db.subproj/btree.subproj/btree.h

   1 /*
   2  * Copyright (c) 1999 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * The contents of this file constitute Original Code as defined in and
   7  * are subject to the Apple Public Source License Version 1.1 (the
   8  * "License").  You may not use this file except in compliance with the
   9  * License.  Please obtain a copy of the License at
  10  * http://www.apple.com/publicsource and read it before using this file.
  11  *
  12  * This Original Code and all software distributed under the License are
  13  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  14  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  15  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  17  * License for the specific language governing rights and limitations
  18  * under the License.
  19  *
  20  * @APPLE_LICENSE_HEADER_END@
  21  */
  22 /*
  23  * Copyright (c) 1991, 1993
  24  *      The Regents of the University of California.  All rights reserved.
  25  *
  26  * This code is derived from software contributed to Berkeley by
  27  * Mike Olson.
  28  *
  29  * Redistribution and use in source and binary forms, with or without
  30  * modification, are permitted provided that the following conditions
  31  * are met:
  32  * 1. Redistributions of source code must retain the above copyright
  33  *    notice, this list of conditions and the following disclaimer.
  34  * 2. Redistributions in binary form must reproduce the above copyright
  35  *    notice, this list of conditions and the following disclaimer in the
  36  *    documentation and/or other materials provided with the distribution.
  37  * 3. All advertising materials mentioning features or use of this software
  38  *    must display the following acknowledgement:
  39  *      This product includes software developed by the University of
  40  *      California, Berkeley and its contributors.
  41  * 4. Neither the name of the University nor the names of its contributors
  42  *    may be used to endorse or promote products derived from this software
  43  *    without specific prior written permission.
  44  *
  45  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  55  * SUCH DAMAGE.
  56  */
  57
  58 #include <mpool.h>
  59
  60 #define DEFMINKEYPAGE   (2)             /* Minimum keys per page */
  61 #define MINCACHE        (5)             /* Minimum cached pages */
  62 #define MINPSIZE        (512)           /* Minimum page size */
  63
  64 /*
  65  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
  66  * used as an out-of-band page, i.e. page pointers that point to nowhere point
  67  * to page 0.  Page 1 is the root of the btree.
  68  */
  69 #define P_INVALID        0              /* Invalid tree page number. */
  70 #define P_META           0              /* Tree metadata page number. */
  71 #define P_ROOT           1              /* Tree root page number. */
  72
  73 /*
  74  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
  75  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
  76  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
  77  * This implementation requires that values within structures NOT be padded.
  78  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
  79  * to do some work to get this package to run.
  80  */
  81 typedef struct _page {
  82         pgno_t  pgno;                   /* this page's page number */
  83         pgno_t  prevpg;                 /* left sibling */
  84         pgno_t  nextpg;                 /* right sibling */
  85
  86 #define P_BINTERNAL     0x01            /* btree internal page */
  87 #define P_BLEAF         0x02            /* leaf page */
  88 #define P_OVERFLOW      0x04            /* overflow page */
  89 #define P_RINTERNAL     0x08            /* recno internal page */
  90 #define P_RLEAF         0x10            /* leaf page */
  91 #define P_TYPE          0x1f            /* type mask */
  92 #define P_PRESERVE      0x20            /* never delete this chain of pages */
  93         u_int32_t flags;
  94
  95         indx_t  lower;                  /* lower bound of free space on page */
  96         indx_t  upper;                  /* upper bound of free space on page */
  97         indx_t  linp[1];                /* indx_t-aligned VAR. LENGTH DATA */
  98 } PAGE;
  99
 100 /* First and next index. */
 101 #define BTDATAOFF       (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \
 102                             sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t))
 103 #define NEXTINDEX(p)    (((p)->lower - BTDATAOFF) / sizeof(indx_t))
 104
 105 /*
 106  * For pages other than overflow pages, there is an array of offsets into the
 107  * rest of the page immediately following the page header.  Each offset is to
 108  * an item which is unique to the type of page.  The h_lower offset is just
 109  * past the last filled-in index.  The h_upper offset is the first item on the
 110  * page.  Offsets are from the beginning of the page.
