]> 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
Do not continue reshape during initrd phase
[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 int freeze_reshape);
49
50 static struct mddev_ident *search_mdstat(struct supertype *st,
51 struct mdinfo *info,
52 char *devname,
53 int verbose, int *rvp);
54
55 int Incremental(char *devname, int verbose, int runstop,
56 struct supertype *st, char *homehost, int require_homehost,
57 int autof, int freeze_reshape)
58 {
59 /* Add this device to an array, creating the array if necessary
60 * and starting the array if sensible or - if runstop>0 - if possible.
61 *
62 * This has several steps:
63 *
64 * 1/ Check if device is permitted by mdadm.conf, reject if not.
65 * 2/ Find metadata, reject if none appropriate (check
66 * version/name from args)
67 * 3/ Check if there is a match in mdadm.conf
68 * 3a/ if not, check for homehost match. If no match, assemble as
69 * a 'foreign' array.
70 * 4/ Determine device number.
71 * - If in mdadm.conf with std name, use that
72 * - UUID in /var/run/mdadm.map use that
73 * - If name is suggestive, use that. unless in use with different uuid.
74 * - Choose a free, high number.
75 * - Use a partitioned device unless strong suggestion not to.
76 * e.g. auto=md
77 * Don't choose partitioned for containers.
78 * 5/ Find out if array already exists
79 * 5a/ if it does not
80 * - choose a name, from mdadm.conf or 'name' field in array.
81 * - create the array
82 * - add the device
83 * 5b/ if it does
84 * - check one drive in array to make sure metadata is a reasonably
85 * close match. Reject if not (e.g. different type)
86 * - add the device
87 * 6/ Make sure /var/run/mdadm.map contains this array.
88 * 7/ Is there enough devices to possibly start the array?
89 * For a container, this means running Incremental_container.
90 * 7a/ if not, finish with success.
91 * 7b/ if yes,
92 * - read all metadata and arrange devices like -A does
93 * - if number of OK devices match expected, or -R and there are enough,
94 * start the array (auto-readonly).
95 */
96 struct stat stb;
97 struct mdinfo info, dinfo;
98 struct mdinfo *sra = NULL, *d;
99 struct mddev_ident *match;
100 char chosen_name[1024];
101 int rv = 1;
102 struct map_ent *mp, *map = NULL;
103 int dfd = -1, mdfd = -1;
104 char *avail = NULL;
105 int active_disks;
106 int trustworthy;
107 char *name_to_use;
108 mdu_array_info_t ainf;
109 struct dev_policy *policy = NULL;
110 struct map_ent target_array;
111 int have_target;
112
113 struct createinfo *ci = conf_get_create_info();
114
115 if (stat(devname, &stb) < 0) {
116 if (verbose >= 0)
117 fprintf(stderr, Name ": stat failed for %s: %s.\n",
118 devname, strerror(errno));
119 return rv;
120 }
121 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
122 if (verbose >= 0)
123 fprintf(stderr, Name ": %s is not a block device.\n",
124 devname);
125 return rv;
126 }
127 dfd = dev_open(devname, O_RDONLY|O_EXCL);
128 if (dfd < 0) {
129 if (verbose >= 0)
130 fprintf(stderr, Name ": cannot open %s: %s.\n",
131 devname, strerror(errno));
132 return rv;
133 }
134 /* If the device is a container, we do something very different */
135 if (must_be_container(dfd)) {
136 if (!st)
137 st = super_by_fd(dfd, NULL);
138 if (st && st->ss->load_container)
139 rv = st->ss->load_container(st, dfd, NULL);
140
141 close(dfd);
142 if (!rv && st->ss->container_content)
143 return Incremental_container(st, devname, homehost,
144 verbose, runstop, autof,
145 freeze_reshape);
146
147 fprintf(stderr, Name ": %s is not part of an md array.\n",
148 devname);
149 return rv;
150 }
151
152 /* 1/ Check if device is permitted by mdadm.conf */
153
154 if (!conf_test_dev(devname)) {
155 if (verbose >= 0)
156 fprintf(stderr, Name
157 ": %s not permitted by mdadm.conf.\n",
158 devname);
159 goto out;
160 }
161
162 /* 2/ Find metadata, reject if none appropriate (check
163 * version/name from args) */
164
165 if (fstat(dfd, &stb) < 0) {
166 if (verbose >= 0)
167 fprintf(stderr, Name ": fstat failed for %s: %s.\n",
168 devname, strerror(errno));
169 goto out;
170 }
171 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
172 if (verbose >= 0)
173 fprintf(stderr, Name ": %s is not a block device.\n",
174 devname);
175 goto out;
176 }
177
178 dinfo.disk.major = major(stb.st_rdev);
179 dinfo.disk.minor = minor(stb.st_rdev);
180
181 policy = disk_policy(&dinfo);
182 have_target = policy_check_path(&dinfo, &target_array);
183
184 if (st == NULL && (st = guess_super(dfd)) == NULL) {
185 if (verbose >= 0)
186 fprintf(stderr, Name
187 ": no recognisable superblock on %s.\n",
188 devname);
189 rv = try_spare(devname, &dfd, policy,
190 have_target ? &target_array : NULL,
191 st, verbose);
192 goto out;
193 }
194 if (st->ss->compare_super == NULL ||
195 st->ss->load_super(st, dfd, NULL)) {
196 if (verbose >= 0)
197 fprintf(stderr, Name ": no RAID superblock on %s.\n",
198 devname);
199 rv = try_spare(devname, &dfd, policy,
200 have_target ? &target_array : NULL,
201 st, verbose);
202 free(st);
203 goto out;
204 }
205 close (dfd); dfd = -1;
206
207 st->ss->getinfo_super(st, &info, NULL);
208
209 /* 3/ Check if there is a match in mdadm.conf */
210 match = search_mdstat(st, &info, devname, verbose, &rv);
211 if (!match && rv == 2)
212 goto out;
213
214 if (match && match->devname
215 && strcasecmp(match->devname, "<ignore>") == 0) {
216 if (verbose >= 0)
217 fprintf(stderr, Name ": array containing %s is explicitly"
218 " ignored by mdadm.conf\n",
219 devname);
220 goto out;
221 }
222
223 /* 3a/ if not, check for homehost match. If no match, continue
224 * but don't trust the 'name' in the array. Thus a 'random' minor
225 * number will be assigned, and the device name will be based
226 * on that. */
227 if (match)
228 trustworthy = LOCAL;
229 else if (st->ss->match_home(st, homehost) == 1)
230 trustworthy = LOCAL;
231 else if (st->ss->match_home(st, "any") == 1)
232 trustworthy = LOCAL_ANY;
233 else
234 trustworthy = FOREIGN;
235
236
237 if (!match && !conf_test_metadata(st->ss->name, policy,
238 (trustworthy == LOCAL))) {
239 if (verbose >= 1)
240 fprintf(stderr, Name
241 ": %s has metadata type %s for which "
242 "auto-assembly is disabled\n",
243 devname, st->ss->name);
244 goto out;
245 }
246 if (trustworthy == LOCAL_ANY)
247 trustworthy = LOCAL;
248
249 /* There are three possible sources for 'autof': command line,
250 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
251 * ARRAY takes precedence, then command line, then
252 * CREATE.
