev = devlist; num_devs = 0; |
207 | while (tmpdev) { | |
da6b5ca9 NB |
208 | if (tmpdev->used) |
209 | tmpdev->used = 2; | |
210 | else | |
211 | num_devs++; | |
5787fa49 NB |
212 | tmpdev = tmpdev->next; |
213 | } | |
5787fa49 NB |
214 | devices = malloc(num_devs * sizeof(*devices)); |
215 | ||
82d9eba6 | 216 | if (!st && ident->st) st = ident->st; |
64c4757e | 217 | |
dab6685f | 218 | if (verbose>0) |
82b27616 | 219 | fprintf(stderr, Name ": looking for devices for %s\n", |
8a46fe84 | 220 | mddev ? mddev : "further assembly"); |
82b27616 | 221 | |
811e6cbe NB |
222 | /* first walk the list of devices to find a consistent set |
223 | * that match the criterea, if that is possible. | |
c30e5369 | 224 | * We flag the ones we like with 'used'. |
811e6cbe NB |
225 | */ |
226 | for (tmpdev = devlist; | |
227 | tmpdev; | |
228 | tmpdev = tmpdev->next) { | |
229 | char *devname = tmpdev->devname; | |
64c4757e NB |
230 | int dfd; |
231 | struct stat stb; | |
3da92f27 | 232 | struct supertype *tst = dup_super(st); |
52826846 | 233 | |
da6b5ca9 NB |
234 | if (tmpdev->used > 1) continue; |
235 | ||
52826846 | 236 | if (ident->devices && |
56eedc1a | 237 | !match_oneof(ident->devices, devname)) { |
52437b4f | 238 | if (report_missmatch) |
56eedc1a | 239 | fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices); |
52826846 | 240 | continue; |
56eedc1a | 241 | } |
4b1ac34b | 242 | |
8b0dabea | 243 | dfd = dev_open(devname, O_RDONLY|O_EXCL); |
64c4757e | 244 | if (dfd < 0) { |
52437b4f | 245 | if (report_missmatch) |
682c7051 | 246 | fprintf(stderr, Name ": cannot open device %s: %s\n", |
64c4757e | 247 | devname, strerror(errno)); |
da6b5ca9 | 248 | tmpdev->used = 2; |
52826846 NB |
249 | } else if (fstat(dfd, &stb)< 0) { |
250 | /* Impossible! */ | |
251 | fprintf(stderr, Name ": fstat failed for %s: %s\n", | |
252 | devname, strerror(errno)); | |
da6b5ca9 | 253 | tmpdev->used = 2; |
cd29a5c8 NB |
254 | } else if ((stb.st_mode & S_IFMT) != S_IFBLK) { |
255 | fprintf(stderr, Name ": %s is not a block device.\n", | |
52826846 | 256 | devname); |
da6b5ca9 | 257 | tmpdev->used = 2; |
82d9eba6 | 258 | } else if (!tst && (tst = guess_super(dfd)) == NULL) { |
52437b4f | 259 | if (report_missmatch) |
da6b5ca9 NB |
260 | fprintf(stderr, Name ": no recogniseable superblock on %s\n", |
261 | devname); | |
262 | tmpdev->used = 2; | |
1ff98339 N |
263 | } else if (tst->ss->load_super(tst,dfd, NULL)) { |
264 | if (report_missmatch) | |
265 | fprintf( stderr, Name ": no RAID superblock on %s\n", | |
266 | devname); | |
31015d57 | 267 | } else if (auto_assem && st == NULL && |
d1d3482b N |
268 | !conf_test_metadata(tst->ss->name, |
269 | tst->ss->match_home(tst, homehost) == 1)) { | |
31015d57 N |
270 | if (report_missmatch) |
271 | fprintf(stderr, Name ": %s has metadata type %s for which " | |
272 | "auto-assembly is disabled\n", | |
273 | devname, tst->ss->name); | |
274 | tmpdev->used = 2; | |
52826846 | 275 | } else { |
98dbd966 DW |
276 | content = &info; |
277 | memset(content, 0, sizeof(*content)); | |
278 | tst->ss->getinfo_super(tst, content); | |
64c4757e | 279 | } |
c06487ce | 280 | if (dfd >= 0) close(dfd); |
52826846 | 281 | |
9008ed1c N |
282 | if (tst && tst->sb && tst->ss->container_content |
283 | && tst->loaded_container) { | |
284 | /* tmpdev is a container. We need to be either | |
285 | * looking for a member, or auto-assembling | |
286 | */ | |
287 | if (st) { | |
288 | /* already found some components, this cannot | |
289 | * be another one. | |
290 | */ | |
291 | if (report_missmatch) | |
292 | fprintf(stderr, Name ": %s is a container, but we are looking for components\n", | |
293 | devname); | |
294 | goto loop; | |
295 | } | |
296 | ||
297 | if (ident->container) { | |
298 | if (ident->container[0] == '/' && | |
299 | !same_dev(ident->container, devname)) { | |
300 | if (report_missmatch) | |
301 | fprintf(stderr, Name ": %s is not the container required (%s)\n", | |
302 | devname, ident->container); | |
303 | goto loop; | |
304 | } | |
305 | if (ident->container[0] != '/') { | |
306 | /* we have a uuid */ | |
307 | int uuid[4]; | |
308 | if (!parse_uuid(ident->container, uuid) || | |
309 | !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) { | |
310 | if (report_missmatch) | |
311 | fprintf(stderr, Name ": %s has wrong UUID to be required container\n", | |
312 | devname); | |
313 | goto loop; | |
314 | } | |
315 | } | |
316 | } | |
317 | /* It is worth looking inside this container. | |
318 | */ | |
7636b5a8 N |
319 | if (verbose > 0) |
320 | fprintf(stderr, Name ": looking in container %s\n", | |
321 | devname); | |
9008ed1c N |
322 | next_member: |
323 | if (tmpdev->content) | |
324 | content = tmpdev->content; | |
325 | else | |
326 | content = tst->ss->container_content(tst); | |
3ef383aa DW |
327 | if (!content) |
328 | goto loop; /* empty container */ | |
9008ed1c N |
329 | |
330 | tmpdev->content = content->next; | |
331 | if (tmpdev->content == NULL) | |
f7ad3ccc | 332 | tmpdev->used = 2; |
9008ed1c N |
333 | |
334 | } else if (ident->container || ident->member) { | |
335 | /* No chance of this matching if we don't have | |
336 | * a container */ | |
337 | if (report_missmatch) | |
338 | fprintf(stderr, Name "%s is not a container, and one is required.\n", | |
339 | devname); | |
340 | goto loop; | |
341 | } | |
342 | ||
838acbc2 | 343 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && |
3da92f27 | 344 | (!tst || !tst->sb || |
98dbd966 | 345 | same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) { |
52437b4f | 346 | if (report_missmatch) |
52826846 NB |
347 | fprintf(stderr, Name ": %s has wrong uuid.\n", |
348 | devname); | |
df37ffc0 | 349 | goto loop; |
82b27616 | 350 | } |
c4f12c13 | 351 | if (ident->name[0] && (!update || strcmp(update, "name")!= 0) && |
3da92f27 | 352 | (!tst || !tst->sb || |
