xref: /dragonfly/lib/libc/db/btree/bt_delete.c (revision 3badf6b7)
1 /*-
2  * Copyright (c) 1990, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Mike Olson.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * @(#)bt_delete.c	8.13 (Berkeley) 7/28/94
33  * $FreeBSD: head/lib/libc/db/btree/bt_delete.c 189327 2009-03-04 00:58:04Z delphij $
34  */
35 
36 #include <sys/param.h>
37 
38 #include <errno.h>
39 #include <stdio.h>
40 #include <string.h>
41 
42 #include <db.h>
43 #include "btree.h"
44 
45 static int __bt_bdelete(BTREE *, const DBT *);
46 static int __bt_curdel(BTREE *, const DBT *, PAGE *, unsigned int);
47 static int __bt_pdelete(BTREE *, PAGE *);
48 static int __bt_relink(BTREE *, PAGE *);
49 static int __bt_stkacq(BTREE *, PAGE **, CURSOR *);
50 
51 /*
52  * __bt_delete
53  *	Delete the item(s) referenced by a key.
54  *
55  * Return RET_SPECIAL if the key is not found.
56  */
57 int
58 __bt_delete(const DB *dbp, const DBT *key, unsigned int flags)
59 {
60 	BTREE *t;
61 	CURSOR *c;
62 	PAGE *h;
63 	int status;
64 
65 	t = dbp->internal;
66 
67 	/* Toss any page pinned across calls. */
68 	if (t->bt_pinned != NULL) {
69 		mpool_put(t->bt_mp, t->bt_pinned, 0);
70 		t->bt_pinned = NULL;
71 	}
72 
73 	/* Check for change to a read-only tree. */
74 	if (F_ISSET(t, B_RDONLY)) {
75 		errno = EPERM;
76 		return (RET_ERROR);
77 	}
78 
79 	switch (flags) {
80 	case 0:
81 		status = __bt_bdelete(t, key);
82 		break;
83 	case R_CURSOR:
84 		/*
85 		 * If flags is R_CURSOR, delete the cursor.  Must already
86 		 * have started a scan and not have already deleted it.
87 		 */
88 		c = &t->bt_cursor;
89 		if (F_ISSET(c, CURS_INIT)) {
90 			if (F_ISSET(c, CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
91 				return (RET_SPECIAL);
92 			if ((h = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL)
93 				return (RET_ERROR);
94 
95 			/*
96 			 * If the page is about to be emptied, we'll need to
97 			 * delete it, which means we have to acquire a stack.
98 			 */
99 			if (NEXTINDEX(h) == 1)
100 				if (__bt_stkacq(t, &h, &t->bt_cursor))
101 					return (RET_ERROR);
102 
103 			status = __bt_dleaf(t, NULL, h, c->pg.index);
104 
105 			if (NEXTINDEX(h) == 0 && status == RET_SUCCESS) {
106 				if (__bt_pdelete(t, h))
107 					return (RET_ERROR);
108 			} else
109 				mpool_put(t->bt_mp,
110 				    h, status == RET_SUCCESS ? MPOOL_DIRTY : 0);
111 			break;
112 		}
113 		/* FALLTHROUGH */
114 	default:
115 		errno = EINVAL;
116 		return (RET_ERROR);
117 	}
118 	if (status == RET_SUCCESS)
119 		F_SET(t, B_MODIFIED);
120 	return (status);
121 }
122 
123 /*
124  * __bt_stkacq --
125  *	Acquire a stack so we can delete a cursor entry.
126  *
127  * Parameters:
128  *	  t:	tree
129  *	 hp:	pointer to current, pinned PAGE pointer
130  *	  c:	pointer to the cursor
131  *
132  * Returns:
133  *	0 on success, 1 on failure
134  */
135 static int
136 __bt_stkacq(BTREE *t, PAGE **hp, CURSOR *c)
137 {
138 	BINTERNAL *bi;
139 	EPG *e;
140 	EPGNO *parent;
141 	PAGE *h;
142 	indx_t idx;
143 	pgno_t pgno;
144 	recno_t nextpg, prevpg;
145 	int exact, level;
146 
147 	/*
148 	 * Find the first occurrence of the key in the tree.  Toss the
149 	 * currently locked page so we don't hit an already-locked page.
