bt_put.c revision 1.9 1 1.9 christos /* $NetBSD: bt_put.c,v 1.9 1997/07/13 18:51:55 christos Exp $ */
2 1.7 cgd
3 1.1 cgd /*-
4 1.6 cgd * Copyright (c) 1990, 1993, 1994
5 1.1 cgd * The Regents of the University of California. All rights reserved.
6 1.1 cgd *
7 1.1 cgd * This code is derived from software contributed to Berkeley by
8 1.1 cgd * Mike Olson.
9 1.1 cgd *
10 1.1 cgd * Redistribution and use in source and binary forms, with or without
11 1.1 cgd * modification, are permitted provided that the following conditions
12 1.1 cgd * are met:
13 1.1 cgd * 1. Redistributions of source code must retain the above copyright
14 1.1 cgd * notice, this list of conditions and the following disclaimer.
15 1.1 cgd * 2. Redistributions in binary form must reproduce the above copyright
16 1.1 cgd * notice, this list of conditions and the following disclaimer in the
17 1.1 cgd * documentation and/or other materials provided with the distribution.
18 1.1 cgd * 3. All advertising materials mentioning features or use of this software
19 1.1 cgd * must display the following acknowledgement:
20 1.1 cgd * This product includes software developed by the University of
21 1.1 cgd * California, Berkeley and its contributors.
22 1.1 cgd * 4. Neither the name of the University nor the names of its contributors
23 1.1 cgd * may be used to endorse or promote products derived from this software
24 1.1 cgd * without specific prior written permission.
25 1.1 cgd *
26 1.1 cgd * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 1.1 cgd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 1.1 cgd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 1.1 cgd * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 1.1 cgd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 1.1 cgd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 1.1 cgd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 1.1 cgd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 1.1 cgd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 1.1 cgd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 1.1 cgd * SUCH DAMAGE.
37 1.1 cgd */
38 1.1 cgd
39 1.9 christos #include <sys/cdefs.h>
40 1.1 cgd #if defined(LIBC_SCCS) && !defined(lint)
41 1.7 cgd #if 0
42 1.8 cgd static char sccsid[] = "@(#)bt_put.c 8.8 (Berkeley) 7/26/94";
43 1.7 cgd #else
44 1.9 christos __RCSID("$NetBSD: bt_put.c,v 1.9 1997/07/13 18:51:55 christos Exp $");
45 1.7 cgd #endif
46 1.1 cgd #endif /* LIBC_SCCS and not lint */
47 1.1 cgd
48 1.1 cgd #include <sys/types.h>
49 1.1 cgd
50 1.1 cgd #include <errno.h>
51 1.1 cgd #include <stdio.h>
52 1.1 cgd #include <stdlib.h>
53 1.1 cgd #include <string.h>
54 1.1 cgd
55 1.1 cgd #include <db.h>
56 1.1 cgd #include "btree.h"
57 1.1 cgd
58 1.1 cgd static EPG *bt_fast __P((BTREE *, const DBT *, const DBT *, int *));
59 1.1 cgd
60 1.1 cgd /*
61 1.1 cgd * __BT_PUT -- Add a btree item to the tree.
62 1.1 cgd *
63 1.1 cgd * Parameters:
64 1.1 cgd * dbp: pointer to access method
65 1.1 cgd * key: key
66 1.1 cgd * data: data
67 1.1 cgd * flag: R_NOOVERWRITE
68 1.1 cgd *
69 1.1 cgd * Returns:
70 1.1 cgd * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is already in the
71 1.1 cgd * tree and R_NOOVERWRITE specified.