98dbd966 | 353 | name_matches(content->name, ident->name, homehost)==0)) { |
52437b4f | 354 | if (report_missmatch) |
947fd4dd NB |
355 | fprintf(stderr, Name ": %s has wrong name.\n", |
356 | devname); | |
df37ffc0 | 357 | goto loop; |
947fd4dd | 358 | } |
98c6faba | 359 | if (ident->super_minor != UnSet && |
3da92f27 | 360 | (!tst || !tst->sb || |
98dbd966 | 361 | ident->super_minor != content->array.md_minor)) { |
52437b4f | 362 | if (report_missmatch) |
52826846 | 363 | fprintf(stderr, Name ": %s has wrong super-minor.\n", |
64c4757e | 364 | devname); |
df37ffc0 | 365 | goto loop; |
64c4757e | 366 | } |
98c6faba | 367 | if (ident->level != UnSet && |
3da92f27 | 368 | (!tst || !tst->sb || |
98dbd966 | 369 | ident->level != content->array.level)) { |
52437b4f | 370 | if (report_missmatch) |
cd29a5c8 NB |
371 | fprintf(stderr, Name ": %s has wrong raid level.\n", |
372 | devname); | |
df37ffc0 | 373 | goto loop; |
cd29a5c8 | 374 | } |
98c6faba | 375 | if (ident->raid_disks != UnSet && |
3da92f27 | 376 | (!tst || !tst->sb || |
98dbd966 | 377 | ident->raid_disks!= content->array.raid_disks)) { |
52437b4f | 378 | if (report_missmatch) |
cd29a5c8 NB |
379 | fprintf(stderr, Name ": %s requires wrong number of drives.\n", |
380 | devname); | |
df37ffc0 | 381 | goto loop; |
cd29a5c8 | 382 | } |
c30e5369 | 383 | if (auto_assem) { |
3da92f27 | 384 | if (tst == NULL || tst->sb == NULL) |
da6b5ca9 | 385 | continue; |
da6b5ca9 | 386 | } |
83b6208e | 387 | /* If we are this far, then we are nearly commited to this device. |
52826846 | 388 | * If the super_block doesn't exist, or doesn't match others, |
83b6208e NB |
389 | * then we probably cannot continue |
390 | * However if one of the arrays is for the homehost, and | |
391 | * the other isn't that can disambiguate. | |
52826846 | 392 | */ |
52826846 | 393 | |
3da92f27 | 394 | if (!tst || !tst->sb) { |
52826846 NB |
395 | fprintf(stderr, Name ": %s has no superblock - assembly aborted\n", |
396 | devname); | |
ddd1a492 NB |
397 | if (st) |
398 | st->ss->free_super(st); | |
52826846 NB |
399 | return 1; |
400 | } | |
838acbc2 | 401 | |
9008ed1c N |
402 | if (tst && tst->sb && tst->ss->container_content |
403 | && tst->loaded_container) { | |
404 | /* we have the one container we need, don't keep | |
405 | * looking. If the chosen member is active, skip. | |
406 | */ | |
407 | if (is_member_busy(content->text_version)) { | |
f7ad3ccc N |
408 | if (report_missmatch) |
409 | fprintf(stderr, Name ": member %s in %s is already assembled\n", | |
410 | content->text_version, | |
411 | devname); | |
8a0a0ded N |
412 | skip: |
413 | if (tmpdev->content) | |
414 | goto next_member; | |
f7ad3ccc N |
415 | tst->ss->free_super(tst); |
416 | tst = NULL; | |
417 | content = NULL; | |
9008ed1c N |
418 | if (auto_assem) |
419 | goto loop; | |
9008ed1c N |
420 | return 1; |
421 | } | |
8a0a0ded N |
422 | if (ident->member && ident->member[0]) { |
423 | char *s = strchr(content->text_version+1, '/'); | |
424 | if (s == NULL) { | |
425 | fprintf(stderr, Name ": badly formatted version: %s\n", | |
426 | content->text_version); | |
427 | goto skip; | |
428 | } | |
429 | if (strcmp(ident->member, s+1) != 0) { | |
430 | if (report_missmatch) | |
431 | fprintf(stderr, | |
432 | Name ": skipping wrong member %s\n", | |
433 | content->text_version); | |
434 | goto skip; | |
435 | } | |
436 | } | |
9008ed1c | 437 | st = tst; tst = NULL; |
4c1c3ad8 | 438 | if (!auto_assem && inargv && tmpdev->next != NULL) { |
9008ed1c N |
439 | fprintf(stderr, Name ": %s is a container, but is not " |
440 | "only device given: confused and aborting\n", | |
441 | devname); | |
442 | st->ss->free_super(st); | |
443 | return 1; | |
444 | } | |
7636b5a8 N |
445 | if (verbose > 0) |
446 | fprintf(stderr, Name ": found match on member %s in %s\n", | |
447 | content->text_version, devname); | |
9008ed1c N |
448 | break; |
449 | } | |
83b6208e | 450 | if (st == NULL) |
3da92f27 NB |
451 | st = dup_super(tst); |
452 | if (st->minor_version == -1) | |
453 | st->minor_version = tst->minor_version; | |
83b6208e NB |
454 | if (st->ss != tst->ss || |
455 | st->minor_version != tst->minor_version || | |
64557c33 | 456 | st->ss->compare_super(st, tst) != 0) { |
83b6208e | 457 | /* Some mismatch. If exactly one array matches this host, |
da6b5ca9 NB |
458 | * we can resolve on that one. |
459 | * Or, if we are auto assembling, we just ignore the second | |
460 | * for now. | |
83b6208e | 461 | */ |
c30e5369 | 462 | if (auto_assem) |
df37ffc0 | 463 | goto loop; |
83b6208e | 464 | if (homehost) { |
3da92f27 NB |
465 | int first = st->ss->match_home(st, homehost); |
466 | int last = tst->ss->match_home(tst, homehost); | |
9362c1c8 N |
467 | if (first != last && |
468 | (first == 1 || last == 1)) { | |
83b6208e NB |
469 | /* We can do something */ |
470 | if (first) {/* just ignore this one */ | |
52437b4f | 471 | if (report_missmatch) |
83b6208e NB |
472 | fprintf(stderr, Name ": %s misses out due to wrong homehost\n", |
473 | devname); | |
df37ffc0 | 474 | goto loop; |
83b6208e NB |
475 | } else { /* reject all those sofar */ |
476 | mddev_dev_t td; | |
52437b4f | 477 | if (report_missmatch) |
83b6208e NB |
478 | fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n", |
479 | devname); | |
480 | for (td=devlist; td != tmpdev; td=td->next) | |
481 | if (td->used == 1) | |
482 | td->used = 0; | |
483 | tmpdev->used = 1; | |
df37ffc0 | 484 | goto loop; |
83b6208e NB |
485 | } |
486 | } | |
487 | } | |
52826846 NB |
488 | fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n", |
489 | devname); | |
3da92f27 NB |
490 | tst->ss->free_super(tst); |
491 | st->ss->free_super(st); | |
52826846 | 492 | return 1; |
64c4757e NB |
493 | } |
494 | ||
811e6cbe | 495 | tmpdev->used = 1; |
df37ffc0 NB |
496 | |
497 | loop: | |
9008ed1c N |
498 | if (tmpdev->content) |
499 | goto next_member; | |
3d2b16e7 NB |
500 | if (tst) |
