8b5224234862f67717c8373fc0265dd6c30cb2d4
[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 <poll.h>
35 #include <ctype.h>
36 #include <dirent.h>
37 #include <signal.h>
38 #include <dlfcn.h>
39
40
41 /*
42 * following taken from linux/blkpg.h because they aren't
43 * anywhere else and it isn't safe to #include linux/ * stuff.
44 */
45
46 #define BLKPG _IO(0x12,105)
47
48 /* The argument structure */
49 struct blkpg_ioctl_arg {
50 int op;
51 int flags;
52 int datalen;
53 void *data;
54 };
55
56 /* The subfunctions (for the op field) */
57 #define BLKPG_ADD_PARTITION 1
58 #define BLKPG_DEL_PARTITION 2
59
60 /* Sizes of name fields. Unused at present. */
61 #define BLKPG_DEVNAMELTH 64
62 #define BLKPG_VOLNAMELTH 64
63
64 /* The data structure for ADD_PARTITION and DEL_PARTITION */
65 struct blkpg_partition {
66 long long start; /* starting offset in bytes */
67 long long length; /* length in bytes */
68 int pno; /* partition number */
69 char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2,
70 to be used in kernel messages */
71 char volname[BLKPG_VOLNAMELTH]; /* volume label */
72 };
73
74 #include "part.h"
75
76 /* Force a compilation error if condition is true */
77 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
78
79 /* Force a compilation error if condition is true, but also produce a
80 result (of value 0 and type size_t), so the expression can be used
81 e.g. in a structure initializer (or where-ever else comma expressions
82 aren't permitted). */
83 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
84
85 static int is_dlm_hooks_ready = 0;
86
87 int dlm_funs_ready(void)
88 {
89 return is_dlm_hooks_ready ? 1 : 0;
90 }
91
92 #ifndef MDASSEMBLE
93 static struct dlm_hooks *dlm_hooks = NULL;
94 struct dlm_lock_resource *dlm_lock_res = NULL;
95 static int ast_called = 0;
96
97 struct dlm_lock_resource {
98 dlm_lshandle_t *ls;
99 struct dlm_lksb lksb;
100 };
101
102 /* Using poll(2) to wait for and dispatch ASTs */
103 static int poll_for_ast(dlm_lshandle_t ls)
104 {
105 struct pollfd pfd;
106
107 pfd.fd = dlm_hooks->ls_get_fd(ls);
108 pfd.events = POLLIN;
109
110 while (!ast_called)
111 {
112 if (poll(&pfd, 1, 0) < 0)
113 {
114 perror("poll");
115 return -1;
116 }
117 dlm_hooks->dispatch(dlm_hooks->ls_get_fd(ls));
118 }
119 ast_called = 0;
120
121 return 0;
122 }
123
124 static void dlm_ast(void *arg)
125 {
126 ast_called = 1;
127 }
128
129 static char *cluster_name = NULL;
130 /* Create the lockspace, take bitmapXXX locks on all the bitmaps. */
131 int cluster_get_dlmlock(int *lockid)
132 {
133 int ret = -1;
134 char str[64];
135 int flags = LKF_NOQUEUE;
136
137 ret = get_cluster_name(&cluster_name);
138 if (ret) {
139 pr_err("The md can't get cluster name\n");
140 return -1;
141 }
142
143 dlm_lock_res = xmalloc(sizeof(struct dlm_lock_resource));
144 dlm_lock_res->ls = dlm_hooks->create_lockspace(cluster_name, O_RDWR);
145 if (!dlm_lock_res->ls) {
146 pr_err("%s failed to create lockspace\n", cluster_name);
147 return -ENOMEM;
148 }
149
150 snprintf(str, 64, "bitmap%s", cluster_name);
151 ret = dlm_hooks->ls_lock(dlm_lock_res->ls, LKM_PWMODE, &dlm_lock_res->lksb,
152 flags, str, strlen(str), 0, dlm_ast,
153 dlm_lock_res, NULL, NULL);
154 if (ret) {
155 pr_err("error %d when get PW mode on lock %s\n", errno, str);
156 dlm_hooks->release_lockspace(cluster_name, dlm_lock_res->ls, 1);
157 return ret;
158 }
159
160 /* Wait for it to complete */
161 poll_for_ast(dlm_lock_res->ls);
162 *lockid = dlm_lock_res->lksb.sb_lkid;
163
164 return dlm_lock_res->lksb.sb_status;
165 }
166
167 int cluster_release_dlmlock(int lockid)
168 {
169 int ret = -1;
170
171 if (!cluster_name)
172 return -1;
173
174 ret = dlm_hooks->ls_unlock(dlm_lock_res->ls, lockid, 0,
175 &dlm_lock_res->lksb, dlm_lock_res);
176 if (ret) {
177 pr_err("error %d happened when unlock\n", errno);
178 /* XXX make sure the lock is unlocked eventually */
179 goto out;
180 }
181
182 /* Wait for it to complete */
183 poll_for_ast(dlm_lock_res->ls);
184
185 errno = dlm_lock_res->lksb.sb_status;
186 if (errno != EUNLOCK) {
187 pr_err("error %d happened in ast when unlock lockspace\n", errno);
188 /* XXX make sure the lockspace is unlocked eventually */
189 goto out;
190 }
191
192 ret = dlm_hooks->release_lockspace(cluster_name, dlm_lock_res->ls, 1);
193 if (ret) {
194 pr_err("error %d happened when release lockspace\n", errno);
195 /* XXX make sure the lockspace is released eventually */
196 goto out;
197 }
198 free(dlm_lock_res);
199
200 out:
201 return ret;
202 }
203 #else
204 int cluster_get_dlmlock(int *lockid)
205 {
206 return -1;
207 }
208 int cluster_release_dlmlock(int lockid)
209 {
210 return -1;
211 }
212 #endif
213
214 /*
215 * Parse a 128 bit uuid in 4 integers
216 * format is 32 hexx nibbles with options :.<space> separator
217 * If not exactly 32 hex digits are found, return 0
218 * else return 1
219 */
220 int parse_uuid(char *str, int uuid[4])
221 {
222 int hit = 0; /* number of Hex digIT */
223 int i;
224 char c;
225 for (i = 0; i < 4; i++)
226 uuid[i] = 0;
227
228 while ((c = *str++) != 0) {
229 int n;
230 if (c >= '0' && c <= '9')
231 n = c-'0';
232 else if (c >= 'a' && c <= 'f')
233 n = 10 + c - 'a';
234 else if (c >= 'A' && c <= 'F')
235 n = 10 + c - 'A';
236 else if (strchr(":. -", c))
237 continue;
238 else return 0;
239
240 if (hit<32) {
241 uuid[hit/8] <<= 4;
242 uuid[hit/8] += n;
243 }
244 hit++;
245 }
246 if (hit == 32)
247 return 1;
248 return 0;
249 }
250
251 /*
252 * Get the md version number.
253 * We use the RAID_VERSION ioctl if it is supported
254 * If not, but we have a block device with major '9', we assume
255 * 0.36.0
256 *
257 * Return version number as 24 but number - assume version parts
258 * always < 255
259 */
260
261 int md_get_version(int fd)
262 {
263 struct stat stb;
264 mdu_version_t vers;
265
266 if (fstat(fd, &stb)<0)
267 return -1;
268 if ((S_IFMT&stb.st_mode) != S_IFBLK)
269 return -1;
270
271 if (ioctl(fd, RAID_VERSION, &vers) == 0)
272 return (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
273 if (errno == EACCES)
274 return -1;
275 if (major(stb.st_rdev) == MD_MAJOR)
276 return (3600);
277 return -1;
278 }
279
280 int get_linux_version()
281 {
282 struct utsname name;
283 char *cp;
284 int a = 0, b = 0,c = 0;
285 if (uname(&name) <0)
286 return -1;
287
288 cp = name.release;
289 a = strtoul(cp, &cp, 10);
290 if (*cp == '.')
291 b = strtoul(cp+1, &cp, 10);
292 if (*cp == '.')
293 c = strtoul(cp+1, &cp, 10);
294
295 return (a*1000000)+(b*1000)+c;
296 }
297
298 #ifndef MDASSEMBLE
299 int mdadm_version(char *version)
300 {
301 int a, b, c;
302 char *cp;
303
304 if (!version)
305 version = Version;
306
307 cp = strchr(version, '-');
308 if (!cp || *(cp+1) != ' ' || *(cp+2) != 'v')
309 return -1;
310 cp += 3;
311 a = strtoul(cp, &cp, 10);
312 if (*cp != '.')
313 return -1;
314 b = strtoul(cp+1, &cp, 10);
315 if (*cp == '.')
