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()
149 int a
= 0, b
= 0,c
= 0;
154 a
= strtoul(cp
, &cp
, 10);
156 b
= strtoul(cp
+1, &cp
, 10);
158 c
= strtoul(cp
+1, &cp
, 10);
160 return (a
*1000000)+(b
*1000)+c
;
164 int mdadm_version(char *version
)
172 cp
= strchr(version
, '-');
173 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
176 a
= strtoul(cp
, &cp
, 10);
179 b
= strtoul(cp
+1, &cp
, 10);
181 c
= strtoul(cp
+1, &cp
, 10);
184 if (*cp
!= ' ' && *cp
!= '-')
186 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 */
298 int test_partition_from_id(dev_t id
)
303 sprintf(buf
, "%d:%d", major(id
), minor(id
));
304 fd
= dev_open(buf
, O_RDONLY
);
307 rv
= test_partition(fd
);
312 int enough(int level
, int raid_disks
, int layout
, int clean
,
313 char *avail
, int avail_disks
)
318 /* This is the tricky one - we need to check
319 * which actual disks are present.
321 copies
= (layout
&255)* ((layout
>>8) & 255);
324 /* there must be one of the 'copies' form 'first' */
330 first
= (first
+1) % raid_disks
;
335 } while (first
!= 0);
338 case LEVEL_MULTIPATH
:
339 return avail_disks
>= 1;
342 return avail_disks
== raid_disks
;
344 return avail_disks
>= 1;
348 return avail_disks
>= raid_disks
-1;
350 return avail_disks
>= raid_disks
;
353 return avail_disks
>= raid_disks
-2;
355 return avail_disks
>= raid_disks
;
361 int enough_fd(int fd
)
363 struct mdu_array_info_s array
;
364 struct mdu_disk_info_s disk
;
369 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0 ||
370 array
.raid_disks
<= 0)
372 avail
= calloc(array
.raid_disks
, 1);
373 for (i
=0; i
< 1024 && array
.nr_disks
> 0; i
++) {
375 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
377 if (disk
.major
== 0 && disk
.minor
== 0)
381 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
383 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
386 avail
[disk
.raid_disk
] = 1;
388 /* This is used on an active array, so assume it is clean */
389 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
390 1, avail
, avail_disks
);
396 const int uuid_zero
[4] = { 0, 0, 0, 0 };
398 int same_uuid(int a
[4], int b
[4], int swapuuid
)
401 /* parse uuids are hostendian.
402 * uuid's from some superblocks are big-ending
403 * if there is a difference, we need to swap..
405 unsigned char *ac
= (unsigned char *)a
;
406 unsigned char *bc
= (unsigned char *)b
;
408 for (i
=0; i
<16; i
+= 4) {
409 if (ac
[i
+0] != bc
[i
+3] ||
410 ac
[i
+1] != bc
[i
+2] ||
411 ac
[i
+2] != bc
[i
+1] ||
425 void copy_uuid(void *a
, int b
[4], int swapuuid
)
428 /* parse uuids are hostendian.
429 * uuid's from some superblocks are big-ending
430 * if there is a difference, we need to swap..
432 unsigned char *ac
= (unsigned char *)a
;
433 unsigned char *bc
= (unsigned char *)b
;
435 for (i
=0; i
<16; i
+= 4) {
445 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
452 copy_uuid(uuid
, id
, swap
);
453 for (i
= 0; i
< 4; i
++) {
456 for (j
= 3; j
>= 0; j
--) {
457 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
465 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
467 // dirty hack to work around an issue with super1 superblocks...
468 // super1 superblocks need swapuuid set in order for assembly to
469 // work, but can't have it set if we want this printout to match
470 // all the other uuid printouts in super1.c, so we force swapuuid
471 // to 1 to make our printout match the rest of super1
472 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
476 int check_ext2(int fd
, char *name
)
479 * Check for an ext2fs file system.
