]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de> | |
5 | * | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neilb@cse.unsw.edu.au> | |
23 | * Paper: Neil Brown | |
24 | * School of Computer Science and Engineering | |
25 | * The University of New South Wales | |
26 | * Sydney, 2052 | |
27 | * Australia | |
28 | */ | |
29 | ||
30 | #include "mdadm.h" | |
31 | #include "md_p.h" | |
32 | #include <sys/socket.h> | |
33 | #include <sys/utsname.h> | |
34 | #include <sys/wait.h> | |
35 | #include <sys/un.h> | |
36 | #include <ctype.h> | |
37 | #include <dirent.h> | |
38 | #include <signal.h> | |
39 | ||
40 | /* | |
41 | * following taken from linux/blkpg.h because they aren't | |
42 | * anywhere else and it isn't safe to #include linux/ * stuff. | |
43 | */ | |
44 | ||
45 | #define BLKPG _IO(0x12,105) | |
46 | ||
47 | /* The argument structure */ | |
48 | struct blkpg_ioctl_arg { | |
49 | int op; | |
50 | int flags; | |
51 | int datalen; | |
52 | void *data; | |
53 | }; | |
54 | ||
55 | /* The subfunctions (for the op field) */ | |
56 | #define BLKPG_ADD_PARTITION 1 | |
57 | #define BLKPG_DEL_PARTITION 2 | |
58 | ||
59 | /* Sizes of name fields. Unused at present. */ | |
60 | #define BLKPG_DEVNAMELTH 64 | |
61 | #define BLKPG_VOLNAMELTH 64 | |
62 | ||
63 | /* The data structure for ADD_PARTITION and DEL_PARTITION */ | |
64 | struct blkpg_partition { | |
65 | long long start; /* starting offset in bytes */ | |
66 | long long length; /* length in bytes */ | |
67 | int pno; /* partition number */ | |
68 | char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2, | |
69 | to be used in kernel messages */ | |
70 | char volname[BLKPG_VOLNAMELTH]; /* volume label */ | |
71 | }; | |
72 | ||
73 | /* | |
74 | * Parse a 128 bit uuid in 4 integers | |
75 | * format is 32 hexx nibbles with options :.<space> separator | |
76 | * If not exactly 32 hex digits are found, return 0 | |
77 | * else return 1 | |
78 | */ | |
79 | int parse_uuid(char *str, int uuid[4]) | |
80 | { | |
81 | int hit = 0; /* number of Hex digIT */ | |
82 | int i; | |
83 | char c; | |
84 | for (i=0; i<4; i++) uuid[i]=0; | |
85 | ||
86 | while ((c= *str++)) { | |
87 | int n; | |
88 | if (c>='0' && c<='9') | |
89 | n = c-'0'; | |
90 | else if (c>='a' && c <= 'f') | |
91 | n = 10 + c - 'a'; | |
92 | else if (c>='A' && c <= 'F') | |
93 | n = 10 + c - 'A'; | |
94 | else if (strchr(":. -", c)) | |
95 | continue; | |
96 | else return 0; | |
97 | ||
98 | if (hit<32) { | |
99 | uuid[hit/8] <<= 4; | |
100 | uuid[hit/8] += n; | |
101 | } | |
102 | hit++; | |
103 | } | |
104 | if (hit == 32) | |
105 | return 1; | |
106 | return 0; | |
107 | } | |
108 | ||
109 | ||
110 | /* | |
111 | * Get the md version number. | |
112 | * We use the RAID_VERSION ioctl if it is supported | |
113 | * If not, but we have a block device with major '9', we assume | |
114 | * 0.36.0 | |
115 | * | |
116 | * Return version number as 24 but number - assume version parts | |
117 | * always < 255 | |
118 | */ | |
119 | ||
120 | int md_get_version(int fd) | |
121 | { | |
122 | struct stat stb; | |
123 | mdu_version_t vers; | |
124 | ||
125 | if (fstat(fd, &stb)<0) | |
126 | return -1; | |
127 | if ((S_IFMT&stb.st_mode) != S_IFBLK) | |
128 | return -1; | |
129 | ||
130 | if (ioctl(fd, RAID_VERSION, &vers) == 0) | |
131 | return (vers.major*10000) + (vers.minor*100) + vers.patchlevel; | |
132 | if (errno == EACCES) | |
133 | return -1; | |
134 | if (major(stb.st_rdev) == MD_MAJOR) | |
135 | return (3600); | |
136 | return -1; | |
137 | } | |
138 | ||
139 | int get_linux_version() | |
140 | { | |
141 | struct utsname name; | |
142 | char *cp; | |
143 | int a,b,c; | |
144 | if (uname(&name) <0) | |
145 | return -1; | |
146 | ||
147 | cp = name.release; | |
148 | a = strtoul(cp, &cp, 10); | |
149 | if (*cp != '.') return -1; | |
150 | b = strtoul(cp+1, &cp, 10); | |
151 | if (*cp != '.') return -1; | |
152 | c = strtoul(cp+1, NULL, 10); | |
153 | ||
154 | return (a*1000000)+(b*1000)+c; | |
155 | } | |
156 | ||
157 | void remove_partitions(int fd) | |
158 | { | |
159 | /* remove partitions from this block devices. | |
160 | * This is used for components added to an array | |
161 | */ | |
162 | #ifdef BLKPG_DEL_PARTITION | |
163 | struct blkpg_ioctl_arg a; | |
164 | struct blkpg_partition p; | |
165 | ||
166 | a.op = BLKPG_DEL_PARTITION; | |
167 | a.data = (void*)&p; | |
168 | a.datalen = sizeof(p); | |
169 | a.flags = 0; | |
170 | memset(a.data, 0, a.datalen); | |
171 | for (p.pno=0; p.pno < 16; p.pno++) | |
172 | ioctl(fd, BLKPG, &a); | |
173 | #endif | |
174 | } | |
175 | ||
