2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2012 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
++)
95 while ((c
= *str
++) != 0) {
97 if (c
>= '0' && c
<= '9')
99 else if (c
>= 'a' && c
<= 'f')
101 else if (c
>= 'A' && c
<= 'F')
103 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)
197 * -1 returned on error.
200 long long s
= strtoll(size
, &c
, 10);
214 s
*= 1024 * 1024 * 2;
223 int parse_layout_10(char *layout
)
227 /* Parse the layout string for raid10 */
228 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
229 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
230 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
234 if (layout
[0] == 'n')
236 else if (layout
[0] == 'o')
237 rv
= 0x10000 + (copies
<<8) + 1;
239 rv
= 1 + (copies
<<8);
243 int parse_layout_faulty(char *layout
)
245 /* Parse the layout string for 'faulty' */
246 int ln
= strcspn(layout
, "0123456789");
247 char *m
= xstrdup(layout
);
250 mode
= map_name(faultylayout
, m
);
254 return mode
| (atoi(layout
+ln
)<< ModeShift
);
257 long parse_num(char *num
)
259 /* Either return a valid number, or -1 */
261 long rv
= strtol(num
, &c
, 10);
262 if (rv
< 0 || *c
|| !num
[0])
269 void remove_partitions(int fd
)
271 /* remove partitions from this block devices.
272 * This is used for components added to an array
274 #ifdef BLKPG_DEL_PARTITION
275 struct blkpg_ioctl_arg a
;
276 struct blkpg_partition p
;
278 a
.op
= BLKPG_DEL_PARTITION
;
280 a
.datalen
= sizeof(p
);
282 memset(a
.data
, 0, a
.datalen
);
283 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
284 ioctl(fd
, BLKPG
, &a
);
288 int test_partition(int fd
)
290 /* Check if fd is a whole-disk or a partition.
291 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
292 * will return ENXIO on an invalid partition number.
294 struct blkpg_ioctl_arg a
;
295 struct blkpg_partition p
;
296 a
.op
= BLKPG_DEL_PARTITION
;
298 a
.datalen
= sizeof(p
);
300 memset(a
.data
, 0, a
.datalen
);
302 if (ioctl(fd
, BLKPG
, &a
) == 0)
303 /* Very unlikely, but not a partition */
306 /* not a partition */
312 int test_partition_from_id(dev_t id
)
317 sprintf(buf
, "%d:%d", major(id
), minor(id
));
318 fd
= dev_open(buf
, O_RDONLY
);
321 rv
= test_partition(fd
);
326 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
332 for (i
= 0; i
< raid_disks
; i
++)
333 avail_disks
+= !!avail
[i
];
337 /* This is the tricky one - we need to check
338 * which actual disks are present.
340 copies
= (layout
&255)* ((layout
>>8) & 255);
343 /* there must be one of the 'copies' form 'first' */
349 first
= (first
+1) % raid_disks
;
354 } while (first
!= 0);
357 case LEVEL_MULTIPATH
:
358 return avail_disks
>= 1;
361 return avail_disks
== raid_disks
;
363 return avail_disks
>= 1;
367 return avail_disks
>= raid_disks
-1;
369 return avail_disks
>= raid_disks
;
372 return avail_disks
>= raid_disks
-2;
374 return avail_disks
>= raid_disks
;
380 int enough_fd(int fd
)
382 struct mdu_array_info_s array
;
383 struct mdu_disk_info_s disk
;
388 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0 ||
389 array
.raid_disks
<= 0)
391 avail
= xcalloc(array
.raid_disks
, 1);
392 for (i
= 0; i
< MAX_DISKS
&& array
.nr_disks
> 0; i
++) {
394 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
396 if (disk
.major
== 0 && disk
.minor
== 0)
400 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
402 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
405 avail
[disk
.raid_disk
] = 1;
407 /* This is used on an active array, so assume it is clean */
408 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
414 const int uuid_zero
[4] = { 0, 0, 0, 0 };
416 int same_uuid(int a
[4], int b
[4], int swapuuid
)
419 /* parse uuids are hostendian.
