fs/xfs/xfs_btree.h

   1 /*
   2  * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #ifndef __XFS_BTREE_H__
  19 #define __XFS_BTREE_H__
  20
  21 struct xfs_buf;
  22 struct xfs_bmap_free;
  23 struct xfs_inode;
  24 struct xfs_mount;
  25 struct xfs_trans;
  26
  27 extern kmem_zone_t      *xfs_btree_cur_zone;
  28
  29 /*
  30  * This nonsense is to make -wlint happy.
  31  */
  32 #define XFS_LOOKUP_EQ   ((xfs_lookup_t)XFS_LOOKUP_EQi)
  33 #define XFS_LOOKUP_LE   ((xfs_lookup_t)XFS_LOOKUP_LEi)
  34 #define XFS_LOOKUP_GE   ((xfs_lookup_t)XFS_LOOKUP_GEi)
  35
  36 #define XFS_BTNUM_BNO   ((xfs_btnum_t)XFS_BTNUM_BNOi)
  37 #define XFS_BTNUM_CNT   ((xfs_btnum_t)XFS_BTNUM_CNTi)
  38 #define XFS_BTNUM_BMAP  ((xfs_btnum_t)XFS_BTNUM_BMAPi)
  39 #define XFS_BTNUM_INO   ((xfs_btnum_t)XFS_BTNUM_INOi)
  40
  41 /*
  42  * Generic btree header.
  43  *
  44  * This is a combination of the actual format used on disk for short and long
  45  * format btrees.  The first three fields are shared by both format, but the
  46  * pointers are different and should be used with care.
  47  *
  48  * To get the size of the actual short or long form headers please use the size
  49  * macros below.  Never use sizeof(xfs_btree_block).
  50  *
  51  * The blkno, crc, lsn, owner and uuid fields are only available in filesystems
  52  * with the crc feature bit, and all accesses to them must be conditional on
  53  * that flag.
  54  */
  55 struct xfs_btree_block {
  56         __be32          bb_magic;       /* magic number for block type */
  57         __be16          bb_level;       /* 0 is a leaf */
  58         __be16          bb_numrecs;     /* current # of data records */
  59         union {
  60                 struct {
  61                         __be32          bb_leftsib;
  62                         __be32          bb_rightsib;
  63
  64                         __be64          bb_blkno;
  65                         __be64          bb_lsn;
  66                         uuid_t          bb_uuid;
  67                         __be32          bb_owner;
  68                         __le32          bb_crc;
  69                 } s;                    /* short form pointers */
  70                 struct  {
  71                         __be64          bb_leftsib;
  72                         __be64          bb_rightsib;
  73
  74                         __be64          bb_blkno;
  75                         __be64          bb_lsn;
  76                         uuid_t          bb_uuid;
  77                         __be64          bb_owner;
  78                         __le32          bb_crc;
  79                         __be32          bb_pad; /* padding for alignment */
  80                 } l;                    /* long form pointers */
  81         } bb_u;                         /* rest */
  82 };
  83
  84 #define XFS_BTREE_SBLOCK_LEN    16      /* size of a short form block */
  85 #define XFS_BTREE_LBLOCK_LEN    24      /* size of a long form block */
  86
  87 /* sizes of CRC enabled btree blocks */
  88 #define XFS_BTREE_SBLOCK_CRC_LEN        (XFS_BTREE_SBLOCK_LEN + 40)
  89 #define XFS_BTREE_LBLOCK_CRC_LEN        (XFS_BTREE_LBLOCK_LEN + 48)
  90
  91
  92 #define XFS_BTREE_SBLOCK_CRC_OFF \
  93         offsetof(struct xfs_btree_block, bb_u.s.bb_crc)
  94 #define XFS_BTREE_LBLOCK_CRC_OFF \
  95         offsetof(struct xfs_btree_block, bb_u.l.bb_crc)
  96
  97
  98 /*
  99  * Generic key, ptr and record wrapper structures.
 100  *
 101  * These are disk format structures, and are converted where necessary
 102  * by the btree specific code that needs to interpret them.