176 | int enough(int level, int raid_disks, int layout, int clean, | |
177 | char *avail, int avail_disks) | |
178 | { | |
179 | int copies, first; | |
180 | switch (level) { | |
181 | case 10: | |
182 | /* This is the tricky one - we need to check | |
183 | * which actual disks are present. | |
184 | */ | |
185 | copies = (layout&255)* ((layout>>8) & 255); | |
186 | first=0; | |
187 | do { | |
188 | /* there must be one of the 'copies' form 'first' */ | |
189 | int n = copies; | |
190 | int cnt=0; | |
191 | while (n--) { | |
192 | if (avail[first]) | |
193 | cnt++; | |
194 | first = (first+1) % raid_disks; | |
195 | } | |
196 | if (cnt == 0) | |
197 | return 0; | |
198 | ||
199 | } while (first != 0); | |
200 | return 1; | |
201 | ||
202 | case -4: | |
203 | return avail_disks>= 1; | |
204 | case -1: | |
205 | case 0: | |
206 | return avail_disks == raid_disks; | |
207 | case 1: | |
208 | return avail_disks >= 1; | |
209 | case 4: | |
210 | case 5: | |
211 | if (clean) | |
212 | return avail_disks >= raid_disks-1; | |
213 | else | |
214 | return avail_disks >= raid_disks; | |
215 | case 6: | |
216 | if (clean) | |
217 | return avail_disks >= raid_disks-2; | |
218 | else | |
219 | return avail_disks >= raid_disks; | |
220 | default: | |
221 | return 0; | |
222 | } | |
223 | } | |
224 | ||
225 | int same_uuid(int a[4], int b[4], int swapuuid) | |
226 | { | |
227 | if (swapuuid) { | |
228 | /* parse uuids are hostendian. | |
229 | * uuid's from some superblocks are big-ending | |
230 | * if there is a difference, we need to swap.. | |
231 | */ | |
232 | unsigned char *ac = (unsigned char *)a; | |
233 | unsigned char *bc = (unsigned char *)b; | |
234 | int i; | |
235 | for (i=0; i<16; i+= 4) { | |
236 | if (ac[i+0] != bc[i+3] || | |
237 | ac[i+1] != bc[i+2] || | |
238 | ac[i+2] != bc[i+1] || | |
239 | ac[i+3] != bc[i+0]) | |
240 | return 0; | |
241 | } | |
242 | return 1; | |
243 | } else { | |
244 | if (a[0]==b[0] && | |
245 | a[1]==b[1] && | |
246 | a[2]==b[2] && | |
247 | a[3]==b[3]) | |
248 | return 1; | |
249 | return 0; | |
250 | } | |
251 | } | |
252 | void copy_uuid(void *a, int b[4], int swapuuid) | |
253 | { | |
254 | if (swapuuid) { | |
255 | /* parse uuids are hostendian. | |
256 | * uuid's from some superblocks are big-ending | |
257 | * if there is a difference, we need to swap.. | |
258 | */ | |
259 | unsigned char *ac = (unsigned char *)a; | |
260 | unsigned char *bc = (unsigned char *)b; | |
261 | int i; | |
262 | for (i=0; i<16; i+= 4) { | |
263 | ac[i+0] = bc[i+3]; | |
264 | ac[i+1] = bc[i+2]; | |
265 | ac[i+2] = bc[i+1]; | |
266 | ac[i+3] = bc[i+0]; | |
267 | } | |
268 | } else | |
269 | memcpy(a, b, 16); | |
270 | } | |
271 | ||
272 | #ifndef MDASSEMBLE | |
273 | int check_ext2(int fd, char *name) | |
274 | { | |
275 | /* | |
276 | * Check for an ext2fs file system. | |
277 | * Superblock is always 1K at 1K offset | |
278 | * | |
279 | * s_magic is le16 at 56 == 0xEF53 | |
280 | * report mtime - le32 at 44 | |
281 | * blocks - le32 at 4 | |
282 | * logblksize - le32 at 24 | |
283 | */ | |
284 | unsigned char sb[1024]; | |
285 | time_t mtime; | |
286 | int size, bsize; | |
287 | if (lseek(fd, 1024,0)!= 1024) | |
288 | return 0; | |
289 | if (read(fd, sb, 1024)!= 1024) | |
290 | return 0; | |
291 | if (sb[56] != 0x53 || sb[57] != 0xef) | |
292 | return 0; | |
293 | ||
294 | mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8; | |
295 | bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8; | |
296 | size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8; | |
297 | fprintf(stderr, Name ": %s appears to contain an ext2fs file system\n", | |
298 | name); | |
299 | fprintf(stderr," size=%dK mtime=%s", | |
300 | size*(1<<bsize), ctime(&mtime)); | |
301 | return 1; | |
302 | } | |
303 | ||
304 | int check_reiser(int fd, char *name) | |
305 | { | |
306 | /* | |
307 | * superblock is at 64K | |
308 | * size is 1024; | |
309 | * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52 | |
310 | * | |
311 | */ | |
312 | unsigned char sb[1024]; | |
313 | unsigned long size; | |
314 | if (lseek(fd, 64*1024, 0) != 64*1024) | |
315 | return 0; | |
316 | if (read(fd, sb, 1024) != 1024) | |
317 | return 0; | |
318 | if (strncmp((char*)sb+52, "ReIsErFs",8)!=0 && | |
319 | strncmp((char*)sb+52, "ReIsEr2Fs",9)!=0) | |
320 | return 0; | |
321 | fprintf(stderr, Name ": %s appears to contain a reiserfs file system\n",name); | |
322 | size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8; | |
323 | fprintf(stderr, " size = %luK\n", size*4); | |
324 | ||
325 | return 1; | |
326 | } | |
327 | ||
328 | int check_raid(int fd, char *name) | |
