]> 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
New RESHAPE_NO_BACKUP flag to track when backup action is needed.
[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) 2006-2012 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 struct context *c);
47
48 int Incremental(char *devname, struct context *c,
49 struct supertype *st)
50 {
51 /* Add this device to an array, creating the array if necessary
52 * and starting the array if sensible or - if runstop>0 - if possible.
53 *
54 * This has several steps:
55 *
56 * 1/ Check if device is permitted by mdadm.conf, reject if not.
57 * 2/ Find metadata, reject if none appropriate (check
58 * version/name from args)
59 * 3/ Check if there is a match in mdadm.conf
60 * 3a/ if not, check for homehost match. If no match, assemble as
61 * a 'foreign' array.
62 * 4/ Determine device number.
63 * - If in mdadm.conf with std name, use that
64 * - UUID in /var/run/mdadm.map use that
65 * - If name is suggestive, use that. unless in use with different uuid.
66 * - Choose a free, high number.
67 * - Use a partitioned device unless strong suggestion not to.
68 * e.g. auto=md
69 * Don't choose partitioned for containers.
70 * 5/ Find out if array already exists
71 * 5a/ if it does not
72 * - choose a name, from mdadm.conf or 'name' field in array.
73 * - create the array
74 * - add the device
75 * 5b/ if it does
76 * - check one drive in array to make sure metadata is a reasonably
77 * close match. Reject if not (e.g. different type)
78 * - add the device
79 * 6/ Make sure /var/run/mdadm.map contains this array.
80 * 7/ Is there enough devices to possibly start the array?
81 * For a container, this means running Incremental_container.
82 * 7a/ if not, finish with success.
83 * 7b/ if yes,
84 * - read all metadata and arrange devices like -A does
85 * - if number of OK devices match expected, or -R and there are enough,
86 * start the array (auto-readonly).
87 */
88 struct stat stb;
89 struct mdinfo info, dinfo;
90 struct mdinfo *sra = NULL, *d;
91 struct mddev_ident *match;
92 char chosen_name[1024];
93 int rv = 1;
94 struct map_ent *mp, *map = NULL;
95 int dfd = -1, mdfd = -1;
96 char *avail = NULL;
97 int active_disks;
98 int trustworthy;
99 char *name_to_use;
100 mdu_array_info_t ainf;
101 struct dev_policy *policy = NULL;
102 struct map_ent target_array;
103 int have_target;
104
105 struct createinfo *ci = conf_get_create_info();
106
107 if (stat(devname, &stb) < 0) {
108 if (c->verbose >= 0)
109 pr_err("stat failed for %s: %s.\n",
110 devname, strerror(errno));
111 return rv;
112 }
113 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
114 if (c->verbose >= 0)
115 pr_err("%s is not a block device.\n",
116 devname);
117 return rv;
118 }
119 dfd = dev_open(devname, O_RDONLY|O_EXCL);
120 if (dfd < 0) {
121 if (c->verbose >= 0)
122 pr_err("cannot open %s: %s.\n",
123 devname, strerror(errno));
124 return rv;
125 }
126 /* If the device is a container, we do something very different */
127 if (must_be_container(dfd)) {
128 if (!st)
129 st = super_by_fd(dfd, NULL);
130 if (st && st->ss->load_container)
131 rv = st->ss->load_container(st, dfd, NULL);
132
133 close(dfd);
134 if (!rv && st->ss->container_content) {
135 if (map_lock(&map))
136 pr_err("failed to get "
137 "exclusive lock on mapfile\n");
138 rv = Incremental_container(st, devname, c);
139 map_unlock(&map);
140 return rv;
141 }
142
143 pr_err("%s is not part of an md array.\n",
144 devname);
145 return rv;
146 }
147
148 /* 1/ Check if device is permitted by mdadm.conf */
149
150 if (!conf_test_dev(devname)) {
151 if (c->verbose >= 0)
152 pr_err("%s not permitted by mdadm.conf.\n",
153 devname);
154 goto out;
155 }
156
157 /* 2/ Find metadata, reject if none appropriate (check
158 * version/name from args) */
159
160 if (fstat(dfd, &stb) < 0) {
161 if (c->verbose >= 0)
162 pr_err("fstat failed for %s: %s.\n",
163 devname, strerror(errno));
164 goto out;
165 }
166 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
167 if (c->verbose >= 0)
168 pr_err("%s is not a block device.\n",
169 devname);
170 goto out;
171 }
172
173 dinfo.disk.major = major(stb.st_rdev);
174 dinfo.disk.minor = minor(stb.st_rdev);
175
176 policy = disk_policy(&dinfo);
177 have_target = policy_check_path(&dinfo, &target_array);
178
179 if (st == NULL && (st = guess_super(dfd)) == NULL) {
180 if (c->verbose >= 0)
181 pr_err("no recognisable superblock on %s.\n",
182 devname);
183 rv = try_spare(devname, &dfd, policy,
184 have_target ? &target_array : NULL,
185 st, c->verbose);
186 goto out;
187 }
188 if (st->ss->compare_super == NULL ||
189 st->ss->load_super(st, dfd, NULL)) {
190 if (c->verbose >= 0)
191 pr_err("no RAID superblock on %s.\n",
192 devname);
193 rv = try_spare(devname, &dfd, policy,
194 have_target ? &target_array : NULL,
195 st, c->verbose);
196 free(st);
197 goto out;
198 }
199 close (dfd); dfd = -1;
200
201 st->ss->getinfo_super(st, &info, NULL);
202
203 /* 3/ Check if there is a match in mdadm.conf */
204 match = conf_match(st, &info, devname, c->verbose, &rv);
205 if (!match && rv == 2)
206 goto out;
207
208 if (match && match->devname
209 && strcasecmp(match->devname, "<ignore>") == 0) {
210 if (c->verbose >= 0)
211 pr_err("array containing %s is explicitly"
212 " ignored by mdadm.conf\n",
213 devname);
214 goto out;
215 }
216
217 /* 3a/ if not, check for homehost match. If no match, continue
218 * but don't trust the 'name' in the array. Thus a 'random' minor
219 * number will be assigned, and the device name will be based
220 * on that. */
221 if (match)
222 trustworthy = LOCAL;
223 else if (st->ss->match_home(st, c->homehost) == 1)
224 trustworthy = LOCAL;
225 else if (st->ss->match_home(st, "any") == 1)
226 trustworthy = LOCAL_ANY;
227 else
228 trustworthy = FOREIGN;
229
230 if (!match && !conf_test_metadata(st->ss->name, policy,
231 (trustworthy == LOCAL))) {
232 if (c->verbose >= 1)
233 pr_err("%s has metadata type %s for which "
234 "auto-assembly is disabled\n",
235 devname, st->ss->name);
236 goto out;
237 }
238 if (trustworthy == LOCAL_ANY)
239 trustworthy = LOCAL;
240
241 /* There are three possible sources for 'autof': command line,
242 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
243 * ARRAY takes precedence, then command line, then
244 * CREATE.
