Update internal BDB to version 4.5.20

13 files changed, 846 insertions, 661 deletions

diff --git a/db/qam/qam.c b/db/qam/qam.c
index 3ae2f90d0..08c5e3417 100644
--- a/db/qam/qam.c
+++ b/db/qam/qam.c
@@ -1,23 +1,16 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam.c,v 11.187 2004/10/18 20:21:13 ubell Exp $
+ * $Id: qam.c,v 12.38 2006/09/09 14:28:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#include <string.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_page.h"
-#include "dbinc/db_shash.h"
 #include "dbinc/btree.h"
 #include "dbinc/lock.h"
 #include "dbinc/log.h"
@@ -33,18 +26,22 @@ static int __qam_c_put __P((DBC *, DBT *, DBT *, u_int32_t, db_pgno_t *));
 static int __qam_consume __P((DBC *, QMETA *, db_recno_t));
 static int __qam_getno __P((DB *, const DBT *, db_recno_t *));
 
+#define	DONT_NEED_LOCKS(dbc) ((dbc)->txn == NULL ||			\
+	F_ISSET(dbc, DBC_READ_COMMITTED | DBC_READ_UNCOMMITTED))
+
 /*
  * __qam_position --
  *	Position a queued access method cursor at a record.  This returns
  *	the page locked.  *exactp will be set if the record is valid.
  * PUBLIC: int __qam_position
- * PUBLIC:       __P((DBC *, db_recno_t *, qam_position_mode, int *));
+ * PUBLIC:      __P((DBC *, db_recno_t *, db_lockmode_t, u_int32_t, int *));
  */
 int
-__qam_position(dbc, recnop, mode, exactp)
+__qam_position(dbc, recnop, lock_mode, get_mode, exactp)
 	DBC *dbc;		/* open cursor */
 	db_recno_t *recnop;	/* pointer to recno to find */
-	qam_position_mode mode;/* locking: read or write */
+	db_lockmode_t lock_mode;/* locking: read or write */
+	u_int32_t get_mode;	/* flags to __memp_fget */
 	int *exactp;		/* indicate if it was found */
 {
 	QUEUE_CURSOR *cp;
@@ -59,14 +56,13 @@ __qam_position(dbc, recnop, mode, exactp)
 	/* Fetch the page for this recno. */
 	pg = QAM_RECNO_PAGE(dbp, *recnop);
 
-	if ((ret = __db_lget(dbc, 0, pg, mode == QAM_READ ?
-	    DB_LOCK_READ : DB_LOCK_WRITE, 0, &cp->lock)) != 0)
+	if ((ret = __db_lget(dbc, 0, pg, lock_mode, 0, &cp->lock)) != 0)
 		return (ret);
 	cp->page = NULL;
 	*exactp = 0;
 	if ((ret = __qam_fget(dbp, &pg,
-	    mode == QAM_WRITE ? DB_MPOOL_CREATE : 0, &cp->page)) != 0) {
-		if (mode != QAM_WRITE &&
+	    dbc->txn, get_mode, &cp->page)) != 0) {
+		if (!FLD_ISSET(get_mode, DB_MPOOL_CREATE) &&
 		    (ret == DB_PAGE_NOTFOUND || ret == ENOENT))
 			ret = 0;
 
@@ -79,10 +75,16 @@ __qam_position(dbc, recnop, mode, exactp)
 	cp->indx = QAM_RECNO_INDEX(dbp, pg, *recnop);
 
 	if (PGNO(cp->page) == 0) {
-		if (F_ISSET(dbp, DB_AM_RDONLY)) {
+		/*
+		 * We have read an uninitialized page: set the page number if
+		 * we're creating the page.  Otherwise, we know that the record
+		 * doesn't exist yet.
+		 */
+		if (!FLD_ISSET(get_mode, DB_MPOOL_CREATE)) {
 			*exactp = 0;
 			return (0);
 		}
+		DB_ASSERT(dbp->dbenv, FLD_ISSET(get_mode, DB_MPOOL_CREATE));
 		PGNO(cp->page) = pg;
 		TYPE(cp->page) = P_QAMDATA;
 	}
@@ -102,7 +104,7 @@ __qam_position(dbc, recnop, mode, exactp)
  *   pagep must be write locked
  *
  * PUBLIC: int __qam_pitem
- * PUBLIC:     __P((DBC *,  QPAGE *, u_int32_t, db_recno_t, DBT *));
+ * PUBLIC:    __P((DBC *,  QPAGE *, u_int32_t, db_recno_t, DBT *));
  */
 int
 __qam_pitem(dbc, pagep, indx, recno, data)
@@ -133,7 +135,7 @@ __qam_pitem(dbc, pagep, indx, recno, data)
 	datap = data;
 	if (F_ISSET(data, DB_DBT_PARTIAL)) {
 		if (data->doff + data->dlen > t->re_len) {
-			__db_err(dbenv,
+			__db_errx(dbenv,
 		"%s: data offset plus length larger than record size of %lu",
 			    "Record length error", (u_long)t->re_len);
 			return (EINVAL);
@@ -175,7 +177,7 @@ __qam_pitem(dbc, pagep, indx, recno, data)
 			if (F_ISSET(qp, QAM_VALID))
 				memcpy(dest, p, t->re_len);
 			else
-				memset(dest, t->re_pad, t->re_len);
+				memset(dest, (int)t->re_pad, t->re_len);
 
 			dest += data->doff;
 			memcpy(dest, data->data, data->size);
@@ -202,7 +204,8 @@ no_partial:
 	F_SET(qp, QAM_VALID | QAM_SET);
 	memcpy(p, datap->data, datap->size);
 	if (!F_ISSET(data, DB_DBT_PARTIAL))
-		memset(p + datap->size,  t->re_pad, t->re_len - datap->size);
+		memset(p + datap->size,
+		    (int)t->re_pad, t->re_len - datap->size);
 
 err:	if (allocated)
 		__os_free(dbenv, datap->data);
@@ -222,6 +225,7 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 	db_pgno_t *pgnop;
 {
 	DB *dbp;
+	DB_ENV *dbenv;
 	DB_LOCK lock;
 	DB_MPOOLFILE *mpf;
 	QMETA *meta;
@@ -229,9 +233,10 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 	db_pgno_t pg;
 	db_recno_t new_cur, new_first;
 	u_int32_t opcode;
-	int exact, ret, t_ret;
+	int exact, ret, t_ret, writelock;
 
 	dbp = dbc->dbp;
+	dbenv = dbp->dbenv;
 	mpf = dbp->mpf;
 	if (pgnop != NULL)
 		*pgnop = PGNO_INVALID;
@@ -241,6 +246,7 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 	switch (flags) {
 	case DB_KEYFIRST:
 	case DB_KEYLAST:
+	case DB_NOOVERWRITE:
 		if ((ret = __qam_getno(dbp, key, &cp->recno)) != 0)
 			return (ret);
 		/* FALLTHROUGH */
@@ -248,29 +254,34 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 		break;
 	default:
 		/* The interface shouldn't let anything else through. */
-		DB_ASSERT(0);
-		return (__db_ferr(dbp->dbenv, "DBC->put", 0));
+		return (__db_ferr(dbenv, "DBC->put", 0));
 	}
 
 	/* Write lock the record. */
-	if ((ret = __db_lget(dbc,
-	    0, cp->recno, DB_LOCK_WRITE, DB_LOCK_RECORD, &lock)) != 0)
+	if ((ret = __db_lget(dbc, LCK_COUPLE,
+	    cp->recno, DB_LOCK_WRITE, DB_LOCK_RECORD, &cp->lock)) != 0)
 		return (ret);
 
-	if ((ret = __qam_position(dbc, &cp->recno, QAM_WRITE, &exact)) != 0) {
+	lock = cp->lock;
+
+	if ((ret = __qam_position(dbc, &cp->recno, DB_LOCK_WRITE,
+	    DB_MPOOL_CREATE | DB_MPOOL_DIRTY, &exact)) != 0) {
 		/* We could not get the page, we can release the record lock. */
 		(void)__LPUT(dbc, lock);
 		return (ret);
 	}
 
-	/* Put the item on the page. */
-	ret = __qam_pitem(dbc, (QPAGE *)cp->page, cp->indx, cp->recno, data);
+	if (exact != 0 && flags == DB_NOOVERWRITE)
+		ret = DB_KEYEXIST;
+	else
+		/* Put the item on the page. */
+		ret = __qam_pitem(dbc,
+		     (QPAGE *)cp->page, cp->indx, cp->recno, data);
 
 	/* Doing record locking, release the page lock */
 	if ((t_ret = __LPUT(dbc, cp->lock)) != 0 && ret == 0)
 		ret = t_ret;
-	if ((t_ret = __qam_fput(
-	    dbp, cp->pgno, cp->page, DB_MPOOL_DIRTY)) != 0 && ret == 0)
+	if ((t_ret = __qam_fput(dbp, cp->pgno, cp->page, 0)) != 0 && ret == 0)
 		ret = t_ret;
 	cp->page = NULL;
 	cp->lock = lock;
@@ -282,12 +293,14 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 	pg = ((QUEUE *)dbp->q_internal)->q_meta;
 
 	/*
-	 * Get the meta page first, we don't want to write lock it while
-	 * trying to pin it.
+	 * Get the meta page first, we don't want to lock it while trying
+	 * to pin it.
 	 */
-	if ((ret = __memp_fget(mpf, &pg, 0, &meta)) != 0)
+	writelock = 0;
+	if ((ret = __memp_fget(mpf, &pg, dbc->txn, 0, &meta)) != 0)
 		return (ret);
-	if ((ret = __db_lget(dbc, 0, pg,  DB_LOCK_WRITE, 0, &lock)) != 0) {
+	if ((ret = __db_lget(dbc, LCK_COUPLE,
+	    pg,  DB_LOCK_READ, 0, &cp->lock)) != 0) {
 		(void)__memp_fput(mpf, meta, 0);
 		return (ret);
 	}
@@ -303,6 +316,7 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 	 * insert is.
 	 */
 
+recheck:
 	if (meta->first_recno == meta->cur_recno) {
 		new_first = cp->recno;
 		new_cur = cp->recno + 1;
@@ -311,19 +325,12 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 		opcode |= QAM_SETFIRST;
 		opcode |= QAM_SETCUR;
 	} else {
-		if (QAM_BEFORE_FIRST(meta, cp->recno) &&
-		    (meta->first_recno <= meta->cur_recno ||
-		    meta->first_recno - cp->recno <
-		    cp->recno - meta->cur_recno)) {
+		if (QAM_BEFORE_FIRST(meta, cp->recno)) {
 			new_first = cp->recno;
 			opcode |= QAM_SETFIRST;
 		}
 
-		if (meta->cur_recno == cp->recno ||
-		    (QAM_AFTER_CURRENT(meta, cp->recno) &&
-		    (meta->first_recno <= meta->cur_recno ||
-		    cp->recno - meta->cur_recno <=
-		    meta->first_recno - cp->recno))) {
+		if (QAM_AFTER_CURRENT(meta, cp->recno)) {
 			new_cur = cp->recno + 1;
 			if (new_cur == RECNO_OOB)
 				new_cur++;
@@ -331,25 +338,36 @@ __qam_c_put(dbc, key, data, flags, pgnop)
 		}
 	}
 
-	if (opcode != 0 && DBC_LOGGING(dbc)) {
-		ret = __qam_mvptr_log(dbp, dbc->txn, &meta->dbmeta.lsn,
-		    0, opcode, meta->first_recno, new_first,
-		    meta->cur_recno, new_cur, &meta->dbmeta.lsn, PGNO_BASE_MD);
-		if (ret != 0)
-			opcode = 0;
+	if (opcode == 0)
+		goto done;
+
+	/* Drop the read lock and get the a write lock on the meta page. */
+	if (writelock == 0 && (ret = __db_lget(dbc, LCK_COUPLE_ALWAYS,
+	     pg,  DB_LOCK_WRITE, 0, &cp->lock)) != 0) {
+		(void)__memp_fput(mpf, meta, 0);
+		return (ret);
 	}
+	if (writelock++ == 0)
+		goto recheck;
+
+	if (((ret = __memp_dirty(mpf, &meta, dbc->txn, DB_MPOOL_DIRTY)) != 0 ||
+	    (DBC_LOGGING(dbc) &&
+	    (ret = __qam_mvptr_log(dbp, dbc->txn,
+	    &meta->dbmeta.lsn, 0, opcode, meta->first_recno,
+	    new_first, meta->cur_recno, new_cur,
+	    &meta->dbmeta.lsn, PGNO_BASE_MD)) != 0)))
+		opcode = 0;
 
 	if (opcode & QAM_SETCUR)
 		meta->cur_recno = new_cur;
 	if (opcode & QAM_SETFIRST)
 		meta->first_recno = new_first;
 
-	if ((t_ret = __memp_fput(
-	    mpf, meta, opcode != 0 ? DB_MPOOL_DIRTY : 0)) != 0 && ret == 0)
+done:	if ((t_ret = __memp_fput(mpf, meta, 0)) != 0 && ret == 0)
 		ret = t_ret;
 
 	/* Don't hold the meta page long term. */
-	if ((t_ret = __LPUT(dbc, lock)) != 0 && ret == 0)
+	if ((t_ret = __LPUT(dbc, cp->lock)) != 0 && ret == 0)
 		ret = t_ret;
 	return (ret);
 }
@@ -385,7 +403,7 @@ __qam_append(dbc, key, data)
 	 * Get the meta page first, we don't want to write lock it while
 	 * trying to pin it.
 	 */
-	if ((ret = __memp_fget(mpf, &pg, 0, &meta)) != 0)
+	if ((ret = __memp_fget(mpf, &pg, dbc->txn, DB_MPOOL_DIRTY, &meta)) != 0)
 		return (ret);
 	/* Write lock the meta page. */
 	if ((ret = __db_lget(dbc, 0, pg,  DB_LOCK_WRITE, 0, &lock)) != 0) {
@@ -443,7 +461,8 @@ __qam_append(dbc, key, data)
 	/* Fetch and write lock the data page. */
 	if ((ret = __db_lget(dbc, 0, pg,  DB_LOCK_WRITE, 0, &lock)) != 0)
 		goto err;
-	if ((ret = __qam_fget(dbp, &pg, DB_MPOOL_CREATE, &page)) != 0) {
+	if ((ret = __qam_fget(dbp, &pg, dbc->txn,
+	    DB_MPOOL_CREATE | DB_MPOOL_DIRTY, &page)) != 0) {
 		/* We did not fetch it, we can release the lock. */
 		(void)__LPUT(dbc, lock);
 		goto err;
@@ -464,7 +483,7 @@ __qam_append(dbc, key, data)
 		ret = t_ret;
 
 	if ((t_ret =
-	    __qam_fput(dbp, pg, page, DB_MPOOL_DIRTY)) != 0 && ret == 0)
+	    __qam_fput(dbp, pg, page, 0)) != 0 && ret == 0)
 		ret = t_ret;
 
 	/* Return the record number to the user. */
@@ -491,7 +510,7 @@ __qam_append(dbc, key, data)
 	}
 
 err:	/* Release the meta page. */
-	if ((t_ret = __memp_fput(mpf, meta, DB_MPOOL_DIRTY)) != 0 && ret == 0)
+	if ((t_ret = __memp_fput(mpf, meta, 0)) != 0 && ret == 0)
 		ret = t_ret;
 
 	return (ret);
@@ -514,19 +533,19 @@ __qam_c_del(dbc)
 	QMETA *meta;
 	QUEUE_CURSOR *cp;
 	db_pgno_t pg;
-	db_recno_t first;
 	int exact, ret, t_ret;
 
 	dbp = dbc->dbp;
 	mpf = dbp->mpf;
 	cp = (QUEUE_CURSOR *)dbc->internal;
+	LOCK_INIT(lock);
 
 	pg = ((QUEUE *)dbp->q_internal)->q_meta;
 	/*
 	 * Get the meta page first, we don't want to write lock it while
 	 * trying to pin it.
 	 */
-	if ((ret = __memp_fget(mpf, &pg, 0, &meta)) != 0)
+	if ((ret = __memp_fget(mpf, &pg, dbc->txn, 0, &meta)) != 0)
 		return (ret);
 	/* Write lock the meta page. */
 	if ((ret = __db_lget(dbc, 0, pg,  DB_LOCK_READ, 0, &metalock)) != 0) {
@@ -537,8 +556,6 @@ __qam_c_del(dbc)
 	if (QAM_NOT_VALID(meta, cp->recno))
 		ret = DB_NOTFOUND;
 
-	first = meta->first_recno;
-
 	/* Don't hold the meta page long term. */
 	if ((t_ret = __LPUT(dbc, metalock)) != 0 && ret == 0)
 		ret = t_ret;
@@ -546,13 +563,15 @@ __qam_c_del(dbc)
 	if (ret != 0)
 		goto err;
 
-	if ((ret = __db_lget(dbc,
-	    0, cp->recno, DB_LOCK_WRITE, DB_LOCK_RECORD, &lock)) != 0)
+	if ((ret = __db_lget(dbc, LCK_COUPLE,
+	    cp->recno, DB_LOCK_WRITE, DB_LOCK_RECORD, &cp->lock)) != 0)
 		goto err;
 	cp->lock_mode = DB_LOCK_WRITE;
+	lock = cp->lock;
 
 	/* Find the record ; delete only deletes exact matches. */
-	if ((ret = __qam_position(dbc, &cp->recno, QAM_WRITE, &exact)) != 0)
+	if ((ret = __qam_position(dbc, &cp->recno, DB_LOCK_WRITE,
+	    DB_MPOOL_CREATE | DB_MPOOL_DIRTY, &exact)) != 0)
 		goto err;
 
 	if (!exact) {
@@ -582,20 +601,30 @@ __qam_c_del(dbc)
 
 	F_CLR(qp, QAM_VALID);
 
-	if (cp->recno == first) {
+	/*
+	 * Peek at the first_recno before locking the meta page.
+	 * Other threads cannot move first_recno past
+	 * our position while we have the record locked.
+	 * If it's pointing at the deleted record then lock
+	 * the metapage and check again as lower numbered
+recheck:
+	 * record may have been inserted.
+	 */
+	if (cp->recno == meta->first_recno) {
 		pg = ((QUEUE *)dbp->q_internal)->q_meta;
 		if ((ret =
 		    __db_lget(dbc, 0, pg,  DB_LOCK_WRITE, 0, &metalock)) != 0)
 			goto err;
-		ret = __qam_consume(dbc, meta, first);
+		if (cp->recno == meta->first_recno)
+			ret = __qam_consume(dbc, meta, meta->first_recno);
 		if ((t_ret = __LPUT(dbc, metalock)) != 0 && ret == 0)
 			ret = t_ret;
 	}
 
 err:	if ((t_ret = __memp_fput(mpf, meta, 0)) != 0 && ret == 0)
 		ret = t_ret;
-	if (cp->page != NULL && (t_ret = __qam_fput(dbp, cp->pgno,
-	    cp->page, ret == 0 ? DB_MPOOL_DIRTY : 0)) != 0 && ret == 0)
+	if (cp->page != NULL &&
+	    (t_ret = __qam_fput(dbp, cp->pgno, cp->page, 0)) != 0 && ret == 0)
 		ret = t_ret;
 	cp->page = NULL;
 
