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) {
169 sleep_for(10, 0, true);
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
)
430 /* Parse the layout string for 'faulty' */
431 ln
= strcspn(layout
, "0123456789");
434 mode
= map_name(faultylayout
, m
);
440 return mode
| (atoi(layout
+ln
)<< ModeShift
);
443 int parse_cluster_confirm_arg(char *input
, char **devname
, int *slot
)
446 *slot
= strtoul(input
, &dev
, 10);
447 if (dev
== input
|| dev
[0] != ':')
453 void remove_partitions(int fd
)
455 /* remove partitions from this block devices.
456 * This is used for components added to an array
458 #ifdef BLKPG_DEL_PARTITION
459 struct blkpg_ioctl_arg a
;
460 struct blkpg_partition p
;
462 a
.op
= BLKPG_DEL_PARTITION
;
464 a
.datalen
= sizeof(p
);
466 memset(a
.data
, 0, a
.datalen
);
467 for (p
.pno
= 0; p
.pno
< 16; p
.pno
++)
468 ioctl(fd
, BLKPG
, &a
);
472 int test_partition(int fd
)
474 /* Check if fd is a whole-disk or a partition.
475 * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
476 * will return ENXIO on an invalid partition number.
478 struct blkpg_ioctl_arg a
;
479 struct blkpg_partition p
;
480 a
.op
= BLKPG_DEL_PARTITION
;
482 a
.datalen
= sizeof(p
);
484 memset(a
.data
, 0, a
.datalen
);
486 if (ioctl(fd
, BLKPG
, &a
) == 0)
487 /* Very unlikely, but not a partition */
489 if (errno
== ENXIO
|| errno
== ENOTTY
)
490 /* not a partition */
496 int test_partition_from_id(dev_t id
)
501 sprintf(buf
, "%d:%d", major(id
), minor(id
));
502 fd
= dev_open(buf
, O_RDONLY
);
505 rv
= test_partition(fd
);
510 int enough(int level
, int raid_disks
, int layout
, int clean
, char *avail
)
516 for (i
= 0; i
< raid_disks
; i
++)
517 avail_disks
+= !!avail
[i
];
521 /* This is the tricky one - we need to check
522 * which actual disks are present.
524 copies
= (layout
&255)* ((layout
>>8) & 255);
527 /* there must be one of the 'copies' form 'first' */
534 this = (this+1) % raid_disks
;
538 first
= (first
+(layout
&255)) % raid_disks
;
539 } while (first
!= 0);
542 case LEVEL_MULTIPATH
:
543 return avail_disks
>= 1;
546 return avail_disks
== raid_disks
;
548 return avail_disks
>= 1;
550 if (avail_disks
== raid_disks
- 1 &&
551 !avail
[raid_disks
- 1])
552 /* If just the parity device is missing, then we
553 * have enough, even if not clean
559 return avail_disks
>= raid_disks
-1;
561 return avail_disks
>= raid_disks
;
564 return avail_disks
>= raid_disks
-2;
566 return avail_disks
>= raid_disks
;
572 char *__fname_from_uuid(int id
[4], int swap
, char *buf
, char sep
)
579 copy_uuid(uuid
, id
, swap
);
580 for (i
= 0; i
< 4; i
++) {
583 for (j
= 3; j
>= 0; j
--) {
584 sprintf(c
,"%02x", (unsigned char) uuid
[j
+4*i
]);
593 * fname_from_uuid() - generate uuid string. Should not be used with super1.
594 * @info: info with uuid
597 * This routine should not be used with super1. See detail_fname_from_uuid() for details. It does
598 * not use superswitch swapuuid as it should be 0 but it has to do UUID conversion if host is big
599 * endian- left for backward compatibility.
601 char *fname_from_uuid(struct mdinfo
*info
, char *buf
)
603 #if __BYTE_ORDER == BIG_ENDIAN
604 return __fname_from_uuid(info
->uuid
, true, buf
, ':');
606 return __fname_from_uuid(info
->uuid
, false, buf
, ':');
610 int check_ext2(int fd
, char *name
)
613 * Check for an ext2fs file system.
