xref: /freebsd/lib/libc/db/btree/bt_split.c (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
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 #if defined(LIBC_SCCS) && !defined(lint)
36 static char sccsid[] = "@(#)bt_split.c	8.10 (Berkeley) 1/9/95";
37 #endif /* LIBC_SCCS and not lint */
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include <sys/param.h>
42 
43 #include <limits.h>
44 #include <stdio.h>
45 #include <stdlib.h>
46 #include <string.h>
47 
48 #include <db.h>
49 #include "btree.h"
50 
51 static int	 bt_broot(BTREE *, PAGE *, PAGE *, PAGE *);
52 static PAGE	*bt_page(BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t);
53 static int	 bt_preserve(BTREE *, pgno_t);
54 static PAGE	*bt_psplit(BTREE *, PAGE *, PAGE *, PAGE *, indx_t *, size_t);
55 static PAGE	*bt_root(BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t);
56 static int	 bt_rroot(BTREE *, PAGE *, PAGE *, PAGE *);
57 static recno_t	 rec_total(PAGE *);
58 
59 #ifdef STATISTICS
60 u_long	bt_rootsplit, bt_split, bt_sortsplit, bt_pfxsaved;
61 #endif
62 
63 /*
64  * __BT_SPLIT -- Split the tree.
65  *
66  * Parameters:
67  *	t:	tree
68  *	sp:	page to split
69  *	key:	key to insert
70  *	data:	data to insert
71  *	flags:	BIGKEY/BIGDATA flags
72  *	ilen:	insert length
73  *	skip:	index to leave open
74  *
75  * Returns:
76  *	RET_ERROR, RET_SUCCESS
77  */
78 int
79 __bt_split(BTREE *t, PAGE *sp, const DBT *key, const DBT *data, int flags,
80     size_t ilen, u_int32_t argskip)
81 {
82 	BINTERNAL *bi;
83 	BLEAF *bl, *tbl;
84 	DBT a, b;
85 	EPGNO *parent;
86 	PAGE *h, *l, *r, *lchild, *rchild;
87 	indx_t nxtindex;
88 	u_int16_t skip;
89 	u_int32_t n, nbytes, nksize;
90 	int parentsplit;
91 	char *dest;
92 
93 	/*
94 	 * Split the page into two pages, l and r.  The split routines return
95 	 * a pointer to the page into which the key should be inserted and with
96 	 * skip set to the offset which should be used.  Additionally, l and r
97 	 * are pinned.
98 	 */
99 	skip = argskip;
100 	h = sp->pgno == P_ROOT ?
101 	    bt_root(t, sp, &l, &r, &skip, ilen) :
102 	    bt_page(t, sp, &l, &r, &skip, ilen);
103 	if (h == NULL)
104 		return (RET_ERROR);
105 
106 	/*
107 	 * Insert the new key/data pair into the leaf page.  (Key inserts
108 	 * always cause a leaf page to split first.)
109 	 */
110 	h->linp[skip] = h->upper -= ilen;
111 	dest = (char *)h + h->upper;
112 	if (F_ISSET(t, R_RECNO))
113 		WR_RLEAF(dest, data, flags)
114 	else
115 		WR_BLEAF(dest, key, data, flags)
116 
117 	/* If the root page was split, make it look right. */
118 	if (sp->pgno == P_ROOT &&
119 	    (F_ISSET(t, R_RECNO) ?
120 	    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
121 		goto err2;
122 
123 	/*
124 	 * Now we walk the parent page stack -- a LIFO stack of the pages that
125 	 * were traversed when we searched for the page that split.  Each stack
126 	 * entry is a page number and a page index offset.  The offset is for
127 	 * the page traversed on the search.  We've just split a page, so we
128 	 * have to insert a new key into the parent page.
129 	 *
130 	 * If the insert into the parent page causes it to split, may have to
131 	 * continue splitting all the way up the tree.  We stop if the root
132 	 * splits or the page inserted into didn't have to split to hold the
133 	 * new key.  Some algorithms replace the key for the old page as well
134 	 * as the new page.  We don't, as there's no reason to believe that the
135 	 * first key on the old page is any better than the key we have, and,
136 	 * in the case of a key being placed at index 0 causing the split, the
137 	 * key is unavailable.
