HDFS-6549. Add support for accessing the NFS gateway from the AIX NFS client. Contributed by Aaron T. Myers.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1604023 13f79535-47bb-0310-9956-ffa450edef68

hadoop-hdfs-project/hadoop-hdfs-nfs/src/main/java/org/apache/hadoop/hdfs/nfs/conf/NfsConfigKeys.java

@@ -55,4 +55,6 @@ public class NfsConfigKeys {
   public static final String DFS_NFS_PORT_MONITORING_DISABLED_KEY = "nfs.port.monitoring.disabled";
   public static final boolean DFS_NFS_PORT_MONITORING_DISABLED_DEFAULT = true;
 
-}
+  public static final String  AIX_COMPAT_MODE_KEY = "nfs.aix.compatibility.mode.enabled";
+  public static final boolean AIX_COMPAT_MODE_DEFAULT = false;
+}

hadoop-hdfs-project/hadoop-hdfs-nfs/src/main/java/org/apache/hadoop/hdfs/nfs/nfs3/OpenFileCtx.java

@@ -95,6 +95,7 @@ class OpenFileCtx {
    */
   private AtomicLong nextOffset;
   private final HdfsDataOutputStream fos;
+  private final boolean aixCompatMode;
   
   // It's updated after each sync to HDFS
   private Nfs3FileAttributes latestAttr;
@@ -199,8 +200,15 @@ class OpenFileCtx {
   
   OpenFileCtx(HdfsDataOutputStream fos, Nfs3FileAttributes latestAttr,
       String dumpFilePath, DFSClient client, IdUserGroup iug) {
+    this(fos, latestAttr, dumpFilePath, client, iug, false);
+  }
+  
+  OpenFileCtx(HdfsDataOutputStream fos, Nfs3FileAttributes latestAttr,
+      String dumpFilePath, DFSClient client, IdUserGroup iug,
+      boolean aixCompatMode) {
     this.fos = fos;
     this.latestAttr = latestAttr;
+    this.aixCompatMode = aixCompatMode;
     // We use the ReverseComparatorOnMin as the comparator of the map. In this
     // way, we first dump the data with larger offset. In the meanwhile, we
     // retrieve the last element to write back to HDFS.
@@ -788,15 +796,29 @@ class OpenFileCtx {
     }
 
     if (commitOffset > 0) {
-      if (commitOffset > flushed) {
+      if (aixCompatMode) {
-        if (!fromRead) {
+        // The AIX NFS client misinterprets RFC-1813 and will always send 4096
-          CommitCtx commitCtx = new CommitCtx(commitOffset, channel, xid,
+        // for the commitOffset even if fewer bytes than that have ever (or will
-              preOpAttr);
+        // ever) be sent by the client. So, if in AIX compatibility mode, we
-          pendingCommits.put(commitOffset, commitCtx);
+        // will always DO_SYNC if the number of bytes to commit have already all
+        // been flushed, else we will fall through to the logic below which
+        // checks for pending writes in the case that we're being asked to
+        // commit more bytes than have so far been flushed. See HDFS-6549 for
+        // more info.
+        if (commitOffset <= flushed) {
+          return COMMIT_STATUS.COMMIT_DO_SYNC;
         }
-        return COMMIT_STATUS.COMMIT_WAIT;
       } else {
-        return COMMIT_STATUS.COMMIT_DO_SYNC;
+        if (commitOffset > flushed) {
+          if (!fromRead) {
+            CommitCtx commitCtx = new CommitCtx(commitOffset, channel, xid,
+                preOpAttr);
+            pendingCommits.put(commitOffset, commitCtx);
+          }
+          return COMMIT_STATUS.COMMIT_WAIT;
+        } else {
+          return COMMIT_STATUS.COMMIT_DO_SYNC;
+        } 
       }
     }
 

hadoop-hdfs-project/hadoop-hdfs-nfs/src/main/java/org/apache/hadoop/hdfs/nfs/nfs3/RpcProgramNfs3.java

@@ -153,6 +153,7 @@ public class RpcProgramNfs3 extends RpcProgram implements Nfs3Interface {
   private final short replication;
   private final long blockSize;
   private final int bufferSize;
+  private final boolean aixCompatMode;
   private Statistics statistics;
   private String writeDumpDir; // The dir save dump files
   
@@ -170,8 +171,11 @@ public class RpcProgramNfs3 extends RpcProgram implements Nfs3Interface {
     config.set(FsPermission.UMASK_LABEL, "000");
     iug = new IdUserGroup(config);
     
+    aixCompatMode = config.getBoolean(
+        NfsConfigKeys.AIX_COMPAT_MODE_KEY,
+        NfsConfigKeys.AIX_COMPAT_MODE_DEFAULT);
     exports = NfsExports.getInstance(config);
-    writeManager = new WriteManager(iug, config);
+    writeManager = new WriteManager(iug, config, aixCompatMode);
     clientCache = new DFSClientCache(config);
     replication = (short) config.getInt(DFSConfigKeys.DFS_REPLICATION_KEY,
         DFSConfigKeys.DFS_REPLICATION_DEFAULT);
@@ -900,7 +904,8 @@ public class RpcProgramNfs3 extends RpcProgram implements Nfs3Interface {
       
