]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com> | |
5 | * | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neilb@suse.de> | |
23 | */ | |
24 | ||
25 | #include "mdadm.h" | |
26 | #include <ctype.h> | |
27 | ||
28 | static int name_matches(char *found, char *required, char *homehost, int require_homehost) | |
29 | { | |
30 | /* See if the name found matches the required name, possibly | |
31 | * prefixed with 'homehost' | |
32 | */ | |
33 | char *sep; | |
34 | unsigned int l; | |
35 | ||
36 | if (strcmp(found, required)==0) | |
37 | return 1; | |
38 | sep = strchr(found, ':'); | |
39 | if (!sep) | |
40 | return 0; | |
41 | l = sep - found; | |
42 | if (strncmp(found, "any:", 4) == 0 || | |
43 | (homehost && strcmp(homehost, "any") == 0) || | |
44 | !require_homehost || | |
45 | (homehost && strlen(homehost) == l && | |
46 | strncmp(found, homehost, l) == 0)) { | |
47 | /* matching homehost */ | |
48 | if (strcmp(sep+1, required) == 0) | |
49 | return 1; | |
50 | } | |
51 | return 0; | |
52 | } | |
53 | ||
54 | static int is_member_busy(char *metadata_version) | |
55 | { | |
56 | /* check if the given member array is active */ | |
57 | struct mdstat_ent *mdstat = mdstat_read(0, 0); | |
58 | struct mdstat_ent *ent; | |
59 | int busy = 0; | |
60 | ||
61 | for (ent = mdstat; ent; ent = ent->next) { | |
62 | if (ent->metadata_version == NULL) | |
63 | continue; | |
64 | if (strncmp(ent->metadata_version, "external:", 9) != 0) | |
65 | continue; | |
66 | if (!is_subarray(&ent->metadata_version[9])) | |
67 | continue; | |
68 | /* Skip first char - it can be '/' or '-' */ | |
69 | if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) { | |
70 | busy = 1; | |
71 | break; | |
72 | } | |
73 | } | |
74 | free_mdstat(mdstat); | |
75 | ||
76 | return busy; | |
77 | } | |
78 | ||
79 | static int ident_matches(struct mddev_ident *ident, | |
80 | struct mdinfo *content, | |
81 | struct supertype *tst, | |
82 | char *homehost, int require_homehost, | |
83 | char *update, char *devname) | |
84 | { | |
85 | ||
86 | if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) && | |
87 | same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 && | |
88 | memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) { | |
89 | if (devname) | |
90 | pr_err("%s has wrong uuid.\n", devname); | |
91 | return 0; | |
92 | } | |
93 | if (ident->name[0] && (!update || strcmp(update, "name")!= 0) && | |
94 | name_matches(content->name, ident->name, homehost, require_homehost)==0) { | |
95 | if (devname) | |
96 | pr_err("%s has wrong name.\n", devname); | |
97 | return 0; | |
98 | } | |
99 | if (ident->super_minor != UnSet && | |
100 | ident->super_minor != content->array.md_minor) { | |
101 | if (devname) | |
102 | pr_err("%s has wrong super-minor.\n", | |
103 | devname); | |
104 | return 0; | |
105 | } | |
106 | if (ident->level != UnSet && | |
107 | ident->level != content->array.level) { | |
108 | if (devname) | |
109 | pr_err("%s has wrong raid level.\n", | |
110 | devname); | |
111 | return 0; | |
112 | } | |
113 | if (ident->raid_disks != UnSet && | |
114 | content->array.raid_disks != 0 && /* metadata doesn't know how many to expect */ | |
115 | ident->raid_disks!= content->array.raid_disks) { | |
116 | if (devname) | |
117 | pr_err("%s requires wrong number of drives.\n", | |
118 | devname); | |
119 | return 0; | |
120 | } | |
121 | if (ident->member && ident->member[0]) { | |
122 | /* content->text_version must match */ | |
123 | char *s = strchr(content->text_version+1, '/'); | |
124 | if (s == NULL) { | |
125 | if (devname) | |
126 | pr_err("%s is not a container and one is required.\n", | |
127 | devname); | |
128 | return 0; | |
129 | } else if (strcmp(ident->member, s+1) != 0) { | |
130 | if (devname) | |
131 | pr_err("skipping wrong member %s is %s\n", | |
132 | content->text_version, devname); | |
133 | return 0; | |
134 | } | |
135 | } | |
136 | return 1; | |
137 | } | |
138 | ||
139 | static int select_devices(struct mddev_dev *devlist, | |
140 | struct mddev_ident *ident, | |
141 | struct supertype **stp, | |
142 | struct mdinfo **contentp, | |
143 | struct context *c, | |
144 | int inargv, int auto_assem) | |
145 | { | |
146 | struct mddev_dev *tmpdev; | |
147 | int num_devs; | |
148 | struct supertype *st = *stp; | |
149 | struct mdinfo *content = NULL; | |
150 | int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0); | |
151 | struct domainlist *domains = NULL; | |
152 | dev_t rdev; | |
153 | ||
154 | tmpdev = devlist; num_devs = 0; | |
155 | while (tmpdev) { | |
156 | if (tmpdev->used) | |
157 | tmpdev->used = 2; | |
158 | else | |
159 | num_devs++; | |
160 | tmpdev->disposition = 0; | |
161 | tmpdev = tmpdev->next; | |
162 | } | |
163 | ||
164 | /* first walk the list of devices to find a consistent set | |
165 | * that match the criterea, if that is possible. | |
166 | * We flag the ones we like with 'used'. | |
167 | */ | |
168 | for (tmpdev = devlist; | |
169 | tmpdev; | |
170 | tmpdev = tmpdev ? tmpdev->next : NULL) { | |
171 | char *devname = tmpdev->devname; | |
172 | int dfd; | |
173 | struct supertype *tst; | |
174 | struct dev_policy *pol = NULL; | |
175 | int found_container = 0; | |
176 | ||
177 | if (tmpdev->used > 1) | |
178 | continue; | |
179 | ||
180 | if (ident->container) { | |
181 | if (ident->container[0] == '/' && | |
182 | !same_dev(ident->container, devname)) { | |
183 | if (report_mismatch) | |
184 | pr_err("%s is not the container required (%s)\n", | |
185 | devname, ident->container); | |
186 | continue; | |
187 | } | |
188 | } else if (ident->devices && | |
189 | !match_oneof(ident->devices, devname)) { | |
190 | /* Note that we ignore the "device=" identifier if a | |
191 | * "container=" is given. Checking both is unnecessarily | |
192 | * complicated. | |
193 | */ | |
194 | if (report_mismatch) | |
195 | pr_err("%s is not one of %s\n", devname, ident->devices); | |
196 | continue; | |
197 | } | |
198 | ||
199 | tst = dup_super(st); | |
200 | ||
201 | dfd = dev_open(devname, O_RDONLY); | |
202 | if (dfd < 0) { | |
203 | if (report_mismatch) | |
204 | pr_err("cannot open device %s: %s\n", | |
205 | devname, strerror(errno)); | |
206 | tmpdev->used = 2; | |
207 | } else if (!fstat_is_blkdev(dfd, devname, &rdev)) { | |
208 | tmpdev->used = 2; | |
209 | } else if (must_be_container(dfd)) { | |
210 | if (st) { | |
211 | /* already found some components, this cannot | |
212 | * be another one. | |
213 | */ | |
214 | if (report_mismatch) | |
215 | pr_err("%s is a container, but we are looking for components\n", | |
216 | devname); | |
217 | tmpdev->used = 2; | |
218 | } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) { | |
219 | if (report_mismatch) | |
220 | pr_err("not a recognisable container: %s\n", | |
221 | devname); | |
222 | tmpdev->used = 2; | |
223 | } else if (!tst->ss->load_container || | |
224 | tst->ss->load_container(tst, dfd, NULL)) { | |
225 | if (report_mismatch) | |
226 | pr_err("no correct container type: %s\n", | |
227 | devname); | |
228 | tmpdev->used = 2; | |
229 | } else if (auto_assem && | |
230 | !conf_test_metadata(tst->ss->name, | |
231 | (pol = devid_policy(rdev)), | |
232 | tst->ss->match_home(tst, c->homehost) == 1)) { | |
233 | if (report_mismatch) | |
234 | pr_err("%s has metadata type %s for which auto-assembly is disabled\n", | |
235 | devname, tst->ss->name); | |
236 | tmpdev->used = 2; | |
237 | } else | |
238 | found_container = 1; | |
239 | } else { | |
240 | if (!tst && (tst = guess_super(dfd)) == NULL) { | |
241 | if (report_mismatch) | |
242 | pr_err("no recogniseable superblock on %s\n", | |
243 | devname); | |
244 | tmpdev->used = 2; | |
245 | } else if ((tst->ignore_hw_compat = 0), | |
246 | tst->ss->load_super(tst, dfd, | |
247 | report_mismatch ? devname : NULL)) { | |
248 | if (report_mismatch) | |
249 | pr_err("no RAID superblock on %s\n", | |
250 | devname); | |
251 | tmpdev->used = 2; | |
252 | } else if (tst->ss->compare_super == NULL) { | |
253 | if (report_mismatch) | |
254 | pr_err("Cannot assemble %s metadata on %s\n", | |
255 | tst->ss->name, devname); | |
256 | tmpdev->used = 2; | |
257 | } else if (auto_assem && st == NULL && | |
258 | !conf_test_metadata(tst->ss->name, | |
259 | (pol = devid_policy(rdev)), | |
260 | tst->ss->match_home(tst, c->homehost) == 1)) { | |
261 | if (report_mismatch) | |
262 | pr_err("%s has metadata type %s for which auto-assembly is disabled\n", | |
263 | devname, tst->ss->name); | |
264 | tmpdev->used = 2; | |
265 | } | |
266 | } | |
267 | if (dfd >= 0) close(dfd); | |
268 | if (tmpdev->used == 2) { | |
269 | if (auto_assem || !inargv) | |
270 | /* Ignore unrecognised devices during auto-assembly */ | |
271 | goto loop; | |
272 | if (ident->uuid_set || ident->name[0] || | |
273 | ident->super_minor != UnSet) | |
274 | /* Ignore unrecognised device if looking for | |
275 | * specific array */ | |
276 | goto loop; | |
277 | ||
278 | pr_err("%s has no superblock - assembly aborted\n", | |
279 | devname); | |
280 | if (st) | |
281 | st->ss->free_super(st); | |
282 | dev_policy_free(pol); | |
283 | domain_free(domains); | |
284 | return -1; | |
285 | } | |
286 | ||
287 | if (found_container) { | |