 103  */
 104 union xfs_btree_ptr {
 105         __be32                  s;      /* short form ptr */
 106         __be64                  l;      /* long form ptr */
 107 };
 108
 109 union xfs_btree_key {
 110         xfs_bmbt_key_t          bmbt;
 111         xfs_bmdr_key_t          bmbr;   /* bmbt root block */
 112         xfs_alloc_key_t         alloc;
 113         xfs_inobt_key_t         inobt;
 114 };
 115
 116 union xfs_btree_rec {
 117         xfs_bmbt_rec_t          bmbt;
 118         xfs_bmdr_rec_t          bmbr;   /* bmbt root block */
 119         xfs_alloc_rec_t         alloc;
 120         xfs_inobt_rec_t         inobt;
 121 };
 122
 123 /*
 124  * For logging record fields.
 125  */
 126 #define XFS_BB_MAGIC            0x01
 127 #define XFS_BB_LEVEL            0x02
 128 #define XFS_BB_NUMRECS          0x04
 129 #define XFS_BB_LEFTSIB          0x08
 130 #define XFS_BB_RIGHTSIB         0x10
 131 #define XFS_BB_BLKNO            0x20
 132 #define XFS_BB_NUM_BITS         5
 133 #define XFS_BB_ALL_BITS         ((1 << XFS_BB_NUM_BITS) - 1)
 134 #define XFS_BB_NUM_BITS_CRC     8
 135 #define XFS_BB_ALL_BITS_CRC     ((1 << XFS_BB_NUM_BITS_CRC) - 1)
 136
 137 /*
 138  * Generic stats interface
 139  */
 140 #define __XFS_BTREE_STATS_INC(type, stat) \
 141         XFS_STATS_INC(xs_ ## type ## _2_ ## stat)
 142 #define XFS_BTREE_STATS_INC(cur, stat)  \
 143 do {    \
 144         switch (cur->bc_btnum) {  \
 145         case XFS_BTNUM_BNO: __XFS_BTREE_STATS_INC(abtb, stat); break;   \
 146         case XFS_BTNUM_CNT: __XFS_BTREE_STATS_INC(abtc, stat); break;   \
 147         case XFS_BTNUM_BMAP: __XFS_BTREE_STATS_INC(bmbt, stat); break;  \
 148         case XFS_BTNUM_INO: __XFS_BTREE_STATS_INC(ibt, stat); break;    \
 149         case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break;       \
 150         }       \
 151 } while (0)
 152
 153 #define __XFS_BTREE_STATS_ADD(type, stat, val) \
 154         XFS_STATS_ADD(xs_ ## type ## _2_ ## stat, val)
 155 #define XFS_BTREE_STATS_ADD(cur, stat, val)  \
 156 do {    \
 157         switch (cur->bc_btnum) {  \
 158         case XFS_BTNUM_BNO: __XFS_BTREE_STATS_ADD(abtb, stat, val); break; \
 159         case XFS_BTNUM_CNT: __XFS_BTREE_STATS_ADD(abtc, stat, val); break; \
 160         case XFS_BTNUM_BMAP: __XFS_BTREE_STATS_ADD(bmbt, stat, val); break; \
 161         case XFS_BTNUM_INO: __XFS_BTREE_STATS_ADD(ibt, stat, val); break; \
 162         case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break;       \
 163         }       \
 164 } while (0)
 165
 166 #define XFS_BTREE_MAXLEVELS     8       /* max of all btrees */
 167
 168 struct xfs_btree_ops {
 169         /* size of the key and record structures */
 170         size_t  key_len;
 171         size_t  rec_len;
 172
 173         /* cursor operations */
 174         struct xfs_btree_cur *(*dup_cursor)(struct xfs_btree_cur *);
 175         void    (*update_cursor)(struct xfs_btree_cur *src,
 176                                  struct xfs_btree_cur *dst);
 177
 178         /* update btree root pointer */
 179         void    (*set_root)(struct xfs_btree_cur *cur,
 180                             union xfs_btree_ptr *nptr, int level_change);
 181
 182         /* block allocation / freeing */
 183         int     (*alloc_block)(struct xfs_btree_cur *cur,
 184                                union xfs_btree_ptr *start_bno,
 185                                union xfs_btree_ptr *new_bno,
 186                                int length, int *stat);
 187         int     (*free_block)(struct xfs_btree_cur *cur, struct xfs_buf *bp);
 188
 189         /* update last record information */
