]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
mdadm: improve the dlm locking mechanism for clustered raid
[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 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 } else if (devices[j].i.reshape_active !=
850 content->reshape_active ||
851 (devices[j].i.reshape_active &&
852 devices[j].i.reshape_progress !=
853 content->reshape_progress))
854 /* Here, it may be a source of data. If two
855 * devices claim different progresses, it
856 * means that reshape boundaries differ for
857 * their own devices. Kernel will only treat
858 * the first one as reshape progress and
859 * go on. It may cause disaster, so avoid it.
860 */
861 continue;
862 if (chosen_drive < 0 ||
863 devices[j].i.events
864 > devices[chosen_drive].i.events)
865 chosen_drive = j;
866 }
867 if (chosen_drive < 0)
868 break;
869 current_events = devices[chosen_drive].i.events;
870 add_another:
871 if (c->verbose >= 0)
872 pr_err("forcing event count in %s(%d) from %d upto %d\n",
873 devices[chosen_drive].devname,
874 devices[chosen_drive].i.disk.raid_disk,
875 (int)(devices[chosen_drive].i.events),
876 (int)(devices[most_recent].i.events));
877 fd = dev_open(devices[chosen_drive].devname,
878 devices[chosen_drive].included ? O_RDWR
879 : (O_RDWR|O_EXCL));
880 if (fd < 0) {
881 pr_err("Couldn't open %s for write - not updating\n",
882 devices[chosen_drive].devname);
883 devices[chosen_drive].i.events = 0;
884 continue;
885 }
886 tst = dup_super(st);
887 if (tst->ss->load_super(tst,fd, NULL)) {
888 close(fd);
889 pr_err("RAID superblock disappeared from %s - not updating.\n",
890 devices[chosen_drive].devname);
891 devices[chosen_drive].i.events = 0;
892 continue;
893 }
894 content->events = devices[most_recent].i.events;
895 tst->ss->update_super(tst, content, "force-one",
896 devices[chosen_drive].devname, c->verbose,
897 0, NULL);
898
899 if (tst->ss->store_super(tst, fd)) {
900 close(fd);
901 pr_err("Could not re-write superblock on %s\n",
902 devices[chosen_drive].devname);
903 devices[chosen_drive].i.events = 0;
904 tst->ss->free_super(tst);
905 continue;
906 }
907 close(fd);
908 devices[chosen_drive].i.events = devices[most_recent].i.events;
909 devices[chosen_drive].uptodate = 1;
910 avail[chosen_drive] = 1;
911 okcnt++;
912 tst->ss->free_super(tst);
913 /* If there are any other drives of the same vintage,
914 * add them in as well. We can't lose and we might gain
915 */
916 for (i = 0;
917 i < content->array.raid_disks * 2 && i < bestcnt ;
918 i += 2) {
919 int j = best[i];
920 if (j >= 0 &&
921 !devices[j].uptodate &&
922 devices[j].i.recovery_start == MaxSector &&
923 devices[j].i.events == current_events &&
924 ((!devices[j].i.reshape_active &&
925 !content->reshape_active) ||
926 (devices[j].i.reshape_active ==
927 content->reshape_active &&
928 devices[j].i.reshape_progress ==
929 content->reshape_progress))) {
930 chosen_drive = j;
931 goto add_another;
932 }
933 }
934 }
935 return okcnt;
936 }
937
938 static int start_array(int mdfd,
939 char *mddev,
940 struct mdinfo *content,
941 struct supertype *st,
942 struct mddev_ident *ident,
943 int *best, int bestcnt,
944 int chosen_drive,
945 struct devs *devices,
946 unsigned int okcnt,
947 unsigned int sparecnt,
948 unsigned int rebuilding_cnt,
949 unsigned int journalcnt,
950 struct context *c,
951 int clean, char *avail,
952 int start_partial_ok,
953 int err_ok,
954 int was_forced
955 )
956 {
957 int rv;
958 int i;
959 unsigned int req_cnt;
960
961 if (content->journal_device_required && (content->journal_clean == 0)) {
962 if (!c->force) {
963 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
964 return 1;
965 }
966 pr_err("Journal is missing or stale, starting array read only.\n");
967 c->readonly = 1;
968 }
969
970 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
971 clean = 1;
972
973 rv = set_array_info(mdfd, st, content);
974 if (rv && !err_ok) {
975 pr_err("failed to set array info for %s: %s\n",
976 mddev, strerror(errno));
977 return 1;
978 }
979 if (ident->bitmap_fd >= 0) {
980 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
981 pr_err("SET_BITMAP_FILE failed.\n");
982 return 1;
983 }
984 } else if (ident->bitmap_file) {
985 /* From config file */
986 int bmfd = open(ident->bitmap_file, O_RDWR);
987 if (bmfd < 0) {
988 pr_err("Could not open bitmap file %s\n",
989 ident->bitmap_file);
990 return 1;
991 }
992 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
993 pr_err("Failed to set bitmapfile for %s\n", mddev);
994 close(bmfd);
995 return 1;
996 }
997 close(bmfd);
998 }
999
1000 /* First, add the raid disks, but add the chosen one last */
1001 for (i = 0; i <= bestcnt; i++) {
1002 int j;
1003 if (i < bestcnt) {
1004 j = best[i];
1005 if (j == chosen_drive)
1006 continue;
1007 } else
1008 j = chosen_drive;
1009
1010 if (j >= 0 && !devices[j].included) {
1011 int dfd;
1012
1013 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1014 if (dfd >= 0) {
1015 remove_partitions(dfd);
1016 close(dfd);
1017 }
1018 rv = add_disk(mdfd, st, content, &devices[j].i);
1019
1020 if (rv) {
1021 pr_err("failed to add %s to %s: %s\n",
1022 devices[j].devname, mddev,
1023 strerror(errno));
1024 if (i < content->array.raid_disks * 2 ||
1025 i == bestcnt)
1026 okcnt--;
1027 else
1028 sparecnt--;
1029 } else if (c->verbose > 0) {
1030 pr_err("added %s to %s as %d%s%s\n",
1031 devices[j].devname, mddev,
1032 devices[j].i.disk.raid_disk,
1033 devices[j].uptodate?"":
1034 " (possibly out of date)",
1035 (devices[j].i.disk.state &
1036 (1<<MD_DISK_REPLACEMENT)) ?
