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>
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 snprintf(str
, 64, "bitmap%s", cluster_name
);
151 ret
= dlm_hooks
->ls_lock(dlm_lock_res
->ls
, LKM_PWMODE
, &dlm_lock_res
->lksb
,
152 flags
, str
, strlen(str
), 0, dlm_ast
,
153 dlm_lock_res
, NULL
, NULL
);
155 pr_err("error %d when get PW mode on lock %s\n", errno
, str
);
156 dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
160 /* Wait for it to complete */
161 poll_for_ast(dlm_lock_res
->ls
);
162 *lockid
= dlm_lock_res
->lksb
.sb_lkid
;
164 return dlm_lock_res
->lksb
.sb_status
;
167 int cluster_release_dlmlock(int lockid
)
174 ret
= dlm_hooks
->ls_unlock(dlm_lock_res
->ls
, lockid
, 0,
175 &dlm_lock_res
->lksb
, dlm_lock_res
);
177 pr_err("error %d happened when unlock\n", errno
);
178 /* XXX make sure the lock is unlocked eventually */
182 /* Wait for it to complete */
183 poll_for_ast(dlm_lock_res
->ls
);
185 errno
= dlm_lock_res
->lksb
.sb_status
;
186 if (errno
!= EUNLOCK
) {
187 pr_err("error %d happened in ast when unlock lockspace\n", errno
);
188 /* XXX make sure the lockspace is unlocked eventually */
192 ret
= dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
194 pr_err("error %d happened when release lockspace\n", errno
);
195 /* XXX make sure the lockspace is released eventually */
204 int cluster_get_dlmlock(int *lockid
)
208 int cluster_release_dlmlock(int lockid
)
215 * Get array info from the kernel. Longer term we want to deprecate the
216 * ioctl and get it from sysfs.
218 int md_get_array_info(int fd
, struct mdu_array_info_s
*array
)
220 return ioctl(fd
, GET_ARRAY_INFO
, array
);
224 * Get disk info from the kernel.
226 int md_get_disk_info(int fd
, struct mdu_disk_info_s
*disk
)
228 return ioctl(fd
, GET_DISK_INFO
, disk
);
232 * Parse a 128 bit uuid in 4 integers
233 * format is 32 hexx nibbles with options :.<space> separator
234 * If not exactly 32 hex digits are found, return 0
237 int parse_uuid(char *str
, int uuid
[4])
239 int hit
= 0; /* number of Hex digIT */
242 for (i
= 0; i
< 4; i
++)
245 while ((c
= *str
++) != 0) {
247 if (c
>= '0' && c
<= '9')
249 else if (c
>= 'a' && c
<= 'f')
251 else if (c
>= 'A' && c
<= 'F')
253 else if (strchr(":. -", c
))
269 * Get the md version number.
270 * We use the RAID_VERSION ioctl if it is supported
271 * If not, but we have a block device with major '9', we assume
274 * Return version number as 24 but number - assume version parts
278 int md_get_version(int fd
)
283 if (fstat(fd
, &stb
)<0)
285 if ((S_IFMT
&stb
.st_mode
) != S_IFBLK
)
288 if (ioctl(fd
, RAID_VERSION
, &vers
) == 0)
289 return (vers
.major
*10000) + (vers
.minor
*100) + vers
.patchlevel
;
292 if (major(stb
.st_rdev
) == MD_MAJOR
)
297 int get_linux_version()
301 int a
= 0, b
= 0,c
= 0;
306 a
= strtoul(cp
, &cp
, 10);
308 b
= strtoul(cp
+1, &cp
, 10);
310 c
= strtoul(cp
+1, &cp
, 10);
312 return (a
*1000000)+(b
*1000)+c
;
316 int mdadm_version(char *version
)
324 cp
= strchr(version
, '-');
325 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
328 a
= strtoul(cp
, &cp
, 10);
331 b
= strtoul(cp
+1, &cp
, 10);
333 c
= strtoul(cp
+1, &cp
, 10);
336 if (*cp
!= ' ' && *cp
!= '-')
338 return (a
*1000000)+(b
*1000)+c
;
341 unsigned long long parse_size(char *size
)
343 /* parse 'size' which should be a number optionally
344 * followed by 'K', 'M', or 'G'.
345 * Without a suffix, K is assumed.
346 * Number returned is in sectors (half-K)
347 * INVALID_SECTORS returned on error.
350 long long s
= strtoll(size
, &c
, 10);
364 s
*= 1024 * 1024 * 2;
366 case 's': /* sectors */
377 int parse_layout_10(char *layout
)
381 /* Parse the layout string for raid10 */
382 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
383 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
384 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
388 if (layout
[0] == 'n')
390 else if (layout
[0] == 'o')
391 rv
= 0x10000 + (copies
<<8) + 1;
393 rv
= 1 + (copies
<<8);
397 int parse_layout_faulty(char *layout
)
399 /* Parse the layout string for 'faulty' */
400 int ln
= strcspn(layout
, "0123456789");
401 char *m
= xstrdup(layout
);
404 mode
= map_name(faultylayout
, m
);
408 return mode
| (atoi(layout
+ln
)<< ModeShift
);
411 long parse_num(char *num
)
413 /* Either return a valid number, or -1 */
415 long rv
= strtol(num
, &c
, 10);
416 if (rv
< 0 || *c
|| !num
[0])
423 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
426 *slot
= strtoul(input
, &dev
, 10);
427 if (dev
== input
|| dev
[0] != ':')
433 void remove_partitions(int fd
)
435 /* remove partitions from this block devices.
