]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Incremental.c - support --incremental. Part of: | |
3 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
4 | * | |
5 | * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de> | |
6 | * | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify | |
9 | * it under the terms of the GNU General Public License as published by | |
10 | * the Free Software Foundation; either version 2 of the License, or | |
11 | * (at your option) any later version. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
21 | * | |
22 | * Author: Neil Brown | |
23 | * Email: <neilb@suse.de> | |
24 | * Paper: Neil Brown | |
25 | * Novell Inc | |
26 | * GPO Box Q1283 | |
27 | * QVB Post Office, NSW 1230 | |
28 | * Australia | |
29 | */ | |
30 | ||
31 | #include "mdadm.h" | |
32 | #include <dirent.h> | |
33 | #include <ctype.h> | |
34 | ||
35 | static int count_active(struct supertype *st, struct mdinfo *sra, | |
36 | int mdfd, char **availp, | |
37 | struct mdinfo *info); | |
38 | static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra, | |
39 | int number, __u64 events, int verbose, | |
40 | char *array_name); | |
41 | static int try_spare(char *devname, int *dfdp, struct dev_policy *pol, | |
42 | struct map_ent *target, | |
43 | struct supertype *st, int verbose); | |
44 | ||
45 | static int Incremental_container(struct supertype *st, char *devname, | |
46 | char *homehost, | |
47 | int verbose, int runstop, int autof, | |
48 | int freeze_reshape); | |
49 | ||
50 | int Incremental(char *devname, int verbose, int runstop, | |
51 | struct supertype *st, char *homehost, int require_homehost, | |
52 | int autof, int freeze_reshape) | |
53 | { | |
54 | /* Add this device to an array, creating the array if necessary | |
55 | * and starting the array if sensible or - if runstop>0 - if possible. | |
56 | * | |
57 | * This has several steps: | |
58 | * | |
59 | * 1/ Check if device is permitted by mdadm.conf, reject if not. | |
60 | * 2/ Find metadata, reject if none appropriate (check | |
61 | * version/name from args) | |
62 | * 3/ Check if there is a match in mdadm.conf | |
63 | * 3a/ if not, check for homehost match. If no match, assemble as | |
64 | * a 'foreign' array. | |
65 | * 4/ Determine device number. | |
66 | * - If in mdadm.conf with std name, use that | |
67 | * - UUID in /var/run/mdadm.map use that | |
68 | * - If name is suggestive, use that. unless in use with different uuid. | |
69 | * - Choose a free, high number. | |
70 | * - Use a partitioned device unless strong suggestion not to. | |
71 | * e.g. auto=md | |
72 | * Don't choose partitioned for containers. | |
73 | * 5/ Find out if array already exists | |
74 | * 5a/ if it does not | |
75 | * - choose a name, from mdadm.conf or 'name' field in array. | |
76 | * - create the array | |
77 | * - add the device | |
78 | * 5b/ if it does | |
79 | * - check one drive in array to make sure metadata is a reasonably | |
80 | * close match. Reject if not (e.g. different type) | |
81 | * - add the device | |
82 | * 6/ Make sure /var/run/mdadm.map contains this array. | |
83 | * 7/ Is there enough devices to possibly start the array? | |
84 | * For a container, this means running Incremental_container. | |
85 | * 7a/ if not, finish with success. | |
86 | * 7b/ if yes, | |
87 | * - read all metadata and arrange devices like -A does | |
88 | * - if number of OK devices match expected, or -R and there are enough, | |
89 | * start the array (auto-readonly). | |
90 | */ | |
91 | struct stat stb; | |
92 | struct mdinfo info, dinfo; | |
93 | struct mdinfo *sra = NULL, *d; | |
94 | struct mddev_ident *match; | |
95 | char chosen_name[1024]; | |
96 | int rv = 1; | |
97 | struct map_ent *mp, *map = NULL; | |
98 | int dfd = -1, mdfd = -1; | |
99 | char *avail = NULL; | |
100 | int active_disks; | |
101 | int trustworthy; | |
102 | char *name_to_use; | |
103 | mdu_array_info_t ainf; | |
104 | struct dev_policy *policy = NULL; | |
105 | struct map_ent target_array; | |
106 | int have_target; | |
107 | ||
108 | struct createinfo *ci = conf_get_create_info(); | |
109 | ||
110 | if (stat(devname, &stb) < 0) { | |
111 | if (verbose >= 0) | |
112 | pr_err("stat failed for %s: %s.\n", | |
113 | devname, strerror(errno)); | |
114 | return rv; | |
115 | } | |
116 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
117 | if (verbose >= 0) | |
118 | pr_err("%s is not a block device.\n", | |
119 | devname); | |
120 | return rv; | |
121 | } | |
122 | dfd = dev_open(devname, O_RDONLY|O_EXCL); | |
123 | if (dfd < 0) { | |
124 | if (verbose >= 0) | |
125 | pr_err("cannot open %s: %s.\n", | |
126 | devname, strerror(errno)); | |
127 | return rv; | |
128 | } | |
129 | /* If the device is a container, we do something very different */ | |
130 | if (must_be_container(dfd)) { | |
131 | if (!st) | |
132 | st = super_by_fd(dfd, NULL); | |
133 | if (st && st->ss->load_container) | |
134 | rv = st->ss->load_container(st, dfd, NULL); | |
135 | ||
136 | close(dfd); | |
137 | if (!rv && st->ss->container_content) { | |
138 | if (map_lock(&map)) | |
139 | pr_err("failed to get " | |
140 | "exclusive lock on mapfile\n"); | |
141 | rv = Incremental_container(st, devname, homehost, | |
142 | verbose, runstop, autof, | |
143 | freeze_reshape); | |
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 | if (!conf_test_dev(devname)) { | |
156 | if (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(dfd, &stb) < 0) { | |
166 | if (verbose >= 0) | |
167 | pr_err("fstat failed for %s: %s.\n", | |
168 | devname, strerror(errno)); | |
169 | goto out; | |
170 | } | |
171 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
172 | if (verbose >= 0) | |
173 | pr_err("%s is not a block device.\n", | |
174 | devname); | |
175 | goto out; | |
176 | } | |
177 | ||
178 | dinfo.disk.major = major(stb.st_rdev); | |
179 | dinfo.disk.minor = minor(stb.st_rdev); | |
180 | ||
181 | policy = disk_policy(&dinfo); | |
182 | have_target = policy_check_path(&dinfo, &target_array); | |
183 | ||
184 | if (st == NULL && (st = guess_super(dfd)) == NULL) { | |
185 | if (verbose >= 0) | |
186 | pr_err("no recognisable superblock on %s.\n", | |
187 | devname); | |
188 | rv = try_spare(devname, &dfd, policy, | |
189 | have_target ? &target_array : NULL, | |
190 | st, verbose); | |
191 | goto out; | |
192 | } | |
193 | if (st->ss->compare_super == NULL || | |
194 | st->ss->load_super(st, dfd, NULL)) { | |
195 | if (verbose >= 0) | |
196 | pr_err("no RAID superblock on %s.\n", | |
197 | devname); | |
198 | rv = try_spare(devname, &dfd, policy, | |
199 | have_target ? &target_array : NULL, | |
200 | st, verbose); | |
201 | free(st); | |
202 | goto out; | |
203 | } | |
204 | close (dfd); dfd = -1; | |
205 | ||
206 | st->ss->getinfo_super(st, &info, NULL); | |
207 | ||
208 | /* 3/ Check if there is a match in mdadm.conf */ | |
209 | match = conf_match(st, &info, devname, verbose, &rv); | |