420 * uuid's from some superblocks are big-ending
421 * if there is a difference, we need to swap..
423 unsigned char *ac
= (unsigned char *)a
;
424 unsigned char *bc
= (unsigned char *)b
;
426 for (i
= 0; i
< 16; i
+= 4) {
427 if (ac
[i
+0] != bc
[i
+3] ||
428 ac
[i
+1] != bc
[i
+2] ||
429 ac
[i
+2] != bc
[i
+1] ||
444 void copy_uuid(void *a
, int b
[4], int swapuuid
)
447 /* parse uuids are hostendian.
448 * uuid's from some superblocks are big-ending
449 * if there is a difference, we need to swap..
451 unsigned char *ac
= (unsigned char *)a
;
452 unsigned char *bc
= (unsigned char *)b
;
454 for (i
= 0; i
< 16; i
+= 4) {
464 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
471 copy_uuid(uuid
, id
, swap
);
472 for (i
= 0; i
< 4; i
++) {
475 for (j
= 3; j
>= 0; j
--) {
476 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
484 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
486 // dirty hack to work around an issue with super1 superblocks...
487 // super1 superblocks need swapuuid set in order for assembly to
488 // work, but can't have it set if we want this printout to match
489 // all the other uuid printouts in super1.c, so we force swapuuid
490 // to 1 to make our printout match the rest of super1
491 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
495 int check_ext2(int fd
, char *name
)
498 * Check for an ext2fs file system.
499 * Superblock is always 1K at 1K offset
501 * s_magic is le16 at 56 == 0xEF53
502 * report mtime - le32 at 44
504 * logblksize - le32 at 24
506 unsigned char sb
[1024];
509 if (lseek(fd
, 1024,0)!= 1024)
511 if (read(fd
, sb
, 1024)!= 1024)
513 if (sb
[56] != 0x53 || sb
[57] != 0xef)
516 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
517 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
518 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
519 pr_err("%s appears to contain an ext2fs file system\n",
521 fprintf(stderr
," size=%dK mtime=%s",
522 size
*(1<<bsize
), ctime(&mtime
));
526 int check_reiser(int fd
, char *name
)
529 * superblock is at 64K
531 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
534 unsigned char sb
[1024];
536 if (lseek(fd
, 64*1024, 0) != 64*1024)
538 if (read(fd
, sb
, 1024) != 1024)
540 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
541 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
543 pr_err("%s appears to contain a reiserfs file system\n",name
);
544 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
545 fprintf(stderr
, " size = %luK\n", size
*4);
550 int check_raid(int fd
, char *name
)
555 struct supertype
*st
= guess_super(fd
);
558 st
->ignore_hw_compat
= 1;
559 st
->ss
->load_super(st
, fd
, name
);
560 /* Looks like a raid array .. */
561 pr_err("%s appears to be part of a raid array:\n",
563 st
->ss
->getinfo_super(st
, &info
, NULL
);
564 st
->ss
->free_super(st
);
565 crtime
= info
.array
.ctime
;
566 level
= map_num(pers
, info
.array
.level
);
567 if (!level
) level
= "-unknown-";
568 fprintf(stderr
, " level=%s devices=%d ctime=%s",
569 level
, info
.array
.raid_disks
, ctime(&crtime
));
577 for (i
= 0; i
< 5; i
++) {
579 fprintf(stderr
, "%s%s", mesg
, add
);
581 if (fgets(buf
, 100, stdin
)==NULL
)
583 if (buf
[0]=='y' || buf
[0]=='Y')
585 if (buf
[0]=='n' || buf
[0]=='N')
589 pr_err("assuming 'no'\n");
592 #endif /* MDASSEMBLE */
594 int is_standard(char *dev
, int *nump
)
596 /* tests if dev is a "standard" md dev name.
597 * i.e if the last component is "/dNN" or "/mdNN",
598 * where NN is a string of digits
599 * Returns 1 if a partitionable standard,
600 * -1 if non-partitonable,
601 * 0 if not a standard name.