@@ -633,37 +662,32 @@ __qam_c_get(dbc, key, data, flags, pgnop)
 	QMETA *meta;
 	QUEUE *t;
 	QUEUE_CURSOR *cp;
-	db_lockmode_t lock_mode;
+	db_lockmode_t lock_mode, meta_mode;
 	db_pgno_t metapno;
 	db_recno_t first;
-	qam_position_mode mode;
 	int exact, inorder, is_first, locked, ret, t_ret, wait, with_delete;
-	int put_mode, retrying;
+	int retrying;
 
 	dbp = dbc->dbp;
 	dbenv = dbp->dbenv;
 	mpf = dbp->mpf;
 	cp = (QUEUE_CURSOR *)dbc->internal;
+	LOCK_INIT(lock);
+	LOCK_INIT(pglock);
 
 	PANIC_CHECK(dbenv);
 
 	wait = 0;
 	with_delete = 0;
 	retrying = 0;
-	lock_mode = DB_LOCK_READ;
+	lock_mode = F_ISSET(dbc, DBC_RMW) ? DB_LOCK_WRITE : DB_LOCK_READ;
+	meta_mode = DB_LOCK_READ;
 	meta = NULL;
 	inorder = F_ISSET(dbp, DB_AM_INORDER);
-	put_mode = 0;
 	t_ret = 0;
 	*pgnop = 0;
 	pg = NULL;
 
-	mode = QAM_READ;
-	if (F_ISSET(dbc, DBC_RMW)) {
-		lock_mode = DB_LOCK_WRITE;
-		mode = QAM_WRITE;
-	}
-
 	if (flags == DB_CONSUME_WAIT) {
 		wait = 1;
 		flags = DB_CONSUME;
@@ -671,8 +695,7 @@ __qam_c_get(dbc, key, data, flags, pgnop)
 	if (flags == DB_CONSUME) {
 		with_delete = 1;
 		flags = DB_FIRST;
-		lock_mode = DB_LOCK_WRITE;
-		mode = QAM_CONSUME;
+		meta_mode = lock_mode = DB_LOCK_WRITE;
 	}
 
 	DEBUG_LREAD(dbc, dbc->txn, "qam_c_get",
@@ -682,22 +705,36 @@ __qam_c_get(dbc, key, data, flags, pgnop)
 	locked = 0;
 
 	is_first = 0;
+	first = 0;
 
 	t = (QUEUE *)dbp->q_internal;
 	metapno = t->q_meta;
+	LOCK_INIT(metalock);
 
 	/*
 	 * Get the meta page first, we don't want to write lock it while
 	 * trying to pin it.  This is because someone my have it pinned
 	 * but not locked.
 	 */
-	if ((ret = __memp_fget(mpf, &metapno, 0, &meta)) != 0)
+	if ((ret = __memp_fget(mpf, &metapno, dbc->txn, 0, &meta)) != 0)
 		return (ret);
-	if ((ret = __db_lget(dbc, 0, metapno, lock_mode, 0, &metalock)) != 0)
-		goto err;
-	locked = 1;
 
-	first = 0;
+get_next:
+	switch (flags) {
+	case DB_NEXT:
+	case DB_NEXT_NODUP:
+	case DB_FIRST:
+	case DB_PREV:
+	case DB_PREV_NODUP:
+	case DB_LAST:
+		if ((ret = __db_lget(dbc,
+		    0, metapno, meta_mode, 0, &metalock)) != 0)
+			goto err;
+		locked = 1;
+		break;
+	default:
+		break;
+	}
 
 	/* Release any previous lock if not in a transaction. */
 	if ((ret = __TLPUT(dbc, cp->lock)) != 0)
@@ -713,7 +750,7 @@ retry:	/* Update the record number. */
 		/* NOTREACHED */
 	case DB_NEXT:
 	case DB_NEXT_NODUP:
-get_next:	if (cp->recno != RECNO_OOB) {
+		if (cp->recno != RECNO_OOB) {
 			++cp->recno;
 			/* Wrap around, skipping zero. */
 			if (cp->recno == RECNO_OOB)
@@ -721,8 +758,7 @@ get_next:	if (cp->recno != RECNO_OOB) {
 			/*
 			 * Check to see if we are out of data.
 			 */
-			if (cp->recno == meta->cur_recno ||
-			    QAM_AFTER_CURRENT(meta, cp->recno)) {
+			if (QAM_AFTER_CURRENT(meta, cp->recno)) {
 				pg = NULL;
 				if (!wait) {
 					ret = DB_NOTFOUND;
@@ -742,18 +778,18 @@ get_next:	if (cp->recno != RECNO_OOB) {
 
 				if (CDB_LOCKING(dbenv)) {
 					/* Drop the metapage before we wait. */
-					if ((ret =
-					    __memp_fput(mpf, meta, 0)) != 0)
-						goto err;
+					ret = __memp_fput(mpf, meta, 0);
 					meta = NULL;
+					if (ret != 0)
+						goto err;
 					if ((ret = __lock_get(
 					    dbenv, dbc->locker,
 					    DB_LOCK_SWITCH, &dbc->lock_dbt,
 					    DB_LOCK_WAIT, &dbc->mylock)) != 0)
 						goto err;
 
-					if ((ret = __memp_fget(mpf,
-					     &metapno, 0, &meta)) != 0)
+					if ((ret = __memp_fget(mpf, &metapno,
+					    dbc->txn, 0, &meta)) != 0)
 						goto err;
 					if ((ret = __lock_get(
 					    dbenv, dbc->locker,
@@ -770,31 +806,31 @@ get_next:	if (cp->recno != RECNO_OOB) {
 				 */
 				if (locked == 0) {
 					if ((ret = __db_lget(dbc, 0, metapno,
-					     lock_mode, 0, &metalock)) != 0)
+					    meta_mode, 0, &metalock)) != 0)
 						goto err;
 					locked = 1;
-					if (cp->recno != meta->cur_recno &&
-					    cp->recno != RECNO_OOB &&
+					if (cp->recno != RECNO_OOB &&
 					    !QAM_AFTER_CURRENT(meta, cp->recno))
 						goto retry;
 				}
 				/* Drop the metapage before we wait. */
-				if ((ret = __memp_fput(mpf, meta, 0)) != 0)
-					goto err;
+				ret = __memp_fput(mpf, meta, 0);
 				meta = NULL;
+				if (ret != 0)
+					goto err;
 				if ((ret = __db_lget(dbc,
-				     0, metapno, DB_LOCK_WAIT,
-				     DB_LOCK_SWITCH, &metalock)) != 0) {
+				    0, metapno, DB_LOCK_WAIT,
+				    DB_LOCK_SWITCH, &metalock)) != 0) {
 					if (ret == DB_LOCK_DEADLOCK)
 						ret = DB_LOCK_NOTGRANTED;
 					goto err;
 				}
-				if ((ret = __memp_fget(
-				     mpf, &metapno, 0, &meta)) != 0)
+				if ((ret = __memp_fget(mpf, &metapno, dbc->txn,
+				    0, &meta)) != 0)
 					goto err;
 				if ((ret = __db_lget(dbc, 0,
-				     PGNO_INVALID, DB_LOCK_WRITE,
-				     DB_LOCK_UPGRADE, &metalock)) != 0) {
+				    PGNO_INVALID, DB_LOCK_WRITE,
+				    DB_LOCK_UPGRADE, &metalock)) != 0) {
 					if (ret == DB_LOCK_DEADLOCK)
 						ret = DB_LOCK_NOTGRANTED;
 					goto err;
@@ -843,10 +879,6 @@ get_next:	if (cp->recno != RECNO_OOB) {
 	case DB_GET_BOTH_RANGE:
 		if ((ret = __qam_getno(dbp, key, &cp->recno)) != 0)
 			goto err;
-		if (QAM_NOT_VALID(meta, cp->recno)) {
-			ret = DB_NOTFOUND;
-			goto err;
-		}
 		break;
 	default:
 		ret = __db_unknown_flag(dbenv, "__qam_c_get", flags);
@@ -861,18 +893,20 @@ get_next:	if (cp->recno != RECNO_OOB) {
 	}
 
 	/* Lock the record. */
-	if ((ret = __db_lget(dbc, 0, cp->recno, lock_mode,
+	if (((ret = __db_lget(dbc, LCK_COUPLE, cp->recno, lock_mode,
 	    (with_delete && !retrying) ?
 	    DB_LOCK_NOWAIT | DB_LOCK_RECORD : DB_LOCK_RECORD,
-	    &lock)) == DB_LOCK_DEADLOCK && with_delete) {
+	    &lock)) == DB_LOCK_DEADLOCK || ret == DB_LOCK_NOTGRANTED) &&
+	    with_delete) {
 #ifdef QDEBUG
-		__db_logmsg(dbenv,
-		    dbc->txn, "Queue S", 0, "%x %d %d %d",
-		    dbc->locker, cp->recno, first, meta->first_recno);
+		if (DBC_LOGGING(dbc))
+			(void)__log_printf(dbenv,
+			    dbc->txn, "Queue S: %x %d %d %d",
+			    dbc->locker, cp->recno, first, meta->first_recno);
 #endif
 		first = 0;
 		if ((ret =
-		    __db_lget(dbc, 0, metapno, lock_mode, 0, &metalock)) != 0)
+		    __db_lget(dbc, 0, metapno, meta_mode, 0, &metalock)) != 0)
 			goto err;
 		locked = 1;
 		goto retry;
@@ -886,27 +920,55 @@ get_next:	if (cp->recno != RECNO_OOB) {
 	 * since the first/last may have moved while we slept.
 	 * We release our locks and try again.
 	 */
-	if (((inorder || !with_delete) && is_first) || flags == DB_LAST) {
-get_first:
+	switch (flags) {
+	default:
+		if ((!inorder && with_delete) || !is_first)
+			break;
+		/* FALLTHROUGH */
+	case DB_SET:
+	case DB_SET_RANGE:
+	case DB_GET_BOTH:
+	case DB_GET_BOTH_RANGE:
+	case DB_LAST:
 		if ((ret =
-		    __db_lget(dbc, 0, metapno, lock_mode, 0, &metalock)) != 0)
-			goto err;
-		if (cp->recno !=
-		    (is_first ? meta->first_recno : (meta->cur_recno - 1))) {
+		    __db_lget(dbc, 0, metapno, meta_mode, 0, &metalock)) != 0)
+			goto lerr;
+		locked = 1;
+		if ((is_first && cp->recno != meta->first_recno) ||
+		    (flags == DB_LAST && cp->recno != meta->cur_recno - 1)) {
 			if ((ret = __LPUT(dbc, lock)) != 0)
 				goto err;
 			if (is_first)
 				flags = DB_FIRST;
-			locked = 1;
 			goto retry;
+		} else if (!is_first && flags != DB_LAST) {
+			if (QAM_BEFORE_FIRST(meta, cp->recno)) {
+				if (flags == DB_SET_RANGE ||
+				    flags == DB_GET_BOTH_RANGE) {
+					if ((ret = __LPUT(dbc, metalock)) != 0)
+						goto err;
+					locked = 0;
+					cp->lock = lock;
+					LOCK_INIT(lock);
+					goto release_retry;
+				}
+				ret = DB_NOTFOUND;
+				goto lerr;
+			}
+			if (QAM_AFTER_CURRENT(meta, cp->recno)) {
+				ret = DB_NOTFOUND;
+				goto lerr;
+			}
 		}
 		/* Don't hold the meta page long term. */
 		if ((ret = __LPUT(dbc, metalock)) != 0)
 			goto err;
+		locked = 0;
 	}
 
 	/* Position the cursor on the record. */
-	if ((ret = __qam_position(dbc, &cp->recno, mode, &exact)) != 0) {
+	if ((ret = __qam_position(dbc, &cp->recno,
+	    lock_mode, 0, &exact)) != 0) {
 		/* We cannot get the page, release the record lock. */
 		(void)__LPUT(dbc, lock);
 		goto err;
@@ -916,6 +978,7 @@ get_first:
 	pglock = cp->lock;
 	cp->lock = lock;
 	cp->lock_mode = lock_mode;
+	LOCK_INIT(lock);
 
 	if (!exact) {
 release_retry:	/* Release locks and retry, if possible. */
@@ -924,38 +987,53 @@ release_retry:	/* Release locks and retry, if possible. */
 		cp->page = pg = NULL;
 		if ((ret = __LPUT(dbc, pglock)) != 0)
 			goto err1;
+		if (with_delete) {
+			if ((ret = __LPUT(dbc, cp->lock)) != 0)
+				goto err1;
+		} else if ((ret = __TLPUT(dbc, cp->lock)) != 0)
+			goto err1;
 
+		/*
+		 * If we don't need locks and we are out of range
+		 * then we can just skip to the FIRST/LAST record
+		 * otherwise we must iterate to lock the records
+		 * and get serializability.
+		 */
 		switch (flags) {
-		case DB_GET_BOTH_RANGE:
-			flags = DB_SET_RANGE;
-			/* FALLTHROUGH */
 		case DB_NEXT:
 		case DB_NEXT_NODUP:
-		case DB_SET_RANGE:
 			if (!with_delete)
 				is_first = 0;
-			/* Peek at the meta page unlocked. */
-			if (QAM_BEFORE_FIRST(meta, cp->recno))
-				goto get_first;
-			/* FALLTHROUGH */
+			if (QAM_BEFORE_FIRST(meta, cp->recno) &&
+			    DONT_NEED_LOCKS(dbc))
+				flags = DB_FIRST;
+			break;
+		case DB_LAST:
 		case DB_PREV:
 		case DB_PREV_NODUP:
-		case DB_LAST:
-			if (flags == DB_LAST)
+			if (QAM_AFTER_CURRENT(meta, cp->recno) &&
+			    DONT_NEED_LOCKS(dbc))
+				flags = DB_LAST;
+			else
 				flags = DB_PREV;
-			retrying = 0;
-			if ((ret = __LPUT(dbc, cp->lock)) != 0)
-				goto err1;
-			if (flags == DB_SET_RANGE)
-				goto get_next;
+			break;
+
+		case DB_GET_BOTH_RANGE:
+		case DB_SET_RANGE:
+			if (QAM_BEFORE_FIRST(meta, cp->recno) &&
+			    DONT_NEED_LOCKS(dbc))
+				flags = DB_FIRST;
 			else
-				goto retry;
+				flags = DB_NEXT;
+			break;
 
 		default:
 			/* this is for the SET and GET_BOTH cases */
 			ret = DB_KEYEMPTY;
 			goto err1;
 		}
+		retrying = 0;
+		goto get_next;
 	}
 
 	qp = QAM_GET_RECORD(dbp, pg, cp->indx);
@@ -976,18 +1054,18 @@ release_retry:	/* Release locks and retry, if possible. */
 	}
 
 	/* Return the key if the user didn't give us one. */
-	if (key != NULL) {
-		if (flags != DB_GET_BOTH && flags != DB_SET &&
-		    (ret = __db_retcopy(dbp->dbenv,
+	if (key != NULL && !F_ISSET(key, DB_DBT_ISSET)) {
+		if ((ret = __db_retcopy(dbp->dbenv,
 		    key, &cp->recno, sizeof(cp->recno),
 		    &dbc->rkey->data, &dbc->rkey->ulen)) != 0)
 			goto err1;
 		F_SET(key, DB_DBT_ISSET);
 	}
 
-	if (data != NULL) {
-		if (!F_ISSET(dbc, DBC_MULTIPLE|DBC_MULTIPLE_KEY) &&
-		    (ret = __db_retcopy(dbp->dbenv, data, qp->data, t->re_len,
+	if (data != NULL &&
+	    !F_ISSET(dbc, DBC_MULTIPLE|DBC_MULTIPLE_KEY) &&
+	    !F_ISSET(data, DB_DBT_ISSET)) {
+		if ((ret = __db_retcopy(dbp->dbenv, data, qp->data, t->re_len,
 		    &dbc->rdata->data, &dbc->rdata->ulen)) != 0)
 			goto err1;
 		F_SET(data, DB_DBT_ISSET);
@@ -1001,7 +1079,12 @@ release_retry:	/* Release locks and retry, if possible. */
 		 * DB_APPEND there;  attempting one should be forbidden by
 		 * the interface.
 		 */
-		DB_ASSERT(!F_ISSET(dbp, DB_AM_SECONDARY));
+		DB_ASSERT(dbenv, !F_ISSET(dbp, DB_AM_SECONDARY));
+
+		if ((ret = __qam_dirty(dbp, cp->pgno, &cp->page,
+		    dbc->txn, 0)) != 0)
+			goto err1;
+		pg = cp->page;
 
 		/*
 		 * Check and see if we *have* any secondary indices.
@@ -1048,7 +1131,6 @@ release_retry:	/* Release locks and retry, if possible. */
 		}
 
 		F_CLR(qp, QAM_VALID);
-		put_mode = DB_MPOOL_DIRTY;
 
 		if ((ret = __LPUT(dbc, pglock)) != 0)
 			goto err1;
@@ -1066,14 +1148,15 @@ release_retry:	/* Release locks and retry, if possible. */
 		 */
 
 		if (locked == 0 && (ret = __db_lget(
-		    dbc, 0, metapno, lock_mode, 0, &metalock)) != 0)
+		    dbc, 0, metapno, meta_mode, 0, &metalock)) != 0)
 			goto err1;
 		locked = 1;
 
 #ifdef QDEBUG
-		__db_logmsg(dbenv,
-		    dbc->txn, "Queue D", 0, "%x %d %d %d",
-		    dbc->locker, cp->recno, first, meta->first_recno);
+		if (DBC_LOGGING(dbc))
+			(void)__log_printf(dbenv,
+			    dbc->txn, "Queue D: %x %d %d %d",
+			    dbc->locker, cp->recno, first, meta->first_recno);
 #endif
 		/*
 		 * See if we deleted the "first" record.  If
@@ -1092,8 +1175,8 @@ release_retry:	/* Release locks and retry, if possible. */
 
 done:
 err1:	if (cp->page != NULL) {
-		if ((t_ret = __qam_fput(
-		    dbp, cp->pgno, cp->page, put_mode)) != 0 && ret == 0)
+		if ((t_ret = __qam_fput(dbp, cp->pgno, cp->page, 0)) != 0 &&
+		    ret == 0)
 			ret = t_ret;
 
