1 /*- 2 * Copyright (c) 1991, 1993 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 * %sccs.include.redist.c% 9 * 10 * @(#)btree.h 8.5 (Berkeley) 02/21/94 11 */ 12 13 #include <mpool.h> 14 15 #define DEFMINKEYPAGE (2) /* Minimum keys per page */ 16 #define MINCACHE (5) /* Minimum cached pages */ 17 #define MINPSIZE (512) /* Minimum page size */ 18 19 /* 20 * Page 0 of a btree file contains a copy of the meta-data. This page is also 21 * used as an out-of-band page, i.e. page pointers that point to nowhere point 22 * to page 0. Page 1 is the root of the btree. 23 */ 24 #define P_INVALID 0 /* Invalid tree page number. */ 25 #define P_META 0 /* Tree metadata page number. */ 26 #define P_ROOT 1 /* Tree root page number. */ 27 28 /* 29 * There are five page layouts in the btree: btree internal pages (BINTERNAL), 30 * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages 31 * (RLEAF) and overflow pages. All five page types have a page header (PAGE). 32 * This implementation requires that values within structures NOT be padded. 33 * (ANSI C permits random padding.) If your compiler pads randomly you'll have 34 * to do some work to get this package to run. 35 */ 36 typedef struct _page { 37 pgno_t pgno; /* this page's page number */ 38 pgno_t prevpg; /* left sibling */ 39 pgno_t nextpg; /* right sibling */ 40 41 #define P_BINTERNAL 0x01 /* btree internal page */ 42 #define P_BLEAF 0x02 /* leaf page */ 43 #define P_OVERFLOW 0x04 /* overflow page */ 44 #define P_RINTERNAL 0x08 /* recno internal page */ 45 #define P_RLEAF 0x10 /* leaf page */ 46 #define P_TYPE 0x1f /* type mask */ 47 #define P_PRESERVE 0x20 /* never delete this chain of pages */ 48 u_int32_t flags; 49 50 indx_t lower; /* lower bound of free space on page */ 51 indx_t upper; /* upper bound of free space on page */ 52 indx_t linp[1]; /* indx_t-aligned VAR. LENGTH DATA */ 53 } PAGE; 54 55 /* First and next index. */ 56 #define BTDATAOFF (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \ 57 sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t)) 58 #define NEXTINDEX(p) (((p)->lower - BTDATAOFF) / sizeof(indx_t)) 59 60 /* 61 * For pages other than overflow pages, there is an array of offsets into the 62 * rest of the page immediately following the page header. Each offset is to 63 * an item which is unique to the type of page. The h_lower offset is just 64 * past the last filled-in index. The h_upper offset is the first item on the 65 * page. Offsets are from the beginning of the page. 66 * 67 * If an item is too big to store on a single page, a flag is set and the item 68 * is a { page, size } pair such that the page is the first page of an overflow 69 * chain with size bytes of item. Overflow pages are simply bytes without any 70 * external structure. 71 * 72 * The page number and size fields in the items are pgno_t-aligned so they can 73 * be manipulated without copying. (This presumes that 32 bit items can be 74 * manipulated on this system.) 75 */ 76 #define LALIGN(n) \ 77 (((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1)) 78 #define NOVFLSIZE (sizeof(pgno_t) + sizeof(size_t)) 79 80 /* 81 * For the btree internal pages, the item is a key. BINTERNALs are {key, pgno} 82 * pairs, such that the key compares less than or equal to all of the records 83 * on that page. For a tree without duplicate keys, an internal page with two 84 * consecutive keys, a and b, will have all records greater than or equal to a 85 * and less than b stored on the page associated with a. Duplicate keys are 86 * somewhat special and can cause duplicate internal and leaf page records and 87 * some minor modifications of the above rule. 88 */ 89 typedef struct _binternal { 90 size_t ksize; /* key size */ 91 pgno_t pgno; /* page number stored on */ 92 #define P_BIGDATA 0x01 /* overflow data */ 93 #define P_BIGKEY 0x02 /* overflow key */ 94 u_char flags; 95 char bytes[1]; /* data */ 96 } BINTERNAL; 97 98 /* Get the page's BINTERNAL structure at index indx. */ 99 #define GETBINTERNAL(pg, indx) \ 100 ((BINTERNAL *)((char *)(pg) + (pg)->linp[indx])) 101 102 /* Get the number of bytes in the entry. */ 103 #define NBINTERNAL(len) \ 104 LALIGN(sizeof(size_t) + sizeof(pgno_t) + sizeof(u_char) + (len)) 105 106 /* Copy a BINTERNAL entry to the page. */ 107 #define WR_BINTERNAL(p, size, pgno, flags) { \ 108 *(size_t *)p = size; \ 109 p += sizeof(size_t); \ 110 *(pgno_t *)p = pgno; \ 111 p += sizeof(pgno_t); \ 112 *(u_char *)p = flags; \ 113 p += sizeof(u_char); \ 114 } 115 116 /* 117 * For the recno internal pages, the item is a page number with the number of 118 * keys found on that page and below. 