]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mdadm/util: unify fstat checking blkdev into function
[thirdparty/mdadm.git] / Assemble.c
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 struct stat stb;
516 if (tmpdev->used != 3)
517 continue;
518 if (stat(tmpdev->devname, &stb)< 0) {
519 pr_err("fstat failed for %s: %s\n",
520 tmpdev->devname, strerror(errno));
521 tmpdev->used = 2;
522 } else {
523 struct dev_policy *pol = devid_policy(stb.st_rdev);
524 int dt = domain_test(domains, pol, NULL);
525 if (inargv && dt != 0)
526 /* take this spare as domains match
527 * if there are any */
528 tmpdev->used = 1;
529 else if (!inargv && dt == 1)
530 /* device wasn't explicitly listed, so need
531 * explicit domain match - which we have */
532 tmpdev->used = 1;
533 else
534 /* if domains don't match mark as unused */
535 tmpdev->used = 0;
536 dev_policy_free(pol);
537 }
538 }
539 domain_free(domains);
540 *stp = st;
541 if (st && st->sb && content == *contentp)
542 st->ss->getinfo_super(st, content, NULL);
543 *contentp = content;
544
545 return num_devs;
546 }
547
548 struct devs {
549 char *devname;
550 int uptodate; /* set once we decide that this device is as
551 * recent as everything else in the array.
552 */
553 int included; /* set if the device is already in the array
554 * due to a previous '-I'
555 */
556 struct mdinfo i;
557 };
558
559 static int load_devices(struct devs *devices, char *devmap,
560 struct mddev_ident *ident, struct supertype **stp,
561 struct mddev_dev *devlist, struct context *c,
562 struct mdinfo *content,
563 int mdfd, char *mddev,
564 int *most_recentp, int *bestcntp, int **bestp,
565 int inargv)
566 {
567 struct mddev_dev *tmpdev;
568 int devcnt = 0;
569 int nextspare = 0;
570 int bitmap_done = 0;
571 int most_recent = -1;
572 int bestcnt = 0;
573 int *best = *bestp;
574 struct supertype *st = *stp;
575
576 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
577 char *devname = tmpdev->devname;
578 struct stat stb;
579 struct supertype *tst;
580 int i;
581 int dfd;
582
583 if (tmpdev->used != 1)
584 continue;
585 /* looks like a good enough match to update the super block if needed */
586 if (c->update) {
587 /* prepare useful information in info structures */
588 struct stat stb2;
589 int err;
590 fstat(mdfd, &stb2);
591
592 if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
593 random_uuid((__u8 *)ident->uuid);
594
595 if (strcmp(c->update, "ppl") == 0 &&
596 ident->bitmap_fd >= 0) {
597 pr_err("PPL is not compatible with bitmap\n");
598 return -1;
599 }
600
601 dfd = dev_open(devname,
602 tmpdev->disposition == 'I'
603 ? O_RDWR : (O_RDWR|O_EXCL));
604
605 tst = dup_super(st);
606 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
607 pr_err("cannot re-read metadata from %s - aborting\n",
608 devname);
609 if (dfd >= 0)
610 close(dfd);
611 close(mdfd);
612 free(devices);
613 free(devmap);
614 tst->ss->free_super(tst);
615 free(tst);
616 *stp = st;
617 return -1;
618 }
619 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
620
621 memcpy(content->uuid, ident->uuid, 16);
622 strcpy(content->name, ident->name);
623 content->array.md_minor = minor(stb2.st_rdev);
624
625 if (strcmp(c->update, "byteorder") == 0)
626 err = 0;
627 else if (strcmp(c->update, "home-cluster") == 0) {
628 tst->cluster_name = c->homecluster;
629 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
630 } else if (strcmp(c->update, "nodes") == 0) {
631 tst->nodes = c->nodes;
632 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
633 } else if (strcmp(c->update, "revert-reshape") == 0 &&
634 c->invalid_backup)
635 err = tst->ss->update_super(tst, content,
636 "revert-reshape-nobackup",
637 devname, c->verbose,
638 ident->uuid_set,
639 c->homehost);
640 else
641 err = tst->ss->update_super(tst, content, c->update,
642 devname, c->verbose,
643 ident->uuid_set,
644 c->homehost);
645 if (err < 0) {
646 if (err == -1)
647 pr_err("--update=%s not understood for %s metadata\n",
648 c->update, tst->ss->name);
649 tst->ss->free_super(tst);
650 free(tst);
651 close(mdfd);
652 close(dfd);
653 free(devices);
654 free(devmap);
655 *stp = st;
656 return -1;
657 }
658 if (strcmp(c->update, "uuid")==0 &&
659 !ident->uuid_set) {
660 ident->uuid_set = 1;
661 memcpy(ident->uuid, content->uuid, 16);
662 }
663 if (tst->ss->store_super(tst, dfd))
664 pr_err("Could not re-write superblock on %s.\n",
665 devname);
666
667 if (strcmp(c->update, "uuid")==0 &&
668 ident->bitmap_fd >= 0 && !bitmap_done) {
669 if (bitmap_update_uuid(ident->bitmap_fd,
670 content->uuid,
671 tst->ss->swapuuid) != 0)
672 pr_err("Could not update uuid on external bitmap.\n");
673 else
674 bitmap_done = 1;
675 }
676 } else {
677 dfd = dev_open(devname,
678 tmpdev->disposition == 'I'
679 ? O_RDWR : (O_RDWR|O_EXCL));
680 tst = dup_super(st);
681
682 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
683 pr_err("cannot re-read metadata from %s - aborting\n",
684 devname);
685 if (dfd >= 0)
686 close(dfd);
687 close(mdfd);
688 free(devices);
689 free(devmap);
690 tst->ss->free_super(tst);
691 free(tst);
692 *stp = st;
693 return -1;
694 }
695 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
696 }
697
698 fstat(dfd, &stb);
699 close(dfd);
700
701 if (c->verbose > 0)
702 pr_err("%s is identified as a member of %s, slot %d%s.\n",
703 devname, mddev, content->disk.raid_disk,
704 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
705 devices[devcnt].devname = devname;
706 devices[devcnt].uptodate = 0;
707 devices[devcnt].included = (tmpdev->disposition == 'I');
708 devices[devcnt].i = *content;
709 devices[devcnt].i.disk.major = major(stb.st_rdev);
710 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
711
712 if (devices[devcnt].i.disk.state == 6) {
713 if (most_recent < 0 ||
714 devices[devcnt].i.events
715 > devices[most_recent].i.events) {
716 struct supertype *tmp = tst;
717 tst = st;
718 st = tmp;
719 most_recent = devcnt;
720 }
721 }
722 tst->ss->free_super(tst);
723 free(tst);
724
725 if (content->array.level == LEVEL_MULTIPATH)
726 /* with multipath, the raid_disk from the superblock is meaningless */
727 i = devcnt;
728 else
729 i = devices[devcnt].i.disk.raid_disk;
730 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
731 if (nextspare < content->array.raid_disks*2)
732 nextspare = content->array.raid_disks*2;
733 i = nextspare++;
734 } else {
735 /* i is raid_disk - double it so there is room for
736 * replacements */
737 i *= 2;
738 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
739 i++;
740 if (i >= content->array.raid_disks*2 &&
741 i >= nextspare)
742 nextspare = i+1;
743 }
744 if (i < 10000) {
745 if (i >= bestcnt) {
746 int newbestcnt = i+10;
747 int *newbest = xmalloc(sizeof(int)*newbestcnt);
748 int c;
749 for (c=0; c < newbestcnt; c++)
750 if (c < bestcnt)
751 newbest[c] = best[c];
752 else
753 newbest[c] = -1;
754 if (best)free(best);
755 best = newbest;
756 bestcnt = newbestcnt;
757 }
758 if (best[i] >=0 &&
759 devices[best[i]].i.events
760 == devices[devcnt].i.events
761 && (devices[best[i]].i.disk.minor
762 != devices[devcnt].i.disk.minor)
763 && st->ss == &super0
764 && content->array.level != LEVEL_MULTIPATH) {
765 /* two different devices with identical superblock.
