]> 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
util: Introduce md_get_disk_info()
[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) 2006-2013 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 <sys/wait.h>
33 #include <dirent.h>
34 #include <ctype.h>
35
36 static int count_active(struct supertype *st, struct mdinfo *sra,
37 int mdfd, char **availp,
38 struct mdinfo *info);
39 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
40 int number, __u64 events, int verbose,
41 char *array_name);
42 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
43 struct map_ent *target,
44 struct supertype *st, int verbose);
45
46 static int Incremental_container(struct supertype *st, char *devname,
47 struct context *c, char *only);
48
49 int Incremental(struct mddev_dev *devlist, struct context *c,
50 struct supertype *st)
51 {
52 /* Add this device to an array, creating the array if necessary
53 * and starting the array if sensible or - if runstop>0 - if possible.
54 *
55 * This has several steps:
56 *
57 * 1/ Check if device is permitted by mdadm.conf, reject if not.
58 * 2/ Find metadata, reject if none appropriate (check
59 * version/name from args)
60 * 3/ Check if there is a match in mdadm.conf
61 * 3a/ if not, check for homehost match. If no match, assemble as
62 * a 'foreign' array.
63 * 4/ Determine device number.
64 * - If in mdadm.conf with std name, use that
65 * - UUID in /var/run/mdadm.map use that
66 * - If name is suggestive, use that. unless in use with different uuid.
67 * - Choose a free, high number.
68 * - Use a partitioned device unless strong suggestion not to.
69 * e.g. auto=md
70 * Don't choose partitioned for containers.
71 * 5/ Find out if array already exists
72 * 5a/ if it does not
73 * - choose a name, from mdadm.conf or 'name' field in array.
74 * - create the array
75 * - add the device
76 * 5b/ if it does
77 * - check one drive in array to make sure metadata is a reasonably
78 * close match. Reject if not (e.g. different type)
79 * - add the device
80 * 6/ Make sure /var/run/mdadm.map contains this array.
81 * 7/ Is there enough devices to possibly start the array?
82 * For a container, this means running Incremental_container.
83 * 7a/ if not, finish with success.
84 * 7b/ if yes,
85 * - read all metadata and arrange devices like -A does
86 * - if number of OK devices match expected, or -R and there are enough,
87 * start the array (auto-readonly).
88 */
89 struct stat stb;
90 struct mdinfo info, dinfo;
91 struct mdinfo *sra = NULL, *d;
92 struct mddev_ident *match;
93 char chosen_name[1024];
94 char *md_devname;
95 int rv = 1;
96 struct map_ent *mp, *map = NULL;
97 int dfd = -1, mdfd = -1;
98 char *avail = NULL;
99 int active_disks;
100 int trustworthy;
101 char *name_to_use;
102 mdu_array_info_t ainf;
103 struct dev_policy *policy = NULL;
104 struct map_ent target_array;
105 int have_target;
106 char *devname = devlist->devname;
107 int journal_device_missing = 0;
108
109 struct createinfo *ci = conf_get_create_info();
110
111 if (stat(devname, &stb) < 0) {
112 if (c->verbose >= 0)
113 pr_err("stat failed for %s: %s.\n",
114 devname, strerror(errno));
115 return rv;
116 }
117 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
118 if (c->verbose >= 0)
119 pr_err("%s is not a block device.\n",
120 devname);
121 return rv;
122 }
123 dfd = dev_open(devname, O_RDONLY);
124 if (dfd < 0) {
125 if (c->verbose >= 0)
126 pr_err("cannot open %s: %s.\n",
127 devname, strerror(errno));
128 return rv;
129 }
130 /* If the device is a container, we do something very different */
131 if (must_be_container(dfd)) {
132 if (!st)
133 st = super_by_fd(dfd, NULL);
134 if (st && st->ss->load_container)
135 rv = st->ss->load_container(st, dfd, NULL);
136
137 close(dfd);
138 if (!rv && st->ss->container_content) {
139 if (map_lock(&map))
140 pr_err("failed to get exclusive lock on mapfile\n");
141 if (c->export)
142 printf("MD_DEVNAME=%s\n", devname);
143 rv = Incremental_container(st, devname, c, NULL);
144 map_unlock(&map);
145 return rv;
146 }
147
148 pr_err("%s is not part of an md array.\n",
149 devname);
150 return rv;
151 }
152
153 /* 1/ Check if device is permitted by mdadm.conf */
154
155 for (;devlist; devlist = devlist->next)
156 if (conf_test_dev(devlist->devname))
157 break;
158 if (!devlist) {
159 devlist = conf_get_devs();
160 for (;devlist; devlist = devlist->next) {
161 struct stat st2;
162 if (stat(devlist->devname, &st2) == 0 &&
163 (st2.st_mode & S_IFMT) == S_IFBLK &&
164 st2.st_rdev == stb.st_rdev)
165 break;
166 }
167 }
168 if (!devlist) {
169 if (c->verbose >= 0)
170 pr_err("%s not permitted by mdadm.conf.\n",
171 devname);
172 goto out;
173 }
174
175 /* 2/ Find metadata, reject if none appropriate (check
176 * version/name from args) */
177
178 if (fstat(dfd, &stb) < 0) {
179 if (c->verbose >= 0)
180 pr_err("fstat failed for %s: %s.\n",
181 devname, strerror(errno));
182 goto out;
183 }
184 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
185 if (c->verbose >= 0)
186 pr_err("%s is not a block device.\n",
187 devname);
188 goto out;
189 }
190
191 dinfo.disk.major = major(stb.st_rdev);
192 dinfo.disk.minor = minor(stb.st_rdev);
193
194 policy = disk_policy(&dinfo);
195 have_target = policy_check_path(&dinfo, &target_array);
196
197 if (st == NULL && (st = guess_super_type(dfd, guess_array)) == NULL) {
198 if (c->verbose >= 0)
199 pr_err("no recognisable superblock on %s.\n",
200 devname);
201 rv = try_spare(devname, &dfd, policy,
202 have_target ? &target_array : NULL,
203 NULL, c->verbose);
204 goto out;
205 }
206 st->ignore_hw_compat = 0;
207
208 if (st->ss->compare_super == NULL ||
209 st->ss->load_super(st, dfd, c->verbose >= 0 ? devname : NULL)) {
210 if (c->verbose >= 0)
211 pr_err("no RAID superblock on %s.\n",
212 devname);
213 rv = try_spare(devname, &dfd, policy,
214 have_target ? &target_array : NULL,
215 st, c->verbose);
216 free(st);
217 goto out;
218 }
219 close (dfd); dfd = -1;
220
221 st->ss->getinfo_super(st, &info, NULL);
222
223 /* 3/ Check if there is a match in mdadm.conf */
224 match = conf_match(st, &info, devname, c->verbose, &rv);
225 if (!match && rv == 2)
226 goto out;
227
228 if (match && match->devname
229 && strcasecmp(match->devname, "<ignore>") == 0) {
230 if (c->verbose >= 0)
231 pr_err("array containing %s is explicitly ignored by mdadm.conf\n",
232 devname);
233 goto out;
234 }
235
236 /* 3a/ if not, check for homehost match. If no match, continue
237 * but don't trust the 'name' in the array. Thus a 'random' minor
238 * number will be assigned, and the device name will be based
239 * on that. */
240 if (match)
241 trustworthy = LOCAL;
242 else if (st->ss->match_home(st, c->homehost) == 1)
243 trustworthy = LOCAL;
244 else if (st->ss->match_home(st, "any") == 1)
245 trustworthy = LOCAL_ANY;
246 else
247 trustworthy = FOREIGN;
248
249 if (!match && !conf_test_metadata(st->ss->name, policy,
250 (trustworthy == LOCAL))) {
251 if (c->verbose >= 1)
252 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
253 devname, st->ss->name);
254 goto out;
255 }
256 if (trustworthy == LOCAL_ANY)
257 trustworthy = LOCAL;
258
259 /* There are three possible sources for 'autof': command line,
260 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
261 * ARRAY takes precedence, then command line, then
262 * CREATE.
