xref: /freebsd/lib/libc/db/btree/bt_split.c (revision 3494f7c0)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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/param.h>
36 
37 #include <limits.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 
42 #include <db.h>
43 #include "btree.h"
44 
45 static int	 bt_broot(BTREE *, PAGE *, PAGE *, PAGE *);
46 static PAGE	*bt_page(BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t);
47 static int	 bt_preserve(BTREE *, pgno_t);
48 static PAGE	*bt_psplit(BTREE *, PAGE *, PAGE *, PAGE *, indx_t *, size_t);
49 static PAGE	*bt_root(BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t);
50 static int	 bt_rroot(BTREE *, PAGE *, PAGE *, PAGE *);
51 static recno_t	 rec_total(PAGE *);
52 
53 #ifdef STATISTICS
54 u_long	bt_rootsplit, bt_split, bt_sortsplit, bt_pfxsaved;
55 #endif
56 
57 /*
58  * __BT_SPLIT -- Split the tree.
59  *
60  * Parameters:
61  *	t:	tree
62  *	sp:	page to split
63  *	key:	key to insert
64  *	data:	data to insert
65  *	flags:	BIGKEY/BIGDATA flags
66  *	ilen:	insert length
67  *	skip:	index to leave open
68  *
69  * Returns:
70  *	RET_ERROR, RET_SUCCESS
71  */
72 int
73 __bt_split(BTREE *t, PAGE *sp, const DBT *key, const DBT *data, int flags,
74     size_t ilen, u_int32_t argskip)
75 {
76 	BINTERNAL *bi;
77 	BLEAF *bl, *tbl;
78 	DBT a, b;
79 	EPGNO *parent;
80 	PAGE *h, *l, *r, *lchild, *rchild;
81 	indx_t nxtindex;
82 	u_int16_t skip;
83 	u_int32_t n, nbytes, nksize;
84 	int parentsplit;
85 	char *dest;
86 
87 	/*
88 	 * Split the page into two pages, l and r.  The split routines return
89 	 * a pointer to the page into which the key should be inserted and with
90 	 * skip set to the offset which should be used.  Additionally, l and r
91 	 * are pinned.
92 	 */
93 	skip = argskip;
94 	h = sp->pgno == P_ROOT ?
95 	    bt_root(t, sp, &l, &r, &skip, ilen) :
96 	    bt_page(t, sp, &l, &r, &skip, ilen);
97 	if (h == NULL)
98 		return (RET_ERROR);
99 
100 	/*
101 	 * Insert the new key/data pair into the leaf page.  (Key inserts
102 	 * always cause a leaf page to split first.)
103 	 */
104 	h->linp[skip] = h->upper -= ilen;
105 	dest = (char *)h + h->upper;
106 	if (F_ISSET(t, R_RECNO))
107 		WR_RLEAF(dest, data, flags)
108 	else
109 		WR_BLEAF(dest, key, data, flags)
110 
111 	/* If the root page was split, make it look right. */
112 	if (sp->pgno == P_ROOT &&
113 	    (F_ISSET(t, R_RECNO) ?
114 	    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
115 		goto err2;
116 
117 	/*
118 	 * Now we walk the parent page stack -- a LIFO stack of the pages that
119 	 * were traversed when we searched for the page that split.  Each stack
120 	 * entry is a page number and a page index offset.  The offset is for
121 	 * the page traversed on the search.  We've just split a page, so we
122 	 * have to insert a new key into the parent page.
123 	 *
124 	 * If the insert into the parent page causes it to split, may have to
125 	 * continue splitting all the way up the tree.  We stop if the root
126 	 * splits or the page inserted into didn't have to split to hold the
127 	 * new key.  Some algorithms replace the key for the old page as well
128 	 * as the new page.  We don't, as there's no reason to believe that the
129 	 * first key on the old page is any better than the key we have, and,
130 	 * in the case of a key being placed at index 0 causing the split, the
131 	 * key is unavailable.
132 	 *
133 	 * There are a maximum of 5 pages pinned at any time.  We keep the left
134 	 * and right pages pinned while working on the parent.   The 5 are the
135 	 * two children, left parent and right parent (when the parent splits)
136 	 * and the root page or the overflow key page when calling bt_preserve.
