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
37 #if defined(LIBC_SCCS) && !defined(lint)
38 static char sccsid
[] = "@(#)bt_delete.c 8.13 (Berkeley) 7/28/94";
39 #endif /* LIBC_SCCS and not lint */
41 #include <sys/types.h>
50 static int __bt_bdelete
__P((BTREE
*, const DBT
*));
51 static int __bt_curdel
__P((BTREE
*, const DBT
*, PAGE
*, u_int
));
52 static int __bt_pdelete
__P((BTREE
*, PAGE
*));
53 static int __bt_relink
__P((BTREE
*, PAGE
*));
54 static int __bt_stkacq
__P((BTREE
*, PAGE
**, CURSOR
*));
58 * Delete the item(s) referenced by a key.
60 * Return RET_SPECIAL if the key is not found.
63 __bt_delete(dbp
, key
, flags
)
75 /* Toss any page pinned across calls. */
76 if (t
->bt_pinned
!= NULL
) {
77 mpool_put(t
->bt_mp
, t
->bt_pinned
, 0);
81 /* Check for change to a read-only tree. */
82 if (F_ISSET(t
, B_RDONLY
)) {
89 status
= __bt_bdelete(t
, key
);
93 * If flags is R_CURSOR, delete the cursor. Must already
94 * have started a scan and not have already deleted it.
97 if (F_ISSET(c
, CURS_INIT
)) {
98 if (F_ISSET(c
, CURS_ACQUIRE
| CURS_AFTER
| CURS_BEFORE
))
100 if ((h
= mpool_get(t
->bt_mp
, c
->pg
.pgno
, 0)) == NULL
)
104 * If the page is about to be emptied, we'll need to
105 * delete it, which means we have to acquire a stack.
107 if (NEXTINDEX(h
) == 1)
108 if (__bt_stkacq(t
, &h
, &t
->bt_cursor
))
111 status
= __bt_dleaf(t
, NULL
, h
, c
->pg
.index
);
113 if (NEXTINDEX(h
) == 0 && status
== RET_SUCCESS
) {
114 if (__bt_pdelete(t
, h
))
118 h
, status
== RET_SUCCESS
? MPOOL_DIRTY
: 0);
126 if (status
== RET_SUCCESS
)
127 F_SET(t
, B_MODIFIED
);
133 * Acquire a stack so we can delete a cursor entry.
137 * hp: pointer to current, pinned PAGE pointer
138 * c: pointer to the cursor
141 * 0 on success, 1 on failure
144 __bt_stkacq(t
, hp
, c
)
155 recno_t nextpg
, prevpg
;
159 * Find the first occurrence of the key in the tree. Toss the
160 * currently locked page so we don't hit an already-locked page.
163 mpool_put(t
->bt_mp
, h
, 0);
164 if ((e
= __bt_search(t
, &c
->key
, &exact
)) == NULL
)
168 /* See if we got it in one shot. */
169 if (h
->pgno
== c
->pg
.pgno
)
173 * Move right, looking for the page. At each move we have to move
174 * up the stack until we don't have to move to the next page. If
175 * we have to change pages at an internal level, we have to fix the
178 while (h
->pgno
!= c
->pg
.pgno
) {
179 if ((nextpg
= h
->nextpg
) == P_INVALID
)
181 mpool_put(t
->bt_mp
, h
, 0);
183 /* Move up the stack. */
184 for (level
= 0; (parent
= BT_POP(t
)) != NULL
; ++level
) {
185 /* Get the parent page. */
186 if ((h
= mpool_get(t
->bt_mp
, parent
->pgno
, 0)) == NULL
)
189 /* Move to the next index. */
190 if (parent
->index
!= NEXTINDEX(h
) - 1) {
191 index
= parent
->index
+ 1;
192 BT_PUSH(t
, h
->pgno
, index
);
195 mpool_put(t
->bt_mp
, h
, 0);
198 /* Restore the stack. */
200 /* Push the next level down onto the stack. */
201 bi
= GETBINTERNAL(h
, index
);
205 /* Lose the currently pinned page. */
206 mpool_put(t
->bt_mp
, h
, 0);
208 /* Get the next level down. */
209 if ((h
= mpool_get(t
->bt_mp
, pgno
, 0)) == NULL
)
213 mpool_put(t
->bt_mp
, h
, 0);
214 if ((h
= mpool_get(t
->bt_mp
, nextpg
, 0)) == NULL
)
218 if (h
->pgno
== c
->pg
.pgno
)
221 /* Reacquire the original stack. */
222 mpool_put(t
->bt_mp
, h
, 0);
223 if ((e
= __bt_search(t
, &c
->key
, &exact
)) == NULL
)
228 * Move left, looking for the page. At each move we have to move
229 * up the stack until we don't have to change pages to move to the
230 * next page. If we have to change pages at an internal level, we
231 * have to fix the stack back up.