263 */
264 if (match && match->autof)
265 c->autof = match->autof;
266 if (c->autof == 0)
267 c->autof = ci->autof;
268
269 name_to_use = info.name;
270 if (name_to_use[0] == 0 &&
271 info.array.level == LEVEL_CONTAINER) {
272 name_to_use = info.text_version;
273 trustworthy = METADATA;
274 }
275 if (name_to_use[0] && trustworthy != LOCAL &&
276 ! c->require_homehost &&
277 conf_name_is_free(name_to_use))
278 trustworthy = LOCAL;
279
280 /* strip "hostname:" prefix from name if we have decided
281 * to treat it as LOCAL
282 */
283 if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
284 name_to_use = strchr(name_to_use, ':')+1;
285
286 /* 4/ Check if array exists.
287 */
288 if (map_lock(&map))
289 pr_err("failed to get exclusive lock on mapfile\n");
290 /* Now check we can get O_EXCL. If not, probably "mdadm -A" has
291 * taken over
292 */
293 dfd = dev_open(devname, O_RDONLY|O_EXCL);
294 if (dfd < 0) {
295 if (c->verbose >= 0)
296 pr_err("cannot reopen %s: %s.\n",
297 devname, strerror(errno));
298 goto out_unlock;
299 }
300 /* Cannot hold it open while we add the device to the array,
301 * so we must release the O_EXCL and depend on the map_lock()
302 * So now is the best time to remove any partitions.
303 */
304 remove_partitions(dfd);
305 close(dfd);
306 dfd = -1;
307
308 mp = map_by_uuid(&map, info.uuid);
309 if (mp)
310 mdfd = open_dev(mp->devnm);
311 else
312 mdfd = -1;
313
314 if (mdfd < 0) {
315
316 /* Skip the clustered ones. This should be started by
317 * clustering resource agents
318 */
319 if (info.array.state & (1 << MD_SB_CLUSTERED))
320 goto out;
321
322 /* Couldn't find an existing array, maybe make a new one */
323 mdfd = create_mddev(match ? match->devname : NULL,
324 name_to_use, c->autof, trustworthy, chosen_name);
325
326 if (mdfd < 0)
327 goto out_unlock;
328
329 sysfs_init(&info, mdfd, NULL);
330
331 if (set_array_info(mdfd, st, &info) != 0) {
332 pr_err("failed to set array info for %s: %s\n",
333 chosen_name, strerror(errno));
334 rv = 2;
335 goto out_unlock;
336 }
337
338 dinfo = info;
339 dinfo.disk.major = major(stb.st_rdev);
340 dinfo.disk.minor = minor(stb.st_rdev);
341 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
342 pr_err("failed to add %s to new array %s: %s.\n",
343 devname, chosen_name, strerror(errno));
344 ioctl(mdfd, STOP_ARRAY, 0);
345 rv = 2;
346 goto out_unlock;
347 }
348 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
349 GET_OFFSET | GET_SIZE));
350
351 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
352 /* It really should be 'none' - must be old buggy
353 * kernel, and mdadm -I may not be able to complete.
354 * So reject it.
355 */
356 ioctl(mdfd, STOP_ARRAY, NULL);
357 pr_err("You have an old buggy kernel which cannot support\n --incremental reliably. Aborting.\n");
358 rv = 2;
359 goto out_unlock;
360 }
361 info.array.working_disks = 1;
362 /* 6/ Make sure /var/run/mdadm.map contains this array. */
363 map_update(&map, fd2devnm(mdfd),
364 info.text_version,
365 info.uuid, chosen_name);
366 } else {
367 /* 5b/ if it does */
368 /* - check one drive in array to make sure metadata is a reasonably */
369 /* close match. Reject if not (e.g. different type) */
370 /* - add the device */
371 char dn[20];
372 int dfd2;
373 int err;
374 struct supertype *st2;
375 struct mdinfo info2, *d;
376
377 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
378 GET_OFFSET | GET_SIZE));
379
380 if (mp->path)
381 strcpy(chosen_name, mp->path);
382 else
383 strcpy(chosen_name, mp->devnm);
384
385 /* It is generally not OK to add non-spare drives to a
386 * running array as they are probably missing because
387 * they failed. However if runstop is 1, then the
388 * array was possibly started early and our best bet is
389 * to add this anyway.
390 * Also if action policy is re-add or better we allow
391 * re-add.
392 * This doesn't apply to containers as the 'non-spare'
393 * flag has a different meaning. The test has to happen
394 * at the device level there
395 */
396 if (!st->ss->external
397 && (info.disk.state & (1<<MD_DISK_SYNC)) != 0
398 && ! policy_action_allows(policy, st->ss->name,
399 act_re_add)
400 && c->runstop < 1) {
401 if (md_get_array_info(mdfd, &ainf) == 0) {
402 pr_err("not adding %s to active array (without --run) %s\n",
403 devname, chosen_name);
404 rv = 2;
405 goto out_unlock;
406 }
407 }
408 if (!sra) {
409 rv = 2;
410 goto out_unlock;
411 }
412 if (sra->devs) {
413 sprintf(dn, "%d:%d", sra->devs->disk.major,
414 sra->devs->disk.minor);
415 dfd2 = dev_open(dn, O_RDONLY);
416 if (dfd2 < 0) {
417 pr_err("unable to open %s\n", devname);
418 rv = 2;
419 goto out_unlock;
420 }
421 st2 = dup_super(st);
422 if (st2->ss->load_super(st2, dfd2, NULL) ||
423 st->ss->compare_super(st, st2) != 0) {
424 pr_err("metadata mismatch between %s and chosen array %s\n",
425 devname, chosen_name);
426 close(dfd2);
427 rv = 2;
428 goto out_unlock;
429 }
430 close(dfd2);
431 st2->ss->getinfo_super(st2, &info2, NULL);
432 st2->ss->free_super(st2);
433 if (info.array.level != info2.array.level ||
434 memcmp(info.uuid, info2.uuid, 16) != 0 ||
435 info.array.raid_disks != info2.array.raid_disks) {
436 pr_err("unexpected difference between %s and %s.\n",
437 chosen_name, devname);
438 rv = 2;
439 goto out_unlock;
440 }
441 }
442 info.disk.major = major(stb.st_rdev);
443 info.disk.minor = minor(stb.st_rdev);
444 /* add disk needs to know about containers */
445 if (st->ss->external)
446 sra->array.level = LEVEL_CONTAINER;
447
448 if (info.array.state & (1 << MD_SB_CLUSTERED))
449 info.disk.state |= (1 << MD_DISK_CLUSTER_ADD);
450
451 err = add_disk(mdfd, st, sra, &info);
452 if (err < 0 && errno == EBUSY) {
453 /* could be another device present with the same
454 * disk.number. Find and reject any such
455 */
456 find_reject(mdfd, st, sra, info.disk.number,
457 info.events, c->verbose, chosen_name);
458 err = add_disk(mdfd, st, sra, &info);
459 }
460 if (err < 0 && errno == EINVAL &&
461 info.disk.state & (1<<MD_DISK_SYNC)) {
462 /* Maybe it needs to be added as a spare */
463 if (policy_action_allows(policy, st->ss->name,
464 act_force_spare)) {
465 info.disk.state &= ~(1<<MD_DISK_SYNC);
466 err = add_disk(mdfd, st, sra, &info);
