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(void)
135 int flags
= LKF_NOQUEUE
;
138 if (!dlm_funs_ready()) {
139 pr_err("Something wrong with dlm library\n");
143 ret
= get_cluster_name(&cluster_name
);
145 pr_err("The md can't get cluster name\n");
149 dlm_lock_res
= xmalloc(sizeof(struct dlm_lock_resource
));
150 dlm_lock_res
->ls
= dlm_hooks
->open_lockspace(cluster_name
);
151 if (!dlm_lock_res
->ls
) {
152 dlm_lock_res
->ls
= dlm_hooks
->create_lockspace(cluster_name
, O_RDWR
);
153 if (!dlm_lock_res
->ls
) {
154 pr_err("%s failed to create lockspace\n", cluster_name
);
158 pr_err("open existed %s lockspace\n", cluster_name
);
161 snprintf(str
, 64, "bitmap%s", cluster_name
);
163 ret
= dlm_hooks
->ls_lock(dlm_lock_res
->ls
, LKM_PWMODE
,
164 &dlm_lock_res
->lksb
, flags
, str
, strlen(str
),
165 0, dlm_ast
, dlm_lock_res
, NULL
, NULL
);
167 pr_err("error %d when get PW mode on lock %s\n", errno
, str
);
168 /* let's try several times if EAGAIN happened */
169 if (dlm_lock_res
->lksb
.sb_status
== EAGAIN
&& retry_count
< 10) {
174 dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
178 /* Wait for it to complete */
179 poll_for_ast(dlm_lock_res
->ls
);
181 if (dlm_lock_res
->lksb
.sb_status
) {
182 pr_err("failed to lock cluster\n");
188 int cluster_release_dlmlock(void)
195 if (!dlm_lock_res
->lksb
.sb_lkid
)
198 ret
= dlm_hooks
->ls_unlock_wait(dlm_lock_res
->ls
,
199 dlm_lock_res
->lksb
.sb_lkid
, 0,
200 &dlm_lock_res
->lksb
);
202 pr_err("error %d happened when unlock\n", errno
);
203 /* XXX make sure the lock is unlocked eventually */
207 /* Wait for it to complete */
208 poll_for_ast(dlm_lock_res
->ls
);
210 errno
= dlm_lock_res
->lksb
.sb_status
;
211 if (errno
!= EUNLOCK
) {
212 pr_err("error %d happened in ast when unlock lockspace\n",
214 /* XXX make sure the lockspace is unlocked eventually */
218 ret
= dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
220 pr_err("error %d happened when release lockspace\n", errno
);
221 /* XXX make sure the lockspace is released eventually */
230 int md_array_valid(int fd
)
235 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
237 if (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
, RAID_VERSION
);
254 int md_array_active(int fd
)
257 struct mdu_array_info_s array
;
260 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
262 if (!md_array_is_active(sra
))
268 * GET_ARRAY_INFO doesn't provide access to the proper state
269 * information, so fallback to a basic check for raid_disks != 0
271 ret
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
277 int md_array_is_active(struct mdinfo
*info
)
279 return (info
->array_state
!= ARRAY_CLEAR
&&
280 info
->array_state
!= ARRAY_INACTIVE
&&
281 info
->array_state
!= ARRAY_UNKNOWN_STATE
);
285 * Get array info from the kernel. Longer term we want to deprecate the
286 * ioctl and get it from sysfs.
288 int md_get_array_info(int fd
, struct mdu_array_info_s
*array
)
290 return ioctl(fd
, GET_ARRAY_INFO
, array
);
296 int md_set_array_info(int fd
, struct mdu_array_info_s
*array
)
298 return ioctl(fd
, SET_ARRAY_INFO
, array
);
302 * Get disk info from the kernel.
304 int md_get_disk_info(int fd
, struct mdu_disk_info_s
*disk
)
306 return ioctl(fd
, GET_DISK_INFO
, disk
);
309 int get_linux_version()
313 int a
= 0, b
= 0,c
= 0;
318 a
= strtoul(cp
, &cp
, 10);
320 b
= strtoul(cp
+1, &cp
, 10);
322 c
= strtoul(cp
+1, &cp
, 10);
324 return (a
*1000000)+(b
*1000)+c
;
327 int mdadm_version(char *version
)
335 cp
= strchr(version
, '-');
336 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
339 a
= strtoul(cp
, &cp
, 10);
342 b
= strtoul(cp
+1, &cp
, 10);
344 c
= strtoul(cp
+1, &cp
, 10);
347 if (*cp
!= ' ' && *cp
!= '-')
349 return (a
*1000000)+(b
*1000)+c
;
352 unsigned long long parse_size(char *size
)
354 /* parse 'size' which should be a number optionally
355 * followed by 'K', 'M'. 'G' or 'T'.
356 * Without a suffix, K is assumed.
357 * Number returned is in sectors (half-K)
358 * INVALID_SECTORS returned on error.
