Home | History | Annotate | Line # | Download | only in btree
bt_split.c revision 1.7
      1 /*	$NetBSD: bt_split.c,v 1.7 1997/07/13 18:51:59 christos Exp $	*/
      2 
      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. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the University of
     21  *	California, Berkeley and its contributors.
     22  * 4. Neither the name of the University nor the names of its contributors
     23  *    may be used to endorse or promote products derived from this software
     24  *    without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36  * SUCH DAMAGE.
     37  */
     38 
     39 #include <sys/cdefs.h>
     40 #if defined(LIBC_SCCS) && !defined(lint)
     41 #if 0
     42 static char sccsid[] = "@(#)bt_split.c	8.9 (Berkeley) 7/26/94";
     43 #else
     44 __RCSID("$NetBSD: bt_split.c,v 1.7 1997/07/13 18:51:59 christos Exp $");
     45 #endif
     46 #endif /* LIBC_SCCS and not lint */
     47 
     48 #include <sys/types.h>
     49 
     50 #include <limits.h>
     51 #include <stdio.h>
     52 #include <stdlib.h>
     53 #include <string.h>
     54 
     55 #include <db.h>
     56 #include "btree.h"
     57 
     58 static int	 bt_broot __P((BTREE *, PAGE *, PAGE *, PAGE *));
     59 static PAGE	*bt_page
     60 		    __P((BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t));
     61 static int	 bt_preserve __P((BTREE *, pgno_t));
     62 static PAGE	*bt_psplit
     63 		    __P((BTREE *, PAGE *, PAGE *, PAGE *, indx_t *, size_t));
     64 static PAGE	*bt_root
     65 		    __P((BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t));
     66 static int	 bt_rroot __P((BTREE *, PAGE *, PAGE *, PAGE *));
     67 static recno_t	 rec_total __P((PAGE *));
     68 
     69 #ifdef STATISTICS
     70 u_long	bt_rootsplit, bt_split, bt_sortsplit, bt_pfxsaved;
     71 #endif
     72 
     73 /*
     74  * __BT_SPLIT -- Split the tree.
     75  *
     76  * Parameters:
     77  *	t:	tree
     78  *	sp:	page to split
     79  *	key:	key to insert
     80  *	data:	data to insert
     81  *	flags:	BIGKEY/BIGDATA flags
     82  *	ilen:	insert length
     83  *	skip:	index to leave open
     84  *
     85  * Returns:
     86  *	RET_ERROR, RET_SUCCESS
     87  */
     88 int
     89 __bt_split(t, sp, key, data, flags, ilen, argskip)
     90 	BTREE *t;
     91 	PAGE *sp;
     92 	const DBT *key, *data;
     93 	int flags;
     94 	size_t ilen;
     95 	u_int32_t argskip;
     96 {
     97 	BINTERNAL *bi = NULL;	/* pacify gcc */
     98 	BLEAF *bl = NULL, *tbl;	/* pacify gcc */
     99 	DBT a, b;
    100 	EPGNO *parent;
    101 	PAGE *h, *l, *r, *lchild, *rchild;
    102 	indx_t nxtindex;
    103 	u_int16_t skip;
    104 	u_int32_t n, nbytes, nksize = 0; /* pacify gcc */
    105 	int parentsplit;
    106 	char *dest;
    107 
    108 	/*
    109 	 * Split the page into two pages, l and r.  The split routines return
    110 	 * a pointer to the page into which the key should be inserted and with
    111 	 * skip set to the offset which should be used.  Additionally, l and r
    112 	 * are pinned.
    113 	 */
    114 	skip = argskip;
    115 	h = sp->pgno == P_ROOT ?
    116 	    bt_root(t, sp, &l, &r, &skip, ilen) :
    117 	    bt_page(t, sp, &l, &r, &skip, ilen);
    118 	if (h == NULL)
    119 		return (RET_ERROR);
    120 
    121 	/*
    122 	 * Insert the new key/data pair into the leaf page.  (Key inserts
    123 	 * always cause a leaf page to split first.)