467 } else
468 if (c->verbose >= 0)
469 pr_err("can only add %s to %s as a spare, and force-spare is not set.\n",
470 devname, chosen_name);
471 }
472 if (err < 0) {
473 pr_err("failed to add %s to existing array %s: %s.\n",
474 devname, chosen_name, strerror(errno));
475 rv = 2;
476 goto out_unlock;
477 }
478 info.array.working_disks = 0;
479 for (d = sra->devs; d; d=d->next)
480 info.array.working_disks ++;
481
482 }
483 if (strncmp(chosen_name, "/dev/md/", 8) == 0)
484 md_devname = chosen_name+8;
485 else
486 md_devname = chosen_name;
487 if (c->export) {
488 printf("MD_DEVICE=%s\n", fd2devnm(mdfd));
489 printf("MD_DEVNAME=%s\n", md_devname);
490 printf("MD_FOREIGN=%s\n", trustworthy == FOREIGN ? "yes" : "no");
491 }
492
493 /* 7/ Is there enough devices to possibly start the array? */
494 /* 7a/ if not, finish with success. */
495 if (info.array.level == LEVEL_CONTAINER) {
496 char devnm[32];
497 /* Try to assemble within the container */
498 sysfs_uevent(sra, "change");
499 if (!c->export && c->verbose >= 0)
500 pr_err("container %s now has %d device%s\n",
501 chosen_name, info.array.working_disks,
502 info.array.working_disks == 1?"":"s");
503 wait_for(chosen_name, mdfd);
504 if (st->ss->external)
505 strcpy(devnm, fd2devnm(mdfd));
506 if (st->ss->load_container)
507 rv = st->ss->load_container(st, mdfd, NULL);
508 close(mdfd);
509 sysfs_free(sra);
510 if (!rv)
511 rv = Incremental_container(st, chosen_name, c, NULL);
512 map_unlock(&map);
513 /* after spare is added, ping monitor for external metadata
514 * so that it can eg. try to rebuild degraded array */
515 if (st->ss->external)
516 ping_monitor(devnm);
517 return rv;
518 }
519
520 /* We have added something to the array, so need to re-read the
521 * state. Eventually this state should be kept up-to-date as
522 * things change.
523 */
524 sysfs_free(sra);
525 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
526 GET_OFFSET | GET_SIZE));
527 active_disks = count_active(st, sra, mdfd, &avail, &info);
528
529 journal_device_missing = (info.journal_device_required) && (info.journal_clean == 0);
530
531 if (info.consistency_policy == CONSISTENCY_POLICY_PPL)
532 info.array.state |= 1;
533
534 if (enough(info.array.level, info.array.raid_disks,
535 info.array.layout, info.array.state & 1,
536 avail) == 0) {
537 if (c->export) {
538 printf("MD_STARTED=no\n");
539 } else if (c->verbose >= 0)
540 pr_err("%s attached to %s, not enough to start (%d).\n",
541 devname, chosen_name, active_disks);
542 rv = 0;
543 goto out_unlock;
544 }
545
546 /* 7b/ if yes, */
547 /* - if number of OK devices match expected, or -R and there */
548 /* are enough, */
549 /* + add any bitmap file */
550 /* + start the array (auto-readonly). */
551
552 if (md_get_array_info(mdfd, &ainf) == 0) {
553 if (c->export) {
554 printf("MD_STARTED=already\n");
555 } else if (c->verbose >= 0)
556 pr_err("%s attached to %s which is already active.\n",
557 devname, chosen_name);
558 rv = 0;
559 goto out_unlock;
560 }
561
562 map_unlock(&map);
563 if (c->runstop > 0 || (!journal_device_missing && active_disks >= info.array.working_disks)) {
564 struct mdinfo *dsk;
565 /* Let's try to start it */
566
567 if (journal_device_missing)
568 pr_err("Trying to run with missing journal device\n");
569 if (info.reshape_active && !(info.reshape_active & RESHAPE_NO_BACKUP)) {
570 pr_err("%s: This array is being reshaped and cannot be started\n",
571 chosen_name);
572 cont_err("by --incremental. Please use --assemble\n");
573 goto out;
574 }
575 if (match && match->bitmap_file) {
576 int bmfd = open(match->bitmap_file, O_RDWR);
577 if (bmfd < 0) {
578 pr_err("Could not open bitmap file %s.\n",
579 match->bitmap_file);
580 goto out;
581 }
582 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
583 close(bmfd);
584 pr_err("Failed to set bitmapfile for %s.\n",
585 chosen_name);
586 goto out;
587 }
588 close(bmfd);
589 }
590 /* Need to remove from the array any devices which
591 * 'count_active' discerned were too old or inappropriate
592 */
593 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
594 if (d->disk.state & (1<<MD_DISK_REMOVED))
595 remove_disk(mdfd, st, sra, d);
596
597 if ((sra == NULL || active_disks >= info.array.working_disks)
598 && trustworthy != FOREIGN)
599 rv = ioctl(mdfd, RUN_ARRAY, NULL);
600 else
601 rv = sysfs_set_str(sra, NULL,
602 "array_state", "read-auto");
603 /* Array might be O_EXCL which will interfere with
604 * fsck and mount. So re-open without O_EXCL.
605 */
606 reopen_mddev(mdfd);
607 if (rv == 0) {
608 if (c->export) {
609 printf("MD_STARTED=yes\n");
610 } else if (c->verbose >= 0)
611 pr_err("%s attached to %s, which has been started.\n",
612 devname, chosen_name);
613 rv = 0;
614 wait_for(chosen_name, mdfd);
615 /* We just started the array, so some devices
616 * might have been evicted from the array
617 * because their event counts were too old.
618 * If the action=re-add policy is in-force for
619 * those devices we should re-add them now.
620 */
621 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
622 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
623 add_disk(mdfd, st, sra, dsk) == 0)
624 pr_err("%s re-added to %s\n",
625 dsk->sys_name, chosen_name);
626 }
627 } else {
628 pr_err("%s attached to %s, but failed to start: %s.\n",
629 devname, chosen_name, strerror(errno));
630 rv = 1;
631 }
632 } else {
633 if (c->export) {
634 printf("MD_STARTED=unsafe\n");
635 } else if (journal_device_missing) {
636 pr_err("Journal device is missing, not safe to start yet.\n");
637 } else if (c->verbose >= 0)
638 pr_err("%s attached to %s, not enough to start safely.\n",
639 devname, chosen_name);
640 rv = 0;
641 }
642 out:
643 free(avail);
644 if (dfd >= 0)
645 close(dfd);
646 if (mdfd >= 0)
647 close(mdfd);
648 if (policy)
649 dev_policy_free(policy);
650 sysfs_free(sra);
651 return rv;
652 out_unlock:
653 map_unlock(&map);
654 goto out;
655 }
656
657 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
658 int number, __u64 events, int verbose,
659 char *array_name)
660 {
661 /* Find a device attached to this array with a disk.number of number
662 * and events less than the passed events, and remove the device.
