xref: /illumos-gate/usr/src/cmd/sendmail/db/btree/bt_delete.c (revision 8b80e8cb6855118d46f605e91b5ed4ce83417395)
1 /*-
2  * See the file LICENSE for redistribution information.
3  *
4  * Copyright (c) 1996, 1997, 1998
5  *	Sleepycat Software.  All rights reserved.
6  */
7 /*
8  * Copyright (c) 1990, 1993, 1994, 1995, 1996
9  *	Keith Bostic.  All rights reserved.
10  */
11 /*
12  * Copyright (c) 1990, 1993, 1994, 1995
13  *	The Regents of the University of California.  All rights reserved.
14  *
15  * This code is derived from software contributed to Berkeley by
16  * Mike Olson.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  * 3. All advertising materials mentioning features or use of this software
27  *    must display the following acknowledgement:
28  *	This product includes software developed by the University of
29  *	California, Berkeley and its contributors.
30  * 4. Neither the name of the University nor the names of its contributors
31  *    may be used to endorse or promote products derived from this software
32  *    without specific prior written permission.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
35  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
36  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
37  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
38  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
39  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
40  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
41  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
42  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
43  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44  * SUCH DAMAGE.
45  */
46 
47 #include "config.h"
48 
49 #ifndef lint
50 static const char sccsid[] = "@(#)bt_delete.c	10.43 (Sleepycat) 12/7/98";
51 #endif /* not lint */
52 
53 #ifndef NO_SYSTEM_INCLUDES
54 #include <sys/types.h>
55 
56 #include <string.h>
57 #endif
58 
59 #include "db_int.h"
60 #include "db_page.h"
61 #include "btree.h"
62 
63 /*
64  * __bam_delete --
65  *	Delete the items referenced by a key.
66  *
67  * PUBLIC: int __bam_delete __P((DB *, DB_TXN *, DBT *, u_int32_t));
68  */
69 int
70 __bam_delete(dbp, txn, key, flags)
71 	DB *dbp;
72 	DB_TXN *txn;
73 	DBT *key;
74 	u_int32_t flags;
75 {
76 	DBC *dbc;
77 	DBT data;
78 	u_int32_t f_init, f_next;
79 	int ret, t_ret;
80 
81 	DB_PANIC_CHECK(dbp);
82 
83 	/* Check for invalid flags. */
84 	if ((ret =
85 	    __db_delchk(dbp, key, flags, F_ISSET(dbp, DB_AM_RDONLY))) != 0)
86 		return (ret);
87 
88 	/* Allocate a cursor. */
89 	if ((ret = dbp->cursor(dbp, txn, &dbc, DB_WRITELOCK)) != 0)
90 		return (ret);
91 
92 	DEBUG_LWRITE(dbc, txn, "bam_delete", key, NULL, flags);
93 
94 	/*
95 	 * Walk a cursor through the key/data pairs, deleting as we go.  Set
96 	 * the DB_DBT_USERMEM flag, as this might be a threaded application
97 	 * and the flags checking will catch us.  We don't actually want the
98 	 * keys or data, so request a partial of length 0.
99 	 */
100 	memset(&data, 0, sizeof(data));
101 	F_SET(&data, DB_DBT_USERMEM | DB_DBT_PARTIAL);
102 
103 	/* If locking, set read-modify-write flag. */
104 	f_init = DB_SET;
105 	f_next = DB_NEXT_DUP;
106 	if (dbp->dbenv != NULL && dbp->dbenv->lk_info != NULL) {
107 		f_init |= DB_RMW;
108 		f_next |= DB_RMW;
109 	}
110 
111 	/* Walk through the set of key/data pairs, deleting as we go. */
112 	if ((ret = dbc->c_get(dbc, key, &data, f_init)) != 0)
113 		goto err;
114 	for (;;) {
115 		if ((ret = dbc->c_del(dbc, 0)) != 0)
116 			goto err;
117 		if ((ret = dbc->c_get(dbc, key, &data, f_next)) != 0) {
118 			if (ret == DB_NOTFOUND) {
119 				ret = 0;
120 				break;
121 			}
122 			goto err;
123 		}
124 	}
125 
126 err:	/* Discard the cursor. */
127 	if ((t_ret = dbc->c_close(dbc)) != 0 &&
128 	    (ret == 0 || ret == DB_NOTFOUND))
129 		ret = t_ret;
130 
131 	return (ret);
132 }
133 
134 /*
135  * __bam_ditem --
136  *	Delete one or more entries from a page.
