18 files changed, 195 insertions, 167 deletions

diff --git a/src/btree/bt_compact.c b/src/btree/bt_compact.c
index d8b3a638de3..1528d65b8c8 100644
--- a/src/btree/bt_compact.c
+++ b/src/btree/bt_compact.c
@@ -76,7 +76,7 @@ __wt_compact(WT_SESSION_IMPL *session, const char *cfg[])
 	WT_CONNECTION_IMPL *conn;
 	WT_DECL_RET;
 	WT_REF *ref;
-	int block_manager_begin, skip;
+	int block_manager_begin, evict_reset, skip;
 
 	WT_UNUSED(cfg);
 
@@ -133,8 +133,9 @@ __wt_compact(WT_SESSION_IMPL *session, const char *cfg[])
 	 * then let eviction continue;
 	 */
 	conn->compact_in_memory_pass = 1;
-	WT_ERR(__wt_evict_file_exclusive_on(session));
-	__wt_evict_file_exclusive_off(session);
+	WT_ERR(__wt_evict_file_exclusive_on(session, &evict_reset));
+	if (evict_reset)
+		__wt_evict_file_exclusive_off(session);
 
 	/* Start compaction. */
 	WT_ERR(bm->compact_start(bm, session));
diff --git a/src/btree/bt_cursor.c b/src/btree/bt_cursor.c
index 1960e4605ef..8409b0cd6a2 100644
--- a/src/btree/bt_cursor.c
+++ b/src/btree/bt_cursor.c
@@ -936,36 +936,22 @@ __cursor_truncate(WT_SESSION_IMPL *session,
 	 * instantiated the end cursor, so we know that page is pinned in memory
 	 * and we can proceed without concern.
 	 */
-	if (start == NULL) {
-		do {
-			WT_RET(__wt_btcur_remove(stop));
-			for (;;) {
-				if ((ret = __wt_btcur_prev(stop, 1)) != 0)
-					break;
-				stop->compare = 0;	/* Exact match */
-				if ((ret = rmfunc(session, stop, 1)) != 0)
-					break;
-			}
-		} while (ret == WT_RESTART);
-	} else {
-		do {
-			WT_RET(__wt_btcur_remove(start));
-			/*
-			 * Reset ret each time through so that we don't loop
-			 * forever in the cursor equals case.
-			 */
-			for (ret = 0;;) {
-				if (stop != NULL &&
-				    __cursor_equals(start, stop))
-					break;
-				if ((ret = __wt_btcur_next(start, 1)) != 0)
-					break;
-				start->compare = 0;	/* Exact match */
-				if ((ret = rmfunc(session, start, 1)) != 0)
-					break;
-			}
-		} while (ret == WT_RESTART);
-	}
+	do {
+		WT_RET(__wt_btcur_remove(start));
+		/*
+		 * Reset ret each time through so that we don't loop forever in
+		 * the cursor equals case.
+		 */
+		for (ret = 0;;) {
+			if (stop != NULL && __cursor_equals(start, stop))
+				break;
+			if ((ret = __wt_btcur_next(start, 1)) != 0)
+				break;
+			start->compare = 0;	/* Exact match */
+			if ((ret = rmfunc(session, start, 1)) != 0)
+				break;
+		}
+	} while (ret == WT_RESTART);
 
