]> git.ipfire.org Git - thirdparty/mdadm.git/blob - util.c
Incremental: restore assembly for inactive containers, block active
[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 /* partition table structures so we can check metadata position
69 * against the end of the last partition.
70 * Only handle MBR ant GPT partition tables.
71 */
72 struct MBR_part_record {
73 __u8 bootable;
74 __u8 first_head;
75 __u8 first_sector;
76 __u8 first_cyl;
77 __u8 part_type;
78 __u8 last_head;
79 __u8 last_sector;
80 __u8 last_cyl;
81 __u32 first_sect_lba;
82 __u32 blocks_num;
83 };
84
85 struct MBR {
86 __u8 pad[446];
87 struct MBR_part_record parts[4];
88 __u16 magic;
89 } __attribute__((packed));
90
91 struct GPT_part_entry {
92 unsigned char type_guid[16];
93 unsigned char partition_guid[16];
94 __u64 starting_lba;
95 __u64 ending_lba;
96 unsigned char attr_bits[8];
97 unsigned char name[72];
98 } __attribute__((packed));
99
100 struct GPT {
101 __u64 magic;
102 __u32 revision;
103 __u32 header_size;
104 __u32 crc;
105 __u32 pad1;
106 __u64 current_lba;
107 __u64 backup_lba;
108 __u64 first_lba;
109 __u64 last_lba;
110 __u8 guid[16];
111 __u64 part_start;
112 __u32 part_cnt;
113 __u32 part_size;
114 __u32 part_crc;
115 __u8 pad2[420];
116 } __attribute__((packed));
117
118 /* Force a compilation error if condition is true */
119 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
120
121 /* Force a compilation error if condition is true, but also produce a
122 result (of value 0 and type size_t), so the expression can be used
123 e.g. in a structure initializer (or where-ever else comma expressions
124 aren't permitted). */
125 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
126
127
128 /* MBR/GPT magic numbers */
129 #define MBR_SIGNATURE_MAGIC __cpu_to_le16(0xAA55)
130 #define GPT_SIGNATURE_MAGIC __cpu_to_le64(0x5452415020494645ULL)
131
132 #define MBR_PARTITIONS 4
133 #define MBR_GPT_PARTITION_TYPE 0xEE
134
135 /*
136 * Parse a 128 bit uuid in 4 integers
137 * format is 32 hexx nibbles with options :.<space> separator
138 * If not exactly 32 hex digits are found, return 0
139 * else return 1
140 */
141 int parse_uuid(char *str, int uuid[4])
142 {
143 int hit = 0; /* number of Hex digIT */
144 int i;
145 char c;
146 for (i=0; i<4; i++) uuid[i]=0;
147
148 while ((c= *str++)) {
149 int n;
150 if (c>='0' && c<='9')
151 n = c-'0';
152 else if (c>='a' && c <= 'f')
153 n = 10 + c - 'a';
154 else if (c>='A' && c <= 'F')
155 n = 10 + c - 'A';
156 else if (strchr(":. -", c))
157 continue;
158 else return 0;
159
160 if (hit<32) {
161 uuid[hit/8] <<= 4;
162 uuid[hit/8] += n;
163 }
164 hit++;
165 }
166 if (hit == 32)
167 return 1;
168 return 0;
169 }
170
171
172 /*
173 * Get the md version number.
174 * We use the RAID_VERSION ioctl if it is supported
175 * If not, but we have a block device with major '9', we assume
176 * 0.36.0
177 *
178 * Return version number as 24 but number - assume version parts
179 * always < 255
180 */
181
182 int md_get_version(int fd)
183 {
184 struct stat stb;
185 mdu_version_t vers;
186
187 if (fstat(fd, &stb)<0)
188 return -1;
189 if ((S_IFMT&stb.st_mode) != S_IFBLK)
190 return -1;
191
192 if (ioctl(fd, RAID_VERSION, &vers) == 0)
193 return (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
194 if (errno == EACCES)
195 return -1;
196 if (major(stb.st_rdev) == MD_MAJOR)
197 return (3600);
198 return -1;
199 }
200
201 int get_linux_version()
202 {
203 struct utsname name;
204 char *cp;
205 int a,b,c;
206 if (uname(&name) <0)
207 return -1;
208
209 cp = name.release;
210 a = strtoul(cp, &cp, 10);
211 if (*cp != '.') return -1;
212 b = strtoul(cp+1, &cp, 10);
213 if (*cp != '.') return -1;
214 c = strtoul(cp+1, NULL, 10);
215
216 return (a*1000000)+(b*1000)+c;
217 }
218
219 #ifndef MDASSEMBLE
220 long long parse_size(char *size)
221 {
222 /* parse 'size' which should be a number optionally
223 * followed by 'K', 'M', or 'G'.
224 * Without a suffix, K is assumed.
225 * Number returned is in sectors (half-K)
226 */
227 char *c;
228 long long s = strtoll(size, &c, 10);
229 if (s > 0) {
230 switch (*c) {
231 case 'K':
232 c++;
233 default:
234 s *= 2;
235 break;
236 case 'M':
237 c++;
238 s *= 1024 * 2;
239 break;
240 case 'G':
241 c++;
242 s *= 1024 * 1024 * 2;
243 break;
244 }
245 }
246 if (*c)
247 s = 0;
248 return s;
249 }
250
251 int parse_layout_10(char *layout)
252 {
253 int copies, rv;
254 char *cp;
255 /* Parse the layout string for raid10 */
256 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
257 if ((layout[0] != 'n' && layout[0] != 'f' && layout[0] != 'o') ||
258 (copies = strtoul(layout+1, &cp, 10)) < 1 ||
259 copies > 200 ||
260 *cp)
261 return -1;
262 if (layout[0] == 'n')
263 rv = 256 + copies;
264 else if (layout[0] == 'o')
265 rv = 0x10000 + (copies<<8) + 1;
266 else
267 rv = 1 + (copies<<8);
268 return rv;
269 }
270
271 int parse_layout_faulty(char *layout)
272 {
273 /* Parse the layout string for 'faulty' */
274 int ln = strcspn(layout, "0123456789");
275 char *m = strdup(layout);
276 int mode;
277 m[ln] = 0;
278 mode = map_name(faultylayout, m);
279 if (mode == UnSet)
280 return -1;
281
282 return mode | (atoi(layout+ln)<< ModeShift);
283 }
284 #endif
285
286 void remove_partitions(int fd)
287 {
288 /* remove partitions from this block devices.
