xref: /freebsd/lib/libc/db/btree/btree.h (revision e0c4386e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /* Macros to set/clear/test flags. */
36 #define	F_SET(p, f)	(p)->flags |= (f)
37 #define	F_CLR(p, f)	(p)->flags &= ~(f)
38 #define	F_ISSET(p, f)	((p)->flags & (f))
39 
40 #include <mpool.h>
41 
42 #define	DEFMINKEYPAGE	(2)		/* Minimum keys per page */
43 #define	MINCACHE	(5)		/* Minimum cached pages */
44 #define	MINPSIZE	(512)		/* Minimum page size */
45 
46 /*
47  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
48  * used as an out-of-band page, i.e. page pointers that point to nowhere point
49  * to page 0.  Page 1 is the root of the btree.
50  */
51 #define	P_INVALID	 0		/* Invalid tree page number. */
52 #define	P_META		 0		/* Tree metadata page number. */
53 #define	P_ROOT		 1		/* Tree root page number. */
54 
55 /*
56  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
57  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
58  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
59  * This implementation requires that values within structures NOT be padded.
60  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
61  * to do some work to get this package to run.
62  */
63 typedef struct _page {
64 	pgno_t	pgno;			/* this page's page number */
65 	pgno_t	prevpg;			/* left sibling */
66 	pgno_t	nextpg;			/* right sibling */
67 
68 #define	P_BINTERNAL	0x01		/* btree internal page */
69 #define	P_BLEAF		0x02		/* leaf page */
70 #define	P_OVERFLOW	0x04		/* overflow page */
71 #define	P_RINTERNAL	0x08		/* recno internal page */
72 #define	P_RLEAF		0x10		/* leaf page */
73 #define P_TYPE		0x1f		/* type mask */
74 #define	P_PRESERVE	0x20		/* never delete this chain of pages */
75 	u_int32_t flags;
76 
77 	indx_t	lower;			/* lower bound of free space on page */
78 	indx_t	upper;			/* upper bound of free space on page */
79 	indx_t	linp[1];		/* indx_t-aligned VAR. LENGTH DATA */
80 } PAGE;
81 
82 /* First and next index. */
83 #define	BTDATAOFF							\
84 	(sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) +		\
85 	    sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t))
86 #define	NEXTINDEX(p)	(((p)->lower - BTDATAOFF) / sizeof(indx_t))
87 
88 /*
89  * For pages other than overflow pages, there is an array of offsets into the
90  * rest of the page immediately following the page header.  Each offset is to
91  * an item which is unique to the type of page.  The h_lower offset is just
92  * past the last filled-in index.  The h_upper offset is the first item on the
93  * page.  Offsets are from the beginning of the page.
94  *
95  * If an item is too big to store on a single page, a flag is set and the item
96  * is a { page, size } pair such that the page is the first page of an overflow
97  * chain with size bytes of item.  Overflow pages are simply bytes without any
98  * external structure.
99  *
100  * The page number and size fields in the items are pgno_t-aligned so they can
101  * be manipulated without copying.  (This presumes that 32 bit items can be
102  * manipulated on this system.)
103  */
104 #define	LALIGN(n)	(((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1))
105 #define	NOVFLSIZE	(sizeof(pgno_t) + sizeof(u_int32_t))
106 
107 /*
108  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
109  * pairs, such that the key compares less than or equal to all of the records
110  * on that page.  For a tree without duplicate keys, an internal page with two
111  * consecutive keys, a and b, will have all records greater than or equal to a
112  * and less than b stored on the page associated with a.  Duplicate keys are
113  * somewhat special and can cause duplicate internal and leaf page records and
114  * some minor modifications of the above rule.
