]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Incremental.c
Incremental - avoid including wayward devices.
[thirdparty/mdadm.git] / Incremental.c
1 /*
2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
4 *
5 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
6 *
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 *
22 * Author: Neil Brown
23 * Email: <neilb@suse.de>
24 * Paper: Neil Brown
25 * Novell Inc
26 * GPO Box Q1283
27 * QVB Post Office, NSW 1230
28 * Australia
29 */
30
31 #include "mdadm.h"
32 #include <dirent.h>
33 #include <ctype.h>
34
35 static int count_active(struct supertype *st, struct mdinfo *sra,
36 int mdfd, char **availp,
37 struct mdinfo *info);
38 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
39 int number, __u64 events, int verbose,
40 char *array_name);
41 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
42 struct map_ent *target,
43 struct supertype *st, int verbose);
44
45 static int Incremental_container(struct supertype *st, char *devname,
46 char *homehost,
47 int verbose, int runstop, int autof);
48
49 static struct mddev_ident *search_mdstat(struct supertype *st,
50 struct mdinfo *info,
51 char *devname,
52 int verbose, int *rvp);
53
54 int Incremental(char *devname, int verbose, int runstop,
55 struct supertype *st, char *homehost, int require_homehost,
56 int autof)
57 {
58 /* Add this device to an array, creating the array if necessary
59 * and starting the array if sensible or - if runstop>0 - if possible.
60 *
61 * This has several steps:
62 *
63 * 1/ Check if device is permitted by mdadm.conf, reject if not.
64 * 2/ Find metadata, reject if none appropriate (check
65 * version/name from args)
66 * 3/ Check if there is a match in mdadm.conf
67 * 3a/ if not, check for homehost match. If no match, assemble as
68 * a 'foreign' array.
69 * 4/ Determine device number.
70 * - If in mdadm.conf with std name, use that
71 * - UUID in /var/run/mdadm.map use that
72 * - If name is suggestive, use that. unless in use with different uuid.
73 * - Choose a free, high number.
74 * - Use a partitioned device unless strong suggestion not to.
75 * e.g. auto=md
76 * Don't choose partitioned for containers.
77 * 5/ Find out if array already exists
78 * 5a/ if it does not
79 * - choose a name, from mdadm.conf or 'name' field in array.
80 * - create the array
81 * - add the device
82 * 5b/ if it does
83 * - check one drive in array to make sure metadata is a reasonably
84 * close match. Reject if not (e.g. different type)
85 * - add the device
86 * 6/ Make sure /var/run/mdadm.map contains this array.
87 * 7/ Is there enough devices to possibly start the array?
88 * For a container, this means running Incremental_container.
89 * 7a/ if not, finish with success.
90 * 7b/ if yes,
91 * - read all metadata and arrange devices like -A does
92 * - if number of OK devices match expected, or -R and there are enough,
93 * start the array (auto-readonly).
94 */
95 struct stat stb;
96 struct mdinfo info, dinfo;
97 struct mdinfo *sra = NULL, *d;
98 struct mddev_ident *match;
99 char chosen_name[1024];
100 int rv = 1;
101 struct map_ent *mp, *map = NULL;
102 int dfd = -1, mdfd = -1;
103 char *avail = NULL;
104 int active_disks;
105 int trustworthy;
106 char *name_to_use;
107 mdu_array_info_t ainf;
108 struct dev_policy *policy = NULL;
109 struct map_ent target_array;
110 int have_target;
111
112 struct createinfo *ci = conf_get_create_info();
113
114 if (stat(devname, &stb) < 0) {
115 if (verbose >= 0)
116 fprintf(stderr, Name ": stat failed for %s: %s.\n",
117 devname, strerror(errno));
118 return rv;
119 }
120 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
121 if (verbose >= 0)
122 fprintf(stderr, Name ": %s is not a block device.\n",
123 devname);
124 return rv;
125 }
126 dfd = dev_open(devname, O_RDONLY|O_EXCL);
127 if (dfd < 0) {
128 if (verbose >= 0)
129 fprintf(stderr, Name ": cannot open %s: %s.\n",
130 devname, strerror(errno));
131 return rv;
132 }
133 /* If the device is a container, we do something very different */
134 if (must_be_container(dfd)) {
135 if (!st)
136 st = super_by_fd(dfd, NULL);
137 if (st)
138 rv = st->ss->load_container(st, dfd, NULL);
139
140 close(dfd);
141 if (!rv && st->ss->container_content)
142 return Incremental_container(st, devname, homehost,
143 verbose, runstop, autof);
144
145 fprintf(stderr, Name ": %s is not part of an md array.\n",
146 devname);
147 return rv;
148 }
149
150 /* 1/ Check if device is permitted by mdadm.conf */
151
152 if (!conf_test_dev(devname)) {
153 if (verbose >= 0)
154 fprintf(stderr, Name
155 ": %s not permitted by mdadm.conf.\n",
156 devname);
157 goto out;
158 }
159
160 /* 2/ Find metadata, reject if none appropriate (check
161 * version/name from args) */
162
163 if (fstat(dfd, &stb) < 0) {
164 if (verbose >= 0)
165 fprintf(stderr, Name ": fstat failed for %s: %s.\n",
166 devname, strerror(errno));
167 goto out;
168 }
169 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
170 if (verbose >= 0)
171 fprintf(stderr, Name ": %s is not a block device.\n",
172 devname);
173 goto out;
174 }
175
176 dinfo.disk.major = major(stb.st_rdev);
177 dinfo.disk.minor = minor(stb.st_rdev);
178
179 policy = disk_policy(&dinfo);
180 have_target = policy_check_path(&dinfo, &target_array);
181
182 if (st == NULL && (st = guess_super(dfd)) == NULL) {
183 if (verbose >= 0)
184 fprintf(stderr, Name
185 ": no recognisable superblock on %s.\n",
186 devname);
187 rv = try_spare(devname, &dfd, policy,
188 have_target ? &target_array : NULL,
189 st, verbose);
190 goto out;
191 }
192 if (st->ss->compare_super == NULL ||
193 st->ss->load_super(st, dfd, NULL)) {
194 if (verbose >= 0)
195 fprintf(stderr, Name ": no RAID superblock on %s.\n",
196 devname);
197 rv = try_spare(devname, &dfd, policy,
198 have_target ? &target_array : NULL,
199 st, verbose);
200 free(st);
201 goto out;
202 }
203 close (dfd); dfd = -1;
204
205 memset(&info, 0, sizeof(info));
206 st->ss->getinfo_super(st, &info, NULL);
207
208 /* 3/ Check if there is a match in mdadm.conf */
209 match = search_mdstat(st, &info, devname, verbose, &rv);
210 if (!match && rv == 2)
211 goto out;
212
213 if (match && match->devname
214 && strcasecmp(match->devname, "<ignore>") == 0) {
215 if (verbose >= 0)
216 fprintf(stderr, Name ": array containing %s is explicitly"
217 " ignored by mdadm.conf\n",
218 devname);
219 goto out;
220 }
221
222 /* 3a/ if not, check for homehost match. If no match, continue
223 * but don't trust the 'name' in the array. Thus a 'random' minor
224 * number will be assigned, and the device name will be based
225 * on that. */
226 if (match)
227 trustworthy = LOCAL;
228 else if (st->ss->match_home(st, homehost) == 1)
229 trustworthy = LOCAL;
230 else if (st->ss->match_home(st, "any") == 1)
231 trustworthy = LOCAL_ANY;
232 else
233 trustworthy = FOREIGN;
234
235
236 if (!match && !conf_test_metadata(st->ss->name, policy,
237 (trustworthy == LOCAL))) {
238 if (verbose >= 1)
239 fprintf(stderr, Name
240 ": %s has metadata type %s for which "
241 "auto-assembly is disabled\n",
242 devname, st->ss->name);
243 goto out;
244 }
245 if (trustworthy == LOCAL_ANY)
246 trustworthy = LOCAL;
247
248 /* There are three possible sources for 'autof': command line,
249 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
250 * ARRAY takes precedence, then command line, then
251 * CREATE.
