Update internal BDB to version 4.6.18.

12 files changed, 533 insertions, 295 deletions

diff --git a/db/txn/txn.c b/db/txn/txn.c
index 527c483d1..2154f0962 100644
--- a/db/txn/txn.c
+++ b/db/txn/txn.c
@@ -1,8 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 1996,2007 Oracle.  All rights reserved.
  */
 /*
  * Copyright (c) 1995, 1996
@@ -35,7 +34,7 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- * $Id: txn.c,v 12.55 2006/08/24 14:46:52 bostic Exp $
+ * $Id: txn.c,v 12.78 2007/06/29 00:25:02 margo Exp $
  */
 
 #include "db_config.h"
@@ -50,21 +49,10 @@
 #include "dbinc/mp.h"
 #include "dbinc/txn.h"
 
-#define	SET_LOG_FLAGS(dbenv, txn, lflags)				\
-	do {								\
-		lflags = DB_LOG_COMMIT;					\
-		if (F_ISSET(txn, TXN_SYNC))				\
-			lflags |= DB_FLUSH;				\
-		else if (F_ISSET(txn, TXN_WRITE_NOSYNC))		\
-			lflags |= DB_LOG_WRNOSYNC;			\
-		else if (!F_ISSET(txn, TXN_NOSYNC) &&			\
-		    !F_ISSET(dbenv, DB_ENV_TXN_NOSYNC)) {		\
-			if (F_ISSET(dbenv, DB_ENV_TXN_WRITE_NOSYNC))	\
-				lflags |= DB_LOG_WRNOSYNC;		\
-			else						\
-				lflags |= DB_FLUSH;			\
-		}							\
-	} while (0)
+#define	LOG_FLAGS(txn)						\
+		(DB_LOG_COMMIT | (F_ISSET(txn, TXN_SYNC) ?	\
+		DB_FLUSH : (F_ISSET(txn, TXN_WRITE_NOSYNC) ?	\
+		DB_LOG_WRNOSYNC : 0)))
 
 /*
  * __txn_isvalid enumerated types.  We cannot simply use the transaction
@@ -79,11 +67,11 @@ typedef enum {
 } txnop_t;
 
 static int  __txn_abort_pp __P((DB_TXN *));
-static int  __txn_begin_int __P((DB_TXN *, int));
+static int  __txn_begin_int __P((DB_TXN *));
 static int  __txn_commit_pp __P((DB_TXN *, u_int32_t));
 static int  __txn_discard __P((DB_TXN *, u_int32_t));
 static int  __txn_dispatch_undo
-		__P((DB_ENV *, DB_TXN *, DBT *, DB_LSN *, void *));
+		__P((DB_ENV *, DB_TXN *, DBT *, DB_LSN *, DB_TXNHEAD *));
 static int  __txn_end __P((DB_TXN *, int));
 static int  __txn_isvalid __P((const DB_TXN *, txnop_t));
 static int  __txn_undo __P((DB_TXN *));
@@ -109,9 +97,9 @@ __txn_begin_pp(dbenv, parent, txnpp, flags)
 
 	if ((ret = __db_fchk(dbenv,
 	    "txn_begin", flags,
-	    DB_READ_COMMITTED | DB_READ_UNCOMMITTED | DB_TXN_NOWAIT |
+	    DB_READ_COMMITTED | DB_READ_UNCOMMITTED |
 	    DB_TXN_NOSYNC | DB_TXN_SNAPSHOT | DB_TXN_SYNC |
-	    DB_TXN_WRITE_NOSYNC)) != 0)
+	    DB_TXN_WAIT | DB_TXN_WRITE_NOSYNC | DB_TXN_NOWAIT)) != 0)
 		return (ret);
 	if ((ret = __db_fcchk(dbenv, "txn_begin", flags,
 	    DB_TXN_WRITE_NOSYNC | DB_TXN_NOSYNC, DB_TXN_SYNC)) != 0)
@@ -182,23 +170,36 @@ __txn_begin(dbenv, parent, txnpp, flags)
 	TAILQ_INIT(&txn->events);
 	STAILQ_INIT(&txn->logs);
 	txn->flags = TXN_MALLOC;
+
+	/*
+	 * Set the sync mode for commit.  Any local bits override those
+	 * in the environment.  SYNC is the default.
+	 */
+	if (LF_ISSET(DB_TXN_SYNC))
+		F_SET(txn, TXN_SYNC);
+	else if (LF_ISSET(DB_TXN_NOSYNC))
+		F_SET(txn, TXN_NOSYNC);
+	else if (LF_ISSET(DB_TXN_WRITE_NOSYNC))
+		F_SET(txn, TXN_WRITE_NOSYNC);
+	else if (F_ISSET(dbenv, DB_ENV_TXN_NOSYNC))
+		F_SET(txn, TXN_NOSYNC);
+	else if (F_ISSET(dbenv, DB_ENV_TXN_WRITE_NOSYNC))
+		F_SET(txn, TXN_WRITE_NOSYNC);
+	else
+		F_SET(txn, TXN_SYNC);
+
+	if (LF_ISSET(DB_TXN_NOWAIT) ||
+	    (F_ISSET(dbenv, DB_ENV_TXN_NOWAIT) && !LF_ISSET(DB_TXN_WAIT)))
+		F_SET(txn, TXN_NOWAIT);
 	if (LF_ISSET(DB_READ_COMMITTED))
 		F_SET(txn, TXN_READ_COMMITTED);
 	if (LF_ISSET(DB_READ_UNCOMMITTED))
 		F_SET(txn, TXN_READ_UNCOMMITTED);
-	if (LF_ISSET(DB_TXN_NOSYNC))
-		F_SET(txn, TXN_NOSYNC);
 	if (LF_ISSET(DB_TXN_SNAPSHOT) || F_ISSET(dbenv, DB_ENV_TXN_SNAPSHOT) ||
 	    (parent != NULL && F_ISSET(parent, TXN_SNAPSHOT)))
 		F_SET(txn, TXN_SNAPSHOT);
-	if (LF_ISSET(DB_TXN_SYNC))
-		F_SET(txn, TXN_SYNC);
-	if (LF_ISSET(DB_TXN_NOWAIT))
-		F_SET(txn, TXN_NOWAIT);
-	if (LF_ISSET(DB_TXN_WRITE_NOSYNC))
-		F_SET(txn, TXN_WRITE_NOSYNC);
 
-	if ((ret = __txn_begin_int(txn, 0)) != 0)
+	if ((ret = __txn_begin_int(txn)) != 0)
 		goto err;
 	td = txn->td;
 
@@ -212,7 +213,7 @@ __txn_begin(dbenv, parent, txnpp, flags)
 		region = dbenv->lk_handle->reginfo.primary;
 		if (parent != NULL) {
 			ret = __lock_inherit_timeout(dbenv,
-			    parent->txnid, txn->txnid);
+			    parent->locker, txn->locker);
 			/* No parent locker set yet. */
 			if (ret == EINVAL) {
 				parent = NULL;
@@ -227,7 +228,7 @@ __txn_begin(dbenv, parent, txnpp, flags)
 		 * or it has no timeouts set.
 		 */
 		if (parent == NULL && region->tx_timeout != 0)
-			if ((ret = __lock_set_timeout(dbenv, txn->txnid,
+			if ((ret = __lock_set_timeout(dbenv, txn->locker,
 			    region->tx_timeout, DB_SET_TXN_TIMEOUT)) != 0)
 				goto err;
 	}
@@ -267,7 +268,7 @@ __txn_xa_begin(dbenv, txn)
 	memset(&txn->lock_timeout, 0, sizeof(db_timeout_t));
 	memset(&txn->expire, 0, sizeof(db_timeout_t));
 
-	return (__txn_begin_int(txn, 0));
+	return (__txn_begin_int(txn));
 }
 
 /*
@@ -342,7 +343,7 @@ __txn_compensate_begin(dbenv, txnpp)
 	txn->flags = TXN_COMPENSATE | TXN_MALLOC;
 
 	*txnpp = txn;
-	return (__txn_begin_int(txn, 1));
+	return (__txn_begin_int(txn));
 }
 
 /*
@@ -350,9 +351,8 @@ __txn_compensate_begin(dbenv, txnpp)
  *	Normal DB version of txn_begin.
  */
 static int
-__txn_begin_int(txn, internal)
+__txn_begin_int(txn)
 	DB_TXN *txn;
-	int internal;
 {
 	DB_ENV *dbenv;
 	DB_TXNMGR *mgr;
@@ -372,14 +372,6 @@ __txn_begin_int(txn, internal)
 		goto err;
 	}
 
-	/* Make sure that we aren't still recovering prepared transactions. */
-	if (!internal && region->stat.st_nrestores != 0) {
-		__db_errx(dbenv,
-    "recovery of prepared but not yet committed transactions is incomplete");
-		ret = EINVAL;
-		goto err;
-	}
-
 	/*
 	 * Allocate a new transaction id. Our current valid range can span
 	 * the maximum valid value, so check for it and wrap manually.
@@ -394,7 +386,7 @@ __txn_begin_int(txn, internal)
 
 	/* Allocate a new transaction detail structure. */
 	if ((ret =
-	    __db_shalloc(&mgr->reginfo, sizeof(TXN_DETAIL), 0, &td)) != 0) {
+	    __env_alloc(&mgr->reginfo, sizeof(TXN_DETAIL), &td)) != 0) {
 		__db_errx(dbenv,
 		    "Unable to allocate memory for transaction detail");
 		goto err;
@@ -404,12 +396,21 @@ __txn_begin_int(txn, internal)
 	SH_TAILQ_INSERT_HEAD(&region->active_txn, td, links, __txn_detail);
 
 	id = ++region->last_txnid;
+
+#ifdef HAVE_STATISTICS
 	++region->stat.st_nbegins;
 	if (++region->stat.st_nactive > region->stat.st_maxnactive)
 		region->stat.st_maxnactive = region->stat.st_nactive;
+#endif
 
 	td->txnid = id;
 	dbenv->thread_id(dbenv, &td->pid, &td->tid);
+
+	/* allocate a locker for this txn */
+	if (LOCKING_ON(dbenv) && (ret =
+	    __lock_getlocker(dbenv->lk_handle, id, 1, &txn->locker)) != 0)
+		goto err;
+
 	ZERO_LSN(td->last_lsn);
 	ZERO_LSN(td->begin_lsn);
 	SH_TAILQ_INIT(&td->kids);
@@ -425,6 +426,9 @@ __txn_begin_int(txn, internal)
 	td->status = TXN_RUNNING;
 	td->flags = 0;
 	td->xa_status = 0;
+	td->nlog_dbs = 0;
+	td->nlog_slots = TXN_NSLOTS;
+	td->log_dbs = R_OFFSET(&mgr->reginfo, td->slots);
 
 	TXN_SYSTEM_UNLOCK(dbenv);
 