766 * Could be a mis-detection caused by overlapping
767 * partitions. fail-safe.
768 */
769 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
770 " If they are really different, please --zero the superblock on one\n"
771 " If they are the same or overlap, please remove one from %s.\n",
772 devices[best[i]].devname, devname,
773 inargv ? "the list" :
774 "the\n DEVICE list in mdadm.conf"
775 );
776 close(mdfd);
777 free(devices);
778 free(devmap);
779 *stp = st;
780 return -1;
781 }
782 if (best[i] == -1
783 || (devices[best[i]].i.events
784 < devices[devcnt].i.events))
785 best[i] = devcnt;
786 }
787 devcnt++;
788 }
789 if (most_recent >= 0)
790 *most_recentp = most_recent;
791 *bestcntp = bestcnt;
792 *bestp = best;
793 *stp = st;
794 return devcnt;
795 }
796
797 static int force_array(struct mdinfo *content,
798 struct devs *devices,
799 int *best, int bestcnt, char *avail,
800 int most_recent,
801 struct supertype *st,
802 struct context *c)
803 {
804 int okcnt = 0;
805 while (!enough(content->array.level, content->array.raid_disks,
806 content->array.layout, 1,
807 avail)
808 ||
809 (content->reshape_active && content->delta_disks > 0 &&
810 !enough(content->array.level, (content->array.raid_disks
811 - content->delta_disks),
812 content->new_layout, 1,
813 avail)
814 )) {
815 /* Choose the newest best drive which is
816 * not up-to-date, update the superblock
817 * and add it.
818 */
819 int fd;
820 struct supertype *tst;
821 unsigned long long current_events;
822 int chosen_drive = -1;
823 int i;
824
825 for (i = 0;
826 i < content->array.raid_disks * 2 && i < bestcnt;
827 i += 2) {
828 int j = best[i];
829 if (j < 0)
830 continue;
831 if (devices[j].uptodate)
832 continue;
833 if (devices[j].i.recovery_start != MaxSector) {
834 int delta;
835 if (!devices[j].i.reshape_active ||
836 devices[j].i.delta_disks <= 0)
837 continue;
838 /* When increasing number of devices, an
839 * added device also appears to be
840 * recovering. It is safe to include it
841 * as long as it won't be a source of
842 * data.
843 * For now, just allow for last data
844 * devices in RAID4 or last devices in RAID4/5/6.
845 */
846 delta = devices[j].i.delta_disks;
847 if (devices[j].i.array.level >= 4 &&
848 devices[j].i.array.level <= 6 &&
849 i/2 >= content->array.raid_disks - delta)
850 /* OK */;
851 else if (devices[j].i.array.level == 4 &&
852 i/2 >= content->array.raid_disks - delta - 1)
853 /* OK */;
854 else
855 continue;
856 }
857 if (chosen_drive < 0 ||
858 devices[j].i.events
859 > devices[chosen_drive].i.events)
860 chosen_drive = j;
861 }
862 if (chosen_drive < 0)
863 break;
864 current_events = devices[chosen_drive].i.events;
865 add_another:
866 if (c->verbose >= 0)
867 pr_err("forcing event count in %s(%d) from %d upto %d\n",
868 devices[chosen_drive].devname,
869 devices[chosen_drive].i.disk.raid_disk,
870 (int)(devices[chosen_drive].i.events),
871 (int)(devices[most_recent].i.events));
872 fd = dev_open(devices[chosen_drive].devname,
873 devices[chosen_drive].included ? O_RDWR
874 : (O_RDWR|O_EXCL));
875 if (fd < 0) {
876 pr_err("Couldn't open %s for write - not updating\n",
877 devices[chosen_drive].devname);
878 devices[chosen_drive].i.events = 0;
879 continue;
880 }
881 tst = dup_super(st);
882 if (tst->ss->load_super(tst,fd, NULL)) {
883 close(fd);
884 pr_err("RAID superblock disappeared from %s - not updating.\n",
885 devices[chosen_drive].devname);
886 devices[chosen_drive].i.events = 0;
887 continue;
888 }
889 content->events = devices[most_recent].i.events;
890 tst->ss->update_super(tst, content, "force-one",
891 devices[chosen_drive].devname, c->verbose,
892 0, NULL);
893
894 if (tst->ss->store_super(tst, fd)) {
895 close(fd);
896 pr_err("Could not re-write superblock on %s\n",
897 devices[chosen_drive].devname);
898 devices[chosen_drive].i.events = 0;
899 tst->ss->free_super(tst);
900 continue;
901 }
902 close(fd);
903 devices[chosen_drive].i.events = devices[most_recent].i.events;
904 devices[chosen_drive].uptodate = 1;
905 avail[chosen_drive] = 1;
906 okcnt++;
907 tst->ss->free_super(tst);
908 /* If there are any other drives of the same vintage,
909 * add them in as well. We can't lose and we might gain
910 */
911 for (i = 0;
912 i < content->array.raid_disks * 2 && i < bestcnt ;
913 i += 2) {
914 int j = best[i];
915 if (j >= 0 &&
916 !devices[j].uptodate &&
917 devices[j].i.recovery_start == MaxSector &&
918 devices[j].i.events == current_events) {
919 chosen_drive = j;
920 goto add_another;
921 }
922 }
923 }
924 return okcnt;
925 }
926
927 static int start_array(int mdfd,
928 char *mddev,
929 struct mdinfo *content,
930 struct supertype *st,
