xref: /freebsd/lib/libc/db/btree/bt_put.c (revision 7661de35d15f582ab33e3bd6b8d909601557e436)
1 /*-
2  * Copyright (c) 1990, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Mike Olson.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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  * 4. 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.
19  *
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
30  * SUCH DAMAGE.
31  */
32 
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char sccsid[] = "@(#)bt_put.c	8.8 (Berkeley) 7/26/94";
35 #endif /* LIBC_SCCS and not lint */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/types.h>
40 
41 #include <errno.h>
42 #include <stdio.h>
43 #include <stdlib.h>
44 #include <string.h>
45 
46 #include <db.h>
47 #include "btree.h"
48 
49 static EPG *bt_fast(BTREE *, const DBT *, const DBT *, int *);
50 
51 /*
52  * __BT_PUT -- Add a btree item to the tree.
53  *
54  * Parameters:
55  *	dbp:	pointer to access method
56  *	key:	key
57  *	data:	data
58  *	flag:	R_NOOVERWRITE
59  *
60  * Returns:
61  *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is already in the
62  *	tree and R_NOOVERWRITE specified.
63  */
64 int
65 __bt_put(const DB *dbp, DBT *key, const DBT *data, u_int flags)
66 {
67 	BTREE *t;
68 	DBT tkey, tdata;
69 	EPG *e;
70 	PAGE *h;
71 	indx_t idx, nxtindex;
72 	pgno_t pg;
73 	u_int32_t nbytes, tmp;
74 	int dflags, exact, status;
75 	char *dest, db[NOVFLSIZE], kb[NOVFLSIZE];
76 
77 	t = dbp->internal;
78 
79 	/* Toss any page pinned across calls. */
80 	if (t->bt_pinned != NULL) {
81 		mpool_put(t->bt_mp, t->bt_pinned, 0);
82 		t->bt_pinned = NULL;
83 	}
84 
85 	/* Check for change to a read-only tree. */
86 	if (F_ISSET(t, B_RDONLY)) {
87 		errno = EPERM;
88 		return (RET_ERROR);
89 	}
90 
91 	switch (flags) {
92 	case 0:
93 	case R_NOOVERWRITE:
94 		break;
95 	case R_CURSOR:
96 		/*
97 		 * If flags is R_CURSOR, put the cursor.  Must already
98 		 * have started a scan and not have already deleted it.
99 		 */
100 		if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
101 		    !F_ISSET(&t->bt_cursor,
102 			CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
103 			break;
104 		/* FALLTHROUGH */
105 	default:
106 		errno = EINVAL;
107 		return (RET_ERROR);
108 	}
109 
110 	/*
111 	 * If the key/data pair won't fit on a page, store it on overflow
112 	 * pages.  Only put the key on the overflow page if the pair are
113 	 * still too big after moving the data to an overflow page.
114 	 *
115 	 * XXX
116 	 * If the insert fails later on, the overflow pages aren't recovered.
117 	 */
118 	dflags = 0;
119 	if (key->size + data->size > t->bt_ovflsize) {
120 		if (key->size > t->bt_ovflsize) {
121 storekey:		if (__ovfl_put(t, key, &pg) == RET_ERROR)
122 				return (RET_ERROR);
123 			tkey.data = kb;
124 			tkey.size = NOVFLSIZE;
125 			memmove(kb, &pg, sizeof(pgno_t));
126 			tmp = key->size;
127 			memmove(kb + sizeof(pgno_t),
128 			    &tmp, sizeof(u_int32_t));
129 			dflags |= P_BIGKEY;
130 			key = &tkey;
131 		}
132 		if (key->size + data->size > t->bt_ovflsize) {
133 			if (__ovfl_put(t, data, &pg) == RET_ERROR)
134 				return (RET_ERROR);
135 			tdata.data = db;
136 			tdata.size = NOVFLSIZE;
137 			memmove(db, &pg, sizeof(pgno_t));
138 			tmp = data->size;
139 			memmove(db + sizeof(pgno_t),
140 			    &tmp, sizeof(u_int32_t));
141 			dflags |= P_BIGDATA;
142 			data = &tdata;
143 		}
144 		if (key->size + data->size > t->bt_ovflsize)
145 			goto storekey;
146 	}
147 
148 	/* Replace the cursor. */
149 	if (flags == R_CURSOR) {
150 		if ((h = mpool_get(t->bt_mp, t->bt_cursor.pg.pgno, 0)) == NULL)
151 			return (RET_ERROR);
152 		idx = t->bt_cursor.pg.index;
153 		goto delete;
154 	}
155 
156 	/*
157 	 * Find the key to delete, or, the location at which to insert.
158 	 * Bt_fast and __bt_search both pin the returned page.
159 	 */
160 	if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL)
161 		if ((e = __bt_search(t, key, &exact)) == NULL)
162 			return (RET_ERROR);
163 	h = e->page;
164 	idx = e->index;
165 
166 	/*
167 	 * Add the key/data pair to the tree.  If an identical key is already
168 	 * in the tree, and R_NOOVERWRITE is set, an error is returned.  If
169 	 * R_NOOVERWRITE is not set, the key is either added (if duplicates are
170 	 * permitted) or an error is returned.
