]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Grow: Frozen array can't be idle
[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 return -1;
598 }
599
600 dfd = dev_open(devname,
601 tmpdev->disposition == 'I'
602 ? O_RDWR : (O_RDWR|O_EXCL));
603
604 tst = dup_super(st);
605 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
606 pr_err("cannot re-read metadata from %s - aborting\n",
607 devname);
608 if (dfd >= 0)
609 close(dfd);
610 close(mdfd);
611 free(devices);
612 free(devmap);
613 tst->ss->free_super(tst);
614 free(tst);
615 *stp = st;
616 return -1;
617 }
618 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
619
620 memcpy(content->uuid, ident->uuid, 16);
621 strcpy(content->name, ident->name);
622 content->array.md_minor = minor(stb2.st_rdev);
623
624 if (strcmp(c->update, "byteorder") == 0)
625 err = 0;
626 else if (strcmp(c->update, "home-cluster") == 0) {
627 tst->cluster_name = c->homecluster;
628 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
629 } else if (strcmp(c->update, "nodes") == 0) {
630 tst->nodes = c->nodes;
631 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
632 } else if (strcmp(c->update, "revert-reshape") == 0 &&
633 c->invalid_backup)
634 err = tst->ss->update_super(tst, content,
635 "revert-reshape-nobackup",
636 devname, c->verbose,
637 ident->uuid_set,
638 c->homehost);
639 else
640 err = tst->ss->update_super(tst, content, c->update,
641 devname, c->verbose,
642 ident->uuid_set,
643 c->homehost);
644 if (err < 0) {
645 if (err == -1)
646 pr_err("--update=%s not understood for %s metadata\n",
647 c->update, tst->ss->name);
648 tst->ss->free_super(tst);
649 free(tst);
650 close(mdfd);
651 close(dfd);
652 free(devices);
653 free(devmap);
654 *stp = st;
655 return -1;
656 }
657 if (strcmp(c->update, "uuid")==0 &&
658 !ident->uuid_set) {
659 ident->uuid_set = 1;
660 memcpy(ident->uuid, content->uuid, 16);
661 }
662 if (tst->ss->store_super(tst, dfd))
663 pr_err("Could not re-write superblock on %s.\n",
664 devname);
665
666 if (strcmp(c->update, "uuid")==0 &&
667 ident->bitmap_fd >= 0 && !bitmap_done) {
668 if (bitmap_update_uuid(ident->bitmap_fd,
669 content->uuid,
670 tst->ss->swapuuid) != 0)
671 pr_err("Could not update uuid on external bitmap.\n");
672 else
673 bitmap_done = 1;
674 }
675 } else {
676 dfd = dev_open(devname,
677 tmpdev->disposition == 'I'
678 ? O_RDWR : (O_RDWR|O_EXCL));
679 tst = dup_super(st);
680
681 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
682 pr_err("cannot re-read metadata from %s - aborting\n",
683 devname);
684 if (dfd >= 0)
685 close(dfd);
686 close(mdfd);
687 free(devices);
688 free(devmap);
689 tst->ss->free_super(tst);
690 free(tst);
691 *stp = st;
692 return -1;
693 }
694 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
695 }
696
697 fstat(dfd, &stb);
698 close(dfd);
699
700 if (c->verbose > 0)
701 pr_err("%s is identified as a member of %s, slot %d%s.\n",
702 devname, mddev, content->disk.raid_disk,
703 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
704 devices[devcnt].devname = devname;
705 devices[devcnt].uptodate = 0;
706 devices[devcnt].included = (tmpdev->disposition == 'I');
707 devices[devcnt].i = *content;
708 devices[devcnt].i.disk.major = major(stb.st_rdev);
709 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
710
711 if (devices[devcnt].i.disk.state == 6) {
712 if (most_recent < 0 ||
713 devices[devcnt].i.events
714 > devices[most_recent].i.events) {
715 struct supertype *tmp = tst;
716 tst = st;
717 st = tmp;
718 most_recent = devcnt;
719 }
720 }
721 tst->ss->free_super(tst);
722 free(tst);
723
724 if (content->array.level == LEVEL_MULTIPATH)
725 /* with multipath, the raid_disk from the superblock is meaningless */
726 i = devcnt;
727 else
728 i = devices[devcnt].i.disk.raid_disk;
729 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
730 if (nextspare < content->array.raid_disks*2)
731 nextspare = content->array.raid_disks*2;
732 i = nextspare++;
733 } else {
734 /* i is raid_disk - double it so there is room for
735 * replacements */
736 i *= 2;
737 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
738 i++;
739 if (i >= content->array.raid_disks*2 &&
740 i >= nextspare)
741 nextspare = i+1;
742 }
743 if (i < 10000) {
744 if (i >= bestcnt) {
745 int newbestcnt = i+10;
746 int *newbest = xmalloc(sizeof(int)*newbestcnt);
747 int c;
748 for (c=0; c < newbestcnt; c++)
749 if (c < bestcnt)
750 newbest[c] = best[c];
751 else
752 newbest[c] = -1;
753 if (best)free(best);
754 best = newbest;
755 bestcnt = newbestcnt;
756 }
757 if (best[i] >=0 &&
758 devices[best[i]].i.events ==
759 devices[devcnt].i.events &&
760 (devices[best[i]].i.disk.minor !=
761 devices[devcnt].i.disk.minor) &&
762 st->ss == &super0 &&
763 content->array.level != LEVEL_MULTIPATH) {
764 /* two different devices with identical superblock.
765 * Could be a mis-detection caused by overlapping
766 * partitions. fail-safe.
767 */
768 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
769 " If they are really different, please --zero the superblock on one\n"
770 " If they are the same or overlap, please remove one from %s.\n",
771 devices[best[i]].devname, devname,
772 inargv ? "the list" :
773 "the\n DEVICE list in mdadm.conf"
774 );
775 close(mdfd);
776 free(devices);
777 free(devmap);
778 *stp = st;
779 return -1;
780 }
781 if (best[i] == -1 || (devices[best[i]].i.events
782 < devices[devcnt].i.events))
783 best[i] = devcnt;
784 else if (st->ss == &super_imsm)
785 best[i+1] = devcnt;
786 }
787 devcnt++;
788 }
789 if (most_recent >= 0)
790 *most_recentp = most_recent;
791 *bestcntp = bestcnt;
792 *bestp = best;
793 *stp = st;
794 return devcnt;
795 }
796
797 static int force_array(struct mdinfo *content,
798 struct devs *devices,
799 int *best, int bestcnt, char *avail,
800 int most_recent,
801 struct supertype *st,
802 struct context *c)
803 {
804 int okcnt = 0;
805 while (!enough(content->array.level, content->array.raid_disks,
806 content->array.layout, 1,
807 avail) ||
808 (content->reshape_active && content->delta_disks > 0 &&
809 !enough(content->array.level, (content->array.raid_disks
810 - content->delta_disks),
811 content->new_layout, 1, avail))) {
812 /* Choose the newest best drive which is
813 * not up-to-date, update the superblock
814 * and add it.
