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>
34 #include <linux/magic.h>
43 * following taken from linux/blkpg.h because they aren't
44 * anywhere else and it isn't safe to #include linux/ * stuff.
47 #define BLKPG _IO(0x12,105)
49 /* The argument structure */
50 struct blkpg_ioctl_arg
{
57 /* The subfunctions (for the op field) */
58 #define BLKPG_ADD_PARTITION 1
59 #define BLKPG_DEL_PARTITION 2
61 /* Sizes of name fields. Unused at present. */
62 #define BLKPG_DEVNAMELTH 64
63 #define BLKPG_VOLNAMELTH 64
65 /* The data structure for ADD_PARTITION and DEL_PARTITION */
66 struct blkpg_partition
{
67 long long start
; /* starting offset in bytes */
68 long long length
; /* length in bytes */
69 int pno
; /* partition number */
70 char devname
[BLKPG_DEVNAMELTH
]; /* partition name, like sda5 or c0d1p2,
71 to be used in kernel messages */
72 char volname
[BLKPG_VOLNAMELTH
]; /* volume label */
77 /* Force a compilation error if condition is true */
78 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
80 /* Force a compilation error if condition is true, but also produce a
81 result (of value 0 and type size_t), so the expression can be used
82 e.g. in a structure initializer (or where-ever else comma expressions
84 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
86 static int is_dlm_hooks_ready
= 0;
88 int dlm_funs_ready(void)
90 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
,
152 &dlm_lock_res
->lksb
, flags
, str
, strlen(str
),
153 0, dlm_ast
, 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",
189 /* XXX make sure the lockspace is unlocked eventually */
193 ret
= dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
195 pr_err("error %d happened when release lockspace\n", errno
);
196 /* XXX make sure the lockspace is released eventually */
205 int md_array_valid(int fd
)
210 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
212 if (sra
->array_state
!= ARRAY_UNKNOWN_STATE
)
220 * GET_ARRAY_INFO doesn't provide access to the proper state
221 * information, so fallback to a basic check for raid_disks != 0
223 ret
= ioctl(fd
, RAID_VERSION
);
229 int md_array_active(int fd
)
232 struct mdu_array_info_s array
;
235 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
237 if (!md_array_is_active(sra
))
243 * GET_ARRAY_INFO doesn't provide access to the proper state
244 * information, so fallback to a basic check for raid_disks != 0
246 ret
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
252 int md_array_is_active(struct mdinfo
*info
)
254 return (info
->array_state
!= ARRAY_CLEAR
&&
255 info
->array_state
!= ARRAY_INACTIVE
&&
256 info
->array_state
!= ARRAY_UNKNOWN_STATE
);
260 * Get array info from the kernel. Longer term we want to deprecate the
261 * ioctl and get it from sysfs.
263 int md_get_array_info(int fd
, struct mdu_array_info_s
*array
)
265 return ioctl(fd
, GET_ARRAY_INFO
, array
);
271 int md_set_array_info(int fd
, struct mdu_array_info_s
*array
)
273 return ioctl(fd
, SET_ARRAY_INFO
, array
);
277 * Get disk info from the kernel.
279 int md_get_disk_info(int fd
, struct mdu_disk_info_s
*disk
)
281 return ioctl(fd
, GET_DISK_INFO
, disk
);
285 * Parse a 128 bit uuid in 4 integers
286 * format is 32 hexx nibbles with options :.<space> separator
287 * If not exactly 32 hex digits are found, return 0
290 int parse_uuid(char *str
, int uuid
[4])
292 int hit
= 0; /* number of Hex digIT */
295 for (i
= 0; i
< 4; i
++)
298 while ((c
= *str
++) != 0) {
300 if (c
>= '0' && c
<= '9')
302 else if (c
>= 'a' && c
<= 'f')
304 else if (c
>= 'A' && c
<= 'F')
306 else if (strchr(":. -", c
))
321 int get_linux_version()
325 int a
= 0, b
= 0,c
= 0;
330 a
= strtoul(cp
, &cp
, 10);
332 b
= strtoul(cp
+1, &cp
, 10);
334 c
= strtoul(cp
+1, &cp
, 10);
336 return (a
*1000000)+(b
*1000)+c
;
339 int mdadm_version(char *version
)
347 cp
= strchr(version
, '-');
348 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
351 a
= strtoul(cp
, &cp
, 10);
354 b
= strtoul(cp
+1, &cp
, 10);
356 c
= strtoul(cp
+1, &cp
, 10);
359 if (*cp
!= ' ' && *cp
!= '-')
361 return (a
*1000000)+(b
*1000)+c
;
364 unsigned long long parse_size(char *size
)
366 /* parse 'size' which should be a number optionally
367 * followed by 'K', 'M', or 'G'.
368 * Without a suffix, K is assumed.
369 * Number returned is in sectors (half-K)
370 * INVALID_SECTORS returned on error.
