]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Incremental.c
Remove stray 'free' in block_monitor.
[thirdparty/mdadm.git] / Incremental.c
CommitLineData
8382f19b
NB
1/*
2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
4 *
e736b623 5 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
8382f19b
NB
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"
61018da0
N
32#include <dirent.h>
33#include <ctype.h>
8382f19b 34
2bf740b3
N
35static int count_active(struct supertype *st, struct mdinfo *sra,
36 int mdfd, char **availp,
8382f19b 37 struct mdinfo *info);
7e0f6979 38static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
8382f19b
NB
39 int number, __u64 events, int verbose,
40 char *array_name);
f08605b3 41static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
d2db3045 42 struct map_ent *target,
f08605b3 43 struct supertype *st, int verbose);
8382f19b 44
7d91c3f5 45static int Incremental_container(struct supertype *st, char *devname,
47c74f3f
N
46 char *homehost,
47 int verbose, int runstop, int autof);
7d91c3f5 48
fa56eddb 49static struct mddev_ident *search_mdstat(struct supertype *st,
3a97f210
N
50 struct mdinfo *info,
51 char *devname,
52 int verbose, int *rvp);
53
8382f19b 54int Incremental(char *devname, int verbose, int runstop,
0ac91628
N
55 struct supertype *st, char *homehost, int require_homehost,
56 int autof)
8382f19b
NB
57{
58 /* Add this device to an array, creating the array if necessary
598f0d58 59 * and starting the array if sensible or - if runstop>0 - if possible.
8382f19b
NB
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
215bb3f7
N
67 * 3a/ if not, check for homehost match. If no match, assemble as
68 * a 'foreign' array.
8382f19b
NB
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
352452c3 76 * Don't choose partitioned for containers.
8382f19b
NB
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?
352452c3 88 * For a container, this means running Incremental_container.
8382f19b
NB
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;
7e83544b 96 struct mdinfo info, dinfo;
de6ae750 97 struct mdinfo *sra = NULL, *d;
fa56eddb 98 struct mddev_ident *match;
8382f19b 99 char chosen_name[1024];
15d4a7e4 100 int rv = 1;
8382f19b 101 struct map_ent *mp, *map = NULL;
15d4a7e4 102 int dfd = -1, mdfd = -1;
de6ae750 103 char *avail = NULL;
8382f19b 104 int active_disks;
47c74f3f 105 int trustworthy;
d7288ddc 106 char *name_to_use;
ad5bc697 107 mdu_array_info_t ainf;
7e83544b 108 struct dev_policy *policy = NULL;
d2db3045
N
109 struct map_ent target_array;
110 int have_target;
8382f19b
NB
111
112 struct createinfo *ci = conf_get_create_info();
113
47c74f3f
N
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 */
3a371610 134 if (must_be_container(dfd)) {
47c74f3f
N
135 if (!st)
136 st = super_by_fd(dfd, NULL);
a92b2112 137 if (st && st->ss->load_container)
47c74f3f
N
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 }
8382f19b 149
352452c3 150 /* 1/ Check if device is permitted by mdadm.conf */
8382f19b
NB
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);
15d4a7e4 157 goto out;
8382f19b
NB
158 }
159
160 /* 2/ Find metadata, reject if none appropriate (check
161 * version/name from args) */
162
8382f19b
NB
163 if (fstat(dfd, &stb) < 0) {
164 if (verbose >= 0)
165 fprintf(stderr, Name ": fstat failed for %s: %s.\n",
166 devname, strerror(errno));
15d4a7e4 167 goto out;
8382f19b
NB
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);
15d4a7e4 173 goto out;
8382f19b
NB
174 }
175
7e83544b
N
176 dinfo.disk.major = major(stb.st_rdev);
177 dinfo.disk.minor = minor(stb.st_rdev);
178
179 policy = disk_policy(&dinfo);
d2db3045 180 have_target = policy_check_path(&dinfo, &target_array);
7e83544b 181
8382f19b
NB
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);
d2db3045
N
187 rv = try_spare(devname, &dfd, policy,
188 have_target ? &target_array : NULL,
189 st, verbose);
15d4a7e4 190 goto out;
8382f19b 191 }
0f22b998
N
192 if (st->ss->compare_super == NULL ||
193 st->ss->load_super(st, dfd, NULL)) {
8382f19b
NB
194 if (verbose >= 0)
195 fprintf(stderr, Name ": no RAID superblock on %s.\n",
196 devname);
d2db3045
N
197 rv = try_spare(devname, &dfd, policy,
198 have_target ? &target_array : NULL,
199 st, verbose);
0f22b998 200 free(st);
15d4a7e4 201 goto out;
8382f19b 202 }
15d4a7e4 203 close (dfd); dfd = -1;
8382f19b 204
f35f2525 205 memset(&info, 0, sizeof(info));
a5d85af7 206 st->ss->getinfo_super(st, &info, NULL);
8382f19b 207
3a97f210
N
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;
8382f19b 212
112cace6
N
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);
15d4a7e4 219 goto out;
112cace6
N
220 }
221
7b403fef
N
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. */
215bb3f7
N
226 if (match)
227 trustworthy = LOCAL;
d1d3482b
N
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;
215bb3f7 232 else
d1d3482b
N
233 trustworthy = FOREIGN;
234
235
4e8d9f0a 236 if (!match && !conf_test_metadata(st->ss->name, policy,
d1d3482b
N
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);
15d4a7e4 243 goto out;
d1d3482b
N
244 }
245 if (trustworthy == LOCAL_ANY)
ecb02e31 246 trustworthy = LOCAL;
e12e1d1d 247
4ef2f11e
N
248 /* There are three possible sources for 'autof': command line,
249 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
350ac35d
N
250 * ARRAY takes precedence, then command line, then
251 * CREATE.
4ef2f11e 252 */
350ac35d 253 if (match && match->autof)
4ef2f11e
N
254 autof = match->autof;
255 if (autof == 0)
256 autof = ci->autof;
257
7cdc0872 258 name_to_use = info.name;
0ac91628 259 if (name_to_use[0] == 0 &&
ecb02e31
N
260 info.array.level == LEVEL_CONTAINER &&
261 trustworthy == LOCAL) {
262 name_to_use = info.text_version;
263 trustworthy = METADATA;
264 }
0ac91628
N
265 if (name_to_use[0] && trustworthy != LOCAL &&
266 ! require_homehost &&
267 conf_name_is_free(name_to_use))
268 trustworthy = LOCAL;
ecb02e31 269
7cdc0872
N
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
ad5bc697 276 /* 4/ Check if array exists.
