2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
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.
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.
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
22 * Email: <neilb@suse.de>
27 #include <sys/socket.h>
28 #include <sys/utsname.h>
36 * following taken from linux/blkpg.h because they aren't
37 * anywhere else and it isn't safe to #include linux/ * stuff.
40 #define BLKPG _IO(0x12,105)
42 /* The argument structure */
43 struct blkpg_ioctl_arg
{
50 /* The subfunctions (for the op field) */
51 #define BLKPG_ADD_PARTITION 1
52 #define BLKPG_DEL_PARTITION 2
54 /* Sizes of name fields. Unused at present. */
55 #define BLKPG_DEVNAMELTH 64
56 #define BLKPG_VOLNAMELTH 64
58 /* The data structure for ADD_PARTITION and DEL_PARTITION */
59 struct blkpg_partition
{
60 long long start
; /* starting offset in bytes */
61 long long length
; /* length in bytes */
62 int pno
; /* partition number */
63 char devname
[BLKPG_DEVNAMELTH
]; /* partition name, like sda5 or c0d1p2,
64 to be used in kernel messages */
65 char volname
[BLKPG_VOLNAMELTH
]; /* volume label */
70 /* Force a compilation error if condition is true */
71 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
73 /* Force a compilation error if condition is true, but also produce a
74 result (of value 0 and type size_t), so the expression can be used
75 e.g. in a structure initializer (or where-ever else comma expressions
77 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
80 * Parse a 128 bit uuid in 4 integers
81 * format is 32 hexx nibbles with options :.<space> separator
82 * If not exactly 32 hex digits are found, return 0
85 int parse_uuid(char *str
, int uuid
[4])
87 int hit
= 0; /* number of Hex digIT */
90 for (i
=0; i
<4; i
++) uuid
[i
]=0;
96 else if (c
>='a' && c
<= 'f')
98 else if (c
>='A' && c
<= 'F')
100 else if (strchr(":. -", c
))
117 * Get the md version number.
118 * We use the RAID_VERSION ioctl if it is supported
119 * If not, but we have a block device with major '9', we assume
122 * Return version number as 24 but number - assume version parts
126 int md_get_version(int fd
)
131 if (fstat(fd
, &stb
)<0)
133 if ((S_IFMT
&stb
.st_mode
) != S_IFBLK
)
136 if (ioctl(fd
, RAID_VERSION
, &vers
) == 0)
137 return (vers
.major
*10000) + (vers
.minor
*100) + vers
.patchlevel
;
140 if (major(stb
.st_rdev
) == MD_MAJOR
)
145 int get_linux_version()
154 a
= strtoul(cp
, &cp
, 10);
155 if (*cp
!= '.') return -1;
156 b
= strtoul(cp
+1, &cp
, 10);
157 if (*cp
!= '.') return -1;
158 c
= strtoul(cp
+1, NULL
, 10);
160 return (a
*1000000)+(b
*1000)+c
;
163 int mdadm_version(char *version
)
171 cp
= strchr(version
, '-');
172 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
175 a
= strtoul(cp
, &cp
, 10);
178 b
= strtoul(cp
+1, &cp
, 10);
180 c
= strtoul(cp
+1, &cp
, 10);
183 if (*cp
!= ' ' && *cp
!= '-')
185 return (a
*1000000)+(b
*1000)+c
;
189 long long parse_size(char *size
)
191 /* parse 'size' which should be a number optionally
192 * followed by 'K', 'M', or 'G'.
193 * Without a suffix, K is assumed.
194 * Number returned is in sectors (half-K)
197 long long s
= strtoll(size
, &c
, 10);
211 s
*= 1024 * 1024 * 2;
220 int parse_layout_10(char *layout
)
224 /* Parse the layout string for raid10 */
225 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
226 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
227 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
231 if (layout
[0] == 'n')
233 else if (layout
[0] == 'o')
234 rv
= 0x10000 + (copies
<<8) + 1;
236 rv
= 1 + (copies
<<8);
240 int parse_layout_faulty(char *layout
)
242 /* Parse the layout string for 'faulty' */
243 int ln
= strcspn(layout
, "0123456789");
244 char *m
= strdup(layout
);
247 mode
= map_name(faultylayout
, m
);
251 return mode
| (atoi(layout
+ln
)<< ModeShift
);
255 void remove_partitions(int fd
)
257 /* remove partitions from this block devices.
258 * This is used for components added to an array
260 #ifdef BLKPG_DEL_PARTITION
261 struct blkpg_ioctl_arg a
;
262 struct blkpg_partition p
;
264 a
.op
= BLKPG_DEL_PARTITION
;
266 a
.datalen
= sizeof(p
);
268 memset(a
.data
, 0, a
.datalen
);
269 for (p
.pno
=0; p
.pno
< 16; p
.pno
++)
270 ioctl(fd
, BLKPG
, &a
);
274 int test_partition(int fd
)
276 /* Check if fd is a whole-disk or a partition.
