mdadm.h

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
  23  *    Paper: Neil Brown
  24  *           School of Computer Science and Engineering
  25  *           The University of New South Wales
  26  *           Sydney, 2052
  27  *           Australia
  28  */
  29
  30 #define _GNU_SOURCE
  31 #define _FILE_OFFSET_BITS 64
  32 #include        <unistd.h>
  33 #if !defined(__dietlibc__) && !defined(__KLIBC__)
  34 extern __off64_t lseek64 __P ((int __fd, __off64_t __offset, int __whence));
  35 #else
  36 # if defined(__NO_STAT64) || __WORDSIZE != 32
  37 # define lseek64 lseek
  38 # endif
  39 #endif
  40
  41 #include        <sys/types.h>
  42 #include        <sys/stat.h>
  43 #include        <stdlib.h>
  44 #include        <time.h>
  45 #include        <sys/time.h>
  46 #include        <getopt.h>
  47 #include        <fcntl.h>
  48 #include        <stdio.h>
  49 #include        <errno.h>
  50 #include        <string.h>
  51 #include        <syslog.h>
  52 #ifdef __dietlibc__
  53 #include        <strings.h>
  54 /* dietlibc has deprecated random and srandom!! */
  55 #define random rand
  56 #define srandom srand
  57 #endif
  58
  59
  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>
  65 #define MD_MAJOR 9
  66 #define MdpMinorShift 6
  67
  68 #ifndef BLKGETSIZE64
  69 #define BLKGETSIZE64 _IOR(0x12,114,size_t) /* return device size in bytes (u64 *arg) */
  70 #endif
  71
  72 #define DEFAULT_BITMAP_CHUNK 4096
  73 #define DEFAULT_BITMAP_DELAY 5
  74 #define DEFAULT_MAX_WRITE_BEHIND 256
  75
  76 #include        "md_u.h"
  77 #include        "md_p.h"
  78 #include        "bitmap.h"
  79 #include        "msg.h"
  80
  81 #include <endian.h>
  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 | \
  88                      ((x) & 0xff00U) >> 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)
 101
 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)
 110
 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)
 124
 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)
 131 #else
 132 #  error "unknown endianness."
 133 #endif
 134 #endif /* __KLIBC__ */
 135
 136
 137
 138 /* general information that might be extracted from a superblock */
 139 struct mdinfo {
 140         mdu_array_info_t        array;
 141         mdu_disk_info_t         disk;
 142         __u64                   events;
 143         int                     uuid[4];
 144         char                    name[33];
 145         unsigned long long      data_offset;
 146         unsigned long long      component_size; /* same as array.size, except in
 147                                                  * sectors and up to 64bits.
 148                                                  */
 149         int                     reshape_active;
 150         unsigned long long      reshape_progress;
 151         unsigned long long      resync_start;
 152         unsigned long           safe_mode_delay; /* ms delay to mark clean */
 153         int                     new_level, delta_disks, new_layout, new_chunk;
 154         int                     errors;
 155         int                     cache_size; /* size of raid456 stripe cache*/
 156         int                     mismatch_cnt;
 157         char                    text_version[50];
 158
 159         int container_member; /* for assembling external-metatdata arrays
 160                                * This is to be used internally by metadata
 161                                * handler only */
 162
 163         char            sys_name[20];
 164         struct mdinfo *devs;
 165         struct mdinfo *next;
 166
 167         /* Device info for mdmon: */
 168         int state_fd;
 169         #define DS_FAULTY       1
 170         #define DS_INSYNC       2
 171         #define DS_WRITE_MOSTLY 4
 172         #define DS_SPARE        8
 173         #define DS_BLOCKED      16
 174         #define DS_REMOVE       1024
 175         #define DS_UNBLOCK      2048
 176         int prev_state, curr_state, next_state;
 177
 178 };
 179
 180 struct createinfo {
 181         int     uid;
 182         int     gid;
 183         int     autof;
 184         int     mode;
 185         int     symlinks;
 186         struct supertype *supertype;
 187 };
 188
 189 #define Name "mdadm"
 190
 191 enum mode {
 192         ASSEMBLE=1,
 193         BUILD,
 194         CREATE,
 195         MANAGE,
 196         MISC,
 197         MONITOR,
 198         GROW,
 199         INCREMENTAL,
 200         AUTODETECT,
 201 };
 202
 203 extern char short_options[];
 204 extern char short_bitmap_options[];
 205 extern char short_bitmap_auto_options[];
 206 extern struct option long_options[];
 207 extern char Version[], Usage[], Help[], OptionHelp[],
 208         Help_create[], Help_build[], Help_assemble[], Help_grow[],
 209         Help_incr[],
 210         Help_manage[], Help_misc[], Help_monitor[], Help_config[];
 211
 212 /* for option that don't have short equivilents, we assign arbitrary
 213  * small numbers.  '1' means an undecorated option, so we start at '2'.