 111  *
 112  * If an item is too big to store on a single page, a flag is set and the item
 113  * is a { page, size } pair such that the page is the first page of an overflow
 114  * chain with size bytes of item.  Overflow pages are simply bytes without any
 115  * external structure.
 116  *
 117  * The page number and size fields in the items are pgno_t-aligned so they can
 118  * be manipulated without copying.  (This presumes that 32 bit items can be
 119  * manipulated on this system.)
 120  */
 121 #define LALIGN(n) \
 122         (((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1))
 123 #define NOVFLSIZE       (sizeof(pgno_t) + sizeof(size_t))
 124
 125 /*
 126  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
 127  * pairs, such that the key compares less than or equal to all of the records
 128  * on that page.  For a tree without duplicate keys, an internal page with two
 129  * consecutive keys, a and b, will have all records greater than or equal to a
 130  * and less than b stored on the page associated with a.  Duplicate keys are
 131  * somewhat special and can cause duplicate internal and leaf page records and
 132  * some minor modifications of the above rule.
 133  */
 134 typedef struct _binternal {
 135         size_t  ksize;                  /* key size */
 136         pgno_t  pgno;                   /* page number stored on */
 137 #define P_BIGDATA       0x01            /* overflow data */
 138 #define P_BIGKEY        0x02            /* overflow key */
 139         u_char  flags;
 140         char    bytes[1];               /* data */
 141 } BINTERNAL;
 142
 143 /* Get the page's BINTERNAL structure at index indx. */
 144 #define GETBINTERNAL(pg, indx) \
 145         ((BINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
 146
 147 /* Get the number of bytes in the entry. */
 148 #define NBINTERNAL(len) \
 149         LALIGN(sizeof(size_t) + sizeof(pgno_t) + sizeof(u_char) + (len))
 150
 151 /* Copy a BINTERNAL entry to the page. */
 152 #define WR_BINTERNAL(p, size, pgno, flags) { \
 153         *(size_t *)p = size; \
 154         p += sizeof(size_t); \
 155         *(pgno_t *)p = pgno; \
 156         p += sizeof(pgno_t); \
 157         *(u_char *)p = flags; \
 158         p += sizeof(u_char); \
 159 }
 160
 161 /*
 162  * For the recno internal pages, the item is a page number with the number of
 163  * keys found on that page and below.
 164  */
 165 typedef struct _rinternal {
 166         recno_t nrecs;                  /* number of records */
 167         pgno_t  pgno;                   /* page number stored below */
 168 } RINTERNAL;
 169
 170 /* Get the page's RINTERNAL structure at index indx. */
 171 #define GETRINTERNAL(pg, indx) \
 172         ((RINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
 173
 174 /* Get the number of bytes in the entry. */
 175 #define NRINTERNAL \
 176         LALIGN(sizeof(recno_t) + sizeof(pgno_t))
 177
 178 /* Copy a RINTERAL entry to the page. */
 179 #define WR_RINTERNAL(p, nrecs, pgno) { \
 180         *(recno_t *)p = nrecs; \
 181         p += sizeof(recno_t); \
 182         *(pgno_t *)p = pgno; \
 183 }
 184
 185 /* For the btree leaf pages, the item is a key and data pair. */
 186 typedef struct _bleaf {
 187         size_t  ksize;                  /* size of key */
 188         size_t  dsize;                  /* size of data */
 189         u_char  flags;                  /* P_BIGDATA, P_BIGKEY */
 190         char    bytes[1];               /* data */
 191 } BLEAF;
