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;
92 static struct dlm_hooks
*dlm_hooks
= NULL
;
93 struct dlm_lock_resource
*dlm_lock_res
= NULL
;
94 static int ast_called
= 0;
96 struct dlm_lock_resource
{
101 /* Using poll(2) to wait for and dispatch ASTs */
102 static int poll_for_ast(dlm_lshandle_t ls
)
106 pfd
.fd
= dlm_hooks
->ls_get_fd(ls
);
111 if (poll(&pfd
, 1, 0) < 0)
116 dlm_hooks
->dispatch(dlm_hooks
->ls_get_fd(ls
));
123 static void dlm_ast(void *arg
)
128 static char *cluster_name
= NULL
;
129 /* Create the lockspace, take bitmapXXX locks on all the bitmaps. */
130 int cluster_get_dlmlock(int *lockid
)
134 int flags
= LKF_NOQUEUE
;
136 ret
= get_cluster_name(&cluster_name
);
138 pr_err("The md can't get cluster name\n");
142 dlm_lock_res
= xmalloc(sizeof(struct dlm_lock_resource
));
143 dlm_lock_res
->ls
= dlm_hooks
->create_lockspace(cluster_name
, O_RDWR
);
144 if (!dlm_lock_res
->ls
) {
145 pr_err("%s failed to create lockspace\n", cluster_name
);
149 snprintf(str
, 64, "bitmap%s", cluster_name
);
150 ret
= dlm_hooks
->ls_lock(dlm_lock_res
->ls
, LKM_PWMODE
, &dlm_lock_res
->lksb
,
151 flags
, str
, strlen(str
), 0, dlm_ast
,
152 dlm_lock_res
, NULL
, NULL
);
154 pr_err("error %d when get PW mode on lock %s\n", errno
, str
);
155 dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
159 /* Wait for it to complete */
160 poll_for_ast(dlm_lock_res
->ls
);
161 *lockid
= dlm_lock_res
->lksb
.sb_lkid
;
163 return dlm_lock_res
->lksb
.sb_status
;
166 int cluster_release_dlmlock(int lockid
)
173 ret
= dlm_hooks
->ls_unlock(dlm_lock_res
->ls
, lockid
, 0,
174 &dlm_lock_res
->lksb
, dlm_lock_res
);
176 pr_err("error %d happened when unlock\n", errno
);
177 /* XXX make sure the lock is unlocked eventually */
181 /* Wait for it to complete */
182 poll_for_ast(dlm_lock_res
->ls
);
184 errno
= dlm_lock_res
->lksb
.sb_status
;
185 if (errno
!= EUNLOCK
) {
186 pr_err("error %d happened in ast when unlock lockspace\n", errno
);
187 /* XXX make sure the lockspace is unlocked eventually */
191 ret
= dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
193 pr_err("error %d happened when release lockspace\n", errno
);
194 /* XXX make sure the lockspace is released eventually */
203 int md_array_valid(int fd
)
208 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
210 if (sra
->array_state
!= ARRAY_UNKNOWN_STATE
)
218 * GET_ARRAY_INFO doesn't provide access to the proper state
219 * information, so fallback to a basic check for raid_disks != 0
221 ret
= ioctl(fd
, RAID_VERSION
);
227 int md_array_active(int fd
)
230 struct mdu_array_info_s array
;
233 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
235 if (sra
->array_state
!= ARRAY_CLEAR
&&
236 sra
->array_state
!= ARRAY_INACTIVE
&&
237 sra
->array_state
!= ARRAY_UNKNOWN_STATE
)
245 * GET_ARRAY_INFO doesn't provide access to the proper state
246 * information, so fallback to a basic check for raid_disks != 0
248 ret
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
255 * Get array info from the kernel. Longer term we want to deprecate the
256 * ioctl and get it from sysfs.
258 int md_get_array_info(int fd
, struct mdu_array_info_s
*array
)
260 return ioctl(fd
, GET_ARRAY_INFO
, array
);
266 int md_set_array_info(int fd
, struct mdu_array_info_s
*array
)
268 return ioctl(fd
, SET_ARRAY_INFO
, array
);
272 * Get disk info from the kernel.