288 | /* tmpdev is a container. We need to be either | |
289 | * looking for a member, or auto-assembling | |
290 | */ | |
291 | /* should be safe to try an exclusive open now, we | |
292 | * have rejected anything that some other mdadm might | |
293 | * be looking at | |
294 | */ | |
295 | dfd = dev_open(devname, O_RDONLY | O_EXCL); | |
296 | if (dfd < 0) { | |
297 | if (report_mismatch) | |
298 | pr_err("%s is busy - skipping\n", devname); | |
299 | goto loop; | |
300 | } | |
301 | close(dfd); | |
302 | ||
303 | if (ident->container && ident->container[0] != '/') { | |
304 | /* we have a uuid */ | |
305 | int uuid[4]; | |
306 | ||
307 | content = *contentp; | |
308 | tst->ss->getinfo_super(tst, content, NULL); | |
309 | ||
310 | if (!parse_uuid(ident->container, uuid) || | |
311 | !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) { | |
312 | if (report_mismatch) | |
313 | pr_err("%s has wrong UUID to be required container\n", | |
314 | devname); | |
315 | goto loop; | |
316 | } | |
317 | } | |
318 | /* It is worth looking inside this container. | |
319 | */ | |
320 | if (c->verbose > 0) | |
321 | pr_err("looking in container %s\n", | |
322 | devname); | |
323 | ||
324 | for (content = tst->ss->container_content(tst, NULL); | |
325 | content; | |
326 | content = content->next) { | |
327 | ||
328 | if (!ident_matches(ident, content, tst, | |
329 | c->homehost, c->require_homehost, | |
330 | c->update, | |
331 | report_mismatch ? devname : NULL)) | |
332 | /* message already printed */; | |
333 | else if (is_member_busy(content->text_version)) { | |
334 | if (report_mismatch) | |
335 | pr_err("member %s in %s is already assembled\n", | |
336 | content->text_version, | |
337 | devname); | |
338 | } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) { | |
339 | /* do not assemble arrays with unsupported configurations */ | |
340 | pr_err("Cannot activate member %s in %s.\n", | |
341 | content->text_version, | |
342 | devname); | |
343 | } else | |
344 | break; | |
345 | } | |
346 | if (!content) { | |
347 | tmpdev->used = 2; | |
348 | goto loop; /* empty container */ | |
349 | } | |
350 | ||
351 | st = tst; tst = NULL; | |
352 | if (!auto_assem && inargv && tmpdev->next != NULL) { | |
353 | pr_err("%s is a container, but is not only device given: confused and aborting\n", | |
354 | devname); | |
355 | st->ss->free_super(st); | |
356 | dev_policy_free(pol); | |
357 | domain_free(domains); | |
358 | return -1; | |
359 | } | |
360 | if (c->verbose > 0) | |
361 | pr_err("found match on member %s in %s\n", | |
362 | content->text_version, devname); | |
363 | ||
364 | /* make sure we finished the loop */ | |
365 | tmpdev = NULL; | |
366 | goto loop; | |
367 | } else { | |
368 | content = *contentp; | |
369 | tst->ss->getinfo_super(tst, content, NULL); | |
370 | ||
371 | if (!ident_matches(ident, content, tst, | |
372 | c->homehost, c->require_homehost, | |
373 | c->update, | |
374 | report_mismatch ? devname : NULL)) | |
375 | goto loop; | |
376 | ||
377 | if (auto_assem) { | |
378 | /* Never auto-assemble things that conflict | |
379 | * with mdadm.conf in some way | |
380 | */ | |
381 | struct mddev_ident *match; | |
382 | int rv = 0; | |
383 | ||
384 | match = conf_match(tst, content, devname, | |
385 | report_mismatch ? c->verbose : -1, | |
386 | &rv); | |
387 | if (!match && rv == 2) | |
388 | goto loop; | |
389 | if (match && match->devname && | |
390 | strcasecmp(match->devname, "<ignore>") == 0) { | |
391 | if (report_mismatch) | |
392 | pr_err("%s is a member of an explicitly ignored array\n", | |
393 | devname); | |
394 | goto loop; | |
395 | } | |
396 | if (match && !ident_matches(match, content, tst, | |
397 | c->homehost, c->require_homehost, | |
398 | c->update, | |
399 | report_mismatch ? devname : NULL)) | |
400 | /* Array exists in mdadm.conf but some | |
401 | * details don't match, so reject it | |
402 | */ | |
403 | goto loop; | |
404 | } | |
405 | ||
406 | /* should be safe to try an exclusive open now, we | |
407 | * have rejected anything that some other mdadm might | |
408 | * be looking at | |
409 | */ | |
410 | dfd = dev_open(devname, O_RDONLY | O_EXCL); | |
411 | if (dfd < 0) { | |
412 | if (report_mismatch) | |
413 | pr_err("%s is busy - skipping\n", devname); | |
414 | goto loop; | |
415 | } | |
416 | close(dfd); | |
417 | ||
418 | if (st == NULL) | |
419 | st = dup_super(tst); | |
420 | if (st->minor_version == -1) | |
421 | st->minor_version = tst->minor_version; | |
422 | ||
423 | if (memcmp(content->uuid, uuid_zero, | |
424 | sizeof(int[4])) == 0) { | |
425 | /* this is a floating spare. It cannot define | |
426 | * an array unless there are no more arrays of | |
427 | * this type to be found. It can be included | |
428 | * in an array of this type though. | |
429 | */ | |
430 | tmpdev->used = 3; | |
431 | goto loop; | |
432 | } | |
433 | ||
434 | if (st->ss != tst->ss || | |
435 | st->minor_version != tst->minor_version || | |
436 | st->ss->compare_super(st, tst) != 0) { | |
437 | /* Some mismatch. If exactly one array matches this host, | |
438 | * we can resolve on that one. | |
439 | * Or, if we are auto assembling, we just ignore the second | |
440 | * for now. | |
441 | */ | |
442 | if (auto_assem) | |
443 | goto loop; | |
444 | if (c->homehost) { | |
445 | int first = st->ss->match_home(st, c->homehost); | |
446 | int last = tst->ss->match_home(tst, c->homehost); | |
447 | if (first != last && | |
448 | (first == 1 || last == 1)) { | |
449 | /* We can do something */ | |
450 | if (first) {/* just ignore this one */ | |
451 | if (report_mismatch) | |
452 | pr_err("%s misses out due to wrong homehost\n", | |
453 | devname); | |
454 | goto loop; | |
455 | } else { /* reject all those sofar */ | |
456 | struct mddev_dev *td; | |
457 | if (report_mismatch) | |
458 | pr_err("%s overrides previous devices due to good homehost\n", | |
459 | devname); | |
460 | for (td=devlist; td != tmpdev; td=td->next) | |
461 | if (td->used == 1) | |
462 | td->used = 0; | |
463 | tmpdev->used = 1; | |
464 | goto loop; | |
465 | } | |
466 | } | |
467 | } | |
468 | pr_err("superblock on %s doesn't match others - assembly aborted\n", | |
469 | devname); | |
470 | tst->ss->free_super(tst); | |
471 | st->ss->free_super(st); | |
472 | dev_policy_free(pol); | |
473 | domain_free(domains); | |
474 | return -1; | |
475 | } | |
476 | tmpdev->used = 1; | |
477 | } | |
478 | loop: | |
479 | /* Collect domain information from members only */ | |
480 | if (tmpdev && tmpdev->used == 1) { | |
481 | if (!pol) | |
482 | pol = devid_policy(rdev); | |
483 | domain_merge(&domains, pol, tst?tst->ss->name:NULL); | |
484 | } | |
485 | dev_policy_free(pol); | |
486 | pol = NULL; | |
487 | if (tst) | |
488 | tst->ss->free_super(tst); | |
489 | } | |
490 | ||
491 | /* Check if we found some imsm spares but no members */ | |
492 | if ((auto_assem || | |
493 | (ident->uuid_set && | |
494 | memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) && | |
495 | (!st || !st->sb)) | |
496 | for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) { | |
497 | if (tmpdev->used != 3) | |
498 | continue; | |
499 | tmpdev->used = 1; | |
500 | content = *contentp; | |
501 | ||
502 | if (!st->sb) { | |
503 | /* we need sb from one of the spares */ | |
504 | int dfd = dev_open(tmpdev->devname, O_RDONLY); | |
505 | if (dfd < 0 || | |
506 | st->ss->load_super(st, dfd, NULL)) | |
507 | tmpdev->used = 2; | |
508 | if (dfd > 0) | |
509 | close(dfd); | |
510 | } | |
511 | } | |
512 | ||
513 | /* Now reject spares that don't match domains of identified members */ | |
514 | for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) { | |
515 | if (tmpdev->used != 3) | |
516 | continue; | |
517 | if (!stat_is_blkdev(tmpdev->devname, &rdev)) { | |
518 | tmpdev->used = 2; | |
519 | } else { | |
520 | struct dev_policy *pol = devid_policy(rdev); | |
521 | int dt = domain_test(domains, pol, NULL); | |
522 | if (inargv && dt != 0) | |
523 | /* take this spare as domains match | |
524 | * if there are any */ | |
525 | tmpdev->used = 1; | |
526 | else if (!inargv && dt == 1) | |
527 | /* device wasn't explicitly listed, so need | |
528 | * explicit domain match - which we have */ | |
529 | tmpdev->used = 1; | |
530 | else | |
531 | /* if domains don't match mark as unused */ | |
532 | tmpdev->used = 0; | |
533 | dev_policy_free(pol); | |
534 | } | |
535 | } | |
536 | domain_free(domains); | |
537 | *stp = st; | |
538 | if (st && st->sb && content == *contentp) | |
539 | st->ss->getinfo_super(st, content, NULL); | |
540 | *contentp = content; | |
541 | ||
542 | return num_devs; | |
543 | } | |
544 | ||
545 | struct devs { | |
546 | char *devname; | |
547 | int uptodate; /* set once we decide that this device is as | |
548 | * recent as everything else in the array. | |
549 | */ | |
550 | int included; /* set if the device is already in the array | |
551 | * due to a previous '-I' | |
552 | */ | |
553 | struct mdinfo i; | |
554 | }; | |
555 | ||
556 | static int load_devices(struct devs *devices, char *devmap, | |
557 | struct mddev_ident *ident, struct supertype **stp, | |
558 | struct mddev_dev *devlist, struct context *c, | |
559 | struct mdinfo *content, | |
560 | int mdfd, char *mddev, | |
