bt_delete.c revision 1.5 1 1.1 cgd /*-
2 1.5 cgd * Copyright (c) 1990, 1993, 1994
3 1.1 cgd * The Regents of the University of California. All rights reserved.
4 1.1 cgd *
5 1.1 cgd * This code is derived from software contributed to Berkeley by
6 1.1 cgd * Mike Olson.
7 1.1 cgd *
8 1.1 cgd * Redistribution and use in source and binary forms, with or without
9 1.1 cgd * modification, are permitted provided that the following conditions
10 1.1 cgd * are met:
11 1.1 cgd * 1. Redistributions of source code must retain the above copyright
12 1.1 cgd * notice, this list of conditions and the following disclaimer.
13 1.1 cgd * 2. Redistributions in binary form must reproduce the above copyright
14 1.1 cgd * notice, this list of conditions and the following disclaimer in the
15 1.1 cgd * documentation and/or other materials provided with the distribution.
16 1.1 cgd * 3. All advertising materials mentioning features or use of this software
17 1.1 cgd * must display the following acknowledgement:
18 1.1 cgd * This product includes software developed by the University of
19 1.1 cgd * California, Berkeley and its contributors.
20 1.1 cgd * 4. Neither the name of the University nor the names of its contributors
21 1.1 cgd * may be used to endorse or promote products derived from this software
22 1.1 cgd * without specific prior written permission.
23 1.1 cgd *
24 1.1 cgd * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 1.1 cgd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 1.1 cgd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 1.1 cgd * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 1.1 cgd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 1.1 cgd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 1.1 cgd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 1.1 cgd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 1.1 cgd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 1.1 cgd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 1.1 cgd * SUCH DAMAGE.
35 1.1 cgd */
36 1.1 cgd
37 1.1 cgd #if defined(LIBC_SCCS) && !defined(lint)
38 1.5 cgd static char sccsid[] = "@(#)bt_delete.c 8.4 (Berkeley) 5/31/94";
39 1.1 cgd #endif /* LIBC_SCCS and not lint */
40 1.1 cgd
41 1.1 cgd #include <sys/types.h>
42 1.1 cgd
43 1.1 cgd #include <errno.h>
44 1.1 cgd #include <stdio.h>
45 1.1 cgd #include <string.h>
46 1.1 cgd
47 1.1 cgd #include <db.h>
48 1.1 cgd #include "btree.h"
49 1.1 cgd
50 1.1 cgd static int bt_bdelete __P((BTREE *, const DBT *));
51 1.1 cgd
52 1.1 cgd /*
53 1.1 cgd * __BT_DELETE -- Delete the item(s) referenced by a key.
54 1.1 cgd *
55 1.1 cgd * Parameters:
56 1.1 cgd * dbp: pointer to access method
57 1.1 cgd * key: key to delete
58 1.1 cgd * flags: R_CURSOR if deleting what the cursor references
59 1.1 cgd *
60 1.1 cgd * Returns:
61 1.1 cgd * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
62 1.1 cgd */
63 1.1 cgd int
64 1.1 cgd __bt_delete(dbp, key, flags)
65 1.1 cgd const DB *dbp;
66 1.1 cgd const DBT *key;
67 1.1 cgd u_int flags;
68 1.1 cgd {
69 1.1 cgd BTREE *t;
70 1.1 cgd int status;
71 1.1 cgd
72 1.1 cgd t = dbp->internal;
73 1.4 cgd
74 1.4 cgd /* Toss any page pinned across calls. */
75 1.4 cgd if (t->bt_pinned != NULL) {
76 1.4 cgd mpool_put(t->bt_mp, t->bt_pinned, 0);
77 1.4 cgd t->bt_pinned = NULL;
78 1.4 cgd }
79 1.4 cgd
80 1.1 cgd if (ISSET(t, B_RDONLY)) {
81 1.1 cgd errno = EPERM;
82 1.1 cgd return (RET_ERROR);
83 1.1 cgd }
84 1.4 cgd
85 1.1 cgd switch(flags) {
86 1.1 cgd case 0:
87 1.1 cgd status = bt_bdelete(t, key);
88 1.1 cgd break;
89 1.1 cgd case R_CURSOR:
90 1.1 cgd /*
91 1.1 cgd * If flags is R_CURSOR, delete the cursor; must already have
92 1.1 cgd * started a scan and not have already deleted the record. For
93 1.1 cgd * the delete cursor bit to have been set requires that the
94 1.1 cgd * scan be initialized, so no reason to check.
