mdadm.h

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
   5  *
   6  *
   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.
  11  *
  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.
  16  *
  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
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@suse.de>
  23  */
  24
  25 #define _GNU_SOURCE
  26 #define _FILE_OFFSET_BITS 64
  27 #include        <unistd.h>
  28 #if !defined(__dietlibc__) && !defined(__KLIBC__)
  29 extern __off64_t lseek64 __P ((int __fd, __off64_t __offset, int __whence));
  30 #else
  31 # if defined(__NO_STAT64) || __WORDSIZE != 32
  32 # define lseek64 lseek
  33 # endif
  34 #endif
  35
  36 #include        <sys/types.h>
  37 #include        <sys/stat.h>
  38 #include        <stdlib.h>
  39 #include        <time.h>
  40 #include        <sys/time.h>
  41 #include        <getopt.h>
  42 #include        <fcntl.h>
  43 #include        <stdio.h>
  44 #include        <errno.h>
  45 #include        <string.h>
  46 #include        <syslog.h>
  47 #ifdef __dietlibc__
  48 #include        <strings.h>
  49 /* dietlibc has deprecated random and srandom!! */
  50 #define random rand
  51 #define srandom srand
  52 #endif
  53
  54
  55 #include        <linux/kdev_t.h>
  56 /*#include      <linux/fs.h> */
  57 #include        <sys/mount.h>
  58 #include        <asm/types.h>
  59 #include        <sys/ioctl.h>
  60 #define MD_MAJOR 9
  61 #define MdpMinorShift 6
  62
  63 #ifndef BLKGETSIZE64
  64 #define BLKGETSIZE64 _IOR(0x12,114,size_t) /* return device size in bytes (u64 *arg) */
  65 #endif
  66
  67 #define DEFAULT_BITMAP_CHUNK 4096
  68 #define DEFAULT_BITMAP_DELAY 5
  69 #define DEFAULT_MAX_WRITE_BEHIND 256
  70
  71 #define VAR_RUN "/var/run/mdadm"
  72 /* ALT_RUN should be somewhere that persists across the pivotroot
  73  * from early boot to late boot.
  74  * If you don't have /lib/init/rw you might want to use /dev/.something
  75  */
  76 #ifndef ALT_RUN
  77 #define ALT_RUN "/lib/init/rw/mdadm"
  78 #endif /* ALT_RUN */
  79
  80 #include        "md_u.h"
  81 #include        "md_p.h"
  82 #include        "bitmap.h"
  83 #include        "msg.h"
  84
  85 #include <endian.h>
  86 /* Redhat don't like to #include <asm/byteorder.h>, and
  87  * some time include <linux/byteorder/xxx_endian.h> isn't enough,
  88  * and there is no standard conversion function so... */
  89 /* And dietlibc doesn't think byteswap is ok, so.. */
  90 /*  #include <byteswap.h> */
  91 #define bswap_16(x) (((x) & 0x00ffU) << 8 | \
  92                      ((x) & 0xff00U) >> 8)
  93 #define bswap_32(x) (((x) & 0x000000ffU) << 24 | \
  94                      ((x) & 0xff000000U) >> 24 | \
  95                      ((x) & 0x0000ff00U) << 8  | \
  96                      ((x) & 0x00ff0000U) >> 8)
  97 #define bswap_64(x) (((x) & 0x00000000000000ffULL) << 56 | \
  98                      ((x) & 0xff00000000000000ULL) >> 56 | \
  99                      ((x) & 0x000000000000ff00ULL) << 40 | \
 100                      ((x) & 0x00ff000000000000ULL) >> 40 | \
 101                      ((x) & 0x0000000000ff0000ULL) << 24 | \
 102                      ((x) & 0x0000ff0000000000ULL) >> 24 | \
 103                      ((x) & 0x00000000ff000000ULL) << 8 | \
 104                      ((x) & 0x000000ff00000000ULL) >> 8)
 105
 106 #if !defined(__KLIBC__)
 107 #if BYTE_ORDER == LITTLE_ENDIAN
 108 #define __cpu_to_le16(_x) (_x)
 109 #define __cpu_to_le32(_x) (_x)
 110 #define __cpu_to_le64(_x) (_x)
 111 #define __le16_to_cpu(_x) (_x)
 112 #define __le32_to_cpu(_x) (_x)
 113 #define __le64_to_cpu(_x) (_x)
 114
 115 #define __cpu_to_be16(_x) bswap_16(_x)
 116 #define __cpu_to_be32(_x) bswap_32(_x)
 117 #define __cpu_to_be64(_x) bswap_64(_x)
 118 #define __be16_to_cpu(_x) bswap_16(_x)
 119 #define __be32_to_cpu(_x) bswap_32(_x)
 120 #define __be64_to_cpu(_x) bswap_64(_x)
 121 #elif BYTE_ORDER == BIG_ENDIAN
 122 #define __cpu_to_le16(_x) bswap_16(_x)
 123 #define __cpu_to_le32(_x) bswap_32(_x)
 124 #define __cpu_to_le64(_x) bswap_64(_x)
 125 #define __le16_to_cpu(_x) bswap_16(_x)
 126 #define __le32_to_cpu(_x) bswap_32(_x)
 127 #define __le64_to_cpu(_x) bswap_64(_x)
 128
 129 #define __cpu_to_be16(_x) (_x)
 130 #define __cpu_to_be32(_x) (_x)
 131 #define __cpu_to_be64(_x) (_x)
 132 #define __be16_to_cpu(_x) (_x)
 133 #define __be32_to_cpu(_x) (_x)
 134 #define __be64_to_cpu(_x) (_x)
 135 #else
 136 #  error "unknown endianness."
