2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 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>
31 #include <sys/resource.h>
33 #include <linux/magic.h>
39 * following taken from linux/blkpg.h because they aren't
40 * anywhere else and it isn't safe to #include linux/ * stuff.
43 #define BLKPG _IO(0x12,105)
45 /* The argument structure */
46 struct blkpg_ioctl_arg
{
53 /* The subfunctions (for the op field) */
54 #define BLKPG_ADD_PARTITION 1
55 #define BLKPG_DEL_PARTITION 2
57 /* Sizes of name fields. Unused at present. */
58 #define BLKPG_DEVNAMELTH 64
59 #define BLKPG_VOLNAMELTH 64
61 /* The data structure for ADD_PARTITION and DEL_PARTITION */
62 struct blkpg_partition
{
63 long long start
; /* starting offset in bytes */
64 long long length
; /* length in bytes */
65 int pno
; /* partition number */
66 char devname
[BLKPG_DEVNAMELTH
]; /* partition name, like sda5 or c0d1p2,
67 to be used in kernel messages */
68 char volname
[BLKPG_VOLNAMELTH
]; /* volume label */
73 /* Force a compilation error if condition is true */
74 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
76 /* Force a compilation error if condition is true, but also produce a
77 result (of value 0 and type size_t), so the expression can be used
78 e.g. in a structure initializer (or where-ever else comma expressions
80 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
83 * Parse a 128 bit uuid in 4 integers
84 * format is 32 hexx nibbles with options :.<space> separator
85 * If not exactly 32 hex digits are found, return 0
88 int parse_uuid(char *str
, int uuid
[4])
90 int hit
= 0; /* number of Hex digIT */
93 for (i
= 0; i
< 4; i
++)
96 while ((c
= *str
++) != 0) {
98 if (c
>= '0' && c
<= '9')
100 else if (c
>= 'a' && c
<= 'f')
102 else if (c
>= 'A' && c
<= 'F')
104 else if (strchr(":. -", c
))
120 * Get the md version number.
121 * We use the RAID_VERSION ioctl if it is supported
122 * If not, but we have a block device with major '9', we assume
125 * Return version number as 24 but number - assume version parts
129 int md_get_version(int fd
)
134 if (fstat(fd
, &stb
)<0)
136 if ((S_IFMT
&stb
.st_mode
) != S_IFBLK
)
139 if (ioctl(fd
, RAID_VERSION
, &vers
) == 0)
140 return (vers
.major
*10000) + (vers
.minor
*100) + vers
.patchlevel
;
143 if (major(stb
.st_rdev
) == MD_MAJOR
)
148 int get_linux_version()
152 int a
= 0, b
= 0,c
= 0;
157 a
= strtoul(cp
, &cp
, 10);
159 b
= strtoul(cp
+1, &cp
, 10);
161 c
= strtoul(cp
+1, &cp
, 10);
163 return (a
*1000000)+(b
*1000)+c
;
167 int mdadm_version(char *version
)
175 cp
= strchr(version
, '-');
176 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
179 a
= strtoul(cp
, &cp
, 10);
182 b
= strtoul(cp
+1, &cp
, 10);
184 c
= strtoul(cp
+1, &cp
, 10);
187 if (*cp
!= ' ' && *cp
!= '-')
189 return (a
*1000000)+(b
*1000)+c
;
192 unsigned long long parse_size(char *size
)
194 /* parse 'size' which should be a number optionally
195 * followed by 'K', 'M', or 'G'.
196 * Without a suffix, K is assumed.
197 * Number returned is in sectors (half-K)
198 * INVALID_SECTORS returned on error.
201 long long s
= strtoll(size
, &c
, 10);
215 s
*= 1024 * 1024 * 2;
217 case 's': /* sectors */
228 int parse_layout_10(char *layout
)
232 /* Parse the layout string for raid10 */
233 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
234 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
235 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
239 if (layout
[0] == 'n')
241 else if (layout
[0] == 'o')
242 rv
= 0x10000 + (copies
<<8) + 1;
244 rv
= 1 + (copies
<<8);
248 int parse_layout_faulty(char *layout
)
250 /* Parse the layout string for 'faulty' */
251 int ln
= strcspn(layout
, "0123456789");
252 char *m
= xstrdup(layout
);
255 mode
= map_name(faultylayout
, m
);
259 return mode
| (atoi(layout
+ln
)<< ModeShift
);
262 long parse_num(char *num
)
264 /* Either return a valid number, or -1 */
266 long rv
= strtol(num
, &c
, 10);
267 if (rv
< 0 || *c
|| !num
[0])
274 void remove_partitions(int fd
)
276 /* remove partitions from this block devices.