       // Add open stream
       OpenFileCtx openFileCtx = new OpenFileCtx(fos, postOpObjAttr,
-          writeDumpDir + "/" + postOpObjAttr.getFileId(), dfsClient, iug);
+          writeDumpDir + "/" + postOpObjAttr.getFileId(), dfsClient, iug,
+          aixCompatMode);
       fileHandle = new FileHandle(postOpObjAttr.getFileId());
       if (!writeManager.addOpenFileStream(fileHandle, openFileCtx)) {
         LOG.warn("Can't add more stream, close it."
@@ -1438,9 +1443,24 @@ public class RpcProgramNfs3 extends RpcProgram implements Nfs3Interface {
       }
       long cookieVerf = request.getCookieVerf();
       if ((cookieVerf != 0) && (cookieVerf != dirStatus.getModificationTime())) {
-        LOG.error("CookierVerf mismatch. request cookierVerf:" + cookieVerf
+        if (aixCompatMode) {
-            + " dir cookieVerf:" + dirStatus.getModificationTime());
+          // The AIX NFS client misinterprets RFC-1813 and will repeatedly send
-        return new READDIR3Response(Nfs3Status.NFS3ERR_BAD_COOKIE);
+          // the same cookieverf value even across VFS-level readdir calls,
+          // instead of getting a new cookieverf for every VFS-level readdir
+          // call, and reusing the cookieverf only in the event that multiple
+          // incremental NFS-level readdir calls must be made to fetch all of
+          // the directory entries. This means that whenever a readdir call is
+          // made by an AIX NFS client for a given directory, and that directory
+          // is subsequently modified, thus changing its mtime, no later readdir
+          // calls will succeed from AIX for that directory until the FS is
+          // unmounted/remounted. See HDFS-6549 for more info.
+          LOG.warn("AIX compatibility mode enabled, ignoring cookieverf " +
+              "mismatches.");
+        } else {
+          LOG.error("CookieVerf mismatch. request cookieVerf: " + cookieVerf
+              + " dir cookieVerf: " + dirStatus.getModificationTime());
+          return new READDIR3Response(Nfs3Status.NFS3ERR_BAD_COOKIE);
+        }
       }
 
       if (cookie == 0) {
@@ -1588,9 +1608,22 @@ public class RpcProgramNfs3 extends RpcProgram implements Nfs3Interface {
       }
       long cookieVerf = request.getCookieVerf();
       if ((cookieVerf != 0) && (cookieVerf != dirStatus.getModificationTime())) {
-        LOG.error("CookierVerf mismatch. request cookierVerf:" + cookieVerf
+        if (aixCompatMode) {
-            + " dir cookieVerf:" + dirStatus.getModificationTime());
+          // The AIX NFS client misinterprets RFC-1813 and will repeatedly send
-        return new READDIRPLUS3Response(Nfs3Status.NFS3ERR_BAD_COOKIE);
+          // the same cookieverf value even across VFS-level readdir calls,
+          // instead of getting a new cookieverf for every VFS-level readdir
+          // call. This means that whenever a readdir call is made by an AIX NFS
+          // client for a given directory, and that directory is subsequently
+          // modified, thus changing its mtime, no later readdir calls will
+          // succeed for that directory from AIX until the FS is
+          // unmounted/remounted. See HDFS-6549 for more info.
+          LOG.warn("AIX compatibility mode enabled, ignoring cookieverf " +
+              "mismatches.");
+        } else {
+          LOG.error("cookieverf mismatch. request cookieverf: " + cookieVerf
+              + " dir cookieverf: " + dirStatus.getModificationTime());
+          return new READDIRPLUS3Response(Nfs3Status.NFS3ERR_BAD_COOKIE);
+        }
       }
 
       if (cookie == 0) {

hadoop-hdfs-project/hadoop-hdfs-nfs/src/main/java/org/apache/hadoop/hdfs/nfs/nfs3/WriteManager.java

@@ -58,6 +58,7 @@ public class WriteManager {
   private boolean asyncDataServiceStarted = false;
 
   private final int maxStreams;
+  private final boolean aixCompatMode;
 
   /**
    * The time limit to wait for accumulate reordered sequential writes to the
@@ -79,9 +80,11 @@ public class WriteManager {
     return fileContextCache.put(h, ctx);
   }
   