137 	 * This code must make sure that all pins are released other than the
138 	 * root page or overflow page which is unlocked elsewhere.
139 	 */
140 	while ((parent = BT_POP(t)) != NULL) {
141 		lchild = l;
142 		rchild = r;
143 
144 		/* Get the parent page. */
145 		if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
146 			goto err2;
147 
148 		/*
149 		 * The new key goes ONE AFTER the index, because the split
150 		 * was to the right.
151 		 */
152 		skip = parent->index + 1;
153 
154 		/*
155 		 * Calculate the space needed on the parent page.
156 		 *
157 		 * Prefix trees: space hack when inserting into BINTERNAL
158 		 * pages.  Retain only what's needed to distinguish between
159 		 * the new entry and the LAST entry on the page to its left.
160 		 * If the keys compare equal, retain the entire key.  Note,
161 		 * we don't touch overflow keys, and the entire key must be
162 		 * retained for the next-to-left most key on the leftmost
163 		 * page of each level, or the search will fail.  Applicable
164 		 * ONLY to internal pages that have leaf pages as children.
165 		 * Further reduction of the key between pairs of internal
166 		 * pages loses too much information.
167 		 */
168 		switch (rchild->flags & P_TYPE) {
169 		case P_BINTERNAL:
170 			bi = GETBINTERNAL(rchild, 0);
171 			nbytes = NBINTERNAL(bi->ksize);
172 			break;
173 		case P_BLEAF:
174 			bl = GETBLEAF(rchild, 0);
175 			nbytes = NBINTERNAL(bl->ksize);
176 			if (t->bt_pfx && !(bl->flags & P_BIGKEY) &&
177 			    (h->prevpg != P_INVALID || skip > 1)) {
178 				tbl = GETBLEAF(lchild, NEXTINDEX(lchild) - 1);
179 				a.size = tbl->ksize;
180 				a.data = tbl->bytes;
181 				b.size = bl->ksize;
182 				b.data = bl->bytes;
183 				nksize = t->bt_pfx(&a, &b);
184 				n = NBINTERNAL(nksize);
185 				if (n < nbytes) {
186 #ifdef STATISTICS
187 					bt_pfxsaved += nbytes - n;
188 #endif
189 					nbytes = n;
190 				} else
191 					nksize = 0;
192 			} else
193 				nksize = 0;
194 			break;
195 		case P_RINTERNAL:
196 		case P_RLEAF:
197 			nbytes = NRINTERNAL;
198 			break;
199 		default:
200 			abort();
201 		}
202 
203 		/* Split the parent page if necessary or shift the indices. */
204 		if ((u_int32_t)(h->upper - h->lower) < nbytes + sizeof(indx_t)) {
205 			sp = h;
206 			h = h->pgno == P_ROOT ?
207 			    bt_root(t, h, &l, &r, &skip, nbytes) :
208 			    bt_page(t, h, &l, &r, &skip, nbytes);
209 			if (h == NULL)
210 				goto err1;
211 			parentsplit = 1;
212 		} else {
213 			if (skip < (nxtindex = NEXTINDEX(h)))
214 				memmove(h->linp + skip + 1, h->linp + skip,
215 				    (nxtindex - skip) * sizeof(indx_t));
216 			h->lower += sizeof(indx_t);
217 			parentsplit = 0;
218 		}
219 
220 		/* Insert the key into the parent page. */
221 		switch (rchild->flags & P_TYPE) {
222 		case P_BINTERNAL:
223 			h->linp[skip] = h->upper -= nbytes;
224 			dest = (char *)h + h->linp[skip];
225 			memmove(dest, bi, nbytes);
226 			((BINTERNAL *)dest)->pgno = rchild->pgno;
227 			break;
228 		case P_BLEAF:
229 			h->linp[skip] = h->upper -= nbytes;
230 			dest = (char *)h + h->linp[skip];
231 			WR_BINTERNAL(dest, nksize ? nksize : bl->ksize,
232 			    rchild->pgno, bl->flags & P_BIGKEY);
233 			memmove(dest, bl->bytes, nksize ? nksize : bl->ksize);
234 			if (bl->flags & P_BIGKEY) {
235 				pgno_t pgno;
236 				memcpy(&pgno, bl->bytes, sizeof(pgno));
237 				if (bt_preserve(t, pgno) == RET_ERROR)
238 					goto err1;
239 			}
240 			break;
241 		case P_RINTERNAL:
242 			/*
243 			 * Update the left page count.  If split
244 			 * added at index 0, fix the correct page.
