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>
38 * following taken from linux/blkpg.h because they aren't
39 * anywhere else and it isn't safe to #include linux/ * stuff.
42 #define BLKPG _IO(0x12,105)
44 /* The argument structure */
45 struct blkpg_ioctl_arg
{
52 /* The subfunctions (for the op field) */
53 #define BLKPG_ADD_PARTITION 1
54 #define BLKPG_DEL_PARTITION 2
56 /* Sizes of name fields. Unused at present. */
57 #define BLKPG_DEVNAMELTH 64
58 #define BLKPG_VOLNAMELTH 64
60 /* The data structure for ADD_PARTITION and DEL_PARTITION */
61 struct blkpg_partition
{
62 long long start
; /* starting offset in bytes */
63 long long length
; /* length in bytes */
64 int pno
; /* partition number */
65 char devname
[BLKPG_DEVNAMELTH
]; /* partition name, like sda5 or c0d1p2,
66 to be used in kernel messages */
67 char volname
[BLKPG_VOLNAMELTH
]; /* volume label */
72 /* Force a compilation error if condition is true */
73 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
75 /* Force a compilation error if condition is true, but also produce a
76 result (of value 0 and type size_t), so the expression can be used
77 e.g. in a structure initializer (or where-ever else comma expressions
79 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
82 * Parse a 128 bit uuid in 4 integers
83 * format is 32 hexx nibbles with options :.<space> separator
84 * If not exactly 32 hex digits are found, return 0
87 int parse_uuid(char *str
, int uuid
[4])
89 int hit
= 0; /* number of Hex digIT */
92 for (i
=0; i
<4; i
++) uuid
[i
]=0;
98 else if (c
>='a' && c
<= 'f')
100 else if (c
>='A' && c
<= 'F')
102 else if (strchr(":. -", c
))
119 * Get the md version number.
120 * We use the RAID_VERSION ioctl if it is supported
121 * If not, but we have a block device with major '9', we assume
124 * Return version number as 24 but number - assume version parts
128 int md_get_version(int fd
)
133 if (fstat(fd
, &stb
)<0)
135 if ((S_IFMT
&stb
.st_mode
) != S_IFBLK
)
138 if (ioctl(fd
, RAID_VERSION
, &vers
) == 0)
139 return (vers
.major
*10000) + (vers
.minor
*100) + vers
.patchlevel
;
142 if (major(stb
.st_rdev
) == MD_MAJOR
)
147 int get_linux_version()
151 int a
= 0, b
= 0,c
= 0;
156 a
= strtoul(cp
, &cp
, 10);
158 b
= strtoul(cp
+1, &cp
, 10);
160 c
= strtoul(cp
+1, &cp
, 10);
162 return (a
*1000000)+(b
*1000)+c
;
166 int mdadm_version(char *version
)
174 cp
= strchr(version
, '-');
175 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
178 a
= strtoul(cp
, &cp
, 10);
181 b
= strtoul(cp
+1, &cp
, 10);
183 c
= strtoul(cp
+1, &cp
, 10);
186 if (*cp
!= ' ' && *cp
!= '-')
188 return (a
*1000000)+(b
*1000)+c
;
191 long long parse_size(char *size
)
193 /* parse 'size' which should be a number optionally
194 * followed by 'K', 'M', or 'G'.
195 * Without a suffix, K is assumed.
196 * Number returned is in sectors (half-K)
199 long long s
= strtoll(size
, &c
, 10);
213 s
*= 1024 * 1024 * 2;
222 int parse_layout_10(char *layout
)
226 /* Parse the layout string for raid10 */
227 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
228 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
229 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
233 if (layout
[0] == 'n')
235 else if (layout
[0] == 'o')
236 rv
= 0x10000 + (copies
<<8) + 1;
238 rv
= 1 + (copies
<<8);
242 int parse_layout_faulty(char *layout
)
244 /* Parse the layout string for 'faulty' */
245 int ln
= strcspn(layout
, "0123456789");
246 char *m
= strdup(layout
);
249 mode
= map_name(faultylayout
, m
);
253 return mode
| (atoi(layout
+ln
)<< ModeShift
);
257 void remove_partitions(int fd
)
259 /* remove partitions from this block devices.