329 | { | |
330 | struct mdinfo info; | |
331 | time_t crtime; | |
332 | char *level; | |
333 | struct supertype *st = guess_super(fd); | |
334 | ||
335 | if (!st) return 0; | |
336 | st->ss->load_super(st, fd, name); | |
337 | /* Looks like a raid array .. */ | |
338 | fprintf(stderr, Name ": %s appears to be part of a raid array:\n", | |
339 | name); | |
340 | st->ss->getinfo_super(st, &info); | |
341 | st->ss->free_super(st); | |
342 | crtime = info.array.ctime; | |
343 | level = map_num(pers, info.array.level); | |
344 | if (!level) level = "-unknown-"; | |
345 | fprintf(stderr, " level=%s devices=%d ctime=%s", | |
346 | level, info.array.raid_disks, ctime(&crtime)); | |
347 | return 1; | |
348 | } | |
349 | ||
350 | int ask(char *mesg) | |
351 | { | |
352 | char *add = ""; | |
353 | int i; | |
354 | for (i=0; i<5; i++) { | |
355 | char buf[100]; | |
356 | fprintf(stderr, "%s%s", mesg, add); | |
357 | fflush(stderr); | |
358 | if (fgets(buf, 100, stdin)==NULL) | |
359 | return 0; | |
360 | if (buf[0]=='y' || buf[0]=='Y') | |
361 | return 1; | |
362 | if (buf[0]=='n' || buf[0]=='N') | |
363 | return 0; | |
364 | add = "(y/n) "; | |
365 | } | |
366 | fprintf(stderr, Name ": assuming 'no'\n"); | |
367 | return 0; | |
368 | } | |
369 | #endif /* MDASSEMBLE */ | |
370 | ||
371 | char *map_num(mapping_t *map, int num) | |
372 | { | |
373 | while (map->name) { | |
374 | if (map->num == num) | |
375 | return map->name; | |
376 | map++; | |
377 | } | |
378 | return NULL; | |
379 | } | |
380 | ||
381 | int map_name(mapping_t *map, char *name) | |
382 | { | |
383 | while (map->name) { | |
384 | if (strcmp(map->name, name)==0) | |
385 | return map->num; | |
386 | map++; | |
387 | } | |
388 | return UnSet; | |
389 | } | |
390 | ||
391 | ||
392 | int is_standard(char *dev, int *nump) | |
393 | { | |
394 | /* tests if dev is a "standard" md dev name. | |
395 | * i.e if the last component is "/dNN" or "/mdNN", | |
396 | * where NN is a string of digits | |
397 | * Returns 1 if a partitionable standard, | |
398 | * -1 if non-partitonable, | |
399 | * 0 if not a standard name. | |
400 | */ | |
401 | char *d = strrchr(dev, '/'); | |
402 | int type=0; | |
403 | int num; | |
404 | if (!d) | |
405 | return 0; | |
406 | if (strncmp(d, "/d",2)==0) | |
407 | d += 2, type=1; /* /dev/md/dN{pM} */ | |
408 | else if (strncmp(d, "/md_d", 5)==0) | |
409 | d += 5, type=1; /* /dev/md_dNpM */ | |
410 | else if (strncmp(d, "/md", 3)==0) | |
411 | d += 3, type=-1; /* /dev/mdN */ | |
412 | else if (d-dev > 3 && strncmp(d-2, "md/", 3)==0) | |
413 | d += 1, type=-1; /* /dev/md/N */ | |
414 | else | |
415 | return 0; | |
416 | if (!*d) | |
417 | return 0; | |
418 | num = atoi(d); | |
419 | while (isdigit(*d)) | |
420 | d++; | |
421 | if (*d) | |
422 | return 0; | |
423 | if (nump) *nump = num; | |
424 | ||
425 | return type; | |
426 | } | |
427 | ||
428 | ||
429 | /* | |
430 | * convert a major/minor pair for a block device into a name in /dev, if possible. | |
431 | * On the first call, walk /dev collecting name. | |
432 | * Put them in a simple linked listfor now. | |
433 | */ | |
434 | struct devmap { | |
435 | int major, minor; | |
436 | char *name; | |
437 | struct devmap *next; | |
438 | } *devlist = NULL; | |
439 | int devlist_ready = 0; | |
440 | ||
441 | int add_dev(const char *name, const struct stat *stb, int flag, struct FTW *s) | |
442 | { | |
443 | struct stat st; | |
444 | if (S_ISLNK(stb->st_mode)) { | |
445 | stat(name, &st); | |
446 | stb = &st; | |
447 | } | |
448 | ||
449 | if ((stb->st_mode&S_IFMT)== S_IFBLK) { | |
450 | char *n = strdup(name); | |
451 | struct devmap *dm = malloc(sizeof(*dm)); | |
452 | if (strncmp(n, "/dev/./", 7)==0) | |
453 | strcpy(n+4, name+6); | |
454 | if (dm) { | |
455 | dm->major = major(stb->st_rdev); | |
456 | dm->minor = minor(stb->st_rdev); | |
457 | dm->name = n; | |
458 | dm->next = devlist; | |
459 | devlist = dm; | |
460 | } | |
461 | } | |
462 | return 0; | |
463 | } | |
464 | ||
465 | #ifndef HAVE_NFTW | |
466 | #ifdef HAVE_FTW | |
467 | int add_dev_1(const char *name, const struct stat *stb, int flag) | |
468 | { | |
469 | return add_dev(name, stb, flag, NULL); | |
470 | } | |
471 | int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags) | |
472 | { | |
473 | return ftw(path, add_dev_1, nopenfd); | |
474 | } | |
475 | #else | |
476 | int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags) | |
477 | { | |
478 | return 0; | |
479 | } | |
480 | #endif /* HAVE_FTW */ | |
481 | #endif /* HAVE_NFTW */ | |
482 | ||
483 | /* | |
484 | * Find a block device with the right major/minor number. | |
