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>
28 #include <sys/socket.h>
29 #include <sys/utsname.h>
32 #include <sys/resource.h>
34 #include <linux/magic.h>
42 * following taken from linux/blkpg.h because they aren't
43 * anywhere else and it isn't safe to #include linux/ * stuff.
46 #define BLKPG _IO(0x12,105)
48 /* The argument structure */
49 struct blkpg_ioctl_arg
{
56 /* The subfunctions (for the op field) */
57 #define BLKPG_ADD_PARTITION 1
58 #define BLKPG_DEL_PARTITION 2
60 /* Sizes of name fields. Unused at present. */
61 #define BLKPG_DEVNAMELTH 64
62 #define BLKPG_VOLNAMELTH 64
64 /* The data structure for ADD_PARTITION and DEL_PARTITION */
65 struct blkpg_partition
{
66 long long start
; /* starting offset in bytes */
67 long long length
; /* length in bytes */
68 int pno
; /* partition number */
69 char devname
[BLKPG_DEVNAMELTH
]; /* partition name, like sda5 or c0d1p2,
70 to be used in kernel messages */
71 char volname
[BLKPG_VOLNAMELTH
]; /* volume label */
76 /* Force a compilation error if condition is true */
77 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
79 /* Force a compilation error if condition is true, but also produce a
80 result (of value 0 and type size_t), so the expression can be used
81 e.g. in a structure initializer (or where-ever else comma expressions
83 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
85 static int is_dlm_hooks_ready
= 0;
87 int dlm_funs_ready(void)
89 return is_dlm_hooks_ready
? 1 : 0;
93 static struct dlm_hooks
*dlm_hooks
= NULL
;
94 struct dlm_lock_resource
*dlm_lock_res
= NULL
;
95 static int ast_called
= 0;
97 struct dlm_lock_resource
{
102 /* Using poll(2) to wait for and dispatch ASTs */
103 static int poll_for_ast(dlm_lshandle_t ls
)
107 pfd
.fd
= dlm_hooks
->ls_get_fd(ls
);
112 if (poll(&pfd
, 1, 0) < 0)
117 dlm_hooks
->dispatch(dlm_hooks
->ls_get_fd(ls
));
124 static void dlm_ast(void *arg
)
129 static char *cluster_name
= NULL
;
130 /* Create the lockspace, take bitmapXXX locks on all the bitmaps. */
131 int cluster_get_dlmlock(int *lockid
)
135 int flags
= LKF_NOQUEUE
;
137 ret
= get_cluster_name(&cluster_name
);
139 pr_err("The md can't get cluster name\n");
143 dlm_lock_res
= xmalloc(sizeof(struct dlm_lock_resource
));
144 dlm_lock_res
->ls
= dlm_hooks
->create_lockspace(cluster_name
, O_RDWR
);
145 if (!dlm_lock_res
->ls
) {
146 pr_err("%s failed to create lockspace\n", cluster_name
);
150 /* Conversions need the lockid in the LKSB */
151 if (flags
& LKF_CONVERT
)
152 dlm_lock_res
->lksb
.sb_lkid
= *lockid
;
154 snprintf(str
, 64, "bitmap%s", cluster_name
);
155 /* if flags with LKF_CONVERT causes below return ENOENT which means
156 * "No such file or directory" */
157 ret
= dlm_hooks
->ls_lock(dlm_lock_res
->ls
, LKM_PWMODE
, &dlm_lock_res
->lksb
,
158 flags
, str
, strlen(str
), 0, dlm_ast
,
159 dlm_lock_res
, NULL
, NULL
);
161 pr_err("error %d when get PW mode on lock %s\n", errno
, str
);
162 dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
166 /* Wait for it to complete */
167 poll_for_ast(dlm_lock_res
->ls
);
168 *lockid
= dlm_lock_res
->lksb
.sb_lkid
;
170 return dlm_lock_res
->lksb
.sb_status
;
173 int cluster_release_dlmlock(int lockid
)
180 /* if flags with LKF_CONVERT causes below return EINVAL which means
181 * "Invalid argument" */
182 ret
= dlm_hooks
->ls_unlock(dlm_lock_res
->ls
, lockid
, 0,
183 &dlm_lock_res
->lksb
, dlm_lock_res
);
185 pr_err("error %d happened when unlock\n", errno
);
186 /* XXX make sure the lock is unlocked eventually */
190 /* Wait for it to complete */
191 poll_for_ast(dlm_lock_res
->ls
);
193 errno
= dlm_lock_res
->lksb
.sb_status
;
194 if (errno
!= EUNLOCK
) {
195 pr_err("error %d happened in ast when unlock lockspace\n", errno
);
196 /* XXX make sure the lockspace is unlocked eventually */
200 ret
= dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
202 pr_err("error %d happened when release lockspace\n", errno
);
203 /* XXX make sure the lockspace is released eventually */
212 int cluster_get_dlmlock(int *lockid
)
216 int cluster_release_dlmlock(int lockid
)
223 * Parse a 128 bit uuid in 4 integers
224 * format is 32 hexx nibbles with options :.<space> separator
225 * If not exactly 32 hex digits are found, return 0
228 int parse_uuid(char *str
, int uuid
[4])
230 int hit
= 0; /* number of Hex digIT */
233 for (i
= 0; i
< 4; i
++)
236 while ((c
= *str
++) != 0) {
238 if (c
>= '0' && c
<= '9')
240 else if (c
>= 'a' && c
<= 'f')
242 else if (c
>= 'A' && c
<= 'F')
244 else if (strchr(":. -", c
))
260 * Get the md version number.