316 c = strtoul(cp+1, &cp, 10);
317 else
318 c = 0;
319 if (*cp != ' ' && *cp != '-')
320 return -1;
321 return (a*1000000)+(b*1000)+c;
322 }
323
324 unsigned long long parse_size(char *size)
325 {
326 /* parse 'size' which should be a number optionally
327 * followed by 'K', 'M', or 'G'.
328 * Without a suffix, K is assumed.
329 * Number returned is in sectors (half-K)
330 * INVALID_SECTORS returned on error.
331 */
332 char *c;
333 long long s = strtoll(size, &c, 10);
334 if (s > 0) {
335 switch (*c) {
336 case 'K':
337 c++;
338 default:
339 s *= 2;
340 break;
341 case 'M':
342 c++;
343 s *= 1024 * 2;
344 break;
345 case 'G':
346 c++;
347 s *= 1024 * 1024 * 2;
348 break;
349 case 's': /* sectors */
350 c++;
351 break;
352 }
353 } else
354 s = INVALID_SECTORS;
355 if (*c)
356 s = INVALID_SECTORS;
357 return s;
358 }
359
360 int parse_layout_10(char *layout)
361 {
362 int copies, rv;
363 char *cp;
364 /* Parse the layout string for raid10 */
365 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
366 if ((layout[0] != 'n' && layout[0] != 'f' && layout[0] != 'o') ||
367 (copies = strtoul(layout+1, &cp, 10)) < 1 ||
368 copies > 200 ||
369 *cp)
370 return -1;
371 if (layout[0] == 'n')
372 rv = 256 + copies;
373 else if (layout[0] == 'o')
374 rv = 0x10000 + (copies<<8) + 1;
375 else
376 rv = 1 + (copies<<8);
377 return rv;
378 }
379
380 int parse_layout_faulty(char *layout)
381 {
382 /* Parse the layout string for 'faulty' */
383 int ln = strcspn(layout, "0123456789");
384 char *m = xstrdup(layout);
385 int mode;
386 m[ln] = 0;
387 mode = map_name(faultylayout, m);
388 if (mode == UnSet)
389 return -1;
390
391 return mode | (atoi(layout+ln)<< ModeShift);
392 }
393
394 long parse_num(char *num)
395 {
396 /* Either return a valid number, or -1 */
397 char *c;
398 long rv = strtol(num, &c, 10);
399 if (rv < 0 || *c || !num[0])
400 return -1;
401 else
402 return rv;
403 }
404 #endif
405
406 int parse_cluster_confirm_arg(char *input, char **devname, int *slot)
407 {
408 char *dev;
409 *slot = strtoul(input, &dev, 10);
410 if (dev == input || dev[0] != ':')
411 return -1;
412 *devname = dev+1;
413 return 0;
414 }
415
416 void remove_partitions(int fd)
417 {
418 /* remove partitions from this block devices.
419 * This is used for components added to an array
420 */
421 #ifdef BLKPG_DEL_PARTITION
422 struct blkpg_ioctl_arg a;
423 struct blkpg_partition p;
424
425 a.op = BLKPG_DEL_PARTITION;
426 a.data = (void*)&p;
427 a.datalen = sizeof(p);
428 a.flags = 0;
429 memset(a.data, 0, a.datalen);
430 for (p.pno = 0; p.pno < 16; p.pno++)
431 ioctl(fd, BLKPG, &a);
432 #endif
433 }
434
435 int test_partition(int fd)
436 {
437 /* Check if fd is a whole-disk or a partition.
438 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
439 * will return ENXIO on an invalid partition number.
440 */
441 struct blkpg_ioctl_arg a;
442 struct blkpg_partition p;
443 a.op = BLKPG_DEL_PARTITION;
444 a.data = (void*)&p;
445 a.datalen = sizeof(p);
446 a.flags = 0;
447 memset(a.data, 0, a.datalen);
448 p.pno = 1<<30;
449 if (ioctl(fd, BLKPG, &a) == 0)
450 /* Very unlikely, but not a partition */
451 return 0;
452 if (errno == ENXIO || errno == ENOTTY)
453 /* not a partition */
454 return 0;
455
456 return 1;
457 }
458
459 int test_partition_from_id(dev_t id)
460 {
461 char buf[20];
462 int fd, rv;
463
464 sprintf(buf, "%d:%d", major(id), minor(id));
465 fd = dev_open(buf, O_RDONLY);
466 if (fd < 0)
467 return -1;
468 rv = test_partition(fd);
469 close(fd);
470 return rv;
471 }
472
473 int enough(int level, int raid_disks, int layout, int clean, char *avail)
474 {
475 int copies, first;
476 int i;
477 int avail_disks = 0;
478
479 for (i = 0; i < raid_disks; i++)
480 avail_disks += !!avail[i];
481
482 switch (level) {
483 case 10:
484 /* This is the tricky one - we need to check
485 * which actual disks are present.
486 */
487 copies = (layout&255)* ((layout>>8) & 255);
488 first = 0;
489 do {
490 /* there must be one of the 'copies' form 'first' */
491 int n = copies;
492 int cnt = 0;
493 int this = first;
494 while (n--) {
495 if (avail[this])
496 cnt++;
497 this = (this+1) % raid_disks;
498 }
499 if (cnt == 0)
500 return 0;
501 first = (first+(layout&255)) % raid_disks;
502 } while (first != 0);
503 return 1;
504
505 case LEVEL_MULTIPATH:
506 return avail_disks>= 1;
507 case LEVEL_LINEAR:
508 case 0:
509 return avail_disks == raid_disks;
510 case 1:
511 return avail_disks >= 1;
512 case 4:
513 if (avail_disks == raid_disks - 1 &&
514 !avail[raid_disks - 1])
515 /* If just the parity device is missing, then we
516 * have enough, even if not clean
517 */
518 return 1;
519 /* FALL THROUGH */
520 case 5:
521 if (clean)
522 return avail_disks >= raid_disks-1;
523 else
524 return avail_disks >= raid_disks;
525 case 6:
526 if (clean)
527 return avail_disks >= raid_disks-2;
528 else
529 return avail_disks >= raid_disks;
530 default:
531 return 0;
532 }
533 }
534
535 int enough_fd(int fd)
536 {
537 struct mdu_array_info_s array;
538 struct mdu_disk_info_s disk;
539 int i, rv;
540 char *avail;
541
542 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0 ||
543 array.raid_disks <= 0)
544 return 0;
545 avail = xcalloc(array.raid_disks, 1);
546 for (i = 0; i < MAX_DISKS && array.nr_disks > 0; i++) {
547 disk.number = i;
548 if (ioctl(fd, GET_DISK_INFO, &disk) != 0)
549 continue;
550 if (disk.major == 0 && disk.minor == 0)
551 continue;
552 array.nr_disks--;
553
554 if (! (disk.state & (1<<MD_DISK_SYNC)))
555 continue;
556 if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks)
557 continue;
558 avail[disk.raid_disk] = 1;
559 }
560 /* This is used on an active array, so assume it is clean */
561 rv = enough(array.level, array.raid_disks, array.layout,
562 1, avail);
563 free(avail);
564 return rv;
565 }
566
567 const int uuid_zero[4] = { 0, 0, 0, 0 };
568
569 int same_uuid(int a[4], int b[4], int swapuuid)
570 {
571 if (swapuuid) {
572 /* parse uuids are hostendian.
573 * uuid's from some superblocks are big-ending
574 * if there is a difference, we need to swap..
575 */
576 unsigned char *ac = (unsigned char *)a;
577 unsigned char *bc = (unsigned char *)b;
578 int i;
579 for (i = 0; i < 16; i += 4) {
580 if (ac[i+0] != bc[i+3] ||
581 ac[i+1] != bc[i+2] ||
582 ac[i+2] != bc[i+1] ||
583 ac[i+3] != bc[i+0])
584 return 0;
585 }
586 return 1;
587 } else {
588 if (a[0]==b[0] &&
589 a[1]==b[1] &&
590 a[2]==b[2] &&
591 a[3]==b[3])
592 return 1;
593 return 0;
594 }
595 }
596
597 void copy_uuid(void *a, int b[4], int swapuuid)
598 {
599 if (swapuuid) {
600 /* parse uuids are hostendian.
601 * uuid's from some superblocks are big-ending
602 * if there is a difference, we need to swap..