274 int md_get_disk_info(int fd
, struct mdu_disk_info_s
*disk
)
276 return ioctl(fd
, GET_DISK_INFO
, disk
);
280 * Parse a 128 bit uuid in 4 integers
281 * format is 32 hexx nibbles with options :.<space> separator
282 * If not exactly 32 hex digits are found, return 0
285 int parse_uuid(char *str
, int uuid
[4])
287 int hit
= 0; /* number of Hex digIT */
290 for (i
= 0; i
< 4; i
++)
293 while ((c
= *str
++) != 0) {
295 if (c
>= '0' && c
<= '9')
297 else if (c
>= 'a' && c
<= 'f')
299 else if (c
>= 'A' && c
<= 'F')
301 else if (strchr(":. -", c
))
316 int get_linux_version()
320 int a
= 0, b
= 0,c
= 0;
325 a
= strtoul(cp
, &cp
, 10);
327 b
= strtoul(cp
+1, &cp
, 10);
329 c
= strtoul(cp
+1, &cp
, 10);
331 return (a
*1000000)+(b
*1000)+c
;
334 int mdadm_version(char *version
)
342 cp
= strchr(version
, '-');
343 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
346 a
= strtoul(cp
, &cp
, 10);
349 b
= strtoul(cp
+1, &cp
, 10);
351 c
= strtoul(cp
+1, &cp
, 10);
354 if (*cp
!= ' ' && *cp
!= '-')
356 return (a
*1000000)+(b
*1000)+c
;
359 unsigned long long parse_size(char *size
)
361 /* parse 'size' which should be a number optionally
362 * followed by 'K', 'M', or 'G'.
363 * Without a suffix, K is assumed.
364 * Number returned is in sectors (half-K)
365 * INVALID_SECTORS returned on error.
368 long long s
= strtoll(size
, &c
, 10);
382 s
*= 1024 * 1024 * 2;
384 case 's': /* sectors */
395 int parse_layout_10(char *layout
)
399 /* Parse the layout string for raid10 */
400 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
401 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
402 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
406 if (layout
[0] == 'n')
408 else if (layout
[0] == 'o')
409 rv
= 0x10000 + (copies
<<8) + 1;
411 rv
= 1 + (copies
<<8);
415 int parse_layout_faulty(char *layout
)
417 /* Parse the layout string for 'faulty' */
418 int ln
= strcspn(layout
, "0123456789");
419 char *m
= xstrdup(layout
);
422 mode
= map_name(faultylayout
, m
);
426 return mode
| (atoi(layout
+ln
)<< ModeShift
);
429 long parse_num(char *num
)
431 /* Either return a valid number, or -1 */
433 long rv
= strtol(num
, &c
, 10);
434 if (rv
< 0 || *c
|| !num
[0])
440 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
443 *slot
= strtoul(input
, &dev
, 10);
444 if (dev
== input
|| dev
[0] != ':')
450 void remove_partitions(int fd
)
452 /* remove partitions from this block devices.
453 * This is used for components added to an array
455 #ifdef BLKPG_DEL_PARTITION
456 struct blkpg_ioctl_arg a
;
457 struct blkpg_partition p
;
459 a
.op
= BLKPG_DEL_PARTITION
;
461 a
.datalen
= sizeof(p
);
463 memset(a
.data
, 0, a
.datalen
);
464 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
465 ioctl(fd
, BLKPG
, &a
);
469 int test_partition(int fd
)
471 /* Check if fd is a whole-disk or a partition.
472 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
473 * will return ENXIO on an invalid partition number.
475 struct blkpg_ioctl_arg a
;
476 struct blkpg_partition p
;
477 a
.op
= BLKPG_DEL_PARTITION
;
479 a
.datalen
= sizeof(p
);
481 memset(a
.data
, 0, a
.datalen
);
483 if (ioctl(fd
, BLKPG
, &a
) == 0)
484 /* Very unlikely, but not a partition */
486 if (errno
== ENXIO
|| errno
== ENOTTY
)
487 /* not a partition */
493 int test_partition_from_id(dev_t id
)
498 sprintf(buf
, "%d:%d", major(id
), minor(id
));
499 fd
= dev_open(buf
, O_RDONLY
);
502 rv
= test_partition(fd
);
507 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
513 for (i
= 0; i
< raid_disks
; i
++)
514 avail_disks
+= !!avail
[i
];
518 /* This is the tricky one - we need to check
519 * which actual disks are present.
521 copies
= (layout
&255)* ((layout
>>8) & 255);
524 /* there must be one of the 'copies' form 'first' */
531 this = (this+1) % raid_disks
;
535 first
= (first
+(layout
&255)) % raid_disks
;
536 } while (first
!= 0);
539 case LEVEL_MULTIPATH
:
540 return avail_disks
>= 1;
543 return avail_disks
== raid_disks
;
545 return avail_disks
>= 1;
547 if (avail_disks
== raid_disks
- 1 &&
548 !avail
[raid_disks
- 1])
549 /* If just the parity device is missing, then we
550 * have enough, even if not clean
556 return avail_disks
>= raid_disks
-1;
558 return avail_disks
>= raid_disks
;
561 return avail_disks
>= raid_disks
-2;
563 return avail_disks
>= raid_disks
;
569 const int uuid_zero
[4] = { 0, 0, 0, 0 };
571 int same_uuid(int a
[4], int b
[4], int swapuuid
)
574 /* parse uuids are hostendian.
575 * uuid's from some superblocks are big-ending
576 * if there is a difference, we need to swap..