614 * Superblock is always 1K at 1K offset
616 * s_magic is le16 at 56 == 0xEF53
617 * report mtime - le32 at 44
619 * logblksize - le32 at 24
621 unsigned char sb
[1024];
623 unsigned long long size
;
625 if (lseek(fd
, 1024,0)!= 1024)
627 if (read(fd
, sb
, 1024)!= 1024)
629 if (sb
[56] != 0x53 || sb
[57] != 0xef)
632 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
633 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
634 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
636 pr_err("%s appears to contain an ext2fs file system\n",
638 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
642 int check_reiser(int fd
, char *name
)
645 * superblock is at 64K
647 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
650 unsigned char sb
[1024];
651 unsigned long long size
;
652 if (lseek(fd
, 64*1024, 0) != 64*1024)
654 if (read(fd
, sb
, 1024) != 1024)
656 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
657 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
659 pr_err("%s appears to contain a reiserfs file system\n",name
);
660 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
661 cont_err("size = %lluK\n", size
*4);
666 int check_raid(int fd
, char *name
)
671 struct supertype
*st
= guess_super(fd
);
675 if (st
->ss
->add_to_super
!= NULL
) {
676 st
->ss
->load_super(st
, fd
, name
);
677 /* Looks like a raid array .. */
678 pr_err("%s appears to be part of a raid array:\n", name
);
679 st
->ss
->getinfo_super(st
, &info
, NULL
);
680 st
->ss
->free_super(st
);
681 crtime
= info
.array
.ctime
;
682 level
= map_num(pers
, info
.array
.level
);
685 cont_err("level=%s devices=%d ctime=%s",
686 level
, info
.array
.raid_disks
, ctime(&crtime
));
688 /* Looks like GPT or MBR */
689 pr_err("partition table exists on %s\n", name
);
694 int fstat_is_blkdev(int fd
, char *devname
, dev_t
*rdev
)
698 if (fstat(fd
, &stb
) != 0) {
699 pr_err("fstat failed for %s: %s\n", devname
, strerror(errno
));
702 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
703 pr_err("%s is not a block device.\n", devname
);
711 int stat_is_blkdev(char *devname
, dev_t
*rdev
)
715 if (stat(devname
, &stb
) != 0) {
716 pr_err("stat failed for %s: %s\n", devname
, strerror(errno
));
719 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
720 pr_err("%s is not a block device.\n", devname
);
732 for (i
= 0; i
< 5; i
++) {
734 fprintf(stderr
, "%s%s", mesg
, add
);
736 if (fgets(buf
, 100, stdin
)==NULL
)
738 if (buf
[0]=='y' || buf
[0]=='Y')
740 if (buf
[0]=='n' || buf
[0]=='N')
744 pr_err("assuming 'no'\n");
748 int is_standard(char *dev
, int *nump
)
750 /* tests if dev is a "standard" md dev name.
751 * i.e if the last component is "/dNN" or "/mdNN",
752 * where NN is a string of digits
753 * Returns 1 if a partitionable standard,
754 * -1 if non-partitonable,
755 * 0 if not a standard name.
757 char *d
= strrchr(dev
, '/');
762 if (strncmp(d
, "/d",2) == 0)
763 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
764 else if (strncmp(d
, "/md_d", 5) == 0)
765 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
766 else if (strncmp(d
, "/md", 3) == 0)
767 d
+= 3, type
= -1; /* /dev/mdN */
768 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
769 d
+= 1, type
= -1; /* /dev/md/N */
779 if (nump
) *nump
= num
;
784 unsigned long calc_csum(void *super
, int bytes
)
786 unsigned long long newcsum
= 0;
789 unsigned int *superc
= (unsigned int*) super
;
791 for(i
= 0; i
< bytes
/4; i
++)
792 newcsum
+= superc
[i
];
793 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
795 /* The in-kernel checksum calculation is always 16bit on
796 * the alpha, though it is 32 bit on i386...
797 * I wonder what it is elsewhere... (it uses an API in
798 * a way that it shouldn't).
800 csum
= (csum
& 0xffff) + (csum
>> 16);
801 csum
= (csum
& 0xffff) + (csum
>> 16);
806 char *human_size(long long bytes
)
810 /* We convert bytes to either centi-M{ega,ibi}bytes,
811 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
812 * with appropriate rounding, and then print
813 * 1/100th of those as a decimal.
814 * We allow upto 2048Megabytes before converting to
815 * gigabytes and 2048Gigabytes before converting to
816 * terabytes, as that shows more precision and isn't
817 * too large a number.
820 if (bytes
< 5000*1024)
822 else if (bytes
< 2*1024LL*1024LL*1024LL) {
823 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
824 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
825 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
826 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
827 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
828 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
829 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
830 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
831 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
833 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
834 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
835 snprintf(buf
, sizeof(buf
), " (%ld.%02ld TiB %ld.%02ld TB)",
836 cTiB
/100, cTiB
% 100, cTB
/100, cTB
% 100);
841 char *human_size_brief(long long bytes
, int prefix
)
845 /* We convert bytes to either centi-M{ega,ibi}bytes,
846 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
847 * with appropriate rounding, and then print
848 * 1/100th of those as a decimal.
849 * We allow upto 2048Megabytes before converting to
850 * gigabytes and 2048Gigabytes before converting to
851 * terabytes, as that shows more precision and isn't
852 * too large a number.