210 | if (!match && rv == 2) | |
211 | goto out; | |
212 | ||
213 | if (match && match->devname | |
214 | && strcasecmp(match->devname, "<ignore>") == 0) { | |
215 | if (verbose >= 0) | |
216 | pr_err("array containing %s is explicitly" | |
217 | " ignored by mdadm.conf\n", | |
218 | devname); | |
219 | goto out; | |
220 | } | |
221 | ||
222 | /* 3a/ if not, check for homehost match. If no match, continue | |
223 | * but don't trust the 'name' in the array. Thus a 'random' minor | |
224 | * number will be assigned, and the device name will be based | |
225 | * on that. */ | |
226 | if (match) | |
227 | trustworthy = LOCAL; | |
228 | else if (st->ss->match_home(st, homehost) == 1) | |
229 | trustworthy = LOCAL; | |
230 | else if (st->ss->match_home(st, "any") == 1) | |
231 | trustworthy = LOCAL_ANY; | |
232 | else | |
233 | trustworthy = FOREIGN; | |
234 | ||
235 | ||
236 | if (!match && !conf_test_metadata(st->ss->name, policy, | |
237 | (trustworthy == LOCAL))) { | |
238 | if (verbose >= 1) | |
239 | pr_err("%s has metadata type %s for which " | |
240 | "auto-assembly is disabled\n", | |
241 | devname, st->ss->name); | |
242 | goto out; | |
243 | } | |
244 | if (trustworthy == LOCAL_ANY) | |
245 | trustworthy = LOCAL; | |
246 | ||
247 | /* There are three possible sources for 'autof': command line, | |
248 | * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf. | |
249 | * ARRAY takes precedence, then command line, then | |
250 | * CREATE. | |
251 | */ | |
252 | if (match && match->autof) | |
253 | autof = match->autof; | |
254 | if (autof == 0) | |
255 | autof = ci->autof; | |
256 | ||
257 | name_to_use = info.name; | |
258 | if (name_to_use[0] == 0 && | |
259 | info.array.level == LEVEL_CONTAINER) { | |
260 | name_to_use = info.text_version; | |
261 | trustworthy = METADATA; | |
262 | } | |
263 | if (name_to_use[0] && trustworthy != LOCAL && | |
264 | ! require_homehost && | |
265 | conf_name_is_free(name_to_use)) | |
266 | trustworthy = LOCAL; | |
267 | ||
268 | /* strip "hostname:" prefix from name if we have decided | |
269 | * to treat it as LOCAL | |
270 | */ | |
271 | if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL) | |
272 | name_to_use = strchr(name_to_use, ':')+1; | |
273 | ||
274 | /* 4/ Check if array exists. | |
275 | */ | |
276 | if (map_lock(&map)) | |
277 | pr_err("failed to get exclusive lock on " | |
278 | "mapfile\n"); | |
279 | mp = map_by_uuid(&map, info.uuid); | |
280 | if (mp) | |
281 | mdfd = open_dev(mp->devnum); | |
282 | else | |
283 | mdfd = -1; | |
284 | ||
285 | if (mdfd < 0) { | |
286 | ||
287 | /* Couldn't find an existing array, maybe make a new one */ | |
288 | mdfd = create_mddev(match ? match->devname : NULL, | |
289 | name_to_use, autof, trustworthy, chosen_name); | |
290 | ||
291 | if (mdfd < 0) | |
292 | goto out_unlock; | |
293 | ||
294 | sysfs_init(&info, mdfd, 0); | |
295 | ||
296 | if (set_array_info(mdfd, st, &info) != 0) { | |
297 | pr_err("failed to set array info for %s: %s\n", | |
298 | chosen_name, strerror(errno)); | |
299 | rv = 2; | |
300 | goto out_unlock; | |
301 | } | |
302 | ||
303 | dinfo = info; | |
304 | dinfo.disk.major = major(stb.st_rdev); | |
305 | dinfo.disk.minor = minor(stb.st_rdev); | |
306 | if (add_disk(mdfd, st, &info, &dinfo) != 0) { | |
307 | pr_err("failed to add %s to %s: %s.\n", | |
308 | devname, chosen_name, strerror(errno)); | |
309 | ioctl(mdfd, STOP_ARRAY, 0); | |
310 | rv = 2; | |
311 | goto out_unlock; | |
312 | } | |
313 | sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE | | |
314 | GET_OFFSET | GET_SIZE)); | |
315 | ||
316 | if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) { | |
317 | /* It really should be 'none' - must be old buggy | |
318 | * kernel, and mdadm -I may not be able to complete. | |
319 | * So reject it. | |
320 | */ | |
321 | ioctl(mdfd, STOP_ARRAY, NULL); | |
322 | pr_err("You have an old buggy kernel which cannot support\n" | |
323 | " --incremental reliably. Aborting.\n"); | |
324 | rv = 2; | |
325 | goto out_unlock; | |
326 | } | |
327 | info.array.working_disks = 1; | |
328 | /* 6/ Make sure /var/run/mdadm.map contains this array. */ | |
329 | map_update(&map, fd2devnum(mdfd), | |
330 | info.text_version, | |
331 | info.uuid, chosen_name); | |
332 | } else { | |
333 | /* 5b/ if it does */ | |
334 | /* - check one drive in array to make sure metadata is a reasonably */ | |
335 | /* close match. Reject if not (e.g. different type) */ | |
336 | /* - add the device */ | |
337 | char dn[20]; | |
338 | int dfd2; | |
339 | int err; | |
340 | struct supertype *st2; | |
341 | struct mdinfo info2, *d; | |
342 | ||
343 | sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE | | |
344 | GET_OFFSET | GET_SIZE)); | |
345 | ||
346 | if (mp->path) | |
347 | strcpy(chosen_name, mp->path); | |
348 | else | |
349 | strcpy(chosen_name, devnum2devname(mp->devnum)); | |
350 | ||
351 | /* It is generally not OK to add non-spare drives to a | |
352 | * running array as they are probably missing because | |
353 | * they failed. However if runstop is 1, then the | |
354 | * array was possibly started early and our best bet is | |
355 | * to add this anyway. | |
356 | * Also if action policy is re-add or better we allow | |
357 | * re-add. | |
358 | * This doesn't apply to containers as the 'non-spare' | |
359 | * flag has a different meaning. The test has to happen | |
360 | * at the device level there | |
361 | */ | |
362 | if (!st->ss->external | |
363 | && (info.disk.state & (1<<MD_DISK_SYNC)) != 0 | |
364 | && ! policy_action_allows(policy, st->ss->name, | |
365 | act_re_add) | |
366 | && runstop < 1) { | |
367 | if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) { | |
368 | pr_err("not adding %s to active array (without --run) %s\n", | |
369 | devname, chosen_name); | |
370 | rv = 2; | |
371 | goto out_unlock; | |
372 | } | |
373 | } | |
374 | if (!sra) { | |
375 | rv = 2; | |
376 | goto out_unlock; | |
377 | } | |
378 | if (sra->devs) { | |
379 | sprintf(dn, "%d:%d", sra->devs->disk.major, | |
380 | sra->devs->disk.minor); | |
381 | dfd2 = dev_open(dn, O_RDONLY); | |
382 | if (dfd2 < 0) { | |
383 | pr_err("unable to open %s\n", devname); | |
384 | rv = 2; | |
385 | goto out_unlock; | |
386 | } | |
387 | st2 = dup_super(st); | |
388 | if (st2->ss->load_super(st2, dfd2, NULL) || | |
389 | st->ss->compare_super(st, st2) != 0) { | |
390 | pr_err("metadata mismatch between %s and " | |
391 | "chosen array %s\n", | |
392 | devname, chosen_name); | |
393 | close(dfd2); | |
394 | rv = 2; | |
395 | goto out_unlock; | |
396 | } | |
397 | close(dfd2); | |
398 | st2->ss->getinfo_super(st2, &info2, NULL); | |
399 | st2->ss->free_super(st2); | |
400 | if (info.array.level != info2.array.level || | |
401 | memcmp(info.uuid, info2.uuid, 16) != 0 || | |
402 | info.array.raid_disks != info2.array.raid_disks) { | |
403 | pr_err("unexpected difference between %s and %s.\n", | |
404 | chosen_name, devname); | |
405 | rv = 2; | |