373 long long s
= strtoll(size
, &c
, 10);
387 s
*= 1024 * 1024 * 2;
389 case 's': /* sectors */
400 int is_near_layout_10(int layout
)
404 fc
= (layout
>> 8) & 255;
405 fo
= layout
& (1 << 16);
406 if (fc
> 1 || fo
> 0)
411 int parse_layout_10(char *layout
)
415 /* Parse the layout string for raid10 */
416 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
417 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
418 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
422 if (layout
[0] == 'n')
424 else if (layout
[0] == 'o')
425 rv
= 0x10000 + (copies
<<8) + 1;
427 rv
= 1 + (copies
<<8);
431 int parse_layout_faulty(char *layout
)
433 /* Parse the layout string for 'faulty' */
434 int ln
= strcspn(layout
, "0123456789");
435 char *m
= xstrdup(layout
);
438 mode
= map_name(faultylayout
, m
);
442 return mode
| (atoi(layout
+ln
)<< ModeShift
);
445 long parse_num(char *num
)
447 /* Either return a valid number, or -1 */
449 long rv
= strtol(num
, &c
, 10);
450 if (rv
< 0 || *c
|| !num
[0])
456 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
459 *slot
= strtoul(input
, &dev
, 10);
460 if (dev
== input
|| dev
[0] != ':')
466 void remove_partitions(int fd
)
468 /* remove partitions from this block devices.
469 * This is used for components added to an array
471 #ifdef BLKPG_DEL_PARTITION
472 struct blkpg_ioctl_arg a
;
473 struct blkpg_partition p
;
475 a
.op
= BLKPG_DEL_PARTITION
;
477 a
.datalen
= sizeof(p
);
479 memset(a
.data
, 0, a
.datalen
);
480 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
481 ioctl(fd
, BLKPG
, &a
);
485 int test_partition(int fd
)
487 /* Check if fd is a whole-disk or a partition.
488 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
489 * will return ENXIO on an invalid partition number.
491 struct blkpg_ioctl_arg a
;
492 struct blkpg_partition p
;
493 a
.op
= BLKPG_DEL_PARTITION
;
495 a
.datalen
= sizeof(p
);
497 memset(a
.data
, 0, a
.datalen
);
499 if (ioctl(fd
, BLKPG
, &a
) == 0)
500 /* Very unlikely, but not a partition */
502 if (errno
== ENXIO
|| errno
== ENOTTY
)
503 /* not a partition */
509 int test_partition_from_id(dev_t id
)
514 sprintf(buf
, "%d:%d", major(id
), minor(id
));
515 fd
= dev_open(buf
, O_RDONLY
);
518 rv
= test_partition(fd
);
523 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
529 for (i
= 0; i
< raid_disks
; i
++)
530 avail_disks
+= !!avail
[i
];
534 /* This is the tricky one - we need to check
535 * which actual disks are present.
537 copies
= (layout
&255)* ((layout
>>8) & 255);
540 /* there must be one of the 'copies' form 'first' */
547 this = (this+1) % raid_disks
;
551 first
= (first
+(layout
&255)) % raid_disks
;
552 } while (first
!= 0);
555 case LEVEL_MULTIPATH
:
556 return avail_disks
>= 1;
559 return avail_disks
== raid_disks
;
561 return avail_disks
>= 1;
563 if (avail_disks
== raid_disks
- 1 &&
564 !avail
[raid_disks
- 1])
565 /* If just the parity device is missing, then we
566 * have enough, even if not clean
572 return avail_disks
>= raid_disks
-1;
574 return avail_disks
>= raid_disks
;
577 return avail_disks
>= raid_disks
-2;
579 return avail_disks
>= raid_disks
;
585 const int uuid_zero
[4] = { 0, 0, 0, 0 };
587 int same_uuid(int a
[4], int b
[4], int swapuuid
)
590 /* parse uuids are hostendian.
591 * uuid's from some superblocks are big-ending
592 * if there is a difference, we need to swap..
594 unsigned char *ac
= (unsigned char *)a
;
595 unsigned char *bc
= (unsigned char *)b
;
597 for (i
= 0; i
< 16; i
+= 4) {
598 if (ac
[i
+0] != bc
[i
+3] ||
599 ac
[i
+1] != bc
[i
+2] ||
600 ac
[i
+2] != bc
[i
+1] ||
615 void copy_uuid(void *a
, int b
[4], int swapuuid
)
618 /* parse uuids are hostendian.
619 * uuid's from some superblocks are big-ending
620 * if there is a difference, we need to swap..
622 unsigned char *ac
= (unsigned char *)a
;
623 unsigned char *bc
= (unsigned char *)b
;
625 for (i
= 0; i
< 16; i
+= 4) {
635 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
642 copy_uuid(uuid
, id
, swap
);
643 for (i
= 0; i
< 4; i
++) {
646 for (j
= 3; j
>= 0; j
--) {
647 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
655 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
,
658 // dirty hack to work around an issue with super1 superblocks...
659 // super1 superblocks need swapuuid set in order for assembly to
660 // work, but can't have it set if we want this printout to match
661 // all the other uuid printouts in super1.c, so we force swapuuid
662 // to 1 to make our printout match the rest of super1
663 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 :
664 st
->ss
->swapuuid
, buf
, sep
);
667 int check_ext2(int fd
, char *name
)
670 * Check for an ext2fs file system.
