]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2009 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@suse.de> | |
23 | */ | |
24 | ||
25 | #include "mdadm.h" | |
26 | #include "md_p.h" | |
27 | #include <sys/socket.h> | |
28 | #include <sys/utsname.h> | |
29 | #include <sys/wait.h> | |
30 | #include <sys/un.h> | |
31 | #include <ctype.h> | |
32 | #include <dirent.h> | |
33 | #include <signal.h> | |
34 | ||
35 | /* | |
36 | * following taken from linux/blkpg.h because they aren't | |
37 | * anywhere else and it isn't safe to #include linux/ * stuff. | |
38 | */ | |
39 | ||
40 | #define BLKPG _IO(0x12,105) | |
41 | ||
42 | /* The argument structure */ | |
43 | struct blkpg_ioctl_arg { | |
44 | int op; | |
45 | int flags; | |
46 | int datalen; | |
47 | void *data; | |
48 | }; | |
49 | ||
50 | /* The subfunctions (for the op field) */ | |
51 | #define BLKPG_ADD_PARTITION 1 | |
52 | #define BLKPG_DEL_PARTITION 2 | |
53 | ||
54 | /* Sizes of name fields. Unused at present. */ | |
55 | #define BLKPG_DEVNAMELTH 64 | |
56 | #define BLKPG_VOLNAMELTH 64 | |
57 | ||
58 | /* The data structure for ADD_PARTITION and DEL_PARTITION */ | |
59 | struct blkpg_partition { | |
60 | long long start; /* starting offset in bytes */ | |
61 | long long length; /* length in bytes */ | |
62 | int pno; /* partition number */ | |
63 | char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2, | |
64 | to be used in kernel messages */ | |
65 | char volname[BLKPG_VOLNAMELTH]; /* volume label */ | |
66 | }; | |
67 | ||
68 | #include "part.h" | |
69 | ||
70 | /* Force a compilation error if condition is true */ | |
71 | #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition)) | |
72 | ||
73 | /* Force a compilation error if condition is true, but also produce a | |
74 | result (of value 0 and type size_t), so the expression can be used | |
75 | e.g. in a structure initializer (or where-ever else comma expressions | |
76 | aren't permitted). */ | |
77 | #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); })) | |
78 | ||
79 | /* | |
80 | * Parse a 128 bit uuid in 4 integers | |
81 | * format is 32 hexx nibbles with options :.<space> separator | |
82 | * If not exactly 32 hex digits are found, return 0 | |
83 | * else return 1 | |
84 | */ | |
85 | int parse_uuid(char *str, int uuid[4]) | |
86 | { | |
87 | int hit = 0; /* number of Hex digIT */ | |
88 | int i; | |
89 | char c; | |
90 | for (i=0; i<4; i++) uuid[i]=0; | |
91 | ||
92 | while ((c= *str++)) { | |
93 | int n; | |
94 | if (c>='0' && c<='9') | |
95 | n = c-'0'; | |
96 | else if (c>='a' && c <= 'f') | |
97 | n = 10 + c - 'a'; | |
98 | else if (c>='A' && c <= 'F') | |
99 | n = 10 + c - 'A'; | |
100 | else if (strchr(":. -", c)) | |
101 | continue; | |
102 | else return 0; | |
103 | ||
104 | if (hit<32) { | |
105 | uuid[hit/8] <<= 4; | |
106 | uuid[hit/8] += n; | |
107 | } | |
108 | hit++; | |
109 | } | |
110 | if (hit == 32) | |
111 | return 1; | |
112 | return 0; | |
113 | } | |
114 | ||
115 | ||
116 | /* | |
117 | * Get the md version number. | |
118 | * We use the RAID_VERSION ioctl if it is supported | |
119 | * If not, but we have a block device with major '9', we assume | |
120 | * 0.36.0 | |
121 | * | |
122 | * Return version number as 24 but number - assume version parts | |
123 | * always < 255 | |
124 | */ | |
125 | ||
126 | int md_get_version(int fd) | |
127 | { | |
128 | struct stat stb; | |
129 | mdu_version_t vers; | |
130 | ||
131 | if (fstat(fd, &stb)<0) | |
132 | return -1; | |
133 | if ((S_IFMT&stb.st_mode) != S_IFBLK) | |
134 | return -1; | |
135 | ||
136 | if (ioctl(fd, RAID_VERSION, &vers) == 0) | |
137 | return (vers.major*10000) + (vers.minor*100) + vers.patchlevel; | |
138 | if (errno == EACCES) | |
139 | return -1; | |
140 | if (major(stb.st_rdev) == MD_MAJOR) | |
141 | return (3600); | |
142 | return -1; | |
143 | } | |
144 | ||
145 | int get_linux_version() | |
146 | { | |
147 | struct utsname name; | |
148 | char *cp; | |
149 | int a = 0, b = 0,c = 0; | |
150 | if (uname(&name) <0) | |
151 | return -1; | |
152 | ||
153 | cp = name.release; | |
154 | a = strtoul(cp, &cp, 10); | |
155 | if (*cp == '.') | |
156 | b = strtoul(cp+1, &cp, 10); | |
157 | if (*cp == '.') | |
158 | c = strtoul(cp+1, &cp, 10); | |
159 | ||
160 | return (a*1000000)+(b*1000)+c; | |
161 | } | |
162 | ||
163 | #ifndef MDASSEMBLE | |
164 | int mdadm_version(char *version) | |
165 | { | |
166 | int a, b, c; | |
167 | char *cp; | |
168 | ||
169 | if (!version) | |
170 | version = Version; | |
171 | ||
172 | cp = strchr(version, '-'); | |
173 | if (!cp || *(cp+1) != ' ' || *(cp+2) != 'v') | |
174 | return -1; | |
175 | cp += 3; | |
176 | a = strtoul(cp, &cp, 10); | |
177 | if (*cp != '.') | |
178 | return -1; | |
179 | b = strtoul(cp+1, &cp, 10); | |
180 | if (*cp == '.') | |
181 | c = strtoul(cp+1, &cp, 10); | |
182 | else | |
183 | c = 0; | |
184 | if (*cp != ' ' && *cp != '-') | |
185 | return -1; | |
186 | return (a*1000000)+(b*1000)+c; | |
187 | } | |
188 | ||
189 | long long parse_size(char *size) | |
190 | { | |
191 | /* parse 'size' which should be a number optionally | |
192 | * followed by 'K', 'M', or 'G'. | |
193 | * Without a suffix, K is assumed. | |
194 | * Number returned is in sectors (half-K) | |
195 | */ | |
196 | char *c; | |
197 | long long s = strtoll(size, &c, 10); | |
198 | if (s > 0) { | |
199 | switch (*c) { | |
200 | case 'K': | |
201 | c++; | |
202 | default: | |
203 | s *= 2; | |
204 | break; | |
205 | case 'M': | |
206 | c++; | |
207 | s *= 1024 * 2; | |
208 | break; | |
209 | case 'G': | |
210 | c++; | |
211 | s *= 1024 * 1024 * 2; | |
212 | break; | |
213 | } | |
214 | } | |
215 | if (*c) | |
216 | s = 0; | |
217 | return s; | |
218 | } | |
219 | ||
220 | int parse_layout_10(char *layout) | |
221 | { | |
222 | int copies, rv; | |
223 | char *cp; | |
224 | /* Parse the layout string for raid10 */ | |
225 | /* 'f', 'o' or 'n' followed by a number <= raid_disks */ | |
226 | if ((layout[0] != 'n' && layout[0] != 'f' && layout[0] != 'o') || | |
227 | (copies = strtoul(layout+1, &cp, 10)) < 1 || | |
228 | copies > 200 || | |
229 | *cp) | |
230 | return -1; | |
231 | if (layout[0] == 'n') | |
232 | rv = 256 + copies; | |
233 | else if (layout[0] == 'o') | |
234 | rv = 0x10000 + (copies<<8) + 1; | |
235 | else | |
236 | rv = 1 + (copies<<8); | |
237 | return rv; | |
238 | } | |
239 | ||
240 | int parse_layout_faulty(char *layout) | |
241 | { | |
242 | /* Parse the layout string for 'faulty' */ | |
243 | int ln = strcspn(layout, "0123456789"); | |
244 | char *m = strdup(layout); | |
245 | int mode; | |
246 | m[ln] = 0; | |
247 | mode = map_name(faultylayout, m); | |
248 | if (mode == UnSet) | |
249 | return -1; | |
250 | ||
251 | return mode | (atoi(layout+ln)<< ModeShift); | |