253 */
254 if (match && match->autof)
255 autof = match->autof;
256 if (autof == 0)
257 autof = ci->autof;
258
259 name_to_use = info.name;
260 if (name_to_use[0] == 0 &&
261 info.array.level == LEVEL_CONTAINER &&
262 trustworthy == LOCAL) {
263 name_to_use = info.text_version;
264 trustworthy = METADATA;
265 }
266 if (name_to_use[0] && trustworthy != LOCAL &&
267 ! require_homehost &&
268 conf_name_is_free(name_to_use))
269 trustworthy = LOCAL;
270
271 /* strip "hostname:" prefix from name if we have decided
272 * to treat it as LOCAL
273 */
274 if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
275 name_to_use = strchr(name_to_use, ':')+1;
276
277 /* 4/ Check if array exists.
278 */
279 if (map_lock(&map))
280 fprintf(stderr, Name ": failed to get exclusive lock on "
281 "mapfile\n");
282 mp = map_by_uuid(&map, info.uuid);
283 if (mp)
284 mdfd = open_dev(mp->devnum);
285 else
286 mdfd = -1;
287
288 if (mdfd < 0) {
289
290 /* Couldn't find an existing array, maybe make a new one */
291 mdfd = create_mddev(match ? match->devname : NULL,
292 name_to_use, autof, trustworthy, chosen_name);
293
294 if (mdfd < 0)
295 goto out;
296
297 sysfs_init(&info, mdfd, 0);
298
299 if (set_array_info(mdfd, st, &info) != 0) {
300 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
301 chosen_name, strerror(errno));
302 rv = 2;
303 goto out;
304 }
305
306 dinfo = info;
307 dinfo.disk.major = major(stb.st_rdev);
308 dinfo.disk.minor = minor(stb.st_rdev);
309 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
310 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
311 devname, chosen_name, strerror(errno));
312 ioctl(mdfd, STOP_ARRAY, 0);
313 rv = 2;
314 goto out;
315 }
316 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
317 GET_OFFSET | GET_SIZE));
318
319 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
320 /* It really should be 'none' - must be old buggy
321 * kernel, and mdadm -I may not be able to complete.
322 * So reject it.
323 */
324 ioctl(mdfd, STOP_ARRAY, NULL);
325 fprintf(stderr, Name
326 ": You have an old buggy kernel which cannot support\n"
327 " --incremental reliably. Aborting.\n");
328 sysfs_free(sra);
329 rv = 2;
330 goto out;
331 }
332 info.array.working_disks = 1;
333 /* 6/ Make sure /var/run/mdadm.map contains this array. */
334 map_update(&map, fd2devnum(mdfd),
335 info.text_version,
336 info.uuid, chosen_name);
337 } else {
338 /* 5b/ if it does */
339 /* - check one drive in array to make sure metadata is a reasonably */
340 /* close match. Reject if not (e.g. different type) */
341 /* - add the device */
342 char dn[20];
343 int dfd2;
344 int err;
345 struct supertype *st2;
346 struct mdinfo info2, *d;
347
348 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
349 GET_OFFSET | GET_SIZE));
350
351 if (mp->path)
352 strcpy(chosen_name, mp->path);
353 else
354 strcpy(chosen_name, devnum2devname(mp->devnum));
355
356 /* It is generally not OK to add non-spare drives to a
357 * running array as they are probably missing because
358 * they failed. However if runstop is 1, then the
359 * array was possibly started early and our best bet is
360 * to add this anyway.
361 * Also if action policy is re-add or better we allow
362 * re-add.
363 * This doesn't apply to containers as the 'non-spare'
364 * flag has a different meaning. The test has to happen
365 * at the device level there
366 */
367 if (!st->ss->external
368 && (info.disk.state & (1<<MD_DISK_SYNC)) != 0
369 && ! policy_action_allows(policy, st->ss->name,
370 act_re_add)
371 && runstop < 1) {
372 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
373 fprintf(stderr, Name
374 ": not adding %s to active array (without --run) %s\n",
375 devname, chosen_name);
376 rv = 2;
377 goto out;
378 }
379 }
380 if (!sra) {
381 rv = 2;
382 goto out;
383 }
384 if (sra->devs) {
385 sprintf(dn, "%d:%d", sra->devs->disk.major,
386 sra->devs->disk.minor);
387 dfd2 = dev_open(dn, O_RDONLY);
388 st2 = dup_super(st);
389 if (st2->ss->load_super(st2, dfd2, NULL) ||
390 st->ss->compare_super(st, st2) != 0) {
391 fprintf(stderr, Name
392 ": metadata mismatch between %s and "
393 "chosen array %s\n",
394 devname, chosen_name);
395 close(dfd2);
396 rv = 2;
397 goto out;
398 }
399 close(dfd2);
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 freeze_reshape);
457 if (rv == 1)
458 /* Don't fail the whole -I if a subarray didn't
459 * have enough devices to start yet
460 */
461 rv = 0;
462 /* after spare is added, ping monitor for external metadata
463 * so that it can eg. try to rebuild degraded array */
464 if (st->ss->external)
465 ping_monitor_by_id(devnum);
466 return rv;
467 }
468
469 /* We have added something to the array, so need to re-read the
470 * state. Eventually this state should be kept up-to-date as
471 * things change.
