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