72 1.1 cgd */
73 1.1 cgd int
74 1.1 cgd __bt_put(dbp, key, data, flags)
75 1.1 cgd const DB *dbp;
76 1.1 cgd DBT *key;
77 1.1 cgd const DBT *data;
78 1.1 cgd u_int flags;
79 1.1 cgd {
80 1.1 cgd BTREE *t;
81 1.1 cgd DBT tkey, tdata;
82 1.9 christos EPG *e = NULL; /* pacify gcc */
83 1.1 cgd PAGE *h;
84 1.1 cgd indx_t index, nxtindex;
85 1.1 cgd pgno_t pg;
86 1.6 cgd u_int32_t nbytes;
87 1.1 cgd int dflags, exact, status;
88 1.1 cgd char *dest, db[NOVFLSIZE], kb[NOVFLSIZE];
89 1.1 cgd
90 1.1 cgd t = dbp->internal;
91 1.4 cgd
92 1.4 cgd /* Toss any page pinned across calls. */
93 1.4 cgd if (t->bt_pinned != NULL) {
94 1.4 cgd mpool_put(t->bt_mp, t->bt_pinned, 0);
95 1.4 cgd t->bt_pinned = NULL;
96 1.4 cgd }
97 1.1 cgd
98 1.8 cgd /* Check for change to a read-only tree. */
99 1.8 cgd if (F_ISSET(t, B_RDONLY)) {
100 1.8 cgd errno = EPERM;
101 1.8 cgd return (RET_ERROR);
102 1.8 cgd }
103 1.8 cgd
104 1.1 cgd switch (flags) {
105 1.1 cgd case 0:
106 1.1 cgd case R_NOOVERWRITE:
107 1.1 cgd break;
108 1.8 cgd case R_CURSOR:
109 1.8 cgd /*
110 1.8 cgd * If flags is R_CURSOR, put the cursor. Must already
111 1.8 cgd * have started a scan and not have already deleted it.
112 1.8 cgd */
113 1.8 cgd if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
114 1.8 cgd !F_ISSET(&t->bt_cursor,
115 1.8 cgd CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
116 1.8 cgd break;
117 1.8 cgd /* FALLTHROUGH */
118 1.1 cgd default:
119 1.8 cgd errno = EINVAL;
120 1.1 cgd return (RET_ERROR);
121 1.1 cgd }
122 1.1 cgd
123 1.1 cgd /*
124 1.8 cgd * If the key/data pair won't fit on a page, store it on overflow
125 1.8 cgd * pages. Only put the key on the overflow page if the pair are
126 1.8 cgd * still too big after moving the data to an overflow page.
127 1.1 cgd *
128 1.1 cgd * XXX
129 1.8 cgd * If the insert fails later on, the overflow pages aren't recovered.
130 1.1 cgd */
131 1.1 cgd dflags = 0;
132 1.1 cgd if (key->size + data->size > t->bt_ovflsize) {
133 1.1 cgd if (key->size > t->bt_ovflsize) {
134 1.1 cgd storekey: if (__ovfl_put(t, key, &pg) == RET_ERROR)
135 1.1 cgd return (RET_ERROR);
136 1.1 cgd tkey.data = kb;
137 1.1 cgd tkey.size = NOVFLSIZE;
138 1.1 cgd memmove(kb, &pg, sizeof(pgno_t));
139 1.1 cgd memmove(kb + sizeof(pgno_t),
140 1.6 cgd &key->size, sizeof(u_int32_t));
141 1.1 cgd dflags |= P_BIGKEY;
142 1.1 cgd key = &tkey;
143 1.1 cgd }
144 1.1 cgd if (key->size + data->size > t->bt_ovflsize) {
145 1.1 cgd if (__ovfl_put(t, data, &pg) == RET_ERROR)
146 1.1 cgd return (RET_ERROR);
147 1.1 cgd tdata.data = db;
148 1.1 cgd tdata.size = NOVFLSIZE;
149 1.1 cgd memmove(db, &pg, sizeof(pgno_t));
150 1.1 cgd memmove(db + sizeof(pgno_t),
151 1.6 cgd &data->size, sizeof(u_int32_t));
152 1.1 cgd dflags |= P_BIGDATA;
153 1.1 cgd data = &tdata;
154 1.1 cgd }
155 1.1 cgd if (key->size + data->size > t->bt_ovflsize)
156 1.1 cgd goto storekey;
157 1.1 cgd }
158 1.1 cgd
159 1.1 cgd /* Replace the cursor. */
160 1.1 cgd if (flags == R_CURSOR) {
161 1.8 cgd if ((h = mpool_get(t->bt_mp, t->bt_cursor.pg.pgno, 0)) == NULL)
162 1.1 cgd return (RET_ERROR);
163 1.8 cgd index = t->bt_cursor.pg.index;
164 1.1 cgd goto delete;
165 1.1 cgd }
166 1.1 cgd
167 1.1 cgd /*
168 1.8 cgd * Find the key to delete, or, the location at which to insert.