@@ -467,14 +471,18 @@ err:	TXN_SYSTEM_UNLOCK(dbenv);
  *	Fill in the fields of the local transaction structure given
  *	the detail transaction structure.
  *
- * PUBLIC: void __txn_continue __P((DB_ENV *, DB_TXN *, TXN_DETAIL *));
+ * PUBLIC: int __txn_continue __P((DB_ENV *, DB_TXN *, TXN_DETAIL *));
  */
-void
+int
 __txn_continue(env, txn, td)
 	DB_ENV *env;
 	DB_TXN *txn;
 	TXN_DETAIL *td;
 {
+	int ret;
+
+	ret = 0;
+
 	txn->mgrp = env->tx_handle;
 	txn->parent = NULL;
 	txn->txnid = td->txnid;
@@ -489,8 +497,19 @@ __txn_continue(env, txn, td)
 	txn->set_name = __txn_set_name;
 
 	txn->flags = 0;
+	/*
+	 * If this is a restored transaction, we need to propagate that fact
+	 * to the process-local structure.  However, if it's not a restored
+	 * transaction, then we're running in XA and we need to make sure
+	 * that we have a locker associated with this transaction.
+	 */
 	if (F_ISSET(td, TXN_DTL_RESTORED))
 		F_SET(txn, TXN_RESTORED);
+	else
+		ret = __lock_getlocker(env->lk_handle,
+		    txn->txnid, 0, &txn->locker);
+
+	return (ret);
 }
 
 /*
@@ -537,7 +556,7 @@ __txn_commit(txn, flags)
 	REGENV *renv;
 	REGINFO *infop;
 	TXN_DETAIL *td;
-	u_int32_t id, lflags;
+	u_int32_t id;
 	int ret, t_ret;
 
 	dbenv = txn->mgrp->dbenv;
@@ -550,7 +569,7 @@ __txn_commit(txn, flags)
 	 * return.  If the transaction deadlocked, they want abort, not commit.
 	 */
 	if (F_ISSET(txn, TXN_DEADLOCK)) {
-		ret = __db_txn_deadlock_err(dbenv);
+		ret = __db_txn_deadlock_err(dbenv, txn);
 		goto err;
 	}
 
@@ -633,13 +652,13 @@ __txn_commit(txn, flags)
 					request.obj = &list_dbt;
 				}
 				ret = __lock_vec(dbenv,
-				    txn->txnid, 0, &request, 1, NULL);
+				    txn->locker, 0, &request, 1, NULL);
 			}
 
 			if (ret == 0 && !IS_ZERO_LSN(td->last_lsn)) {
-				SET_LOG_FLAGS(dbenv, txn, lflags);
 				ret = __txn_regop_log(dbenv, txn,
-				    &td->visible_lsn, lflags, TXN_COMMIT,
+				    &td->visible_lsn, LOG_FLAGS(txn),
+				    TXN_COMMIT,
 				    (int32_t)time(NULL), id, request.obj);
 				if (ret == 0)
 					td->last_lsn = td->visible_lsn;
@@ -762,7 +781,7 @@ __txn_abort(txn)
 	REGENV *renv;
 	REGINFO *infop;
 	TXN_DETAIL *td;
-	u_int32_t id, lflags;
+	u_int32_t id;
 	int ret;
 
 	dbenv = txn->mgrp->dbenv;
@@ -806,6 +825,11 @@ __txn_abort(txn)
 	}
 
 	if (LOCKING_ON(dbenv)) {
+		/* Allocate a locker for this restored txn if necessary. */
+		if (txn->locker == NULL &&
+		    (ret = __lock_getlocker(dbenv->lk_handle,
+		    txn->txnid, 1, &txn->locker)) != 0)
+			return (__db_panic(dbenv, ret));
 		/*
 		 * We are about to free all the read locks for this transaction
 		 * below.  Some of those locks might be handle locks which
@@ -818,17 +842,17 @@ __txn_abort(txn)
 
 		/* Turn off timeouts. */
 		if ((ret = __lock_set_timeout(dbenv,
-		    txn->txnid, 0, DB_SET_TXN_TIMEOUT)) != 0)
+		    txn->locker, 0, DB_SET_TXN_TIMEOUT)) != 0)
 			return (__db_panic(dbenv, ret));
 
 		if ((ret = __lock_set_timeout(dbenv,
-		    txn->txnid, 0, DB_SET_LOCK_TIMEOUT)) != 0)
+		    txn->locker, 0, DB_SET_LOCK_TIMEOUT)) != 0)
 			return (__db_panic(dbenv, ret));
 
 		request.op = DB_LOCK_UPGRADE_WRITE;
 		request.obj = NULL;
 		if ((ret = __lock_vec(
-		    dbenv, txn->txnid, DB_LOCK_ABORT, &request, 1, NULL)) != 0)
+		    dbenv, txn->locker, DB_LOCK_ABORT, &request, 1, NULL)) != 0)
 			return (__db_panic(dbenv, ret));
 	}
 undo:	if ((ret = __txn_undo(txn)) != 0)
@@ -840,10 +864,9 @@ undo:	if ((ret = __txn_undo(txn)) != 0)
 	 * then we log the abort so we know that this transaction
 	 * was actually completed.
 	 */
-done:	SET_LOG_FLAGS(dbenv, txn, lflags);
-	if (DBENV_LOGGING(dbenv) && td->status == TXN_PREPARED &&
+done:	 if (DBENV_LOGGING(dbenv) && td->status == TXN_PREPARED &&
 	    (ret = __txn_regop_log(dbenv, txn, &td->last_lsn,
-	    lflags, TXN_ABORT, (int32_t)time(NULL), id, NULL)) != 0)
+	    LOG_FLAGS(txn), TXN_ABORT, (int32_t)time(NULL), id, NULL)) != 0)
 		return (__db_panic(dbenv, ret));
 
 	/* __txn_end always panics if it errors, so pass the return along. */
@@ -861,12 +884,15 @@ __txn_discard(txn, flags)
 {
 	DB_ENV *dbenv;
 	DB_THREAD_INFO *ip;
-	int ret;
+	int ret, t_ret;
 
 	dbenv = txn->mgrp->dbenv;
 
 	ENV_ENTER(dbenv, ip);
 	ret = __txn_discard_int(txn, flags);
+	if (IS_ENV_REPLICATED(dbenv) &&
+	    (t_ret = __op_rep_exit(dbenv)) != 0 && ret == 0)
+		ret = t_ret;
 	ENV_LEAVE(dbenv, ip);
 	return (ret);
 }
@@ -943,7 +969,7 @@ __txn_prepare(txn, gid)
 	if ((ret = __txn_isvalid(txn, TXN_OP_PREPARE)) != 0)
 		return (ret);
 	if (F_ISSET(txn, TXN_DEADLOCK))
-		return (__db_txn_deadlock_err(dbenv));
+		return (__db_txn_deadlock_err(dbenv, txn));
 
 	ENV_ENTER(dbenv, ip);
 
@@ -971,13 +997,12 @@ __txn_prepare(txn, gid)
 	memset(&request, 0, sizeof(request));
 	if (LOCKING_ON(dbenv)) {
 		request.op = DB_LOCK_PUT_READ;
-		if (IS_REP_MASTER(dbenv) &&
-		    !IS_ZERO_LSN(td->last_lsn)) {
+		if (!IS_ZERO_LSN(td->last_lsn)) {
 			memset(&list_dbt, 0, sizeof(list_dbt));
 			request.obj = &list_dbt;
 		}
 		if ((ret = __lock_vec(dbenv,
-		    txn->txnid, 0, &request, 1, NULL)) != 0)
+		    txn->locker, 0, &request, 1, NULL)) != 0)
 			goto err;
 
 	}
@@ -1072,11 +1097,11 @@ __txn_set_name(txn, name)
 	ENV_ENTER(dbenv, ip);
 	TXN_SYSTEM_LOCK(dbenv);
 	if (td->name != INVALID_ROFF) {
-		__db_shalloc_free(
+		__env_alloc_free(
 		    &mgr->reginfo, R_ADDR(&mgr->reginfo, td->name));
 		td->name = INVALID_ROFF;
 	}
-	if ((ret = __db_shalloc(&mgr->reginfo, len, 0, &p)) != 0) {
+	if ((ret = __env_alloc(&mgr->reginfo, len, &p)) != 0) {
 		TXN_SYSTEM_UNLOCK(dbenv);
 		__db_errx(dbenv,
 		    "Unable to allocate memory for transaction name");
@@ -1116,15 +1141,17 @@ __txn_set_timeout(txn, timeout, op)
 	db_timeout_t timeout;
 	u_int32_t op;
 {
+	DB_ENV *dbenv;
 	DB_THREAD_INFO *ip;
 	int ret;
 
-	if (op != DB_SET_TXN_TIMEOUT &&  op != DB_SET_LOCK_TIMEOUT)
-		return (__db_ferr(txn->mgrp->dbenv, "DB_TXN->set_timeout", 0));
+	dbenv = txn->mgrp->dbenv;
 
-	ENV_ENTER(txn->mgrp->dbenv, ip);
-	ret = __lock_set_timeout(
-	    txn->mgrp->dbenv, txn->txnid, timeout, op);
+	if (op != DB_SET_TXN_TIMEOUT && op != DB_SET_LOCK_TIMEOUT)
+		return (__db_ferr(dbenv, "DB_TXN->set_timeout", 0));
+
+	ENV_ENTER(dbenv, ip);
+	ret = __lock_set_timeout( dbenv, txn->locker, timeout, op);
 	ENV_LEAVE(txn->mgrp->dbenv, ip);
 	return (ret);
 }
@@ -1276,18 +1303,26 @@ __txn_end(txn, is_commit)
 	 * so DB_LOCK_DEADLOCK is just as fatal as any other error.
 	 */
 	if (LOCKING_ON(dbenv)) {
+		/* Allocate a locker for this restored txn if necessary. */
+		if (txn->locker == NULL &&
+		    (ret = __lock_getlocker(dbenv->lk_handle,
+		    txn->txnid, 1, &txn->locker)) != 0)
+			return (__db_panic(dbenv, ret));
 		request.op = txn->parent == NULL ||
 		    is_commit == 0 ? DB_LOCK_PUT_ALL : DB_LOCK_INHERIT;
 		request.obj = NULL;
 		if ((ret = __lock_vec(dbenv,
-		    txn->txnid, 0, &request, 1, NULL)) != 0)
+		    txn->locker, 0, &request, 1, NULL)) != 0)
 			return (__db_panic(dbenv, ret));
 	}
 
 	/* End the transaction. */
+	td = txn->td;
+	if (td->nlog_dbs != 0 && (ret = __txn_dref_fname(dbenv, txn)) != 0)
+		return (__db_panic(dbenv, ret));
+
 	TXN_SYSTEM_LOCK(dbenv);
 