361 long long s
= strtoll(size
, &c
, 10);
375 s
*= 1024 * 1024 * 2;
379 s
*= 1024 * 1024 * 1024 * 2LL;
381 case 's': /* sectors */
392 int is_near_layout_10(int layout
)
396 fc
= (layout
>> 8) & 255;
397 fo
= layout
& (1 << 16);
398 if (fc
> 1 || fo
> 0)
403 int parse_layout_10(char *layout
)
407 /* Parse the layout string for raid10 */
408 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
409 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
410 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
414 if (layout
[0] == 'n')
416 else if (layout
[0] == 'o')
417 rv
= 0x10000 + (copies
<<8) + 1;
419 rv
= 1 + (copies
<<8);
423 int parse_layout_faulty(char *layout
)
425 /* Parse the layout string for 'faulty' */
426 int ln
= strcspn(layout
, "0123456789");
427 char *m
= xstrdup(layout
);
430 mode
= map_name(faultylayout
, m
);
434 return mode
| (atoi(layout
+ln
)<< ModeShift
);
437 long parse_num(char *num
)
439 /* Either return a valid number, or -1 */
441 long rv
= strtol(num
, &c
, 10);
442 if (rv
< 0 || *c
|| !num
[0])
448 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
451 *slot
= strtoul(input
, &dev
, 10);
452 if (dev
== input
|| dev
[0] != ':')
458 void remove_partitions(int fd
)
460 /* remove partitions from this block devices.
461 * This is used for components added to an array
463 #ifdef BLKPG_DEL_PARTITION
464 struct blkpg_ioctl_arg a
;
465 struct blkpg_partition p
;
467 a
.op
= BLKPG_DEL_PARTITION
;
469 a
.datalen
= sizeof(p
);
471 memset(a
.data
, 0, a
.datalen
);
472 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
473 ioctl(fd
, BLKPG
, &a
);
477 int test_partition(int fd
)
479 /* Check if fd is a whole-disk or a partition.
480 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
481 * will return ENXIO on an invalid partition number.
483 struct blkpg_ioctl_arg a
;
484 struct blkpg_partition p
;
485 a
.op
= BLKPG_DEL_PARTITION
;
487 a
.datalen
= sizeof(p
);
489 memset(a
.data
, 0, a
.datalen
);
491 if (ioctl(fd
, BLKPG
, &a
) == 0)
492 /* Very unlikely, but not a partition */
494 if (errno
== ENXIO
|| errno
== ENOTTY
)
495 /* not a partition */
501 int test_partition_from_id(dev_t id
)
506 sprintf(buf
, "%d:%d", major(id
), minor(id
));
507 fd
= dev_open(buf
, O_RDONLY
);
510 rv
= test_partition(fd
);
515 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
521 for (i
= 0; i
< raid_disks
; i
++)
522 avail_disks
+= !!avail
[i
];
526 /* This is the tricky one - we need to check
527 * which actual disks are present.
529 copies
= (layout
&255)* ((layout
>>8) & 255);
532 /* there must be one of the 'copies' form 'first' */
539 this = (this+1) % raid_disks
;
543 first
= (first
+(layout
&255)) % raid_disks
;
544 } while (first
!= 0);
547 case LEVEL_MULTIPATH
:
548 return avail_disks
>= 1;
551 return avail_disks
== raid_disks
;
553 return avail_disks
>= 1;
555 if (avail_disks
== raid_disks
- 1 &&
556 !avail
[raid_disks
- 1])
557 /* If just the parity device is missing, then we
558 * have enough, even if not clean
564 return avail_disks
>= raid_disks
-1;
566 return avail_disks
>= raid_disks
;
569 return avail_disks
>= raid_disks
-2;
571 return avail_disks
>= raid_disks
;
577 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
584 copy_uuid(uuid
, id
, swap
);
585 for (i
= 0; i
< 4; i
++) {
588 for (j
= 3; j
>= 0; j
--) {
589 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
597 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
,
600 // dirty hack to work around an issue with super1 superblocks...
601 // super1 superblocks need swapuuid set in order for assembly to
602 // work, but can't have it set if we want this printout to match
603 // all the other uuid printouts in super1.c, so we force swapuuid
604 // to 1 to make our printout match the rest of super1
605 #if __BYTE_ORDER == BIG_ENDIAN
606 return __fname_from_uuid(info
->uuid
, 1, buf
, sep
);
608 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 :
609 st
->ss
->swapuuid
, buf
, sep
);
613 int check_ext2(int fd
, char *name
)
616 * Check for an ext2fs file system.