138 	 *
139 	 * There are a maximum of 5 pages pinned at any time.  We keep the left
140 	 * and right pages pinned while working on the parent.   The 5 are the
141 	 * two children, left parent and right parent (when the parent splits)
142 	 * and the root page or the overflow key page when calling bt_preserve.
143 	 * This code must make sure that all pins are released other than the
144 	 * root page or overflow page which is unlocked elsewhere.
145 	 */
146 	while ((parent = BT_POP(t)) != NULL) {
147 		lchild = l;
148 		rchild = r;
149 
150 		/* Get the parent page. */
151 		if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
152 			goto err2;
153 
154 		/*
155 		 * The new key goes ONE AFTER the index, because the split
156 		 * was to the right.
157 		 */
158 		skip = parent->index + 1;
159 
160 		/*
161 		 * Calculate the space needed on the parent page.
162 		 *
163 		 * Prefix trees: space hack when inserting into BINTERNAL
164 		 * pages.  Retain only what's needed to distinguish between
165 		 * the new entry and the LAST entry on the page to its left.
166 		 * If the keys compare equal, retain the entire key.  Note,
167 		 * we don't touch overflow keys, and the entire key must be
168 		 * retained for the next-to-left most key on the leftmost
169 		 * page of each level, or the search will fail.  Applicable
170 		 * ONLY to internal pages that have leaf pages as children.
171 		 * Further reduction of the key between pairs of internal
172 		 * pages loses too much information.
173 		 */
174 		switch (rchild->flags & P_TYPE) {
175 		case P_BINTERNAL:
176 			bi = GETBINTERNAL(rchild, 0);
177 			nbytes = NBINTERNAL(bi->ksize);
178 			break;
179 		case P_BLEAF:
180 			bl = GETBLEAF(rchild, 0);
181 			nbytes = NBINTERNAL(bl->ksize);
182 			if (t->bt_pfx && !(bl->flags & P_BIGKEY) &&
183 			    (h->prevpg != P_INVALID || skip > 1)) {
184 				tbl = GETBLEAF(lchild, NEXTINDEX(lchild) - 1);
185 				a.size = tbl->ksize;
186 				a.data = tbl->bytes;
187 				b.size = bl->ksize;
188 				b.data = bl->bytes;
189 				nksize = t->bt_pfx(&a, &b);
190 				n = NBINTERNAL(nksize);
191 				if (n < nbytes) {
192 #ifdef STATISTICS
193 					bt_pfxsaved += nbytes - n;
194 #endif
195 					nbytes = n;
196 				} else
197 					nksize = 0;
198 			} else
199 				nksize = 0;
200 			break;
201 		case P_RINTERNAL:
202 		case P_RLEAF:
203 			nbytes = NRINTERNAL;
204 			break;
205 		default:
206 			abort();
207 		}
208 
209 		/* Split the parent page if necessary or shift the indices. */
210 		if ((u_int32_t)(h->upper - h->lower) < nbytes + sizeof(indx_t)) {
211 			sp = h;
212 			h = h->pgno == P_ROOT ?
213 			    bt_root(t, h, &l, &r, &skip, nbytes) :
214 			    bt_page(t, h, &l, &r, &skip, nbytes);
215 			if (h == NULL)
216 				goto err1;
217 			parentsplit = 1;
218 		} else {
219 			if (skip < (nxtindex = NEXTINDEX(h)))
220 				memmove(h->linp + skip + 1, h->linp + skip,
221 				    (nxtindex - skip) * sizeof(indx_t));
222 			h->lower += sizeof(indx_t);
223 			parentsplit = 0;
224 		}
225 
226 		/* Insert the key into the parent page. */
227 		switch (rchild->flags & P_TYPE) {
228 		case P_BINTERNAL:
229 			h->linp[skip] = h->upper -= nbytes;
230 			dest = (char *)h + h->linp[skip];
231 			memmove(dest, bi, nbytes);
232 			((BINTERNAL *)dest)->pgno = rchild->pgno;
233 			break;
234 		case P_BLEAF:
235 			h->linp[skip] = h->upper -= nbytes;
236 			dest = (char *)h + h->linp[skip];
237 			WR_BINTERNAL(dest, nksize ? nksize : bl->ksize,
238 			    rchild->pgno, bl->flags & P_BIGKEY);
239 			memmove(dest, bl->bytes, nksize ? nksize : bl->ksize);
240 			if (bl->flags & P_BIGKEY) {
241 				pgno_t pgno;
242 				memcpy(&pgno, bl->bytes, sizeof(pgno));
243 				if (bt_preserve(t, pgno) == RET_ERROR)
244 					goto err1;
245 			}
246 			break;
247 		case P_RINTERNAL:
248 			/*
249 			 * Update the left page count.  If split
250 			 * added at index 0, fix the correct page.
