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
);
226 int md_set_array_info(int fd
, struct mdu_array_info_s
*array
)
228 return ioctl(fd
, SET_ARRAY_INFO
, array
);
232 * Get disk info from the kernel.
234 int md_get_disk_info(int fd
, struct mdu_disk_info_s
*disk
)
236 return ioctl(fd
, GET_DISK_INFO
, disk
);
240 * Parse a 128 bit uuid in 4 integers
241 * format is 32 hexx nibbles with options :.<space> separator
242 * If not exactly 32 hex digits are found, return 0
245 int parse_uuid(char *str
, int uuid
[4])
247 int hit
= 0; /* number of Hex digIT */
250 for (i
= 0; i
< 4; i
++)
253 while ((c
= *str
++) != 0) {
255 if (c
>= '0' && c
<= '9')
257 else if (c
>= 'a' && c
<= 'f')
259 else if (c
>= 'A' && c
<= 'F')
261 else if (strchr(":. -", c
))
277 * Get the md version number.
278 * We use the RAID_VERSION ioctl if it is supported
279 * If not, but we have a block device with major '9', we assume
282 * Return version number as 24 but number - assume version parts
286 int md_get_version(int fd
)
291 if (fstat(fd
, &stb
)<0)
293 if ((S_IFMT
&stb
.st_mode
) != S_IFBLK
)
296 if (ioctl(fd
, RAID_VERSION
, &vers
) == 0)
297 return (vers
.major
*10000) + (vers
.minor
*100) + vers
.patchlevel
;
300 if (major(stb
.st_rdev
) == MD_MAJOR
)
305 int get_linux_version()
309 int a
= 0, b
= 0,c
= 0;
314 a
= strtoul(cp
, &cp
, 10);
316 b
= strtoul(cp
+1, &cp
, 10);
318 c
= strtoul(cp
+1, &cp
, 10);
320 return (a
*1000000)+(b
*1000)+c
;
324 int mdadm_version(char *version
)
332 cp
= strchr(version
, '-');
333 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
336 a
= strtoul(cp
, &cp
, 10);
339 b
= strtoul(cp
+1, &cp
, 10);
341 c
= strtoul(cp
+1, &cp
, 10);
344 if (*cp
!= ' ' && *cp
!= '-')
346 return (a
*1000000)+(b
*1000)+c
;
349 unsigned long long parse_size(char *size
)
351 /* parse 'size' which should be a number optionally
352 * followed by 'K', 'M', or 'G'.
353 * Without a suffix, K is assumed.
354 * Number returned is in sectors (half-K)
355 * INVALID_SECTORS returned on error.
358 long long s
= strtoll(size
, &c
, 10);
372 s
*= 1024 * 1024 * 2;
374 case 's': /* sectors */
385 int parse_layout_10(char *layout
)
389 /* Parse the layout string for raid10 */
390 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
391 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
392 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
396 if (layout
[0] == 'n')
398 else if (layout
[0] == 'o')
399 rv
= 0x10000 + (copies
<<8) + 1;
401 rv
= 1 + (copies
<<8);
405 int parse_layout_faulty(char *layout
)
407 /* Parse the layout string for 'faulty' */
408 int ln
= strcspn(layout
, "0123456789");
409 char *m
= xstrdup(layout
);
412 mode
= map_name(faultylayout
, m
);
416 return mode
| (atoi(layout
+ln
)<< ModeShift
);
419 long parse_num(char *num
)
421 /* Either return a valid number, or -1 */
423 long rv
= strtol(num
, &c
, 10);
424 if (rv
< 0 || *c
|| !num
[0])
431 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
434 *slot
= strtoul(input
, &dev
, 10);
435 if (dev
== input
|| dev
[0] != ':')
441 void remove_partitions(int fd
)
443 /* remove partitions from this block devices.
444 * This is used for components added to an array
446 #ifdef BLKPG_DEL_PARTITION
447 struct blkpg_ioctl_arg a
;
448 struct blkpg_partition p
;
450 a
.op
= BLKPG_DEL_PARTITION
;
452 a
.datalen
= sizeof(p
);
454 memset(a
.data
, 0, a
.datalen
);
455 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
456 ioctl(fd
, BLKPG
, &a
);
460 int test_partition(int fd
)
462 /* Check if fd is a whole-disk or a partition.
463 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
464 * will return ENXIO on an invalid partition number.
466 struct blkpg_ioctl_arg a
;
467 struct blkpg_partition p
;
468 a
.op
= BLKPG_DEL_PARTITION
;
470 a
.datalen
= sizeof(p
);
472 memset(a
.data
, 0, a
.datalen
);
474 if (ioctl(fd
, BLKPG
, &a
) == 0)
475 /* Very unlikely, but not a partition */
477 if (errno
== ENXIO
|| errno
== ENOTTY
)
478 /* not a partition */
484 int test_partition_from_id(dev_t id
)
489 sprintf(buf
, "%d:%d", major(id
), minor(id
));
490 fd
= dev_open(buf
, O_RDONLY
);
493 rv
= test_partition(fd
);
498 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
504 for (i
= 0; i
< raid_disks
; i
++)
505 avail_disks
+= !!avail
[i
];
509 /* This is the tricky one - we need to check
510 * which actual disks are present.
