]> git.ipfire.org Git - thirdparty/mdadm.git/blob - util.c
simplify calculating array_blocks
[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 int __offroot;
36
37 /*
38 * following taken from linux/blkpg.h because they aren't
39 * anywhere else and it isn't safe to #include linux/ * stuff.
40 */
41
42 #define BLKPG _IO(0x12,105)
43
44 /* The argument structure */
45 struct blkpg_ioctl_arg {
46 int op;
47 int flags;
48 int datalen;
49 void *data;
50 };
51
52 /* The subfunctions (for the op field) */
53 #define BLKPG_ADD_PARTITION 1
54 #define BLKPG_DEL_PARTITION 2
55
56 /* Sizes of name fields. Unused at present. */
57 #define BLKPG_DEVNAMELTH 64
58 #define BLKPG_VOLNAMELTH 64
59
60 /* The data structure for ADD_PARTITION and DEL_PARTITION */
61 struct blkpg_partition {
62 long long start; /* starting offset in bytes */
63 long long length; /* length in bytes */
64 int pno; /* partition number */
65 char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2,
66 to be used in kernel messages */
67 char volname[BLKPG_VOLNAMELTH]; /* volume label */
68 };
69
70 #include "part.h"
71
72 /* Force a compilation error if condition is true */
73 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
74
75 /* Force a compilation error if condition is true, but also produce a
76 result (of value 0 and type size_t), so the expression can be used
77 e.g. in a structure initializer (or where-ever else comma expressions
78 aren't permitted). */
79 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
80
81 /*
82 * Parse a 128 bit uuid in 4 integers
83 * format is 32 hexx nibbles with options :.<space> separator
84 * If not exactly 32 hex digits are found, return 0
85 * else return 1
86 */
87 int parse_uuid(char *str, int uuid[4])
88 {
89 int hit = 0; /* number of Hex digIT */
90 int i;
91 char c;
92 for (i=0; i<4; i++) uuid[i]=0;
93
94 while ((c= *str++)) {
95 int n;
96 if (c>='0' && c<='9')
97 n = c-'0';
98 else if (c>='a' && c <= 'f')
99 n = 10 + c - 'a';
100 else if (c>='A' && c <= 'F')
101 n = 10 + c - 'A';
102 else if (strchr(":. -", c))
103 continue;
104 else return 0;
105
106 if (hit<32) {
107 uuid[hit/8] <<= 4;
108 uuid[hit/8] += n;
109 }
110 hit++;
111 }
112 if (hit == 32)
113 return 1;
114 return 0;
115 }
116
117
118 /*
119 * Get the md version number.
120 * We use the RAID_VERSION ioctl if it is supported
121 * If not, but we have a block device with major '9', we assume
122 * 0.36.0
123 *
124 * Return version number as 24 but number - assume version parts
125 * always < 255
126 */
127
128 int md_get_version(int fd)
129 {
130 struct stat stb;
131 mdu_version_t vers;
132
133 if (fstat(fd, &stb)<0)
134 return -1;
135 if ((S_IFMT&stb.st_mode) != S_IFBLK)
136 return -1;
137
138 if (ioctl(fd, RAID_VERSION, &vers) == 0)
139 return (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
140 if (errno == EACCES)
141 return -1;
142 if (major(stb.st_rdev) == MD_MAJOR)
143 return (3600);
144 return -1;
145 }
146
147 int get_linux_version()
148 {
149 struct utsname name;
150 char *cp;
151 int a = 0, b = 0,c = 0;
152 if (uname(&name) <0)
153 return -1;
154
155 cp = name.release;
156 a = strtoul(cp, &cp, 10);
157 if (*cp == '.')
158 b = strtoul(cp+1, &cp, 10);
159 if (*cp == '.')
160 c = strtoul(cp+1, &cp, 10);
161
162 return (a*1000000)+(b*1000)+c;
163 }
164
165 #ifndef MDASSEMBLE
166 int mdadm_version(char *version)
167 {
168 int a, b, c;
169 char *cp;
170
171 if (!version)
172 version = Version;
173
174 cp = strchr(version, '-');
175 if (!cp || *(cp+1) != ' ' || *(cp+2) != 'v')
176 return -1;
177 cp += 3;
178 a = strtoul(cp, &cp, 10);
179 if (*cp != '.')
180 return -1;
181 b = strtoul(cp+1, &cp, 10);
182 if (*cp == '.')
183 c = strtoul(cp+1, &cp, 10);
184 else
185 c = 0;
186 if (*cp != ' ' && *cp != '-')
187 return -1;
188 return (a*1000000)+(b*1000)+c;
189 }
190
191 long long parse_size(char *size)
192 {
193 /* parse 'size' which should be a number optionally
194 * followed by 'K', 'M', or 'G'.
195 * Without a suffix, K is assumed.
196 * Number returned is in sectors (half-K)
197 */
198 char *c;
199 long long s = strtoll(size, &c, 10);
200 if (s > 0) {
201 switch (*c) {
202 case 'K':
203 c++;
204 default:
205 s *= 2;
206 break;
207 case 'M':
208 c++;
209 s *= 1024 * 2;
210 break;
211 case 'G':
212 c++;
213 s *= 1024 * 1024 * 2;
214 break;
215 }
216 }
217 if (*c)
218 s = 0;
219 return s;
220 }
221
222 int parse_layout_10(char *layout)
223 {
224 int copies, rv;
225 char *cp;
226 /* Parse the layout string for raid10 */
227 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
228 if ((layout[0] != 'n' && layout[0] != 'f' && layout[0] != 'o') ||
229 (copies = strtoul(layout+1, &cp, 10)) < 1 ||
230 copies > 200 ||
231 *cp)
232 return -1;
233 if (layout[0] == 'n')
234 rv = 256 + copies;
235 else if (layout[0] == 'o')
236 rv = 0x10000 + (copies<<8) + 1;
237 else
238 rv = 1 + (copies<<8);
239 return rv;
240 }
241
242 int parse_layout_faulty(char *layout)
243 {
244 /* Parse the layout string for 'faulty' */
245 int ln = strcspn(layout, "0123456789");
246 char *m = strdup(layout);
247 int mode;
248 m[ln] = 0;
249 mode = map_name(faultylayout, m);
250 if (mode == UnSet)
251 return -1;
252
253 return mode | (atoi(layout+ln)<< ModeShift);
254 }
255 #endif
256
257 void remove_partitions(int fd)
258 {
259 /* remove partitions from this block devices.