485 | * If we find multiple names, choose the shortest. | |
486 | * If we find a non-standard name, it is probably there | |
487 | * deliberately so prefer it over a standard name. | |
488 | * This applies only to names for MD devices. | |
489 | */ | |
490 | char *map_dev(int major, int minor, int create) | |
491 | { | |
492 | struct devmap *p; | |
493 | char *std = NULL, *nonstd=NULL; | |
494 | int did_check = 0; | |
495 | ||
496 | if (major == 0 && minor == 0) | |
497 | return NULL; | |
498 | ||
499 | retry: | |
500 | if (!devlist_ready) { | |
501 | char *dev = "/dev"; | |
502 | struct stat stb; | |
503 | while(devlist) { | |
504 | struct devmap *d = devlist; | |
505 | devlist = d->next; | |
506 | free(d->name); | |
507 | free(d); | |
508 | } | |
509 | if (lstat(dev, &stb)==0 && | |
510 | S_ISLNK(stb.st_mode)) | |
511 | dev = "/dev/."; | |
512 | nftw(dev, add_dev, 10, FTW_PHYS); | |
513 | devlist_ready=1; | |
514 | did_check = 1; | |
515 | } | |
516 | ||
517 | for (p=devlist; p; p=p->next) | |
518 | if (p->major == major && | |
519 | p->minor == minor) { | |
520 | if (is_standard(p->name, NULL)) { | |
521 | if (std == NULL || | |
522 | strlen(p->name) < strlen(std)) | |
523 | std = p->name; | |
524 | } else { | |
525 | if (nonstd == NULL || | |
526 | strlen(p->name) < strlen(nonstd)) | |
527 | nonstd = p->name; | |
528 | } | |
529 | } | |
530 | if (!std && !nonstd && !did_check) { | |
531 | devlist_ready = 0; | |
532 | goto retry; | |
533 | } | |
534 | if (create && !std && !nonstd) { | |
535 | static char buf[30]; | |
536 | snprintf(buf, sizeof(buf), "%d:%d", major, minor); | |
537 | nonstd = buf; | |
538 | } | |
539 | ||
540 | return nonstd ? nonstd : std; | |
541 | } | |
542 | ||
543 | unsigned long calc_csum(void *super, int bytes) | |
544 | { | |
545 | unsigned long long newcsum = 0; | |
546 | int i; | |
547 | unsigned int csum; | |
548 | unsigned int *superc = (unsigned int*) super; | |
549 | ||
550 | for(i=0; i<bytes/4; i++) | |
551 | newcsum+= superc[i]; | |
552 | csum = (newcsum& 0xffffffff) + (newcsum>>32); | |
553 | #ifdef __alpha__ | |
554 | /* The in-kernel checksum calculation is always 16bit on | |
555 | * the alpha, though it is 32 bit on i386... | |
556 | * I wonder what it is elsewhere... (it uses and API in | |
557 | * a way that it shouldn't). | |
558 | */ | |
559 | csum = (csum & 0xffff) + (csum >> 16); | |
560 | csum = (csum & 0xffff) + (csum >> 16); | |
561 | #endif | |
562 | return csum; | |
563 | } | |
564 | ||
565 | #ifndef MDASSEMBLE | |
566 | char *human_size(long long bytes) | |
567 | { | |
568 | static char buf[30]; | |
569 | ||
570 | /* We convert bytes to either centi-M{ega,ibi}bytes or | |
571 | * centi-G{igi,ibi}bytes, with appropriate rounding, | |
572 | * and then print 1/100th of those as a decimal. | |
573 | * We allow upto 2048Megabytes before converting to | |
574 | * gigabytes, as that shows more precision and isn't | |
575 | * too large a number. | |
576 | * Terrabytes are not yet handled. | |
577 | */ | |
578 | ||
579 | if (bytes < 5000*1024) | |
580 | buf[0]=0; | |
581 | else if (bytes < 2*1024LL*1024LL*1024LL) { | |
582 | long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2; | |
583 | long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2; | |
584 | snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)", | |
585 | cMiB/100 , cMiB % 100, | |
586 | cMB/100, cMB % 100); | |
587 | } else { | |
588 | long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2; | |
589 | long cGB = (bytes / (1000000000LL/200LL ) +1) /2; | |
590 | snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)", | |
591 | cGiB/100 , cGiB % 100, | |
592 | cGB/100, cGB % 100); | |
593 | } | |
594 | return buf; | |
595 | } | |
596 | ||
597 | char *human_size_brief(long long bytes) | |
598 | { | |
599 | static char buf[30]; | |
600 | ||
601 | if (bytes < 5000*1024) | |
602 | snprintf(buf, sizeof(buf), "%ld.%02ldKiB", | |
603 | (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024) | |
604 | ); | |
605 | else if (bytes < 2*1024LL*1024LL*1024LL) | |
606 | snprintf(buf, sizeof(buf), "%ld.%02ldMiB", | |
607 | (long)(bytes>>20), | |
608 | (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100) | |
609 | ); | |
610 | else | |
611 | snprintf(buf, sizeof(buf), "%ld.%02ldGiB", | |
612 | (long)(bytes>>30), | |
613 | (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100) | |
614 | ); | |
615 | return buf; | |
616 | } | |
617 | #endif | |
618 | ||
619 | unsigned long long calc_array_size(int level, int raid_disks, int layout, | |
620 | int chunksize, unsigned long long devsize) | |