561 | int *most_recentp, int *bestcntp, int **bestp, | |
562 | int inargv) | |
563 | { | |
564 | struct mddev_dev *tmpdev; | |
565 | int devcnt = 0; | |
566 | int nextspare = 0; | |
567 | int bitmap_done = 0; | |
568 | int most_recent = -1; | |
569 | int bestcnt = 0; | |
570 | int *best = *bestp; | |
571 | struct supertype *st = *stp; | |
572 | ||
573 | for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) { | |
574 | char *devname = tmpdev->devname; | |
575 | struct stat stb; | |
576 | struct supertype *tst; | |
577 | int i; | |
578 | int dfd; | |
579 | ||
580 | if (tmpdev->used != 1) | |
581 | continue; | |
582 | /* looks like a good enough match to update the super block if needed */ | |
583 | if (c->update) { | |
584 | /* prepare useful information in info structures */ | |
585 | struct stat stb2; | |
586 | int err; | |
587 | fstat(mdfd, &stb2); | |
588 | ||
589 | if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set) | |
590 | random_uuid((__u8 *)ident->uuid); | |
591 | ||
592 | if (strcmp(c->update, "ppl") == 0 && | |
593 | ident->bitmap_fd >= 0) { | |
594 | pr_err("PPL is not compatible with bitmap\n"); | |
595 | return -1; | |
596 | } | |
597 | ||
598 | dfd = dev_open(devname, | |
599 | tmpdev->disposition == 'I' | |
600 | ? O_RDWR : (O_RDWR|O_EXCL)); | |
601 | ||
602 | tst = dup_super(st); | |
603 | if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) { | |
604 | pr_err("cannot re-read metadata from %s - aborting\n", | |
605 | devname); | |
606 | if (dfd >= 0) | |
607 | close(dfd); | |
608 | close(mdfd); | |
609 | free(devices); | |
610 | free(devmap); | |
611 | tst->ss->free_super(tst); | |
612 | free(tst); | |
613 | *stp = st; | |
614 | return -1; | |
615 | } | |
616 | tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks); | |
617 | ||
618 | memcpy(content->uuid, ident->uuid, 16); | |
619 | strcpy(content->name, ident->name); | |
620 | content->array.md_minor = minor(stb2.st_rdev); | |
621 | ||
622 | if (strcmp(c->update, "byteorder") == 0) | |
623 | err = 0; | |
624 | else if (strcmp(c->update, "home-cluster") == 0) { | |
625 | tst->cluster_name = c->homecluster; | |
626 | err = tst->ss->write_bitmap(tst, dfd, NameUpdate); | |
627 | } else if (strcmp(c->update, "nodes") == 0) { | |
628 | tst->nodes = c->nodes; | |
629 | err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate); | |
630 | } else if (strcmp(c->update, "revert-reshape") == 0 && | |
631 | c->invalid_backup) | |
632 | err = tst->ss->update_super(tst, content, | |
633 | "revert-reshape-nobackup", | |
634 | devname, c->verbose, | |
635 | ident->uuid_set, | |
636 | c->homehost); | |
637 | else | |
638 | err = tst->ss->update_super(tst, content, c->update, | |
639 | devname, c->verbose, | |
640 | ident->uuid_set, | |
641 | c->homehost); | |
642 | if (err < 0) { | |
643 | if (err == -1) | |
644 | pr_err("--update=%s not understood for %s metadata\n", | |
645 | c->update, tst->ss->name); | |
646 | tst->ss->free_super(tst); | |
647 | free(tst); | |
648 | close(mdfd); | |
649 | close(dfd); | |
650 | free(devices); | |
651 | free(devmap); | |
652 | *stp = st; | |
653 | return -1; | |
654 | } | |
655 | if (strcmp(c->update, "uuid")==0 && | |
656 | !ident->uuid_set) { | |
657 | ident->uuid_set = 1; | |
658 | memcpy(ident->uuid, content->uuid, 16); | |
659 | } | |
660 | if (tst->ss->store_super(tst, dfd)) | |
661 | pr_err("Could not re-write superblock on %s.\n", | |
662 | devname); | |
663 | ||
664 | if (strcmp(c->update, "uuid")==0 && | |
665 | ident->bitmap_fd >= 0 && !bitmap_done) { | |
666 | if (bitmap_update_uuid(ident->bitmap_fd, | |
667 | content->uuid, | |
668 | tst->ss->swapuuid) != 0) | |
669 | pr_err("Could not update uuid on external bitmap.\n"); | |
670 | else | |
671 | bitmap_done = 1; | |
672 | } | |
673 | } else { | |
674 | dfd = dev_open(devname, | |
675 | tmpdev->disposition == 'I' | |
676 | ? O_RDWR : (O_RDWR|O_EXCL)); | |
677 | tst = dup_super(st); | |
678 | ||
679 | if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) { | |
680 | pr_err("cannot re-read metadata from %s - aborting\n", | |
681 | devname); | |
682 | if (dfd >= 0) | |
683 | close(dfd); | |
684 | close(mdfd); | |
685 | free(devices); | |
686 | free(devmap); | |
687 | tst->ss->free_super(tst); | |
688 | free(tst); | |
689 | *stp = st; | |
690 | return -1; | |
691 | } | |
692 | tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks); | |
693 | } | |
694 | ||
695 | fstat(dfd, &stb); | |
696 | close(dfd); | |
697 | ||
698 | if (c->verbose > 0) | |
699 | pr_err("%s is identified as a member of %s, slot %d%s.\n", | |
700 | devname, mddev, content->disk.raid_disk, | |
701 | (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":""); | |
702 | devices[devcnt].devname = devname; | |
703 | devices[devcnt].uptodate = 0; | |
704 | devices[devcnt].included = (tmpdev->disposition == 'I'); | |
705 | devices[devcnt].i = *content; | |
706 | devices[devcnt].i.disk.major = major(stb.st_rdev); | |
707 | devices[devcnt].i.disk.minor = minor(stb.st_rdev); | |
708 | ||
709 | if (devices[devcnt].i.disk.state == 6) { | |
710 | if (most_recent < 0 || | |
711 | devices[devcnt].i.events | |
712 | > devices[most_recent].i.events) { | |
713 | struct supertype *tmp = tst; | |
714 | tst = st; | |
715 | st = tmp; | |
716 | most_recent = devcnt; | |
717 | } | |
718 | } | |
719 | tst->ss->free_super(tst); | |
720 | free(tst); | |
721 | ||
722 | if (content->array.level == LEVEL_MULTIPATH) | |
723 | /* with multipath, the raid_disk from the superblock is meaningless */ | |
724 | i = devcnt; | |
725 | else | |
726 | i = devices[devcnt].i.disk.raid_disk; | |
727 | if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) { | |
728 | if (nextspare < content->array.raid_disks*2) | |
729 | nextspare = content->array.raid_disks*2; | |
730 | i = nextspare++; | |
731 | } else { | |
732 | /* i is raid_disk - double it so there is room for | |
733 | * replacements */ | |
734 | i *= 2; | |
735 | if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT)) | |
736 | i++; | |
737 | if (i >= content->array.raid_disks*2 && | |
738 | i >= nextspare) | |
739 | nextspare = i+1; | |
740 | } | |
741 | if (i < 10000) { | |
742 | if (i >= bestcnt) { | |
743 | int newbestcnt = i+10; | |
744 | int *newbest = xmalloc(sizeof(int)*newbestcnt); | |
745 | int c; | |
746 | for (c=0; c < newbestcnt; c++) | |
747 | if (c < bestcnt) | |
748 | newbest[c] = best[c]; | |
749 | else | |
750 | newbest[c] = -1; | |
751 | if (best)free(best); | |
752 | best = newbest; | |
753 | bestcnt = newbestcnt; | |
754 | } | |
755 | if (best[i] >=0 && | |
756 | devices[best[i]].i.events == | |
757 | devices[devcnt].i.events && | |
758 | (devices[best[i]].i.disk.minor != | |
759 | devices[devcnt].i.disk.minor) && | |
760 | st->ss == &super0 && | |
761 | content->array.level != LEVEL_MULTIPATH) { | |
762 | /* two different devices with identical superblock. | |
763 | * Could be a mis-detection caused by overlapping | |
764 | * partitions. fail-safe. | |
765 | */ | |
766 | pr_err("WARNING %s and %s appear to have very similar superblocks.\n" | |
767 | " If they are really different, please --zero the superblock on one\n" | |
768 | " If they are the same or overlap, please remove one from %s.\n", | |
769 | devices[best[i]].devname, devname, | |
770 | inargv ? "the list" : | |
771 | "the\n DEVICE list in mdadm.conf" | |
772 | ); | |
773 | close(mdfd); | |
774 | free(devices); | |
775 | free(devmap); | |
776 | *stp = st; | |
777 | return -1; | |
778 | } | |
779 | if (best[i] == -1 || (devices[best[i]].i.events | |
780 | < devices[devcnt].i.events)) | |
781 | best[i] = devcnt; | |
782 | } | |
783 | devcnt++; | |
784 | } | |
785 | if (most_recent >= 0) | |
786 | *most_recentp = most_recent; | |
787 | *bestcntp = bestcnt; | |
788 | *bestp = best; | |
789 | *stp = st; | |
790 | return devcnt; | |
791 | } | |
792 | ||
793 | static int force_array(struct mdinfo *content, | |
794 | struct devs *devices, | |
795 | int *best, int bestcnt, char *avail, | |
796 | int most_recent, | |
797 | struct supertype *st, | |
798 | struct context *c) | |
799 | { | |
800 | int okcnt = 0; | |
801 | while (!enough(content->array.level, content->array.raid_disks, | |
802 | content->array.layout, 1, | |
803 | avail) || | |
804 | (content->reshape_active && content->delta_disks > 0 && | |
805 | !enough(content->array.level, (content->array.raid_disks | |
806 | - content->delta_disks), | |
807 | content->new_layout, 1, avail))) { | |
808 | /* Choose the newest best drive which is | |
809 | * not up-to-date, update the superblock | |
810 | * and add it. | |
811 | */ | |
812 | int fd; | |
813 | struct supertype *tst; | |
814 | unsigned long long current_events; | |
815 | int chosen_drive = -1; | |
816 | int i; | |
817 | ||
818 | for (i = 0; | |
819 | i < content->array.raid_disks * 2 && i < bestcnt; | |
820 | i += 2) { | |
821 | int j = best[i]; | |
822 | if (j < 0) | |
823 | continue; | |
824 | if (devices[j].uptodate) | |
825 | continue; | |
826 | if (devices[j].i.recovery_start != MaxSector) { | |
827 | int delta; | |
828 | if (!devices[j].i.reshape_active || | |
829 | devices[j].i.delta_disks <= 0) | |
830 | continue; | |
831 | /* When increasing number of devices, an | |
832 | * added device also appears to be | |