512 copies
= (layout
&255)* ((layout
>>8) & 255);
515 /* there must be one of the 'copies' form 'first' */
522 this = (this+1) % raid_disks
;
526 first
= (first
+(layout
&255)) % raid_disks
;
527 } while (first
!= 0);
530 case LEVEL_MULTIPATH
:
531 return avail_disks
>= 1;
534 return avail_disks
== raid_disks
;
536 return avail_disks
>= 1;
538 if (avail_disks
== raid_disks
- 1 &&
539 !avail
[raid_disks
- 1])
540 /* If just the parity device is missing, then we
541 * have enough, even if not clean
547 return avail_disks
>= raid_disks
-1;
549 return avail_disks
>= raid_disks
;
552 return avail_disks
>= raid_disks
-2;
554 return avail_disks
>= raid_disks
;
560 int enough_fd(int fd
)
562 struct mdu_array_info_s array
;
563 struct mdu_disk_info_s disk
;
567 if (md_get_array_info(fd
, &array
) != 0 || array
.raid_disks
<= 0)
569 avail
= xcalloc(array
.raid_disks
, 1);
570 for (i
= 0; i
< MAX_DISKS
&& array
.nr_disks
> 0; i
++) {
572 if (md_get_disk_info(fd
, &disk
) != 0)
574 if (disk
.major
== 0 && disk
.minor
== 0)
578 if (! (disk
.state
& (1<<MD_DISK_SYNC
)))
580 if (disk
.raid_disk
< 0 || disk
.raid_disk
>= array
.raid_disks
)
582 avail
[disk
.raid_disk
] = 1;
584 /* This is used on an active array, so assume it is clean */
585 rv
= enough(array
.level
, array
.raid_disks
, array
.layout
,
591 const int uuid_zero
[4] = { 0, 0, 0, 0 };
593 int same_uuid(int a
[4], int b
[4], int swapuuid
)
596 /* parse uuids are hostendian.
597 * uuid's from some superblocks are big-ending
598 * if there is a difference, we need to swap..
600 unsigned char *ac
= (unsigned char *)a
;
601 unsigned char *bc
= (unsigned char *)b
;
603 for (i
= 0; i
< 16; i
+= 4) {
604 if (ac
[i
+0] != bc
[i
+3] ||
605 ac
[i
+1] != bc
[i
+2] ||
606 ac
[i
+2] != bc
[i
+1] ||
621 void copy_uuid(void *a
, int b
[4], int swapuuid
)
624 /* parse uuids are hostendian.
625 * uuid's from some superblocks are big-ending
626 * if there is a difference, we need to swap..
628 unsigned char *ac
= (unsigned char *)a
;
629 unsigned char *bc
= (unsigned char *)b
;
631 for (i
= 0; i
< 16; i
+= 4) {
641 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
648 copy_uuid(uuid
, id
, swap
);
649 for (i
= 0; i
< 4; i
++) {
652 for (j
= 3; j
>= 0; j
--) {
653 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
661 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
663 // dirty hack to work around an issue with super1 superblocks...
664 // super1 superblocks need swapuuid set in order for assembly to
665 // work, but can't have it set if we want this printout to match
666 // all the other uuid printouts in super1.c, so we force swapuuid
667 // to 1 to make our printout match the rest of super1
668 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
672 int check_ext2(int fd
, char *name
)
675 * Check for an ext2fs file system.