215bb3f7 277 */
93c861ee
DL
278 if (map_lock(&map))
279 fprintf(stderr, Name ": failed to get exclusive lock on "
280 "mapfile\n");
215bb3f7 281 mp = map_by_uuid(&map, info.uuid);
339c2d6c
N
282 if (mp)
283 mdfd = open_dev(mp->devnum);
284 else
215bb3f7 285 mdfd = -1;
d7288ddc 286
8382f19b 287 if (mdfd < 0) {
8382f19b 288
215bb3f7 289 /* Couldn't find an existing array, maybe make a new one */
ffb3e494 290 mdfd = create_mddev(match ? match->devname : NULL,
e12e1d1d 291 name_to_use, autof, trustworthy, chosen_name);
215bb3f7
N
292
293 if (mdfd < 0)
15d4a7e4 294 goto out;
215bb3f7
N
295
296 sysfs_init(&info, mdfd, 0);
297
f35f2525
N
298 if (set_array_info(mdfd, st, &info) != 0) {
299 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
8382f19b 300 chosen_name, strerror(errno));
15d4a7e4
N
301 rv = 2;
302 goto out;
8382f19b 303 }
7801ac20 304
c5afc314
N
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) {
8382f19b
NB
309 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
310 devname, chosen_name, strerror(errno));
311 ioctl(mdfd, STOP_ARRAY, 0);
15d4a7e4
N
312 rv = 2;
313 goto out;
8382f19b 314 }
2bf740b3
N
315 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
316 GET_OFFSET | GET_SIZE));
317
06c7f68e 318 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
8382f19b
NB
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");
8382f19b 327 sysfs_free(sra);
15d4a7e4
N
328 rv = 2;
329 goto out;
8382f19b 330 }
c5afc314 331 info.array.working_disks = 1;
ad5bc697
N
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);
8382f19b
NB
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;
8382f19b 343 int err;
3da92f27 344 struct supertype *st2;
c5afc314 345 struct mdinfo info2, *d;
215bb3f7 346
2bf740b3
N
347 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
348 GET_OFFSET | GET_SIZE));
349
339c2d6c
N
350 if (mp->path)
351 strcpy(chosen_name, mp->path);
352 else
353 strcpy(chosen_name, devnum2devname(mp->devnum));
215bb3f7 354
ef83fe7c
N
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
7e83544b
N
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
3a6ec29a 362 */
ef83fe7c 363 if ((info.disk.state & (1<<MD_DISK_SYNC)) != 0
7e83544b
N
364 && ! policy_action_allows(policy, st->ss->name,
365 act_re_add)
ef83fe7c 366 && runstop < 1) {
1dccfff9
DW
367 int active = 0;
368
369 if (st->ss->external) {
370 char *devname = devnum2devname(fd2devnum(mdfd));
371
372 active = devname && is_container_active(devname);
373 free(devname);
374 } else if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0)
375 active = 1;
376 if (active) {
3a6ec29a
N
377 fprintf(stderr, Name
378 ": not adding %s to active array (without --run) %s\n",
379 devname, chosen_name);
15d4a7e4
N
380 rv = 2;
381 goto out;
3a6ec29a
N
382 }
383 }
15d4a7e4
N
384 if (!sra) {
385 rv = 2;
386 goto out;
387 }
7c548327
N
388 if (sra->devs) {
389 sprintf(dn, "%d:%d", sra->devs->disk.major,
390 sra->devs->disk.minor);
391 dfd2 = dev_open(dn, O_RDONLY);
392 st2 = dup_super(st);
393 if (st2->ss->load_super(st2, dfd2, NULL) ||
394 st->ss->compare_super(st, st2) != 0) {
395 fprintf(stderr, Name
396 ": metadata mismatch between %s and "
397 "chosen array %s\n",
398 devname, chosen_name);
7c548327 399 close(dfd2);
15d4a7e4
N
400 rv = 2;
401 goto out;
7c548327 402 }
8382f19b 403 close(dfd2);
7c548327 404 memset(&info2, 0, sizeof(info2));
a5d85af7 405 st2->ss->getinfo_super(st2, &info2, NULL);
7c548327
N
406 st2->ss->free_super(st2);
407 if (info.array.level != info2.array.level ||
408 memcmp(info.uuid, info2.uuid, 16) != 0 ||
409 info.array.raid_disks != info2.array.raid_disks) {
410 fprintf(stderr, Name
411 ": unexpected difference between %s and %s.\n",
412 chosen_name, devname);
15d4a7e4
N
413 rv = 2;
414 goto out;
7c548327 415 }
8382f19b 416 }
7801ac20
N
417 info2.disk.major = major(stb.st_rdev);
418 info2.disk.minor = minor(stb.st_rdev);
c5afc314
N
419 /* add disk needs to know about containers */
420 if (st->ss->external)
421 sra->array.level = LEVEL_CONTAINER;
ac7de9d9 422 err = add_disk(mdfd, st, sra, &info2);
8382f19b
NB
423 if (err < 0 && errno == EBUSY) {
424 /* could be another device present with the same
425 * disk.number. Find and reject any such
426 */
427 find_reject(mdfd, st, sra, info.disk.number,
428 info.events, verbose, chosen_name);
ac7de9d9 429 err = add_disk(mdfd, st, sra, &info2);
8382f19b
NB
430 }
431 if (err < 0) {
432 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
433 devname, chosen_name, strerror(errno));
15d4a7e4
N
434 rv = 2;
435 goto out;
8382f19b 436 }
c5afc314
N
437 info.array.working_disks = 0;
438 for (d = sra->devs; d; d=d->next)
439 info.array.working_disks ++;
440
8382f19b 441 }
8382f19b
NB
442
443 /* 7/ Is there enough devices to possibly start the array? */
444 /* 7a/ if not, finish with success. */
352452c3 445 if (info.array.level == LEVEL_CONTAINER) {
ed690d36 446 char *devname = NULL;
352452c3 447 /* Try to assemble within the container */
ad5bc697 448 map_unlock(&map);
97590376 449 sysfs_uevent(&info, "change");
c5afc314
N
450 if (verbose >= 0)
451 fprintf(stderr, Name
452 ": container %s now has %d devices\n",
453 chosen_name, info.array.working_disks);
a7c6e3fb 454 wait_for(chosen_name, mdfd);
ed690d36
PC
455 if (st->ss->external)
456 devname = devnum2devname(fd2devnum(mdfd));
a7c6e3fb 457 close(mdfd);
2bf740b3 458 sysfs_free(sra);
03b7f6c6 459 rv = Incremental(chosen_name, verbose, runstop,
0ac91628 460 NULL, homehost, require_homehost, autof);
03b7f6c6
N
461 if (rv == 1)
462 /* Don't fail the whole -I if a subarray didn't
463 * have enough devices to start yet
464 */
465 rv = 0;
ed690d36
PC
466 /* after spare is added, ping monitor for external metadata
467 * so that it can eg. try to rebuild degraded array */
468 if (st->ss->external) {
469 ping_monitor(devname);
470 free(devname);
471 }
03b7f6c6 472 return rv;
352452c3 473 }
de6ae750 474
2bf740b3
N
475 /* We have added something to the array, so need to re-read the
476 * state. Eventually this state should be kept up-to-date as
477 * things change.
