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
]);
592 char *fname_from_uuid(struct supertype
*st
, struct mdinfo
*info
,
595 // dirty hack to work around an issue with super1 superblocks...
596 // super1 superblocks need swapuuid set in order for assembly to
597 // work, but can't have it set if we want this printout to match
598 // all the other uuid printouts in super1.c, so we force swapuuid
599 // to 1 to make our printout match the rest of super1
600 #if __BYTE_ORDER == BIG_ENDIAN
601 return __fname_from_uuid(info
->uuid
, 1, buf
, sep
);
603 return __fname_from_uuid(info
->uuid
, (st
->ss
== &super1
) ? 1 :
604 st
->ss
->swapuuid
, buf
, sep
);
608 int check_ext2(int fd
, char *name
)
611 * Check for an ext2fs file system.
612 * Superblock is always 1K at 1K offset
614 * s_magic is le16 at 56 == 0xEF53
615 * report mtime - le32 at 44
617 * logblksize - le32 at 24
619 unsigned char sb
[1024];
621 unsigned long long size
;
623 if (lseek(fd
, 1024,0)!= 1024)
625 if (read(fd
, sb
, 1024)!= 1024)
627 if (sb
[56] != 0x53 || sb
[57] != 0xef)
630 mtime
= sb
[44]|(sb
[45]|(sb
[46]|sb
[47]<<8)<<8)<<8;
631 bsize
= sb
[24]|(sb
[25]|(sb
[26]|sb
[27]<<8)<<8)<<8;
632 size
= sb
[4]|(sb
[5]|(sb
[6]|sb
[7]<<8)<<8)<<8;
634 pr_err("%s appears to contain an ext2fs file system\n",
636 cont_err("size=%lluK mtime=%s", size
, ctime(&mtime
));
640 int check_reiser(int fd
, char *name
)
643 * superblock is at 64K
645 * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
648 unsigned char sb
[1024];
649 unsigned long long size
;
650 if (lseek(fd
, 64*1024, 0) != 64*1024)
652 if (read(fd
, sb
, 1024) != 1024)
654 if (strncmp((char*)sb
+52, "ReIsErFs",8) != 0 &&
655 strncmp((char*)sb
+52, "ReIsEr2Fs",9) != 0)
657 pr_err("%s appears to contain a reiserfs file system\n",name
);
658 size
= sb
[0]|(sb
[1]|(sb
[2]|sb
[3]<<8)<<8)<<8;
659 cont_err("size = %lluK\n", size
*4);
664 int check_raid(int fd
, char *name
)
669 struct supertype
*st
= guess_super(fd
);
673 if (st
->ss
->add_to_super
!= NULL
) {
674 st
->ss
->load_super(st
, fd
, name
);
675 /* Looks like a raid array .. */
676 pr_err("%s appears to be part of a raid array:\n", name
);
677 st
->ss
->getinfo_super(st
, &info
, NULL
);
678 st
->ss
->free_super(st
);
679 crtime
= info
.array
.ctime
;
680 level
= map_num(pers
, info
.array
.level
);
683 cont_err("level=%s devices=%d ctime=%s",
684 level
, info
.array
.raid_disks
, ctime(&crtime
));
686 /* Looks like GPT or MBR */
687 pr_err("partition table exists on %s\n", name
);
692 int fstat_is_blkdev(int fd
, char *devname
, dev_t
*rdev
)
696 if (fstat(fd
, &stb
) != 0) {
697 pr_err("fstat failed for %s: %s\n", devname
, strerror(errno
));
700 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
701 pr_err("%s is not a block device.\n", devname
);
709 int stat_is_blkdev(char *devname
, dev_t
*rdev
)
713 if (stat(devname
, &stb
) != 0) {
714 pr_err("stat failed for %s: %s\n", devname
, strerror(errno
));
717 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
) {
718 pr_err("%s is not a block device.\n", devname
);
730 for (i
= 0; i
< 5; i
++) {
732 fprintf(stderr
, "%s%s", mesg
, add
);
734 if (fgets(buf
, 100, stdin
)==NULL
)
736 if (buf
[0]=='y' || buf
[0]=='Y')
738 if (buf
[0]=='n' || buf
[0]=='N')
742 pr_err("assuming 'no'\n");
746 int is_standard(char *dev
, int *nump
)
748 /* tests if dev is a "standard" md dev name.
749 * i.e if the last component is "/dNN" or "/mdNN",
750 * where NN is a string of digits
751 * Returns 1 if a partitionable standard,
752 * -1 if non-partitonable,
753 * 0 if not a standard name.
755 char *d
= strrchr(dev
, '/');
760 if (strncmp(d
, "/d",2) == 0)
761 d
+= 2, type
= 1; /* /dev/md/dN{pM} */
762 else if (strncmp(d
, "/md_d", 5) == 0)
763 d
+= 5, type
= 1; /* /dev/md_dN{pM} */
764 else if (strncmp(d
, "/md", 3) == 0)
765 d
+= 3, type
= -1; /* /dev/mdN */
766 else if (d
-dev
> 3 && strncmp(d
-2, "md/", 3) == 0)
767 d
+= 1, type
= -1; /* /dev/md/N */
777 if (nump
) *nump
= num
;
782 unsigned long calc_csum(void *super
, int bytes
)
784 unsigned long long newcsum
= 0;
787 unsigned int *superc
= (unsigned int*) super
;
789 for(i
= 0; i
< bytes
/4; i
++)
790 newcsum
+= superc
[i
];
791 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
793 /* The in-kernel checksum calculation is always 16bit on
794 * the alpha, though it is 32 bit on i386...
795 * I wonder what it is elsewhere... (it uses an API in
796 * a way that it shouldn't).
798 csum
= (csum
& 0xffff) + (csum
>> 16);
799 csum
= (csum
& 0xffff) + (csum
>> 16);
804 char *human_size(long long bytes
)
808 /* We convert bytes to either centi-M{ega,ibi}bytes,
809 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
810 * with appropriate rounding, and then print
811 * 1/100th of those as a decimal.