472 */
473 sysfs_free(sra);
474 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
475 GET_OFFSET | GET_SIZE));
476 active_disks = count_active(st, sra, mdfd, &avail, &info);
477 if (enough(info.array.level, info.array.raid_disks,
478 info.array.layout, info.array.state & 1,
479 avail, active_disks) == 0) {
480 if (verbose >= 0)
481 fprintf(stderr, Name
482 ": %s attached to %s, not enough to start (%d).\n",
483 devname, chosen_name, active_disks);
484 map_unlock(&map);
485 rv = 0;
486 goto out;
487 }
488
489 /* 7b/ if yes, */
490 /* - if number of OK devices match expected, or -R and there */
491 /* are enough, */
492 /* + add any bitmap file */
493 /* + start the array (auto-readonly). */
494
495 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
496 if (verbose >= 0)
497 fprintf(stderr, Name
498 ": %s attached to %s which is already active.\n",
499 devname, chosen_name);
500 map_unlock(&map);
501 rv = 0;
502 goto out;
503 }
504
505 map_unlock(&map);
506 if (runstop > 0 || active_disks >= info.array.working_disks) {
507 struct mdinfo *dsk;
508 /* Let's try to start it */
509 if (match && match->bitmap_file) {
510 int bmfd = open(match->bitmap_file, O_RDWR);
511 if (bmfd < 0) {
512 fprintf(stderr, Name
513 ": Could not open bitmap file %s.\n",
514 match->bitmap_file);
515 goto out;
516 }
517 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
518 close(bmfd);
519 fprintf(stderr, Name
520 ": Failed to set bitmapfile for %s.\n",
521 chosen_name);
522 goto out;
523 }
524 close(bmfd);
525 }
526 /* Need to remove from the array any devices which
527 * 'count_active' discerned were too old or inappropriate
528 */
529 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
530 if (d->disk.state & (1<<MD_DISK_REMOVED))
531 remove_disk(mdfd, st, sra, d);
532
533 if ((sra == NULL || active_disks >= info.array.working_disks)
534 && trustworthy != FOREIGN)
535 rv = ioctl(mdfd, RUN_ARRAY, NULL);
536 else
537 rv = sysfs_set_str(sra, NULL,
538 "array_state", "read-auto");
539 if (rv == 0) {
540 if (verbose >= 0)
541 fprintf(stderr, Name
542 ": %s attached to %s, which has been started.\n",
543 devname, chosen_name);
544 rv = 0;
545 wait_for(chosen_name, mdfd);
546 /* We just started the array, so some devices
547 * might have been evicted from the array
548 * because their event counts were too old.
549 * If the action=re-add policy is in-force for
550 * those devices we should re-add them now.
551 */
552 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
553 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
554 add_disk(mdfd, st, sra, dsk) == 0)
555 fprintf(stderr, Name
556 ": %s re-added to %s\n",
557 dsk->sys_name, chosen_name);
558 }
559 } else {
560 fprintf(stderr, Name
561 ": %s attached to %s, but failed to start: %s.\n",
562 devname, chosen_name, strerror(errno));
563 rv = 1;
564 }
565 } else {
566 if (verbose >= 0)
567 fprintf(stderr, Name
568 ": %s attached to %s, not enough to start safely.\n",
569 devname, chosen_name);
570 rv = 0;
571 }
572 out:
573 free(avail);
574 if (dfd >= 0)
575 close(dfd);
576 if (mdfd >= 0)
577 close(mdfd);
578 if (policy)
579 dev_policy_free(policy);
580 if (sra)
581 sysfs_free(sra);
582 return rv;
583 }
584
585 static struct mddev_ident *search_mdstat(struct supertype *st,
586 struct mdinfo *info,
587 char *devname,
588 int verbose, int *rvp)
589 {
590 struct mddev_ident *array_list, *match;
591 array_list = conf_get_ident(NULL);
592 match = NULL;
593 for (; array_list; array_list = array_list->next) {
594 if (array_list->uuid_set &&
595 same_uuid(array_list->uuid, info->uuid, st->ss->swapuuid)
596 == 0) {
597 if (verbose >= 2 && array_list->devname)
598 fprintf(stderr, Name
599 ": UUID differs from %s.\n",
600 array_list->devname);
601 continue;
602 }
603 if (array_list->name[0] &&
604 strcasecmp(array_list->name, info->name) != 0) {
605 if (verbose >= 2 && array_list->devname)
606 fprintf(stderr, Name
607 ": Name differs from %s.\n",
608 array_list->devname);
609 continue;
610 }
611 if (array_list->devices &&
612 !match_oneof(array_list->devices, devname)) {
613 if (verbose >= 2 && array_list->devname)
614 fprintf(stderr, Name
615 ": Not a listed device for %s.\n",
616 array_list->devname);
617 continue;
618 }
619 if (array_list->super_minor != UnSet &&
620 array_list->super_minor != info->array.md_minor) {
621 if (verbose >= 2 && array_list->devname)
622 fprintf(stderr, Name
623 ": Different super-minor to %s.\n",
624 array_list->devname);
625 continue;
626 }
627 if (!array_list->uuid_set &&
628 !array_list->name[0] &&
629 !array_list->devices &&
630 array_list->super_minor == UnSet) {
631 if (verbose >= 2 && array_list->devname)
632 fprintf(stderr, Name
633 ": %s doesn't have any identifying information.\n",
634 array_list->devname);
635 continue;
636 }
637 /* FIXME, should I check raid_disks and level too?? */
638
639 if (match) {
640 if (verbose >= 0) {
641 if (match->devname && array_list->devname)
642 fprintf(stderr, Name
643 ": we match both %s and %s - cannot decide which to use.\n",
644 match->devname, array_list->devname);
645 else
646 fprintf(stderr, Name
647 ": multiple lines in mdadm.conf match\n");
648 }
649 *rvp = 2;
650 match = NULL;
651 break;
652 }
653 match = array_list;
654 }
655 return match;
656 }
657
658 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
659 int number, __u64 events, int verbose,
660 char *array_name)
661 {
662 /* Find a device attached to this array with a disk.number of number
663 * and events less than the passed events, and remove the device.