-	td = txn->td;
 	td->status = is_commit ? TXN_COMMITTED : TXN_ABORTED;
 	SH_TAILQ_REMOVE(&region->active_txn, td, links, __txn_detail);
 	if (F_ISSET(td, TXN_DTL_RESTORED)) {
@@ -1296,7 +1331,7 @@ __txn_end(txn, is_commit)
 	}
 
 	if (td->name != INVALID_ROFF) {
-		__db_shalloc_free(
+		__env_alloc_free(
 		    &mgr->reginfo, R_ADDR(&mgr->reginfo, td->name));
 		td->name = INVALID_ROFF;
 	}
@@ -1311,10 +1346,12 @@ __txn_end(txn, is_commit)
 		if (td->mvcc_ref != 0) {
 			SH_TAILQ_INSERT_HEAD(&region->mvcc_txn,
 			    td, links, __txn_detail);
+#ifdef HAVE_STATISTICS
 			if (++region->stat.st_nsnapshot >
 			    region->stat.st_maxnsnapshot)
 				region->stat.st_maxnsnapshot =
 				    region->stat.st_nsnapshot;
+#endif
 			td = NULL;
 		}
 		MUTEX_UNLOCK(dbenv, mvcc_mtx);
@@ -1323,14 +1360,20 @@ __txn_end(txn, is_commit)
 				return (__db_panic(dbenv, ret));
 	}
 
-	if (td != NULL)
-		__db_shalloc_free(&mgr->reginfo, td);
+	if (td != NULL) {
+		if (td->nlog_slots != TXN_NSLOTS)
+			__env_alloc_free(&mgr->reginfo,
+			    R_ADDR(&mgr->reginfo, td->log_dbs));
+		__env_alloc_free(&mgr->reginfo, td);
+	}
 
+#ifdef HAVE_STATISTICS
 	if (is_commit)
 		region->stat.st_ncommits++;
 	else
 		region->stat.st_naborts++;
 	--region->stat.st_nactive;
+#endif
 
 	TXN_SYSTEM_UNLOCK(dbenv);
 
@@ -1339,7 +1382,7 @@ __txn_end(txn, is_commit)
 	 * if any.
 	 */
 	if (LOCKING_ON(dbenv) && (ret =
-	    __lock_freefamilylocker(dbenv->lk_handle, txn->txnid)) != 0)
+	    __lock_freefamilylocker(dbenv->lk_handle, txn->locker)) != 0)
 		return (__db_panic(dbenv, ret));
 	if (txn->parent != NULL)
 		TAILQ_REMOVE(&txn->parent->kids, txn, klinks);
@@ -1362,11 +1405,19 @@ __txn_end(txn, is_commit)
 	}
 
 	if (do_closefiles) {
-		F_SET(dbenv->lg_handle, DBLOG_RECOVER);
-		(void)__dbreg_close_files(dbenv);
-		F_CLR(dbenv->lg_handle, DBLOG_RECOVER);
+		/*
+		 * Otherwise, we have resolved the last outstanding prepared
+		 * txn and need to invalidate the fileids that were left
+		 * open for those txns and then close them.
+		 */
+		(void)__dbreg_invalidate_files(dbenv, 1);
+		(void)__dbreg_close_files(dbenv, 1);
+		if (IS_REP_MASTER(dbenv))
+			F_CLR(dbenv->rep_handle, DBREP_OPENFILES);
+		F_CLR(dbenv->lg_handle, DBLOG_OPENFILES);
 		mgr->n_discards = 0;
-		(void)__txn_checkpoint(dbenv, 0, 0, DB_FORCE);
+		(void)__txn_checkpoint(dbenv, 0, 0,
+		    DB_CKP_INTERNAL | DB_FORCE);
 	}
 
 	return (0);
@@ -1378,10 +1429,11 @@ __txn_dispatch_undo(dbenv, txn, rdbt, key_lsn, txnlist)
 	DB_TXN *txn;
 	DBT *rdbt;
 	DB_LSN *key_lsn;
-	void *txnlist;
+	DB_TXNHEAD *txnlist;
 {
 	int ret;
 
+	txnlist->td = txn->td;
 	ret = __db_dispatch(dbenv, dbenv->recover_dtab,
 	    dbenv->recover_dtab_size, rdbt, key_lsn, DB_TXN_ABORT, txnlist);
 	if (ret == DB_SURPRISE_KID) {
@@ -1418,7 +1470,7 @@ __txn_undo(txn)
 	txnlist = NULL;
 	ret = 0;
 
-	if (!DBENV_LOGGING(dbenv))
+	if (!LOGGING_ON(dbenv))
 		return (0);
 
 	/*
@@ -1474,7 +1526,7 @@ __txn_undo(txn)
 		 * The dispatch routine returns the lsn of the record
 		 * before the current one in the key_lsn argument.
 		 */
-		if ((ret = __log_c_get(logc, &key_lsn, &rdbt, DB_SET)) == 0) {
+		if ((ret = __logc_get(logc, &key_lsn, &rdbt, DB_SET)) == 0) {
 			ret = __txn_dispatch_undo(dbenv,
 			    txn, &rdbt, &key_lsn, txnlist);
 		}
@@ -1491,7 +1543,7 @@ __txn_undo(txn)
 	ret = __db_do_the_limbo(dbenv, ptxn, txn, txnlist, LIMBO_NORMAL);
 #endif
 
-err:	if (logc != NULL && (t_ret = __log_c_close(logc)) != 0 && ret == 0)
+err:	if (logc != NULL && (t_ret = __logc_close(logc)) != 0 && ret == 0)
 		ret = t_ret;
 
 	if (ptxn == NULL && txnlist != NULL)
@@ -1586,10 +1638,10 @@ __txn_force_abort(dbenv, buffer)
 }
 
 /*
- * __txn_preclose
- *	Before we can close an environment, we need to check if we
- * were in the midst of taking care of restored transactions.  If
- * so, then we need to close the files that we opened.
+ * __txn_preclose --
+ *	Before we can close an environment, we need to check if we were in the
+ *	middle of taking care of restored transactions.  If so, close the files
+ *	we opened.
  *
  * PUBLIC: int __txn_preclose __P((DB_ENV *));
  */
@@ -1614,12 +1666,12 @@ __txn_preclose(dbenv)
 
 	if (do_closefiles) {
 		/*
-		 * Set the DBLOG_RECOVER flag while closing these
-		 * files so they do not create additional log records
-		 * that will confuse future recoveries.
+		 * Set the DBLOG_RECOVER flag while closing these files so they
+		 * do not create additional log records that will confuse future
+		 * recoveries.
 		 */
 		F_SET(dbenv->lg_handle, DBLOG_RECOVER);
-		ret = __dbreg_close_files(dbenv);
+		ret = __dbreg_close_files(dbenv, 0);
 		F_CLR(dbenv->lg_handle, DBLOG_RECOVER);
 	} else
 		ret = 0;
diff --git a/db/txn/txn.src b/db/txn/txn.src
index 0cd2bbcaa..ada4862d5 100644
--- a/db/txn/txn.src
+++ b/db/txn/txn.src
@@ -1,10 +1,9 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 1996,2007 Oracle.  All rights reserved.
  *
- * $Id: txn.src,v 12.7 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn.src,v 12.11 2007/06/01 17:35:16 sue Exp $
  */
 
 PREFIX	__txn
@@ -29,15 +28,15 @@ INCLUDE
  *	Environment ID of this operation (4.4+).
  */
 BEGIN_COMPAT regop		42	10
-ARG	opcode		u_int32_t	ld
+ARG	opcode		u_int32_t	lu
 TIME	timestamp	int32_t		ld
 LOCKS	locks		DBT		s
 END
 
 BEGIN regop		44	10
-ARG	opcode		u_int32_t	ld
+ARG	opcode		u_int32_t	lu
 TIME	timestamp	int32_t		ld
-ARG	envid		u_int32_t	ld
+ARG	envid		u_int32_t	lu
 LOCKS	locks		DBT		s
 END
 
@@ -57,21 +56,22 @@ END
  * envid:
  *	Environment ID of this checkpoint (4.3+).
  * rep_gen:
- *	Persistent replication generation number.
+ *	Persistent replication generation number (4.2-4.5 only).
+ *	Renamed to 'spare' in 4.6.
  */
 BEGIN_COMPAT ckp		42	11
 POINTER	ckp_lsn		DB_LSN *	lu
 POINTER	last_ckp	DB_LSN *	lu
 TIME	timestamp	int32_t		ld
-ARG	rep_gen		u_int32_t	ld
+ARG	rep_gen		u_int32_t	lu
 END
 
 BEGIN ckp		43	11
 POINTER	ckp_lsn		DB_LSN *	lu
 POINTER	last_ckp	DB_LSN *	lu
 TIME	timestamp	int32_t		ld
-ARG	envid		u_int32_t	ld
-ARG	rep_gen		u_int32_t	ld
+ARG	envid		u_int32_t	lu
+ARG	spare		u_int32_t	lu
 END
 
 /*
@@ -93,8 +93,8 @@ BEGIN xa_regop		42	13
 ARG	opcode		u_int32_t	lu
 DBT	xid		DBT		s
 ARG	formatID	int32_t		ld
-ARG	gtrid		u_int32_t	u
-ARG	bqual		u_int32_t	u
+ARG	gtrid		u_int32_t	lu
+ARG	bqual		u_int32_t	lu
 POINTER	begin_lsn	DB_LSN *	lu
 LOCKS	locks		DBT		s
 END
@@ -103,6 +103,6 @@ END
  * Log the fact that we are recycling txnids.
  */
 BEGIN recycle		42	14
-ARG	min		u_int32_t	u
-ARG	max		u_int32_t	u
+ARG	min		u_int32_t	lu
+ARG	max		u_int32_t	lu
 END
diff --git a/db/txn/txn_auto.c b/db/txn/txn_auto.c
index 3a8cda986..d72869d5b 100644
--- a/db/txn/txn_auto.c
+++ b/db/txn/txn_auto.c
@@ -196,6 +196,7 @@ __txn_regop_log(dbenv, txnp, ret_lsnp, flags,
 				 *ret_lsnp = *rlsnp;
 		}
 	} else {
+		ret = 0;
 #ifdef DIAGNOSTIC
 		/*
 		 * Set the debug bit if we are going to log non-durable
@@ -205,10 +206,9 @@ __txn_regop_log(dbenv, txnp, ret_lsnp, flags,
 		rectype |= DB_debug_FLAG;
 		memcpy(logrec.data, &rectype, sizeof(rectype));
 