 137 #endif
 138 #endif /* __KLIBC__ */
 139
 140
 141 /*
 142  * min()/max()/clamp() macros that also do
 143  * strict type-checking.. See the
 144  * "unnecessary" pointer comparison.
 145  */
 146 #define min(x, y) ({                            \
 147         typeof(x) _min1 = (x);                  \
 148         typeof(y) _min2 = (y);                  \
 149         (void) (&_min1 == &_min2);              \
 150         _min1 < _min2 ? _min1 : _min2; })
 151
 152 #define max(x, y) ({                            \
 153         typeof(x) _max1 = (x);                  \
 154         typeof(y) _max2 = (y);                  \
 155         (void) (&_max1 == &_max2);              \
 156         _max1 > _max2 ? _max1 : _max2; })
 157
 158 /* general information that might be extracted from a superblock */
 159 struct mdinfo {
 160         mdu_array_info_t        array;
 161         mdu_disk_info_t         disk;
 162         __u64                   events;
 163         int                     uuid[4];
 164         char                    name[33];
 165         unsigned long long      data_offset;
 166         unsigned long long      component_size; /* same as array.size, except in
 167                                                  * sectors and up to 64bits.
 168                                                  */
 169         unsigned long long      custom_array_size; /* size for non-default sized
 170                                                     * arrays (in sectors)
 171                                                     */
 172         int                     reshape_active;
 173         unsigned long long      reshape_progress;
 174         union {
 175                 unsigned long long resync_start; /* per-array resync position */
 176                 unsigned long long recovery_start; /* per-device rebuild position */
 177                 #define MaxSector  (~0ULL) /* resync/recovery complete position */
 178         };
 179         unsigned long           safe_mode_delay; /* ms delay to mark clean */
 180         int                     new_level, delta_disks, new_layout, new_chunk;
 181         int                     errors;
 182         int                     cache_size; /* size of raid456 stripe cache*/
 183         int                     mismatch_cnt;
 184         char                    text_version[50];
 185         void                    *update_private; /* for passing metadata-format
 186                                                   * specific update data
 187                                                   * between successive calls to
 188                                                   * update_super()
 189                                                   */
 190
 191         int container_member; /* for assembling external-metatdata arrays
 192                                * This is to be used internally by metadata
 193                                * handler only */
 194
 195         char            sys_name[20];
 196         struct mdinfo *devs;
 197         struct mdinfo *next;
 198
 199         /* Device info for mdmon: */
 200         int recovery_fd;
 201         int state_fd;
 202         #define DS_FAULTY       1
 203         #define DS_INSYNC       2
 204         #define DS_WRITE_MOSTLY 4
 205         #define DS_SPARE        8
 206         #define DS_BLOCKED      16
 207         #define DS_REMOVE       1024
 208         #define DS_UNBLOCK      2048
 209         int prev_state, curr_state, next_state;
 210
 211 };
 212
 213 struct createinfo {
 214         int     uid;
 215         int     gid;
 216         int     autof;
 217         int     mode;
 218         int     symlinks;
 219         struct supertype *supertype;
 220 };
 221
 222 #define Name "mdadm"
 223
 224 enum mode {
 225         ASSEMBLE=1,
 226         BUILD,
 227         CREATE,
 228         MANAGE,
 229         MISC,
 230         MONITOR,
 231         GROW,
 232         INCREMENTAL,
 233         AUTODETECT,
 234 };
 235
 236 extern char short_options[];
 237 extern char short_bitmap_options[];
 238 extern char short_bitmap_auto_options[];
 239 extern struct option long_options[];
 240 extern char Version[], Usage[], Help[], OptionHelp[],
 241         Help_create[], Help_build[], Help_assemble[], Help_grow[],
 242         Help_incr[],
 243         Help_manage[], Help_misc[], Help_monitor[], Help_config[];
 244
 245 /* for option that don't have short equivilents, we assign arbitrary
 246  * small numbers.  '1' means an undecorated option, so we start at '2'.
 247  */
 248 enum special_options {
 249         AssumeClean = 2,
 250         BitmapChunk,
 251         WriteBehind,
 252         ReAdd,
 253         NoDegraded,
 254         Sparc22,
 255         BackupFile,
 256         HomeHost,
 257         AutoHomeHost,
 258         Symlinks,
 259         AutoDetect,
 260         Waitclean,
 261         DetailPlatform,
 262 };
 263
 264 /* structures read from config file */
 265 /* List of mddevice names and identifiers
 266  * Identifiers can be:
 267  *    uuid=128-hex-uuid
 268  *    super-minor=decimal-minor-number-from-superblock
 269  *    devices=comma,separated,list,of,device,names,with,wildcards
 270  *
 271  * If multiple fields are present, the intersection of all matching
 272  * devices is considered
 273  */
 274 #define UnSet (0xfffe)
 275 typedef struct mddev_ident_s {
 276         char    *devname;
 277
 278         int     uuid_set;
 279         int     uuid[4];
 280         char    name[33];
 281
 282         unsigned int super_minor;
 283
 284         char    *devices;       /* comma separated list of device
 285                                  * names with wild cards
 286                                  */
 287         int     level;
 288         unsigned int raid_disks;
 289         unsigned int spare_disks;
 290         struct supertype *st;
 291         int     autof;          /* 1 for normal, 2 for partitioned */
 292         char    *spare_group;
 293         char    *bitmap_file;
 294         int     bitmap_fd;
 295
 296         char    *container;     /* /dev/whatever name of container, or
 297                                  * uuid of container.  You would expect
 298                                  * this to be the 'devname' or UUID
 299                                  * of some other entry.