150 	 */
151 	h = *hp;
152 	mpool_put(t->bt_mp, h, 0);
153 	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
154 		return (1);
155 	h = e->page;
156 
157 	/* See if we got it in one shot. */
158 	if (h->pgno == c->pg.pgno)
159 		goto ret;
160 
161 	/*
162 	 * Move right, looking for the page.  At each move we have to move
163 	 * up the stack until we don't have to move to the next page.  If
164 	 * we have to change pages at an internal level, we have to fix the
165 	 * stack back up.
166 	 */
167 	while (h->pgno != c->pg.pgno) {
168 		if ((nextpg = h->nextpg) == P_INVALID)
169 			break;
170 		mpool_put(t->bt_mp, h, 0);
171 
172 		/* Move up the stack. */
173 		for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
174 			/* Get the parent page. */
175 			if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
176 				return (1);
177 
178 			/* Move to the next index. */
179 			if (parent->index != NEXTINDEX(h) - 1) {
180 				idx = parent->index + 1;
181 				BT_PUSH(t, h->pgno, idx);
182 				break;
183 			}
184 			mpool_put(t->bt_mp, h, 0);
185 		}
186 
187 		/* Restore the stack. */
188 		while (level--) {
189 			/* Push the next level down onto the stack. */
190 			bi = GETBINTERNAL(h, idx);
191 			pgno = bi->pgno;
192 			BT_PUSH(t, pgno, 0);
193 
194 			/* Lose the currently pinned page. */
195 			mpool_put(t->bt_mp, h, 0);
196 
197 			/* Get the next level down. */
198 			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
199 				return (1);
200 			idx = 0;
201 		}
202 		mpool_put(t->bt_mp, h, 0);
203 		if ((h = mpool_get(t->bt_mp, nextpg, 0)) == NULL)
204 			return (1);
205 	}
206 
207 	if (h->pgno == c->pg.pgno)
208 		goto ret;
209 
210 	/* Reacquire the original stack. */
211 	mpool_put(t->bt_mp, h, 0);
212 	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
213 		return (1);
214 	h = e->page;
215 
216 	/*
217 	 * Move left, looking for the page.  At each move we have to move
218 	 * up the stack until we don't have to change pages to move to the
219 	 * next page.  If we have to change pages at an internal level, we
220 	 * have to fix the stack back up.
221 	 */
222 	while (h->pgno != c->pg.pgno) {
223 		if ((prevpg = h->prevpg) == P_INVALID)
224 			break;
225 		mpool_put(t->bt_mp, h, 0);
226 
227 		/* Move up the stack. */
228 		for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
229 			/* Get the parent page. */
230 			if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
231 				return (1);
232 
233 			/* Move to the next index. */
234 			if (parent->index != 0) {
235 				idx = parent->index - 1;
236 				BT_PUSH(t, h->pgno, idx);
237 				break;
238 			}
239 			mpool_put(t->bt_mp, h, 0);
240 		}
241 
242 		/* Restore the stack. */
243 		while (level--) {
244 			/* Push the next level down onto the stack. */
245 			bi = GETBINTERNAL(h, idx);
246 			pgno = bi->pgno;
247 
248 			/* Lose the currently pinned page. */
249 			mpool_put(t->bt_mp, h, 0);
250 
251 			/* Get the next level down. */
252 			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
253 				return (1);
254 
255 			idx = NEXTINDEX(h) - 1;
256 			BT_PUSH(t, pgno, idx);
257 		}
258 		mpool_put(t->bt_mp, h, 0);
259 		if ((h = mpool_get(t->bt_mp, prevpg, 0)) == NULL)
260 			return (1);
261 	}
262 
263 
264 ret:	mpool_put(t->bt_mp, h, 0);
265 	return ((*hp = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL);
266 }
267 
268 /*
269  * __bt_bdelete --
270  *	Delete all key/data pairs matching the specified key.