 190         void    (*update_lastrec)(struct xfs_btree_cur *cur,
 191                                   struct xfs_btree_block *block,
 192                                   union xfs_btree_rec *rec,
 193                                   int ptr, int reason);
 194
 195         /* records in block/level */
 196         int     (*get_minrecs)(struct xfs_btree_cur *cur, int level);
 197         int     (*get_maxrecs)(struct xfs_btree_cur *cur, int level);
 198
 199         /* records on disk.  Matter for the root in inode case. */
 200         int     (*get_dmaxrecs)(struct xfs_btree_cur *cur, int level);
 201
 202         /* init values of btree structures */
 203         void    (*init_key_from_rec)(union xfs_btree_key *key,
 204                                      union xfs_btree_rec *rec);
 205         void    (*init_rec_from_key)(union xfs_btree_key *key,
 206                                      union xfs_btree_rec *rec);
 207         void    (*init_rec_from_cur)(struct xfs_btree_cur *cur,
 208                                      union xfs_btree_rec *rec);
 209         void    (*init_ptr_from_cur)(struct xfs_btree_cur *cur,
 210                                      union xfs_btree_ptr *ptr);
 211
 212         /* difference between key value and cursor value */
 213         __int64_t (*key_diff)(struct xfs_btree_cur *cur,
 214                               union xfs_btree_key *key);
 215
 216         const struct xfs_buf_ops        *buf_ops;
 217
 218 #if defined(DEBUG) || defined(XFS_WARN)
 219         /* check that k1 is lower than k2 */
 220         int     (*keys_inorder)(struct xfs_btree_cur *cur,
 221                                 union xfs_btree_key *k1,
 222                                 union xfs_btree_key *k2);
 223
 224         /* check that r1 is lower than r2 */
 225         int     (*recs_inorder)(struct xfs_btree_cur *cur,
 226                                 union xfs_btree_rec *r1,
 227                                 union xfs_btree_rec *r2);
 228 #endif
 229 };
 230
 231 /*
 232  * Reasons for the update_lastrec method to be called.
 233  */
 234 #define LASTREC_UPDATE  0
 235 #define LASTREC_INSREC  1
 236 #define LASTREC_DELREC  2
 237
 238
 239 /*
 240  * Btree cursor structure.
 241  * This collects all information needed by the btree code in one place.
 242  */
 243 typedef struct xfs_btree_cur
 244 {
 245         struct xfs_trans        *bc_tp; /* transaction we're in, if any */
 246         struct xfs_mount        *bc_mp; /* file system mount struct */
 247         const struct xfs_btree_ops *bc_ops;
 248         uint                    bc_flags; /* btree features - below */
 249         union {
 250                 xfs_alloc_rec_incore_t  a;
 251                 xfs_bmbt_irec_t         b;
 252                 xfs_inobt_rec_incore_t  i;
 253         }               bc_rec;         /* current insert/search record value */
 254         struct xfs_buf  *bc_bufs[XFS_BTREE_MAXLEVELS];  /* buf ptr per level */
 255         int             bc_ptrs[XFS_BTREE_MAXLEVELS];   /* key/record # */
 256         __uint8_t       bc_ra[XFS_BTREE_MAXLEVELS];     /* readahead bits */
 257 #define XFS_BTCUR_LEFTRA        1       /* left sibling has been read-ahead */
 258 #define XFS_BTCUR_RIGHTRA       2       /* right sibling has been read-ahead */
 259         __uint8_t       bc_nlevels;     /* number of levels in the tree */
 260         __uint8_t       bc_blocklog;    /* log2(blocksize) of btree blocks */
 261         xfs_btnum_t     bc_btnum;       /* identifies which btree type */
 262         union {
 263                 struct {                        /* needed for BNO, CNT, INO */