676 * Superblock is always 1K at 1K offset
678 * s_magic is le16 at 56 == 0xEF53
679 * report mtime - le32 at 44
681 * logblksize - le32 at 24
683 unsigned char sb
[1024];
685 unsigned long long size
;
687 if (lseek(fd
, 1024,0)!= 1024)
689 if (read(fd
, sb
, 1024)!= 1024)
691 if (sb
[56] != 0x53 || sb
[57] != 0xef)
694 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
695 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
696 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
698 pr_err("%s appears to contain an ext2fs file system\n",
700 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
704 int check_reiser(int fd
, char *name
)
707 * superblock is at 64K
709 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
712 unsigned char sb
[1024];
713 unsigned long long size
;
714 if (lseek(fd
, 64*1024, 0) != 64*1024)
716 if (read(fd
, sb
, 1024) != 1024)
718 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
719 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
721 pr_err("%s appears to contain a reiserfs file system\n",name
);
722 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
723 cont_err("size = %lluK\n", size
*4);
728 int check_raid(int fd
, char *name
)
733 struct supertype
*st
= guess_super(fd
);
737 if (st
->ss
->add_to_super
!= NULL
) {
738 st
->ss
->load_super(st
, fd
, name
);
739 /* Looks like a raid array .. */
740 pr_err("%s appears to be part of a raid array:\n", name
);
741 st
->ss
->getinfo_super(st
, &info
, NULL
);
742 st
->ss
->free_super(st
);
743 crtime
= info
.array
.ctime
;
744 level
= map_num(pers
, info
.array
.level
);
747 cont_err("level=%s devices=%d ctime=%s",
748 level
, info
.array
.raid_disks
, ctime(&crtime
));
750 /* Looks like GPT or MBR */
751 pr_err("partition table exists on %s\n", name
);
760 for (i
= 0; i
< 5; i
++) {
762 fprintf(stderr
, "%s%s", mesg
, add
);
764 if (fgets(buf
, 100, stdin
)==NULL
)
766 if (buf
[0]=='y' || buf
[0]=='Y')
768 if (buf
[0]=='n' || buf
[0]=='N')
772 pr_err("assuming 'no'\n");
775 #endif /* MDASSEMBLE */
777 int is_standard(char *dev
, int *nump
)
779 /* tests if dev is a "standard" md dev name.
780 * i.e if the last component is "/dNN" or "/mdNN",
781 * where NN is a string of digits
782 * Returns 1 if a partitionable standard,
783 * -1 if non-partitonable,
784 * 0 if not a standard name.
786 char *d
= strrchr(dev
, '/');
791 if (strncmp(d
, "/d",2) == 0)
792 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
793 else if (strncmp(d
, "/md_d", 5) == 0)
794 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
795 else if (strncmp(d
, "/md", 3) == 0)
796 d
+= 3, type
= -1; /* /dev/mdN */
797 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
798 d
+= 1, type
= -1; /* /dev/md/N */
808 if (nump
) *nump
= num
;
813 unsigned long calc_csum(void *super
, int bytes
)
815 unsigned long long newcsum
= 0;
818 unsigned int *superc
= (unsigned int*) super
;
820 for(i
= 0; i
< bytes
/4; i
++)
821 newcsum
+= superc
[i
];
822 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
824 /* The in-kernel checksum calculation is always 16bit on
825 * the alpha, though it is 32 bit on i386...
826 * I wonder what it is elsewhere... (it uses an API in
827 * a way that it shouldn't).
829 csum
= (csum
& 0xffff) + (csum
>> 16);
830 csum
= (csum
& 0xffff) + (csum
>> 16);
836 char *human_size(long long bytes
)
840 /* We convert bytes to either centi-M{ega,ibi}bytes or
841 * centi-G{igi,ibi}bytes, with appropriate rounding,
842 * and then print 1/100th of those as a decimal.
843 * We allow upto 2048Megabytes before converting to
844 * gigabytes, as that shows more precision and isn't
845 * too large a number.
846 * Terabytes are not yet handled.
849 if (bytes
< 5000*1024)
851 else if (bytes
< 2*1024LL*1024LL*1024LL) {
852 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
853 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
854 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
855 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
857 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
858 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
859 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
860 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
865 char *human_size_brief(long long bytes
, int prefix
)
869 /* We convert bytes to either centi-M{ega,ibi}bytes or
870 * centi-G{igi,ibi}bytes, with appropriate rounding,
871 * and then print 1/100th of those as a decimal.
872 * We allow upto 2048Megabytes before converting to
873 * gigabytes, as that shows more precision and isn't
874 * too large a number.
875 * Terabytes are not yet handled.