271  *
272  * Parameters:
273  *	  t:	tree
274  *	key:	key to delete
275  *
276  * Returns:
277  *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
278  */
279 static int
280 __bt_bdelete(BTREE *t, const DBT *key)
281 {
282 	EPG *e;
283 	PAGE *h;
284 	int deleted, exact, redo;
285 
286 	deleted = 0;
287 
288 	/* Find any matching record; __bt_search pins the page. */
289 loop:	if ((e = __bt_search(t, key, &exact)) == NULL)
290 		return (deleted ? RET_SUCCESS : RET_ERROR);
291 	if (!exact) {
292 		mpool_put(t->bt_mp, e->page, 0);
293 		return (deleted ? RET_SUCCESS : RET_SPECIAL);
294 	}
295 
296 	/*
297 	 * Delete forward, then delete backward, from the found key.  If
298 	 * there are duplicates and we reach either side of the page, do
299 	 * the key search again, so that we get them all.
300 	 */
301 	redo = 0;
302 	h = e->page;
303 	do {
304 		if (__bt_dleaf(t, key, h, e->index)) {
305 			mpool_put(t->bt_mp, h, 0);
306 			return (RET_ERROR);
307 		}
308 		if (F_ISSET(t, B_NODUPS)) {
309 			if (NEXTINDEX(h) == 0) {
310 				if (__bt_pdelete(t, h))
311 					return (RET_ERROR);
312 			} else
313 				mpool_put(t->bt_mp, h, MPOOL_DIRTY);
314 			return (RET_SUCCESS);
315 		}
316 		deleted = 1;
317 	} while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0);
318 
319 	/* Check for right-hand edge of the page. */
320 	if (e->index == NEXTINDEX(h))
321 		redo = 1;
322 
323 	/* Delete from the key to the beginning of the page. */
324 	while (e->index-- > 0) {
325 		if (__bt_cmp(t, key, e) != 0)
326 			break;
327 		if (__bt_dleaf(t, key, h, e->index) == RET_ERROR) {
328 			mpool_put(t->bt_mp, h, 0);
329 			return (RET_ERROR);
330 		}
331 		if (e->index == 0)
332 			redo = 1;
333 	}
334 
335 	/* Check for an empty page. */
336 	if (NEXTINDEX(h) == 0) {
337 		if (__bt_pdelete(t, h))
338 			return (RET_ERROR);
339 		goto loop;
340 	}
341 
342 	/* Put the page. */
343 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
344 
345 	if (redo)
346 		goto loop;
347 	return (RET_SUCCESS);
348 }
349 
350 /*
351  * __bt_pdelete --
352  *	Delete a single page from the tree.
353  *
354  * Parameters:
355  *	t:	tree
356  *	h:	leaf page
357  *
358  * Returns:
359  *	RET_SUCCESS, RET_ERROR.
360  *
361  * Side-effects:
362  *	mpool_put's the page
363  */
364 static int
365 __bt_pdelete(BTREE *t, PAGE *h)
366 {
367 	BINTERNAL *bi;
368 	PAGE *pg;
369 	EPGNO *parent;
370 	indx_t cnt, idx, *ip, offset;
371 	uint32_t nksize;
372 	char *from;
373 
374 	/*
375 	 * Walk the parent page stack -- a LIFO stack of the pages that were
376 	 * traversed when we searched for the page where the delete occurred.
377 	 * Each stack entry is a page number and a page index offset.  The
378 	 * offset is for the page traversed on the search.  We've just deleted
379 	 * a page, so we have to delete the key from the parent page.
380 	 *
381 	 * If the delete from the parent page makes it empty, this process may
382 	 * continue all the way up the tree.  We stop if we reach the root page
383 	 * (which is never deleted, it's just not worth the effort) or if the
384 	 * delete does not empty the page.
385 	 */
386 	while ((parent = BT_POP(t)) != NULL) {
387 		/* Get the parent page. */
388 		if ((pg = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
389 			return (RET_ERROR);
390 
391 		idx = parent->index;
392 		bi = GETBINTERNAL(pg, idx);
393 
394 		/* Free any overflow pages. */
395 		if (bi->flags & P_BIGKEY &&
396 		    __ovfl_delete(t, bi->bytes) == RET_ERROR) {
397 			mpool_put(t->bt_mp, pg, 0);
398 			return (RET_ERROR);
399 		}
400 
401 		/*
402 		 * Free the parent if it has only the one key and it's not the
403 		 * root page. If it's the rootpage, turn it back into an empty
404 		 * leaf page.
