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>
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
++)
93 while ((c
= *str
++) != 0) {
95 if (c
>= '0' && c
<= '9')
97 else if (c
>= 'a' && c
<= 'f')
99 else if (c
>= 'A' && c
<= 'F')
101 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 unsigned 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)
195 * INVALID_SECTORS returned on error.
198 long long s
= strtoll(size
, &c
, 10);
212 s
*= 1024 * 1024 * 2;
214 case 's': /* sectors */
225 int parse_layout_10(char *layout
)
229 /* Parse the layout string for raid10 */
230 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
231 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
232 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
236 if (layout
[0] == 'n')
238 else if (layout
[0] == 'o')
239 rv
= 0x10000 + (copies
<<8) + 1;
241 rv
= 1 + (copies
<<8);
245 int parse_layout_faulty(char *layout
)
247 /* Parse the layout string for 'faulty' */
248 int ln
= strcspn(layout
, "0123456789");
249 char *m
= xstrdup(layout
);
252 mode
= map_name(faultylayout
, m
);
256 return mode
| (atoi(layout
+ln
)<< ModeShift
);
259 long parse_num(char *num
)
261 /* Either return a valid number, or -1 */
263 long rv
= strtol(num
, &c
, 10);
264 if (rv
< 0 || *c
|| !num
[0])
271 void remove_partitions(int fd
)
273 /* remove partitions from this block devices.
274 * This is used for components added to an array
276 #ifdef BLKPG_DEL_PARTITION
277 struct blkpg_ioctl_arg a
;
278 struct blkpg_partition p
;
280 a
.op
= BLKPG_DEL_PARTITION
;
282 a
.datalen
= sizeof(p
);
284 memset(a
.data
, 0, a
.datalen
);
285 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
286 ioctl(fd
, BLKPG
, &a
);
290 int test_partition(int fd
)
292 /* Check if fd is a whole-disk or a partition.
293 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
294 * will return ENXIO on an invalid partition number.
296 struct blkpg_ioctl_arg a
;
297 struct blkpg_partition p
;
298 a
.op
= BLKPG_DEL_PARTITION
;
300 a
.datalen
= sizeof(p
);
302 memset(a
.data
, 0, a
.datalen
);
304 if (ioctl(fd
, BLKPG
, &a
) == 0)
305 /* Very unlikely, but not a partition */
308 /* not a partition */
314 int test_partition_from_id(dev_t id
)
319 sprintf(buf
, "%d:%d", major(id
), minor(id
));
320 fd
= dev_open(buf
, O_RDONLY
);
323 rv
= test_partition(fd
);
328 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
334 for (i
= 0; i
< raid_disks
; i
++)
335 avail_disks
+= !!avail
[i
];
339 /* This is the tricky one - we need to check
340 * which actual disks are present.
342 copies
= (layout
&255)* ((layout
>>8) & 255);
345 /* there must be one of the 'copies' form 'first' */
352 this = (this+1) % raid_disks
;
356 first
= (first
+(layout
&255)) % raid_disks
;
357 } while (first
!= 0);
360 case LEVEL_MULTIPATH
:
361 return avail_disks
>= 1;
364 return avail_disks
== raid_disks
;
366 return avail_disks
>= 1;
370 return avail_disks
>= raid_disks
-1;
372 return avail_disks
>= raid_disks
;
375 return avail_disks
>= raid_disks
-2;
377 return avail_disks
>= raid_disks
;
383 int enough_fd(int fd
)
385 struct mdu_array_info_s array
;
386 struct mdu_disk_info_s disk
;
391 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0 ||
392 array
.raid_disks
<= 0)
394 avail
= xcalloc(array
.raid_disks
, 1);
395 for (i
= 0; i
< MAX_DISKS
&& array
.nr_disks
> 0; i
++) {
397 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
399 if (disk
.major
== 0 && disk
.minor
== 0)
403 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
405 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
408 avail
[disk
.raid_disk
] = 1;
410 /* This is used on an active array, so assume it is clean */
411 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
417 const int uuid_zero
[4] = { 0, 0, 0, 0 };
419 int same_uuid(int a
[4], int b
[4], int swapuuid
)
422 /* parse uuids are hostendian.
