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>
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(void)
134 int flags
= LKF_NOQUEUE
;
137 if (!dlm_funs_ready()) {
138 pr_err("Something wrong with dlm library\n");
142 ret
= get_cluster_name(&cluster_name
);
144 pr_err("The md can't get cluster name\n");
148 dlm_lock_res
= xmalloc(sizeof(struct dlm_lock_resource
));
149 dlm_lock_res
->ls
= dlm_hooks
->open_lockspace(cluster_name
);
150 if (!dlm_lock_res
->ls
) {
151 dlm_lock_res
->ls
= dlm_hooks
->create_lockspace(cluster_name
, O_RDWR
);
152 if (!dlm_lock_res
->ls
) {
153 pr_err("%s failed to create lockspace\n", cluster_name
);
157 pr_err("open existed %s lockspace\n", cluster_name
);
160 snprintf(str
, 64, "bitmap%s", cluster_name
);
162 ret
= dlm_hooks
->ls_lock(dlm_lock_res
->ls
, LKM_PWMODE
,
163 &dlm_lock_res
->lksb
, flags
, str
, strlen(str
),
164 0, dlm_ast
, dlm_lock_res
, NULL
, NULL
);
166 pr_err("error %d when get PW mode on lock %s\n", errno
, str
);
167 /* let's try several times if EAGAIN happened */
168 if (dlm_lock_res
->lksb
.sb_status
== EAGAIN
&& retry_count
< 10) {
173 dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
177 /* Wait for it to complete */
178 poll_for_ast(dlm_lock_res
->ls
);
180 if (dlm_lock_res
->lksb
.sb_status
) {
181 pr_err("failed to lock cluster\n");
187 int cluster_release_dlmlock(void)
194 if (!dlm_lock_res
->lksb
.sb_lkid
)
197 ret
= dlm_hooks
->ls_unlock_wait(dlm_lock_res
->ls
,
198 dlm_lock_res
->lksb
.sb_lkid
, 0,
199 &dlm_lock_res
->lksb
);
201 pr_err("error %d happened when unlock\n", errno
);
202 /* XXX make sure the lock is unlocked eventually */
206 /* Wait for it to complete */
207 poll_for_ast(dlm_lock_res
->ls
);
209 errno
= dlm_lock_res
->lksb
.sb_status
;
210 if (errno
!= EUNLOCK
) {
211 pr_err("error %d happened in ast when unlock lockspace\n",
213 /* XXX make sure the lockspace is unlocked eventually */
217 ret
= dlm_hooks
->release_lockspace(cluster_name
, dlm_lock_res
->ls
, 1);
219 pr_err("error %d happened when release lockspace\n", errno
);
220 /* XXX make sure the lockspace is released eventually */
229 int md_array_valid(int fd
)
234 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
236 if (sra
->array_state
!= ARRAY_UNKNOWN_STATE
)
244 * GET_ARRAY_INFO doesn't provide access to the proper state
245 * information, so fallback to a basic check for raid_disks != 0
247 ret
= ioctl(fd
, RAID_VERSION
);
253 int md_array_active(int fd
)
256 struct mdu_array_info_s array
;
259 sra
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
261 if (!md_array_is_active(sra
))
267 * GET_ARRAY_INFO doesn't provide access to the proper state
268 * information, so fallback to a basic check for raid_disks != 0
270 ret
= md_get_array_info(fd
, &array
);
276 int md_array_is_active(struct mdinfo
*info
)
278 return (info
->array_state
!= ARRAY_CLEAR
&&
279 info
->array_state
!= ARRAY_INACTIVE
&&
280 info
->array_state
!= ARRAY_UNKNOWN_STATE
);
284 * Get array info from the kernel. Longer term we want to deprecate the
285 * ioctl and get it from sysfs.
287 int md_get_array_info(int fd
, struct mdu_array_info_s
*array
)
289 return ioctl(fd
, GET_ARRAY_INFO
, array
);
295 int md_set_array_info(int fd
, struct mdu_array_info_s
*array
)
297 return ioctl(fd
, SET_ARRAY_INFO
, array
);
301 * Get disk info from the kernel.
303 int md_get_disk_info(int fd
, struct mdu_disk_info_s
*disk
)
305 return ioctl(fd
, GET_DISK_INFO
, disk
);
308 int get_linux_version()
312 int a
= 0, b
= 0,c
= 0;
317 a
= strtoul(cp
, &cp
, 10);
319 b
= strtoul(cp
+1, &cp
, 10);
321 c
= strtoul(cp
+1, &cp
, 10);
323 return (a
*1000000)+(b
*1000)+c
;
326 int mdadm_version(char *version
)
334 cp
= strchr(version
, '-');
335 if (!cp
|| *(cp
+1) != ' ' || *(cp
+2) != 'v')
338 a
= strtoul(cp
, &cp
, 10);
341 b
= strtoul(cp
+1, &cp
, 10);
343 c
= strtoul(cp
+1, &cp
, 10);
346 if (*cp
!= ' ' && *cp
!= '-')
348 return (a
*1000000)+(b
*1000)+c
;
351 unsigned long long parse_size(char *size
)
353 /* parse 'size' which should be a number optionally
354 * followed by 'K', 'M'. 'G' or 'T'.
355 * Without a suffix, K is assumed.
356 * Number returned is in sectors (half-K)
357 * INVALID_SECTORS returned on error.