260 * This is used for components added to an array
262 #ifdef BLKPG_DEL_PARTITION
263 struct blkpg_ioctl_arg a
;
264 struct blkpg_partition p
;
266 a
.op
= BLKPG_DEL_PARTITION
;
268 a
.datalen
= sizeof(p
);
270 memset(a
.data
, 0, a
.datalen
);
271 for (p
.pno
=0; p
.pno
< 16; p
.pno
++)
272 ioctl(fd
, BLKPG
, &a
);
276 int test_partition(int fd
)
278 /* Check if fd is a whole-disk or a partition.
279 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
280 * will return ENXIO on an invalid partition number.
282 struct blkpg_ioctl_arg a
;
283 struct blkpg_partition p
;
284 a
.op
= BLKPG_DEL_PARTITION
;
286 a
.datalen
= sizeof(p
);
288 memset(a
.data
, 0, a
.datalen
);
290 if (ioctl(fd
, BLKPG
, &a
) == 0)
291 /* Very unlikely, but not a partition */
294 /* not a partition */
300 int test_partition_from_id(dev_t id
)
305 sprintf(buf
, "%d:%d", major(id
), minor(id
));
306 fd
= dev_open(buf
, O_RDONLY
);
309 rv
= test_partition(fd
);
314 int enough(int level
, int raid_disks
, int layout
, int clean
,
315 char *avail
, int avail_disks
)
320 /* This is the tricky one - we need to check
321 * which actual disks are present.
323 copies
= (layout
&255)* ((layout
>>8) & 255);
326 /* there must be one of the 'copies' form 'first' */
332 first
= (first
+1) % raid_disks
;
337 } while (first
!= 0);
340 case LEVEL_MULTIPATH
:
341 return avail_disks
>= 1;
344 return avail_disks
== raid_disks
;
346 return avail_disks
>= 1;
350 return avail_disks
>= raid_disks
-1;
352 return avail_disks
>= raid_disks
;
355 return avail_disks
>= raid_disks
-2;
357 return avail_disks
>= raid_disks
;
363 int enough_fd(int fd
)
365 struct mdu_array_info_s array
;
366 struct mdu_disk_info_s disk
;
371 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0 ||
372 array
.raid_disks
<= 0)
374 avail
= calloc(array
.raid_disks
, 1);
375 for (i
=0; i
< 1024 && array
.nr_disks
> 0; i
++) {
377 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
379 if (disk
.major
== 0 && disk
.minor
== 0)
383 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
385 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
388 avail
[disk
.raid_disk
] = 1;
390 /* This is used on an active array, so assume it is clean */
391 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
392 1, avail
, avail_disks
);
398 const int uuid_zero
[4] = { 0, 0, 0, 0 };
400 int same_uuid(int a
[4], int b
[4], int swapuuid
)
403 /* parse uuids are hostendian.
404 * uuid's from some superblocks are big-ending
405 * if there is a difference, we need to swap..
407 unsigned char *ac
= (unsigned char *)a
;
408 unsigned char *bc
= (unsigned char *)b
;
410 for (i
=0; i
<16; i
+= 4) {
411 if (ac
[i
+0] != bc
[i
+3] ||
412 ac
[i
+1] != bc
[i
+2] ||
413 ac
[i
+2] != bc
[i
+1] ||
427 void copy_uuid(void *a
, int b
[4], int swapuuid
)
430 /* parse uuids are hostendian.
431 * uuid's from some superblocks are big-ending
432 * if there is a difference, we need to swap..
434 unsigned char *ac
= (unsigned char *)a
;
435 unsigned char *bc
= (unsigned char *)b
;
437 for (i
=0; i
<16; i
+= 4) {
447 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
454 copy_uuid(uuid
, id
, swap
);
455 for (i
= 0; i
< 4; i
++) {
458 for (j
= 3; j
>= 0; j
--) {
459 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
467 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
469 // dirty hack to work around an issue with super1 superblocks...
470 // super1 superblocks need swapuuid set in order for assembly to
471 // work, but can't have it set if we want this printout to match
472 // all the other uuid printouts in super1.c, so we force swapuuid
473 // to 1 to make our printout match the rest of super1
474 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
478 int check_ext2(int fd
, char *name
)
481 * Check for an ext2fs file system.
