xref: /src/lib/libc/db/hash/hash_page.c

/*	$NetBSD: hash_page.c,v 1.23 2008/09/11 12:58:00 joerg Exp $	*/

/*-
 * Copyright (c) 1990, 1993, 1994
 *	The Regents of the University of California.  All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Margo Seltzer.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

#if HAVE_NBTOOL_CONFIG_H
#include "nbtool_config.h"
#endif

#include <sys/cdefs.h>
__RCSID("$NetBSD: hash_page.c,v 1.23 2008/09/11 12:58:00 joerg Exp $");

/*
 * PACKAGE:  hashing
 *
 * DESCRIPTION:
 *	Page manipulation for hashing package.
 *
 * ROUTINES:
 *
 * External
 *	__get_page
 *	__add_ovflpage
 * Internal
 *	overflow_page
 *	open_temp
 */

#include "namespace.h"

#include <sys/types.h>

#include <errno.h>
#include <fcntl.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <paths.h>
#include <assert.h>

#include <db.h>
#include "hash.h"
#include "page.h"
#include "extern.h"

static uint32_t	*fetch_bitmap(HTAB *, int);
static uint32_t	 first_free(uint32_t);
static int	 open_temp(HTAB *);
static uint16_t	 overflow_page(HTAB *);
static void	 putpair(char *, const DBT *, const DBT *);
static void	 squeeze_key(uint16_t *, const DBT *, const DBT *);
static int	 ugly_split(HTAB *, uint32_t, BUFHEAD *, BUFHEAD *, int, int);

#define	PAGE_INIT(P) { \
	((uint16_t *)(void *)(P))[0] = 0; \
	temp = 3 * sizeof(uint16_t); \
	_DIAGASSERT(hashp->BSIZE >= temp); \
	((uint16_t *)(void *)(P))[1] = (uint16_t)(hashp->BSIZE - temp); \
	((uint16_t *)(void *)(P))[2] = hashp->BSIZE; \
}

/*
 * This is called AFTER we have verified that there is room on the page for
 * the pair (PAIRFITS has returned true) so we go right ahead and start moving
 * stuff on.
 */
static void
putpair(char *p, const DBT *key, const DBT *val)
{
	uint16_t *bp, n, off;
	size_t temp;

	bp = (uint16_t *)(void *)p;

	/* Enter the key first. */
	n = bp[0];

	temp = OFFSET(bp);
	_DIAGASSERT(temp >= key->size);
	off = (uint16_t)(temp - key->size);
	memmove(p + off, key->data, key->size);
	bp[++n] = off;

	/* Now the data. */
	_DIAGASSERT(off >= val->size);
	off -= (uint16_t)val->size;
	memmove(p + off, val->data, val->size);
	bp[++n] = off;

	/* Adjust page info. */
	bp[0] = n;
	temp = (n + 3) * sizeof(uint16_t);
	_DIAGASSERT(off >= temp);
	bp[n + 1] = (uint16_t)(off - temp);
	bp[n + 2] = off;
}

/*
 * Returns:
 *	 0 OK
 *	-1 error
 */
int
__delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
{
	uint16_t *bp, newoff;
	int n;
	uint16_t pairlen;
	size_t temp;

	bp = (uint16_t *)(void *)bufp->page;
	n = bp[0];

	if (bp[ndx + 1] < REAL_KEY)
		return (__big_delete(hashp, bufp));
	if (ndx != 1)
		newoff = bp[ndx - 1];
	else
		newoff = hashp->BSIZE;
	pairlen = newoff - bp[ndx + 1];

	if (ndx != (n - 1)) {
		/* Hard Case -- need to shuffle keys */
		int i;
		char *src = bufp->page + (int)OFFSET(bp);
		char *dst = src + (int)pairlen;
		memmove(dst, src, (size_t)(bp[ndx + 1] - OFFSET(bp)));

