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 * 4. Neither the name of the University nor the names of its contributors 171573Srgrimes * may be used to endorse or promote products derived from this software 181573Srgrimes * without specific prior written permission. 191573Srgrimes * 201573Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 211573Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 221573Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 231573Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 241573Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 251573Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 261573Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 271573Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 281573Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 291573Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 301573Srgrimes * SUCH DAMAGE. 311573Srgrimes * 3214272Spst * @(#)btree.h 8.11 (Berkeley) 8/17/94 3392991Sobrien * $FreeBSD$ 341573Srgrimes */ 351573Srgrimes 3614272Spst/* Macros to set/clear/test flags. */ 3714272Spst#define F_SET(p, f) (p)->flags |= (f) 3814272Spst#define F_CLR(p, f) (p)->flags &= ~(f) 3914272Spst#define F_ISSET(p, f) ((p)->flags & (f)) 4014272Spst 411573Srgrimes#include <mpool.h> 421573Srgrimes 431573Srgrimes#define DEFMINKEYPAGE (2) /* Minimum keys per page */ 441573Srgrimes#define MINCACHE (5) /* Minimum cached pages */ 451573Srgrimes#define MINPSIZE (512) /* Minimum page size */ 461573Srgrimes 471573Srgrimes/* 481573Srgrimes * Page 0 of a btree file contains a copy of the meta-data. This page is also 491573Srgrimes * used as an out-of-band page, i.e. page pointers that point to nowhere point 501573Srgrimes * to page 0. Page 1 is the root of the btree. 511573Srgrimes */ 521573Srgrimes#define P_INVALID 0 /* Invalid tree page number. */ 531573Srgrimes#define P_META 0 /* Tree metadata page number. */ 541573Srgrimes#define P_ROOT 1 /* Tree root page number. */ 551573Srgrimes 561573Srgrimes/* 571573Srgrimes * There are five page layouts in the btree: btree internal pages (BINTERNAL), 581573Srgrimes * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages 591573Srgrimes * (RLEAF) and overflow pages. All five page types have a page header (PAGE). 601573Srgrimes * This implementation requires that values within structures NOT be padded. 611573Srgrimes * (ANSI C permits random padding.) If your compiler pads randomly you'll have 621573Srgrimes * to do some work to get this package to run. 631573Srgrimes */ 641573Srgrimestypedef struct _page { 651573Srgrimes pgno_t pgno; /* this page's page number */ 661573Srgrimes pgno_t prevpg; /* left sibling */ 671573Srgrimes pgno_t nextpg; /* right sibling */ 681573Srgrimes 691573Srgrimes#define P_BINTERNAL 0x01 /* btree internal page */ 701573Srgrimes#define P_BLEAF 0x02 /* leaf page */ 711573Srgrimes#define P_OVERFLOW 0x04 /* overflow page */ 721573Srgrimes#define P_RINTERNAL 0x08 /* recno internal page */ 731573Srgrimes#define P_RLEAF 0x10 /* leaf page */ 741573Srgrimes#define P_TYPE 0x1f /* type mask */ 751573Srgrimes#define P_PRESERVE 0x20 /* never delete this chain of pages */ 761573Srgrimes u_int32_t flags; 771573Srgrimes 781573Srgrimes indx_t lower; /* lower bound of free space on page */ 791573Srgrimes indx_t upper; /* upper bound of free space on page */ 801573Srgrimes indx_t linp[1]; /* indx_t-aligned VAR. LENGTH DATA */ 811573Srgrimes} PAGE; 821573Srgrimes 831573Srgrimes/* First and next index. */ 8414272Spst#define BTDATAOFF \ 8514272Spst (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \ 