xref: /freebsd/lib/libc/db/btree/btree.h (revision afe61c15161c324a7af299a9b8457aba5afc92db)
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  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)btree.h	8.5 (Berkeley) 2/21/94
37  */
38 
39 #include <mpool.h>
40 
41 #define	DEFMINKEYPAGE	(2)		/* Minimum keys per page */
42 #define	MINCACHE	(5)		/* Minimum cached pages */
43 #define	MINPSIZE	(512)		/* Minimum page size */
44 
45 /*
46  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
47  * used as an out-of-band page, i.e. page pointers that point to nowhere point
48  * to page 0.  Page 1 is the root of the btree.
49  */
50 #define	P_INVALID	 0		/* Invalid tree page number. */
51 #define	P_META		 0		/* Tree metadata page number. */
52 #define	P_ROOT		 1		/* Tree root page number. */
53 
54 /*
55  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
56  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
57  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
58  * This implementation requires that values within structures NOT be padded.
59  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
60  * to do some work to get this package to run.
61  */
62 typedef struct _page {
63 	pgno_t	pgno;			/* this page's page number */
64 	pgno_t	prevpg;			/* left sibling */
65 	pgno_t	nextpg;			/* right sibling */
66 
67 #define	P_BINTERNAL	0x01		/* btree internal page */
68 #define	P_BLEAF		0x02		/* leaf page */
69 #define	P_OVERFLOW	0x04		/* overflow page */
70 #define	P_RINTERNAL	0x08		/* recno internal page */
71 #define	P_RLEAF		0x10		/* leaf page */
72 #define P_TYPE		0x1f		/* type mask */
73 #define	P_PRESERVE	0x20		/* never delete this chain of pages */
74 	u_int32_t flags;
75 
76 	indx_t	lower;			/* lower bound of free space on page */
77 	indx_t	upper;			/* upper bound of free space on page */
78 	indx_t	linp[1];		/* indx_t-aligned VAR. LENGTH DATA */
79 } PAGE;
80 
81 /* First and next index. */
82 #define	BTDATAOFF	(sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \
83 			    sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t))
84 #define	NEXTINDEX(p)	(((p)->lower - BTDATAOFF) / sizeof(indx_t))
85 
86 /*
87  * For pages other than overflow pages, there is an array of offsets into the
88  * rest of the page immediately following the page header.  Each offset is to
89  * an item which is unique to the type of page.  The h_lower offset is just
90  * past the last filled-in index.  The h_upper offset is the first item on the
91  * page.  Offsets are from the beginning of the page.
92  *
93  * If an item is too big to store on a single page, a flag is set and the item
94  * is a { page, size } pair such that the page is the first page of an overflow
95  * chain with size bytes of item.  Overflow pages are simply bytes without any
96  * external structure.
97  *
98  * The page number and size fields in the items are pgno_t-aligned so they can
99  * be manipulated without copying.  (This presumes that 32 bit items can be
100  * manipulated on this system.)
101  */
102 #define	LALIGN(n) \
103 	(((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1))
104 #define	NOVFLSIZE	(sizeof(pgno_t) + sizeof(size_t))
105 
106 /*
107  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
108  * pairs, such that the key compares less than or equal to all of the records
109  * on that page.  For a tree without duplicate keys, an internal page with two
110  * consecutive keys, a and b, will have all records greater than or equal to a
111  * and less than b stored on the page associated with a.  Duplicate keys are
112  * somewhat special and can cause duplicate internal and leaf page records and
113  * some minor modifications of the above rule.
114  */
115 typedef struct _binternal {
116 	size_t	ksize;			/* key size */
117 	pgno_t	pgno;			/* page number stored on */
118 #define	P_BIGDATA	0x01		/* overflow data */
119 #define	P_BIGKEY	0x02		/* overflow key */
120 	u_char	flags;
121 	char	bytes[1];		/* data */
122 } BINTERNAL;
123 
124 /* Get the page's BINTERNAL structure at index indx. */
125 #define	GETBINTERNAL(pg, indx) \
126 	((BINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
127 
128 /* Get the number of bytes in the entry. */
129 #define NBINTERNAL(len) \
130 	LALIGN(sizeof(size_t) + sizeof(pgno_t) + sizeof(u_char) + (len))
131 
132 /* Copy a BINTERNAL entry to the page. */
133 #define	WR_BINTERNAL(p, size, pgno, flags) { \
134 	*(size_t *)p = size; \
135 	p += sizeof(size_t); \
136 	*(pgno_t *)p = pgno; \
137 	p += sizeof(pgno_t); \
138 	*(u_char *)p = flags; \
139 	p += sizeof(u_char); \
140 }
141 
142 /*
143  * For the recno internal pages, the item is a page number with the number of
144  * keys found on that page and below.