252 */
253 if (match && match->autof)
254 autof = match->autof;
255 if (autof == 0)
256 autof = ci->autof;
257
258 name_to_use = info.name;
259 if (name_to_use[0] == 0 &&
260 info.array.level == LEVEL_CONTAINER &&
261 trustworthy == LOCAL) {
262 name_to_use = info.text_version;
263 trustworthy = METADATA;
264 }
265 if (name_to_use[0] && trustworthy != LOCAL &&
266 ! require_homehost &&
267 conf_name_is_free(name_to_use))
268 trustworthy = LOCAL;
269
270 /* strip "hostname:" prefix from name if we have decided
271 * to treat it as LOCAL
272 */
273 if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
274 name_to_use = strchr(name_to_use, ':')+1;
275
276 /* 4/ Check if array exists.
277 */
278 if (map_lock(&map))
279 fprintf(stderr, Name ": failed to get exclusive lock on "
280 "mapfile\n");
281 mp = map_by_uuid(&map, info.uuid);
282 if (mp)
283 mdfd = open_dev(mp->devnum);
284 else
285 mdfd = -1;
286
287 if (mdfd < 0) {
288
289 /* Couldn't find an existing array, maybe make a new one */
290 mdfd = create_mddev(match ? match->devname : NULL,
291 name_to_use, autof, trustworthy, chosen_name);
292
293 if (mdfd < 0)
294 goto out;
295
296 sysfs_init(&info, mdfd, 0);
297
298 if (set_array_info(mdfd, st, &info) != 0) {
299 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
300 chosen_name, strerror(errno));
301 rv = 2;
302 goto out;
303 }
304
305 dinfo = info;
306 dinfo.disk.major = major(stb.st_rdev);
307 dinfo.disk.minor = minor(stb.st_rdev);
308 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
309 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
310 devname, chosen_name, strerror(errno));
311 ioctl(mdfd, STOP_ARRAY, 0);
312 rv = 2;
313 goto out;
314 }
315 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
316 GET_OFFSET | GET_SIZE));
317
318 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
319 /* It really should be 'none' - must be old buggy
320 * kernel, and mdadm -I may not be able to complete.
321 * So reject it.
322 */
323 ioctl(mdfd, STOP_ARRAY, NULL);
324 fprintf(stderr, Name
325 ": You have an old buggy kernel which cannot support\n"
326 " --incremental reliably. Aborting.\n");
327 sysfs_free(sra);
328 rv = 2;
329 goto out;
330 }
331 info.array.working_disks = 1;
332 /* 6/ Make sure /var/run/mdadm.map contains this array. */
333 map_update(&map, fd2devnum(mdfd),
334 info.text_version,
335 info.uuid, chosen_name);
336 } else {
337 /* 5b/ if it does */
338 /* - check one drive in array to make sure metadata is a reasonably */
339 /* close match. Reject if not (e.g. different type) */
340 /* - add the device */
341 char dn[20];
342 int dfd2;
343 int err;
344 struct supertype *st2;
345 struct mdinfo info2, *d;
346
347 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
348 GET_OFFSET | GET_SIZE));
349
350 if (mp->path)
351 strcpy(chosen_name, mp->path);
352 else
353 strcpy(chosen_name, devnum2devname(mp->devnum));
354
355 /* It is generally not OK to add non-spare drives to a
356 * running array as they are probably missing because
357 * they failed. However if runstop is 1, then the
358 * array was possibly started early and our best bet is
359 * to add this anyway.