 		/* Doing record locking, release the page lock */
@@ -1101,6 +1184,9 @@ err1:	if (cp->page != NULL) {
 			ret = t_ret;
 		cp->page = NULL;
 	}
+	if (0) {
+lerr:		(void)__LPUT(dbc, lock);
+	}
 
 err:	if (meta) {
 		/* Release the meta page. */
@@ -1112,11 +1198,11 @@ err:	if (meta) {
 			if ((t_ret = __LPUT(dbc, metalock)) != 0 && ret == 0)
 				ret = t_ret;
 	}
-	DB_ASSERT(!LOCK_ISSET(metalock));
+	DB_ASSERT(dbenv, !LOCK_ISSET(metalock));
 
 	return ((ret == DB_LOCK_NOTGRANTED &&
-	     !F_ISSET(dbenv, DB_ENV_TIME_NOTGRANTED)) ?
-	     DB_LOCK_DEADLOCK : ret);
+	    !F_ISSET(dbenv, DB_ENV_TIME_NOTGRANTED)) ?
+	    DB_LOCK_DEADLOCK : ret);
 }
 
 /*
@@ -1142,7 +1228,7 @@ __qam_consume(dbc, meta, first)
 	dbp = dbc->dbp;
 	mpf = dbp->mpf;
 	cp = (QUEUE_CURSOR *)dbc->internal;
-	put_mode = DB_MPOOL_DIRTY;
+	put_mode = 0;
 	ret = 0;
 
 	save_page = cp->pgno;
@@ -1165,12 +1251,11 @@ __qam_consume(dbc, meta, first)
 		if (ret != 0)
 			goto done;
 		if ((ret =
-		    __qam_fput(dbp, cp->pgno, cp->page, put_mode)) != 0)
+		    __qam_fput(dbp, cp->pgno, cp->page, 0)) != 0)
 			goto done;
 		cp->page = NULL;
-		put_mode = 0;
 		if ((ret = __qam_position(dbc,
-		    &first, QAM_READ, &exact)) != 0 || exact != 0) {
+		    &first, DB_LOCK_READ, 0, &exact)) != 0 || exact != 0) {
 			(void)__LPUT(dbc, lock);
 			goto done;
 		}
@@ -1198,20 +1283,20 @@ __qam_consume(dbc, meta, first)
 		 */
 		if (cp->page != NULL && rec_extent != 0 &&
 		    ((exact = (first % rec_extent == 0)) ||
-		    first % meta->rec_page == 0 ||
+		    (first % meta->rec_page == 0) ||
 		    first == UINT32_MAX)) {
 			if (exact == 1 && (ret = __db_lget(dbc,
 			    0, cp->pgno, DB_LOCK_WRITE, 0, &cp->lock)) != 0)
 				break;
-
 #ifdef QDEBUG
-			__db_logmsg(dbp->dbenv,
-			    dbc->txn, "Queue R", 0, "%x %d %d %d",
-			    dbc->locker, cp->pgno, first, meta->first_recno);
+			if (DBC_LOGGING(dbc))
+				(void)__log_printf(dbp->dbenv, dbc->txn,
+				    "Queue R: %x %d %d %d", dbc->locker,
+				    cp->pgno, first, meta->first_recno);
 #endif
 			put_mode |= DB_MPOOL_DISCARD;
-			if ((ret = __qam_fput(dbp,
-			    cp->pgno, cp->page, put_mode)) != 0)
+			if ((ret =
+			    __qam_fput(dbp, cp->pgno, cp->page, put_mode)) != 0)
 				break;
 			cp->page = NULL;
 
@@ -1250,7 +1335,7 @@ __qam_consume(dbc, meta, first)
 			break;
 
 		if ((ret = __qam_position(dbc,
-		    &first, QAM_READ, &exact)) != 0) {
+		    &first, DB_LOCK_READ, 0, &exact)) != 0) {
 			(void)__LPUT(dbc, lock);
 			break;
 		}
@@ -1275,10 +1360,13 @@ __qam_consume(dbc, meta, first)
 	 * Advance first_recno to this point.
 	 */
 	if (ret == 0 && meta->first_recno != first) {
+		if ((ret = __memp_dirty(mpf, &meta, dbc->txn, 0)) != 0)
+			goto done;
 #ifdef QDEBUG
-		__db_logmsg(dbp->dbenv, dbc->txn, "Queue M",
-		    0, "%x %d %d %d", dbc->locker, cp->recno,
-		    first, meta->first_recno);
+		if (DBC_LOGGING(dbc))
+			(void)__log_printf(dbp->dbenv, dbc->txn,
+			    "Queue M: %x %d %d %d", dbc->locker, cp->recno,
+			    first, meta->first_recno);
 #endif
 		if (DBC_LOGGING(dbc))
 			if ((ret = __qam_incfirst_log(dbp,
@@ -1286,7 +1374,6 @@ __qam_consume(dbc, meta, first)
 			    cp->recno, PGNO_BASE_MD)) != 0)
 				goto done;
 		meta->first_recno = first;
-		(void)__memp_fset(mpf, meta, DB_MPOOL_DIRTY);
 	}
 
 done:
@@ -1309,22 +1396,17 @@ __qam_bulk(dbc, data, flags)
 	db_indx_t indx;
 	db_lockmode_t lkmode;
 	db_pgno_t metapno;
-	qam_position_mode mode;
-	int32_t  *endp, *offp;
+	u_int32_t  *endp, *offp;
+	u_int32_t pagesize, re_len, recs;
 	u_int8_t *dbuf, *dp, *np;
-	int exact, recs, re_len, ret, t_ret, valid;
-	int is_key, need_pg, pagesize, size, space;
+	int exact, ret, t_ret, valid;
+	int is_key, need_pg, size, space;
 
 	dbp = dbc->dbp;
 	mpf = dbp->mpf;
 	cp = (QUEUE_CURSOR *)dbc->internal;
 
-	mode = QAM_READ;
-	lkmode = DB_LOCK_READ;
-	if (F_ISSET(dbc, DBC_RMW)) {
-		mode = QAM_WRITE;
-		lkmode = DB_LOCK_WRITE;
-	}
+	lkmode = F_ISSET(dbc, DBC_RMW) ? DB_LOCK_WRITE : DB_LOCK_READ;
 
 	pagesize = dbp->pgsize;
 	re_len = ((QUEUE *)dbp->q_internal)->re_len;
@@ -1336,7 +1418,7 @@ __qam_bulk(dbc, data, flags)
 
 	if ((ret = __db_lget(dbc, 0, metapno, DB_LOCK_READ, 0, &metalock)) != 0)
 		return (ret);
-	if ((ret = __memp_fget(mpf, &metapno, 0, &meta)) != 0) {
+	if ((ret = __memp_fget(mpf, &metapno, dbc->txn, 0, &meta)) != 0) {
 		/* We did not fetch it, we can release the lock. */
 		(void)__LPUT(dbc, metalock);
 		return (ret);
@@ -1346,11 +1428,11 @@ __qam_bulk(dbc, data, flags)
 	np = dp = dbuf;
 
 	/* Keep track of space that is left.  There is an termination entry */
-	space = data->ulen;
-	space -= sizeof(*offp);
+	space = (int)data->ulen;
+	space -= (int)sizeof(*offp);
 
-	/* Build the offset/size table form the end up. */
-	endp = (int32_t *) ((u_int8_t *)dbuf + data->ulen);
+	/* Build the offset/size table from the end up. */
+	endp = (u_int32_t *)((u_int8_t *)dbuf + data->ulen);
 	endp--;
 	offp = endp;
 	/* Save the lock on the current position of the cursor. */
@@ -1361,7 +1443,7 @@ next_pg:
 	/* Wrap around, skipping zero. */
 	if (cp->recno == RECNO_OOB)
 		cp->recno++;
-	if ((ret = __qam_position(dbc, &cp->recno, mode, &exact)) != 0)
+	if ((ret = __qam_position(dbc, &cp->recno, lkmode, 0, &exact)) != 0)
 		goto done;
 
 	pg = cp->page;
@@ -1379,23 +1461,24 @@ next_pg:
 
 		if (pg != NULL) {
 			if ((ret = __db_lget(dbc, LCK_COUPLE,
-			     cp->recno, lkmode, DB_LOCK_RECORD, &rlock)) != 0)
+			    cp->recno, lkmode, DB_LOCK_RECORD, &rlock)) != 0)
 				goto done;
 			qp = QAM_GET_RECORD(dbp, pg, indx);
 			if (F_ISSET(qp, QAM_VALID)) {
 				valid = 1;
-				space -= (is_key ? 3 : 2) * sizeof(*offp);
+				space -= (int)
+				    ((is_key ? 3 : 2) * sizeof(*offp));
 				if (space < 0)
 					goto get_space;
 				if (need_pg) {
 					dp = np;
-					size = pagesize - QPAGE_SZ(dbp);
+					size = (int)pagesize - QPAGE_SZ(dbp);
 					if (space < size) {
 get_space:
 						if (offp == endp) {
 							data->size = (u_int32_t)
-							    DB_ALIGN(size +
-							    pagesize,
+							    DB_ALIGN((u_int32_t)
+							    size + pagesize,
 							    sizeof(u_int32_t));
 							ret = DB_BUFFER_SMALL;
 							break;
@@ -1407,16 +1490,17 @@ get_space:
 						break;
 					}
 					memcpy(dp,
-					    (char *)pg + QPAGE_SZ(dbp), size);
+					    (u_int8_t *)pg + QPAGE_SZ(dbp),
+					    (u_int)size);
 					need_pg = 0;
 					space -= size;
 					np += size;
 				}
 				if (is_key)
 					*offp-- = cp->recno;
-				*offp-- = (int32_t)((u_int8_t*)qp -
-				    (u_int8_t*)pg - QPAGE_SZ(dbp) +
-				    dp - dbuf + SSZA(QAMDATA, data));
+				*offp-- = (u_int32_t)((((u_int8_t *)qp -
+				    (u_int8_t *)pg) - QPAGE_SZ(dbp)) +
+				    (dp - dbuf) + SSZA(QAMDATA, data));
 				*offp-- = re_len;
 			}
 		}
@@ -1426,7 +1510,6 @@ get_space:
 		}
 		cp->recno++;
 	} while (++indx < recs && cp->recno != RECNO_OOB &&
-	    cp->recno != meta->cur_recno &&
 	    !QAM_AFTER_CURRENT(meta, cp->recno));
 
 	/* Drop the page lock. */
@@ -1442,7 +1525,6 @@ get_space:
 
 	if (ret == 0 && space > 0 &&
 	    (indx >= recs || cp->recno == RECNO_OOB) &&
-	    cp->recno != meta->cur_recno &&
 	    !QAM_AFTER_CURRENT(meta, cp->recno))
 		goto next_pg;
 
@@ -1457,7 +1539,7 @@ get_space:
 	if (is_key == 1)
 		*offp = RECNO_OOB;
 	else
-		*offp = -1;
+		*offp = (u_int32_t)-1;
 
 done:	/* Release the meta page. */
 	if ((t_ret = __memp_fput(mpf, meta, 0)) != 0 && ret == 0)
@@ -1514,19 +1596,12 @@ __qam_c_dup(orig_dbc, new_dbc)
 	DBC *orig_dbc, *new_dbc;
 {
 	QUEUE_CURSOR *orig, *new;
-	int ret;
 
 	orig = (QUEUE_CURSOR *)orig_dbc->internal;
 	new = (QUEUE_CURSOR *)new_dbc->internal;
 
 	new->recno = orig->recno;
 
-	/* Acquire the long term lock if we are not in a transaction. */
-	if (orig_dbc->txn == NULL && LOCK_ISSET(orig->lock))
-		if ((ret = __db_lget(new_dbc, 0, new->recno,
-		    new->lock_mode, DB_LOCK_RECORD, &new->lock)) != 0)
-			return (ret);
-
 	return (0);
 }
 
@@ -1555,7 +1630,7 @@ __qam_c_init(dbc)
 	}
 
 	/* Initialize methods. */
-	dbc->c_close = __db_c_close;
+	dbc->c_close = __db_c_close_pp;
 	dbc->c_count = __db_c_count_pp;
 	dbc->c_del = __db_c_del_pp;
 	dbc->c_dup = __db_c_dup_pp;
@@ -1598,7 +1673,7 @@ __qam_getno(dbp, key, rep)
 	db_recno_t *rep;
 {
 	if ((*rep = *(db_recno_t *)key->data) == 0) {
-		__db_err(dbp->dbenv, "illegal record number of 0");
+		__db_errx(dbp->dbenv, "illegal record number of 0");
 		return (EINVAL);
 	}
 	return (0);
@@ -1639,7 +1714,8 @@ __qam_truncate(dbc, countp)
 		return (ret);
 
 	mpf = dbp->mpf;
-	if ((ret = __memp_fget(mpf, &metapno, 0, &meta)) != 0) {
+	if ((ret = __memp_fget(mpf, &metapno, dbc->txn,
+	    DB_MPOOL_DIRTY, &meta)) != 0) {
 		/* We did not fetch it, we can release the lock. */
 		(void)__LPUT(dbc, metalock);
 		return (ret);
@@ -1647,7 +1723,7 @@ __qam_truncate(dbc, countp)
 	/* Remove the last extent file. */
 	if (meta->cur_recno > 1 && ((QUEUE *)dbp->q_internal)->page_ext != 0) {
 		if ((ret = __qam_fremove(dbp,
-		     QAM_RECNO_PAGE(dbp, meta->cur_recno - 1))) != 0)
+		    QAM_RECNO_PAGE(dbp, meta->cur_recno - 1))) != 0)
 			return (ret);
 	}
 
@@ -1659,13 +1735,36 @@ __qam_truncate(dbc, countp)
 	if (ret == 0)
 		meta->first_recno = meta->cur_recno = 1;
 
-	if ((t_ret = __memp_fput(mpf,
-	    meta, ret == 0 ? DB_MPOOL_DIRTY: 0)) != 0 && ret == 0)
+	if ((t_ret = __memp_fput(mpf, meta, 0)) != 0 && ret == 0)
 		ret = t_ret;
 	if ((t_ret = __LPUT(dbc, metalock)) != 0 && ret == 0)
 		ret = t_ret;
 
-	*countp = count;
+	if (countp != NULL)
+		*countp = count;
 
 	return (ret);
 }
+
+/*
+ * __qam_delete --
+ *	Queue fast delete function.
+ *
+ * PUBLIC: int __qam_delete __P((DBC *,  DBT *));
+ */
+int
+__qam_delete(dbc, key)
+	DBC *dbc;
+	DBT *key;
+{
+	QUEUE_CURSOR *cp;
+	int ret;
+
+	cp = (QUEUE_CURSOR *)dbc->internal;
+	if ((ret = __qam_getno(dbc->dbp, key, &cp->recno)) != 0)
+		goto err;
+
+	ret = __qam_c_del(dbc);
+
+err:	return (ret);
+}
diff --git a/db/qam/qam.src b/db/qam/qam.src
index 71063f0b6..4e795e553 100644
--- a/db/qam/qam.src
+++ b/db/qam/qam.src
@@ -1,22 +1,15 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam.src,v 11.33 2004/06/17 17:35:22 bostic Exp $
+ * $Id: qam.src,v 12.5 2006/08/24 14:46:23 bostic Exp $
  */
 
 PREFIX	__qam
 DBPRIVATE
 
-INCLUDE #ifndef NO_SYSTEM_INCLUDES
-INCLUDE #include <sys/types.h>
-INCLUDE
-INCLUDE #include <ctype.h>
-INCLUDE #include <string.h>
-INCLUDE #endif
-INCLUDE
 INCLUDE #include "db_int.h"
 INCLUDE #include "dbinc/crypto.h"
 INCLUDE #include "dbinc/db_page.h"
@@ -31,7 +24,7 @@ INCLUDE
  * incfirst
  * Used when we increment first_recno.
  */
-BEGIN incfirst		84
+BEGIN incfirst		42	84
 DB	fileid		int32_t		ld
 ARG	recno		db_recno_t	lu
 ARG	meta_pgno	db_pgno_t	lu
@@ -41,7 +34,7 @@ END
  * mvptr
  * Used when we change one or both of cur_recno and first_recno.
  */
-BEGIN mvptr		85
+BEGIN mvptr		42	85
 ARG	opcode		u_int32_t	lu
 DB	fileid		int32_t		ld
 ARG	old_first	db_recno_t	lu
@@ -58,7 +51,7 @@ END
  * Used when we delete a record.
  * recno is the record that is being deleted.
  */
-BEGIN del		79
+BEGIN del		42	79
 DB	fileid		int32_t		ld
 POINTER	lsn		DB_LSN *	lu
 ARG	pgno		db_pgno_t	lu
@@ -72,7 +65,7 @@ END
  * recno is the record being added.
  * data is the record itself.
  */
-BEGIN add		80
+BEGIN add		42	80
 DB	fileid		int32_t		ld
 POINTER	lsn		DB_LSN *	lu
 ARG	pgno		db_pgno_t	lu
@@ -88,7 +81,7 @@ END
  * Used when we delete a record in extent based queue.
  * recno is the record that is being deleted.
  */
-BEGIN delext		83
+BEGIN delext		42	83
 DB	fileid		int32_t		ld
 POINTER	lsn		DB_LSN *	lu
 ARG	pgno		db_pgno_t	lu
diff --git a/db/qam/qam_auto.c b/db/qam/qam_auto.c
index fe3dcf6fd..acca78332 100644
--- a/db/qam/qam_auto.c
+++ b/db/qam/qam_auto.c
@@ -2,13 +2,6 @@
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#include <ctype.h>
-#include <string.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/crypto.h"
 #include "dbinc/db_page.h"
@@ -23,9 +16,9 @@
  * PUBLIC:     u_int32_t, db_recno_t, db_pgno_t));
  */
 int
-__qam_incfirst_log(dbp, txnid, ret_lsnp, flags, recno, meta_pgno)
+__qam_incfirst_log(dbp, txnp, ret_lsnp, flags, recno, meta_pgno)
 	DB *dbp;
-	DB_TXN *txnid;
+	DB_TXN *txnp;
 	DB_LSN *ret_lsnp;
 	u_int32_t flags;
 	db_recno_t recno;
@@ -51,30 +44,31 @@ __qam_incfirst_log(dbp, txnid, ret_lsnp, flags, recno, meta_pgno)
 
 	if (LF_ISSET(DB_LOG_NOT_DURABLE) ||
 	    F_ISSET(dbp, DB_AM_NOT_DURABLE)) {
+		if (txnp == NULL)
+			return (0);
 		is_durable = 0;
 	} else
 		is_durable = 1;
 
