xref: /minix/lib/libc/db/btree/btree.h (revision 84d9c625)
1 /*	$NetBSD: btree.h,v 1.17 2013/09/04 13:03:22 ryoon 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. 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  *	@(#)btree.h	8.11 (Berkeley) 8/17/94
35  */
36 
37 #if HAVE_NBTOOL_CONFIG_H
38 #include "nbtool_config.h"
39 #endif
40 
41 /* Macros to set/clear/test flags. */
42 #define	F_SET(p, f)	(p)->flags |= (f)
43 #define	F_CLR(p, f)	(p)->flags &= ~(f)
44 #define	F_ISSET(p, f)	((p)->flags & (f))
45 
46 #include <mpool.h>
47 
48 #define	DEFMINKEYPAGE	(2)		/* Minimum keys per page */
49 #define	MINCACHE	(5)		/* Minimum cached pages */
50 #define	MINPSIZE	(512)		/* Minimum page size */
51 
52 /*
53  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
54  * used as an out-of-band page, i.e. page pointers that point to nowhere point
55  * to page 0.  Page 1 is the root of the btree.
56  */
57 #define	P_INVALID	 0		/* Invalid tree page number. */
58 #define	P_META		 0		/* Tree metadata page number. */
59 #define	P_ROOT		 1		/* Tree root page number. */
60 
61 /*
62  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
63  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
64  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
65  * This implementation requires that values within structures NOT be padded.
66  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
67  * to do some work to get this package to run.
68  */
69 typedef struct _page {
70 	pgno_t	pgno;			/* this page's page number */
71 	pgno_t	prevpg;			/* left sibling */
72 	pgno_t	nextpg;			/* right sibling */
73 
74 #define	P_BINTERNAL	0x01		/* btree internal page */
75 #define	P_BLEAF		0x02		/* leaf page */
76 #define	P_OVERFLOW	0x04		/* overflow page */
77 #define	P_RINTERNAL	0x08		/* recno internal page */
78 #define	P_RLEAF		0x10		/* leaf page */
79 #define P_TYPE		0x1f		/* type mask */
80 #define	P_PRESERVE	0x20		/* never delete this chain of pages */
81 	uint32_t flags;
82 
83 	indx_t	lower;			/* lower bound of free space on page */
84 	indx_t	upper;			/* upper bound of free space on page */
85 	indx_t	linp[1];		/* indx_t-aligned VAR. LENGTH DATA */
86 } PAGE;
87 
88 /* First and next index. */
89 #define	BTDATAOFF							\
90 	(sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) +		\
91 	    sizeof(uint32_t) + sizeof(indx_t) + sizeof(indx_t))
92 
93 #define	_NEXTINDEX(p) (((p)->lower - BTDATAOFF) / sizeof(indx_t))
94 #ifdef _DIAGNOSTIC
95 static __inline indx_t
NEXTINDEX(const PAGE * p)96 NEXTINDEX(const PAGE *p) {
97 	size_t x = _NEXTINDEX(p);
98 	_DBFIT(x, indx_t);
99 	return (indx_t)x;
100 }
101 #else
102 #define	NEXTINDEX(p) (indx_t)_NEXTINDEX(p)
103 #endif
104 
105 /*
106  * For pages other than overflow pages, there is an array of offsets into the
107  * rest of the page immediately following the page header.  Each offset is to
108  * an item which is unique to the type of page.  The h_lower offset is just
109  * past the last filled-in index.  The h_upper offset is the first item on the
110  * page.  Offsets are from the beginning of the page.
111  *
112  * If an item is too big to store on a single page, a flag is set and the item
113  * is a { page, size } pair such that the page is the first page of an overflow
114  * chain with size bytes of item.  Overflow pages are simply bytes without any
115  * external structure.
116  *
117  * The page number and size fields in the items are pgno_t-aligned so they can
118  * be manipulated without copying.  (This presumes that 32 bit items can be
119  * manipulated on this system.)
120  */
121 #define	BTLALIGN(n)	(((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1))
122 #define	NOVFLSIZE	(sizeof(pgno_t) + sizeof(uint32_t))
123 
124 /*
125  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
126  * pairs, such that the key compares less than or equal to all of the records
127  * on that page.  For a tree without duplicate keys, an internal page with two
128  * consecutive keys, a and b, will have all records greater than or equal to a
129  * and less than b stored on the page associated with a.  Duplicate keys are
130  * somewhat special and can cause duplicate internal and leaf page records and
131  * some minor modifications of the above rule.
