btree.h revision 14272
11573Srgrimes/*- 214272Spst * Copyright (c) 1991, 1993, 1994 31573Srgrimes * The Regents of the University of California. All rights reserved. 41573Srgrimes * 51573Srgrimes * This code is derived from software contributed to Berkeley by 61573Srgrimes * Mike Olson. 71573Srgrimes * 81573Srgrimes * Redistribution and use in source and binary forms, with or without 91573Srgrimes * modification, are permitted provided that the following conditions 101573Srgrimes * are met: 111573Srgrimes * 1. Redistributions of source code must retain the above copyright 121573Srgrimes * notice, this list of conditions and the following disclaimer. 131573Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 141573Srgrimes * notice, this list of conditions and the following disclaimer in the 151573Srgrimes * documentation and/or other materials provided with the distribution. 161573Srgrimes * 3. All advertising materials mentioning features or use of this software 171573Srgrimes * must display the following acknowledgement: 181573Srgrimes * This product includes software developed by the University of 191573Srgrimes * California, Berkeley and its contributors. 201573Srgrimes * 4. Neither the name of the University nor the names of its contributors 211573Srgrimes * may be used to endorse or promote products derived from this software 221573Srgrimes * without specific prior written permission. 231573Srgrimes * 241573Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 251573Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 261573Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 271573Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 281573Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 291573Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 301573Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 311573Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 321573Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 331573Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 341573Srgrimes * SUCH DAMAGE. 351573Srgrimes * 3614272Spst * @(#)btree.h 8.11 (Berkeley) 8/17/94 371573Srgrimes */ 381573Srgrimes 3914272Spst/* Macros to set/clear/test flags. */ 4014272Spst#define F_SET(p, f) (p)->flags |= (f) 4114272Spst#define F_CLR(p, f) (p)->flags &= ~(f) 4214272Spst#define F_ISSET(p, f) ((p)->flags & (f)) 4314272Spst 441573Srgrimes#include <mpool.h> 451573Srgrimes 461573Srgrimes#define DEFMINKEYPAGE (2) /* Minimum keys per page */ 471573Srgrimes#define MINCACHE (5) /* Minimum cached pages */ 481573Srgrimes#define MINPSIZE (512) /* Minimum page size */ 491573Srgrimes 501573Srgrimes/* 511573Srgrimes * Page 0 of a btree file contains a copy of the meta-data. This page is also 521573Srgrimes * used as an out-of-band page, i.e. page pointers that point to nowhere point 531573Srgrimes * to page 0. Page 1 is the root of the btree. 541573Srgrimes */ 551573Srgrimes#define P_INVALID 0 /* Invalid tree page number. */ 561573Srgrimes#define P_META 0 /* Tree metadata page number. */ 571573Srgrimes#define P_ROOT 1 /* Tree root page number. */ 581573Srgrimes 591573Srgrimes/* 601573Srgrimes * There are five page layouts in the btree: btree internal pages (BINTERNAL), 611573Srgrimes * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages 621573Srgrimes * (RLEAF) and overflow pages. All five page types have a page header (PAGE). 631573Srgrimes * This implementation requires that values within structures NOT be padded. 