1 /*
2  * Copyright (c) 1997-2000 by Sun Microsystems, Inc.
3  * All rights reserved.
4  */
5 
6 #ifndef _KRB5_BTREE_H
7 #define	_KRB5_BTREE_H
8 
9 #pragma ident	"%Z%%M%	%I%	%E% SMI"
10 
11 #ifdef	__cplusplus
12 extern "C" {
13 #endif
14 
15 
16 /*-
17  * Copyright (c) 1991, 1993, 1994
18  *	The Regents of the University of California.  All rights reserved.
19  *
20  * This code is derived from software contributed to Berkeley by
21  * Mike Olson.
22  *
23  * Redistribution and use in source and binary forms, with or without
24  * modification, are permitted provided that the following conditions
25  * are met:
26  * 1. Redistributions of source code must retain the above copyright
27  *    notice, this list of conditions and the following disclaimer.
28  * 2. Redistributions in binary form must reproduce the above copyright
29  *    notice, this list of conditions and the following disclaimer in the
30  *    documentation and/or other materials provided with the distribution.
31  * 3. All advertising materials mentioning features or use of this software
32  *    must display the following acknowledgement:
33  *	This product includes software developed by the University of
34  *	California, Berkeley and its contributors.
35  * 4. Neither the name of the University nor the names of its contributors
36  *    may be used to endorse or promote products derived from this software
37  *    without specific prior written permission.
38  *
39  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
40  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
41  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
42  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
43  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
44  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
45  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
46  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
47  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
48  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
49  * SUCH DAMAGE.
50  *
51  *	@(#)btree.h	8.11 (Berkeley) 8/17/94
52  */
53 
54 /* Macros to set/clear/test flags. */
55 #define	F_SET(p, f)	(p)->flags |= (f)
56 #define	F_CLR(p, f)	(p)->flags &= ~(f)
57 #define	F_ISSET(p, f)	((p)->flags & (f))
58 
59 #include "mpool.h"
60 
61 #define	DEFMINKEYPAGE	(2)		/* Minimum keys per page */
62 #define	MINCACHE	(5)		/* Minimum cached pages */
63 #define	MINPSIZE	(512)		/* Minimum page size */
64 
65 /*
66  * Page 0 of a btree file contains a copy of the meta-data.  This page is also
67  * used as an out-of-band page, i.e. page pointers that point to nowhere point
68  * to page 0.  Page 1 is the root of the btree.
69  */
70 #define	P_INVALID	 0		/* Invalid tree page number. */
71 #define	P_META		 0		/* Tree metadata page number. */
72 #define	P_ROOT		 1		/* Tree root page number. */
73 
74 /*
75  * There are five page layouts in the btree: btree internal pages (BINTERNAL),
76  * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages
77  * (RLEAF) and overflow pages.  All five page types have a page header (PAGE).
78  * This implementation requires that values within structures NOT be padded.
79  * (ANSI C permits random padding.)  If your compiler pads randomly you'll have
80  * to do some work to get this package to run.
81  */
82 typedef struct _page {
83 	db_pgno_t	pgno;			/* this page's page number */
84 	db_pgno_t	prevpg;			/* left sibling */
85 	db_pgno_t	nextpg;			/* right sibling */
86 
87 #define	P_BINTERNAL	0x01		/* btree internal page */
88 #define	P_BLEAF		0x02		/* leaf page */
89 #define	P_OVERFLOW	0x04		/* overflow page */
90 #define	P_RINTERNAL	0x08		/* recno internal page */
91 #define	P_RLEAF		0x10		/* leaf page */
92 #define P_TYPE		0x1f		/* type mask */
93 #define	P_PRESERVE	0x20		/* never delete this chain of pages */
94 	u_int32_t flags;
95 
96 	indx_t	lower;			/* lower bound of free space on page */
97 	indx_t	upper;			/* upper bound of free space on page */
98 	indx_t	linp[1];		/* indx_t-aligned VAR. LENGTH DATA */
99 } PAGE;
100 
101 /* First and next index. */
102 #define	BTDATAOFF							\
103 	(sizeof(db_pgno_t) + sizeof(db_pgno_t) + sizeof(db_pgno_t) +		\
104 	    sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t))
105 #define	NEXTINDEX(p)	(((p)->lower - BTDATAOFF) / sizeof(indx_t))
106 
107 /*
108  * For pages other than overflow pages, there is an array of offsets into the
109  * rest of the page immediately following the page header.  Each offset is to
110  * an item which is unique to the type of page.  The h_lower offset is just
111  * past the last filled-in index.  The h_upper offset is the first item on the
112  * page.  Offsets are from the beginning of the page.
