]> git.ipfire.org Git - thirdparty/mdadm.git/blob - mapfile.c
Bugfix: mapfile locking is broken/racy
[thirdparty/mdadm.git] / mapfile.c
1 /*
2 * mapfile - manage /var/run/mdadm.map. 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 /* /var/run/mdadm.map is used to track arrays being created in --incremental
32 * mode. It particularly allows lookup from UUID to array device, but
33 * also allows the array device name to be easily found.
34 *
35 * The map file is line based with space separated fields. The fields are:
36 * Device id - mdX or mdpX where X is a number.
37 * metadata - 0.90 1.0 1.1 1.2 ddf ...
38 * UUID - uuid of the array
39 * path - path where device created: /dev/md/home
40 *
41 * The preferred location for the map file is /var/run/mdadm.map.
42 * However /var/run may not exist or be writable in early boot. And if
43 * no-one has created /var/run/mdadm, we still want to survive.
44 * So possible locations are:
45 * /var/run/mdadm/map /var/run/mdadm.map /lib/initrw/madam/map
46 * The last can easily be change at compile to e.g. somewhere in /dev.
47 * We read from the first one that exists and write to the first
48 * one that we can.
49 */
50 #include "mdadm.h"
51 #include <sys/file.h>
52 #include <ctype.h>
53
54 #define mapnames(base) { base, base ".new", base ".lock"}
55 char *mapname[2][3] = {
56 mapnames(MAP_DIR "/" MAP_FILE),
57 mapnames("/var/run/mdadm.map")
58 };
59 char *mapdir[2] = { MAP_DIR, NULL };
60
61 int mapmode[3] = { O_RDONLY, O_RDWR|O_CREAT, O_RDWR|O_CREAT|O_TRUNC };
62 char *mapsmode[3] = { "r", "w", "w"};
63
64 FILE *open_map(int modenum, int *choice)
65 {
66 int i;
67
68 for (i = 0 ; i < 2 ; i++) {
69 int fd;
70 if ((mapmode[modenum] & O_CREAT) && mapdir[i])
71 /* Attempt to create directory, don't worry about
72 * failure.
73 */
74 mkdir(mapdir[i], 0755);
75 fd = open(mapname[i][modenum], mapmode[modenum], 0600);
76 if (fd >= 0) {
77 *choice = i;
78 return fdopen(fd, mapsmode[modenum]);
79 }
80 }
81 return NULL;
82 }
83
84 int map_write(struct map_ent *mel)
85 {
86 FILE *f;
87 int err;
88 int which;
89
90 f = open_map(1, &which);
91
92 if (!f)
93 return 0;
94 for (; mel; mel = mel->next) {
95 if (mel->bad)
96 continue;
97 if (mel->devnum < 0)
98 fprintf(f, "mdp%d ", -1-mel->devnum);
99 else
100 fprintf(f, "md%d ", mel->devnum);
101 fprintf(f, "%s ", mel->metadata);
102 fprintf(f, "%08x:%08x:%08x:%08x ", mel->uuid[0],
103 mel->uuid[1], mel->uuid[2], mel->uuid[3]);
104 fprintf(f, "%s\n", mel->path?:"");
105 }
106 fflush(f);
107 err = ferror(f);
108 fclose(f);
109 if (err) {
110 unlink(mapname[which][1]);
111 return 0;
112 }
113 return rename(mapname[which][1],
114 mapname[which][0]) == 0;
115 }
116
117
118 static FILE *lf = NULL;
119 static int lwhich = 0;
120 int map_lock(struct map_ent **melp)
121 {
122 while (lf == NULL) {
123 struct stat buf;
124 lf = open_map(2, &lwhich);
125 if (lf == NULL)
126 return -1;
127 if (flock(fileno(lf), LOCK_EX) != 0) {
128 fclose(lf);
129 lf = NULL;
130 return -1;
131 }
132 if (fstat(fileno(lf), &buf) != 0 ||
133 buf.st_nlink == 0) {
134 /* The owner of the lock unlinked it,
135 * so we have a lock on a stale file,
136 * try again
137 */
138 fclose(lf);
139 lf = NULL;
140 }
141 }
142 if (*melp)
143 map_free(*melp);
144 map_read(melp);
145 return 0;
146 }
147
148 void map_unlock(struct map_ent **melp)
149 {
150 if (lf) {
151 /* must unlink before closing the file,
152 * as only the owner of the lock may
153 * unlink the file
154 */
155 unlink(mapname[lwhich][2]);
156 fclose(lf);
157 }
158 lf = NULL;
159 }
160
161 void map_add(struct map_ent **melp,
162 int devnum, char *metadata, int uuid[4], char *path)
163 {
164 struct map_ent *me = malloc(sizeof(*me));
165
166 me->devnum = devnum;
167 strcpy(me->metadata, metadata);
168 memcpy(me->uuid, uuid, 16);
169 me->path = path ? strdup(path) : NULL;