641573Srgrimes * (ANSI C permits random padding.) If your compiler pads randomly you'll have 651573Srgrimes * to do some work to get this package to run. 661573Srgrimes */ 671573Srgrimestypedef struct _page { 681573Srgrimes pgno_t pgno; /* this page's page number */ 691573Srgrimes pgno_t prevpg; /* left sibling */ 701573Srgrimes pgno_t nextpg; /* right sibling */ 711573Srgrimes 721573Srgrimes#define P_BINTERNAL 0x01 /* btree internal page */ 731573Srgrimes#define P_BLEAF 0x02 /* leaf page */ 741573Srgrimes#define P_OVERFLOW 0x04 /* overflow page */ 751573Srgrimes#define P_RINTERNAL 0x08 /* recno internal page */ 761573Srgrimes#define P_RLEAF 0x10 /* leaf page */ 771573Srgrimes#define P_TYPE 0x1f /* type mask */ 781573Srgrimes#define P_PRESERVE 0x20 /* never delete this chain of pages */ 791573Srgrimes u_int32_t flags; 801573Srgrimes 811573Srgrimes indx_t lower; /* lower bound of free space on page */ 821573Srgrimes indx_t upper; /* upper bound of free space on page */ 831573Srgrimes indx_t linp[1]; /* indx_t-aligned VAR. LENGTH DATA */ 841573Srgrimes} PAGE; 851573Srgrimes 861573Srgrimes/* First and next index. */ 8714272Spst#define BTDATAOFF \ 8814272Spst (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \ 8914272Spst sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t)) 901573Srgrimes#define NEXTINDEX(p) (((p)->lower - BTDATAOFF) / sizeof(indx_t)) 911573Srgrimes 921573Srgrimes/* 931573Srgrimes * For pages other than overflow pages, there is an array of offsets into the 941573Srgrimes * rest of the page immediately following the page header. Each offset is to 951573Srgrimes * an item which is unique to the type of page. The h_lower offset is just 961573Srgrimes * past the last filled-in index. The h_upper offset is the first item on the 971573Srgrimes * page. Offsets are from the beginning of the page. 981573Srgrimes * 991573Srgrimes * If an item is too big to store on a single page, a flag is set and the item 1001573Srgrimes * is a { page, size } pair such that the page is the first page of an overflow 1011573Srgrimes * chain with size bytes of item. Overflow pages are simply bytes without any 1021573Srgrimes * external structure. 1031573Srgrimes * 1041573Srgrimes * The page number and size fields in the items are pgno_t-aligned so they can 1051573Srgrimes * be manipulated without copying. (This presumes that 32 bit items can be 1061573Srgrimes * manipulated on this system.) 1071573Srgrimes */ 10814272Spst#define LALIGN(n) (((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1)) 10914272Spst#define NOVFLSIZE (sizeof(pgno_t) + sizeof(u_int32_t)) 1101573Srgrimes 1111573Srgrimes/* 1121573Srgrimes * For the btree internal pages, the item is a key. BINTERNALs are {key, pgno} 1131573Srgrimes * pairs, such that the key compares less than or equal to all of the records 1141573Srgrimes * on that page. For a tree without duplicate keys, an internal page with two 1151573Srgrimes * consecutive keys, a and b, will have all records greater than or equal to a 1161573Srgrimes * and less than b stored on the page associated with a. Duplicate keys are 1171573Srgrimes * somewhat special and can cause duplicate internal and leaf page records and 1181573Srgrimes * some minor modifications of the above rule. 1191573Srgrimes */ 1201573Srgrimestypedef struct _binternal { 12114272Spst u_int32_t ksize; /* key size */ 1221573Srgrimes pgno_t pgno; /* page number stored on */ 1231573Srgrimes#define P_BIGDATA 0x01 /* overflow data */ 1241573Srgrimes#define P_BIGKEY 0x02 /* overflow key */ 1251573Srgrimes u_char flags; 1261573Srgrimes char bytes[1]; /* data */ 1271573Srgrimes} BINTERNAL; 1281573Srgrimes 1291573Srgrimes/* Get the page's BINTERNAL structure at index indx. */ 13014272Spst#define GETBINTERNAL(pg, indx) \ 1311573Srgrimes ((BINTERNAL *)((char *)(pg) + (pg)->linp[indx])) 1321573Srgrimes 1331573Srgrimes/* Get the number of bytes in the entry. */ 13414272Spst#define NBINTERNAL(len) \ 13514272Spst LALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len)) 1361573Srgrimes 1371573Srgrimes/* Copy a BINTERNAL entry to the page. */ 13814272Spst#define WR_BINTERNAL(p, size, pgno, flags) { \ 13914272Spst *(u_int32_t *)p = size; \ 14014272Spst p += sizeof(u_int32_t); \ 14114272Spst *(pgno_t *)p = pgno; \ 14214272Spst p += sizeof(pgno_t); \ 14314272Spst *(u_char *)p = flags; \ 14414272Spst p += sizeof(u_char); \ 1451573Srgrimes} 1461573Srgrimes 1471573Srgrimes/* 1481573Srgrimes * For the recno internal pages, the item is a page number with the number of 1491573Srgrimes * keys found on that page and below. 