251 			 */
252 			if (skip > 0)
253 				dest = (char *)h + h->linp[skip - 1];
254 			else
255 				dest = (char *)l + l->linp[NEXTINDEX(l) - 1];
256 			((RINTERNAL *)dest)->nrecs = rec_total(lchild);
257 			((RINTERNAL *)dest)->pgno = lchild->pgno;
258 
259 			/* Update the right page count. */
260 			h->linp[skip] = h->upper -= nbytes;
261 			dest = (char *)h + h->linp[skip];
262 			((RINTERNAL *)dest)->nrecs = rec_total(rchild);
263 			((RINTERNAL *)dest)->pgno = rchild->pgno;
264 			break;
265 		case P_RLEAF:
266 			/*
267 			 * Update the left page count.  If split
268 			 * added at index 0, fix the correct page.
269 			 */
270 			if (skip > 0)
271 				dest = (char *)h + h->linp[skip - 1];
272 			else
273 				dest = (char *)l + l->linp[NEXTINDEX(l) - 1];
274 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(lchild);
275 			((RINTERNAL *)dest)->pgno = lchild->pgno;
276 
277 			/* Update the right page count. */
278 			h->linp[skip] = h->upper -= nbytes;
279 			dest = (char *)h + h->linp[skip];
280 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(rchild);
281 			((RINTERNAL *)dest)->pgno = rchild->pgno;
282 			break;
283 		default:
284 			abort();
285 		}
286 
287 		/* Unpin the held pages. */
288 		if (!parentsplit) {
289 			mpool_put(t->bt_mp, h, MPOOL_DIRTY);
290 			break;
291 		}
292 
293 		/* If the root page was split, make it look right. */
294 		if (sp->pgno == P_ROOT &&
295 		    (F_ISSET(t, R_RECNO) ?
296 		    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
297 			goto err1;
298 
299 		mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
300 		mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
301 	}
302 
303 	/* Unpin the held pages. */
304 	mpool_put(t->bt_mp, l, MPOOL_DIRTY);
305 	mpool_put(t->bt_mp, r, MPOOL_DIRTY);
306 
307 	/* Clear any pages left on the stack. */
308 	return (RET_SUCCESS);
309 
310 	/*
311 	 * If something fails in the above loop we were already walking back
312 	 * up the tree and the tree is now inconsistent.  Nothing much we can
313 	 * do about it but release any memory we're holding.
314 	 */
315 err1:	mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
316 	mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
317 
318 err2:	mpool_put(t->bt_mp, l, 0);
319 	mpool_put(t->bt_mp, r, 0);
320 	__dbpanic(t->bt_dbp);
321 	return (RET_ERROR);
322 }
323 
324 /*
325  * BT_PAGE -- Split a non-root page of a btree.
326  *
327  * Parameters:
328  *	t:	tree
329  *	h:	root page
330  *	lp:	pointer to left page pointer
331  *	rp:	pointer to right page pointer
332  *	skip:	pointer to index to leave open
333  *	ilen:	insert length
334  *
335  * Returns:
336  *	Pointer to page in which to insert or NULL on error.