171 	 */
172 	switch (flags) {
173 	case R_NOOVERWRITE:
174 		if (!exact)
175 			break;
176 		mpool_put(t->bt_mp, h, 0);
177 		return (RET_SPECIAL);
178 	default:
179 		if (!exact || !F_ISSET(t, B_NODUPS))
180 			break;
181 		/*
182 		 * !!!
183 		 * Note, the delete may empty the page, so we need to put a
184 		 * new entry into the page immediately.
185 		 */
186 delete:		if (__bt_dleaf(t, key, h, idx) == RET_ERROR) {
187 			mpool_put(t->bt_mp, h, 0);
188 			return (RET_ERROR);
189 		}
190 		break;
191 	}
192 
193 	/*
194 	 * If not enough room, or the user has put a ceiling on the number of
195 	 * keys permitted in the page, split the page.  The split code will
196 	 * insert the key and data and unpin the current page.  If inserting
197 	 * into the offset array, shift the pointers up.
198 	 */
199 	nbytes = NBLEAFDBT(key->size, data->size);
200 	if ((u_int32_t)(h->upper - h->lower) < nbytes + sizeof(indx_t)) {
201 		if ((status = __bt_split(t, h, key,
202 		    data, dflags, nbytes, idx)) != RET_SUCCESS)
203 			return (status);
204 		goto success;
205 	}
206 
207 	if (idx < (nxtindex = NEXTINDEX(h)))
208 		memmove(h->linp + idx + 1, h->linp + idx,
209 		    (nxtindex - idx) * sizeof(indx_t));
210 	h->lower += sizeof(indx_t);
211 
212 	h->linp[idx] = h->upper -= nbytes;
213 	dest = (char *)h + h->upper;
214 	WR_BLEAF(dest, key, data, dflags);
215 
216 	/* If the cursor is on this page, adjust it as necessary. */
217 	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
218 	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
219 	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index >= idx)
220 		++t->bt_cursor.pg.index;
221 
222 	if (t->bt_order == NOT) {
223 		if (h->nextpg == P_INVALID) {
224 			if (idx == NEXTINDEX(h) - 1) {
225 				t->bt_order = FORWARD;
226 				t->bt_last.index = idx;
227 				t->bt_last.pgno = h->pgno;
228 			}
229 		} else if (h->prevpg == P_INVALID) {
230 			if (idx == 0) {
231 				t->bt_order = BACK;
232 				t->bt_last.index = 0;
233 				t->bt_last.pgno = h->pgno;
234 			}
235 		}
236 	}
237 
238 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
239 
240 success:
241 	if (flags == R_SETCURSOR)
242 		__bt_setcur(t, e->page->pgno, e->index);
243 
244 	F_SET(t, B_MODIFIED);
245 	return (RET_SUCCESS);
246 }
247 
248 #ifdef STATISTICS
249 u_long bt_cache_hit, bt_cache_miss;
250 #endif
251 
252 /*
253  * BT_FAST -- Do a quick check for sorted data.
254  *
255  * Parameters:
256  *	t:	tree
257  *	key:	key to insert
258  *
259  * Returns:
260  *	EPG for new record or NULL if not found.
261  */
262 static EPG *
263 bt_fast(BTREE *t, const DBT *key, const DBT *data, int *exactp)
264 {
265 	PAGE *h;
266 	u_int32_t nbytes;
267 	int cmp;
268 
269 	if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) {
270 		t->bt_order = NOT;
271 		return (NULL);
272 	}
273 	t->bt_cur.page = h;
274 	t->bt_cur.index = t->bt_last.index;
275 
276 	/*
277 	 * If won't fit in this page or have too many keys in this page,
278 	 * have to search to get split stack.
279 	 */
280 	nbytes = NBLEAFDBT(key->size, data->size);
281 	if ((u_int32_t)(h->upper - h->lower) < nbytes + sizeof(indx_t))
282 		goto miss;
283 
284 	if (t->bt_order == FORWARD) {
285 		if (t->bt_cur.page->nextpg != P_INVALID)
286 			goto miss;
287 		if (t->bt_cur.index != NEXTINDEX(h) - 1)
288 			goto miss;
289 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) < 0)
290 			goto miss;
291 		t->bt_last.index = cmp ? ++t->bt_cur.index : t->bt_cur.index;
292 	} else {
293 		if (t->bt_cur.page->prevpg != P_INVALID)
294 			goto miss;
295 		if (t->bt_cur.index != 0)
296 			goto miss;
297 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) > 0)
298 			goto miss;
299 		t->bt_last.index = 0;
300 	}
301 	*exactp = cmp == 0;
302 #ifdef STATISTICS
303 	++bt_cache_hit;
304 #endif
305 	return (&t->bt_cur);
306 
307 miss:
308 #ifdef STATISTICS
309 	++bt_cache_miss;
310 #endif
311 	t->bt_order = NOT;
312 	mpool_put(t->bt_mp, h, 0);
313 	return (NULL);
314 }
315