812 * We allow upto 2048Megabytes before converting to
813 * gigabytes and 2048Gigabytes before converting to
814 * terabytes, as that shows more precision and isn't
815 * too large a number.
818 if (bytes
< 5000*1024)
820 else if (bytes
< 2*1024LL*1024LL*1024LL) {
821 long cMiB
= (bytes
* 200LL / (1LL<<20) + 1) / 2;
822 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
823 snprintf(buf
, sizeof(buf
), " (%ld.%02ld MiB %ld.%02ld MB)",
824 cMiB
/100, cMiB
% 100, cMB
/100, cMB
% 100);
825 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
826 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) / 2;
827 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
828 snprintf(buf
, sizeof(buf
), " (%ld.%02ld GiB %ld.%02ld GB)",
829 cGiB
/100, cGiB
% 100, cGB
/100, cGB
% 100);
831 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
832 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
833 snprintf(buf
, sizeof(buf
), " (%ld.%02ld TiB %ld.%02ld TB)",
834 cTiB
/100, cTiB
% 100, cTB
/100, cTB
% 100);
839 char *human_size_brief(long long bytes
, int prefix
)
843 /* We convert bytes to either centi-M{ega,ibi}bytes,
844 * centi-G{igi,ibi}bytes or centi-T{era,ebi}bytes
845 * with appropriate rounding, and then print
846 * 1/100th of those as a decimal.
847 * We allow upto 2048Megabytes before converting to
848 * gigabytes and 2048Gigabytes before converting to
849 * terabytes, as that shows more precision and isn't
850 * too large a number.
852 * If prefix == IEC, we mean prefixes like kibi,mebi,gibi etc.
853 * If prefix == JEDEC, we mean prefixes like kilo,mega,giga etc.
856 if (bytes
< 5000*1024)
858 else if (prefix
== IEC
) {
859 if (bytes
< 2*1024LL*1024LL*1024LL) {
860 long cMiB
= (bytes
* 200LL / (1LL<<20) +1) /2;
861 snprintf(buf
, sizeof(buf
), "%ld.%02ldMiB",
862 cMiB
/100, cMiB
% 100);
863 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
864 long cGiB
= (bytes
* 200LL / (1LL<<30) +1) /2;
865 snprintf(buf
, sizeof(buf
), "%ld.%02ldGiB",
866 cGiB
/100, cGiB
% 100);
868 long cTiB
= (bytes
* 200LL / (1LL<<40) + 1) / 2;
869 snprintf(buf
, sizeof(buf
), "%ld.%02ldTiB",
870 cTiB
/100, cTiB
% 100);
873 else if (prefix
== JEDEC
) {
874 if (bytes
< 2*1024LL*1024LL*1024LL) {
875 long cMB
= (bytes
/ ( 1000000LL / 200LL ) +1) /2;
876 snprintf(buf
, sizeof(buf
), "%ld.%02ldMB",
878 } else if (bytes
< 2*1024LL*1024LL*1024LL*1024LL) {
879 long cGB
= (bytes
/ (1000000000LL/200LL ) +1) /2;
880 snprintf(buf
, sizeof(buf
), "%ld.%02ldGB",
883 long cTB
= (bytes
/ (1000000000000LL / 200LL) + 1) / 2;
884 snprintf(buf
, sizeof(buf
), "%ld.%02ldTB",
894 void print_r10_layout(int layout
)
896 int near
= layout
& 255;
897 int far
= (layout
>> 8) & 255;
898 int offset
= (layout
&0x10000);
902 printf("%s near=%d", sep
, near
);
906 printf("%s %s=%d", sep
, offset
?"offset":"far", far
);
908 printf("NO REDUNDANCY");
911 unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
912 int chunksize
, unsigned long long devsize
)
916 devsize
&= ~(unsigned long long)((chunksize
>>9)-1);
917 return get_data_disks(level
, layout
, raid_disks
) * devsize
;
920 int get_data_disks(int level
, int layout
, int raid_disks
)
924 case 0: data_disks
= raid_disks
;
926 case 1: data_disks
= 1;
929 case 5: data_disks
= raid_disks
- 1;
931 case 6: data_disks
= raid_disks
- 2;
933 case 10: data_disks
= raid_disks
/ (layout
& 255) / ((layout
>>8)&255);
940 dev_t
devnm2devid(char *devnm
)
942 /* First look in /sys/block/$DEVNM/dev for %d:%d
943 * If that fails, try parsing out a number
950 snprintf(path
, sizeof(path
), "/sys/block/%s/dev", devnm
);
951 fd
= open(path
, O_RDONLY
);
954 int n
= read(fd
, buf
, sizeof(buf
));
958 if (n
> 0 && sscanf(buf
, "%d:%d\n", &mjr
, &mnr
) == 2)
959 return makedev(mjr
, mnr
);
961 if (strncmp(devnm
, "md_d", 4) == 0 &&
963 (mnr
= strtoul(devnm
+4, &ep
, 10)) >= 0 &&
964 ep
> devnm
&& *ep
== 0)
965 return makedev(get_mdp_major(), mnr
<< MdpMinorShift
);
967 if (strncmp(devnm
, "md", 2) == 0 &&
969 (mnr
= strtoul(devnm
+2, &ep
, 10)) >= 0 &&
970 ep
> devnm
&& *ep
== 0)
971 return makedev(MD_MAJOR
, mnr
);
977 * is_devname_numbered() - helper for numbered devname verification.
978 * @devname: path or name to check.
979 * @pref: expected devname prefix.
980 * @pref_len: prefix len.
982 static bool is_devname_numbered(const char *devname
, const char *pref
, const int pref_len
)
986 assert(devname
&& pref
);
988 if (strncmp(devname
, pref
, pref_len
) != 0)
991 if (parse_num(&val
, devname
+ pref_len
) != 0)
1001 * is_devname_md_numbered() - check if &devname is numbered MD device (md).
1002 * @devname: path or name to check.
