2 * Copyright (c) 1990, 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. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)bt_delete.c 8.13 (Berkeley) 7/28/94
39 #include <sys/types.h>
48 static int __bt_bdelete __P((BTREE *, const DBT *));
49 static int __bt_curdel __P((BTREE *, const DBT *, PAGE *, u_int));
50 static int __bt_pdelete __P((BTREE *, PAGE *));
51 static int __bt_relink __P((BTREE *, PAGE *));
52 static int __bt_stkacq __P((BTREE *, PAGE **, CURSOR *));
56 * Delete the item(s) referenced by a key.
58 * Return RET_SPECIAL if the key is not found.
61 __bt_delete(dbp, key, flags)
73 /* Toss any page pinned across calls. */
74 if (t->bt_pinned != NULL) {
75 mpool_put(t->bt_mp, t->bt_pinned, 0);
79 /* Check for change to a read-only tree. */
80 if (F_ISSET(t, B_RDONLY)) {
87 status = __bt_bdelete(t, key);
91 * If flags is R_CURSOR, delete the cursor. Must already
92 * have started a scan and not have already deleted it.
95 if (F_ISSET(c, CURS_INIT)) {
96 if (F_ISSET(c, CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
98 if ((h = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL)
102 * If the page is about to be emptied, we'll need to
103 * delete it, which means we have to acquire a stack.
105 if (NEXTINDEX(h) == 1)
106 if (__bt_stkacq(t, &h, &t->bt_cursor))
109 status = __bt_dleaf(t, NULL, h, c->pg.index);
111 if (NEXTINDEX(h) == 0 && status == RET_SUCCESS) {
112 if (__bt_pdelete(t, h))
116 h, status == RET_SUCCESS ? MPOOL_DIRTY : 0);
124 if (status == RET_SUCCESS)
125 F_SET(t, B_MODIFIED);
131 * Acquire a stack so we can delete a cursor entry.
135 * hp: pointer to current, pinned PAGE pointer
136 * c: pointer to the cursor
139 * 0 on success, 1 on failure
142 __bt_stkacq(t, hp, c)
153 recno_t nextpg, prevpg;
157 * Find the first occurrence of the key in the tree. Toss the
158 * currently locked page so we don't hit an already-locked page.
161 mpool_put(t->bt_mp, h, 0);
162 if ((e = __bt_search(t, &c->key, &exact)) == NULL)
166 /* See if we got it in one shot. */
167 if (h->pgno == c->pg.pgno)
171 * Move right, looking for the page. At each move we have to move
172 * up the stack until we don't have to move to the next page. If
173 * we have to change pages at an internal level, we have to fix the
176 while (h->pgno != c->pg.pgno) {
177 if ((nextpg = h->nextpg) == P_INVALID)
179 mpool_put(t->bt_mp, h, 0);
181 /* Move up the stack. */
182 for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
183 /* Get the parent page. */
184 if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
187 /* Move to the next index. */
188 if (parent->index != NEXTINDEX(h) - 1) {
189 index = parent->index + 1;
190 BT_PUSH(t, h->pgno, index);
193 mpool_put(t->bt_mp, h, 0);
196 /* Restore the stack. */
198 /* Push the next level down onto the stack. */
199 bi = GETBINTERNAL(h, index);
203 /* Lose the currently pinned page. */
204 mpool_put(t->bt_mp, h, 0);
206 /* Get the next level down. */
207 if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
211 mpool_put(t->bt_mp, h, 0);
212 if ((h = mpool_get(t->bt_mp, nextpg, 0)) == NULL)
216 if (h->pgno == c->pg.pgno)
219 /* Reacquire the original stack. */
220 mpool_put(t->bt_mp, h, 0);
221 if ((e = __bt_search(t, &c->key, &exact)) == NULL)
226 * Move left, looking for the page. At each move we have to move
227 * up the stack until we don't have to change pages to move to the
228 * next page. If we have to change pages at an internal level, we
229 * have to fix the stack back up.
