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
  46  * the 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
  49  * the size macros below.  Never use sizeof(xfs_btree_block).
  50  */
  51 struct xfs_btree_block {
  52         __be32          bb_magic;       /* magic number for block type */
  53         __be16          bb_level;       /* 0 is a leaf */
  54         __be16          bb_numrecs;     /* current # of data records */
  55         union {
  56                 struct {
  57                         __be32          bb_leftsib;
  58                         __be32          bb_rightsib;
  59                 } s;                    /* short form pointers */
  60                 struct  {
  61                         __be64          bb_leftsib;
  62                         __be64          bb_rightsib;
  63                 } l;                    /* long form pointers */
  64         } bb_u;                         /* rest */
  65 };
  66
  67 #define XFS_BTREE_SBLOCK_LEN    16      /* size of a short form block */
  68 #define XFS_BTREE_LBLOCK_LEN    24      /* size of a long form block */
  69
  70
  71 /*
  72  * Generic key, ptr and record wrapper structures.
  73  *
  74  * These are disk format structures, and are converted where necessary
  75  * by the btree specific code that needs to interpret them.
  76  */
  77 union xfs_btree_ptr {
  78         __be32                  s;      /* short form ptr */
  79         __be64                  l;      /* long form ptr */
  80 };
  81
  82 union xfs_btree_key {
  83         xfs_bmbt_key_t          bmbt;
  84         xfs_bmdr_key_t          bmbr;   /* bmbt root block */
  85         xfs_alloc_key_t         alloc;
  86         xfs_inobt_key_t         inobt;
  87 };
  88
  89 union xfs_btree_rec {
  90         xfs_bmbt_rec_t          bmbt;
  91         xfs_bmdr_rec_t          bmbr;   /* bmbt root block */
  92         xfs_alloc_rec_t         alloc;
  93         xfs_inobt_rec_t         inobt;
  94 };
  95
  96 /*
  97  * For logging record fields.
  98  */
  99 #define XFS_BB_MAGIC            0x01
 100 #define XFS_BB_LEVEL            0x02
 101 #define XFS_BB_NUMRECS          0x04
 102 #define XFS_BB_LEFTSIB          0x08
 103 #define XFS_BB_RIGHTSIB         0x10
 104 #define XFS_BB_NUM_BITS         5
 105 #define XFS_BB_ALL_BITS         ((1 << XFS_BB_NUM_BITS) - 1)
 106
 107 /*
 108  * Magic numbers for btree blocks.
 109  */
 110 extern const __uint32_t xfs_magics[];
 111
 112 /*
 113  * Generic stats interface
 114  */
 115 #define __XFS_BTREE_STATS_INC(type, stat) \
 116         XFS_STATS_INC(xs_ ## type ## _2_ ## stat)
 117 #define XFS_BTREE_STATS_INC(cur, stat)  \
 118 do {    \
 119         switch (cur->bc_btnum) {  \
 120         case XFS_BTNUM_BNO: __XFS_BTREE_STATS_INC(abtb, stat); break;   \
 121         case XFS_BTNUM_CNT: __XFS_BTREE_STATS_INC(abtc, stat); break;   \
 122         case XFS_BTNUM_BMAP: __XFS_BTREE_STATS_INC(bmbt, stat); break;  \
 123         case XFS_BTNUM_INO: __XFS_BTREE_STATS_INC(ibt, stat); break;    \
 124         case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break;       \
 125         }       \
 126 } while (0)
 127
 128 #define __XFS_BTREE_STATS_ADD(type, stat, val) \
 129         XFS_STATS_ADD(xs_ ## type ## _2_ ## stat, val)
 130 #define XFS_BTREE_STATS_ADD(cur, stat, val)  \
 131 do {    \
 132         switch (cur->bc_btnum) {  \
 133         case XFS_BTNUM_BNO: __XFS_BTREE_STATS_ADD(abtb, stat, val); break; \
 134         case XFS_BTNUM_CNT: __XFS_BTREE_STATS_ADD(abtc, stat, val); break; \
 135         case XFS_BTNUM_BMAP: __XFS_BTREE_STATS_ADD(bmbt, stat, val); break; \