245 */
246 if (match && match->autof)
247 c->autof = match->autof;
248 if (c->autof == 0)
249 c->autof = ci->autof;
250
251 name_to_use = info.name;
252 if (name_to_use[0] == 0 &&
253 info.array.level == LEVEL_CONTAINER) {
254 name_to_use = info.text_version;
255 trustworthy = METADATA;
256 }
257 if (name_to_use[0] && trustworthy != LOCAL &&
258 ! c->require_homehost &&
259 conf_name_is_free(name_to_use))
260 trustworthy = LOCAL;
261
262 /* strip "hostname:" prefix from name if we have decided
263 * to treat it as LOCAL
264 */
265 if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
266 name_to_use = strchr(name_to_use, ':')+1;
267
268 /* 4/ Check if array exists.
269 */
270 if (map_lock(&map))
271 pr_err("failed to get exclusive lock on "
272 "mapfile\n");
273 mp = map_by_uuid(&map, info.uuid);
274 if (mp)
275 mdfd = open_dev(mp->devnum);
276 else
277 mdfd = -1;
278
279 if (mdfd < 0) {
280
281 /* Couldn't find an existing array, maybe make a new one */
282 mdfd = create_mddev(match ? match->devname : NULL,
283 name_to_use, c->autof, trustworthy, chosen_name);
284
285 if (mdfd < 0)
286 goto out_unlock;
287
288 sysfs_init(&info, mdfd, 0);
289
290 if (set_array_info(mdfd, st, &info) != 0) {
291 pr_err("failed to set array info for %s: %s\n",
292 chosen_name, strerror(errno));
293 rv = 2;
294 goto out_unlock;
295 }
296
297 dinfo = info;
298 dinfo.disk.major = major(stb.st_rdev);
299 dinfo.disk.minor = minor(stb.st_rdev);
300 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
301 pr_err("failed to add %s to %s: %s.\n",
302 devname, chosen_name, strerror(errno));
303 ioctl(mdfd, STOP_ARRAY, 0);
304 rv = 2;
305 goto out_unlock;
306 }
307 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
308 GET_OFFSET | GET_SIZE));
309
310 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
311 /* It really should be 'none' - must be old buggy
312 * kernel, and mdadm -I may not be able to complete.
313 * So reject it.
314 */
315 ioctl(mdfd, STOP_ARRAY, NULL);
316 pr_err("You have an old buggy kernel which cannot support\n"
317 " --incremental reliably. Aborting.\n");
318 rv = 2;
319 goto out_unlock;
320 }
321 info.array.working_disks = 1;
322 /* 6/ Make sure /var/run/mdadm.map contains this array. */
323 map_update(&map, fd2devnum(mdfd),
324 info.text_version,
325 info.uuid, chosen_name);
326 } else {
327 /* 5b/ if it does */
328 /* - check one drive in array to make sure metadata is a reasonably */
329 /* close match. Reject if not (e.g. different type) */
330 /* - add the device */
331 char dn[20];
332 int dfd2;
333 int err;
334 struct supertype *st2;
335 struct mdinfo info2, *d;
336
337 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
338 GET_OFFSET | GET_SIZE));
339
340 if (mp->path)
341 strcpy(chosen_name, mp->path);
342 else
343 strcpy(chosen_name, devnum2devname(mp->devnum));
344
345 /* It is generally not OK to add non-spare drives to a
346 * running array as they are probably missing because
347 * they failed. However if runstop is 1, then the
348 * array was possibly started early and our best bet is
349 * to add this anyway.
350 * Also if action policy is re-add or better we allow
351 * re-add.
352 * This doesn't apply to containers as the 'non-spare'
353 * flag has a different meaning. The test has to happen
354 * at the device level there
355 */
356 if (!st->ss->external
357 && (info.disk.state & (1<<MD_DISK_SYNC)) != 0
358 && ! policy_action_allows(policy, st->ss->name,
359 act_re_add)
360 && c->runstop < 1) {
361 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
362 pr_err("not adding %s to active array (without --run) %s\n",
363 devname, chosen_name);
364 rv = 2;
365 goto out_unlock;
366 }
367 }
368 if (!sra) {
369 rv = 2;
370 goto out_unlock;
371 }
372 if (sra->devs) {
373 sprintf(dn, "%d:%d", sra->devs->disk.major,
374 sra->devs->disk.minor);
375 dfd2 = dev_open(dn, O_RDONLY);
376 if (dfd2 < 0) {
377 pr_err("unable to open %s\n", devname);
378 rv = 2;
379 goto out_unlock;
380 }
381 st2 = dup_super(st);
382 if (st2->ss->load_super(st2, dfd2, NULL) ||
383 st->ss->compare_super(st, st2) != 0) {
384 pr_err("metadata mismatch between %s and "
385 "chosen array %s\n",
386 devname, chosen_name);
387 close(dfd2);
388 rv = 2;
389 goto out_unlock;
390 }
391 close(dfd2);
392 st2->ss->getinfo_super(st2, &info2, NULL);
393 st2->ss->free_super(st2);
394 if (info.array.level != info2.array.level ||
395 memcmp(info.uuid, info2.uuid, 16) != 0 ||
396 info.array.raid_disks != info2.array.raid_disks) {
397 pr_err("unexpected difference between %s and %s.\n",
398 chosen_name, devname);
399 rv = 2;
400 goto out_unlock;
401 }
402 }
403 info.disk.major = major(stb.st_rdev);
404 info.disk.minor = minor(stb.st_rdev);
405 /* add disk needs to know about containers */
406 if (st->ss->external)
407 sra->array.level = LEVEL_CONTAINER;
408 err = add_disk(mdfd, st, sra, &info);
409 if (err < 0 && errno == EBUSY) {
410 /* could be another device present with the same
411 * disk.number. Find and reject any such
412 */
413 find_reject(mdfd, st, sra, info.disk.number,
414 info.events, c->verbose, chosen_name);
415 err = add_disk(mdfd, st, sra, &info);
416 }
417 if (err < 0) {
418 pr_err("failed to add %s to %s: %s.\n",
419 devname, chosen_name, strerror(errno));
420 rv = 2;
421 goto out_unlock;
422 }
423 info.array.working_disks = 0;
424 for (d = sra->devs; d; d=d->next)
425 info.array.working_disks ++;
426
427 }
428
429 /* 7/ Is there enough devices to possibly start the array? */
430 /* 7a/ if not, finish with success. */
431 if (info.array.level == LEVEL_CONTAINER) {
432 int devnum = devnum; /* defined and used iff ->external */
433 /* Try to assemble within the container */
434 sysfs_uevent(sra, "change");
435 if (c->verbose >= 0)
436 pr_err("container %s now has %d device%s\n",
437 chosen_name, info.array.working_disks,
438 info.array.working_disks == 1?"":"s");
439 wait_for(chosen_name, mdfd);
440 if (st->ss->external)
441 devnum = fd2devnum(mdfd);
442 if (st->ss->load_container)
443 rv = st->ss->load_container(st, mdfd, NULL);
444 close(mdfd);
445 sysfs_free(sra);
446 if (!rv)
447 rv = Incremental_container(st, chosen_name, c);
448 map_unlock(&map);
449 if (rv == 1)
450 /* Don't fail the whole -I if a subarray didn't
451 * have enough devices to start yet
452 */
453 rv = 0;
454 /* after spare is added, ping monitor for external metadata
455 * so that it can eg. try to rebuild degraded array */
456 if (st->ss->external)
457 ping_monitor_by_id(devnum);
458 return rv;
459 }
460
461 /* We have added something to the array, so need to re-read the
462 * state. Eventually this state should be kept up-to-date as
463 * things change.