931 struct mddev_ident *ident,
932 int *best, int bestcnt,
933 int chosen_drive,
934 struct devs *devices,
935 unsigned int okcnt,
936 unsigned int sparecnt,
937 unsigned int rebuilding_cnt,
938 unsigned int journalcnt,
939 struct context *c,
940 int clean, char *avail,
941 int start_partial_ok,
942 int err_ok,
943 int was_forced
944 )
945 {
946 int rv;
947 int i;
948 unsigned int req_cnt;
949
950 if (content->journal_device_required && (content->journal_clean == 0)) {
951 if (!c->force) {
952 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
953 return 1;
954 }
955 pr_err("Journal is missing or stale, starting array read only.\n");
956 c->readonly = 1;
957 }
958
959 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
960 clean = 1;
961
962 rv = set_array_info(mdfd, st, content);
963 if (rv && !err_ok) {
964 pr_err("failed to set array info for %s: %s\n",
965 mddev, strerror(errno));
966 return 1;
967 }
968 if (ident->bitmap_fd >= 0) {
969 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
970 pr_err("SET_BITMAP_FILE failed.\n");
971 return 1;
972 }
973 } else if (ident->bitmap_file) {
974 /* From config file */
975 int bmfd = open(ident->bitmap_file, O_RDWR);
976 if (bmfd < 0) {
977 pr_err("Could not open bitmap file %s\n",
978 ident->bitmap_file);
979 return 1;
980 }
981 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
982 pr_err("Failed to set bitmapfile for %s\n", mddev);
983 close(bmfd);
984 return 1;
985 }
986 close(bmfd);
987 }
988
989 /* First, add the raid disks, but add the chosen one last */
990 for (i = 0; i <= bestcnt; i++) {
991 int j;
992 if (i < bestcnt) {
993 j = best[i];
994 if (j == chosen_drive)
995 continue;
996 } else
997 j = chosen_drive;
998
999 if (j >= 0 && !devices[j].included) {
1000 int dfd;
1001
1002 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1003 if (dfd >= 0) {
1004 remove_partitions(dfd);
1005 close(dfd);
1006 }
1007 rv = add_disk(mdfd, st, content, &devices[j].i);
1008
1009 if (rv) {
1010 pr_err("failed to add %s to %s: %s\n",
1011 devices[j].devname, mddev,
1012 strerror(errno));
1013 if (i < content->array.raid_disks * 2 ||
1014 i == bestcnt)
1015 okcnt--;
1016 else
1017 sparecnt--;
1018 } else if (c->verbose > 0) {
1019 pr_err("added %s to %s as %d%s%s\n",
1020 devices[j].devname, mddev,
1021 devices[j].i.disk.raid_disk,
1022 devices[j].uptodate?"":
1023 " (possibly out of date)",
1024 (devices[j].i.disk.state &
1025 (1<<MD_DISK_REPLACEMENT)) ?
1026 " replacement":"");
1027 }
1028 } else if (j >= 0) {
1029 if (c->verbose > 0)
1030 pr_err("%s is already in %s as %d\n",
1031 devices[j].devname, mddev,
1032 devices[j].i.disk.raid_disk);
1033 } else if (c->verbose > 0 &&
1034 i < content->array.raid_disks * 2 && (i & 1) == 0)
1035 pr_err("no uptodate device for slot %d of %s\n",
1036 i/2, mddev);
1037 }
1038
1039 if (content->array.level == LEVEL_CONTAINER) {
1040 if (c->verbose >= 0) {
1041 pr_err("Container %s has been assembled with %d drive%s",
1042 mddev, okcnt + sparecnt + journalcnt,
1043 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1044 if (okcnt < (unsigned)content->array.raid_disks)
1045 fprintf(stderr, " (out of %d)",
1046 content->array.raid_disks);
1047 fprintf(stderr, "\n");
1048 }
1049
1050 if (st->ss->validate_container) {
1051 struct mdinfo *devices_list;
1052 struct mdinfo *info_devices;
1053 unsigned int count;
1054
1055 devices_list = NULL;
1056 info_devices = xmalloc(sizeof(struct mdinfo) *
1057 (okcnt + sparecnt));
1058 for (count = 0; count < okcnt + sparecnt; count++) {
1059 info_devices[count] = devices[count].i;
1060 info_devices[count].next = devices_list;
1061 devices_list = &info_devices[count];
1062 }
1063 if (st->ss->validate_container(devices_list))
1064 pr_err("Mismatch detected!\n");
1065 free(info_devices);
1066 }
1067
1068 st->ss->free_super(st);
1069 sysfs_uevent(content, "change");
1070 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1071 /* Was partial, is still partial, so signal an error
1072 * to ensure we don't retry */
1073 return 1;
1074 return 0;
1075 }
1076
1077 /* Get number of in-sync devices according to the superblock.
1078 * We must have this number to start the array without -s or -R
1079 */
1080 req_cnt = content->array.working_disks;
1081
1082 if (c->runstop == 1 ||
1083 (c->runstop <= 0 &&
1084 (enough(content->array.level, content->array.raid_disks,
1085 content->array.layout, clean, avail) &&
1086 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1087 /* This array is good-to-go.
1088 * If a reshape is in progress then we might need to
1089 * continue monitoring it. In that case we start
1090 * it read-only and let the grow code make it writable.