360 * Also if action policy is re-add or better we allow
361 * re-add
362 */
363 if ((info.disk.state & (1<<MD_DISK_SYNC)) != 0
364 && ! policy_action_allows(policy, st->ss->name,
365 act_re_add)
366 && runstop < 1) {
367 int active = 0;
368
369 if (st->ss->external) {
370 char *devname = devnum2devname(fd2devnum(mdfd));
371
372 active = devname && is_container_active(devname);
373 free(devname);
374 } else if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0)
375 active = 1;
376 if (active) {
377 fprintf(stderr, Name
378 ": not adding %s to active array (without --run) %s\n",
379 devname, chosen_name);
380 rv = 2;
381 goto out;
382 }
383 }
384 if (!sra) {
385 rv = 2;
386 goto out;
387 }
388 if (sra->devs) {
389 sprintf(dn, "%d:%d", sra->devs->disk.major,
390 sra->devs->disk.minor);
391 dfd2 = dev_open(dn, O_RDONLY);
392 st2 = dup_super(st);
393 if (st2->ss->load_super(st2, dfd2, NULL) ||
394 st->ss->compare_super(st, st2) != 0) {
395 fprintf(stderr, Name
396 ": metadata mismatch between %s and "
397 "chosen array %s\n",
398 devname, chosen_name);
399 close(dfd2);
400 rv = 2;
401 goto out;
402 }
403 close(dfd2);
404 memset(&info2, 0, sizeof(info2));
405 st2->ss->getinfo_super(st2, &info2, NULL);
406 st2->ss->free_super(st2);
407 if (info.array.level != info2.array.level ||
408 memcmp(info.uuid, info2.uuid, 16) != 0 ||
409 info.array.raid_disks != info2.array.raid_disks) {
410 fprintf(stderr, Name
411 ": unexpected difference between %s and %s.\n",
412 chosen_name, devname);
413 rv = 2;
414 goto out;
415 }
416 }
417 info2.disk.major = major(stb.st_rdev);
418 info2.disk.minor = minor(stb.st_rdev);
419 /* add disk needs to know about containers */
420 if (st->ss->external)
421 sra->array.level = LEVEL_CONTAINER;
422 err = add_disk(mdfd, st, sra, &info2);
423 if (err < 0 && errno == EBUSY) {
424 /* could be another device present with the same
425 * disk.number. Find and reject any such
426 */
427 find_reject(mdfd, st, sra, info.disk.number,
428 info.events, verbose, chosen_name);
429 err = add_disk(mdfd, st, sra, &info2);
430 }
431 if (err < 0) {
432 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
433 devname, chosen_name, strerror(errno));
434 rv = 2;
435 goto out;
436 }
437 info.array.working_disks = 0;
438 for (d = sra->devs; d; d=d->next)
439 info.array.working_disks ++;
440
441 }
442
443 /* 7/ Is there enough devices to possibly start the array? */
444 /* 7a/ if not, finish with success. */
445 if (info.array.level == LEVEL_CONTAINER) {
446 /* Try to assemble within the container */
447 map_unlock(&map);
448 sysfs_uevent(&info, "change");
449 if (verbose >= 0)
450 fprintf(stderr, Name
451 ": container %s now has %d devices\n",
452 chosen_name, info.array.working_disks);
453 wait_for(chosen_name, mdfd);
454 close(mdfd);
455 sysfs_free(sra);
456 rv = Incremental(chosen_name, verbose, runstop,
457 NULL, homehost, require_homehost, autof);
458 if (rv == 1)
459 /* Don't fail the whole -I if a subarray didn't
460 * have enough devices to start yet
461 */
462 rv = 0;
463 return rv;
464 }
465
466 /* We have added something to the array, so need to re-read the
467 * state. Eventually this state should be kept up-to-date as
468 * things change.
469 */
470 sysfs_free(sra);
471 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
472 GET_OFFSET | GET_SIZE));
473 active_disks = count_active(st, sra, mdfd, &avail, &info);
474 if (enough(info.array.level, info.array.raid_disks,
475 info.array.layout, info.array.state & 1,
476 avail, active_disks) == 0) {
477 if (verbose >= 0)
478 fprintf(stderr, Name
479 ": %s attached to %s, not enough to start (%d).\n",
480 devname, chosen_name, active_disks);
481 map_unlock(&map);
482 rv = 0;
483 goto out;
484 }
485
486 /* 7b/ if yes, */
487 /* - if number of OK devices match expected, or -R and there */
488 /* are enough, */
489 /* + add any bitmap file */
490 /* + start the array (auto-readonly). */
491
492 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
493 if (verbose >= 0)
494 fprintf(stderr, Name
495 ": %s attached to %s which is already active.\n",
496 devname, chosen_name);
497 map_unlock(&map);
498 rv = 0;
499 goto out;
500 }
501
502 map_unlock(&map);
503 if (runstop > 0 || active_disks >= info.array.working_disks) {
504 struct mdinfo *dsk;
505 /* Let's try to start it */
506 if (match && match->bitmap_file) {
507 int bmfd = open(match->bitmap_file, O_RDWR);
508 if (bmfd < 0) {
509 fprintf(stderr, Name
510 ": Could not open bitmap file %s.\n",
511 match->bitmap_file);
512 goto out;
513 }
514 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
515 close(bmfd);
516 fprintf(stderr, Name
517 ": Failed to set bitmapfile for %s.\n",
518 chosen_name);
519 goto out;
520 }
521 close(bmfd);
522 }
523 /* Need to remove from the array any devices which
524 * 'count_active' discerned were too old or inappropriate
525 */
526 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
527 if (d->disk.state & (1<<MD_DISK_REMOVED))
528 remove_disk(mdfd, st, sra, d);
529
530 if ((sra == NULL || active_disks >= info.array.working_disks)
531 && trustworthy != FOREIGN)
532 rv = ioctl(mdfd, RUN_ARRAY, NULL);
533 else
534 rv = sysfs_set_str(sra, NULL,
535 "array_state", "read-auto");
536 if (rv == 0) {
537 if (verbose >= 0)
538 fprintf(stderr, Name
539 ": %s attached to %s, which has been started.\n",
540 devname, chosen_name);
541 rv = 0;
542 wait_for(chosen_name, mdfd);
543 /* We just started the array, so some devices
544 * might have been evicted from the array
545 * because their event counts were too old.
546 * If the action=re-add policy is in-force for
547 * those devices we should re-add them now.
