btree.h revision 1.8 1 /* $NetBSD: btree.h,v 1.8 1995/02/27 13:21:08 cgd 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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)btree.h 8.6 (Berkeley) 5/31/94
39 */
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 (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) \
105 (((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 * One comment about cursors. The cursor key is never removed from the tree,
239 * even if deleted. This is because it is quite difficult to decide where the
240 * cursor should be when other keys have been inserted/deleted in the tree;
241 * duplicate keys make it impossible. This scheme does require extra work
242 * though, to make sure that we don't perform an operation on a deleted key.
243 */
244 typedef struct _epgno {
245 pgno_t pgno; /* the page number */
246 indx_t index; /* the index on the page */
247 } EPGNO;
248
249 typedef struct _epg {
250 PAGE *page; /* the (pinned) page */
251 indx_t index; /* the index on the page */
252 } EPG;
253
254 /*
255 * The metadata of the tree. The m_nrecs field is used only by the RECNO code.
256 * This is because the btree doesn't really need it and it requires that every
257 * put or delete call modify the metadata.
258 */
259 typedef struct _btmeta {
260 u_int32_t m_magic; /* magic number */
261 u_int32_t m_version; /* version */
262 u_int32_t m_psize; /* page size */
263 u_int32_t m_free; /* page number of first free page */
264 u_int32_t m_nrecs; /* R: number of records */
265 #define SAVEMETA (B_NODUPS | R_RECNO)
266 u_int32_t m_flags; /* bt_flags & SAVEMETA */
267 u_int32_t m_unused; /* unused */
268 } BTMETA;
269
270 /* The in-memory btree/recno data structure. */
271 typedef struct _btree {
272 MPOOL *bt_mp; /* memory pool cookie */
273
274 DB *bt_dbp; /* pointer to enclosing DB */
275
276 EPG bt_cur; /* current (pinned) page */
277 PAGE *bt_pinned; /* page pinned across calls */
278
279 EPGNO bt_bcursor; /* B: btree cursor */
280 recno_t bt_rcursor; /* R: recno cursor (1-based) */
281
282 #define BT_POP(t) (t->bt_sp ? t->bt_stack + --t->bt_sp : NULL)
283 #define BT_CLR(t) (t->bt_sp = 0)
284 EPGNO *bt_stack; /* stack of parent pages */
285 u_int bt_sp; /* current stack pointer */
286 u_int bt_maxstack; /* largest stack */
287
288 char *bt_kbuf; /* key buffer */
289 size_t bt_kbufsz; /* key buffer size */
290 char *bt_dbuf; /* data buffer */
291 size_t bt_dbufsz; /* data buffer size */
292
293 int bt_fd; /* tree file descriptor */
294
295 pgno_t bt_free; /* next free page */
296 u_int32_t bt_psize; /* page size */
297 indx_t bt_ovflsize; /* cut-off for key/data overflow */
298 int bt_lorder; /* byte order */
299 /* sorted order */
300 enum { NOT, BACK, FORWARD } bt_order;
301 EPGNO bt_last; /* last insert */
302
303 /* B: key comparison function */
304 int (*bt_cmp) __P((const DBT *, const DBT *));
305 /* B: prefix comparison function */
306 size_t (*bt_pfx) __P((const DBT *, const DBT *));
307 /* R: recno input function */
308 int (*bt_irec) __P((struct _btree *, recno_t));
309
310 FILE *bt_rfp; /* R: record FILE pointer */
311 int bt_rfd; /* R: record file descriptor */
312
313 caddr_t bt_cmap; /* R: current point in mapped space */
314 caddr_t bt_smap; /* R: start of mapped space */
315 caddr_t bt_emap; /* R: end of mapped space */
316 size_t bt_msize; /* R: size of mapped region. */
317
318 recno_t bt_nrecs; /* R: number of records */
319 size_t bt_reclen; /* R: fixed record length */
320 u_char bt_bval; /* R: delimiting byte/pad character */
321
322 /*
323 * NB:
324 * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
325 */
326 #define B_DELCRSR 0x00001 /* cursor has been deleted */
327 #define B_INMEM 0x00002 /* in-memory tree */
328 #define B_METADIRTY 0x00004 /* need to write metadata */
329 #define B_MODIFIED 0x00008 /* tree modified */
330 #define B_NEEDSWAP 0x00010 /* if byte order requires swapping */
331 #define B_NODUPS 0x00020 /* no duplicate keys permitted */
332 #define B_RDONLY 0x00040 /* read-only tree */
333 #define R_RECNO 0x00080 /* record oriented tree */
334 #define B_SEQINIT 0x00100 /* sequential scan initialized */
335
336 #define R_CLOSEFP 0x00200 /* opened a file pointer */
337 #define R_EOF 0x00400 /* end of input file reached. */
338 #define R_FIXLEN 0x00800 /* fixed length records */
339 #define R_MEMMAPPED 0x01000 /* memory mapped file. */
340 #define R_INMEM 0x02000 /* in-memory file */
341 #define R_MODIFIED 0x04000 /* modified file */
342 #define R_RDONLY 0x08000 /* read-only file */
343
344 #define B_DB_LOCK 0x10000 /* DB_LOCK specified. */
345 #define B_DB_SHMEM 0x20000 /* DB_SHMEM specified. */
346 #define B_DB_TXN 0x40000 /* DB_TXN specified. */
347
348 u_int32_t bt_flags; /* btree state */
349 } BTREE;
350
351 #define SET(t, f) ((t)->bt_flags |= (f))
352 #define CLR(t, f) ((t)->bt_flags &= ~(f))
353 #define ISSET(t, f) ((t)->bt_flags & (f))
354
355 #include "extern.h"
356