233 while (h
->pgno
!= c
->pg
.pgno
) {
234 if ((prevpg
= h
->prevpg
) == P_INVALID
)
236 mpool_put(t
->bt_mp
, h
, 0);
238 /* Move up the stack. */
239 for (level
= 0; (parent
= BT_POP(t
)) != NULL
; ++level
) {
240 /* Get the parent page. */
241 if ((h
= mpool_get(t
->bt_mp
, parent
->pgno
, 0)) == NULL
)
244 /* Move to the next index. */
245 if (parent
->index
!= 0) {
246 index
= parent
->index
- 1;
247 BT_PUSH(t
, h
->pgno
, index
);
250 mpool_put(t
->bt_mp
, h
, 0);
253 /* Restore the stack. */
255 /* Push the next level down onto the stack. */
256 bi
= GETBINTERNAL(h
, index
);
259 /* Lose the currently pinned page. */
260 mpool_put(t
->bt_mp
, h
, 0);
262 /* Get the next level down. */
263 if ((h
= mpool_get(t
->bt_mp
, pgno
, 0)) == NULL
)
266 index
= NEXTINDEX(h
) - 1;
267 BT_PUSH(t
, pgno
, index
);
269 mpool_put(t
->bt_mp
, h
, 0);
270 if ((h
= mpool_get(t
->bt_mp
, prevpg
, 0)) == NULL
)
275 ret
: mpool_put(t
->bt_mp
, h
, 0);
276 return ((*hp
= mpool_get(t
->bt_mp
, c
->pg
.pgno
, 0)) == NULL
);
281 * Delete all key/data pairs matching the specified key.
288 * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
297 int deleted
, exact
, redo
;
301 /* Find any matching record; __bt_search pins the page. */
302 loop
: if ((e
= __bt_search(t
, key
, &exact
)) == NULL
)
303 return (deleted
? RET_SUCCESS
: RET_ERROR
);
305 mpool_put(t
->bt_mp
, e
->page
, 0);
306 return (deleted
? RET_SUCCESS
: RET_SPECIAL
);
310 * Delete forward, then delete backward, from the found key. If
311 * there are duplicates and we reach either side of the page, do
312 * the key search again, so that we get them all.
317 if (__bt_dleaf(t
, key
, h
, e
->index
)) {
318 mpool_put(t
->bt_mp
, h
, 0);
321 if (F_ISSET(t
, B_NODUPS
)) {
322 if (NEXTINDEX(h
) == 0) {
323 if (__bt_pdelete(t
, h
))
326 mpool_put(t
->bt_mp
, h
, MPOOL_DIRTY
);
327 return (RET_SUCCESS
);
330 } while (e
->index
< NEXTINDEX(h
) && __bt_cmp(t
, key
, e
) == 0);
332 /* Check for right-hand edge of the page. */
333 if (e
->index
== NEXTINDEX(h
))
336 /* Delete from the key to the beginning of the page. */
337 while (e
->index
-- > 0) {
338 if (__bt_cmp(t
, key
, e
) != 0)
340 if (__bt_dleaf(t
, key
, h
, e
->index
) == RET_ERROR
) {
341 mpool_put(t
->bt_mp
, h
, 0);
348 /* Check for an empty page. */
349 if (NEXTINDEX(h
) == 0) {
350 if (__bt_pdelete(t
, h
))
356 mpool_put(t
->bt_mp
, h
, MPOOL_DIRTY
);
360 return (RET_SUCCESS
);
365 * Delete a single page from the tree.