877 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
878 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
881 if (bytes
< 5000*1024)
883 else if (prefix
== IEC
) {
884 if (bytes
< 2*1024LL*1024LL*1024LL) {
885 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
886 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
887 cMiB
/100, cMiB
% 100);
889 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
890 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
891 cGiB
/100, cGiB
% 100);
894 else if (prefix
== JEDEC
) {
895 if (bytes
< 2*1024LL*1024LL*1024LL) {
896 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
897 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
900 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
901 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
911 void print_r10_layout(int layout
)
913 int near
= layout
& 255;
914 int far
= (layout
>> 8) & 255;
915 int offset
= (layout
&0x10000);
919 printf("%s near=%d", sep
, near
);
923 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
925 printf("NO REDUNDANCY");
929 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
930 int chunksize
, unsigned long long devsize
)
934 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
935 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
938 int get_data_disks(int level
, int layout
, int raid_disks
)
942 case 0: data_disks
= raid_disks
;
944 case 1: data_disks
= 1;
947 case 5: data_disks
= raid_disks
- 1;
949 case 6: data_disks
= raid_disks
- 2;
951 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
958 dev_t
devnm2devid(char *devnm
)
960 /* First look in /sys/block/$DEVNM/dev for %d:%d
961 * If that fails, try parsing out a number
968 sprintf(path
, "/sys/block/%s/dev", devnm
);
969 fd
= open(path
, O_RDONLY
);
972 int n
= read(fd
, buf
, sizeof(buf
));
976 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
977 return makedev(mjr
, mnr
);
979 if (strncmp(devnm
, "md_d", 4) == 0 &&
981 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
982 ep
> devnm
&& *ep
== 0)
983 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
985 if (strncmp(devnm
, "md", 2) == 0 &&
987 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
988 ep
> devnm
&& *ep
== 0)
989 return makedev(MD_MAJOR
, mnr
);
994 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
995 char *get_md_name(char *devnm
)
997 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
998 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
1000 static char devname
[50];
1002 dev_t rdev
= devnm2devid(devnm
);
1007 if (strncmp(devnm
, "md_", 3) == 0) {
1008 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
1010 if (stat(devname
, &stb
) == 0
1011 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1012 && (stb
.st_rdev
== rdev
))
1015 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
1016 if (stat(devname
, &stb
) == 0
1017 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1018 && (stb
.st_rdev
== rdev
))
1021 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
1022 if (stat(devname
, &stb
) == 0
1023 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1024 && (stb
.st_rdev
== rdev
))
1027 dn
= map_dev(major(rdev
), minor(rdev
), 0);
1030 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
1031 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
1032 if (errno
!= EEXIST
)
1035 if (stat(devname
, &stb
) == 0
1036 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1037 && (stb
.st_rdev
== rdev
))
1043 void put_md_name(char *name
)
1045 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1048 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1050 int get_maj_min(char *dev
, int *major
, int *minor
)
1053 *major
= strtoul(dev
, &e
, 0);
1054 return (e
> dev
&& *e
== ':' && e
[1] &&
1055 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1059 int dev_open(char *dev
, int flags
)
1061 /* like 'open', but if 'dev' matches %d:%d, create a temp
1062 * block device and open that
1073 if (get_maj_min(dev
, &major
, &minor
)) {
1074 snprintf(devname
, sizeof(devname
), "/dev/.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
);
1081 /* Try /tmp as /dev appear to be read-only */
1082 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
1083 (int)getpid(), major
, minor
);
1084 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1085 fd
= open(devname
, flags
);
1090 fd
= open(dev
, flags
);
1094 int open_dev_flags(char *devnm
, int flags
)
1099 devid
= devnm2devid(devnm
);
1100 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1101 return dev_open(buf
, flags
);
1104 int open_dev(char *devnm
)
1106 return open_dev_flags(devnm
, O_RDONLY
);
1109 int open_dev_excl(char *devnm
)
1114 dev_t devid
= devnm2devid(devnm
);
1117 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1118 for (i
= 0; i
< 25; i
++) {
1119 int fd
= dev_open(buf
, flags
|O_EXCL
);
1122 if (errno
== EACCES
&& flags
== O_RDWR
) {
1135 int same_dev(char *one
, char *two
)
1137 struct stat st1
, st2
;
1138 if (stat(one
, &st1
) != 0)
1140 if (stat(two
, &st2
) != 0)
1142 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1144 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1146 return st1
.st_rdev
== st2
.st_rdev
;
1149 void wait_for(char *dev
, int fd
)
1152 struct stat stb_want
;
1155 if (fstat(fd
, &stb_want
) != 0 ||
1156 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1159 for (i
= 0; i
< 25; i
++) {
1161 if (stat(dev
, &stb
) == 0 &&
1162 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1163 (stb
.st_rdev
== stb_want
.st_rdev
))
1170 dprintf("timeout waiting for %s\n", dev
);
1173 struct superswitch
*superlist
[] =
1176 &super_ddf
, &super_imsm
,
1180 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
1182 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1184 mdu_array_info_t array
;
1187 struct supertype
*st
= NULL
;
1192 char *subarray
= NULL
;