    124 	 */
    125 	h->linp[skip] = h->upper -= ilen;
    126 	dest = (char *)h + h->upper;
    127 	if (F_ISSET(t, R_RECNO))
    128 		WR_RLEAF(dest, data, flags)
    129 	else
    130 		WR_BLEAF(dest, key, data, flags)
    131 
    132 	/* If the root page was split, make it look right. */
    133 	if (sp->pgno == P_ROOT &&
    134 	    (F_ISSET(t, R_RECNO) ?
    135 	    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
    136 		goto err2;
    137 
    138 	/*
    139 	 * Now we walk the parent page stack -- a LIFO stack of the pages that
    140 	 * were traversed when we searched for the page that split.  Each stack
    141 	 * entry is a page number and a page index offset.  The offset is for
    142 	 * the page traversed on the search.  We've just split a page, so we
    143 	 * have to insert a new key into the parent page.
    144 	 *
    145 	 * If the insert into the parent page causes it to split, may have to
    146 	 * continue splitting all the way up the tree.  We stop if the root
    147 	 * splits or the page inserted into didn't have to split to hold the
    148 	 * new key.  Some algorithms replace the key for the old page as well
    149 	 * as the new page.  We don't, as there's no reason to believe that the
    150 	 * first key on the old page is any better than the key we have, and,
    151 	 * in the case of a key being placed at index 0 causing the split, the
    152 	 * key is unavailable.
    153 	 *
    154 	 * There are a maximum of 5 pages pinned at any time.  We keep the left
    155 	 * and right pages pinned while working on the parent.   The 5 are the
    156 	 * two children, left parent and right parent (when the parent splits)
    157 	 * and the root page or the overflow key page when calling bt_preserve.
    158 	 * This code must make sure that all pins are released other than the
    159 	 * root page or overflow page which is unlocked elsewhere.
    160 	 */
    161 	while ((parent = BT_POP(t)) != NULL) {
    162 		lchild = l;
    163 		rchild = r;
    164 
    165 		/* Get the parent page. */
    166 		if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
    167 			goto err2;
    168 
    169 	 	/*
    170 		 * The new key goes ONE AFTER the index, because the split
    171 		 * was to the right.
    172 		 */
    173 		skip = parent->index + 1;
    174 
    175 		/*
    176 		 * Calculate the space needed on the parent page.
    177 		 *
    178 		 * Prefix trees: space hack when inserting into BINTERNAL
    179 		 * pages.  Retain only what's needed to distinguish between
    180 		 * the new entry and the LAST entry on the page to its left.
    181 		 * If the keys compare equal, retain the entire key.  Note,
    182 		 * we don't touch overflow keys, and the entire key must be
    183 		 * retained for the next-to-left most key on the leftmost
    184 		 * page of each level, or the search will fail.  Applicable
    185 		 * ONLY to internal pages that have leaf pages as children.
    186 		 * Further reduction of the key between pairs of internal
    187 		 * pages loses too much information.
