Home | History | Annotate | Line # | Download | only in btree
bt_split.c revision 1.13
      1 /*	$NetBSD: bt_split.c,v 1.13 2003/08/07 16:42:41 agc 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. 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/cdefs.h>
     36 #if defined(LIBC_SCCS) && !defined(lint)
     37 #if 0
     38 static char sccsid[] = "@(#)bt_split.c	8.9 (Berkeley) 7/26/94";
     39 #else
     40 __RCSID("$NetBSD: bt_split.c,v 1.13 2003/08/07 16:42:41 agc Exp $");
     41 #endif
     42 #endif /* LIBC_SCCS and not lint */
     43 
     44 #include "namespace.h"
     45 #include <sys/types.h>
     46 
     47 #include <limits.h>
     48 #include <stdio.h>
     49 #include <stdlib.h>
     50 #include <string.h>
     51 
     52 #include <db.h>
     53 #include "btree.h"
     54 
     55 static int	 bt_broot __P((BTREE *, PAGE *, PAGE *, PAGE *));
     56 static PAGE	*bt_page
     57 		    __P((BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t));
     58 static int	 bt_preserve __P((BTREE *, pgno_t));
     59 static PAGE	*bt_psplit
     60 		    __P((BTREE *, PAGE *, PAGE *, PAGE *, indx_t *, size_t));
     61 static PAGE	*bt_root
     62 		    __P((BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t));
     63 static int	 bt_rroot __P((BTREE *, PAGE *, PAGE *, PAGE *));
     64 static recno_t	 rec_total __P((PAGE *));
     65 
     66 #ifdef STATISTICS
     67 u_long	bt_rootsplit, bt_split, bt_sortsplit, bt_pfxsaved;
     68 #endif
     69 
     70 /*
     71  * __BT_SPLIT -- Split the tree.
     72  *
     73  * Parameters:
     74  *	t:	tree
     75  *	sp:	page to split
     76  *	key:	key to insert
     77  *	data:	data to insert
     78  *	flags:	BIGKEY/BIGDATA flags
     79  *	ilen:	insert length
     80  *	skip:	index to leave open
     81  *
     82  * Returns:
     83  *	RET_ERROR, RET_SUCCESS
     84  */
     85 int
     86 __bt_split(t, sp, key, data, flags, ilen, argskip)
     87 	BTREE *t;
     88 	PAGE *sp;
     89 	const DBT *key, *data;
     90 	int flags;
     91 	size_t ilen;
     92 	u_int32_t argskip;
     93 {
     94 	BINTERNAL *bi = NULL;	/* pacify gcc */
     95 	BLEAF *bl = NULL, *tbl;	/* pacify gcc */
     96 	DBT a, b;
     97 	EPGNO *parent;
     98 	PAGE *h, *l, *r, *lchild, *rchild;
     99 	indx_t nxtindex;
    100 	u_int16_t skip;
    101 	u_int32_t n, nbytes, nksize = 0; /* pacify gcc */
    102 	int parentsplit;
    103 	char *dest;
    104 
    105 	/*
    106 	 * Split the page into two pages, l and r.  The split routines return
    107 	 * a pointer to the page into which the key should be inserted and with
    108 	 * skip set to the offset which should be used.  Additionally, l and r
    109 	 * are pinned.
    110 	 */
    111 	skip = argskip;
    112 	h = sp->pgno == P_ROOT ?
    113 	    bt_root(t, sp, &l, &r, &skip, ilen) :
    114 	    bt_page(t, sp, &l, &r, &skip, ilen);
    115 	if (h == NULL)
    116 		return (RET_ERROR);
    117 
    118 	/*
    119 	 * Insert the new key/data pair into the leaf page.  (Key inserts
    120 	 * always cause a leaf page to split first.)
    121 	 */
    122 	h->linp[skip] = h->upper -= ilen;
    123 	dest = (char *)(void *)h + h->upper;
    124 	if (F_ISSET(t, R_RECNO))
    125 		WR_RLEAF(dest, data, flags)
    126 	else
    127 		WR_BLEAF(dest, key, data, flags)
    128 
    129 	/* If the root page was split, make it look right. */
    130 	if (sp->pgno == P_ROOT &&
    131 	    (F_ISSET(t, R_RECNO) ?
