From 85bd8e0b728408185a5717361bdd757f46a4709e Mon Sep 17 00:00:00 2001
From: Eli Collins <eli@apache.org>
Date: Tue, 16 Oct 2012 02:25:25 +0000
Subject: [PATCH] HDFS-4058. DirectoryScanner may fail with IOOB if the
 directory scanning threads return out of volume order. Contributed by Eli
 Collins

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1398617 13f79535-47bb-0310-9956-ffa450edef68
---
 hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt    |  3 +++
 .../hdfs/server/datanode/DirectoryScanner.java | 18 +++++++++---------
 2 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt b/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt
index 5ab7edeaa68..b4b720b6d02 100644
--- a/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt
+++ b/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt
@@ -114,6 +114,9 @@ Release 2.0.3-alpha - Unreleased
 
     HDFS-3678. Edit log files are never being purged from 2NN. (atm)
 
+    HDFS-4058. DirectoryScanner may fail with IOOB if the directory
+    scanning threads return out of volume order. (eli)
+
 Release 2.0.2-alpha - 2012-09-07 
   
   INCOMPATIBLE CHANGES
diff --git a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DirectoryScanner.java b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DirectoryScanner.java
index 004af654e63..5d870d771e8 100644
--- a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DirectoryScanner.java
+++ b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DirectoryScanner.java
@@ -431,16 +431,16 @@ public class DirectoryScanner implements Runnable {
   private Map<String, ScanInfo[]> getDiskReport() {
     // First get list of data directories
     final List<? extends FsVolumeSpi> volumes = dataset.getVolumes();
-    ArrayList<ScanInfoPerBlockPool> dirReports =
-      new ArrayList<ScanInfoPerBlockPool>(volumes.size());
-    
+
+    // Use an array since the threads may return out of order and
+    // compilersInProgress#keySet may return out of order as well.
+    ScanInfoPerBlockPool[] dirReports = new ScanInfoPerBlockPool[volumes.size()];
+
     Map<Integer, Future<ScanInfoPerBlockPool>> compilersInProgress =
       new HashMap<Integer, Future<ScanInfoPerBlockPool>>();
+
     for (int i = 0; i < volumes.size(); i++) {
-      if (!isValid(dataset, volumes.get(i))) {
-        // volume is invalid
-        dirReports.add(i, null);
-      } else {
+      if (isValid(dataset, volumes.get(i))) {
         ReportCompiler reportCompiler =
           new ReportCompiler(volumes.get(i));
         Future<ScanInfoPerBlockPool> result = 
@@ -452,7 +452,7 @@ public class DirectoryScanner implements Runnable {
     for (Entry<Integer, Future<ScanInfoPerBlockPool>> report :
         compilersInProgress.entrySet()) {
       try {
-        dirReports.add(report.getKey(), report.getValue().get());
+        dirReports[report.getKey()] = report.getValue().get();
       } catch (Exception ex) {
         LOG.error("Error compiling report", ex);
         // Propagate ex to DataBlockScanner to deal with
@@ -465,7 +465,7 @@ public class DirectoryScanner implements Runnable {
     for (int i = 0; i < volumes.size(); i++) {
       if (isValid(dataset, volumes.get(i))) {
         // volume is still valid
-        list.addAll(dirReports.get(i));
+        list.addAll(dirReports[i]);
       }
     }