2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
31 #define _FILE_OFFSET_BITS 64
33 #if !defined(__dietlibc__) && !defined(__KLIBC__)
34 extern __off64_t lseek64
__P ((int __fd
, __off64_t __offset
, int __whence
));
36 # if defined(__NO_STAT64) || __WORDSIZE != 32
37 # define lseek64 lseek
41 #include <sys/types.h>
54 /* dietlibc has deprecated random and srandom!! */
60 #include <linux/kdev_t.h>
61 /*#include <linux/fs.h> */
62 #include <sys/mount.h>
63 #include <asm/types.h>
64 #include <sys/ioctl.h>
66 #define MdpMinorShift 6
69 #define BLKGETSIZE64 _IOR(0x12,114,size_t) /* return device size in bytes (u64 *arg) */
72 #define DEFAULT_BITMAP_CHUNK 4096
73 #define DEFAULT_BITMAP_DELAY 5
74 #define DEFAULT_MAX_WRITE_BEHIND 256
82 /* Redhat don't like to #include <asm/byteorder.h>, and
83 * some time include <linux/byteorder/xxx_endian.h> isn't enough,
84 * and there is no standard conversion function so... */
85 /* And dietlibc doesn't think byteswap is ok, so.. */
86 /* #include <byteswap.h> */
87 #define bswap_16(x) (((x) & 0x00ffU) << 8 | \
89 #define bswap_32(x) (((x) & 0x000000ffU) << 24 | \
90 ((x) & 0xff000000U) >> 24 | \
91 ((x) & 0x0000ff00U) << 8 | \
92 ((x) & 0x00ff0000U) >> 8)
93 #define bswap_64(x) (((x) & 0x00000000000000ffULL) << 56 | \
94 ((x) & 0xff00000000000000ULL) >> 56 | \
95 ((x) & 0x000000000000ff00ULL) << 40 | \
96 ((x) & 0x00ff000000000000ULL) >> 40 | \
97 ((x) & 0x0000000000ff0000ULL) << 24 | \
98 ((x) & 0x0000ff0000000000ULL) >> 24 | \
99 ((x) & 0x00000000ff000000ULL) << 8 | \
100 ((x) & 0x000000ff00000000ULL) >> 8)
102 #if !defined(__KLIBC__)
103 #if BYTE_ORDER == LITTLE_ENDIAN
104 #define __cpu_to_le16(_x) (_x)
105 #define __cpu_to_le32(_x) (_x)
106 #define __cpu_to_le64(_x) (_x)
107 #define __le16_to_cpu(_x) (_x)
108 #define __le32_to_cpu(_x) (_x)
109 #define __le64_to_cpu(_x) (_x)
111 #define __cpu_to_be16(_x) bswap_16(_x)
112 #define __cpu_to_be32(_x) bswap_32(_x)
113 #define __cpu_to_be64(_x) bswap_64(_x)
114 #define __be16_to_cpu(_x) bswap_16(_x)
115 #define __be32_to_cpu(_x) bswap_32(_x)
116 #define __be64_to_cpu(_x) bswap_64(_x)
117 #elif BYTE_ORDER == BIG_ENDIAN
118 #define __cpu_to_le16(_x) bswap_16(_x)
119 #define __cpu_to_le32(_x) bswap_32(_x)
120 #define __cpu_to_le64(_x) bswap_64(_x)
121 #define __le16_to_cpu(_x) bswap_16(_x)
122 #define __le32_to_cpu(_x) bswap_32(_x)
123 #define __le64_to_cpu(_x) bswap_64(_x)
125 #define __cpu_to_be16(_x) (_x)
126 #define __cpu_to_be32(_x) (_x)
127 #define __cpu_to_be64(_x) (_x)
128 #define __be16_to_cpu(_x) (_x)
129 #define __be32_to_cpu(_x) (_x)
130 #define __be64_to_cpu(_x) (_x)
132 # error "unknown endianness."
134 #endif /* __KLIBC__ */
138 /* general information that might be extracted from a superblock */
140 mdu_array_info_t array
;
141 mdu_disk_info_t disk
;
145 unsigned long long data_offset
;
146 unsigned long long component_size
; /* same as array.size, except in
147 * sectors and up to 64bits.