854 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
855 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
858 if (bytes
< 5000*1024)
860 else if (prefix
== IEC
) {
861 if (bytes
< 2*1024LL*1024LL*1024LL) {
862 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
863 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
864 cMiB
/100, cMiB
% 100);
865 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
866 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
867 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
868 cGiB
/100, cGiB
% 100);
870 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
871 snprintf(buf
, sizeof(buf
), "%ld.%02ldTiB",
872 cTiB
/100, cTiB
% 100);
875 else if (prefix
== JEDEC
) {
876 if (bytes
< 2*1024LL*1024LL*1024LL) {
877 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
878 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
880 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
881 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
882 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
885 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
886 snprintf(buf
, sizeof(buf
), "%ld.%02ldTB",
896 void print_r10_layout(int layout
)
898 int near
= layout
& 255;
899 int far
= (layout
>> 8) & 255;
900 int offset
= (layout
&0x10000);
904 printf("%s near=%d", sep
, near
);
908 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
910 printf("NO REDUNDANCY");
913 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
914 int chunksize
, unsigned long long devsize
)
918 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
919 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
922 int get_data_disks(int level
, int layout
, int raid_disks
)
926 case 0: data_disks
= raid_disks
;
928 case 1: data_disks
= 1;
931 case 5: data_disks
= raid_disks
- 1;
933 case 6: data_disks
= raid_disks
- 2;
935 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
942 dev_t
devnm2devid(char *devnm
)
944 /* First look in /sys/block/$DEVNM/dev for %d:%d
945 * If that fails, try parsing out a number
952 snprintf(path
, sizeof(path
), "/sys/block/%s/dev", devnm
);
953 fd
= open(path
, O_RDONLY
);
956 int n
= read(fd
, buf
, sizeof(buf
));
960 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
961 return makedev(mjr
, mnr
);
963 if (strncmp(devnm
, "md_d", 4) == 0 &&
965 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
966 ep
> devnm
&& *ep
== 0)
967 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
969 if (strncmp(devnm
, "md", 2) == 0 &&
971 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
972 ep
> devnm
&& *ep
== 0)
973 return makedev(MD_MAJOR
, mnr
);
979 * is_devname_numbered() - helper for numbered devname verification.
980 * @devname: path or name to check.
981 * @pref: expected devname prefix.
982 * @pref_len: prefix len.
984 static bool is_devname_numbered(const char *devname
, const char *pref
, const int pref_len
)
988 assert(devname
&& pref
);
990 if (strncmp(devname
, pref
, pref_len
) != 0)
993 if (parse_num(&val
, devname
+ pref_len
) != 0)
1003 * is_devname_md_numbered() - check if &devname is numbered MD device (md).
1004 * @devname: path or name to check.
1006 bool is_devname_md_numbered(const char *devname
)
1008 return is_devname_numbered(devname
, DEV_NUM_PREF
, DEV_NUM_PREF_LEN
);
1012 * is_devname_md_d_numbered() - check if &devname is secondary numbered MD device (md_d).
1013 * @devname: path or name to check.
1015 bool is_devname_md_d_numbered(const char *devname
)
1017 static const char d_dev
[] = DEV_NUM_PREF
"_d";
1019 return is_devname_numbered(devname
, d_dev
, sizeof(d_dev
) - 1);
1023 * get_md_name() - Get main dev node of the md device.
1024 * @devnm: Md device name or path.
1026 * Function checks if the full name was passed and returns md name
1027 * if it is the MD device.
1029 * Return: Main dev node of the md device or NULL if not found.
1031 char *get_md_name(char *devnm
)
1033 static char devname
[NAME_MAX
];
1036 if (strncmp(devnm
, "/dev/", 5) == 0)
1037 snprintf(devname
, sizeof(devname
), "%s", devnm
);
1039 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
1041 if (!is_mddev(devname
))
1043 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
)
1049 void put_md_name(char *name
)
1051 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1055 int get_maj_min(char *dev
, int *major
, int *minor
)
1058 *major
= strtoul(dev
, &e
, 0);
1059 return (e
> dev
&& *e
== ':' && e
[1] &&
1060 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1065 * is_bit_set() - get bit value by index.
1067 * @index: index of the bit (LSB numbering).
1069 * Return: bit value.
1071 bool is_bit_set(int *val
, unsigned char index
)
1073 if ((*val
) & (1 << index
))
1078 int dev_open(char *dev
, int flags
)
1080 /* like 'open', but if 'dev' matches %d:%d, create a temp
1081 * block device and open that
1092 if (get_maj_min(dev
, &major
, &minor
)) {
1093 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1094 (int)getpid(), major
, minor
);
1095 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1096 fd
= open(devname
, flags
);
1100 /* Try /tmp as /dev appear to be read-only */
1101 snprintf(devname
, sizeof(devname
),
1102 "/tmp/.tmp.md.%d:%d:%d",
1103 (int)getpid(), major
, minor
);
1104 if (mknod(devname
, S_IFBLK
|0600,
1105 makedev(major
, minor
)) == 0) {
1106 fd
= open(devname
, flags
);
1111 fd
= open(dev
, flags
);
1115 int open_dev_flags(char *devnm
, int flags
)
1120 devid
= devnm2devid(devnm
);
1121 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1122 return dev_open(buf
, flags
);
1125 int open_dev(char *devnm
)
1127 return open_dev_flags(devnm
, O_RDONLY
);
1130 int open_dev_excl(char *devnm
)
1135 dev_t devid
= devnm2devid(devnm
);
1136 unsigned int delay
= 1; // miliseconds
1138 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1139 for (i
= 0; i
< 25; i
++) {
1140 int fd
= dev_open(buf
, flags
|O_EXCL
);
1143 if (errno
== EACCES
&& flags
== O_RDWR
) {
1149 sleep_for(0, MSEC_TO_NSEC(delay