231 while (h->pgno != c->pg.pgno) {
232 if ((prevpg = h->prevpg) == P_INVALID)
234 mpool_put(t->bt_mp, h, 0);
236 /* Move up the stack. */
237 for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
238 /* Get the parent page. */
239 if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
242 /* Move to the next index. */
243 if (parent->index != 0) {
244 index = parent->index - 1;
245 BT_PUSH(t, h->pgno, index);
248 mpool_put(t->bt_mp, h, 0);
251 /* Restore the stack. */
253 /* Push the next level down onto the stack. */
254 bi = GETBINTERNAL(h, index);
257 /* Lose the currently pinned page. */
258 mpool_put(t->bt_mp, h, 0);
260 /* Get the next level down. */
261 if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
264 index = NEXTINDEX(h) - 1;
265 BT_PUSH(t, pgno, index);
267 mpool_put(t->bt_mp, h, 0);
268 if ((h = mpool_get(t->bt_mp, prevpg, 0)) == NULL)
273 ret: mpool_put(t->bt_mp, h, 0);
274 return ((*hp = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL);
279 * Delete all key/data pairs matching the specified key.
286 * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
295 int deleted, exact, redo;
299 /* Find any matching record; __bt_search pins the page. */
300 loop: if ((e = __bt_search(t, key, &exact)) == NULL)
301 return (deleted ? RET_SUCCESS : RET_ERROR);
303 mpool_put(t->bt_mp, e->page, 0);
304 return (deleted ? RET_SUCCESS : RET_SPECIAL);
308 * Delete forward, then delete backward, from the found key. If
309 * there are duplicates and we reach either side of the page, do
310 * the key search again, so that we get them all.
315 if (__bt_dleaf(t, key, h, e->index)) {
316 mpool_put(t->bt_mp, h, 0);
319 if (F_ISSET(t, B_NODUPS)) {
320 if (NEXTINDEX(h) == 0) {
321 if (__bt_pdelete(t, h))
324 mpool_put(t->bt_mp, h, MPOOL_DIRTY);
325 return (RET_SUCCESS);
328 } while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0);
330 /* Check for right-hand edge of the page. */
331 if (e->index == NEXTINDEX(h))
334 /* Delete from the key to the beginning of the page. */
335 while (e->index-- > 0) {
336 if (__bt_cmp(t, key, e) != 0)
338 if (__bt_dleaf(t, key, h, e->index) == RET_ERROR) {
339 mpool_put(t->bt_mp, h, 0);
346 /* Check for an empty page. */
347 if (NEXTINDEX(h) == 0) {
348 if (__bt_pdelete(t, h))
354 mpool_put(t->bt_mp, h, MPOOL_DIRTY);
358 return (RET_SUCCESS);
363 * Delete a single page from the tree.
370 * RET_SUCCESS, RET_ERROR.
373 * mpool_put's the page
383 indx_t cnt, index, *ip, offset;
388 * Walk the parent page stack -- a LIFO stack of the pages that were
389 * traversed when we searched for the page where the delete occurred.
390 * Each stack entry is a page number and a page index offset. The
391 * offset is for the page traversed on the search. We've just deleted
392 * a page, so we have to delete the key from the parent page.
394 * If the delete from the parent page makes it empty, this process may
395 * continue all the way up the tree. We stop if we reach the root page
396 * (which is never deleted, it's just not worth the effort) or if the
397 * delete does not empty the page.
399 while ((parent = BT_POP(t)) != NULL) {
400 /* Get the parent page. */
401 if ((pg = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
404 index = parent->index;
405 bi = GETBINTERNAL(pg, index);
407 /* Free any overflow pages. */
408 if (bi->flags & P_BIGKEY &&
409 __ovfl_delete(t, bi->bytes) == RET_ERROR) {
410 mpool_put(t->bt_mp, pg, 0);
415 * Free the parent if it has only the one key and it's not the
416 * root page. If it's the rootpage, turn it back into an empty
419 if (NEXTINDEX(pg) == 1)
420 if (pg->pgno == P_ROOT) {
421 pg->lower = BTDATAOFF;
422 pg->upper = t->bt_psize;
425 if (__bt_relink(t, pg) || __bt_free(t, pg))
430 /* Pack remaining key items at the end of the page. */
431 nksize = NBINTERNAL(bi->ksize);
432 from = (char *)pg + pg->upper;
433 memmove(from + nksize, from, (char *)bi - from);
436 /* Adjust indices' offsets, shift the indices down. */
437 offset = pg->linp[index];
438 for (cnt = index, ip = &pg->linp[0]; cnt--; ++ip)
441 for (cnt = NEXTINDEX(pg) - index; --cnt; ++ip)
442 ip[0] = ip[1] < offset ? ip[1] + nksize : ip[1];
443 pg->lower -= sizeof(indx_t);
446 mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
450 /* Free the leaf page, as long as it wasn't the root. */
451 if (h->pgno == P_ROOT) {
452 mpool_put(t->bt_mp, h, MPOOL_DIRTY);
453 return (RET_SUCCESS);
455 return (__bt_relink(t, h) || __bt_free(t, h));
460 * Delete a single record from a leaf page.