1004 bool is_devname_md_numbered(const char *devname
)
1006 return is_devname_numbered(devname
, DEV_NUM_PREF
, DEV_NUM_PREF_LEN
);
1010 * is_devname_md_d_numbered() - check if &devname is secondary numbered MD device (md_d).
1011 * @devname: path or name to check.
1013 bool is_devname_md_d_numbered(const char *devname
)
1015 static const char d_dev
[] = DEV_NUM_PREF
"_d";
1017 return is_devname_numbered(devname
, d_dev
, sizeof(d_dev
) - 1);
1021 * get_md_name() - Get main dev node of the md device.
1022 * @devnm: Md device name or path.
1024 * Function checks if the full name was passed and returns md name
1025 * if it is the MD device.
1027 * Return: Main dev node of the md device or NULL if not found.
1029 char *get_md_name(char *devnm
)
1031 static char devname
[NAME_MAX
];
1034 if (strncmp(devnm
, "/dev/", 5) == 0)
1035 snprintf(devname
, sizeof(devname
), "%s", devnm
);
1037 snprintf(devname
, sizeof(devname
), "/dev/%s", devnm
);
1039 if (!is_mddev(devname
))
1041 if (stat(devname
, &stb
) == 0 && (S_IFMT
&stb
.st_mode
) == S_IFBLK
)
1047 void put_md_name(char *name
)
1049 if (strncmp(name
, "/dev/.tmp.md", 12) == 0)
1053 int get_maj_min(char *dev
, int *major
, int *minor
)
1056 *major
= strtoul(dev
, &e
, 0);
1057 return (e
> dev
&& *e
== ':' && e
[1] &&
1058 (*minor
= strtoul(e
+1, &e
, 0)) >= 0 &&
1063 * is_bit_set() - get bit value by index.
1065 * @index: index of the bit (LSB numbering).
1067 * Return: bit value.
1069 bool is_bit_set(int *val
, unsigned char index
)
1071 if ((*val
) & (1 << index
))
1076 int dev_open(char *dev
, int flags
)
1078 /* like 'open', but if 'dev' matches %d:%d, create a temp
1079 * block device and open that
1090 if (get_maj_min(dev
, &major
, &minor
)) {
1091 snprintf(devname
, sizeof(devname
), "/dev/.tmp.md.%d:%d:%d",
1092 (int)getpid(), major
, minor
);
1093 if (mknod(devname
, S_IFBLK
|0600, makedev(major
, minor
)) == 0) {
1094 fd
= open(devname
, flags
);
1098 /* Try /tmp as /dev appear to be read-only */
1099 snprintf(devname
, sizeof(devname
),
1100 "/tmp/.tmp.md.%d:%d:%d",
1101 (int)getpid(), major
, minor
);
1102 if (mknod(devname
, S_IFBLK
|0600,
1103 makedev(major
, minor
)) == 0) {
1104 fd
= open(devname
, flags
);
1109 fd
= open(dev
, flags
);
1113 int open_dev_flags(char *devnm
, int flags
)
1118 devid
= devnm2devid(devnm
);
1119 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1120 return dev_open(buf
, flags
);
1123 int open_dev(char *devnm
)
1125 return open_dev_flags(devnm
, O_RDONLY
);
1128 int open_dev_excl(char *devnm
)
1133 dev_t devid
= devnm2devid(devnm
);
1134 unsigned int delay
= 1; // miliseconds
1136 sprintf(buf
, "%d:%d", major(devid
), minor(devid
));
1137 for (i
= 0; i
< 25; i
++) {
1138 int fd
= dev_open(buf
, flags
|O_EXCL
);
1141 if (errno
== EACCES
&& flags
== O_RDWR
) {
1147 sleep_for(0, MSEC_TO_NSEC(delay
), true);
1154 int same_dev(char *one
, char *two
)
1156 struct stat st1
, st2
;
1157 if (stat(one
, &st1
) != 0)
1159 if (stat(two
, &st2
) != 0)
1161 if ((st1
.st_mode
& S_IFMT
) != S_IFBLK
)
1163 if ((st2
.st_mode
& S_IFMT
) != S_IFBLK
)
1165 return st1
.st_rdev
== st2
.st_rdev
;
1168 void wait_for(char *dev
, int fd
)
1171 struct stat stb_want
;
1172 unsigned int delay
= 1; // miliseconds
1174 if (fstat(fd
, &stb_want
) != 0 ||
1175 (stb_want
.st_mode
& S_IFMT
) != S_IFBLK
)
1178 for (i
= 0; i
< 25; i
++) {
1180 if (stat(dev
, &stb
) == 0 &&
1181 (stb
.st_mode
& S_IFMT
) == S_IFBLK
&&
1182 (stb
.st_rdev
== stb_want
.st_rdev
))
1184 sleep_for(0, MSEC_TO_NSEC(delay
), true);
1189 pr_err("timeout waiting for %s\n", dev
);
1192 struct superswitch
*superlist
[] =
1195 &super_ddf
, &super_imsm
,
1200 struct supertype
*super_by_fd(int fd
, char **subarrayp
)
1202 mdu_array_info_t array
;
1205 struct supertype
*st
= NULL
;
1210 char *subarray
= NULL
;
1211 char container
[32] = "";
1214 devnm
= fd2devnm(fd
);
1218 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
1221 vers
= sra
->array
.major_version
;
1222 minor
= sra
->array
.minor_version
;
1223 verstr
= sra
->text_version
;
1225 if (md_get_array_info(fd
, &array
))
1226 array
.major_version
= array
.minor_version
= 0;
1227 vers
= array
.major_version
;
1228 minor
= array
.minor_version
;
1233 sprintf(version
, "%d.%d", vers
, minor
);
1236 if (minor
== -2 && is_subarray(verstr
)) {
1237 char *dev
= verstr
+1;
1239 subarray
= strchr(dev
, '/');
1242 subarray
= xstrdup(subarray
);
