1 /* $NetBSD: btree.h,v 1.14 2003/10/27 00:12:42 lukem Exp $ */ 2 3 /*- 4 * Copyright (c) 1991, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Mike Olson. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)btree.h 8.11 (Berkeley) 8/17/94 35 */ 36 37 #if HAVE_NBTOOL_CONFIG_H 38 #include "nbtool_config.h" 39 #endif 40 41 /* Macros to set/clear/test flags. */ 42 #define F_SET(p, f) (p)->flags |= (f) 43 #define F_CLR(p, f) (p)->flags &= ~(f) 44 #define F_ISSET(p, f) ((p)->flags & (f)) 45 46 #include <mpool.h> 47 48 #define DEFMINKEYPAGE (2) /* Minimum keys per page */ 49 #define MINCACHE (5) /* Minimum cached pages */ 50 #define MINPSIZE (512) /* Minimum page size */ 51 52 /* 53 * Page 0 of a btree file contains a copy of the meta-data. This page is also 54 * used as an out-of-band page, i.e. page pointers that point to nowhere point 55 * to page 0. Page 1 is the root of the btree. 56 */ 57 #define P_INVALID 0 /* Invalid tree page number. */ 58 #define P_META 0 /* Tree metadata page number. */ 59 #define P_ROOT 1 /* Tree root page number. */ 60 61 /* 62 * There are five page layouts in the btree: btree internal pages (BINTERNAL), 63 * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages 64 * (RLEAF) and overflow pages. All five page types have a page header (PAGE). 65 * This implementation requires that values within structures NOT be padded. 66 * (ANSI C permits random padding.) If your compiler pads randomly you'll have 67 * to do some work to get this package to run. 68 */ 69 typedef struct _page { 70 pgno_t pgno; /* this page's page number */ 71 pgno_t prevpg; /* left sibling */ 72 pgno_t nextpg; /* right sibling */ 73 74 #define P_BINTERNAL 0x01 /* btree internal page */ 75 #define P_BLEAF 0x02 /* leaf page */ 76 #define P_OVERFLOW 0x04 /* overflow page */ 77 #define P_RINTERNAL 0x08 /* recno internal page */ 78 #define P_RLEAF 0x10 /* leaf page */ 79 #define P_TYPE 0x1f /* type mask */ 80 #define P_PRESERVE 0x20 /* never delete this chain of pages */ 81 u_int32_t flags; 82 83 indx_t lower; /* lower bound of free space on page */ 84 indx_t upper; /* upper bound of free space on page */ 85 indx_t linp[1]; /* indx_t-aligned VAR. LENGTH DATA */ 86 } PAGE; 87 88 /* First and next index. */ 89 #define BTDATAOFF \ 90 (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \ 91 sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t)) 92 #define NEXTINDEX(p) (((p)->lower - BTDATAOFF) / sizeof(indx_t)) 93 94 /* 95 * For pages other than overflow pages, there is an array of offsets into the 96 * rest of the page immediately following the page header. Each offset is to 97 * an item which is unique to the type of page. The h_lower offset is just 98 * past the last filled-in index. The h_upper offset is the first item on the 99 * page. Offsets are from the beginning of the page. 100 * 101 * If an item is too big to store on a single page, a flag is set and the item 102 * is a { page, size } pair such that the page is the first page of an overflow 103 * chain with size bytes of item. Overflow pages are simply bytes without any 104 * external structure. 105 * 106 * The page number and size fields in the items are pgno_t-aligned so they can 107 * be manipulated without copying. (This presumes that 32 bit items can be 108 * manipulated on this system.) 