548 */
549 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
550 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
551 add_disk(mdfd, st, sra, dsk) == 0)
552 fprintf(stderr, Name
553 ": %s re-added to %s\n",
554 dsk->sys_name, chosen_name);
555 }
556 } else {
557 fprintf(stderr, Name
558 ": %s attached to %s, but failed to start: %s.\n",
559 devname, chosen_name, strerror(errno));
560 rv = 1;
561 }
562 } else {
563 if (verbose >= 0)
564 fprintf(stderr, Name
565 ": %s attached to %s, not enough to start safely.\n",
566 devname, chosen_name);
567 rv = 0;
568 }
569 out:
570 free(avail);
571 if (dfd >= 0)
572 close(dfd);
573 if (mdfd >= 0)
574 close(mdfd);
575 if (policy)
576 dev_policy_free(policy);
577 if (sra)
578 sysfs_free(sra);
579 return rv;
580 }
581
582 static struct mddev_ident *search_mdstat(struct supertype *st,
583 struct mdinfo *info,
584 char *devname,
585 int verbose, int *rvp)
586 {
587 struct mddev_ident *array_list, *match;
588 array_list = conf_get_ident(NULL);
589 match = NULL;
590 for (; array_list; array_list = array_list->next) {
591 if (array_list->uuid_set &&
592 same_uuid(array_list->uuid, info->uuid, st->ss->swapuuid)
593 == 0) {
594 if (verbose >= 2 && array_list->devname)
595 fprintf(stderr, Name
596 ": UUID differs from %s.\n",
597 array_list->devname);
598 continue;
599 }
600 if (array_list->name[0] &&
601 strcasecmp(array_list->name, info->name) != 0) {
602 if (verbose >= 2 && array_list->devname)
603 fprintf(stderr, Name
604 ": Name differs from %s.\n",
605 array_list->devname);
606 continue;
607 }
608 if (array_list->devices &&
609 !match_oneof(array_list->devices, devname)) {
610 if (verbose >= 2 && array_list->devname)
611 fprintf(stderr, Name
612 ": Not a listed device for %s.\n",
613 array_list->devname);
614 continue;
615 }
616 if (array_list->super_minor != UnSet &&
617 array_list->super_minor != info->array.md_minor) {
618 if (verbose >= 2 && array_list->devname)
619 fprintf(stderr, Name
620 ": Different super-minor to %s.\n",
621 array_list->devname);
622 continue;
623 }
624 if (!array_list->uuid_set &&
625 !array_list->name[0] &&
626 !array_list->devices &&
627 array_list->super_minor == UnSet) {
628 if (verbose >= 2 && array_list->devname)
629 fprintf(stderr, Name
630 ": %s doesn't have any identifying information.\n",
631 array_list->devname);
632 continue;
633 }
634 /* FIXME, should I check raid_disks and level too?? */
635
636 if (match) {
637 if (verbose >= 0) {
638 if (match->devname && array_list->devname)
639 fprintf(stderr, Name
640 ": we match both %s and %s - cannot decide which to use.\n",
641 match->devname, array_list->devname);
642 else
643 fprintf(stderr, Name
644 ": multiple lines in mdadm.conf match\n");
645 }
646 *rvp = 2;
647 match = NULL;
648 break;
649 }
650 match = array_list;
651 }
652 return match;
653 }
654
655 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
656 int number, __u64 events, int verbose,
657 char *array_name)
658 {
659 /* Find a device attached to this array with a disk.number of number
660 * and events less than the passed events, and remove the device.
661 */
662 struct mdinfo *d;
663 mdu_array_info_t ra;
664
665 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
666 return; /* not safe to remove from active arrays
667 * without thinking more */
668
669 for (d = sra->devs; d ; d = d->next) {
670 char dn[10];
671 int dfd;
672 struct mdinfo info;
673 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
674 dfd = dev_open(dn, O_RDONLY);
675 if (dfd < 0)
676 continue;
677 if (st->ss->load_super(st, dfd, NULL)) {
678 close(dfd);
679 continue;
680 }
681 st->ss->getinfo_super(st, &info, NULL);
682 st->ss->free_super(st);
683 close(dfd);
684
685 if (info.disk.number != number ||
686 info.events >= events)
687 continue;
688
689 if (d->disk.raid_disk > -1)
690 sysfs_set_str(sra, d, "slot", "none");
691 if (sysfs_set_str(sra, d, "state", "remove") == 0)
692 if (verbose >= 0)
693 fprintf(stderr, Name
694 ": removing old device %s from %s\n",
695 d->sys_name+4, array_name);
696 }
697 }
698
699 static int count_active(struct supertype *st, struct mdinfo *sra,
700 int mdfd, char **availp,
701 struct mdinfo *bestinfo)
702 {
703 /* count how many devices in sra think they are active */
704 struct mdinfo *d;
705 int cnt = 0;
706 __u64 max_events = 0;
707 char *avail = NULL;
708 int *best;
709 char *devmap = NULL;
710 int numdevs = 0;
711 int devnum;
712 int b, i;
713 int raid_disks = 0;
714
715 if (!sra)
716 return 0;
717
718 for (d = sra->devs ; d ; d = d->next)
719 numdevs++;
720 for (d = sra->devs, devnum=0 ; d ; d = d->next, devnum++) {
721 char dn[30];
722 int dfd;
723 int ok;
724 struct mdinfo info;
725
726 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
727 dfd = dev_open(dn, O_RDONLY);
728 if (dfd < 0)
729 continue;
730 ok = st->ss->load_super(st, dfd, NULL);
731 close(dfd);
732 if (ok != 0)
733 continue;
734 info.array.raid_disks = raid_disks;
735 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
736 if (!avail) {
737 raid_disks = info.array.raid_disks;
738 avail = calloc(raid_disks, 1);
739 if (!avail) {
740 fprintf(stderr, Name ": out of memory.\n");
741 exit(1);
742 }
743 *availp = avail;
744
745 best = calloc(raid_disks, sizeof(int));
746 devmap = calloc(raid_disks * numdevs, 1);
747
748 st->ss->getinfo_super(st, &info, devmap);
749 }
750
751 if (info.disk.state & (1<<MD_DISK_SYNC))
752 {
753 if (cnt == 0) {
754 cnt++;
755 max_events = info.events;
756 avail[info.disk.raid_disk] = 2;
757 best[info.disk.raid_disk] = devnum;
758 st->ss->getinfo_super(st, bestinfo, NULL);
759 } else if (info.events == max_events) {
760 avail[info.disk.raid_disk] = 2;
761 best[info.disk.raid_disk] = devnum;
762 } else if (info.events == max_events-1) {
763 if (avail[info.disk.raid_disk] == 0) {
764 avail[info.disk.raid_disk] = 1;
765 best[info.disk.raid_disk] = devnum;
766 }
767 } else if (info.events < max_events - 1)
768 ;
769 else if (info.events == max_events+1) {
770 int i;
771 max_events = info.events;
772 for (i=0; i < raid_disks; i++)
773 if (avail[i])
774 avail[i]--;
775 avail[info.disk.raid_disk] = 2;
776 best[info.disk.raid_disk] = devnum;
777 st->ss->getinfo_super(st, bestinfo, NULL);
778 } else { /* info.events much bigger */
779 memset(avail, 0, info.disk.raid_disk);
780 max_events = info.events;
781 avail[info.disk.raid_disk] = 2;
782 st->ss->getinfo_super(st, bestinfo, NULL);
783 }
784 }
785 st->ss->free_super(st);
786 }
787 if (!avail)
788 return 0;
789 /* We need to reject any device that thinks the best device is
790 * failed or missing */
791 for (b = 0; b < raid_disks; b++)
792 if (avail[b] == 2)
793 break;
794 cnt = 0;
795 for (i = 0 ; i < raid_disks ; i++) {
796 if (i != b && avail[i])
797 if (devmap[raid_disks * best[i] + b] == 0) {
798 /* This device thinks 'b' is failed -
799 * don't use it */
800 devnum = best[i];
801 for (d=sra->devs ; devnum; d = d->next)
802 devnum--;
803 d->disk.state |= (1 << MD_DISK_REMOVED);
804 avail[i] = 0;
805 }
806 if (avail[i])
807 cnt++;
808 }
809 free(best);
810 free(devmap);
811 return cnt;
812 }
813
814 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
815 struct map_ent *target, int bare,
816 struct supertype *st, int verbose)
817 {
818 /* This device doesn't have any md metadata
819 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
820 * If 'st' is not set, then we only know that some metadata allows this,
821 * others possibly don't.
