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