xref: /freebsd/lib/libc/db/btree/btree.h (revision 3bdf775801b218aa5a89564839405b122f4b233e)
1 /*-
2  * Copyright (c) 1991, 1993, 1994
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)btree.h	8.11 (Berkeley) 8/17/94
33  * $FreeBSD$
34  */
35 
36 /* Macros to set/clear/test flags. */
37 #define	F_SET(p, f)	(p)->flags |= (f)
38 #define	F_CLR(p, f)	(p)->flags &= ~(f)
39 #define	F_ISSET(p, f)	((p)->flags & (f))
40 
41 #include <mpool.h>
42 
43 #define	DEFMINKEYPAGE	(2)		/* Minimum keys per page */
44 #define	MINCACHE	(5)		/* Minimum cached pages */
45 #define	MINPSIZE	(512)		/* Minimum page size */
46 
47 /*
48  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
49  * used as an out-of-band page, i.e. page pointers that point to nowhere point
50  * to page 0.  Page 1 is the root of the btree.
51  */
52 #define	P_INVALID	 0		/* Invalid tree page number. */
53 #define	P_META		 0		/* Tree metadata page number. */
54 #define	P_ROOT		 1		/* Tree root page number. */
55 
56 /*
57  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
58  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
59  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
60  * This implementation requires that values within structures NOT be padded.
61  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
62  * to do some work to get this package to run.
63  */
64 typedef struct _page {
65 	pgno_t	pgno;			/* this page's page number */
66 	pgno_t	prevpg;			/* left sibling */
67 	pgno_t	nextpg;			/* right sibling */
68 
69 #define	P_BINTERNAL	0x01		/* btree internal page */
70 #define	P_BLEAF		0x02		/* leaf page */
71 #define	P_OVERFLOW	0x04		/* overflow page */
72 #define	P_RINTERNAL	0x08		/* recno internal page */
73 #define	P_RLEAF		0x10		/* leaf page */
74 #define P_TYPE		0x1f		/* type mask */
75 #define	P_PRESERVE	0x20		/* never delete this chain of pages */
76 	u_int32_t flags;
77 
78 	indx_t	lower;			/* lower bound of free space on page */
79 	indx_t	upper;			/* upper bound of free space on page */
80 	indx_t	linp[1];		/* indx_t-aligned VAR. LENGTH DATA */
81 } PAGE;
82 
83 /* First and next index. */
84 #define	BTDATAOFF							\
85 	(sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) +		\
86 	    sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t))
87 #define	NEXTINDEX(p)	(((p)->lower - BTDATAOFF) / sizeof(indx_t))
88 
89 /*
90  * For pages other than overflow pages, there is an array of offsets into the
91  * rest of the page immediately following the page header.  Each offset is to
92  * an item which is unique to the type of page.  The h_lower offset is just
93  * past the last filled-in index.  The h_upper offset is the first item on the
94  * page.  Offsets are from the beginning of the page.
95  *
96  * If an item is too big to store on a single page, a flag is set and the item
97  * is a { page, size } pair such that the page is the first page of an overflow
98  * chain with size bytes of item.  Overflow pages are simply bytes without any
99  * external structure.
100  *
101  * The page number and size fields in the items are pgno_t-aligned so they can
102  * be manipulated without copying.  (This presumes that 32 bit items can be
103  * manipulated on this system.)
104  */
105 #define	LALIGN(n)	(((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1))
106 #define	NOVFLSIZE	(sizeof(pgno_t) + sizeof(u_int32_t))
107 
108 /*
109  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
110  * pairs, such that the key compares less than or equal to all of the records
111  * on that page.  For a tree without duplicate keys, an internal page with two
112  * consecutive keys, a and b, will have all records greater than or equal to a
113  * and less than b stored on the page associated with a.  Duplicate keys are
114  * somewhat special and can cause duplicate internal and leaf page records and
115  * some minor modifications of the above rule.