169 1.8 cgd * Bt_fast and __bt_search both pin the returned page.
170 1.1 cgd */
171 1.1 cgd if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL)
172 1.1 cgd if ((e = __bt_search(t, key, &exact)) == NULL)
173 1.1 cgd return (RET_ERROR);
174 1.1 cgd h = e->page;
175 1.1 cgd index = e->index;
176 1.1 cgd
177 1.1 cgd /*
178 1.8 cgd * Add the key/data pair to the tree. If an identical key is already
179 1.8 cgd * in the tree, and R_NOOVERWRITE is set, an error is returned. If
180 1.8 cgd * R_NOOVERWRITE is not set, the key is either added (if duplicates are
181 1.8 cgd * permitted) or an error is returned.
182 1.1 cgd */
183 1.1 cgd switch (flags) {
184 1.1 cgd case R_NOOVERWRITE:
185 1.1 cgd if (!exact)
186 1.1 cgd break;
187 1.1 cgd mpool_put(t->bt_mp, h, 0);
188 1.1 cgd return (RET_SPECIAL);
189 1.1 cgd default:
190 1.8 cgd if (!exact || !F_ISSET(t, B_NODUPS))
191 1.1 cgd break;
192 1.8 cgd /*
193 1.8 cgd * !!!
194 1.8 cgd * Note, the delete may empty the page, so we need to put a
195 1.8 cgd * new entry into the page immediately.
196 1.8 cgd */
197 1.8 cgd delete: if (__bt_dleaf(t, key, h, index) == RET_ERROR) {
198 1.1 cgd mpool_put(t->bt_mp, h, 0);
199 1.1 cgd return (RET_ERROR);
200 1.1 cgd }
201 1.1 cgd break;
202 1.1 cgd }
203 1.1 cgd
204 1.1 cgd /*
205 1.1 cgd * If not enough room, or the user has put a ceiling on the number of
206 1.1 cgd * keys permitted in the page, split the page. The split code will
207 1.1 cgd * insert the key and data and unpin the current page. If inserting
208 1.1 cgd * into the offset array, shift the pointers up.
209 1.1 cgd */
210 1.1 cgd nbytes = NBLEAFDBT(key->size, data->size);
211 1.1 cgd if (h->upper - h->lower < nbytes + sizeof(indx_t)) {
212 1.1 cgd if ((status = __bt_split(t, h, key,
213 1.1 cgd data, dflags, nbytes, index)) != RET_SUCCESS)
214 1.1 cgd return (status);
215 1.1 cgd goto success;
216 1.1 cgd }
217 1.1 cgd
218 1.1 cgd if (index < (nxtindex = NEXTINDEX(h)))
219 1.1 cgd memmove(h->linp + index + 1, h->linp + index,
220 1.1 cgd (nxtindex - index) * sizeof(indx_t));
221 1.1 cgd h->lower += sizeof(indx_t);
222 1.1 cgd
223 1.1 cgd h->linp[index] = h->upper -= nbytes;
224 1.1 cgd dest = (char *)h + h->upper;
225 1.1 cgd WR_BLEAF(dest, key, data, dflags);
226 1.1 cgd
227 1.8 cgd /* If the cursor is on this page, adjust it as necessary. */
228 1.8 cgd if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
229 1.8 cgd !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
230 1.8 cgd t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index >= index)
231 1.8 cgd ++t->bt_cursor.pg.index;
232 1.8 cgd
233 1.1 cgd if (t->bt_order == NOT)
234 1.1 cgd if (h->nextpg == P_INVALID) {
235 1.1 cgd if (index == NEXTINDEX(h) - 1) {
236 1.1 cgd t->bt_order = FORWARD;
237 1.1 cgd t->bt_last.index = index;
238 1.1 cgd t->bt_last.pgno = h->pgno;
239 1.1 cgd }
240 1.1 cgd } else if (h->prevpg == P_INVALID) {
241 1.1 cgd if (index == 0) {
242 1.1 cgd t->bt_order = BACK;
243 1.1 cgd t->bt_last.index = 0;
244 1.1 cgd t->bt_last.pgno = h->pgno;
245 1.1 cgd }
246 1.1 cgd }
247 1.1 cgd
248 1.1 cgd mpool_put(t->bt_mp, h, MPOOL_DIRTY);
249 1.1 cgd
250 1.1 cgd success:
251 1.8 cgd if (flags == R_SETCURSOR)
252 1.8 cgd __bt_setcur(t, e->page->pgno, e->index);
253 1.8 cgd
254 1.8 cgd F_SET(t, B_MODIFIED);
255 1.1 cgd return (RET_SUCCESS);
256 1.1 cgd }
257 1.1 cgd
258 1.1 cgd #ifdef STATISTICS
259 1.1 cgd u_long bt_cache_hit, bt_cache_miss;
260 1.1 cgd #endif
261 1.1 cgd
262 1.1 cgd /*
263 1.1 cgd * BT_FAST -- Do a quick check for sorted data.