 214  */
 215 enum special_options {
 216         AssumeClean = 2,
 217         BitmapChunk,
 218         WriteBehind,
 219         ReAdd,
 220         NoDegraded,
 221         Sparc22,
 222         BackupFile,
 223         HomeHost,
 224         AutoHomeHost,
 225         Symlinks,
 226         AutoDetect,
 227         Waitclean,
 228 };
 229
 230 /* structures read from config file */
 231 /* List of mddevice names and identifiers
 232  * Identifiers can be:
 233  *    uuid=128-hex-uuid
 234  *    super-minor=decimal-minor-number-from-superblock
 235  *    devices=comma,separated,list,of,device,names,with,wildcards
 236  *
 237  * If multiple fields are present, the intersection of all matching
 238  * devices is considered
 239  */
 240 #define UnSet (0xfffe)
 241 typedef struct mddev_ident_s {
 242         char    *devname;
 243
 244         int     uuid_set;
 245         int     uuid[4];
 246         char    name[33];
 247
 248         unsigned int super_minor;
 249
 250         char    *devices;       /* comma separated list of device
 251                                  * names with wild cards
 252                                  */
 253         int     level;
 254         unsigned int raid_disks;
 255         unsigned int spare_disks;
 256         struct supertype *st;
 257         int     autof;          /* 1 for normal, 2 for partitioned */
 258         char    *spare_group;
 259         char    *bitmap_file;
 260         int     bitmap_fd;
 261
 262         char    *container;     /* /dev/whatever name of container, or
 263                                  * uuid of container.  You would expect
 264                                  * this to be the 'devname' or UUID
 265                                  * of some other entry.
 266                                  */
 267         char    *member;        /* subarray within a container */
 268
 269         struct mddev_ident_s *next;
 270 } *mddev_ident_t;
 271
 272 /* List of device names - wildcards expanded */
 273 typedef struct mddev_dev_s {
 274         char *devname;
 275         char disposition;       /* 'a' for add, 'r' for remove, 'f' for fail.
 276                                  * Not set for names read from .config
 277                                  */
 278         char writemostly;       /* 1 for 'set writemostly', 2 for 'clear writemostly' */
 279         char re_add;
 280         char used;              /* set when used */
 281         struct mdinfo *content; /* If devname is a container, this might list
 282                                  * the remaining member arrays. */
 283         struct mddev_dev_s *next;
 284 } *mddev_dev_t;
 285
 286 typedef struct mapping {
 287         char *name;
 288         int num;
 289 } mapping_t;
 290
 291
 292 struct mdstat_ent {
 293         char            *dev;
 294         int             devnum;
 295         int             active;
 296         char            *level;
 297         char            *pattern; /* U or up, _ for down */
 298         int             percent; /* -1 if no resync */
 299         int             resync; /* 1 if resync, 0 if recovery */
 300         int             devcnt;
 301         int             raid_disks;
 302         int             chunk_size;
 303         char *          metadata_version;
 304         struct mdstat_ent *next;
 305 };
 306
 307 extern struct mdstat_ent *mdstat_read(int hold, int start);
 308 extern void free_mdstat(struct mdstat_ent *ms);
 309 extern void mdstat_wait(int seconds);
 310 extern void mdstat_wait_fd(int fd, const sigset_t *sigmask);
 311 extern int mddev_busy(int devnum);
 312
 313 struct map_ent {
 314         struct map_ent *next;
 315         int     devnum;
 316         char    metadata[20];
 317         int     uuid[4];
 318         int     bad;
 319         char    *path;
 320 };
 321 extern int map_update(struct map_ent **mpp, int devnum, char *metadata,
 322                       int uuid[4], char *path);
 323 extern struct map_ent *map_by_uuid(struct map_ent **map, int uuid[4]);
 324 extern struct map_ent *map_by_devnum(struct map_ent **map, int devnum);