603 */
604 unsigned char *ac = (unsigned char *)a;
605 unsigned char *bc = (unsigned char *)b;
606 int i;
607 for (i = 0; i < 16; i += 4) {
608 ac[i+0] = bc[i+3];
609 ac[i+1] = bc[i+2];
610 ac[i+2] = bc[i+1];
611 ac[i+3] = bc[i+0];
612 }
613 } else
614 memcpy(a, b, 16);
615 }
616
617 char *__fname_from_uuid(int id[4], int swap, char *buf, char sep)
618 {
619 int i, j;
620 char uuid[16];
621 char *c = buf;
622 strcpy(c, "UUID-");
623 c += strlen(c);
624 copy_uuid(uuid, id, swap);
625 for (i = 0; i < 4; i++) {
626 if (i)
627 *c++ = sep;
628 for (j = 3; j >= 0; j--) {
629 sprintf(c,"%02x", (unsigned char) uuid[j+4*i]);
630 c+= 2;
631 }
632 }
633 return buf;
634
635 }
636
637 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
638 {
639 // dirty hack to work around an issue with super1 superblocks...
640 // super1 superblocks need swapuuid set in order for assembly to
641 // work, but can't have it set if we want this printout to match
642 // all the other uuid printouts in super1.c, so we force swapuuid
643 // to 1 to make our printout match the rest of super1
644 return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep);
645 }
646
647 #ifndef MDASSEMBLE
648 int check_ext2(int fd, char *name)
649 {
650 /*
651 * Check for an ext2fs file system.
652 * Superblock is always 1K at 1K offset
653 *
654 * s_magic is le16 at 56 == 0xEF53
655 * report mtime - le32 at 44
656 * blocks - le32 at 4
657 * logblksize - le32 at 24
658 */
659 unsigned char sb[1024];
660 time_t mtime;
661 unsigned long long size;
662 int bsize;
663 if (lseek(fd, 1024,0)!= 1024)
664 return 0;
665 if (read(fd, sb, 1024)!= 1024)
666 return 0;
667 if (sb[56] != 0x53 || sb[57] != 0xef)
668 return 0;
669
670 mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
671 bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
672 size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
673 size <<= bsize;
674 pr_err("%s appears to contain an ext2fs file system\n",
675 name);
676 cont_err("size=%lluK mtime=%s", size, ctime(&mtime));
677 return 1;
678 }
679
680 int check_reiser(int fd, char *name)
681 {
682 /*
683 * superblock is at 64K
684 * size is 1024;
685 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
686 *
687 */
688 unsigned char sb[1024];
689 unsigned long long size;
690 if (lseek(fd, 64*1024, 0) != 64*1024)
691 return 0;
692 if (read(fd, sb, 1024) != 1024)
693 return 0;
694 if (strncmp((char*)sb+52, "ReIsErFs",8) != 0 &&
695 strncmp((char*)sb+52, "ReIsEr2Fs",9) != 0)
696 return 0;
697 pr_err("%s appears to contain a reiserfs file system\n",name);
698 size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
699 cont_err("size = %lluK\n", size*4);
700
701 return 1;
702 }
703
704 int check_raid(int fd, char *name)
705 {
706 struct mdinfo info;
707 time_t crtime;
708 char *level;
709 struct supertype *st = guess_super(fd);
710
711 if (!st)
712 return 0;
713 st->ss->load_super(st, fd, name);
714 /* Looks like a raid array .. */
715 pr_err("%s appears to be part of a raid array:\n",
716 name);
717 st->ss->getinfo_super(st, &info, NULL);
718 st->ss->free_super(st);
719 crtime = info.array.ctime;
720 level = map_num(pers, info.array.level);
721 if (!level) level = "-unknown-";
722 cont_err("level=%s devices=%d ctime=%s",
723 level, info.array.raid_disks, ctime(&crtime));
724 return 1;
725 }
726
727 int ask(char *mesg)
728 {
729 char *add = "";
730 int i;
731 for (i = 0; i < 5; i++) {
732 char buf[100];
733 fprintf(stderr, "%s%s", mesg, add);
734 fflush(stderr);
735 if (fgets(buf, 100, stdin)==NULL)
736 return 0;
737 if (buf[0]=='y' || buf[0]=='Y')
738 return 1;
739 if (buf[0]=='n' || buf[0]=='N')
740 return 0;
741 add = "(y/n) ";
742 }
743 pr_err("assuming 'no'\n");
744 return 0;
745 }
746 #endif /* MDASSEMBLE */
747
748 int is_standard(char *dev, int *nump)
749 {
750 /* tests if dev is a "standard" md dev name.
751 * i.e if the last component is "/dNN" or "/mdNN",
752 * where NN is a string of digits
753 * Returns 1 if a partitionable standard,
754 * -1 if non-partitonable,
755 * 0 if not a standard name.
756 */
757 char *d = strrchr(dev, '/');
758 int type = 0;
759 int num;
760 if (!d)
761 return 0;
762 if (strncmp(d, "/d",2) == 0)
763 d += 2, type = 1; /* /dev/md/dN{pM} */
764 else if (strncmp(d, "/md_d", 5) == 0)
765 d += 5, type = 1; /* /dev/md_dN{pM} */
766 else if (strncmp(d, "/md", 3) == 0)
767 d += 3, type = -1; /* /dev/mdN */
768 else if (d-dev > 3 && strncmp(d-2, "md/", 3) == 0)
769 d += 1, type = -1; /* /dev/md/N */
770 else
771 return 0;
772 if (!*d)
773 return 0;
774 num = atoi(d);
775 while (isdigit(*d))
776 d++;
777 if (*d)
778 return 0;
779 if (nump) *nump = num;
780
781 return type;
782 }
783
784 unsigned long calc_csum(void *super, int bytes)
785 {
786 unsigned long long newcsum = 0;
787 int i;
788 unsigned int csum;
789 unsigned int *superc = (unsigned int*) super;
790
791 for(i = 0; i < bytes/4; i++)
792 newcsum += superc[i];
793 csum = (newcsum& 0xffffffff) + (newcsum>>32);
794 #ifdef __alpha__
795 /* The in-kernel checksum calculation is always 16bit on
796 * the alpha, though it is 32 bit on i386...
797 * I wonder what it is elsewhere... (it uses an API in
798 * a way that it shouldn't).
799 */
800 csum = (csum & 0xffff) + (csum >> 16);
801 csum = (csum & 0xffff) + (csum >> 16);
802 #endif
803 return csum;
804 }
805
806 #ifndef MDASSEMBLE
807 char *human_size(long long bytes)
808 {
809 static char buf[30];
810
811 /* We convert bytes to either centi-M{ega,ibi}bytes or
812 * centi-G{igi,ibi}bytes, with appropriate rounding,
813 * and then print 1/100th of those as a decimal.
814 * We allow upto 2048Megabytes before converting to
815 * gigabytes, as that shows more precision and isn't
816 * too large a number.
817 * Terabytes are not yet handled.
818 */
819
820 if (bytes < 5000*1024)
821 buf[0] = 0;
822 else if (bytes < 2*1024LL*1024LL*1024LL) {
823 long cMiB = (bytes * 200LL / (1LL<<20) + 1) / 2;
824 long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2;
825 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
826 cMiB/100 , cMiB % 100,
827 cMB/100, cMB % 100);
828 } else {
829 long cGiB = (bytes * 200LL / (1LL<<30) +1) / 2;
830 long cGB = (bytes / (1000000000LL/200LL ) +1) /2;
831 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
832 cGiB/100 , cGiB % 100,
833 cGB/100, cGB % 100);
834 }
835 return buf;
836 }
837
838 char *human_size_brief(long long bytes, int prefix)
839 {
840 static char buf[30];
841
842 /* We convert bytes to either centi-M{ega,ibi}bytes or
843 * centi-G{igi,ibi}bytes, with appropriate rounding,
844 * and then print 1/100th of those as a decimal.
845 * We allow upto 2048Megabytes before converting to
846 * gigabytes, as that shows more precision and isn't
847 * too large a number.
848 * Terabytes are not yet handled.