480 * Superblock is always 1K at 1K offset
482 * s_magic is le16 at 56 == 0xEF53
483 * report mtime - le32 at 44
485 * logblksize - le32 at 24
487 unsigned char sb
[1024];
490 if (lseek(fd
, 1024,0)!= 1024)
492 if (read(fd
, sb
, 1024)!= 1024)
494 if (sb
[56] != 0x53 || sb
[57] != 0xef)
497 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
498 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
499 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
500 fprintf(stderr
, Name
": %s appears to contain an ext2fs file system\n",
502 fprintf(stderr
," size=%dK mtime=%s",
503 size
*(1<<bsize
), ctime(&mtime
));
507 int check_reiser(int fd
, char *name
)
510 * superblock is at 64K
512 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
515 unsigned char sb
[1024];
517 if (lseek(fd
, 64*1024, 0) != 64*1024)
519 if (read(fd
, sb
, 1024) != 1024)
521 if (strncmp((char*)sb
+52, "ReIsErFs",8)!=0 &&
522 strncmp((char*)sb
+52, "ReIsEr2Fs",9)!=0)
524 fprintf(stderr
, Name
": %s appears to contain a reiserfs file system\n",name
);
525 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
526 fprintf(stderr
, " size = %luK\n", size
*4);
531 int check_raid(int fd
, char *name
)
536 struct supertype
*st
= guess_super(fd
);
539 st
->ignore_hw_compat
= 1;
540 st
->ss
->load_super(st
, fd
, name
);
541 /* Looks like a raid array .. */
542 fprintf(stderr
, Name
": %s appears to be part of a raid array:\n",
544 st
->ss
->getinfo_super(st
, &info
, NULL
);
545 st
->ss
->free_super(st
);
546 crtime
= info
.array
.ctime
;
547 level
= map_num(pers
, info
.array
.level
);
548 if (!level
) level
= "-unknown-";
549 fprintf(stderr
, " level=%s devices=%d ctime=%s",
550 level
, info
.array
.raid_disks
, ctime(&crtime
));
558 for (i
=0; i
<5; i
++) {
560 fprintf(stderr
, "%s%s", mesg
, add
);
562 if (fgets(buf
, 100, stdin
)==NULL
)
564 if (buf
[0]=='y' || buf
[0]=='Y')
566 if (buf
[0]=='n' || buf
[0]=='N')
570 fprintf(stderr
, Name
": assuming 'no'\n");
573 #endif /* MDASSEMBLE */
575 int is_standard(char *dev
, int *nump
)
577 /* tests if dev is a "standard" md dev name.
578 * i.e if the last component is "/dNN" or "/mdNN",
579 * where NN is a string of digits
580 * Returns 1 if a partitionable standard,
581 * -1 if non-partitonable,
582 * 0 if not a standard name.
584 char *d
= strrchr(dev
, '/');
589 if (strncmp(d
, "/d",2)==0)
590 d
+= 2, type
=1; /* /dev/md/dN{pM} */
591 else if (strncmp(d
, "/md_d", 5)==0)
592 d
+= 5, type
=1; /* /dev/md_dN{pM} */
593 else if (strncmp(d
, "/md", 3)==0)
594 d
+= 3, type
=-1; /* /dev/mdN */
595 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3)==0)
596 d
+= 1, type
=-1; /* /dev/md/N */
606 if (nump
) *nump
= num
;
611 unsigned long calc_csum(void *super
, int bytes
)
613 unsigned long long newcsum
= 0;
616 unsigned int *superc
= (unsigned int*) super
;
618 for(i
=0; i
<bytes
/4; i
++)
620 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
622 /* The in-kernel checksum calculation is always 16bit on
623 * the alpha, though it is 32 bit on i386...
624 * I wonder what it is elsewhere... (it uses and API in
625 * a way that it shouldn't).
627 csum
= (csum
& 0xffff) + (csum
>> 16);
628 csum
= (csum
& 0xffff) + (csum
>> 16);
634 char *human_size(long long bytes
)
638 /* We convert bytes to either centi-M{ega,ibi}bytes or
639 * centi-G{igi,ibi}bytes, with appropriate rounding,
640 * and then print 1/100th of those as a decimal.
641 * We allow upto 2048Megabytes before converting to
642 * gigabytes, as that shows more precision and isn't
643 * too large a number.
644 * Terabytes are not yet handled.
