4ccb1bbf9f07e122a99f1c349ad546516dea9c0e
[thirdparty/mdadm.git] / util.c
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 /*
69 * Parse a 128 bit uuid in 4 integers
70 * format is 32 hexx nibbles with options :.<space> separator
71 * If not exactly 32 hex digits are found, return 0
72 * else return 1
73 */
74 int parse_uuid(char *str, int uuid[4])
75 {
76 int hit = 0; /* number of Hex digIT */
77 int i;
78 char c;
79 for (i=0; i<4; i++) uuid[i]=0;
80
81 while ((c= *str++)) {
82 int n;
83 if (c>='0' && c<='9')
84 n = c-'0';
85 else if (c>='a' && c <= 'f')
86 n = 10 + c - 'a';
87 else if (c>='A' && c <= 'F')
88 n = 10 + c - 'A';
89 else if (strchr(":. -", c))
90 continue;
91 else return 0;
92
93 if (hit<32) {
94 uuid[hit/8] <<= 4;
95 uuid[hit/8] += n;
96 }
97 hit++;
98 }
99 if (hit == 32)
100 return 1;
101 return 0;
102 }
103
104
105 /*
106 * Get the md version number.
107 * We use the RAID_VERSION ioctl if it is supported
108 * If not, but we have a block device with major '9', we assume
109 * 0.36.0
110 *
111 * Return version number as 24 but number - assume version parts
112 * always < 255
113 */
114
115 int md_get_version(int fd)
116 {
117 struct stat stb;
118 mdu_version_t vers;
119
120 if (fstat(fd, &stb)<0)
121 return -1;
122 if ((S_IFMT&stb.st_mode) != S_IFBLK)
123 return -1;
124
125 if (ioctl(fd, RAID_VERSION, &vers) == 0)
126 return (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
127 if (errno == EACCES)
128 return -1;
129 if (major(stb.st_rdev) == MD_MAJOR)
130 return (3600);
131 return -1;
132 }
133
134 int get_linux_version()
135 {
136 struct utsname name;
137 char *cp;
138 int a,b,c;
139 if (uname(&name) <0)
140 return -1;
141
142 cp = name.release;
143 a = strtoul(cp, &cp, 10);
144 if (*cp != '.') return -1;
145 b = strtoul(cp+1, &cp, 10);
146 if (*cp != '.') return -1;
147 c = strtoul(cp+1, NULL, 10);
148
149 return (a*1000000)+(b*1000)+c;
150 }
151
152 void remove_partitions(int fd)
153 {
154 /* remove partitions from this block devices.
155 * This is used for components added to an array
156 */
157 #ifdef BLKPG_DEL_PARTITION
158 struct blkpg_ioctl_arg a;
159 struct blkpg_partition p;
160
161 a.op = BLKPG_DEL_PARTITION;
162 a.data = (void*)&p;
163 a.datalen = sizeof(p);
164 a.flags = 0;
165 memset(a.data, 0, a.datalen);
166 for (p.pno=0; p.pno < 16; p.pno++)
167 ioctl(fd, BLKPG, &a);
168 #endif
169 }
170
171 int enough(int level, int raid_disks, int layout, int clean,
172 char *avail, int avail_disks)
173 {
174 int copies, first;
175 switch (level) {
176 case 10:
177 /* This is the tricky one - we need to check
178 * which actual disks are present.
179 */
180 copies = (layout&255)* ((layout>>8) & 255);
181 first=0;
182 do {
183 /* there must be one of the 'copies' form 'first' */
184 int n = copies;
185 int cnt=0;
186 while (n--) {
187 if (avail[first])
188 cnt++;
189 first = (first+1) % raid_disks;
190 }
191 if (cnt == 0)
192 return 0;
193
194 } while (first != 0);
195 return 1;
196
197 case -4:
198 return avail_disks>= 1;
199 case -1:
200 case 0:
201 return avail_disks == raid_disks;
202 case 1:
203 return avail_disks >= 1;
204 case 4:
205 case 5:
206 if (clean)
207 return avail_disks >= raid_disks-1;
208 else
209 return avail_disks >= raid_disks;
210 case 6:
211 if (clean)
212 return avail_disks >= raid_disks-2;
213 else
214 return avail_disks >= raid_disks;
215 default:
216 return 0;
217 }
218 }
219
220 const int uuid_match_any[4] = { ~0, ~0, ~0, ~0 };
221 int same_uuid(int a[4], int b[4], int swapuuid)
222 {
223 if (memcmp(a, uuid_match_any, sizeof(int[4])) == 0 ||
224 memcmp(b, uuid_match_any, sizeof(int[4])) == 0)
225 return 1;
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 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
273 {
274 int i, j;
275 int id;
276 char uuid[16];
277 char *c = buf;
278 strcpy(c, "UUID-");
279 c += strlen(c);
280 copy_uuid(uuid, info->uuid, st->ss->swapuuid);
281 for (i = 0; i < 4; i++) {
282 id = uuid[i];
283 if (i)
284 *c++ = sep;
285 for (j = 3; j >= 0; j--) {
286 sprintf(c,"%02x", (unsigned char) uuid[j+4*i]);
287 c+= 2;
288 }
289 }
290 return buf;
291 }
292
293 #ifndef MDASSEMBLE
294 int check_ext2(int fd, char *name)
295 {
296 /*
297 * Check for an ext2fs file system.