436 * This is used for components added to an array
438 #ifdef BLKPG_DEL_PARTITION
439 struct blkpg_ioctl_arg a
;
440 struct blkpg_partition p
;
442 a
.op
= BLKPG_DEL_PARTITION
;
444 a
.datalen
= sizeof(p
);
446 memset(a
.data
, 0, a
.datalen
);
447 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
448 ioctl(fd
, BLKPG
, &a
);
452 int test_partition(int fd
)
454 /* Check if fd is a whole-disk or a partition.
455 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
456 * will return ENXIO on an invalid partition number.
458 struct blkpg_ioctl_arg a
;
459 struct blkpg_partition p
;
460 a
.op
= BLKPG_DEL_PARTITION
;
462 a
.datalen
= sizeof(p
);
464 memset(a
.data
, 0, a
.datalen
);
466 if (ioctl(fd
, BLKPG
, &a
) == 0)
467 /* Very unlikely, but not a partition */
469 if (errno
== ENXIO
|| errno
== ENOTTY
)
470 /* not a partition */
476 int test_partition_from_id(dev_t id
)
481 sprintf(buf
, "%d:%d", major(id
), minor(id
));
482 fd
= dev_open(buf
, O_RDONLY
);
485 rv
= test_partition(fd
);
490 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
496 for (i
= 0; i
< raid_disks
; i
++)
497 avail_disks
+= !!avail
[i
];
501 /* This is the tricky one - we need to check
502 * which actual disks are present.
504 copies
= (layout
&255)* ((layout
>>8) & 255);
507 /* there must be one of the 'copies' form 'first' */
514 this = (this+1) % raid_disks
;
518 first
= (first
+(layout
&255)) % raid_disks
;
519 } while (first
!= 0);
522 case LEVEL_MULTIPATH
:
523 return avail_disks
>= 1;
526 return avail_disks
== raid_disks
;
528 return avail_disks
>= 1;
530 if (avail_disks
== raid_disks
- 1 &&
531 !avail
[raid_disks
- 1])
532 /* If just the parity device is missing, then we
533 * have enough, even if not clean
539 return avail_disks
>= raid_disks
-1;
541 return avail_disks
>= raid_disks
;
544 return avail_disks
>= raid_disks
-2;
546 return avail_disks
>= raid_disks
;
552 int enough_fd(int fd
)
554 struct mdu_array_info_s array
;
555 struct mdu_disk_info_s disk
;
559 if (md_get_array_info(fd
, &array
) != 0 || array
.raid_disks
<= 0)
561 avail
= xcalloc(array
.raid_disks
, 1);
562 for (i
= 0; i
< MAX_DISKS
&& array
.nr_disks
> 0; i
++) {
564 if (md_get_disk_info(fd
, &disk
) != 0)
566 if (disk
.major
== 0 && disk
.minor
== 0)
570 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
572 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
574 avail
[disk
.raid_disk
] = 1;
576 /* This is used on an active array, so assume it is clean */
577 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
583 const int uuid_zero
[4] = { 0, 0, 0, 0 };
585 int same_uuid(int a
[4], int b
[4], int swapuuid
)
588 /* parse uuids are hostendian.
589 * uuid's from some superblocks are big-ending
590 * if there is a difference, we need to swap..
592 unsigned char *ac
= (unsigned char *)a
;
593 unsigned char *bc
= (unsigned char *)b
;
595 for (i
= 0; i
< 16; i
+= 4) {
596 if (ac
[i
+0] != bc
[i
+3] ||
597 ac
[i
+1] != bc
[i
+2] ||
598 ac
[i
+2] != bc
[i
+1] ||
613 void copy_uuid(void *a
, int b
[4], int swapuuid
)
616 /* parse uuids are hostendian.
617 * uuid's from some superblocks are big-ending
618 * if there is a difference, we need to swap..
620 unsigned char *ac
= (unsigned char *)a
;
621 unsigned char *bc
= (unsigned char *)b
;
623 for (i
= 0; i
< 16; i
+= 4) {
633 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
640 copy_uuid(uuid
, id
, swap
);
641 for (i
= 0; i
< 4; i
++) {
644 for (j
= 3; j
>= 0; j
--) {
645 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
653 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
655 // dirty hack to work around an issue with super1 superblocks...
656 // super1 superblocks need swapuuid set in order for assembly to
657 // work, but can't have it set if we want this printout to match
658 // all the other uuid printouts in super1.c, so we force swapuuid
659 // to 1 to make our printout match the rest of super1
660 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
664 int check_ext2(int fd
, char *name
)
667 * Check for an ext2fs file system.