464 * key: referenced key
466 * index: index on page to delete
469 * RET_SUCCESS, RET_ERROR.
472 __bt_dleaf(t, key, h, index)
479 indx_t cnt, *ip, offset;
484 /* If this record is referenced by the cursor, delete the cursor. */
485 if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
486 !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
487 t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index == index &&
488 __bt_curdel(t, key, h, index))
491 /* If the entry uses overflow pages, make them available for reuse. */
492 to = bl = GETBLEAF(h, index);
493 if (bl->flags & P_BIGKEY && __ovfl_delete(t, bl->bytes) == RET_ERROR)
495 if (bl->flags & P_BIGDATA &&
496 __ovfl_delete(t, bl->bytes + bl->ksize) == RET_ERROR)
499 /* Pack the remaining key/data items at the end of the page. */
501 from = (char *)h + h->upper;
502 memmove(from + nbytes, from, (char *)to - from);
505 /* Adjust the indices' offsets, shift the indices down. */
506 offset = h->linp[index];
507 for (cnt = index, ip = &h->linp[0]; cnt--; ++ip)
510 for (cnt = NEXTINDEX(h) - index; --cnt; ++ip)
511 ip[0] = ip[1] < offset ? ip[1] + nbytes : ip[1];
512 h->lower -= sizeof(indx_t);
514 /* If the cursor is on this page, adjust it as necessary. */
515 if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
516 !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
517 t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index > index)
518 --t->bt_cursor.pg.index;
520 return (RET_SUCCESS);
529 * key: referenced key (or NULL)
531 * index: index on page to delete
534 * RET_SUCCESS, RET_ERROR.
537 __bt_curdel(t, key, h, index)
549 * If there are duplicates, move forward or backward to one.
550 * Otherwise, copy the key into the cursor area.
553 F_CLR(c, CURS_AFTER | CURS_BEFORE | CURS_ACQUIRE);
556 if (!F_ISSET(t, B_NODUPS)) {
558 * We're going to have to do comparisons. If we weren't
559 * provided a copy of the key, i.e. the user is deleting
560 * the current cursor position, get one.
565 if ((status = __bt_ret(t, &e,
566 &c->key, &c->key, NULL, NULL, 1)) != RET_SUCCESS)
571 /* Check previous key, if not at the beginning of the page. */
575 if (__bt_cmp(t, key, &e) == 0) {
576 F_SET(c, CURS_BEFORE);
580 /* Check next key, if not at the end of the page. */
581 if (index < NEXTINDEX(h) - 1) {
584 if (__bt_cmp(t, key, &e) == 0) {
585 F_SET(c, CURS_AFTER);
589 /* Check previous key if at the beginning of the page. */
590 if (index == 0 && h->prevpg != P_INVALID) {
591 if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
594 e.index = NEXTINDEX(pg) - 1;
595 if (__bt_cmp(t, key, &e) == 0) {
596 F_SET(c, CURS_BEFORE);
599 mpool_put(t->bt_mp, pg, 0);
601 /* Check next key if at the end of the page. */
602 if (index == NEXTINDEX(h) - 1 && h->nextpg != P_INVALID) {
603 if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
607 if (__bt_cmp(t, key, &e) == 0) {
608 F_SET(c, CURS_AFTER);
609 dup1: mpool_put(t->bt_mp, pg, 0);
610 dup2: c->pg.pgno = e.page->pgno;
611 c->pg.index = e.index;
612 return (RET_SUCCESS);
614 mpool_put(t->bt_mp, pg, 0);
619 if (curcopy || (status =
620 __bt_ret(t, &e, &c->key, &c->key, NULL, NULL, 1)) == RET_SUCCESS) {
621 F_SET(c, CURS_ACQUIRE);
622 return (RET_SUCCESS);
629 * Link around a deleted page.
633 * h: page to be deleted
642 if (h->nextpg != P_INVALID) {
643 if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
645 pg->prevpg = h->prevpg;
646 mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
648 if (h->prevpg != P_INVALID) {
649 if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
651 pg->nextpg = h->nextpg;
652 mpool_put(t->bt_mp, pg, MPOOL_DIRTY);