Consistently print program Name and __func__ in debug messages.
[thirdparty/mdadm.git] / util.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 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 <sys/resource.h>
32 #include <sys/vfs.h>
33 #include <linux/magic.h>
34 #include <ctype.h>
35 #include <dirent.h>
36 #include <signal.h>
37
38 /*
39 * following taken from linux/blkpg.h because they aren't
40 * anywhere else and it isn't safe to #include linux/ * stuff.
41 */
42
43 #define BLKPG _IO(0x12,105)
44
45 /* The argument structure */
46 struct blkpg_ioctl_arg {
47 int op;
48 int flags;
49 int datalen;
50 void *data;
51 };
52
53 /* The subfunctions (for the op field) */
54 #define BLKPG_ADD_PARTITION 1
55 #define BLKPG_DEL_PARTITION 2
56
57 /* Sizes of name fields. Unused at present. */
58 #define BLKPG_DEVNAMELTH 64
59 #define BLKPG_VOLNAMELTH 64
60
61 /* The data structure for ADD_PARTITION and DEL_PARTITION */
62 struct blkpg_partition {
63 long long start; /* starting offset in bytes */
64 long long length; /* length in bytes */
65 int pno; /* partition number */
66 char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2,
67 to be used in kernel messages */
68 char volname[BLKPG_VOLNAMELTH]; /* volume label */
69 };
70
71 #include "part.h"
72
73 /* Force a compilation error if condition is true */
74 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
75
76 /* Force a compilation error if condition is true, but also produce a
77 result (of value 0 and type size_t), so the expression can be used
78 e.g. in a structure initializer (or where-ever else comma expressions
79 aren't permitted). */
80 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
81
82 /*
83 * Parse a 128 bit uuid in 4 integers
84 * format is 32 hexx nibbles with options :.<space> separator
85 * If not exactly 32 hex digits are found, return 0
86 * else return 1
87 */
88 int parse_uuid(char *str, int uuid[4])
89 {
90 int hit = 0; /* number of Hex digIT */
91 int i;
92 char c;
93 for (i = 0; i < 4; i++)
94 uuid[i] = 0;
95
96 while ((c = *str++) != 0) {
97 int n;
98 if (c >= '0' && c <= '9')
99 n = c-'0';
100 else if (c >= 'a' && c <= 'f')
101 n = 10 + c - 'a';
102 else if (c >= 'A' && c <= 'F')
103 n = 10 + c - 'A';
104 else if (strchr(":. -", c))
105 continue;
106 else return 0;
107
108 if (hit<32) {
109 uuid[hit/8] <<= 4;
110 uuid[hit/8] += n;
111 }
112 hit++;
113 }
114 if (hit == 32)
115 return 1;
116 return 0;
117 }
118
119 /*
120 * Get the md version number.
121 * We use the RAID_VERSION ioctl if it is supported
122 * If not, but we have a block device with major '9', we assume
123 * 0.36.0
124 *
125 * Return version number as 24 but number - assume version parts
126 * always < 255
127 */
128
129 int md_get_version(int fd)
130 {
131 struct stat stb;
132 mdu_version_t vers;
133
134 if (fstat(fd, &stb)<0)
135 return -1;
136 if ((S_IFMT&stb.st_mode) != S_IFBLK)
137 return -1;
138
139 if (ioctl(fd, RAID_VERSION, &vers) == 0)
140 return (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
141 if (errno == EACCES)
142 return -1;
143 if (major(stb.st_rdev) == MD_MAJOR)
144 return (3600);
145 return -1;
146 }
147
148 int get_linux_version()
149 {
150 struct utsname name;
151 char *cp;
152 int a = 0, b = 0,c = 0;
153 if (uname(&name) <0)
154 return -1;
155
156 cp = name.release;
157 a = strtoul(cp, &cp, 10);
158 if (*cp == '.')
159 b = strtoul(cp+1, &cp, 10);
160 if (*cp == '.')
161 c = strtoul(cp+1, &cp, 10);
162
163 return (a*1000000)+(b*1000)+c;
164 }
165
166 #ifndef MDASSEMBLE
167 int mdadm_version(char *version)
168 {
169 int a, b, c;
170 char *cp;
171
172 if (!version)
173 version = Version;
174
175 cp = strchr(version, '-');
176 if (!cp || *(cp+1) != ' ' || *(cp+2) != 'v')
177 return -1;
178 cp += 3;
179 a = strtoul(cp, &cp, 10);
180 if (*cp != '.')
181 return -1;
182 b = strtoul(cp+1, &cp, 10);
183 if (*cp == '.')
184 c = strtoul(cp+1, &cp, 10);
185 else
186 c = 0;
187 if (*cp != ' ' && *cp != '-')
188 return -1;
189 return (a*1000000)+(b*1000)+c;
190 }
191
192 unsigned long long parse_size(char *size)
193 {
194 /* parse 'size' which should be a number optionally
195 * followed by 'K', 'M', or 'G'.
196 * Without a suffix, K is assumed.
197 * Number returned is in sectors (half-K)
198 * INVALID_SECTORS returned on error.
199 */
200 char *c;
201 long long s = strtoll(size, &c, 10);
202 if (s > 0) {
203 switch (*c) {
204 case 'K':
205 c++;
206 default:
207 s *= 2;
208 break;
209 case 'M':
210 c++;
211 s *= 1024 * 2;
212 break;
213 case 'G':
214 c++;
215 s *= 1024 * 1024 * 2;
216 break;
217 case 's': /* sectors */
218 c++;
219 break;
220 }
221 } else
222 s = INVALID_SECTORS;
223 if (*c)
224 s = INVALID_SECTORS;
225 return s;
226 }
227
228 int parse_layout_10(char *layout)
229 {
230 int copies, rv;
231 char *cp;
232 /* Parse the layout string for raid10 */
233 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
234 if ((layout[0] != 'n' && layout[0] != 'f' && layout[0] != 'o') ||
235 (copies = strtoul(layout+1, &cp, 10)) < 1 ||
236 copies > 200 ||
237 *cp)
238 return -1;
239 if (layout[0] == 'n')
240 rv = 256 + copies;
241 else if (layout[0] == 'o')
242 rv = 0x10000 + (copies<<8) + 1;
243 else
244 rv = 1 + (copies<<8);
245 return rv;
246 }
247
248 int parse_layout_faulty(char *layout)
249 {
250 /* Parse the layout string for 'faulty' */
251 int ln = strcspn(layout, "0123456789");
252 char *m = xstrdup(layout);
253 int mode;
254 m[ln] = 0;
255 mode = map_name(faultylayout, m);
256 if (mode == UnSet)
257 return -1;
258
259 return mode | (atoi(layout+ln)<< ModeShift);
260 }
261
262 long parse_num(char *num)
263 {
264 /* Either return a valid number, or -1 */
265 char *c;
266 long rv = strtol(num, &c, 10);
267 if (rv < 0 || *c || !num[0])
268 return -1;
269 else
270 return rv;
271 }
272 #endif
273
274 void remove_partitions(int fd)
275 {
276 /* remove partitions from this block devices.
277 * This is used for components added to an array
278 */
279 #ifdef BLKPG_DEL_PARTITION
280 struct blkpg_ioctl_arg a;
281 struct blkpg_partition p;
282
283 a.op = BLKPG_DEL_PARTITION;
284 a.data = (void*)&p;
285 a.datalen = sizeof(p);
286 a.flags = 0;
287 memset(a.data, 0, a.datalen);
288 for (p.pno = 0; p.pno < 16; p.pno++)
289 ioctl(fd, BLKPG, &a);
290 #endif
291 }
292
293 int test_partition(int fd)
294 {
295 /* Check if fd is a whole-disk or a partition.
