Minor changes from Berkeley DB 1.86 and further improvements from OpenBSD.

This does not include the new hash routines since they will cause problems
when reading old hash files.

Since mpool(3) has been changed, provide a compatibility shim for older
binaries.

Obtained from:	OpenBSD

11 files changed, 237 insertions, 130 deletions

diff --git a/include/mpool.h b/include/mpool.h
index b5be330..c74764d 100644
--- a/include/mpool.h
+++ b/include/mpool.h
@@ -30,7 +30,7 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- *	@(#)mpool.h	8.2 (Berkeley) 7/14/94
+ *	@(#)mpool.h	8.4 (Berkeley) 11/2/95
  * $FreeBSD$
  */
 
@@ -47,7 +47,7 @@
  * pool is handed an opaque MPOOL cookie which stores all of this information.
  */
 #define	HASHSIZE	128
-#define	HASHKEY(pgno)	((pgno - 1) % HASHSIZE)
+#define	HASHKEY(pgno)	((pgno - 1 + HASHSIZE) % HASHSIZE)
 
 /* The BKT structures are the elements of the queues. */
 typedef struct _bkt {
@@ -58,6 +58,7 @@ typedef struct _bkt {
 
 #define	MPOOL_DIRTY	0x01		/* page needs to be written */
 #define	MPOOL_PINNED	0x02		/* page is pinned into memory */
+#define	MPOOL_INUSE	0x04		/* page address is valid */
 	u_int8_t flags;			/* flags */
 } BKT;
 
@@ -68,7 +69,7 @@ typedef struct MPOOL {
 	pgno_t	curcache;		/* current number of cached pages */
 	pgno_t	maxcache;		/* max number of cached pages */
 	pgno_t	npages;			/* number of pages in the file */
-	u_long	pagesize;		/* file page size */
+	unsigned long	pagesize;	/* file page size */
 	int	fd;			/* file descriptor */
 					/* page in conversion routine */
 	void    (*pgin)(void *, pgno_t, void *);
@@ -76,25 +77,32 @@ typedef struct MPOOL {
 	void    (*pgout)(void *, pgno_t, void *);
 	void	*pgcookie;		/* cookie for page in/out routines */
 #ifdef STATISTICS
-	u_long	cachehit;
-	u_long	cachemiss;
-	u_long	pagealloc;
-	u_long	pageflush;
-	u_long	pageget;
-	u_long	pagenew;
-	u_long	pageput;
-	u_long	pageread;
-	u_long	pagewrite;
+	unsigned long	cachehit;
+	unsigned long	cachemiss;
+	unsigned long	pagealloc;
+	unsigned long	pageflush;
+	unsigned long	pageget;
+	unsigned long	pagenew;
+	unsigned long	pageput;
+	unsigned long	pageread;
+	unsigned long	pagewrite;
 #endif
 } MPOOL;
 
+#define	MPOOL_IGNOREPIN	0x01		/* Ignore if the page is pinned. */
+#define	MPOOL_PAGE_REQUEST	0x01	/* Allocate a new page with a
+					   specific page number. */
+#define	MPOOL_PAGE_NEXT		0x02	/* Allocate a new page with the next
+					  page number. */
+
 __BEGIN_DECLS
 MPOOL	*mpool_open(void *, int, pgno_t, pgno_t);
 void	 mpool_filter(MPOOL *, void (*)(void *, pgno_t, void *),
 	    void (*)(void *, pgno_t, void *), void *);
-void	*mpool_new(MPOOL *, pgno_t *);
-void	*mpool_get(MPOOL *, pgno_t, u_int);
-int	 mpool_put(MPOOL *, void *, u_int);
+void	*mpool_new(MPOOL *, pgno_t *, unsigned int);
+void	*mpool_get(MPOOL *, pgno_t, unsigned int);
+int	 mpool_delete(MPOOL *, void *);
+int	 mpool_put(MPOOL *, void *, unsigned int);
 int	 mpool_sync(MPOOL *);
 int	 mpool_close(MPOOL *);
 #ifdef STATISTICS
diff --git a/lib/libc/db/Symbol.map b/lib/libc/db/Symbol.map
index 14df29f..225fd40 100644
--- a/lib/libc/db/Symbol.map
+++ b/lib/libc/db/Symbol.map
@@ -16,7 +16,6 @@ FBSD_1.0 {
 	dbm_dirfno;
 	mpool_open;
 	mpool_filter;
-	mpool_new;
 	mpool_get;
 	mpool_put;
 	mpool_close;
@@ -24,6 +23,11 @@ FBSD_1.0 {
 	mpool_stat;
 };
 