603 char *d
= strrchr(dev
, '/');
608 if (strncmp(d
, "/d",2) == 0)
609 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
610 else if (strncmp(d
, "/md_d", 5) == 0)
611 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
612 else if (strncmp(d
, "/md", 3) == 0)
613 d
+= 3, type
= -1; /* /dev/mdN */
614 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
615 d
+= 1, type
= -1; /* /dev/md/N */
625 if (nump
) *nump
= num
;
630 unsigned long calc_csum(void *super
, int bytes
)
632 unsigned long long newcsum
= 0;
635 unsigned int *superc
= (unsigned int*) super
;
637 for(i
= 0; i
< bytes
/4; i
++)
638 newcsum
+= superc
[i
];
639 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
641 /* The in-kernel checksum calculation is always 16bit on
642 * the alpha, though it is 32 bit on i386...
643 * I wonder what it is elsewhere... (it uses an API in
644 * a way that it shouldn't).
646 csum
= (csum
& 0xffff) + (csum
>> 16);
647 csum
= (csum
& 0xffff) + (csum
>> 16);
653 char *human_size(long long bytes
)
657 /* We convert bytes to either centi-M{ega,ibi}bytes or
658 * centi-G{igi,ibi}bytes, with appropriate rounding,
659 * and then print 1/100th of those as a decimal.
660 * We allow upto 2048Megabytes before converting to
661 * gigabytes, as that shows more precision and isn't
662 * too large a number.
663 * Terabytes are not yet handled.
666 if (bytes
< 5000*1024)
668 else if (bytes
< 2*1024LL*1024LL*1024LL) {
669 long cMiB
= (bytes
/ ( (1LL<<20) / 200LL ) +1) /2;
670 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
671 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
672 cMiB
/100 , cMiB
% 100,
675 long cGiB
= (bytes
/ ( (1LL<<30) / 200LL ) +1) /2;
676 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
677 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
678 cGiB
/100 , cGiB
% 100,
684 char *human_size_brief(long long bytes
)
688 if (bytes
< 5000*1024)
689 snprintf(buf
, sizeof(buf
), "%ld.%02ldKiB",
690 (long)(bytes
>>10), (long)(((bytes
&1023)*100+512)/1024)
692 else if (bytes
< 2*1024LL*1024LL*1024LL)
693 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
695 (long)((bytes
&0xfffff)+0x100000/200)/(0x100000/100)
698 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
700 (long)(((bytes
>>10)&0xfffff)+0x100000/200)/(0x100000/100)
705 void print_r10_layout(int layout
)
707 int near
= layout
& 255;
708 int far
= (layout
>> 8) & 255;
709 int offset
= (layout
&0x10000);
713 printf("%s near=%d", sep
, near
);
717 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
719 printf("NO REDUNDANCY");
723 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
724 int chunksize
, unsigned long long devsize
)
728 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
729 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
732 int get_data_disks(int level
, int layout
, int raid_disks
)
736 case 0: data_disks
= raid_disks
;
738 case 1: data_disks
= 1;
741 case 5: data_disks
= raid_disks
- 1;
743 case 6: data_disks
= raid_disks
- 2;
745 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
752 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
753 char *get_md_name(int dev
)
755 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
756 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
757 static char devname
[50];
763 int mdp
= get_mdp_major();
764 if (mdp
< 0) return NULL
;
765 rdev
= makedev(mdp
, (-1-dev
)<<6);
766 snprintf(devname
, sizeof(devname
), "/dev/md/d%d", -1-dev
);
767 if (stat(devname
, &stb
) == 0
768 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
769 && (stb
.st_rdev
== rdev
))
772 rdev
= makedev(MD_MAJOR
, dev
);
773 snprintf(devname
, sizeof(devname
), "/dev/md%d", dev
);
774 if (stat(devname
, &stb
) == 0
775 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
776 && (stb
.st_rdev
== rdev
))
779 snprintf(devname
, sizeof(devname
), "/dev/md/%d", dev
);
780 if (stat(devname