621 | { | |
622 | int data_disks = 0; | |
623 | switch (level) { | |
624 | case 0: data_disks = raid_disks; break; | |
625 | case 1: data_disks = 1; break; | |
626 | case 4: | |
627 | case 5: data_disks = raid_disks - 1; break; | |
628 | case 6: data_disks = raid_disks - 2; break; | |
629 | case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255); | |
630 | break; | |
631 | } | |
632 | devsize &= ~(unsigned long long)((chunksize>>9)-1); | |
633 | return data_disks * devsize; | |
634 | } | |
635 | ||
636 | int get_mdp_major(void) | |
637 | { | |
638 | static int mdp_major = -1; | |
639 | FILE *fl; | |
640 | char *w; | |
641 | int have_block = 0; | |
642 | int have_devices = 0; | |
643 | int last_num = -1; | |
644 | ||
645 | if (mdp_major != -1) | |
646 | return mdp_major; | |
647 | fl = fopen("/proc/devices", "r"); | |
648 | if (!fl) | |
649 | return -1; | |
650 | while ((w = conf_word(fl, 1))) { | |
651 | if (have_block && strcmp(w, "devices:")==0) | |
652 | have_devices = 1; | |
653 | have_block = (strcmp(w, "Block")==0); | |
654 | if (isdigit(w[0])) | |
655 | last_num = atoi(w); | |
656 | if (have_devices && strcmp(w, "mdp")==0) | |
657 | mdp_major = last_num; | |
658 | free(w); | |
659 | } | |
660 | fclose(fl); | |
661 | return mdp_major; | |
662 | } | |
663 | ||
664 | #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) | |
665 | char *get_md_name(int dev) | |
666 | { | |
667 | /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */ | |
668 | /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */ | |
669 | static char devname[50]; | |
670 | struct stat stb; | |
671 | dev_t rdev; | |
672 | char *dn; | |
673 | ||
674 | if (dev < 0) { | |
675 | int mdp = get_mdp_major(); | |
676 | if (mdp < 0) return NULL; | |
677 | rdev = makedev(mdp, (-1-dev)<<6); | |
678 | snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev); | |
679 | if (stat(devname, &stb) == 0 | |
680 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
681 | && (stb.st_rdev == rdev)) | |
682 | return devname; | |
683 | } else { | |
684 | rdev = makedev(MD_MAJOR, dev); | |
685 | snprintf(devname, sizeof(devname), "/dev/md%d", dev); | |
686 | if (stat(devname, &stb) == 0 | |
687 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
688 | && (stb.st_rdev == rdev)) | |
689 | return devname; | |
690 | ||
691 | snprintf(devname, sizeof(devname), "/dev/md/%d", dev); | |
692 | if (stat(devname, &stb) == 0 | |
693 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
694 | && (stb.st_rdev == rdev)) | |
695 | return devname; | |
696 | } | |
697 | dn = map_dev(major(rdev), minor(rdev), 0); | |
698 | if (dn) | |
699 | return dn; | |
700 | snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev); | |
701 | if (mknod(devname, S_IFBLK | 0600, rdev) == -1) | |
702 | if (errno != EEXIST) | |
703 | return NULL; | |
704 | ||
705 | if (stat(devname, &stb) == 0 | |
706 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
707 | && (stb.st_rdev == rdev)) | |
708 | return devname; | |
709 | unlink(devname); | |
710 | return NULL; | |
711 | } | |
712 | ||
713 | void put_md_name(char *name) | |
714 | { | |
715 | if (strncmp(name, "/dev/.tmp.md", 12)==0) | |
716 | unlink(name); | |
717 | } | |
718 | ||
719 | int find_free_devnum(int use_partitions) | |
720 | { | |
721 | int devnum; | |
722 | for (devnum = 127; devnum != 128; | |
723 | devnum = devnum ? devnum-1 : (1<<22)-1) { | |
724 | char *dn; | |
725 | int _devnum; | |
726 | ||
727 | _devnum = use_partitions ? (-1-devnum) : devnum; | |
728 | if (mddev_busy(_devnum)) | |
729 | continue; | |
730 | /* make sure it is new to /dev too, at least as a | |
731 | * non-standard */ | |
732 | dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0); | |
733 | if (dn && ! is_standard(dn, NULL)) | |
734 | continue; | |
735 | break; | |
736 | } | |
737 | if (devnum == 128) | |
738 | return NoMdDev; | |
739 | return use_partitions ? (-1-devnum) : devnum; | |
740 | } | |
741 | #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */ | |
742 | ||
743 | int dev_open(char *dev, int flags) | |
744 | { | |
745 | /* like 'open', but if 'dev' matches %d:%d, create a temp | |
746 | * block device and open that | |
747 | */ | |
748 | char *e; | |
749 | int fd = -1; | |
750 | char devname[32]; | |
751 | int major; | |
752 | int minor; | |
753 | ||
754 | if (!dev) return -1; | |
755 | ||
756 | major = strtoul(dev, &e, 0); | |
757 | if (e > dev && *e == ':' && e[1] && | |
758 | (minor = strtoul(e+1, &e, 0)) >= 0 && | |
759 | *e == 0) { | |
760 | snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d", | |
761 | (int)getpid(), major, minor); | |