668 * Superblock is always 1K at 1K offset
670 * s_magic is le16 at 56 == 0xEF53
671 * report mtime - le32 at 44
673 * logblksize - le32 at 24
675 unsigned char sb
[1024];
677 unsigned long long size
;
679 if (lseek(fd
, 1024,0)!= 1024)
681 if (read(fd
, sb
, 1024)!= 1024)
683 if (sb
[56] != 0x53 || sb
[57] != 0xef)
686 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
687 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
688 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
690 pr_err("%s appears to contain an ext2fs file system\n",
692 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
696 int check_reiser(int fd
, char *name
)
699 * superblock is at 64K
701 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
704 unsigned char sb
[1024];
705 unsigned long long size
;
706 if (lseek(fd
, 64*1024, 0) != 64*1024)
708 if (read(fd
, sb
, 1024) != 1024)
710 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
711 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
713 pr_err("%s appears to contain a reiserfs file system\n",name
);
714 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
715 cont_err("size = %lluK\n", size
*4);
720 int check_raid(int fd
, char *name
)
725 struct supertype
*st
= guess_super(fd
);
729 if (st
->ss
->add_to_super
!= NULL
) {
730 st
->ss
->load_super(st
, fd
, name
);
731 /* Looks like a raid array .. */
732 pr_err("%s appears to be part of a raid array:\n", name
);
733 st
->ss
->getinfo_super(st
, &info
, NULL
);
734 st
->ss
->free_super(st
);
735 crtime
= info
.array
.ctime
;
736 level
= map_num(pers
, info
.array
.level
);
739 cont_err("level=%s devices=%d ctime=%s",
740 level
, info
.array
.raid_disks
, ctime(&crtime
));
742 /* Looks like GPT or MBR */
743 pr_err("partition table exists on %s\n", name
);
752 for (i
= 0; i
< 5; i
++) {
754 fprintf(stderr
, "%s%s", mesg
, add
);
756 if (fgets(buf
, 100, stdin
)==NULL
)
758 if (buf
[0]=='y' || buf
[0]=='Y')
760 if (buf
[0]=='n' || buf
[0]=='N')
764 pr_err("assuming 'no'\n");
767 #endif /* MDASSEMBLE */
769 int is_standard(char *dev
, int *nump
)
771 /* tests if dev is a "standard" md dev name.
772 * i.e if the last component is "/dNN" or "/mdNN",
773 * where NN is a string of digits
774 * Returns 1 if a partitionable standard,
775 * -1 if non-partitonable,
776 * 0 if not a standard name.
778 char *d
= strrchr(dev
, '/');
783 if (strncmp(d
, "/d",2) == 0)
784 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
785 else if (strncmp(d
, "/md_d", 5) == 0)
786 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
787 else if (strncmp(d
, "/md", 3) == 0)
788 d
+= 3, type
= -1; /* /dev/mdN */
789 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
790 d
+= 1, type
= -1; /* /dev/md/N */
800 if (nump
) *nump
= num
;
805 unsigned long calc_csum(void *super
, int bytes
)
807 unsigned long long newcsum
= 0;
810 unsigned int *superc
= (unsigned int*) super
;
812 for(i
= 0; i
< bytes
/4; i
++)
813 newcsum
+= superc
[i
];
814 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
816 /* The in-kernel checksum calculation is always 16bit on
817 * the alpha, though it is 32 bit on i386...
818 * I wonder what it is elsewhere... (it uses an API in
819 * a way that it shouldn't).
821 csum
= (csum
& 0xffff) + (csum
>> 16);
822 csum
= (csum
& 0xffff) + (csum
>> 16);
828 char *human_size(long long bytes
)
832 /* We convert bytes to either centi-M{ega,ibi}bytes or
833 * centi-G{igi,ibi}bytes, with appropriate rounding,
834 * and then print 1/100th of those as a decimal.
835 * We allow upto 2048Megabytes before converting to
836 * gigabytes, as that shows more precision and isn't
837 * too large a number.
838 * Terabytes are not yet handled.
841 if (bytes
< 5000*1024)
843 else if (bytes
< 2*1024LL*1024LL*1024LL) {
844 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
845 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
846 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
847 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
849 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
850 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
851 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
852 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
857 char *human_size_brief(long long bytes
, int prefix
)
861 /* We convert bytes to either centi-M{ega,ibi}bytes or
862 * centi-G{igi,ibi}bytes, with appropriate rounding,
863 * and then print 1/100th of those as a decimal.
864 * We allow upto 2048Megabytes before converting to
865 * gigabytes, as that shows more precision and isn't
866 * too large a number.
867 * Terabytes are not yet handled.