406 | goto out_unlock; | |
407 | } | |
408 | } | |
409 | info.disk.major = major(stb.st_rdev); | |
410 | info.disk.minor = minor(stb.st_rdev); | |
411 | /* add disk needs to know about containers */ | |
412 | if (st->ss->external) | |
413 | sra->array.level = LEVEL_CONTAINER; | |
414 | err = add_disk(mdfd, st, sra, &info); | |
415 | if (err < 0 && errno == EBUSY) { | |
416 | /* could be another device present with the same | |
417 | * disk.number. Find and reject any such | |
418 | */ | |
419 | find_reject(mdfd, st, sra, info.disk.number, | |
420 | info.events, verbose, chosen_name); | |
421 | err = add_disk(mdfd, st, sra, &info); | |
422 | } | |
423 | if (err < 0) { | |
424 | pr_err("failed to add %s to %s: %s.\n", | |
425 | devname, chosen_name, strerror(errno)); | |
426 | rv = 2; | |
427 | goto out_unlock; | |
428 | } | |
429 | info.array.working_disks = 0; | |
430 | for (d = sra->devs; d; d=d->next) | |
431 | info.array.working_disks ++; | |
432 | ||
433 | } | |
434 | ||
435 | /* 7/ Is there enough devices to possibly start the array? */ | |
436 | /* 7a/ if not, finish with success. */ | |
437 | if (info.array.level == LEVEL_CONTAINER) { | |
438 | int devnum = devnum; /* defined and used iff ->external */ | |
439 | /* Try to assemble within the container */ | |
440 | sysfs_uevent(sra, "change"); | |
441 | if (verbose >= 0) | |
442 | pr_err("container %s now has %d device%s\n", | |
443 | chosen_name, info.array.working_disks, | |
444 | info.array.working_disks==1?"":"s"); | |
445 | wait_for(chosen_name, mdfd); | |
446 | if (st->ss->external) | |
447 | devnum = fd2devnum(mdfd); | |
448 | if (st->ss->load_container) | |
449 | rv = st->ss->load_container(st, mdfd, NULL); | |
450 | close(mdfd); | |
451 | sysfs_free(sra); | |
452 | if (!rv) | |
453 | rv = Incremental_container(st, chosen_name, homehost, | |
454 | verbose, runstop, autof, | |
455 | freeze_reshape); | |
456 | map_unlock(&map); | |
457 | if (rv == 1) | |
458 | /* Don't fail the whole -I if a subarray didn't | |
459 | * have enough devices to start yet | |
460 | */ | |
461 | rv = 0; | |
462 | /* after spare is added, ping monitor for external metadata | |
463 | * so that it can eg. try to rebuild degraded array */ | |
464 | if (st->ss->external) | |
465 | ping_monitor_by_id(devnum); | |
466 | return rv; | |
467 | } | |
468 | ||
469 | /* We have added something to the array, so need to re-read the | |
470 | * state. Eventually this state should be kept up-to-date as | |
471 | * things change. | |
472 | */ | |
473 | sysfs_free(sra); | |
474 | sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE | | |
475 | GET_OFFSET | GET_SIZE)); | |
476 | active_disks = count_active(st, sra, mdfd, &avail, &info); | |
477 | if (enough(info.array.level, info.array.raid_disks, | |
478 | info.array.layout, info.array.state & 1, | |
479 | avail) == 0) { | |
480 | if (verbose >= 0) | |
481 | pr_err("%s attached to %s, not enough to start (%d).\n", | |
482 | devname, chosen_name, active_disks); | |
483 | rv = 0; | |
484 | goto out_unlock; | |
485 | } | |
486 | ||
487 | /* 7b/ if yes, */ | |
488 | /* - if number of OK devices match expected, or -R and there */ | |
489 | /* are enough, */ | |
490 | /* + add any bitmap file */ | |
491 | /* + start the array (auto-readonly). */ | |
492 | ||
493 | if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) { | |
494 | if (verbose >= 0) | |
495 | pr_err("%s attached to %s which is already active.\n", | |
496 | devname, chosen_name); | |
497 | rv = 0; | |
498 | goto out_unlock; | |
499 | } | |
500 | ||
501 | map_unlock(&map); | |
502 | if (runstop > 0 || active_disks >= info.array.working_disks) { | |
503 | struct mdinfo *dsk; | |
504 | /* Let's try to start it */ | |
505 | if (match && match->bitmap_file) { | |
506 | int bmfd = open(match->bitmap_file, O_RDWR); | |
507 | if (bmfd < 0) { | |
508 | pr_err("Could not open bitmap file %s.\n", | |
509 | match->bitmap_file); | |
510 | goto out; | |
511 | } | |
512 | if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) { | |
513 | close(bmfd); | |
514 | pr_err("Failed to set bitmapfile for %s.\n", | |
515 | chosen_name); | |
516 | goto out; | |
517 | } | |
518 | close(bmfd); | |
519 | } | |
520 | /* Need to remove from the array any devices which | |
521 | * 'count_active' discerned were too old or inappropriate | |
522 | */ | |
523 | for (d = sra ? sra->devs : NULL ; d ; d = d->next) | |
524 | if (d->disk.state & (1<<MD_DISK_REMOVED)) | |
525 | remove_disk(mdfd, st, sra, d); | |
526 | ||
527 | if ((sra == NULL || active_disks >= info.array.working_disks) | |
528 | && trustworthy != FOREIGN) | |
529 | rv = ioctl(mdfd, RUN_ARRAY, NULL); | |
530 | else | |
531 | rv = sysfs_set_str(sra, NULL, | |
532 | "array_state", "read-auto"); | |
533 | if (rv == 0) { | |
534 | if (verbose >= 0) | |
535 | pr_err("%s attached to %s, which has been started.\n", | |
536 | devname, chosen_name); | |
537 | rv = 0; | |
538 | wait_for(chosen_name, mdfd); | |
539 | /* We just started the array, so some devices | |
540 | * might have been evicted from the array | |
541 | * because their event counts were too old. | |
542 | * If the action=re-add policy is in-force for | |
543 | * those devices we should re-add them now. | |
544 | */ | |
545 | for (dsk = sra->devs; dsk ; dsk = dsk->next) { | |
546 | if (disk_action_allows(dsk, st->ss->name, act_re_add) && | |
547 | add_disk(mdfd, st, sra, dsk) == 0) | |
548 | pr_err("%s re-added to %s\n", | |
549 | dsk->sys_name, chosen_name); | |
550 | } | |
551 | } else { | |
552 | pr_err("%s attached to %s, but failed to start: %s.\n", | |
553 | devname, chosen_name, strerror(errno)); | |
554 | rv = 1; | |
555 | } | |
556 | } else { | |
557 | if (verbose >= 0) | |
558 | pr_err("%s attached to %s, not enough to start safely.\n", | |
559 | devname, chosen_name); | |
560 | rv = 0; | |
561 | } | |
562 | out: | |
563 | free(avail); | |
564 | if (dfd >= 0) | |
565 | close(dfd); | |
566 | if (mdfd >= 0) | |
567 | close(mdfd); | |
568 | if (policy) | |
569 | dev_policy_free(policy); | |
570 | if (sra) | |
571 | sysfs_free(sra); | |
572 | return rv; | |
573 | out_unlock: | |
574 | map_unlock(&map); | |
575 | goto out; | |
576 | } | |
577 | ||
578 | static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra, | |
579 | int number, __u64 events, int verbose, | |
580 | char *array_name) | |
581 | { | |
582 | /* Find a device attached to this array with a disk.number of number | |
583 | * and events less than the passed events, and remove the device. | |
584 | */ | |
585 | struct mdinfo *d; | |
586 | mdu_array_info_t ra; | |
587 | ||
588 | if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0) | |
589 | return; /* not safe to remove from active arrays | |
590 | * without thinking more */ | |
591 | ||
592 | for (d = sra->devs; d ; d = d->next) { | |
593 | char dn[10]; | |
594 | int dfd; | |
595 | struct mdinfo info; | |
596 | sprintf(dn, "%d:%d", d->disk.major, d->disk.minor); | |