849 *
850 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
851 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
852 */
853
854 if (bytes < 5000*1024)
855 buf[0] = 0;
856 else if (prefix == IEC) {
857 if (bytes < 2*1024LL*1024LL*1024LL) {
858 long cMiB = (bytes * 200LL / (1LL<<20) +1) /2;
859 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
860 cMiB/100 , cMiB % 100);
861 } else {
862 long cGiB = (bytes * 200LL / (1LL<<30) +1) /2;
863 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
864 cGiB/100 , cGiB % 100);
865 }
866 }
867 else if (prefix == JEDEC) {
868 if (bytes < 2*1024LL*1024LL*1024LL) {
869 long cMB = (bytes / ( 1000000LL / 200LL ) +1) /2;
870 snprintf(buf, sizeof(buf), "%ld.%02ldMB",
871 cMB/100, cMB % 100);
872 } else {
873 long cGB = (bytes / (1000000000LL/200LL ) +1) /2;
874 snprintf(buf, sizeof(buf), "%ld.%02ldGB",
875 cGB/100 , cGB % 100);
876 }
877 }
878 else
879 buf[0] = 0;
880
881 return buf;
882 }
883
884 void print_r10_layout(int layout)
885 {
886 int near = layout & 255;
887 int far = (layout >> 8) & 255;
888 int offset = (layout&0x10000);
889 char *sep = "";
890
891 if (near != 1) {
892 printf("%s near=%d", sep, near);
893 sep = ",";
894 }
895 if (far != 1)
896 printf("%s %s=%d", sep, offset?"offset":"far", far);
897 if (near*far == 1)
898 printf("NO REDUNDANCY");
899 }
900 #endif
901
902 unsigned long long calc_array_size(int level, int raid_disks, int layout,
903 int chunksize, unsigned long long devsize)
904 {
905 if (level == 1)
906 return devsize;
907 devsize &= ~(unsigned long long)((chunksize>>9)-1);
908 return get_data_disks(level, layout, raid_disks) * devsize;
909 }
910
911 int get_data_disks(int level, int layout, int raid_disks)
912 {
913 int data_disks = 0;
914 switch (level) {
915 case 0: data_disks = raid_disks;
916 break;
917 case 1: data_disks = 1;
918 break;
919 case 4:
920 case 5: data_disks = raid_disks - 1;
921 break;
922 case 6: data_disks = raid_disks - 2;
923 break;
924 case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
925 break;
926 }
927
928 return data_disks;
929 }
930
931 dev_t devnm2devid(char *devnm)
932 {
933 /* First look in /sys/block/$DEVNM/dev for %d:%d
934 * If that fails, try parsing out a number
935 */
936 char path[100];
937 char *ep;
938 int fd;
939 int mjr,mnr;
940
941 sprintf(path, "/sys/block/%s/dev", devnm);
942 fd = open(path, O_RDONLY);
943 if (fd >= 0) {
944 char buf[20];
945 int n = read(fd, buf, sizeof(buf));
946 close(fd);
947 if (n > 0)
948 buf[n] = 0;
949 if (n > 0 && sscanf(buf, "%d:%d\n", &mjr, &mnr) == 2)
950 return makedev(mjr, mnr);
951 }
952 if (strncmp(devnm, "md_d", 4) == 0 &&
953 isdigit(devnm[4]) &&
954 (mnr = strtoul(devnm+4, &ep, 10)) >= 0 &&
955 ep > devnm && *ep == 0)
956 return makedev(get_mdp_major(), mnr << MdpMinorShift);
957
958 if (strncmp(devnm, "md", 2) == 0 &&
959 isdigit(devnm[2]) &&
960 (mnr = strtoul(devnm+2, &ep, 10)) >= 0 &&
961 ep > devnm && *ep == 0)
962 return makedev(MD_MAJOR, mnr);
963
964 return 0;
965 }
966
967 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
968 char *get_md_name(char *devnm)
969 {
970 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
971 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
972
973 static char devname[50];
974 struct stat stb;
975 dev_t rdev = devnm2devid(devnm);
976 char *dn;
977
978 if (rdev == 0)
979 return 0;
980 if (strncmp(devnm, "md_", 3) == 0) {
981 snprintf(devname, sizeof(devname), "/dev/md/%s",
982 devnm + 3);
983 if (stat(devname, &stb) == 0
984 && (S_IFMT&stb.st_mode) == S_IFBLK
985 && (stb.st_rdev == rdev))
986 return devname;
987 }
988 snprintf(devname, sizeof(devname), "/dev/%s", devnm);
989 if (stat(devname, &stb) == 0
990 && (S_IFMT&stb.st_mode) == S_IFBLK
991 && (stb.st_rdev == rdev))
992 return devname;
993
994 snprintf(devname, sizeof(devname), "/dev/md/%s", devnm+2);
995 if (stat(devname, &stb) == 0
996 && (S_IFMT&stb.st_mode) == S_IFBLK
997 && (stb.st_rdev == rdev))
998 return devname;
999
1000 dn = map_dev(major(rdev), minor(rdev), 0);
1001 if (dn)
1002 return dn;
1003 snprintf(devname, sizeof(devname), "/dev/.tmp.%s", devnm);
1004 if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
1005 if (errno != EEXIST)
1006 return NULL;
1007
1008 if (stat(devname, &stb) == 0
1009 && (S_IFMT&stb.st_mode) == S_IFBLK
1010 && (stb.st_rdev == rdev))
1011 return devname;
1012 unlink(devname);
1013 return NULL;
1014 }
1015
1016 void put_md_name(char *name)
1017 {
1018 if (strncmp(name, "/dev/.tmp.md", 12) == 0)
1019 unlink(name);
1020 }
1021 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1022
1023 int get_maj_min(char *dev, int *major, int *minor)
1024 {
1025 char *e;
1026 *major = strtoul(dev, &e, 0);
1027 return (e > dev && *e == ':' && e[1] &&
1028 (*minor = strtoul(e+1, &e, 0)) >= 0 &&
1029 *e == 0);
1030 }
1031
1032 int dev_open(char *dev, int flags)
1033 {
1034 /* like 'open', but if 'dev' matches %d:%d, create a temp
1035 * block device and open that
1036 */
1037 int fd = -1;
1038 char devname[32];
1039 int major;
1040 int minor;
1041
1042 if (!dev)
1043 return -1;
1044 flags |= O_DIRECT;
1045
1046 if (get_maj_min(dev, &major, &minor)) {
1047 snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d",
1048 (int)getpid(), major, minor);
1049 if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) {
1050 fd = open(devname, flags);
1051 unlink(devname);
1052 }
1053 if (fd < 0) {
1054 /* Try /tmp as /dev appear to be read-only */
1055 snprintf(devname, sizeof(devname), "/tmp/.tmp.md.%d:%d:%d",
1056 (int)getpid(), major, minor);
1057 if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) {
1058 fd = open(devname, flags);
1059 unlink(devname);
1060 }
1061 }
1062 } else
1063 fd = open(dev, flags);
1064 return fd;
1065 }
1066
1067 int open_dev_flags(char *devnm, int flags)
1068 {
1069 dev_t devid;
1070 char buf[20];
1071
1072 devid = devnm2devid(devnm);
1073 sprintf(buf, "%d:%d", major(devid), minor(devid));
1074 return dev_open(buf, flags);
1075 }
1076
1077 int open_dev(char *devnm)
1078 {
1079 return open_dev_flags(devnm, O_RDONLY);
1080 }
1081
1082 int open_dev_excl(char *devnm)
1083 {
1084 char buf[20];
1085 int i;
1086 int flags = O_RDWR;
1087 dev_t devid = devnm2devid(devnm);
1088 long delay = 1000;
1089
1090 sprintf(buf, "%d:%d", major(devid), minor(devid));
1091 for (i = 0 ; i < 25 ; i++) {
1092 int fd = dev_open(buf, flags|O_EXCL);
1093 if (fd >= 0)
1094 return fd;
1095 if (errno == EACCES && flags == O_RDWR) {
1096 flags = O_RDONLY;
1097 continue;
1098 }
1099 if (errno != EBUSY)
1100 return fd;
1101 usleep(delay);
1102 if (delay < 200000)
1103 delay *= 2;
1104 }
1105 return -1;
1106 }
1107
1108 int same_dev(char *one, char *two)
1109 {
1110 struct stat st1, st2;
1111 if (stat(one, &st1) != 0)
1112 return 0;
1113 if (stat(two, &st2) != 0)
1114 return 0;
1115 if ((st1.st_mode & S_IFMT) != S_IFBLK)
1116 return 0;
1117 if ((st2.st_mode & S_IFMT) != S_IFBLK)
1118 return 0;
1119 return st1.st_rdev == st2.st_rdev;
1120 }
1121
1122 void wait_for(char *dev, int fd)
1123 {
1124 int i;
1125 struct stat stb_want;
1126 long delay = 1000;
1127
