Home | History | Annotate | Line # | Download | only in btree
btree.h revision 1.12
      1 /*	$NetBSD: btree.h,v 1.12 2002/01/21 21:33:42 tv 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. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the University of
     21  *	California, Berkeley and its contributors.
     22  * 4. Neither the name of the University nor the names of its contributors
     23  *    may be used to endorse or promote products derived from this software
     24  *    without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36  * SUCH DAMAGE.
     37  *
     38  *	@(#)btree.h	8.11 (Berkeley) 8/17/94
     39  */
     40 
     41 #if HAVE_CONFIG_H
     42 #include "config.h"
     43 #endif
     44 
     45 /* Macros to set/clear/test flags. */
     46 #define	F_SET(p, f)	(p)->flags |= (f)
     47 #define	F_CLR(p, f)	(p)->flags &= ~(f)
     48 #define	F_ISSET(p, f)	((p)->flags & (f))
     49 
     50 #include <mpool.h>
     51 
     52 #define	DEFMINKEYPAGE	(2)		/* Minimum keys per page */
     53 #define	MINCACHE	(5)		/* Minimum cached pages */
     54 #define	MINPSIZE	(512)		/* Minimum page size */
     55 
     56 /*
     57  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
     58  * used as an out-of-band page, i.e. page pointers that point to nowhere point
     59  * to page 0.  Page 1 is the root of the btree.
     60  */
     61 #define	P_INVALID	 0		/* Invalid tree page number. */
     62 #define	P_META		 0		/* Tree metadata page number. */
     63 #define	P_ROOT		 1		/* Tree root page number. */
     64 
     65 /*
     66  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
     67  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
     68  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
     69  * This implementation requires that values within structures NOT be padded.
     70  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
     71  * to do some work to get this package to run.
     72  */
     73 typedef struct _page {
     74 	pgno_t	pgno;			/* this page's page number */
     75 	pgno_t	prevpg;			/* left sibling */
     76 	pgno_t	nextpg;			/* right sibling */
     77 
     78 #define	P_BINTERNAL	0x01		/* btree internal page */
     79 #define	P_BLEAF		0x02		/* leaf page */
     80 #define	P_OVERFLOW	0x04		/* overflow page */
     81 #define	P_RINTERNAL	0x08		/* recno internal page */
     82 #define	P_RLEAF		0x10		/* leaf page */
     83 #define P_TYPE		0x1f		/* type mask */
     84 #define	P_PRESERVE	0x20		/* never delete this chain of pages */
     85 	u_int32_t flags;
     86 
     87 	indx_t	lower;			/* lower bound of free space on page */
     88 	indx_t	upper;			/* upper bound of free space on page */
     89 	indx_t	linp[1];		/* indx_t-aligned VAR. LENGTH DATA */
     90 } PAGE;
     91 
     92 /* First and next index. */
     93 #define	BTDATAOFF							\
     94 	(sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) +		\
     95 	    sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t))
     96 #define	NEXTINDEX(p)	(((p)->lower - BTDATAOFF) / sizeof(indx_t))
     97 
     98 /*
     99  * For pages other than overflow pages, there is an array of offsets into the
    100  * rest of the page immediately following the page header.  Each offset is to
    101  * an item which is unique to the type of page.  The h_lower offset is just
    102  * past the last filled-in index.  The h_upper offset is the first item on the
    103  * page.  Offsets are from the beginning of the page.
    104  *
    105  * If an item is too big to store on a single page, a flag is set and the item
    106  * is a { page, size } pair such that the page is the first page of an overflow
    107  * chain with size bytes of item.  Overflow pages are simply bytes without any
    108  * external structure.
    109  *
    110  * The page number and size fields in the items are pgno_t-aligned so they can
    111  * be manipulated without copying.  (This presumes that 32 bit items can be
    112  * manipulated on this system.)
    113  */
    114 #define	BTLALIGN(n)	(((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1))
    115 #define	NOVFLSIZE	(sizeof(pgno_t) + sizeof(u_int32_t))
    116 
    117 /*
    118  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
    119  * pairs, such that the key compares less than or equal to all of the records
    120  * on that page.  For a tree without duplicate keys, an internal page with two
    121  * consecutive keys, a and b, will have all records greater than or equal to a
    122  * and less than b stored on the page associated with a.  Duplicate keys are
    123  * somewhat special and can cause duplicate internal and leaf page records and
    124  * some minor modifications of the above rule.
