xref: /original-bsd/lib/libc/db/btree/bt_split.c (revision c3e32dec)
1 /*-
2  * Copyright (c) 1990, 1993
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  * %sccs.include.redist.c%
9  */
10 
11 #if defined(LIBC_SCCS) && !defined(lint)
12 static char sccsid[] = "@(#)bt_split.c	8.1 (Berkeley) 06/04/93";
13 #endif /* LIBC_SCCS and not lint */
14 
15 #include <sys/types.h>
16 
17 #define	__DBINTERFACE_PRIVATE
18 #include <limits.h>
19 #include <stdio.h>
20 #include <stdlib.h>
21 #include <string.h>
22 
23 #include <db.h>
24 #include "btree.h"
25 
26 static int	 bt_broot __P((BTREE *, PAGE *, PAGE *, PAGE *));
27 static PAGE	*bt_page
28 		    __P((BTREE *, PAGE *, PAGE **, PAGE **, u_int *, size_t));
29 static int	 bt_preserve __P((BTREE *, pgno_t));
30 static PAGE	*bt_psplit
31 		    __P((BTREE *, PAGE *, PAGE *, PAGE *, u_int *, size_t));
32 static PAGE	*bt_root
33 		    __P((BTREE *, PAGE *, PAGE **, PAGE **, u_int *, size_t));
34 static int	 bt_rroot __P((BTREE *, PAGE *, PAGE *, PAGE *));
35 static recno_t	 rec_total __P((PAGE *));
36 
37 #ifdef STATISTICS
38 u_long	bt_rootsplit, bt_split, bt_sortsplit, bt_pfxsaved;
39 #endif
40 
41 /*
42  * __BT_SPLIT -- Split the tree.
43  *
44  * Parameters:
45  *	t:	tree
46  *	sp:	page to split
47  *	key:	key to insert
48  *	data:	data to insert
49  *	flags:	BIGKEY/BIGDATA flags
50  *	ilen:	insert length
51  *	skip:	index to leave open
52  *
53  * Returns:
54  *	RET_ERROR, RET_SUCCESS
55  */
56 int
57 __bt_split(t, sp, key, data, flags, ilen, skip)
58 	BTREE *t;
59 	PAGE *sp;
60 	const DBT *key, *data;
61 	u_long flags;
62 	size_t ilen;
63 	u_int skip;
64 {
65 	BINTERNAL *bi;
66 	BLEAF *bl, *tbl;
67 	DBT a, b;
68 	EPGNO *parent;
69 	PAGE *h, *l, *r, *lchild, *rchild;
70 	indx_t nxtindex;
71 	size_t n, nbytes, nksize;
72 	int parentsplit;
73 	char *dest;
74 
75 	/*
76 	 * Split the page into two pages, l and r.  The split routines return
77 	 * a pointer to the page into which the key should be inserted and with
78 	 * skip set to the offset which should be used.  Additionally, l and r
79 	 * are pinned.
80 	 */
81 	h = sp->pgno == P_ROOT ?
82 	    bt_root(t, sp, &l, &r, &skip, ilen) :
83 	    bt_page(t, sp, &l, &r, &skip, ilen);
84 	if (h == NULL)
85 		return (RET_ERROR);
86 
87 	/*
88 	 * Insert the new key/data pair into the leaf page.  (Key inserts
89 	 * always cause a leaf page to split first.)
90 	 */
91 	h->linp[skip] = h->upper -= ilen;
92 	dest = (char *)h + h->upper;
93 	if (ISSET(t, R_RECNO))
94 		WR_RLEAF(dest, data, flags)
95 	else
96 		WR_BLEAF(dest, key, data, flags)
97 
98 	/* If the root page was split, make it look right. */
99 	if (sp->pgno == P_ROOT &&
100 	    (ISSET(t, R_RECNO) ?
101 	    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
102 		goto err2;
103 
104 	/*
105 	 * Now we walk the parent page stack -- a LIFO stack of the pages that
106 	 * were traversed when we searched for the page that split.  Each stack
107 	 * entry is a page number and a page index offset.  The offset is for
108 	 * the page traversed on the search.  We've just split a page, so we
109 	 * have to insert a new key into the parent page.