		/* Now adjust the pointers */
		for (i = ndx + 2; i <= n; i += 2) {
			if (bp[i + 1] == OVFLPAGE) {
				bp[i - 2] = bp[i];
				bp[i - 1] = bp[i + 1];
			} else {
				bp[i - 2] = bp[i] + pairlen;
				bp[i - 1] = bp[i + 1] + pairlen;
			}
		}
	}
	/* Finally adjust the page data */
	bp[n] = OFFSET(bp) + pairlen;
	temp = bp[n + 1] + pairlen + 2 * sizeof(uint16_t);
	_DIAGASSERT(temp <= 0xffff);
	bp[n - 1] = (uint16_t)temp;
	bp[0] = n - 2;
	hashp->NKEYS--;

	bufp->flags |= BUF_MOD;
	return (0);
}
/*
 * Returns:
 *	 0 ==> OK
 *	-1 ==> Error
 */
int
__split_page(HTAB *hashp, uint32_t obucket, uint32_t nbucket)
{
	BUFHEAD *new_bufp, *old_bufp;
	uint16_t *ino;
	char *np;
	DBT key, val;
	int n, ndx, retval;
	uint16_t copyto, diff, off, moved;
	char *op;
	size_t temp;

	copyto = (uint16_t)hashp->BSIZE;
	off = (uint16_t)hashp->BSIZE;
	old_bufp = __get_buf(hashp, obucket, NULL, 0);
	if (old_bufp == NULL)
		return (-1);
	new_bufp = __get_buf(hashp, nbucket, NULL, 0);
	if (new_bufp == NULL)
		return (-1);

	old_bufp->flags |= (BUF_MOD | BUF_PIN);
	new_bufp->flags |= (BUF_MOD | BUF_PIN);

	ino = (uint16_t *)(void *)(op = old_bufp->page);
	np = new_bufp->page;

	moved = 0;

	for (n = 1, ndx = 1; n < ino[0]; n += 2) {
		if (ino[n + 1] < REAL_KEY) {
			retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
			    (int)copyto, (int)moved);
			old_bufp->flags &= ~BUF_PIN;
			new_bufp->flags &= ~BUF_PIN;
			return (retval);

		}
		key.data = (uint8_t *)op + ino[n];
		key.size = off - ino[n];

		if (__call_hash(hashp, key.data, (int)key.size) == obucket) {
			/* Don't switch page */
			diff = copyto - off;
			if (diff) {
				copyto = ino[n + 1] + diff;
				memmove(op + copyto, op + ino[n + 1],
				    (size_t)(off - ino[n + 1]));
				ino[ndx] = copyto + ino[n] - ino[n + 1];
				ino[ndx + 1] = copyto;
			} else
				copyto = ino[n + 1];
			ndx += 2;
		} else {
			/* Switch page */
			val.data = (uint8_t *)op + ino[n + 1];
			val.size = ino[n] - ino[n + 1];
			putpair(np, &key, &val);
			moved += 2;
		}

		off = ino[n + 1];
	}

	/* Now clean up the page */
	ino[0] -= moved;
	temp = sizeof(uint16_t) * (ino[0] + 3);
	_DIAGASSERT(copyto >= temp);
	FREESPACE(ino) = (uint16_t)(copyto - temp);
	OFFSET(ino) = copyto;

#ifdef DEBUG3
	(void)fprintf(stderr, "split %d/%d\n",
	    ((uint16_t *)np)[0] / 2,
	    ((uint16_t *)op)[0] / 2);
#endif
	/* unpin both pages */
	old_bufp->flags &= ~BUF_PIN;
	new_bufp->flags &= ~BUF_PIN;
	return (0);
}

/*
 * Called when we encounter an overflow or big key/data page during split
 * handling.  This is special cased since we have to begin checking whether
 * the key/data pairs fit on their respective pages and because we may need
 * overflow pages for both the old and new pages.
 *
 * The first page might be a page with regular key/data pairs in which case
 * we have a regular overflow condition and just need to go on to the next
 * page or it might be a big key/data pair in which case we need to fix the
 * big key/data pair.
 *
 * Returns:
 *	 0 ==> success
 *	-1 ==> failure
 */
static int
ugly_split(
	HTAB *hashp,
	uint32_t obucket,	/* Same as __split_page. */
	BUFHEAD *old_bufp,
	BUFHEAD *new_bufp,
	int copyto,	/* First byte on page which contains key/data values. */
	int moved	/* Number of pairs moved to new page. */
)
{
	BUFHEAD *bufp;	/* Buffer header for ino */
	uint16_t *ino;	/* Page keys come off of */
	uint16_t *np;	/* New page */
	uint16_t *op;	/* Page keys go on to if they aren't moving */
	size_t temp;