617 * Superblock is always 1K at 1K offset
619 * s_magic is le16 at 56 == 0xEF53
620 * report mtime - le32 at 44
622 * logblksize - le32 at 24
624 unsigned char sb
[1024];
626 unsigned long long size
;
628 if (lseek(fd
, 1024,0)!= 1024)
630 if (read(fd
, sb
, 1024)!= 1024)
632 if (sb
[56] != 0x53 || sb
[57] != 0xef)
635 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
636 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
637 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
639 pr_err("%s appears to contain an ext2fs file system\n",
641 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
645 int check_reiser(int fd
, char *name
)
648 * superblock is at 64K
650 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
653 unsigned char sb
[1024];
654 unsigned long long size
;
655 if (lseek(fd
, 64*1024, 0) != 64*1024)
657 if (read(fd
, sb
, 1024) != 1024)
659 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
660 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
662 pr_err("%s appears to contain a reiserfs file system\n",name
);
663 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
664 cont_err("size = %lluK\n", size
*4);
669 int check_raid(int fd
, char *name
)
674 struct supertype
*st
= guess_super(fd
);
678 if (st
->ss
->add_to_super
!= NULL
) {
679 st
->ss
->load_super(st
, fd
, name
);
680 /* Looks like a raid array .. */
681 pr_err("%s appears to be part of a raid array:\n", name
);
682 st
->ss
->getinfo_super(st
, &info
, NULL
);
683 st
->ss
->free_super(st
);
684 crtime
= info
.array
.ctime
;
685 level
= map_num(pers
, info
.array
.level
);
688 cont_err("level=%s devices=%d ctime=%s",
689 level
, info
.array
.raid_disks
, ctime(&crtime
));
691 /* Looks like GPT or MBR */
692 pr_err("partition table exists on %s\n", name
);
697 int fstat_is_blkdev(int fd
, char *devname
, dev_t
*rdev
)
701 if (fstat(fd
, &stb
) != 0) {
702 pr_err("fstat failed for %s: %s\n", devname
, strerror(errno
));
705 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
706 pr_err("%s is not a block device.\n", devname
);
714 int stat_is_blkdev(char *devname
, dev_t
*rdev
)
718 if (stat(devname
, &stb
) != 0) {
719 pr_err("stat failed for %s: %s\n", devname
, strerror(errno
));
722 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
723 pr_err("%s is not a block device.\n", devname
);
735 for (i
= 0; i
< 5; i
++) {
737 fprintf(stderr
, "%s%s", mesg
, add
);
739 if (fgets(buf
, 100, stdin
)==NULL
)
741 if (buf
[0]=='y' || buf
[0]=='Y')
743 if (buf
[0]=='n' || buf
[0]=='N')
747 pr_err("assuming 'no'\n");
751 int is_standard(char *dev
, int *nump
)
753 /* tests if dev is a "standard" md dev name.
754 * i.e if the last component is "/dNN" or "/mdNN",
755 * where NN is a string of digits
756 * Returns 1 if a partitionable standard,
757 * -1 if non-partitonable,
758 * 0 if not a standard name.
760 char *d
= strrchr(dev
, '/');
765 if (strncmp(d
, "/d",2) == 0)
766 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
767 else if (strncmp(d
, "/md_d", 5) == 0)
768 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
769 else if (strncmp(d
, "/md", 3) == 0)
770 d
+= 3, type
= -1; /* /dev/mdN */
771 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
772 d
+= 1, type
= -1; /* /dev/md/N */
782 if (nump
) *nump
= num
;
787 unsigned long calc_csum(void *super
, int bytes
)
789 unsigned long long newcsum
= 0;
792 unsigned int *superc
= (unsigned int*) super
;
794 for(i
= 0; i
< bytes
/4; i
++)
795 newcsum
+= superc
[i
];
796 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
798 /* The in-kernel checksum calculation is always 16bit on
799 * the alpha, though it is 32 bit on i386...
800 * I wonder what it is elsewhere... (it uses an API in
801 * a way that it shouldn't).
803 csum
= (csum
& 0xffff) + (csum
>> 16);
804 csum
= (csum
& 0xffff) + (csum
>> 16);
809 char *human_size(long long bytes
)
813 /* We convert bytes to either centi-M{ega,ibi}bytes,
814 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
815 * with appropriate rounding, and then print
816 * 1/100th of those as a decimal.
817 * We allow upto 2048Megabytes before converting to
818 * gigabytes and 2048Gigabytes before converting to
819 * terabytes, as that shows more precision and isn't
820 * too large a number.
823 if (bytes
< 5000*1024)
825 else if (bytes
< 2*1024LL*1024LL*1024LL) {
826 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
827 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
828 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
829 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
830 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
831 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
832 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
833 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
834 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
836 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
837 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
838 snprintf(buf
, sizeof(buf
), " (%ld.%02ld TiB %ld.%02ld TB)",
839 cTiB
/100, cTiB
% 100, cTB
/100, cTB
% 100);
844 char *human_size_brief(long long bytes
, int prefix
)
848 /* We convert bytes to either centi-M{ega,ibi}bytes,
849 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
850 * with appropriate rounding, and then print
851 * 1/100th of those as a decimal.
852 * We allow upto 2048Megabytes before converting to
853 * gigabytes and 2048Gigabytes before converting to
854 * terabytes, as that shows more precision and isn't
855 * too large a number.