822 * So look for a container or array to attach the device to.
823 * Prefer 'target' if that is set and the array is found.
824 *
825 * If st is set, then only arrays of that type are considered
826 * Return 0 on success, or some exit code on failure, probably 1.
827 */
828 int rv = -1;
829 struct stat stb;
830 struct map_ent *mp, *map = NULL;
831 struct mdinfo *chosen = NULL;
832 int dfd = *dfdp;
833
834 if (fstat(dfd, &stb) != 0)
835 return 1;
836
837 /*
838 * Now we need to find a suitable array to add this to.
839 * We only accept arrays that:
840 * - match 'st'
841 * - are in the same domains as the device
842 * - are of an size for which the device will be useful
843 * and we choose the one that is the most degraded
844 */
845
846 if (map_lock(&map)) {
847 fprintf(stderr, Name ": failed to get exclusive lock on "
848 "mapfile\n");
849 return 1;
850 }
851 for (mp = map ; mp ; mp = mp->next) {
852 struct supertype *st2;
853 struct domainlist *dl = NULL;
854 struct mdinfo *sra;
855 unsigned long long devsize;
856
857 if (is_subarray(mp->metadata))
858 continue;
859 if (st) {
860 st2 = st->ss->match_metadata_desc(mp->metadata);
861 if (!st2 ||
862 (st->minor_version >= 0 &&
863 st->minor_version != st2->minor_version)) {
864 if (verbose > 1)
865 fprintf(stderr, Name ": not adding %s to %s as metadata type doesn't match\n",
866 devname, mp->path);
867 free(st2);
868 continue;
869 }
870 free(st2);
871 }
872 sra = sysfs_read(-1, mp->devnum,
873 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
874 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
875 if (!sra) {
876 /* Probably a container - no degraded info */
877 sra = sysfs_read(-1, mp->devnum,
878 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
879 GET_COMPONENT|GET_VERSION);
880 if (sra)
881 sra->array.failed_disks = 0;
882 }
883 if (!sra)
884 continue;
885 if (st == NULL) {
886 int i;
887 st2 = NULL;
888 for(i=0; !st2 && superlist[i]; i++)
889 st2 = superlist[i]->match_metadata_desc(
890 sra->text_version);
891 if (!st2) {
892 if (verbose > 1)
893 fprintf(stderr, Name ": not adding %s to %s"
894 " as metadata not recognised.\n",
895 devname, mp->path);
896 goto next;
897 }
898 /* Need to double check the 'act_spare' permissions applies
899 * to this metadata.
900 */
901 if (!policy_action_allows(pol, st2->ss->name, act_spare))
902 goto next;
903 if (!bare && !policy_action_allows(pol, st2->ss->name,
904 act_spare_same_slot))
905 goto next;
906 } else
907 st2 = st;
908 get_dev_size(dfd, NULL, &devsize);
909 if (st2->ss->avail_size(st2, devsize) < sra->component_size) {
910 if (verbose > 1)
911 fprintf(stderr, Name ": not adding %s to %s as it is too small\n",
912 devname, mp->path);
913 goto next;
914 }
915 dl = domain_from_array(sra, st2->ss->name);
916 if (!domain_test(dl, pol, st2->ss->name)) {
917 /* domain test fails */
918 if (verbose > 1)
919 fprintf(stderr, Name ": not adding %s to %s as it is not in a compatible domain\n",
920 devname, mp->path);
921
922 goto next;
923 }
924 /* test against target.
925 * If 'target' is set and 'bare' is false, we only accept
926 * arrays/containers that match 'target'.
927 * If 'target' is set and 'bare' is true, we prefer the
928 * array which matches 'target'.
929 */
930 if (target) {
931 if (strcmp(target->metadata, mp->metadata) == 0 &&
932 memcmp(target->uuid, mp->uuid,
933 sizeof(target->uuid)) == 0) {
934 /* This is our target!! */
935 if (chosen)
936 sysfs_free(chosen);
937 chosen = sra;
938 sra = NULL;
939 /* skip to end so we don't check any more */
940 while (mp->next)
941 mp = mp->next;
942 goto next;
943 }
944 /* not our target */
945 if (!bare)
946 goto next;
947 }
948
949 /* all tests passed, OK to add to this array */
950 if (!chosen) {
951 chosen = sra;
952 sra = NULL;
953 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
954 sysfs_free(chosen);
955 chosen = sra;
956 sra = NULL;
957 }
958 next:
959 if (sra)
960 sysfs_free(sra);
961 if (st != st2)
962 free(st2);
963 if (dl)
964 domain_free(dl);
965 }
966 if (chosen) {
967 /* add current device to chosen array as a spare */
968 int mdfd = open_dev(devname2devnum(chosen->sys_name));
969 if (mdfd >= 0) {
970 struct mddev_dev devlist;
971 char devname[20];
972 devlist.next = NULL;
973 devlist.used = 0;
974 devlist.re_add = 0;
975 devlist.writemostly = 0;
976 devlist.devname = devname;
977 sprintf(devname, "%d:%d", major(stb.st_rdev),
978 minor(stb.st_rdev));
979 devlist.disposition = 'a';
980 close(dfd);
981 *dfdp = -1;
982 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
983 -1, 0);
984 close(mdfd);
985 }
986 if (verbose > 0) {
987 if (rv == 0)
988 fprintf(stderr, Name ": added %s as spare for %s\n",
989 devname, chosen->sys_name);
990 else
991 fprintf(stderr, Name ": failed to add %s as spare for %s\n",
992 devname, chosen->sys_name);
993 }
994 sysfs_free(chosen);
995 }
996 return rv ? 0 : 1;
997 }
998
999 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1000 struct supertype *st, int verbose)
1001 {
1002 /* we know that at least one partition virtual-metadata is
1003 * allowed to incorporate spares like this device. We need to
1004 * find a suitable device to copy partition information from.
