]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Incremental.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Check all member devices in enough_fd
[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 && st->ss->load_container)
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 * This doesn't apply to containers as the 'non-spare'
363 * flag has a different meaning. The test has to happen
364 * at the device level there
365 */
366 if (!st->ss->external
367 && (info.disk.state & (1<<MD_DISK_SYNC)) != 0
368 && ! policy_action_allows(policy, st->ss->name,
369 act_re_add)
370 && runstop < 1) {
371 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
372 fprintf(stderr, Name
373 ": not adding %s to active array (without --run) %s\n",
374 devname, chosen_name);
375 rv = 2;
376 goto out;
377 }
378 }
379 if (!sra) {
380 rv = 2;
381 goto out;
382 }
383 if (sra->devs) {
384 sprintf(dn, "%d:%d", sra->devs->disk.major,
385 sra->devs->disk.minor);
386 dfd2 = dev_open(dn, O_RDONLY);
387 st2 = dup_super(st);
388 if (st2->ss->load_super(st2, dfd2, NULL) ||
389 st->ss->compare_super(st, st2) != 0) {
390 fprintf(stderr, Name
391 ": metadata mismatch between %s and "
392 "chosen array %s\n",
393 devname, chosen_name);
394 close(dfd2);
395 rv = 2;
396 goto out;
397 }
398 close(dfd2);
399 memset(&info2, 0, sizeof(info2));
400 st2->ss->getinfo_super(st2, &info2, NULL);
401 st2->ss->free_super(st2);
402 if (info.array.level != info2.array.level ||
403 memcmp(info.uuid, info2.uuid, 16) != 0 ||
404 info.array.raid_disks != info2.array.raid_disks) {
405 fprintf(stderr, Name
406 ": unexpected difference between %s and %s.\n",
407 chosen_name, devname);
408 rv = 2;
409 goto out;
410 }
411 }
412 info2.disk.major = major(stb.st_rdev);
413 info2.disk.minor = minor(stb.st_rdev);
414 /* add disk needs to know about containers */
415 if (st->ss->external)
416 sra->array.level = LEVEL_CONTAINER;
417 err = add_disk(mdfd, st, sra, &info2);
418 if (err < 0 && errno == EBUSY) {
419 /* could be another device present with the same
420 * disk.number. Find and reject any such
421 */
422 find_reject(mdfd, st, sra, info.disk.number,
423 info.events, verbose, chosen_name);
424 err = add_disk(mdfd, st, sra, &info2);
425 }
426 if (err < 0) {
427 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
428 devname, chosen_name, strerror(errno));
429 rv = 2;
430 goto out;
431 }
432 info.array.working_disks = 0;
433 for (d = sra->devs; d; d=d->next)
434 info.array.working_disks ++;
435
436 }
437
438 /* 7/ Is there enough devices to possibly start the array? */
439 /* 7a/ if not, finish with success. */
440 if (info.array.level == LEVEL_CONTAINER) {
441 int devnum = devnum; /* defined and used iff ->external */
442 /* Try to assemble within the container */
443 map_unlock(&map);
444 sysfs_uevent(&info, "change");
445 if (verbose >= 0)
446 fprintf(stderr, Name
447 ": container %s now has %d devices\n",
448 chosen_name, info.array.working_disks);
449 wait_for(chosen_name, mdfd);
450 if (st->ss->external)
451 devnum = fd2devnum(mdfd);
452 close(mdfd);
453 sysfs_free(sra);
454 rv = Incremental(chosen_name, verbose, runstop,
455 NULL, homehost, require_homehost, autof);
456 if (rv == 1)
457 /* Don't fail the whole -I if a subarray didn't
458 * have enough devices to start yet
459 */
460 rv = 0;
461 /* after spare is added, ping monitor for external metadata
462 * so that it can eg. try to rebuild degraded array */
463 if (st->ss->external)
464 ping_monitor_by_id(devnum);
465 return rv;
466 }
467
468 /* We have added something to the array, so need to re-read the
469 * state. Eventually this state should be kept up-to-date as
470 * things change.
471 */
472 sysfs_free(sra);
473 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
474 GET_OFFSET | GET_SIZE));
475 active_disks = count_active(st, sra, mdfd, &avail, &info);
476 if (enough(info.array.level, info.array.raid_disks,
477 info.array.layout, info.array.state & 1,
478 avail, active_disks) == 0) {
479 if (verbose >= 0)
480 fprintf(stderr, Name
481 ": %s attached to %s, not enough to start (%d).\n",
482 devname, chosen_name, active_disks);
483 map_unlock(&map);
484 rv = 0;
485 goto out;
486 }
487
488 /* 7b/ if yes, */
489 /* - if number of OK devices match expected, or -R and there */
490 /* are enough, */
491 /* + add any bitmap file */
492 /* + start the array (auto-readonly). */
493
494 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
495 if (verbose >= 0)
496 fprintf(stderr, Name
497 ": %s attached to %s which is already active.\n",
498 devname, chosen_name);
499 map_unlock(&map);
500 rv = 0;
501 goto out;
502 }
503
504 map_unlock(&map);
505 if (runstop > 0 || active_disks >= info.array.working_disks) {
506 struct mdinfo *dsk;
507 /* Let's try to start it */
508 if (match && match->bitmap_file) {
509 int bmfd = open(match->bitmap_file, O_RDWR);
510 if (bmfd < 0) {
511 fprintf(stderr, Name
512 ": Could not open bitmap file %s.\n",
513 match->bitmap_file);
514 goto out;
515 }
516 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
517 close(bmfd);
518 fprintf(stderr, Name
519 ": Failed to set bitmapfile for %s.\n",
520 chosen_name);
521 goto out;
522 }
523 close(bmfd);
524 }
525 /* Need to remove from the array any devices which
526 * 'count_active' discerned were too old or inappropriate
527 */
528 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
529 if (d->disk.state & (1<<MD_DISK_REMOVED))
530 remove_disk(mdfd, st, sra, d);
531
532 if ((sra == NULL || active_disks >= info.array.working_disks)
533 && trustworthy != FOREIGN)
534 rv = ioctl(mdfd, RUN_ARRAY, NULL);
535 else
536 rv = sysfs_set_str(sra, NULL,
537 "array_state", "read-auto");
538 if (rv == 0) {
539 if (verbose >= 0)
540 fprintf(stderr, Name
541 ": %s attached to %s, which has been started.\n",
542 devname, chosen_name);
543 rv = 0;
544 wait_for(chosen_name, mdfd);
545 /* We just started the array, so some devices
546 * might have been evicted from the array
547 * because their event counts were too old.