 192
 193 /* Get the page's BLEAF structure at index indx. */
 194 #define GETBLEAF(pg, indx) \
 195         ((BLEAF *)((char *)(pg) + (pg)->linp[indx]))
 196
 197 /* Get the number of bytes in the entry. */
 198 #define NBLEAF(p)       NBLEAFDBT((p)->ksize, (p)->dsize)
 199
 200 /* Get the number of bytes in the user's key/data pair. */
 201 #define NBLEAFDBT(ksize, dsize) \
 202         LALIGN(sizeof(size_t) + sizeof(size_t) + sizeof(u_char) + \
 203             (ksize) + (dsize))
 204
 205 /* Copy a BLEAF entry to the page. */
 206 #define WR_BLEAF(p, key, data, flags) { \
 207         *(size_t *)p = key->size; \
 208         p += sizeof(size_t); \
 209         *(size_t *)p = data->size; \
 210         p += sizeof(size_t); \
 211         *(u_char *)p = flags; \
 212         p += sizeof(u_char); \
 213         memmove(p, key->data, key->size); \
 214         p += key->size; \
 215         memmove(p, data->data, data->size); \
 216 }
 217
 218 /* For the recno leaf pages, the item is a data entry. */
 219 typedef struct _rleaf {
 220         size_t  dsize;                  /* size of data */
 221         u_char  flags;                  /* P_BIGDATA */
 222         char    bytes[1];
 223 } RLEAF;
 224
 225 /* Get the page's RLEAF structure at index indx. */
 226 #define GETRLEAF(pg, indx) \
 227         ((RLEAF *)((char *)(pg) + (pg)->linp[indx]))
 228
 229 /* Get the number of bytes in the entry. */
 230 #define NRLEAF(p)       NRLEAFDBT((p)->dsize)
 231
 232 /* Get the number of bytes from the user's data. */
 233 #define NRLEAFDBT(dsize) \
 234         LALIGN(sizeof(size_t) + sizeof(u_char) + (dsize))
 235
 236 /* Copy a RLEAF entry to the page. */
 237 #define WR_RLEAF(p, data, flags) { \
 238         *(size_t *)p = data->size; \
 239         p += sizeof(size_t); \
 240         *(u_char *)p = flags; \
 241         p += sizeof(u_char); \
 242         memmove(p, data->data, data->size); \
 243 }
 244
 245 /*
 246  * A record in the tree is either a pointer to a page and an index in the page
 247  * or a page number and an index.  These structures are used as a cursor, stack
 248  * entry and search returns as well as to pass records to other routines.
 249  *
 250  * One comment about searches.  Internal page searches must find the largest
 251  * record less than key in the tree so that descents work.  Leaf page searches
 252  * must find the smallest record greater than key so that the returned index
 253  * is the record's correct position for insertion.
 254  *
 255  * One comment about cursors.  The cursor key is never removed from the tree,
 256  * even if deleted.  This is because it is quite difficult to decide where the
 257  * cursor should be when other keys have been inserted/deleted in the tree;
 258  * duplicate keys make it impossible.  This scheme does require extra work
 259  * though, to make sure that we don't perform an operation on a deleted key.
 260  */
 261 typedef struct _epgno {
 262         pgno_t  pgno;                   /* the page number */
 263         indx_t  index;                  /* the index on the page */
 264 } EPGNO;
 265
 266 typedef struct _epg {
 267         PAGE    *page;                  /* the (pinned) page */
 268         indx_t   index;                 /* the index on the page */
 269 } EPG;
 270
 271 /*
 272  * The metadata of the tree.  The m_nrecs field is used only by the RECNO code.
 273  * This is because the btree doesn't really need it and it requires that every
 274  * put or delete call modify the metadata.