    132 	    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
    133 		goto err2;
    134 
    135 	/*
    136 	 * Now we walk the parent page stack -- a LIFO stack of the pages that
    137 	 * were traversed when we searched for the page that split.  Each stack
    138 	 * entry is a page number and a page index offset.  The offset is for
    139 	 * the page traversed on the search.  We've just split a page, so we
    140 	 * have to insert a new key into the parent page.
    141 	 *
    142 	 * If the insert into the parent page causes it to split, may have to
    143 	 * continue splitting all the way up the tree.  We stop if the root
    144 	 * splits or the page inserted into didn't have to split to hold the
    145 	 * new key.  Some algorithms replace the key for the old page as well
    146 	 * as the new page.  We don't, as there's no reason to believe that the
    147 	 * first key on the old page is any better than the key we have, and,
    148 	 * in the case of a key being placed at index 0 causing the split, the
    149 	 * key is unavailable.
    150 	 *
    151 	 * There are a maximum of 5 pages pinned at any time.  We keep the left
    152 	 * and right pages pinned while working on the parent.   The 5 are the
    153 	 * two children, left parent and right parent (when the parent splits)
    154 	 * and the root page or the overflow key page when calling bt_preserve.
    155 	 * This code must make sure that all pins are released other than the
    156 	 * root page or overflow page which is unlocked elsewhere.
    157 	 */
    158 	while ((parent = BT_POP(t)) != NULL) {
    159 		lchild = l;
    160 		rchild = r;
    161 
    162 		/* Get the parent page. */
    163 		if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
    164 			goto err2;
    165 
    166 	 	/*
    167 		 * The new key goes ONE AFTER the index, because the split
    168 		 * was to the right.
    169 		 */
    170 		skip = parent->index + 1;
    171 
    172 		/*
    173 		 * Calculate the space needed on the parent page.
    174 		 *
    175 		 * Prefix trees: space hack when inserting into BINTERNAL
    176 		 * pages.  Retain only what's needed to distinguish between
    177 		 * the new entry and the LAST entry on the page to its left.
    178 		 * If the keys compare equal, retain the entire key.  Note,
    179 		 * we don't touch overflow keys, and the entire key must be
    180 		 * retained for the next-to-left most key on the leftmost
    181 		 * page of each level, or the search will fail.  Applicable
    182 		 * ONLY to internal pages that have leaf pages as children.
    183 		 * Further reduction of the key between pairs of internal
    184 		 * pages loses too much information.
    185 		 */
    186 		switch (rchild->flags & P_TYPE) {
    187 		case P_BINTERNAL:
    188 			bi = GETBINTERNAL(rchild, 0);
    189 			nbytes = NBINTERNAL(bi->ksize);
    190 			break;
    191 		case P_BLEAF:
    192 			bl = GETBLEAF(rchild, 0);
    193 			nbytes = NBINTERNAL(bl->ksize);
    194 			if (t->bt_pfx && !(bl->flags & P_BIGKEY) &&
    195 			    (h->prevpg != P_INVALID || skip > 1)) {
    196 				tbl = GETBLEAF(lchild, NEXTINDEX(lchild) - 1);
    197 				a.size = tbl->ksize;
    198 				a.data = tbl->bytes;
    199 				b.size = bl->ksize;
    200 				b.data = bl->bytes;
    201 				nksize = t->bt_pfx(&a, &b);
    202 				n = NBINTERNAL(nksize);
    203 				if (n < nbytes) {
    204 #ifdef STATISTICS
    205 					bt_pfxsaved += nbytes - n;
    206 #endif
    207 					nbytes = n;
    208 				} else
    209 					nksize = 0;
    210 			} else
    211 				nksize = 0;
    212 			break;
    213 		case P_RINTERNAL:
    214 		case P_RLEAF:
    215 			nbytes = NRINTERNAL;
    216 			break;
    217 		default:
    218 			abort();
    219 		}
    220 
    221 		/* Split the parent page if necessary or shift the indices. */
    222 		if (h->upper - h->lower < nbytes + sizeof(indx_t)) {
    223 			sp = h;
    224 			h = h->pgno == P_ROOT ?