833 | * recovering. It is safe to include it | |
834 | * as long as it won't be a source of | |
835 | * data. | |
836 | * For now, just allow for last data | |
837 | * devices in RAID4 or last devices in RAID4/5/6. | |
838 | */ | |
839 | delta = devices[j].i.delta_disks; | |
840 | if (devices[j].i.array.level >= 4 && | |
841 | devices[j].i.array.level <= 6 && | |
842 | i/2 >= content->array.raid_disks - delta) | |
843 | /* OK */; | |
844 | else if (devices[j].i.array.level == 4 && | |
845 | i/2 >= content->array.raid_disks - delta - 1) | |
846 | /* OK */; | |
847 | else | |
848 | continue; | |
849 | } | |
850 | if (chosen_drive < 0 || | |
851 | devices[j].i.events | |
852 | > devices[chosen_drive].i.events) | |
853 | chosen_drive = j; | |
854 | } | |
855 | if (chosen_drive < 0) | |
856 | break; | |
857 | current_events = devices[chosen_drive].i.events; | |
858 | add_another: | |
859 | if (c->verbose >= 0) | |
860 | pr_err("forcing event count in %s(%d) from %d upto %d\n", | |
861 | devices[chosen_drive].devname, | |
862 | devices[chosen_drive].i.disk.raid_disk, | |
863 | (int)(devices[chosen_drive].i.events), | |
864 | (int)(devices[most_recent].i.events)); | |
865 | fd = dev_open(devices[chosen_drive].devname, | |
866 | devices[chosen_drive].included ? O_RDWR | |
867 | : (O_RDWR|O_EXCL)); | |
868 | if (fd < 0) { | |
869 | pr_err("Couldn't open %s for write - not updating\n", | |
870 | devices[chosen_drive].devname); | |
871 | devices[chosen_drive].i.events = 0; | |
872 | continue; | |
873 | } | |
874 | tst = dup_super(st); | |
875 | if (tst->ss->load_super(tst,fd, NULL)) { | |
876 | close(fd); | |
877 | pr_err("RAID superblock disappeared from %s - not updating.\n", | |
878 | devices[chosen_drive].devname); | |
879 | devices[chosen_drive].i.events = 0; | |
880 | continue; | |
881 | } | |
882 | content->events = devices[most_recent].i.events; | |
883 | tst->ss->update_super(tst, content, "force-one", | |
884 | devices[chosen_drive].devname, c->verbose, | |
885 | 0, NULL); | |
886 | ||
887 | if (tst->ss->store_super(tst, fd)) { | |
888 | close(fd); | |
889 | pr_err("Could not re-write superblock on %s\n", | |
890 | devices[chosen_drive].devname); | |
891 | devices[chosen_drive].i.events = 0; | |
892 | tst->ss->free_super(tst); | |
893 | continue; | |
894 | } | |
895 | close(fd); | |
896 | devices[chosen_drive].i.events = devices[most_recent].i.events; | |
897 | devices[chosen_drive].uptodate = 1; | |
898 | avail[chosen_drive] = 1; | |
899 | okcnt++; | |
900 | tst->ss->free_super(tst); | |
901 | /* If there are any other drives of the same vintage, | |
902 | * add them in as well. We can't lose and we might gain | |
903 | */ | |
904 | for (i = 0; | |
905 | i < content->array.raid_disks * 2 && i < bestcnt ; | |
906 | i += 2) { | |
907 | int j = best[i]; | |
908 | if (j >= 0 && | |
909 | !devices[j].uptodate && | |
910 | devices[j].i.recovery_start == MaxSector && | |
911 | devices[j].i.events == current_events) { | |
912 | chosen_drive = j; | |
913 | goto add_another; | |
914 | } | |
915 | } | |
916 | } | |
917 | return okcnt; | |
918 | } | |
919 | ||
920 | static int start_array(int mdfd, | |
921 | char *mddev, | |
922 | struct mdinfo *content, | |
923 | struct supertype *st, | |
924 | struct mddev_ident *ident, | |
925 | int *best, int bestcnt, | |
926 | int chosen_drive, | |
927 | struct devs *devices, | |
928 | unsigned int okcnt, | |
929 | unsigned int sparecnt, | |
930 | unsigned int rebuilding_cnt, | |
931 | unsigned int journalcnt, | |
932 | struct context *c, | |
933 | int clean, char *avail, | |
934 | int start_partial_ok, | |
935 | int err_ok, | |
936 | int was_forced | |
937 | ) | |
938 | { | |
939 | int rv; | |
940 | int i; | |
941 | unsigned int req_cnt; | |
942 | ||
943 | if (content->journal_device_required && (content->journal_clean == 0)) { | |
944 | if (!c->force) { | |
945 | pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n"); | |
946 | return 1; | |
947 | } | |
948 | pr_err("Journal is missing or stale, starting array read only.\n"); | |
949 | c->readonly = 1; | |
950 | } | |
951 | ||
952 | if (content->consistency_policy == CONSISTENCY_POLICY_PPL) | |
953 | clean = 1; | |
954 | ||
955 | rv = set_array_info(mdfd, st, content); | |
956 | if (rv && !err_ok) { | |
957 | pr_err("failed to set array info for %s: %s\n", | |
958 | mddev, strerror(errno)); | |
959 | return 1; | |
960 | } | |
961 | if (ident->bitmap_fd >= 0) { | |
962 | if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) { | |
963 | pr_err("SET_BITMAP_FILE failed.\n"); | |
964 | return 1; | |
965 | } | |
966 | } else if (ident->bitmap_file) { | |
967 | /* From config file */ | |
968 | int bmfd = open(ident->bitmap_file, O_RDWR); | |
969 | if (bmfd < 0) { | |
970 | pr_err("Could not open bitmap file %s\n", | |
971 | ident->bitmap_file); | |
972 | return 1; | |
973 | } | |
974 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
975 | pr_err("Failed to set bitmapfile for %s\n", mddev); | |
976 | close(bmfd); | |
977 | return 1; | |
978 | } | |
979 | close(bmfd); | |
980 | } | |
981 | ||
982 | /* First, add the raid disks, but add the chosen one last */ | |
983 | for (i = 0; i <= bestcnt; i++) { | |
984 | int j; | |
985 | if (i < bestcnt) { | |
986 | j = best[i]; | |
987 | if (j == chosen_drive) | |
988 | continue; | |
989 | } else | |
990 | j = chosen_drive; | |
991 | ||
992 | if (j >= 0 && !devices[j].included) { | |
993 | int dfd; | |
994 | ||
995 | dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL); | |
996 | if (dfd >= 0) { | |
997 | remove_partitions(dfd); | |
998 | close(dfd); | |
999 | } | |
1000 | rv = add_disk(mdfd, st, content, &devices[j].i); | |
1001 | ||
1002 | if (rv) { | |
1003 | pr_err("failed to add %s to %s: %s\n", | |
1004 | devices[j].devname, mddev, | |
1005 | strerror(errno)); | |
1006 | if (i < content->array.raid_disks * 2 || | |
1007 | i == bestcnt) | |
1008 | okcnt--; | |
1009 | else | |
1010 | sparecnt--; | |
1011 | } else if (c->verbose > 0) { | |
1012 | pr_err("added %s to %s as %d%s%s\n", | |
1013 | devices[j].devname, mddev, | |
1014 | devices[j].i.disk.raid_disk, | |
1015 | devices[j].uptodate?"": | |
1016 | " (possibly out of date)", | |
1017 | (devices[j].i.disk.state & | |
1018 | (1<<MD_DISK_REPLACEMENT)) ? | |
1019 | " replacement":""); | |
1020 | } | |
1021 | } else if (j >= 0) { | |
1022 | if (c->verbose > 0) | |
1023 | pr_err("%s is already in %s as %d\n", | |
1024 | devices[j].devname, mddev, | |
1025 | devices[j].i.disk.raid_disk); | |
1026 | } else if (c->verbose > 0 && | |
1027 | i < content->array.raid_disks * 2 && (i & 1) == 0) | |
1028 | pr_err("no uptodate device for slot %d of %s\n", | |
1029 | i/2, mddev); | |
1030 | } | |
1031 | ||
1032 | if (content->array.level == LEVEL_CONTAINER) { | |
1033 | if (c->verbose >= 0) { | |
1034 | pr_err("Container %s has been assembled with %d drive%s", | |
1035 | mddev, okcnt + sparecnt + journalcnt, | |
1036 | okcnt + sparecnt + journalcnt == 1 ? "" : "s"); | |
1037 | if (okcnt < (unsigned)content->array.raid_disks) | |
1038 | fprintf(stderr, " (out of %d)", | |
1039 | content->array.raid_disks); | |
1040 | fprintf(stderr, "\n"); | |
1041 | } | |
1042 | ||
1043 | if (st->ss->validate_container) { | |
1044 | struct mdinfo *devices_list; | |
1045 | struct mdinfo *info_devices; | |
1046 | unsigned int count; | |
1047 | ||
1048 | devices_list = NULL; | |
1049 | info_devices = xmalloc(sizeof(struct mdinfo) * | |
1050 | (okcnt + sparecnt)); | |
1051 | for (count = 0; count < okcnt + sparecnt; count++) { | |
1052 | info_devices[count] = devices[count].i; | |
1053 | info_devices[count].next = devices_list; | |
1054 | devices_list = &info_devices[count]; | |
1055 | } | |
1056 | if (st->ss->validate_container(devices_list)) | |
1057 | pr_err("Mismatch detected!\n"); | |
1058 | free(info_devices); | |
1059 | } | |
1060 | ||
1061 | st->ss->free_super(st); | |
1062 | sysfs_uevent(content, "change"); | |
1063 | if (err_ok && okcnt < (unsigned)content->array.raid_disks) | |
1064 | /* Was partial, is still partial, so signal an error | |
1065 | * to ensure we don't retry */ | |
1066 | return 1; | |
1067 | return 0; | |
1068 | } | |
1069 | ||
1070 | /* Get number of in-sync devices according to the superblock. | |
1071 | * We must have this number to start the array without -s or -R | |
1072 | */ | |
1073 | req_cnt = content->array.working_disks; | |
1074 | ||
1075 | if (c->runstop == 1 || | |
1076 | (c->runstop <= 0 && | |
1077 | (enough(content->array.level, content->array.raid_disks, | |
1078 | content->array.layout, clean, avail) && | |
1079 | (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) { | |
1080 | /* This array is good-to-go. | |
1081 | * If a reshape is in progress then we might need to | |
1082 | * continue monitoring it. In that case we start | |
1083 | * it read-only and let the grow code make it writable. | |
1084 | */ | |
1085 | int rv; | |
1086 | ||
1087 | if (content->reshape_active && | |
1088 | !(content->reshape_active & RESHAPE_NO_BACKUP) && | |
1089 | content->delta_disks <= 0) { | |
1090 | if (!c->backup_file) { | |
1091 | pr_err("%s: Need a backup file to complete reshape of this array.\n", | |
1092 | mddev); | |