337  */
338 static PAGE *
339 bt_page(BTREE *t, PAGE *h, PAGE **lp, PAGE **rp, indx_t *skip, size_t ilen)
340 {
341 	PAGE *l, *r, *tp;
342 	pgno_t npg;
343 
344 #ifdef STATISTICS
345 	++bt_split;
346 #endif
347 	/* Put the new right page for the split into place. */
348 	if ((r = __bt_new(t, &npg)) == NULL)
349 		return (NULL);
350 	r->pgno = npg;
351 	r->lower = BTDATAOFF;
352 	r->upper = t->bt_psize;
353 	r->nextpg = h->nextpg;
354 	r->prevpg = h->pgno;
355 	r->flags = h->flags & P_TYPE;
356 
357 	/*
358 	 * If we're splitting the last page on a level because we're appending
359 	 * a key to it (skip is NEXTINDEX()), it's likely that the data is
360 	 * sorted.  Adding an empty page on the side of the level is less work
361 	 * and can push the fill factor much higher than normal.  If we're
362 	 * wrong it's no big deal, we'll just do the split the right way next
363 	 * time.  It may look like it's equally easy to do a similar hack for
364 	 * reverse sorted data, that is, split the tree left, but it's not.
365 	 * Don't even try.
366 	 */
367 	if (h->nextpg == P_INVALID && *skip == NEXTINDEX(h)) {
368 #ifdef STATISTICS
369 		++bt_sortsplit;
370 #endif
371 		h->nextpg = r->pgno;
372 		r->lower = BTDATAOFF + sizeof(indx_t);
373 		*skip = 0;
374 		*lp = h;
375 		*rp = r;
376 		return (r);
377 	}
378 
379 	/* Put the new left page for the split into place. */
380 	if ((l = (PAGE *)calloc(1, t->bt_psize)) == NULL) {
381 		mpool_put(t->bt_mp, r, 0);
382 		return (NULL);
383 	}
384 	l->pgno = h->pgno;
385 	l->nextpg = r->pgno;
386 	l->prevpg = h->prevpg;
387 	l->lower = BTDATAOFF;
388 	l->upper = t->bt_psize;
389 	l->flags = h->flags & P_TYPE;
390 
391 	/* Fix up the previous pointer of the page after the split page. */
392 	if (h->nextpg != P_INVALID) {
393 		if ((tp = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL) {
394 			free(l);
395 			/* XXX mpool_free(t->bt_mp, r->pgno); */
396 			return (NULL);
397 		}
398 		tp->prevpg = r->pgno;
399 		mpool_put(t->bt_mp, tp, MPOOL_DIRTY);
400 	}
401 
402 	/*
403 	 * Split right.  The key/data pairs aren't sorted in the btree page so
404 	 * it's simpler to copy the data from the split page onto two new pages
405 	 * instead of copying half the data to the right page and compacting
406 	 * the left page in place.  Since the left page can't change, we have
407 	 * to swap the original and the allocated left page after the split.
408 	 */
409 	tp = bt_psplit(t, h, l, r, skip, ilen);
410 
411 	/* Move the new left page onto the old left page. */
412 	memmove(h, l, t->bt_psize);
413 	if (tp == l)
414 		tp = h;
415 	free(l);
416 
417 	*lp = h;
418 	*rp = r;
419 	return (tp);
420 }
421 
422 /*
423  * BT_ROOT -- Split the root page of a btree.
424  *
425  * Parameters:
426  *	t:	tree
427  *	h:	root page
428  *	lp:	pointer to left page pointer
429  *	rp:	pointer to right page pointer
430  *	skip:	pointer to index to leave open
431  *	ilen:	insert length
432  *
433  * Returns:
434  *	Pointer to page in which to insert or NULL on error.
435  */
436 static PAGE *
437 bt_root(BTREE *t, PAGE *h, PAGE **lp, PAGE **rp, indx_t *skip, size_t ilen)
438 {
439 	PAGE *l, *r, *tp;
440 	pgno_t lnpg, rnpg;
441 
442 #ifdef STATISTICS
443 	++bt_split;
444 	++bt_rootsplit;
445 #endif
446 	/* Put the new left and right pages for the split into place. */
447 	if ((l = __bt_new(t, &lnpg)) == NULL ||
448 	    (r = __bt_new(t, &rnpg)) == NULL)
449 		return (NULL);
450 	l->pgno = lnpg;
451 	r->pgno = rnpg;
452 	l->nextpg = r->pgno;
453 	r->prevpg = l->pgno;
454 	l->prevpg = r->nextpg = P_INVALID;
455 	l->lower = r->lower = BTDATAOFF;
456 	l->upper = r->upper = t->bt_psize;
457 	l->flags = r->flags = h->flags & P_TYPE;
458 
459 	/* Split the root page. */
460 	tp = bt_psplit(t, h, l, r, skip, ilen);
461 
462 	*lp = l;
463 	*rp = r;
464 	return (tp);
465 }
466 
467 /*
468  * BT_RROOT -- Fix up the recno root page after it has been split.
