]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Incremental.c
imsm: Remove --dump/--restore implementation
[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 dev_t rdev, rdev2;
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 struct dev_policy *policy = NULL;
103 struct map_ent target_array;
104 int have_target;
105 char *devname = devlist->devname;
106 int journal_device_missing = 0;
107
108 struct createinfo *ci = conf_get_create_info();
109
110 if (!stat_is_blkdev(devname, &rdev))
111 return rv;
112 dfd = dev_open(devname, O_RDONLY);
113 if (dfd < 0) {
114 if (c->verbose >= 0)
115 pr_err("cannot open %s: %s.\n",
116 devname, strerror(errno));
117 return rv;
118 }
119 /* If the device is a container, we do something very different */
120 if (must_be_container(dfd)) {
121 if (!st)
122 st = super_by_fd(dfd, NULL);
123 if (st && st->ss->load_container)
124 rv = st->ss->load_container(st, dfd, NULL);
125
126 close(dfd);
127 if (!rv && st->ss->container_content) {
128 if (map_lock(&map))
129 pr_err("failed to get exclusive lock on mapfile\n");
130 if (c->export)
131 printf("MD_DEVNAME=%s\n", devname);
132 rv = Incremental_container(st, devname, c, NULL);
133 map_unlock(&map);
134 return rv;
135 }
136
137 pr_err("%s is not part of an md array.\n",
138 devname);
139 return rv;
140 }
141
142 /* 1/ Check if device is permitted by mdadm.conf */
143
144 for (;devlist; devlist = devlist->next)
145 if (conf_test_dev(devlist->devname))
146 break;
147 if (!devlist) {
148 devlist = conf_get_devs();
149 for (;devlist; devlist = devlist->next) {
150 if (stat_is_blkdev(devlist->devname, &rdev2) &&
151 rdev2 == rdev)
152 break;
153 }
154 }
155 if (!devlist) {
156 if (c->verbose >= 0)
157 pr_err("%s not permitted by mdadm.conf.\n",
158 devname);
159 goto out;
160 }
161
162 /* 2/ Find metadata, reject if none appropriate (check
163 * version/name from args) */
164
165 if (!fstat_is_blkdev(dfd, devname, &rdev))
166 goto out;
167
168 dinfo.disk.major = major(rdev);
169 dinfo.disk.minor = minor(rdev);
170
171 policy = disk_policy(&dinfo);
172 have_target = policy_check_path(&dinfo, &target_array);
173
174 if (st == NULL && (st = guess_super_type(dfd, guess_array)) == NULL) {
175 if (c->verbose >= 0)
176 pr_err("no recognisable superblock on %s.\n",
177 devname);
178 rv = try_spare(devname, &dfd, policy,
179 have_target ? &target_array : NULL,
180 NULL, c->verbose);
181 goto out;
182 }
183 st->ignore_hw_compat = 0;
184
185 if (st->ss->compare_super == NULL ||
186 st->ss->load_super(st, dfd, c->verbose >= 0 ? devname : NULL)) {
187 if (c->verbose >= 0)
188 pr_err("no RAID superblock on %s.\n",
189 devname);
190 rv = try_spare(devname, &dfd, policy,
191 have_target ? &target_array : NULL,
192 st, c->verbose);
193 free(st);
194 goto out;
195 }
196 close (dfd); dfd = -1;
197
198 st->ss->getinfo_super(st, &info, NULL);
199
200 /* 3/ Check if there is a match in mdadm.conf */
201 match = conf_match(st, &info, devname, c->verbose, &rv);
202 if (!match && rv == 2)
203 goto out;
204
205 if (match && match->devname &&
206 strcasecmp(match->devname, "<ignore>") == 0) {
207 if (c->verbose >= 0)
208 pr_err("array containing %s is explicitly ignored by mdadm.conf\n",
209 devname);
210 goto out;
211 }
212
213 /* 3a/ if not, check for homehost match. If no match, continue
214 * but don't trust the 'name' in the array. Thus a 'random' minor
215 * number will be assigned, and the device name will be based
216 * on that. */
217 if (match)
218 trustworthy = LOCAL;
219 else if (st->ss->match_home(st, c->homehost) == 1)
220 trustworthy = LOCAL;
221 else if (st->ss->match_home(st, "any") == 1)
222 trustworthy = LOCAL_ANY;
223 else
224 trustworthy = FOREIGN;
225
226 if (!match && !conf_test_metadata(st->ss->name, policy,
227 (trustworthy == LOCAL))) {
228 if (c->verbose >= 1)
229 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
230 devname, st->ss->name);
231 goto out;
232 }
233 if (trustworthy == LOCAL_ANY)
234 trustworthy = LOCAL;
235
236 /* There are three possible sources for 'autof': command line,
237 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
238 * ARRAY takes precedence, then command line, then
239 * CREATE.
240 */
241 if (match && match->autof)
242 c->autof = match->autof;
243 if (c->autof == 0)
244 c->autof = ci->autof;
245
246 name_to_use = info.name;
247 if (name_to_use[0] == 0 && info.array.level == LEVEL_CONTAINER) {
248 name_to_use = info.text_version;
249 trustworthy = METADATA;
250 }
251 if (name_to_use[0] && trustworthy != LOCAL &&
252 ! c->require_homehost &&
253 conf_name_is_free(name_to_use))
254 trustworthy = LOCAL;
255
256 /* strip "hostname:" prefix from name if we have decided
257 * to treat it as LOCAL
258 */
259 if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
260 name_to_use = strchr(name_to_use, ':')+1;
261
262 /* 4/ Check if array exists.
263 */
264 if (map_lock(&map))
265 pr_err("failed to get exclusive lock on mapfile\n");
266 /* Now check we can get O_EXCL. If not, probably "mdadm -A" has
267 * taken over
268 */
269 dfd = dev_open(devname, O_RDONLY|O_EXCL);
270 if (dfd < 0) {
271 if (c->verbose >= 0)
272 pr_err("cannot reopen %s: %s.\n",
273 devname, strerror(errno));
274 goto out_unlock;
275 }
276 /* Cannot hold it open while we add the device to the array,
277 * so we must release the O_EXCL and depend on the map_lock()
278 * So now is the best time to remove any partitions.
279 */
280 remove_partitions(dfd);
281 close(dfd);
282 dfd = -1;
283
284 mp = map_by_uuid(&map, info.uuid);
285 if (mp)
286 mdfd = open_dev(mp->devnm);
287 else
288 mdfd = -1;
289
290 if (mdfd < 0) {
291
292 /* Skip the clustered ones. This should be started by
293 * clustering resource agents
294 */
295 if (info.array.state & (1 << MD_SB_CLUSTERED))
296 goto out;
297
298 /* Couldn't find an existing array, maybe make a new one */
299 mdfd = create_mddev(match ? match->devname : NULL,
300 name_to_use, c->autof, trustworthy, chosen_name, 0);
301
302 if (mdfd < 0)
303 goto out_unlock;
304
305 if (sysfs_init(&info, mdfd, NULL)) {
306 pr_err("unable to initialize sysfs for %s\n",
307 chosen_name);
308 rv = 2;
309 goto out_unlock;
310 }
311
312 if (set_array_info(mdfd, st, &info) != 0) {
313 pr_err("failed to set array info for %s: %s\n",
314 chosen_name, strerror(errno));
315 rv = 2;
316 goto out_unlock;
317 }
318
319 dinfo = info;
320 dinfo.disk.major = major(rdev);
321 dinfo.disk.minor = minor(rdev);
322 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
323 pr_err("failed to add %s to new array %s: %s.\n",
324 devname, chosen_name, strerror(errno));
325 ioctl(mdfd, STOP_ARRAY, 0);
326 rv = 2;
327 goto out_unlock;
328 }
329 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
330 GET_OFFSET | GET_SIZE));
331
332 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
333 /* It really should be 'none' - must be old buggy
334 * kernel, and mdadm -I may not be able to complete.