150 unsigned long long reshape_progress
;
151 unsigned long long resync_start
;
152 int new_level
, delta_disks
, new_layout
, new_chunk
;
154 int cache_size
; /* size of raid456 stripe cache*/
156 char text_version
[50];
158 int container_member
; /* for assembling external-metatdata arrays
159 * This is to be used internally by metadata
166 /* Device info for mdmon: */
170 #define DS_WRITE_MOSTLY 4
172 #define DS_BLOCKED 16
173 #define DS_REMOVE 1024
174 int prev_state
, curr_state
, next_state
;
184 struct supertype
*supertype
;
201 extern char short_options
[];
202 extern char short_bitmap_options
[];
203 extern char short_bitmap_auto_options
[];
204 extern struct option long_options
[];
205 extern char Version
[], Usage
[], Help
[], OptionHelp
[],
206 Help_create
[], Help_build
[], Help_assemble
[], Help_grow
[],
208 Help_manage
[], Help_misc
[], Help_monitor
[], Help_config
[];
210 /* for option that don't have short equivilents, we assign arbitrary
211 * small numbers. '1' means an undecorated option, so we start at '2'.
213 enum special_options
{
227 /* structures read from config file */
228 /* List of mddevice names and identifiers
229 * Identifiers can be:
231 * super-minor=decimal-minor-number-from-superblock
232 * devices=comma,separated,list,of,device,names,with,wildcards
234 * If multiple fields are present, the intersection of all matching
235 * devices is considered
237 #define UnSet (0xfffe)
238 typedef struct mddev_ident_s
{
245 unsigned int super_minor
;
247 char *devices
; /* comma separated list of device
248 * names with wild cards
251 unsigned int raid_disks
;
252 unsigned int spare_disks
;
253 struct supertype
*st
;
254 int autof
; /* 1 for normal, 2 for partitioned */
259 struct mddev_ident_s
*next
;
262 /* List of device names - wildcards expanded */
263 typedef struct mddev_dev_s
{
265 char disposition
; /* 'a' for add, 'r' for remove, 'f' for fail.
266 * Not set for names read from .config
270 char used
; /* set when used */
271 struct mddev_dev_s
*next
;
274 typedef struct mapping
{
285 char *pattern
; /* U or up, _ for down */
286 int percent
; /* -1 if no resync */
287 int resync
; /* 1 if resync, 0 if recovery */
291 char * metadata_version
;
292 struct mdstat_ent
*next
;
295 extern struct mdstat_ent
*mdstat_read(int hold
, int start
);
296 extern void free_mdstat(struct mdstat_ent
*ms
);
297 extern void mdstat_wait(int seconds
);
298 extern void mdstat_wait_fd(int fd
, const sigset_t
*sigmask
);
299 extern int mddev_busy(int devnum
);
302 struct map_ent
*next
;
308 extern int map_update(struct map_ent
**mpp
, int devnum
, int major
, int minor
,
309 int uuid
[4], char *path
);
310 extern struct map_ent
*map_by_uuid(struct map_ent
**map
, int uuid
[4]);
311 extern void map_read(struct map_ent
**melp
);
312 extern int map_write(struct map_ent
*mel
);
313 extern void map_delete(struct map_ent
**mapp
, int devnum
);
314 extern void map_free(struct map_ent
*map
);
315 extern void map_add(struct map_ent
**melp
,
316 int devnum
, int major
, int minor
, int uuid
[4], char *path
);
318 /* various details can be requested */
321 #define GET_COMPONENT 4
324 #define GET_MISMATCH 32
325 #define GET_VERSION 64
326 #define GET_DISKS 128
328 #define GET_DEVS 1024 /* gets role, major, minor */
329 #define GET_OFFSET 2048
330 #define GET_SIZE 4096
331 #define GET_STATE 8192
332 #define GET_ERROR 16384
334 /* If fd >= 0, get the array it is open on,
335 * else use devnum. >=0 -> major9. <0.....