    188 		 */
    189 		switch (rchild->flags & P_TYPE) {
    190 		case P_BINTERNAL:
    191 			bi = GETBINTERNAL(rchild, 0);
    192 			nbytes = NBINTERNAL(bi->ksize);
    193 			break;
    194 		case P_BLEAF:
    195 			bl = GETBLEAF(rchild, 0);
    196 			nbytes = NBINTERNAL(bl->ksize);
    197 			if (t->bt_pfx && !(bl->flags & P_BIGKEY) &&
    198 			    (h->prevpg != P_INVALID || skip > 1)) {
    199 				tbl = GETBLEAF(lchild, NEXTINDEX(lchild) - 1);
    200 				a.size = tbl->ksize;
    201 				a.data = tbl->bytes;
    202 				b.size = bl->ksize;
    203 				b.data = bl->bytes;
    204 				nksize = t->bt_pfx(&a, &b);
    205 				n = NBINTERNAL(nksize);
    206 				if (n < nbytes) {
    207 #ifdef STATISTICS
    208 					bt_pfxsaved += nbytes - n;
    209 #endif
    210 					nbytes = n;
    211 				} else
    212 					nksize = 0;
    213 			} else
    214 				nksize = 0;
    215 			break;
    216 		case P_RINTERNAL:
    217 		case P_RLEAF:
    218 			nbytes = NRINTERNAL;
    219 			break;
    220 		default:
    221 			abort();
    222 		}
    223 
    224 		/* Split the parent page if necessary or shift the indices. */
    225 		if (h->upper - h->lower < nbytes + sizeof(indx_t)) {
    226 			sp = h;
    227 			h = h->pgno == P_ROOT ?
    228 			    bt_root(t, h, &l, &r, &skip, nbytes) :
    229 			    bt_page(t, h, &l, &r, &skip, nbytes);
    230 			if (h == NULL)
    231 				goto err1;
    232 			parentsplit = 1;
    233 		} else {
    234 			if (skip < (nxtindex = NEXTINDEX(h)))
    235 				memmove(h->linp + skip + 1, h->linp + skip,
    236 				    (nxtindex - skip) * sizeof(indx_t));
    237 			h->lower += sizeof(indx_t);
    238 			parentsplit = 0;
    239 		}
    240 
    241 		/* Insert the key into the parent page. */
    242 		switch (rchild->flags & P_TYPE) {
    243 		case P_BINTERNAL:
    244 			h->linp[skip] = h->upper -= nbytes;
    245 			dest = (char *)h + h->linp[skip];
    246 			memmove(dest, bi, nbytes);
    247 			((BINTERNAL *)dest)->pgno = rchild->pgno;
    248 			break;
    249 		case P_BLEAF:
    250 			h->linp[skip] = h->upper -= nbytes;
    251 			dest = (char *)h + h->linp[skip];
    252 			WR_BINTERNAL(dest, nksize ? nksize : bl->ksize,
    253 			    rchild->pgno, bl->flags & P_BIGKEY);
    254 			memmove(dest, bl->bytes, nksize ? nksize : bl->ksize);
    255 			if (bl->flags & P_BIGKEY &&
    256 			    bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR)
    257 				goto err1;
    258 			break;
    259 		case P_RINTERNAL:
    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 = rec_total(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 = rec_total(rchild);
    275 			((RINTERNAL *)dest)->pgno = rchild->pgno;
    276 			break;
    277 		case P_RLEAF:
    278 			/*
    279 			 * Update the left page count.  If split
    280 			 * added at index 0, fix the correct page.
    281 			 */
    282 			if (skip > 0)
    283 				dest = (char *)h + h->linp[skip - 1];
    284 			else
    285 				dest = (char *)l + l->linp[NEXTINDEX(l) - 1];
    286 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(lchild);
    287 			((RINTERNAL *)dest)->pgno = lchild->pgno;
    288 
    289 			/* Update the right page count. */
    290 			h->linp[skip] = h->upper -= nbytes;
    291 			dest = (char *)h + h->linp[skip];
    292 			((RINTERNAL *)dest)->nrecs = NEXTINDEX(rchild);
    293 			((RINTERNAL *)dest)->pgno = rchild->pgno;
    294 			break;
    295 		default:
    296 			abort();
    297 		}
    298 
    299 		/* Unpin the held pages. */
    300 		if (!parentsplit) {
    301 			mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    302 			break;
    303 		}
    304 
    305 		/* If the root page was split, make it look right. */
    306 		if (sp->pgno == P_ROOT &&
    307 		    (F_ISSET(t, R_RECNO) ?
