]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Assemble: free resources in load_devices
[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 if (tst)
285 tst->ss->free_super(tst);
286 return -1;
287 }
288
289 if (found_container) {
290 /* tmpdev is a container. We need to be either
291 * looking for a member, or auto-assembling
292 */
293 /* should be safe to try an exclusive open now, we
294 * have rejected anything that some other mdadm might
295 * be looking at
296 */
297 dfd = dev_open(devname, O_RDONLY | O_EXCL);
298 if (dfd < 0) {
299 if (report_mismatch)
300 pr_err("%s is busy - skipping\n", devname);
301 goto loop;
302 }
303 close(dfd);
304
305 if (ident->container && ident->container[0] != '/') {
306 /* we have a uuid */
307 int uuid[4];
308
309 content = *contentp;
310 tst->ss->getinfo_super(tst, content, NULL);
311
312 if (!parse_uuid(ident->container, uuid) ||
313 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
314 if (report_mismatch)
315 pr_err("%s has wrong UUID to be required container\n",
316 devname);
317 goto loop;
318 }
319 }
320 /* It is worth looking inside this container.
321 */
322 if (c->verbose > 0)
323 pr_err("looking in container %s\n",
324 devname);
325
326 for (content = tst->ss->container_content(tst, NULL);
327 content;
328 content = content->next) {
329
330 if (!ident_matches(ident, content, tst,
331 c->homehost, c->require_homehost,
332 c->update,
333 report_mismatch ? devname : NULL))
334 /* message already printed */;
335 else if (is_member_busy(content->text_version)) {
336 if (report_mismatch)
337 pr_err("member %s in %s is already assembled\n",
338 content->text_version,
339 devname);
340 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
341 /* do not assemble arrays with unsupported configurations */
342 pr_err("Cannot activate member %s in %s.\n",
343 content->text_version,
344 devname);
345 } else
346 break;
347 }
348 if (!content) {
349 tmpdev->used = 2;
350 goto loop; /* empty container */
351 }
352
353 st = tst; tst = NULL;
354 if (!auto_assem && inargv && tmpdev->next != NULL) {
355 pr_err("%s is a container, but is not only device given: confused and aborting\n",
356 devname);
357 st->ss->free_super(st);
358 dev_policy_free(pol);
359 domain_free(domains);
360 return -1;
361 }
362 if (c->verbose > 0)
363 pr_err("found match on member %s in %s\n",
364 content->text_version, devname);
365
366 /* make sure we finished the loop */
367 tmpdev = NULL;
368 goto loop;
369 } else {
370 content = *contentp;
371 tst->ss->getinfo_super(tst, content, NULL);
372
373 if (!ident_matches(ident, content, tst,
374 c->homehost, c->require_homehost,
375 c->update,
376 report_mismatch ? devname : NULL))
377 goto loop;
378
379 if (auto_assem) {
380 /* Never auto-assemble things that conflict
381 * with mdadm.conf in some way
382 */
383 struct mddev_ident *match;
384 int rv = 0;
385
386 match = conf_match(tst, content, devname,
387 report_mismatch ? c->verbose : -1,
388 &rv);
389 if (!match && rv == 2)
390 goto loop;
391 if (match && match->devname &&
392 strcasecmp(match->devname, "<ignore>") == 0) {
393 if (report_mismatch)
394 pr_err("%s is a member of an explicitly ignored array\n",
395 devname);
396 goto loop;
397 }
398 if (match && !ident_matches(match, content, tst,
399 c->homehost, c->require_homehost,
400 c->update,
401 report_mismatch ? devname : NULL))
402 /* Array exists in mdadm.conf but some
403 * details don't match, so reject it
404 */
405 goto loop;
406 }
407
408 /* should be safe to try an exclusive open now, we
409 * have rejected anything that some other mdadm might
410 * be looking at
411 */
412 dfd = dev_open(devname, O_RDONLY | O_EXCL);
413 if (dfd < 0) {
414 if (report_mismatch)
415 pr_err("%s is busy - skipping\n", devname);
416 goto loop;
417 }
418 close(dfd);
419
420 if (st == NULL)
421 st = dup_super(tst);
422 if (st->minor_version == -1)
423 st->minor_version = tst->minor_version;
424
425 if (memcmp(content->uuid, uuid_zero,
426 sizeof(int[4])) == 0) {
427 /* this is a floating spare. It cannot define
428 * an array unless there are no more arrays of
429 * this type to be found. It can be included
430 * in an array of this type though.
431 */
432 tmpdev->used = 3;
433 goto loop;
434 }
435
436 if (st->ss != tst->ss ||
437 st->minor_version != tst->minor_version ||
438 st->ss->compare_super(st, tst) != 0) {
439 /* Some mismatch. If exactly one array matches this host,
440 * we can resolve on that one.
441 * Or, if we are auto assembling, we just ignore the second
442 * for now.
443 */
444 if (auto_assem)
445 goto loop;
446 if (c->homehost) {
447 int first = st->ss->match_home(st, c->homehost);
448 int last = tst->ss->match_home(tst, c->homehost);
449 if (first != last &&
450 (first == 1 || last == 1)) {
451 /* We can do something */
452 if (first) {/* just ignore this one */
453 if (report_mismatch)
454 pr_err("%s misses out due to wrong homehost\n",
455 devname);
456 goto loop;
457 } else { /* reject all those sofar */
458 struct mddev_dev *td;
459 if (report_mismatch)
460 pr_err("%s overrides previous devices due to good homehost\n",
461 devname);
462 for (td=devlist; td != tmpdev; td=td->next)
463 if (td->used == 1)
464 td->used = 0;
465 tmpdev->used = 1;
466 goto loop;
467 }
468 }
469 }
470 pr_err("superblock on %s doesn't match others - assembly aborted\n",
471 devname);
472 tst->ss->free_super(tst);
473 st->ss->free_super(st);
474 dev_policy_free(pol);
475 domain_free(domains);
476 return -1;
477 }
478 tmpdev->used = 1;
479 }
480 loop:
481 /* Collect domain information from members only */
482 if (tmpdev && tmpdev->used == 1) {
483 if (!pol)
484 pol = devid_policy(rdev);
485 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
486 }
487 dev_policy_free(pol);
488 pol = NULL;
489 if (tst)
490 tst->ss->free_super(tst);
491 }
492
493 /* Check if we found some imsm spares but no members */
494 if ((auto_assem ||
495 (ident->uuid_set &&
496 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
497 (!st || !st->sb))
498 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
499 if (tmpdev->used != 3)
500 continue;
501 tmpdev->used = 1;
502 content = *contentp;
503
504 if (!st->sb) {
505 /* we need sb from one of the spares */
506 int dfd = dev_open(tmpdev->devname, O_RDONLY);
507 if (dfd < 0 ||
508 st->ss->load_super(st, dfd, NULL))
509 tmpdev->used = 2;
510 if (dfd > 0)
511 close(dfd);
512 }
513 }
514
515 /* Now reject spares that don't match domains of identified members */
516 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
517 if (tmpdev->used != 3)
518 continue;
519 if (!stat_is_blkdev(tmpdev->devname, &rdev)) {
520 tmpdev->used = 2;
521 } else {
522 struct dev_policy *pol = devid_policy(rdev);
523 int dt = domain_test(domains, pol, NULL);
524 if (inargv && dt != 0)
525 /* take this spare as domains match
526 * if there are any */
527 tmpdev->used = 1;
528 else if (!inargv && dt == 1)
529 /* device wasn't explicitly listed, so need
530 * explicit domain match - which we have */
531 tmpdev->used = 1;
532 else
533 /* if domains don't match mark as unused */
534 tmpdev->used = 0;
535 dev_policy_free(pol);
536 }
537 }
538 domain_free(domains);
539 *stp = st;
540 if (st && st->sb && content == *contentp)
541 st->ss->getinfo_super(st, content, NULL);
542 *contentp = content;
543
544 return num_devs;
545 }
546
547 struct devs {
548 char *devname;
549 int uptodate; /* set once we decide that this device is as
550 * recent as everything else in the array.