578 unsigned char *ac
= (unsigned char *)a
;
579 unsigned char *bc
= (unsigned char *)b
;
581 for (i
= 0; i
< 16; i
+= 4) {
582 if (ac
[i
+0] != bc
[i
+3] ||
583 ac
[i
+1] != bc
[i
+2] ||
584 ac
[i
+2] != bc
[i
+1] ||
599 void copy_uuid(void *a
, int b
[4], int swapuuid
)
602 /* parse uuids are hostendian.
603 * uuid's from some superblocks are big-ending
604 * if there is a difference, we need to swap..
606 unsigned char *ac
= (unsigned char *)a
;
607 unsigned char *bc
= (unsigned char *)b
;
609 for (i
= 0; i
< 16; i
+= 4) {
619 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
626 copy_uuid(uuid
, id
, swap
);
627 for (i
= 0; i
< 4; i
++) {
630 for (j
= 3; j
>= 0; j
--) {
631 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
639 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
, char *buf
, char sep
)
641 // dirty hack to work around an issue with super1 superblocks...
642 // super1 superblocks need swapuuid set in order for assembly to
643 // work, but can't have it set if we want this printout to match
644 // all the other uuid printouts in super1.c, so we force swapuuid
645 // to 1 to make our printout match the rest of super1
646 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 : st
->ss
->swapuuid
, buf
, sep
);
649 int check_ext2(int fd
, char *name
)
652 * Check for an ext2fs file system.
653 * Superblock is always 1K at 1K offset
655 * s_magic is le16 at 56 == 0xEF53
656 * report mtime - le32 at 44
658 * logblksize - le32 at 24
660 unsigned char sb
[1024];
662 unsigned long long size
;
664 if (lseek(fd
, 1024,0)!= 1024)
666 if (read(fd
, sb
, 1024)!= 1024)
668 if (sb
[56] != 0x53 || sb
[57] != 0xef)
671 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
672 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
673 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
675 pr_err("%s appears to contain an ext2fs file system\n",
677 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
681 int check_reiser(int fd
, char *name
)
684 * superblock is at 64K
686 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
689 unsigned char sb
[1024];
690 unsigned long long size
;
691 if (lseek(fd
, 64*1024, 0) != 64*1024)
693 if (read(fd
, sb
, 1024) != 1024)
695 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
696 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
698 pr_err("%s appears to contain a reiserfs file system\n",name
);
699 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
700 cont_err("size = %lluK\n", size
*4);
705 int check_raid(int fd
, char *name
)
710 struct supertype
*st
= guess_super(fd
);
714 if (st
->ss
->add_to_super
!= NULL
) {
715 st
->ss
->load_super(st
, fd
, name
);
716 /* Looks like a raid array .. */
717 pr_err("%s appears to be part of a raid array:\n", name
);
718 st
->ss
->getinfo_super(st
, &info
, NULL
);
719 st
->ss
->free_super(st
);
720 crtime
= info
.array
.ctime
;
721 level
= map_num(pers
, info
.array
.level
);
724 cont_err("level=%s devices=%d ctime=%s",
725 level
, info
.array
.raid_disks
, ctime(&crtime
));
727 /* Looks like GPT or MBR */
728 pr_err("partition table exists on %s\n", name
);
733 int fstat_is_blkdev(int fd
, char *devname
, dev_t
*rdev
)
737 if (fstat(fd
, &stb
) != 0) {
738 pr_err("fstat failed for %s: %s\n", devname
, strerror(errno
));
741 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
742 pr_err("%s is not a block device.\n", devname
);
750 int stat_is_blkdev(char *devname
, dev_t
*rdev
)
754 if (stat(devname
, &stb
) != 0) {
755 pr_err("stat failed for %s: %s\n", devname
, strerror(errno
));
758 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
759 pr_err("%s is not a block device.\n", devname
);
771 for (i
= 0; i
< 5; i
++) {
773 fprintf(stderr
, "%s%s", mesg
, add
);
775 if (fgets(buf
, 100, stdin
)==NULL
)
777 if (buf
[0]=='y' || buf
[0]=='Y')
779 if (buf
[0]=='n' || buf
[0]=='N')
783 pr_err("assuming 'no'\n");
787 int is_standard(char *dev
, int *nump
)
789 /* tests if dev is a "standard" md dev name.
790 * i.e if the last component is "/dNN" or "/mdNN",
791 * where NN is a string of digits
792 * Returns 1 if a partitionable standard,
793 * -1 if non-partitonable,
794 * 0 if not a standard name.