110 	 *
111 	 * If the insert into the parent page causes it to split, may have to
112 	 * continue splitting all the way up the tree.  We stop if the root
113 	 * splits or the page inserted into didn't have to split to hold the
114 	 * new key.  Some algorithms replace the key for the old page as well
115 	 * as the new page.  We don't, as there's no reason to believe that the
116 	 * first key on the old page is any better than the key we have, and,
117 	 * in the case of a key being placed at index 0 causing the split, the
118 	 * key is unavailable.
119 	 *
120 	 * There are a maximum of 5 pages pinned at any time.  We keep the left
121 	 * and right pages pinned while working on the parent.   The 5 are the
122 	 * two children, left parent and right parent (when the parent splits)
123 	 * and the root page or the overflow key page when calling bt_preserve.
124 	 * This code must make sure that all pins are released other than the
125 	 * root page or overflow page which is unlocked elsewhere.
126 	 */
127 	while ((parent = BT_POP(t)) != NULL) {
128 		lchild = l;
129 		rchild = r;
130 
131 		/* Get the parent page. */
132 		if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
133 			goto err2;
134 
135 	 	/*
136 		 * The new key goes ONE AFTER the index, because the split
137 		 * was to the right.
138 		 */
139 		skip = parent->index + 1;
140 
141 		/*
142 		 * Calculate the space needed on the parent page.
143 		 *
144 		 * Prefix trees: space hack when inserting into BINTERNAL
145 		 * pages.  Retain only what's needed to distinguish between
146 		 * the new entry and the LAST entry on the page to its left.
147 		 * If the keys compare equal, retain the entire key.  Note,
148 		 * we don't touch overflow keys, and the entire key must be
149 		 * retained for the next-to-left most key on the leftmost
150 		 * page of each level, or the search will fail.  Applicable
151 		 * ONLY to internal pages that have leaf pages as children.
152 		 * Further reduction of the key between pairs of internal
153 		 * pages loses too much information.
154 		 */
155 		switch (rchild->flags & P_TYPE) {
156 		case P_BINTERNAL:
157 			bi = GETBINTERNAL(rchild, 0);
158 			nbytes = NBINTERNAL(bi->ksize);
159 			break;
160 		case P_BLEAF:
161 			bl = GETBLEAF(rchild, 0);
162 			nbytes = NBINTERNAL(bl->ksize);
163 			if (t->bt_pfx && !(bl->flags & P_BIGKEY) &&
164 			    (h->prevpg != P_INVALID || skip > 1)) {
165 				tbl = GETBLEAF(lchild, NEXTINDEX(lchild) - 1);
166 				a.size = tbl->ksize;
167 				a.data = tbl->bytes;
168 				b.size = bl->ksize;
169 				b.data = bl->bytes;
170 				nksize = t->bt_pfx(&a, &b);
171 				n = NBINTERNAL(nksize);
172 				if (n < nbytes) {
173 #ifdef STATISTICS
174 					bt_pfxsaved += nbytes - n;
175 #endif
176 					nbytes = n;
177 				} else
178 					nksize = 0;
179 			} else
180 				nksize = 0;
181 			break;
182 		case P_RINTERNAL:
183 		case P_RLEAF:
184 			nbytes = NRINTERNAL;
185 			break;
186 		default:
187 			abort();
188 		}
189 
190 		/* Split the parent page if necessary or shift the indices. */
191 		if (h->upper - h->lower < nbytes + sizeof(indx_t)) {
192 			sp = h;
193 			h = h->pgno == P_ROOT ?