115  */
116 typedef struct _binternal {
117 	u_int32_t ksize;		/* key size */
118 	pgno_t	pgno;			/* page number stored on */
119 #define	P_BIGDATA	0x01		/* overflow data */
120 #define	P_BIGKEY	0x02		/* overflow key */
121 	u_char	flags;
122 	char	bytes[1];		/* data */
123 } BINTERNAL;
124 
125 /* Get the page's BINTERNAL structure at index indx. */
126 #define	GETBINTERNAL(pg, indx)						\
127 	((BINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
128 
129 /* Get the number of bytes in the entry. */
130 #define NBINTERNAL(len)							\
131 	LALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len))
132 
133 /* Copy a BINTERNAL entry to the page. */
134 #define	WR_BINTERNAL(p, size, pgno, flags) {				\
135 	*(u_int32_t *)p = size;						\
136 	p += sizeof(u_int32_t);						\
137 	*(pgno_t *)p = pgno;						\
138 	p += sizeof(pgno_t);						\
139 	*(u_char *)p = flags;						\
140 	p += sizeof(u_char);						\
141 }
142 
143 /*
144  * For the recno internal pages, the item is a page number with the number of
145  * keys found on that page and below.
146  */
147 typedef struct _rinternal {
148 	recno_t	nrecs;			/* number of records */
149 	pgno_t	pgno;			/* page number stored below */
150 } RINTERNAL;
151 
152 /* Get the page's RINTERNAL structure at index indx. */
153 #define	GETRINTERNAL(pg, indx)						\
154 	((RINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
155 
156 /* Get the number of bytes in the entry. */
157 #define NRINTERNAL							\
158 	LALIGN(sizeof(recno_t) + sizeof(pgno_t))
159 
160 /* Copy a RINTERAL entry to the page. */
161 #define	WR_RINTERNAL(p, nrecs, pgno) {					\
162 	*(recno_t *)p = nrecs;						\
163 	p += sizeof(recno_t);						\
164 	*(pgno_t *)p = pgno;						\
165 }
166 
167 /* For the btree leaf pages, the item is a key and data pair. */
168 typedef struct _bleaf {
169 	u_int32_t	ksize;		/* size of key */
170 	u_int32_t	dsize;		/* size of data */
171 	u_char	flags;			/* P_BIGDATA, P_BIGKEY */
172 	char	bytes[1];		/* data */
173 } BLEAF;
174 
175 /* Get the page's BLEAF structure at index indx. */
176 #define	GETBLEAF(pg, indx)						\
177 	((BLEAF *)((char *)(pg) + (pg)->linp[indx]))
178 
179 /* Get the number of bytes in the entry. */
180 #define NBLEAF(p)	NBLEAFDBT((p)->ksize, (p)->dsize)
181 
182 /* Get the number of bytes in the user's key/data pair. */
183 #define NBLEAFDBT(ksize, dsize)						\
184 	LALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) +	\
185 	    (ksize) + (dsize))
186 
187 /* Copy a BLEAF entry to the page. */
188 #define	WR_BLEAF(p, key, data, flags) {					\
189 	*(u_int32_t *)p = key->size;					\
190 	p += sizeof(u_int32_t);						\
191 	*(u_int32_t *)p = data->size;					\
192 	p += sizeof(u_int32_t);						\
193 	*(u_char *)p = flags;						\
194 	p += sizeof(u_char);						\
195 	memmove(p, key->data, key->size);				\
196 	p += key->size;							\
197 	memmove(p, data->data, data->size);				\
198 }
199 
200 /* For the recno leaf pages, the item is a data entry. */
201 typedef struct _rleaf {
202 	u_int32_t	dsize;		/* size of data */
203 	u_char	flags;			/* P_BIGDATA */
204 	char	bytes[1];
205 } RLEAF;
206 
207 /* Get the page's RLEAF structure at index indx. */
208 #define	GETRLEAF(pg, indx)						\
209 	((RLEAF *)((char *)(pg) + (pg)->linp[indx]))
210 
211 /* Get the number of bytes in the entry. */
212 #define NRLEAF(p)	NRLEAFDBT((p)->dsize)
213 
214 /* Get the number of bytes from the user's data. */
215 #define	NRLEAFDBT(dsize)						\
216 	LALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize))
217 
218 /* Copy a RLEAF entry to the page. */
219 #define	WR_RLEAF(p, data, flags) {					\
220 	*(u_int32_t *)p = data->size;					\
221 	p += sizeof(u_int32_t);						\
222 	*(u_char *)p = flags;						\
223 	p += sizeof(u_char);						\
224 	memmove(p, data->data, data->size);				\
225 }
226 
227 /*
228  * A record in the tree is either a pointer to a page and an index in the page
229  * or a page number and an index.  These structures are used as a cursor, stack
230  * entry and search returns as well as to pass records to other routines.