277 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
278 * will return ENXIO on an invalid partition number.
280 struct blkpg_ioctl_arg a
;
281 struct blkpg_partition p
;
282 a
.op
= BLKPG_DEL_PARTITION
;
284 a
.datalen
= sizeof(p
);
286 memset(a
.data
, 0, a
.datalen
);
288 if (ioctl(fd
, BLKPG
, &a
) == 0)
289 /* Very unlikely, but not a partition */
292 /* not a partition */
299 int enough(int level
, int raid_disks
, int layout
, int clean
,
300 char *avail
, int avail_disks
)
305 /* This is the tricky one - we need to check
306 * which actual disks are present.
308 copies
= (layout
&255)* ((layout
>>8) & 255);
311 /* there must be one of the 'copies' form 'first' */
317 first
= (first
+1) % raid_disks
;
322 } while (first
!= 0);
325 case LEVEL_MULTIPATH
:
326 return avail_disks
>= 1;
329 return avail_disks
== raid_disks
;
331 return avail_disks
>= 1;
335 return avail_disks
>= raid_disks
-1;
337 return avail_disks
>= raid_disks
;
340 return avail_disks
>= raid_disks
-2;
342 return avail_disks
>= raid_disks
;
348 int enough_fd(int fd
)
350 struct mdu_array_info_s array
;
351 struct mdu_disk_info_s disk
;
356 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0 ||
357 array
.raid_disks
<= 0)
359 avail
= calloc(array
.raid_disks
, 1);
360 for (i
=0; i
<array
.raid_disks
+ array
.nr_disks
; i
++) {
362 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
364 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
366 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
369 avail
[disk
.raid_disk
] = 1;
371 /* This is used on an active array, so assume it is clean */
372 return enough(array
.level
, array
.raid_disks
, array
.layout
,
378 const int uuid_match_any
[4] = { ~0, ~0, ~0, ~0 };
379 int same_uuid(int a
[4], int b
[4], int swapuuid
)
381 if (memcmp(a
, uuid_match_any
, sizeof(int[4])) == 0 ||
382 memcmp(b
, uuid_match_any
, sizeof(int[4])) == 0)
386 /* parse uuids are hostendian.
387 * uuid's from some superblocks are big-ending
388 * if there is a difference, we need to swap..
390 unsigned char *ac
= (unsigned char *)a
;
391 unsigned char *bc
= (unsigned char *)b
;
393 for (i
=0; i
<16; i
+= 4) {
394 if (ac
[i
+0] != bc
[i
+3] ||
395 ac
[i
+1] != bc
[i
+2] ||
396 ac
[i
+2] != bc
[i
+1] ||
410 void copy_uuid(void *a
, int b
[4], int swapuuid
)
413 /* parse uuids are hostendian.
414 * uuid's from some superblocks are big-ending
415 * if there is a difference, we need to swap..
417 unsigned char *ac
= (unsigned char *)a
;
418 unsigned char *bc
= (unsigned char *)b
;
420 for (i
=0; i
<16; i
+= 4) {
430 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
437 copy_uuid(uuid
, id
, swap
);
438 for (i
= 0; i
< 4; i
++) {
441 for (j
= 3; j
>= 0; j
--) {
442 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
450 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
452 // dirty hack to work around an issue with super1 superblocks...
453 // super1 superblocks need swapuuid set in order for assembly to
454 // work, but can't have it set if we want this printout to match
455 // all the other uuid printouts in super1.c, so we force swapuuid
456 // to 1 to make our printout match the rest of super1
457 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
461 int check_ext2(int fd
, char *name
)
464 * Check for an ext2fs file system.