869 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
870 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
873 if (bytes
< 5000*1024)
875 else if (prefix
== IEC
) {
876 if (bytes
< 2*1024LL*1024LL*1024LL) {
877 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
878 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
879 cMiB
/100, cMiB
% 100);
881 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
882 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
883 cGiB
/100, cGiB
% 100);
886 else if (prefix
== JEDEC
) {
887 if (bytes
< 2*1024LL*1024LL*1024LL) {
888 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
889 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
892 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
893 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
903 void print_r10_layout(int layout
)
905 int near
= layout
& 255;
906 int far
= (layout
>> 8) & 255;
907 int offset
= (layout
&0x10000);
911 printf("%s near=%d", sep
, near
);
915 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
917 printf("NO REDUNDANCY");
921 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
922 int chunksize
, unsigned long long devsize
)
926 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
927 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
930 int get_data_disks(int level
, int layout
, int raid_disks
)
934 case 0: data_disks
= raid_disks
;
936 case 1: data_disks
= 1;
939 case 5: data_disks
= raid_disks
- 1;
941 case 6: data_disks
= raid_disks
- 2;
943 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
950 dev_t
devnm2devid(char *devnm
)
952 /* First look in /sys/block/$DEVNM/dev for %d:%d
953 * If that fails, try parsing out a number
960 sprintf(path
, "/sys/block/%s/dev", devnm
);
961 fd
= open(path
, O_RDONLY
);
964 int n
= read(fd
, buf
, sizeof(buf
));
968 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
969 return makedev(mjr
, mnr
);
971 if (strncmp(devnm
, "md_d", 4) == 0 &&
973 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
974 ep
> devnm
&& *ep
== 0)
975 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
977 if (strncmp(devnm
, "md", 2) == 0 &&
979 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
980 ep
> devnm
&& *ep
== 0)
981 return makedev(MD_MAJOR
, mnr
);
986 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
987 char *get_md_name(char *devnm
)
989 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
990 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
992 static char devname
[50];
994 dev_t rdev
= devnm2devid(devnm
);
999 if (strncmp(devnm
, "md_", 3) == 0) {
1000 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
1002 if (stat(devname
, &stb
) == 0
1003 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1004 && (stb
.st_rdev
== rdev
))
1007 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
1008 if (stat(devname
, &stb
) == 0
1009 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1010 && (stb
.st_rdev
== rdev
))
1013 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
1014 if (stat(devname
, &stb
) == 0
1015 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1016 && (stb
.st_rdev
== rdev
))
1019 dn
= map_dev(major(rdev
), minor(rdev
), 0);
1022 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
1023 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
1024 if (errno
!= EEXIST
)
1027 if (stat(devname
, &stb
) == 0
1028 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1029 && (stb
.st_rdev
== rdev
))
1035 void put_md_name(char *name
)
1037 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1040 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1042 int get_maj_min(char *dev
, int *major
, int *minor
)
1045 *major
= strtoul(dev
, &e
, 0);
1046 return (e
> dev
&& *e
== ':' && e
[1] &&
1047 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1051 int dev_open(char *dev
, int flags
)
1053 /* like 'open', but if 'dev' matches %d:%d, create a temp
1054 * block device and open that
1065 if (get_maj_min(dev
, &major
, &minor
)) {
1066 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1067 (int)getpid(), major
, minor
);
1068 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1069 fd
= open(devname
, flags
);
1073 /* Try /tmp as /dev appear to be read-only */
1074 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
1075 (int)getpid(), major
, minor
);
1076 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1077 fd
= open(devname
, flags
);
1082 fd
= open(dev
, flags
);
1086 int open_dev_flags(char *devnm
, int flags
)
1091 devid
= devnm2devid(devnm
);
1092 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1093 return dev_open(buf
, flags
);
1096 int open_dev(char *devnm
)
1098 return open_dev_flags(devnm
, O_RDONLY
);
1101 int open_dev_excl(char *devnm
)
1106 dev_t devid
= devnm2devid(devnm
);
1109 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1110 for (i
= 0; i
< 25; i
++) {
1111 int fd
= dev_open(buf
, flags
|O_EXCL
);
1114 if (errno
== EACCES
&& flags
== O_RDWR
) {
1127 int same_dev(char *one
, char *two
)
1129 struct stat st1
, st2
;
1130 if (stat(one
, &st1
) != 0)
1132 if (stat(two
, &st2
) != 0)
1134 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1136 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1138 return st1
.st_rdev
== st2
.st_rdev
;
1141 void wait_for(char *dev
, int fd
)
1144 struct stat stb_want
;
1147 if (fstat(fd
, &stb_want
) != 0 ||
1148 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1151 for (i
= 0; i
< 25; i
++) {
1153 if (stat(dev
, &stb
) == 0 &&
1154 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1155 (stb