 300                                  */
 301         char    *member;        /* subarray within a container */
 302
 303         struct mddev_ident_s *next;
 304         union {
 305                 /* fields needed by different users of this structure */
 306                 int assembled;  /* set when assembly succeeds */
 307         };
 308 } *mddev_ident_t;
 309
 310 /* List of device names - wildcards expanded */
 311 typedef struct mddev_dev_s {
 312         char *devname;
 313         char disposition;       /* 'a' for add, 'r' for remove, 'f' for fail.
 314                                  * Not set for names read from .config
 315                                  */
 316         char writemostly;       /* 1 for 'set writemostly', 2 for 'clear writemostly' */
 317         char re_add;
 318         char used;              /* set when used */
 319         struct mdinfo *content; /* If devname is a container, this might list
 320                                  * the remaining member arrays. */
 321         struct mddev_dev_s *next;
 322 } *mddev_dev_t;
 323
 324 typedef struct mapping {
 325         char *name;
 326         int num;
 327 } mapping_t;
 328
 329
 330 struct mdstat_ent {
 331         char            *dev;
 332         int             devnum;
 333         int             active;
 334         char            *level;
 335         char            *pattern; /* U or up, _ for down */
 336         int             percent; /* -1 if no resync */
 337         int             resync; /* 1 if resync, 0 if recovery */
 338         int             devcnt;
 339         int             raid_disks;
 340         int             chunk_size;
 341         char *          metadata_version;
 342         struct mdstat_ent *next;
 343 };
 344
 345 extern struct mdstat_ent *mdstat_read(int hold, int start);
 346 extern void free_mdstat(struct mdstat_ent *ms);
 347 extern void mdstat_wait(int seconds);
 348 extern void mdstat_wait_fd(int fd, const sigset_t *sigmask);
 349 extern int mddev_busy(int devnum);
 350
 351 struct map_ent {
 352         struct map_ent *next;
 353         int     devnum;
 354         char    metadata[20];
 355         int     uuid[4];
 356         int     bad;
 357         char    *path;
 358 };
 359 extern int map_update(struct map_ent **mpp, int devnum, char *metadata,
 360                       int uuid[4], char *path);
 361 extern struct map_ent *map_by_uuid(struct map_ent **map, int uuid[4]);
 362 extern struct map_ent *map_by_devnum(struct map_ent **map, int devnum);
 363 extern struct map_ent *map_by_name(struct map_ent **map, char *name);
 364 extern void map_read(struct map_ent **melp);
 365 extern int map_write(struct map_ent *mel);
 366 extern void map_delete(struct map_ent **mapp, int devnum);
 367 extern void map_free(struct map_ent *map);
 368 extern void map_add(struct map_ent **melp,
 369                     int devnum, char *metadata, int uuid[4], char *path);
 370 extern int map_lock(struct map_ent **melp);
 371 extern void map_unlock(struct map_ent **melp);
 372
 373 /* various details can be requested */
 374 enum sysfs_read_flags {
 375         GET_LEVEL       = (1 << 0),
 376         GET_LAYOUT      = (1 << 1),
 377         GET_COMPONENT   = (1 << 2),
 378         GET_CHUNK       = (1 << 3),
 379         GET_CACHE       = (1 << 4),
 380         GET_MISMATCH    = (1 << 5),
 381         GET_VERSION     = (1 << 6),
 382         GET_DISKS       = (1 << 7),
 383         GET_DEGRADED    = (1 << 8),
 384         GET_SAFEMODE    = (1 << 9),
 385         GET_DEVS        = (1 << 10), /* gets role, major, minor */
 386         GET_OFFSET      = (1 << 11),
 387         GET_SIZE        = (1 << 12),
 388         GET_STATE       = (1 << 13),
 389         GET_ERROR       = (1 << 14),
 390         SKIP_GONE_DEVS  = (1 << 15),
 391 };
 392
 393 /* If fd >= 0, get the array it is open on,
 394  * else use devnum. >=0 -> major9. <0.....