260 * This is used for components added to an array
261 */
262 #ifdef BLKPG_DEL_PARTITION
263 struct blkpg_ioctl_arg a;
264 struct blkpg_partition p;
265
266 a.op = BLKPG_DEL_PARTITION;
267 a.data = (void*)&p;
268 a.datalen = sizeof(p);
269 a.flags = 0;
270 memset(a.data, 0, a.datalen);
271 for (p.pno=0; p.pno < 16; p.pno++)
272 ioctl(fd, BLKPG, &a);
273 #endif
274 }
275
276 int test_partition(int fd)
277 {
278 /* Check if fd is a whole-disk or a partition.
279 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
280 * will return ENXIO on an invalid partition number.
281 */
282 struct blkpg_ioctl_arg a;
283 struct blkpg_partition p;
284 a.op = BLKPG_DEL_PARTITION;
285 a.data = (void*)&p;
286 a.datalen = sizeof(p);
287 a.flags = 0;
288 memset(a.data, 0, a.datalen);
289 p.pno = 1<<30;
290 if (ioctl(fd, BLKPG, &a) == 0)
291 /* Very unlikely, but not a partition */
292 return 0;
293 if (errno == ENXIO)
294 /* not a partition */
295 return 0;
296
297 return 1;
298 }
299
300 int test_partition_from_id(dev_t id)
301 {
302 char buf[20];
303 int fd, rv;
304
305 sprintf(buf, "%d:%d", major(id), minor(id));
306 fd = dev_open(buf, O_RDONLY);
307 if (fd < 0)
308 return -1;
309 rv = test_partition(fd);
310 close(fd);
311 return rv;
312 }
313
314 int enough(int level, int raid_disks, int layout, int clean, char *avail)
315 {
316 int copies, first;
317 int i;
318 int avail_disks = 0;
319
320 for (i = 0; i < raid_disks; i++)
321 avail_disks += !!avail[i];
322
323 switch (level) {
324 case 10:
325 /* This is the tricky one - we need to check
326 * which actual disks are present.
327 */
328 copies = (layout&255)* ((layout>>8) & 255);
329 first=0;
330 do {
331 /* there must be one of the 'copies' form 'first' */
332 int n = copies;
333 int cnt=0;
334 while (n--) {
335 if (avail[first])
336 cnt++;
337 first = (first+1) % raid_disks;
338 }
339 if (cnt == 0)
340 return 0;
341
342 } while (first != 0);
343 return 1;
344
345 case LEVEL_MULTIPATH:
346 return avail_disks>= 1;
347 case LEVEL_LINEAR:
348 case 0:
349 return avail_disks == raid_disks;
350 case 1:
351 return avail_disks >= 1;
352 case 4:
353 case 5:
354 if (clean)
355 return avail_disks >= raid_disks-1;
356 else
357 return avail_disks >= raid_disks;
358 case 6:
359 if (clean)
360 return avail_disks >= raid_disks-2;
361 else
362 return avail_disks >= raid_disks;
363 default:
364 return 0;
365 }
366 }
367
368 int enough_fd(int fd)
369 {
370 struct mdu_array_info_s array;
371 struct mdu_disk_info_s disk;
372 int avail_disks = 0;
373 int i, rv;
374 char *avail;
375
376 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0 ||
377 array.raid_disks <= 0)
378 return 0;
379 avail = calloc(array.raid_disks, 1);
380 for (i=0; i < 1024 && array.nr_disks > 0; i++) {
381 disk.number = i;
382 if (ioctl(fd, GET_DISK_INFO, &disk) != 0)
383 continue;
384 if (disk.major == 0 && disk.minor == 0)
385 continue;
386 array.nr_disks--;
387
388 if (! (disk.state & (1<<MD_DISK_SYNC)))
389 continue;
390 if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks)
391 continue;
392 avail_disks++;
393 avail[disk.raid_disk] = 1;
394 }
395 /* This is used on an active array, so assume it is clean */
396 rv = enough(array.level, array.raid_disks, array.layout,
397 1, avail);
398 free(avail);
399 return rv;
400 }
401
402
403 const int uuid_zero[4] = { 0, 0, 0, 0 };
404
405 int same_uuid(int a[4], int b[4], int swapuuid)
406 {
407 if (swapuuid) {
408 /* parse uuids are hostendian.
409 * uuid's from some superblocks are big-ending
410 * if there is a difference, we need to swap..
411 */
412 unsigned char *ac = (unsigned char *)a;
413 unsigned char *bc = (unsigned char *)b;
414 int i;
415 for (i=0; i<16; i+= 4) {
416 if (ac[i+0] != bc[i+3] ||
417 ac[i+1] != bc[i+2] ||
418 ac[i+2] != bc[i+1] ||
419 ac[i+3] != bc[i+0])
420 return 0;
421 }
422 return 1;
423 } else {
424 if (a[0]==b[0] &&
425 a[1]==b[1] &&
426 a[2]==b[2] &&
427 a[3]==b[3])
428 return 1;
429 return 0;
430 }
431 }
432 void copy_uuid(void *a, int b[4], int swapuuid)
433 {
434 if (swapuuid) {
435 /* parse uuids are hostendian.
436 * uuid's from some superblocks are big-ending
437 * if there is a difference, we need to swap..
438 */
439 unsigned char *ac = (unsigned char *)a;
440 unsigned char *bc = (unsigned char *)b;
441 int i;
442 for (i=0; i<16; i+= 4) {
443 ac[i+0] = bc[i+3];
444 ac[i+1] = bc[i+2];
445 ac[i+2] = bc[i+1];
446 ac[i+3] = bc[i+0];
447 }
448 } else
449 memcpy(a, b, 16);
450 }
451
452 char *__fname_from_uuid(int id[4], int swap, char *buf, char sep)
453 {
454 int i, j;
455 char uuid[16];
456 char *c = buf;
457 strcpy(c, "UUID-");
458 c += strlen(c);
459 copy_uuid(uuid, id, swap);
460 for (i = 0; i < 4; i++) {
461 if (i)
462 *c++ = sep;
463 for (j = 3; j >= 0; j--) {
464 sprintf(c,"%02x", (unsigned char) uuid[j+4*i]);
465 c+= 2;
466 }
467 }
468 return buf;
469
470 }
471
472 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
473 {
474 // dirty hack to work around an issue with super1 superblocks...
475 // super1 superblocks need swapuuid set in order for assembly to
476 // work, but can't have it set if we want this printout to match
477 // all the other uuid printouts in super1.c, so we force swapuuid
478 // to 1 to make our printout match the rest of super1
479 return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep);
480 }
481
482 #ifndef MDASSEMBLE
483 int check_ext2(int fd, char *name)
484 {
485 /*
486 * Check for an ext2fs file system.