469  *
470  * Parameters:
471  *	t:	tree
472  *	h:	root page
473  *	l:	left page
474  *	r:	right page
475  *
476  * Returns:
477  *	RET_ERROR, RET_SUCCESS
478  */
479 static int
480 bt_rroot(BTREE *t, PAGE *h, PAGE *l, PAGE *r)
481 {
482 	char *dest;
483 
484 	/* Insert the left and right keys, set the header information. */
485 	h->linp[0] = h->upper = t->bt_psize - NRINTERNAL;
486 	dest = (char *)h + h->upper;
487 	WR_RINTERNAL(dest,
488 	    l->flags & P_RLEAF ? NEXTINDEX(l) : rec_total(l), l->pgno);
489 
490 	__PAST_END(h->linp, 1) = h->upper -= NRINTERNAL;
491 	dest = (char *)h + h->upper;
492 	WR_RINTERNAL(dest,
493 	    r->flags & P_RLEAF ? NEXTINDEX(r) : rec_total(r), r->pgno);
494 
495 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
496 
497 	/* Unpin the root page, set to recno internal page. */
498 	h->flags &= ~P_TYPE;
499 	h->flags |= P_RINTERNAL;
500 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
501 
502 	return (RET_SUCCESS);
503 }
504 
505 /*
506  * BT_BROOT -- Fix up the btree root page after it has been split.
507  *
508  * Parameters:
509  *	t:	tree
510  *	h:	root page
511  *	l:	left page
512  *	r:	right page
513  *
514  * Returns:
515  *	RET_ERROR, RET_SUCCESS
516  */
517 static int
518 bt_broot(BTREE *t, PAGE *h, PAGE *l, PAGE *r)
519 {
520 	BINTERNAL *bi;
521 	BLEAF *bl;
522 	u_int32_t nbytes;
523 	char *dest;
524 
525 	/*
526 	 * If the root page was a leaf page, change it into an internal page.
527 	 * We copy the key we split on (but not the key's data, in the case of
528 	 * a leaf page) to the new root page.
529 	 *
530 	 * The btree comparison code guarantees that the left-most key on any
531 	 * level of the tree is never used, so it doesn't need to be filled in.
532 	 */
533 	nbytes = NBINTERNAL(0);
534 	h->linp[0] = h->upper = t->bt_psize - nbytes;
535 	dest = (char *)h + h->upper;
536 	WR_BINTERNAL(dest, 0, l->pgno, 0);
537 
538 	switch (h->flags & P_TYPE) {
539 	case P_BLEAF:
540 		bl = GETBLEAF(r, 0);
541 		nbytes = NBINTERNAL(bl->ksize);
542 		__PAST_END(h->linp, 1) = h->upper -= nbytes;
543 		dest = (char *)h + h->upper;
544 		WR_BINTERNAL(dest, bl->ksize, r->pgno, 0);
545 		memmove(dest, bl->bytes, bl->ksize);
546 
547 		/*
548 		 * If the key is on an overflow page, mark the overflow chain
549 		 * so it isn't deleted when the leaf copy of the key is deleted.