231  *
232  * One comment about searches.  Internal page searches must find the largest
233  * record less than key in the tree so that descents work.  Leaf page searches
234  * must find the smallest record greater than key so that the returned index
235  * is the record's correct position for insertion.
236  */
237 typedef struct _epgno {
238 	pgno_t	pgno;			/* the page number */
239 	indx_t	index;			/* the index on the page */
240 } EPGNO;
241 
242 typedef struct _epg {
243 	PAGE	*page;			/* the (pinned) page */
244 	indx_t	 index;			/* the index on the page */
245 } EPG;
246 
247 /*
248  * About cursors.  The cursor (and the page that contained the key/data pair
249  * that it referenced) can be deleted, which makes things a bit tricky.  If
250  * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set
251  * or there simply aren't any duplicates of the key) we copy the key that it
252  * referenced when it's deleted, and reacquire a new cursor key if the cursor
253  * is used again.  If there are duplicates keys, we move to the next/previous
254  * key, and set a flag so that we know what happened.  NOTE: if duplicate (to
255  * the cursor) keys are added to the tree during this process, it is undefined
256  * if they will be returned or not in a cursor scan.
257  *
258  * The flags determine the possible states of the cursor:
259  *
260  * CURS_INIT	The cursor references *something*.
261  * CURS_ACQUIRE	The cursor was deleted, and a key has been saved so that
262  *		we can reacquire the right position in the tree.
263  * CURS_AFTER, CURS_BEFORE
264  *		The cursor was deleted, and now references a key/data pair
265  *		that has not yet been returned, either before or after the
266  *		deleted key/data pair.
267  * XXX
268  * This structure is broken out so that we can eventually offer multiple
269  * cursors as part of the DB interface.
270  */
271 typedef struct _cursor {
272 	EPGNO	 pg;			/* B: Saved tree reference. */
273 	DBT	 key;			/* B: Saved key, or key.data == NULL. */
274 	recno_t	 rcursor;		/* R: recno cursor (1-based) */
275 
276 #define	CURS_ACQUIRE	0x01		/*  B: Cursor needs to be reacquired. */
277 #define	CURS_AFTER	0x02		/*  B: Unreturned cursor after key. */
278 #define	CURS_BEFORE	0x04		/*  B: Unreturned cursor before key. */
279 #define	CURS_INIT	0x08		/* RB: Cursor initialized. */
280 	u_int8_t flags;
281 } CURSOR;
282 
283 /*
284  * The metadata of the tree.  The nrecs field is used only by the RECNO code.
285  * This is because the btree doesn't really need it and it requires that every
286  * put or delete call modify the metadata.