857 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
858 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
861 if (bytes
< 5000*1024)
863 else if (prefix
== IEC
) {
864 if (bytes
< 2*1024LL*1024LL*1024LL) {
865 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
866 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
867 cMiB
/100, cMiB
% 100);
868 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
869 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
870 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
871 cGiB
/100, cGiB
% 100);
873 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
874 snprintf(buf
, sizeof(buf
), "%ld.%02ldTiB",
875 cTiB
/100, cTiB
% 100);
878 else if (prefix
== JEDEC
) {
879 if (bytes
< 2*1024LL*1024LL*1024LL) {
880 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
881 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
883 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
884 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
885 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
888 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
889 snprintf(buf
, sizeof(buf
), "%ld.%02ldTB",
899 void print_r10_layout(int layout
)
901 int near
= layout
& 255;
902 int far
= (layout
>> 8) & 255;
903 int offset
= (layout
&0x10000);
907 printf("%s near=%d", sep
, near
);
911 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
913 printf("NO REDUNDANCY");
916 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
917 int chunksize
, unsigned long long devsize
)
921 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
922 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
925 int get_data_disks(int level
, int layout
, int raid_disks
)
929 case 0: data_disks
= raid_disks
;
931 case 1: data_disks
= 1;
934 case 5: data_disks
= raid_disks
- 1;
936 case 6: data_disks
= raid_disks
- 2;
938 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
945 dev_t
devnm2devid(char *devnm
)
947 /* First look in /sys/block/$DEVNM/dev for %d:%d
948 * If that fails, try parsing out a number
955 sprintf(path
, "/sys/block/%s/dev", devnm
);
956 fd
= open(path
, O_RDONLY
);
959 int n
= read(fd
, buf
, sizeof(buf
));
963 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
964 return makedev(mjr
, mnr
);
966 if (strncmp(devnm
, "md_d", 4) == 0 &&
968 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
969 ep
> devnm
&& *ep
== 0)
970 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
972 if (strncmp(devnm
, "md", 2) == 0 &&
974 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
975 ep
> devnm
&& *ep
== 0)
976 return makedev(MD_MAJOR
, mnr
);
981 char *get_md_name(char *devnm
)
983 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
984 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
986 static char devname
[50];
988 dev_t rdev
= devnm2devid(devnm
);
993 if (strncmp(devnm
, "md_", 3) == 0) {
994 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
996 if (stat(devname
, &stb
) == 0 &&
997 (S_IFMT
&stb
.st_mode
) == S_IFBLK
&& (stb
.st_rdev
== rdev
))
1000 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
1001 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
1002 (stb
.st_rdev
== rdev
))
1005 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
1006 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
1007 (stb
.st_rdev
== rdev
))
1010 dn
= map_dev(major(rdev
), minor(rdev
), 0);
1013 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
1014 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
1015 if (errno
!= EEXIST
)
1018 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
1019 (stb
.st_rdev
== rdev
))
1025 void put_md_name(char *name
)
1027 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1031 int get_maj_min(char *dev
, int *major
, int *minor
)
1034 *major
= strtoul(dev
, &e
, 0);
1035 return (e
> dev
&& *e
== ':' && e
[1] &&
1036 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1040 int dev_open(char *dev
, int flags
)
1042 /* like 'open', but if 'dev' matches %d:%d, create a temp
1043 * block device and open that
1054 if (get_maj_min(dev
, &major
, &minor
)) {
1055 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1056 (int)getpid(), major
, minor
);
1057 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1058 fd
= open(devname
, flags
);
1062 /* Try /tmp as /dev appear to be read-only */
1063 snprintf(devname
, sizeof(devname
),
1064 "/tmp/.tmp.md.%d:%d:%d",
1065 (int)getpid(), major
, minor
);
1066 if (mknod(devname
, S_IFBLK
|0600,
1067 makedev(major
, minor
)) == 0) {
1068 fd
= open(devname
, flags
);
1073 fd
= open(dev
, flags
);
1077 int open_dev_flags(char *devnm
, int flags
)
1082 devid
= devnm2devid(devnm
);
1083 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1084 return dev_open(buf
, flags
);
1087 int open_dev(char *devnm
)
1089 return open_dev_flags(devnm
, O_RDONLY
);
1092 int open_dev_excl(char *devnm
)
1097 dev_t devid
= devnm2devid(devnm
);
1100 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1101 for (i
= 0; i
< 25; i
++) {
1102 int fd
= dev_open(buf
, flags
|O_EXCL
);
1105 if (errno
== EACCES
&& flags
== O_RDWR
) {
1118 int same_dev(char *one
, char *two
)
1120 struct stat st1
, st2
;
1121 if (stat(one
, &st1
) != 0)
1123 if (stat(two
, &st2
) != 0)
1125 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1127 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1129 return st1
.st_rdev
== st2
.st_rdev
;
1132 void wait_for(char *dev
, int fd
)
1135 struct stat stb_want
;
1138 if (fstat(fd
, &stb_want
) != 0 ||
1139 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1142 for (i
= 0; i
< 25; i
++) {
1144 if (stat(dev
, &stb
) == 0 &&
1145 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1146 (stb
.st_rdev
== stb_want
.st_rdev
))
1153 pr_err("timeout waiting for %s\n", dev
);
1156 struct superswitch
*superlist
[] =
1159 &super_ddf
, &super_imsm
,
1164 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1166 mdu_array_info_t array
;
1169 struct supertype
*st
= NULL
;
1174 char *subarray
= NULL
;