815 */
816 int fd;
817 struct supertype *tst;
818 unsigned long long current_events;
819 int chosen_drive = -1;
820 int i;
821
822 for (i = 0;
823 i < content->array.raid_disks * 2 && i < bestcnt;
824 i += 2) {
825 int j = best[i];
826 if (j < 0)
827 continue;
828 if (devices[j].uptodate)
829 continue;
830 if (devices[j].i.recovery_start != MaxSector) {
831 int delta;
832 if (!devices[j].i.reshape_active ||
833 devices[j].i.delta_disks <= 0)
834 continue;
835 /* When increasing number of devices, an
836 * added device also appears to be
837 * recovering. It is safe to include it
838 * as long as it won't be a source of
839 * data.
840 * For now, just allow for last data
841 * devices in RAID4 or last devices in RAID4/5/6.
842 */
843 delta = devices[j].i.delta_disks;
844 if (devices[j].i.array.level >= 4 &&
845 devices[j].i.array.level <= 6 &&
846 i/2 >= content->array.raid_disks - delta)
847 /* OK */;
848 else if (devices[j].i.array.level == 4 &&
849 i/2 >= content->array.raid_disks - delta - 1)
850 /* OK */;
851 else
852 continue;
853 } else if (devices[j].i.reshape_active !=
854 content->reshape_active ||
855 (devices[j].i.reshape_active &&
856 devices[j].i.reshape_progress !=
857 content->reshape_progress))
858 /* Here, it may be a source of data. If two
859 * devices claim different progresses, it
860 * means that reshape boundaries differ for
861 * their own devices. Kernel will only treat
862 * the first one as reshape progress and
863 * go on. It may cause disaster, so avoid it.
864 */
865 continue;
866 if (chosen_drive < 0 ||
867 devices[j].i.events
868 > devices[chosen_drive].i.events)
869 chosen_drive = j;
870 }
871 if (chosen_drive < 0)
872 break;
873 current_events = devices[chosen_drive].i.events;
874 add_another:
875 if (c->verbose >= 0)
876 pr_err("forcing event count in %s(%d) from %d upto %d\n",
877 devices[chosen_drive].devname,
878 devices[chosen_drive].i.disk.raid_disk,
879 (int)(devices[chosen_drive].i.events),
880 (int)(devices[most_recent].i.events));
881 fd = dev_open(devices[chosen_drive].devname,
882 devices[chosen_drive].included ? O_RDWR
883 : (O_RDWR|O_EXCL));
884 if (fd < 0) {
885 pr_err("Couldn't open %s for write - not updating\n",
886 devices[chosen_drive].devname);
887 devices[chosen_drive].i.events = 0;
888 continue;
889 }
890 tst = dup_super(st);
891 if (tst->ss->load_super(tst,fd, NULL)) {
892 close(fd);
893 pr_err("RAID superblock disappeared from %s - not updating.\n",
894 devices[chosen_drive].devname);
895 devices[chosen_drive].i.events = 0;
896 continue;
897 }
898 content->events = devices[most_recent].i.events;
899 tst->ss->update_super(tst, content, "force-one",
900 devices[chosen_drive].devname, c->verbose,
901 0, NULL);
902
903 if (tst->ss->store_super(tst, fd)) {
904 close(fd);
905 pr_err("Could not re-write superblock on %s\n",
906 devices[chosen_drive].devname);
907 devices[chosen_drive].i.events = 0;
908 tst->ss->free_super(tst);
909 continue;
910 }
911 close(fd);
912 devices[chosen_drive].i.events = devices[most_recent].i.events;
913 devices[chosen_drive].uptodate = 1;
914 avail[chosen_drive] = 1;
915 okcnt++;
916 tst->ss->free_super(tst);
917 /* If there are any other drives of the same vintage,
918 * add them in as well. We can't lose and we might gain
919 */
920 for (i = 0;
921 i < content->array.raid_disks * 2 && i < bestcnt ;
922 i += 2) {
923 int j = best[i];
924 if (j >= 0 &&
925 !devices[j].uptodate &&
926 devices[j].i.recovery_start == MaxSector &&
927 devices[j].i.events == current_events &&
928 ((!devices[j].i.reshape_active &&
929 !content->reshape_active) ||
930 (devices[j].i.reshape_active ==
931 content->reshape_active &&
932 devices[j].i.reshape_progress ==
933 content->reshape_progress))) {
934 chosen_drive = j;
935 goto add_another;
936 }
937 }
938 }
939 return okcnt;
940 }
941
942 static int start_array(int mdfd,
943 char *mddev,
944 struct mdinfo *content,
945 struct supertype *st,
946 struct mddev_ident *ident,
947 int *best, int bestcnt,
948 int chosen_drive,
949 struct devs *devices,
950 unsigned int okcnt,
951 unsigned int sparecnt,
952 unsigned int rebuilding_cnt,
953 unsigned int journalcnt,
954 struct context *c,
955 int clean, char *avail,
956 int start_partial_ok,
957 int err_ok,
958 int was_forced
959 )
960 {
961 int rv;
962 int i;
963 unsigned int req_cnt;
964
965 if (content->journal_device_required && (content->journal_clean == 0)) {
966 if (!c->force) {
967 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
968 return 1;
969 }
970 pr_err("Journal is missing or stale, starting array read only.\n");
971 c->readonly = 1;
972 }
973
974 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
975 clean = 1;
976
977 rv = set_array_info(mdfd, st, content);
978 if (rv && !err_ok) {
979 pr_err("failed to set array info for %s: %s\n",
980 mddev, strerror(errno));
981 return 1;
982 }
983 if (ident->bitmap_fd >= 0) {
984 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
985 pr_err("SET_BITMAP_FILE failed.\n");
986 return 1;
987 }
988 } else if (ident->bitmap_file) {
989 /* From config file */
990 int bmfd = open(ident->bitmap_file, O_RDWR);
991 if (bmfd < 0) {
992 pr_err("Could not open bitmap file %s\n",
993 ident->bitmap_file);
994 return 1;
995 }
996 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
997 pr_err("Failed to set bitmapfile for %s\n", mddev);
998 close(bmfd);
999 return 1;
1000 }
1001 close(bmfd);
1002 }
1003
1004 /* First, add the raid disks, but add the chosen one last */
1005 for (i = 0; i <= bestcnt; i++) {
1006 int j;
1007 if (i < bestcnt) {
1008 j = best[i];
1009 if (j == chosen_drive)
1010 continue;
1011 } else
1012 j = chosen_drive;
1013
1014 if (j >= 0 && !devices[j].included) {
1015 int dfd;
1016
1017 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1018 if (dfd >= 0) {
1019 remove_partitions(dfd);
1020 close(dfd);
1021 }
1022 rv = add_disk(mdfd, st, content, &devices[j].i);
1023
1024 if (rv) {
1025 pr_err("failed to add %s to %s: %s\n",
1026 devices[j].devname, mddev,
1027 strerror(errno));
1028 if (i < content->array.raid_disks * 2 ||
1029 i == bestcnt)
1030 okcnt--;
1031 else
1032 sparecnt--;
1033 } else if (c->verbose > 0) {
1034 pr_err("added %s to %s as %d%s%s\n",
1035 devices[j].devname, mddev,
1036 devices[j].i.disk.raid_disk,
1037 devices[j].uptodate?"":
1038 " (possibly out of date)",
1039 (devices[j].i.disk.state &
1040 (1<<MD_DISK_REPLACEMENT)) ?