1091 */
1092 int rv;
1093
1094 if (content->reshape_active &&
1095 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1096 content->delta_disks <= 0) {
1097 if (!c->backup_file) {
1098 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1099 mddev);
1100 pr_err("Please provided one with \"--backup-file=...\"\n");
1101 if (c->update &&
1102 strcmp(c->update, "revert-reshape") == 0)
1103 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1104 return 1;
1105 }
1106 rv = sysfs_set_str(content, NULL,
1107 "array_state", "readonly");
1108 if (rv == 0)
1109 rv = Grow_continue(mdfd, st, content,
1110 c->backup_file, 0,
1111 c->freeze_reshape);
1112 } else if (c->readonly &&
1113 sysfs_attribute_available(content, NULL,
1114 "array_state")) {
1115 rv = sysfs_set_str(content, NULL,
1116 "array_state", "readonly");
1117 } else
1118 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1119 reopen_mddev(mdfd); /* drop O_EXCL */
1120 if (rv == 0) {
1121 if (c->verbose >= 0) {
1122 pr_err("%s has been started with %d drive%s",
1123 mddev, okcnt, okcnt==1?"":"s");
1124 if (okcnt < (unsigned)content->array.raid_disks)
1125 fprintf(stderr, " (out of %d)",
1126 content->array.raid_disks);
1127 if (rebuilding_cnt)
1128 fprintf(stderr, "%s %d rebuilding",
1129 sparecnt?",":" and",
1130 rebuilding_cnt);
1131 if (sparecnt)
1132 fprintf(stderr, " and %d spare%s",
1133 sparecnt,
1134 sparecnt == 1 ? "" : "s");
1135 if (content->journal_clean)
1136 fprintf(stderr, " and %d journal",
1137 journalcnt);
1138 fprintf(stderr, ".\n");
1139 }
1140 if (content->reshape_active &&
1141 content->array.level >= 4 &&
1142 content->array.level <= 6) {
1143 /* might need to increase the size
1144 * of the stripe cache - default is 256
1145 */
1146 int chunk_size = content->array.chunk_size;
1147
1148 if (content->reshape_active &&
1149 content->new_chunk > chunk_size)
1150 chunk_size = content->new_chunk;
1151 if (256 < 4 * ((chunk_size+4065)/4096)) {
1152 struct mdinfo *sra;
1153
1154 sra = sysfs_read(mdfd, NULL, 0);
1155 if (sra)
1156 sysfs_set_num(sra, NULL,
1157 "stripe_cache_size",
1158 (4 * chunk_size / 4096) + 1);
1159 sysfs_free(sra);
1160 }
1161 }
1162 if (okcnt < (unsigned)content->array.raid_disks) {
1163 /* If any devices did not get added
1164 * because the kernel rejected them based
1165 * on event count, try adding them
1166 * again providing the action policy is
1167 * 're-add' or greater. The bitmap
1168 * might allow them to be included, or
1169 * they will become spares.
1170 */
1171 for (i = 0; i < bestcnt; i++) {
1172 int j = best[i];
1173 if (j >= 0 && !devices[j].uptodate) {
1174 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1175 continue;
1176 rv = add_disk(mdfd, st, content,
1177 &devices[j].i);
1178 if (rv == 0 && c->verbose >= 0)
1179 pr_err("%s has been re-added.\n",
1180 devices[j].devname);
1181 }
1182 }
1183 }
1184 if (content->array.level == 6 &&
1185 okcnt + 1 == (unsigned)content->array.raid_disks &&
1186 was_forced) {
1187 struct mdinfo *sra;
1188
1189 sra = sysfs_read(mdfd, NULL, 0);
1190 if (sra)
1191 sysfs_set_str(sra, NULL,
1192 "sync_action", "repair");
1193 sysfs_free(sra);
1194 }
1195 return 0;
1196 }
1197 pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno));
1198
1199 if (!enough(content->array.level, content->array.raid_disks,
1200 content->array.layout, 1, avail))
1201 pr_err("Not enough devices to start the array.\n");
1202 else if (!enough(content->array.level,
1203 content->array.raid_disks,
1204 content->array.layout, clean, avail))
1205 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1206
1207 return 1;
1208 }
1209 if (c->runstop == -1) {
1210 pr_err("%s assembled from %d drive%s",
1211 mddev, okcnt, okcnt == 1 ? "" : "s");
1212 if (okcnt != (unsigned)content->array.raid_disks)
1213 fprintf(stderr, " (out of %d)",
1214 content->array.raid_disks);
1215 fprintf(stderr, ", but not started.\n");
1216 return 2;
1217 }
1218 if (c->verbose >= -1) {
1219 pr_err("%s assembled from %d drive%s",
1220 mddev, okcnt, okcnt == 1 ? "" : "s");
1221 if (rebuilding_cnt)
1222 fprintf(stderr, "%s %d rebuilding",
1223 sparecnt ? "," : " and", rebuilding_cnt);
1224 if (sparecnt)
1225 fprintf(stderr, " and %d spare%s", sparecnt,
1226 sparecnt == 1 ? "" : "s");
1227 if (!enough(content->array.level, content->array.raid_disks,
1228 content->array.layout, 1, avail))
1229 fprintf(stderr, " - not enough to start the array.\n");
1230 else if (!enough(content->array.level,
1231 content->array.raid_disks,
1232 content->array.layout, clean, avail))
1233 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1234 else {
1235 if (req_cnt == (unsigned)content->array.raid_disks)
1236 fprintf(stderr, " - need all %d to start it",
1237 req_cnt);
1238 else
1239 fprintf(stderr, " - need %d to start", req_cnt);
1240 fprintf(stderr, " (use --run to insist).\n");
1241 }
1242 }
1243 return 1;
1244 }
1245
1246 int Assemble(struct supertype *st, char *mddev,
1247 struct mddev_ident *ident,
1248 struct mddev_dev *devlist,
1249 struct context *c)
1250 {
1251 /*
1252 * The task of Assemble is to find a collection of
1253 * devices that should (according to their superblocks)
1254 * form an array, and to give this collection to the MD driver.
1255 * In Linux-2.4 and later, this involves submitting a
1256 * SET_ARRAY_INFO ioctl with no arg - to prepare
1257 * the array - and then submit a number of
1258 * ADD_NEW_DISK ioctls to add disks into
1259 * the array. Finally RUN_ARRAY might
1260 * be submitted to start the array.
1261 *
1262 * Much of the work of Assemble is in finding and/or
1263 * checking the disks to make sure they look right.
1264 *
1265 * If mddev is not set, then scan must be set and we
1266 * read through the config file for dev+uuid mapping
1267 * We recurse, setting mddev, for each device that
1268 * - isn't running
1269 * - has a valid uuid (or any uuid if !uuidset)
1270 *
1271 * If mddev is set, we try to determine state of md.
1272 * check version - must be at least 0.90.0
1273 * check kernel version. must be at least 2.4.
1274 * If not, we can possibly fall back on START_ARRAY
1275 * Try to GET_ARRAY_INFO.
1276 * If possible, give up
1277 * If not, try to STOP_ARRAY just to make sure
1278 *
1279 * If !uuidset and scan, look in conf-file for uuid
1280 * If not found, give up
1281 * If !devlist and scan and uuidset, get list of devs from conf-file
1282 *
1283 * For each device:
1284 * Check superblock - discard if bad
1285 * Check uuid (set if we don't have one) - discard if no match
1286 * Check superblock similarity if we have a superblock - discard if different
1287 * Record events, devicenum
1288 * This should give us a list of devices for the array
1289 * We should collect the most recent event number
1290 *
1291 * Count disks with recent enough event count
1292 * While force && !enough disks
1293 * Choose newest rejected disks, update event count
1294 * mark clean and rewrite superblock
1295 * If recent kernel:
1296 * SET_ARRAY_INFO
1297 * foreach device with recent events : ADD_NEW_DISK
1298 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1299 * If old kernel:
1300 * Check the device numbers in superblock are right
1301 * update superblock if any changes
1302 * START_ARRAY
1303 *
1304 */
1305 int rv;
1306 int mdfd;
1307 int clean;
1308 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1309 ident->super_minor == UnSet && ident->name[0] == 0
1310 && (ident->container == NULL || ident->member == NULL));
1311 struct devs *devices;
1312 char *devmap;
1313 int *best = NULL; /* indexed by raid_disk */
1314 int bestcnt = 0;
1315 int devcnt;
1316 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1317 int journal_clean = 0;
1318 int i;
1319 int was_forced = 0;
1320 int most_recent = 0;
1321 int chosen_drive;
1322 int change = 0;
1323 int inargv = 0;
1324 int start_partial_ok = (c->runstop >= 0) &&
1325 (c->force || devlist==NULL || auto_assem);
1326 int num_devs;
1327 struct mddev_dev *tmpdev;
1328 struct mdinfo info;
1329 struct mdinfo *content = NULL;
1330 struct mdinfo *pre_exist = NULL;
1331 char *avail;
1332 char *name = NULL;
1333 char chosen_name[1024];
1334 struct map_ent *map = NULL;
1335 struct map_ent *mp;
1336
1337 /*
1338 * If any subdevs are listed, then any that don't
1339 * match ident are discarded. Remainder must all match and
1340 * become the array.