8614272Spst sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t)) 871573Srgrimes#define NEXTINDEX(p) (((p)->lower - BTDATAOFF) / sizeof(indx_t)) 881573Srgrimes 891573Srgrimes/* 901573Srgrimes * For pages other than overflow pages, there is an array of offsets into the 911573Srgrimes * rest of the page immediately following the page header. Each offset is to 921573Srgrimes * an item which is unique to the type of page. The h_lower offset is just 931573Srgrimes * past the last filled-in index. The h_upper offset is the first item on the 941573Srgrimes * page. Offsets are from the beginning of the page. 951573Srgrimes * 961573Srgrimes * If an item is too big to store on a single page, a flag is set and the item 971573Srgrimes * is a { page, size } pair such that the page is the first page of an overflow 981573Srgrimes * chain with size bytes of item. Overflow pages are simply bytes without any 991573Srgrimes * external structure. 1001573Srgrimes * 1011573Srgrimes * The page number and size fields in the items are pgno_t-aligned so they can 1021573Srgrimes * be manipulated without copying. (This presumes that 32 bit items can be 1031573Srgrimes * manipulated on this system.) 1041573Srgrimes */ 10514272Spst#define LALIGN(n) (((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1)) 10614272Spst#define NOVFLSIZE (sizeof(pgno_t) + sizeof(u_int32_t)) 1071573Srgrimes 1081573Srgrimes/* 1091573Srgrimes * For the btree internal pages, the item is a key. BINTERNALs are {key, pgno} 1101573Srgrimes * pairs, such that the key compares less than or equal to all of the records 1111573Srgrimes * on that page. For a tree without duplicate keys, an internal page with two 1121573Srgrimes * consecutive keys, a and b, will have all records greater than or equal to a 1131573Srgrimes * and less than b stored on the page associated with a. Duplicate keys are 1141573Srgrimes * somewhat special and can cause duplicate internal and leaf page records and 1151573Srgrimes * some minor modifications of the above rule. 1161573Srgrimes */ 1171573Srgrimestypedef struct _binternal { 11814272Spst u_int32_t ksize; /* key size */ 1191573Srgrimes pgno_t pgno; /* page number stored on */ 1201573Srgrimes#define P_BIGDATA 0x01 /* overflow data */ 1211573Srgrimes#define P_BIGKEY 0x02 /* overflow key */ 1221573Srgrimes u_char flags; 1231573Srgrimes char bytes[1]; /* data */ 1241573Srgrimes} BINTERNAL; 1251573Srgrimes 1261573Srgrimes/* Get the page's BINTERNAL structure at index indx. */ 12714272Spst#define GETBINTERNAL(pg, indx) \ 1281573Srgrimes ((BINTERNAL *)((char *)(pg) + (pg)->linp[indx])) 1291573Srgrimes 1301573Srgrimes/* Get the number of bytes in the entry. */ 13114272Spst#define NBINTERNAL(len) \ 13214272Spst LALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len)) 1331573Srgrimes 1341573Srgrimes/* Copy a BINTERNAL entry to the page. */ 13514272Spst#define WR_BINTERNAL(p, size, pgno, flags) { \ 13614272Spst *(u_int32_t *)p = size; \ 13714272Spst p += sizeof(u_int32_t); \ 13814272Spst *(pgno_t *)p = pgno; \ 13914272Spst p += sizeof(pgno_t); \ 14014272Spst *(u_char *)p = flags; \ 14114272Spst p += sizeof(u_char); \ 1421573Srgrimes} 1431573Srgrimes 1441573Srgrimes/* 1451573Srgrimes * For the recno internal pages, the item is a page number with the number of 1461573Srgrimes * keys found on that page and below. 