597 | dfd = dev_open(dn, O_RDONLY); | |
598 | if (dfd < 0) | |
599 | continue; | |
600 | if (st->ss->load_super(st, dfd, NULL)) { | |
601 | close(dfd); | |
602 | continue; | |
603 | } | |
604 | st->ss->getinfo_super(st, &info, NULL); | |
605 | st->ss->free_super(st); | |
606 | close(dfd); | |
607 | ||
608 | if (info.disk.number != number || | |
609 | info.events >= events) | |
610 | continue; | |
611 | ||
612 | if (d->disk.raid_disk > -1) | |
613 | sysfs_set_str(sra, d, "slot", "none"); | |
614 | if (sysfs_set_str(sra, d, "state", "remove") == 0) | |
615 | if (verbose >= 0) | |
616 | pr_err("removing old device %s from %s\n", | |
617 | d->sys_name+4, array_name); | |
618 | } | |
619 | } | |
620 | ||
621 | static int count_active(struct supertype *st, struct mdinfo *sra, | |
622 | int mdfd, char **availp, | |
623 | struct mdinfo *bestinfo) | |
624 | { | |
625 | /* count how many devices in sra think they are active */ | |
626 | struct mdinfo *d; | |
627 | int cnt = 0; | |
628 | __u64 max_events = 0; | |
629 | char *avail = NULL; | |
630 | int *best = NULL; | |
631 | char *devmap = NULL; | |
632 | int numdevs = 0; | |
633 | int devnum; | |
634 | int b, i; | |
635 | int raid_disks = 0; | |
636 | ||
637 | if (!sra) | |
638 | return 0; | |
639 | ||
640 | for (d = sra->devs ; d ; d = d->next) | |
641 | numdevs++; | |
642 | for (d = sra->devs, devnum=0 ; d ; d = d->next, devnum++) { | |
643 | char dn[30]; | |
644 | int dfd; | |
645 | int ok; | |
646 | struct mdinfo info; | |
647 | ||
648 | sprintf(dn, "%d:%d", d->disk.major, d->disk.minor); | |
649 | dfd = dev_open(dn, O_RDONLY); | |
650 | if (dfd < 0) | |
651 | continue; | |
652 | ok = st->ss->load_super(st, dfd, NULL); | |
653 | close(dfd); | |
654 | if (ok != 0) | |
655 | continue; | |
656 | info.array.raid_disks = raid_disks; | |
657 | st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum); | |
658 | if (!avail) { | |
659 | raid_disks = info.array.raid_disks; | |
660 | avail = calloc(raid_disks, 1); | |
661 | if (!avail) { | |
662 | pr_err("out of memory.\n"); | |
663 | exit(1); | |
664 | } | |
665 | *availp = avail; | |
666 | ||
667 | best = calloc(raid_disks, sizeof(int)); | |
668 | devmap = calloc(raid_disks * numdevs, 1); | |
669 | ||
670 | st->ss->getinfo_super(st, &info, devmap); | |
671 | } | |
672 | ||
673 | if (info.disk.state & (1<<MD_DISK_SYNC)) | |
674 | { | |
675 | if (cnt == 0) { | |
676 | cnt++; | |
677 | max_events = info.events; | |
678 | avail[info.disk.raid_disk] = 2; | |
679 | best[info.disk.raid_disk] = devnum; | |
680 | st->ss->getinfo_super(st, bestinfo, NULL); | |
681 | } else if (info.events == max_events) { | |
682 | avail[info.disk.raid_disk] = 2; | |
683 | best[info.disk.raid_disk] = devnum; | |
684 | } else if (info.events == max_events-1) { | |
685 | if (avail[info.disk.raid_disk] == 0) { | |
686 | avail[info.disk.raid_disk] = 1; | |
687 | best[info.disk.raid_disk] = devnum; | |
688 | } | |
689 | } else if (info.events < max_events - 1) | |
690 | ; | |
691 | else if (info.events == max_events+1) { | |
692 | int i; | |
693 | max_events = info.events; | |
694 | for (i=0; i < raid_disks; i++) | |
695 | if (avail[i]) | |
696 | avail[i]--; | |
697 | avail[info.disk.raid_disk] = 2; | |
698 | best[info.disk.raid_disk] = devnum; | |
699 | st->ss->getinfo_super(st, bestinfo, NULL); | |
700 | } else { /* info.events much bigger */ | |
701 | memset(avail, 0, raid_disks); | |
702 | max_events = info.events; | |
703 | avail[info.disk.raid_disk] = 2; | |
704 | best[info.disk.raid_disk] = devnum; | |
705 | st->ss->getinfo_super(st, bestinfo, NULL); | |
706 | } | |
707 | } | |
708 | st->ss->free_super(st); | |
709 | } | |
710 | if (!avail) | |
711 | return 0; | |
712 | /* We need to reject any device that thinks the best device is | |
713 | * failed or missing */ | |
714 | for (b = 0; b < raid_disks; b++) | |
715 | if (avail[b] == 2) | |
716 | break; | |
717 | cnt = 0; | |
718 | for (i = 0 ; i < raid_disks ; i++) { | |
719 | if (i != b && avail[i]) | |
720 | if (devmap[raid_disks * best[i] + b] == 0) { | |
721 | /* This device thinks 'b' is failed - | |
722 | * don't use it */ | |
723 | devnum = best[i]; | |
724 | for (d=sra->devs ; devnum; d = d->next) | |
725 | devnum--; | |
726 | d->disk.state |= (1 << MD_DISK_REMOVED); | |
727 | avail[i] = 0; | |
728 | } | |
729 | if (avail[i]) | |
730 | cnt++; | |
731 | } | |
732 | free(best); | |
733 | free(devmap); | |
734 | return cnt; | |
735 | } | |
736 | ||
737 | /* test if container has degraded member(s) */ | |
738 | static int container_members_max_degradation(struct map_ent *map, struct map_ent *me) | |
739 | { | |
740 | mdu_array_info_t array; | |
741 | int afd; | |
742 | int max_degraded = 0; | |
743 | ||
744 | for(; map; map = map->next) { | |
745 | if (!is_subarray(map->metadata) || | |
746 | devname2devnum(map->metadata+1) != me->devnum) | |
747 | continue; | |
748 | afd = open_dev(map->devnum); | |
749 | if (afd < 0) | |
750 | continue; | |
751 | /* most accurate information regarding array degradation */ | |
752 | if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) { | |
753 | int degraded = array.raid_disks - array.active_disks - | |
754 | array.spare_disks; | |
755 | if (degraded > max_degraded) | |
756 | max_degraded = degraded; | |
757 | } | |
758 | close(afd); | |
759 | } | |
760 | return (max_degraded); | |
761 | } | |
762 | ||
763 | static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol, | |
764 | struct map_ent *target, int bare, | |
765 | struct supertype *st, int verbose) | |
766 | { | |
767 | /* This device doesn't have any md metadata | |
768 | * The device policy allows 'spare' and if !bare, it allows spare-same-slot. | |
769 | * If 'st' is not set, then we only know that some metadata allows this, | |
770 | * others possibly don't. | |
771 | * So look for a container or array to attach the device to. | |
772 | * Prefer 'target' if that is set and the array is found. | |
773 | * | |
774 | * If st is set, then only arrays of that type are considered | |
775 | * Return 0 on success, or some exit code on failure, probably 1. | |
776 | */ | |
777 | int rv = 1; | |
778 | struct stat stb; | |
779 | struct map_ent *mp, *map = NULL; | |
780 | struct mdinfo *chosen = NULL; | |
781 | int dfd = *dfdp; | |
782 | ||
783 | if (fstat(dfd, &stb) != 0) | |
784 | return 1; | |
785 | ||
786 | /* | |
787 | * Now we need to find a suitable array to add this to. | |
788 | * We only accept arrays that: | |
789 | * - match 'st' | |
790 | * - are in the same domains as the device | |
791 | * - are of an size for which the device will be useful | |
792 | * and we choose the one that is the most degraded | |
793 | */ | |
794 | ||
795 | if (map_lock(&map)) { | |
796 | pr_err("failed to get exclusive lock on " | |
797 | "mapfile\n"); | |
798 | return 1; | |
799 | } | |
800 | for (mp = map ; mp ; mp = mp->next) { | |
801 | struct supertype *st2; | |
802 | struct domainlist *dl = NULL; | |