289 * This is used for components added to an array
290 */
291 #ifdef BLKPG_DEL_PARTITION
292 struct blkpg_ioctl_arg a;
293 struct blkpg_partition p;
294
295 a.op = BLKPG_DEL_PARTITION;
296 a.data = (void*)&p;
297 a.datalen = sizeof(p);
298 a.flags = 0;
299 memset(a.data, 0, a.datalen);
300 for (p.pno=0; p.pno < 16; p.pno++)
301 ioctl(fd, BLKPG, &a);
302 #endif
303 }
304
305 int test_partition(int fd)
306 {
307 /* Check if fd is a whole-disk or a partition.
308 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
309 * will return ENXIO on an invalid partition number.
310 */
311 struct blkpg_ioctl_arg a;
312 struct blkpg_partition p;
313 a.op = BLKPG_DEL_PARTITION;
314 a.data = (void*)&p;
315 a.datalen = sizeof(p);
316 a.flags = 0;
317 memset(a.data, 0, a.datalen);
318 p.pno = 1<<30;
319 if (ioctl(fd, BLKPG, &a) == 0)
320 /* Very unlikely, but not a partition */
321 return 0;
322 if (errno == ENXIO)
323 /* not a partition */
324 return 0;
325
326 return 1;
327 }
328
329
330 int enough(int level, int raid_disks, int layout, int clean,
331 char *avail, int avail_disks)
332 {
333 int copies, first;
334 switch (level) {
335 case 10:
336 /* This is the tricky one - we need to check
337 * which actual disks are present.
338 */
339 copies = (layout&255)* ((layout>>8) & 255);
340 first=0;
341 do {
342 /* there must be one of the 'copies' form 'first' */
343 int n = copies;
344 int cnt=0;
345 while (n--) {
346 if (avail[first])
347 cnt++;
348 first = (first+1) % raid_disks;
349 }
350 if (cnt == 0)
351 return 0;
352
353 } while (first != 0);
354 return 1;
355
356 case LEVEL_MULTIPATH:
357 return avail_disks>= 1;
358 case LEVEL_LINEAR:
359 case 0:
360 return avail_disks == raid_disks;
361 case 1:
362 return avail_disks >= 1;
363 case 4:
364 case 5:
365 if (clean)
366 return avail_disks >= raid_disks-1;
367 else
368 return avail_disks >= raid_disks;
369 case 6:
370 if (clean)
371 return avail_disks >= raid_disks-2;
372 else
373 return avail_disks >= raid_disks;
374 default:
375 return 0;
376 }
377 }
378
379 const int uuid_match_any[4] = { ~0, ~0, ~0, ~0 };
380 int same_uuid(int a[4], int b[4], int swapuuid)
381 {
382 if (memcmp(a, uuid_match_any, sizeof(int[4])) == 0 ||
383 memcmp(b, uuid_match_any, sizeof(int[4])) == 0)
384 return 1;
385
386 if (swapuuid) {
387 /* parse uuids are hostendian.
388 * uuid's from some superblocks are big-ending
389 * if there is a difference, we need to swap..
390 */
391 unsigned char *ac = (unsigned char *)a;
392 unsigned char *bc = (unsigned char *)b;
393 int i;
394 for (i=0; i<16; i+= 4) {
395 if (ac[i+0] != bc[i+3] ||
396 ac[i+1] != bc[i+2] ||
397 ac[i+2] != bc[i+1] ||
398 ac[i+3] != bc[i+0])
399 return 0;
400 }
401 return 1;
402 } else {
403 if (a[0]==b[0] &&
404 a[1]==b[1] &&
405 a[2]==b[2] &&
406 a[3]==b[3])
407 return 1;
408 return 0;
409 }
410 }
411 void copy_uuid(void *a, int b[4], int swapuuid)
412 {
413 if (swapuuid) {
414 /* parse uuids are hostendian.
415 * uuid's from some superblocks are big-ending
416 * if there is a difference, we need to swap..
417 */
418 unsigned char *ac = (unsigned char *)a;
419 unsigned char *bc = (unsigned char *)b;
420 int i;
421 for (i=0; i<16; i+= 4) {
422 ac[i+0] = bc[i+3];
423 ac[i+1] = bc[i+2];
424 ac[i+2] = bc[i+1];
425 ac[i+3] = bc[i+0];
426 }
427 } else
428 memcpy(a, b, 16);
429 }
430
431 char *__fname_from_uuid(int id[4], int swap, char *buf, char sep)
432 {
433 int i, j;
434 char uuid[16];
435 char *c = buf;
436 strcpy(c, "UUID-");
437 c += strlen(c);
438 copy_uuid(uuid, id, swap);
439 for (i = 0; i < 4; i++) {
440 if (i)
441 *c++ = sep;
442 for (j = 3; j >= 0; j--) {
443 sprintf(c,"%02x", (unsigned char) uuid[j+4*i]);
444 c+= 2;
445 }
446 }
447 return buf;
448
449 }
450
451 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
452 {
453 // dirty hack to work around an issue with super1 superblocks...
454 // super1 superblocks need swapuuid set in order for assembly to
455 // work, but can't have it set if we want this printout to match
456 // all the other uuid printouts in super1.c, so we force swapuuid
457 // to 1 to make our printout match the rest of super1
458 return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep);
459 }
460
461 #ifndef MDASSEMBLE
462 int check_ext2(int fd, char *name)
463 {
464 /*
465 * Check for an ext2fs file system.