145  */
146 typedef struct _rinternal {
147 	recno_t	nrecs;			/* number of records */
148 	pgno_t	pgno;			/* page number stored below */
149 } RINTERNAL;
150 
151 /* Get the page's RINTERNAL structure at index indx. */
152 #define	GETRINTERNAL(pg, indx) \
153 	((RINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
154 
155 /* Get the number of bytes in the entry. */
156 #define NRINTERNAL \
157 	LALIGN(sizeof(recno_t) + sizeof(pgno_t))
158 
159 /* Copy a RINTERAL entry to the page. */
160 #define	WR_RINTERNAL(p, nrecs, pgno) { \
161 	*(recno_t *)p = nrecs; \
162 	p += sizeof(recno_t); \
163 	*(pgno_t *)p = pgno; \
164 }
165 
166 /* For the btree leaf pages, the item is a key and data pair. */
167 typedef struct _bleaf {
168 	size_t	ksize;			/* size of key */
169 	size_t	dsize;			/* size of data */
170 	u_char	flags;			/* P_BIGDATA, P_BIGKEY */
171 	char	bytes[1];		/* data */
172 } BLEAF;
173 
174 /* Get the page's BLEAF structure at index indx. */
175 #define	GETBLEAF(pg, indx) \
176 	((BLEAF *)((char *)(pg) + (pg)->linp[indx]))
177 
178 /* Get the number of bytes in the entry. */
179 #define NBLEAF(p)	NBLEAFDBT((p)->ksize, (p)->dsize)
180 
181 /* Get the number of bytes in the user's key/data pair. */
182 #define NBLEAFDBT(ksize, dsize) \
183 	LALIGN(sizeof(size_t) + sizeof(size_t) + sizeof(u_char) + \
184 	    (ksize) + (dsize))
185 
186 /* Copy a BLEAF entry to the page. */
187 #define	WR_BLEAF(p, key, data, flags) { \
188 	*(size_t *)p = key->size; \
189 	p += sizeof(size_t); \
190 	*(size_t *)p = data->size; \
191 	p += sizeof(size_t); \
192 	*(u_char *)p = flags; \
193 	p += sizeof(u_char); \
194 	memmove(p, key->data, key->size); \
195 	p += key->size; \
196 	memmove(p, data->data, data->size); \
197 }
198 
199 /* For the recno leaf pages, the item is a data entry. */
200 typedef struct _rleaf {
201 	size_t	dsize;			/* size of data */
202 	u_char	flags;			/* P_BIGDATA */
203 	char	bytes[1];
204 } RLEAF;
205 
206 /* Get the page's RLEAF structure at index indx. */
207 #define	GETRLEAF(pg, indx) \
208 	((RLEAF *)((char *)(pg) + (pg)->linp[indx]))
209 
210 /* Get the number of bytes in the entry. */
211 #define NRLEAF(p)	NRLEAFDBT((p)->dsize)
212 
213 /* Get the number of bytes from the user's data. */
214 #define	NRLEAFDBT(dsize) \
215 	LALIGN(sizeof(size_t) + sizeof(u_char) + (dsize))
216 
217 /* Copy a RLEAF entry to the page. */
218 #define	WR_RLEAF(p, data, flags) { \
219 	*(size_t *)p = data->size; \
220 	p += sizeof(size_t); \
221 	*(u_char *)p = flags; \
222 	p += sizeof(u_char); \
223 	memmove(p, data->data, data->size); \
224 }
225 
226 /*
227  * A record in the tree is either a pointer to a page and an index in the page
228  * or a page number and an index.  These structures are used as a cursor, stack
229  * entry and search returns as well as to pass records to other routines.
230  *
231  * One comment about searches.  Internal page searches must find the largest
232  * record less than key in the tree so that descents work.  Leaf page searches
233  * must find the smallest record greater than key so that the returned index
234  * is the record's correct position for insertion.
235  *
236  * One comment about cursors.  The cursor key is never removed from the tree,
237  * even if deleted.  This is because it is quite difficult to decide where the
238  * cursor should be when other keys have been inserted/deleted in the tree;
239  * duplicate keys make it impossible.  This scheme does require extra work
240  * though, to make sure that we don't perform an operation on a deleted key.
241  */
242 typedef struct _epgno {
243 	pgno_t	pgno;			/* the page number */
244 	indx_t	index;			/* the index on the page */
245 } EPGNO;
246 
247 typedef struct _epg {
248 	PAGE	*page;			/* the (pinned) page */
249 	indx_t	 index;			/* the index on the page */
250 } EPG;
251 
252 /*
253  * The metadata of the tree.  The m_nrecs field is used only by the RECNO code.
254  * This is because the btree doesn't really need it and it requires that every
255  * put or delete call modify the metadata.