550 		 */
551 	if (bl->flags & P_BIGKEY) {
552 			pgno_t pgno;
553 			memcpy(&pgno, bl->bytes, sizeof(pgno));
554 			if (bt_preserve(t, pgno) == RET_ERROR)
555 				return (RET_ERROR);
556 		}
557 		break;
558 	case P_BINTERNAL:
559 		bi = GETBINTERNAL(r, 0);
560 		nbytes = NBINTERNAL(bi->ksize);
561 		__PAST_END(h->linp, 1) = h->upper -= nbytes;
562 		dest = (char *)h + h->upper;
563 		memmove(dest, bi, nbytes);
564 		((BINTERNAL *)dest)->pgno = r->pgno;
565 		break;
566 	default:
567 		abort();
568 	}
569 
570 	/* There are two keys on the page. */
571 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
572 
573 	/* Unpin the root page, set to btree internal page. */
574 	h->flags &= ~P_TYPE;
575 	h->flags |= P_BINTERNAL;
576 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
577 
578 	return (RET_SUCCESS);
579 }
580 
581 /*
582  * BT_PSPLIT -- Do the real work of splitting the page.
583  *
584  * Parameters:
585  *	t:	tree
586  *	h:	page to be split
587  *	l:	page to put lower half of data
588  *	r:	page to put upper half of data
589  *	pskip:	pointer to index to leave open
590  *	ilen:	insert length
591  *
592  * Returns:
593  *	Pointer to page in which to insert.
594  */
595 static PAGE *
596 bt_psplit(BTREE *t, PAGE *h, PAGE *l, PAGE *r, indx_t *pskip, size_t ilen)
597 {
598 	BINTERNAL *bi;
599 	BLEAF *bl;
600 	CURSOR *c;
601 	RLEAF *rl;
602 	PAGE *rval;
603 	void *src;
604 	indx_t full, half, nxt, off, skip, top, used;
605 	u_int32_t nbytes;
606 	int bigkeycnt, isbigkey;
607 
608 	/*
609 	 * Split the data to the left and right pages.  Leave the skip index
610 	 * open.  Additionally, make some effort not to split on an overflow
611 	 * key.  This makes internal page processing faster and can save
612 	 * space as overflow keys used by internal pages are never deleted.
613 	 */
614 	bigkeycnt = 0;
615 	skip = *pskip;
616 	full = t->bt_psize - BTDATAOFF;
617 	half = full / 2;
618 	used = 0;
619 	for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) {
620 		if (skip == off) {
621 			nbytes = ilen;
622 			isbigkey = 0;		/* XXX: not really known. */
623 		} else
624 			switch (h->flags & P_TYPE) {
625 			case P_BINTERNAL:
626 				src = bi = GETBINTERNAL(h, nxt);
627 				nbytes = NBINTERNAL(bi->ksize);
628 				isbigkey = bi->flags & P_BIGKEY;
629 				break;
630 			case P_BLEAF:
631 				src = bl = GETBLEAF(h, nxt);
632 				nbytes = NBLEAF(bl);
633 				isbigkey = bl->flags & P_BIGKEY;
634 				break;
635 			case P_RINTERNAL:
636 				src = GETRINTERNAL(h, nxt);
637 				nbytes = NRINTERNAL;
638 				isbigkey = 0;
639 				break;
640 			case P_RLEAF:
641 				src = rl = GETRLEAF(h, nxt);
642 				nbytes = NRLEAF(rl);
643 				isbigkey = 0;
644 				break;
645 			default:
646 				abort();
647 			}
648 
649 		/*
650 		 * If the key/data pairs are substantial fractions of the max
651 		 * possible size for the page, it's possible to get situations
652 		 * where we decide to try and copy too much onto the left page.
653 		 * Make sure that doesn't happen.
654 		 */
655 		if ((skip <= off && used + nbytes + sizeof(indx_t) >= full) ||
656 		    nxt == top - 1) {
657 			--off;
658 			break;
659 		}
660 
661 		/* Copy the key/data pair, if not the skipped index. */
662 		if (skip != off) {
663 			++nxt;
664 
665 			l->linp[off] = l->upper -= nbytes;
666 			memmove((char *)l + l->upper, src, nbytes);
667 		}
668 
669 		used += nbytes + sizeof(indx_t);
670 		if (used >= half) {
671 			if (!isbigkey || bigkeycnt == 3)
672 				break;
673 			else
674 				++bigkeycnt;
675 		}
676 	}
677 
678 	/*
679 	 * Off is the last offset that's valid for the left page.
680 	 * Nxt is the first offset to be placed on the right page.