360 long long s
= strtoll(size
, &c
, 10);
374 s
*= 1024 * 1024 * 2;
378 s
*= 1024 * 1024 * 1024 * 2LL;
380 case 's': /* sectors */
391 int is_near_layout_10(int layout
)
395 fc
= (layout
>> 8) & 255;
396 fo
= layout
& (1 << 16);
397 if (fc
> 1 || fo
> 0)
402 int parse_layout_10(char *layout
)
406 /* Parse the layout string for raid10 */
407 /* 'f', 'o' or 'n' followed by a number <= raid_disks */
408 if ((layout
[0] != 'n' && layout
[0] != 'f' && layout
[0] != 'o') ||
409 (copies
= strtoul(layout
+1, &cp
, 10)) < 1 ||
413 if (layout
[0] == 'n')
415 else if (layout
[0] == 'o')
416 rv
= 0x10000 + (copies
<<8) + 1;
418 rv
= 1 + (copies
<<8);
422 int parse_layout_faulty(char *layout
)
426 /* Parse the layout string for 'faulty' */
427 int ln
= strcspn(layout
, "0123456789");
428 char *m
= xstrdup(layout
);
431 mode
= map_name(faultylayout
, m
);
435 return mode
| (atoi(layout
+ln
)<< ModeShift
);
438 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
441 *slot
= strtoul(input
, &dev
, 10);
442 if (dev
== input
|| dev
[0] != ':')
448 void remove_partitions(int fd
)
450 /* remove partitions from this block devices.
451 * This is used for components added to an array
453 #ifdef BLKPG_DEL_PARTITION
454 struct blkpg_ioctl_arg a
;
455 struct blkpg_partition p
;
457 a
.op
= BLKPG_DEL_PARTITION
;
459 a
.datalen
= sizeof(p
);
461 memset(a
.data
, 0, a
.datalen
);
462 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
463 ioctl(fd
, BLKPG
, &a
);
467 int test_partition(int fd
)
469 /* Check if fd is a whole-disk or a partition.
470 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
471 * will return ENXIO on an invalid partition number.
473 struct blkpg_ioctl_arg a
;
474 struct blkpg_partition p
;
475 a
.op
= BLKPG_DEL_PARTITION
;
477 a
.datalen
= sizeof(p
);
479 memset(a
.data
, 0, a
.datalen
);
481 if (ioctl(fd
, BLKPG
, &a
) == 0)
482 /* Very unlikely, but not a partition */
484 if (errno
== ENXIO
|| errno
== ENOTTY
)
485 /* not a partition */
491 int test_partition_from_id(dev_t id
)
496 sprintf(buf
, "%d:%d", major(id
), minor(id
));
497 fd
= dev_open(buf
, O_RDONLY
);
500 rv
= test_partition(fd
);
505 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
511 for (i
= 0; i
< raid_disks
; i
++)
512 avail_disks
+= !!avail
[i
];
516 /* This is the tricky one - we need to check
517 * which actual disks are present.
519 copies
= (layout
&255)* ((layout
>>8) & 255);
522 /* there must be one of the 'copies' form 'first' */
529 this = (this+1) % raid_disks
;
533 first
= (first
+(layout
&255)) % raid_disks
;
534 } while (first
!= 0);
537 case LEVEL_MULTIPATH
:
538 return avail_disks
>= 1;
541 return avail_disks
== raid_disks
;
543 return avail_disks
>= 1;
545 if (avail_disks
== raid_disks
- 1 &&
546 !avail
[raid_disks
- 1])
547 /* If just the parity device is missing, then we
548 * have enough, even if not clean
554 return avail_disks
>= raid_disks
-1;
556 return avail_disks
>= raid_disks
;
559 return avail_disks
>= raid_disks
-2;
561 return avail_disks
>= raid_disks
;
567 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
574 copy_uuid(uuid
, id
, swap
);
575 for (i
= 0; i
< 4; i
++) {
578 for (j
= 3; j
>= 0; j
--) {
579 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
587 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
,
590 // dirty hack to work around an issue with super1 superblocks...
591 // super1 superblocks need swapuuid set in order for assembly to
592 // work, but can't have it set if we want this printout to match
593 // all the other uuid printouts in super1.c, so we force swapuuid
594 // to 1 to make our printout match the rest of super1
595 #if __BYTE_ORDER == BIG_ENDIAN
596 return __fname_from_uuid(info
->uuid
, 1, buf
, sep
);
598 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 :
599 st
->ss
->swapuuid
, buf
, sep
);
603 int check_ext2(int fd
, char *name
)
606 * Check for an ext2fs file system.