.st_rdev
== stb_want
.st_rdev
))
1162 dprintf("timeout waiting for %s\n", dev
);
1165 struct superswitch
*superlist
[] =
1168 &super_ddf
, &super_imsm
,
1172 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1174 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1176 mdu_array_info_t array
;
1179 struct supertype
*st
= NULL
;
1184 char *subarray
= NULL
;
1185 char container
[32] = "";
1187 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1190 vers
= sra
->array
.major_version
;
1191 minor
= sra
->array
.minor_version
;
1192 verstr
= sra
->text_version
;
1194 if (md_get_array_info(fd
, &array
))
1195 array
.major_version
= array
.minor_version
= 0;
1196 vers
= array
.major_version
;
1197 minor
= array
.minor_version
;
1202 sprintf(version
, "%d.%d", vers
, minor
);
1205 if (minor
== -2 && is_subarray(verstr
)) {
1206 char *dev
= verstr
+1;
1208 subarray
= strchr(dev
, '/');
1211 subarray
= xstrdup(subarray
);
1213 strcpy(container
, dev
);
1215 sra
= sysfs_read(-1, container
, GET_VERSION
);
1216 if (sra
&& sra
->text_version
[0])
1217 verstr
= sra
->text_version
;
1219 verstr
= "-no-metadata-";
1222 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1223 st
= superlist
[i
]->match_metadata_desc(verstr
);
1229 *subarrayp
= subarray
;
1230 strcpy(st
->container_devnm
, container
);
1231 strcpy(st
->devnm
, fd2devnm(fd
));
1237 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1239 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1244 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1245 fd
= dev_open(buf
, O_RDONLY
);
1248 if (get_dev_size(fd
, NULL
, size
)) {
1256 struct supertype
*dup_super(struct supertype
*orig
)
1258 struct supertype
*st
;
1262 st
= xcalloc(1, sizeof(*st
));
1264 st
->max_devs
= orig
->max_devs
;
1265 st
->minor_version
= orig
->minor_version
;
1266 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1267 st
->data_offset
= orig
->data_offset
;
1273 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1275 /* try each load_super to find the best match,
1276 * and return the best superswitch
1278 struct superswitch
*ss
;
1279 struct supertype
*st
;
1280 unsigned int besttime
= 0;
1284 st
= xcalloc(1, sizeof(*st
));
1285 st
->container_devnm
[0] = 0;
1287 for (i
= 0; superlist
[i
]; i
++) {
1290 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1292 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1294 memset(st
, 0, sizeof(*st
));
1295 st
->ignore_hw_compat
= 1;
1296 rv
= ss
->load_super(st
, fd
, NULL
);
1299 st
->ss
->getinfo_super(st
, &info
, NULL
);
1300 if (bestsuper
== -1 ||
1301 besttime
< info
.array
.ctime
) {
1303 besttime
= info
.array
.ctime
;
1308 if (bestsuper
!= -1) {
1310 memset(st
, 0, sizeof(*st
));
1311 st
->ignore_hw_compat
= 1;
1312 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1314 superlist
[bestsuper
]->free_super(st
);
1322 /* Return size of device in bytes */
1323 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1325 unsigned long long ldsize
;
1328 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1329 ldsize
= (unsigned long long)st
.st_size
;
1332 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1335 unsigned long dsize
;
1336 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1341 pr_err("Cannot get size of %s: %s\n",
1342 dname
, strerror(errno
));
1350 /* Return sector size of device in bytes */
1351 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1353 unsigned int sectsize
;
1355 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1357 pr_err("Cannot get sector size of %s: %s\n",
1358 dname
, strerror(errno
));
1362 *sectsizep
= sectsize
;
1366 /* Return true if this can only be a container, not a member device.
1367 * i.e. is and md device and size is zero
1369 int must_be_container(int fd
)
1371 unsigned long long size
;
1372 if (md_get_version(fd
) < 0)
1374 if (get_dev_size(fd
, NULL
, &size
) == 0)
1381 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1382 * Returns: 1 if successful
1383 * -1 for unknown partition type
1384 * 0 for other errors
1386 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1389 unsigned char empty_gpt_entry
[16]= {0};
1390 struct GPT_part_entry
*part
;
1392 unsigned long long curr_part_end
;
1393 unsigned all_partitions
, entry_size
;
1395 unsigned int sector_size
= 0;
1399 BUILD_BUG_ON(sizeof(gpt
) != 512);
1400 /* skip protective MBR */
1401 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1403 lseek(fd
, sector_size
, SEEK_SET
);
1404 /* read GPT header */
1405 if (read(fd
, &gpt
, 512) != 512)
1408 /* get the number of partition entries and the entry size */
1409 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1410 entry_size
= __le32_to_cpu(gpt
.part_size
);
1412 /* Check GPT signature*/
1413 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1417 if (all_partitions
> 1024 ||
1418 entry_size
> sizeof(buf
))
1421 part
= (struct GPT_part_entry
*)buf
;
1423 /* set offset to third block (GPT entries) */
1424 lseek(fd
, sector_size
*2, SEEK_SET
);
1425 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1426 /* read partition entry */
1427 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1430 /* is this valid partition? */
1431 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1432 /* check the last lba for the current partition */
1433 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1434 if (curr_part_end
> *endofpart
)
1435 *endofpart
= curr_part_end
;
1442 /* Sets endofpart parameter to the last block used by the last partition on the device.