+FBSD_1.1 {
+	mpool_new;
+	mpool_delete;
+};
+
 FBSDprivate_1.0 {
 	__bt_open;
 	__dbpanic;
diff --git a/lib/libc/db/btree/bt_debug.c b/lib/libc/db/btree/bt_debug.c
index ff5357a..dc8b83d 100644
--- a/lib/libc/db/btree/bt_debug.c
+++ b/lib/libc/db/btree/bt_debug.c
@@ -83,10 +83,9 @@ __bt_dump(DB *dbp)
 	}
 #undef X
 
-	for (i = P_ROOT; (h = mpool_get(t->bt_mp, i, 0)) != NULL; ++i) {
+	for (i = P_ROOT;
+	    (h = mpool_get(t->bt_mp, i, MPOOL_IGNOREPIN)) != NULL; ++i)
 		__bt_dpage(h);
-		(void)mpool_put(t->bt_mp, h, 0);
-	}
 }
 
 /*
@@ -135,10 +134,8 @@ __bt_dnpage(DB *dbp, pgno_t pgno)
 	PAGE *h;
 
 	t = dbp->internal;
-	if ((h = mpool_get(t->bt_mp, pgno, 0)) != NULL) {
+	if ((h = mpool_get(t->bt_mp, pgno, MPOOL_IGNOREPIN)) != NULL)
 		__bt_dpage(h);
-		(void)mpool_put(t->bt_mp, h, 0);
-	}
 }
 
 /*
@@ -257,7 +254,8 @@ __bt_stat(DB *dbp)
 	t = dbp->internal;
 	pcont = pinternal = pleaf = 0;
 	nkeys = ifree = lfree = 0;
-	for (i = P_ROOT; (h = mpool_get(t->bt_mp, i, 0)) != NULL; ++i) {
+	for (i = P_ROOT;
+	    (h = mpool_get(t->bt_mp, i, MPOOL_IGNOREPIN)) != NULL; ++i)
 		switch (h->flags & P_TYPE) {
 		case P_BINTERNAL:
 		case P_RINTERNAL:
@@ -274,22 +272,18 @@ __bt_stat(DB *dbp)
 			++pcont;
 			break;
 		}
-		(void)mpool_put(t->bt_mp, h, 0);
-	}
 
 	/* Count the levels of the tree. */
 	for (i = P_ROOT, levels = 0 ;; ++levels) {
-		h = mpool_get(t->bt_mp, i, 0);
+		h = mpool_get(t->bt_mp, i, MPOOL_IGNOREPIN);
 		if (h->flags & (P_BLEAF|P_RLEAF)) {
 			if (levels == 0)
 				levels = 1;
-			(void)mpool_put(t->bt_mp, h, 0);
 			break;
 		}
 		i = F_ISSET(t, R_RECNO) ?
 		    GETRINTERNAL(h, 0)->pgno :
 		    GETBINTERNAL(h, 0)->pgno;
-		(void)mpool_put(t->bt_mp, h, 0);
 	}
 
 	(void)fprintf(stderr, "%d level%s with %lu keys",
diff --git a/lib/libc/db/btree/bt_open.c b/lib/libc/db/btree/bt_open.c
index 7422ca4..47f3646 100644
--- a/lib/libc/db/btree/bt_open.c
+++ b/lib/libc/db/btree/bt_open.c
@@ -352,18 +352,25 @@ nroot(BTREE *t)
 	PAGE *meta, *root;
 	pgno_t npg;
 
-	if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) {
-		mpool_put(t->bt_mp, meta, 0);
-		return (RET_SUCCESS);
+	if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) {
+		if (root->lower == 0 &&
+		    root->pgno == 0 &&
+		    root->linp[0] == 0) {
+			mpool_delete(t->bt_mp, root);
+			errno = EINVAL;
+		} else {
+			mpool_put(t->bt_mp, root, 0);
+			return (RET_SUCCESS);
+		}
 	}
 	if (errno != EINVAL)		/* It's OK to not exist. */
 		return (RET_ERROR);
 	errno = 0;
 
-	if ((meta = mpool_new(t->bt_mp, &npg)) == NULL)
+	if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
 		return (RET_ERROR);
 