109 */ 110 #define BTLALIGN(n) (((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1)) 111 #define NOVFLSIZE (sizeof(pgno_t) + sizeof(u_int32_t)) 112 113 /* 114 * For the btree internal pages, the item is a key. BINTERNALs are {key, pgno} 115 * pairs, such that the key compares less than or equal to all of the records 116 * on that page. For a tree without duplicate keys, an internal page with two 117 * consecutive keys, a and b, will have all records greater than or equal to a 118 * and less than b stored on the page associated with a. Duplicate keys are 119 * somewhat special and can cause duplicate internal and leaf page records and 120 * some minor modifications of the above rule. 121 */ 122 typedef struct _binternal { 123 u_int32_t ksize; /* key size */ 124 pgno_t pgno; /* page number stored on */ 125 #define P_BIGDATA 0x01 /* overflow data */ 126 #define P_BIGKEY 0x02 /* overflow key */ 127 u_char flags; 128 char bytes[1]; /* data */ 129 } BINTERNAL; 130 131 /* Get the page's BINTERNAL structure at index indx. */ 132 #define GETBINTERNAL(pg, indx) \ 133 ((BINTERNAL *)(void *)((char *)(void *)(pg) + (pg)->linp[indx])) 134 135 /* Get the number of bytes in the entry. */ 136 #define NBINTERNAL(len) \ 137 BTLALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len)) 138 139 /* Copy a BINTERNAL entry to the page. */ 140 #define WR_BINTERNAL(p, size, pgno, flags) { \ 141 *(u_int32_t *)(void *)p = size; \ 142 p += sizeof(u_int32_t); \ 143 *(pgno_t *)(void *)p = pgno; \ 144 p += sizeof(pgno_t); \ 145 *(u_char *)(void *)p = flags; \ 146 p += sizeof(u_char); \ 147 } 148 149 /* 150 * For the recno internal pages, the item is a page number with the number of 151 * keys found on that page and below. 152 */ 153 typedef struct _rinternal { 154 recno_t nrecs; /* number of records */ 155 pgno_t pgno; /* page number stored below */ 156 } RINTERNAL; 157 158 /* Get the page's RINTERNAL structure at index indx. */ 159 #define GETRINTERNAL(pg, indx) \ 160 ((RINTERNAL *)(void *)((char *)(void *)(pg) + (pg)->linp[indx])) 161 162 /* Get the number of bytes in the entry. */ 163 #define NRINTERNAL \ 164 BTLALIGN(sizeof(recno_t) + sizeof(pgno_t)) 165 166 /* Copy a RINTERAL entry to the page. */ 167 #define WR_RINTERNAL(p, nrecs, pgno) { \ 168 *(recno_t *)(void *)p = nrecs; \ 169 p += sizeof(recno_t); \ 170 *(pgno_t *)(void *)p = pgno; \ 171 } 172 173 /* For the btree leaf pages, the item is a key and data pair. */ 174 typedef struct _bleaf { 175 u_int32_t ksize; /* size of key */ 176 u_int32_t dsize; /* size of data */ 177 u_char flags; /* P_BIGDATA, P_BIGKEY */ 178 char bytes[1]; /* data */ 179 } BLEAF; 180 181 /* Get the page's BLEAF structure at index indx. */ 182 #define GETBLEAF(pg, indx) \ 183 ((BLEAF *)(void *)((char *)(void *)(pg) + (pg)->linp[indx])) 184 185 /* Get the number of bytes in the entry. */ 186 #define NBLEAF(p) NBLEAFDBT((p)->ksize, (p)->dsize) 187 188 /* Get the number of bytes in the user's key/data pair. */ 189 #define NBLEAFDBT(ksize, dsize) \ 190 BTLALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) + \ 191 (ksize) + (dsize)) 192 193 /* Copy a BLEAF entry to the page. */ 194 #define WR_BLEAF(p, key, data, flags) { \ 195 *(u_int32_t *)(void *)p = key->size; \ 196 p += sizeof(u_int32_t); \ 197 *(u_int32_t *)(void *)p = data->size; \ 198 p += sizeof(u_int32_t); \ 199 *(u_char *)(void *)p = flags; \ 200 p += sizeof(u_char); \ 201 memmove(p, key->data, key->size); \ 202 p += key->size; \ 203 memmove(p, data->data, data->size); \ 204 } 205 206 /* For the recno leaf pages, the item is a data entry. */ 