, &stb
) == 0
781 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
782 && (stb
.st_rdev
== rdev
))
785 dn
= map_dev(major(rdev
), minor(rdev
), 0);
788 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md%d", dev
);
789 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
793 if (stat(devname
, &stb
) == 0
794 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
795 && (stb
.st_rdev
== rdev
))
801 void put_md_name(char *name
)
803 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
807 int find_free_devnum(int use_partitions
)
810 for (devnum
= 127; devnum
!= 128;
811 devnum
= devnum
? devnum
-1 : (1<<20)-1) {
816 _devnum
= use_partitions
? (-1-devnum
) : devnum
;
817 if (mddev_busy(_devnum
))
819 sprintf(nbuf
, "%s%d", use_partitions
?"mdp":"md", devnum
);
820 if (!conf_name_is_free(nbuf
))
822 /* make sure it is new to /dev too, at least as a
824 dn
= map_dev(dev2major(_devnum
), dev2minor(_devnum
), 0);
825 if (dn
&& ! is_standard(dn
, NULL
))
831 return use_partitions
? (-1-devnum
) : devnum
;
833 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
835 int dev_open(char *dev
, int flags
)
837 /* like 'open', but if 'dev' matches %d:%d, create a temp
838 * block device and open that
849 major
= strtoul(dev
, &e
, 0);
850 if (e
> dev
&& *e
== ':' && e
[1] &&
851 (minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
853 char *path
= map_dev(major
, minor
, 0);
855 fd
= open(path
, flags
);
857 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
858 (int)getpid(), major
, minor
);
859 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
860 fd
= open(devname
, flags
);
865 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
866 (int)getpid(), major
, minor
);
867 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
868 fd
= open(devname
, flags
);
873 fd
= open(dev
, flags
);
877 int open_dev_flags(int devnum
, int flags
)
881 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
882 return dev_open(buf
, flags
);
885 int open_dev(int devnum
)
887 return open_dev_flags(devnum
, O_RDONLY
);
890 int open_dev_excl(int devnum
)
896 sprintf(buf
, "%d:%d", dev2major(devnum
), dev2minor(devnum
));
897 for (i
= 0 ; i
< 25 ; i
++) {
898 int fd
= dev_open(buf
, flags
|O_EXCL
);
901 if (errno
== EACCES
&& flags
== O_RDWR
) {
912 int same_dev(char *one
, char *two
)
914 struct stat st1
, st2
;
915 if (stat(one
, &st1
) != 0)
917 if (stat(two
, &st2
) != 0)
919 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
921 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
923 return st1
.st_rdev
== st2
.st_rdev
;
926 void wait_for(char *dev
, int fd
)
929 struct stat stb_want
;
931 if (fstat(fd
, &stb_want
) != 0 ||
932 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
935 for (i
= 0 ; i
< 25 ; i
++) {
937 if (stat(dev
, &stb
) == 0 &&
938 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
939 (stb
.st_rdev
== stb_want
.st_rdev
))
944 dprintf("%s: timeout waiting for %s\n", __func__
, dev
);
947 struct superswitch
*superlist
[] =
950 &super_ddf
, &super_imsm
,
954 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
956 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
958 mdu_array_info_t array
;
961 struct supertype
*st
= NULL
;
966 char *subarray
= NULL
;
967 int container
= NoMdDev
;
969 sra
= sysfs_read(fd
, 0, GET_VERSION
);
972 vers
= sra
->array
.major_version
;
973 minor
= sra
->array
.minor_version
;
974 verstr
= sra
->text_version
;
976 if (ioctl(fd
, GET_ARRAY_INFO
, &array
))
977 array
.major_version
= array
.minor_version
= 0;
978 vers
= array
.major_version
;
979 minor
= array
.minor_version
;
984 sprintf(version
, "%d.%d", vers
, minor
);
987 if (minor
== -2 && is_subarray(verstr
)) {
988 char *dev
= verstr
+1;
990 subarray
= strchr(dev
, '/');
993 subarray
= xstrdup(subarray
);
995 container
= devname2devnum(dev
);
998 sra
= sysfs_read(-1, container
, GET_VERSION
);
999 if (sra