-	if ((root = mpool_new(t->bt_mp, &npg)) == NULL)
+	if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
 		return (RET_ERROR);
 
 	if (npg != P_ROOT)
diff --git a/lib/libc/db/btree/bt_page.c b/lib/libc/db/btree/bt_page.c
index face4b1..ab5e359 100644
--- a/lib/libc/db/btree/bt_page.c
+++ b/lib/libc/db/btree/bt_page.c
@@ -90,5 +90,5 @@ __bt_new(BTREE *t, pgno_t *npg)
 		F_SET(t, B_METADIRTY);
 		return (h);
 	}
-	return (mpool_new(t->bt_mp, npg));
+	return (mpool_new(t->bt_mp, npg, MPOOL_PAGE_NEXT));
 }
diff --git a/lib/libc/db/hash/README b/lib/libc/db/hash/README
index f29ccf7..80d6743 100644
--- a/lib/libc/db/hash/README
+++ b/lib/libc/db/hash/README
@@ -1,4 +1,5 @@
 #	@(#)README	8.1 (Berkeley) 6/4/93
+# $FreeBSD$
 
 This package implements a superset of the hsearch and dbm/ndbm libraries.
 
@@ -44,10 +45,6 @@ Test Programs:
 
 NOTES:
 
-The file search.h is provided for using the hsearch compatible interface
-on BSD systems.  On System V derived systems, search.h should appear in 
-/usr/include.
-
 The man page ../man/db.3 explains the interface to the hashing system.
 The file hash.ps is a postscript copy of a paper explaining
 the history, implementation, and performance of the hash package.
diff --git a/lib/libc/db/hash/hash_func.c b/lib/libc/db/hash/hash_func.c
index 2ca0a48..db32a75 100644
--- a/lib/libc/db/hash/hash_func.c
+++ b/lib/libc/db/hash/hash_func.c
@@ -43,60 +43,57 @@ __FBSDID("$FreeBSD$");
 #include "page.h"
 #include "extern.h"
 
+#ifdef notdef
 static u_int32_t hash1(const void *, size_t) __unused;
 static u_int32_t hash2(const void *, size_t) __unused;
 static u_int32_t hash3(const void *, size_t) __unused;
+#endif
 static u_int32_t hash4(const void *, size_t);
 
-/* Global default hash function */
+/* Default hash function. */
 u_int32_t (*__default_hash)(const void *, size_t) = hash4;
 
+#ifdef notdef
 /*
- * HASH FUNCTIONS
- *
  * Assume that we've already split the bucket to which this key hashes,
  * calculate that bucket, and check that in fact we did already split it.
  *
- * This came from ejb's hsearch.
+ * EJB's original hsearch hash.
  */
-
 #define PRIME1		37
 #define PRIME2		1048583
 
-static u_int32_t
-hash1(keyarg, len)
-	const void *keyarg;
-	size_t len;
+u_int32_t
+hash1(const void *key, size_t len)
 {
-	const u_char *key;
 	u_int32_t h;
+	u_int8_t *k;
 
+	h = 0;
+	k = (u_int8_t *)key;
 	/* Convert string to integer */
-	for (key = keyarg, h = 0; len--;)
-		h = h * PRIME1 ^ (*key++ - ' ');
+	while (len--)
+		h = h * PRIME1 ^ (*k++ - ' ');
 	h %= PRIME2;
 	return (h);
 }
 
 /*
- * Phong's linear congruential hash
+ * Phong Vo's linear congruential hash
  */
 #define dcharhash(h, c)	((h) = 0x63c63cd9*(h) + 0x9c39c33d + (c))
 
-static u_int32_t
-hash2(keyarg, len)
-	const void *keyarg;
-	size_t len;
+u_int32_t
+hash2(const void *key, size_t len)
 {
-	const u_char *e, *key;
 	u_int32_t h;
-	u_char c;
+	u_int8_t *e, c, *k;
 
-	key = keyarg;
-	e = key + len;
-	for (h = 0; key != e;) {
-		c = *key++;
-		if (!c && key > e)
+	k = (u_int8_t *)key;
+	e = k + len;
+	for (h = 0; k != e;) {
+		c = *k++;
+		if (!c && k > e)
 			break;
 		dcharhash(h, c);
 	}
@@ -110,101 +107,84 @@ hash2(keyarg, len)
  * all 8 bytes.  Essentially, this saves us 7 cmp & branch instructions.  If
  * this routine is heavily used enough, it's worth the ugly coding.
  *
- * OZ's original sdbm hash
+ * Ozan Yigit's original sdbm hash.
  */
-static u_int32_t
-hash3(keyarg, len)
-	const void *keyarg;
-	size_t len;
+u_int32_t
+hash3(const void *key, size_t len)
 {
-	const u_char *key;
-	size_t loop;
-	u_int32_t h;
+	u_int32_t n, loop;
+	u_int8_t *k;
 