256  */
257 typedef struct _btmeta {
258 	u_int32_t	m_magic;	/* magic number */
259 	u_int32_t	m_version;	/* version */
260 	u_int32_t	m_psize;	/* page size */
261 	u_int32_t	m_free;		/* page number of first free page */
262 	u_int32_t	m_nrecs;	/* R: number of records */
263 #define	SAVEMETA	(B_NODUPS | R_RECNO)
264 	u_int32_t	m_flags;	/* bt_flags & SAVEMETA */
265 	u_int32_t	m_unused;	/* unused */
266 } BTMETA;
267 
268 /* The in-memory btree/recno data structure. */
269 typedef struct _btree {
270 	MPOOL	*bt_mp;			/* memory pool cookie */
271 
272 	DB	*bt_dbp;		/* pointer to enclosing DB */
273 
274 	EPG	bt_cur;			/* current (pinned) page */
275 	PAGE	*bt_pinned;		/* page pinned across calls */
276 
277 	EPGNO	bt_bcursor;		/* B: btree cursor */
278 	recno_t	bt_rcursor;		/* R: recno cursor (1-based) */
279 
280 #define	BT_POP(t)	(t->bt_sp ? t->bt_stack + --t->bt_sp : NULL)
281 #define	BT_CLR(t)	(t->bt_sp = 0)
282 	EPGNO	*bt_stack;		/* stack of parent pages */
283 	u_int	bt_sp;			/* current stack pointer */
284 	u_int	bt_maxstack;		/* largest stack */
285 
286 	char	*bt_kbuf;		/* key buffer */
287 	size_t	bt_kbufsz;		/* key buffer size */
288 	char	*bt_dbuf;		/* data buffer */
289 	size_t	bt_dbufsz;		/* data buffer size */
290 
291 	int	bt_fd;			/* tree file descriptor */
292 
293 	pgno_t	bt_free;		/* next free page */
294 	u_int32_t bt_psize;		/* page size */
295 	indx_t	bt_ovflsize;		/* cut-off for key/data overflow */
296 	int	bt_lorder;		/* byte order */
297 					/* sorted order */
298 	enum { NOT, BACK, FORWARD } bt_order;
299 	EPGNO	bt_last;		/* last insert */
300 
301 					/* B: key comparison function */
302 	int	(*bt_cmp) __P((const DBT *, const DBT *));
303 					/* B: prefix comparison function */
304 	size_t	(*bt_pfx) __P((const DBT *, const DBT *));
305 					/* R: recno input function */
306 	int	(*bt_irec) __P((struct _btree *, recno_t));
307 
308 	FILE	*bt_rfp;		/* R: record FILE pointer */
309 	int	bt_rfd;			/* R: record file descriptor */
310 
311 	caddr_t	bt_cmap;		/* R: current point in mapped space */
312 	caddr_t	bt_smap;		/* R: start of mapped space */
313 	caddr_t bt_emap;		/* R: end of mapped space */
314 	size_t	bt_msize;		/* R: size of mapped region. */
315 
316 	recno_t	bt_nrecs;		/* R: number of records */
317 	size_t	bt_reclen;		/* R: fixed record length */
318 	u_char	bt_bval;		/* R: delimiting byte/pad character */
319 
320 /*
321  * NB:
322  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
323  */
324 #define	B_DELCRSR	0x00001		/* cursor has been deleted */
325 #define	B_INMEM		0x00002		/* in-memory tree */
326 #define	B_METADIRTY	0x00004		/* need to write metadata */
327 #define	B_MODIFIED	0x00008		/* tree modified */
328 #define	B_NEEDSWAP	0x00010		/* if byte order requires swapping */
329 #define	B_NODUPS	0x00020		/* no duplicate keys permitted */
330 #define	B_RDONLY	0x00040		/* read-only tree */
331 #define	R_RECNO		0x00080		/* record oriented tree */
332 #define	B_SEQINIT	0x00100		/* sequential scan initialized */
333 
334 #define	R_CLOSEFP	0x00200		/* opened a file pointer */
335 #define	R_EOF		0x00400		/* end of input file reached. */
336 #define	R_FIXLEN	0x00800		/* fixed length records */
337 #define	R_MEMMAPPED	0x01000		/* memory mapped file. */
338 #define	R_INMEM		0x02000		/* in-memory file */
339 #define	R_MODIFIED	0x04000		/* modified file */
340 #define	R_RDONLY	0x08000		/* read-only file */
341 
342 #define	B_DB_LOCK	0x10000		/* DB_LOCK specified. */
343 #define	B_DB_SHMEM	0x20000		/* DB_SHMEM specified. */
344 #define	B_DB_TXN	0x40000		/* DB_TXN specified. */
345 
346 	u_int32_t	bt_flags;	/* btree state */
347 } BTREE;
348 
349 #define	SET(t, f)	((t)->bt_flags |= (f))
350 #define	CLR(t, f)	((t)->bt_flags &= ~(f))
351 #define	ISSET(t, f)	((t)->bt_flags & (f))
352 
353 #include "extern.h"
354