95 1.1 cgd */
96 1.1 cgd if (!ISSET(t, B_SEQINIT))
97 1.1 cgd goto einval;
98 1.1 cgd status = ISSET(t, B_DELCRSR) ?
99 1.1 cgd RET_SPECIAL : __bt_crsrdel(t, &t->bt_bcursor);
100 1.1 cgd break;
101 1.1 cgd default:
102 1.1 cgd einval: errno = EINVAL;
103 1.1 cgd return (RET_ERROR);
104 1.1 cgd }
105 1.1 cgd if (status == RET_SUCCESS)
106 1.1 cgd SET(t, B_MODIFIED);
107 1.1 cgd return (status);
108 1.1 cgd }
109 1.1 cgd
110 1.1 cgd /*
111 1.1 cgd * BT_BDELETE -- Delete all key/data pairs matching the specified key.
112 1.1 cgd *
113 1.1 cgd * Parameters:
114 1.1 cgd * tree: tree
115 1.1 cgd * key: key to delete
116 1.1 cgd *
117 1.1 cgd * Returns:
118 1.1 cgd * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
119 1.1 cgd */
120 1.1 cgd static int
121 1.1 cgd bt_bdelete(t, key)
122 1.1 cgd BTREE *t;
123 1.1 cgd const DBT *key;
124 1.1 cgd {
125 1.1 cgd EPG *e, save;
126 1.1 cgd PAGE *h;
127 1.1 cgd pgno_t cpgno, pg;
128 1.1 cgd indx_t cindex;
129 1.1 cgd int deleted, dirty1, dirty2, exact;
130 1.1 cgd
131 1.1 cgd /* Find any matching record; __bt_search pins the page. */
132 1.1 cgd if ((e = __bt_search(t, key, &exact)) == NULL)
133 1.1 cgd return (RET_ERROR);
134 1.1 cgd if (!exact) {
135 1.1 cgd mpool_put(t->bt_mp, e->page, 0);
136 1.1 cgd return (RET_SPECIAL);
137 1.1 cgd }
138 1.1 cgd
139 1.1 cgd /*
140 1.1 cgd * Delete forward, then delete backward, from the found key. The
141 1.1 cgd * ordering is so that the deletions don't mess up the page refs.
142 1.1 cgd * The first loop deletes the key from the original page, the second
143 1.1 cgd * unpins the original page. In the first loop, dirty1 is set if
144 1.1 cgd * the original page is modified, and dirty2 is set if any subsequent
145 1.1 cgd * pages are modified. In the second loop, dirty1 starts off set if
146 1.1 cgd * the original page has been modified, and is set if any subsequent
147 1.1 cgd * pages are modified.
148 1.1 cgd *
149 1.1 cgd * If find the key referenced by the cursor, don't delete it, just
150 1.1 cgd * flag it for future deletion. The cursor page number is P_INVALID
151 1.1 cgd * unless the sequential scan is initialized, so no reason to check.
152 1.1 cgd * A special case is when the already deleted cursor record was the
153 1.1 cgd * only record found. If so, then the delete opertion fails as no
154 1.1 cgd * records were deleted.
155 1.1 cgd *
156 1.1 cgd * Cycle in place in the current page until the current record doesn't
157 1.1 cgd * match the key or the page is empty. If the latter, walk forward,
158 1.1 cgd * skipping empty pages and repeating until a record doesn't match
159 1.1 cgd * the key or the end of the tree is reached.