298 * Superblock is always 1K at 1K offset
299 *
300 * s_magic is le16 at 56 == 0xEF53
301 * report mtime - le32 at 44
302 * blocks - le32 at 4
303 * logblksize - le32 at 24
304 */
305 unsigned char sb[1024];
306 time_t mtime;
307 int size, bsize;
308 if (lseek(fd, 1024,0)!= 1024)
309 return 0;
310 if (read(fd, sb, 1024)!= 1024)
311 return 0;
312 if (sb[56] != 0x53 || sb[57] != 0xef)
313 return 0;
314
315 mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
316 bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
317 size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
318 fprintf(stderr, Name ": %s appears to contain an ext2fs file system\n",
319 name);
320 fprintf(stderr," size=%dK mtime=%s",
321 size*(1<<bsize), ctime(&mtime));
322 return 1;
323 }
324
325 int check_reiser(int fd, char *name)
326 {
327 /*
328 * superblock is at 64K
329 * size is 1024;
330 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
331 *
332 */
333 unsigned char sb[1024];
334 unsigned long size;
335 if (lseek(fd, 64*1024, 0) != 64*1024)
336 return 0;
337 if (read(fd, sb, 1024) != 1024)
338 return 0;
339 if (strncmp((char*)sb+52, "ReIsErFs",8)!=0 &&
340 strncmp((char*)sb+52, "ReIsEr2Fs",9)!=0)
341 return 0;
342 fprintf(stderr, Name ": %s appears to contain a reiserfs file system\n",name);
343 size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
344 fprintf(stderr, " size = %luK\n", size*4);
345
346 return 1;
347 }
348
349 int check_raid(int fd, char *name)
350 {
351 struct mdinfo info;
352 time_t crtime;
353 char *level;
354 struct supertype *st = guess_super(fd);
355
356 if (!st) return 0;
357 st->ss->load_super(st, fd, name);
358 /* Looks like a raid array .. */
359 fprintf(stderr, Name ": %s appears to be part of a raid array:\n",
360 name);
361 st->ss->getinfo_super(st, &info);
362 st->ss->free_super(st);
363 crtime = info.array.ctime;
364 level = map_num(pers, info.array.level);
365 if (!level) level = "-unknown-";
366 fprintf(stderr, " level=%s devices=%d ctime=%s",
367 level, info.array.raid_disks, ctime(&crtime));
368 return 1;
369 }
370
371 int ask(char *mesg)
372 {
373 char *add = "";
374 int i;
375 for (i=0; i<5; i++) {
376 char buf[100];
377 fprintf(stderr, "%s%s", mesg, add);
378 fflush(stderr);
379 if (fgets(buf, 100, stdin)==NULL)
380 return 0;
381 if (buf[0]=='y' || buf[0]=='Y')
382 return 1;
383 if (buf[0]=='n' || buf[0]=='N')
384 return 0;
385 add = "(y/n) ";
386 }
387 fprintf(stderr, Name ": assuming 'no'\n");
388 return 0;
389 }
390 #endif /* MDASSEMBLE */
391
392 char *map_num(mapping_t *map, int num)
393 {
394 while (map->name) {
395 if (map->num == num)
396 return map->name;
397 map++;
398 }
399 return NULL;
400 }
401
402 int map_name(mapping_t *map, char *name)
403 {
404 while (map->name) {
405 if (strcmp(map->name, name)==0)
406 return map->num;
407 map++;
408 }
409 return UnSet;
410 }
411
412
413 int is_standard(char *dev, int *nump)
414 {
415 /* tests if dev is a "standard" md dev name.
416 * i.e if the last component is "/dNN" or "/mdNN",
417 * where NN is a string of digits
418 * Returns 1 if a partitionable standard,
419 * -1 if non-partitonable,
420 * 0 if not a standard name.