372 * RET_SUCCESS, RET_ERROR.
375 * mpool_put's the page
385 indx_t cnt
, index
, *ip
, offset
;
390 * Walk the parent page stack -- a LIFO stack of the pages that were
391 * traversed when we searched for the page where the delete occurred.
392 * Each stack entry is a page number and a page index offset. The
393 * offset is for the page traversed on the search. We've just deleted
394 * a page, so we have to delete the key from the parent page.
396 * If the delete from the parent page makes it empty, this process may
397 * continue all the way up the tree. We stop if we reach the root page
398 * (which is never deleted, it's just not worth the effort) or if the
399 * delete does not empty the page.
401 while ((parent
= BT_POP(t
)) != NULL
) {
402 /* Get the parent page. */
403 if ((pg
= mpool_get(t
->bt_mp
, parent
->pgno
, 0)) == NULL
)
406 index
= parent
->index
;
407 bi
= GETBINTERNAL(pg
, index
);
409 /* Free any overflow pages. */
410 if (bi
->flags
& P_BIGKEY
&&
411 __ovfl_delete(t
, bi
->bytes
) == RET_ERROR
) {
412 mpool_put(t
->bt_mp
, pg
, 0);
417 * Free the parent if it has only the one key and it's not the
418 * root page. If it's the rootpage, turn it back into an empty
421 if (NEXTINDEX(pg
) == 1)
422 if (pg
->pgno
== P_ROOT
) {
423 pg
->lower
= BTDATAOFF
;
424 pg
->upper
= t
->bt_psize
;
427 if (__bt_relink(t
, pg
) || __bt_free(t
, pg
))
432 /* Pack remaining key items at the end of the page. */
433 nksize
= NBINTERNAL(bi
->ksize
);
434 from
= (char *)pg
+ pg
->upper
;
435 memmove(from
+ nksize
, from
, (char *)bi
- from
);
438 /* Adjust indices' offsets, shift the indices down. */
439 offset
= pg
->linp
[index
];
440 for (cnt
= index
, ip
= &pg
->linp
[0]; cnt
--; ++ip
)
443 for (cnt
= NEXTINDEX(pg
) - index
; --cnt
; ++ip
)
444 ip
[0] = ip
[1] < offset
? ip
[1] + nksize
: ip
[1];
445 pg
->lower
-= sizeof(indx_t
);
448 mpool_put(t
->bt_mp
, pg
, MPOOL_DIRTY
);
452 /* Free the leaf page, as long as it wasn't the root. */
453 if (h
->pgno
== P_ROOT
) {
454 mpool_put(t
->bt_mp
, h
, MPOOL_DIRTY
);
455 return (RET_SUCCESS
);
457 return (__bt_relink(t
, h
) || __bt_free(t
, h
));
462 * Delete a single record from a leaf page.
466 * key: referenced key
468 * index: index on page to delete
471 * RET_SUCCESS, RET_ERROR.
474 __bt_dleaf(t
, key
, h
, index
)
481 indx_t cnt
, *ip
, offset
;
486 /* If this record is referenced by the cursor, delete the cursor. */
487 if (F_ISSET(&t
->bt_cursor
, CURS_INIT
) &&
488 !F_ISSET(&t
->bt_cursor
, CURS_ACQUIRE
) &&
489 t
->bt_cursor
.pg
.pgno
== h
->pgno
&& t
->bt_cursor
.pg
.index
== index
&&
490 __bt_curdel(t
, key
, h
, index
))
493 /* If the entry uses overflow pages, make them available for reuse. */
494 to
= bl
= GETBLEAF(h
, index
);
495 if (bl
->flags
& P_BIGKEY
&& __ovfl_delete(t
, bl
->bytes
) == RET_ERROR
)
497 if (bl
->flags
& P_BIGDATA
&&
498 __ovfl_delete(t
, bl
->bytes
+ bl
->ksize
) == RET_ERROR
)
501 /* Pack the remaining key/data items at the end of the page. */
503 from
= (char *)h
+ h
->upper
;
504 memmove(from
+ nbytes
, from
, (char *)to
- from
);
507 /* Adjust the indices' offsets, shift the indices down. */
508 offset
= h
->linp
[index
];
509 for (cnt
= index
, ip
= &h
->linp
[0]; cnt
--; ++ip
)
512 for (cnt
= NEXTINDEX(h
) - index
; --cnt
; ++ip
)
513 ip
[0] = ip
[1] < offset
? ip
[1] + nbytes
: ip
[1];
514 h
->lower
-= sizeof(indx_t
);
516 /* If the cursor is on this page, adjust it as necessary. */
517 if (F_ISSET(&t
->bt_cursor
, CURS_INIT
) &&
518 !F_ISSET(&t
->bt_cursor
, CURS_ACQUIRE
) &&
519 t
->bt_cursor
.pg
.pgno
== h
->pgno
&& t
->bt_cursor
.pg
.index
> index
)
520 --t
->bt_cursor
.pg
.index
;
522 return (RET_SUCCESS
);
531 * key: referenced key (or NULL)
533 * index: index on page to delete
536 * RET_SUCCESS, RET_ERROR.