1037 " replacement":"");
1038 }
1039 } else if (j >= 0) {
1040 if (c->verbose > 0)
1041 pr_err("%s is already in %s as %d\n",
1042 devices[j].devname, mddev,
1043 devices[j].i.disk.raid_disk);
1044 } else if (c->verbose > 0 &&
1045 i < content->array.raid_disks * 2 && (i & 1) == 0)
1046 pr_err("no uptodate device for slot %d of %s\n",
1047 i/2, mddev);
1048 }
1049
1050 if (content->array.level == LEVEL_CONTAINER) {
1051 if (c->verbose >= 0) {
1052 pr_err("Container %s has been assembled with %d drive%s",
1053 mddev, okcnt + sparecnt + journalcnt,
1054 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1055 if (okcnt < (unsigned)content->array.raid_disks)
1056 fprintf(stderr, " (out of %d)",
1057 content->array.raid_disks);
1058 fprintf(stderr, "\n");
1059 }
1060
1061 if (st->ss->validate_container) {
1062 struct mdinfo *devices_list;
1063 struct mdinfo *info_devices;
1064 unsigned int count;
1065
1066 devices_list = NULL;
1067 info_devices = xmalloc(sizeof(struct mdinfo) *
1068 (okcnt + sparecnt));
1069 for (count = 0; count < okcnt + sparecnt; count++) {
1070 info_devices[count] = devices[count].i;
1071 info_devices[count].next = devices_list;
1072 devices_list = &info_devices[count];
1073 }
1074 if (st->ss->validate_container(devices_list))
1075 pr_err("Mismatch detected!\n");
1076 free(info_devices);
1077 }
1078
1079 st->ss->free_super(st);
1080 sysfs_uevent(content, "change");
1081 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1082 /* Was partial, is still partial, so signal an error
1083 * to ensure we don't retry */
1084 return 1;
1085 return 0;
1086 }
1087
1088 /* Get number of in-sync devices according to the superblock.
1089 * We must have this number to start the array without -s or -R
1090 */
1091 req_cnt = content->array.working_disks;
1092
1093 if (c->runstop == 1 ||
1094 (c->runstop <= 0 &&
1095 (enough(content->array.level, content->array.raid_disks,
1096 content->array.layout, clean, avail) &&
1097 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1098 /* This array is good-to-go.
1099 * If a reshape is in progress then we might need to
1100 * continue monitoring it. In that case we start
1101 * it read-only and let the grow code make it writable.
1102 */
1103 int rv;
1104
1105 if (content->reshape_active &&
1106 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1107 content->delta_disks <= 0) {
1108 if (!c->backup_file) {
1109 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1110 mddev);
1111 pr_err("Please provided one with \"--backup-file=...\"\n");
1112 if (c->update &&
1113 strcmp(c->update, "revert-reshape") == 0)
1114 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1115 return 1;
1116 }
1117 rv = sysfs_set_str(content, NULL,
1118 "array_state", "readonly");
1119 if (rv == 0)
1120 rv = Grow_continue(mdfd, st, content,
1121 c->backup_file, 0,
1122 c->freeze_reshape);
1123 } else if (c->readonly &&
1124 sysfs_attribute_available(content, NULL,
1125 "array_state")) {
1126 rv = sysfs_set_str(content, NULL,
1127 "array_state", "readonly");
1128 } else
1129 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1130 reopen_mddev(mdfd); /* drop O_EXCL */
1131 if (rv == 0) {
1132 if (c->verbose >= 0) {
1133 pr_err("%s has been started with %d drive%s",
1134 mddev, okcnt, okcnt==1?"":"s");
1135 if (okcnt < (unsigned)content->array.raid_disks)
1136 fprintf(stderr, " (out of %d)",
1137 content->array.raid_disks);
1138 if (rebuilding_cnt)
1139 fprintf(stderr, "%s %d rebuilding",
1140 sparecnt?",":" and",
1141 rebuilding_cnt);
1142 if (sparecnt)
1143 fprintf(stderr, " and %d spare%s",
1144 sparecnt,
1145 sparecnt == 1 ? "" : "s");
1146 if (content->journal_clean)
1147 fprintf(stderr, " and %d journal",
1148 journalcnt);
1149 fprintf(stderr, ".\n");
1150 }
1151 if (content->reshape_active &&
1152 content->array.level >= 4 &&
1153 content->array.level <= 6) {
1154 /* might need to increase the size
1155 * of the stripe cache - default is 256
1156 */
1157 int chunk_size = content->array.chunk_size;
1158
1159 if (content->reshape_active &&
1160 content->new_chunk > chunk_size)
1161 chunk_size = content->new_chunk;
1162 if (256 < 4 * ((chunk_size+4065)/4096)) {
1163 struct mdinfo *sra;
1164
1165 sra = sysfs_read(mdfd, NULL, 0);
1166 if (sra)
1167 sysfs_set_num(sra, NULL,
1168 "stripe_cache_size",
1169 (4 * chunk_size / 4096) + 1);
1170 sysfs_free(sra);
1171 }
1172 }
1173 if (okcnt < (unsigned)content->array.raid_disks) {
1174 /* If any devices did not get added
1175 * because the kernel rejected them based
1176 * on event count, try adding them
1177 * again providing the action policy is
1178 * 're-add' or greater. The bitmap
1179 * might allow them to be included, or
1180 * they will become spares.