252 | } | |
253 | #endif | |
254 | ||
255 | void remove_partitions(int fd) | |
256 | { | |
257 | /* remove partitions from this block devices. | |
258 | * This is used for components added to an array | |
259 | */ | |
260 | #ifdef BLKPG_DEL_PARTITION | |
261 | struct blkpg_ioctl_arg a; | |
262 | struct blkpg_partition p; | |
263 | ||
264 | a.op = BLKPG_DEL_PARTITION; | |
265 | a.data = (void*)&p; | |
266 | a.datalen = sizeof(p); | |
267 | a.flags = 0; | |
268 | memset(a.data, 0, a.datalen); | |
269 | for (p.pno=0; p.pno < 16; p.pno++) | |
270 | ioctl(fd, BLKPG, &a); | |
271 | #endif | |
272 | } | |
273 | ||
274 | int test_partition(int fd) | |
275 | { | |
276 | /* Check if fd is a whole-disk or a partition. | |
277 | * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION | |
278 | * will return ENXIO on an invalid partition number. | |
279 | */ | |
280 | struct blkpg_ioctl_arg a; | |
281 | struct blkpg_partition p; | |
282 | a.op = BLKPG_DEL_PARTITION; | |
283 | a.data = (void*)&p; | |
284 | a.datalen = sizeof(p); | |
285 | a.flags = 0; | |
286 | memset(a.data, 0, a.datalen); | |
287 | p.pno = 1<<30; | |
288 | if (ioctl(fd, BLKPG, &a) == 0) | |
289 | /* Very unlikely, but not a partition */ | |
290 | return 0; | |
291 | if (errno == ENXIO) | |
292 | /* not a partition */ | |
293 | return 0; | |
294 | ||
295 | return 1; | |
296 | } | |
297 | ||
298 | int test_partition_from_id(dev_t id) | |
299 | { | |
300 | char buf[20]; | |
301 | int fd, rv; | |
302 | ||
303 | sprintf(buf, "%d:%d", major(id), minor(id)); | |
304 | fd = dev_open(buf, O_RDONLY); | |
305 | if (fd < 0) | |
306 | return -1; | |
307 | rv = test_partition(fd); | |
308 | close(fd); | |
309 | return rv; | |
310 | } | |
311 | ||
312 | int enough(int level, int raid_disks, int layout, int clean, | |
313 | char *avail, int avail_disks) | |
314 | { | |
315 | int copies, first; | |
316 | switch (level) { | |
317 | case 10: | |
318 | /* This is the tricky one - we need to check | |
319 | * which actual disks are present. | |
320 | */ | |
321 | copies = (layout&255)* ((layout>>8) & 255); | |
322 | first=0; | |
323 | do { | |
324 | /* there must be one of the 'copies' form 'first' */ | |
325 | int n = copies; | |
326 | int cnt=0; | |
327 | while (n--) { | |
328 | if (avail[first]) | |
329 | cnt++; | |
330 | first = (first+1) % raid_disks; | |
331 | } | |
332 | if (cnt == 0) | |
333 | return 0; | |
334 | ||
335 | } while (first != 0); | |
336 | return 1; | |
337 | ||
338 | case LEVEL_MULTIPATH: | |
339 | return avail_disks>= 1; | |
340 | case LEVEL_LINEAR: | |
341 | case 0: | |
342 | return avail_disks == raid_disks; | |
343 | case 1: | |
344 | return avail_disks >= 1; | |
345 | case 4: | |
346 | case 5: | |
347 | if (clean) | |
348 | return avail_disks >= raid_disks-1; | |
349 | else | |
350 | return avail_disks >= raid_disks; | |
351 | case 6: | |
352 | if (clean) | |
353 | return avail_disks >= raid_disks-2; | |
354 | else | |
355 | return avail_disks >= raid_disks; | |
356 | default: | |
357 | return 0; | |
358 | } | |
359 | } | |
360 | ||
361 | int enough_fd(int fd) | |
362 | { | |
363 | struct mdu_array_info_s array; | |
364 | struct mdu_disk_info_s disk; | |
365 | int avail_disks = 0; | |
366 | int i; | |
367 | char *avail; | |
368 | ||
369 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0 || | |
370 | array.raid_disks <= 0) | |
371 | return 0; | |
372 | avail = calloc(array.raid_disks, 1); | |
373 | for (i=0; i < 1024 && array.nr_disks > 0; i++) { | |
374 | disk.number = i; | |
375 | if (ioctl(fd, GET_DISK_INFO, &disk) != 0) | |
376 | continue; | |
377 | if (disk.major == 0 && disk.minor == 0) | |
378 | continue; | |
379 | array.nr_disks--; | |
380 | ||
381 | if (! (disk.state & (1<<MD_DISK_SYNC))) | |
382 | continue; | |
383 | if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks) | |
384 | continue; | |
385 | avail_disks++; | |
386 | avail[disk.raid_disk] = 1; | |
387 | } | |
388 | /* This is used on an active array, so assume it is clean */ | |
389 | return enough(array.level, array.raid_disks, array.layout, | |
390 | 1, | |
391 | avail, avail_disks); | |
392 | } | |
393 | ||
394 | ||
395 | const int uuid_zero[4] = { 0, 0, 0, 0 }; | |
396 | ||
397 | int same_uuid(int a[4], int b[4], int swapuuid) | |
398 | { | |
399 | if (swapuuid) { | |
400 | /* parse uuids are hostendian. | |
401 | * uuid's from some superblocks are big-ending | |
402 | * if there is a difference, we need to swap.. | |
403 | */ | |
404 | unsigned char *ac = (unsigned char *)a; | |
405 | unsigned char *bc = (unsigned char *)b; | |
406 | int i; | |
407 | for (i=0; i<16; i+= 4) { | |
408 | if (ac[i+0] != bc[i+3] || | |
409 | ac[i+1] != bc[i+2] || | |
410 | ac[i+2] != bc[i+1] || | |
411 | ac[i+3] != bc[i+0]) | |
412 | return 0; | |
413 | } | |
414 | return 1; | |
415 | } else { | |
416 | if (a[0]==b[0] && | |
417 | a[1]==b[1] && | |
418 | a[2]==b[2] && | |
419 | a[3]==b[3]) | |
420 | return 1; | |
421 | return 0; | |
422 | } | |
423 | } | |
424 | void copy_uuid(void *a, int b[4], int swapuuid) | |
425 | { | |
426 | if (swapuuid) { | |
427 | /* parse uuids are hostendian. | |
428 | * uuid's from some superblocks are big-ending | |
429 | * if there is a difference, we need to swap.. | |
430 | */ | |
431 | unsigned char *ac = (unsigned char *)a; | |
432 | unsigned char *bc = (unsigned char *)b; | |
433 | int i; | |
434 | for (i=0; i<16; i+= 4) { | |
435 | ac[i+0] = bc[i+3]; | |
436 | ac[i+1] = bc[i+2]; | |
437 | ac[i+2] = bc[i+1]; | |
438 | ac[i+3] = bc[i+0]; | |
439 | } | |
440 | } else | |
441 | memcpy(a, b, 16); | |
442 | } | |
443 | ||
444 | char *__fname_from_uuid(int id[4], int swap, char *buf, char sep) | |
445 | { | |
446 | int i, j; | |
447 | char uuid[16]; | |
448 | char *c = buf; | |
449 | strcpy(c, "UUID-"); | |
450 | c += strlen(c); | |
451 | copy_uuid(uuid, id, swap); | |
452 | for (i = 0; i < 4; i++) { | |
453 | if (i) | |
454 | *c++ = sep; | |
455 | for (j = 3; j >= 0; j--) { | |
456 | sprintf(c,"%02x", (unsigned char) uuid[j+4*i]); | |
457 | c+= 2; | |
458 | } | |
459 | } | |
460 | return buf; | |
461 | ||
462 | } | |
463 | ||
464 | char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep) | |
465 | { | |
466 | // dirty hack to work around an issue with super1 superblocks... | |
467 | // super1 superblocks need swapuuid set in order for assembly to | |
468 | // work, but can't have it set if we want this printout to match | |
469 | // all the other uuid printouts in super1.c, so we force swapuuid | |
470 | // to 1 to make our printout match the rest of super1 | |
471 | return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep); | |
472 | } | |
473 | ||
474 | #ifndef MDASSEMBLE | |
475 | int check_ext2(int fd, char *name) | |
476 | { | |
477 | /* | |