487 * Superblock is always 1K at 1K offset
488 *
489 * s_magic is le16 at 56 == 0xEF53
490 * report mtime - le32 at 44
491 * blocks - le32 at 4
492 * logblksize - le32 at 24
493 */
494 unsigned char sb[1024];
495 time_t mtime;
496 int size, bsize;
497 if (lseek(fd, 1024,0)!= 1024)
498 return 0;
499 if (read(fd, sb, 1024)!= 1024)
500 return 0;
501 if (sb[56] != 0x53 || sb[57] != 0xef)
502 return 0;
503
504 mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
505 bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
506 size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
507 fprintf(stderr, Name ": %s appears to contain an ext2fs file system\n",
508 name);
509 fprintf(stderr," size=%dK mtime=%s",
510 size*(1<<bsize), ctime(&mtime));
511 return 1;
512 }
513
514 int check_reiser(int fd, char *name)
515 {
516 /*
517 * superblock is at 64K
518 * size is 1024;
519 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
520 *
521 */
522 unsigned char sb[1024];
523 unsigned long size;
524 if (lseek(fd, 64*1024, 0) != 64*1024)
525 return 0;
526 if (read(fd, sb, 1024) != 1024)
527 return 0;
528 if (strncmp((char*)sb+52, "ReIsErFs",8)!=0 &&
529 strncmp((char*)sb+52, "ReIsEr2Fs",9)!=0)
530 return 0;
531 fprintf(stderr, Name ": %s appears to contain a reiserfs file system\n",name);
532 size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
533 fprintf(stderr, " size = %luK\n", size*4);
534
535 return 1;
536 }
537
538 int check_raid(int fd, char *name)
539 {
540 struct mdinfo info;
541 time_t crtime;
542 char *level;
543 struct supertype *st = guess_super(fd);
544
545 if (!st) return 0;
546 st->ignore_hw_compat = 1;
547 st->ss->load_super(st, fd, name);
548 /* Looks like a raid array .. */
549 fprintf(stderr, Name ": %s appears to be part of a raid array:\n",
550 name);
551 st->ss->getinfo_super(st, &info, NULL);
552 st->ss->free_super(st);
553 crtime = info.array.ctime;
554 level = map_num(pers, info.array.level);
555 if (!level) level = "-unknown-";
556 fprintf(stderr, " level=%s devices=%d ctime=%s",
557 level, info.array.raid_disks, ctime(&crtime));
558 return 1;
559 }
560
561 int ask(char *mesg)
562 {
563 char *add = "";
564 int i;
565 for (i=0; i<5; i++) {
566 char buf[100];
567 fprintf(stderr, "%s%s", mesg, add);
568 fflush(stderr);
569 if (fgets(buf, 100, stdin)==NULL)
570 return 0;
571 if (buf[0]=='y' || buf[0]=='Y')
572 return 1;
573 if (buf[0]=='n' || buf[0]=='N')
574 return 0;
575 add = "(y/n) ";
576 }
577 fprintf(stderr, Name ": assuming 'no'\n");
578 return 0;
579 }
580 #endif /* MDASSEMBLE */
581
582 int is_standard(char *dev, int *nump)
583 {
584 /* tests if dev is a "standard" md dev name.
585 * i.e if the last component is "/dNN" or "/mdNN",
586 * where NN is a string of digits
587 * Returns 1 if a partitionable standard,
588 * -1 if non-partitonable,
589 * 0 if not a standard name.
590 */
591 char *d = strrchr(dev, '/');
592 int type=0;
593 int num;
594 if (!d)
595 return 0;
596 if (strncmp(d, "/d",2)==0)
597 d += 2, type=1; /* /dev/md/dN{pM} */
598 else if (strncmp(d, "/md_d", 5)==0)
599 d += 5, type=1; /* /dev/md_dN{pM} */
600 else if (strncmp(d, "/md", 3)==0)
601 d += 3, type=-1; /* /dev/mdN */
602 else if (d-dev > 3 && strncmp(d-2, "md/", 3)==0)
603 d += 1, type=-1; /* /dev/md/N */
604 else
605 return 0;
606 if (!*d)
607 return 0;
608 num = atoi(d);
609 while (isdigit(*d))
610 d++;
611 if (*d)
612 return 0;
613 if (nump) *nump = num;
614
615 return type;
616 }
617
618 unsigned long calc_csum(void *super, int bytes)
619 {
620 unsigned long long newcsum = 0;
621 int i;
622 unsigned int csum;
623 unsigned int *superc = (unsigned int*) super;
624
625 for(i=0; i<bytes/4; i++)
626 newcsum+= superc[i];
627 csum = (newcsum& 0xffffffff) + (newcsum>>32);
628 #ifdef __alpha__
629 /* The in-kernel checksum calculation is always 16bit on
630 * the alpha, though it is 32 bit on i386...
631 * I wonder what it is elsewhere... (it uses and API in
632 * a way that it shouldn't).
633 */
634 csum = (csum & 0xffff) + (csum >> 16);
635 csum = (csum & 0xffff) + (csum >> 16);
636 #endif
637 return csum;
638 }
639
640 #ifndef MDASSEMBLE
641 char *human_size(long long bytes)
642 {
643 static char buf[30];
644
645 /* We convert bytes to either centi-M{ega,ibi}bytes or
646 * centi-G{igi,ibi}bytes, with appropriate rounding,
647 * and then print 1/100th of those as a decimal.
648 * We allow upto 2048Megabytes before converting to
649 * gigabytes, as that shows more precision and isn't
650 * too large a number.
651 * Terabytes are not yet handled.