548 * If the action=re-add policy is in-force for
549 * those devices we should re-add them now.
550 */
551 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
552 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
553 add_disk(mdfd, st, sra, dsk) == 0)
554 fprintf(stderr, Name
555 ": %s re-added to %s\n",
556 dsk->sys_name, chosen_name);
557 }
558 } else {
559 fprintf(stderr, Name
560 ": %s attached to %s, but failed to start: %s.\n",
561 devname, chosen_name, strerror(errno));
562 rv = 1;
563 }
564 } else {
565 if (verbose >= 0)
566 fprintf(stderr, Name
567 ": %s attached to %s, not enough to start safely.\n",
568 devname, chosen_name);
569 rv = 0;
570 }
571 out:
572 free(avail);
573 if (dfd >= 0)
574 close(dfd);
575 if (mdfd >= 0)
576 close(mdfd);
577 if (policy)
578 dev_policy_free(policy);
579 if (sra)
580 sysfs_free(sra);
581 return rv;
582 }
583
584 static struct mddev_ident *search_mdstat(struct supertype *st,
585 struct mdinfo *info,
586 char *devname,
587 int verbose, int *rvp)
588 {
589 struct mddev_ident *array_list, *match;
590 array_list = conf_get_ident(NULL);
591 match = NULL;
592 for (; array_list; array_list = array_list->next) {
593 if (array_list->uuid_set &&
594 same_uuid(array_list->uuid, info->uuid, st->ss->swapuuid)
595 == 0) {
596 if (verbose >= 2 && array_list->devname)
597 fprintf(stderr, Name
598 ": UUID differs from %s.\n",
599 array_list->devname);
600 continue;
601 }
602 if (array_list->name[0] &&
603 strcasecmp(array_list->name, info->name) != 0) {
604 if (verbose >= 2 && array_list->devname)
605 fprintf(stderr, Name
606 ": Name differs from %s.\n",
607 array_list->devname);
608 continue;
609 }
610 if (array_list->devices &&
611 !match_oneof(array_list->devices, devname)) {
612 if (verbose >= 2 && array_list->devname)
613 fprintf(stderr, Name
614 ": Not a listed device for %s.\n",
615 array_list->devname);
616 continue;
617 }
618 if (array_list->super_minor != UnSet &&
619 array_list->super_minor != info->array.md_minor) {
620 if (verbose >= 2 && array_list->devname)
621 fprintf(stderr, Name
622 ": Different super-minor to %s.\n",
623 array_list->devname);
624 continue;
625 }
626 if (!array_list->uuid_set &&
627 !array_list->name[0] &&
628 !array_list->devices &&
629 array_list->super_minor == UnSet) {
630 if (verbose >= 2 && array_list->devname)
631 fprintf(stderr, Name
632 ": %s doesn't have any identifying information.\n",
633 array_list->devname);
634 continue;
635 }
636 /* FIXME, should I check raid_disks and level too?? */
637
638 if (match) {
639 if (verbose >= 0) {
640 if (match->devname && array_list->devname)
641 fprintf(stderr, Name
642 ": we match both %s and %s - cannot decide which to use.\n",
643 match->devname, array_list->devname);
644 else
645 fprintf(stderr, Name
646 ": multiple lines in mdadm.conf match\n");
647 }
648 *rvp = 2;
649 match = NULL;
650 break;
651 }
652 match = array_list;
653 }
654 return match;
655 }
656
657 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
658 int number, __u64 events, int verbose,
659 char *array_name)
660 {
661 /* Find a device attached to this array with a disk.number of number
662 * and events less than the passed events, and remove the device.