&& sra
->text_version
[0])
1000 verstr
= sra
->text_version
;
1002 verstr
= "-no-metadata-";
1005 for (i
= 0; st
== NULL
&& superlist
[i
] ; i
++)
1006 st
= superlist
[i
]->match_metadata_desc(verstr
);
1013 *subarrayp
= subarray
;
1014 st
->container_dev
= container
;
1015 st
->devnum
= fd2devnum(fd
);
1021 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1023 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1028 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1029 fd
= dev_open(buf
, O_RDONLY
);
1032 if (get_dev_size(fd
, NULL
, size
)) {
1040 struct supertype
*dup_super(struct supertype
*orig
)
1042 struct supertype
*st
;
1046 st
= xcalloc(1, sizeof(*st
));
1048 st
->max_devs
= orig
->max_devs
;
1049 st
->minor_version
= orig
->minor_version
;
1055 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1057 /* try each load_super to find the best match,
1058 * and return the best superswitch
1060 struct superswitch
*ss
;
1061 struct supertype
*st
;
1062 time_t besttime
= 0;
1066 st
= xcalloc(1, sizeof(*st
));
1067 st
->container_dev
= NoMdDev
;
1069 for (i
= 0 ; superlist
[i
]; i
++) {
1072 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1074 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1076 memset(st
, 0, sizeof(*st
));
1077 st
->ignore_hw_compat
= 1;
1078 rv
= ss
->load_super(st
, fd
, NULL
);
1081 st
->ss
->getinfo_super(st
, &info
, NULL
);
1082 if (bestsuper
== -1 ||
1083 besttime
< info
.array
.ctime
) {
1085 besttime
= info
.array
.ctime
;
1090 if (bestsuper
!= -1) {
1092 memset(st
, 0, sizeof(*st
));
1093 st
->ignore_hw_compat
= 1;
1094 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1096 superlist
[bestsuper
]->free_super(st
);
1097 st
->ignore_hw_compat
= 0;
1105 /* Return size of device in bytes */
1106 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1108 unsigned long long ldsize
;
1111 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1112 ldsize
= (unsigned long long)st
.st_size
;
1115 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1118 unsigned long dsize
;
1119 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1124 pr_err("Cannot get size of %s: %s\b",
1125 dname
, strerror(errno
));
1133 /* Return true if this can only be a container, not a member device.
1134 * i.e. is and md device and size is zero
1136 int must_be_container(int fd
)
1138 unsigned long long size
;
1139 if (md_get_version(fd
) < 0)
1141 if (get_dev_size(fd
, NULL
, &size
) == 0)
1148 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1149 * Returns: 1 if successful
1150 * -1 for unknown partition type
1151 * 0 for other errors
1153 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1156 unsigned char empty_gpt_entry
[16]= {0};
1157 struct GPT_part_entry
*part
;
1159 unsigned long long curr_part_end
;
1160 unsigned all_partitions
, entry_size
;
1165 BUILD_BUG_ON(sizeof(gpt
) != 512);
1166 /* skip protective MBR */
1167 lseek(fd
, 512, SEEK_SET
);
1168 /* read GPT header */
1169 if (read(fd
, &gpt
, 512) != 512)
1172 /* get the number of partition entries and the entry size */
1173 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1174 entry_size
= __le32_to_cpu(gpt
.part_size
);
1176 /* Check GPT signature*/
1177 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1181 if (all_partitions
> 1024 ||
1182 entry_size
> sizeof(buf
))
1185 part
= (struct GPT_part_entry
*)buf
;
1187 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1188 /* read partition entry */
1189 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1192 /* is this valid partition? */
1193 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1194 /* check the last lba for the current partition */
1195 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1196 if (curr_part_end
> *endofpart
)
1197 *endofpart
= curr_part_end
;
1204 /* Sets endofpart parameter to the last block used by the last partition on the device.