261 * We use the RAID_VERSION ioctl if it is supported
262 * If not, but we have a block device with major '9', we assume
265 * Return version number as 24 but number - assume version parts
269 int md_get_version(int fd
)
274 if (fstat(fd
, &stb
)<0)
276 if ((S_IFMT
&stb
.st_mode
) != S_IFBLK
)
279 if (ioctl(fd
, RAID_VERSION
, &vers
) == 0)
280 return (vers
.major
*10000) + (vers
.minor
*100) + vers
.patchlevel
;
283 if (major(stb
.st_rdev
) == MD_MAJOR
)
288 int get_linux_version()
292 int a
= 0, b
= 0,c
= 0;
297 a
= strtoul(cp
, &cp
, 10);
299 b
= strtoul(cp
+1, &cp
, 10);
301 c
= strtoul(cp
+1, &cp
, 10);
303 return (a
*1000000)+(b
*1000)+c
;
307 int mdadm_version(char *version
)
315 cp
= strchr(version
, '-');
316 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
319 a
= strtoul(cp
, &cp
, 10);
322 b
= strtoul(cp
+1, &cp
, 10);
324 c
= strtoul(cp
+1, &cp
, 10);
327 if (*cp
!= ' ' && *cp
!= '-')
329 return (a
*1000000)+(b
*1000)+c
;
332 unsigned long long parse_size(char *size
)
334 /* parse 'size' which should be a number optionally
335 * followed by 'K', 'M', or 'G'.
336 * Without a suffix, K is assumed.
337 * Number returned is in sectors (half-K)
338 * INVALID_SECTORS returned on error.
341 long long s
= strtoll(size
, &c
, 10);
355 s
*= 1024 * 1024 * 2;
357 case 's': /* sectors */
368 int parse_layout_10(char *layout
)
372 /* Parse the layout string for raid10 */
373 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
374 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
375 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
379 if (layout
[0] == 'n')
381 else if (layout
[0] == 'o')
382 rv
= 0x10000 + (copies
<<8) + 1;
384 rv
= 1 + (copies
<<8);
388 int parse_layout_faulty(char *layout
)
390 /* Parse the layout string for 'faulty' */
391 int ln
= strcspn(layout
, "0123456789");
392 char *m
= xstrdup(layout
);
395 mode
= map_name(faultylayout
, m
);
399 return mode
| (atoi(layout
+ln
)<< ModeShift
);
402 long parse_num(char *num
)
404 /* Either return a valid number, or -1 */
406 long rv
= strtol(num
, &c
, 10);
407 if (rv
< 0 || *c
|| !num
[0])
414 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
417 *slot
= strtoul(input
, &dev
, 10);
418 if (dev
== input
|| dev
[0] != ':')
424 void remove_partitions(int fd
)
426 /* remove partitions from this block devices.
427 * This is used for components added to an array
429 #ifdef BLKPG_DEL_PARTITION
430 struct blkpg_ioctl_arg a
;
431 struct blkpg_partition p
;
433 a
.op
= BLKPG_DEL_PARTITION
;
435 a
.datalen
= sizeof(p
);
437 memset(a
.data
, 0, a
.datalen
);
438 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
439 ioctl(fd
, BLKPG
, &a
);
443 int test_partition(int fd
)
445 /* Check if fd is a whole-disk or a partition.
446 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
447 * will return ENXIO on an invalid partition number.
449 struct blkpg_ioctl_arg a
;
450 struct blkpg_partition p
;
451 a
.op
= BLKPG_DEL_PARTITION
;
453 a
.datalen
= sizeof(p
);
455 memset(a
.data
, 0, a
.datalen
);
457 if (ioctl(fd
, BLKPG
, &a
) == 0)
458 /* Very unlikely, but not a partition */
460 if (errno
== ENXIO
|| errno
== ENOTTY
)
461 /* not a partition */
467 int test_partition_from_id(dev_t id
)
472 sprintf(buf
, "%d:%d", major(id
), minor(id
));
473 fd
= dev_open(buf
, O_RDONLY
);
476 rv
= test_partition(fd
);
481 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
487 for (i
= 0; i
< raid_disks
; i
++)
488 avail_disks
+= !!avail
[i
];
492 /* This is the tricky one - we need to check
493 * which actual disks are present.