264 1.1 cgd *
265 1.1 cgd * Parameters:
266 1.1 cgd * t: tree
267 1.1 cgd * key: key to insert
268 1.1 cgd *
269 1.1 cgd * Returns:
270 1.1 cgd * EPG for new record or NULL if not found.
271 1.1 cgd */
272 1.1 cgd static EPG *
273 1.1 cgd bt_fast(t, key, data, exactp)
274 1.1 cgd BTREE *t;
275 1.1 cgd const DBT *key, *data;
276 1.1 cgd int *exactp;
277 1.1 cgd {
278 1.1 cgd PAGE *h;
279 1.6 cgd u_int32_t nbytes;
280 1.1 cgd int cmp;
281 1.1 cgd
282 1.1 cgd if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) {
283 1.1 cgd t->bt_order = NOT;
284 1.1 cgd return (NULL);
285 1.1 cgd }
286 1.5 cgd t->bt_cur.page = h;
287 1.5 cgd t->bt_cur.index = t->bt_last.index;
288 1.1 cgd
289 1.1 cgd /*
290 1.8 cgd * If won't fit in this page or have too many keys in this page,
291 1.8 cgd * have to search to get split stack.
292 1.1 cgd */
293 1.1 cgd nbytes = NBLEAFDBT(key->size, data->size);
294 1.1 cgd if (h->upper - h->lower < nbytes + sizeof(indx_t))
295 1.1 cgd goto miss;
296 1.1 cgd
297 1.1 cgd if (t->bt_order == FORWARD) {
298 1.5 cgd if (t->bt_cur.page->nextpg != P_INVALID)
299 1.1 cgd goto miss;
300 1.5 cgd if (t->bt_cur.index != NEXTINDEX(h) - 1)
301 1.1 cgd goto miss;
302 1.5 cgd if ((cmp = __bt_cmp(t, key, &t->bt_cur)) < 0)
303 1.1 cgd goto miss;
304 1.5 cgd t->bt_last.index = cmp ? ++t->bt_cur.index : t->bt_cur.index;
305 1.1 cgd } else {
306 1.5 cgd if (t->bt_cur.page->prevpg != P_INVALID)
307 1.1 cgd goto miss;
308 1.5 cgd if (t->bt_cur.index != 0)
309 1.1 cgd goto miss;
310 1.5 cgd if ((cmp = __bt_cmp(t, key, &t->bt_cur)) > 0)
311 1.1 cgd goto miss;
312 1.1 cgd t->bt_last.index = 0;
313 1.1 cgd }
314 1.1 cgd *exactp = cmp == 0;
315 1.1 cgd #ifdef STATISTICS
316 1.1 cgd ++bt_cache_hit;
317 1.1 cgd #endif
318 1.5 cgd return (&t->bt_cur);
319 1.1 cgd
320 1.1 cgd miss:
321 1.1 cgd #ifdef STATISTICS
322 1.1 cgd ++bt_cache_miss;
323 1.1 cgd #endif
324 1.1 cgd t->bt_order = NOT;
325 1.1 cgd mpool_put(t->bt_mp, h, 0);
326 1.1 cgd return (NULL);
327 1.1 cgd }
328