bt_delete.c revision 1.16
1/*	$NetBSD: bt_delete.c,v 1.16 2008/09/11 12:58:00 joerg 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#if HAVE_NBTOOL_CONFIG_H
36#include "nbtool_config.h"
37#endif
38
39#include <sys/cdefs.h>
40__RCSID("$NetBSD: bt_delete.c,v 1.16 2008/09/11 12:58:00 joerg Exp $");
41
42#include "namespace.h"
43#include <sys/types.h>
44
45#include <assert.h>
46#include <errno.h>
47#include <stdio.h>
48#include <string.h>
49
50#include <db.h>
51#include "btree.h"
52
53static int __bt_bdelete(BTREE *, const DBT *);
54static int __bt_curdel(BTREE *, const DBT *, PAGE *, u_int);
55static int __bt_pdelete(BTREE *, PAGE *);
56static int __bt_relink(BTREE *, PAGE *);
57static int __bt_stkacq(BTREE *, PAGE **, CURSOR *);
58
59/*
60 * __bt_delete
61 *	Delete the item(s) referenced by a key.
62 *
63 * Return RET_SPECIAL if the key is not found.
64 */
65int
66__bt_delete(const DB *dbp, const DBT *key, u_int flags)
67{
68	BTREE *t;
69	CURSOR *c;
70	PAGE *h;
71	int status;
72
73	t = dbp->internal;
74
75	/* Toss any page pinned across calls. */
76	if (t->bt_pinned != NULL) {
77		mpool_put(t->bt_mp, t->bt_pinned, 0);
78		t->bt_pinned = NULL;
79	}
80
81	/* Check for change to a read-only tree. */
82	if (F_ISSET(t, B_RDONLY)) {
83		errno = EPERM;
84		return (RET_ERROR);
85	}
86
87	switch (flags) {
88	case 0:
89		status = __bt_bdelete(t, key);
90		break;
91	case R_CURSOR:
92		/*
93		 * If flags is R_CURSOR, delete the cursor.  Must already
94		 * have started a scan and not have already deleted it.
95		 */
96		c = &t->bt_cursor;
97		if (F_ISSET(c, CURS_INIT)) {
98			if (F_ISSET(c, CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
99				return (RET_SPECIAL);
100			if ((h = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL)
101				return (RET_ERROR);
102
103			/*
104			 * If the page is about to be emptied, we'll need to
105			 * delete it, which means we have to acquire a stack.
106			 */
107			if (NEXTINDEX(h) == 1)
108				if (__bt_stkacq(t, &h, &t->bt_cursor))
109					return (RET_ERROR);
110
111			status = __bt_dleaf(t, NULL, h, (u_int)c->pg.index);
112
113			if (NEXTINDEX(h) == 0 && status == RET_SUCCESS) {
114				if (__bt_pdelete(t, h))
115					return (RET_ERROR);
116			} else
117				mpool_put(t->bt_mp, h,
118				    (u_int)(status == RET_SUCCESS ?
119				    MPOOL_DIRTY : 0));
120			break;
121		}
122		/* FALLTHROUGH */
123	default:
124		errno = EINVAL;
125		return (RET_ERROR);
126	}
127	if (status == RET_SUCCESS)
128		F_SET(t, B_MODIFIED);
129	return (status);
130}
131
132/*
133 * __bt_stkacq --
134 *	Acquire a stack so we can delete a cursor entry.
135 *
136 * Parameters:
137 *	  t:	tree
138 *	 hp:	pointer to current, pinned PAGE pointer
139 *	  c:	pointer to the cursor
140 *
141 * Returns:
142 *	0 on success, 1 on failure
143 */
144static int
145__bt_stkacq(BTREE *t, PAGE **hp, CURSOR *c)
146{
147	BINTERNAL *bi;
148	EPG *e;
149	EPGNO *parent;
150	PAGE *h;
151	indx_t idx = 0;	/* Pacify gcc */
152	pgno_t pgno;
153	recno_t nextpg, prevpg;
154	int exact, level;
155
156	/*
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.
159	 */
160	h = *hp;
161	mpool_put(t->bt_mp, h, 0);
162	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
163		return (1);
164	h = e->page;
165
166	/* See if we got it in one shot. */
167	if (h->pgno == c->pg.pgno)
168		goto ret;
169
170	/*
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
174	 * stack back up.