681 	 */
682 	l->lower += (off + 1) * sizeof(indx_t);
683 
684 	/*
685 	 * If splitting the page that the cursor was on, the cursor has to be
686 	 * adjusted to point to the same record as before the split.  If the
687 	 * cursor is at or past the skipped slot, the cursor is incremented by
688 	 * one.  If the cursor is on the right page, it is decremented by the
689 	 * number of records split to the left page.
690 	 */
691 	c = &t->bt_cursor;
692 	if (F_ISSET(c, CURS_INIT) && c->pg.pgno == h->pgno) {
693 		if (c->pg.index >= skip)
694 			++c->pg.index;
695 		if (c->pg.index < nxt)			/* Left page. */
696 			c->pg.pgno = l->pgno;
697 		else {					/* Right page. */
698 			c->pg.pgno = r->pgno;
699 			c->pg.index -= nxt;
700 		}
701 	}
702 
703 	/*
704 	 * If the skipped index was on the left page, just return that page.
705 	 * Otherwise, adjust the skip index to reflect the new position on
706 	 * the right page.
707 	 */
708 	if (skip <= off) {
709 		skip = MAX_PAGE_OFFSET;
710 		rval = l;
711 	} else {
712 		rval = r;
713 		*pskip -= nxt;
714 	}
715 
716 	for (off = 0; nxt < top; ++off) {
717 		if (skip == nxt) {
718 			++off;
719 			skip = MAX_PAGE_OFFSET;
720 		}
721 		switch (h->flags & P_TYPE) {
722 		case P_BINTERNAL:
723 			src = bi = GETBINTERNAL(h, nxt);
724 			nbytes = NBINTERNAL(bi->ksize);
725 			break;
726 		case P_BLEAF:
727 			src = bl = GETBLEAF(h, nxt);
728 			nbytes = NBLEAF(bl);
729 			break;
730 		case P_RINTERNAL:
731 			src = GETRINTERNAL(h, nxt);
732 			nbytes = NRINTERNAL;
733 			break;
734 		case P_RLEAF:
735 			src = rl = GETRLEAF(h, nxt);
736 			nbytes = NRLEAF(rl);
737 			break;
738 		default:
739 			abort();
740 		}
741 		++nxt;
742 		r->linp[off] = r->upper -= nbytes;
743 		memmove((char *)r + r->upper, src, nbytes);
744 	}
745 	r->lower += off * sizeof(indx_t);
746 
747 	/* If the key is being appended to the page, adjust the index. */
748 	if (skip == top)
749 		r->lower += sizeof(indx_t);
750 
751 	return (rval);
752 }
753 
754 /*
755  * BT_PRESERVE -- Mark a chain of pages as used by an internal node.
756  *
757  * Chains of indirect blocks pointed to by leaf nodes get reclaimed when the
758  * record that references them gets deleted.  Chains pointed to by internal
759  * pages never get deleted.  This routine marks a chain as pointed to by an
760  * internal page.
761  *
762  * Parameters:
763  *	t:	tree
764  *	pg:	page number of first page in the chain.
765  *
766  * Returns:
767  *	RET_SUCCESS, RET_ERROR.
768  */
769 static int
770 bt_preserve(BTREE *t, pgno_t pg)
771 {
772 	PAGE *h;
773 
774 	if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL)
775 		return (RET_ERROR);
776 	h->flags |= P_PRESERVE;
777 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
778 	return (RET_SUCCESS);
779 }
780 
781 /*
782  * REC_TOTAL -- Return the number of recno entries below a page.
783  *
784  * Parameters:
785  *	h:	page
786  *
787  * Returns:
788  *	The number of recno entries below a page.
789  *
790  * XXX
791  * These values could be set by the bt_psplit routine.  The problem is that the
792  * entry has to be popped off of the stack etc. or the values have to be passed
793  * all the way back to bt_split/bt_rroot and it's not very clean.
794  */
795 static recno_t
796 rec_total(PAGE *h)
797 {
798 	recno_t recs;
799 	indx_t nxt, top;
800 
801 	for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt)
802 		recs += GETRINTERNAL(h, nxt)->nrecs;
803 	return (recs);
804 }
805