495 copies
= (layout
&255)* ((layout
>>8) & 255);
498 /* there must be one of the 'copies' form 'first' */
505 this = (this+1) % raid_disks
;
509 first
= (first
+(layout
&255)) % raid_disks
;
510 } while (first
!= 0);
513 case LEVEL_MULTIPATH
:
514 return avail_disks
>= 1;
517 return avail_disks
== raid_disks
;
519 return avail_disks
>= 1;
521 if (avail_disks
== raid_disks
- 1 &&
522 !avail
[raid_disks
- 1])
523 /* If just the parity device is missing, then we
524 * have enough, even if not clean
530 return avail_disks
>= raid_disks
-1;
532 return avail_disks
>= raid_disks
;
535 return avail_disks
>= raid_disks
-2;
537 return avail_disks
>= raid_disks
;
543 int enough_fd(int fd
)
545 struct mdu_array_info_s array
;
546 struct mdu_disk_info_s disk
;
550 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0 ||
551 array
.raid_disks
<= 0)
553 avail
= xcalloc(array
.raid_disks
, 1);
554 for (i
= 0; i
< MAX_DISKS
&& array
.nr_disks
> 0; i
++) {
556 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
558 if (disk
.major
== 0 && disk
.minor
== 0)
562 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
564 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
566 avail
[disk
.raid_disk
] = 1;
568 /* This is used on an active array, so assume it is clean */
569 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
575 const int uuid_zero
[4] = { 0, 0, 0, 0 };
577 int same_uuid(int a
[4], int b
[4], int swapuuid
)
580 /* parse uuids are hostendian.
581 * uuid's from some superblocks are big-ending
582 * if there is a difference, we need to swap..
584 unsigned char *ac
= (unsigned char *)a
;
585 unsigned char *bc
= (unsigned char *)b
;
587 for (i
= 0; i
< 16; i
+= 4) {
588 if (ac
[i
+0] != bc
[i
+3] ||
589 ac
[i
+1] != bc
[i
+2] ||
590 ac
[i
+2] != bc
[i
+1] ||
605 void copy_uuid(void *a
, int b
[4], int swapuuid
)
608 /* parse uuids are hostendian.
609 * uuid's from some superblocks are big-ending
610 * if there is a difference, we need to swap..
612 unsigned char *ac
= (unsigned char *)a
;
613 unsigned char *bc
= (unsigned char *)b
;
615 for (i
= 0; i
< 16; i
+= 4) {
625 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
632 copy_uuid(uuid
, id
, swap
);
633 for (i
= 0; i
< 4; i
++) {
636 for (j
= 3; j
>= 0; j
--) {
637 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
645 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
647 // dirty hack to work around an issue with super1 superblocks...
648 // super1 superblocks need swapuuid set in order for assembly to
649 // work, but can't have it set if we want this printout to match
650 // all the other uuid printouts in super1.c, so we force swapuuid
651 // to 1 to make our printout match the rest of super1
652 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
656 int check_ext2(int fd
, char *name
)
659 * Check for an ext2fs file system.
660 * Superblock is always 1K at 1K offset
662 * s_magic is le16 at 56 == 0xEF53
663 * report mtime - le32 at 44
665 * logblksize - le32 at 24
667 unsigned char sb
[1024];
669 unsigned long long size
;
671 if (lseek(fd
, 1024,0)!= 1024)
673 if (read(fd
, sb
, 1024)!= 1024)
675 if (sb
[56] != 0x53 || sb
[57] != 0xef)
678 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
679 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
680 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
682 pr_err("%s appears to contain an ext2fs file system\n",
684 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
688 int check_reiser(int fd
, char *name
)
691 * superblock is at 64K
693 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
696 unsigned char sb
[1024];
697 unsigned long long size
;
698 if (lseek(fd
, 64*1024, 0) != 64*1024)
700 if (read(fd
, sb
, 1024) != 1024)
702 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
703 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
705 pr_err("%s appears to contain a reiserfs file system\n",name
);
706 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
707 cont_err("size = %lluK\n", size
*4);
712 int check_raid(int fd
, char *name
)
717 struct supertype
*st
= guess_super(fd
);
721 st
->ss
->load_super(st
, fd
, name
);
722 /* Looks like a raid array .. */
723 pr_err("%s appears to be part of a raid array:\n",
725 st
->ss
->getinfo_super(st
, &info
, NULL
);
726 st
->ss
->free_super(st
);
727 crtime
= info
.array
.ctime
;
728 level
= map_num(pers
, info
.array
.level
);
729 if (!level
) level
= "-unknown-";
730 cont_err("level=%s devices=%d ctime=%s",
731 level
, info
.array
.raid_disks
, ctime(&crtime
));
739 for (i
= 0; i
< 5; i
++) {
741 fprintf(stderr
, "%s%s", mesg
, add
);
743 if (fgets(buf
, 100, stdin
)==NULL
)
745 if (buf
[0]=='y' || buf
[0]=='Y')
747 if (buf
[0]=='n' || buf
[0]=='N')
751 pr_err("assuming 'no'\n");
754 #endif /* MDASSEMBLE */
756 int is_standard(char *dev
, int *nump
)
758 /* tests if dev is a "standard" md dev name.
759 * i.e if the last component is "/dNN" or "/mdNN",
760 * where NN is a string of digits
761 * Returns 1 if a partitionable standard,
762 * -1 if non-partitonable,
763 * 0 if not a standard name.
765 char *d
= strrchr(dev
, '/');
770 if (strncmp(d
, "/d",2) == 0)
771 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
772 else if (strncmp(d
, "/md_d", 5) == 0)
773 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
774 else if (strncmp(d
, "/md", 3) == 0)
775 d
+= 3, type
= -1; /* /dev/mdN */
776 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
777 d
+= 1, type
= -1; /* /dev/md/N */
787 if (nump
) *nump
= num
;
792 unsigned long calc_csum(void *super
, int bytes
)
794 unsigned long long newcsum
= 0;
797 unsigned int *superc
= (unsigned int*) super
;
799 for(i
= 0; i
< bytes
/4; i
++)
800 newcsum
+= superc
[i
];
801 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
803 /* The in-kernel checksum calculation is always 16bit on
804 * the alpha, though it is 32 bit on i386...