663 */
664 struct mdinfo *d;
665 mdu_array_info_t ra;
666
667 if (md_get_array_info(mdfd, &ra) == 0)
668 return; /* not safe to remove from active arrays
669 * without thinking more */
670
671 for (d = sra->devs; d ; d = d->next) {
672 char dn[24]; // 2*11 bytes for ints (including sign) + colon + null byte
673 int dfd;
674 struct mdinfo info;
675 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
676 dfd = dev_open(dn, O_RDONLY);
677 if (dfd < 0)
678 continue;
679 if (st->ss->load_super(st, dfd, NULL)) {
680 close(dfd);
681 continue;
682 }
683 st->ss->getinfo_super(st, &info, NULL);
684 st->ss->free_super(st);
685 close(dfd);
686
687 if (info.disk.number != number ||
688 info.events >= events)
689 continue;
690
691 if (d->disk.raid_disk > -1)
692 sysfs_set_str(sra, d, "slot", "none");
693 if (sysfs_set_str(sra, d, "state", "remove") == 0)
694 if (verbose >= 0)
695 pr_err("removing old device %s from %s\n",
696 d->sys_name+4, array_name);
697 }
698 }
699
700 static int count_active(struct supertype *st, struct mdinfo *sra,
701 int mdfd, char **availp,
702 struct mdinfo *bestinfo)
703 {
704 /* count how many devices in sra think they are active */
705 struct mdinfo *d;
706 int cnt = 0;
707 int replcnt = 0;
708 __u64 max_events = 0;
709 char *avail = NULL;
710 int *best = NULL;
711 char *devmap = NULL;
712 int numdevs = 0;
713 int devnum;
714 int b, i;
715 int raid_disks = 0;
716
717 if (!sra)
718 return 0;
719
720 for (d = sra->devs ; d ; d = d->next)
721 numdevs++;
722 for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
723 char dn[30];
724 int dfd;
725 int ok;
726 struct mdinfo info;
727
728 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
729 dfd = dev_open(dn, O_RDONLY);
730 if (dfd < 0)
731 continue;
732 ok = st->ss->load_super(st, dfd, NULL);
733 close(dfd);
734 if (ok != 0)
735 continue;
736
737 info.array.raid_disks = raid_disks;
738 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
739 if (info.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
740 bestinfo->journal_clean = 1;
741 if (!avail) {
742 raid_disks = info.array.raid_disks;
743 avail = xcalloc(raid_disks, 1);
744 *availp = avail;
745
746 best = xcalloc(raid_disks, sizeof(int));
747 devmap = xcalloc(raid_disks, numdevs);
748
749 st->ss->getinfo_super(st, &info, devmap);
750 }
751
752 if (info.disk.state & (1<<MD_DISK_SYNC))
753 {
754 if (cnt == 0) {
755 cnt++;
756 max_events = info.events;
757 avail[info.disk.raid_disk] = 2;
758 best[info.disk.raid_disk] = devnum;
759 st->ss->getinfo_super(st, bestinfo, NULL);
760 } else if (info.events == max_events) {
761 avail[info.disk.raid_disk] = 2;
762 best[info.disk.raid_disk] = devnum;
763 } else if (info.events == max_events-1) {
764 if (avail[info.disk.raid_disk] == 0) {
765 avail[info.disk.raid_disk] = 1;
766 best[info.disk.raid_disk] = devnum;
767 }
768 } else if (info.events < max_events - 1)
769 ;
770 else if (info.events == max_events+1) {
771 int i;
772 max_events = info.events;
773 for (i = 0; i < raid_disks; i++)
774 if (avail[i])
775 avail[i]--;
776 avail[info.disk.raid_disk] = 2;
777 best[info.disk.raid_disk] = devnum;
778 st->ss->getinfo_super(st, bestinfo, NULL);
779 } else { /* info.events much bigger */
780 memset(avail, 0, raid_disks);
781 max_events = info.events;
782 avail[info.disk.raid_disk] = 2;
783 best[info.disk.raid_disk] = devnum;
784 st->ss->getinfo_super(st, bestinfo, NULL);
785 }
786 } else if (info.disk.state & (1<<MD_DISK_REPLACEMENT))
787 replcnt++;
788 st->ss->free_super(st);
789 }
790
791 if (!avail)
792 return 0;
793 /* We need to reject any device that thinks the best device is
794 * failed or missing */
795 for (b = 0; b < raid_disks; b++)
796 if (avail[b] == 2)
797 break;
798 cnt = 0;
799 for (i = 0 ; i < raid_disks ; i++) {
800 if (i != b && avail[i])
801 if (devmap[raid_disks * best[i] + b] == 0) {
802 /* This device thinks 'b' is failed -
803 * don't use it */
804 devnum = best[i];
805 for (d=sra->devs ; devnum; d = d->next)
806 devnum--;
807 d->disk.state |= (1 << MD_DISK_REMOVED);
808 avail[i] = 0;
809 }
810 if (avail[i])
811 cnt++;
812 }
813 /* Also need to reject any spare device with an event count that
814 * is too high
815 */
816 for (d = sra->devs; d; d = d->next) {
817 if (!(d->disk.state & (1<<MD_DISK_SYNC)) &&
818 d->events > max_events)
819 d->disk.state |= (1 << MD_DISK_REMOVED);
820 }
821 free(best);
822 free(devmap);
823 return cnt + replcnt;
824 }
825
826 /* test if container has degraded member(s) */
827 static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
828 {
829 mdu_array_info_t array;
830 int afd;
831 int max_degraded = 0;
832
833 for(; map; map = map->next) {
834 if (!metadata_container_matches(map->metadata, me->devnm))
835 continue;
836 afd = open_dev(map->devnm);
837 if (afd < 0)
838 continue;
839 /* most accurate information regarding array degradation */
840 if (md_get_array_info(afd, &array) >= 0) {
841 int degraded = array.raid_disks - array.active_disks -
842 array.spare_disks;
843 if (degraded > max_degraded)
844 max_degraded = degraded;
845 }
846 close(afd);
847 }
848 return (max_degraded);
849 }
850
851 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
852 struct map_ent *target, int bare,
853 struct supertype *st, int verbose)
854 {
855 /* This device doesn't have any md metadata
856 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
857 * If 'st' is not set, then we only know that some metadata allows this,
858 * others possibly don't.
859 * So look for a container or array to attach the device to.
860 * Prefer 'target' if that is set and the array is found.
861 *
862 * If st is set, then only arrays of that type are considered
863 * Return 0 on success, or some exit code on failure, probably 1.
864 */
865 int rv = 1;
866 struct stat stb;
867 struct map_ent *mp, *map = NULL;
868 struct mdinfo *chosen = NULL;
869 int dfd = *dfdp;
870
871 if (fstat(dfd, &stb) != 0)
872 return 1;
873
874 /*
875 * Now we need to find a suitable array to add this to.
876 * We only accept arrays that:
877 * - match 'st'
878 * - are in the same domains as the device
879 * - are of an size for which the device will be useful
880 * and we choose the one that is the most degraded
881 */
882
883 if (map_lock(&map)) {
884 pr_err("failed to get exclusive lock on mapfile\n");
885 return 1;
886 }
887 for (mp = map ; mp ; mp = mp->next) {
888 struct supertype *st2;
889 struct domainlist *dl = NULL;
890 struct mdinfo *sra;
891 unsigned long long devsize;
892 unsigned long long component_size = 0;
893
894 if (is_subarray(mp->metadata))
895 continue;
896 if (st) {
897 st2 = st->ss->match_metadata_desc(mp->metadata);
898 if (!st2 ||
899 (st->minor_version >= 0 &&
900 st->minor_version != st2->minor_version)) {
901 if (verbose > 1)
902 pr_err("not adding %s to %s as metadata type doesn't match\n",
903 devname, mp->path);
904 free(st2);
905 continue;
906 }
907 free(st2);
908 }
909 sra = sysfs_read(-1, mp->devnm,
910 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
911 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
912 if (!sra) {
913 /* Probably a container - no degraded info */
914 sra = sysfs_read(-1, mp->devnm,
915 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
916 GET_COMPONENT|GET_VERSION);
917 if (sra)
918 sra->array.failed_disks = -1;
919 }
920 if (!sra)
921 continue;
922 if (st == NULL) {
923 int i;
924 st2 = NULL;
925 for(i = 0; !st2 && superlist[i]; i++)
926 st2 = superlist[i]->match_metadata_desc(
927 sra->text_version);
928 if (!st2) {
929 if (verbose > 1)
930 pr_err("not adding %s to %s as metadata not recognised.\n",
931 devname, mp->path);
932 goto next;
933 }
934 /* Need to double check the 'act_spare' permissions applies
935 * to this metadata.