652 */
653
654 if (bytes < 5000*1024)
655 buf[0]=0;
656 else if (bytes < 2*1024LL*1024LL*1024LL) {
657 long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2;
658 long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2;
659 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
660 cMiB/100 , cMiB % 100,
661 cMB/100, cMB % 100);
662 } else {
663 long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2;
664 long cGB = (bytes / (1000000000LL/200LL ) +1) /2;
665 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
666 cGiB/100 , cGiB % 100,
667 cGB/100, cGB % 100);
668 }
669 return buf;
670 }
671
672 char *human_size_brief(long long bytes)
673 {
674 static char buf[30];
675
676 if (bytes < 5000*1024)
677 snprintf(buf, sizeof(buf), "%ld.%02ldKiB",
678 (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024)
679 );
680 else if (bytes < 2*1024LL*1024LL*1024LL)
681 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
682 (long)(bytes>>20),
683 (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100)
684 );
685 else
686 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
687 (long)(bytes>>30),
688 (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100)
689 );
690 return buf;
691 }
692
693 void print_r10_layout(int layout)
694 {
695 int near = layout & 255;
696 int far = (layout >> 8) & 255;
697 int offset = (layout&0x10000);
698 char *sep = "";
699
700 if (near != 1) {
701 printf("%s near=%d", sep, near);
702 sep = ",";
703 }
704 if (far != 1)
705 printf("%s %s=%d", sep, offset?"offset":"far", far);
706 if (near*far == 1)
707 printf("NO REDUNDANCY");
708 }
709 #endif
710
711 unsigned long long calc_array_size(int level, int raid_disks, int layout,
712 int chunksize, unsigned long long devsize)
713 {
714 if (level == 1)
715 return devsize;
716 devsize &= ~(unsigned long long)((chunksize>>9)-1);
717 return get_data_disks(level, layout, raid_disks) * devsize;
718 }
719
720 int get_data_disks(int level, int layout, int raid_disks)
721 {
722 int data_disks = 0;
723 switch (level) {
724 case 0: data_disks = raid_disks; break;
725 case 1: data_disks = 1; break;
726 case 4:
727 case 5: data_disks = raid_disks - 1; break;
728 case 6: data_disks = raid_disks - 2; break;
729 case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
730 break;
731 }
732
733 return data_disks;
734 }
735
736 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
737 char *get_md_name(int dev)
738 {
739 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
740 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
741 static char devname[50];
742 struct stat stb;
743 dev_t rdev;
744 char *dn;
745
746 if (dev < 0) {
747 int mdp = get_mdp_major();
748 if (mdp < 0) return NULL;
749 rdev = makedev(mdp, (-1-dev)<<6);
750 snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev);
751 if (stat(devname, &stb) == 0
752 && (S_IFMT&stb.st_mode) == S_IFBLK
753 && (stb.st_rdev == rdev))
754 return devname;
755 } else {
756 rdev = makedev(MD_MAJOR, dev);
757 snprintf(devname, sizeof(devname), "/dev/md%d", dev);
758 if (stat(devname, &stb) == 0
759 && (S_IFMT&stb.st_mode) == S_IFBLK
760 && (stb.st_rdev == rdev))
761 return devname;
762
763 snprintf(devname, sizeof(devname), "/dev/md/%d", dev);
764 if (stat(devname, &stb) == 0
765 && (S_IFMT&stb.st_mode) == S_IFBLK
766 && (stb.st_rdev == rdev))
767 return devname;
768 }
769 dn = map_dev(major(rdev), minor(rdev), 0);
770 if (dn)
771 return dn;
772 snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev);
773 if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
774 if (errno != EEXIST)
775 return NULL;
776
777 if (stat(devname, &stb) == 0
778 && (S_IFMT&stb.st_mode) == S_IFBLK
779 && (stb.st_rdev == rdev))
780 return devname;
781 unlink(devname);
782 return NULL;
783 }
784
785 void put_md_name(char *name)
786 {
787 if (strncmp(name, "/dev/.tmp.md", 12)==0)
788 unlink(name);
789 }
790
791 int find_free_devnum(int use_partitions)
792 {
793 int devnum;
794 for (devnum = 127; devnum != 128;
795 devnum = devnum ? devnum-1 : (1<<20)-1) {
796 char *dn;
797 int _devnum;
798
799 _devnum = use_partitions ? (-1-devnum) : devnum;
800 if (mddev_busy(_devnum))
801 continue;
802 /* make sure it is new to /dev too, at least as a
803 * non-standard */
804 dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0);
805 if (dn && ! is_standard(dn, NULL))
806 continue;
807 break;
808 }
809 if (devnum == 128)
810 return NoMdDev;
811 return use_partitions ? (-1-devnum) : devnum;
812 }
813 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
814
815 int dev_open(char *dev, int flags)
816 {
817 /* like 'open', but if 'dev' matches %d:%d, create a temp
818 * block device and open that
819 */
820 char *e;
821 int fd = -1;
822 char devname[32];
823 int major;
824 int minor;
825
826 if (!dev) return -1;
827 flags |= O_DIRECT;
828
829 major = strtoul(dev, &e, 0);
830 if (e > dev && *e == ':' && e[1] &&
831 (minor = strtoul(e+1, &e, 0)) >= 0 &&
832 *e == 0) {
833 char *path = map_dev(major, minor, 0);
834 if (path)
835 fd = open(path, flags);
836 if (fd < 0) {
837 snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d",
838 (int)getpid(), major, minor);
839 if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) {
840 fd = open(devname, flags);
841 unlink(devname);
842 }
843 }
844 if (fd < 0) {
845 snprintf(devname, sizeof(devname), "/tmp/.tmp.md.%d:%d:%d",
846 (int)getpid(), major, minor);
847 if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) {
848 fd = open(devname, flags);
849 unlink(devname);
850 }
851 }
852 } else
853 fd = open(dev, flags);
854 return fd;
855 }
856
857 int open_dev_flags(int devnum, int flags)
858 {
859 char buf[20];
860
861 sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
862 return dev_open(buf, flags);
863 }
864
865 int open_dev(int devnum)
866 {
867 return open_dev_flags(devnum, O_RDONLY);
868 }
869
870 int open_dev_excl(int devnum)
871 {
872 char buf[20];
873 int i;
874 int flags = O_RDWR;
875
876 sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
877 for (i=0 ; i<25 ; i++) {
878 int fd = dev_open(buf, flags|O_EXCL);
879 if (fd >= 0)
880 return fd;
881 if (errno == EACCES && flags == O_RDWR) {
882 flags = O_RDONLY;
883 continue;
884 }
885 if (errno != EBUSY)
886 return fd;
887 usleep(200000);
888 }
889 return -1;
890 }
891
892 int same_dev(char *one, char *two)
893 {
894 struct stat st1, st2;
895 if (stat(one, &st1) != 0)
896 return 0;
897 if (stat(two, &st2) != 0)
898 return 0;
899 if ((st1.st_mode & S_IFMT) != S_IFBLK)
900 return 0;
901 if ((st2.st_mode & S_IFMT) != S_IFBLK)
902 return 0;
903 return st1.st_rdev == st2.st_rdev;
904 }
905
906 void wait_for(char *dev, int fd)
907 {
908 int i;
909 struct stat stb_want;
910
911 if (fstat(fd, &stb_want) != 0 ||
912 (stb_want.st_mode & S_IFMT) != S_IFBLK)
913 return;
914
915 for (i=0 ; i<25 ; i++) {
916 struct stat stb;
917 if (stat(dev, &stb) == 0 &&