207 typedef struct _rleaf { 208 u_int32_t dsize; /* size of data */ 209 u_char flags; /* P_BIGDATA */ 210 char bytes[1]; 211 } RLEAF; 212 213 /* Get the page's RLEAF structure at index indx. */ 214 #define GETRLEAF(pg, indx) \ 215 ((RLEAF *)(void *)((char *)(void *)(pg) + (pg)->linp[indx])) 216 217 /* Get the number of bytes in the entry. */ 218 #define NRLEAF(p) NRLEAFDBT((p)->dsize) 219 220 /* Get the number of bytes from the user's data. */ 221 #define NRLEAFDBT(dsize) \ 222 BTLALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize)) 223 224 /* Copy a RLEAF entry to the page. */ 225 #define WR_RLEAF(p, data, flags) { \ 226 *(u_int32_t *)(void *)p = data->size; \ 227 p += sizeof(u_int32_t); \ 228 *(u_char *)(void *)p = flags; \ 229 p += sizeof(u_char); \ 230 memmove(p, data->data, data->size); \ 231 } 232 233 /* 234 * A record in the tree is either a pointer to a page and an index in the page 235 * or a page number and an index. These structures are used as a cursor, stack 236 * entry and search returns as well as to pass records to other routines. 237 * 238 * One comment about searches. Internal page searches must find the largest 239 * record less than key in the tree so that descents work. Leaf page searches 240 * must find the smallest record greater than key so that the returned index 241 * is the record's correct position for insertion. 242 */ 243 typedef struct _epgno { 244 pgno_t pgno; /* the page number */ 245 indx_t index; /* the index on the page */ 246 } EPGNO; 247 248 typedef struct _epg { 249 PAGE *page; /* the (pinned) page */ 250 indx_t index; /* the index on the page */ 251 } EPG; 252 253 /* 254 * About cursors. The cursor (and the page that contained the key/data pair 255 * that it referenced) can be deleted, which makes things a bit tricky. If 256 * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set 257 * or there simply aren't any duplicates of the key) we copy the key that it 258 * referenced when it's deleted, and reacquire a new cursor key if the cursor 259 * is used again. If there are duplicates keys, we move to the next/previous 260 * key, and set a flag so that we know what happened. NOTE: if duplicate (to 261 * the cursor) keys are added to the tree during this process, it is undefined 262 * if they will be returned or not in a cursor scan. 263 * 264 * The flags determine the possible states of the cursor: 265 * 266 * CURS_INIT The cursor references *something*. 267 * CURS_ACQUIRE The cursor was deleted, and a key has been saved so that 268 * we can reacquire the right position in the tree. 269 * CURS_AFTER, CURS_BEFORE 270 * The cursor was deleted, and now references a key/data pair 271 * that has not yet been returned, either before or after the 272 * deleted key/data pair. 273 * XXX 274 * This structure is broken out so that we can eventually offer multiple 275 * cursors as part of the DB interface. 276 */ 277 typedef struct _cursor { 278 EPGNO pg; /* B: Saved tree reference. */ 279 DBT key; /* B: Saved key, or key.data == NULL. */ 280 recno_t rcursor; /* R: recno cursor (1-based) */ 281 282 #define CURS_ACQUIRE 0x01 /* B: Cursor needs to be reacquired. */ 283 #define CURS_AFTER 0x02 /* B: Unreturned cursor after key. */ 284 #define CURS_BEFORE 0x04 /* B: Unreturned cursor before key. */ 285 #define CURS_INIT 0x08 /* RB: Cursor initialized. */ 286 u_int8_t flags; 287 } CURSOR; 288 289 /* 290 * The metadata of the tree. The nrecs field is used only by the RECNO code. 291 * This is because the btree doesn't really need it and it requires that every 292 * put or delete call modify the metadata. 