-		ret = __log_put(dbenv,
-		    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
-#else
-		ret = 0;
+		if (!IS_REP_CLIENT(dbenv))
+			ret = __log_put(dbenv,
+			    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
 #endif
 		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
 		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
@@ -336,7 +336,7 @@ __txn_ckp_42_read(dbenv, recbuf, argpp)
  */
 int
 __txn_ckp_log(dbenv, txnp, ret_lsnp, flags,
-    ckp_lsn, last_ckp, timestamp, envid, rep_gen)
+    ckp_lsn, last_ckp, timestamp, envid, spare)
 	DB_ENV *dbenv;
 	DB_TXN *txnp;
 	DB_LSN *ret_lsnp;
@@ -345,7 +345,7 @@ __txn_ckp_log(dbenv, txnp, ret_lsnp, flags,
 	DB_LSN * last_ckp;
 	int32_t timestamp;
 	u_int32_t envid;
-	u_int32_t rep_gen;
+	u_int32_t spare;
 {
 	DBT logrec;
 	DB_TXNLOGREC *lr;
@@ -454,7 +454,7 @@ __txn_ckp_log(dbenv, txnp, ret_lsnp, flags,
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
-	uinttmp = (u_int32_t)rep_gen;
+	uinttmp = (u_int32_t)spare;
 	memcpy(bp, &uinttmp, sizeof(uinttmp));
 	bp += sizeof(uinttmp);
 
@@ -469,6 +469,7 @@ __txn_ckp_log(dbenv, txnp, ret_lsnp, flags,
 				 *ret_lsnp = *rlsnp;
 		}
 	} else {
+		ret = 0;
 #ifdef DIAGNOSTIC
 		/*
 		 * Set the debug bit if we are going to log non-durable
@@ -478,10 +479,9 @@ __txn_ckp_log(dbenv, txnp, ret_lsnp, flags,
 		rectype |= DB_debug_FLAG;
 		memcpy(logrec.data, &rectype, sizeof(rectype));
 
-		ret = __log_put(dbenv,
-		    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
-#else
-		ret = 0;
+		if (!IS_REP_CLIENT(dbenv))
+			ret = __log_put(dbenv,
+			    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
 #endif
 		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
 		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
@@ -548,7 +548,7 @@ __txn_ckp_read(dbenv, recbuf, argpp)
 	bp += sizeof(uinttmp);
 
 	memcpy(&uinttmp, bp, sizeof(uinttmp));
-	argp->rep_gen = (u_int32_t)uinttmp;
+	argp->spare = (u_int32_t)uinttmp;
 	bp += sizeof(uinttmp);
 
 	*argpp = argp;
@@ -674,6 +674,7 @@ __txn_child_log(dbenv, txnp, ret_lsnp, flags,
 				 *ret_lsnp = *rlsnp;
 		}
 	} else {
+		ret = 0;
 #ifdef DIAGNOSTIC
 		/*
 		 * Set the debug bit if we are going to log non-durable
@@ -683,10 +684,9 @@ __txn_child_log(dbenv, txnp, ret_lsnp, flags,
 		rectype |= DB_debug_FLAG;
 		memcpy(logrec.data, &rectype, sizeof(rectype));
 
-		ret = __log_put(dbenv,
-		    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
-#else
-		ret = 0;
+		if (!IS_REP_CLIENT(dbenv))
+			ret = __log_put(dbenv,
+			    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
 #endif
 		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
 		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
@@ -914,6 +914,7 @@ __txn_xa_regop_log(dbenv, txnp, ret_lsnp, flags,
 				 *ret_lsnp = *rlsnp;
 		}
 	} else {
+		ret = 0;
 #ifdef DIAGNOSTIC
 		/*
 		 * Set the debug bit if we are going to log non-durable
@@ -923,10 +924,9 @@ __txn_xa_regop_log(dbenv, txnp, ret_lsnp, flags,
 		rectype |= DB_debug_FLAG;
 		memcpy(logrec.data, &rectype, sizeof(rectype));
 
-		ret = __log_put(dbenv,
-		    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
-#else
-		ret = 0;
+		if (!IS_REP_CLIENT(dbenv))
+			ret = __log_put(dbenv,
+			    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
 #endif
 		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
 		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
@@ -1131,6 +1131,7 @@ __txn_recycle_log(dbenv, txnp, ret_lsnp, flags,
 				 *ret_lsnp = *rlsnp;
 		}
 	} else {
+		ret = 0;
 #ifdef DIAGNOSTIC
 		/*
 		 * Set the debug bit if we are going to log non-durable
@@ -1140,10 +1141,9 @@ __txn_recycle_log(dbenv, txnp, ret_lsnp, flags,
 		rectype |= DB_debug_FLAG;
 		memcpy(logrec.data, &rectype, sizeof(rectype));
 
-		ret = __log_put(dbenv,
-		    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
-#else
-		ret = 0;
+		if (!IS_REP_CLIENT(dbenv))
+			ret = __log_put(dbenv,
+			    rlsnp, (DBT *)&logrec, flags | DB_LOG_NOCOPY);
 #endif
 		STAILQ_INSERT_HEAD(&txnp->logs, lr, links);
 		F_SET((TXN_DETAIL *)txnp->td, TXN_DTL_INMEMORY);
diff --git a/db/txn/txn_autop.c b/db/txn/txn_autop.c
index c993e0fad..1cc878bd5 100644
--- a/db/txn/txn_autop.c
+++ b/db/txn/txn_autop.c
@@ -41,7 +41,7 @@ __txn_regop_42_print(dbenv, dbtp, lsnp, notused2, notused3)
 	    (u_long)argp->type,
 	    (u_long)argp->txnp->txnid,
 	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
-	(void)printf("\topcode: %ld\n", (long)argp->opcode);
+	(void)printf("\topcode: %lu\n", (u_long)argp->opcode);
 	timeval = (time_t)argp->timestamp;
 	lt = localtime(&timeval);
 	(void)printf(
@@ -87,7 +87,7 @@ __txn_regop_print(dbenv, dbtp, lsnp, notused2, notused3)
 	    (u_long)argp->type,
 	    (u_long)argp->txnp->txnid,
 	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
-	(void)printf("\topcode: %ld\n", (long)argp->opcode);
+	(void)printf("\topcode: %lu\n", (u_long)argp->opcode);
 	timeval = (time_t)argp->timestamp;
 	lt = localtime(&timeval);
 	(void)printf(
@@ -96,7 +96,7 @@ __txn_regop_print(dbenv, dbtp, lsnp, notused2, notused3)
 	    (u_long)lt->tm_year - 100, (u_long)lt->tm_mon+1,
 	    (u_long)lt->tm_mday, (u_long)lt->tm_hour,
 	    (u_long)lt->tm_min, (u_long)lt->tm_sec);
-	(void)printf("\tenvid: %ld\n", (long)argp->envid);
+	(void)printf("\tenvid: %lu\n", (u_long)argp->envid);
 	(void)printf("\tlocks: \n");
 	__lock_list_print(dbenv, &argp->locks);
 	(void)printf("\n");
@@ -146,7 +146,7 @@ __txn_ckp_42_print(dbenv, dbtp, lsnp, notused2, notused3)
 	    (u_long)lt->tm_year - 100, (u_long)lt->tm_mon+1,
 	    (u_long)lt->tm_mday, (u_long)lt->tm_hour,
 	    (u_long)lt->tm_min, (u_long)lt->tm_sec);
-	(void)printf("\trep_gen: %ld\n", (long)argp->rep_gen);
+	(void)printf("\trep_gen: %lu\n", (u_long)argp->rep_gen);
 	(void)printf("\n");
 	__os_free(dbenv, argp);
 	return (0);
@@ -194,8 +194,8 @@ __txn_ckp_print(dbenv, dbtp, lsnp, notused2, notused3)
 	    (u_long)lt->tm_year - 100, (u_long)lt->tm_mon+1,
 	    (u_long)lt->tm_mday, (u_long)lt->tm_hour,
 	    (u_long)lt->tm_min, (u_long)lt->tm_sec);
-	(void)printf("\tenvid: %ld\n", (long)argp->envid);
-	(void)printf("\trep_gen: %ld\n", (long)argp->rep_gen);
+	(void)printf("\tenvid: %lu\n", (u_long)argp->envid);
+	(void)printf("\tspare: %lu\n", (u_long)argp->spare);
 	(void)printf("\n");
 	__os_free(dbenv, argp);
 	return (0);
@@ -273,8 +273,8 @@ __txn_xa_regop_print(dbenv, dbtp, lsnp, notused2, notused3)
 	}
 	(void)printf("\n");
 	(void)printf("\tformatID: %ld\n", (long)argp->formatID);
-	(void)printf("\tgtrid: %u\n", argp->gtrid);
-	(void)printf("\tbqual: %u\n", argp->bqual);
+	(void)printf("\tgtrid: %lu\n", (u_long)argp->gtrid);
+	(void)printf("\tbqual: %lu\n", (u_long)argp->bqual);
 	(void)printf("\tbegin_lsn: [%lu][%lu]\n",
 	    (u_long)argp->begin_lsn.file, (u_long)argp->begin_lsn.offset);
 	(void)printf("\tlocks: \n");
@@ -311,8 +311,8 @@ __txn_recycle_print(dbenv, dbtp, lsnp, notused2, notused3)
 	    (u_long)argp->type,
 	    (u_long)argp->txnp->txnid,
 	    (u_long)argp->prev_lsn.file, (u_long)argp->prev_lsn.offset);
-	(void)printf("\tmin: %u\n", argp->min);
-	(void)printf("\tmax: %u\n", argp->max);
+	(void)printf("\tmin: %lu\n", (u_long)argp->min);
+	(void)printf("\tmax: %lu\n", (u_long)argp->max);
 	(void)printf("\n");
 	__os_free(dbenv, argp);
 	return (0);
diff --git a/db/txn/txn_chkpt.c b/db/txn/txn_chkpt.c
index 59ddf5509..89cf670e6 100644
--- a/db/txn/txn_chkpt.c
+++ b/db/txn/txn_chkpt.c
@@ -1,8 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 1996,2007 Oracle.  All rights reserved.
  */
 /*
  * Copyright (c) 1995, 1996
@@ -35,7 +34,7 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- * $Id: txn_chkpt.c,v 12.27 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_chkpt.c,v 12.47 2007/06/08 17:34:57 bostic Exp $
  */
 
 #include "db_config.h"
@@ -101,22 +100,11 @@ __txn_checkpoint(dbenv, kbytes, minutes, flags)
 	REGENV *renv;
 	REGINFO *infop;
 	time_t last_ckp_time, now;
-	u_int32_t bytes, gen, id, logflags, mbytes;
+	u_int32_t bytes, id, logflags, mbytes, op;
 	int ret;
 