277 * This is used for components added to an array
279 #ifdef BLKPG_DEL_PARTITION
280 struct blkpg_ioctl_arg a
;
281 struct blkpg_partition p
;
283 a
.op
= BLKPG_DEL_PARTITION
;
285 a
.datalen
= sizeof(p
);
287 memset(a
.data
, 0, a
.datalen
);
288 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
289 ioctl(fd
, BLKPG
, &a
);
293 int test_partition(int fd
)
295 /* Check if fd is a whole-disk or a partition.
296 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
297 * will return ENXIO on an invalid partition number.
299 struct blkpg_ioctl_arg a
;
300 struct blkpg_partition p
;
301 a
.op
= BLKPG_DEL_PARTITION
;
303 a
.datalen
= sizeof(p
);
305 memset(a
.data
, 0, a
.datalen
);
307 if (ioctl(fd
, BLKPG
, &a
) == 0)
308 /* Very unlikely, but not a partition */
310 if (errno
== ENXIO
|| errno
== ENOTTY
)
311 /* not a partition */
317 int test_partition_from_id(dev_t id
)
322 sprintf(buf
, "%d:%d", major(id
), minor(id
));
323 fd
= dev_open(buf
, O_RDONLY
);
326 rv
= test_partition(fd
);
331 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
337 for (i
= 0; i
< raid_disks
; i
++)
338 avail_disks
+= !!avail
[i
];
342 /* This is the tricky one - we need to check
343 * which actual disks are present.
345 copies
= (layout
&255)* ((layout
>>8) & 255);
348 /* there must be one of the 'copies' form 'first' */
355 this = (this+1) % raid_disks
;
359 first
= (first
+(layout
&255)) % raid_disks
;
360 } while (first
!= 0);
363 case LEVEL_MULTIPATH
:
364 return avail_disks
>= 1;
367 return avail_disks
== raid_disks
;
369 return avail_disks
>= 1;
373 return avail_disks
>= raid_disks
-1;
375 return avail_disks
>= raid_disks
;
378 return avail_disks
>= raid_disks
-2;
380 return avail_disks
>= raid_disks
;
386 int enough_fd(int fd
)
388 struct mdu_array_info_s array
;
389 struct mdu_disk_info_s disk
;
393 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0 ||
394 array
.raid_disks
<= 0)
396 avail
= xcalloc(array
.raid_disks
, 1);
397 for (i
= 0; i
< MAX_DISKS
&& array
.nr_disks
> 0; i
++) {
399 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
401 if (disk
.major
== 0 && disk
.minor
== 0)
405 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
407 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
409 avail
[disk
.raid_disk
] = 1;
411 /* This is used on an active array, so assume it is clean */
412 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
418 const int uuid_zero
[4] = { 0, 0, 0, 0 };
420 int same_uuid(int a
[4], int b
[4], int swapuuid
)
423 /* parse uuids are hostendian.
424 * uuid's from some superblocks are big-ending
425 * if there is a difference, we need to swap..
427 unsigned char *ac
= (unsigned char *)a
;
428 unsigned char *bc
= (unsigned char *)b
;
430 for (i
= 0; i
< 16; i
+= 4) {
431 if (ac
[i
+0] != bc
[i
+3] ||
432 ac
[i
+1] != bc
[i
+2] ||
433 ac
[i
+2] != bc
[i
+1] ||
448 void copy_uuid(void *a
, int b
[4], int swapuuid
)
451 /* parse uuids are hostendian.
452 * uuid's from some superblocks are big-ending
453 * if there is a difference, we need to swap..
455 unsigned char *ac
= (unsigned char *)a
;
456 unsigned char *bc
= (unsigned char *)b
;
458 for (i
= 0; i
< 16; i
+= 4) {
468 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
475 copy_uuid(uuid
, id
, swap
);
476 for (i
= 0; i
< 4; i
++) {
479 for (j
= 3; j
>= 0; j
--) {
480 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
488 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
490 // dirty hack to work around an issue with super1 superblocks...
491 // super1 superblocks need swapuuid set in order for assembly to
492 // work, but can't have it set if we want this printout to match
493 // all the other uuid printouts in super1.c, so we force swapuuid
494 // to 1 to make our printout match the rest of super1
495 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
499 int check_ext2(int fd
, char *name
)
502 * Check for an ext2fs file system.