    225 			    bt_root(t, h, &l, &r, &skip, nbytes) :
    226 			    bt_page(t, h, &l, &r, &skip, nbytes);
    227 			if (h == NULL)
    228 				goto err1;
    229 			parentsplit = 1;
    230 		} else {
    231 			if (skip < (nxtindex = NEXTINDEX(h)))
    232 				memmove(h->linp + skip + 1, h->linp + skip,
    233 				    (nxtindex - skip) * sizeof(indx_t));
    234 			h->lower += sizeof(indx_t);
    235 			parentsplit = 0;
    236 		}
    237 
    238 		/* Insert the key into the parent page. */
    239 		switch (rchild->flags & P_TYPE) {
    240 		case P_BINTERNAL:
    241 			h->linp[skip] = h->upper -= nbytes;
    242 			dest = (char *)(void *)h + h->linp[skip];
    243 			memmove(dest, bi, nbytes);
    244 			((BINTERNAL *)(void *)dest)->pgno = rchild->pgno;
    245 			break;
    246 		case P_BLEAF:
    247 			h->linp[skip] = h->upper -= nbytes;
    248 			dest = (char *)(void *)h + h->linp[skip];
    249 			WR_BINTERNAL(dest, nksize ? nksize : bl->ksize,
    250 			    rchild->pgno, bl->flags & P_BIGKEY);
    251 			memmove(dest, bl->bytes, nksize ? nksize : bl->ksize);
    252 			if (bl->flags & P_BIGKEY &&
    253 			    bt_preserve(t, *(pgno_t *)(void *)bl->bytes) ==
    254 			    RET_ERROR)
    255 				goto err1;
    256 			break;
    257 		case P_RINTERNAL:
    258 			/*
    259 			 * Update the left page count.  If split
    260 			 * added at index 0, fix the correct page.
    261 			 */
    262 			if (skip > 0)
    263 				dest = (char *)(void *)h + h->linp[skip - 1];
    264 			else
    265 				dest = (char *)(void *)l + l->linp[NEXTINDEX(l) - 1];
    266 			((RINTERNAL *)(void *)dest)->nrecs = rec_total(lchild);
    267 			((RINTERNAL *)(void *)dest)->pgno = lchild->pgno;
    268 
    269 			/* Update the right page count. */
    270 			h->linp[skip] = h->upper -= nbytes;
    271 			dest = (char *)(void *)h + h->linp[skip];
    272 			((RINTERNAL *)(void *)dest)->nrecs = rec_total(rchild);
    273 			((RINTERNAL *)(void *)dest)->pgno = rchild->pgno;
    274 			break;
    275 		case P_RLEAF:
    276 			/*
    277 			 * Update the left page count.  If split
    278 			 * added at index 0, fix the correct page.
    279 			 */
    280 			if (skip > 0)
    281 				dest = (char *)(void *)h + h->linp[skip - 1];
    282 			else
    283 				dest = (char *)(void *)l + l->linp[NEXTINDEX(l) - 1];
    284 			((RINTERNAL *)(void *)dest)->nrecs = NEXTINDEX(lchild);
    285 			((RINTERNAL *)(void *)dest)->pgno = lchild->pgno;
    286 
    287 			/* Update the right page count. */
    288 			h->linp[skip] = h->upper -= nbytes;
    289 			dest = (char *)(void *)h + h->linp[skip];
    290 			((RINTERNAL *)(void *)dest)->nrecs = NEXTINDEX(rchild);
    291 			((RINTERNAL *)(void *)dest)->pgno = rchild->pgno;
    292 			break;
    293 		default:
    294 			abort();
    295 		}
    296 
    297 		/* Unpin the held pages. */
    298 		if (!parentsplit) {
    299 			mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    300 			break;
    301 		}
    302 
    303 		/* If the root page was split, make it look right. */
    304 		if (sp->pgno == P_ROOT &&
    305 		    (F_ISSET(t, R_RECNO) ?