335 * So reject it.
336 */
337 ioctl(mdfd, STOP_ARRAY, NULL);
338 pr_err("You have an old buggy kernel which cannot support\n --incremental reliably. Aborting.\n");
339 rv = 2;
340 goto out_unlock;
341 }
342 info.array.working_disks = 1;
343 /* 6/ Make sure /var/run/mdadm.map contains this array. */
344 map_update(&map, fd2devnm(mdfd),
345 info.text_version,
346 info.uuid, chosen_name);
347 } else {
348 /* 5b/ if it does */
349 /* - check one drive in array to make sure metadata is a reasonably */
350 /* close match. Reject if not (e.g. different type) */
351 /* - add the device */
352 char dn[20];
353 int dfd2;
354 int err;
355 struct supertype *st2;
356 struct mdinfo info2, *d;
357
358 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
359 GET_OFFSET | GET_SIZE));
360
361 if (mp->path)
362 strcpy(chosen_name, mp->path);
363 else
364 strcpy(chosen_name, mp->devnm);
365
366 /* It is generally not OK to add non-spare drives to a
367 * running array as they are probably missing because
368 * they failed. However if runstop is 1, then the
369 * array was possibly started early and our best bet is
370 * to add this anyway.
371 * Also if action policy is re-add or better we allow
372 * re-add.
373 * This doesn't apply to containers as the 'non-spare'
374 * flag has a different meaning. The test has to happen
375 * at the device level there
376 */
377 if (!st->ss->external &&
378 (info.disk.state & (1 << MD_DISK_SYNC)) != 0 &&
379 !policy_action_allows(policy, st->ss->name, act_re_add) &&
380 c->runstop < 1) {
381 if (md_array_active(mdfd)) {
382 pr_err("not adding %s to active array (without --run) %s\n",
383 devname, chosen_name);
384 rv = 2;
385 goto out_unlock;
386 }
387 }
388 if (!sra) {
389 rv = 2;
390 goto out_unlock;
391 }
392 if (sra->devs) {
393 sprintf(dn, "%d:%d", sra->devs->disk.major,
394 sra->devs->disk.minor);
395 dfd2 = dev_open(dn, O_RDONLY);
396 if (dfd2 < 0) {
397 pr_err("unable to open %s\n", devname);
398 rv = 2;
399 goto out_unlock;
400 }
401 st2 = dup_super(st);
402 if (st2->ss->load_super(st2, dfd2, NULL) ||
403 st->ss->compare_super(st, st2) != 0) {
404 pr_err("metadata mismatch between %s and chosen array %s\n",
405 devname, chosen_name);
406 close(dfd2);
407 rv = 2;
408 goto out_unlock;
409 }
410 close(dfd2);
411 st2->ss->getinfo_super(st2, &info2, NULL);
412 st2->ss->free_super(st2);
413 if (info.array.level != info2.array.level ||
414 memcmp(info.uuid, info2.uuid, 16) != 0 ||
415 info.array.raid_disks != info2.array.raid_disks) {
416 pr_err("unexpected difference between %s and %s.\n",
417 chosen_name, devname);
418 rv = 2;
419 goto out_unlock;
420 }
421 }
422 info.disk.major = major(rdev);
423 info.disk.minor = minor(rdev);
424 /* add disk needs to know about containers */
425 if (st->ss->external)
426 sra->array.level = LEVEL_CONTAINER;
427
428 if (info.array.state & (1 << MD_SB_CLUSTERED))
429 info.disk.state |= (1 << MD_DISK_CLUSTER_ADD);
430
431 err = add_disk(mdfd, st, sra, &info);
432 if (err < 0 && errno == EBUSY) {
433 /* could be another device present with the same
434 * disk.number. Find and reject any such
435 */
436 find_reject(mdfd, st, sra, info.disk.number,
437 info.events, c->verbose, chosen_name);
438 err = add_disk(mdfd, st, sra, &info);
439 }
440 if (err < 0 && errno == EINVAL &&
441 info.disk.state & (1<<MD_DISK_SYNC)) {
442 /* Maybe it needs to be added as a spare */
443 if (policy_action_allows(policy, st->ss->name,
444 act_force_spare)) {
445 info.disk.state &= ~(1<<MD_DISK_SYNC);
446 err = add_disk(mdfd, st, sra, &info);
447 } else
448 if (c->verbose >= 0)
449 pr_err("can only add %s to %s as a spare, and force-spare is not set.\n",
450 devname, chosen_name);
451 }
452 if (err < 0) {
453 pr_err("failed to add %s to existing array %s: %s.\n",
454 devname, chosen_name, strerror(errno));
455 rv = 2;
456 goto out_unlock;
457 }
458 info.array.working_disks = 0;
459 for (d = sra->devs; d; d=d->next)
460 info.array.working_disks ++;
461
462 }
463 if (strncmp(chosen_name, "/dev/md/", 8) == 0)
464 md_devname = chosen_name+8;
465 else
466 md_devname = chosen_name;
467 if (c->export) {
468 printf("MD_DEVICE=%s\n", fd2devnm(mdfd));
469 printf("MD_DEVNAME=%s\n", md_devname);
470 printf("MD_FOREIGN=%s\n", trustworthy == FOREIGN ? "yes" : "no");
471 }
472
473 /* 7/ Is there enough devices to possibly start the array? */
474 /* 7a/ if not, finish with success. */
475 if (info.array.level == LEVEL_CONTAINER) {
476 char devnm[32];
477 /* Try to assemble within the container */
478 sysfs_uevent(sra, "change");
479 if (!c->export && c->verbose >= 0)
480 pr_err("container %s now has %d device%s\n",
481 chosen_name, info.array.working_disks,
482 info.array.working_disks == 1?"":"s");
483 sysfs_rules_apply(chosen_name, &info);
484 wait_for(chosen_name, mdfd);
485 if (st->ss->external)
486 strcpy(devnm, fd2devnm(mdfd));
487 if (st->ss->load_container)
488 rv = st->ss->load_container(st, mdfd, NULL);
489 close(mdfd);
490 sysfs_free(sra);
491 if (!rv)
492 rv = Incremental_container(st, chosen_name, c, NULL);
493 map_unlock(&map);
494 /* after spare is added, ping monitor for external metadata
495 * so that it can eg. try to rebuild degraded array */
496 if (st->ss->external)
497 ping_monitor(devnm);
498 return rv;
499 }
500
501 /* We have added something to the array, so need to re-read the
502 * state. Eventually this state should be kept up-to-date as
503 * things change.
