xref: /freebsd/lib/libc/db/btree/bt_put.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
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  * 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.
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, R_SETCURSOR, R_CURSOR
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 	case R_SETCURSOR:
95 		break;
96 	case R_CURSOR:
97 		/*
98 		 * If flags is R_CURSOR, put the cursor.  Must already
99 		 * have started a scan and not have already deleted it.
100 		 */
101 		if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
102 		    !F_ISSET(&t->bt_cursor,
103 			CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
104 			break;
105 		/* FALLTHROUGH */
106 	default:
107 		errno = EINVAL;
108 		return (RET_ERROR);
109 	}
110 
111 	/*
112 	 * If the key/data pair won't fit on a page, store it on overflow
113 	 * pages.  Only put the key on the overflow page if the pair are
114 	 * still too big after moving the data to an overflow page.
115 	 *
116 	 * XXX
117 	 * If the insert fails later on, the overflow pages aren't recovered.
118 	 */
119 	dflags = 0;
120 	if (key->size + data->size > t->bt_ovflsize) {
121 		if (key->size > t->bt_ovflsize) {
122 storekey:		if (__ovfl_put(t, key, &pg) == RET_ERROR)
123 				return (RET_ERROR);
124 			tkey.data = kb;
125 			tkey.size = NOVFLSIZE;
126 			memmove(kb, &pg, sizeof(pgno_t));
127 			tmp = key->size;
128 			memmove(kb + sizeof(pgno_t),
129 			    &tmp, sizeof(u_int32_t));
130 			dflags |= P_BIGKEY;
131 			key = &tkey;
132 		}
133 		if (key->size + data->size > t->bt_ovflsize) {
134 			if (__ovfl_put(t, data, &pg) == RET_ERROR)
135 				return (RET_ERROR);
136 			tdata.data = db;
137 			tdata.size = NOVFLSIZE;
138 			memmove(db, &pg, sizeof(pgno_t));
139 			tmp = data->size;
140 			memmove(db + sizeof(pgno_t),
141 			    &tmp, sizeof(u_int32_t));
142 			dflags |= P_BIGDATA;
143 			data = &tdata;
144 		}
145 		if (key->size + data->size > t->bt_ovflsize)
146 			goto storekey;
147 	}
148 
149 	/* Replace the cursor. */
150 	if (flags == R_CURSOR) {
151 		if ((h = mpool_get(t->bt_mp, t->bt_cursor.pg.pgno, 0)) == NULL)
152 			return (RET_ERROR);
153 		idx = t->bt_cursor.pg.index;
154 		goto delete;
155 	}
156 
157 	/*
158 	 * Find the key to delete, or, the location at which to insert.
159 	 * Bt_fast and __bt_search both pin the returned page.
160 	 */
161 	if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL)
162 		if ((e = __bt_search(t, key, &exact)) == NULL)
163 			return (RET_ERROR);
164 	h = e->page;
165 	idx = e->index;
166 
167 	/*
168 	 * Add the key/data pair to the tree.  If an identical key is already
169 	 * in the tree, and R_NOOVERWRITE is set, an error is returned.  If
170 	 * R_NOOVERWRITE is not set, the key is either added (if duplicates are
171 	 * permitted) or an error is returned.
172 	 */
173 	switch (flags) {
174 	case R_NOOVERWRITE:
175 		if (!exact)
176 			break;
177 		mpool_put(t->bt_mp, h, 0);
178 		return (RET_SPECIAL);
179 	default:
180 		if (!exact || !F_ISSET(t, B_NODUPS))
181 			break;
182 		/*
183 		 * !!!
184 		 * Note, the delete may empty the page, so we need to put a
185 		 * new entry into the page immediately.
186 		 */
187 delete:		if (__bt_dleaf(t, key, h, idx) == RET_ERROR) {
188 			mpool_put(t->bt_mp, h, 0);
189 			return (RET_ERROR);
190 		}
191 		break;
192 	}
193 
194 	/*
195 	 * If not enough room, or the user has put a ceiling on the number of
196 	 * keys permitted in the page, split the page.  The split code will
197 	 * insert the key and data and unpin the current page.  If inserting
198 	 * into the offset array, shift the pointers up.