1181 */
1182 for (i = 0; i < bestcnt; i++) {
1183 int j = best[i];
1184 if (j >= 0 && !devices[j].uptodate) {
1185 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1186 continue;
1187 rv = add_disk(mdfd, st, content,
1188 &devices[j].i);
1189 if (rv == 0 && c->verbose >= 0)
1190 pr_err("%s has been re-added.\n",
1191 devices[j].devname);
1192 }
1193 }
1194 }
1195 if (content->array.level == 6 &&
1196 okcnt + 1 == (unsigned)content->array.raid_disks &&
1197 was_forced) {
1198 struct mdinfo *sra;
1199
1200 sra = sysfs_read(mdfd, NULL, 0);
1201 if (sra)
1202 sysfs_set_str(sra, NULL,
1203 "sync_action", "repair");
1204 sysfs_free(sra);
1205 }
1206 return 0;
1207 }
1208 pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno));
1209
1210 if (!enough(content->array.level, content->array.raid_disks,
1211 content->array.layout, 1, avail))
1212 pr_err("Not enough devices to start the array.\n");
1213 else if (!enough(content->array.level,
1214 content->array.raid_disks,
1215 content->array.layout, clean, avail))
1216 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1217
1218 return 1;
1219 }
1220 if (c->runstop == -1) {
1221 pr_err("%s assembled from %d drive%s",
1222 mddev, okcnt, okcnt == 1 ? "" : "s");
1223 if (okcnt != (unsigned)content->array.raid_disks)
1224 fprintf(stderr, " (out of %d)",
1225 content->array.raid_disks);
1226 fprintf(stderr, ", but not started.\n");
1227 return 2;
1228 }
1229 if (c->verbose >= -1) {
1230 pr_err("%s assembled from %d drive%s",
1231 mddev, okcnt, okcnt == 1 ? "" : "s");
1232 if (rebuilding_cnt)
1233 fprintf(stderr, "%s %d rebuilding",
1234 sparecnt ? "," : " and", rebuilding_cnt);
1235 if (sparecnt)
1236 fprintf(stderr, " and %d spare%s", sparecnt,
1237 sparecnt == 1 ? "" : "s");
1238 if (!enough(content->array.level, content->array.raid_disks,
1239 content->array.layout, 1, avail))
1240 fprintf(stderr, " - not enough to start the array.\n");
1241 else if (!enough(content->array.level,
1242 content->array.raid_disks,
1243 content->array.layout, clean, avail))
1244 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1245 else {
1246 if (req_cnt == (unsigned)content->array.raid_disks)
1247 fprintf(stderr, " - need all %d to start it",
1248 req_cnt);
1249 else
1250 fprintf(stderr, " - need %d to start", req_cnt);
1251 fprintf(stderr, " (use --run to insist).\n");
1252 }
1253 }
1254 return 1;
1255 }
1256
1257 int Assemble(struct supertype *st, char *mddev,
1258 struct mddev_ident *ident,
1259 struct mddev_dev *devlist,
1260 struct context *c)
1261 {
1262 /*
1263 * The task of Assemble is to find a collection of
1264 * devices that should (according to their superblocks)
1265 * form an array, and to give this collection to the MD driver.
1266 * In Linux-2.4 and later, this involves submitting a
1267 * SET_ARRAY_INFO ioctl with no arg - to prepare
1268 * the array - and then submit a number of
1269 * ADD_NEW_DISK ioctls to add disks into
1270 * the array. Finally RUN_ARRAY might
1271 * be submitted to start the array.
1272 *
1273 * Much of the work of Assemble is in finding and/or
1274 * checking the disks to make sure they look right.
1275 *
1276 * If mddev is not set, then scan must be set and we
1277 * read through the config file for dev+uuid mapping
1278 * We recurse, setting mddev, for each device that
1279 * - isn't running
1280 * - has a valid uuid (or any uuid if !uuidset)
1281 *
1282 * If mddev is set, we try to determine state of md.
1283 * check version - must be at least 0.90.0
1284 * check kernel version. must be at least 2.4.
1285 * If not, we can possibly fall back on START_ARRAY
1286 * Try to GET_ARRAY_INFO.