 395  */
 396 extern int sysfs_open(int devnum, char *devname, char *attr);
 397 extern void sysfs_init(struct mdinfo *mdi, int fd, int devnum);
 398 extern void sysfs_free(struct mdinfo *sra);
 399 extern struct mdinfo *sysfs_read(int fd, int devnum, unsigned long options);
 400 extern int sysfs_attr_match(const char *attr, const char *str);
 401 extern int sysfs_match_word(const char *word, char **list);
 402 extern int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 403                          char *name, char *val);
 404 extern int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 405                          char *name, unsigned long long val);
 406 extern int sysfs_uevent(struct mdinfo *sra, char *event);
 407 extern int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 408                         char *name);
 409 extern int sysfs_fd_get_ll(int fd, unsigned long long *val);
 410 extern int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 411                         char *name, unsigned long long *val);
 412 extern int sysfs_fd_get_str(int fd, char *val, int size);
 413 extern int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 414                          char *name, char *val, int size);
 415 extern int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms);
 416 extern int sysfs_set_array(struct mdinfo *info, int vers);
 417 extern int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume);
 418 extern int sysfs_disk_to_scsi_id(int fd, __u32 *id);
 419 extern int sysfs_unique_holder(int devnum, long rdev);
 420 extern int load_sys(char *path, char *buf);
 421
 422
 423 extern int save_stripes(int *source, unsigned long long *offsets,
 424                         int raid_disks, int chunk_size, int level, int layout,
 425                         int nwrites, int *dest,
 426                         unsigned long long start, unsigned long long length,
 427                         char *buf);
 428 extern int restore_stripes(int *dest, unsigned long long *offsets,
 429                            int raid_disks, int chunk_size, int level, int layout,
 430                            int source, unsigned long long read_offset,
 431                            unsigned long long start, unsigned long long length);
 432
 433 #ifndef Sendmail
 434 #define Sendmail "/usr/lib/sendmail -t"
 435 #endif
 436
 437 #define SYSLOG_FACILITY LOG_DAEMON
 438
 439 extern char *map_num(mapping_t *map, int num);
 440 extern int map_name(mapping_t *map, char *name);
 441 extern mapping_t r5layout[], r6layout[], pers[], modes[], faultylayout[];
 442
 443 extern char *map_dev(int major, int minor, int create);
 444
 445 struct active_array;
 446 struct metadata_update;
 447
 448 /* A superswitch provides entry point the a metadata handler.
 449  *
 450  * The super_switch primarily operates on some "metadata" that
 451  * is accessed via the 'supertype'.
 452  * This metadata has one of three possible sources.
 453  * 1/ It is read from a single device.  In this case it may not completely
 454  *    describe the array or arrays as some information might be on other
 455  *    devices.
 456  * 2/ It is read from all devices in a container.  In this case all
 457  *    information is present.
 458  * 3/ It is created by ->init_super / ->add_to_super.  In this case it will
 459  *    be complete once enough ->add_to_super calls have completed.
 460  *
 461  * When creating an array inside a container, the metadata will be
 462  * formed by a combination of 2 and 3.  The metadata or the array is read,
 463  * then new information is added.
 464  *
 465  * The metadata must sometimes have a concept of a 'current' array
 466  * and a 'current' device.
 467  * The 'current' array is set by init_super to be the newly created array,
 468  * or is set by super_by_fd when it finds it is looking at an array inside
 469  * a container.
 470  *
 471  * The 'current' device is either the device that the metadata was read from
 472  * in case 1, or the last device added by add_to_super in case 3.
 473  * Case 2 does not identify a 'current' device.
 474  */
 475 extern struct superswitch {
 476
 477         /* Used to report details of metadata read from a component
 478          * device. ->load_super has been called.
 479          */
 480         void (*examine_super)(struct supertype *st, char *homehost);
 481         void (*brief_examine_super)(struct supertype *st, int verbose);
 482         void (*brief_examine_subarrays)(struct supertype *st, int verbose);
 483         void (*export_examine_super)(struct supertype *st);
 484
 485         /* Used to report details of an active array.
 486          * ->load_super was possibly given a 'component' string.
 487          */
 488         void (*detail_super)(struct supertype *st, char *homehost);
 489         void (*brief_detail_super)(struct supertype *st);
 490         void (*export_detail_super)(struct supertype *st);
 491
 492         /* Optional: platform hardware / firmware details */
 493         int (*detail_platform)(int verbose, int enumerate_only);
 494
 495         /* Used:
 496          *   to get uuid to storing in bitmap metadata
 497          *   and 'reshape' backup-data metadata
 498          *   To see if a device is being re-added to an array it was part of.
 499          */
 500         void (*uuid_from_super)(struct supertype *st, int uuid[4]);
 501
 502         /* Extract generic details from metadata.  This could be details about
 503          * the container, or about an individual array within the container.
 504          * The determination is made either by:
 505          *   load_super being given a 'component' string.
 506          *   validate_geometry determining what to create.
 507          * The info includes both array information and device information.
 508          * The particular device should be:
 509          *   The last device added by add_to_super
 510          *   The device the metadata was loaded from by load_super
 511          */
 512         void (*getinfo_super)(struct supertype *st, struct mdinfo *info);
 513
 514         /* Check if the given metadata is flagged as belonging to "this"
 515          * host.  0 for 'no', 1 for 'yes', -1 for "Don't record homehost"
 516          */
 517         int (*match_home)(struct supertype *st, char *homehost);
 518
 519         /* Make one of several generic modifications to metadata
 520          * prior to assembly (or other times).
 521          *   sparc2.2  - first bug in early 0.90 metadata
 522          *   super-minor - change name of 0.90 metadata
 523          *   summaries - 'correct' any redundant data
 524          *   resync - mark array as dirty to trigger a resync.
 525          *   uuid - set new uuid - only 0.90 or 1.x
 526          *   name - change the name of the array (where supported)
 527          *   homehost - change which host this array is tied to.
 528          *   devicesize - If metadata is at start of device, change recorded
 529          *               device size to match actual device size
 530          *   byteorder - swap bytes for 0.90 metadata
 531          *
 532          *   force-one  - mark that device as uptodate, not old or failed.
 533          *   force-array - mark array as clean if it would not otherwise
 534          *               assemble
 535          *   assemble   - not sure how this is different from force-one...
 536          *   linear-grow-new - add a new device to a linear array, but don't
 537          *                   change the size: so superblock still matches
 538          *   linear-grow-update - now change the size of the array.