1443 * Returns: 1 if successful
1444 * -1 for unknown partition type
1445 * 0 for other errors
1447 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1449 struct MBR boot_sect
;
1450 unsigned long long curr_part_end
;
1452 unsigned int sector_size
;
1457 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1460 if (read(fd
, &boot_sect
, 512) != 512)
1463 /* check MBP signature */
1464 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1466 /* found the correct signature */
1468 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1470 * Have to make every access through boot_sect rather
1471 * than using a pointer to the partition table (or an
1472 * entry), since the entries are not properly aligned.
1475 /* check for GPT type */
1476 if (boot_sect
.parts
[part_nr
].part_type
==
1477 MBR_GPT_PARTITION_TYPE
) {
1478 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1481 /* check the last used lba for the current partition */
1483 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1484 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1485 if (curr_part_end
> *endofpart
)
1486 *endofpart
= curr_part_end
;
1489 /* Unknown partition table */
1492 /* calculate number of 512-byte blocks */
1493 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1494 *endofpart
*= (sector_size
/ 512);
1499 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1500 unsigned long long size
)
1503 * Check where the last partition ends
1505 unsigned long long endofpart
;
1507 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1508 /* There appears to be a partition table here */
1509 if (freesize
== 0) {
1510 /* partitions will not be visible in new device */
1511 pr_err("partition table exists on %s but will be lost or\n"
1512 " meaningless after creating array\n",
1515 } else if (endofpart
> freesize
) {
1516 /* last partition overlaps metadata */
1517 pr_err("metadata will over-write last partition on %s.\n",
1520 } else if (size
&& endofpart
> size
) {
1521 /* partitions will be truncated in new device */
1522 pr_err("array size is too small to cover all partitions on %s.\n",
1530 int open_container(int fd
)
1532 /* 'fd' is a block device. Find out if it is in use
1533 * by a container, and return an open fd on that container.
1544 if (fstat(fd
, &st
) != 0)
1546 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1547 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1548 e
= path
+ strlen(path
);
1550 dir
= opendir(path
);
1553 while ((de
= readdir(dir
))) {
1556 if (de
->d_name
[0] == '.')
1558 /* Need to make sure it is a container and not a volume */
1559 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1560 dfd
= open(path
, O_RDONLY
);
1563 n
= read(dfd
, buf
, sizeof(buf
));
1565 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1568 if (strncmp(buf
, "external", 8) != 0 ||
1572 sprintf(e
, "/%s/dev", de
->d_name
);
1573 dfd
= open(path
, O_RDONLY
);
1576 n
= read(dfd
, buf
, sizeof(buf
));
1578 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1581 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1583 sprintf(buf
, "%d:%d", major
, minor
);
1584 dfd
= dev_open(buf
, O_RDONLY
);
1594 struct superswitch
*version_to_superswitch(char *vers
)
1598 for (i
= 0; superlist
[i
]; i
++) {
1599 struct superswitch
*ss
= superlist
[i
];
1601 if (strcmp(vers
, ss
->name
) == 0)
1608 int metadata_container_matches(char *metadata
, char *devnm
)
1610 /* Check if 'devnm' is the container named in 'metadata'
1612 * /containername/componentname or
1613 * -containername/componentname
1616 if (*metadata
!= '/' && *metadata
!= '-')
1619 if (strncmp(metadata
+1, devnm
, l
) != 0)
1621 if (metadata
[l
+1] != '/')
1626 int metadata_subdev_matches(char *metadata
, char *devnm
)
1628 /* Check if 'devnm' is the subdev named in 'metadata'
1630 * /containername/subdev or
1631 * -containername/subdev
1634 if (*metadata
!= '/' && *metadata
!= '-')
1636 sl
= strchr(metadata
+1, '/');
1639 if (strcmp(sl
+1, devnm
) == 0)
1644 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1646 if (mdstat
->metadata_version
== NULL
||
1647 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1648 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1654 int is_subarray_active(char *subarray
, char *container
)
1656 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1657 struct mdstat_ent
*ent
;
1659 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1660 if (is_container_member(ent
, container
))
1661 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1664 free_mdstat(mdstat
);
1669 /* open_subarray - opens a subarray in a container
1670 * @dev: container device name
1671 * @st: empty supertype
1672 * @quiet: block reporting errors flag
1674 * On success returns an fd to a container and fills in *st
1676 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1679 struct mdinfo
*info
;
1683 fd
= open(dev
, O_RDWR
|O_EXCL
);
1686 pr_err("Couldn't open %s, aborting\n",
1691 _devnm
= fd2devnm(fd
);
1692 if (_devnm
== NULL
) {
1694 pr_err("Failed to determine device number for %s\n",
1698 strcpy(st
->devnm
, _devnm
);
1700 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1703 pr_err("Failed to read sysfs for %s\n",
1708 if (mdi
->array
.level
!= UnSet
) {
1710 pr_err("%s is not a container\n", dev
);
1714 st
->ss
= version_to_superswitch(mdi
->text_version
);
1717 pr_err("Operation not supported for %s metadata\n",
1722 if (st
->devnm
[0] == 0) {
1724 pr_err("Failed to allocate device name\n");
1728 if (!st
->ss
->load_container
) {
1730 pr_err("%s is not a container\n", dev
);
1734 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1736 pr_err("Failed to load metadata for %s\n",
1741 info
= st
->ss
->container_content(st
, subarray
);
1744 pr_err("Failed to find subarray-%s in %s\n",
1754 st
->ss
->free_super(st
);
1767 int add_disk(int mdfd
, struct supertype
*st
,
1768 struct mdinfo
*sra
, struct mdinfo
*info
)
1770 /* Add a device to an array, in one of 2 ways. */
1773 if (st
->ss
->external
) {
1774 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1775 info
->recovery_start
= MaxSector
;
1777 info
->recovery_start
= 0;
1778 rv
= sysfs_add_disk(sra
, info
, 0);
1781 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1785 sd2
= xmalloc(sizeof(*sd2
));
1787 sd2
->next
= sra
->devs
;
1793 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1797 int remove_disk(int mdfd
, struct supertype
*st
,
1798 struct mdinfo
*sra
, struct mdinfo
*info
)
1801 /* Remove the disk given by 'info' from the array */
1803 if (st
->ss
->external
)
1804 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1807 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1812 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1814 int cnt
= force
? 500 : 5;
1817 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1818 * outstanding IO requests to the device.