465 * Superblock is always 1K at 1K offset
467 * s_magic is le16 at 56 == 0xEF53
468 * report mtime - le32 at 44
470 * logblksize - le32 at 24
472 unsigned char sb
[1024];
475 if (lseek(fd
, 1024,0)!= 1024)
477 if (read(fd
, sb
, 1024)!= 1024)
479 if (sb
[56] != 0x53 || sb
[57] != 0xef)
482 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
483 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
484 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
485 fprintf(stderr
, Name
": %s appears to contain an ext2fs file system\n",
487 fprintf(stderr
," size=%dK mtime=%s",
488 size
*(1<<bsize
), ctime(&mtime
));
492 int check_reiser(int fd
, char *name
)
495 * superblock is at 64K
497 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
500 unsigned char sb
[1024];
502 if (lseek(fd
, 64*1024, 0) != 64*1024)
504 if (read(fd
, sb
, 1024) != 1024)
506 if (strncmp((char*)sb
+52, "ReIsErFs",8)!=0 &&
507 strncmp((char*)sb
+52, "ReIsEr2Fs",9)!=0)
509 fprintf(stderr
, Name
": %s appears to contain a reiserfs file system\n",name
);
510 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
511 fprintf(stderr
, " size = %luK\n", size
*4);
516 int check_raid(int fd
, char *name
)
521 struct supertype
*st
= guess_super(fd
);
524 st
->ss
->load_super(st
, fd
, name
);
525 /* Looks like a raid array .. */
526 fprintf(stderr
, Name
": %s appears to be part of a raid array:\n",
528 st
->ss
->getinfo_super(st
, &info
, NULL
);
529 st
->ss
->free_super(st
);
530 crtime
= info
.array
.ctime
;
531 level
= map_num(pers
, info
.array
.level
);
532 if (!level
) level
= "-unknown-";
533 fprintf(stderr
, " level=%s devices=%d ctime=%s",
534 level
, info
.array
.raid_disks
, ctime(&crtime
));
542 for (i
=0; i
<5; i
++) {
544 fprintf(stderr
, "%s%s", mesg
, add
);
546 if (fgets(buf
, 100, stdin
)==NULL
)
548 if (buf
[0]=='y' || buf
[0]=='Y')
550 if (buf
[0]=='n' || buf
[0]=='N')
554 fprintf(stderr
, Name
": assuming 'no'\n");
557 #endif /* MDASSEMBLE */
559 char *map_num(mapping_t
*map
, int num
)
569 int map_name(mapping_t
*map
, char *name
)
572 if (strcmp(map
->name
, name
)==0)
580 int is_standard(char *dev
, int *nump
)
582 /* tests if dev is a "standard" md dev name.
583 * i.e if the last component is "/dNN" or "/mdNN",
584 * where NN is a string of digits
585 * Returns 1 if a partitionable standard,
586 * -1 if non-partitonable,
587 * 0 if not a standard name.
589 char *d
= strrchr(dev
, '/');
594 if (strncmp(d
, "/d",2)==0)
595 d
+= 2, type
=1; /* /dev/md/dN{pM} */
596 else if (strncmp(d
, "/md_d", 5)==0)
597 d
+= 5, type
=1; /* /dev/md_dN{pM} */
598 else if (strncmp(d
, "/md", 3)==0)
599 d
+= 3, type
=-1; /* /dev/mdN */
600 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3)==0)
601 d
+= 1, type
=-1; /* /dev/md/N */
611 if (nump
) *nump
= num
;
618 * convert a major/minor pair for a block device into a name in /dev, if possible.
619 * On the first call, walk /dev collecting name.
620 * Put them in a simple linked listfor now.
627 int devlist_ready
= 0;
629 int add_dev(const char *name
, const struct stat
*stb
, int flag
, struct FTW
*s
)
633 if (S_ISLNK(stb
->st_mode
)) {
634 if (stat(name
, &st
) != 0)
639 if ((stb
->st_mode
&S_IFMT
)== S_IFBLK
) {
640 char *n
= strdup(name
);
641 struct devmap
*dm
= malloc(sizeof(*dm
));
642 if (strncmp(n
, "/dev/./", 7)==0)
645 dm
->major
= major(stb
->st_rdev
);
646 dm
->minor
= minor(stb
->st_rdev
);
657 int add_dev_1(const char *name
, const struct stat
*stb
, int flag
)
659 return add_dev(name
, stb
, flag
, NULL
);
661 int nftw(const char *path
, int (*han
)(const char *name
, const struct stat
*stb
, int flag
, struct FTW
*s
), int nopenfd
, int flags
)
663 return ftw(path
, add_dev_1
, nopenfd
);
666 int nftw(const char *path
, int (*han
)(const char *name
, const struct stat
*stb
, int flag
, struct FTW
*s
), int nopenfd
, int flags
)
670 #endif /* HAVE_FTW */
671 #endif /* HAVE_NFTW */
674 * Find a block device with the right major/minor number.
675 * If we find multiple names, choose the shortest.
676 * If we find a name in /dev/md/, we prefer that.
677 * This applies only to names for MD devices.