647 if (bytes
< 5000*1024)
649 else if (bytes
< 2*1024LL*1024LL*1024LL) {
650 long cMiB
= (bytes
/ ( (1LL<<20) / 200LL ) +1) /2;
651 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
652 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
653 cMiB
/100 , cMiB
% 100,
656 long cGiB
= (bytes
/ ( (1LL<<30) / 200LL ) +1) /2;
657 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
658 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
659 cGiB
/100 , cGiB
% 100,
665 char *human_size_brief(long long bytes
)
669 if (bytes
< 5000*1024)
670 snprintf(buf
, sizeof(buf
), "%ld.%02ldKiB",
671 (long)(bytes
>>10), (long)(((bytes
&1023)*100+512)/1024)
673 else if (bytes
< 2*1024LL*1024LL*1024LL)
674 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
676 (long)((bytes
&0xfffff)+0x100000/200)/(0x100000/100)
679 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
681 (long)(((bytes
>>10)&0xfffff)+0x100000/200)/(0x100000/100)
686 void print_r10_layout(int layout
)
688 int near
= layout
& 255;
689 int far
= (layout
>> 8) & 255;
690 int offset
= (layout
&0x10000);
694 printf("%s near=%d", sep
, near
);
698 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
700 printf("NO REDUNDANCY");
704 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
705 int chunksize
, unsigned long long devsize
)
707 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
708 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
711 int get_data_disks(int level
, int layout
, int raid_disks
)
715 case 0: data_disks
= raid_disks
; break;
716 case 1: data_disks
= 1; break;
718 case 5: data_disks
= raid_disks
- 1; break;
719 case 6: data_disks
= raid_disks
- 2; break;
720 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
727 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
728 char *get_md_name(int dev
)
730 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
731 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
732 static char devname
[50];
738 int mdp
= get_mdp_major();
739 if (mdp
< 0) return NULL
;
740 rdev
= makedev(mdp
, (-1-dev
)<<6);
741 snprintf(devname
, sizeof(devname
), "/dev/md/d%d", -1-dev
);
742 if (stat(devname
, &stb
) == 0
743 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
744 && (stb
.st_rdev
== rdev
))
747 rdev
= makedev(MD_MAJOR
, dev
);
748 snprintf(devname
, sizeof(devname
), "/dev/md%d", dev
);
749 if (stat(devname
, &stb
) == 0
750 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
751 && (stb
.st_rdev
== rdev
))
754 snprintf(devname
, sizeof(devname
), "/dev/md/%d", dev
);
755 if (stat(devname
, &stb
) == 0
756 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
757 && (stb
.st_rdev
== rdev
))
760 dn
= map_dev(major(rdev
), minor(rdev
), 0);
763 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md%d", dev
);
764 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
768 if (stat(devname
, &stb
) == 0
769 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
770 && (stb
.st_rdev
== rdev
))
776 void put_md_name(char *name
)
778 if (strncmp(name
, "/dev/.tmp.md", 12)==0)
782 int find_free_devnum(int use_partitions
)
785 for (devnum
= 127; devnum
!= 128;
786 devnum
= devnum
? devnum
-1 : (1<<20)-1) {
790 _devnum
= use_partitions
? (-1-devnum
) : devnum
;
791 if (mddev_busy(_devnum
))
793 /* make sure it is new to /dev too, at least as a
795 dn
= map_dev(dev2major(_devnum
), dev2minor(_devnum
), 0);
796 if (dn
&& ! is_standard(dn
, NULL
))
802 return use_partitions
? (-1-devnum
) : devnum
;
804 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
806 int dev_open(char *dev
, int flags
)
808 /* like 'open', but if 'dev' matches %d:%d, create a temp
809 * block device and open that
820 major
= strtoul(dev
, &e
, 0);
821 if (e
> dev
&& *e
== ':' && e
[1] &&
822 (minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
824 char *path
= map_dev(major
, minor
, 0);
826 fd
= open(path
, flags
);
828 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
829 (int)getpid(), major
, minor
);
830 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
))==0) {
831 fd
= open(devname
, flags
);
836 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
837 (int)getpid(), major
, minor
);
838 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
))==0) {
839 fd
= open(devname
, flags
);
844 fd
= open(dev
, flags
);
848 int open_dev_flags(int devnum