 325 extern struct map_ent *map_by_name(struct map_ent **map, char *name);
 326 extern void map_read(struct map_ent **melp);
 327 extern int map_write(struct map_ent *mel);
 328 extern void map_delete(struct map_ent **mapp, int devnum);
 329 extern void map_free(struct map_ent *map);
 330 extern void map_add(struct map_ent **melp,
 331                     int devnum, char *metadata, int uuid[4], char *path);
 332 extern int map_lock(struct map_ent **melp);
 333 extern void map_unlock(struct map_ent **melp);
 334
 335 /* various details can be requested */
 336 #define GET_LEVEL       1
 337 #define GET_LAYOUT      2
 338 #define GET_COMPONENT   4
 339 #define GET_CHUNK       8
 340 #define GET_CACHE       16
 341 #define GET_MISMATCH    32
 342 #define GET_VERSION     64
 343 #define GET_DISKS       128
 344 #define GET_DEGRADED    256
 345 #define GET_SAFEMODE    512
 346
 347 #define GET_DEVS        1024 /* gets role, major, minor */
 348 #define GET_OFFSET      2048
 349 #define GET_SIZE        4096
 350 #define GET_STATE       8192
 351 #define GET_ERROR       16384
 352
 353 /* If fd >= 0, get the array it is open on,
 354  * else use devnum. >=0 -> major9. <0.....
 355  */
 356 extern int sysfs_open(int devnum, char *devname, char *attr);
 357 extern void sysfs_init(struct mdinfo *mdi, int fd, int devnum);
 358 extern void sysfs_free(struct mdinfo *sra);
 359 extern struct mdinfo *sysfs_read(int fd, int devnum, unsigned long options);
 360 extern int sysfs_attr_match(const char *attr, const char *str);
 361 extern int sysfs_match_word(const char *word, char **list);
 362 extern int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 363                          char *name, char *val);
 364 extern int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 365                          char *name, unsigned long long val);
 366 extern int sysfs_uevent(struct mdinfo *sra, char *event);
 367 extern int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 368                         char *name, unsigned long long *val);
 369 extern int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms);
 370 extern int sysfs_set_array(struct mdinfo *info, int vers);
 371 extern int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd);
 372 extern int sysfs_disk_to_scsi_id(int fd, __u32 *id);
 373 extern int sysfs_unique_holder(int devnum, long rdev);
 374 extern int load_sys(char *path, char *buf);
 375
 376
 377 extern int save_stripes(int *source, unsigned long long *offsets,
 378                         int raid_disks, int chunk_size, int level, int layout,
 379                         int nwrites, int *dest,
 380                         unsigned long long start, unsigned long long length);
 381 extern int restore_stripes(int *dest, unsigned long long *offsets,
 382                            int raid_disks, int chunk_size, int level, int layout,
 383                            int source, unsigned long long read_offset,
 384                            unsigned long long start, unsigned long long length);
 385
 386 #ifndef Sendmail
 387 #define Sendmail "/usr/lib/sendmail -t"
 388 #endif
 389
 390 #define SYSLOG_FACILITY LOG_DAEMON
 391
 392 extern char *map_num(mapping_t *map, int num);
 393 extern int map_name(mapping_t *map, char *name);
 394 extern mapping_t r5layout[], pers[], modes[], faultylayout[];
 395
 396 extern char *map_dev(int major, int minor, int create);
 397
 398 struct active_array;
 399 struct metadata_update;
 400
 401 /* A superswitch provides entry point the a metadata handler.
 402  *
 403  * The super_switch primarily operates on some "metadata" that
 404  * is accessed via the 'supertype'.
 405  * This metadata has one of three possible sources.
 406  * 1/ It is read from a single device.  In this case it may not completely
 407  *    describe the array or arrays as some information might be on other
 408  *    devices.
 409  * 2/ It is read from all devices in a container.  In this case all
 410  *    information is present.
 411  * 3/ It is created by ->init_super / ->add_to_super.  In this case it will
 412  *    be complete once enough ->add_to_super calls have completed.