175	 */
176	while (h->pgno != c->pg.pgno) {
177		if ((nextpg = h->nextpg) == P_INVALID)
178			break;
179		mpool_put(t->bt_mp, h, 0);
180
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)
185				return (1);
186
187			/* Move to the next index. */
188			if (parent->index != NEXTINDEX(h) - 1) {
189				idx = parent->index + 1;
190				BT_PUSH(t, h->pgno, idx);
191				break;
192			}
193			mpool_put(t->bt_mp, h, 0);
194		}
195
196		/* Restore the stack. */
197		while (level--) {
198			/* Push the next level down onto the stack. */
199			bi = GETBINTERNAL(h, idx);
200			pgno = bi->pgno;
201			BT_PUSH(t, pgno, 0);
202
203			/* Lose the currently pinned page. */
204			mpool_put(t->bt_mp, h, 0);
205
206			/* Get the next level down. */
207			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
208				return (1);
209			idx = 0;
210		}
211		mpool_put(t->bt_mp, h, 0);
212		if ((h = mpool_get(t->bt_mp, nextpg, 0)) == NULL)
213			return (1);
214	}
215
216	if (h->pgno == c->pg.pgno)
217		goto ret;
218
219	/* Reacquire the original stack. */
220	mpool_put(t->bt_mp, h, 0);
221	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
222		return (1);
223	h = e->page;
224
225	/*
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.
230	 */
231	while (h->pgno != c->pg.pgno) {
232		if ((prevpg = h->prevpg) == P_INVALID)
233			break;
234		mpool_put(t->bt_mp, h, 0);
235
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)
240				return (1);
241
242			/* Move to the next index. */
243			if (parent->index != 0) {
244				idx = parent->index - 1;
245				BT_PUSH(t, h->pgno, idx);
246				break;
247			}
248			mpool_put(t->bt_mp, h, 0);
249		}
250
251		/* Restore the stack. */
252		while (level--) {
253			/* Push the next level down onto the stack. */
254			bi = GETBINTERNAL(h, idx);
255			pgno = bi->pgno;
256
257			/* Lose the currently pinned page. */
258			mpool_put(t->bt_mp, h, 0);
259
260			/* Get the next level down. */
261			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
262				return (1);
263
264			idx = NEXTINDEX(h) - 1;
265			BT_PUSH(t, pgno, idx);
266		}
267		mpool_put(t->bt_mp, h, 0);
268		if ((h = mpool_get(t->bt_mp, prevpg, 0)) == NULL)
269			return (1);
270	}
271
272
273ret:	mpool_put(t->bt_mp, h, 0);
274	return ((*hp = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL);
275}
276
277/*
278 * __bt_bdelete --
279 *	Delete all key/data pairs matching the specified key.
280 *
281 * Parameters:
282 *	  t:	tree
283 *	key:	key to delete
284 *
285 * Returns:
286 *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
287 */
288static int
289__bt_bdelete(BTREE *t, const DBT *key)
290{
291	EPG *e;
292	PAGE *h;
293	int deleted, exact, redo;
294
295	deleted = 0;
296
297	/* Find any matching record; __bt_search pins the page. */
298loop:	if ((e = __bt_search(t, key, &exact)) == NULL)
299		return (deleted ? RET_SUCCESS : RET_ERROR);
300	if (!exact) {
301		mpool_put(t->bt_mp, e->page, 0);
302		return (deleted ? RET_SUCCESS : RET_SPECIAL);
303	}
304
305	/*
306	 * Delete forward, then delete backward, from the found key.  If
307	 * there are duplicates and we reach either side of the page, do
308	 * the key search again, so that we get them all.
309	 */
310	redo = 0;
311	h = e->page;
312	do {
313		if (__bt_dleaf(t, key, h, (u_int)e->index)) {
314			mpool_put(t->bt_mp, h, 0);
315			return (RET_ERROR);
316		}
317		if (F_ISSET(t, B_NODUPS)) {
318			if (NEXTINDEX(h) == 0) {
319				if (__bt_pdelete(t, h))
320					return (RET_ERROR);
321			} else
322				mpool_put(t->bt_mp, h, MPOOL_DIRTY);
323			return (RET_SUCCESS);
324		}
325		deleted = 1;
326	} while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0);
327
328	/* Check for right-hand edge of the page. */
329	if (e->index == NEXTINDEX(h))
330		redo = 1;
331
332	/* Delete from the key to the beginning of the page. */
333	while (e->index-- > 0) {
334		if (__bt_cmp(t, key, e) != 0)
335			break;
336		if (__bt_dleaf(t, key, h, (u_int)e->index) == RET_ERROR) {
337			mpool_put(t->bt_mp, h, 0);
338			return (RET_ERROR);
339		}
340		if (e->index == 0)
341			redo = 1;
342	}
343
344	/* Check for an empty page. */
345	if (NEXTINDEX(h) == 0) {
346		if (__bt_pdelete(t, h))
347			return (RET_ERROR);
348		goto loop;
349	}
350
351	/* Put the page. */
352	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
353
354	if (redo)
355		goto loop;
356	return (RET_SUCCESS);
357}
358
359/*
360 * __bt_pdelete --
361 *	Delete a single page from the tree.
