2 * Copyright (c) 1991, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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 * 3. 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.
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
32 * @(#)btree.h 8.11 (Berkeley) 8/17/94
33 * $FreeBSD: head/lib/libc/db/btree/btree.h 189327 2009-03-04 00:58:04Z delphij $
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))
43 #define DEFMINKEYPAGE (2) /* Minimum keys per page */
44 #define MINCACHE (5) /* Minimum cached pages */
45 #define MINPSIZE (512) /* Minimum page size */
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.
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. */
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.
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 */
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 */
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 */
83 /* First and next index. */
85 (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \
86 sizeof(uint32_t) + sizeof(indx_t) + sizeof(indx_t))
87 #define NEXTINDEX(p) (((p)->lower - BTDATAOFF) / sizeof(indx_t))
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.
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
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.)
105 #define LALIGN(n) roundup2(n, sizeof(pgno_t))
106 #define NOVFLSIZE (sizeof(pgno_t) + sizeof(uint32_t))
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.
117 typedef struct _binternal {
118 uint32_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 */
123 char bytes[1]; /* data */
126 /* Get the page's BINTERNAL structure at index indx. */
127 #define GETBINTERNAL(pg, indx) \
128 ((BINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
130 /* Get the number of bytes in the entry. */
131 #define NBINTERNAL(len) \
132 LALIGN(sizeof(uint32_t) + sizeof(pgno_t) + \
133 sizeof(unsigned char) + (len))
135 /* Copy a BINTERNAL entry to the page. */
136 #define WR_BINTERNAL(p, size, pgno, flags) { \
137 *(uint32_t *)p = size; \
138 p += sizeof(uint32_t); \
139 *(pgno_t *)p = pgno; \
140 p += sizeof(pgno_t); \
141 *(unsigned char *)p = flags; \
142 p += sizeof(unsigned char); \
146 * For the recno internal pages, the item is a page number with the number of
147 * keys found on that page and below.
149 typedef struct _rinternal {
150 recno_t nrecs; /* number of records */
151 pgno_t pgno; /* page number stored below */
154 /* Get the page's RINTERNAL structure at index indx. */
155 #define GETRINTERNAL(pg, indx) \
156 ((RINTERNAL *)((char *)(pg) + (pg)->linp[indx]))
158 /* Get the number of bytes in the entry. */
160 LALIGN(sizeof(recno_t) + sizeof(pgno_t))
162 /* Copy a RINTERAL entry to the page. */
163 #define WR_RINTERNAL(p, nrecs, pgno) { \
164 *(recno_t *)p = nrecs; \
165 p += sizeof(recno_t); \
166 *(pgno_t *)p = pgno; \
169 /* For the btree leaf pages, the item is a key and data pair. */
170 typedef struct _bleaf {
171 uint32_t ksize; /* size of key */
172 uint32_t dsize; /* size of data */
173 unsigned char flags; /* P_BIGDATA, P_BIGKEY */
174 char bytes[1]; /* data */
177 /* Get the page's BLEAF structure at index indx. */
178 #define GETBLEAF(pg, indx) \
179 ((BLEAF *)((char *)(pg) + (pg)->linp[indx]))
181 /* Get the number of bytes in the entry. */
182 #define NBLEAF(p) NBLEAFDBT((p)->ksize, (p)->dsize)
184 /* Get the number of bytes in the user's key/data pair. */
185 #define NBLEAFDBT(ksize, dsize) \
186 LALIGN(sizeof(uint32_t) + sizeof(uint32_t) + \
187 sizeof(unsigned char) + (ksize) + (dsize))
189 /* Copy a BLEAF entry to the page. */
190 #define WR_BLEAF(p, key, data, flags) { \
191 *(uint32_t *)p = key->size; \
192 p += sizeof(uint32_t); \
193 *(uint32_t *)p = data->size; \
194 p += sizeof(uint32_t); \
195 *(unsigned char *)p = flags; \
196 p += sizeof(unsigned char); \
197 memmove(p, key->data, key->size); \
199 memmove(p, data->data, data->size); \
202 /* For the recno leaf pages, the item is a data entry. */
203 typedef struct _rleaf {
204 uint32_t dsize; /* size of data */
205 unsigned char flags; /* P_BIGDATA */
209 /* Get the page's RLEAF structure at index indx. */
210 #define GETRLEAF(pg, indx) \
211 ((RLEAF *)((char *)(pg) + (pg)->linp[indx]))
213 /* Get the number of bytes in the entry. */
214 #define NRLEAF(p) NRLEAFDBT((p)->dsize)
216 /* Get the number of bytes from the user's data. */
217 #define NRLEAFDBT(dsize) \
218 LALIGN(sizeof(uint32_t) + sizeof(unsigned char) + (dsize))
220 /* Copy a RLEAF entry to the page. */
221 #define WR_RLEAF(p, data, flags) { \
222 *(uint32_t *)p = data->size; \
223 p += sizeof(uint32_t); \
224 *(unsigned char *)p = flags; \
225 p += sizeof(unsigned char); \
226 memmove(p, data->data, data->size); \
230 * A record in the tree is either a pointer to a page and an index in the page
231 * or a page number and an index. These structures are used as a cursor, stack
232 * entry and search returns as well as to pass records to other routines.
234 * One comment about searches. Internal page searches must find the largest
235 * record less than key in the tree so that descents work. Leaf page searches
236 * must find the smallest record greater than key so that the returned index
237 * is the record's correct position for insertion.