245 			 */
246 			if (skip > 0)
247 				dest = (char *)h + h->linp[skip - 1];
248 			else
249 				dest = (char *)l + l->linp[NEXTINDEX(l) - 1];
250 			((RINTERNAL *)dest)->nrecs = rec_total(lchild);
251 			((RINTERNAL *)dest)->pgno = lchild->pgno;
252 
253 			/* Update the right page count. */
254 			h->linp[skip] = h->upper -= nbytes;
255 			dest = (char *)h + h->linp[skip];
256 			((RINTERNAL *)dest)->nrecs = rec_total(rchild);
257 			((RINTERNAL *)dest)->pgno = rchild->pgno;
258 			break;
259 		case P_RLEAF:
260 			/*
261 			 * Update the left page count.  If split
262 			 * added at index 0, fix the correct page.
263 			 */
264 			if (skip > 0)
265 				dest = (char *)h + h->linp[skip - 1];
266 			else
267 				dest = (char *)l + l->linp[NEXTINDEX(l) - 1];
268 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(lchild);
269 			((RINTERNAL *)dest)->pgno = lchild->pgno;
270 
271 			/* Update the right page count. */
272 			h->linp[skip] = h->upper -= nbytes;
273 			dest = (char *)h + h->linp[skip];
274 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(rchild);
275 			((RINTERNAL *)dest)->pgno = rchild->pgno;
276 			break;
277 		default:
278 			abort();
279 		}
280 
281 		/* Unpin the held pages. */
282 		if (!parentsplit) {
283 			mpool_put(t->bt_mp, h, MPOOL_DIRTY);
284 			break;
285 		}
286 
287 		/* If the root page was split, make it look right. */
288 		if (sp->pgno == P_ROOT &&
289 		    (F_ISSET(t, R_RECNO) ?
290 		    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
291 			goto err1;
292 
293 		mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
294 		mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
295 	}
296 
297 	/* Unpin the held pages. */
298 	mpool_put(t->bt_mp, l, MPOOL_DIRTY);
299 	mpool_put(t->bt_mp, r, MPOOL_DIRTY);
300 
301 	/* Clear any pages left on the stack. */
302 	return (RET_SUCCESS);
303 
304 	/*
305 	 * If something fails in the above loop we were already walking back
306 	 * up the tree and the tree is now inconsistent.  Nothing much we can
307 	 * do about it but release any memory we're holding.
308 	 */
309 err1:	mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
310 	mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
311 
312 err2:	mpool_put(t->bt_mp, l, 0);
313 	mpool_put(t->bt_mp, r, 0);
314 	__dbpanic(t->bt_dbp);
315 	return (RET_ERROR);
316 }
317 
318 /*
319  * BT_PAGE -- Split a non-root page of a btree.
320  *
321  * Parameters:
322  *	t:	tree
323  *	h:	root page
324  *	lp:	pointer to left page pointer
325  *	rp:	pointer to right page pointer
326  *	skip:	pointer to index to leave open
327  *	ilen:	insert length
328  *
329  * Returns:
330  *	Pointer to page in which to insert or NULL on error.