194 			    bt_root(t, h, &l, &r, &skip, nbytes) :
195 			    bt_page(t, h, &l, &r, &skip, nbytes);
196 			if (h == NULL)
197 				goto err1;
198 			parentsplit = 1;
199 		} else {
200 			if (skip < (nxtindex = NEXTINDEX(h)))
201 				memmove(h->linp + skip + 1, h->linp + skip,
202 				    (nxtindex - skip) * sizeof(indx_t));
203 			h->lower += sizeof(indx_t);
204 			parentsplit = 0;
205 		}
206 
207 		/* Insert the key into the parent page. */
208 		switch(rchild->flags & P_TYPE) {
209 		case P_BINTERNAL:
210 			h->linp[skip] = h->upper -= nbytes;
211 			dest = (char *)h + h->linp[skip];
212 			memmove(dest, bi, nbytes);
213 			((BINTERNAL *)dest)->pgno = rchild->pgno;
214 			break;
215 		case P_BLEAF:
216 			h->linp[skip] = h->upper -= nbytes;
217 			dest = (char *)h + h->linp[skip];
218 			WR_BINTERNAL(dest, nksize ? nksize : bl->ksize,
219 			    rchild->pgno, bl->flags & P_BIGKEY);
220 			memmove(dest, bl->bytes, nksize ? nksize : bl->ksize);
221 			if (bl->flags & P_BIGKEY &&
222 			    bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR)
223 				goto err1;
224 			break;
225 		case P_RINTERNAL:
226 			/*
227 			 * Update the left page count.  If split
228 			 * added at index 0, fix the correct page.
229 			 */
230 			if (skip > 0)
231 				dest = (char *)h + h->linp[skip - 1];
232 			else
233 				dest = (char *)l + l->linp[NEXTINDEX(l) - 1];
234 			((RINTERNAL *)dest)->nrecs = rec_total(lchild);
235 			((RINTERNAL *)dest)->pgno = lchild->pgno;
236 
237 			/* Update the right page count. */
238 			h->linp[skip] = h->upper -= nbytes;
239 			dest = (char *)h + h->linp[skip];
240 			((RINTERNAL *)dest)->nrecs = rec_total(rchild);
241 			((RINTERNAL *)dest)->pgno = rchild->pgno;
242 			break;
243 		case P_RLEAF:
244 			/*
245 			 * Update the left page count.  If split
246 			 * added at index 0, fix the correct page.
247 			 */
248 			if (skip > 0)
249 				dest = (char *)h + h->linp[skip - 1];
250 			else
251 				dest = (char *)l + l->linp[NEXTINDEX(l) - 1];
252 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(lchild);
253 			((RINTERNAL *)dest)->pgno = lchild->pgno;
254 
255 			/* Update the right page count. */
256 			h->linp[skip] = h->upper -= nbytes;
257 			dest = (char *)h + h->linp[skip];
258 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(rchild);
259 			((RINTERNAL *)dest)->pgno = rchild->pgno;
260 			break;
261 		default:
262 			abort();
263 		}
264 
265 		/* Unpin the held pages. */
266 		if (!parentsplit) {
267 			mpool_put(t->bt_mp, h, MPOOL_DIRTY);
268 			break;
269 		}
270 
271 		/* If the root page was split, make it look right. */
272 		if (sp->pgno == P_ROOT &&
273 		    (ISSET(t, R_RECNO) ?
274 		    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
275 			goto err1;
276 
277 		mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
278 		mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
279 	}
280 
281 	/* Unpin the held pages. */
282 	mpool_put(t->bt_mp, l, MPOOL_DIRTY);
283 	mpool_put(t->bt_mp, r, MPOOL_DIRTY);
284 
285 	/* Clear any pages left on the stack. */
286 	return (RET_SUCCESS);
287 
288 	/*
289 	 * If something fails in the above loop we were already walking back
290 	 * up the tree and the tree is now inconsistent.  Nothing much we can
291 	 * do about it but release any memory we're holding.
292 	 */
293 err1:	mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
294 	mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
295 
296 err2:	mpool_put(t->bt_mp, l, 0);
297 	mpool_put(t->bt_mp, r, 0);
298 	__dbpanic(t->bt_dbp);
299 	return (RET_ERROR);
300 }
301 
302 /*
303  * BT_PAGE -- Split a non-root page of a btree.
304  *
305  * Parameters:
306  *	t:	tree
307  *	h:	root page
308  *	lp:	pointer to left page pointer
309  *	rp:	pointer to right page pointer
310  *	skip:	pointer to index to leave open
311  *	ilen:	insert length
312  *
313  * Returns:
314  *	Pointer to page in which to insert or NULL on error.