679 char *map_dev(int major
, int minor
, int create
)
682 char *regular
= NULL
, *preferred
=NULL
;
685 if (major
== 0 && minor
== 0)
689 if (!devlist_ready
) {
693 struct devmap
*d
= devlist
;
698 if (lstat(dev
, &stb
)==0 &&
699 S_ISLNK(stb
.st_mode
))
701 nftw(dev
, add_dev
, 10, FTW_PHYS
);
706 for (p
=devlist
; p
; p
=p
->next
)
707 if (p
->major
== major
&&
709 if (strncmp(p
->name
, "/dev/md/",8) == 0) {
710 if (preferred
== NULL
||
711 strlen(p
->name
) < strlen(preferred
))
714 if (regular
== NULL
||
715 strlen(p
->name
) < strlen(regular
))
719 if (!regular
&& !preferred
&& !did_check
) {
723 if (create
&& !regular
&& !preferred
) {
725 snprintf(buf
, sizeof(buf
), "%d:%d", major
, minor
);
729 return preferred
? preferred
: regular
;
732 unsigned long calc_csum(void *super
, int bytes
)
734 unsigned long long newcsum
= 0;
737 unsigned int *superc
= (unsigned int*) super
;
739 for(i
=0; i
<bytes
/4; i
++)
741 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
743 /* The in-kernel checksum calculation is always 16bit on
744 * the alpha, though it is 32 bit on i386...
745 * I wonder what it is elsewhere... (it uses and API in
746 * a way that it shouldn't).
748 csum
= (csum
& 0xffff) + (csum
>> 16);
749 csum
= (csum
& 0xffff) + (csum
>> 16);
755 char *human_size(long long bytes
)
759 /* We convert bytes to either centi-M{ega,ibi}bytes or
760 * centi-G{igi,ibi}bytes, with appropriate rounding,
761 * and then print 1/100th of those as a decimal.
762 * We allow upto 2048Megabytes before converting to
763 * gigabytes, as that shows more precision and isn't
764 * too large a number.
765 * Terrabytes are not yet handled.
768 if (bytes
< 5000*1024)
770 else if (bytes
< 2*1024LL*1024LL*1024LL) {
771 long cMiB
= (bytes
/ ( (1LL<<20) / 200LL ) +1) /2;
772 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
773 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
774 cMiB
/100 , cMiB
% 100,
777 long cGiB
= (bytes
/ ( (1LL<<30) / 200LL ) +1) /2;
778 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
779 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
780 cGiB
/100 , cGiB
% 100,
786 char *human_size_brief(long long bytes
)
790 if (bytes
< 5000*1024)
791 snprintf(buf
, sizeof(buf
), "%ld.%02ldKiB",
792 (long)(bytes
>>10), (long)(((bytes
&1023)*100+512)/1024)
794 else if (bytes
< 2*1024LL*1024LL*1024LL)
795 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
797 (long)((bytes
&0xfffff)+0x100000/200)/(0x100000/100)
800 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
802 (long)(((bytes
>>10)&0xfffff)+0x100000/200)/(0x100000/100)
807 void print_r10_layout(int layout
)
809 int near
= layout
& 255;
810 int far
= (layout
>> 8) & 255;
811 int offset
= (layout
&0x10000);
815 printf("%s near=%d", sep
, near
);
819 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
821 printf("NO REDUNDANCY");
825 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
826 int chunksize
, unsigned long long devsize
)
830 case 0: data_disks
= raid_disks
; break;
831 case 1: data_disks
= 1; break;
833 case 5: data_disks
= raid_disks
- 1; break;
834 case 6: data_disks
= raid_disks
- 2; break;
835 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
838 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
839 return data_disks
* devsize
;
842 int get_mdp_major(void)
844 static int mdp_major
= -1;
848 int have_devices
= 0;
853 fl
= fopen("/proc/devices", "r");
856 while ((w
= conf_word(fl
, 1))) {
857 if (have_block
&& strcmp(w
, "devices:")==0)
859 have_block
= (strcmp(w
, "Block")==0);
862 if (have_devices
&& strcmp(w
, "mdp")==0)
863 mdp_major
= last_num
;
870 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
871 char *get_md_name(int dev
)
873 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
874 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
875 static char devname
[50];
881 int mdp
= get_mdp_major();
882 if (mdp
< 0) return NULL
;
883 rdev
= makedev(mdp
, (-1-dev
)<<6);
884 snprintf(devname
, sizeof(devname
), "/dev/md/d%d", -1-dev
);
885 if (stat(devname
, &stb
) == 0
886 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
887 && (stb
.st_rdev
== rdev
))
890 rdev
= makedev(MD_MAJOR
, dev
);
891 snprintf(devname
, sizeof(devname
), "/dev/md%d", dev
);
892 if (stat(devname
, &stb
) == 0
893 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
894 && (stb
.st_rdev
== rdev
))
897 snprintf(devname
, sizeof(devname
), "/dev/md/%d", dev
);
898 if (stat(devname
, &stb
) == 0
899 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
900 && (stb
.st_rdev
== rdev
))
903 dn
= map_dev(major(rdev
), minor(rdev
), 0);
906 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md%d", dev
);
907 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
