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