503 * Superblock is always 1K at 1K offset
505 * s_magic is le16 at 56 == 0xEF53
506 * report mtime - le32 at 44
508 * logblksize - le32 at 24
510 unsigned char sb
[1024];
512 unsigned long long size
;
514 if (lseek(fd
, 1024,0)!= 1024)
516 if (read(fd
, sb
, 1024)!= 1024)
518 if (sb
[56] != 0x53 || sb
[57] != 0xef)
521 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
522 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
523 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
525 pr_err("%s appears to contain an ext2fs file system\n",
527 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
531 int check_reiser(int fd
, char *name
)
534 * superblock is at 64K
536 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
539 unsigned char sb
[1024];
540 unsigned long long size
;
541 if (lseek(fd
, 64*1024, 0) != 64*1024)
543 if (read(fd
, sb
, 1024) != 1024)
545 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
546 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
548 pr_err("%s appears to contain a reiserfs file system\n",name
);
549 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
550 cont_err("size = %lluK\n", size
*4);
555 int check_raid(int fd
, char *name
)
560 struct supertype
*st
= guess_super(fd
);
564 st
->ss
->load_super(st
, fd
, name
);
565 /* Looks like a raid array .. */
566 pr_err("%s appears to be part of a raid array:\n",
568 st
->ss
->getinfo_super(st
, &info
, NULL
);
569 st
->ss
->free_super(st
);
570 crtime
= info
.array
.ctime
;
571 level
= map_num(pers
, info
.array
.level
);
572 if (!level
) level
= "-unknown-";
573 cont_err("level=%s devices=%d ctime=%s",
574 level
, info
.array
.raid_disks
, ctime(&crtime
));
582 for (i
= 0; i
< 5; i
++) {
584 fprintf(stderr
, "%s%s", mesg
, add
);
586 if (fgets(buf
, 100, stdin
)==NULL
)
588 if (buf
[0]=='y' || buf
[0]=='Y')
590 if (buf
[0]=='n' || buf
[0]=='N')
594 pr_err("assuming 'no'\n");
597 #endif /* MDASSEMBLE */
599 int is_standard(char *dev
, int *nump
)
601 /* tests if dev is a "standard" md dev name.
602 * i.e if the last component is "/dNN" or "/mdNN",
603 * where NN is a string of digits
604 * Returns 1 if a partitionable standard,
605 * -1 if non-partitonable,
606 * 0 if not a standard name.
608 char *d
= strrchr(dev
, '/');
613 if (strncmp(d
, "/d",2) == 0)
614 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
615 else if (strncmp(d
, "/md_d", 5) == 0)
616 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
617 else if (strncmp(d
, "/md", 3) == 0)
618 d
+= 3, type
= -1; /* /dev/mdN */
619 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
620 d
+= 1, type
= -1; /* /dev/md/N */
630 if (nump
) *nump
= num
;
635 unsigned long calc_csum(void *super
, int bytes
)
637 unsigned long long newcsum
= 0;
640 unsigned int *superc
= (unsigned int*) super
;
642 for(i
= 0; i
< bytes
/4; i
++)
643 newcsum
+= superc
[i
];
644 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
646 /* The in-kernel checksum calculation is always 16bit on
647 * the alpha, though it is 32 bit on i386...
648 * I wonder what it is elsewhere... (it uses an API in
649 * a way that it shouldn't).
651 csum
= (csum
& 0xffff) + (csum
>> 16);
652 csum
= (csum
& 0xffff) + (csum
>> 16);
658 char *human_size(long long bytes
)
662 /* We convert bytes to either centi-M{ega,ibi}bytes or
663 * centi-G{igi,ibi}bytes, with appropriate rounding,
664 * and then print 1/100th of those as a decimal.
665 * We allow upto 2048Megabytes before converting to
666 * gigabytes, as that shows more precision and isn't
667 * too large a number.
668 * Terabytes are not yet handled.
671 if (bytes
< 5000*1024)
673 else if (bytes
< 2*1024LL*1024LL*1024LL) {
674 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
675 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
676 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
677 cMiB
/100 , cMiB
% 100,
680 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
681 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
682 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
683 cGiB
/100 , cGiB
% 100,
689 char *human_size_brief(long long bytes
, int prefix
)
693 /* We convert bytes to either centi-M{ega,ibi}bytes or
694 * centi-G{igi,ibi}bytes, with appropriate rounding,
695 * and then print 1/100th of those as a decimal.