803 | struct mdinfo *sra; | |
804 | unsigned long long devsize; | |
805 | unsigned long long component_size = 0; | |
806 | ||
807 | if (is_subarray(mp->metadata)) | |
808 | continue; | |
809 | if (st) { | |
810 | st2 = st->ss->match_metadata_desc(mp->metadata); | |
811 | if (!st2 || | |
812 | (st->minor_version >= 0 && | |
813 | st->minor_version != st2->minor_version)) { | |
814 | if (verbose > 1) | |
815 | pr_err("not adding %s to %s as metadata type doesn't match\n", | |
816 | devname, mp->path); | |
817 | free(st2); | |
818 | continue; | |
819 | } | |
820 | free(st2); | |
821 | } | |
822 | sra = sysfs_read(-1, mp->devnum, | |
823 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE| | |
824 | GET_DEGRADED|GET_COMPONENT|GET_VERSION); | |
825 | if (!sra) { | |
826 | /* Probably a container - no degraded info */ | |
827 | sra = sysfs_read(-1, mp->devnum, | |
828 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE| | |
829 | GET_COMPONENT|GET_VERSION); | |
830 | if (sra) | |
831 | sra->array.failed_disks = -1; | |
832 | } | |
833 | if (!sra) | |
834 | continue; | |
835 | if (st == NULL) { | |
836 | int i; | |
837 | st2 = NULL; | |
838 | for(i=0; !st2 && superlist[i]; i++) | |
839 | st2 = superlist[i]->match_metadata_desc( | |
840 | sra->text_version); | |
841 | if (!st2) { | |
842 | if (verbose > 1) | |
843 | pr_err("not adding %s to %s" | |
844 | " as metadata not recognised.\n", | |
845 | devname, mp->path); | |
846 | goto next; | |
847 | } | |
848 | /* Need to double check the 'act_spare' permissions applies | |
849 | * to this metadata. | |
850 | */ | |
851 | if (!policy_action_allows(pol, st2->ss->name, act_spare)) | |
852 | goto next; | |
853 | if (!bare && !policy_action_allows(pol, st2->ss->name, | |
854 | act_spare_same_slot)) | |
855 | goto next; | |
856 | } else | |
857 | st2 = st; | |
858 | /* update number of failed disks for mostly degraded | |
859 | * container member */ | |
860 | if (sra->array.failed_disks == -1) | |
861 | sra->array.failed_disks = container_members_max_degradation(map, mp); | |
862 | ||
863 | get_dev_size(dfd, NULL, &devsize); | |
864 | if (sra->component_size == 0) { | |
865 | /* true for containers, here we must read superblock | |
866 | * to obtain minimum spare size */ | |
867 | struct supertype *st3 = dup_super(st2); | |
868 | int mdfd = open_dev(mp->devnum); | |
869 | if (mdfd < 0) { | |
870 | free(st3); | |
871 | goto next; | |
872 | } | |
873 | if (st3->ss->load_container && | |
874 | !st3->ss->load_container(st3, mdfd, mp->path)) { | |
875 | component_size = st3->ss->min_acceptable_spare_size(st3); | |
876 | st3->ss->free_super(st3); | |
877 | } | |
878 | free(st3); | |
879 | close(mdfd); | |
880 | } | |
881 | if ((sra->component_size > 0 && | |
882 | st2->ss->avail_size(st2, devsize) < sra->component_size) | |
883 | || | |
884 | (sra->component_size == 0 && devsize < component_size)) { | |
885 | if (verbose > 1) | |
886 | pr_err("not adding %s to %s as it is too small\n", | |
887 | devname, mp->path); | |
888 | goto next; | |
889 | } | |
890 | /* test against target. | |
891 | * If 'target' is set and 'bare' is false, we only accept | |
892 | * arrays/containers that match 'target'. | |
893 | * If 'target' is set and 'bare' is true, we prefer the | |
894 | * array which matches 'target'. | |
895 | * target is considered only if we deal with degraded array | |
896 | */ | |
897 | if (target && policy_action_allows(pol, st2->ss->name, | |
898 | act_spare_same_slot)) { | |
899 | if (strcmp(target->metadata, mp->metadata) == 0 && | |
900 | memcmp(target->uuid, mp->uuid, | |
901 | sizeof(target->uuid)) == 0 && | |
902 | sra->array.failed_disks > 0) { | |
903 | /* This is our target!! */ | |
904 | if (chosen) | |
905 | sysfs_free(chosen); | |
906 | chosen = sra; | |
907 | sra = NULL; | |
908 | /* skip to end so we don't check any more */ | |
909 | while (mp->next) | |
910 | mp = mp->next; | |
911 | goto next; | |
912 | } | |
913 | /* not our target */ | |
914 | if (!bare) | |
915 | goto next; | |
916 | } | |
917 | ||
918 | dl = domain_from_array(sra, st2->ss->name); | |
919 | if (domain_test(dl, pol, st2->ss->name) != 1) { | |
920 | /* domain test fails */ | |
921 | if (verbose > 1) | |
922 | pr_err("not adding %s to %s as" | |
923 | " it is not in a compatible domain\n", | |
924 | devname, mp->path); | |
925 | ||
926 | goto next; | |
927 | } | |
928 | /* all tests passed, OK to add to this array */ | |
929 | if (!chosen) { | |
930 | chosen = sra; | |
931 | sra = NULL; | |
932 | } else if (chosen->array.failed_disks < sra->array.failed_disks) { | |
933 | sysfs_free(chosen); | |
934 | chosen = sra; | |
935 | sra = NULL; | |
936 | } | |
937 | next: | |
938 | if (sra) | |
939 | sysfs_free(sra); | |
940 | if (st != st2) | |
941 | free(st2); | |
942 | if (dl) | |
943 | domain_free(dl); | |
944 | } | |
945 | if (chosen) { | |
946 | /* add current device to chosen array as a spare */ | |
947 | int mdfd = open_dev(devname2devnum(chosen->sys_name)); | |
948 | if (mdfd >= 0) { | |
949 | struct mddev_dev devlist; | |
950 | char devname[20]; | |
951 | devlist.next = NULL; | |
952 | devlist.used = 0; | |
953 | devlist.writemostly = 0; | |
954 | devlist.devname = devname; | |
955 | sprintf(devname, "%d:%d", major(stb.st_rdev), | |
956 | minor(stb.st_rdev)); | |
957 | devlist.disposition = 'a'; | |
958 | close(dfd); | |
959 | *dfdp = -1; | |
960 | rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist, | |
961 | -1, 0, NULL, 0); | |
962 | close(mdfd); | |
963 | } | |
964 | if (verbose > 0) { | |
965 | if (rv == 0) | |
966 | pr_err("added %s as spare for %s\n", | |
967 | devname, chosen->sys_name); | |
968 | else | |
969 | pr_err("failed to add %s as spare for %s\n", | |
970 | devname, chosen->sys_name); | |
971 | } | |
972 | sysfs_free(chosen); | |
973 | } | |
974 | map_unlock(&map); | |
975 | return rv; | |
976 | } | |
977 | ||
978 | static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol, | |
979 | struct supertype *st, int verbose) | |
980 | { | |
981 | /* we know that at least one partition virtual-metadata is | |
982 | * allowed to incorporate spares like this device. We need to | |
983 | * find a suitable device to copy partition information from. | |
984 | * | |
985 | * Getting a list of all disk (not partition) devices is | |
986 | * slightly non-trivial. We could look at /sys/block, but | |
987 | * that is theoretically due to be removed. Maybe best to use | |
988 | * /dev/disk/by-path/?* and ignore names ending '-partNN' as | |
989 | * we depend on this directory of 'path' info. But that fails | |
990 | * to find loop devices and probably others. Maybe don't | |
991 | * worry about that, they aren't the real target. | |
992 | * | |
993 | * So: check things in /dev/disk/by-path to see if they are in | |
994 | * a compatible domain, then load the partition table and see | |
995 | * if it is OK for the new device, and choose the largest | |
996 | * partition table that fits. | |