1341 * If no subdevs, then we scan all devices in the config file, but
1342 * there must be something in the identity
1343 */
1344
1345 if (!devlist &&
1346 ident->uuid_set == 0 &&
1347 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1348 ident->name[0] == 0 &&
1349 (ident->container == NULL || ident->member == NULL) &&
1350 ident->devices == NULL) {
1351 pr_err("No identity information available for %s - cannot assemble.\n",
1352 mddev ? mddev : "further assembly");
1353 return 1;
1354 }
1355
1356 if (devlist == NULL)
1357 devlist = conf_get_devs();
1358 else if (mddev)
1359 inargv = 1;
1360
1361 try_again:
1362 /* We come back here when doing auto-assembly and attempting some
1363 * set of devices failed. Those are now marked as ->used==2 and
1364 * we ignore them and try again
1365 */
1366 if (!st && ident->st)
1367 st = ident->st;
1368 if (c->verbose>0)
1369 pr_err("looking for devices for %s\n",
1370 mddev ? mddev : "further assembly");
1371
1372 content = &info;
1373 if (st && c->force)
1374 st->ignore_hw_compat = 1;
1375 num_devs = select_devices(devlist, ident, &st, &content, c,
1376 inargv, auto_assem);
1377 if (num_devs < 0)
1378 return 1;
1379
1380 if (!st || !st->sb || !content)
1381 return 2;
1382
1383 /* We have a full set of devices - we now need to find the
1384 * array device.
1385 * However there is a risk that we are racing with "mdadm -I"
1386 * and the array is already partially assembled - we will have
1387 * rejected any devices already in this address.
1388 * So we take a lock on the map file - to prevent further races -
1389 * and look for the uuid in there. If found and the array is
1390 * active, we abort. If found and the array is not active
1391 * we commit to that md device and add all the contained devices
1392 * to our list. We flag them so that we don't try to re-add,
1393 * but can remove if they turn out to not be wanted.
1394 */
1395 if (map_lock(&map))
1396 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1397 if (c->update && strcmp(c->update,"uuid") == 0)
1398 mp = NULL;
1399 else
1400 mp = map_by_uuid(&map, content->uuid);
1401 if (mp) {
1402 struct mdinfo *dv;
1403 /* array already exists. */
1404 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1405 if (pre_exist->array.level != UnSet) {
1406 pr_err("Found some drive for an array that is already active: %s\n",
1407 mp->path);
1408 pr_err("giving up.\n");
1409 return 1;
1410 }
1411 for (dv = pre_exist->devs; dv; dv = dv->next) {
1412 /* We want to add this device to our list,
1413 * but it could already be there if "mdadm -I"
1414 * started *after* we checked for O_EXCL.
1415 * If we add it to the top of the list
1416 * it will be preferred over later copies.
1417 */
1418 struct mddev_dev *newdev;
1419 char *devname = map_dev(dv->disk.major,
1420 dv->disk.minor,
1421 0);
1422 if (!devname)
1423 continue;
1424 newdev = xmalloc(sizeof(*newdev));
1425 newdev->devname = devname;
1426 newdev->disposition = 'I';
1427 newdev->used = 1;
1428 newdev->next = devlist;
1429 devlist = newdev;
1430 num_devs++;
1431 }
1432 strcpy(chosen_name, mp->path);
1433 if (c->verbose > 0 || mddev == NULL ||
1434 strcmp(mddev, chosen_name) != 0)
1435 pr_err("Merging with already-assembled %s\n",
1436 chosen_name);
1437 mdfd = open_dev_excl(mp->devnm);
1438 } else {
1439 int trustworthy = FOREIGN;
1440 name = content->name;
1441 switch (st->ss->match_home(st, c->homehost)
1442 ?: st->ss->match_home(st, "any")) {
1443 case 1:
1444 trustworthy = LOCAL;
1445 name = strchr(content->name, ':');
1446 if (name)
1447 name++;
1448 else
1449 name = content->name;
1450 break;
1451 }
1452 if (!auto_assem)
1453 /* If the array is listed in mdadm.conf or on
1454 * command line, then we trust the name
1455 * even if the array doesn't look local
1456 */
1457 trustworthy = LOCAL;
1458
1459 if (name[0] == 0 &&
1460 content->array.level == LEVEL_CONTAINER) {
1461 name = content->text_version;
1462 trustworthy = METADATA;
1463 }
1464
1465 if (name[0] && trustworthy != LOCAL &&
1466 ! c->require_homehost &&
1467 conf_name_is_free(name))
1468 trustworthy = LOCAL;
1469
1470 if (trustworthy == LOCAL &&
1471 strchr(name, ':'))
1472 /* Ignore 'host:' prefix of name */
1473 name = strchr(name, ':')+1;
1474
1475 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1476 chosen_name, 0);
1477 }
1478 if (mdfd < 0) {
1479 st->ss->free_super(st);
1480 if (auto_assem)
1481 goto try_again;
1482 return 1;
1483 }
1484 mddev = chosen_name;
1485 if (pre_exist == NULL) {
1486 if (mddev_busy(fd2devnm(mdfd))) {
1487 pr_err("%s already active, cannot restart it!\n",
1488 mddev);
1489 for (tmpdev = devlist ;
1490 tmpdev && tmpdev->used != 1;
1491 tmpdev = tmpdev->next)
1492 ;
1493 if (tmpdev && auto_assem)
1494 pr_err("%s needed for %s...\n",
1495 mddev, tmpdev->devname);
1496 close(mdfd);
1497 mdfd = -3;
1498 st->ss->free_super(st);
1499 if (auto_assem)
1500 goto try_again;
1501 return 1;
1502 }
1503 /* just incase it was started but has no content */
1504 ioctl(mdfd, STOP_ARRAY, NULL);
1505 }
1506
1507 if (content != &info) {
1508 /* This is a member of a container. Try starting the array. */
1509 int err;
1510 err = assemble_container_content(st, mdfd, content, c,
1511 chosen_name, NULL);
1512 close(mdfd);
1513 return err;
1514 }
1515
1516 /* Ok, no bad inconsistancy, we can try updating etc */
1517 devices = xcalloc(num_devs, sizeof(*devices));
1518 devmap = xcalloc(num_devs, content->array.raid_disks);
1519 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1520 c, content, mdfd, mddev,
1521 &most_recent, &bestcnt, &best, inargv);
1522 if (devcnt < 0)
1523 return 1;
1524
1525 if (devcnt == 0) {
1526 pr_err("no devices found for %s\n",
1527 mddev);
1528 if (st)
1529 st->ss->free_super(st);
1530 close(mdfd);
1531 free(devices);
1532 free(devmap);
1533 return 1;
1534 }
1535
1536 if (c->update && strcmp(c->update, "byteorder")==0)
1537 st->minor_version = 90;
1538
1539 st->ss->getinfo_super(st, content, NULL);
1540 clean = content->array.state & 1;
1541
1542 /* now we have some devices that might be suitable.
