db/btree/btree.h

   1 /*-
   2  * Copyright (c) 1991, 1993, 1994
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Mike Olson.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      @(#)btree.h     8.11 (Berkeley) 8/17/94
  37  */
  38
  39 /* Macros to set/clear/test flags. */
  40 #define F_SET(p, f)     (p)->flags |= (f)
  41 #define F_CLR(p, f)     (p)->flags &= ~(f)
  42 #define F_ISSET(p, f)   ((p)->flags & (f))
  43
  44 #include <mpool.h>
  45
  46 #define DEFMINKEYPAGE   (2)             /* Minimum keys per page */
  47 #define MINCACHE        (5)             /* Minimum cached pages */
  48 #define MINPSIZE        (512)           /* Minimum page size */
  49
  50 /*
  51  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
  52  * used as an out-of-band page, i.e. page pointers that point to nowhere point
  53  * to page 0.  Page 1 is the root of the btree.
  54  */
  55 #define P_INVALID        0              /* Invalid tree page number. */
  56 #define P_META           0              /* Tree metadata page number. */
  57 #define P_ROOT           1              /* Tree root page number. */
  58
  59 /*
  60  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
  61  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
  62  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
  63  * This implementation requires that values within structures NOT be padded.
  64  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
  65  * to do some work to get this package to run.
  66  */
  67 typedef struct _page {
  68         pgno_t  pgno;                   /* this page's page number */
  69         pgno_t  prevpg;                 /* left sibling */
  70         pgno_t  nextpg;                 /* right sibling */
  71
  72 #define P_BINTERNAL     0x01            /* btree internal page */
  73 #define P_BLEAF         0x02            /* leaf page */
  74 #define P_OVERFLOW      0x04            /* overflow page */
  75 #define P_RINTERNAL     0x08            /* recno internal page */
  76 #define P_RLEAF         0x10            /* leaf page */
  77 #define P_TYPE          0x1f            /* type mask */
  78 #define P_PRESERVE      0x20            /* never delete this chain of pages */
  79         u_int32_t flags;
  80
  81         indx_t  lower;                  /* lower bound of free space on page */
  82         indx_t  upper;                  /* upper bound of free space on page */
  83         indx_t  linp[1];                /* indx_t-aligned VAR. LENGTH DATA */
  84 } PAGE;
  85
  86 /* First and next index. */
  87 #define BTDATAOFF                                                       \
  88         (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) +             \
  89             sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t))
  90 #define NEXTINDEX(p)    (((p)->lower - BTDATAOFF) / sizeof(indx_t))
  91
  92 /*
  93  * For pages other than overflow pages, there is an array of offsets into the
  94  * rest of the page immediately following the page header.  Each offset is to
  95  * an item which is unique to the type of page.  The h_lower offset is just
  96  * past the last filled-in index.  The h_upper offset is the first item on the
  97  * page.  Offsets are from the beginning of the page.
  98  *
  99  * If an item is too big to store on a single page, a flag is set and the item
 100  * is a { page, size } pair such that the page is the first page of an overflow
 101  * chain with size bytes of item.  Overflow pages are simply bytes without any
 102  * external structure.
 103  *
 104  * The page number and size fields in the items are pgno_t-aligned so they can
 105  * be manipulated without copying.  (This presumes that 32 bit items can be
 106  * manipulated on this system.)
 107  */
 108 #define LALIGN(n)       (((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1))
 109 #define NOVFLSIZE       (sizeof(pgno_t) + sizeof(u_int32_t))
 110
 111 /*
 112  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
 113  * pairs, such that the key compares less than or equal to all of the records
 114  * on that page.  For a tree without duplicate keys, an internal page with two
 115  * consecutive keys, a and b, will have all records greater than or equal to a
 116  * and less than b stored on the page associated with a.  Duplicate keys are
 117  * somewhat special and can cause duplicate internal and leaf page records and
 118  * some minor modifications of the above rule.