671 * Superblock is always 1K at 1K offset
673 * s_magic is le16 at 56 == 0xEF53
674 * report mtime - le32 at 44
676 * logblksize - le32 at 24
678 unsigned char sb
[1024];
680 unsigned long long size
;
682 if (lseek(fd
, 1024,0)!= 1024)
684 if (read(fd
, sb
, 1024)!= 1024)
686 if (sb
[56] != 0x53 || sb
[57] != 0xef)
689 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
690 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
691 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
693 pr_err("%s appears to contain an ext2fs file system\n",
695 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
699 int check_reiser(int fd
, char *name
)
702 * superblock is at 64K
704 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
707 unsigned char sb
[1024];
708 unsigned long long size
;
709 if (lseek(fd
, 64*1024, 0) != 64*1024)
711 if (read(fd
, sb
, 1024) != 1024)
713 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
714 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
716 pr_err("%s appears to contain a reiserfs file system\n",name
);
717 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
718 cont_err("size = %lluK\n", size
*4);
723 int check_raid(int fd
, char *name
)
728 struct supertype
*st
= guess_super(fd
);
732 if (st
->ss
->add_to_super
!= NULL
) {
733 st
->ss
->load_super(st
, fd
, name
);
734 /* Looks like a raid array .. */
735 pr_err("%s appears to be part of a raid array:\n", name
);
736 st
->ss
->getinfo_super(st
, &info
, NULL
);
737 st
->ss
->free_super(st
);
738 crtime
= info
.array
.ctime
;
739 level
= map_num(pers
, info
.array
.level
);
742 cont_err("level=%s devices=%d ctime=%s",
743 level
, info
.array
.raid_disks
, ctime(&crtime
));
745 /* Looks like GPT or MBR */
746 pr_err("partition table exists on %s\n", name
);
751 int fstat_is_blkdev(int fd
, char *devname
, dev_t
*rdev
)
755 if (fstat(fd
, &stb
) != 0) {
756 pr_err("fstat failed for %s: %s\n", devname
, strerror(errno
));
759 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
760 pr_err("%s is not a block device.\n", devname
);
768 int stat_is_blkdev(char *devname
, dev_t
*rdev
)
772 if (stat(devname
, &stb
) != 0) {
773 pr_err("stat failed for %s: %s\n", devname
, strerror(errno
));
776 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
777 pr_err("%s is not a block device.\n", devname
);
789 for (i
= 0; i
< 5; i
++) {
791 fprintf(stderr
, "%s%s", mesg
, add
);
793 if (fgets(buf
, 100, stdin
)==NULL
)
795 if (buf
[0]=='y' || buf
[0]=='Y')
797 if (buf
[0]=='n' || buf
[0]=='N')
801 pr_err("assuming 'no'\n");
805 int is_standard(char *dev
, int *nump
)
807 /* tests if dev is a "standard" md dev name.
808 * i.e if the last component is "/dNN" or "/mdNN",
809 * where NN is a string of digits
810 * Returns 1 if a partitionable standard,
811 * -1 if non-partitonable,
812 * 0 if not a standard name.
814 char *d
= strrchr(dev
, '/');
819 if (strncmp(d
, "/d",2) == 0)
820 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
821 else if (strncmp(d
, "/md_d", 5) == 0)
822 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
823 else if (strncmp(d
, "/md", 3) == 0)
824 d
+= 3, type
= -1; /* /dev/mdN */
825 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
826 d
+= 1, type
= -1; /* /dev/md/N */
836 if (nump
) *nump
= num
;
841 unsigned long calc_csum(void *super
, int bytes
)
843 unsigned long long newcsum
= 0;
846 unsigned int *superc
= (unsigned int*) super
;
848 for(i
= 0; i
< bytes
/4; i
++)
849 newcsum
+= superc
[i
];
850 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
852 /* The in-kernel checksum calculation is always 16bit on
853 * the alpha, though it is 32 bit on i386...
854 * I wonder what it is elsewhere... (it uses an API in
855 * a way that it shouldn't).
857 csum
= (csum
& 0xffff) + (csum
>> 16);
858 csum
= (csum
& 0xffff) + (csum
>> 16);
863 char *human_size(long long bytes
)
867 /* We convert bytes to either centi-M{ega,ibi}bytes or
868 * centi-G{igi,ibi}bytes, with appropriate rounding,
869 * and then print 1/100th of those as a decimal.
870 * We allow upto 2048Megabytes before converting to
871 * gigabytes, as that shows more precision and isn't
872 * too large a number.
873 * Terabytes are not yet handled.
876 if (bytes
< 5000*1024)
878 else if (bytes
< 2*1024LL*1024LL*1024LL) {
879 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
880 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
881 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
882 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
884 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
885 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
886 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
887 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
892 char *human_size_brief(long long bytes
, int prefix
)
896 /* We convert bytes to either centi-M{ega,ibi}bytes or
897 * centi-G{igi,ibi}bytes, with appropriate rounding,
898 * and then print 1/100th of those as a decimal.
899 * We allow upto 2048Megabytes before converting to
900 * gigabytes, as that shows more precision and isn't
901 * too large a number.
902 * Terabytes are not yet handled.