551 */
552 int included; /* set if the device is already in the array
553 * due to a previous '-I'
554 */
555 struct mdinfo i;
556 };
557
558 static int load_devices(struct devs *devices, char *devmap,
559 struct mddev_ident *ident, struct supertype **stp,
560 struct mddev_dev *devlist, struct context *c,
561 struct mdinfo *content,
562 int mdfd, char *mddev,
563 int *most_recentp, int *bestcntp, int **bestp,
564 int inargv)
565 {
566 struct mddev_dev *tmpdev;
567 int devcnt = 0;
568 int nextspare = 0;
569 int bitmap_done = 0;
570 int most_recent = -1;
571 int bestcnt = 0;
572 int *best = *bestp;
573 struct supertype *st = *stp;
574
575 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
576 char *devname = tmpdev->devname;
577 struct stat stb;
578 struct supertype *tst;
579 int i;
580 int dfd;
581
582 if (tmpdev->used != 1)
583 continue;
584 /* looks like a good enough match to update the super block if needed */
585 if (c->update) {
586 /* prepare useful information in info structures */
587 struct stat stb2;
588 int err;
589 fstat(mdfd, &stb2);
590
591 if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
592 random_uuid((__u8 *)ident->uuid);
593
594 if (strcmp(c->update, "ppl") == 0 &&
595 ident->bitmap_fd >= 0) {
596 pr_err("PPL is not compatible with bitmap\n");
597 close(mdfd);
598 free(devices);
599 free(devmap);
600 return -1;
601 }
602
603 dfd = dev_open(devname,
604 tmpdev->disposition == 'I'
605 ? O_RDWR : (O_RDWR|O_EXCL));
606
607 tst = dup_super(st);
608 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
609 pr_err("cannot re-read metadata from %s - aborting\n",
610 devname);
611 if (dfd >= 0)
612 close(dfd);
613 close(mdfd);
614 free(devices);
615 free(devmap);
616 tst->ss->free_super(tst);
617 free(tst);
618 *stp = st;
619 return -1;
620 }
621 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
622
623 memcpy(content->uuid, ident->uuid, 16);
624 strcpy(content->name, ident->name);
625 content->array.md_minor = minor(stb2.st_rdev);
626
627 if (strcmp(c->update, "byteorder") == 0)
628 err = 0;
629 else if (strcmp(c->update, "home-cluster") == 0) {
630 tst->cluster_name = c->homecluster;
631 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
632 } else if (strcmp(c->update, "nodes") == 0) {
633 tst->nodes = c->nodes;
634 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
635 } else if (strcmp(c->update, "revert-reshape") == 0 &&
636 c->invalid_backup)
637 err = tst->ss->update_super(tst, content,
638 "revert-reshape-nobackup",
639 devname, c->verbose,
640 ident->uuid_set,
641 c->homehost);
642 else
643 err = tst->ss->update_super(tst, content, c->update,
644 devname, c->verbose,
645 ident->uuid_set,
646 c->homehost);
647 if (err < 0) {
648 if (err == -1)
649 pr_err("--update=%s not understood for %s metadata\n",
650 c->update, tst->ss->name);
651 tst->ss->free_super(tst);
652 free(tst);
653 close(mdfd);
654 close(dfd);
655 free(devices);
656 free(devmap);
657 *stp = st;
658 return -1;
659 }
660 if (strcmp(c->update, "uuid")==0 &&
661 !ident->uuid_set) {
662 ident->uuid_set = 1;
663 memcpy(ident->uuid, content->uuid, 16);
664 }
665 if (tst->ss->store_super(tst, dfd))
666 pr_err("Could not re-write superblock on %s.\n",
667 devname);
668
669 if (strcmp(c->update, "uuid")==0 &&
670 ident->bitmap_fd >= 0 && !bitmap_done) {
671 if (bitmap_update_uuid(ident->bitmap_fd,
672 content->uuid,
673 tst->ss->swapuuid) != 0)
674 pr_err("Could not update uuid on external bitmap.\n");
675 else
676 bitmap_done = 1;
677 }
678 } else {
679 dfd = dev_open(devname,
680 tmpdev->disposition == 'I'
681 ? O_RDWR : (O_RDWR|O_EXCL));
682 tst = dup_super(st);
683
684 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
685 pr_err("cannot re-read metadata from %s - aborting\n",
686 devname);
687 if (dfd >= 0)
688 close(dfd);
689 close(mdfd);
690 free(devices);
691 free(devmap);
692 tst->ss->free_super(tst);
693 free(tst);
694 *stp = st;
695 return -1;
696 }
697 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
698 }
699
700 fstat(dfd, &stb);
701 close(dfd);
702
703 if (c->verbose > 0)
704 pr_err("%s is identified as a member of %s, slot %d%s.\n",
705 devname, mddev, content->disk.raid_disk,
706 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
707 devices[devcnt].devname = devname;
708 devices[devcnt].uptodate = 0;
709 devices[devcnt].included = (tmpdev->disposition == 'I');
710 devices[devcnt].i = *content;
711 devices[devcnt].i.disk.major = major(stb.st_rdev);
712 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
713
714 if (devices[devcnt].i.disk.state == 6) {
715 if (most_recent < 0 ||
716 devices[devcnt].i.events
717 > devices[most_recent].i.events) {
718 struct supertype *tmp = tst;
719 tst = st;
720 st = tmp;
721 most_recent = devcnt;
722 }
723 }
724 tst->ss->free_super(tst);
725 free(tst);
726
727 if (content->array.level == LEVEL_MULTIPATH)
728 /* with multipath, the raid_disk from the superblock is meaningless */
729 i = devcnt;
730 else
731 i = devices[devcnt].i.disk.raid_disk;
732 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
733 if (nextspare < content->array.raid_disks*2)
734 nextspare = content->array.raid_disks*2;
735 i = nextspare++;
736 } else {
737 /* i is raid_disk - double it so there is room for
738 * replacements */
739 i *= 2;
740 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
741 i++;
742 if (i >= content->array.raid_disks*2 &&
743 i >= nextspare)
744 nextspare = i+1;
745 }
746 if (i < 10000) {
747 if (i >= bestcnt) {
748 int newbestcnt = i+10;
749 int *newbest = xmalloc(sizeof(int)*newbestcnt);
750 int c;
751 for (c=0; c < newbestcnt; c++)
752 if (c < bestcnt)
753 newbest[c] = best[c];
754 else
755 newbest[c] = -1;
756 if (best)free(best);
757 best = newbest;
758 bestcnt = newbestcnt;
759 }
760 if (best[i] >=0 &&
761 devices[best[i]].i.events ==
762 devices[devcnt].i.events &&
763 (devices[best[i]].i.disk.minor !=
764 devices[devcnt].i.disk.minor) &&
765 st->ss == &super0 &&
766 content->array.level != LEVEL_MULTIPATH) {
767 /* two different devices with identical superblock.