116  */
117 typedef struct _binternal {
118 	u_int32_t ksize;		/* key size */
119 	pgno_t	pgno;			/* page number stored on */
120 #define	P_BIGDATA	0x01		/* overflow data */
121 #define	P_BIGKEY	0x02		/* overflow key */
122 	u_char	flags;
123 	char	bytes[1];		/* data */
124 } BINTERNAL;
125 
126 /* Get the page's BINTERNAL structure at index indx. */
127 #define	GETBINTERNAL(pg, indx)						\
128 	((BINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
129 
130 /* Get the number of bytes in the entry. */
131 #define NBINTERNAL(len)							\
132 	LALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len))
133 
134 /* Copy a BINTERNAL entry to the page. */
135 #define	WR_BINTERNAL(p, size, pgno, flags) {				\
136 	*(u_int32_t *)p = size;						\
137 	p += sizeof(u_int32_t);						\
138 	*(pgno_t *)p = pgno;						\
139 	p += sizeof(pgno_t);						\
140 	*(u_char *)p = flags;						\
141 	p += sizeof(u_char);						\
142 }
143 
144 /*
145  * For the recno internal pages, the item is a page number with the number of
146  * keys found on that page and below.
147  */
148 typedef struct _rinternal {
149 	recno_t	nrecs;			/* number of records */
150 	pgno_t	pgno;			/* page number stored below */
151 } RINTERNAL;
152 
153 /* Get the page's RINTERNAL structure at index indx. */
154 #define	GETRINTERNAL(pg, indx)						\
155 	((RINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
156 
157 /* Get the number of bytes in the entry. */
158 #define NRINTERNAL							\
159 	LALIGN(sizeof(recno_t) + sizeof(pgno_t))
160 
161 /* Copy a RINTERAL entry to the page. */
162 #define	WR_RINTERNAL(p, nrecs, pgno) {					\
163 	*(recno_t *)p = nrecs;						\
164 	p += sizeof(recno_t);						\
165 	*(pgno_t *)p = pgno;						\
166 }
167 
168 /* For the btree leaf pages, the item is a key and data pair. */
169 typedef struct _bleaf {
170 	u_int32_t	ksize;		/* size of key */
171 	u_int32_t	dsize;		/* size of data */
172 	u_char	flags;			/* P_BIGDATA, P_BIGKEY */
173 	char	bytes[1];		/* data */
174 } BLEAF;
175 
176 /* Get the page's BLEAF structure at index indx. */
177 #define	GETBLEAF(pg, indx)						\
178 	((BLEAF *)((char *)(pg) + (pg)->linp[indx]))
179 
180 /* Get the number of bytes in the entry. */
181 #define NBLEAF(p)	NBLEAFDBT((p)->ksize, (p)->dsize)
182 
183 /* Get the number of bytes in the user's key/data pair. */
184 #define NBLEAFDBT(ksize, dsize)						\
185 	LALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) +	\
186 	    (ksize) + (dsize))
187 
188 /* Copy a BLEAF entry to the page. */
189 #define	WR_BLEAF(p, key, data, flags) {					\
190 	*(u_int32_t *)p = key->size;					\
191 	p += sizeof(u_int32_t);						\
192 	*(u_int32_t *)p = data->size;					\
193 	p += sizeof(u_int32_t);						\
194 	*(u_char *)p = flags;						\
195 	p += sizeof(u_char);						\
196 	memmove(p, key->data, key->size);				\
197 	p += key->size;							\
198 	memmove(p, data->data, data->size);				\
199 }
200 
201 /* For the recno leaf pages, the item is a data entry. */
202 typedef struct _rleaf {
203 	u_int32_t	dsize;		/* size of data */
204 	u_char	flags;			/* P_BIGDATA */
205 	char	bytes[1];
206 } RLEAF;
207 
208 /* Get the page's RLEAF structure at index indx. */
209 #define	GETRLEAF(pg, indx)						\
210 	((RLEAF *)((char *)(pg) + (pg)->linp[indx]))
211 
212 /* Get the number of bytes in the entry. */
213 #define NRLEAF(p)	NRLEAFDBT((p)->dsize)
214 
215 /* Get the number of bytes from the user's data. */
216 #define	NRLEAFDBT(dsize)						\
217 	LALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize))
218 
219 /* Copy a RLEAF entry to the page. */
220 #define	WR_RLEAF(p, data, flags) {					\
221 	*(u_int32_t *)p = data->size;					\
222 	p += sizeof(u_int32_t);						\
223 	*(u_char *)p = flags;						\
224 	p += sizeof(u_char);						\
225 	memmove(p, data->data, data->size);				\
226 }
227 
228 /*
229  * A record in the tree is either a pointer to a page and an index in the page
230  * or a page number and an index.  These structures are used as a cursor, stack
231  * entry and search returns as well as to pass records to other routines.