904 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
905 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
908 if (bytes
< 5000*1024)
910 else if (prefix
== IEC
) {
911 if (bytes
< 2*1024LL*1024LL*1024LL) {
912 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
913 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
914 cMiB
/100, cMiB
% 100);
916 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
917 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
918 cGiB
/100, cGiB
% 100);
921 else if (prefix
== JEDEC
) {
922 if (bytes
< 2*1024LL*1024LL*1024LL) {
923 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
924 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
927 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
928 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
938 void print_r10_layout(int layout
)
940 int near
= layout
& 255;
941 int far
= (layout
>> 8) & 255;
942 int offset
= (layout
&0x10000);
946 printf("%s near=%d", sep
, near
);
950 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
952 printf("NO REDUNDANCY");
955 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
956 int chunksize
, unsigned long long devsize
)
960 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
961 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
964 int get_data_disks(int level
, int layout
, int raid_disks
)
968 case 0: data_disks
= raid_disks
;
970 case 1: data_disks
= 1;
973 case 5: data_disks
= raid_disks
- 1;
975 case 6: data_disks
= raid_disks
- 2;
977 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
984 dev_t
devnm2devid(char *devnm
)
986 /* First look in /sys/block/$DEVNM/dev for %d:%d
987 * If that fails, try parsing out a number
994 sprintf(path
, "/sys/block/%s/dev", devnm
);
995 fd
= open(path
, O_RDONLY
);
998 int n
= read(fd
, buf
, sizeof(buf
));
1002 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
1003 return makedev(mjr
, mnr
);
1005 if (strncmp(devnm
, "md_d", 4) == 0 &&
1006 isdigit(devnm
[4]) &&
1007 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
1008 ep
> devnm
&& *ep
== 0)
1009 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
1011 if (strncmp(devnm
, "md", 2) == 0 &&
1012 isdigit(devnm
[2]) &&
1013 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
1014 ep
> devnm
&& *ep
== 0)
1015 return makedev(MD_MAJOR
, mnr
);
1020 char *get_md_name(char *devnm
)
1022 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
1023 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
1025 static char devname
[50];
1027 dev_t rdev
= devnm2devid(devnm
);
1032 if (strncmp(devnm
, "md_", 3) == 0) {
1033 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
1035 if (stat(devname
, &stb
) == 0 &&
1036 (S_IFMT
&stb
.st_mode
) == S_IFBLK
&& (stb
.st_rdev
== rdev
))
1039 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
1040 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
1041 (stb
.st_rdev
== rdev
))
1044 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
1045 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
1046 (stb
.st_rdev
== rdev
))
1049 dn
= map_dev(major(rdev
), minor(rdev
), 0);
1052 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
1053 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
1054 if (errno
!= EEXIST
)
1057 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
1058 (stb
.st_rdev
== rdev
))
1064 void put_md_name(char *name
)
1066 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1070 int get_maj_min(char *dev
, int *major
, int *minor
)
1073 *major
= strtoul(dev
, &e
, 0);
1074 return (e
> dev
&& *e
== ':' && e
[1] &&
1075 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1079 int dev_open(char *dev
, int flags
)
1081 /* like 'open', but if 'dev' matches %d:%d, create a temp
1082 * block device and open that
1093 if (get_maj_min(dev
, &major
, &minor
)) {
1094 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1095 (int)getpid(), major
, minor
);
1096 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1097 fd
= open(devname
, flags
);
1101 /* Try /tmp as /dev appear to be read-only */
1102 snprintf(devname
, sizeof(devname
),
1103 "/tmp/.tmp.md.%d:%d:%d",
1104 (int)getpid(), major
, minor
);
1105 if (mknod(devname
, S_IFBLK
|0600,
1106 makedev(major
, minor
)) == 0) {
1107 fd
= open(devname
, flags
);
1112 fd
= open(dev
, flags
);
1116 int open_dev_flags(char *devnm
, int flags
)
1121 devid
= devnm2devid(devnm
);
1122 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1123 return dev_open(buf
, flags
);
1126 int open_dev(char *devnm
)
1128 return open_dev_flags(devnm
, O_RDONLY
);
1131 int open_dev_excl(char *devnm
)
1136 dev_t devid
= devnm2devid(devnm
);
1139 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1140 for (i
= 0; i
< 25; i
++) {
1141 int fd
= dev_open(buf
, flags
|O_EXCL
);
1144 if (errno
== EACCES
&& flags
== O_RDWR
) {
1157 int same_dev(char *one
, char *two
)
1159 struct stat st1
, st2
;
1160 if (stat(one
, &st1
) != 0)
1162 if (stat(two
, &st2
) != 0)
1164 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1166 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1168 return st1
.st_rdev
== st2
.st_rdev
;
1171 void wait_for(char *dev
, int fd
)
1174 struct stat stb_want
;
1177 if (fstat(fd
, &stb_want
) != 0 ||
1178 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1181 for (i
= 0; i
< 25; i
++) {
1183 if (stat(dev
, &stb
) == 0 &&
1184 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1185 (stb
.st_rdev
== stb_want
.st_rdev
))
1192 pr_err("timeout waiting for %s\n", dev
);
1195 struct superswitch
*superlist
[] =
1198 &super_ddf
, &super_imsm
,
1203 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1205 mdu_array_info_t array
;
1208 struct supertype
*st
= NULL
;
1213 char *subarray
= NULL
;
1214 char container
[32] = "";
1216 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1219 vers
= sra
->array
.major_version
;
1220 minor
= sra