1287 * If possible, give up
1288 * If not, try to STOP_ARRAY just to make sure
1289 *
1290 * If !uuidset and scan, look in conf-file for uuid
1291 * If not found, give up
1292 * If !devlist and scan and uuidset, get list of devs from conf-file
1293 *
1294 * For each device:
1295 * Check superblock - discard if bad
1296 * Check uuid (set if we don't have one) - discard if no match
1297 * Check superblock similarity if we have a superblock - discard if different
1298 * Record events, devicenum
1299 * This should give us a list of devices for the array
1300 * We should collect the most recent event number
1301 *
1302 * Count disks with recent enough event count
1303 * While force && !enough disks
1304 * Choose newest rejected disks, update event count
1305 * mark clean and rewrite superblock
1306 * If recent kernel:
1307 * SET_ARRAY_INFO
1308 * foreach device with recent events : ADD_NEW_DISK
1309 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1310 * If old kernel:
1311 * Check the device numbers in superblock are right
1312 * update superblock if any changes
1313 * START_ARRAY
1314 *
1315 */
1316 int rv;
1317 int mdfd;
1318 int clean;
1319 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1320 ident->super_minor == UnSet && ident->name[0] == 0 &&
1321 (ident->container == NULL || ident->member == NULL));
1322 struct devs *devices;
1323 char *devmap;
1324 int *best = NULL; /* indexed by raid_disk */
1325 int bestcnt = 0;
1326 int devcnt;
1327 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1328 int journal_clean = 0;
1329 int i;
1330 int was_forced = 0;
1331 int most_recent = 0;
1332 int chosen_drive;
1333 int change = 0;
1334 int inargv = 0;
1335 int start_partial_ok = (c->runstop >= 0) &&
1336 (c->force || devlist==NULL || auto_assem);
1337 int num_devs;
1338 struct mddev_dev *tmpdev;
1339 struct mdinfo info;
1340 struct mdinfo *content = NULL;
1341 struct mdinfo *pre_exist = NULL;
1342 char *avail;
1343 char *name = NULL;
1344 char chosen_name[1024];
1345 struct map_ent *map = NULL;
1346 struct map_ent *mp;
1347
1348 /*
1349 * If any subdevs are listed, then any that don't
1350 * match ident are discarded. Remainder must all match and
1351 * become the array.
1352 * If no subdevs, then we scan all devices in the config file, but
1353 * there must be something in the identity
1354 */
1355
1356 if (!devlist &&
1357 ident->uuid_set == 0 &&
1358 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1359 ident->name[0] == 0 &&
1360 (ident->container == NULL || ident->member == NULL) &&
1361 ident->devices == NULL) {
1362 pr_err("No identity information available for %s - cannot assemble.\n",
1363 mddev ? mddev : "further assembly");
1364 return 1;
1365 }
1366
1367 if (devlist == NULL)
1368 devlist = conf_get_devs();
1369 else if (mddev)
1370 inargv = 1;
1371
1372 try_again:
1373 /* We come back here when doing auto-assembly and attempting some
1374 * set of devices failed. Those are now marked as ->used==2 and
1375 * we ignore them and try again
1376 */
1377 if (!st && ident->st)
1378 st = ident->st;
1379 if (c->verbose>0)
1380 pr_err("looking for devices for %s\n",
1381 mddev ? mddev : "further assembly");
1382
1383 content = &info;
1384 if (st && c->force)
1385 st->ignore_hw_compat = 1;
1386 num_devs = select_devices(devlist, ident, &st, &content, c,
1387 inargv, auto_assem);
1388 if (num_devs < 0)
1389 return 1;
1390
1391 if (!st || !st->sb || !content)
1392 return 2;
1393
1394 /* We have a full set of devices - we now need to find the
1395 * array device.
1396 * However there is a risk that we are racing with "mdadm -I"
1397 * and the array is already partially assembled - we will have
1398 * rejected any devices already in this address.
1399 * So we take a lock on the map file - to prevent further races -
1400 * and look for the uuid in there. If found and the array is
1401 * active, we abort. If found and the array is not active
1402 * we commit to that md device and add all the contained devices
1403 * to our list. We flag them so that we don't try to re-add,
1404 * but can remove if they turn out to not be wanted.
1405 */
1406 if (map_lock(&map))
1407 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1408 if (c->update && strcmp(c->update,"uuid") == 0)
1409 mp = NULL;
1410 else
1411 mp = map_by_uuid(&map, content->uuid);
1412 if (mp) {
1413 struct mdinfo *dv;
1414 /* array already exists. */
1415 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1416 if (pre_exist->array.level != UnSet) {
1417 pr_err("Found some drive for an array that is already active: %s\n",
1418 mp->path);
1419 pr_err("giving up.\n");
1420 return 1;
1421 }
1422 for (dv = pre_exist->devs; dv; dv = dv->next) {
1423 /* We want to add this device to our list,
1424 * but it could already be there if "mdadm -I"
1425 * started *after* we checked for O_EXCL.
1426 * If we add it to the top of the list
1427 * it will be preferred over later copies.