    308 		    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
    309 			goto err1;
    310 
    311 		mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
    312 		mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
    313 	}
    314 
    315 	/* Unpin the held pages. */
    316 	mpool_put(t->bt_mp, l, MPOOL_DIRTY);
    317 	mpool_put(t->bt_mp, r, MPOOL_DIRTY);
    318 
    319 	/* Clear any pages left on the stack. */
    320 	return (RET_SUCCESS);
    321 
    322 	/*
    323 	 * If something fails in the above loop we were already walking back
    324 	 * up the tree and the tree is now inconsistent.  Nothing much we can
    325 	 * do about it but release any memory we're holding.
    326 	 */
    327 err1:	mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
    328 	mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
    329 
    330 err2:	mpool_put(t->bt_mp, l, 0);
    331 	mpool_put(t->bt_mp, r, 0);
    332 	__dbpanic(t->bt_dbp);
    333 	return (RET_ERROR);
    334 }
    335 
    336 /*
    337  * BT_PAGE -- Split a non-root page of a btree.
    338  *
    339  * Parameters:
    340  *	t:	tree
    341  *	h:	root page
    342  *	lp:	pointer to left page pointer
    343  *	rp:	pointer to right page pointer
    344  *	skip:	pointer to index to leave open
    345  *	ilen:	insert length
    346  *
    347  * Returns:
    348  *	Pointer to page in which to insert or NULL on error.
    349  */
    350 static PAGE *
    351 bt_page(t, h, lp, rp, skip, ilen)
    352 	BTREE *t;
    353 	PAGE *h, **lp, **rp;
    354 	indx_t *skip;
    355 	size_t ilen;
    356 {
    357 	PAGE *l, *r, *tp;
    358 	pgno_t npg;
    359 
    360 #ifdef STATISTICS
    361 	++bt_split;
    362 #endif
    363 	/* Put the new right page for the split into place. */
    364 	if ((r = __bt_new(t, &npg)) == NULL)
    365 		return (NULL);
    366 	r->pgno = npg;
    367 	r->lower = BTDATAOFF;
    368 	r->upper = t->bt_psize;
    369 	r->nextpg = h->nextpg;
    370 	r->prevpg = h->pgno;
    371 	r->flags = h->flags & P_TYPE;
    372 
    373 	/*
    374 	 * If we're splitting the last page on a level because we're appending
    375 	 * a key to it (skip is NEXTINDEX()), it's likely that the data is
    376 	 * sorted.  Adding an empty page on the side of the level is less work
    377 	 * and can push the fill factor much higher than normal.  If we're
    378 	 * wrong it's no big deal, we'll just do the split the right way next
    379 	 * time.  It may look like it's equally easy to do a similar hack for
    380 	 * reverse sorted data, that is, split the tree left, but it's not.
    381 	 * Don't even try.
    382 	 */
    383 	if (h->nextpg == P_INVALID && *skip == NEXTINDEX(h)) {
    384 #ifdef STATISTICS
    385 		++bt_sortsplit;
    386 #endif
    387 		h->nextpg = r->pgno;
    388 		r->lower = BTDATAOFF + sizeof(indx_t);
    389 		*skip = 0;
    390 		*lp = h;
    391 		*rp = r;
    392 		return (r);
    393 	}
    394 
    395 	/* Put the new left page for the split into place. */
    396 	if ((l = (PAGE *)malloc(t->bt_psize)) == NULL) {
    397 		mpool_put(t->bt_mp, r, 0);
    398 		return (NULL);
    399 	}
    400 #ifdef PURIFY
    401 	memset(l, 0xff, t->bt_psize);
    402 #endif
    403 	l->pgno = h->pgno;
    404 	l->nextpg = r->pgno;
    405 	l->prevpg = h->prevpg;
    406 	l->lower = BTDATAOFF;
    407 	l->upper = t->bt_psize;
    408 	l->flags = h->flags & P_TYPE;
    409 
    410 	/* Fix up the previous pointer of the page after the split page. */
    411 	if (h->nextpg != P_INVALID) {
    412 		if ((tp = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL) {
    413 			free(l);
    414 			/* XXX mpool_free(t->bt_mp, r->pgno); */
    415 			return (NULL);
    416 		}
    417 		tp->prevpg = r->pgno;
    418 		mpool_put(t->bt_mp, tp, MPOOL_DIRTY);
    419 	}
    420 
    421 	/*
    422 	 * Split right.  The key/data pairs aren't sorted in the btree page so
    423 	 * it's simpler to copy the data from the split page onto two new pages
    424 	 * instead of copying half the data to the right page and compacting
    425 	 * the left page in place.  Since the left page can't change, we have
    426 	 * to swap the original and the allocated left page after the split.