696 * We allow upto 2048Megabytes before converting to
697 * gigabytes, as that shows more precision and isn't
698 * too large a number.
699 * Terabytes are not yet handled.
701 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
702 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
705 if (bytes
< 5000*1024)
707 else if (prefix
== IEC
) {
708 if (bytes
< 2*1024LL*1024LL*1024LL) {
709 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
710 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
711 cMiB
/100 , cMiB
% 100);
713 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
714 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
715 cGiB
/100 , cGiB
% 100);
718 else if (prefix
== JEDEC
) {
719 if (bytes
< 2*1024LL*1024LL*1024LL) {
720 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
721 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
724 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
725 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
726 cGB
/100 , cGB
% 100);
735 void print_r10_layout(int layout
)
737 int near
= layout
& 255;
738 int far
= (layout
>> 8) & 255;
739 int offset
= (layout
&0x10000);
743 printf("%s near=%d", sep
, near
);
747 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
749 printf("NO REDUNDANCY");
753 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
754 int chunksize
, unsigned long long devsize
)
758 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
759 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
762 int get_data_disks(int level
, int layout
, int raid_disks
)
766 case 0: data_disks
= raid_disks
;
768 case 1: data_disks
= 1;
771 case 5: data_disks
= raid_disks
- 1;
773 case 6: data_disks
= raid_disks
- 2;
775 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
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 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
819 char *get_md_name(char *devnm
)
821 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
822 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
824 static char devname
[50];
826 dev_t rdev
= devnm2devid(devnm
);
831 if (strncmp(devnm
, "md_", 3) == 0) {
832 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
834 if (stat(devname
, &stb
) == 0
835 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
836 && (stb
.st_rdev
== rdev
))
839 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
840 if (stat(devname
, &stb
) == 0
841 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
842 && (stb
.st_rdev
== rdev
))
845 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
846 if (stat(devname
, &stb
) == 0
847 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
848 && (stb
.st_rdev
== rdev
))
851 dn
= map_dev(major(rdev
), minor(rdev
), 0);
854 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
855 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
859 if (stat(devname
, &stb
) == 0
860 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
861 && (stb
.st_rdev
== rdev
))
867 void put_md_name(char *name
)
869 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
872 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
874 int get_maj_min(char *dev
, int *major
, int *minor
)
877 *major
= strtoul(dev
, &e
, 0);
878 return (e
> dev
&& *e
== ':' && e
[1] &&
879 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
883 int dev_open(char *dev
, int flags
)
885 /* like 'open', but if 'dev' matches %d:%d, create a temp
886 * block device and open that
896 if (get_maj_min(dev
, &major
, &minor
)) {
897 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
898 (int)getpid(), major
, minor
);
899 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
900 fd
= open(devname
, flags
);
904 /* Try /tmp as /dev appear to be read-only */
905 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
906 (int)getpid(), major
, minor
);
907 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
908 fd
= open(devname
, flags
);
913 fd
= open(dev
, flags
);
917 int open_dev_flags(char *devnm
, int flags
)
922 devid
= devnm2devid(devnm
);
923 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
924 return dev_open(buf
, flags
);
927 int open_dev(char *devnm
)
929 return open_dev_flags(devnm
, O_RDONLY
);
932 int open_dev_excl(char *devnm
)
937 int devid
= devnm2devid(devnm
);
940 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
941 for (i
= 0 ; i
< 25 ; i
++) {
942 int fd
= dev_open(buf
, flags
|O_EXCL
);
945 if (errno
== EACCES
&& flags
== O_RDWR
) {
958 int same_dev(char *one
, char *two
)
960 struct stat st1
, st2
;
961 if (stat(one
, &st1
) != 0)
963 if (stat(two
, &st2
) != 0)
965 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
967 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
969 return st1
.st_rdev
== st2
.st_rdev
;
972 void wait_for(char *dev
, int fd
)
975 struct stat stb_want
;
978 if (fstat(fd
, &stb_want
) != 0 ||
979 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
982 for (i
= 0 ; i
< 25 ; i
++) {
984 if (stat(dev
, &stb
) == 0 &&
985 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
