bt_delete.c revision 1.13
1/*	$NetBSD: bt_delete.c,v 1.13 2007/02/03 23:46:09 christos Exp $	*/
2
3/*-
4 * Copyright (c) 1990, 1993, 1994
5 *	The Regents of the University of California.  All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Mike Olson.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 */
34
35#include <sys/cdefs.h>
36#if defined(LIBC_SCCS) && !defined(lint)
37#if 0
38static char sccsid[] = "@(#)bt_delete.c	8.13 (Berkeley) 7/28/94";
39#else
40__RCSID("$NetBSD: bt_delete.c,v 1.13 2007/02/03 23:46:09 christos Exp $");
41#endif
42#endif /* LIBC_SCCS and not lint */
43
44#include "namespace.h"
45#include <sys/types.h>
46
47#include <assert.h>
48#include <errno.h>
49#include <stdio.h>
50#include <string.h>
51
52#include <db.h>
53#include "btree.h"
54
55static int __bt_bdelete(BTREE *, const DBT *);
56static int __bt_curdel(BTREE *, const DBT *, PAGE *, u_int);
57static int __bt_pdelete(BTREE *, PAGE *);
58static int __bt_relink(BTREE *, PAGE *);
59static int __bt_stkacq(BTREE *, PAGE **, CURSOR *);
60
61/*
62 * __bt_delete
63 *	Delete the item(s) referenced by a key.
64 *
65 * Return RET_SPECIAL if the key is not found.
66 */
67int
68__bt_delete(const DB *dbp, const DBT *key, u_int flags)
69{
70	BTREE *t;
71	CURSOR *c;
72	PAGE *h;
73	int status;
74
75	t = dbp->internal;
76
77	/* Toss any page pinned across calls. */
78	if (t->bt_pinned != NULL) {
79		mpool_put(t->bt_mp, t->bt_pinned, 0);
80		t->bt_pinned = NULL;
81	}
82
83	/* Check for change to a read-only tree. */
84	if (F_ISSET(t, B_RDONLY)) {
85		errno = EPERM;
86		return (RET_ERROR);
87	}
88
89	switch (flags) {
90	case 0:
91		status = __bt_bdelete(t, key);
92		break;
93	case R_CURSOR:
94		/*
95		 * If flags is R_CURSOR, delete the cursor.  Must already
96		 * have started a scan and not have already deleted it.
97		 */
98		c = &t->bt_cursor;
99		if (F_ISSET(c, CURS_INIT)) {
100			if (F_ISSET(c, CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
101				return (RET_SPECIAL);
102			if ((h = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL)
103				return (RET_ERROR);
104
105			/*
106			 * If the page is about to be emptied, we'll need to
107			 * delete it, which means we have to acquire a stack.
108			 */
109			if (NEXTINDEX(h) == 1)
110				if (__bt_stkacq(t, &h, &t->bt_cursor))
111					return (RET_ERROR);
112
113			status = __bt_dleaf(t, NULL, h, (u_int)c->pg.index);
114
115			if (NEXTINDEX(h) == 0 && status == RET_SUCCESS) {
116				if (__bt_pdelete(t, h))
117					return (RET_ERROR);
118			} else
119				mpool_put(t->bt_mp, h,
120				    (u_int)(status == RET_SUCCESS ?
121				    MPOOL_DIRTY : 0));
122			break;
123		}
124		/* FALLTHROUGH */
125	default:
126		errno = EINVAL;
127		return (RET_ERROR);
128	}
129	if (status == RET_SUCCESS)
130		F_SET(t, B_MODIFIED);
131	return (status);
132}
133
134/*
135 * __bt_stkacq --
136 *	Acquire a stack so we can delete a cursor entry.