    427 	 */
    428 	tp = bt_psplit(t, h, l, r, skip, ilen);
    429 
    430 	/* Move the new left page onto the old left page. */
    431 	memmove(h, l, t->bt_psize);
    432 	if (tp == l)
    433 		tp = h;
    434 	free(l);
    435 
    436 	*lp = h;
    437 	*rp = r;
    438 	return (tp);
    439 }
    440 
    441 /*
    442  * BT_ROOT -- Split the root page of a btree.
    443  *
    444  * Parameters:
    445  *	t:	tree
    446  *	h:	root page
    447  *	lp:	pointer to left page pointer
    448  *	rp:	pointer to right page pointer
    449  *	skip:	pointer to index to leave open
    450  *	ilen:	insert length
    451  *
    452  * Returns:
    453  *	Pointer to page in which to insert or NULL on error.
    454  */
    455 static PAGE *
    456 bt_root(t, h, lp, rp, skip, ilen)
    457 	BTREE *t;
    458 	PAGE *h, **lp, **rp;
    459 	indx_t *skip;
    460 	size_t ilen;
    461 {
    462 	PAGE *l, *r, *tp;
    463 	pgno_t lnpg, rnpg;
    464 
    465 #ifdef STATISTICS
    466 	++bt_split;
    467 	++bt_rootsplit;
    468 #endif
    469 	/* Put the new left and right pages for the split into place. */
    470 	if ((l = __bt_new(t, &lnpg)) == NULL ||
    471 	    (r = __bt_new(t, &rnpg)) == NULL)
    472 		return (NULL);
    473 	l->pgno = lnpg;
    474 	r->pgno = rnpg;
    475 	l->nextpg = r->pgno;
    476 	r->prevpg = l->pgno;
    477 	l->prevpg = r->nextpg = P_INVALID;
    478 	l->lower = r->lower = BTDATAOFF;
    479 	l->upper = r->upper = t->bt_psize;
    480 	l->flags = r->flags = h->flags & P_TYPE;
    481 
    482 	/* Split the root page. */
    483 	tp = bt_psplit(t, h, l, r, skip, ilen);
    484 
    485 	*lp = l;
    486 	*rp = r;
    487 	return (tp);
    488 }
    489 
    490 /*
    491  * BT_RROOT -- Fix up the recno root page after it has been split.
    492  *
    493  * Parameters:
    494  *	t:	tree
    495  *	h:	root page
    496  *	l:	left page
    497  *	r:	right page
    498  *
    499  * Returns:
    500  *	RET_ERROR, RET_SUCCESS
    501  */
    502 static int
    503 bt_rroot(t, h, l, r)
    504 	BTREE *t;
    505 	PAGE *h, *l, *r;
    506 {
    507 	char *dest;
    508 
    509 	/* Insert the left and right keys, set the header information. */
    510 	h->linp[0] = h->upper = t->bt_psize - NRINTERNAL;
    511 	dest = (char *)h + h->upper;
    512 	WR_RINTERNAL(dest,
    513 	    l->flags & P_RLEAF ? NEXTINDEX(l) : rec_total(l), l->pgno);
    514 
    515 	h->linp[1] = h->upper -= NRINTERNAL;
    516 	dest = (char *)h + h->upper;
    517 	WR_RINTERNAL(dest,
    518 	    r->flags & P_RLEAF ? NEXTINDEX(r) : rec_total(r), r->pgno);
    519 
    520 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
    521 
    522 	/* Unpin the root page, set to recno internal page. */
    523 	h->flags &= ~P_TYPE;
    524 	h->flags |= P_RINTERNAL;
    525 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    526 
    527 	return (RET_SUCCESS);
    528 }
    529 
    530 /*
    531  * BT_BROOT -- Fix up the btree root page after it has been split.