-	if (txnid == NULL) {
+	if (txnp == NULL) {
 		txn_num = 0;
 		lsnp = &null_lsn;
 		null_lsn.file = null_lsn.offset = 0;
 	} else {
-		if (TAILQ_FIRST(&txnid->kids) != NULL &&
-		    (ret = __txn_activekids(dbenv, rectype, txnid)) != 0)
+		if (TAILQ_FIRST(&txnp->kids) != NULL &&
+		    (ret = __txn_activekids(dbenv, rectype, txnp)) != 0)
 			return (ret);
 		/*
 		 * We need to assign begin_lsn while holding region mutex.
 		 * That assignment is done inside the DbEnv->log_put call,
 		 * so pass in the appropriate memory location to be filled
 		 * in by the log_put code.
-		*/
-		DB_SET_BEGIN_LSNP(txnid, &rlsnp);
-		txn_num = txnid->txnid;
-		lsnp = &txnid->last_lsn;
+		 */
+		DB_SET_TXN_LSNP(txnp, &rlsnp, &lsnp);
+		txn_num = txnp->txnid;
 	}
 
-	DB_ASSERT(dbp->log_filename != NULL);
+	DB_ASSERT(dbenv, dbp->log_filename != NULL);
 	if (dbp->log_filename->id == DB_LOGFILEID_INVALID &&
 	    (ret = __dbreg_lazy_id(dbp)) != 0)
 		return (ret);
@@ -89,7 +83,7 @@ __qam_incfirst_log(dbp, txnid, ret_lsnp, flags, recno, meta_pgno)
 		logrec.size += npad;
 	}
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret =
 		    __os_malloc(dbenv, logrec.size, &logrec.data)) != 0)
 			return (ret);
@@ -133,12 +127,13 @@ __qam_incfirst_log(dbp, txnid, ret_lsnp, flags, recno, meta_pgno)
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	DB_ASSERT((u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
+	DB_ASSERT(dbenv,
+	    (u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret = __log_put(dbenv, rlsnp,(DBT *)&logrec,
-		    flags | DB_LOG_NOCOPY)) == 0 && txnid != NULL) {
-			txnid->last_lsn = *rlsnp;
+		    flags | DB_LOG_NOCOPY)) == 0 && txnp != NULL) {
+			*lsnp = *rlsnp;
 			if (rlsnp != ret_lsnp)
 				 *ret_lsnp = *rlsnp;
 		}
@@ -157,20 +152,21 @@ __qam_incfirst_log(dbp, txnid, ret_lsnp, flags, recno, meta_pgno)
 #else
 		ret = 0;
 #endif
-		STAILQ_INSERT_HEAD(&txnid->logs, lr, links);
+		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
+		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
 		LSN_NOT_LOGGED(*ret_lsnp);
 	}
 
 #ifdef LOG_DIAGNOSTIC
 	if (ret != 0)
 		(void)__qam_incfirst_print(dbenv,
-		    (DBT *)&logrec, ret_lsnp, NULL, NULL);
+		    (DBT *)&logrec, ret_lsnp, DB_TXN_PRINT, NULL);
 #endif
 
 #ifdef DIAGNOSTIC
 	__os_free(dbenv, logrec.data);
 #else
-	if (is_durable || txnid == NULL)
+	if (is_durable || txnp == NULL)
 		__os_free(dbenv, logrec.data);
 #endif
 	return (ret);
@@ -195,13 +191,14 @@ __qam_incfirst_read(dbenv, recbuf, argpp)
 	    sizeof(__qam_incfirst_args) + sizeof(DB_TXN), &argp)) != 0)
 		return (ret);
 	bp = recbuf;
-	argp->txnid = (DB_TXN *)&argp[1];
+	argp->txnp = (DB_TXN *)&argp[1];
+	memset(argp->txnp, 0, sizeof(DB_TXN));
 
 	memcpy(&argp->type, bp, sizeof(argp->type));
 	bp += sizeof(argp->type);
 
-	memcpy(&argp->txnid->txnid,  bp, sizeof(argp->txnid->txnid));
-	bp += sizeof(argp->txnid->txnid);
+	memcpy(&argp->txnp->txnid, bp, sizeof(argp->txnp->txnid));
+	bp += sizeof(argp->txnp->txnid);
 
 	memcpy(&argp->prev_lsn, bp, sizeof(DB_LSN));
 	bp += sizeof(DB_LSN);
@@ -228,11 +225,11 @@ __qam_incfirst_read(dbenv, recbuf, argpp)
  * PUBLIC:     db_recno_t, DB_LSN *, db_pgno_t));
  */
 int
-__qam_mvptr_log(dbp, txnid, ret_lsnp, flags,
+__qam_mvptr_log(dbp, txnp, ret_lsnp, flags,
     opcode, old_first, new_first, old_cur, new_cur,
     metalsn, meta_pgno)
 	DB *dbp;
-	DB_TXN *txnid;
+	DB_TXN *txnp;
 	DB_LSN *ret_lsnp;
 	u_int32_t flags;
 	u_int32_t opcode;
@@ -263,30 +260,31 @@ __qam_mvptr_log(dbp, txnid, ret_lsnp, flags,
 
 	if (LF_ISSET(DB_LOG_NOT_DURABLE) ||
 	    F_ISSET(dbp, DB_AM_NOT_DURABLE)) {
+		if (txnp == NULL)
+			return (0);
 		is_durable = 0;
 	} else
 		is_durable = 1;
 
-	if (txnid == NULL) {
+	if (txnp == NULL) {
 		txn_num = 0;
 		lsnp = &null_lsn;
 		null_lsn.file = null_lsn.offset = 0;
 	} else {
-		if (TAILQ_FIRST(&txnid->kids) != NULL &&
-		    (ret = __txn_activekids(dbenv, rectype, txnid)) != 0)
+		if (TAILQ_FIRST(&txnp->kids) != NULL &&
+		    (ret = __txn_activekids(dbenv, rectype, txnp)) != 0)
 			return (ret);
 		/*
 		 * We need to assign begin_lsn while holding region mutex.
 		 * That assignment is done inside the DbEnv->log_put call,
 		 * so pass in the appropriate memory location to be filled
 		 * in by the log_put code.
-		*/
-		DB_SET_BEGIN_LSNP(txnid, &rlsnp);
-		txn_num = txnid->txnid;
-		lsnp = &txnid->last_lsn;
+		 */
+		DB_SET_TXN_LSNP(txnp, &rlsnp, &lsnp);
+		txn_num = txnp->txnid;
 	}
 
-	DB_ASSERT(dbp->log_filename != NULL);
+	DB_ASSERT(dbenv, dbp->log_filename != NULL);
 	if (dbp->log_filename->id == DB_LOGFILEID_INVALID &&
 	    (ret = __dbreg_lazy_id(dbp)) != 0)
 		return (ret);
@@ -306,7 +304,7 @@ __qam_mvptr_log(dbp, txnid, ret_lsnp, flags,
 		logrec.size += npad;
 	}
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret =
 		    __os_malloc(dbenv, logrec.size, &logrec.data)) != 0)
 			return (ret);
@@ -362,9 +360,15 @@ __qam_mvptr_log(dbp, txnid, ret_lsnp, flags,
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	if (metalsn != NULL)
+	if (metalsn != NULL) {
+		if (txnp != NULL) {
+			LOG *lp = dbenv->lg_handle->reginfo.primary;
+			if (LOG_COMPARE(metalsn, &lp->lsn) >= 0 && (ret =
+			    __log_check_page_lsn(dbenv, dbp, metalsn) != 0))
+				return (ret);
+		}
 		memcpy(bp, metalsn, sizeof(*metalsn));
-	else
+	} else
 		memset(bp, 0, sizeof(*metalsn));
 	bp += sizeof(*metalsn);
 
@@ -372,12 +376,13 @@ __qam_mvptr_log(dbp, txnid, ret_lsnp, flags,
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	DB_ASSERT((u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
+	DB_ASSERT(dbenv,
+	    (u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret = __log_put(dbenv, rlsnp,(DBT *)&logrec,
-		    flags | DB_LOG_NOCOPY)) == 0 && txnid != NULL) {
-			txnid->last_lsn = *rlsnp;
+		    flags | DB_LOG_NOCOPY)) == 0 && txnp != NULL) {
+			*lsnp = *rlsnp;
 			if (rlsnp != ret_lsnp)
 				 *ret_lsnp = *rlsnp;
 		}
@@ -396,20 +401,21 @@ __qam_mvptr_log(dbp, txnid, ret_lsnp, flags,
 #else
 		ret = 0;
 #endif
-		STAILQ_INSERT_HEAD(&txnid->logs, lr, links);
+		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
+		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
 		LSN_NOT_LOGGED(*ret_lsnp);
 	}
 
 #ifdef LOG_DIAGNOSTIC
 	if (ret != 0)
 		(void)__qam_mvptr_print(dbenv,
-		    (DBT *)&logrec, ret_lsnp, NULL, NULL);
+		    (DBT *)&logrec, ret_lsnp, DB_TXN_PRINT, NULL);
 #endif
 
 #ifdef DIAGNOSTIC
 	__os_free(dbenv, logrec.data);
 #else
-	if (is_durable || txnid == NULL)
+	if (is_durable || txnp == NULL)
 		__os_free(dbenv, logrec.data);
 #endif
 	return (ret);
@@ -433,13 +439,14 @@ __qam_mvptr_read(dbenv, recbuf, argpp)
 	    sizeof(__qam_mvptr_args) + sizeof(DB_TXN), &argp)) != 0)
 		return (ret);
 	bp = recbuf;
-	argp->txnid = (DB_TXN *)&argp[1];
+	argp->txnp = (DB_TXN *)&argp[1];
+	memset(argp->txnp, 0, sizeof(DB_TXN));
 
 	memcpy(&argp->type, bp, sizeof(argp->type));
 	bp += sizeof(argp->type);
 
-	memcpy(&argp->txnid->txnid,  bp, sizeof(argp->txnid->txnid));
-	bp += sizeof(argp->txnid->txnid);
+	memcpy(&argp->txnp->txnid, bp, sizeof(argp->txnp->txnid));
+	bp += sizeof(argp->txnp->txnid);
 
 	memcpy(&argp->prev_lsn, bp, sizeof(DB_LSN));
 	bp += sizeof(DB_LSN);
@@ -484,9 +491,9 @@ __qam_mvptr_read(dbenv, recbuf, argpp)
  * PUBLIC:     u_int32_t, DB_LSN *, db_pgno_t, u_int32_t, db_recno_t));
  */
 int
-__qam_del_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno)
+__qam_del_log(dbp, txnp, ret_lsnp, flags, lsn, pgno, indx, recno)
 	DB *dbp;
-	DB_TXN *txnid;
+	DB_TXN *txnp;
 	DB_LSN *ret_lsnp;
 	u_int32_t flags;
 	DB_LSN * lsn;
@@ -514,30 +521,31 @@ __qam_del_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno)
 
 	if (LF_ISSET(DB_LOG_NOT_DURABLE) ||
 	    F_ISSET(dbp, DB_AM_NOT_DURABLE)) {
+		if (txnp == NULL)
+			return (0);
 		is_durable = 0;
 	} else
 		is_durable = 1;
 
-	if (txnid == NULL) {
+	if (txnp == NULL) {
 		txn_num = 0;
 		lsnp = &null_lsn;
 		null_lsn.file = null_lsn.offset = 0;
 	} else {
-		if (TAILQ_FIRST(&txnid->kids) != NULL &&
-		    (ret = __txn_activekids(dbenv, rectype, txnid)) != 0)
+		if (TAILQ_FIRST(&txnp->kids) != NULL &&
+		    (ret = __txn_activekids(dbenv, rectype, txnp)) != 0)
 			return (ret);
 		/*
 		 * We need to assign begin_lsn while holding region mutex.
 		 * That assignment is done inside the DbEnv->log_put call,
 		 * so pass in the appropriate memory location to be filled
 		 * in by the log_put code.
-		*/
-		DB_SET_BEGIN_LSNP(txnid, &rlsnp);
-		txn_num = txnid->txnid;
-		lsnp = &txnid->last_lsn;
+		 */
+		DB_SET_TXN_LSNP(txnp, &rlsnp, &lsnp);
+		txn_num = txnp->txnid;
 	}
 
-	DB_ASSERT(dbp->log_filename != NULL);
+	DB_ASSERT(dbenv, dbp->log_filename != NULL);
 	if (dbp->log_filename->id == DB_LOGFILEID_INVALID &&
 	    (ret = __dbreg_lazy_id(dbp)) != 0)
 		return (ret);
@@ -554,7 +562,7 @@ __qam_del_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno)
 		logrec.size += npad;
 	}
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret =
 		    __os_malloc(dbenv, logrec.size, &logrec.data)) != 0)
 			return (ret);
@@ -590,9 +598,15 @@ __qam_del_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno)
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	if (lsn != NULL)
+	if (lsn != NULL) {
+		if (txnp != NULL) {
+			LOG *lp = dbenv->lg_handle->reginfo.primary;
+			if (LOG_COMPARE(lsn, &lp->lsn) >= 0 && (ret =
+			    __log_check_page_lsn(dbenv, dbp, lsn) != 0))
+				return (ret);
+		}
 		memcpy(bp, lsn, sizeof(*lsn));
-	else
+	} else
 		memset(bp, 0, sizeof(*lsn));
 	bp += sizeof(*lsn);
 
@@ -608,12 +622,13 @@ __qam_del_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno)
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	DB_ASSERT((u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
+	DB_ASSERT(dbenv,
+	    (u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret = __log_put(dbenv, rlsnp,(DBT *)&logrec,
-		    flags | DB_LOG_NOCOPY)) == 0 && txnid != NULL) {
-			txnid->last_lsn = *rlsnp;
+		    flags | DB_LOG_NOCOPY)) == 0 && txnp != NULL) {
+			*lsnp = *rlsnp;
 			if (rlsnp != ret_lsnp)
 				 *ret_lsnp = *rlsnp;
 		}
@@ -632,20 +647,21 @@ __qam_del_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno)
 #else
 		ret = 0;
 #endif
-		STAILQ_INSERT_HEAD(&txnid->logs, lr, links);
+		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
+		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
 		LSN_NOT_LOGGED(*ret_lsnp);
 	}
 
 #ifdef LOG_DIAGNOSTIC
 	if (ret != 0)
 		(void)__qam_del_print(dbenv,
-		    (DBT *)&logrec, ret_lsnp, NULL, NULL);
+		    (DBT *)&logrec, ret_lsnp, DB_TXN_PRINT, NULL);
 #endif
 
 #ifdef DIAGNOSTIC
 	__os_free(dbenv, logrec.data);
 #else
-	if (is_durable || txnid == NULL)
+	if (is_durable || txnp == NULL)
 		__os_free(dbenv, logrec.data);
 #endif
 	return (ret);
@@ -669,13 +685,14 @@ __qam_del_read(dbenv, recbuf, argpp)
 	    sizeof(__qam_del_args) + sizeof(DB_TXN), &argp)) != 0)
 		return (ret);
 	bp = recbuf;
-	argp->txnid = (DB_TXN *)&argp[1];
+	argp->txnp = (DB_TXN *)&argp[1];
+	memset(argp->txnp, 0, sizeof(DB_TXN));
 
 	memcpy(&argp->type, bp, sizeof(argp->type));
 	bp += sizeof(argp->type);
 
-	memcpy(&argp->txnid->txnid,  bp, sizeof(argp->txnid->txnid));
-	bp += sizeof(argp->txnid->txnid);
+	memcpy(&argp->txnp->txnid, bp, sizeof(argp->txnp->txnid));
+	bp += sizeof(argp->txnp->txnid);
 
 	memcpy(&argp->prev_lsn, bp, sizeof(DB_LSN));
 	bp += sizeof(DB_LSN);
@@ -709,10 +726,10 @@ __qam_del_read(dbenv, recbuf, argpp)
  * PUBLIC:     const DBT *, u_int32_t, const DBT *));
  */
 int
-__qam_add_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data,
+__qam_add_log(dbp, txnp, ret_lsnp, flags, lsn, pgno, indx, recno, data,
     vflag, olddata)
 	DB *dbp;
-	DB_TXN *txnid;
+	DB_TXN *txnp;
 	DB_LSN *ret_lsnp;
 	u_int32_t flags;
 	DB_LSN * lsn;
@@ -743,30 +760,31 @@ __qam_add_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data,
 
 	if (LF_ISSET(DB_LOG_NOT_DURABLE) ||
 	    F_ISSET(dbp, DB_AM_NOT_DURABLE)) {
+		if (txnp == NULL)
+			return (0);
 		is_durable = 0;
 	} else
 		is_durable = 1;
 
-	if (txnid == NULL) {
+	if (txnp == NULL) {
 		txn_num = 0;
 		lsnp = &null_lsn;
 		null_lsn.file = null_lsn.offset = 0;
 	} else {
-		if (TAILQ_FIRST(&txnid->kids) != NULL &&
-		    (ret = __txn_activekids(dbenv, rectype, txnid)) != 0)
+		if (TAILQ_FIRST(&txnp->kids) != NULL &&
+		    (ret = __txn_activekids(dbenv, rectype, txnp)) != 0)
 			return (ret);
 		/*
 		 * We need to assign begin_lsn while holding region mutex.
 		 * That assignment is done inside the DbEnv->log_put call,
 		 * so pass in the appropriate memory location to be filled
 		 * in by the log_put code.
-		*/
-		DB_SET_BEGIN_LSNP(txnid, &rlsnp);
-		txn_num = txnid->txnid;
-		lsnp = &txnid->last_lsn;
+		 */
+		DB_SET_TXN_LSNP(txnp, &rlsnp, &lsnp);
+		txn_num = txnp->txnid;
 	}
 
-	DB_ASSERT(dbp->log_filename != NULL);
+	DB_ASSERT(dbenv, dbp->log_filename != NULL);
 	if (dbp->log_filename->id == DB_LOGFILEID_INVALID &&
 	    (ret = __dbreg_lazy_id(dbp)) != 0)
 		return (ret);
@@ -786,7 +804,7 @@ __qam_add_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data,
 		logrec.size += npad;
 	}
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret =
 		    __os_malloc(dbenv, logrec.size, &logrec.data)) != 0)
 			return (ret);
@@ -822,9 +840,15 @@ __qam_add_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data,
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	if (lsn != NULL)
+	if (lsn != NULL) {
+		if (txnp != NULL) {
+			LOG *lp = dbenv->lg_handle->reginfo.primary;
+			if (LOG_COMPARE(lsn, &lp->lsn) >= 0 && (ret =
+			    __log_check_page_lsn(dbenv, dbp, lsn) != 0))
+				return (ret);
+		}
 		memcpy(bp, lsn, sizeof(*lsn));
-	else
+	} else
 		memset(bp, 0, sizeof(*lsn));
 	bp += sizeof(*lsn);
 
@@ -866,12 +890,13 @@ __qam_add_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data,
 		bp += olddata->size;
 	}
 
-	DB_ASSERT((u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
+	DB_ASSERT(dbenv,
+	    (u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret = __log_put(dbenv, rlsnp,(DBT *)&logrec,
-		    flags | DB_LOG_NOCOPY)) == 0 && txnid != NULL) {
-			txnid->last_lsn = *rlsnp;
+		    flags | DB_LOG_NOCOPY)) == 0 && txnp != NULL) {
+			*lsnp = *rlsnp;
 			if (rlsnp != ret_lsnp)
 				 *ret_lsnp = *rlsnp;
 		}
@@ -890,20 +915,21 @@ __qam_add_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data,
 #else
 		ret = 0;
 #endif
-		STAILQ_INSERT_HEAD(&txnid->logs, lr, links);
+		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
+		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
 		LSN_NOT_LOGGED(*ret_lsnp);
 	}
 