 136         case XFS_BTNUM_INO: __XFS_BTREE_STATS_ADD(ibt, stat, val); break; \
 137         case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break;       \
 138         }       \
 139 } while (0)
 140
 141 #define XFS_BTREE_MAXLEVELS     8       /* max of all btrees */
 142
 143 struct xfs_btree_ops {
 144         /* size of the key and record structures */
 145         size_t  key_len;
 146         size_t  rec_len;
 147
 148         /* cursor operations */
 149         struct xfs_btree_cur *(*dup_cursor)(struct xfs_btree_cur *);
 150         void    (*update_cursor)(struct xfs_btree_cur *src,
 151                                  struct xfs_btree_cur *dst);
 152
 153         /* update btree root pointer */
 154         void    (*set_root)(struct xfs_btree_cur *cur,
 155                             union xfs_btree_ptr *nptr, int level_change);
 156
 157         /* block allocation / freeing */
 158         int     (*alloc_block)(struct xfs_btree_cur *cur,
 159                                union xfs_btree_ptr *start_bno,
 160                                union xfs_btree_ptr *new_bno,
 161                                int length, int *stat);
 162         int     (*free_block)(struct xfs_btree_cur *cur, struct xfs_buf *bp);
 163
 164         /* update last record information */
 165         void    (*update_lastrec)(struct xfs_btree_cur *cur,
 166                                   struct xfs_btree_block *block,
 167                                   union xfs_btree_rec *rec,
 168                                   int ptr, int reason);
 169
 170         /* records in block/level */
 171         int     (*get_minrecs)(struct xfs_btree_cur *cur, int level);
 172         int     (*get_maxrecs)(struct xfs_btree_cur *cur, int level);
 173
 174         /* records on disk.  Matter for the root in inode case. */
 175         int     (*get_dmaxrecs)(struct xfs_btree_cur *cur, int level);
 176
 177         /* init values of btree structures */
 178         void    (*init_key_from_rec)(union xfs_btree_key *key,
 179                                      union xfs_btree_rec *rec);
 180         void    (*init_rec_from_key)(union xfs_btree_key *key,
 181                                      union xfs_btree_rec *rec);
 182         void    (*init_rec_from_cur)(struct xfs_btree_cur *cur,
 183                                      union xfs_btree_rec *rec);
 184         void    (*init_ptr_from_cur)(struct xfs_btree_cur *cur,
 185                                      union xfs_btree_ptr *ptr);
 186
 187         /* difference between key value and cursor value */
 188         __int64_t (*key_diff)(struct xfs_btree_cur *cur,
 189                               union xfs_btree_key *key);
 190
 191         const struct xfs_buf_ops        *buf_ops;
 192
 193 #ifdef DEBUG
 194         /* check that k1 is lower than k2 */
 195         int     (*keys_inorder)(struct xfs_btree_cur *cur,
 196                                 union xfs_btree_key *k1,
 197                                 union xfs_btree_key *k2);
 198
 199         /* check that r1 is lower than r2 */
 200         int     (*recs_inorder)(struct xfs_btree_cur *cur,
 201                                 union xfs_btree_rec *r1,
 202                                 union xfs_btree_rec *r2);
 203 #endif
 204 };
 205
 206 /*
 207  * Reasons for the update_lastrec method to be called.
 208  */
 209 #define LASTREC_UPDATE  0
 210 #define LASTREC_INSREC  1
 211 #define LASTREC_DELREC  2
 212
 213
 214 /*
 215  * Btree cursor structure.
 216  * This collects all information needed by the btree code in one place.