663 */
664 struct mdinfo *d;
665 mdu_array_info_t ra;
666
667 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
668 return; /* not safe to remove from active arrays
669 * without thinking more */
670
671 for (d = sra->devs; d ; d = d->next) {
672 char dn[10];
673 int dfd;
674 struct mdinfo info;
675 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
676 dfd = dev_open(dn, O_RDONLY);
677 if (dfd < 0)
678 continue;
679 if (st->ss->load_super(st, dfd, NULL)) {
680 close(dfd);
681 continue;
682 }
683 st->ss->getinfo_super(st, &info, NULL);
684 st->ss->free_super(st);
685 close(dfd);
686
687 if (info.disk.number != number ||
688 info.events >= events)
689 continue;
690
691 if (d->disk.raid_disk > -1)
692 sysfs_set_str(sra, d, "slot", "none");
693 if (sysfs_set_str(sra, d, "state", "remove") == 0)
694 if (verbose >= 0)
695 fprintf(stderr, Name
696 ": removing old device %s from %s\n",
697 d->sys_name+4, array_name);
698 }
699 }
700
701 static int count_active(struct supertype *st, struct mdinfo *sra,
702 int mdfd, char **availp,
703 struct mdinfo *bestinfo)
704 {
705 /* count how many devices in sra think they are active */
706 struct mdinfo *d;
707 int cnt = 0;
708 __u64 max_events = 0;
709 char *avail = NULL;
710 int *best;
711 char *devmap = NULL;
712 int numdevs = 0;
713 int devnum;
714 int b, i;
715 int raid_disks = 0;
716
717 if (!sra)
718 return 0;
719
720 for (d = sra->devs ; d ; d = d->next)
721 numdevs++;
722 for (d = sra->devs, devnum=0 ; d ; d = d->next, devnum++) {
723 char dn[30];
724 int dfd;
725 int ok;
726 struct mdinfo info;
727
728 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
729 dfd = dev_open(dn, O_RDONLY);
730 if (dfd < 0)
731 continue;
732 ok = st->ss->load_super(st, dfd, NULL);
733 close(dfd);
734 if (ok != 0)
735 continue;
736 info.array.raid_disks = raid_disks;
737 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
738 if (!avail) {
739 raid_disks = info.array.raid_disks;
740 avail = calloc(raid_disks, 1);
741 if (!avail) {
742 fprintf(stderr, Name ": out of memory.\n");
743 exit(1);
744 }
745 *availp = avail;
746
747 best = calloc(raid_disks, sizeof(int));
748 devmap = calloc(raid_disks * numdevs, 1);
749
750 st->ss->getinfo_super(st, &info, devmap);
751 }
752
753 if (info.disk.state & (1<<MD_DISK_SYNC))
754 {
755 if (cnt == 0) {
756 cnt++;
757 max_events = info.events;
758 avail[info.disk.raid_disk] = 2;
759 best[info.disk.raid_disk] = devnum;
760 st->ss->getinfo_super(st, bestinfo, NULL);
761 } else if (info.events == max_events) {
762 avail[info.disk.raid_disk] = 2;
763 best[info.disk.raid_disk] = devnum;
764 } else if (info.events == max_events-1) {
765 if (avail[info.disk.raid_disk] == 0) {
766 avail[info.disk.raid_disk] = 1;
767 best[info.disk.raid_disk] = devnum;
768 }
769 } else if (info.events < max_events - 1)
770 ;
771 else if (info.events == max_events+1) {
772 int i;
773 max_events = info.events;
774 for (i=0; i < raid_disks; i++)
775 if (avail[i])
776 avail[i]--;
777 avail[info.disk.raid_disk] = 2;
778 best[info.disk.raid_disk] = devnum;
779 st->ss->getinfo_super(st, bestinfo, NULL);
780 } else { /* info.events much bigger */
781 memset(avail, 0, info.disk.raid_disk);
782 max_events = info.events;
783 avail[info.disk.raid_disk] = 2;
784 st->ss->getinfo_super(st, bestinfo, NULL);
785 }
786 }
787 st->ss->free_super(st);
788 }
789 if (!avail)
790 return 0;
791 /* We need to reject any device that thinks the best device is
792 * failed or missing */
793 for (b = 0; b < raid_disks; b++)
794 if (avail[b] == 2)
795 break;
796 cnt = 0;
797 for (i = 0 ; i < raid_disks ; i++) {
798 if (i != b && avail[i])
799 if (devmap[raid_disks * best[i] + b] == 0) {
800 /* This device thinks 'b' is failed -
801 * don't use it */
802 devnum = best[i];
803 for (d=sra->devs ; devnum; d = d->next)
804 devnum--;
805 d->disk.state |= (1 << MD_DISK_REMOVED);
806 avail[i] = 0;
807 }
808 if (avail[i])
809 cnt++;
810 }
811 free(best);
812 free(devmap);
813 return cnt;
814 }
815
816 /* test if container has degraded member(s) */
817 static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
818 {
819 mdu_array_info_t array;
820 int afd;
821 int max_degraded = 0;
822
823 for(; map; map = map->next) {
824 if (!is_subarray(map->metadata) ||
825 devname2devnum(map->metadata+1) != me->devnum)
826 continue;
827 afd = open_dev(map->devnum);
828 if (afd < 0)
829 continue;
830 /* most accurate information regarding array degradation */
831 if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) {
832 int degraded = array.raid_disks - array.active_disks -
833 array.spare_disks;
834 if (degraded > max_degraded)
835 max_degraded = degraded;
836 }
837 close(afd);
838 }
839 return (max_degraded);
840 }
841
842 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
843 struct map_ent *target, int bare,
844 struct supertype *st, int verbose)
845 {
846 /* This device doesn't have any md metadata
847 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
848 * If 'st' is not set, then we only know that some metadata allows this,
849 * others possibly don't.
850 * So look for a container or array to attach the device to.
851 * Prefer 'target' if that is set and the array is found.
852 *
853 * If st is set, then only arrays of that type are considered
854 * Return 0 on success, or some exit code on failure, probably 1.
855 */
856 int rv = 1;
857 struct stat stb;
858 struct map_ent *mp, *map = NULL;
859 struct mdinfo *chosen = NULL;
860 int dfd = *dfdp;
861
862 if (fstat(dfd, &stb) != 0)
863 return 1;
864
865 /*
866 * Now we need to find a suitable array to add this to.