501 | tst->ss->free_super(tst); | |
811e6cbe NB |
502 | } |
503 | ||
98dbd966 | 504 | if (!st || !st->sb || !content) |
c30e5369 N |
505 | return 2; |
506 | ||
05833051 | 507 | /* Now need to open the array device. Use create_mddev */ |
98dbd966 DW |
508 | if (content == &info) |
509 | st->ss->getinfo_super(st, content); | |
3d2c4fc7 | 510 | |
69207ff6 | 511 | trustworthy = FOREIGN; |
05833051 | 512 | name = content->name; |
745f72f6 N |
513 | switch (st->ss->match_home(st, homehost) |
514 | ?: st->ss->match_home(st, "any")) { | |
69207ff6 N |
515 | case 1: |
516 | trustworthy = LOCAL; | |
98dbd966 | 517 | name = strchr(content->name, ':'); |
69207ff6 N |
518 | if (name) |
519 | name++; | |
3d2c4fc7 | 520 | else |
98dbd966 | 521 | name = content->name; |
69207ff6 | 522 | break; |
c30e5369 | 523 | } |
05833051 | 524 | if (!auto_assem) |
aa7c284c | 525 | /* If the array is listed in mdadm.conf or on |
ac2ecf55 N |
526 | * command line, then we trust the name |
527 | * even if the array doesn't look local | |
528 | */ | |
529 | trustworthy = LOCAL; | |
530 | ||
05833051 | 531 | if (name[0] == 0 && |
98dbd966 DW |
532 | content->array.level == LEVEL_CONTAINER) { |
533 | name = content->text_version; | |
a4bc1720 N |
534 | trustworthy = METADATA; |
535 | } | |
0ac91628 N |
536 | |
537 | if (name[0] && trustworthy != LOCAL && | |
538 | ! require_homehost && | |
539 | conf_name_is_free(name)) | |
540 | trustworthy = LOCAL; | |
541 | ||
7cdc0872 N |
542 | if (trustworthy == LOCAL && |
543 | strchr(name, ':')) | |
544 | /* Ignore 'host:' prefix of name */ | |
545 | name = strchr(name, ':')+1; | |
546 | ||
a04d5763 N |
547 | mdfd = create_mddev(mddev, name, ident->autof, trustworthy, |
548 | chosen_name); | |
c30e5369 N |
549 | if (mdfd < 0) { |
550 | st->ss->free_super(st); | |
551 | free(devices); | |
552 | if (auto_assem) | |
60e1bc1a | 553 | goto try_again; |
c30e5369 N |
554 | return 1; |
555 | } | |
a04d5763 | 556 | mddev = chosen_name; |
c30e5369 N |
557 | vers = md_get_version(mdfd); |
558 | if (vers < 9000) { | |
559 | fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n" | |
560 | " Upgrade your kernel or try --build\n"); | |
561 | close(mdfd); | |
562 | return 1; | |
563 | } | |
66afdfa9 | 564 | if (mddev_busy(fd2devnum(mdfd))) { |
c30e5369 N |
565 | fprintf(stderr, Name ": %s already active, cannot restart it!\n", |
566 | mddev); | |
567 | for (tmpdev = devlist ; | |
568 | tmpdev && tmpdev->used != 1; | |
569 | tmpdev = tmpdev->next) | |
570 | ; | |
571 | if (tmpdev && auto_assem) | |
572 | fprintf(stderr, Name ": %s needed for %s...\n", | |
573 | mddev, tmpdev->devname); | |
574 | close(mdfd); | |
575 | mdfd = -3; | |
576 | st->ss->free_super(st); | |
577 | free(devices); | |
578 | if (auto_assem) | |
579 | goto try_again; | |
580 | return 1; | |
8a46fe84 | 581 | } |
c30e5369 | 582 | ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */ |
8a46fe84 | 583 | |
9008ed1c N |
584 | #ifndef MDASSEMBLE |
585 | if (content != &info) { | |
586 | /* This is a member of a container. Try starting the array. */ | |
587 | return assemble_container_content(st, mdfd, content, runstop, | |
588 | chosen_name, verbose); | |
589 | } | |
590 | #endif | |
811e6cbe | 591 | /* Ok, no bad inconsistancy, we can try updating etc */ |
bf4fb153 | 592 | bitmap_done = 0; |
6e46bf34 | 593 | content->update_private = NULL; |
da6b5ca9 | 594 | for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) { |
811e6cbe NB |
595 | char *devname = tmpdev->devname; |
596 | struct stat stb; | |
5787fa49 | 597 | /* looks like a good enough match to update the super block if needed */ |
d1f1011b | 598 | #ifndef MDASSEMBLE |
5787fa49 | 599 | if (update) { |
811e6cbe | 600 | int dfd; |
4b1ac34b NB |
601 | /* prepare useful information in info structures */ |
602 | struct stat stb2; | |
3da92f27 | 603 | struct supertype *tst; |
4b1ac34b | 604 | fstat(mdfd, &stb2); |
7d99579f NB |
605 | |
606 | if (strcmp(update, "uuid")==0 && | |
607 | !ident->uuid_set) { | |
608 | int rfd; | |
609 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
610 | read(rfd, ident->uuid, 16) != 16) { | |
611 | *(__u32*)(ident->uuid) = random(); | |
612 | *(__u32*)(ident->uuid+1) = random(); | |
613 | *(__u32*)(ident->uuid+2) = random(); | |
614 | *(__u32*)(ident->uuid+3) = random(); | |
615 | } | |
616 | if (rfd >= 0) close(rfd); | |
7d99579f | 617 | } |
811e6cbe NB |
618 | dfd = dev_open(devname, O_RDWR|O_EXCL); |
619 | ||
0430ed48 NB |
620 | remove_partitions(dfd); |
621 | ||
3da92f27 | 622 | tst = dup_super(st); |
9f22b13f N |
623 | if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) { |
624 | fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n", | |
625 | devname); | |
626 | if (dfd >= 0) | |
627 | close(dfd); | |
628 | close(mdfd); | |
629 | return 1; | |
630 | } | |
98dbd966 | 631 | tst->ss->getinfo_super(tst, content); |
811e6cbe | 632 | |
98dbd966 DW |
633 | memcpy(content->uuid, ident->uuid, 16); |
634 | strcpy(content->name, ident->name); | |
635 | content->array.md_minor = minor(stb2.st_rdev); | |
811e6cbe | 636 | |
98dbd966 | 637 | tst->ss->update_super(tst, content, update, |
3da92f27 NB |
638 | devname, verbose, |
639 | ident->uuid_set, homehost); | |
e5eac01f NB |
640 | if (strcmp(update, "uuid")==0 && |
641 | !ident->uuid_set) { | |
642 | ident->uuid_set = 1; | |
98dbd966 | 643 | memcpy(ident->uuid, content->uuid, 16); |
e5eac01f | 644 | } |
b8a8ccf9 | 645 | if (dfd < 0) |
5787fa49 NB |
646 | fprintf(stderr, Name ": Cannot open %s for superblock update\n", |
647 | devname); | |
3da92f27 | 648 | else if (tst->ss->store_super(tst, dfd)) |
5787fa49 NB |
649 | fprintf(stderr, Name ": Could not re-write superblock on %s.\n", |
650 | devname); | |
651 | if (dfd >= 0) | |
652 | close(dfd); | |
8131b493 NB |
653 | |
654 | if (strcmp(update, "uuid")==0 && | |
bf4fb153 | 655 | ident->bitmap_fd >= 0 && !bitmap_done) { |