1428 */
1429 struct mddev_dev *newdev;
1430 char *devname = map_dev(dv->disk.major,
1431 dv->disk.minor,
1432 0);
1433 if (!devname)
1434 continue;
1435 newdev = xmalloc(sizeof(*newdev));
1436 newdev->devname = devname;
1437 newdev->disposition = 'I';
1438 newdev->used = 1;
1439 newdev->next = devlist;
1440 devlist = newdev;
1441 num_devs++;
1442 }
1443 strcpy(chosen_name, mp->path);
1444 if (c->verbose > 0 || mddev == NULL ||
1445 strcmp(mddev, chosen_name) != 0)
1446 pr_err("Merging with already-assembled %s\n",
1447 chosen_name);
1448 mdfd = open_dev_excl(mp->devnm);
1449 } else {
1450 int trustworthy = FOREIGN;
1451 name = content->name;
1452 switch (st->ss->match_home(st, c->homehost)
1453 ?: st->ss->match_home(st, "any")) {
1454 case 1:
1455 trustworthy = LOCAL;
1456 name = strchr(content->name, ':');
1457 if (name)
1458 name++;
1459 else
1460 name = content->name;
1461 break;
1462 }
1463 if (!auto_assem)
1464 /* If the array is listed in mdadm.conf or on
1465 * command line, then we trust the name
1466 * even if the array doesn't look local
1467 */
1468 trustworthy = LOCAL;
1469
1470 if (name[0] == 0 &&
1471 content->array.level == LEVEL_CONTAINER) {
1472 name = content->text_version;
1473 trustworthy = METADATA;
1474 }
1475
1476 if (name[0] && trustworthy != LOCAL &&
1477 ! c->require_homehost &&
1478 conf_name_is_free(name))
1479 trustworthy = LOCAL;
1480
1481 if (trustworthy == LOCAL &&
1482 strchr(name, ':'))
1483 /* Ignore 'host:' prefix of name */
1484 name = strchr(name, ':')+1;
1485
1486 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1487 chosen_name, 0);
1488 }
1489 if (mdfd < 0) {
1490 st->ss->free_super(st);
1491 if (auto_assem)
1492 goto try_again;
1493 return 1;
1494 }
1495 mddev = chosen_name;
1496 if (pre_exist == NULL) {
1497 if (mddev_busy(fd2devnm(mdfd))) {
1498 pr_err("%s already active, cannot restart it!\n",
1499 mddev);
1500 for (tmpdev = devlist ;
1501 tmpdev && tmpdev->used != 1;
1502 tmpdev = tmpdev->next)
1503 ;
1504 if (tmpdev && auto_assem)
1505 pr_err("%s needed for %s...\n",
1506 mddev, tmpdev->devname);
1507 close(mdfd);
1508 mdfd = -3;
1509 st->ss->free_super(st);
1510 if (auto_assem)
1511 goto try_again;
1512 return 1;
1513 }
1514 /* just incase it was started but has no content */
1515 ioctl(mdfd, STOP_ARRAY, NULL);
1516 }
1517
1518 if (content != &info) {
1519 /* This is a member of a container. Try starting the array. */
1520 int err;
1521 err = assemble_container_content(st, mdfd, content, c,
1522 chosen_name, NULL);
1523 close(mdfd);
1524 return err;
1525 }
1526
1527 /* Ok, no bad inconsistancy, we can try updating etc */
1528 devices = xcalloc(num_devs, sizeof(*devices));
1529 devmap = xcalloc(num_devs, content->array.raid_disks);
1530 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1531 c, content, mdfd, mddev,
1532 &most_recent, &bestcnt, &best, inargv);
1533 if (devcnt < 0)
1534 return 1;
1535
1536 if (devcnt == 0) {
1537 pr_err("no devices found for %s\n",
1538 mddev);
1539 if (st)
1540 st->ss->free_super(st);
1541 close(mdfd);
1542 free(devices);
1543 free(devmap);
1544 return 1;
1545 }
1546
1547 if (c->update && strcmp(c->update, "byteorder")==0)
1548 st->minor_version = 90;
1549
1550 st->ss->getinfo_super(st, content, NULL);
1551 clean = content->array.state & 1;
1552
1553 /* now we have some devices that might be suitable.
1554 * I wonder how many
1555 */
1556 avail = xcalloc(content->array.raid_disks, 1);
1557 okcnt = 0;
1558 replcnt = 0;
1559 sparecnt=0;
1560 journalcnt=0;
1561 rebuilding_cnt=0;
1562 for (i=0; i< bestcnt; i++) {
1563 int j = best[i];
1564 int event_margin = 1; /* always allow a difference of '1'
1565 * like the kernel does
1566 */
1567 if (j < 0) continue;
1568 /* note: we ignore error flags in multipath arrays
1569 * as they don't make sense
1570 */
1571 if (content->array.level != LEVEL_MULTIPATH) {
1572 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1573 if (content->journal_device_required)
1574 journalcnt++;
1575 else /* unexpected journal, mark as faulty */
1576 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1577 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1578 if (!(devices[j].i.disk.state
1579 & (1<<MD_DISK_FAULTY))) {
1580 devices[j].uptodate = 1;
1581 sparecnt++;
1582 }
1583 continue;
1584 }
1585 }
1586 /* If this device thinks that 'most_recent' has failed, then
1587 * we must reject this device.
1588 */
1589 if (j != most_recent && !c->force &&
1590 content->array.raid_disks > 0 &&
1591 devices[most_recent].i.disk.raid_disk >= 0 &&
1592 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1593 if (c->verbose > -1)
1594 pr_err("ignoring %s as it reports %s as failed\n",
1595 devices[j].devname, devices[most_recent].devname);
1596 best[i] = -1;
1597 continue;
1598 }
1599 /* Require event counter to be same as, or just less than,
1600 * most recent. If it is bigger, it must be a stray spare and
1601 * should be ignored.