1543 * I wonder how many
1544 */
1545 avail = xcalloc(content->array.raid_disks, 1);
1546 okcnt = 0;
1547 replcnt = 0;
1548 sparecnt=0;
1549 journalcnt=0;
1550 rebuilding_cnt=0;
1551 for (i=0; i< bestcnt; i++) {
1552 int j = best[i];
1553 int event_margin = 1; /* always allow a difference of '1'
1554 * like the kernel does
1555 */
1556 if (j < 0) continue;
1557 /* note: we ignore error flags in multipath arrays
1558 * as they don't make sense
1559 */
1560 if (content->array.level != LEVEL_MULTIPATH) {
1561 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1562 if (content->journal_device_required)
1563 journalcnt++;
1564 else /* unexpected journal, mark as faulty */
1565 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1566 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1567 if (!(devices[j].i.disk.state
1568 & (1<<MD_DISK_FAULTY))) {
1569 devices[j].uptodate = 1;
1570 sparecnt++;
1571 }
1572 continue;
1573 }
1574 }
1575 /* If this device thinks that 'most_recent' has failed, then
1576 * we must reject this device.
1577 */
1578 if (j != most_recent && !c->force &&
1579 content->array.raid_disks > 0 &&
1580 devices[most_recent].i.disk.raid_disk >= 0 &&
1581 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1582 if (c->verbose > -1)
1583 pr_err("ignoring %s as it reports %s as failed\n",
1584 devices[j].devname, devices[most_recent].devname);
1585 best[i] = -1;
1586 continue;
1587 }
1588 /* Require event counter to be same as, or just less than,
1589 * most recent. If it is bigger, it must be a stray spare and
1590 * should be ignored.
1591 */
1592 if (devices[j].i.events+event_margin >=
1593 devices[most_recent].i.events &&
1594 devices[j].i.events <=
1595 devices[most_recent].i.events
1596 ) {
1597 devices[j].uptodate = 1;
1598 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1599 journal_clean = 1;
1600 if (i < content->array.raid_disks * 2) {
1601 if (devices[j].i.recovery_start == MaxSector ||
1602 (content->reshape_active &&
1603 i >= content->array.raid_disks - content->delta_disks)) {
1604 if (!avail[i/2]) {
1605 okcnt++;
1606 avail[i/2]=1;
1607 } else
1608 replcnt++;
1609 } else
1610 rebuilding_cnt++;
1611 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1612 sparecnt++;
1613 }
1614 }
1615 free(devmap);
1616 if (c->force) {
1617 int force_ok = force_array(content, devices, best, bestcnt,
1618 avail, most_recent, st, c);
1619 okcnt += force_ok;
1620 if (force_ok)
1621 was_forced = 1;
1622 }
1623 /* Now we want to look at the superblock which the kernel will base things on
1624 * and compare the devices that we think are working with the devices that the
1625 * superblock thinks are working.
1626 * If there are differences and --force is given, then update this chosen
1627 * superblock.
1628 */
1629 chosen_drive = -1;
1630 st->ss->free_super(st);
1631 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1632 int j = best[i];
1633 int fd;
1634
1635 if (j<0)
1636 continue;
1637 if (!devices[j].uptodate)
1638 continue;
1639 if (devices[j].i.events < devices[most_recent].i.events)
1640 continue;
1641 chosen_drive = j;
1642 if ((fd=dev_open(devices[j].devname,
1643 devices[j].included ? O_RDONLY
1644 : (O_RDONLY|O_EXCL)))< 0) {
1645 pr_err("Cannot open %s: %s\n",
1646 devices[j].devname, strerror(errno));
1647 close(mdfd);
1648 free(devices);
1649 return 1;
1650 }
1651 if (st->ss->load_super(st,fd, NULL)) {
1652 close(fd);
1653 pr_err("RAID superblock has disappeared from %s\n",
1654 devices[j].devname);
1655 close(mdfd);
1656 free(devices);
1657 return 1;
1658 }
1659 close(fd);
1660 }
1661 if (st->sb == NULL) {
1662 pr_err("No suitable drives found for %s\n", mddev);
1663 close(mdfd);
1664 free(devices);
1665 return 1;
1666 }
1667 st->ss->getinfo_super(st, content, NULL);
1668 if (sysfs_init(content, mdfd, NULL)) {
1669 pr_err("Unable to initialize sysfs\n");
1670 close(mdfd);
1671 free(devices);
1672 return 1;
1673 }
1674
1675 /* after reload context, store journal_clean in context */
1676 content->journal_clean = journal_clean;
1677 for (i=0; i<bestcnt; i++) {
1678 int j = best[i];
1679 unsigned int desired_state;
1680
1681 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1682 desired_state = (1<<MD_DISK_JOURNAL);
1683 else if (i >= content->array.raid_disks * 2)
1684 desired_state = 0;
1685 else if (i & 1)
1686 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1687 else
1688 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1689
1690 if (j<0)
1691 continue;
1692 if (!devices[j].uptodate)
1693 continue;
1694
1695 devices[j].i.disk.state = desired_state;
1696 if (!(devices[j].i.array.state & 1))
1697 clean = 0;
1698
1699 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1700 c->verbose, 0, NULL)) {
1701 if (c->force) {
1702 if (c->verbose >= 0)
1703 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1704 j, mddev, devices[j].devname);
1705 change = 1;
1706 } else {
1707 if (c->verbose >= -1)
1708 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1709 i, mddev, devices[j].devname);
1710 }
1711 }
1712 #if 0
1713 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1714 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1715 i, mddev);
1716 }
1717 #endif
1718 }
1719 if (c->force && !clean &&
1720 !enough(content->array.level, content->array.raid_disks,
1721 content->array.layout, clean,
1722 avail)) {
1723 change += st->ss->update_super(st, content, "force-array",
1724 devices[chosen_drive].devname, c->verbose,
1725 0, NULL);
1726 was_forced = 1;
1727 clean = 1;
1728 }
1729
1730 if (change) {
1731 int fd;
1732 fd = dev_open(devices[chosen_drive].devname,
1733 devices[chosen_drive].included ?