160 1.1 cgd */
161 1.1 cgd cpgno = t->bt_bcursor.pgno;
162 1.1 cgd cindex = t->bt_bcursor.index;
163 1.1 cgd save = *e;
164 1.1 cgd dirty1 = 0;
165 1.1 cgd for (h = e->page, deleted = 0;;) {
166 1.1 cgd dirty2 = 0;
167 1.1 cgd do {
168 1.1 cgd if (h->pgno == cpgno && e->index == cindex) {
169 1.1 cgd if (!ISSET(t, B_DELCRSR)) {
170 1.1 cgd SET(t, B_DELCRSR);
171 1.1 cgd deleted = 1;
172 1.1 cgd }
173 1.1 cgd ++e->index;
174 1.1 cgd } else {
175 1.1 cgd if (__bt_dleaf(t, h, e->index)) {
176 1.1 cgd if (h->pgno != save.page->pgno)
177 1.1 cgd mpool_put(t->bt_mp, h, dirty2);
178 1.1 cgd mpool_put(t->bt_mp, save.page, dirty1);
179 1.1 cgd return (RET_ERROR);
180 1.1 cgd }
181 1.1 cgd if (h->pgno == save.page->pgno)
182 1.1 cgd dirty1 = MPOOL_DIRTY;
183 1.1 cgd else
184 1.1 cgd dirty2 = MPOOL_DIRTY;
185 1.1 cgd deleted = 1;
186 1.1 cgd }
187 1.1 cgd } while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0);
188 1.1 cgd
189 1.1 cgd /*
190 1.1 cgd * Quit if didn't find a match, no next page, or first key on
191 1.1 cgd * the next page doesn't match. Don't unpin the original page
192 1.1 cgd * unless an error occurs.
193 1.1 cgd */
194 1.1 cgd if (e->index < NEXTINDEX(h))
195 1.1 cgd break;
196 1.1 cgd for (;;) {
197 1.1 cgd if ((pg = h->nextpg) == P_INVALID)
198 1.1 cgd goto done1;
199 1.1 cgd if (h->pgno != save.page->pgno)
200 1.1 cgd mpool_put(t->bt_mp, h, dirty2);
201 1.1 cgd if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) {
202 1.1 cgd mpool_put(t->bt_mp, save.page, dirty1);
203 1.1 cgd return (RET_ERROR);
204 1.1 cgd }
205 1.1 cgd if (NEXTINDEX(h) != 0) {
206 1.1 cgd e->page = h;
207 1.1 cgd e->index = 0;
208 1.1 cgd break;
209 1.1 cgd }
210 1.1 cgd }
211 1.1 cgd
212 1.1 cgd if (__bt_cmp(t, key, e) != 0)
213 1.1 cgd break;
214 1.1 cgd }
215 1.1 cgd
216 1.1 cgd /*
217 1.1 cgd * Reach here with the original page and the last page referenced
218 1.1 cgd * pinned (they may be the same). Release it if not the original.
219 1.1 cgd */
220 1.1 cgd done1: if (h->pgno != save.page->pgno)
221 1.1 cgd mpool_put(t->bt_mp, h, dirty2);
222 1.1 cgd
223 1.1 cgd /*
224 1.1 cgd * Walk backwards from the record previous to the record returned by
225 1.1 cgd * __bt_search, skipping empty pages, until a record doesn't match
226 1.1 cgd * the key or reach the beginning of the tree.
227 1.1 cgd */
228 1.1 cgd *e = save;
229 1.1 cgd for (;;) {
230 1.1 cgd if (e->index)
231 1.1 cgd --e->index;
232 1.1 cgd for (h = e->page; e->index; --e->index) {
233 1.1 cgd if (__bt_cmp(t, key, e) != 0)
234 1.1 cgd goto done2;
235 1.1 cgd if (h->pgno == cpgno && e->index == cindex) {
236 1.1 cgd if (!ISSET(t, B_DELCRSR)) {
237 1.1 cgd SET(t, B_DELCRSR);
238 1.1 cgd deleted = 1;
239 1.1 cgd }
240 1.1 cgd } else {
241 1.1 cgd if (__bt_dleaf(t, h, e->index) == RET_ERROR) {
242 1.1 cgd mpool_put(t->bt_mp, h, dirty1);
243 1.1 cgd return (RET_ERROR);
244 1.1 cgd }
245 1.1 cgd if (h->pgno == save.page->pgno)
246 1.1 cgd dirty1 = MPOOL_DIRTY;
247 1.1 cgd deleted = 1;
248 1.1 cgd }
249 1.1 cgd }
250 1.1 cgd
251 1.1 cgd if ((pg = h->prevpg) == P_INVALID)
252 1.1 cgd goto done2;
253 1.1 cgd mpool_put(t->bt_mp, h, dirty1);
254 1.1 cgd dirty1 = 0;
255 1.1 cgd if ((e->page = mpool_get(t->bt_mp, pg, 0)) == NULL)
256 1.1 cgd return (RET_ERROR);
257 1.1 cgd e->index = NEXTINDEX(e->page);
258 1.1 cgd }
259 1.1 cgd
260 1.1 cgd /*
261 1.1 cgd * Reach here with the last page that was looked at pinned. Release
262 1.1 cgd * it.