 217  */
 218 typedef struct xfs_btree_cur
 219 {
 220         struct xfs_trans        *bc_tp; /* transaction we're in, if any */
 221         struct xfs_mount        *bc_mp; /* file system mount struct */
 222         const struct xfs_btree_ops *bc_ops;
 223         uint                    bc_flags; /* btree features - below */
 224         union {
 225                 xfs_alloc_rec_incore_t  a;
 226                 xfs_bmbt_irec_t         b;
 227                 xfs_inobt_rec_incore_t  i;
 228         }               bc_rec;         /* current insert/search record value */
 229         struct xfs_buf  *bc_bufs[XFS_BTREE_MAXLEVELS];  /* buf ptr per level */
 230         int             bc_ptrs[XFS_BTREE_MAXLEVELS];   /* key/record # */
 231         __uint8_t       bc_ra[XFS_BTREE_MAXLEVELS];     /* readahead bits */
 232 #define XFS_BTCUR_LEFTRA        1       /* left sibling has been read-ahead */
 233 #define XFS_BTCUR_RIGHTRA       2       /* right sibling has been read-ahead */
 234         __uint8_t       bc_nlevels;     /* number of levels in the tree */
 235         __uint8_t       bc_blocklog;    /* log2(blocksize) of btree blocks */
 236         xfs_btnum_t     bc_btnum;       /* identifies which btree type */
 237         union {
 238                 struct {                        /* needed for BNO, CNT, INO */
 239                         struct xfs_buf  *agbp;  /* agf/agi buffer pointer */
 240                         xfs_agnumber_t  agno;   /* ag number */
 241                 } a;
 242                 struct {                        /* needed for BMAP */
 243                         struct xfs_inode *ip;   /* pointer to our inode */
 244                         struct xfs_bmap_free *flist;    /* list to free after */
 245                         xfs_fsblock_t   firstblock;     /* 1st blk allocated */
 246                         int             allocated;      /* count of alloced */
 247                         short           forksize;       /* fork's inode space */
 248                         char            whichfork;      /* data or attr fork */
 249                         char            flags;          /* flags */
 250 #define XFS_BTCUR_BPRV_WASDEL   1                       /* was delayed */
 251                 } b;
 252         }               bc_private;     /* per-btree type data */
 253 } xfs_btree_cur_t;
 254
 255 /* cursor flags */
 256 #define XFS_BTREE_LONG_PTRS             (1<<0)  /* pointers are 64bits long */
 257 #define XFS_BTREE_ROOT_IN_INODE         (1<<1)  /* root may be variable size */
 258 #define XFS_BTREE_LASTREC_UPDATE        (1<<2)  /* track last rec externally */
 259
 260
 261 #define XFS_BTREE_NOERROR       0
 262 #define XFS_BTREE_ERROR         1
 263
 264 /*
 265  * Convert from buffer to btree block header.
 266  */
 267 #define XFS_BUF_TO_BLOCK(bp)    ((struct xfs_btree_block *)((bp)->b_addr))
 268
 269
 270 /*
 271  * Check that block header is ok.
 272  */
 273 int
 274 xfs_btree_check_block(
 275         struct xfs_btree_cur    *cur,   /* btree cursor */
 276         struct xfs_btree_block  *block, /* generic btree block pointer */
 277         int                     level,  /* level of the btree block */
 278         struct xfs_buf          *bp);   /* buffer containing block, if any */
 279
 280 /*
 281  * Check that (long) pointer is ok.
 282  */
 283 int                                     /* error (0 or EFSCORRUPTED) */
 284 xfs_btree_check_lptr(
 285         struct xfs_btree_cur    *cur,   /* btree cursor */
 286         xfs_dfsbno_t            ptr,    /* btree block disk address */
 287         int                     level); /* btree block level */
 288
 289 /*
 290  * Delete the btree cursor.
 291  */
 292 void
 293 xfs_btree_del_cursor(
 294         xfs_btree_cur_t         *cur,   /* btree cursor */
 295         int                     error); /* del because of error */
 296
 297 /*
 298  * Duplicate the btree cursor.
 299  * Allocate a new one, copy the record, re-get the buffers.
 300  */
 301 int                                     /* error */
 302 xfs_btree_dup_cursor(
 303         xfs_btree_cur_t         *cur,   /* input cursor */
 304         xfs_btree_cur_t         **ncur);/* output cursor */
 305
 306 /*
 307  * Get a buffer for the block, return it with no data read.