796 char *d
= strrchr(dev
, '/');
801 if (strncmp(d
, "/d",2) == 0)
802 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
803 else if (strncmp(d
, "/md_d", 5) == 0)
804 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
805 else if (strncmp(d
, "/md", 3) == 0)
806 d
+= 3, type
= -1; /* /dev/mdN */
807 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
808 d
+= 1, type
= -1; /* /dev/md/N */
818 if (nump
) *nump
= num
;
823 unsigned long calc_csum(void *super
, int bytes
)
825 unsigned long long newcsum
= 0;
828 unsigned int *superc
= (unsigned int*) super
;
830 for(i
= 0; i
< bytes
/4; i
++)
831 newcsum
+= superc
[i
];
832 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
834 /* The in-kernel checksum calculation is always 16bit on
835 * the alpha, though it is 32 bit on i386...
836 * I wonder what it is elsewhere... (it uses an API in
837 * a way that it shouldn't).
839 csum
= (csum
& 0xffff) + (csum
>> 16);
840 csum
= (csum
& 0xffff) + (csum
>> 16);
845 char *human_size(long long bytes
)
849 /* We convert bytes to either centi-M{ega,ibi}bytes or
850 * centi-G{igi,ibi}bytes, with appropriate rounding,
851 * and then print 1/100th of those as a decimal.
852 * We allow upto 2048Megabytes before converting to
853 * gigabytes, as that shows more precision and isn't
854 * too large a number.
855 * Terabytes are not yet handled.
858 if (bytes
< 5000*1024)
860 else if (bytes
< 2*1024LL*1024LL*1024LL) {
861 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
862 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
863 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
864 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
866 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
867 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
868 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
869 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
874 char *human_size_brief(long long bytes
, int prefix
)
878 /* We convert bytes to either centi-M{ega,ibi}bytes or
879 * centi-G{igi,ibi}bytes, with appropriate rounding,
880 * and then print 1/100th of those as a decimal.
881 * We allow upto 2048Megabytes before converting to
882 * gigabytes, as that shows more precision and isn't
883 * too large a number.
884 * Terabytes are not yet handled.
886 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
887 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
890 if (bytes
< 5000*1024)
892 else if (prefix
== IEC
) {
893 if (bytes
< 2*1024LL*1024LL*1024LL) {
894 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
895 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
896 cMiB
/100, cMiB
% 100);
898 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
899 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
900 cGiB
/100, cGiB
% 100);
903 else if (prefix
== JEDEC
) {
904 if (bytes
< 2*1024LL*1024LL*1024LL) {
905 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
906 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
909 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
910 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
920 void print_r10_layout(int layout
)
922 int near
= layout
& 255;
923 int far
= (layout
>> 8) & 255;
924 int offset
= (layout
&0x10000);
928 printf("%s near=%d", sep
, near
);
932 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
934 printf("NO REDUNDANCY");
937 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
938 int chunksize
, unsigned long long devsize
)
942 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
943 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
946 int get_data_disks(int level
, int layout
, int raid_disks
)
950 case 0: data_disks
= raid_disks
;
952 case 1: data_disks
= 1;
955 case 5: data_disks
= raid_disks
- 1;
957 case 6: data_disks
= raid_disks
- 2;
959 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
966 dev_t
devnm2devid(char *devnm
)
968 /* First look in /sys/block/$DEVNM/dev for %d:%d
969 * If that fails, try parsing out a number
976 sprintf(path
, "/sys/block/%s/dev", devnm
);
977 fd
= open(path
, O_RDONLY
);
980 int n
= read(fd
, buf
, sizeof(buf
));
984 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
985 return makedev(mjr
, mnr
);
987 if (strncmp(devnm
, "md_d", 4) == 0 &&
989 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
990 ep
> devnm
&& *ep
== 0)
991 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
993 if (strncmp(devnm
, "md", 2) == 0 &&
995 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
996 ep
> devnm
&& *ep
== 0)
997 return makedev(MD_MAJOR
, mnr
);
1002 char *get_md_name(char *devnm
)
1004 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
1005 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
1007 static char devname
[50];
1009 dev_t rdev
= devnm2devid(devnm
);
1014 if (strncmp(devnm
, "md_", 3) == 0) {
1015 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
1017 if (stat(devname
, &stb
) == 0
1018 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1019 && (stb
.st_rdev
== rdev
))
1022 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
1023 if (stat(devname
, &stb
) == 0
1024 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1025 && (stb
.st_rdev
== rdev
))
1028 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
1029 if (stat(devname
, &stb
) == 0
1030 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1031 && (stb
.st_rdev
== rdev
))
1034 dn
= map_dev(major(rdev
), minor(rdev
), 0);
1037 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
1038 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
1039 if (errno
!= EEXIST
)
1042 if (stat(devname
, &stb
) == 0
1043 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
