xref: /dragonfly/lib/libc/db/btree/bt_delete.c (revision e3146d3a)
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  * $DragonFly: src/lib/libc/db/btree/bt_delete.c,v 1.7 2005/09/19 09:20:37 asmodai Exp $
34  */
35 
36 #include <sys/types.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 *, u_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(dbp, key, flags)
59 	const DB *dbp;
60 	const DBT *key;
61 	u_int flags;
62 {
63 	BTREE *t;
64 	CURSOR *c;
65 	PAGE *h;
66 	int status;
67 
68 	t = dbp->internal;
69 
70 	/* Toss any page pinned across calls. */
71 	if (t->bt_pinned != NULL) {
72 		mpool_put(t->bt_mp, t->bt_pinned, 0);
73 		t->bt_pinned = NULL;
74 	}
75 
76 	/* Check for change to a read-only tree. */
77 	if (F_ISSET(t, B_RDONLY)) {
78 		errno = EPERM;
79 		return (RET_ERROR);
80 	}
81 
82 	switch (flags) {
83 	case 0:
84 		status = __bt_bdelete(t, key);
85 		break;
86 	case R_CURSOR:
87 		/*
88 		 * If flags is R_CURSOR, delete the cursor.  Must already
89 		 * have started a scan and not have already deleted it.
90 		 */
91 		c = &t->bt_cursor;
92 		if (F_ISSET(c, CURS_INIT)) {
93 			if (F_ISSET(c, CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
94 				return (RET_SPECIAL);
95 			if ((h = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL)
96 				return (RET_ERROR);
97 
98 			/*
99 			 * If the page is about to be emptied, we'll need to
100 			 * delete it, which means we have to acquire a stack.
101 			 */
102 			if (NEXTINDEX(h) == 1)
103 				if (__bt_stkacq(t, &h, &t->bt_cursor))
104 					return (RET_ERROR);
105 
106 			status = __bt_dleaf(t, NULL, h, c->pg.index);
107 
108 			if (NEXTINDEX(h) == 0 && status == RET_SUCCESS) {
109 				if (__bt_pdelete(t, h))
110 					return (RET_ERROR);
111 			} else
112 				mpool_put(t->bt_mp,
113 				    h, status == RET_SUCCESS ? MPOOL_DIRTY : 0);
114 			break;
115 		}
116 		/* FALLTHROUGH */
117 	default:
118 		errno = EINVAL;
119 		return (RET_ERROR);
120 	}
121 	if (status == RET_SUCCESS)
122 		F_SET(t, B_MODIFIED);
123 	return (status);
124 }
125 
126 /*
127  * __bt_stkacq --
128  *	Acquire a stack so we can delete a cursor entry.
129  *
130  * Parameters:
131  *	  t:	tree
132  *	 hp:	pointer to current, pinned PAGE pointer
133  *	  c:	pointer to the cursor
134  *
135  * Returns:
136  *	0 on success, 1 on failure
137  */
138 static int
139 __bt_stkacq(t, hp, c)
140 	BTREE *t;
141 	PAGE **hp;
142 	CURSOR *c;
143 {
144 	BINTERNAL *bi;
145 	EPG *e;
146 	EPGNO *parent;
147 	PAGE *h;
148 	indx_t idx = 0;
149 	pgno_t pgno;
150 	recno_t nextpg, prevpg;
151 	int exact, level;
152 
153 	/*
154 	 * Find the first occurrence of the key in the tree.  Toss the
155 	 * currently locked page so we don't hit an already-locked page.
156 	 */
157 	h = *hp;
158 	mpool_put(t->bt_mp, h, 0);
159 	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
160 		return (1);
161 	h = e->page;
162 
163 	/* See if we got it in one shot. */
164 	if (h->pgno == c->pg.pgno)
165 		goto ret;
166 
167 	/*
168 	 * Move right, looking for the page.  At each move we have to move
169 	 * up the stack until we don't have to move to the next page.  If
170 	 * we have to change pages at an internal level, we have to fix the
171 	 * stack back up.