113  *
114  * If an item is too big to store on a single page, a flag is set and the item
115  * is a { page, size } pair such that the page is the first page of an overflow
116  * chain with size bytes of item.  Overflow pages are simply bytes without any
117  * external structure.
118  *
119  * The page number and size fields in the items are db_pgno_t-aligned so they can
120  * be manipulated without copying.  (This presumes that 32 bit items can be
121  * manipulated on this system.)
122  */
123 #define	LALIGN(n)	(((n) + sizeof(db_pgno_t) - 1) & ~(sizeof(db_pgno_t) - 1))
124 #define	NOVFLSIZE	(sizeof(db_pgno_t) + sizeof(u_int32_t))
125 
126 /*
127  * For the btree internal pages, the item is a key.  BINTERNALs are {key, pgno}
128  * pairs, such that the key compares less than or equal to all of the records
129  * on that page.  For a tree without duplicate keys, an internal page with two
130  * consecutive keys, a and b, will have all records greater than or equal to a
131  * and less than b stored on the page associated with a.  Duplicate keys are
132  * somewhat special and can cause duplicate internal and leaf page records and
133  * some minor modifications of the above rule.
134  */
135 typedef struct _binternal {
136 	u_int32_t ksize;		/* key size */
137 	db_pgno_t	pgno;			/* page number stored on */
138 #define	P_BIGDATA	0x01		/* overflow data */
139 #define	P_BIGKEY	0x02		/* overflow key */
140 	u_char	flags;
141 	char	bytes[1];		/* data */
142 } BINTERNAL;
143 
144 /* Get the page's BINTERNAL structure at index indx. */
145 #define	GETBINTERNAL(pg, indx)						\
146 	((BINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
147 
148 /* Get the number of bytes in the entry. */
149 #define NBINTERNAL(len)							\
150 	LALIGN(sizeof(u_int32_t) + sizeof(db_pgno_t) + sizeof(u_char) + (len))
151 
152 /* Copy a BINTERNAL entry to the page. */
153 #define	WR_BINTERNAL(p, size, pgno, flags) {				\
154 	*(u_int32_t *)p = size;						\
155 	p += sizeof(u_int32_t);						\
156 	*(db_pgno_t *)p = pgno;						\
157 	p += sizeof(db_pgno_t);						\
158 	*(u_char *)p = flags;						\
159 	p += sizeof(u_char);						\
160 }
161 
162 /*
163  * For the recno internal pages, the item is a page number with the number of
164  * keys found on that page and below.
165  */
166 typedef struct _rinternal {
167 	recno_t	nrecs;			/* number of records */
168 	db_pgno_t	pgno;			/* page number stored below */
169 } RINTERNAL;
170 
171 /* Get the page's RINTERNAL structure at index indx. */
172 #define	GETRINTERNAL(pg, indx)						\
173 	((RINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
174 
175 /* Get the number of bytes in the entry. */
176 #define NRINTERNAL							\
177 	LALIGN(sizeof(recno_t) + sizeof(db_pgno_t))
178 
179 /* Copy a RINTERAL entry to the page. */
180 #define	WR_RINTERNAL(p, nrecs, pgno) {					\
181 	*(recno_t *)p = nrecs;						\
182 	p += sizeof(recno_t);						\
183 	*(db_pgno_t *)p = pgno;						\
184 }
185 
186 /* For the btree leaf pages, the item is a key and data pair. */
187 typedef struct _bleaf {
188 	u_int32_t	ksize;		/* size of key */
189 	u_int32_t	dsize;		/* size of data */
190 	u_char	flags;			/* P_BIGDATA, P_BIGKEY */
191 	char	bytes[1];		/* data */
192 } BLEAF;
193 
194 /* Get the page's BLEAF structure at index indx. */
195 #define	GETBLEAF(pg, indx)						\
196 	((BLEAF *)((char *)(pg) + (pg)->linp[indx]))
197 
198 /* Get the number of bytes in the entry. */
199 #define NBLEAF(p)	NBLEAFDBT((p)->ksize, (p)->dsize)
200 
201 /* Get the number of bytes in the user's key/data pair. */
202 #define NBLEAFDBT(ksize, dsize)						\
203 	LALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) +	\
204 	    (ksize) + (dsize))
205 
206 /* Copy a BLEAF entry to the page. */
207 #define	WR_BLEAF(p, key, data, flags) {					\
208 	*(u_int32_t *)p = key->size;					\
209 	p += sizeof(u_int32_t);						\
210 	*(u_int32_t *)p = data->size;					\
211 	p += sizeof(u_int32_t);						\