1175 char container
[32] = "";
1177 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1180 vers
= sra
->array
.major_version
;
1181 minor
= sra
->array
.minor_version
;
1182 verstr
= sra
->text_version
;
1184 if (md_get_array_info(fd
, &array
))
1185 array
.major_version
= array
.minor_version
= 0;
1186 vers
= array
.major_version
;
1187 minor
= array
.minor_version
;
1192 sprintf(version
, "%d.%d", vers
, minor
);
1195 if (minor
== -2 && is_subarray(verstr
)) {
1196 char *dev
= verstr
+1;
1198 subarray
= strchr(dev
, '/');
1201 subarray
= xstrdup(subarray
);
1203 strcpy(container
, dev
);
1205 sra
= sysfs_read(-1, container
, GET_VERSION
);
1206 if (sra
&& sra
->text_version
[0])
1207 verstr
= sra
->text_version
;
1209 verstr
= "-no-metadata-";
1212 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1213 st
= superlist
[i
]->match_metadata_desc(verstr
);
1219 *subarrayp
= subarray
;
1220 strcpy(st
->container_devnm
, container
);
1221 strcpy(st
->devnm
, fd2devnm(fd
));
1228 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1233 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1234 fd
= dev_open(buf
, O_RDONLY
);
1237 if (get_dev_size(fd
, NULL
, size
)) {
1245 int dev_sector_size_from_id(dev_t id
, unsigned int *size
)
1250 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1251 fd
= dev_open(buf
, O_RDONLY
);
1254 if (get_dev_sector_size(fd
, NULL
, size
)) {
1262 struct supertype
*dup_super(struct supertype
*orig
)
1264 struct supertype
*st
;
1268 st
= xcalloc(1, sizeof(*st
));
1270 st
->max_devs
= orig
->max_devs
;
1271 st
->minor_version
= orig
->minor_version
;
1272 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1273 st
->data_offset
= orig
->data_offset
;
1279 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1281 /* try each load_super to find the best match,
1282 * and return the best superswitch
1284 struct superswitch
*ss
;
1285 struct supertype
*st
;
1286 unsigned int besttime
= 0;
1290 st
= xcalloc(1, sizeof(*st
));
1291 st
->container_devnm
[0] = 0;
1293 for (i
= 0; superlist
[i
]; i
++) {
1296 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1298 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1300 memset(st
, 0, sizeof(*st
));
1301 st
->ignore_hw_compat
= 1;
1302 rv
= ss
->load_super(st
, fd
, NULL
);
1305 st
->ss
->getinfo_super(st
, &info
, NULL
);
1306 if (bestsuper
== -1 ||
1307 besttime
< info
.array
.ctime
) {
1309 besttime
= info
.array
.ctime
;
1314 if (bestsuper
!= -1) {
1316 memset(st
, 0, sizeof(*st
));
1317 st
->ignore_hw_compat
= 1;
1318 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1320 superlist
[bestsuper
]->free_super(st
);
1328 /* Return size of device in bytes */
1329 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1331 unsigned long long ldsize
;
1334 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1335 ldsize
= (unsigned long long)st
.st_size
;
1338 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1341 unsigned long dsize
;
1342 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1347 pr_err("Cannot get size of %s: %s\n",
1348 dname
, strerror(errno
));
1356 /* Return sector size of device in bytes */
1357 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1359 unsigned int sectsize
;
1361 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1363 pr_err("Cannot get sector size of %s: %s\n",
1364 dname
, strerror(errno
));
1368 *sectsizep
= sectsize
;
1372 /* Return true if this can only be a container, not a member device.
1373 * i.e. is and md device and size is zero
1375 int must_be_container(int fd
)
1378 unsigned long long size
;
1380 mdi
= sysfs_read(fd
, NULL
, GET_VERSION
);
1385 if (get_dev_size(fd
, NULL
, &size
) == 0)
1392 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1393 * Returns: 1 if successful
1394 * -1 for unknown partition type
1395 * 0 for other errors
1397 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1400 unsigned char empty_gpt_entry
[16]= {0};
1401 struct GPT_part_entry
*part
;
1403 unsigned long long curr_part_end
;
1404 unsigned all_partitions
, entry_size
;
1406 unsigned int sector_size
= 0;
1410 BUILD_BUG_ON(sizeof(gpt
) != 512);
1411 /* skip protective MBR */
1412 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1414 lseek(fd
, sector_size
, SEEK_SET
);
1415 /* read GPT header */
1416 if (read(fd
, &gpt
, 512) != 512)
1419 /* get the number of partition entries and the entry size */
1420 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1421 entry_size
= __le32_to_cpu(gpt
.part_size
);
1423 /* Check GPT signature*/
1424 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1428 if (all_partitions
> 1024 ||
1429 entry_size
> sizeof(buf
))
1432 part
= (struct GPT_part_entry
*)buf
;
1434 /* set offset to third block (GPT entries) */
1435 lseek(fd
, sector_size
*2, SEEK_SET
);
1436 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1437 /* read partition entry */
1438 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1441 /* is this valid partition? */
1442 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1443 /* check the last lba for the current partition */
1444 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1445 if (curr_part_end
> *endofpart
)
1446 *endofpart
= curr_part_end
;
1453 /* Sets endofpart parameter to the last block used by the last partition on the device.
1454 * Returns: 1 if successful
1455 * -1 for unknown partition type
1456 * 0 for other errors
1458 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1460 struct MBR boot_sect
;
1461 unsigned long long curr_part_end
;
1463 unsigned int sector_size
;
1468 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1471 if (read(fd
, &boot_sect
, 512) != 512)
1474 /* check MBP signature */
1475 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1477 /* found the correct signature */
1479 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1481 * Have to make every access through boot_sect rather
1482 * than using a pointer to the partition table (or an
1483 * entry), since the entries are not properly aligned.