132  */
133 typedef struct _binternal {
134 	uint32_t ksize;			/* key size */
135 	pgno_t	pgno;			/* page number stored on */
136 #define	P_BIGDATA	0x01		/* overflow data */
137 #define	P_BIGKEY	0x02		/* overflow key */
138 	uint8_t	flags;
139 	char	bytes[1];		/* data */
140 } BINTERNAL;
141 
142 /* Get the page's BINTERNAL structure at index indx. */
143 #define	GETBINTERNAL(pg, indx)						\
144 	((BINTERNAL *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
145 
146 /* Get the number of bytes in the entry. */
147 #define _NBINTERNAL(len)						\
148     BTLALIGN(sizeof(uint32_t) + sizeof(pgno_t) + sizeof(uint8_t) + (len))
149 #ifdef _DIAGNOSTIC
150 static __inline uint32_t
NBINTERNAL(uint32_t len)151 NBINTERNAL(uint32_t len) {
152 	size_t x = _NBINTERNAL(len);
153 	_DBFIT(x, uint32_t);
154 	return (uint32_t)x;
155 }
156 #else
157 #define NBINTERNAL(len)	(uint32_t)_NBINTERNAL(len)
158 #endif
159 
160 /* Copy a BINTERNAL entry to the page. */
161 #define	WR_BINTERNAL(p, size, pgno, flags) do {				\
162 	_DBFIT(size, uint32_t);						\
163 	*(uint32_t *)(void *)p = (uint32_t)size;			\
164 	p += sizeof(uint32_t);						\
165 	*(pgno_t *)(void *)p = pgno;					\
166 	p += sizeof(pgno_t);						\
167 	*(uint8_t *)(void *)p = flags;					\
168 	p += sizeof(uint8_t);						\
169 } while (/*CONSTCOND*/0)
170 
171 /*
172  * For the recno internal pages, the item is a page number with the number of
173  * keys found on that page and below.
174  */
175 typedef struct _rinternal {
176 	recno_t	nrecs;			/* number of records */
177 	pgno_t	pgno;			/* page number stored below */
178 } RINTERNAL;
179 
180 /* Get the page's RINTERNAL structure at index indx. */
181 #define	GETRINTERNAL(pg, indx)						\
182 	((RINTERNAL *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
183 
184 /* Get the number of bytes in the entry. */
185 #define NRINTERNAL							\
186 	BTLALIGN(sizeof(recno_t) + sizeof(pgno_t))
187 
188 /* Copy a RINTERNAL entry to the page. */
189 #define	WR_RINTERNAL(p, nrecs, pgno) do {				\
190 	*(recno_t *)(void *)p = nrecs;					\
191 	p += sizeof(recno_t);						\
192 	*(pgno_t *)(void *)p = pgno;					\
193 } while (/*CONSTCOND*/0)
194 
195 /* For the btree leaf pages, the item is a key and data pair. */
196 typedef struct _bleaf {
197 	uint32_t	ksize;		/* size of key */
198 	uint32_t	dsize;		/* size of data */
199 	uint8_t	flags;			/* P_BIGDATA, P_BIGKEY */
200 	char	bytes[1];		/* data */
201 } BLEAF;
202 
203 /* Get the page's BLEAF structure at index indx. */
204 #define	GETBLEAF(pg, indx)						\
205 	((BLEAF *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
206 
207 
208 /* Get the number of bytes in the user's key/data pair. */
209 #define _NBLEAFDBT(ksize, dsize)					\
210     BTLALIGN(sizeof(uint32_t) + sizeof(uint32_t) + sizeof(uint8_t) +	\
211 	    (ksize) + (dsize))
212 #ifdef _DIAGNOSTIC
213 static __inline uint32_t
NBLEAFDBT(size_t k,size_t d)214 NBLEAFDBT(size_t k, size_t d) {
215 	size_t x = _NBLEAFDBT(k, d);
216 	_DBFIT(x, uint32_t);
217 	return (uint32_t)x;
218 }
219 #else
220 #define NBLEAFDBT(p, q)	(uint32_t)_NBLEAFDBT(p, q)
221 #endif
222 
223 /* Get the number of bytes in the entry. */
224 #define NBLEAF(p)	NBLEAFDBT((p)->ksize, (p)->dsize)
225 
226 /* Copy a BLEAF entry to the page. */
227 #define	WR_BLEAF(p, key, data, flags) do {				\
228 	_DBFIT(key->size, uint32_t);					\
229 	*(uint32_t *)(void *)p = (uint32_t)key->size;			\
230 	p += sizeof(uint32_t);						\
231 	_DBFIT(data->size, uint32_t);					\
232 	*(uint32_t *)(void *)p = (uint32_t)data->size;			\
233 	p += sizeof(uint32_t);						\
234 	*(uint8_t *)(void *)p = flags;					\
235 	p += sizeof(uint8_t);						\
236 	(void)memmove(p, key->data, key->size);				\
237 	p += key->size;							\
238 	(void)memmove(p, data->data, data->size);			\
239 } while (/*CONSTCOND*/0)
240 
241 /* For the recno leaf pages, the item is a data entry. */
242 typedef struct _rleaf {