3da92f27 | 656 | if (bitmap_update_uuid(ident->bitmap_fd, |
98dbd966 | 657 | content->uuid, |
3da92f27 | 658 | tst->ss->swapuuid) != 0) |
bf4fb153 NB |
659 | fprintf(stderr, Name ": Could not update uuid on external bitmap.\n"); |
660 | else | |
661 | bitmap_done = 1; | |
662 | } | |
3da92f27 | 663 | tst->ss->free_super(tst); |
d1f1011b NB |
664 | } else |
665 | #endif | |
666 | { | |
30e1b9a5 | 667 | struct supertype *tst = dup_super(st); |
da6b5ca9 NB |
668 | int dfd; |
669 | dfd = dev_open(devname, O_RDWR|O_EXCL); | |
670 | ||
0430ed48 NB |
671 | remove_partitions(dfd); |
672 | ||
9f22b13f N |
673 | if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) { |
674 | fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n", | |
675 | devname); | |
676 | if (dfd >= 0) | |
677 | close(dfd); | |
678 | close(mdfd); | |
679 | return 1; | |
680 | } | |
98dbd966 | 681 | tst->ss->getinfo_super(tst, content); |
3da92f27 | 682 | tst->ss->free_super(tst); |
da6b5ca9 | 683 | close(dfd); |
5787fa49 NB |
684 | } |
685 | ||
811e6cbe NB |
686 | stat(devname, &stb); |
687 | ||
dab6685f | 688 | if (verbose > 0) |
cd29a5c8 | 689 | fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n", |
98dbd966 | 690 | devname, mddev, content->disk.raid_disk); |
64c4757e | 691 | devices[devcnt].devname = devname; |
64c4757e | 692 | devices[devcnt].uptodate = 0; |
98dbd966 | 693 | devices[devcnt].i = *content; |
213ee40b NB |
694 | devices[devcnt].i.disk.major = major(stb.st_rdev); |
695 | devices[devcnt].i.disk.minor = minor(stb.st_rdev); | |
64c4757e | 696 | if (most_recent < devcnt) { |
213ee40b NB |
697 | if (devices[devcnt].i.events |
698 | > devices[most_recent].i.events) | |
64c4757e NB |
699 | most_recent = devcnt; |
700 | } | |
df0d4ea0 | 701 | if (content->array.level == LEVEL_MULTIPATH) |
5787fa49 NB |
702 | /* with multipath, the raid_disk from the superblock is meaningless */ |
703 | i = devcnt; | |
704 | else | |
213ee40b | 705 | i = devices[devcnt].i.disk.raid_disk; |
308e1801 | 706 | if (i+1 == 0) { |
98dbd966 DW |
707 | if (nextspare < content->array.raid_disks) |
708 | nextspare = content->array.raid_disks; | |
308e1801 | 709 | i = nextspare++; |
08110d41 | 710 | } else { |
98dbd966 | 711 | if (i >= content->array.raid_disks && |
08110d41 NB |
712 | i >= nextspare) |
713 | nextspare = i+1; | |
308e1801 | 714 | } |
98c6faba | 715 | if (i < 10000) { |
56eedc1a | 716 | if (i >= bestcnt) { |
98c6faba | 717 | unsigned int newbestcnt = i+10; |
56eedc1a | 718 | int *newbest = malloc(sizeof(int)*newbestcnt); |
98c6faba | 719 | unsigned int c; |
56eedc1a NB |
720 | for (c=0; c < newbestcnt; c++) |
721 | if (c < bestcnt) | |
722 | newbest[c] = best[c]; | |
723 | else | |
724 | newbest[c] = -1; | |
725 | if (best)free(best); | |
726 | best = newbest; | |
727 | bestcnt = newbestcnt; | |
728 | } | |
6a255b69 | 729 | if (best[i] >=0 && |
213ee40b NB |
730 | devices[best[i]].i.events |
731 | == devices[devcnt].i.events | |
732 | && (devices[best[i]].i.disk.minor | |
733 | != devices[devcnt].i.disk.minor) | |
b8ac1967 | 734 | && st->ss == &super0 |
98dbd966 | 735 | && content->array.level != LEVEL_MULTIPATH) { |
6a255b69 NB |
736 | /* two different devices with identical superblock. |
737 | * Could be a mis-detection caused by overlapping | |
738 | * partitions. fail-safe. | |
739 | */ | |
740 | fprintf(stderr, Name ": WARNING %s and %s appear" | |
741 | " to have very similar superblocks.\n" | |
742 | " If they are really different, " | |
743 | "please --zero the superblock on one\n" | |
d2df86e0 NB |
744 | " If they are the same or overlap," |
745 | " please remove one from %s.\n", | |
746 | devices[best[i]].devname, devname, | |
747 | inargv ? "the list" : | |
748 | "the\n DEVICE list in mdadm.conf" | |
749 | ); | |
7f91af49 | 750 | close(mdfd); |
6a255b69 NB |
751 | return 1; |
752 | } | |
52826846 | 753 | if (best[i] == -1 |
213ee40b NB |
754 | || (devices[best[i]].i.events |
755 | < devices[devcnt].i.events)) | |
52826846 | 756 | best[i] = devcnt; |
56eedc1a | 757 | } |
64c4757e | 758 | devcnt++; |
df37ffc0 | 759 | } |
6e46bf34 DW |
760 | free(content->update_private); |
761 | content->update_private = NULL; | |
4b1ac34b | 762 | |
64c4757e | 763 | if (devcnt == 0) { |
682c7051 | 764 | fprintf(stderr, Name ": no devices found for %s\n", |
64c4757e | 765 | mddev); |
3d2b16e7 NB |
766 | if (st) |
767 | st->ss->free_super(st); | |
7f91af49 | 768 | close(mdfd); |
64c4757e NB |
769 | return 1; |
770 | } | |
4b1ac34b | 771 | |
3d2b16e7 NB |
772 | if (update && strcmp(update, "byteorder")==0) |
773 | st->minor_version = 90; | |
774 | ||
98dbd966 DW |
775 | st->ss->getinfo_super(st, content); |
776 | clean = content->array.state & 1; | |
4b1ac34b | 777 | |
64c4757e NB |
778 | /* now we have some devices that might be suitable. |
779 | * I wonder how many | |
780 | */ | |
98dbd966 DW |
781 | avail = malloc(content->array.raid_disks); |
782 | memset(avail, 0, content->array.raid_disks); | |
64c4757e | 783 | okcnt = 0; |
52826846 | 784 | sparecnt=0; |
1ff98339 | 785 | rebuilding_cnt=0; |
56eedc1a | 786 | for (i=0; i< bestcnt ;i++) { |
64c4757e | 787 | int j = best[i]; |
ee04451c NB |
788 | int event_margin = 1; /* always allow a difference of '1' |
789 | * like the kernel does | |
790 | */ | |
64c4757e | 791 | if (j < 0) continue; |
d013a55e NB |
792 | /* note: we ignore error flags in multipath arrays |
793 | * as they don't make sense | |
794 | */ | |
df0d4ea0 | 795 | if (content->array.level != LEVEL_MULTIPATH) |
f22385f9 | 796 | if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) { |
213ee40b NB |
797 | if (!(devices[j].i.disk.state |
798 | & (1<<MD_DISK_FAULTY))) | |
aa88f531 | 799 | sparecnt++; |
d013a55e | 800 | continue; |
aa88f531 | 801 | } |
213ee40b NB |
802 | if (devices[j].i.events+event_margin >= |
803 | devices[most_recent].i.events) { | |