1128 if (fstat(fd, &stb_want) != 0 ||
1129 (stb_want.st_mode & S_IFMT) != S_IFBLK)
1130 return;
1131
1132 for (i = 0 ; i < 25 ; i++) {
1133 struct stat stb;
1134 if (stat(dev, &stb) == 0 &&
1135 (stb.st_mode & S_IFMT) == S_IFBLK &&
1136 (stb.st_rdev == stb_want.st_rdev))
1137 return;
1138 usleep(delay);
1139 if (delay < 200000)
1140 delay *= 2;
1141 }
1142 if (i == 25)
1143 dprintf("timeout waiting for %s\n", dev);
1144 }
1145
1146 struct superswitch *superlist[] =
1147 {
1148 &super0, &super1,
1149 &super_ddf, &super_imsm,
1150 &mbr, &gpt,
1151 NULL };
1152
1153 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1154
1155 struct supertype *super_by_fd(int fd, char **subarrayp)
1156 {
1157 mdu_array_info_t array;
1158 int vers;
1159 int minor;
1160 struct supertype *st = NULL;
1161 struct mdinfo *sra;
1162 char *verstr;
1163 char version[20];
1164 int i;
1165 char *subarray = NULL;
1166 char container[32] = "";
1167
1168 sra = sysfs_read(fd, NULL, GET_VERSION);
1169
1170 if (sra) {
1171 vers = sra->array.major_version;
1172 minor = sra->array.minor_version;
1173 verstr = sra->text_version;
1174 } else {
1175 if (ioctl(fd, GET_ARRAY_INFO, &array))
1176 array.major_version = array.minor_version = 0;
1177 vers = array.major_version;
1178 minor = array.minor_version;
1179 verstr = "";
1180 }
1181
1182 if (vers != -1) {
1183 sprintf(version, "%d.%d", vers, minor);
1184 verstr = version;
1185 }
1186 if (minor == -2 && is_subarray(verstr)) {
1187 char *dev = verstr+1;
1188
1189 subarray = strchr(dev, '/');
1190 if (subarray) {
1191 *subarray++ = '\0';
1192 subarray = xstrdup(subarray);
1193 }
1194 strcpy(container, dev);
1195 sysfs_free(sra);
1196 sra = sysfs_read(-1, container, GET_VERSION);
1197 if (sra && sra->text_version[0])
1198 verstr = sra->text_version;
1199 else
1200 verstr = "-no-metadata-";
1201 }
1202
1203 for (i = 0; st == NULL && superlist[i] ; i++)
1204 st = superlist[i]->match_metadata_desc(verstr);
1205
1206 sysfs_free(sra);
1207 if (st) {
1208 st->sb = NULL;
1209 if (subarrayp)
1210 *subarrayp = subarray;
1211 strcpy(st->container_devnm, container);
1212 strcpy(st->devnm, fd2devnm(fd));
1213 } else
1214 free(subarray);
1215
1216 return st;
1217 }
1218 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1219
1220 int dev_size_from_id(dev_t id, unsigned long long *size)
1221 {
1222 char buf[20];
1223 int fd;
1224
1225 sprintf(buf, "%d:%d", major(id), minor(id));
1226 fd = dev_open(buf, O_RDONLY);
1227 if (fd < 0)
1228 return 0;
1229 if (get_dev_size(fd, NULL, size)) {
1230 close(fd);
1231 return 1;
1232 }
1233 close(fd);
1234 return 0;
1235 }
1236
1237 struct supertype *dup_super(struct supertype *orig)
1238 {
1239 struct supertype *st;
1240
1241 if (!orig)
1242 return orig;
1243 st = xcalloc(1, sizeof(*st));
1244 st->ss = orig->ss;
1245 st->max_devs = orig->max_devs;
1246 st->minor_version = orig->minor_version;
1247 st->ignore_hw_compat = orig->ignore_hw_compat;
1248 st->data_offset = orig->data_offset;
1249 st->sb = NULL;
1250 st->info = NULL;
1251 return st;
1252 }
1253
1254 struct supertype *guess_super_type(int fd, enum guess_types guess_type)
1255 {
1256 /* try each load_super to find the best match,
1257 * and return the best superswitch
1258 */
1259 struct superswitch *ss;
1260 struct supertype *st;
1261 unsigned int besttime = 0;
1262 int bestsuper = -1;
1263 int i;
1264
1265 st = xcalloc(1, sizeof(*st));
1266 st->container_devnm[0] = 0;
1267
1268 for (i = 0 ; superlist[i]; i++) {
1269 int rv;
1270 ss = superlist[i];
1271 if (guess_type == guess_array && ss->add_to_super == NULL)
1272 continue;
1273 if (guess_type == guess_partitions && ss->add_to_super != NULL)
1274 continue;
1275 memset(st, 0, sizeof(*st));
1276 st->ignore_hw_compat = 1;
1277 rv = ss->load_super(st, fd, NULL);
1278 if (rv == 0) {
1279 struct mdinfo info;
1280 st->ss->getinfo_super(st, &info, NULL);
1281 if (bestsuper == -1 ||
1282 besttime < info.array.ctime) {
1283 bestsuper = i;
1284 besttime = info.array.ctime;
1285 }
1286 ss->free_super(st);
1287 }
1288 }
1289 if (bestsuper != -1) {
1290 int rv;
1291 memset(st, 0, sizeof(*st));
1292 st->ignore_hw_compat = 1;
1293 rv = superlist[bestsuper]->load_super(st, fd, NULL);
1294 if (rv == 0) {
1295 superlist[bestsuper]->free_super(st);
1296 return st;
1297 }
1298 }
1299 free(st);
1300 return NULL;
1301 }
1302
1303 /* Return size of device in bytes */
1304 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
1305 {
1306 unsigned long long ldsize;
1307 struct stat st;
1308
1309 if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
1310 ldsize = (unsigned long long)st.st_size;
1311 else
1312 #ifdef BLKGETSIZE64
1313 if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
1314 #endif
1315 {
1316 unsigned long dsize;
1317 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
1318 ldsize = dsize;
1319 ldsize <<= 9;
1320 } else {
1321 if (dname)
1322 pr_err("Cannot get size of %s: %s\b",
1323 dname, strerror(errno));
1324 return 0;
1325 }
1326 }
1327 *sizep = ldsize;
1328 return 1;
1329 }
1330
1331 /* Return true if this can only be a container, not a member device.
1332 * i.e. is and md device and size is zero
1333 */
1334 int must_be_container(int fd)
1335 {
1336 unsigned long long size;
1337 if (md_get_version(fd) < 0)
1338 return 0;
1339 if (get_dev_size(fd, NULL, &size) == 0)
1340 return 1;
1341 if (size == 0)
1342 return 1;
1343 return 0;
1344 }
1345
1346 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1347 * Returns: 1 if successful
1348 * -1 for unknown partition type
1349 * 0 for other errors
1350 */
1351 static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart)
1352 {
1353 struct GPT gpt;
1354 unsigned char empty_gpt_entry[16]= {0};
1355 struct GPT_part_entry *part;
1356 char buf[512];
1357 unsigned long long curr_part_end;
1358 unsigned all_partitions, entry_size;
1359 unsigned part_nr;
1360
1361 *endofpart = 0;
1362
1363 BUILD_BUG_ON(sizeof(gpt) != 512);
1364 /* skip protective MBR */
1365 lseek(fd, 512, SEEK_SET);
1366 /* read GPT header */
1367 if (read(fd, &gpt, 512) != 512)
1368 return 0;
1369
1370 /* get the number of partition entries and the entry size */
1371 all_partitions = __le32_to_cpu(gpt.part_cnt);
1372 entry_size = __le32_to_cpu(gpt.part_size);
1373
1374 /* Check GPT signature*/
1375 if (gpt.magic != GPT_SIGNATURE_MAGIC)
1376 return -1;
1377
1378 /* sanity checks */
1379 if (all_partitions > 1024 ||
1380 entry_size > sizeof(buf))
1381 return -1;
1382
1383 part = (struct GPT_part_entry *)buf;
1384
1385 for (part_nr = 0; part_nr < all_partitions; part_nr++) {
1386 /* read partition entry */
1387 if (read(fd, buf, entry_size) != (ssize_t)entry_size)
1388 return 0;
1389
1390 /* is this valid partition? */
1391 if (memcmp(part->type_guid, empty_gpt_entry, 16) != 0) {
1392 /* check the last lba for the current partition */
1393 curr_part_end = __le64_to_cpu(part->ending_lba);
1394 if (curr_part_end > *endofpart)
1395 *endofpart = curr_part_end;
1396 }
1397
1398 }
1399 return 1;
1400 }
1401
1402 /* Sets endofpart parameter to the last block used by the last partition on the device.