199 	 */
200 	nbytes = NBLEAFDBT(key->size, data->size);
201 	if ((u_int32_t)(h->upper - h->lower) < nbytes + sizeof(indx_t)) {
202 		if ((status = __bt_split(t, h, key,
203 		    data, dflags, nbytes, idx)) != RET_SUCCESS)
204 			return (status);
205 		goto success;
206 	}
207 
208 	if (idx < (nxtindex = NEXTINDEX(h)))
209 		memmove(h->linp + idx + 1, h->linp + idx,
210 		    (nxtindex - idx) * sizeof(indx_t));
211 	h->lower += sizeof(indx_t);
212 
213 	h->linp[idx] = h->upper -= nbytes;
214 	dest = (char *)h + h->upper;
215 	WR_BLEAF(dest, key, data, dflags);
216 
217 	/* If the cursor is on this page, adjust it as necessary. */
218 	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
219 	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
220 	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index >= idx)
221 		++t->bt_cursor.pg.index;
222 
223 	if (t->bt_order == NOT) {
224 		if (h->nextpg == P_INVALID) {
225 			if (idx == NEXTINDEX(h) - 1) {
226 				t->bt_order = FORWARD;
227 				t->bt_last.index = idx;
228 				t->bt_last.pgno = h->pgno;
229 			}
230 		} else if (h->prevpg == P_INVALID) {
231 			if (idx == 0) {
232 				t->bt_order = BACK;
233 				t->bt_last.index = 0;
234 				t->bt_last.pgno = h->pgno;
235 			}
236 		}
237 	}
238 
239 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
240 
241 success:
242 	if (flags == R_SETCURSOR)
243 		__bt_setcur(t, e->page->pgno, e->index);
244 
245 	F_SET(t, B_MODIFIED);
246 	return (RET_SUCCESS);
247 }
248 
249 #ifdef STATISTICS
250 u_long bt_cache_hit, bt_cache_miss;
251 #endif
252 
253 /*
254  * BT_FAST -- Do a quick check for sorted data.
255  *
256  * Parameters:
257  *	t:	tree
258  *	key:	key to insert
259  *
260  * Returns:
261  *	EPG for new record or NULL if not found.
262  */
263 static EPG *
264 bt_fast(BTREE *t, const DBT *key, const DBT *data, int *exactp)
265 {
266 	PAGE *h;
267 	u_int32_t nbytes;
268 	int cmp;
269 
270 	if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) {
271 		t->bt_order = NOT;
272 		return (NULL);
273 	}
274 	t->bt_cur.page = h;
275 	t->bt_cur.index = t->bt_last.index;
276 
277 	/*
278 	 * If won't fit in this page or have too many keys in this page,
279 	 * have to search to get split stack.
280 	 */
281 	nbytes = NBLEAFDBT(key->size, data->size);
282 	if ((u_int32_t)(h->upper - h->lower) < nbytes + sizeof(indx_t))
283 		goto miss;
284 
285 	if (t->bt_order == FORWARD) {
286 		if (t->bt_cur.page->nextpg != P_INVALID)
287 			goto miss;
288 		if (t->bt_cur.index != NEXTINDEX(h) - 1)
289 			goto miss;
290 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) < 0)
291 			goto miss;
292 		t->bt_last.index = cmp ? ++t->bt_cur.index : t->bt_cur.index;
293 	} else {
294 		if (t->bt_cur.page->prevpg != P_INVALID)
295 			goto miss;
296 		if (t->bt_cur.index != 0)
297 			goto miss;
298 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) > 0)
299 			goto miss;
300 		t->bt_last.index = 0;
301 	}
302 	*exactp = cmp == 0;
303 #ifdef STATISTICS
304 	++bt_cache_hit;
305 #endif
306 	return (&t->bt_cur);
307 
308 miss:
309 #ifdef STATISTICS
310 	++bt_cache_miss;
311 #endif
312 	t->bt_order = NOT;
313 	mpool_put(t->bt_mp, h, 0);
314 	return (NULL);
315 }
316