170 me->next = *melp;
171 me->bad = 0;
172 *melp = me;
173 }
174
175 void map_read(struct map_ent **melp)
176 {
177 FILE *f;
178 char buf[8192];
179 char path[200];
180 int devnum, uuid[4];
181 char metadata[30];
182 char nam[4];
183 int which;
184
185 *melp = NULL;
186
187 f = open_map(0, &which);
188 if (!f) {
189 RebuildMap();
190 f = open_map(0, &which);
191 }
192 if (!f)
193 return;
194
195 while (fgets(buf, sizeof(buf), f)) {
196 path[0] = 0;
197 if (sscanf(buf, " %3[mdp]%d %s %x:%x:%x:%x %200s",
198 nam, &devnum, metadata, uuid, uuid+1,
199 uuid+2, uuid+3, path) >= 7) {
200 if (strncmp(nam, "md", 2) != 0)
201 continue;
202 if (nam[2] == 'p')
203 devnum = -1 - devnum;
204 map_add(melp, devnum, metadata, uuid, path);
205 }
206 }
207 fclose(f);
208 }
209
210 void map_free(struct map_ent *map)
211 {
212 while (map) {
213 struct map_ent *mp = map;
214 map = mp->next;
215 free(mp->path);
216 free(mp);
217 }
218 }
219
220 int map_update(struct map_ent **mpp, int devnum, char *metadata,
221 int *uuid, char *path)
222 {
223 struct map_ent *map, *mp;
224 int rv;
225
226 if (mpp && *mpp)
227 map = *mpp;
228 else
229 map_read(&map);
230
231 for (mp = map ; mp ; mp=mp->next)
232 if (mp->devnum == devnum) {
233 strcpy(mp->metadata, metadata);
234 memcpy(mp->uuid, uuid, 16);
235 free(mp->path);
236 mp->path = path ? strdup(path) : NULL;
237 break;
238 }
239 if (!mp)
240 map_add(&map, devnum, metadata, uuid, path);
241 if (mpp)
242 *mpp = NULL;
243 rv = map_write(map);
244 map_free(map);
245 return rv;
246 }
247
248 void map_delete(struct map_ent **mapp, int devnum)
249 {
250 struct map_ent *mp;
251
252 if (*mapp == NULL)
253 map_read(mapp);
254
255 for (mp = *mapp; mp; mp = *mapp) {
256 if (mp->devnum == devnum) {
257 *mapp = mp->next;
258 free(mp->path);
259 free(mp);
260 } else
261 mapp = & mp->next;
262 }
263 }
264
265 void map_remove(struct map_ent **mapp, int devnum)
266 {
267 if (devnum == NoMdDev)
268 return;
269
270 map_delete(mapp, devnum);
271 map_write(*mapp);
272 map_free(*mapp);
273 }
274
275 struct map_ent *map_by_uuid(struct map_ent **map, int uuid[4])
276 {
277 struct map_ent *mp;
278 if (!*map)
279 map_read(map);
280
281 for (mp = *map ; mp ; mp = mp->next) {
282 if (memcmp(uuid, mp->uuid, 16) != 0)
283 continue;
284 if (!mddev_busy(mp->devnum)) {
285 mp->bad = 1;
286 continue;
287 }
288 return mp;
289 }
290 return NULL;
291 }
292
293 struct map_ent *map_by_devnum(struct map_ent **map, int devnum)
294 {
295 struct map_ent *mp;
296 if (!*map)
297 map_read(map);
298
299 for (mp = *map ; mp ; mp = mp->next) {
300 if (mp->devnum != devnum)
301 continue;
302 if (!mddev_busy(mp->devnum)) {
303 mp->bad = 1;
304 continue;
305 }
306 return mp;
307 }
308 return NULL;
309 }
310
311 struct map_ent *map_by_name(struct map_ent **map, char *name)
312 {
313 struct map_ent *mp;
314 if (!*map)
315 map_read(map);
316
317 for (mp = *map ; mp ; mp = mp->next) {
318 if (!mp->path)
319 continue;
320 if (strncmp(mp->path, "/dev/md/", 8) != 0)
321 continue;
322 if (strcmp(mp->path+8, name) != 0)
323 continue;
324 if (!mddev_busy(mp->devnum)) {
325 mp->bad = 1;
326 continue;
327 }
328 return mp;
329 }
330 return NULL;
331 }
332
333 /* sets the proper subarray and container_dev according to the metadata
334 * version super_by_fd does this automatically, this routine is meant as
335 * a supplement for guess_super()
336 */
337 static void set_member_info(struct supertype *st, struct mdstat_ent *ent)
338 {
339
340 st->subarray[0] = '\0';
341
342 if (ent->metadata_version == NULL ||
343 strncmp(ent->metadata_version, "external:", 9) != 0)
344 return;
345
346 if (is_subarray(&ent->metadata_version[9])) {
347 char version[strlen(ent->metadata_version)+1];
348 char *subarray;
349 char *name = &version[10];
350
351 strcpy(version, ent->metadata_version);