1403 * Returns: 1 if successful
1404 * -1 for unknown partition type
1405 * 0 for other errors
1406 */
1407 static int get_last_partition_end(int fd, unsigned long long *endofpart)
1408 {
1409 struct MBR boot_sect;
1410 struct MBR_part_record *part;
1411 unsigned long long curr_part_end;
1412 unsigned part_nr;
1413 int retval = 0;
1414
1415 *endofpart = 0;
1416
1417 BUILD_BUG_ON(sizeof(boot_sect) != 512);
1418 /* read MBR */
1419 lseek(fd, 0, 0);
1420 if (read(fd, &boot_sect, 512) != 512)
1421 goto abort;
1422
1423 /* check MBP signature */
1424 if (boot_sect.magic == MBR_SIGNATURE_MAGIC) {
1425 retval = 1;
1426 /* found the correct signature */
1427 part = boot_sect.parts;
1428
1429 for (part_nr = 0; part_nr < MBR_PARTITIONS; part_nr++) {
1430 /* check for GPT type */
1431 if (part->part_type == MBR_GPT_PARTITION_TYPE) {
1432 retval = get_gpt_last_partition_end(fd, endofpart);
1433 break;
1434 }
1435 /* check the last used lba for the current partition */
1436 curr_part_end = __le32_to_cpu(part->first_sect_lba) +
1437 __le32_to_cpu(part->blocks_num);
1438 if (curr_part_end > *endofpart)
1439 *endofpart = curr_part_end;
1440
1441 part++;
1442 }
1443 } else {
1444 /* Unknown partition table */
1445 retval = -1;
1446 }
1447 abort:
1448 return retval;
1449 }
1450
1451 int check_partitions(int fd, char *dname, unsigned long long freesize,
1452 unsigned long long size)
1453 {
1454 /*
1455 * Check where the last partition ends
1456 */
1457 unsigned long long endofpart;
1458 int ret;
1459
1460 if ((ret = get_last_partition_end(fd, &endofpart)) > 0) {
1461 /* There appears to be a partition table here */
1462 if (freesize == 0) {
1463 /* partitions will not be visible in new device */
1464 pr_err("partition table exists on %s but will be lost or\n"
1465 " meaningless after creating array\n",
1466 dname);
1467 return 1;
1468 } else if (endofpart > freesize) {
1469 /* last partition overlaps metadata */
1470 pr_err("metadata will over-write last partition on %s.\n",
1471 dname);
1472 return 1;
1473 } else if (size && endofpart > size) {
1474 /* partitions will be truncated in new device */
1475 pr_err("array size is too small to cover all partitions on %s.\n",
1476 dname);
1477 return 1;
1478 }
1479 }
1480 return 0;
1481 }
1482
1483 int open_container(int fd)
1484 {
1485 /* 'fd' is a block device. Find out if it is in use
1486 * by a container, and return an open fd on that container.
1487 */
1488 char path[256];
1489 char *e;
1490 DIR *dir;
1491 struct dirent *de;
1492 int dfd, n;
1493 char buf[200];
1494 int major, minor;
1495 struct stat st;
1496
1497 if (fstat(fd, &st) != 0)
1498 return -1;
1499 sprintf(path, "/sys/dev/block/%d:%d/holders",
1500 (int)major(st.st_rdev), (int)minor(st.st_rdev));
1501 e = path + strlen(path);
1502
1503 dir = opendir(path);
1504 if (!dir)
1505 return -1;
1506 while ((de = readdir(dir))) {
1507 if (de->d_ino == 0)
1508 continue;
1509 if (de->d_name[0] == '.')
1510 continue;
1511 /* Need to make sure it is a container and not a volume */
1512 sprintf(e, "/%s/md/metadata_version", de->d_name);
1513 dfd = open(path, O_RDONLY);
1514 if (dfd < 0)
1515 continue;
1516 n = read(dfd, buf, sizeof(buf));
1517 close(dfd);
1518 if (n <= 0 || (unsigned)n >= sizeof(buf))
1519 continue;
1520 buf[n] = 0;
1521 if (strncmp(buf, "external", 8) != 0 ||
1522 n < 10 ||
1523 buf[9] == '/')
1524 continue;
1525 sprintf(e, "/%s/dev", de->d_name);
1526 dfd = open(path, O_RDONLY);
1527 if (dfd < 0)
1528 continue;
1529 n = read(dfd, buf, sizeof(buf));
1530 close(dfd);
1531 if (n <= 0 || (unsigned)n >= sizeof(buf))
1532 continue;
1533 buf[n] = 0;
1534 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
1535 continue;
1536 sprintf(buf, "%d:%d", major, minor);
1537 dfd = dev_open(buf, O_RDONLY);
1538 if (dfd >= 0) {
1539 closedir(dir);
1540 return dfd;
1541 }
1542 }
1543 closedir(dir);
1544 return -1;
1545 }
1546
1547 struct superswitch *version_to_superswitch(char *vers)
1548 {
1549 int i;
1550
1551 for (i = 0; superlist[i]; i++) {
1552 struct superswitch *ss = superlist[i];
1553
1554 if (strcmp(vers, ss->name) == 0)
1555 return ss;
1556 }
1557
1558 return NULL;
1559 }
1560
1561 int metadata_container_matches(char *metadata, char *devnm)
1562 {
1563 /* Check if 'devnm' is the container named in 'metadata'
1564 * which is
1565 * /containername/componentname or
1566 * -containername/componentname
1567 */
1568 int l;
1569 if (*metadata != '/' && *metadata != '-')
1570 return 0;
1571 l = strlen(devnm);
1572 if (strncmp(metadata+1, devnm, l) != 0)
1573 return 0;
1574 if (metadata[l+1] != '/')
1575 return 0;
1576 return 1;
1577 }
1578
1579 int metadata_subdev_matches(char *metadata, char *devnm)
1580 {
1581 /* Check if 'devnm' is the subdev named in 'metadata'
1582 * which is
1583 * /containername/subdev or
1584 * -containername/subdev
1585 */
1586 char *sl;
1587 if (*metadata != '/' && *metadata != '-')
1588 return 0;
1589 sl = strchr(metadata+1, '/');
1590 if (!sl)
1591 return 0;
1592 if (strcmp(sl+1, devnm) == 0)
1593 return 1;
1594 return 0;
1595 }
1596
1597 int is_container_member(struct mdstat_ent *mdstat, char *container)
1598 {
1599 if (mdstat->metadata_version == NULL ||
1600 strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
1601 !metadata_container_matches(mdstat->metadata_version+9, container))
1602 return 0;
1603
1604 return 1;
1605 }
1606
1607 int is_subarray_active(char *subarray, char *container)
1608 {
1609 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1610 struct mdstat_ent *ent;
1611
1612 for (ent = mdstat; ent; ent = ent->next)
1613 if (is_container_member(ent, container))
1614 if (strcmp(to_subarray(ent, container), subarray) == 0)
1615 break;
1616
1617 free_mdstat(mdstat);
1618
1619 return ent != NULL;
1620 }
1621
1622 /* open_subarray - opens a subarray in a container
1623 * @dev: container device name
1624 * @st: empty supertype
1625 * @quiet: block reporting errors flag
1626 *
1627 * On success returns an fd to a container and fills in *st
1628 */
1629 int open_subarray(char *dev, char *subarray, struct supertype *st, int quiet)
1630 {
1631 struct mdinfo *mdi;
1632 struct mdinfo *info;
1633 int fd, err = 1;
1634 char *_devnm;
1635
1636 fd = open(dev, O_RDWR|O_EXCL);
1637 if (fd < 0) {
1638 if (!quiet)
1639 pr_err("Couldn't open %s, aborting\n",
1640 dev);
1641 return -1;
1642 }
1643
1644 _devnm = fd2devnm(fd);
1645 if (_devnm == NULL) {
1646 if (!quiet)
1647 pr_err("Failed to determine device number for %s\n",
1648 dev);
1649 goto close_fd;
1650 }
1651 strcpy(st->devnm, _devnm);
1652
1653 mdi = sysfs_read(fd, st->devnm, GET_VERSION|GET_LEVEL);
1654 if (!mdi) {
1655 if (!quiet)
1656 pr_err("Failed to read sysfs for %s\n",
1657 dev);
1658 goto close_fd;
1659 }
1660
1661 if (mdi->array.level != UnSet) {
1662 if (!quiet)
1663 pr_err("%s is not a container\n", dev);
1664 goto free_sysfs;
1665 }
1666
1667 st->ss = version_to_superswitch(mdi->text_version);
1668 if (!st->ss) {
1669 if (!quiet)
1670 pr_err("Operation not supported for %s metadata\n",
1671 mdi->text_version);
1672 goto free_sysfs;
1673 }
1674
1675 if (st->devnm[0] == 0) {
1676 if (!quiet)
1677 pr_err("Failed to allocate device name\n");
1678 goto free_sysfs;
1679 }
1680
1681 if (!st->ss->load_container) {
1682 if (!quiet)
1683 pr_err("%s is not a container\n", dev);
1684 goto free_sysfs;
1685 }
1686
1687 if (st->ss->load_container(st, fd, NULL)) {
1688 if (!quiet)
1689 pr_err("Failed to load metadata for %s\n",
1690 dev);
1691 goto free_sysfs;
1692 }
1693
1694 info = st->ss->container_content(st, subarray);
1695 if (!info) {
1696 if (!quiet)
1697 pr_err("Failed to find subarray-%s in %s\n",
1698 subarray, dev);
1699 goto free_super;
1700 }
1701 free(info);
1702
1703 err = 0;
1704
1705 free_super:
1706 if (err)
1707 st->ss->free_super(st);
1708 free_sysfs:
1709 sysfs_free(mdi);
1710 close_fd:
1711 if (err)
1712 close(fd);
1713
1714 if (err)
1715 return -1;
1716 else
1717 return fd;
1718 }
1719
1720 int add_disk(int mdfd, struct supertype *st,
1721 struct mdinfo *sra, struct mdinfo *info)
1722 {
1723 /* Add a device to an array, in one of 2 ways. */
1724 int rv;
1725 #ifndef MDASSEMBLE
1726 if (st->ss->external) {
1727 if (info->disk.state & (1<<MD_DISK_SYNC))
1728 info->recovery_start = MaxSector;
1729 else
1730 info->recovery_start = 0;
1731 rv = sysfs_add_disk(sra, info, 0);
1732 if (! rv) {
1733 struct mdinfo *sd2;
1734 for (sd2 = sra->devs; sd2; sd2=sd2->next)
1735 if (sd2 == info)
1736 break;
1737 if (sd2 == NULL) {
1738 sd2 = xmalloc(sizeof(*sd2));
1739 *sd2 = *info;
1740 sd2->next = sra->devs;
1741 sra->devs = sd2;
1742 }
1743 }
1744 } else
1745 #endif
1746 rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk);
1747 return rv;
1748 }
1749
1750 int remove_disk(int mdfd, struct supertype *st,
1751 struct mdinfo *sra, struct mdinfo *info)
1752 {
1753 int rv;
1754 /* Remove the disk given by 'info' from the array */
1755 #ifndef MDASSEMBLE
1756 if (st->ss->external)
1757 rv = sysfs_set_str(sra, info, "slot", "none");
1758 else
1759 #endif
1760 rv = ioctl(mdfd, HOT_REMOVE_DISK, makedev(info->disk.major,
1761 info->disk.minor));
1762 return rv;
1763 }
1764
1765 int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
1766 {
1767 /* Initialise kernel's knowledge of array.
