001/**
002 * Licensed to the Apache Software Foundation (ASF) under one
003 * or more contributor license agreements.  See the NOTICE file
004 * distributed with this work for additional information
005 * regarding copyright ownership.  The ASF licenses this file
006 * to you under the Apache License, Version 2.0 (the
007 * "License"); you may not use this file except in compliance
008 * with the License.  You may obtain a copy of the License at
009 *
010 *     http://www.apache.org/licenses/LICENSE-2.0
011 *
012 * Unless required by applicable law or agreed to in writing, software
013 * distributed under the License is distributed on an "AS IS" BASIS,
014 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
015 * See the License for the specific language governing permissions and
016 * limitations under the License.
017 */
018package org.apache.hadoop.hdfs.server.namenode;
019
020import static org.apache.hadoop.util.Time.now;
021
022import java.io.DataInput;
023import java.io.DataInputStream;
024import java.io.DataOutputStream;
025import java.io.File;
026import java.io.FileInputStream;
027import java.io.FileNotFoundException;
028import java.io.FileOutputStream;
029import java.io.IOException;
030import java.security.DigestInputStream;
031import java.security.DigestOutputStream;
032import java.security.MessageDigest;
033import java.util.ArrayList;
034import java.util.Arrays;
035import java.util.Collection;
036import java.util.HashMap;
037import java.util.List;
038import java.util.Map;
039import java.util.TreeMap;
040
041import org.apache.commons.logging.Log;
042import org.apache.hadoop.classification.InterfaceAudience;
043import org.apache.hadoop.classification.InterfaceStability;
044import org.apache.hadoop.conf.Configuration;
045import org.apache.hadoop.fs.FileSystem;
046import org.apache.hadoop.fs.Path;
047import org.apache.hadoop.fs.PathIsNotDirectoryException;
048import org.apache.hadoop.fs.UnresolvedLinkException;
049import org.apache.hadoop.fs.permission.PermissionStatus;
050import org.apache.hadoop.hdfs.DFSUtil;
051import org.apache.hadoop.hdfs.protocol.HdfsConstants;
052import org.apache.hadoop.hdfs.protocol.LayoutFlags;
053import org.apache.hadoop.hdfs.protocol.LayoutVersion;
054import org.apache.hadoop.hdfs.protocol.LayoutVersion.Feature;
055import org.apache.hadoop.hdfs.server.blockmanagement.BlockInfo;
056import org.apache.hadoop.hdfs.server.blockmanagement.BlockInfoUnderConstruction;
057import org.apache.hadoop.hdfs.server.blockmanagement.BlockManager;
058import org.apache.hadoop.hdfs.server.common.HdfsServerConstants.StartupOption;
059import org.apache.hadoop.hdfs.server.common.InconsistentFSStateException;
060import org.apache.hadoop.hdfs.server.namenode.snapshot.DirectoryWithSnapshotFeature;
061import org.apache.hadoop.hdfs.server.namenode.snapshot.FileDiffList;
062import org.apache.hadoop.hdfs.server.namenode.snapshot.Snapshot;
063import org.apache.hadoop.hdfs.server.namenode.snapshot.SnapshotFSImageFormat;
064import org.apache.hadoop.hdfs.server.namenode.snapshot.SnapshotFSImageFormat.ReferenceMap;
065import org.apache.hadoop.hdfs.server.namenode.startupprogress.Phase;
066import org.apache.hadoop.hdfs.server.namenode.startupprogress.StartupProgress;
067import org.apache.hadoop.hdfs.server.namenode.startupprogress.StartupProgress.Counter;
068import org.apache.hadoop.hdfs.server.namenode.startupprogress.Step;
069import org.apache.hadoop.hdfs.server.namenode.startupprogress.StepType;
070import org.apache.hadoop.hdfs.util.ReadOnlyList;
071import org.apache.hadoop.io.IOUtils;
072import org.apache.hadoop.io.MD5Hash;
073import org.apache.hadoop.io.Text;
074import org.apache.hadoop.util.StringUtils;
075
076import com.google.common.annotations.VisibleForTesting;
077import com.google.common.base.Preconditions;
078
079/**
080 * Contains inner classes for reading or writing the on-disk format for
081 * FSImages.