1193 char container
[32] = "";
1195 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1198 vers
= sra
->array
.major_version
;
1199 minor
= sra
->array
.minor_version
;
1200 verstr
= sra
->text_version
;
1202 if (md_get_array_info(fd
, &array
))
1203 array
.major_version
= array
.minor_version
= 0;
1204 vers
= array
.major_version
;
1205 minor
= array
.minor_version
;
1210 sprintf(version
, "%d.%d", vers
, minor
);
1213 if (minor
== -2 && is_subarray(verstr
)) {
1214 char *dev
= verstr
+1;
1216 subarray
= strchr(dev
, '/');
1219 subarray
= xstrdup(subarray
);
1221 strcpy(container
, dev
);
1223 sra
= sysfs_read(-1, container
, GET_VERSION
);
1224 if (sra
&& sra
->text_version
[0])
1225 verstr
= sra
->text_version
;
1227 verstr
= "-no-metadata-";
1230 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1231 st
= superlist
[i
]->match_metadata_desc(verstr
);
1237 *subarrayp
= subarray
;
1238 strcpy(st
->container_devnm
, container
);
1239 strcpy(st
->devnm
, fd2devnm(fd
));
1245 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1247 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1252 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1253 fd
= dev_open(buf
, O_RDONLY
);
1256 if (get_dev_size(fd
, NULL
, size
)) {
1264 struct supertype
*dup_super(struct supertype
*orig
)
1266 struct supertype
*st
;
1270 st
= xcalloc(1, sizeof(*st
));
1272 st
->max_devs
= orig
->max_devs
;
1273 st
->minor_version
= orig
->minor_version
;
1274 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1275 st
->data_offset
= orig
->data_offset
;
1281 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1283 /* try each load_super to find the best match,
1284 * and return the best superswitch
1286 struct superswitch
*ss
;
1287 struct supertype
*st
;
1288 unsigned int besttime
= 0;
1292 st
= xcalloc(1, sizeof(*st
));
1293 st
->container_devnm
[0] = 0;
1295 for (i
= 0; superlist
[i
]; i
++) {
1298 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1300 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1302 memset(st
, 0, sizeof(*st
));
1303 st
->ignore_hw_compat
= 1;
1304 rv
= ss
->load_super(st
, fd
, NULL
);
1307 st
->ss
->getinfo_super(st
, &info
, NULL
);
1308 if (bestsuper
== -1 ||
1309 besttime
< info
.array
.ctime
) {
1311 besttime
= info
.array
.ctime
;
1316 if (bestsuper
!= -1) {
1318 memset(st
, 0, sizeof(*st
));
1319 st
->ignore_hw_compat
= 1;
1320 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1322 superlist
[bestsuper
]->free_super(st
);
1330 /* Return size of device in bytes */
1331 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1333 unsigned long long ldsize
;
1336 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1337 ldsize
= (unsigned long long)st
.st_size
;
1340 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1343 unsigned long dsize
;
1344 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1349 pr_err("Cannot get size of %s: %s\n",
1350 dname
, strerror(errno
));
1358 /* Return sector size of device in bytes */
1359 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1361 unsigned int sectsize
;
1363 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1365 pr_err("Cannot get sector size of %s: %s\n",
1366 dname
, strerror(errno
));
1370 *sectsizep
= sectsize
;
1374 /* Return true if this can only be a container, not a member device.
1375 * i.e. is and md device and size is zero
1377 int must_be_container(int fd
)
1379 unsigned long long size
;
1380 if (md_get_version(fd
) < 0)
1382 if (get_dev_size(fd
, NULL
, &size
) == 0)
1389 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1390 * Returns: 1 if successful
1391 * -1 for unknown partition type
1392 * 0 for other errors
1394 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1397 unsigned char empty_gpt_entry
[16]= {0};
1398 struct GPT_part_entry
*part
;
1400 unsigned long long curr_part_end
;
1401 unsigned all_partitions
, entry_size
;
1403 unsigned int sector_size
= 0;
1407 BUILD_BUG_ON(sizeof(gpt
) != 512);
1408 /* skip protective MBR */
1409 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1411 lseek(fd
, sector_size
, SEEK_SET
);
1412 /* read GPT header */
1413 if (read(fd
, &gpt
, 512) != 512)
1416 /* get the number of partition entries and the entry size */
1417 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1418 entry_size
= __le32_to_cpu(gpt
.part_size
);
1420 /* Check GPT signature*/
1421 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1425 if (all_partitions
> 1024 ||
1426 entry_size
> sizeof(buf
))
1429 part
= (struct GPT_part_entry
*)buf
;
1431 /* set offset to third block (GPT entries) */
1432 lseek(fd
, sector_size
*2, SEEK_SET
);
1433 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1434 /* read partition entry */
1435 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1438 /* is this valid partition? */
1439 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1440 /* check the last lba for the current partition */
1441 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1442 if (curr_part_end
> *endofpart
)
1443 *endofpart
= curr_part_end
;
1450 /* Sets endofpart parameter to the last block used by the last partition on the device.
1451 * Returns: 1 if successful
1452 * -1 for unknown partition type
1453 * 0 for other errors
1455 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1457 struct MBR boot_sect
;
1458 unsigned long long curr_part_end
;
1460 unsigned int sector_size
;
1465 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1468 if (read(fd
, &boot_sect
, 512) != 512)
1471 /* check MBP signature */
1472 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1474 /* found the correct signature */
1476 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1478 * Have to make every access through boot_sect rather
1479 * than using a pointer to the partition table (or an
1480 * entry), since the entries are not properly aligned.