    125  */
    126 typedef struct _binternal {
    127 	u_int32_t ksize;		/* key size */
    128 	pgno_t	pgno;			/* page number stored on */
    129 #define	P_BIGDATA	0x01		/* overflow data */
    130 #define	P_BIGKEY	0x02		/* overflow key */
    131 	u_char	flags;
    132 	char	bytes[1];		/* data */
    133 } BINTERNAL;
    134 
    135 /* Get the page's BINTERNAL structure at index indx. */
    136 #define	GETBINTERNAL(pg, indx)						\
    137 	((BINTERNAL *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
    138 
    139 /* Get the number of bytes in the entry. */
    140 #define NBINTERNAL(len)							\
    141 	BTLALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len))
    142 
    143 /* Copy a BINTERNAL entry to the page. */
    144 #define	WR_BINTERNAL(p, size, pgno, flags) {				\
    145 	*(u_int32_t *)(void *)p = size;					\
    146 	p += sizeof(u_int32_t);						\
    147 	*(pgno_t *)(void *)p = pgno;					\
    148 	p += sizeof(pgno_t);						\
    149 	*(u_char *)(void *)p = flags;					\
    150 	p += sizeof(u_char);						\
    151 }
    152 
    153 /*
    154  * For the recno internal pages, the item is a page number with the number of
    155  * keys found on that page and below.
    156  */
    157 typedef struct _rinternal {
    158 	recno_t	nrecs;			/* number of records */
    159 	pgno_t	pgno;			/* page number stored below */
    160 } RINTERNAL;
    161 
    162 /* Get the page's RINTERNAL structure at index indx. */
    163 #define	GETRINTERNAL(pg, indx)						\
    164 	((RINTERNAL *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
    165 
    166 /* Get the number of bytes in the entry. */
    167 #define NRINTERNAL							\
    168 	BTLALIGN(sizeof(recno_t) + sizeof(pgno_t))
    169 
    170 /* Copy a RINTERAL entry to the page. */
    171 #define	WR_RINTERNAL(p, nrecs, pgno) {					\
    172 	*(recno_t *)(void *)p = nrecs;					\
    173 	p += sizeof(recno_t);						\
    174 	*(pgno_t *)(void *)p = pgno;					\
    175 }
    176 
    177 /* For the btree leaf pages, the item is a key and data pair. */
    178 typedef struct _bleaf {
    179 	u_int32_t	ksize;		/* size of key */
    180 	u_int32_t	dsize;		/* size of data */
    181 	u_char	flags;			/* P_BIGDATA, P_BIGKEY */
    182 	char	bytes[1];		/* data */
    183 } BLEAF;
    184 
    185 /* Get the page's BLEAF structure at index indx. */
    186 #define	GETBLEAF(pg, indx)						\
    187 	((BLEAF *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
    188 
    189 /* Get the number of bytes in the entry. */
    190 #define NBLEAF(p)	NBLEAFDBT((p)->ksize, (p)->dsize)
    191 
    192 /* Get the number of bytes in the user's key/data pair. */
    193 #define NBLEAFDBT(ksize, dsize)						\
    194 	BTLALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) +	\
    195 	    (ksize) + (dsize))
    196 
    197 /* Copy a BLEAF entry to the page. */
    198 #define	WR_BLEAF(p, key, data, flags) {					\
    199 	*(u_int32_t *)(void *)p = key->size;				\
    200 	p += sizeof(u_int32_t);						\
    201 	*(u_int32_t *)(void *)p = data->size;				\
    202 	p += sizeof(u_int32_t);						\
    203 	*(u_char *)(void *)p = flags;					\
    204 	p += sizeof(u_char);						\
    205 	memmove(p, key->data, key->size);				\
    206 	p += key->size;							\
    207 	memmove(p, data->data, data->size);				\
    208 }
    209 
    210 /* For the recno leaf pages, the item is a data entry. */
    211 typedef struct _rleaf {
    212 	u_int32_t	dsize;		/* size of data */
    213 	u_char	flags;			/* P_BIGDATA */
    214 	char	bytes[1];
    215 } RLEAF;
    216 
    217 /* Get the page's RLEAF structure at index indx. */
    218 #define	GETRLEAF(pg, indx)						\
    219 	((RLEAF *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
    220 
    221 /* Get the number of bytes in the entry. */
    222 #define NRLEAF(p)	NRLEAFDBT((p)->dsize)
    223 
    224 /* Get the number of bytes from the user's data. */
    225 #define	NRLEAFDBT(dsize)						\
    226 	BTLALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize))
    227 
    228 /* Copy a RLEAF entry to the page. */
    229 #define	WR_RLEAF(p, data, flags) {					\
    230 	*(u_int32_t *)(void *)p = data->size;				\
    231 	p += sizeof(u_int32_t);						\
    232 	*(u_char *)(void *)p = flags;					\
    233 	p += sizeof(u_char);						\
    234 	memmove(p, data->data, data->size);				\
    235 }
    236 
    237 /*
    238  * A record in the tree is either a pointer to a page and an index in the page
    239  * or a page number and an index.  These structures are used as a cursor, stack
    240  * entry and search returns as well as to pass records to other routines.