911 if (stat(devname
, &stb
) == 0
912 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
913 && (stb
.st_rdev
== rdev
))
919 void put_md_name(char *name
)
921 if (strncmp(name
, "/dev/.tmp.md", 12)==0)
925 int find_free_devnum(int use_partitions
)
928 for (devnum
= 127; devnum
!= 128;
929 devnum
= devnum
? devnum
-1 : (1<<20)-1) {
933 _devnum
= use_partitions
? (-1-devnum
) : devnum
;
934 if (mddev_busy(_devnum
))
936 /* make sure it is new to /dev too, at least as a
938 dn
= map_dev(dev2major(_devnum
), dev2minor(_devnum
), 0);
939 if (dn
&& ! is_standard(dn
, NULL
))
945 return use_partitions
? (-1-devnum
) : devnum
;
947 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
949 int dev_open(char *dev
, int flags
)
951 /* like 'open', but if 'dev' matches %d:%d, create a temp
952 * block device and open that
963 major
= strtoul(dev
, &e
, 0);
964 if (e
> dev
&& *e
== ':' && e
[1] &&
965 (minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
967 char *path
= map_dev(major
, minor
, 0);
969 fd
= open(path
, flags
);
971 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
972 (int)getpid(), major
, minor
);
973 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
))==0) {
974 fd
= open(devname
, flags
);
979 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
980 (int)getpid(), major
, minor
);
981 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
))==0) {
982 fd
= open(devname
, flags
);
987 fd
= open(dev
, flags
);
991 int open_dev(int devnum
)
995 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
996 return dev_open(buf
, O_RDWR
);
999 int open_dev_excl(int devnum
)
1004 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
1005 for (i
=0 ; i
<25 ; i
++) {
1006 int fd
= dev_open(buf
, O_RDWR
|O_EXCL
);
1016 int same_dev(char *one
, char *two
)
1018 struct stat st1
, st2
;
1019 if (stat(one
, &st1
) != 0)
1021 if (stat(two
, &st2
) != 0)
1023 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1025 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1027 return st1
.st_rdev
== st2
.st_rdev
;
1030 void wait_for(char *dev
, int fd
)
1033 struct stat stb_want
;
1035 if (fstat(fd
, &stb_want
) != 0 ||
1036 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1039 for (i
=0 ; i
<25 ; i
++) {
1041 if (stat(dev
, &stb
) == 0 &&
1042 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1043 (stb
.st_rdev
== stb_want
.st_rdev
))
1048 dprintf("%s: timeout waiting for %s\n", __func__
, dev
);
1051 struct superswitch
*superlist
[] =
1054 &super_ddf
, &super_imsm
,
1058 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1060 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1062 mdu_array_info_t array
;
1065 struct supertype
*st
= NULL
;
1070 char *subarray
= NULL
;
1071 int container
= NoMdDev
;
1073 sra
= sysfs_read(fd
, 0, GET_VERSION
);
1076 vers
= sra
->array
.major_version
;
1077 minor
= sra
->array
.minor_version
;
1078 verstr
= sra
->text_version
;
1080 if (ioctl(fd
, GET_ARRAY_INFO
, &array
))
1081 array
.major_version
= array
.minor_version
= 0;
1082 vers
= array
.major_version
;
1083 minor
= array
.minor_version
;
1088 sprintf(version
, "%d.%d", vers
, minor
);
1091 if (minor
== -2 && is_subarray(verstr
)) {
1092 char *dev
= verstr
+1;
1094 subarray
= strchr(dev
, '/');
1097 subarray
= strdup(subarray
);
1098 container
= devname2devnum(dev
);
1101 sra
= sysfs_read(-1, container
, GET_VERSION
);
1102 if (sra
&& sra
->text_version
[0])
1103 verstr
= sra
->text_version
;
1105 verstr
= "-no-metadata-";
1108 for (i
= 0; st
== NULL
&& superlist
[i
] ; i
++)
1109 st
= superlist
[i
]->match_metadata_desc(verstr
);
1116 *subarrayp
= subarray
;
1117 st
->container_dev
= container
;
1118 st
->devnum
= fd2devnum(fd
);
1124 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1126 int dev_size_from_id(unsigned int id
, unsigned long long *size
)
1131 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1132 fd
= dev_open(buf
, O_RDONLY
);
1135 if (get_dev_size(fd
, NULL
, size
)) {
1143 struct supertype
*dup_super(struct supertype
*orig
)
1145 struct supertype
*st
;
1149 st
= malloc(sizeof(*st
));
1152 memset(st
, 0, sizeof(*st
));
1154 st
->max_devs
= orig
->max_devs
;
1155 st
->minor_version
= orig
->minor_version
;
1161 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1163 /* try each load_super to find the best match,
1164 * and return the best superswitch
1166 struct superswitch
*ss
;
1167 struct supertype
*st
;
1168 time_t besttime
= 0;
1172 st
= malloc(sizeof(*st
));
1173 memset(st
, 0, sizeof(*st
));
1174 st
->container_dev
= NoMdDev
;
1176 for (i
=0 ; superlist
[i
]; i
++) {
1179 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1181 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1183 memset(st
, 0, sizeof(*st
));