), true);
1156 int same_dev(char *one
, char *two
)
1158 struct stat st1
, st2
;
1159 if (stat(one
, &st1
) != 0)
1161 if (stat(two
, &st2
) != 0)
1163 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1165 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1167 return st1
.st_rdev
== st2
.st_rdev
;
1170 void wait_for(char *dev
, int fd
)
1173 struct stat stb_want
;
1174 unsigned int delay
= 1; // miliseconds
1176 if (fstat(fd
, &stb_want
) != 0 ||
1177 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1180 for (i
= 0; i
< 25; i
++) {
1182 if (stat(dev
, &stb
) == 0 &&
1183 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1184 (stb
.st_rdev
== stb_want
.st_rdev
))
1186 sleep_for(0, MSEC_TO_NSEC(delay
), true);
1191 pr_err("timeout waiting for %s\n", dev
);
1194 struct superswitch
*superlist
[] =
1197 &super_ddf
, &super_imsm
,
1202 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1204 mdu_array_info_t array
;
1207 struct supertype
*st
= NULL
;
1212 char *subarray
= NULL
;
1213 char container
[32] = "";
1216 devnm
= fd2devnm(fd
);
1220 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1223 vers
= sra
->array
.major_version
;
1224 minor
= sra
->array
.minor_version
;
1225 verstr
= sra
->text_version
;
1227 if (md_get_array_info(fd
, &array
))
1228 array
.major_version
= array
.minor_version
= 0;
1229 vers
= array
.major_version
;
1230 minor
= array
.minor_version
;
1235 sprintf(version
, "%d.%d", vers
, minor
);
1238 if (minor
== -2 && is_subarray(verstr
)) {
1239 char *dev
= verstr
+1;
1241 subarray
= strchr(dev
, '/');
1244 subarray
= xstrdup(subarray
);
1246 strcpy(container
, dev
);
1248 sra
= sysfs_read(-1, container
, GET_VERSION
);
1249 if (sra
&& sra
->text_version
[0])
1250 verstr
= sra
->text_version
;
1252 verstr
= "-no-metadata-";
1255 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1256 st
= superlist
[i
]->match_metadata_desc(verstr
);
1262 *subarrayp
= subarray
;
1263 strcpy(st
->container_devnm
, container
);
1264 strncpy(st
->devnm
, devnm
, MD_NAME_MAX
- 1);
1271 struct supertype
*dup_super(struct supertype
*orig
)
1273 struct supertype
*st
;
1277 st
= xcalloc(1, sizeof(*st
));
1279 st
->max_devs
= orig
->max_devs
;
1280 st
->minor_version
= orig
->minor_version
;
1281 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1282 st
->data_offset
= orig
->data_offset
;
1288 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1290 /* try each load_super to find the best match,
1291 * and return the best superswitch
1293 struct superswitch
*ss
;
1294 struct supertype
*st
;
1295 unsigned int besttime
= 0;
1299 st
= xcalloc(1, sizeof(*st
));
1300 st
->container_devnm
[0] = 0;
1302 for (i
= 0; superlist
[i
]; i
++) {
1305 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1307 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1309 memset(st
, 0, sizeof(*st
));
1310 st
->ignore_hw_compat
= 1;
1311 rv
= ss
->load_super(st
, fd
, NULL
);
1314 st
->ss
->getinfo_super(st
, &info
, NULL
);
1315 if (bestsuper
== -1 ||
1316 besttime
< info
.array
.ctime
) {
1318 besttime
= info
.array
.ctime
;
1323 if (bestsuper
!= -1) {
1325 memset(st
, 0, sizeof(*st
));
1326 st
->ignore_hw_compat
= 1;
1327 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1329 superlist
[bestsuper
]->free_super(st
);
1337 /* Return size of device in bytes */
1338 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1340 unsigned long long ldsize
;
1343 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1344 ldsize
= (unsigned long long)st
.st_size
;
1347 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1350 unsigned long dsize
;
1351 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1356 pr_err("Cannot get size of %s: %s\n",
1357 dname
, strerror(errno
));
1365 /* Return sector size of device in bytes */
1366 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1368 unsigned int sectsize
;
1370 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1372 pr_err("Cannot get sector size of %s: %s\n",
1373 dname
, strerror(errno
));
1377 *sectsizep
= sectsize
;
1381 /* Return true if this can only be a container, not a member device.
1382 * i.e. is and md device and size is zero
1384 int must_be_container(int fd
)
1387 unsigned long long size
;
1389 mdi
= sysfs_read(fd
, NULL
, GET_VERSION
);
1394 if (get_dev_size(fd
, NULL
, &size
) == 0)
1401 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1402 * Returns: 1 if successful
1403 * -1 for unknown partition type
1404 * 0 for other errors
1406 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1409 unsigned char empty_gpt_entry
[16]= {0};
1410 struct GPT_part_entry
*part
;
1412 unsigned long long curr_part_end
;
1413 unsigned all_partitions
, entry_size
;
1415 unsigned int sector_size
= 0;
1419 BUILD_BUG_ON(sizeof(gpt
) != 512);
1420 /* skip protective MBR */
1421 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1423 lseek(fd
, sector_size
, SEEK_SET
);
1424 /* read GPT header */
1425 if (read(fd
, &gpt
, 512) != 512)
1428 /* get the number of partition entries and the entry size */
1429 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1430 entry_size
= __le32_to_cpu(gpt
.part_size
);
1432 /* Check GPT signature*/
1433 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1437 if (all_partitions
> 1024 ||
1438 entry_size
> sizeof(buf
))
1441 part
= (struct GPT_part_entry
*)buf
;
1443 /* set offset to third block (GPT entries) */
1444 lseek(fd
, sector_size
*2, SEEK_SET
);
1445 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1446 /* read partition entry */
1447 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1450 /* is this valid partition? */
1451 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1452 /* check the last lba for the current partition */
1453 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1454 if (curr_part_end
> *endofpart
)
1455 *endofpart
= curr_part_end
;
1462 /* Sets endofpart parameter to the last block used by the last partition on the device.