805 * I wonder what it is elsewhere... (it uses an API in
806 * a way that it shouldn't).
808 csum
= (csum
& 0xffff) + (csum
>> 16);
809 csum
= (csum
& 0xffff) + (csum
>> 16);
815 char *human_size(long long bytes
)
819 /* We convert bytes to either centi-M{ega,ibi}bytes or
820 * centi-G{igi,ibi}bytes, with appropriate rounding,
821 * and then print 1/100th of those as a decimal.
822 * We allow upto 2048Megabytes before converting to
823 * gigabytes, as that shows more precision and isn't
824 * too large a number.
825 * Terabytes are not yet handled.
828 if (bytes
< 5000*1024)
830 else if (bytes
< 2*1024LL*1024LL*1024LL) {
831 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
832 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
833 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
834 cMiB
/100 , cMiB
% 100,
837 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
838 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
839 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
840 cGiB
/100 , cGiB
% 100,
846 char *human_size_brief(long long bytes
, int prefix
)
850 /* We convert bytes to either centi-M{ega,ibi}bytes or
851 * centi-G{igi,ibi}bytes, with appropriate rounding,
852 * and then print 1/100th of those as a decimal.
853 * We allow upto 2048Megabytes before converting to
854 * gigabytes, as that shows more precision and isn't
855 * too large a number.
856 * Terabytes are not yet handled.
858 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
859 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
862 if (bytes
< 5000*1024)
864 else if (prefix
== IEC
) {
865 if (bytes
< 2*1024LL*1024LL*1024LL) {
866 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
867 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
868 cMiB
/100 , cMiB
% 100);
870 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
871 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
872 cGiB
/100 , cGiB
% 100);
875 else if (prefix
== JEDEC
) {
876 if (bytes
< 2*1024LL*1024LL*1024LL) {
877 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
878 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
881 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
882 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
883 cGB
/100 , cGB
% 100);
892 void print_r10_layout(int layout
)
894 int near
= layout
& 255;
895 int far
= (layout
>> 8) & 255;
896 int offset
= (layout
&0x10000);
900 printf("%s near=%d", sep
, near
);
904 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
906 printf("NO REDUNDANCY");
910 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
911 int chunksize
, unsigned long long devsize
)
915 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
916 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
919 int get_data_disks(int level
, int layout
, int raid_disks
)
923 case 0: data_disks
= raid_disks
;
925 case 1: data_disks
= 1;
928 case 5: data_disks
= raid_disks
- 1;
930 case 6: data_disks
= raid_disks
- 2;
932 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
939 int devnm2devid(char *devnm
)
941 /* First look in /sys/block/$DEVNM/dev for %d:%d
942 * If that fails, try parsing out a number
949 sprintf(path
, "/sys/block/%s/dev", devnm
);
950 fd
= open(path
, O_RDONLY
);
953 int n
= read(fd
, buf
, sizeof(buf
));
957 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
958 return makedev(mjr
, mnr
);
960 if (strncmp(devnm
, "md_d", 4) == 0 &&
962 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
963 ep
> devnm
&& *ep
== 0)
964 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
966 if (strncmp(devnm
, "md", 2) == 0 &&
968 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
969 ep
> devnm
&& *ep
== 0)
970 return makedev(MD_MAJOR
, mnr
);
975 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
976 char *get_md_name(char *devnm
)
978 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
979 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
981 static char devname
[50];
983 dev_t rdev
= devnm2devid(devnm
);
988 if (strncmp(devnm
, "md_", 3) == 0) {
989 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
991 if (stat(devname
, &stb
) == 0
992 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
993 && (stb
.st_rdev
== rdev
))
996 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
997 if (stat(devname
, &stb
) == 0
998 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
999 && (stb
.st_rdev
== rdev
))
1002 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
1003 if (stat(devname
, &stb
) == 0
1004 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1005 && (stb
.st_rdev
== rdev
))
1008 dn
= map_dev(major(rdev
), minor(rdev
), 0);
1011 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
1012 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
1013 if (errno
!= EEXIST
)
1016 if (stat(devname
, &stb
) == 0
1017 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1018 && (stb
.st_rdev
== rdev
))
1024 void put_md_name(char *name
)
1026 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1029 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1031 int get_maj_min(char *dev
, int *major
, int *minor
)
1034 *major
= strtoul(dev
, &e
, 0);
1035 return (e
> dev
&& *e
== ':' && e
[1] &&