137 *
138 * Parameters:
139 *	  t:	tree
140 *	 hp:	pointer to current, pinned PAGE pointer
141 *	  c:	pointer to the cursor
142 *
143 * Returns:
144 *	0 on success, 1 on failure
145 */
146static int
147__bt_stkacq(BTREE *t, PAGE **hp, CURSOR *c)
148{
149	BINTERNAL *bi;
150	EPG *e;
151	EPGNO *parent;
152	PAGE *h;
153	indx_t idx = 0;	/* Pacify gcc */
154	pgno_t pgno;
155	recno_t nextpg, prevpg;
156	int exact, level;
157
158	/*
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.
161	 */
162	h = *hp;
163	mpool_put(t->bt_mp, h, 0);
164	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
165		return (1);
166	h = e->page;
167
168	/* See if we got it in one shot. */
169	if (h->pgno == c->pg.pgno)
170		goto ret;
171
172	/*
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
176	 * stack back up.
177	 */
178	while (h->pgno != c->pg.pgno) {
179		if ((nextpg = h->nextpg) == P_INVALID)
180			break;
181		mpool_put(t->bt_mp, h, 0);
182
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)
187				return (1);
188
189			/* Move to the next index. */
190			if (parent->index != NEXTINDEX(h) - 1) {
191				idx = parent->index + 1;
192				BT_PUSH(t, h->pgno, idx);
193				break;
194			}
195			mpool_put(t->bt_mp, h, 0);
196		}
197
198		/* Restore the stack. */
199		while (level--) {
200			/* Push the next level down onto the stack. */
201			bi = GETBINTERNAL(h, idx);
202			pgno = bi->pgno;
203			BT_PUSH(t, pgno, 0);
204
205			/* Lose the currently pinned page. */
206			mpool_put(t->bt_mp, h, 0);
207
208			/* Get the next level down. */
209			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
210				return (1);
211			idx = 0;
212		}
213		mpool_put(t->bt_mp, h, 0);
214		if ((h = mpool_get(t->bt_mp, nextpg, 0)) == NULL)
215			return (1);
216	}
217
218	if (h->pgno == c->pg.pgno)
219		goto ret;
220
221	/* Reacquire the original stack. */
222	mpool_put(t->bt_mp, h, 0);
223	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
224		return (1);
225	h = e->page;
226
227	/*
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.
232	 */
233	while (h->pgno != c->pg.pgno) {
234		if ((prevpg = h->prevpg) == P_INVALID)
235			break;
236		mpool_put(t->bt_mp, h, 0);
237
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)
242				return (1);
243
244			/* Move to the next index. */
245			if (parent->index != 0) {
246				idx = parent->index - 1;
247				BT_PUSH(t, h->pgno, idx);
248				break;
249			}
250			mpool_put(t->bt_mp, h, 0);
251		}
252
253		/* Restore the stack. */
254		while (level--) {
255			/* Push the next level down onto the stack. */
256			bi = GETBINTERNAL(h, idx);
257			pgno = bi->pgno;
258
259			/* Lose the currently pinned page. */
260			mpool_put(t->bt_mp, h, 0);
261
262			/* Get the next level down. */
263			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
264				return (1);
265
266			idx = NEXTINDEX(h) - 1;
267			BT_PUSH(t, pgno, idx);
268		}
269		mpool_put(t->bt_mp, h, 0);
270		if ((h = mpool_get(t->bt_mp, prevpg, 0)) == NULL)
271			return (1);
272	}
273
274
275ret:	mpool_put(t->bt_mp, h, 0);
276	return ((*hp = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL);
277}
278
279/*
280 * __bt_bdelete --
281 *	Delete all key/data pairs matching the specified key.
282 *
283 * Parameters:
284 *	  t:	tree
285 *	key:	key to delete
286 *
287 * Returns:
288 *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
289 */
290static int
291__bt_bdelete(BTREE *t, const DBT *key)
292{
293	EPG *e;
294	PAGE *h;
295	int deleted, exact, redo;
296
297	deleted = 0;
298
299	/* Find any matching record; __bt_search pins the page. */
300loop:	if ((e = __bt_search(t, key, &exact)) == NULL)
301		return (deleted ? RET_SUCCESS : RET_ERROR);
302	if (!exact) {
303		mpool_put(t->bt_mp, e->page, 0);
304		return (deleted ? RET_SUCCESS : RET_SPECIAL);
305	}
306
307	/*
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.