1093 | pr_err("Please provided one with \"--backup-file=...\"\n"); | |
1094 | if (c->update && | |
1095 | strcmp(c->update, "revert-reshape") == 0) | |
1096 | pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n"); | |
1097 | return 1; | |
1098 | } | |
1099 | rv = sysfs_set_str(content, NULL, | |
1100 | "array_state", "readonly"); | |
1101 | if (rv == 0) | |
1102 | rv = Grow_continue(mdfd, st, content, | |
1103 | c->backup_file, 0, | |
1104 | c->freeze_reshape); | |
1105 | } else if (c->readonly && | |
1106 | sysfs_attribute_available(content, NULL, | |
1107 | "array_state")) { | |
1108 | rv = sysfs_set_str(content, NULL, | |
1109 | "array_state", "readonly"); | |
1110 | } else | |
1111 | rv = ioctl(mdfd, RUN_ARRAY, NULL); | |
1112 | reopen_mddev(mdfd); /* drop O_EXCL */ | |
1113 | if (rv == 0) { | |
1114 | if (c->verbose >= 0) { | |
1115 | pr_err("%s has been started with %d drive%s", | |
1116 | mddev, okcnt, okcnt==1?"":"s"); | |
1117 | if (okcnt < (unsigned)content->array.raid_disks) | |
1118 | fprintf(stderr, " (out of %d)", | |
1119 | content->array.raid_disks); | |
1120 | if (rebuilding_cnt) | |
1121 | fprintf(stderr, "%s %d rebuilding", | |
1122 | sparecnt?",":" and", | |
1123 | rebuilding_cnt); | |
1124 | if (sparecnt) | |
1125 | fprintf(stderr, " and %d spare%s", | |
1126 | sparecnt, | |
1127 | sparecnt == 1 ? "" : "s"); | |
1128 | if (content->journal_clean) | |
1129 | fprintf(stderr, " and %d journal", | |
1130 | journalcnt); | |
1131 | fprintf(stderr, ".\n"); | |
1132 | } | |
1133 | if (content->reshape_active && | |
1134 | content->array.level >= 4 && | |
1135 | content->array.level <= 6) { | |
1136 | /* might need to increase the size | |
1137 | * of the stripe cache - default is 256 | |
1138 | */ | |
1139 | int chunk_size = content->array.chunk_size; | |
1140 | ||
1141 | if (content->reshape_active && | |
1142 | content->new_chunk > chunk_size) | |
1143 | chunk_size = content->new_chunk; | |
1144 | if (256 < 4 * ((chunk_size+4065)/4096)) { | |
1145 | struct mdinfo *sra; | |
1146 | ||
1147 | sra = sysfs_read(mdfd, NULL, 0); | |
1148 | if (sra) | |
1149 | sysfs_set_num(sra, NULL, | |
1150 | "stripe_cache_size", | |
1151 | (4 * chunk_size / 4096) + 1); | |
1152 | sysfs_free(sra); | |
1153 | } | |
1154 | } | |
1155 | if (okcnt < (unsigned)content->array.raid_disks) { | |
1156 | /* If any devices did not get added | |
1157 | * because the kernel rejected them based | |
1158 | * on event count, try adding them | |
1159 | * again providing the action policy is | |
1160 | * 're-add' or greater. The bitmap | |
1161 | * might allow them to be included, or | |
1162 | * they will become spares. | |
1163 | */ | |
1164 | for (i = 0; i < bestcnt; i++) { | |
1165 | int j = best[i]; | |
1166 | if (j >= 0 && !devices[j].uptodate) { | |
1167 | if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add)) | |
1168 | continue; | |
1169 | rv = add_disk(mdfd, st, content, | |
1170 | &devices[j].i); | |
1171 | if (rv == 0 && c->verbose >= 0) | |
1172 | pr_err("%s has been re-added.\n", | |
1173 | devices[j].devname); | |
1174 | } | |
1175 | } | |
1176 | } | |
1177 | if (content->array.level == 6 && | |
1178 | okcnt + 1 == (unsigned)content->array.raid_disks && | |
1179 | was_forced) { | |
1180 | struct mdinfo *sra; | |
1181 | ||
1182 | sra = sysfs_read(mdfd, NULL, 0); | |
1183 | if (sra) | |
1184 | sysfs_set_str(sra, NULL, | |
1185 | "sync_action", "repair"); | |
1186 | sysfs_free(sra); | |
1187 | } | |
1188 | return 0; | |
1189 | } | |
1190 | pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno)); | |
1191 | ||
1192 | if (!enough(content->array.level, content->array.raid_disks, | |
1193 | content->array.layout, 1, avail)) | |
1194 | pr_err("Not enough devices to start the array.\n"); | |
1195 | else if (!enough(content->array.level, | |
1196 | content->array.raid_disks, | |
1197 | content->array.layout, clean, avail)) | |
1198 | pr_err("Not enough devices to start the array while not clean - consider --force.\n"); | |
1199 | ||
1200 | return 1; | |
1201 | } | |
1202 | if (c->runstop == -1) { | |
1203 | pr_err("%s assembled from %d drive%s", | |
1204 | mddev, okcnt, okcnt == 1 ? "" : "s"); | |
1205 | if (okcnt != (unsigned)content->array.raid_disks) | |
1206 | fprintf(stderr, " (out of %d)", | |
1207 | content->array.raid_disks); | |
1208 | fprintf(stderr, ", but not started.\n"); | |
1209 | return 2; | |
1210 | } | |
1211 | if (c->verbose >= -1) { | |
1212 | pr_err("%s assembled from %d drive%s", | |
1213 | mddev, okcnt, okcnt == 1 ? "" : "s"); | |
1214 | if (rebuilding_cnt) | |
1215 | fprintf(stderr, "%s %d rebuilding", | |
1216 | sparecnt ? "," : " and", rebuilding_cnt); | |
1217 | if (sparecnt) | |
1218 | fprintf(stderr, " and %d spare%s", sparecnt, | |
1219 | sparecnt == 1 ? "" : "s"); | |
1220 | if (!enough(content->array.level, content->array.raid_disks, | |
1221 | content->array.layout, 1, avail)) | |
1222 | fprintf(stderr, " - not enough to start the array.\n"); | |
1223 | else if (!enough(content->array.level, | |
1224 | content->array.raid_disks, | |
1225 | content->array.layout, clean, avail)) | |
1226 | fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n"); | |
1227 | else { | |
1228 | if (req_cnt == (unsigned)content->array.raid_disks) | |
1229 | fprintf(stderr, " - need all %d to start it", | |
1230 | req_cnt); | |
1231 | else | |
1232 | fprintf(stderr, " - need %d to start", req_cnt); | |
1233 | fprintf(stderr, " (use --run to insist).\n"); | |
1234 | } | |
1235 | } | |
1236 | return 1; | |
1237 | } | |
1238 | ||
1239 | int Assemble(struct supertype *st, char *mddev, | |
1240 | struct mddev_ident *ident, | |
1241 | struct mddev_dev *devlist, | |
1242 | struct context *c) | |
1243 | { | |
1244 | /* | |
1245 | * The task of Assemble is to find a collection of | |
1246 | * devices that should (according to their superblocks) | |
1247 | * form an array, and to give this collection to the MD driver. | |
1248 | * In Linux-2.4 and later, this involves submitting a | |
1249 | * SET_ARRAY_INFO ioctl with no arg - to prepare | |
1250 | * the array - and then submit a number of | |
1251 | * ADD_NEW_DISK ioctls to add disks into | |
1252 | * the array. Finally RUN_ARRAY might | |
1253 | * be submitted to start the array. | |
1254 | * | |
1255 | * Much of the work of Assemble is in finding and/or | |
1256 | * checking the disks to make sure they look right. | |
1257 | * | |
1258 | * If mddev is not set, then scan must be set and we | |
1259 | * read through the config file for dev+uuid mapping | |
1260 | * We recurse, setting mddev, for each device that | |
1261 | * - isn't running | |
1262 | * - has a valid uuid (or any uuid if !uuidset) | |
1263 | * | |
1264 | * If mddev is set, we try to determine state of md. | |
1265 | * check version - must be at least 0.90.0 | |
1266 | * check kernel version. must be at least 2.4. | |
1267 | * If not, we can possibly fall back on START_ARRAY | |
1268 | * Try to GET_ARRAY_INFO. | |
1269 | * If possible, give up | |
1270 | * If not, try to STOP_ARRAY just to make sure | |
1271 | * | |
1272 | * If !uuidset and scan, look in conf-file for uuid | |
1273 | * If not found, give up | |
1274 | * If !devlist and scan and uuidset, get list of devs from conf-file | |
1275 | * | |
1276 | * For each device: | |
1277 | * Check superblock - discard if bad | |
1278 | * Check uuid (set if we don't have one) - discard if no match | |
1279 | * Check superblock similarity if we have a superblock - discard if different | |
1280 | * Record events, devicenum | |
1281 | * This should give us a list of devices for the array | |
1282 | * We should collect the most recent event number | |
1283 | * | |
1284 | * Count disks with recent enough event count | |
1285 | * While force && !enough disks | |
1286 | * Choose newest rejected disks, update event count | |
1287 | * mark clean and rewrite superblock | |
1288 | * If recent kernel: | |
1289 | * SET_ARRAY_INFO | |
1290 | * foreach device with recent events : ADD_NEW_DISK | |
1291 | * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY | |
1292 | * If old kernel: | |
1293 | * Check the device numbers in superblock are right | |
1294 | * update superblock if any changes | |
1295 | * START_ARRAY | |
1296 | * | |
1297 | */ | |
1298 | int rv; | |
1299 | int mdfd; | |
1300 | int clean; | |
1301 | int auto_assem = (mddev == NULL && !ident->uuid_set && | |
1302 | ident->super_minor == UnSet && ident->name[0] == 0 && | |
1303 | (ident->container == NULL || ident->member == NULL)); | |
1304 | struct devs *devices; | |
1305 | char *devmap; | |
1306 | int *best = NULL; /* indexed by raid_disk */ | |
1307 | int bestcnt = 0; | |
1308 | int devcnt; | |
1309 | unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt; | |
1310 | int journal_clean = 0; | |
1311 | int i; | |
1312 | int was_forced = 0; | |
1313 | int most_recent = 0; | |
1314 | int chosen_drive; | |
1315 | int change = 0; | |
1316 | int inargv = 0; | |
1317 | int start_partial_ok = (c->runstop >= 0) && | |
1318 | (c->force || devlist==NULL || auto_assem); | |
1319 | int num_devs; | |
1320 | struct mddev_dev *tmpdev; | |
1321 | struct mdinfo info; | |
1322 | struct mdinfo *content = NULL; | |
1323 | struct mdinfo *pre_exist = NULL; | |
1324 | char *avail; | |