->array
.minor_version
;
1221 verstr
= sra
->text_version
;
1223 if (md_get_array_info(fd
, &array
))
1224 array
.major_version
= array
.minor_version
= 0;
1225 vers
= array
.major_version
;
1226 minor
= array
.minor_version
;
1231 sprintf(version
, "%d.%d", vers
, minor
);
1234 if (minor
== -2 && is_subarray(verstr
)) {
1235 char *dev
= verstr
+1;
1237 subarray
= strchr(dev
, '/');
1240 subarray
= xstrdup(subarray
);
1242 strcpy(container
, dev
);
1244 sra
= sysfs_read(-1, container
, GET_VERSION
);
1245 if (sra
&& sra
->text_version
[0])
1246 verstr
= sra
->text_version
;
1248 verstr
= "-no-metadata-";
1251 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1252 st
= superlist
[i
]->match_metadata_desc(verstr
);
1258 *subarrayp
= subarray
;
1259 strcpy(st
->container_devnm
, container
);
1260 strcpy(st
->devnm
, fd2devnm(fd
));
1267 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1272 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1273 fd
= dev_open(buf
, O_RDONLY
);
1276 if (get_dev_size(fd
, NULL
, size
)) {
1284 int dev_sector_size_from_id(dev_t id
, unsigned int *size
)
1289 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1290 fd
= dev_open(buf
, O_RDONLY
);
1293 if (get_dev_sector_size(fd
, NULL
, size
)) {
1301 struct supertype
*dup_super(struct supertype
*orig
)
1303 struct supertype
*st
;
1307 st
= xcalloc(1, sizeof(*st
));
1309 st
->max_devs
= orig
->max_devs
;
1310 st
->minor_version
= orig
->minor_version
;
1311 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1312 st
->data_offset
= orig
->data_offset
;
1318 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1320 /* try each load_super to find the best match,
1321 * and return the best superswitch
1323 struct superswitch
*ss
;
1324 struct supertype
*st
;
1325 unsigned int besttime
= 0;
1329 st
= xcalloc(1, sizeof(*st
));
1330 st
->container_devnm
[0] = 0;
1332 for (i
= 0; superlist
[i
]; i
++) {
1335 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1337 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1339 memset(st
, 0, sizeof(*st
));
1340 st
->ignore_hw_compat
= 1;
1341 rv
= ss
->load_super(st
, fd
, NULL
);
1344 st
->ss
->getinfo_super(st
, &info
, NULL
);
1345 if (bestsuper
== -1 ||
1346 besttime
< info
.array
.ctime
) {
1348 besttime
= info
.array
.ctime
;
1353 if (bestsuper
!= -1) {
1355 memset(st
, 0, sizeof(*st
));
1356 st
->ignore_hw_compat
= 1;
1357 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1359 superlist
[bestsuper
]->free_super(st
);
1367 /* Return size of device in bytes */
1368 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1370 unsigned long long ldsize
;
1373 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1374 ldsize
= (unsigned long long)st
.st_size
;
1377 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1380 unsigned long dsize
;
1381 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1386 pr_err("Cannot get size of %s: %s\n",
1387 dname
, strerror(errno
));
1395 /* Return sector size of device in bytes */
1396 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1398 unsigned int sectsize
;
1400 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1402 pr_err("Cannot get sector size of %s: %s\n",
1403 dname
, strerror(errno
));
1407 *sectsizep
= sectsize
;
1411 /* Return true if this can only be a container, not a member device.
1412 * i.e. is and md device and size is zero
1414 int must_be_container(int fd
)
1417 unsigned long long size
;
1419 mdi
= sysfs_read(fd
, NULL
, GET_VERSION
);
1424 if (get_dev_size(fd
, NULL
, &size
) == 0)
1431 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1432 * Returns: 1 if successful
1433 * -1 for unknown partition type
1434 * 0 for other errors
1436 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1439 unsigned char empty_gpt_entry
[16]= {0};
1440 struct GPT_part_entry
*part
;
1442 unsigned long long curr_part_end
;
1443 unsigned all_partitions
, entry_size
;
1445 unsigned int sector_size
= 0;
1449 BUILD_BUG_ON(sizeof(gpt
) != 512);
1450 /* skip protective MBR */
1451 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1453 lseek(fd
, sector_size
, SEEK_SET
);
1454 /* read GPT header */
1455 if (read(fd
, &gpt
, 512) != 512)
1458 /* get the number of partition entries and the entry size */
1459 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1460 entry_size
= __le32_to_cpu(gpt
.part_size
);
1462 /* Check GPT signature*/
1463 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1467 if (all_partitions
> 1024 ||
1468 entry_size
> sizeof(buf
))
1471 part
= (struct GPT_part_entry
*)buf
;
1473 /* set offset to third block (GPT entries) */
1474 lseek(fd
, sector_size
*2, SEEK_SET
);
1475 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1476 /* read partition entry */
1477 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1480 /* is this valid partition? */
1481 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1482 /* check the last lba for the current partition */
1483 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1484 if (curr_part_end
> *endofpart
)
1485 *endofpart
= curr_part_end
;
1492 /* Sets endofpart parameter to the last block used by the last partition on the device.
1493 * Returns: 1 if successful
1494 * -1 for unknown partition type
1495 * 0 for other errors
1497 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1499 struct MBR boot_sect
;
1500 unsigned long long curr_part_end
;
1502 unsigned int sector_size
;
1507 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1510 if (read(fd
, &boot_sect
, 512) != 512)
1513 /* check MBP signature */
1514 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1516 /* found the correct signature */
1518 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1520 * Have to make every access through boot_sect rather
1521 * than using a pointer to the partition table (or an
1522 * entry), since the entries are not properly aligned.