 	WT_RET_NOTFOUND_OK(ret);
 	return (0);
@@ -999,40 +985,24 @@ __cursor_truncate_fix(WT_SESSION_IMPL *session,
 	 * other thread of control; in that case, repeat the full search to
 	 * refresh the page's modification information.
 	 */
-	if (start == NULL) {
-		do {
-			WT_RET(__wt_btcur_remove(stop));
-			for (;;) {
-				if ((ret = __wt_btcur_prev(stop, 1)) != 0)
-					break;
-				stop->compare = 0;	/* Exact match */
-				value = (uint8_t *)stop->iface.value.data;
-				if (*value != 0 &&
-				    (ret = rmfunc(session, stop, 1)) != 0)
-					break;
-			}
-		} while (ret == WT_RESTART);
-	} else {
-		do {
-			WT_RET(__wt_btcur_remove(start));
-			/*
-			 * Reset ret each time through so that we don't loop
-			 * forever in the cursor equals case.
-			 */
-			for (ret = 0;;) {
-				if (stop != NULL &&
-				    __cursor_equals(start, stop))
-					break;
-				if ((ret = __wt_btcur_next(start, 1)) != 0)
-					break;
-				start->compare = 0;	/* Exact match */
-				value = (uint8_t *)start->iface.value.data;
-				if (*value != 0 &&
-				    (ret = rmfunc(session, start, 1)) != 0)
-					break;
-			}
-		} while (ret == WT_RESTART);
-	}
+	do {
+		WT_RET(__wt_btcur_remove(start));
+		/*
+		 * Reset ret each time through so that we don't loop forever in
+		 * the cursor equals case.
+		 */
+		for (ret = 0;;) {
+			if (stop != NULL && __cursor_equals(start, stop))
+				break;
+			if ((ret = __wt_btcur_next(start, 1)) != 0)
+				break;
+			start->compare = 0;	/* Exact match */
+			value = (uint8_t *)start->iface.value.data;
+			if (*value != 0 &&
+			    (ret = rmfunc(session, start, 1)) != 0)
+				break;
+		}
+	} while (ret == WT_RESTART);
 
 	WT_RET_NOTFOUND_OK(ret);
 	return (0);
@@ -1055,9 +1025,15 @@ __wt_btcur_range_truncate(WT_CURSOR_BTREE *start, WT_CURSOR_BTREE *stop)
 	btree = cbt->btree;
 
 	/*
-	 * For recovery, we log the start and stop keys for a truncate
-	 * operation, not the individual records removed.  On the other hand,
-	 * for rollback we need to keep track of all the in-memory operations.
+	 * We always delete in a forward direction because it's faster, assert
+	 * our caller provided us with a start cursor.
+	 */
+	WT_ASSERT(session, start != NULL);
+
+	/*
+	 * For recovery, log the start and stop keys for a truncate operation,
+	 * not the individual records removed.  On the other hand, for rollback
+	 * we need to keep track of all the in-memory operations.
 	 *
 	 * We deal with this here by logging the truncate range first, then (in
 	 * the logging code) disabling writing of the in-memory remove records
@@ -1081,15 +1057,13 @@ __wt_btcur_range_truncate(WT_CURSOR_BTREE *start, WT_CURSOR_BTREE *stop)
 		 * fully instantiated when truncating row-store objects because
 		 * it's comparing page and/or skiplist positions, not keys. (Key
 		 * comparison would work, it's only that a key comparison would
-		 * be relatively expensive.  Column-store objects have record
-		 * number keys, so the key comparison is cheap.)  Cursors may
-		 * have only had their keys set, so we must ensure the cursors
-		 * are positioned in the tree.
+		 * be relatively expensive, especially with custom collators.
+		 * Column-store objects have record number keys, so the key
+		 * comparison is cheap.)  The session truncate code did cursor
+		 * searches when setting up the truncate so we're good to go: if
+		 * that ever changes, we'd need to do something here to ensure a
+		 * fully instantiated cursor.
 		 */
-		if (start != NULL)
-			WT_ERR(__wt_btcur_search(start));
-		if (stop != NULL)
-			WT_ERR(__wt_btcur_search(stop));
 		WT_ERR(__cursor_truncate(
 		    session, start, stop, __cursor_row_modify));
 		break;
diff --git a/src/btree/bt_page.c b/src/btree/bt_page.c
index 2f2ce4cf4f7..b5140beb792 100644
--- a/src/btree/bt_page.c
+++ b/src/btree/bt_page.c
@@ -129,15 +129,24 @@ __wt_page_in_func(WT_SESSION_IMPL *session, WT_REF *ref, uint32_t flags
 			    __evict_force_check(session, page, flags)) {
 				++force_attempts;
 				ret = __wt_page_release_evict(session, ref);
+				/* If forced eviction fails, stall. */
 				if (ret == EBUSY) {
-					/* If forced eviction fails, stall. */
 					ret = 0;
 					wait_cnt += 1000;
+					WT_STAT_FAST_CONN_INCR(session,
+					    page_forcible_evict_blocked);
+					break;
 				} else
 					WT_RET(ret);
-				WT_STAT_FAST_CONN_INCR(
-				    session, page_forcible_evict_blocked);
-				break;
+
+				/*
+				 * The result of a successful forced eviction
+				 * is a page-state transition (potentially to
+				 * an in-memory page we can use, or a restart
+				 * return for our caller), continue the outer
+				 * page-acquisition loop.
+				 */
+				continue;
 			}
 