64c4757e | 804 | devices[j].uptodate = 1; |
1ff98339 N |
805 | if (i < content->array.raid_disks) { |
806 | if (devices[j].i.recovery_start == MaxSector) { | |
807 | okcnt++; | |
808 | avail[i]=1; | |
809 | } else | |
810 | rebuilding_cnt++; | |
265e0f17 | 811 | } else |
52826846 | 812 | sparecnt++; |
64c4757e NB |
813 | } |
814 | } | |
98dbd966 DW |
815 | while (force && !enough(content->array.level, content->array.raid_disks, |
816 | content->array.layout, 1, | |
265e0f17 | 817 | avail, okcnt)) { |
64c4757e NB |
818 | /* Choose the newest best drive which is |
819 | * not up-to-date, update the superblock | |
820 | * and add it. | |
821 | */ | |
52826846 | 822 | int fd; |
3da92f27 | 823 | struct supertype *tst; |
c5a6f9a6 | 824 | long long current_events; |
cd29a5c8 | 825 | chosen_drive = -1; |
98dbd966 | 826 | for (i=0; i<content->array.raid_disks && i < bestcnt; i++) { |
52826846 NB |
827 | int j = best[i]; |
828 | if (j>=0 && | |
829 | !devices[j].uptodate && | |
921d9e16 | 830 | devices[j].i.recovery_start == MaxSector && |
52826846 | 831 | (chosen_drive < 0 || |
213ee40b NB |
832 | devices[j].i.events |
833 | > devices[chosen_drive].i.events)) | |
52826846 NB |
834 | chosen_drive = j; |
835 | } | |
836 | if (chosen_drive < 0) | |
837 | break; | |
213ee40b | 838 | current_events = devices[chosen_drive].i.events; |
c5a6f9a6 | 839 | add_another: |
dab6685f NB |
840 | if (verbose >= 0) |
841 | fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n", | |
213ee40b NB |
842 | devices[chosen_drive].devname, |
843 | devices[chosen_drive].i.disk.raid_disk, | |
844 | (int)(devices[chosen_drive].i.events), | |
845 | (int)(devices[most_recent].i.events)); | |
8b0dabea | 846 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 NB |
847 | if (fd < 0) { |
848 | fprintf(stderr, Name ": Couldn't open %s for write - not updating\n", | |
849 | devices[chosen_drive].devname); | |
213ee40b | 850 | devices[chosen_drive].i.events = 0; |
52826846 NB |
851 | continue; |
852 | } | |
3da92f27 | 853 | tst = dup_super(st); |
60b435db | 854 | if (tst->ss->load_super(tst,fd, NULL)) { |
52826846 NB |
855 | close(fd); |
856 | fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n", | |
857 | devices[chosen_drive].devname); | |
213ee40b | 858 | devices[chosen_drive].i.events = 0; |
52826846 NB |
859 | continue; |
860 | } | |
98dbd966 DW |
861 | content->events = devices[most_recent].i.events; |
862 | tst->ss->update_super(tst, content, "force-one", | |
67a8c82d NB |
863 | devices[chosen_drive].devname, verbose, |
864 | 0, NULL); | |
4b1ac34b | 865 | |
3da92f27 | 866 | if (tst->ss->store_super(tst, fd)) { |
52826846 NB |
867 | close(fd); |
868 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
869 | devices[chosen_drive].devname); | |
213ee40b | 870 | devices[chosen_drive].i.events = 0; |
3da92f27 | 871 | tst->ss->free_super(tst); |
52826846 NB |
872 | continue; |
873 | } | |
874 | close(fd); | |
213ee40b | 875 | devices[chosen_drive].i.events = devices[most_recent].i.events; |
52826846 | 876 | devices[chosen_drive].uptodate = 1; |
265e0f17 | 877 | avail[chosen_drive] = 1; |
52826846 | 878 | okcnt++; |
3da92f27 | 879 | tst->ss->free_super(tst); |
c5a6f9a6 NB |
880 | |
881 | /* If there are any other drives of the same vintage, | |
882 | * add them in as well. We can't lose and we might gain | |
883 | */ | |
98dbd966 | 884 | for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) { |
c5a6f9a6 NB |
885 | int j = best[i]; |
886 | if (j >= 0 && | |
887 | !devices[j].uptodate && | |
213ee40b | 888 | devices[j].i.events == current_events) { |
c5a6f9a6 NB |
889 | chosen_drive = j; |
890 | goto add_another; | |
891 | } | |
892 | } | |
64c4757e | 893 | } |
52826846 NB |
894 | |
895 | /* Now we want to look at the superblock which the kernel will base things on | |
896 | * and compare the devices that we think are working with the devices that the | |
897 | * superblock thinks are working. | |
898 | * If there are differences and --force is given, then update this chosen | |
899 | * superblock. | |
900 | */ | |
cd29a5c8 | 901 | chosen_drive = -1; |
3da92f27 | 902 | st->ss->free_super(st); |
56eedc1a | 903 | for (i=0; chosen_drive < 0 && i<bestcnt; i++) { |
52826846 NB |
904 | int j = best[i]; |
905 | int fd; | |
4b1ac34b | 906 | |
52826846 NB |
907 | if (j<0) |
908 | continue; | |
909 | if (!devices[j].uptodate) | |
910 | continue; | |
911 | chosen_drive = j; | |
8b0dabea | 912 | if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) { |
52826846 NB |
913 | fprintf(stderr, Name ": Cannot open %s: %s\n", |
914 | devices[j].devname, strerror(errno)); | |
7f91af49 | 915 | close(mdfd); |
52826846 NB |
916 | return 1; |
917 | } | |
3da92f27 | 918 | if (st->ss->load_super(st,fd, NULL)) { |
52826846 NB |
919 | close(fd); |
920 | fprintf(stderr, Name ": RAID superblock has disappeared from %s\n", | |
921 | devices[j].devname); | |
7f91af49 | 922 | close(mdfd); |
52826846 NB |
923 | return 1; |
924 | } | |
925 | close(fd); | |
926 | } | |
3da92f27 | 927 | if (st->sb == NULL) { |
4b1ac34b | 928 | fprintf(stderr, Name ": No suitable drives found for %s\n", mddev); |
7f91af49 | 929 | close(mdfd); |
4b1ac34b NB |
930 | return 1; |
931 | } | |
98dbd966 | 932 | st->ss->getinfo_super(st, content); |
f35f2525 | 933 | #ifndef MDASSEMBLE |
98dbd966 | 934 | sysfs_init(content, mdfd, 0); |
f35f2525 | 935 | #endif |
56eedc1a | 936 | for (i=0; i<bestcnt; i++) { |
52826846 | 937 | int j = best[i]; |
98c6faba | 938 | unsigned int desired_state; |
11a3e71d | 939 | |
98dbd966 | 940 | if (i < content->array.raid_disks) |
11a3e71d NB |
941 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); |
942 | else | |
943 | desired_state = 0; | |
944 | ||
52826846 NB |
945 | if (j<0) |
946 | continue; | |
947 | if (!devices[j].uptodate) | |
948 | continue; | |
4b1ac34b | 949 | |
213ee40b | 950 | devices[j].i.disk.state = desired_state; |