504 */
505 sysfs_free(sra);
506 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
507 GET_OFFSET | GET_SIZE));
508 active_disks = count_active(st, sra, mdfd, &avail, &info);
509
510 journal_device_missing = (info.journal_device_required) && (info.journal_clean == 0);
511
512 if (info.consistency_policy == CONSISTENCY_POLICY_PPL)
513 info.array.state |= 1;
514
515 if (enough(info.array.level, info.array.raid_disks,
516 info.array.layout, info.array.state & 1, avail) == 0) {
517 if (c->export) {
518 printf("MD_STARTED=no\n");
519 } else if (c->verbose >= 0)
520 pr_err("%s attached to %s, not enough to start (%d).\n",
521 devname, chosen_name, active_disks);
522 rv = 0;
523 goto out_unlock;
524 }
525
526 /* 7b/ if yes, */
527 /* - if number of OK devices match expected, or -R and there */
528 /* are enough, */
529 /* + add any bitmap file */
530 /* + start the array (auto-readonly). */
531
532 if (md_array_active(mdfd)) {
533 if (c->export) {
534 printf("MD_STARTED=already\n");
535 } else if (c->verbose >= 0)
536 pr_err("%s attached to %s which is already active.\n",
537 devname, chosen_name);
538 rv = 0;
539 goto out_unlock;
540 }
541
542 map_unlock(&map);
543 if (c->runstop > 0 || (!journal_device_missing && active_disks >= info.array.working_disks)) {
544 struct mdinfo *dsk;
545 /* Let's try to start it */
546
547 if (journal_device_missing)
548 pr_err("Trying to run with missing journal device\n");
549 if (info.reshape_active && !(info.reshape_active & RESHAPE_NO_BACKUP)) {
550 pr_err("%s: This array is being reshaped and cannot be started\n",
551 chosen_name);
552 cont_err("by --incremental. Please use --assemble\n");
553 goto out;
554 }
555 if (match && match->bitmap_file) {
556 int bmfd = open(match->bitmap_file, O_RDWR);
557 if (bmfd < 0) {
558 pr_err("Could not open bitmap file %s.\n",
559 match->bitmap_file);
560 goto out;
561 }
562 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
563 close(bmfd);
564 pr_err("Failed to set bitmapfile for %s.\n",
565 chosen_name);
566 goto out;
567 }
568 close(bmfd);
569 }
570 /* Need to remove from the array any devices which
571 * 'count_active' discerned were too old or inappropriate
572 */
573 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
574 if (d->disk.state & (1<<MD_DISK_REMOVED))
575 remove_disk(mdfd, st, sra, d);
576
577 if ((sra == NULL || active_disks >= info.array.working_disks) &&
578 trustworthy != FOREIGN)
579 rv = ioctl(mdfd, RUN_ARRAY, NULL);
580 else
581 rv = sysfs_set_str(sra, NULL,
582 "array_state", "read-auto");
583 /* Array might be O_EXCL which will interfere with
584 * fsck and mount. So re-open without O_EXCL.
585 */
586 reopen_mddev(mdfd);
587 if (rv == 0) {
588 if (c->export) {
589 printf("MD_STARTED=yes\n");
590 } else if (c->verbose >= 0)
591 pr_err("%s attached to %s, which has been started.\n",
592 devname, chosen_name);
593 rv = 0;
594 wait_for(chosen_name, mdfd);
595 /* We just started the array, so some devices
596 * might have been evicted from the array
597 * because their event counts were too old.
598 * If the action=re-add policy is in-force for
599 * those devices we should re-add them now.
600 */
601 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
602 if (disk_action_allows(dsk, st->ss->name,
603 act_re_add) &&
604 add_disk(mdfd, st, sra, dsk) == 0)
605 pr_err("%s re-added to %s\n",
606 dsk->sys_name, chosen_name);
607 }
608 } else {
609 pr_err("%s attached to %s, but failed to start: %s.\n",
610 devname, chosen_name, strerror(errno));
611 rv = 1;
612 }
613 } else {
614 if (c->export) {
615 printf("MD_STARTED=unsafe\n");
616 } else if (journal_device_missing) {
617 pr_err("Journal device is missing, not safe to start yet.\n");
618 } else if (c->verbose >= 0)
619 pr_err("%s attached to %s, not enough to start safely.\n",
620 devname, chosen_name);
621 rv = 0;
622 }
623 out:
624 free(avail);
625 if (dfd >= 0)
626 close(dfd);
627 if (mdfd >= 0)
628 close(mdfd);
629 if (policy)
630 dev_policy_free(policy);
631 sysfs_free(sra);
632 return rv;
633 out_unlock:
634 map_unlock(&map);
635 goto out;
636 }
637
638 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
639 int number, __u64 events, int verbose,
640 char *array_name)
641 {
642 /* Find a device attached to this array with a disk.number of number
643 * and events less than the passed events, and remove the device.
644 */
645 struct mdinfo *d;
646
647 if (md_array_active(mdfd))
648 return; /* not safe to remove from active arrays
649 * without thinking more */
650
651 for (d = sra->devs; d ; d = d->next) {
652 char dn[24]; // 2*11 bytes for ints (including sign) + colon + null byte
653 int dfd;
654 struct mdinfo info;
655 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
656 dfd = dev_open(dn, O_RDONLY);
657 if (dfd < 0)
658 continue;
659 if (st->ss->load_super(st, dfd, NULL)) {
660 close(dfd);
661 continue;
662 }
663 st->ss->getinfo_super(st, &info, NULL);
664 st->ss->free_super(st);
665 close(dfd);
666
667 if (info.disk.number != number || info.events >= events)
668 continue;
669
670 if (d->disk.raid_disk > -1)
671 sysfs_set_str(sra, d, "slot", "none");
672 if (sysfs_set_str(sra, d, "state", "remove") == 0)
673 if (verbose >= 0)
674 pr_err("removing old device %s from %s\n",
675 d->sys_name+4, array_name);
676 }
677 }
678
679 static int count_active(struct supertype *st, struct mdinfo *sra,
680 int mdfd, char **availp,
681 struct mdinfo *bestinfo)
682 {
683 /* count how many devices in sra think they are active */
684 struct mdinfo *d;
685 int cnt = 0;
686 int replcnt = 0;
687 __u64 max_events = 0;
688 __u64 max_journal_events = 0;
689 char *avail = NULL;
690 int *best = NULL;
691 char *devmap = NULL;
692 int numdevs = 0;
693 int devnum;
694 int b, i;
695 int raid_disks = 0;
696
697 if (!sra)
698 return 0;
699
700 for (d = sra->devs ; d ; d = d->next)
701 numdevs++;
702 for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
703 char dn[30];
704 int dfd;
705 int ok;
706 struct mdinfo info;
707
708 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
709 dfd = dev_open(dn, O_RDONLY);
710 if (dfd < 0)
711 continue;
712 ok = st->ss->load_super(st, dfd, NULL);
713 close(dfd);
714 if (ok != 0)
715 continue;
716
717 info.array.raid_disks = raid_disks;
718 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
719 if (info.disk.raid_disk == MD_DISK_ROLE_JOURNAL &&
720 info.events > max_journal_events)
721 max_journal_events = info.events;
722 if (!avail) {
723 raid_disks = info.array.raid_disks;
724 avail = xcalloc(raid_disks, 1);
725 *availp = avail;
726
727 best = xcalloc(raid_disks, sizeof(int));
728 devmap = xcalloc(raid_disks, numdevs);
729
730 st->ss->getinfo_super(st, &info, devmap);
731 }
732
733 if (info.disk.state & (1<<MD_DISK_SYNC))
734 {
735 if (cnt == 0) {