 539          */
 540         int (*update_super)(struct supertype *st, struct mdinfo *info,
 541                             char *update,
 542                             char *devname, int verbose,
 543                             int uuid_set, char *homehost);
 544
 545         /* Create new metadata for new array as described.  This could
 546          * be a new container, or an array in a pre-existing container.
 547          * Also used to zero metadata prior to writing it to invalidate old
 548          * metadata.
 549          */
 550         int (*init_super)(struct supertype *st, mdu_array_info_t *info,
 551                           unsigned long long size, char *name,
 552                           char *homehost, int *uuid);
 553
 554         /* update the metadata to include new device, either at create or
 555          * when hot-adding a spare.
 556          */
 557         int (*add_to_super)(struct supertype *st, mdu_disk_info_t *dinfo,
 558                              int fd, char *devname);
 559
 560         /* Write metadata to one device when fixing problems or adding
 561          * a new device.
 562          */
 563         int (*store_super)(struct supertype *st, int fd);
 564
 565         /*  Write all metadata for this array.
 566          */
 567         int (*write_init_super)(struct supertype *st);
 568         int (*compare_super)(struct supertype *st, struct supertype *tst);
 569         int (*load_super)(struct supertype *st, int fd, char *devname);
 570         struct supertype * (*match_metadata_desc)(char *arg);
 571         __u64 (*avail_size)(struct supertype *st, __u64 size);
 572         int (*add_internal_bitmap)(struct supertype *st, int *chunkp,
 573                                    int delay, int write_behind,
 574                                    unsigned long long size, int may_change, int major);
 575         void (*locate_bitmap)(struct supertype *st, int fd);
 576         int (*write_bitmap)(struct supertype *st, int fd);
 577         void (*free_super)(struct supertype *st);
 578
 579         /* validate_geometry is called with an st returned by
 580          * match_metadata_desc.
 581          * It should check that the geometry described in compatible with
 582          * the metadata type.  It will be called repeatedly as devices
 583          * added to validate changing size and new devices.  If there are
 584          * inter-device dependencies, it should record sufficient details
 585          * so these can be validated.
 586          * Both 'size' and '*freesize' are in sectors.  chunk is bytes.
 587          */
 588         int (*validate_geometry)(struct supertype *st, int level, int layout,
 589                                  int raiddisks,
 590                                  int chunk, unsigned long long size,
 591                                  char *subdev, unsigned long long *freesize,
 592                                  int verbose);
 593
 594         struct mdinfo *(*container_content)(struct supertype *st);
 595         /* Allow a metadata handler to override mdadm's default layouts */
 596         int (*default_layout)(int level); /* optional */
 597
 598 /* for mdmon */
 599         int (*open_new)(struct supertype *c, struct active_array *a,
 600                         char *inst);
 601
 602         /* Tell the metadata handler the current state of the array.
 603          * This covers whether it is known to be consistent (no pending writes)
 604          * and how far along a resync is known to have progressed
 605          * (in a->resync_start).
 606          * resync status is really irrelevant if the array is not consistent,
 607          * but some metadata (DDF!) have a place to record the distinction.
 608          * If 'consistent' is '2', then the array can mark it dirty if a
 609          * resync/recovery/whatever is required, or leave it clean if not.
 610          * Return value is 0 dirty (not consistent) and 1 if clean.
 611          * it is only really important if consistent is passed in as '2'.
 612          */
 613         int (*set_array_state)(struct active_array *a, int consistent);
 614
 615         /* When the state of a device might have changed, we call set_disk to
 616          * tell the metadata what the current state is.
 617          * Typically this happens on spare->in_sync and (spare|in_sync)->faulty
 618          * transitions.
 619          * set_disk might be called when the state of the particular disk has
 620          * not in fact changed.
 621          */
 622         void (*set_disk)(struct active_array *a, int n, int state);
 623         void (*sync_metadata)(struct supertype *st);
 624         void (*process_update)(struct supertype *st,
 625                                struct metadata_update *update);
 626         void (*prepare_update)(struct supertype *st,
 627                                struct metadata_update *update);
 628
 629         /* activate_spare will check if the array is degraded and, if it
 630          * is, try to find some spare space in the container.
 631          * On success, it add appropriate updates (For process_update) to
 632          * to the 'updates' list and returns a list of 'mdinfo' identifying
 633          * the device, or devices as there might be multiple missing
 634          * devices and multiple spares available.
 635          */
 636         struct mdinfo *(*activate_spare)(struct active_array *a,
 637                                          struct metadata_update **updates);
 638
 639         int swapuuid; /* true if uuid is bigending rather than hostendian */
 640         int external;
 641         const char *name; /* canonical metadata name */
 642 } super0, super1, super_ddf, *superlist[];
 643
 644 extern struct superswitch super_imsm;
 645
 646 struct metadata_update {
 647         int     len;
 648         char    *buf;
 649         void    *space; /* allocated space that monitor will use */
 650         struct metadata_update *next;
 651 };
 652
 653 /* A supertype holds a particular collection of metadata.
 654  * It identifies the metadata type by the superswitch, and the particular
 655  * sub-version of that metadata type.
 656  * metadata read in or created is stored in 'sb' and 'info'.
 657  * There are also fields used by mdmon to track containers.
 658  *
 659  * A supertype may refer to:
 660  *   Just an array, possibly in a container
 661  *   A container, not identifying any particular array
 662  *   Info read from just one device, not yet fully describing the array/container.