478 | * Check for an ext2fs file system. | |
479 | * Superblock is always 1K at 1K offset | |
480 | * | |
481 | * s_magic is le16 at 56 == 0xEF53 | |
482 | * report mtime - le32 at 44 | |
483 | * blocks - le32 at 4 | |
484 | * logblksize - le32 at 24 | |
485 | */ | |
486 | unsigned char sb[1024]; | |
487 | time_t mtime; | |
488 | int size, bsize; | |
489 | if (lseek(fd, 1024,0)!= 1024) | |
490 | return 0; | |
491 | if (read(fd, sb, 1024)!= 1024) | |
492 | return 0; | |
493 | if (sb[56] != 0x53 || sb[57] != 0xef) | |
494 | return 0; | |
495 | ||
496 | mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8; | |
497 | bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8; | |
498 | size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8; | |
499 | fprintf(stderr, Name ": %s appears to contain an ext2fs file system\n", | |
500 | name); | |
501 | fprintf(stderr," size=%dK mtime=%s", | |
502 | size*(1<<bsize), ctime(&mtime)); | |
503 | return 1; | |
504 | } | |
505 | ||
506 | int check_reiser(int fd, char *name) | |
507 | { | |
508 | /* | |
509 | * superblock is at 64K | |
510 | * size is 1024; | |
511 | * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52 | |
512 | * | |
513 | */ | |
514 | unsigned char sb[1024]; | |
515 | unsigned long size; | |
516 | if (lseek(fd, 64*1024, 0) != 64*1024) | |
517 | return 0; | |
518 | if (read(fd, sb, 1024) != 1024) | |
519 | return 0; | |
520 | if (strncmp((char*)sb+52, "ReIsErFs",8)!=0 && | |
521 | strncmp((char*)sb+52, "ReIsEr2Fs",9)!=0) | |
522 | return 0; | |
523 | fprintf(stderr, Name ": %s appears to contain a reiserfs file system\n",name); | |
524 | size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8; | |
525 | fprintf(stderr, " size = %luK\n", size*4); | |
526 | ||
527 | return 1; | |
528 | } | |
529 | ||
530 | int check_raid(int fd, char *name) | |
531 | { | |
532 | struct mdinfo info; | |
533 | time_t crtime; | |
534 | char *level; | |
535 | struct supertype *st = guess_super(fd); | |
536 | ||
537 | if (!st) return 0; | |
538 | st->ignore_hw_compat = 1; | |
539 | st->ss->load_super(st, fd, name); | |
540 | /* Looks like a raid array .. */ | |
541 | fprintf(stderr, Name ": %s appears to be part of a raid array:\n", | |
542 | name); | |
543 | st->ss->getinfo_super(st, &info, NULL); | |
544 | st->ss->free_super(st); | |
545 | crtime = info.array.ctime; | |
546 | level = map_num(pers, info.array.level); | |
547 | if (!level) level = "-unknown-"; | |
548 | fprintf(stderr, " level=%s devices=%d ctime=%s", | |
549 | level, info.array.raid_disks, ctime(&crtime)); | |
550 | return 1; | |
551 | } | |
552 | ||
553 | int ask(char *mesg) | |
554 | { | |
555 | char *add = ""; | |
556 | int i; | |
557 | for (i=0; i<5; i++) { | |
558 | char buf[100]; | |
559 | fprintf(stderr, "%s%s", mesg, add); | |
560 | fflush(stderr); | |
561 | if (fgets(buf, 100, stdin)==NULL) | |
562 | return 0; | |
563 | if (buf[0]=='y' || buf[0]=='Y') | |
564 | return 1; | |
565 | if (buf[0]=='n' || buf[0]=='N') | |
566 | return 0; | |
567 | add = "(y/n) "; | |
568 | } | |
569 | fprintf(stderr, Name ": assuming 'no'\n"); | |
570 | return 0; | |
571 | } | |
572 | #endif /* MDASSEMBLE */ | |
573 | ||
574 | int is_standard(char *dev, int *nump) | |
575 | { | |
576 | /* tests if dev is a "standard" md dev name. | |
577 | * i.e if the last component is "/dNN" or "/mdNN", | |
578 | * where NN is a string of digits | |
579 | * Returns 1 if a partitionable standard, | |
580 | * -1 if non-partitonable, | |
581 | * 0 if not a standard name. | |
582 | */ | |
583 | char *d = strrchr(dev, '/'); | |
584 | int type=0; | |
585 | int num; | |
586 | if (!d) | |
587 | return 0; | |
588 | if (strncmp(d, "/d",2)==0) | |
589 | d += 2, type=1; /* /dev/md/dN{pM} */ | |
590 | else if (strncmp(d, "/md_d", 5)==0) | |
591 | d += 5, type=1; /* /dev/md_dN{pM} */ | |
592 | else if (strncmp(d, "/md", 3)==0) | |
593 | d += 3, type=-1; /* /dev/mdN */ | |
594 | else if (d-dev > 3 && strncmp(d-2, "md/", 3)==0) | |
595 | d += 1, type=-1; /* /dev/md/N */ | |
596 | else | |
597 | return 0; | |
598 | if (!*d) | |
599 | return 0; | |
600 | num = atoi(d); | |
601 | while (isdigit(*d)) | |
602 | d++; | |
603 | if (*d) | |
604 | return 0; | |
605 | if (nump) *nump = num; | |
606 | ||
607 | return type; | |
608 | } | |
609 | ||
610 | unsigned long calc_csum(void *super, int bytes) | |
611 | { | |
612 | unsigned long long newcsum = 0; | |
613 | int i; | |
614 | unsigned int csum; | |
615 | unsigned int *superc = (unsigned int*) super; | |
616 | ||
617 | for(i=0; i<bytes/4; i++) | |
618 | newcsum+= superc[i]; | |
619 | csum = (newcsum& 0xffffffff) + (newcsum>>32); | |
620 | #ifdef __alpha__ | |
621 | /* The in-kernel checksum calculation is always 16bit on | |
622 | * the alpha, though it is 32 bit on i386... | |
623 | * I wonder what it is elsewhere... (it uses and API in | |
624 | * a way that it shouldn't). | |
625 | */ | |
626 | csum = (csum & 0xffff) + (csum >> 16); | |
627 | csum = (csum & 0xffff) + (csum >> 16); | |
628 | #endif | |
629 | return csum; | |
630 | } | |
631 | ||
632 | #ifndef MDASSEMBLE | |
633 | char *human_size(long long bytes) | |
634 | { | |
635 | static char buf[30]; | |
636 | ||
637 | /* We convert bytes to either centi-M{ega,ibi}bytes or | |
638 | * centi-G{igi,ibi}bytes, with appropriate rounding, | |
639 | * and then print 1/100th of those as a decimal. | |
640 | * We allow upto 2048Megabytes before converting to | |
641 | * gigabytes, as that shows more precision and isn't | |
642 | * too large a number. | |
643 | * Terrabytes are not yet handled. | |
644 | */ | |
645 | ||
646 | if (bytes < 5000*1024) | |
647 | buf[0]=0; | |
648 | else if (bytes < 2*1024LL*1024LL*1024LL) { | |
649 | long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2; | |
650 | long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2; | |
651 | snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)", | |
652 | cMiB/100 , cMiB % 100, | |
653 | cMB/100, cMB % 100); | |
654 | } else { | |
655 | long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2; | |
656 | long cGB = (bytes / (1000000000LL/200LL ) +1) /2; | |
657 | snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)", | |
658 | cGiB/100 , cGiB % 100, | |
659 | cGB/100, cGB % 100); | |
660 | } | |
661 | return buf; | |
662 | } | |
663 | ||
664 | char *human_size_brief(long long bytes) | |
665 | { | |
666 | static char buf[30]; | |
667 | ||
668 | if (bytes < 5000*1024) | |
669 | snprintf(buf, sizeof(buf), "%ld.%02ldKiB", | |
670 | (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024) | |
671 | ); | |
672 | else if (bytes < 2*1024LL*1024LL*1024LL) | |
673 | snprintf(buf, sizeof(buf), "%ld.%02ldMiB", | |
674 | (long)(bytes>>20), | |
675 | (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100) | |