607 * Superblock is always 1K at 1K offset
609 * s_magic is le16 at 56 == 0xEF53
610 * report mtime - le32 at 44
612 * logblksize - le32 at 24
614 unsigned char sb
[1024];
616 unsigned long long size
;
618 if (lseek(fd
, 1024,0)!= 1024)
620 if (read(fd
, sb
, 1024)!= 1024)
622 if (sb
[56] != 0x53 || sb
[57] != 0xef)
625 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
626 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
627 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
629 pr_err("%s appears to contain an ext2fs file system\n",
631 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
635 int check_reiser(int fd
, char *name
)
638 * superblock is at 64K
640 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
643 unsigned char sb
[1024];
644 unsigned long long size
;
645 if (lseek(fd
, 64*1024, 0) != 64*1024)
647 if (read(fd
, sb
, 1024) != 1024)
649 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
650 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
652 pr_err("%s appears to contain a reiserfs file system\n",name
);
653 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
654 cont_err("size = %lluK\n", size
*4);
659 int check_raid(int fd
, char *name
)
664 struct supertype
*st
= guess_super(fd
);
668 if (st
->ss
->add_to_super
!= NULL
) {
669 st
->ss
->load_super(st
, fd
, name
);
670 /* Looks like a raid array .. */
671 pr_err("%s appears to be part of a raid array:\n", name
);
672 st
->ss
->getinfo_super(st
, &info
, NULL
);
673 st
->ss
->free_super(st
);
674 crtime
= info
.array
.ctime
;
675 level
= map_num(pers
, info
.array
.level
);
678 cont_err("level=%s devices=%d ctime=%s",
679 level
, info
.array
.raid_disks
, ctime(&crtime
));
681 /* Looks like GPT or MBR */
682 pr_err("partition table exists on %s\n", name
);
687 int fstat_is_blkdev(int fd
, char *devname
, dev_t
*rdev
)
691 if (fstat(fd
, &stb
) != 0) {
692 pr_err("fstat failed for %s: %s\n", devname
, strerror(errno
));
695 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
696 pr_err("%s is not a block device.\n", devname
);
704 int stat_is_blkdev(char *devname
, dev_t
*rdev
)
708 if (stat(devname
, &stb
) != 0) {
709 pr_err("stat failed for %s: %s\n", devname
, strerror(errno
));
712 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
713 pr_err("%s is not a block device.\n", devname
);
725 for (i
= 0; i
< 5; i
++) {
727 fprintf(stderr
, "%s%s", mesg
, add
);
729 if (fgets(buf
, 100, stdin
)==NULL
)
731 if (buf
[0]=='y' || buf
[0]=='Y')
733 if (buf
[0]=='n' || buf
[0]=='N')
737 pr_err("assuming 'no'\n");
741 int is_standard(char *dev
, int *nump
)
743 /* tests if dev is a "standard" md dev name.
744 * i.e if the last component is "/dNN" or "/mdNN",
745 * where NN is a string of digits
746 * Returns 1 if a partitionable standard,
747 * -1 if non-partitonable,
748 * 0 if not a standard name.
750 char *d
= strrchr(dev
, '/');
755 if (strncmp(d
, "/d",2) == 0)
756 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
757 else if (strncmp(d
, "/md_d", 5) == 0)
758 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
759 else if (strncmp(d
, "/md", 3) == 0)
760 d
+= 3, type
= -1; /* /dev/mdN */
761 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
762 d
+= 1, type
= -1; /* /dev/md/N */
772 if (nump
) *nump
= num
;
777 unsigned long calc_csum(void *super
, int bytes
)
779 unsigned long long newcsum
= 0;
782 unsigned int *superc
= (unsigned int*) super
;
784 for(i
= 0; i
< bytes
/4; i
++)
785 newcsum
+= superc
[i
];
786 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
788 /* The in-kernel checksum calculation is always 16bit on
789 * the alpha, though it is 32 bit on i386...
790 * I wonder what it is elsewhere... (it uses an API in
791 * a way that it shouldn't).
793 csum
= (csum
& 0xffff) + (csum
>> 16);
794 csum
= (csum
& 0xffff) + (csum
>> 16);
799 char *human_size(long long bytes
)
803 /* We convert bytes to either centi-M{ega,ibi}bytes,
804 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
805 * with appropriate rounding, and then print
806 * 1/100th of those as a decimal.
807 * We allow upto 2048Megabytes before converting to
808 * gigabytes and 2048Gigabytes before converting to
809 * terabytes, as that shows more precision and isn't
810 * too large a number.
813 if (bytes
< 5000*1024)
815 else if (bytes
< 2*1024LL*1024LL*1024LL) {
816 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
817 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
818 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
819 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
820 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
821 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
822 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
823 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
824 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
826 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
827 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
828 snprintf(buf
, sizeof(buf
), " (%ld.%02ld TiB %ld.%02ld TB)",
829 cTiB
/100, cTiB
% 100, cTB
/100, cTB
% 100);
834 char *human_size_brief(long long bytes
, int prefix
)
838 /* We convert bytes to either centi-M{ega,ibi}bytes,
839 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
840 * with appropriate rounding, and then print
841 * 1/100th of those as a decimal.
842 * We allow upto 2048Megabytes before converting to
843 * gigabytes and 2048Gigabytes before converting to
844 * terabytes, as that shows more precision and isn't
845 * too large a number.