296 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
297 * will return ENXIO on an invalid partition number.
298 */
299 struct blkpg_ioctl_arg a;
300 struct blkpg_partition p;
301 a.op = BLKPG_DEL_PARTITION;
302 a.data = (void*)&p;
303 a.datalen = sizeof(p);
304 a.flags = 0;
305 memset(a.data, 0, a.datalen);
306 p.pno = 1<<30;
307 if (ioctl(fd, BLKPG, &a) == 0)
308 /* Very unlikely, but not a partition */
309 return 0;
310 if (errno == ENXIO || errno == ENOTTY)
311 /* not a partition */
312 return 0;
313
314 return 1;
315 }
316
317 int test_partition_from_id(dev_t id)
318 {
319 char buf[20];
320 int fd, rv;
321
322 sprintf(buf, "%d:%d", major(id), minor(id));
323 fd = dev_open(buf, O_RDONLY);
324 if (fd < 0)
325 return -1;
326 rv = test_partition(fd);
327 close(fd);
328 return rv;
329 }
330
331 int enough(int level, int raid_disks, int layout, int clean, char *avail)
332 {
333 int copies, first;
334 int i;
335 int avail_disks = 0;
336
337 for (i = 0; i < raid_disks; i++)
338 avail_disks += !!avail[i];
339
340 switch (level) {
341 case 10:
342 /* This is the tricky one - we need to check
343 * which actual disks are present.
344 */
345 copies = (layout&255)* ((layout>>8) & 255);
346 first = 0;
347 do {
348 /* there must be one of the 'copies' form 'first' */
349 int n = copies;
350 int cnt = 0;
351 int this = first;
352 while (n--) {
353 if (avail[this])
354 cnt++;
355 this = (this+1) % raid_disks;
356 }
357 if (cnt == 0)
358 return 0;
359 first = (first+(layout&255)) % raid_disks;
360 } while (first != 0);
361 return 1;
362
363 case LEVEL_MULTIPATH:
364 return avail_disks>= 1;
365 case LEVEL_LINEAR:
366 case 0:
367 return avail_disks == raid_disks;
368 case 1:
369 return avail_disks >= 1;
370 case 4:
371 case 5:
372 if (clean)
373 return avail_disks >= raid_disks-1;
374 else
375 return avail_disks >= raid_disks;
376 case 6:
377 if (clean)
378 return avail_disks >= raid_disks-2;
379 else
380 return avail_disks >= raid_disks;
381 default:
382 return 0;
383 }
384 }
385
386 int enough_fd(int fd)
387 {
388 struct mdu_array_info_s array;
389 struct mdu_disk_info_s disk;
390 int i, rv;
391 char *avail;
392
393 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0 ||
394 array.raid_disks <= 0)
395 return 0;
396 avail = xcalloc(array.raid_disks, 1);
397 for (i = 0; i < MAX_DISKS && array.nr_disks > 0; i++) {
398 disk.number = i;
399 if (ioctl(fd, GET_DISK_INFO, &disk) != 0)
400 continue;
401 if (disk.major == 0 && disk.minor == 0)
402 continue;
403 array.nr_disks--;
404
405 if (! (disk.state & (1<<MD_DISK_SYNC)))
406 continue;
407 if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks)
408 continue;
409 avail[disk.raid_disk] = 1;
410 }
411 /* This is used on an active array, so assume it is clean */
412 rv = enough(array.level, array.raid_disks, array.layout,
413 1, avail);
414 free(avail);
415 return rv;
416 }
417
418 const int uuid_zero[4] = { 0, 0, 0, 0 };
419
420 int same_uuid(int a[4], int b[4], int swapuuid)
421 {
422 if (swapuuid) {
423 /* parse uuids are hostendian.
424 * uuid's from some superblocks are big-ending
425 * if there is a difference, we need to swap..
426 */
427 unsigned char *ac = (unsigned char *)a;
428 unsigned char *bc = (unsigned char *)b;
429 int i;
430 for (i = 0; i < 16; i += 4) {
431 if (ac[i+0] != bc[i+3] ||
432 ac[i+1] != bc[i+2] ||
433 ac[i+2] != bc[i+1] ||
434 ac[i+3] != bc[i+0])
435 return 0;
436 }
437 return 1;
438 } else {
439 if (a[0]==b[0] &&
440 a[1]==b[1] &&
441 a[2]==b[2] &&
442 a[3]==b[3])
443 return 1;
444 return 0;
445 }
446 }
447
448 void copy_uuid(void *a, int b[4], int swapuuid)
449 {
450 if (swapuuid) {
451 /* parse uuids are hostendian.
452 * uuid's from some superblocks are big-ending
453 * if there is a difference, we need to swap..
454 */
455 unsigned char *ac = (unsigned char *)a;
456 unsigned char *bc = (unsigned char *)b;
457 int i;
458 for (i = 0; i < 16; i += 4) {
459 ac[i+0] = bc[i+3];
460 ac[i+1] = bc[i+2];
461 ac[i+2] = bc[i+1];
462 ac[i+3] = bc[i+0];
463 }
464 } else
465 memcpy(a, b, 16);
466 }
467
468 char *__fname_from_uuid(int id[4], int swap, char *buf, char sep)
469 {
470 int i, j;
471 char uuid[16];
472 char *c = buf;
473 strcpy(c, "UUID-");
474 c += strlen(c);
475 copy_uuid(uuid, id, swap);
476 for (i = 0; i < 4; i++) {
477 if (i)
478 *c++ = sep;
479 for (j = 3; j >= 0; j--) {
480 sprintf(c,"%02x", (unsigned char) uuid[j+4*i]);
481 c+= 2;
482 }
483 }
484 return buf;
485
486 }
487
488 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
489 {
490 // dirty hack to work around an issue with super1 superblocks...
491 // super1 superblocks need swapuuid set in order for assembly to
492 // work, but can't have it set if we want this printout to match
493 // all the other uuid printouts in super1.c, so we force swapuuid
494 // to 1 to make our printout match the rest of super1
495 return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep);
496 }
497
498 #ifndef MDASSEMBLE
499 int check_ext2(int fd, char *name)
500 {
501 /*
502 * Check for an ext2fs file system.