315  */
316 static PAGE *
317 bt_page(t, h, lp, rp, skip, ilen)
318 	BTREE *t;
319 	PAGE *h, **lp, **rp;
320 	u_int *skip;
321 	size_t ilen;
322 {
323 	PAGE *l, *r, *tp;
324 	pgno_t npg;
325 
326 #ifdef STATISTICS
327 	++bt_split;
328 #endif
329 	/* Put the new right page for the split into place. */
330 	if ((r = __bt_new(t, &npg)) == NULL)
331 		return (NULL);
332 	r->pgno = npg;
333 	r->lower = BTDATAOFF;
334 	r->upper = t->bt_psize;
335 	r->nextpg = h->nextpg;
336 	r->prevpg = h->pgno;
337 	r->flags = h->flags & P_TYPE;
338 
339 	/*
340 	 * If we're splitting the last page on a level because we're appending
341 	 * a key to it (skip is NEXTINDEX()), it's likely that the data is
342 	 * sorted.  Adding an empty page on the side of the level is less work
343 	 * and can push the fill factor much higher than normal.  If we're
344 	 * wrong it's no big deal, we'll just do the split the right way next
345 	 * time.  It may look like it's equally easy to do a similar hack for
346 	 * reverse sorted data, that is, split the tree left, but it's not.
347 	 * Don't even try.
348 	 */
349 	if (h->nextpg == P_INVALID && *skip == NEXTINDEX(h)) {
350 #ifdef STATISTICS
351 		++bt_sortsplit;
352 #endif
353 		h->nextpg = r->pgno;
354 		r->lower = BTDATAOFF + sizeof(indx_t);
355 		*skip = 0;
356 		*lp = h;
357 		*rp = r;
358 		return (r);
359 	}
360 
361 	/* Put the new left page for the split into place. */
362 	if ((l = malloc(t->bt_psize)) == NULL) {
363 		mpool_put(t->bt_mp, r, 0);
364 		return (NULL);
365 	}
366 	l->pgno = h->pgno;
367 	l->nextpg = r->pgno;
368 	l->prevpg = h->prevpg;
369 	l->lower = BTDATAOFF;
370 	l->upper = t->bt_psize;
371 	l->flags = h->flags & P_TYPE;
372 
373 	/* Fix up the previous pointer of the page after the split page. */
374 	if (h->nextpg != P_INVALID) {
375 		if ((tp = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL) {
376 			free(l);
377 			/* XXX mpool_free(t->bt_mp, r->pgno); */
378 			return (NULL);
379 		}
380 		tp->prevpg = r->pgno;
381 		mpool_put(t->bt_mp, tp, 0);
382 	}
383 
384 	/*
385 	 * Split right.  The key/data pairs aren't sorted in the btree page so
386 	 * it's simpler to copy the data from the split page onto two new pages
387 	 * instead of copying half the data to the right page and compacting
388 	 * the left page in place.  Since the left page can't change, we have
389 	 * to swap the original and the allocated left page after the split.
390 	 */
391 	tp = bt_psplit(t, h, l, r, skip, ilen);
392 
393 	/* Move the new left page onto the old left page. */
394 	memmove(h, l, t->bt_psize);
395 	if (tp == l)
396 		tp = h;
397 	free(l);
398 
399 	*lp = h;
400 	*rp = r;
401 	return (tp);
402 }
403 
404 /*
405  * BT_ROOT -- Split the root page of a btree.
406  *
407  * Parameters:
408  *	t:	tree
409  *	h:	root page
410  *	lp:	pointer to left page pointer
411  *	rp:	pointer to right page pointer
412  *	skip:	pointer to index to leave open
413  *	ilen:	insert length
414  *
415  * Returns:
416  *	Pointer to page in which to insert or NULL on error.