423 * uuid's from some superblocks are big-ending
424 * if there is a difference, we need to swap..
426 unsigned char *ac
= (unsigned char *)a
;
427 unsigned char *bc
= (unsigned char *)b
;
429 for (i
= 0; i
< 16; i
+= 4) {
430 if (ac
[i
+0] != bc
[i
+3] ||
431 ac
[i
+1] != bc
[i
+2] ||
432 ac
[i
+2] != bc
[i
+1] ||
447 void copy_uuid(void *a
, int b
[4], int swapuuid
)
450 /* parse uuids are hostendian.
451 * uuid's from some superblocks are big-ending
452 * if there is a difference, we need to swap..
454 unsigned char *ac
= (unsigned char *)a
;
455 unsigned char *bc
= (unsigned char *)b
;
457 for (i
= 0; i
< 16; i
+= 4) {
467 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
474 copy_uuid(uuid
, id
, swap
);
475 for (i
= 0; i
< 4; i
++) {
478 for (j
= 3; j
>= 0; j
--) {
479 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
487 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
489 // dirty hack to work around an issue with super1 superblocks...
490 // super1 superblocks need swapuuid set in order for assembly to
491 // work, but can't have it set if we want this printout to match
492 // all the other uuid printouts in super1.c, so we force swapuuid
493 // to 1 to make our printout match the rest of super1
494 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
498 int check_ext2(int fd
, char *name
)
501 * Check for an ext2fs file system.
502 * Superblock is always 1K at 1K offset
504 * s_magic is le16 at 56 == 0xEF53
505 * report mtime - le32 at 44
507 * logblksize - le32 at 24
509 unsigned char sb
[1024];
512 if (lseek(fd
, 1024,0)!= 1024)
514 if (read(fd
, sb
, 1024)!= 1024)
516 if (sb
[56] != 0x53 || sb
[57] != 0xef)
519 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
520 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
521 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
522 pr_err("%s appears to contain an ext2fs file system\n",
524 fprintf(stderr
," size=%dK mtime=%s",
525 size
*(1<<bsize
), ctime(&mtime
));
529 int check_reiser(int fd
, char *name
)
532 * superblock is at 64K
534 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
537 unsigned char sb
[1024];
539 if (lseek(fd
, 64*1024, 0) != 64*1024)
541 if (read(fd
, sb
, 1024) != 1024)
543 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
544 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
546 pr_err("%s appears to contain a reiserfs file system\n",name
);
547 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
548 fprintf(stderr
, " size = %luK\n", size
*4);
553 int check_raid(int fd
, char *name
)
558 struct supertype
*st
= guess_super(fd
);
562 st
->ss
->load_super(st
, fd
, name
);
563 /* Looks like a raid array .. */
564 pr_err("%s appears to be part of a raid array:\n",
566 st
->ss
->getinfo_super(st
, &info
, NULL
);
567 st
->ss
->free_super(st
);
568 crtime
= info
.array
.ctime
;
569 level
= map_num(pers
, info
.array
.level
);
570 if (!level
) level
= "-unknown-";
571 fprintf(stderr
, " level=%s devices=%d ctime=%s",
572 level
, info
.array
.raid_disks
, ctime(&crtime
));
580 for (i
= 0; i
< 5; i
++) {
582 fprintf(stderr
, "%s%s", mesg
, add
);
584 if (fgets(buf
, 100, stdin
)==NULL
)
586 if (buf
[0]=='y' || buf
[0]=='Y')
588 if (buf
[0]=='n' || buf
[0]=='N')
592 pr_err("assuming 'no'\n");
595 #endif /* MDASSEMBLE */
597 int is_standard(char *dev
, int *nump
)
599 /* tests if dev is a "standard" md dev name.
600 * i.e if the last component is "/dNN" or "/mdNN",
601 * where NN is a string of digits
602 * Returns 1 if a partitionable standard,
603 * -1 if non-partitonable,
604 * 0 if not a standard name.