311	 */
312	redo = 0;
313	h = e->page;
314	do {
315		if (__bt_dleaf(t, key, h, (u_int)e->index)) {
316			mpool_put(t->bt_mp, h, 0);
317			return (RET_ERROR);
318		}
319		if (F_ISSET(t, B_NODUPS)) {
320			if (NEXTINDEX(h) == 0) {
321				if (__bt_pdelete(t, h))
322					return (RET_ERROR);
323			} else
324				mpool_put(t->bt_mp, h, MPOOL_DIRTY);
325			return (RET_SUCCESS);
326		}
327		deleted = 1;
328	} while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0);
329
330	/* Check for right-hand edge of the page. */
331	if (e->index == NEXTINDEX(h))
332		redo = 1;
333
334	/* Delete from the key to the beginning of the page. */
335	while (e->index-- > 0) {
336		if (__bt_cmp(t, key, e) != 0)
337			break;
338		if (__bt_dleaf(t, key, h, (u_int)e->index) == RET_ERROR) {
339			mpool_put(t->bt_mp, h, 0);
340			return (RET_ERROR);
341		}
342		if (e->index == 0)
343			redo = 1;
344	}
345
346	/* Check for an empty page. */
347	if (NEXTINDEX(h) == 0) {
348		if (__bt_pdelete(t, h))
349			return (RET_ERROR);
350		goto loop;
351	}
352
353	/* Put the page. */
354	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
355
356	if (redo)
357		goto loop;
358	return (RET_SUCCESS);
359}
360
361/*
362 * __bt_pdelete --
363 *	Delete a single page from the tree.
364 *
365 * Parameters:
366 *	t:	tree
367 *	h:	leaf page
368 *
369 * Returns:
370 *	RET_SUCCESS, RET_ERROR.
371 *
372 * Side-effects:
373 *	mpool_put's the page
374 */
375static int
376__bt_pdelete(BTREE *t, PAGE *h)
377{
378	BINTERNAL *bi;
379	PAGE *pg;
380	EPGNO *parent;
381	indx_t cnt, idx, *ip, offset;
382	u_int32_t nksize;
383	char *from;
384
385	/*
386	 * Walk the parent page stack -- a LIFO stack of the pages that were
387	 * traversed when we searched for the page where the delete occurred.
388	 * Each stack entry is a page number and a page index offset.  The
389	 * offset is for the page traversed on the search.  We've just deleted
390	 * a page, so we have to delete the key from the parent page.
391	 *
392	 * If the delete from the parent page makes it empty, this process may
393	 * continue all the way up the tree.  We stop if we reach the root page
394	 * (which is never deleted, it's just not worth the effort) or if the
395	 * delete does not empty the page.
396	 */
397	while ((parent = BT_POP(t)) != NULL) {
398		/* Get the parent page. */
399		if ((pg = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
400			return (RET_ERROR);
401
402		idx = parent->index;
403		bi = GETBINTERNAL(pg, idx);
404
405		/* Free any overflow pages. */
406		if (bi->flags & P_BIGKEY &&
407		    __ovfl_delete(t, bi->bytes) == RET_ERROR) {
408			mpool_put(t->bt_mp, pg, 0);
409			return (RET_ERROR);
410		}
411
412		/*
413		 * Free the parent if it has only the one key and it's not the
414		 * root page. If it's the rootpage, turn it back into an empty
415		 * leaf page.