137  *
138  * PUBLIC: int __bam_ditem __P((DBC *, PAGE *, u_int32_t));
139  */
140 int
141 __bam_ditem(dbc, h, indx)
142 	DBC *dbc;
143 	PAGE *h;
144 	u_int32_t indx;
145 {
146 	BINTERNAL *bi;
147 	BKEYDATA *bk;
148 	BOVERFLOW *bo;
149 	DB *dbp;
150 	u_int32_t nbytes;
151 	int ret;
152 
153 	dbp = dbc->dbp;
154 
155 	switch (TYPE(h)) {
156 	case P_IBTREE:
157 		bi = GET_BINTERNAL(h, indx);
158 		switch (B_TYPE(bi->type)) {
159 		case B_DUPLICATE:
160 		case B_OVERFLOW:
161 			nbytes = BINTERNAL_SIZE(bi->len);
162 			bo = (BOVERFLOW *)bi->data;
163 			goto offpage;
164 		case B_KEYDATA:
165 			nbytes = BINTERNAL_SIZE(bi->len);
166 			break;
167 		default:
168 			return (__db_pgfmt(dbp, h->pgno));
169 		}
170 		break;
171 	case P_IRECNO:
172 		nbytes = RINTERNAL_SIZE;
173 		break;
174 	case P_LBTREE:
175 		/*
176 		 * If it's a duplicate key, discard the index and don't touch
177 		 * the actual page item.
178 		 *
179 		 * XXX
180 		 * This works because no data item can have an index matching
181 		 * any other index so even if the data item is in a key "slot",
182 		 * it won't match any other index.
183 		 */
184 		if ((indx % 2) == 0) {
185 			/*
186 			 * Check for a duplicate after us on the page.  NOTE:
187 			 * we have to delete the key item before deleting the
188 			 * data item, otherwise the "indx + P_INDX" calculation
189 			 * won't work!
190 			 */
191 			if (indx + P_INDX < (u_int32_t)NUM_ENT(h) &&
192 			    h->inp[indx] == h->inp[indx + P_INDX])
193 				return (__bam_adjindx(dbc,
194 				    h, indx, indx + O_INDX, 0));
195 			/*
196 			 * Check for a duplicate before us on the page.  It
197 			 * doesn't matter if we delete the key item before or
198 			 * after the data item for the purposes of this one.
199 			 */
200 			if (indx > 0 && h->inp[indx] == h->inp[indx - P_INDX])
201 				return (__bam_adjindx(dbc,
202 				    h, indx, indx - P_INDX, 0));
203 		}
204 		/* FALLTHROUGH */
205 	case P_LRECNO:
206 		bk = GET_BKEYDATA(h, indx);
207 		switch (B_TYPE(bk->type)) {
208 		case B_DUPLICATE:
209 		case B_OVERFLOW:
210 			nbytes = BOVERFLOW_SIZE;
211 			bo = GET_BOVERFLOW(h, indx);
212 
213 offpage:		/* Delete duplicate/offpage chains. */
214 			if (B_TYPE(bo->type) == B_DUPLICATE) {
215 				if ((ret =
216 				    __db_ddup(dbc, bo->pgno, __bam_free)) != 0)
217 					return (ret);
218 			} else
219 				if ((ret =
220 				    __db_doff(dbc, bo->pgno, __bam_free)) != 0)
221 					return (ret);
222 			break;
223 		case B_KEYDATA:
224 			nbytes = BKEYDATA_SIZE(bk->len);
225 			break;
226 		default:
227 			return (__db_pgfmt(dbp, h->pgno));
228 		}
229 		break;
230 	default:
231 		return (__db_pgfmt(dbp, h->pgno));
232 	}
233 
234 	/* Delete the item. */
235 	if ((ret = __db_ditem(dbc, h, indx, nbytes)) != 0)
236 		return (ret);
237 
238 	/* Mark the page dirty. */
239 	return (memp_fset(dbp->mpf, h, DB_MPOOL_DIRTY));
240 }
241 
242 /*
243  * __bam_adjindx --
244  *	Adjust an index on the page.