482 * Superblock is always 1K at 1K offset
484 * s_magic is le16 at 56 == 0xEF53
485 * report mtime - le32 at 44
487 * logblksize - le32 at 24
489 unsigned char sb
[1024];
492 if (lseek(fd
, 1024,0)!= 1024)
494 if (read(fd
, sb
, 1024)!= 1024)
496 if (sb
[56] != 0x53 || sb
[57] != 0xef)
499 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
500 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
501 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
502 fprintf(stderr
, Name
": %s appears to contain an ext2fs file system\n",
504 fprintf(stderr
," size=%dK mtime=%s",
505 size
*(1<<bsize
), ctime(&mtime
));
509 int check_reiser(int fd
, char *name
)
512 * superblock is at 64K
514 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
517 unsigned char sb
[1024];
519 if (lseek(fd
, 64*1024, 0) != 64*1024)
521 if (read(fd
, sb
, 1024) != 1024)
523 if (strncmp((char*)sb
+52, "ReIsErFs",8)!=0 &&
524 strncmp((char*)sb
+52, "ReIsEr2Fs",9)!=0)
526 fprintf(stderr
, Name
": %s appears to contain a reiserfs file system\n",name
);
527 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
528 fprintf(stderr
, " size = %luK\n", size
*4);
533 int check_raid(int fd
, char *name
)
538 struct supertype
*st
= guess_super(fd
);
541 st
->ignore_hw_compat
= 1;
542 st
->ss
->load_super(st
, fd
, name
);
543 /* Looks like a raid array .. */
544 fprintf(stderr
, Name
": %s appears to be part of a raid array:\n",
546 st
->ss
->getinfo_super(st
, &info
, NULL
);
547 st
->ss
->free_super(st
);
548 crtime
= info
.array
.ctime
;
549 level
= map_num(pers
, info
.array
.level
);
550 if (!level
) level
= "-unknown-";
551 fprintf(stderr
, " level=%s devices=%d ctime=%s",
552 level
, info
.array
.raid_disks
, ctime(&crtime
));
560 for (i
=0; i
<5; i
++) {
562 fprintf(stderr
, "%s%s", mesg
, add
);
564 if (fgets(buf
, 100, stdin
)==NULL
)
566 if (buf
[0]=='y' || buf
[0]=='Y')
568 if (buf
[0]=='n' || buf
[0]=='N')
572 fprintf(stderr
, Name
": assuming 'no'\n");
575 #endif /* MDASSEMBLE */
577 int is_standard(char *dev
, int *nump
)
579 /* tests if dev is a "standard" md dev name.
580 * i.e if the last component is "/dNN" or "/mdNN",
581 * where NN is a string of digits
582 * Returns 1 if a partitionable standard,
583 * -1 if non-partitonable,
584 * 0 if not a standard name.
586 char *d
= strrchr(dev
, '/');
591 if (strncmp(d
, "/d",2)==0)
592 d
+= 2, type
=1; /* /dev/md/dN{pM} */
593 else if (strncmp(d
, "/md_d", 5)==0)
594 d
+= 5, type
=1; /* /dev/md_dN{pM} */
595 else if (strncmp(d
, "/md", 3)==0)
596 d
+= 3, type
=-1; /* /dev/mdN */
597 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3)==0)
598 d
+= 1, type
=-1; /* /dev/md/N */
608 if (nump
) *nump
= num
;
613 unsigned long calc_csum(void *super
, int bytes
)
615 unsigned long long newcsum
= 0;
618 unsigned int *superc
= (unsigned int*) super
;
620 for(i
=0; i
<bytes
/4; i
++)
622 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
624 /* The in-kernel checksum calculation is always 16bit on
625 * the alpha, though it is 32 bit on i386...
626 * I wonder what it is elsewhere... (it uses and API in
627 * a way that it shouldn't).
629 csum
= (csum
& 0xffff) + (csum
>> 16);
630 csum
= (csum
& 0xffff) + (csum
>> 16);
636 char *human_size(long long bytes
)
640 /* We convert bytes to either centi-M{ega,ibi}bytes or
641 * centi-G{igi,ibi}bytes, with appropriate rounding,
642 * and then print 1/100th of those as a decimal.
643 * We allow upto 2048Megabytes before converting to
644 * gigabytes, as that shows more precision and isn't
645 * too large a number.
646 * Terabytes are not yet handled.