918 (stb.st_mode & S_IFMT) == S_IFBLK &&
919 (stb.st_rdev == stb_want.st_rdev))
920 return;
921 usleep(200000);
922 }
923 if (i == 25)
924 dprintf("%s: timeout waiting for %s\n", __func__, dev);
925 }
926
927 struct superswitch *superlist[] =
928 {
929 &super0, &super1,
930 &super_ddf, &super_imsm,
931 &mbr, &gpt,
932 NULL };
933
934 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
935
936 struct supertype *super_by_fd(int fd, char **subarrayp)
937 {
938 mdu_array_info_t array;
939 int vers;
940 int minor;
941 struct supertype *st = NULL;
942 struct mdinfo *sra;
943 char *verstr;
944 char version[20];
945 int i;
946 char *subarray = NULL;
947 int container = NoMdDev;
948
949 sra = sysfs_read(fd, 0, GET_VERSION);
950
951 if (sra) {
952 vers = sra->array.major_version;
953 minor = sra->array.minor_version;
954 verstr = sra->text_version;
955 } else {
956 if (ioctl(fd, GET_ARRAY_INFO, &array))
957 array.major_version = array.minor_version = 0;
958 vers = array.major_version;
959 minor = array.minor_version;
960 verstr = "";
961 }
962
963 if (vers != -1) {
964 sprintf(version, "%d.%d", vers, minor);
965 verstr = version;
966 }
967 if (minor == -2 && is_subarray(verstr)) {
968 char *dev = verstr+1;
969
970 subarray = strchr(dev, '/');
971 if (subarray) {
972 *subarray++ = '\0';
973 subarray = strdup(subarray);
974 }
975 container = devname2devnum(dev);
976 if (sra)
977 sysfs_free(sra);
978 sra = sysfs_read(-1, container, GET_VERSION);
979 if (sra && sra->text_version[0])
980 verstr = sra->text_version;
981 else
982 verstr = "-no-metadata-";
983 }
984
985 for (i = 0; st == NULL && superlist[i] ; i++)
986 st = superlist[i]->match_metadata_desc(verstr);
987
988 if (sra)
989 sysfs_free(sra);
990 if (st) {
991 st->sb = NULL;
992 if (subarrayp)
993 *subarrayp = subarray;
994 st->container_dev = container;
995 st->devnum = fd2devnum(fd);
996 } else
997 free(subarray);
998
999 return st;
1000 }
1001 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1002
1003 int dev_size_from_id(dev_t id, unsigned long long *size)
1004 {
1005 char buf[20];
1006 int fd;
1007
1008 sprintf(buf, "%d:%d", major(id), minor(id));
1009 fd = dev_open(buf, O_RDONLY);
1010 if (fd < 0)
1011 return 0;
1012 if (get_dev_size(fd, NULL, size)) {
1013 close(fd);
1014 return 1;
1015 }
1016 close(fd);
1017 return 0;
1018 }
1019
1020 struct supertype *dup_super(struct supertype *orig)
1021 {
1022 struct supertype *st;
1023
1024 if (!orig)
1025 return orig;
1026 st = malloc(sizeof(*st));
1027 if (!st)
1028 return st;
1029 memset(st, 0, sizeof(*st));
1030 st->ss = orig->ss;
1031 st->max_devs = orig->max_devs;
1032 st->minor_version = orig->minor_version;
1033 st->sb = NULL;
1034 st->info = NULL;
1035 return st;
1036 }
1037
1038 struct supertype *guess_super_type(int fd, enum guess_types guess_type)
1039 {
1040 /* try each load_super to find the best match,
1041 * and return the best superswitch
1042 */
1043 struct superswitch *ss;
1044 struct supertype *st;
1045 time_t besttime = 0;
1046 int bestsuper = -1;
1047 int i;
1048
1049 st = malloc(sizeof(*st));
1050 memset(st, 0, sizeof(*st));
1051 st->container_dev = NoMdDev;
1052
1053 for (i=0 ; superlist[i]; i++) {
1054 int rv;
1055 ss = superlist[i];
1056 if (guess_type == guess_array && ss->add_to_super == NULL)
1057 continue;
1058 if (guess_type == guess_partitions && ss->add_to_super != NULL)
1059 continue;
1060 memset(st, 0, sizeof(*st));
1061 st->ignore_hw_compat = 1;
1062 rv = ss->load_super(st, fd, NULL);
1063 if (rv == 0) {
1064 struct mdinfo info;
1065 st->ss->getinfo_super(st, &info, NULL);
1066 if (bestsuper == -1 ||
1067 besttime < info.array.ctime) {
1068 bestsuper = i;
1069 besttime = info.array.ctime;
1070 }
1071 ss->free_super(st);
1072 }
1073 }
1074 if (bestsuper != -1) {
1075 int rv;
1076 memset(st, 0, sizeof(*st));
1077 st->ignore_hw_compat = 1;
1078 rv = superlist[bestsuper]->load_super(st, fd, NULL);
1079 if (rv == 0) {
1080 superlist[bestsuper]->free_super(st);
1081 st->ignore_hw_compat = 0;
1082 return st;
1083 }
1084 }
1085 free(st);
1086 return NULL;
1087 }
1088
1089 /* Return size of device in bytes */
1090 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
1091 {
1092 unsigned long long ldsize;
1093 struct stat st;
1094
1095 if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
1096 ldsize = (unsigned long long)st.st_size;
1097 else
1098 #ifdef BLKGETSIZE64
1099 if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
1100 #endif
1101 {
1102 unsigned long dsize;
1103 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
1104 ldsize = dsize;
1105 ldsize <<= 9;
1106 } else {
1107 if (dname)
1108 fprintf(stderr, Name ": Cannot get size of %s: %s\b",
1109 dname, strerror(errno));
1110 return 0;
1111 }
1112 }
1113 *sizep = ldsize;
1114 return 1;
1115 }
1116
1117 /* Return true if this can only be a container, not a member device.
1118 * i.e. is and md device and size is zero
1119 */
1120 int must_be_container(int fd)
1121 {
1122 unsigned long long size;
1123 if (md_get_version(fd) < 0)
1124 return 0;
1125 if (get_dev_size(fd, NULL, &size) == 0)
1126 return 1;
1127 if (size == 0)
1128 return 1;
1129 return 0;
1130 }
1131
1132 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1133 * Returns: 1 if successful
1134 * -1 for unknown partition type
1135 * 0 for other errors
1136 */
1137 static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart)
1138 {
1139 struct GPT gpt;
1140 unsigned char empty_gpt_entry[16]= {0};
1141 struct GPT_part_entry *part;
1142 char buf[512];
1143 unsigned long long curr_part_end;
1144 unsigned all_partitions, entry_size;
1145 unsigned part_nr;
1146
1147 *endofpart = 0;
1148
1149 BUILD_BUG_ON(sizeof(gpt) != 512);
1150 /* skip protective MBR */
1151 lseek(fd, 512, SEEK_SET);
1152 /* read GPT header */
1153 if (read(fd, &gpt, 512) != 512)
1154 return 0;
1155
1156 /* get the number of partition entries and the entry size */
1157 all_partitions = __le32_to_cpu(gpt.part_cnt);
1158 entry_size = __le32_to_cpu(gpt.part_size);
1159
1160 /* Check GPT signature*/
1161 if (gpt.magic != GPT_SIGNATURE_MAGIC)
1162 return -1;
1163
1164 /* sanity checks */
1165 if (all_partitions > 1024 ||
1166 entry_size > sizeof(buf))
1167 return -1;
1168
1169 part = (struct GPT_part_entry *)buf;
1170
1171 for (part_nr=0; part_nr < all_partitions; part_nr++) {
1172 /* read partition entry */
1173 if (read(fd, buf, entry_size) != (ssize_t)entry_size)
1174 return 0;
1175
1176 /* is this valid partition? */
1177 if (memcmp(part->type_guid, empty_gpt_entry, 16) != 0) {
1178 /* check the last lba for the current partition */
1179 curr_part_end = __le64_to_cpu(part->ending_lba);
1180 if (curr_part_end > *endofpart)
1181 *endofpart = curr_part_end;
1182 }
1183
1184 }
1185 return 1;
1186 }
1187
1188 /* Sets endofpart parameter to the last block used by the last partition on the device.