 #ifdef LOG_DIAGNOSTIC
 	if (ret != 0)
 		(void)__qam_add_print(dbenv,
-		    (DBT *)&logrec, ret_lsnp, NULL, NULL);
+		    (DBT *)&logrec, ret_lsnp, DB_TXN_PRINT, NULL);
 #endif
 
 #ifdef DIAGNOSTIC
 	__os_free(dbenv, logrec.data);
 #else
-	if (is_durable || txnid == NULL)
+	if (is_durable || txnp == NULL)
 		__os_free(dbenv, logrec.data);
 #endif
 	return (ret);
@@ -927,13 +953,14 @@ __qam_add_read(dbenv, recbuf, argpp)
 	    sizeof(__qam_add_args) + sizeof(DB_TXN), &argp)) != 0)
 		return (ret);
 	bp = recbuf;
-	argp->txnid = (DB_TXN *)&argp[1];
+	argp->txnp = (DB_TXN *)&argp[1];
+	memset(argp->txnp, 0, sizeof(DB_TXN));
 
 	memcpy(&argp->type, bp, sizeof(argp->type));
 	bp += sizeof(argp->type);
 
-	memcpy(&argp->txnid->txnid,  bp, sizeof(argp->txnid->txnid));
-	bp += sizeof(argp->txnid->txnid);
+	memcpy(&argp->txnp->txnid, bp, sizeof(argp->txnp->txnid));
+	bp += sizeof(argp->txnp->txnid);
 
 	memcpy(&argp->prev_lsn, bp, sizeof(DB_LSN));
 	bp += sizeof(DB_LSN);
@@ -983,9 +1010,9 @@ __qam_add_read(dbenv, recbuf, argpp)
  * PUBLIC:     const DBT *));
  */
 int
-__qam_delext_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data)
+__qam_delext_log(dbp, txnp, ret_lsnp, flags, lsn, pgno, indx, recno, data)
 	DB *dbp;
-	DB_TXN *txnid;
+	DB_TXN *txnp;
 	DB_LSN *ret_lsnp;
 	u_int32_t flags;
 	DB_LSN * lsn;
@@ -1014,30 +1041,31 @@ __qam_delext_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data)
 
 	if (LF_ISSET(DB_LOG_NOT_DURABLE) ||
 	    F_ISSET(dbp, DB_AM_NOT_DURABLE)) {
+		if (txnp == NULL)
+			return (0);
 		is_durable = 0;
 	} else
 		is_durable = 1;
 
-	if (txnid == NULL) {
+	if (txnp == NULL) {
 		txn_num = 0;
 		lsnp = &null_lsn;
 		null_lsn.file = null_lsn.offset = 0;
 	} else {
-		if (TAILQ_FIRST(&txnid->kids) != NULL &&
-		    (ret = __txn_activekids(dbenv, rectype, txnid)) != 0)
+		if (TAILQ_FIRST(&txnp->kids) != NULL &&
+		    (ret = __txn_activekids(dbenv, rectype, txnp)) != 0)
 			return (ret);
 		/*
 		 * We need to assign begin_lsn while holding region mutex.
 		 * That assignment is done inside the DbEnv->log_put call,
 		 * so pass in the appropriate memory location to be filled
 		 * in by the log_put code.
-		*/
-		DB_SET_BEGIN_LSNP(txnid, &rlsnp);
-		txn_num = txnid->txnid;
-		lsnp = &txnid->last_lsn;
+		 */
+		DB_SET_TXN_LSNP(txnp, &rlsnp, &lsnp);
+		txn_num = txnp->txnid;
 	}
 
-	DB_ASSERT(dbp->log_filename != NULL);
+	DB_ASSERT(dbenv, dbp->log_filename != NULL);
 	if (dbp->log_filename->id == DB_LOGFILEID_INVALID &&
 	    (ret = __dbreg_lazy_id(dbp)) != 0)
 		return (ret);
@@ -1055,7 +1083,7 @@ __qam_delext_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data)
 		logrec.size += npad;
 	}
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret =
 		    __os_malloc(dbenv, logrec.size, &logrec.data)) != 0)
 			return (ret);
@@ -1091,9 +1119,15 @@ __qam_delext_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data)
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	if (lsn != NULL)
+	if (lsn != NULL) {
+		if (txnp != NULL) {
+			LOG *lp = dbenv->lg_handle->reginfo.primary;
+			if (LOG_COMPARE(lsn, &lp->lsn) >= 0 && (ret =
+			    __log_check_page_lsn(dbenv, dbp, lsn) != 0))
+				return (ret);
+		}
 		memcpy(bp, lsn, sizeof(*lsn));
-	else
+	} else
 		memset(bp, 0, sizeof(*lsn));
 	bp += sizeof(*lsn);
 
@@ -1120,12 +1154,13 @@ __qam_delext_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data)
 		bp += data->size;
 	}
 
-	DB_ASSERT((u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
+	DB_ASSERT(dbenv,
+	    (u_int32_t)(bp - (u_int8_t *)logrec.data) <= logrec.size);
 
-	if (is_durable || txnid == NULL) {
+	if (is_durable || txnp == NULL) {
 		if ((ret = __log_put(dbenv, rlsnp,(DBT *)&logrec,
-		    flags | DB_LOG_NOCOPY)) == 0 && txnid != NULL) {
-			txnid->last_lsn = *rlsnp;
+		    flags | DB_LOG_NOCOPY)) == 0 && txnp != NULL) {
+			*lsnp = *rlsnp;
 			if (rlsnp != ret_lsnp)
 				 *ret_lsnp = *rlsnp;
 		}
@@ -1144,20 +1179,21 @@ __qam_delext_log(dbp, txnid, ret_lsnp, flags, lsn, pgno, indx, recno, data)
 #else
 		ret = 0;
 #endif
-		STAILQ_INSERT_HEAD(&txnid->logs, lr, links);
+		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
+		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
 		LSN_NOT_LOGGED(*ret_lsnp);
 	}
 
 #ifdef LOG_DIAGNOSTIC
 	if (ret != 0)
 		(void)__qam_delext_print(dbenv,
-		    (DBT *)&logrec, ret_lsnp, NULL, NULL);
+		    (DBT *)&logrec, ret_lsnp, DB_TXN_PRINT, NULL);
 #endif
 
 #ifdef DIAGNOSTIC
 	__os_free(dbenv, logrec.data);
 #else
-	if (is_durable || txnid == NULL)
+	if (is_durable || txnp == NULL)
 		__os_free(dbenv, logrec.data);
 #endif
 	return (ret);
@@ -1181,13 +1217,14 @@ __qam_delext_read(dbenv, recbuf, argpp)
 	    sizeof(__qam_delext_args) + sizeof(DB_TXN), &argp)) != 0)
 		return (ret);
 	bp = recbuf;
-	argp->txnid = (DB_TXN *)&argp[1];
+	argp->txnp = (DB_TXN *)&argp[1];
+	memset(argp->txnp, 0, sizeof(DB_TXN));
 
 	memcpy(&argp->type, bp, sizeof(argp->type));
 	bp += sizeof(argp->type);
 
-	memcpy(&argp->txnid->txnid,  bp, sizeof(argp->txnid->txnid));
-	bp += sizeof(argp->txnid->txnid);
+	memcpy(&argp->txnp->txnid, bp, sizeof(argp->txnp->txnid));
+	bp += sizeof(argp->txnp->txnid);
 
 	memcpy(&argp->prev_lsn, bp, sizeof(DB_LSN));
 	bp += sizeof(DB_LSN);
diff --git a/db/qam/qam_autop.c b/db/qam/qam_autop.c
index e9c122385..8a1f14f28 100644
--- a/db/qam/qam_autop.c
+++ b/db/qam/qam_autop.c
@@ -3,13 +3,6 @@
 #include "db_config.h"
 
 #ifdef HAVE_QUEUE
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#include <ctype.h>
-#include <string.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/crypto.h"
 #include "dbinc/db_page.h"
@@ -34,20 +27,18 @@ __qam_incfirst_print(dbenv, dbtp, lsnp, notused2, notused3)
 	__qam_incfirst_args *argp;
 	int ret;
 
-	notused2 = DB_TXN_ABORT;
+	notused2 = DB_TXN_PRINT;
 	notused3 = NULL;
 
 	if ((ret = __qam_incfirst_read(dbenv, dbtp->data, &argp)) != 0)
 		return (ret);
 	(void)printf(
-	    "[%lu][%lu]__qam_incfirst%s: rec: %lu txnid %lx prevlsn [%lu][%lu]\n",
-	    (u_long)lsnp->file,
-	    (u_long)lsnp->offset,
+    "[%lu][%lu]__qam_incfirst%s: rec: %lu txnp %lx prevlsn [%lu][%lu]\n",
+	    (u_long)lsnp->file, (u_long)lsnp->offset,
 	    (argp->type & DB_debug_FLAG) ? "_debug" : "",
 	    (u_long)argp->type,
-	    (u_long)argp->txnid->txnid,
-	    (u_long)argp->prev_lsn.file,
-	    (u_long)argp->prev_lsn.offset);
+	    (u_long)argp->txnp->txnid,
+	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
 	(void)printf("\tfileid: %ld\n", (long)argp->fileid);
 	(void)printf("\trecno: %lu\n", (u_long)argp->recno);
 	(void)printf("\tmeta_pgno: %lu\n", (u_long)argp->meta_pgno);
@@ -71,20 +62,18 @@ __qam_mvptr_print(dbenv, dbtp, lsnp, notused2, notused3)
 	__qam_mvptr_args *argp;
 	int ret;
 
-	notused2 = DB_TXN_ABORT;
+	notused2 = DB_TXN_PRINT;
 	notused3 = NULL;
 
 	if ((ret = __qam_mvptr_read(dbenv, dbtp->data, &argp)) != 0)
 		return (ret);
 	(void)printf(
-	    "[%lu][%lu]__qam_mvptr%s: rec: %lu txnid %lx prevlsn [%lu][%lu]\n",
-	    (u_long)lsnp->file,
-	    (u_long)lsnp->offset,
+    "[%lu][%lu]__qam_mvptr%s: rec: %lu txnp %lx prevlsn [%lu][%lu]\n",
+	    (u_long)lsnp->file, (u_long)lsnp->offset,
 	    (argp->type & DB_debug_FLAG) ? "_debug" : "",
 	    (u_long)argp->type,
-	    (u_long)argp->txnid->txnid,
-	    (u_long)argp->prev_lsn.file,
-	    (u_long)argp->prev_lsn.offset);
+	    (u_long)argp->txnp->txnid,
+	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
 	(void)printf("\topcode: %lu\n", (u_long)argp->opcode);
 	(void)printf("\tfileid: %ld\n", (long)argp->fileid);
 	(void)printf("\told_first: %lu\n", (u_long)argp->old_first);
@@ -114,20 +103,18 @@ __qam_del_print(dbenv, dbtp, lsnp, notused2, notused3)
 	__qam_del_args *argp;
 	int ret;
 
-	notused2 = DB_TXN_ABORT;
+	notused2 = DB_TXN_PRINT;
 	notused3 = NULL;
 
 	if ((ret = __qam_del_read(dbenv, dbtp->data, &argp)) != 0)
 		return (ret);
 	(void)printf(
-	    "[%lu][%lu]__qam_del%s: rec: %lu txnid %lx prevlsn [%lu][%lu]\n",
-	    (u_long)lsnp->file,
-	    (u_long)lsnp->offset,
+    "[%lu][%lu]__qam_del%s: rec: %lu txnp %lx prevlsn [%lu][%lu]\n",
+	    (u_long)lsnp->file, (u_long)lsnp->offset,
 	    (argp->type & DB_debug_FLAG) ? "_debug" : "",
 	    (u_long)argp->type,
-	    (u_long)argp->txnid->txnid,
-	    (u_long)argp->prev_lsn.file,
-	    (u_long)argp->prev_lsn.offset);
+	    (u_long)argp->txnp->txnid,
+	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
 	(void)printf("\tfileid: %ld\n", (long)argp->fileid);
 	(void)printf("\tlsn: [%lu][%lu]\n",
 	    (u_long)argp->lsn.file, (u_long)argp->lsn.offset);
@@ -156,20 +143,18 @@ __qam_add_print(dbenv, dbtp, lsnp, notused2, notused3)
 	int ch;
 	int ret;
 
-	notused2 = DB_TXN_ABORT;
+	notused2 = DB_TXN_PRINT;
 	notused3 = NULL;
 
 	if ((ret = __qam_add_read(dbenv, dbtp->data, &argp)) != 0)
 		return (ret);
 	(void)printf(
-	    "[%lu][%lu]__qam_add%s: rec: %lu txnid %lx prevlsn [%lu][%lu]\n",
-	    (u_long)lsnp->file,
-	    (u_long)lsnp->offset,
+    "[%lu][%lu]__qam_add%s: rec: %lu txnp %lx prevlsn [%lu][%lu]\n",
+	    (u_long)lsnp->file, (u_long)lsnp->offset,
 	    (argp->type & DB_debug_FLAG) ? "_debug" : "",
 	    (u_long)argp->type,
-	    (u_long)argp->txnid->txnid,
-	    (u_long)argp->prev_lsn.file,
-	    (u_long)argp->prev_lsn.offset);
+	    (u_long)argp->txnp->txnid,
+	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
 	(void)printf("\tfileid: %ld\n", (long)argp->fileid);
 	(void)printf("\tlsn: [%lu][%lu]\n",
 	    (u_long)argp->lsn.file, (u_long)argp->lsn.offset);
@@ -211,20 +196,18 @@ __qam_delext_print(dbenv, dbtp, lsnp, notused2, notused3)
 	int ch;
 	int ret;
 
-	notused2 = DB_TXN_ABORT;
+	notused2 = DB_TXN_PRINT;
 	notused3 = NULL;
 
 	if ((ret = __qam_delext_read(dbenv, dbtp->data, &argp)) != 0)
 		return (ret);
 	(void)printf(
-	    "[%lu][%lu]__qam_delext%s: rec: %lu txnid %lx prevlsn [%lu][%lu]\n",
-	    (u_long)lsnp->file,
-	    (u_long)lsnp->offset,
+    "[%lu][%lu]__qam_delext%s: rec: %lu txnp %lx prevlsn [%lu][%lu]\n",
+	    (u_long)lsnp->file, (u_long)lsnp->offset,
 	    (argp->type & DB_debug_FLAG) ? "_debug" : "",
 	    (u_long)argp->type,
-	    (u_long)argp->txnid->txnid,
-	    (u_long)argp->prev_lsn.file,
-	    (u_long)argp->prev_lsn.offset);
+	    (u_long)argp->txnp->txnid,
+	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
 	(void)printf("\tfileid: %ld\n", (long)argp->fileid);
 	(void)printf("\tlsn: [%lu][%lu]\n",
 	    (u_long)argp->lsn.file, (u_long)argp->lsn.offset);
diff --git a/db/qam/qam_conv.c b/db/qam/qam_conv.c
index c2b7d53a4..85fb47871 100644
--- a/db/qam/qam_conv.c
+++ b/db/qam/qam_conv.c
@@ -1,18 +1,14 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_conv.c,v 11.17 2004/01/28 03:36:19 bostic Exp $
+ * $Id: qam_conv.c,v 12.4 2006/08/24 14:46:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_page.h"
 #include "dbinc/db_swap.h"
diff --git a/db/qam/qam_files.c b/db/qam/qam_files.c
index d3f040603..62ad8ed8a 100644
--- a/db/qam/qam_files.c
+++ b/db/qam/qam_files.c
@@ -1,25 +1,16 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_files.c,v 1.88 2004/10/21 14:54:42 bostic Exp $
+ * $Id: qam_files.c,v 12.17 2006/08/24 14:46:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-#include <stdlib.h>
-
-#include <string.h>
-#include <ctype.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_page.h"
-#include "dbinc/db_shash.h"
 #include "dbinc/db_am.h"
 #include "dbinc/log.h"
 #include "dbinc/fop.h"
@@ -35,13 +26,14 @@
  *
  * Calculate which extent the page is in, open and create if necessary.
  *
- * PUBLIC: int __qam_fprobe
- * PUBLIC:	   __P((DB *, db_pgno_t, void *, qam_probe_mode, u_int32_t));
+ * PUBLIC: int __qam_fprobe __P((DB *, db_pgno_t,
+ * PUBLIC:     DB_TXN *, void *, qam_probe_mode, u_int32_t));
  */
 int
-__qam_fprobe(dbp, pgno, addrp, mode, flags)
+__qam_fprobe(dbp, pgno, txn, addrp, mode, flags)
 	DB *dbp;
 	db_pgno_t pgno;
+	DB_TXN *txn;
 	void *addrp;
 	qam_probe_mode mode;
 	u_int32_t flags;
@@ -51,8 +43,8 @@ __qam_fprobe(dbp, pgno, addrp, mode, flags)
 	MPFARRAY *array;
 	QUEUE *qp;
 	u_int8_t fid[DB_FILE_ID_LEN];
-	u_int32_t extid, maxext, numext, offset, oldext, openflags;
-	char buf[MAXPATHLEN];
+	u_int32_t i, extid, maxext, numext, lflags, offset, oldext, openflags;
+	char buf[DB_MAXPATHLEN];
 	int ftype, less, ret, t_ret;
 
 	dbenv = dbp->dbenv;
@@ -61,9 +53,17 @@ __qam_fprobe(dbp, pgno, addrp, mode, flags)
 
 	if (qp->page_ext == 0) {
 		mpf = dbp->mpf;
-		return (mode == QAM_PROBE_GET ?
-		    __memp_fget(mpf, &pgno, flags, addrp) :
-		    __memp_fput(mpf, addrp, flags));
+		switch (mode) {
+		case QAM_PROBE_GET:
+			return (__memp_fget(mpf, &pgno, txn, flags, addrp));
+		case QAM_PROBE_PUT:
+			return (__memp_fput(mpf, addrp, flags));
+		case QAM_PROBE_DIRTY:
+			return (__memp_dirty(mpf, addrp, txn, flags));
+		case QAM_PROBE_MPF:
+			*(DB_MPOOLFILE **)addrp = mpf;
+			return (0);
+		}
 	}
 
 	mpf = NULL;
@@ -73,7 +73,7 @@ __qam_fprobe(dbp, pgno, addrp, mode, flags)
 	 * The file cannot go away because we must have a record locked
 	 * in that file.
 	 */
-	MUTEX_THREAD_LOCK(dbenv, dbp->mutexp);
+	MUTEX_LOCK(dbenv, dbp->mutex);
 	extid = QAM_PAGE_EXTENT(dbp, pgno);
 