-	ret = gen = 0;
-	/*
-	 * A client will only call through here during recovery,
-	 * so just sync the Mpool and go home.
-	 */
-	if (IS_REP_CLIENT(dbenv)) {
-		if (MPOOL_ON(dbenv) && (ret = __memp_sync(dbenv, NULL)) != 0) {
-			__db_err(dbenv, ret,
-		    "txn_checkpoint: failed to flush the buffer cache");
-			return (ret);
-		}
-		return (0);
-	}
+	DB_ASSERT(dbenv, !IS_REP_CLIENT(dbenv));
+	ret = 0;
 
 	mgr = dbenv->tx_handle;
 	region = mgr->reginfo.primary;
@@ -177,17 +165,68 @@ __txn_checkpoint(dbenv, kbytes, minutes, flags)
 	 * sees a later chk_lsn but competes first.  An archive process could
 	 * then remove a log this checkpoint depends on.
 	 */
-do_ckp:	MUTEX_LOCK(dbenv, region->mtx_ckp);
+do_ckp:
+	MUTEX_LOCK(dbenv, region->mtx_ckp);
 	if ((ret = __txn_getactive(dbenv, &ckp_lsn)) != 0)
 		goto err;
 
-	if (MPOOL_ON(dbenv) && (ret = __memp_sync(dbenv, NULL)) != 0) {
+	/*
+	 * Checkpoints in replication groups can cause performance problems.
+	 *
+	 * As on the master, checkpoint on the replica requires the cache be
+	 * flushed.  The problem occurs when a client has dirty cache pages
+	 * to write when the checkpoint record arrives, and the client's PERM
+	 * response is necessary in order to meet the system's durability
+	 * guarantees.  In this case, the master will have to wait until the
+	 * client completes its cache flush and writes the checkpoint record
+	 * before subsequent transactions can be committed.  The delay may
+	 * cause transactions to timeout waiting on client response, which
+	 * can cause nasty ripple effects in the system's overall throughput.
+	 * [#15338]
+	 *
+	 * First, we send a start-sync record when the checkpoint starts so
+	 * clients can start flushing their cache in preparation for the
+	 * arrival of the checkpoint record.
+	 */
+	if (LOGGING_ON(dbenv) &&
+	    IS_REP_MASTER(dbenv) && dbenv->rep_handle->send != NULL)
+		(void)__rep_send_message(dbenv,
+		    DB_EID_BROADCAST, REP_START_SYNC, &ckp_lsn, NULL, 0, 0);
+
+	/* Flush the cache. */
+	if (MPOOL_ON(dbenv) &&
+	    (ret = __memp_sync_int(
+		dbenv, NULL, 0, DB_SYNC_CHECKPOINT, NULL, NULL)) != 0) {
 		__db_err(dbenv, ret,
 		    "txn_checkpoint: failed to flush the buffer cache");
 		goto err;
 	}
 
 	/*
+	 * The client won't have more dirty pages to flush from its cache than
+	 * the master did, but there may be differences between the hardware,
+	 * I/O configuration and workload on the master and the client that
+	 * can result in the client being unable to finish its cache flush as
+	 * fast as the master.  A way to avoid the problem is to pause after
+	 * the master completes its checkpoint and before the actual checkpoint
+	 * record is logged, giving the replicas additional time to finish.
+	 *
+	 * !!!
+	 * We do not currently surface an API to modify this value,
+	 *
+	 * !!!
+	 * Currently turned off when testing, because it makes the test suite
+	 * take a long time to run.
+	 */
+#ifndef	CONFIG_TEST
+	if (LOGGING_ON(dbenv) &&
+	    IS_REP_MASTER(dbenv) && dbenv->rep_handle->send != NULL &&
+	    !LF_ISSET(DB_CKP_INTERNAL) &&
+	    dbenv->rep_handle->region->chkpt_delay != 0)
+		__os_sleep(dbenv, dbenv->rep_handle->region->chkpt_delay, 0);
+#endif
+
+	/*
 	 * Because we can't be a replication client here, and because
 	 * recovery (somewhat unusually) calls txn_checkpoint and expects
 	 * it to write a log message, LOGGING_ON is the correct macro here.
@@ -196,9 +235,6 @@ do_ckp:	MUTEX_LOCK(dbenv, region->mtx_ckp);
 		TXN_SYSTEM_LOCK(dbenv);
 		last_ckp = region->last_ckp;
 		TXN_SYSTEM_UNLOCK(dbenv);
-		if (REP_ON(dbenv) && (ret = __rep_get_gen(dbenv, &gen)) != 0)
-			goto err;
-
 		/*
 		 * Put out records for the open files before we log
 		 * the checkpoint.  The records are certain to be at
@@ -208,11 +244,20 @@ do_ckp:	MUTEX_LOCK(dbenv, region->mtx_ckp);
 		 * checkpoint.
 		 */
 		logflags = DB_LOG_CHKPNT;
+		/*
+		 * If this is a normal checkpoint, log files as checkpoints.
+		 * If we are recovering, only log as DBREG_RCLOSE if
+		 * there are no prepared txns.  Otherwise, it should
+		 * stay as DBREG_CHKPNT.
+		 */
+		op = DBREG_CHKPNT;
 		if (!IS_RECOVERING(dbenv))
 			logflags |= DB_FLUSH;
-		if ((ret = __dbreg_log_files(dbenv)) != 0 ||
+		else if (region->stat.st_nrestores == 0)
+			op = DBREG_RCLOSE;
+		if ((ret = __dbreg_log_files(dbenv, op)) != 0 ||
 		    (ret = __txn_ckp_log(dbenv, NULL, &ckp_lsn, logflags,
-		    &ckp_lsn, &last_ckp, (int32_t)time(NULL), id, gen)) != 0) {
+		    &ckp_lsn, &last_ckp, (int32_t)time(NULL), id, 0)) != 0) {
 			__db_err(dbenv, ret,
 			    "txn_checkpoint: log failed at LSN [%ld %ld]",
 			    (long)ckp_lsn.file, (long)ckp_lsn.offset);
diff --git a/db/txn/txn_failchk.c b/db/txn/txn_failchk.c
index 4301fdab8..92bb24d3a 100644
--- a/db/txn/txn_failchk.c
+++ b/db/txn/txn_failchk.c
@@ -1,10 +1,9 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 2005-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 2005,2007 Oracle.  All rights reserved.
  *
- * $Id: txn_failchk.c,v 12.6 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_failchk.c,v 12.9 2007/06/29 00:25:02 margo Exp $
  */
 
 #include "db_config.h"
@@ -63,7 +62,8 @@ retry:	TXN_SYSTEM_LOCK(dbenv);
 		TXN_SYSTEM_UNLOCK(dbenv);
 		if ((ret = __os_calloc(dbenv, 1, sizeof(DB_TXN), &txn)) != 0)
 			return (ret);
-		__txn_continue(dbenv, txn, td);
+		if ((ret = __txn_continue(dbenv, txn, td)) != 0)
+			return (ret);
 		F_SET(txn, TXN_MALLOC);
 		SH_TAILQ_FOREACH(ktd, &td->kids, klinks, __txn_detail) {
 			if (F_ISSET(ktd, TXN_DTL_INMEMORY))
@@ -73,7 +73,8 @@ retry:	TXN_SYSTEM_LOCK(dbenv);
 			if ((ret =
 			    __os_calloc(dbenv, 1, sizeof(DB_TXN), &ktxn)) != 0)
 				return (ret);
-			__txn_continue(dbenv, ktxn, ktd);
+			if ((ret = __txn_continue(dbenv, ktxn, ktd)) != 0)
+				return (ret);
 			F_SET(ktxn, TXN_MALLOC);
 			ktxn->parent = txn;
 			TAILQ_INSERT_HEAD(&txn->kids, txn, klinks);
diff --git a/db/txn/txn_method.c b/db/txn/txn_method.c
index f56867473..d1c46da2b 100644
--- a/db/txn/txn_method.c
+++ b/db/txn/txn_method.c
@@ -1,10 +1,9 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 1996,2007 Oracle.  All rights reserved.
  *
- * $Id: txn_method.c,v 12.6 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_method.c,v 12.9 2007/05/17 15:16:00 bostic Exp $
  */
 
 #include "db_config.h"
@@ -13,13 +12,13 @@
 #include "dbinc/txn.h"
 
 /*
- * __txn_dbenv_create --
+ * __txn_env_create --
  *	Transaction specific initialization of the DB_ENV structure.
  *
- * PUBLIC: int __txn_dbenv_create __P((DB_ENV *));
+ * PUBLIC: int __txn_env_create __P((DB_ENV *));
  */
 int
-__txn_dbenv_create(dbenv)
+__txn_env_create(dbenv)
 	DB_ENV *dbenv;
 {
 	/*
@@ -34,13 +33,13 @@ __txn_dbenv_create(dbenv)
 }
 
 /*
- * __txn_dbenv_destroy --
+ * __txn_env_destroy --
  *	Transaction specific destruction of the DB_ENV structure.
  *
- * PUBLIC: void __txn_dbenv_destroy __P((DB_ENV *));
+ * PUBLIC: void __txn_env_destroy __P((DB_ENV *));
  */
 void
-__txn_dbenv_destroy(dbenv)
+__txn_env_destroy(dbenv)
 	DB_ENV *dbenv;
 {
 	COMPQUIET(dbenv, NULL);
diff --git a/db/txn/txn_rec.c b/db/txn/txn_rec.c
index c81a9dcae..263ae28f2 100644
--- a/db/txn/txn_rec.c
+++ b/db/txn/txn_rec.c
@@ -1,8 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 1996,2007 Oracle.  All rights reserved.
  */
 /*
  * Copyright (c) 1996
@@ -32,13 +31,14 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- * $Id: txn_rec.c,v 12.14 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_rec.c,v 12.25 2007/06/01 15:36:52 sue Exp $
  */
 
 #include "db_config.h"
 
 #include "db_int.h"
 #include "dbinc/db_page.h"
+#include "dbinc/lock.h"
 #include "dbinc/txn.h"
 #include "dbinc/db_am.h"
 
@@ -142,6 +142,9 @@ __txn_xa_regop_recover(dbenv, dbtp, lsnp, op, info)
 	db_recops op;
 	void *info;
 {
+	DBT *lock_dbt;
+	DB_TXNHEAD *headp;
+	DB_LOCKTAB *lt;
 	__txn_xa_regop_args *argp;
 	int ret;
 	u_int32_t status;
@@ -157,6 +160,7 @@ __txn_xa_regop_recover(dbenv, dbtp, lsnp, op, info)
 		ret = EINVAL;
 		goto err;
 	}
+	headp = info;
 
 	/*
 	 * The return value here is either a DB_NOTFOUND or it is
@@ -169,9 +173,8 @@ __txn_xa_regop_recover(dbenv, dbtp, lsnp, op, info)
 