-  WriteManager(IdUserGroup iug, final NfsConfiguration config) {
+  WriteManager(IdUserGroup iug, final NfsConfiguration config,
+      boolean aixCompatMode) {
     this.iug = iug;
     this.config = config;
+    this.aixCompatMode = aixCompatMode;
     streamTimeout = config.getLong(NfsConfigKeys.DFS_NFS_STREAM_TIMEOUT_KEY,
         NfsConfigKeys.DFS_NFS_STREAM_TIMEOUT_DEFAULT);
     LOG.info("Stream timeout is " + streamTimeout + "ms.");
@@ -175,7 +178,7 @@ public class WriteManager {
       String writeDumpDir = config.get(NfsConfigKeys.DFS_NFS_FILE_DUMP_DIR_KEY,
           NfsConfigKeys.DFS_NFS_FILE_DUMP_DIR_DEFAULT);
       openFileCtx = new OpenFileCtx(fos, latestAttr, writeDumpDir + "/"
-          + fileHandle.getFileId(), dfsClient, iug);
+          + fileHandle.getFileId(), dfsClient, iug, aixCompatMode);
 
       if (!addOpenFileStream(fileHandle, openFileCtx)) {
         LOG.info("Can't add new stream. Close it. Tell client to retry.");

hadoop-hdfs-project/hadoop-hdfs-nfs/src/test/java/org/apache/hadoop/hdfs/nfs/nfs3/TestWrites.java

@@ -190,6 +190,29 @@ public class TestWrites {
     ret = ctx.checkCommit(dfsClient, 0, ch, 1, attr, false);
     Assert.assertTrue(ret == COMMIT_STATUS.COMMIT_FINISHED);
   }
+  
+  @Test
+  public void testCheckCommitAixCompatMode() throws IOException {
+    DFSClient dfsClient = Mockito.mock(DFSClient.class);
+    Nfs3FileAttributes attr = new Nfs3FileAttributes();
+    HdfsDataOutputStream fos = Mockito.mock(HdfsDataOutputStream.class);
+
+    // Last argument "true" here to enable AIX compatibility mode.
+    OpenFileCtx ctx = new OpenFileCtx(fos, attr, "/dumpFilePath", dfsClient,
+        new IdUserGroup(new NfsConfiguration()), true);
+    
+    // Test fall-through to pendingWrites check in the event that commitOffset
+    // is greater than the number of bytes we've so far flushed.
+    Mockito.when(fos.getPos()).thenReturn((long) 2);
+    COMMIT_STATUS status = ctx.checkCommitInternal(5, null, 1, attr, false);
+    Assert.assertTrue(status == COMMIT_STATUS.COMMIT_FINISHED);
+    
+    // Test the case when we actually have received more bytes than we're trying
+    // to commit.
+    Mockito.when(fos.getPos()).thenReturn((long) 10);
+    status = ctx.checkCommitInternal(5, null, 1, attr, false);
+    Assert.assertTrue(status == COMMIT_STATUS.COMMIT_DO_SYNC);
+  }
 
   @Test
   // Validate all the commit check return codes OpenFileCtx.COMMIT_STATUS, which
@@ -207,7 +230,7 @@ public class TestWrites {
 
     FileHandle h = new FileHandle(1); // fake handle for "/dumpFilePath"
     COMMIT_STATUS ret;
-    WriteManager wm = new WriteManager(new IdUserGroup(config), config);
+    WriteManager wm = new WriteManager(new IdUserGroup(config), config, false);
     assertTrue(wm.addOpenFileStream(h, ctx));
     
     // Test inactive open file context

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

@@ -421,6 +421,9 @@ Release 2.5.0 - UNRELEASED
     HDFS-3848. A Bug in recoverLeaseInternal method of FSNameSystem class
     (Hooman Peiro Sajjad  and Chen He via kihwal)
 
+    HDFS-6549. Add support for accessing the NFS gateway from the AIX NFS
+    client. (atm)
+
   BREAKDOWN OF HDFS-2006 SUBTASKS AND RELATED JIRAS
 
     HDFS-6299. Protobuf for XAttr and client-side implementation. (Yi Liu via umamahesh)

hadoop-hdfs-project/hadoop-hdfs/src/site/apt/HdfsNfsGateway.apt.vm

@@ -88,6 +88,25 @@ HDFS NFS Gateway
   </property>
 ----
 
+   The AIX NFS client has a {{{https://issues.apache.org/jira/browse/HDFS-6549}few known issues}}
+   that prevent it from working correctly by default with the HDFS NFS
+   Gateway. If you want to be able to access the HDFS NFS Gateway from AIX, you
+   should set the following configuration setting to enable work-arounds for these
+   issues:
+
+----
+<property>
+  <name>nfs.aix.compatibility.mode.enabled</name>
+  <value>true</value>
+</property>
+----
+
+   Note that regular, non-AIX clients should NOT enable AIX compatibility mode.
+   The work-arounds implemented by AIX compatibility mode effectively disable
+   safeguards to ensure that listing of directory contents via NFS returns
+   consistent results, and that all data sent to the NFS server can be assured to
+   have been committed.
+
    It's strongly recommended for the users to update a few configuration properties based on their use
    cases. All the related configuration properties can be added or updated in hdfs-site.xml.