 			/* Check if we need an autocommit transaction. */
diff --git a/src/btree/bt_sync.c b/src/btree/bt_sync.c
index a75af03d8c8..d925eefc2fe 100644
--- a/src/btree/bt_sync.c
+++ b/src/btree/bt_sync.c
@@ -25,6 +25,7 @@ __sync_file(WT_SESSION_IMPL *session, int syncop)
 	uint64_t internal_bytes, leaf_bytes;
 	uint64_t internal_pages, leaf_pages;
 	uint32_t flags;
+	int evict_reset;
 
 	btree = S2BT(session);
 
@@ -99,11 +100,11 @@ __sync_file(WT_SESSION_IMPL *session, int syncop)
 		 * eviction to complete.
 		 */
 		btree->checkpointing = 1;
+		WT_FULL_BARRIER();
 
-		if (!F_ISSET(btree, WT_BTREE_NO_EVICTION)) {
-			WT_ERR(__wt_evict_file_exclusive_on(session));
+		WT_ERR(__wt_evict_file_exclusive_on(session, &evict_reset));
+		if (evict_reset)
 			__wt_evict_file_exclusive_off(session);
-		}
 
 		/* Write all dirty in-cache pages. */
 		flags |= WT_READ_NO_EVICT;
diff --git a/src/conn/conn_dhandle.c b/src/conn/conn_dhandle.c
index 8ed656d6416..a5512352f2c 100644
--- a/src/conn/conn_dhandle.c
+++ b/src/conn/conn_dhandle.c
@@ -475,16 +475,15 @@ __conn_btree_apply_internal(WT_SESSION_IMPL *session, WT_DATA_HANDLE *dhandle,
 	WT_DECL_RET;
 
 	/*
-	 * We need to pull the handle into the session handle
-	 * cache and make sure it's referenced to stop other
-	 * internal code dropping the handle (e.g in LSM when
-	 * cleaning up obsolete chunks). Holding the metadata
-	 * lock isn't enough.
+	 * We need to pull the handle into the session handle cache and make
+	 * sure it's referenced to stop other internal code dropping the handle
+	 * (e.g in LSM when cleaning up obsolete chunks).
 	 */
 	ret = __wt_session_get_btree(session,
 	    dhandle->name, dhandle->checkpoint, NULL, 0);
 	if (ret == 0) {
-		ret = func(session, cfg);
+		WT_SAVE_DHANDLE(session,
+		    ret = func(session, cfg));
 		if (WT_META_TRACKING(session))
 			WT_TRET(__wt_meta_track_handle_lock(session, 0));
 		else
@@ -550,12 +549,11 @@ __wt_conn_btree_apply_single(WT_SESSION_IMPL *session,
     int (*func)(WT_SESSION_IMPL *, const char *[]), const char *cfg[])
 {
 	WT_CONNECTION_IMPL *conn;
-	WT_DATA_HANDLE *dhandle, *saved_dhandle;
+	WT_DATA_HANDLE *dhandle;
 	WT_DECL_RET;
 	uint64_t bucket, hash;
 
 	conn = S2C(session);
-	saved_dhandle = session->dhandle;
 
 	WT_ASSERT(session, F_ISSET(session, WT_SESSION_HANDLE_LIST_LOCKED));
 