245  *
246  * PUBLIC: int __bam_adjindx __P((DBC *, PAGE *, u_int32_t, u_int32_t, int));
247  */
248 int
249 __bam_adjindx(dbc, h, indx, indx_copy, is_insert)
250 	DBC *dbc;
251 	PAGE *h;
252 	u_int32_t indx, indx_copy;
253 	int is_insert;
254 {
255 	DB *dbp;
256 	db_indx_t copy;
257 	int ret;
258 
259 	dbp = dbc->dbp;
260 
261 	/* Log the change. */
262 	if (DB_LOGGING(dbc) &&
263 	    (ret = __bam_adj_log(dbp->dbenv->lg_info, dbc->txn, &LSN(h),
264 	    0, dbp->log_fileid, PGNO(h), &LSN(h), indx, indx_copy,
265 	    (u_int32_t)is_insert)) != 0)
266 		return (ret);
267 
268 	if (is_insert) {
269 		copy = h->inp[indx_copy];
270 		if (indx != NUM_ENT(h))
271 			memmove(&h->inp[indx + O_INDX], &h->inp[indx],
272 			    sizeof(db_indx_t) * (NUM_ENT(h) - indx));
273 		h->inp[indx] = copy;
274 		++NUM_ENT(h);
275 	} else {
276 		--NUM_ENT(h);
277 		if (indx != NUM_ENT(h))
278 			memmove(&h->inp[indx], &h->inp[indx + O_INDX],
279 			    sizeof(db_indx_t) * (NUM_ENT(h) - indx));
280 	}
281 
282 	/* Mark the page dirty. */
283 	ret = memp_fset(dbp->mpf, h, DB_MPOOL_DIRTY);
284 
285 	/* Adjust the cursors. */
286 	__bam_ca_di(dbp, h->pgno, indx, is_insert ? 1 : -1);
287 	return (0);
288 }
289 
290 /*
291  * __bam_dpage --
292  *	Delete a page from the tree.
293  *
294  * PUBLIC: int __bam_dpage __P((DBC *, const DBT *));
295  */
296 int
297 __bam_dpage(dbc, key)
298 	DBC *dbc;
299 	const DBT *key;
300 {
301 	CURSOR *cp;
302 	DB *dbp;
303 	DB_LOCK lock;
304 	PAGE *h;
305 	db_pgno_t pgno;
306 	int level;		/* !!!: has to hold number of tree levels. */
307 	int exact, ret;
308 
309 	dbp = dbc->dbp;
310 	cp = dbc->internal;
311 	ret = 0;
312 
313 	/*
314 	 * The locking protocol is that we acquire locks by walking down the
315 	 * tree, to avoid the obvious deadlocks.
316 	 *
317 	 * Call __bam_search to reacquire the empty leaf page, but this time
318 	 * get both the leaf page and it's parent, locked.  Walk back up the
319 	 * tree, until we have the top pair of pages that we want to delete.
320 	 * Once we have the top page that we want to delete locked, lock the
321 	 * underlying pages and check to make sure they're still empty.  If
322 	 * they are, delete them.
323 	 */
324 	for (level = LEAFLEVEL;; ++level) {
325 		/* Acquire a page and its parent, locked. */
326 		if ((ret =
327 		    __bam_search(dbc, key, S_WRPAIR, level, NULL, &exact)) != 0)
328 			return (ret);
329 
330 		/*
331 		 * If we reach the root or the page isn't going to be empty
332 		 * when we delete one record, quit.
333 		 */
334 		h = cp->csp[-1].page;
335 		if (h->pgno == PGNO_ROOT || NUM_ENT(h) != 1)
336 			break;
337 
338 		/* Release the two locked pages. */
339 		(void)memp_fput(dbp->mpf, cp->csp[-1].page, 0);
340 		(void)__BT_TLPUT(dbc, cp->csp[-1].lock);
341 		(void)memp_fput(dbp->mpf, cp->csp[0].page, 0);
342 		(void)__BT_TLPUT(dbc, cp->csp[0].lock);
343 	}
344 
345 	/*
346 	 * Leave the stack pointer one after the last entry, we may be about
347 	 * to push more items on the stack.
348 	 */
349 	++cp->csp;
350 
351 	/*
352 	 * cp->csp[-2].page is the top page, which we're not going to delete,
353 	 * and cp->csp[-1].page is the first page we are going to delete.
354 	 *
355 	 * Walk down the chain, acquiring the rest of the pages until we've
356 	 * retrieved the leaf page.  If we find any pages that aren't going
357 	 * to be emptied by the delete, someone else added something while we
358 	 * were walking the tree, and we discontinue the delete.
359 	 */
360 	for (h = cp->csp[-1].page;;) {
361 		if (ISLEAF(h)) {
362 			if (NUM_ENT(h) != 0)
363 				goto release;
364 			break;
365 		} else
366 			if (NUM_ENT(h) != 1)
367 				goto release;
368 
369 		/*
370 		 * Get the next page, write lock it and push it onto the stack.
371 		 * We know it's index 0, because it can only have one element.