1471573Srgrimes */ 1481573Srgrimestypedef struct _rinternal { 1491573Srgrimes recno_t nrecs; /* number of records */ 1501573Srgrimes pgno_t pgno; /* page number stored below */ 1511573Srgrimes} RINTERNAL; 1521573Srgrimes 1531573Srgrimes/* Get the page's RINTERNAL structure at index indx. */ 15414272Spst#define GETRINTERNAL(pg, indx) \ 1551573Srgrimes ((RINTERNAL *)((char *)(pg) + (pg)->linp[indx])) 1561573Srgrimes 1571573Srgrimes/* Get the number of bytes in the entry. */ 15814272Spst#define NRINTERNAL \ 1591573Srgrimes LALIGN(sizeof(recno_t) + sizeof(pgno_t)) 1601573Srgrimes 1611573Srgrimes/* Copy a RINTERAL entry to the page. */ 16214272Spst#define WR_RINTERNAL(p, nrecs, pgno) { \ 16314272Spst *(recno_t *)p = nrecs; \ 16414272Spst p += sizeof(recno_t); \ 16514272Spst *(pgno_t *)p = pgno; \ 1661573Srgrimes} 1671573Srgrimes 1681573Srgrimes/* For the btree leaf pages, the item is a key and data pair. */ 1691573Srgrimestypedef struct _bleaf { 17014272Spst u_int32_t ksize; /* size of key */ 17114272Spst u_int32_t dsize; /* size of data */ 1721573Srgrimes u_char flags; /* P_BIGDATA, P_BIGKEY */ 1731573Srgrimes char bytes[1]; /* data */ 1741573Srgrimes} BLEAF; 1751573Srgrimes 1761573Srgrimes/* Get the page's BLEAF structure at index indx. */ 17714272Spst#define GETBLEAF(pg, indx) \ 1781573Srgrimes ((BLEAF *)((char *)(pg) + (pg)->linp[indx])) 1791573Srgrimes 1801573Srgrimes/* Get the number of bytes in the entry. */ 1811573Srgrimes#define NBLEAF(p) NBLEAFDBT((p)->ksize, (p)->dsize) 1821573Srgrimes 1831573Srgrimes/* Get the number of bytes in the user's key/data pair. */ 18414272Spst#define NBLEAFDBT(ksize, dsize) \ 18514272Spst LALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) + \ 1861573Srgrimes (ksize) + (dsize)) 1871573Srgrimes 1881573Srgrimes/* Copy a BLEAF entry to the page. */ 18914272Spst#define WR_BLEAF(p, key, data, flags) { \ 19014272Spst *(u_int32_t *)p = key->size; \ 19114272Spst p += sizeof(u_int32_t); \ 19214272Spst *(u_int32_t *)p = data->size; \ 19314272Spst p += sizeof(u_int32_t); \ 19414272Spst *(u_char *)p = flags; \ 19514272Spst p += sizeof(u_char); \ 19614272Spst memmove(p, key->data, key->size); \ 19714272Spst p += key->size; \ 19814272Spst memmove(p, data->data, data->size); \ 1991573Srgrimes} 2001573Srgrimes 2011573Srgrimes/* For the recno leaf pages, the item is a data entry. */ 2021573Srgrimestypedef struct _rleaf { 20314272Spst u_int32_t dsize; /* size of data */ 2041573Srgrimes u_char flags; /* P_BIGDATA */ 2051573Srgrimes char bytes[1]; 2061573Srgrimes} RLEAF; 2071573Srgrimes 2081573Srgrimes/* Get the page's RLEAF structure at index indx. */ 20914272Spst#define GETRLEAF(pg, indx) \ 2101573Srgrimes ((RLEAF *)((char *)(pg) + (pg)->linp[indx])) 2111573Srgrimes 2121573Srgrimes/* Get the number of bytes in the entry. */ 2131573Srgrimes#define NRLEAF(p) NRLEAFDBT((p)->dsize) 2141573Srgrimes 2151573Srgrimes/* Get the number of bytes from the user's data. */ 21614272Spst#define NRLEAFDBT(dsize) \ 21714272Spst LALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize)) 2181573Srgrimes 2191573Srgrimes/* Copy a RLEAF entry to the page. */ 22014272Spst#define WR_RLEAF(p, data, flags) { \ 22114272Spst *(u_int32_t *)p = data->size; \ 22214272Spst p += sizeof(u_int32_t); \ 22314272Spst *(u_char *)p = flags; \ 22414272Spst p += sizeof(u_char); \ 22514272Spst memmove(p, data->data, data->size); \ 2261573Srgrimes} 2271573Srgrimes 2281573Srgrimes/* 2291573Srgrimes * A record in the tree is either a pointer to a page and an index in the page 2301573Srgrimes * or a page number and an index. These structures are used as a cursor, stack 2311573Srgrimes * entry and search returns as well as to pass records to other routines. 