4e9a6ff7 N |
951 | if (!(devices[j].i.array.state & 1)) |
952 | clean = 0; | |
213ee40b NB |
953 | |
954 | if (st->ss->update_super(st, &devices[j].i, "assemble", NULL, | |
68c7d6d7 | 955 | verbose, 0, NULL)) { |
52826846 | 956 | if (force) { |
dab6685f NB |
957 | if (verbose >= 0) |
958 | fprintf(stderr, Name ": " | |
959 | "clearing FAULTY flag for device %d in %s for %s\n", | |
960 | j, mddev, devices[j].devname); | |
4b1ac34b | 961 | change = 1; |
52826846 | 962 | } else { |
dab6685f NB |
963 | if (verbose >= -1) |
964 | fprintf(stderr, Name ": " | |
965 | "device %d in %s has wrong state in superblock, but %s seems ok\n", | |
966 | i, mddev, devices[j].devname); | |
52826846 NB |
967 | } |
968 | } | |
4b1ac34b | 969 | #if 0 |
213ee40b | 970 | if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) { |
52826846 NB |
971 | fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", |
972 | i, mddev); | |
973 | } | |
4b1ac34b | 974 | #endif |
52826846 | 975 | } |
c5a6f9a6 | 976 | if (force && !clean && |
98dbd966 DW |
977 | !enough(content->array.level, content->array.raid_disks, |
978 | content->array.layout, clean, | |
c5a6f9a6 | 979 | avail, okcnt)) { |
98dbd966 | 980 | change += st->ss->update_super(st, content, "force-array", |
67a8c82d NB |
981 | devices[chosen_drive].devname, verbose, |
982 | 0, NULL); | |
583315d9 | 983 | clean = 1; |
d013a55e | 984 | } |
52826846 | 985 | |
4b1ac34b | 986 | if (change) { |
52826846 | 987 | int fd; |
8b0dabea | 988 | fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL); |
52826846 | 989 | if (fd < 0) { |
353632d9 | 990 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", |
52826846 | 991 | devices[chosen_drive].devname); |
7f91af49 | 992 | close(mdfd); |
52826846 NB |
993 | return 1; |
994 | } | |
3da92f27 | 995 | if (st->ss->store_super(st, fd)) { |
52826846 NB |
996 | close(fd); |
997 | fprintf(stderr, Name ": Could not re-write superblock on %s\n", | |
998 | devices[chosen_drive].devname); | |
7f91af49 | 999 | close(mdfd); |
52826846 NB |
1000 | return 1; |
1001 | } | |
1002 | close(fd); | |
52826846 NB |
1003 | } |
1004 | ||
353632d9 NB |
1005 | /* If we are in the middle of a reshape we may need to restore saved data |
1006 | * that was moved aside due to the reshape overwriting live data | |
1007 | * The code of doing this lives in Grow.c | |
1008 | */ | |
39c5a909 | 1009 | #ifndef MDASSEMBLE |
98dbd966 | 1010 | if (content->reshape_active) { |
353632d9 NB |
1011 | int err = 0; |
1012 | int *fdlist = malloc(sizeof(int)* bestcnt); | |
cd77ac4e | 1013 | if (verbose > 0) |
ea0ebe96 N |
1014 | fprintf(stderr, Name ":%s has an active reshape - checking " |
1015 | "if critical section needs to be restored\n", | |
1016 | chosen_name); | |
353632d9 NB |
1017 | for (i=0; i<bestcnt; i++) { |
1018 | int j = best[i]; | |
1019 | if (j >= 0) { | |
1020 | fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
1021 | if (fdlist[i] < 0) { | |
1022 | fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n", | |
1023 | devices[j].devname); | |
1024 | err = 1; | |
1025 | break; | |
1026 | } | |
1027 | } else | |
1028 | fdlist[i] = -1; | |
1029 | } | |
1030 | if (!err) | |
cd77ac4e | 1031 | err = Grow_restart(st, content, fdlist, bestcnt, backup_file, verbose > 0); |
353632d9 NB |
1032 | while (i>0) { |
1033 | i--; | |
1034 | if (fdlist[i]>=0) close(fdlist[i]); | |
1035 | } | |
1036 | if (err) { | |
1037 | fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n"); | |
e9e43ec3 N |
1038 | if (backup_file == NULL) |
1039 | fprintf(stderr," Possibly you needed to specify the --backup-file\n"); | |
7f91af49 | 1040 | close(mdfd); |
353632d9 NB |
1041 | return err; |
1042 | } | |
1043 | } | |
39c5a909 | 1044 | #endif |
aa88f531 NB |
1045 | /* count number of in-sync devices according to the superblock. |
1046 | * We must have this number to start the array without -s or -R | |
1047 | */ | |
98dbd966 | 1048 | req_cnt = content->array.working_disks; |
aa88f531 | 1049 | |
64c4757e NB |
1050 | /* Almost ready to actually *do* something */ |
1051 | if (!old_linux) { | |
fbf8a0b7 | 1052 | int rv; |
a19c88b8 | 1053 | |
a04d5763 N |
1054 | /* First, fill in the map, so that udev can find our name |
1055 | * as soon as we become active. | |
1056 | */ | |
98dbd966 DW |
1057 | map_update(NULL, fd2devnum(mdfd), content->text_version, |
1058 | content->uuid, chosen_name); | |
a04d5763 | 1059 | |
98dbd966 | 1060 | rv = set_array_info(mdfd, st, content); |
fbf8a0b7 | 1061 | if (rv) { |
f35f2525 | 1062 | fprintf(stderr, Name ": failed to set array info for %s: %s\n", |
64c4757e | 1063 | mddev, strerror(errno)); |
24af7a87 | 1064 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1065 | close(mdfd); |
64c4757e NB |
1066 | return 1; |
1067 | } | |
11484a63 | 1068 | if (ident->bitmap_fd >= 0) { |
c82f047c NB |
1069 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { |
1070 | fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n"); | |
24af7a87 | 1071 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1072 | close(mdfd); |
c82f047c NB |
1073 | return 1; |
1074 | } | |
7ef02d01 NB |
1075 | } else if (ident->bitmap_file) { |
1076 | /* From config file */ | |
1077 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
1078 | if (bmfd < 0) { | |
1079 | fprintf(stderr, Name ": Could not open bitmap file %s\n", | |
1080 | ident->bitmap_file); | |
24af7a87 | 1081 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1082 | close(mdfd); |
7ef02d01 NB |
1083 | return 1; |
1084 | } | |
1085 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
1086 | fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev); | |
1087 | close(bmfd); | |
24af7a87 | 1088 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1089 | close(mdfd); |
7ef02d01 NB |
1090 | return 1; |
1091 | } | |
1092 | close(bmfd); | |
c82f047c | 1093 | } |
7ef02d01 | 1094 | |