1041 " replacement":"");
1042 }
1043 } else if (j >= 0) {
1044 if (c->verbose > 0)
1045 pr_err("%s is already in %s as %d\n",
1046 devices[j].devname, mddev,
1047 devices[j].i.disk.raid_disk);
1048 } else if (c->verbose > 0 &&
1049 i < content->array.raid_disks * 2 && (i & 1) == 0)
1050 pr_err("no uptodate device for slot %d of %s\n",
1051 i/2, mddev);
1052 }
1053
1054 if (content->array.level == LEVEL_CONTAINER) {
1055 if (c->verbose >= 0) {
1056 pr_err("Container %s has been assembled with %d drive%s",
1057 mddev, okcnt + sparecnt + journalcnt,
1058 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1059 if (okcnt < (unsigned)content->array.raid_disks)
1060 fprintf(stderr, " (out of %d)",
1061 content->array.raid_disks);
1062 fprintf(stderr, "\n");
1063 }
1064
1065 if (st->ss->validate_container) {
1066 struct mdinfo *devices_list;
1067 struct mdinfo *info_devices;
1068 unsigned int count;
1069
1070 devices_list = NULL;
1071 info_devices = xmalloc(sizeof(struct mdinfo) *
1072 (okcnt + sparecnt));
1073 for (count = 0; count < okcnt + sparecnt; count++) {
1074 info_devices[count] = devices[count].i;
1075 info_devices[count].next = devices_list;
1076 devices_list = &info_devices[count];
1077 }
1078 if (st->ss->validate_container(devices_list))
1079 pr_err("Mismatch detected!\n");
1080 free(info_devices);
1081 }
1082
1083 st->ss->free_super(st);
1084 sysfs_uevent(content, "change");
1085 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1086 /* Was partial, is still partial, so signal an error
1087 * to ensure we don't retry */
1088 return 1;
1089 return 0;
1090 }
1091
1092 /* Get number of in-sync devices according to the superblock.
1093 * We must have this number to start the array without -s or -R
1094 */
1095 req_cnt = content->array.working_disks;
1096
1097 if (c->runstop == 1 ||
1098 (c->runstop <= 0 &&
1099 (enough(content->array.level, content->array.raid_disks,
1100 content->array.layout, clean, avail) &&
1101 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1102 /* This array is good-to-go.
1103 * If a reshape is in progress then we might need to
1104 * continue monitoring it. In that case we start
1105 * it read-only and let the grow code make it writable.
1106 */
1107 int rv;
1108
1109 if (content->reshape_active &&
1110 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1111 content->delta_disks <= 0) {
1112 if (!c->backup_file) {
1113 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1114 mddev);
1115 pr_err("Please provided one with \"--backup-file=...\"\n");
1116 if (c->update &&
1117 strcmp(c->update, "revert-reshape") == 0)
1118 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1119 return 1;
1120 }
1121 rv = sysfs_set_str(content, NULL,
1122 "array_state", "readonly");
1123 if (rv == 0)
1124 rv = Grow_continue(mdfd, st, content,
1125 c->backup_file, 0,
1126 c->freeze_reshape);
1127 } else if (c->readonly &&
1128 sysfs_attribute_available(content, NULL,
1129 "array_state")) {
1130 rv = sysfs_set_str(content, NULL,
1131 "array_state", "readonly");
1132 } else
1133 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1134 reopen_mddev(mdfd); /* drop O_EXCL */
1135 if (rv == 0) {
1136 if (c->verbose >= 0) {
1137 pr_err("%s has been started with %d drive%s",
1138 mddev, okcnt, okcnt==1?"":"s");
1139 if (okcnt < (unsigned)content->array.raid_disks)
1140 fprintf(stderr, " (out of %d)",
1141 content->array.raid_disks);
1142 if (rebuilding_cnt)
1143 fprintf(stderr, "%s %d rebuilding",
1144 sparecnt?",":" and",
1145 rebuilding_cnt);
1146 if (sparecnt)
1147 fprintf(stderr, " and %d spare%s",
1148 sparecnt,
1149 sparecnt == 1 ? "" : "s");
1150 if (content->journal_clean)
1151 fprintf(stderr, " and %d journal",
1152 journalcnt);
1153 fprintf(stderr, ".\n");
1154 }
1155 if (content->reshape_active &&
1156 content->array.level >= 4 &&
1157 content->array.level <= 6) {
1158 /* might need to increase the size
1159 * of the stripe cache - default is 256
1160 */
1161 int chunk_size = content->array.chunk_size;
1162
1163 if (content->reshape_active &&
1164 content->new_chunk > chunk_size)
1165 chunk_size = content->new_chunk;
1166 if (256 < 4 * ((chunk_size+4065)/4096)) {
1167 struct mdinfo *sra;
1168
1169 sra = sysfs_read(mdfd, NULL, 0);
1170 if (sra)
1171 sysfs_set_num(sra, NULL,
1172 "stripe_cache_size",
1173 (4 * chunk_size / 4096) + 1);
1174 sysfs_free(sra);
1175 }
1176 }
1177 if (okcnt < (unsigned)content->array.raid_disks) {
1178 /* If any devices did not get added
1179 * because the kernel rejected them based
1180 * on event count, try adding them
1181 * again providing the action policy is
1182 * 're-add' or greater. The bitmap
1183 * might allow them to be included, or
1184 * they will become spares.
