HDFS-6097. Zero-copy reads are incorrectly disabled on file offsets above 2GB (cmccabe)

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1577355 13f79535-47bb-0310-9956-ffa450edef68

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

@@ -370,6 +370,9 @@ Release 2.4.0 - UNRELEASED
     HDFS-5244. TestNNStorageRetentionManager#testPurgeMultipleDirs fails.
     (Jinghui Wang via suresh)
 
+    HDFS-6097. zero-copy reads are incorrectly disabled on file offsets above
+    2GB (cmccabe)
+
   BREAKDOWN OF HDFS-5698 SUBTASKS AND RELATED JIRAS
 
     HDFS-5717. Save FSImage header in protobuf. (Haohui Mai via jing9)

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/DFSInputStream.java

@@ -1601,28 +1601,63 @@ implements ByteBufferReadable, CanSetDropBehind, CanSetReadahead,
 
   private synchronized ByteBuffer tryReadZeroCopy(int maxLength,
       EnumSet<ReadOption> opts) throws IOException {
-    // Java ByteBuffers can't be longer than 2 GB, because they use
-    // 4-byte signed integers to represent capacity, etc.
-    // So we can't mmap the parts of the block higher than the 2 GB offset.
-    // FIXME: we could work around this with multiple memory maps.
-    // See HDFS-5101.
-    long blockEnd32 = Math.min(Integer.MAX_VALUE, blockEnd);
-    long curPos = pos;
-    long blockLeft = blockEnd32 - curPos + 1;
-    if (blockLeft <= 0) {
+    // Copy 'pos' and 'blockEnd' to local variables to make it easier for the
+    // JVM to optimize this function.
+    final long curPos = pos;
+    final long curEnd = blockEnd;
+    final long blockStartInFile = currentLocatedBlock.getStartOffset();
+    final long blockPos = curPos - blockStartInFile;
+
+    // Shorten this read if the end of the block is nearby.
+    long length63;
+    if ((curPos + maxLength) <= (curEnd + 1)) {
+      length63 = maxLength;
+    } else {
+      length63 = 1 + curEnd - curPos;
+      if (length63 <= 0) {
+        if (DFSClient.LOG.isDebugEnabled()) {
+          DFSClient.LOG.debug("Unable to perform a zero-copy read from offset " +
+            curPos + " of " + src + "; " + length63 + " bytes left in block.  " +
+            "blockPos=" + blockPos + "; curPos=" + curPos +
+            "; curEnd=" + curEnd);
+        }
+        return null;
+      }
       if (DFSClient.LOG.isDebugEnabled()) {
-        DFSClient.LOG.debug("unable to perform a zero-copy read from offset " +
-          curPos + " of " + src + "; blockLeft = " + blockLeft +
-          "; blockEnd32 = " + blockEnd32 + ", blockEnd = " + blockEnd +
-          "; maxLength = " + maxLength);
+        DFSClient.LOG.debug("Reducing read length from " + maxLength +
+            " to " + length63 + " to avoid going more than one byte " +
+            "past the end of the block.  blockPos=" + blockPos +
+            "; curPos=" + curPos + "; curEnd=" + curEnd);
+      }
+    }
+    // Make sure that don't go beyond 31-bit offsets in the MappedByteBuffer.
+    int length;
+    if (blockPos + length63 <= Integer.MAX_VALUE) {
+      length = (int)length63;
+    } else {
+      long length31 = Integer.MAX_VALUE - blockPos;
+      if (length31 <= 0) {
+        // Java ByteBuffers can't be longer than 2 GB, because they use
+        // 4-byte signed integers to represent capacity, etc.
+        // So we can't mmap the parts of the block higher than the 2 GB offset.
+        // FIXME: we could work around this with multiple memory maps.
+        // See HDFS-5101.
+        if (DFSClient.LOG.isDebugEnabled()) {
+          DFSClient.LOG.debug("Unable to perform a zero-copy read from offset " +
+            curPos + " of " + src + "; 31-bit MappedByteBuffer limit " +
+            "exceeded.  blockPos=" + blockPos + ", curEnd=" + curEnd);
+        }
+        return null;
+      }
+      length = (int)length31;
+      if (DFSClient.LOG.isDebugEnabled()) {
+        DFSClient.LOG.debug("Reducing read length from " + maxLength +
+            " to " + length + " to avoid 31-bit limit.  " +
+            "blockPos=" + blockPos + "; curPos=" + curPos +
+            "; curEnd=" + curEnd);
       }
-      return null;
     }
-    int length = Math.min((int)blockLeft, maxLength);
-    long blockStartInFile = currentLocatedBlock.getStartOffset();
-    long blockPos = curPos - blockStartInFile;
-    long limit = blockPos + length;
-    ClientMmap clientMmap = blockReader.getClientMmap(opts);
+    final ClientMmap clientMmap = blockReader.getClientMmap(opts);
     if (clientMmap == null) {
       if (DFSClient.LOG.isDebugEnabled()) {
         DFSClient.LOG.debug("unable to perform a zero-copy read from offset " +
@@ -1634,16 +1669,16 @@ implements ByteBufferReadable, CanSetDropBehind, CanSetReadahead,
     boolean success = false;
     ByteBuffer buffer;
     try {
-      seek(pos + length);
+      seek(curPos + length);
       buffer = clientMmap.getMappedByteBuffer().asReadOnlyBuffer();
       buffer.position((int)blockPos);
-      buffer.limit((int)limit);
+      buffer.limit((int)(blockPos + length));
       extendedReadBuffers.put(buffer, clientMmap);
       readStatistics.addZeroCopyBytes(length);
       if (DFSClient.LOG.isDebugEnabled()) {
-        DFSClient.LOG.debug("readZeroCopy read " + maxLength + " bytes from " +
-            "offset " + curPos + " via the zero-copy read path.  " +
-            "blockEnd = " + blockEnd);
+        DFSClient.LOG.debug("readZeroCopy read " + length + 
+            " bytes from offset " + curPos + " via the zero-copy read " +
+            "path.  blockEnd = " + blockEnd);
       }
       success = true;
     } finally {

