fs/reiserfs/reiserfs_private.h

   1 /*
   2  *  Copyright 2000-2002 by Hans Reiser, licensing governed by reiserfs/README
   3  *
   4  *  GRUB  --  GRand Unified Bootloader
   5  *  Copyright (C) 2000, 2001  Free Software Foundation, Inc.
   6  *
   7  *  (C) Copyright 2003 - 2004
   8  *  Sysgo AG, <www.elinos.com>, Pavel Bartusek <pba@sysgo.com>
   9  *
  10  *
  11  * SPDX-License-Identifier:     GPL-2.0+
  12  */
  13
  14 /* An implementation for the ReiserFS filesystem ported from GRUB.
  15  * Some parts of this code (mainly the structures and defines) are
  16  * from the original reiser fs code, as found in the linux kernel.
  17  */
  18
  19 #ifndef __BYTE_ORDER
  20 #if defined(__LITTLE_ENDIAN) && !defined(__BIG_ENDIAN)
  21 #define __BYTE_ORDER __LITTLE_ENDIAN
  22 #elif defined(__BIG_ENDIAN) && !defined(__LITTLE_ENDIAN)
  23 #define __BYTE_ORDER __BIG_ENDIAN
  24 #else
  25 #error "unable to define __BYTE_ORDER"
  26 #endif
  27 #endif /* not __BYTE_ORDER */
  28
  29 #define FSYS_BUFLEN  0x8000
  30 #define FSYS_BUF     fsys_buf
  31
  32 /* This is the new super block of a journaling reiserfs system */
  33 struct reiserfs_super_block
  34 {
  35   __u32 s_block_count;                  /* blocks count         */
  36   __u32 s_free_blocks;                  /* free blocks count    */
  37   __u32 s_root_block;                   /* root block number    */
  38   __u32 s_journal_block;                /* journal block number    */
  39   __u32 s_journal_dev;                  /* journal device number  */
  40   __u32 s_journal_size;                 /* size of the journal on FS creation.  used to make sure they don't overflow it */
  41   __u32 s_journal_trans_max;            /* max number of blocks in a transaction.  */
  42   __u32 s_journal_magic;                /* random value made on fs creation */
  43   __u32 s_journal_max_batch;            /* max number of blocks to batch into a trans */
  44   __u32 s_journal_max_commit_age;       /* in seconds, how old can an async commit be */
  45   __u32 s_journal_max_trans_age;        /* in seconds, how old can a transaction be */
  46   __u16 s_blocksize;                    /* block size           */
  47   __u16 s_oid_maxsize;                  /* max size of object id array  */
  48   __u16 s_oid_cursize;                  /* current size of object id array */
  49   __u16 s_state;                        /* valid or error       */
  50   char s_magic[16];                     /* reiserfs magic string indicates that file system is reiserfs */
  51   __u16 s_tree_height;                  /* height of disk tree */
  52   __u16 s_bmap_nr;                      /* amount of bitmap blocks needed to address each block of file system */
  53   __u16 s_version;
  54   char s_unused[128];                   /* zero filled by mkreiserfs */
  55 };
  56
  57
  58 #define sb_root_block(sbp)            (__le32_to_cpu((sbp)->s_root_block))
  59 #define sb_journal_block(sbp)         (__le32_to_cpu((sbp)->s_journal_block))
  60 #define set_sb_journal_block(sbp,v)   ((sbp)->s_journal_block = __cpu_to_le32(v))
  61 #define sb_journal_size(sbp)          (__le32_to_cpu((sbp)->s_journal_size))
  62 #define sb_blocksize(sbp)             (__le16_to_cpu((sbp)->s_blocksize))
  63 #define set_sb_blocksize(sbp,v)       ((sbp)->s_blocksize = __cpu_to_le16(v))
  64 #define sb_version(sbp)               (__le16_to_cpu((sbp)->s_version))
  65 #define set_sb_version(sbp,v)         ((sbp)->s_version = __cpu_to_le16(v))
  66
  67
  68 #define REISERFS_MAX_SUPPORTED_VERSION 2
  69 #define REISERFS_SUPER_MAGIC_STRING "ReIsErFs"
  70 #define REISER2FS_SUPER_MAGIC_STRING "ReIsEr2Fs"
  71 #define REISER3FS_SUPER_MAGIC_STRING "ReIsEr3Fs"
  72