, int flags
)
852 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
853 return dev_open(buf
, flags
);
856 int open_dev(int devnum
)
858 return open_dev_flags(devnum
, O_RDONLY
);
861 int open_dev_excl(int devnum
)
867 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
868 for (i
=0 ; i
<25 ; i
++) {
869 int fd
= dev_open(buf
, flags
|O_EXCL
);
872 if (errno
== EACCES
&& flags
== O_RDWR
) {
883 int same_dev(char *one
, char *two
)
885 struct stat st1
, st2
;
886 if (stat(one
, &st1
) != 0)
888 if (stat(two
, &st2
) != 0)
890 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
892 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
894 return st1
.st_rdev
== st2
.st_rdev
;
897 void wait_for(char *dev
, int fd
)
900 struct stat stb_want
;
902 if (fstat(fd
, &stb_want
) != 0 ||
903 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
906 for (i
=0 ; i
<25 ; i
++) {
908 if (stat(dev
, &stb
) == 0 &&
909 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
910 (stb
.st_rdev
== stb_want
.st_rdev
))
915 dprintf("%s: timeout waiting for %s\n", __func__
, dev
);
918 struct superswitch
*superlist
[] =
921 &super_ddf
, &super_imsm
,
925 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
927 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
929 mdu_array_info_t array
;
932 struct supertype
*st
= NULL
;
937 char *subarray
= NULL
;
938 int container
= NoMdDev
;
940 sra
= sysfs_read(fd
, 0, GET_VERSION
);
943 vers
= sra
->array
.major_version
;
944 minor
= sra
->array
.minor_version
;
945 verstr
= sra
->text_version
;
947 if (ioctl(fd
, GET_ARRAY_INFO
, &array
))
948 array
.major_version
= array
.minor_version
= 0;
949 vers
= array
.major_version
;
950 minor
= array
.minor_version
;
955 sprintf(version
, "%d.%d", vers
, minor
);
958 if (minor
== -2 && is_subarray(verstr
)) {
959 char *dev
= verstr
+1;
961 subarray
= strchr(dev
, '/');
964 subarray
= strdup(subarray
);
965 container
= devname2devnum(dev
);
968 sra
= sysfs_read(-1, container
, GET_VERSION
);
969 if (sra
&& sra
->text_version
[0])
970 verstr
= sra
->text_version
;
972 verstr
= "-no-metadata-";
975 for (i
= 0; st
== NULL
&& superlist
[i
] ; i
++)
976 st
= superlist
[i
]->match_metadata_desc(verstr
);
983 *subarrayp
= subarray
;
984 st
->container_dev
= container
;
985 st
->devnum
= fd2devnum(fd
);
991 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
993 int dev_size_from_id(dev_t id
, unsigned long long *size
)
998 sprintf(buf
, "%d:%d", major(id
), minor(id
));
999 fd
= dev_open(buf
, O_RDONLY
);
1002 if (get_dev_size(fd
, NULL
, size
)) {
1010 struct supertype
*dup_super(struct supertype
*orig
)
1012 struct supertype
*st
;
1016 st
= malloc(sizeof(*st
));
1019 memset(st
, 0, sizeof(*st
));
1021 st
->max_devs
= orig
->max_devs
;
1022 st
->minor_version
= orig
->minor_version
;
1028 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1030 /* try each load_super to find the best match,
1031 * and return the best superswitch
1033 struct superswitch
*ss
;
1034 struct supertype
*st
;
1035 time_t besttime
= 0;
1039 st
= malloc(sizeof(*st
));
1040 memset(st
, 0, sizeof(*st
));
1041 st
->container_dev
= NoMdDev
;
1043 for (i
=0 ; superlist
[i
]; i
++) {
1046 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1048 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1050 memset(st
, 0, sizeof(*st
));
1051 st
->ignore_hw_compat
= 1;
1052 rv
= ss
->load_super(st
, fd
, NULL
);
1055 st
->ss
->getinfo_super(st
, &info
, NULL
);
1056 if (bestsuper
== -1 ||
1057 besttime
< info
.array
.ctime
) {
1059 besttime
= info
.array
.ctime
;
1064 if (bestsuper
!= -1) {
1066 memset(st
, 0, sizeof(*st
));
1067 st
->ignore_hw_compat
= 1;
1068 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1070 superlist
[bestsuper
]->free_super(st
);
1071 st
->ignore_hw_compat
= 0;
1079 /* Return size of device in bytes */
1080 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1082 unsigned long long ldsize
;
1085 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1086 ldsize
= (unsigned long long)st
.st_size
;
1089 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1092 unsigned long dsize
;
1093 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1098 fprintf(stderr
, Name
": Cannot get size of %s: %s\b",
1099 dname
, strerror(errno
));
1107 /* Return true if this can only be a container, not a member device.