1189 * Returns: 1 if successful
1190 * -1 for unknown partition type
1191 * 0 for other errors
1192 */
1193 static int get_last_partition_end(int fd, unsigned long long *endofpart)
1194 {
1195 struct MBR boot_sect;
1196 struct MBR_part_record *part;
1197 unsigned long long curr_part_end;
1198 unsigned part_nr;
1199 int retval = 0;
1200
1201 *endofpart = 0;
1202
1203 BUILD_BUG_ON(sizeof(boot_sect) != 512);
1204 /* read MBR */
1205 lseek(fd, 0, 0);
1206 if (read(fd, &boot_sect, 512) != 512)
1207 goto abort;
1208
1209 /* check MBP signature */
1210 if (boot_sect.magic == MBR_SIGNATURE_MAGIC) {
1211 retval = 1;
1212 /* found the correct signature */
1213 part = boot_sect.parts;
1214
1215 for (part_nr=0; part_nr < MBR_PARTITIONS; part_nr++) {
1216 /* check for GPT type */
1217 if (part->part_type == MBR_GPT_PARTITION_TYPE) {
1218 retval = get_gpt_last_partition_end(fd, endofpart);
1219 break;
1220 }
1221 /* check the last used lba for the current partition */
1222 curr_part_end = __le32_to_cpu(part->first_sect_lba) +
1223 __le32_to_cpu(part->blocks_num);
1224 if (curr_part_end > *endofpart)
1225 *endofpart = curr_part_end;
1226
1227 part++;
1228 }
1229 } else {
1230 /* Unknown partition table */
1231 retval = -1;
1232 }
1233 abort:
1234 return retval;
1235 }
1236
1237 int check_partitions(int fd, char *dname, unsigned long long freesize,
1238 unsigned long long size)
1239 {
1240 /*
1241 * Check where the last partition ends
1242 */
1243 unsigned long long endofpart;
1244 int ret;
1245
1246 if ((ret = get_last_partition_end(fd, &endofpart)) > 0) {
1247 /* There appears to be a partition table here */
1248 if (freesize == 0) {
1249 /* partitions will not be visible in new device */
1250 fprintf(stderr,
1251 Name ": partition table exists on %s but will be lost or\n"
1252 " meaningless after creating array\n",
1253 dname);
1254 return 1;
1255 } else if (endofpart > freesize) {
1256 /* last partition overlaps metadata */
1257 fprintf(stderr,
1258 Name ": metadata will over-write last partition on %s.\n",
1259 dname);
1260 return 1;
1261 } else if (size && endofpart > size) {
1262 /* partitions will be truncated in new device */
1263 fprintf(stderr,
1264 Name ": array size is too small to cover all partitions on %s.\n",
1265 dname);
1266 return 1;
1267 }
1268 }
1269 return 0;
1270 }
1271
1272 void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk)
1273 {
1274 int d;
1275
1276 ioctl(mdfd, GET_ARRAY_INFO, ainf);
1277 for (d = 0 ; d < 1024 ; d++) {
1278 if (ioctl(mdfd, GET_DISK_INFO, disk) == 0 &&
1279 (disk->major || disk->minor))
1280 return;
1281 }
1282 }
1283
1284 int open_container(int fd)
1285 {
1286 /* 'fd' is a block device. Find out if it is in use
1287 * by a container, and return an open fd on that container.
1288 */
1289 char path[256];
1290 char *e;
1291 DIR *dir;
1292 struct dirent *de;
1293 int dfd, n;
1294 char buf[200];
1295 int major, minor;
1296 struct stat st;
1297
1298 if (fstat(fd, &st) != 0)
1299 return -1;
1300 sprintf(path, "/sys/dev/block/%d:%d/holders",
1301 (int)major(st.st_rdev), (int)minor(st.st_rdev));
1302 e = path + strlen(path);
1303
1304 dir = opendir(path);
1305 if (!dir)
1306 return -1;
1307 while ((de = readdir(dir))) {
1308 if (de->d_ino == 0)
1309 continue;
1310 if (de->d_name[0] == '.')