    306 		    bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR)
    307 			goto err1;
    308 
    309 		mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
    310 		mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
    311 	}
    312 
    313 	/* Unpin the held pages. */
    314 	mpool_put(t->bt_mp, l, MPOOL_DIRTY);
    315 	mpool_put(t->bt_mp, r, MPOOL_DIRTY);
    316 
    317 	/* Clear any pages left on the stack. */
    318 	return (RET_SUCCESS);
    319 
    320 	/*
    321 	 * If something fails in the above loop we were already walking back
    322 	 * up the tree and the tree is now inconsistent.  Nothing much we can
    323 	 * do about it but release any memory we're holding.
    324 	 */
    325 err1:	mpool_put(t->bt_mp, lchild, MPOOL_DIRTY);
    326 	mpool_put(t->bt_mp, rchild, MPOOL_DIRTY);
    327 
    328 err2:	mpool_put(t->bt_mp, l, 0);
    329 	mpool_put(t->bt_mp, r, 0);
    330 	__dbpanic(t->bt_dbp);
    331 	return (RET_ERROR);
    332 }
    333 
    334 /*
    335  * BT_PAGE -- Split a non-root page of a btree.
    336  *
    337  * Parameters:
    338  *	t:	tree
    339  *	h:	root page
    340  *	lp:	pointer to left page pointer
    341  *	rp:	pointer to right page pointer
    342  *	skip:	pointer to index to leave open
    343  *	ilen:	insert length
    344  *
    345  * Returns:
    346  *	Pointer to page in which to insert or NULL on error.
    347  */
    348 static PAGE *
    349 bt_page(t, h, lp, rp, skip, ilen)
    350 	BTREE *t;
    351 	PAGE *h, **lp, **rp;
    352 	indx_t *skip;
    353 	size_t ilen;
    354 {
    355 	PAGE *l, *r, *tp;
    356 	pgno_t npg;
    357 
    358 #ifdef STATISTICS
    359 	++bt_split;
    360 #endif
    361 	/* Put the new right page for the split into place. */
    362 	if ((r = __bt_new(t, &npg)) == NULL)
    363 		return (NULL);
    364 	r->pgno = npg;
    365 	r->lower = BTDATAOFF;
    366 	r->upper = t->bt_psize;
    367 	r->nextpg = h->nextpg;
    368 	r->prevpg = h->pgno;
    369 	r->flags = h->flags & P_TYPE;
    370 
    371 	/*
    372 	 * If we're splitting the last page on a level because we're appending
    373 	 * a key to it (skip is NEXTINDEX()), it's likely that the data is
    374 	 * sorted.  Adding an empty page on the side of the level is less work
    375 	 * and can push the fill factor much higher than normal.  If we're
    376 	 * wrong it's no big deal, we'll just do the split the right way next
    377 	 * time.  It may look like it's equally easy to do a similar hack for
    378 	 * reverse sorted data, that is, split the tree left, but it's not.
    379 	 * Don't even try.