503 * Superblock is always 1K at 1K offset
504 *
505 * s_magic is le16 at 56 == 0xEF53
506 * report mtime - le32 at 44
507 * blocks - le32 at 4
508 * logblksize - le32 at 24
509 */
510 unsigned char sb[1024];
511 time_t mtime;
512 unsigned long long size;
513 int bsize;
514 if (lseek(fd, 1024,0)!= 1024)
515 return 0;
516 if (read(fd, sb, 1024)!= 1024)
517 return 0;
518 if (sb[56] != 0x53 || sb[57] != 0xef)
519 return 0;
520
521 mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
522 bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
523 size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
524 size <<= bsize;
525 pr_err("%s appears to contain an ext2fs file system\n",
526 name);
527 cont_err("size=%lluK mtime=%s", size, ctime(&mtime));
528 return 1;
529 }
530
531 int check_reiser(int fd, char *name)
532 {
533 /*
534 * superblock is at 64K
535 * size is 1024;
536 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
537 *
538 */
539 unsigned char sb[1024];
540 unsigned long long size;
541 if (lseek(fd, 64*1024, 0) != 64*1024)
542 return 0;
543 if (read(fd, sb, 1024) != 1024)
544 return 0;
545 if (strncmp((char*)sb+52, "ReIsErFs",8) != 0 &&
546 strncmp((char*)sb+52, "ReIsEr2Fs",9) != 0)
547 return 0;
548 pr_err("%s appears to contain a reiserfs file system\n",name);
549 size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
550 cont_err("size = %lluK\n", size*4);
551
552 return 1;
553 }
554
555 int check_raid(int fd, char *name)
556 {
557 struct mdinfo info;
558 time_t crtime;
559 char *level;
560 struct supertype *st = guess_super(fd);
561
562 if (!st)
563 return 0;
564 st->ss->load_super(st, fd, name);
565 /* Looks like a raid array .. */
566 pr_err("%s appears to be part of a raid array:\n",
567 name);
568 st->ss->getinfo_super(st, &info, NULL);
569 st->ss->free_super(st);
570 crtime = info.array.ctime;
571 level = map_num(pers, info.array.level);
572 if (!level) level = "-unknown-";
573 cont_err("level=%s devices=%d ctime=%s",
574 level, info.array.raid_disks, ctime(&crtime));
575 return 1;
576 }
577
578 int ask(char *mesg)
579 {
580 char *add = "";
581 int i;
582 for (i = 0; i < 5; i++) {
583 char buf[100];
584 fprintf(stderr, "%s%s", mesg, add);
585 fflush(stderr);
586 if (fgets(buf, 100, stdin)==NULL)
587 return 0;
588 if (buf[0]=='y' || buf[0]=='Y')
589 return 1;
590 if (buf[0]=='n' || buf[0]=='N')
591 return 0;
592 add = "(y/n) ";
593 }
594 pr_err("assuming 'no'\n");
595 return 0;
596 }
597 #endif /* MDASSEMBLE */
598
599 int is_standard(char *dev, int *nump)
600 {
601 /* tests if dev is a "standard" md dev name.
602 * i.e if the last component is "/dNN" or "/mdNN",
603 * where NN is a string of digits
604 * Returns 1 if a partitionable standard,
605 * -1 if non-partitonable,
606 * 0 if not a standard name.
607 */
608 char *d = strrchr(dev, '/');
609 int type = 0;
610 int num;
611 if (!d)
612 return 0;
613 if (strncmp(d, "/d",2) == 0)
614 d += 2, type = 1; /* /dev/md/dN{pM} */
615 else if (strncmp(d, "/md_d", 5) == 0)
616 d += 5, type = 1; /* /dev/md_dN{pM} */
617 else if (strncmp(d, "/md", 3) == 0)
618 d += 3, type = -1; /* /dev/mdN */
619 else if (d-dev > 3 && strncmp(d-2, "md/", 3) == 0)
620 d += 1, type = -1; /* /dev/md/N */
621 else
622 return 0;
623 if (!*d)
624 return 0;
625 num = atoi(d);
626 while (isdigit(*d))
627 d++;
628 if (*d)
629 return 0;
630 if (nump) *nump = num;
631
632 return type;
633 }
634
635 unsigned long calc_csum(void *super, int bytes)
636 {
637 unsigned long long newcsum = 0;
638 int i;
639 unsigned int csum;
640 unsigned int *superc = (unsigned int*) super;
641
642 for(i = 0; i < bytes/4; i++)
643 newcsum += superc[i];
644 csum = (newcsum& 0xffffffff) + (newcsum>>32);
645 #ifdef __alpha__
646 /* The in-kernel checksum calculation is always 16bit on
647 * the alpha, though it is 32 bit on i386...
648 * I wonder what it is elsewhere... (it uses an API in
649 * a way that it shouldn't).
650 */
651 csum = (csum & 0xffff) + (csum >> 16);
652 csum = (csum & 0xffff) + (csum >> 16);
653 #endif
654 return csum;
655 }
656
657 #ifndef MDASSEMBLE
658 char *human_size(long long bytes)
659 {
660 static char buf[30];
661
662 /* We convert bytes to either centi-M{ega,ibi}bytes or
663 * centi-G{igi,ibi}bytes, with appropriate rounding,
664 * and then print 1/100th of those as a decimal.
665 * We allow upto 2048Megabytes before converting to
666 * gigabytes, as that shows more precision and isn't
667 * too large a number.
668 * Terabytes are not yet handled.
669 */
670
671 if (bytes < 5000*1024)
672 buf[0] = 0;
673 else if (bytes < 2*1024LL*1024LL*1024LL) {
674 long cMiB = (bytes * 200LL / (1LL<<20) + 1) / 2;
675 long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2;
676 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
677 cMiB/100 , cMiB % 100,
678 cMB/100, cMB % 100);
679 } else {
680 long cGiB = (bytes * 200LL / (1LL<<30) +1) / 2;
681 long cGB = (bytes / (1000000000LL/200LL ) +1) /2;
682 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
683 cGiB/100 , cGiB % 100,
684 cGB/100, cGB % 100);
685 }
686 return buf;
687 }
688
689 char *human_size_brief(long long bytes, int prefix)
690 {
691 static char buf[30];
692
693 /* We convert bytes to either centi-M{ega,ibi}bytes or
694 * centi-G{igi,ibi}bytes, with appropriate rounding,
695 * and then print 1/100th of those as a decimal.
696 * We allow upto 2048Megabytes before converting to
697 * gigabytes, as that shows more precision and isn't
698 * too large a number.
699 * Terabytes are not yet handled.
700 *
701 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
702 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
703 */
704
705 if (bytes < 5000*1024)
706 buf[0] = 0;
707 else if (prefix == IEC) {
708 if (bytes < 2*1024LL*1024LL*1024LL) {
709 long cMiB = (bytes * 200LL / (1LL<<20) +1) /2;
710 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
711 cMiB/100 , cMiB % 100);
712 } else {
713 long cGiB = (bytes * 200LL / (1LL<<30) +1) /2;
714 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
715 cGiB/100 , cGiB % 100);
716 }
717 }
718 else if (prefix == JEDEC) {
719 if (bytes < 2*1024LL*1024LL*1024LL) {
720 long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2;
721 snprintf(buf, sizeof(buf), "%ld.%02ldMB",
722 cMB/100, cMB % 100);
723 } else {
724 long cGB = (bytes / (1000000000LL/200LL ) +1) /2;
725 snprintf(buf, sizeof(buf), "%ld.%02ldGB",
726 cGB/100 , cGB % 100);
727 }
728 }
729 else
730 buf[0] = 0;
731
732 return buf;
733 }
734
735 void print_r10_layout(int layout)
736 {