1486 /* check for GPT type */
1487 if (boot_sect
.parts
[part_nr
].part_type
==
1488 MBR_GPT_PARTITION_TYPE
) {
1489 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1492 /* check the last used lba for the current partition */
1494 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1495 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1496 if (curr_part_end
> *endofpart
)
1497 *endofpart
= curr_part_end
;
1500 /* Unknown partition table */
1503 /* calculate number of 512-byte blocks */
1504 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1505 *endofpart
*= (sector_size
/ 512);
1510 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1511 unsigned long long size
)
1514 * Check where the last partition ends
1516 unsigned long long endofpart
;
1518 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1519 /* There appears to be a partition table here */
1520 if (freesize
== 0) {
1521 /* partitions will not be visible in new device */
1522 pr_err("partition table exists on %s but will be lost or\n"
1523 " meaningless after creating array\n",
1526 } else if (endofpart
> freesize
) {
1527 /* last partition overlaps metadata */
1528 pr_err("metadata will over-write last partition on %s.\n",
1531 } else if (size
&& endofpart
> size
) {
1532 /* partitions will be truncated in new device */
1533 pr_err("array size is too small to cover all partitions on %s.\n",
1541 int open_container(int fd
)
1543 /* 'fd' is a block device. Find out if it is in use
1544 * by a container, and return an open fd on that container.
1555 if (fstat(fd
, &st
) != 0)
1557 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1558 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1559 e
= path
+ strlen(path
);
1561 dir
= opendir(path
);
1564 while ((de
= readdir(dir
))) {
1567 if (de
->d_name
[0] == '.')
1569 /* Need to make sure it is a container and not a volume */
1570 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1571 dfd
= open(path
, O_RDONLY
);
1574 n
= read(dfd
, buf
, sizeof(buf
));
1576 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1579 if (strncmp(buf
, "external", 8) != 0 ||
1583 sprintf(e
, "/%s/dev", de
->d_name
);
1584 dfd
= open(path
, O_RDONLY
);
1587 n
= read(dfd
, buf
, sizeof(buf
));
1589 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1592 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1594 sprintf(buf
, "%d:%d", major
, minor
);
1595 dfd
= dev_open(buf
, O_RDONLY
);
1605 struct superswitch
*version_to_superswitch(char *vers
)
1609 for (i
= 0; superlist
[i
]; i
++) {
1610 struct superswitch
*ss
= superlist
[i
];
1612 if (strcmp(vers
, ss
->name
) == 0)
1619 int metadata_container_matches(char *metadata
, char *devnm
)
1621 /* Check if 'devnm' is the container named in 'metadata'
1623 * /containername/componentname or
1624 * -containername/componentname
1627 if (*metadata
!= '/' && *metadata
!= '-')
1630 if (strncmp(metadata
+1, devnm
, l
) != 0)
1632 if (metadata
[l
+1] != '/')
1637 int metadata_subdev_matches(char *metadata
, char *devnm
)
1639 /* Check if 'devnm' is the subdev named in 'metadata'
1641 * /containername/subdev or
1642 * -containername/subdev
1645 if (*metadata
!= '/' && *metadata
!= '-')
1647 sl
= strchr(metadata
+1, '/');
1650 if (strcmp(sl
+1, devnm
) == 0)
1655 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1657 if (mdstat
->metadata_version
== NULL
||
1658 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1659 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1665 int is_subarray_active(char *subarray
, char *container
)
1667 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1668 struct mdstat_ent
*ent
;
1670 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1671 if (is_container_member(ent
, container
))
1672 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1675 free_mdstat(mdstat
);
1680 /* open_subarray - opens a subarray in a container
1681 * @dev: container device name
1682 * @st: empty supertype
1683 * @quiet: block reporting errors flag
1685 * On success returns an fd to a container and fills in *st
1687 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1690 struct mdinfo
*info
;
1694 fd
= open(dev
, O_RDWR
|O_EXCL
);
1697 pr_err("Couldn't open %s, aborting\n",
1702 _devnm
= fd2devnm(fd
);
1703 if (_devnm
== NULL
) {
1705 pr_err("Failed to determine device number for %s\n",
1709 strcpy(st
->devnm
, _devnm
);
1711 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1714 pr_err("Failed to read sysfs for %s\n",
1719 if (mdi
->array
.level
!= UnSet
) {
1721 pr_err("%s is not a container\n", dev
);
1725 st
->ss
= version_to_superswitch(mdi
->text_version
);
1728 pr_err("Operation not supported for %s metadata\n",
1733 if (st
->devnm
[0] == 0) {
1735 pr_err("Failed to allocate device name\n");
1739 if (!st
->ss
->load_container
) {
1741 pr_err("%s is not a container\n", dev
);
1745 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1747 pr_err("Failed to load metadata for %s\n",
1752 info
= st
->ss
->container_content(st
, subarray
);
1755 pr_err("Failed to find subarray-%s in %s\n",
1765 st
->ss
->free_super(st
);
1778 int add_disk(int mdfd
, struct supertype
*st
,
1779 struct mdinfo
*sra
, struct mdinfo
*info
)
1781 /* Add a device to an array, in one of 2 ways. */
1784 if (st
->ss
->external
) {
1785 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1786 info
->recovery_start
= MaxSector
;
1788 info
->recovery_start
= 0;
1789 rv
= sysfs_add_disk(sra
, info
, 0);
1792 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1796 sd2
= xmalloc(sizeof(*sd2
));
1798 sd2
->next
= sra
->devs
;
1803 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1807 int remove_disk(int mdfd
, struct supertype
*st
,
1808 struct mdinfo
*sra
, struct mdinfo
*info
)
1812 /* Remove the disk given by 'info' from the array */
1813 if (st
->ss
->external
)
1814 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1816 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1821 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1823 int cnt
= force
? 500 : 5;
1826 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1827 * outstanding IO requests to the device.
1828 * In this case, it can be helpful to wait a little while,
1829 * up to 5 seconds if 'force' is set, or 50 msec if not.