1525 /* check for GPT type */
1526 if (boot_sect
.parts
[part_nr
].part_type
==
1527 MBR_GPT_PARTITION_TYPE
) {
1528 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1531 /* check the last used lba for the current partition */
1533 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1534 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1535 if (curr_part_end
> *endofpart
)
1536 *endofpart
= curr_part_end
;
1539 /* Unknown partition table */
1542 /* calculate number of 512-byte blocks */
1543 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1544 *endofpart
*= (sector_size
/ 512);
1549 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1550 unsigned long long size
)
1553 * Check where the last partition ends
1555 unsigned long long endofpart
;
1557 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1558 /* There appears to be a partition table here */
1559 if (freesize
== 0) {
1560 /* partitions will not be visible in new device */
1561 pr_err("partition table exists on %s but will be lost or\n"
1562 " meaningless after creating array\n",
1565 } else if (endofpart
> freesize
) {
1566 /* last partition overlaps metadata */
1567 pr_err("metadata will over-write last partition on %s.\n",
1570 } else if (size
&& endofpart
> size
) {
1571 /* partitions will be truncated in new device */
1572 pr_err("array size is too small to cover all partitions on %s.\n",
1580 int open_container(int fd
)
1582 /* 'fd' is a block device. Find out if it is in use
1583 * by a container, and return an open fd on that container.
1594 if (fstat(fd
, &st
) != 0)
1596 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1597 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1598 e
= path
+ strlen(path
);
1600 dir
= opendir(path
);
1603 while ((de
= readdir(dir
))) {
1606 if (de
->d_name
[0] == '.')
1608 /* Need to make sure it is a container and not a volume */
1609 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1610 dfd
= open(path
, O_RDONLY
);
1613 n
= read(dfd
, buf
, sizeof(buf
));
1615 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1618 if (strncmp(buf
, "external", 8) != 0 ||
1622 sprintf(e
, "/%s/dev", de
->d_name
);
1623 dfd
= open(path
, O_RDONLY
);
1626 n
= read(dfd
, buf
, sizeof(buf
));
1628 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1631 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1633 sprintf(buf
, "%d:%d", major
, minor
);
1634 dfd
= dev_open(buf
, O_RDONLY
);
1644 struct superswitch
*version_to_superswitch(char *vers
)
1648 for (i
= 0; superlist
[i
]; i
++) {
1649 struct superswitch
*ss
= superlist
[i
];
1651 if (strcmp(vers
, ss
->name
) == 0)
1658 int metadata_container_matches(char *metadata
, char *devnm
)
1660 /* Check if 'devnm' is the container named in 'metadata'
1662 * /containername/componentname or
1663 * -containername/componentname
1666 if (*metadata
!= '/' && *metadata
!= '-')
1669 if (strncmp(metadata
+1, devnm
, l
) != 0)
1671 if (metadata
[l
+1] != '/')
1676 int metadata_subdev_matches(char *metadata
, char *devnm
)
1678 /* Check if 'devnm' is the subdev named in 'metadata'
1680 * /containername/subdev or
1681 * -containername/subdev
1684 if (*metadata
!= '/' && *metadata
!= '-')
1686 sl
= strchr(metadata
+1, '/');
1689 if (strcmp(sl
+1, devnm
) == 0)
1694 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1696 if (mdstat
->metadata_version
== NULL
||
1697 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1698 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1704 int is_subarray_active(char *subarray
, char *container
)
1706 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1707 struct mdstat_ent
*ent
;
1709 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1710 if (is_container_member(ent
, container
))
1711 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1714 free_mdstat(mdstat
);
1719 /* open_subarray - opens a subarray in a container
1720 * @dev: container device name
1721 * @st: empty supertype
1722 * @quiet: block reporting errors flag
1724 * On success returns an fd to a container and fills in *st
1726 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1729 struct mdinfo
*info
;
1733 fd
= open(dev
, O_RDWR
|O_EXCL
);
1736 pr_err("Couldn't open %s, aborting\n",
1741 _devnm
= fd2devnm(fd
);
1742 if (_devnm
== NULL
) {
1744 pr_err("Failed to determine device number for %s\n",
1748 strcpy(st
->devnm
, _devnm
);
1750 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1753 pr_err("Failed to read sysfs for %s\n",
1758 if (mdi
->array
.level
!= UnSet
) {
1760 pr_err("%s is not a container\n", dev
);
1764 st
->ss
= version_to_superswitch(mdi
->text_version
);
1767 pr_err("Operation not supported for %s metadata\n",
1772 if (st
->devnm
[0] == 0) {
1774 pr_err("Failed to allocate device name\n");
1778 if (!st
->ss
->load_container
) {
1780 pr_err("%s is not a container\n", dev
);
1784 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1786 pr_err("Failed to load metadata for %s\n",
1791 info
= st
->ss
->container_content(st
, subarray
);
1794 pr_err("Failed to find subarray-%s in %s\n",
1804 st
->ss
->free_super(st
);
1817 int add_disk(int mdfd
, struct supertype
*st
,
1818 struct mdinfo
*sra
, struct mdinfo
*info
)
1820 /* Add a device to an array, in one of 2 ways. */
1823 if (st
->ss
->external
) {
1824 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1825 info
->recovery_start
= MaxSector
;
1827 info
->recovery_start
= 0;
1828 rv
= sysfs_add_disk(sra
, info
, 0);
1831 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1835 sd2
= xmalloc(sizeof(*sd2
));
1837 sd2
->next
= sra
->devs
;
1842 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1846 int remove_disk(int mdfd
, struct supertype
*st
,
1847 struct mdinfo
*sra
, struct mdinfo
*info
)
1851 /* Remove the disk given by 'info' from the array */
1852 if (st
->ss
->external
)
1853 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1855 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1860 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1862 int cnt
= force
? 500 : 5;
1865 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1866 * outstanding IO requests to the device.