478 */
479 sysfs_free(sra);
480 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
481 GET_OFFSET | GET_SIZE));
482 active_disks = count_active(st, sra, mdfd, &avail, &info);
8382f19b
NB
483 if (enough(info.array.level, info.array.raid_disks,
484 info.array.layout, info.array.state & 1,
3288b419 485 avail, active_disks) == 0) {
8382f19b
NB
486 if (verbose >= 0)
487 fprintf(stderr, Name
488 ": %s attached to %s, not enough to start (%d).\n",
489 devname, chosen_name, active_disks);
ad5bc697 490 map_unlock(&map);
15d4a7e4
N
491 rv = 0;
492 goto out;
8382f19b 493 }
8382f19b
NB
494
495 /* 7b/ if yes, */
496 /* - if number of OK devices match expected, or -R and there */
497 /* are enough, */
498 /* + add any bitmap file */
499 /* + start the array (auto-readonly). */
8382f19b
NB
500
501 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
502 if (verbose >= 0)
503 fprintf(stderr, Name
504 ": %s attached to %s which is already active.\n",
505 devname, chosen_name);
ad5bc697 506 map_unlock(&map);
15d4a7e4
N
507 rv = 0;
508 goto out;
8382f19b 509 }
ad5bc697
N
510
511 map_unlock(&map);
8382f19b 512 if (runstop > 0 || active_disks >= info.array.working_disks) {
2bf740b3 513 struct mdinfo *dsk;
8382f19b
NB
514 /* Let's try to start it */
515 if (match && match->bitmap_file) {
516 int bmfd = open(match->bitmap_file, O_RDWR);
517 if (bmfd < 0) {
518 fprintf(stderr, Name
519 ": Could not open bitmap file %s.\n",
520 match->bitmap_file);
15d4a7e4 521 goto out;
8382f19b
NB
522 }
523 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
524 close(bmfd);
525 fprintf(stderr, Name
526 ": Failed to set bitmapfile for %s.\n",
527 chosen_name);
15d4a7e4 528 goto out;
8382f19b
NB
529 }
530 close(bmfd);
531 }
de6ae750
N
532 /* Need to remove from the array any devices which
533 * 'count_active' discerned were too old or inappropriate
534 */
535 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
536 if (d->disk.state & (1<<MD_DISK_REMOVED))
537 remove_disk(mdfd, st, sra, d);
2bf740b3 538
7b403fef 539 if ((sra == NULL || active_disks >= info.array.working_disks)
215bb3f7 540 && trustworthy != FOREIGN)
8382f19b
NB
541 rv = ioctl(mdfd, RUN_ARRAY, NULL);
542 else
543 rv = sysfs_set_str(sra, NULL,
544 "array_state", "read-auto");
545 if (rv == 0) {
546 if (verbose >= 0)
547 fprintf(stderr, Name
7e83544b 548 ": %s attached to %s, which has been started.\n",
8382f19b
NB
549 devname, chosen_name);
550 rv = 0;
a7c6e3fb 551 wait_for(chosen_name, mdfd);
7e83544b
N
552 /* We just started the array, so some devices
553 * might have been evicted from the array
554 * because their event counts were too old.
555 * If the action=re-add policy is in-force for
556 * those devices we should re-add them now.
557 */
558 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
559 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
560 add_disk(mdfd, st, sra, dsk) == 0)
561 fprintf(stderr, Name
562 ": %s re-added to %s\n",
563 dsk->sys_name, chosen_name);
564 }
8382f19b
NB
565 } else {
566 fprintf(stderr, Name
567 ": %s attached to %s, but failed to start: %s.\n",
568 devname, chosen_name, strerror(errno));
569 rv = 1;
570 }
571 } else {
572 if (verbose >= 0)
573 fprintf(stderr, Name
574 ": %s attached to %s, not enough to start safely.\n",
575 devname, chosen_name);
576 rv = 0;
577 }
15d4a7e4 578out:
de6ae750 579 free(avail);
15d4a7e4
N
580 if (dfd >= 0)
581 close(dfd);
582 if (mdfd >= 0)
583 close(mdfd);
7e83544b
N
584 if (policy)
585 dev_policy_free(policy);
2bf740b3
N
586 if (sra)
587 sysfs_free(sra);
8382f19b
NB
588 return rv;
589}
590
fa56eddb 591static struct mddev_ident *search_mdstat(struct supertype *st,
3a97f210
N
592 struct mdinfo *info,
593 char *devname,
594 int verbose, int *rvp)
595{
fa56eddb 596 struct mddev_ident *array_list, *match;
3a97f210
N
597 array_list = conf_get_ident(NULL);
598 match = NULL;
599 for (; array_list; array_list = array_list->next) {
600 if (array_list->uuid_set &&
601 same_uuid(array_list->uuid, info->uuid, st->ss->swapuuid)
602 == 0) {
603 if (verbose >= 2 && array_list->devname)
604 fprintf(stderr, Name
605 ": UUID differs from %s.\n",
606 array_list->devname);
607 continue;
608 }
609 if (array_list->name[0] &&
610 strcasecmp(array_list->name, info->name) != 0) {
611 if (verbose >= 2 && array_list->devname)
612 fprintf(stderr, Name
613 ": Name differs from %s.\n",
614 array_list->devname);
615 continue;
616 }
617 if (array_list->devices &&
618 !match_oneof(array_list->devices, devname)) {
619 if (verbose >= 2 && array_list->devname)
620 fprintf(stderr, Name
621 ": Not a listed device for %s.\n",
622 array_list->devname);
623 continue;
624 }
625 if (array_list->super_minor != UnSet &&
626 array_list->super_minor != info->array.md_minor) {
627 if (verbose >= 2 && array_list->devname)
628 fprintf(stderr, Name
629 ": Different super-minor to %s.\n",
630 array_list->devname);
631 continue;
632 }
633 if (!array_list->uuid_set &&
634 !array_list->name[0] &&
635 !array_list->devices &&
636 array_list->super_minor == UnSet) {
637 if (verbose >= 2 && array_list->devname)
638 fprintf(stderr, Name
639 ": %s doesn't have any identifying information.\n",
640 array_list->devname);
641 continue;
642 }
643 /* FIXME, should I check raid_disks and level too?? */
644
645 if (match) {
646 if (verbose >= 0) {
647 if (match->devname && array_list->devname)
648 fprintf(stderr, Name
649 ": we match both %s and %s - cannot decide which to use.\n",
650 match->devname, array_list->devname);
651 else
652 fprintf(stderr, Name
653 ": multiple lines in mdadm.conf match\n");
654 }
655 *rvp = 2;
656 match = NULL;
657 break;
658 }
659 match = array_list;
660 }
661 return match;
662}
663
7e0f6979 664static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
8382f19b
NB
665 int number, __u64 events, int verbose,
666 char *array_name)
667{
3da92f27 668 /* Find a device attached to this array with a disk.number of number
8382f19b
NB
669 * and events less than the passed events, and remove the device.