1005 *
1006 * Getting a list of all disk (not partition) devices is
1007 * slightly non-trivial. We could look at /sys/block, but
1008 * that is theoretically due to be removed. Maybe best to use
1009 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1010 * we depend on this directory of 'path' info. But that fails
1011 * to find loop devices and probably others. Maybe don't
1012 * worry about that, they aren't the real target.
1013 *
1014 * So: check things in /dev/disk/by-path to see if they are in
1015 * a compatible domain, then load the partition table and see
1016 * if it is OK for the new device, and choose the largest
1017 * partition table that fits.
1018 */
1019 DIR *dir;
1020 struct dirent *de;
1021 char *chosen = NULL;
1022 unsigned long long chosen_size;
1023 struct supertype *chosen_st = NULL;
1024 int fd;
1025
1026 dir = opendir("/dev/disk/by-path");
1027 if (!dir)
1028 return 1;
1029 while ((de = readdir(dir)) != NULL) {
1030 char *ep;
1031 struct dev_policy *pol2 = NULL;
1032 struct domainlist *domlist = NULL;
1033 int fd = -1;
1034 struct mdinfo info;
1035 struct supertype *st2 = NULL;
1036 char *devname = NULL;
1037 unsigned long long devsectors;
1038
1039 if (de->d_ino == 0 ||
1040 de->d_name[0] == '.' ||
1041 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1042 goto next;
1043
1044 ep = de->d_name + strlen(de->d_name);
1045 while (ep > de->d_name &&
1046 isdigit(ep[-1]))
1047 ep--;
1048 if (ep > de->d_name + 5 &&
1049 strncmp(ep-5, "-part", 5) == 0)
1050 /* This is a partition - skip it */
1051 goto next;
1052
1053 pol2 = path_policy(de->d_name, type_disk);
1054
1055 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1056 if (domain_test(domlist, pol, st ? st->ss->name : NULL) == 0)
1057 /* new device is incompatible with this device. */
1058 goto next;
1059
1060 domain_free(domlist);
1061 domlist = NULL;
1062
1063 asprintf(&devname, "/dev/disk/by-path/%s", de->d_name);
1064 fd = open(devname, O_RDONLY);
1065 if (fd < 0)
1066 goto next;
1067 if (get_dev_size(fd, devname, &devsectors) == 0)
1068 goto next;
1069 devsectors >>= 9;
1070
1071 if (st)
1072 st2 = dup_super(st);
1073 else
1074 st2 = guess_super_type(fd, guess_partitions);
1075 if (st2 == NULL ||
1076 st2->ss->load_super(st2, fd, NULL) < 0)
1077 goto next;
1078
1079 if (!st) {
1080 /* Check domain policy again, this time referring to metadata */
1081 domain_merge(&domlist, pol2, st2->ss->name);
1082 if (domain_test(domlist, pol, st2->ss->name) == 0)
1083 /* Incompatible devices for this metadata type */
1084 goto next;
1085 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1086 /* Some partition types allow sparing, but not
1087 * this one.
1088 */
1089 goto next;
1090 }
1091
1092 st2->ss->getinfo_super(st2, &info, NULL);
1093 if (info.component_size > devsectors)
1094 /* This partitioning doesn't fit in the device */
1095 goto next;
1096
1097 /* This is an acceptable device to copy partition
1098 * metadata from. We could just stop here, but I
1099 * think I want to keep looking incase a larger
1100 * metadata which makes better use of the device can
1101 * be found.
1102 */
1103 if (chosen == NULL ||
1104 chosen_size < info.component_size) {
1105 chosen_size = info.component_size;
1106 free(chosen);
1107 chosen = devname;
1108 devname = NULL;
1109 if (chosen_st) {
1110 chosen_st->ss->free_super(chosen_st);
1111 free(chosen_st);
1112 }
1113 chosen_st = st2;
1114 st2 = NULL;
1115 }
1116
1117 next:
1118 free(devname);
1119 domain_free(domlist);
1120 dev_policy_free(pol2);
1121 if (st2)
1122 st2->ss->free_super(st2);
1123 free(st2);
1124
1125 if (fd >= 0)
1126 close(fd);
1127 }
1128
1129 if (!chosen)
1130 return 1;
1131
1132 /* 'chosen' is the best device we can find. Let's write its
1133 * metadata to devname dfd is read-only so don't use that
1134 */
1135 fd = open(devname, O_RDWR);
1136 if (fd >= 0) {
1137 chosen_st->ss->store_super(chosen_st, fd);
1138 close(fd);
1139 }
1140 free(chosen);
1141 chosen_st->ss->free_super(chosen_st);
1142 free(chosen_st);
1143 return 0;
1144 }
1145
1146 static int is_bare(int dfd)
1147 {
1148 unsigned long long size = 0;
1149 char bufpad[4096 + 4096];
1150 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1151
1152 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1153 read(dfd, buf, 4096) != 4096)
1154 return 0;
1155
1156 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1157 return 0;
1158 if (memcmp(buf, buf+1, 4095) != 0)
1159 return 0;
1160
1161 /* OK, first 4K appear blank, try the end. */
1162 get_dev_size(dfd, NULL, &size);
1163 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1164 read(dfd, buf, 4096) != 4096)
1165 return 0;
1166
1167 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1168 return 0;
1169 if (memcmp(buf, buf+1, 4095) != 0)
1170 return 0;
1171
1172 return 1;
1173 }
1174
1175 /* adding a spare to a regular array is quite different from adding one to
1176 * a set-of-partitions virtual array.
1177 * This function determines which is worth trying and tries as appropriate.
1178 * Arrays are given priority over partitions.