082 *
083 * In particular, the format of the FSImage looks like:
084 * <pre>
085 * FSImage {
086 *   layoutVersion: int, namespaceID: int, numberItemsInFSDirectoryTree: long,
087 *   namesystemGenerationStampV1: long, namesystemGenerationStampV2: long,
088 *   generationStampAtBlockIdSwitch:long, lastAllocatedBlockId:
089 *   long transactionID: long, snapshotCounter: int, numberOfSnapshots: int,
090 *   numOfSnapshottableDirs: int,
091 *   {FSDirectoryTree, FilesUnderConstruction, SecretManagerState} (can be compressed)
092 * }
093 *
094 * FSDirectoryTree (if {@link Feature#FSIMAGE_NAME_OPTIMIZATION} is supported) {
095 *   INodeInfo of root, numberOfChildren of root: int
096 *   [list of INodeInfo of root's children],
097 *   [list of INodeDirectoryInfo of root's directory children]
098 * }
099 *
100 * FSDirectoryTree (if {@link Feature#FSIMAGE_NAME_OPTIMIZATION} not supported){
101 *   [list of INodeInfo of INodes in topological order]
102 * }
103 *
104 * INodeInfo {
105 *   {
106 *     localName: short + byte[]
107 *   } when {@link Feature#FSIMAGE_NAME_OPTIMIZATION} is supported
108 *   or
109 *   {
110 *     fullPath: byte[]
111 *   } when {@link Feature#FSIMAGE_NAME_OPTIMIZATION} is not supported
112 *   replicationFactor: short, modificationTime: long,
113 *   accessTime: long, preferredBlockSize: long,
114 *   numberOfBlocks: int (-1 for INodeDirectory, -2 for INodeSymLink),
115 *   {
116 *     nsQuota: long, dsQuota: long,
117 *     {
118 *       isINodeSnapshottable: byte,
119 *       isINodeWithSnapshot: byte (if isINodeSnapshottable is false)
120 *     } (when {@link Feature#SNAPSHOT} is supported),
121 *     fsPermission: short, PermissionStatus
122 *   } for INodeDirectory
123 *   or
124 *   {
125 *     symlinkString, fsPermission: short, PermissionStatus
126 *   } for INodeSymlink
127 *   or
128 *   {
129 *     [list of BlockInfo]
130 *     [list of FileDiff]
131 *     {
132 *       isINodeFileUnderConstructionSnapshot: byte,
133 *       {clientName: short + byte[], clientMachine: short + byte[]} (when
134 *       isINodeFileUnderConstructionSnapshot is true),
135 *     } (when {@link Feature#SNAPSHOT} is supported and writing snapshotINode),
136 *     fsPermission: short, PermissionStatus
137 *   } for INodeFile
138 * }
139 *
140 * INodeDirectoryInfo {
141 *   fullPath of the directory: short + byte[],
142 *   numberOfChildren: int, [list of INodeInfo of children INode],
143 *   {
144 *     numberOfSnapshots: int,
145 *     [list of Snapshot] (when NumberOfSnapshots is positive),
146 *     numberOfDirectoryDiffs: int,
147 *     [list of DirectoryDiff] (NumberOfDirectoryDiffs is positive),
148 *     number of children that are directories,
149 *     [list of INodeDirectoryInfo of the directory children] (includes
150 *     snapshot copies of deleted sub-directories)
151 *   } (when {@link Feature#SNAPSHOT} is supported),
152 * }
153 *
154 * Snapshot {
155 *   snapshotID: int, root of Snapshot: INodeDirectoryInfo (its local name is
156 *   the name of the snapshot)
157 * }
158 *
159 * DirectoryDiff {
160 *   full path of the root of the associated Snapshot: short + byte[],
161 *   childrenSize: int,
162 *   isSnapshotRoot: byte,
163 *   snapshotINodeIsNotNull: byte (when isSnapshotRoot is false),
164 *   snapshotINode: INodeDirectory (when SnapshotINodeIsNotNull is true), Diff
165 * }
166 *
167 * Diff {
168 *   createdListSize: int, [Local name of INode in created list],
169 *   deletedListSize: int, [INode in deleted list: INodeInfo]
170 * }
171 *
172 * FileDiff {
173 *   full path of the root of the associated Snapshot: short + byte[],
174 *   fileSize: long,
175 *   snapshotINodeIsNotNull: byte,
176 *   snapshotINode: INodeFile (when SnapshotINodeIsNotNull is true), Diff
177 * }
178 * </pre>
179 */
180@InterfaceAudience.Private