  73 #define MAX_HEIGHT 7
  74
  75 /* must be correct to keep the desc and commit structs at 4k */
  76 #define JOURNAL_TRANS_HALF 1018
  77
  78 /* first block written in a commit.  */
  79 struct reiserfs_journal_desc {
  80   __u32 j_trans_id;                     /* id of commit */
  81   __u32 j_len;                          /* length of commit. len +1 is the commit block */
  82   __u32 j_mount_id;                     /* mount id of this trans*/
  83   __u32 j_realblock[JOURNAL_TRANS_HALF]; /* real locations for the first blocks */
  84   char j_magic[12];
  85 };
  86
  87 /* last block written in a commit */
  88 struct reiserfs_journal_commit {
  89   __u32 j_trans_id;                     /* must match j_trans_id from the desc block */
  90   __u32 j_len;                  /* ditto */
  91   __u32 j_realblock[JOURNAL_TRANS_HALF]; /* real locations for the last blocks */
  92   char j_digest[16];                    /* md5 sum of all the blocks involved, including desc and commit. not used, kill it */
  93 };
  94
  95 /* this header block gets written whenever a transaction is considered
  96    fully flushed, and is more recent than the last fully flushed
  97    transaction.
  98    fully flushed means all the log blocks and all the real blocks are
  99    on disk, and this transaction does not need to be replayed.
 100 */
 101 struct reiserfs_journal_header {
 102   /* id of last fully flushed transaction */
 103   __u32 j_last_flush_trans_id;
 104   /* offset in the log of where to start replay after a crash */
 105   __u32 j_first_unflushed_offset;
 106   /* mount id to detect very old transactions */
 107   __u32 j_mount_id;
 108 };
 109
 110 /* magic string to find desc blocks in the journal */
 111 #define JOURNAL_DESC_MAGIC "ReIsErLB"
 112
 113
 114 /*
 115  * directories use this key as well as old files
 116  */
 117 struct offset_v1
 118 {
 119   /*
 120    * for regular files this is the offset to the first byte of the
 121    * body, contained in the object-item, as measured from the start of
 122    * the entire body of the object.
 123    *
 124    * for directory entries, k_offset consists of hash derived from
 125    * hashing the name and using few bits (23 or more) of the resulting
 126    * hash, and generation number that allows distinguishing names with
 127    * hash collisions. If number of collisions overflows generation
 128    * number, we return EEXIST.  High order bit is 0 always
 129    */
 130   __u32 k_offset;
 131   __u32 k_uniqueness;
 132 };
 133
 134 struct offset_v2 {
 135   /*
 136    * for regular files this is the offset to the first byte of the
 137    * body, contained in the object-item, as measured from the start of
 138    * the entire body of the object.
 139    *
 140    * for directory entries, k_offset consists of hash derived from
 141    * hashing the name and using few bits (23 or more) of the resulting
 142    * hash, and generation number that allows distinguishing names with
 143    * hash collisions. If number of collisions overflows generation
 144    * number, we return EEXIST.  High order bit is 0 always
 145    */
 146
 147 #if defined(__LITTLE_ENDIAN_BITFIELD)
 148             /* little endian version */
 149             __u64 k_offset:60;
 150             __u64 k_type: 4;
 151 #elif defined(__BIG_ENDIAN_BITFIELD)
 152             /* big endian version */
 153             __u64 k_type: 4;
 154             __u64 k_offset:60;
 155 #else
 156 #error "__LITTLE_ENDIAN_BITFIELD or __BIG_ENDIAN_BITFIELD must be defined"
 157 #endif
 158 } __attribute__ ((__packed__));
 159
 160 #define TYPE_MAXTYPE 3
 161 #define TYPE_ANY 15
 162
 163 #if (__BYTE_ORDER == __BIG_ENDIAN)
 164 typedef union {