1044 && (stb
.st_rdev
== rdev
))
1050 void put_md_name(char *name
)
1052 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1056 int get_maj_min(char *dev
, int *major
, int *minor
)
1059 *major
= strtoul(dev
, &e
, 0);
1060 return (e
> dev
&& *e
== ':' && e
[1] &&
1061 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1065 int dev_open(char *dev
, int flags
)
1067 /* like 'open', but if 'dev' matches %d:%d, create a temp
1068 * block device and open that
1079 if (get_maj_min(dev
, &major
, &minor
)) {
1080 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1081 (int)getpid(), major
, minor
);
1082 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1083 fd
= open(devname
, flags
);
1087 /* Try /tmp as /dev appear to be read-only */
1088 snprintf(devname
, sizeof(devname
), "/tmp/.tmp.md.%d:%d:%d",
1089 (int)getpid(), major
, minor
);
1090 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1091 fd
= open(devname
, flags
);
1096 fd
= open(dev
, flags
);
1100 int open_dev_flags(char *devnm
, int flags
)
1105 devid
= devnm2devid(devnm
);
1106 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1107 return dev_open(buf
, flags
);
1110 int open_dev(char *devnm
)
1112 return open_dev_flags(devnm
, O_RDONLY
);
1115 int open_dev_excl(char *devnm
)
1120 dev_t devid
= devnm2devid(devnm
);
1123 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1124 for (i
= 0; i
< 25; i
++) {
1125 int fd
= dev_open(buf
, flags
|O_EXCL
);
1128 if (errno
== EACCES
&& flags
== O_RDWR
) {
1141 int same_dev(char *one
, char *two
)
1143 struct stat st1
, st2
;
1144 if (stat(one
, &st1
) != 0)
1146 if (stat(two
, &st2
) != 0)
1148 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1150 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1152 return st1
.st_rdev
== st2
.st_rdev
;
1155 void wait_for(char *dev
, int fd
)
1158 struct stat stb_want
;
1161 if (fstat(fd
, &stb_want
) != 0 ||
1162 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1165 for (i
= 0; i
< 25; i
++) {
1167 if (stat(dev
, &stb
) == 0 &&
1168 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1169 (stb
.st_rdev
== stb_want
.st_rdev
))
1176 dprintf("timeout waiting for %s\n", dev
);
1179 struct superswitch
*superlist
[] =
1182 &super_ddf
, &super_imsm
,
1187 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1189 mdu_array_info_t array
;
1192 struct supertype
*st
= NULL
;
1197 char *subarray
= NULL
;
1198 char container
[32] = "";
1200 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1203 vers
= sra
->array
.major_version
;
1204 minor
= sra
->array
.minor_version
;
1205 verstr
= sra
->text_version
;
1207 if (md_get_array_info(fd
, &array
))
1208 array
.major_version
= array
.minor_version
= 0;
1209 vers
= array
.major_version
;
1210 minor
= array
.minor_version
;
1215 sprintf(version
, "%d.%d", vers
, minor
);
1218 if (minor
== -2 && is_subarray(verstr
)) {
1219 char *dev
= verstr
+1;
1221 subarray
= strchr(dev
, '/');
1224 subarray
= xstrdup(subarray
);
1226 strcpy(container
, dev
);
1228 sra
= sysfs_read(-1, container
, GET_VERSION
);
1229 if (sra
&& sra
->text_version
[0])
1230 verstr
= sra
->text_version
;
1232 verstr
= "-no-metadata-";
1235 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1236 st
= superlist
[i
]->match_metadata_desc(verstr
);
1242 *subarrayp
= subarray
;
1243 strcpy(st
->container_devnm
, container
);
1244 strcpy(st
->devnm
, fd2devnm(fd
));
1251 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1256 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1257 fd
= dev_open(buf
, O_RDONLY
);
1260 if (get_dev_size(fd
, NULL
, size
)) {
1268 struct supertype
*dup_super(struct supertype
*orig
)
1270 struct supertype
*st
;
1274 st
= xcalloc(1, sizeof(*st
));
1276 st
->max_devs
= orig
->max_devs
;
1277 st
->minor_version
= orig
->minor_version
;
1278 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1279 st
->data_offset
= orig
->data_offset
;
1285 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1287 /* try each load_super to find the best match,
1288 * and return the best superswitch
1290 struct superswitch
*ss
;
1291 struct supertype
*st
;
1292 unsigned int besttime
= 0;
1296 st
= xcalloc(1, sizeof(*st
));
1297 st
->container_devnm
[0] = 0;
1299 for (i
= 0; superlist
[i
]; i
++) {
1302 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1304 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1306 memset(st
, 0, sizeof(*st
));
1307 st
->ignore_hw_compat
= 1;
1308 rv
= ss
->load_super(st
, fd
, NULL
);
1311 st
->ss
->getinfo_super(st
, &info
, NULL
);
1312 if (bestsuper
== -1 ||
1313 besttime
< info
.array
.ctime
) {
1315 besttime
= info
.array
.ctime
;
1320 if (bestsuper
!= -1) {
1322 memset(st
, 0, sizeof(*st
));
1323 st
->ignore_hw_compat
= 1;
1324 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1326 superlist
[bestsuper
]->free_super(st
);
1334 /* Return size of device in bytes */
1335 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1337 unsigned long long ldsize
;
1340 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1341 ldsize
= (unsigned long long)st
.st_size
;
1344 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1347 unsigned long dsize
;
1348 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1353 pr_err("Cannot get size of %s: %s\n",
1354 dname
, strerror(errno
));
1362 /* Return sector size of device in bytes */
1363 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1365 unsigned int sectsize
;
1367 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1369 pr_err("Cannot get sector size of %s: %s\n",
1370 dname
, strerror(errno
));
1374 *sectsizep
= sectsize
;
1378 /* Return true if this can only be a container, not a member device.
