Merge branch 'develop' into mongodb-3.0

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));

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;

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. */

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;

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);
 }
 
 /*

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);
 }
 

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 @@ __wt_evict_file(WT_SESSION_IMPL *session, int syncop)
 			    session, next_ref, WT_READ_NO_EVICT));
 	}
 
-	if (eviction_enabled)
+	if (evict_reset)
 		__wt_evict_file_exclusive_off(session);
 
 	return (ret);