    532  *
    533  * Parameters:
    534  *	t:	tree
    535  *	h:	root page
    536  *	l:	left page
    537  *	r:	right page
    538  *
    539  * Returns:
    540  *	RET_ERROR, RET_SUCCESS
    541  */
    542 static int
    543 bt_broot(t, h, l, r)
    544 	BTREE *t;
    545 	PAGE *h, *l, *r;
    546 {
    547 	BINTERNAL *bi = NULL;	/* pacify gcc */
    548 	BLEAF *bl;
    549 	u_int32_t nbytes;
    550 	char *dest;
    551 
    552 	/*
    553 	 * If the root page was a leaf page, change it into an internal page.
    554 	 * We copy the key we split on (but not the key's data, in the case of
    555 	 * a leaf page) to the new root page.
    556 	 *
    557 	 * The btree comparison code guarantees that the left-most key on any
    558 	 * level of the tree is never used, so it doesn't need to be filled in.
    559 	 */
    560 	nbytes = NBINTERNAL(0);
    561 	h->linp[0] = h->upper = t->bt_psize - nbytes;
    562 	dest = (char *)h + h->upper;
    563 	WR_BINTERNAL(dest, 0, l->pgno, 0);
    564 
    565 	switch (h->flags & P_TYPE) {
    566 	case P_BLEAF:
    567 		bl = GETBLEAF(r, 0);
    568 		nbytes = NBINTERNAL(bl->ksize);
    569 		h->linp[1] = h->upper -= nbytes;
    570 		dest = (char *)h + h->upper;
    571 		WR_BINTERNAL(dest, bl->ksize, r->pgno, 0);
    572 		memmove(dest, bl->bytes, bl->ksize);
    573 
    574 		/*
    575 		 * If the key is on an overflow page, mark the overflow chain
    576 		 * so it isn't deleted when the leaf copy of the key is deleted.
    577 		 */
    578 		if (bl->flags & P_BIGKEY &&
    579 		    bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR)
    580 			return (RET_ERROR);
    581 		break;
    582 	case P_BINTERNAL:
    583 		bi = GETBINTERNAL(r, 0);
    584 		nbytes = NBINTERNAL(bi->ksize);
    585 		h->linp[1] = h->upper -= nbytes;
    586 		dest = (char *)h + h->upper;
    587 		memmove(dest, bi, nbytes);
    588 		((BINTERNAL *)dest)->pgno = r->pgno;
    589 		break;
    590 	default:
    591 		abort();
    592 	}
    593 
    594 	/* There are two keys on the page. */
    595 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
    596 
    597 	/* Unpin the root page, set to btree internal page. */
    598 	h->flags &= ~P_TYPE;
    599 	h->flags |= P_BINTERNAL;
    600 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    601 
    602 	return (RET_SUCCESS);
    603 }
    604 
    605 /*
    606  * BT_PSPLIT -- Do the real work of splitting the page.
    607  *
    608  * Parameters:
    609  *	t:	tree
    610  *	h:	page to be split
    611  *	l:	page to put lower half of data
    612  *	r:	page to put upper half of data
    613  *	pskip:	pointer to index to leave open
    614  *	ilen:	insert length
    615  *
    616  * Returns:
    617  *	Pointer to page in which to insert.