    241  *
    242  * One comment about searches.  Internal page searches must find the largest
    243  * record less than key in the tree so that descents work.  Leaf page searches
    244  * must find the smallest record greater than key so that the returned index
    245  * is the record's correct position for insertion.
    246  */
    247 typedef struct _epgno {
    248 	pgno_t	pgno;			/* the page number */
    249 	indx_t	index;			/* the index on the page */
    250 } EPGNO;
    251 
    252 typedef struct _epg {
    253 	PAGE	*page;			/* the (pinned) page */
    254 	indx_t	 index;			/* the index on the page */
    255 } EPG;
    256 
    257 /*
    258  * About cursors.  The cursor (and the page that contained the key/data pair
    259  * that it referenced) can be deleted, which makes things a bit tricky.  If
    260  * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set
    261  * or there simply aren't any duplicates of the key) we copy the key that it
    262  * referenced when it's deleted, and reacquire a new cursor key if the cursor
    263  * is used again.  If there are duplicates keys, we move to the next/previous
    264  * key, and set a flag so that we know what happened.  NOTE: if duplicate (to
    265  * the cursor) keys are added to the tree during this process, it is undefined
    266  * if they will be returned or not in a cursor scan.
    267  *
    268  * The flags determine the possible states of the cursor:
    269  *
    270  * CURS_INIT	The cursor references *something*.
    271  * CURS_ACQUIRE	The cursor was deleted, and a key has been saved so that
    272  *		we can reacquire the right position in the tree.
    273  * CURS_AFTER, CURS_BEFORE
    274  *		The cursor was deleted, and now references a key/data pair
    275  *		that has not yet been returned, either before or after the
    276  *		deleted key/data pair.
    277  * XXX
    278  * This structure is broken out so that we can eventually offer multiple
    279  * cursors as part of the DB interface.
    280  */
    281 typedef struct _cursor {
    282 	EPGNO	 pg;			/* B: Saved tree reference. */
    283 	DBT	 key;			/* B: Saved key, or key.data == NULL. */
    284 	recno_t	 rcursor;		/* R: recno cursor (1-based) */
    285 
    286 #define	CURS_ACQUIRE	0x01		/*  B: Cursor needs to be reacquired. */
    287 #define	CURS_AFTER	0x02		/*  B: Unreturned cursor after key. */
    288 #define	CURS_BEFORE	0x04		/*  B: Unreturned cursor before key. */
    289 #define	CURS_INIT	0x08		/* RB: Cursor initialized. */
    290 	u_int8_t flags;
    291 } CURSOR;
    292 
    293 /*
    294  * The metadata of the tree.  The nrecs field is used only by the RECNO code.
    295  * This is because the btree doesn't really need it and it requires that every
    296  * put or delete call modify the metadata.