1819 * In this case, it can be helpful to wait a little while,
1820 * up to 5 seconds if 'force' is set, or 50 msec if not.
1822 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1830 int sys_hot_remove_disk(int statefd
, int force
)
1832 int cnt
= force
? 500 : 5;
1835 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1839 return ret
== 6 ? 0 : -1;
1842 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1844 /* Initialise kernel's knowledge of array.
1845 * This varies between externally managed arrays
1848 int vers
= md_get_version(mdfd
);
1852 if (st
->ss
->external
)
1853 rv
= sysfs_set_array(info
, vers
);
1856 if ((vers
% 100) >= 1) { /* can use different versions */
1857 mdu_array_info_t inf
;
1858 memset(&inf
, 0, sizeof(inf
));
1859 inf
.major_version
= info
->array
.major_version
;
1860 inf
.minor_version
= info
->array
.minor_version
;
1861 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
1863 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
1867 unsigned long long min_recovery_start(struct mdinfo
*array
)
1869 /* find the minimum recovery_start in an array for metadata
1870 * formats that only record per-array recovery progress instead
1873 unsigned long long recovery_start
= MaxSector
;
1876 for (d
= array
->devs
; d
; d
= d
->next
)
1877 recovery_start
= min(recovery_start
, d
->recovery_start
);
1879 return recovery_start
;
1882 int mdmon_pid(char *devnm
)
1889 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1891 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1895 n
= read(fd
, pid
, 9);
1902 int mdmon_running(char *devnm
)
1904 int pid
= mdmon_pid(devnm
);
1907 if (kill(pid
, 0) == 0)
1912 int start_mdmon(char *devnm
)
1926 if (check_env("MDADM_NO_MDMON"))
1929 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1933 sl
= strrchr(pathbuf
, '/');
1938 strcpy(sl
, "mdmon");
1942 /* First try to run systemctl */
1943 if (!check_env("MDADM_NO_SYSTEMCTL"))
1946 /* FIXME yuk. CLOSE_EXEC?? */
1948 for (i
= 3; skipped
< 20; i
++)
1954 /* Don't want to see error messages from
1955 * systemctl. If the service doesn't exist,
1956 * we start mdmon ourselves.
1959 open("/dev/null", O_WRONLY
);
1960 snprintf(pathbuf
, sizeof(pathbuf
), "mdmon@%s.service",
1962 status
= execl("/usr/bin/systemctl", "systemctl",
1965 status
= execl("/bin/systemctl", "systemctl", "start",
1968 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1970 default: /* parent - good */
1971 pid
= wait(&status
);
1972 if (pid
>= 0 && status
== 0)
1976 /* That failed, try running mdmon directly */
1979 /* FIXME yuk. CLOSE_EXEC?? */
1981 for (i
= 3; skipped
< 20; i
++)
1987 for (i
= 0; paths
[i
]; i
++)
1989 execl(paths
[i
], paths
[i
],
1993 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1995 default: /* parent - good */
1996 pid
= wait(&status
);
1997 if (pid
< 0 || status
!= 0) {
1998 pr_err("failed to launch mdmon. Array remains readonly\n");
2005 __u32
random32(void)
2008 int rfd
= open("/dev/urandom", O_RDONLY
);
2009 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
2016 void random_uuid(__u8
*buf
)
2021 fd
= open("/dev/urandom", O_RDONLY
);
2024 len
= read(fd
, buf
, 16);
2032 for (i
= 0; i
< 4; i
++)
2038 int flush_metadata_updates(struct supertype
*st
)
2042 st
->update_tail
= NULL
;
2046 sfd
= connect_monitor(st
->container_devnm
);
2050 while (st
->updates
) {
2051 struct metadata_update
*mu
= st
->updates
;
2052 st
->updates
= mu
->next
;
2054 send_message(sfd
, mu
, 0);
2062 st
->update_tail
= NULL
;
2066 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2069 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2074 mu
->space_list
= NULL
;
2076 *st
->update_tail
= mu
;
2077 st
->update_tail
= &mu
->next
;
2079 #endif /* MDASSEMBLE */
2082 /* tinyc doesn't optimize this check in ioctl.h out ... */
2083 unsigned int __invalid_size_argument_for_IOC
= 0;
2086 int experimental(void)
2088 if (check_env("MDADM_EXPERIMENTAL"))
2091 pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n");
2096 /* Pick all spares matching given criteria from a container
2097 * if min_size == 0 do not check size
2098 * if domlist == NULL do not check domains
2099 * if spare_group given add it to domains of each spare