1036 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1040 int dev_open(char *dev
, int flags
)
1042 /* like 'open', but if 'dev' matches %d:%d, create a temp
1043 * block device and open that
1050 if (!dev
) return -1;
1053 if (get_maj_min(dev
, &major
, &minor
)) {
1054 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1055 (int)getpid(), major
, minor
);
1056 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1057 fd
= open(devname
, flags
);
1061 /* Try /tmp as /dev appear to be read-only */
1062 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
1063 (int)getpid(), major
, minor
);
1064 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1065 fd
= open(devname
, flags
);
1070 fd
= open(dev
, flags
);
1074 int open_dev_flags(char *devnm
, int flags
)
1079 devid
= devnm2devid(devnm
);
1080 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1081 return dev_open(buf
, flags
);
1084 int open_dev(char *devnm
)
1086 return open_dev_flags(devnm
, O_RDONLY
);
1089 int open_dev_excl(char *devnm
)
1094 int devid
= devnm2devid(devnm
);
1097 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1098 for (i
= 0 ; i
< 25 ; i
++) {
1099 int fd
= dev_open(buf
, flags
|O_EXCL
);
1102 if (errno
== EACCES
&& flags
== O_RDWR
) {
1115 int same_dev(char *one
, char *two
)
1117 struct stat st1
, st2
;
1118 if (stat(one
, &st1
) != 0)
1120 if (stat(two
, &st2
) != 0)
1122 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1124 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1126 return st1
.st_rdev
== st2
.st_rdev
;
1129 void wait_for(char *dev
, int fd
)
1132 struct stat stb_want
;
1135 if (fstat(fd
, &stb_want
) != 0 ||
1136 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1139 for (i
= 0 ; i
< 25 ; i
++) {
1141 if (stat(dev
, &stb
) == 0 &&
1142 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1143 (stb
.st_rdev
== stb_want
.st_rdev
))
1150 dprintf("timeout waiting for %s\n", dev
);
1153 struct superswitch
*superlist
[] =
1156 &super_ddf
, &super_imsm
,
1160 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1162 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1164 mdu_array_info_t array
;
1167 struct supertype
*st
= NULL
;
1172 char *subarray
= NULL
;
1173 char container
[32] = "";
1175 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1178 vers
= sra
->array
.major_version
;
1179 minor
= sra
->array
.minor_version
;
1180 verstr
= sra
->text_version
;
1182 if (ioctl(fd
, GET_ARRAY_INFO
, &array
))
1183 array
.major_version
= array
.minor_version
= 0;
1184 vers
= array
.major_version
;
1185 minor
= array
.minor_version
;
1190 sprintf(version
, "%d.%d", vers
, minor
);
1193 if (minor
== -2 && is_subarray(verstr
)) {
1194 char *dev
= verstr
+1;
1196 subarray
= strchr(dev
, '/');
1199 subarray
= xstrdup(subarray
);
1201 strcpy(container
, dev
);
1204 sra
= sysfs_read(-1, container
, GET_VERSION
);
1205 if (sra
&& sra
->text_version
[0])
1206 verstr
= sra
->text_version
;
1208 verstr
= "-no-metadata-";
1211 for (i
= 0; st
== NULL
&& superlist
[i
] ; i
++)
1212 st
= superlist
[i
]->match_metadata_desc(verstr
);
1219 *subarrayp
= subarray
;
1220 strcpy(st
->container_devnm
, container
);
1221 strcpy(st
->devnm
, fd2devnm(fd
));
1227 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1229 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1234 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1235 fd
= dev_open(buf
, O_RDONLY
);
1238 if (get_dev_size(fd
, NULL
, size
)) {
1246 struct supertype
*dup_super(struct supertype
*orig
)
1248 struct supertype
*st
;
1252 st
= xcalloc(1, sizeof(*st
));
1254 st
->max_devs
= orig
->max_devs
;
1255 st
->minor_version
= orig
->minor_version
;
1256 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1257 st
->data_offset
= orig
->data_offset
;
1263 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1265 /* try each load_super to find the best match,
1266 * and return the best superswitch
1268 struct superswitch
*ss
;
1269 struct supertype
*st
;
1270 time_t besttime
= 0;
1274 st
= xcalloc(1, sizeof(*st
));
1275 st
->container_devnm
[0] = 0;
1277 for (i
= 0 ; superlist
[i
]; i
++) {
1280 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1282 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1284 memset(st
, 0, sizeof(*st
));
1285 st
->ignore_hw_compat
= 1;
1286 rv
= ss
->load_super(st
, fd
, NULL
);
1289 st
->ss
->getinfo_super(st
, &info
, NULL
);
1290 if (bestsuper
== -1 ||
1291 besttime
< info
.array
.ctime
) {
1293 besttime
= info
.array
.ctime
;
1298 if (bestsuper
!= -1) {
1300 memset(st
, 0, sizeof(*st
));
1301 st
->ignore_hw_compat
= 1;
1302 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1304 superlist
[bestsuper
]->free_super(st
);
1312 /* Return size of device in bytes */
1313 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1315 unsigned long long ldsize
;
1318 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1319 ldsize
= (unsigned long long)st
.st_size
;
1322 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1325 unsigned long dsize
;
1326 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1331 pr_err("Cannot get size of %s: %s\b",
1332 dname
, strerror(errno
));
1340 /* Return true if this can only be a container, not a member device.