331  */
332 static PAGE *
333 bt_page(BTREE *t, PAGE *h, PAGE **lp, PAGE **rp, indx_t *skip, size_t ilen)
334 {
335 	PAGE *l, *r, *tp;
336 	pgno_t npg;
337 
338 #ifdef STATISTICS
339 	++bt_split;
340 #endif
341 	/* Put the new right page for the split into place. */
342 	if ((r = __bt_new(t, &npg)) == NULL)
343 		return (NULL);
344 	r->pgno = npg;
345 	r->lower = BTDATAOFF;
346 	r->upper = t->bt_psize;
347 	r->nextpg = h->nextpg;
348 	r->prevpg = h->pgno;
349 	r->flags = h->flags & P_TYPE;
350 
351 	/*
352 	 * If we're splitting the last page on a level because we're appending
353 	 * a key to it (skip is NEXTINDEX()), it's likely that the data is
354 	 * sorted.  Adding an empty page on the side of the level is less work
355 	 * and can push the fill factor much higher than normal.  If we're
356 	 * wrong it's no big deal, we'll just do the split the right way next
357 	 * time.  It may look like it's equally easy to do a similar hack for
358 	 * reverse sorted data, that is, split the tree left, but it's not.
359 	 * Don't even try.
360 	 */
361 	if (h->nextpg == P_INVALID && *skip == NEXTINDEX(h)) {
362 #ifdef STATISTICS
363 		++bt_sortsplit;
364 #endif
365 		h->nextpg = r->pgno;
366 		r->lower = BTDATAOFF + sizeof(indx_t);
367 		*skip = 0;
368 		*lp = h;
369 		*rp = r;
370 		return (r);
371 	}
372 
373 	/* Put the new left page for the split into place. */
374 	if ((l = (PAGE *)calloc(1, t->bt_psize)) == NULL) {
375 		mpool_put(t->bt_mp, r, 0);
376 		return (NULL);
377 	}
378 	l->pgno = h->pgno;
379 	l->nextpg = r->pgno;
380 	l->prevpg = h->prevpg;
381 	l->lower = BTDATAOFF;
382 	l->upper = t->bt_psize;
383 	l->flags = h->flags & P_TYPE;
384 
385 	/* Fix up the previous pointer of the page after the split page. */
386 	if (h->nextpg != P_INVALID) {
387 		if ((tp = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL) {
388 			free(l);
389 			/* XXX mpool_free(t->bt_mp, r->pgno); */
390 			return (NULL);
391 		}
392 		tp->prevpg = r->pgno;
393 		mpool_put(t->bt_mp, tp, MPOOL_DIRTY);
394 	}
395 
396 	/*
397 	 * Split right.  The key/data pairs aren't sorted in the btree page so
398 	 * it's simpler to copy the data from the split page onto two new pages
399 	 * instead of copying half the data to the right page and compacting
400 	 * the left page in place.  Since the left page can't change, we have
401 	 * to swap the original and the allocated left page after the split.
402 	 */
403 	tp = bt_psplit(t, h, l, r, skip, ilen);
404 
405 	/* Move the new left page onto the old left page. */
406 	memmove(h, l, t->bt_psize);
407 	if (tp == l)
408 		tp = h;
409 	free(l);
410 
411 	*lp = h;
412 	*rp = r;
413 	return (tp);
414 }
415 
416 /*
417  * BT_ROOT -- Split the root page of a btree.
418  *
419  * Parameters:
420  *	t:	tree
421  *	h:	root page
422  *	lp:	pointer to left page pointer
423  *	rp:	pointer to right page pointer
424  *	skip:	pointer to index to leave open
425  *	ilen:	insert length
426  *
427  * Returns:
428  *	Pointer to page in which to insert or NULL on error.
429  */
430 static PAGE *
431 bt_root(BTREE *t, PAGE *h, PAGE **lp, PAGE **rp, indx_t *skip, size_t ilen)
432 {
433 	PAGE *l, *r, *tp;
434 	pgno_t lnpg, rnpg;
435 
436 #ifdef STATISTICS
437 	++bt_split;
438 	++bt_rootsplit;
439 #endif
440 	/* Put the new left and right pages for the split into place. */
441 	if ((l = __bt_new(t, &lnpg)) == NULL ||
442 	    (r = __bt_new(t, &rnpg)) == NULL)
443 		return (NULL);
444 	l->pgno = lnpg;
445 	r->pgno = rnpg;
446 	l->nextpg = r->pgno;
447 	r->prevpg = l->pgno;
448 	l->prevpg = r->nextpg = P_INVALID;
449 	l->lower = r->lower = BTDATAOFF;
450 	l->upper = r->upper = t->bt_psize;
451 	l->flags = r->flags = h->flags & P_TYPE;
452 
453 	/* Split the root page. */
454 	tp = bt_psplit(t, h, l, r, skip, ilen);
455 
456 	*lp = l;
457 	*rp = r;
458 	return (tp);
459 }
460 
461 /*
462  * BT_RROOT -- Fix up the recno root page after it has been split.