936 */
937 if (!policy_action_allows(pol, st2->ss->name, act_spare))
938 goto next;
939 if (!bare && !policy_action_allows(pol, st2->ss->name,
940 act_spare_same_slot))
941 goto next;
942 } else
943 st2 = st;
944 /* update number of failed disks for mostly degraded
945 * container member */
946 if (sra->array.failed_disks == -1)
947 sra->array.failed_disks = container_members_max_degradation(map, mp);
948
949 get_dev_size(dfd, NULL, &devsize);
950 if (sra->component_size == 0) {
951 /* true for containers, here we must read superblock
952 * to obtain minimum spare size */
953 struct supertype *st3 = dup_super(st2);
954 int mdfd = open_dev(mp->devnm);
955 if (mdfd < 0) {
956 free(st3);
957 goto next;
958 }
959 if (st3->ss->load_container &&
960 !st3->ss->load_container(st3, mdfd, mp->path)) {
961 component_size = st3->ss->min_acceptable_spare_size(st3);
962 st3->ss->free_super(st3);
963 }
964 free(st3);
965 close(mdfd);
966 }
967 if ((sra->component_size > 0 &&
968 st2->ss->avail_size(st2, devsize,
969 sra->devs
970 ? sra->devs->data_offset
971 : INVALID_SECTORS)
972 < sra->component_size)
973 ||
974 (sra->component_size == 0 && devsize < component_size)) {
975 if (verbose > 1)
976 pr_err("not adding %s to %s as it is too small\n",
977 devname, mp->path);
978 goto next;
979 }
980 /* test against target.
981 * If 'target' is set and 'bare' is false, we only accept
982 * arrays/containers that match 'target'.
983 * If 'target' is set and 'bare' is true, we prefer the
984 * array which matches 'target'.
985 * target is considered only if we deal with degraded array
986 */
987 if (target && policy_action_allows(pol, st2->ss->name,
988 act_spare_same_slot)) {
989 if (strcmp(target->metadata, mp->metadata) == 0 &&
990 memcmp(target->uuid, mp->uuid,
991 sizeof(target->uuid)) == 0 &&
992 sra->array.failed_disks > 0) {
993 /* This is our target!! */
994 sysfs_free(chosen);
995 chosen = sra;
996 sra = NULL;
997 /* skip to end so we don't check any more */
998 while (mp->next)
999 mp = mp->next;
1000 goto next;
1001 }
1002 /* not our target */
1003 if (!bare)
1004 goto next;
1005 }
1006
1007 dl = domain_from_array(sra, st2->ss->name);
1008 if (domain_test(dl, pol, st2->ss->name) != 1) {
1009 /* domain test fails */
1010 if (verbose > 1)
1011 pr_err("not adding %s to %s as it is not in a compatible domain\n",
1012 devname, mp->path);
1013
1014 goto next;
1015 }
1016 /* all tests passed, OK to add to this array */
1017 if (!chosen) {
1018 chosen = sra;
1019 sra = NULL;
1020 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
1021 sysfs_free(chosen);
1022 chosen = sra;
1023 sra = NULL;
1024 }
1025 next:
1026 sysfs_free(sra);
1027 if (st != st2)
1028 free(st2);
1029 if (dl)
1030 domain_free(dl);
1031 }
1032 if (chosen) {
1033 /* add current device to chosen array as a spare */
1034 int mdfd = open_dev(chosen->sys_name);
1035 if (mdfd >= 0) {
1036 struct mddev_dev devlist;
1037 char chosen_devname[24]; // 2*11 for int (including signs) + colon + null
1038 devlist.next = NULL;
1039 devlist.used = 0;
1040 devlist.writemostly = FlagDefault;
1041 devlist.failfast = FlagDefault;
1042 devlist.devname = chosen_devname;
1043 sprintf(chosen_devname, "%d:%d", major(stb.st_rdev),
1044 minor(stb.st_rdev));
1045 devlist.disposition = 'a';
1046 close(dfd);
1047 *dfdp = -1;
1048 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1049 -1, 0, NULL, 0);
1050 close(mdfd);
1051 }
1052 if (verbose > 0) {
1053 if (rv == 0)
1054 pr_err("added %s as spare for %s\n",
1055 devname, chosen->sys_name);
1056 else
1057 pr_err("failed to add %s as spare for %s\n",
1058 devname, chosen->sys_name);
1059 }
1060 sysfs_free(chosen);
1061 }
1062 map_unlock(&map);
1063 return rv;
1064 }
1065
1066 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1067 struct supertype *st, int verbose)
1068 {
1069 /* we know that at least one partition virtual-metadata is
1070 * allowed to incorporate spares like this device. We need to
1071 * find a suitable device to copy partition information from.
1072 *
1073 * Getting a list of all disk (not partition) devices is
1074 * slightly non-trivial. We could look at /sys/block, but
1075 * that is theoretically due to be removed. Maybe best to use
1076 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1077 * we depend on this directory of 'path' info. But that fails
1078 * to find loop devices and probably others. Maybe don't
1079 * worry about that, they aren't the real target.
1080 *
1081 * So: check things in /dev/disk/by-path to see if they are in
1082 * a compatible domain, then load the partition table and see
1083 * if it is OK for the new device, and choose the largest
1084 * partition table that fits.
1085 */
1086 DIR *dir;
1087 struct dirent *de;
1088 char *chosen = NULL;
1089 unsigned long long chosen_size = 0;
1090 struct supertype *chosen_st = NULL;
1091 int fd;
1092
1093 dir = opendir("/dev/disk/by-path");
1094 if (!dir)
1095 return 1;
1096 while ((de = readdir(dir)) != NULL) {
1097 char *ep;
1098 struct dev_policy *pol2 = NULL;
1099 struct domainlist *domlist = NULL;
1100 int fd = -1;
1101 struct mdinfo info;
1102 struct supertype *st2 = NULL;
1103 char *devname = NULL;
1104 unsigned long long devsectors;
1105
1106 if (de->d_ino == 0 ||
1107 de->d_name[0] == '.' ||
1108 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1109 goto next;
1110
1111 ep = de->d_name + strlen(de->d_name);
1112 while (ep > de->d_name &&
1113 isdigit(ep[-1]))
1114 ep--;
1115 if (ep > de->d_name + 5 &&
1116 strncmp(ep-5, "-part", 5) == 0)
1117 /* This is a partition - skip it */
1118 goto next;
1119
1120 pol2 = path_policy(de->d_name, type_disk);
1121
1122 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1123 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1124 /* new device is incompatible with this device. */
1125 goto next;
1126
1127 domain_free(domlist);
1128 domlist = NULL;
1129
1130 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1131 devname = NULL;
1132 goto next;
1133 }
1134 fd = open(devname, O_RDONLY);
1135 if (fd < 0)
1136 goto next;
1137 if (get_dev_size(fd, devname, &devsectors) == 0)
1138 goto next;
1139 devsectors >>= 9;
1140
1141 if (st)
1142 st2 = dup_super(st);
1143 else
1144 st2 = guess_super_type(fd, guess_partitions);
1145 if (st2 == NULL ||
1146 st2->ss->load_super(st2, fd, NULL) < 0)
1147 goto next;
1148 st2->ignore_hw_compat = 0;
1149
1150 if (!st) {
1151 /* Check domain policy again, this time referring to metadata */
1152 domain_merge(&domlist, pol2, st2->ss->name);
1153 if (domain_test(domlist, pol, st2->ss->name) != 1)
1154 /* Incompatible devices for this metadata type */
1155 goto next;
1156 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1157 /* Some partition types allow sparing, but not
1158 * this one.