    380 	 */
    381 	if (h->nextpg == P_INVALID && *skip == NEXTINDEX(h)) {
    382 #ifdef STATISTICS
    383 		++bt_sortsplit;
    384 #endif
    385 		h->nextpg = r->pgno;
    386 		r->lower = BTDATAOFF + sizeof(indx_t);
    387 		*skip = 0;
    388 		*lp = h;
    389 		*rp = r;
    390 		return (r);
    391 	}
    392 
    393 	/* Put the new left page for the split into place. */
    394 	if ((l = (PAGE *)malloc(t->bt_psize)) == NULL) {
    395 		mpool_put(t->bt_mp, r, 0);
    396 		return (NULL);
    397 	}
    398 #ifdef PURIFY
    399 	memset(l, 0xff, t->bt_psize);
    400 #endif
    401 	l->pgno = h->pgno;
    402 	l->nextpg = r->pgno;
    403 	l->prevpg = h->prevpg;
    404 	l->lower = BTDATAOFF;
    405 	l->upper = t->bt_psize;
    406 	l->flags = h->flags & P_TYPE;
    407 
    408 	/* Fix up the previous pointer of the page after the split page. */
    409 	if (h->nextpg != P_INVALID) {
    410 		if ((tp = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL) {
    411 			free(l);
    412 			/* XXX mpool_free(t->bt_mp, r->pgno); */
    413 			return (NULL);
    414 		}
    415 		tp->prevpg = r->pgno;
    416 		mpool_put(t->bt_mp, tp, MPOOL_DIRTY);
    417 	}
    418 
    419 	/*
    420 	 * Split right.  The key/data pairs aren't sorted in the btree page so
    421 	 * it's simpler to copy the data from the split page onto two new pages
    422 	 * instead of copying half the data to the right page and compacting
    423 	 * the left page in place.  Since the left page can't change, we have
    424 	 * to swap the original and the allocated left page after the split.
    425 	 */
    426 	tp = bt_psplit(t, h, l, r, skip, ilen);
    427 
    428 	/* Move the new left page onto the old left page. */
    429 	memmove(h, l, t->bt_psize);
    430 	if (tp == l)
    431 		tp = h;
    432 	free(l);
    433 
    434 	*lp = h;
    435 	*rp = r;
    436 	return (tp);
    437 }
    438 
    439 /*
    440  * BT_ROOT -- Split the root page of a btree.
    441  *
    442  * Parameters:
    443  *	t:	tree
    444  *	h:	root page
    445  *	lp:	pointer to left page pointer
    446  *	rp:	pointer to right page pointer
    447  *	skip:	pointer to index to leave open
    448  *	ilen:	insert length
    449  *
    450  * Returns:
    451  *	Pointer to page in which to insert or NULL on error.
    452  */
    453 static PAGE *
    454 bt_root(t, h, lp, rp, skip, ilen)
    455 	BTREE *t;
    456 	PAGE *h, **lp, **rp;
    457 	indx_t *skip;
    458 	size_t ilen;
    459 {
    460 	PAGE *l, *r, *tp;
    461 	pgno_t lnpg, rnpg;
    462 
    463 #ifdef STATISTICS
    464 	++bt_split;
    465 	++bt_rootsplit;
    466 #endif
    467 	/* Put the new left and right pages for the split into place. */
    468 	if ((l = __bt_new(t, &lnpg)) == NULL ||
    469 	    (r = __bt_new(t, &rnpg)) == NULL)
    470 		return (NULL);
    471 	l->pgno = lnpg;
    472 	r->pgno = rnpg;
    473 	l->nextpg = r->pgno;
    474 	r->prevpg = l->pgno;
    475 	l->prevpg = r->nextpg = P_INVALID;
    476 	l->lower = r->lower = BTDATAOFF;
    477 	l->upper = r->upper = t->bt_psize;
    478 	l->flags = r->flags = h->flags & P_TYPE;
    479 
    480 	/* Split the root page. */
    481 	tp = bt_psplit(t, h, l, r, skip, ilen);
    482 
    483 	*lp = l;
    484 	*rp = r;
    485 	return (tp);
    486 }
    487 
    488 /*
    489  * BT_RROOT -- Fix up the recno root page after it has been split.
    490  *
    491  * Parameters:
    492  *	t:	tree
    493  *	h:	root page
    494  *	l:	left page
    495  *	r:	right page
    496  *
    497  * Returns:
    498  *	RET_ERROR, RET_SUCCESS
    499  */
    500 static int
    501 bt_rroot(t, h, l, r)
    502 	BTREE *t;
    503 	PAGE *h, *l, *r;
    504 {
    505 	char *dest;
    506 
    507 	/* Insert the left and right keys, set the header information. */
    508 	h->linp[0] = h->upper = t->bt_psize - NRINTERNAL;
    509 	dest = (char *)(void *)h + h->upper;
    510 	WR_RINTERNAL(dest,
    511 	    l->flags & P_RLEAF ? NEXTINDEX(l) : rec_total(l), l->pgno);
    512 
    513 	h->linp[1] = h->upper -= NRINTERNAL;
    514 	dest = (char *)(void *)h + h->upper;
    515 	WR_RINTERNAL(dest,
    516 	    r->flags & P_RLEAF ? NEXTINDEX(r) : rec_total(r), r->pgno);
    517 
    518 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
    519 
    520 	/* Unpin the root page, set to recno internal page. */
    521 	h->flags &= ~P_TYPE;
    522 	h->flags |= P_RINTERNAL;
    523 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    524 
    525 	return (RET_SUCCESS);
    526 }
    527 
    528 /*
    529  * BT_BROOT -- Fix up the btree root page after it has been split.