1205 * Returns: 1 if successful
1206 * -1 for unknown partition type
1207 * 0 for other errors
1209 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1211 struct MBR boot_sect
;
1212 struct MBR_part_record
*part
;
1213 unsigned long long curr_part_end
;
1219 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1222 if (read(fd
, &boot_sect
, 512) != 512)
1225 /* check MBP signature */
1226 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1228 /* found the correct signature */
1229 part
= boot_sect
.parts
;
1231 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1232 /* check for GPT type */
1233 if (part
->part_type
== MBR_GPT_PARTITION_TYPE
) {
1234 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1237 /* check the last used lba for the current partition */
1238 curr_part_end
= __le32_to_cpu(part
->first_sect_lba
) +
1239 __le32_to_cpu(part
->blocks_num
);
1240 if (curr_part_end
> *endofpart
)
1241 *endofpart
= curr_part_end
;
1246 /* Unknown partition table */
1253 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1254 unsigned long long size
)
1257 * Check where the last partition ends
1259 unsigned long long endofpart
;
1262 if ((ret
= get_last_partition_end(fd
, &endofpart
)) > 0) {
1263 /* There appears to be a partition table here */
1264 if (freesize
== 0) {
1265 /* partitions will not be visible in new device */
1266 pr_err("partition table exists on %s but will be lost or\n"
1267 " meaningless after creating array\n",
1270 } else if (endofpart
> freesize
) {
1271 /* last partition overlaps metadata */
1272 pr_err("metadata will over-write last partition on %s.\n",
1275 } else if (size
&& endofpart
> size
) {
1276 /* partitions will be truncated in new device */
1277 pr_err("array size is too small to cover all partitions on %s.\n",
1285 void get_one_disk(int mdfd
, mdu_array_info_t
*ainf
, mdu_disk_info_t
*disk
)
1289 ioctl(mdfd
, GET_ARRAY_INFO
, ainf
);
1290 for (d
= 0 ; d
< MAX_DISKS
; d
++) {
1291 if (ioctl(mdfd
, GET_DISK_INFO
, disk
) == 0 &&
1292 (disk
->major
|| disk
->minor
))
1297 int open_container(int fd
)
1299 /* 'fd' is a block device. Find out if it is in use
1300 * by a container, and return an open fd on that container.
1311 if (fstat(fd
, &st
) != 0)
1313 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1314 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1315 e
= path
+ strlen(path
);
1317 dir
= opendir(path
);
1320 while ((de
= readdir(dir
))) {
1323 if (de
->d_name
[0] == '.')
1325 sprintf(e
, "/%s/dev", de
->d_name
);
1326 dfd
= open(path
, O_RDONLY
);
1329 n
= read(dfd
, buf
, sizeof(buf
));
1331 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1334 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1336 sprintf(buf
, "%d:%d", major
, minor
);
1337 dfd
= dev_open(buf
, O_RDONLY
);
1347 struct superswitch
*version_to_superswitch(char *vers
)
1351 for (i
= 0; superlist
[i
]; i
++) {
1352 struct superswitch
*ss
= superlist
[i
];
1354 if (strcmp(vers
, ss
->name
) == 0)
1361 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1363 if (mdstat
->metadata_version
== NULL
||
1364 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1365 !is_subarray(mdstat
->metadata_version
+9) ||
1366 strncmp(mdstat
->metadata_version
+10, container
, strlen(container
)) != 0 ||
1367 mdstat
->metadata_version
[10+strlen(container
)] != '/')
1373 int is_subarray_active(char *subarray
, char *container
)
1375 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1376 struct mdstat_ent
*ent
;
1378 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1379 if (is_container_member(ent
, container
))