1867 * In this case, it can be helpful to wait a little while,
1868 * up to 5 seconds if 'force' is set, or 50 msec if not.
1870 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1878 int sys_hot_remove_disk(int statefd
, int force
)
1880 int cnt
= force
? 500 : 5;
1883 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1887 return ret
== 6 ? 0 : -1;
1890 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1892 /* Initialise kernel's knowledge of array.
1893 * This varies between externally managed arrays
1896 mdu_array_info_t inf
;
1899 if (st
->ss
->external
)
1900 return sysfs_set_array(info
, 9003);
1902 memset(&inf
, 0, sizeof(inf
));
1903 inf
.major_version
= info
->array
.major_version
;
1904 inf
.minor_version
= info
->array
.minor_version
;
1905 rv
= md_set_array_info(mdfd
, &inf
);
1910 unsigned long long min_recovery_start(struct mdinfo
*array
)
1912 /* find the minimum recovery_start in an array for metadata
1913 * formats that only record per-array recovery progress instead
1916 unsigned long long recovery_start
= MaxSector
;
1919 for (d
= array
->devs
; d
; d
= d
->next
)
1920 recovery_start
= min(recovery_start
, d
->recovery_start
);
1922 return recovery_start
;
1925 int mdmon_pid(char *devnm
)
1932 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1934 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1938 n
= read(fd
, pid
, 9);
1945 int mdmon_running(char *devnm
)
1947 int pid
= mdmon_pid(devnm
);
1950 if (kill(pid
, 0) == 0)
1955 int start_mdmon(char *devnm
)
1969 if (check_env("MDADM_NO_MDMON"))
1972 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1976 sl
= strrchr(pathbuf
, '/');
1981 strcpy(sl
, "mdmon");
1985 /* First try to run systemctl */
1986 if (!check_env("MDADM_NO_SYSTEMCTL"))
1989 /* FIXME yuk. CLOSE_EXEC?? */
1991 for (i
= 3; skipped
< 20; i
++)
1997 /* Don't want to see error messages from
1998 * systemctl. If the service doesn't exist,
1999 * we start mdmon ourselves.
2002 open("/dev/null", O_WRONLY
);
2003 snprintf(pathbuf
, sizeof(pathbuf
), "mdmon@%s.service",
2005 status
= execl("/usr/bin/systemctl", "systemctl",
2008 status
= execl("/bin/systemctl", "systemctl", "start",
2011 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
2013 default: /* parent - good */
2014 pid
= wait(&status
);
2015 if (pid
>= 0 && status
== 0)
2019 /* That failed, try running mdmon directly */
2022 /* FIXME yuk. CLOSE_EXEC?? */
2024 for (i
= 3; skipped
< 20; i
++)
2030 for (i
= 0; paths
[i
]; i
++)
2032 execl(paths
[i
], paths
[i
],
2036 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
2038 default: /* parent - good */
2039 pid
= wait(&status
);
2040 if (pid
< 0 || status
!= 0) {
2041 pr_err("failed to launch mdmon. Array remains readonly\n");
2048 __u32
random32(void)
2051 int rfd
= open("/dev/urandom", O_RDONLY
);
2052 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
2059 void random_uuid(__u8
*buf
)
2064 fd
= open("/dev/urandom", O_RDONLY
);
2067 len
= read(fd
, buf
, 16);
2075 for (i
= 0; i
< 4; i
++)
2080 int flush_metadata_updates(struct supertype
*st
)
2084 st
->update_tail
= NULL
;
2088 sfd
= connect_monitor(st
->container_devnm
);
2092 while (st
->updates
) {
2093 struct metadata_update
*mu
= st
->updates
;
2094 st
->updates
= mu
->next
;
2096 send_message(sfd
, mu
, 0);
2104 st
->update_tail
= NULL
;
2108 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2111 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2116 mu
->space_list
= NULL
;
2118 *st
->update_tail
= mu
;
2119 st
->update_tail
= &mu
->next
;
2123 /* tinyc doesn't optimize this check in ioctl.h out ... */
2124 unsigned int __invalid_size_argument_for_IOC
= 0;
2127 int experimental(void)
2129 if (check_env("MDADM_EXPERIMENTAL"))
2132 pr_err("To use this feature MDADM_EXPERIMENTAL environment variable has to be defined.\n");
2137 /* Pick all spares matching given criteria from a container
2138 * if min_size == 0 do not check size
2139 * if domlist == NULL do not check domains
2140 * if spare_group given add it to domains of each spare
2141 * metadata allows to test domains using metadata of destination array */
2142 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2143 struct spare_criteria
*criteria
,
2144 struct domainlist
*domlist
,
2146 const char *metadata
, int get_one
)
2148 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2150 /* get list of all disks in container */
2151 if (st
->ss
->getinfo_super_disks
)
2152 disks
= st
->ss
->getinfo_super_disks(st
);
2156 /* find spare devices on the list */
2158 disks
->array
.spare_disks
= 0;
2162 if (d
->disk
.state
== 0) {
2163 /* check if size is acceptable */
2164 unsigned long long dev_size
;
2165 unsigned int dev_sector_size
;
2167 int sector_size_valid
= 0;
2169 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2171 if (!criteria
->min_size
||
2172 (dev_size_from_id(dev
, &dev_size
) &&
2173 dev_size
>= criteria
->min_size
))
2176 if (!criteria
->sector_size
||
2177 (dev_sector_size_from_id(dev
, &dev_sector_size
) &&
2178 criteria
->sector_size
== dev_sector_size
))
2179 sector_size_valid
= 1;
2181 found
= size_valid
&& sector_size_valid
;
2183 /* check if domain matches */
2184 if (found
&& domlist
) {
2185 struct dev_policy
*pol
= devid_policy(dev
);
2187 pol_add(&pol
, pol_domain
,
2189 if (domain_test(domlist
, pol
, metadata
) != 1)
2191 dev_policy_free(pol
);
2196 disks
->array
.spare_disks
++;
2210 /* Checks if paths point to the same device
2211 * Returns 0 if they do.