867 * We only accept arrays that:
868 * - match 'st'
869 * - are in the same domains as the device
870 * - are of an size for which the device will be useful
871 * and we choose the one that is the most degraded
872 */
873
874 if (map_lock(&map)) {
875 fprintf(stderr, Name ": failed to get exclusive lock on "
876 "mapfile\n");
877 return 1;
878 }
879 for (mp = map ; mp ; mp = mp->next) {
880 struct supertype *st2;
881 struct domainlist *dl = NULL;
882 struct mdinfo *sra;
883 unsigned long long devsize;
884 unsigned long long component_size = 0;
885
886 if (is_subarray(mp->metadata))
887 continue;
888 if (st) {
889 st2 = st->ss->match_metadata_desc(mp->metadata);
890 if (!st2 ||
891 (st->minor_version >= 0 &&
892 st->minor_version != st2->minor_version)) {
893 if (verbose > 1)
894 fprintf(stderr, Name ": not adding %s to %s as metadata type doesn't match\n",
895 devname, mp->path);
896 free(st2);
897 continue;
898 }
899 free(st2);
900 }
901 sra = sysfs_read(-1, mp->devnum,
902 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
903 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
904 if (!sra) {
905 /* Probably a container - no degraded info */
906 sra = sysfs_read(-1, mp->devnum,
907 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
908 GET_COMPONENT|GET_VERSION);
909 if (sra)
910 sra->array.failed_disks = -1;
911 }
912 if (!sra)
913 continue;
914 if (st == NULL) {
915 int i;
916 st2 = NULL;
917 for(i=0; !st2 && superlist[i]; i++)
918 st2 = superlist[i]->match_metadata_desc(
919 sra->text_version);
920 if (!st2) {
921 if (verbose > 1)
922 fprintf(stderr, Name ": not adding %s to %s"
923 " as metadata not recognised.\n",
924 devname, mp->path);
925 goto next;
926 }
927 /* Need to double check the 'act_spare' permissions applies
928 * to this metadata.
929 */
930 if (!policy_action_allows(pol, st2->ss->name, act_spare))
931 goto next;
932 if (!bare && !policy_action_allows(pol, st2->ss->name,
933 act_spare_same_slot))
934 goto next;
935 } else
936 st2 = st;
937 /* update number of failed disks for mostly degraded
938 * container member */
939 if (sra->array.failed_disks == -1)
940 sra->array.failed_disks = container_members_max_degradation(map, mp);
941
942 get_dev_size(dfd, NULL, &devsize);
943 if (sra->component_size == 0) {
944 /* true for containers, here we must read superblock
945 * to obtain minimum spare size */
946 struct supertype *st3 = dup_super(st2);
947 int mdfd = open_dev(mp->devnum);
948 if (!mdfd)
949 goto next;
950 if (st3->ss->load_container &&
951 !st3->ss->load_container(st3, mdfd, mp->path)) {
952 component_size = st3->ss->min_acceptable_spare_size(st3);
953 st3->ss->free_super(st3);
954 }
955 free(st3);
956 close(mdfd);
957 }
958 if ((sra->component_size > 0 &&
959 st2->ss->avail_size(st2, devsize) < sra->component_size)
960 ||
961 (sra->component_size == 0 && devsize < component_size)) {
962 if (verbose > 1)
963 fprintf(stderr, Name ": not adding %s to %s as it is too small\n",
964 devname, mp->path);
965 goto next;
966 }
967 /* test against target.
968 * If 'target' is set and 'bare' is false, we only accept
969 * arrays/containers that match 'target'.
970 * If 'target' is set and 'bare' is true, we prefer the
971 * array which matches 'target'.
972 * target is considered only if we deal with degraded array
973 */
974 if (target && policy_action_allows(pol, st2->ss->name,
975 act_spare_same_slot)) {
976 if (strcmp(target->metadata, mp->metadata) == 0 &&
977 memcmp(target->uuid, mp->uuid,
978 sizeof(target->uuid)) == 0 &&
979 sra->array.failed_disks > 0) {
980 /* This is our target!! */
981 if (chosen)
982 sysfs_free(chosen);
983 chosen = sra;
984 sra = NULL;
985 /* skip to end so we don't check any more */
986 while (mp->next)
987 mp = mp->next;
988 goto next;
989 }
990 /* not our target */
991 if (!bare)
992 goto next;
993 }
994
995 dl = domain_from_array(sra, st2->ss->name);
996 if (domain_test(dl, pol, st2->ss->name) != 1) {
997 /* domain test fails */
998 if (verbose > 1)
999 fprintf(stderr, Name ": not adding %s to %s as"
1000 " it is not in a compatible domain\n",
1001 devname, mp->path);
1002
1003 goto next;
1004 }
1005 /* all tests passed, OK to add to this array */
1006 if (!chosen) {
1007 chosen = sra;
1008 sra = NULL;
1009 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
1010 sysfs_free(chosen);
1011 chosen = sra;
1012 sra = NULL;
1013 }
1014 next:
1015 if (sra)
1016 sysfs_free(sra);
1017 if (st != st2)
1018 free(st2);
1019 if (dl)
1020 domain_free(dl);
1021 }
1022 if (chosen) {
1023 /* add current device to chosen array as a spare */
1024 int mdfd = open_dev(devname2devnum(chosen->sys_name));
1025 if (mdfd >= 0) {
1026 struct mddev_dev devlist;
1027 char devname[20];
1028 devlist.next = NULL;
1029 devlist.used = 0;
1030 devlist.re_add = 0;
1031 devlist.writemostly = 0;
1032 devlist.devname = devname;
1033 sprintf(devname, "%d:%d", major(stb.st_rdev),
1034 minor(stb.st_rdev));
1035 devlist.disposition = 'a';
1036 close(dfd);
1037 *dfdp = -1;
1038 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1039 -1, 0, NULL);
1040 close(mdfd);
1041 }
1042 if (verbose > 0) {
1043 if (rv == 0)
1044 fprintf(stderr, Name ": added %s as spare for %s\n",
1045 devname, chosen->sys_name);
1046 else
1047 fprintf(stderr, Name ": failed to add %s as spare for %s\n",
1048 devname, chosen->sys_name);
1049 }
1050 sysfs_free(chosen);
1051 }
1052 return rv;
1053 }
1054
1055 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1056 struct supertype *st, int verbose)
1057 {
1058 /* we know that at least one partition virtual-metadata is
1059 * allowed to incorporate spares like this device. We need to
1060 * find a suitable device to copy partition information from.