 165     struct offset_v2 offset_v2;
 166     __u64 linear;
 167 } __attribute__ ((__packed__)) offset_v2_esafe_overlay;
 168
 169 static inline __u16 offset_v2_k_type( const struct offset_v2 *v2 )
 170 {
 171     offset_v2_esafe_overlay tmp = *(const offset_v2_esafe_overlay *)v2;
 172     tmp.linear = __le64_to_cpu( tmp.linear );
 173     return (tmp.offset_v2.k_type <= TYPE_MAXTYPE)?tmp.offset_v2.k_type:TYPE_ANY;
 174 }
 175
 176 static inline loff_t offset_v2_k_offset( const struct offset_v2 *v2 )
 177 {
 178     offset_v2_esafe_overlay tmp = *(const offset_v2_esafe_overlay *)v2;
 179     tmp.linear = __le64_to_cpu( tmp.linear );
 180     return tmp.offset_v2.k_offset;
 181 }
 182 #elif (__BYTE_ORDER == __LITTLE_ENDIAN)
 183 # define offset_v2_k_type(v2)           ((v2)->k_type)
 184 # define offset_v2_k_offset(v2)         ((v2)->k_offset)
 185 #else
 186 #error "__BYTE_ORDER must be __LITTLE_ENDIAN or __BIG_ENDIAN"
 187 #endif
 188
 189 struct key
 190 {
 191   /* packing locality: by default parent directory object id */
 192   __u32 k_dir_id;
 193   /* object identifier */
 194   __u32 k_objectid;
 195   /* the offset and node type (old and new form) */
 196   union
 197   {
 198     struct offset_v1 v1;
 199     struct offset_v2 v2;
 200   }
 201   u;
 202 };
 203
 204 #define KEY_SIZE (sizeof (struct key))
 205
 206 /* Header of a disk block.  More precisely, header of a formatted leaf
 207    or internal node, and not the header of an unformatted node. */
 208 struct block_head
 209 {
 210   __u16 blk_level;        /* Level of a block in the tree. */
 211   __u16 blk_nr_item;      /* Number of keys/items in a block. */
 212   __u16 blk_free_space;   /* Block free space in bytes. */
 213   struct key  blk_right_delim_key; /* Right delimiting key for this block (supported for leaf level nodes
 214                                       only) */
 215 };
 216 #define BLKH_SIZE (sizeof (struct block_head))
 217 #define DISK_LEAF_NODE_LEVEL  1 /* Leaf node level.                       */
 218
 219 struct item_head
 220 {
 221         /* Everything in the tree is found by searching for it based on
 222          * its key.*/
 223         struct key ih_key;
 224         union {
 225                 /* The free space in the last unformatted node of an
 226                    indirect item if this is an indirect item.  This
 227                    equals 0xFFFF iff this is a direct item or stat data
 228                    item. Note that the key, not this field, is used to
 229                    determine the item type, and thus which field this
 230                    union contains. */
 231                 __u16 ih_free_space;
 232                 /* Iff this is a directory item, this field equals the
 233                    number of directory entries in the directory item. */
 234                 __u16 ih_entry_count;
 235         } __attribute__ ((__packed__)) u;
 236         __u16 ih_item_len;           /* total size of the item body */
 237         __u16 ih_item_location;      /* an offset to the item body
 238                                       * within the block */
 239         __u16 ih_version;            /* 0 for all old items, 2 for new
 240                                         ones. Highest bit is set by fsck
 241                                         temporary, cleaned after all
 242                                         done */
 243 } __attribute__ ((__packed__));
 244
 245 /* size of item header     */
 246 #define IH_SIZE (sizeof (struct item_head))
 247
 248 #define ITEM_VERSION_1 0
 249 #define ITEM_VERSION_2 1
 250
 251 #define ih_version(ih)    (__le16_to_cpu((ih)->ih_version))
 252
 253 #define IH_KEY_OFFSET(ih) (ih_version(ih) == ITEM_VERSION_1 \