@@ -578,15 +576,14 @@ __wt_conn_btree_apply_single(WT_SESSION_IMPL *session,
 			 */
 			__wt_spin_lock(session, &dhandle->close_lock);
 			if (F_ISSET(dhandle, WT_DHANDLE_OPEN)) {
-				session->dhandle = dhandle;
-				ret = func(session, cfg);
+				WT_WITH_DHANDLE(session, dhandle,
+				    ret = func(session, cfg));
 			}
 			__wt_spin_unlock(session, &dhandle->close_lock);
 			WT_ERR(ret);
 		}
 
-err:	session->dhandle = saved_dhandle;
-	return (ret);
+err:	return (ret);
 }
 
 /*
diff --git a/src/conn/conn_stat.c b/src/conn/conn_stat.c
index be2172ea8d8..67814dc330b 100644
--- a/src/conn/conn_stat.c
+++ b/src/conn/conn_stat.c
@@ -193,6 +193,7 @@ static int
 __statlog_apply(WT_SESSION_IMPL *session, const char *cfg[])
 {
 	WT_DATA_HANDLE *dhandle;
+	WT_DECL_RET;
 	char **p;
 
 	WT_UNUSED(cfg);
@@ -201,8 +202,11 @@ __statlog_apply(WT_SESSION_IMPL *session, const char *cfg[])
 
 	/* Check for a match on the set of sources. */
 	for (p = S2C(session)->stat_sources; *p != NULL; ++p)
-		if (WT_PREFIX_MATCH(dhandle->name, *p))
-			return (__statlog_dump(session, dhandle->name, 0));
+		if (WT_PREFIX_MATCH(dhandle->name, *p)) {
+			WT_WITHOUT_DHANDLE(session,
+			    ret = __statlog_dump(session, dhandle->name, 0));
+			WT_RET(ret);
+		}
 	return (0);
 }
 
diff --git a/src/evict/evict_file.c b/src/evict/evict_file.c
index 910aef070ca..9e39fcc7a2c 100644
--- a/src/evict/evict_file.c
+++ b/src/evict/evict_file.c
@@ -15,21 +15,16 @@
 int
 __wt_evict_file(WT_SESSION_IMPL *session, int syncop)
 {
-	WT_BTREE *btree;
 	WT_DECL_RET;
 	WT_PAGE *page;
 	WT_REF *next_ref, *ref;
-	int eviction_enabled;
-
-	btree = S2BT(session);
-	eviction_enabled = !F_ISSET(btree, WT_BTREE_NO_EVICTION);
+	int evict_reset;
 
 	/*
 	 * We need exclusive access to the file -- disable ordinary eviction
 	 * and drain any blocks already queued.
 	 */
-	if (eviction_enabled)
-		WT_RET(__wt_evict_file_exclusive_on(session));
+	WT_RET(__wt_evict_file_exclusive_on(session, &evict_reset));
 
 	/* Make sure the oldest transaction ID is up-to-date. */
 	__wt_txn_update_oldest(session);
@@ -140,7 +135,7 @@ err:		/* On error, clear any left-over tree walk. */
 			    session, next_ref, WT_READ_NO_EVICT));
 	}
 
-	if (eviction_enabled)
+	if (evict_reset)
 		__wt_evict_file_exclusive_off(session);
 
 	return (ret);
diff --git a/src/evict/evict_lru.c b/src/evict/evict_lru.c
index 83a9aa5c8c5..35be3df9185 100644
--- a/src/evict/evict_lru.c
+++ b/src/evict/evict_lru.c
@@ -653,7 +653,7 @@ __wt_evict_page(WT_SESSION_IMPL *session, WT_REF *ref)
  *	blocks queued for eviction.
  */
 int
-__wt_evict_file_exclusive_on(WT_SESSION_IMPL *session)
+__wt_evict_file_exclusive_on(WT_SESSION_IMPL *session, int *evict_resetp)
 {
 	WT_BTREE *btree;
 	WT_CACHE *cache;
@@ -664,6 +664,15 @@ __wt_evict_file_exclusive_on(WT_SESSION_IMPL *session)
 	cache = S2C(session)->cache;
 