466 * Superblock is always 1K at 1K offset
467 *
468 * s_magic is le16 at 56 == 0xEF53
469 * report mtime - le32 at 44
470 * blocks - le32 at 4
471 * logblksize - le32 at 24
472 */
473 unsigned char sb[1024];
474 time_t mtime;
475 int size, bsize;
476 if (lseek(fd, 1024,0)!= 1024)
477 return 0;
478 if (read(fd, sb, 1024)!= 1024)
479 return 0;
480 if (sb[56] != 0x53 || sb[57] != 0xef)
481 return 0;
482
483 mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
484 bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
485 size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
486 fprintf(stderr, Name ": %s appears to contain an ext2fs file system\n",
487 name);
488 fprintf(stderr," size=%dK mtime=%s",
489 size*(1<<bsize), ctime(&mtime));
490 return 1;
491 }
492
493 int check_reiser(int fd, char *name)
494 {
495 /*
496 * superblock is at 64K
497 * size is 1024;
498 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
499 *
500 */
501 unsigned char sb[1024];
502 unsigned long size;
503 if (lseek(fd, 64*1024, 0) != 64*1024)
504 return 0;
505 if (read(fd, sb, 1024) != 1024)
506 return 0;
507 if (strncmp((char*)sb+52, "ReIsErFs",8)!=0 &&
508 strncmp((char*)sb+52, "ReIsEr2Fs",9)!=0)
509 return 0;
510 fprintf(stderr, Name ": %s appears to contain a reiserfs file system\n",name);
511 size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
512 fprintf(stderr, " size = %luK\n", size*4);
513
514 return 1;
515 }
516
517 int check_raid(int fd, char *name)
518 {
519 struct mdinfo info;
520 time_t crtime;
521 char *level;
522 struct supertype *st = guess_super(fd);
523
524 if (!st) return 0;
525 st->ss->load_super(st, fd, name);
526 /* Looks like a raid array .. */
527 fprintf(stderr, Name ": %s appears to be part of a raid array:\n",
528 name);
529 st->ss->getinfo_super(st, &info);
530 st->ss->free_super(st);
531 crtime = info.array.ctime;
532 level = map_num(pers, info.array.level);
533 if (!level) level = "-unknown-";
534 fprintf(stderr, " level=%s devices=%d ctime=%s",
535 level, info.array.raid_disks, ctime(&crtime));
536 return 1;
537 }
538
539 int ask(char *mesg)
540 {
541 char *add = "";
542 int i;
543 for (i=0; i<5; i++) {
544 char buf[100];
545 fprintf(stderr, "%s%s", mesg, add);
546 fflush(stderr);
547 if (fgets(buf, 100, stdin)==NULL)
548 return 0;
549 if (buf[0]=='y' || buf[0]=='Y')
550 return 1;
551 if (buf[0]=='n' || buf[0]=='N')
552 return 0;
553 add = "(y/n) ";
554 }
555 fprintf(stderr, Name ": assuming 'no'\n");
556 return 0;
557 }
558 #endif /* MDASSEMBLE */
559
560 char *map_num(mapping_t *map, int num)
561 {
562 while (map->name) {
563 if (map->num == num)
564 return map->name;
565 map++;
566 }
567 return NULL;
568 }
569
570 int map_name(mapping_t *map, char *name)
571 {
572 while (map->name) {
573 if (strcmp(map->name, name)==0)
574 return map->num;
575 map++;
576 }
577 return UnSet;
578 }
579
580
581 int is_standard(char *dev, int *nump)
582 {
583 /* tests if dev is a "standard" md dev name.
584 * i.e if the last component is "/dNN" or "/mdNN",
585 * where NN is a string of digits
586 * Returns 1 if a partitionable standard,
587 * -1 if non-partitonable,
588 * 0 if not a standard name.
589 */
590 char *d = strrchr(dev, '/');
591 int type=0;
592 int num;
593 if (!d)
594 return 0;
595 if (strncmp(d, "/d",2)==0)
596 d += 2, type=1; /* /dev/md/dN{pM} */
597 else if (strncmp(d, "/md_d", 5)==0)
598 d += 5, type=1; /* /dev/md_dN{pM} */
599 else if (strncmp(d, "/md", 3)==0)
600 d += 3, type=-1; /* /dev/mdN */
601 else if (d-dev > 3 && strncmp(d-2, "md/", 3)==0)
602 d += 1, type=-1; /* /dev/md/N */
603 else
604 return 0;
605 if (!*d)
606 return 0;
607 num = atoi(d);
608 while (isdigit(*d))
609 d++;
610 if (*d)
611 return 0;
612 if (nump) *nump = num;
613
614 return type;
615 }
616
617
618 /*
619 * convert a major/minor pair for a block device into a name in /dev, if possible.
620 * On the first call, walk /dev collecting name.
621 * Put them in a simple linked listfor now.
622 */
623 struct devmap {
624 int major, minor;
625 char *name;
626 struct devmap *next;
627 } *devlist = NULL;
628 int devlist_ready = 0;
629
630 int add_dev(const char *name, const struct stat *stb, int flag, struct FTW *s)
631 {
632 struct stat st;
633
634 if (S_ISLNK(stb->st_mode)) {
635 if (stat(name, &st) != 0)
636 return 0;
637 stb = &st;
638 }
639
640 if ((stb->st_mode&S_IFMT)== S_IFBLK) {
641 char *n = strdup(name);
642 struct devmap *dm = malloc(sizeof(*dm));
643 if (strncmp(n, "/dev/./", 7)==0)
644 strcpy(n+4, name+6);
645 if (dm) {
646 dm->major = major(stb->st_rdev);
647 dm->minor = minor(stb->st_rdev);
648 dm->name = n;
649 dm->next = devlist;
650 devlist = dm;
651 }
652 }
653 return 0;
654 }
655
656 #ifndef HAVE_NFTW
657 #ifdef HAVE_FTW
658 int add_dev_1(const char *name, const struct stat *stb, int flag)
659 {
660 return add_dev(name, stb, flag, NULL);
661 }
662 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
663 {
664 return ftw(path, add_dev_1, nopenfd);
665 }
666 #else
667 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
668 {
669 return 0;
670 }
671 #endif /* HAVE_FTW */
672 #endif /* HAVE_NFTW */
673
674 /*
675 * Find a block device with the right major/minor number.
676 * If we find multiple names, choose the shortest.
677 * If we find a name in /dev/md/, we prefer that.
678 * This applies only to names for MD devices.
679 */
680 char *map_dev(int major, int minor, int create)
681 {
682 struct devmap *p;
683 char *regular = NULL, *preferred=NULL;
684 int did_check = 0;
685
686 if (major == 0 && minor == 0)
687 return NULL;
688
689 retry:
690 if (!devlist_ready) {
691 char *dev = "/dev";
692 struct stat stb;
693 while(devlist) {
694 struct devmap *d = devlist;
695 devlist = d->next;
696 free(d->name);
697 free(d);
698 }
699 if (lstat(dev, &stb)==0 &&
700 S_ISLNK(stb.st_mode))
701 dev = "/dev/.";
702 nftw(dev, add_dev, 10, FTW_PHYS);
703 devlist_ready=1;
704 did_check = 1;
705 }
706
707 for (p=devlist; p; p=p->next)
708 if (p->major == major &&
709 p->minor == minor) {
710 if (strncmp(p->name, "/dev/md/",8) == 0) {
711 if (preferred == NULL ||
712 strlen(p->name) < strlen(preferred))
713 preferred = p->name;
714 } else {
715 if (regular == NULL ||
716 strlen(p->name) < strlen(regular))
717 regular = p->name;
718 }
719 }
720 if (!regular && !preferred && !did_check) {
721 devlist_ready = 0;
722 goto retry;
723 }
724 if (create && !regular && !preferred) {
725 static char buf[30];
726 snprintf(buf, sizeof(buf), "%d:%d", major, minor);
727 regular = buf;
728 }
729
730 return preferred ? preferred : regular;
731 }
732
733 unsigned long calc_csum(void *super, int bytes)
734 {
735 unsigned long long newcsum = 0;
736 int i;
737 unsigned int csum;
738 unsigned int *superc = (unsigned int*) super;
739
740 for(i=0; i<bytes/4; i++)
741 newcsum+= superc[i];
742 csum = (newcsum& 0xffffffff) + (newcsum>>32);
743 #ifdef __alpha__
744 /* The in-kernel checksum calculation is always 16bit on
745 * the alpha, though it is 32 bit on i386...