239 typedef struct _epgno {
240 pgno_t pgno; /* the page number */
241 indx_t index; /* the index on the page */
244 typedef struct _epg {
245 PAGE *page; /* the (pinned) page */
246 indx_t index; /* the index on the page */
250 * About cursors. The cursor (and the page that contained the key/data pair
251 * that it referenced) can be deleted, which makes things a bit tricky. If
252 * there are no duplicates of the cursor key in the tree (i.e. B_NODUPS is set
253 * or there simply aren't any duplicates of the key) we copy the key that it
254 * referenced when it's deleted, and reacquire a new cursor key if the cursor
255 * is used again. If there are duplicates keys, we move to the next/previous
256 * key, and set a flag so that we know what happened. NOTE: if duplicate (to
257 * the cursor) keys are added to the tree during this process, it is undefined
258 * if they will be returned or not in a cursor scan.
260 * The flags determine the possible states of the cursor:
262 * CURS_INIT The cursor references *something*.
263 * CURS_ACQUIRE The cursor was deleted, and a key has been saved so that
264 * we can reacquire the right position in the tree.
265 * CURS_AFTER, CURS_BEFORE
266 * The cursor was deleted, and now references a key/data pair
267 * that has not yet been returned, either before or after the
268 * deleted key/data pair.
270 * This structure is broken out so that we can eventually offer multiple
271 * cursors as part of the DB interface.
273 typedef struct _cursor {
274 EPGNO pg; /* B: Saved tree reference. */
275 DBT key; /* B: Saved key, or key.data == NULL. */
276 recno_t rcursor; /* R: recno cursor (1-based) */
278 #define CURS_ACQUIRE 0x01 /* B: Cursor needs to be reacquired. */
279 #define CURS_AFTER 0x02 /* B: Unreturned cursor after key. */
280 #define CURS_BEFORE 0x04 /* B: Unreturned cursor before key. */
281 #define CURS_INIT 0x08 /* RB: Cursor initialized. */
286 * The metadata of the tree. The nrecs field is used only by the RECNO code.
287 * This is because the btree doesn't really need it and it requires that every
288 * put or delete call modify the metadata.
290 typedef struct _btmeta {
291 uint32_t magic; /* magic number */
292 uint32_t version; /* version */
293 uint32_t psize; /* page size */
294 uint32_t free; /* page number of first free page */
295 uint32_t nrecs; /* R: number of records */
297 #define SAVEMETA (B_NODUPS | R_RECNO)
298 uint32_t flags; /* bt_flags & SAVEMETA */
301 /* The in-memory btree/recno data structure. */
302 typedef struct _btree {
303 MPOOL *bt_mp; /* memory pool cookie */
305 DB *bt_dbp; /* pointer to enclosing DB */
307 EPG bt_cur; /* current (pinned) page */
308 PAGE *bt_pinned; /* page pinned across calls */
310 CURSOR bt_cursor; /* cursor */
312 #define BT_PUSH(t, p, i) { \
313 t->bt_sp->pgno = p; \
314 t->bt_sp->index = i; \
317 #define BT_POP(t) (t->bt_sp == t->bt_stack ? NULL : --t->bt_sp)
318 #define BT_CLR(t) (t->bt_sp = t->bt_stack)
319 EPGNO bt_stack[50]; /* stack of parent pages */
320 EPGNO *bt_sp; /* current stack pointer */
322 DBT bt_rkey; /* returned key */
323 DBT bt_rdata; /* returned data */
325 int bt_fd; /* tree file descriptor */
327 pgno_t bt_free; /* next free page */
328 uint32_t bt_psize; /* page size */
329 indx_t bt_ovflsize; /* cut-off for key/data overflow */
330 int bt_lorder; /* byte order */
332 enum { NOT, BACK, FORWARD } bt_order;
333 EPGNO bt_last; /* last insert */
335 /* B: key comparison function */
336 int (*bt_cmp)(const DBT *, const DBT *);
337 /* B: prefix comparison function */
338 size_t (*bt_pfx)(const DBT *, const DBT *);
339 /* R: recno input function */
340 int (*bt_irec)(struct _btree *, recno_t);
342 FILE *bt_rfp; /* R: record FILE pointer */
343 int bt_rfd; /* R: record file descriptor */
345 caddr_t bt_cmap; /* R: current point in mapped space */
346 caddr_t bt_smap; /* R: start of mapped space */
347 caddr_t bt_emap; /* R: end of mapped space */
348 size_t bt_msize; /* R: size of mapped region. */
350 recno_t bt_nrecs; /* R: number of records */
351 size_t bt_reclen; /* R: fixed record length */
352 unsigned char bt_bval; /* R: delimiting byte/pad character */
356 * B_NODUPS and R_RECNO are stored on disk, and may not be changed.
358 #define B_INMEM 0x00001 /* in-memory tree */
359 #define B_METADIRTY 0x00002 /* need to write metadata */
360 #define B_MODIFIED 0x00004 /* tree modified */
361 #define B_NEEDSWAP 0x00008 /* if byte order requires swapping */
362 #define B_RDONLY 0x00010 /* read-only tree */
364 #define B_NODUPS 0x00020 /* no duplicate keys permitted */
365 #define R_RECNO 0x00080 /* record oriented tree */
367 #define R_CLOSEFP 0x00040 /* opened a file pointer */
368 #define R_EOF 0x00100 /* end of input file reached. */
369 #define R_FIXLEN 0x00200 /* fixed length records */
370 #define R_MEMMAPPED 0x00400 /* memory mapped file. */
371 #define R_INMEM 0x00800 /* in-memory file */
372 #define R_MODIFIED 0x01000 /* modified file */
373 #define R_RDONLY 0x02000 /* read-only file */
375 #define B_DB_LOCK 0x04000 /* DB_LOCK specified. */
376 #define B_DB_SHMEM 0x08000 /* DB_SHMEM specified. */
377 #define B_DB_TXN 0x10000 /* DB_TXN specified. */