 	/*
+	 * If the file isn't evictable, there's no work to do.
+	 */
+	if (F_ISSET(btree, WT_BTREE_NO_EVICTION)) {
+		*evict_resetp = 0;
+		return (0);
+	}
+	*evict_resetp = 1;
+
+	/*
 	 * Hold the walk lock to set the "no eviction" flag: no new pages from
 	 * the file will be queued for eviction after this point.
 	 */
@@ -979,7 +988,7 @@ retry:	while (slot < max_entries && ret == 0) {
 		 * exclusive access when a handle is being closed.
 		 */
 		if (!F_ISSET(btree, WT_BTREE_NO_EVICTION)) {
-			WT_WITH_BTREE(session, btree,
+			WT_WITH_DHANDLE(session, dhandle,
 			    ret = __evict_walk_file(session, &slot, flags));
 			WT_ASSERT(session, session->split_gen == 0);
 		}
@@ -1072,34 +1081,37 @@ __evict_walk_file(WT_SESSION_IMPL *session, u_int *slotp, uint32_t flags)
 	WT_PAGE_MODIFY *mod;
 	uint64_t pages_walked;
 	uint32_t walk_flags;
-	int internal_pages, modified, restarts;
+	int enough, internal_pages, modified, restarts;
 
 	btree = S2BT(session);
 	cache = S2C(session)->cache;
 	start = cache->evict + *slotp;
 	end = WT_MIN(start + WT_EVICT_WALK_PER_FILE,
 	    cache->evict + cache->evict_slots);
+	enough = internal_pages = restarts = 0;
 
 	walk_flags =
 	    WT_READ_CACHE | WT_READ_NO_EVICT | WT_READ_NO_GEN | WT_READ_NO_WAIT;
 
 	/*
 	 * Get some more eviction candidate pages.
+	 *
+	 * !!! Take care terminating this loop.
+	 *
+	 * Don't make an extra call to __wt_tree_walk after we hit the end of a
+	 * tree: that will leave a page pinned, which may prevent any work from
+	 * being done.
+	 *
+	 * Once we hit the page limit, do one more step through the walk in
+	 * case we are appending and only the last page in the file is live.
 	 */
-	for (evict = start, pages_walked = 0, internal_pages = restarts = 0;
-	    evict < end && pages_walked < WT_EVICT_MAX_PER_FILE &&
-	    (ret == 0 || ret == WT_NOTFOUND);
+	for (evict = start, pages_walked = 0;
+	    evict < end && !enough && (ret == 0 || ret == WT_NOTFOUND);
 	    ret = __wt_tree_walk(
 	    session, &btree->evict_ref, &pages_walked, walk_flags)) {
+		enough = (pages_walked > WT_EVICT_MAX_PER_FILE);
 		if (btree->evict_ref == NULL) {
-			/*
-			 * Take care with terminating this loop.
-			 *
-			 * Don't make an extra call to __wt_tree_walk: that will
-			 * leave a page pinned, which may prevent any work from
-			 * being done.
-			 */
-			if (++restarts == 2)
+			if (++restarts == 2 || enough)
 				break;
 			continue;
 		}
diff --git a/src/include/dhandle.h b/src/include/dhandle.h
index 423bb84d00d..300e8e735b9 100644
--- a/src/include/dhandle.h
+++ b/src/include/dhandle.h
@@ -6,6 +6,10 @@
  * See the file LICENSE for redistribution information.
  */
 
+/*
+ * Helpers for calling a function with a data handle in session->dhandle
+ * then restoring afterwards.
+ */
 #define	WT_WITH_DHANDLE(s, d, e) do {					\
 	WT_DATA_HANDLE *__saved_dhandle = (s)->dhandle;			\
 	(s)->dhandle = (d);						\
@@ -15,6 +19,15 @@
 
 #define	WT_WITH_BTREE(s, b, e)	WT_WITH_DHANDLE(s, (b)->dhandle, e)
 