736 cnt++;
737 max_events = info.events;
738 avail[info.disk.raid_disk] = 2;
739 best[info.disk.raid_disk] = devnum;
740 st->ss->getinfo_super(st, bestinfo, NULL);
741 } else if (info.events == max_events) {
742 avail[info.disk.raid_disk] = 2;
743 best[info.disk.raid_disk] = devnum;
744 } else if (info.events == max_events-1) {
745 if (avail[info.disk.raid_disk] == 0) {
746 avail[info.disk.raid_disk] = 1;
747 best[info.disk.raid_disk] = devnum;
748 }
749 } else if (info.events < max_events - 1)
750 ;
751 else if (info.events == max_events+1) {
752 int i;
753 max_events = info.events;
754 for (i = 0; i < raid_disks; i++)
755 if (avail[i])
756 avail[i]--;
757 avail[info.disk.raid_disk] = 2;
758 best[info.disk.raid_disk] = devnum;
759 st->ss->getinfo_super(st, bestinfo, NULL);
760 } else { /* info.events much bigger */
761 memset(avail, 0, raid_disks);
762 max_events = info.events;
763 avail[info.disk.raid_disk] = 2;
764 best[info.disk.raid_disk] = devnum;
765 st->ss->getinfo_super(st, bestinfo, NULL);
766 }
767 } else if (info.disk.state & (1<<MD_DISK_REPLACEMENT))
768 replcnt++;
769 st->ss->free_super(st);
770 }
771 if (max_journal_events >= max_events - 1)
772 bestinfo->journal_clean = 1;
773
774 if (!avail)
775 return 0;
776 /* We need to reject any device that thinks the best device is
777 * failed or missing */
778 for (b = 0; b < raid_disks; b++)
779 if (avail[b] == 2)
780 break;
781 cnt = 0;
782 for (i = 0 ; i < raid_disks ; i++) {
783 if (i != b && avail[i])
784 if (devmap[raid_disks * best[i] + b] == 0) {
785 /* This device thinks 'b' is failed -
786 * don't use it */
787 devnum = best[i];
788 for (d=sra->devs ; devnum; d = d->next)
789 devnum--;
790 d->disk.state |= (1 << MD_DISK_REMOVED);
791 avail[i] = 0;
792 }
793 if (avail[i])
794 cnt++;
795 }
796 /* Also need to reject any spare device with an event count that
797 * is too high
798 */
799 for (d = sra->devs; d; d = d->next) {
800 if (!(d->disk.state & (1<<MD_DISK_SYNC)) &&
801 d->events > max_events)
802 d->disk.state |= (1 << MD_DISK_REMOVED);
803 }
804 free(best);
805 free(devmap);
806 return cnt + replcnt;
807 }
808
809 /* test if container has degraded member(s) */
810 static int
811 container_members_max_degradation(struct map_ent *map, struct map_ent *me)
812 {
813 struct mdinfo *sra;
814 int degraded, max_degraded = 0;
815
816 for(; map; map = map->next) {
817 if (!metadata_container_matches(map->metadata, me->devnm))
818 continue;
819 /* most accurate information regarding array degradation */
820 sra = sysfs_read(-1, map->devnm,
821 GET_DISKS | GET_DEVS | GET_STATE);
822 if (!sra)
823 continue;
824 degraded = sra->array.raid_disks - sra->array.active_disks -
825 sra->array.spare_disks;
826 if (degraded > max_degraded)
827 max_degraded = degraded;
828 sysfs_free(sra);
829 }
830
831 return max_degraded;
832 }
833
834 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
835 struct map_ent *target, int bare,
836 struct supertype *st, int verbose)
837 {
838 /* This device doesn't have any md metadata
839 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
840 * If 'st' is not set, then we only know that some metadata allows this,
841 * others possibly don't.
842 * So look for a container or array to attach the device to.
843 * Prefer 'target' if that is set and the array is found.
844 *
845 * If st is set, then only arrays of that type are considered
846 * Return 0 on success, or some exit code on failure, probably 1.
847 */
848 int rv = 1;
849 dev_t rdev;
850 struct map_ent *mp, *map = NULL;
851 struct mdinfo *chosen = NULL;
852 int dfd = *dfdp;
853
854 if (!fstat_is_blkdev(dfd, devname, &rdev))
855 return 1;
856
857 /*
858 * Now we need to find a suitable array to add this to.
859 * We only accept arrays that:
860 * - match 'st'
861 * - are in the same domains as the device
862 * - are of an size for which the device will be useful
863 * and we choose the one that is the most degraded
864 */
865
866 if (map_lock(&map)) {
867 pr_err("failed to get exclusive lock on mapfile\n");
868 return 1;
869 }
870 for (mp = map ; mp ; mp = mp->next) {
871 struct supertype *st2;
872 struct domainlist *dl = NULL;
873 struct mdinfo *sra;
874 unsigned long long devsize, freesize = 0;
875 struct spare_criteria sc = {0, 0};
876
877 if (is_subarray(mp->metadata))
878 continue;
879 if (st) {
880 st2 = st->ss->match_metadata_desc(mp->metadata);
881 if (!st2 ||
882 (st->minor_version >= 0 &&
883 st->minor_version != st2->minor_version)) {
884 if (verbose > 1)
885 pr_err("not adding %s to %s as metadata type doesn't match\n",
886 devname, mp->path);
887 free(st2);
888 continue;
889 }
890 free(st2);
891 }
892 sra = sysfs_read(-1, mp->devnm,
893 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
894 GET_COMPONENT|GET_VERSION);
895 if (sra)
896 sra->array.failed_disks = -1;
897 else
898 continue;
899 if (st == NULL) {
900 int i;
901 st2 = NULL;
902 for(i = 0; !st2 && superlist[i]; i++)
903 st2 = superlist[i]->match_metadata_desc(
904 sra->text_version);
905 if (!st2) {
906 if (verbose > 1)
907 pr_err("not adding %s to %s as metadata not recognised.\n",
908 devname, mp->path);
909 goto next;
910 }
911 /* Need to double check the 'act_spare' permissions applies
912 * to this metadata.
913 */
914 if (!policy_action_allows(pol, st2->ss->name, act_spare))
915 goto next;
916 if (!bare && !policy_action_allows(pol, st2->ss->name,
917 act_spare_same_slot))
918 goto next;
919 } else
920 st2 = st;
921 /* update number of failed disks for mostly degraded
922 * container member */
923 if (sra->array.failed_disks == -1)
924 sra->array.failed_disks = container_members_max_degradation(map, mp);
925
926 get_dev_size(dfd, NULL, &devsize);
927 if (sra->component_size == 0) {
928 /* true for containers, here we must read superblock
929 * to obtain minimum spare size */
930 struct supertype *st3 = dup_super(st2);
931 int mdfd = open_dev(mp->devnm);
932 if (mdfd < 0) {
933 free(st3);
934 goto next;
935 }
936 if (st3->ss->load_container &&
937 !st3->ss->load_container(st3, mdfd, mp->path)) {
938 if (st3->ss->get_spare_criteria)
939 st3->ss->get_spare_criteria(st3, &sc);
940 st3->ss->free_super(st3);
941 }
942 free(st3);
943 close(mdfd);
944 }
945 if ((sra->component_size > 0 &&
946 st2->ss->validate_geometry(st2, sra->array.level, sra->array.layout,
947 sra->array.raid_disks, &sra->array.chunk_size,
948 sra->component_size,
949 sra->devs ? sra->devs->data_offset : INVALID_SECTORS,
950 devname, &freesize, sra->consistency_policy,
951 0) &&
952 freesize < sra->component_size) ||
953 (sra->component_size == 0 && devsize < sc.min_size)) {
954 if (verbose > 1)
955 pr_err("not adding %s to %s as it is too small\n",
956 devname, mp->path);
957 goto next;
958 }
959 /* test against target.