464 */
465 sysfs_free(sra);
466 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
467 GET_OFFSET | GET_SIZE));
468 active_disks = count_active(st, sra, mdfd, &avail, &info);
469 if (enough(info.array.level, info.array.raid_disks,
470 info.array.layout, info.array.state & 1,
471 avail) == 0) {
472 if (c->verbose >= 0)
473 pr_err("%s attached to %s, not enough to start (%d).\n",
474 devname, chosen_name, active_disks);
475 rv = 0;
476 goto out_unlock;
477 }
478
479 /* 7b/ if yes, */
480 /* - if number of OK devices match expected, or -R and there */
481 /* are enough, */
482 /* + add any bitmap file */
483 /* + start the array (auto-readonly). */
484
485 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
486 if (c->verbose >= 0)
487 pr_err("%s attached to %s which is already active.\n",
488 devname, chosen_name);
489 rv = 0;
490 goto out_unlock;
491 }
492
493 map_unlock(&map);
494 if (c->runstop > 0 || active_disks >= info.array.working_disks) {
495 struct mdinfo *dsk;
496 /* Let's try to start it */
497
498 if (info.reshape_active && !(info.reshape_active & RESHAPE_NO_BACKUP)) {
499 fprintf(stderr, Name
500 ": %s: This array is being reshaped and cannot be started\n"
501 " by --incremental. Please use --assemble\n",
502 chosen_name);
503 goto out;
504 }
505 if (match && match->bitmap_file) {
506 int bmfd = open(match->bitmap_file, O_RDWR);
507 if (bmfd < 0) {
508 pr_err("Could not open bitmap file %s.\n",
509 match->bitmap_file);
510 goto out;
511 }
512 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
513 close(bmfd);
514 pr_err("Failed to set bitmapfile for %s.\n",
515 chosen_name);
516 goto out;
517 }
518 close(bmfd);
519 }
520 /* Need to remove from the array any devices which
521 * 'count_active' discerned were too old or inappropriate
522 */
523 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
524 if (d->disk.state & (1<<MD_DISK_REMOVED))
525 remove_disk(mdfd, st, sra, d);
526
527 if ((sra == NULL || active_disks >= info.array.working_disks)
528 && trustworthy != FOREIGN)
529 rv = ioctl(mdfd, RUN_ARRAY, NULL);
530 else
531 rv = sysfs_set_str(sra, NULL,
532 "array_state", "read-auto");
533 if (rv == 0) {
534 if (c->verbose >= 0)
535 pr_err("%s attached to %s, which has been started.\n",
536 devname, chosen_name);
537 rv = 0;
538 wait_for(chosen_name, mdfd);
539 /* We just started the array, so some devices
540 * might have been evicted from the array
541 * because their event counts were too old.
542 * If the action=re-add policy is in-force for
543 * those devices we should re-add them now.
544 */
545 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
546 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
547 add_disk(mdfd, st, sra, dsk) == 0)
548 pr_err("%s re-added to %s\n",
549 dsk->sys_name, chosen_name);
550 }
551 } else {
552 pr_err("%s attached to %s, but failed to start: %s.\n",
553 devname, chosen_name, strerror(errno));
554 rv = 1;
555 }
556 } else {
557 if (c->verbose >= 0)
558 pr_err("%s attached to %s, not enough to start safely.\n",
559 devname, chosen_name);
560 rv = 0;
561 }
562 out:
563 free(avail);
564 if (dfd >= 0)
565 close(dfd);
566 if (mdfd >= 0)
567 close(mdfd);
568 if (policy)
569 dev_policy_free(policy);
570 if (sra)
571 sysfs_free(sra);
572 return rv;
573 out_unlock:
574 map_unlock(&map);
575 goto out;
576 }
577
578 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
579 int number, __u64 events, int verbose,
580 char *array_name)
581 {
582 /* Find a device attached to this array with a disk.number of number
583 * and events less than the passed events, and remove the device.