606 char *d
= strrchr(dev
, '/');
611 if (strncmp(d
, "/d",2) == 0)
612 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
613 else if (strncmp(d
, "/md_d", 5) == 0)
614 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
615 else if (strncmp(d
, "/md", 3) == 0)
616 d
+= 3, type
= -1; /* /dev/mdN */
617 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
618 d
+= 1, type
= -1; /* /dev/md/N */
628 if (nump
) *nump
= num
;
633 unsigned long calc_csum(void *super
, int bytes
)
635 unsigned long long newcsum
= 0;
638 unsigned int *superc
= (unsigned int*) super
;
640 for(i
= 0; i
< bytes
/4; i
++)
641 newcsum
+= superc
[i
];
642 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
644 /* The in-kernel checksum calculation is always 16bit on
645 * the alpha, though it is 32 bit on i386...
646 * I wonder what it is elsewhere... (it uses an API in
647 * a way that it shouldn't).
649 csum
= (csum
& 0xffff) + (csum
>> 16);
650 csum
= (csum
& 0xffff) + (csum
>> 16);
656 char *human_size(long long bytes
)
660 /* We convert bytes to either centi-M{ega,ibi}bytes or
661 * centi-G{igi,ibi}bytes, with appropriate rounding,
662 * and then print 1/100th of those as a decimal.
663 * We allow upto 2048Megabytes before converting to
664 * gigabytes, as that shows more precision and isn't
665 * too large a number.
666 * Terabytes are not yet handled.
669 if (bytes
< 5000*1024)
671 else if (bytes
< 2*1024LL*1024LL*1024LL) {
672 long cMiB
= (bytes
/ ( (1LL<<20) / 200LL ) +1) /2;
673 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
674 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
675 cMiB
/100 , cMiB
% 100,
678 long cGiB
= (bytes
/ ( (1LL<<30) / 200LL ) +1) /2;
679 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
680 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
681 cGiB
/100 , cGiB
% 100,
687 char *human_size_brief(long long bytes
, int prefix
)
691 /* We convert bytes to either centi-M{ega,ibi}bytes or
692 * centi-G{igi,ibi}bytes, with appropriate rounding,
693 * and then print 1/100th of those as a decimal.
694 * We allow upto 2048Megabytes before converting to
695 * gigabytes, as that shows more precision and isn't
696 * too large a number.
697 * Terabytes are not yet handled.
699 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
700 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
703 if (bytes
< 5000*1024)
705 else if (prefix
== IEC
) {
706 if (bytes
< 2*1024LL*1024LL*1024LL) {
707 long cMiB
= (bytes
/ ( (1LL<<20) / 200LL ) +1) /2;
708 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
709 cMiB
/100 , cMiB
% 100);
711 long cGiB
= (bytes
/ ( (1LL<<30) / 200LL ) +1) /2;
712 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
713 cGiB
/100 , cGiB
% 100);
716 else if (prefix
== JEDEC
) {
717 if (bytes
< 2*1024LL*1024LL*1024LL) {
718 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
719 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
722 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
723 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
724 cGB
/100 , cGB
% 100);
733 void print_r10_layout(int layout
)
735 int near
= layout
& 255;
736 int far
= (layout
>> 8) & 255;
737 int offset
= (layout
&0x10000);
741 printf("%s near=%d", sep
, near
);
745 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
747 printf("NO REDUNDANCY");
751 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
752 int chunksize
, unsigned long long devsize
)
756 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
757 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
760 int get_data_disks(int level
, int layout
, int raid_disks
)
764 case 0: data_disks
= raid_disks
;
766 case 1: data_disks
= 1;
769 case 5: data_disks
= raid_disks
- 1;
771 case 6: data_disks
= raid_disks
- 2;
773 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
780 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
782 int devnm2devid(char *devnm
)
784 /* First look in /sys/block/$DEVNM/dev for %d:%d
785 * If that fails, try parsing out a number
792 sprintf(path
, "/sys/block/%s/dev", devnm
);
793 fd
= open(path
, O_RDONLY
);
796 int n
= read(fd
, buf
, sizeof(buf
));
800 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
801 return makedev(mjr
, mnr
);
803 if (strncmp(devnm
, "md_d", 4) == 0 &&
805 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
806 ep
> devnm
&& *ep
== 0)