746 * I wonder what it is elsewhere... (it uses and API in
747 * a way that it shouldn't).
748 */
749 csum = (csum & 0xffff) + (csum >> 16);
750 csum = (csum & 0xffff) + (csum >> 16);
751 #endif
752 return csum;
753 }
754
755 #ifndef MDASSEMBLE
756 char *human_size(long long bytes)
757 {
758 static char buf[30];
759
760 /* We convert bytes to either centi-M{ega,ibi}bytes or
761 * centi-G{igi,ibi}bytes, with appropriate rounding,
762 * and then print 1/100th of those as a decimal.
763 * We allow upto 2048Megabytes before converting to
764 * gigabytes, as that shows more precision and isn't
765 * too large a number.
766 * Terrabytes are not yet handled.
767 */
768
769 if (bytes < 5000*1024)
770 buf[0]=0;
771 else if (bytes < 2*1024LL*1024LL*1024LL) {
772 long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2;
773 long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2;
774 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
775 cMiB/100 , cMiB % 100,
776 cMB/100, cMB % 100);
777 } else {
778 long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2;
779 long cGB = (bytes / (1000000000LL/200LL ) +1) /2;
780 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
781 cGiB/100 , cGiB % 100,
782 cGB/100, cGB % 100);
783 }
784 return buf;
785 }
786
787 char *human_size_brief(long long bytes)
788 {
789 static char buf[30];
790
791 if (bytes < 5000*1024)
792 snprintf(buf, sizeof(buf), "%ld.%02ldKiB",
793 (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024)
794 );
795 else if (bytes < 2*1024LL*1024LL*1024LL)
796 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
797 (long)(bytes>>20),
798 (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100)
799 );
800 else
801 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
802 (long)(bytes>>30),
803 (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100)
804 );
805 return buf;
806 }
807
808 void print_r10_layout(int layout)
809 {
810 int near = layout & 255;
811 int far = (layout >> 8) & 255;
812 int offset = (layout&0x10000);
813 char *sep = "";
814
815 if (near != 1) {
816 printf("%s near=%d", sep, near);
817 sep = ",";
818 }
819 if (far != 1)
820 printf("%s %s=%d", sep, offset?"offset":"far", far);
821 if (near*far == 1)
822 printf("NO REDUNDANCY");
823 }
824 #endif
825
826 unsigned long long calc_array_size(int level, int raid_disks, int layout,
827 int chunksize, unsigned long long devsize)
828 {
829 int data_disks = 0;
830 switch (level) {
831 case 0: data_disks = raid_disks; break;
832 case 1: data_disks = 1; break;
833 case 4:
834 case 5: data_disks = raid_disks - 1; break;
835 case 6: data_disks = raid_disks - 2; break;
836 case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
837 break;
838 }
839 devsize &= ~(unsigned long long)((chunksize>>9)-1);
840 return data_disks * devsize;
841 }
842
843 int get_mdp_major(void)
844 {
845 static int mdp_major = -1;
846 FILE *fl;
847 char *w;
848 int have_block = 0;
849 int have_devices = 0;
850 int last_num = -1;
851
852 if (mdp_major != -1)
853 return mdp_major;
854 fl = fopen("/proc/devices", "r");
855 if (!fl)
856 return -1;
857 while ((w = conf_word(fl, 1))) {
858 if (have_block && strcmp(w, "devices:")==0)
859 have_devices = 1;
860 have_block = (strcmp(w, "Block")==0);
861 if (isdigit(w[0]))
862 last_num = atoi(w);
863 if (have_devices && strcmp(w, "mdp")==0)
864 mdp_major = last_num;
865 free(w);
866 }
867 fclose(fl);
868 return mdp_major;
869 }
870
871 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
872 char *get_md_name(int dev)
873 {
874 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
875 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
876 static char devname[50];
877 struct stat stb;
878 dev_t rdev;
879 char *dn;
880
881 if (dev < 0) {
882 int mdp = get_mdp_major();
883 if (mdp < 0) return NULL;
884 rdev = makedev(mdp, (-1-dev)<<6);
885 snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev);
886 if (stat(devname, &stb) == 0
887 && (S_IFMT&stb.st_mode) == S_IFBLK
888 && (stb.st_rdev == rdev))
889 return devname;
890 } else {
891 rdev = makedev(MD_MAJOR, dev);
892 snprintf(devname, sizeof(devname), "/dev/md%d", dev);
893 if (stat(devname, &stb) == 0
894 && (S_IFMT&stb.st_mode) == S_IFBLK
895 && (stb.st_rdev == rdev))
896 return devname;
897
898 snprintf(devname, sizeof(devname), "/dev/md/%d", dev);
899 if (stat(devname, &stb) == 0
900 && (S_IFMT&stb.st_mode) == S_IFBLK
901 && (stb.st_rdev == rdev))
902 return devname;
903 }
904 dn = map_dev(major(rdev), minor(rdev), 0);
905 if (dn)
906 return dn;
907 snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev);
908 if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
909 if (errno != EEXIST)
910 return NULL;
911
912 if (stat(devname, &stb) == 0
913 && (S_IFMT&stb.st_mode) == S_IFBLK
914 && (stb.st_rdev == rdev))
915 return devname;
916 unlink(devname);
917 return NULL;
918 }
919
920 void put_md_name(char *name)
921 {
922 if (strncmp(name, "/dev/.tmp.md", 12)==0)
923 unlink(name);
924 }
925
926 int find_free_devnum(int use_partitions)
927 {
928 int devnum;
929 for (devnum = 127; devnum != 128;
930 devnum = devnum ? devnum-1 : (1<<20)-1) {
931 char *dn;
932 int _devnum;
933
934 _devnum = use_partitions ? (-1-devnum) : devnum;
935 if (mddev_busy(_devnum))
936 continue;
937 /* make sure it is new to /dev too, at least as a
938 * non-standard */
939 dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0);
940 if (dn && ! is_standard(dn, NULL))
941 continue;
942 break;
943 }