1179 */
1180 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1181 struct map_ent *target,
1182 struct supertype *st, int verbose)
1183 {
1184 int i;
1185 int rv;
1186 int arrays_ok = 0;
1187 int partitions_ok = 0;
1188 int dfd = *dfdp;
1189 int bare;
1190
1191 /* Can only add a spare if device has at least one domain */
1192 if (pol_find(pol, pol_domain) == NULL)
1193 return 1;
1194 /* And only if some action allows spares */
1195 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1196 return 1;
1197
1198 /* Now check if the device is bare.
1199 * bare devices can always be added as a spare
1200 * non-bare devices can only be added if spare-same-slot is permitted,
1201 * and this device is replacing a previous device - in which case 'target'
1202 * will be set.
1203 */
1204 if (!is_bare(dfd)) {
1205 /* Must have a target and allow same_slot */
1206 /* Later - may allow force_spare without target */
1207 if (!target ||
1208 !policy_action_allows(pol, st?st->ss->name:NULL,
1209 act_spare_same_slot)) {
1210 if (verbose > 1)
1211 fprintf(stderr, Name ": %s is not bare, so not "
1212 "considering as a spare\n",
1213 devname);
1214 return 1;
1215 }
1216 bare = 0;
1217 } else
1218 bare = 1;
1219
1220 /* It might be OK to add this device to an array - need to see
1221 * what arrays might be candidates.
1222 */
1223 if (st) {
1224 /* just try try 'array' or 'partition' based on this metadata */
1225 if (st->ss->add_to_super)
1226 return array_try_spare(devname, dfdp, pol, target, bare,
1227 st, verbose);
1228 else
1229 return partition_try_spare(devname, dfdp, pol,
1230 st, verbose);
1231 }
1232 /* No metadata was specified or found so options are open.
1233 * Check for whether any array metadata, or any partition metadata
1234 * might allow adding the spare. This check is just help to avoid
1235 * a more costly scan of all arrays when we can be sure that will
1236 * fail.
1237 */
1238 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1239 if (superlist[i]->add_to_super && !arrays_ok &&
1240 policy_action_allows(pol, superlist[i]->name, act_spare))
1241 arrays_ok = 1;
1242 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1243 policy_action_allows(pol, superlist[i]->name, act_spare))
1244 partitions_ok = 1;
1245 }
1246 rv = 1;
1247 if (arrays_ok)
1248 rv = array_try_spare(devname, dfdp, pol, target, bare,
1249 st, verbose);
1250 if (rv != 0 && partitions_ok)
1251 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1252 return rv;
1253 }
1254
1255 int IncrementalScan(int verbose)
1256 {
1257 /* look at every device listed in the 'map' file.
1258 * If one is found that is not running then:
1259 * look in mdadm.conf for bitmap file.
1260 * if one exists, but array has none, add it.
1261 * try to start array in auto-readonly mode
1262 */
1263 struct map_ent *mapl = NULL;
1264 struct map_ent *me;
1265 struct mddev_ident *devs, *mddev;
1266 int rv = 0;
1267
1268 map_read(&mapl);
1269 devs = conf_get_ident(NULL);
1270
1271 for (me = mapl ; me ; me = me->next) {
1272 mdu_array_info_t array;
1273 mdu_bitmap_file_t bmf;
1274 struct mdinfo *sra;
1275 int mdfd = open_dev(me->devnum);
1276
1277 if (mdfd < 0)
1278 continue;
1279 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1280 errno != ENODEV) {
1281 close(mdfd);
1282 continue;
1283 }
1284 /* Ok, we can try this one. Maybe it needs a bitmap */
1285 for (mddev = devs ; mddev ; mddev = mddev->next)
1286 if (mddev->devname && me->path
1287 && devname_matches(mddev->devname, me->path))
1288 break;
1289 if (mddev && mddev->bitmap_file) {
1290 /*
1291 * Note: early kernels will wrongly fail this, so it
1292 * is a hint only
1293 */
1294 int added = -1;
1295 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1296 int bmfd = open(mddev->bitmap_file, O_RDWR);
1297 if (bmfd >= 0) {
1298 added = ioctl(mdfd, SET_BITMAP_FILE,
1299 bmfd);
1300 close(bmfd);
1301 }
1302 }
1303 if (verbose >= 0) {
1304 if (added == 0)
1305 fprintf(stderr, Name
1306 ": Added bitmap %s to %s\n",
1307 mddev->bitmap_file, me->path);
1308 else if (errno != EEXIST)
1309 fprintf(stderr, Name
1310 ": Failed to add bitmap to %s: %s\n",
1311 me->path, strerror(errno));
1312 }
1313 }
1314 sra = sysfs_read(mdfd, 0, 0);
1315 if (sra) {
1316 if (sysfs_set_str(sra, NULL,
1317 "array_state", "read-auto") == 0) {
1318 if (verbose >= 0)
1319 fprintf(stderr, Name
1320 ": started array %s\n",
1321 me->path ?: devnum2devname(me->devnum));
1322 } else {
1323 fprintf(stderr, Name
1324 ": failed to start array %s: %s\n",
1325 me->path ?: devnum2devname(me->devnum),
1326 strerror(errno));
1327 rv = 1;
1328 }
1329 }
1330 }
1331 return rv;
1332 }
1333
1334 static char *container2devname(char *devname)
1335 {
1336 char *mdname = NULL;
1337
1338 if (devname[0] == '/') {
1339 int fd = open(devname, O_RDONLY);
1340 if (fd >= 0) {
1341 mdname = devnum2devname(fd2devnum(fd));
1342 close(fd);
1343 }
1344 } else {
1345 int uuid[4];
1346 struct map_ent *mp, *map = NULL;
1347
1348 if (!parse_uuid(devname, uuid))
1349 return mdname;
1350 mp = map_by_uuid(&map, uuid);
1351 if (mp)
1352 mdname = devnum2devname(mp->devnum);
1353 map_free(map);
1354 }
1355
1356 return mdname;
1357 }
1358
1359 static int Incremental_container(struct supertype *st, char *devname,
1360 char *homehost, int verbose,
1361 int runstop, int autof)
1362 {
1363 /* Collect the contents of this container and for each
1364 * array, choose a device name and assemble the array.