 413  *
 414  * When creating an array inside a container, the metadata will be
 415  * formed by a combination of 2 and 3.  The metadata or the array is read,
 416  * then new information is added.
 417  *
 418  * The metadata must sometimes have a concept of a 'current' array
 419  * and a 'current' device.
 420  * The 'current' array is set by init_super to be the newly created array,
 421  * or is set by super_by_fd when it finds it is looking at an array inside
 422  * a container.
 423  *
 424  * The 'current' device is either the device that the metadata was read from
 425  * in case 1, or the last device added by add_to_super in case 3.
 426  * Case 2 does not identify a 'current' device.
 427  */
 428 extern struct superswitch {
 429
 430         /* Used to report details of metadata read from a component
 431          * device. ->load_super has been called.
 432          */
 433         void (*examine_super)(struct supertype *st, char *homehost);
 434         void (*brief_examine_super)(struct supertype *st);
 435         void (*export_examine_super)(struct supertype *st);
 436
 437         /* Used to report details of an active array.
 438          * ->load_super was possibly given a 'component' string.
 439          */
 440         void (*detail_super)(struct supertype *st, char *homehost);
 441         void (*brief_detail_super)(struct supertype *st);
 442         void (*export_detail_super)(struct supertype *st);
 443
 444         /* Used:
 445          *   to get uuid to storing in bitmap metadata
 446          *   and 'reshape' backup-data metadata
 447          *   To see if a device is being re-added to an array it was part of.
 448          */
 449         void (*uuid_from_super)(struct supertype *st, int uuid[4]);
 450
 451         /* Extra generic details from metadata.  This could be details about
 452          * the container, or about an individual array within the container.
 453          * The determination is made either by:
 454          *   load_super being given a 'component' string.
 455          *   validate_geometry determining what to create.
 456          * The info includes both array information and device information.
 457          * The particular device should be:
 458          *   The last device added by add_to_super
 459          *   The device the metadata was loaded from by load_super
 460          */
 461         void (*getinfo_super)(struct supertype *st, struct mdinfo *info);
 462
 463         /* Check if the given metadata is flagged as belonging to "this"
 464          * host.  0 for 'no', 1 for 'yes', -1 for "Don't record homehost"
 465          */
 466         int (*match_home)(struct supertype *st, char *homehost);
 467
 468         /* Make one of several generic modifications to metadata
 469          * prior to assembly (or other times).
 470          *   sparc2.2  - first bug in early 0.90 metadata
 471          *   super-minor - change name of 0.90 metadata
 472          *   summaries - 'correct' any redundant data
 473          *   resync - mark array as dirty to trigger a resync.
 474          *   uuid - set new uuid - only 0.90 or 1.x
 475          *   name - change the name of the array (where supported)
 476          *   homehost - change which host this array is tied to.
 477          *   devicesize - If metadata is at start of device, change recorded
 478          *               device size to match actual device size
 479          *   byteorder - swap bytes for 0.90 metadata
 480          *
 481          *   force-one  - mark that device as uptodate, not old or failed.
 482          *   force-array - mark array as clean if it would not otherwise
 483          *               assemble
 484          *   assemble   - not sure how this is different from force-one...
 485          *   linear-grow-new - add a new device to a linear array, but don't
 486          *                   change the size: so superblock still matches
 487          *   linear-grow-update - now change the size of the array.
 488          */
 489         int (*update_super)(struct supertype *st, struct mdinfo *info,
 490                             char *update,
 491                             char *devname, int verbose,
 492                             int uuid_set, char *homehost);
 493
 494         /* Create new metadata for new array as described.  This could
 495          * be a new container, or an array in a pre-existing container.
 496          * Also used to zero metadata prior to writing it to invalidate old
 497          * metadata.
 498          */
 499         int (*init_super)(struct supertype *st, mdu_array_info_t *info,
 500                           unsigned long long size, char *name,
 501                           char *homehost, int *uuid);
 502
 503         /* update the metadata to include new device, either at create or
 504          * when hot-adding a spare.
 505          */
 506         void (*add_to_super)(struct supertype *st, mdu_disk_info_t *dinfo,
 507                              int fd, char *devname);
 508
 509         /* Write metadata to one device when fixing problems or adding
 510          * a new device.