181@InterfaceStability.Evolving
182public class FSImageFormat {
183  private static final Log LOG = FSImage.LOG;
184
185  // Static-only class
186  private FSImageFormat() {}
187
188  interface AbstractLoader {
189    MD5Hash getLoadedImageMd5();
190    long getLoadedImageTxId();
191  }
192
193  static class LoaderDelegator implements AbstractLoader {
194    private AbstractLoader impl;
195    private final Configuration conf;
196    private final FSNamesystem fsn;
197
198    LoaderDelegator(Configuration conf, FSNamesystem fsn) {
199      this.conf = conf;
200      this.fsn = fsn;
201    }
202
203    @Override
204    public MD5Hash getLoadedImageMd5() {
205      return impl.getLoadedImageMd5();
206    }
207
208    @Override
209    public long getLoadedImageTxId() {
210      return impl.getLoadedImageTxId();
211    }
212
213    public void load(File file, boolean requireSameLayoutVersion)
214        throws IOException {
215      Preconditions.checkState(impl == null, "Image already loaded!");
216
217      FileInputStream is = null;
218      try {
219        is = new FileInputStream(file);
220        byte[] magic = new byte[FSImageUtil.MAGIC_HEADER.length];
221        IOUtils.readFully(is, magic, 0, magic.length);
222        if (Arrays.equals(magic, FSImageUtil.MAGIC_HEADER)) {
223          FSImageFormatProtobuf.Loader loader = new FSImageFormatProtobuf.Loader(
224              conf, fsn, requireSameLayoutVersion);
225          impl = loader;
226          loader.load(file);
227        } else {
228          Loader loader = new Loader(conf, fsn);
229          impl = loader;
230          loader.load(file);
231        }
232      } finally {
233        IOUtils.cleanup(LOG, is);
234      }
235    }
236  }
237
238  /**
239   * Construct a loader class to load the image. It chooses the loader based on
240   * the layout version.
241   */
242  public static LoaderDelegator newLoader(Configuration conf, FSNamesystem fsn) {
243    return new LoaderDelegator(conf, fsn);
244  }
245
246  /**
247   * A one-shot class responsible for loading an image. The load() function
248   * should be called once, after which the getter methods may be used to retrieve
249   * information about the image that was loaded, if loading was successful.
250   */
251  public static class Loader implements AbstractLoader {
252    private final Configuration conf;
253    /** which namesystem this loader is working for */
254    private final FSNamesystem namesystem;
255
256    /** Set to true once a file has been loaded using this loader. */
257    private boolean loaded = false;
258
259    /** The transaction ID of the last edit represented by the loaded file */
260    private long imgTxId;
261    /** The MD5 sum of the loaded file */
262    private MD5Hash imgDigest;
263    
264    private Map<Integer, Snapshot> snapshotMap = null;
265    private final ReferenceMap referenceMap = new ReferenceMap();
266
267    Loader(Configuration conf, FSNamesystem namesystem) {
268      this.conf = conf;
269      this.namesystem = namesystem;
270    }
271
272    /**
273     * Return the MD5 checksum of the image that has been loaded.
274     * @throws IllegalStateException if load() has not yet been called.
275     */
276    @Override
277    public MD5Hash getLoadedImageMd5() {
278      checkLoaded();
279      return imgDigest;
280    }
281
282    @Override
283    public long getLoadedImageTxId() {
284      checkLoaded();
285      return imgTxId;
286    }
287
288    /**
289     * Throw IllegalStateException if load() has not yet been called.
290     */
291    private void checkLoaded() {
292      if (!loaded) {
293        throw new IllegalStateException("Image not yet loaded!");
294      }
295    }
296
297    /**
298     * Throw IllegalStateException if load() has already been called.