405 		 */
406 		if (NEXTINDEX(pg) == 1) {
407 			if (pg->pgno == P_ROOT) {
408 				pg->lower = BTDATAOFF;
409 				pg->upper = t->bt_psize;
410 				pg->flags = P_BLEAF;
411 			} else {
412 				if (__bt_relink(t, pg) || __bt_free(t, pg))
413 					return (RET_ERROR);
414 				continue;
415 			}
416 		} else {
417 			/* Pack remaining key items at the end of the page. */
418 			nksize = NBINTERNAL(bi->ksize);
419 			from = (char *)pg + pg->upper;
420 			memmove(from + nksize, from, (char *)bi - from);
421 			pg->upper += nksize;
422 
423 			/* Adjust indices' offsets, shift the indices down. */
424 			offset = pg->linp[idx];
425 			for (cnt = idx, ip = &pg->linp[0]; cnt--; ++ip)
426 				if (ip[0] < offset)
427 					ip[0] += nksize;
428 			for (cnt = NEXTINDEX(pg) - idx; --cnt; ++ip)
429 				ip[0] = ip[1] < offset ? ip[1] + nksize : ip[1];
430 			pg->lower -= sizeof(indx_t);
431 		}
432 
433 		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
434 		break;
435 	}
436 
437 	/* Free the leaf page, as long as it wasn't the root. */
438 	if (h->pgno == P_ROOT) {
439 		mpool_put(t->bt_mp, h, MPOOL_DIRTY);
440 		return (RET_SUCCESS);
441 	}
442 	return (__bt_relink(t, h) || __bt_free(t, h));
443 }
444 
445 /*
446  * __bt_dleaf --
447  *	Delete a single record from a leaf page.
448  *
449  * Parameters:
450  *	t:	tree
451  *    key:	referenced key
452  *	h:	page
453  *	idx:	index on page to delete
454  *
455  * Returns:
456  *	RET_SUCCESS, RET_ERROR.
457  */
458 int
459 __bt_dleaf(BTREE *t, const DBT *key, PAGE *h, unsigned int idx)
460 {
461 	BLEAF *bl;
462 	indx_t cnt, *ip, offset;
463 	uint32_t nbytes;
464 	void *to;
465 	char *from;
466 
467 	/* If this record is referenced by the cursor, delete the cursor. */
468 	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
469 	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
470 	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index == idx &&
471 	    __bt_curdel(t, key, h, idx))
472 		return (RET_ERROR);
473 
474 	/* If the entry uses overflow pages, make them available for reuse. */
475 	to = bl = GETBLEAF(h, idx);
476 	if (bl->flags & P_BIGKEY && __ovfl_delete(t, bl->bytes) == RET_ERROR)
477 		return (RET_ERROR);
478 	if (bl->flags & P_BIGDATA &&
479 	    __ovfl_delete(t, bl->bytes + bl->ksize) == RET_ERROR)
480 		return (RET_ERROR);
481 
482 	/* Pack the remaining key/data items at the end of the page. */
483 	nbytes = NBLEAF(bl);
484 	from = (char *)h + h->upper;
485 	memmove(from + nbytes, from, (char *)to - from);
486 	h->upper += nbytes;
487 
488 	/* Adjust the indices' offsets, shift the indices down. */
489 	offset = h->linp[idx];
490 	for (cnt = idx, ip = &h->linp[0]; cnt--; ++ip)
491 		if (ip[0] < offset)
492 			ip[0] += nbytes;
493 	for (cnt = NEXTINDEX(h) - idx; --cnt; ++ip)
494 		ip[0] = ip[1] < offset ? ip[1] + nbytes : ip[1];
495 	h->lower -= sizeof(indx_t);
496 
497 	/* If the cursor is on this page, adjust it as necessary. */
498 	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
499 	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
500 	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index > idx)
501 		--t->bt_cursor.pg.index;
502 
503 	return (RET_SUCCESS);
504 }
505 
506 /*
507  * __bt_curdel --
508  *	Delete the cursor.