768 * Could be a mis-detection caused by overlapping
769 * partitions. fail-safe.
770 */
771 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
772 " If they are really different, please --zero the superblock on one\n"
773 " If they are the same or overlap, please remove one from %s.\n",
774 devices[best[i]].devname, devname,
775 inargv ? "the list" :
776 "the\n DEVICE list in mdadm.conf"
777 );
778 close(mdfd);
779 free(devices);
780 free(devmap);
781 *stp = st;
782 return -1;
783 }
784 if (best[i] == -1 || (devices[best[i]].i.events
785 < devices[devcnt].i.events))
786 best[i] = devcnt;
787 else if (st->ss == &super_imsm)
788 best[i+1] = devcnt;
789 }
790 devcnt++;
791 }
792 if (most_recent >= 0)
793 *most_recentp = most_recent;
794 *bestcntp = bestcnt;
795 *bestp = best;
796 *stp = st;
797 return devcnt;
798 }
799
800 static int force_array(struct mdinfo *content,
801 struct devs *devices,
802 int *best, int bestcnt, char *avail,
803 int most_recent,
804 struct supertype *st,
805 struct context *c)
806 {
807 int okcnt = 0;
808 while (!enough(content->array.level, content->array.raid_disks,
809 content->array.layout, 1,
810 avail) ||
811 (content->reshape_active && content->delta_disks > 0 &&
812 !enough(content->array.level, (content->array.raid_disks
813 - content->delta_disks),
814 content->new_layout, 1, avail))) {
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 } else if (devices[j].i.reshape_active !=
857 content->reshape_active ||
858 (devices[j].i.reshape_active &&
859 devices[j].i.reshape_progress !=
860 content->reshape_progress))
861 /* Here, it may be a source of data. If two
862 * devices claim different progresses, it
863 * means that reshape boundaries differ for
864 * their own devices. Kernel will only treat
865 * the first one as reshape progress and
866 * go on. It may cause disaster, so avoid it.
867 */
868 continue;
869 if (chosen_drive < 0 ||
870 devices[j].i.events
871 > devices[chosen_drive].i.events)
872 chosen_drive = j;
873 }
874 if (chosen_drive < 0)
875 break;
876 current_events = devices[chosen_drive].i.events;
877 add_another:
878 if (c->verbose >= 0)
879 pr_err("forcing event count in %s(%d) from %d upto %d\n",
880 devices[chosen_drive].devname,
881 devices[chosen_drive].i.disk.raid_disk,
882 (int)(devices[chosen_drive].i.events),
883 (int)(devices[most_recent].i.events));
884 fd = dev_open(devices[chosen_drive].devname,
885 devices[chosen_drive].included ? O_RDWR
886 : (O_RDWR|O_EXCL));
887 if (fd < 0) {
888 pr_err("Couldn't open %s for write - not updating\n",
889 devices[chosen_drive].devname);
890 devices[chosen_drive].i.events = 0;
891 continue;
892 }
893 tst = dup_super(st);
894 if (tst->ss->load_super(tst,fd, NULL)) {
895 close(fd);
896 pr_err("RAID superblock disappeared from %s - not updating.\n",
897 devices[chosen_drive].devname);
898 devices[chosen_drive].i.events = 0;
899 continue;
900 }
901 content->events = devices[most_recent].i.events;
902 tst->ss->update_super(tst, content, "force-one",
903 devices[chosen_drive].devname, c->verbose,
904 0, NULL);
905
906 if (tst->ss->store_super(tst, fd)) {
907 close(fd);
908 pr_err("Could not re-write superblock on %s\n",
909 devices[chosen_drive].devname);
910 devices[chosen_drive].i.events = 0;
911 tst->ss->free_super(tst);
912 continue;
913 }
914 close(fd);
915 devices[chosen_drive].i.events = devices[most_recent].i.events;
916 devices[chosen_drive].uptodate = 1;
917 avail[chosen_drive] = 1;
918 okcnt++;
919 tst->ss->free_super(tst);
920 /* If there are any other drives of the same vintage,
921 * add them in as well. We can't lose and we might gain
922 */
923 for (i = 0;
924 i < content->array.raid_disks * 2 && i < bestcnt ;
925 i += 2) {
926 int j = best[i];
927 if (j >= 0 &&
928 !devices[j].uptodate &&
929 devices[j].i.recovery_start == MaxSector &&
930 devices[j].i.events == current_events &&
931 ((!devices[j].i.reshape_active &&
932 !content->reshape_active) ||
933 (devices[j].i.reshape_active ==
934 content->reshape_active &&
935 devices[j].i.reshape_progress ==
936 content->reshape_progress))) {
937 chosen_drive = j;
938 goto add_another;
939 }
940 }
941 }
942 return okcnt;
943 }
944
945 static int start_array(int mdfd,
946 char *mddev,
947 struct mdinfo *content,
948 struct supertype *st,
949 struct mddev_ident *ident,
950 int *best, int bestcnt,
951 int chosen_drive,
952 struct devs *devices,
953 unsigned int okcnt,
954 unsigned int sparecnt,
955 unsigned int rebuilding_cnt,
956 unsigned int journalcnt,
957 struct context *c,
958 int clean, char *avail,
959 int start_partial_ok,
960 int err_ok,
961 int was_forced
962 )
963 {
964 int rv;
965 int i;
966 unsigned int req_cnt;
967
968 if (content->journal_device_required && (content->journal_clean == 0)) {
969 if (!c->force) {
970 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
971 return 1;
972 }
973 pr_err("Journal is missing or stale, starting array read only.\n");
974 c->readonly = 1;
975 }
976
977 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
978 clean = 1;
979
980 rv = set_array_info(mdfd, st, content);
981 if (rv && !err_ok) {
982 pr_err("failed to set array info for %s: %s\n",
983 mddev, strerror(errno));
984 return 1;
985 }
986 if (ident->bitmap_fd >= 0) {
987 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
988 pr_err("SET_BITMAP_FILE failed.\n");
989 return 1;
990 }
991 } else if (ident->bitmap_file) {
992 /* From config file */
993 int bmfd = open(ident->bitmap_file, O_RDWR);
994 if (bmfd < 0) {
995 pr_err("Could not open bitmap file %s\n",
996 ident->bitmap_file);
997 return 1;
998 }
999 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1000 pr_err("Failed to set bitmapfile for %s\n", mddev);
1001 close(bmfd);
1002 return 1;
1003 }
1004 close(bmfd);
1005 }
1006
1007 /* First, add the raid disks, but add the chosen one last */
1008 for (i = 0; i <= bestcnt; i++) {
1009 int j;
1010 if (i < bestcnt) {
1011 j = best[i];
1012 if (j == chosen_drive)
1013 continue;
1014 } else
1015 j = chosen_drive;
1016
1017 if (j >= 0 && !devices[j].included) {
1018 int dfd;
1019
1020 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1021 if (dfd >= 0) {
1022 remove_partitions(dfd);
1023 close(dfd);
1024 }
1025 rv = add_disk(mdfd, st, content, &devices[j].i);
1026
1027 if (rv) {
1028 pr_err("failed to add %s to %s: %s\n",
1029 devices[j].devname, mddev,
1030 strerror(errno));
1031 if (i < content->array.raid_disks * 2 ||
1032 i == bestcnt)
1033 okcnt--;
1034 else
1035 sparecnt--;
1036 } else if (c->verbose > 0) {
1037 pr_err("added %s to %s as %d%s%s\n",
1038 devices[j].devname, mddev,
1039 devices[j].i.disk.raid_disk,
1040 devices[j].uptodate?"":
1041 " (possibly out of date)",
1042 (devices[j].i.disk.state &
1043 (1<<MD_DISK_REPLACEMENT)) ?