 	/* Array1 will always be in use if array2 is in use. */
@@ -87,6 +87,7 @@ __qam_fprobe(dbp, pgno, addrp, mode, flags)
 		goto alloc;
 	}
 
+retry:
 	if (extid < array->low_extent) {
 		less = 1;
 		offset = array->low_extent - extid;
@@ -129,16 +130,12 @@ __qam_fprobe(dbp, pgno, addrp, mode, flags)
 			    * sizeof(array->mpfarray[0]));
 			offset = 0;
 		} else if (less == 0 && offset == array->n_extent &&
-		    mode != QAM_PROBE_MPF && array->mpfarray[0].pinref == 0) {
+		    (mode == QAM_PROBE_GET || mode == QAM_PROBE_PUT) &&
+		    array->mpfarray[0].pinref == 0) {
 			/*
 			 * If this is at the end of the array and the file at
 			 * the beginning has a zero pin count we can close
 			 * the bottom extent and put this one at the end.
-			 * TODO: If this process is "slow" then it might be
-			 * appending but miss one or more extents.
-			 * We could check to see if all the extents
-			 * are unpinned and close them in the else
-			 * clause below.
 			 */
 			mpf = array->mpfarray[0].mpf;
 			if (mpf != NULL && (ret = __memp_fclose(mpf, 0)) != 0)
@@ -160,18 +157,49 @@ __qam_fprobe(dbp, pgno, addrp, mode, flags)
 			    / (qp->page_ext * qp->rec_page);
 			if (offset >= maxext/2) {
 				array = &qp->array2;
-				DB_ASSERT(array->n_extent == 0);
+				DB_ASSERT(dbenv, array->n_extent == 0);
 				oldext = 0;
 				array->n_extent = 4;
 				array->low_extent = extid;
 				offset = 0;
 				numext = 0;
+			} else if (array->mpfarray[0].pinref == 0) {
+				/*
+				 * Check to see if there are extents marked
+				 * for deletion at the beginning of the cache.
+				 * If so close them so they will go away.
+				 */
+				for (i = 0; i < array->n_extent; i++) {
+					if (array->mpfarray[i].pinref != 0)
+						break;
+					mpf = array->mpfarray[i].mpf;
+					if (mpf == NULL)
+						continue;
+					(void)__memp_get_flags(mpf, &lflags);
+					if (!FLD_ISSET(lflags, DB_MPOOL_UNLINK))
+						break;
+
+					array->mpfarray[i].mpf = NULL;
+					if ((ret = __memp_fclose(mpf, 0)) != 0)
+						goto err;
+				}
+				if (i == 0)
+					goto increase;
+				memmove(&array->mpfarray[0],
+				     &array->mpfarray[i],
+				    (array->n_extent - i) *
+				    sizeof(array->mpfarray[0]));
+				memset(&array->mpfarray[array->n_extent - i],
+				     '\0', i * sizeof(array->mpfarray[0]));
+				array->low_extent += i;
+				array->hi_extent += i;
+				goto retry;
 			} else {
 				/*
 				 * Increase the size to at least include
 				 * the new one and double it.
 				 */
-				array->n_extent += offset;
+increase:			array->n_extent += offset;
 				array->n_extent <<= 2;
 			}
 alloc:			if ((ret = __os_realloc(dbenv,
@@ -219,6 +247,7 @@ alloc:			if ((ret = __os_realloc(dbenv,
 		(void)__memp_set_pgcookie(mpf, &qp->pgcookie);
 		(void)__memp_get_ftype(dbp->mpf, &ftype);
 		(void)__memp_set_ftype(mpf, ftype);
+		(void)__memp_set_clear_len(mpf, dbp->pgsize);
 
 		/* Set up the fileid for this extent. */
 		__qam_exid(dbp, fid, extid);
@@ -255,25 +284,31 @@ alloc:			if ((ret = __os_realloc(dbenv,
 		(void)__memp_set_flags(mpf, DB_MPOOL_UNLINK, 0);
 
 err:
-	MUTEX_THREAD_UNLOCK(dbenv, dbp->mutexp);
+	MUTEX_UNLOCK(dbenv, dbp->mutex);
 
 	if (ret == 0) {
-		if (mode == QAM_PROBE_MPF) {
-			*(DB_MPOOLFILE **)addrp = mpf;
-			return (0);
-		}
 		pgno--;
 		pgno %= qp->page_ext;
-		if (mode == QAM_PROBE_GET) {
-			if ((ret = __memp_fget(mpf, &pgno, flags, addrp)) == 0)
-				return (ret);
-		} else
+		switch (mode) {
+		case QAM_PROBE_GET:
+			ret = __memp_fget(mpf, &pgno, txn, flags, addrp);
+			if (ret == 0)
+				return (0);
+			break;
+		case QAM_PROBE_PUT:
 			ret = __memp_fput(mpf, addrp, flags);
+			break;
+		case QAM_PROBE_DIRTY:
+			return (__memp_dirty(mpf, addrp, txn, flags));
+		case QAM_PROBE_MPF:
+			*(DB_MPOOLFILE **)addrp = mpf;
+			return (0);
+		}
 
-		MUTEX_THREAD_LOCK(dbenv, dbp->mutexp);
+		MUTEX_LOCK(dbenv, dbp->mutex);
 		/* Recalculate because we dropped the lock. */
 		offset = extid - array->low_extent;
-		DB_ASSERT(array->mpfarray[offset].pinref > 0);
+		DB_ASSERT(dbenv, array->mpfarray[offset].pinref > 0);
 		if (--array->mpfarray[offset].pinref == 0 &&
 		    (mode == QAM_PROBE_GET || ret == 0)) {
 			/* Check to see if this file will be unlinked. */
@@ -285,7 +320,7 @@ err:
 					ret = t_ret;
 			}
 		}
-		MUTEX_THREAD_UNLOCK(dbenv, dbp->mutexp);
+		MUTEX_UNLOCK(dbenv, dbp->mutex);
 	}
 	return (ret);
 }
@@ -314,7 +349,7 @@ __qam_fclose(dbp, pgnoaddr)
 	dbenv = dbp->dbenv;
 	qp = (QUEUE *)dbp->q_internal;
 
-	MUTEX_THREAD_LOCK(dbenv, dbp->mutexp);
+	MUTEX_LOCK(dbenv, dbp->mutex);
 
 	extid = QAM_PAGE_EXTENT(dbp, pgnoaddr);
 	array = &qp->array1;
@@ -322,7 +357,8 @@ __qam_fclose(dbp, pgnoaddr)
 		array = &qp->array2;
 	offset = extid - array->low_extent;
 
-	DB_ASSERT(extid >= array->low_extent && offset < array->n_extent);
+	DB_ASSERT(dbenv,
+	    extid >= array->low_extent && offset < array->n_extent);
 
 	/* If other threads are still using this file, leave it. */
 	if (array->mpfarray[offset].pinref != 0)
@@ -333,7 +369,7 @@ __qam_fclose(dbp, pgnoaddr)
 	ret = __memp_fclose(mpf, 0);
 
 done:
-	MUTEX_THREAD_UNLOCK(dbenv, dbp->mutexp);
+	MUTEX_UNLOCK(dbenv, dbp->mutex);
 	return (ret);
 }
 
@@ -356,16 +392,13 @@ __qam_fremove(dbp, pgnoaddr)
 	MPFARRAY *array;
 	QUEUE *qp;
 	u_int32_t extid, offset;
-#ifdef CONFIG_TEST
-	char buf[MAXPATHLEN], *real_name;
-#endif
 	int ret;
 
 	qp = (QUEUE *)dbp->q_internal;
 	dbenv = dbp->dbenv;
 	ret = 0;
 
-	MUTEX_THREAD_LOCK(dbenv, dbp->mutexp);
+	MUTEX_LOCK(dbenv, dbp->mutex);
 
 	extid = QAM_PAGE_EXTENT(dbp, pgnoaddr);
 	array = &qp->array1;
@@ -373,16 +406,14 @@ __qam_fremove(dbp, pgnoaddr)
 		array = &qp->array2;
 	offset = extid - array->low_extent;
 
-	DB_ASSERT(extid >= array->low_extent && offset < array->n_extent);
+	DB_ASSERT(dbenv,
+	    extid >= array->low_extent && offset < array->n_extent);
 
-#ifdef CONFIG_TEST
-	real_name = NULL;
-	/* Find the real name of the file. */
-	QAM_EXNAME(qp, extid, buf, sizeof(buf));
-	if ((ret = __db_appname(dbenv,
-	    DB_APP_DATA, buf, 0, NULL, &real_name)) != 0)
+	mpf = array->mpfarray[offset].mpf;
+	/* This extent my already be marked for delete and closed. */
+	if (mpf == NULL)
 		goto err;
-#endif
+
 	/*
 	 * The log must be flushed before the file is deleted.  We depend on
 	 * the log record of the last delete to recreate the file if we crash.
@@ -390,7 +421,6 @@ __qam_fremove(dbp, pgnoaddr)
 	if (LOGGING_ON(dbenv) && (ret = __log_flush(dbenv, NULL)) != 0)
 		goto err;
 
-	mpf = array->mpfarray[offset].mpf;
 	(void)__memp_set_flags(mpf, DB_MPOOL_UNLINK, 1);
 	/* Someone could be real slow, let them close it down. */
 	if (array->mpfarray[offset].pinref != 0)
@@ -416,12 +446,8 @@ __qam_fremove(dbp, pgnoaddr)
 			array->hi_extent--;
 	}
 
-err:
-	MUTEX_THREAD_UNLOCK(dbenv, dbp->mutexp);
-#ifdef CONFIG_TEST
-	if (real_name != NULL)
-		__os_free(dbenv, real_name);
-#endif
+err:	MUTEX_UNLOCK(dbenv, dbp->mutex);
+
 	return (ret);
 }
 
@@ -491,7 +517,7 @@ __qam_gen_filelist(dbp, filelistp)
 
 	/* Find out the first and last record numbers in the database. */
 	i = PGNO_BASE_MD;
-	if ((ret = __memp_fget(mpf, &i, 0, &meta)) != 0)
+	if ((ret = __memp_fget(mpf, &i, NULL, 0, &meta)) != 0)
 		return (ret);
 
 	current = meta->cur_recno;
@@ -536,15 +562,15 @@ again:
 	first += stop % rec_extent;
 
 	for (i = first; i >= first && i <= stop; i += rec_extent) {
-		if ((ret = __qam_fprobe(dbp, QAM_RECNO_PAGE(dbp, i), &fp->mpf,
-		    QAM_PROBE_MPF, 0)) != 0) {
+		if ((ret = __qam_fprobe(dbp, QAM_RECNO_PAGE(dbp, i), NULL,
+		    &fp->mpf, QAM_PROBE_MPF, 0)) != 0) {
 			if (ret == ENOENT)
 				continue;
 			return (ret);
 		}
 		fp->id = QAM_RECNO_EXTENT(dbp, i);
 		fp++;
-		DB_ASSERT((size_t)(fp - *filelistp) < extent_cnt);
+		DB_ASSERT(dbenv, (size_t)(fp - *filelistp) < extent_cnt);
 	}
 
 	if (current < first) {
@@ -571,7 +597,7 @@ __qam_extent_names(dbenv, name, namelistp)
 	QUEUE_FILELIST *filelist, *fp;
 	size_t len;
 	int cnt, ret, t_ret;
-	char buf[MAXPATHLEN], **cp, *freep;
+	char buf[DB_MAXPATHLEN], **cp, *freep;
 
 	*namelistp = NULL;
 	filelist = NULL;
@@ -579,7 +605,7 @@ __qam_extent_names(dbenv, name, namelistp)
 		return (ret);
 	if ((ret = __db_open(dbp,
 	    NULL, name, NULL, DB_QUEUE, DB_RDONLY, 0, PGNO_BASE_MD)) != 0)
-		return (ret);
+		goto done;
 	qp = dbp->q_internal;
 	if (qp->page_ext == 0)
 		goto done;
@@ -677,7 +703,7 @@ int __qam_nameop(dbp, txn, newname, op)
 	u_int8_t fid[DB_FILE_ID_LEN];
 	u_int32_t exid;
 	int cnt, i, ret, t_ret;
-	char buf[MAXPATHLEN], nbuf[MAXPATHLEN], sepsave;
+	char buf[DB_MAXPATHLEN], nbuf[DB_MAXPATHLEN], sepsave;
 	char *endname, *endpath, *exname, *fullname, **names;
 	char *ndir, *namep, *new, *cp;
 
@@ -769,7 +795,7 @@ int __qam_nameop(dbp, txn, newname, op)
 			continue;
 		/* Make sure we have all numbers. foo.db vs. foo.db.0. */
 		for (cp = &names[i][len]; *cp != '\0'; cp++)
-			if (!isdigit(*cp))
+			if (!isdigit((int)*cp))
 				break;
 		if (*cp != '\0')
 			continue;
@@ -786,7 +812,8 @@ int __qam_nameop(dbp, txn, newname, op)
 			snprintf(exname, exlen,
 			     "%s%s", fullname, names[i] + len);
 			if ((t_ret = __memp_nameop(dbenv,
-			    fid, NULL, exname, NULL)) != 0 && ret == 0)
+			    fid, NULL, exname, NULL,
+			    F_ISSET(dbp, DB_AM_INMEM))) != 0 && ret == 0)
 				ret = t_ret;
 			break;
 
diff --git a/db/qam/qam_method.c b/db/qam/qam_method.c
index e3526fa3c..ff5129517 100644
--- a/db/qam/qam_method.c
+++ b/db/qam/qam_method.c
@@ -1,21 +1,16 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_method.c,v 11.84 2004/10/14 18:09:32 bostic Exp $
+ * $Id: qam_method.c,v 12.8 2006/08/24 14:46:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_page.h"
-#include "dbinc/db_shash.h"
 #include "dbinc/db_am.h"
 #include "dbinc/lock.h"
 #include "dbinc/mp.h"
@@ -130,7 +125,7 @@ __qam_set_extentsize(dbp, extentsize)
 	DB_ILLEGAL_AFTER_OPEN(dbp, "DB->set_extentsize");
 
 	if (extentsize < 1) {
-		__db_err(dbp->dbenv, "Extent size must be at least 1");
+		__db_errx(dbp->dbenv, "Extent size must be at least 1");
 		return (EINVAL);
 	}
 
@@ -163,7 +158,7 @@ __queue_pageinfo(dbp, firstp, lastp, emptyp, prpage, flags)
 
 	/* Find out the page number of the last page in the database. */
 	i = PGNO_BASE_MD;
-	if ((ret = __memp_fget(mpf, &i, 0, &meta)) != 0)
+	if ((ret = __memp_fget(mpf, &i, NULL, 0, &meta)) != 0)
 		return (ret);
 
 	first = QAM_RECNO_PAGE(dbp, meta->first_recno);
@@ -222,7 +217,7 @@ __db_prqueue(dbp, flags)
 	/* Dump each page. */
 begin:
 	for (; i <= stop; ++i) {
-		if ((ret = __qam_fget(dbp, &i, 0, &h)) != 0) {
+		if ((ret = __qam_fget(dbp, &i, NULL, 0, &h)) != 0) {
 			pg_ext = ((QUEUE *)dbp->q_internal)->page_ext;
 			if (pg_ext == 0) {
 				if (ret == DB_PAGE_NOTFOUND && first == last)
@@ -302,8 +297,8 @@ __qam_rr(dbp, txn, name, subdb, newname, op)
 
 	PANIC_CHECK(dbenv);
 
-	if (subdb != NULL) {
-		__db_err(dbenv,
+	if (subdb != NULL && name != NULL) {
+		__db_errx(dbenv,
 		    "Queue does not support multiple databases per file");
 		return (EINVAL);
 	}
diff --git a/db/qam/qam_open.c b/db/qam/qam_open.c
index 595d74dac..75df8e41c 100644
--- a/db/qam/qam_open.c
+++ b/db/qam/qam_open.c
@@ -1,24 +1,17 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_open.c,v 11.68 2004/02/27 12:38:31 bostic Exp $
+ * $Id: qam_open.c,v 12.11 2006/08/24 14:46:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#include <string.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/crypto.h"
 #include "dbinc/db_page.h"
-#include "dbinc/db_shash.h"
 #include "dbinc/db_swap.h"
 #include "dbinc/db_am.h"
 #include "dbinc/lock.h"
@@ -58,11 +51,17 @@ __qam_open(dbp, txn, name, base_pgno, mode, flags)
 	qmeta = NULL;
 
 	if (name == NULL && t->page_ext != 0) {
-		__db_err(dbenv,
+		__db_errx(dbenv,
 	"Extent size may not be specified for in-memory queue database");
 		return (EINVAL);
 	}
 
+	if (MULTIVERSION(dbp)) {
+		__db_errx(dbenv,
+		    "Multiversion queue databases are not supported");
+		return (EINVAL);
+	}
+
 	/* Initialize the remaining fields/methods of the DB. */
 	dbp->db_am_remove = __qam_remove;
 	dbp->db_am_rename = __qam_rename;
@@ -86,12 +85,12 @@ __qam_open(dbp, txn, name, base_pgno, mode, flags)
 	if ((ret =
 	    __db_lget(dbc, 0, base_pgno, DB_LOCK_READ, 0, &metalock)) != 0)
 		goto err;
-	if ((ret = __memp_fget(mpf, &base_pgno, 0, &qmeta)) != 0)
+	if ((ret = __memp_fget(mpf, &base_pgno, dbc->txn, 0, &qmeta)) != 0)
 		goto err;
 
 	/* If the magic number is incorrect, that's a fatal error. */
 	if (qmeta->dbmeta.magic != DB_QAMMAGIC) {
-		__db_err(dbenv, "%s: unexpected file type or format", name);
+		__db_errx(dbenv, "%s: unexpected file type or format", name);
 		ret = EINVAL;
 		goto err;
 	}
@@ -103,7 +102,7 @@ __qam_open(dbp, txn, name, base_pgno, mode, flags)
 		goto err;
 
 	if (mode == 0)
-		mode = __db_omode("rwrw--");
+		mode = __db_omode("rw-rw----");
 	t->mode = mode;
 	t->re_pad = qmeta->re_pad;
 	t->re_len = qmeta->re_len;
@@ -188,7 +187,7 @@ __qam_metachk(dbp, name, qmeta)
 	switch (vers) {
 	case 1:
 	case 2:
-		__db_err(dbenv,
+		__db_errx(dbenv,
 		    "%s: queue version %lu requires a version upgrade",
 		    name, (u_long)vers);
 		return (DB_OLD_VERSION);
@@ -196,7 +195,7 @@ __qam_metachk(dbp, name, qmeta)
 	case 4:
 		break;
 	default:
-		__db_err(dbenv,
+		__db_errx(dbenv,
 		    "%s: unsupported qam version: %lu", name, (u_long)vers);
 		return (EINVAL);
 	}
@@ -233,8 +232,10 @@ __qam_init_meta(dbp, meta)
 	DB *dbp;
 	QMETA *meta;
 {
+	DB_ENV *dbenv;
 	QUEUE *t;
 