944 if (devnum == 128)
945 return NoMdDev;
946 return use_partitions ? (-1-devnum) : devnum;
947 }
948 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
949
950 int dev_open(char *dev, int flags)
951 {
952 /* like 'open', but if 'dev' matches %d:%d, create a temp
953 * block device and open that
954 */
955 char *e;
956 int fd = -1;
957 char devname[32];
958 int major;
959 int minor;
960
961 if (!dev) return -1;
962
963 major = strtoul(dev, &e, 0);
964 if (e > dev && *e == ':' && e[1] &&
965 (minor = strtoul(e+1, &e, 0)) >= 0 &&
966 *e == 0) {
967 snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d",
968 (int)getpid(), major, minor);
969 if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) {
970 fd = open(devname, flags|O_DIRECT);
971 unlink(devname);
972 }
973 } else
974 fd = open(dev, flags|O_DIRECT);
975 return fd;
976 }
977
978 int open_dev(int devnum)
979 {
980 char buf[20];
981
982 sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
983 return dev_open(buf, O_RDWR);
984 }
985
986 int open_dev_excl(int devnum)
987 {
988 char buf[20];
989 int i;
990
991 sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
992 for (i=0 ; i<25 ; i++) {
993 int fd = dev_open(buf, O_RDWR|O_EXCL);
994 if (fd >= 0)
995 return fd;
996 if (errno != EBUSY)
997 return fd;
998 usleep(200000);
999 }
1000 return -1;
1001 }
1002
1003 int same_dev(char *one, char *two)
1004 {
1005 struct stat st1, st2;
1006 if (stat(one, &st1) != 0)
1007 return 0;
1008 if (stat(two, &st2) != 0)
1009 return 0;
1010 if ((st1.st_mode & S_IFMT) != S_IFBLK)
1011 return 0;
1012 if ((st2.st_mode & S_IFMT) != S_IFBLK)
1013 return 0;
1014 return st1.st_rdev == st2.st_rdev;
1015 }
1016
1017 void wait_for(char *dev, int fd)
1018 {
1019 int i;
1020 struct stat stb_want;
1021
1022 if (fstat(fd, &stb_want) != 0 ||
1023 (stb_want.st_mode & S_IFMT) != S_IFBLK)
1024 return;
1025
1026 for (i=0 ; i<25 ; i++) {
1027 struct stat stb;
1028 if (stat(dev, &stb) == 0 &&
1029 (stb.st_mode & S_IFMT) == S_IFBLK &&
1030 (stb.st_rdev == stb_want.st_rdev))
1031 return;
1032 usleep(200000);
1033 }
1034 if (i == 25)
1035 dprintf("%s: timeout waiting for %s\n", __func__, dev);
1036 }
1037
1038 struct superswitch *superlist[] = { &super0, &super1, &super_ddf, &super_imsm, NULL };
1039
1040 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1041
1042 struct supertype *super_by_fd(int fd)
1043 {
1044 mdu_array_info_t array;
1045 int vers;
1046 int minor;
1047 struct supertype *st = NULL;
1048 struct mdinfo *sra;
1049 char *verstr;
1050 char version[20];
1051 int i;
1052 char *subarray = NULL;
1053
1054 sra = sysfs_read(fd, 0, GET_VERSION);
1055
1056 if (sra) {
1057 vers = sra->array.major_version;
1058 minor = sra->array.minor_version;
1059 verstr = sra->text_version;
1060 } else {
1061 if (ioctl(fd, GET_ARRAY_INFO, &array))
1062 array.major_version = array.minor_version = 0;
1063 vers = array.major_version;
1064 minor = array.minor_version;
1065 verstr = "";
1066 }
1067
1068 if (vers != -1) {
1069 sprintf(version, "%d.%d", vers, minor);
1070 verstr = version;
1071 }
1072 if (minor == -2 && is_subarray(verstr)) {
1073 char *dev = verstr+1;
1074 subarray = strchr(dev, '/');
1075 int devnum;
1076 if (subarray)
1077 *subarray++ = '\0';
1078 devnum = devname2devnum(dev);
1079 subarray = strdup(subarray);
1080 if (sra)
1081 sysfs_free(sra);
1082 sra = sysfs_read(-1, devnum, GET_VERSION);
1083 if (sra && sra->text_version[0])
1084 verstr = sra->text_version;
1085 else
1086 verstr = "-no-metadata-";
1087 }
1088
1089 for (i = 0; st == NULL && superlist[i] ; i++)
1090 st = superlist[i]->match_metadata_desc(verstr);
1091
1092 if (sra)
1093 sysfs_free(sra);
1094 if (st) {
1095 st->sb = NULL;
1096 if (subarray) {
1097 strncpy(st->subarray, subarray, 32);
1098 st->subarray[31] = 0;
1099 free(subarray);
1100 } else
1101 st->subarray[0] = 0;
1102 }
1103 return st;
1104 }
1105 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1106
1107
1108 struct supertype *dup_super(struct supertype *orig)
1109 {
1110 struct supertype *st;
1111
1112 if (!orig)
1113 return orig;
1114 st = malloc(sizeof(*st));
1115 if (!st)
1116 return st;
1117 memset(st, 0, sizeof(*st));
1118 st->ss = orig->ss;
1119 st->max_devs = orig->max_devs;
1120 st->minor_version = orig->minor_version;
1121 strcpy(st->subarray, orig->subarray);
1122 st->sb = NULL;
1123 st->info = NULL;
1124 return st;
1125 }
1126
1127 struct supertype *guess_super(int fd)
1128 {
1129 /* try each load_super to find the best match,
1130 * and return the best superswitch
1131 */
1132 struct superswitch *ss;
1133 struct supertype *st;
1134 unsigned long besttime = 0;
1135 int bestsuper = -1;
1136 int i;
1137
1138 st = malloc(sizeof(*st));
1139 for (i=0 ; superlist[i]; i++) {
1140 int rv;
1141 ss = superlist[i];
1142 memset(st, 0, sizeof(*st));
1143 rv = ss->load_super(st, fd, NULL);
1144 if (rv == 0) {
1145 struct mdinfo info;
1146 st->ss->getinfo_super(st, &info);
1147 if (bestsuper == -1 ||
1148 besttime < info.array.ctime) {
1149 bestsuper = i;
1150 besttime = info.array.ctime;
1151 }
1152 ss->free_super(st);
1153 }
1154 }
1155 if (bestsuper != -1) {
1156 int rv;
1157 memset(st, 0, sizeof(*st));
1158 rv = superlist[bestsuper]->load_super(st, fd, NULL);
1159 if (rv == 0) {
1160 superlist[bestsuper]->free_super(st);
1161 return st;
1162 }
1163 }
1164 free(st);
1165 return NULL;
1166 }
1167
1168 /* Return size of device in bytes */
1169 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
1170 {