1061 *
1062 * Getting a list of all disk (not partition) devices is
1063 * slightly non-trivial. We could look at /sys/block, but
1064 * that is theoretically due to be removed. Maybe best to use
1065 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1066 * we depend on this directory of 'path' info. But that fails
1067 * to find loop devices and probably others. Maybe don't
1068 * worry about that, they aren't the real target.
1069 *
1070 * So: check things in /dev/disk/by-path to see if they are in
1071 * a compatible domain, then load the partition table and see
1072 * if it is OK for the new device, and choose the largest
1073 * partition table that fits.
1074 */
1075 DIR *dir;
1076 struct dirent *de;
1077 char *chosen = NULL;
1078 unsigned long long chosen_size = 0;
1079 struct supertype *chosen_st = NULL;
1080 int fd;
1081
1082 dir = opendir("/dev/disk/by-path");
1083 if (!dir)
1084 return 1;
1085 while ((de = readdir(dir)) != NULL) {
1086 char *ep;
1087 struct dev_policy *pol2 = NULL;
1088 struct domainlist *domlist = NULL;
1089 int fd = -1;
1090 struct mdinfo info;
1091 struct supertype *st2 = NULL;
1092 char *devname = NULL;
1093 unsigned long long devsectors;
1094
1095 if (de->d_ino == 0 ||
1096 de->d_name[0] == '.' ||
1097 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1098 goto next;
1099
1100 ep = de->d_name + strlen(de->d_name);
1101 while (ep > de->d_name &&
1102 isdigit(ep[-1]))
1103 ep--;
1104 if (ep > de->d_name + 5 &&
1105 strncmp(ep-5, "-part", 5) == 0)
1106 /* This is a partition - skip it */
1107 goto next;
1108
1109 pol2 = path_policy(de->d_name, type_disk);
1110
1111 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1112 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1113 /* new device is incompatible with this device. */
1114 goto next;
1115
1116 domain_free(domlist);
1117 domlist = NULL;
1118
1119 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1120 devname = NULL;
1121 goto next;
1122 }
1123 fd = open(devname, O_RDONLY);
1124 if (fd < 0)
1125 goto next;
1126 if (get_dev_size(fd, devname, &devsectors) == 0)
1127 goto next;
1128 devsectors >>= 9;
1129
1130 if (st)
1131 st2 = dup_super(st);
1132 else
1133 st2 = guess_super_type(fd, guess_partitions);
1134 if (st2 == NULL ||
1135 st2->ss->load_super(st2, fd, NULL) < 0)
1136 goto next;
1137
1138 if (!st) {
1139 /* Check domain policy again, this time referring to metadata */
1140 domain_merge(&domlist, pol2, st2->ss->name);
1141 if (domain_test(domlist, pol, st2->ss->name) != 1)
1142 /* Incompatible devices for this metadata type */
1143 goto next;
1144 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1145 /* Some partition types allow sparing, but not
1146 * this one.
1147 */
1148 goto next;
1149 }
1150
1151 st2->ss->getinfo_super(st2, &info, NULL);
1152 if (info.component_size > devsectors)
1153 /* This partitioning doesn't fit in the device */
1154 goto next;
1155
1156 /* This is an acceptable device to copy partition
1157 * metadata from. We could just stop here, but I
1158 * think I want to keep looking incase a larger
1159 * metadata which makes better use of the device can
1160 * be found.