1341 * i.e. is and md device and size is zero
1343 int must_be_container(int fd
)
1345 unsigned long long size
;
1346 if (md_get_version(fd
) < 0)
1348 if (get_dev_size(fd
, NULL
, &size
) == 0)
1355 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1356 * Returns: 1 if successful
1357 * -1 for unknown partition type
1358 * 0 for other errors
1360 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1363 unsigned char empty_gpt_entry
[16]= {0};
1364 struct GPT_part_entry
*part
;
1366 unsigned long long curr_part_end
;
1367 unsigned all_partitions
, entry_size
;
1372 BUILD_BUG_ON(sizeof(gpt
) != 512);
1373 /* skip protective MBR */
1374 lseek(fd
, 512, SEEK_SET
);
1375 /* read GPT header */
1376 if (read(fd
, &gpt
, 512) != 512)
1379 /* get the number of partition entries and the entry size */
1380 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1381 entry_size
= __le32_to_cpu(gpt
.part_size
);
1383 /* Check GPT signature*/
1384 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1388 if (all_partitions
> 1024 ||
1389 entry_size
> sizeof(buf
))
1392 part
= (struct GPT_part_entry
*)buf
;
1394 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1395 /* read partition entry */
1396 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1399 /* is this valid partition? */
1400 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1401 /* check the last lba for the current partition */
1402 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1403 if (curr_part_end
> *endofpart
)
1404 *endofpart
= curr_part_end
;
1411 /* Sets endofpart parameter to the last block used by the last partition on the device.
1412 * Returns: 1 if successful
1413 * -1 for unknown partition type
1414 * 0 for other errors
1416 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1418 struct MBR boot_sect
;
1419 struct MBR_part_record
*part
;
1420 unsigned long long curr_part_end
;
1426 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1429 if (read(fd
, &boot_sect
, 512) != 512)
1432 /* check MBP signature */
1433 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1435 /* found the correct signature */
1436 part
= boot_sect
.parts
;
1438 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1439 /* check for GPT type */
1440 if (part
->part_type
== MBR_GPT_PARTITION_TYPE
) {
1441 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1444 /* check the last used lba for the current partition */
1445 curr_part_end
= __le32_to_cpu(part
->first_sect_lba
) +
1446 __le32_to_cpu(part
->blocks_num
);
1447 if (curr_part_end
> *endofpart
)
1448 *endofpart
= curr_part_end
;
1453 /* Unknown partition table */
1460 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1461 unsigned long long size
)
1464 * Check where the last partition ends
1466 unsigned long long endofpart
;
1469 if ((ret
= get_last_partition_end(fd
, &endofpart
)) > 0) {
1470 /* There appears to be a partition table here */
1471 if (freesize
== 0) {
1472 /* partitions will not be visible in new device */
1473 pr_err("partition table exists on %s but will be lost or\n"
1474 " meaningless after creating array\n",
1477 } else if (endofpart
> freesize
) {
1478 /* last partition overlaps metadata */
1479 pr_err("metadata will over-write last partition on %s.\n",
1482 } else if (size
&& endofpart
> size
) {
1483 /* partitions will be truncated in new device */
1484 pr_err("array size is too small to cover all partitions on %s.\n",
1492 int open_container(int fd
)
1494 /* 'fd' is a block device. Find out if it is in use
1495 * by a container, and return an open fd on that container.
1506 if (fstat(fd
, &st
) != 0)
1508 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1509 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1510 e
= path
+ strlen(path
);
1512 dir
= opendir(path
);
1515 while ((de
= readdir(dir
))) {
1518 if (de
->d_name
[0] == '.')
1520 /* Need to make sure it is a container and not a volume */
1521 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1522 dfd
= open(path
, O_RDONLY
);
1525 n
= read(dfd
, buf
, sizeof(buf
));
1527 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1530 if (strncmp(buf
, "external", 8) != 0 ||
1534 sprintf(e
, "/%s/dev", de
->d_name
);
1535 dfd
= open(path
, O_RDONLY
);
1538 n
= read(dfd
, buf
, sizeof(buf
));
1540 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1543 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1545 sprintf(buf
, "%d:%d", major
, minor
);
1546 dfd
= dev_open(buf
, O_RDONLY
);
1556 struct superswitch
*version_to_superswitch(char *vers
)
1560 for (i
= 0; superlist
[i
]; i
++) {
1561 struct superswitch
*ss
= superlist
[i
];
1563 if (strcmp(vers
, ss
->name
) == 0)
1570 int metadata_container_matches(char *metadata
, char *devnm
)
1572 /* Check if 'devnm' is the container named in 'metadata'
1574 * /containername/componentname or
1575 * -containername/componentname