1380 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1383 free_mdstat(mdstat
);
1388 /* open_subarray - opens a subarray in a container
1389 * @dev: container device name
1390 * @st: empty supertype
1391 * @quiet: block reporting errors flag
1393 * On success returns an fd to a container and fills in *st
1395 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1398 struct mdinfo
*info
;
1401 fd
= open(dev
, O_RDWR
|O_EXCL
);
1404 pr_err("Couldn't open %s, aborting\n",
1409 st
->devnum
= fd2devnum(fd
);
1410 if (st
->devnum
== NoMdDev
) {
1412 pr_err("Failed to determine device number for %s\n",
1417 mdi
= sysfs_read(fd
, st
->devnum
, GET_VERSION
|GET_LEVEL
);
1420 pr_err("Failed to read sysfs for %s\n",
1425 if (mdi
->array
.level
!= UnSet
) {
1427 pr_err("%s is not a container\n", dev
);
1431 st
->ss
= version_to_superswitch(mdi
->text_version
);
1434 pr_err("Operation not supported for %s metadata\n",
1439 st
->devname
= devnum2devname(st
->devnum
);
1442 pr_err("Failed to allocate device name\n");
1446 if (!st
->ss
->load_container
) {
1448 pr_err("%s is not a container\n", dev
);
1452 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1454 pr_err("Failed to load metadata for %s\n",
1459 info
= st
->ss
->container_content(st
, subarray
);
1462 pr_err("Failed to find subarray-%s in %s\n",
1472 st
->ss
->free_super(st
);
1488 int add_disk(int mdfd
, struct supertype
*st
,
1489 struct mdinfo
*sra
, struct mdinfo
*info
)
1491 /* Add a device to an array, in one of 2 ways. */
1494 if (st
->ss
->external
) {
1495 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1496 info
->recovery_start
= MaxSector
;
1498 info
->recovery_start
= 0;
1499 rv
= sysfs_add_disk(sra
, info
, 0);
1502 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1506 sd2
= xmalloc(sizeof(*sd2
));
1508 sd2
->next
= sra
->devs
;
1514 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1518 int remove_disk(int mdfd
, struct supertype
*st
,
1519 struct mdinfo
*sra
, struct mdinfo
*info
)
1522 /* Remove the disk given by 'info' from the array */
1524 if (st
->ss
->external
)
1525 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1528 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1533 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1535 /* Initialise kernel's knowledge of array.
1536 * This varies between externally managed arrays
1539 int vers
= md_get_version(mdfd
);
1543 if (st
->ss
->external
)
1544 rv
= sysfs_set_array(info
, vers
);
1547 if ((vers
% 100) >= 1) { /* can use different versions */
1548 mdu_array_info_t inf
;
1549 memset(&inf
, 0, sizeof(inf
));
1550 inf
.major_version
= info
->array
.major_version
;
1551 inf
.minor_version
= info
->array
.minor_version
;
1552 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1554 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1558 unsigned long long min_recovery_start(struct mdinfo
*array
)
1560 /* find the minimum recovery_start in an array for metadata
1561 * formats that only record per-array recovery progress instead
1564 unsigned long long recovery_start
= MaxSector
;
1567 for (d
= array
->devs
; d
; d
= d
->next
)
1568 recovery_start
= min(recovery_start
, d
->recovery_start
);
1570 return recovery_start
;
1573 int mdmon_pid(int devnum
)
1579 char *devname
= devnum2devname(devnum
);
1581 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devname
);
1584 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1588 n
= read(fd
, pid
, 9);
1595 int mdmon_running(int devnum
)
1597 int pid
= mdmon_pid(devnum
);
1600 if (kill(pid
, 0) == 0)
1605 int start_mdmon(int devnum
)