172 	 */
173 	while (h->pgno != c->pg.pgno) {
174 		if ((nextpg = h->nextpg) == P_INVALID)
175 			break;
176 		mpool_put(t->bt_mp, h, 0);
177 
178 		/* Move up the stack. */
179 		for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
180 			/* Get the parent page. */
181 			if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
182 				return (1);
183 
184 			/* Move to the next index. */
185 			if (parent->index != NEXTINDEX(h) - 1) {
186 				idx = parent->index + 1;
187 				BT_PUSH(t, h->pgno, idx);
188 				break;
189 			}
190 			mpool_put(t->bt_mp, h, 0);
191 		}
192 
193 		/* Restore the stack. */
194 		while (level--) {
195 			/* Push the next level down onto the stack. */
196 			bi = GETBINTERNAL(h, idx);
197 			pgno = bi->pgno;
198 			BT_PUSH(t, pgno, 0);
199 
200 			/* Lose the currently pinned page. */
201 			mpool_put(t->bt_mp, h, 0);
202 
203 			/* Get the next level down. */
204 			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
205 				return (1);
206 			idx = 0;
207 		}
208 		mpool_put(t->bt_mp, h, 0);
209 		if ((h = mpool_get(t->bt_mp, nextpg, 0)) == NULL)
210 			return (1);
211 	}
212 
213 	if (h->pgno == c->pg.pgno)
214 		goto ret;
215 
216 	/* Reacquire the original stack. */
217 	mpool_put(t->bt_mp, h, 0);
218 	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
219 		return (1);
220 	h = e->page;
221 
222 	/*
223 	 * Move left, looking for the page.  At each move we have to move
224 	 * up the stack until we don't have to change pages to move to the
225 	 * next page.  If we have to change pages at an internal level, we
226 	 * have to fix the stack back up.
227 	 */
228 	while (h->pgno != c->pg.pgno) {
229 		if ((prevpg = h->prevpg) == P_INVALID)
230 			break;
231 		mpool_put(t->bt_mp, h, 0);
232 
233 		/* Move up the stack. */
234 		for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
235 			/* Get the parent page. */
236 			if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
237 				return (1);
238 
239 			/* Move to the next index. */
240 			if (parent->index != 0) {
241 				idx = parent->index - 1;
242 				BT_PUSH(t, h->pgno, idx);
243 				break;
244 			}
245 			mpool_put(t->bt_mp, h, 0);
246 		}
247 
248 		/* Restore the stack. */
249 		while (level--) {
250 			/* Push the next level down onto the stack. */
251 			bi = GETBINTERNAL(h, idx);
252 			pgno = bi->pgno;
253 
254 			/* Lose the currently pinned page. */
255 			mpool_put(t->bt_mp, h, 0);
256 
257 			/* Get the next level down. */
258 			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
259 				return (1);
260 
261 			idx = NEXTINDEX(h) - 1;
262 			BT_PUSH(t, pgno, idx);
263 		}
264 		mpool_put(t->bt_mp, h, 0);
265 		if ((h = mpool_get(t->bt_mp, prevpg, 0)) == NULL)
266 			return (1);
267 	}
268 
269 
270 ret:	mpool_put(t->bt_mp, h, 0);
271 	return ((*hp = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL);
272 }
273 
274 /*
275  * __bt_bdelete --
276  *	Delete all key/data pairs matching the specified key.
277  *
278  * Parameters:
279  *	  t:	tree
280  *	key:	key to delete
281  *
282  * Returns:
283  *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found.
284  */
285 static int
286 __bt_bdelete(t, key)
287 	BTREE *t;
288 	const DBT *key;
289 {
290 	EPG *e;
291 	PAGE *h;
292 	int deleted, exact, redo;
293 
294 	deleted = 0;
295 
296 	/* Find any matching record; __bt_search pins the page. */
297 loop:	if ((e = __bt_search(t, key, &exact)) == NULL)
298 		return (deleted ? RET_SUCCESS : RET_ERROR);
299 	if (!exact) {
300 		mpool_put(t->bt_mp, e->page, 0);
301 		return (deleted ? RET_SUCCESS : RET_SPECIAL);
302 	}
303 
304 	/*
305 	 * Delete forward, then delete backward, from the found key.  If
306 	 * there are duplicates and we reach either side of the page, do
307 	 * the key search again, so that we get them all.