1244 strcpy(container
, dev
);
1246 sra
= sysfs_read(-1, container
, GET_VERSION
);
1247 if (sra
&& sra
->text_version
[0])
1248 verstr
= sra
->text_version
;
1250 verstr
= "-no-metadata-";
1253 for (i
= 0; st
== NULL
&& superlist
[i
]; i
++)
1254 st
= superlist
[i
]->match_metadata_desc(verstr
);
1260 *subarrayp
= subarray
;
1261 strcpy(st
->container_devnm
, container
);
1262 strncpy(st
->devnm
, devnm
, MD_NAME_MAX
- 1);
1269 struct supertype
*dup_super(struct supertype
*orig
)
1271 struct supertype
*st
;
1275 st
= xcalloc(1, sizeof(*st
));
1277 st
->max_devs
= orig
->max_devs
;
1278 st
->minor_version
= orig
->minor_version
;
1279 st
->ignore_hw_compat
= orig
->ignore_hw_compat
;
1280 st
->data_offset
= orig
->data_offset
;
1286 struct supertype
*guess_super_type(int fd
, enum guess_types guess_type
)
1288 /* try each load_super to find the best match,
1289 * and return the best superswitch
1291 struct superswitch
*ss
;
1292 struct supertype
*st
;
1293 unsigned int besttime
= 0;
1297 st
= xcalloc(1, sizeof(*st
));
1298 st
->container_devnm
[0] = 0;
1300 for (i
= 0; superlist
[i
]; i
++) {
1303 if (guess_type
== guess_array
&& ss
->add_to_super
== NULL
)
1305 if (guess_type
== guess_partitions
&& ss
->add_to_super
!= NULL
)
1307 memset(st
, 0, sizeof(*st
));
1308 st
->ignore_hw_compat
= 1;
1309 rv
= ss
->load_super(st
, fd
, NULL
);
1312 st
->ss
->getinfo_super(st
, &info
, NULL
);
1313 if (bestsuper
== -1 ||
1314 besttime
< info
.array
.ctime
) {
1316 besttime
= info
.array
.ctime
;
1321 if (bestsuper
!= -1) {
1323 memset(st
, 0, sizeof(*st
));
1324 st
->ignore_hw_compat
= 1;
1325 rv
= superlist
[bestsuper
]->load_super(st
, fd
, NULL
);
1327 superlist
[bestsuper
]->free_super(st
);
1335 /* Return size of device in bytes */
1336 int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
)
1338 unsigned long long ldsize
;
1341 if (fstat(fd
, &st
) != -1 && S_ISREG(st
.st_mode
))
1342 ldsize
= (unsigned long long)st
.st_size
;
1345 if (ioctl(fd
, BLKGETSIZE64
, &ldsize
) != 0)
1348 unsigned long dsize
;
1349 if (ioctl(fd
, BLKGETSIZE
, &dsize
) == 0) {
1354 pr_err("Cannot get size of %s: %s\n",
1355 dname
, strerror(errno
));
1363 /* Return sector size of device in bytes */
1364 int get_dev_sector_size(int fd
, char *dname
, unsigned int *sectsizep
)
1366 unsigned int sectsize
;
1368 if (ioctl(fd
, BLKSSZGET
, §size
) != 0) {
1370 pr_err("Cannot get sector size of %s: %s\n",
1371 dname
, strerror(errno
));
1375 *sectsizep
= sectsize
;
1379 /* Return true if this can only be a container, not a member device.
1380 * i.e. is and md device and size is zero
1382 int must_be_container(int fd
)
1385 unsigned long long size
;
1387 mdi
= sysfs_read(fd
, NULL
, GET_VERSION
);
1392 if (get_dev_size(fd
, NULL
, &size
) == 0)
1399 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1400 * Returns: 1 if successful
1401 * -1 for unknown partition type
1402 * 0 for other errors
1404 static int get_gpt_last_partition_end(int fd
, unsigned long long *endofpart
)
1407 unsigned char empty_gpt_entry
[16]= {0};
1408 struct GPT_part_entry
*part
;
1410 unsigned long long curr_part_end
;
1411 unsigned all_partitions
, entry_size
;
1413 unsigned int sector_size
= 0;
1417 BUILD_BUG_ON(sizeof(gpt
) != 512);
1418 /* skip protective MBR */
1419 if (!get_dev_sector_size(fd
, NULL
, §or_size
))
1421 lseek(fd
, sector_size
, SEEK_SET
);
1422 /* read GPT header */
1423 if (read(fd
, &gpt
, 512) != 512)
1426 /* get the number of partition entries and the entry size */
1427 all_partitions
= __le32_to_cpu(gpt
.part_cnt
);
1428 entry_size
= __le32_to_cpu(gpt
.part_size
);
1430 /* Check GPT signature*/
1431 if (gpt
.magic
!= GPT_SIGNATURE_MAGIC
)
1435 if (all_partitions
> 1024 ||
1436 entry_size
> sizeof(buf
))
1439 part
= (struct GPT_part_entry
*)buf
;
1441 /* set offset to third block (GPT entries) */
1442 lseek(fd
, sector_size
*2, SEEK_SET
);
1443 for (part_nr
= 0; part_nr
< all_partitions
; part_nr
++) {
1444 /* read partition entry */
1445 if (read(fd
, buf
, entry_size
) != (ssize_t
)entry_size
)
1448 /* is this valid partition? */
1449 if (memcmp(part
->type_guid
, empty_gpt_entry
, 16) != 0) {
1450 /* check the last lba for the current partition */
1451 curr_part_end
= __le64_to_cpu(part
->ending_lba
);
1452 if (curr_part_end
> *endofpart
)
1453 *endofpart
= curr_part_end
;
1460 /* Sets endofpart parameter to the last block used by the last partition on the device.