 511          */
 512         int (*store_super)(struct supertype *st, int fd);
 513
 514         /*  Write all metadata for this array.
 515          */
 516         int (*write_init_super)(struct supertype *st);
 517         int (*compare_super)(struct supertype *st, struct supertype *tst);
 518         int (*load_super)(struct supertype *st, int fd, char *devname);
 519         struct supertype * (*match_metadata_desc)(char *arg);
 520         __u64 (*avail_size)(struct supertype *st, __u64 size);
 521         int (*add_internal_bitmap)(struct supertype *st, int *chunkp,
 522                                    int delay, int write_behind,
 523                                    unsigned long long size, int may_change, int major);
 524         void (*locate_bitmap)(struct supertype *st, int fd);
 525         int (*write_bitmap)(struct supertype *st, int fd);
 526         void (*free_super)(struct supertype *st);
 527
 528         /* validate_geometry is called with an st returned by
 529          * match_metadata_desc.
 530          * It should check that the geometry described in compatible with
 531          * the metadata type.  It will be called repeatedly as devices
 532          * added to validate changing size and new devices.  If there are
 533          * inter-device dependencies, it should record sufficient details
 534          * so these can be validated.
 535          */
 536         int (*validate_geometry)(struct supertype *st, int level, int layout,
 537                                  int raiddisks,
 538                                  int chunk, unsigned long long size,
 539                                  char *subdev, unsigned long long *freesize,
 540                                  int verbose);
 541
 542         struct mdinfo *(*container_content)(struct supertype *st);
 543
 544 /* for mdmon */
 545         int (*open_new)(struct supertype *c, struct active_array *a,
 546                         char *inst);
 547
 548         /* Tell the metadata handler the current state of the array.
 549          * This covers whether it is known to be consistent (no pending writes)
 550          * and how far along a resync is known to have progressed
 551          * (in a->resync_start).
 552          * resync status is really irrelevant if the array is not consistent,
 553          * but some metadata (DDF!) have a place to record the distinction.
 554          * If 'consistent' is '2', then the array can mark it dirty if a
 555          * resync/recovery/whatever is required, or leave it clean if not.
 556          * Return value is 0 dirty (not consistent) and 1 if clean.
 557          * it is only really important if consistent is passed in as '2'.
 558          */
 559         int (*set_array_state)(struct active_array *a, int consistent);
 560
 561         /* When the state of a device might have changed, we call set_disk to
 562          * tell the metadata what the current state is.
 563          * Typically this happens on spare->in_sync and (spare|in_sync)->faulty
 564          * transitions.
 565          * set_disk might be called when the state of the particular disk has
 566          * not in fact changed.
 567          */
 568         void (*set_disk)(struct active_array *a, int n, int state);
 569         void (*sync_metadata)(struct supertype *st);
 570         void (*process_update)(struct supertype *st,
 571                                struct metadata_update *update);
 572         void (*prepare_update)(struct supertype *st,
 573                                struct metadata_update *update);
 574
 575         /* activate_spare will check if the array is degraded and, if it
 576          * is, try to find some spare space in the container.
 577          * On success, it add appropriate updates (For process_update) to
 578          * to the 'updates' list and returns a list of 'mdinfo' identifying
 579          * the device, or devices as there might be multiple missing
 580          * devices and multiple spares available.
 581          */
 582         struct mdinfo *(*activate_spare)(struct active_array *a,
 583                                          struct metadata_update **updates);
 584
 585         int swapuuid; /* true if uuid is bigending rather than hostendian */
 586         int external;
 587 } super0, super1, super_ddf, *superlist[];
 588
 589 extern struct superswitch super_imsm;
 590
 591 struct metadata_update {
 592         int     len;
 593         char    *buf;
 594         void    *space; /* allocated space that monitor will use */
 595         struct metadata_update *next;
 596 };
 597
 598 /* A supertype holds a particular collection of metadata.
 599  * It identifies the metadata type by the superswitch, and the particular
 600  * sub-version of that metadata type.
 601  * metadata read in or created is stored in 'sb' and 'info'.
 602  * There are also fields used by mdmon to track containers.
 603  *
 604  * A supertype may refer to:
 605  *   Just an array, possibly in a container
 606  *   A container, not identifying any particular array
 607  *   Info read from just one device, not yet fully describing the array/container.
