HDFS-2229. Fix a deadlock in namenode by enforcing lock acquisition ordering.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1156847 13f79535-47bb-0310-9956-ffa450edef68

hdfs/CHANGES.txt

@@ -954,6 +954,9 @@ Trunk (unreleased changes)
     HDFS-2245. Fix a NullPointerException in BlockManager.chooseTarget(..).
     (szetszwo)
 
+    HDFS-2229. Fix a deadlock in namenode by enforcing lock acquisition
+    ordering.  (szetszwo)
+
   BREAKDOWN OF HDFS-1073 SUBTASKS
 
     HDFS-1521. Persist transaction ID on disk between NN restarts.

hdfs/src/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java

@@ -1829,9 +1829,9 @@ private void invalidateCorruptReplicas(Block blk) {
    * over or under replicated. Place it into the respective queue.
    */
   public void processMisReplicatedBlocks() {
+    assert namesystem.hasWriteLock();
+
     long nrInvalid = 0, nrOverReplicated = 0, nrUnderReplicated = 0;
-    namesystem.writeLock();
-    try {
     neededReplications.clear();
     for (BlockInfo block : blocksMap.getBlocks()) {
       INodeFile fileINode = block.getINode();
@@ -1859,9 +1859,7 @@ public void processMisReplicatedBlocks() {
         processOverReplicatedBlock(block, expectedReplication, null, null);
       }
     }
-    } finally {
-      namesystem.writeUnlock();
-    }
+
     LOG.info("Total number of blocks            = " + blocksMap.size());
     LOG.info("Number of invalid blocks          = " + nrInvalid);
     LOG.info("Number of under-replicated blocks = " + nrUnderReplicated);

hdfs/src/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java

@@ -313,13 +313,18 @@ void activateSecretManager() throws IOException {
    * Activate FSNamesystem daemons.
    */
   void activate(Configuration conf) throws IOException {
+    writeLock();
+    try {
       setBlockTotal();
       blockManager.activate(conf);
+
       this.lmthread = new Daemon(leaseManager.new Monitor());
       lmthread.start();
-
       this.nnrmthread = new Daemon(new NameNodeResourceMonitor());
       nnrmthread.start();
+    } finally {
+      writeUnlock();
+    }
     
     registerMXBean();
     DefaultMetricsSystem.instance().register(this);