986 (stb
.st_rdev
== stb_want
.st_rdev
))
993 dprintf("timeout waiting for %s\n", dev
);
996 struct superswitch
*superlist
[] =
999 &super_ddf
, &super_imsm
,
1003 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1005 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1007 mdu_array_info_t array
;
1010 struct supertype
*st
= NULL
;
1015 char *subarray
= NULL
;
1016 char container
[32] = "";
1018 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1021 vers
= sra
->array
.major_version
;
1022 minor
= sra
->array
.minor_version
;
1023 verstr
= sra
->text_version
;
1025 if (ioctl(fd
, GET_ARRAY_INFO
, &array
))
1026 array
.major_version
= array
.minor_version
= 0;
1027 vers
= array
.major_version
;
1028 minor
= array
.minor_version
;
1033 sprintf(version
, "%d.%d", vers
, minor
);
1036 if (minor
== -2 && is_subarray(verstr
)) {
1037 char *dev
= verstr
+1;
1039 subarray
= strchr(dev
, '/');
1042 subarray
= xstrdup(subarray
);
1044 strcpy(container
, dev
);
1047 sra
= sysfs_read(-1, container
, GET_VERSION
);
1048 if (sra
&& sra
->text_version
[0])
1049 verstr
= sra
->text_version
;
1051 verstr
= "-no-metadata-";
1054 for (i
= 0; st
== NULL
&& superlist
[i
] ; i
++)
1055 st
= superlist
[i
]->match_metadata_desc(verstr
);
1062 *subarrayp
= subarray
;
1063 strcpy(st
->container_devnm
, container
);
1064 strcpy(st
->devnm
, fd2devnm(fd
));
1070 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1072 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1077 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1078 fd
= dev_open(buf
, O_RDONLY
);
1081 if (get_dev_size(fd
, NULL
, size
)) {
1089 struct supertype
*dup_super(struct supertype
*orig
)
1091 struct supertype
*st
;
1095 st
= xcalloc(1, sizeof(*st
));
1097 st
->max_devs
= orig
->max_devs
;
1098 st
->minor_version
= orig
->minor_version
;
1099 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1100 st
->data_offset
= orig
->data_offset
;
1106 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1108 /* try each load_super to find the best match,
1109 * and return the best superswitch
1111 struct superswitch
*ss
;
1112 struct supertype
*st
;
1113 time_t besttime
= 0;
1117 st
= xcalloc(1, sizeof(*st
));
1118 st
->container_devnm
[0] = 0;
1120 for (i
= 0 ; superlist
[i
]; i
++) {
1123 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1125 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1127 memset(st
, 0, sizeof(*st
));
1128 st
->ignore_hw_compat
= 1;
1129 rv
= ss
->load_super(st
, fd
, NULL
);
1132 st
->ss
->getinfo_super(st
, &info
, NULL
);
1133 if (bestsuper
== -1 ||
1134 besttime
< info
.array
.ctime
) {
1136 besttime
= info
.array
.ctime
;
1141 if (bestsuper
!= -1) {
1143 memset(st
, 0, sizeof(*st
));
1144 st
->ignore_hw_compat
= 1;
1145 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1147 superlist
[bestsuper
]->free_super(st
);
1155 /* Return size of device in bytes */
1156 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1158 unsigned long long ldsize
;
1161 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1162 ldsize
= (unsigned long long)st
.st_size
;
1165 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1168 unsigned long dsize
;
1169 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1174 pr_err("Cannot get size of %s: %s\b",
1175 dname
, strerror(errno
));
1183 /* Return true if this can only be a container, not a member device.
1184 * i.e. is and md device and size is zero
1186 int must_be_container(int fd
)
1188 unsigned long long size
;
1189 if (md_get_version(fd
) < 0)
1191 if (get_dev_size(fd
, NULL
, &size
) == 0)
1198 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1199 * Returns: 1 if successful
1200 * -1 for unknown partition type
1201 * 0 for other errors
1203 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1206 unsigned char empty_gpt_entry
[16]= {0};
1207 struct GPT_part_entry
*part
;
1209 unsigned long long curr_part_end
;
1210 unsigned all_partitions
, entry_size
;
1215 BUILD_BUG_ON(sizeof(gpt
) != 512);
1216 /* skip protective MBR */
1217 lseek(fd
, 512, SEEK_SET
);
1218 /* read GPT header */
1219 if (read(fd
, &gpt
, 512) != 512)
1222 /* get the number of partition entries and the entry size */
1223 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1224 entry_size
= __le32_to_cpu(gpt
.part_size
);
1226 /* Check GPT signature*/
1227 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1231 if (all_partitions
> 1024 ||
1232 entry_size
> sizeof(buf
))
1235 part
= (struct GPT_part_entry
*)buf
;
1237 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1238 /* read partition entry */
1239 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1242 /* is this valid partition? */
1243 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1244 /* check the last lba for the current partition */
1245 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1246 if (curr_part_end
> *endofpart
)
1247 *endofpart
= curr_part_end
;
1254 /* Sets endofpart parameter to the last block used by the last partition on the device.