    530  *
    531  * Parameters:
    532  *	t:	tree
    533  *	h:	root page
    534  *	l:	left page
    535  *	r:	right page
    536  *
    537  * Returns:
    538  *	RET_ERROR, RET_SUCCESS
    539  */
    540 static int
    541 bt_broot(t, h, l, r)
    542 	BTREE *t;
    543 	PAGE *h, *l, *r;
    544 {
    545 	BINTERNAL *bi = NULL;	/* pacify gcc */
    546 	BLEAF *bl;
    547 	u_int32_t nbytes;
    548 	char *dest;
    549 
    550 	/*
    551 	 * If the root page was a leaf page, change it into an internal page.
    552 	 * We copy the key we split on (but not the key's data, in the case of
    553 	 * a leaf page) to the new root page.
    554 	 *
    555 	 * The btree comparison code guarantees that the left-most key on any
    556 	 * level of the tree is never used, so it doesn't need to be filled in.
    557 	 */
    558 	nbytes = NBINTERNAL(0);
    559 	h->linp[0] = h->upper = t->bt_psize - nbytes;
    560 	dest = (char *)(void *)h + h->upper;
    561 	WR_BINTERNAL(dest, 0, l->pgno, 0);
    562 
    563 	switch (h->flags & P_TYPE) {
    564 	case P_BLEAF:
    565 		bl = GETBLEAF(r, 0);
    566 		nbytes = NBINTERNAL(bl->ksize);
    567 		h->linp[1] = h->upper -= nbytes;
    568 		dest = (char *)(void *)h + h->upper;
    569 		WR_BINTERNAL(dest, bl->ksize, r->pgno, 0);
    570 		memmove(dest, bl->bytes, bl->ksize);
    571 
    572 		/*
    573 		 * If the key is on an overflow page, mark the overflow chain
    574 		 * so it isn't deleted when the leaf copy of the key is deleted.
    575 		 */
    576 		if (bl->flags & P_BIGKEY &&
    577 		    bt_preserve(t, *(pgno_t *)(void *)bl->bytes) == RET_ERROR)
    578 			return (RET_ERROR);
    579 		break;
    580 	case P_BINTERNAL:
    581 		bi = GETBINTERNAL(r, 0);
    582 		nbytes = NBINTERNAL(bi->ksize);
    583 		h->linp[1] = h->upper -= nbytes;
    584 		dest = (char *)(void *)h + h->upper;
    585 		memmove(dest, bi, nbytes);
    586 		((BINTERNAL *)(void *)dest)->pgno = r->pgno;
    587 		break;
    588 	default:
    589 		abort();
    590 	}
    591 
    592 	/* There are two keys on the page. */
    593 	h->lower = BTDATAOFF + 2 * sizeof(indx_t);
    594 
    595 	/* Unpin the root page, set to btree internal page. */
    596 	h->flags &= ~P_TYPE;
    597 	h->flags |= P_BINTERNAL;
    598 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    599 
    600 	return (RET_SUCCESS);
    601 }
    602 
    603 /*
    604  * BT_PSPLIT -- Do the real work of splitting the page.
    605  *
    606  * Parameters:
    607  *	t:	tree
    608  *	h:	page to be split
    609  *	l:	page to put lower half of data
    610  *	r:	page to put upper half of data
    611  *	pskip:	pointer to index to leave open
    612  *	ilen:	insert length
    613  *
    614  * Returns:
    615  *	Pointer to page in which to insert.