 663  *
 664  *
 665  * A supertype is created by:
 666  *   super_by_fd
 667  *   guess_super
 668  *   dup_super
 669  */
 670 struct supertype {
 671         struct superswitch *ss;
 672         int minor_version;
 673         int max_devs;
 674         int container_dev;    /* devnum of container */
 675         char subarray[32];      /* name of array inside container */
 676         void *sb;
 677         void *info;
 678         int loaded_container;   /* Set if load_super found a container,
 679                                  * not just one device */
 680
 681         struct metadata_update *updates;
 682         struct metadata_update **update_tail;
 683
 684         /* extra stuff used by mdmon */
 685         struct active_array *arrays;
 686         int sock; /* listen to external programs */
 687         int devnum;
 688         char *devname; /* e.g. md0.  This appears in metadata_verison:
 689                         *  external:/md0/12
 690                         */
 691         int devcnt;
 692
 693         struct mdinfo *devs;
 694
 695 };
 696
 697 extern struct supertype *super_by_fd(int fd);
 698 extern struct supertype *guess_super(int fd);
 699 extern struct supertype *dup_super(struct supertype *st);
 700 extern int get_dev_size(int fd, char *dname, unsigned long long *sizep);
 701 extern void get_one_disk(int mdfd, mdu_array_info_t *ainf,
 702                          mdu_disk_info_t *disk);
 703 void wait_for(char *dev, int fd);
 704
 705 #if __GNUC__ < 3
 706 struct stat64;
 707 #endif
 708
 709 #define HAVE_NFTW  we assume
 710 #define HAVE_FTW
 711
 712 #ifdef __UCLIBC__
 713 # include <features.h>
 714 # ifndef __UCLIBC_HAS_LFS__
 715 #  define lseek64 lseek
 716 # endif
 717 # ifndef  __UCLIBC_HAS_FTW__
 718 #  undef HAVE_FTW
 719 #  undef HAVE_NFTW
 720 # endif
 721 #endif
 722
 723 #ifdef __dietlibc__
 724 # undef HAVE_NFTW
 725 #endif
 726
 727 #if defined(__KLIBC__)
 728 # undef HAVE_NFTW
 729 # undef HAVE_FTW
 730 #endif
 731
 732 #ifndef HAVE_NFTW
 733 # define FTW_PHYS 1
 734 # ifndef HAVE_FTW
 735   struct FTW {};
 736 # endif
 737 #endif
 738
 739 #ifdef HAVE_FTW
 740 # include <ftw.h>
 741 #endif
 742
 743 extern int add_dev(const char *name, const struct stat *stb, int flag, struct FTW *s);
 744
 745
 746 extern int Manage_ro(char *devname, int fd, int readonly);
 747 extern int Manage_runstop(char *devname, int fd, int runstop, int quiet);
 748 extern int Manage_resize(char *devname, int fd, long long size, int raid_disks);
 749 extern int Manage_subdevs(char *devname, int fd,
 750                           mddev_dev_t devlist, int verbose);
 751 extern int autodetect(void);
 752 extern int Grow_Add_device(char *devname, int fd, char *newdev);
 753 extern int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force);
 754 extern int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
 755                         long long size,
 756                         int level, char *layout_str, int chunksize, int raid_disks);
 757 extern int Grow_restart(struct supertype *st, struct mdinfo *info,
 758                         int *fdlist, int cnt, char *backup_file, int verbose);
 759 extern int Grow_continue(int mdfd, struct supertype *st,
 760                          struct mdinfo *info, char *backup_file);
 761
 762 extern int Assemble(struct supertype *st, char *mddev,
 763                     mddev_ident_t ident,
 764                     mddev_dev_t devlist, char *backup_file,
 765                     int readonly, int runstop,
 766                     char *update, char *homehost, int require_homehost,
 767                     int verbose, int force);
 768
 769 extern int Build(char *mddev, int chunk, int level, int layout,
 770                  int raiddisks, mddev_dev_t devlist, int assume_clean,
 771                  char *bitmap_file, int bitmap_chunk, int write_behind,
 772                  int delay, int verbose, int autof, unsigned long long size);
 773
 774
 775 extern int Create(struct supertype *st, char *mddev,
 776                   int chunk, int level, int layout, unsigned long long size, int raiddisks, int sparedisks,
 777                   char *name, char *homehost, int *uuid,
 778                   int subdevs, mddev_dev_t devlist,
 779                   int runstop, int verbose, int force, int assume_clean,
 780                   char *bitmap_file, int bitmap_chunk, int write_behind, int delay, int autof);
 781
 782 extern int Detail(char *dev, int brief, int export, int test, char *homehost);
 783 extern int Detail_Platform(struct superswitch *ss, int scan, int verbose);
 784 extern int Query(char *dev);
 785 extern int Examine(mddev_dev_t devlist, int brief, int export, int scan,
 786                    int SparcAdjust, struct supertype *forcest, char *homehost);
 787 extern int Monitor(mddev_dev_t devlist,
 788                    char *mailaddr, char *alert_cmd,
 789                    int period, int daemonise, int scan, int oneshot,
 790                    int dosyslog, int test, char *pidfile, int increments);
 791
 792 extern int Kill(char *dev, struct supertype *st, int force, int quiet, int noexcl);
 793 extern int Wait(char *dev);
 794 extern int WaitClean(char *dev, int sock, int verbose);
 795
 796 extern int Incremental(char *devname, int verbose, int runstop,
 797                        struct supertype *st, char *homehost, int require_homehost,
 798                        int autof);
 799 extern int Incremental_container(struct supertype *st, char *devname,
 800                                  int verbose, int runstop, int autof,
 801                                  int trustworthy);
 802 extern void RebuildMap(void);
 803 extern int IncrementalScan(int verbose);