 264                         struct xfs_buf  *agbp;  /* agf/agi buffer pointer */
 265                         xfs_agnumber_t  agno;   /* ag number */
 266                 } a;
 267                 struct {                        /* needed for BMAP */
 268                         struct xfs_inode *ip;   /* pointer to our inode */
 269                         struct xfs_bmap_free *flist;    /* list to free after */
 270                         xfs_fsblock_t   firstblock;     /* 1st blk allocated */
 271                         int             allocated;      /* count of alloced */
 272                         short           forksize;       /* fork's inode space */
 273                         char            whichfork;      /* data or attr fork */
 274                         char            flags;          /* flags */
 275 #define XFS_BTCUR_BPRV_WASDEL   1                       /* was delayed */
 276                 } b;
 277         }               bc_private;     /* per-btree type data */
 278 } xfs_btree_cur_t;
 279
 280 /* cursor flags */
 281 #define XFS_BTREE_LONG_PTRS             (1<<0)  /* pointers are 64bits long */
 282 #define XFS_BTREE_ROOT_IN_INODE         (1<<1)  /* root may be variable size */
 283 #define XFS_BTREE_LASTREC_UPDATE        (1<<2)  /* track last rec externally */
 284 #define XFS_BTREE_CRC_BLOCKS            (1<<3)  /* uses extended btree blocks */
 285
 286
 287 #define XFS_BTREE_NOERROR       0
 288 #define XFS_BTREE_ERROR         1
 289
 290 /*
 291  * Convert from buffer to btree block header.
 292  */
 293 #define XFS_BUF_TO_BLOCK(bp)    ((struct xfs_btree_block *)((bp)->b_addr))
 294
 295
 296 /*
 297  * Check that block header is ok.
 298  */
 299 int
 300 xfs_btree_check_block(
 301         struct xfs_btree_cur    *cur,   /* btree cursor */
 302         struct xfs_btree_block  *block, /* generic btree block pointer */
 303         int                     level,  /* level of the btree block */
 304         struct xfs_buf          *bp);   /* buffer containing block, if any */
 305
 306 /*
 307  * Check that (long) pointer is ok.
 308  */
 309 int                                     /* error (0 or EFSCORRUPTED) */
 310 xfs_btree_check_lptr(
 311         struct xfs_btree_cur    *cur,   /* btree cursor */
 312         xfs_dfsbno_t            ptr,    /* btree block disk address */
 313         int                     level); /* btree block level */
 314
 315 /*
 316  * Delete the btree cursor.
 317  */
 318 void
 319 xfs_btree_del_cursor(
 320         xfs_btree_cur_t         *cur,   /* btree cursor */
 321         int                     error); /* del because of error */
 322
 323 /*
 324  * Duplicate the btree cursor.
 325  * Allocate a new one, copy the record, re-get the buffers.
 326  */
 327 int                                     /* error */
 328 xfs_btree_dup_cursor(
 329         xfs_btree_cur_t         *cur,   /* input cursor */
 330         xfs_btree_cur_t         **ncur);/* output cursor */
 331
 332 /*
 333  * Get a buffer for the block, return it with no data read.
 334  * Long-form addressing.
 335  */
 336 struct xfs_buf *                                /* buffer for fsbno */
 337 xfs_btree_get_bufl(
 338         struct xfs_mount        *mp,    /* file system mount point */
 339         struct xfs_trans        *tp,    /* transaction pointer */
 340         xfs_fsblock_t           fsbno,  /* file system block number */
 341         uint                    lock);  /* lock flags for get_buf */
 342
 343 /*
 344  * Get a buffer for the block, return it with no data read.
 345  * Short-form addressing.