	BUFHEAD *last_bfp;	/* Last buf header OVFL needing to be freed */
	DBT key, val;
	SPLIT_RETURN ret;
	uint16_t n, off, ov_addr, scopyto;
	char *cino;		/* Character value of ino */

	bufp = old_bufp;
	ino = (uint16_t *)(void *)old_bufp->page;
	np = (uint16_t *)(void *)new_bufp->page;
	op = (uint16_t *)(void *)old_bufp->page;
	last_bfp = NULL;
	scopyto = (uint16_t)copyto;	/* ANSI */

	n = ino[0] - 1;
	while (n < ino[0]) {
		if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
			if (__big_split(hashp, old_bufp,
			    new_bufp, bufp, (int)bufp->addr, obucket, &ret))
				return (-1);
			old_bufp = ret.oldp;
			if (!old_bufp)
				return (-1);
			op = (uint16_t *)(void *)old_bufp->page;
			new_bufp = ret.newp;
			if (!new_bufp)
				return (-1);
			np = (uint16_t *)(void *)new_bufp->page;
			bufp = ret.nextp;
			if (!bufp)
				return (0);
			cino = (char *)bufp->page;
			ino = (uint16_t *)(void *)cino;
			last_bfp = ret.nextp;
		} else if (ino[n + 1] == OVFLPAGE) {
			ov_addr = ino[n];
			/*
			 * Fix up the old page -- the extra 2 are the fields
			 * which contained the overflow information.
			 */
			ino[0] -= (moved + 2);
			temp = sizeof(uint16_t) * (ino[0] + 3);
			_DIAGASSERT(scopyto >= temp);
			FREESPACE(ino) = (uint16_t)(scopyto - temp);
			OFFSET(ino) = scopyto;

			bufp = __get_buf(hashp, (uint32_t)ov_addr, bufp, 0);
			if (!bufp)
				return (-1);

			ino = (uint16_t *)(void *)bufp->page;
			n = 1;
			scopyto = hashp->BSIZE;
			moved = 0;

			if (last_bfp)
				__free_ovflpage(hashp, last_bfp);
			last_bfp = bufp;
		}
		/* Move regular sized pairs of there are any */
		off = hashp->BSIZE;
		for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
			cino = (char *)(void *)ino;
			key.data = (uint8_t *)cino + ino[n];
			key.size = off - ino[n];
			val.data = (uint8_t *)cino + ino[n + 1];
			val.size = ino[n] - ino[n + 1];
			off = ino[n + 1];

			if (__call_hash(hashp, key.data, (int)key.size) == obucket) {
				/* Keep on old page */
				if (PAIRFITS(op, (&key), (&val)))
					putpair((char *)(void *)op, &key, &val);
				else {
					old_bufp =
					    __add_ovflpage(hashp, old_bufp);
					if (!old_bufp)
						return (-1);
					op = (uint16_t *)(void *)old_bufp->page;
					putpair((char *)(void *)op, &key, &val);
				}
				old_bufp->flags |= BUF_MOD;
			} else {
				/* Move to new page */
				if (PAIRFITS(np, (&key), (&val)))
					putpair((char *)(void *)np, &key, &val);
				else {
					new_bufp =
					    __add_ovflpage(hashp, new_bufp);
					if (!new_bufp)
						return (-1);
					np = (uint16_t *)(void *)new_bufp->page;
					putpair((char *)(void *)np, &key, &val);
				}
				new_bufp->flags |= BUF_MOD;
			}
		}
	}
	if (last_bfp)
		__free_ovflpage(hashp, last_bfp);
	return (0);
}