649 if (bytes
< 5000*1024)
651 else if (bytes
< 2*1024LL*1024LL*1024LL) {
652 long cMiB
= (bytes
/ ( (1LL<<20) / 200LL ) +1) /2;
653 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
654 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
655 cMiB
/100 , cMiB
% 100,
658 long cGiB
= (bytes
/ ( (1LL<<30) / 200LL ) +1) /2;
659 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
660 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
661 cGiB
/100 , cGiB
% 100,
667 char *human_size_brief(long long bytes
)
671 if (bytes
< 5000*1024)
672 snprintf(buf
, sizeof(buf
), "%ld.%02ldKiB",
673 (long)(bytes
>>10), (long)(((bytes
&1023)*100+512)/1024)
675 else if (bytes
< 2*1024LL*1024LL*1024LL)
676 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
678 (long)((bytes
&0xfffff)+0x100000/200)/(0x100000/100)
681 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
683 (long)(((bytes
>>10)&0xfffff)+0x100000/200)/(0x100000/100)
688 void print_r10_layout(int layout
)
690 int near
= layout
& 255;
691 int far
= (layout
>> 8) & 255;
692 int offset
= (layout
&0x10000);
696 printf("%s near=%d", sep
, near
);
700 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
702 printf("NO REDUNDANCY");
706 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
707 int chunksize
, unsigned long long devsize
)
709 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
710 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
713 int get_data_disks(int level
, int layout
, int raid_disks
)
717 case 0: data_disks
= raid_disks
; break;
718 case 1: data_disks
= 1; break;
720 case 5: data_disks
= raid_disks
- 1; break;
721 case 6: data_disks
= raid_disks
- 2; break;
722 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
729 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
730 char *get_md_name(int dev
)
732 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
733 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
734 static char devname
[50];
740 int mdp
= get_mdp_major();
741 if (mdp
< 0) return NULL
;
742 rdev
= makedev(mdp
, (-1-dev
)<<6);
743 snprintf(devname
, sizeof(devname
), "/dev/md/d%d", -1-dev
);
744 if (stat(devname
, &stb
) == 0
745 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
746 && (stb
.st_rdev
== rdev
))
749 rdev
= makedev(MD_MAJOR
, dev
);
750 snprintf(devname
, sizeof(devname
), "/dev/md%d", dev
);
751 if (stat(devname
, &stb
) == 0
752 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
753 && (stb
.st_rdev
== rdev
))
756 snprintf(devname
, sizeof(devname
), "/dev/md/%d", dev
);
757 if (stat(devname
, &stb
) == 0
758 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
759 && (stb
.st_rdev
== rdev
))
762 dn
= map_dev(major(rdev
), minor(rdev
), 0);
765 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md%d", dev
);
766 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
770 if (stat(devname
, &stb
) == 0
771 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
772 && (stb
.st_rdev
== rdev
))
778 void put_md_name(char *name
)
780 if (strncmp(name
, "/dev/.tmp.md", 12)==0)
784 int find_free_devnum(int use_partitions
)
787 for (devnum
= 127; devnum
!= 128;
788 devnum
= devnum
? devnum
-1 : (1<<20)-1) {
792 _devnum
= use_partitions
? (-1-devnum
) : devnum
;
793 if (mddev_busy(_devnum
))
795 /* make sure it is new to /dev too, at least as a
797 dn
= map_dev(dev2major(_devnum
), dev2minor(_devnum
), 0);
798 if (dn
&& ! is_standard(dn
, NULL
))
804 return use_partitions
? (-1-devnum
) : devnum
;
806 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
808 int dev_open(char *dev
, int flags
)
810 /* like 'open', but if 'dev' matches %d:%d, create a temp
811 * block device and open that
822 major
= strtoul(dev
, &e
, 0);
823 if (e
> dev
&& *e
== ':' && e
[1] &&
824 (minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
826 char *path
= map_dev(major
, minor
, 0);
828 fd
= open(path
, flags
);
830 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
831 (int)getpid(), major
, minor
);
832 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
))==0) {
833 fd
= open(devname
, flags
);
838 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
839 (int)getpid(), major
, minor
);
840 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
))==0) {
841 fd
= open(devname
, flags
);
846 fd
= open(dev
, flags
);
850 int open_dev_flags(int devnum