1184 rv
= ss
->load_super(st
, fd
, NULL
);
1187 st
->ss
->getinfo_super(st
, &info
, NULL
);
1188 if (bestsuper
== -1 ||
1189 besttime
< info
.array
.ctime
) {
1191 besttime
= info
.array
.ctime
;
1196 if (bestsuper
!= -1) {
1198 memset(st
, 0, sizeof(*st
));
1199 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1201 superlist
[bestsuper
]->free_super(st
);
1209 /* Return size of device in bytes */
1210 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1212 unsigned long long ldsize
;
1215 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1216 ldsize
= (unsigned long long)st
.st_size
;
1219 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1222 unsigned long dsize
;
1223 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1228 fprintf(stderr
, Name
": Cannot get size of %s: %s\b",
1229 dname
, strerror(errno
));
1237 /* Return true if this can only be a container, not a member device.
1238 * i.e. is and md device and size is zero
1240 int must_be_container(int fd
)
1242 unsigned long long size
;
1243 if (md_get_version(fd
) < 0)
1245 if (get_dev_size(fd
, NULL
, &size
) == 0)
1252 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1253 * Returns: 1 if successful
1254 * -1 for unknown partition type
1255 * 0 for other errors
1257 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1260 unsigned char buf
[512];
1261 unsigned char empty_gpt_entry
[16]= {0};
1262 struct GPT_part_entry
*part
;
1263 unsigned long long curr_part_end
;
1264 unsigned all_partitions
, entry_size
;
1269 BUILD_BUG_ON(sizeof(gpt
) != 512);
1270 /* read GPT header */
1271 lseek(fd
, 512, SEEK_SET
);
1272 if (read(fd
, &gpt
, 512) != 512)
1275 /* get the number of partition entries and the entry size */
1276 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1277 entry_size
= __le32_to_cpu(gpt
.part_size
);
1279 /* Check GPT signature*/
1280 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1284 if (all_partitions
> 1024 ||
1288 /* read first GPT partition entries */
1289 if (read(fd
, buf
, 512) != 512)
1292 part
= (struct GPT_part_entry
*)buf
;
1294 for (part_nr
=0; part_nr
< all_partitions
; part_nr
++) {
1295 /* is this valid partition? */
1296 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1297 /* check the last lba for the current partition */
1298 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1299 if (curr_part_end
> *endofpart
)
1300 *endofpart
= curr_part_end
;
1303 part
= (struct GPT_part_entry
*)((unsigned char*)part
+ entry_size
);
1305 if ((unsigned char *)part
>= buf
+ 512) {
1306 if (read(fd
, buf
, 512) != 512)
1308 part
= (struct GPT_part_entry
*)buf
;
1314 /* Sets endofpart parameter to the last block used by the last partition on the device.
1315 * Returns: 1 if successful
1316 * -1 for unknown partition type
1317 * 0 for other errors
1319 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1321 struct MBR boot_sect
;
1322 struct MBR_part_record
*part
;
1323 unsigned long long curr_part_end
;
1329 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1332 if (read(fd
, &boot_sect
, 512) != 512)
1335 /* check MBP signature */
1336 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1338 /* found the correct signature */
1339 part
= boot_sect
.parts
;
1341 for (part_nr
=0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1342 /* check for GPT type */
1343 if (part
->part_type
== MBR_GPT_PARTITION_TYPE
) {
1344 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1347 /* check the last used lba for the current partition */
1348 curr_part_end
= __le32_to_cpu(part
->first_sect_lba
) +
1349 __le32_to_cpu(part
->blocks_num
);
1350 if (curr_part_end
> *endofpart
)
1351 *endofpart
= curr_part_end
;
1356 /* Unknown partition table */
1363 int check_partitions(int fd
, char *dname
, unsigned long long freesize
)
1366 * Check where the last partition ends
1368 unsigned long long endofpart
;
1371 if ((ret
= get_last_partition_end(fd
, &endofpart
)) > 0) {
1372 /* There appears to be a partition table here */
1373 if (freesize
== 0) {
1374 /* partitions will not be visible in new device */
1376 Name
": partition table exists on %s but will be lost or\n"
1377 " meaningless after creating array\n",
1380 } else if (endofpart
> freesize
) {
1381 /* last partition overlaps metadata */
1383 Name
": metadata will over-write last partition on %s.\n",
1391 void get_one_disk(int mdfd
, mdu_array_info_t
*ainf
, mdu_disk_info_t
*disk
)
1394 ioctl(mdfd
, GET_ARRAY_INFO
, ainf
);
1395 for (d
= 0 ; d
< ainf
->raid_disks
+ ainf
->nr_disks
; d
++)
1396 if (ioctl(mdfd
, GET_DISK_INFO
, disk
) == 0)
1400 int open_container(int fd
)
1402 /* 'fd' is a block device. Find out if it is in use
1403 * by a container, and return an open fd on that container.