1379 * i.e. is and md device and size is zero
1381 int must_be_container(int fd
)
1384 unsigned long long size
;
1386 mdi
= sysfs_read(fd
, NULL
, GET_VERSION
);
1391 if (get_dev_size(fd
, NULL
, &size
) == 0)
1398 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1399 * Returns: 1 if successful
1400 * -1 for unknown partition type
1401 * 0 for other errors
1403 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1406 unsigned char empty_gpt_entry
[16]= {0};
1407 struct GPT_part_entry
*part
;
1409 unsigned long long curr_part_end
;
1410 unsigned all_partitions
, entry_size
;
1412 unsigned int sector_size
= 0;
1416 BUILD_BUG_ON(sizeof(gpt
) != 512);
1417 /* skip protective MBR */
1418 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1420 lseek(fd
, sector_size
, SEEK_SET
);
1421 /* read GPT header */
1422 if (read(fd
, &gpt
, 512) != 512)
1425 /* get the number of partition entries and the entry size */
1426 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1427 entry_size
= __le32_to_cpu(gpt
.part_size
);
1429 /* Check GPT signature*/
1430 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1434 if (all_partitions
> 1024 ||
1435 entry_size
> sizeof(buf
))
1438 part
= (struct GPT_part_entry
*)buf
;
1440 /* set offset to third block (GPT entries) */
1441 lseek(fd
, sector_size
*2, SEEK_SET
);
1442 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1443 /* read partition entry */
1444 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1447 /* is this valid partition? */
1448 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1449 /* check the last lba for the current partition */
1450 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1451 if (curr_part_end
> *endofpart
)
1452 *endofpart
= curr_part_end
;
1459 /* Sets endofpart parameter to the last block used by the last partition on the device.
1460 * Returns: 1 if successful
1461 * -1 for unknown partition type
1462 * 0 for other errors
1464 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1466 struct MBR boot_sect
;
1467 unsigned long long curr_part_end
;
1469 unsigned int sector_size
;
1474 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1477 if (read(fd
, &boot_sect
, 512) != 512)
1480 /* check MBP signature */
1481 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1483 /* found the correct signature */
1485 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1487 * Have to make every access through boot_sect rather
1488 * than using a pointer to the partition table (or an
1489 * entry), since the entries are not properly aligned.
1492 /* check for GPT type */
1493 if (boot_sect
.parts
[part_nr
].part_type
==
1494 MBR_GPT_PARTITION_TYPE
) {
1495 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1498 /* check the last used lba for the current partition */
1500 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1501 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1502 if (curr_part_end
> *endofpart
)
1503 *endofpart
= curr_part_end
;
1506 /* Unknown partition table */
1509 /* calculate number of 512-byte blocks */
1510 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1511 *endofpart
*= (sector_size
/ 512);
1516 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1517 unsigned long long size
)
1520 * Check where the last partition ends
1522 unsigned long long endofpart
;
1524 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1525 /* There appears to be a partition table here */
1526 if (freesize
== 0) {
1527 /* partitions will not be visible in new device */
1528 pr_err("partition table exists on %s but will be lost or\n"
1529 " meaningless after creating array\n",
1532 } else if (endofpart
> freesize
) {
1533 /* last partition overlaps metadata */
1534 pr_err("metadata will over-write last partition on %s.\n",
1537 } else if (size
&& endofpart
> size
) {
1538 /* partitions will be truncated in new device */
1539 pr_err("array size is too small to cover all partitions on %s.\n",
1547 int open_container(int fd
)
1549 /* 'fd' is a block device. Find out if it is in use
1550 * by a container, and return an open fd on that container.
1561 if (fstat(fd
, &st
) != 0)
1563 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1564 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1565 e
= path
+ strlen(path
);
1567 dir
= opendir(path
);
1570 while ((de
= readdir(dir
))) {
1573 if (de
->d_name
[0] == '.')