960 * If 'target' is set and 'bare' is false, we only accept
961 * arrays/containers that match 'target'.
962 * If 'target' is set and 'bare' is true, we prefer the
963 * array which matches 'target'.
964 * target is considered only if we deal with degraded array
965 */
966 if (target && policy_action_allows(pol, st2->ss->name,
967 act_spare_same_slot)) {
968 if (strcmp(target->metadata, mp->metadata) == 0 &&
969 memcmp(target->uuid, mp->uuid,
970 sizeof(target->uuid)) == 0 &&
971 sra->array.failed_disks > 0) {
972 /* This is our target!! */
973 sysfs_free(chosen);
974 chosen = sra;
975 sra = NULL;
976 /* skip to end so we don't check any more */
977 while (mp->next)
978 mp = mp->next;
979 goto next;
980 }
981 /* not our target */
982 if (!bare)
983 goto next;
984 }
985
986 dl = domain_from_array(sra, st2->ss->name);
987 if (domain_test(dl, pol, st2->ss->name) != 1) {
988 /* domain test fails */
989 if (verbose > 1)
990 pr_err("not adding %s to %s as it is not in a compatible domain\n",
991 devname, mp->path);
992
993 goto next;
994 }
995 /* all tests passed, OK to add to this array */
996 if (!chosen) {
997 chosen = sra;
998 sra = NULL;
999 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
1000 sysfs_free(chosen);
1001 chosen = sra;
1002 sra = NULL;
1003 }
1004 next:
1005 sysfs_free(sra);
1006 if (st != st2)
1007 free(st2);
1008 if (dl)
1009 domain_free(dl);
1010 }
1011 if (chosen) {
1012 /* add current device to chosen array as a spare */
1013 int mdfd = open_dev(chosen->sys_name);
1014 if (mdfd >= 0) {
1015 struct mddev_dev devlist;
1016 char chosen_devname[24]; // 2*11 for int (including signs) + colon + null
1017 devlist.next = NULL;
1018 devlist.used = 0;
1019 devlist.writemostly = FlagDefault;
1020 devlist.failfast = FlagDefault;
1021 devlist.devname = chosen_devname;
1022 sprintf(chosen_devname, "%d:%d", major(rdev),
1023 minor(rdev));
1024 devlist.disposition = 'a';
1025 close(dfd);
1026 *dfdp = -1;
1027 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1028 -1, 0, NULL, 0);
1029 close(mdfd);
1030 }
1031 if (verbose > 0) {
1032 if (rv == 0)
1033 pr_err("added %s as spare for %s\n",
1034 devname, chosen->sys_name);
1035 else
1036 pr_err("failed to add %s as spare for %s\n",
1037 devname, chosen->sys_name);
1038 }
1039 sysfs_free(chosen);
1040 }
1041 map_unlock(&map);
1042 return rv;
1043 }
1044
1045 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1046 struct supertype *st, int verbose)
1047 {
1048 /* we know that at least one partition virtual-metadata is
1049 * allowed to incorporate spares like this device. We need to
1050 * find a suitable device to copy partition information from.
1051 *
1052 * Getting a list of all disk (not partition) devices is
1053 * slightly non-trivial. We could look at /sys/block, but
1054 * that is theoretically due to be removed. Maybe best to use
1055 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1056 * we depend on this directory of 'path' info. But that fails
1057 * to find loop devices and probably others. Maybe don't
1058 * worry about that, they aren't the real target.
1059 *
1060 * So: check things in /dev/disk/by-path to see if they are in
1061 * a compatible domain, then load the partition table and see
1062 * if it is OK for the new device, and choose the largest
1063 * partition table that fits.
1064 */
1065 DIR *dir;
1066 struct dirent *de;
1067 char *chosen = NULL;
1068 unsigned long long chosen_size = 0;
1069 struct supertype *chosen_st = NULL;
1070 int fd;
1071
1072 dir = opendir("/dev/disk/by-path");
1073 if (!dir)
1074 return 1;
1075 while ((de = readdir(dir)) != NULL) {
1076 char *ep;
1077 struct dev_policy *pol2 = NULL;
1078 struct domainlist *domlist = NULL;
1079 int fd = -1;
1080 struct mdinfo info;
1081 struct supertype *st2 = NULL;
1082 char *devname = NULL;
1083 unsigned long long devsectors;
1084 char *pathlist[2];
1085
1086 if (de->d_ino == 0 || de->d_name[0] == '.' ||
1087 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1088 goto next;
1089
1090 ep = de->d_name + strlen(de->d_name);
1091 while (ep > de->d_name &&
1092 isdigit(ep[-1]))
1093 ep--;
1094 if (ep > de->d_name + 5 &&
1095 strncmp(ep-5, "-part", 5) == 0)
1096 /* This is a partition - skip it */
1097 goto next;
1098
1099 pathlist[0] = de->d_name;
1100 pathlist[1] = NULL;
1101 pol2 = path_policy(pathlist, type_disk);
1102
1103 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1104 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1105 /* new device is incompatible with this device. */
1106 goto next;
1107
1108 domain_free(domlist);
1109 domlist = NULL;
1110
1111 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1112 devname = NULL;
1113 goto next;
1114 }
1115 fd = open(devname, O_RDONLY);
1116 if (fd < 0)
1117 goto next;
1118 if (get_dev_size(fd, devname, &devsectors) == 0)
1119 goto next;
1120 devsectors >>= 9;
1121
1122 if (st)
1123 st2 = dup_super(st);
1124 else
1125 st2 = guess_super_type(fd, guess_partitions);
1126 if (st2 == NULL || st2->ss->load_super(st2, fd, NULL) < 0)
1127 goto next;
1128 st2->ignore_hw_compat = 0;
1129
1130 if (!st) {
1131 /* Check domain policy again, this time referring to metadata */
1132 domain_merge(&domlist, pol2, st2->ss->name);
1133 if (domain_test(domlist, pol, st2->ss->name) != 1)
1134 /* Incompatible devices for this metadata type */
1135 goto next;
1136 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1137 /* Some partition types allow sparing, but not
1138 * this one.
1139 */
1140 goto next;
1141 }
1142
1143 st2->ss->getinfo_super(st2, &info, NULL);
1144 if (info.component_size > devsectors)
1145 /* This partitioning doesn't fit in the device */
1146 goto next;
1147
1148 /* This is an acceptable device to copy partition
1149 * metadata from. We could just stop here, but I
1150 * think I want to keep looking incase a larger
1151 * metadata which makes better use of the device can
1152 * be found.