1108 * i.e. is and md device and size is zero
1110 int must_be_container(int fd
)
1112 unsigned long long size
;
1113 if (md_get_version(fd
) < 0)
1115 if (get_dev_size(fd
, NULL
, &size
) == 0)
1122 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1123 * Returns: 1 if successful
1124 * -1 for unknown partition type
1125 * 0 for other errors
1127 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1130 unsigned char empty_gpt_entry
[16]= {0};
1131 struct GPT_part_entry
*part
;
1133 unsigned long long curr_part_end
;
1134 unsigned all_partitions
, entry_size
;
1139 BUILD_BUG_ON(sizeof(gpt
) != 512);
1140 /* skip protective MBR */
1141 lseek(fd
, 512, SEEK_SET
);
1142 /* read GPT header */
1143 if (read(fd
, &gpt
, 512) != 512)
1146 /* get the number of partition entries and the entry size */
1147 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1148 entry_size
= __le32_to_cpu(gpt
.part_size
);
1150 /* Check GPT signature*/
1151 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1155 if (all_partitions
> 1024 ||
1156 entry_size
> sizeof(buf
))
1159 part
= (struct GPT_part_entry
*)buf
;
1161 for (part_nr
=0; part_nr
< all_partitions
; part_nr
++) {
1162 /* read partition entry */
1163 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1166 /* is this valid partition? */
1167 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1168 /* check the last lba for the current partition */
1169 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1170 if (curr_part_end
> *endofpart
)
1171 *endofpart
= curr_part_end
;
1178 /* Sets endofpart parameter to the last block used by the last partition on the device.
1179 * Returns: 1 if successful
1180 * -1 for unknown partition type
1181 * 0 for other errors
1183 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1185 struct MBR boot_sect
;
1186 struct MBR_part_record
*part
;
1187 unsigned long long curr_part_end
;
1193 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1196 if (read(fd
, &boot_sect
, 512) != 512)
1199 /* check MBP signature */
1200 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1202 /* found the correct signature */
1203 part
= boot_sect
.parts
;
1205 for (part_nr
=0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1206 /* check for GPT type */
1207 if (part
->part_type
== MBR_GPT_PARTITION_TYPE
) {
1208 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1211 /* check the last used lba for the current partition */
1212 curr_part_end
= __le32_to_cpu(part
->first_sect_lba
) +
1213 __le32_to_cpu(part
->blocks_num
);
1214 if (curr_part_end
> *endofpart
)
1215 *endofpart
= curr_part_end
;
1220 /* Unknown partition table */
1227 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1228 unsigned long long size
)
1231 * Check where the last partition ends
1233 unsigned long long endofpart
;
1236 if ((ret
= get_last_partition_end(fd
, &endofpart
)) > 0) {
1237 /* There appears to be a partition table here */
1238 if (freesize
== 0) {
1239 /* partitions will not be visible in new device */
1241 Name
": partition table exists on %s but will be lost or\n"
1242 " meaningless after creating array\n",
1245 } else if (endofpart
> freesize
) {
1246 /* last partition overlaps metadata */
1248 Name
": metadata will over-write last partition on %s.\n",
1251 } else if (size
&& endofpart
> size
) {
1252 /* partitions will be truncated in new device */
1254 Name
": array size is too small to cover all partitions on %s.\n",
1262 void get_one_disk(int mdfd
, mdu_array_info_t
*ainf
, mdu_disk_info_t
*disk
)
1266 ioctl(mdfd
, GET_ARRAY_INFO
, ainf
);
1267 for (d
= 0 ; d
< 1024 ; d
++) {
1268 if (ioctl(mdfd
, GET_DISK_INFO
, disk
) == 0 &&
1269 (disk
->major
|| disk
->minor
))
1274 int open_container(int fd
)
1276 /* 'fd' is a block device. Find out if it is in use
1277 * by a container, and return an open fd on that container.
1288 if (fstat(fd
, &st
) != 0)
1290 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1291 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1292 e
= path
+ strlen(path
);
1294 dir
= opendir(path
);
1297 while ((de
= readdir(dir
))) {
1300 if (de
->d_name
[0] == '.')