1159 */
1160 goto next;
1161 }
1162
1163 st2->ss->getinfo_super(st2, &info, NULL);
1164 if (info.component_size > devsectors)
1165 /* This partitioning doesn't fit in the device */
1166 goto next;
1167
1168 /* This is an acceptable device to copy partition
1169 * metadata from. We could just stop here, but I
1170 * think I want to keep looking incase a larger
1171 * metadata which makes better use of the device can
1172 * be found.
1173 */
1174 if (chosen == NULL ||
1175 chosen_size < info.component_size) {
1176 chosen_size = info.component_size;
1177 free(chosen);
1178 chosen = devname;
1179 devname = NULL;
1180 if (chosen_st) {
1181 chosen_st->ss->free_super(chosen_st);
1182 free(chosen_st);
1183 }
1184 chosen_st = st2;
1185 st2 = NULL;
1186 }
1187
1188 next:
1189 free(devname);
1190 domain_free(domlist);
1191 dev_policy_free(pol2);
1192 if (st2)
1193 st2->ss->free_super(st2);
1194 free(st2);
1195
1196 if (fd >= 0)
1197 close(fd);
1198 }
1199
1200 closedir(dir);
1201
1202 if (!chosen)
1203 return 1;
1204
1205 /* 'chosen' is the best device we can find. Let's write its
1206 * metadata to devname dfd is read-only so don't use that
1207 */
1208 fd = open(devname, O_RDWR);
1209 if (fd >= 0) {
1210 chosen_st->ss->store_super(chosen_st, fd);
1211 close(fd);
1212 }
1213 free(chosen);
1214 chosen_st->ss->free_super(chosen_st);
1215 free(chosen_st);
1216 return 0;
1217 }
1218
1219 static int is_bare(int dfd)
1220 {
1221 unsigned long long size = 0;
1222 char bufpad[4096 + 4096];
1223 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1224
1225 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1226 read(dfd, buf, 4096) != 4096)
1227 return 0;
1228
1229 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1230 return 0;
1231 if (memcmp(buf, buf+1, 4095) != 0)
1232 return 0;
1233
1234 /* OK, first 4K appear blank, try the end. */
1235 get_dev_size(dfd, NULL, &size);
1236 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1237 read(dfd, buf, 4096) != 4096)
1238 return 0;
1239
1240 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1241 return 0;
1242 if (memcmp(buf, buf+1, 4095) != 0)
1243 return 0;
1244
1245 return 1;
1246 }
1247
1248 /* adding a spare to a regular array is quite different from adding one to
1249 * a set-of-partitions virtual array.
1250 * This function determines which is worth trying and tries as appropriate.
1251 * Arrays are given priority over partitions.
1252 */
1253 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1254 struct map_ent *target,
1255 struct supertype *st, int verbose)
1256 {
1257 int i;
1258 int rv;
1259 int arrays_ok = 0;
1260 int partitions_ok = 0;
1261 int dfd = *dfdp;
1262 int bare;
1263
1264 /* Can only add a spare if device has at least one domain */
1265 if (pol_find(pol, pol_domain) == NULL)
1266 return 1;
1267 /* And only if some action allows spares */
1268 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1269 return 1;
1270
1271 /* Now check if the device is bare.
1272 * bare devices can always be added as a spare
1273 * non-bare devices can only be added if spare-same-slot is permitted,
1274 * and this device is replacing a previous device - in which case 'target'
1275 * will be set.
1276 */
1277 if (!is_bare(dfd)) {
1278 /* Must have a target and allow same_slot */
1279 /* Later - may allow force_spare without target */
1280 if (!target ||
1281 !policy_action_allows(pol, st?st->ss->name:NULL,
1282 act_spare_same_slot)) {
1283 if (verbose > 1)
1284 pr_err("%s is not bare, so not considering as a spare\n",
1285 devname);
1286 return 1;
1287 }
1288 bare = 0;
1289 } else
1290 bare = 1;
1291
1292 /* It might be OK to add this device to an array - need to see
1293 * what arrays might be candidates.
1294 */
1295 if (st) {
1296 /* just try try 'array' or 'partition' based on this metadata */
1297 if (st->ss->add_to_super)
1298 return array_try_spare(devname, dfdp, pol, target, bare,
1299 st, verbose);
1300 else
1301 return partition_try_spare(devname, dfdp, pol,
1302 st, verbose);
1303 }
1304 /* No metadata was specified or found so options are open.
1305 * Check for whether any array metadata, or any partition metadata
1306 * might allow adding the spare. This check is just help to avoid
1307 * a more costly scan of all arrays when we can be sure that will
1308 * fail.
1309 */
1310 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1311 if (superlist[i]->add_to_super && !arrays_ok &&
1312 policy_action_allows(pol, superlist[i]->name, act_spare))
1313 arrays_ok = 1;
1314 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1315 policy_action_allows(pol, superlist[i]->name, act_spare))
1316 partitions_ok = 1;
1317 }
1318 rv = 1;
1319 if (arrays_ok)
1320 rv = array_try_spare(devname, dfdp, pol, target, bare,
1321 st, verbose);
1322 if (rv != 0 && partitions_ok)
1323 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1324 return rv;
1325 }
1326
1327 int IncrementalScan(struct context *c, char *devnm)
1328 {
1329 /* look at every device listed in the 'map' file.
1330 * If one is found that is not running then:
1331 * look in mdadm.conf for bitmap file.
1332 * if one exists, but array has none, add it.