584 */
585 struct mdinfo *d;
586 mdu_array_info_t ra;
587
588 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
589 return; /* not safe to remove from active arrays
590 * without thinking more */
591
592 for (d = sra->devs; d ; d = d->next) {
593 char dn[10];
594 int dfd;
595 struct mdinfo info;
596 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
597 dfd = dev_open(dn, O_RDONLY);
598 if (dfd < 0)
599 continue;
600 if (st->ss->load_super(st, dfd, NULL)) {
601 close(dfd);
602 continue;
603 }
604 st->ss->getinfo_super(st, &info, NULL);
605 st->ss->free_super(st);
606 close(dfd);
607
608 if (info.disk.number != number ||
609 info.events >= events)
610 continue;
611
612 if (d->disk.raid_disk > -1)
613 sysfs_set_str(sra, d, "slot", "none");
614 if (sysfs_set_str(sra, d, "state", "remove") == 0)
615 if (verbose >= 0)
616 pr_err("removing old device %s from %s\n",
617 d->sys_name+4, array_name);
618 }
619 }
620
621 static int count_active(struct supertype *st, struct mdinfo *sra,
622 int mdfd, char **availp,
623 struct mdinfo *bestinfo)
624 {
625 /* count how many devices in sra think they are active */
626 struct mdinfo *d;
627 int cnt = 0;
628 __u64 max_events = 0;
629 char *avail = NULL;
630 int *best = NULL;
631 char *devmap = NULL;
632 int numdevs = 0;
633 int devnum;
634 int b, i;
635 int raid_disks = 0;
636
637 if (!sra)
638 return 0;
639
640 for (d = sra->devs ; d ; d = d->next)
641 numdevs++;
642 for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
643 char dn[30];
644 int dfd;
645 int ok;
646 struct mdinfo info;
647
648 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
649 dfd = dev_open(dn, O_RDONLY);
650 if (dfd < 0)
651 continue;
652 ok = st->ss->load_super(st, dfd, NULL);
653 close(dfd);
654 if (ok != 0)
655 continue;
656 info.array.raid_disks = raid_disks;
657 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
658 if (!avail) {
659 raid_disks = info.array.raid_disks;
660 avail = xcalloc(raid_disks, 1);
661 *availp = avail;
662
663 best = xcalloc(raid_disks, sizeof(int));
664 devmap = xcalloc(raid_disks, numdevs);
665
666 st->ss->getinfo_super(st, &info, devmap);
667 }
668
669 if (info.disk.state & (1<<MD_DISK_SYNC))
670 {
671 if (cnt == 0) {
672 cnt++;
673 max_events = info.events;
674 avail[info.disk.raid_disk] = 2;
675 best[info.disk.raid_disk] = devnum;
676 st->ss->getinfo_super(st, bestinfo, NULL);
677 } else if (info.events == max_events) {
678 avail[info.disk.raid_disk] = 2;
679 best[info.disk.raid_disk] = devnum;
680 } else if (info.events == max_events-1) {
681 if (avail[info.disk.raid_disk] == 0) {
682 avail[info.disk.raid_disk] = 1;
683 best[info.disk.raid_disk] = devnum;
684 }
685 } else if (info.events < max_events - 1)
686 ;
687 else if (info.events == max_events+1) {
688 int i;
689 max_events = info.events;
690 for (i = 0; i < raid_disks; i++)
691 if (avail[i])
692 avail[i]--;
693 avail[info.disk.raid_disk] = 2;
694 best[info.disk.raid_disk] = devnum;
695 st->ss->getinfo_super(st, bestinfo, NULL);
696 } else { /* info.events much bigger */
697 memset(avail, 0, raid_disks);
698 max_events = info.events;
699 avail[info.disk.raid_disk] = 2;
700 best[info.disk.raid_disk] = devnum;
701 st->ss->getinfo_super(st, bestinfo, NULL);
702 }
703 }
704 st->ss->free_super(st);
705 }
706 if (!avail)
707 return 0;
708 /* We need to reject any device that thinks the best device is
709 * failed or missing */
710 for (b = 0; b < raid_disks; b++)
711 if (avail[b] == 2)
712 break;
713 cnt = 0;
714 for (i = 0 ; i < raid_disks ; i++) {
715 if (i != b && avail[i])
716 if (devmap[raid_disks * best[i] + b] == 0) {
717 /* This device thinks 'b' is failed -
718 * don't use it */
719 devnum = best[i];
720 for (d=sra->devs ; devnum; d = d->next)
721 devnum--;
722 d->disk.state |= (1 << MD_DISK_REMOVED);
723 avail[i] = 0;
724 }
725 if (avail[i])
726 cnt++;
727 }
728 free(best);
729 free(devmap);
730 return cnt;
731 }
732
733 /* test if container has degraded member(s) */
734 static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
735 {
736 mdu_array_info_t array;
737 int afd;
738 int max_degraded = 0;
739
740 for(; map; map = map->next) {
741 if (!is_subarray(map->metadata) ||
742 devname2devnum(map->metadata+1) != me->devnum)
743 continue;
744 afd = open_dev(map->devnum);
745 if (afd < 0)
746 continue;
747 /* most accurate information regarding array degradation */
748 if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) {
749 int degraded = array.raid_disks - array.active_disks -
750 array.spare_disks;
751 if (degraded > max_degraded)
752 max_degraded = degraded;
753 }
754 close(afd);
755 }
756 return (max_degraded);
757 }
758
759 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
760 struct map_ent *target, int bare,
761 struct supertype *st, int verbose)
762 {
763 /* This device doesn't have any md metadata
764 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
765 * If 'st' is not set, then we only know that some metadata allows this,
766 * others possibly don't.
767 * So look for a container or array to attach the device to.
768 * Prefer 'target' if that is set and the array is found.
769 *
770 * If st is set, then only arrays of that type are considered
771 * Return 0 on success, or some exit code on failure, probably 1.
772 */
773 int rv = 1;
774 struct stat stb;
775 struct map_ent *mp, *map = NULL;
776 struct mdinfo *chosen = NULL;
777 int dfd = *dfdp;
778
779 if (fstat(dfd, &stb) != 0)
780 return 1;
781
782 /*
783 * Now we need to find a suitable array to add this to.
784 * We only accept arrays that:
785 * - match 'st'
786 * - are in the same domains as the device
787 * - are of an size for which the device will be useful
788 * and we choose the one that is the most degraded
789 */
790
791 if (map_lock(&map)) {
792 pr_err("failed to get exclusive lock on "
793 "mapfile\n");
794 return 1;
795 }
796 for (mp = map ; mp ; mp = mp->next) {
797 struct supertype *st2;
798 struct domainlist *dl = NULL;
799 struct mdinfo *sra;
800 unsigned long long devsize;
801 unsigned long long component_size = 0;
802
803 if (is_subarray(mp->metadata))
804 continue;
805 if (st) {
806 st2 = st->ss->match_metadata_desc(mp->metadata);
807 if (!st2 ||
808 (st->minor_version >= 0 &&
809 st->minor_version != st2->minor_version)) {
810 if (verbose > 1)
811 pr_err("not adding %s to %s as metadata type doesn't match\n",
812 devname, mp->path);
813 free(st2);
814 continue;
815 }
816 free(st2);
817 }
818 sra = sysfs_read(-1, mp->devnum,
819 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
820 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
821 if (!sra) {
822 /* Probably a container - no degraded info */
823 sra = sysfs_read(-1, mp->devnum,
824 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
825 GET_COMPONENT|GET_VERSION);
826 if (sra)
827 sra->array.failed_disks = -1;
828 }
829 if (!sra)
830 continue;
831 if (st == NULL) {
832 int i;
833 st2 = NULL;
834 for(i = 0; !st2 && superlist[i]; i++)
835 st2 = superlist[i]->match_metadata_desc(
836 sra->text_version);
837 if (!st2) {
838 if (verbose > 1)
839 pr_err("not adding %s to %s"
840 " as metadata not recognised.\n",
841 devname, mp->path);
842 goto next;
843 }
844 /* Need to double check the 'act_spare' permissions applies
845 * to this metadata.