1414 if (fstat(fd
, &st
) != 0)
1416 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1417 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1418 e
= path
+ strlen(path
);
1420 dir
= opendir(path
);
1423 while ((de
= readdir(dir
))) {
1426 if (de
->d_name
[0] == '.')
1428 sprintf(e
, "/%s/dev", de
->d_name
);
1429 dfd
= open(path
, O_RDONLY
);
1432 n
= read(dfd
, buf
, sizeof(buf
));
1434 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1437 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1439 sprintf(buf
, "%d:%d", major
, minor
);
1440 dfd
= dev_open(buf
, O_RDONLY
);
1450 struct superswitch
*version_to_superswitch(char *vers
)
1454 for (i
= 0; superlist
[i
]; i
++) {
1455 struct superswitch
*ss
= superlist
[i
];
1457 if (strcmp(vers
, ss
->name
) == 0)
1464 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1466 if (mdstat
->metadata_version
== NULL
||
1467 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1468 !is_subarray(mdstat
->metadata_version
+9) ||
1469 strncmp(mdstat
->metadata_version
+10, container
, strlen(container
)) != 0 ||
1470 mdstat
->metadata_version
[10+strlen(container
)] != '/')
1476 int is_subarray_active(char *subarray
, char *container
)
1478 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1479 struct mdstat_ent
*ent
;
1481 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1482 if (is_container_member(ent
, container
))
1484 strcmp(to_subarray(ent
, container
), subarray
) == 0)
1487 free_mdstat(mdstat
);
1492 int is_container_active(char *container
)
1494 return is_subarray_active(NULL
, container
);
1497 /* open_subarray - opens a subarray in a container
1498 * @dev: container device name
1499 * @st: empty supertype
1500 * @quiet: block reporting errors flag
1502 * On success returns an fd to a container and fills in *st
1504 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1507 struct mdinfo
*info
;
1510 fd
= open(dev
, O_RDWR
|O_EXCL
);
1513 fprintf(stderr
, Name
": Couldn't open %s, aborting\n",
1518 st
->devnum
= fd2devnum(fd
);
1519 if (st
->devnum
== NoMdDev
) {
1522 Name
": Failed to determine device number for %s\n",
1527 mdi
= sysfs_read(fd
, st
->devnum
, GET_VERSION
|GET_LEVEL
);
1530 fprintf(stderr
, Name
": Failed to read sysfs for %s\n",
1535 if (mdi
->array
.level
!= UnSet
) {
1537 fprintf(stderr
, Name
": %s is not a container\n", dev
);
1541 st
->ss
= version_to_superswitch(mdi
->text_version
);
1545 Name
": Operation not supported for %s metadata\n",
1550 st
->devname
= devnum2devname(st
->devnum
);
1553 fprintf(stderr
, Name
": Failed to allocate device name\n");
1557 if (!st
->ss
->load_container
) {
1559 fprintf(stderr
, Name
": %s is not a container\n", dev
);
1563 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1565 fprintf(stderr
, Name
": Failed to load metadata for %s\n",
1570 info
= st
->ss
->container_content(st
, subarray
);
1573 fprintf(stderr
, Name
": Failed to find subarray-%s in %s\n",
1583 st
->ss
->free_super(st
);
1599 int add_disk(int mdfd
, struct supertype
*st
,
1600 struct mdinfo
*sra
, struct mdinfo
*info
)
1602 /* Add a device to an array, in one of 2 ways. */
1605 if (st
->ss
->external
) {
1606 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1607 info
->recovery_start
= MaxSector
;
1609 info
->recovery_start
= 0;
1610 rv
= sysfs_add_disk(sra
, info
, 0);
1613 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1617 sd2
= malloc(sizeof(*sd2
));
1619 sd2
->next
= sra
->devs
;
1625 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1629 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1631 /* Initialise kernel's knowledge of array.