1185 */
1186 for (i = 0; i < bestcnt; i++) {
1187 int j = best[i];
1188 if (j >= 0 && !devices[j].uptodate) {
1189 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1190 continue;
1191 rv = add_disk(mdfd, st, content,
1192 &devices[j].i);
1193 if (rv == 0 && c->verbose >= 0)
1194 pr_err("%s has been re-added.\n",
1195 devices[j].devname);
1196 }
1197 }
1198 }
1199 if (content->array.level == 6 &&
1200 okcnt + 1 == (unsigned)content->array.raid_disks &&
1201 was_forced) {
1202 struct mdinfo *sra;
1203
1204 sra = sysfs_read(mdfd, NULL, 0);
1205 if (sra)
1206 sysfs_set_str(sra, NULL,
1207 "sync_action", "repair");
1208 sysfs_free(sra);
1209 }
1210 return 0;
1211 }
1212 pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno));
1213
1214 if (!enough(content->array.level, content->array.raid_disks,
1215 content->array.layout, 1, avail))
1216 pr_err("Not enough devices to start the array.\n");
1217 else if (!enough(content->array.level,
1218 content->array.raid_disks,
1219 content->array.layout, clean, avail))
1220 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1221
1222 return 1;
1223 }
1224 if (c->runstop == -1) {
1225 pr_err("%s assembled from %d drive%s",
1226 mddev, okcnt, okcnt == 1 ? "" : "s");
1227 if (okcnt != (unsigned)content->array.raid_disks)
1228 fprintf(stderr, " (out of %d)",
1229 content->array.raid_disks);
1230 fprintf(stderr, ", but not started.\n");
1231 return 2;
1232 }
1233 if (c->verbose >= -1) {
1234 pr_err("%s assembled from %d drive%s",
1235 mddev, okcnt, okcnt == 1 ? "" : "s");
1236 if (rebuilding_cnt)
1237 fprintf(stderr, "%s %d rebuilding",
1238 sparecnt ? "," : " and", rebuilding_cnt);
1239 if (sparecnt)
1240 fprintf(stderr, " and %d spare%s", sparecnt,
1241 sparecnt == 1 ? "" : "s");
1242 if (!enough(content->array.level, content->array.raid_disks,
1243 content->array.layout, 1, avail))
1244 fprintf(stderr, " - not enough to start the array.\n");
1245 else if (!enough(content->array.level,
1246 content->array.raid_disks,
1247 content->array.layout, clean, avail))
1248 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1249 else {
1250 if (req_cnt == (unsigned)content->array.raid_disks)
1251 fprintf(stderr, " - need all %d to start it",
1252 req_cnt);
1253 else
1254 fprintf(stderr, " - need %d to start", req_cnt);
1255 fprintf(stderr, " (use --run to insist).\n");
1256 }
1257 }
1258 return 1;
1259 }
1260
1261 int Assemble(struct supertype *st, char *mddev,
1262 struct mddev_ident *ident,
1263 struct mddev_dev *devlist,
1264 struct context *c)
1265 {
1266 /*
1267 * The task of Assemble is to find a collection of
1268 * devices that should (according to their superblocks)
1269 * form an array, and to give this collection to the MD driver.
1270 * In Linux-2.4 and later, this involves submitting a
1271 * SET_ARRAY_INFO ioctl with no arg - to prepare
1272 * the array - and then submit a number of
1273 * ADD_NEW_DISK ioctls to add disks into
1274 * the array. Finally RUN_ARRAY might
1275 * be submitted to start the array.
1276 *
1277 * Much of the work of Assemble is in finding and/or
1278 * checking the disks to make sure they look right.
1279 *
1280 * If mddev is not set, then scan must be set and we
1281 * read through the config file for dev+uuid mapping
1282 * We recurse, setting mddev, for each device that
1283 * - isn't running
1284 * - has a valid uuid (or any uuid if !uuidset)
1285 *
1286 * If mddev is set, we try to determine state of md.
1287 * check version - must be at least 0.90.0
1288 * check kernel version. must be at least 2.4.
1289 * If not, we can possibly fall back on START_ARRAY
1290 * Try to GET_ARRAY_INFO.
1291 * If possible, give up
1292 * If not, try to STOP_ARRAY just to make sure
1293 *
1294 * If !uuidset and scan, look in conf-file for uuid
1295 * If not found, give up
1296 * If !devlist and scan and uuidset, get list of devs from conf-file
1297 *
1298 * For each device:
1299 * Check superblock - discard if bad
1300 * Check uuid (set if we don't have one) - discard if no match
1301 * Check superblock similarity if we have a superblock - discard if different
1302 * Record events, devicenum
1303 * This should give us a list of devices for the array
1304 * We should collect the most recent event number
1305 *
1306 * Count disks with recent enough event count
1307 * While force && !enough disks
1308 * Choose newest rejected disks, update event count
1309 * mark clean and rewrite superblock
1310 * If recent kernel:
1311 * SET_ARRAY_INFO
1312 * foreach device with recent events : ADD_NEW_DISK
1313 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1314 * If old kernel:
1315 * Check the device numbers in superblock are right
1316 * update superblock if any changes
1317 * START_ARRAY
1318 *
1319 */
1320 int rv = -1;
1321 int mdfd = -1;
1322 int clean;
1323 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1324 ident->super_minor == UnSet && ident->name[0] == 0 &&
1325 (ident->container == NULL || ident->member == NULL));
1326 struct devs *devices = NULL;
1327 char *devmap;
1328 int *best = NULL; /* indexed by raid_disk */
1329 int bestcnt = 0;
1330 int devcnt;
1331 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1332 int journal_clean = 0;
1333 int i;
1334 int was_forced = 0;
1335 int most_recent = 0;
1336 int chosen_drive;
1337 int change = 0;
1338 int inargv = 0;
1339 int start_partial_ok = (c->runstop >= 0) &&
1340 (c->force || devlist==NULL || auto_assem);
1341 int num_devs;
1342 struct mddev_dev *tmpdev;
1343 struct mdinfo info;
1344 struct mdinfo *content = NULL;
1345 struct mdinfo *pre_exist = NULL;
1346 char *avail;
1347 char *name = NULL;
1348 char chosen_name[1024];
1349 struct map_ent *map = NULL;
1350 struct map_ent *mp;
1351
1352 /*
1353 * If any subdevs are listed, then any that don't
1354 * match ident are discarded. Remainder must all match and
1355 * become the array.