676 | ); | |
677 | else | |
678 | snprintf(buf, sizeof(buf), "%ld.%02ldGiB", | |
679 | (long)(bytes>>30), | |
680 | (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100) | |
681 | ); | |
682 | return buf; | |
683 | } | |
684 | ||
685 | void print_r10_layout(int layout) | |
686 | { | |
687 | int near = layout & 255; | |
688 | int far = (layout >> 8) & 255; | |
689 | int offset = (layout&0x10000); | |
690 | char *sep = ""; | |
691 | ||
692 | if (near != 1) { | |
693 | printf("%s near=%d", sep, near); | |
694 | sep = ","; | |
695 | } | |
696 | if (far != 1) | |
697 | printf("%s %s=%d", sep, offset?"offset":"far", far); | |
698 | if (near*far == 1) | |
699 | printf("NO REDUNDANCY"); | |
700 | } | |
701 | #endif | |
702 | ||
703 | unsigned long long calc_array_size(int level, int raid_disks, int layout, | |
704 | int chunksize, unsigned long long devsize) | |
705 | { | |
706 | int data_disks = 0; | |
707 | switch (level) { | |
708 | case 0: data_disks = raid_disks; break; | |
709 | case 1: data_disks = 1; break; | |
710 | case 4: | |
711 | case 5: data_disks = raid_disks - 1; break; | |
712 | case 6: data_disks = raid_disks - 2; break; | |
713 | case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255); | |
714 | break; | |
715 | } | |
716 | devsize &= ~(unsigned long long)((chunksize>>9)-1); | |
717 | return data_disks * devsize; | |
718 | } | |
719 | ||
720 | #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) | |
721 | char *get_md_name(int dev) | |
722 | { | |
723 | /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */ | |
724 | /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */ | |
725 | static char devname[50]; | |
726 | struct stat stb; | |
727 | dev_t rdev; | |
728 | char *dn; | |
729 | ||
730 | if (dev < 0) { | |
731 | int mdp = get_mdp_major(); | |
732 | if (mdp < 0) return NULL; | |
733 | rdev = makedev(mdp, (-1-dev)<<6); | |
734 | snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev); | |
735 | if (stat(devname, &stb) == 0 | |
736 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
737 | && (stb.st_rdev == rdev)) | |
738 | return devname; | |
739 | } else { | |
740 | rdev = makedev(MD_MAJOR, dev); | |
741 | snprintf(devname, sizeof(devname), "/dev/md%d", dev); | |
742 | if (stat(devname, &stb) == 0 | |
743 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
744 | && (stb.st_rdev == rdev)) | |
745 | return devname; | |
746 | ||
747 | snprintf(devname, sizeof(devname), "/dev/md/%d", dev); | |
748 | if (stat(devname, &stb) == 0 | |
749 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
750 | && (stb.st_rdev == rdev)) | |
751 | return devname; | |
752 | } | |
753 | dn = map_dev(major(rdev), minor(rdev), 0); | |
754 | if (dn) | |
755 | return dn; | |
756 | snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev); | |
757 | if (mknod(devname, S_IFBLK | 0600, rdev) == -1) | |
758 | if (errno != EEXIST) | |
759 | return NULL; | |
760 | ||
761 | if (stat(devname, &stb) == 0 | |
762 | && (S_IFMT&stb.st_mode) == S_IFBLK | |
763 | && (stb.st_rdev == rdev)) | |
764 | return devname; | |
765 | unlink(devname); | |
766 | return NULL; | |
767 | } | |
768 | ||
769 | void put_md_name(char *name) | |
770 | { | |
771 | if (strncmp(name, "/dev/.tmp.md", 12)==0) | |
772 | unlink(name); | |
773 | } | |
774 | ||
775 | int find_free_devnum(int use_partitions) | |
776 | { | |
777 | int devnum; | |
778 | for (devnum = 127; devnum != 128; | |
779 | devnum = devnum ? devnum-1 : (1<<20)-1) { | |
780 | char *dn; | |
781 | int _devnum; | |
782 | ||
783 | _devnum = use_partitions ? (-1-devnum) : devnum; | |
784 | if (mddev_busy(_devnum)) | |
785 | continue; | |
786 | /* make sure it is new to /dev too, at least as a | |
787 | * non-standard */ | |
788 | dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0); | |
789 | if (dn && ! is_standard(dn, NULL)) | |
790 | continue; | |
791 | break; | |
792 | } | |
793 | if (devnum == 128) | |
794 | return NoMdDev; | |
795 | return use_partitions ? (-1-devnum) : devnum; | |
796 | } | |
797 | #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */ | |
798 | ||
799 | int dev_open(char *dev, int flags) | |
800 | { | |
801 | /* like 'open', but if 'dev' matches %d:%d, create a temp | |
802 | * block device and open that | |
803 | */ | |
804 | char *e; | |
805 | int fd = -1; | |
806 | char devname[32]; | |
807 | int major; | |
808 | int minor; | |
809 | ||
810 | if (!dev) return -1; | |
811 | flags |= O_DIRECT; | |
812 | ||
813 | major = strtoul(dev, &e, 0); | |
814 | if (e > dev && *e == ':' && e[1] && | |
815 | (minor = strtoul(e+1, &e, 0)) >= 0 && | |
816 | *e == 0) { | |
817 | char *path = map_dev(major, minor, 0); | |
818 | if (path) | |
819 | fd = open(path, flags); | |
820 | if (fd < 0) { | |
821 | snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d", | |
822 | (int)getpid(), major, minor); | |
823 | if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) { | |
824 | fd = open(devname, flags); | |
825 | unlink(devname); | |
826 | } | |
827 | } | |
828 | if (fd < 0) { | |
829 | snprintf(devname, sizeof(devname), "/tmp/.tmp.md.%d:%d:%d", | |
830 | (int)getpid(), major, minor); | |
831 | if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) { | |
832 | fd = open(devname, flags); | |
833 | unlink(devname); | |
834 | } | |
835 | } | |
836 | } else | |
837 | fd = open(dev, flags); | |
838 | return fd; | |
839 | } | |
840 | ||
841 | int open_dev_flags(int devnum, int flags) | |
842 | { | |
843 | char buf[20]; | |
844 | ||
845 | sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum)); | |
846 | return dev_open(buf, flags); | |
847 | } | |
848 | ||
849 | int open_dev(int devnum) | |
850 | { | |
851 | return open_dev_flags(devnum, O_RDONLY); | |
852 | } | |
853 | ||
854 | int open_dev_excl(int devnum) | |
855 | { | |
856 | char buf[20]; | |
857 | int i; | |
858 | int flags = O_RDWR; | |
859 | ||
860 | sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum)); | |
861 | for (i=0 ; i<25 ; i++) { | |
862 | int fd = dev_open(buf, flags|O_EXCL); | |
863 | if (fd >= 0) | |
864 | return fd; | |
865 | if (errno == EACCES && flags == O_RDWR) { | |
866 | flags = O_RDONLY; | |
867 | continue; | |
868 | } | |
869 | if (errno != EBUSY) | |
870 | return fd; | |
871 | usleep(200000); | |
872 | } | |
873 | return -1; | |
874 | } | |
875 | ||
876 | int same_dev(char *one, char *two) | |
877 | { | |
878 | struct stat st1, st2; | |
879 | if (stat(one, &st1) != 0) | |
880 | return 0; | |
881 | if (stat(two, &st2) != 0) | |
882 | return 0; | |
883 | if ((st1.st_mode & S_IFMT) != S_IFBLK) | |
884 | return 0; | |
885 | if ((st2.st_mode & S_IFMT) != S_IFBLK) | |
886 | return 0; | |
887 | return st1.st_rdev == st2.st_rdev; | |
888 | } | |
889 | ||
890 | void wait_for(char *dev, int fd) | |
891 | { | |