737 int near = layout & 255;
738 int far = (layout >> 8) & 255;
739 int offset = (layout&0x10000);
740 char *sep = "";
741
742 if (near != 1) {
743 printf("%s near=%d", sep, near);
744 sep = ",";
745 }
746 if (far != 1)
747 printf("%s %s=%d", sep, offset?"offset":"far", far);
748 if (near*far == 1)
749 printf("NO REDUNDANCY");
750 }
751 #endif
752
753 unsigned long long calc_array_size(int level, int raid_disks, int layout,
754 int chunksize, unsigned long long devsize)
755 {
756 if (level == 1)
757 return devsize;
758 devsize &= ~(unsigned long long)((chunksize>>9)-1);
759 return get_data_disks(level, layout, raid_disks) * devsize;
760 }
761
762 int get_data_disks(int level, int layout, int raid_disks)
763 {
764 int data_disks = 0;
765 switch (level) {
766 case 0: data_disks = raid_disks;
767 break;
768 case 1: data_disks = 1;
769 break;
770 case 4:
771 case 5: data_disks = raid_disks - 1;
772 break;
773 case 6: data_disks = raid_disks - 2;
774 break;
775 case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
776 break;
777 }
778
779 return data_disks;
780 }
781
782 int devnm2devid(char *devnm)
783 {
784 /* First look in /sys/block/$DEVNM/dev for %d:%d
785 * If that fails, try parsing out a number
786 */
787 char path[100];
788 char *ep;
789 int fd;
790 int mjr,mnr;
791
792 sprintf(path, "/sys/block/%s/dev", devnm);
793 fd = open(path, O_RDONLY);
794 if (fd >= 0) {
795 char buf[20];
796 int n = read(fd, buf, sizeof(buf));
797 close(fd);
798 if (n > 0)
799 buf[n] = 0;
800 if (n > 0 && sscanf(buf, "%d:%d\n", &mjr, &mnr) == 2)
801 return makedev(mjr, mnr);
802 }
803 if (strncmp(devnm, "md_d", 4) == 0 &&
804 isdigit(devnm[4]) &&
805 (mnr = strtoul(devnm+4, &ep, 10)) >= 0 &&
806 ep > devnm && *ep == 0)
807 return makedev(get_mdp_major(), mnr << MdpMinorShift);
808
809 if (strncmp(devnm, "md", 2) == 0 &&
810 isdigit(devnm[2]) &&
811 (mnr = strtoul(devnm+2, &ep, 10)) >= 0 &&
812 ep > devnm && *ep == 0)
813 return makedev(MD_MAJOR, mnr);
814
815 return 0;
816 }
817
818 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
819 char *get_md_name(char *devnm)
820 {
821 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
822 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
823
824 static char devname[50];
825 struct stat stb;
826 dev_t rdev = devnm2devid(devnm);
827 char *dn;
828
829 if (rdev == 0)
830 return 0;
831 if (strncmp(devnm, "md_", 3) == 0) {
832 snprintf(devname, sizeof(devname), "/dev/md/%s",
833 devnm + 3);
834 if (stat(devname, &stb) == 0
835 && (S_IFMT&stb.st_mode) == S_IFBLK
836 && (stb.st_rdev == rdev))
837 return devname;
838 }
839 snprintf(devname, sizeof(devname), "/dev/%s", devnm);
840 if (stat(devname, &stb) == 0
841 && (S_IFMT&stb.st_mode) == S_IFBLK
842 && (stb.st_rdev == rdev))
843 return devname;
844
845 snprintf(devname, sizeof(devname), "/dev/md/%s", devnm+2);
846 if (stat(devname, &stb) == 0
847 && (S_IFMT&stb.st_mode) == S_IFBLK
848 && (stb.st_rdev == rdev))
849 return devname;
850
851 dn = map_dev(major(rdev), minor(rdev), 0);
852 if (dn)
853 return dn;
854 snprintf(devname, sizeof(devname), "/dev/.tmp.%s", devnm);
855 if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
856 if (errno != EEXIST)
857 return NULL;
858
859 if (stat(devname, &stb) == 0
860 && (S_IFMT&stb.st_mode) == S_IFBLK
861 && (stb.st_rdev == rdev))
862 return devname;
863 unlink(devname);
864 return NULL;
865 }
866
867 void put_md_name(char *name)
868 {
869 if (strncmp(name, "/dev/.tmp.md", 12) == 0)
870 unlink(name);
871 }
872 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
873
874 int get_maj_min(char *dev, int *major, int *minor)
875 {
876 char *e;
877 *major = strtoul(dev, &e, 0);
878 return (e > dev && *e == ':' && e[1] &&
879 (*minor = strtoul(e+1, &e, 0)) >= 0 &&
880 *e == 0);
881 }
882
883 int dev_open(char *dev, int flags)
884 {
885 /* like 'open', but if 'dev' matches %d:%d, create a temp
886 * block device and open that
887 */
888 int fd = -1;
889 char devname[32];
890 int major;
891 int minor;
892
893 if (!dev) return -1;
894 flags |= O_DIRECT;
895
896 if (get_maj_min(dev, &major, &minor)) {
897 snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d",
898 (int)getpid(), major, minor);
899 if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) {
900 fd = open(devname, flags);
901 unlink(devname);
902 }
903 if (fd < 0) {
904 /* Try /tmp as /dev appear to be read-only */
905 snprintf(devname, sizeof(devname), "/tmp/.tmp.md.%d:%d:%d",
906 (int)getpid(), major, minor);
907 if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) {
908 fd = open(devname, flags);
909 unlink(devname);
910 }
911 }
912 } else
913 fd = open(dev, flags);
914 return fd;
915 }
916
917 int open_dev_flags(char *devnm, int flags)
918 {
919 int devid;
920 char buf[20];
921
922 devid = devnm2devid(devnm);
923 sprintf(buf, "%d:%d", major(devid), minor(devid));
924 return dev_open(buf, flags);
925 }
926
927 int open_dev(char *devnm)
928 {
929 return open_dev_flags(devnm, O_RDONLY);
930 }
931
932 int open_dev_excl(char *devnm)
933 {
934 char buf[20];
935 int i;
936 int flags = O_RDWR;
937 int devid = devnm2devid(devnm);
938 long delay = 1000;
939
940 sprintf(buf, "%d:%d", major(devid), minor(devid));
941 for (i = 0 ; i < 25 ; i++) {
942 int fd = dev_open(buf, flags|O_EXCL);
943 if (fd >= 0)
944 return fd;
945 if (errno == EACCES && flags == O_RDWR) {
946 flags = O_RDONLY;
947 continue;
948 }
949 if (errno != EBUSY)
950 return fd;
951 usleep(delay);
952 if (delay < 200000)
953 delay *= 2;
954 }
955 return -1;
956 }
957
958 int same_dev(char *one, char *two)
959 {
960 struct stat st1, st2;
961 if (stat(one, &st1) != 0)
962 return 0;
963 if (stat(two, &st2) != 0)
964 return 0;
965 if ((st1.st_mode & S_IFMT) != S_IFBLK)
966 return 0;
967 if ((st2.st_mode & S_IFMT) != S_IFBLK)
968 return 0;
969 return st1.st_rdev == st2.st_rdev;
970 }
971
972 void wait_for(char *dev, int fd)
973 {
974 int i;
975 struct stat stb_want;
976 long delay = 1000;
977
978 if (fstat(fd, &stb_want) != 0 ||
979 (stb_want.st_mode & S_IFMT) != S_IFBLK)
980 return;
981
982 for (i = 0 ; i < 25 ; i++) {
983 struct stat stb;
984 if (stat(dev, &stb) == 0 &&
985 (stb.st_mode & S_IFMT) == S_IFBLK &&
986 (stb.st_rdev == stb_want.st_rdev))
987 return;
988 usleep(delay);
989 if (delay < 200000)
990 delay *= 2;
991 }
992 if (i == 25)
993 dprintf("timeout waiting for %s\n", dev);
994 }
995
996 struct superswitch *superlist[] =
997 {
998 &super0, &super1,
999 &super_ddf, &super_imsm,
1000 &mbr, &gpt,
1001 NULL };
1002
1003 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1004