1356 * If no subdevs, then we scan all devices in the config file, but
1357 * there must be something in the identity
1358 */
1359
1360 if (!devlist &&
1361 ident->uuid_set == 0 &&
1362 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1363 ident->name[0] == 0 &&
1364 (ident->container == NULL || ident->member == NULL) &&
1365 ident->devices == NULL) {
1366 pr_err("No identity information available for %s - cannot assemble.\n",
1367 mddev ? mddev : "further assembly");
1368 return 1;
1369 }
1370
1371 if (devlist == NULL)
1372 devlist = conf_get_devs();
1373 else if (mddev)
1374 inargv = 1;
1375
1376 try_again:
1377 /* We come back here when doing auto-assembly and attempting some
1378 * set of devices failed. Those are now marked as ->used==2 and
1379 * we ignore them and try again
1380 */
1381 if (!st && ident->st)
1382 st = ident->st;
1383 if (c->verbose>0)
1384 pr_err("looking for devices for %s\n",
1385 mddev ? mddev : "further assembly");
1386
1387 content = &info;
1388 if (st && c->force)
1389 st->ignore_hw_compat = 1;
1390 num_devs = select_devices(devlist, ident, &st, &content, c,
1391 inargv, auto_assem);
1392 if (num_devs < 0)
1393 return 1;
1394
1395 if (!st || !st->sb || !content)
1396 return 2;
1397
1398 /* We have a full set of devices - we now need to find the
1399 * array device.
1400 * However there is a risk that we are racing with "mdadm -I"
1401 * and the array is already partially assembled - we will have
1402 * rejected any devices already in this address.
1403 * So we take a lock on the map file - to prevent further races -
1404 * and look for the uuid in there. If found and the array is
1405 * active, we abort. If found and the array is not active
1406 * we commit to that md device and add all the contained devices
1407 * to our list. We flag them so that we don't try to re-add,
1408 * but can remove if they turn out to not be wanted.
1409 */
1410 if (map_lock(&map))
1411 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1412 if (c->update && strcmp(c->update,"uuid") == 0)
1413 mp = NULL;
1414 else
1415 mp = map_by_uuid(&map, content->uuid);
1416 if (mp) {
1417 struct mdinfo *dv;
1418 /* array already exists. */
1419 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1420 if (pre_exist->array.level != UnSet) {
1421 pr_err("Found some drive for an array that is already active: %s\n",
1422 mp->path);
1423 pr_err("giving up.\n");
1424 goto out;
1425 }
1426 for (dv = pre_exist->devs; dv; dv = dv->next) {
1427 /* We want to add this device to our list,
1428 * but it could already be there if "mdadm -I"
1429 * started *after* we checked for O_EXCL.
1430 * If we add it to the top of the list
1431 * it will be preferred over later copies.
1432 */
1433 struct mddev_dev *newdev;
1434 char *devname = map_dev(dv->disk.major,
1435 dv->disk.minor,
1436 0);
1437 if (!devname)
1438 continue;
1439 newdev = xmalloc(sizeof(*newdev));
1440 newdev->devname = devname;
1441 newdev->disposition = 'I';
1442 newdev->used = 1;
1443 newdev->next = devlist;
1444 devlist = newdev;
1445 num_devs++;
1446 }
1447 strcpy(chosen_name, mp->path);
1448 if (c->verbose > 0 || mddev == NULL ||
1449 strcmp(mddev, chosen_name) != 0)
1450 pr_err("Merging with already-assembled %s\n",
1451 chosen_name);
1452 mdfd = open_dev_excl(mp->devnm);
1453 } else {
1454 int trustworthy = FOREIGN;
1455 name = content->name;
1456 switch (st->ss->match_home(st, c->homehost)
1457 ?: st->ss->match_home(st, "any")) {
1458 case 1:
1459 trustworthy = LOCAL;
1460 name = strchr(content->name, ':');
1461 if (name)
1462 name++;
1463 else
1464 name = content->name;
1465 break;
1466 }
1467 if (!auto_assem)
1468 /* If the array is listed in mdadm.conf or on
1469 * command line, then we trust the name
1470 * even if the array doesn't look local
1471 */
1472 trustworthy = LOCAL;
1473
1474 if (name[0] == 0 &&
1475 content->array.level == LEVEL_CONTAINER) {
1476 name = content->text_version;
1477 trustworthy = METADATA;
1478 }
1479
1480 if (name[0] && trustworthy != LOCAL &&
1481 ! c->require_homehost &&
1482 conf_name_is_free(name))
1483 trustworthy = LOCAL;
1484
1485 if (trustworthy == LOCAL &&
1486 strchr(name, ':'))
1487 /* Ignore 'host:' prefix of name */
1488 name = strchr(name, ':')+1;
1489
1490 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1491 chosen_name, 0);
1492 }
1493 if (mdfd < 0) {
1494 st->ss->free_super(st);
1495 if (auto_assem)
1496 goto try_again;
1497 goto out;
1498 }
1499 mddev = chosen_name;
1500 if (pre_exist == NULL) {
1501 if (mddev_busy(fd2devnm(mdfd))) {
1502 pr_err("%s already active, cannot restart it!\n",
1503 mddev);
1504 for (tmpdev = devlist ;
1505 tmpdev && tmpdev->used != 1;
1506 tmpdev = tmpdev->next)
1507 ;
1508 if (tmpdev && auto_assem)
1509 pr_err("%s needed for %s...\n",
1510 mddev, tmpdev->devname);
1511 close(mdfd);
1512 mdfd = -3;
1513 st->ss->free_super(st);
1514 if (auto_assem)
1515 goto try_again;
1516 goto out;
1517 }
1518 /* just incase it was started but has no content */
1519 ioctl(mdfd, STOP_ARRAY, NULL);
1520 }
1521
1522 if (content != &info) {
1523 /* This is a member of a container. Try starting the array. */
1524 int err;
1525 err = assemble_container_content(st, mdfd, content, c,
1526 chosen_name, NULL);
1527 close(mdfd);
1528 return err;
1529 }
1530
1531 /* Ok, no bad inconsistancy, we can try updating etc */
1532 devices = xcalloc(num_devs, sizeof(*devices));
1533 devmap = xcalloc(num_devs, content->array.raid_disks);
1534 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1535 c, content, mdfd, mddev,
1536 &most_recent, &bestcnt, &best, inargv);
1537 if (devcnt < 0) {
1538 mdfd = -3;
1539 goto out;
1540 }
1541
1542 if (devcnt == 0) {
1543 pr_err("no devices found for %s\n",
1544 mddev);
1545 if (st)
1546 st->ss->free_super(st);
1547 free(devmap);
1548 goto out;
1549 }
1550
1551 if (c->update && strcmp(c->update, "byteorder")==0)
1552 st->minor_version = 90;
1553
1554 st->ss->getinfo_super(st, content, NULL);
1555 clean = content->array.state & 1;
1556
1557 /* now we have some devices that might be suitable.