1602 */
1603 if (devices[j].i.events+event_margin >=
1604 devices[most_recent].i.events &&
1605 devices[j].i.events <=
1606 devices[most_recent].i.events
1607 ) {
1608 devices[j].uptodate = 1;
1609 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1610 journal_clean = 1;
1611 if (i < content->array.raid_disks * 2) {
1612 if (devices[j].i.recovery_start == MaxSector ||
1613 (content->reshape_active &&
1614 i >= content->array.raid_disks - content->delta_disks)) {
1615 if (!avail[i/2]) {
1616 okcnt++;
1617 avail[i/2]=1;
1618 } else
1619 replcnt++;
1620 } else
1621 rebuilding_cnt++;
1622 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1623 sparecnt++;
1624 }
1625 }
1626 free(devmap);
1627 if (c->force) {
1628 int force_ok = force_array(content, devices, best, bestcnt,
1629 avail, most_recent, st, c);
1630 okcnt += force_ok;
1631 if (force_ok)
1632 was_forced = 1;
1633 }
1634 /* Now we want to look at the superblock which the kernel will base things on
1635 * and compare the devices that we think are working with the devices that the
1636 * superblock thinks are working.
1637 * If there are differences and --force is given, then update this chosen
1638 * superblock.
1639 */
1640 chosen_drive = -1;
1641 st->ss->free_super(st);
1642 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1643 int j = best[i];
1644 int fd;
1645
1646 if (j<0)
1647 continue;
1648 if (!devices[j].uptodate)
1649 continue;
1650 if (devices[j].i.events < devices[most_recent].i.events)
1651 continue;
1652 chosen_drive = j;
1653 if ((fd=dev_open(devices[j].devname,
1654 devices[j].included ? O_RDONLY
1655 : (O_RDONLY|O_EXCL)))< 0) {
1656 pr_err("Cannot open %s: %s\n",
1657 devices[j].devname, strerror(errno));
1658 close(mdfd);
1659 free(devices);
1660 return 1;
1661 }
1662 if (st->ss->load_super(st,fd, NULL)) {
1663 close(fd);
1664 pr_err("RAID superblock has disappeared from %s\n",
1665 devices[j].devname);
1666 close(mdfd);
1667 free(devices);
1668 return 1;
1669 }
1670 close(fd);
1671 }
1672 if (st->sb == NULL) {
1673 pr_err("No suitable drives found for %s\n", mddev);
1674 close(mdfd);
1675 free(devices);
1676 return 1;
1677 }
1678 st->ss->getinfo_super(st, content, NULL);
1679 if (sysfs_init(content, mdfd, NULL)) {
1680 pr_err("Unable to initialize sysfs\n");
1681 close(mdfd);
1682 free(devices);
1683 return 1;
1684 }
1685
1686 /* after reload context, store journal_clean in context */
1687 content->journal_clean = journal_clean;
1688 for (i=0; i<bestcnt; i++) {
1689 int j = best[i];
1690 unsigned int desired_state;
1691
1692 if (j < 0)
1693 continue;
1694 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1695 desired_state = (1<<MD_DISK_JOURNAL);
1696 else if (i >= content->array.raid_disks * 2)
1697 desired_state = 0;
1698 else if (i & 1)
1699 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1700 else
1701 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1702
1703 if (!devices[j].uptodate)
1704 continue;
1705
1706 devices[j].i.disk.state = desired_state;
1707 if (!(devices[j].i.array.state & 1))
1708 clean = 0;
1709
1710 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1711 c->verbose, 0, NULL)) {
1712 if (c->force) {
1713 if (c->verbose >= 0)
1714 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1715 j, mddev, devices[j].devname);
1716 change = 1;
1717 } else {
1718 if (c->verbose >= -1)
1719 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1720 i, mddev, devices[j].devname);
1721 }
1722 }
1723 #if 0
1724 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1725 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1726 i, mddev);
1727 }
1728 #endif
1729 }
1730 if (c->force && !clean &&
1731 !enough(content->array.level, content->array.raid_disks,
1732 content->array.layout, clean,
1733 avail)) {
1734 change += st->ss->update_super(st, content, "force-array",
1735 devices[chosen_drive].devname, c->verbose,
1736 0, NULL);
1737 was_forced = 1;
1738 clean = 1;
1739 }
1740
1741 if (change) {
1742 int fd;
1743 fd = dev_open(devices[chosen_drive].devname,
1744 devices[chosen_drive].included ?
1745 O_RDWR : (O_RDWR|O_EXCL));
1746 if (fd < 0) {
1747 pr_err("Could not open %s for write - cannot Assemble array.\n",
1748 devices[chosen_drive].devname);
1749 close(mdfd);
1750 free(devices);
1751 return 1;
1752 }
1753 if (st->ss->store_super(st, fd)) {
1754 close(fd);
1755 pr_err("Could not re-write superblock on %s\n",
1756 devices[chosen_drive].devname);
1757 close(mdfd);
1758 free(devices);
1759 return 1;
1760 }
1761 if (c->verbose >= 0)
1762 pr_err("Marking array %s as 'clean'\n",
1763 mddev);
1764 close(fd);
1765 }
1766
1767 /* If we are in the middle of a reshape we may need to restore saved data
1768 * that was moved aside due to the reshape overwriting live data
1769 * The code of doing this lives in Grow.c
1770 */
1771 if (content->reshape_active &&
1772 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1773 int err = 0;
1774 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1775 if (c->verbose > 0)
1776 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1777 chosen_name);
1778 if (!c->backup_file)
1779 c->backup_file = locate_backup(content->sys_name);
1780 enable_fds(bestcnt/2);
1781 for (i = 0; i < bestcnt/2; i++) {
1782 int j = best[i*2];
1783 if (j >= 0) {
1784 fdlist[i] = dev_open(devices[j].devname,
1785 devices[j].included
1786 ? O_RDWR : (O_RDWR|O_EXCL));
1787 if (fdlist[i] < 0) {
1788 pr_err("Could not open %s for write - cannot Assemble array.\n",
1789 devices[j].devname);
1790 err = 1;
1791 break;
1792 }
1793 } else
1794 fdlist[i] = -1;
1795 }
1796 if (!err) {
1797 if (st->ss->external && st->ss->recover_backup)
1798 err = st->ss->recover_backup(st, content);
1799 else
1800 err = Grow_restart(st, content, fdlist, bestcnt/2,
1801 c->backup_file, c->verbose > 0);
1802 if (err && c->invalid_backup) {
1803 if (c->verbose > 0)
1804 pr_err("continuing without restoring backup\n");
1805 err = 0;
1806 }
1807 }
1808 while (i>0) {
1809 i--;
1810 if (fdlist[i]>=0) close(fdlist[i]);
1811 }
1812 free(fdlist);
1813 if (err) {
1814 pr_err("Failed to restore critical section for reshape, sorry.\n");
1815 if (c->backup_file == NULL)
1816 cont_err("Possibly you needed to specify the --backup-file\n");
1817 close(mdfd);
1818 free(devices);
1819 return err;
1820 }
1821 }
1822
1823 /* Almost ready to actually *do* something */
1824 /* First, fill in the map, so that udev can find our name
1825 * as soon as we become active.