 608  *
 609  *
 610  * A supertype is created by:
 611  *   super_by_fd
 612  *   guess_super
 613  *   dup_super
 614  */
 615 struct supertype {
 616         struct superswitch *ss;
 617         int minor_version;
 618         int max_devs;
 619         int container_dev;    /* devnum of container */
 620         char subarray[32];      /* name of array inside container */
 621         void *sb;
 622         void *info;
 623         int loaded_container;   /* Set if load_super found a container,
 624                                  * not just one device */
 625
 626         struct metadata_update *updates;
 627         struct metadata_update **update_tail;
 628
 629         /* extra stuff used by mdmon */
 630         struct active_array *arrays;
 631         int sock; /* listen to external programs */
 632         int devnum;
 633         char *devname; /* e.g. md0.  This appears in metadata_verison:
 634                         *  external:/md0/12
 635                         */
 636         int devcnt;
 637         char *device_name; /* e.g. /dev/md/whatever */
 638
 639         struct mdinfo *devs;
 640
 641 };
 642
 643 extern struct supertype *super_by_fd(int fd);
 644 extern struct supertype *guess_super(int fd);
 645 extern struct supertype *dup_super(struct supertype *st);
 646 extern int get_dev_size(int fd, char *dname, unsigned long long *sizep);
 647 extern void get_one_disk(int mdfd, mdu_array_info_t *ainf,
 648                          mdu_disk_info_t *disk);
 649 void wait_for(char *dev);
 650
 651 #if __GNUC__ < 3
 652 struct stat64;
 653 #endif
 654
 655 #define HAVE_NFTW  we assume
 656 #define HAVE_FTW
 657
 658 #ifdef UCLIBC
 659 # include <features.h>
 660 # ifndef  __UCLIBC_HAS_FTW__
 661 #  undef HAVE_FTW
 662 #  undef HAVE_NFTW
 663 # endif
 664 #endif
 665
 666 #ifdef __dietlibc__
 667 # undef HAVE_NFTW
 668 #endif
 669
 670 #if defined(__KLIBC__)
 671 # undef HAVE_NFTW
 672 # undef HAVE_FTW
 673 #endif
 674
 675 #ifndef HAVE_NFTW
 676 # define FTW_PHYS 1
 677 # ifndef HAVE_FTW
 678   struct FTW {};
 679 # endif
 680 #endif
 681
 682 #ifdef HAVE_FTW
 683 # include <ftw.h>
 684 #endif
 685
 686 extern int add_dev(const char *name, const struct stat *stb, int flag, struct FTW *s);
 687
 688
 689 extern int Manage_ro(char *devname, int fd, int readonly);
 690 extern int Manage_runstop(char *devname, int fd, int runstop, int quiet);
 691 extern int Manage_resize(char *devname, int fd, long long size, int raid_disks);
 692 extern int Manage_reconfig(char *devname, int fd, int layout);
 693 extern int Manage_subdevs(char *devname, int fd,
 694                           mddev_dev_t devlist, int verbose);
 695 extern int autodetect(void);
 696 extern int Grow_Add_device(char *devname, int fd, char *newdev);
 697 extern int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force);
 698 extern int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
 699                         long long size,
 700                         int level, int layout, int chunksize, int raid_disks);
 701 extern int Grow_restart(struct supertype *st, struct mdinfo *info,
 702                         int *fdlist, int cnt, char *backup_file);
 703
 704
 705 extern int Assemble(struct supertype *st, char *mddev,
 706                     mddev_ident_t ident,
 707                     mddev_dev_t devlist, char *backup_file,
 708                     int readonly, int runstop,
 709                     char *update, char *homehost,
 710                     int verbose, int force);
 711
 712 extern int Build(char *mddev, int chunk, int level, int layout,
 713                  int raiddisks, mddev_dev_t devlist, int assume_clean,
 714                  char *bitmap_file, int bitmap_chunk, int write_behind,
 715                  int delay, int verbose, int autof);
 716
 717
 718 extern int Create(struct supertype *st, char *mddev,
 719                   int chunk, int level, int layout, unsigned long long size, int raiddisks, int sparedisks,