670 */
06c7f68e 671 struct mdinfo *d;
8382f19b
NB
672 mdu_array_info_t ra;
673
674 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
675 return; /* not safe to remove from active arrays
676 * without thinking more */
677
678 for (d = sra->devs; d ; d = d->next) {
679 char dn[10];
680 int dfd;
8382f19b 681 struct mdinfo info;
06c7f68e 682 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
8382f19b
NB
683 dfd = dev_open(dn, O_RDONLY);
684 if (dfd < 0)
685 continue;
3da92f27 686 if (st->ss->load_super(st, dfd, NULL)) {
8382f19b
NB
687 close(dfd);
688 continue;
689 }
a5d85af7 690 st->ss->getinfo_super(st, &info, NULL);
3da92f27 691 st->ss->free_super(st);
8382f19b
NB
692 close(dfd);
693
694 if (info.disk.number != number ||
695 info.events >= events)
696 continue;
697
06c7f68e 698 if (d->disk.raid_disk > -1)
8382f19b
NB
699 sysfs_set_str(sra, d, "slot", "none");
700 if (sysfs_set_str(sra, d, "state", "remove") == 0)
701 if (verbose >= 0)
702 fprintf(stderr, Name
703 ": removing old device %s from %s\n",
06c7f68e 704 d->sys_name+4, array_name);
8382f19b
NB
705 }
706}
707
2bf740b3
N
708static int count_active(struct supertype *st, struct mdinfo *sra,
709 int mdfd, char **availp,
8382f19b
NB
710 struct mdinfo *bestinfo)
711{
712 /* count how many devices in sra think they are active */
06c7f68e 713 struct mdinfo *d;
de6ae750 714 int cnt = 0;
8382f19b 715 __u64 max_events = 0;
8382f19b 716 char *avail = NULL;
de6ae750
N
717 int *best;
718 char *devmap = NULL;
719 int numdevs = 0;
720 int devnum;
721 int b, i;
722 int raid_disks = 0;
8382f19b 723
b526e52d
DW
724 if (!sra)
725 return 0;
726
de6ae750
N
727 for (d = sra->devs ; d ; d = d->next)
728 numdevs++;
729 for (d = sra->devs, devnum=0 ; d ; d = d->next, devnum++) {
8382f19b
NB
730 char dn[30];
731 int dfd;
8382f19b
NB
732 int ok;
733 struct mdinfo info;
734
06c7f68e 735 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
8382f19b
NB
736 dfd = dev_open(dn, O_RDONLY);
737 if (dfd < 0)
738 continue;
3da92f27 739 ok = st->ss->load_super(st, dfd, NULL);
8382f19b
NB
740 close(dfd);
741 if (ok != 0)
742 continue;
de6ae750
N
743 info.array.raid_disks = raid_disks;
744 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
43aaf431 745 if (!avail) {
de6ae750
N
746 raid_disks = info.array.raid_disks;
747 avail = calloc(raid_disks, 1);
43aaf431
DL
748 if (!avail) {
749 fprintf(stderr, Name ": out of memory.\n");
750 exit(1);
751 }
43aaf431 752 *availp = avail;
de6ae750
N
753
754 best = calloc(raid_disks, sizeof(int));
755 devmap = calloc(raid_disks * numdevs, 1);
756
757 st->ss->getinfo_super(st, &info, devmap);
43aaf431
DL
758 }
759
8382f19b
NB
760 if (info.disk.state & (1<<MD_DISK_SYNC))
761 {
8382f19b
NB
762 if (cnt == 0) {
763 cnt++;
764 max_events = info.events;
765 avail[info.disk.raid_disk] = 2;
de6ae750 766 best[info.disk.raid_disk] = devnum;
a5d85af7 767 st->ss->getinfo_super(st, bestinfo, NULL);
8382f19b 768 } else if (info.events == max_events) {
8382f19b 769 avail[info.disk.raid_disk] = 2;
de6ae750 770 best[info.disk.raid_disk] = devnum;
8382f19b 771 } else if (info.events == max_events-1) {
de6ae750
N
772 if (avail[info.disk.raid_disk] == 0) {
773 avail[info.disk.raid_disk] = 1;
774 best[info.disk.raid_disk] = devnum;
775 }
8382f19b
NB
776 } else if (info.events < max_events - 1)
777 ;
778 else if (info.events == max_events+1) {
779 int i;
8382f19b 780 max_events = info.events;
de6ae750 781 for (i=0; i < raid_disks; i++)
8382f19b
NB
782 if (avail[i])
783 avail[i]--;
784 avail[info.disk.raid_disk] = 2;
de6ae750 785 best[info.disk.raid_disk] = devnum;
a5d85af7 786 st->ss->getinfo_super(st, bestinfo, NULL);
8382f19b 787 } else { /* info.events much bigger */
8382f19b
NB
788 memset(avail, 0, info.disk.raid_disk);
789 max_events = info.events;
1d1a9f87 790 avail[info.disk.raid_disk] = 2;
a5d85af7 791 st->ss->getinfo_super(st, bestinfo, NULL);
8382f19b
NB
792 }
793 }
3da92f27 794 st->ss->free_super(st);
8382f19b 795 }
de6ae750
N
796 if (!avail)
797 return 0;
798 /* We need to reject any device that thinks the best device is
799 * failed or missing */
800 for (b = 0; b < raid_disks; b++)
801 if (avail[b] == 2)
802 break;
803 cnt = 0;
804 for (i = 0 ; i < raid_disks ; i++) {
805 if (i != b && avail[i])
806 if (devmap[raid_disks * best[i] + b] == 0) {
807 /* This device thinks 'b' is failed -
808 * don't use it */
809 devnum = best[i];
810 for (d=sra->devs ; devnum; d = d->next)
811 devnum--;
812 d->disk.state |= (1 << MD_DISK_REMOVED);
813 avail[i] = 0;
814 }
815 if (avail[i])
816 cnt++;
817 }
818 free(best);
819 free(devmap);
820 return cnt;
8382f19b
NB
821}
822
56e8be85 823static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
d2db3045 824 struct map_ent *target, int bare,
56e8be85 825 struct supertype *st, int verbose)
f08605b3
N
826{
827 /* This device doesn't have any md metadata
d2db3045
N
828 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
829 * If 'st' is not set, then we only know that some metadata allows this,
830 * others possibly don't.