299     */
300    private void checkNotLoaded() {
301      if (loaded) {
302        throw new IllegalStateException("Image already loaded!");
303      }
304    }
305
306    public void load(File curFile) throws IOException {
307      checkNotLoaded();
308      assert curFile != null : "curFile is null";
309
310      StartupProgress prog = NameNode.getStartupProgress();
311      Step step = new Step(StepType.INODES);
312      prog.beginStep(Phase.LOADING_FSIMAGE, step);
313      long startTime = now();
314
315      //
316      // Load in bits
317      //
318      MessageDigest digester = MD5Hash.getDigester();
319      DigestInputStream fin = new DigestInputStream(
320           new FileInputStream(curFile), digester);
321
322      DataInputStream in = new DataInputStream(fin);
323      try {
324        // read image version: first appeared in version -1
325        int imgVersion = in.readInt();
326        if (getLayoutVersion() != imgVersion) {
327          throw new InconsistentFSStateException(curFile, 
328              "imgVersion " + imgVersion +
329              " expected to be " + getLayoutVersion());
330        }
331        boolean supportSnapshot = NameNodeLayoutVersion.supports(
332            LayoutVersion.Feature.SNAPSHOT, imgVersion);
333        if (NameNodeLayoutVersion.supports(
334            LayoutVersion.Feature.ADD_LAYOUT_FLAGS, imgVersion)) {
335          LayoutFlags.read(in);
336        }
337
338        // read namespaceID: first appeared in version -2
339        in.readInt();
340
341        long numFiles = in.readLong();
342
343        // read in the last generation stamp for legacy blocks.
344        long genstamp = in.readLong();
345        namesystem.setGenerationStampV1(genstamp);
346        
347        if (NameNodeLayoutVersion.supports(
348            LayoutVersion.Feature.SEQUENTIAL_BLOCK_ID, imgVersion)) {
349          // read the starting generation stamp for sequential block IDs
350          genstamp = in.readLong();
351          namesystem.setGenerationStampV2(genstamp);
352
353          // read the last generation stamp for blocks created after
354          // the switch to sequential block IDs.
355          long stampAtIdSwitch = in.readLong();
356          namesystem.setGenerationStampV1Limit(stampAtIdSwitch);
357
358          // read the max sequential block ID.
359          long maxSequentialBlockId = in.readLong();
360          namesystem.setLastAllocatedBlockId(maxSequentialBlockId);
361        } else {
362          long startingGenStamp = namesystem.upgradeGenerationStampToV2();
363          // This is an upgrade.
364          LOG.info("Upgrading to sequential block IDs. Generation stamp " +
365                   "for new blocks set to " + startingGenStamp);
366        }
367
368        // read the transaction ID of the last edit represented by
369        // this image
370        if (NameNodeLayoutVersion.supports(
371            LayoutVersion.Feature.STORED_TXIDS, imgVersion)) {
372          imgTxId = in.readLong();
373        } else {
374          imgTxId = 0;
375        }
376
377        // read the last allocated inode id in the fsimage
378        if (NameNodeLayoutVersion.supports(
379            LayoutVersion.Feature.ADD_INODE_ID, imgVersion)) {
380          long lastInodeId = in.readLong();
381          namesystem.resetLastInodeId(lastInodeId);