1501573Srgrimes */ 1511573Srgrimestypedef struct _rinternal { 1521573Srgrimes recno_t nrecs; /* number of records */ 1531573Srgrimes pgno_t pgno; /* page number stored below */ 1541573Srgrimes} RINTERNAL; 1551573Srgrimes 1561573Srgrimes/* Get the page's RINTERNAL structure at index indx. */ 15714272Spst#define GETRINTERNAL(pg, indx) \ 1581573Srgrimes ((RINTERNAL *)((char *)(pg) + (pg)->linp[indx])) 1591573Srgrimes 1601573Srgrimes/* Get the number of bytes in the entry. */ 16114272Spst#define NRINTERNAL \ 1621573Srgrimes LALIGN(sizeof(recno_t) + sizeof(pgno_t)) 1631573Srgrimes 1641573Srgrimes/* Copy a RINTERAL entry to the page. */ 16514272Spst#define WR_RINTERNAL(p, nrecs, pgno) { \ 16614272Spst *(recno_t *)p = nrecs; \ 16714272Spst p += sizeof(recno_t); \ 16814272Spst *(pgno_t *)p = pgno; \ 1691573Srgrimes} 1701573Srgrimes 1711573Srgrimes/* For the btree leaf pages, the item is a key and data pair. */ 1721573Srgrimestypedef struct _bleaf { 17314272Spst u_int32_t ksize; /* size of key */ 17414272Spst u_int32_t dsize; /* size of data */ 1751573Srgrimes u_char flags; /* P_BIGDATA, P_BIGKEY */ 1761573Srgrimes char bytes[1]; /* data */ 1771573Srgrimes} BLEAF; 1781573Srgrimes 1791573Srgrimes/* Get the page's BLEAF structure at index indx. */ 18014272Spst#define GETBLEAF(pg, indx) \ 1811573Srgrimes ((BLEAF *)((char *)(pg) + (pg)->linp[indx])) 1821573Srgrimes 1831573Srgrimes/* Get the number of bytes in the entry. */ 1841573Srgrimes#define NBLEAF(p) NBLEAFDBT((p)->ksize, (p)->dsize) 1851573Srgrimes 1861573Srgrimes/* Get the number of bytes in the user's key/data pair. */ 18714272Spst#define NBLEAFDBT(ksize, dsize) \ 18814272Spst LALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) + \ 1891573Srgrimes (ksize) + (dsize)) 1901573Srgrimes 1911573Srgrimes/* Copy a BLEAF entry to the page. */ 19214272Spst#define WR_BLEAF(p, key, data, flags) { \ 19314272Spst *(u_int32_t *)p = key->size; \ 19414272Spst p += sizeof(u_int32_t); \ 19514272Spst *(u_int32_t *)p = data->size; \ 19614272Spst p += sizeof(u_int32_t); \ 19714272Spst *(u_char *)p = flags; \ 19814272Spst p += sizeof(u_char); \ 19914272Spst memmove(p, key->data, key->size); \ 20014272Spst p += key->size; \ 20114272Spst memmove(p, data->data, data->size); \ 2021573Srgrimes} 2031573Srgrimes 2041573Srgrimes/* For the recno leaf pages, the item is a data entry. */ 2051573Srgrimestypedef struct _rleaf { 20614272Spst u_int32_t dsize; /* size of data */ 2071573Srgrimes u_char flags; /* P_BIGDATA */ 2081573Srgrimes char bytes[1]; 2091573Srgrimes} RLEAF; 2101573Srgrimes 2111573Srgrimes/* Get the page's RLEAF structure at index indx. */ 21214272Spst#define GETRLEAF(pg, indx) \ 2131573Srgrimes ((RLEAF *)((char *)(pg) + (pg)->linp[indx])) 2141573Srgrimes 2151573Srgrimes/* Get the number of bytes in the entry. */ 2161573Srgrimes#define NRLEAF(p) NRLEAFDBT((p)->dsize) 2171573Srgrimes 2181573Srgrimes/* Get the number of bytes from the user's data. */ 21914272Spst#define NRLEAFDBT(dsize) \ 22014272Spst LALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize)) 2211573Srgrimes 2221573Srgrimes/* Copy a RLEAF entry to the page. */ 22314272Spst#define WR_RLEAF(p, data, flags) { \ 22414272Spst *(u_int32_t *)p = data->size; \ 22514272Spst p += sizeof(u_int32_t); \ 22614272Spst *(u_char *)p = flags; \ 22714272Spst p += sizeof(u_char); \ 22814272Spst memmove(p, data->data, data->size); \ 2291573Srgrimes} 2301573Srgrimes 2311573Srgrimes/* 2321573Srgrimes * A record in the tree is either a pointer to a page and an index in the page 2331573Srgrimes * or a page number and an index. These structures are used as a cursor, stack 2341573Srgrimes * entry and search returns as well as to pass records to other routines. 