831 * So look for a container or array to attach the device to.
832 * Prefer 'target' if that is set and the array is found.
833 *
f08605b3
N
834 * If st is set, then only arrays of that type are considered
835 * Return 0 on success, or some exit code on failure, probably 1.
836 */
8659d089 837 int rv = 1;
f08605b3
N
838 struct stat stb;
839 struct map_ent *mp, *map = NULL;
840 struct mdinfo *chosen = NULL;
841 int dfd = *dfdp;
842
f08605b3
N
843 if (fstat(dfd, &stb) != 0)
844 return 1;
f08605b3 845
56e8be85 846 /*
f08605b3
N
847 * Now we need to find a suitable array to add this to.
848 * We only accept arrays that:
849 * - match 'st'
850 * - are in the same domains as the device
851 * - are of an size for which the device will be useful
852 * and we choose the one that is the most degraded
853 */
854
855 if (map_lock(&map)) {
856 fprintf(stderr, Name ": failed to get exclusive lock on "
857 "mapfile\n");
858 return 1;
859 }
860 for (mp = map ; mp ; mp = mp->next) {
861 struct supertype *st2;
862 struct domainlist *dl = NULL;
863 struct mdinfo *sra;
864 unsigned long long devsize;
865
866 if (is_subarray(mp->metadata))
867 continue;
868 if (st) {
869 st2 = st->ss->match_metadata_desc(mp->metadata);
870 if (!st2 ||
871 (st->minor_version >= 0 &&
872 st->minor_version != st2->minor_version)) {
873 if (verbose > 1)
874 fprintf(stderr, Name ": not adding %s to %s as metadata type doesn't match\n",
875 devname, mp->path);
876 free(st2);
877 continue;
878 }
879 free(st2);
880 }
881 sra = sysfs_read(-1, mp->devnum,
882 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
883 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
884 if (!sra) {
885 /* Probably a container - no degraded info */
886 sra = sysfs_read(-1, mp->devnum,
887 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
888 GET_COMPONENT|GET_VERSION);
889 if (sra)
890 sra->array.failed_disks = 0;
891 }
892 if (!sra)
893 continue;
894 if (st == NULL) {
895 int i;
896 st2 = NULL;
897 for(i=0; !st2 && superlist[i]; i++)
898 st2 = superlist[i]->match_metadata_desc(
899 sra->text_version);
6e57f80a
N
900 if (!st2) {
901 if (verbose > 1)
902 fprintf(stderr, Name ": not adding %s to %s"
903 " as metadata not recognised.\n",
904 devname, mp->path);
905 goto next;
906 }
625b2507
N
907 /* Need to double check the 'act_spare' permissions applies
908 * to this metadata.
909 */
910 if (!policy_action_allows(pol, st2->ss->name, act_spare))
911 goto next;
d2db3045
N
912 if (!bare && !policy_action_allows(pol, st2->ss->name,
913 act_spare_same_slot))
914 goto next;
f08605b3
N
915 } else
916 st2 = st;
917 get_dev_size(dfd, NULL, &devsize);
918 if (st2->ss->avail_size(st2, devsize) < sra->component_size) {
919 if (verbose > 1)
920 fprintf(stderr, Name ": not adding %s to %s as it is too small\n",
921 devname, mp->path);
922 goto next;
923 }
924 dl = domain_from_array(sra, st2->ss->name);
925 if (!domain_test(dl, pol, st2->ss->name)) {
926 /* domain test fails */
927 if (verbose > 1)
928 fprintf(stderr, Name ": not adding %s to %s as it is not in a compatible domain\n",
929 devname, mp->path);
930
931 goto next;
932 }
d2db3045
N
933 /* test against target.
934 * If 'target' is set and 'bare' is false, we only accept
935 * arrays/containers that match 'target'.
936 * If 'target' is set and 'bare' is true, we prefer the
937 * array which matches 'target'.
938 */
939 if (target) {
940 if (strcmp(target->metadata, mp->metadata) == 0 &&
941 memcmp(target->uuid, mp->uuid,
942 sizeof(target->uuid)) == 0) {
943 /* This is our target!! */
944 if (chosen)
945 sysfs_free(chosen);
946 chosen = sra;
947 sra = NULL;
948 /* skip to end so we don't check any more */
949 while (mp->next)
950 mp = mp->next;
951 goto next;
952 }
953 /* not our target */
954 if (!bare)
955 goto next;
956 }
957
f08605b3
N
958 /* all tests passed, OK to add to this array */
959 if (!chosen) {
960 chosen = sra;
961 sra = NULL;
962 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
963 sysfs_free(chosen);
964 chosen = sra;
965 sra = NULL;
966 }
967 next:
968 if (sra)
969 sysfs_free(sra);
970 if (st != st2)
971 free(st2);
972 if (dl)
973 domain_free(dl);
974 }
975 if (chosen) {
976 /* add current device to chosen array as a spare */
977 int mdfd = open_dev(devname2devnum(chosen->sys_name));
978 if (mdfd >= 0) {
a655e550 979 struct mddev_dev devlist;
f08605b3
N
980 char devname[20];
981 devlist.next = NULL;
982 devlist.used = 0;
983 devlist.re_add = 0;
984 devlist.writemostly = 0;
985 devlist.devname = devname;
986 sprintf(devname, "%d:%d", major(stb.st_rdev),
987 minor(stb.st_rdev));
988 devlist.disposition = 'a';
989 close(dfd);
990 *dfdp = -1;
991 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
833bb0f8 992 -1, 0, NULL);
f08605b3
N
993 close(mdfd);
994 }
995 if (verbose > 0) {
996 if (rv == 0)
997 fprintf(stderr, Name ": added %s as spare for %s\n",
998 devname, chosen->sys_name);
999 else
1000 fprintf(stderr, Name ": failed to add %s as spare for %s\n",
1001 devname, chosen->sys_name);
1002 }
1003 sysfs_free(chosen);
1004 }
8659d089 1005 return rv;
f08605b3
N
1006}
1007
56e8be85
N
1008static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1009 struct supertype *st, int verbose)
1010{
61018da0
N
1011 /* we know that at least one partition virtual-metadata is
1012 * allowed to incorporate spares like this device. We need to
1013 * find a suitable device to copy partition information from.
1014 *
1015 * Getting a list of all disk (not partition) devices is
1016 * slightly non-trivial. We could look at /sys/block, but
1017 * that is theoretically due to be removed. Maybe best to use
1018 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1019 * we depend on this directory of 'path' info. But that fails
1020 * to find loop devices and probably others. Maybe don't
1021 * worry about that, they aren't the real target.