382          if (LOG.isDebugEnabled()) {
383            LOG.debug("load last allocated InodeId from fsimage:" + lastInodeId);
384          }
385        } else {
386          if (LOG.isDebugEnabled()) {
387            LOG.debug("Old layout version doesn't have inode id."
388                + " Will assign new id for each inode.");
389          }
390        }
391        
392        if (supportSnapshot) {
393          snapshotMap = namesystem.getSnapshotManager().read(in, this);
394        }
395
396        // read compression related info
397        FSImageCompression compression;
398        if (NameNodeLayoutVersion.supports(
399            LayoutVersion.Feature.FSIMAGE_COMPRESSION, imgVersion)) {
400          compression = FSImageCompression.readCompressionHeader(conf, in);
401        } else {
402          compression = FSImageCompression.createNoopCompression();
403        }
404        in = compression.unwrapInputStream(fin);
405
406        LOG.info("Loading image file " + curFile + " using " + compression);
407        
408        // load all inodes
409        LOG.info("Number of files = " + numFiles);
410        prog.setTotal(Phase.LOADING_FSIMAGE, step, numFiles);
411        Counter counter = prog.getCounter(Phase.LOADING_FSIMAGE, step);
412        if (NameNodeLayoutVersion.supports(
413            LayoutVersion.Feature.FSIMAGE_NAME_OPTIMIZATION, imgVersion)) {
414          if (supportSnapshot) {
415            loadLocalNameINodesWithSnapshot(numFiles, in, counter);
416          } else {
417            loadLocalNameINodes(numFiles, in, counter);
418          }
419        } else {
420          loadFullNameINodes(numFiles, in, counter);
421        }
422
423        loadFilesUnderConstruction(in, supportSnapshot, counter);
424        prog.endStep(Phase.LOADING_FSIMAGE, step);
425        // Now that the step is finished, set counter equal to total to adjust
426        // for possible under-counting due to reference inodes.
427        prog.setCount(Phase.LOADING_FSIMAGE, step, numFiles);
428
429        loadSecretManagerState(in);
430
431        loadCacheManagerState(in);
432
433        // make sure to read to the end of file
434        boolean eof = (in.read() == -1);
435        assert eof : "Should have reached the end of image file " + curFile;
436      } finally {
437        in.close();
438      }
439
440      imgDigest = new MD5Hash(digester.digest());
441      loaded = true;
442      
443      LOG.info("Image file " + curFile + " of size " + curFile.length() +
444          " bytes loaded in " + (now() - startTime)/1000 + " seconds.");
445    }
446
447  /** Update the root node's attributes */
448  private void updateRootAttr(INodeWithAdditionalFields root) {                                                           
449    final Quota.Counts q = root.getQuotaCounts();
450    final long nsQuota = q.get(Quota.NAMESPACE);
451    final long dsQuota = q.get(Quota.DISKSPACE);
452    FSDirectory fsDir = namesystem.dir;
453    if (nsQuota != -1 || dsQuota != -1) {
454      fsDir.rootDir.getDirectoryWithQuotaFeature().setQuota(nsQuota, dsQuota);
455    }
456    fsDir.rootDir.cloneModificationTime(root);
457    fsDir.rootDir.clonePermissionStatus(root);    
458  }
459  
460    /**
461     * Load fsimage files when 1) only local names are stored, 
462     * and 2) snapshot is supported.
463     * 
464     * @param numFiles number of files expected to be read
465     * @param in Image input stream
466     * @param counter Counter to increment for namenode startup progress
467     */
468    private void loadLocalNameINodesWithSnapshot(long numFiles, DataInput in,
469        Counter counter) throws IOException {
470      assert NameNodeLayoutVersion.supports(
471          LayoutVersion.Feature.FSIMAGE_NAME_OPTIMIZATION, getLayoutVersion());
472      assert NameNodeLayoutVersion.supports(
473          LayoutVersion.Feature.SNAPSHOT, getLayoutVersion());
474      
475      // load root
476      loadRoot(in, counter);
477      // load rest of the nodes recursively
478      loadDirectoryWithSnapshot(in, counter);
479    }
480    
481  /** 
482   * load fsimage files assuming only local names are stored. Used when
483   * snapshots are not supported by the layout version.
484   *   
485   * @param numFiles number of files expected to be read
486   * @param in image input stream
487   * @param counter Counter to increment for namenode startup progress
488   * @throws IOException
489   */  
490   private void loadLocalNameINodes(long numFiles, DataInput in, Counter counter)
491       throws IOException {
492     assert NameNodeLayoutVersion.supports(
493         LayoutVersion.Feature.FSIMAGE_NAME_OPTIMIZATION, getLayoutVersion());
494     assert numFiles > 0;
495
496     // load root
497     loadRoot(in, counter);
498     // have loaded the first file (the root)
499     numFiles--; 
500
501     // load rest of the nodes directory by directory
502     while (numFiles > 0) {
503       numFiles -= loadDirectory(in, counter);
504     }
505     if (numFiles != 0) {
506       throw new IOException("Read unexpect number of files: " + -numFiles);
507     }
508   }
509   
510    /**
511     * Load information about root, and use the information to update the root
512     * directory of NameSystem.