1044 " replacement":"");
1045 }
1046 } else if (j >= 0) {
1047 if (c->verbose > 0)
1048 pr_err("%s is already in %s as %d\n",
1049 devices[j].devname, mddev,
1050 devices[j].i.disk.raid_disk);
1051 } else if (c->verbose > 0 &&
1052 i < content->array.raid_disks * 2 && (i & 1) == 0)
1053 pr_err("no uptodate device for slot %d of %s\n",
1054 i/2, mddev);
1055 }
1056
1057 if (content->array.level == LEVEL_CONTAINER) {
1058 if (c->verbose >= 0) {
1059 pr_err("Container %s has been assembled with %d drive%s",
1060 mddev, okcnt + sparecnt + journalcnt,
1061 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1062 if (okcnt < (unsigned)content->array.raid_disks)
1063 fprintf(stderr, " (out of %d)",
1064 content->array.raid_disks);
1065 fprintf(stderr, "\n");
1066 }
1067
1068 if (st->ss->validate_container) {
1069 struct mdinfo *devices_list;
1070 struct mdinfo *info_devices;
1071 unsigned int count;
1072
1073 devices_list = NULL;
1074 info_devices = xmalloc(sizeof(struct mdinfo) *
1075 (okcnt + sparecnt));
1076 for (count = 0; count < okcnt + sparecnt; count++) {
1077 info_devices[count] = devices[count].i;
1078 info_devices[count].next = devices_list;
1079 devices_list = &info_devices[count];
1080 }
1081 if (st->ss->validate_container(devices_list))
1082 pr_err("Mismatch detected!\n");
1083 free(info_devices);
1084 }
1085
1086 st->ss->free_super(st);
1087 sysfs_uevent(content, "change");
1088 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1089 /* Was partial, is still partial, so signal an error
1090 * to ensure we don't retry */
1091 return 1;
1092 return 0;
1093 }
1094
1095 /* Get number of in-sync devices according to the superblock.
1096 * We must have this number to start the array without -s or -R
1097 */
1098 req_cnt = content->array.working_disks;
1099
1100 if (c->runstop == 1 ||
1101 (c->runstop <= 0 &&
1102 (enough(content->array.level, content->array.raid_disks,
1103 content->array.layout, clean, avail) &&
1104 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1105 /* This array is good-to-go.
1106 * If a reshape is in progress then we might need to
1107 * continue monitoring it. In that case we start
1108 * it read-only and let the grow code make it writable.
1109 */
1110 int rv;
1111
1112 if (content->reshape_active &&
1113 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1114 content->delta_disks <= 0) {
1115 if (!c->backup_file) {
1116 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1117 mddev);
1118 pr_err("Please provided one with \"--backup-file=...\"\n");
1119 if (c->update &&
1120 strcmp(c->update, "revert-reshape") == 0)
1121 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1122 return 1;
1123 }
1124 rv = sysfs_set_str(content, NULL,
1125 "array_state", "readonly");
1126 if (rv == 0)
1127 rv = Grow_continue(mdfd, st, content,
1128 c->backup_file, 0,
1129 c->freeze_reshape);
1130 } else if (c->readonly &&
1131 sysfs_attribute_available(content, NULL,
1132 "array_state")) {
1133 rv = sysfs_set_str(content, NULL,
1134 "array_state", "readonly");
1135 } else
1136 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1137 reopen_mddev(mdfd); /* drop O_EXCL */
1138 if (rv == 0) {
1139 if (c->verbose >= 0) {
1140 pr_err("%s has been started with %d drive%s",
1141 mddev, okcnt, okcnt==1?"":"s");
1142 if (okcnt < (unsigned)content->array.raid_disks)
1143 fprintf(stderr, " (out of %d)",
1144 content->array.raid_disks);
1145 if (rebuilding_cnt)
1146 fprintf(stderr, "%s %d rebuilding",
1147 sparecnt?",":" and",
1148 rebuilding_cnt);
1149 if (sparecnt)
1150 fprintf(stderr, " and %d spare%s",
1151 sparecnt,
1152 sparecnt == 1 ? "" : "s");
1153 if (content->journal_clean)
1154 fprintf(stderr, " and %d journal",
1155 journalcnt);
1156 fprintf(stderr, ".\n");
1157 }
1158 if (content->reshape_active &&
1159 content->array.level >= 4 &&
1160 content->array.level <= 6) {
1161 /* might need to increase the size
1162 * of the stripe cache - default is 256
1163 */
1164 int chunk_size = content->array.chunk_size;
1165
1166 if (content->reshape_active &&
1167 content->new_chunk > chunk_size)
1168 chunk_size = content->new_chunk;
1169 if (256 < 4 * ((chunk_size+4065)/4096)) {
1170 struct mdinfo *sra;
1171
1172 sra = sysfs_read(mdfd, NULL, 0);
1173 if (sra)
1174 sysfs_set_num(sra, NULL,
1175 "stripe_cache_size",
1176 (4 * chunk_size / 4096) + 1);
1177 sysfs_free(sra);
1178 }
1179 }
1180 if (okcnt < (unsigned)content->array.raid_disks) {
1181 /* If any devices did not get added
1182 * because the kernel rejected them based
1183 * on event count, try adding them
1184 * again providing the action policy is
1185 * 're-add' or greater. The bitmap
1186 * might allow them to be included, or
1187 * they will become spares.
1188 */
1189 for (i = 0; i < bestcnt; i++) {
1190 int j = best[i];
1191 if (j >= 0 && !devices[j].uptodate) {
1192 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1193 continue;
1194 rv = add_disk(mdfd, st, content,
1195 &devices[j].i);
1196 if (rv == 0 && c->verbose >= 0)
1197 pr_err("%s has been re-added.\n",
1198 devices[j].devname);
1199 }
1200 }
1201 }
1202 if (content->array.level == 6 &&
1203 okcnt + 1 == (unsigned)content->array.raid_disks &&
1204 was_forced) {
1205 struct mdinfo *sra;
1206
1207 sra = sysfs_read(mdfd, NULL, 0);
1208 if (sra)
1209 sysfs_set_str(sra, NULL,
1210 "sync_action", "repair");
1211 sysfs_free(sra);
1212 }
1213 return 0;
1214 }
1215 pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno));
1216
1217 if (!enough(content->array.level, content->array.raid_disks,
1218 content->array.layout, 1, avail))
1219 pr_err("Not enough devices to start the array.\n");
1220 else if (!enough(content->array.level,
1221 content->array.raid_disks,
1222 content->array.layout, clean, avail))
1223 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1224
1225 return 1;
1226 }
1227 if (c->runstop == -1) {
1228 pr_err("%s assembled from %d drive%s",
1229 mddev, okcnt, okcnt == 1 ? "" : "s");
1230 if (okcnt != (unsigned)content->array.raid_disks)
1231 fprintf(stderr, " (out of %d)",
1232 content->array.raid_disks);
1233 fprintf(stderr, ", but not started.\n");
1234 return 2;
1235 }
1236 if (c->verbose >= -1) {
1237 pr_err("%s assembled from %d drive%s",
1238 mddev, okcnt, okcnt == 1 ? "" : "s");
1239 if (rebuilding_cnt)
1240 fprintf(stderr, "%s %d rebuilding",
1241 sparecnt ? "," : " and", rebuilding_cnt);
1242 if (sparecnt)
1243 fprintf(stderr, " and %d spare%s", sparecnt,
1244 sparecnt == 1 ? "" : "s");
1245 if (!enough(content->array.level, content->array.raid_disks,
1246 content->array.layout, 1, avail))
1247 fprintf(stderr, " - not enough to start the array.\n");
1248 else if (!enough(content->array.level,
1249 content->array.raid_disks,
1250 content->array.layout, clean, avail))
1251 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1252 else {
1253 if (req_cnt == (unsigned)content->array.raid_disks)
1254 fprintf(stderr, " - need all %d to start it",
1255 req_cnt);
1256 else
1257 fprintf(stderr, " - need %d to start", req_cnt);
1258 fprintf(stderr, " (use --run to insist).\n");
1259 }
1260 }
1261 return 1;
1262 }
1263
1264 int Assemble(struct supertype *st, char *mddev,
1265 struct mddev_ident *ident,
1266 struct mddev_dev *devlist,
1267 struct context *c)
1268 {
1269 /*
1270 * The task of Assemble is to find a collection of
1271 * devices that should (according to their superblocks)
1272 * form an array, and to give this collection to the MD driver.