463  *
464  * Parameters:
465  *	t:	tree
466  *	h:	root page
467  *	l:	left page
468  *	r:	right page
469  *
470  * Returns:
471  *	RET_ERROR, RET_SUCCESS
472  */
473 static int
474 bt_rroot(BTREE *t, PAGE *h, PAGE *l, PAGE *r)
475 {
476 	char *dest;
477 
478 	/* Insert the left and right keys, set the header information. */
479 	h->linp[0] = h->upper = t->bt_psize - NRINTERNAL;
480 	dest = (char *)h + h->upper;
481 	WR_RINTERNAL(dest,
482 	    l->flags & P_RLEAF ? NEXTINDEX(l) : rec_total(l), l->pgno);
483 
484 	__PAST_END(h->linp, 1) = h->upper -= NRINTERNAL;
485 	dest = (char *)h + h->upper;
486 	WR_RINTERNAL(dest,
487 	    r->flags & P_RLEAF ? NEXTINDEX(r) : rec_total(r), r->pgno);
488 
489 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
490 
491 	/* Unpin the root page, set to recno internal page. */
492 	h->flags &= ~P_TYPE;
493 	h->flags |= P_RINTERNAL;
494 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
495 
496 	return (RET_SUCCESS);
497 }
498 
499 /*
500  * BT_BROOT -- Fix up the btree root page after it has been split.
501  *
502  * Parameters:
503  *	t:	tree
504  *	h:	root page
505  *	l:	left page
506  *	r:	right page
507  *
508  * Returns:
509  *	RET_ERROR, RET_SUCCESS
510  */
511 static int
512 bt_broot(BTREE *t, PAGE *h, PAGE *l, PAGE *r)
513 {
514 	BINTERNAL *bi;
515 	BLEAF *bl;
516 	u_int32_t nbytes;
517 	char *dest;
518 
519 	/*
520 	 * If the root page was a leaf page, change it into an internal page.
521 	 * We copy the key we split on (but not the key's data, in the case of
522 	 * a leaf page) to the new root page.
523 	 *
524 	 * The btree comparison code guarantees that the left-most key on any
525 	 * level of the tree is never used, so it doesn't need to be filled in.
526 	 */
527 	nbytes = NBINTERNAL(0);
528 	h->linp[0] = h->upper = t->bt_psize - nbytes;
529 	dest = (char *)h + h->upper;
530 	WR_BINTERNAL(dest, 0, l->pgno, 0);
531 
532 	switch (h->flags & P_TYPE) {
533 	case P_BLEAF:
534 		bl = GETBLEAF(r, 0);
535 		nbytes = NBINTERNAL(bl->ksize);
536 		__PAST_END(h->linp, 1) = h->upper -= nbytes;
537 		dest = (char *)h + h->upper;
538 		WR_BINTERNAL(dest, bl->ksize, r->pgno, 0);
539 		memmove(dest, bl->bytes, bl->ksize);
540 
541 		/*
542 		 * If the key is on an overflow page, mark the overflow chain
543 		 * so it isn't deleted when the leaf copy of the key is deleted.