 	/*
 	 * If we are rolling forward, then an aborted prepare
-	 * indicates that this may the last record we'll see for
-	 * this transaction ID, so we should remove it from the
-	 * list.
+	 * indicates that this may be the last record we'll see for
+	 * this transaction ID, so we should remove it from the list.
 	 */
 
 	if (op == DB_TXN_FORWARD_ROLL) {
@@ -211,9 +214,27 @@ txn_err:		__db_errx(dbenv,
 			    "transaction not in list %lx",
 			    (u_long)argp->txnp->txnid);
 			ret = DB_NOTFOUND;
-		} else if ((ret = __db_txnlist_add(dbenv,
-		   info, argp->txnp->txnid, TXN_COMMIT, lsnp)) == 0)
-			ret = __txn_restore_txn(dbenv, lsnp, argp);
+		} else if (IS_ZERO_LSN(headp->trunc_lsn) ||
+		    LOG_COMPARE(&headp->trunc_lsn, lsnp) >= 0) {
+			if ((ret = __db_txnlist_add(dbenv,
+			   info, argp->txnp->txnid, TXN_COMMIT, lsnp)) == 0) {
+				/* Re-acquire the locks for this transaction. */
+				lock_dbt = &argp->locks;
+				if (LOCKING_ON(dbenv)) {
+					lt = dbenv->lk_handle;
+					if ((ret = __lock_getlocker(lt,
+						argp->txnp->txnid, 1,
+						&argp->txnp->locker)) != 0)
+						goto err;
+					if ((ret = __lock_get_list(dbenv,
+					    argp->txnp->locker, 0,
+					    DB_LOCK_WRITE, lock_dbt)) != 0)
+						goto err;
+				}
+
+				ret = __txn_restore_txn(dbenv, lsnp, argp);
+			}
+		}
 	} else
 		ret = 0;
 
@@ -237,8 +258,6 @@ __txn_ckp_recover(dbenv, dbtp, lsnp, op, info)
 	db_recops op;
 	void *info;
 {
-	DB_REP *db_rep;
-	REP *rep;
 	__txn_ckp_args *argp;
 	int ret;
 
@@ -251,16 +270,6 @@ __txn_ckp_recover(dbenv, dbtp, lsnp, op, info)
 	if (op == DB_TXN_BACKWARD_ROLL)
 		__db_txnlist_ckp(dbenv, info, lsnp);
 
-	if (op == DB_TXN_FORWARD_ROLL) {
-		/* Record the max generation number that we've seen. */
-		if (REP_ON(dbenv)) {
-			db_rep = dbenv->rep_handle;
-			rep = db_rep->region;
-			if (argp->rep_gen > rep->recover_gen)
-				rep->recover_gen = argp->rep_gen;
-		}
-	}
-
 	*lsnp = argp->last_ckp;
 	__os_free(dbenv, argp);
 	return (DB_TXN_CKP);
@@ -418,8 +427,7 @@ __txn_restore_txn(dbenv, lsnp, argp)
 	TXN_SYSTEM_LOCK(dbenv);
 
 	/* Allocate a new transaction detail structure. */
-	if ((ret =
-	    __db_shalloc(&mgr->reginfo, sizeof(TXN_DETAIL), 0, &td)) != 0) {
+	if ((ret = __env_alloc(&mgr->reginfo, sizeof(TXN_DETAIL), &td)) != 0) {
 		TXN_SYSTEM_UNLOCK(dbenv);
 		return (ret);
 	}
@@ -445,11 +453,16 @@ __txn_restore_txn(dbenv, lsnp, argp)
 	td->bqual = argp->bqual;
 	td->gtrid = argp->gtrid;
 	td->format = argp->formatID;
+	td->nlog_dbs = 0;
+	td->nlog_slots = TXN_NSLOTS;
+	td->log_dbs = R_OFFSET(&mgr->reginfo, td->slots);
 
 	region->stat.st_nrestores++;
+#ifdef HAVE_STATISTICS
 	region->stat.st_nactive++;
 	if (region->stat.st_nactive > region->stat.st_maxnactive)
 		region->stat.st_maxnactive = region->stat.st_nactive;
+#endif
 	TXN_SYSTEM_UNLOCK(dbenv);
 	return (0);
 }
@@ -587,8 +600,6 @@ __txn_ckp_42_recover(dbenv, dbtp, lsnp, op, info)
 	db_recops op;
 	void *info;
 {
-	DB_REP *db_rep;
-	REP *rep;
 	__txn_ckp_42_args *argp;
 	int ret;
 
@@ -601,16 +612,6 @@ __txn_ckp_42_recover(dbenv, dbtp, lsnp, op, info)
 	if (op == DB_TXN_BACKWARD_ROLL)
 		__db_txnlist_ckp(dbenv, info, lsnp);
 
-	if (op == DB_TXN_FORWARD_ROLL) {
-		/* Record the max generation number that we've seen. */
-		if (REP_ON(dbenv)) {
-			db_rep = dbenv->rep_handle;
-			rep = db_rep->region;
-			if (argp->rep_gen > rep->recover_gen)
-				rep->recover_gen = argp->rep_gen;
-		}
-	}
-
 	*lsnp = argp->last_ckp;
 	__os_free(dbenv, argp);
 	return (DB_TXN_CKP);
diff --git a/db/txn/txn_recover.c b/db/txn/txn_recover.c
index 961f0c39c..81c8d0c6e 100644
--- a/db/txn/txn_recover.c
+++ b/db/txn/txn_recover.c
@@ -1,10 +1,9 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 2001-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 2001,2007 Oracle.  All rights reserved.
  *
- * $Id: txn_recover.c,v 12.19 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_recover.c,v 12.29 2007/06/29 00:25:02 margo Exp $
  */
 
 #include "db_config.h"
@@ -134,7 +133,7 @@ __txn_get_prepared(dbenv, xids, txns, count, retp, flags)
 	XID *xids;
 	DB_PREPLIST *txns;
 	long count;		/* This is long for XA compatibility. */
-	long  *retp;
+	long *retp;
 	u_int32_t flags;
 {
 	DB_LSN min;
@@ -144,15 +143,14 @@ __txn_get_prepared(dbenv, xids, txns, count, retp, flags)
 	TXN_DETAIL *td;
 	XID *xidp;
 	long i;
-	int nrestores, open_files, ret;
+	int restored, ret;
 
 	*retp = 0;
 
 	MAX_LSN(min);
 	prepp = txns;
 	xidp = xids;
-	nrestores = ret = 0;
-	open_files = 1;
+	restored = ret = 0;
 
 	/*
 	 * If we are starting a scan, then we traverse the active transaction
@@ -173,24 +171,18 @@ __txn_get_prepared(dbenv, xids, txns, count, retp, flags)
 	 * restored, then we never crashed; just the main server did).
 	 */
 	TXN_SYSTEM_LOCK(dbenv);
-	if (flags == DB_FIRST) {
-		SH_TAILQ_FOREACH(td, &region->active_txn, links, __txn_detail) {
-			if (F_ISSET(td, TXN_DTL_RESTORED))
-				nrestores++;
-			F_CLR(td, TXN_DTL_COLLECTED);
-		}
-		mgr->n_discards = 0;
-	} else
-		open_files = 0;
 
 	/* Now begin collecting active transactions. */
 	for (td = SH_TAILQ_FIRST(&region->active_txn, __txn_detail);
 	    td != NULL && *retp < count;
 	    td = SH_TAILQ_NEXT(td, links, __txn_detail)) {
 		if (td->status != TXN_PREPARED ||
-		    F_ISSET(td, TXN_DTL_COLLECTED))
+		    (flags != DB_FIRST && F_ISSET(td, TXN_DTL_COLLECTED)))
 			continue;
 
+		if (F_ISSET(td, TXN_DTL_RESTORED))
+			restored = 1;
+
 		if (xids != NULL) {
 			xidp->formatID = td->format;
 			/*
@@ -210,7 +202,8 @@ __txn_get_prepared(dbenv, xids, txns, count, retp, flags)
 				TXN_SYSTEM_UNLOCK(dbenv);
 				goto err;
 			}
-			__txn_continue(dbenv, prepp->txn, td);
+			if ((ret = __txn_continue(dbenv, prepp->txn, td)) != 0)
+				goto err;
 			F_SET(prepp->txn, TXN_MALLOC);
 			memcpy(prepp->gid, td->xid, sizeof(td->xid));
 			prepp++;
@@ -222,30 +215,43 @@ __txn_get_prepared(dbenv, xids, txns, count, retp, flags)
 
 		(*retp)++;
 		F_SET(td, TXN_DTL_COLLECTED);
-		if (IS_ENV_REPLICATED(dbenv) &&
-		    (ret = __op_rep_enter(dbenv)) != 0)
-			goto err;
 	}
+	if (flags == DB_FIRST)
+		for (; td != NULL; td = SH_TAILQ_NEXT(td, links, __txn_detail))
+			F_CLR(td, TXN_DTL_COLLECTED);
 	TXN_SYSTEM_UNLOCK(dbenv);
 
 	/*
 	 * Now link all the transactions into the transaction manager's list.
 	 */
-	if (txns != NULL) {
+	if (txns != NULL && *retp != 0) {
 		MUTEX_LOCK(dbenv, mgr->mutex);
 		for (i = 0; i < *retp; i++)
 			TAILQ_INSERT_TAIL(&mgr->txn_chain, txns[i].txn, links);
 		MUTEX_UNLOCK(dbenv, mgr->mutex);
-	}
 
-	if (open_files && nrestores && *retp != 0 && !IS_MAX_LSN(min)) {
-		F_SET(dbenv->lg_handle, DBLOG_RECOVER);
-		ret = __txn_openfiles(dbenv, &min, 0);
-		F_CLR(dbenv->lg_handle, DBLOG_RECOVER);
+		/*
+		 * If we are restoring, update our count of outstanding
+		 * transactions.
+		 */
+		if (REP_ON(dbenv)) {
+			REP_SYSTEM_LOCK(dbenv);
+			dbenv->rep_handle->region->op_cnt += (u_long)*retp;
+			REP_SYSTEM_UNLOCK(dbenv);
+		}
+
 	}
-	return (0);
+	/*
+	 * If recovery already opened the files for us, don't
+	 * do it here.
+	 */
+	if (restored != 0 && flags == DB_FIRST &&
+	    !F_ISSET(dbenv->lg_handle, DBLOG_OPENFILES))
+		ret = __txn_openfiles(dbenv, &min, 0);
 
-err:	TXN_SYSTEM_UNLOCK(dbenv);
+	if (0) {
+err:		TXN_SYSTEM_UNLOCK(dbenv);
+	}
 	return (ret);
 }
 