1153 */
1154 if (chosen == NULL || chosen_size < info.component_size) {
1155 chosen_size = info.component_size;
1156 free(chosen);
1157 chosen = devname;
1158 devname = NULL;
1159 if (chosen_st) {
1160 chosen_st->ss->free_super(chosen_st);
1161 free(chosen_st);
1162 }
1163 chosen_st = st2;
1164 st2 = NULL;
1165 }
1166
1167 next:
1168 free(devname);
1169 domain_free(domlist);
1170 dev_policy_free(pol2);
1171 if (st2)
1172 st2->ss->free_super(st2);
1173 free(st2);
1174
1175 if (fd >= 0)
1176 close(fd);
1177 }
1178
1179 closedir(dir);
1180
1181 if (!chosen)
1182 return 1;
1183
1184 /* 'chosen' is the best device we can find. Let's write its
1185 * metadata to devname dfd is read-only so don't use that
1186 */
1187 fd = open(devname, O_RDWR);
1188 if (fd >= 0) {
1189 chosen_st->ss->store_super(chosen_st, fd);
1190 close(fd);
1191 }
1192 free(chosen);
1193 chosen_st->ss->free_super(chosen_st);
1194 free(chosen_st);
1195 return 0;
1196 }
1197
1198 static int is_bare(int dfd)
1199 {
1200 unsigned long long size = 0;
1201 char bufpad[4096 + 4096];
1202 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1203
1204 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1205 read(dfd, buf, 4096) != 4096)
1206 return 0;
1207
1208 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1209 return 0;
1210 if (memcmp(buf, buf+1, 4095) != 0)
1211 return 0;
1212
1213 /* OK, first 4K appear blank, try the end. */
1214 get_dev_size(dfd, NULL, &size);
1215 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1216 read(dfd, buf, 4096) != 4096)
1217 return 0;
1218
1219 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1220 return 0;
1221 if (memcmp(buf, buf+1, 4095) != 0)
1222 return 0;
1223
1224 return 1;
1225 }
1226
1227 /* adding a spare to a regular array is quite different from adding one to
1228 * a set-of-partitions virtual array.
1229 * This function determines which is worth trying and tries as appropriate.
1230 * Arrays are given priority over partitions.
1231 */
1232 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1233 struct map_ent *target,
1234 struct supertype *st, int verbose)
1235 {
1236 int i;
1237 int rv;
1238 int arrays_ok = 0;
1239 int partitions_ok = 0;
1240 int dfd = *dfdp;
1241 int bare;
1242
1243 /* Can only add a spare if device has at least one domain */
1244 if (pol_find(pol, pol_domain) == NULL)
1245 return 1;
1246 /* And only if some action allows spares */
1247 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1248 return 1;
1249
1250 /* Now check if the device is bare.
1251 * bare devices can always be added as a spare
1252 * non-bare devices can only be added if spare-same-slot is permitted,
1253 * and this device is replacing a previous device - in which case 'target'
1254 * will be set.
1255 */
1256 if (!is_bare(dfd)) {
1257 /* Must have a target and allow same_slot */
1258 /* Later - may allow force_spare without target */
1259 if (!target ||
1260 !policy_action_allows(pol, st?st->ss->name:NULL,
1261 act_spare_same_slot)) {
1262 if (verbose > 1)
1263 pr_err("%s is not bare, so not considering as a spare\n",
1264 devname);
1265 return 1;
1266 }
1267 bare = 0;
1268 } else
1269 bare = 1;
1270
1271 /* It might be OK to add this device to an array - need to see
1272 * what arrays might be candidates.
1273 */
1274 if (st) {
1275 /* just try to add 'array' or 'partition' based on this metadata */
1276 if (st->ss->add_to_super)
1277 return array_try_spare(devname, dfdp, pol, target, bare,
1278 st, verbose);
1279 else
1280 return partition_try_spare(devname, dfdp, pol,
1281 st, verbose);
1282 }
1283 /* No metadata was specified or found so options are open.
1284 * Check for whether any array metadata, or any partition metadata
1285 * might allow adding the spare. This check is just help to avoid
1286 * a more costly scan of all arrays when we can be sure that will
1287 * fail.
1288 */
1289 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1290 if (superlist[i]->add_to_super && !arrays_ok &&
1291 policy_action_allows(pol, superlist[i]->name, act_spare))
1292 arrays_ok = 1;
1293 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1294 policy_action_allows(pol, superlist[i]->name, act_spare))
1295 partitions_ok = 1;
1296 }
1297 rv = 1;
1298 if (arrays_ok)
1299 rv = array_try_spare(devname, dfdp, pol, target, bare,
1300 st, verbose);
1301 if (rv != 0 && partitions_ok)
1302 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1303 return rv;
1304 }
1305
1306 int IncrementalScan(struct context *c, char *devnm)
1307 {
1308 /* look at every device listed in the 'map' file.
1309 * If one is found that is not running then:
1310 * look in mdadm.conf for bitmap file.
1311 * if one exists, but array has none, add it.
1312 * try to start array in auto-readonly mode
1313 */
1314 struct map_ent *mapl = NULL;
1315 struct map_ent *me;
1316 struct mddev_ident *devs, *mddev;
1317 int rv = 0;
1318 char container[32];
1319 char *only = NULL;
1320
1321 map_read(&mapl);
1322 devs = conf_get_ident(NULL);
1323
1324 restart:
1325 for (me = mapl ; me ; me = me->next) {
1326 struct mdinfo *sra;
1327 int mdfd;
1328
1329 if (devnm && strcmp(devnm, me->devnm) != 0)
1330 continue;
1331 if (me->metadata[0] == '/') {
1332 char *sl;
1333
1334 if (!devnm)
1335 continue;
1336
1337 /* member array, need to work on container */
1338 strncpy(container, me->metadata+1, 32);
1339 container[31] = 0;
1340 sl = strchr(container, '/');
1341 if (sl)
1342 *sl = 0;
1343 only = devnm;
1344 devnm = container;
1345 goto restart;
1346 }
1347 mdfd = open_dev(me->devnm);
1348
1349 if (mdfd < 0)
1350 continue;
1351 if (!isdigit(me->metadata[0])) {
1352 /* must be a container */
1353 struct supertype *st = super_by_fd(mdfd, NULL);
1354 int ret = 0;
1355 struct map_ent *map = NULL;
1356
1357 if (st && st->ss->load_container)
1358 ret = st->ss->load_container(st, mdfd, NULL);
1359 close(mdfd);
1360 if (!ret && st && st->ss->container_content) {
1361 if (map_lock(&map))
1362 pr_err("failed to get exclusive lock on mapfile\n");
1363 ret = Incremental_container(st, me->path, c, only);
1364 map_unlock(&map);
1365 }
1366 if (ret)
1367 rv = 1;
1368 continue;
1369 }
1370 if (md_array_active(mdfd)) {
1371 close(mdfd);
1372 continue;
1373 }
1374 /* Ok, we can try this one. Maybe it needs a bitmap */
1375 for (mddev = devs ; mddev ; mddev = mddev->next)
1376 if (mddev->devname && me->path &&
1377 devname_matches(mddev->devname, me->path))
1378 break;
1379 if (mddev && mddev->bitmap_file) {
1380 /*
1381 * Note: early kernels will wrongly fail this, so it
1382 * is a hint only
1383 */
1384 int added = -1;
1385 int bmfd;
1386
1387 bmfd = open(mddev->bitmap_file, O_RDWR);
1388 if (bmfd >= 0) {
1389 added = ioctl(mdfd, SET_BITMAP_FILE, bmfd);
1390 close(bmfd);
1391 }
1392 if (c->verbose >= 0) {
1393 if (added == 0)
1394 pr_err("Added bitmap %s to %s\n",
1395 mddev->bitmap_file, me->path);
1396 else if (errno != EEXIST)
1397 pr_err("Failed to add bitmap to %s: %s\n",
1398 me->path, strerror(errno));
1399 }
1400 }
1401 /* FIXME check for reshape_active and consider not
1402 * starting array.