846 */
847 if (!policy_action_allows(pol, st2->ss->name, act_spare))
848 goto next;
849 if (!bare && !policy_action_allows(pol, st2->ss->name,
850 act_spare_same_slot))
851 goto next;
852 } else
853 st2 = st;
854 /* update number of failed disks for mostly degraded
855 * container member */
856 if (sra->array.failed_disks == -1)
857 sra->array.failed_disks = container_members_max_degradation(map, mp);
858
859 get_dev_size(dfd, NULL, &devsize);
860 if (sra->component_size == 0) {
861 /* true for containers, here we must read superblock
862 * to obtain minimum spare size */
863 struct supertype *st3 = dup_super(st2);
864 int mdfd = open_dev(mp->devnum);
865 if (mdfd < 0) {
866 free(st3);
867 goto next;
868 }
869 if (st3->ss->load_container &&
870 !st3->ss->load_container(st3, mdfd, mp->path)) {
871 component_size = st3->ss->min_acceptable_spare_size(st3);
872 st3->ss->free_super(st3);
873 }
874 free(st3);
875 close(mdfd);
876 }
877 if ((sra->component_size > 0 &&
878 st2->ss->avail_size(st2, devsize,
879 sra->devs
880 ? sra->devs->data_offset
881 : INVALID_SECTORS)
882 < sra->component_size)
883 ||
884 (sra->component_size == 0 && devsize < component_size)) {
885 if (verbose > 1)
886 pr_err("not adding %s to %s as it is too small\n",
887 devname, mp->path);
888 goto next;
889 }
890 /* test against target.
891 * If 'target' is set and 'bare' is false, we only accept
892 * arrays/containers that match 'target'.
893 * If 'target' is set and 'bare' is true, we prefer the
894 * array which matches 'target'.
895 * target is considered only if we deal with degraded array
896 */
897 if (target && policy_action_allows(pol, st2->ss->name,
898 act_spare_same_slot)) {
899 if (strcmp(target->metadata, mp->metadata) == 0 &&
900 memcmp(target->uuid, mp->uuid,
901 sizeof(target->uuid)) == 0 &&
902 sra->array.failed_disks > 0) {
903 /* This is our target!! */
904 if (chosen)
905 sysfs_free(chosen);
906 chosen = sra;
907 sra = NULL;
908 /* skip to end so we don't check any more */
909 while (mp->next)
910 mp = mp->next;
911 goto next;
912 }
913 /* not our target */
914 if (!bare)
915 goto next;
916 }
917
918 dl = domain_from_array(sra, st2->ss->name);
919 if (domain_test(dl, pol, st2->ss->name) != 1) {
920 /* domain test fails */
921 if (verbose > 1)
922 pr_err("not adding %s to %s as"
923 " it is not in a compatible domain\n",
924 devname, mp->path);
925
926 goto next;
927 }
928 /* all tests passed, OK to add to this array */
929 if (!chosen) {
930 chosen = sra;
931 sra = NULL;
932 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
933 sysfs_free(chosen);
934 chosen = sra;
935 sra = NULL;
936 }
937 next:
938 if (sra)
939 sysfs_free(sra);
940 if (st != st2)
941 free(st2);
942 if (dl)
943 domain_free(dl);
944 }
945 if (chosen) {
946 /* add current device to chosen array as a spare */
947 int mdfd = open_dev(devname2devnum(chosen->sys_name));
948 if (mdfd >= 0) {
949 struct mddev_dev devlist;
950 char devname[20];
951 devlist.next = NULL;
952 devlist.used = 0;
953 devlist.writemostly = 0;
954 devlist.devname = devname;
955 sprintf(devname, "%d:%d", major(stb.st_rdev),
956 minor(stb.st_rdev));
957 devlist.disposition = 'a';
958 close(dfd);
959 *dfdp = -1;
960 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
961 -1, 0, NULL, 0);
962 close(mdfd);
963 }
964 if (verbose > 0) {
965 if (rv == 0)
966 pr_err("added %s as spare for %s\n",
967 devname, chosen->sys_name);
968 else
969 pr_err("failed to add %s as spare for %s\n",
970 devname, chosen->sys_name);
971 }
972 sysfs_free(chosen);
973 }
974 map_unlock(&map);
975 return rv;
976 }
977
978 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
979 struct supertype *st, int verbose)
980 {
981 /* we know that at least one partition virtual-metadata is
982 * allowed to incorporate spares like this device. We need to
983 * find a suitable device to copy partition information from.
984 *
985 * Getting a list of all disk (not partition) devices is
986 * slightly non-trivial. We could look at /sys/block, but
987 * that is theoretically due to be removed. Maybe best to use
988 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
989 * we depend on this directory of 'path' info. But that fails
990 * to find loop devices and probably others. Maybe don't
991 * worry about that, they aren't the real target.
992 *
993 * So: check things in /dev/disk/by-path to see if they are in
994 * a compatible domain, then load the partition table and see
995 * if it is OK for the new device, and choose the largest
996 * partition table that fits.