664 */
665 struct mdinfo *d;
666 mdu_array_info_t ra;
667
668 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
669 return; /* not safe to remove from active arrays
670 * without thinking more */
671
672 for (d = sra->devs; d ; d = d->next) {
673 char dn[10];
674 int dfd;
675 struct mdinfo info;
676 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
677 dfd = dev_open(dn, O_RDONLY);
678 if (dfd < 0)
679 continue;
680 if (st->ss->load_super(st, dfd, NULL)) {
681 close(dfd);
682 continue;
683 }
684 st->ss->getinfo_super(st, &info, NULL);
685 st->ss->free_super(st);
686 close(dfd);
687
688 if (info.disk.number != number ||
689 info.events >= events)
690 continue;
691
692 if (d->disk.raid_disk > -1)
693 sysfs_set_str(sra, d, "slot", "none");
694 if (sysfs_set_str(sra, d, "state", "remove") == 0)
695 if (verbose >= 0)
696 fprintf(stderr, Name
697 ": removing old device %s from %s\n",
698 d->sys_name+4, array_name);
699 }
700 }
701
702 static int count_active(struct supertype *st, struct mdinfo *sra,
703 int mdfd, char **availp,
704 struct mdinfo *bestinfo)
705 {
706 /* count how many devices in sra think they are active */
707 struct mdinfo *d;
708 int cnt = 0;
709 __u64 max_events = 0;
710 char *avail = NULL;
711 int *best = NULL;
712 char *devmap = NULL;
713 int numdevs = 0;
714 int devnum;
715 int b, i;
716 int raid_disks = 0;
717
718 if (!sra)
719 return 0;
720
721 for (d = sra->devs ; d ; d = d->next)
722 numdevs++;
723 for (d = sra->devs, devnum=0 ; d ; d = d->next, devnum++) {
724 char dn[30];
725 int dfd;
726 int ok;
727 struct mdinfo info;
728
729 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
730 dfd = dev_open(dn, O_RDONLY);
731 if (dfd < 0)
732 continue;
733 ok = st->ss->load_super(st, dfd, NULL);
734 close(dfd);
735 if (ok != 0)
736 continue;
737 info.array.raid_disks = raid_disks;
738 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
739 if (!avail) {
740 raid_disks = info.array.raid_disks;
741 avail = calloc(raid_disks, 1);
742 if (!avail) {
743 fprintf(stderr, Name ": out of memory.\n");
744 exit(1);
745 }
746 *availp = avail;
747
748 best = calloc(raid_disks, sizeof(int));
749 devmap = calloc(raid_disks * numdevs, 1);
750
751 st->ss->getinfo_super(st, &info, devmap);
752 }
753
754 if (info.disk.state & (1<<MD_DISK_SYNC))
755 {
756 if (cnt == 0) {
757 cnt++;
758 max_events = info.events;
759 avail[info.disk.raid_disk] = 2;
760 best[info.disk.raid_disk] = devnum;
761 st->ss->getinfo_super(st, bestinfo, NULL);
762 } else if (info.events == max_events) {
763 avail[info.disk.raid_disk] = 2;
764 best[info.disk.raid_disk] = devnum;
765 } else if (info.events == max_events-1) {
766 if (avail[info.disk.raid_disk] == 0) {
767 avail[info.disk.raid_disk] = 1;
768 best[info.disk.raid_disk] = devnum;
769 }
770 } else if (info.events < max_events - 1)
771 ;
772 else if (info.events == max_events+1) {
773 int i;
774 max_events = info.events;
775 for (i=0; i < raid_disks; i++)
776 if (avail[i])
777 avail[i]--;
778 avail[info.disk.raid_disk] = 2;
779 best[info.disk.raid_disk] = devnum;
780 st->ss->getinfo_super(st, bestinfo, NULL);
781 } else { /* info.events much bigger */
782 memset(avail, 0, raid_disks);
783 max_events = info.events;
784 avail[info.disk.raid_disk] = 2;
785 best[info.disk.raid_disk] = devnum;
786 st->ss->getinfo_super(st, bestinfo, NULL);
787 }
788 }
789 st->ss->free_super(st);
790 }
791 if (!avail)
792 return 0;
793 /* We need to reject any device that thinks the best device is
794 * failed or missing */
795 for (b = 0; b < raid_disks; b++)
796 if (avail[b] == 2)
797 break;
798 cnt = 0;
799 for (i = 0 ; i < raid_disks ; i++) {
800 if (i != b && avail[i])
801 if (devmap[raid_disks * best[i] + b] == 0) {
802 /* This device thinks 'b' is failed -
803 * don't use it */
804 devnum = best[i];
805 for (d=sra->devs ; devnum; d = d->next)
806 devnum--;
807 d->disk.state |= (1 << MD_DISK_REMOVED);
808 avail[i] = 0;
809 }
810 if (avail[i])
811 cnt++;
812 }
813 free(best);
814 free(devmap);
815 return cnt;
816 }
817
818 /* test if container has degraded member(s) */
819 static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
820 {
821 mdu_array_info_t array;
822 int afd;
823 int max_degraded = 0;
824
825 for(; map; map = map->next) {
826 if (!is_subarray(map->metadata) ||
827 devname2devnum(map->metadata+1) != me->devnum)
828 continue;
829 afd = open_dev(map->devnum);
830 if (afd < 0)
831 continue;
832 /* most accurate information regarding array degradation */
833 if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) {
834 int degraded = array.raid_disks - array.active_disks -
835 array.spare_disks;
836 if (degraded > max_degraded)
837 max_degraded = degraded;
838 }
839 close(afd);
840 }
841 return (max_degraded);
842 }
843
844 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
845 struct map_ent *target, int bare,
846 struct supertype *st, int verbose)
847 {
848 /* This device doesn't have any md metadata
849 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
850 * If 'st' is not set, then we only know that some metadata allows this,
851 * others possibly don't.
852 * So look for a container or array to attach the device to.
853 * Prefer 'target' if that is set and the array is found.
854 *
855 * If st is set, then only arrays of that type are considered
856 * Return 0 on success, or some exit code on failure, probably 1.
857 */
858 int rv = 1;
859 struct stat stb;
860 struct map_ent *mp, *map = NULL;
861 struct mdinfo *chosen = NULL;
862 int dfd = *dfdp;
863
864 if (fstat(dfd, &stb) != 0)
865 return 1;
866
867 /*
868 * Now we need to find a suitable array to add this to.
869 * We only accept arrays that:
870 * - match 'st'
871 * - are in the same domains as the device
872 * - are of an size for which the device will be useful
873 * and we choose the one that is the most degraded
874 */
875
876 if (map_lock(&map)) {
877 fprintf(stderr, Name ": failed to get exclusive lock on "
878 "mapfile\n");
879 return 1;
880 }
881 for (mp = map ; mp ; mp = mp->next) {
882 struct supertype *st2;
883 struct domainlist *dl = NULL;
884 struct mdinfo *sra;
885 unsigned long long devsize;
886 unsigned long long component_size = 0;
887
888 if (is_subarray(mp->metadata))
889 continue;
890 if (st) {
891 st2 = st->ss->match_metadata_desc(mp->metadata);
892 if (!st2 ||
893 (st->minor_version >= 0 &&
894 st->minor_version != st2->minor_version)) {
895 if (verbose > 1)
896 fprintf(stderr, Name ": not adding %s to %s as metadata type doesn't match\n",
897 devname, mp->path);
898 free(st2);
899 continue;
900 }
901 free(st2);
902 }
903 sra = sysfs_read(-1, mp->devnum,
904 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
905 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
906 if (!sra) {
907 /* Probably a container - no degraded info */
908 sra = sysfs_read(-1, mp->devnum,
909 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
910 GET_COMPONENT|GET_VERSION);
911 if (sra)
912 sra->array.failed_disks = -1;
913 }
914 if (!sra)
915 continue;
916 if (st == NULL) {
917 int i;
918 st2 = NULL;
919 for(i=0; !st2 && superlist[i]; i++)
920 st2 = superlist[i]->match_metadata_desc(
921 sra->text_version);
922 if (!st2) {
923 if (verbose > 1)
924 fprintf(stderr, Name ": not adding %s to %s"
925 " as metadata not recognised.\n",
926 devname, mp->path);
927 goto next;
928 }
929 /* Need to double check the 'act_spare' permissions applies
930 * to this metadata.