1255 * Returns: 1 if successful
1256 * -1 for unknown partition type
1257 * 0 for other errors
1259 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1261 struct MBR boot_sect
;
1262 struct MBR_part_record
*part
;
1263 unsigned long long curr_part_end
;
1269 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1272 if (read(fd
, &boot_sect
, 512) != 512)
1275 /* check MBP signature */
1276 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1278 /* found the correct signature */
1279 part
= boot_sect
.parts
;
1281 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1282 /* check for GPT type */
1283 if (part
->part_type
== MBR_GPT_PARTITION_TYPE
) {
1284 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1287 /* check the last used lba for the current partition */
1288 curr_part_end
= __le32_to_cpu(part
->first_sect_lba
) +
1289 __le32_to_cpu(part
->blocks_num
);
1290 if (curr_part_end
> *endofpart
)
1291 *endofpart
= curr_part_end
;
1296 /* Unknown partition table */
1303 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1304 unsigned long long size
)
1307 * Check where the last partition ends
1309 unsigned long long endofpart
;
1312 if ((ret
= get_last_partition_end(fd
, &endofpart
)) > 0) {
1313 /* There appears to be a partition table here */
1314 if (freesize
== 0) {
1315 /* partitions will not be visible in new device */
1316 pr_err("partition table exists on %s but will be lost or\n"
1317 " meaningless after creating array\n",
1320 } else if (endofpart
> freesize
) {
1321 /* last partition overlaps metadata */
1322 pr_err("metadata will over-write last partition on %s.\n",
1325 } else if (size
&& endofpart
> size
) {
1326 /* partitions will be truncated in new device */
1327 pr_err("array size is too small to cover all partitions on %s.\n",
1335 int open_container(int fd
)
1337 /* 'fd' is a block device. Find out if it is in use
1338 * by a container, and return an open fd on that container.
1349 if (fstat(fd
, &st
) != 0)
1351 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1352 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1353 e
= path
+ strlen(path
);
1355 dir
= opendir(path
);
1358 while ((de
= readdir(dir
))) {
1361 if (de
->d_name
[0] == '.')
1363 /* Need to make sure it is a container and not a volume */
1364 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1365 dfd
= open(path
, O_RDONLY
);
1368 n
= read(dfd
, buf
, sizeof(buf
));
1370 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1373 if (strncmp(buf
, "external", 8) != 0 ||
1377 sprintf(e
, "/%s/dev", de
->d_name
);
1378 dfd
= open(path
, O_RDONLY
);
1381 n
= read(dfd
, buf
, sizeof(buf
));
1383 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1386 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1388 sprintf(buf
, "%d:%d", major
, minor
);
1389 dfd
= dev_open(buf
, O_RDONLY
);
1399 struct superswitch
*version_to_superswitch(char *vers
)
1403 for (i
= 0; superlist
[i
]; i
++) {
1404 struct superswitch
*ss
= superlist
[i
];
1406 if (strcmp(vers
, ss
->name
) == 0)
1413 int metadata_container_matches(char *metadata
, char *devnm
)
1415 /* Check if 'devnm' is the container named in 'metadata'
1417 * /containername/componentname or
1418 * -containername/componentname
1421 if (*metadata
!= '/' && *metadata
!= '-')
1424 if (strncmp(metadata
+1, devnm
, l
) != 0)
1426 if (metadata
[l
+1] != '/')
1431 int metadata_subdev_matches(char *metadata
, char *devnm
)
1433 /* Check if 'devnm' is the subdev named in 'metadata'
1435 * /containername/subdev or
1436 * -containername/subdev
1439 if (*metadata
!= '/' && *metadata
!= '-')
1441 sl
= strchr(metadata
+1, '/');
1444 if (strcmp(sl
+1, devnm
) == 0)
1449 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1451 if (mdstat
->metadata_version
== NULL
||
1452 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1453 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1459 int is_subarray_active(char *subarray
, char *container
)
1461 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1462 struct mdstat_ent
*ent
;