337 extern int sysfs_open(int devnum
, char *devname
, char *attr
);
338 extern void sysfs_free(struct mdinfo
*sra
);
339 extern struct mdinfo
*sysfs_read(int fd
, int devnum
, unsigned long options
);
340 extern int sysfs_set_str(struct mdinfo
*sra
, struct mdinfo
*dev
,
341 char *name
, char *val
);
342 extern int sysfs_set_num(struct mdinfo
*sra
, struct mdinfo
*dev
,
343 char *name
, unsigned long long val
);
344 extern int sysfs_get_ll(struct mdinfo
*sra
, struct mdinfo
*dev
,
345 char *name
, unsigned long long *val
);
346 extern int sysfs_set_array(struct mdinfo
*sra
,
347 struct mdinfo
*info
);
348 extern int sysfs_add_disk(struct mdinfo
*sra
, struct mdinfo
*sd
);
353 extern int save_stripes(int *source
, unsigned long long *offsets
,
354 int raid_disks
, int chunk_size
, int level
, int layout
,
355 int nwrites
, int *dest
,
356 unsigned long long start
, unsigned long long length
);
357 extern int restore_stripes(int *dest
, unsigned long long *offsets
,
358 int raid_disks
, int chunk_size
, int level
, int layout
,
359 int source
, unsigned long long read_offset
,
360 unsigned long long start
, unsigned long long length
);
363 #define Sendmail "/usr/lib/sendmail -t"
366 #define SYSLOG_FACILITY LOG_DAEMON
368 extern char *map_num(mapping_t
*map
, int num
);
369 extern int map_name(mapping_t
*map
, char *name
);
370 extern mapping_t r5layout
[], pers
[], modes
[], faultylayout
[];
372 extern char *map_dev(int major
, int minor
, int create
);
375 struct metadata_update
;
377 extern struct superswitch
{
378 void (*examine_super
)(struct supertype
*st
, char *homehost
);
379 void (*brief_examine_super
)(struct supertype
*st
);
380 void (*export_examine_super
)(struct supertype
*st
);
381 void (*detail_super
)(struct supertype
*st
, char *homehost
);
382 void (*brief_detail_super
)(struct supertype
*st
);
383 void (*export_detail_super
)(struct supertype
*st
);
384 void (*uuid_from_super
)(struct supertype
*st
, int uuid
[4]);
385 void (*getinfo_super
)(struct supertype
*st
, struct mdinfo
*info
);
386 void (*getinfo_super_n
)(struct supertype
*st
, struct mdinfo
*info
);
387 int (*match_home
)(struct supertype
*st
, char *homehost
);
388 int (*update_super
)(struct supertype
*st
, struct mdinfo
*info
,
390 char *devname
, int verbose
,
391 int uuid_set
, char *homehost
);
392 int (*init_super
)(struct supertype
*st
, mdu_array_info_t
*info
,
393 unsigned long long size
, char *name
,
394 char *homehost
, int *uuid
);
395 void (*add_to_super
)(struct supertype
*st
, mdu_disk_info_t
*dinfo
,
396 int fd
, char *devname
);
397 int (*store_super
)(struct supertype
*st
, int fd
);
398 int (*write_init_super
)(struct supertype
*st
);
399 int (*compare_super
)(struct supertype
*st
, struct supertype
*tst
);
400 int (*load_super
)(struct supertype
*st
, int fd
, char *devname
);
401 struct supertype
* (*match_metadata_desc
)(char *arg
);
402 __u64 (*avail_size
)(struct supertype
*st
, __u64 size
);
403 int (*add_internal_bitmap
)(struct supertype
*st
, int *chunkp
,
404 int delay
, int write_behind
,
405 unsigned long long size
, int may_change
, int major
);
406 void (*locate_bitmap
)(struct supertype
*st
, int fd
);
407 int (*write_bitmap
)(struct supertype
*st
, int fd
);
408 void (*free_super
)(struct supertype
*st
);
409 int (*validate_geometry
)(struct supertype
*st
, int level
, int layout
,
411 int chunk
, unsigned long long size
,
412 char *subdev
, unsigned long long *freesize
);
414 struct mdinfo
*(*container_content
)(struct supertype
*st
);
417 int (*open_new
)(struct supertype
*c
, struct active_array
*a
,
420 /* Tell the metadata handler the current state of the array.
421 * This covers whether it is known to be consistent (no pending writes)
422 * when how far along a resync is known to have progressed
423 * (in a->resync_start).
424 * resync status is really irrelevant if the array is not consistent,
425 * but some metadata (DDF!) have a place to record the distinction.
427 void (*set_array_state
)(struct active_array
*a
, int consistent
);
429 /* When the state of a device might have changed, we call set_disk to
430 * tell the metadata what the current state is.
431 * Typically this happens on spare->in_sync and (spare|in_sync)->faulty
433 * set_disk might be called when the state of the particular disk has
434 * not in fact changed.
436 void (*set_disk
)(struct active_array
*a
, int n
, int state
);
437 void (*sync_metadata
)(struct supertype
*st
);
438 void (*process_update
)(struct supertype
*st
,
439 struct metadata_update
*update
);
441 /* activate_spare will check if the array is degraded and, if it
442 * is, try to find some spare space in the container.
443 * On success, it add appropriate updates (For process_update) to
444 * to the 'updates' list and returns a list of 'mdinfo' identifying
445 * the device, or devices as there might be multiple missing
446 * devices and multiple spares available.