997 */
998 DIR *dir;
999 struct dirent *de;
1000 char *chosen = NULL;
1001 unsigned long long chosen_size = 0;
1002 struct supertype *chosen_st = NULL;
1003 int fd;
1004
1005 dir = opendir("/dev/disk/by-path");
1006 if (!dir)
1007 return 1;
1008 while ((de = readdir(dir)) != NULL) {
1009 char *ep;
1010 struct dev_policy *pol2 = NULL;
1011 struct domainlist *domlist = NULL;
1012 int fd = -1;
1013 struct mdinfo info;
1014 struct supertype *st2 = NULL;
1015 char *devname = NULL;
1016 unsigned long long devsectors;
1017
1018 if (de->d_ino == 0 ||
1019 de->d_name[0] == '.' ||
1020 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1021 goto next;
1022
1023 ep = de->d_name + strlen(de->d_name);
1024 while (ep > de->d_name &&
1025 isdigit(ep[-1]))
1026 ep--;
1027 if (ep > de->d_name + 5 &&
1028 strncmp(ep-5, "-part", 5) == 0)
1029 /* This is a partition - skip it */
1030 goto next;
1031
1032 pol2 = path_policy(de->d_name, type_disk);
1033
1034 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1035 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1036 /* new device is incompatible with this device. */
1037 goto next;
1038
1039 domain_free(domlist);
1040 domlist = NULL;
1041
1042 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1043 devname = NULL;
1044 goto next;
1045 }
1046 fd = open(devname, O_RDONLY);
1047 if (fd < 0)
1048 goto next;
1049 if (get_dev_size(fd, devname, &devsectors) == 0)
1050 goto next;
1051 devsectors >>= 9;
1052
1053 if (st)
1054 st2 = dup_super(st);
1055 else
1056 st2 = guess_super_type(fd, guess_partitions);
1057 if (st2 == NULL ||
1058 st2->ss->load_super(st2, fd, NULL) < 0)
1059 goto next;
1060
1061 if (!st) {
1062 /* Check domain policy again, this time referring to metadata */
1063 domain_merge(&domlist, pol2, st2->ss->name);
1064 if (domain_test(domlist, pol, st2->ss->name) != 1)
1065 /* Incompatible devices for this metadata type */
1066 goto next;
1067 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1068 /* Some partition types allow sparing, but not
1069 * this one.
1070 */
1071 goto next;
1072 }
1073
1074 st2->ss->getinfo_super(st2, &info, NULL);
1075 if (info.component_size > devsectors)
1076 /* This partitioning doesn't fit in the device */
1077 goto next;
1078
1079 /* This is an acceptable device to copy partition
1080 * metadata from. We could just stop here, but I
1081 * think I want to keep looking incase a larger
1082 * metadata which makes better use of the device can
1083 * be found.
1084 */
1085 if (chosen == NULL ||
1086 chosen_size < info.component_size) {
1087 chosen_size = info.component_size;
1088 free(chosen);
1089 chosen = devname;
1090 devname = NULL;
1091 if (chosen_st) {
1092 chosen_st->ss->free_super(chosen_st);
1093 free(chosen_st);
1094 }
1095 chosen_st = st2;
1096 st2 = NULL;
1097 }
1098
1099 next:
1100 free(devname);
1101 domain_free(domlist);
1102 dev_policy_free(pol2);
1103 if (st2)
1104 st2->ss->free_super(st2);
1105 free(st2);
1106
1107 if (fd >= 0)
1108 close(fd);
1109 }
1110
1111 closedir(dir);
1112
1113 if (!chosen)
1114 return 1;
1115
1116 /* 'chosen' is the best device we can find. Let's write its
1117 * metadata to devname dfd is read-only so don't use that
1118 */
1119 fd = open(devname, O_RDWR);
1120 if (fd >= 0) {
1121 chosen_st->ss->store_super(chosen_st, fd);
1122 close(fd);
1123 }
1124 free(chosen);
1125 chosen_st->ss->free_super(chosen_st);
1126 free(chosen_st);
1127 return 0;
1128 }
1129
1130 static int is_bare(int dfd)
1131 {
1132 unsigned long long size = 0;
1133 char bufpad[4096 + 4096];
1134 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1135
1136 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1137 read(dfd, buf, 4096) != 4096)
1138 return 0;
1139
1140 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1141 return 0;
1142 if (memcmp(buf, buf+1, 4095) != 0)
1143 return 0;
1144
1145 /* OK, first 4K appear blank, try the end. */
1146 get_dev_size(dfd, NULL, &size);
1147 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1148 read(dfd, buf, 4096) != 4096)
1149 return 0;
1150
1151 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1152 return 0;
1153 if (memcmp(buf, buf+1, 4095) != 0)
1154 return 0;
1155
1156 return 1;
1157 }
1158
1159 /* adding a spare to a regular array is quite different from adding one to
1160 * a set-of-partitions virtual array.
1161 * This function determines which is worth trying and tries as appropriate.
1162 * Arrays are given priority over partitions.
1163 */
1164 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1165 struct map_ent *target,
1166 struct supertype *st, int verbose)
1167 {
1168 int i;
1169 int rv;
1170 int arrays_ok = 0;
1171 int partitions_ok = 0;
1172 int dfd = *dfdp;
1173 int bare;
1174
1175 /* Can only add a spare if device has at least one domain */
1176 if (pol_find(pol, pol_domain) == NULL)
1177 return 1;
1178 /* And only if some action allows spares */
1179 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1180 return 1;
1181
1182 /* Now check if the device is bare.
1183 * bare devices can always be added as a spare
1184 * non-bare devices can only be added if spare-same-slot is permitted,
1185 * and this device is replacing a previous device - in which case 'target'
1186 * will be set.
1187 */
1188 if (!is_bare(dfd)) {
1189 /* Must have a target and allow same_slot */
1190 /* Later - may allow force_spare without target */
1191 if (!target ||
1192 !policy_action_allows(pol, st?st->ss->name:NULL,
1193 act_spare_same_slot)) {
1194 if (verbose > 1)
1195 pr_err("%s is not bare, so not "
1196 "considering as a spare\n",
1197 devname);
1198 return 1;
1199 }
1200 bare = 0;
1201 } else
1202 bare = 1;
1203
1204 /* It might be OK to add this device to an array - need to see
1205 * what arrays might be candidates.
1206 */
1207 if (st) {
1208 /* just try try 'array' or 'partition' based on this metadata */
1209 if (st->ss->add_to_super)
1210 return array_try_spare(devname, dfdp, pol, target, bare,
1211 st, verbose);
1212 else
1213 return partition_try_spare(devname, dfdp, pol,
1214 st, verbose);
1215 }
1216 /* No metadata was specified or found so options are open.
1217 * Check for whether any array metadata, or any partition metadata
1218 * might allow adding the spare. This check is just help to avoid
1219 * a more costly scan of all arrays when we can be sure that will
1220 * fail.