1826 */
1827 if (c->update && strcmp(c->update, "metadata")==0) {
1828 content->array.major_version = 1;
1829 content->array.minor_version = 0;
1830 strcpy(content->text_version, "1.0");
1831 }
1832
1833 map_update(&map, fd2devnm(mdfd), content->text_version,
1834 content->uuid, chosen_name);
1835
1836 rv = start_array(mdfd, mddev, content,
1837 st, ident, best, bestcnt,
1838 chosen_drive, devices, okcnt, sparecnt,
1839 rebuilding_cnt, journalcnt,
1840 c,
1841 clean, avail, start_partial_ok,
1842 pre_exist != NULL,
1843 was_forced);
1844 if (rv == 1 && !pre_exist)
1845 ioctl(mdfd, STOP_ARRAY, NULL);
1846 free(devices);
1847 map_unlock(&map);
1848 if (rv == 0) {
1849 wait_for(chosen_name, mdfd);
1850 close(mdfd);
1851 if (auto_assem) {
1852 int usecs = 1;
1853 /* There is a nasty race with 'mdadm --monitor'.
1854 * If it opens this device before we close it,
1855 * it gets an incomplete open on which IO
1856 * doesn't work and the capacity is
1857 * wrong.
1858 * If we reopen (to check for layered devices)
1859 * before --monitor closes, we loose.
1860 *
1861 * So: wait upto 1 second for there to be
1862 * a non-zero capacity.
1863 */
1864 while (usecs < 1000) {
1865 mdfd = open(mddev, O_RDONLY);
1866 if (mdfd >= 0) {
1867 unsigned long long size;
1868 if (get_dev_size(mdfd, NULL, &size) &&
1869 size > 0)
1870 break;
1871 close(mdfd);
1872 }
1873 usleep(usecs);
1874 usecs <<= 1;
1875 }
1876 }
1877 } else
1878 close(mdfd);
1879
1880 /* '2' means 'OK, but not started yet' */
1881 return rv == 2 ? 0 : rv;
1882 }
1883
1884 int assemble_container_content(struct supertype *st, int mdfd,
1885 struct mdinfo *content, struct context *c,
1886 char *chosen_name, int *result)
1887 {
1888 struct mdinfo *dev, *sra, *dev2;
1889 int working = 0, preexist = 0;
1890 int expansion = 0;
1891 int old_raid_disks;
1892 int start_reshape;
1893 char *avail;
1894 int err;
1895
1896 if (sysfs_init(content, mdfd, NULL)) {
1897 pr_err("Unable to initialize sysfs\n");
1898 return 1;
1899 }
1900
1901 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1902 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1903 if (content->array.major_version == -1 &&
1904 content->array.minor_version == -2 &&
1905 c->readonly &&
1906 content->text_version[0] == '/')
1907 content->text_version[0] = '-';
1908 if (sysfs_set_array(content, 9003) != 0) {
1909 sysfs_free(sra);
1910 return 1;
1911 }
1912 }
1913
1914 /* There are two types of reshape: container wide or sub-array specific
1915 * Check if metadata requests blocking container wide reshapes
1916 */
1917 start_reshape = (content->reshape_active &&
1918 !((content->reshape_active == CONTAINER_RESHAPE) &&
1919 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1920
1921 /* Block subarray here if it is under reshape now
1922 * Do not allow for any changes in this array
1923 */
1924 if (st->ss->external && content->recovery_blocked && start_reshape)
1925 block_subarray(content);
1926
1927 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1928 for (dev = content->devs; dev; dev = dev->next)
1929 if (dev2->disk.major == dev->disk.major &&
1930 dev2->disk.minor == dev->disk.minor)
1931 break;
1932 if (dev)
1933 continue;
1934 /* Don't want this one any more */
1935 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1936 errno == EBUSY) {
1937 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1938 sysfs_free(sra);
1939 return 1;
1940 }
1941 sysfs_set_str(sra, dev2, "state", "remove");
1942 }
1943 old_raid_disks = content->array.raid_disks - content->delta_disks;
1944 avail = xcalloc(content->array.raid_disks, 1);
1945 for (dev = content->devs; dev; dev = dev->next) {
1946 if (dev->disk.raid_disk >= 0)
1947 avail[dev->disk.raid_disk] = 1;
1948 if (sysfs_add_disk(content, dev, 1) == 0) {
1949 if (dev->disk.raid_disk >= old_raid_disks &&
1950 content->reshape_active)
1951 expansion++;
1952 else
1953 working++;
1954 } else if (errno == EEXIST)
1955 preexist++;
1956 }
1957 sysfs_free(sra);
1958 if (working + expansion == 0 && c->runstop <= 0) {
1959 free(avail);
1960 return 1;/* Nothing new, don't try to start */
1961 }
1962 map_update(NULL, fd2devnm(mdfd), content->text_version,