, int flags
)
854 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
855 return dev_open(buf
, flags
);
858 int open_dev(int devnum
)
860 return open_dev_flags(devnum
, O_RDONLY
);
863 int open_dev_excl(int devnum
)
869 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
870 for (i
=0 ; i
<25 ; i
++) {
871 int fd
= dev_open(buf
, flags
|O_EXCL
);
874 if (errno
== EACCES
&& flags
== O_RDWR
) {
885 int same_dev(char *one
, char *two
)
887 struct stat st1
, st2
;
888 if (stat(one
, &st1
) != 0)
890 if (stat(two
, &st2
) != 0)
892 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
894 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
896 return st1
.st_rdev
== st2
.st_rdev
;
899 void wait_for(char *dev
, int fd
)
902 struct stat stb_want
;
904 if (fstat(fd
, &stb_want
) != 0 ||
905 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
908 for (i
=0 ; i
<25 ; i
++) {
910 if (stat(dev
, &stb
) == 0 &&
911 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
912 (stb
.st_rdev
== stb_want
.st_rdev
))
917 dprintf("%s: timeout waiting for %s\n", __func__
, dev
);
920 struct superswitch
*superlist
[] =
923 &super_ddf
, &super_imsm
,
927 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
929 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
931 mdu_array_info_t array
;
934 struct supertype
*st
= NULL
;
939 char *subarray
= NULL
;
940 int container
= NoMdDev
;
942 sra
= sysfs_read(fd
, 0, GET_VERSION
);
945 vers
= sra
->array
.major_version
;
946 minor
= sra
->array
.minor_version
;
947 verstr
= sra
->text_version
;
949 if (ioctl(fd
, GET_ARRAY_INFO
, &array
))
950 array
.major_version
= array
.minor_version
= 0;
951 vers
= array
.major_version
;
952 minor
= array
.minor_version
;
957 sprintf(version
, "%d.%d", vers
, minor
);
960 if (minor
== -2 && is_subarray(verstr
)) {
961 char *dev
= verstr
+1;
963 subarray
= strchr(dev
, '/');
966 subarray
= strdup(subarray
);
967 container
= devname2devnum(dev
);
970 sra
= sysfs_read(-1, container
, GET_VERSION
);
971 if (sra
&& sra
->text_version
[0])
972 verstr
= sra
->text_version
;
974 verstr
= "-no-metadata-";
977 for (i
= 0; st
== NULL
&& superlist
[i
] ; i
++)
978 st
= superlist
[i
]->match_metadata_desc(verstr
);
985 *subarrayp
= subarray
;
986 st
->container_dev
= container
;
987 st
->devnum
= fd2devnum(fd
);
993 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
995 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1000 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1001 fd
= dev_open(buf
, O_RDONLY
);
1004 if (get_dev_size(fd
, NULL
, size
)) {
1012 struct supertype
*dup_super(struct supertype
*orig
)
1014 struct supertype
*st
;
1018 st
= malloc(sizeof(*st
));
1021 memset(st
, 0, sizeof(*st
));
1023 st
->max_devs
= orig
->max_devs
;
1024 st
->minor_version
= orig
->minor_version
;
1030 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1032 /* try each load_super to find the best match,
1033 * and return the best superswitch
1035 struct superswitch
*ss
;
1036 struct supertype
*st
;
1037 time_t besttime
= 0;
1041 st
= malloc(sizeof(*st
));
1042 memset(st
, 0, sizeof(*st
));
1043 st
->container_dev
= NoMdDev
;
1045 for (i
=0 ; superlist
[i
]; i
++) {
1048 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1050 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1052 memset(st
, 0, sizeof(*st
));
1053 st
->ignore_hw_compat
= 1;
1054 rv
= ss
->load_super(st
, fd
, NULL
);
1057 st
->ss
->getinfo_super(st
, &info
, NULL
);
1058 if (bestsuper
== -1 ||
1059 besttime
< info
.array
.ctime
) {
1061 besttime
= info
.array
.ctime
;
1066 if (bestsuper
!= -1) {
1068 memset(st
, 0, sizeof(*st
));
1069 st
->ignore_hw_compat
= 1;
1070 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1072 superlist
[bestsuper
]->free_super(st
);
1073 st
->ignore_hw_compat
= 0;
1081 /* Return size of device in bytes */
1082 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1084 unsigned long long ldsize
;
1087 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1088 ldsize
= (unsigned long long)st
.st_size
;
1091 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1094 unsigned long dsize
;
1095 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1100 fprintf(stderr
, Name
": Cannot get size of %s: %s\b",
1101 dname
, strerror(errno
));
1109 /* Return true if this can only be a container, not a member device.