-#define HASHC   h = *key++ + 65599 * h
+#define HASHC   n = *k++ + 65599 * n
 
-	h = 0;
-	key = keyarg;
+	n = 0;
+	k = (u_int8_t *)key;
 	if (len > 0) {
 		loop = (len + 8 - 1) >> 3;
 
 		switch (len & (8 - 1)) {
 		case 0:
-			do {
+			do {	/* All fall throughs */
 				HASHC;
-				/* FALLTHROUGH */
 		case 7:
 				HASHC;
-				/* FALLTHROUGH */
 		case 6:
 				HASHC;
-				/* FALLTHROUGH */
 		case 5:
 				HASHC;
-				/* FALLTHROUGH */
 		case 4:
 				HASHC;
-				/* FALLTHROUGH */
 		case 3:
 				HASHC;
-				/* FALLTHROUGH */
 		case 2:
 				HASHC;
-				/* FALLTHROUGH */
 		case 1:
 				HASHC;
 			} while (--loop);
 		}
+
 	}
-	return (h);
+	return (n);
 }
+#endif /* notdef */
 
-/* Hash function from Chris Torek. */
-static u_int32_t
-hash4(keyarg, len)
-	const void *keyarg;
-	size_t len;
+/* Chris Torek's hash function. */
+u_int32_t
+hash4(const void *key, size_t len)
 {
-	const u_char *key;
-	size_t loop;
-	u_int32_t h;
+	u_int32_t h, loop;
+	const u_int8_t *k;
 
-#define HASH4a   h = (h << 5) - h + *key++;
-#define HASH4b   h = (h << 5) + h + *key++;
+#define HASH4a   h = (h << 5) - h + *k++;
+#define HASH4b   h = (h << 5) + h + *k++;
 #define HASH4 HASH4b
 
 	h = 0;
-	key = keyarg;
+	k = key;
 	if (len > 0) {
 		loop = (len + 8 - 1) >> 3;
 
 		switch (len & (8 - 1)) {
 		case 0:
-			do {
+			do {	/* All fall throughs */
 				HASH4;
-				/* FALLTHROUGH */
 		case 7:
 				HASH4;
-				/* FALLTHROUGH */
 		case 6:
 				HASH4;
-				/* FALLTHROUGH */
 		case 5:
 				HASH4;
-				/* FALLTHROUGH */
 		case 4:
 				HASH4;
-				/* FALLTHROUGH */
 		case 3:
 				HASH4;
-				/* FALLTHROUGH */
 		case 2:
 				HASH4;
-				/* FALLTHROUGH */
 		case 1:
 				HASH4;
 			} while (--loop);
 		}
+
 	}
 	return (h);
 }
diff --git a/lib/libc/db/man/mpool.3 b/lib/libc/db/man/mpool.3
index e9918a9..3de51c3 100644
--- a/lib/libc/db/man/mpool.3
+++ b/lib/libc/db/man/mpool.3
@@ -28,7 +28,7 @@
 .\"	@(#)mpool.3	8.1 (Berkeley) 6/4/93
 .\" $FreeBSD$
 .\"
-.Dd June 4, 1993
+.Dd February 25, 1999
 .Dt MPOOL 3
 .Os
 .Sh NAME
@@ -47,7 +47,9 @@
 .Fa "void *pgcookie"
 .Fc
 .Ft void *
-.Fn mpool_new "MPOOL *mp" "pgno_t *pgnoaddr"
+.Fn mpool_new "MPOOL *mp" "pgno_t *pgnoaddr" "u_int flags"
+.Ft int
+.Fn mpool_delete "MPOOL *mp" "void *page"
 .Ft void *
 .Fn mpool_get "MPOOL *mp" "pgno_t pgno" "u_int flags"
 .Ft int
@@ -99,11 +101,11 @@ Both functions are called with the
 .Fa pgcookie
 pointer, the page number and a pointer to the page to being read or written.
 .Pp
-The
+The function
 .Fn mpool_new
-function takes an
-.Ft MPOOL
-pointer and an address as arguments.
+takes an
+.Dv MPOOL
+pointer, an address, and a set of flags as arguments.
 If a new page can be allocated, a pointer to the page is returned and
 the page number is stored into the
 .Fa pgnoaddr
@@ -113,6 +115,24 @@ Otherwise,
 is returned and
 .Va errno
 is set.
+The flags value is formed by
+.Tn OR Ns 'ing
+the following values:
+.Bl -tag -width Ds
+.It Dv MPOOL_PAGE_REQUEST
+Allocate a new page with a specific page number.
+.It Dv MPOOL_PAGE_NEXT
+Allocate a new page with the next page number.
+.El
+.Pp
+The function
+.Fn mpool_delete
+deletes the specified page from a pool and frees the page.
+It takes an
+.Dv MPOOL
+pointer and a page as arguments.
+The page must have been generated by
+.Fn mpool_new .
 .Pp
 The
 .Fn mpool_get
diff --git a/lib/libc/db/mpool/Makefile.inc b/lib/libc/db/mpool/Makefile.inc
index 9fef712..832e54d 100644
--- a/lib/libc/db/mpool/Makefile.inc
+++ b/lib/libc/db/mpool/Makefile.inc
@@ -3,4 +3,4 @@
 