1333 * try to start array in auto-readonly mode
1334 */
1335 struct map_ent *mapl = NULL;
1336 struct map_ent *me;
1337 struct mddev_ident *devs, *mddev;
1338 int rv = 0;
1339 char container[32];
1340 char *only = NULL;
1341
1342 map_read(&mapl);
1343 devs = conf_get_ident(NULL);
1344
1345 restart:
1346 for (me = mapl ; me ; me = me->next) {
1347 mdu_array_info_t array;
1348 struct mdinfo *sra;
1349 int mdfd;
1350
1351 if (devnm && strcmp(devnm, me->devnm) != 0)
1352 continue;
1353 if (me->metadata[0] == '/') {
1354 char *sl;
1355
1356 if (!devnm)
1357 continue;
1358
1359 /* member array, need to work on container */
1360 strncpy(container, me->metadata+1, 32);
1361 container[31] = 0;
1362 sl = strchr(container, '/');
1363 if (sl)
1364 *sl = 0;
1365 only = devnm;
1366 devnm = container;
1367 goto restart;
1368 }
1369 mdfd = open_dev(me->devnm);
1370
1371 if (mdfd < 0)
1372 continue;
1373 if (!isdigit(me->metadata[0])) {
1374 /* must be a container */
1375 struct supertype *st = super_by_fd(mdfd, NULL);
1376 int ret = 0;
1377 struct map_ent *map = NULL;
1378
1379 if (st && st->ss->load_container)
1380 ret = st->ss->load_container(st, mdfd, NULL);
1381 close(mdfd);
1382 if (!ret && st && st->ss->container_content) {
1383 if (map_lock(&map))
1384 pr_err("failed to get exclusive lock on mapfile\n");
1385 ret = Incremental_container(st, me->path, c, only);
1386 map_unlock(&map);
1387 }
1388 if (ret)
1389 rv = 1;
1390 continue;
1391 }
1392 if (md_get_array_info(mdfd, &array) == 0 || errno != ENODEV) {
1393 close(mdfd);
1394 continue;
1395 }
1396 /* Ok, we can try this one. Maybe it needs a bitmap */
1397 for (mddev = devs ; mddev ; mddev = mddev->next)
1398 if (mddev->devname && me->path
1399 && devname_matches(mddev->devname, me->path))
1400 break;
1401 if (mddev && mddev->bitmap_file) {
1402 /*
1403 * Note: early kernels will wrongly fail this, so it
1404 * is a hint only
1405 */
1406 int added = -1;
1407 int bmfd;
1408
1409 bmfd = open(mddev->bitmap_file, O_RDWR);
1410 if (bmfd >= 0) {
1411 added = ioctl(mdfd, SET_BITMAP_FILE, bmfd);
1412 close(bmfd);
1413 }
1414 if (c->verbose >= 0) {
1415 if (added == 0)
1416 pr_err("Added bitmap %s to %s\n",
1417 mddev->bitmap_file, me->path);
1418 else if (errno != EEXIST)
1419 pr_err("Failed to add bitmap to %s: %s\n",
1420 me->path, strerror(errno));
1421 }
1422 }
1423 /* FIXME check for reshape_active and consider not
1424 * starting array.
1425 */
1426 sra = sysfs_read(mdfd, NULL, 0);
1427 if (sra) {
1428 if (sysfs_set_str(sra, NULL,
1429 "array_state", "read-auto") == 0) {
1430 if (c->verbose >= 0)
1431 pr_err("started array %s\n",
1432 me->path ?: me->devnm);
1433 } else {
1434 pr_err("failed to start array %s: %s\n",
1435 me->path ?: me->devnm,
1436 strerror(errno));
1437 rv = 1;
1438 }
1439 sysfs_free(sra);
1440 }
1441 }
1442 return rv;
1443 }
1444
1445 static char *container2devname(char *devname)
1446 {
1447 char *mdname = NULL;
1448
1449 if (devname[0] == '/') {
1450 int fd = open(devname, O_RDONLY);
1451 if (fd >= 0) {
1452 mdname = xstrdup(fd2devnm(fd));
1453 close(fd);
1454 }
1455 } else {
1456 int uuid[4];
1457 struct map_ent *mp, *map = NULL;
1458
1459 if (!parse_uuid(devname, uuid))
1460 return mdname;
1461 mp = map_by_uuid(&map, uuid);
1462 if (mp)
1463 mdname = xstrdup(mp->devnm);
1464 map_free(map);
1465 }
1466
1467 return mdname;
1468 }
1469
1470 static int Incremental_container(struct supertype *st, char *devname,
1471 struct context *c, char *only)
1472 {
1473 /* Collect the contents of this container and for each
1474 * array, choose a device name and assemble the array.
1475 */
1476
1477 struct mdinfo *list;
1478 struct mdinfo *ra;
1479 struct map_ent *map = NULL;
1480 struct mdinfo info;
1481 int trustworthy;
1482 struct mddev_ident *match;
1483 int rv = 0;
1484 struct domainlist *domains;
1485 struct map_ent *smp;
1486 int suuid[4];
1487 int sfd;
1488 int ra_blocked = 0;
1489 int ra_all = 0;
1490 int result = 0;
1491
1492 st->ss->getinfo_super(st, &info, NULL);
1493
1494 if ((c->runstop > 0 && info.container_enough >= 0) ||
1495 info.container_enough > 0)
1496 /* pass */;
1497 else {
1498 if (c->export) {
1499 printf("MD_STARTED=no\n");
1500 } else if (c->verbose)
1501 pr_err("not enough devices to start the container\n");
1502 return 0;
1503 }
1504
1505 match = conf_match(st, &info, devname, c->verbose, &rv);
1506 if (match == NULL && rv == 2)
1507 return rv;
1508
1509 /* Need to compute 'trustworthy' */
1510 if (match)
1511 trustworthy = LOCAL;
1512 else if (st->ss->match_home(st, c->homehost) == 1)
1513 trustworthy = LOCAL;
1514 else if (st->ss->match_home(st, "any") == 1)
1515 trustworthy = LOCAL;
1516 else
1517 trustworthy = FOREIGN;
1518
1519 list = st->ss->container_content(st, NULL);
1520 /* when nothing to activate - quit */
1521 if (list == NULL) {
1522 if (c->export) {
1523 printf("MD_STARTED=nothing\n");
1524 }
1525 return 0;
1526 }
1527 for (ra = list ; ra ; ra = ra->next) {
1528 int mdfd;
1529 char chosen_name[1024];
1530 struct map_ent *mp;
1531 struct mddev_ident *match = NULL;
1532
1533 ra_all++;
1534 /* do not activate arrays blocked by metadata handler */
1535 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1536 pr_err("Cannot activate array %s in %s.\n",
1537 ra->text_version, devname);
1538 ra_blocked++;
1539 continue;
1540 }
1541 mp = map_by_uuid(&map, ra->uuid);
1542
1543 if (mp) {
1544 mdfd = open_dev(mp->devnm);
1545 if (mp->path)
1546 strcpy(chosen_name, mp->path);
1547 else
1548 strcpy(chosen_name, mp->devnm);
1549 } else if (!only) {
1550
1551 /* Check in mdadm.conf for container == devname and
1552 * member == ra->text_version after second slash.