1578 if (*metadata
!= '/' && *metadata
!= '-')
1581 if (strncmp(metadata
+1, devnm
, l
) != 0)
1583 if (metadata
[l
+1] != '/')
1588 int metadata_subdev_matches(char *metadata
, char *devnm
)
1590 /* Check if 'devnm' is the subdev named in 'metadata'
1592 * /containername/subdev or
1593 * -containername/subdev
1596 if (*metadata
!= '/' && *metadata
!= '-')
1598 sl
= strchr(metadata
+1, '/');
1601 if (strcmp(sl
+1, devnm
) == 0)
1606 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1608 if (mdstat
->metadata_version
== NULL
||
1609 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1610 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1616 int is_subarray_active(char *subarray
, char *container
)
1618 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1619 struct mdstat_ent
*ent
;
1621 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1622 if (is_container_member(ent
, container
))
1623 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1626 free_mdstat(mdstat
);
1631 /* open_subarray - opens a subarray in a container
1632 * @dev: container device name
1633 * @st: empty supertype
1634 * @quiet: block reporting errors flag
1636 * On success returns an fd to a container and fills in *st
1638 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1641 struct mdinfo
*info
;
1645 fd
= open(dev
, O_RDWR
|O_EXCL
);
1648 pr_err("Couldn't open %s, aborting\n",
1653 _devnm
= fd2devnm(fd
);
1654 if (_devnm
== NULL
) {
1656 pr_err("Failed to determine device number for %s\n",
1660 strcpy(st
->devnm
, _devnm
);
1662 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1665 pr_err("Failed to read sysfs for %s\n",
1670 if (mdi
->array
.level
!= UnSet
) {
1672 pr_err("%s is not a container\n", dev
);
1676 st
->ss
= version_to_superswitch(mdi
->text_version
);
1679 pr_err("Operation not supported for %s metadata\n",
1684 if (st
->devnm
[0] == 0) {
1686 pr_err("Failed to allocate device name\n");
1690 if (!st
->ss
->load_container
) {
1692 pr_err("%s is not a container\n", dev
);
1696 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1698 pr_err("Failed to load metadata for %s\n",
1703 info
= st
->ss
->container_content(st
, subarray
);
1706 pr_err("Failed to find subarray-%s in %s\n",
1716 st
->ss
->free_super(st
);
1729 int add_disk(int mdfd
, struct supertype
*st
,
1730 struct mdinfo
*sra
, struct mdinfo
*info
)
1732 /* Add a device to an array, in one of 2 ways. */
1735 if (st
->ss
->external
) {
1736 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1737 info
->recovery_start
= MaxSector
;
1739 info
->recovery_start
= 0;
1740 rv
= sysfs_add_disk(sra
, info
, 0);
1743 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1747 sd2
= xmalloc(sizeof(*sd2
));
1749 sd2
->next
= sra
->devs
;
1755 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1759 int remove_disk(int mdfd
, struct supertype
*st
,
1760 struct mdinfo
*sra
, struct mdinfo
*info
)
1763 /* Remove the disk given by 'info' from the array */
1765 if (st
->ss
->external
)
1766 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1769 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1774 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1776 /* Initialise kernel's knowledge of array.
1777 * This varies between externally managed arrays
1780 int vers
= md_get_version(mdfd
);
1784 if (st
->ss
->external
)
1785 rv
= sysfs_set_array(info
, vers
);
1788 if ((vers
% 100) >= 1) { /* can use different versions */
1789 mdu_array_info_t inf
;
1790 memset(&inf
, 0, sizeof(inf
));
1791 inf
.major_version
= info
->array
.major_version
;
1792 inf
.minor_version
= info
->array
.minor_version
;
1793 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1795 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1799 unsigned long long min_recovery_start(struct mdinfo
*array
)
1801 /* find the minimum recovery_start in an array for metadata
1802 * formats that only record per-array recovery progress instead
1805 unsigned long long recovery_start
= MaxSector
;
1808 for (d
= array
->devs
; d
; d
= d
->next
)
1809 recovery_start
= min(recovery_start
, d
->recovery_start
);
1811 return recovery_start
;
1814 int mdmon_pid(char *devnm
)
1821 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1823 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1827 n
= read(fd
, pid
, 9);
1834 int mdmon_running(char *devnm
)
1836 int pid
= mdmon_pid(devnm
);
1839 if (kill(pid
, 0) == 0)
1844 int start_mdmon(char *devnm
)
1858 if (check_env("MDADM_NO_MDMON"))
1861 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1865 sl
= strrchr(pathbuf
, '/');
1870 strcpy(sl
, "mdmon");
1874 /* First try to run systemctl */
1875 if (!check_env("MDADM_NO_SYSTEMCTL"))
1878 /* FIXME yuk. CLOSE_EXEC?? */
1880 for (i
= 3; skipped
< 20; i
++)
1886 /* Don't want to see error messages from
1887 * systemctl. If the service doesn't exist,
1888 * we start mdmon ourselves.