/*
 * Add the given pair to the page
 *
 * Returns:
 *	0 ==> OK
 *	1 ==> failure
 */
int
__addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
{
	uint16_t *bp, *sop;
	int do_expand;

	bp = (uint16_t *)(void *)bufp->page;
	do_expand = 0;
	while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
		/* Exception case */
		if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
			/* This is the last page of a big key/data pair
			   and we need to add another page */
			break;
		else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
			bufp = __get_buf(hashp, (uint32_t)bp[bp[0] - 1], bufp,
			    0);
			if (!bufp)
				return (-1);
			bp = (uint16_t *)(void *)bufp->page;
		} else if (bp[bp[0]] != OVFLPAGE) {
			/* Short key/data pairs, no more pages */
			break;
		} else {
			/* Try to squeeze key on this page */
			if (bp[2] >= REAL_KEY &&
			    FREESPACE(bp) >= PAIRSIZE(key, val)) {
				squeeze_key(bp, key, val);
				goto stats;
			} else {
				bufp = __get_buf(hashp,
				    (uint32_t)bp[bp[0] - 1], bufp, 0);
				if (!bufp)
					return (-1);
				bp = (uint16_t *)(void *)bufp->page;
			}
		}

	if (PAIRFITS(bp, key, val))
		putpair(bufp->page, key, val);
	else {
		do_expand = 1;
		bufp = __add_ovflpage(hashp, bufp);
		if (!bufp)
			return (-1);
		sop = (uint16_t *)(void *)bufp->page;

		if (PAIRFITS(sop, key, val))
			putpair((char *)(void *)sop, key, val);
		else
			if (__big_insert(hashp, bufp, key, val))
				return (-1);
	}
stats:
	bufp->flags |= BUF_MOD;
	/*
	 * If the average number of keys per bucket exceeds the fill factor,
	 * expand the table.
	 */
	hashp->NKEYS++;
	if (do_expand ||
	    (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
		return (__expand_table(hashp));
	return (0);
}

/*
 *
 * Returns:
 *	pointer on success
 *	NULL on error
 */
BUFHEAD *
__add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
{
	uint16_t *sp;
	uint16_t ndx, ovfl_num;
	size_t temp;
#ifdef DEBUG1
	int tmp1, tmp2;
#endif
	sp = (uint16_t *)(void *)bufp->page;

	/* Check if we are dynamically determining the fill factor */
	if (hashp->FFACTOR == DEF_FFACTOR) {
		hashp->FFACTOR = (uint32_t)sp[0] >> 1;
		if (hashp->FFACTOR < MIN_FFACTOR)
			hashp->FFACTOR = MIN_FFACTOR;
	}
	bufp->flags |= BUF_MOD;
	ovfl_num = overflow_page(hashp);
#ifdef DEBUG1
	tmp1 = bufp->addr;
	tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
#endif
	if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, (uint32_t)ovfl_num,
	    bufp, 1)))
		return (NULL);
	bufp->ovfl->flags |= BUF_MOD;
#ifdef DEBUG1
	(void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
	    tmp1, tmp2, bufp->ovfl->addr);
#endif
	ndx = sp[0];
	/*
	 * Since a pair is allocated on a page only if there's room to add
	 * an overflow page, we know that the OVFL information will fit on
	 * the page.
	 */
	sp[ndx + 4] = OFFSET(sp);
	temp = FREESPACE(sp);
	_DIAGASSERT(temp >= OVFLSIZE);
	sp[ndx + 3] = (uint16_t)(temp - OVFLSIZE);
	sp[ndx + 1] = ovfl_num;
	sp[ndx + 2] = OVFLPAGE;
	sp[0] = ndx + 2;
#ifdef HASH_STATISTICS
	hash_overflows++;
#endif
	return (bufp->ovfl);
}

/*
 * Returns:
 *	 0 indicates SUCCESS
 *	-1 indicates FAILURE
 */
int
__get_page(HTAB *hashp, char *p, uint32_t bucket, int is_bucket, int is_disk,
    int is_bitmap)
{
	int fd, page, size;
	ssize_t rsize;
	uint16_t *bp;
	size_t temp;

	fd = hashp->fp;
	size = hashp->BSIZE;