847 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
848 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
851 if (bytes
< 5000*1024)
853 else if (prefix
== IEC
) {
854 if (bytes
< 2*1024LL*1024LL*1024LL) {
855 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
856 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
857 cMiB
/100, cMiB
% 100);
858 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
859 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
860 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
861 cGiB
/100, cGiB
% 100);
863 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
864 snprintf(buf
, sizeof(buf
), "%ld.%02ldTiB",
865 cTiB
/100, cTiB
% 100);
868 else if (prefix
== JEDEC
) {
869 if (bytes
< 2*1024LL*1024LL*1024LL) {
870 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
871 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
873 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
874 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
875 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
878 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
879 snprintf(buf
, sizeof(buf
), "%ld.%02ldTB",
889 void print_r10_layout(int layout
)
891 int near
= layout
& 255;
892 int far
= (layout
>> 8) & 255;
893 int offset
= (layout
&0x10000);
897 printf("%s near=%d", sep
, near
);
901 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
903 printf("NO REDUNDANCY");
906 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
907 int chunksize
, unsigned long long devsize
)
911 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
912 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
915 int get_data_disks(int level
, int layout
, int raid_disks
)
919 case 0: data_disks
= raid_disks
;
921 case 1: data_disks
= 1;
924 case 5: data_disks
= raid_disks
- 1;
926 case 6: data_disks
= raid_disks
- 2;
928 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
935 dev_t
devnm2devid(char *devnm
)
937 /* First look in /sys/block/$DEVNM/dev for %d:%d
938 * If that fails, try parsing out a number
945 snprintf(path
, sizeof(path
), "/sys/block/%s/dev", devnm
);
946 fd
= open(path
, O_RDONLY
);
949 int n
= read(fd
, buf
, sizeof(buf
));
953 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
954 return makedev(mjr
, mnr
);
956 if (strncmp(devnm
, "md_d", 4) == 0 &&
958 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
959 ep
> devnm
&& *ep
== 0)
960 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
962 if (strncmp(devnm
, "md", 2) == 0 &&
964 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
965 ep
> devnm
&& *ep
== 0)
966 return makedev(MD_MAJOR
, mnr
);
971 char *get_md_name(char *devnm
)
973 /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
974 /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
976 static char devname
[50];
978 dev_t rdev
= devnm2devid(devnm
);
983 if (strncmp(devnm
, "md_", 3) == 0) {
984 snprintf(devname
, sizeof(devname
), "/dev/md/%s",
986 if (stat(devname
, &stb
) == 0 &&
987 (S_IFMT
&stb
.st_mode
) == S_IFBLK
&& (stb
.st_rdev
== rdev
))
990 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
991 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
992 (stb
.st_rdev
== rdev
))
995 snprintf(devname
, sizeof(devname
), "/dev/md/%s", devnm
+2);
996 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
997 (stb
.st_rdev
== rdev
))
1000 dn
= map_dev(major(rdev
), minor(rdev
), 0);
1003 snprintf(devname
, sizeof(devname
), "/dev/.tmp.%s", devnm
);
1004 if (mknod(devname
, S_IFBLK
| 0600, rdev
) == -1)
1005 if (errno
!= EEXIST
)
1008 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
&&
1009 (stb
.st_rdev
== rdev
))
1015 void put_md_name(char *name
)
1017 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1021 int get_maj_min(char *dev
, int *major
, int *minor
)
1024 *major
= strtoul(dev
, &e
, 0);
1025 return (e
> dev
&& *e
== ':' && e
[1] &&
1026 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1030 int dev_open(char *dev
, int flags
)
1032 /* like 'open', but if 'dev' matches %d:%d, create a temp
1033 * block device and open that
1044 if (get_maj_min(dev
, &major
, &minor
)) {
1045 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1046 (int)getpid(), major
, minor
);
1047 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1048 fd
= open(devname
, flags
);
1052 /* Try /tmp as /dev appear to be read-only */
1053 snprintf(devname
, sizeof(devname
),
1054 "/tmp/.tmp.md.%d:%d:%d",
1055 (int)getpid(), major
, minor
);
1056 if (mknod(devname
, S_IFBLK
|0600,
1057 makedev(major
, minor
)) == 0) {
1058 fd
= open(devname
, flags
);
1063 fd
= open(dev
, flags
);
1067 int open_dev_flags(char *devnm
, int flags
)
1072 devid
= devnm2devid(devnm
);
1073 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1074 return dev_open(buf
, flags
);
1077 int open_dev(char *devnm
)
1079 return open_dev_flags(devnm
, O_RDONLY
);
1082 int open_dev_excl(char *devnm
)
1087 dev_t devid
= devnm2devid(devnm
);
1090 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1091 for (i
= 0; i
< 25; i
++) {
1092 int fd
= dev_open(buf
, flags
|O_EXCL
);
1095 if (errno
== EACCES
&& flags
== O_RDWR
) {
1108 int same_dev(char *one
, char *two
)
1110 struct stat st1
, st2
;
1111 if (stat(one
, &st1
) != 0)
1113 if (stat(two
, &st2
) != 0)
1115 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1117 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1119 return st1
.st_rdev
== st2
.st_rdev
;
1122 void wait_for(char *dev
, int fd
)
1125 struct stat stb_want
;
1128 if (fstat(fd
, &stb_want
) != 0 ||
1129 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1132 for (i
= 0; i
< 25; i
++) {
1134 if (stat(dev
, &stb
) == 0 &&
1135 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1136 (stb
.st_rdev
== stb_want
.st_rdev
))
1143 pr_err("timeout waiting for %s\n", dev
);
1146 struct superswitch
*superlist
[] =
1149 &super_ddf
, &super_imsm
,
1154 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1156 mdu_array_info_t array
;
1159 struct supertype
*st
= NULL
;
1164 char *subarray
= NULL
