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