	if ((fd == -1) || !is_disk) {
		PAGE_INIT(p);
		return (0);
	}
	if (is_bucket)
		page = BUCKET_TO_PAGE(bucket);
	else
		page = OADDR_TO_PAGE(bucket);
	if ((rsize = pread(fd, p, (size_t)size, (off_t)page << hashp->BSHIFT)) == -1)
		return (-1);
	bp = (uint16_t *)(void *)p;
	if (!rsize)
		bp[0] = 0;	/* We hit the EOF, so initialize a new page */
	else
		if (rsize != size) {
			errno = EFTYPE;
			return (-1);
		}
	if (!is_bitmap && !bp[0]) {
		PAGE_INIT(p);
	} else
		if (hashp->LORDER != BYTE_ORDER) {
			int i, max;

			if (is_bitmap) {
				max = (uint32_t)hashp->BSIZE >> 2; /* divide by 4 */
				for (i = 0; i < max; i++)
					M_32_SWAP(((int *)(void *)p)[i]);
			} else {
				M_16_SWAP(bp[0]);
				max = bp[0] + 2;
				for (i = 1; i <= max; i++)
					M_16_SWAP(bp[i]);
			}
		}
	return (0);
}

/*
 * Write page p to disk
 *
 * Returns:
 *	 0 ==> OK
 *	-1 ==>failure
 */
int
__put_page(HTAB *hashp, char *p, uint32_t bucket, int is_bucket, int is_bitmap)
{
	int fd, page, size;
	ssize_t wsize;

	size = hashp->BSIZE;
	if ((hashp->fp == -1) && open_temp(hashp))
		return (-1);
	fd = hashp->fp;

	if (hashp->LORDER != BYTE_ORDER) {
		int i;
		int max;

		if (is_bitmap) {
			max = (uint32_t)hashp->BSIZE >> 2;	/* divide by 4 */
			for (i = 0; i < max; i++)
				M_32_SWAP(((int *)(void *)p)[i]);
		} else {
			max = ((uint16_t *)(void *)p)[0] + 2;
			for (i = 0; i <= max; i++)
				M_16_SWAP(((uint16_t *)(void *)p)[i]);
		}
	}
	if (is_bucket)
		page = BUCKET_TO_PAGE(bucket);
	else
		page = OADDR_TO_PAGE(bucket);
	if ((wsize = pwrite(fd, p, (size_t)size, (off_t)page << hashp->BSHIFT)) == -1)
		/* Errno is set */
		return (-1);
	if (wsize != size) {
		errno = EFTYPE;
		return (-1);
	}
	return (0);
}

#define BYTE_MASK	((1 << INT_BYTE_SHIFT) -1)
/*
 * Initialize a new bitmap page.  Bitmap pages are left in memory
 * once they are read in.
 */
int
__ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
{
	uint32_t *ip;
	int clearbytes, clearints;

	if ((ip = malloc((size_t)hashp->BSIZE)) == NULL)
		return (1);
	hashp->nmaps++;
	clearints = ((uint32_t)(nbits - 1) >> INT_BYTE_SHIFT) + 1;
	clearbytes = clearints << INT_TO_BYTE;
	(void)memset(ip, 0, (size_t)clearbytes);
	(void)memset(((char *)(void *)ip) + clearbytes, 0xFF,
	    (size_t)(hashp->BSIZE - clearbytes));
	ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
	SETBIT(ip, 0);
	hashp->BITMAPS[ndx] = (uint16_t)pnum;
	hashp->mapp[ndx] = ip;
	return (0);
}

static uint32_t
first_free(uint32_t map)
{
	uint32_t i, mask;

	mask = 0x1;
	for (i = 0; i < BITS_PER_MAP; i++) {
		if (!(mask & map))
			return (i);
		mask = mask << 1;
	}
	return (i);
}

static uint16_t
overflow_page(HTAB *hashp)
{
	uint32_t *freep = NULL;
	int max_free, offset, splitnum;
	uint16_t addr;
	int bit, first_page, free_bit, free_page, i, in_use_bits, j;
#ifdef DEBUG2
	int tmp1, tmp2;
#endif
	splitnum = hashp->OVFL_POINT;
	max_free = hashp->SPARES[splitnum];

	free_page = (uint32_t)(max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
	free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);