 804
 805 extern int CreateBitmap(char *filename, int force, char uuid[16],
 806                         unsigned long chunksize, unsigned long daemon_sleep,
 807                         unsigned long write_behind,
 808                         unsigned long long array_size,
 809                         int major);
 810 extern int ExamineBitmap(char *filename, int brief, struct supertype *st);
 811 extern int bitmap_update_uuid(int fd, int *uuid, int swap);
 812 extern unsigned long bitmap_sectors(struct bitmap_super_s *bsb);
 813
 814 extern int md_get_version(int fd);
 815 extern int get_linux_version(void);
 816 extern long long parse_size(char *size);
 817 extern int parse_uuid(char *str, int uuid[4]);
 818 extern int parse_layout_10(char *layout);
 819 extern int parse_layout_faulty(char *layout);
 820 extern int check_ext2(int fd, char *name);
 821 extern int check_reiser(int fd, char *name);
 822 extern int check_raid(int fd, char *name);
 823 extern int check_partitions(int fd, char *dname, unsigned long long freesize);
 824
 825 extern int get_mdp_major(void);
 826 extern int dev_open(char *dev, int flags);
 827 extern int open_dev(int devnum);
 828 extern int open_dev_excl(int devnum);
 829 extern int is_standard(char *dev, int *nump);
 830 extern int same_dev(char *one, char *two);
 831
 832 extern int parse_auto(char *str, char *msg, int config);
 833 extern mddev_ident_t conf_get_ident(char *dev);
 834 extern mddev_dev_t conf_get_devs(void);
 835 extern int conf_test_dev(char *devname);
 836 extern int conf_test_metadata(const char *version);
 837 extern struct createinfo *conf_get_create_info(void);
 838 extern void set_conffile(char *file);
 839 extern char *conf_get_mailaddr(void);
 840 extern char *conf_get_mailfrom(void);
 841 extern char *conf_get_program(void);
 842 extern char *conf_get_homehost(int *require_homehostp);
 843 extern char *conf_line(FILE *file);
 844 extern char *conf_word(FILE *file, int allow_key);
 845 extern int conf_name_is_free(char *name);
 846 extern int devname_matches(char *name, char *match);
 847 extern struct mddev_ident_s *conf_match(struct mdinfo *info, struct supertype *st);
 848
 849 extern void free_line(char *line);
 850 extern int match_oneof(char *devices, char *devname);
 851 extern void uuid_from_super(int uuid[4], mdp_super_t *super);
 852 extern const int uuid_match_any[4];
 853 extern int same_uuid(int a[4], int b[4], int swapuuid);
 854 extern void copy_uuid(void *a, int b[4], int swapuuid);
 855 extern char *__fname_from_uuid(int id[4], int swap, char *buf, char sep);
 856 extern char *fname_from_uuid(struct supertype *st,
 857                              struct mdinfo *info, char *buf, char sep);
 858 extern unsigned long calc_csum(void *super, int bytes);
 859 extern int enough(int level, int raid_disks, int layout, int clean,
 860                    char *avail, int avail_disks);
 861 extern int ask(char *mesg);
 862 extern unsigned long long get_component_size(int fd);
 863 extern void remove_partitions(int fd);
 864 extern unsigned long long calc_array_size(int level, int raid_disks, int layout,
 865                                    int chunksize, unsigned long long devsize);
 866 extern int flush_metadata_updates(struct supertype *st);
 867 extern void append_metadata_update(struct supertype *st, void *buf, int len);
 868 extern int assemble_container_content(struct supertype *st, int mdfd,
 869                                       struct mdinfo *content, int runstop,
 870                                       char *chosen_name, int verbose);
 871
 872 extern int add_disk(int mdfd, struct supertype *st,
 873                     struct mdinfo *sra, struct mdinfo *info);
 874 extern int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info);
 875 unsigned long long min_recovery_start(struct mdinfo *array);
 876
 877 extern char *human_size(long long bytes);
 878 extern char *human_size_brief(long long bytes);
 879 extern void print_r10_layout(int layout);
 880
 881 #define NoMdDev (1<<23)
 882 extern int find_free_devnum(int use_partitions);
 883
 884 extern void put_md_name(char *name);
 885 extern char *get_md_name(int dev);
 886
 887 extern char DefaultConfFile[];
 888
 889 extern int create_mddev(char *dev, char *name, int autof, int trustworthy,
 890                         char *chosen);
 891 /* values for 'trustworthy' */
 892 #define LOCAL   1
 893 #define FOREIGN 2
 894 #define METADATA 3
 895 extern int open_mddev(char *dev, int report_errors);
 896 extern int open_container(int fd);
 897
 898 extern char *pid_dir;
 899 extern int mdmon_running(int devnum);
 900 extern int mdmon_pid(int devnum);
 901 extern int check_env(char *name);
 902 extern __u32 random32(void);
 903 extern int start_mdmon(int devnum);
 904
 905 extern char *devnum2devname(int num);
 906 extern int devname2devnum(char *name);
 907 extern int stat2devnum(struct stat *st);
 908 extern int fd2devnum(int fd);
 909
 910 static inline int dev2major(int d)
 911 {
 912         if (d >= 0)
 913                 return MD_MAJOR;
 914         else
 915                 return get_mdp_major();
 916 }
 917
 918 static inline int dev2minor(int d)
 919 {
 920         if (d >= 0)
 921                 return d;
 922         return (-1-d) << MdpMinorShift;
 923 }
 924
 925 static inline int ROUND_UP(int a, int base)
 926 {
 927         return ((a+base-1)/base)*base;
 928 }
 929
 930 static inline int is_subarray(char *vers)
 931 {
 932         /* The version string for a 'subarray' (an array in a container)
 933          * is
 934          *    /containername/componentname    for normal read-write arrays
 935          *    -containername/componentname    for read-only arrays.