1464 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1465 if (is_container_member(ent
, container
))
1466 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1469 free_mdstat(mdstat
);
1474 /* open_subarray - opens a subarray in a container
1475 * @dev: container device name
1476 * @st: empty supertype
1477 * @quiet: block reporting errors flag
1479 * On success returns an fd to a container and fills in *st
1481 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1484 struct mdinfo
*info
;
1488 fd
= open(dev
, O_RDWR
|O_EXCL
);
1491 pr_err("Couldn't open %s, aborting\n",
1496 _devnm
= fd2devnm(fd
);
1497 if (_devnm
== NULL
) {
1499 pr_err("Failed to determine device number for %s\n",
1503 strcpy(st
->devnm
, _devnm
);
1505 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1508 pr_err("Failed to read sysfs for %s\n",
1513 if (mdi
->array
.level
!= UnSet
) {
1515 pr_err("%s is not a container\n", dev
);
1519 st
->ss
= version_to_superswitch(mdi
->text_version
);
1522 pr_err("Operation not supported for %s metadata\n",
1527 if (st
->devnm
[0] == 0) {
1529 pr_err("Failed to allocate device name\n");
1533 if (!st
->ss
->load_container
) {
1535 pr_err("%s is not a container\n", dev
);
1539 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1541 pr_err("Failed to load metadata for %s\n",
1546 info
= st
->ss
->container_content(st
, subarray
);
1549 pr_err("Failed to find subarray-%s in %s\n",
1559 st
->ss
->free_super(st
);
1572 int add_disk(int mdfd
, struct supertype
*st
,
1573 struct mdinfo
*sra
, struct mdinfo
*info
)
1575 /* Add a device to an array, in one of 2 ways. */
1578 if (st
->ss
->external
) {
1579 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1580 info
->recovery_start
= MaxSector
;
1582 info
->recovery_start
= 0;
1583 rv
= sysfs_add_disk(sra
, info
, 0);
1586 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1590 sd2
= xmalloc(sizeof(*sd2
));
1592 sd2
->next
= sra
->devs
;
1598 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1602 int remove_disk(int mdfd
, struct supertype
*st
,
1603 struct mdinfo
*sra
, struct mdinfo
*info
)
1606 /* Remove the disk given by 'info' from the array */
1608 if (st
->ss
->external
)
1609 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1612 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1617 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1619 /* Initialise kernel's knowledge of array.
1620 * This varies between externally managed arrays
1623 int vers
= md_get_version(mdfd
);
1627 if (st
->ss
->external
)
1628 rv
= sysfs_set_array(info
, vers
);
1631 if ((vers
% 100) >= 1) { /* can use different versions */
1632 mdu_array_info_t inf
;
1633 memset(&inf
, 0, sizeof(inf
));
1634 inf
.major_version
= info
->array
.major_version
;
1635 inf
.minor_version
= info
->array
.minor_version
;
1636 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1638 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1642 unsigned long long min_recovery_start(struct mdinfo
*array
)
1644 /* find the minimum recovery_start in an array for metadata
1645 * formats that only record per-array recovery progress instead
1648 unsigned long long recovery_start
= MaxSector
;
1651 for (d
= array
->devs
; d
; d
= d
->next
)
1652 recovery_start
= min(recovery_start
, d
->recovery_start
);
1654 return recovery_start
;
1657 int mdmon_pid(char *devnm
)
1664 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1666 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1670 n
= read(fd
, pid
, 9);
1677 int mdmon_running(char *devnm
)
1679 int pid
= mdmon_pid(devnm
);
1682 if (kill(pid
, 0) == 0)
1687 int start_mdmon(char *devnm
)
1701 if (check_env("MDADM_NO_MDMON"))
1704 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1708 sl
= strrchr(pathbuf
, '/');
1713 strcpy(sl
, "mdmon");
1717 /* First try to run systemctl */
1718 if (!check_env("MDADM_NO_SYSTEMCTL"))
1721 /* FIXME yuk. CLOSE_EXEC?? */
1723 for (i
= 3; skipped
< 20; i
++)
1729 /* Don't want to see error messages from
1730 * systemctl. If the service doesn't exist,
1731 * we start mdmon ourselves.