1461 * Returns: 1 if successful
1462 * -1 for unknown partition type
1463 * 0 for other errors
1465 static int get_last_partition_end(int fd
, unsigned long long *endofpart
)
1467 struct MBR boot_sect
;
1468 unsigned long long curr_part_end
;
1470 unsigned int sector_size
;
1475 BUILD_BUG_ON(sizeof(boot_sect
) != 512);
1478 if (read(fd
, &boot_sect
, 512) != 512)
1481 /* check MBP signature */
1482 if (boot_sect
.magic
== MBR_SIGNATURE_MAGIC
) {
1484 /* found the correct signature */
1486 for (part_nr
= 0; part_nr
< MBR_PARTITIONS
; part_nr
++) {
1488 * Have to make every access through boot_sect rather
1489 * than using a pointer to the partition table (or an
1490 * entry), since the entries are not properly aligned.
1493 /* check for GPT type */
1494 if (boot_sect
.parts
[part_nr
].part_type
==
1495 MBR_GPT_PARTITION_TYPE
) {
1496 retval
= get_gpt_last_partition_end(fd
, endofpart
);
1499 /* check the last used lba for the current partition */
1501 __le32_to_cpu(boot_sect
.parts
[part_nr
].first_sect_lba
) +
1502 __le32_to_cpu(boot_sect
.parts
[part_nr
].blocks_num
);
1503 if (curr_part_end
> *endofpart
)
1504 *endofpart
= curr_part_end
;
1507 /* Unknown partition table */
1510 /* calculate number of 512-byte blocks */
1511 if (get_dev_sector_size(fd
, NULL
, §or_size
))
1512 *endofpart
*= (sector_size
/ 512);
1517 int check_partitions(int fd
, char *dname
, unsigned long long freesize
,
1518 unsigned long long size
)
1521 * Check where the last partition ends
1523 unsigned long long endofpart
;
1525 if (get_last_partition_end(fd
, &endofpart
) > 0) {
1526 /* There appears to be a partition table here */
1527 if (freesize
== 0) {
1528 /* partitions will not be visible in new device */
1529 pr_err("partition table exists on %s but will be lost or\n"
1530 " meaningless after creating array\n",
1533 } else if (endofpart
> freesize
) {
1534 /* last partition overlaps metadata */
1535 pr_err("metadata will over-write last partition on %s.\n",
1538 } else if (size
&& endofpart
> size
) {
1539 /* partitions will be truncated in new device */
1540 pr_err("array size is too small to cover all partitions on %s.\n",
1548 int open_container(int fd
)
1550 /* 'fd' is a block device. Find out if it is in use
1551 * by a container, and return an open fd on that container.
1562 if (fstat(fd
, &st
) != 0)
1564 sprintf(path
, "/sys/dev/block/%d:%d/holders",
1565 (int)major(st
.st_rdev
), (int)minor(st
.st_rdev
));
1566 e
= path
+ strlen(path
);
1568 dir
= opendir(path
);
1571 while ((de
= readdir(dir
))) {
1574 if (de
->d_name
[0] == '.')
1576 /* Need to make sure it is a container and not a volume */
1577 sprintf(e
, "/%s/md/metadata_version", de
->d_name
);
1578 dfd
= open(path
, O_RDONLY
);
1581 n
= read(dfd
, buf
, sizeof(buf
));
1583 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1586 if (strncmp(buf
, "external", 8) != 0 ||
1590 sprintf(e
, "/%s/dev", de
->d_name
);
1591 dfd
= open(path
, O_RDONLY
);
1594 n
= read(dfd
, buf
, sizeof(buf
));
1596 if (n
<= 0 || (unsigned)n
>= sizeof(buf
))
1599 if (sscanf(buf
, "%d:%d", &major
, &minor
) != 2)
1601 sprintf(buf
, "%d:%d", major
, minor
);
1602 dfd
= dev_open(buf
, O_RDONLY
);
1612 struct superswitch
*version_to_superswitch(char *vers
)
1616 for (i
= 0; superlist
[i
]; i
++) {
1617 struct superswitch
*ss
= superlist
[i
];
1619 if (strcmp(vers
, ss
->name
) == 0)
1626 int metadata_container_matches(char *metadata
, char *devnm
)
1628 /* Check if 'devnm' is the container named in 'metadata'
1630 * /containername/componentname or
1631 * -containername/componentname
1634 if (*metadata
!= '/' && *metadata
!= '-')
1637 if (strncmp(metadata
+1, devnm
, l
) != 0)
1639 if (metadata
[l
+1] != '/')
1644 int metadata_subdev_matches(char *metadata
, char *devnm
)
1646 /* Check if 'devnm' is the subdev named in 'metadata'
1648 * /containername/subdev or
1649 * -containername/subdev
1652 if (*metadata
!= '/' && *metadata
!= '-')
1654 sl
= strchr(metadata
+1, '/');
1657 if (strcmp(sl
+1, devnm
) == 0)
1662 int is_container_member(struct mdstat_ent
*mdstat
, char *container
)
1664 if (mdstat
->metadata_version
== NULL
||
1665 strncmp(mdstat
->metadata_version
, "external:", 9) != 0 ||
1666 !metadata_container_matches(mdstat
->metadata_version
+9, container
))
1672 int is_subarray_active(char *subarray
, char *container
)
1674 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1675 struct mdstat_ent
*ent
;
1677 for (ent
= mdstat
; ent
; ent
= ent
->next
)
1678 if (is_container_member(ent
, container
))
1679 if (strcmp(to_subarray(ent
, container
), subarray
) == 0)
1682 free_mdstat(mdstat
);
1687 /* open_subarray - opens a subarray in a container
1688 * @dev: container device name
1689 * @st: empty supertype
1690 * @quiet: block reporting errors flag
1692 * On success returns an fd to a container and fills in *st
1694 int open_subarray(char *dev
, char *subarray
, struct supertype
*st
, int quiet
)
1697 struct mdinfo
*info
;
1701 fd
= open(dev
, O_RDWR
|O_EXCL
);
1704 pr_err("Couldn't open %s, aborting\n",
1709 _devnm
= fd2devnm(fd
);
1710 if (_devnm
== NULL
) {