1558 * I wonder how many
1559 */
1560 avail = xcalloc(content->array.raid_disks, 1);
1561 okcnt = 0;
1562 replcnt = 0;
1563 sparecnt=0;
1564 journalcnt=0;
1565 rebuilding_cnt=0;
1566 for (i=0; i< bestcnt; i++) {
1567 int j = best[i];
1568 int event_margin = 1; /* always allow a difference of '1'
1569 * like the kernel does
1570 */
1571 if (j < 0) continue;
1572 /* note: we ignore error flags in multipath arrays
1573 * as they don't make sense
1574 */
1575 if (content->array.level != LEVEL_MULTIPATH) {
1576 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1577 if (content->journal_device_required)
1578 journalcnt++;
1579 else /* unexpected journal, mark as faulty */
1580 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1581 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1582 if (!(devices[j].i.disk.state
1583 & (1<<MD_DISK_FAULTY))) {
1584 devices[j].uptodate = 1;
1585 sparecnt++;
1586 }
1587 continue;
1588 }
1589 }
1590 /* If this device thinks that 'most_recent' has failed, then
1591 * we must reject this device.
1592 */
1593 if (j != most_recent && !c->force &&
1594 content->array.raid_disks > 0 &&
1595 devices[most_recent].i.disk.raid_disk >= 0 &&
1596 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1597 if (c->verbose > -1)
1598 pr_err("ignoring %s as it reports %s as failed\n",
1599 devices[j].devname, devices[most_recent].devname);
1600 best[i] = -1;
1601 continue;
1602 }
1603 /* Require event counter to be same as, or just less than,
1604 * most recent. If it is bigger, it must be a stray spare and
1605 * should be ignored.
1606 */
1607 if (devices[j].i.events+event_margin >=
1608 devices[most_recent].i.events &&
1609 devices[j].i.events <=
1610 devices[most_recent].i.events
1611 ) {
1612 devices[j].uptodate = 1;
1613 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1614 journal_clean = 1;
1615 if (i < content->array.raid_disks * 2) {
1616 if (devices[j].i.recovery_start == MaxSector ||
1617 (content->reshape_active &&
1618 i >= content->array.raid_disks - content->delta_disks)) {
1619 if (!avail[i/2]) {
1620 okcnt++;
1621 avail[i/2]=1;
1622 } else
1623 replcnt++;
1624 } else
1625 rebuilding_cnt++;
1626 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1627 sparecnt++;
1628 }
1629 }
1630 free(devmap);
1631 if (c->force) {
1632 int force_ok = force_array(content, devices, best, bestcnt,
1633 avail, most_recent, st, c);
1634 okcnt += force_ok;
1635 if (force_ok)
1636 was_forced = 1;
1637 }
1638 /* Now we want to look at the superblock which the kernel will base things on
1639 * and compare the devices that we think are working with the devices that the
1640 * superblock thinks are working.
1641 * If there are differences and --force is given, then update this chosen
1642 * superblock.
1643 */
1644 chosen_drive = -1;
1645 st->ss->free_super(st);
1646 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1647 int j = best[i];
1648 int fd;
1649
1650 if (j<0)
1651 continue;
1652 if (!devices[j].uptodate)
1653 continue;
1654 if (devices[j].i.events < devices[most_recent].i.events)
1655 continue;
1656 chosen_drive = j;
1657 if ((fd=dev_open(devices[j].devname,
1658 devices[j].included ? O_RDONLY
1659 : (O_RDONLY|O_EXCL)))< 0) {
1660 pr_err("Cannot open %s: %s\n",
1661 devices[j].devname, strerror(errno));
1662 goto out;
1663 }
1664 if (st->ss->load_super(st,fd, NULL)) {
1665 close(fd);
1666 pr_err("RAID superblock has disappeared from %s\n",
1667 devices[j].devname);
1668 goto out;
1669 }
1670 close(fd);
1671 }
1672 if (st->sb == NULL) {
1673 pr_err("No suitable drives found for %s\n", mddev);
1674 goto out;
1675 }
1676 st->ss->getinfo_super(st, content, NULL);
1677 if (sysfs_init(content, mdfd, NULL)) {
1678 pr_err("Unable to initialize sysfs\n");
1679 goto out;
1680 }
1681
1682 /* after reload context, store journal_clean in context */
1683 content->journal_clean = journal_clean;
1684 for (i=0; i<bestcnt; i++) {
1685 int j = best[i];
1686 unsigned int desired_state;
1687
1688 if (j < 0)
1689 continue;
1690 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1691 desired_state = (1<<MD_DISK_JOURNAL);
1692 else if (i >= content->array.raid_disks * 2)
1693 desired_state = 0;
1694 else if (i & 1)
1695 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1696 else
1697 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1698
1699 if (!devices[j].uptodate)
1700 continue;
1701
1702 devices[j].i.disk.state = desired_state;
1703 if (!(devices[j].i.array.state & 1))
1704 clean = 0;
1705
1706 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1707 c->verbose, 0, NULL)) {
1708 if (c->force) {
1709 if (c->verbose >= 0)
1710 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1711 j, mddev, devices[j].devname);
1712 change = 1;
1713 } else {
1714 if (c->verbose >= -1)
1715 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1716 i, mddev, devices[j].devname);
1717 }
1718 }
1719 #if 0
1720 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1721 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1722 i, mddev);
1723 }
1724 #endif
1725 }
1726 if (c->force && !clean &&
1727 !enough(content->array.level, content->array.raid_disks,
1728 content->array.layout, clean,
1729 avail)) {
1730 change += st->ss->update_super(st, content, "force-array",
1731 devices[chosen_drive].devname, c->verbose,
1732 0, NULL);
1733 was_forced = 1;
1734 clean = 1;
1735 }
1736
1737 if (change) {
1738 int fd;
1739 fd = dev_open(devices[chosen_drive].devname,
1740 devices[chosen_drive].included ?