 254                            ? __le32_to_cpu((ih)->ih_key.u.v1.k_offset) \
 255                            : offset_v2_k_offset(&((ih)->ih_key.u.v2)))
 256
 257 #define IH_KEY_ISTYPE(ih, type) (ih_version(ih) == ITEM_VERSION_1 \
 258                                  ? __le32_to_cpu((ih)->ih_key.u.v1.k_uniqueness) == V1_##type \
 259                                  : offset_v2_k_type(&((ih)->ih_key.u.v2)) == V2_##type)
 260
 261 /***************************************************************************/
 262 /*                      DISK CHILD                                         */
 263 /***************************************************************************/
 264 /* Disk child pointer: The pointer from an internal node of the tree
 265    to a node that is on disk. */
 266 struct disk_child {
 267   __u32       dc_block_number;              /* Disk child's block number. */
 268   __u16       dc_size;                      /* Disk child's used space.   */
 269   __u16       dc_reserved;
 270 };
 271
 272 #define DC_SIZE (sizeof(struct disk_child))
 273 #define dc_block_number(dc_p)   (__le32_to_cpu((dc_p)->dc_block_number))
 274
 275
 276 /*
 277  * old stat data is 32 bytes long. We are going to distinguish new one by
 278  * different size
 279  */
 280 struct stat_data_v1
 281 {
 282     __u16 sd_mode;      /* file type, permissions */
 283     __u16 sd_nlink;     /* number of hard links */
 284     __u16 sd_uid;               /* owner */
 285     __u16 sd_gid;               /* group */
 286     __u32 sd_size;      /* file size */
 287     __u32 sd_atime;     /* time of last access */
 288     __u32 sd_mtime;     /* time file was last modified  */
 289     __u32 sd_ctime;     /* time inode (stat data) was last changed (except changes to sd_atime and sd_mtime) */
 290     union {
 291         __u32 sd_rdev;
 292         __u32 sd_blocks;        /* number of blocks file uses */
 293     } __attribute__ ((__packed__)) u;
 294     __u32 sd_first_direct_byte; /* first byte of file which is stored
 295                                    in a direct item: except that if it
 296                                    equals 1 it is a symlink and if it
 297                                    equals ~(__u32)0 there is no
 298                                    direct item.  The existence of this
 299                                    field really grates on me. Let's
 300                                    replace it with a macro based on
 301                                    sd_size and our tail suppression
 302                                    policy.  Someday.  -Hans */
 303 } __attribute__ ((__packed__));
 304
 305 #define stat_data_v1(ih)        (ih_version(ih) == ITEM_VERSION_1)
 306 #define sd_v1_mode(sdp)         ((sdp)->sd_mode)
 307 #define sd_v1_nlink(sdp)        (__le16_to_cpu((sdp)->sd_nlink))
 308 #define sd_v1_uid(sdp)          (__le16_to_cpu((sdp)->sd_uid))
 309 #define sd_v1_gid(sdp)          (__le16_to_cpu((sdp)->sd_gid))
 310 #define sd_v1_size(sdp)         (__le32_to_cpu((sdp)->sd_size))
 311 #define sd_v1_mtime(sdp)        (__le32_to_cpu((sdp)->sd_mtime))
 312
 313 /* Stat Data on disk (reiserfs version of UFS disk inode minus the
 314    address blocks) */
 315 struct stat_data {
 316     __u16 sd_mode;      /* file type, permissions */
 317     __u16 sd_attrs;     /* persistent inode flags */
 318     __u32 sd_nlink;     /* number of hard links */
 319     __u64 sd_size;      /* file size */
 320     __u32 sd_uid;               /* owner */
 321     __u32 sd_gid;               /* group */
 322     __u32 sd_atime;     /* time of last access */
 323     __u32 sd_mtime;     /* time file was last modified  */