1273 * In Linux-2.4 and later, this involves submitting a
1274 * SET_ARRAY_INFO ioctl with no arg - to prepare
1275 * the array - and then submit a number of
1276 * ADD_NEW_DISK ioctls to add disks into
1277 * the array. Finally RUN_ARRAY might
1278 * be submitted to start the array.
1279 *
1280 * Much of the work of Assemble is in finding and/or
1281 * checking the disks to make sure they look right.
1282 *
1283 * If mddev is not set, then scan must be set and we
1284 * read through the config file for dev+uuid mapping
1285 * We recurse, setting mddev, for each device that
1286 * - isn't running
1287 * - has a valid uuid (or any uuid if !uuidset)
1288 *
1289 * If mddev is set, we try to determine state of md.
1290 * check version - must be at least 0.90.0
1291 * check kernel version. must be at least 2.4.
1292 * If not, we can possibly fall back on START_ARRAY
1293 * Try to GET_ARRAY_INFO.
1294 * If possible, give up
1295 * If not, try to STOP_ARRAY just to make sure
1296 *
1297 * If !uuidset and scan, look in conf-file for uuid
1298 * If not found, give up
1299 * If !devlist and scan and uuidset, get list of devs from conf-file
1300 *
1301 * For each device:
1302 * Check superblock - discard if bad
1303 * Check uuid (set if we don't have one) - discard if no match
1304 * Check superblock similarity if we have a superblock - discard if different
1305 * Record events, devicenum
1306 * This should give us a list of devices for the array
1307 * We should collect the most recent event number
1308 *
1309 * Count disks with recent enough event count
1310 * While force && !enough disks
1311 * Choose newest rejected disks, update event count
1312 * mark clean and rewrite superblock
1313 * If recent kernel:
1314 * SET_ARRAY_INFO
1315 * foreach device with recent events : ADD_NEW_DISK
1316 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1317 * If old kernel:
1318 * Check the device numbers in superblock are right
1319 * update superblock if any changes
1320 * START_ARRAY
1321 *
1322 */
1323 int rv = -1;
1324 int mdfd = -1;
1325 int clean;
1326 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1327 ident->super_minor == UnSet && ident->name[0] == 0 &&
1328 (ident->container == NULL || ident->member == NULL));
1329 struct devs *devices = NULL;
1330 char *devmap;
1331 int *best = NULL; /* indexed by raid_disk */
1332 int bestcnt = 0;
1333 int devcnt;
1334 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1335 int journal_clean = 0;
1336 int i;
1337 int was_forced = 0;
1338 int most_recent = 0;
1339 int chosen_drive;
1340 int change = 0;
1341 int inargv = 0;
1342 int start_partial_ok = (c->runstop >= 0) &&
1343 (c->force || devlist==NULL || auto_assem);
1344 int num_devs;
1345 struct mddev_dev *tmpdev;
1346 struct mdinfo info;
1347 struct mdinfo *content = NULL;
1348 struct mdinfo *pre_exist = NULL;
1349 char *avail;
1350 char *name = NULL;
1351 char chosen_name[1024];
1352 struct map_ent *map = NULL;
1353 struct map_ent *mp;
1354
1355 /*
1356 * If any subdevs are listed, then any that don't
1357 * match ident are discarded. Remainder must all match and
1358 * become the array.
1359 * If no subdevs, then we scan all devices in the config file, but
1360 * there must be something in the identity
1361 */
1362
1363 if (!devlist &&
1364 ident->uuid_set == 0 &&
1365 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1366 ident->name[0] == 0 &&
1367 (ident->container == NULL || ident->member == NULL) &&
1368 ident->devices == NULL) {
1369 pr_err("No identity information available for %s - cannot assemble.\n",
1370 mddev ? mddev : "further assembly");
1371 return 1;
1372 }
1373
1374 if (devlist == NULL)
1375 devlist = conf_get_devs();
1376 else if (mddev)
1377 inargv = 1;
1378
1379 try_again:
1380 /* We come back here when doing auto-assembly and attempting some
1381 * set of devices failed. Those are now marked as ->used==2 and
1382 * we ignore them and try again
1383 */
1384 if (!st && ident->st)
1385 st = ident->st;
1386 if (c->verbose>0)
1387 pr_err("looking for devices for %s\n",
1388 mddev ? mddev : "further assembly");
1389
1390 content = &info;
1391 if (st && c->force)
1392 st->ignore_hw_compat = 1;
1393 num_devs = select_devices(devlist, ident, &st, &content, c,
1394 inargv, auto_assem);
1395 if (num_devs < 0)
1396 return 1;
1397
1398 if (!st || !st->sb || !content)
1399 return 2;
1400
1401 /* We have a full set of devices - we now need to find the
1402 * array device.
1403 * However there is a risk that we are racing with "mdadm -I"
1404 * and the array is already partially assembled - we will have
1405 * rejected any devices already in this address.
1406 * So we take a lock on the map file - to prevent further races -
1407 * and look for the uuid in there. If found and the array is
1408 * active, we abort. If found and the array is not active
1409 * we commit to that md device and add all the contained devices
1410 * to our list. We flag them so that we don't try to re-add,
1411 * but can remove if they turn out to not be wanted.
1412 */
1413 if (map_lock(&map))
1414 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1415 if (c->update && strcmp(c->update,"uuid") == 0)
1416 mp = NULL;
1417 else
1418 mp = map_by_uuid(&map, content->uuid);
1419 if (mp) {
1420 struct mdinfo *dv;
1421 /* array already exists. */
1422 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1423 if (pre_exist->array.level != UnSet) {
1424 pr_err("Found some drive for an array that is already active: %s\n",
1425 mp->path);
1426 pr_err("giving up.\n");
1427 goto out;
1428 }
1429 for (dv = pre_exist->devs; dv; dv = dv->next) {
1430 /* We want to add this device to our list,
1431 * but it could already be there if "mdadm -I"
1432 * started *after* we checked for O_EXCL.
1433 * If we add it to the top of the list
1434 * it will be preferred over later copies.