1311 continue;
1312 sprintf(e, "/%s/dev", de->d_name);
1313 dfd = open(path, O_RDONLY);
1314 if (dfd < 0)
1315 continue;
1316 n = read(dfd, buf, sizeof(buf));
1317 close(dfd);
1318 if (n <= 0 || (unsigned)n >= sizeof(buf))
1319 continue;
1320 buf[n] = 0;
1321 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
1322 continue;
1323 sprintf(buf, "%d:%d", major, minor);
1324 dfd = dev_open(buf, O_RDONLY);
1325 if (dfd >= 0) {
1326 closedir(dir);
1327 return dfd;
1328 }
1329 }
1330 closedir(dir);
1331 return -1;
1332 }
1333
1334 struct superswitch *version_to_superswitch(char *vers)
1335 {
1336 int i;
1337
1338 for (i = 0; superlist[i]; i++) {
1339 struct superswitch *ss = superlist[i];
1340
1341 if (strcmp(vers, ss->name) == 0)
1342 return ss;
1343 }
1344
1345 return NULL;
1346 }
1347
1348 int is_container_member(struct mdstat_ent *mdstat, char *container)
1349 {
1350 if (mdstat->metadata_version == NULL ||
1351 strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
1352 !is_subarray(mdstat->metadata_version+9) ||
1353 strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
1354 mdstat->metadata_version[10+strlen(container)] != '/')
1355 return 0;
1356
1357 return 1;
1358 }
1359
1360 int is_subarray_active(char *subarray, char *container)
1361 {
1362 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1363 struct mdstat_ent *ent;
1364
1365 for (ent = mdstat; ent; ent = ent->next)
1366 if (is_container_member(ent, container))
1367 if (strcmp(to_subarray(ent, container), subarray) == 0)
1368 break;
1369
1370 free_mdstat(mdstat);
1371
1372 return ent != NULL;
1373 }
1374
1375 /* open_subarray - opens a subarray in a container
1376 * @dev: container device name
1377 * @st: empty supertype
1378 * @quiet: block reporting errors flag
1379 *
1380 * On success returns an fd to a container and fills in *st
1381 */
1382 int open_subarray(char *dev, char *subarray, struct supertype *st, int quiet)
1383 {
1384 struct mdinfo *mdi;
1385 struct mdinfo *info;
1386 int fd, err = 1;
1387
1388 fd = open(dev, O_RDWR|O_EXCL);
1389 if (fd < 0) {
1390 if (!quiet)
1391 fprintf(stderr, Name ": Couldn't open %s, aborting\n",
1392 dev);
1393 return -1;
1394 }
1395
1396 st->devnum = fd2devnum(fd);
1397 if (st->devnum == NoMdDev) {
1398 if (!quiet)
1399 fprintf(stderr,
1400 Name ": Failed to determine device number for %s\n",
1401 dev);
1402 goto close_fd;
1403 }
1404
1405 mdi = sysfs_read(fd, st->devnum, GET_VERSION|GET_LEVEL);
1406 if (!mdi) {
1407 if (!quiet)
1408 fprintf(stderr, Name ": Failed to read sysfs for %s\n",
1409 dev);
1410 goto close_fd;
1411 }
1412
1413 if (mdi->array.level != UnSet) {
1414 if (!quiet)
1415 fprintf(stderr, Name ": %s is not a container\n", dev);
1416 goto free_sysfs;
1417 }
1418
1419 st->ss = version_to_superswitch(mdi->text_version);
1420 if (!st->ss) {
1421 if (!quiet)
1422 fprintf(stderr,
1423 Name ": Operation not supported for %s metadata\n",
1424 mdi->text_version);
1425 goto free_sysfs;
1426 }
1427
1428 st->devname = devnum2devname(st->devnum);
1429 if (!st->devname) {
1430 if (!quiet)
1431 fprintf(stderr, Name ": Failed to allocate device name\n");
1432 goto free_sysfs;
1433 }
1434
1435 if (!st->ss->load_container) {
1436 if (!quiet)
1437 fprintf(stderr, Name ": %s is not a container\n", dev);
1438 goto free_name;
1439 }
1440
1441 if (st->ss->load_container(st, fd, NULL)) {
1442 if (!quiet)
1443 fprintf(stderr, Name ": Failed to load metadata for %s\n",
1444 dev);
1445 goto free_name;
1446 }
1447
1448 info = st->ss->container_content(st, subarray);
1449 if (!info) {
1450 if (!quiet)
1451 fprintf(stderr, Name ": Failed to find subarray-%s in %s\n",
1452 subarray, dev);
1453 goto free_super;
1454 }
1455 free(info);
1456
1457 err = 0;
1458
1459 free_super:
1460 if (err)
1461 st->ss->free_super(st);
1462 free_name:
1463 if (err)
1464 free(st->devname);
1465 free_sysfs:
1466 sysfs_free(mdi);
1467 close_fd:
1468 if (err)
1469 close(fd);
1470
1471 if (err)
1472 return -1;
1473 else
1474 return fd;
1475 }
1476
1477 int add_disk(int mdfd, struct supertype *st,
1478 struct mdinfo *sra, struct mdinfo *info)
1479 {
1480 /* Add a device to an array, in one of 2 ways. */
1481 int rv;
1482 #ifndef MDASSEMBLE
1483 if (st->ss->external) {
1484 if (info->disk.state & (1<<MD_DISK_SYNC))
1485 info->recovery_start = MaxSector;
1486 else
1487 info->recovery_start = 0;
1488 rv = sysfs_add_disk(sra, info, 0);
1489 if (! rv) {
1490 struct mdinfo *sd2;
1491 for (sd2 = sra->devs; sd2; sd2=sd2->next)
1492 if (sd2 == info)
1493 break;
1494 if (sd2 == NULL) {
1495 sd2 = malloc(sizeof(*sd2));
1496 *sd2 = *info;
1497 sd2->next = sra->devs;
1498 sra->devs = sd2;
1499 }
1500 }
1501 } else
1502 #endif
1503 rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk);
1504 return rv;
1505 }
1506
1507 int remove_disk(int mdfd, struct supertype *st,
1508 struct mdinfo *sra, struct mdinfo *info)
1509 {
1510 int rv;
1511 /* Remove the disk given by 'info' from the array */
1512 #ifndef MDASSEMBLE
1513 if (st->ss->external)
1514 rv = sysfs_set_str(sra, info, "slot", "none");
1515 else
1516 #endif
1517 rv = ioctl(mdfd, HOT_REMOVE_DISK, makedev(info->disk.major,
1518 info->disk.minor));
1519 return rv;
1520 }
1521
1522 int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
1523 {
1524 /* Initialise kernel's knowledge of array.