 324     __u32 sd_ctime;     /* time inode (stat data) was last changed (except changes to sd_atime and sd_mtime) */
 325     __u32 sd_blocks;
 326     union {
 327         __u32 sd_rdev;
 328         __u32 sd_generation;
 329       /*__u32 sd_first_direct_byte; */
 330       /* first byte of file which is stored in a
 331                                        direct item: except that if it equals 1
 332                                        it is a symlink and if it equals
 333                                        ~(__u32)0 there is no direct item.  The
 334                                        existence of this field really grates
 335                                        on me. Let's replace it with a macro
 336                                        based on sd_size and our tail
 337                                        suppression policy? */
 338   } __attribute__ ((__packed__)) u;
 339 } __attribute__ ((__packed__));
 340
 341 #define stat_data_v2(ih)        (ih_version(ih) == ITEM_VERSION_2)
 342 #define sd_v2_mode(sdp)         (__le16_to_cpu((sdp)->sd_mode))
 343 #define sd_v2_nlink(sdp)        (__le32_to_cpu((sdp)->sd_nlink))
 344 #define sd_v2_size(sdp)         (__le64_to_cpu((sdp)->sd_size))
 345 #define sd_v2_uid(sdp)          (__le32_to_cpu((sdp)->sd_uid))
 346 #define sd_v2_gid(sdp)          (__le32_to_cpu((sdp)->sd_gid))
 347 #define sd_v2_mtime(sdp)        (__le32_to_cpu((sdp)->sd_mtime))
 348
 349 #define sd_mode(sdp)         (__le16_to_cpu((sdp)->sd_mode))
 350 #define sd_size(sdp)         (__le32_to_cpu((sdp)->sd_size))
 351 #define sd_size_hi(sdp)      (__le32_to_cpu((sdp)->sd_size_hi))
 352
 353 struct reiserfs_de_head
 354 {
 355   __u32 deh_offset;  /* third component of the directory entry key */
 356   __u32 deh_dir_id;  /* objectid of the parent directory of the
 357                         object, that is referenced by directory entry */
 358   __u32 deh_objectid;/* objectid of the object, that is referenced by
 359                         directory entry */
 360   __u16 deh_location;/* offset of name in the whole item */
 361   __u16 deh_state;   /* whether 1) entry contains stat data (for
 362                         future), and 2) whether entry is hidden
 363                         (unlinked) */
 364 };
 365
 366 #define DEH_SIZE (sizeof (struct reiserfs_de_head))
 367 #define deh_offset(p_deh)         (__le32_to_cpu((p_deh)->deh_offset))
 368 #define deh_dir_id(p_deh)         (__le32_to_cpu((p_deh)->deh_dir_id))
 369 #define deh_objectid(p_deh)       (__le32_to_cpu((p_deh)->deh_objectid))
 370 #define deh_location(p_deh)       (__le16_to_cpu((p_deh)->deh_location))
 371 #define deh_state(p_deh)          (__le16_to_cpu((p_deh)->deh_state))
 372
 373
 374 #define DEH_Statdata (1 << 0)                   /* not used now */
 375 #define DEH_Visible  (1 << 2)
 376
 377 #define SD_OFFSET  0
 378 #define SD_UNIQUENESS 0
 379 #define DOT_OFFSET 1
 380 #define DOT_DOT_OFFSET 2
 381 #define DIRENTRY_UNIQUENESS 500
 382
 383 #define V1_TYPE_STAT_DATA 0x0
 384 #define V1_TYPE_DIRECT 0xffffffff
 385 #define V1_TYPE_INDIRECT 0xfffffffe
 386 #define V1_TYPE_DIRECTORY_MAX 0xfffffffd
 387 #define V2_TYPE_STAT_DATA 0
 388 #define V2_TYPE_INDIRECT 1
 389 #define V2_TYPE_DIRECT 2
 390 #define V2_TYPE_DIRENTRY 3
 391
 392 #define REISERFS_ROOT_OBJECTID 2
 393 #define REISERFS_ROOT_PARENT_OBJECTID 1
 394 #define REISERFS_DISK_OFFSET_IN_BYTES (64 * 1024)
 395 /* the spot for the super in versions 3.5 - 3.5.11 (inclusive) */
 396 #define REISERFS_OLD_DISK_OFFSET_IN_BYTES (8 * 1024)
 397 #define REISERFS_OLD_BLOCKSIZE 4096
 398
 399 #define S_ISREG(mode) (((mode) & 0170000) == 0100000)
 400 #define S_ISDIR(mode) (((mode) & 0170000) == 0040000)