    297  */
    298 typedef struct _btmeta {
    299 	u_int32_t	magic;		/* magic number */
    300 	u_int32_t	version;	/* version */
    301 	u_int32_t	psize;		/* page size */
    302 	u_int32_t	free;		/* page number of first free page */
    303 	u_int32_t	nrecs;		/* R: number of records */
    304 
    305 #define	SAVEMETA	(B_NODUPS | R_RECNO)
    306 	u_int32_t	flags;		/* bt_flags & SAVEMETA */
    307 } BTMETA;
    308 
    309 /* The in-memory btree/recno data structure. */
    310 typedef struct _btree {
    311 	MPOOL	 *bt_mp;		/* memory pool cookie */
    312 
    313 	DB	 *bt_dbp;		/* pointer to enclosing DB */
    314 
    315 	EPG	  bt_cur;		/* current (pinned) page */
    316 	PAGE	 *bt_pinned;		/* page pinned across calls */
    317 
    318 	CURSOR	  bt_cursor;		/* cursor */
    319 
    320 #define	BT_PUSH(t, p, i) {						\
    321 	t->bt_sp->pgno = p; 						\
    322 	t->bt_sp->index = i; 						\
    323 	++t->bt_sp;							\
    324 }
    325 #define	BT_POP(t)	(t->bt_sp == t->bt_stack ? NULL : --t->bt_sp)
    326 #define	BT_CLR(t)	(t->bt_sp = t->bt_stack)
    327 	EPGNO	  bt_stack[50];		/* stack of parent pages */
    328 	EPGNO	 *bt_sp;		/* current stack pointer */
    329 
    330 	DBT	  bt_rkey;		/* returned key */
    331 	DBT	  bt_rdata;		/* returned data */
    332 
    333 	int	  bt_fd;		/* tree file descriptor */
    334 
    335 	pgno_t	  bt_free;		/* next free page */
    336 	u_int32_t bt_psize;		/* page size */
    337 	indx_t	  bt_ovflsize;		/* cut-off for key/data overflow */
    338 	int	  bt_lorder;		/* byte order */
    339 					/* sorted order */
    340 	enum { NOT, BACK, FORWARD } bt_order;
    341 	EPGNO	  bt_last;		/* last insert */
    342 
    343 					/* B: key comparison function */
    344 	int	(*bt_cmp) __P((const DBT *, const DBT *));
    345 					/* B: prefix comparison function */
    346 	size_t	(*bt_pfx) __P((const DBT *, const DBT *));
    347 					/* R: recno input function */
    348 	int	(*bt_irec) __P((struct _btree *, recno_t));
    349 
    350 	FILE	 *bt_rfp;		/* R: record FILE pointer */
    351 	int	  bt_rfd;		/* R: record file descriptor */
    352 
    353 	caddr_t	  bt_cmap;		/* R: current point in mapped space */
    354 	caddr_t	  bt_smap;		/* R: start of mapped space */
    355 	caddr_t   bt_emap;		/* R: end of mapped space */
    356 	size_t	  bt_msize;		/* R: size of mapped region. */
    357 
    358 	recno_t	  bt_nrecs;		/* R: number of records */
    359 	size_t	  bt_reclen;		/* R: fixed record length */
    360 	u_char	  bt_bval;		/* R: delimiting byte/pad character */
    361 
    362 /*
    363  * NB:
    364  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
    365  */
    366 #define	B_INMEM		0x00001		/* in-memory tree */
    367 #define	B_METADIRTY	0x00002		/* need to write metadata */
    368 #define	B_MODIFIED	0x00004		/* tree modified */
    369 #define	B_NEEDSWAP	0x00008		/* if byte order requires swapping */
    370 #define	B_RDONLY	0x00010		/* read-only tree */
    371 
    372 #define	B_NODUPS	0x00020		/* no duplicate keys permitted */
    373 #define	R_RECNO		0x00080		/* record oriented tree */
    374 
    375 #define	R_CLOSEFP	0x00040		/* opened a file pointer */
    376 #define	R_EOF		0x00100		/* end of input file reached. */
    377 #define	R_FIXLEN	0x00200		/* fixed length records */
    378 #define	R_MEMMAPPED	0x00400		/* memory mapped file. */
    379 #define	R_INMEM		0x00800		/* in-memory file */
    380 #define	R_MODIFIED	0x01000		/* modified file */
    381 #define	R_RDONLY	0x02000		/* read-only file */
    382 
    383 #define	B_DB_LOCK	0x04000		/* DB_LOCK specified. */
    384 #define	B_DB_SHMEM	0x08000		/* DB_SHMEM specified. */
    385 #define	B_DB_TXN	0x10000		/* DB_TXN specified. */
    386 	u_int32_t flags;
    387 } BTREE;
    388 
    389 #include "extern.h"
    390