448 struct mdinfo
*(*activate_spare
)(struct active_array
*a
,
449 struct metadata_update
**updates
);
452 int swapuuid
; /* true if uuid is bigending rather than hostendian */
454 } super0
, super1
, super_ddf
, super_ddf_bvd
, super_ddf_svd
, *superlist
[];
456 extern struct superswitch super_imsm
, super_imsm_raid
;
459 struct superswitch
*ss
;
462 int container_dev
; /* devnum of container */
466 /* extra stuff used by mdmon */
467 struct active_array
*arrays
;
468 int sock
; /* listen to external programs */
469 int mgr_pipe
[2]; /* communicate between threads */
470 int mon_pipe
[2]; /* communicate between threads */
472 char *devname
; /* e.g. md0. This appears in metadata_verison:
476 char *device_name
; /* e.g. /dev/md/whatever */
482 extern struct supertype supertype_container_member
;
483 extern struct supertype
*super_by_fd(int fd
);
484 extern struct supertype
*guess_super(int fd
);
485 extern struct supertype
*dup_super(struct supertype
*st
);
486 extern int get_dev_size(int fd
, char *dname
, unsigned long long *sizep
);
487 extern void get_one_disk(int mdfd
, mdu_array_info_t
*ainf
,
488 mdu_disk_info_t
*disk
);
494 #define HAVE_NFTW we assume
498 # include <features.h>
499 # ifndef __UCLIBC_HAS_FTW__
509 #if defined(__KLIBC__)
525 extern int add_dev(const char *name
, const struct stat
*stb
, int flag
, struct FTW
*s
);
528 extern int Manage_ro(char *devname
, int fd
, int readonly
);
529 extern int Manage_runstop(char *devname
, int fd
, int runstop
, int quiet
);
530 extern int Manage_resize(char *devname
, int fd
, long long size
, int raid_disks
);
531 extern int Manage_reconfig(char *devname
, int fd
, int layout
);
532 extern int Manage_subdevs(char *devname
, int fd
,
533 mddev_dev_t devlist
, int verbose
);
534 extern int autodetect(void);
535 extern int Grow_Add_device(char *devname
, int fd
, char *newdev
);
536 extern int Grow_addbitmap(char *devname
, int fd
, char *file
, int chunk
, int delay
, int write_behind
, int force
);
537 extern int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
539 int level
, int layout
, int chunksize
, int raid_disks
);
540 extern int Grow_restart(struct supertype
*st
, struct mdinfo
*info
,
541 int *fdlist
, int cnt
, char *backup_file
);
544 extern int Assemble(struct supertype
*st
, char *mddev
, int mdfd
,
546 mddev_dev_t devlist
, char *backup_file
,
547 int readonly
, int runstop
,
548 char *update
, char *homehost
,
549 int verbose
, int force
);
551 extern int Build(char *mddev
, int mdfd
, int chunk
, int level
, int layout
,
553 mddev_dev_t devlist
, int assume_clean
,
554 char *bitmap_file
, int bitmap_chunk
, int write_behind
, int delay
, int verbose
);
557 extern int Create(struct supertype
*st
, char *mddev
, int mdfd
,
558 int chunk
, int level
, int layout
, unsigned long long size
, int raiddisks
, int sparedisks
,
559 char *name
, char *homehost
, int *uuid
,
560 int subdevs
, mddev_dev_t devlist
,
561 int runstop
, int verbose
, int force
, int assume_clean
,
562 char *bitmap_file
, int bitmap_chunk
, int write_behind
, int delay
);
564 extern int Detail(char *dev
, int brief
, int export
, int test
, char *homehost
);
565 extern int Query(char *dev
);
566 extern int Examine(mddev_dev_t devlist
, int brief
, int export
, int scan
,
567 int SparcAdjust
, struct supertype
*forcest
, char *homehost
);
568 extern int Monitor(mddev_dev_t devlist
,
569 char *mailaddr
, char *alert_cmd
,
570 int period
, int daemonise
, int scan
, int oneshot
,
571 int dosyslog
, int test
, char *pidfile
);
573 extern int Kill(char *dev
, int force
, int quiet
, int noexcl
);
574 extern int Wait(char *dev
);
576 extern int Incremental(char *devname
, int verbose
, int runstop
,
577 struct supertype
*st
, char *homehost
, int autof
);
578 extern int Incremental_container(struct supertype
*st
, char *devname
,
579 int verbose
, int runstop