 275  */
 276 typedef struct _btmeta {
 277         u_int32_t       m_magic;        /* magic number */
 278         u_int32_t       m_version;      /* version */
 279         u_int32_t       m_psize;        /* page size */
 280         u_int32_t       m_free;         /* page number of first free page */
 281         u_int32_t       m_nrecs;        /* R: number of records */
 282 #define SAVEMETA        (B_NODUPS | R_RECNO)
 283         u_int32_t       m_flags;        /* bt_flags & SAVEMETA */
 284         u_int32_t       m_unused;       /* unused */
 285 } BTMETA;
 286
 287 /* The in-memory btree/recno data structure. */
 288 typedef struct _btree {
 289         MPOOL   *bt_mp;                 /* memory pool cookie */
 290
 291         DB      *bt_dbp;                /* pointer to enclosing DB */
 292
 293         EPG     bt_cur;                 /* current (pinned) page */
 294         PAGE    *bt_pinned;             /* page pinned across calls */
 295
 296         EPGNO   bt_bcursor;             /* B: btree cursor */
 297         recno_t bt_rcursor;             /* R: recno cursor (1-based) */
 298
 299 #define BT_POP(t)       (t->bt_sp ? t->bt_stack + --t->bt_sp : NULL)
 300 #define BT_CLR(t)       (t->bt_sp = 0)
 301         EPGNO   *bt_stack;              /* stack of parent pages */
 302         u_int   bt_sp;                  /* current stack pointer */
 303         u_int   bt_maxstack;            /* largest stack */
 304
 305         char    *bt_kbuf;               /* key buffer */
 306         size_t  bt_kbufsz;              /* key buffer size */
 307         char    *bt_dbuf;               /* data buffer */
 308         size_t  bt_dbufsz;              /* data buffer size */
 309
 310         int     bt_fd;                  /* tree file descriptor */
 311
 312         pgno_t  bt_free;                /* next free page */
 313         u_int32_t bt_psize;             /* page size */
 314         indx_t  bt_ovflsize;            /* cut-off for key/data overflow */
 315         int     bt_lorder;              /* byte order */
 316                                         /* sorted order */
 317         enum { NOT, BACK, FORWARD } bt_order;
 318         EPGNO   bt_last;                /* last insert */
 319
 320                                         /* B: key comparison function */
 321         int     (*bt_cmp) __P((const DBT *, const DBT *));
 322                                         /* B: prefix comparison function */
 323         size_t  (*bt_pfx) __P((const DBT *, const DBT *));
 324                                         /* R: recno input function */
 325         int     (*bt_irec) __P((struct _btree *, recno_t));
 326
 327         FILE    *bt_rfp;                /* R: record FILE pointer */
 328         int     bt_rfd;                 /* R: record file descriptor */
 329
 330         caddr_t bt_cmap;                /* R: current point in mapped space */
 331         caddr_t bt_smap;                /* R: start of mapped space */
 332         caddr_t bt_emap;                /* R: end of mapped space */
 333         size_t  bt_msize;               /* R: size of mapped region. */
 334
 335         recno_t bt_nrecs;               /* R: number of records */
 336         size_t  bt_reclen;              /* R: fixed record length */
 337         u_char  bt_bval;                /* R: delimiting byte/pad character */
 338
 339 /*
 340  * NB:
 341  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
 342  */
 343 #define B_DELCRSR       0x00001         /* cursor has been deleted */
 344 #define B_INMEM         0x00002         /* in-memory tree */
 345 #define B_METADIRTY     0x00004         /* need to write metadata */
 346 #define B_MODIFIED      0x00008         /* tree modified */
 347 #define B_NEEDSWAP      0x00010         /* if byte order requires swapping */
 348 #define B_NODUPS        0x00020         /* no duplicate keys permitted */
 349 #define B_RDONLY        0x00040         /* read-only tree */
 350 #define R_RECNO         0x00080         /* record oriented tree */
 351 #define B_SEQINIT       0x00100         /* sequential scan initialized */
 352
 353 #define R_CLOSEFP       0x00200         /* opened a file pointer */
 354 #define R_EOF           0x00400         /* end of input file reached. */
 355 #define R_FIXLEN        0x00800         /* fixed length records */
 356 #define R_MEMMAPPED     0x01000         /* memory mapped file. */
 357 #define R_INMEM         0x02000         /* in-memory file */
 358 #define R_MODIFIED      0x04000         /* modified file */
 359 #define R_RDONLY        0x08000         /* read-only file */
 360
 361 #define B_DB_LOCK       0x10000         /* DB_LOCK specified. */
 362 #define B_DB_SHMEM      0x20000         /* DB_SHMEM specified. */
 363 #define B_DB_TXN        0x40000         /* DB_TXN specified. */
 364
 365         u_int32_t       bt_flags;       /* btree state */
 366 } BTREE;
 367
 368 #define SET(t, f)       ((t)->bt_flags |= (f))
 369 #define CLR(t, f)       ((t)->bt_flags &= ~(f))
 370 #define ISSET(t, f)     ((t)->bt_flags & (f))
 371
 372 #include "bt_extern.h"