 .PATH: ${.CURDIR}/db/mpool
 
-SRCS+=	mpool.c
+SRCS+=	mpool.c mpool-compat.c
diff --git a/lib/libc/db/mpool/mpool-compat.c b/lib/libc/db/mpool/mpool-compat.c
new file mode 100644
index 0000000..9df76ff
--- /dev/null
+++ b/lib/libc/db/mpool/mpool-compat.c
@@ -0,0 +1,43 @@
+/*-
+ * Copyright (c) 2009 Xin LI <delphij@FreeBSD.org>
+ * All rights reserved.
+ *
+ * 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.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
+ *
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <db.h>
+#include <mpool.h>
+
+void *__mpool_new__44bsd(MPOOL *, pgno_t *);
+
+void *
+__mpool_new__44bsd(MPOOL *mp, pgno_t *pgnoaddr)
+{
+
+	return (mpool_new(mp, pgnoaddr, MPOOL_PAGE_NEXT));
+}
+
+__sym_compat(mpool_new, __mpool_new_44bsd, FBSD_1.0);
diff --git a/lib/libc/db/mpool/mpool.c b/lib/libc/db/mpool/mpool.c
index 5610eb5..09f888b 100644
--- a/lib/libc/db/mpool/mpool.c
+++ b/lib/libc/db/mpool/mpool.c
@@ -110,9 +110,7 @@ mpool_filter(MPOOL *mp, void (*pgin) (void *, pgno_t, void *),
  *	Get a new page of memory.
  */
 void *
-mpool_new(mp, pgnoaddr)
-	MPOOL *mp;
-	pgno_t *pgnoaddr;
+mpool_new(MPOOL *mp, pgno_t *pgnoaddr, u_int flags)
 {
 	struct _hqh *head;
 	BKT *bp;
@@ -131,8 +129,13 @@ mpool_new(mp, pgnoaddr)
 	 */
 	if ((bp = mpool_bkt(mp)) == NULL)
 		return (NULL);
-	*pgnoaddr = bp->pgno = mp->npages++;
-	bp->flags = MPOOL_PINNED;
+	if (flags == MPOOL_PAGE_REQUEST) {
+		mp->npages++;
+		bp->pgno = *pgnoaddr;
+	} else
+		bp->pgno = *pgnoaddr = mp->npages++;
+
+	bp->flags = MPOOL_PINNED | MPOOL_INUSE;
 
 	head = &mp->hqh[HASHKEY(bp->pgno)];
 	TAILQ_INSERT_HEAD(head, bp, hq);
@@ -140,6 +143,32 @@ mpool_new(mp, pgnoaddr)
 	return (bp->page);
 }
 
+int
+mpool_delete(MPOOL *mp, void *page)
+{
+	struct _hqh *head;
+	BKT *bp;
+
+	bp = (BKT *)((char *)page - sizeof(BKT));
+
+#ifdef DEBUG
+	if (!(bp->flags & MPOOL_PINNED)) {
+		(void)fprintf(stderr,
+		    "mpool_delete: page %d not pinned\n", bp->pgno);
+		abort();
+	}
+#endif
+
+	/* Remove from the hash and lru queues. */
+	head = &mp->hqh[HASHKEY(bp->pgno)];
+	TAILQ_REMOVE(head, bp, hq);
+	TAILQ_REMOVE(&mp->lqh, bp, q);
+
+	free(bp);
+	mp->curcache--;
+	return (RET_SUCCESS);
+}	
+	
 /*
  * mpool_get
  *	Get a page.
@@ -154,12 +183,6 @@ mpool_get(MPOOL *mp, pgno_t pgno,
 	off_t off;
 	int nr;
 