1435 */
1436 struct mddev_dev *newdev;
1437 char *devname = map_dev(dv->disk.major,
1438 dv->disk.minor,
1439 0);
1440 if (!devname)
1441 continue;
1442 newdev = xmalloc(sizeof(*newdev));
1443 newdev->devname = devname;
1444 newdev->disposition = 'I';
1445 newdev->used = 1;
1446 newdev->next = devlist;
1447 devlist = newdev;
1448 num_devs++;
1449 }
1450 strcpy(chosen_name, mp->path);
1451 if (c->verbose > 0 || mddev == NULL ||
1452 strcmp(mddev, chosen_name) != 0)
1453 pr_err("Merging with already-assembled %s\n",
1454 chosen_name);
1455 mdfd = open_dev_excl(mp->devnm);
1456 } else {
1457 int trustworthy = FOREIGN;
1458 name = content->name;
1459 switch (st->ss->match_home(st, c->homehost)
1460 ?: st->ss->match_home(st, "any")) {
1461 case 1:
1462 trustworthy = LOCAL;
1463 name = strchr(content->name, ':');
1464 if (name)
1465 name++;
1466 else
1467 name = content->name;
1468 break;
1469 }
1470 if (!auto_assem)
1471 /* If the array is listed in mdadm.conf or on
1472 * command line, then we trust the name
1473 * even if the array doesn't look local
1474 */
1475 trustworthy = LOCAL;
1476
1477 if (name[0] == 0 &&
1478 content->array.level == LEVEL_CONTAINER) {
1479 name = content->text_version;
1480 trustworthy = METADATA;
1481 }
1482
1483 if (name[0] && trustworthy != LOCAL &&
1484 ! c->require_homehost &&
1485 conf_name_is_free(name))
1486 trustworthy = LOCAL;
1487
1488 if (trustworthy == LOCAL &&
1489 strchr(name, ':'))
1490 /* Ignore 'host:' prefix of name */
1491 name = strchr(name, ':')+1;
1492
1493 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1494 chosen_name, 0);
1495 }
1496 if (mdfd < 0) {
1497 st->ss->free_super(st);
1498 if (auto_assem)
1499 goto try_again;
1500 goto out;
1501 }
1502 mddev = chosen_name;
1503 if (pre_exist == NULL) {
1504 if (mddev_busy(fd2devnm(mdfd))) {
1505 pr_err("%s already active, cannot restart it!\n",
1506 mddev);
1507 for (tmpdev = devlist ;
1508 tmpdev && tmpdev->used != 1;
1509 tmpdev = tmpdev->next)
1510 ;
1511 if (tmpdev && auto_assem)
1512 pr_err("%s needed for %s...\n",
1513 mddev, tmpdev->devname);
1514 close(mdfd);
1515 mdfd = -3;
1516 st->ss->free_super(st);
1517 if (auto_assem)
1518 goto try_again;
1519 goto out;
1520 }
1521 /* just incase it was started but has no content */
1522 ioctl(mdfd, STOP_ARRAY, NULL);
1523 }
1524
1525 if (content != &info) {
1526 /* This is a member of a container. Try starting the array. */
1527 int err;
1528 err = assemble_container_content(st, mdfd, content, c,
1529 chosen_name, NULL);
1530 close(mdfd);
1531 return err;
1532 }
1533
1534 /* Ok, no bad inconsistancy, we can try updating etc */
1535 devices = xcalloc(num_devs, sizeof(*devices));
1536 devmap = xcalloc(num_devs, content->array.raid_disks);
1537 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1538 c, content, mdfd, mddev,
1539 &most_recent, &bestcnt, &best, inargv);
1540 if (devcnt < 0) {
1541 mdfd = -3;
1542 goto out;
1543 }
1544
1545 if (devcnt == 0) {
1546 pr_err("no devices found for %s\n",
1547 mddev);
1548 if (st)
1549 st->ss->free_super(st);
1550 free(devmap);
1551 goto out;
1552 }
1553
1554 if (c->update && strcmp(c->update, "byteorder")==0)
1555 st->minor_version = 90;
1556
1557 st->ss->getinfo_super(st, content, NULL);
1558 clean = content->array.state & 1;
1559
1560 /* now we have some devices that might be suitable.
1561 * I wonder how many
1562 */
1563 avail = xcalloc(content->array.raid_disks, 1);
1564 okcnt = 0;
1565 replcnt = 0;
1566 sparecnt=0;
1567 journalcnt=0;
1568 rebuilding_cnt=0;
1569 for (i=0; i< bestcnt; i++) {
1570 int j = best[i];
1571 int event_margin = 1; /* always allow a difference of '1'
1572 * like the kernel does
1573 */
1574 if (j < 0) continue;
1575 /* note: we ignore error flags in multipath arrays
1576 * as they don't make sense
1577 */
1578 if (content->array.level != LEVEL_MULTIPATH) {
1579 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1580 if (content->journal_device_required)
1581 journalcnt++;
1582 else /* unexpected journal, mark as faulty */
1583 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1584 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1585 if (!(devices[j].i.disk.state
1586 & (1<<MD_DISK_FAULTY))) {
1587 devices[j].uptodate = 1;
1588 sparecnt++;
1589 }
1590 continue;
1591 }
1592 }
1593 /* If this device thinks that 'most_recent' has failed, then
1594 * we must reject this device.
1595 */
1596 if (j != most_recent && !c->force &&
1597 content->array.raid_disks > 0 &&
1598 devices[most_recent].i.disk.raid_disk >= 0 &&
1599 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1600 if (c->verbose > -1)
1601 pr_err("ignoring %s as it reports %s as failed\n",
1602 devices[j].devname, devices[most_recent].devname);
1603 best[i] = -1;
1604 continue;
1605 }
1606 /* Require event counter to be same as, or just less than,
1607 * most recent. If it is bigger, it must be a stray spare and
1608 * should be ignored.
1609 */
1610 if (devices[j].i.events+event_margin >=
1611 devices[most_recent].i.events &&
1612 devices[j].i.events <=
1613 devices[most_recent].i.events
1614 ) {
1615 devices[j].uptodate = 1;
1616 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1617 journal_clean = 1;
1618 if (i < content->array.raid_disks * 2) {
1619 if (devices[j].i.recovery_start == MaxSector ||
1620 (content->reshape_active &&
1621 i >= content->array.raid_disks - content->delta_disks)) {
1622 if (!avail[i/2]) {
1623 okcnt++;
1624 avail[i/2]=1;
1625 } else
1626 replcnt++;
1627 } else
1628 rebuilding_cnt++;
1629 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1630 sparecnt++;
1631 }
1632 }
1633 free(devmap);
1634 if (c->force) {
1635 int force_ok = force_array(content, devices, best, bestcnt,
1636 avail, most_recent, st, c);
1637 okcnt += force_ok;
1638 if (force_ok)
1639 was_forced = 1;
1640 }
1641 /* Now we want to look at the superblock which the kernel will base things on
1642 * and compare the devices that we think are working with the devices that the
1643 * superblock thinks are working.
1644 * If there are differences and --force is given, then update this chosen
1645 * superblock.