 936          * containername is e.g. md0, md_d1
 937          * componentname is dependant on the metadata. e.g. '1' 'S1' ...
 938          */
 939         return (*vers == '/' || *vers == '-');
 940 }
 941
 942 #ifdef DEBUG
 943 #define dprintf(fmt, arg...) \
 944         fprintf(stderr, fmt, ##arg)
 945 #else
 946 #define dprintf(fmt, arg...) \
 947         ({ if (0) fprintf(stderr, fmt, ##arg); 0; })
 948 #endif
 949 #include <assert.h>
 950 #include <stdarg.h>
 951 static inline int xasprintf(char **strp, const char *fmt, ...) {
 952         va_list ap;
 953         int ret;
 954         va_start(ap, fmt);
 955         ret = vasprintf(strp, fmt, ap);
 956         va_end(ap);
 957         assert(ret >= 0);
 958         return ret;
 959 }
 960
 961 #define LEVEL_MULTIPATH         (-4)
 962 #define LEVEL_LINEAR            (-1)
 963 #define LEVEL_FAULTY            (-5)
 964
 965 /* kernel module doesn't know about these */
 966 #define LEVEL_CONTAINER         (-100)
 967 #define LEVEL_UNSUPPORTED       (-200)
 968
 969
 970 /* faulty stuff */
 971
 972 #define WriteTransient  0
 973 #define ReadTransient   1
 974 #define WritePersistent 2
 975 #define ReadPersistent  3
 976 #define WriteAll        4 /* doesn't go to device */
 977 #define ReadFixable     5
 978 #define Modes   6
 979
 980 #define ClearErrors     31
 981 #define ClearFaults     30
 982
 983 #define AllPersist      100 /* internal use only */
 984 #define NoPersist       101
 985
 986 #define ModeMask        0x1f
 987 #define ModeShift       5
 988
 989
 990 #ifdef __TINYC__
 991 #undef minor
 992 #undef major
 993 #undef makedev
 994 #define minor(x) ((x)&0xff)
 995 #define major(x) (((x)>>8)&0xff)
 996 #define makedev(M,m) (((M)<<8) | (m))
 997 #endif
 998
 999 /* for raid4/5/6 */
1000 #define ALGORITHM_LEFT_ASYMMETRIC       0
1001 #define ALGORITHM_RIGHT_ASYMMETRIC      1
1002 #define ALGORITHM_LEFT_SYMMETRIC        2
1003 #define ALGORITHM_RIGHT_SYMMETRIC       3
1004
1005 /* Define non-rotating (raid4) algorithms.  These allow
1006  * conversion of raid4 to raid5.
1007  */
1008 #define ALGORITHM_PARITY_0              4 /* P or P,Q are initial devices */
1009 #define ALGORITHM_PARITY_N              5 /* P or P,Q are final devices. */
1010
1011 /* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
1012  * Firstly, the exact positioning of the parity block is slightly
1013  * different between the 'LEFT_*' modes of md and the "_N_*" modes
1014  * of DDF.
1015  * Secondly, or order of datablocks over which the Q syndrome is computed
1016  * is different.
1017  * Consequently we have different layouts for DDF/raid6 than md/raid6.
1018  * These layouts are from the DDFv1.2 spec.
1019  * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
1020  * leaves RLQ=3 as 'Vendor Specific'
1021  */
1022
1023 #define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
1024 #define ALGORITHM_ROTATING_N_RESTART    9 /* DDF PRL=6 RLQ=2 */
1025 #define ALGORITHM_ROTATING_N_CONTINUE   10 /*DDF PRL=6 RLQ=3 */
1026
1027
1028 /* For every RAID5 algorithm we define a RAID6 algorithm
1029  * with exactly the same layout for data and parity, and
1030  * with the Q block always on the last device (N-1).
1031  * This allows trivial conversion from RAID5 to RAID6
1032  */
1033 #define ALGORITHM_LEFT_ASYMMETRIC_6     16
1034 #define ALGORITHM_RIGHT_ASYMMETRIC_6    17
1035 #define ALGORITHM_LEFT_SYMMETRIC_6      18
1036 #define ALGORITHM_RIGHT_SYMMETRIC_6     19
1037 #define ALGORITHM_PARITY_0_6            20
1038 #define ALGORITHM_PARITY_N_6            ALGORITHM_PARITY_N
1039
1040 /* Define PATH_MAX in case we don't use glibc or standard library does
1041  * not have PATH_MAX defined. Assume max path length is 4K characters.
1042  */
1043 #ifndef PATH_MAX
1044 #define PATH_MAX        4096
1045 #endif
1046