1575 /* Need to make sure it is a container and not a volume */
1576 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1577 dfd
= open(path
, O_RDONLY
);
1580 n
= read(dfd
, buf
, sizeof(buf
));
1582 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1585 if (strncmp(buf
, "external", 8) != 0 ||
1589 sprintf(e
, "/%s/dev", de
->d_name
);
1590 dfd
= open(path
, O_RDONLY
);
1593 n
= read(dfd
, buf
, sizeof(buf
));
1595 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1598 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1600 sprintf(buf
, "%d:%d", major
, minor
);
1601 dfd
= dev_open(buf
, O_RDONLY
);
1611 struct superswitch
*version_to_superswitch(char *vers
)
1615 for (i
= 0; superlist
[i
]; i
++) {
1616 struct superswitch
*ss
= superlist
[i
];
1618 if (strcmp(vers
, ss
->name
) == 0)
1625 int metadata_container_matches(char *metadata
, char *devnm
)
1627 /* Check if 'devnm' is the container named in 'metadata'
1629 * /containername/componentname or
1630 * -containername/componentname
1633 if (*metadata
!= '/' && *metadata
!= '-')
1636 if (strncmp(metadata
+1, devnm
, l
) != 0)
1638 if (metadata
[l
+1] != '/')
1643 int metadata_subdev_matches(char *metadata
, char *devnm
)
1645 /* Check if 'devnm' is the subdev named in 'metadata'
1647 * /containername/subdev or
1648 * -containername/subdev
1651 if (*metadata
!= '/' && *metadata
!= '-')
1653 sl
= strchr(metadata
+1, '/');
1656 if (strcmp(sl
+1, devnm
) == 0)
1661 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1663 if (mdstat
->metadata_version
== NULL
||
1664 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1665 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1671 int is_subarray_active(char *subarray
, char *container
)
1673 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1674 struct mdstat_ent
*ent
;
1676 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1677 if (is_container_member(ent
, container
))
1678 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1681 free_mdstat(mdstat
);
1686 /* open_subarray - opens a subarray in a container
1687 * @dev: container device name
1688 * @st: empty supertype
1689 * @quiet: block reporting errors flag
1691 * On success returns an fd to a container and fills in *st
1693 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1696 struct mdinfo
*info
;
1700 fd
= open(dev
, O_RDWR
|O_EXCL
);
1703 pr_err("Couldn't open %s, aborting\n",
1708 _devnm
= fd2devnm(fd
);
1709 if (_devnm
== NULL
) {
1711 pr_err("Failed to determine device number for %s\n",
1715 strcpy(st
->devnm
, _devnm
);
1717 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1720 pr_err("Failed to read sysfs for %s\n",
1725 if (mdi
->array
.level
!= UnSet
) {
1727 pr_err("%s is not a container\n", dev
);
1731 st
->ss
= version_to_superswitch(mdi
->text_version
);
1734 pr_err("Operation not supported for %s metadata\n",
1739 if (st
->devnm
[0] == 0) {
1741 pr_err("Failed to allocate device name\n");
1745 if (!st
->ss
->load_container
) {
1747 pr_err("%s is not a container\n", dev
);
1751 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1753 pr_err("Failed to load metadata for %s\n",
1758 info
= st
->ss
->container_content(st
, subarray
);
1761 pr_err("Failed to find subarray-%s in %s\n",
1771 st
->ss
->free_super(st
);
1784 int add_disk(int mdfd
, struct supertype
*st
,
1785 struct mdinfo
*sra
, struct mdinfo
*info
)
1787 /* Add a device to an array, in one of 2 ways. */
1790 if (st
->ss
->external
) {
1791 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1792 info
->recovery_start
= MaxSector
;
1794 info
->recovery_start
= 0;
1795 rv
= sysfs_add_disk(sra
, info
, 0);
1798 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1802 sd2
= xmalloc(sizeof(*sd2
));
1804 sd2
->next
= sra
->devs
;
1809 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1813 int remove_disk(int mdfd
, struct supertype
*st
,
1814 struct mdinfo
*sra
, struct mdinfo
*info
)
1818 /* Remove the disk given by 'info' from the array */
1819 if (st
->ss
->external
)
1820 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1822 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1827 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1829 int cnt
= force
? 500 : 5;
1832 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1833 * outstanding IO requests to the device.
1834 * In this case, it can be helpful to wait a little while,
1835 * up to 5 seconds if 'force' is set, or 50 msec if not.
1837 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1845 int sys_hot_remove_disk(int statefd
, int force
)
1847 int cnt
= force
? 500 : 5;
1850 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1854 return ret
== 6 ? 0 : -1;
1857 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1859 /* Initialise kernel's knowledge of array.