+/* Call a function without the caller's data handle, restore afterwards. */
+#define	WT_WITHOUT_DHANDLE(s, e) WT_WITH_DHANDLE(s, NULL, e)
+
+/*
+ * Call a function with the caller's data handle, restore it afterwards in case
+ * it is overwritten.
+ */
+#define	WT_SAVE_DHANDLE(s, e) WT_WITH_DHANDLE(s, (s)->dhandle, e)
+
 /*
  * WT_DATA_HANDLE --
  *	A handle for a generic named data source.
diff --git a/src/include/extern.h b/src/include/extern.h
index 23bb36623e5..5d3ee5bc8f8 100644
--- a/src/include/extern.h
+++ b/src/include/extern.h
@@ -297,7 +297,7 @@ extern int __wt_evict_server_wake(WT_SESSION_IMPL *session);
 extern int __wt_evict_create(WT_SESSION_IMPL *session);
 extern int __wt_evict_destroy(WT_SESSION_IMPL *session);
 extern int __wt_evict_page(WT_SESSION_IMPL *session, WT_REF *ref);
-extern int __wt_evict_file_exclusive_on(WT_SESSION_IMPL *session);
+extern int __wt_evict_file_exclusive_on(WT_SESSION_IMPL *session, int *evict_resetp);
 extern void __wt_evict_file_exclusive_off(WT_SESSION_IMPL *session);
 extern int __wt_evict_lru_page(WT_SESSION_IMPL *session, int is_server);
 extern int __wt_cache_wait(WT_SESSION_IMPL *session, int full);
diff --git a/src/log/log.c b/src/log/log.c
index f6c8602faff..f76ec402b0d 100644
--- a/src/log/log.c
+++ b/src/log/log.c
@@ -50,27 +50,29 @@ __wt_log_needs_recovery(WT_SESSION_IMPL *session, WT_LSN *ckp_lsn, int *rec)
 
 	conn = S2C(session);
 	log = conn->log;
-	c = NULL;
+
 	/*
-	 * Default is to run recovery always.
+	 * Default is to run recovery always (regardless of whether this
+	 * connection has logging enabled).
 	 */
 	*rec = 1;
-
 	if (log == NULL)
 		return (0);
+
 	WT_RET(__wt_curlog_open(session, "log:", NULL, &c));
 	c->set_key(c, ckp_lsn->file, ckp_lsn->offset, 0);
 	WT_ERR(c->search(c));
+
 	/*
-	 * If the checkpoint LSN we're given is the last record,
-	 * then recovery is not needed.
+	 * If the checkpoint LSN we're given is the last record, then recovery
+	 * is not needed.
 	 */
 	if ((ret = c->next(c)) == WT_NOTFOUND) {
 		*rec = 0;
 		ret = 0;
 	}
-err:	if (c != NULL)
-		(void)c->close(c);
+
+err:	WT_TRET(c->close(c));
 	return (ret);
 }
 
diff --git a/src/log/log_slot.c b/src/log/log_slot.c
index ff61afb698c..8dcb2f9f165 100644
--- a/src/log/log_slot.c
+++ b/src/log/log_slot.c
@@ -203,8 +203,7 @@ retry:
 		 * churn is used to change how long we pause before closing
 		 * the slot - which leads to more consolidation and less churn.
 		 */
-		if (++switch_fails % SLOT_POOL == 0 &&
-		    switch_fails != 0 && slot->slot_churn < 5)
+		if (++switch_fails % SLOT_POOL == 0 && slot->slot_churn < 5)
 			++slot->slot_churn;
 		__wt_yield();
 		goto retry;
diff --git a/src/meta/meta_apply.c b/src/meta/meta_apply.c
index 8ef5cc38db4..6d08ce3aa6a 100644
--- a/src/meta/meta_apply.c
+++ b/src/meta/meta_apply.c
@@ -43,7 +43,8 @@ __wt_meta_btree_apply(WT_SESSION_IMPL *session,
 		 */
 		ret = __wt_session_get_btree(session, uri, NULL, NULL, 0);
 		if (ret == 0) {
-			ret = func(session, cfg);
+			WT_SAVE_DHANDLE(session,
+			    ret = func(session, cfg));
 			if (WT_META_TRACKING(session))
 				WT_TRET(
 				    __wt_meta_track_handle_lock(session, 0));
diff --git a/src/schema/schema_truncate.c b/src/schema/schema_truncate.c
index 383e860cc7b..1eb76226aad 100644
--- a/src/schema/schema_truncate.c
+++ b/src/schema/schema_truncate.c
@@ -170,11 +170,15 @@ __wt_schema_range_truncate(
 	cursor = (start != NULL) ? start : stop;
 	uri = cursor->internal_uri;
 