1161 */
1162 if (chosen == NULL ||
1163 chosen_size < info.component_size) {
1164 chosen_size = info.component_size;
1165 free(chosen);
1166 chosen = devname;
1167 devname = NULL;
1168 if (chosen_st) {
1169 chosen_st->ss->free_super(chosen_st);
1170 free(chosen_st);
1171 }
1172 chosen_st = st2;
1173 st2 = NULL;
1174 }
1175
1176 next:
1177 free(devname);
1178 domain_free(domlist);
1179 dev_policy_free(pol2);
1180 if (st2)
1181 st2->ss->free_super(st2);
1182 free(st2);
1183
1184 if (fd >= 0)
1185 close(fd);
1186 }
1187
1188 if (!chosen)
1189 return 1;
1190
1191 /* 'chosen' is the best device we can find. Let's write its
1192 * metadata to devname dfd is read-only so don't use that
1193 */
1194 fd = open(devname, O_RDWR);
1195 if (fd >= 0) {
1196 chosen_st->ss->store_super(chosen_st, fd);
1197 close(fd);
1198 }
1199 free(chosen);
1200 chosen_st->ss->free_super(chosen_st);
1201 free(chosen_st);
1202 return 0;
1203 }
1204
1205 static int is_bare(int dfd)
1206 {
1207 unsigned long long size = 0;
1208 char bufpad[4096 + 4096];
1209 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1210
1211 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1212 read(dfd, buf, 4096) != 4096)
1213 return 0;
1214
1215 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1216 return 0;
1217 if (memcmp(buf, buf+1, 4095) != 0)
1218 return 0;
1219
1220 /* OK, first 4K appear blank, try the end. */
1221 get_dev_size(dfd, NULL, &size);
1222 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1223 read(dfd, buf, 4096) != 4096)
1224 return 0;
1225
1226 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1227 return 0;
1228 if (memcmp(buf, buf+1, 4095) != 0)
1229 return 0;
1230
1231 return 1;
1232 }
1233
1234 /* adding a spare to a regular array is quite different from adding one to
1235 * a set-of-partitions virtual array.
1236 * This function determines which is worth trying and tries as appropriate.
1237 * Arrays are given priority over partitions.
1238 */
1239 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1240 struct map_ent *target,
1241 struct supertype *st, int verbose)
1242 {
1243 int i;
1244 int rv;
1245 int arrays_ok = 0;
1246 int partitions_ok = 0;
1247 int dfd = *dfdp;
1248 int bare;
1249
1250 /* Can only add a spare if device has at least one domain */
1251 if (pol_find(pol, pol_domain) == NULL)
1252 return 1;
1253 /* And only if some action allows spares */
1254 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1255 return 1;
1256
1257 /* Now check if the device is bare.
1258 * bare devices can always be added as a spare
1259 * non-bare devices can only be added if spare-same-slot is permitted,
1260 * and this device is replacing a previous device - in which case 'target'
1261 * will be set.
1262 */
1263 if (!is_bare(dfd)) {
1264 /* Must have a target and allow same_slot */
1265 /* Later - may allow force_spare without target */
1266 if (!target ||
1267 !policy_action_allows(pol, st?st->ss->name:NULL,
1268 act_spare_same_slot)) {
1269 if (verbose > 1)
1270 fprintf(stderr, Name ": %s is not bare, so not "
1271 "considering as a spare\n",
1272 devname);
1273 return 1;
1274 }
1275 bare = 0;
1276 } else
1277 bare = 1;
1278
1279 /* It might be OK to add this device to an array - need to see
1280 * what arrays might be candidates.
1281 */
1282 if (st) {
1283 /* just try try 'array' or 'partition' based on this metadata */
1284 if (st->ss->add_to_super)
1285 return array_try_spare(devname, dfdp, pol, target, bare,
1286 st, verbose);
1287 else
1288 return partition_try_spare(devname, dfdp, pol,
1289 st, verbose);
1290 }
1291 /* No metadata was specified or found so options are open.
1292 * Check for whether any array metadata, or any partition metadata
1293 * might allow adding the spare. This check is just help to avoid
1294 * a more costly scan of all arrays when we can be sure that will
1295 * fail.
1296 */
1297 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1298 if (superlist[i]->add_to_super && !arrays_ok &&
1299 policy_action_allows(pol, superlist[i]->name, act_spare))
1300 arrays_ok = 1;
1301 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1302 policy_action_allows(pol, superlist[i]->name, act_spare))
1303 partitions_ok = 1;
1304 }
1305 rv = 1;
1306 if (arrays_ok)
1307 rv = array_try_spare(devname, dfdp, pol, target, bare,
1308 st, verbose);
1309 if (rv != 0 && partitions_ok)
1310 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1311 return rv;
1312 }
1313
1314 int IncrementalScan(int verbose)
1315 {
1316 /* look at every device listed in the 'map' file.
1317 * If one is found that is not running then:
1318 * look in mdadm.conf for bitmap file.
1319 * if one exists, but array has none, add it.