1221 */
1222 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1223 if (superlist[i]->add_to_super && !arrays_ok &&
1224 policy_action_allows(pol, superlist[i]->name, act_spare))
1225 arrays_ok = 1;
1226 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1227 policy_action_allows(pol, superlist[i]->name, act_spare))
1228 partitions_ok = 1;
1229 }
1230 rv = 1;
1231 if (arrays_ok)
1232 rv = array_try_spare(devname, dfdp, pol, target, bare,
1233 st, verbose);
1234 if (rv != 0 && partitions_ok)
1235 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1236 return rv;
1237 }
1238
1239 int IncrementalScan(int verbose)
1240 {
1241 /* look at every device listed in the 'map' file.
1242 * If one is found that is not running then:
1243 * look in mdadm.conf for bitmap file.
1244 * if one exists, but array has none, add it.
1245 * try to start array in auto-readonly mode
1246 */
1247 struct map_ent *mapl = NULL;
1248 struct map_ent *me;
1249 struct mddev_ident *devs, *mddev;
1250 int rv = 0;
1251
1252 map_read(&mapl);
1253 devs = conf_get_ident(NULL);
1254
1255 for (me = mapl ; me ; me = me->next) {
1256 mdu_array_info_t array;
1257 mdu_bitmap_file_t bmf;
1258 struct mdinfo *sra;
1259 int mdfd = open_dev(me->devnum);
1260
1261 if (mdfd < 0)
1262 continue;
1263 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1264 errno != ENODEV) {
1265 close(mdfd);
1266 continue;
1267 }
1268 /* Ok, we can try this one. Maybe it needs a bitmap */
1269 for (mddev = devs ; mddev ; mddev = mddev->next)
1270 if (mddev->devname && me->path
1271 && devname_matches(mddev->devname, me->path))
1272 break;
1273 if (mddev && mddev->bitmap_file) {
1274 /*
1275 * Note: early kernels will wrongly fail this, so it
1276 * is a hint only
1277 */
1278 int added = -1;
1279 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1280 int bmfd = open(mddev->bitmap_file, O_RDWR);
1281 if (bmfd >= 0) {
1282 added = ioctl(mdfd, SET_BITMAP_FILE,
1283 bmfd);
1284 close(bmfd);
1285 }
1286 }
1287 if (verbose >= 0) {
1288 if (added == 0)
1289 pr_err("Added bitmap %s to %s\n",
1290 mddev->bitmap_file, me->path);
1291 else if (errno != EEXIST)
1292 pr_err("Failed to add bitmap to %s: %s\n",
1293 me->path, strerror(errno));
1294 }
1295 }
1296 /* FIXME check for reshape_active and consider not
1297 * starting array.
1298 */
1299 sra = sysfs_read(mdfd, 0, 0);
1300 if (sra) {
1301 if (sysfs_set_str(sra, NULL,
1302 "array_state", "read-auto") == 0) {
1303 if (verbose >= 0)
1304 pr_err("started array %s\n",
1305 me->path ?: devnum2devname(me->devnum));
1306 } else {
1307 pr_err("failed to start array %s: %s\n",
1308 me->path ?: devnum2devname(me->devnum),
1309 strerror(errno));
1310 rv = 1;
1311 }
1312 sysfs_free(sra);
1313 }
1314 }
1315 return rv;
1316 }
1317
1318 static char *container2devname(char *devname)
1319 {
1320 char *mdname = NULL;
1321
1322 if (devname[0] == '/') {
1323 int fd = open(devname, O_RDONLY);
1324 if (fd >= 0) {
1325 mdname = devnum2devname(fd2devnum(fd));
1326 close(fd);
1327 }
1328 } else {
1329 int uuid[4];
1330 struct map_ent *mp, *map = NULL;
1331
1332 if (!parse_uuid(devname, uuid))
1333 return mdname;
1334 mp = map_by_uuid(&map, uuid);
1335 if (mp)
1336 mdname = devnum2devname(mp->devnum);
1337 map_free(map);
1338 }
1339
1340 return mdname;
1341 }
1342
1343 static int Incremental_container(struct supertype *st, char *devname,
1344 struct context *c)
1345 {
1346 /* Collect the contents of this container and for each
1347 * array, choose a device name and assemble the array.
1348 */
1349
1350 struct mdinfo *list;
1351 struct mdinfo *ra;
1352 struct map_ent *map = NULL;
1353 struct mdinfo info;
1354 int trustworthy;
1355 struct mddev_ident *match;
1356 int rv = 0;
1357 struct domainlist *domains;
1358 struct map_ent *smp;
1359 int suuid[4];
1360 int sfd;
1361 int ra_blocked = 0;
1362 int ra_all = 0;
1363
1364 st->ss->getinfo_super(st, &info, NULL);
1365
1366 if ((c->runstop > 0 && info.container_enough >= 0) ||
1367 info.container_enough > 0)
1368 /* pass */;
1369 else {
1370 if (c->verbose)
1371 pr_err("not enough devices to start the container\n");
1372 return 0;
1373 }
1374
1375 match = conf_match(st, &info, devname, c->verbose, &rv);
1376 if (match == NULL && rv == 2)
1377 return rv;
1378
1379 /* Need to compute 'trustworthy' */
1380 if (match)
1381 trustworthy = LOCAL;
1382 else if (st->ss->match_home(st, c->homehost) == 1)
1383 trustworthy = LOCAL;
1384 else if (st->ss->match_home(st, "any") == 1)
1385 trustworthy = LOCAL;
1386 else
1387 trustworthy = FOREIGN;
1388
1389 list = st->ss->container_content(st, NULL);
1390 /* when nothing to activate - quit */
1391 if (list == NULL)
1392 return 0;
1393 for (ra = list ; ra ; ra = ra->next) {
1394 int mdfd;
1395 char chosen_name[1024];
1396 struct map_ent *mp;
1397 struct mddev_ident *match = NULL;
1398
1399 ra_all++;
1400 /* do not activate arrays blocked by metadata handler */
1401 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1402 pr_err("Cannot activate array %s in %s.\n",
1403 ra->text_version, devname);
1404 ra_blocked++;
1405 continue;
1406 }
1407 mp = map_by_uuid(&map, ra->uuid);
1408
1409 if (mp) {
1410 mdfd = open_dev(mp->devnum);
1411 if (mp->path)
1412 strcpy(chosen_name, mp->path);
1413 else
1414 strcpy(chosen_name, devnum2devname(mp->devnum));
1415 } else {
1416
1417 /* Check in mdadm.conf for container == devname and
1418 * member == ra->text_version after second slash.