1712 pr_err("Failed to determine device number for %s\n",
1716 strcpy(st
->devnm
, _devnm
);
1718 mdi
= sysfs_read(fd
, st
->devnm
, GET_VERSION
|GET_LEVEL
);
1721 pr_err("Failed to read sysfs for %s\n",
1726 if (mdi
->array
.level
!= UnSet
) {
1728 pr_err("%s is not a container\n", dev
);
1732 st
->ss
= version_to_superswitch(mdi
->text_version
);
1735 pr_err("Operation not supported for %s metadata\n",
1740 if (st
->devnm
[0] == 0) {
1742 pr_err("Failed to allocate device name\n");
1746 if (!st
->ss
->load_container
) {
1748 pr_err("%s is not a container\n", dev
);
1752 if (st
->ss
->load_container(st
, fd
, NULL
)) {
1754 pr_err("Failed to load metadata for %s\n",
1759 info
= st
->ss
->container_content(st
, subarray
);
1762 pr_err("Failed to find subarray-%s in %s\n",
1772 st
->ss
->free_super(st
);
1785 int add_disk(int mdfd
, struct supertype
*st
,
1786 struct mdinfo
*sra
, struct mdinfo
*info
)
1788 /* Add a device to an array, in one of 2 ways. */
1791 if (st
->ss
->external
) {
1792 if (info
->disk
.state
& (1<<MD_DISK_SYNC
))
1793 info
->recovery_start
= MaxSector
;
1795 info
->recovery_start
= 0;
1796 rv
= sysfs_add_disk(sra
, info
, 0);
1799 for (sd2
= sra
->devs
; sd2
; sd2
=sd2
->next
)
1803 sd2
= xmalloc(sizeof(*sd2
));
1805 sd2
->next
= sra
->devs
;
1810 rv
= ioctl(mdfd
, ADD_NEW_DISK
, &info
->disk
);
1814 int remove_disk(int mdfd
, struct supertype
*st
,
1815 struct mdinfo
*sra
, struct mdinfo
*info
)
1819 /* Remove the disk given by 'info' from the array */
1820 if (st
->ss
->external
)
1821 rv
= sysfs_set_str(sra
, info
, "slot", STR_COMMON_NONE
);
1823 rv
= ioctl(mdfd
, HOT_REMOVE_DISK
, makedev(info
->disk
.major
,
1828 int hot_remove_disk(int mdfd
, unsigned long dev
, int force
)
1830 int cnt
= force
? 500 : 5;
1833 /* HOT_REMOVE_DISK can fail with EBUSY if there are
1834 * outstanding IO requests to the device.
1835 * In this case, it can be helpful to wait a little while,
1836 * up to 5 seconds if 'force' is set, or 50 msec if not.
1838 while ((ret
= ioctl(mdfd
, HOT_REMOVE_DISK
, dev
)) == -1 &&
1841 sleep_for(0, MSEC_TO_NSEC(10), true);
1846 int sys_hot_remove_disk(int statefd
, int force
)
1848 int cnt
= force
? 500 : 5;
1851 while ((ret
= write(statefd
, "remove", 6)) == -1 &&
1854 sleep_for(0, MSEC_TO_NSEC(10), true);
1855 return ret
== 6 ? 0 : -1;
1858 int set_array_info(int mdfd
, struct supertype
*st
, struct mdinfo
*info
)
1860 /* Initialise kernel's knowledge of array.
1861 * This varies between externally managed arrays
1864 mdu_array_info_t inf
;
1867 if (st
->ss
->external
)
1868 return sysfs_set_array(info
, 9003);
1870 memset(&inf
, 0, sizeof(inf
));
1871 inf
.major_version
= info
->array
.major_version
;
1872 inf
.minor_version
= info
->array
.minor_version
;
1873 rv
= md_set_array_info(mdfd
, &inf
);
1878 unsigned long long min_recovery_start(struct mdinfo
*array
)
1880 /* find the minimum recovery_start in an array for metadata
1881 * formats that only record per-array recovery progress instead
1884 unsigned long long recovery_start
= MaxSector
;
1887 for (d
= array
->devs
; d
; d
= d
->next
)
1888 recovery_start
= min(recovery_start
, d
->recovery_start
);
1890 return recovery_start
;
1893 int mdmon_pid(char *devnm
)
1900 sprintf(path
, "%s/%s.pid", MDMON_DIR
, devnm
);
1902 fd
= open(path
, O_RDONLY
| O_NOATIME
, 0);
1906 n
= read(fd
, pid
, 9);
1913 int mdmon_running(char *devnm
)
1915 int pid
= mdmon_pid(devnm
);
1918 if (kill(pid
, 0) == 0)
1923 int start_mdmon(char *devnm
)
1929 char *prefix
= in_initrd() ? "initrd-" : "";
1938 if (check_env("MDADM_NO_MDMON"))
1940 if (continue_via_systemd(devnm
, MDMON_SERVICE
, prefix
))
1943 /* That failed, try running mdmon directly */
1944 len
= readlink("/proc/self/exe", pathbuf
, sizeof(pathbuf
)-1);
1948 sl
= strrchr(pathbuf
, '/');
1953 strcpy(sl
, "mdmon");
1960 for (i
= 0; paths
[i
]; i
++)
1962 execl(paths
[i
], paths
[i
],
1966 case -1: pr_err("cannot run mdmon. Array remains readonly\n");
1968 default: /* parent - good */
1969 pid
= wait(&status
);
1970 if (pid
< 0 || status
!= 0) {
1971 pr_err("failed to launch mdmon. Array remains readonly\n");
1978 __u32
random32(void)
1981 int rfd
= open("/dev/urandom", O_RDONLY
);
1982 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1989 void random_uuid(__u8
*buf
)
1994 fd
= open("/dev/urandom", O_RDONLY
);
1997 len
= read(fd
, buf
, 16);
2005 for (i
= 0; i
< 4; i
++)
2010 int flush_metadata_updates(struct supertype
*st
)
2014 st
->update_tail
= NULL
;
2018 sfd
= connect_monitor(st
->container_devnm
);
2022 while (st
->updates
) {
2023 struct metadata_update
*mu
= st
->updates
;
2024 st
->updates
= mu
->next
;
2026 send_message(sfd
, mu
, 0);
2034 st
->update_tail
= NULL
;
2038 void append_metadata_update(struct supertype
*st
, void *buf
, int len
)
2041 struct metadata_update
*mu
= xmalloc(sizeof(*mu
));
2046 mu
->space_list
= NULL
;
2048 *st
->update_tail
= mu
;
2049 st
->update_tail
= &mu
->next
;
2053 /* tinyc doesn't optimize this check in ioctl.h out ... */
2054 unsigned int __invalid_size_argument_for_IOC
= 0;
2058 * disk_fd_matches_criteria() - check if device matches spare criteria.