1403 */
1404 sra = sysfs_read(mdfd, NULL, 0);
1405 if (sra) {
1406 if (sysfs_set_str(sra, NULL,
1407 "array_state", "read-auto") == 0) {
1408 if (c->verbose >= 0)
1409 pr_err("started array %s\n",
1410 me->path ?: me->devnm);
1411 } else {
1412 pr_err("failed to start array %s: %s\n",
1413 me->path ?: me->devnm,
1414 strerror(errno));
1415 rv = 1;
1416 }
1417 sysfs_free(sra);
1418 }
1419 }
1420 map_free(mapl);
1421 return rv;
1422 }
1423
1424 static char *container2devname(char *devname)
1425 {
1426 char *mdname = NULL;
1427
1428 if (devname[0] == '/') {
1429 int fd = open(devname, O_RDONLY);
1430 if (fd >= 0) {
1431 mdname = xstrdup(fd2devnm(fd));
1432 close(fd);
1433 }
1434 } else {
1435 int uuid[4];
1436 struct map_ent *mp, *map = NULL;
1437
1438 if (!parse_uuid(devname, uuid))
1439 return mdname;
1440 mp = map_by_uuid(&map, uuid);
1441 if (mp)
1442 mdname = xstrdup(mp->devnm);
1443 map_free(map);
1444 }
1445
1446 return mdname;
1447 }
1448
1449 static int Incremental_container(struct supertype *st, char *devname,
1450 struct context *c, char *only)
1451 {
1452 /* Collect the contents of this container and for each
1453 * array, choose a device name and assemble the array.
1454 */
1455
1456 struct mdinfo *list;
1457 struct mdinfo *ra;
1458 struct map_ent *map = NULL;
1459 struct mdinfo info;
1460 int trustworthy;
1461 struct mddev_ident *match;
1462 int rv = 0;
1463 struct domainlist *domains;
1464 struct map_ent *smp;
1465 int suuid[4];
1466 int sfd;
1467 int ra_blocked = 0;
1468 int ra_all = 0;
1469 int result = 0;
1470
1471 st->ss->getinfo_super(st, &info, NULL);
1472
1473 if ((c->runstop > 0 && info.container_enough >= 0) ||
1474 info.container_enough > 0)
1475 /* pass */;
1476 else {
1477 if (c->export) {
1478 printf("MD_STARTED=no\n");
1479 } else if (c->verbose)
1480 pr_err("not enough devices to start the container\n");
1481 return 0;
1482 }
1483
1484 match = conf_match(st, &info, devname, c->verbose, &rv);
1485 if (match == NULL && rv == 2)
1486 return rv;
1487
1488 /* Need to compute 'trustworthy' */
1489 if (match)
1490 trustworthy = LOCAL;
1491 else if (st->ss->match_home(st, c->homehost) == 1)
1492 trustworthy = LOCAL;
1493 else if (st->ss->match_home(st, "any") == 1)
1494 trustworthy = LOCAL;
1495 else
1496 trustworthy = FOREIGN;
1497
1498 list = st->ss->container_content(st, NULL);
1499 /* when nothing to activate - quit */
1500 if (list == NULL) {
1501 if (c->export) {
1502 printf("MD_STARTED=nothing\n");
1503 }
1504 return 0;
1505 }
1506 for (ra = list ; ra ; ra = ra->next) {
1507 int mdfd;
1508 char chosen_name[1024];
1509 struct map_ent *mp;
1510 struct mddev_ident *match = NULL;
1511
1512 ra_all++;
1513 /* do not activate arrays blocked by metadata handler */
1514 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1515 pr_err("Cannot activate array %s in %s.\n",
1516 ra->text_version, devname);
1517 ra_blocked++;
1518 continue;
1519 }
1520 mp = map_by_uuid(&map, ra->uuid);
1521
1522 if (mp) {
1523 mdfd = open_dev(mp->devnm);
1524 if (mp->path)
1525 strcpy(chosen_name, mp->path);
1526 else
1527 strcpy(chosen_name, mp->devnm);
1528 } else if (!only) {
1529
1530 /* Check in mdadm.conf for container == devname and
1531 * member == ra->text_version after second slash.
1532 */
1533 char *sub = strchr(ra->text_version+1, '/');
1534 struct mddev_ident *array_list;
1535 if (sub) {
1536 sub++;
1537 array_list = conf_get_ident(NULL);
1538 } else
1539 array_list = NULL;
1540 for(; array_list ; array_list = array_list->next) {
1541 char *dn;
1542 if (array_list->member == NULL ||
1543 array_list->container == NULL)
1544 continue;
1545 if (strcmp(array_list->member, sub) != 0)
1546 continue;
1547 if (array_list->uuid_set &&
1548 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1549 continue;
1550 dn = container2devname(array_list->container);
1551 if (dn == NULL)
1552 continue;
1553 if (strncmp(dn, ra->text_version+1,
1554 strlen(dn)) != 0 ||
1555 ra->text_version[strlen(dn)+1] != '/') {
1556 free(dn);
1557 continue;
1558 }
1559 free(dn);
1560 /* we have a match */
1561 match = array_list;
1562 if (c->verbose>0)
1563 pr_err("match found for member %s\n",
1564 array_list->member);
1565 break;
1566 }
1567
1568 if (match && match->devname &&
1569 strcasecmp(match->devname, "<ignore>") == 0) {
1570 if (c->verbose > 0)
1571 pr_err("array %s/%s is explicitly ignored by mdadm.conf\n",
1572 match->container, match->member);
1573 continue;
1574 }
1575 if (match)
1576 trustworthy = LOCAL;
1577
1578 mdfd = create_mddev(match ? match->devname : NULL,
1579 ra->name,
1580 c->autof,
1581 trustworthy,
1582 chosen_name, 0);
1583 }
1584 if (only && (!mp || strcmp(mp->devnm, only) != 0))
1585 continue;
1586
1587 if (mdfd < 0) {
1588 pr_err("failed to open %s: %s.\n",
1589 chosen_name, strerror(errno));
1590 return 2;
1591 }
1592
1593 assemble_container_content(st, mdfd, ra, c,
1594 chosen_name, &result);
1595 map_free(map);
1596 map = NULL;
1597 close(mdfd);
1598 }
1599 if (c->export && result) {
1600 char sep = '=';
1601 printf("MD_STARTED");
1602 if (result & INCR_NO) {
1603 printf("%cno", sep);
1604 sep = ',';
1605 }
1606 if (result & INCR_UNSAFE) {
1607 printf("%cunsafe", sep);
1608 sep = ',';
1609 }
1610 if (result & INCR_ALREADY) {
1611 printf("%calready", sep);
1612 sep = ',';
1613 }
1614 if (result & INCR_YES) {
1615 printf("%cyes", sep);
1616 sep = ',';
1617 }
1618 printf("\n");
1619 }
1620
1621 /* don't move spares to container with volume being activated
1622 when all volumes are blocked */
1623 if (ra_all == ra_blocked)
1624 return 0;
1625