1768 * This varies between externally managed arrays
1769 * and older kernels
1770 */
1771 int vers = md_get_version(mdfd);
1772 int rv;
1773
1774 #ifndef MDASSEMBLE
1775 if (st->ss->external)
1776 rv = sysfs_set_array(info, vers);
1777 else
1778 #endif
1779 if ((vers % 100) >= 1) { /* can use different versions */
1780 mdu_array_info_t inf;
1781 memset(&inf, 0, sizeof(inf));
1782 inf.major_version = info->array.major_version;
1783 inf.minor_version = info->array.minor_version;
1784 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
1785 } else
1786 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
1787 return rv;
1788 }
1789
1790 unsigned long long min_recovery_start(struct mdinfo *array)
1791 {
1792 /* find the minimum recovery_start in an array for metadata
1793 * formats that only record per-array recovery progress instead
1794 * of per-device
1795 */
1796 unsigned long long recovery_start = MaxSector;
1797 struct mdinfo *d;
1798
1799 for (d = array->devs; d; d = d->next)
1800 recovery_start = min(recovery_start, d->recovery_start);
1801
1802 return recovery_start;
1803 }
1804
1805 int mdmon_pid(char *devnm)
1806 {
1807 char path[100];
1808 char pid[10];
1809 int fd;
1810 int n;
1811
1812 sprintf(path, "%s/%s.pid", MDMON_DIR, devnm);
1813
1814 fd = open(path, O_RDONLY | O_NOATIME, 0);
1815
1816 if (fd < 0)
1817 return -1;
1818 n = read(fd, pid, 9);
1819 close(fd);
1820 if (n <= 0)
1821 return -1;
1822 return atoi(pid);
1823 }
1824
1825 int mdmon_running(char *devnm)
1826 {
1827 int pid = mdmon_pid(devnm);
1828 if (pid <= 0)
1829 return 0;
1830 if (kill(pid, 0) == 0)
1831 return 1;
1832 return 0;
1833 }
1834
1835 int start_mdmon(char *devnm)
1836 {
1837 int i, skipped;
1838 int len;
1839 pid_t pid;
1840 int status;
1841 char pathbuf[1024];
1842 char *paths[4] = {
1843 pathbuf,
1844 BINDIR "/mdmon",
1845 "./mdmon",
1846 NULL
1847 };
1848
1849 if (check_env("MDADM_NO_MDMON"))
1850 return 0;
1851
1852 len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf)-1);
1853 if (len > 0) {
1854 char *sl;
1855 pathbuf[len] = 0;
1856 sl = strrchr(pathbuf, '/');
1857 if (sl)
1858 sl++;
1859 else
1860 sl = pathbuf;
1861 strcpy(sl, "mdmon");
1862 } else
1863 pathbuf[0] = '\0';
1864
1865 /* First try to run systemctl */
1866 if (!check_env("MDADM_NO_SYSTEMCTL"))
1867 switch(fork()) {
1868 case 0:
1869 /* FIXME yuk. CLOSE_EXEC?? */
1870 skipped = 0;
1871 for (i = 3; skipped < 20; i++)
1872 if (close(i) < 0)
1873 skipped++;
1874 else
1875 skipped = 0;
1876
1877 /* Don't want to see error messages from
1878 * systemctl. If the service doesn't exist,
1879 * we start mdmon ourselves.
1880 */
1881 close(2);
1882 open("/dev/null", O_WRONLY);
1883 snprintf(pathbuf, sizeof(pathbuf), "mdmon@%s.service",
1884 devnm);
1885 status = execl("/usr/bin/systemctl", "systemctl",
1886 "start",
1887 pathbuf, NULL);
1888 status = execl("/bin/systemctl", "systemctl", "start",
1889 pathbuf, NULL);
1890 exit(1);
1891 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1892 return -1;
1893 default: /* parent - good */
1894 pid = wait(&status);
1895 if (pid >= 0 && status == 0)
1896 return 0;
1897 }
1898
1899 /* That failed, try running mdmon directly */
1900 switch(fork()) {
1901 case 0:
1902 /* FIXME yuk. CLOSE_EXEC?? */
1903 skipped = 0;
1904 for (i = 3; skipped < 20; i++)
1905 if (close(i) < 0)
1906 skipped++;
1907 else
1908 skipped = 0;
1909
1910 for (i = 0; paths[i]; i++)
1911 if (paths[i][0]) {
1912 execl(paths[i], paths[i],
1913 devnm, NULL);
1914 }
1915 exit(1);
1916 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1917 return -1;
1918 default: /* parent - good */
1919 pid = wait(&status);
1920 if (pid < 0 || status != 0) {
1921 pr_err("failed to launch mdmon. Array remains readonly\n");
1922 return -1;
1923 }
1924 }
1925 return 0;
1926 }
1927
1928 __u32 random32(void)
1929 {
1930 __u32 rv;
1931 int rfd = open("/dev/urandom", O_RDONLY);
1932 if (rfd < 0 || read(rfd, &rv, 4) != 4)
1933 rv = random();
1934 if (rfd >= 0)
1935 close(rfd);
1936 return rv;
1937 }
1938
1939 void random_uuid(__u8 *buf)
1940 {
1941 int fd, i, len;
1942 __u32 r[4];
1943
1944 fd = open("/dev/urandom", O_RDONLY);
1945 if (fd < 0)
1946 goto use_random;
1947 len = read(fd, buf, 16);
1948 close(fd);
1949 if (len != 16)
1950 goto use_random;
1951
1952 return;
1953
1954 use_random:
1955 for (i = 0; i < 4; i++)
1956 r[i] = random();
1957 memcpy(buf, r, 16);
1958 }
1959
1960 #ifndef MDASSEMBLE
1961 int flush_metadata_updates(struct supertype *st)
1962 {
1963 int sfd;
1964 if (!st->updates) {
1965 st->update_tail = NULL;
1966 return -1;
1967 }
1968
1969 sfd = connect_monitor(st->container_devnm);
1970 if (sfd < 0)
1971 return -1;
1972
1973 while (st->updates) {
1974 struct metadata_update *mu = st->updates;
1975 st->updates = mu->next;
1976
1977 send_message(sfd, mu, 0);
1978 wait_reply(sfd, 0);
1979 free(mu->buf);
1980 free(mu);
1981 }
1982 ack(sfd, 0);
1983 wait_reply(sfd, 0);
1984 close(sfd);
1985 st->update_tail = NULL;
1986 return 0;
1987 }
1988
1989 void append_metadata_update(struct supertype *st, void *buf, int len)
1990 {
1991
1992 struct metadata_update *mu = xmalloc(sizeof(*mu));
1993
1994 mu->buf = buf;
1995 mu->len = len;
1996 mu->space = NULL;
1997 mu->space_list = NULL;
1998 mu->next = NULL;
1999 *st->update_tail = mu;
2000 st->update_tail = &mu->next;
2001 }
2002 #endif /* MDASSEMBLE */
2003
2004 #ifdef __TINYC__
2005 /* tinyc doesn't optimize this check in ioctl.h out ... */
2006 unsigned int __invalid_size_argument_for_IOC = 0;
2007 #endif
2008
2009 int experimental(void)
2010 {
2011 if (check_env("MDADM_EXPERIMENTAL"))
2012 return 1;
2013 else {
2014 pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n");
2015 return 0;
2016 }
2017 }
2018