1734 O_RDWR : (O_RDWR|O_EXCL));
1735 if (fd < 0) {
1736 pr_err("Could not open %s for write - cannot Assemble array.\n",
1737 devices[chosen_drive].devname);
1738 close(mdfd);
1739 free(devices);
1740 return 1;
1741 }
1742 if (st->ss->store_super(st, fd)) {
1743 close(fd);
1744 pr_err("Could not re-write superblock on %s\n",
1745 devices[chosen_drive].devname);
1746 close(mdfd);
1747 free(devices);
1748 return 1;
1749 }
1750 if (c->verbose >= 0)
1751 pr_err("Marking array %s as 'clean'\n",
1752 mddev);
1753 close(fd);
1754 }
1755
1756 /* If we are in the middle of a reshape we may need to restore saved data
1757 * that was moved aside due to the reshape overwriting live data
1758 * The code of doing this lives in Grow.c
1759 */
1760 if (content->reshape_active &&
1761 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1762 int err = 0;
1763 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1764 if (c->verbose > 0)
1765 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1766 chosen_name);
1767 if (!c->backup_file)
1768 c->backup_file = locate_backup(content->sys_name);
1769 enable_fds(bestcnt/2);
1770 for (i = 0; i < bestcnt/2; i++) {
1771 int j = best[i*2];
1772 if (j >= 0) {
1773 fdlist[i] = dev_open(devices[j].devname,
1774 devices[j].included
1775 ? O_RDWR : (O_RDWR|O_EXCL));
1776 if (fdlist[i] < 0) {
1777 pr_err("Could not open %s for write - cannot Assemble array.\n",
1778 devices[j].devname);
1779 err = 1;
1780 break;
1781 }
1782 } else
1783 fdlist[i] = -1;
1784 }
1785 if (!err) {
1786 if (st->ss->external && st->ss->recover_backup)
1787 err = st->ss->recover_backup(st, content);
1788 else
1789 err = Grow_restart(st, content, fdlist, bestcnt/2,
1790 c->backup_file, c->verbose > 0);
1791 if (err && c->invalid_backup) {
1792 if (c->verbose > 0)
1793 pr_err("continuing without restoring backup\n");
1794 err = 0;
1795 }
1796 }
1797 while (i>0) {
1798 i--;
1799 if (fdlist[i]>=0) close(fdlist[i]);
1800 }
1801 free(fdlist);
1802 if (err) {
1803 pr_err("Failed to restore critical section for reshape, sorry.\n");
1804 if (c->backup_file == NULL)
1805 cont_err("Possibly you needed to specify the --backup-file\n");
1806 close(mdfd);
1807 free(devices);
1808 return err;
1809 }
1810 }
1811
1812 /* Almost ready to actually *do* something */
1813 /* First, fill in the map, so that udev can find our name
1814 * as soon as we become active.
1815 */
1816 if (c->update && strcmp(c->update, "metadata")==0) {
1817 content->array.major_version = 1;
1818 content->array.minor_version = 0;
1819 strcpy(content->text_version, "1.0");
1820 }
1821
1822 map_update(&map, fd2devnm(mdfd), content->text_version,
1823 content->uuid, chosen_name);
1824
1825 rv = start_array(mdfd, mddev, content,
1826 st, ident, best, bestcnt,
1827 chosen_drive, devices, okcnt, sparecnt,
1828 rebuilding_cnt, journalcnt,
1829 c,
1830 clean, avail, start_partial_ok,
1831 pre_exist != NULL,
1832 was_forced);
1833 if (rv == 1 && !pre_exist)
1834 ioctl(mdfd, STOP_ARRAY, NULL);
1835 free(devices);
1836 map_unlock(&map);
1837 if (rv == 0) {
1838 wait_for(chosen_name, mdfd);
1839 close(mdfd);
1840 if (auto_assem) {
1841 int usecs = 1;
1842 /* There is a nasty race with 'mdadm --monitor'.
1843 * If it opens this device before we close it,
1844 * it gets an incomplete open on which IO
1845 * doesn't work and the capacity is
1846 * wrong.
1847 * If we reopen (to check for layered devices)
1848 * before --monitor closes, we loose.
1849 *
1850 * So: wait upto 1 second for there to be
1851 * a non-zero capacity.