;
1165 char container
[32] = "";
1167 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1170 vers
= sra
->array
.major_version
;
1171 minor
= sra
->array
.minor_version
;
1172 verstr
= sra
->text_version
;
1174 if (md_get_array_info(fd
, &array
))
1175 array
.major_version
= array
.minor_version
= 0;
1176 vers
= array
.major_version
;
1177 minor
= array
.minor_version
;
1182 sprintf(version
, "%d.%d", vers
, minor
);
1185 if (minor
== -2 && is_subarray(verstr
)) {
1186 char *dev
= verstr
+1;
1188 subarray
= strchr(dev
, '/');
1191 subarray
= xstrdup(subarray
);
1193 strcpy(container
, dev
);
1195 sra
= sysfs_read(-1, container
, GET_VERSION
);
1196 if (sra
&& sra
->text_version
[0])
1197 verstr
= sra
->text_version
;
1199 verstr
= "-no-metadata-";
1202 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1203 st
= superlist
[i
]->match_metadata_desc(verstr
);
1209 *subarrayp
= subarray
;
1210 strcpy(st
->container_devnm
, container
);
1211 strcpy(st
->devnm
, fd2devnm(fd
));
1218 int dev_size_from_id(dev_t id
, unsigned long long *size
)
1223 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1224 fd
= dev_open(buf
, O_RDONLY
);
1227 if (get_dev_size(fd
, NULL
, size
)) {
1235 int dev_sector_size_from_id(dev_t id
, unsigned int *size
)
1240 sprintf(buf
, "%d:%d", major(id
), minor(id
));
1241 fd
= dev_open(buf
, O_RDONLY
);
1244 if (get_dev_sector_size(fd
, NULL
, size
)) {
1252 struct supertype
*dup_super(struct supertype
*orig
)
1254 struct supertype
*st
;
1258 st
= xcalloc(1, sizeof(*st
));
1260 st
->max_devs
= orig
->max_devs
;
1261 st
->minor_version
= orig
->minor_version
;
1262 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1263 st
->data_offset
= orig
->data_offset
;
1269 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1271 /* try each load_super to find the best match,
1272 * and return the best superswitch
1274 struct superswitch
*ss
;
1275 struct supertype
*st
;
1276 unsigned int besttime
= 0;
1280 st
= xcalloc(1, sizeof(*st
));
1281 st
->container_devnm
[0] = 0;
1283 for (i
= 0; superlist
[i
]; i
++) {
1286 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1288 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1290 memset(st
, 0, sizeof(*st
));
1291 st
->ignore_hw_compat
= 1;
1292 rv
= ss
->load_super(st
, fd
, NULL
);
1295 st
->ss
->getinfo_super(st
, &info
, NULL
);
1296 if (bestsuper
== -1 ||
1297 besttime
< info
.array
.ctime
) {
1299 besttime
= info
.array
.ctime
;
1304 if (bestsuper
!= -1) {
1306 memset(st
, 0, sizeof(*st
));
1307 st
->ignore_hw_compat
= 1;
1308 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1310 superlist
[bestsuper
]->free_super(st
);
1318 /* Return size of device in bytes */
1319 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1321 unsigned long long ldsize
;
1324 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1325 ldsize
= (unsigned long long)st
.st_size
;
1328 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1331 unsigned long dsize
;
1332 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1337 pr_err("Cannot get size of %s: %s\n",
1338 dname
, strerror(errno
));
1346 /* Return sector size of device in bytes */
1347 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1349 unsigned int sectsize
;
1351 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1353 pr_err("Cannot get sector size of %s: %s\n",
1354 dname
, strerror(errno
));
1358 *sectsizep
= sectsize
;
1362 /* Return true if this can only be a container, not a member device.
1363 * i.e. is and md device and size is zero
1365 int must_be_container(int fd
)
1368 unsigned long long size
;
1370 mdi
= sysfs_read(fd
, NULL
, GET_VERSION
);
1375 if (get_dev_size(fd
, NULL
, &size
) == 0)
1382 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1383 * Returns: 1 if successful
1384 * -1 for unknown partition type
1385 * 0 for other errors
1387 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1390 unsigned char empty_gpt_entry
[16]= {0};
1391 struct GPT_part_entry
*part
;
1393 unsigned long long curr_part_end
;
1394 unsigned all_partitions
, entry_size
;
1396 unsigned int sector_size
= 0;
1400 BUILD_BUG_ON(sizeof(gpt
) != 512);
1401 /* skip protective MBR */
1402 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1404 lseek(fd
, sector_size
, SEEK_SET
);
1405 /* read GPT header */
1406 if (read(fd
, &gpt
, 512) != 512)
1409 /* get the number of partition entries and the entry size */
1410 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1411 entry_size
= __le32_to_cpu(gpt
.part_size
);
1413 /* Check GPT signature*/
1414 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1418 if (all_partitions
> 1024 ||
1419 entry_size
> sizeof(buf
))
1422 part
= (struct GPT_part_entry
*)buf
;
1424 /* set offset to third block (GPT entries) */
1425 lseek(fd
, sector_size
*2, SEEK_SET
);
1426 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1427 /* read partition entry */
1428 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1431 /* is this valid partition? */
1432 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1433 /* check the last lba for the current partition */
1434 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1435 if (curr_part_end
> *endofpart
)
1436 *endofpart
= curr_part_end
;
1443 /* Sets endofpart parameter to the last block used by the last partition on the device.
1444 * Returns: 1 if successful
1445 * -1 for unknown partition type
1446 * 0 for other errors
1448 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1450 struct MBR boot_sect
;
1451 unsigned long long curr_part_end
;
1453 unsigned int sector_size
;
1458 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1461 if (read(fd
, &boot_sect
, 512) != 512)
1464 /* check MBP signature */
1465 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1467 /* found the correct signature */
1469 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1471 * Have to make every access through boot_sect rather
1472 * than using a pointer to the partition table (or an
1473 * entry), since the entries are not properly aligned.