1022 *
1023 * So: check things in /dev/disk/by-path to see if they are in
1024 * a compatible domain, then load the partition table and see
1025 * if it is OK for the new device, and choose the largest
1026 * partition table that fits.
1027 */
1028 DIR *dir;
1029 struct dirent *de;
1030 char *chosen = NULL;
1031 unsigned long long chosen_size;
1032 struct supertype *chosen_st = NULL;
1033 int fd;
1034
1035 dir = opendir("/dev/disk/by-path");
1036 if (!dir)
1037 return 1;
1038 while ((de = readdir(dir)) != NULL) {
1039 char *ep;
1040 struct dev_policy *pol2 = NULL;
1041 struct domainlist *domlist = NULL;
1042 int fd = -1;
1043 struct mdinfo info;
1044 struct supertype *st2 = NULL;
1045 char *devname = NULL;
1046 unsigned long long devsectors;
1047
1048 if (de->d_ino == 0 ||
1049 de->d_name[0] == '.' ||
1050 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1051 goto next;
1052
1053 ep = de->d_name + strlen(de->d_name);
1054 while (ep > de->d_name &&
1055 isdigit(ep[-1]))
1056 ep--;
1057 if (ep > de->d_name + 5 &&
1058 strncmp(ep-5, "-part", 5) == 0)
1059 /* This is a partition - skip it */
1060 goto next;
1061
1062 pol2 = path_policy(de->d_name, type_disk);
1063
1064 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1065 if (domain_test(domlist, pol, st ? st->ss->name : NULL) == 0)
1066 /* new device is incompatible with this device. */
1067 goto next;
1068
1069 domain_free(domlist);
1070 domlist = NULL;
1071
1072 asprintf(&devname, "/dev/disk/by-path/%s", de->d_name);
1073 fd = open(devname, O_RDONLY);
1074 if (fd < 0)
1075 goto next;
1076 if (get_dev_size(fd, devname, &devsectors) == 0)
1077 goto next;
1078 devsectors >>= 9;
1079
1080 if (st)
1081 st2 = dup_super(st);
1082 else
1083 st2 = guess_super_type(fd, guess_partitions);
1084 if (st2 == NULL ||
1085 st2->ss->load_super(st2, fd, NULL) < 0)
1086 goto next;
1087
1088 if (!st) {
1089 /* Check domain policy again, this time referring to metadata */
1090 domain_merge(&domlist, pol2, st2->ss->name);
1091 if (domain_test(domlist, pol, st2->ss->name) == 0)
1092 /* Incompatible devices for this metadata type */
1093 goto next;
625b2507
N
1094 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1095 /* Some partition types allow sparing, but not
1096 * this one.
1097 */
1098 goto next;
61018da0
N
1099 }
1100
a5d85af7 1101 st2->ss->getinfo_super(st2, &info, NULL);
61018da0
N
1102 if (info.component_size > devsectors)
1103 /* This partitioning doesn't fit in the device */
1104 goto next;
1105
1106 /* This is an acceptable device to copy partition
1107 * metadata from. We could just stop here, but I
1108 * think I want to keep looking incase a larger
1109 * metadata which makes better use of the device can
1110 * be found.
1111 */
1112 if (chosen == NULL ||
1113 chosen_size < info.component_size) {
1114 chosen_size = info.component_size;
1115 free(chosen);
1116 chosen = devname;
1117 devname = NULL;
1118 if (chosen_st) {
1119 chosen_st->ss->free_super(chosen_st);
1120 free(chosen_st);
1121 }
1122 chosen_st = st2;
1123 st2 = NULL;
1124 }
1125
1126 next:
1127 free(devname);
1128 domain_free(domlist);
1129 dev_policy_free(pol2);
1130 if (st2)
1131 st2->ss->free_super(st2);
1132 free(st2);
1133
1134 if (fd >= 0)
1135 close(fd);
1136 }
1137
1138 if (!chosen)
1139 return 1;
1140
1141 /* 'chosen' is the best device we can find. Let's write its
1142 * metadata to devname dfd is read-only so don't use that
1143 */
1144 fd = open(devname, O_RDWR);
1145 if (fd >= 0) {
1146 chosen_st->ss->store_super(chosen_st, fd);
1147 close(fd);
1148 }
1149 free(chosen);
1150 chosen_st->ss->free_super(chosen_st);
1151 free(chosen_st);
1152 return 0;
56e8be85
N
1153}
1154
52e965c2
N
1155static int is_bare(int dfd)
1156{
1157 unsigned long long size = 0;
1158 char bufpad[4096 + 4096];
1159 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1160
1161 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1162 read(dfd, buf, 4096) != 4096)
1163 return 0;
1164
1165 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1166 return 0;
1167 if (memcmp(buf, buf+1, 4095) != 0)
1168 return 0;
1169
1170 /* OK, first 4K appear blank, try the end. */
1171 get_dev_size(dfd, NULL, &size);
1172 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1173 read(dfd, buf, 4096) != 4096)
1174 return 0;
1175
1176 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1177 return 0;
1178 if (memcmp(buf, buf+1, 4095) != 0)
1179 return 0;
1180
1181 return 1;
1182}
56e8be85
N
1183
1184/* adding a spare to a regular array is quite different from adding one to
1185 * a set-of-partitions virtual array.
1186 * This function determines which is worth trying and tries as appropriate.
1187 * Arrays are given priority over partitions.
1188 */
1189static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
d2db3045 1190 struct map_ent *target,
56e8be85
N
1191 struct supertype *st, int verbose)
1192{
1193 int i;
1194 int rv;
1195 int arrays_ok = 0;
1196 int partitions_ok = 0;
56e8be85 1197 int dfd = *dfdp;
d2db3045 1198 int bare;
56e8be85 1199
d2db3045 1200 /* Can only add a spare if device has at least one domain */
56e8be85
N
1201 if (pol_find(pol, pol_domain) == NULL)
1202 return 1;
1203 /* And only if some action allows spares */
1204 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1205 return 1;
1206
d2db3045
N
1207 /* Now check if the device is bare.
1208 * bare devices can always be added as a spare
1209 * non-bare devices can only be added if spare-same-slot is permitted,
1210 * and this device is replacing a previous device - in which case 'target'
1211 * will be set.
56e8be85 1212 */
52e965c2 1213 if (!is_bare(dfd)) {
d2db3045
N
1214 /* Must have a target and allow same_slot */
1215 /* Later - may allow force_spare without target */
1216 if (!target ||
1217 !policy_action_allows(pol, st?st->ss->name:NULL,
1218 act_spare_same_slot)) {
1219 if (verbose > 1)
1220 fprintf(stderr, Name ": %s is not bare, so not "
1221 "considering as a spare\n",
1222 devname);
1223 return 1;
1224 }
1225 bare = 0;
1226 } else
1227 bare = 1;
56e8be85 1228
d2db3045
N
1229 /* It might be OK to add this device to an array - need to see
1230 * what arrays might be candidates.