1302 sprintf(e
, "/%s/dev", de
->d_name
);
1303 dfd
= open(path
, O_RDONLY
);
1306 n
= read(dfd
, buf
, sizeof(buf
));
1308 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1311 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1313 sprintf(buf
, "%d:%d", major
, minor
);
1314 dfd
= dev_open(buf
, O_RDONLY
);
1324 struct superswitch
*version_to_superswitch(char *vers
)
1328 for (i
= 0; superlist
[i
]; i
++) {
1329 struct superswitch
*ss
= superlist
[i
];
1331 if (strcmp(vers
, ss
->name
) == 0)
1338 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1340 if (mdstat
->metadata_version
== NULL
||
1341 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1342 !is_subarray(mdstat
->metadata_version
+9) ||
1343 strncmp(mdstat
->metadata_version
+10, container
, strlen(container
)) != 0 ||
1344 mdstat
->metadata_version
[10+strlen(container
)] != '/')
1350 int is_subarray_active(char *subarray
, char *container
)
1352 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1353 struct mdstat_ent
*ent
;
1355 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1356 if (is_container_member(ent
, container
))
1357 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1360 free_mdstat(mdstat
);
1365 /* open_subarray - opens a subarray in a container
1366 * @dev: container device name
1367 * @st: empty supertype
1368 * @quiet: block reporting errors flag
1370 * On success returns an fd to a container and fills in *st
1372 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1375 struct mdinfo
*info
;
1378 fd
= open(dev
, O_RDWR
|O_EXCL
);
1381 fprintf(stderr
, Name
": Couldn't open %s, aborting\n",
1386 st
->devnum
= fd2devnum(fd
);
1387 if (st
->devnum
== NoMdDev
) {
1390 Name
": Failed to determine device number for %s\n",
1395 mdi
= sysfs_read(fd
, st
->devnum
, GET_VERSION
|GET_LEVEL
);
1398 fprintf(stderr
, Name
": Failed to read sysfs for %s\n",
1403 if (mdi
->array
.level
!= UnSet
) {
1405 fprintf(stderr
, Name
": %s is not a container\n", dev
);
1409 st
->ss
= version_to_superswitch(mdi
->text_version
);
1413 Name
": Operation not supported for %s metadata\n",
1418 st
->devname
= devnum2devname(st
->devnum
);
1421 fprintf(stderr
, Name
": Failed to allocate device name\n");
1425 if (!st
->ss
->load_container
) {
1427 fprintf(stderr
, Name
": %s is not a container\n", dev
);
1431 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1433 fprintf(stderr
, Name
": Failed to load metadata for %s\n",
1438 info
= st
->ss
->container_content(st
, subarray
);
1441 fprintf(stderr
, Name
": Failed to find subarray-%s in %s\n",
1451 st
->ss
->free_super(st
);
1467 int add_disk(int mdfd
, struct supertype
*st
,
1468 struct mdinfo
*sra
, struct mdinfo
*info
)
1470 /* Add a device to an array, in one of 2 ways. */
1473 if (st
->ss
->external
) {
1474 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1475 info
->recovery_start
= MaxSector
;
1477 info
->recovery_start
= 0;
1478 rv
= sysfs_add_disk(sra
, info
, 0);
1481 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1485 sd2
= malloc(sizeof(*sd2
));
1487 sd2
->next
= sra
->devs
;
1493 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1497 int remove_disk(int mdfd
, struct supertype
*st
,
1498 struct mdinfo
*sra
, struct mdinfo
*info
)
1501 /* Remove the disk given by 'info' from the array */
1503 if (st
->ss
->external
)
1504 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1507 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1512 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1514 /* Initialise kernel's knowledge of array.