243 	uint32_t	dsize;		/* size of data */
244 	uint8_t	flags;			/* P_BIGDATA */
245 	char	bytes[1];
246 } RLEAF;
247 
248 /* Get the page's RLEAF structure at index indx. */
249 #define	GETRLEAF(pg, indx)						\
250 	((RLEAF *)(void *)((char *)(void *)(pg) + (pg)->linp[indx]))
251 
252 #define	_NRLEAFDBT(dsize)						\
253 	BTLALIGN(sizeof(uint32_t) + sizeof(uint8_t) + (dsize))
254 
255 #ifdef _DIAGNOSTIC
256 static __inline uint32_t
NRLEAFDBT(size_t d)257 NRLEAFDBT(size_t d) {
258 	size_t x = _NRLEAFDBT(d);
259 	_DBFIT(x, uint32_t);
260 	return (uint32_t)x;
261 }
262 #else
263 #define NRLEAFDBT(d)	(uint32_t)_NRLEAFDBT(d)
264 #endif
265 
266 /* Get the number of bytes in the entry. */
267 #define NRLEAF(p)	NRLEAFDBT((p)->dsize)
268 
269 /* Get the number of bytes from the user's data. */
270 
271 /* Copy a RLEAF entry to the page. */
272 #define	WR_RLEAF(p, data, flags) do {					\
273 	_DBFIT(data->size, uint32_t);					\
274 	*(uint32_t *)(void *)p = (uint32_t)data->size;			\
275 	p += sizeof(uint32_t);						\
276 	*(uint8_t *)(void *)p = flags;					\
277 	p += sizeof(uint8_t);						\
278 	memmove(p, data->data, data->size);				\
279 } while (/*CONSTCOND*/0)
280 
281 /*
282  * A record in the tree is either a pointer to a page and an index in the page
283  * or a page number and an index.  These structures are used as a cursor, stack
284  * entry and search returns as well as to pass records to other routines.
285  *
286  * One comment about searches.  Internal page searches must find the largest
287  * record less than key in the tree so that descents work.  Leaf page searches
288  * must find the smallest record greater than key so that the returned index
289  * is the record's correct position for insertion.
290  */
291 typedef struct _epgno {
292 	pgno_t	pgno;			/* the page number */
293 	indx_t	index;			/* the index on the page */
294 } EPGNO;
295 
296 typedef struct _epg {
297 	PAGE	*page;			/* the (pinned) page */
298 	indx_t	 index;			/* the index on the page */
299 } EPG;
300 
301 /*
302  * About cursors.  The cursor (and the page that contained the key/data pair
303  * that it referenced) can be deleted, which makes things a bit tricky.  If
304  * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set
305  * or there simply aren't any duplicates of the key) we copy the key that it
306  * referenced when it's deleted, and reacquire a new cursor key if the cursor
307  * is used again.  If there are duplicates keys, we move to the next/previous
308  * key, and set a flag so that we know what happened.  NOTE: if duplicate (to
309  * the cursor) keys are added to the tree during this process, it is undefined
310  * if they will be returned or not in a cursor scan.
311  *
312  * The flags determine the possible states of the cursor:
313  *
314  * CURS_INIT	The cursor references *something*.
315  * CURS_ACQUIRE	The cursor was deleted, and a key has been saved so that
316  *		we can reacquire the right position in the tree.
317  * CURS_AFTER, CURS_BEFORE
318  *		The cursor was deleted, and now references a key/data pair
319  *		that has not yet been returned, either before or after the
320  *		deleted key/data pair.
321  * XXX
322  * This structure is broken out so that we can eventually offer multiple
323  * cursors as part of the DB interface.
324  */
325 typedef struct _cursor {
326 	EPGNO	 pg;			/* B: Saved tree reference. */
327 	DBT	 key;			/* B: Saved key, or key.data == NULL. */
328 	recno_t	 rcursor;		/* R: recno cursor (1-based) */
329 
330 #define	CURS_ACQUIRE	0x01		/*  B: Cursor needs to be reacquired. */
331 #define	CURS_AFTER	0x02		/*  B: Unreturned cursor after key. */
332 #define	CURS_BEFORE	0x04		/*  B: Unreturned cursor before key. */
333 #define	CURS_INIT	0x08		/* RB: Cursor initialized. */
334 	uint8_t flags;
335 } CURSOR;
336 
337 /*
338  * The metadata of the tree.  The nrecs field is used only by the RECNO code.
