HDFS-7385. ThreadLocal used in FSEditLog class causes FSImage permission mess up. Contributed by jiangyu.

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

@@ -1446,6 +1446,9 @@ Release 2.6.0 - 2014-11-15
 
     HDFS-7391. Renable SSLv2Hello in HttpFS. (rkanter via acmurthy)
 
+    HDFS-7385. ThreadLocal used in FSEditLog class causes FSImage permission mess
+    up. (jiangyu via cnauroth)
+
 Release 2.5.2 - 2014-11-10
 
   INCOMPATIBLE CHANGES

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

@@ -708,6 +708,7 @@ public class FSEditLog implements LogsPurgeable {
     Preconditions.checkArgument(newNode.isUnderConstruction());
     PermissionStatus permissions = newNode.getPermissionStatus();
     AddOp op = AddOp.getInstance(cache.get())
+      .reset()
       .setInodeId(newNode.getId())
       .setPath(path)
       .setReplication(newNode.getFileReplication())
@@ -778,6 +779,7 @@ public class FSEditLog implements LogsPurgeable {
   public void logMkDir(String path, INode newNode) {
     PermissionStatus permissions = newNode.getPermissionStatus();
     MkdirOp op = MkdirOp.getInstance(cache.get())
+      .reset()
       .setInodeId(newNode.getId())
       .setPath(path)
       .setTimestamp(newNode.getModificationTime())

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

@@ -419,6 +419,12 @@ public abstract class FSEditLogOp {
       assert(opCode == OP_ADD || opCode == OP_CLOSE);
     }
     
+    <T extends AddCloseOp> T reset() {
+      this.aclEntries = null;
+      this.xAttrs = null;
+      return (T)this;
+    }
+
     <T extends AddCloseOp> T setInodeId(long inodeId) {
       this.inodeId = inodeId;
       return (T)this;
@@ -1410,6 +1416,12 @@ public abstract class FSEditLogOp {
       return (MkdirOp)cache.get(OP_MKDIR);
     }
 
+    MkdirOp reset() {
+      this.aclEntries = null;
+      this.xAttrs = null;
+      return this;
+    }
+
     MkdirOp setInodeId(long inodeId) {
       this.inodeId = inodeId;
       return this;

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestEditLog.java

@@ -17,9 +17,14 @@
  */
 package org.apache.hadoop.hdfs.server.namenode;
 
+import static org.apache.hadoop.fs.permission.AclEntryScope.*;
+import static org.apache.hadoop.fs.permission.AclEntryType.*;
+import static org.apache.hadoop.fs.permission.FsAction.*;
+import static org.apache.hadoop.hdfs.server.namenode.AclTestHelpers.*;
 import static org.apache.hadoop.test.MetricsAsserts.assertCounter;
 import static org.apache.hadoop.test.MetricsAsserts.getMetrics;
 import static org.junit.Assert.assertEquals;
+import static org.junit.Assert.assertFalse;
 import static org.junit.Assert.assertNotNull;
 import static org.junit.Assert.assertTrue;
 import static org.junit.Assert.fail;
@@ -57,6 +62,7 @@ import org.apache.hadoop.fs.ChecksumException;
 import org.apache.hadoop.fs.FileSystem;
 import org.apache.hadoop.fs.FileUtil;
 import org.apache.hadoop.fs.Path;
+import org.apache.hadoop.fs.permission.AclEntry;
 import org.apache.hadoop.fs.permission.FsPermission;
 import org.apache.hadoop.fs.permission.PermissionStatus;
 import org.apache.hadoop.hdfs.DFSConfigKeys;
@@ -1501,4 +1507,65 @@ public class TestEditLog {
     LOG.info(String.format("loaded %d edit log segments in %.2f seconds",
         NUM_EDIT_LOG_ROLLS, delta));
   }
+
+  /**
+   * Edit log op instances are cached internally using thread-local storage.
+   * This test checks that the cached instances are reset in between different
+   * transactions processed on the same thread, so that we don't accidentally
+   * apply incorrect attributes to an inode.
+   *
+   * @throws IOException if there is an I/O error
+   */
+  @Test
+  public void testResetThreadLocalCachedOps() throws IOException {
+    Configuration conf = new HdfsConfiguration();
+    conf.setBoolean(DFSConfigKeys.DFS_NAMENODE_ACLS_ENABLED_KEY, true);
+    // Set single handler thread, so all transactions hit same thread-local ops.
+    conf.setInt(DFSConfigKeys.DFS_NAMENODE_HANDLER_COUNT_KEY, 1);
+    MiniDFSCluster cluster = null;
+    FileSystem fileSys = null;
+    try {
+      cluster = new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
+      cluster.waitActive();
+      fileSys = cluster.getFileSystem();
+
+      // Create /dir1 with a default ACL.
+      Path dir1 = new Path("/dir1");
+      fileSys.mkdirs(dir1);
+      List<AclEntry> aclSpec = Lists.newArrayList(
+          aclEntry(DEFAULT, USER, "foo", READ_EXECUTE));
+      fileSys.modifyAclEntries(dir1, aclSpec);
+
+      // /dir1/dir2 is expected to clone the default ACL.
+      Path dir2 = new Path("/dir1/dir2");
+      fileSys.mkdirs(dir2);
+
+      // /dir1/file1 is expected to clone the default ACL.
+      Path file1 = new Path("/dir1/file1");
+      fileSys.create(file1).close();
+
+      // /dir3 is not a child of /dir1, so must not clone the default ACL.
+      Path dir3 = new Path("/dir3");
+      fileSys.mkdirs(dir3);
+
+      // /file2 is not a child of /dir1, so must not clone the default ACL.
+      Path file2 = new Path("/file2");
+      fileSys.create(file2).close();
+
+      // Restart and assert the above stated expectations.
+      IOUtils.cleanup(LOG, fileSys);
+      cluster.restartNameNode();
+      fileSys = cluster.getFileSystem();
+      assertFalse(fileSys.getAclStatus(dir1).getEntries().isEmpty());
+      assertFalse(fileSys.getAclStatus(dir2).getEntries().isEmpty());
+      assertFalse(fileSys.getAclStatus(file1).getEntries().isEmpty());
+      assertTrue(fileSys.getAclStatus(dir3).getEntries().isEmpty());
+      assertTrue(fileSys.getAclStatus(file2).getEntries().isEmpty());
+    } finally {
+      IOUtils.cleanup(LOG, fileSys);
+      if (cluster != null) {
+        cluster.shutdown();
+      }
+    }
+  }
 }