421 */
422 char *d = strrchr(dev, '/');
423 int type=0;
424 int num;
425 if (!d)
426 return 0;
427 if (strncmp(d, "/d",2)==0)
428 d += 2, type=1; /* /dev/md/dN{pM} */
429 else if (strncmp(d, "/md_d", 5)==0)
430 d += 5, type=1; /* /dev/md_dN{pM} */
431 else if (strncmp(d, "/md", 3)==0)
432 d += 3, type=-1; /* /dev/mdN */
433 else if (d-dev > 3 && strncmp(d-2, "md/", 3)==0)
434 d += 1, type=-1; /* /dev/md/N */
435 else
436 return 0;
437 if (!*d)
438 return 0;
439 num = atoi(d);
440 while (isdigit(*d))
441 d++;
442 if (*d)
443 return 0;
444 if (nump) *nump = num;
445
446 return type;
447 }
448
449
450 /*
451 * convert a major/minor pair for a block device into a name in /dev, if possible.
452 * On the first call, walk /dev collecting name.
453 * Put them in a simple linked listfor now.
454 */
455 struct devmap {
456 int major, minor;
457 char *name;
458 struct devmap *next;
459 } *devlist = NULL;
460 int devlist_ready = 0;
461
462 int add_dev(const char *name, const struct stat *stb, int flag, struct FTW *s)
463 {
464 struct stat st;
465
466 if (S_ISLNK(stb->st_mode)) {
467 if (stat(name, &st) != 0)
468 return 0;
469 stb = &st;
470 }
471
472 if ((stb->st_mode&S_IFMT)== S_IFBLK) {
473 char *n = strdup(name);
474 struct devmap *dm = malloc(sizeof(*dm));
475 if (strncmp(n, "/dev/./", 7)==0)
476 strcpy(n+4, name+6);
477 if (dm) {
478 dm->major = major(stb->st_rdev);
479 dm->minor = minor(stb->st_rdev);
480 dm->name = n;
481 dm->next = devlist;
482 devlist = dm;
483 }
484 }
485 return 0;
486 }
487
488 #ifndef HAVE_NFTW
489 #ifdef HAVE_FTW
490 int add_dev_1(const char *name, const struct stat *stb, int flag)
491 {
492 return add_dev(name, stb, flag, NULL);
493 }
494 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
495 {
496 return ftw(path, add_dev_1, nopenfd);
497 }
498 #else
499 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
500 {
501 return 0;
502 }
503 #endif /* HAVE_FTW */
504 #endif /* HAVE_NFTW */
505
506 /*
507 * Find a block device with the right major/minor number.
508 * If we find multiple names, choose the shortest.
509 * If we find a name in /dev/md/, we prefer that.
510 * This applies only to names for MD devices.
511 */
512 char *map_dev(int major, int minor, int create)
513 {
514 struct devmap *p;
515 char *regular = NULL, *preferred=NULL;
516 int did_check = 0;
517
518 if (major == 0 && minor == 0)
519 return NULL;
520
521 retry:
522 if (!devlist_ready) {
523 char *dev = "/dev";
524 struct stat stb;
525 while(devlist) {
526 struct devmap *d = devlist;
527 devlist = d->next;
528 free(d->name);
529 free(d);
530 }
531 if (lstat(dev, &stb)==0 &&
532 S_ISLNK(stb.st_mode))
533 dev = "/dev/.";
534 nftw(dev, add_dev, 10, FTW_PHYS);
535 devlist_ready=1;
536 did_check = 1;
537 }
538
539 for (p=devlist; p; p=p->next)
540 if (p->major == major &&
541 p->minor == minor) {
542 if (strncmp(p->name, "/dev/md/",8) == 0) {
543 if (preferred == NULL ||
544 strlen(p->name) < strlen(preferred))
545 preferred = p->name;
546 } else {
547 if (regular == NULL ||
548 strlen(p->name) < strlen(regular))
549 regular = p->name;
550 }
551 }
552 if (!regular && !preferred && !did_check) {
553 devlist_ready = 0;
554 goto retry;
555 }
556 if (create && !regular && !preferred) {
557 static char buf[30];
558 snprintf(buf, sizeof(buf), "%d:%d", major, minor);
559 regular = buf;
560 }
561
562 return preferred ? preferred : regular;
563 }
564
565 unsigned long calc_csum(void *super, int bytes)
566 {
567 unsigned long long newcsum = 0;
568 int i;
569 unsigned int csum;
570 unsigned int *superc = (unsigned int*) super;
571
572 for(i=0; i<bytes/4; i++)
573 newcsum+= superc[i];
574 csum = (newcsum& 0xffffffff) + (newcsum>>32);
575 #ifdef __alpha__
576 /* The in-kernel checksum calculation is always 16bit on
577 * the alpha, though it is 32 bit on i386...
578 * I wonder what it is elsewhere... (it uses and API in
579 * a way that it shouldn't).