1553 */
1554 char *sub = strchr(ra->text_version+1, '/');
1555 struct mddev_ident *array_list;
1556 if (sub) {
1557 sub++;
1558 array_list = conf_get_ident(NULL);
1559 } else
1560 array_list = NULL;
1561 for(; array_list ; array_list = array_list->next) {
1562 char *dn;
1563 if (array_list->member == NULL ||
1564 array_list->container == NULL)
1565 continue;
1566 if (strcmp(array_list->member, sub) != 0)
1567 continue;
1568 if (array_list->uuid_set &&
1569 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1570 continue;
1571 dn = container2devname(array_list->container);
1572 if (dn == NULL)
1573 continue;
1574 if (strncmp(dn, ra->text_version+1,
1575 strlen(dn)) != 0 ||
1576 ra->text_version[strlen(dn)+1] != '/') {
1577 free(dn);
1578 continue;
1579 }
1580 free(dn);
1581 /* we have a match */
1582 match = array_list;
1583 if (c->verbose>0)
1584 pr_err("match found for member %s\n",
1585 array_list->member);
1586 break;
1587 }
1588
1589 if (match && match->devname &&
1590 strcasecmp(match->devname, "<ignore>") == 0) {
1591 if (c->verbose > 0)
1592 pr_err("array %s/%s is explicitly ignored by mdadm.conf\n",
1593 match->container, match->member);
1594 continue;
1595 }
1596 if (match)
1597 trustworthy = LOCAL;
1598
1599 mdfd = create_mddev(match ? match->devname : NULL,
1600 ra->name,
1601 c->autof,
1602 trustworthy,
1603 chosen_name);
1604 }
1605 if (only && (!mp || strcmp(mp->devnm, only) != 0))
1606 continue;
1607
1608 if (mdfd < 0) {
1609 pr_err("failed to open %s: %s.\n",
1610 chosen_name, strerror(errno));
1611 return 2;
1612 }
1613
1614 assemble_container_content(st, mdfd, ra, c,
1615 chosen_name, &result);
1616 close(mdfd);
1617 }
1618 if (c->export && result) {
1619 char sep = '=';
1620 printf("MD_STARTED");
1621 if (result & INCR_NO) {
1622 printf("%cno", sep);
1623 sep = ',';
1624 }
1625 if (result & INCR_UNSAFE) {
1626 printf("%cunsafe", sep);
1627 sep = ',';
1628 }
1629 if (result & INCR_ALREADY) {
1630 printf("%calready", sep);
1631 sep = ',';
1632 }
1633 if (result & INCR_YES) {
1634 printf("%cyes", sep);
1635 sep = ',';
1636 }
1637 printf("\n");
1638 }
1639
1640 /* don't move spares to container with volume being activated
1641 when all volumes are blocked */
1642 if (ra_all == ra_blocked)
1643 return 0;
1644
1645 /* Now move all suitable spares from spare container */
1646 domains = domain_from_array(list, st->ss->name);
1647 memcpy(suuid, uuid_zero, sizeof(int[4]));
1648 if (domains &&
1649 (smp = map_by_uuid(&map, suuid)) != NULL &&
1650 (sfd = open(smp->path, O_RDONLY)) >= 0) {
1651 /* spare container found */
1652 struct supertype *sst =
1653 super_imsm.match_metadata_desc("imsm");
1654 struct mdinfo *sinfo;
1655 unsigned long long min_size = 0;
1656 if (st->ss->min_acceptable_spare_size)
1657 min_size = st->ss->min_acceptable_spare_size(st);
1658 if (!sst->ss->load_container(sst, sfd, NULL)) {
1659 close(sfd);
1660 sinfo = container_choose_spares(sst, min_size,
1661 domains, NULL,
1662 st->ss->name, 0);
1663 sst->ss->free_super(sst);
1664 if (sinfo){
1665 int count = 0;
1666 struct mdinfo *disks = sinfo->devs;
1667 while (disks) {
1668 /* move spare from spare
1669 * container to currently
1670 * assembled one
1671 */
1672 if (move_spare(
1673 smp->path,
1674 devname,
1675 makedev(disks->disk.major,
1676 disks->disk.minor)))
1677 count++;
1678 disks = disks->next;
1679 }
1680 if (count)
1681 pr_err("Added %d spare%s to %s\n",
1682 count, count>1?"s":"", devname);
1683 }
1684 sysfs_free(sinfo);
1685 } else
1686 close(sfd);
1687 }
1688 domain_free(domains);
1689 return 0;
1690 }
1691
1692 static void run_udisks(char *arg1, char *arg2)
1693 {
1694 int pid = fork();
1695 int status;
1696 if (pid == 0) {
1697 execl("/usr/bin/udisks", "udisks", arg1, arg2, NULL);
1698 execl("/bin/udisks", "udisks", arg1, arg2, NULL);
1699 exit(1);
1700 }
1701 while (pid > 0 && wait(&status) != pid)
1702 ;
1703 }
1704
1705 /*
1706 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1707 * raid arrays, and if so first fail (if needed) and then remove the device.
1708 *
1709 * @devname - The device we want to remove
1710 * @id_path - name as found in /dev/disk/by-path for this device
1711 *
1712 * Note: the device name must be a kernel name like "sda", so
1713 * that we can find it in /proc/mdstat
1714 */
1715 int IncrementalRemove(char *devname, char *id_path, int verbose)
1716 {
1717 int mdfd;
1718 int rv = 0;
1719 struct mdstat_ent *ent;
1720 struct mddev_dev devlist;
1721 struct mdinfo mdi;
1722 char buf[32];
1723
1724 if (!id_path)
1725 dprintf("incremental removal without --path <id_path> lacks the possibility to re-add new device in this port\n");
1726
1727 if (strchr(devname, '/')) {
1728 pr_err("incremental removal requires a kernel device name, not a file: %s\n", devname);
1729 return 1;
1730 }
1731 ent = mdstat_by_component(devname);
1732 if (!ent) {
1733 if (verbose >= 0)
1734 pr_err("%s does not appear to be a component of any array\n", devname);
1735 return 1;
1736 }
1737 sysfs_init(&mdi, -1, ent->devnm);
1738 mdfd = open_dev_excl(ent->devnm);
1739 if (mdfd > 0) {
1740 close(mdfd);
1741 if (sysfs_get_str(&mdi, NULL, "array_state",
1742 buf, sizeof(buf)) > 0) {
1743 if (strncmp(buf, "active", 6) == 0 ||
1744 strncmp(buf, "clean", 5) == 0)
1745 sysfs_set_str(&mdi, NULL,
1746 "array_state", "read-auto");
1747 }
1748 }
1749 mdfd = open_dev(ent->devnm);
1750 if (mdfd < 0) {
1751 if (verbose >= 0)
1752 pr_err("Cannot open array %s!!\n", ent->devnm);
1753 free_mdstat(ent);
1754 return 1;
1755 }
1756
1757 if (id_path) {
1758 struct map_ent *map = NULL, *me;
1759 me = map_by_devnm(&map, ent->devnm);
1760 if (me)
1761 policy_save_path(id_path, me);
1762 map_free(map);
1763 }
1764
1765 memset(&devlist, 0, sizeof(devlist));
1766 devlist.devname = devname;
1767 devlist.disposition = 'f';
1768 /* for a container, we must fail each member array */
1769 if (ent->metadata_version &&
1770 strncmp(ent->metadata_version, "external:", 9) == 0) {
1771 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1772 struct mdstat_ent *memb;
1773 for (memb = mdstat ; memb ; memb = memb->next)
1774 if (is_container_member(memb, ent->devnm)) {
1775 int subfd = open_dev(memb->devnm);
1776 if (subfd >= 0) {
1777 rv |= Manage_subdevs(
1778 memb->devnm, subfd,
1779 &devlist, verbose, 0,
1780 NULL, 0);
1781 close(subfd);
1782 }
1783 }
1784 free_mdstat(mdstat);
1785 } else
1786 rv |= Manage_subdevs(ent->devnm, mdfd, &devlist,
1787 verbose, 0, NULL, 0);
1788 if (rv & 2) {
1789 /* Failed due to EBUSY, try to stop the array.
1790 * Give udisks a chance to unmount it first.
1791 */
1792 int devid = devnm2devid(ent->devnm);
1793 run_udisks("--unmount", map_dev(major(devid),minor(devid), 0));
1794 rv = Manage_stop(ent->devnm, mdfd, verbose, 1);
1795 if (rv)
1796 /* At least we can try to trigger a 'remove' */
1797 sysfs_uevent(&mdi, "remove");
1798 if (verbose) {
1799 if (rv)
1800 pr_err("Fail to stop %s too.\n", ent->devnm);
1801 }
1802 } else {
1803 devlist.disposition = 'r';
1804 rv = Manage_subdevs(ent->devnm, mdfd, &devlist,
1805 verbose, 0, NULL, 0);
1806 }
1807 close(mdfd);
1808 free_mdstat(ent);
1809 return rv;
1810 }
Unnamed repository; edit this file 'description' to name the repository.