762 | if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) { | |
763 | fd = open(devname, flags|O_DIRECT); | |
764 | unlink(devname); | |
765 | } | |
766 | } else | |
767 | fd = open(dev, flags|O_DIRECT); | |
768 | return fd; | |
769 | } | |
770 | ||
771 | int open_dev_excl(int devnum) | |
772 | { | |
773 | char buf[20]; | |
774 | int i; | |
775 | ||
776 | sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum)); | |
777 | for (i=0 ; i<25 ; i++) { | |
778 | int fd = dev_open(buf, O_RDWR|O_EXCL); | |
779 | if (fd >= 0) | |
780 | return fd; | |
781 | if (errno != EBUSY) | |
782 | return fd; | |
783 | usleep(200000); | |
784 | } | |
785 | return -1; | |
786 | } | |
787 | ||
788 | struct superswitch *superlist[] = { &super0, &super1, &super_ddf, &super_imsm, NULL }; | |
789 | ||
790 | #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) | |
791 | ||
792 | struct supertype *super_by_fd(int fd) | |
793 | { | |
794 | mdu_array_info_t array; | |
795 | int vers; | |
796 | int minor; | |
797 | struct supertype *st = NULL; | |
798 | struct mdinfo *sra; | |
799 | char *verstr; | |
800 | char version[20]; | |
801 | int i; | |
802 | char *subarray = NULL; | |
803 | ||
804 | sra = sysfs_read(fd, 0, GET_VERSION); | |
805 | ||
806 | if (sra) { | |
807 | vers = sra->array.major_version; | |
808 | minor = sra->array.minor_version; | |
809 | verstr = sra->text_version; | |
810 | } else { | |
811 | if (ioctl(fd, GET_ARRAY_INFO, &array)) | |
812 | array.major_version = array.minor_version = 0; | |
813 | vers = array.major_version; | |
814 | minor = array.minor_version; | |
815 | verstr = ""; | |
816 | } | |
817 | ||
818 | if (vers != -1) { | |
819 | sprintf(version, "%d.%d", vers, minor); | |
820 | verstr = version; | |
821 | } | |
822 | if (minor == -2 && is_subarray(verstr)) { | |
823 | char *dev = verstr+1; | |
824 | subarray = strchr(dev, '/'); | |
825 | int devnum; | |
826 | if (subarray) | |
827 | *subarray++ = '\0'; | |
828 | devnum = devname2devnum(dev); | |
829 | subarray = strdup(subarray); | |
830 | if (sra) | |
831 | sysfs_free(sra); | |
832 | sra = sysfs_read(-1, devnum, GET_VERSION); | |
833 | verstr = sra->text_version ? : "-no-metadata-"; | |
834 | } | |
835 | ||
836 | for (i = 0; st == NULL && superlist[i] ; i++) | |
837 | st = superlist[i]->match_metadata_desc(verstr); | |
838 | ||
839 | if (sra) | |
840 | sysfs_free(sra); | |
841 | if (st) { | |
842 | st->sb = NULL; | |
843 | if (subarray) { | |
844 | strncpy(st->subarray, subarray, 32); | |
845 | st->subarray[31] = 0; | |
846 | free(subarray); | |
847 | } else | |
848 | st->subarray[0] = 0; | |
849 | } | |
850 | return st; | |
851 | } | |
852 | #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */ | |
853 | ||
854 | ||
855 | struct supertype *dup_super(struct supertype *orig) | |
856 | { | |
857 | struct supertype *st; | |
858 | ||
859 | if (!orig) | |
860 | return orig; | |
861 | st = malloc(sizeof(*st)); | |
862 | if (!st) | |
863 | return st; | |
864 | memset(st, 0, sizeof(*st)); | |
865 | st->ss = orig->ss; | |
866 | st->max_devs = orig->max_devs; | |
867 | st->minor_version = orig->minor_version; | |
868 | strcpy(st->subarray, orig->subarray); | |
869 | st->sb = NULL; | |
870 | st->info = NULL; | |
871 | return st; | |
872 | } | |
873 | ||
874 | struct supertype *guess_super(int fd) | |
875 | { | |
876 | /* try each load_super to find the best match, | |
877 | * and return the best superswitch | |
878 | */ | |
879 | struct superswitch *ss; | |
880 | struct supertype *st; | |
881 | unsigned long besttime = 0; | |
882 | int bestsuper = -1; | |
883 | int i; | |
884 | ||
885 | st = malloc(sizeof(*st)); | |
886 | for (i=0 ; superlist[i]; i++) { | |
887 | int rv; | |
888 | ss = superlist[i]; | |
889 | memset(st, 0, sizeof(*st)); | |
890 | rv = ss->load_super(st, fd, NULL); | |
891 | if (rv == 0) { | |
892 | struct mdinfo info; | |
893 | st->ss->getinfo_super(st, &info); | |
894 | if (bestsuper == -1 || | |
895 | besttime < info.array.ctime) { | |
896 | bestsuper = i; | |
897 | besttime = info.array.ctime; | |
898 | } | |
899 | ss->free_super(st); | |
900 | } | |
901 | } | |
902 | if (bestsuper != -1) { | |
903 | int rv; | |
904 | memset(st, 0, sizeof(*st)); | |
905 | rv = superlist[bestsuper]->load_super(st, fd, NULL); | |
906 | if (rv == 0) { | |
907 | superlist[bestsuper]->free_super(st); | |
908 | return st; | |
909 | } | |
910 | } | |
911 | free(st); | |
912 | return NULL; | |
913 | } | |
914 | ||
915 | /* Return size of device in bytes */ | |
916 | int get_dev_size(int fd, char *dname, unsigned long long *sizep) | |
917 | { | |
918 | unsigned long long ldsize; | |