+	dbenv = dbp->dbenv;
 	t = dbp->q_internal;
 
 	memset(meta, 0, sizeof(QMETA));
@@ -248,8 +249,8 @@ __qam_init_meta(dbp, meta)
 		FLD_SET(meta->dbmeta.metaflags, DBMETA_CHKSUM);
 	if (F_ISSET(dbp, DB_AM_ENCRYPT)) {
 		meta->dbmeta.encrypt_alg =
-		   ((DB_CIPHER *)dbp->dbenv->crypto_handle)->alg;
-		DB_ASSERT(meta->dbmeta.encrypt_alg != 0);
+		   ((DB_CIPHER *)dbenv->crypto_handle)->alg;
+		DB_ASSERT(dbenv, meta->dbmeta.encrypt_alg != 0);
 		meta->crypto_magic = meta->dbmeta.magic;
 	}
 	meta->dbmeta.type = P_QAMMETA;
@@ -264,7 +265,7 @@ __qam_init_meta(dbp, meta)
 
 	/* Verify that we can fit at least one record per page. */
 	if (QAM_RECNO_PER_PAGE(dbp) < 1) {
-		__db_err(dbp->dbenv,
+		__db_errx(dbenv,
 		    "Record size of %lu too large for page size of %lu",
 		    (u_long)t->re_len, (u_long)dbp->pgsize);
 		return (EINVAL);
@@ -308,11 +309,12 @@ __qam_new_file(dbp, txn, fhp, name)
 
 	/* Build meta-data page. */
 
-	if (name == NULL) {
+	if (F_ISSET(dbp, DB_AM_INMEM)) {
 		pgno = PGNO_BASE_MD;
-		ret = __memp_fget(mpf, &pgno, DB_MPOOL_CREATE, &meta);
+		ret = __memp_fget(mpf, &pgno, txn,
+		    DB_MPOOL_CREATE | DB_MPOOL_DIRTY, &meta);
 	} else {
-		ret = __os_calloc(dbp->dbenv, 1, dbp->pgsize, &buf);
+		ret = __os_calloc(dbenv, 1, dbp->pgsize, &buf);
 		meta = (QMETA *)buf;
 	}
 	if (ret != 0)
@@ -321,9 +323,12 @@ __qam_new_file(dbp, txn, fhp, name)
 	if ((ret = __qam_init_meta(dbp, meta)) != 0)
 		goto err;
 
-	if (name == NULL)
-		ret = __memp_fput(mpf, meta, DB_MPOOL_DIRTY);
-	else {
+	if (F_ISSET(dbp, DB_AM_INMEM)) {
+		if ((ret = __db_log_page(dbp,
+		    txn, &meta->dbmeta.lsn, pgno, (PAGE *)meta)) != 0)
+			goto err;
+		ret = __memp_fput(mpf, meta, 0);
+	} else {
 		pginfo.db_pagesize = dbp->pgsize;
 		pginfo.flags =
 		    F_ISSET(dbp, (DB_AM_CHKSUM | DB_AM_ENCRYPT | DB_AM_SWAP));
diff --git a/db/qam/qam_rec.c b/db/qam/qam_rec.c
index e92141ddd..88a7bf261 100644
--- a/db/qam/qam_rec.c
+++ b/db/qam/qam_rec.c
@@ -1,28 +1,42 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_rec.c,v 11.78 2004/05/11 14:04:51 bostic Exp $
+ * $Id: qam_rec.c,v 12.20 2006/09/09 14:28:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#include <string.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_page.h"
-#include "dbinc/db_shash.h"
 #include "dbinc/db_am.h"
 #include "dbinc/lock.h"
 #include "dbinc/log.h"
 #include "dbinc/mp.h"
 #include "dbinc/qam.h"
+#include "dbinc/txn.h"
+
+/*
+ * LSNs in queue data pages are advisory.  They do not have to be accurate
+ * as all operations are idempotent on records.  They should not be rolled
+ * forward during recovery as committed transaction may obscure updates from
+ * an incomplete transaction that updates the same page.  The incomplete
+ * transaction may be completed during a later hot backup cycle.
+ */
+
+/* Determine if we are restoring prepared transactions from __txn_recover. */
+#define	IS_IN_RESTORE(dbenv)						\
+	(((DB_TXNREGION *)						\
+	     (dbenv)->tx_handle->reginfo.primary)->stat.st_nrestores != 0)
+
+/* Queue version of REC_DIRTY -- needs to probe the correct file. */
+#define	QAM_DIRTY(dbp, pgno, pagep)					\
+	if ((ret = __qam_dirty((dbp), pgno, pagep, NULL, 0)) != 0) {	\
+		ret = __db_pgerr((dbp), (pgno), ret);			\
+		goto out;						\
+	}
 
 /*
  * __qam_incfirst_recover --
@@ -49,20 +63,23 @@ __qam_incfirst_recover(dbenv, dbtp, lsnp, op, info)
 	QUEUE_CURSOR *cp;
 	db_pgno_t metapg;
 	u_int32_t rec_ext;
-	int exact, modified, ret, t_ret;
+	int exact, ret, t_ret;
 
+	LOCK_INIT(lock);
+	COMPQUIET(meta, NULL);
 	REC_PRINT(__qam_incfirst_print);
-	REC_INTRO(__qam_incfirst_read, 1);
+	REC_INTRO(__qam_incfirst_read, 1, 1);
 
 	metapg = ((QUEUE *)file_dbp->q_internal)->q_meta;
 
 	if ((ret = __db_lget(dbc,
 	    LCK_ROLLBACK, metapg,  DB_LOCK_WRITE, 0, &lock)) != 0)
 		goto done;
-	if ((ret = __memp_fget(mpf, &metapg, 0, &meta)) != 0) {
+	if ((ret = __memp_fget(mpf, &metapg, NULL,
+	    0, &meta)) != 0) {
 		if (DB_REDO(op)) {
-			if ((ret = __memp_fget(mpf,
-			    &metapg, DB_MPOOL_CREATE, &meta)) != 0) {
+			if ((ret = __memp_fget(mpf, &metapg, NULL,
+			    DB_MPOOL_CREATE, &meta)) != 0) {
 				(void)__LPUT(dbc, lock);
 				goto out;
 			}
@@ -75,8 +92,6 @@ __qam_incfirst_recover(dbenv, dbtp, lsnp, op, info)
 		}
 	}
 
-	modified = 0;
-
 	/*
 	 * Only move first_recno backwards so we pick up the aborted delete.
 	 * When going forward we need to be careful since
@@ -84,21 +99,21 @@ __qam_incfirst_recover(dbenv, dbtp, lsnp, op, info)
 	 */
 	if (DB_UNDO(op)) {
 		if (QAM_BEFORE_FIRST(meta, argp->recno)) {
+			REC_DIRTY(mpf, &meta);
 			meta->first_recno = argp->recno;
-			modified = 1;
 		}
 
 		trunc_lsn = ((DB_TXNHEAD *)info)->trunc_lsn;
 		/* if we are truncating, update the LSN */
 		if (!IS_ZERO_LSN(trunc_lsn) &&
-		    log_compare(&LSN(meta), &trunc_lsn) > 0) {
+		    LOG_COMPARE(&LSN(meta), &trunc_lsn) > 0) {
+			REC_DIRTY(mpf, &meta);
 			LSN(meta) = trunc_lsn;
-			modified = 1;
 		}
 	} else {
-		if (log_compare(&LSN(meta), lsnp) < 0) {
+		if (LOG_COMPARE(&LSN(meta), lsnp) < 0) {
+			REC_DIRTY(mpf, &meta);
 			LSN(meta) = *lsnp;
-			modified = 1;
 		}
 		if (meta->page_ext == 0)
 			rec_ext = 0;
@@ -110,7 +125,7 @@ __qam_incfirst_recover(dbenv, dbtp, lsnp, op, info)
 		while (meta->first_recno != meta->cur_recno &&
 		    !QAM_BEFORE_FIRST(meta, argp->recno + 1)) {
 			if ((ret = __qam_position(dbc,
-			    &meta->first_recno, QAM_READ, &exact)) != 0)
+			    &meta->first_recno, DB_LOCK_READ, 0, &exact)) != 0)
 				goto err;
 			if (cp->page != NULL && (ret =
 			    __qam_fput(file_dbp, cp->pgno, cp->page, 0)) != 0)
@@ -123,14 +138,14 @@ __qam_incfirst_recover(dbenv, dbtp, lsnp, op, info)
 				if ((ret =
 				    __qam_fremove(file_dbp, cp->pgno)) != 0)
 					goto err;
+			REC_DIRTY(mpf, &meta);
 			meta->first_recno++;
 			if (meta->first_recno == RECNO_OOB)
 				meta->first_recno++;
-			modified = 1;
 		}
 	}
 
-	ret = __memp_fput(mpf, meta, modified ? DB_MPOOL_DIRTY : 0);
+	ret = __memp_fput(mpf, meta, 0);
 	if ((t_ret = __LPUT(dbc, lock)) != 0 && ret == 0)
 		ret = t_ret;
 	if (ret != 0)
@@ -169,21 +184,23 @@ __qam_mvptr_recover(dbenv, dbtp, lsnp, op, info)
 	DB_LOCK lock;
 	DB_MPOOLFILE *mpf;
 	QMETA *meta;
+	QUEUE_CURSOR *cp;
 	db_pgno_t metapg;
-	int cmp_n, cmp_p, modified, ret;
+	int cmp_n, cmp_p, exact, ret;
 
 	REC_PRINT(__qam_mvptr_print);
-	REC_INTRO(__qam_mvptr_read, 1);
+	REC_INTRO(__qam_mvptr_read, 1, 1);
 
 	metapg = ((QUEUE *)file_dbp->q_internal)->q_meta;
 
 	if ((ret = __db_lget(dbc,
 	    LCK_ROLLBACK, metapg,  DB_LOCK_WRITE, 0, &lock)) != 0)
 		goto done;
-	if ((ret = __memp_fget(mpf, &metapg, 0, &meta)) != 0) {
+	if ((ret = __memp_fget(mpf, &metapg, NULL,
+	    0, &meta)) != 0) {
 		if (DB_REDO(op)) {
-			if ((ret = __memp_fget(mpf,
-			    &metapg, DB_MPOOL_CREATE, &meta)) != 0) {
+			if ((ret = __memp_fget(mpf, &metapg, NULL,
+			    DB_MPOOL_CREATE, &meta)) != 0) {
 				(void)__LPUT(dbc, lock);
 				goto out;
 			}
@@ -196,43 +213,75 @@ __qam_mvptr_recover(dbenv, dbtp, lsnp, op, info)
 		}
 	}
 
-	modified = 0;
-	cmp_n = log_compare(lsnp, &LSN(meta));
-	cmp_p = log_compare(&LSN(meta), &argp->metalsn);
+	cmp_n = LOG_COMPARE(lsnp, &LSN(meta));
+	cmp_p = LOG_COMPARE(&LSN(meta), &argp->metalsn);
 
 	/*
 	 * Under normal circumstances, we never undo a movement of one of
 	 * the pointers.  Just move them along regardless of abort/commit.
+	 * When going forward we need to verify that this is really where
+	 * the pointer belongs.  A transaction may roll back and reinsert
+	 * a record that was missing at the time of this action.
 	 *
 	 * If we're undoing a truncate, we need to reset the pointers to
 	 * their state before the truncate.
 	 */
 	if (DB_UNDO(op)) {
 		if ((argp->opcode & QAM_TRUNCATE) && cmp_n <= 0) {
+			REC_DIRTY(mpf, &meta);
 			meta->first_recno = argp->old_first;
 			meta->cur_recno = argp->old_cur;
 			LSN(meta) = argp->metalsn;
-			modified = 1;
 		}
 		/* If the page lsn is beyond the truncate point, move it back */
 		trunc_lsn = ((DB_TXNHEAD *)info)->trunc_lsn;
 		if (!IS_ZERO_LSN(trunc_lsn) &&
-		    log_compare(&trunc_lsn, &LSN(meta)) < 0) {
+		    LOG_COMPARE(&trunc_lsn, &LSN(meta)) < 0) {
+			REC_DIRTY(mpf, &meta);
 			LSN(meta) = argp->metalsn;
-			modified = 1;
 		}
 	} else if (op == DB_TXN_APPLY || cmp_p == 0) {
-		if (argp->opcode & QAM_SETFIRST)
-			meta->first_recno = argp->new_first;
+		REC_DIRTY(mpf, &meta);
+		cp = (QUEUE_CURSOR *)dbc->internal;
+		if ((argp->opcode & QAM_SETFIRST) &&
+		    meta->first_recno == argp->old_first) {
+			if (argp->old_first > argp->new_first)
+				meta->first_recno = argp->new_first;
+			else {
+				if ((ret = __qam_position(dbc,
+				    &meta->first_recno, DB_LOCK_READ, 0,
+				    &exact)) != 0)
+					goto err;
+				if (!exact)
+					meta->first_recno = argp->new_first;
+				if (cp->page != NULL &&
+				    (ret = __qam_fput(file_dbp,
+				    cp->pgno, cp->page, 0)) != 0)
+					goto err;
+			}
+		}
 
-		if (argp->opcode & QAM_SETCUR)
-			meta->cur_recno = argp->new_cur;
+		if ((argp->opcode & QAM_SETCUR) &&
+		    meta->cur_recno == argp->old_cur) {
+			if (argp->old_cur < argp->new_cur)
+				meta->cur_recno = argp->new_cur;
+			else {
+				if ((ret = __qam_position(dbc, &meta->cur_recno,
+				    DB_LOCK_READ, 0, &exact)) != 0)
+					goto err;
+				if (!exact)
+					meta->cur_recno = argp->new_cur;
+				if (cp->page != NULL &&
+				    (ret = __qam_fput(file_dbp,
+				    cp->pgno, cp->page, 0)) != 0)
+					goto err;
+			}
+		}
 
-		modified = 1;
 		meta->dbmeta.lsn = *lsnp;
 	}
 
-	if ((ret = __memp_fput(mpf, meta, modified ? DB_MPOOL_DIRTY : 0)) != 0)
+	if ((ret = __memp_fput(mpf, meta, 0)) != 0)
 		goto out;
 
 	if ((ret = __LPUT(dbc, lock)) != 0)
@@ -241,6 +290,11 @@ __qam_mvptr_recover(dbenv, dbtp, lsnp, op, info)
 done:	*lsnp = argp->prev_lsn;
 	ret = 0;
 
+	if (0) {
+err:		(void)__memp_fput(mpf, meta, 0);
+		(void)__LPUT(dbc, lock);
+	}
+
 out:	REC_CLOSE;
 }
 
@@ -269,24 +323,24 @@ __qam_del_recover(dbenv, dbtp, lsnp, op, info)
 	QMETA *meta;
 	QPAGE *pagep;
 	db_pgno_t metapg;
-	int cmp_n, modified, ret, t_ret;
+	int cmp_n, ret, t_ret;
 
 	COMPQUIET(info, NULL);
+	COMPQUIET(pagep, NULL);
 	REC_PRINT(__qam_del_print);
-	REC_INTRO(__qam_del_read, 1);
+	REC_INTRO(__qam_del_read, 1, 1);
 
-	if ((ret = __qam_fget(file_dbp,
-	    &argp->pgno, DB_MPOOL_CREATE, &pagep)) != 0)
+	if ((ret = __qam_fget(file_dbp, &argp->pgno, NULL,
+	    DB_MPOOL_CREATE, &pagep)) != 0)
 		goto out;
 
-	modified = 0;
 	if (pagep->pgno == PGNO_INVALID) {
+		QAM_DIRTY(file_dbp, argp->pgno, &pagep);
 		pagep->pgno = argp->pgno;
 		pagep->type = P_QAMDATA;
-		modified = 1;
 	}
 
-	cmp_n = log_compare(lsnp, &LSN(pagep));
+	cmp_n = LOG_COMPARE(lsnp, &LSN(pagep));
 
 	if (DB_UNDO(op)) {
 		/* make sure first is behind us */
@@ -294,7 +348,8 @@ __qam_del_recover(dbenv, dbtp, lsnp, op, info)
 		if ((ret = __db_lget(dbc,
 		    LCK_ROLLBACK, metapg, DB_LOCK_WRITE, 0, &lock)) != 0)
 			goto err;
-		if ((ret = __memp_fget(mpf, &metapg, 0, &meta)) != 0) {
+		if ((ret = __memp_fget(mpf, &metapg, NULL,
+		    DB_MPOOL_EDIT, &meta)) != 0) {
 			(void)__LPUT(dbc, lock);
 			goto err;
 		}
@@ -303,8 +358,9 @@ __qam_del_recover(dbenv, dbtp, lsnp, op, info)
 		    (meta->first_recno <= meta->cur_recno ||
 		    meta->first_recno -
 		    argp->recno < argp->recno - meta->cur_recno))) {
+			REC_DIRTY(mpf, &meta);
 			meta->first_recno = argp->recno;
-			ret = __memp_fput(mpf, meta, DB_MPOOL_DIRTY);
+			ret = __memp_fput(mpf, meta, 0);
 		} else
 			ret = __memp_fput(mpf, meta, 0);
 		if ((t_ret = __LPUT(dbc, lock)) != 0 && ret == 0)
@@ -313,6 +369,7 @@ __qam_del_recover(dbenv, dbtp, lsnp, op, info)
 			goto err;
 
 		/* Need to undo delete - mark the record as present */
+		QAM_DIRTY(file_dbp, pagep->pgno, &pagep);
 		qp = QAM_GET_RECORD(file_dbp, pagep, argp->indx);
 		F_SET(qp, QAM_VALID);
 
@@ -323,19 +380,28 @@ __qam_del_recover(dbenv, dbtp, lsnp, op, info)
 		 * foul up a concurrent put.  Having too late an LSN
 		 * is harmless in queue except when we're determining
 		 * what we need to roll forward during recovery.  [#2588]
+		 * If we are aborting a restored transaction then it
+		 * might get rolled forward later so the LSN needs to
+		 * be correct in that case too. [#12181]
 		 */
-		if (op == DB_TXN_BACKWARD_ROLL && cmp_n <= 0)
+		if (cmp_n <= 0 &&
+		      (op == DB_TXN_BACKWARD_ROLL || IS_IN_RESTORE(dbenv)))
 			LSN(pagep) = argp->lsn;
-		modified = 1;
 	} else if (op == DB_TXN_APPLY || (cmp_n > 0 && DB_REDO(op))) {
 		/* Need to redo delete - clear the valid bit */
+		QAM_DIRTY(file_dbp, pagep->pgno, &pagep);
 		qp = QAM_GET_RECORD(file_dbp, pagep, argp->indx);
 		F_CLR(qp, QAM_VALID);
-		LSN(pagep) = *lsnp;
-		modified = 1;
+		/*
+		 * We only move the LSN forward during replication.
+		 * During recovery we could obscure an update from
+		 * a partially completed transaction while processing
+		 * a hot backup.  [#13823]
+		 */
+		if (op == DB_TXN_APPLY)
+			LSN(pagep) = *lsnp;
 	}
-	if ((ret = __qam_fput(file_dbp,
-	    argp->pgno, pagep, modified ? DB_MPOOL_DIRTY : 0)) != 0)
+	if ((ret = __qam_fput(file_dbp, argp->pgno, pagep, 0)) != 0)
 		goto out;
 
 done:	*lsnp = argp->prev_lsn;
@@ -371,34 +437,31 @@ __qam_delext_recover(dbenv, dbtp, lsnp, op, info)
 	QMETA *meta;
 	QPAGE *pagep;
 	db_pgno_t metapg;
-	int cmp_n, modified, ret, t_ret;
+	int cmp_n, ret, t_ret;
 