 346  */
 347 struct xfs_buf *                                /* buffer for agno/agbno */
 348 xfs_btree_get_bufs(
 349         struct xfs_mount        *mp,    /* file system mount point */
 350         struct xfs_trans        *tp,    /* transaction pointer */
 351         xfs_agnumber_t          agno,   /* allocation group number */
 352         xfs_agblock_t           agbno,  /* allocation group block number */
 353         uint                    lock);  /* lock flags for get_buf */
 354
 355 /*
 356  * Check for the cursor referring to the last block at the given level.
 357  */
 358 int                                     /* 1=is last block, 0=not last block */
 359 xfs_btree_islastblock(
 360         xfs_btree_cur_t         *cur,   /* btree cursor */
 361         int                     level); /* level to check */
 362
 363 /*
 364  * Compute first and last byte offsets for the fields given.
 365  * Interprets the offsets table, which contains struct field offsets.
 366  */
 367 void
 368 xfs_btree_offsets(
 369         __int64_t               fields, /* bitmask of fields */
 370         const short             *offsets,/* table of field offsets */
 371         int                     nbits,  /* number of bits to inspect */
 372         int                     *first, /* output: first byte offset */
 373         int                     *last); /* output: last byte offset */
 374
 375 /*
 376  * Get a buffer for the block, return it read in.
 377  * Long-form addressing.
 378  */
 379 int                                     /* error */
 380 xfs_btree_read_bufl(
 381         struct xfs_mount        *mp,    /* file system mount point */
 382         struct xfs_trans        *tp,    /* transaction pointer */
 383         xfs_fsblock_t           fsbno,  /* file system block number */
 384         uint                    lock,   /* lock flags for read_buf */
 385         struct xfs_buf          **bpp,  /* buffer for fsbno */
 386         int                     refval, /* ref count value for buffer */
 387         const struct xfs_buf_ops *ops);
 388
 389 /*
 390  * Read-ahead the block, don't wait for it, don't return a buffer.
 391  * Long-form addressing.
 392  */
 393 void                                    /* error */
 394 xfs_btree_reada_bufl(
 395         struct xfs_mount        *mp,    /* file system mount point */
 396         xfs_fsblock_t           fsbno,  /* file system block number */
 397         xfs_extlen_t            count,  /* count of filesystem blocks */
 398         const struct xfs_buf_ops *ops);
 399
 400 /*
 401  * Read-ahead the block, don't wait for it, don't return a buffer.
 402  * Short-form addressing.
 403  */
 404 void                                    /* error */
 405 xfs_btree_reada_bufs(
 406         struct xfs_mount        *mp,    /* file system mount point */
 407         xfs_agnumber_t          agno,   /* allocation group number */
 408         xfs_agblock_t           agbno,  /* allocation group block number */
 409         xfs_extlen_t            count,  /* count of filesystem blocks */
 410         const struct xfs_buf_ops *ops);
 411
 412 /*
 413  * Initialise a new btree block header
 414  */
 415 void
 416 xfs_btree_init_block(
 417         struct xfs_mount *mp,
 418         struct xfs_buf  *bp,
 419         __u32           magic,
 420         __u16           level,
 421         __u16           numrecs,
 422         __u64           owner,
 423         unsigned int    flags);
 424
 425 void
 426 xfs_btree_init_block_int(
 427         struct xfs_mount        *mp,
 428         struct xfs_btree_block  *buf,
 429         xfs_daddr_t             blkno,
 430         __u32                   magic,
 431         __u16                   level,
 432         __u16                   numrecs,
 433         __u64                   owner,
 434         unsigned int            flags);
 435
 436 /*
 437  * Common btree core entry points.