 308  * Long-form addressing.
 309  */
 310 struct xfs_buf *                                /* buffer for fsbno */
 311 xfs_btree_get_bufl(
 312         struct xfs_mount        *mp,    /* file system mount point */
 313         struct xfs_trans        *tp,    /* transaction pointer */
 314         xfs_fsblock_t           fsbno,  /* file system block number */
 315         uint                    lock);  /* lock flags for get_buf */
 316
 317 /*
 318  * Get a buffer for the block, return it with no data read.
 319  * Short-form addressing.
 320  */
 321 struct xfs_buf *                                /* buffer for agno/agbno */
 322 xfs_btree_get_bufs(
 323         struct xfs_mount        *mp,    /* file system mount point */
 324         struct xfs_trans        *tp,    /* transaction pointer */
 325         xfs_agnumber_t          agno,   /* allocation group number */
 326         xfs_agblock_t           agbno,  /* allocation group block number */
 327         uint                    lock);  /* lock flags for get_buf */
 328
 329 /*
 330  * Check for the cursor referring to the last block at the given level.
 331  */
 332 int                                     /* 1=is last block, 0=not last block */
 333 xfs_btree_islastblock(
 334         xfs_btree_cur_t         *cur,   /* btree cursor */
 335         int                     level); /* level to check */
 336
 337 /*
 338  * Compute first and last byte offsets for the fields given.
 339  * Interprets the offsets table, which contains struct field offsets.
 340  */
 341 void
 342 xfs_btree_offsets(
 343         __int64_t               fields, /* bitmask of fields */
 344         const short             *offsets,/* table of field offsets */
 345         int                     nbits,  /* number of bits to inspect */
 346         int                     *first, /* output: first byte offset */
 347         int                     *last); /* output: last byte offset */
 348
 349 /*
 350  * Get a buffer for the block, return it read in.
 351  * Long-form addressing.
 352  */
 353 int                                     /* error */
 354 xfs_btree_read_bufl(
 355         struct xfs_mount        *mp,    /* file system mount point */
 356         struct xfs_trans        *tp,    /* transaction pointer */
 357         xfs_fsblock_t           fsbno,  /* file system block number */
 358         uint                    lock,   /* lock flags for read_buf */
 359         struct xfs_buf          **bpp,  /* buffer for fsbno */
 360         int                     refval, /* ref count value for buffer */
 361         const struct xfs_buf_ops *ops);
 362
 363 /*
 364  * Read-ahead the block, don't wait for it, don't return a buffer.
 365  * Long-form addressing.
 366  */
 367 void                                    /* error */
 368 xfs_btree_reada_bufl(
 369         struct xfs_mount        *mp,    /* file system mount point */
 370         xfs_fsblock_t           fsbno,  /* file system block number */
 371         xfs_extlen_t            count,  /* count of filesystem blocks */
 372         const struct xfs_buf_ops *ops);
 373
 374 /*
 375  * Read-ahead the block, don't wait for it, don't return a buffer.
 376  * Short-form addressing.
 377  */
 378 void                                    /* error */
 379 xfs_btree_reada_bufs(
 380         struct xfs_mount        *mp,    /* file system mount point */
 381         xfs_agnumber_t          agno,   /* allocation group number */
 382         xfs_agblock_t           agbno,  /* allocation group block number */
 383         xfs_extlen_t            count,  /* count of filesystem blocks */
 384         const struct xfs_buf_ops *ops);
 385
 386 /*
 387  * Initialise a new btree block header
 388  */
 389 void
 390 xfs_btree_init_block(
 391         struct xfs_mount *mp,
 392         struct xfs_buf  *bp,
 393         __u32           magic,
 394         __u16           level,
 395         __u16           numrecs,
 396         unsigned int    flags);
 397
 398 /*
 399  * Common btree core entry points.