1365 */
1366
1367 struct mdinfo *list;
1368 struct mdinfo *ra;
1369 struct map_ent *map = NULL;
1370 struct mdinfo info;
1371 int trustworthy;
1372 struct mddev_ident *match;
1373 int rv = 0;
1374
1375 memset(&info, 0, sizeof(info));
1376 st->ss->getinfo_super(st, &info, NULL);
1377
1378 if ((runstop > 0 && info.container_enough >= 0) ||
1379 info.container_enough > 0)
1380 /* pass */;
1381 else {
1382 if (verbose)
1383 fprintf(stderr, Name ": not enough devices to start the container\n");
1384 return 0;
1385 }
1386
1387 match = search_mdstat(st, &info, devname, verbose, &rv);
1388 if (match == NULL && rv == 2)
1389 return rv;
1390
1391 /* Need to compute 'trustworthy' */
1392 if (match)
1393 trustworthy = LOCAL;
1394 else if (st->ss->match_home(st, homehost) == 1)
1395 trustworthy = LOCAL;
1396 else if (st->ss->match_home(st, "any") == 1)
1397 trustworthy = LOCAL;
1398 else
1399 trustworthy = FOREIGN;
1400
1401 list = st->ss->container_content(st, NULL);
1402 if (map_lock(&map))
1403 fprintf(stderr, Name ": failed to get exclusive lock on "
1404 "mapfile\n");
1405
1406 for (ra = list ; ra ; ra = ra->next) {
1407 int mdfd;
1408 char chosen_name[1024];
1409 struct map_ent *mp;
1410 struct mddev_ident *match = NULL;
1411
1412 mp = map_by_uuid(&map, ra->uuid);
1413
1414 if (mp) {
1415 mdfd = open_dev(mp->devnum);
1416 if (mp->path)
1417 strcpy(chosen_name, mp->path);
1418 else
1419 strcpy(chosen_name, devnum2devname(mp->devnum));
1420 } else {
1421
1422 /* Check in mdadm.conf for container == devname and
1423 * member == ra->text_version after second slash.
1424 */
1425 char *sub = strchr(ra->text_version+1, '/');
1426 struct mddev_ident *array_list;
1427 if (sub) {
1428 sub++;
1429 array_list = conf_get_ident(NULL);
1430 } else
1431 array_list = NULL;
1432 for(; array_list ; array_list = array_list->next) {
1433 char *dn;
1434 if (array_list->member == NULL ||
1435 array_list->container == NULL)
1436 continue;
1437 if (strcmp(array_list->member, sub) != 0)
1438 continue;
1439 if (array_list->uuid_set &&
1440 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1441 continue;
1442 dn = container2devname(array_list->container);
1443 if (dn == NULL)
1444 continue;
1445 if (strncmp(dn, ra->text_version+1,
1446 strlen(dn)) != 0 ||
1447 ra->text_version[strlen(dn)+1] != '/') {
1448 free(dn);
1449 continue;
1450 }
1451 free(dn);
1452 /* we have a match */
1453 match = array_list;
1454 if (verbose>0)
1455 fprintf(stderr, Name ": match found for member %s\n",
1456 array_list->member);
1457 break;
1458 }
1459
1460 if (match && match->devname &&
1461 strcasecmp(match->devname, "<ignore>") == 0) {
1462 if (verbose > 0)
1463 fprintf(stderr, Name ": array %s/%s is "
1464 "explicitly ignored by mdadm.conf\n",
1465 match->container, match->member);
1466 return 2;
1467 }
1468 if (match)
1469 trustworthy = LOCAL;
1470
1471 mdfd = create_mddev(match ? match->devname : NULL,
1472 ra->name,
1473 autof,
1474 trustworthy,
1475 chosen_name);
1476 }
1477
1478 if (mdfd < 0) {
1479 fprintf(stderr, Name ": failed to open %s: %s.\n",
1480 chosen_name, strerror(errno));
1481 return 2;
1482 }
1483
1484 assemble_container_content(st, mdfd, ra, runstop,
1485 chosen_name, verbose);
1486 }
1487 map_unlock(&map);
1488 return 0;
1489 }
1490
1491 /*
1492 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1493 * raid arrays, and if so first fail (if needed) and then remove the device.
1494 *
1495 * @devname - The device we want to remove
1496 * @id_path - name as found in /dev/disk/by-path for this device
1497 *
1498 * Note: the device name must be a kernel name like "sda", so
1499 * that we can find it in /proc/mdstat
1500 */
1501 int IncrementalRemove(char *devname, char *id_path, int verbose)
1502 {
1503 int mdfd;
1504 int rv;
1505 struct mdstat_ent *ent;
1506 struct mddev_dev devlist;
1507
1508 if (!id_path)
1509 dprintf(Name ": incremental removal without --path <id_path> "
1510 "lacks the possibility to re-add new device in this "
1511 "port\n");
1512
1513 if (strchr(devname, '/')) {
1514 fprintf(stderr, Name ": incremental removal requires a "
1515 "kernel device name, not a file: %s\n", devname);
1516 return 1;
1517 }
1518 ent = mdstat_by_component(devname);
1519 if (!ent) {
1520 fprintf(stderr, Name ": %s does not appear to be a component "
1521 "of any array\n", devname);
1522 return 1;
1523 }
1524 mdfd = open_dev(ent->devnum);
1525 if (mdfd < 0) {
1526 fprintf(stderr, Name ": Cannot open array %s!!\n", ent->dev);
1527 free_mdstat(ent);
1528 return 1;
1529 }
1530
1531 if (id_path) {
1532 struct map_ent *map = NULL, *me;
1533 me = map_by_devnum(&map, ent->devnum);
1534 if (me)
1535 policy_save_path(id_path, me);
1536 map_free(map);
1537 }
1538
1539 memset(&devlist, 0, sizeof(devlist));
1540 devlist.devname = devname;
1541 devlist.disposition = 'f';
1542 /* for a container, we must fail each member array */
1543 if (ent->metadata_version &&
1544 strncmp(ent->metadata_version, "external:", 9) == 0) {
1545 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1546 struct mdstat_ent *memb;
1547 for (memb = mdstat ; memb ; memb = memb->next)
1548 if (is_container_member(memb, ent->dev)) {
1549 int subfd = open_dev(memb->devnum);
1550 if (subfd >= 0) {
1551 Manage_subdevs(memb->dev, subfd,
1552 &devlist, verbose, 0);
1553 close(subfd);
1554 }
1555 }
1556 free_mdstat(mdstat);
1557 } else
1558 Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0);
1559 devlist.disposition = 'r';
1560 rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0);
1561 close(mdfd);
1562 free_mdstat(ent);
1563 return rv;
1564 }