417  */
418 static PAGE *
419 bt_root(t, h, lp, rp, skip, ilen)
420 	BTREE *t;
421 	PAGE *h, **lp, **rp;
422 	u_int *skip;
423 	size_t ilen;
424 {
425 	PAGE *l, *r, *tp;
426 	pgno_t lnpg, rnpg;
427 
428 #ifdef STATISTICS
429 	++bt_split;
430 	++bt_rootsplit;
431 #endif
432 	/* Put the new left and right pages for the split into place. */
433 	if ((l = __bt_new(t, &lnpg)) == NULL ||
434 	    (r = __bt_new(t, &rnpg)) == NULL)
435 		return (NULL);
436 	l->pgno = lnpg;
437 	r->pgno = rnpg;
438 	l->nextpg = r->pgno;
439 	r->prevpg = l->pgno;
440 	l->prevpg = r->nextpg = P_INVALID;
441 	l->lower = r->lower = BTDATAOFF;
442 	l->upper = r->upper = t->bt_psize;
443 	l->flags = r->flags = h->flags & P_TYPE;
444 
445 	/* Split the root page. */
446 	tp = bt_psplit(t, h, l, r, skip, ilen);
447 
448 	*lp = l;
449 	*rp = r;
450 	return (tp);
451 }
452 
453 /*
454  * BT_RROOT -- Fix up the recno root page after it has been split.
455  *
456  * Parameters:
457  *	t:	tree
458  *	h:	root page
459  *	l:	left page
460  *	r:	right page
461  *
462  * Returns:
463  *	RET_ERROR, RET_SUCCESS
464  */
465 static int
466 bt_rroot(t, h, l, r)
467 	BTREE *t;
468 	PAGE *h, *l, *r;
469 {
470 	char *dest;
471 
472 	/* Insert the left and right keys, set the header information. */
473 	h->linp[0] = h->upper = t->bt_psize - NRINTERNAL;
474 	dest = (char *)h + h->upper;
475 	WR_RINTERNAL(dest,
476 	    l->flags & P_RLEAF ? NEXTINDEX(l) : rec_total(l), l->pgno);
477 
478 	h->linp[1] = h->upper -= NRINTERNAL;
479 	dest = (char *)h + h->upper;
480 	WR_RINTERNAL(dest,
481 	    r->flags & P_RLEAF ? NEXTINDEX(r) : rec_total(r), r->pgno);
482 
483 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
484 
485 	/* Unpin the root page, set to recno internal page. */
486 	h->flags &= ~P_TYPE;
487 	h->flags |= P_RINTERNAL;
488 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
489 
490 	return (RET_SUCCESS);
491 }
492 
493 /*
494  * BT_BROOT -- Fix up the btree root page after it has been split.
495  *
496  * Parameters:
497  *	t:	tree
498  *	h:	root page
499  *	l:	left page
500  *	r:	right page
501  *
502  * Returns:
503  *	RET_ERROR, RET_SUCCESS
504  */
505 static int
506 bt_broot(t, h, l, r)
507 	BTREE *t;
508 	PAGE *h, *l, *r;
509 {
510 	BINTERNAL *bi;
511 	BLEAF *bl;
512 	size_t nbytes;
513 	char *dest;
514 
515 	/*
516 	 * If the root page was a leaf page, change it into an internal page.
517 	 * We copy the key we split on (but not the key's data, in the case of
518 	 * a leaf page) to the new root page.
519 	 *
520 	 * The btree comparison code guarantees that the left-most key on any
521 	 * level of the tree is never used, so it doesn't need to be filled in.
522 	 */
523 	nbytes = NBINTERNAL(0);
524 	h->linp[0] = h->upper = t->bt_psize - nbytes;
525 	dest = (char *)h + h->upper;
526 	WR_BINTERNAL(dest, 0, l->pgno, 0);
527 
528 	switch(h->flags & P_TYPE) {
529 	case P_BLEAF:
530 		bl = GETBLEAF(r, 0);
531 		nbytes = NBINTERNAL(bl->ksize);
532 		h->linp[1] = h->upper -= nbytes;
533 		dest = (char *)h + h->upper;
534 		WR_BINTERNAL(dest, bl->ksize, r->pgno, 0);
535 		memmove(dest, bl->bytes, bl->ksize);
536 
537 		/*
538 		 * If the key is on an overflow page, mark the overflow chain
539 		 * so it isn't deleted when the leaf copy of the key is deleted.