    618  */
    619 static PAGE *
    620 bt_psplit(t, h, l, r, pskip, ilen)
    621 	BTREE *t;
    622 	PAGE *h, *l, *r;
    623 	indx_t *pskip;
    624 	size_t ilen;
    625 {
    626 	BINTERNAL *bi;
    627 	BLEAF *bl;
    628 	CURSOR *c;
    629 	RLEAF *rl;
    630 	PAGE *rval;
    631 	void *src = NULL;	/* pacify gcc */
    632 	indx_t full, half, nxt, off, skip, top, used;
    633 	u_int32_t nbytes;
    634 	int bigkeycnt, isbigkey;
    635 
    636 	/*
    637 	 * Split the data to the left and right pages.  Leave the skip index
    638 	 * open.  Additionally, make some effort not to split on an overflow
    639 	 * key.  This makes internal page processing faster and can save
    640 	 * space as overflow keys used by internal pages are never deleted.
    641 	 */
    642 	bigkeycnt = 0;
    643 	skip = *pskip;
    644 	full = t->bt_psize - BTDATAOFF;
    645 	half = full / 2;
    646 	used = 0;
    647 	for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) {
    648 		if (skip == off) {
    649 			nbytes = ilen;
    650 			isbigkey = 0;		/* XXX: not really known. */
    651 		} else
    652 			switch (h->flags & P_TYPE) {
    653 			case P_BINTERNAL:
    654 				src = bi = GETBINTERNAL(h, nxt);
    655 				nbytes = NBINTERNAL(bi->ksize);
    656 				isbigkey = bi->flags & P_BIGKEY;
    657 				break;
    658 			case P_BLEAF:
    659 				src = bl = GETBLEAF(h, nxt);
    660 				nbytes = NBLEAF(bl);
    661 				isbigkey = bl->flags & P_BIGKEY;
    662 				break;
    663 			case P_RINTERNAL:
    664 				src = GETRINTERNAL(h, nxt);
    665 				nbytes = NRINTERNAL;
    666 				isbigkey = 0;
    667 				break;
    668 			case P_RLEAF:
    669 				src = rl = GETRLEAF(h, nxt);
    670 				nbytes = NRLEAF(rl);
    671 				isbigkey = 0;
    672 				break;
    673 			default:
    674 				abort();
    675 			}
    676 
    677 		/*
    678 		 * If the key/data pairs are substantial fractions of the max
    679 		 * possible size for the page, it's possible to get situations
    680 		 * where we decide to try and copy too much onto the left page.
    681 		 * Make sure that doesn't happen.
    682 		 */
    683 		if (skip <= off && used + nbytes >= full) {
    684 			--off;
    685 			break;
    686 		}
    687 
    688 		/* Copy the key/data pair, if not the skipped index. */
    689 		if (skip != off) {
    690 			++nxt;
    691 
    692 			l->linp[off] = l->upper -= nbytes;
    693 			memmove((char *)l + l->upper, src, nbytes);
    694 		}
    695 
    696 		used += nbytes;
    697 		if (used >= half) {
    698 			if (!isbigkey || bigkeycnt == 3)
    699 				break;
    700 			else
    701 				++bigkeycnt;
    702 		}
    703 	}
    704 
    705 	/*
    706 	 * Off is the last offset that's valid for the left page.
    707 	 * Nxt is the first offset to be placed on the right page.
    708 	 */
    709 	l->lower += (off + 1) * sizeof(indx_t);
    710 
    711 	/*
    712 	 * If splitting the page that the cursor was on, the cursor has to be
    713 	 * adjusted to point to the same record as before the split.  If the
    714 	 * cursor is at or past the skipped slot, the cursor is incremented by
    715 	 * one.  If the cursor is on the right page, it is decremented by the
    716 	 * number of records split to the left page.