1110 * i.e. is and md device and size is zero
1112 int must_be_container(int fd
)
1114 unsigned long long size
;
1115 if (md_get_version(fd
) < 0)
1117 if (get_dev_size(fd
, NULL
, &size
) == 0)
1124 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1125 * Returns: 1 if successful
1126 * -1 for unknown partition type
1127 * 0 for other errors
1129 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1132 unsigned char empty_gpt_entry
[16]= {0};
1133 struct GPT_part_entry
*part
;
1135 unsigned long long curr_part_end
;
1136 unsigned all_partitions
, entry_size
;
1141 BUILD_BUG_ON(sizeof(gpt
) != 512);
1142 /* skip protective MBR */
1143 lseek(fd
, 512, SEEK_SET
);
1144 /* read GPT header */
1145 if (read(fd
, &gpt
, 512) != 512)
1148 /* get the number of partition entries and the entry size */
1149 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1150 entry_size
= __le32_to_cpu(gpt
.part_size
);
1152 /* Check GPT signature*/
1153 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1157 if (all_partitions
> 1024 ||
1158 entry_size
> sizeof(buf
))
1161 part
= (struct GPT_part_entry
*)buf
;
1163 for (part_nr
=0; part_nr
< all_partitions
; part_nr
++) {
1164 /* read partition entry */
1165 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1168 /* is this valid partition? */
1169 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1170 /* check the last lba for the current partition */
1171 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1172 if (curr_part_end
> *endofpart
)
1173 *endofpart
= curr_part_end
;
1180 /* Sets endofpart parameter to the last block used by the last partition on the device.
1181 * Returns: 1 if successful
1182 * -1 for unknown partition type
1183 * 0 for other errors
1185 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1187 struct MBR boot_sect
;
1188 struct MBR_part_record
*part
;
1189 unsigned long long curr_part_end
;
1195 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1198 if (read(fd
, &boot_sect
, 512) != 512)
1201 /* check MBP signature */
1202 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1204 /* found the correct signature */
1205 part
= boot_sect
.parts
;
1207 for (part_nr
=0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1208 /* check for GPT type */
1209 if (part
->part_type
== MBR_GPT_PARTITION_TYPE
) {
1210 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1213 /* check the last used lba for the current partition */
1214 curr_part_end
= __le32_to_cpu(part
->first_sect_lba
) +
1215 __le32_to_cpu(part
->blocks_num
);
1216 if (curr_part_end
> *endofpart
)
1217 *endofpart
= curr_part_end
;
1222 /* Unknown partition table */
1229 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1230 unsigned long long size
)
1233 * Check where the last partition ends
1235 unsigned long long endofpart
;
1238 if ((ret
= get_last_partition_end(fd
, &endofpart
)) > 0) {
1239 /* There appears to be a partition table here */
1240 if (freesize
== 0) {
1241 /* partitions will not be visible in new device */
1243 Name
": partition table exists on %s but will be lost or\n"
1244 " meaningless after creating array\n",
1247 } else if (endofpart
> freesize
) {
1248 /* last partition overlaps metadata */
1250 Name
": metadata will over-write last partition on %s.\n",
1253 } else if (size
&& endofpart
> size
) {
1254 /* partitions will be truncated in new device */
1256 Name
": array size is too small to cover all partitions on %s.\n",
1264 void get_one_disk(int mdfd
, mdu_array_info_t
*ainf
, mdu_disk_info_t
*disk
)
1268 ioctl(mdfd
, GET_ARRAY_INFO
, ainf
);
1269 for (d
= 0 ; d
< 1024 ; d
++) {
1270 if (ioctl(mdfd
, GET_DISK_INFO
, disk
) == 0 &&
1271 (disk
->major
|| disk
->minor
))
1276 int open_container(int fd
)
1278 /* 'fd' is a block device. Find out if it is in use
1279 * by a container, and return an open fd on that container.
1290 if (fstat(fd
, &st
) != 0)
1292 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1293 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1294 e
= path
+ strlen(path
);
1296 dir
= opendir(path
);
1299 while ((de
= readdir(dir
))) {
1302 if (de
->d_name
[0] == '.')