 401 #define S_ISLNK(mode) (((mode) & 0170000) == 0120000)
 402
 403 #define PATH_MAX       1024     /* include/linux/limits.h */
 404 #define MAX_LINK_COUNT    5     /* number of symbolic links to follow */
 405
 406 /* The size of the node cache */
 407 #define FSYSREISER_CACHE_SIZE 24*1024
 408 #define FSYSREISER_MIN_BLOCKSIZE SECTOR_SIZE
 409 #define FSYSREISER_MAX_BLOCKSIZE FSYSREISER_CACHE_SIZE / 3
 410
 411 /* Info about currently opened file */
 412 struct fsys_reiser_fileinfo
 413 {
 414   __u32 k_dir_id;
 415   __u32 k_objectid;
 416 };
 417
 418 /* In memory info about the currently mounted filesystem */
 419 struct fsys_reiser_info
 420 {
 421   /* The last read item head */
 422   struct item_head *current_ih;
 423   /* The last read item */
 424   char *current_item;
 425   /* The information for the currently opened file */
 426   struct fsys_reiser_fileinfo fileinfo;
 427   /* The start of the journal */
 428   __u32 journal_block;
 429   /* The size of the journal */
 430   __u32 journal_block_count;
 431   /* The first valid descriptor block in journal
 432      (relative to journal_block) */
 433   __u32 journal_first_desc;
 434
 435   /* The ReiserFS version. */
 436   __u16 version;
 437   /* The current depth of the reiser tree. */
 438   __u16 tree_depth;
 439   /* SECTOR_SIZE << blocksize_shift == blocksize. */
 440   __u8  blocksize_shift;
 441   /* 1 << full_blocksize_shift == blocksize. */
 442   __u8  fullblocksize_shift;
 443   /* The reiserfs block size  (must be a power of 2) */
 444   __u16 blocksize;
 445   /* The number of cached tree nodes */
 446   __u16 cached_slots;
 447   /* The number of valid transactions in journal */
 448   __u16 journal_transactions;
 449
 450   unsigned int blocks[MAX_HEIGHT];
 451   unsigned int next_key_nr[MAX_HEIGHT];
 452 };
 453
 454 /* The cached s+tree blocks in FSYS_BUF,  see below
 455  * for a more detailed description.
 456  */
 457 #define ROOT     ((char *) ((int) FSYS_BUF))
 458 #define CACHE(i) (ROOT + ((i) << INFO->fullblocksize_shift))
 459 #define LEAF     CACHE (DISK_LEAF_NODE_LEVEL)
 460
 461 #define BLOCKHEAD(cache) ((struct block_head *) cache)
 462 #define ITEMHEAD         ((struct item_head  *) ((int) LEAF + BLKH_SIZE))
 463 #define KEY(cache)       ((struct key        *) ((int) cache + BLKH_SIZE))
 464 #define DC(cache)        ((struct disk_child *) \
 465                           ((int) cache + BLKH_SIZE + KEY_SIZE * nr_item))
 466 /* The fsys_reiser_info block.
 467  */
 468 #define INFO \
 469     ((struct fsys_reiser_info *) ((int) FSYS_BUF + FSYSREISER_CACHE_SIZE))
 470 /*
 471  * The journal cache.  For each transaction it contains the number of
 472  * blocks followed by the real block numbers of this transaction.
 473  *
 474  * If the block numbers of some transaction won't fit in this space,
 475  * this list is stopped with a 0xffffffff marker and the remaining
 476  * uncommitted transactions aren't cached.
 477  */
 478 #define JOURNAL_START    ((__u32 *) (INFO + 1))
 479 #define JOURNAL_END      ((__u32 *) (FSYS_BUF + FSYS_BUFLEN))
 480
 481
 482 static __inline__ unsigned long
 483 log2 (unsigned long word)
 484 {
 485 #ifdef __I386__
 486   __asm__ ("bsfl %1,%0"
 487            : "=r" (word)
 488            : "r" (word));
 489   return word;
 490 #else
 491   int i;
 492
 493   for(i=0; i<(8*sizeof(word)); i++)
 494     if ((1<<i) & word)
 495       return i;
 496
 497   return 0;
 498 #endif
 499 }
 500
 501 static __inline__ int
 502 is_power_of_two (unsigned long word)
 503 {
 504   return (word & -word) == word;
 505 }
 506
 507 extern const char *bb_mode_string(int mode);
 508 extern int reiserfs_devread (int sector, int byte_offset, int byte_len, char *buf);