    717 	 */
    718 	c = &t->bt_cursor;
    719 	if (F_ISSET(c, CURS_INIT) && c->pg.pgno == h->pgno) {
    720 		if (c->pg.index >= skip)
    721 			++c->pg.index;
    722 		if (c->pg.index < nxt)			/* Left page. */
    723 			c->pg.pgno = l->pgno;
    724 		else {					/* Right page. */
    725 			c->pg.pgno = r->pgno;
    726 			c->pg.index -= nxt;
    727 		}
    728 	}
    729 
    730 	/*
    731 	 * If the skipped index was on the left page, just return that page.
    732 	 * Otherwise, adjust the skip index to reflect the new position on
    733 	 * the right page.
    734 	 */
    735 	if (skip <= off) {
    736 		skip = 0;
    737 		rval = l;
    738 	} else {
    739 		rval = r;
    740 		*pskip -= nxt;
    741 	}
    742 
    743 	for (off = 0; nxt < top; ++off) {
    744 		if (skip == nxt) {
    745 			++off;
    746 			skip = 0;
    747 		}
    748 		switch (h->flags & P_TYPE) {
    749 		case P_BINTERNAL:
    750 			src = bi = GETBINTERNAL(h, nxt);
    751 			nbytes = NBINTERNAL(bi->ksize);
    752 			break;
    753 		case P_BLEAF:
    754 			src = bl = GETBLEAF(h, nxt);
    755 			nbytes = NBLEAF(bl);
    756 			break;
    757 		case P_RINTERNAL:
    758 			src = GETRINTERNAL(h, nxt);
    759 			nbytes = NRINTERNAL;
    760 			break;
    761 		case P_RLEAF:
    762 			src = rl = GETRLEAF(h, nxt);
    763 			nbytes = NRLEAF(rl);
    764 			break;
    765 		default:
    766 			abort();
    767 		}
    768 		++nxt;
    769 		r->linp[off] = r->upper -= nbytes;
    770 		memmove((char *)r + r->upper, src, nbytes);
    771 	}
    772 	r->lower += off * sizeof(indx_t);
    773 
    774 	/* If the key is being appended to the page, adjust the index. */
    775 	if (skip == top)
    776 		r->lower += sizeof(indx_t);
    777 
    778 	return (rval);
    779 }
    780 
    781 /*
    782  * BT_PRESERVE -- Mark a chain of pages as used by an internal node.
    783  *
    784  * Chains of indirect blocks pointed to by leaf nodes get reclaimed when the
    785  * record that references them gets deleted.  Chains pointed to by internal
    786  * pages never get deleted.  This routine marks a chain as pointed to by an
    787  * internal page.
    788  *
    789  * Parameters:
    790  *	t:	tree
    791  *	pg:	page number of first page in the chain.
    792  *
    793  * Returns:
    794  *	RET_SUCCESS, RET_ERROR.
    795  */
    796 static int
    797 bt_preserve(t, pg)
    798 	BTREE *t;
    799 	pgno_t pg;
    800 {
    801 	PAGE *h;
    802 
    803 	if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL)
    804 		return (RET_ERROR);
    805 	h->flags |= P_PRESERVE;
    806 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    807 	return (RET_SUCCESS);
    808 }
    809 
    810 /*
    811  * REC_TOTAL -- Return the number of recno entries below a page.
    812  *
    813  * Parameters:
    814  *	h:	page
    815  *
    816  * Returns:
    817  *	The number of recno entries below a page.
    818  *
    819  * XXX
    820  * These values could be set by the bt_psplit routine.  The problem is that the
    821  * entry has to be popped off of the stack etc. or the values have to be passed
    822  * all the way back to bt_split/bt_rroot and it's not very clean.
    823  */
    824 static recno_t
    825 rec_total(h)
    826 	PAGE *h;
    827 {
    828 	recno_t recs;
    829 	indx_t nxt, top;
    830 
    831 	for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt)
    832 		recs += GETRINTERNAL(h, nxt)->nrecs;
    833 	return (recs);
    834 }
    835