416		 */
417		if (NEXTINDEX(pg) == 1) {
418			if (pg->pgno == P_ROOT) {
419				pg->lower = BTDATAOFF;
420				pg->upper = t->bt_psize;
421				pg->flags = P_BLEAF;
422			} else {
423				if (__bt_relink(t, pg) || __bt_free(t, pg))
424					return (RET_ERROR);
425				continue;
426			}
427		} else {
428			/* Pack remaining key items at the end of the page. */
429			nksize = NBINTERNAL(bi->ksize);
430			from = (char *)(void *)pg + pg->upper;
431			memmove(from + nksize, from,
432			(size_t)((char *)(void *)bi - from));
433			pg->upper += nksize;
434
435			/* Adjust indices' offsets, shift the indices down. */
436			offset = pg->linp[idx];
437			for (cnt = idx, ip = &pg->linp[0]; cnt--; ++ip)
438				if (ip[0] < offset)
439					ip[0] += nksize;
440			for (cnt = NEXTINDEX(pg) - idx; --cnt; ++ip)
441				ip[0] = ip[1] < offset ? ip[1] + nksize : ip[1];
442			pg->lower -= sizeof(indx_t);
443		}
444
445		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
446		break;
447	}
448
449	/* Free the leaf page, as long as it wasn't the root. */
450	if (h->pgno == P_ROOT) {
451		mpool_put(t->bt_mp, h, MPOOL_DIRTY);
452		return (RET_SUCCESS);
453	}
454	return (__bt_relink(t, h) || __bt_free(t, h));
455}
456
457/*
458 * __bt_dleaf --
459 *	Delete a single record from a leaf page.
460 *
461 * Parameters:
462 *	t:	tree
463 *    key:	referenced key
464 *	h:	page
465 *	idx:	index on page to delete
466 *
467 * Returns:
468 *	RET_SUCCESS, RET_ERROR.
469 */
470int
471__bt_dleaf(BTREE *t, const DBT *key, PAGE *h, u_int idx)
472{
473	BLEAF *bl;
474	indx_t cnt, *ip, offset;
475	u_int32_t nbytes;
476	void *to;
477	char *from;
478
479	/* If this record is referenced by the cursor, delete the cursor. */
480	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
481	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
482	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index == idx &&
483	    __bt_curdel(t, key, h, idx))
484		return (RET_ERROR);
485
486	/* If the entry uses overflow pages, make them available for reuse. */
487	to = bl = GETBLEAF(h, idx);
488	if (bl->flags & P_BIGKEY && __ovfl_delete(t, bl->bytes) == RET_ERROR)
489		return (RET_ERROR);
490	if (bl->flags & P_BIGDATA &&
491	    __ovfl_delete(t, bl->bytes + bl->ksize) == RET_ERROR)
492		return (RET_ERROR);
493
494	/* Pack the remaining key/data items at the end of the page. */
495	nbytes = NBLEAF(bl);
496	from = (char *)(void *)h + h->upper;
497	memmove(from + nbytes, from, (size_t)((char *)(void *)to - from));
498	h->upper += nbytes;
499
500	/* Adjust the indices' offsets, shift the indices down. */
501	offset = h->linp[idx];
502	for (cnt = idx, ip = &h->linp[0]; cnt--; ++ip)
503		if (ip[0] < offset)
504			ip[0] += nbytes;
505	for (cnt = NEXTINDEX(h) - idx; --cnt; ++ip)
506		ip[0] = ip[1] < offset ? ip[1] + nbytes : ip[1];
507	h->lower -= sizeof(indx_t);
508
509	/* If the cursor is on this page, adjust it as necessary. */
510	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
511	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
512	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index > idx)
513		--t->bt_cursor.pg.index;
514
515	return (RET_SUCCESS);
516}
517
518/*
519 * __bt_curdel --
520 *	Delete the cursor.
521 *
522 * Parameters:
523 *	t:	tree
524 *    key:	referenced key (or NULL)
525 *	h:	page
526 *  idx:	index on page to delete
527 *
528 * Returns:
529 *	RET_SUCCESS, RET_ERROR.