1525 * This varies between externally managed arrays
1526 * and older kernels
1527 */
1528 int vers = md_get_version(mdfd);
1529 int rv;
1530
1531 #ifndef MDASSEMBLE
1532 if (st->ss->external)
1533 rv = sysfs_set_array(info, vers);
1534 else
1535 #endif
1536 if ((vers % 100) >= 1) { /* can use different versions */
1537 mdu_array_info_t inf;
1538 memset(&inf, 0, sizeof(inf));
1539 inf.major_version = info->array.major_version;
1540 inf.minor_version = info->array.minor_version;
1541 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
1542 } else
1543 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
1544 return rv;
1545 }
1546
1547 unsigned long long min_recovery_start(struct mdinfo *array)
1548 {
1549 /* find the minimum recovery_start in an array for metadata
1550 * formats that only record per-array recovery progress instead
1551 * of per-device
1552 */
1553 unsigned long long recovery_start = MaxSector;
1554 struct mdinfo *d;
1555
1556 for (d = array->devs; d; d = d->next)
1557 recovery_start = min(recovery_start, d->recovery_start);
1558
1559 return recovery_start;
1560 }
1561
1562 int mdmon_pid(int devnum)
1563 {
1564 char path[100];
1565 char pid[10];
1566 int fd;
1567 int n;
1568 char *devname = devnum2devname(devnum);
1569
1570 sprintf(path, "%s/%s.pid", MDMON_DIR, devname);
1571 free(devname);
1572
1573 fd = open(path, O_RDONLY | O_NOATIME, 0);
1574
1575 if (fd < 0)
1576 return -1;
1577 n = read(fd, pid, 9);
1578 close(fd);
1579 if (n <= 0)
1580 return -1;
1581 return atoi(pid);
1582 }
1583
1584 int mdmon_running(int devnum)
1585 {
1586 int pid = mdmon_pid(devnum);
1587 if (pid <= 0)
1588 return 0;
1589 if (kill(pid, 0) == 0)
1590 return 1;
1591 return 0;
1592 }
1593
1594 int start_mdmon(int devnum)
1595 {
1596 int i, skipped;
1597 int len;
1598 pid_t pid;
1599 int status;
1600 char pathbuf[1024];
1601 char *paths[4] = {
1602 pathbuf,
1603 "/sbin/mdmon",
1604 "mdmon",
1605 NULL
1606 };
1607
1608 if (check_env("MDADM_NO_MDMON"))
1609 return 0;
1610
1611 len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf)-1);
1612 if (len > 0) {
1613 char *sl;
1614 pathbuf[len] = 0;
1615 sl = strrchr(pathbuf, '/');
1616 if (sl)
1617 sl++;
1618 else
1619 sl = pathbuf;
1620 strcpy(sl, "mdmon");
1621 } else
1622 pathbuf[0] = '\0';
1623
1624 switch(fork()) {
1625 case 0:
1626 /* FIXME yuk. CLOSE_EXEC?? */
1627 skipped = 0;
1628 for (i=3; skipped < 20; i++)
1629 if (close(i) < 0)
1630 skipped++;
1631 else
1632 skipped = 0;
1633
1634 for (i=0; paths[i]; i++)
1635 if (paths[i][0]) {
1636 if (__offroot) {
1637 execl(paths[i], "mdmon", "--offroot",
1638 devnum2devname(devnum),
1639 NULL);
1640 } else {
1641 execl(paths[i], "mdmon",
1642 devnum2devname(devnum),
1643 NULL);
1644 }
1645 }
1646 exit(1);
1647 case -1: fprintf(stderr, Name ": cannot run mdmon. "
1648 "Array remains readonly\n");
1649 return -1;
1650 default: /* parent - good */
1651 pid = wait(&status);
1652 if (pid < 0 || status != 0)
1653 return -1;
1654 }
1655 return 0;
1656 }
1657
1658 int check_env(char *name)
1659 {
1660 char *val = getenv(name);
1661
1662 if (val && atoi(val) == 1)
1663 return 1;
1664
1665 return 0;
1666 }
1667
1668 __u32 random32(void)
1669 {
1670 __u32 rv;
1671 int rfd = open("/dev/urandom", O_RDONLY);
1672 if (rfd < 0 || read(rfd, &rv, 4) != 4)
1673 rv = random();
1674 if (rfd >= 0)
1675 close(rfd);
1676 return rv;
1677 }
1678
1679 #ifndef MDASSEMBLE
1680 int flush_metadata_updates(struct supertype *st)
1681 {
1682 int sfd;
1683 if (!st->updates) {
1684 st->update_tail = NULL;
1685 return -1;
1686 }
1687
1688 sfd = connect_monitor(devnum2devname(st->container_dev));
1689 if (sfd < 0)
1690 return -1;
1691
1692 while (st->updates) {
1693 struct metadata_update *mu = st->updates;
1694 st->updates = mu->next;
1695
1696 send_message(sfd, mu, 0);
1697 wait_reply(sfd, 0);
1698 free(mu->buf);
1699 free(mu);
1700 }
1701 ack(sfd, 0);
1702 wait_reply(sfd, 0);
1703 close(sfd);
1704 st->update_tail = NULL;
1705 return 0;
1706 }
1707
1708 void append_metadata_update(struct supertype *st, void *buf, int len)
1709 {
1710
1711 struct metadata_update *mu = malloc(sizeof(*mu));
1712
1713 mu->buf = buf;
1714 mu->len = len;
1715 mu->space = NULL;
1716 mu->space_list = NULL;
1717 mu->next = NULL;
1718 *st->update_tail = mu;
1719 st->update_tail = &mu->next;
1720 }
1721 #endif /* MDASSEMBLE */
1722
1723 #ifdef __TINYC__
1724 /* tinyc doesn't optimize this check in ioctl.h out ... */
1725 unsigned int __invalid_size_argument_for_IOC = 0;
1726 #endif
1727
1728 int experimental(void)
1729 {
1730 if (check_env("MDADM_EXPERIMENTAL"))
1731 return 1;
1732 else {
1733 fprintf(stderr, Name ": To use this feature MDADM_EXPERIMENTAL"
1734 " environment variable has to be defined.\n");
1735 return 0;
1736 }
1737 }
1738
1739 /* Pick all spares matching given criteria from a container
1740 * if min_size == 0 do not check size
1741 * if domlist == NULL do not check domains
1742 * if spare_group given add it to domains of each spare
1743 * metadata allows to test domains using metadata of destination array */
1744 struct mdinfo *container_choose_spares(struct supertype *st,
1745 unsigned long long min_size,
1746 struct domainlist *domlist,
1747 char *spare_group,
1748 const char *metadata, int get_one)
1749 {
1750 struct mdinfo *d, **dp, *disks = NULL;
1751
1752 /* get list of all disks in container */
1753 if (st->ss->getinfo_super_disks)
1754 disks = st->ss->getinfo_super_disks(st);
1755
1756 if (!disks)
1757 return disks;
1758 /* find spare devices on the list */
1759 dp = &disks->devs;
1760 disks->array.spare_disks = 0;
1761 while (*dp) {
1762 int found = 0;
1763 d = *dp;
1764 if (d->disk.state == 0) {
1765 /* check if size is acceptable */
1766 unsigned long long dev_size;
1767 dev_t dev = makedev(d->disk.major,d->disk.minor);
1768
1769 if (!min_size ||
1770 (dev_size_from_id(dev, &dev_size) &&
1771 dev_size >= min_size))
1772 found = 1;
1773 /* check if domain matches */
1774 if (found && domlist) {
1775 struct dev_policy *pol = devnum_policy(dev);
1776 if (spare_group)
1777 pol_add(&pol, pol_domain,
1778 spare_group, NULL);
1779 if (domain_test(domlist, pol, metadata) != 1)
1780 found = 0;
1781 dev_policy_free(pol);
1782 }
1783 }
1784 if (found) {
1785 dp = &d->next;
1786 disks->array.spare_disks++;
1787 if (get_one) {
1788 sysfs_free(*dp);
1789 d->next = NULL;
1790 }
1791 } else {
1792 *dp = d->next;
1793 d->next = NULL;
1794 sysfs_free(d);
1795 }
1796 }
1797 return disks;
1798 }