@@ -279,7 +285,7 @@ __txn_openfiles(dbenv, min, force)
 	memset(&data, 0, sizeof(data));
 	if ((ret = __txn_getckp(dbenv, &open_lsn)) == 0)
 		while (!IS_ZERO_LSN(open_lsn) && (ret =
-		    __log_c_get(logc, &open_lsn, &data, DB_SET)) == 0 &&
+		    __logc_get(logc, &open_lsn, &data, DB_SET)) == 0 &&
 		    (force ||
 		    (min != NULL && LOG_COMPARE(min, &open_lsn) < 0))) {
 			/* Format the log record. */
@@ -300,7 +306,7 @@ __txn_openfiles(dbenv, min, force)
 			    ckp_args->last_ckp;
 			__os_free(dbenv, ckp_args);
 			if (force) {
-				if ((ret = __log_c_get(logc, &open_lsn,
+				if ((ret = __logc_get(logc, &open_lsn,
 				    &data, DB_SET)) != 0)
 					goto err;
 				break;
@@ -317,7 +323,7 @@ __txn_openfiles(dbenv, min, force)
 	 * - We are forcing an openfiles and we have our ckp_lsn.
 	 */
 	if ((ret == DB_NOTFOUND || IS_ZERO_LSN(open_lsn)) && (ret =
-	    __log_c_get(logc, &open_lsn, &data, DB_FIRST)) != 0) {
+	    __logc_get(logc, &open_lsn, &data, DB_FIRST)) != 0) {
 		__db_errx(dbenv, "No log records");
 		goto err;
 	}
@@ -330,7 +336,7 @@ __txn_openfiles(dbenv, min, force)
 		__db_txnlist_end(dbenv, txninfo);
 
 err:
-	if (logc != NULL && (t_ret = __log_c_close(logc)) != 0 && ret == 0)
+	if (logc != NULL && (t_ret = __logc_close(logc)) != 0 && ret == 0)
 		ret = t_ret;
 	return (ret);
 }
diff --git a/db/txn/txn_region.c b/db/txn/txn_region.c
index 0b5547bb5..bbe908048 100644
--- a/db/txn/txn_region.c
+++ b/db/txn/txn_region.c
@@ -1,10 +1,9 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 1996,2007 Oracle.  All rights reserved.
  *
- * $Id: txn_region.c,v 12.20 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_region.c,v 12.31 2007/05/17 15:16:00 bostic Exp $
  */
 
 #include "db_config.h"
@@ -20,11 +19,12 @@ static size_t __txn_region_size __P((DB_ENV *));
  * __txn_open --
  *	Open a transaction region.
  *
- * PUBLIC: int __txn_open __P((DB_ENV *));
+ * PUBLIC: int __txn_open __P((DB_ENV *, int));
  */
 int
-__txn_open(dbenv)
+__txn_open(dbenv, create_ok)
 	DB_ENV *dbenv;
+	int create_ok;
 {
 	DB_TXNMGR *mgr;
 	int ret;
@@ -40,9 +40,9 @@ __txn_open(dbenv)
 	mgr->reginfo.type = REGION_TYPE_TXN;
 	mgr->reginfo.id = INVALID_REGION_ID;
 	mgr->reginfo.flags = REGION_JOIN_OK;
-	if (F_ISSET(dbenv, DB_ENV_CREATE))
+	if (create_ok)
 		F_SET(&mgr->reginfo, REGION_CREATE_OK);
-	if ((ret = __db_r_attach(dbenv,
+	if ((ret = __env_region_attach(dbenv,
 	    &mgr->reginfo, __txn_region_size(dbenv))) != 0)
 		goto err;
 
@@ -65,7 +65,7 @@ __txn_open(dbenv)
 
 err:	dbenv->tx_handle = NULL;
 	if (mgr->reginfo.addr != NULL)
-		(void)__db_r_detach(dbenv, &mgr->reginfo, 0);
+		(void)__env_region_detach(dbenv, &mgr->reginfo, 0);
 
 	(void)__mutex_free(dbenv, &mgr->mutex);
 	__os_free(dbenv, mgr);
@@ -106,8 +106,8 @@ __txn_init(dbenv, mgr)
 			return (ret);
 	}
 
-	if ((ret = __db_shalloc(&mgr->reginfo,
-	    sizeof(DB_TXNREGION), 0, &mgr->reginfo.primary)) != 0) {
+	if ((ret = __env_alloc(&mgr->reginfo,
+	    sizeof(DB_TXNREGION), &mgr->reginfo.primary)) != 0) {
 		__db_errx(dbenv,
 		    "Unable to allocate memory for the transaction region");
 		return (ret);
@@ -132,7 +132,9 @@ __txn_init(dbenv, mgr)
 	region->time_ckp = time(NULL);
 
 	memset(&region->stat, 0, sizeof(region->stat));
+#ifdef HAVE_STATISTICS
 	region->stat.st_maxtxns = region->maxtxns;
+#endif
 
 	SH_TAILQ_INIT(&region->active_txn);
 	SH_TAILQ_INIT(&region->mvcc_txn);
@@ -168,10 +170,10 @@ __txn_findlastckp(dbenv, lsnp, max_lsn)
 	memset(&dbt, 0, sizeof(dbt));
 	if (max_lsn != NULL) {
 		lsn = *max_lsn;
-		if ((ret = __log_c_get(logc, &lsn, &dbt, DB_SET)) != 0)
+		if ((ret = __logc_get(logc, &lsn, &dbt, DB_SET)) != 0)
 			goto err;
 	} else {
-		if ((ret = __log_c_get(logc, &lsn, &dbt, DB_LAST)) != 0)
+		if ((ret = __logc_get(logc, &lsn, &dbt, DB_LAST)) != 0)
 			goto err;
 		/*
 		 * Twiddle the last LSN so it points to the beginning of the
@@ -182,7 +184,7 @@ __txn_findlastckp(dbenv, lsnp, max_lsn)
 	}
 
 	/* Read backwards, looking for checkpoints. */
-	while ((ret = __log_c_get(logc, &lsn, &dbt, DB_PREV)) == 0) {
+	while ((ret = __logc_get(logc, &lsn, &dbt, DB_PREV)) == 0) {
 		if (dbt.size < sizeof(u_int32_t))
 			continue;
 		memcpy(&rectype, dbt.data, sizeof(u_int32_t));
@@ -192,7 +194,7 @@ __txn_findlastckp(dbenv, lsnp, max_lsn)
 		}
 	}
 
-err:	if ((t_ret = __log_c_close(logc)) != 0 && ret == 0)
+err:	if ((t_ret = __logc_close(logc)) != 0 && ret == 0)
 		ret = t_ret;
 
 	/*
@@ -203,13 +205,13 @@ err:	if ((t_ret = __log_c_close(logc)) != 0 && ret == 0)
 }
 
 /*
- * __txn_dbenv_refresh --
+ * __txn_env_refresh --
  *	Clean up after the transaction system on a close or failed open.
  *
- * PUBLIC: int __txn_dbenv_refresh __P((DB_ENV *));
+ * PUBLIC: int __txn_env_refresh __P((DB_ENV *));
  */
 int
-__txn_dbenv_refresh(dbenv)
+__txn_env_refresh(dbenv)
 	DB_ENV *dbenv;
 {
 	DB_TXN *txn;
@@ -269,7 +271,7 @@ __txn_dbenv_refresh(dbenv)
 		ret = t_ret;
 
 	/* Detach from the region. */
-	if ((t_ret = __db_r_detach(dbenv, reginfo, 0)) != 0 && ret == 0)
+	if ((t_ret = __env_region_detach(dbenv, reginfo, 0)) != 0 && ret == 0)
 		ret = t_ret;
 
 	__os_free(dbenv, mgr);
@@ -279,12 +281,25 @@ __txn_dbenv_refresh(dbenv)
 }
 
 /*
+ * __txn_region_mutex_count --
+ *	Return the number of mutexes the txn region will need.
+ *
+ * PUBLIC: u_int32_t __txn_region_mutex_count __P((DB_ENV *));
+ */
+u_int32_t
+__txn_region_mutex_count(dbenv)
+	DB_ENV *dbenv;
+{
+	/*
+	 * We need a MVCC mutex for each TXN_DETAIL structure, a mutex for
+	 * DB_TXNMGR structure, two mutexes for the DB_TXNREGION structure.
+	 */
+	return (dbenv->tx_max + 1 + 2);
+}
+
+/*
  * __txn_region_size --
- *	 Return the amount of space needed for the txn region.  Make the
- *	 region large enough to hold txn_max transaction detail structures
- *	 plus some space to hold thread handles and the beginning of the
- *	 shalloc region and anything we need for mutex system resource
- *	 recording.
+ *	 Return the amount of space needed for the txn region.
  */
 static size_t
 __txn_region_size(dbenv)
@@ -292,8 +307,16 @@ __txn_region_size(dbenv)
 {
 	size_t s;
 
+	/*
+	 * Make the region large enough to hold the primary transaction region
+	 * structure, txn_max transaction detail structures, txn_max chunks of
+	 * overhead required by the underlying shared region allocator for each
+	 * chunk of memory, txn_max transaction names, at an average of 20
+	 * bytes each, and 10KB for safety.
+	 */
 	s = sizeof(DB_TXNREGION) +
-	    dbenv->tx_max * sizeof(TXN_DETAIL) + 10 * 1024;
+	    dbenv->tx_max * (sizeof(TXN_DETAIL) + __env_alloc_overhead() + 20) +
+	    10 * 1024;
 	return (s);
 }
 
@@ -377,7 +400,8 @@ __txn_oldest_reader(dbenv, lsnp)
  *
  * PUBLIC: int __txn_add_buffer __P((DB_ENV *, TXN_DETAIL *));
  */
-int __txn_add_buffer(dbenv, td)
+int
+__txn_add_buffer(dbenv, td)
 	DB_ENV *dbenv;
 	TXN_DETAIL *td;
 {
@@ -388,6 +412,7 @@ int __txn_add_buffer(dbenv, td)
 	++td->mvcc_ref;
 	MUTEX_UNLOCK(dbenv, td->mvcc_mtx);
 
+	COMPQUIET(dbenv, NULL);
 	return (0);
 }
 
@@ -397,7 +422,8 @@ int __txn_add_buffer(dbenv, td)
  *
  * PUBLIC: int __txn_remove_buffer __P((DB_ENV *, TXN_DETAIL *, db_mutex_t));
  */
-int __txn_remove_buffer(dbenv, td, hash_mtx)
+int
+__txn_remove_buffer(dbenv, td, hash_mtx)
 	DB_ENV *dbenv;
 	TXN_DETAIL *td;
 	db_mutex_t hash_mtx;
@@ -425,8 +451,10 @@ int __txn_remove_buffer(dbenv, td, hash_mtx)
 
 		TXN_SYSTEM_LOCK(dbenv);
 		SH_TAILQ_REMOVE(&region->mvcc_txn, td, links, __txn_detail);
+#ifdef HAVE_STATISTICS
 		--region->stat.st_nsnapshot;
-		__db_shalloc_free(&mgr->reginfo, td);
+#endif
+		__env_alloc_free(&mgr->reginfo, td);
 		TXN_SYSTEM_UNLOCK(dbenv);
 