212 	*(u_char *)p = flags;						\
213 	p += sizeof(u_char);						\
214 	memmove(p, key->data, key->size);				\
215 	p += key->size;							\
216 	memmove(p, data->data, data->size);				\
217 }
218 
219 /* For the recno leaf pages, the item is a data entry. */
220 typedef struct _rleaf {
221 	u_int32_t	dsize;		/* size of data */
222 	u_char	flags;			/* P_BIGDATA */
223 	char	bytes[1];
224 } RLEAF;
225 
226 /* Get the page's RLEAF structure at index indx. */
227 #define	GETRLEAF(pg, indx)						\
228 	((RLEAF *)((char *)(pg) + (pg)->linp[indx]))
229 
230 /* Get the number of bytes in the entry. */
231 #define NRLEAF(p)	NRLEAFDBT((p)->dsize)
232 
233 /* Get the number of bytes from the user's data. */
234 #define	NRLEAFDBT(dsize)						\
235 	LALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize))
236 
237 /* Copy a RLEAF entry to the page. */
238 #define	WR_RLEAF(p, data, flags) {					\
239 	*(u_int32_t *)p = data->size;					\
240 	p += sizeof(u_int32_t);						\
241 	*(u_char *)p = flags;						\
242 	p += sizeof(u_char);						\
243 	memmove(p, data->data, data->size);				\
244 }
245 
246 /*
247  * A record in the tree is either a pointer to a page and an index in the page
248  * or a page number and an index.  These structures are used as a cursor, stack
249  * entry and search returns as well as to pass records to other routines.
250  *
251  * One comment about searches.  Internal page searches must find the largest
252  * record less than key in the tree so that descents work.  Leaf page searches
253  * must find the smallest record greater than key so that the returned index
254  * is the record's correct position for insertion.
255  */
256 typedef struct _epgno {
257 	db_pgno_t	pgno;			/* the page number */
258 	indx_t	index;			/* the index on the page */
259 } EPGNO;
260 
261 typedef struct _epg {
262 	PAGE	*page;			/* the (pinned) page */
263 	indx_t	 index;			/* the index on the page */
264 } EPG;
265 
266 /*
267  * About cursors.  The cursor (and the page that contained the key/data pair
268  * that it referenced) can be deleted, which makes things a bit tricky.  If
269  * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set
270  * or there simply aren't any duplicates of the key) we copy the key that it
271  * referenced when it's deleted, and reacquire a new cursor key if the cursor
272  * is used again.  If there are duplicates keys, we move to the next/previous
273  * key, and set a flag so that we know what happened.  NOTE: if duplicate (to
274  * the cursor) keys are added to the tree during this process, it is undefined
275  * if they will be returned or not in a cursor scan.
276  *
277  * The flags determine the possible states of the cursor:
278  *
279  * CURS_INIT	The cursor references *something*.
280  * CURS_ACQUIRE	The cursor was deleted, and a key has been saved so that
281  *		we can reacquire the right position in the tree.
282  * CURS_AFTER, CURS_BEFORE
283  *		The cursor was deleted, and now references a key/data pair
284  *		that has not yet been returned, either before or after the
285  *		deleted key/data pair.
286  * XXX
287  * This structure is broken out so that we can eventually offer multiple
288  * cursors as part of the DB interface.
289  */
290 typedef struct _cursor {
291 	EPGNO	 pg;			/* B: Saved tree reference. */
292 	DBT	 key;			/* B: Saved key, or key.data == NULL. */
293 	recno_t	 rcursor;		/* R: recno cursor (1-based) */
294 
295 #define	CURS_ACQUIRE	0x01		/*  B: Cursor needs to be reacquired. */
296 #define	CURS_AFTER	0x02		/*  B: Unreturned cursor after key. */
297 #define	CURS_BEFORE	0x04		/*  B: Unreturned cursor before key. */
298 #define	CURS_INIT	0x08		/* RB: Cursor initialized. */
299 	u_int8_t flags;
300 } CURSOR;
301 
302 /*
303  * The metadata of the tree.  The nrecs field is used only by the RECNO code.
304  * This is because the btree doesn't really need it and it requires that every
305  * put or delete call modify the metadata.