1325 | char *name = NULL; | |
1326 | char chosen_name[1024]; | |
1327 | struct map_ent *map = NULL; | |
1328 | struct map_ent *mp; | |
1329 | ||
1330 | /* | |
1331 | * If any subdevs are listed, then any that don't | |
1332 | * match ident are discarded. Remainder must all match and | |
1333 | * become the array. | |
1334 | * If no subdevs, then we scan all devices in the config file, but | |
1335 | * there must be something in the identity | |
1336 | */ | |
1337 | ||
1338 | if (!devlist && | |
1339 | ident->uuid_set == 0 && | |
1340 | (ident->super_minor < 0 || ident->super_minor == UnSet) && | |
1341 | ident->name[0] == 0 && | |
1342 | (ident->container == NULL || ident->member == NULL) && | |
1343 | ident->devices == NULL) { | |
1344 | pr_err("No identity information available for %s - cannot assemble.\n", | |
1345 | mddev ? mddev : "further assembly"); | |
1346 | return 1; | |
1347 | } | |
1348 | ||
1349 | if (devlist == NULL) | |
1350 | devlist = conf_get_devs(); | |
1351 | else if (mddev) | |
1352 | inargv = 1; | |
1353 | ||
1354 | try_again: | |
1355 | /* We come back here when doing auto-assembly and attempting some | |
1356 | * set of devices failed. Those are now marked as ->used==2 and | |
1357 | * we ignore them and try again | |
1358 | */ | |
1359 | if (!st && ident->st) | |
1360 | st = ident->st; | |
1361 | if (c->verbose>0) | |
1362 | pr_err("looking for devices for %s\n", | |
1363 | mddev ? mddev : "further assembly"); | |
1364 | ||
1365 | content = &info; | |
1366 | if (st && c->force) | |
1367 | st->ignore_hw_compat = 1; | |
1368 | num_devs = select_devices(devlist, ident, &st, &content, c, | |
1369 | inargv, auto_assem); | |
1370 | if (num_devs < 0) | |
1371 | return 1; | |
1372 | ||
1373 | if (!st || !st->sb || !content) | |
1374 | return 2; | |
1375 | ||
1376 | /* We have a full set of devices - we now need to find the | |
1377 | * array device. | |
1378 | * However there is a risk that we are racing with "mdadm -I" | |
1379 | * and the array is already partially assembled - we will have | |
1380 | * rejected any devices already in this address. | |
1381 | * So we take a lock on the map file - to prevent further races - | |
1382 | * and look for the uuid in there. If found and the array is | |
1383 | * active, we abort. If found and the array is not active | |
1384 | * we commit to that md device and add all the contained devices | |
1385 | * to our list. We flag them so that we don't try to re-add, | |
1386 | * but can remove if they turn out to not be wanted. | |
1387 | */ | |
1388 | if (map_lock(&map)) | |
1389 | pr_err("failed to get exclusive lock on mapfile - continue anyway...\n"); | |
1390 | if (c->update && strcmp(c->update,"uuid") == 0) | |
1391 | mp = NULL; | |
1392 | else | |
1393 | mp = map_by_uuid(&map, content->uuid); | |
1394 | if (mp) { | |
1395 | struct mdinfo *dv; | |
1396 | /* array already exists. */ | |
1397 | pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS); | |
1398 | if (pre_exist->array.level != UnSet) { | |
1399 | pr_err("Found some drive for an array that is already active: %s\n", | |
1400 | mp->path); | |
1401 | pr_err("giving up.\n"); | |
1402 | return 1; | |
1403 | } | |
1404 | for (dv = pre_exist->devs; dv; dv = dv->next) { | |
1405 | /* We want to add this device to our list, | |
1406 | * but it could already be there if "mdadm -I" | |
1407 | * started *after* we checked for O_EXCL. | |
1408 | * If we add it to the top of the list | |
1409 | * it will be preferred over later copies. | |
1410 | */ | |
1411 | struct mddev_dev *newdev; | |
1412 | char *devname = map_dev(dv->disk.major, | |
1413 | dv->disk.minor, | |
1414 | 0); | |
1415 | if (!devname) | |
1416 | continue; | |
1417 | newdev = xmalloc(sizeof(*newdev)); | |
1418 | newdev->devname = devname; | |
1419 | newdev->disposition = 'I'; | |
1420 | newdev->used = 1; | |
1421 | newdev->next = devlist; | |
1422 | devlist = newdev; | |
1423 | num_devs++; | |
1424 | } | |
1425 | strcpy(chosen_name, mp->path); | |
1426 | if (c->verbose > 0 || mddev == NULL || | |
1427 | strcmp(mddev, chosen_name) != 0) | |
1428 | pr_err("Merging with already-assembled %s\n", | |
1429 | chosen_name); | |
1430 | mdfd = open_dev_excl(mp->devnm); | |
1431 | } else { | |
1432 | int trustworthy = FOREIGN; | |
1433 | name = content->name; | |
1434 | switch (st->ss->match_home(st, c->homehost) | |
1435 | ?: st->ss->match_home(st, "any")) { | |
1436 | case 1: | |
1437 | trustworthy = LOCAL; | |
1438 | name = strchr(content->name, ':'); | |
1439 | if (name) | |
1440 | name++; | |
1441 | else | |
1442 | name = content->name; | |
1443 | break; | |
1444 | } | |
1445 | if (!auto_assem) | |
1446 | /* If the array is listed in mdadm.conf or on | |
1447 | * command line, then we trust the name | |
1448 | * even if the array doesn't look local | |
1449 | */ | |
1450 | trustworthy = LOCAL; | |
1451 | ||
1452 | if (name[0] == 0 && | |
1453 | content->array.level == LEVEL_CONTAINER) { | |
1454 | name = content->text_version; | |
1455 | trustworthy = METADATA; | |
1456 | } | |
1457 | ||
1458 | if (name[0] && trustworthy != LOCAL && | |
1459 | ! c->require_homehost && | |
1460 | conf_name_is_free(name)) | |
1461 | trustworthy = LOCAL; | |
1462 | ||
1463 | if (trustworthy == LOCAL && | |
1464 | strchr(name, ':')) | |
1465 | /* Ignore 'host:' prefix of name */ | |
1466 | name = strchr(name, ':')+1; | |
1467 | ||
1468 | mdfd = create_mddev(mddev, name, ident->autof, trustworthy, | |
1469 | chosen_name, 0); | |
1470 | } | |
1471 | if (mdfd < 0) { | |
1472 | st->ss->free_super(st); | |
1473 | if (auto_assem) | |
1474 | goto try_again; | |
1475 | return 1; | |
1476 | } | |
1477 | mddev = chosen_name; | |
1478 | if (pre_exist == NULL) { | |
1479 | if (mddev_busy(fd2devnm(mdfd))) { | |
1480 | pr_err("%s already active, cannot restart it!\n", | |
1481 | mddev); | |
1482 | for (tmpdev = devlist ; | |
1483 | tmpdev && tmpdev->used != 1; | |
1484 | tmpdev = tmpdev->next) | |
1485 | ; | |
1486 | if (tmpdev && auto_assem) | |
1487 | pr_err("%s needed for %s...\n", | |
1488 | mddev, tmpdev->devname); | |
1489 | close(mdfd); | |
1490 | mdfd = -3; | |
1491 | st->ss->free_super(st); | |
1492 | if (auto_assem) | |
1493 | goto try_again; | |
1494 | return 1; | |
1495 | } | |
1496 | /* just incase it was started but has no content */ | |
1497 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1498 | } | |
1499 | ||
1500 | if (content != &info) { | |
1501 | /* This is a member of a container. Try starting the array. */ | |
1502 | int err; | |
1503 | err = assemble_container_content(st, mdfd, content, c, | |
1504 | chosen_name, NULL); | |
1505 | close(mdfd); | |
1506 | return err; | |
1507 | } | |
1508 | ||
1509 | /* Ok, no bad inconsistancy, we can try updating etc */ | |
1510 | devices = xcalloc(num_devs, sizeof(*devices)); | |
1511 | devmap = xcalloc(num_devs, content->array.raid_disks); | |
1512 | devcnt = load_devices(devices, devmap, ident, &st, devlist, | |
1513 | c, content, mdfd, mddev, | |
1514 | &most_recent, &bestcnt, &best, inargv); | |
1515 | if (devcnt < 0) | |
1516 | return 1; | |
1517 | ||
1518 | if (devcnt == 0) { | |
1519 | pr_err("no devices found for %s\n", | |
1520 | mddev); | |
1521 | if (st) | |
1522 | st->ss->free_super(st); | |
1523 | close(mdfd); | |
1524 | free(devices); | |
1525 | free(devmap); | |
1526 | return 1; | |
1527 | } | |
1528 | ||
1529 | if (c->update && strcmp(c->update, "byteorder")==0) | |
1530 | st->minor_version = 90; | |
1531 | ||
1532 | st->ss->getinfo_super(st, content, NULL); | |
1533 | clean = content->array.state & 1; | |
1534 | ||
1535 | /* now we have some devices that might be suitable. | |
1536 | * I wonder how many | |
1537 | */ | |
1538 | avail = xcalloc(content->array.raid_disks, 1); | |
1539 | okcnt = 0; | |
1540 | replcnt = 0; | |
1541 | sparecnt=0; | |
1542 | journalcnt=0; | |
1543 | rebuilding_cnt=0; | |
1544 | for (i=0; i< bestcnt; i++) { | |
1545 | int j = best[i]; | |
1546 | int event_margin = 1; /* always allow a difference of '1' | |
1547 | * like the kernel does | |
1548 | */ | |
1549 | if (j < 0) continue; | |
1550 | /* note: we ignore error flags in multipath arrays | |
1551 | * as they don't make sense | |
1552 | */ | |
1553 | if (content->array.level != LEVEL_MULTIPATH) { | |
1554 | if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) { | |
1555 | if (content->journal_device_required) | |
1556 | journalcnt++; | |
1557 | else /* unexpected journal, mark as faulty */ | |
1558 | devices[j].i.disk.state |= (1<<MD_DISK_FAULTY); | |
1559 | } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) { | |
1560 | if (!(devices[j].i.disk.state | |
1561 | & (1<<MD_DISK_FAULTY))) { | |
1562 | devices[j].uptodate = 1; | |
1563 | sparecnt++; | |
1564 | } | |
1565 | continue; | |
1566 | } | |
1567 | } | |
1568 | /* If this device thinks that 'most_recent' has failed, then | |
1569 | * we must reject this device. | |
1570 | */ | |
1571 | if (j != most_recent && !c->force && | |
1572 | content->array.raid_disks > 0 && | |
1573 | devices[most_recent].i.disk.raid_disk >= 0 && | |
1574 | devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) { | |
1575 | if (c->verbose > -1) | |
1576 | pr_err("ignoring %s as it reports %s as failed\n", | |