540 		 */
541 		if (bl->flags & P_BIGKEY &&
542 		    bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR)
543 			return (RET_ERROR);
544 		break;
545 	case P_BINTERNAL:
546 		bi = GETBINTERNAL(r, 0);
547 		nbytes = NBINTERNAL(bi->ksize);
548 		h->linp[1] = h->upper -= nbytes;
549 		dest = (char *)h + h->upper;
550 		memmove(dest, bi, nbytes);
551 		((BINTERNAL *)dest)->pgno = r->pgno;
552 		break;
553 	default:
554 		abort();
555 	}
556 
557 	/* There are two keys on the page. */
558 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
559 
560 	/* Unpin the root page, set to btree internal page. */
561 	h->flags &= ~P_TYPE;
562 	h->flags |= P_BINTERNAL;
563 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
564 
565 	return (RET_SUCCESS);
566 }
567 
568 /*
569  * BT_PSPLIT -- Do the real work of splitting the page.
570  *
571  * Parameters:
572  *	t:	tree
573  *	h:	page to be split
574  *	l:	page to put lower half of data
575  *	r:	page to put upper half of data
576  *	pskip:	pointer to index to leave open
577  *	ilen:	insert length
578  *
579  * Returns:
580  *	Pointer to page in which to insert.
581  */
582 static PAGE *
583 bt_psplit(t, h, l, r, pskip, ilen)
584 	BTREE *t;
585 	PAGE *h, *l, *r;
586 	u_int *pskip;
587 	size_t ilen;
588 {
589 	BINTERNAL *bi;
590 	BLEAF *bl;
591 	RLEAF *rl;
592 	EPGNO *c;
593 	PAGE *rval;
594 	void *src;
595 	indx_t full, half, nxt, off, skip, top, used;
596 	size_t nbytes;
597 	int bigkeycnt, isbigkey;
598 
599 	/*
600 	 * Split the data to the left and right pages.  Leave the skip index
601 	 * open.  Additionally, make some effort not to split on an overflow
602 	 * key.  This makes internal page processing faster and can save
603 	 * space as overflow keys used by internal pages are never deleted.
604 	 */
605 	bigkeycnt = 0;
606 	skip = *pskip;
607 	full = t->bt_psize - BTDATAOFF;
608 	half = full / 2;
609 	used = 0;
610 	for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) {
611 		if (skip == off) {
612 			nbytes = ilen;
613 			isbigkey = 0;		/* XXX: not really known. */
614 		} else
615 			switch (h->flags & P_TYPE) {
616 			case P_BINTERNAL:
617 				src = bi = GETBINTERNAL(h, nxt);
618 				nbytes = NBINTERNAL(bi->ksize);
619 				isbigkey = bi->flags & P_BIGKEY;
620 				break;
621 			case P_BLEAF:
622 				src = bl = GETBLEAF(h, nxt);
623 				nbytes = NBLEAF(bl);
624 				isbigkey = bl->flags & P_BIGKEY;
625 				break;
626 			case P_RINTERNAL:
627 				src = GETRINTERNAL(h, nxt);
628 				nbytes = NRINTERNAL;
629 				isbigkey = 0;
630 				break;
631 			case P_RLEAF:
632 				src = rl = GETRLEAF(h, nxt);
633 				nbytes = NRLEAF(rl);
634 				isbigkey = 0;
635 				break;
636 			default:
637 				abort();
638 			}
639 
640 		/*
641 		 * If the key/data pairs are substantial fractions of the max
642 		 * possible size for the page, it's possible to get situations
643 		 * where we decide to try and copy too much onto the left page.
644 		 * Make sure that doesn't happen.
645 		 */
646 		if (skip <= off && used + nbytes >= full) {
647 			--off;
648 			break;
649 		}
650 
651 		/* Copy the key/data pair, if not the skipped index. */
652 		if (skip != off) {
653 			++nxt;
654 
655 			l->linp[off] = l->upper -= nbytes;
656 			memmove((char *)l + l->upper, src, nbytes);
657 		}
658 
659 		used += nbytes;
660 		if (used >= half) {
661 			if (!isbigkey || bigkeycnt == 3)
662 				break;
663 			else
664 				++bigkeycnt;
665 		}
666 	}
667 
668 	/*
669 	 * Off is the last offset that's valid for the left page.
670 	 * Nxt is the first offset to be placed on the right page.