997 | */ | |
998 | DIR *dir; | |
999 | struct dirent *de; | |
1000 | char *chosen = NULL; | |
1001 | unsigned long long chosen_size = 0; | |
1002 | struct supertype *chosen_st = NULL; | |
1003 | int fd; | |
1004 | ||
1005 | dir = opendir("/dev/disk/by-path"); | |
1006 | if (!dir) | |
1007 | return 1; | |
1008 | while ((de = readdir(dir)) != NULL) { | |
1009 | char *ep; | |
1010 | struct dev_policy *pol2 = NULL; | |
1011 | struct domainlist *domlist = NULL; | |
1012 | int fd = -1; | |
1013 | struct mdinfo info; | |
1014 | struct supertype *st2 = NULL; | |
1015 | char *devname = NULL; | |
1016 | unsigned long long devsectors; | |
1017 | ||
1018 | if (de->d_ino == 0 || | |
1019 | de->d_name[0] == '.' || | |
1020 | (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN)) | |
1021 | goto next; | |
1022 | ||
1023 | ep = de->d_name + strlen(de->d_name); | |
1024 | while (ep > de->d_name && | |
1025 | isdigit(ep[-1])) | |
1026 | ep--; | |
1027 | if (ep > de->d_name + 5 && | |
1028 | strncmp(ep-5, "-part", 5) == 0) | |
1029 | /* This is a partition - skip it */ | |
1030 | goto next; | |
1031 | ||
1032 | pol2 = path_policy(de->d_name, type_disk); | |
1033 | ||
1034 | domain_merge(&domlist, pol2, st ? st->ss->name : NULL); | |
1035 | if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1) | |
1036 | /* new device is incompatible with this device. */ | |
1037 | goto next; | |
1038 | ||
1039 | domain_free(domlist); | |
1040 | domlist = NULL; | |
1041 | ||
1042 | if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) { | |
1043 | devname = NULL; | |
1044 | goto next; | |
1045 | } | |
1046 | fd = open(devname, O_RDONLY); | |
1047 | if (fd < 0) | |
1048 | goto next; | |
1049 | if (get_dev_size(fd, devname, &devsectors) == 0) | |
1050 | goto next; | |
1051 | devsectors >>= 9; | |
1052 | ||
1053 | if (st) | |
1054 | st2 = dup_super(st); | |
1055 | else | |
1056 | st2 = guess_super_type(fd, guess_partitions); | |
1057 | if (st2 == NULL || | |
1058 | st2->ss->load_super(st2, fd, NULL) < 0) | |
1059 | goto next; | |
1060 | ||
1061 | if (!st) { | |
1062 | /* Check domain policy again, this time referring to metadata */ | |
1063 | domain_merge(&domlist, pol2, st2->ss->name); | |
1064 | if (domain_test(domlist, pol, st2->ss->name) != 1) | |
1065 | /* Incompatible devices for this metadata type */ | |
1066 | goto next; | |
1067 | if (!policy_action_allows(pol, st2->ss->name, act_spare)) | |
1068 | /* Some partition types allow sparing, but not | |
1069 | * this one. | |
1070 | */ | |
1071 | goto next; | |
1072 | } | |
1073 | ||
1074 | st2->ss->getinfo_super(st2, &info, NULL); | |
1075 | if (info.component_size > devsectors) | |
1076 | /* This partitioning doesn't fit in the device */ | |
1077 | goto next; | |
1078 | ||
1079 | /* This is an acceptable device to copy partition | |
1080 | * metadata from. We could just stop here, but I | |
1081 | * think I want to keep looking incase a larger | |
1082 | * metadata which makes better use of the device can | |
1083 | * be found. | |
1084 | */ | |
1085 | if (chosen == NULL || | |
1086 | chosen_size < info.component_size) { | |
1087 | chosen_size = info.component_size; | |
1088 | free(chosen); | |
1089 | chosen = devname; | |
1090 | devname = NULL; | |
1091 | if (chosen_st) { | |
1092 | chosen_st->ss->free_super(chosen_st); | |
1093 | free(chosen_st); | |
1094 | } | |
1095 | chosen_st = st2; | |
1096 | st2 = NULL; | |
1097 | } | |
1098 | ||
1099 | next: | |
1100 | free(devname); | |
1101 | domain_free(domlist); | |
1102 | dev_policy_free(pol2); | |
1103 | if (st2) | |
1104 | st2->ss->free_super(st2); | |
1105 | free(st2); | |
1106 | ||
1107 | if (fd >= 0) | |
1108 | close(fd); | |
1109 | } | |
1110 | ||
1111 | closedir(dir); | |
1112 | ||
1113 | if (!chosen) | |
1114 | return 1; | |
1115 | ||
1116 | /* 'chosen' is the best device we can find. Let's write its | |
1117 | * metadata to devname dfd is read-only so don't use that | |
1118 | */ | |
1119 | fd = open(devname, O_RDWR); | |
1120 | if (fd >= 0) { | |
1121 | chosen_st->ss->store_super(chosen_st, fd); | |
1122 | close(fd); | |
1123 | } | |
1124 | free(chosen); | |
1125 | chosen_st->ss->free_super(chosen_st); | |
1126 | free(chosen_st); | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | static int is_bare(int dfd) | |
1131 | { | |
1132 | unsigned long long size = 0; | |
1133 | char bufpad[4096 + 4096]; | |
1134 | char *buf = (char*)(((long)bufpad + 4096) & ~4095); | |
1135 | ||
1136 | if (lseek(dfd, 0, SEEK_SET) != 0 || | |
1137 | read(dfd, buf, 4096) != 4096) | |
1138 | return 0; | |
1139 | ||
1140 | if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff') | |
1141 | return 0; | |
1142 | if (memcmp(buf, buf+1, 4095) != 0) | |
1143 | return 0; | |
1144 | ||
1145 | /* OK, first 4K appear blank, try the end. */ | |
1146 | get_dev_size(dfd, NULL, &size); | |
1147 | if (lseek(dfd, size-4096, SEEK_SET) < 0 || | |
1148 | read(dfd, buf, 4096) != 4096) | |
1149 | return 0; | |
1150 | ||
1151 | if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff') | |
1152 | return 0; | |
1153 | if (memcmp(buf, buf+1, 4095) != 0) | |
1154 | return 0; | |
1155 | ||
1156 | return 1; | |
1157 | } | |
1158 | ||
1159 | /* adding a spare to a regular array is quite different from adding one to | |
1160 | * a set-of-partitions virtual array. | |
1161 | * This function determines which is worth trying and tries as appropriate. | |
1162 | * Arrays are given priority over partitions. | |
1163 | */ | |
1164 | static int try_spare(char *devname, int *dfdp, struct dev_policy *pol, | |
1165 | struct map_ent *target, | |
1166 | struct supertype *st, int verbose) | |
1167 | { | |
1168 | int i; | |
1169 | int rv; | |
1170 | int arrays_ok = 0; | |
1171 | int partitions_ok = 0; | |
1172 | int dfd = *dfdp; | |
1173 | int bare; | |
1174 | ||
1175 | /* Can only add a spare if device has at least one domain */ | |
1176 | if (pol_find(pol, pol_domain) == NULL) | |
1177 | return 1; | |
1178 | /* And only if some action allows spares */ | |
1179 | if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare)) | |
1180 | return 1; | |
1181 | ||
1182 | /* Now check if the device is bare. | |
1183 | * bare devices can always be added as a spare | |
1184 | * non-bare devices can only be added if spare-same-slot is permitted, | |
1185 | * and this device is replacing a previous device - in which case 'target' | |
1186 | * will be set. | |
1187 | */ | |
1188 | if (!is_bare(dfd)) { | |
1189 | /* Must have a target and allow same_slot */ | |
1190 | /* Later - may allow force_spare without target */ | |
1191 | if (!target || | |
1192 | !policy_action_allows(pol, st?st->ss->name:NULL, | |
1193 | act_spare_same_slot)) { | |
1194 | if (verbose > 1) | |
1195 | pr_err("%s is not bare, so not " | |
1196 | "considering as a spare\n", | |
1197 | devname); | |
1198 | return 1; | |
1199 | } | |
1200 | bare = 0; | |
1201 | } else | |
1202 | bare = 1; | |
1203 | ||