580 */
581 csum = (csum & 0xffff) + (csum >> 16);
582 csum = (csum & 0xffff) + (csum >> 16);
583 #endif
584 return csum;
585 }
586
587 #ifndef MDASSEMBLE
588 char *human_size(long long bytes)
589 {
590 static char buf[30];
591
592 /* We convert bytes to either centi-M{ega,ibi}bytes or
593 * centi-G{igi,ibi}bytes, with appropriate rounding,
594 * and then print 1/100th of those as a decimal.
595 * We allow upto 2048Megabytes before converting to
596 * gigabytes, as that shows more precision and isn't
597 * too large a number.
598 * Terrabytes are not yet handled.
599 */
600
601 if (bytes < 5000*1024)
602 buf[0]=0;
603 else if (bytes < 2*1024LL*1024LL*1024LL) {
604 long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2;
605 long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2;
606 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
607 cMiB/100 , cMiB % 100,
608 cMB/100, cMB % 100);
609 } else {
610 long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2;
611 long cGB = (bytes / (1000000000LL/200LL ) +1) /2;
612 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
613 cGiB/100 , cGiB % 100,
614 cGB/100, cGB % 100);
615 }
616 return buf;
617 }
618
619 char *human_size_brief(long long bytes)
620 {
621 static char buf[30];
622
623 if (bytes < 5000*1024)
624 snprintf(buf, sizeof(buf), "%ld.%02ldKiB",
625 (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024)
626 );
627 else if (bytes < 2*1024LL*1024LL*1024LL)
628 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
629 (long)(bytes>>20),
630 (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100)
631 );
632 else
633 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
634 (long)(bytes>>30),
635 (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100)
636 );
637 return buf;
638 }
639
640 void print_r10_layout(int layout)
641 {
642 int near = layout & 255;
643 int far = (layout >> 8) & 255;
644 int offset = (layout&0x10000);
645 char *sep = "";
646
647 if (near != 1) {
648 printf("%s near=%d", sep, near);
649 sep = ",";
650 }
651 if (far != 1)
652 printf("%s %s=%d", sep, offset?"offset":"far", far);
653 if (near*far == 1)
654 printf("NO REDUNDANCY");
655 }
656 #endif
657
658 unsigned long long calc_array_size(int level, int raid_disks, int layout,
659 int chunksize, unsigned long long devsize)
660 {
661 int data_disks = 0;
662 switch (level) {
663 case 0: data_disks = raid_disks; break;
664 case 1: data_disks = 1; break;
665 case 4:
666 case 5: data_disks = raid_disks - 1; break;
667 case 6: data_disks = raid_disks - 2; break;
668 case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
669 break;
670 }
671 devsize &= ~(unsigned long long)((chunksize>>9)-1);
672 return data_disks * devsize;
673 }
674
675 int get_mdp_major(void)
676 {
677 static int mdp_major = -1;
678 FILE *fl;
679 char *w;
680 int have_block = 0;
681 int have_devices = 0;
682 int last_num = -1;
683
684 if (mdp_major != -1)
685 return mdp_major;
686 fl = fopen("/proc/devices", "r");
687 if (!fl)
688 return -1;
689 while ((w = conf_word(fl, 1))) {
690 if (have_block && strcmp(w, "devices:")==0)
691 have_devices = 1;
692 have_block = (strcmp(w, "Block")==0);
693 if (isdigit(w[0]))
694 last_num = atoi(w);
695 if (have_devices && strcmp(w, "mdp")==0)
696 mdp_major = last_num;
697 free(w);
698 }
699 fclose(fl);
700 return mdp_major;
701 }
702
703 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
704 char *get_md_name(int dev)
705 {
706 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
707 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
708 static char devname[50];
709 struct stat stb;
710 dev_t rdev;
711 char *dn;
712
713 if (dev < 0) {
714 int mdp = get_mdp_major();
715 if (mdp < 0) return NULL;
716 rdev = makedev(mdp, (-1-dev)<<6);
717 snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev);
718 if (stat(devname, &stb) == 0
719 && (S_IFMT&stb.st_mode) == S_IFBLK
720 && (stb.st_rdev == rdev))
721 return devname;
722 } else {
723 rdev = makedev(MD_MAJOR, dev);
724 snprintf(devname, sizeof(devname), "/dev/md%d", dev);
725 if (stat(devname, &stb) == 0