 119  */
 120 typedef struct _binternal {
 121         u_int32_t ksize;                /* key size */
 122         pgno_t  pgno;                   /* page number stored on */
 123 #define P_BIGDATA       0x01            /* overflow data */
 124 #define P_BIGKEY        0x02            /* overflow key */
 125         u_char  flags;
 126         char    bytes[1];               /* data */
 127 } BINTERNAL;
 128
 129 /* Get the page's BINTERNAL structure at index indx. */
 130 #define GETBINTERNAL(pg, indx)                                          \
 131         ((BINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
 132
 133 /* Get the number of bytes in the entry. */
 134 #define NBINTERNAL(len)                                                 \
 135         LALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len))
 136
 137 /* Copy a BINTERNAL entry to the page. */
 138 #define WR_BINTERNAL(p, size, pgno, flags) {                            \
 139         *(u_int32_t *)p = size;                                         \
 140         p += sizeof(u_int32_t);                                         \
 141         *(pgno_t *)p = pgno;                                            \
 142         p += sizeof(pgno_t);                                            \
 143         *(u_char *)p = flags;                                           \
 144         p += sizeof(u_char);                                            \
 145 }
 146
 147 /*
 148  * For the recno internal pages, the item is a page number with the number of
 149  * keys found on that page and below.
 150  */
 151 typedef struct _rinternal {
 152         recno_t nrecs;                  /* number of records */
 153         pgno_t  pgno;                   /* page number stored below */
 154 } RINTERNAL;
 155
 156 /* Get the page's RINTERNAL structure at index indx. */
 157 #define GETRINTERNAL(pg, indx)                                          \
 158         ((RINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
 159
 160 /* Get the number of bytes in the entry. */
 161 #define NRINTERNAL                                                      \
 162         LALIGN(sizeof(recno_t) + sizeof(pgno_t))
 163
 164 /* Copy a RINTERAL entry to the page. */
 165 #define WR_RINTERNAL(p, nrecs, pgno) {                                  \
 166         *(recno_t *)p = nrecs;                                          \
 167         p += sizeof(recno_t);                                           \
 168         *(pgno_t *)p = pgno;                                            \
 169 }
 170
 171 /* For the btree leaf pages, the item is a key and data pair. */
 172 typedef struct _bleaf {
 173         u_int32_t       ksize;          /* size of key */
 174         u_int32_t       dsize;          /* size of data */
 175         u_char  flags;                  /* P_BIGDATA, P_BIGKEY */
 176         char    bytes[1];               /* data */
 177 } BLEAF;
 178
 179 /* Get the page's BLEAF structure at index indx. */
 180 #define GETBLEAF(pg, indx)                                              \
 181         ((BLEAF *)((char *)(pg) + (pg)->linp[indx]))
 182
 183 /* Get the number of bytes in the entry. */
 184 #define NBLEAF(p)       NBLEAFDBT((p)->ksize, (p)->dsize)
 185
 186 /* Get the number of bytes in the user's key/data pair. */
 187 #define NBLEAFDBT(ksize, dsize)                                         \
 188         LALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) + \
 189             (ksize) + (dsize))
 190
 191 /* Copy a BLEAF entry to the page. */
 192 #define WR_BLEAF(p, key, data, flags) {                                 \
 193         *(u_int32_t *)p = key->size;                                    \
 194         p += sizeof(u_int32_t);                                         \
 195         *(u_int32_t *)p = data->size;                                   \
 196         p += sizeof(u_int32_t);                                         \
 197         *(u_char *)p = flags;                                           \
 198         p += sizeof(u_char);                                            \
 199         memmove(p, key->data, key->size);                               \
 200         p += key->size;                                                 \
 201         memmove(p, data->data, data->size);                             \
 202 }
 203
 204 /* For the recno leaf pages, the item is a data entry. */
 205 typedef struct _rleaf {
 206         u_int32_t       dsize;          /* size of data */
 207         u_char  flags;                  /* P_BIGDATA */
 208         char    bytes[1];
 209 } RLEAF;
 210
 211 /* Get the page's RLEAF structure at index indx. */
 212 #define GETRLEAF(pg, indx)                                              \
 213         ((RLEAF *)((char *)(pg) + (pg)->linp[indx]))
 214
 215 /* Get the number of bytes in the entry. */
 216 #define NRLEAF(p)       NRLEAFDBT((p)->dsize)
 217
 218 /* Get the number of bytes from the user's data. */
 219 #define NRLEAFDBT(dsize)                                                \
 220         LALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize))
 221
 222 /* Copy a RLEAF entry to the page. */
 223 #define WR_RLEAF(p, data, flags) {                                      \
 224         *(u_int32_t *)p = data->size;                                   \
 225         p += sizeof(u_int32_t);                                         \
 226         *(u_char *)p = flags;                                           \
 227         p += sizeof(u_char);                                            \
 228         memmove(p, data->data, data->size);                             \
 229 }
 230
 231 /*
 232  * A record in the tree is either a pointer to a page and an index in the page
 233  * or a page number and an index.  These structures are used as a cursor, stack
 234  * entry and search returns as well as to pass records to other routines.