1963 content->uuid, chosen_name);
1964
1965 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1966 st->ss->validate_ppl) {
1967 content->array.state |= 1;
1968 err = 0;
1969
1970 for (dev = content->devs; dev; dev = dev->next) {
1971 int dfd;
1972 char *devpath;
1973 int ret;
1974
1975 ret = st->ss->validate_ppl(st, content, dev);
1976 if (ret == 0)
1977 continue;
1978
1979 if (ret < 0) {
1980 err = 1;
1981 break;
1982 }
1983
1984 if (!c->force) {
1985 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1986 chosen_name);
1987 content->array.state &= ~1;
1988 avail[dev->disk.raid_disk] = 0;
1989 break;
1990 }
1991
1992 /* have --force - overwrite the invalid ppl */
1993 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
1994 dfd = dev_open(devpath, O_RDWR);
1995 if (dfd < 0) {
1996 pr_err("Failed to open %s\n", devpath);
1997 err = 1;
1998 break;
1999 }
2000
2001 err = st->ss->write_init_ppl(st, content, dfd);
2002 close(dfd);
2003
2004 if (err)
2005 break;
2006 }
2007
2008 if (err) {
2009 free(avail);
2010 return err;
2011 }
2012 }
2013
2014 if (enough(content->array.level, content->array.raid_disks,
2015 content->array.layout, content->array.state & 1, avail) == 0) {
2016 if (c->export && result)
2017 *result |= INCR_NO;
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 started\n");
2025 }
2026 free(avail);
2027 return 1;
2028 }
2029 free(avail);
2030
2031 if (c->runstop <= 0 &&
2032 (working + preexist + expansion) <
2033 content->array.working_disks) {
2034 if (c->export && result)
2035 *result |= INCR_UNSAFE;
2036 else if (c->verbose >= 0) {
2037 pr_err("%s assembled with %d device%s",
2038 chosen_name, preexist + working,
2039 preexist + working == 1 ? "":"s");
2040 if (preexist)
2041 fprintf(stderr, " (%d new)", working);
2042 fprintf(stderr, " but not safe to start\n");
2043 }
2044 return 1;
2045 }
2046
2047
2048 if (start_reshape) {
2049 int spare = content->array.raid_disks + expansion;
2050 if (restore_backup(st, content,
2051 working,
2052 spare, &c->backup_file, c->verbose) == 1)
2053 return 1;
2054
2055 err = sysfs_set_str(content, NULL,
2056 "array_state", "readonly");
2057 if (err)
2058 return 1;
2059
2060 if (st->ss->external) {
2061 if (!mdmon_running(st->container_devnm))
2062 start_mdmon(st->container_devnm);
2063 ping_monitor(st->container_devnm);
2064 if (mdmon_running(st->container_devnm) &&
2065 st->update_tail == NULL)
2066 st->update_tail = &st->updates;
2067 }
2068
2069 err = Grow_continue(mdfd, st, content, c->backup_file,
2070 0, c->freeze_reshape);
2071 } else switch(content->array.level) {
2072 case LEVEL_LINEAR:
2073 case LEVEL_MULTIPATH:
2074 case 0:
2075 err = sysfs_set_str(content, NULL, "array_state",
2076 c->readonly ? "readonly" : "active");
2077 break;
2078 default:
2079 err = sysfs_set_str(content, NULL, "array_state",
2080 "readonly");
2081 /* start mdmon if needed. */
2082 if (!err) {
2083 if (!mdmon_running(st->container_devnm))
2084 start_mdmon(st->container_devnm);
2085 ping_monitor(st->container_devnm);
2086 }
2087 break;
2088 }
2089 if (!err)
2090 sysfs_set_safemode(content, content->safe_mode_delay);
2091
2092 /* Block subarray here if it is not reshaped now
2093 * It has be blocked a little later to allow mdmon to switch in
2094 * in to R/W state
2095 */
2096 if (st->ss->external && content->recovery_blocked &&
2097 !start_reshape)
2098 block_subarray(content);
2099
2100 if (c->export && result) {
2101 if (err)
2102 *result |= INCR_NO;
2103 else
2104 *result |= INCR_YES;
2105 } else if (c->verbose >= 0) {
2106 if (err)
2107 pr_err("array %s now has %d device%s",
2108 chosen_name, working + preexist,
2109 working + preexist == 1 ? "":"s");
2110 else
2111 pr_err("Started %s with %d device%s",
2112 chosen_name, working + preexist,
2113 working + preexist == 1 ? "":"s");
2114 if (preexist)
2115 fprintf(stderr, " (%d new)", working);
2116 if (expansion)
2117 fprintf(stderr, " ( + %d for expansion)",
2118 expansion);
2119 fprintf(stderr, "\n");
2120 }
2121 if (!err)
2122 wait_for(chosen_name, mdfd);
2123 return err;
2124 /* FIXME should have an O_EXCL and wait for read-auto */
2125 }