1515 * This varies between externally managed arrays
1518 int vers
= md_get_version(mdfd
);
1522 if (st
->ss
->external
)
1523 rv
= sysfs_set_array(info
, vers
);
1526 if ((vers
% 100) >= 1) { /* can use different versions */
1527 mdu_array_info_t inf
;
1528 memset(&inf
, 0, sizeof(inf
));
1529 inf
.major_version
= info
->array
.major_version
;
1530 inf
.minor_version
= info
->array
.minor_version
;
1531 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1533 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1537 unsigned long long min_recovery_start(struct mdinfo
*array
)
1539 /* find the minimum recovery_start in an array for metadata
1540 * formats that only record per-array recovery progress instead
1543 unsigned long long recovery_start
= MaxSector
;
1546 for (d
= array
->devs
; d
; d
= d
->next
)
1547 recovery_start
= min(recovery_start
, d
->recovery_start
);
1549 return recovery_start
;
1552 int mdmon_pid(int devnum
)
1558 char *devname
= devnum2devname(devnum
);
1560 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devname
);
1563 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1567 n
= read(fd
, pid
, 9);
1574 int mdmon_running(int devnum
)
1576 int pid
= mdmon_pid(devnum
);
1579 if (kill(pid
, 0) == 0)
1584 int start_mdmon(int devnum
)
1598 if (check_env("MDADM_NO_MDMON"))
1601 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1605 sl
= strrchr(pathbuf
, '/');
1610 strcpy(sl
, "mdmon");
1616 /* FIXME yuk. CLOSE_EXEC?? */
1618 for (i
=3; skipped
< 20; i
++)
1624 for (i
=0; paths
[i
]; i
++)
1626 execl(paths
[i
], "mdmon",
1627 devnum2devname(devnum
),
1630 case -1: fprintf(stderr
, Name
": cannot run mdmon. "
1631 "Array remains readonly\n");
1633 default: /* parent - good */
1634 pid
= wait(&status
);
1635 if (pid
< 0 || status
!= 0)
1641 int check_env(char *name
)
1643 char *val
= getenv(name
);
1645 if (val
&& atoi(val
) == 1)
1651 __u32
random32(void)
1654 int rfd
= open("/dev/urandom", O_RDONLY
);
1655 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1663 int flush_metadata_updates(struct supertype
*st
)
1667 st
->update_tail
= NULL
;
1671 sfd
= connect_monitor(devnum2devname(st
->container_dev
));
1675 while (st
->updates
) {
1676 struct metadata_update
*mu
= st
->updates
;
1677 st
->updates
= mu
->next
;
1679 send_message(sfd
, mu
, 0);
1687 st
->update_tail
= NULL
;
1691 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
1694 struct metadata_update
*mu
= malloc(sizeof(*mu
));
1699 mu
->space_list
= NULL
;
1701 *st
->update_tail
= mu
;
1702 st
->update_tail
= &mu
->next
;
1704 #endif /* MDASSEMBLE */
1707 /* tinyc doesn't optimize this check in ioctl.h out ... */
1708 unsigned int __invalid_size_argument_for_IOC
= 0;
1711 int experimental(void)
1713 if (check_env("MDADM_EXPERIMENTAL"))
1716 fprintf(stderr
, Name
": To use this feature MDADM_EXPERIMENTAL"
1717 " environment variable has to be defined.\n");
1722 /* Pick all spares matching given criteria from a container
1723 * if min_size == 0 do not check size
1724 * if domlist == NULL do not check domains
1725 * if spare_group given add it to domains of each spare
1726 * metadata allows to test domains using metadata of destination array */
1727 struct mdinfo
*container_choose_spares(struct supertype
*st
,
1728 unsigned long long min_size
,
1729 struct domainlist
*domlist
,
1731 const char *metadata
, int get_one
)
1733 struct mdinfo
*d
, **dp
, *disks
= NULL
;
1735 /* get list of all disks in container */
1736 if (st
->ss
->getinfo_super_disks
)
1737 disks
= st
->ss
->getinfo_super_disks(st
);
1741 /* find spare devices on the list */
1743 disks
->array
.spare_disks
= 0;
1747 if (d
->disk
.state
== 0) {
1748 /* check if size is acceptable */
1749 unsigned long long dev_size
;
1750 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
1753 (dev_size_from_id(dev
, &dev_size
) &&
1754 dev_size
>= min_size
))
1756 /* check if domain matches */
1757 if (found
&& domlist
) {
1758 struct dev_policy
*pol
= devnum_policy(dev
);
1760 pol_add(&pol
, pol_domain
,
1762 if (domain_test(domlist
, pol
, metadata
) != 1)
1764 dev_policy_free(pol
);
1769 disks
->array
.spare_disks
++;