HDFS-4845. FSNamesystem.deleteInternal should acquire write-lock before changing the inode map. Contributed by Arpit Agarwal

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1492941 13f79535-47bb-0310-9956-ffa450edef68

hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt

@@ -571,6 +571,9 @@ Release 2.1.0-beta - UNRELEASED
     HDFS-4586. TestDataDirs.testGetDataDirsFromURIs fails with all directories
     in dfs.datanode.data.dir are invalid. (Ivan Mitic via atm)
 
+    HDFS-4845. FSNamesystem.deleteInternal should acquire write-lock before
+    changing the inode map.  (Arpit Agarwal via szetszwo)
+
   BREAKDOWN OF HDFS-347 SUBTASKS AND RELATED JIRAS
 
     HDFS-4353. Encapsulate connections to peers in Peer and PeerServer classes.

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java

@@ -2933,7 +2933,12 @@ public class FSNamesystem implements Namesystem, FSClusterStats,
     getEditLog().logSync(); 
     removeBlocks(collectedBlocks); // Incremental deletion of blocks
     collectedBlocks.clear();
-    dir.removeFromInodeMap(removedINodes);
+    dir.writeLock();
+    try {
+      dir.removeFromInodeMap(removedINodes);
+    } finally {
+      dir.writeUnlock();
+    }
     removedINodes.clear();
     if (NameNode.stateChangeLog.isDebugEnabled()) {
       NameNode.stateChangeLog.debug("DIR* Namesystem.delete: "