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