1626 /* Now move all suitable spares from spare container */
1627 domains = domain_from_array(list, st->ss->name);
1628 memcpy(suuid, uuid_zero, sizeof(int[4]));
1629 if (domains &&
1630 (smp = map_by_uuid(&map, suuid)) != NULL &&
1631 (sfd = open(smp->path, O_RDONLY)) >= 0) {
1632 /* spare container found */
1633 struct supertype *sst =
1634 super_imsm.match_metadata_desc("imsm");
1635 struct mdinfo *sinfo;
1636
1637 if (!sst->ss->load_container(sst, sfd, NULL)) {
1638 struct spare_criteria sc = {0, 0};
1639
1640 if (st->ss->get_spare_criteria)
1641 st->ss->get_spare_criteria(st, &sc);
1642
1643 close(sfd);
1644 sinfo = container_choose_spares(sst, &sc,
1645 domains, NULL,
1646 st->ss->name, 0);
1647 sst->ss->free_super(sst);
1648 if (sinfo){
1649 int count = 0;
1650 struct mdinfo *disks = sinfo->devs;
1651 while (disks) {
1652 /* move spare from spare
1653 * container to currently
1654 * assembled one
1655 */
1656 if (move_spare(
1657 smp->path,
1658 devname,
1659 makedev(disks->disk.major,
1660 disks->disk.minor)))
1661 count++;
1662 disks = disks->next;
1663 }
1664 if (count)
1665 pr_err("Added %d spare%s to %s\n",
1666 count, count>1?"s":"", devname);
1667 }
1668 sysfs_free(sinfo);
1669 } else
1670 close(sfd);
1671 }
1672 domain_free(domains);
1673 map_free(map);
1674 return 0;
1675 }
1676
1677 static void run_udisks(char *arg1, char *arg2)
1678 {
1679 int pid = fork();
1680 int status;
1681 if (pid == 0) {
1682 execl("/usr/bin/udisks", "udisks", arg1, arg2, NULL);
1683 execl("/bin/udisks", "udisks", arg1, arg2, NULL);
1684 exit(1);
1685 }
1686 while (pid > 0 && wait(&status) != pid)
1687 ;
1688 }
1689
1690 static int force_remove(char *devnm, int fd, struct mdinfo *mdi, int verbose)
1691 {
1692 int rv;
1693 int devid = devnm2devid(devnm);
1694
1695 run_udisks("--unmount", map_dev(major(devid), minor(devid), 0));
1696 rv = Manage_stop(devnm, fd, verbose, 1);
1697 if (rv) {
1698 /* At least we can try to trigger a 'remove' */
1699 sysfs_uevent(mdi, "remove");
1700 if (verbose)
1701 pr_err("Fail to stop %s too.\n", devnm);
1702 }
1703 return rv;
1704 }
1705
1706 static void remove_from_member_array(struct mdstat_ent *memb,
1707 struct mddev_dev *devlist, int verbose)
1708 {
1709 int rv;
1710 struct mdinfo mmdi;
1711 int subfd = open_dev(memb->devnm);
1712
1713 if (subfd >= 0) {
1714 rv = Manage_subdevs(memb->devnm, subfd, devlist, verbose,
1715 0, NULL, 0);
1716 if (rv & 2) {
1717 if (sysfs_init(&mmdi, -1, memb->devnm))
1718 pr_err("unable to initialize sysfs for: %s\n",
1719 memb->devnm);
1720 else
1721 force_remove(memb->devnm, subfd, &mmdi,
1722 verbose);
1723 }
1724 close(subfd);
1725 }
1726 }
1727
1728 /*
1729 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1730 * raid arrays, and if so first fail (if needed) and then remove the device.
1731 *
1732 * @devname - The device we want to remove
1733 * @id_path - name as found in /dev/disk/by-path for this device
1734 *
1735 * Note: the device name must be a kernel name like "sda", so
1736 * that we can find it in /proc/mdstat
1737 */
1738 int IncrementalRemove(char *devname, char *id_path, int verbose)
1739 {
1740 int mdfd;
1741 int rv = 0;
1742 struct mdstat_ent *ent;
1743 struct mddev_dev devlist;
1744 struct mdinfo mdi;
1745 char buf[32];
1746
1747 if (!id_path)
1748 dprintf("incremental removal without --path <id_path> lacks the possibility to re-add new device in this port\n");
1749
1750 if (strchr(devname, '/')) {
1751 pr_err("incremental removal requires a kernel device name, not a file: %s\n", devname);
1752 return 1;
1753 }
1754 ent = mdstat_by_component(devname);
1755 if (!ent) {
1756 if (verbose >= 0)
1757 pr_err("%s does not appear to be a component of any array\n", devname);
1758 return 1;
1759 }
1760 if (sysfs_init(&mdi, -1, ent->devnm)) {
1761 pr_err("unable to initialize sysfs for: %s\n", devname);
1762 return 1;
1763 }
1764 mdfd = open_dev_excl(ent->devnm);
1765 if (mdfd > 0) {
1766 close(mdfd);
1767 if (sysfs_get_str(&mdi, NULL, "array_state",
1768 buf, sizeof(buf)) > 0) {
1769 if (strncmp(buf, "active", 6) == 0 ||
1770 strncmp(buf, "clean", 5) == 0)
1771 sysfs_set_str(&mdi, NULL,
1772 "array_state", "read-auto");
1773 }
1774 }
1775 mdfd = open_dev(ent->devnm);
1776 if (mdfd < 0) {
1777 if (verbose >= 0)
1778 pr_err("Cannot open array %s!!\n", ent->devnm);
1779 free_mdstat(ent);
1780 return 1;
1781 }
1782
1783 if (id_path) {
1784 struct map_ent *map = NULL, *me;
1785 me = map_by_devnm(&map, ent->devnm);
1786 if (me)
1787 policy_save_path(id_path, me);
1788 map_free(map);
1789 }
1790
1791 memset(&devlist, 0, sizeof(devlist));
1792 devlist.devname = devname;
1793 devlist.disposition = 'f';
1794 /* for a container, we must fail each member array */
1795 if (ent->metadata_version &&
1796 strncmp(ent->metadata_version, "external:", 9) == 0) {
1797 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1798 struct mdstat_ent *memb;
1799 for (memb = mdstat ; memb ; memb = memb->next) {
1800 if (is_container_member(memb, ent->devnm))
1801 remove_from_member_array(memb,
1802 &devlist, verbose);
1803 }
1804 free_mdstat(mdstat);
1805 } else {
1806 rv |= Manage_subdevs(ent->devnm, mdfd, &devlist,
1807 verbose, 0, NULL, 0);
1808 if (rv & 2) {
1809 /* Failed due to EBUSY, try to stop the array.
1810 * Give udisks a chance to unmount it first.
1811 */
1812 rv = force_remove(ent->devnm, mdfd, &mdi, verbose);
1813 goto end;
1814 }
1815 }
1816
1817 devlist.disposition = 'r';
1818 rv = Manage_subdevs(ent->devnm, mdfd, &devlist,
1819 verbose, 0, NULL, 0);
1820 end:
1821 close(mdfd);
1822 free_mdstat(ent);
1823 return rv;
1824 }