919 | struct stat st; | |
920 | ||
921 | if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode)) | |
922 | ldsize = (unsigned long long)st.st_size; | |
923 | else | |
924 | #ifdef BLKGETSIZE64 | |
925 | if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0) | |
926 | #endif | |
927 | { | |
928 | unsigned long dsize; | |
929 | if (ioctl(fd, BLKGETSIZE, &dsize) == 0) { | |
930 | ldsize = dsize; | |
931 | ldsize <<= 9; | |
932 | } else { | |
933 | if (dname) | |
934 | fprintf(stderr, Name ": Cannot get size of %s: %s\b", | |
935 | dname, strerror(errno)); | |
936 | return 0; | |
937 | } | |
938 | } | |
939 | *sizep = ldsize; | |
940 | return 1; | |
941 | } | |
942 | ||
943 | void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk) | |
944 | { | |
945 | int d; | |
946 | ioctl(mdfd, GET_ARRAY_INFO, ainf); | |
947 | for (d = 0 ; d < ainf->raid_disks + ainf->nr_disks ; d++) | |
948 | if (ioctl(mdfd, GET_DISK_INFO, disk) == 0) | |
949 | return; | |
950 | } | |
951 | ||
952 | int open_container(int fd) | |
953 | { | |
954 | /* 'fd' is a block device. Find out if it is in use | |
955 | * by a container, and return an open fd on that container. | |
956 | */ | |
957 | char path[256]; | |
958 | char *e; | |
959 | DIR *dir; | |
960 | struct dirent *de; | |
961 | int dfd, n; | |
962 | char buf[200]; | |
963 | int major, minor; | |
964 | struct stat st; | |
965 | ||
966 | if (fstat(fd, &st) != 0) | |
967 | return -1; | |
968 | sprintf(path, "/sys/dev/block/%d:%d/holders", | |
969 | (int)major(st.st_rdev), (int)minor(st.st_rdev)); | |
970 | e = path + strlen(path); | |
971 | ||
972 | dir = opendir(path); | |
973 | if (!dir) | |
974 | return -1; | |
975 | while ((de = readdir(dir))) { | |
976 | if (de->d_ino == 0) | |
977 | continue; | |
978 | if (de->d_name[0] == '.') | |
979 | continue; | |
980 | sprintf(e, "/%s/dev", de->d_name); | |
981 | dfd = open(path, O_RDONLY); | |
982 | if (dfd < 0) | |
983 | continue; | |
984 | n = read(dfd, buf, sizeof(buf)); | |
985 | close(dfd); | |
986 | if (n <= 0 || n >= sizeof(buf)) | |
987 | continue; | |
988 | buf[n] = 0; | |
989 | if (sscanf(buf, "%d:%d", &major, &minor) != 2) | |
990 | continue; | |
991 | sprintf(buf, "%d:%d", major, minor); | |
992 | dfd = dev_open(buf, O_RDONLY); | |
993 | if (dfd >= 0) { | |
994 | closedir(dir); | |
995 | return dfd; | |
996 | } | |
997 | } | |
998 | closedir(dir); | |
999 | return -1; | |
1000 | } | |
1001 | ||
1002 | int add_disk(int mdfd, struct supertype *st, | |
1003 | struct mdinfo *sra, struct mdinfo *info) | |
1004 | { | |
1005 | /* Add a device to an array, in one of 2 ways. */ | |
1006 | int rv; | |
1007 | #ifndef MDASSEMBLE | |
1008 | if (st->ss->external) { | |
1009 | rv = sysfs_add_disk(sra, info); | |
1010 | if (! rv) { | |
1011 | struct mdinfo *sd2; | |
1012 | for (sd2 = sra->devs; sd2; sd2=sd2->next) | |
1013 | if (sd2 == info) | |
1014 | break; | |
1015 | if (sd2 == NULL) { | |
1016 | sd2 = malloc(sizeof(*sd2)); | |
1017 | *sd2 = *info; | |
1018 | sd2->next = sra->devs; | |
1019 | sra->devs = sd2; | |
1020 | } | |
1021 | } | |
1022 | } else | |
1023 | #endif | |
1024 | rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk); | |
1025 | return rv; | |
1026 | } | |
1027 | ||
1028 | int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info) | |
1029 | { | |
1030 | /* Initialise kernel's knowledge of array. | |
1031 | * This varies between externally managed arrays | |
1032 | * and older kernels | |
1033 | */ | |
1034 | int vers = md_get_version(mdfd); | |
1035 | int rv; | |
1036 | ||
1037 | #ifndef MDASSEMBLE | |
1038 | if (st->ss->external) | |
1039 | rv = sysfs_set_array(info, vers); | |
1040 | else | |
1041 | #endif | |
1042 | if ((vers % 100) >= 1) { /* can use different versions */ | |
1043 | mdu_array_info_t inf; | |
1044 | memset(&inf, 0, sizeof(inf)); | |
1045 | inf.major_version = info->array.major_version; | |
1046 | inf.minor_version = info->array.minor_version; | |
1047 | rv = ioctl(mdfd, SET_ARRAY_INFO, &inf); | |
1048 | } else | |
1049 | rv = ioctl(mdfd, SET_ARRAY_INFO, NULL); | |
1050 | return rv; | |
1051 | } | |
1052 | ||
1053 | char *devnum2devname(int num) | |
1054 | { | |
1055 | char name[100]; | |
1056 | if (num > 0) | |
1057 | sprintf(name, "md%d", num); | |
1058 | else | |
1059 | sprintf(name, "md_d%d", -1-num); | |
1060 | return strdup(name); | |
1061 | } | |
1062 | ||
1063 | int devname2devnum(char *name) | |
1064 | { | |
1065 | char *ep; | |
1066 | int num; | |
1067 | if (strncmp(name, "md_d", 4)==0) | |
1068 | num = -1-strtoul(name+4, &ep, 10); | |
1069 | else | |
1070 | num = strtoul(name+2, &ep, 10); | |