1005 struct supertype *super_by_fd(int fd, char **subarrayp)
1006 {
1007 mdu_array_info_t array;
1008 int vers;
1009 int minor;
1010 struct supertype *st = NULL;
1011 struct mdinfo *sra;
1012 char *verstr;
1013 char version[20];
1014 int i;
1015 char *subarray = NULL;
1016 char container[32] = "";
1017
1018 sra = sysfs_read(fd, NULL, GET_VERSION);
1019
1020 if (sra) {
1021 vers = sra->array.major_version;
1022 minor = sra->array.minor_version;
1023 verstr = sra->text_version;
1024 } else {
1025 if (ioctl(fd, GET_ARRAY_INFO, &array))
1026 array.major_version = array.minor_version = 0;
1027 vers = array.major_version;
1028 minor = array.minor_version;
1029 verstr = "";
1030 }
1031
1032 if (vers != -1) {
1033 sprintf(version, "%d.%d", vers, minor);
1034 verstr = version;
1035 }
1036 if (minor == -2 && is_subarray(verstr)) {
1037 char *dev = verstr+1;
1038
1039 subarray = strchr(dev, '/');
1040 if (subarray) {
1041 *subarray++ = '\0';
1042 subarray = xstrdup(subarray);
1043 }
1044 strcpy(container, dev);
1045 if (sra)
1046 sysfs_free(sra);
1047 sra = sysfs_read(-1, container, GET_VERSION);
1048 if (sra && sra->text_version[0])
1049 verstr = sra->text_version;
1050 else
1051 verstr = "-no-metadata-";
1052 }
1053
1054 for (i = 0; st == NULL && superlist[i] ; i++)
1055 st = superlist[i]->match_metadata_desc(verstr);
1056
1057 if (sra)
1058 sysfs_free(sra);
1059 if (st) {
1060 st->sb = NULL;
1061 if (subarrayp)
1062 *subarrayp = subarray;
1063 strcpy(st->container_devnm, container);
1064 strcpy(st->devnm, fd2devnm(fd));
1065 } else
1066 free(subarray);
1067
1068 return st;
1069 }
1070 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1071
1072 int dev_size_from_id(dev_t id, unsigned long long *size)
1073 {
1074 char buf[20];
1075 int fd;
1076
1077 sprintf(buf, "%d:%d", major(id), minor(id));
1078 fd = dev_open(buf, O_RDONLY);
1079 if (fd < 0)
1080 return 0;
1081 if (get_dev_size(fd, NULL, size)) {
1082 close(fd);
1083 return 1;
1084 }
1085 close(fd);
1086 return 0;
1087 }
1088
1089 struct supertype *dup_super(struct supertype *orig)
1090 {
1091 struct supertype *st;
1092
1093 if (!orig)
1094 return orig;
1095 st = xcalloc(1, sizeof(*st));
1096 st->ss = orig->ss;
1097 st->max_devs = orig->max_devs;
1098 st->minor_version = orig->minor_version;
1099 st->ignore_hw_compat = orig->ignore_hw_compat;
1100 st->data_offset = orig->data_offset;
1101 st->sb = NULL;
1102 st->info = NULL;
1103 return st;
1104 }
1105
1106 struct supertype *guess_super_type(int fd, enum guess_types guess_type)
1107 {
1108 /* try each load_super to find the best match,
1109 * and return the best superswitch
1110 */
1111 struct superswitch *ss;
1112 struct supertype *st;
1113 time_t besttime = 0;
1114 int bestsuper = -1;
1115 int i;
1116
1117 st = xcalloc(1, sizeof(*st));
1118 st->container_devnm[0] = 0;
1119
1120 for (i = 0 ; superlist[i]; i++) {
1121 int rv;
1122 ss = superlist[i];
1123 if (guess_type == guess_array && ss->add_to_super == NULL)
1124 continue;
1125 if (guess_type == guess_partitions && ss->add_to_super != NULL)
1126 continue;
1127 memset(st, 0, sizeof(*st));
1128 st->ignore_hw_compat = 1;
1129 rv = ss->load_super(st, fd, NULL);
1130 if (rv == 0) {
1131 struct mdinfo info;
1132 st->ss->getinfo_super(st, &info, NULL);
1133 if (bestsuper == -1 ||
1134 besttime < info.array.ctime) {
1135 bestsuper = i;
1136 besttime = info.array.ctime;
1137 }
1138 ss->free_super(st);
1139 }
1140 }
1141 if (bestsuper != -1) {
1142 int rv;
1143 memset(st, 0, sizeof(*st));
1144 st->ignore_hw_compat = 1;
1145 rv = superlist[bestsuper]->load_super(st, fd, NULL);
1146 if (rv == 0) {
1147 superlist[bestsuper]->free_super(st);
1148 return st;
1149 }
1150 }
1151 free(st);
1152 return NULL;
1153 }
1154
1155 /* Return size of device in bytes */
1156 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
1157 {
1158 unsigned long long ldsize;
1159 struct stat st;
1160
1161 if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
1162 ldsize = (unsigned long long)st.st_size;
1163 else
1164 #ifdef BLKGETSIZE64
1165 if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
1166 #endif
1167 {
1168 unsigned long dsize;
1169 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
1170 ldsize = dsize;
1171 ldsize <<= 9;
1172 } else {
1173 if (dname)
1174 pr_err("Cannot get size of %s: %s\b",
1175 dname, strerror(errno));
1176 return 0;
1177 }
1178 }
1179 *sizep = ldsize;
1180 return 1;
1181 }
1182
1183 /* Return true if this can only be a container, not a member device.
1184 * i.e. is and md device and size is zero
1185 */
1186 int must_be_container(int fd)
1187 {
1188 unsigned long long size;
1189 if (md_get_version(fd) < 0)
1190 return 0;
1191 if (get_dev_size(fd, NULL, &size) == 0)
1192 return 1;
1193 if (size == 0)
1194 return 1;
1195 return 0;
1196 }
1197
1198 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1199 * Returns: 1 if successful
1200 * -1 for unknown partition type
1201 * 0 for other errors
1202 */
1203 static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart)
1204 {
1205 struct GPT gpt;
1206 unsigned char empty_gpt_entry[16]= {0};
1207 struct GPT_part_entry *part;
1208 char buf[512];
1209 unsigned long long curr_part_end;
1210 unsigned all_partitions, entry_size;
1211 unsigned part_nr;
1212
1213 *endofpart = 0;
1214
1215 BUILD_BUG_ON(sizeof(gpt) != 512);
1216 /* skip protective MBR */
1217 lseek(fd, 512, SEEK_SET);
1218 /* read GPT header */
1219 if (read(fd, &gpt, 512) != 512)
1220 return 0;
1221
1222 /* get the number of partition entries and the entry size */
1223 all_partitions = __le32_to_cpu(gpt.part_cnt);
1224 entry_size = __le32_to_cpu(gpt.part_size);
1225
1226 /* Check GPT signature*/
1227 if (gpt.magic != GPT_SIGNATURE_MAGIC)
1228 return -1;
1229
1230 /* sanity checks */
1231 if (all_partitions > 1024 ||
1232 entry_size > sizeof(buf))
1233 return -1;
1234
1235 part = (struct GPT_part_entry *)buf;
1236
1237 for (part_nr = 0; part_nr < all_partitions; part_nr++) {
1238 /* read partition entry */
1239 if (read(fd, buf, entry_size) != (ssize_t)entry_size)
1240 return 0;
1241
1242 /* is this valid partition? */
1243 if (memcmp(part->type_guid, empty_gpt_entry, 16) != 0) {
1244 /* check the last lba for the current partition */
1245 curr_part_end = __le64_to_cpu(part->ending_lba);
1246 if (curr_part_end > *endofpart)
1247 *endofpart = curr_part_end;
1248 }
1249
1250 }
1251 return 1;
1252 }
1253
1254 /* Sets endofpart parameter to the last block used by the last partition on the device.