807 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
809 if (strncmp(devnm
, "md", 2) == 0 &&
811 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
812 ep
> devnm
&& *ep
== 0)
813 return makedev(MD_MAJOR
, mnr
);
818 char *get_md_name(char *devnm
)
820 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
821 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
823 static char devname
[50];
825 dev_t rdev
= devnm2devid(devnm
);
830 if (strncmp(devnm
, "md_", 3) == 0) {
831 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
833 if (stat(devname
, &stb
) == 0
834 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
835 && (stb
.st_rdev
== rdev
))
838 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
839 if (stat(devname
, &stb
) == 0
840 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
841 && (stb
.st_rdev
== rdev
))
844 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
845 if (stat(devname
, &stb
) == 0
846 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
847 && (stb
.st_rdev
== rdev
))
850 dn
= map_dev(major(rdev
), minor(rdev
), 0);
853 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
854 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
858 if (stat(devname
, &stb
) == 0
859 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
860 && (stb
.st_rdev
== rdev
))
866 void put_md_name(char *name
)
868 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
872 char *find_free_devnm(int use_partitions
)
874 static char devnm
[32];
876 for (devnum
= 127; devnum
!= 128;
877 devnum
= devnum
? devnum
-1 : (1<<20)-1) {
880 sprintf(devnm
, "md_d%d", devnum
);
882 sprintf(devnm
, "md%d", devnum
);
883 if (mddev_busy(devnm
))
885 if (!conf_name_is_free(devnm
))
888 /* make sure it is new to /dev too, at least as a
890 int devid
= devnm2devid(devnm
);
892 char *dn
= map_dev(major(devid
),
894 if (dn
&& ! is_standard(dn
, NULL
))
904 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
906 int dev_open(char *dev
, int flags
)
908 /* like 'open', but if 'dev' matches %d:%d, create a temp
909 * block device and open that
920 major
= strtoul(dev
, &e
, 0);
921 if (e
> dev
&& *e
== ':' && e
[1] &&
922 (minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
924 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
925 (int)getpid(), major
, minor
);
926 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
927 fd
= open(devname
, flags
);
931 /* Try /tmp as /dev appear to be read-only */
932 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
933 (int)getpid(), major
, minor
);
934 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
935 fd
= open(devname
, flags
);
940 fd
= open(dev
, flags
);
944 int open_dev_flags(char *devnm
, int flags
)
949 devid
= devnm2devid(devnm
);
950 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
951 return dev_open(buf
, flags
);
954 int open_dev(char *devnm
)
956 return open_dev_flags(devnm
, O_RDONLY
);
959 int open_dev_excl(char *devnm
)
964 int devid
= devnm2devid(devnm
);
966 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
967 for (i
= 0 ; i
< 25 ; i
++) {
968 int fd
= dev_open(buf
, flags
|O_EXCL
);
971 if (errno
== EACCES
&& flags
== O_RDWR
) {
982 int same_dev(char *one
, char *two
)
984 struct stat st1
, st2
;
985 if (stat(one
, &st1
) != 0)
987 if (stat(two
, &st2
) != 0)
989 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
991 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
993 return st1
.st_rdev
== st2
.st_rdev
;
996 void wait_for(char *dev
, int fd
)
999 struct stat stb_want
;
1001 if (fstat(fd
, &stb_want
) != 0 ||
1002 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1005 for (i
= 0 ; i
< 25 ; i
++) {
1007 if (stat(dev
, &stb
) == 0 &&
1008 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1009 (stb
.st_rdev
== stb_want
.st_rdev
))
1014 dprintf("%s: timeout waiting for %s\n", __func__
, dev
);
1017 struct superswitch
*superlist
[] =
1020 &super_ddf
, &super_imsm
,
1024 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1026 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1028 mdu_array_info_t array
;
1031 struct supertype
*st
= NULL
;
1036 char *subarray
= NULL
;
1037 char container
[32] = "";
1039 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1042 vers
= sra
->array
.major_version
;
1043 minor
= sra
->array
.minor_version
;
1044 verstr
= sra
->text_version
;
1046 if (ioctl(fd
, GET_ARRAY_INFO