726 && (S_IFMT&stb.st_mode) == S_IFBLK
727 && (stb.st_rdev == rdev))
728 return devname;
729
730 snprintf(devname, sizeof(devname), "/dev/md/%d", dev);
731 if (stat(devname, &stb) == 0
732 && (S_IFMT&stb.st_mode) == S_IFBLK
733 && (stb.st_rdev == rdev))
734 return devname;
735 }
736 dn = map_dev(major(rdev), minor(rdev), 0);
737 if (dn)
738 return dn;
739 snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev);
740 if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
741 if (errno != EEXIST)
742 return NULL;
743
744 if (stat(devname, &stb) == 0
745 && (S_IFMT&stb.st_mode) == S_IFBLK
746 && (stb.st_rdev == rdev))
747 return devname;
748 unlink(devname);
749 return NULL;
750 }
751
752 void put_md_name(char *name)
753 {
754 if (strncmp(name, "/dev/.tmp.md", 12)==0)
755 unlink(name);
756 }
757
758 int find_free_devnum(int use_partitions)
759 {
760 int devnum;
761 for (devnum = 127; devnum != 128;
762 devnum = devnum ? devnum-1 : (1<<20)-1) {
763 char *dn;
764 int _devnum;
765
766 _devnum = use_partitions ? (-1-devnum) : devnum;
767 if (mddev_busy(_devnum))
768 continue;
769 /* make sure it is new to /dev too, at least as a
770 * non-standard */
771 dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0);
772 if (dn && ! is_standard(dn, NULL))
773 continue;
774 break;
775 }
776 if (devnum == 128)
777 return NoMdDev;
778 return use_partitions ? (-1-devnum) : devnum;
779 }
780 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
781
782 int dev_open(char *dev, int flags)
783 {
784 /* like 'open', but if 'dev' matches %d:%d, create a temp
785 * block device and open that
786 */
787 char *e;
788 int fd = -1;
789 char devname[32];
790 int major;
791 int minor;
792
793 if (!dev) return -1;
794
795 major = strtoul(dev, &e, 0);
796 if (e > dev && *e == ':' && e[1] &&
797 (minor = strtoul(e+1, &e, 0)) >= 0 &&
798 *e == 0) {
799 snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d",
800 (int)getpid(), major, minor);
801 if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) {
802 fd = open(devname, flags|O_DIRECT);
803 unlink(devname);
804 }
805 } else
806 fd = open(dev, flags|O_DIRECT);
807 return fd;
808 }
809
810 int open_dev(int devnum)
811 {
812 char buf[20];
813
814 sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
815 return dev_open(buf, O_RDWR);
816 }
817
818 int open_dev_excl(int devnum)
819 {
820 char buf[20];
821 int i;
822
823 sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
824 for (i=0 ; i<25 ; i++) {
825 int fd = dev_open(buf, O_RDWR|O_EXCL);
826 if (fd >= 0)
827 return fd;
828 if (errno != EBUSY)
829 return fd;
830 usleep(200000);
831 }
832 return -1;
833 }
834
835 int same_dev(char *one, char *two)
836 {
837 struct stat st1, st2;
838 if (stat(one, &st1) != 0)
839 return 0;
840 if (stat(two, &st2) != 0)
841 return 0;
842 if ((st1.st_mode & S_IFMT) != S_IFBLK)
843 return 0;
844 if ((st2.st_mode & S_IFMT) != S_IFBLK)
845 return 0;
846 return st1.st_rdev == st2.st_rdev;
847 }
848
849 void wait_for(char *dev, int fd)
850 {
851 int i;
852 struct stat stb_want;
853
854 if (fstat(fd, &stb_want) != 0 ||
855 (stb_want.st_mode & S_IFMT) != S_IFBLK)
856 return;
857
858 for (i=0 ; i<25 ; i++) {
859 struct stat stb;
860 if (stat(dev, &stb) == 0 &&
861 (stb.st_mode & S_IFMT) == S_IFBLK &&
862 (stb.st_rdev == stb_want.st_rdev))
863 return;
864 usleep(200000);
865 }
866 if (i == 25)
867 dprintf("%s: timeout waiting for %s\n", __func__, dev);
868 }
869
870 struct superswitch *superlist[] = { &super0, &super1, &super_ddf, &super_imsm, NULL };
871
872 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
873
874 struct supertype *super_by_fd(int fd)
875 {
876 mdu_array_info_t array;
877 int vers;
878 int minor;
879 struct supertype *st = NULL;
880 struct mdinfo *sra;
881 char *verstr;
882 char version[20];
883 int i;
884 char *subarray = NULL;
885
886 sra = sysfs_read(fd, 0, GET_VERSION);
887
888 if (sra) {
889 vers = sra->array.major_version;
890 minor = sra->array.minor_version;
891 verstr = sra->text_version;
892 } else {
893 if (ioctl(fd, GET_ARRAY_INFO, &array))