1619 if (check_env("MDADM_NO_MDMON"))
1622 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1626 sl
= strrchr(pathbuf
, '/');
1631 strcpy(sl
, "mdmon");
1637 /* FIXME yuk. CLOSE_EXEC?? */
1639 for (i
= 3; skipped
< 20; i
++)
1645 for (i
= 0; paths
[i
]; i
++)
1648 execl(paths
[i
], "mdmon", "--offroot",
1649 devnum2devname(devnum
),
1652 execl(paths
[i
], "mdmon",
1653 devnum2devname(devnum
),
1658 case -1: pr_err("cannot run mdmon. "
1659 "Array remains readonly\n");
1661 default: /* parent - good */
1662 pid
= wait(&status
);
1663 if (pid
< 0 || status
!= 0)
1669 int check_env(char *name
)
1671 char *val
= getenv(name
);
1673 if (val
&& atoi(val
) == 1)
1679 __u32
random32(void)
1682 int rfd
= open("/dev/urandom", O_RDONLY
);
1683 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1691 int flush_metadata_updates(struct supertype
*st
)
1695 st
->update_tail
= NULL
;
1699 sfd
= connect_monitor(devnum2devname(st
->container_dev
));
1703 while (st
->updates
) {
1704 struct metadata_update
*mu
= st
->updates
;
1705 st
->updates
= mu
->next
;
1707 send_message(sfd
, mu
, 0);
1715 st
->update_tail
= NULL
;
1719 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
1722 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
1727 mu
->space_list
= NULL
;
1729 *st
->update_tail
= mu
;
1730 st
->update_tail
= &mu
->next
;
1732 #endif /* MDASSEMBLE */
1735 /* tinyc doesn't optimize this check in ioctl.h out ... */
1736 unsigned int __invalid_size_argument_for_IOC
= 0;
1739 int experimental(void)
1741 if (check_env("MDADM_EXPERIMENTAL"))
1744 pr_err("To use this feature MDADM_EXPERIMENTAL"
1745 " environment variable has to be defined.\n");
1750 /* Pick all spares matching given criteria from a container
1751 * if min_size == 0 do not check size
1752 * if domlist == NULL do not check domains
1753 * if spare_group given add it to domains of each spare
1754 * metadata allows to test domains using metadata of destination array */
1755 struct mdinfo
*container_choose_spares(struct supertype
*st
,
1756 unsigned long long min_size
,
1757 struct domainlist
*domlist
,
1759 const char *metadata
, int get_one
)
1761 struct mdinfo
*d
, **dp
, *disks
= NULL
;
1763 /* get list of all disks in container */
1764 if (st
->ss
->getinfo_super_disks
)
1765 disks
= st
->ss
->getinfo_super_disks(st
);
1769 /* find spare devices on the list */
1771 disks
->array
.spare_disks
= 0;
1775 if (d
->disk
.state
== 0) {
1776 /* check if size is acceptable */
1777 unsigned long long dev_size
;
1778 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
1781 (dev_size_from_id(dev
, &dev_size
) &&
1782 dev_size
>= min_size
))
1784 /* check if domain matches */
1785 if (found
&& domlist
) {
1786 struct dev_policy
*pol
= devnum_policy(dev
);
1788 pol_add(&pol
, pol_domain
,
1790 if (domain_test(domlist
, pol
, metadata
) != 1)
1792 dev_policy_free(pol
);
1797 disks
->array
.spare_disks
++;
1811 void *xmalloc(size_t len
)
1813 void *rv
= malloc(len
);
1817 msg
= Name
": memory allocation failure - aborting\n";
1818 exit(4+!!write(2, msg
, strlen(msg
)));
1821 void *xrealloc(void *ptr
, size_t len
)
1823 void *rv
= realloc(ptr
, len
);
1827 msg
= Name
": memory allocation failure - aborting\n";
1828 exit(4+!!write(2, msg
, strlen(msg
)));
1831 void *xcalloc(size_t num
, size_t size
)
1833 void *rv
= calloc(num
, size
);
1837 msg
= Name
": memory allocation failure - aborting\n";
1838 exit(4+!!write(2, msg
, strlen(msg
)));
1841 char *xstrdup(const char *str
)
1843 char *rv
= strdup(str
);
1847 msg
= Name
": memory allocation failure - aborting\n";
1848 exit(4+!!write(2, msg
, strlen(msg
)));