232  *
233  * One comment about searches.  Internal page searches must find the largest
234  * record less than key in the tree so that descents work.  Leaf page searches
235  * must find the smallest record greater than key so that the returned index
236  * is the record's correct position for insertion.
237  */
238 typedef struct _epgno {
239 	pgno_t	pgno;			/* the page number */
240 	indx_t	index;			/* the index on the page */
241 } EPGNO;
242 
243 typedef struct _epg {
244 	PAGE	*page;			/* the (pinned) page */
245 	indx_t	 index;			/* the index on the page */
246 } EPG;
247 
248 /*
249  * About cursors.  The cursor (and the page that contained the key/data pair
250  * that it referenced) can be deleted, which makes things a bit tricky.  If
251  * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set
252  * or there simply aren't any duplicates of the key) we copy the key that it
253  * referenced when it's deleted, and reacquire a new cursor key if the cursor
254  * is used again.  If there are duplicates keys, we move to the next/previous
255  * key, and set a flag so that we know what happened.  NOTE: if duplicate (to
256  * the cursor) keys are added to the tree during this process, it is undefined
257  * if they will be returned or not in a cursor scan.
258  *
259  * The flags determine the possible states of the cursor:
260  *
261  * CURS_INIT	The cursor references *something*.
262  * CURS_ACQUIRE	The cursor was deleted, and a key has been saved so that
263  *		we can reacquire the right position in the tree.
264  * CURS_AFTER, CURS_BEFORE
265  *		The cursor was deleted, and now references a key/data pair
266  *		that has not yet been returned, either before or after the
267  *		deleted key/data pair.
268  * XXX
269  * This structure is broken out so that we can eventually offer multiple
270  * cursors as part of the DB interface.
271  */
272 typedef struct _cursor {
273 	EPGNO	 pg;			/* B: Saved tree reference. */
274 	DBT	 key;			/* B: Saved key, or key.data == NULL. */
275 	recno_t	 rcursor;		/* R: recno cursor (1-based) */
276 
277 #define	CURS_ACQUIRE	0x01		/*  B: Cursor needs to be reacquired. */
278 #define	CURS_AFTER	0x02		/*  B: Unreturned cursor after key. */
279 #define	CURS_BEFORE	0x04		/*  B: Unreturned cursor before key. */
280 #define	CURS_INIT	0x08		/* RB: Cursor initialized. */
281 	u_int8_t flags;
282 } CURSOR;
283 
284 /*
285  * The metadata of the tree.  The nrecs field is used only by the RECNO code.
286  * This is because the btree doesn't really need it and it requires that every
287  * put or delete call modify the metadata.