1171 unsigned long long ldsize;
1172 struct stat st;
1173
1174 if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
1175 ldsize = (unsigned long long)st.st_size;
1176 else
1177 #ifdef BLKGETSIZE64
1178 if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
1179 #endif
1180 {
1181 unsigned long dsize;
1182 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
1183 ldsize = dsize;
1184 ldsize <<= 9;
1185 } else {
1186 if (dname)
1187 fprintf(stderr, Name ": Cannot get size of %s: %s\b",
1188 dname, strerror(errno));
1189 return 0;
1190 }
1191 }
1192 *sizep = ldsize;
1193 return 1;
1194 }
1195
1196
1197 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1198 * Returns: 1 if successful
1199 * -1 for unknown partition type
1200 * 0 for other errors
1201 */
1202 static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart)
1203 {
1204 struct GPT gpt;
1205 unsigned char buf[512];
1206 unsigned char empty_gpt_entry[16]= {0};
1207 struct GPT_part_entry *part;
1208 unsigned long long curr_part_end;
1209 unsigned all_partitions, entry_size;
1210 int part_nr;
1211
1212 *endofpart = 0;
1213
1214 BUILD_BUG_ON(sizeof(gpt) != 512);
1215 /* read GPT header */
1216 lseek(fd, 512, SEEK_SET);
1217 if (read(fd, &gpt, 512) != 512)
1218 return 0;
1219
1220 /* get the number of partition entries and the entry size */
1221 all_partitions = __le32_to_cpu(gpt.part_cnt);
1222 entry_size = __le32_to_cpu(gpt.part_size);
1223
1224 /* Check GPT signature*/
1225 if (gpt.magic != GPT_SIGNATURE_MAGIC)
1226 return -1;
1227
1228 /* sanity checks */
1229 if (all_partitions > 1024 ||
1230 entry_size > 512)
1231 return -1;
1232
1233 /* read first GPT partition entries */
1234 if (read(fd, buf, 512) != 512)
1235 return 0;
1236
1237 part = (struct GPT_part_entry*)buf;
1238
1239 for (part_nr=0; part_nr < all_partitions; part_nr++) {
1240 /* is this valid partition? */
1241 if (memcmp(part->type_guid, empty_gpt_entry, 16) != 0) {
1242 /* check the last lba for the current partition */
1243 curr_part_end = __le64_to_cpu(part->ending_lba);
1244 if (curr_part_end > *endofpart)
1245 *endofpart = curr_part_end;
1246 }
1247
1248 part = (struct GPT_part_entry*)((unsigned char*)part + entry_size);
1249
1250 if ((unsigned char *)part >= buf + 512) {
1251 if (read(fd, buf, 512) != 512)
1252 return 0;
1253 part = (struct GPT_part_entry*)buf;
1254 }
1255 }
1256 return 1;
1257 }
1258
1259 /* Sets endofpart parameter to the last block used by the last partition on the device.
1260 * Returns: 1 if successful
1261 * -1 for unknown partition type
1262 * 0 for other errors
1263 */
1264 static int get_last_partition_end(int fd, unsigned long long *endofpart)
1265 {
1266 struct MBR boot_sect;
1267 struct MBR_part_record *part;
1268 unsigned long long curr_part_end;
1269 int part_nr;
1270 int retval = 0;
1271
1272 *endofpart = 0;
1273
1274 BUILD_BUG_ON(sizeof(boot_sect) != 512);
1275 /* read MBR */
1276 lseek(fd, 0, 0);
1277 if (read(fd, &boot_sect, 512) != 512)
1278 goto abort;
1279
1280 /* check MBP signature */
1281 if (boot_sect.magic == MBR_SIGNATURE_MAGIC) {
1282 retval = 1;
1283 /* found the correct signature */
1284 part = boot_sect.parts;
1285
1286 for (part_nr=0; part_nr < MBR_PARTITIONS; part_nr++) {
1287 /* check for GPT type */
1288 if (part->part_type == MBR_GPT_PARTITION_TYPE) {
1289 retval = get_gpt_last_partition_end(fd, endofpart);
1290 break;
1291 }
1292 /* check the last used lba for the current partition */
1293 curr_part_end = __le32_to_cpu(part->first_sect_lba) +
1294 __le32_to_cpu(part->blocks_num);
1295 if (curr_part_end > *endofpart)
1296 *endofpart = curr_part_end;
1297
1298 part++;
1299 }
1300 } else {
1301 /* Unknown partition table */
1302 retval = -1;
1303 }
1304 abort:
1305 return retval;
1306 }
1307
1308 int check_partitions(int fd, char *dname, unsigned long long freesize)
1309 {
1310 /*
1311 * Check where the last partition ends
1312 */
1313 unsigned long long endofpart;
1314 int ret;
1315
1316 if ((ret = get_last_partition_end(fd, &endofpart)) > 0) {
1317 /* There appears to be a partition table here */
1318 if (freesize == 0) {
1319 /* partitions will not be visible in new device */
1320 fprintf(stderr,
1321 Name ": partition table exists on %s but will be lost or\n"
1322 " meaningless after creating array\n",
1323 dname);
1324 return 1;
1325 } else if (endofpart > freesize) {
1326 /* last partition overlaps metadata */
1327 fprintf(stderr,
1328 Name ": metadata will over-write last partition on %s.\n",
1329 dname);
1330 return 1;
1331 }
1332 }
1333 return 0;
1334 }
1335
1336 void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk)
1337 {
1338 int d;
1339 ioctl(mdfd, GET_ARRAY_INFO, ainf);
1340 for (d = 0 ; d < ainf->raid_disks + ainf->nr_disks ; d++)
1341 if (ioctl(mdfd, GET_DISK_INFO, disk) == 0)
1342 return;
1343 }
1344
1345 int open_container(int fd)
1346 {
1347 /* 'fd' is a block device. Find out if it is in use
1348 * by a container, and return an open fd on that container.
1349 */
1350 char path[256];
1351 char *e;
1352 DIR *dir;
1353 struct dirent *de;
1354 int dfd, n;
1355 char buf[200];
1356 int major, minor;
1357 struct stat st;
1358
1359 if (fstat(fd, &st) != 0)
1360 return -1;
1361 sprintf(path, "/sys/dev/block/%d:%d/holders",
1362 (int)major(st.st_rdev), (int)minor(st.st_rdev));
1363 e = path + strlen(path);
1364
1365 dir = opendir(path);
1366 if (!dir)
1367 return -1;
1368 while ((de = readdir(dir))) {
1369 if (de->d_ino == 0)
1370 continue;
1371 if (de->d_name[0] == '.')