2321573Srgrimes * 2331573Srgrimes * One comment about searches. Internal page searches must find the largest 2341573Srgrimes * record less than key in the tree so that descents work. Leaf page searches 2351573Srgrimes * must find the smallest record greater than key so that the returned index 2361573Srgrimes * is the record's correct position for insertion. 2371573Srgrimes */ 2381573Srgrimestypedef struct _epgno { 2391573Srgrimes pgno_t pgno; /* the page number */ 2401573Srgrimes indx_t index; /* the index on the page */ 2411573Srgrimes} EPGNO; 2421573Srgrimes 2431573Srgrimestypedef struct _epg { 2441573Srgrimes PAGE *page; /* the (pinned) page */ 2451573Srgrimes indx_t index; /* the index on the page */ 2461573Srgrimes} EPG; 2471573Srgrimes 2481573Srgrimes/* 24914272Spst * About cursors. The cursor (and the page that contained the key/data pair 25014272Spst * that it referenced) can be deleted, which makes things a bit tricky. If 25114272Spst * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set 25214272Spst * or there simply aren't any duplicates of the key) we copy the key that it 25314272Spst * referenced when it's deleted, and reacquire a new cursor key if the cursor 25414272Spst * is used again. If there are duplicates keys, we move to the next/previous 25514272Spst * key, and set a flag so that we know what happened. NOTE: if duplicate (to 25614272Spst * the cursor) keys are added to the tree during this process, it is undefined 25714272Spst * if they will be returned or not in a cursor scan. 25814272Spst * 25914272Spst * The flags determine the possible states of the cursor: 26014272Spst * 26114272Spst * CURS_INIT The cursor references *something*. 26214272Spst * CURS_ACQUIRE The cursor was deleted, and a key has been saved so that 26314272Spst * we can reacquire the right position in the tree. 26414272Spst * CURS_AFTER, CURS_BEFORE 26514272Spst * The cursor was deleted, and now references a key/data pair 26614272Spst * that has not yet been returned, either before or after the 26714272Spst * deleted key/data pair. 26814272Spst * XXX 26914272Spst * This structure is broken out so that we can eventually offer multiple 27014272Spst * cursors as part of the DB interface. 27114272Spst */ 27214272Spsttypedef struct _cursor { 27314272Spst EPGNO pg; /* B: Saved tree reference. */ 27414272Spst DBT key; /* B: Saved key, or key.data == NULL. */ 27514272Spst recno_t rcursor; /* R: recno cursor (1-based) */ 27614272Spst 27714272Spst#define CURS_ACQUIRE 0x01 /* B: Cursor needs to be reacquired. */ 27814272Spst#define CURS_AFTER 0x02 /* B: Unreturned cursor after key. */ 27914272Spst#define CURS_BEFORE 0x04 /* B: Unreturned cursor before key. */ 28014272Spst#define CURS_INIT 0x08 /* RB: Cursor initialized. */ 28114272Spst u_int8_t flags; 28214272Spst} CURSOR; 28314272Spst 28414272Spst/* 28514272Spst * The metadata of the tree. The nrecs field is used only by the RECNO code. 2861573Srgrimes * This is because the btree doesn't really need it and it requires that every 2871573Srgrimes * put or delete call modify the metadata. 2881573Srgrimes */ 2891573Srgrimestypedef struct _btmeta { 29014272Spst u_int32_t magic; /* magic number */ 29114272Spst u_int32_t version; /* version */ 29214272Spst u_int32_t psize; /* page size */ 29314272Spst u_int32_t free; /* page number of first free page */ 29414272Spst u_int32_t nrecs; /* R: number of records */ 29514272Spst 2961573Srgrimes#define SAVEMETA (B_NODUPS | R_RECNO) 29714272Spst u_int32_t flags; /* bt_flags & SAVEMETA */ 2981573Srgrimes} BTMETA; 2991573Srgrimes 3001573Srgrimes/* The in-memory btree/recno data structure. */ 3011573Srgrimestypedef struct _btree { 30214272Spst MPOOL *bt_mp; /* memory