1476 /* check for GPT type */
1477 if (boot_sect
.parts
[part_nr
].part_type
==
1478 MBR_GPT_PARTITION_TYPE
) {
1479 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1482 /* check the last used lba for the current partition */
1484 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1485 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1486 if (curr_part_end
> *endofpart
)
1487 *endofpart
= curr_part_end
;
1490 /* Unknown partition table */
1493 /* calculate number of 512-byte blocks */
1494 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1495 *endofpart
*= (sector_size
/ 512);
1500 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1501 unsigned long long size
)
1504 * Check where the last partition ends
1506 unsigned long long endofpart
;
1508 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1509 /* There appears to be a partition table here */
1510 if (freesize
== 0) {
1511 /* partitions will not be visible in new device */
1512 pr_err("partition table exists on %s but will be lost or\n"
1513 " meaningless after creating array\n",
1516 } else if (endofpart
> freesize
) {
1517 /* last partition overlaps metadata */
1518 pr_err("metadata will over-write last partition on %s.\n",
1521 } else if (size
&& endofpart
> size
) {
1522 /* partitions will be truncated in new device */
1523 pr_err("array size is too small to cover all partitions on %s.\n",
1531 int open_container(int fd
)
1533 /* 'fd' is a block device. Find out if it is in use
1534 * by a container, and return an open fd on that container.
1545 if (fstat(fd
, &st
) != 0)
1547 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1548 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1549 e
= path
+ strlen(path
);
1551 dir
= opendir(path
);
1554 while ((de
= readdir(dir
))) {
1557 if (de
->d_name
[0] == '.')
1559 /* Need to make sure it is a container and not a volume */
1560 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1561 dfd
= open(path
, O_RDONLY
);
1564 n
= read(dfd
, buf
, sizeof(buf
));
1566 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1569 if (strncmp(buf
, "external", 8) != 0 ||
1573 sprintf(e
, "/%s/dev", de
->d_name
);
1574 dfd
= open(path
, O_RDONLY
);
1577 n
= read(dfd
, buf
, sizeof(buf
));
1579 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1582 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1584 sprintf(buf
, "%d:%d", major
, minor
);
1585 dfd
= dev_open(buf
, O_RDONLY
);
1595 struct superswitch
*version_to_superswitch(char *vers
)
1599 for (i
= 0; superlist
[i
]; i
++) {
1600 struct superswitch
*ss
= superlist
[i
];
1602 if (strcmp(vers
, ss
->name
) == 0)
1609 int metadata_container_matches(char *metadata
, char *devnm
)
1611 /* Check if 'devnm' is the container named in 'metadata'
1613 * /containername/componentname or
1614 * -containername/componentname
1617 if (*metadata
!= '/' && *metadata
!= '-')
1620 if (strncmp(metadata
+1, devnm
, l
) != 0)
1622 if (metadata
[l
+1] != '/')
1627 int metadata_subdev_matches(char *metadata
, char *devnm
)
1629 /* Check if 'devnm' is the subdev named in 'metadata'
1631 * /containername/subdev or
1632 * -containername/subdev
1635 if (*metadata
!= '/' && *metadata
!= '-')
1637 sl
= strchr(metadata
+1, '/');
1640 if (strcmp(sl
+1, devnm
) == 0)
1645 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1647 if (mdstat
->metadata_version
== NULL
||
1648 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1649 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1655 int is_subarray_active(char *subarray
, char *container
)
1657 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1658 struct mdstat_ent
*ent
;
1660 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1661 if (is_container_member(ent
, container
))
1662 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1665 free_mdstat(mdstat
);
1670 /* open_subarray - opens a subarray in a container
1671 * @dev: container device name
1672 * @st: empty supertype
1673 * @quiet: block reporting errors flag
1675 * On success returns an fd to a container and fills in *st
1677 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1680 struct mdinfo
*info
;
1684 fd
= open(dev
, O_RDWR
|O_EXCL
);
1687 pr_err("Couldn't open %s, aborting\n",
1692 _devnm
= fd2devnm(fd
);
1693 if (_devnm
== NULL
) {
1695 pr_err("Failed to determine device number for %s\n",
1699 strcpy(st
->devnm
, _devnm
);
1701 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1704 pr_err("Failed to read sysfs for %s\n",
1709 if (mdi
->array
.level
!= UnSet
) {
1711 pr_err("%s is not a container\n", dev
);
1715 st
->ss
= version_to_superswitch(mdi
->text_version
);
1718 pr_err("Operation not supported for %s metadata\n",
1723 if (st
->devnm
[0] == 0) {
1725 pr_err("Failed to allocate device name\n");
1729 if (!st
->ss
->load_container
) {
1731 pr_err("%s is not a container\n", dev
);
1735 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1737 pr_err("Failed to load metadata for %s\n",
1742 info
= st
->ss
->container_content(st
, subarray
);
1745 pr_err("Failed to find subarray-%s in %s\n",
1755 st
->ss
->free_super(st
);
1768 int add_disk(int mdfd
, struct supertype
*st
,
1769 struct mdinfo
*sra
, struct mdinfo
*info
)
1771 /* Add a device to an array, in one of 2 ways. */
1774 if (st
->ss
->external
) {
1775 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1776 info
->recovery_start
= MaxSector
;
1778 info
->recovery_start
= 0;
1779 rv
= sysfs_add_disk(sra
, info
, 0);
1782 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1786 sd2
= xmalloc(sizeof(*sd2
));
1788 sd2
->next
= sra
->devs
;
1793 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1797 int remove_disk(int mdfd
, struct supertype
*st
,
1798 struct mdinfo
*sra
, struct mdinfo
*info
)
1802 /* Remove the disk given by 'info' from the array */
1803 if (st
->ss
->external
)
1804 rv
= sysfs_set_str(sra
, info
, "slot", "none");
1806 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1811 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1813 int cnt
= force
? 500 : 5;
1816 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1817 * outstanding IO requests to the device.