-	/* Check for attempt to retrieve a non-existent page. */
-	if (pgno >= mp->npages) {
-		errno = EINVAL;
-		return (NULL);
-	}
-
 #ifdef STATISTICS
 	++mp->pageget;
 #endif
@@ -167,7 +190,7 @@ mpool_get(MPOOL *mp, pgno_t pgno,
 	/* Check for a page that is cached. */
 	if ((bp = mpool_look(mp, pgno)) != NULL) {
 #ifdef DEBUG
-		if (bp->flags & MPOOL_PINNED) {
+		if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) {
 			(void)fprintf(stderr,
 			    "mpool_get: page %d already pinned\n", bp->pgno);
 			abort();
@@ -193,20 +216,38 @@ mpool_get(MPOOL *mp, pgno_t pgno,
 		return (NULL);
 
 	/* Read in the contents. */
+	off = mp->pagesize * pgno;
+	if ((nr = pread(mp->fd, bp->page, mp->pagesize, off)) != mp->pagesize) {
+		switch (nr) {
+		case -1:
+			/* errno is set for us by pread(). */
+			free(bp);
+			mp->curcache--;
+			return (NULL);
+		case 0:
+			/*
+			 * A zero-length read means you need to create a
+			 * new page.
+			 */
+			memset(bp->page, 0, mp->pagesize);
+			break;
+		default:
+			/* A partial read is definitely bad. */
+			free(bp);
+			mp->curcache--;
+			errno = EINVAL;
+			return (NULL);
+		}
+	}
 #ifdef STATISTICS
 	++mp->pageread;
 #endif
-	off = mp->pagesize * pgno;
-	nr = pread(mp->fd, bp->page, mp->pagesize, off);
-	if (nr != mp->pagesize) {
-		if (nr >= 0)
-			errno = EFTYPE;
-		return (NULL);
-	}
 
 	/* Set the page number, pin the page. */
 	bp->pgno = pgno;
-	bp->flags = MPOOL_PINNED;
+	if (!(flags & MPOOL_IGNOREPIN))
+		bp->flags = MPOOL_PINNED;
+	bp->flags |= MPOOL_INUSE;
 
 	/*
 	 * Add the page to the head of the hash chain and the tail
@@ -245,7 +286,8 @@ mpool_put(MPOOL *mp, void *page, u_int flags)
 	}
 #endif
 	bp->flags &= ~MPOOL_PINNED;
-	bp->flags |= flags & MPOOL_DIRTY;
+	if (flags & MPOOL_DIRTY)
+		bp->flags |= flags & MPOOL_DIRTY;
 	return (RET_SUCCESS);
 }
 
@@ -329,6 +371,7 @@ mpool_bkt(MPOOL *mp)
 				bp->page = spage;
 			}
 #endif
+			bp->flags = 0;
 			return (bp);
 		}
 
@@ -338,6 +381,7 @@ new:	if ((bp = (BKT *)calloc(1, sizeof(BKT) + mp->pagesize)) == NULL)
 	++mp->pagealloc;
 #endif
 	bp->page = (char *)bp + sizeof(BKT);
+	bp->flags = 0;
 	++mp->curcache;
 	return (bp);
 }
@@ -363,6 +407,15 @@ mpool_write(MPOOL *mp, BKT *bp)
 	if (pwrite(mp->fd, bp->page, mp->pagesize, off) != mp->pagesize)
 		return (RET_ERROR);
 
+	/*
+	 * Re-run through the input filter since this page may soon be
+	 * accessed via the cache, and whatever the user's output filter
+	 * did may screw things up if we don't let the input filter
+	 * restore the in-core copy.
+	 */
+	if (mp->pgin)
+		(mp->pgin)(mp->pgcookie, bp->pgno, bp->page);
+
 	bp->flags &= ~MPOOL_DIRTY;
 	return (RET_SUCCESS);
 }
@@ -379,7 +432,8 @@ mpool_look(MPOOL *mp, pgno_t pgno)
 
 	head = &mp->hqh[HASHKEY(pgno)];
 	TAILQ_FOREACH(bp, head, hq)
-		if (bp->pgno == pgno) {
+		if ((bp->pgno == pgno) &&
+			((bp->flags & MPOOL_INUSE) == MPOOL_INUSE)) {
 #ifdef STATISTICS
 			++mp->cachehit;
 #endif