1463 * Returns: 1 if successful
1464 * -1 for unknown partition type
1465 * 0 for other errors
1467 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1469 struct MBR boot_sect
;
1470 unsigned long long curr_part_end
;
1472 unsigned int sector_size
;
1477 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1480 if (read(fd
, &boot_sect
, 512) != 512)
1483 /* check MBP signature */
1484 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1486 /* found the correct signature */
1488 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1490 * Have to make every access through boot_sect rather
1491 * than using a pointer to the partition table (or an
1492 * entry), since the entries are not properly aligned.
1495 /* check for GPT type */
1496 if (boot_sect
.parts
[part_nr
].part_type
==
1497 MBR_GPT_PARTITION_TYPE
) {
1498 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1501 /* check the last used lba for the current partition */
1503 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1504 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1505 if (curr_part_end
> *endofpart
)
1506 *endofpart
= curr_part_end
;
1509 /* Unknown partition table */
1512 /* calculate number of 512-byte blocks */
1513 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1514 *endofpart
*= (sector_size
/ 512);
1519 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1520 unsigned long long size
)
1523 * Check where the last partition ends
1525 unsigned long long endofpart
;
1527 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1528 /* There appears to be a partition table here */
1529 if (freesize
== 0) {
1530 /* partitions will not be visible in new device */
1531 pr_err("partition table exists on %s but will be lost or\n"
1532 " meaningless after creating array\n",
1535 } else if (endofpart
> freesize
) {
1536 /* last partition overlaps metadata */
1537 pr_err("metadata will over-write last partition on %s.\n",
1540 } else if (size
&& endofpart
> size
) {
1541 /* partitions will be truncated in new device */
1542 pr_err("array size is too small to cover all partitions on %s.\n",
1550 int open_container(int fd
)
1552 /* 'fd' is a block device. Find out if it is in use
1553 * by a container, and return an open fd on that container.
1564 if (fstat(fd
, &st
) != 0)
1566 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1567 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1568 e
= path
+ strlen(path
);
1570 dir
= opendir(path
);
1573 while ((de
= readdir(dir
))) {
1576 if (de
->d_name
[0] == '.')
1578 /* Need to make sure it is a container and not a volume */
1579 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1580 dfd
= open(path
, O_RDONLY
);
1583 n
= read(dfd
, buf
, sizeof(buf
));
1585 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1588 if (strncmp(buf
, "external", 8) != 0 ||
1592 sprintf(e
, "/%s/dev", de
->d_name
);
1593 dfd
= open(path
, O_RDONLY
);
1596 n
= read(dfd
, buf
, sizeof(buf
));
1598 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1601 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1603 sprintf(buf
, "%d:%d", major
, minor
);
1604 dfd
= dev_open(buf
, O_RDONLY
);
1614 struct superswitch
*version_to_superswitch(char *vers
)
1618 for (i
= 0; superlist
[i
]; i
++) {
1619 struct superswitch
*ss
= superlist
[i
];
1621 if (strcmp(vers
, ss
->name
) == 0)
1628 int metadata_container_matches(char *metadata
, char *devnm
)
1630 /* Check if 'devnm' is the container named in 'metadata'
1632 * /containername/componentname or
1633 * -containername/componentname
1636 if (*metadata
!= '/' && *metadata
!= '-')
1639 if (strncmp(metadata
+1, devnm
, l
) != 0)
1641 if (metadata
[l
+1] != '/')
1646 int metadata_subdev_matches(char *metadata
, char *devnm
)
1648 /* Check if 'devnm' is the subdev named in 'metadata'
1650 * /containername/subdev or
1651 * -containername/subdev
1654 if (*metadata
!= '/' && *metadata
!= '-')
1656 sl
= strchr(metadata
+1, '/');
1659 if (strcmp(sl
+1, devnm
) == 0)
1664 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1666 if (mdstat
->metadata_version
== NULL
||
1667 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1668 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1674 int is_subarray_active(char *subarray
, char *container
)
1676 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1677 struct mdstat_ent
*ent
;
1679 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1680 if (is_container_member(ent
, container
))
1681 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1684 free_mdstat(mdstat
);
1689 /* open_subarray - opens a subarray in a container
1690 * @dev: container device name
1691 * @st: empty supertype
1692 * @quiet: block reporting errors flag
1694 * On success returns an fd to a container and fills in *st
1696 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1699 struct mdinfo
*info
;
1703 fd
= open(dev
, O_RDWR
|O_EXCL
);
1706 pr_err("Couldn't open %s, aborting\n",
1711 _devnm
= fd2devnm(fd
);
1712 if (_devnm
== NULL
) {
1714 pr_err("Failed to determine device number for %s\n",