1831 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1839 int sys_hot_remove_disk(int statefd
, int force
)
1841 int cnt
= force
? 500 : 5;
1844 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1848 return ret
== 6 ? 0 : -1;
1851 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1853 /* Initialise kernel's knowledge of array.
1854 * This varies between externally managed arrays
1857 mdu_array_info_t inf
;
1860 if (st
->ss
->external
)
1861 return sysfs_set_array(info
, 9003);
1863 memset(&inf
, 0, sizeof(inf
));
1864 inf
.major_version
= info
->array
.major_version
;
1865 inf
.minor_version
= info
->array
.minor_version
;
1866 rv
= md_set_array_info(mdfd
, &inf
);
1871 unsigned long long min_recovery_start(struct mdinfo
*array
)
1873 /* find the minimum recovery_start in an array for metadata
1874 * formats that only record per-array recovery progress instead
1877 unsigned long long recovery_start
= MaxSector
;
1880 for (d
= array
->devs
; d
; d
= d
->next
)
1881 recovery_start
= min(recovery_start
, d
->recovery_start
);
1883 return recovery_start
;
1886 int mdmon_pid(char *devnm
)
1893 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1895 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1899 n
= read(fd
, pid
, 9);
1906 int mdmon_running(char *devnm
)
1908 int pid
= mdmon_pid(devnm
);
1911 if (kill(pid
, 0) == 0)
1916 int start_mdmon(char *devnm
)
1930 if (check_env("MDADM_NO_MDMON"))
1932 if (continue_via_systemd(devnm
, MDMON_SERVICE
))
1935 /* That failed, try running mdmon directly */
1936 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1940 sl
= strrchr(pathbuf
, '/');
1945 strcpy(sl
, "mdmon");
1952 for (i
= 0; paths
[i
]; i
++)
1954 execl(paths
[i
], paths
[i
],
1958 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1960 default: /* parent - good */
1961 pid
= wait(&status
);
1962 if (pid
< 0 || status
!= 0) {
1963 pr_err("failed to launch mdmon. Array remains readonly\n");
1970 __u32
random32(void)
1973 int rfd
= open("/dev/urandom", O_RDONLY
);
1974 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1981 void random_uuid(__u8
*buf
)
1986 fd
= open("/dev/urandom", O_RDONLY
);
1989 len
= read(fd
, buf
, 16);
1997 for (i
= 0; i
< 4; i
++)
2002 int flush_metadata_updates(struct supertype
*st
)
2006 st
->update_tail
= NULL
;
2010 sfd
= connect_monitor(st
->container_devnm
);
2014 while (st
->updates
) {
2015 struct metadata_update
*mu
= st
->updates
;
2016 st
->updates
= mu
->next
;
2018 send_message(sfd
, mu
, 0);
2026 st
->update_tail
= NULL
;
2030 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2033 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2038 mu
->space_list
= NULL
;
2040 *st
->update_tail
= mu
;
2041 st
->update_tail
= &mu
->next
;
2045 /* tinyc doesn't optimize this check in ioctl.h out ... */
2046 unsigned int __invalid_size_argument_for_IOC
= 0;
2049 /* Pick all spares matching given criteria from a container
2050 * if min_size == 0 do not check size
2051 * if domlist == NULL do not check domains
2052 * if spare_group given add it to domains of each spare
2053 * metadata allows to test domains using metadata of destination array */
2054 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2055 struct spare_criteria
*criteria
,
2056 struct domainlist
*domlist
,
2058 const char *metadata
, int get_one
)
2060 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2062 /* get list of all disks in container */
2063 if (st
->ss
->getinfo_super_disks
)
2064 disks
= st
->ss
->getinfo_super_disks(st
);
2068 /* find spare devices on the list */
2070 disks
->array
.spare_disks
= 0;
2074 if (d
->disk
.state
== 0) {
2075 /* check if size is acceptable */
2076 unsigned long long dev_size
;
2077 unsigned int dev_sector_size
;
2079 int sector_size_valid
= 0;
2081 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2083 if (!criteria
->min_size
||
2084 (dev_size_from_id(dev
, &dev_size
) &&
2085 dev_size
>= criteria
->min_size
))
2088 if (!criteria
->sector_size
||
2089 (dev_sector_size_from_id(dev
, &dev_sector_size
) &&
2090 criteria
->sector_size
== dev_sector_size
))
2091 sector_size_valid
= 1;
2093 found
= size_valid
&& sector_size_valid
;
2095 /* check if domain matches */
2096 if (found
&& domlist
) {
2097 struct dev_policy
*pol
= devid_policy(dev
);
2099 pol_add(&pol
, pol_domain
,
2101 if (domain_test(domlist
, pol
, metadata
) != 1)
2103 dev_policy_free(pol
);
2108 disks
->array
.spare_disks
++;
2122 /* Checks if paths point to the same device
2123 * Returns 0 if they do.
2124 * Returns 1 if they don't.
2125 * Returns -1 if something went wrong,
2126 * e.g. paths are empty or the files
2127 * they point to don't exist */
2128 int compare_paths (char* path1
, char* path2
)
2130 struct stat st1
,st2
;
2132 if (path1
== NULL
|| path2
== NULL
)
2134 if (stat(path1
,&st1
) != 0)
2136 if (stat(path2
,&st2
) != 0)
2138 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2143 /* Make sure we can open as many devices as needed */
2144 void enable_fds(int devices
)
2146 unsigned int fds
= 20 + devices
;
2148 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0 || lim
.rlim_cur
>= fds
)
2150 if (lim
.rlim_max
< fds
)
2153 setrlimit(RLIMIT_NOFILE
, &lim
);
2156 /* Close all opened descriptors if needed and redirect
2157 * streams to /dev/null.