1320 * try to start array in auto-readonly mode
1321 */
1322 struct map_ent *mapl = NULL;
1323 struct map_ent *me;
1324 struct mddev_ident *devs, *mddev;
1325 int rv = 0;
1326
1327 map_read(&mapl);
1328 devs = conf_get_ident(NULL);
1329
1330 for (me = mapl ; me ; me = me->next) {
1331 mdu_array_info_t array;
1332 mdu_bitmap_file_t bmf;
1333 struct mdinfo *sra;
1334 int mdfd = open_dev(me->devnum);
1335
1336 if (mdfd < 0)
1337 continue;
1338 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1339 errno != ENODEV) {
1340 close(mdfd);
1341 continue;
1342 }
1343 /* Ok, we can try this one. Maybe it needs a bitmap */
1344 for (mddev = devs ; mddev ; mddev = mddev->next)
1345 if (mddev->devname && me->path
1346 && devname_matches(mddev->devname, me->path))
1347 break;
1348 if (mddev && mddev->bitmap_file) {
1349 /*
1350 * Note: early kernels will wrongly fail this, so it
1351 * is a hint only
1352 */
1353 int added = -1;
1354 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1355 int bmfd = open(mddev->bitmap_file, O_RDWR);
1356 if (bmfd >= 0) {
1357 added = ioctl(mdfd, SET_BITMAP_FILE,
1358 bmfd);
1359 close(bmfd);
1360 }
1361 }
1362 if (verbose >= 0) {
1363 if (added == 0)
1364 fprintf(stderr, Name
1365 ": Added bitmap %s to %s\n",
1366 mddev->bitmap_file, me->path);
1367 else if (errno != EEXIST)
1368 fprintf(stderr, Name
1369 ": Failed to add bitmap to %s: %s\n",
1370 me->path, strerror(errno));
1371 }
1372 }
1373 sra = sysfs_read(mdfd, 0, 0);
1374 if (sra) {
1375 if (sysfs_set_str(sra, NULL,
1376 "array_state", "read-auto") == 0) {
1377 if (verbose >= 0)
1378 fprintf(stderr, Name
1379 ": started array %s\n",
1380 me->path ?: devnum2devname(me->devnum));
1381 } else {
1382 fprintf(stderr, Name
1383 ": failed to start array %s: %s\n",
1384 me->path ?: devnum2devname(me->devnum),
1385 strerror(errno));
1386 rv = 1;
1387 }
1388 }
1389 }
1390 return rv;
1391 }
1392
1393 static char *container2devname(char *devname)
1394 {
1395 char *mdname = NULL;
1396
1397 if (devname[0] == '/') {
1398 int fd = open(devname, O_RDONLY);
1399 if (fd >= 0) {
1400 mdname = devnum2devname(fd2devnum(fd));
1401 close(fd);
1402 }
1403 } else {
1404 int uuid[4];
1405 struct map_ent *mp, *map = NULL;
1406
1407 if (!parse_uuid(devname, uuid))
1408 return mdname;
1409 mp = map_by_uuid(&map, uuid);
1410 if (mp)
1411 mdname = devnum2devname(mp->devnum);
1412 map_free(map);
1413 }
1414
1415 return mdname;
1416 }
1417
1418 static int Incremental_container(struct supertype *st, char *devname,
1419 char *homehost, int verbose,
1420 int runstop, int autof)
1421 {
1422 /* Collect the contents of this container and for each
1423 * array, choose a device name and assemble the array.
1424 */
1425
1426 struct mdinfo *list;
1427 struct mdinfo *ra;
1428 struct map_ent *map = NULL;
1429 struct mdinfo info;
1430 int trustworthy;
1431 struct mddev_ident *match;
1432 int rv = 0;
1433 struct domainlist *domains;
1434 struct map_ent *smp;
1435 int suuid[4];
1436 int sfd;
1437
1438 memset(&info, 0, sizeof(info));
1439 st->ss->getinfo_super(st, &info, NULL);
1440
1441 if ((runstop > 0 && info.container_enough >= 0) ||
1442 info.container_enough > 0)
1443 /* pass */;
1444 else {
1445 if (verbose)
1446 fprintf(stderr, Name ": not enough devices to start the container\n");
1447 return 0;
1448 }
1449
1450 match = search_mdstat(st, &info, devname, verbose, &rv);
1451 if (match == NULL && rv == 2)
1452 return rv;
1453
1454 /* Need to compute 'trustworthy' */
1455 if (match)
1456 trustworthy = LOCAL;
1457 else if (st->ss->match_home(st, homehost) == 1)
1458 trustworthy = LOCAL;
1459 else if (st->ss->match_home(st, "any") == 1)
1460 trustworthy = LOCAL;
1461 else
1462 trustworthy = FOREIGN;
1463
1464 list = st->ss->container_content(st, NULL);
1465 if (map_lock(&map))
1466 fprintf(stderr, Name ": failed to get exclusive lock on "
1467 "mapfile\n");
1468 /* do not assemble arrays that might have bad blocks */
1469 if (list->array.state & (1<<MD_SB_BBM_ERRORS)) {
1470 fprintf(stderr, Name ": BBM log found in metadata. "
1471 "Cannot activate array(s).\n");
1472 /* free container data and exit */
1473 sysfs_free(list);
1474 return 2;
1475 }
1476
1477 for (ra = list ; ra ; ra = ra->next) {
1478 int mdfd;
1479 char chosen_name[1024];
1480 struct map_ent *mp;
1481 struct mddev_ident *match = NULL;
1482
1483 mp = map_by_uuid(&map, ra->uuid);
1484
1485 if (mp) {
1486 mdfd = open_dev(mp->devnum);
1487 if (mp->path)
1488 strcpy(chosen_name, mp->path);
1489 else
1490 strcpy(chosen_name, devnum2devname(mp->devnum));
1491 } else {
1492
1493 /* Check in mdadm.conf for container == devname and
1494 * member == ra->text_version after second slash.