931 */
932 if (!policy_action_allows(pol, st2->ss->name, act_spare))
933 goto next;
934 if (!bare && !policy_action_allows(pol, st2->ss->name,
935 act_spare_same_slot))
936 goto next;
937 } else
938 st2 = st;
939 /* update number of failed disks for mostly degraded
940 * container member */
941 if (sra->array.failed_disks == -1)
942 sra->array.failed_disks = container_members_max_degradation(map, mp);
943
944 get_dev_size(dfd, NULL, &devsize);
945 if (sra->component_size == 0) {
946 /* true for containers, here we must read superblock
947 * to obtain minimum spare size */
948 struct supertype *st3 = dup_super(st2);
949 int mdfd = open_dev(mp->devnum);
950 if (!mdfd)
951 goto next;
952 if (st3->ss->load_container &&
953 !st3->ss->load_container(st3, mdfd, mp->path)) {
954 component_size = st3->ss->min_acceptable_spare_size(st3);
955 st3->ss->free_super(st3);
956 }
957 free(st3);
958 close(mdfd);
959 }
960 if ((sra->component_size > 0 &&
961 st2->ss->avail_size(st2, devsize) < sra->component_size)
962 ||
963 (sra->component_size == 0 && devsize < component_size)) {
964 if (verbose > 1)
965 fprintf(stderr, Name ": not adding %s to %s as it is too small\n",
966 devname, mp->path);
967 goto next;
968 }
969 /* test against target.
970 * If 'target' is set and 'bare' is false, we only accept
971 * arrays/containers that match 'target'.
972 * If 'target' is set and 'bare' is true, we prefer the
973 * array which matches 'target'.
974 * target is considered only if we deal with degraded array
975 */
976 if (target && policy_action_allows(pol, st2->ss->name,
977 act_spare_same_slot)) {
978 if (strcmp(target->metadata, mp->metadata) == 0 &&
979 memcmp(target->uuid, mp->uuid,
980 sizeof(target->uuid)) == 0 &&
981 sra->array.failed_disks > 0) {
982 /* This is our target!! */
983 if (chosen)
984 sysfs_free(chosen);
985 chosen = sra;
986 sra = NULL;
987 /* skip to end so we don't check any more */
988 while (mp->next)
989 mp = mp->next;
990 goto next;
991 }
992 /* not our target */
993 if (!bare)
994 goto next;
995 }
996
997 dl = domain_from_array(sra, st2->ss->name);
998 if (domain_test(dl, pol, st2->ss->name) != 1) {
999 /* domain test fails */
1000 if (verbose > 1)
1001 fprintf(stderr, Name ": not adding %s to %s as"
1002 " it is not in a compatible domain\n",
1003 devname, mp->path);
1004
1005 goto next;
1006 }
1007 /* all tests passed, OK to add to this array */
1008 if (!chosen) {
1009 chosen = sra;
1010 sra = NULL;
1011 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
1012 sysfs_free(chosen);
1013 chosen = sra;
1014 sra = NULL;
1015 }
1016 next:
1017 if (sra)
1018 sysfs_free(sra);
1019 if (st != st2)
1020 free(st2);
1021 if (dl)
1022 domain_free(dl);
1023 }
1024 if (chosen) {
1025 /* add current device to chosen array as a spare */
1026 int mdfd = open_dev(devname2devnum(chosen->sys_name));
1027 if (mdfd >= 0) {
1028 struct mddev_dev devlist;
1029 char devname[20];
1030 devlist.next = NULL;
1031 devlist.used = 0;
1032 devlist.re_add = 0;
1033 devlist.writemostly = 0;
1034 devlist.devname = devname;
1035 sprintf(devname, "%d:%d", major(stb.st_rdev),
1036 minor(stb.st_rdev));
1037 devlist.disposition = 'a';
1038 close(dfd);
1039 *dfdp = -1;
1040 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1041 -1, 0, NULL, 0);
1042 close(mdfd);
1043 }
1044 if (verbose > 0) {
1045 if (rv == 0)
1046 fprintf(stderr, Name ": added %s as spare for %s\n",
1047 devname, chosen->sys_name);
1048 else
1049 fprintf(stderr, Name ": failed to add %s as spare for %s\n",
1050 devname, chosen->sys_name);
1051 }
1052 sysfs_free(chosen);
1053 }
1054 return rv;
1055 }
1056
1057 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1058 struct supertype *st, int verbose)
1059 {
1060 /* we know that at least one partition virtual-metadata is
1061 * allowed to incorporate spares like this device. We need to
1062 * find a suitable device to copy partition information from.
1063 *
1064 * Getting a list of all disk (not partition) devices is
1065 * slightly non-trivial. We could look at /sys/block, but
1066 * that is theoretically due to be removed. Maybe best to use
1067 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1068 * we depend on this directory of 'path' info. But that fails
1069 * to find loop devices and probably others. Maybe don't
1070 * worry about that, they aren't the real target.
1071 *
1072 * So: check things in /dev/disk/by-path to see if they are in
1073 * a compatible domain, then load the partition table and see
1074 * if it is OK for the new device, and choose the largest
1075 * partition table that fits.