pool cookie */ 3031573Srgrimes 30414272Spst DB *bt_dbp; /* pointer to enclosing DB */ 3051573Srgrimes 30614272Spst EPG bt_cur; /* current (pinned) page */ 30714272Spst PAGE *bt_pinned; /* page pinned across calls */ 3081573Srgrimes 30914272Spst CURSOR bt_cursor; /* cursor */ 3101573Srgrimes 31114272Spst#define BT_PUSH(t, p, i) { \ 312189327Sdelphij t->bt_sp->pgno = p; \ 313189327Sdelphij t->bt_sp->index = i; \ 31414272Spst ++t->bt_sp; \ 31514272Spst} 31614272Spst#define BT_POP(t) (t->bt_sp == t->bt_stack ? NULL : --t->bt_sp) 31714272Spst#define BT_CLR(t) (t->bt_sp = t->bt_stack) 31814272Spst EPGNO bt_stack[50]; /* stack of parent pages */ 31914272Spst EPGNO *bt_sp; /* current stack pointer */ 3201573Srgrimes 32114272Spst DBT bt_rkey; /* returned key */ 32214272Spst DBT bt_rdata; /* returned data */ 3231573Srgrimes 32414272Spst int bt_fd; /* tree file descriptor */ 3251573Srgrimes 32614272Spst pgno_t bt_free; /* next free page */ 3271573Srgrimes u_int32_t bt_psize; /* page size */ 32814272Spst indx_t bt_ovflsize; /* cut-off for key/data overflow */ 32914272Spst int bt_lorder; /* byte order */ 3301573Srgrimes /* sorted order */ 3311573Srgrimes enum { NOT, BACK, FORWARD } bt_order; 33214272Spst EPGNO bt_last; /* last insert */ 3331573Srgrimes 3341573Srgrimes /* B: key comparison function */ 33592905Sobrien int (*bt_cmp)(const DBT *, const DBT *); 3361573Srgrimes /* B: prefix comparison function */ 33792905Sobrien size_t (*bt_pfx)(const DBT *, const DBT *); 3381573Srgrimes /* R: recno input function */ 33992905Sobrien int (*bt_irec)(struct _btree *, recno_t); 3401573Srgrimes 34114272Spst FILE *bt_rfp; /* R: record FILE pointer */ 34214272Spst int bt_rfd; /* R: record file descriptor */ 3431573Srgrimes 34414272Spst caddr_t bt_cmap; /* R: current point in mapped space */ 34514272Spst caddr_t bt_smap; /* R: start of mapped space */ 34614272Spst caddr_t bt_emap; /* R: end of mapped space */ 34714272Spst size_t bt_msize; /* R: size of mapped region. */ 3481573Srgrimes 34914272Spst recno_t bt_nrecs; /* R: number of records */ 35014272Spst size_t bt_reclen; /* R: fixed record length */ 35114272Spst u_char bt_bval; /* R: delimiting byte/pad character */ 3521573Srgrimes 3531573Srgrimes/* 3541573Srgrimes * NB: 3551573Srgrimes * B_NODUPS and R_RECNO are stored on disk, and may not be changed. 3561573Srgrimes */ 35714272Spst#define B_INMEM 0x00001 /* in-memory tree */ 35814272Spst#define B_METADIRTY 0x00002 /* need to write metadata */ 35914272Spst#define B_MODIFIED 0x00004 /* tree modified */ 36014272Spst#define B_NEEDSWAP 0x00008 /* if byte order requires swapping */ 36114272Spst#define B_RDONLY 0x00010 /* read-only tree */ 36214272Spst 3631573Srgrimes#define B_NODUPS 0x00020 /* no duplicate keys permitted */ 3641573Srgrimes#define R_RECNO 0x00080 /* record oriented tree */ 3651573Srgrimes 36614272Spst#define R_CLOSEFP 0x00040 /* opened a file pointer */ 36714272Spst#define R_EOF 0x00100 /* end of input file reached. */ 36814272Spst#define R_FIXLEN 0x00200 /* fixed length records */ 36914272Spst#define R_MEMMAPPED 0x00400 /* memory mapped file. */ 37014272Spst#define R_INMEM 0x00800 /* in-memory file */ 37114272Spst#define R_MODIFIED 0x01000 /* modified file */ 37214272Spst#define R_RDONLY 0x02000 /* read-only file */ 3731573Srgrimes 37414272Spst#define B_DB_LOCK 0x04000 /* DB_LOCK specified. */ 37514272Spst#define B_DB_SHMEM 0x08000 /* DB_SHMEM specified. */ 37614272Spst#define B_DB_TXN 0x10000 /* DB_TXN specified. */ 37714272Spst u_int32_t flags; 3781573Srgrimes} BTREE; 3791573Srgrimes 3801573Srgrimes#include "extern.h" 381