2100 * metadata allows to test domains using metadata of destination array */
2101 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2102 unsigned long long min_size
,
2103 struct domainlist
*domlist
,
2105 const char *metadata
, int get_one
)
2107 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2109 /* get list of all disks in container */
2110 if (st
->ss
->getinfo_super_disks
)
2111 disks
= st
->ss
->getinfo_super_disks(st
);
2115 /* find spare devices on the list */
2117 disks
->array
.spare_disks
= 0;
2121 if (d
->disk
.state
== 0) {
2122 /* check if size is acceptable */
2123 unsigned long long dev_size
;
2124 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2127 (dev_size_from_id(dev
, &dev_size
) &&
2128 dev_size
>= min_size
))
2130 /* check if domain matches */
2131 if (found
&& domlist
) {
2132 struct dev_policy
*pol
= devid_policy(dev
);
2134 pol_add(&pol
, pol_domain
,
2136 if (domain_test(domlist
, pol
, metadata
) != 1)
2138 dev_policy_free(pol
);
2143 disks
->array
.spare_disks
++;
2157 /* Checks if paths point to the same device
2158 * Returns 0 if they do.
2159 * Returns 1 if they don't.
2160 * Returns -1 if something went wrong,
2161 * e.g. paths are empty or the files
2162 * they point to don't exist */
2163 int compare_paths (char* path1
, char* path2
)
2165 struct stat st1
,st2
;
2167 if (path1
== NULL
|| path2
== NULL
)
2169 if (stat(path1
,&st1
) != 0)
2171 if (stat(path2
,&st2
) != 0)
2173 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2178 /* Make sure we can open as many devices as needed */
2179 void enable_fds(int devices
)
2181 unsigned int fds
= 20 + devices
;
2183 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0
2184 || lim
.rlim_cur
>= fds
)
2186 if (lim
.rlim_max
< fds
)
2189 setrlimit(RLIMIT_NOFILE
, &lim
);
2194 /* This is based on similar function in systemd. */
2196 /* statfs.f_type is signed long on s390x and MIPS, causing all
2197 sorts of sign extension problems with RAMFS_MAGIC being
2198 defined as 0x858458f6 */
2199 return statfs("/", &s
) >= 0 &&
2200 ((unsigned long)s
.f_type
== TMPFS_MAGIC
||
2201 ((unsigned long)s
.f_type
& 0xFFFFFFFFUL
) ==
2202 ((unsigned long)RAMFS_MAGIC
& 0xFFFFFFFFUL
));
2205 void reopen_mddev(int mdfd
)
2207 /* Re-open without any O_EXCL, but keep
2212 devnm
= fd2devnm(mdfd
);
2214 fd
= open_dev(devnm
);
2215 if (fd
>= 0 && fd
!= mdfd
)
2220 static struct cmap_hooks
*cmap_hooks
= NULL
;
2221 static int is_cmap_hooks_ready
= 0;
2223 void set_cmap_hooks(void)
2225 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2226 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2227 if (!cmap_hooks
->cmap_handle
)
2230 cmap_hooks
->initialize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2231 cmap_hooks
->get_string
= dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2232 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2234 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2235 !cmap_hooks
->finalize
)
2236 dlclose(cmap_hooks
->cmap_handle
);
2238 is_cmap_hooks_ready
= 1;
2241 int get_cluster_name(char **cluster_name
)
2244 cmap_handle_t handle
;
2246 if (!is_cmap_hooks_ready
)
2249 rv
= cmap_hooks
->initialize(&handle
);
2253 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2255 free(*cluster_name
);
2262 cmap_hooks
->finalize(handle
);
2267 void set_dlm_hooks(void)
2269 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2270 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2271 if (!dlm_hooks
->dlm_handle
)
2274 dlm_hooks
->create_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2275 dlm_hooks
->release_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2276 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2277 dlm_hooks
->ls_unlock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock");
2278 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2279 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2281 if (!dlm_hooks
->create_lockspace
|| !dlm_hooks
->ls_lock
||
2282 !dlm_hooks
->ls_unlock
|| !dlm_hooks
->release_lockspace
||
2283 !dlm_hooks
->ls_get_fd
|| !dlm_hooks
->dispatch
)
2284 dlclose(dlm_hooks
->dlm_handle
);
2286 is_dlm_hooks_ready
= 1;
2289 void set_hooks(void)