 438  */
 439 int xfs_btree_increment(struct xfs_btree_cur *, int, int *);
 440 int xfs_btree_decrement(struct xfs_btree_cur *, int, int *);
 441 int xfs_btree_lookup(struct xfs_btree_cur *, xfs_lookup_t, int *);
 442 int xfs_btree_update(struct xfs_btree_cur *, union xfs_btree_rec *);
 443 int xfs_btree_new_iroot(struct xfs_btree_cur *, int *, int *);
 444 int xfs_btree_insert(struct xfs_btree_cur *, int *);
 445 int xfs_btree_delete(struct xfs_btree_cur *, int *);
 446 int xfs_btree_get_rec(struct xfs_btree_cur *, union xfs_btree_rec **, int *);
 447
 448 /*
 449  * btree block CRC helpers
 450  */
 451 void xfs_btree_lblock_calc_crc(struct xfs_buf *);
 452 bool xfs_btree_lblock_verify_crc(struct xfs_buf *);
 453 void xfs_btree_sblock_calc_crc(struct xfs_buf *);
 454 bool xfs_btree_sblock_verify_crc(struct xfs_buf *);
 455
 456 /*
 457  * Internal btree helpers also used by xfs_bmap.c.
 458  */
 459 void xfs_btree_log_block(struct xfs_btree_cur *, struct xfs_buf *, int);
 460 void xfs_btree_log_recs(struct xfs_btree_cur *, struct xfs_buf *, int, int);
 461
 462 /*
 463  * Helpers.
 464  */
 465 static inline int xfs_btree_get_numrecs(struct xfs_btree_block *block)
 466 {
 467         return be16_to_cpu(block->bb_numrecs);
 468 }
 469
 470 static inline void xfs_btree_set_numrecs(struct xfs_btree_block *block,
 471                 __uint16_t numrecs)
 472 {
 473         block->bb_numrecs = cpu_to_be16(numrecs);
 474 }
 475
 476 static inline int xfs_btree_get_level(struct xfs_btree_block *block)
 477 {
 478         return be16_to_cpu(block->bb_level);
 479 }
 480
 481
 482 /*
 483  * Min and max functions for extlen, agblock, fileoff, and filblks types.
 484  */
 485 #define XFS_EXTLEN_MIN(a,b)     min_t(xfs_extlen_t, (a), (b))
 486 #define XFS_EXTLEN_MAX(a,b)     max_t(xfs_extlen_t, (a), (b))
 487 #define XFS_AGBLOCK_MIN(a,b)    min_t(xfs_agblock_t, (a), (b))
 488 #define XFS_AGBLOCK_MAX(a,b)    max_t(xfs_agblock_t, (a), (b))
 489 #define XFS_FILEOFF_MIN(a,b)    min_t(xfs_fileoff_t, (a), (b))
 490 #define XFS_FILEOFF_MAX(a,b)    max_t(xfs_fileoff_t, (a), (b))
 491 #define XFS_FILBLKS_MIN(a,b)    min_t(xfs_filblks_t, (a), (b))
 492 #define XFS_FILBLKS_MAX(a,b)    max_t(xfs_filblks_t, (a), (b))
 493
 494 #define XFS_FSB_SANITY_CHECK(mp,fsb)    \
 495         (XFS_FSB_TO_AGNO(mp, fsb) < mp->m_sb.sb_agcount && \
 496                 XFS_FSB_TO_AGBNO(mp, fsb) < mp->m_sb.sb_agblocks)
 497
 498 /*
 499  * Trace hooks.  Currently not implemented as they need to be ported
 500  * over to the generic tracing functionality, which is some effort.
 501  *
 502  * i,j = integer (32 bit)
 503  * b = btree block buffer (xfs_buf_t)
 504  * p = btree ptr
 505  * r = btree record
 506  * k = btree key
 507  */
 508 #define XFS_BTREE_TRACE_ARGBI(c, b, i)
 509 #define XFS_BTREE_TRACE_ARGBII(c, b, i, j)
 510 #define XFS_BTREE_TRACE_ARGI(c, i)
 511 #define XFS_BTREE_TRACE_ARGIPK(c, i, p, s)
 512 #define XFS_BTREE_TRACE_ARGIPR(c, i, p, r)
 513 #define XFS_BTREE_TRACE_ARGIK(c, i, k)
 514 #define XFS_BTREE_TRACE_ARGR(c, r)
 515 #define XFS_BTREE_TRACE_CURSOR(c, t)
 516
 517 #endif  /* __XFS_BTREE_H__ */