2212 * Returns 1 if they don't.
2213 * Returns -1 if something went wrong,
2214 * e.g. paths are empty or the files
2215 * they point to don't exist */
2216 int compare_paths (char* path1
, char* path2
)
2218 struct stat st1
,st2
;
2220 if (path1
== NULL
|| path2
== NULL
)
2222 if (stat(path1
,&st1
) != 0)
2224 if (stat(path2
,&st2
) != 0)
2226 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2231 /* Make sure we can open as many devices as needed */
2232 void enable_fds(int devices
)
2234 unsigned int fds
= 20 + devices
;
2236 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0 || lim
.rlim_cur
>= fds
)
2238 if (lim
.rlim_max
< fds
)
2241 setrlimit(RLIMIT_NOFILE
, &lim
);
2246 /* This is based on similar function in systemd. */
2248 /* statfs.f_type is signed long on s390x and MIPS, causing all
2249 sorts of sign extension problems with RAMFS_MAGIC being
2250 defined as 0x858458f6 */
2251 return statfs("/", &s
) >= 0 &&
2252 ((unsigned long)s
.f_type
== TMPFS_MAGIC
||
2253 ((unsigned long)s
.f_type
& 0xFFFFFFFFUL
) ==
2254 ((unsigned long)RAMFS_MAGIC
& 0xFFFFFFFFUL
));
2257 void reopen_mddev(int mdfd
)
2259 /* Re-open without any O_EXCL, but keep
2264 devnm
= fd2devnm(mdfd
);
2266 fd
= open_dev(devnm
);
2267 if (fd
>= 0 && fd
!= mdfd
)
2271 static struct cmap_hooks
*cmap_hooks
= NULL
;
2272 static int is_cmap_hooks_ready
= 0;
2274 void set_cmap_hooks(void)
2276 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2277 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2278 if (!cmap_hooks
->cmap_handle
)
2281 cmap_hooks
->initialize
=
2282 dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2283 cmap_hooks
->get_string
=
2284 dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2285 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2287 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2288 !cmap_hooks
->finalize
)
2289 dlclose(cmap_hooks
->cmap_handle
);
2291 is_cmap_hooks_ready
= 1;
2294 int get_cluster_name(char **cluster_name
)
2297 cmap_handle_t handle
;
2299 if (!is_cmap_hooks_ready
)
2302 rv
= cmap_hooks
->initialize(&handle
);
2306 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2308 free(*cluster_name
);
2315 cmap_hooks
->finalize(handle
);
2320 void set_dlm_hooks(void)
2322 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2323 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2324 if (!dlm_hooks
->dlm_handle
)
2327 dlm_hooks
->create_lockspace
=
2328 dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2329 dlm_hooks
->release_lockspace
=
2330 dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2331 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2332 dlm_hooks
->ls_unlock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock");
2333 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2334 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2336 if (!dlm_hooks
->create_lockspace
|| !dlm_hooks
->ls_lock
||
2337 !dlm_hooks
->ls_unlock
|| !dlm_hooks
->release_lockspace
||
2338 !dlm_hooks
->ls_get_fd
|| !dlm_hooks
->dispatch
)
2339 dlclose(dlm_hooks
->dlm_handle
);
2341 is_dlm_hooks_ready
= 1;
2344 void set_hooks(void)
2350 int zero_disk_range(int fd
, unsigned long long sector
, size_t count
)
2356 size_t len
= count
* 512;
2359 fd_zero
= open("/dev/zero", O_RDONLY
);
2361 pr_err("Cannot open /dev/zero\n");
2365 if (lseek64(fd
, sector
* 512, SEEK_SET
) < 0) {
2367 pr_err("Failed to seek offset for zeroing\n");
2371 addr
= mmap(NULL
, len
, PROT_READ
, MAP_PRIVATE
, fd_zero
, 0);
2373 if (addr
== MAP_FAILED
) {
2375 pr_err("Mapping /dev/zero failed\n");
2380 n
= write(fd
, addr
+ written
, len
- written
);
2385 pr_err("Zeroing disk range failed\n");
2389 } while (written
!= len
);