2351573Srgrimes * 2361573Srgrimes * One comment about searches. Internal page searches must find the largest 2371573Srgrimes * record less than key in the tree so that descents work. Leaf page searches 2381573Srgrimes * must find the smallest record greater than key so that the returned index 2391573Srgrimes * is the record's correct position for insertion. 2401573Srgrimes */ 2411573Srgrimestypedef struct _epgno { 2421573Srgrimes pgno_t pgno; /* the page number */ 2431573Srgrimes indx_t index; /* the index on the page */ 2441573Srgrimes} EPGNO; 2451573Srgrimes 2461573Srgrimestypedef struct _epg { 2471573Srgrimes PAGE *page; /* the (pinned) page */ 2481573Srgrimes indx_t index; /* the index on the page */ 2491573Srgrimes} EPG; 2501573Srgrimes 2511573Srgrimes/* 25214272Spst * About cursors. The cursor (and the page that contained the key/data pair 25314272Spst * that it referenced) can be deleted, which makes things a bit tricky. If 25414272Spst * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set 25514272Spst * or there simply aren't any duplicates of the key) we copy the key that it 25614272Spst * referenced when it's deleted, and reacquire a new cursor key if the cursor 25714272Spst * is used again. If there are duplicates keys, we move to the next/previous 25814272Spst * key, and set a flag so that we know what happened. NOTE: if duplicate (to 25914272Spst * the cursor) keys are added to the tree during this process, it is undefined 26014272Spst * if they will be returned or not in a cursor scan. 26114272Spst * 26214272Spst * The flags determine the possible states of the cursor: 26314272Spst * 26414272Spst * CURS_INIT The cursor references *something*. 26514272Spst * CURS_ACQUIRE The cursor was deleted, and a key has been saved so that 26614272Spst * we can reacquire the right position in the tree. 26714272Spst * CURS_AFTER, CURS_BEFORE 26814272Spst * The cursor was deleted, and now references a key/data pair 26914272Spst * that has not yet been returned, either before or after the 27014272Spst * deleted key/data pair. 27114272Spst * XXX 27214272Spst * This structure is broken out so that we can eventually offer multiple 27314272Spst * cursors as part of the DB interface. 27414272Spst */ 27514272Spsttypedef struct _cursor { 27614272Spst EPGNO pg; /* B: Saved tree reference. */ 27714272Spst DBT key; /* B: Saved key, or key.data == NULL. */ 27814272Spst recno_t rcursor; /* R: recno cursor (1-based) */ 27914272Spst 28014272Spst#define CURS_ACQUIRE 0x01 /* B: Cursor needs to be reacquired. */ 28114272Spst#define CURS_AFTER 0x02 /* B: Unreturned cursor after key. */ 28214272Spst#define CURS_BEFORE 0x04 /* B: Unreturned cursor before key. */ 28314272Spst#define CURS_INIT 0x08 /* RB: Cursor initialized. */ 28414272Spst u_int8_t flags; 28514272Spst} CURSOR; 28614272Spst 28714272Spst/* 28814272Spst * The metadata of the tree. The nrecs field is used only by the RECNO code. 2891573Srgrimes * This is because the btree doesn't really need it and it requires that every 2901573Srgrimes * put or delete call modify the metadata. 