288  */
289 typedef struct _btmeta {
290 	u_int32_t	magic;		/* magic number */
291 	u_int32_t	version;	/* version */
292 	u_int32_t	psize;		/* page size */
293 	u_int32_t	free;		/* page number of first free page */
294 	u_int32_t	nrecs;		/* R: number of records */
295 
296 #define	SAVEMETA	(B_NODUPS | R_RECNO)
297 	u_int32_t	flags;		/* bt_flags & SAVEMETA */
298 } BTMETA;
299 
300 /* The in-memory btree/recno data structure. */
301 typedef struct _btree {
302 	MPOOL	 *bt_mp;		/* memory pool cookie */
303 
304 	DB	 *bt_dbp;		/* pointer to enclosing DB */
305 
306 	EPG	  bt_cur;		/* current (pinned) page */
307 	PAGE	 *bt_pinned;		/* page pinned across calls */
308 
309 	CURSOR	  bt_cursor;		/* cursor */
310 
311 #define	BT_PUSH(t, p, i) {						\
312 	t->bt_sp->pgno = p;						\
313 	t->bt_sp->index = i;						\
314 	++t->bt_sp;							\
315 }
316 #define	BT_POP(t)	(t->bt_sp == t->bt_stack ? NULL : --t->bt_sp)
317 #define	BT_CLR(t)	(t->bt_sp = t->bt_stack)
318 	EPGNO	  bt_stack[50];		/* stack of parent pages */
319 	EPGNO	 *bt_sp;		/* current stack pointer */
320 
321 	DBT	  bt_rkey;		/* returned key */
322 	DBT	  bt_rdata;		/* returned data */
323 
324 	int	  bt_fd;		/* tree file descriptor */
325 
326 	pgno_t	  bt_free;		/* next free page */
327 	u_int32_t bt_psize;		/* page size */
328 	indx_t	  bt_ovflsize;		/* cut-off for key/data overflow */
329 	int	  bt_lorder;		/* byte order */
330 					/* sorted order */
331 	enum { NOT, BACK, FORWARD } bt_order;
332 	EPGNO	  bt_last;		/* last insert */
333 
334 					/* B: key comparison function */
335 	int	(*bt_cmp)(const DBT *, const DBT *);
336 					/* B: prefix comparison function */
337 	size_t	(*bt_pfx)(const DBT *, const DBT *);
338 					/* R: recno input function */
339 	int	(*bt_irec)(struct _btree *, recno_t);
340 
341 	FILE	 *bt_rfp;		/* R: record FILE pointer */
342 	int	  bt_rfd;		/* R: record file descriptor */
343 
344 	caddr_t	  bt_cmap;		/* R: current point in mapped space */
345 	caddr_t	  bt_smap;		/* R: start of mapped space */
346 	caddr_t   bt_emap;		/* R: end of mapped space */
347 	size_t	  bt_msize;		/* R: size of mapped region. */
348 
349 	recno_t	  bt_nrecs;		/* R: number of records */
350 	size_t	  bt_reclen;		/* R: fixed record length */
351 	u_char	  bt_bval;		/* R: delimiting byte/pad character */
352 
353 /*
354  * NB:
355  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
356  */
357 #define	B_INMEM		0x00001		/* in-memory tree */
358 #define	B_METADIRTY	0x00002		/* need to write metadata */
359 #define	B_MODIFIED	0x00004		/* tree modified */
360 #define	B_NEEDSWAP	0x00008		/* if byte order requires swapping */
361 #define	B_RDONLY	0x00010		/* read-only tree */
362 
363 #define	B_NODUPS	0x00020		/* no duplicate keys permitted */
364 #define	R_RECNO		0x00080		/* record oriented tree */
365 
366 #define	R_CLOSEFP	0x00040		/* opened a file pointer */
367 #define	R_EOF		0x00100		/* end of input file reached. */
368 #define	R_FIXLEN	0x00200		/* fixed length records */
369 #define	R_MEMMAPPED	0x00400		/* memory mapped file. */
370 #define	R_INMEM		0x00800		/* in-memory file */
371 #define	R_MODIFIED	0x01000		/* modified file */
372 #define	R_RDONLY	0x02000		/* read-only file */
373 
374 #define	B_DB_LOCK	0x04000		/* DB_LOCK specified. */
375 #define	B_DB_SHMEM	0x08000		/* DB_SHMEM specified. */
376 #define	B_DB_TXN	0x10000		/* DB_TXN specified. */
377 	u_int32_t flags;
378 } BTREE;
379 
380 #include "extern.h"
381