-	if (WT_PREFIX_MATCH(uri, "file:"))
+	if (WT_PREFIX_MATCH(uri, "file:")) {
+		if (start != NULL)
+			WT_CURSOR_NEEDKEY(start);
+		if (stop != NULL)
+			WT_CURSOR_NEEDKEY(stop);
 		WT_WITH_BTREE(session, ((WT_CURSOR_BTREE *)cursor)->btree,
 		    ret = __wt_btcur_range_truncate(
 			(WT_CURSOR_BTREE *)start, (WT_CURSOR_BTREE *)stop));
-	else if (WT_PREFIX_MATCH(uri, "table:"))
+	} else if (WT_PREFIX_MATCH(uri, "table:"))
 		ret = __wt_table_range_truncate(
 		    (WT_CURSOR_TABLE *)start, (WT_CURSOR_TABLE *)stop);
 	else if ((dsrc = __wt_schema_get_source(session, uri)) != NULL &&
@@ -182,6 +186,6 @@ __wt_schema_range_truncate(
 		ret = dsrc->range_truncate(dsrc, &session->iface, start, stop);
 	else
 		ret = __wt_range_truncate(start, stop);
-
+err:
 	return (ret);
 }
diff --git a/src/schema/schema_worker.c b/src/schema/schema_worker.c
index 94eb3170175..3dfd068cf9c 100644
--- a/src/schema/schema_worker.c
+++ b/src/schema/schema_worker.c
@@ -57,7 +57,8 @@ __wt_schema_worker(WT_SESSION_IMPL *session,
 
 			WT_ERR(__wt_session_get_btree_ckpt(
 			    session, uri, cfg, open_flags));
-			ret = file_func(session, cfg);
+			WT_SAVE_DHANDLE(session,
+			    ret = file_func(session, cfg));
 			WT_TRET(__wt_session_release_btree(session));
 		}
 	} else if (WT_PREFIX_MATCH(uri, "colgroup:")) {
diff --git a/src/session/session_api.c b/src/session/session_api.c
index e54553aa071..e95bea3f75b 100644
--- a/src/session/session_api.c
+++ b/src/session/session_api.c
@@ -570,7 +570,9 @@ __session_truncate(WT_SESSION *wt_session,
 	WT_DECL_RET;
 	WT_SESSION_IMPL *session;
 	WT_CURSOR *cursor;
-	int cmp;
+	int cmp, local_start;
+
+	local_start = 0;
 
 	session = (WT_SESSION_IMPL *)wt_session;
 	SESSION_TXN_API_CALL(session, truncate, config, cfg);
@@ -642,9 +644,7 @@ __session_truncate(WT_SESSION *wt_session,
 	 * what records currently appear in the object.  For this reason, do a
 	 * search-near, rather than a search.  Additionally, we have to correct
 	 * after calling search-near, to position the start/stop cursors on the
-	 * next record greater than/less than the original key.  If the cursors
-	 * hit the beginning/end of the object, or the start/stop keys cross,
-	 * we're done, the range must be empty.
+	 * next record greater than/less than the original key.
 	 */
 	if (start != NULL) {
 		WT_ERR(start->search_near(start, &cmp));
@@ -659,12 +659,28 @@ __session_truncate(WT_SESSION *wt_session,
 			WT_ERR_NOTFOUND_OK(ret);
 			goto done;
 		}
+	}
 