    616  */
    617 static PAGE *
    618 bt_psplit(t, h, l, r, pskip, ilen)
    619 	BTREE *t;
    620 	PAGE *h, *l, *r;
    621 	indx_t *pskip;
    622 	size_t ilen;
    623 {
    624 	BINTERNAL *bi;
    625 	BLEAF *bl;
    626 	CURSOR *c;
    627 	RLEAF *rl;
    628 	PAGE *rval;
    629 	void *src = NULL;	/* pacify gcc */
    630 	indx_t full, half, nxt, off, skip, top, used;
    631 	u_int32_t nbytes;
    632 	int bigkeycnt, isbigkey;
    633 
    634 	/*
    635 	 * Split the data to the left and right pages.  Leave the skip index
    636 	 * open.  Additionally, make some effort not to split on an overflow
    637 	 * key.  This makes internal page processing faster and can save
    638 	 * space as overflow keys used by internal pages are never deleted.
    639 	 */
    640 	bigkeycnt = 0;
    641 	skip = *pskip;
    642 	full = t->bt_psize - BTDATAOFF;
    643 	half = full / 2;
    644 	used = 0;
    645 	for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) {
    646 		if (skip == off) {
    647 			nbytes = ilen;
    648 			isbigkey = 0;		/* XXX: not really known. */
    649 		} else
    650 			switch (h->flags & P_TYPE) {
    651 			case P_BINTERNAL:
    652 				src = bi = GETBINTERNAL(h, nxt);
    653 				nbytes = NBINTERNAL(bi->ksize);
    654 				isbigkey = bi->flags & P_BIGKEY;
    655 				break;
    656 			case P_BLEAF:
    657 				src = bl = GETBLEAF(h, nxt);
    658 				nbytes = NBLEAF(bl);
    659 				isbigkey = bl->flags & P_BIGKEY;
    660 				break;
    661 			case P_RINTERNAL:
    662 				src = GETRINTERNAL(h, nxt);
    663 				nbytes = NRINTERNAL;
    664 				isbigkey = 0;
    665 				break;
    666 			case P_RLEAF:
    667 				src = rl = GETRLEAF(h, nxt);
    668 				nbytes = NRLEAF(rl);
    669 				isbigkey = 0;
    670 				break;
    671 			default:
    672 				abort();
    673 			}
    674 
    675 		/*
    676 		 * If the key/data pairs are substantial fractions of the max
    677 		 * possible size for the page, it's possible to get situations
    678 		 * where we decide to try and copy too much onto the left page.
    679 		 * Make sure that doesn't happen.
    680 		 */
    681 		if ((skip <= off && used + nbytes + sizeof(indx_t) >= full) ||
    682 		    nxt == top - 1) {
    683 			--off;
    684 			break;
    685 		}
    686 
    687 		/* Copy the key/data pair, if not the skipped index. */
    688 		if (skip != off) {
    689 			++nxt;
    690 
    691 			l->linp[off] = l->upper -= nbytes;
    692 			memmove((char *)(void *)l + l->upper, src, nbytes);
    693 		}
    694 
    695 		used += nbytes + sizeof(indx_t);
    696 		if (used >= half) {
    697 			if (!isbigkey || bigkeycnt == 3)
    698 				break;
    699 			else
    700 				++bigkeycnt;
    701 		}
    702 	}
    703 
    704 	/*
    705 	 * Off is the last offset that's valid for the left page.
    706 	 * Nxt is the first offset to be placed on the right page.
    707 	 */
    708 	l->lower += (off + 1) * sizeof(indx_t);
    709 
    710 	/*
    711 	 * If splitting the page that the cursor was on, the cursor has to be
    712 	 * adjusted to point to the same record as before the split.  If the
    713 	 * cursor is at or past the skipped slot, the cursor is incremented by
    714 	 * one.  If the cursor is on the right page, it is decremented by the
    715 	 * number of records split to the left page.