892 | int i; | |
893 | struct stat stb_want; | |
894 | ||
895 | if (fstat(fd, &stb_want) != 0 || | |
896 | (stb_want.st_mode & S_IFMT) != S_IFBLK) | |
897 | return; | |
898 | ||
899 | for (i=0 ; i<25 ; i++) { | |
900 | struct stat stb; | |
901 | if (stat(dev, &stb) == 0 && | |
902 | (stb.st_mode & S_IFMT) == S_IFBLK && | |
903 | (stb.st_rdev == stb_want.st_rdev)) | |
904 | return; | |
905 | usleep(200000); | |
906 | } | |
907 | if (i == 25) | |
908 | dprintf("%s: timeout waiting for %s\n", __func__, dev); | |
909 | } | |
910 | ||
911 | struct superswitch *superlist[] = | |
912 | { | |
913 | &super0, &super1, | |
914 | &super_ddf, &super_imsm, | |
915 | &mbr, &gpt, | |
916 | NULL }; | |
917 | ||
918 | #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) | |
919 | ||
920 | struct supertype *super_by_fd(int fd, char **subarrayp) | |
921 | { | |
922 | mdu_array_info_t array; | |
923 | int vers; | |
924 | int minor; | |
925 | struct supertype *st = NULL; | |
926 | struct mdinfo *sra; | |
927 | char *verstr; | |
928 | char version[20]; | |
929 | int i; | |
930 | char *subarray = NULL; | |
931 | int container = NoMdDev; | |
932 | ||
933 | sra = sysfs_read(fd, 0, GET_VERSION); | |
934 | ||
935 | if (sra) { | |
936 | vers = sra->array.major_version; | |
937 | minor = sra->array.minor_version; | |
938 | verstr = sra->text_version; | |
939 | } else { | |
940 | if (ioctl(fd, GET_ARRAY_INFO, &array)) | |
941 | array.major_version = array.minor_version = 0; | |
942 | vers = array.major_version; | |
943 | minor = array.minor_version; | |
944 | verstr = ""; | |
945 | } | |
946 | ||
947 | if (vers != -1) { | |
948 | sprintf(version, "%d.%d", vers, minor); | |
949 | verstr = version; | |
950 | } | |
951 | if (minor == -2 && is_subarray(verstr)) { | |
952 | char *dev = verstr+1; | |
953 | ||
954 | subarray = strchr(dev, '/'); | |
955 | if (subarray) | |
956 | *subarray++ = '\0'; | |
957 | subarray = strdup(subarray); | |
958 | container = devname2devnum(dev); | |
959 | if (sra) | |
960 | sysfs_free(sra); | |
961 | sra = sysfs_read(-1, container, GET_VERSION); | |
962 | if (sra && sra->text_version[0]) | |
963 | verstr = sra->text_version; | |
964 | else | |
965 | verstr = "-no-metadata-"; | |
966 | } | |
967 | ||
968 | for (i = 0; st == NULL && superlist[i] ; i++) | |
969 | st = superlist[i]->match_metadata_desc(verstr); | |
970 | ||
971 | if (sra) | |
972 | sysfs_free(sra); | |
973 | if (st) { | |
974 | st->sb = NULL; | |
975 | if (subarrayp) | |
976 | *subarrayp = subarray; | |
977 | st->container_dev = container; | |
978 | st->devnum = fd2devnum(fd); | |
979 | } else | |
980 | free(subarray); | |
981 | ||
982 | return st; | |
983 | } | |
984 | #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */ | |
985 | ||
986 | int dev_size_from_id(dev_t id, unsigned long long *size) | |
987 | { | |
988 | char buf[20]; | |
989 | int fd; | |
990 | ||
991 | sprintf(buf, "%d:%d", major(id), minor(id)); | |
992 | fd = dev_open(buf, O_RDONLY); | |
993 | if (fd < 0) | |
994 | return 0; | |
995 | if (get_dev_size(fd, NULL, size)) { | |
996 | close(fd); | |
997 | return 1; | |
998 | } | |
999 | close(fd); | |
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | struct supertype *dup_super(struct supertype *orig) | |
1004 | { | |
1005 | struct supertype *st; | |
1006 | ||
1007 | if (!orig) | |
1008 | return orig; | |
1009 | st = malloc(sizeof(*st)); | |
1010 | if (!st) | |
1011 | return st; | |
1012 | memset(st, 0, sizeof(*st)); | |
1013 | st->ss = orig->ss; | |
1014 | st->max_devs = orig->max_devs; | |
1015 | st->minor_version = orig->minor_version; | |
1016 | st->sb = NULL; | |
1017 | st->info = NULL; | |
1018 | return st; | |
1019 | } | |
1020 | ||
1021 | struct supertype *guess_super_type(int fd, enum guess_types guess_type) | |
1022 | { | |
1023 | /* try each load_super to find the best match, | |
1024 | * and return the best superswitch | |
1025 | */ | |
1026 | struct superswitch *ss; | |
1027 | struct supertype *st; | |
1028 | time_t besttime = 0; | |
1029 | int bestsuper = -1; | |
1030 | int i; | |
1031 | ||
1032 | st = malloc(sizeof(*st)); | |
1033 | memset(st, 0, sizeof(*st)); | |
1034 | st->container_dev = NoMdDev; | |
1035 | ||
1036 | for (i=0 ; superlist[i]; i++) { | |
1037 | int rv; | |
1038 | ss = superlist[i]; | |
1039 | if (guess_type == guess_array && ss->add_to_super == NULL) | |
1040 | continue; | |
1041 | if (guess_type == guess_partitions && ss->add_to_super != NULL) | |
1042 | continue; | |
1043 | memset(st, 0, sizeof(*st)); | |
1044 | st->ignore_hw_compat = 1; | |
1045 | rv = ss->load_super(st, fd, NULL); | |
1046 | if (rv == 0) { | |
1047 | struct mdinfo info; | |
1048 | st->ss->getinfo_super(st, &info, NULL); | |
1049 | if (bestsuper == -1 || | |
1050 | besttime < info.array.ctime) { | |
1051 | bestsuper = i; | |
1052 | besttime = info.array.ctime; | |
1053 | } | |
1054 | ss->free_super(st); | |
1055 | } | |
1056 | } | |
1057 | if (bestsuper != -1) { | |
1058 | int rv; | |
1059 | memset(st, 0, sizeof(*st)); | |
1060 | st->ignore_hw_compat = 1; | |
1061 | rv = superlist[bestsuper]->load_super(st, fd, NULL); | |
1062 | if (rv == 0) { | |
1063 | superlist[bestsuper]->free_super(st); | |
1064 | st->ignore_hw_compat = 0; | |
1065 | return st; | |
1066 | } | |
1067 | } | |
1068 | free(st); | |
1069 | return NULL; | |
1070 | } | |
1071 | ||
1072 | /* Return size of device in bytes */ | |
1073 | int get_dev_size(int fd, char *dname, unsigned long long *sizep) | |
1074 | { | |
1075 | unsigned long long ldsize; | |
1076 | struct stat st; | |
1077 | ||
1078 | if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode)) | |
1079 | ldsize = (unsigned long long)st.st_size; | |
1080 | else | |
1081 | #ifdef BLKGETSIZE64 | |
1082 | if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0) | |
1083 | #endif | |
1084 | { | |
1085 | unsigned long dsize; | |
1086 | if (ioctl(fd, BLKGETSIZE, &dsize) == 0) { | |
1087 | ldsize = dsize; | |
1088 | ldsize <<= 9; | |
1089 | } else { | |
1090 | if (dname) | |
1091 | fprintf(stderr, Name ": Cannot get size of %s: %s\b", | |
1092 | dname, strerror(errno)); | |
1093 | return 0; | |
1094 | } | |
1095 | } | |
1096 | *sizep = ldsize; | |
1097 | return 1; | |
1098 | } | |
1099 | ||
1100 | /* Return true if this can only be a container, not a member device. | |
1101 | * i.e. is and md device and size is zero | |
1102 | */ | |
1103 | int must_be_container(int fd) | |
1104 | { | |
1105 | unsigned long long size; | |
1106 | if (md_get_version(fd) < 0) | |
1107 | return 0; | |
1108 | if (get_dev_size(fd, NULL, &size) == 0) | |
1109 | return 1; | |
1110 | if (size == 0) | |
1111 | return 1; | |
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | /* Sets endofpart parameter to the last block used by the last GPT partition on the device. | |