-		if (start != NULL) {
-			WT_ERR(start->compare(start, stop, &cmp));
-			if (cmp > 0)
-				goto done;
-		}
+	/*
+	 * We always truncate in the forward direction because the underlying
+	 * data structures can move through pages faster forward than backward.
+	 * If we don't have a start cursor, create one and position it at the
+	 * first record.
+	 */
+	if (start == NULL) {
+		WT_ERR(__session_open_cursor(
+		    wt_session, stop->uri, NULL, NULL, &start));
+		local_start = 1;
+		WT_ERR(start->next(start));
+	}
+
+	/*
+	 * If the start/stop keys cross, we're done, the range must be empty.
+	 */
+	if (stop != NULL) {
+		WT_ERR(start->compare(start, stop, &cmp));
+		if (cmp > 0)
+			goto done;
 	}
 
 	WT_ERR(__wt_schema_range_truncate(session, start, stop));
@@ -673,6 +689,12 @@ done:
 err:	TXN_API_END_RETRY(session, ret, 0);
 
 	/*
+	 * Close any locally-opened start cursor.
+	 */
+	if (local_start)
+		WT_TRET(start->close(start));
+
+	/*
 	 * Only map WT_NOTFOUND to ENOENT if a URI was specified.
 	 */
 	return (ret == WT_NOTFOUND && uri != NULL ? ENOENT : ret);
diff --git a/src/txn/txn_ckpt.c b/src/txn/txn_ckpt.c
index eae21d0b9f5..fb590e1a297 100644
--- a/src/txn/txn_ckpt.c
+++ b/src/txn/txn_ckpt.c
@@ -237,7 +237,6 @@ __checkpoint_data_source(WT_SESSION_IMPL *session, const char *cfg[])
 int
 __wt_checkpoint_list(WT_SESSION_IMPL *session, const char *cfg[])
 {
-	WT_DATA_HANDLE *saved_dhandle;
 	WT_DECL_RET;
 	const char *name;
 
@@ -253,7 +252,6 @@ __wt_checkpoint_list(WT_SESSION_IMPL *session, const char *cfg[])
 
 	/* Not strictly necessary, but cleaner to clear the current handle. */
 	name = session->dhandle->name;
-	saved_dhandle = session->dhandle;
 	session->dhandle = NULL;
 
 	/* Record busy file names, we'll deal with them in the checkpoint. */
@@ -264,8 +262,7 @@ __wt_checkpoint_list(WT_SESSION_IMPL *session, const char *cfg[])
 		WT_ERR(__wt_strdup(session, name,
 		    &session->ckpt_handle[session->ckpt_handle_next++].name));
 
-err:	session->dhandle = saved_dhandle;
-	return (ret);
+err:	return (ret);
 }
 
 /*
diff --git a/src/utilities/util_main.c b/src/utilities/util_main.c
index 3274f3a0fd0..51a11f82c43 100644
--- a/src/utilities/util_main.c
+++ b/src/utilities/util_main.c
@@ -169,25 +169,21 @@ main(int argc, char *argv[])
 	if (func == NULL)
 		return (usage());
 
-	/* Build the configuration string, as necessary. */
-	if (cmd_config != NULL || rec_config != NULL) {
-		len = 10;		/* some slop */
-		if (config != NULL)
-			len += strlen(config);
-		if (cmd_config != NULL)
-			len += strlen(cmd_config);
-		if (rec_config != NULL)
-			len += strlen(rec_config);
-		if ((p = malloc(len)) == NULL) {
-			ret = util_err(errno, NULL);
-			goto err;
-		}
-		(void)snprintf(p, len, "%s,%s,%s",
-		    config == NULL ? "" : config,
-		    cmd_config == NULL ? "" : cmd_config,
-		    rec_config == NULL ? "" : rec_config);
-		config = p;
+	/* Build the configuration string. */
+	len = 10;					/* some slop */
+	if (config != NULL)
+		len += strlen(config);
+	if (cmd_config != NULL)
+		len += strlen(cmd_config);
+	len += strlen(rec_config);
+	if ((p = malloc(len)) == NULL) {
+		ret = util_err(errno, NULL);
+		goto err;
 	}
+	(void)snprintf(p, len, "%s,%s,%s",
+	    config == NULL ? "" : config,
+	    cmd_config == NULL ? "" : cmd_config, rec_config);
+	config = p;
 
 	/* Open the database and a session. */
 	if ((ret = wiredtiger_open(home,