372 		 */
373 		pgno = TYPE(h) == P_IBTREE ?
374 		    GET_BINTERNAL(h, 0)->pgno : GET_RINTERNAL(h, 0)->pgno;
375 
376 		if ((ret = __bam_lget(dbc, 0, pgno, DB_LOCK_WRITE, &lock)) != 0)
377 			goto release;
378 		if ((ret = memp_fget(dbp->mpf, &pgno, 0, &h)) != 0)
379 			goto release;
380 		BT_STK_PUSH(cp, h, 0, lock, ret);
381 	}
382 
383 	/* Adjust back to reference the last page on the stack. */
384 	BT_STK_POP(cp);
385 
386 	/* Delete the pages. */
387 	return (__bam_dpages(dbc));
388 
389 release:
390 	/* Adjust back to reference the last page on the stack. */
391 	BT_STK_POP(cp);
392 
393 	/* Discard any locked pages and return. */
394 	__bam_stkrel(dbc, 0);
395 
396 	return (ret);
397 }
398 
399 /*
400  * __bam_dpages --
401  *	Delete a set of locked pages.
402  *
403  * PUBLIC: int __bam_dpages __P((DBC *));
404  */
405 int
406 __bam_dpages(dbc)
407 	DBC *dbc;
408 {
409 	CURSOR *cp;
410 	DB *dbp;
411 	DBT a, b;
412 	DB_LOCK c_lock, p_lock;
413 	EPG *epg;
414 	PAGE *child, *parent;
415 	db_indx_t nitems;
416 	db_pgno_t pgno;
417 	db_recno_t rcnt;
418 	int done, ret;
419 
420 	dbp = dbc->dbp;
421 	cp = dbc->internal;
422 	epg = cp->sp;
423 
424 	/*
425 	 * !!!
426 	 * There is an interesting deadlock situation here.  We have to relink
427 	 * the leaf page chain around the leaf page being deleted.  Consider
428 	 * a cursor walking through the leaf pages, that has the previous page
429 	 * read-locked and is waiting on a lock for the page we're deleting.
430 	 * It will deadlock here.  This is a problem, because if our process is
431 	 * selected to resolve the deadlock, we'll leave an empty leaf page
432 	 * that we can never again access by walking down the tree.  So, before
433 	 * we unlink the subtree, we relink the leaf page chain.
434 	 */
435 	if ((ret = __db_relink(dbc, DB_REM_PAGE, cp->csp->page, NULL, 1)) != 0)
436 		goto release;
437 
438 	/*
439 	 * We have the entire stack of deletable pages locked.
440 	 *
441 	 * Delete the highest page in the tree's reference to the underlying
442 	 * stack of pages.  Then, release that page, letting the rest of the
443 	 * tree get back to business.
444 	 */
445 	if ((ret = __bam_ditem(dbc, epg->page, epg->indx)) != 0) {
446 release:	(void)__bam_stkrel(dbc, 0);
447 		return (ret);
448 	}
449 
450 	pgno = epg->page->pgno;
451 	nitems = NUM_ENT(epg->page);
452 
453 	(void)memp_fput(dbp->mpf, epg->page, 0);
454 	(void)__BT_TLPUT(dbc, epg->lock);
455 
456 	/*
457 	 * Free the rest of the stack of pages.
458 	 *
459 	 * !!!
460 	 * Don't bother checking for errors.  We've unlinked the subtree from
461 	 * the tree, and there's no possibility of recovery outside of doing
462 	 * TXN rollback.
463 	 */
464 	while (++epg <= cp->csp) {
465 		/*
466 		 * Delete page entries so they will be restored as part of
467 		 * recovery.
468 		 */
469 		if (NUM_ENT(epg->page) != 0)
470 			(void)__bam_ditem(dbc, epg->page, epg->indx);
471 
472 		(void)__bam_free(dbc, epg->page);
473 		(void)__BT_TLPUT(dbc, epg->lock);
474 	}
475 	BT_STK_CLR(cp);
476 
477 	/*
478 	 * Try and collapse the tree a level -- this is only applicable
479 	 * if we've deleted the next-to-last element from the root page.
480 	 *
481 	 * There are two cases when collapsing a tree.
482 	 *
483 	 * If we've just deleted the last item from the root page, there is no
484 	 * further work to be done.  The code above has emptied the root page
485 	 * and freed all pages below it.