2911573Srgrimes */ 2921573Srgrimestypedef struct _btmeta { 29314272Spst u_int32_t magic; /* magic number */ 29414272Spst u_int32_t version; /* version */ 29514272Spst u_int32_t psize; /* page size */ 29614272Spst u_int32_t free; /* page number of first free page */ 29714272Spst u_int32_t nrecs; /* R: number of records */ 29814272Spst 2991573Srgrimes#define SAVEMETA (B_NODUPS | R_RECNO) 30014272Spst u_int32_t flags; /* bt_flags & SAVEMETA */ 3011573Srgrimes} BTMETA; 3021573Srgrimes 3031573Srgrimes/* The in-memory btree/recno data structure. */ 3041573Srgrimestypedef struct _btree { 30514272Spst MPOOL *bt_mp; /* memory pool cookie */ 3061573Srgrimes 30714272Spst DB *bt_dbp; /* pointer to enclosing DB */ 3081573Srgrimes 30914272Spst EPG bt_cur; /* current (pinned) page */ 31014272Spst PAGE *bt_pinned; /* page pinned across calls */ 3111573Srgrimes 31214272Spst CURSOR bt_cursor; /* cursor */ 3131573Srgrimes 31414272Spst#define BT_PUSH(t, p, i) { \ 31514272Spst t->bt_sp->pgno = p; \ 31614272Spst t->bt_sp->index = i; \ 31714272Spst ++t->bt_sp; \ 31814272Spst} 31914272Spst#define BT_POP(t) (t->bt_sp == t->bt_stack ? NULL : --t->bt_sp) 32014272Spst#define BT_CLR(t) (t->bt_sp = t->bt_stack) 32114272Spst EPGNO bt_stack[50]; /* stack of parent pages */ 32214272Spst EPGNO *bt_sp; /* current stack pointer */ 3231573Srgrimes 32414272Spst DBT bt_rkey; /* returned key */ 32514272Spst DBT bt_rdata; /* returned data */ 3261573Srgrimes 32714272Spst int bt_fd; /* tree file descriptor */ 3281573Srgrimes 32914272Spst pgno_t bt_free; /* next free page */ 3301573Srgrimes u_int32_t bt_psize; /* page size */ 33114272Spst indx_t bt_ovflsize; /* cut-off for key/data overflow */ 33214272Spst int bt_lorder; /* byte order */ 3331573Srgrimes /* sorted order */ 3341573Srgrimes enum { NOT, BACK, FORWARD } bt_order; 33514272Spst EPGNO bt_last; /* last insert */ 3361573Srgrimes 3371573Srgrimes /* B: key comparison function */ 3381573Srgrimes int (*bt_cmp) __P((const DBT *, const DBT *)); 3391573Srgrimes /* B: prefix comparison function */ 3401573Srgrimes size_t (*bt_pfx) __P((const DBT *, const DBT *)); 3411573Srgrimes /* R: recno input function */ 3421573Srgrimes int (*bt_irec) __P((struct _btree *, recno_t)); 3431573Srgrimes 34414272Spst FILE *bt_rfp; /* R: record FILE pointer */ 34514272Spst int bt_rfd; /* R: record file descriptor */ 3461573Srgrimes 34714272Spst caddr_t bt_cmap; /* R: current point in mapped space */ 34814272Spst caddr_t bt_smap; /* R: start of mapped space */ 34914272Spst caddr_t bt_emap; /* R: end of mapped space */ 35014272Spst size_t bt_msize; /* R: size of mapped region. */ 3511573Srgrimes 35214272Spst recno_t bt_nrecs; /* R: number of records */ 35314272Spst size_t bt_reclen; /* R: fixed record length */ 35414272Spst u_char bt_bval; /* R: delimiting byte/pad character */ 3551573Srgrimes 3561573Srgrimes/* 3571573Srgrimes * NB: 3581573Srgrimes * B_NODUPS and R_RECNO are stored on disk, and may not be changed. 3591573Srgrimes */ 36014272Spst#define B_INMEM 0x00001 /* in-memory tree */ 36114272Spst#define B_METADIRTY 0x00002 /* need to write metadata */ 36214272Spst#define B_MODIFIED 0x00004 /* tree modified */ 36314272Spst#define B_NEEDSWAP 0x00008 /* if byte order requires swapping */ 36414272Spst#define B_RDONLY 0x00010 /* read-only tree */ 36514272Spst 3661573Srgrimes#define B_NODUPS 0x00020 /* no duplicate keys permitted */ 3671573Srgrimes#define R_RECNO 0x00080 /* record oriented tree */ 3681573Srgrimes 36914272Spst#define R_CLOSEFP 0x00040 /* opened a file pointer */ 37014272Spst#define R_EOF 0x00100 /* end of input file reached. */ 37114272Spst#define R_FIXLEN 0x00200 /* fixed length records */ 37214272Spst#define R_MEMMAPPED 0x00400 /* memory mapped file. */ 37314272Spst#define R_INMEM 0x00800 /* in-memory file */ 37414272Spst#define R_MODIFIED 0x01000 /* modified file */ 37514272Spst#define R_RDONLY 0x02000 /* read-only file */ 3761573Srgrimes 37714272Spst#define B_DB_LOCK 0x04000 /* DB_LOCK specified. */ 37814272Spst#define B_DB_SHMEM 0x08000 /* DB_SHMEM specified. */ 37914272Spst#define B_DB_TXN 0x10000 /* DB_TXN specified. */ 38014272Spst u_int32_t flags; 3811573Srgrimes} BTREE; 3821573Srgrimes 3831573Srgrimes#include "extern.h" 384