1076 */
1077 DIR *dir;
1078 struct dirent *de;
1079 char *chosen = NULL;
1080 unsigned long long chosen_size = 0;
1081 struct supertype *chosen_st = NULL;
1082 int fd;
1083
1084 dir = opendir("/dev/disk/by-path");
1085 if (!dir)
1086 return 1;
1087 while ((de = readdir(dir)) != NULL) {
1088 char *ep;
1089 struct dev_policy *pol2 = NULL;
1090 struct domainlist *domlist = NULL;
1091 int fd = -1;
1092 struct mdinfo info;
1093 struct supertype *st2 = NULL;
1094 char *devname = NULL;
1095 unsigned long long devsectors;
1096
1097 if (de->d_ino == 0 ||
1098 de->d_name[0] == '.' ||
1099 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1100 goto next;
1101
1102 ep = de->d_name + strlen(de->d_name);
1103 while (ep > de->d_name &&
1104 isdigit(ep[-1]))
1105 ep--;
1106 if (ep > de->d_name + 5 &&
1107 strncmp(ep-5, "-part", 5) == 0)
1108 /* This is a partition - skip it */
1109 goto next;
1110
1111 pol2 = path_policy(de->d_name, type_disk);
1112
1113 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1114 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1115 /* new device is incompatible with this device. */
1116 goto next;
1117
1118 domain_free(domlist);
1119 domlist = NULL;
1120
1121 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1122 devname = NULL;
1123 goto next;
1124 }
1125 fd = open(devname, O_RDONLY);
1126 if (fd < 0)
1127 goto next;
1128 if (get_dev_size(fd, devname, &devsectors) == 0)
1129 goto next;
1130 devsectors >>= 9;
1131
1132 if (st)
1133 st2 = dup_super(st);
1134 else
1135 st2 = guess_super_type(fd, guess_partitions);
1136 if (st2 == NULL ||
1137 st2->ss->load_super(st2, fd, NULL) < 0)
1138 goto next;
1139
1140 if (!st) {
1141 /* Check domain policy again, this time referring to metadata */
1142 domain_merge(&domlist, pol2, st2->ss->name);
1143 if (domain_test(domlist, pol, st2->ss->name) != 1)
1144 /* Incompatible devices for this metadata type */
1145 goto next;
1146 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1147 /* Some partition types allow sparing, but not
1148 * this one.
1149 */
1150 goto next;
1151 }
1152
1153 st2->ss->getinfo_super(st2, &info, NULL);
1154 if (info.component_size > devsectors)
1155 /* This partitioning doesn't fit in the device */
1156 goto next;
1157
1158 /* This is an acceptable device to copy partition
1159 * metadata from. We could just stop here, but I
1160 * think I want to keep looking incase a larger
1161 * metadata which makes better use of the device can
1162 * be found.
1163 */
1164 if (chosen == NULL ||
1165 chosen_size < info.component_size) {
1166 chosen_size = info.component_size;
1167 free(chosen);
1168 chosen = devname;
1169 devname = NULL;
1170 if (chosen_st) {
1171 chosen_st->ss->free_super(chosen_st);
1172 free(chosen_st);
1173 }
1174 chosen_st = st2;
1175 st2 = NULL;
1176 }
1177
1178 next:
1179 free(devname);
1180 domain_free(domlist);
1181 dev_policy_free(pol2);
1182 if (st2)
1183 st2->ss->free_super(st2);
1184 free(st2);
1185
1186 if (fd >= 0)
1187 close(fd);
1188 }
1189
1190 if (!chosen)
1191 return 1;
1192
1193 /* 'chosen' is the best device we can find. Let's write its
1194 * metadata to devname dfd is read-only so don't use that
1195 */
1196 fd = open(devname, O_RDWR);
1197 if (fd >= 0) {
1198 chosen_st->ss->store_super(chosen_st, fd);
1199 close(fd);
1200 }
1201 free(chosen);
1202 chosen_st->ss->free_super(chosen_st);
1203 free(chosen_st);
1204 return 0;
1205 }
1206
1207 static int is_bare(int dfd)
1208 {
1209 unsigned long long size = 0;
1210 char bufpad[4096 + 4096];
1211 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1212
1213 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1214 read(dfd, buf, 4096) != 4096)
1215 return 0;
1216
1217 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1218 return 0;
1219 if (memcmp(buf, buf+1, 4095) != 0)
1220 return 0;
1221
1222 /* OK, first 4K appear blank, try the end. */
1223 get_dev_size(dfd, NULL, &size);
1224 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1225 read(dfd, buf, 4096) != 4096)
1226 return 0;
1227
1228 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1229 return 0;
1230 if (memcmp(buf, buf+1, 4095) != 0)
1231 return 0;
1232
1233 return 1;
1234 }
1235
1236 /* adding a spare to a regular array is quite different from adding one to
1237 * a set-of-partitions virtual array.
1238 * This function determines which is worth trying and tries as appropriate.
1239 * Arrays are given priority over partitions.
1240 */
1241 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1242 struct map_ent *target,
1243 struct supertype *st, int verbose)
1244 {
1245 int i;
1246 int rv;
1247 int arrays_ok = 0;
1248 int partitions_ok = 0;
1249 int dfd = *dfdp;
1250 int bare;
1251
1252 /* Can only add a spare if device has at least one domain */
1253 if (pol_find(pol, pol_domain) == NULL)
1254 return 1;
1255 /* And only if some action allows spares */
1256 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1257 return 1;
1258
1259 /* Now check if the device is bare.
1260 * bare devices can always be added as a spare
1261 * non-bare devices can only be added if spare-same-slot is permitted,
1262 * and this device is replacing a previous device - in which case 'target'
1263 * will be set.
1264 */
1265 if (!is_bare(dfd)) {
1266 /* Must have a target and allow same_slot */
1267 /* Later - may allow force_spare without target */
1268 if (!target ||
1269 !policy_action_allows(pol, st?st->ss->name:NULL,
1270 act_spare_same_slot)) {
1271 if (verbose > 1)
1272 fprintf(stderr, Name ": %s is not bare, so not "
1273 "considering as a spare\n",
1274 devname);
1275 return 1;
1276 }
1277 bare = 0;
1278 } else
1279 bare = 1;
1280
1281 /* It might be OK to add this device to an array - need to see
1282 * what arrays might be candidates.
1283 */
1284 if (st) {
1285 /* just try try 'array' or 'partition' based on this metadata */
1286 if (st->ss->add_to_super)
1287 return array_try_spare(devname, dfdp, pol, target, bare,
1288 st, verbose);
1289 else
1290 return partition_try_spare(devname, dfdp, pol,
1291 st, verbose);
1292 }
1293 /* No metadata was specified or found so options are open.
1294 * Check for whether any array metadata, or any partition metadata
1295 * might allow adding the spare. This check is just help to avoid
1296 * a more costly scan of all arrays when we can be sure that will
1297 * fail.