1204 | /* It might be OK to add this device to an array - need to see | |
1205 | * what arrays might be candidates. | |
1206 | */ | |
1207 | if (st) { | |
1208 | /* just try try 'array' or 'partition' based on this metadata */ | |
1209 | if (st->ss->add_to_super) | |
1210 | return array_try_spare(devname, dfdp, pol, target, bare, | |
1211 | st, verbose); | |
1212 | else | |
1213 | return partition_try_spare(devname, dfdp, pol, | |
1214 | st, verbose); | |
1215 | } | |
1216 | /* No metadata was specified or found so options are open. | |
1217 | * Check for whether any array metadata, or any partition metadata | |
1218 | * might allow adding the spare. This check is just help to avoid | |
1219 | * a more costly scan of all arrays when we can be sure that will | |
1220 | * fail. | |
1221 | */ | |
1222 | for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) { | |
1223 | if (superlist[i]->add_to_super && !arrays_ok && | |
1224 | policy_action_allows(pol, superlist[i]->name, act_spare)) | |
1225 | arrays_ok = 1; | |
1226 | if (superlist[i]->add_to_super == NULL && !partitions_ok && | |
1227 | policy_action_allows(pol, superlist[i]->name, act_spare)) | |
1228 | partitions_ok = 1; | |
1229 | } | |
1230 | rv = 1; | |
1231 | if (arrays_ok) | |
1232 | rv = array_try_spare(devname, dfdp, pol, target, bare, | |
1233 | st, verbose); | |
1234 | if (rv != 0 && partitions_ok) | |
1235 | rv = partition_try_spare(devname, dfdp, pol, st, verbose); | |
1236 | return rv; | |
1237 | } | |
1238 | ||
1239 | int IncrementalScan(int verbose) | |
1240 | { | |
1241 | /* look at every device listed in the 'map' file. | |
1242 | * If one is found that is not running then: | |
1243 | * look in mdadm.conf for bitmap file. | |
1244 | * if one exists, but array has none, add it. | |
1245 | * try to start array in auto-readonly mode | |
1246 | */ | |
1247 | struct map_ent *mapl = NULL; | |
1248 | struct map_ent *me; | |
1249 | struct mddev_ident *devs, *mddev; | |
1250 | int rv = 0; | |
1251 | ||
1252 | map_read(&mapl); | |
1253 | devs = conf_get_ident(NULL); | |
1254 | ||
1255 | for (me = mapl ; me ; me = me->next) { | |
1256 | mdu_array_info_t array; | |
1257 | mdu_bitmap_file_t bmf; | |
1258 | struct mdinfo *sra; | |
1259 | int mdfd = open_dev(me->devnum); | |
1260 | ||
1261 | if (mdfd < 0) | |
1262 | continue; | |
1263 | if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 || | |
1264 | errno != ENODEV) { | |
1265 | close(mdfd); | |
1266 | continue; | |
1267 | } | |
1268 | /* Ok, we can try this one. Maybe it needs a bitmap */ | |
1269 | for (mddev = devs ; mddev ; mddev = mddev->next) | |
1270 | if (mddev->devname && me->path | |
1271 | && devname_matches(mddev->devname, me->path)) | |
1272 | break; | |
1273 | if (mddev && mddev->bitmap_file) { | |
1274 | /* | |
1275 | * Note: early kernels will wrongly fail this, so it | |
1276 | * is a hint only | |
1277 | */ | |
1278 | int added = -1; | |
1279 | if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) { | |
1280 | int bmfd = open(mddev->bitmap_file, O_RDWR); | |
1281 | if (bmfd >= 0) { | |
1282 | added = ioctl(mdfd, SET_BITMAP_FILE, | |
1283 | bmfd); | |
1284 | close(bmfd); | |
1285 | } | |
1286 | } | |
1287 | if (verbose >= 0) { | |
1288 | if (added == 0) | |
1289 | pr_err("Added bitmap %s to %s\n", | |
1290 | mddev->bitmap_file, me->path); | |
1291 | else if (errno != EEXIST) | |
1292 | pr_err("Failed to add bitmap to %s: %s\n", | |
1293 | me->path, strerror(errno)); | |
1294 | } | |
1295 | } | |
1296 | sra = sysfs_read(mdfd, 0, 0); | |
1297 | if (sra) { | |
1298 | if (sysfs_set_str(sra, NULL, | |
1299 | "array_state", "read-auto") == 0) { | |
1300 | if (verbose >= 0) | |
1301 | pr_err("started array %s\n", | |
1302 | me->path ?: devnum2devname(me->devnum)); | |
1303 | } else { | |
1304 | pr_err("failed to start array %s: %s\n", | |
1305 | me->path ?: devnum2devname(me->devnum), | |
1306 | strerror(errno)); | |
1307 | rv = 1; | |
1308 | } | |
1309 | sysfs_free(sra); | |
1310 | } | |
1311 | } | |
1312 | return rv; | |
1313 | } | |
1314 | ||
1315 | static char *container2devname(char *devname) | |
1316 | { | |
1317 | char *mdname = NULL; | |
1318 | ||
1319 | if (devname[0] == '/') { | |
1320 | int fd = open(devname, O_RDONLY); | |
1321 | if (fd >= 0) { | |
1322 | mdname = devnum2devname(fd2devnum(fd)); | |
1323 | close(fd); | |
1324 | } | |
1325 | } else { | |
1326 | int uuid[4]; | |
1327 | struct map_ent *mp, *map = NULL; | |
1328 | ||
1329 | if (!parse_uuid(devname, uuid)) | |
1330 | return mdname; | |
1331 | mp = map_by_uuid(&map, uuid); | |
1332 | if (mp) | |
1333 | mdname = devnum2devname(mp->devnum); | |
1334 | map_free(map); | |
1335 | } | |
1336 | ||
1337 | return mdname; | |
1338 | } | |
1339 | ||
1340 | static int Incremental_container(struct supertype *st, char *devname, | |
1341 | char *homehost, int verbose, | |
1342 | int runstop, int autof, int freeze_reshape) | |
1343 | { | |
1344 | /* Collect the contents of this container and for each | |
1345 | * array, choose a device name and assemble the array. | |
1346 | */ | |
1347 | ||
1348 | struct mdinfo *list; | |
1349 | struct mdinfo *ra; | |
1350 | struct map_ent *map = NULL; | |
1351 | struct mdinfo info; | |
1352 | int trustworthy; | |
1353 | struct mddev_ident *match; | |
1354 | int rv = 0; | |
1355 | struct domainlist *domains; | |
1356 | struct map_ent *smp; | |
1357 | int suuid[4]; | |
1358 | int sfd; | |
1359 | int ra_blocked = 0; | |
1360 | int ra_all = 0; | |
1361 | ||
1362 | st->ss->getinfo_super(st, &info, NULL); | |
1363 | ||
1364 | if ((runstop > 0 && info.container_enough >= 0) || | |
1365 | info.container_enough > 0) | |
1366 | /* pass */; | |
1367 | else { | |
1368 | if (verbose) | |
1369 | pr_err("not enough devices to start the container\n"); | |
1370 | return 0; | |
1371 | } | |
1372 | ||
1373 | match = conf_match(st, &info, devname, verbose, &rv); | |
1374 | if (match == NULL && rv == 2) | |
1375 | return rv; | |
1376 | ||
1377 | /* Need to compute 'trustworthy' */ | |
1378 | if (match) | |
1379 | trustworthy = LOCAL; | |
1380 | else if (st->ss->match_home(st, homehost) == 1) | |
1381 | trustworthy = LOCAL; | |
1382 | else if (st->ss->match_home(st, "any") == 1) | |
1383 | trustworthy = LOCAL; | |
1384 | else | |
1385 | trustworthy = FOREIGN; | |
1386 | ||
1387 | list = st->ss->container_content(st, NULL); | |
1388 | /* when nothing to activate - quit */ | |
1389 | if (list == NULL) | |
1390 | return 0; | |
1391 | for (ra = list ; ra ; ra = ra->next) { | |
1392 | int mdfd; | |
1393 | char chosen_name[1024]; | |
1394 | struct map_ent *mp; | |
1395 | struct mddev_ident *match = NULL; | |
1396 | ||
1397 | ra_all++; | |
1398 | /* do not activate arrays blocked by metadata handler */ | |
1399 | if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) { | |
1400 | pr_err("Cannot activate array %s in %s.\n", | |
1401 | ra->text_version, devname); | |
1402 | ra_blocked++; | |
1403 | continue; | |
1404 | } | |