1891 open("/dev/null", O_WRONLY
);
1892 snprintf(pathbuf
, sizeof(pathbuf
), "mdmon@%s.service",
1894 status
= execl("/usr/bin/systemctl", "systemctl",
1897 status
= execl("/bin/systemctl", "systemctl", "start",
1900 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1902 default: /* parent - good */
1903 pid
= wait(&status
);
1904 if (pid
>= 0 && status
== 0)
1908 /* That failed, try running mdmon directly */
1911 /* FIXME yuk. CLOSE_EXEC?? */
1913 for (i
= 3; skipped
< 20; i
++)
1919 for (i
= 0; paths
[i
]; i
++)
1921 execl(paths
[i
], paths
[i
],
1925 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1927 default: /* parent - good */
1928 pid
= wait(&status
);
1929 if (pid
< 0 || status
!= 0) {
1930 pr_err("failed to launch mdmon. Array remains readonly\n");
1937 __u32
random32(void)
1940 int rfd
= open("/dev/urandom", O_RDONLY
);
1941 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1949 int flush_metadata_updates(struct supertype
*st
)
1953 st
->update_tail
= NULL
;
1957 sfd
= connect_monitor(st
->container_devnm
);
1961 while (st
->updates
) {
1962 struct metadata_update
*mu
= st
->updates
;
1963 st
->updates
= mu
->next
;
1965 send_message(sfd
, mu
, 0);
1973 st
->update_tail
= NULL
;
1977 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
1980 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
1985 mu
->space_list
= NULL
;
1987 *st
->update_tail
= mu
;
1988 st
->update_tail
= &mu
->next
;
1990 #endif /* MDASSEMBLE */
1993 /* tinyc doesn't optimize this check in ioctl.h out ... */
1994 unsigned int __invalid_size_argument_for_IOC
= 0;
1997 int experimental(void)
1999 if (check_env("MDADM_EXPERIMENTAL"))
2002 pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n");
2007 /* Pick all spares matching given criteria from a container
2008 * if min_size == 0 do not check size
2009 * if domlist == NULL do not check domains
2010 * if spare_group given add it to domains of each spare
2011 * metadata allows to test domains using metadata of destination array */
2012 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2013 unsigned long long min_size
,
2014 struct domainlist
*domlist
,
2016 const char *metadata
, int get_one
)
2018 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2020 /* get list of all disks in container */
2021 if (st
->ss
->getinfo_super_disks
)
2022 disks
= st
->ss
->getinfo_super_disks(st
);
2026 /* find spare devices on the list */
2028 disks
->array
.spare_disks
= 0;
2032 if (d
->disk
.state
== 0) {
2033 /* check if size is acceptable */
2034 unsigned long long dev_size
;
2035 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2038 (dev_size_from_id(dev
, &dev_size
) &&
2039 dev_size
>= min_size
))
2041 /* check if domain matches */
2042 if (found
&& domlist
) {
2043 struct dev_policy
*pol
= devid_policy(dev
);
2045 pol_add(&pol
, pol_domain
,
2047 if (domain_test(domlist
, pol
, metadata
) != 1)
2049 dev_policy_free(pol
);
2054 disks
->array
.spare_disks
++;
2068 /* Checks if paths point to the same device
2069 * Returns 0 if they do.
2070 * Returns 1 if they don't.
2071 * Returns -1 if something went wrong,
2072 * e.g. paths are empty or the files
2073 * they point to don't exist */
2074 int compare_paths (char* path1
, char* path2
)
2076 struct stat st1
,st2
;
2078 if (path1
== NULL
|| path2
== NULL
)
2080 if (stat(path1
,&st1
) != 0)
2082 if (stat(path2
,&st2
) != 0)
2084 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2089 /* Make sure we can open as many devices as needed */
2090 void enable_fds(int devices
)
2092 unsigned int fds
= 20 + devices
;
2094 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0
2095 || lim
.rlim_cur
>= fds
)
2097 if (lim
.rlim_max
< fds
)
2100 setrlimit(RLIMIT_NOFILE
, &lim
);
2105 /* This is based on similar function in systemd. */
2107 /* statfs.f_type is signed long on s390x and MIPS, causing all
2108 sorts of sign extension problems with RAMFS_MAGIC being
2109 defined as 0x858458f6 */
2110 return statfs("/", &s
) >= 0 &&
2111 ((unsigned long)s
.f_type
== TMPFS_MAGIC
||
2112 ((unsigned long)s
.f_type
& 0xFFFFFFFFUL
) ==
2113 ((unsigned long)RAMFS_MAGIC
& 0xFFFFFFFFUL
));
2116 void reopen_mddev(int mdfd
)
2118 /* Re-open without any O_EXCL, but keep
2123 devnm
= fd2devnm(mdfd
);
2125 fd
= open_dev(devnm
);
2126 if (fd
>= 0 && fd
!= mdfd
)
2131 static struct cmap_hooks
*cmap_hooks
= NULL
;
2132 static int is_cmap_hooks_ready
= 0;
2134 void set_cmap_hooks(void)
2136 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2137 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2138 if (!cmap_hooks
->cmap_handle
)
2141 cmap_hooks
->initialize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2142 cmap_hooks
->get_string
= dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2143 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2145 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2146 !cmap_hooks
->finalize
)
2147 dlclose(cmap_hooks
->cmap_handle
);
2149 is_cmap_hooks_ready
= 1;
2152 int get_cluster_name(char **cluster_name
)
2155 cmap_handle_t handle
;
2157 if (!is_cmap_hooks_ready
)
2160 rv
= cmap_hooks
->initialize(&handle
);
2164 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2166 free(*cluster_name
);
2173 cmap_hooks
->finalize(handle
);
2178 void set_dlm_hooks(void)
2180 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2181 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2182 if (!dlm_hooks
->dlm_handle
)
2185 dlm_hooks
->create_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2186 dlm_hooks
->release_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2187 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2188 dlm_hooks
->ls_unlock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock");
2189 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2190 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2192 if (!dlm_hooks
->create_lockspace
|| !dlm_hooks
->ls_lock
||
2193 !dlm_hooks
->ls_unlock
|| !dlm_hooks
->release_lockspace
||
2194 !dlm_hooks
->ls_get_fd
|| !dlm_hooks
->dispatch
)
2195 dlclose(dlm_hooks
->dlm_handle
);
2197 is_dlm_hooks_ready
= 1;
2200 void set_hooks(void)