539 __bt_curdel(t
, key
, h
, index
)
551 * If there are duplicates, move forward or backward to one.
552 * Otherwise, copy the key into the cursor area.
555 F_CLR(c
, CURS_AFTER
| CURS_BEFORE
| CURS_ACQUIRE
);
558 if (!F_ISSET(t
, B_NODUPS
)) {
560 * We're going to have to do comparisons. If we weren't
561 * provided a copy of the key, i.e. the user is deleting
562 * the current cursor position, get one.
567 if ((status
= __bt_ret(t
, &e
,
568 &c
->key
, &c
->key
, NULL
, NULL
, 1)) != RET_SUCCESS
)
573 /* Check previous key, if not at the beginning of the page. */
577 if (__bt_cmp(t
, key
, &e
) == 0) {
578 F_SET(c
, CURS_BEFORE
);
582 /* Check next key, if not at the end of the page. */
583 if (index
< NEXTINDEX(h
) - 1) {
586 if (__bt_cmp(t
, key
, &e
) == 0) {
587 F_SET(c
, CURS_AFTER
);
591 /* Check previous key if at the beginning of the page. */
592 if (index
== 0 && h
->prevpg
!= P_INVALID
) {
593 if ((pg
= mpool_get(t
->bt_mp
, h
->prevpg
, 0)) == NULL
)
596 e
.index
= NEXTINDEX(pg
) - 1;
597 if (__bt_cmp(t
, key
, &e
) == 0) {
598 F_SET(c
, CURS_BEFORE
);
601 mpool_put(t
->bt_mp
, pg
, 0);
603 /* Check next key if at the end of the page. */
604 if (index
== NEXTINDEX(h
) - 1 && h
->nextpg
!= P_INVALID
) {
605 if ((pg
= mpool_get(t
->bt_mp
, h
->nextpg
, 0)) == NULL
)
609 if (__bt_cmp(t
, key
, &e
) == 0) {
610 F_SET(c
, CURS_AFTER
);
611 dup1
: mpool_put(t
->bt_mp
, pg
, 0);
612 dup2
: c
->pg
.pgno
= e
.page
->pgno
;
613 c
->pg
.index
= e
.index
;
614 return (RET_SUCCESS
);
616 mpool_put(t
->bt_mp
, pg
, 0);
621 if (curcopy
|| (status
=
622 __bt_ret(t
, &e
, &c
->key
, &c
->key
, NULL
, NULL
, 1)) == RET_SUCCESS
) {
623 F_SET(c
, CURS_ACQUIRE
);
624 return (RET_SUCCESS
);
631 * Link around a deleted page.
635 * h: page to be deleted
644 if (h
->nextpg
!= P_INVALID
) {
645 if ((pg
= mpool_get(t
->bt_mp
, h
->nextpg
, 0)) == NULL
)
647 pg
->prevpg
= h
->prevpg
;
648 mpool_put(t
->bt_mp
, pg
, MPOOL_DIRTY
);
650 if (h
->prevpg
!= P_INVALID
) {
651 if ((pg
= mpool_get(t
->bt_mp
, h
->prevpg
, 0)) == NULL
)
653 pg
->nextpg
= h
->nextpg
;
654 mpool_put(t
->bt_mp
, pg
, MPOOL_DIRTY
);