2158 * For debug purposed, leave STDOUT and STDERR untouched
2160 * 1- if any error occurred
2163 void manage_fork_fds(int close_all
)
2166 struct dirent
*dirent
;
2169 open("/dev/null", O_RDWR
);
2179 dir
= opendir("/proc/self/fd");
2181 pr_err("Cannot open /proc/self/fd directory.\n");
2184 for (dirent
= readdir(dir
); dirent
; dirent
= readdir(dir
)) {
2187 if ((strcmp(dirent
->d_name
, ".") == 0) ||
2188 (strcmp(dirent
->d_name
, "..")) == 0)
2191 fd
= strtol(dirent
->d_name
, NULL
, 10);
2197 /* In a systemd/udev world, it is best to get systemd to
2198 * run daemon rather than running in the background.
2200 * 1- if systemd service has been started
2203 int continue_via_systemd(char *devnm
, char *service_name
)
2208 /* Simply return that service cannot be started */
2209 if (check_env("MDADM_NO_SYSTEMCTL"))
2214 snprintf(pathbuf
, sizeof(pathbuf
),
2215 "%s@%s.service", service_name
, devnm
);
2216 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
2218 status
= execl("/bin/systemctl", "systemctl", "restart",
2221 case -1: /* Just do it ourselves. */
2223 default: /* parent - good */
2224 pid
= wait(&status
);
2225 if (pid
>= 0 && status
== 0)
2233 /* This is based on similar function in systemd. */
2235 /* statfs.f_type is signed long on s390x and MIPS, causing all
2236 sorts of sign extension problems with RAMFS_MAGIC being
2237 defined as 0x858458f6 */
2238 return statfs("/", &s
) >= 0 &&
2239 ((unsigned long)s
.f_type
== TMPFS_MAGIC
||
2240 ((unsigned long)s
.f_type
& 0xFFFFFFFFUL
) ==
2241 ((unsigned long)RAMFS_MAGIC
& 0xFFFFFFFFUL
));
2244 void reopen_mddev(int mdfd
)
2246 /* Re-open without any O_EXCL, but keep
2251 devnm
= fd2devnm(mdfd
);
2253 fd
= open_dev(devnm
);
2254 if (fd
>= 0 && fd
!= mdfd
)
2258 static struct cmap_hooks
*cmap_hooks
= NULL
;
2259 static int is_cmap_hooks_ready
= 0;
2261 void set_cmap_hooks(void)
2263 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2264 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2265 if (!cmap_hooks
->cmap_handle
)
2268 cmap_hooks
->initialize
=
2269 dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2270 cmap_hooks
->get_string
=
2271 dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2272 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2274 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2275 !cmap_hooks
->finalize
)
2276 dlclose(cmap_hooks
->cmap_handle
);
2278 is_cmap_hooks_ready
= 1;
2281 int get_cluster_name(char **cluster_name
)
2284 cmap_handle_t handle
;
2286 if (!is_cmap_hooks_ready
)
2289 rv
= cmap_hooks
->initialize(&handle
);
2293 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2295 free(*cluster_name
);
2302 cmap_hooks
->finalize(handle
);
2307 void set_dlm_hooks(void)
2309 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2310 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2311 if (!dlm_hooks
->dlm_handle
)
2314 dlm_hooks
->open_lockspace
=
2315 dlsym(dlm_hooks
->dlm_handle
, "dlm_open_lockspace");
2316 dlm_hooks
->create_lockspace
=
2317 dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2318 dlm_hooks
->release_lockspace
=
2319 dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2320 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2321 dlm_hooks
->ls_unlock_wait
=
2322 dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock_wait");
2323 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2324 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2326 if (!dlm_hooks
->open_lockspace
|| !dlm_hooks
->create_lockspace
||
2327 !dlm_hooks
->ls_lock
|| !dlm_hooks
->ls_unlock_wait
||
2328 !dlm_hooks
->release_lockspace
|| !dlm_hooks
->ls_get_fd
||
2329 !dlm_hooks
->dispatch
)
2330 dlclose(dlm_hooks
->dlm_handle
);
2332 is_dlm_hooks_ready
= 1;
2335 void set_hooks(void)
2341 int zero_disk_range(int fd
, unsigned long long sector
, size_t count
)
2347 size_t len
= count
* 512;
2350 fd_zero
= open("/dev/zero", O_RDONLY
);
2352 pr_err("Cannot open /dev/zero\n");
2356 if (lseek64(fd
, sector
* 512, SEEK_SET
) < 0) {
2358 pr_err("Failed to seek offset for zeroing\n");
2362 addr
= mmap(NULL
, len
, PROT_READ
, MAP_PRIVATE
, fd_zero
, 0);
2364 if (addr
== MAP_FAILED
) {
2366 pr_err("Mapping /dev/zero failed\n");
2371 n
= write(fd
, addr
+ written
, len
- written
);
2376 pr_err("Zeroing disk range failed\n");
2380 } while (written
!= len
);