1116 | * Returns: 1 if successful | |
1117 | * -1 for unknown partition type | |
1118 | * 0 for other errors | |
1119 | */ | |
1120 | static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart) | |
1121 | { | |
1122 | struct GPT gpt; | |
1123 | unsigned char empty_gpt_entry[16]= {0}; | |
1124 | struct GPT_part_entry part; | |
1125 | unsigned long long curr_part_end; | |
1126 | unsigned all_partitions, entry_size; | |
1127 | unsigned part_nr; | |
1128 | ||
1129 | *endofpart = 0; | |
1130 | ||
1131 | BUILD_BUG_ON(sizeof(gpt) != 512); | |
1132 | /* skip protective MBR */ | |
1133 | lseek(fd, 512, SEEK_SET); | |
1134 | /* read GPT header */ | |
1135 | if (read(fd, &gpt, 512) != 512) | |
1136 | return 0; | |
1137 | ||
1138 | /* get the number of partition entries and the entry size */ | |
1139 | all_partitions = __le32_to_cpu(gpt.part_cnt); | |
1140 | entry_size = __le32_to_cpu(gpt.part_size); | |
1141 | ||
1142 | /* Check GPT signature*/ | |
1143 | if (gpt.magic != GPT_SIGNATURE_MAGIC) | |
1144 | return -1; | |
1145 | ||
1146 | /* sanity checks */ | |
1147 | if (all_partitions > 1024 || | |
1148 | entry_size > 512) | |
1149 | return -1; | |
1150 | ||
1151 | for (part_nr=0; part_nr < all_partitions; part_nr++) { | |
1152 | /* read partition entry */ | |
1153 | if (read(fd, &part, entry_size) != (ssize_t)entry_size) | |
1154 | return 0; | |
1155 | ||
1156 | /* is this valid partition? */ | |
1157 | if (memcmp(part.type_guid, empty_gpt_entry, 16) != 0) { | |
1158 | /* check the last lba for the current partition */ | |
1159 | curr_part_end = __le64_to_cpu(part.ending_lba); | |
1160 | if (curr_part_end > *endofpart) | |
1161 | *endofpart = curr_part_end; | |
1162 | } | |
1163 | ||
1164 | } | |
1165 | return 1; | |
1166 | } | |
1167 | ||
1168 | /* Sets endofpart parameter to the last block used by the last partition on the device. | |
1169 | * Returns: 1 if successful | |
1170 | * -1 for unknown partition type | |
1171 | * 0 for other errors | |
1172 | */ | |
1173 | static int get_last_partition_end(int fd, unsigned long long *endofpart) | |
1174 | { | |
1175 | struct MBR boot_sect; | |
1176 | struct MBR_part_record *part; | |
1177 | unsigned long long curr_part_end; | |
1178 | unsigned part_nr; | |
1179 | int retval = 0; | |
1180 | ||
1181 | *endofpart = 0; | |
1182 | ||
1183 | BUILD_BUG_ON(sizeof(boot_sect) != 512); | |
1184 | /* read MBR */ | |
1185 | lseek(fd, 0, 0); | |
1186 | if (read(fd, &boot_sect, 512) != 512) | |
1187 | goto abort; | |
1188 | ||
1189 | /* check MBP signature */ | |
1190 | if (boot_sect.magic == MBR_SIGNATURE_MAGIC) { | |
1191 | retval = 1; | |
1192 | /* found the correct signature */ | |
1193 | part = boot_sect.parts; | |
1194 | ||
1195 | for (part_nr=0; part_nr < MBR_PARTITIONS; part_nr++) { | |
1196 | /* check for GPT type */ | |
1197 | if (part->part_type == MBR_GPT_PARTITION_TYPE) { | |
1198 | retval = get_gpt_last_partition_end(fd, endofpart); | |
1199 | break; | |
1200 | } | |
1201 | /* check the last used lba for the current partition */ | |
1202 | curr_part_end = __le32_to_cpu(part->first_sect_lba) + | |
1203 | __le32_to_cpu(part->blocks_num); | |
1204 | if (curr_part_end > *endofpart) | |
1205 | *endofpart = curr_part_end; | |
1206 | ||
1207 | part++; | |
1208 | } | |
1209 | } else { | |
1210 | /* Unknown partition table */ | |
1211 | retval = -1; | |
1212 | } | |
1213 | abort: | |
1214 | return retval; | |
1215 | } | |
1216 | ||
1217 | int check_partitions(int fd, char *dname, unsigned long long freesize, | |
1218 | unsigned long long size) | |
1219 | { | |
1220 | /* | |
1221 | * Check where the last partition ends | |
1222 | */ | |
1223 | unsigned long long endofpart; | |
1224 | int ret; | |
1225 | ||
1226 | if ((ret = get_last_partition_end(fd, &endofpart)) > 0) { | |
1227 | /* There appears to be a partition table here */ | |
1228 | if (freesize == 0) { | |
1229 | /* partitions will not be visible in new device */ | |
1230 | fprintf(stderr, | |
1231 | Name ": partition table exists on %s but will be lost or\n" | |
1232 | " meaningless after creating array\n", | |
1233 | dname); | |
1234 | return 1; | |
1235 | } else if (endofpart > freesize) { | |
1236 | /* last partition overlaps metadata */ | |
1237 | fprintf(stderr, | |
1238 | Name ": metadata will over-write last partition on %s.\n", | |
1239 | dname); | |
1240 | return 1; | |
1241 | } else if (size && endofpart > size) { | |
1242 | /* partitions will be truncated in new device */ | |
1243 | fprintf(stderr, | |
1244 | Name ": array size is too small to cover all partitions on %s.\n", | |
1245 | dname); | |
1246 | return 1; | |
1247 | } | |
1248 | } | |
1249 | return 0; | |
1250 | } | |
1251 | ||
1252 | void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk) | |
1253 | { | |
1254 | int d; | |
1255 | ||
1256 | ioctl(mdfd, GET_ARRAY_INFO, ainf); | |
1257 | for (d = 0 ; d < 1024 ; d++) { | |
1258 | if (ioctl(mdfd, GET_DISK_INFO, disk) == 0 && | |
1259 | (disk->major || disk->minor)) | |
1260 | return; | |
1261 | } | |
1262 | } | |
1263 | ||
1264 | int open_container(int fd) | |
1265 | { | |
1266 | /* 'fd' is a block device. Find out if it is in use | |
1267 | * by a container, and return an open fd on that container. | |
1268 | */ | |
1269 | char path[256]; | |
1270 | char *e; | |
1271 | DIR *dir; | |
1272 | struct dirent *de; | |
1273 | int dfd, n; | |
1274 | char buf[200]; | |
1275 | int major, minor; | |
1276 | struct stat st; | |
1277 | ||
1278 | if (fstat(fd, &st) != 0) | |
1279 | return -1; | |
1280 | sprintf(path, "/sys/dev/block/%d:%d/holders", | |
1281 | (int)major(st.st_rdev), (int)minor(st.st_rdev)); | |
1282 | e = path + strlen(path); | |
1283 | ||
1284 | dir = opendir(path); | |
1285 | if (!dir) | |
1286 | return -1; | |
1287 | while ((de = readdir(dir))) { | |
1288 | if (de->d_ino == 0) | |
1289 | continue; | |
1290 | if (de->d_name[0] == '.') | |
1291 | continue; | |
1292 | sprintf(e, "/%s/dev", de->d_name); | |
1293 | dfd = open(path, O_RDONLY); | |
1294 | if (dfd < 0) | |
1295 | continue; | |
1296 | n = read(dfd, buf, sizeof(buf)); | |
1297 | close(dfd); | |
1298 | if (n <= 0 || (unsigned)n >= sizeof(buf)) | |
1299 | continue; | |
1300 | buf[n] = 0; | |
1301 | if (sscanf(buf, "%d:%d", &major, &minor) != 2) | |
1302 | continue; | |
1303 | sprintf(buf, "%d:%d", major, minor); | |
1304 | dfd = dev_open(buf, O_RDONLY); | |
1305 | if (dfd >= 0) { | |
1306 | closedir(dir); | |
1307 | return dfd; | |
1308 | } | |
1309 | } | |
1310 | closedir(dir); | |
1311 | return -1; | |
1312 | } | |
1313 | ||
1314 | struct superswitch *version_to_superswitch(char *vers) | |
1315 | { | |
1316 | int i; | |
1317 | ||
1318 | for (i = 0; superlist[i]; i++) { | |
1319 | struct superswitch *ss = superlist[i]; | |
1320 | ||