530 */
531static int
532__bt_curdel(BTREE *t, const DBT *key, PAGE *h, u_int idx)
533{
534	CURSOR *c;
535	EPG e;
536	PAGE *pg;
537	int curcopy, status;
538
539	/*
540	 * If there are duplicates, move forward or backward to one.
541	 * Otherwise, copy the key into the cursor area.
542	 */
543	c = &t->bt_cursor;
544	F_CLR(c, CURS_AFTER | CURS_BEFORE | CURS_ACQUIRE);
545
546	curcopy = 0;
547	if (!F_ISSET(t, B_NODUPS)) {
548		/*
549		 * We're going to have to do comparisons.  If we weren't
550		 * provided a copy of the key, i.e. the user is deleting
551		 * the current cursor position, get one.
552		 */
553		if (key == NULL) {
554			e.page = h;
555			e.index = idx;
556			if ((status = __bt_ret(t, &e,
557			    &c->key, &c->key, NULL, NULL, 1)) != RET_SUCCESS)
558				return (status);
559			curcopy = 1;
560			key = &c->key;
561		}
562		/* Check previous key, if not at the beginning of the page. */
563		if (idx > 0) {
564			e.page = h;
565			e.index = idx - 1;
566			if (__bt_cmp(t, key, &e) == 0) {
567				F_SET(c, CURS_BEFORE);
568				goto dup2;
569			}
570		}
571		/* Check next key, if not at the end of the page. */
572		if (idx < NEXTINDEX(h) - 1) {
573			e.page = h;
574			e.index = idx + 1;
575			if (__bt_cmp(t, key, &e) == 0) {
576				F_SET(c, CURS_AFTER);
577				goto dup2;
578			}
579		}
580		/* Check previous key if at the beginning of the page. */
581		if (idx == 0 && h->prevpg != P_INVALID) {
582			if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
583				return (RET_ERROR);
584			e.page = pg;
585			e.index = NEXTINDEX(pg) - 1;
586			if (__bt_cmp(t, key, &e) == 0) {
587				F_SET(c, CURS_BEFORE);
588				goto dup1;
589			}
590			mpool_put(t->bt_mp, pg, 0);
591		}
592		/* Check next key if at the end of the page. */
593		if (idx == NEXTINDEX(h) - 1 && h->nextpg != P_INVALID) {
594			if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
595				return (RET_ERROR);
596			e.page = pg;
597			e.index = 0;
598			if (__bt_cmp(t, key, &e) == 0) {
599				F_SET(c, CURS_AFTER);
600dup1:				mpool_put(t->bt_mp, pg, 0);
601dup2:				c->pg.pgno = e.page->pgno;
602				c->pg.index = e.index;
603				return (RET_SUCCESS);
604			}
605			mpool_put(t->bt_mp, pg, 0);
606		}
607	}
608	e.page = h;
609	e.index = idx;
610	if (curcopy || (status =
611	    __bt_ret(t, &e, &c->key, &c->key, NULL, NULL, 1)) == RET_SUCCESS) {
612		F_SET(c, CURS_ACQUIRE);
613		return (RET_SUCCESS);
614	}
615	return (status);
616}
617
618/*
619 * __bt_relink --
620 *	Link around a deleted page.
621 *
622 * Parameters:
623 *	t:	tree
624 *	h:	page to be deleted
625 */
626static int
627__bt_relink(BTREE *t, PAGE *h)
628{
629	PAGE *pg;
630
631	if (h->nextpg != P_INVALID) {
632		if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
633			return (RET_ERROR);
634		pg->prevpg = h->prevpg;
635		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
636	}
637	if (h->prevpg != P_INVALID) {
638		if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
639			return (RET_ERROR);
640		pg->nextpg = h->nextpg;
641		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
642	}
643	return (0);
644}
645