486 	 */
487 	if (pgno != PGNO_ROOT || nitems != 1)
488 		return (0);
489 
490 	/*
491 	 * If we just deleted the next-to-last item from the root page, the
492 	 * tree can collapse one or more levels.  While there remains only a
493 	 * single item on the root page, write lock the last page referenced
494 	 * by the root page and copy it over the root page.  If we can't get a
495 	 * write lock, that's okay, the tree just stays deeper than we'd like.
496 	 */
497 	for (done = 0; !done;) {
498 		/* Initialize. */
499 		parent = child = NULL;
500 		p_lock = c_lock = LOCK_INVALID;
501 
502 		/* Lock the root. */
503 		pgno = PGNO_ROOT;
504 		if ((ret =
505 		    __bam_lget(dbc, 0, pgno, DB_LOCK_WRITE, &p_lock)) != 0)
506 			goto stop;
507 		if ((ret = memp_fget(dbp->mpf, &pgno, 0, &parent)) != 0)
508 			goto stop;
509 
510 		if (NUM_ENT(parent) != 1 ||
511 		    (TYPE(parent) != P_IBTREE && TYPE(parent) != P_IRECNO))
512 			goto stop;
513 
514 		pgno = TYPE(parent) == P_IBTREE ?
515 		    GET_BINTERNAL(parent, 0)->pgno :
516 		    GET_RINTERNAL(parent, 0)->pgno;
517 
518 		/* Lock the child page. */
519 		if ((ret =
520 		    __bam_lget(dbc, 0, pgno, DB_LOCK_WRITE, &c_lock)) != 0)
521 			goto stop;
522 		if ((ret = memp_fget(dbp->mpf, &pgno, 0, &child)) != 0)
523 			goto stop;
524 
525 		/* Log the change. */
526 		if (DB_LOGGING(dbc)) {
527 			memset(&a, 0, sizeof(a));
528 			a.data = child;
529 			a.size = dbp->pgsize;
530 			memset(&b, 0, sizeof(b));
531 			b.data = P_ENTRY(parent, 0);
532 			b.size = BINTERNAL_SIZE(((BINTERNAL *)b.data)->len);
533 			__bam_rsplit_log(dbp->dbenv->lg_info, dbc->txn,
534 			   &child->lsn, 0, dbp->log_fileid, child->pgno, &a,
535 			   RE_NREC(parent), &b, &parent->lsn);
536 		}
537 
538 		/*
539 		 * Make the switch.
540 		 *
541 		 * One fixup -- if the tree has record numbers and we're not
542 		 * converting to a leaf page, we have to preserve the total
543 		 * record count.  Note that we are about to overwrite everything
544 		 * on the parent, including its LSN.  This is actually OK,
545 		 * because the above log message, which describes this update,
546 		 * stores its LSN on the child page.  When the child is copied
547 		 * to the parent, the correct LSN is going to copied into
548 		 * place in the parent.
549 		 */
550 		COMPQUIET(rcnt, 0);
551 		if (TYPE(child) == P_IRECNO ||
552 		    (TYPE(child) == P_IBTREE && F_ISSET(dbp, DB_BT_RECNUM)))
553 			rcnt = RE_NREC(parent);
554 		memcpy(parent, child, dbp->pgsize);
555 		parent->pgno = PGNO_ROOT;
556 		if (TYPE(child) == P_IRECNO ||
557 		    (TYPE(child) == P_IBTREE && F_ISSET(dbp, DB_BT_RECNUM)))
558 			RE_NREC_SET(parent, rcnt);
559 
560 		/* Mark the pages dirty. */
561 		memp_fset(dbp->mpf, parent, DB_MPOOL_DIRTY);
562 		memp_fset(dbp->mpf, child, DB_MPOOL_DIRTY);
563 
564 		/* Adjust the cursors. */
565 		__bam_ca_rsplit(dbp, child->pgno, PGNO_ROOT);
566 
567 		/*
568 		 * Free the page copied onto the root page and discard its
569 		 * lock.  (The call to __bam_free() discards our reference
570 		 * to the page.)
571 		 */
572 		(void)__bam_free(dbc, child);
573 		child = NULL;
574 
575 		if (0) {
576 stop:			done = 1;
577 		}
578 		if (p_lock != LOCK_INVALID)
579 			(void)__BT_TLPUT(dbc, p_lock);
580 		if (parent != NULL)
581 			memp_fput(dbp->mpf, parent, 0);
582 		if (c_lock != LOCK_INVALID)
583 			(void)__BT_TLPUT(dbc, c_lock);
584 		if (child != NULL)
585 			memp_fput(dbp->mpf, child, 0);
586 	}
587 
588 	return (0);
589 }
590