 720                   char *name, char *homehost, int *uuid,
 721                   int subdevs, mddev_dev_t devlist,
 722                   int runstop, int verbose, int force, int assume_clean,
 723                   char *bitmap_file, int bitmap_chunk, int write_behind, int delay, int autof);
 724
 725 extern int Detail(char *dev, int brief, int export, int test, char *homehost);
 726 extern int Query(char *dev);
 727 extern int Examine(mddev_dev_t devlist, int brief, int export, int scan,
 728                    int SparcAdjust, struct supertype *forcest, char *homehost);
 729 extern int Monitor(mddev_dev_t devlist,
 730                    char *mailaddr, char *alert_cmd,
 731                    int period, int daemonise, int scan, int oneshot,
 732                    int dosyslog, int test, char *pidfile);
 733
 734 extern int Kill(char *dev, int force, int quiet, int noexcl);
 735 extern int Wait(char *dev);
 736 extern int WaitClean(char *dev, int verbose);
 737
 738 extern int Incremental(char *devname, int verbose, int runstop,
 739                        struct supertype *st, char *homehost, int autof);
 740 extern int Incremental_container(struct supertype *st, char *devname,
 741                                  int verbose, int runstop, int autof,
 742                                  int trustworthy);
 743 extern void RebuildMap(void);
 744 extern int IncrementalScan(int verbose);
 745
 746 extern int CreateBitmap(char *filename, int force, char uuid[16],
 747                         unsigned long chunksize, unsigned long daemon_sleep,
 748                         unsigned long write_behind,
 749                         unsigned long long array_size,
 750                         int major);
 751 extern int ExamineBitmap(char *filename, int brief, struct supertype *st);
 752 extern int bitmap_update_uuid(int fd, int *uuid, int swap);
 753 extern unsigned long bitmap_sectors(struct bitmap_super_s *bsb);
 754
 755 extern int md_get_version(int fd);
 756 extern int get_linux_version(void);
 757 extern int parse_uuid(char *str, int uuid[4]);
 758 extern int check_ext2(int fd, char *name);
 759 extern int check_reiser(int fd, char *name);
 760 extern int check_raid(int fd, char *name);
 761
 762 extern int get_mdp_major(void);
 763 extern int dev_open(char *dev, int flags);
 764 extern int open_dev_excl(int devnum);
 765 extern int is_standard(char *dev, int *nump);
 766 extern int same_dev(char *one, char *two);
 767
 768 extern int parse_auto(char *str, char *msg, int config);
 769 extern mddev_ident_t conf_get_ident(char *dev);
 770 extern mddev_dev_t conf_get_devs(void);
 771 extern int conf_test_dev(char *devname);
 772 extern struct createinfo *conf_get_create_info(void);
 773 extern void set_conffile(char *file);
 774 extern char *conf_get_mailaddr(void);
 775 extern char *conf_get_mailfrom(void);
 776 extern char *conf_get_program(void);
 777 extern char *conf_get_homehost(void);
 778 extern char *conf_line(FILE *file);
 779 extern char *conf_word(FILE *file, int allow_key);
 780 extern void free_line(char *line);
 781 extern int match_oneof(char *devices, char *devname);
 782 extern void uuid_from_super(int uuid[4], mdp_super_t *super);
 783 extern const int uuid_match_any[4];
 784 extern int same_uuid(int a[4], int b[4], int swapuuid);
 785 extern void copy_uuid(void *a, int b[4], int swapuuid);
 786 extern char *fname_from_uuid(struct supertype *st,
 787                              struct mdinfo *info, char *buf, char sep);
 788 extern unsigned long calc_csum(void *super, int bytes);
 789 extern int enough(int level, int raid_disks, int layout, int clean,
 790                    char *avail, int avail_disks);
 791 extern int ask(char *mesg);
 792 extern unsigned long long get_component_size(int fd);
 793 extern void remove_partitions(int fd);
 794 extern unsigned long long calc_array_size(int level, int raid_disks, int layout,
 795                                    int chunksize, unsigned long long devsize);