1298 */
1299 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1300 if (superlist[i]->add_to_super && !arrays_ok &&
1301 policy_action_allows(pol, superlist[i]->name, act_spare))
1302 arrays_ok = 1;
1303 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1304 policy_action_allows(pol, superlist[i]->name, act_spare))
1305 partitions_ok = 1;
1306 }
1307 rv = 1;
1308 if (arrays_ok)
1309 rv = array_try_spare(devname, dfdp, pol, target, bare,
1310 st, verbose);
1311 if (rv != 0 && partitions_ok)
1312 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1313 return rv;
1314 }
1315
1316 int IncrementalScan(int verbose)
1317 {
1318 /* look at every device listed in the 'map' file.
1319 * If one is found that is not running then:
1320 * look in mdadm.conf for bitmap file.
1321 * if one exists, but array has none, add it.
1322 * try to start array in auto-readonly mode
1323 */
1324 struct map_ent *mapl = NULL;
1325 struct map_ent *me;
1326 struct mddev_ident *devs, *mddev;
1327 int rv = 0;
1328
1329 map_read(&mapl);
1330 devs = conf_get_ident(NULL);
1331
1332 for (me = mapl ; me ; me = me->next) {
1333 mdu_array_info_t array;
1334 mdu_bitmap_file_t bmf;
1335 struct mdinfo *sra;
1336 int mdfd = open_dev(me->devnum);
1337
1338 if (mdfd < 0)
1339 continue;
1340 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1341 errno != ENODEV) {
1342 close(mdfd);
1343 continue;
1344 }
1345 /* Ok, we can try this one. Maybe it needs a bitmap */
1346 for (mddev = devs ; mddev ; mddev = mddev->next)
1347 if (mddev->devname && me->path
1348 && devname_matches(mddev->devname, me->path))
1349 break;
1350 if (mddev && mddev->bitmap_file) {
1351 /*
1352 * Note: early kernels will wrongly fail this, so it
1353 * is a hint only
1354 */
1355 int added = -1;
1356 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1357 int bmfd = open(mddev->bitmap_file, O_RDWR);
1358 if (bmfd >= 0) {
1359 added = ioctl(mdfd, SET_BITMAP_FILE,
1360 bmfd);
1361 close(bmfd);
1362 }
1363 }
1364 if (verbose >= 0) {
1365 if (added == 0)
1366 fprintf(stderr, Name
1367 ": Added bitmap %s to %s\n",
1368 mddev->bitmap_file, me->path);
1369 else if (errno != EEXIST)
1370 fprintf(stderr, Name
1371 ": Failed to add bitmap to %s: %s\n",
1372 me->path, strerror(errno));
1373 }
1374 }
1375 sra = sysfs_read(mdfd, 0, 0);
1376 if (sra) {
1377 if (sysfs_set_str(sra, NULL,
1378 "array_state", "read-auto") == 0) {
1379 if (verbose >= 0)
1380 fprintf(stderr, Name
1381 ": started array %s\n",
1382 me->path ?: devnum2devname(me->devnum));
1383 } else {
1384 fprintf(stderr, Name
1385 ": failed to start array %s: %s\n",
1386 me->path ?: devnum2devname(me->devnum),
1387 strerror(errno));
1388 rv = 1;
1389 }
1390 }
1391 }
1392 return rv;
1393 }
1394
1395 static char *container2devname(char *devname)
1396 {
1397 char *mdname = NULL;
1398
1399 if (devname[0] == '/') {
1400 int fd = open(devname, O_RDONLY);
1401 if (fd >= 0) {
1402 mdname = devnum2devname(fd2devnum(fd));
1403 close(fd);
1404 }
1405 } else {
1406 int uuid[4];
1407 struct map_ent *mp, *map = NULL;
1408
1409 if (!parse_uuid(devname, uuid))
1410 return mdname;
1411 mp = map_by_uuid(&map, uuid);
1412 if (mp)
1413 mdname = devnum2devname(mp->devnum);
1414 map_free(map);
1415 }
1416
1417 return mdname;
1418 }
1419
1420 static int Incremental_container(struct supertype *st, char *devname,
1421 char *homehost, int verbose,
1422 int runstop, int autof, int freeze_reshape)
1423 {
1424 /* Collect the contents of this container and for each
1425 * array, choose a device name and assemble the array.
1426 */
1427
1428 struct mdinfo *list;
1429 struct mdinfo *ra;
1430 struct map_ent *map = NULL;
1431 struct mdinfo info;
1432 int trustworthy;
1433 struct mddev_ident *match;
1434 int rv = 0;
1435 struct domainlist *domains;
1436 struct map_ent *smp;
1437 int suuid[4];
1438 int sfd;
1439
1440 st->ss->getinfo_super(st, &info, NULL);
1441
1442 if ((runstop > 0 && info.container_enough >= 0) ||
1443 info.container_enough > 0)
1444 /* pass */;
1445 else {
1446 if (verbose)
1447 fprintf(stderr, Name ": not enough devices to start the container\n");
1448 return 0;
1449 }
1450
1451 match = search_mdstat(st, &info, devname, verbose, &rv);
1452 if (match == NULL && rv == 2)
1453 return rv;
1454
1455 /* Need to compute 'trustworthy' */
1456 if (match)
1457 trustworthy = LOCAL;
1458 else if (st->ss->match_home(st, homehost) == 1)
1459 trustworthy = LOCAL;
1460 else if (st->ss->match_home(st, "any") == 1)
1461 trustworthy = LOCAL;
1462 else
1463 trustworthy = FOREIGN;
1464
1465 list = st->ss->container_content(st, NULL);
1466 if (map_lock(&map))
1467 fprintf(stderr, Name ": failed to get exclusive lock on "
1468 "mapfile\n");
1469 /* do not assemble arrays that might have bad blocks */
1470 if (list && list->array.state & (1<<MD_SB_BBM_ERRORS)) {
1471 fprintf(stderr, Name ": BBM log found in metadata. "
1472 "Cannot activate array(s).\n");
1473 /* free container data and exit */
1474 sysfs_free(list);
1475 map_unlock(&map);
1476 return 2;
1477 }
1478
1479 for (ra = list ; ra ; ra = ra->next) {
1480 int mdfd;
1481 char chosen_name[1024];
1482 struct map_ent *mp;
1483 struct mddev_ident *match = NULL;
1484
1485 mp = map_by_uuid(&map, ra->uuid);
1486
1487 if (mp) {
1488 mdfd = open_dev(mp->devnum);
1489 if (mp->path)
1490 strcpy(chosen_name, mp->path);
1491 else
1492 strcpy(chosen_name, devnum2devname(mp->devnum));
1493 } else {
1494
1495 /* Check in mdadm.conf for container == devname and
1496 * member == ra->text_version after second slash.