1718 strcpy(st
->devnm
, _devnm
);
1720 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1723 pr_err("Failed to read sysfs for %s\n",
1728 if (mdi
->array
.level
!= UnSet
) {
1730 pr_err("%s is not a container\n", dev
);
1734 st
->ss
= version_to_superswitch(mdi
->text_version
);
1737 pr_err("Operation not supported for %s metadata\n",
1742 if (st
->devnm
[0] == 0) {
1744 pr_err("Failed to allocate device name\n");
1748 if (!st
->ss
->load_container
) {
1750 pr_err("%s is not a container\n", dev
);
1754 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1756 pr_err("Failed to load metadata for %s\n",
1761 info
= st
->ss
->container_content(st
, subarray
);
1764 pr_err("Failed to find subarray-%s in %s\n",
1774 st
->ss
->free_super(st
);
1787 int add_disk(int mdfd
, struct supertype
*st
,
1788 struct mdinfo
*sra
, struct mdinfo
*info
)
1790 /* Add a device to an array, in one of 2 ways. */
1793 if (st
->ss
->external
) {
1794 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1795 info
->recovery_start
= MaxSector
;
1797 info
->recovery_start
= 0;
1798 rv
= sysfs_add_disk(sra
, info
, 0);
1801 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1805 sd2
= xmalloc(sizeof(*sd2
));
1807 sd2
->next
= sra
->devs
;
1812 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1816 int remove_disk(int mdfd
, struct supertype
*st
,
1817 struct mdinfo
*sra
, struct mdinfo
*info
)
1821 /* Remove the disk given by 'info' from the array */
1822 if (st
->ss
->external
)
1823 rv
= sysfs_set_str(sra
, info
, "slot", STR_COMMON_NONE
);
1825 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1830 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1832 int cnt
= force
? 500 : 5;
1835 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1836 * outstanding IO requests to the device.
1837 * In this case, it can be helpful to wait a little while,
1838 * up to 5 seconds if 'force' is set, or 50 msec if not.
1840 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1843 sleep_for(0, MSEC_TO_NSEC(10), true);
1848 int sys_hot_remove_disk(int statefd
, int force
)
1850 int cnt
= force
? 500 : 5;
1853 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1856 sleep_for(0, MSEC_TO_NSEC(10), true);
1857 return ret
== 6 ? 0 : -1;
1860 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1862 /* Initialise kernel's knowledge of array.
1863 * This varies between externally managed arrays
1866 mdu_array_info_t inf
;
1869 if (st
->ss
->external
)
1870 return sysfs_set_array(info
, 9003);
1872 memset(&inf
, 0, sizeof(inf
));
1873 inf
.major_version
= info
->array
.major_version
;
1874 inf
.minor_version
= info
->array
.minor_version
;
1875 rv
= md_set_array_info(mdfd
, &inf
);
1880 unsigned long long min_recovery_start(struct mdinfo
*array
)
1882 /* find the minimum recovery_start in an array for metadata
1883 * formats that only record per-array recovery progress instead
1886 unsigned long long recovery_start
= MaxSector
;
1889 for (d
= array
->devs
; d
; d
= d
->next
)
1890 recovery_start
= min(recovery_start
, d
->recovery_start
);
1892 return recovery_start
;
1895 int mdmon_pid(char *devnm
)
1902 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1904 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1908 n
= read(fd
, pid
, 9);
1915 int mdmon_running(char *devnm
)
1917 int pid
= mdmon_pid(devnm
);
1920 if (kill(pid
, 0) == 0)
1925 int start_mdmon(char *devnm
)
1931 char *prefix
= in_initrd() ? "initrd-" : "";
1940 if (check_env("MDADM_NO_MDMON"))
1942 if (continue_via_systemd(devnm
, MDMON_SERVICE
, prefix
))
1945 /* That failed, try running mdmon directly */
1946 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1950 sl
= strrchr(pathbuf
, '/');
1955 strcpy(sl
, "mdmon");
1962 for (i
= 0; paths
[i
]; i
++)
1964 execl(paths
[i
], paths
[i
],
1968 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1970 default: /* parent - good */
1971 pid
= wait(&status
);
1972 if (pid
< 0 || status
!= 0) {
1973 pr_err("failed to launch mdmon. Array remains readonly\n");
1980 __u32
random32(void)
1983 int rfd
= open("/dev/urandom", O_RDONLY
);
1984 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1991 void random_uuid(__u8
*buf
)
1996 fd
= open("/dev/urandom", O_RDONLY
);
1999 len
= read(fd
, buf
, 16);
2007 for (i
= 0; i
< 4; i
++)
2012 int flush_metadata_updates(struct supertype
*st
)
2016 st
->update_tail
= NULL
;
2020 sfd
= connect_monitor(st
->container_devnm
);
2024 while (st
->updates
) {
2025 struct metadata_update
*mu
= st
->updates
;
2026 st
->updates
= mu
->next
;
2028 send_message(sfd
, mu
, 0);
2036 st
->update_tail
= NULL
;
2040 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2043 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2048 mu
->space_list
= NULL
;
2050 *st
->update_tail
= mu
;
2051 st
->update_tail
= &mu
->next
;
2055 /* tinyc doesn't optimize this check in ioctl.h out ... */
2056 unsigned int __invalid_size_argument_for_IOC
= 0;
2060 * disk_fd_matches_criteria() - check if device matches spare criteria.