1577 | devices[j].devname, devices[most_recent].devname); | |
1578 | best[i] = -1; | |
1579 | continue; | |
1580 | } | |
1581 | /* Require event counter to be same as, or just less than, | |
1582 | * most recent. If it is bigger, it must be a stray spare and | |
1583 | * should be ignored. | |
1584 | */ | |
1585 | if (devices[j].i.events+event_margin >= | |
1586 | devices[most_recent].i.events && | |
1587 | devices[j].i.events <= | |
1588 | devices[most_recent].i.events | |
1589 | ) { | |
1590 | devices[j].uptodate = 1; | |
1591 | if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) | |
1592 | journal_clean = 1; | |
1593 | if (i < content->array.raid_disks * 2) { | |
1594 | if (devices[j].i.recovery_start == MaxSector || | |
1595 | (content->reshape_active && | |
1596 | i >= content->array.raid_disks - content->delta_disks)) { | |
1597 | if (!avail[i/2]) { | |
1598 | okcnt++; | |
1599 | avail[i/2]=1; | |
1600 | } else | |
1601 | replcnt++; | |
1602 | } else | |
1603 | rebuilding_cnt++; | |
1604 | } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL) | |
1605 | sparecnt++; | |
1606 | } | |
1607 | } | |
1608 | free(devmap); | |
1609 | if (c->force) { | |
1610 | int force_ok = force_array(content, devices, best, bestcnt, | |
1611 | avail, most_recent, st, c); | |
1612 | okcnt += force_ok; | |
1613 | if (force_ok) | |
1614 | was_forced = 1; | |
1615 | } | |
1616 | /* Now we want to look at the superblock which the kernel will base things on | |
1617 | * and compare the devices that we think are working with the devices that the | |
1618 | * superblock thinks are working. | |
1619 | * If there are differences and --force is given, then update this chosen | |
1620 | * superblock. | |
1621 | */ | |
1622 | chosen_drive = -1; | |
1623 | st->ss->free_super(st); | |
1624 | for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) { | |
1625 | int j = best[i]; | |
1626 | int fd; | |
1627 | ||
1628 | if (j<0) | |
1629 | continue; | |
1630 | if (!devices[j].uptodate) | |
1631 | continue; | |
1632 | if (devices[j].i.events < devices[most_recent].i.events) | |
1633 | continue; | |
1634 | chosen_drive = j; | |
1635 | if ((fd=dev_open(devices[j].devname, | |
1636 | devices[j].included ? O_RDONLY | |
1637 | : (O_RDONLY|O_EXCL)))< 0) { | |
1638 | pr_err("Cannot open %s: %s\n", | |
1639 | devices[j].devname, strerror(errno)); | |
1640 | close(mdfd); | |
1641 | free(devices); | |
1642 | return 1; | |
1643 | } | |
1644 | if (st->ss->load_super(st,fd, NULL)) { | |
1645 | close(fd); | |
1646 | pr_err("RAID superblock has disappeared from %s\n", | |
1647 | devices[j].devname); | |
1648 | close(mdfd); | |
1649 | free(devices); | |
1650 | return 1; | |
1651 | } | |
1652 | close(fd); | |
1653 | } | |
1654 | if (st->sb == NULL) { | |
1655 | pr_err("No suitable drives found for %s\n", mddev); | |
1656 | close(mdfd); | |
1657 | free(devices); | |
1658 | return 1; | |
1659 | } | |
1660 | st->ss->getinfo_super(st, content, NULL); | |
1661 | if (sysfs_init(content, mdfd, NULL)) { | |
1662 | pr_err("Unable to initialize sysfs\n"); | |
1663 | close(mdfd); | |
1664 | free(devices); | |
1665 | return 1; | |
1666 | } | |
1667 | ||
1668 | /* after reload context, store journal_clean in context */ | |
1669 | content->journal_clean = journal_clean; | |
1670 | for (i=0; i<bestcnt; i++) { | |
1671 | int j = best[i]; | |
1672 | unsigned int desired_state; | |
1673 | ||
1674 | if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL) | |
1675 | desired_state = (1<<MD_DISK_JOURNAL); | |
1676 | else if (i >= content->array.raid_disks * 2) | |
1677 | desired_state = 0; | |
1678 | else if (i & 1) | |
1679 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT); | |
1680 | else | |
1681 | desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC); | |
1682 | ||
1683 | if (j<0) | |
1684 | continue; | |
1685 | if (!devices[j].uptodate) | |
1686 | continue; | |
1687 | ||
1688 | devices[j].i.disk.state = desired_state; | |
1689 | if (!(devices[j].i.array.state & 1)) | |
1690 | clean = 0; | |
1691 | ||
1692 | if (st->ss->update_super(st, &devices[j].i, "assemble", NULL, | |
1693 | c->verbose, 0, NULL)) { | |
1694 | if (c->force) { | |
1695 | if (c->verbose >= 0) | |
1696 | pr_err("clearing FAULTY flag for device %d in %s for %s\n", | |
1697 | j, mddev, devices[j].devname); | |
1698 | change = 1; | |
1699 | } else { | |
1700 | if (c->verbose >= -1) | |
1701 | pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n", | |
1702 | i, mddev, devices[j].devname); | |
1703 | } | |
1704 | } | |
1705 | #if 0 | |
1706 | if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) { | |
1707 | pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n", | |
1708 | i, mddev); | |
1709 | } | |
1710 | #endif | |
1711 | } | |
1712 | if (c->force && !clean && | |
1713 | !enough(content->array.level, content->array.raid_disks, | |
1714 | content->array.layout, clean, | |
1715 | avail)) { | |
1716 | change += st->ss->update_super(st, content, "force-array", | |
1717 | devices[chosen_drive].devname, c->verbose, | |
1718 | 0, NULL); | |
1719 | was_forced = 1; | |
1720 | clean = 1; | |
1721 | } | |
1722 | ||
1723 | if (change) { | |
1724 | int fd; | |
1725 | fd = dev_open(devices[chosen_drive].devname, | |
1726 | devices[chosen_drive].included ? | |
1727 | O_RDWR : (O_RDWR|O_EXCL)); | |
1728 | if (fd < 0) { | |
1729 | pr_err("Could not open %s for write - cannot Assemble array.\n", | |
1730 | devices[chosen_drive].devname); | |
1731 | close(mdfd); | |
1732 | free(devices); | |
1733 | return 1; | |
1734 | } | |
1735 | if (st->ss->store_super(st, fd)) { | |
1736 | close(fd); | |
1737 | pr_err("Could not re-write superblock on %s\n", | |
1738 | devices[chosen_drive].devname); | |
1739 | close(mdfd); | |
1740 | free(devices); | |
1741 | return 1; | |
1742 | } | |
1743 | if (c->verbose >= 0) | |
1744 | pr_err("Marking array %s as 'clean'\n", | |
1745 | mddev); | |
1746 | close(fd); | |
1747 | } | |
1748 | ||
1749 | /* If we are in the middle of a reshape we may need to restore saved data | |
1750 | * that was moved aside due to the reshape overwriting live data | |
1751 | * The code of doing this lives in Grow.c | |
1752 | */ | |
1753 | if (content->reshape_active && | |
1754 | !(content->reshape_active & RESHAPE_NO_BACKUP)) { | |
1755 | int err = 0; | |
1756 | int *fdlist = xmalloc(sizeof(int)* bestcnt); | |
1757 | if (c->verbose > 0) | |
1758 | pr_err("%s has an active reshape - checking if critical section needs to be restored\n", | |
1759 | chosen_name); | |
1760 | if (!c->backup_file) | |
1761 | c->backup_file = locate_backup(content->sys_name); | |
1762 | enable_fds(bestcnt/2); | |
1763 | for (i = 0; i < bestcnt/2; i++) { | |
1764 | int j = best[i*2]; | |
1765 | if (j >= 0) { | |
1766 | fdlist[i] = dev_open(devices[j].devname, | |
1767 | devices[j].included | |
1768 | ? O_RDWR : (O_RDWR|O_EXCL)); | |
1769 | if (fdlist[i] < 0) { | |
1770 | pr_err("Could not open %s for write - cannot Assemble array.\n", | |
1771 | devices[j].devname); | |
1772 | err = 1; | |
1773 | break; | |
1774 | } | |
1775 | } else | |
1776 | fdlist[i] = -1; | |
1777 | } | |
1778 | if (!err) { | |
1779 | if (st->ss->external && st->ss->recover_backup) | |
1780 | err = st->ss->recover_backup(st, content); | |
1781 | else | |
1782 | err = Grow_restart(st, content, fdlist, bestcnt/2, | |
1783 | c->backup_file, c->verbose > 0); | |
1784 | if (err && c->invalid_backup) { | |
1785 | if (c->verbose > 0) | |
1786 | pr_err("continuing without restoring backup\n"); | |
1787 | err = 0; | |
1788 | } | |
1789 | } | |
1790 | while (i>0) { | |
1791 | i--; | |
1792 | if (fdlist[i]>=0) close(fdlist[i]); | |
1793 | } | |
1794 | free(fdlist); | |
1795 | if (err) { | |
1796 | pr_err("Failed to restore critical section for reshape, sorry.\n"); | |
1797 | if (c->backup_file == NULL) | |
1798 | cont_err("Possibly you needed to specify the --backup-file\n"); | |
1799 | close(mdfd); | |
1800 | free(devices); | |
1801 | return err; | |
1802 | } | |
1803 | } | |
1804 | ||
1805 | /* Almost ready to actually *do* something */ | |
1806 | /* First, fill in the map, so that udev can find our name | |
1807 | * as soon as we become active. | |
1808 | */ | |
1809 | if (c->update && strcmp(c->update, "metadata")==0) { | |
1810 | content->array.major_version = 1; | |
1811 | content->array.minor_version = 0; | |
1812 | strcpy(content->text_version, "1.0"); | |
1813 | } | |
1814 | ||
1815 | map_update(&map, fd2devnm(mdfd), content->text_version, | |
1816 | content->uuid, chosen_name); | |
1817 | ||
1818 | rv = start_array(mdfd, mddev, content, | |
1819 | st, ident, best, bestcnt, | |
1820 | chosen_drive, devices, okcnt, sparecnt, | |
1821 | rebuilding_cnt, journalcnt, | |
1822 | c, | |
1823 | clean, avail, start_partial_ok, | |
1824 | pre_exist != NULL, | |
1825 | was_forced); | |
1826 | if (rv == 1 && !pre_exist) | |
1827 | ioctl(mdfd, STOP_ARRAY, NULL); | |
1828 | free(devices); | |
1829 | map_unlock(&map); | |
1830 | if (rv == 0) { | |
1831 | wait_for(chosen_name, mdfd); | |
1832 | close(mdfd); | |
1833 | if (auto_assem) { | |
1834 | int usecs = 1; | |
1835 | /* There is a nasty race with 'mdadm --monitor'. | |
1836 | * If it opens this device before we close it, | |