 		MUTEX_LOCK(dbenv, hash_mtx);
diff --git a/db/txn/txn_stat.c b/db/txn/txn_stat.c
index 2f7cf95e5..a2c4f1fe2 100644
--- a/db/txn/txn_stat.c
+++ b/db/txn/txn_stat.c
@@ -1,10 +1,9 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 1996,2007 Oracle.  All rights reserved.
  *
- * $Id: txn_stat.c,v 12.18 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_stat.c,v 12.23 2007/06/22 18:27:51 bostic Exp $
  */
 
 #include "db_config.h"
@@ -134,7 +133,8 @@ __txn_stat(dbenv, statp, flags)
 	    &stats->st_region_wait, &stats->st_region_nowait);
 	stats->st_regsize = mgr->reginfo.rp->size;
 	if (LF_ISSET(DB_STAT_CLEAR)) {
-		__mutex_clear(dbenv, region->mtx_region);
+		if (!LF_ISSET(DB_STAT_SUBSYSTEM))
+			__mutex_clear(dbenv, region->mtx_region);
 		memset(&region->stat, 0, sizeof(region->stat));
 		region->stat.st_maxtxns = region->maxtxns;
 		region->stat.st_maxnactive =
@@ -167,7 +167,7 @@ __txn_stat_print_pp(dbenv, flags)
 	ENV_REQUIRES_CONFIG(dbenv,
 	    dbenv->tx_handle, "DB_ENV->txn_stat_print", DB_INIT_TXN);
 
-	if ((ret = __db_fchk(dbenv, "DB_ENV->txn_stat",
+	if ((ret = __db_fchk(dbenv, "DB_ENV->txn_stat_print",
 	    flags, DB_STAT_ALL | DB_STAT_CLEAR)) != 0)
 		return (ret);
 
@@ -192,7 +192,7 @@ __txn_stat_print(dbenv, flags)
 	int ret;
 
 	orig_flags = flags;
-	LF_CLR(DB_STAT_CLEAR);
+	LF_CLR(DB_STAT_CLEAR | DB_STAT_SUBSYSTEM);
 	if (flags == 0 || LF_ISSET(DB_STAT_ALL)) {
 		ret = __txn_print_stats(dbenv, orig_flags);
 		if (flags == 0 || ret != 0)
@@ -317,7 +317,7 @@ __txn_print_all(dbenv, flags)
 
 	TXN_SYSTEM_LOCK(dbenv);
 
-	__db_print_reginfo(dbenv, &mgr->reginfo, "Transaction");
+	__db_print_reginfo(dbenv, &mgr->reginfo, "Transaction", flags);
 
 	__db_msg(dbenv, "%s", DB_GLOBAL(db_line));
 	__db_msg(dbenv, "DB_TXNMGR handle information:");
diff --git a/db/txn/txn_util.c b/db/txn/txn_util.c
index fea21c76f..a455a2ff2 100644
--- a/db/txn/txn_util.c
+++ b/db/txn/txn_util.c
@@ -1,10 +1,9 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 2001-2006
- *	Oracle Corporation.  All rights reserved.
+ * Copyright (c) 2001,2007 Oracle.  All rights reserved.
  *
- * $Id: txn_util.c,v 12.8 2006/08/24 14:46:53 bostic Exp $
+ * $Id: txn_util.c,v 12.18 2007/06/13 18:21:32 ubell Exp $
  */
 
 #include "db_config.h"
@@ -14,6 +13,7 @@
 #include "dbinc/lock.h"
 #include "dbinc/mp.h"
 #include "dbinc/txn.h"
+#include "dbinc/log.h"
 #include "dbinc/db_am.h"
 
 typedef struct __txn_event TXN_EVENT;
@@ -34,7 +34,7 @@ struct __txn_event {
 		struct {
 			/* Lock event. */
 			DB_LOCK lock;
-			u_int32_t locker;
+			DB_LOCKER *locker;
 			DB *dbp;
 		} t;
 	} u;
@@ -149,7 +149,7 @@ __txn_remrem(dbenv, txn, name)
  * trade.
  *
  * PUBLIC: int __txn_lockevent __P((DB_ENV *,
- * PUBLIC:     DB_TXN *, DB *, DB_LOCK *, u_int32_t));
+ * PUBLIC:     DB_TXN *, DB *, DB_LOCK *, DB_LOCKER *));
  */
 int
 __txn_lockevent(dbenv, txn, dbp, lock, locker)
@@ -157,7 +157,7 @@ __txn_lockevent(dbenv, txn, dbp, lock, locker)
 	DB_TXN *txn;
 	DB *dbp;
 	DB_LOCK *lock;
-	u_int32_t locker;
+	DB_LOCKER *locker;
 {
 	int ret;
 	TXN_EVENT *e;
@@ -174,6 +174,7 @@ __txn_lockevent(dbenv, txn, dbp, lock, locker)
 	e->u.t.dbp = dbp;
 	e->op = TXN_TRADE;
 	TAILQ_INSERT_TAIL(&txn->events, e, links);
+	dbp->cur_txn = txn;
 
 	return (0);
 }
@@ -183,14 +184,14 @@ __txn_lockevent(dbenv, txn, dbp, lock, locker)
  *	Remove a lock event because the locker is going away.  We can remove
  * by lock (using offset) or by locker_id (or by both).
  *
- * PUBLIC: void __txn_remlock __P((DB_ENV *, DB_TXN *, DB_LOCK *, u_int32_t));
+ * PUBLIC: void __txn_remlock __P((DB_ENV *, DB_TXN *, DB_LOCK *, DB_LOCKER *));
  */
 void
 __txn_remlock(dbenv, txn, lock, locker)
 	DB_ENV *dbenv;
 	DB_TXN *txn;
 	DB_LOCK *lock;
-	u_int32_t locker;
+	DB_LOCKER *locker;
 {
 	TXN_EVENT *e, *next_e;
 
@@ -218,9 +219,10 @@ __txn_remlock(dbenv, txn, lock, locker)
 	req.lock = e->u.t.lock;						\
 	req.op = DB_LOCK_TRADE;						\
 	t_ret = __lock_vec(dbenv, e->u.t.locker, 0, &req, 1, NULL);	\
-	if (t_ret == 0)							\
-		e->u.t.dbp->cur_lid = e->u.t.locker;			\
-	else if (t_ret == DB_NOTFOUND)					\
+	if (t_ret == 0)	{						\
+		e->u.t.dbp->cur_locker = e->u.t.locker;			\
+		e->u.t.dbp->cur_txn = NULL;				\
+	} else if (t_ret == DB_NOTFOUND)				\
 		t_ret = 0;						\
 	if (t_ret != 0 && ret == 0)					\
 		ret = t_ret;						\
@@ -275,8 +277,6 @@ __txn_doevents(dbenv, txn, opcode, preprocess)
 			goto dofree;
 		switch (e->op) {
 		case TXN_CLOSE:
-			/* If we didn't abort this txn, we screwed up badly. */
-			DB_ASSERT(dbenv, opcode == TXN_ABORT);
 			if ((t_ret = __db_close(e->u.c.dbp,
 			    NULL, DB_NOSYNC)) != 0 && ret == 0)
 				ret = t_ret;
@@ -323,3 +323,109 @@ dofree:
 
 	return (ret);
 }
+
+/*
+ * PUBLIC: int __txn_record_fname __P((DB_ENV *, DB_TXN *, FNAME *));
+ */
+int
+__txn_record_fname(dbenv, txn, fname)
+	DB_ENV *dbenv;
+	DB_TXN *txn;
+	FNAME *fname;
+{
+	DB_TXNMGR *mgr;
+	DB_LOG *dblp;
+	TXN_DETAIL *td;
+	roff_t fname_off;
+	roff_t *np, *ldbs;
+	u_int32_t i;
+	int ret;
+
+	if ((td = txn->td) == NULL)
+		return (0);
+	mgr = dbenv->tx_handle;
+	dblp = dbenv->lg_handle;
+	fname_off = R_OFFSET(&dblp->reginfo, fname);
+
+	/* See if we already have a ref to this DB handle. */
+	ldbs = R_ADDR(&mgr->reginfo, td->log_dbs);
+	for (i = 0, np = ldbs; i < td->nlog_dbs; i++, np++)
+		if (*np == fname_off)
+			return (0);
+
+	if (td->nlog_slots <= td->nlog_dbs) {
+		TXN_SYSTEM_LOCK(dbenv);
+		if ((ret = __env_alloc(&mgr->reginfo,
+		    sizeof(roff_t) * (td->nlog_slots << 1), &np)) != 0)
+			return (ret);
+		memcpy(np, ldbs, td->nlog_dbs * sizeof(roff_t));
+		if (td->nlog_slots > TXN_NSLOTS)
+			__env_alloc_free(&mgr->reginfo, ldbs);
+
+		TXN_SYSTEM_UNLOCK(dbenv);
+		td->log_dbs = R_OFFSET(&mgr->reginfo, np);
+		ldbs = np;
+		td->nlog_slots = td->nlog_slots << 1;
+	}
+
+	ldbs[td->nlog_dbs] = fname_off;
+	td->nlog_dbs++;
+	fname->txn_ref++;
+
+	return (0);
+}
+
+/*
+ * __txn_dref_fnam --
+ *	Either pass the fname to our parent txn or decrement the refcount
+ * and close the fileid if it goes to zero.
+ *
+ * PUBLIC: int __txn_dref_fname __P((DB_ENV *, DB_TXN *));
+ */
+int
+__txn_dref_fname(dbenv, txn)
+	DB_ENV *dbenv;
+	DB_TXN *txn;
+{
+	DB_TXNMGR *mgr;
+	DB_LOG *dblp;
+	FNAME *fname;
+	roff_t *np;
+	TXN_DETAIL *ptd, *td;
+	u_int32_t i;
+	int ret;
+
+	td = txn->td;
+
+	if (td->nlog_dbs == 0)
+		return (0);
+
+	mgr = dbenv->tx_handle;
+	dblp = dbenv->lg_handle;
+	ret = 0;
+
+	ptd = txn->parent != NULL ? txn->parent->td : NULL;
+
+	np = R_ADDR(&mgr->reginfo, td->log_dbs);
+	for (i = 0; i < td->nlog_dbs; i++, np++) {
+		fname = R_ADDR(&dblp->reginfo, *np);
+		MUTEX_LOCK(dbenv, fname->mutex);
+		if (ptd != NULL) {
+			fname->txn_ref--;
+			ret = __txn_record_fname(dbenv, txn->parent, fname);
+			MUTEX_UNLOCK(dbenv, fname->mutex);
+		} else if (fname->txn_ref == 1) {
+			MUTEX_UNLOCK(dbenv, fname->mutex);
+			DB_ASSERT(dbenv, fname->txn_ref != 0);
+			ret = __dbreg_close_id_int(
+			    dbenv, fname, DBREG_CLOSE, 0);
+		} else {
+			fname->txn_ref--;
+			MUTEX_UNLOCK(dbenv, fname->mutex);
+		}
+		if (ret != 0)
+			break;
+	}
+
+	return (ret);
+}