1646 */
1647 chosen_drive = -1;
1648 st->ss->free_super(st);
1649 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1650 int j = best[i];
1651 int fd;
1652
1653 if (j<0)
1654 continue;
1655 if (!devices[j].uptodate)
1656 continue;
1657 if (devices[j].i.events < devices[most_recent].i.events)
1658 continue;
1659 chosen_drive = j;
1660 if ((fd=dev_open(devices[j].devname,
1661 devices[j].included ? O_RDONLY
1662 : (O_RDONLY|O_EXCL)))< 0) {
1663 pr_err("Cannot open %s: %s\n",
1664 devices[j].devname, strerror(errno));
1665 goto out;
1666 }
1667 if (st->ss->load_super(st,fd, NULL)) {
1668 close(fd);
1669 pr_err("RAID superblock has disappeared from %s\n",
1670 devices[j].devname);
1671 goto out;
1672 }
1673 close(fd);
1674 }
1675 if (st->sb == NULL) {
1676 pr_err("No suitable drives found for %s\n", mddev);
1677 goto out;
1678 }
1679 st->ss->getinfo_super(st, content, NULL);
1680 if (sysfs_init(content, mdfd, NULL)) {
1681 pr_err("Unable to initialize sysfs\n");
1682 goto out;
1683 }
1684
1685 /* after reload context, store journal_clean in context */
1686 content->journal_clean = journal_clean;
1687 for (i=0; i<bestcnt; i++) {
1688 int j = best[i];
1689 unsigned int desired_state;
1690
1691 if (j < 0)
1692 continue;
1693 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1694 desired_state = (1<<MD_DISK_JOURNAL);
1695 else if (i >= content->array.raid_disks * 2)
1696 desired_state = 0;
1697 else if (i & 1)
1698 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1699 else
1700 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1701
1702 if (!devices[j].uptodate)
1703 continue;
1704
1705 devices[j].i.disk.state = desired_state;
1706 if (!(devices[j].i.array.state & 1))
1707 clean = 0;
1708
1709 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1710 c->verbose, 0, NULL)) {
1711 if (c->force) {
1712 if (c->verbose >= 0)
1713 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1714 j, mddev, devices[j].devname);
1715 change = 1;
1716 } else {
1717 if (c->verbose >= -1)
1718 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1719 i, mddev, devices[j].devname);
1720 }
1721 }
1722 #if 0
1723 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1724 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1725 i, mddev);
1726 }
1727 #endif
1728 }
1729 if (c->force && !clean &&
1730 !enough(content->array.level, content->array.raid_disks,
1731 content->array.layout, clean,
1732 avail)) {
1733 change += st->ss->update_super(st, content, "force-array",
1734 devices[chosen_drive].devname, c->verbose,
1735 0, NULL);
1736 was_forced = 1;
1737 clean = 1;
1738 }
1739
1740 if (change) {
1741 int fd;
1742 fd = dev_open(devices[chosen_drive].devname,
1743 devices[chosen_drive].included ?
1744 O_RDWR : (O_RDWR|O_EXCL));
1745 if (fd < 0) {
1746 pr_err("Could not open %s for write - cannot Assemble array.\n",
1747 devices[chosen_drive].devname);
1748 goto out;
1749 }
1750 if (st->ss->store_super(st, fd)) {
1751 close(fd);
1752 pr_err("Could not re-write superblock on %s\n",
1753 devices[chosen_drive].devname);
1754 goto out;
1755 }
1756 if (c->verbose >= 0)
1757 pr_err("Marking array %s as 'clean'\n",
1758 mddev);
1759 close(fd);
1760 }
1761
1762 /* If we are in the middle of a reshape we may need to restore saved data
1763 * that was moved aside due to the reshape overwriting live data
1764 * The code of doing this lives in Grow.c
1765 */
1766 if (content->reshape_active &&
1767 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1768 int err = 0;
1769 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1770 if (c->verbose > 0)
1771 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1772 chosen_name);
1773 if (!c->backup_file)
1774 c->backup_file = locate_backup(content->sys_name);
1775 enable_fds(bestcnt/2);
1776 for (i = 0; i < bestcnt/2; i++) {
1777 int j = best[i*2];
1778 if (j >= 0) {
1779 fdlist[i] = dev_open(devices[j].devname,
1780 devices[j].included
1781 ? O_RDWR : (O_RDWR|O_EXCL));
1782 if (fdlist[i] < 0) {
1783 pr_err("Could not open %s for write - cannot Assemble array.\n",
1784 devices[j].devname);
1785 err = 1;
1786 break;
1787 }
1788 } else
1789 fdlist[i] = -1;
1790 }
1791 if (!err) {
1792 if (st->ss->external && st->ss->recover_backup)
1793 err = st->ss->recover_backup(st, content);
1794 else
1795 err = Grow_restart(st, content, fdlist, bestcnt/2,
1796 c->backup_file, c->verbose > 0);
1797 if (err && c->invalid_backup) {
1798 if (c->verbose > 0)
1799 pr_err("continuing without restoring backup\n");
1800 err = 0;
1801 }
1802 }
1803 while (i>0) {
1804 i--;
1805 if (fdlist[i]>=0) close(fdlist[i]);
1806 }
1807 free(fdlist);
1808 if (err) {
1809 pr_err("Failed to restore critical section for reshape, sorry.\n");
1810 if (c->backup_file == NULL)
1811 cont_err("Possibly you needed to specify the --backup-file\n");
1812 goto out;
1813 }
1814 }
1815
1816 /* Almost ready to actually *do* something */
1817 /* First, fill in the map, so that udev can find our name
1818 * as soon as we become active.
1819 */
1820 if (c->update && strcmp(c->update, "metadata")==0) {
1821 content->array.major_version = 1;
1822 content->array.minor_version = 0;
1823 strcpy(content->text_version, "1.0");
1824 }
1825
1826 map_update(&map, fd2devnm(mdfd), content->text_version,
1827 content->uuid, chosen_name);
1828
1829 rv = start_array(mdfd, mddev, content,
1830 st, ident, best, bestcnt,
1831 chosen_drive, devices, okcnt, sparecnt,
1832 rebuilding_cnt, journalcnt,
1833 c,
1834 clean, avail, start_partial_ok,
1835 pre_exist != NULL,
1836 was_forced);
1837 if (rv == 1 && !pre_exist)
1838 ioctl(mdfd, STOP_ARRAY, NULL);
1839 free(devices);
1840 out:
1841 map_unlock(&map);
1842 if (rv == 0) {
1843 wait_for(chosen_name, mdfd);
1844 close(mdfd);
1845 if (auto_assem) {
1846 int usecs = 1;
1847 /* There is a nasty race with 'mdadm --monitor'.
1848 * If it opens this device before we close it,
1849 * it gets an incomplete open on which IO
1850 * doesn't work and the capacity is
1851 * wrong.
1852 * If we reopen (to check for layered devices)
1853 * before --monitor closes, we loose.
1854 *
1855 * So: wait upto 1 second for there to be
1856 * a non-zero capacity.