544 		 */
545 	if (bl->flags & P_BIGKEY) {
546 			pgno_t pgno;
547 			memcpy(&pgno, bl->bytes, sizeof(pgno));
548 			if (bt_preserve(t, pgno) == RET_ERROR)
549 				return (RET_ERROR);
550 		}
551 		break;
552 	case P_BINTERNAL:
553 		bi = GETBINTERNAL(r, 0);
554 		nbytes = NBINTERNAL(bi->ksize);
555 		__PAST_END(h->linp, 1) = h->upper -= nbytes;
556 		dest = (char *)h + h->upper;
557 		memmove(dest, bi, nbytes);
558 		((BINTERNAL *)dest)->pgno = r->pgno;
559 		break;
560 	default:
561 		abort();
562 	}
563 
564 	/* There are two keys on the page. */
565 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
566 
567 	/* Unpin the root page, set to btree internal page. */
568 	h->flags &= ~P_TYPE;
569 	h->flags |= P_BINTERNAL;
570 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
571 
572 	return (RET_SUCCESS);
573 }
574 
575 /*
576  * BT_PSPLIT -- Do the real work of splitting the page.
577  *
578  * Parameters:
579  *	t:	tree
580  *	h:	page to be split
581  *	l:	page to put lower half of data
582  *	r:	page to put upper half of data
583  *	pskip:	pointer to index to leave open
584  *	ilen:	insert length
585  *
586  * Returns:
587  *	Pointer to page in which to insert.
588  */
589 static PAGE *
590 bt_psplit(BTREE *t, PAGE *h, PAGE *l, PAGE *r, indx_t *pskip, size_t ilen)
591 {
592 	BINTERNAL *bi;
593 	BLEAF *bl;
594 	CURSOR *c;
595 	RLEAF *rl;
596 	PAGE *rval;
597 	void *src;
598 	indx_t full, half, nxt, off, skip, top, used;
599 	u_int32_t nbytes;
600 	int bigkeycnt, isbigkey;
601 
602 	/*
603 	 * Split the data to the left and right pages.  Leave the skip index
604 	 * open.  Additionally, make some effort not to split on an overflow
605 	 * key.  This makes internal page processing faster and can save
606 	 * space as overflow keys used by internal pages are never deleted.
607 	 */
608 	bigkeycnt = 0;
609 	skip = *pskip;
610 	full = t->bt_psize - BTDATAOFF;
611 	half = full / 2;
612 	used = 0;
613 	for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) {
614 		if (skip == off) {
615 			nbytes = ilen;
616 			isbigkey = 0;		/* XXX: not really known. */
617 		} else
618 			switch (h->flags & P_TYPE) {
619 			case P_BINTERNAL:
620 				src = bi = GETBINTERNAL(h, nxt);
621 				nbytes = NBINTERNAL(bi->ksize);
622 				isbigkey = bi->flags & P_BIGKEY;
623 				break;
624 			case P_BLEAF:
625 				src = bl = GETBLEAF(h, nxt);
626 				nbytes = NBLEAF(bl);
627 				isbigkey = bl->flags & P_BIGKEY;
628 				break;
629 			case P_RINTERNAL:
630 				src = GETRINTERNAL(h, nxt);
631 				nbytes = NRINTERNAL;
632 				isbigkey = 0;
633 				break;
634 			case P_RLEAF:
635 				src = rl = GETRLEAF(h, nxt);
636 				nbytes = NRLEAF(rl);
637 				isbigkey = 0;
638 				break;
639 			default:
640 				abort();
641 			}
642 
643 		/*
644 		 * If the key/data pairs are substantial fractions of the max
645 		 * possible size for the page, it's possible to get situations
646 		 * where we decide to try and copy too much onto the left page.
647 		 * Make sure that doesn't happen.
648 		 */
649 		if ((skip <= off && used + nbytes + sizeof(indx_t) >= full) ||
650 		    nxt == top - 1) {
651 			--off;
652 			break;
653 		}
654 
655 		/* Copy the key/data pair, if not the skipped index. */
656 		if (skip != off) {
657 			++nxt;
658 
659 			l->linp[off] = l->upper -= nbytes;
660 			memmove((char *)l + l->upper, src, nbytes);
661 		}
662 
663 		used += nbytes + sizeof(indx_t);
664 		if (used >= half) {
665 			if (!isbigkey || bigkeycnt == 3)
666 				break;
667 			else
668 				++bigkeycnt;
669 		}
670 	}
671 
672 	/*
673 	 * Off is the last offset that's valid for the left page.
674 	 * Nxt is the first offset to be placed on the right page.