 400  */
 401 int xfs_btree_increment(struct xfs_btree_cur *, int, int *);
 402 int xfs_btree_decrement(struct xfs_btree_cur *, int, int *);
 403 int xfs_btree_lookup(struct xfs_btree_cur *, xfs_lookup_t, int *);
 404 int xfs_btree_update(struct xfs_btree_cur *, union xfs_btree_rec *);
 405 int xfs_btree_new_iroot(struct xfs_btree_cur *, int *, int *);
 406 int xfs_btree_insert(struct xfs_btree_cur *, int *);
 407 int xfs_btree_delete(struct xfs_btree_cur *, int *);
 408 int xfs_btree_get_rec(struct xfs_btree_cur *, union xfs_btree_rec **, int *);
 409
 410 /*
 411  * Internal btree helpers also used by xfs_bmap.c.
 412  */
 413 void xfs_btree_log_block(struct xfs_btree_cur *, struct xfs_buf *, int);
 414 void xfs_btree_log_recs(struct xfs_btree_cur *, struct xfs_buf *, int, int);
 415
 416 /*
 417  * Helpers.
 418  */
 419 static inline int xfs_btree_get_numrecs(struct xfs_btree_block *block)
 420 {
 421         return be16_to_cpu(block->bb_numrecs);
 422 }
 423
 424 static inline void xfs_btree_set_numrecs(struct xfs_btree_block *block,
 425                 __uint16_t numrecs)
 426 {
 427         block->bb_numrecs = cpu_to_be16(numrecs);
 428 }
 429
 430 static inline int xfs_btree_get_level(struct xfs_btree_block *block)
 431 {
 432         return be16_to_cpu(block->bb_level);
 433 }
 434
 435
 436 /*
 437  * Min and max functions for extlen, agblock, fileoff, and filblks types.
 438  */
 439 #define XFS_EXTLEN_MIN(a,b)     min_t(xfs_extlen_t, (a), (b))
 440 #define XFS_EXTLEN_MAX(a,b)     max_t(xfs_extlen_t, (a), (b))
 441 #define XFS_AGBLOCK_MIN(a,b)    min_t(xfs_agblock_t, (a), (b))
 442 #define XFS_AGBLOCK_MAX(a,b)    max_t(xfs_agblock_t, (a), (b))
 443 #define XFS_FILEOFF_MIN(a,b)    min_t(xfs_fileoff_t, (a), (b))
 444 #define XFS_FILEOFF_MAX(a,b)    max_t(xfs_fileoff_t, (a), (b))
 445 #define XFS_FILBLKS_MIN(a,b)    min_t(xfs_filblks_t, (a), (b))
 446 #define XFS_FILBLKS_MAX(a,b)    max_t(xfs_filblks_t, (a), (b))
 447
 448 #define XFS_FSB_SANITY_CHECK(mp,fsb)    \
 449         (XFS_FSB_TO_AGNO(mp, fsb) < mp->m_sb.sb_agcount && \
 450                 XFS_FSB_TO_AGBNO(mp, fsb) < mp->m_sb.sb_agblocks)
 451
 452 /*
 453  * Trace hooks.  Currently not implemented as they need to be ported
 454  * over to the generic tracing functionality, which is some effort.
 455  *
 456  * i,j = integer (32 bit)
 457  * b = btree block buffer (xfs_buf_t)
 458  * p = btree ptr
 459  * r = btree record
 460  * k = btree key
 461  */
 462 #define XFS_BTREE_TRACE_ARGBI(c, b, i)
 463 #define XFS_BTREE_TRACE_ARGBII(c, b, i, j)
 464 #define XFS_BTREE_TRACE_ARGI(c, i)
 465 #define XFS_BTREE_TRACE_ARGIPK(c, i, p, s)
 466 #define XFS_BTREE_TRACE_ARGIPR(c, i, p, r)
 467 #define XFS_BTREE_TRACE_ARGIK(c, i, k)
 468 #define XFS_BTREE_TRACE_ARGR(c, r)
 469 #define XFS_BTREE_TRACE_CURSOR(c, t)
 470
 471 #endif  /* __XFS_BTREE_H__ */