1857 */
1858 while (usecs < 1000) {
1859 mdfd = open(mddev, O_RDONLY);
1860 if (mdfd >= 0) {
1861 unsigned long long size;
1862 if (get_dev_size(mdfd, NULL, &size) &&
1863 size > 0)
1864 break;
1865 close(mdfd);
1866 }
1867 usleep(usecs);
1868 usecs <<= 1;
1869 }
1870 }
1871 } else if (mdfd >= 0)
1872 close(mdfd);
1873
1874 /* '2' means 'OK, but not started yet' */
1875 if (rv == -1) {
1876 free(devices);
1877 return 1;
1878 }
1879 return rv == 2 ? 0 : rv;
1880 }
1881
1882 int assemble_container_content(struct supertype *st, int mdfd,
1883 struct mdinfo *content, struct context *c,
1884 char *chosen_name, int *result)
1885 {
1886 struct mdinfo *dev, *sra, *dev2;
1887 int working = 0, preexist = 0;
1888 int expansion = 0;
1889 int old_raid_disks;
1890 int start_reshape;
1891 char *avail;
1892 int err;
1893
1894 if (sysfs_init(content, mdfd, NULL)) {
1895 pr_err("Unable to initialize sysfs\n");
1896 return 1;
1897 }
1898
1899 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1900 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1901 if (content->array.major_version == -1 &&
1902 content->array.minor_version == -2 &&
1903 c->readonly &&
1904 content->text_version[0] == '/')
1905 content->text_version[0] = '-';
1906 if (sysfs_set_array(content, 9003) != 0) {
1907 sysfs_free(sra);
1908 return 1;
1909 }
1910 }
1911
1912 /* There are two types of reshape: container wide or sub-array specific
1913 * Check if metadata requests blocking container wide reshapes
1914 */
1915 start_reshape = (content->reshape_active &&
1916 !((content->reshape_active == CONTAINER_RESHAPE) &&
1917 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1918
1919 /* Block subarray here if it is under reshape now
1920 * Do not allow for any changes in this array
1921 */
1922 if (st->ss->external && content->recovery_blocked && start_reshape)
1923 block_subarray(content);
1924
1925 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1926 for (dev = content->devs; dev; dev = dev->next)
1927 if (dev2->disk.major == dev->disk.major &&
1928 dev2->disk.minor == dev->disk.minor)
1929 break;
1930 if (dev)
1931 continue;
1932 /* Don't want this one any more */
1933 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1934 errno == EBUSY) {
1935 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1936 sysfs_free(sra);
1937 return 1;
1938 }
1939 sysfs_set_str(sra, dev2, "state", "remove");
1940 }
1941 old_raid_disks = content->array.raid_disks - content->delta_disks;
1942 avail = xcalloc(content->array.raid_disks, 1);
1943 for (dev = content->devs; dev; dev = dev->next) {
1944 if (dev->disk.raid_disk >= 0)
1945 avail[dev->disk.raid_disk] = 1;
1946 if (sysfs_add_disk(content, dev, 1) == 0) {
1947 if (dev->disk.raid_disk >= old_raid_disks &&
1948 content->reshape_active)
1949 expansion++;
1950 else
1951 working++;
1952 } else if (errno == EEXIST)
1953 preexist++;
1954 }
1955 sysfs_free(sra);
1956 if (working + expansion == 0 && c->runstop <= 0) {
1957 free(avail);
1958 return 1;/* Nothing new, don't try to start */
1959 }
1960 map_update(NULL, fd2devnm(mdfd), content->text_version,
1961 content->uuid, chosen_name);
1962
1963 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1964 st->ss->validate_ppl) {
1965 content->array.state |= 1;
1966 err = 0;
1967
1968 for (dev = content->devs; dev; dev = dev->next) {
1969 int dfd;
1970 char *devpath;
1971 int ret;
1972
1973 ret = st->ss->validate_ppl(st, content, dev);
1974 if (ret == 0)
1975 continue;
1976
1977 if (ret < 0) {
1978 err = 1;
1979 break;
1980 }
1981
1982 if (!c->force) {
1983 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1984 chosen_name);
1985 content->array.state &= ~1;
1986 avail[dev->disk.raid_disk] = 0;
1987 break;
1988 }
1989
1990 /* have --force - overwrite the invalid ppl */
1991 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
1992 dfd = dev_open(devpath, O_RDWR);
1993 if (dfd < 0) {
1994 pr_err("Failed to open %s\n", devpath);
1995 err = 1;
1996 break;
1997 }
1998
1999 err = st->ss->write_init_ppl(st, content, dfd);
2000 close(dfd);
2001
2002 if (err)
2003 break;
2004 }
2005
2006 if (err) {
2007 free(avail);
2008 return err;
2009 }
2010 }
2011
2012 if (enough(content->array.level, content->array.raid_disks,
2013 content->array.layout, content->array.state & 1, avail) == 0) {
2014 if (c->export && result)
2015 *result |= INCR_NO;
2016 else if (c->verbose >= 0) {
2017 pr_err("%s assembled with %d device%s",
2018 chosen_name, preexist + working,
2019 preexist + working == 1 ? "":"s");
2020 if (preexist)
2021 fprintf(stderr, " (%d new)", working);
2022 fprintf(stderr, " but not started\n");
2023 }
2024 free(avail);
2025 return 1;
2026 }
2027 free(avail);
2028
2029 if (c->runstop <= 0 &&
2030 (working + preexist + expansion) <
2031 content->array.working_disks) {
2032 if (c->export && result)
2033 *result |= INCR_UNSAFE;
2034 else if (c->verbose >= 0) {
2035 pr_err("%s assembled with %d device%s",
2036 chosen_name, preexist + working,
2037 preexist + working == 1 ? "":"s");
2038 if (preexist)
2039 fprintf(stderr, " (%d new)", working);
2040 fprintf(stderr, " but not safe to start\n");
2041 }
2042 return 1;
2043 }
2044
2045
2046 if (start_reshape) {
2047 int spare = content->array.raid_disks + expansion;
2048 if (restore_backup(st, content,
2049 working,
2050 spare, &c->backup_file, c->verbose) == 1)
2051 return 1;
2052
2053 err = sysfs_set_str(content, NULL,
2054 "array_state", "readonly");
2055 if (err)
2056 return 1;
2057
2058 if (st->ss->external) {
2059 if (!mdmon_running(st->container_devnm))
2060 start_mdmon(st->container_devnm);
2061 ping_monitor(st->container_devnm);
2062 if (mdmon_running(st->container_devnm) &&
2063 st->update_tail == NULL)
2064 st->update_tail = &st->updates;
2065 }
2066
2067 err = Grow_continue(mdfd, st, content, c->backup_file,
2068 0, c->freeze_reshape);
2069 } else switch(content->array.level) {
2070 case LEVEL_LINEAR:
2071 case LEVEL_MULTIPATH:
2072 case 0:
2073 err = sysfs_set_str(content, NULL, "array_state",
2074 c->readonly ? "readonly" : "active");
2075 break;
2076 default:
2077 err = sysfs_set_str(content, NULL, "array_state",
2078 "readonly");
2079 /* start mdmon if needed. */
2080 if (!err) {
2081 if (!mdmon_running(st->container_devnm))
2082 start_mdmon(st->container_devnm);
2083 ping_monitor(st->container_devnm);
2084 }
2085 break;
2086 }
2087 if (!err)
2088 sysfs_set_safemode(content, content->safe_mode_delay);
2089
2090 /* Block subarray here if it is not reshaped now
2091 * It has be blocked a little later to allow mdmon to switch in
2092 * in to R/W state
2093 */
2094 if (st->ss->external && content->recovery_blocked &&
2095 !start_reshape)
2096 block_subarray(content);
2097
2098 if (c->export && result) {
2099 if (err)
2100 *result |= INCR_NO;
2101 else
2102 *result |= INCR_YES;
2103 } else if (c->verbose >= 0) {
2104 if (err)
2105 pr_err("array %s now has %d device%s",
2106 chosen_name, working + preexist,
2107 working + preexist == 1 ? "":"s");
2108 else
2109 pr_err("Started %s with %d device%s",
2110 chosen_name, working + preexist,
2111 working + preexist == 1 ? "":"s");
2112 if (preexist)
2113 fprintf(stderr, " (%d new)", working);
2114 if (expansion)
2115 fprintf(stderr, " ( + %d for expansion)",
2116 expansion);
2117 fprintf(stderr, "\n");
2118 }
2119 if (!err)
2120 wait_for(chosen_name, mdfd);
2121 return err;
2122 /* FIXME should have an O_EXCL and wait for read-auto */
2123 }