, &array
))
1047 array
.major_version
= array
.minor_version
= 0;
1048 vers
= array
.major_version
;
1049 minor
= array
.minor_version
;
1054 sprintf(version
, "%d.%d", vers
, minor
);
1057 if (minor
== -2 && is_subarray(verstr
)) {
1058 char *dev
= verstr
+1;
1060 subarray
= strchr(dev
, '/');
1063 subarray
= xstrdup(subarray
);
1065 strcpy(container
, dev
);
1068 sra
= sysfs_read(-1, container
, GET_VERSION
);
1069 if (sra
&& sra
->text_version
[0])
1070 verstr
= sra
->text_version
;
1072 verstr
= "-no-metadata-";
1075 for (i
= 0; st
== NULL
&& superlist
[i
] ; i
++)
1076 st
= superlist
[i
]->match_metadata_desc(verstr
);
1083 *subarrayp
= subarray
;
1084 strcpy(st
->container_devnm
, container
);
1085 strcpy(st
->devnm
, fd2devnm(fd
));
1091 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1093 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1098 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1099 fd
= dev_open(buf
, O_RDONLY
);
1102 if (get_dev_size(fd
, NULL
, size
)) {
1110 struct supertype
*dup_super(struct supertype
*orig
)
1112 struct supertype
*st
;
1116 st
= xcalloc(1, sizeof(*st
));
1118 st
->max_devs
= orig
->max_devs
;
1119 st
->minor_version
= orig
->minor_version
;
1120 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1126 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1128 /* try each load_super to find the best match,
1129 * and return the best superswitch
1131 struct superswitch
*ss
;
1132 struct supertype
*st
;
1133 time_t besttime
= 0;
1137 st
= xcalloc(1, sizeof(*st
));
1138 st
->container_devnm
[0] = 0;
1140 for (i
= 0 ; superlist
[i
]; i
++) {
1143 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1145 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1147 memset(st
, 0, sizeof(*st
));
1148 st
->ignore_hw_compat
= 1;
1149 rv
= ss
->load_super(st
, fd
, NULL
);
1152 st
->ss
->getinfo_super(st
, &info
, NULL
);
1153 if (bestsuper
== -1 ||
1154 besttime
< info
.array
.ctime
) {
1156 besttime
= info
.array
.ctime
;
1161 if (bestsuper
!= -1) {
1163 memset(st
, 0, sizeof(*st
));
1164 st
->ignore_hw_compat
= 1;
1165 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1167 superlist
[bestsuper
]->free_super(st
);
1175 /* Return size of device in bytes */
1176 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1178 unsigned long long ldsize
;
1181 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1182 ldsize
= (unsigned long long)st
.st_size
;
1185 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1188 unsigned long dsize
;
1189 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1194 pr_err("Cannot get size of %s: %s\b",
1195 dname
, strerror(errno
));
1203 /* Return true if this can only be a container, not a member device.
1204 * i.e. is and md device and size is zero
1206 int must_be_container(int fd
)
1208 unsigned long long size
;
1209 if (md_get_version(fd
) < 0)
1211 if (get_dev_size(fd
, NULL
, &size
) == 0)
1218 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1219 * Returns: 1 if successful
1220 * -1 for unknown partition type
1221 * 0 for other errors
1223 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1226 unsigned char empty_gpt_entry
[16]= {0};
1227 struct GPT_part_entry
*part
;
1229 unsigned long long curr_part_end
;
1230 unsigned all_partitions
, entry_size
;
1235 BUILD_BUG_ON(sizeof(gpt
) != 512);
1236 /* skip protective MBR */
1237 lseek(fd
, 512, SEEK_SET
);
1238 /* read GPT header */
1239 if (read(fd
, &gpt
, 512) != 512)
1242 /* get the number of partition entries and the entry size */
1243 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1244 entry_size
= __le32_to_cpu(gpt
.part_size
);
1246 /* Check GPT signature*/
1247 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1251 if (all_partitions
> 1024 ||
1252 entry_size
> sizeof(buf
))
1255 part
= (struct GPT_part_entry
*)buf
;
1257 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1258 /* read partition entry */
1259 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1262 /* is this valid partition? */
1263 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1264 /* check the last lba for the current partition */
1265 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1266 if (curr_part_end
> *endofpart
)
1267 *endofpart
= curr_part_end
;
1274 /* Sets endofpart parameter to the last block used by the last partition on the device.