56e8be85
N
1231 */
1232 if (st) {
1233 /* just try try 'array' or 'partition' based on this metadata */
1234 if (st->ss->add_to_super)
d2db3045 1235 return array_try_spare(devname, dfdp, pol, target, bare,
56e8be85
N
1236 st, verbose);
1237 else
1238 return partition_try_spare(devname, dfdp, pol,
1239 st, verbose);
1240 }
d2db3045
N
1241 /* No metadata was specified or found so options are open.
1242 * Check for whether any array metadata, or any partition metadata
1243 * might allow adding the spare. This check is just help to avoid
1244 * a more costly scan of all arrays when we can be sure that will
1245 * fail.
1246 */
56e8be85
N
1247 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1248 if (superlist[i]->add_to_super && !arrays_ok &&
1249 policy_action_allows(pol, superlist[i]->name, act_spare))
1250 arrays_ok = 1;
1251 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1252 policy_action_allows(pol, superlist[i]->name, act_spare))
1253 partitions_ok = 1;
1254 }
aaccda44 1255 rv = 1;
56e8be85 1256 if (arrays_ok)
d2db3045
N
1257 rv = array_try_spare(devname, dfdp, pol, target, bare,
1258 st, verbose);
aaccda44 1259 if (rv != 0 && partitions_ok)
56e8be85
N
1260 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1261 return rv;
1262}
1263
8382f19b
NB
1264int IncrementalScan(int verbose)
1265{
1266 /* look at every device listed in the 'map' file.
1267 * If one is found that is not running then:
1268 * look in mdadm.conf for bitmap file.
1269 * if one exists, but array has none, add it.
1270 * try to start array in auto-readonly mode
1271 */
1272 struct map_ent *mapl = NULL;
1273 struct map_ent *me;
fa56eddb 1274 struct mddev_ident *devs, *mddev;
8382f19b
NB
1275 int rv = 0;
1276
1277 map_read(&mapl);
1278 devs = conf_get_ident(NULL);
1279
1280 for (me = mapl ; me ; me = me->next) {
8382f19b
NB
1281 mdu_array_info_t array;
1282 mdu_bitmap_file_t bmf;
7e0f6979 1283 struct mdinfo *sra;
339c2d6c 1284 int mdfd = open_dev(me->devnum);
215bb3f7 1285
8382f19b
NB
1286 if (mdfd < 0)
1287 continue;
1288 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1289 errno != ENODEV) {
1290 close(mdfd);
1291 continue;
1292 }
1293 /* Ok, we can try this one. Maybe it needs a bitmap */
1294 for (mddev = devs ; mddev ; mddev = mddev->next)
339c2d6c 1295 if (mddev->devname && me->path
2400e6eb 1296 && devname_matches(mddev->devname, me->path))
8382f19b
NB
1297 break;
1298 if (mddev && mddev->bitmap_file) {
1299 /*
1300 * Note: early kernels will wrongly fail this, so it
1301 * is a hint only
1302 */
1303 int added = -1;
1304 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1305 int bmfd = open(mddev->bitmap_file, O_RDWR);
1306 if (bmfd >= 0) {
1307 added = ioctl(mdfd, SET_BITMAP_FILE,
1308 bmfd);
1309 close(bmfd);
1310 }
1311 }
1312 if (verbose >= 0) {
1313 if (added == 0)
1314 fprintf(stderr, Name
1315 ": Added bitmap %s to %s\n",
1316 mddev->bitmap_file, me->path);
1317 else if (errno != EEXIST)
1318 fprintf(stderr, Name
1319 ": Failed to add bitmap to %s: %s\n",
1320 me->path, strerror(errno));
1321 }
1322 }
1323 sra = sysfs_read(mdfd, 0, 0);
1324 if (sra) {
1325 if (sysfs_set_str(sra, NULL,
1326 "array_state", "read-auto") == 0) {
1327 if (verbose >= 0)
1328 fprintf(stderr, Name
1329 ": started array %s\n",
339c2d6c 1330 me->path ?: devnum2devname(me->devnum));
8382f19b
NB
1331 } else {
1332 fprintf(stderr, Name
1333 ": failed to start array %s: %s\n",
339c2d6c
N
1334 me->path ?: devnum2devname(me->devnum),
1335 strerror(errno));
8382f19b
NB
1336 rv = 1;
1337 }
1338 }
1339 }
1340 return rv;
1341}
598f0d58 1342
1771a6e2
N
1343static char *container2devname(char *devname)
1344{
1345 char *mdname = NULL;
1346
1347 if (devname[0] == '/') {
1348 int fd = open(devname, O_RDONLY);
1349 if (fd >= 0) {
1350 mdname = devnum2devname(fd2devnum(fd));
1351 close(fd);
1352 }
1353 } else {
1354 int uuid[4];
1355 struct map_ent *mp, *map = NULL;
1356
1357 if (!parse_uuid(devname, uuid))
1358 return mdname;
1359 mp = map_by_uuid(&map, uuid);
1360 if (mp)
1361 mdname = devnum2devname(mp->devnum);
1362 map_free(map);
1363 }
1364
1365 return mdname;
1366}
1367
47c74f3f
N
1368static int Incremental_container(struct supertype *st, char *devname,
1369 char *homehost, int verbose,
1370 int runstop, int autof)
598f0d58
NB
1371{
1372 /* Collect the contents of this container and for each
1373 * array, choose a device name and assemble the array.