263 1.1 cgd */
264 1.1 cgd done2: mpool_put(t->bt_mp, h, dirty1);
265 1.1 cgd return (deleted ? RET_SUCCESS : RET_SPECIAL);
266 1.1 cgd }
267 1.1 cgd
268 1.1 cgd /*
269 1.1 cgd * __BT_DLEAF -- Delete a single record from a leaf page.
270 1.1 cgd *
271 1.1 cgd * Parameters:
272 1.1 cgd * t: tree
273 1.1 cgd * index: index on current page to delete
274 1.1 cgd *
275 1.1 cgd * Returns:
276 1.1 cgd * RET_SUCCESS, RET_ERROR.
277 1.1 cgd */
278 1.1 cgd int
279 1.1 cgd __bt_dleaf(t, h, index)
280 1.1 cgd BTREE *t;
281 1.1 cgd PAGE *h;
282 1.5 cgd indx_t index;
283 1.1 cgd {
284 1.1 cgd register BLEAF *bl;
285 1.5 cgd register indx_t cnt, *ip, offset;
286 1.5 cgd register u_int32_t nbytes;
287 1.1 cgd char *from;
288 1.1 cgd void *to;
289 1.1 cgd
290 1.1 cgd /*
291 1.1 cgd * Delete a record from a btree leaf page. Internal records are never
292 1.1 cgd * deleted from internal pages, regardless of the records that caused
293 1.1 cgd * them to be added being deleted. Pages made empty by deletion are
294 1.1 cgd * not reclaimed. They are, however, made available for reuse.
295 1.1 cgd *
296 1.1 cgd * Pack the remaining entries at the end of the page, shift the indices
297 1.1 cgd * down, overwriting the deleted record and its index. If the record
298 1.1 cgd * uses overflow pages, make them available for reuse.
299 1.1 cgd */
300 1.1 cgd to = bl = GETBLEAF(h, index);
301 1.1 cgd if (bl->flags & P_BIGKEY && __ovfl_delete(t, bl->bytes) == RET_ERROR)
302 1.1 cgd return (RET_ERROR);
303 1.1 cgd if (bl->flags & P_BIGDATA &&
304 1.1 cgd __ovfl_delete(t, bl->bytes + bl->ksize) == RET_ERROR)
305 1.1 cgd return (RET_ERROR);
306 1.1 cgd nbytes = NBLEAF(bl);
307 1.1 cgd
308 1.1 cgd /*
309 1.1 cgd * Compress the key/data pairs. Compress and adjust the [BR]LEAF
310 1.1 cgd * offsets. Reset the headers.
311 1.1 cgd */
312 1.1 cgd from = (char *)h + h->upper;
313 1.1 cgd memmove(from + nbytes, from, (char *)to - from);
314 1.1 cgd h->upper += nbytes;
315 1.1 cgd
316 1.1 cgd offset = h->linp[index];
317 1.1 cgd for (cnt = index, ip = &h->linp[0]; cnt--; ++ip)
318 1.1 cgd if (ip[0] < offset)
319 1.1 cgd ip[0] += nbytes;
320 1.1 cgd for (cnt = NEXTINDEX(h) - index; --cnt; ++ip)
321 1.1 cgd ip[0] = ip[1] < offset ? ip[1] + nbytes : ip[1];
322 1.1 cgd h->lower -= sizeof(indx_t);
323 1.1 cgd return (RET_SUCCESS);
324 1.1 cgd }
325