1321 | if (strcmp(vers, ss->name) == 0) | |
1322 | return ss; | |
1323 | } | |
1324 | ||
1325 | return NULL; | |
1326 | } | |
1327 | ||
1328 | int is_container_member(struct mdstat_ent *mdstat, char *container) | |
1329 | { | |
1330 | if (mdstat->metadata_version == NULL || | |
1331 | strncmp(mdstat->metadata_version, "external:", 9) != 0 || | |
1332 | !is_subarray(mdstat->metadata_version+9) || | |
1333 | strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 || | |
1334 | mdstat->metadata_version[10+strlen(container)] != '/') | |
1335 | return 0; | |
1336 | ||
1337 | return 1; | |
1338 | } | |
1339 | ||
1340 | int is_subarray_active(char *subarray, char *container) | |
1341 | { | |
1342 | struct mdstat_ent *mdstat = mdstat_read(0, 0); | |
1343 | struct mdstat_ent *ent; | |
1344 | ||
1345 | for (ent = mdstat; ent; ent = ent->next) | |
1346 | if (is_container_member(ent, container)) | |
1347 | if (strcmp(to_subarray(ent, container), subarray) == 0) | |
1348 | break; | |
1349 | ||
1350 | free_mdstat(mdstat); | |
1351 | ||
1352 | return ent != NULL; | |
1353 | } | |
1354 | ||
1355 | /* open_subarray - opens a subarray in a container | |
1356 | * @dev: container device name | |
1357 | * @st: empty supertype | |
1358 | * @quiet: block reporting errors flag | |
1359 | * | |
1360 | * On success returns an fd to a container and fills in *st | |
1361 | */ | |
1362 | int open_subarray(char *dev, char *subarray, struct supertype *st, int quiet) | |
1363 | { | |
1364 | struct mdinfo *mdi; | |
1365 | struct mdinfo *info; | |
1366 | int fd, err = 1; | |
1367 | ||
1368 | fd = open(dev, O_RDWR|O_EXCL); | |
1369 | if (fd < 0) { | |
1370 | if (!quiet) | |
1371 | fprintf(stderr, Name ": Couldn't open %s, aborting\n", | |
1372 | dev); | |
1373 | return -1; | |
1374 | } | |
1375 | ||
1376 | st->devnum = fd2devnum(fd); | |
1377 | if (st->devnum == NoMdDev) { | |
1378 | if (!quiet) | |
1379 | fprintf(stderr, | |
1380 | Name ": Failed to determine device number for %s\n", | |
1381 | dev); | |
1382 | goto close_fd; | |
1383 | } | |
1384 | ||
1385 | mdi = sysfs_read(fd, st->devnum, GET_VERSION|GET_LEVEL); | |
1386 | if (!mdi) { | |
1387 | if (!quiet) | |
1388 | fprintf(stderr, Name ": Failed to read sysfs for %s\n", | |
1389 | dev); | |
1390 | goto close_fd; | |
1391 | } | |
1392 | ||
1393 | if (mdi->array.level != UnSet) { | |
1394 | if (!quiet) | |
1395 | fprintf(stderr, Name ": %s is not a container\n", dev); | |
1396 | goto free_sysfs; | |
1397 | } | |
1398 | ||
1399 | st->ss = version_to_superswitch(mdi->text_version); | |
1400 | if (!st->ss) { | |
1401 | if (!quiet) | |
1402 | fprintf(stderr, | |
1403 | Name ": Operation not supported for %s metadata\n", | |
1404 | mdi->text_version); | |
1405 | goto free_sysfs; | |
1406 | } | |
1407 | ||
1408 | st->devname = devnum2devname(st->devnum); | |
1409 | if (!st->devname) { | |
1410 | if (!quiet) | |
1411 | fprintf(stderr, Name ": Failed to allocate device name\n"); | |
1412 | goto free_sysfs; | |
1413 | } | |
1414 | ||
1415 | if (!st->ss->load_container) { | |
1416 | if (!quiet) | |
1417 | fprintf(stderr, Name ": %s is not a container\n", dev); | |
1418 | goto free_name; | |
1419 | } | |
1420 | ||
1421 | if (st->ss->load_container(st, fd, NULL)) { | |
1422 | if (!quiet) | |
1423 | fprintf(stderr, Name ": Failed to load metadata for %s\n", | |
1424 | dev); | |
1425 | goto free_name; | |
1426 | } | |
1427 | ||
1428 | info = st->ss->container_content(st, subarray); | |
1429 | if (!info) { | |
1430 | if (!quiet) | |
1431 | fprintf(stderr, Name ": Failed to find subarray-%s in %s\n", | |
1432 | subarray, dev); | |
1433 | goto free_super; | |
1434 | } | |
1435 | free(info); | |
1436 | ||
1437 | err = 0; | |
1438 | ||
1439 | free_super: | |
1440 | if (err) | |
1441 | st->ss->free_super(st); | |
1442 | free_name: | |
1443 | if (err) | |
1444 | free(st->devname); | |
1445 | free_sysfs: | |
1446 | sysfs_free(mdi); | |
1447 | close_fd: | |
1448 | if (err) | |
1449 | close(fd); | |
1450 | ||
1451 | if (err) | |
1452 | return -1; | |
1453 | else | |
1454 | return fd; | |
1455 | } | |
1456 | ||
1457 | int add_disk(int mdfd, struct supertype *st, | |
1458 | struct mdinfo *sra, struct mdinfo *info) | |
1459 | { | |
1460 | /* Add a device to an array, in one of 2 ways. */ | |
1461 | int rv; | |
1462 | #ifndef MDASSEMBLE | |
1463 | if (st->ss->external) { | |
1464 | if (info->disk.state & (1<<MD_DISK_SYNC)) | |
1465 | info->recovery_start = MaxSector; | |
1466 | else | |
1467 | info->recovery_start = 0; | |
1468 | rv = sysfs_add_disk(sra, info, 0); | |
1469 | if (! rv) { | |
1470 | struct mdinfo *sd2; | |
1471 | for (sd2 = sra->devs; sd2; sd2=sd2->next) | |
1472 | if (sd2 == info) | |
1473 | break; | |
1474 | if (sd2 == NULL) { | |
1475 | sd2 = malloc(sizeof(*sd2)); | |
1476 | *sd2 = *info; | |
1477 | sd2->next = sra->devs; | |
1478 | sra->devs = sd2; | |
1479 | } | |
1480 | } | |
1481 | } else | |
1482 | #endif | |
1483 | rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk); | |
1484 | return rv; | |
1485 | } | |
1486 | ||
1487 | int remove_disk(int mdfd, struct supertype *st, | |
1488 | struct mdinfo *sra, struct mdinfo *info) | |
1489 | { | |
1490 | int rv; | |
1491 | /* Remove the disk given by 'info' from the array */ | |
1492 | #ifndef MDASSEMBLE | |
1493 | if (st->ss->external) | |
1494 | rv = sysfs_set_str(sra, info, "slot", "none"); | |
1495 | else | |
1496 | #endif | |
1497 | rv = ioctl(mdfd, HOT_REMOVE_DISK, makedev(info->disk.major, | |
1498 | info->disk.minor)); | |
1499 | return rv; | |
1500 | } | |
1501 | ||
1502 | int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info) | |
1503 | { | |
1504 | /* Initialise kernel's knowledge of array. | |
1505 | * This varies between externally managed arrays | |
1506 | * and older kernels | |
1507 | */ | |
1508 | int vers = md_get_version(mdfd); | |
1509 | int rv; | |
1510 | ||
1511 | #ifndef MDASSEMBLE | |
1512 | if (st->ss->external) | |
1513 | rv = sysfs_set_array(info, vers); | |
1514 | else | |
1515 | #endif | |
1516 | if ((vers % 100) >= 1) { /* can use different versions */ | |
1517 | mdu_array_info_t inf; | |
1518 | memset(&inf, 0, sizeof(inf)); | |
1519 | inf.major_version = info->array.major_version; | |
1520 | inf.minor_version = info->array.minor_version; | |
1521 | rv = ioctl(mdfd, SET_ARRAY_INFO, &inf); | |
1522 | } else | |
1523 | rv = ioctl(mdfd, SET_ARRAY_INFO, NULL); | |
1524 | return rv; | |
1525 | } | |
1526 | ||
1527 | unsigned long long min_recovery_start(struct mdinfo *array) | |
1528 | { | |
1529 | /* find the minimum recovery_start in an array for metadata | |
1530 | * formats that only record per-array recovery progress instead | |
1531 | * of per-device | |
1532 | */ | |
1533 | unsigned long long recovery_start = MaxSector; | |