1741 O_RDWR : (O_RDWR|O_EXCL));
1742 if (fd < 0) {
1743 pr_err("Could not open %s for write - cannot Assemble array.\n",
1744 devices[chosen_drive].devname);
1745 goto out;
1746 }
1747 if (st->ss->store_super(st, fd)) {
1748 close(fd);
1749 pr_err("Could not re-write superblock on %s\n",
1750 devices[chosen_drive].devname);
1751 goto out;
1752 }
1753 if (c->verbose >= 0)
1754 pr_err("Marking array %s as 'clean'\n",
1755 mddev);
1756 close(fd);
1757 }
1758
1759 /* If we are in the middle of a reshape we may need to restore saved data
1760 * that was moved aside due to the reshape overwriting live data
1761 * The code of doing this lives in Grow.c
1762 */
1763 if (content->reshape_active &&
1764 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1765 int err = 0;
1766 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1767 if (c->verbose > 0)
1768 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1769 chosen_name);
1770 if (!c->backup_file)
1771 c->backup_file = locate_backup(content->sys_name);
1772 enable_fds(bestcnt/2);
1773 for (i = 0; i < bestcnt/2; i++) {
1774 int j = best[i*2];
1775 if (j >= 0) {
1776 fdlist[i] = dev_open(devices[j].devname,
1777 devices[j].included
1778 ? O_RDWR : (O_RDWR|O_EXCL));
1779 if (fdlist[i] < 0) {
1780 pr_err("Could not open %s for write - cannot Assemble array.\n",
1781 devices[j].devname);
1782 err = 1;
1783 break;
1784 }
1785 } else
1786 fdlist[i] = -1;
1787 }
1788 if (!err) {
1789 if (st->ss->external && st->ss->recover_backup)
1790 err = st->ss->recover_backup(st, content);
1791 else
1792 err = Grow_restart(st, content, fdlist, bestcnt/2,
1793 c->backup_file, c->verbose > 0);
1794 if (err && c->invalid_backup) {
1795 if (c->verbose > 0)
1796 pr_err("continuing without restoring backup\n");
1797 err = 0;
1798 }
1799 }
1800 while (i>0) {
1801 i--;
1802 if (fdlist[i]>=0) close(fdlist[i]);
1803 }
1804 free(fdlist);
1805 if (err) {
1806 pr_err("Failed to restore critical section for reshape, sorry.\n");
1807 if (c->backup_file == NULL)
1808 cont_err("Possibly you needed to specify the --backup-file\n");
1809 goto out;
1810 }
1811 }
1812
1813 /* Almost ready to actually *do* something */
1814 /* First, fill in the map, so that udev can find our name
1815 * as soon as we become active.
1816 */
1817 if (c->update && strcmp(c->update, "metadata")==0) {
1818 content->array.major_version = 1;
1819 content->array.minor_version = 0;
1820 strcpy(content->text_version, "1.0");
1821 }
1822
1823 map_update(&map, fd2devnm(mdfd), content->text_version,
1824 content->uuid, chosen_name);
1825
1826 rv = start_array(mdfd, mddev, content,
1827 st, ident, best, bestcnt,
1828 chosen_drive, devices, okcnt, sparecnt,
1829 rebuilding_cnt, journalcnt,
1830 c,
1831 clean, avail, start_partial_ok,
1832 pre_exist != NULL,
1833 was_forced);
1834 if (rv == 1 && !pre_exist)
1835 ioctl(mdfd, STOP_ARRAY, NULL);
1836 free(devices);
1837 out:
1838 map_unlock(&map);
1839 if (rv == 0) {
1840 wait_for(chosen_name, mdfd);
1841 close(mdfd);
1842 if (auto_assem) {
1843 int usecs = 1;
1844 /* There is a nasty race with 'mdadm --monitor'.
1845 * If it opens this device before we close it,
1846 * it gets an incomplete open on which IO
1847 * doesn't work and the capacity is
1848 * wrong.
1849 * If we reopen (to check for layered devices)
1850 * before --monitor closes, we loose.
1851 *
1852 * So: wait upto 1 second for there to be
1853 * a non-zero capacity.
1854 */
1855 while (usecs < 1000) {
1856 mdfd = open(mddev, O_RDONLY);
1857 if (mdfd >= 0) {
1858 unsigned long long size;
1859 if (get_dev_size(mdfd, NULL, &size) &&
1860 size > 0)
1861 break;
1862 close(mdfd);
1863 }
1864 usleep(usecs);
1865 usecs <<= 1;
1866 }
1867 }
1868 } else if (mdfd >= 0)
1869 close(mdfd);
1870
1871 /* '2' means 'OK, but not started yet' */
1872 if (rv == -1) {
1873 free(devices);
1874 return 1;
1875 }
1876 return rv == 2 ? 0 : rv;
1877 }
1878
1879 int assemble_container_content(struct supertype *st, int mdfd,
1880 struct mdinfo *content, struct context *c,
1881 char *chosen_name, int *result)
1882 {
1883 struct mdinfo *dev, *sra, *dev2;
1884 int working = 0, preexist = 0;
1885 int expansion = 0;
1886 int old_raid_disks;
1887 int start_reshape;
1888 char *avail;
1889 int err;
1890
1891 if (sysfs_init(content, mdfd, NULL)) {
1892 pr_err("Unable to initialize sysfs\n");
1893 return 1;
1894 }
1895
1896 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1897 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1898 if (content->array.major_version == -1 &&
1899 content->array.minor_version == -2 &&
1900 c->readonly &&
1901 content->text_version[0] == '/')
1902 content->text_version[0] = '-';
1903 if (sysfs_set_array(content, 9003) != 0) {
1904 sysfs_free(sra);
1905 return 1;
1906 }
1907 }
1908
1909 /* There are two types of reshape: container wide or sub-array specific
1910 * Check if metadata requests blocking container wide reshapes
1911 */
1912 start_reshape = (content->reshape_active &&
1913 !((content->reshape_active == CONTAINER_RESHAPE) &&
1914 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1915
1916 /* Block subarray here if it is under reshape now
1917 * Do not allow for any changes in this array
1918 */
1919 if (st->ss->external && content->recovery_blocked && start_reshape)
1920 block_subarray(content);
1921
1922 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1923 for (dev = content->devs; dev; dev = dev->next)