308 	 */
309 	redo = 0;
310 	h = e->page;
311 	do {
312 		if (__bt_dleaf(t, key, h, e->index)) {
313 			mpool_put(t->bt_mp, h, 0);
314 			return (RET_ERROR);
315 		}
316 		if (F_ISSET(t, B_NODUPS)) {
317 			if (NEXTINDEX(h) == 0) {
318 				if (__bt_pdelete(t, h))
319 					return (RET_ERROR);
320 			} else
321 				mpool_put(t->bt_mp, h, MPOOL_DIRTY);
322 			return (RET_SUCCESS);
323 		}
324 		deleted = 1;
325 	} while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0);
326 
327 	/* Check for right-hand edge of the page. */
328 	if (e->index == NEXTINDEX(h))
329 		redo = 1;
330 
331 	/* Delete from the key to the beginning of the page. */
332 	while (e->index-- > 0) {
333 		if (__bt_cmp(t, key, e) != 0)
334 			break;
335 		if (__bt_dleaf(t, key, h, e->index) == RET_ERROR) {
336 			mpool_put(t->bt_mp, h, 0);
337 			return (RET_ERROR);
338 		}
339 		if (e->index == 0)
340 			redo = 1;
341 	}
342 
343 	/* Check for an empty page. */
344 	if (NEXTINDEX(h) == 0) {
345 		if (__bt_pdelete(t, h))
346 			return (RET_ERROR);
347 		goto loop;
348 	}
349 
350 	/* Put the page. */
351 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
352 
353 	if (redo)
354 		goto loop;
355 	return (RET_SUCCESS);
356 }
357 
358 /*
359  * __bt_pdelete --
360  *	Delete a single page from the tree.
361  *
362  * Parameters:
363  *	t:	tree
364  *	h:	leaf page
365  *
366  * Returns:
367  *	RET_SUCCESS, RET_ERROR.
368  *
369  * Side-effects:
370  *	mpool_put's the page
371  */
372 static int
373 __bt_pdelete(t, h)
374 	BTREE *t;
375 	PAGE *h;
376 {
377 	BINTERNAL *bi;
378 	PAGE *pg;
379 	EPGNO *parent;
380 	indx_t cnt, idx, *ip, offset;
381 	u_int32_t nksize;
382 	char *from;
383 
384 	/*
385 	 * Walk the parent page stack -- a LIFO stack of the pages that were
386 	 * traversed when we searched for the page where the delete occurred.
387 	 * Each stack entry is a page number and a page index offset.  The
388 	 * offset is for the page traversed on the search.  We've just deleted
389 	 * a page, so we have to delete the key from the parent page.
390 	 *
391 	 * If the delete from the parent page makes it empty, this process may
392 	 * continue all the way up the tree.  We stop if we reach the root page
393 	 * (which is never deleted, it's just not worth the effort) or if the
394 	 * delete does not empty the page.
395 	 */
396 	while ((parent = BT_POP(t)) != NULL) {
397 		/* Get the parent page. */
398 		if ((pg = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
399 			return (RET_ERROR);
400 
401 		idx = parent->index;
402 		bi = GETBINTERNAL(pg, idx);
403 
404 		/* Free any overflow pages. */
405 		if (bi->flags & P_BIGKEY &&
406 		    __ovfl_delete(t, bi->bytes) == RET_ERROR) {
407 			mpool_put(t->bt_mp, pg, 0);
408 			return (RET_ERROR);
409 		}
410 
411 		/*
412 		 * Free the parent if it has only the one key and it's not the
413 		 * root page. If it's the rootpage, turn it back into an empty
414 		 * leaf page.
415 		 */
416 		if (NEXTINDEX(pg) == 1)
417 			if (pg->pgno == P_ROOT) {
418 				pg->lower = BTDATAOFF;
419 				pg->upper = t->bt_psize;
420 				pg->flags = P_BLEAF;
421 			} else {
422 				if (__bt_relink(t, pg) || __bt_free(t, pg))
423 					return (RET_ERROR);
424 				continue;
425 			}
426 		else {
427 			/* Pack remaining key items at the end of the page. */
428 			nksize = NBINTERNAL(bi->ksize);
429 			from = (char *)pg + pg->upper;
430 			memmove(from + nksize, from, (char *)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  */
468 int
469 __bt_dleaf(BTREE *t, const DBT *key, PAGE *h, u_int idx)
470 {
471 	BLEAF *bl;
472 	indx_t cnt, *ip, offset;
473 	u_int32_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 *)h + h->upper;
495 	memmove(from + nbytes, from, (char *)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  */
529 static 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);
598 dup1:				mpool_put(t->bt_mp, pg, 0);
599 dup2:				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  */
624 static int
625 __bt_relink(t, h)
626 	BTREE *t;
627 	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