119 */ 120 typedef struct _rinternal { 121 recno_t nrecs; /* number of records */ 122 pgno_t pgno; /* page number stored below */ 123 } RINTERNAL; 124 125 /* Get the page's RINTERNAL structure at index indx. */ 126 #define GETRINTERNAL(pg, indx) \ 127 ((RINTERNAL *)((char *)(pg) + (pg)->linp[indx])) 128 129 /* Get the number of bytes in the entry. */ 130 #define NRINTERNAL \ 131 LALIGN(sizeof(recno_t) + sizeof(pgno_t)) 132 133 /* Copy a RINTERAL entry to the page. */ 134 #define WR_RINTERNAL(p, nrecs, pgno) { \ 135 *(recno_t *)p = nrecs; \ 136 p += sizeof(recno_t); \ 137 *(pgno_t *)p = pgno; \ 138 } 139 140 /* For the btree leaf pages, the item is a key and data pair. */ 141 typedef struct _bleaf { 142 size_t ksize; /* size of key */ 143 size_t dsize; /* size of data */ 144 u_char flags; /* P_BIGDATA, P_BIGKEY */ 145 char bytes[1]; /* data */ 146 } BLEAF; 147 148 /* Get the page's BLEAF structure at index indx. */ 149 #define GETBLEAF(pg, indx) \ 150 ((BLEAF *)((char *)(pg) + (pg)->linp[indx])) 151 152 /* Get the number of bytes in the entry. */ 153 #define NBLEAF(p) NBLEAFDBT((p)->ksize, (p)->dsize) 154 155 /* Get the number of bytes in the user's key/data pair. */ 156 #define NBLEAFDBT(ksize, dsize) \ 157 LALIGN(sizeof(size_t) + sizeof(size_t) + sizeof(u_char) + \ 158 (ksize) + (dsize)) 159 160 /* Copy a BLEAF entry to the page. */ 161 #define WR_BLEAF(p, key, data, flags) { \ 162 *(size_t *)p = key->size; \ 163 p += sizeof(size_t); \ 164 *(size_t *)p = data->size; \ 165 p += sizeof(size_t); \ 166 *(u_char *)p = flags; \ 167 p += sizeof(u_char); \ 168 memmove(p, key->data, key->size); \ 169 p += key->size; \ 170 memmove(p, data->data, data->size); \ 171 } 172 173 /* For the recno leaf pages, the item is a data entry. */ 174 typedef struct _rleaf { 175 size_t dsize; /* size of data */ 176 u_char flags; /* P_BIGDATA */ 177 char bytes[1]; 178 } RLEAF; 179 180 /* Get the page's RLEAF structure at index indx. */ 181 #define GETRLEAF(pg, indx) \ 182 ((RLEAF *)((char *)(pg) + (pg)->linp[indx])) 183 184 /* Get the number of bytes in the entry. */ 185 #define NRLEAF(p) NRLEAFDBT((p)->dsize) 186 187 /* Get the number of bytes from the user's data. */ 188 #define NRLEAFDBT(dsize) \ 189 LALIGN(sizeof(size_t) + sizeof(u_char) + (dsize)) 190 191 /* Copy a RLEAF entry to the page. */ 192 #define WR_RLEAF(p, data, flags) { \ 193 *(size_t *)p = data->size; \ 194 p += sizeof(size_t); \ 195 *(u_char *)p = flags; \ 196 p += sizeof(u_char); \ 197 memmove(p, data->data, data->size); \ 198 } 199 200 /* 201 * A record in the tree is either a pointer to a page and an index in the page 202 * or a page number and an index. These structures are used as a cursor, stack 203 * entry and search returns as well as to pass records to other routines. 204 * 205 * One comment about searches. Internal page searches must find the largest 206 * record less than key in the tree so that descents work. Leaf page searches 207 * must find the smallest record greater than key so that the returned index 208 * is the record's correct position for insertion. 209 * 210 * One comment about cursors. The cursor key is never removed from the tree, 211 * even if deleted. This is because it is quite difficult to decide where the 212 * cursor should be when other keys have been inserted/deleted in the tree; 213 * duplicate keys make it impossible. This scheme does require extra work 214 * though, to make sure that we don't perform an operation on a deleted key. 215 */ 216 typedef struct _epgno { 217 pgno_t pgno; /* the page number */ 218 indx_t index; /* the index on the page */ 219 } EPGNO; 220 221 typedef struct _epg { 222 PAGE *page; /* the (pinned) page */ 223 indx_t index; /* the index on the page */ 224 } EPG; 225 226 /* 227 * The metadata of the tree. The m_nrecs field is used only by the RECNO code. 228 * This is because the btree doesn't really need it and it requires that every 229 * put or delete call modify the metadata. 