2059 * @disk_fd: file descriptor of the disk.
2060 * @sc: criteria to test.
2062 * Return: true if disk matches criteria, false otherwise.
2064 bool disk_fd_matches_criteria(int disk_fd
, struct spare_criteria
*sc
)
2066 unsigned int dev_sector_size
= 0;
2067 unsigned long long dev_size
= 0;
2069 if (!sc
->criteria_set
)
2072 if (!get_dev_size(disk_fd
, NULL
, &dev_size
) || dev_size
< sc
->min_size
)
2075 if (!get_dev_sector_size(disk_fd
, NULL
, &dev_sector_size
) ||
2076 sc
->sector_size
!= dev_sector_size
)
2083 * devid_matches_criteria() - check if device referenced by devid matches spare criteria.
2084 * @devid: devid of the device to check.
2085 * @sc: criteria to test.
2087 * Return: true if disk matches criteria, false otherwise.
2089 bool devid_matches_criteria(dev_t devid
, struct spare_criteria
*sc
)
2095 if (!sc
->criteria_set
)
2098 snprintf(buf
, NAME_MAX
, "%d:%d", major(devid
), minor(devid
));
2100 fd
= dev_open(buf
, O_RDONLY
);
2101 if (!is_fd_valid(fd
))
2104 /* Error code inherited */
2105 ret
= disk_fd_matches_criteria(fd
, sc
);
2111 /* Pick all spares matching given criteria from a container
2112 * if min_size == 0 do not check size
2113 * if domlist == NULL do not check domains
2114 * if spare_group given add it to domains of each spare
2115 * metadata allows to test domains using metadata of destination array */
2116 struct mdinfo
*container_choose_spares(struct supertype
*st
,
2117 struct spare_criteria
*criteria
,
2118 struct domainlist
*domlist
,
2120 const char *metadata
, int get_one
)
2122 struct mdinfo
*d
, **dp
, *disks
= NULL
;
2124 /* get list of all disks in container */
2125 if (st
->ss
->getinfo_super_disks
)
2126 disks
= st
->ss
->getinfo_super_disks(st
);
2130 /* find spare devices on the list */
2132 disks
->array
.spare_disks
= 0;
2137 if (d
->disk
.state
== 0) {
2138 dev_t dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
2140 found
= devid_matches_criteria(dev
, criteria
);
2142 /* check if domain matches */
2143 if (found
&& domlist
) {
2144 struct dev_policy
*pol
= devid_policy(dev
);
2146 pol_add(&pol
, pol_domain
,
2148 if (domain_test(domlist
, pol
, metadata
) != 1)
2151 dev_policy_free(pol
);
2156 disks
->array
.spare_disks
++;
2170 /* Checks if paths point to the same device
2171 * Returns 0 if they do.
2172 * Returns 1 if they don't.
2173 * Returns -1 if something went wrong,
2174 * e.g. paths are empty or the files
2175 * they point to don't exist */
2176 int compare_paths (char* path1
, char* path2
)
2178 struct stat st1
,st2
;
2180 if (path1
== NULL
|| path2
== NULL
)
2182 if (stat(path1
,&st1
) != 0)
2184 if (stat(path2
,&st2
) != 0)
2186 if ((st1
.st_ino
== st2
.st_ino
) && (st1
.st_dev
== st2
.st_dev
))
2191 /* Make sure we can open as many devices as needed */
2192 void enable_fds(int devices
)
2194 unsigned int fds
= 20 + devices
;
2196 if (getrlimit(RLIMIT_NOFILE
, &lim
) != 0 || lim
.rlim_cur
>= fds
)
2198 if (lim
.rlim_max
< fds
)
2201 setrlimit(RLIMIT_NOFILE
, &lim
);
2204 /* Close all opened descriptors if needed and redirect
2205 * streams to /dev/null.
2206 * For debug purposed, leave STDOUT and STDERR untouched
2208 * 1- if any error occurred
2211 void manage_fork_fds(int close_all
)
2214 struct dirent
*dirent
;
2217 open("/dev/null", O_RDWR
);
2227 dir
= opendir("/proc/self/fd");
2229 pr_err("Cannot open /proc/self/fd directory.\n");
2232 for (dirent
= readdir(dir
); dirent
; dirent
= readdir(dir
)) {
2235 if ((strcmp(dirent
->d_name
, ".") == 0) ||
2236 (strcmp(dirent
->d_name
, "..")) == 0)
2239 fd
= strtol(dirent
->d_name
, NULL
, 10);
2245 /* In a systemd/udev world, it is best to get systemd to
2246 * run daemon rather than running in the background.