1497 */
1498 char *sub = strchr(ra->text_version+1, '/');
1499 struct mddev_ident *array_list;
1500 if (sub) {
1501 sub++;
1502 array_list = conf_get_ident(NULL);
1503 } else
1504 array_list = NULL;
1505 for(; array_list ; array_list = array_list->next) {
1506 char *dn;
1507 if (array_list->member == NULL ||
1508 array_list->container == NULL)
1509 continue;
1510 if (strcmp(array_list->member, sub) != 0)
1511 continue;
1512 if (array_list->uuid_set &&
1513 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1514 continue;
1515 dn = container2devname(array_list->container);
1516 if (dn == NULL)
1517 continue;
1518 if (strncmp(dn, ra->text_version+1,
1519 strlen(dn)) != 0 ||
1520 ra->text_version[strlen(dn)+1] != '/') {
1521 free(dn);
1522 continue;
1523 }
1524 free(dn);
1525 /* we have a match */
1526 match = array_list;
1527 if (verbose>0)
1528 fprintf(stderr, Name ": match found for member %s\n",
1529 array_list->member);
1530 break;
1531 }
1532
1533 if (match && match->devname &&
1534 strcasecmp(match->devname, "<ignore>") == 0) {
1535 if (verbose > 0)
1536 fprintf(stderr, Name ": array %s/%s is "
1537 "explicitly ignored by mdadm.conf\n",
1538 match->container, match->member);
1539 map_unlock(&map);
1540 return 2;
1541 }
1542 if (match)
1543 trustworthy = LOCAL;
1544
1545 mdfd = create_mddev(match ? match->devname : NULL,
1546 ra->name,
1547 autof,
1548 trustworthy,
1549 chosen_name);
1550 }
1551
1552 if (mdfd < 0) {
1553 fprintf(stderr, Name ": failed to open %s: %s.\n",
1554 chosen_name, strerror(errno));
1555 map_unlock(&map);
1556 return 2;
1557 }
1558
1559 assemble_container_content(st, mdfd, ra, runstop,
1560 chosen_name, verbose, NULL,
1561 freeze_reshape);
1562 close(mdfd);
1563 }
1564
1565 /* Now move all suitable spares from spare container */
1566 domains = domain_from_array(list, st->ss->name);
1567 memcpy(suuid, uuid_zero, sizeof(int[4]));
1568 if (domains &&
1569 (smp = map_by_uuid(&map, suuid)) != NULL &&
1570 (sfd = open(smp->path, O_RDONLY)) >= 0) {
1571 /* spare container found */
1572 struct supertype *sst =
1573 super_imsm.match_metadata_desc("imsm");
1574 struct mdinfo *sinfo;
1575 unsigned long long min_size = 0;
1576 if (st->ss->min_acceptable_spare_size)
1577 min_size = st->ss->min_acceptable_spare_size(st);
1578 if (!sst->ss->load_container(sst, sfd, NULL)) {
1579 close(sfd);
1580 sinfo = container_choose_spares(sst, min_size,
1581 domains, NULL,
1582 st->ss->name, 0);
1583 sst->ss->free_super(sst);
1584 if (sinfo){
1585 int count = 0;
1586 struct mdinfo *disks = sinfo->devs;
1587 while (disks) {
1588 /* move spare from spare
1589 * container to currently
1590 * assembled one
1591 */
1592 if (move_spare(
1593 smp->path,
1594 devname,
1595 makedev(disks->disk.major,
1596 disks->disk.minor)))
1597 count++;
1598 disks = disks->next;
1599 }
1600 if (count)
1601 fprintf(stderr, Name
1602 ": Added %d spare%s to %s\n",
1603 count, count>1?"s":"", devname);
1604 }
1605 sysfs_free(sinfo);
1606 } else
1607 close(sfd);
1608 }
1609 domain_free(domains);
1610 map_unlock(&map);
1611 return 0;
1612 }
1613
1614 /*
1615 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1616 * raid arrays, and if so first fail (if needed) and then remove the device.
1617 *
1618 * @devname - The device we want to remove
1619 * @id_path - name as found in /dev/disk/by-path for this device
1620 *
1621 * Note: the device name must be a kernel name like "sda", so
1622 * that we can find it in /proc/mdstat
1623 */
1624 int IncrementalRemove(char *devname, char *id_path, int verbose)
1625 {
1626 int mdfd;
1627 int rv;
1628 struct mdstat_ent *ent;
1629 struct mddev_dev devlist;
1630
1631 if (!id_path)
1632 dprintf(Name ": incremental removal without --path <id_path> "
1633 "lacks the possibility to re-add new device in this "
1634 "port\n");
1635
1636 if (strchr(devname, '/')) {
1637 fprintf(stderr, Name ": incremental removal requires a "
1638 "kernel device name, not a file: %s\n", devname);
1639 return 1;
1640 }
1641 ent = mdstat_by_component(devname);
1642 if (!ent) {
1643 fprintf(stderr, Name ": %s does not appear to be a component "
1644 "of any array\n", devname);
1645 return 1;
1646 }
1647 mdfd = open_dev(ent->devnum);
1648 if (mdfd < 0) {
1649 fprintf(stderr, Name ": Cannot open array %s!!\n", ent->dev);
1650 free_mdstat(ent);
1651 return 1;
1652 }
1653
1654 if (id_path) {
1655 struct map_ent *map = NULL, *me;
1656 me = map_by_devnum(&map, ent->devnum);
1657 if (me)
1658 policy_save_path(id_path, me);
1659 map_free(map);
1660 }
1661
1662 memset(&devlist, 0, sizeof(devlist));
1663 devlist.devname = devname;
1664 devlist.disposition = 'f';
1665 /* for a container, we must fail each member array */
1666 if (ent->metadata_version &&
1667 strncmp(ent->metadata_version, "external:", 9) == 0) {
1668 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1669 struct mdstat_ent *memb;
1670 for (memb = mdstat ; memb ; memb = memb->next)
1671 if (is_container_member(memb, ent->dev)) {
1672 int subfd = open_dev(memb->devnum);
1673 if (subfd >= 0) {
1674 Manage_subdevs(memb->dev, subfd,
1675 &devlist, verbose, 0,
1676 NULL, 0);
1677 close(subfd);
1678 }
1679 }
1680 free_mdstat(mdstat);
1681 } else
1682 Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0);
1683 devlist.disposition = 'r';
1684 rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0);
1685 close(mdfd);
1686 free_mdstat(ent);
1687 return rv;
1688 }
Unnamed repository; edit this file 'description' to name the repository.