 796 extern int flush_metadata_updates(struct supertype *st);
 797 extern void append_metadata_update(struct supertype *st, void *buf, int len);
 798 extern int assemble_container_content(struct supertype *st, int mdfd,
 799                                       struct mdinfo *content, int runstop,
 800                                       char *chosen_name, int verbose);
 801
 802 extern int add_disk(int mdfd, struct supertype *st,
 803                     struct mdinfo *sra, struct mdinfo *info);
 804 extern int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info);
 805
 806 extern char *human_size(long long bytes);
 807 extern char *human_size_brief(long long bytes);
 808 extern void print_r10_layout(int layout);
 809
 810 #define NoMdDev (1<<23)
 811 extern int find_free_devnum(int use_partitions);
 812
 813 extern void put_md_name(char *name);
 814 extern char *get_md_name(int dev);
 815
 816 extern char DefaultConfFile[];
 817
 818 extern int create_mddev(char *dev, char *name, int autof, int trustworthy,
 819                         char *chosen);
 820 /* values for 'trustworthy' */
 821 #define LOCAL   1
 822 #define FOREIGN 2
 823 #define METADATA 3
 824 extern int open_mddev(char *dev, int report_errors);
 825 extern int open_container(int fd);
 826
 827 extern int mdmon_running(int devnum);
 828 extern int signal_mdmon(int devnum);
 829 extern int check_env(char *name);
 830 extern int start_mdmon(int devnum);
 831
 832 extern char *devnum2devname(int num);
 833 extern int devname2devnum(char *name);
 834 extern int stat2devnum(struct stat *st);
 835 extern int fd2devnum(int fd);
 836
 837 static inline int dev2major(int d)
 838 {
 839         if (d >= 0)
 840                 return MD_MAJOR;
 841         else
 842                 return get_mdp_major();
 843 }
 844
 845 static inline int dev2minor(int d)
 846 {
 847         if (d >= 0)
 848                 return d;
 849         return (-1-d) << MdpMinorShift;
 850 }
 851
 852 static inline int ROUND_UP(int a, int base)
 853 {
 854         return ((a+base-1)/base)*base;
 855 }
 856
 857 static inline int is_subarray(char *vers)
 858 {
 859         /* The version string for a 'subarray' (an array in a container)
 860          * is
 861          *    /containername/componentname    for normal read-write arrays
 862          *    -containername/componentname    for read-only arrays.
 863          * containername is e.g. md0, md_d1
 864          * componentname is dependant on the metadata. e.g. '1' 'S1' ...
 865          */
 866         return (*vers == '/' || *vers == '-');
 867 }
 868
 869 #ifdef DEBUG
 870 #define dprintf(fmt, arg...) \
 871         fprintf(stderr, fmt, ##arg)
 872 #else
 873 #define dprintf(fmt, arg...) \
 874         ({ if (0) fprintf(stderr, fmt, ##arg); 0; })
 875 #endif
 876
 877 #define LEVEL_MULTIPATH         (-4)
 878 #define LEVEL_LINEAR            (-1)
 879 #define LEVEL_FAULTY            (-5)
 880
 881 /* kernel module doesn't know about these */
 882 #define LEVEL_CONTAINER         (-100)
 883 #define LEVEL_UNSUPPORTED       (-200)
 884
 885
 886 /* faulty stuff */
 887
 888 #define WriteTransient  0
 889 #define ReadTransient   1
 890 #define WritePersistent 2
 891 #define ReadPersistent  3
 892 #define WriteAll        4 /* doesn't go to device */
 893 #define ReadFixable     5
 894 #define Modes   6
 895
 896 #define ClearErrors     31
 897 #define ClearFaults     30
 898
 899 #define AllPersist      100 /* internal use only */
 900 #define NoPersist       101
 901
 902 #define ModeMask        0x1f
 903 #define ModeShift       5
 904
 905
 906 #ifdef __TINYC__
 907 #undef minor
 908 #undef major
 909 #undef makedev
 910 #define minor(x) ((x)&0xff)
 911 #define major(x) (((x)>>8)&0xff)
 912 #define makedev(M,m) (((M)<<8) | (m))
 913 #endif
 914
 915 /* for raid5 */
 916 #define ALGORITHM_LEFT_ASYMMETRIC       0
 917 #define ALGORITHM_RIGHT_ASYMMETRIC      1
 918 #define ALGORITHM_LEFT_SYMMETRIC        2
 919 #define ALGORITHM_RIGHT_SYMMETRIC       3