1304 sprintf(e
, "/%s/dev", de
->d_name
);
1305 dfd
= open(path
, O_RDONLY
);
1308 n
= read(dfd
, buf
, sizeof(buf
));
1310 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1313 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1315 sprintf(buf
, "%d:%d", major
, minor
);
1316 dfd
= dev_open(buf
, O_RDONLY
);
1326 struct superswitch
*version_to_superswitch(char *vers
)
1330 for (i
= 0; superlist
[i
]; i
++) {
1331 struct superswitch
*ss
= superlist
[i
];
1333 if (strcmp(vers
, ss
->name
) == 0)
1340 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1342 if (mdstat
->metadata_version
== NULL
||
1343 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1344 !is_subarray(mdstat
->metadata_version
+9) ||
1345 strncmp(mdstat
->metadata_version
+10, container
, strlen(container
)) != 0 ||
1346 mdstat
->metadata_version
[10+strlen(container
)] != '/')
1352 int is_subarray_active(char *subarray
, char *container
)
1354 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1355 struct mdstat_ent
*ent
;
1357 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1358 if (is_container_member(ent
, container
))
1359 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1362 free_mdstat(mdstat
);
1367 /* open_subarray - opens a subarray in a container
1368 * @dev: container device name
1369 * @st: empty supertype
1370 * @quiet: block reporting errors flag
1372 * On success returns an fd to a container and fills in *st
1374 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1377 struct mdinfo
*info
;
1380 fd
= open(dev
, O_RDWR
|O_EXCL
);
1383 fprintf(stderr
, Name
": Couldn't open %s, aborting\n",
1388 st
->devnum
= fd2devnum(fd
);
1389 if (st
->devnum
== NoMdDev
) {
1392 Name
": Failed to determine device number for %s\n",
1397 mdi
= sysfs_read(fd
, st
->devnum
, GET_VERSION
|GET_LEVEL
);
1400 fprintf(stderr
, Name
": Failed to read sysfs for %s\n",
1405 if (mdi
->array
.level
!= UnSet
) {
1407 fprintf(stderr
, Name
": %s is not a container\n", dev
);
1411 st
->ss
= version_to_superswitch(mdi
->text_version
);
1415 Name
": Operation not supported for %s metadata\n",
1420 st
->devname
= devnum2devname(st
->devnum
);
1423 fprintf(stderr
, Name
": Failed to allocate device name\n");
1427 if (!st
->ss
->load_container
) {
1429 fprintf(stderr
, Name
": %s is not a container\n", dev
);
1433 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1435 fprintf(stderr
, Name
": Failed to load metadata for %s\n",
1440 info
= st
->ss
->container_content(st
, subarray
);
1443 fprintf(stderr
, Name
": Failed to find subarray-%s in %s\n",
1453 st
->ss
->free_super(st
);
1469 int add_disk(int mdfd
, struct supertype
*st
,
1470 struct mdinfo
*sra
, struct mdinfo
*info
)
1472 /* Add a device to an array, in one of 2 ways. */
1475 if (st
->ss
->external
) {
1476 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1477 info
->recovery_start
= MaxSector
;
1479 info
->recovery_start
= 0;
1480 rv
= sysfs_add_disk(sra
, info
, 0);
1483 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1487 sd2
= malloc(sizeof(*sd2
));
1489 sd2
->next
= sra
->devs
;
1495 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1499 int remove_disk(int mdfd
, struct supertype
*st
,
1500 struct mdinfo
*sra
, struct mdinfo
*info
)
1503 /* Remove the disk given by 'info' from the array */
1505 if (st
->ss
->external
)
1506 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1509 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1514 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1516 /* Initialise kernel's knowledge of array.