1372 continue;
1373 sprintf(e, "/%s/dev", de->d_name);
1374 dfd = open(path, O_RDONLY);
1375 if (dfd < 0)
1376 continue;
1377 n = read(dfd, buf, sizeof(buf));
1378 close(dfd);
1379 if (n <= 0 || n >= sizeof(buf))
1380 continue;
1381 buf[n] = 0;
1382 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
1383 continue;
1384 sprintf(buf, "%d:%d", major, minor);
1385 dfd = dev_open(buf, O_RDONLY);
1386 if (dfd >= 0) {
1387 closedir(dir);
1388 return dfd;
1389 }
1390 }
1391 closedir(dir);
1392 return -1;
1393 }
1394
1395 struct superswitch *version_to_superswitch(char *vers)
1396 {
1397 int i;
1398
1399 for (i = 0; superlist[i]; i++) {
1400 struct superswitch *ss = superlist[i];
1401
1402 if (strcmp(vers, ss->name) == 0)
1403 return ss;
1404 }
1405
1406 return NULL;
1407 }
1408
1409 int is_container_member(struct mdstat_ent *mdstat, char *container)
1410 {
1411 if (mdstat->metadata_version == NULL ||
1412 strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
1413 !is_subarray(mdstat->metadata_version+9) ||
1414 strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
1415 mdstat->metadata_version[10+strlen(container)] != '/')
1416 return 0;
1417
1418 return 1;
1419 }
1420
1421 int is_subarray_active(char *subarray, char *container)
1422 {
1423 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1424 struct mdstat_ent *ent;
1425
1426 for (ent = mdstat; ent; ent = ent->next) {
1427 if (is_container_member(ent, container)) {
1428 char *inst = &ent->metadata_version[10+strlen(container)+1];
1429
1430 if (!subarray || strcmp(inst, subarray) == 0)
1431 break;
1432 }
1433 }
1434
1435 free_mdstat(mdstat);
1436
1437 return ent != NULL;
1438 }
1439
1440 int is_container_active(char *container)
1441 {
1442 return is_subarray_active(NULL, container);
1443 }
1444
1445 /* open_subarray - opens a subarray in a container
1446 * @dev: container device name
1447 * @st: supertype with only ->subarray set
1448 * @quiet: block reporting errors flag
1449 *
1450 * On success returns an fd to a container and fills in *st
1451 */
1452 int open_subarray(char *dev, struct supertype *st, int quiet)
1453 {
1454 struct mdinfo *mdi;
1455 int fd, err = 1;
1456
1457 fd = open(dev, O_RDWR|O_EXCL);
1458 if (fd < 0) {
1459 if (!quiet)
1460 fprintf(stderr, Name ": Couldn't open %s, aborting\n",
1461 dev);
1462 return 2;
1463 }
1464
1465 st->devnum = fd2devnum(fd);
1466 if (st->devnum == NoMdDev) {
1467 if (!quiet)
1468 fprintf(stderr,
1469 Name ": Failed to determine device number for %s\n",
1470 dev);
1471 goto close_fd;
1472 }
1473
1474 mdi = sysfs_read(fd, st->devnum, GET_VERSION|GET_LEVEL);
1475 if (!mdi) {
1476 if (!quiet)
1477 fprintf(stderr, Name ": Failed to read sysfs for %s\n",
1478 dev);
1479 goto close_fd;
1480 }
1481
1482 if (mdi->array.level != UnSet) {
1483 if (!quiet)
1484 fprintf(stderr, Name ": %s is not a container\n", dev);
1485 goto free_sysfs;
1486 }
1487
1488 st->ss = version_to_superswitch(mdi->text_version);
1489 if (!st->ss) {
1490 if (!quiet)
1491 fprintf(stderr,
1492 Name ": Operation not supported for %s metadata\n",
1493 mdi->text_version);
1494 goto free_sysfs;
1495 }
1496
1497 st->devname = devnum2devname(st->devnum);
1498 if (!st->devname) {
1499 if (!quiet)
1500 fprintf(stderr, Name ": Failed to allocate device name\n");
1501 goto free_sysfs;
1502 }
1503
1504 if (st->ss->load_super(st, fd, NULL)) {
1505 if (!quiet)
1506 fprintf(stderr, Name ": Failed to find subarray-%s in %s\n",
1507 st->subarray, dev);
1508 goto free_name;
1509 }
1510
1511 if (!st->loaded_container) {
1512 if (!quiet)
1513 fprintf(stderr, Name ": %s is not a container\n", dev);
1514 goto free_super;
1515 }
1516
1517 err = 0;
1518
1519 free_super:
1520 if (err)
1521 st->ss->free_super(st);
1522 free_name:
1523 if (err)
1524 free(st->devname);
1525 free_sysfs:
1526 sysfs_free(mdi);
1527 close_fd:
1528 if (err)
1529 close(fd);
1530
1531 if (err)
1532 return -1;
1533 else
1534 return fd;
1535 }
1536
1537 int add_disk(int mdfd, struct supertype *st,
1538 struct mdinfo *sra, struct mdinfo *info)
1539 {
1540 /* Add a device to an array, in one of 2 ways. */
1541 int rv;
1542 #ifndef MDASSEMBLE
1543 if (st->ss->external) {
1544 if (info->disk.state & (1<<MD_DISK_SYNC))
1545 info->recovery_start = MaxSector;
1546 else
1547 info->recovery_start = 0;
1548 rv = sysfs_add_disk(sra, info, 0);
1549 if (! rv) {
1550 struct mdinfo *sd2;
1551 for (sd2 = sra->devs; sd2; sd2=sd2->next)
1552 if (sd2 == info)
1553 break;
1554 if (sd2 == NULL) {
1555 sd2 = malloc(sizeof(*sd2));
1556 *sd2 = *info;
1557 sd2->next = sra->devs;
1558 sra->devs = sd2;
1559 }
1560 }
1561 } else
1562 #endif
1563 rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk);
1564 return rv;
1565 }
1566
1567 int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
1568 {
1569 /* Initialise kernel's knowledge of array.