287  */
288 typedef struct _btmeta {
289 	u_int32_t	magic;		/* magic number */
290 	u_int32_t	version;	/* version */
291 	u_int32_t	psize;		/* page size */
292 	u_int32_t	free;		/* page number of first free page */
293 	u_int32_t	nrecs;		/* R: number of records */
294 
295 #define	SAVEMETA	(B_NODUPS | R_RECNO)
296 	u_int32_t	flags;		/* bt_flags & SAVEMETA */
297 } BTMETA;
298 
299 /* The in-memory btree/recno data structure. */
300 typedef struct _btree {
301 	MPOOL	 *bt_mp;		/* memory pool cookie */
302 
303 	DB	 *bt_dbp;		/* pointer to enclosing DB */
304 
305 	EPG	  bt_cur;		/* current (pinned) page */
306 	PAGE	 *bt_pinned;		/* page pinned across calls */
307 
308 	CURSOR	  bt_cursor;		/* cursor */
309 
310 #define	BT_PUSH(t, p, i) {						\
311 	t->bt_sp->pgno = p;						\
312 	t->bt_sp->index = i;						\
313 	++t->bt_sp;							\
314 }
315 #define	BT_POP(t)	(t->bt_sp == t->bt_stack ? NULL : --t->bt_sp)
316 #define	BT_CLR(t)	(t->bt_sp = t->bt_stack)
317 	EPGNO	  bt_stack[50];		/* stack of parent pages */
318 	EPGNO	 *bt_sp;		/* current stack pointer */
319 
320 	DBT	  bt_rkey;		/* returned key */
321 	DBT	  bt_rdata;		/* returned data */
322 
323 	int	  bt_fd;		/* tree file descriptor */
324 
325 	pgno_t	  bt_free;		/* next free page */
326 	u_int32_t bt_psize;		/* page size */
327 	indx_t	  bt_ovflsize;		/* cut-off for key/data overflow */
328 	int	  bt_lorder;		/* byte order */
329 					/* sorted order */
330 	enum { NOT, BACK, FORWARD } bt_order;
331 	EPGNO	  bt_last;		/* last insert */
332 
333 					/* B: key comparison function */
334 	int	(*bt_cmp)(const DBT *, const DBT *);
335 					/* B: prefix comparison function */
336 	size_t	(*bt_pfx)(const DBT *, const DBT *);
337 					/* R: recno input function */
338 	int	(*bt_irec)(struct _btree *, recno_t);
339 
340 	FILE	 *bt_rfp;		/* R: record FILE pointer */
341 	int	  bt_rfd;		/* R: record file descriptor */
342 
343 	caddr_t	  bt_cmap;		/* R: current point in mapped space */
344 	caddr_t	  bt_smap;		/* R: start of mapped space */
345 	caddr_t   bt_emap;		/* R: end of mapped space */
346 	size_t	  bt_msize;		/* R: size of mapped region. */
347 
348 	recno_t	  bt_nrecs;		/* R: number of records */
349 	size_t	  bt_reclen;		/* R: fixed record length */
350 	u_char	  bt_bval;		/* R: delimiting byte/pad character */
351 
352 /*
353  * NB:
354  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
355  */
356 #define	B_INMEM		0x00001		/* in-memory tree */
357 #define	B_METADIRTY	0x00002		/* need to write metadata */
358 #define	B_MODIFIED	0x00004		/* tree modified */
359 #define	B_NEEDSWAP	0x00008		/* if byte order requires swapping */
360 #define	B_RDONLY	0x00010		/* read-only tree */
361 
362 #define	B_NODUPS	0x00020		/* no duplicate keys permitted */
363 #define	R_RECNO		0x00080		/* record oriented tree */
364 
365 #define	R_CLOSEFP	0x00040		/* opened a file pointer */
366 #define	R_EOF		0x00100		/* end of input file reached. */
367 #define	R_FIXLEN	0x00200		/* fixed length records */
368 #define	R_MEMMAPPED	0x00400		/* memory mapped file. */
369 #define	R_INMEM		0x00800		/* in-memory file */
370 #define	R_MODIFIED	0x01000		/* modified file */
371 #define	R_RDONLY	0x02000		/* read-only file */
372 
373 #define	B_DB_LOCK	0x04000		/* DB_LOCK specified. */
374 #define	B_DB_SHMEM	0x08000		/* DB_SHMEM specified. */
375 #define	B_DB_TXN	0x10000		/* DB_TXN specified. */
376 	u_int32_t flags;
377 } BTREE;
378 
379 #include "extern.h"
380