1071 | return num; | |
1072 | } | |
1073 | ||
1074 | int stat2devnum(struct stat *st) | |
1075 | { | |
1076 | if ((S_IFMT & st->st_mode) == S_IFBLK) { | |
1077 | if (major(st->st_rdev) == MD_MAJOR) | |
1078 | return minor(st->st_rdev); | |
1079 | else | |
1080 | return -1- (minor(st->st_rdev)>>6); | |
1081 | } | |
1082 | return -1; | |
1083 | ||
1084 | } | |
1085 | ||
1086 | int fd2devnum(int fd) | |
1087 | { | |
1088 | struct stat stb; | |
1089 | if (fstat(fd, &stb) == 0) | |
1090 | return stat2devnum(&stb); | |
1091 | return -1; | |
1092 | } | |
1093 | ||
1094 | int mdmon_running(int devnum) | |
1095 | { | |
1096 | char path[100]; | |
1097 | char pid[10]; | |
1098 | int fd; | |
1099 | int n; | |
1100 | sprintf(path, "/var/run/mdadm/%s.pid", devnum2devname(devnum)); | |
1101 | fd = open(path, O_RDONLY, 0); | |
1102 | ||
1103 | if (fd < 0) | |
1104 | return 0; | |
1105 | n = read(fd, pid, 9); | |
1106 | close(fd); | |
1107 | if (n <= 0) | |
1108 | return 0; | |
1109 | if (kill(atoi(pid), 0) == 0) | |
1110 | return 1; | |
1111 | return 0; | |
1112 | } | |
1113 | ||
1114 | int signal_mdmon(int devnum) | |
1115 | { | |
1116 | char path[100]; | |
1117 | char pid[10]; | |
1118 | int fd; | |
1119 | int n; | |
1120 | sprintf(path, "/var/run/mdadm/%s.pid", devnum2devname(devnum)); | |
1121 | fd = open(path, O_RDONLY, 0); | |
1122 | ||
1123 | if (fd < 0) | |
1124 | return 0; | |
1125 | n = read(fd, pid, 9); | |
1126 | close(fd); | |
1127 | if (n <= 0) | |
1128 | return 0; | |
1129 | if (kill(atoi(pid), SIGUSR1) == 0) | |
1130 | return 1; | |
1131 | return 0; | |
1132 | } | |
1133 | ||
1134 | int start_mdmon(int devnum) | |
1135 | { | |
1136 | int i; | |
1137 | int len; | |
1138 | pid_t pid; | |
1139 | int status; | |
1140 | char pathbuf[1024]; | |
1141 | char *paths[4] = { | |
1142 | pathbuf, | |
1143 | "/sbin/mdmon", | |
1144 | "mdmon", | |
1145 | NULL | |
1146 | }; | |
1147 | ||
1148 | if (env_no_mdmon()) | |
1149 | return 0; | |
1150 | ||
1151 | len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf)); | |
1152 | if (len > 0) { | |
1153 | char *sl; | |
1154 | pathbuf[len] = 0; | |
1155 | sl = strrchr(pathbuf, '/'); | |
1156 | if (sl) | |
1157 | sl++; | |
1158 | else | |
1159 | sl = pathbuf; | |
1160 | strcpy(sl, "mdmon"); | |
1161 | } else | |
1162 | pathbuf[0] = '\0'; | |
1163 | ||
1164 | switch(fork()) { | |
1165 | case 0: | |
1166 | /* FIXME yuk. CLOSE_EXEC?? */ | |
1167 | for (i=3; i < 100; i++) | |
1168 | close(i); | |
1169 | for (i=0; paths[i]; i++) | |
1170 | if (paths[i][0]) | |
1171 | execl(paths[i], "mdmon", | |
1172 | map_dev(dev2major(devnum), | |
1173 | dev2minor(devnum), | |
1174 | 1), NULL); | |
1175 | exit(1); | |
1176 | case -1: fprintf(stderr, Name ": cannot run mdmon. " | |
1177 | "Array remains readonly\n"); | |
1178 | return -1; | |
1179 | default: /* parent - good */ | |
1180 | pid = wait(&status); | |
1181 | if (pid < 0 || status != 0) | |
1182 | return -1; | |
1183 | } | |
1184 | return 0; | |
1185 | } | |
1186 | ||
1187 | int env_no_mdmon(void) | |
1188 | { | |
1189 | char *val = getenv("MDADM_NO_MDMON"); | |
1190 | ||
1191 | if (val && atoi(val) == 1) | |
1192 | return 1; | |
1193 | ||
1194 | return 0; | |
1195 | } | |
1196 | ||
1197 | #ifndef MDASSEMBLE | |
1198 | int flush_metadata_updates(struct supertype *st) | |
1199 | { | |
1200 | int sfd; | |
1201 | if (!st->updates) { | |
1202 | st->update_tail = NULL; | |
1203 | return -1; | |
1204 | } | |
1205 | ||
1206 | sfd = connect_monitor(devnum2devname(st->container_dev)); | |
1207 | if (sfd < 0) | |
1208 | return -1; | |
1209 | ||
1210 | while (st->updates) { | |
1211 | struct metadata_update *mu = st->updates; | |
1212 | st->updates = mu->next; | |
1213 | ||
1214 | send_message(sfd, mu, 0); | |
1215 | wait_reply(sfd, 0); | |
1216 | free(mu->buf); | |
1217 | free(mu); | |
1218 | } | |
1219 | ack(sfd, 0); | |
1220 | wait_reply(sfd, 0); | |
1221 | close(sfd); | |
1222 | st->update_tail = NULL; | |
1223 | return 0; | |
1224 | } | |
1225 | ||
1226 | void append_metadata_update(struct supertype *st, void *buf, int len) | |
1227 | { | |
1228 | ||
1229 | struct metadata_update *mu = malloc(sizeof(*mu)); | |
1230 | ||
1231 | mu->buf = buf; | |
1232 | mu->len = len; | |
1233 | mu->space = NULL; | |
1234 | mu->next = NULL; | |
1235 | *st->update_tail = mu; | |
1236 | st->update_tail = &mu->next; | |
1237 | } | |
1238 | #endif /* MDASSEMBLE */ | |
1239 | ||
1240 | #ifdef __TINYC__ | |
1241 | /* tinyc doesn't optimize this check in ioctl.h out ... */ | |
1242 | unsigned int __invalid_size_argument_for_IOC = 0; | |
1243 | #endif | |
1244 |