2061 * @st: supertype, not NULL.
2062 * @disk_fd: file descriptor of the disk.
2063 * @sc: criteria to test.
2065 * Return: true if disk matches criteria, false otherwise.
2067 bool disk_fd_matches_criteria(struct supertype
*st
, int disk_fd
, struct spare_criteria
*sc
)
2069 unsigned int dev_sector_size
= 0;
2070 unsigned long long dev_size
= 0;
2072 if (!sc
->criteria_set
)
2075 if (!get_dev_size(disk_fd
, NULL
, &dev_size
) || dev_size
< sc
->min_size
)
2078 if (!get_dev_sector_size(disk_fd
, NULL
, &dev_sector_size
) ||
2079 sc
->sector_size
!= dev_sector_size
)
2082 if (drive_test_and_add_policies(st
, &sc
->pols
, disk_fd
, 0))
2089 * devid_matches_criteria() - check if device referenced by devid matches spare criteria.
2090 * @st: supertype, not NULL.
2091 * @devid: devid of the device to check.
2092 * @sc: criteria to test.
2094 * Return: true if disk matches criteria, false otherwise.
2096 bool devid_matches_criteria(struct supertype
*st
, dev_t devid
, struct spare_criteria
*sc
)
2102 if (!sc
->criteria_set
)
2105 snprintf(buf
, NAME_MAX
, "%d:%d", major(devid
), minor(devid
));
2107 fd
= dev_open(buf
, O_RDONLY
);
2108 if (!is_fd_valid(fd
))
2111 /* Error code inherited */
2112 ret
= disk_fd_matches_criteria(st
, fd
, sc
);
2118 /* Pick all spares matching given criteria from a container
2119 * if min_size == 0 do not check size
2120 * if domlist == NULL do not check domains
2121 * if spare_group given add it to domains of each spare
2122 * metadata allows to test domains using metadata of destination array */
2123 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2124 struct spare_criteria
*criteria
,
2125 struct domainlist
*domlist
,
2127 const char *metadata
, int get_one
)
2129 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2131 /* get list of all disks in container */
2132 if (st
->ss
->getinfo_super_disks
)
2133 disks
= st
->ss
->getinfo_super_disks(st
);
2137 /* find spare devices on the list */
2139 disks
->array
.spare_disks
= 0;
2144 if (d
->disk
.state
== 0) {
2145 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2147 found
= devid_matches_criteria(st
, dev
, criteria
);
2149 /* check if domain matches */
2150 if (found
&& domlist
) {
2151 struct dev_policy
*pol
= devid_policy(dev
);
2153 pol_add(&pol
, pol_domain
,
2155 if (domain_test(domlist
, pol
, metadata
) != 1)
2158 dev_policy_free(pol
);
2163 disks
->array
.spare_disks
++;
2177 /* Checks if paths point to the same device
2178 * Returns 0 if they do.
2179 * Returns 1 if they don't.
2180 * Returns -1 if something went wrong,
2181 * e.g. paths are empty or the files
2182 * they point to don't exist */
2183 int compare_paths (char* path1
, char* path2
)
2185 struct stat st1
,st2
;
2187 if (path1
== NULL
|| path2
== NULL
)
2189 if (stat(path1
,&st1
) != 0)
2191 if (stat(path2
,&st2
) != 0)
2193 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2198 /* Make sure we can open as many devices as needed */
2199 void enable_fds(int devices
)
2201 unsigned int fds
= 20 + devices
;
2203 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0 || lim
.rlim_cur
>= fds
)
2205 if (lim
.rlim_max
< fds
)
2208 setrlimit(RLIMIT_NOFILE
, &lim
);
2211 /* Close all opened descriptors if needed and redirect
2212 * streams to /dev/null.
2213 * For debug purposed, leave STDOUT and STDERR untouched
2215 * 1- if any error occurred
2218 void manage_fork_fds(int close_all
)
2221 struct dirent
*dirent
;
2224 open("/dev/null", O_RDWR
);
2234 dir
= opendir("/proc/self/fd");
2236 pr_err("Cannot open /proc/self/fd directory.\n");
2239 for (dirent
= readdir(dir
); dirent
; dirent
= readdir(dir
)) {
2242 if ((strcmp(dirent
->d_name
, ".") == 0) ||
2243 (strcmp(dirent
->d_name
, "..")) == 0)
2246 fd
= strtol(dirent
->d_name
, NULL
, 10);
2252 /* In a systemd/udev world, it is best to get systemd to
2253 * run daemon rather than running in the background.