 235  *
 236  * One comment about searches.  Internal page searches must find the largest
 237  * record less than key in the tree so that descents work.  Leaf page searches
 238  * must find the smallest record greater than key so that the returned index
 239  * is the record's correct position for insertion.
 240  */
 241 typedef struct _epgno {
 242         pgno_t  pgno;                   /* the page number */
 243         indx_t  index;                  /* the index on the page */
 244 } EPGNO;
 245
 246 typedef struct _epg {
 247         PAGE    *page;                  /* the (pinned) page */
 248         indx_t   index;                 /* the index on the page */
 249 } EPG;
 250
 251 /*
 252  * About cursors.  The cursor (and the page that contained the key/data pair
 253  * that it referenced) can be deleted, which makes things a bit tricky.  If
 254  * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set
 255  * or there simply aren't any duplicates of the key) we copy the key that it
 256  * referenced when it's deleted, and reacquire a new cursor key if the cursor
 257  * is used again.  If there are duplicates keys, we move to the next/previous
 258  * key, and set a flag so that we know what happened.  NOTE: if duplicate (to
 259  * the cursor) keys are added to the tree during this process, it is undefined
 260  * if they will be returned or not in a cursor scan.
 261  *
 262  * The flags determine the possible states of the cursor:
 263  *
 264  * CURS_INIT    The cursor references *something*.
 265  * CURS_ACQUIRE The cursor was deleted, and a key has been saved so that
 266  *              we can reacquire the right position in the tree.
 267  * CURS_AFTER, CURS_BEFORE
 268  *              The cursor was deleted, and now references a key/data pair
 269  *              that has not yet been returned, either before or after the
 270  *              deleted key/data pair.
 271  * XXX
 272  * This structure is broken out so that we can eventually offer multiple
 273  * cursors as part of the DB interface.
 274  */
 275 typedef struct _cursor {
 276         EPGNO    pg;                    /* B: Saved tree reference. */
 277         DBT      key;                   /* B: Saved key, or key.data == NULL. */
 278         recno_t  rcursor;               /* R: recno cursor (1-based) */
 279
 280 #define CURS_ACQUIRE    0x01            /*  B: Cursor needs to be reacquired. */
 281 #define CURS_AFTER      0x02            /*  B: Unreturned cursor after key. */
 282 #define CURS_BEFORE     0x04            /*  B: Unreturned cursor before key. */
 283 #define CURS_INIT       0x08            /* RB: Cursor initialized. */
 284         u_int8_t flags;
 285 } CURSOR;
 286
 287 /*
 288  * The metadata of the tree.  The nrecs field is used only by the RECNO code.
 289  * This is because the btree doesn't really need it and it requires that every
 290  * put or delete call modify the metadata.