293 */ 294 typedef struct _btmeta { 295 u_int32_t magic; /* magic number */ 296 u_int32_t version; /* version */ 297 u_int32_t psize; /* page size */ 298 u_int32_t free; /* page number of first free page */ 299 u_int32_t nrecs; /* R: number of records */ 300 301 #define SAVEMETA (B_NODUPS | R_RECNO) 302 u_int32_t flags; /* bt_flags & SAVEMETA */ 303 } BTMETA; 304 305 /* The in-memory btree/recno data structure. */ 306 typedef struct _btree { 307 MPOOL *bt_mp; /* memory pool cookie */ 308 309 DB *bt_dbp; /* pointer to enclosing DB */ 310 311 EPG bt_cur; /* current (pinned) page */ 312 PAGE *bt_pinned; /* page pinned across calls */ 313 314 CURSOR bt_cursor; /* cursor */ 315 316 #define BT_PUSH(t, p, i) { \ 317 t->bt_sp->pgno = p; \ 318 t->bt_sp->index = i; \ 319 ++t->bt_sp; \ 320 } 321 #define BT_POP(t) (t->bt_sp == t->bt_stack ? NULL : --t->bt_sp) 322 #define BT_CLR(t) (t->bt_sp = t->bt_stack) 323 EPGNO bt_stack[50]; /* stack of parent pages */ 324 EPGNO *bt_sp; /* current stack pointer */ 325 326 DBT bt_rkey; /* returned key */ 327 DBT bt_rdata; /* returned data */ 328 329 int bt_fd; /* tree file descriptor */ 330 331 pgno_t bt_free; /* next free page */ 332 u_int32_t bt_psize; /* page size */ 333 indx_t bt_ovflsize; /* cut-off for key/data overflow */ 334 int bt_lorder; /* byte order */ 335 /* sorted order */ 336 enum { NOT, BACK, FORWARD } bt_order; 337 EPGNO bt_last; /* last insert */ 338 339 /* B: key comparison function */ 340 int (*bt_cmp) __P((const DBT *, const DBT *)); 341 /* B: prefix comparison function */ 342 size_t (*bt_pfx) __P((const DBT *, const DBT *)); 343 /* R: recno input function */ 344 int (*bt_irec) __P((struct _btree *, recno_t)); 345 346 FILE *bt_rfp; /* R: record FILE pointer */ 347 int bt_rfd; /* R: record file descriptor */ 348 349 caddr_t bt_cmap; /* R: current point in mapped space */ 350 caddr_t bt_smap; /* R: start of mapped space */ 351 caddr_t bt_emap; /* R: end of mapped space */ 352 size_t bt_msize; /* R: size of mapped region. */ 353 354 recno_t bt_nrecs; /* R: number of records */ 355 size_t bt_reclen; /* R: fixed record length */ 356 u_char bt_bval; /* R: delimiting byte/pad character */ 357 358 /* 359 * NB: 360 * B_NODUPS and R_RECNO are stored on disk, and may not be changed. 361 */ 362 #define B_INMEM 0x00001 /* in-memory tree */ 363 #define B_METADIRTY 0x00002 /* need to write metadata */ 364 #define B_MODIFIED 0x00004 /* tree modified */ 365 #define B_NEEDSWAP 0x00008 /* if byte order requires swapping */ 366 #define B_RDONLY 0x00010 /* read-only tree */ 367 368 #define B_NODUPS 0x00020 /* no duplicate keys permitted */ 369 #define R_RECNO 0x00080 /* record oriented tree */ 370 371 #define R_CLOSEFP 0x00040 /* opened a file pointer */ 372 #define R_EOF 0x00100 /* end of input file reached. */ 373 #define R_FIXLEN 0x00200 /* fixed length records */ 374 #define R_MEMMAPPED 0x00400 /* memory mapped file. */ 375 #define R_INMEM 0x00800 /* in-memory file */ 376 #define R_MODIFIED 0x01000 /* modified file */ 377 #define R_RDONLY 0x02000 /* read-only file */ 378 379 #define B_DB_LOCK 0x04000 /* DB_LOCK specified. */ 380 #define B_DB_SHMEM 0x08000 /* DB_SHMEM specified. */ 381 #define B_DB_TXN 0x10000 /* DB_TXN specified. */ 382 u_int32_t flags; 383 } BTREE; 384 385 #include "extern.h" 386