1837 | * it gets an incomplete open on which IO | |
1838 | * doesn't work and the capacity is | |
1839 | * wrong. | |
1840 | * If we reopen (to check for layered devices) | |
1841 | * before --monitor closes, we loose. | |
1842 | * | |
1843 | * So: wait upto 1 second for there to be | |
1844 | * a non-zero capacity. | |
1845 | */ | |
1846 | while (usecs < 1000) { | |
1847 | mdfd = open(mddev, O_RDONLY); | |
1848 | if (mdfd >= 0) { | |
1849 | unsigned long long size; | |
1850 | if (get_dev_size(mdfd, NULL, &size) && | |
1851 | size > 0) | |
1852 | break; | |
1853 | close(mdfd); | |
1854 | } | |
1855 | usleep(usecs); | |
1856 | usecs <<= 1; | |
1857 | } | |
1858 | } | |
1859 | } else | |
1860 | close(mdfd); | |
1861 | ||
1862 | /* '2' means 'OK, but not started yet' */ | |
1863 | return rv == 2 ? 0 : rv; | |
1864 | } | |
1865 | ||
1866 | int assemble_container_content(struct supertype *st, int mdfd, | |
1867 | struct mdinfo *content, struct context *c, | |
1868 | char *chosen_name, int *result) | |
1869 | { | |
1870 | struct mdinfo *dev, *sra, *dev2; | |
1871 | int working = 0, preexist = 0; | |
1872 | int expansion = 0; | |
1873 | int old_raid_disks; | |
1874 | int start_reshape; | |
1875 | char *avail; | |
1876 | int err; | |
1877 | ||
1878 | if (sysfs_init(content, mdfd, NULL)) { | |
1879 | pr_err("Unable to initialize sysfs\n"); | |
1880 | return 1; | |
1881 | } | |
1882 | ||
1883 | sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS); | |
1884 | if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) { | |
1885 | if (content->array.major_version == -1 && | |
1886 | content->array.minor_version == -2 && | |
1887 | c->readonly && | |
1888 | content->text_version[0] == '/') | |
1889 | content->text_version[0] = '-'; | |
1890 | if (sysfs_set_array(content, 9003) != 0) { | |
1891 | sysfs_free(sra); | |
1892 | return 1; | |
1893 | } | |
1894 | } | |
1895 | ||
1896 | /* There are two types of reshape: container wide or sub-array specific | |
1897 | * Check if metadata requests blocking container wide reshapes | |
1898 | */ | |
1899 | start_reshape = (content->reshape_active && | |
1900 | !((content->reshape_active == CONTAINER_RESHAPE) && | |
1901 | (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))); | |
1902 | ||
1903 | /* Block subarray here if it is under reshape now | |
1904 | * Do not allow for any changes in this array | |
1905 | */ | |
1906 | if (st->ss->external && content->recovery_blocked && start_reshape) | |
1907 | block_subarray(content); | |
1908 | ||
1909 | for (dev2 = sra->devs; dev2; dev2 = dev2->next) { | |
1910 | for (dev = content->devs; dev; dev = dev->next) | |
1911 | if (dev2->disk.major == dev->disk.major && | |
1912 | dev2->disk.minor == dev->disk.minor) | |
1913 | break; | |
1914 | if (dev) | |
1915 | continue; | |
1916 | /* Don't want this one any more */ | |
1917 | if (sysfs_set_str(sra, dev2, "slot", "none") < 0 && | |
1918 | errno == EBUSY) { | |
1919 | pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name); | |
1920 | sysfs_free(sra); | |
1921 | return 1; | |
1922 | } | |
1923 | sysfs_set_str(sra, dev2, "state", "remove"); | |
1924 | } | |
1925 | old_raid_disks = content->array.raid_disks - content->delta_disks; | |
1926 | avail = xcalloc(content->array.raid_disks, 1); | |
1927 | for (dev = content->devs; dev; dev = dev->next) { | |
1928 | if (dev->disk.raid_disk >= 0) | |
1929 | avail[dev->disk.raid_disk] = 1; | |
1930 | if (sysfs_add_disk(content, dev, 1) == 0) { | |
1931 | if (dev->disk.raid_disk >= old_raid_disks && | |
1932 | content->reshape_active) | |
1933 | expansion++; | |
1934 | else | |
1935 | working++; | |
1936 | } else if (errno == EEXIST) | |
1937 | preexist++; | |
1938 | } | |
1939 | sysfs_free(sra); | |
1940 | if (working + expansion == 0 && c->runstop <= 0) { | |
1941 | free(avail); | |
1942 | return 1;/* Nothing new, don't try to start */ | |
1943 | } | |
1944 | map_update(NULL, fd2devnm(mdfd), content->text_version, | |
1945 | content->uuid, chosen_name); | |
1946 | ||
1947 | if (content->consistency_policy == CONSISTENCY_POLICY_PPL && | |
1948 | st->ss->validate_ppl) { | |
1949 | content->array.state |= 1; | |
1950 | err = 0; | |
1951 | ||
1952 | for (dev = content->devs; dev; dev = dev->next) { | |
1953 | int dfd; | |
1954 | char *devpath; | |
1955 | int ret; | |
1956 | ||
1957 | ret = st->ss->validate_ppl(st, content, dev); | |
1958 | if (ret == 0) | |
1959 | continue; | |
1960 | ||
1961 | if (ret < 0) { | |
1962 | err = 1; | |
1963 | break; | |
1964 | } | |
1965 | ||
1966 | if (!c->force) { | |
1967 | pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n", | |
1968 | chosen_name); | |
1969 | content->array.state &= ~1; | |
1970 | avail[dev->disk.raid_disk] = 0; | |
1971 | break; | |
1972 | } | |
1973 | ||
1974 | /* have --force - overwrite the invalid ppl */ | |
1975 | devpath = map_dev(dev->disk.major, dev->disk.minor, 0); | |
1976 | dfd = dev_open(devpath, O_RDWR); | |
1977 | if (dfd < 0) { | |
1978 | pr_err("Failed to open %s\n", devpath); | |
1979 | err = 1; | |
1980 | break; | |
1981 | } | |
1982 | ||
1983 | err = st->ss->write_init_ppl(st, content, dfd); | |
1984 | close(dfd); | |
1985 | ||
1986 | if (err) | |
1987 | break; | |
1988 | } | |
1989 | ||
1990 | if (err) { | |
1991 | free(avail); | |
1992 | return err; | |
1993 | } | |
1994 | } | |
1995 | ||
1996 | if (enough(content->array.level, content->array.raid_disks, | |
1997 | content->array.layout, content->array.state & 1, avail) == 0) { | |
1998 | if (c->export && result) | |
1999 | *result |= INCR_NO; | |
2000 | else if (c->verbose >= 0) { | |
2001 | pr_err("%s assembled with %d device%s", | |
2002 | chosen_name, preexist + working, | |
2003 | preexist + working == 1 ? "":"s"); | |
2004 | if (preexist) | |
2005 | fprintf(stderr, " (%d new)", working); | |
2006 | fprintf(stderr, " but not started\n"); | |
2007 | } | |
2008 | free(avail); | |
2009 | return 1; | |
2010 | } | |
2011 | free(avail); | |
2012 | ||
2013 | if (c->runstop <= 0 && | |
2014 | (working + preexist + expansion) < | |
2015 | content->array.working_disks) { | |
2016 | if (c->export && result) | |
2017 | *result |= INCR_UNSAFE; | |
2018 | else if (c->verbose >= 0) { | |
2019 | pr_err("%s assembled with %d device%s", | |
2020 | chosen_name, preexist + working, | |
2021 | preexist + working == 1 ? "":"s"); | |
2022 | if (preexist) | |
2023 | fprintf(stderr, " (%d new)", working); | |
2024 | fprintf(stderr, " but not safe to start\n"); | |
2025 | } | |
2026 | return 1; | |
2027 | } | |
2028 | ||
2029 | ||
2030 | if (start_reshape) { | |
2031 | int spare = content->array.raid_disks + expansion; | |
2032 | if (restore_backup(st, content, | |
2033 | working, | |
2034 | spare, &c->backup_file, c->verbose) == 1) | |
2035 | return 1; | |
2036 | ||
2037 | err = sysfs_set_str(content, NULL, | |
2038 | "array_state", "readonly"); | |
2039 | if (err) | |
2040 | return 1; | |
2041 | ||
2042 | if (st->ss->external) { | |
2043 | if (!mdmon_running(st->container_devnm)) | |
2044 | start_mdmon(st->container_devnm); | |
2045 | ping_monitor(st->container_devnm); | |
2046 | if (mdmon_running(st->container_devnm) && | |
2047 | st->update_tail == NULL) | |
2048 | st->update_tail = &st->updates; | |
2049 | } | |
2050 | ||
2051 | err = Grow_continue(mdfd, st, content, c->backup_file, | |
2052 | 0, c->freeze_reshape); | |
2053 | } else switch(content->array.level) { | |
2054 | case LEVEL_LINEAR: | |
2055 | case LEVEL_MULTIPATH: | |
2056 | case 0: | |
2057 | err = sysfs_set_str(content, NULL, "array_state", | |
2058 | c->readonly ? "readonly" : "active"); | |
2059 | break; | |
2060 | default: | |
2061 | err = sysfs_set_str(content, NULL, "array_state", | |
2062 | "readonly"); | |
2063 | /* start mdmon if needed. */ | |
2064 | if (!err) { | |
2065 | if (!mdmon_running(st->container_devnm)) | |
2066 | start_mdmon(st->container_devnm); | |
2067 | ping_monitor(st->container_devnm); | |
2068 | } | |
2069 | break; | |
2070 | } | |
2071 | if (!err) | |
2072 | sysfs_set_safemode(content, content->safe_mode_delay); | |
2073 | ||
2074 | /* Block subarray here if it is not reshaped now | |
2075 | * It has be blocked a little later to allow mdmon to switch in | |
2076 | * in to R/W state | |
2077 | */ | |
2078 | if (st->ss->external && content->recovery_blocked && | |
2079 | !start_reshape) | |
2080 | block_subarray(content); | |
2081 | ||
2082 | if (c->export && result) { | |
2083 | if (err) | |
2084 | *result |= INCR_NO; | |
2085 | else | |
2086 | *result |= INCR_YES; | |
2087 | } else if (c->verbose >= 0) { | |
2088 | if (err) | |
2089 | pr_err("array %s now has %d device%s", | |
2090 | chosen_name, working + preexist, | |
2091 | working + preexist == 1 ? "":"s"); | |
2092 | else | |
2093 | pr_err("Started %s with %d device%s", | |
2094 | chosen_name, working + preexist, | |
2095 | working + preexist == 1 ? "":"s"); | |
2096 | if (preexist) | |
2097 | fprintf(stderr, " (%d new)", working); | |
2098 | if (expansion) | |
2099 | fprintf(stderr, " ( + %d for expansion)", | |
2100 | expansion); | |
2101 | fprintf(stderr, "\n"); | |
2102 | } | |
2103 | if (!err) | |
2104 | wait_for(chosen_name, mdfd); | |
2105 | return err; | |
2106 | /* FIXME should have an O_EXCL and wait for read-auto */ | |
2107 | } |