1495 */
1496 char *sub = strchr(ra->text_version+1, '/');
1497 struct mddev_ident *array_list;
1498 if (sub) {
1499 sub++;
1500 array_list = conf_get_ident(NULL);
1501 } else
1502 array_list = NULL;
1503 for(; array_list ; array_list = array_list->next) {
1504 char *dn;
1505 if (array_list->member == NULL ||
1506 array_list->container == NULL)
1507 continue;
1508 if (strcmp(array_list->member, sub) != 0)
1509 continue;
1510 if (array_list->uuid_set &&
1511 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1512 continue;
1513 dn = container2devname(array_list->container);
1514 if (dn == NULL)
1515 continue;
1516 if (strncmp(dn, ra->text_version+1,
1517 strlen(dn)) != 0 ||
1518 ra->text_version[strlen(dn)+1] != '/') {
1519 free(dn);
1520 continue;
1521 }
1522 free(dn);
1523 /* we have a match */
1524 match = array_list;
1525 if (verbose>0)
1526 fprintf(stderr, Name ": match found for member %s\n",
1527 array_list->member);
1528 break;
1529 }
1530
1531 if (match && match->devname &&
1532 strcasecmp(match->devname, "<ignore>") == 0) {
1533 if (verbose > 0)
1534 fprintf(stderr, Name ": array %s/%s is "
1535 "explicitly ignored by mdadm.conf\n",
1536 match->container, match->member);
1537 return 2;
1538 }
1539 if (match)
1540 trustworthy = LOCAL;
1541
1542 mdfd = create_mddev(match ? match->devname : NULL,
1543 ra->name,
1544 autof,
1545 trustworthy,
1546 chosen_name);
1547 }
1548
1549 if (mdfd < 0) {
1550 fprintf(stderr, Name ": failed to open %s: %s.\n",
1551 chosen_name, strerror(errno));
1552 return 2;
1553 }
1554
1555 assemble_container_content(st, mdfd, ra, runstop,
1556 chosen_name, verbose, NULL);
1557 close(mdfd);
1558 }
1559
1560 /* Now move all suitable spares from spare container */
1561 domains = domain_from_array(list, st->ss->name);
1562 memcpy(suuid, uuid_zero, sizeof(int[4]));
1563 if (domains &&
1564 (smp = map_by_uuid(&map, suuid)) != NULL &&
1565 (sfd = open(smp->path, O_RDONLY)) >= 0) {
1566 /* spare container found */
1567 struct supertype *sst =
1568 super_imsm.match_metadata_desc("imsm");
1569 struct mdinfo *sinfo;
1570 unsigned long long min_size = 0;
1571 if (st->ss->min_acceptable_spare_size)
1572 min_size = st->ss->min_acceptable_spare_size(st);
1573 if (!sst->ss->load_container(sst, sfd, NULL)) {
1574 close(sfd);
1575 sinfo = container_choose_spares(sst, min_size,
1576 domains, NULL,
1577 st->ss->name, 0);
1578 sst->ss->free_super(sst);
1579 if (sinfo){
1580 int count = 0;
1581 struct mdinfo *disks = sinfo->devs;
1582 while (disks) {
1583 /* move spare from spare
1584 * container to currently
1585 * assembled one
1586 */
1587 if (move_spare(
1588 smp->path,
1589 devname,
1590 makedev(disks->disk.major,
1591 disks->disk.minor)))
1592 count++;
1593 disks = disks->next;
1594 }
1595 if (count)
1596 fprintf(stderr, Name
1597 ": Added %d spare%s to %s\n",
1598 count, count>1?"s":"", devname);
1599 }
1600 sysfs_free(sinfo);
1601 } else
1602 close(sfd);
1603 }
1604 domain_free(domains);
1605 map_unlock(&map);
1606 return 0;
1607 }
1608
1609 /*
1610 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1611 * raid arrays, and if so first fail (if needed) and then remove the device.
1612 *
1613 * @devname - The device we want to remove
1614 * @id_path - name as found in /dev/disk/by-path for this device
1615 *
1616 * Note: the device name must be a kernel name like "sda", so
1617 * that we can find it in /proc/mdstat
1618 */
1619 int IncrementalRemove(char *devname, char *id_path, int verbose)
1620 {
1621 int mdfd;
1622 int rv;
1623 struct mdstat_ent *ent;
1624 struct mddev_dev devlist;
1625
1626 if (!id_path)
1627 dprintf(Name ": incremental removal without --path <id_path> "
1628 "lacks the possibility to re-add new device in this "
1629 "port\n");
1630
1631 if (strchr(devname, '/')) {
1632 fprintf(stderr, Name ": incremental removal requires a "
1633 "kernel device name, not a file: %s\n", devname);
1634 return 1;
1635 }
1636 ent = mdstat_by_component(devname);
1637 if (!ent) {
1638 fprintf(stderr, Name ": %s does not appear to be a component "
1639 "of any array\n", devname);
1640 return 1;
1641 }
1642 mdfd = open_dev(ent->devnum);
1643 if (mdfd < 0) {
1644 fprintf(stderr, Name ": Cannot open array %s!!\n", ent->dev);
1645 free_mdstat(ent);
1646 return 1;
1647 }
1648
1649 if (id_path) {
1650 struct map_ent *map = NULL, *me;
1651 me = map_by_devnum(&map, ent->devnum);
1652 if (me)
1653 policy_save_path(id_path, me);
1654 map_free(map);
1655 }
1656
1657 memset(&devlist, 0, sizeof(devlist));
1658 devlist.devname = devname;
1659 devlist.disposition = 'f';
1660 /* for a container, we must fail each member array */
1661 if (ent->metadata_version &&
1662 strncmp(ent->metadata_version, "external:", 9) == 0) {
1663 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1664 struct mdstat_ent *memb;
1665 for (memb = mdstat ; memb ; memb = memb->next)
1666 if (is_container_member(memb, ent->dev)) {
1667 int subfd = open_dev(memb->devnum);
1668 if (subfd >= 0) {
1669 Manage_subdevs(memb->dev, subfd,
1670 &devlist, verbose, 0,
1671 NULL);
1672 close(subfd);
1673 }
1674 }
1675 free_mdstat(mdstat);
1676 } else
1677 Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL);
1678 devlist.disposition = 'r';
1679 rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL);
1680 close(mdfd);
1681 free_mdstat(ent);
1682 return rv;
1683 }
Unnamed repository; edit this file 'description' to name the repository.