 291  */
 292 typedef struct _btmeta {
 293         u_int32_t       magic;          /* magic number */
 294         u_int32_t       version;        /* version */
 295         u_int32_t       psize;          /* page size */
 296         u_int32_t       free;           /* page number of first free page */
 297         u_int32_t       nrecs;          /* R: number of records */
 298
 299 #define SAVEMETA        (B_NODUPS | R_RECNO)
 300         u_int32_t       flags;          /* bt_flags & SAVEMETA */
 301 } BTMETA;
 302
 303 /* The in-memory btree/recno data structure. */
 304 typedef struct _btree {
 305         MPOOL    *bt_mp;                /* memory pool cookie */
 306
 307         DB       *bt_dbp;               /* pointer to enclosing DB */
 308
 309         EPG       bt_cur;               /* current (pinned) page */
 310         PAGE     *bt_pinned;            /* page pinned across calls */
 311
 312         CURSOR    bt_cursor;            /* cursor */
 313
 314 #define BT_PUSH(t, p, i) {                                              \
 315         t->bt_sp->pgno = p;                                             \
 316         t->bt_sp->index = i;                                            \
 317         ++t->bt_sp;                                                     \
 318 }
 319 #define BT_POP(t)       (t->bt_sp == t->bt_stack ? NULL : --t->bt_sp)
 320 #define BT_CLR(t)       (t->bt_sp = t->bt_stack)
 321         EPGNO     bt_stack[50];         /* stack of parent pages */
 322         EPGNO    *bt_sp;                /* current stack pointer */
 323
 324         DBT       bt_rkey;              /* returned key */
 325         DBT       bt_rdata;             /* returned data */
 326
 327         int       bt_fd;                /* tree file descriptor */
 328
 329         pgno_t    bt_free;              /* next free page */
 330         u_int32_t bt_psize;             /* page size */
 331         indx_t    bt_ovflsize;          /* cut-off for key/data overflow */
 332         int       bt_lorder;            /* byte order */
 333                                         /* sorted order */
 334         enum { NOT, BACK, FORWARD } bt_order;
 335         EPGNO     bt_last;              /* last insert */
 336
 337                                         /* B: key comparison function */
 338         int     (*bt_cmp) __P((const DBT *, const DBT *));
 339                                         /* B: prefix comparison function */
 340         size_t  (*bt_pfx) __P((const DBT *, const DBT *));
 341                                         /* R: recno input function */
 342         int     (*bt_irec) __P((struct _btree *, recno_t));
 343
 344         FILE     *bt_rfp;               /* R: record FILE pointer */
 345         int       bt_rfd;               /* R: record file descriptor */
 346
 347         caddr_t   bt_cmap;              /* R: current point in mapped space */
 348         caddr_t   bt_smap;              /* R: start of mapped space */
 349         caddr_t   bt_emap;              /* R: end of mapped space */
 350         size_t    bt_msize;             /* R: size of mapped region. */
 351
 352         recno_t   bt_nrecs;             /* R: number of records */
 353         size_t    bt_reclen;            /* R: fixed record length */
 354         u_char    bt_bval;              /* R: delimiting byte/pad character */
 355
 356 /*
 357  * NB:
 358  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
 359  */
 360 #define B_INMEM         0x00001         /* in-memory tree */
 361 #define B_METADIRTY     0x00002         /* need to write metadata */
 362 #define B_MODIFIED      0x00004         /* tree modified */
 363 #define B_NEEDSWAP      0x00008         /* if byte order requires swapping */
 364 #define B_RDONLY        0x00010         /* read-only tree */
 365
 366 #define B_NODUPS        0x00020         /* no duplicate keys permitted */
 367 #define R_RECNO         0x00080         /* record oriented tree */
 368
 369 #define R_CLOSEFP       0x00040         /* opened a file pointer */
 370 #define R_EOF           0x00100         /* end of input file reached. */
 371 #define R_FIXLEN        0x00200         /* fixed length records */
 372 #define R_MEMMAPPED     0x00400         /* memory mapped file. */
 373 #define R_INMEM         0x00800         /* in-memory file */
 374 #define R_MODIFIED      0x01000         /* modified file */
 375 #define R_RDONLY        0x02000         /* read-only file */
 376
 377 #define B_DB_LOCK       0x04000         /* DB_LOCK specified. */
 378 #define B_DB_SHMEM      0x08000         /* DB_SHMEM specified. */
 379 #define B_DB_TXN        0x10000         /* DB_TXN specified. */
 380         u_int32_t flags;
 381 } BTREE;
 382
 383 #include "extern.h"