1255 * Returns: 1 if successful
1256 * -1 for unknown partition type
1257 * 0 for other errors
1258 */
1259 static int get_last_partition_end(int fd, unsigned long long *endofpart)
1260 {
1261 struct MBR boot_sect;
1262 struct MBR_part_record *part;
1263 unsigned long long curr_part_end;
1264 unsigned part_nr;
1265 int retval = 0;
1266
1267 *endofpart = 0;
1268
1269 BUILD_BUG_ON(sizeof(boot_sect) != 512);
1270 /* read MBR */
1271 lseek(fd, 0, 0);
1272 if (read(fd, &boot_sect, 512) != 512)
1273 goto abort;
1274
1275 /* check MBP signature */
1276 if (boot_sect.magic == MBR_SIGNATURE_MAGIC) {
1277 retval = 1;
1278 /* found the correct signature */
1279 part = boot_sect.parts;
1280
1281 for (part_nr = 0; part_nr < MBR_PARTITIONS; part_nr++) {
1282 /* check for GPT type */
1283 if (part->part_type == MBR_GPT_PARTITION_TYPE) {
1284 retval = get_gpt_last_partition_end(fd, endofpart);
1285 break;
1286 }
1287 /* check the last used lba for the current partition */
1288 curr_part_end = __le32_to_cpu(part->first_sect_lba) +
1289 __le32_to_cpu(part->blocks_num);
1290 if (curr_part_end > *endofpart)
1291 *endofpart = curr_part_end;
1292
1293 part++;
1294 }
1295 } else {
1296 /* Unknown partition table */
1297 retval = -1;
1298 }
1299 abort:
1300 return retval;
1301 }
1302
1303 int check_partitions(int fd, char *dname, unsigned long long freesize,
1304 unsigned long long size)
1305 {
1306 /*
1307 * Check where the last partition ends
1308 */
1309 unsigned long long endofpart;
1310 int ret;
1311
1312 if ((ret = get_last_partition_end(fd, &endofpart)) > 0) {
1313 /* There appears to be a partition table here */
1314 if (freesize == 0) {
1315 /* partitions will not be visible in new device */
1316 pr_err("partition table exists on %s but will be lost or\n"
1317 " meaningless after creating array\n",
1318 dname);
1319 return 1;
1320 } else if (endofpart > freesize) {
1321 /* last partition overlaps metadata */
1322 pr_err("metadata will over-write last partition on %s.\n",
1323 dname);
1324 return 1;
1325 } else if (size && endofpart > size) {
1326 /* partitions will be truncated in new device */
1327 pr_err("array size is too small to cover all partitions on %s.\n",
1328 dname);
1329 return 1;
1330 }
1331 }
1332 return 0;
1333 }
1334
1335 int open_container(int fd)
1336 {
1337 /* 'fd' is a block device. Find out if it is in use
1338 * by a container, and return an open fd on that container.
1339 */
1340 char path[256];
1341 char *e;
1342 DIR *dir;
1343 struct dirent *de;
1344 int dfd, n;
1345 char buf[200];
1346 int major, minor;
1347 struct stat st;
1348
1349 if (fstat(fd, &st) != 0)
1350 return -1;
1351 sprintf(path, "/sys/dev/block/%d:%d/holders",
1352 (int)major(st.st_rdev), (int)minor(st.st_rdev));
1353 e = path + strlen(path);
1354
1355 dir = opendir(path);
1356 if (!dir)
1357 return -1;
1358 while ((de = readdir(dir))) {
1359 if (de->d_ino == 0)
1360 continue;
1361 if (de->d_name[0] == '.')
1362 continue;
1363 /* Need to make sure it is a container and not a volume */
1364 sprintf(e, "/%s/md/metadata_version", de->d_name);
1365 dfd = open(path, O_RDONLY);
1366 if (dfd < 0)
1367 continue;
1368 n = read(dfd, buf, sizeof(buf));
1369 close(dfd);
1370 if (n <= 0 || (unsigned)n >= sizeof(buf))
1371 continue;
1372 buf[n] = 0;
1373 if (strncmp(buf, "external", 8) != 0 ||
1374 n < 10 ||
1375 buf[9] == '/')
1376 continue;
1377 sprintf(e, "/%s/dev", de->d_name);
1378 dfd = open(path, O_RDONLY);
1379 if (dfd < 0)
1380 continue;
1381 n = read(dfd, buf, sizeof(buf));
1382 close(dfd);
1383 if (n <= 0 || (unsigned)n >= sizeof(buf))
1384 continue;
1385 buf[n] = 0;
1386 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
1387 continue;
1388 sprintf(buf, "%d:%d", major, minor);
1389 dfd = dev_open(buf, O_RDONLY);
1390 if (dfd >= 0) {
1391 closedir(dir);
1392 return dfd;
1393 }
1394 }
1395 closedir(dir);
1396 return -1;
1397 }
1398
1399 struct superswitch *version_to_superswitch(char *vers)
1400 {
1401 int i;
1402
1403 for (i = 0; superlist[i]; i++) {
1404 struct superswitch *ss = superlist[i];
1405
1406 if (strcmp(vers, ss->name) == 0)
1407 return ss;
1408 }
1409
1410 return NULL;
1411 }
1412
1413 int metadata_container_matches(char *metadata, char *devnm)
1414 {
1415 /* Check if 'devnm' is the container named in 'metadata'
1416 * which is
1417 * /containername/componentname or
1418 * -containername/componentname
1419 */
1420 int l;
1421 if (*metadata != '/' && *metadata != '-')
1422 return 0;
1423 l = strlen(devnm);
1424 if (strncmp(metadata+1, devnm, l) != 0)
1425 return 0;
1426 if (metadata[l+1] != '/')
1427 return 0;
1428 return 1;
1429 }
1430
1431 int metadata_subdev_matches(char *metadata, char *devnm)
1432 {
1433 /* Check if 'devnm' is the subdev named in 'metadata'
1434 * which is
1435 * /containername/subdev or
1436 * -containername/subdev
1437 */
1438 char *sl;
1439 if (*metadata != '/' && *metadata != '-')
1440 return 0;
1441 sl = strchr(metadata+1, '/');
1442 if (!sl)
1443 return 0;
1444 if (strcmp(sl+1, devnm) == 0)
1445 return 1;
1446 return 0;
1447 }
1448
1449 int is_container_member(struct mdstat_ent *mdstat, char *container)
1450 {
1451 if (mdstat->metadata_version == NULL ||
1452 strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
1453 !metadata_container_matches(mdstat->metadata_version+9, container))
1454 return 0;
1455
1456 return 1;
1457 }
1458
1459 int is_subarray_active(char *subarray, char *container)
1460 {
1461 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1462 struct mdstat_ent *ent;
1463
1464 for (ent = mdstat; ent; ent = ent->next)
1465 if (is_container_member(ent, container))
1466 if (strcmp(to_subarray(ent, container), subarray) == 0)
1467 break;
1468
1469 free_mdstat(mdstat);
1470
1471 return ent != NULL;
1472 }
1473
1474 /* open_subarray - opens a subarray in a container
1475 * @dev: container device name
1476 * @st: empty supertype
1477 * @quiet: block reporting errors flag
1478 *