352 subarray = strrchr(version, '/');
353 name = &version[10];
354
355 if (!subarray)
356 return;
357 *subarray++ = '\0';
358
359 st->container_dev = devname2devnum(name);
360 strncpy(st->subarray, subarray, sizeof(st->subarray));
361 }
362 }
363
364 void RebuildMap(void)
365 {
366 struct mdstat_ent *mdstat = mdstat_read(0, 0);
367 struct mdstat_ent *md;
368 struct map_ent *map = NULL;
369 int mdp = get_mdp_major();
370 int require_homehost;
371 char sys_hostname[256];
372 char *homehost = conf_get_homehost(&require_homehost);
373
374 if (homehost == NULL || strcmp(homehost, "<system>")==0) {
375 if (gethostname(sys_hostname, sizeof(sys_hostname)) == 0) {
376 sys_hostname[sizeof(sys_hostname)-1] = 0;
377 homehost = sys_hostname;
378 }
379 }
380
381 for (md = mdstat ; md ; md = md->next) {
382 struct mdinfo *sra = sysfs_read(-1, md->devnum, GET_DEVS);
383 struct mdinfo *sd;
384
385 if (!sra)
386 continue;
387
388 for (sd = sra->devs ; sd ; sd = sd->next) {
389 char namebuf[100];
390 char dn[30];
391 int dfd;
392 int ok;
393 struct supertype *st;
394 char *path;
395 struct mdinfo info;
396
397 sprintf(dn, "%d:%d", sd->disk.major, sd->disk.minor);
398 dfd = dev_open(dn, O_RDONLY);
399 if (dfd < 0)
400 continue;
401 st = guess_super(dfd);
402 if ( st == NULL)
403 ok = -1;
404 else {
405 set_member_info(st, md);
406 ok = st->ss->load_super(st, dfd, NULL);
407 }
408 close(dfd);
409 if (ok != 0)
410 continue;
411 st->ss->getinfo_super(st, &info);
412 if (md->devnum >= 0)
413 path = map_dev(MD_MAJOR, md->devnum, 0);
414 else
415 path = map_dev(mdp, (-1-md->devnum)<< 6, 0);
416 if (path == NULL ||
417 strncmp(path, "/dev/md/", 8) != 0) {
418 /* We would really like a name that provides
419 * an MD_DEVNAME for udev.
420 * The name needs to be unique both in /dev/md/
421 * and in this mapfile.
422 * It needs to match watch -I or -As would come
423 * up with.
424 * That means:
425 * Check if array is in mdadm.conf
426 * - if so use that.
427 * determine trustworthy from homehost etc
428 * find a unique name based on metadata name.
429 *
430 */
431 struct mddev_ident_s *match = conf_match(&info, st);
432 struct stat stb;
433 if (match && match->devname && match->devname[0] == '/') {
434 path = match->devname;
435 if (path[0] != '/') {
436 strcpy(namebuf, "/dev/md/");
437 strcat(namebuf, path);
438 path = namebuf;
439 }
440 } else {
441 int unum = 0;
442 char *sep = "_";
443 const char *name;
444 int conflict = 1;
445 if ((homehost == NULL ||
446 st->ss->match_home(st, homehost) != 1) &&
447 st->ss->match_home(st, "any") != 1 &&
448 (require_homehost
449 || ! conf_name_is_free(info.name)))
450 /* require a numeric suffix */
451 unum = 0;
452 else
453 /* allow name to be used as-is if no conflict */
454 unum = -1;
455 name = info.name;
456 if (!*name) {
457 name = st->ss->name;
458 if (!isdigit(name[strlen(name)-1]) &&
459 unum == -1) {
460 unum = 0;
461 sep = "";
462 }
463 }
464 if (strchr(name, ':'))
465 /* probably a uniquifying
466 * hostname prefix. Allow
467 * without a suffix
468 */
469 unum = -1;
470
471 while (conflict) {
472 if (unum >= 0)
473 sprintf(namebuf, "/dev/md/%s%s%d",
474 name, sep, unum);
475 else
476 sprintf(namebuf, "/dev/md/%s",
477 name);
478 unum++;
479 if (lstat(namebuf, &stb) != 0 &&
480 (map == NULL ||
481 !map_by_name(&map, namebuf+8)))
482 conflict = 0;
483 }
484 path = namebuf;
485 }
486 }
487 map_add(&map, md->devnum,
488 info.text_version,
489 info.uuid, path);
490 st->ss->free_super(st);
491 break;
492 }
493 sysfs_free(sra);
494 }
495 /* Only trigger a change if we wrote a new map file */
496 if (map_write(map))
497 for (md = mdstat ; md ; md = md->next) {
498 struct mdinfo *sra = sysfs_read(-1, md->devnum,
499 GET_VERSION);
500 if (sra)
501 sysfs_uevent(sra, "change");
502 sysfs_free(sra);
503 }
504 map_free(map);
505 free_mdstat(mdstat);
506 }