362 *
363 * Parameters:
364 *	t:	tree
365 *	h:	leaf page
366 *
367 * Returns:
368 *	RET_SUCCESS, RET_ERROR.
369 *
370 * Side-effects:
371 *	mpool_put's the page
372 */
373static int
374__bt_pdelete(BTREE *t, PAGE *h)
375{
376	BINTERNAL *bi;
377	PAGE *pg;
378	EPGNO *parent;
379	indx_t cnt, idx, *ip, offset;
380	uint32_t nksize;
381	char *from;
382
383	/*
384	 * Walk the parent page stack -- a LIFO stack of the pages that were
385	 * traversed when we searched for the page where the delete occurred.
386	 * Each stack entry is a page number and a page index offset.  The
387	 * offset is for the page traversed on the search.  We've just deleted
388	 * a page, so we have to delete the key from the parent page.
389	 *
390	 * If the delete from the parent page makes it empty, this process may
391	 * continue all the way up the tree.  We stop if we reach the root page
392	 * (which is never deleted, it's just not worth the effort) or if the
393	 * delete does not empty the page.
394	 */
395	while ((parent = BT_POP(t)) != NULL) {
396		/* Get the parent page. */
397		if ((pg = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
398			return (RET_ERROR);
399
400		idx = parent->index;
401		bi = GETBINTERNAL(pg, idx);
402
403		/* Free any overflow pages. */
404		if (bi->flags & P_BIGKEY &&
405		    __ovfl_delete(t, bi->bytes) == RET_ERROR) {
406			mpool_put(t->bt_mp, pg, 0);
407			return (RET_ERROR);
408		}
409
410		/*
411		 * Free the parent if it has only the one key and it's not the
412		 * root page. If it's the rootpage, turn it back into an empty
413		 * leaf page.
414		 */
415		if (NEXTINDEX(pg) == 1) {
416			if (pg->pgno == P_ROOT) {
417				pg->lower = BTDATAOFF;
418				pg->upper = t->bt_psize;
419				pg->flags = P_BLEAF;
420			} else {
421				if (__bt_relink(t, pg) || __bt_free(t, pg))
422					return (RET_ERROR);
423				continue;
424			}
425		} else {
426			/* Pack remaining key items at the end of the page. */
427			nksize = NBINTERNAL(bi->ksize);
428			from = (char *)(void *)pg + pg->upper;
429			memmove(from + nksize, from,
430			(size_t)((char *)(void *)bi - from));
431			pg->upper += nksize;
432
433			/* Adjust indices' offsets, shift the indices down. */
434			offset = pg->linp[idx];
435			for (cnt = idx, ip = &pg->linp[0]; cnt--; ++ip)
436				if (ip[0] < offset)
437					ip[0] += nksize;
438			for (cnt = NEXTINDEX(pg) - idx; --cnt; ++ip)
439				ip[0] = ip[1] < offset ? ip[1] + nksize : ip[1];
440			pg->lower -= sizeof(indx_t);
441		}
442
443		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
444		break;
445	}
446
447	/* Free the leaf page, as long as it wasn't the root. */
448	if (h->pgno == P_ROOT) {
449		mpool_put(t->bt_mp, h, MPOOL_DIRTY);
450		return (RET_SUCCESS);
451	}
452	return (__bt_relink(t, h) || __bt_free(t, h));
453}
454
455/*
456 * __bt_dleaf --
457 *	Delete a single record from a leaf page.
458 *
459 * Parameters:
460 *	t:	tree
461 *    key:	referenced key
462 *	h:	page
463 *	idx:	index on page to delete
464 *
465 * Returns:
466 *	RET_SUCCESS, RET_ERROR.
467 */
468int
469__bt_dleaf(BTREE *t, const DBT *key, PAGE *h, u_int idx)
470{
471	BLEAF *bl;
472	indx_t cnt, *ip, offset;
473	uint32_t nbytes;
474	void *to;
475	char *from;
476
477	/* If this record is referenced by the cursor, delete the cursor. */
478	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
479	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
480	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index == idx &&
481	    __bt_curdel(t, key, h, idx))
482		return (RET_ERROR);
483
484	/* If the entry uses overflow pages, make them available for reuse. */
485	to = bl = GETBLEAF(h, idx);
486	if (bl->flags & P_BIGKEY && __ovfl_delete(t, bl->bytes) == RET_ERROR)
487		return (RET_ERROR);
488	if (bl->flags & P_BIGDATA &&
489	    __ovfl_delete(t, bl->bytes + bl->ksize) == RET_ERROR)
490		return (RET_ERROR);
491
492	/* Pack the remaining key/data items at the end of the page. */
493	nbytes = NBLEAF(bl);
494	from = (char *)(void *)h + h->upper;
495	memmove(from + nbytes, from, (size_t)((char *)(void *)to - from));
496	h->upper += nbytes;
497
498	/* Adjust the indices' offsets, shift the indices down. */
499	offset = h->linp[idx];
500	for (cnt = idx, ip = &h->linp[0]; cnt--; ++ip)
501		if (ip[0] < offset)
502			ip[0] += nbytes;
503	for (cnt = NEXTINDEX(h) - idx; --cnt; ++ip)
504		ip[0] = ip[1] < offset ? ip[1] + nbytes : ip[1];
505	h->lower -= sizeof(indx_t);
506
507	/* If the cursor is on this page, adjust it as necessary. */
508	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
509	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
510	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index > idx)
511		--t->bt_cursor.pg.index;
512
513	return (RET_SUCCESS);
514}
515
516/*
517 * __bt_curdel --
518 *	Delete the cursor.