, int autof
);
580 extern void RebuildMap(void);
581 extern int IncrementalScan(int verbose
);
583 extern int CreateBitmap(char *filename
, int force
, char uuid
[16],
584 unsigned long chunksize
, unsigned long daemon_sleep
,
585 unsigned long write_behind
,
586 unsigned long long array_size
,
588 extern int ExamineBitmap(char *filename
, int brief
, struct supertype
*st
);
589 extern int bitmap_update_uuid(int fd
, int *uuid
, int swap
);
591 extern int md_get_version(int fd
);
592 extern int get_linux_version(void);
593 extern int parse_uuid(char *str
, int uuid
[4]);
594 extern int check_ext2(int fd
, char *name
);
595 extern int check_reiser(int fd
, char *name
);
596 extern int check_raid(int fd
, char *name
);
598 extern int get_mdp_major(void);
599 extern int dev_open(char *dev
, int flags
);
600 extern int open_dev_excl(int devnum
);
601 extern int is_standard(char *dev
, int *nump
);
603 extern int parse_auto(char *str
, char *msg
, int config
);
604 extern mddev_ident_t
conf_get_ident(char *dev
);
605 extern mddev_dev_t
conf_get_devs(void);
606 extern int conf_test_dev(char *devname
);
607 extern struct createinfo
*conf_get_create_info(void);
608 extern void set_conffile(char *file
);
609 extern char *conf_get_mailaddr(void);
610 extern char *conf_get_mailfrom(void);
611 extern char *conf_get_program(void);
612 extern char *conf_get_homehost(void);
613 extern char *conf_line(FILE *file
);
614 extern char *conf_word(FILE *file
, int allow_key
);
615 extern void free_line(char *line
);
616 extern int match_oneof(char *devices
, char *devname
);
617 extern void uuid_from_super(int uuid
[4], mdp_super_t
*super
);
618 extern int same_uuid(int a
[4], int b
[4], int swapuuid
);
619 extern void copy_uuid(void *a
, int b
[4], int swapuuid
);
620 extern unsigned long calc_csum(void *super
, int bytes
);
621 extern int enough(int level
, int raid_disks
, int layout
, int clean
,
622 char *avail
, int avail_disks
);
623 extern int ask(char *mesg
);
624 extern unsigned long long get_component_size(int fd
);
625 extern void remove_partitions(int fd
);
626 extern unsigned long long calc_array_size(int level
, int raid_disks
, int layout
,
627 int chunksize
, unsigned long long devsize
);
630 extern char *human_size(long long bytes
);
631 char *human_size_brief(long long bytes
);
633 #define NoMdDev (1<<23)
634 extern int find_free_devnum(int use_partitions
);
636 extern void put_md_name(char *name
);
637 extern char *get_md_name(int dev
);
639 extern char DefaultConfFile
[];
641 extern int open_mddev(char *dev
, int autof
);
642 extern int open_mddev_devnum(char *devname
, int devnum
, char *name
,
643 char *chosen_name
, int parts
);
644 extern int open_container(int fd
);
646 extern int mdmon_running(int devnum
);
647 extern int signal_mdmon(int devnum
);
649 extern char *devnum2devname(int num
);
650 extern int fd2devnum(int fd
);
652 static inline int dev2major(int d
)
657 return get_mdp_major();
660 static inline int dev2minor(int d
)
664 return (-1-d
) << MdpMinorShift
;
667 #define LEVEL_MULTIPATH (-4)
668 #define LEVEL_LINEAR (-1)
669 #define LEVEL_FAULTY (-5)
671 /* kernel module doesn't know about these */
672 #define LEVEL_CONTAINER (-100)
673 #define LEVEL_UNSUPPORTED (-200)
678 #define WriteTransient 0
679 #define ReadTransient 1
680 #define WritePersistent 2
681 #define ReadPersistent 3
682 #define WriteAll 4 /* doesn't go to device */
683 #define ReadFixable 5
686 #define ClearErrors 31
687 #define ClearFaults 30
689 #define AllPersist 100 /* internal use only */
690 #define NoPersist 101
692 #define ModeMask 0x1f
700 #define minor(x) ((x)&0xff)
701 #define major(x) (((x)>>8)&0xff)
702 #define makedev(M,m) (((M)<<8) | (m))
706 #define ALGORITHM_LEFT_ASYMMETRIC 0
707 #define ALGORITHM_RIGHT_ASYMMETRIC 1
708 #define ALGORITHM_LEFT_SYMMETRIC 2
709 #define ALGORITHM_RIGHT_SYMMETRIC 3