230 */ 231 typedef struct _btmeta { 232 u_int32_t m_magic; /* magic number */ 233 u_int32_t m_version; /* version */ 234 u_int32_t m_psize; /* page size */ 235 u_int32_t m_free; /* page number of first free page */ 236 u_int32_t m_nrecs; /* R: number of records */ 237 #define SAVEMETA (B_NODUPS | R_RECNO) 238 u_int32_t m_flags; /* bt_flags & SAVEMETA */ 239 u_int32_t m_unused; /* unused */ 240 } BTMETA; 241 242 /* The in-memory btree/recno data structure. */ 243 typedef struct _btree { 244 MPOOL *bt_mp; /* memory pool cookie */ 245 246 DB *bt_dbp; /* pointer to enclosing DB */ 247 248 EPG bt_cur; /* current (pinned) page */ 249 PAGE *bt_pinned; /* page pinned across calls */ 250 251 EPGNO bt_bcursor; /* B: btree cursor */ 252 recno_t bt_rcursor; /* R: recno cursor (1-based) */ 253 254 #define BT_POP(t) (t->bt_sp ? t->bt_stack + --t->bt_sp : NULL) 255 #define BT_CLR(t) (t->bt_sp = 0) 256 EPGNO *bt_stack; /* stack of parent pages */ 257 u_int bt_sp; /* current stack pointer */ 258 u_int bt_maxstack; /* largest stack */ 259 260 char *bt_kbuf; /* key buffer */ 261 size_t bt_kbufsz; /* key buffer size */ 262 char *bt_dbuf; /* data buffer */ 263 size_t bt_dbufsz; /* data buffer size */ 264 265 int bt_fd; /* tree file descriptor */ 266 267 pgno_t bt_free; /* next free page */ 268 u_int32_t bt_psize; /* page size */ 269 indx_t bt_ovflsize; /* cut-off for key/data overflow */ 270 int bt_lorder; /* byte order */ 271 /* sorted order */ 272 enum { NOT, BACK, FORWARD } bt_order; 273 EPGNO bt_last; /* last insert */ 274 275 /* B: key comparison function */ 276 int (*bt_cmp) __P((const DBT *, const DBT *)); 277 /* B: prefix comparison function */ 278 size_t (*bt_pfx) __P((const DBT *, const DBT *)); 279 /* R: recno input function */ 280 int (*bt_irec) __P((struct _btree *, recno_t)); 281 282 FILE *bt_rfp; /* R: record FILE pointer */ 283 int bt_rfd; /* R: record file descriptor */ 284 285 caddr_t bt_cmap; /* R: current point in mapped space */ 286 caddr_t bt_smap; /* R: start of mapped space */ 287 caddr_t bt_emap; /* R: end of mapped space */ 288 size_t bt_msize; /* R: size of mapped region. */ 289 290 recno_t bt_nrecs; /* R: number of records */ 291 size_t bt_reclen; /* R: fixed record length */ 292 u_char bt_bval; /* R: delimiting byte/pad character */ 293 294 /* 295 * NB: 296 * B_NODUPS and R_RECNO are stored on disk, and may not be changed. 297 */ 298 #define B_DELCRSR 0x00001 /* cursor has been deleted */ 299 #define B_INMEM 0x00002 /* in-memory tree */ 300 #define B_METADIRTY 0x00004 /* need to write metadata */ 301 #define B_MODIFIED 0x00008 /* tree modified */ 302 #define B_NEEDSWAP 0x00010 /* if byte order requires swapping */ 303 #define B_NODUPS 0x00020 /* no duplicate keys permitted */ 304 #define B_RDONLY 0x00040 /* read-only tree */ 305 #define R_RECNO 0x00080 /* record oriented tree */ 306 #define B_SEQINIT 0x00100 /* sequential scan initialized */ 307 308 #define R_CLOSEFP 0x00200 /* opened a file pointer */ 309 #define R_EOF 0x00400 /* end of input file reached. */ 310 #define R_FIXLEN 0x00800 /* fixed length records */ 311 #define R_MEMMAPPED 0x01000 /* memory mapped file. */ 312 #define R_INMEM 0x02000 /* in-memory file */ 313 #define R_MODIFIED 0x04000 /* modified file */ 314 #define R_RDONLY 0x08000 /* read-only file */ 315 316 #define B_DB_LOCK 0x10000 /* DB_LOCK specified. */ 317 #define B_DB_SHMEM 0x20000 /* DB_SHMEM specified. */ 318 #define B_DB_TXN 0x40000 /* DB_TXN specified. */ 319 320 u_int32_t bt_flags; /* btree state */ 321 } BTREE; 322 323 #define SET(t, f) ((t)->bt_flags |= (f)) 324 #define CLR(t, f) ((t)->bt_flags &= ~(f)) 325 #define ISSET(t, f) ((t)->bt_flags & (f)) 326 327 #include "extern.h" 328