1374 */
1375
47c74f3f 1376 struct mdinfo *list;
598f0d58 1377 struct mdinfo *ra;
ad5bc697 1378 struct map_ent *map = NULL;
47c74f3f
N
1379 struct mdinfo info;
1380 int trustworthy;
fa56eddb 1381 struct mddev_ident *match;
47c74f3f
N
1382 int rv = 0;
1383
1384 memset(&info, 0, sizeof(info));
1385 st->ss->getinfo_super(st, &info, NULL);
1386
1387 if ((runstop > 0 && info.container_enough >= 0) ||
1388 info.container_enough > 0)
1389 /* pass */;
1390 else {
1391 if (verbose)
1392 fprintf(stderr, Name ": not enough devices to start the container\n");
1393 return 0;
1394 }
1395
1396 match = search_mdstat(st, &info, devname, verbose, &rv);
1397 if (match == NULL && rv == 2)
1398 return rv;
1399
1400 /* Need to compute 'trustworthy' */
1401 if (match)
1402 trustworthy = LOCAL;
1403 else if (st->ss->match_home(st, homehost) == 1)
1404 trustworthy = LOCAL;
1405 else if (st->ss->match_home(st, "any") == 1)
1406 trustworthy = LOCAL;
1407 else
1408 trustworthy = FOREIGN;
ad5bc697 1409
47c74f3f 1410 list = st->ss->container_content(st, NULL);
93c861ee
DL
1411 if (map_lock(&map))
1412 fprintf(stderr, Name ": failed to get exclusive lock on "
1413 "mapfile\n");
598f0d58
NB
1414
1415 for (ra = list ; ra ; ra = ra->next) {
598f0d58
NB
1416 int mdfd;
1417 char chosen_name[1024];
ad5bc697 1418 struct map_ent *mp;
fa56eddb 1419 struct mddev_ident *match = NULL;
598f0d58 1420
c5afc314 1421 mp = map_by_uuid(&map, ra->uuid);
d7288ddc 1422
30926600
N
1423 if (mp) {
1424 mdfd = open_dev(mp->devnum);
339c2d6c
N
1425 if (mp->path)
1426 strcpy(chosen_name, mp->path);
1427 else
1428 strcpy(chosen_name, devnum2devname(mp->devnum));
30926600 1429 } else {
dbb44303 1430
112cace6 1431 /* Check in mdadm.conf for container == devname and
dbb44303
N
1432 * member == ra->text_version after second slash.
1433 */
1434 char *sub = strchr(ra->text_version+1, '/');
fa56eddb 1435 struct mddev_ident *array_list;
dbb44303
N
1436 if (sub) {
1437 sub++;
1438 array_list = conf_get_ident(NULL);
1439 } else
1440 array_list = NULL;
1441 for(; array_list ; array_list = array_list->next) {
dbb44303
N
1442 char *dn;
1443 if (array_list->member == NULL ||
1444 array_list->container == NULL)
1445 continue;
1446 if (strcmp(array_list->member, sub) != 0)
1447 continue;
71d60c48
DW
1448 if (array_list->uuid_set &&
1449 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1450 continue;
1771a6e2
N
1451 dn = container2devname(array_list->container);
1452 if (dn == NULL)
dbb44303 1453 continue;
dbb44303
N
1454 if (strncmp(dn, ra->text_version+1,
1455 strlen(dn)) != 0 ||
1456 ra->text_version[strlen(dn)+1] != '/') {
1457 free(dn);
1458 continue;
1459 }
1460 free(dn);
1461 /* we have a match */
1462 match = array_list;
71d60c48
DW
1463 if (verbose>0)
1464 fprintf(stderr, Name ": match found for member %s\n",
1465 array_list->member);
dbb44303
N
1466 break;
1467 }
dbb44303 1468
112cace6
N
1469 if (match && match->devname &&
1470 strcasecmp(match->devname, "<ignore>") == 0) {
1471 if (verbose > 0)
1472 fprintf(stderr, Name ": array %s/%s is "
1473 "explicitly ignored by mdadm.conf\n",
1474 match->container, match->member);
1475 return 2;
1476 }
05833051
N
1477 if (match)
1478 trustworthy = LOCAL;
112cace6 1479
30926600
N
1480 mdfd = create_mddev(match ? match->devname : NULL,
1481 ra->name,
1482 autof,
1483 trustworthy,
1484 chosen_name);
1485 }
598f0d58
NB
1486
1487 if (mdfd < 0) {
1488 fprintf(stderr, Name ": failed to open %s: %s.\n",
1489 chosen_name, strerror(errno));
1490 return 2;
1491 }
1492
03b7f6c6
N
1493 assemble_container_content(st, mdfd, ra, runstop,
1494 chosen_name, verbose);
598f0d58 1495 }
ad5bc697 1496 map_unlock(&map);
598f0d58
NB
1497 return 0;
1498}
29ba4804
N
1499
1500/*
1501 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1502 * raid arrays, and if so first fail (if needed) and then remove the device.
1503 *
1504 * @devname - The device we want to remove
950bc344 1505 * @id_path - name as found in /dev/disk/by-path for this device
29ba4804
N
1506 *
1507 * Note: the device name must be a kernel name like "sda", so
1508 * that we can find it in /proc/mdstat
1509 */
950bc344 1510int IncrementalRemove(char *devname, char *id_path, int verbose)
29ba4804
N
1511{
1512 int mdfd;
aae3cdc3 1513 int rv;
29ba4804 1514 struct mdstat_ent *ent;
a655e550 1515 struct mddev_dev devlist;
29ba4804 1516
950bc344
PC
1517 if (!id_path)
1518 dprintf(Name ": incremental removal without --path <id_path> "
1519 "lacks the possibility to re-add new device in this "
1520 "port\n");
1521
29ba4804
N
1522 if (strchr(devname, '/')) {
1523 fprintf(stderr, Name ": incremental removal requires a "
1524 "kernel device name, not a file: %s\n", devname);
1525 return 1;
1526 }
1527 ent = mdstat_by_component(devname);
1528 if (!ent) {
1529 fprintf(stderr, Name ": %s does not appear to be a component "
1530 "of any array\n", devname);
1531 return 1;
1532 }
1533 mdfd = open_dev(ent->devnum);
1534 if (mdfd < 0) {
1535 fprintf(stderr, Name ": Cannot open array %s!!\n", ent->dev);
08387a04 1536 free_mdstat(ent);
29ba4804
N
1537 return 1;
1538 }
403410eb
PC
1539
1540 if (id_path) {
1541 struct map_ent *map = NULL, *me;
1542 me = map_by_devnum(&map, ent->devnum);
1543 if (me)
1544 policy_save_path(id_path, me);
1545 map_free(map);
1546 }
1547
29ba4804
N
1548 memset(&devlist, 0, sizeof(devlist));
1549 devlist.devname = devname;
1550 devlist.disposition = 'f';
08387a04
N
1551 /* for a container, we must fail each member array */
1552 if (ent->metadata_version &&
1553 strncmp(ent->metadata_version, "external:", 9) == 0) {
1554 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1555 struct mdstat_ent *memb;
1556 for (memb = mdstat ; memb ; memb = memb->next)
1557 if (is_container_member(memb, ent->dev)) {
1558 int subfd = open_dev(memb->devnum);
1559 if (subfd >= 0) {
1560 Manage_subdevs(memb->dev, subfd,
833bb0f8
N
1561 &devlist, verbose, 0,
1562 NULL);
08387a04
N
1563 close(subfd);
1564 }
1565 }
1566 free_mdstat(mdstat);
1567 } else
833bb0f8 1568 Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL);
29ba4804 1569 devlist.disposition = 'r';
833bb0f8 1570 rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL);
aae3cdc3 1571 close(mdfd);
08387a04 1572 free_mdstat(ent);
aae3cdc3 1573 return rv;
29ba4804 1574}