2019 /* Pick all spares matching given criteria from a container
2020 * if min_size == 0 do not check size
2021 * if domlist == NULL do not check domains
2022 * if spare_group given add it to domains of each spare
2023 * metadata allows to test domains using metadata of destination array */
2024 struct mdinfo *container_choose_spares(struct supertype *st,
2025 unsigned long long min_size,
2026 struct domainlist *domlist,
2027 char *spare_group,
2028 const char *metadata, int get_one)
2029 {
2030 struct mdinfo *d, **dp, *disks = NULL;
2031
2032 /* get list of all disks in container */
2033 if (st->ss->getinfo_super_disks)
2034 disks = st->ss->getinfo_super_disks(st);
2035
2036 if (!disks)
2037 return disks;
2038 /* find spare devices on the list */
2039 dp = &disks->devs;
2040 disks->array.spare_disks = 0;
2041 while (*dp) {
2042 int found = 0;
2043 d = *dp;
2044 if (d->disk.state == 0) {
2045 /* check if size is acceptable */
2046 unsigned long long dev_size;
2047 dev_t dev = makedev(d->disk.major,d->disk.minor);
2048
2049 if (!min_size ||
2050 (dev_size_from_id(dev, &dev_size) &&
2051 dev_size >= min_size))
2052 found = 1;
2053 /* check if domain matches */
2054 if (found && domlist) {
2055 struct dev_policy *pol = devid_policy(dev);
2056 if (spare_group)
2057 pol_add(&pol, pol_domain,
2058 spare_group, NULL);
2059 if (domain_test(domlist, pol, metadata) != 1)
2060 found = 0;
2061 dev_policy_free(pol);
2062 }
2063 }
2064 if (found) {
2065 dp = &d->next;
2066 disks->array.spare_disks++;
2067 if (get_one) {
2068 sysfs_free(*dp);
2069 d->next = NULL;
2070 }
2071 } else {
2072 *dp = d->next;
2073 d->next = NULL;
2074 sysfs_free(d);
2075 }
2076 }
2077 return disks;
2078 }
2079
2080 /* Checks if paths point to the same device
2081 * Returns 0 if they do.
2082 * Returns 1 if they don't.
2083 * Returns -1 if something went wrong,
2084 * e.g. paths are empty or the files
2085 * they point to don't exist */
2086 int compare_paths (char* path1, char* path2)
2087 {
2088 struct stat st1,st2;
2089
2090 if (path1 == NULL || path2 == NULL)
2091 return -1;
2092 if (stat(path1,&st1) != 0)
2093 return -1;
2094 if (stat(path2,&st2) != 0)
2095 return -1;
2096 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev))
2097 return 0;
2098 return 1;
2099 }
2100
2101 /* Make sure we can open as many devices as needed */
2102 void enable_fds(int devices)
2103 {
2104 unsigned int fds = 20 + devices;
2105 struct rlimit lim;
2106 if (getrlimit(RLIMIT_NOFILE, &lim) != 0
2107 || lim.rlim_cur >= fds)
2108 return;
2109 if (lim.rlim_max < fds)
2110 lim.rlim_max = fds;
2111 lim.rlim_cur = fds;
2112 setrlimit(RLIMIT_NOFILE, &lim);
2113 }
2114
2115 int in_initrd(void)
2116 {
2117 /* This is based on similar function in systemd. */
2118 struct statfs s;
2119 /* statfs.f_type is signed long on s390x and MIPS, causing all
2120 sorts of sign extension problems with RAMFS_MAGIC being
2121 defined as 0x858458f6 */
2122 return statfs("/", &s) >= 0 &&
2123 ((unsigned long)s.f_type == TMPFS_MAGIC ||
2124 ((unsigned long)s.f_type & 0xFFFFFFFFUL) ==
2125 ((unsigned long)RAMFS_MAGIC & 0xFFFFFFFFUL));
2126 }
2127
2128 void reopen_mddev(int mdfd)
2129 {
2130 /* Re-open without any O_EXCL, but keep
2131 * the same fd
2132 */
2133 char *devnm;
2134 int fd;
2135 devnm = fd2devnm(mdfd);
2136 close(mdfd);
2137 fd = open_dev(devnm);
2138 if (fd >= 0 && fd != mdfd)
2139 dup2(fd, mdfd);
2140 }
2141
2142 #ifndef MDASSEMBLE
2143 static struct cmap_hooks *cmap_hooks = NULL;
2144 static int is_cmap_hooks_ready = 0;
2145
2146 void set_cmap_hooks(void)
2147 {
2148 cmap_hooks = xmalloc(sizeof(struct cmap_hooks));
2149 cmap_hooks->cmap_handle = dlopen("libcmap.so.4", RTLD_NOW | RTLD_LOCAL);
2150 if (!cmap_hooks->cmap_handle)
2151 return;
2152
2153 cmap_hooks->initialize = dlsym(cmap_hooks->cmap_handle, "cmap_initialize");
2154 cmap_hooks->get_string = dlsym(cmap_hooks->cmap_handle, "cmap_get_string");
2155 cmap_hooks->finalize = dlsym(cmap_hooks->cmap_handle, "cmap_finalize");
2156
2157 if (!cmap_hooks->initialize || !cmap_hooks->get_string ||
2158 !cmap_hooks->finalize)
2159 dlclose(cmap_hooks->cmap_handle);
2160 else
2161 is_cmap_hooks_ready = 1;
2162 }
2163
2164 int get_cluster_name(char **cluster_name)
2165 {
2166 int rv = -1;
2167 cmap_handle_t handle;
2168
2169 if (!is_cmap_hooks_ready)
2170 return rv;
2171
2172 rv = cmap_hooks->initialize(&handle);
2173 if (rv != CS_OK)
2174 goto out;
2175
2176 rv = cmap_hooks->get_string(handle, "totem.cluster_name", cluster_name);
2177 if (rv != CS_OK) {
2178 free(*cluster_name);
2179 rv = -1;
2180 goto name_err;
2181 }
2182
2183 rv = 0;
2184 name_err:
2185 cmap_hooks->finalize(handle);
2186 out:
2187 return rv;
2188 }
2189
2190 void set_dlm_hooks(void)
2191 {
2192 dlm_hooks = xmalloc(sizeof(struct dlm_hooks));
2193 dlm_hooks->dlm_handle = dlopen("libdlm_lt.so.3", RTLD_NOW | RTLD_LOCAL);
2194 if (!dlm_hooks->dlm_handle)
2195 return;
2196
2197 dlm_hooks->create_lockspace = dlsym(dlm_hooks->dlm_handle, "dlm_create_lockspace");
2198 dlm_hooks->release_lockspace = dlsym(dlm_hooks->dlm_handle, "dlm_release_lockspace");
2199 dlm_hooks->ls_lock = dlsym(dlm_hooks->dlm_handle, "dlm_ls_lock");
2200 dlm_hooks->ls_unlock = dlsym(dlm_hooks->dlm_handle, "dlm_ls_unlock");
2201 dlm_hooks->ls_get_fd = dlsym(dlm_hooks->dlm_handle, "dlm_ls_get_fd");
2202 dlm_hooks->dispatch = dlsym(dlm_hooks->dlm_handle, "dlm_dispatch");
2203
2204 if (!dlm_hooks->create_lockspace || !dlm_hooks->ls_lock ||
2205 !dlm_hooks->ls_unlock || !dlm_hooks->release_lockspace ||
2206 !dlm_hooks->ls_get_fd || !dlm_hooks->dispatch)
2207 dlclose(dlm_hooks->dlm_handle);
2208 else
2209 is_dlm_hooks_ready = 1;
2210 }
2211
2212 void set_hooks(void)
2213 {
2214 set_dlm_hooks();
2215 set_cmap_hooks();
2216 }
2217 #endif