1632 * This varies between externally managed arrays
1635 int vers
= md_get_version(mdfd
);
1639 if (st
->ss
->external
)
1640 rv
= sysfs_set_array(info
, vers
);
1643 if ((vers
% 100) >= 1) { /* can use different versions */
1644 mdu_array_info_t inf
;
1645 memset(&inf
, 0, sizeof(inf
));
1646 inf
.major_version
= info
->array
.major_version
;
1647 inf
.minor_version
= info
->array
.minor_version
;
1648 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1650 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1654 unsigned long long min_recovery_start(struct mdinfo
*array
)
1656 /* find the minimum recovery_start in an array for metadata
1657 * formats that only record per-array recovery progress instead
1660 unsigned long long recovery_start
= MaxSector
;
1663 for (d
= array
->devs
; d
; d
= d
->next
)
1664 recovery_start
= min(recovery_start
, d
->recovery_start
);
1666 return recovery_start
;
1669 char *devnum2devname(int num
)
1673 sprintf(name
, "md%d", num
);
1675 sprintf(name
, "md_d%d", -1-num
);
1676 return strdup(name
);
1679 int devname2devnum(char *name
)
1683 if (strncmp(name
, "md_d", 4)==0)
1684 num
= -1-strtoul(name
+4, &ep
, 10);
1686 num
= strtoul(name
+2, &ep
, 10);
1690 int stat2devnum(struct stat
*st
)
1697 if ((S_IFMT
& st
->st_mode
) == S_IFBLK
) {
1698 if (major(st
->st_rdev
) == MD_MAJOR
)
1699 return minor(st
->st_rdev
);
1700 else if (major(st
->st_rdev
) == (unsigned)get_mdp_major())
1701 return -1- (minor(st
->st_rdev
)>>MdpMinorShift
);
1703 /* must be an extended-minor partition. Look at the
1704 * /sys/dev/block/%d:%d link which must look like
1705 * ../../block/mdXXX/mdXXXpYY
1707 sprintf(path
, "/sys/dev/block/%d:%d", major(st
->st_rdev
),
1708 minor(st
->st_rdev
));
1709 n
= readlink(path
, link
, sizeof(link
)-1);
1713 cp
= strrchr(link
, '/');
1715 cp
= strchr(link
, '/');
1716 if (cp
&& strncmp(cp
, "/md", 3) == 0)
1717 return devname2devnum(cp
+1);
1723 int fd2devnum(int fd
)
1726 if (fstat(fd
, &stb
) == 0)
1727 return stat2devnum(&stb
);
1731 int mdmon_pid(int devnum
)
1737 char *devname
= devnum2devname(devnum
);
1739 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devname
);
1742 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1746 n
= read(fd
, pid
, 9);
1753 int mdmon_running(int devnum
)
1755 int pid
= mdmon_pid(devnum
);
1758 if (kill(pid
, 0) == 0)
1763 int start_mdmon(int devnum
)
1777 if (check_env("MDADM_NO_MDMON"))
1780 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
));
1784 sl
= strrchr(pathbuf
, '/');
1789 strcpy(sl
, "mdmon");
1795 /* FIXME yuk. CLOSE_EXEC?? */
1796 for (i
=3; i
< 100; i
++)
1798 for (i
=0; paths
[i
]; i
++)
1800 execl(paths
[i
], "mdmon",
1801 devnum2devname(devnum
),
1804 case -1: fprintf(stderr
, Name
": cannot run mdmon. "
1805 "Array remains readonly\n");
1807 default: /* parent - good */
1808 pid
= wait(&status
);
1809 if (pid
< 0 || status
!= 0)
1815 int check_env(char *name
)
1817 char *val
= getenv(name
);
1819 if (val
&& atoi(val
) == 1)
1825 __u32
random32(void)
1828 int rfd
= open("/dev/urandom", O_RDONLY
);
1829 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1837 int flush_metadata_updates(struct supertype
*st
)
1841 st
->update_tail
= NULL
;
1845 sfd
= connect_monitor(devnum2devname(st
->container_dev
));
1849 while (st
->updates
) {
1850 struct metadata_update
*mu
= st
->updates
;
1851 st
->updates
= mu
->next
;
1853 send_message(sfd
, mu
, 0);
1861 st
->update_tail
= NULL
;
1865 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
1868 struct metadata_update
*mu
= malloc(sizeof(*mu
));
1874 *st
->update_tail
= mu
;
1875 st
->update_tail
= &mu
->next
;
1877 #endif /* MDASSEMBLE */
1880 /* tinyc doesn't optimize this check in ioctl.h out ... */
1881 unsigned int __invalid_size_argument_for_IOC
= 0;