1483 /* check for GPT type */
1484 if (boot_sect
.parts
[part_nr
].part_type
==
1485 MBR_GPT_PARTITION_TYPE
) {
1486 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1489 /* check the last used lba for the current partition */
1491 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1492 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1493 if (curr_part_end
> *endofpart
)
1494 *endofpart
= curr_part_end
;
1497 /* Unknown partition table */
1500 /* calculate number of 512-byte blocks */
1501 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1502 *endofpart
*= (sector_size
/ 512);
1507 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1508 unsigned long long size
)
1511 * Check where the last partition ends
1513 unsigned long long endofpart
;
1515 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1516 /* There appears to be a partition table here */
1517 if (freesize
== 0) {
1518 /* partitions will not be visible in new device */
1519 pr_err("partition table exists on %s but will be lost or\n"
1520 " meaningless after creating array\n",
1523 } else if (endofpart
> freesize
) {
1524 /* last partition overlaps metadata */
1525 pr_err("metadata will over-write last partition on %s.\n",
1528 } else if (size
&& endofpart
> size
) {
1529 /* partitions will be truncated in new device */
1530 pr_err("array size is too small to cover all partitions on %s.\n",
1538 int open_container(int fd
)
1540 /* 'fd' is a block device. Find out if it is in use
1541 * by a container, and return an open fd on that container.
1552 if (fstat(fd
, &st
) != 0)
1554 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1555 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1556 e
= path
+ strlen(path
);
1558 dir
= opendir(path
);
1561 while ((de
= readdir(dir
))) {
1564 if (de
->d_name
[0] == '.')
1566 /* Need to make sure it is a container and not a volume */
1567 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1568 dfd
= open(path
, O_RDONLY
);
1571 n
= read(dfd
, buf
, sizeof(buf
));
1573 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1576 if (strncmp(buf
, "external", 8) != 0 ||
1580 sprintf(e
, "/%s/dev", de
->d_name
);
1581 dfd
= open(path
, O_RDONLY
);
1584 n
= read(dfd
, buf
, sizeof(buf
));
1586 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1589 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1591 sprintf(buf
, "%d:%d", major
, minor
);
1592 dfd
= dev_open(buf
, O_RDONLY
);
1602 struct superswitch
*version_to_superswitch(char *vers
)
1606 for (i
= 0; superlist
[i
]; i
++) {
1607 struct superswitch
*ss
= superlist
[i
];
1609 if (strcmp(vers
, ss
->name
) == 0)
1616 int metadata_container_matches(char *metadata
, char *devnm
)
1618 /* Check if 'devnm' is the container named in 'metadata'
1620 * /containername/componentname or
1621 * -containername/componentname
1624 if (*metadata
!= '/' && *metadata
!= '-')
1627 if (strncmp(metadata
+1, devnm
, l
) != 0)
1629 if (metadata
[l
+1] != '/')
1634 int metadata_subdev_matches(char *metadata
, char *devnm
)
1636 /* Check if 'devnm' is the subdev named in 'metadata'
1638 * /containername/subdev or
1639 * -containername/subdev
1642 if (*metadata
!= '/' && *metadata
!= '-')
1644 sl
= strchr(metadata
+1, '/');
1647 if (strcmp(sl
+1, devnm
) == 0)
1652 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1654 if (mdstat
->metadata_version
== NULL
||
1655 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1656 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1662 int is_subarray_active(char *subarray
, char *container
)
1664 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1665 struct mdstat_ent
*ent
;
1667 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1668 if (is_container_member(ent
, container
))
1669 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1672 free_mdstat(mdstat
);
1677 /* open_subarray - opens a subarray in a container
1678 * @dev: container device name
1679 * @st: empty supertype
1680 * @quiet: block reporting errors flag
1682 * On success returns an fd to a container and fills in *st
1684 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1687 struct mdinfo
*info
;
1691 fd
= open(dev
, O_RDWR
|O_EXCL
);
1694 pr_err("Couldn't open %s, aborting\n",
1699 _devnm
= fd2devnm(fd
);
1700 if (_devnm
== NULL
) {
1702 pr_err("Failed to determine device number for %s\n",
1706 strcpy(st
->devnm
, _devnm
);
1708 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1711 pr_err("Failed to read sysfs for %s\n",
1716 if (mdi
->array
.level
!= UnSet
) {
1718 pr_err("%s is not a container\n", dev
);
1722 st
->ss
= version_to_superswitch(mdi
->text_version
);
1725 pr_err("Operation not supported for %s metadata\n",
1730 if (st
->devnm
[0] == 0) {
1732 pr_err("Failed to allocate device name\n");
1736 if (!st
->ss
->load_container
) {
1738 pr_err("%s is not a container\n", dev
);
1742 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1744 pr_err("Failed to load metadata for %s\n",
1749 info
= st
->ss
->container_content(st
, subarray
);
1752 pr_err("Failed to find subarray-%s in %s\n",
1762 st
->ss
->free_super(st
);
1775 int add_disk(int mdfd
, struct supertype
*st
,
1776 struct mdinfo
*sra
, struct mdinfo
*info
)
1778 /* Add a device to an array, in one of 2 ways. */
1781 if (st
->ss
->external
) {
1782 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1783 info
->recovery_start
= MaxSector
;
1785 info
->recovery_start
= 0;
1786 rv
= sysfs_add_disk(sra
, info
, 0);
1789 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1793 sd2
= xmalloc(sizeof(*sd2
));
1795 sd2
->next
= sra
->devs
;
1801 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1805 int remove_disk(int mdfd
, struct supertype
*st
,
1806 struct mdinfo
*sra
, struct mdinfo
*info
)
1809 /* Remove the disk given by 'info' from the array */
1811 if (st
->ss
->external
)
1812 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1815 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1820 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1822 int cnt
= force
? 500 : 5;
1825 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1826 * outstanding IO requests to the device.