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/client/ShortCircuitReplica.java

@@ -285,7 +285,8 @@ public class ShortCircuitReplica {
   MappedByteBuffer loadMmapInternal() {
     try {
       FileChannel channel = dataStream.getChannel();
-      MappedByteBuffer mmap = channel.map(MapMode.READ_ONLY, 0, channel.size());
+      MappedByteBuffer mmap = channel.map(MapMode.READ_ONLY, 0, 
+          Math.min(Integer.MAX_VALUE, channel.size()));
       if (LOG.isTraceEnabled()) {
         LOG.trace(this + ": created mmap of size " + channel.size());
       }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/fs/TestEnhancedByteBufferAccess.java

@@ -776,4 +776,80 @@ public class TestEnhancedByteBufferAccess {
       if (cluster != null) cluster.shutdown();
     }
   }
+  
+  @Test
+  public void test2GBMmapLimit() throws Exception {
+    Assume.assumeTrue(BlockReaderTestUtil.shouldTestLargeFiles());
+    HdfsConfiguration conf = initZeroCopyTest();
+    final long TEST_FILE_LENGTH = 2469605888L;
+    conf.set(DFSConfigKeys.DFS_CHECKSUM_TYPE_KEY, "NULL");
+    conf.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, TEST_FILE_LENGTH);
+    MiniDFSCluster cluster = null;
+    final Path TEST_PATH = new Path("/a");
+    final String CONTEXT = "test2GBMmapLimit";
+    conf.set(DFSConfigKeys.DFS_CLIENT_CONTEXT, CONTEXT);
+
+    FSDataInputStream fsIn = null, fsIn2 = null;
+    ByteBuffer buf1 = null, buf2 = null;
+    try {
+      cluster = new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
+      cluster.waitActive();
+      DistributedFileSystem fs = cluster.getFileSystem();
+      DFSTestUtil.createFile(fs, TEST_PATH, TEST_FILE_LENGTH, (short)1, 0xB);
+      DFSTestUtil.waitReplication(fs, TEST_PATH, (short)1);
+      
+      fsIn = fs.open(TEST_PATH);
+      buf1 = fsIn.read(null, 1, EnumSet.of(ReadOption.SKIP_CHECKSUMS));
+      Assert.assertEquals(1, buf1.remaining());
+      fsIn.releaseBuffer(buf1);
+      buf1 = null;
+      fsIn.seek(2147483640L);
+      buf1 = fsIn.read(null, 1024, EnumSet.of(ReadOption.SKIP_CHECKSUMS));
+      Assert.assertEquals(7, buf1.remaining());
+      Assert.assertEquals(Integer.MAX_VALUE, buf1.limit());
+      fsIn.releaseBuffer(buf1);
+      buf1 = null;
+      Assert.assertEquals(2147483647L, fsIn.getPos());
+      try {
+        buf1 = fsIn.read(null, 1024,
+            EnumSet.of(ReadOption.SKIP_CHECKSUMS));
+        Assert.fail("expected UnsupportedOperationException");
+      } catch (UnsupportedOperationException e) {
+        // expected; can't read past 2GB boundary.
+      }
+      fsIn.close();
+      fsIn = null;
+
+      // Now create another file with normal-sized blocks, and verify we
+      // can read past 2GB
+      final Path TEST_PATH2 = new Path("/b");
+      conf.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, 268435456L);
+      DFSTestUtil.createFile(fs, TEST_PATH2, 1024 * 1024, TEST_FILE_LENGTH,
+          268435456L, (short)1, 0xA);
+      
+      fsIn2 = fs.open(TEST_PATH2);
+      fsIn2.seek(2147483640L);
+      buf2 = fsIn2.read(null, 1024, EnumSet.of(ReadOption.SKIP_CHECKSUMS));
+      Assert.assertEquals(8, buf2.remaining());
+      Assert.assertEquals(2147483648L, fsIn2.getPos());
+      fsIn2.releaseBuffer(buf2);
+      buf2 = null;
+      buf2 = fsIn2.read(null, 1024, EnumSet.of(ReadOption.SKIP_CHECKSUMS));
+      Assert.assertEquals(1024, buf2.remaining());
+      Assert.assertEquals(2147484672L, fsIn2.getPos());
+      fsIn2.releaseBuffer(buf2);
+      buf2 = null;
+    } finally {
+      if (buf1 != null) {
+        fsIn.releaseBuffer(buf1);
+      }
+      if (buf2 != null) {
+        fsIn2.releaseBuffer(buf2);
+      }
+      IOUtils.cleanup(null, fsIn, fsIn2);
+      if (cluster != null) {
+        cluster.shutdown();
+      }
+    }
+  }
 }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/BlockReaderTestUtil.java

@@ -54,6 +54,15 @@ import org.apache.log4j.LogManager;
  * A helper class to setup the cluster, and get to BlockReader and DataNode for a block.
  */
 public class BlockReaderTestUtil {
+  /**
+   * Returns true if we should run tests that generate large files (> 1GB)
+   */
+  static public boolean shouldTestLargeFiles() {
+    String property = System.getProperty("hdfs.test.large.files");
+    if (property == null) return false;
+    if (property.isEmpty()) return true;
+    return Boolean.parseBoolean(property);
+  }
 
   private HdfsConfiguration conf = null;
   private MiniDFSCluster cluster = null;