1534 | struct mdinfo *d; | |
1535 | ||
1536 | for (d = array->devs; d; d = d->next) | |
1537 | recovery_start = min(recovery_start, d->recovery_start); | |
1538 | ||
1539 | return recovery_start; | |
1540 | } | |
1541 | ||
1542 | int mdmon_pid(int devnum) | |
1543 | { | |
1544 | char path[100]; | |
1545 | char pid[10]; | |
1546 | int fd; | |
1547 | int n; | |
1548 | char *devname = devnum2devname(devnum); | |
1549 | ||
1550 | sprintf(path, "%s/%s.pid", MDMON_DIR, devname); | |
1551 | free(devname); | |
1552 | ||
1553 | fd = open(path, O_RDONLY | O_NOATIME, 0); | |
1554 | ||
1555 | if (fd < 0) | |
1556 | return -1; | |
1557 | n = read(fd, pid, 9); | |
1558 | close(fd); | |
1559 | if (n <= 0) | |
1560 | return -1; | |
1561 | return atoi(pid); | |
1562 | } | |
1563 | ||
1564 | int mdmon_running(int devnum) | |
1565 | { | |
1566 | int pid = mdmon_pid(devnum); | |
1567 | if (pid <= 0) | |
1568 | return 0; | |
1569 | if (kill(pid, 0) == 0) | |
1570 | return 1; | |
1571 | return 0; | |
1572 | } | |
1573 | ||
1574 | int start_mdmon(int devnum) | |
1575 | { | |
1576 | int i, skipped; | |
1577 | int len; | |
1578 | pid_t pid; | |
1579 | int status; | |
1580 | char pathbuf[1024]; | |
1581 | char *paths[4] = { | |
1582 | pathbuf, | |
1583 | "/sbin/mdmon", | |
1584 | "mdmon", | |
1585 | NULL | |
1586 | }; | |
1587 | ||
1588 | if (check_env("MDADM_NO_MDMON")) | |
1589 | return 0; | |
1590 | ||
1591 | len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf)); | |
1592 | if (len > 0) { | |
1593 | char *sl; | |
1594 | pathbuf[len] = 0; | |
1595 | sl = strrchr(pathbuf, '/'); | |
1596 | if (sl) | |
1597 | sl++; | |
1598 | else | |
1599 | sl = pathbuf; | |
1600 | strcpy(sl, "mdmon"); | |
1601 | } else | |
1602 | pathbuf[0] = '\0'; | |
1603 | ||
1604 | switch(fork()) { | |
1605 | case 0: | |
1606 | /* FIXME yuk. CLOSE_EXEC?? */ | |
1607 | skipped = 0; | |
1608 | for (i=3; skipped < 20; i++) | |
1609 | if (close(i) < 0) | |
1610 | skipped++; | |
1611 | else | |
1612 | skipped = 0; | |
1613 | ||
1614 | for (i=0; paths[i]; i++) | |
1615 | if (paths[i][0]) | |
1616 | execl(paths[i], "mdmon", | |
1617 | devnum2devname(devnum), | |
1618 | NULL); | |
1619 | exit(1); | |
1620 | case -1: fprintf(stderr, Name ": cannot run mdmon. " | |
1621 | "Array remains readonly\n"); | |
1622 | return -1; | |
1623 | default: /* parent - good */ | |
1624 | pid = wait(&status); | |
1625 | if (pid < 0 || status != 0) | |
1626 | return -1; | |
1627 | } | |
1628 | return 0; | |
1629 | } | |
1630 | ||
1631 | int check_env(char *name) | |
1632 | { | |
1633 | char *val = getenv(name); | |
1634 | ||
1635 | if (val && atoi(val) == 1) | |
1636 | return 1; | |
1637 | ||
1638 | return 0; | |
1639 | } | |
1640 | ||
1641 | __u32 random32(void) | |
1642 | { | |
1643 | __u32 rv; | |
1644 | int rfd = open("/dev/urandom", O_RDONLY); | |
1645 | if (rfd < 0 || read(rfd, &rv, 4) != 4) | |
1646 | rv = random(); | |
1647 | if (rfd >= 0) | |
1648 | close(rfd); | |
1649 | return rv; | |
1650 | } | |
1651 | ||
1652 | #ifndef MDASSEMBLE | |
1653 | int flush_metadata_updates(struct supertype *st) | |
1654 | { | |
1655 | int sfd; | |
1656 | if (!st->updates) { | |
1657 | st->update_tail = NULL; | |
1658 | return -1; | |
1659 | } | |
1660 | ||
1661 | sfd = connect_monitor(devnum2devname(st->container_dev)); | |
1662 | if (sfd < 0) | |
1663 | return -1; | |
1664 | ||
1665 | while (st->updates) { | |
1666 | struct metadata_update *mu = st->updates; | |
1667 | st->updates = mu->next; | |
1668 | ||
1669 | send_message(sfd, mu, 0); | |
1670 | wait_reply(sfd, 0); | |
1671 | free(mu->buf); | |
1672 | free(mu); | |
1673 | } | |
1674 | ack(sfd, 0); | |
1675 | wait_reply(sfd, 0); | |
1676 | close(sfd); | |
1677 | st->update_tail = NULL; | |
1678 | return 0; | |
1679 | } | |
1680 | ||
1681 | void append_metadata_update(struct supertype *st, void *buf, int len) | |
1682 | { | |
1683 | ||
1684 | struct metadata_update *mu = malloc(sizeof(*mu)); | |
1685 | ||
1686 | mu->buf = buf; | |
1687 | mu->len = len; | |
1688 | mu->space = NULL; | |
1689 | mu->space_list = NULL; | |
1690 | mu->next = NULL; | |
1691 | *st->update_tail = mu; | |
1692 | st->update_tail = &mu->next; | |
1693 | } | |
1694 | #endif /* MDASSEMBLE */ | |
1695 | ||
1696 | #ifdef __TINYC__ | |
1697 | /* tinyc doesn't optimize this check in ioctl.h out ... */ | |
1698 | unsigned int __invalid_size_argument_for_IOC = 0; | |
1699 | #endif | |
1700 | ||
1701 | int experimental(void) | |
1702 | { | |
1703 | if (check_env("MDADM_EXPERIMENTAL")) | |
1704 | return 1; | |
1705 | else { | |
1706 | fprintf(stderr, Name ": To use this feature MDADM_EXPERIMENTAL enviroment variable has to defined.\n"); | |
1707 | return 0; | |
1708 | } | |
1709 | } | |
1710 | ||
1711 | /* Pick all spares matching given criteria from a container | |
1712 | * if min_size == 0 do not check size | |
1713 | * if domlist == NULL do not check domains | |
1714 | * if spare_group given add it to domains of each spare | |
1715 | * metadata allows to test domains using metadata of destination array */ | |
1716 | struct mdinfo *container_choose_spares(struct supertype *st, | |
1717 | unsigned long long min_size, | |
1718 | struct domainlist *domlist, | |
1719 | char *spare_group, | |
1720 | const char *metadata, int get_one) | |
1721 | { | |
1722 | struct mdinfo *d, **dp, *disks = NULL; | |
1723 | ||
1724 | /* get list of all disks in container */ | |
1725 | if (st->ss->getinfo_super_disks) | |
1726 | disks = st->ss->getinfo_super_disks(st); | |
1727 | ||
1728 | if (!disks) | |
1729 | return disks; | |
1730 | /* find spare devices on the list */ | |
1731 | dp = &disks->devs; | |
1732 | disks->array.spare_disks = 0; | |
1733 | while (*dp) { | |
1734 | int found = 0; | |
1735 | d = *dp; | |
1736 | if (d->disk.state == 0) { | |
1737 | /* check if size is acceptable */ | |
1738 | unsigned long long dev_size; | |
1739 | dev_t dev = makedev(d->disk.major,d->disk.minor); | |
1740 | ||
1741 | if (!min_size || | |
1742 | (dev_size_from_id(dev, &dev_size) && | |
1743 | dev_size >= min_size)) | |
1744 | found = 1; | |
1745 | /* check if domain matches */ | |
1746 | if (found && domlist) { | |
1747 | struct dev_policy *pol = devnum_policy(dev); | |
1748 | if (spare_group) | |
1749 | pol_add(&pol, pol_domain, | |
1750 | spare_group, NULL); | |
1751 | if (domain_test(domlist, pol, metadata) != 1) | |
1752 | found = 0; | |
1753 | dev_policy_free(pol); | |
1754 | } | |
1755 | } | |
1756 | if (found) { | |
1757 | dp = &d->next; | |
1758 | disks->array.spare_disks++; | |
1759 | if (get_one) { | |
1760 | sysfs_free(*dp); | |
1761 | d->next = NULL; | |
1762 | } | |
1763 | } else { | |
1764 | *dp = d->next; | |
1765 | d->next = NULL; | |
1766 | sysfs_free(d); | |
1767 | } | |
1768 | } | |
1769 | return disks; | |
1770 | } |