1818 * In this case, it can be helpful to wait a little while,
1819 * up to 5 seconds if 'force' is set, or 50 msec if not.
1821 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1829 int sys_hot_remove_disk(int statefd
, int force
)
1831 int cnt
= force
? 500 : 5;
1834 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1838 return ret
== 6 ? 0 : -1;
1841 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1843 /* Initialise kernel's knowledge of array.
1844 * This varies between externally managed arrays
1847 mdu_array_info_t inf
;
1850 if (st
->ss
->external
)
1851 return sysfs_set_array(info
, 9003);
1853 memset(&inf
, 0, sizeof(inf
));
1854 inf
.major_version
= info
->array
.major_version
;
1855 inf
.minor_version
= info
->array
.minor_version
;
1856 rv
= md_set_array_info(mdfd
, &inf
);
1861 unsigned long long min_recovery_start(struct mdinfo
*array
)
1863 /* find the minimum recovery_start in an array for metadata
1864 * formats that only record per-array recovery progress instead
1867 unsigned long long recovery_start
= MaxSector
;
1870 for (d
= array
->devs
; d
; d
= d
->next
)
1871 recovery_start
= min(recovery_start
, d
->recovery_start
);
1873 return recovery_start
;
1876 int mdmon_pid(char *devnm
)
1883 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1885 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1889 n
= read(fd
, pid
, 9);
1896 int mdmon_running(char *devnm
)
1898 int pid
= mdmon_pid(devnm
);
1901 if (kill(pid
, 0) == 0)
1906 int start_mdmon(char *devnm
)
1920 if (check_env("MDADM_NO_MDMON"))
1922 if (continue_via_systemd(devnm
, MDMON_SERVICE
))
1925 /* That failed, try running mdmon directly */
1926 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1930 sl
= strrchr(pathbuf
, '/');
1935 strcpy(sl
, "mdmon");
1942 for (i
= 0; paths
[i
]; i
++)
1944 execl(paths
[i
], paths
[i
],
1948 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1950 default: /* parent - good */
1951 pid
= wait(&status
);
1952 if (pid
< 0 || status
!= 0) {
1953 pr_err("failed to launch mdmon. Array remains readonly\n");
1960 __u32
random32(void)
1963 int rfd
= open("/dev/urandom", O_RDONLY
);
1964 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1971 void random_uuid(__u8
*buf
)
1976 fd
= open("/dev/urandom", O_RDONLY
);
1979 len
= read(fd
, buf
, 16);
1987 for (i
= 0; i
< 4; i
++)
1992 int flush_metadata_updates(struct supertype
*st
)
1996 st
->update_tail
= NULL
;
2000 sfd
= connect_monitor(st
->container_devnm
);
2004 while (st
->updates
) {
2005 struct metadata_update
*mu
= st
->updates
;
2006 st
->updates
= mu
->next
;
2008 send_message(sfd
, mu
, 0);
2016 st
->update_tail
= NULL
;
2020 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2023 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2028 mu
->space_list
= NULL
;
2030 *st
->update_tail
= mu
;
2031 st
->update_tail
= &mu
->next
;
2035 /* tinyc doesn't optimize this check in ioctl.h out ... */
2036 unsigned int __invalid_size_argument_for_IOC
= 0;
2039 /* Pick all spares matching given criteria from a container
2040 * if min_size == 0 do not check size
2041 * if domlist == NULL do not check domains
2042 * if spare_group given add it to domains of each spare
2043 * metadata allows to test domains using metadata of destination array */
2044 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2045 struct spare_criteria
*criteria
,
2046 struct domainlist
*domlist
,
2048 const char *metadata
, int get_one
)
2050 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2052 /* get list of all disks in container */
2053 if (st
->ss
->getinfo_super_disks
)
2054 disks
= st
->ss
->getinfo_super_disks(st
);
2058 /* find spare devices on the list */
2060 disks
->array
.spare_disks
= 0;
2064 if (d
->disk
.state
== 0) {
2065 /* check if size is acceptable */
2066 unsigned long long dev_size
;
2067 unsigned int dev_sector_size
;
2069 int sector_size_valid
= 0;
2071 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2073 if (!criteria
->min_size
||
2074 (dev_size_from_id(dev
, &dev_size
) &&
2075 dev_size
>= criteria
->min_size
))
2078 if (!criteria
->sector_size
||
2079 (dev_sector_size_from_id(dev
, &dev_sector_size
) &&
2080 criteria
->sector_size
== dev_sector_size
))
2081 sector_size_valid
= 1;
2083 found
= size_valid
&& sector_size_valid
;
2085 /* check if domain matches */
2086 if (found
&& domlist
) {
2087 struct dev_policy
*pol
= devid_policy(dev
);
2089 pol_add(&pol
, pol_domain
,
2091 if (domain_test(domlist
, pol
, metadata
) != 1)
2093 dev_policy_free(pol
);
2098 disks
->array
.spare_disks
++;
2112 /* Checks if paths point to the same device
2113 * Returns 0 if they do.
2114 * Returns 1 if they don't.