894 array.major_version = array.minor_version = 0;
895 vers = array.major_version;
896 minor = array.minor_version;
897 verstr = "";
898 }
899
900 if (vers != -1) {
901 sprintf(version, "%d.%d", vers, minor);
902 verstr = version;
903 }
904 if (minor == -2 && is_subarray(verstr)) {
905 char *dev = verstr+1;
906 subarray = strchr(dev, '/');
907 int devnum;
908 if (subarray)
909 *subarray++ = '\0';
910 devnum = devname2devnum(dev);
911 subarray = strdup(subarray);
912 if (sra)
913 sysfs_free(sra);
914 sra = sysfs_read(-1, devnum, GET_VERSION);
915 if (sra && sra->text_version[0])
916 verstr = sra->text_version;
917 else
918 verstr = "-no-metadata-";
919 }
920
921 for (i = 0; st == NULL && superlist[i] ; i++)
922 st = superlist[i]->match_metadata_desc(verstr);
923
924 if (sra)
925 sysfs_free(sra);
926 if (st) {
927 st->sb = NULL;
928 if (subarray) {
929 strncpy(st->subarray, subarray, 32);
930 st->subarray[31] = 0;
931 free(subarray);
932 } else
933 st->subarray[0] = 0;
934 }
935 return st;
936 }
937 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
938
939
940 struct supertype *dup_super(struct supertype *orig)
941 {
942 struct supertype *st;
943
944 if (!orig)
945 return orig;
946 st = malloc(sizeof(*st));
947 if (!st)
948 return st;
949 memset(st, 0, sizeof(*st));
950 st->ss = orig->ss;
951 st->max_devs = orig->max_devs;
952 st->minor_version = orig->minor_version;
953 strcpy(st->subarray, orig->subarray);
954 st->sb = NULL;
955 st->info = NULL;
956 return st;
957 }
958
959 struct supertype *guess_super(int fd)
960 {
961 /* try each load_super to find the best match,
962 * and return the best superswitch
963 */
964 struct superswitch *ss;
965 struct supertype *st;
966 unsigned long besttime = 0;
967 int bestsuper = -1;
968 int i;
969
970 st = malloc(sizeof(*st));
971 for (i=0 ; superlist[i]; i++) {
972 int rv;
973 ss = superlist[i];
974 memset(st, 0, sizeof(*st));
975 rv = ss->load_super(st, fd, NULL);
976 if (rv == 0) {
977 struct mdinfo info;
978 st->ss->getinfo_super(st, &info);
979 if (bestsuper == -1 ||
980 besttime < info.array.ctime) {
981 bestsuper = i;
982 besttime = info.array.ctime;
983 }
984 ss->free_super(st);
985 }
986 }
987 if (bestsuper != -1) {
988 int rv;
989 memset(st, 0, sizeof(*st));
990 rv = superlist[bestsuper]->load_super(st, fd, NULL);
991 if (rv == 0) {
992 superlist[bestsuper]->free_super(st);
993 return st;
994 }
995 }
996 free(st);
997 return NULL;
998 }
999
1000 /* Return size of device in bytes */
1001 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
1002 {
1003 unsigned long long ldsize;
1004 struct stat st;
1005
1006 if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
1007 ldsize = (unsigned long long)st.st_size;
1008 else
1009 #ifdef BLKGETSIZE64
1010 if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
1011 #endif
1012 {
1013 unsigned long dsize;
1014 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
1015 ldsize = dsize;
1016 ldsize <<= 9;
1017 } else {
1018 if (dname)
1019 fprintf(stderr, Name ": Cannot get size of %s: %s\b",
1020 dname, strerror(errno));
1021 return 0;
1022 }
1023 }
1024 *sizep = ldsize;
1025 return 1;
1026 }
1027
1028 void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk)
1029 {
1030 int d;
1031 ioctl(mdfd, GET_ARRAY_INFO, ainf);
1032 for (d = 0 ; d < ainf->raid_disks + ainf->nr_disks ; d++)
1033 if (ioctl(mdfd, GET_DISK_INFO, disk) == 0)
1034 return;
1035 }
1036
1037 int open_container(int fd)
1038 {
1039 /* 'fd' is a block device. Find out if it is in use
1040 * by a container, and return an open fd on that container.
1041 */
1042 char path[256];
1043 char *e;
1044 DIR *dir;
1045 struct dirent *de;
1046 int dfd, n;
1047 char buf[200];
1048 int major, minor;
1049 struct stat st;
1050
1051 if (fstat(fd, &st) != 0)
1052 return -1;
1053 sprintf(path, "/sys/dev/block/%d:%d/holders",
1054 (int)major(st.st_rdev), (int)minor(st.st_rdev));
1055 e = path + strlen(path);
1056
1057 dir = opendir(path);
1058 if (!dir)
1059 return -1;
1060 while ((de = readdir(dir))) {
1061 if (de->d_ino == 0)
1062 continue;
1063 if (de->d_name[0] == '.')