52826846 | 1095 | /* First, add the raid disks, but add the chosen one last */ |
56eedc1a | 1096 | for (i=0; i<= bestcnt; i++) { |
52826846 | 1097 | int j; |
56eedc1a | 1098 | if (i < bestcnt) { |
52826846 NB |
1099 | j = best[i]; |
1100 | if (j == chosen_drive) | |
1101 | continue; | |
1102 | } else | |
1103 | j = chosen_drive; | |
1104 | ||
56eedc1a | 1105 | if (j >= 0 /* && devices[j].uptodate */) { |
98dbd966 | 1106 | rv = add_disk(mdfd, st, content, &devices[j].i); |
7801ac20 | 1107 | |
a19c88b8 | 1108 | if (rv) { |
213ee40b NB |
1109 | fprintf(stderr, Name ": failed to add " |
1110 | "%s to %s: %s\n", | |
64c4757e NB |
1111 | devices[j].devname, |
1112 | mddev, | |
1113 | strerror(errno)); | |
98dbd966 | 1114 | if (i < content->array.raid_disks |
213ee40b | 1115 | || i == bestcnt) |
52826846 NB |
1116 | okcnt--; |
1117 | else | |
1118 | sparecnt--; | |
dab6685f | 1119 | } else if (verbose > 0) |
213ee40b NB |
1120 | fprintf(stderr, Name ": added %s " |
1121 | "to %s as %d\n", | |
1122 | devices[j].devname, mddev, | |
1123 | devices[j].i.disk.raid_disk); | |
98dbd966 | 1124 | } else if (verbose > 0 && i < content->array.raid_disks) |
213ee40b NB |
1125 | fprintf(stderr, Name ": no uptodate device for " |
1126 | "slot %d of %s\n", | |
64c4757e NB |
1127 | i, mddev); |
1128 | } | |
aba69144 | 1129 | |
98dbd966 | 1130 | if (content->array.level == LEVEL_CONTAINER) { |
598f0d58 NB |
1131 | if (verbose >= 0) { |
1132 | fprintf(stderr, Name ": Container %s has been " | |
1133 | "assembled with %d drive%s", | |
57ed8c91 | 1134 | mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s"); |
98dbd966 | 1135 | if (okcnt < content->array.raid_disks) |
598f0d58 | 1136 | fprintf(stderr, " (out of %d)", |
98dbd966 | 1137 | content->array.raid_disks); |
598f0d58 NB |
1138 | fprintf(stderr, "\n"); |
1139 | } | |
98dbd966 | 1140 | sysfs_uevent(content, "change"); |
a7c6e3fb | 1141 | wait_for(chosen_name, mdfd); |
7f91af49 | 1142 | close(mdfd); |
598f0d58 NB |
1143 | return 0; |
1144 | } | |
1145 | ||
64c4757e | 1146 | if (runstop == 1 || |
b8a8ccf9 | 1147 | (runstop <= 0 && |
98dbd966 DW |
1148 | ( enough(content->array.level, content->array.raid_disks, |
1149 | content->array.layout, clean, avail, okcnt) && | |
1ff98339 | 1150 | (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok) |
aa88f531 | 1151 | ))) { |
e9e43ec3 N |
1152 | /* This array is good-to-go. |
1153 | * If a reshape is in progress then we might need to | |
1154 | * continue monitoring it. In that case we start | |
1155 | * it read-only and let the grow code make it writable. | |
1156 | */ | |
1157 | int rv; | |
eb3929a4 | 1158 | #ifndef MDASSEMBLE |
e9e43ec3 N |
1159 | if (content->reshape_active && |
1160 | content->delta_disks <= 0) | |
1161 | rv = Grow_continue(mdfd, st, content, backup_file); | |
1162 | else | |
eb3929a4 | 1163 | #endif |
e9e43ec3 N |
1164 | rv = ioctl(mdfd, RUN_ARRAY, NULL); |
1165 | if (rv == 0) { | |
dab6685f NB |
1166 | if (verbose >= 0) { |
1167 | fprintf(stderr, Name ": %s has been started with %d drive%s", | |
1168 | mddev, okcnt, okcnt==1?"":"s"); | |
98dbd966 DW |
1169 | if (okcnt < content->array.raid_disks) |
1170 | fprintf(stderr, " (out of %d)", content->array.raid_disks); | |
1ff98339 N |
1171 | if (rebuilding_cnt) |
1172 | fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt); | |
dab6685f NB |
1173 | if (sparecnt) |
1174 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
1175 | fprintf(stderr, ".\n"); | |
1176 | } | |
8a659c33 N |
1177 | if (content->reshape_active && |
1178 | content->array.level >= 4 && | |
1179 | content->array.level <= 6) { | |
acee8e89 N |
1180 | /* might need to increase the size |
1181 | * of the stripe cache - default is 256 | |
1182 | */ | |
8a659c33 | 1183 | if (256 < 4 * (content->array.chunk_size/4096)) { |
acee8e89 N |
1184 | struct mdinfo *sra = sysfs_read(mdfd, 0, 0); |
1185 | if (sra) | |
1186 | sysfs_set_num(sra, NULL, | |
1187 | "stripe_cache_size", | |
8a659c33 | 1188 | (4 * content->array.chunk_size / 4096) + 1); |
acee8e89 N |
1189 | } |
1190 | } | |
a7c6e3fb | 1191 | wait_for(mddev, mdfd); |
7f91af49 N |
1192 | close(mdfd); |
1193 | if (auto_assem) { | |
589395d6 | 1194 | int usecs = 1; |
589395d6 NB |
1195 | /* There is a nasty race with 'mdadm --monitor'. |
1196 | * If it opens this device before we close it, | |
1197 | * it gets an incomplete open on which IO | |
598f0d58 NB |
1198 | * doesn't work and the capacity is |
1199 | * wrong. | |
589395d6 NB |
1200 | * If we reopen (to check for layered devices) |
1201 | * before --monitor closes, we loose. | |
1202 | * | |
1203 | * So: wait upto 1 second for there to be | |
1204 | * a non-zero capacity. | |
1205 | */ | |
1206 | while (usecs < 1000) { | |
1207 | mdfd = open(mddev, O_RDONLY); | |
1208 | if (mdfd >= 0) { | |
beae1dfe NB |
1209 | unsigned long long size; |
1210 | if (get_dev_size(mdfd, NULL, &size) && | |
589395d6 NB |
1211 | size > 0) |
1212 | break; | |
1213 | close(mdfd); | |
1214 | } | |
1215 | usleep(usecs); | |
1216 | usecs <<= 1; | |
1217 | } | |
1218 | } | |
64c4757e | 1219 | return 0; |
82b27616 | 1220 | } |
682c7051 | 1221 | fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n", |
64c4757e | 1222 | mddev, strerror(errno)); |
583315d9 | 1223 | |
98dbd966 DW |
1224 | if (!enough(content->array.level, content->array.raid_disks, |
1225 | content->array.layout, 1, avail, okcnt)) | |
583315d9 NB |
1226 | fprintf(stderr, Name ": Not enough devices to " |
1227 | "start the array.\n"); | |
98dbd966 DW |
1228 | else if (!enough(content->array.level, |
1229 | content->array.raid_disks, | |
1230 | content->array.layout, clean, | |
583315d9 NB |
1231 | avail, okcnt)) |
1232 | fprintf(stderr, Name ": Not enough devices to " | |
1233 | "start the array while not clean " | |
1234 | "- consider --force.\n"); | |
1235 | ||
7f91af49 | 1236 | if (auto_assem) |