1852 */
1853 while (usecs < 1000) {
1854 mdfd = open(mddev, O_RDONLY);
1855 if (mdfd >= 0) {
1856 unsigned long long size;
1857 if (get_dev_size(mdfd, NULL, &size) &&
1858 size > 0)
1859 break;
1860 close(mdfd);
1861 }
1862 usleep(usecs);
1863 usecs <<= 1;
1864 }
1865 }
1866 } else
1867 close(mdfd);
1868
1869 /* '2' means 'OK, but not started yet' */
1870 return rv == 2 ? 0 : rv;
1871 }
1872
1873 int assemble_container_content(struct supertype *st, int mdfd,
1874 struct mdinfo *content, struct context *c,
1875 char *chosen_name, int *result)
1876 {
1877 struct mdinfo *dev, *sra, *dev2;
1878 int working = 0, preexist = 0;
1879 int expansion = 0;
1880 int old_raid_disks;
1881 int start_reshape;
1882 char *avail;
1883 int err;
1884
1885 if (sysfs_init(content, mdfd, NULL)) {
1886 pr_err("Unable to initialize sysfs\n");
1887 return 1;
1888 }
1889
1890 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1891 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1892 if (content->array.major_version == -1 &&
1893 content->array.minor_version == -2 &&
1894 c->readonly &&
1895 content->text_version[0] == '/')
1896 content->text_version[0] = '-';
1897 if (sysfs_set_array(content, 9003) != 0) {
1898 sysfs_free(sra);
1899 return 1;
1900 }
1901 }
1902
1903 /* There are two types of reshape: container wide or sub-array specific
1904 * Check if metadata requests blocking container wide reshapes
1905 */
1906 start_reshape = (content->reshape_active &&
1907 !((content->reshape_active == CONTAINER_RESHAPE) &&
1908 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1909
1910 /* Block subarray here if it is under reshape now
1911 * Do not allow for any changes in this array
1912 */
1913 if (st->ss->external && content->recovery_blocked && start_reshape)
1914 block_subarray(content);
1915
1916 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1917 for (dev = content->devs; dev; dev = dev->next)
1918 if (dev2->disk.major == dev->disk.major &&
1919 dev2->disk.minor == dev->disk.minor)
1920 break;
1921 if (dev)
1922 continue;
1923 /* Don't want this one any more */
1924 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1925 errno == EBUSY) {
1926 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1927 sysfs_free(sra);
1928 return 1;
1929 }
1930 sysfs_set_str(sra, dev2, "state", "remove");
1931 }
1932 old_raid_disks = content->array.raid_disks - content->delta_disks;
1933 avail = xcalloc(content->array.raid_disks, 1);
1934 for (dev = content->devs; dev; dev = dev->next) {
1935 if (dev->disk.raid_disk >= 0)
1936 avail[dev->disk.raid_disk] = 1;
1937 if (sysfs_add_disk(content, dev, 1) == 0) {
1938 if (dev->disk.raid_disk >= old_raid_disks &&
1939 content->reshape_active)
1940 expansion++;
1941 else
1942 working++;
1943 } else if (errno == EEXIST)
1944 preexist++;
1945 }
1946 sysfs_free(sra);
1947 if (working + expansion == 0 && c->runstop <= 0) {
1948 free(avail);
1949 return 1;/* Nothing new, don't try to start */
1950 }
1951 map_update(NULL, fd2devnm(mdfd), content->text_version,
1952 content->uuid, chosen_name);
1953
1954 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1955 st->ss->validate_ppl) {
1956 content->array.state |= 1;
1957 err = 0;
1958
1959 for (dev = content->devs; dev; dev = dev->next) {
1960 int dfd;
1961 char *devpath;
1962 int ret;
1963
1964 ret = st->ss->validate_ppl(st, content, dev);
1965 if (ret == 0)
1966 continue;
1967
1968 if (ret < 0) {
1969 err = 1;
1970 break;
1971 }
1972
1973 if (!c->force) {
1974 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1975 chosen_name);
1976 content->array.state &= ~1;
1977 avail[dev->disk.raid_disk] = 0;
1978 break;
1979 }
1980
1981 /* have --force - overwrite the invalid ppl */
1982 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
1983 dfd = dev_open(devpath, O_RDWR);
1984 if (dfd < 0) {
1985 pr_err("Failed to open %s\n", devpath);
1986 err = 1;
1987 break;
1988 }
1989
1990 err = st->ss->write_init_ppl(st, content, dfd);
1991 close(dfd);
1992
1993 if (err)
1994 break;
1995 }
1996
1997 if (err) {
1998 free(avail);
1999 return err;
2000 }
2001 }
2002
2003 if (enough(content->array.level, content->array.raid_disks,
2004 content->array.layout, content->array.state & 1, avail) == 0) {
2005 if (c->export && result)
2006 *result |= INCR_NO;
2007 else if (c->verbose >= 0) {
2008 pr_err("%s assembled with %d device%s",
2009 chosen_name, preexist + working,
2010 preexist + working == 1 ? "":"s");
2011 if (preexist)
2012 fprintf(stderr, " (%d new)", working);
2013 fprintf(stderr, " but not started\n");
2014 }
2015 free(avail);
2016 return 1;
2017 }
2018 free(avail);
2019
2020 if (c->runstop <= 0 &&
2021 (working + preexist + expansion) <
2022 content->array.working_disks) {
2023 if (c->export && result)
2024 *result |= INCR_UNSAFE;
2025 else if (c->verbose >= 0) {
2026 pr_err("%s assembled with %d device%s",
2027 chosen_name, preexist + working,
2028 preexist + working == 1 ? "":"s");
2029 if (preexist)
2030 fprintf(stderr, " (%d new)", working);
2031 fprintf(stderr, " but not safe to start\n");
2032 }
2033 return 1;
2034 }
2035
2036
2037 if (start_reshape) {
2038 int spare = content->array.raid_disks + expansion;
2039 if (restore_backup(st, content,
2040 working,
2041 spare, &c->backup_file, c->verbose) == 1)
2042 return 1;
2043
2044 err = sysfs_set_str(content, NULL,
2045 "array_state", "readonly");
2046 if (err)
2047 return 1;
2048
2049 if (st->ss->external) {
2050 if (!mdmon_running(st->container_devnm))
2051 start_mdmon(st->container_devnm);
2052 ping_monitor(st->container_devnm);
2053 if (mdmon_running(st->container_devnm) &&
2054 st->update_tail == NULL)
2055 st->update_tail = &st->updates;
2056 }
2057
2058 err = Grow_continue(mdfd, st, content, c->backup_file,
2059 0, c->freeze_reshape);
2060 } else switch(content->array.level) {
2061 case LEVEL_LINEAR:
2062 case LEVEL_MULTIPATH:
2063 case 0:
2064 err = sysfs_set_str(content, NULL, "array_state",
2065 c->readonly ? "readonly" : "active");
2066 break;
2067 default:
2068 err = sysfs_set_str(content, NULL, "array_state",
2069 "readonly");
2070 /* start mdmon if needed. */
2071 if (!err) {
2072 if (!mdmon_running(st->container_devnm))
2073 start_mdmon(st->container_devnm);
2074 ping_monitor(st->container_devnm);
2075 }
2076 break;
2077 }
2078 if (!err)
2079 sysfs_set_safemode(content, content->safe_mode_delay);
2080
2081 /* Block subarray here if it is not reshaped now
2082 * It has be blocked a little later to allow mdmon to switch in
2083 * in to R/W state
2084 */
2085 if (st->ss->external && content->recovery_blocked &&
2086 !start_reshape)
2087 block_subarray(content);
2088
2089 if (c->export && result) {
2090 if (err)
2091 *result |= INCR_NO;
2092 else
2093 *result |= INCR_YES;
2094 } else if (c->verbose >= 0) {
2095 if (err)
2096 pr_err("array %s now has %d device%s",
2097 chosen_name, working + preexist,
2098 working + preexist == 1 ? "":"s");
2099 else
2100 pr_err("Started %s with %d device%s",
2101 chosen_name, working + preexist,
2102 working + preexist == 1 ? "":"s");
2103 if (preexist)
2104 fprintf(stderr, " (%d new)", working);
2105 if (expansion)
2106 fprintf(stderr, " ( + %d for expansion)",
2107 expansion);
2108 fprintf(stderr, "\n");
2109 }
2110 if (!err)
2111 wait_for(chosen_name, mdfd);
2112 return err;
2113 /* FIXME should have an O_EXCL and wait for read-auto */
2114 }
Unnamed repository; edit this file 'description' to name the repository.