1419 */
1420 char *sub = strchr(ra->text_version+1, '/');
1421 struct mddev_ident *array_list;
1422 if (sub) {
1423 sub++;
1424 array_list = conf_get_ident(NULL);
1425 } else
1426 array_list = NULL;
1427 for(; array_list ; array_list = array_list->next) {
1428 char *dn;
1429 if (array_list->member == NULL ||
1430 array_list->container == NULL)
1431 continue;
1432 if (strcmp(array_list->member, sub) != 0)
1433 continue;
1434 if (array_list->uuid_set &&
1435 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1436 continue;
1437 dn = container2devname(array_list->container);
1438 if (dn == NULL)
1439 continue;
1440 if (strncmp(dn, ra->text_version+1,
1441 strlen(dn)) != 0 ||
1442 ra->text_version[strlen(dn)+1] != '/') {
1443 free(dn);
1444 continue;
1445 }
1446 free(dn);
1447 /* we have a match */
1448 match = array_list;
1449 if (c->verbose>0)
1450 pr_err("match found for member %s\n",
1451 array_list->member);
1452 break;
1453 }
1454
1455 if (match && match->devname &&
1456 strcasecmp(match->devname, "<ignore>") == 0) {
1457 if (c->verbose > 0)
1458 pr_err("array %s/%s is "
1459 "explicitly ignored by mdadm.conf\n",
1460 match->container, match->member);
1461 return 2;
1462 }
1463 if (match)
1464 trustworthy = LOCAL;
1465
1466 mdfd = create_mddev(match ? match->devname : NULL,
1467 ra->name,
1468 c->autof,
1469 trustworthy,
1470 chosen_name);
1471 }
1472
1473 if (mdfd < 0) {
1474 pr_err("failed to open %s: %s.\n",
1475 chosen_name, strerror(errno));
1476 return 2;
1477 }
1478
1479 assemble_container_content(st, mdfd, ra, c,
1480 chosen_name);
1481 close(mdfd);
1482 }
1483
1484 /* don't move spares to container with volume being activated
1485 when all volumes are blocked */
1486 if (ra_all == ra_blocked)
1487 return 0;
1488
1489 /* Now move all suitable spares from spare container */
1490 domains = domain_from_array(list, st->ss->name);
1491 memcpy(suuid, uuid_zero, sizeof(int[4]));
1492 if (domains &&
1493 (smp = map_by_uuid(&map, suuid)) != NULL &&
1494 (sfd = open(smp->path, O_RDONLY)) >= 0) {
1495 /* spare container found */
1496 struct supertype *sst =
1497 super_imsm.match_metadata_desc("imsm");
1498 struct mdinfo *sinfo;
1499 unsigned long long min_size = 0;
1500 if (st->ss->min_acceptable_spare_size)
1501 min_size = st->ss->min_acceptable_spare_size(st);
1502 if (!sst->ss->load_container(sst, sfd, NULL)) {
1503 close(sfd);
1504 sinfo = container_choose_spares(sst, min_size,
1505 domains, NULL,
1506 st->ss->name, 0);
1507 sst->ss->free_super(sst);
1508 if (sinfo){
1509 int count = 0;
1510 struct mdinfo *disks = sinfo->devs;
1511 while (disks) {
1512 /* move spare from spare
1513 * container to currently
1514 * assembled one
1515 */
1516 if (move_spare(
1517 smp->path,
1518 devname,
1519 makedev(disks->disk.major,
1520 disks->disk.minor)))
1521 count++;
1522 disks = disks->next;
1523 }
1524 if (count)
1525 pr_err("Added %d spare%s to %s\n",
1526 count, count>1?"s":"", devname);
1527 }
1528 sysfs_free(sinfo);
1529 } else
1530 close(sfd);
1531 }
1532 domain_free(domains);
1533 return 0;
1534 }
1535
1536 /*
1537 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1538 * raid arrays, and if so first fail (if needed) and then remove the device.
1539 *
1540 * @devname - The device we want to remove
1541 * @id_path - name as found in /dev/disk/by-path for this device
1542 *
1543 * Note: the device name must be a kernel name like "sda", so
1544 * that we can find it in /proc/mdstat
1545 */
1546 int IncrementalRemove(char *devname, char *id_path, int verbose)
1547 {
1548 int mdfd;
1549 int rv;
1550 struct mdstat_ent *ent;
1551 struct mddev_dev devlist;
1552
1553 if (!id_path)
1554 dprintf(Name ": incremental removal without --path <id_path> "
1555 "lacks the possibility to re-add new device in this "
1556 "port\n");
1557
1558 if (strchr(devname, '/')) {
1559 pr_err("incremental removal requires a "
1560 "kernel device name, not a file: %s\n", devname);
1561 return 1;
1562 }
1563 ent = mdstat_by_component(devname);
1564 if (!ent) {
1565 pr_err("%s does not appear to be a component "
1566 "of any array\n", devname);
1567 return 1;
1568 }
1569 mdfd = open_dev(ent->devnum);
1570 if (mdfd < 0) {
1571 pr_err("Cannot open array %s!!\n", ent->dev);
1572 free_mdstat(ent);
1573 return 1;
1574 }
1575
1576 if (id_path) {
1577 struct map_ent *map = NULL, *me;
1578 me = map_by_devnum(&map, ent->devnum);
1579 if (me)
1580 policy_save_path(id_path, me);
1581 map_free(map);
1582 }
1583
1584 memset(&devlist, 0, sizeof(devlist));
1585 devlist.devname = devname;
1586 devlist.disposition = 'f';
1587 /* for a container, we must fail each member array */
1588 if (ent->metadata_version &&
1589 strncmp(ent->metadata_version, "external:", 9) == 0) {
1590 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1591 struct mdstat_ent *memb;
1592 for (memb = mdstat ; memb ; memb = memb->next)
1593 if (is_container_member(memb, ent->dev)) {
1594 int subfd = open_dev(memb->devnum);
1595 if (subfd >= 0) {
1596 Manage_subdevs(memb->dev, subfd,
1597 &devlist, verbose, 0,
1598 NULL, 0);
1599 close(subfd);
1600 }
1601 }
1602 free_mdstat(mdstat);
1603 } else
1604 Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0);
1605 devlist.disposition = 'r';
1606 rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0);
1607 close(mdfd);
1608 free_mdstat(ent);
1609 return rv;
1610 }
Unnamed repository; edit this file 'description' to name the repository.