1c203a4b | 1237 | ioctl(mdfd, STOP_ARRAY, NULL); |
7f91af49 | 1238 | close(mdfd); |
64c4757e NB |
1239 | return 1; |
1240 | } | |
82b27616 | 1241 | if (runstop == -1) { |
b8a8ccf9 | 1242 | fprintf(stderr, Name ": %s assembled from %d drive%s", |
52826846 | 1243 | mddev, okcnt, okcnt==1?"":"s"); |
98dbd966 DW |
1244 | if (okcnt != content->array.raid_disks) |
1245 | fprintf(stderr, " (out of %d)", content->array.raid_disks); | |
b8a8ccf9 | 1246 | fprintf(stderr, ", but not started.\n"); |
7f91af49 | 1247 | close(mdfd); |
64c4757e | 1248 | return 0; |
82b27616 | 1249 | } |
4855f95c | 1250 | if (verbose >= -1) { |
dab6685f | 1251 | fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s"); |
1ff98339 N |
1252 | if (rebuilding_cnt) |
1253 | fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt); | |
dab6685f NB |
1254 | if (sparecnt) |
1255 | fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s"); | |
98dbd966 DW |
1256 | if (!enough(content->array.level, content->array.raid_disks, |
1257 | content->array.layout, 1, avail, okcnt)) | |
dab6685f | 1258 | fprintf(stderr, " - not enough to start the array.\n"); |
98dbd966 DW |
1259 | else if (!enough(content->array.level, |
1260 | content->array.raid_disks, | |
1261 | content->array.layout, clean, | |
583315d9 NB |
1262 | avail, okcnt)) |
1263 | fprintf(stderr, " - not enough to start the " | |
1264 | "array while not clean - consider " | |
1265 | "--force.\n"); | |
dab6685f | 1266 | else { |
98dbd966 | 1267 | if (req_cnt == content->array.raid_disks) |
dab6685f NB |
1268 | fprintf(stderr, " - need all %d to start it", req_cnt); |
1269 | else | |
98dbd966 | 1270 | fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks); |
dab6685f NB |
1271 | fprintf(stderr, " (use --run to insist).\n"); |
1272 | } | |
aa88f531 | 1273 | } |
7f91af49 | 1274 | if (auto_assem) |
1c203a4b | 1275 | ioctl(mdfd, STOP_ARRAY, NULL); |
56a8da69 | 1276 | close(mdfd); |
64c4757e | 1277 | return 1; |
82b27616 | 1278 | } else { |
52826846 NB |
1279 | /* The "chosen_drive" is a good choice, and if necessary, the superblock has |
1280 | * been updated to point to the current locations of devices. | |
1281 | * so we can just start the array | |
82b27616 | 1282 | */ |
cd29a5c8 | 1283 | unsigned long dev; |
213ee40b NB |
1284 | dev = makedev(devices[chosen_drive].i.disk.major, |
1285 | devices[chosen_drive].i.disk.minor); | |
82b27616 NB |
1286 | if (ioctl(mdfd, START_ARRAY, dev)) { |
1287 | fprintf(stderr, Name ": Cannot start array: %s\n", | |
1288 | strerror(errno)); | |
1289 | } | |
aba69144 | 1290 | |
64c4757e | 1291 | } |
7f91af49 | 1292 | close(mdfd); |
e0d19036 | 1293 | return 0; |
64c4757e | 1294 | } |
6234c63c DW |
1295 | |
1296 | #ifndef MDASSEMBLE | |
1297 | int assemble_container_content(struct supertype *st, int mdfd, | |
1298 | struct mdinfo *content, int runstop, | |
1299 | char *chosen_name, int verbose) | |
1300 | { | |
1301 | struct mdinfo *dev, *sra; | |
1302 | int working = 0, preexist = 0; | |
1303 | struct map_ent *map = NULL; | |
1304 | ||
1305 | sysfs_init(content, mdfd, 0); | |
1306 | ||
1307 | sra = sysfs_read(mdfd, 0, GET_VERSION); | |
1308 | if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) | |
4408ee76 N |
1309 | if (sysfs_set_array(content, md_get_version(mdfd)) != 0) { |
1310 | close(mdfd); | |
6234c63c | 1311 | return 1; |
4408ee76 | 1312 | } |
6234c63c DW |
1313 | if (sra) |
1314 | sysfs_free(sra); | |
1315 | ||
1316 | for (dev = content->devs; dev; dev = dev->next) | |
462906cd | 1317 | if (sysfs_add_disk(content, dev, 1) == 0) |
6234c63c DW |
1318 | working++; |
1319 | else if (errno == EEXIST) | |
1320 | preexist++; | |
4408ee76 N |
1321 | if (working == 0) { |
1322 | close(mdfd); | |
7cb2aa33 | 1323 | return 1;/* Nothing new, don't try to start */ |
8b4e5ea9 N |
1324 | } |
1325 | ||
1326 | map_update(&map, fd2devnum(mdfd), | |
1327 | content->text_version, | |
1328 | content->uuid, chosen_name); | |
1329 | ||
1330 | if (runstop > 0 || | |
6234c63c | 1331 | (working + preexist) >= content->array.working_disks) { |
bb50e5d3 | 1332 | int err; |
a714580e | 1333 | |
6234c63c DW |
1334 | switch(content->array.level) { |
1335 | case LEVEL_LINEAR: | |
1336 | case LEVEL_MULTIPATH: | |
1337 | case 0: | |
bb50e5d3 N |
1338 | err = sysfs_set_str(content, NULL, "array_state", |
1339 | "active"); | |
6234c63c DW |
1340 | break; |
1341 | default: | |
bb50e5d3 | 1342 | err = sysfs_set_str(content, NULL, "array_state", |
6234c63c DW |
1343 | "readonly"); |
1344 | /* start mdmon if needed. */ | |
bb50e5d3 N |
1345 | if (!err) { |
1346 | if (!mdmon_running(st->container_dev)) | |
1347 | start_mdmon(st->container_dev); | |
1348 | ping_monitor(devnum2devname(st->container_dev)); | |
1349 | } | |
6234c63c DW |
1350 | break; |
1351 | } | |
bb50e5d3 N |
1352 | if (!err) |
1353 | sysfs_set_safemode(content, content->safe_mode_delay); | |
6234c63c | 1354 | if (verbose >= 0) { |
bb50e5d3 N |
1355 | if (err) |
1356 | fprintf(stderr, Name | |
1357 | ": array %s now has %d devices", | |
1358 | chosen_name, working + preexist); | |
1359 | else | |
1360 | fprintf(stderr, Name | |
1361 | ": Started %s with %d devices", | |
1362 | chosen_name, working + preexist); | |
6234c63c DW |
1363 | if (preexist) |
1364 | fprintf(stderr, " (%d new)", working); | |
1365 | fprintf(stderr, "\n"); | |
1366 | } | |
bb50e5d3 | 1367 | if (!err) |
a7c6e3fb | 1368 | wait_for(chosen_name, mdfd); |
4408ee76 | 1369 | close(mdfd); |
7cb2aa33 | 1370 | return 0; |
6234c63c | 1371 | /* FIXME should have an O_EXCL and wait for read-auto */ |
7cb2aa33 | 1372 | } else { |
6234c63c DW |
1373 | if (verbose >= 0) |
1374 | fprintf(stderr, Name | |
1375 | ": %s assembled with %d devices but " | |
1376 | "not started\n", | |
1377 | chosen_name, working); | |
4408ee76 | 1378 | close(mdfd); |
7cb2aa33 N |
1379 | return 1; |
1380 | } | |
6234c63c DW |
1381 | } |
1382 | #endif | |
1383 |