2255 * 1- if systemd service has been started
2258 int continue_via_systemd(char *devnm
, char *service_name
, char *prefix
)
2263 /* Simply return that service cannot be started */
2264 if (check_env("MDADM_NO_SYSTEMCTL"))
2269 snprintf(pathbuf
, sizeof(pathbuf
),
2270 "%s@%s%s.service", service_name
, prefix
?: "", devnm
);
2271 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
2273 status
= execl("/bin/systemctl", "systemctl", "restart",
2276 case -1: /* Just do it ourselves. */
2278 default: /* parent - good */
2279 pid
= wait(&status
);
2280 if (pid
>= 0 && status
== 0)
2288 return access("/etc/initrd-release", F_OK
) >= 0;
2291 void reopen_mddev(int mdfd
)
2293 /* Re-open without any O_EXCL, but keep
2298 devnm
= fd2devnm(mdfd
);
2300 fd
= open_dev(devnm
);
2301 if (fd
>= 0 && fd
!= mdfd
)
2305 static struct cmap_hooks
*cmap_hooks
= NULL
;
2306 static int is_cmap_hooks_ready
= 0;
2308 void set_cmap_hooks(void)
2310 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2311 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2312 if (!cmap_hooks
->cmap_handle
)
2315 cmap_hooks
->initialize
=
2316 dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2317 cmap_hooks
->get_string
=
2318 dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2319 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2321 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2322 !cmap_hooks
->finalize
)
2323 dlclose(cmap_hooks
->cmap_handle
);
2325 is_cmap_hooks_ready
= 1;
2328 int get_cluster_name(char **cluster_name
)
2331 cmap_handle_t handle
;
2333 if (!is_cmap_hooks_ready
)
2336 rv
= cmap_hooks
->initialize(&handle
);
2340 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2342 free(*cluster_name
);
2349 cmap_hooks
->finalize(handle
);
2354 void set_dlm_hooks(void)
2356 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2357 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2358 if (!dlm_hooks
->dlm_handle
)
2361 dlm_hooks
->open_lockspace
=
2362 dlsym(dlm_hooks
->dlm_handle
, "dlm_open_lockspace");
2363 dlm_hooks
->create_lockspace
=
2364 dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2365 dlm_hooks
->release_lockspace
=
2366 dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2367 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2368 dlm_hooks
->ls_unlock_wait
=
2369 dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock_wait");
2370 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2371 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2373 if (!dlm_hooks
->open_lockspace
|| !dlm_hooks
->create_lockspace
||
2374 !dlm_hooks
->ls_lock
|| !dlm_hooks
->ls_unlock_wait
||
2375 !dlm_hooks
->release_lockspace
|| !dlm_hooks
->ls_get_fd
||
2376 !dlm_hooks
->dispatch
)
2377 dlclose(dlm_hooks
->dlm_handle
);
2379 is_dlm_hooks_ready
= 1;
2382 void set_hooks(void)
2388 int zero_disk_range(int fd
, unsigned long long sector
, size_t count
)
2394 size_t len
= count
* 512;
2397 fd_zero
= open("/dev/zero", O_RDONLY
);
2399 pr_err("Cannot open /dev/zero\n");
2403 if (lseek64(fd
, sector
* 512, SEEK_SET
) < 0) {
2405 pr_err("Failed to seek offset for zeroing\n");
2409 addr
= mmap(NULL
, len
, PROT_READ
, MAP_PRIVATE
, fd_zero
, 0);
2411 if (addr
== MAP_FAILED
) {
2413 pr_err("Mapping /dev/zero failed\n");
2418 n
= write(fd
, addr
+ written
, len
- written
);
2423 pr_err("Zeroing disk range failed\n");
2427 } while (written
!= len
);
2437 * sleep_for() - Sleeps for specified time.
2438 * @sec: Seconds to sleep for.
2439 * @nsec: Nanoseconds to sleep for, has to be less than one second.
2440 * @wake_after_interrupt: If set, wake up if interrupted.
2442 * Function immediately returns if error different than EINTR occurs.
2444 void sleep_for(unsigned int sec
, long nsec
, bool wake_after_interrupt
)
2446 struct timespec delay
= {.tv_sec
= sec
, .tv_nsec
= nsec
};
2448 assert(nsec
< MSEC_TO_NSEC(1000));
2452 nanosleep(&delay
, &delay
);
2453 if (errno
!= 0 && errno
!= EINTR
) {
2454 pr_err("Error sleeping for %us %ldns: %s\n", sec
, nsec
, strerror(errno
));
2457 } while (!wake_after_interrupt
&& errno
== EINTR
);
2460 /* is_directory() - Checks if directory provided by path is indeed a regular directory.
2461 * @path: directory path to be checked
2463 * Doesn't accept symlinks.
2465 * Return: true if is a directory, false if not
2467 bool is_directory(const char *path
)
2471 if (lstat(path
, &st
) != 0) {
2472 pr_err("%s: %s\n", strerror(errno
), path
);
2476 if (!S_ISDIR(st
.st_mode
))
2483 * is_file() - Checks if file provided by path is indeed a regular file.
2484 * @path: file path to be checked
2486 * Doesn't accept symlinks.
2488 * Return: true if is a file, false if not
2490 bool is_file(const char *path
)
2494 if (lstat(path
, &st
) != 0) {
2495 pr_err("%s: %s\n", strerror(errno
), path
);
2499 if (!S_ISREG(st
.st_mode
))