	/* Look through all the free maps to find the first free block */
	first_page = (uint32_t)hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
	for ( i = first_page; i <= free_page; i++ ) {
		if (!(freep = (uint32_t *)hashp->mapp[i]) &&
		    !(freep = fetch_bitmap(hashp, i)))
			return (0);
		if (i == free_page)
			in_use_bits = free_bit;
		else
			in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;

		if (i == first_page) {
			bit = hashp->LAST_FREED &
			    ((hashp->BSIZE << BYTE_SHIFT) - 1);
			j = bit / BITS_PER_MAP;
			bit = bit & ~(BITS_PER_MAP - 1);
		} else {
			bit = 0;
			j = 0;
		}
		for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
			if (freep[j] != ALL_SET)
				goto found;
	}

	/* No Free Page Found */
	hashp->LAST_FREED = hashp->SPARES[splitnum];
	hashp->SPARES[splitnum]++;
	offset = hashp->SPARES[splitnum] -
	    (splitnum ? hashp->SPARES[splitnum - 1] : 0);

#define	OVMSG	"HASH: Out of overflow pages.  Increase page size\n"
	if (offset > SPLITMASK) {
		if (++splitnum >= NCACHED) {
			(void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
			errno = EFBIG;
			return (0);
		}
		hashp->OVFL_POINT = splitnum;
		hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
		hashp->SPARES[splitnum-1]--;
		offset = 1;
	}

	/* Check if we need to allocate a new bitmap page */
	if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
		free_page++;
		if (free_page >= NCACHED) {
			(void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
			errno = EFBIG;
			return (0);
		}
		/*
		 * This is tricky.  The 1 indicates that you want the new page
		 * allocated with 1 clear bit.  Actually, you are going to
		 * allocate 2 pages from this map.  The first is going to be
		 * the map page, the second is the overflow page we were
		 * looking for.  The init_bitmap routine automatically, sets
		 * the first bit of itself to indicate that the bitmap itself
		 * is in use.  We would explicitly set the second bit, but
		 * don't have to if we tell init_bitmap not to leave it clear
		 * in the first place.
		 */
		if (__ibitmap(hashp,
		    (int)OADDR_OF(splitnum, offset), 1, free_page))
			return (0);
		hashp->SPARES[splitnum]++;
#ifdef DEBUG2
		free_bit = 2;
#endif
		offset++;
		if (offset > SPLITMASK) {
			if (++splitnum >= NCACHED) {
				(void)write(STDERR_FILENO, OVMSG,
				    sizeof(OVMSG) - 1);
				errno = EFBIG;
				return (0);
			}
			hashp->OVFL_POINT = splitnum;
			hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
			hashp->SPARES[splitnum-1]--;
			offset = 0;
		}
	} else {
		/*
		 * Free_bit addresses the last used bit.  Bump it to address
		 * the first available bit.
		 */
		free_bit++;
		SETBIT(freep, free_bit);
	}

	/* Calculate address of the new overflow page */
	addr = OADDR_OF(splitnum, offset);
#ifdef DEBUG2
	(void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
	    addr, free_bit, free_page);
#endif
	return (addr);

found:
	bit = bit + first_free(freep[j]);
	SETBIT(freep, bit);
#ifdef DEBUG2
	tmp1 = bit;
	tmp2 = i;
#endif
	/*
	 * Bits are addressed starting with 0, but overflow pages are addressed
	 * beginning at 1. Bit is a bit addressnumber, so we need to increment
	 * it to convert it to a page number.
	 */
	bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
	if (bit >= hashp->LAST_FREED)
		hashp->LAST_FREED = bit - 1;

	/* Calculate the split number for this page */
	for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
	offset = (i ? bit - hashp->SPARES[i - 1] : bit);
	if (offset >= SPLITMASK) {
		(void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
		errno = EFBIG;
		return (0);	/* Out of overflow pages */
	}
	addr = OADDR_OF(i, offset);
#ifdef DEBUG2
	(void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
	    addr, tmp1, tmp2);
#endif