675 	 */
676 	l->lower += (off + 1) * sizeof(indx_t);
677 
678 	/*
679 	 * If splitting the page that the cursor was on, the cursor has to be
680 	 * adjusted to point to the same record as before the split.  If the
681 	 * cursor is at or past the skipped slot, the cursor is incremented by
682 	 * one.  If the cursor is on the right page, it is decremented by the
683 	 * number of records split to the left page.
684 	 */
685 	c = &t->bt_cursor;
686 	if (F_ISSET(c, CURS_INIT) && c->pg.pgno == h->pgno) {
687 		if (c->pg.index >= skip)
688 			++c->pg.index;
689 		if (c->pg.index < nxt)			/* Left page. */
690 			c->pg.pgno = l->pgno;
691 		else {					/* Right page. */
692 			c->pg.pgno = r->pgno;
693 			c->pg.index -= nxt;
694 		}
695 	}
696 
697 	/*
698 	 * If the skipped index was on the left page, just return that page.
699 	 * Otherwise, adjust the skip index to reflect the new position on
700 	 * the right page.
701 	 */
702 	if (skip <= off) {
703 		skip = MAX_PAGE_OFFSET;
704 		rval = l;
705 	} else {
706 		rval = r;
707 		*pskip -= nxt;
708 	}
709 
710 	for (off = 0; nxt < top; ++off) {
711 		if (skip == nxt) {
712 			++off;
713 			skip = MAX_PAGE_OFFSET;
714 		}
715 		switch (h->flags & P_TYPE) {
716 		case P_BINTERNAL:
717 			src = bi = GETBINTERNAL(h, nxt);
718 			nbytes = NBINTERNAL(bi->ksize);
719 			break;
720 		case P_BLEAF:
721 			src = bl = GETBLEAF(h, nxt);
722 			nbytes = NBLEAF(bl);
723 			break;
724 		case P_RINTERNAL:
725 			src = GETRINTERNAL(h, nxt);
726 			nbytes = NRINTERNAL;
727 			break;
728 		case P_RLEAF:
729 			src = rl = GETRLEAF(h, nxt);
730 			nbytes = NRLEAF(rl);
731 			break;
732 		default:
733 			abort();
734 		}
735 		++nxt;
736 		r->linp[off] = r->upper -= nbytes;
737 		memmove((char *)r + r->upper, src, nbytes);
738 	}
739 	r->lower += off * sizeof(indx_t);
740 
741 	/* If the key is being appended to the page, adjust the index. */
742 	if (skip == top)
743 		r->lower += sizeof(indx_t);
744 
745 	return (rval);
746 }
747 
748 /*
749  * BT_PRESERVE -- Mark a chain of pages as used by an internal node.
750  *
751  * Chains of indirect blocks pointed to by leaf nodes get reclaimed when the
752  * record that references them gets deleted.  Chains pointed to by internal
753  * pages never get deleted.  This routine marks a chain as pointed to by an
754  * internal page.
755  *
756  * Parameters:
757  *	t:	tree
758  *	pg:	page number of first page in the chain.
759  *
760  * Returns:
761  *	RET_SUCCESS, RET_ERROR.
762  */
763 static int
764 bt_preserve(BTREE *t, pgno_t pg)
765 {
766 	PAGE *h;
767 
768 	if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL)
769 		return (RET_ERROR);
770 	h->flags |= P_PRESERVE;
771 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
772 	return (RET_SUCCESS);
773 }
774 
775 /*
776  * REC_TOTAL -- Return the number of recno entries below a page.
777  *
778  * Parameters:
779  *	h:	page
780  *
781  * Returns:
782  *	The number of recno entries below a page.
783  *
784  * XXX
785  * These values could be set by the bt_psplit routine.  The problem is that the
786  * entry has to be popped off of the stack etc. or the values have to be passed
787  * all the way back to bt_split/bt_rroot and it's not very clean.
788  */
789 static recno_t
790 rec_total(PAGE *h)
791 {
792 	recno_t recs;
793 	indx_t nxt, top;
794 
795 	for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt)
796 		recs += GETRINTERNAL(h, nxt)->nrecs;
797 	return (recs);
798 }
799