    716 	 */
    717 	c = &t->bt_cursor;
    718 	if (F_ISSET(c, CURS_INIT) && c->pg.pgno == h->pgno) {
    719 		if (c->pg.index >= skip)
    720 			++c->pg.index;
    721 		if (c->pg.index < nxt)			/* Left page. */
    722 			c->pg.pgno = l->pgno;
    723 		else {					/* Right page. */
    724 			c->pg.pgno = r->pgno;
    725 			c->pg.index -= nxt;
    726 		}
    727 	}
    728 
    729 	/*
    730 	 * If the skipped index was on the left page, just return that page.
    731 	 * Otherwise, adjust the skip index to reflect the new position on
    732 	 * the right page.
    733 	 */
    734 	if (skip <= off) {
    735 		skip = MAX_PAGE_OFFSET;
    736 		rval = l;
    737 	} else {
    738 		rval = r;
    739 		*pskip -= nxt;
    740 	}
    741 
    742 	for (off = 0; nxt < top; ++off) {
    743 		if (skip == nxt) {
    744 			++off;
    745 			skip = MAX_PAGE_OFFSET;
    746 		}
    747 		switch (h->flags & P_TYPE) {
    748 		case P_BINTERNAL:
    749 			src = bi = GETBINTERNAL(h, nxt);
    750 			nbytes = NBINTERNAL(bi->ksize);
    751 			break;
    752 		case P_BLEAF:
    753 			src = bl = GETBLEAF(h, nxt);
    754 			nbytes = NBLEAF(bl);
    755 			break;
    756 		case P_RINTERNAL:
    757 			src = GETRINTERNAL(h, nxt);
    758 			nbytes = NRINTERNAL;
    759 			break;
    760 		case P_RLEAF:
    761 			src = rl = GETRLEAF(h, nxt);
    762 			nbytes = NRLEAF(rl);
    763 			break;
    764 		default:
    765 			abort();
    766 		}
    767 		++nxt;
    768 		r->linp[off] = r->upper -= nbytes;
    769 		memmove((char *)(void *)r + r->upper, src, nbytes);
    770 	}
    771 	r->lower += off * sizeof(indx_t);
    772 
    773 	/* If the key is being appended to the page, adjust the index. */
    774 	if (skip == top)
    775 		r->lower += sizeof(indx_t);
    776 
    777 	return (rval);
    778 }
    779 
    780 /*
    781  * BT_PRESERVE -- Mark a chain of pages as used by an internal node.
    782  *
    783  * Chains of indirect blocks pointed to by leaf nodes get reclaimed when the
    784  * record that references them gets deleted.  Chains pointed to by internal
    785  * pages never get deleted.  This routine marks a chain as pointed to by an
    786  * internal page.
    787  *
    788  * Parameters:
    789  *	t:	tree
    790  *	pg:	page number of first page in the chain.
    791  *
    792  * Returns:
    793  *	RET_SUCCESS, RET_ERROR.
    794  */
    795 static int
    796 bt_preserve(t, pg)
    797 	BTREE *t;
    798 	pgno_t pg;
    799 {
    800 	PAGE *h;
    801 
    802 	if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL)
    803 		return (RET_ERROR);
    804 	h->flags |= P_PRESERVE;
    805 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    806 	return (RET_SUCCESS);
    807 }
    808 
    809 /*
    810  * REC_TOTAL -- Return the number of recno entries below a page.
    811  *
    812  * Parameters:
    813  *	h:	page
    814  *
    815  * Returns:
    816  *	The number of recno entries below a page.
    817  *
    818  * XXX
    819  * These values could be set by the bt_psplit routine.  The problem is that the
    820  * entry has to be popped off of the stack etc. or the values have to be passed
    821  * all the way back to bt_split/bt_rroot and it's not very clean.
    822  */
    823 static recno_t
    824 rec_total(h)
    825 	PAGE *h;
    826 {
    827 	recno_t recs;
    828 	indx_t nxt, top;
    829 
    830 	for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt)
    831 		recs += GETRINTERNAL(h, nxt)->nrecs;
    832 	return (recs);
    833 }
    834