306  */
307 typedef struct _btmeta {
308 	u_int32_t	magic;		/* magic number */
309 	u_int32_t	version;	/* version */
310 	u_int32_t	psize;		/* page size */
311 	u_int32_t	free;		/* page number of first free page */
312 	u_int32_t	nrecs;		/* R: number of records */
313 
314 #define	SAVEMETA	(B_NODUPS | R_RECNO)
315 	u_int32_t	flags;		/* bt_flags & SAVEMETA */
316 } BTMETA;
317 
318 /* The in-memory btree/recno data structure. */
319 typedef struct _btree {
320 	MPOOL	 *bt_mp;		/* memory pool cookie */
321 
322 	DB	 *bt_dbp;		/* pointer to enclosing DB */
323 
324 	EPG	  bt_cur;		/* current (pinned) page */
325 	PAGE	 *bt_pinned;		/* page pinned across calls */
326 
327 	CURSOR	  bt_cursor;		/* cursor */
328 
329 #define	BT_PUSH(t, p, i) {						\
330 	t->bt_sp->pgno = p; 						\
331 	t->bt_sp->index = i; 						\
332 	++t->bt_sp;							\
333 }
334 #define	BT_POP(t)	(t->bt_sp == t->bt_stack ? NULL : --t->bt_sp)
335 #define	BT_CLR(t)	(t->bt_sp = t->bt_stack)
336 	EPGNO	  bt_stack[50];		/* stack of parent pages */
337 	EPGNO	 *bt_sp;		/* current stack pointer */
338 
339 	DBT	  bt_rkey;		/* returned key */
340 	DBT	  bt_rdata;		/* returned data */
341 
342 	int	  bt_fd;		/* tree file descriptor */
343 
344 	db_pgno_t	  bt_free;		/* next free page */
345 	u_int32_t bt_psize;		/* page size */
346 	indx_t	  bt_ovflsize;		/* cut-off for key/data overflow */
347 	int	  bt_lorder;		/* byte order */
348 					/* sorted order */
349 	enum { NOT, BACK, FORWARD } bt_order;
350 	EPGNO	  bt_last;		/* last insert */
351 
352 					/* B: key comparison function */
353 	int	(*bt_cmp) __P((const DBT *, const DBT *));
354 					/* B: prefix comparison function */
355 	size_t	(*bt_pfx) __P((const DBT *, const DBT *));
356 					/* R: recno input function */
357 	int	(*bt_irec) __P((struct _btree *, recno_t));
358 
359 	FILE	 *bt_rfp;		/* R: record FILE pointer */
360 	int	  bt_rfd;		/* R: record file descriptor */
361 
362 	caddr_t	  bt_cmap;		/* R: current point in mapped space */
363 	caddr_t	  bt_smap;		/* R: start of mapped space */
364 	caddr_t   bt_emap;		/* R: end of mapped space */
365 	size_t	  bt_msize;		/* R: size of mapped region. */
366 
367 	recno_t	  bt_nrecs;		/* R: number of records */
368 	size_t	  bt_reclen;		/* R: fixed record length */
369 	u_char	  bt_bval;		/* R: delimiting byte/pad character */
370 
371 /*
372  * NB:
373  * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
374  */
375 #define	B_INMEM		0x00001		/* in-memory tree */
376 #define	B_METADIRTY	0x00002		/* need to write metadata */
377 #define	B_MODIFIED	0x00004		/* tree modified */
378 #define	B_NEEDSWAP	0x00008		/* if byte order requires swapping */
379 #define	B_RDONLY	0x00010		/* read-only tree */
380 
381 #define	B_NODUPS	0x00020		/* no duplicate keys permitted */
382 #define	R_RECNO		0x00080		/* record oriented tree */
383 
384 #define	R_CLOSEFP	0x00040		/* opened a file pointer */
385 #define	R_EOF		0x00100		/* end of input file reached. */
386 #define	R_FIXLEN	0x00200		/* fixed length records */
387 #define	R_MEMMAPPED	0x00400		/* memory mapped file. */
388 #define	R_INMEM		0x00800		/* in-memory file */
389 #define	R_MODIFIED	0x01000		/* modified file */
390 #define	R_RDONLY	0x02000		/* read-only file */
391 
392 #define	B_DB_LOCK	0x04000		/* DB_LOCK specified. */
393 #define	B_DB_SHMEM	0x08000		/* DB_SHMEM specified. */
394 #define	B_DB_TXN	0x10000		/* DB_TXN specified. */
395 	u_int32_t flags;
396 } BTREE;
397 
398 #include "extern.h"
399 
400 #ifdef	__cplusplus
401 }
402 #endif
403 
404 #endif	/* !_KRB5_BTREE_H */
405