2115 * Returns -1 if something went wrong,
2116 * e.g. paths are empty or the files
2117 * they point to don't exist */
2118 int compare_paths (char* path1
, char* path2
)
2120 struct stat st1
,st2
;
2122 if (path1
== NULL
|| path2
== NULL
)
2124 if (stat(path1
,&st1
) != 0)
2126 if (stat(path2
,&st2
) != 0)
2128 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2133 /* Make sure we can open as many devices as needed */
2134 void enable_fds(int devices
)
2136 unsigned int fds
= 20 + devices
;
2138 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0 || lim
.rlim_cur
>= fds
)
2140 if (lim
.rlim_max
< fds
)
2143 setrlimit(RLIMIT_NOFILE
, &lim
);
2146 /* Close all opened descriptors if needed and redirect
2147 * streams to /dev/null.
2148 * For debug purposed, leave STDOUT and STDERR untouched
2150 * 1- if any error occurred
2153 void manage_fork_fds(int close_all
)
2156 struct dirent
*dirent
;
2159 open("/dev/null", O_RDWR
);
2169 dir
= opendir("/proc/self/fd");
2171 pr_err("Cannot open /proc/self/fd directory.\n");
2174 for (dirent
= readdir(dir
); dirent
; dirent
= readdir(dir
)) {
2177 if ((strcmp(dirent
->d_name
, ".") == 0) ||
2178 (strcmp(dirent
->d_name
, "..")) == 0)
2181 fd
= strtol(dirent
->d_name
, NULL
, 10);
2187 /* In a systemd/udev world, it is best to get systemd to
2188 * run daemon rather than running in the background.
2190 * 1- if systemd service has been started
2193 int continue_via_systemd(char *devnm
, char *service_name
)
2198 /* Simply return that service cannot be started */
2199 if (check_env("MDADM_NO_SYSTEMCTL"))
2204 snprintf(pathbuf
, sizeof(pathbuf
),
2205 "%s@%s.service", service_name
, devnm
);
2206 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
2208 status
= execl("/bin/systemctl", "systemctl", "restart",
2211 case -1: /* Just do it ourselves. */
2213 default: /* parent - good */
2214 pid
= wait(&status
);
2215 if (pid
>= 0 && status
== 0)
2223 /* This is based on similar function in systemd. */
2225 /* statfs.f_type is signed long on s390x and MIPS, causing all
2226 sorts of sign extension problems with RAMFS_MAGIC being
2227 defined as 0x858458f6 */
2228 return statfs("/", &s
) >= 0 &&
2229 ((unsigned long)s
.f_type
== TMPFS_MAGIC
||
2230 ((unsigned long)s
.f_type
& 0xFFFFFFFFUL
) ==
2231 ((unsigned long)RAMFS_MAGIC
& 0xFFFFFFFFUL
));
2234 void reopen_mddev(int mdfd
)
2236 /* Re-open without any O_EXCL, but keep
2241 devnm
= fd2devnm(mdfd
);
2243 fd
= open_dev(devnm
);
2244 if (fd
>= 0 && fd
!= mdfd
)
2248 static struct cmap_hooks
*cmap_hooks
= NULL
;
2249 static int is_cmap_hooks_ready
= 0;
2251 void set_cmap_hooks(void)
2253 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2254 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2255 if (!cmap_hooks
->cmap_handle
)
2258 cmap_hooks
->initialize
=
2259 dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2260 cmap_hooks
->get_string
=
2261 dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2262 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2264 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2265 !cmap_hooks
->finalize
)
2266 dlclose(cmap_hooks
->cmap_handle
);
2268 is_cmap_hooks_ready
= 1;
2271 int get_cluster_name(char **cluster_name
)
2274 cmap_handle_t handle
;
2276 if (!is_cmap_hooks_ready
)
2279 rv
= cmap_hooks
->initialize(&handle
);
2283 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2285 free(*cluster_name
);
2292 cmap_hooks
->finalize(handle
);
2297 void set_dlm_hooks(void)
2299 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2300 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2301 if (!dlm_hooks
->dlm_handle
)
2304 dlm_hooks
->open_lockspace
=
2305 dlsym(dlm_hooks
->dlm_handle
, "dlm_open_lockspace");
2306 dlm_hooks
->create_lockspace
=
2307 dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2308 dlm_hooks
->release_lockspace
=
2309 dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2310 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2311 dlm_hooks
->ls_unlock_wait
=
2312 dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock_wait");
2313 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2314 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2316 if (!dlm_hooks
->open_lockspace
|| !dlm_hooks
->create_lockspace
||
2317 !dlm_hooks
->ls_lock
|| !dlm_hooks
->ls_unlock_wait
||
2318 !dlm_hooks
->release_lockspace
|| !dlm_hooks
->ls_get_fd
||
2319 !dlm_hooks
->dispatch
)
2320 dlclose(dlm_hooks
->dlm_handle
);
2322 is_dlm_hooks_ready
= 1;
2325 void set_hooks(void)
2331 int zero_disk_range(int fd
, unsigned long long sector
, size_t count
)
2337 size_t len
= count
* 512;
2340 fd_zero
= open("/dev/zero", O_RDONLY
);
2342 pr_err("Cannot open /dev/zero\n");
2346 if (lseek64(fd
, sector
* 512, SEEK_SET
) < 0) {
2348 pr_err("Failed to seek offset for zeroing\n");
2352 addr
= mmap(NULL
, len
, PROT_READ
, MAP_PRIVATE
, fd_zero
, 0);
2354 if (addr
== MAP_FAILED
) {
2356 pr_err("Mapping /dev/zero failed\n");
2361 n
= write(fd
, addr
+ written
, len
- written
);
2366 pr_err("Zeroing disk range failed\n");
2370 } while (written
!= len
);