1275 * Returns: 1 if successful
1276 * -1 for unknown partition type
1277 * 0 for other errors
1279 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1281 struct MBR boot_sect
;
1282 struct MBR_part_record
*part
;
1283 unsigned long long curr_part_end
;
1289 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1292 if (read(fd
, &boot_sect
, 512) != 512)
1295 /* check MBP signature */
1296 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1298 /* found the correct signature */
1299 part
= boot_sect
.parts
;
1301 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1302 /* check for GPT type */
1303 if (part
->part_type
== MBR_GPT_PARTITION_TYPE
) {
1304 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1307 /* check the last used lba for the current partition */
1308 curr_part_end
= __le32_to_cpu(part
->first_sect_lba
) +
1309 __le32_to_cpu(part
->blocks_num
);
1310 if (curr_part_end
> *endofpart
)
1311 *endofpart
= curr_part_end
;
1316 /* Unknown partition table */
1323 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1324 unsigned long long size
)
1327 * Check where the last partition ends
1329 unsigned long long endofpart
;
1332 if ((ret
= get_last_partition_end(fd
, &endofpart
)) > 0) {
1333 /* There appears to be a partition table here */
1334 if (freesize
== 0) {
1335 /* partitions will not be visible in new device */
1336 pr_err("partition table exists on %s but will be lost or\n"
1337 " meaningless after creating array\n",
1340 } else if (endofpart
> freesize
) {
1341 /* last partition overlaps metadata */
1342 pr_err("metadata will over-write last partition on %s.\n",
1345 } else if (size
&& endofpart
> size
) {
1346 /* partitions will be truncated in new device */
1347 pr_err("array size is too small to cover all partitions on %s.\n",
1355 int open_container(int fd
)
1357 /* 'fd' is a block device. Find out if it is in use
1358 * by a container, and return an open fd on that container.
1369 if (fstat(fd
, &st
) != 0)
1371 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1372 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1373 e
= path
+ strlen(path
);
1375 dir
= opendir(path
);
1378 while ((de
= readdir(dir
))) {
1381 if (de
->d_name
[0] == '.')
1383 /* Need to make sure it is a container and not a volume */
1384 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1385 dfd
= open(path
, O_RDONLY
);
1388 n
= read(dfd
, buf
, sizeof(buf
));
1390 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1393 if (strncmp(buf
, "external", 8) != 0 ||
1397 sprintf(e
, "/%s/dev", de
->d_name
);
1398 dfd
= open(path
, O_RDONLY
);
1401 n
= read(dfd
, buf
, sizeof(buf
));
1403 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1406 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1408 sprintf(buf
, "%d:%d", major
, minor
);
1409 dfd
= dev_open(buf
, O_RDONLY
);
1419 struct superswitch
*version_to_superswitch(char *vers
)
1423 for (i
= 0; superlist
[i
]; i
++) {
1424 struct superswitch
*ss
= superlist
[i
];
1426 if (strcmp(vers
, ss
->name
) == 0)
1433 int metadata_container_matches(char *metadata
, char *devnm
)
1435 /* Check if 'devnm' is the container named in 'metadata'
1437 * /containername/componentname or
1438 * -containername/componentname
1441 if (*metadata
!= '/' && *metadata
!= '-')
1444 if (strncmp(metadata
+1, devnm
, l
) != 0)
1446 if (metadata
[l
+1] != '/')
1451 int metadata_subdev_matches(char *metadata
, char *devnm
)
1453 /* Check if 'devnm' is the subdev named in 'metadata'
1455 * /containername/subdev or
1456 * -containername/subdev
1459 if (*metadata
!= '/' && *metadata
!= '-')
1461 sl
= strchr(metadata
+1, '/');
1464 if (strcmp(sl
+1, devnm
) == 0)
1469 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1471 if (mdstat
->metadata_version
== NULL
||