509  *
510  * Parameters:
511  *	t:	tree
512  *    key:	referenced key (or NULL)
513  *	h:	page
514  *    idx:	index on page to delete
515  *
516  * Returns:
517  *	RET_SUCCESS, RET_ERROR.
518  */
519 static int
520 __bt_curdel(BTREE *t, const DBT *key, PAGE *h, unsigned int idx)
521 {
522 	CURSOR *c;
523 	EPG e;
524 	PAGE *pg;
525 	int curcopy, status;
526 
527 	/*
528 	 * If there are duplicates, move forward or backward to one.
529 	 * Otherwise, copy the key into the cursor area.
530 	 */
531 	c = &t->bt_cursor;
532 	F_CLR(c, CURS_AFTER | CURS_BEFORE | CURS_ACQUIRE);
533 
534 	curcopy = 0;
535 	if (!F_ISSET(t, B_NODUPS)) {
536 		/*
537 		 * We're going to have to do comparisons.  If we weren't
538 		 * provided a copy of the key, i.e. the user is deleting
539 		 * the current cursor position, get one.
540 		 */
541 		if (key == NULL) {
542 			e.page = h;
543 			e.index = idx;
544 			if ((status = __bt_ret(t, &e,
545 			    &c->key, &c->key, NULL, NULL, 1)) != RET_SUCCESS)
546 				return (status);
547 			curcopy = 1;
548 			key = &c->key;
549 		}
550 		/* Check previous key, if not at the beginning of the page. */
551 		if (idx > 0) {
552 			e.page = h;
553 			e.index = idx - 1;
554 			if (__bt_cmp(t, key, &e) == 0) {
555 				F_SET(c, CURS_BEFORE);
556 				goto dup2;
557 			}
558 		}
559 		/* Check next key, if not at the end of the page. */
560 		if (idx < NEXTINDEX(h) - 1) {
561 			e.page = h;
562 			e.index = idx + 1;
563 			if (__bt_cmp(t, key, &e) == 0) {
564 				F_SET(c, CURS_AFTER);
565 				goto dup2;
566 			}
567 		}
568 		/* Check previous key if at the beginning of the page. */
569 		if (idx == 0 && h->prevpg != P_INVALID) {
570 			if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
571 				return (RET_ERROR);
572 			e.page = pg;
573 			e.index = NEXTINDEX(pg) - 1;
574 			if (__bt_cmp(t, key, &e) == 0) {
575 				F_SET(c, CURS_BEFORE);
576 				goto dup1;
577 			}
578 			mpool_put(t->bt_mp, pg, 0);
579 		}
580 		/* Check next key if at the end of the page. */
581 		if (idx == NEXTINDEX(h) - 1 && h->nextpg != P_INVALID) {
582 			if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
583 				return (RET_ERROR);
584 			e.page = pg;
585 			e.index = 0;
586 			if (__bt_cmp(t, key, &e) == 0) {
587 				F_SET(c, CURS_AFTER);
588 dup1:				mpool_put(t->bt_mp, pg, 0);
589 dup2:				c->pg.pgno = e.page->pgno;
590 				c->pg.index = e.index;
591 				return (RET_SUCCESS);
592 			}
593 			mpool_put(t->bt_mp, pg, 0);
594 		}
595 	}
596 	e.page = h;
597 	e.index = idx;
598 	if (curcopy || (status =
599 	    __bt_ret(t, &e, &c->key, &c->key, NULL, NULL, 1)) == RET_SUCCESS) {
600 		F_SET(c, CURS_ACQUIRE);
601 		return (RET_SUCCESS);
602 	}
603 	return (status);
604 }
605 
606 /*
607  * __bt_relink --
608  *	Link around a deleted page.
609  *
610  * Parameters:
611  *	t:	tree
612  *	h:	page to be deleted
613  */
614 static int
615 __bt_relink(BTREE *t, PAGE *h)
616 {
617 	PAGE *pg;
618 
619 	if (h->nextpg != P_INVALID) {
620 		if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
621 			return (RET_ERROR);
622 		pg->prevpg = h->prevpg;
623 		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
624 	}
625 	if (h->prevpg != P_INVALID) {
626 		if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
627 			return (RET_ERROR);
628 		pg->nextpg = h->nextpg;
629 		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
630 	}
631 	return (0);
632 }
633