1405 | mp = map_by_uuid(&map, ra->uuid); | |
1406 | ||
1407 | if (mp) { | |
1408 | mdfd = open_dev(mp->devnum); | |
1409 | if (mp->path) | |
1410 | strcpy(chosen_name, mp->path); | |
1411 | else | |
1412 | strcpy(chosen_name, devnum2devname(mp->devnum)); | |
1413 | } else { | |
1414 | ||
1415 | /* Check in mdadm.conf for container == devname and | |
1416 | * member == ra->text_version after second slash. | |
1417 | */ | |
1418 | char *sub = strchr(ra->text_version+1, '/'); | |
1419 | struct mddev_ident *array_list; | |
1420 | if (sub) { | |
1421 | sub++; | |
1422 | array_list = conf_get_ident(NULL); | |
1423 | } else | |
1424 | array_list = NULL; | |
1425 | for(; array_list ; array_list = array_list->next) { | |
1426 | char *dn; | |
1427 | if (array_list->member == NULL || | |
1428 | array_list->container == NULL) | |
1429 | continue; | |
1430 | if (strcmp(array_list->member, sub) != 0) | |
1431 | continue; | |
1432 | if (array_list->uuid_set && | |
1433 | !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid)) | |
1434 | continue; | |
1435 | dn = container2devname(array_list->container); | |
1436 | if (dn == NULL) | |
1437 | continue; | |
1438 | if (strncmp(dn, ra->text_version+1, | |
1439 | strlen(dn)) != 0 || | |
1440 | ra->text_version[strlen(dn)+1] != '/') { | |
1441 | free(dn); | |
1442 | continue; | |
1443 | } | |
1444 | free(dn); | |
1445 | /* we have a match */ | |
1446 | match = array_list; | |
1447 | if (verbose>0) | |
1448 | pr_err("match found for member %s\n", | |
1449 | array_list->member); | |
1450 | break; | |
1451 | } | |
1452 | ||
1453 | if (match && match->devname && | |
1454 | strcasecmp(match->devname, "<ignore>") == 0) { | |
1455 | if (verbose > 0) | |
1456 | pr_err("array %s/%s is " | |
1457 | "explicitly ignored by mdadm.conf\n", | |
1458 | match->container, match->member); | |
1459 | return 2; | |
1460 | } | |
1461 | if (match) | |
1462 | trustworthy = LOCAL; | |
1463 | ||
1464 | mdfd = create_mddev(match ? match->devname : NULL, | |
1465 | ra->name, | |
1466 | autof, | |
1467 | trustworthy, | |
1468 | chosen_name); | |
1469 | } | |
1470 | ||
1471 | if (mdfd < 0) { | |
1472 | pr_err("failed to open %s: %s.\n", | |
1473 | chosen_name, strerror(errno)); | |
1474 | return 2; | |
1475 | } | |
1476 | ||
1477 | assemble_container_content(st, mdfd, ra, runstop, | |
1478 | chosen_name, verbose, NULL, | |
1479 | freeze_reshape); | |
1480 | close(mdfd); | |
1481 | } | |
1482 | ||
1483 | /* don't move spares to container with volume being activated | |
1484 | when all volumes are blocked */ | |
1485 | if (ra_all == ra_blocked) | |
1486 | return 0; | |
1487 | ||
1488 | /* Now move all suitable spares from spare container */ | |
1489 | domains = domain_from_array(list, st->ss->name); | |
1490 | memcpy(suuid, uuid_zero, sizeof(int[4])); | |
1491 | if (domains && | |
1492 | (smp = map_by_uuid(&map, suuid)) != NULL && | |
1493 | (sfd = open(smp->path, O_RDONLY)) >= 0) { | |
1494 | /* spare container found */ | |
1495 | struct supertype *sst = | |
1496 | super_imsm.match_metadata_desc("imsm"); | |
1497 | struct mdinfo *sinfo; | |
1498 | unsigned long long min_size = 0; | |
1499 | if (st->ss->min_acceptable_spare_size) | |
1500 | min_size = st->ss->min_acceptable_spare_size(st); | |
1501 | if (!sst->ss->load_container(sst, sfd, NULL)) { | |
1502 | close(sfd); | |
1503 | sinfo = container_choose_spares(sst, min_size, | |
1504 | domains, NULL, | |
1505 | st->ss->name, 0); | |
1506 | sst->ss->free_super(sst); | |
1507 | if (sinfo){ | |
1508 | int count = 0; | |
1509 | struct mdinfo *disks = sinfo->devs; | |
1510 | while (disks) { | |
1511 | /* move spare from spare | |
1512 | * container to currently | |
1513 | * assembled one | |
1514 | */ | |
1515 | if (move_spare( | |
1516 | smp->path, | |
1517 | devname, | |
1518 | makedev(disks->disk.major, | |
1519 | disks->disk.minor))) | |
1520 | count++; | |
1521 | disks = disks->next; | |
1522 | } | |
1523 | if (count) | |
1524 | pr_err("Added %d spare%s to %s\n", | |
1525 | count, count>1?"s":"", devname); | |
1526 | } | |
1527 | sysfs_free(sinfo); | |
1528 | } else | |
1529 | close(sfd); | |
1530 | } | |
1531 | domain_free(domains); | |
1532 | return 0; | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | * IncrementalRemove - Attempt to see if the passed in device belongs to any | |
1537 | * raid arrays, and if so first fail (if needed) and then remove the device. | |
1538 | * | |
1539 | * @devname - The device we want to remove | |
1540 | * @id_path - name as found in /dev/disk/by-path for this device | |
1541 | * | |
1542 | * Note: the device name must be a kernel name like "sda", so | |
1543 | * that we can find it in /proc/mdstat | |
1544 | */ | |
1545 | int IncrementalRemove(char *devname, char *id_path, int verbose) | |
1546 | { | |
1547 | int mdfd; | |
1548 | int rv; | |
1549 | struct mdstat_ent *ent; | |
1550 | struct mddev_dev devlist; | |
1551 | ||
1552 | if (!id_path) | |
1553 | dprintf(Name ": incremental removal without --path <id_path> " | |
1554 | "lacks the possibility to re-add new device in this " | |
1555 | "port\n"); | |
1556 | ||
1557 | if (strchr(devname, '/')) { | |
1558 | pr_err("incremental removal requires a " | |
1559 | "kernel device name, not a file: %s\n", devname); | |
1560 | return 1; | |
1561 | } | |
1562 | ent = mdstat_by_component(devname); | |
1563 | if (!ent) { | |
1564 | pr_err("%s does not appear to be a component " | |
1565 | "of any array\n", devname); | |
1566 | return 1; | |
1567 | } | |
1568 | mdfd = open_dev(ent->devnum); | |
1569 | if (mdfd < 0) { | |
1570 | pr_err("Cannot open array %s!!\n", ent->dev); | |
1571 | free_mdstat(ent); | |
1572 | return 1; | |
1573 | } | |
1574 | ||
1575 | if (id_path) { | |
1576 | struct map_ent *map = NULL, *me; | |
1577 | me = map_by_devnum(&map, ent->devnum); | |
1578 | if (me) | |
1579 | policy_save_path(id_path, me); | |
1580 | map_free(map); | |
1581 | } | |
1582 | ||
1583 | memset(&devlist, 0, sizeof(devlist)); | |
1584 | devlist.devname = devname; | |
1585 | devlist.disposition = 'f'; | |
1586 | /* for a container, we must fail each member array */ | |
1587 | if (ent->metadata_version && | |
1588 | strncmp(ent->metadata_version, "external:", 9) == 0) { | |
1589 | struct mdstat_ent *mdstat = mdstat_read(0, 0); | |
1590 | struct mdstat_ent *memb; | |
1591 | for (memb = mdstat ; memb ; memb = memb->next) | |
1592 | if (is_container_member(memb, ent->dev)) { | |
1593 | int subfd = open_dev(memb->devnum); | |
1594 | if (subfd >= 0) { | |
1595 | Manage_subdevs(memb->dev, subfd, | |
1596 | &devlist, verbose, 0, | |
1597 | NULL, 0); | |
1598 | close(subfd); | |
1599 | } | |
1600 | } | |
1601 | free_mdstat(mdstat); | |
1602 | } else | |
1603 | Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0); | |
1604 | devlist.disposition = 'r'; | |
1605 | rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0); | |
1606 | close(mdfd); | |
1607 | free_mdstat(ent); | |
1608 | return rv; | |
1609 | } |