339  * This is because the btree doesn't really need it and it requires that every
340  * put or delete call modify the metadata.
341  */
342 typedef struct _btmeta {
343 	uint32_t	magic;		/* magic number */
344 	uint32_t	version;	/* version */
345 	uint32_t	psize;		/* page size */
346 	uint32_t	free;		/* page number of first free page */
347 	uint32_t	nrecs;		/* R: number of records */
348 
349 #define	SAVEMETA	(B_NODUPS | R_RECNO)
350 	uint32_t	flags;		/* bt_flags & SAVEMETA */
351 } BTMETA;
352 
353 /* The in-memory btree/recno data structure. */
354 typedef struct _btree {
355 	MPOOL	 *bt_mp;		/* memory pool cookie */
356 
357 	DB	 *bt_dbp;		/* pointer to enclosing DB */
358 
359 	EPG	  bt_cur;		/* current (pinned) page */
360 	PAGE	 *bt_pinned;		/* page pinned across calls */
361 
362 	CURSOR	  bt_cursor;		/* cursor */
363 
364 #define	BT_PUSH(t, p, i) {						\
365 	t->bt_sp->pgno = p; 						\
366 	t->bt_sp->index = i; 						\
367 	++t->bt_sp;							\
368 }
369 #define	BT_POP(t)	(t->bt_sp == t->bt_stack ? NULL : --t->bt_sp)
370 #define	BT_CLR(t)	(t->bt_sp = t->bt_stack)
371 	EPGNO	  bt_stack[50];		/* stack of parent pages */
372 	EPGNO	 *bt_sp;		/* current stack pointer */
373 
374 	DBT	  bt_rkey;		/* returned key */
375 	DBT	  bt_rdata;		/* returned data */
376 
377 	int	  bt_fd;		/* tree file descriptor */
378 
379 	pgno_t	  bt_free;		/* next free page */
380 	uint32_t bt_psize;		/* page size */
381 	indx_t	  bt_ovflsize;		/* cut-off for key/data overflow */
382 	int	  bt_lorder;		/* byte order */
383 					/* sorted order */
384 	enum { NOT, BACK, FORWARD } bt_order;
385 	EPGNO	  bt_last;		/* last insert */
386 
387 					/* B: key comparison function */
388 	int	(*bt_cmp)(const DBT *, const DBT *);
389 					/* B: prefix comparison function */
390 	size_t	(*bt_pfx)(const DBT *, const DBT *);
391 					/* R: recno input function */
392 	int	(*bt_irec)(struct _btree *, recno_t);
393 
394 	FILE	 *bt_rfp;		/* R: record FILE pointer */
395 	int	  bt_rfd;		/* R: record file descriptor */
396 
397 	caddr_t	  bt_cmap;		/* R: current point in mapped space */
398 	caddr_t	  bt_smap;		/* R: start of mapped space */
399 	caddr_t   bt_emap;		/* R: end of mapped space */
400 	size_t	  bt_msize;		/* R: size of mapped region. */
401 
402 	recno_t	  bt_nrecs;		/* R: number of records */
403 	size_t	  bt_reclen;		/* R: fixed record length */
404 	uint8_t	  bt_bval;		/* R: delimiting byte/pad character */
405 
406 /*
407  * NB:
408  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
409  */
410 #define	B_INMEM		0x00001		/* in-memory tree */
411 #define	B_METADIRTY	0x00002		/* need to write metadata */
412 #define	B_MODIFIED	0x00004		/* tree modified */
413 #define	B_NEEDSWAP	0x00008		/* if byte order requires swapping */
414 #define	B_RDONLY	0x00010		/* read-only tree */
415 
416 #define	B_NODUPS	0x00020		/* no duplicate keys permitted */
417 #define	R_RECNO		0x00080		/* record oriented tree */
418 
419 #define	R_CLOSEFP	0x00040		/* opened a file pointer */
420 #define	R_EOF		0x00100		/* end of input file reached. */
421 #define	R_FIXLEN	0x00200		/* fixed length records */
422 #define	R_MEMMAPPED	0x00400		/* memory mapped file. */
423 #define	R_INMEM		0x00800		/* in-memory file */
424 #define	R_MODIFIED	0x01000		/* modified file */
425 #define	R_RDONLY	0x02000		/* read-only file */
426 
427 #define	B_DB_LOCK	0x04000		/* DB_LOCK specified. */
428 #define	B_DB_SHMEM	0x08000		/* DB_SHMEM specified. */
429 #define	B_DB_TXN	0x10000		/* DB_TXN specified. */
430 	uint32_t flags;
431 } BTREE;
432 
433 #include "extern.h"
434