1479 * On success returns an fd to a container and fills in *st
1480 */
1481 int open_subarray(char *dev, char *subarray, struct supertype *st, int quiet)
1482 {
1483 struct mdinfo *mdi;
1484 struct mdinfo *info;
1485 int fd, err = 1;
1486 char *_devnm;
1487
1488 fd = open(dev, O_RDWR|O_EXCL);
1489 if (fd < 0) {
1490 if (!quiet)
1491 pr_err("Couldn't open %s, aborting\n",
1492 dev);
1493 return -1;
1494 }
1495
1496 _devnm = fd2devnm(fd);
1497 if (_devnm == NULL) {
1498 if (!quiet)
1499 pr_err("Failed to determine device number for %s\n",
1500 dev);
1501 goto close_fd;
1502 }
1503 strcpy(st->devnm, _devnm);
1504
1505 mdi = sysfs_read(fd, st->devnm, GET_VERSION|GET_LEVEL);
1506 if (!mdi) {
1507 if (!quiet)
1508 pr_err("Failed to read sysfs for %s\n",
1509 dev);
1510 goto close_fd;
1511 }
1512
1513 if (mdi->array.level != UnSet) {
1514 if (!quiet)
1515 pr_err("%s is not a container\n", dev);
1516 goto free_sysfs;
1517 }
1518
1519 st->ss = version_to_superswitch(mdi->text_version);
1520 if (!st->ss) {
1521 if (!quiet)
1522 pr_err("Operation not supported for %s metadata\n",
1523 mdi->text_version);
1524 goto free_sysfs;
1525 }
1526
1527 if (st->devnm[0] == 0) {
1528 if (!quiet)
1529 pr_err("Failed to allocate device name\n");
1530 goto free_sysfs;
1531 }
1532
1533 if (!st->ss->load_container) {
1534 if (!quiet)
1535 pr_err("%s is not a container\n", dev);
1536 goto free_sysfs;
1537 }
1538
1539 if (st->ss->load_container(st, fd, NULL)) {
1540 if (!quiet)
1541 pr_err("Failed to load metadata for %s\n",
1542 dev);
1543 goto free_sysfs;
1544 }
1545
1546 info = st->ss->container_content(st, subarray);
1547 if (!info) {
1548 if (!quiet)
1549 pr_err("Failed to find subarray-%s in %s\n",
1550 subarray, dev);
1551 goto free_super;
1552 }
1553 free(info);
1554
1555 err = 0;
1556
1557 free_super:
1558 if (err)
1559 st->ss->free_super(st);
1560 free_sysfs:
1561 sysfs_free(mdi);
1562 close_fd:
1563 if (err)
1564 close(fd);
1565
1566 if (err)
1567 return -1;
1568 else
1569 return fd;
1570 }
1571
1572 int add_disk(int mdfd, struct supertype *st,
1573 struct mdinfo *sra, struct mdinfo *info)
1574 {
1575 /* Add a device to an array, in one of 2 ways. */
1576 int rv;
1577 #ifndef MDASSEMBLE
1578 if (st->ss->external) {
1579 if (info->disk.state & (1<<MD_DISK_SYNC))
1580 info->recovery_start = MaxSector;
1581 else
1582 info->recovery_start = 0;
1583 rv = sysfs_add_disk(sra, info, 0);
1584 if (! rv) {
1585 struct mdinfo *sd2;
1586 for (sd2 = sra->devs; sd2; sd2=sd2->next)
1587 if (sd2 == info)
1588 break;
1589 if (sd2 == NULL) {
1590 sd2 = xmalloc(sizeof(*sd2));
1591 *sd2 = *info;
1592 sd2->next = sra->devs;
1593 sra->devs = sd2;
1594 }
1595 }
1596 } else
1597 #endif
1598 rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk);
1599 return rv;
1600 }
1601
1602 int remove_disk(int mdfd, struct supertype *st,
1603 struct mdinfo *sra, struct mdinfo *info)
1604 {
1605 int rv;
1606 /* Remove the disk given by 'info' from the array */
1607 #ifndef MDASSEMBLE
1608 if (st->ss->external)
1609 rv = sysfs_set_str(sra, info, "slot", "none");
1610 else
1611 #endif
1612 rv = ioctl(mdfd, HOT_REMOVE_DISK, makedev(info->disk.major,
1613 info->disk.minor));
1614 return rv;
1615 }
1616
1617 int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
1618 {
1619 /* Initialise kernel's knowledge of array.
1620 * This varies between externally managed arrays
1621 * and older kernels
1622 */
1623 int vers = md_get_version(mdfd);
1624 int rv;
1625
1626 #ifndef MDASSEMBLE
1627 if (st->ss->external)
1628 rv = sysfs_set_array(info, vers);
1629 else
1630 #endif
1631 if ((vers % 100) >= 1) { /* can use different versions */
1632 mdu_array_info_t inf;
1633 memset(&inf, 0, sizeof(inf));
1634 inf.major_version = info->array.major_version;
1635 inf.minor_version = info->array.minor_version;
1636 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
1637 } else
1638 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
1639 return rv;
1640 }
1641
1642 unsigned long long min_recovery_start(struct mdinfo *array)
1643 {
1644 /* find the minimum recovery_start in an array for metadata
1645 * formats that only record per-array recovery progress instead
1646 * of per-device
1647 */
1648 unsigned long long recovery_start = MaxSector;
1649 struct mdinfo *d;
1650
1651 for (d = array->devs; d; d = d->next)
1652 recovery_start = min(recovery_start, d->recovery_start);
1653
1654 return recovery_start;
1655 }
1656
1657 int mdmon_pid(char *devnm)
1658 {
1659 char path[100];
1660 char pid[10];
1661 int fd;
1662 int n;
1663
1664 sprintf(path, "%s/%s.pid", MDMON_DIR, devnm);
1665
1666 fd = open(path, O_RDONLY | O_NOATIME, 0);
1667
1668 if (fd < 0)
1669 return -1;
1670 n = read(fd, pid, 9);
1671 close(fd);
1672 if (n <= 0)
1673 return -1;
1674 return atoi(pid);
1675 }
1676
1677 int mdmon_running(char *devnm)
1678 {
1679 int pid = mdmon_pid(devnm);
1680 if (pid <= 0)
1681 return 0;
1682 if (kill(pid, 0) == 0)
1683 return 1;
1684 return 0;
1685 }
1686
1687 int start_mdmon(char *devnm)
1688 {
1689 int i, skipped;
1690 int len;
1691 pid_t pid;
1692 int status;
1693 char pathbuf[1024];
1694 char *paths[4] = {
1695 pathbuf,
1696 BINDIR "/mdmon",
1697 "./mdmon",
1698 NULL
1699 };
1700
1701 if (check_env("MDADM_NO_MDMON"))
1702 return 0;
1703
1704 len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf)-1);
1705 if (len > 0) {
1706 char *sl;
1707 pathbuf[len] = 0;
1708 sl = strrchr(pathbuf, '/');
1709 if (sl)
1710 sl++;
1711 else
1712 sl = pathbuf;
1713 strcpy(sl, "mdmon");
1714 } else
1715 pathbuf[0] = '\0';
1716
1717 /* First try to run systemctl */
1718 if (!check_env("MDADM_NO_SYSTEMCTL"))
1719 switch(fork()) {
1720 case 0:
1721 /* FIXME yuk. CLOSE_EXEC?? */
1722 skipped = 0;
1723 for (i = 3; skipped < 20; i++)
1724 if (close(i) < 0)
1725 skipped++;
1726 else
1727 skipped = 0;
1728
1729 /* Don't want to see error messages from
1730 * systemctl. If the service doesn't exist,
1731 * we start mdmon ourselves.