	/* Allocate and return the overflow page */
	return (addr);
}

/*
 * Mark this overflow page as free.
 */
void
__free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
{
	uint16_t addr;
	uint32_t *freep;
	int bit_address, free_page, free_bit;
	uint16_t ndx;

	addr = obufp->addr;
#ifdef DEBUG1
	(void)fprintf(stderr, "Freeing %d\n", addr);
#endif
	ndx = (((uint32_t)addr) >> SPLITSHIFT);
	bit_address =
	    (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
	 if (bit_address < hashp->LAST_FREED)
		hashp->LAST_FREED = bit_address;
	free_page = ((uint32_t)bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
	free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);

	if (!(freep = hashp->mapp[free_page]))
		freep = fetch_bitmap(hashp, free_page);
	/*
	 * This had better never happen.  It means we tried to read a bitmap
	 * that has already had overflow pages allocated off it, and we
	 * failed to read it from the file.
	 */
	_DIAGASSERT(freep != NULL);
	CLRBIT(freep, free_bit);
#ifdef DEBUG2
	(void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
	    obufp->addr, free_bit, free_page);
#endif
	__reclaim_buf(hashp, obufp);
}

/*
 * Returns:
 *	 0 success
 *	-1 failure
 */
static int
open_temp(HTAB *hashp)
{
	sigset_t set, oset;
	char *envtmp;
	char namestr[PATH_MAX];

	if (issetugid())
		envtmp = NULL;
	else
		envtmp = getenv("TMPDIR");

	if (-1 == snprintf(namestr, sizeof(namestr), "%s/_hashXXXXXX",
	    envtmp ? envtmp : _PATH_TMP))
		return -1;

	/* Block signals; make sure file goes away at process exit. */
	(void)sigfillset(&set);
	(void)sigprocmask(SIG_BLOCK, &set, &oset);
	if ((hashp->fp = mkstemp(namestr)) != -1) {
		(void)unlink(namestr);
		(void)fcntl(hashp->fp, F_SETFD, FD_CLOEXEC);
	}
	(void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
	return (hashp->fp != -1 ? 0 : -1);
}

/*
 * We have to know that the key will fit, but the last entry on the page is
 * an overflow pair, so we need to shift things.
 */
static void
squeeze_key(uint16_t *sp, const DBT *key, const DBT *val)
{
	char *p;
	uint16_t free_space, n, off, pageno;
	size_t temp;

	p = (char *)(void *)sp;
	n = sp[0];
	free_space = FREESPACE(sp);
	off = OFFSET(sp);

	pageno = sp[n - 1];
	_DIAGASSERT(off >= key->size);
	off -= (uint16_t)key->size;
	sp[n - 1] = off;
	memmove(p + off, key->data, key->size);
	_DIAGASSERT(off >= val->size);
	off -= (uint16_t)val->size;
	sp[n] = off;
	memmove(p + off, val->data, val->size);
	sp[0] = n + 2;
	sp[n + 1] = pageno;
	sp[n + 2] = OVFLPAGE;
	temp = PAIRSIZE(key, val);
	_DIAGASSERT(free_space >= temp);
	FREESPACE(sp) = (uint16_t)(free_space - temp);
	OFFSET(sp) = off;
}

static uint32_t *
fetch_bitmap(HTAB *hashp, int ndx)
{
	if (ndx >= hashp->nmaps)
		return (NULL);
	if ((hashp->mapp[ndx] = malloc((size_t)hashp->BSIZE)) == NULL)
		return (NULL);
	if (__get_page(hashp,
	    (char *)(void *)hashp->mapp[ndx], (uint32_t)hashp->BITMAPS[ndx], 0, 1, 1)) {
		free(hashp->mapp[ndx]);
		return (NULL);
	}
	return (hashp->mapp[ndx]);
}

#ifdef DEBUG4
void print_chain(HTAB *, uint32_t);
void
print_chain(HTAB *hashp, uint32_t addr)
{
	BUFHEAD *bufp;
	uint16_t *bp, oaddr;

	(void)fprintf(stderr, "%d ", addr);
	bufp = __get_buf(hashp, addr, NULL, 0);
	bp = (uint16_t *)bufp->page;
	while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
		((bp[0] > 2) && bp[2] < REAL_KEY))) {
		oaddr = bp[bp[0] - 1];
		(void)fprintf(stderr, "%d ", (int)oaddr);
		bufp = __get_buf(hashp, (uint32_t)oaddr, bufp, 0);
		bp = (uint16_t *)bufp->page;
	}
	(void)fprintf(stderr, "\n");
}
#endif