1517 * This varies between externally managed arrays
1520 int vers
= md_get_version(mdfd
);
1524 if (st
->ss
->external
)
1525 rv
= sysfs_set_array(info
, vers
);
1528 if ((vers
% 100) >= 1) { /* can use different versions */
1529 mdu_array_info_t inf
;
1530 memset(&inf
, 0, sizeof(inf
));
1531 inf
.major_version
= info
->array
.major_version
;
1532 inf
.minor_version
= info
->array
.minor_version
;
1533 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1535 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1539 unsigned long long min_recovery_start(struct mdinfo
*array
)
1541 /* find the minimum recovery_start in an array for metadata
1542 * formats that only record per-array recovery progress instead
1545 unsigned long long recovery_start
= MaxSector
;
1548 for (d
= array
->devs
; d
; d
= d
->next
)
1549 recovery_start
= min(recovery_start
, d
->recovery_start
);
1551 return recovery_start
;
1554 int mdmon_pid(int devnum
)
1560 char *devname
= devnum2devname(devnum
);
1562 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devname
);
1565 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1569 n
= read(fd
, pid
, 9);
1576 int mdmon_running(int devnum
)
1578 int pid
= mdmon_pid(devnum
);
1581 if (kill(pid
, 0) == 0)
1586 int start_mdmon(int devnum
)
1600 if (check_env("MDADM_NO_MDMON"))
1603 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1607 sl
= strrchr(pathbuf
, '/');
1612 strcpy(sl
, "mdmon");
1618 /* FIXME yuk. CLOSE_EXEC?? */
1620 for (i
=3; skipped
< 20; i
++)
1626 for (i
=0; paths
[i
]; i
++)
1629 execl(paths
[i
], "mdmon", "--offroot",
1630 devnum2devname(devnum
),
1633 execl(paths
[i
], "mdmon",
1634 devnum2devname(devnum
),
1639 case -1: fprintf(stderr
, Name
": cannot run mdmon. "
1640 "Array remains readonly\n");
1642 default: /* parent - good */
1643 pid
= wait(&status
);
1644 if (pid
< 0 || status
!= 0)
1650 int check_env(char *name
)
1652 char *val
= getenv(name
);
1654 if (val
&& atoi(val
) == 1)
1660 __u32
random32(void)
1663 int rfd
= open("/dev/urandom", O_RDONLY
);
1664 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1672 int flush_metadata_updates(struct supertype
*st
)
1676 st
->update_tail
= NULL
;
1680 sfd
= connect_monitor(devnum2devname(st
->container_dev
));
1684 while (st
->updates
) {
1685 struct metadata_update
*mu
= st
->updates
;
1686 st
->updates
= mu
->next
;
1688 send_message(sfd
, mu
, 0);
1696 st
->update_tail
= NULL
;
1700 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
1703 struct metadata_update
*mu
= malloc(sizeof(*mu
));
1708 mu
->space_list
= NULL
;
1710 *st
->update_tail
= mu
;
1711 st
->update_tail
= &mu
->next
;
1713 #endif /* MDASSEMBLE */
1716 /* tinyc doesn't optimize this check in ioctl.h out ... */
1717 unsigned int __invalid_size_argument_for_IOC
= 0;
1720 int experimental(void)
1722 if (check_env("MDADM_EXPERIMENTAL"))
1725 fprintf(stderr
, Name
": To use this feature MDADM_EXPERIMENTAL"
1726 " environment variable has to be defined.\n");
1731 /* Pick all spares matching given criteria from a container
1732 * if min_size == 0 do not check size
1733 * if domlist == NULL do not check domains
1734 * if spare_group given add it to domains of each spare
1735 * metadata allows to test domains using metadata of destination array */
1736 struct mdinfo
*container_choose_spares(struct supertype
*st
,
1737 unsigned long long min_size
,
1738 struct domainlist
*domlist
,
1740 const char *metadata
, int get_one
)
1742 struct mdinfo
*d
, **dp
, *disks
= NULL
;
1744 /* get list of all disks in container */
1745 if (st
->ss
->getinfo_super_disks
)
1746 disks
= st
->ss
->getinfo_super_disks(st
);
1750 /* find spare devices on the list */
1752 disks
->array
.spare_disks
= 0;
1756 if (d
->disk
.state
== 0) {
1757 /* check if size is acceptable */
1758 unsigned long long dev_size
;
1759 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
1762 (dev_size_from_id(dev
, &dev_size
) &&
1763 dev_size
>= min_size
))
1765 /* check if domain matches */
1766 if (found
&& domlist
) {
1767 struct dev_policy
*pol
= devnum_policy(dev
);
1769 pol_add(&pol
, pol_domain
,
1771 if (domain_test(domlist
, pol
, metadata
) != 1)
1773 dev_policy_free(pol
);
1778 disks
->array
.spare_disks
++;