1827 * In this case, it can be helpful to wait a little while,
1828 * up to 5 seconds if 'force' is set, or 50 msec if not.
1830 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1838 int sys_hot_remove_disk(int statefd
, int force
)
1840 int cnt
= force
? 500 : 5;
1843 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1847 return ret
== 6 ? 0 : -1;
1850 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1852 /* Initialise kernel's knowledge of array.
1853 * This varies between externally managed arrays
1856 int vers
= md_get_version(mdfd
);
1860 if (st
->ss
->external
)
1861 rv
= sysfs_set_array(info
, vers
);
1864 if ((vers
% 100) >= 1) { /* can use different versions */
1865 mdu_array_info_t inf
;
1866 memset(&inf
, 0, sizeof(inf
));
1867 inf
.major_version
= info
->array
.major_version
;
1868 inf
.minor_version
= info
->array
.minor_version
;
1869 rv
= md_set_array_info(mdfd
, &inf
);
1871 rv
= md_set_array_info(mdfd
, NULL
);
1875 unsigned long long min_recovery_start(struct mdinfo
*array
)
1877 /* find the minimum recovery_start in an array for metadata
1878 * formats that only record per-array recovery progress instead
1881 unsigned long long recovery_start
= MaxSector
;
1884 for (d
= array
->devs
; d
; d
= d
->next
)
1885 recovery_start
= min(recovery_start
, d
->recovery_start
);
1887 return recovery_start
;
1890 int mdmon_pid(char *devnm
)
1897 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1899 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1903 n
= read(fd
, pid
, 9);
1910 int mdmon_running(char *devnm
)
1912 int pid
= mdmon_pid(devnm
);
1915 if (kill(pid
, 0) == 0)
1920 int start_mdmon(char *devnm
)
1934 if (check_env("MDADM_NO_MDMON"))
1937 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1941 sl
= strrchr(pathbuf
, '/');
1946 strcpy(sl
, "mdmon");
1950 /* First try to run systemctl */
1951 if (!check_env("MDADM_NO_SYSTEMCTL"))
1954 /* FIXME yuk. CLOSE_EXEC?? */
1956 for (i
= 3; skipped
< 20; i
++)
1962 /* Don't want to see error messages from
1963 * systemctl. If the service doesn't exist,
1964 * we start mdmon ourselves.
1967 open("/dev/null", O_WRONLY
);
1968 snprintf(pathbuf
, sizeof(pathbuf
), "mdmon@%s.service",
1970 status
= execl("/usr/bin/systemctl", "systemctl",
1973 status
= execl("/bin/systemctl", "systemctl", "start",
1976 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1978 default: /* parent - good */
1979 pid
= wait(&status
);
1980 if (pid
>= 0 && status
== 0)
1984 /* That failed, try running mdmon directly */
1987 /* FIXME yuk. CLOSE_EXEC?? */
1989 for (i
= 3; skipped
< 20; i
++)
1995 for (i
= 0; paths
[i
]; i
++)
1997 execl(paths
[i
], paths
[i
],
2001 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
2003 default: /* parent - good */
2004 pid
= wait(&status
);
2005 if (pid
< 0 || status
!= 0) {
2006 pr_err("failed to launch mdmon. Array remains readonly\n");
2013 __u32
random32(void)
2016 int rfd
= open("/dev/urandom", O_RDONLY
);
2017 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
2024 void random_uuid(__u8
*buf
)
2029 fd
= open("/dev/urandom", O_RDONLY
);
2032 len
= read(fd
, buf
, 16);
2040 for (i
= 0; i
< 4; i
++)
2046 int flush_metadata_updates(struct supertype
*st
)
2050 st
->update_tail
= NULL
;
2054 sfd
= connect_monitor(st
->container_devnm
);
2058 while (st
->updates
) {
2059 struct metadata_update
*mu
= st
->updates
;
2060 st
->updates
= mu
->next
;
2062 send_message(sfd
, mu
, 0);
2070 st
->update_tail
= NULL
;
2074 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2077 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2082 mu
->space_list
= NULL
;
2084 *st
->update_tail
= mu
;
2085 st
->update_tail
= &mu
->next
;
2087 #endif /* MDASSEMBLE */
2090 /* tinyc doesn't optimize this check in ioctl.h out ... */
2091 unsigned int __invalid_size_argument_for_IOC
= 0;
2094 int experimental(void)
2096 if (check_env("MDADM_EXPERIMENTAL"))
2099 pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n");
2104 /* Pick all spares matching given criteria from a container
2105 * if min_size == 0 do not check size
2106 * if domlist == NULL do not check domains
2107 * if spare_group given add it to domains of each spare