1734 open("/dev/null", O_WRONLY
);
1735 snprintf(pathbuf
, sizeof(pathbuf
), "mdmon@%s.service",
1737 status
= execl("/usr/bin/systemctl", "systemctl",
1740 status
= execl("/bin/systemctl", "systemctl", "start",
1743 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1745 default: /* parent - good */
1746 pid
= wait(&status
);
1747 if (pid
>= 0 && status
== 0)
1751 /* That failed, try running mdmon directly */
1754 /* FIXME yuk. CLOSE_EXEC?? */
1756 for (i
= 3; skipped
< 20; i
++)
1762 for (i
= 0; paths
[i
]; i
++)
1764 execl(paths
[i
], paths
[i
],
1768 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1770 default: /* parent - good */
1771 pid
= wait(&status
);
1772 if (pid
< 0 || status
!= 0) {
1773 pr_err("failed to launch mdmon. Array remains readonly\n");
1780 __u32
random32(void)
1783 int rfd
= open("/dev/urandom", O_RDONLY
);
1784 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1792 int flush_metadata_updates(struct supertype
*st
)
1796 st
->update_tail
= NULL
;
1800 sfd
= connect_monitor(st
->container_devnm
);
1804 while (st
->updates
) {
1805 struct metadata_update
*mu
= st
->updates
;
1806 st
->updates
= mu
->next
;
1808 send_message(sfd
, mu
, 0);
1816 st
->update_tail
= NULL
;
1820 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
1823 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
1828 mu
->space_list
= NULL
;
1830 *st
->update_tail
= mu
;
1831 st
->update_tail
= &mu
->next
;
1833 #endif /* MDASSEMBLE */
1836 /* tinyc doesn't optimize this check in ioctl.h out ... */
1837 unsigned int __invalid_size_argument_for_IOC
= 0;
1840 int experimental(void)
1842 if (check_env("MDADM_EXPERIMENTAL"))
1845 pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n");
1850 /* Pick all spares matching given criteria from a container
1851 * if min_size == 0 do not check size
1852 * if domlist == NULL do not check domains
1853 * if spare_group given add it to domains of each spare
1854 * metadata allows to test domains using metadata of destination array */
1855 struct mdinfo
*container_choose_spares(struct supertype
*st
,
1856 unsigned long long min_size
,
1857 struct domainlist
*domlist
,
1859 const char *metadata
, int get_one
)
1861 struct mdinfo
*d
, **dp
, *disks
= NULL
;
1863 /* get list of all disks in container */
1864 if (st
->ss
->getinfo_super_disks
)
1865 disks
= st
->ss
->getinfo_super_disks(st
);
1869 /* find spare devices on the list */
1871 disks
->array
.spare_disks
= 0;
1875 if (d
->disk
.state
== 0) {
1876 /* check if size is acceptable */
1877 unsigned long long dev_size
;
1878 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
1881 (dev_size_from_id(dev
, &dev_size
) &&
1882 dev_size
>= min_size
))
1884 /* check if domain matches */
1885 if (found
&& domlist
) {
1886 struct dev_policy
*pol
= devid_policy(dev
);
1888 pol_add(&pol
, pol_domain
,
1890 if (domain_test(domlist
, pol
, metadata
) != 1)
1892 dev_policy_free(pol
);
1897 disks
->array
.spare_disks
++;
1911 /* Checks if paths point to the same device
1912 * Returns 0 if they do.
1913 * Returns 1 if they don't.
1914 * Returns -1 if something went wrong,
1915 * e.g. paths are empty or the files
1916 * they point to don't exist */
1917 int compare_paths (char* path1
, char* path2
)
1919 struct stat st1
,st2
;
1921 if (path1
== NULL
|| path2
== NULL
)
1923 if (stat(path1
,&st1
) != 0)
1925 if (stat(path2
,&st2
) != 0)
1927 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
1932 /* Make sure we can open as many devices as needed */
1933 void enable_fds(int devices
)
1935 unsigned int fds
= 20 + devices
;
1937 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0
1938 || lim
.rlim_cur
>= fds
)
1940 if (lim
.rlim_max
< fds
)
1943 setrlimit(RLIMIT_NOFILE
, &lim
);
1948 /* This is based on similar function in systemd. */
1950 /* statfs.f_type is signed long on s390x and MIPS, causing all
1951 sorts of sign extension problems with RAMFS_MAGIC being
1952 defined as 0x858458f6 */
1953 return statfs("/", &s
) >= 0 &&
1954 ((unsigned long)s
.f_type
== TMPFS_MAGIC
||
1955 ((unsigned long)s
.f_type
& 0xFFFFFFFFUL
) ==
1956 ((unsigned long)RAMFS_MAGIC
& 0xFFFFFFFFUL
));
1959 void reopen_mddev(int mdfd
)
1961 /* Re-open without any O_EXCL, but keep
1966 devnm
= fd2devnm(mdfd
);
1968 fd
= open_dev(devnm
);
1969 if (fd
>= 0 && fd
!= mdfd
)