1064 continue;
1065 sprintf(e, "/%s/dev", de->d_name);
1066 dfd = open(path, O_RDONLY);
1067 if (dfd < 0)
1068 continue;
1069 n = read(dfd, buf, sizeof(buf));
1070 close(dfd);
1071 if (n <= 0 || n >= sizeof(buf))
1072 continue;
1073 buf[n] = 0;
1074 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
1075 continue;
1076 sprintf(buf, "%d:%d", major, minor);
1077 dfd = dev_open(buf, O_RDONLY);
1078 if (dfd >= 0) {
1079 closedir(dir);
1080 return dfd;
1081 }
1082 }
1083 closedir(dir);
1084 return -1;
1085 }
1086
1087 int add_disk(int mdfd, struct supertype *st,
1088 struct mdinfo *sra, struct mdinfo *info)
1089 {
1090 /* Add a device to an array, in one of 2 ways. */
1091 int rv;
1092 #ifndef MDASSEMBLE
1093 if (st->ss->external) {
1094 rv = sysfs_add_disk(sra, info,
1095 info->disk.state & (1<<MD_DISK_SYNC));
1096 if (! rv) {
1097 struct mdinfo *sd2;
1098 for (sd2 = sra->devs; sd2; sd2=sd2->next)
1099 if (sd2 == info)
1100 break;
1101 if (sd2 == NULL) {
1102 sd2 = malloc(sizeof(*sd2));
1103 *sd2 = *info;
1104 sd2->next = sra->devs;
1105 sra->devs = sd2;
1106 }
1107 }
1108 } else
1109 #endif
1110 rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk);
1111 return rv;
1112 }
1113
1114 int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
1115 {
1116 /* Initialise kernel's knowledge of array.
1117 * This varies between externally managed arrays
1118 * and older kernels
1119 */
1120 int vers = md_get_version(mdfd);
1121 int rv;
1122
1123 #ifndef MDASSEMBLE
1124 if (st->ss->external)
1125 rv = sysfs_set_array(info, vers);
1126 else
1127 #endif
1128 if ((vers % 100) >= 1) { /* can use different versions */
1129 mdu_array_info_t inf;
1130 memset(&inf, 0, sizeof(inf));
1131 inf.major_version = info->array.major_version;
1132 inf.minor_version = info->array.minor_version;
1133 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
1134 } else
1135 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
1136 return rv;
1137 }
1138
1139 char *devnum2devname(int num)
1140 {
1141 char name[100];
1142 if (num > 0)
1143 sprintf(name, "md%d", num);
1144 else
1145 sprintf(name, "md_d%d", -1-num);
1146 return strdup(name);
1147 }
1148
1149 int devname2devnum(char *name)
1150 {
1151 char *ep;
1152 int num;
1153 if (strncmp(name, "md_d", 4)==0)
1154 num = -1-strtoul(name+4, &ep, 10);
1155 else
1156 num = strtoul(name+2, &ep, 10);
1157 return num;
1158 }
1159
1160 int stat2devnum(struct stat *st)
1161 {
1162 char path[30];
1163 char link[200];
1164 char *cp;
1165 int n;
1166
1167 if ((S_IFMT & st->st_mode) == S_IFBLK) {
1168 if (major(st->st_rdev) == MD_MAJOR)
1169 return minor(st->st_rdev);
1170 else if (major(st->st_rdev) == get_mdp_major())
1171 return -1- (minor(st->st_rdev)>>MdpMinorShift);
1172
1173 /* must be an extended-minor partition. Look at the
1174 * /sys/dev/block/%d:%d link which must look like
1175 * ../../block/mdXXX/mdXXXpYY
1176 */
1177 sprintf(path, "/sys/dev/block/%d:%d", major(st->st_rdev),
1178 minor(st->st_rdev));
1179 n = readlink(path, link, sizeof(link)-1);
1180 if (n <= 0)
1181 return NoMdDev;
1182 link[n] = 0;
1183 cp = strrchr(link, '/');
1184 if (cp) *cp = 0;
1185 cp = strchr(link, '/');
1186 if (cp && strncmp(cp, "/md", 3) == 0)
1187 return devname2devnum(cp+1);
1188 }
1189 return NoMdDev;
1190
1191 }
1192
1193 int fd2devnum(int fd)
1194 {
1195 struct stat stb;
1196 if (fstat(fd, &stb) == 0)