2108 * metadata allows to test domains using metadata of destination array */
2109 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2110 unsigned long long min_size
,
2111 struct domainlist
*domlist
,
2113 const char *metadata
, int get_one
)
2115 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2117 /* get list of all disks in container */
2118 if (st
->ss
->getinfo_super_disks
)
2119 disks
= st
->ss
->getinfo_super_disks(st
);
2123 /* find spare devices on the list */
2125 disks
->array
.spare_disks
= 0;
2129 if (d
->disk
.state
== 0) {
2130 /* check if size is acceptable */
2131 unsigned long long dev_size
;
2132 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2135 (dev_size_from_id(dev
, &dev_size
) &&
2136 dev_size
>= min_size
))
2138 /* check if domain matches */
2139 if (found
&& domlist
) {
2140 struct dev_policy
*pol
= devid_policy(dev
);
2142 pol_add(&pol
, pol_domain
,
2144 if (domain_test(domlist
, pol
, metadata
) != 1)
2146 dev_policy_free(pol
);
2151 disks
->array
.spare_disks
++;
2165 /* Checks if paths point to the same device
2166 * Returns 0 if they do.
2167 * Returns 1 if they don't.
2168 * Returns -1 if something went wrong,
2169 * e.g. paths are empty or the files
2170 * they point to don't exist */
2171 int compare_paths (char* path1
, char* path2
)
2173 struct stat st1
,st2
;
2175 if (path1
== NULL
|| path2
== NULL
)
2177 if (stat(path1
,&st1
) != 0)
2179 if (stat(path2
,&st2
) != 0)
2181 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2186 /* Make sure we can open as many devices as needed */
2187 void enable_fds(int devices
)
2189 unsigned int fds
= 20 + devices
;
2191 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0
2192 || lim
.rlim_cur
>= fds
)
2194 if (lim
.rlim_max
< fds
)
2197 setrlimit(RLIMIT_NOFILE
, &lim
);
2202 /* This is based on similar function in systemd. */
2204 /* statfs.f_type is signed long on s390x and MIPS, causing all
2205 sorts of sign extension problems with RAMFS_MAGIC being
2206 defined as 0x858458f6 */
2207 return statfs("/", &s
) >= 0 &&
2208 ((unsigned long)s
.f_type
== TMPFS_MAGIC
||
2209 ((unsigned long)s
.f_type
& 0xFFFFFFFFUL
) ==
2210 ((unsigned long)RAMFS_MAGIC
& 0xFFFFFFFFUL
));
2213 void reopen_mddev(int mdfd
)
2215 /* Re-open without any O_EXCL, but keep
2220 devnm
= fd2devnm(mdfd
);
2222 fd
= open_dev(devnm
);
2223 if (fd
>= 0 && fd
!= mdfd
)
2228 static struct cmap_hooks
*cmap_hooks
= NULL
;
2229 static int is_cmap_hooks_ready
= 0;
2231 void set_cmap_hooks(void)
2233 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2234 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2235 if (!cmap_hooks
->cmap_handle
)
2238 cmap_hooks
->initialize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2239 cmap_hooks
->get_string
= dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2240 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2242 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2243 !cmap_hooks
->finalize
)
2244 dlclose(cmap_hooks
->cmap_handle
);
2246 is_cmap_hooks_ready
= 1;
2249 int get_cluster_name(char **cluster_name
)
2252 cmap_handle_t handle
;
2254 if (!is_cmap_hooks_ready
)
2257 rv
= cmap_hooks
->initialize(&handle
);
2261 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2263 free(*cluster_name
);
2270 cmap_hooks
->finalize(handle
);
2275 void set_dlm_hooks(void)
2277 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2278 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2279 if (!dlm_hooks
->dlm_handle
)
2282 dlm_hooks
->create_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2283 dlm_hooks
->release_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2284 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2285 dlm_hooks
->ls_unlock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock");
2286 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2287 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2289 if (!dlm_hooks
->create_lockspace
|| !dlm_hooks
->ls_lock
||
2290 !dlm_hooks
->ls_unlock
|| !dlm_hooks
->release_lockspace
||
2291 !dlm_hooks
->ls_get_fd
|| !dlm_hooks
->dispatch
)
2292 dlclose(dlm_hooks
->dlm_handle
);
2294 is_dlm_hooks_ready
= 1;
2297 void set_hooks(void)