2248 * 1- if systemd service has been started
2251 int continue_via_systemd(char *devnm
, char *service_name
, char *prefix
)
2256 /* Simply return that service cannot be started */
2257 if (check_env("MDADM_NO_SYSTEMCTL"))
2262 snprintf(pathbuf
, sizeof(pathbuf
),
2263 "%s@%s%s.service", service_name
, prefix
?: "", devnm
);
2264 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
2266 status
= execl("/bin/systemctl", "systemctl", "restart",
2269 case -1: /* Just do it ourselves. */
2271 default: /* parent - good */
2272 pid
= wait(&status
);
2273 if (pid
>= 0 && status
== 0)
2281 return access("/etc/initrd-release", F_OK
) >= 0;
2284 void reopen_mddev(int mdfd
)
2286 /* Re-open without any O_EXCL, but keep
2291 devnm
= fd2devnm(mdfd
);
2293 fd
= open_dev(devnm
);
2294 if (fd
>= 0 && fd
!= mdfd
)
2298 static struct cmap_hooks
*cmap_hooks
= NULL
;
2299 static int is_cmap_hooks_ready
= 0;
2301 void set_cmap_hooks(void)
2303 cmap_hooks
= xmalloc(sizeof(struct cmap_hooks
));
2304 cmap_hooks
->cmap_handle
= dlopen("libcmap.so.4", RTLD_NOW
| RTLD_LOCAL
);
2305 if (!cmap_hooks
->cmap_handle
)
2308 cmap_hooks
->initialize
=
2309 dlsym(cmap_hooks
->cmap_handle
, "cmap_initialize");
2310 cmap_hooks
->get_string
=
2311 dlsym(cmap_hooks
->cmap_handle
, "cmap_get_string");
2312 cmap_hooks
->finalize
= dlsym(cmap_hooks
->cmap_handle
, "cmap_finalize");
2314 if (!cmap_hooks
->initialize
|| !cmap_hooks
->get_string
||
2315 !cmap_hooks
->finalize
)
2316 dlclose(cmap_hooks
->cmap_handle
);
2318 is_cmap_hooks_ready
= 1;
2321 int get_cluster_name(char **cluster_name
)
2324 cmap_handle_t handle
;
2326 if (!is_cmap_hooks_ready
)
2329 rv
= cmap_hooks
->initialize(&handle
);
2333 rv
= cmap_hooks
->get_string(handle
, "totem.cluster_name", cluster_name
);
2335 free(*cluster_name
);
2342 cmap_hooks
->finalize(handle
);
2347 void set_dlm_hooks(void)
2349 dlm_hooks
= xmalloc(sizeof(struct dlm_hooks
));
2350 dlm_hooks
->dlm_handle
= dlopen("libdlm_lt.so.3", RTLD_NOW
| RTLD_LOCAL
);
2351 if (!dlm_hooks
->dlm_handle
)
2354 dlm_hooks
->open_lockspace
=
2355 dlsym(dlm_hooks
->dlm_handle
, "dlm_open_lockspace");
2356 dlm_hooks
->create_lockspace
=
2357 dlsym(dlm_hooks
->dlm_handle
, "dlm_create_lockspace");
2358 dlm_hooks
->release_lockspace
=
2359 dlsym(dlm_hooks
->dlm_handle
, "dlm_release_lockspace");
2360 dlm_hooks
->ls_lock
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_lock");
2361 dlm_hooks
->ls_unlock_wait
=
2362 dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_unlock_wait");
2363 dlm_hooks
->ls_get_fd
= dlsym(dlm_hooks
->dlm_handle
, "dlm_ls_get_fd");
2364 dlm_hooks
->dispatch
= dlsym(dlm_hooks
->dlm_handle
, "dlm_dispatch");
2366 if (!dlm_hooks
->open_lockspace
|| !dlm_hooks
->create_lockspace
||
2367 !dlm_hooks
->ls_lock
|| !dlm_hooks
->ls_unlock_wait
||
2368 !dlm_hooks
->release_lockspace
|| !dlm_hooks
->ls_get_fd
||
2369 !dlm_hooks
->dispatch
)
2370 dlclose(dlm_hooks
->dlm_handle
);
2372 is_dlm_hooks_ready
= 1;
2375 void set_hooks(void)
2381 int zero_disk_range(int fd
, unsigned long long sector
, size_t count
)
2387 size_t len
= count
* 512;
2390 fd_zero
= open("/dev/zero", O_RDONLY
);
2392 pr_err("Cannot open /dev/zero\n");
2396 if (lseek64(fd
, sector
* 512, SEEK_SET
) < 0) {
2398 pr_err("Failed to seek offset for zeroing\n");
2402 addr
= mmap(NULL
, len
, PROT_READ
, MAP_PRIVATE
, fd_zero
, 0);
2404 if (addr
== MAP_FAILED
) {
2406 pr_err("Mapping /dev/zero failed\n");
2411 n
= write(fd
, addr
+ written
, len
- written
);
2416 pr_err("Zeroing disk range failed\n");
2420 } while (written
!= len
);
2430 * sleep_for() - Sleeps for specified time.
2431 * @sec: Seconds to sleep for.
2432 * @nsec: Nanoseconds to sleep for, has to be less than one second.
2433 * @wake_after_interrupt: If set, wake up if interrupted.
2435 * Function immediately returns if error different than EINTR occurs.
2437 void sleep_for(unsigned int sec
, long nsec
, bool wake_after_interrupt
)
2439 struct timespec delay
= {.tv_sec
= sec
, .tv_nsec
= nsec
};
2441 assert(nsec
< MSEC_TO_NSEC(1000));
2445 nanosleep(&delay
, &delay
);
2446 if (errno
!= 0 && errno
!= EINTR
) {
2447 pr_err("Error sleeping for %us %ldns: %s\n", sec
, nsec
, strerror(errno
));
2450 } while (!wake_after_interrupt
&& errno
== EINTR
);
2453 /* is_directory() - Checks if directory provided by path is indeed a regular directory.
2454 * @path: directory path to be checked
2456 * Doesn't accept symlinks.
2458 * Return: true if is a directory, false if not
2460 bool is_directory(const char *path
)
2464 if (lstat(path
, &st
) != 0) {
2465 pr_err("%s: %s\n", strerror(errno
), path
);
2469 if (!S_ISDIR(st
.st_mode
))
2476 * is_file() - Checks if file provided by path is indeed a regular file.
2477 * @path: file path to be checked
2479 * Doesn't accept symlinks.
2481 * Return: true if is a file, false if not
2483 bool is_file(const char *path
)
2487 if (lstat(path
, &st
) != 0) {
2488 pr_err("%s: %s\n", strerror(errno
), path
);
2492 if (!S_ISREG(st
.st_mode
))