1860 * This varies between externally managed arrays
1863 mdu_array_info_t inf
;
1866 if (st
->ss
->external
)
1867 return sysfs_set_array(info
, 9003);
1869 memset(&inf
, 0, sizeof(inf
));
1870 inf
.major_version
= info
->array
.major_version
;
1871 inf
.minor_version
= info
->array
.minor_version
;
1872 rv
= md_set_array_info(mdfd
, &inf
);
1877 unsigned long long min_recovery_start(struct mdinfo
*array
)
1879 /* find the minimum recovery_start in an array for metadata
1880 * formats that only record per-array recovery progress instead
1883 unsigned long long recovery_start
= MaxSector
;
1886 for (d
= array
->devs
; d
; d
= d
->next
)
1887 recovery_start
= min(recovery_start
, d
->recovery_start
);
1889 return recovery_start
;
1892 int mdmon_pid(char *devnm
)
1899 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1901 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1905 n
= read(fd
, pid
, 9);
1912 int mdmon_running(char *devnm
)
1914 int pid
= mdmon_pid(devnm
);
1917 if (kill(pid
, 0) == 0)
1922 int start_mdmon(char *devnm
)
1936 if (check_env("MDADM_NO_MDMON"))
1939 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1943 sl
= strrchr(pathbuf
, '/');
1948 strcpy(sl
, "mdmon");
1952 /* First try to run systemctl */
1953 if (!check_env("MDADM_NO_SYSTEMCTL"))
1956 /* FIXME yuk. CLOSE_EXEC?? */
1958 for (i
= 3; skipped
< 20; i
++)
1964 /* Don't want to see error messages from
1965 * systemctl. If the service doesn't exist,
1966 * we start mdmon ourselves.
1969 open("/dev/null", O_WRONLY
);
1970 snprintf(pathbuf
, sizeof(pathbuf
), "mdmon@%s.service",
1972 status
= execl("/usr/bin/systemctl", "systemctl",
1975 status
= execl("/bin/systemctl", "systemctl", "start",
1978 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1980 default: /* parent - good */
1981 pid
= wait(&status
);
1982 if (pid
>= 0 && status
== 0)
1986 /* That failed, try running mdmon directly */
1989 /* FIXME yuk. CLOSE_EXEC?? */
1991 for (i
= 3; skipped
< 20; i
++)
1997 for (i
= 0; paths
[i
]; i
++)
1999 execl(paths
[i
], paths
[i
],
2003 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
2005 default: /* parent - good */
2006 pid
= wait(&status
);
2007 if (pid
< 0 || status
!= 0) {
2008 pr_err("failed to launch mdmon. Array remains readonly\n");
2015 __u32
random32(void)
2018 int rfd
= open("/dev/urandom", O_RDONLY
);
2019 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
2026 void random_uuid(__u8
*buf
)
2031 fd
= open("/dev/urandom", O_RDONLY
);
2034 len
= read(fd
, buf
, 16);
2042 for (i
= 0; i
< 4; i
++)
2047 int flush_metadata_updates(struct supertype
*st
)
2051 st
->update_tail
= NULL
;
2055 sfd
= connect_monitor(st
->container_devnm
);
2059 while (st
->updates
) {
2060 struct metadata_update
*mu
= st
->updates
;
2061 st
->updates
= mu
->next
;
2063 send_message(sfd
, mu
, 0);
2071 st
->update_tail
= NULL
;
2075 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2078 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2083 mu
->space_list
= NULL
;
2085 *st
->update_tail
= mu
;
2086 st
->update_tail
= &mu
->next
;
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
)
2227 static struct cmap_hooks
*cmap_hooks
= NULL
;
2228 static int is_cmap_hooks_ready
= 0;
2230 void set_cmap_hooks(void)
2232 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2233 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2234 if (!cmap_hooks
->cmap_handle
)
2237 cmap_hooks
->initialize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2238 cmap_hooks
->get_string
= dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2239 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2241 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2242 !cmap_hooks
->finalize
)
2243 dlclose(cmap_hooks
->cmap_handle
);
2245 is_cmap_hooks_ready
= 1;
2248 int get_cluster_name(char **cluster_name
)
2251 cmap_handle_t handle
;
2253 if (!is_cmap_hooks_ready
)
2256 rv
= cmap_hooks
->initialize(&handle
);
2260 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2262 free(*cluster_name
);
2269 cmap_hooks
->finalize(handle
);
2274 void set_dlm_hooks(void)
2276 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2277 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2278 if (!dlm_hooks
->dlm_handle
)
2281 dlm_hooks
->create_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2282 dlm_hooks
->release_lockspace
= dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2283 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2284 dlm_hooks
->ls_unlock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock");
2285 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2286 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2288 if (!dlm_hooks
->create_lockspace
|| !dlm_hooks
->ls_lock
||
2289 !dlm_hooks
->ls_unlock
|| !dlm_hooks
->release_lockspace
||
2290 !dlm_hooks
->ls_get_fd
|| !dlm_hooks
->dispatch
)
2291 dlclose(dlm_hooks
->dlm_handle
);
2293 is_dlm_hooks_ready
= 1;
2296 void set_hooks(void)