519 *
520 * Parameters:
521 *	t:	tree
522 *    key:	referenced key (or NULL)
523 *	h:	page
524 *  idx:	index on page to delete
525 *
526 * Returns:
527 *	RET_SUCCESS, RET_ERROR.
528 */
529static int
530__bt_curdel(BTREE *t, const DBT *key, PAGE *h, u_int idx)
531{
532	CURSOR *c;
533	EPG e;
534	PAGE *pg;
535	int curcopy, status;
536
537	/*
538	 * If there are duplicates, move forward or backward to one.
539	 * Otherwise, copy the key into the cursor area.
540	 */
541	c = &t->bt_cursor;
542	F_CLR(c, CURS_AFTER | CURS_BEFORE | CURS_ACQUIRE);
543
544	curcopy = 0;
545	if (!F_ISSET(t, B_NODUPS)) {
546		/*
547		 * We're going to have to do comparisons.  If we weren't
548		 * provided a copy of the key, i.e. the user is deleting
549		 * the current cursor position, get one.
550		 */
551		if (key == NULL) {
552			e.page = h;
553			e.index = idx;
554			if ((status = __bt_ret(t, &e,
555			    &c->key, &c->key, NULL, NULL, 1)) != RET_SUCCESS)
556				return (status);
557			curcopy = 1;
558			key = &c->key;
559		}
560		/* Check previous key, if not at the beginning of the page. */
561		if (idx > 0) {
562			e.page = h;
563			e.index = idx - 1;
564			if (__bt_cmp(t, key, &e) == 0) {
565				F_SET(c, CURS_BEFORE);
566				goto dup2;
567			}
568		}
569		/* Check next key, if not at the end of the page. */
570		if (idx < NEXTINDEX(h) - 1) {
571			e.page = h;
572			e.index = idx + 1;
573			if (__bt_cmp(t, key, &e) == 0) {
574				F_SET(c, CURS_AFTER);
575				goto dup2;
576			}
577		}
578		/* Check previous key if at the beginning of the page. */
579		if (idx == 0 && h->prevpg != P_INVALID) {
580			if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
581				return (RET_ERROR);
582			e.page = pg;
583			e.index = NEXTINDEX(pg) - 1;
584			if (__bt_cmp(t, key, &e) == 0) {
585				F_SET(c, CURS_BEFORE);
586				goto dup1;
587			}
588			mpool_put(t->bt_mp, pg, 0);
589		}
590		/* Check next key if at the end of the page. */
591		if (idx == NEXTINDEX(h) - 1 && h->nextpg != P_INVALID) {
592			if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
593				return (RET_ERROR);
594			e.page = pg;
595			e.index = 0;
596			if (__bt_cmp(t, key, &e) == 0) {
597				F_SET(c, CURS_AFTER);
598dup1:				mpool_put(t->bt_mp, pg, 0);
599dup2:				c->pg.pgno = e.page->pgno;
600				c->pg.index = e.index;
601				return (RET_SUCCESS);
602			}
603			mpool_put(t->bt_mp, pg, 0);
604		}
605	}
606	e.page = h;
607	e.index = idx;
608	if (curcopy || (status =
609	    __bt_ret(t, &e, &c->key, &c->key, NULL, NULL, 1)) == RET_SUCCESS) {
610		F_SET(c, CURS_ACQUIRE);
611		return (RET_SUCCESS);
612	}
613	return (status);
614}
615
616/*
617 * __bt_relink --
618 *	Link around a deleted page.
619 *
620 * Parameters:
621 *	t:	tree
622 *	h:	page to be deleted
623 */
624static int
625__bt_relink(BTREE *t, PAGE *h)
626{
627	PAGE *pg;
628
629	if (h->nextpg != P_INVALID) {
630		if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
631			return (RET_ERROR);
632		pg->prevpg = h->prevpg;
633		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
634	}
635	if (h->prevpg != P_INVALID) {
636		if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
637			return (RET_ERROR);
638		pg->nextpg = h->nextpg;
639		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
640	}
641	return (0);
642}
643