1197 return stat2devnum(&stb);
1198 return NoMdDev;
1199 }
1200
1201 int mdmon_running(int devnum)
1202 {
1203 char path[100];
1204 char pid[10];
1205 int fd;
1206 int n;
1207 sprintf(path, "/var/run/mdadm/%s.pid", devnum2devname(devnum));
1208 fd = open(path, O_RDONLY, 0);
1209
1210 if (fd < 0)
1211 return 0;
1212 n = read(fd, pid, 9);
1213 close(fd);
1214 if (n <= 0)
1215 return 0;
1216 if (kill(atoi(pid), 0) == 0)
1217 return 1;
1218 return 0;
1219 }
1220
1221 int signal_mdmon(int devnum)
1222 {
1223 char path[100];
1224 char pid[10];
1225 int fd;
1226 int n;
1227 sprintf(path, "/var/run/mdadm/%s.pid", devnum2devname(devnum));
1228 fd = open(path, O_RDONLY, 0);
1229
1230 if (fd < 0)
1231 return 0;
1232 n = read(fd, pid, 9);
1233 close(fd);
1234 if (n <= 0)
1235 return 0;
1236 if (kill(atoi(pid), SIGUSR1) == 0)
1237 return 1;
1238 return 0;
1239 }
1240
1241 int start_mdmon(int devnum)
1242 {
1243 int i;
1244 int len;
1245 pid_t pid;
1246 int status;
1247 char pathbuf[1024];
1248 char *paths[4] = {
1249 pathbuf,
1250 "/sbin/mdmon",
1251 "mdmon",
1252 NULL
1253 };
1254
1255 if (check_env("MDADM_NO_MDMON"))
1256 return 0;
1257
1258 len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf));
1259 if (len > 0) {
1260 char *sl;
1261 pathbuf[len] = 0;
1262 sl = strrchr(pathbuf, '/');
1263 if (sl)
1264 sl++;
1265 else
1266 sl = pathbuf;
1267 strcpy(sl, "mdmon");
1268 } else
1269 pathbuf[0] = '\0';
1270
1271 switch(fork()) {
1272 case 0:
1273 /* FIXME yuk. CLOSE_EXEC?? */
1274 for (i=3; i < 100; i++)
1275 close(i);
1276 for (i=0; paths[i]; i++)
1277 if (paths[i][0])
1278 execl(paths[i], "mdmon",
1279 devnum2devname(devnum),
1280 NULL);
1281 exit(1);
1282 case -1: fprintf(stderr, Name ": cannot run mdmon. "
1283 "Array remains readonly\n");
1284 return -1;
1285 default: /* parent - good */
1286 pid = wait(&status);
1287 if (pid < 0 || status != 0)
1288 return -1;
1289 }
1290 return 0;
1291 }
1292
1293 int check_env(char *name)
1294 {
1295 char *val = getenv(name);
1296
1297 if (val && atoi(val) == 1)
1298 return 1;
1299
1300 return 0;
1301 }
1302
1303 __u32 random32(void)
1304 {
1305 __u32 rv;
1306 int rfd = open("/dev/urandom", O_RDONLY);
1307 if (rfd < 0 || read(rfd, &rv, 4) != 4)
1308 rv = random();
1309 if (rfd >= 0)
1310 close(rfd);
1311 return rv;
1312 }
1313
1314 #ifndef MDASSEMBLE
1315 int flush_metadata_updates(struct supertype *st)
1316 {
1317 int sfd;
1318 if (!st->updates) {
1319 st->update_tail = NULL;
1320 return -1;
1321 }
1322
1323 sfd = connect_monitor(devnum2devname(st->container_dev));
1324 if (sfd < 0)
1325 return -1;
1326
1327 while (st->updates) {
1328 struct metadata_update *mu = st->updates;
1329 st->updates = mu->next;
1330
1331 send_message(sfd, mu, 0);
1332 wait_reply(sfd, 0);
1333 free(mu->buf);
1334 free(mu);
1335 }
1336 ack(sfd, 0);
1337 wait_reply(sfd, 0);
1338 close(sfd);
1339 st->update_tail = NULL;
1340 return 0;
1341 }
1342
1343 void append_metadata_update(struct supertype *st, void *buf, int len)
1344 {
1345
1346 struct metadata_update *mu = malloc(sizeof(*mu));
1347
1348 mu->buf = buf;
1349 mu->len = len;
1350 mu->space = NULL;
1351 mu->next = NULL;
1352 *st->update_tail = mu;
1353 st->update_tail = &mu->next;
1354 }
1355 #endif /* MDASSEMBLE */
1356
1357 #ifdef __TINYC__
1358 /* tinyc doesn't optimize this check in ioctl.h out ... */
1359 unsigned int __invalid_size_argument_for_IOC = 0;
1360 #endif
1361