671 	 */
672 	l->lower += (off + 1) * sizeof(indx_t);
673 
674 	/*
675 	 * If splitting the page that the cursor was on, the cursor has to be
676 	 * adjusted to point to the same record as before the split.  If the
677 	 * cursor is at or past the skipped slot, the cursor is incremented by
678 	 * one.  If the cursor is on the right page, it is decremented by the
679 	 * number of records split to the left page.
680 	 *
681 	 * Don't bother checking for the B_SEQINIT flag, the page number will
682 	 * be P_INVALID.
683 	 */
684 	c = &t->bt_bcursor;
685 	if (c->pgno == h->pgno) {
686 		if (c->index >= skip)
687 			++c->index;
688 		if (c->index < nxt)			/* Left page. */
689 			c->pgno = l->pgno;
690 		else {					/* Right page. */
691 			c->pgno = r->pgno;
692 			c->index -= nxt;
693 		}
694 	}
695 
696 	/*
697 	 * If the skipped index was on the left page, just return that page.
698 	 * Otherwise, adjust the skip index to reflect the new position on
699 	 * the right page.
700 	 */
701 	if (skip <= off) {
702 		skip = 0;
703 		rval = l;
704 	} else {
705 		rval = r;
706 		*pskip -= nxt;
707 	}
708 
709 	for (off = 0; nxt < top; ++off) {
710 		if (skip == nxt) {
711 			++off;
712 			skip = 0;
713 		}
714 		switch (h->flags & P_TYPE) {
715 		case P_BINTERNAL:
716 			src = bi = GETBINTERNAL(h, nxt);
717 			nbytes = NBINTERNAL(bi->ksize);
718 			break;
719 		case P_BLEAF:
720 			src = bl = GETBLEAF(h, nxt);
721 			nbytes = NBLEAF(bl);
722 			break;
723 		case P_RINTERNAL:
724 			src = GETRINTERNAL(h, nxt);
725 			nbytes = NRINTERNAL;
726 			break;
727 		case P_RLEAF:
728 			src = rl = GETRLEAF(h, nxt);
729 			nbytes = NRLEAF(rl);
730 			break;
731 		default:
732 			abort();
733 		}
734 		++nxt;
735 		r->linp[off] = r->upper -= nbytes;
736 		memmove((char *)r + r->upper, src, nbytes);
737 	}
738 	r->lower += off * sizeof(indx_t);
739 
740 	/* If the key is being appended to the page, adjust the index. */
741 	if (skip == top)
742 		r->lower += sizeof(indx_t);
743 
744 	return (rval);
745 }
746 
747 /*
748  * BT_PRESERVE -- Mark a chain of pages as used by an internal node.
749  *
750  * Chains of indirect blocks pointed to by leaf nodes get reclaimed when the
751  * record that references them gets deleted.  Chains pointed to by internal
752  * pages never get deleted.  This routine marks a chain as pointed to by an
753  * internal page.
754  *
755  * Parameters:
756  *	t:	tree
757  *	pg:	page number of first page in the chain.
758  *
759  * Returns:
760  *	RET_SUCCESS, RET_ERROR.
761  */
762 static int
763 bt_preserve(t, pg)
764 	BTREE *t;
765 	pgno_t pg;
766 {
767 	PAGE *h;
768 
769 	if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL)
770 		return (RET_ERROR);
771 	h->flags |= P_PRESERVE;
772 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
773 	return (RET_SUCCESS);
774 }
775 
776 /*
777  * REC_TOTAL -- Return the number of recno entries below a page.
778  *
779  * Parameters:
780  *	h:	page
781  *
782  * Returns:
783  *	The number of recno entries below a page.
784  *
785  * XXX
786  * These values could be set by the bt_psplit routine.  The problem is that the
787  * entry has to be popped off of the stack etc. or the values have to be passed
788  * all the way back to bt_split/bt_rroot and it's not very clean.
789  */
790 static recno_t
791 rec_total(h)
792 	PAGE *h;
793 {
794 	recno_t recs;
795 	indx_t nxt, top;
796 
797 	for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt)
798 		recs += GETRINTERNAL(h, nxt)->nrecs;
799 	return (recs);
800 }
801