 	COMPQUIET(info, NULL);
+	COMPQUIET(pagep, NULL);
 	REC_PRINT(__qam_delext_print);
-	REC_INTRO(__qam_delext_read, 1);
+	REC_INTRO(__qam_delext_read, 1, 1);
 
-	if ((ret = __qam_fget(file_dbp, &argp->pgno, 0, &pagep)) != 0) {
-		if (ret != DB_PAGE_NOTFOUND && ret != ENOENT)
-			goto out;
+	if ((ret = __qam_fget(file_dbp, &argp->pgno, NULL,
+	     DB_REDO(op) ? 0 : DB_MPOOL_CREATE, &pagep)) != 0) {
 		/*
 		 * If we are redoing a delete and the page is not there
 		 * we are done.
 		 */
-		if (DB_REDO(op))
+		if (DB_REDO(op) && (ret == DB_PAGE_NOTFOUND || ret == ENOENT))
 			goto done;
-		if ((ret = __qam_fget(file_dbp,
-		    &argp->pgno, DB_MPOOL_CREATE, &pagep)) != 0)
-			goto out;
+		goto out;
 	}
 
-	modified = 0;
 	if (pagep->pgno == PGNO_INVALID) {
+		QAM_DIRTY(file_dbp, argp->pgno, &pagep);
 		pagep->pgno = argp->pgno;
 		pagep->type = P_QAMDATA;
-		modified = 1;
 	}
 
-	cmp_n = log_compare(lsnp, &LSN(pagep));
+	cmp_n = LOG_COMPARE(lsnp, &LSN(pagep));
 
 	if (DB_UNDO(op)) {
 		/* make sure first is behind us */
@@ -406,7 +469,8 @@ __qam_delext_recover(dbenv, dbtp, lsnp, op, info)
 		if ((ret = __db_lget(dbc,
 		    LCK_ROLLBACK, metapg, DB_LOCK_WRITE, 0, &lock)) != 0)
 			goto err;
-		if ((ret = __memp_fget(mpf, &metapg, 0, &meta)) != 0) {
+		if ((ret = __memp_fget(mpf, &metapg, NULL,
+		    DB_MPOOL_EDIT, &meta)) != 0) {
 			(void)__LPUT(dbc, lock);
 			goto err;
 		}
@@ -416,7 +480,7 @@ __qam_delext_recover(dbenv, dbtp, lsnp, op, info)
 		    meta->first_recno -
 		    argp->recno < argp->recno - meta->cur_recno))) {
 			meta->first_recno = argp->recno;
-			ret = __memp_fput(mpf, meta, DB_MPOOL_DIRTY);
+			ret = __memp_fput(mpf, meta, 0);
 		} else
 			ret = __memp_fput(mpf, meta, 0);
 		if ((t_ret = __LPUT(dbc, lock)) != 0 && ret == 0)
@@ -424,6 +488,7 @@ __qam_delext_recover(dbenv, dbtp, lsnp, op, info)
 		if (ret != 0)
 			goto err;
 
+		QAM_DIRTY(file_dbp, pagep->pgno, &pagep);
 		if ((ret = __qam_pitem(dbc, pagep,
 		    argp->indx, argp->recno, &argp->data)) != 0)
 			goto err;
@@ -436,18 +501,24 @@ __qam_delext_recover(dbenv, dbtp, lsnp, op, info)
 		 * is harmless in queue except when we're determining
 		 * what we need to roll forward during recovery.  [#2588]
 		 */
-		if (op == DB_TXN_BACKWARD_ROLL && cmp_n <= 0)
+		if (cmp_n <= 0 &&
+		      (op == DB_TXN_BACKWARD_ROLL || IS_IN_RESTORE(dbenv)))
 			LSN(pagep) = argp->lsn;
-		modified = 1;
 	} else if (op == DB_TXN_APPLY || (cmp_n > 0 && DB_REDO(op))) {
+		QAM_DIRTY(file_dbp, pagep->pgno, &pagep);
 		/* Need to redo delete - clear the valid bit */
 		qp = QAM_GET_RECORD(file_dbp, pagep, argp->indx);
 		F_CLR(qp, QAM_VALID);
-		LSN(pagep) = *lsnp;
-		modified = 1;
+		/*
+		 * We only move the LSN forward during replication.
+		 * During recovery we could obscure an update from
+		 * a partially completed transaction while processing
+		 * a hot backup.  [#13823]
+		 */
+		if (op == DB_TXN_APPLY)
+			LSN(pagep) = *lsnp;
 	}
-	if ((ret = __qam_fput(file_dbp,
-	    argp->pgno, pagep, modified ? DB_MPOOL_DIRTY : 0)) != 0)
+	if ((ret = __qam_fput(file_dbp, argp->pgno, pagep, 0)) != 0)
 		goto out;
 
 done:	*lsnp = argp->prev_lsn;
@@ -482,62 +553,65 @@ __qam_add_recover(dbenv, dbtp, lsnp, op, info)
 	QMETA *meta;
 	QPAGE *pagep;
 	db_pgno_t metapg;
-	int cmp_n, meta_dirty, modified, ret;
+	int cmp_n, ret;
 
 	COMPQUIET(info, NULL);
+	COMPQUIET(pagep, NULL);
 	REC_PRINT(__qam_add_print);
-	REC_INTRO(__qam_add_read, 1);
+	REC_INTRO(__qam_add_read, 1, 1);
 
-	modified = 0;
-	if ((ret = __qam_fget(file_dbp, &argp->pgno, 0, &pagep)) != 0) {
-		if (ret != DB_PAGE_NOTFOUND && ret != ENOENT)
-			goto out;
+	if ((ret = __qam_fget(file_dbp, &argp->pgno, NULL,
+	     DB_UNDO(op) ? 0 : DB_MPOOL_CREATE, &pagep)) != 0) {
 		/*
 		 * If we are undoing an append and the page is not there
 		 * we are done.
 		 */
-		if (DB_UNDO(op))
+		if (DB_UNDO(op) && (ret == DB_PAGE_NOTFOUND || ret == ENOENT))
 			goto done;
-		if ((ret = __qam_fget(file_dbp,
-		    &argp->pgno, DB_MPOOL_CREATE, &pagep)) != 0)
-			goto out;
+		goto out;
 	}
 
 	if (pagep->pgno == PGNO_INVALID) {
+		QAM_DIRTY(file_dbp, argp->pgno, &pagep);
 		pagep->pgno = argp->pgno;
 		pagep->type = P_QAMDATA;
-		modified = 1;
 	}
 
-	cmp_n = log_compare(lsnp, &LSN(pagep));
+	cmp_n = LOG_COMPARE(lsnp, &LSN(pagep));
 
 	if (DB_REDO(op)) {
 		/* Fix meta-data page. */
 		metapg = ((QUEUE *)file_dbp->q_internal)->q_meta;
-		if ((ret = __memp_fget(mpf, &metapg, 0, &meta)) != 0)
+		if ((ret = __memp_fget(mpf, &metapg, NULL,
+		    0, &meta)) != 0)
 			goto err;
-		meta_dirty = 0;
 		if (QAM_BEFORE_FIRST(meta, argp->recno)) {
+			REC_DIRTY(mpf, &meta);
 			meta->first_recno = argp->recno;
-			meta_dirty = 1;
 		}
 		if (argp->recno == meta->cur_recno ||
-		   QAM_AFTER_CURRENT(meta, argp->recno)) {
+		    QAM_AFTER_CURRENT(meta, argp->recno)) {
+			REC_DIRTY(mpf, &meta);
 			meta->cur_recno = argp->recno + 1;
-			meta_dirty = 1;
 		}
-		if ((ret = __memp_fput(mpf,
-		    meta, meta_dirty? DB_MPOOL_DIRTY : 0)) != 0)
+		if ((ret = __memp_fput(mpf, meta, 0)) != 0)
 			goto err;
 
 		/* Now update the actual page if necessary. */
 		if (op == DB_TXN_APPLY || cmp_n > 0) {
+			QAM_DIRTY(file_dbp, pagep->pgno, &pagep);
 			/* Need to redo add - put the record on page */
 			if ((ret = __qam_pitem(dbc,
 			    pagep, argp->indx, argp->recno, &argp->data)) != 0)
 				goto err;
-			LSN(pagep) = *lsnp;
-			modified = 1;
+			/*
+			 * We only move the LSN forward during replication.
+			 * During recovery we could obscure an update from
+			 * a partially completed transaction while processing
+			 * a hot backup.  [#13823]
+			 */
+			if (op == DB_TXN_APPLY)
+				LSN(pagep) = *lsnp;
 		}
 	} else if (DB_UNDO(op)) {
 		/*
@@ -546,6 +620,7 @@ __qam_add_recover(dbenv, dbtp, lsnp, op, info)
 		 *	Otherwise just clear the valid bit
 		 */
 		if (argp->olddata.size != 0) {
+			QAM_DIRTY(file_dbp, pagep->pgno, &pagep);
 			if ((ret = __qam_pitem(dbc, pagep,
 			    argp->indx, argp->recno, &argp->olddata)) != 0)
 				goto err;
@@ -555,11 +630,10 @@ __qam_add_recover(dbenv, dbtp, lsnp, op, info)
 				    file_dbp, pagep, argp->indx);
 				F_CLR(qp, QAM_VALID);
 			}
-			modified = 1;
 		} else {
+			QAM_DIRTY(file_dbp, pagep->pgno, &pagep);
 			qp = QAM_GET_RECORD(file_dbp, pagep, argp->indx);
 			qp->flags = 0;
-			modified = 1;
 		}
 
 		/*
@@ -570,12 +644,13 @@ __qam_add_recover(dbenv, dbtp, lsnp, op, info)
 		 * is harmless in queue except when we're determining
 		 * what we need to roll forward during recovery.  [#2588]
 		 */
-		if (op == DB_TXN_BACKWARD_ROLL && cmp_n <= 0)
+		if (cmp_n <= 0 &&
+		      (op == DB_TXN_BACKWARD_ROLL || IS_IN_RESTORE(dbenv)))
 			LSN(pagep) = argp->lsn;
 	}
 
 	if ((ret = __qam_fput(file_dbp,
-	    argp->pgno, pagep, modified ? DB_MPOOL_DIRTY : 0)) != 0)
+	    argp->pgno, pagep, 0)) != 0)
 		goto out;
 
 done:	*lsnp = argp->prev_lsn;
diff --git a/db/qam/qam_stat.c b/db/qam/qam_stat.c
index c5264bd01..067555a57 100644
--- a/db/qam/qam_stat.c
+++ b/db/qam/qam_stat.c
@@ -1,24 +1,16 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_stat.c,v 11.47 2004/09/22 16:29:47 bostic Exp $
+ * $Id: qam_stat.c,v 12.8 2006/08/24 14:46:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#include <ctype.h>
-#include <string.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_page.h"
-#include "dbinc/db_shash.h"
 #include "dbinc/db_am.h"
 #include "dbinc/lock.h"
 #include "dbinc/mp.h"
@@ -70,10 +62,10 @@ __qam_stat(dbc, spp, flags)
 	/* Determine the last page of the database. */
 	if ((ret = __db_lget(dbc, 0, t->q_meta, DB_LOCK_READ, 0, &lock)) != 0)
 		goto err;
-	if ((ret = __memp_fget(mpf, &t->q_meta, 0, &meta)) != 0)
+	if ((ret = __memp_fget(mpf, &t->q_meta, dbc->txn, 0, &meta)) != 0)
 		goto err;
 
-	if (flags == DB_FAST_STAT || flags == DB_CACHED_COUNTS) {
+	if (flags == DB_FAST_STAT) {
 		sp->qs_nkeys = meta->dbmeta.key_count;
 		sp->qs_ndata = meta->dbmeta.record_count;
 		goto meta_only;
@@ -102,7 +94,7 @@ begin:
 		if ((ret =
 		    __db_lget(dbc, 0, pgno, DB_LOCK_READ, 0, &lock)) != 0)
 			goto err;
-		ret = __qam_fget(dbp, &pgno, 0, &h);
+		ret = __qam_fget(dbp, &pgno, dbc->txn, 0, &h);
 		if (ret == ENOENT) {
 			pgno += pg_ext - 1;
 			continue;
@@ -153,7 +145,8 @@ begin:
 	    0, t->q_meta, F_ISSET(dbp, DB_AM_RDONLY) ?
 	    DB_LOCK_READ : DB_LOCK_WRITE, 0, &lock)) != 0)
 		goto err;
-	if ((ret = __memp_fget(mpf, &t->q_meta, 0, &meta)) != 0)
+	if ((ret = __memp_fget(mpf, &t->q_meta, dbc->txn,
+	    F_ISSET(dbp, DB_AM_RDONLY) ? 0 : DB_MPOOL_DIRTY, &meta)) != 0)
 		goto err;
 
 	if (!F_ISSET(dbp, DB_AM_RDONLY))
@@ -174,8 +167,7 @@ meta_only:
 	sp->qs_cur_recno = meta->cur_recno;
 
 	/* Discard the meta-data page. */
-	ret = __memp_fput(mpf,
-	    meta, F_ISSET(dbp, DB_AM_RDONLY) ? 0 : DB_MPOOL_DIRTY);
+	ret = __memp_fput(mpf, meta, 0);
 	if ((t_ret = __LPUT(dbc, lock)) != 0 && ret == 0)
 		ret = t_ret;
 	if (ret != 0)
@@ -213,7 +205,7 @@ __qam_stat_print(dbc, flags)
 	dbp = dbc->dbp;
 	dbenv = dbp->dbenv;
 
-	if ((ret = __qam_stat(dbc, &sp, 0)) != 0)
+	if ((ret = __qam_stat(dbc, &sp, LF_ISSET(DB_FAST_STAT))) != 0)
 		return (ret);
 
 	if (LF_ISSET(DB_STAT_ALL)) {
diff --git a/db/qam/qam_stub.c b/db/qam/qam_stub.c
index 1c22aaa52..7b02c8ef1 100644
--- a/db/qam/qam_stub.c
+++ b/db/qam/qam_stub.c
@@ -1,18 +1,14 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1996-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_stub.c,v 1.12 2004/06/14 15:23:33 bostic Exp $
+ * $Id: qam_stub.c,v 12.5 2006/08/24 14:46:24 bostic Exp $
  */
 
-#include "db_config.h"
-
 #ifndef	HAVE_QUEUE
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-#endif
+#include "db_config.h"
 
 #include "db_int.h"
 #include "dbinc/db_page.h"
@@ -34,7 +30,7 @@ int
 __db_no_queue_am(dbenv)
 	DB_ENV *dbenv;
 {
-	__db_err(dbenv,
+	__db_errx(dbenv,
     "library build did not include support for the Queue access method");
 	return (DB_OPNOTSUP);
 }
diff --git a/db/qam/qam_upgrade.c b/db/qam/qam_upgrade.c
index dff82404a..0c5777e25 100644
--- a/db/qam/qam_upgrade.c
+++ b/db/qam/qam_upgrade.c
@@ -1,20 +1,14 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1996-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_upgrade.c,v 11.16 2004/05/10 21:29:43 bostic Exp $
+ * $Id: qam_upgrade.c,v 12.4 2006/08/24 14:46:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#include <string.h>
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_upgrade.h"
 #include "dbinc/db_page.h"
diff --git a/db/qam/qam_verify.c b/db/qam/qam_verify.c
index 571157b8f..b6457ce70 100644
--- a/db/qam/qam_verify.c
+++ b/db/qam/qam_verify.c
@@ -1,29 +1,20 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1999-2004
- *	Sleepycat Software.  All rights reserved.
+ * Copyright (c) 1999-2006
+ *	Oracle Corporation.  All rights reserved.
  *
- * $Id: qam_verify.c,v 1.51 2004/10/11 18:47:51 bostic Exp $
+ * $Id: qam_verify.c,v 12.9 2006/08/24 14:46:24 bostic Exp $
  */
 
 #include "db_config.h"
 
-#ifndef NO_SYSTEM_INCLUDES
-#include <sys/types.h>
-
-#endif
-
 #include "db_int.h"
 #include "dbinc/db_page.h"
 #include "dbinc/db_verify.h"
 #include "dbinc/db_am.h"
-#include "dbinc/db_shash.h"
 #include "dbinc/mp.h"
 #include "dbinc/qam.h"
-#include <stdlib.h>
-#include <string.h>
-
 /*
  * __qam_vrfy_meta --
  *	Verify the queue-specific part of a metadata page.
@@ -93,6 +84,8 @@ __qam_vrfy_meta(dbp, vdp, meta, pgno, flags)
 		 * it when handling extents.  It would get set up in open,
 		 * if we called open normally, but we don't.
 		 */
+		vdp->re_pad = meta->re_pad;
+		qp->re_pad = (int)meta->re_pad;
 		qp->re_len = vdp->re_len = meta->re_len;
 		qp->rec_page = vdp->rec_page = meta->rec_page;
 		qp->page_ext = vdp->page_ext = meta->page_ext;
@@ -165,7 +158,7 @@ __qam_vrfy_meta(dbp, vdp, meta, pgno, flags)
 		}
 	}
 	if (nextents > 0)
-		__db_err(dbenv,
+		__db_errx(dbenv,
 		     "Warning: %d extra extent files found", nextents);
 	vdp->nextents = nextents;
 	vdp->extents = extents;
@@ -344,7 +337,7 @@ begin:	for (; i <= stop; i++) {
 		 */
 		if (LF_ISSET(DB_SALVAGE) && (__db_salvage_isdone(vdp, i) != 0))
 			continue;
-		if ((t_ret = __qam_fget(dbp, &i, 0, &h)) != 0) {
+		if ((t_ret = __qam_fget(dbp, &i, NULL, 0, &h)) != 0) {
 			if (t_ret == ENOENT || t_ret == DB_PAGE_NOTFOUND) {
 				i += (pg_ext - ((i - 1) % pg_ext)) - 1;
 				continue;