1924 if (dev2->disk.major == dev->disk.major &&
1925 dev2->disk.minor == dev->disk.minor)
1926 break;
1927 if (dev)
1928 continue;
1929 /* Don't want this one any more */
1930 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1931 errno == EBUSY) {
1932 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1933 sysfs_free(sra);
1934 return 1;
1935 }
1936 sysfs_set_str(sra, dev2, "state", "remove");
1937 }
1938 old_raid_disks = content->array.raid_disks - content->delta_disks;
1939 avail = xcalloc(content->array.raid_disks, 1);
1940 for (dev = content->devs; dev; dev = dev->next) {
1941 if (dev->disk.raid_disk >= 0)
1942 avail[dev->disk.raid_disk] = 1;
1943 if (sysfs_add_disk(content, dev, 1) == 0) {
1944 if (dev->disk.raid_disk >= old_raid_disks &&
1945 content->reshape_active)
1946 expansion++;
1947 else
1948 working++;
1949 } else if (errno == EEXIST)
1950 preexist++;
1951 }
1952 sysfs_free(sra);
1953 if (working + expansion == 0 && c->runstop <= 0) {
1954 free(avail);
1955 return 1;/* Nothing new, don't try to start */
1956 }
1957 map_update(NULL, fd2devnm(mdfd), content->text_version,
1958 content->uuid, chosen_name);
1959
1960 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1961 st->ss->validate_ppl) {
1962 content->array.state |= 1;
1963 err = 0;
1964
1965 for (dev = content->devs; dev; dev = dev->next) {
1966 int dfd;
1967 char *devpath;
1968 int ret;
1969
1970 ret = st->ss->validate_ppl(st, content, dev);
1971 if (ret == 0)
1972 continue;
1973
1974 if (ret < 0) {
1975 err = 1;
1976 break;
1977 }
1978
1979 if (!c->force) {
1980 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1981 chosen_name);
1982 content->array.state &= ~1;
1983 avail[dev->disk.raid_disk] = 0;
1984 break;
1985 }
1986
1987 /* have --force - overwrite the invalid ppl */
1988 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
1989 dfd = dev_open(devpath, O_RDWR);
1990 if (dfd < 0) {
1991 pr_err("Failed to open %s\n", devpath);
1992 err = 1;
1993 break;
1994 }
1995
1996 err = st->ss->write_init_ppl(st, content, dfd);
1997 close(dfd);
1998
1999 if (err)
2000 break;
2001 }
2002
2003 if (err) {
2004 free(avail);
2005 return err;
2006 }
2007 }
2008
2009 if (enough(content->array.level, content->array.raid_disks,
2010 content->array.layout, content->array.state & 1, avail) == 0) {
2011 if (c->export && result)
2012 *result |= INCR_NO;
2013 else if (c->verbose >= 0) {
2014 pr_err("%s assembled with %d device%s",
2015 chosen_name, preexist + working,
2016 preexist + working == 1 ? "":"s");
2017 if (preexist)
2018 fprintf(stderr, " (%d new)", working);
2019 fprintf(stderr, " but not started\n");
2020 }
2021 free(avail);
2022 return 1;
2023 }
2024 free(avail);
2025
2026 if (c->runstop <= 0 &&
2027 (working + preexist + expansion) <
2028 content->array.working_disks) {
2029 if (c->export && result)
2030 *result |= INCR_UNSAFE;
2031 else if (c->verbose >= 0) {
2032 pr_err("%s assembled with %d device%s",
2033 chosen_name, preexist + working,
2034 preexist + working == 1 ? "":"s");
2035 if (preexist)
2036 fprintf(stderr, " (%d new)", working);
2037 fprintf(stderr, " but not safe to start\n");
2038 }
2039 return 1;
2040 }
2041
2042
2043 if (start_reshape) {
2044 int spare = content->array.raid_disks + expansion;
2045 if (restore_backup(st, content,
2046 working,
2047 spare, &c->backup_file, c->verbose) == 1)
2048 return 1;
2049
2050 err = sysfs_set_str(content, NULL,
2051 "array_state", "readonly");
2052 if (err)
2053 return 1;
2054
2055 if (st->ss->external) {
2056 if (!mdmon_running(st->container_devnm))
2057 start_mdmon(st->container_devnm);
2058 ping_monitor(st->container_devnm);
2059 if (mdmon_running(st->container_devnm) &&
2060 st->update_tail == NULL)
2061 st->update_tail = &st->updates;
2062 }
2063
2064 err = Grow_continue(mdfd, st, content, c->backup_file,
2065 0, c->freeze_reshape);
2066 } else switch(content->array.level) {
2067 case LEVEL_LINEAR:
2068 case LEVEL_MULTIPATH:
2069 case 0:
2070 err = sysfs_set_str(content, NULL, "array_state",
2071 c->readonly ? "readonly" : "active");
2072 break;
2073 default:
2074 err = sysfs_set_str(content, NULL, "array_state",
2075 "readonly");
2076 /* start mdmon if needed. */
2077 if (!err) {
2078 if (!mdmon_running(st->container_devnm))
2079 start_mdmon(st->container_devnm);
2080 ping_monitor(st->container_devnm);
2081 }
2082 break;
2083 }
2084 if (!err)
2085 sysfs_set_safemode(content, content->safe_mode_delay);
2086
2087 /* Block subarray here if it is not reshaped now
2088 * It has be blocked a little later to allow mdmon to switch in
2089 * in to R/W state
2090 */
2091 if (st->ss->external && content->recovery_blocked &&
2092 !start_reshape)
2093 block_subarray(content);
2094
2095 if (c->export && result) {
2096 if (err)
2097 *result |= INCR_NO;
2098 else
2099 *result |= INCR_YES;
2100 } else if (c->verbose >= 0) {
2101 if (err)
2102 pr_err("array %s now has %d device%s",
2103 chosen_name, working + preexist,
2104 working + preexist == 1 ? "":"s");
2105 else
2106 pr_err("Started %s with %d device%s",
2107 chosen_name, working + preexist,
2108 working + preexist == 1 ? "":"s");
2109 if (preexist)
2110 fprintf(stderr, " (%d new)", working);
2111 if (expansion)
2112 fprintf(stderr, " ( + %d for expansion)",
2113 expansion);
2114 fprintf(stderr, "\n");
2115 }
2116 if (!err)
2117 wait_for(chosen_name, mdfd);
2118 return err;
2119 /* FIXME should have an O_EXCL and wait for read-auto */
2120 }