1472 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1473 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1479 int is_subarray_active(char *subarray
, char *container
)
1481 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1482 struct mdstat_ent
*ent
;
1484 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1485 if (is_container_member(ent
, container
))
1486 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1489 free_mdstat(mdstat
);
1494 /* open_subarray - opens a subarray in a container
1495 * @dev: container device name
1496 * @st: empty supertype
1497 * @quiet: block reporting errors flag
1499 * On success returns an fd to a container and fills in *st
1501 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1504 struct mdinfo
*info
;
1508 fd
= open(dev
, O_RDWR
|O_EXCL
);
1511 pr_err("Couldn't open %s, aborting\n",
1516 _devnm
= fd2devnm(fd
);
1517 if (_devnm
== NULL
) {
1519 pr_err("Failed to determine device number for %s\n",
1523 strcpy(st
->devnm
, _devnm
);
1525 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1528 pr_err("Failed to read sysfs for %s\n",
1533 if (mdi
->array
.level
!= UnSet
) {
1535 pr_err("%s is not a container\n", dev
);
1539 st
->ss
= version_to_superswitch(mdi
->text_version
);
1542 pr_err("Operation not supported for %s metadata\n",
1547 if (st
->devnm
[0] == 0) {
1549 pr_err("Failed to allocate device name\n");
1553 if (!st
->ss
->load_container
) {
1555 pr_err("%s is not a container\n", dev
);
1559 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1561 pr_err("Failed to load metadata for %s\n",
1566 info
= st
->ss
->container_content(st
, subarray
);
1569 pr_err("Failed to find subarray-%s in %s\n",
1579 st
->ss
->free_super(st
);
1592 int add_disk(int mdfd
, struct supertype
*st
,
1593 struct mdinfo
*sra
, struct mdinfo
*info
)
1595 /* Add a device to an array, in one of 2 ways. */
1598 if (st
->ss
->external
) {
1599 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1600 info
->recovery_start
= MaxSector
;
1602 info
->recovery_start
= 0;
1603 rv
= sysfs_add_disk(sra
, info
, 0);
1606 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1610 sd2
= xmalloc(sizeof(*sd2
));
1612 sd2
->next
= sra
->devs
;
1618 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1622 int remove_disk(int mdfd
, struct supertype
*st
,
1623 struct mdinfo
*sra
, struct mdinfo
*info
)
1626 /* Remove the disk given by 'info' from the array */
1628 if (st
->ss
->external
)
1629 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1632 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1637 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1639 /* Initialise kernel's knowledge of array.
1640 * This varies between externally managed arrays
1643 int vers
= md_get_version(mdfd
);
1647 if (st
->ss
->external
)
1648 rv
= sysfs_set_array(info
, vers
);
1651 if ((vers
% 100) >= 1) { /* can use different versions */
1652 mdu_array_info_t inf
;
1653 memset(&inf
, 0, sizeof(inf
));
1654 inf
.major_version
= info
->array
.major_version
;
1655 inf
.minor_version
= info
->array
.minor_version
;
1656 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1658 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1662 unsigned long long min_recovery_start(struct mdinfo
*array
)
1664 /* find the minimum recovery_start in an array for metadata
1665 * formats that only record per-array recovery progress instead
1668 unsigned long long recovery_start
= MaxSector
;
1671 for (d
= array
->devs
; d
; d
= d
->next
)
1672 recovery_start
= min(recovery_start
, d
->recovery_start
);
1674 return recovery_start
;
1677 int mdmon_pid(char *devnm
)
1684 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1686 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1690 n
= read(fd
, pid
, 9);
1697 int mdmon_running(char *devnm
)
1699 int pid
= mdmon_pid(devnm
);
1702 if (kill(pid
, 0) == 0)
1707 int start_mdmon(char *devnm
)