1570 * This varies between externally managed arrays
1571 * and older kernels
1572 */
1573 int vers = md_get_version(mdfd);
1574 int rv;
1575
1576 #ifndef MDASSEMBLE
1577 if (st->ss->external)
1578 rv = sysfs_set_array(info, vers);
1579 else
1580 #endif
1581 if ((vers % 100) >= 1) { /* can use different versions */
1582 mdu_array_info_t inf;
1583 memset(&inf, 0, sizeof(inf));
1584 inf.major_version = info->array.major_version;
1585 inf.minor_version = info->array.minor_version;
1586 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
1587 } else
1588 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
1589 return rv;
1590 }
1591
1592 unsigned long long min_recovery_start(struct mdinfo *array)
1593 {
1594 /* find the minimum recovery_start in an array for metadata
1595 * formats that only record per-array recovery progress instead
1596 * of per-device
1597 */
1598 unsigned long long recovery_start = MaxSector;
1599 struct mdinfo *d;
1600
1601 for (d = array->devs; d; d = d->next)
1602 recovery_start = min(recovery_start, d->recovery_start);
1603
1604 return recovery_start;
1605 }
1606
1607 char *devnum2devname(int num)
1608 {
1609 char name[100];
1610 if (num >= 0)
1611 sprintf(name, "md%d", num);
1612 else
1613 sprintf(name, "md_d%d", -1-num);
1614 return strdup(name);
1615 }
1616
1617 int devname2devnum(char *name)
1618 {
1619 char *ep;
1620 int num;
1621 if (strncmp(name, "md_d", 4)==0)
1622 num = -1-strtoul(name+4, &ep, 10);
1623 else
1624 num = strtoul(name+2, &ep, 10);
1625 return num;
1626 }
1627
1628 int stat2devnum(struct stat *st)
1629 {
1630 char path[30];
1631 char link[200];
1632 char *cp;
1633 int n;
1634
1635 if ((S_IFMT & st->st_mode) == S_IFBLK) {
1636 if (major(st->st_rdev) == MD_MAJOR)
1637 return minor(st->st_rdev);
1638 else if (major(st->st_rdev) == get_mdp_major())
1639 return -1- (minor(st->st_rdev)>>MdpMinorShift);
1640
1641 /* must be an extended-minor partition. Look at the
1642 * /sys/dev/block/%d:%d link which must look like
1643 * ../../block/mdXXX/mdXXXpYY
1644 */
1645 sprintf(path, "/sys/dev/block/%d:%d", major(st->st_rdev),
1646 minor(st->st_rdev));
1647 n = readlink(path, link, sizeof(link)-1);
1648 if (n <= 0)
1649 return NoMdDev;
1650 link[n] = 0;
1651 cp = strrchr(link, '/');
1652 if (cp) *cp = 0;
1653 cp = strchr(link, '/');
1654 if (cp && strncmp(cp, "/md", 3) == 0)
1655 return devname2devnum(cp+1);
1656 }
1657 return NoMdDev;
1658
1659 }
1660
1661 int fd2devnum(int fd)
1662 {
1663 struct stat stb;
1664 if (fstat(fd, &stb) == 0)
1665 return stat2devnum(&stb);
1666 return NoMdDev;
1667 }
1668
1669 char *pid_dir = VAR_RUN;
1670
1671 int mdmon_pid(int devnum)
1672 {
1673 char path[100];
1674 char pid[10];
1675 int fd;
1676 int n;
1677 char *devname = devnum2devname(devnum);
1678
1679 sprintf(path, "%s/%s.pid", pid_dir, devname);
1680 free(devname);
1681
1682 fd = open(path, O_RDONLY | O_NOATIME, 0);
1683
1684 if (fd < 0)
1685 return -1;
1686 n = read(fd, pid, 9);
1687 close(fd);
1688 if (n <= 0)
1689 return -1;
1690 return atoi(pid);
1691 }
1692
1693 int mdmon_running(int devnum)
1694 {
1695 int pid = mdmon_pid(devnum);
1696 if (pid <= 0)
1697 return 0;
1698 if (kill(pid, 0) == 0)
1699 return 1;
1700 return 0;
1701 }
1702
1703 int start_mdmon(int devnum)
1704 {
1705 int i;
1706 int len;
1707 pid_t pid;
1708 int status;
1709 char pathbuf[1024];
1710 char *paths[4] = {
1711 pathbuf,
1712 "/sbin/mdmon",
1713 "mdmon",
1714 NULL
1715 };
1716
1717 if (check_env("MDADM_NO_MDMON"))
1718 return 0;
1719
1720 len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf));
1721 if (len > 0) {
1722 char *sl;
1723 pathbuf[len] = 0;
1724 sl = strrchr(pathbuf, '/');
1725 if (sl)
1726 sl++;
1727 else
1728 sl = pathbuf;
1729 strcpy(sl, "mdmon");
1730 } else
1731 pathbuf[0] = '\0';
1732
1733 switch(fork()) {
1734 case 0:
1735 /* FIXME yuk. CLOSE_EXEC?? */
1736 for (i=3; i < 100; i++)
1737 close(i);
1738 for (i=0; paths[i]; i++)
1739 if (paths[i][0])
1740 execl(paths[i], "mdmon",
1741 devnum2devname(devnum),
1742 NULL);
1743 exit(1);
1744 case -1: fprintf(stderr, Name ": cannot run mdmon. "
1745 "Array remains readonly\n");
1746 return -1;
1747 default: /* parent - good */
1748 pid = wait(&status);
1749 if (pid < 0 || status != 0)
1750 return -1;
1751 }
1752 return 0;
1753 }
1754
1755 int check_env(char *name)
1756 {
1757 char *val = getenv(name);
1758
1759 if (val && atoi(val) == 1)
1760 return 1;
1761
1762 return 0;
1763 }
1764
1765 __u32 random32(void)
1766 {
1767 __u32 rv;
1768 int rfd = open("/dev/urandom", O_RDONLY);
1769 if (rfd < 0 || read(rfd, &rv, 4) != 4)
1770 rv = random();
1771 if (rfd >= 0)
1772 close(rfd);
1773 return rv;
1774 }
1775
1776 #ifndef MDASSEMBLE
1777 int flush_metadata_updates(struct supertype *st)
1778 {
1779 int sfd;
1780 if (!st->updates) {
1781 st->update_tail = NULL;
1782 return -1;
1783 }
1784
1785 sfd = connect_monitor(devnum2devname(st->container_dev));
1786 if (sfd < 0)
1787 return -1;
1788
1789 while (st->updates) {
1790 struct metadata_update *mu = st->updates;
1791 st->updates = mu->next;
1792
1793 send_message(sfd, mu, 0);
1794 wait_reply(sfd, 0);
1795 free(mu->buf);
1796 free(mu);
1797 }
1798 ack(sfd, 0);
1799 wait_reply(sfd, 0);
1800 close(sfd);
1801 st->update_tail = NULL;
1802 return 0;
1803 }
1804
1805 void append_metadata_update(struct supertype *st, void *buf, int len)
1806 {
1807
1808 struct metadata_update *mu = malloc(sizeof(*mu));
1809
1810 mu->buf = buf;
1811 mu->len = len;
1812 mu->space = NULL;
1813 mu->next = NULL;
1814 *st->update_tail = mu;
1815 st->update_tail = &mu->next;
1816 }
1817 #endif /* MDASSEMBLE */
1818
1819 #ifdef __TINYC__
1820 /* tinyc doesn't optimize this check in ioctl.h out ... */
1821 unsigned int __invalid_size_argument_for_IOC = 0;
1822 #endif
1823