1732 */
1733 close(2);
1734 open("/dev/null", O_WRONLY);
1735 snprintf(pathbuf, sizeof(pathbuf), "mdmon@%s.service",
1736 devnm);
1737 status = execl("/usr/bin/systemctl", "systemctl",
1738 "start",
1739 pathbuf, NULL);
1740 status = execl("/bin/systemctl", "systemctl", "start",
1741 pathbuf, NULL);
1742 exit(1);
1743 case -1: pr_err("cannot run mdmon. "
1744 "Array remains readonly\n");
1745 return -1;
1746 default: /* parent - good */
1747 pid = wait(&status);
1748 if (pid >= 0 && status == 0)
1749 return 0;
1750 }
1751
1752 /* That failed, try running mdmon directly */
1753 switch(fork()) {
1754 case 0:
1755 /* FIXME yuk. CLOSE_EXEC?? */
1756 skipped = 0;
1757 for (i = 3; skipped < 20; i++)
1758 if (close(i) < 0)
1759 skipped++;
1760 else
1761 skipped = 0;
1762
1763 for (i = 0; paths[i]; i++)
1764 if (paths[i][0]) {
1765 execl(paths[i], paths[i],
1766 devnm, NULL);
1767 }
1768 exit(1);
1769 case -1: pr_err("cannot run mdmon. "
1770 "Array remains readonly\n");
1771 return -1;
1772 default: /* parent - good */
1773 pid = wait(&status);
1774 if (pid < 0 || status != 0) {
1775 pr_err("failed to launch mdmon. "
1776 "Array remains readonly\n");
1777 return -1;
1778 }
1779 }
1780 return 0;
1781 }
1782
1783 __u32 random32(void)
1784 {
1785 __u32 rv;
1786 int rfd = open("/dev/urandom", O_RDONLY);
1787 if (rfd < 0 || read(rfd, &rv, 4) != 4)
1788 rv = random();
1789 if (rfd >= 0)
1790 close(rfd);
1791 return rv;
1792 }
1793
1794 #ifndef MDASSEMBLE
1795 int flush_metadata_updates(struct supertype *st)
1796 {
1797 int sfd;
1798 if (!st->updates) {
1799 st->update_tail = NULL;
1800 return -1;
1801 }
1802
1803 sfd = connect_monitor(st->container_devnm);
1804 if (sfd < 0)
1805 return -1;
1806
1807 while (st->updates) {
1808 struct metadata_update *mu = st->updates;
1809 st->updates = mu->next;
1810
1811 send_message(sfd, mu, 0);
1812 wait_reply(sfd, 0);
1813 free(mu->buf);
1814 free(mu);
1815 }
1816 ack(sfd, 0);
1817 wait_reply(sfd, 0);
1818 close(sfd);
1819 st->update_tail = NULL;
1820 return 0;
1821 }
1822
1823 void append_metadata_update(struct supertype *st, void *buf, int len)
1824 {
1825
1826 struct metadata_update *mu = xmalloc(sizeof(*mu));
1827
1828 mu->buf = buf;
1829 mu->len = len;
1830 mu->space = NULL;
1831 mu->space_list = NULL;
1832 mu->next = NULL;
1833 *st->update_tail = mu;
1834 st->update_tail = &mu->next;
1835 }
1836 #endif /* MDASSEMBLE */
1837
1838 #ifdef __TINYC__
1839 /* tinyc doesn't optimize this check in ioctl.h out ... */
1840 unsigned int __invalid_size_argument_for_IOC = 0;
1841 #endif
1842
1843 int experimental(void)
1844 {
1845 if (check_env("MDADM_EXPERIMENTAL"))
1846 return 1;
1847 else {
1848 pr_err("To use this feature MDADM_EXPERIMENTAL"
1849 " environment variable has to be defined.\n");
1850 return 0;
1851 }
1852 }
1853
1854 /* Pick all spares matching given criteria from a container
1855 * if min_size == 0 do not check size
1856 * if domlist == NULL do not check domains
1857 * if spare_group given add it to domains of each spare
1858 * metadata allows to test domains using metadata of destination array */
1859 struct mdinfo *container_choose_spares(struct supertype *st,
1860 unsigned long long min_size,
1861 struct domainlist *domlist,
1862 char *spare_group,
1863 const char *metadata, int get_one)
1864 {
1865 struct mdinfo *d, **dp, *disks = NULL;
1866
1867 /* get list of all disks in container */
1868 if (st->ss->getinfo_super_disks)
1869 disks = st->ss->getinfo_super_disks(st);
1870
1871 if (!disks)
1872 return disks;
1873 /* find spare devices on the list */
1874 dp = &disks->devs;
1875 disks->array.spare_disks = 0;
1876 while (*dp) {
1877 int found = 0;
1878 d = *dp;
1879 if (d->disk.state == 0) {
1880 /* check if size is acceptable */
1881 unsigned long long dev_size;
1882 dev_t dev = makedev(d->disk.major,d->disk.minor);
1883
1884 if (!min_size ||
1885 (dev_size_from_id(dev, &dev_size) &&
1886 dev_size >= min_size))
1887 found = 1;
1888 /* check if domain matches */
1889 if (found && domlist) {
1890 struct dev_policy *pol = devid_policy(dev);
1891 if (spare_group)
1892 pol_add(&pol, pol_domain,
1893 spare_group, NULL);
1894 if (domain_test(domlist, pol, metadata) != 1)
1895 found = 0;
1896 dev_policy_free(pol);
1897 }
1898 }
1899 if (found) {
1900 dp = &d->next;
1901 disks->array.spare_disks++;
1902 if (get_one) {
1903 sysfs_free(*dp);
1904 d->next = NULL;
1905 }
1906 } else {
1907 *dp = d->next;
1908 d->next = NULL;
1909 sysfs_free(d);
1910 }
1911 }
1912 return disks;
1913 }
1914
1915 /* Checks if paths point to the same device
1916 * Returns 0 if they do.
1917 * Returns 1 if they don't.
1918 * Returns -1 if something went wrong,
1919 * e.g. paths are empty or the files
1920 * they point to don't exist */
1921 int compare_paths (char* path1, char* path2)
1922 {
1923 struct stat st1,st2;
1924
1925 if (path1 == NULL || path2 == NULL)
1926 return -1;
1927 if (stat(path1,&st1) != 0)
1928 return -1;
1929 if (stat(path2,&st2) != 0)
1930 return -1;
1931 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev))
1932 return 0;
1933 return 1;
1934 }
1935
1936 /* Make sure we can open as many devices as needed */
1937 void enable_fds(int devices)
1938 {
1939 unsigned int fds = 20 + devices;
1940 struct rlimit lim;
1941 if (getrlimit(RLIMIT_NOFILE, &lim) != 0
1942 || lim.rlim_cur >= fds)
1943 return;
1944 if (lim.rlim_max < fds)
1945 lim.rlim_max = fds;
1946 lim.rlim_cur = fds;
1947 setrlimit(RLIMIT_NOFILE, &lim);
1948 }
1949
1950 int in_initrd(void)
1951 {
1952 /* This is based on similar function in systemd. */
1953 struct statfs s;
1954 /* statfs.f_type is signed long on s390x and MIPS, causing all
1955 sorts of sign extension problems with RAMFS_MAGIC being
1956 defined as 0x858458f6 */
1957 return statfs("/", &s) >= 0 &&
1958 ((unsigned long)s.f_type == TMPFS_MAGIC ||
1959 ((unsigned long)s.f_type & 0xFFFFFFFFUL) ==
1960 ((unsigned long)RAMFS_MAGIC & 0xFFFFFFFFUL));
1961 }
1962
1963 void reopen_mddev(int mdfd)
1964 {
1965 /* Re-open without any O_EXCL, but keep
1966 * the same fd
1967 */
1968 char *devnm;
1969 int fd;
1970 devnm = fd2devnm(mdfd);
1971 close(mdfd);
1972 fd = open_dev(devnm);
1973 if (fd >= 0 && fd != mdfd)
1974 dup2(fd, mdfd);
1975 }