1721 if (check_env("MDADM_NO_MDMON"))
1724 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1728 sl
= strrchr(pathbuf
, '/');
1733 strcpy(sl
, "mdmon");
1737 /* First try to run systemctl */
1740 /* FIXME yuk. CLOSE_EXEC?? */
1742 for (i
= 3; skipped
< 20; i
++)
1748 snprintf(pathbuf
, sizeof(pathbuf
), "mdmon@%s.service",
1750 status
= execl("/usr/bin/systemctl", "systemctl", "start",
1752 status
= execl("/bin/systemctl", "systemctl", "start",
1755 case -1: pr_err("cannot run mdmon. "
1756 "Array remains readonly\n");
1758 default: /* parent - good */
1759 pid
= wait(&status
);
1760 if (pid
>= 0 && status
== 0)
1764 /* That failed, try running mdmon directly */
1767 /* FIXME yuk. CLOSE_EXEC?? */
1769 for (i
= 3; skipped
< 20; i
++)
1775 for (i
= 0; paths
[i
]; i
++)
1777 execl(paths
[i
], "mdmon",
1781 case -1: pr_err("cannot run mdmon. "
1782 "Array remains readonly\n");
1784 default: /* parent - good */
1785 pid
= wait(&status
);
1786 if (pid
< 0 || status
!= 0) {
1787 pr_err("failed to launch mdmon. "
1788 "Array remains readonly\n");
1795 int check_env(char *name
)
1797 char *val
= getenv(name
);
1799 if (val
&& atoi(val
) == 1)
1805 __u32
random32(void)
1808 int rfd
= open("/dev/urandom", O_RDONLY
);
1809 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1817 int flush_metadata_updates(struct supertype
*st
)
1821 st
->update_tail
= NULL
;
1825 sfd
= connect_monitor(st
->container_devnm
);
1829 while (st
->updates
) {
1830 struct metadata_update
*mu
= st
->updates
;
1831 st
->updates
= mu
->next
;
1833 send_message(sfd
, mu
, 0);
1841 st
->update_tail
= NULL
;
1845 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
1848 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
1853 mu
->space_list
= NULL
;
1855 *st
->update_tail
= mu
;
1856 st
->update_tail
= &mu
->next
;
1858 #endif /* MDASSEMBLE */
1861 /* tinyc doesn't optimize this check in ioctl.h out ... */
1862 unsigned int __invalid_size_argument_for_IOC
= 0;
1865 int experimental(void)
1867 if (check_env("MDADM_EXPERIMENTAL"))
1870 pr_err("To use this feature MDADM_EXPERIMENTAL"
1871 " environment variable has to be defined.\n");
1876 /* Pick all spares matching given criteria from a container
1877 * if min_size == 0 do not check size
1878 * if domlist == NULL do not check domains
1879 * if spare_group given add it to domains of each spare
1880 * metadata allows to test domains using metadata of destination array */
1881 struct mdinfo
*container_choose_spares(struct supertype
*st
,
1882 unsigned long long min_size
,
1883 struct domainlist
*domlist
,
1885 const char *metadata
, int get_one
)
1887 struct mdinfo
*d
, **dp
, *disks
= NULL
;
1889 /* get list of all disks in container */
1890 if (st
->ss
->getinfo_super_disks
)
1891 disks
= st
->ss
->getinfo_super_disks(st
);
1895 /* find spare devices on the list */
1897 disks
->array
.spare_disks
= 0;
1901 if (d
->disk
.state
== 0) {
1902 /* check if size is acceptable */
1903 unsigned long long dev_size
;
1904 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
1907 (dev_size_from_id(dev
, &dev_size
) &&
1908 dev_size
>= min_size
))
1910 /* check if domain matches */
1911 if (found
&& domlist
) {
1912 struct dev_policy
*pol
= devid_policy(dev
);
1914 pol_add(&pol
, pol_domain
,
1916 if (domain_test(domlist
, pol
, metadata
) != 1)
1918 dev_policy_free(pol
);
1923 disks
->array
.spare_disks
++;
1937 /* Checks if paths point to the same device
1938 * Returns 0 if they do.
1939 * Returns 1 if they don't.
1940 * Returns -1 if something went wrong,
1941 * e.g. paths are empty or the files
1942 * they point to don't exist */
1943 int compare_paths (char* path1
, char* path2
)
1945 struct stat st1
,st2
;
1947 if (path1
== NULL
|| path2
== NULL
)
1949 if (stat(path1
,&st1
) != 0)
1951 if (stat(path2
,&st2
) != 0)
1953 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))