513     * @param in The {@link DataInput} instance to read.
514     * @param counter Counter to increment for namenode startup progress
515     */
516    private void loadRoot(DataInput in, Counter counter)
517        throws IOException {
518      // load root
519      if (in.readShort() != 0) {
520        throw new IOException("First node is not root");
521      }
522      final INodeDirectory root = loadINode(null, false, in, counter)
523        .asDirectory();
524      // update the root's attributes
525      updateRootAttr(root);
526    }
527   
528    /** Load children nodes for the parent directory. */
529    private int loadChildren(INodeDirectory parent, DataInput in,
530        Counter counter) throws IOException {
531      int numChildren = in.readInt();
532      for (int i = 0; i < numChildren; i++) {
533        // load single inode
534        INode newNode = loadINodeWithLocalName(false, in, true, counter);
535        addToParent(parent, newNode);
536      }
537      return numChildren;
538    }
539    
540    /**
541     * Load a directory when snapshot is supported.
542     * @param in The {@link DataInput} instance to read.
543     * @param counter Counter to increment for namenode startup progress
544     */
545    private void loadDirectoryWithSnapshot(DataInput in, Counter counter)
546        throws IOException {
547      // Step 1. Identify the parent INode
548      long inodeId = in.readLong();
549      final INodeDirectory parent = this.namesystem.dir.getInode(inodeId)
550          .asDirectory();
551      
552      // Check if the whole subtree has been saved (for reference nodes)
553      boolean toLoadSubtree = referenceMap.toProcessSubtree(parent.getId());
554      if (!toLoadSubtree) {
555        return;
556      }
557
558      // Step 2. Load snapshots if parent is snapshottable
559      int numSnapshots = in.readInt();
560      if (numSnapshots >= 0) {
561        // load snapshots and snapshotQuota
562        SnapshotFSImageFormat.loadSnapshotList(parent, numSnapshots, in, this);
563        if (parent.getDirectorySnapshottableFeature().getSnapshotQuota() > 0) {
564          // add the directory to the snapshottable directory list in 
565          // SnapshotManager. Note that we only add root when its snapshot quota
566          // is positive.
567          this.namesystem.getSnapshotManager().addSnapshottable(parent);
568        }
569      }
570
571      // Step 3. Load children nodes under parent
572      loadChildren(parent, in, counter);
573      
574      // Step 4. load Directory Diff List
575      SnapshotFSImageFormat.loadDirectoryDiffList(parent, in, this);
576      
577      // Recursively load sub-directories, including snapshot copies of deleted
578      // directories
579      int numSubTree = in.readInt();
580      for (int i = 0; i < numSubTree; i++) {
581        loadDirectoryWithSnapshot(in, counter);
582      }
583    }
584    
585   /**
586    * Load all children of a directory
587    * 
588    * @param in input to load from
589    * @param counter Counter to increment for namenode startup progress
590    * @return number of child inodes read
591    * @throws IOException
592    */
593   private int loadDirectory(DataInput in, Counter counter) throws IOException {
594     String parentPath = FSImageSerialization.readString(in);
595     // Rename .snapshot paths if we're doing an upgrade
596     parentPath = renameReservedPathsOnUpgrade(parentPath, getLayoutVersion());
597     final INodeDirectory parent = INodeDirectory.valueOf(
598         namesystem.dir.getNode(parentPath, true), parentPath);
599     return loadChildren(parent, in, counter);
600   }
601
602  /**
603   * load fsimage files assuming full path names are stored
604   * 
605   * @param numFiles total number of files to load
606   * @param in data input stream
607   * @param counter Counter to increment for namenode startup progress
608   * @throws IOException if any error occurs
609   */
610  private void loadFullNameINodes(long numFiles, DataInput in, Counter counter)
611      throws IOException {
612    byte[][] pathComponents;
613    byte[][] parentPath = {{}};      
614    FSDirectory fsDir = namesystem.dir;
615    INodeDirectory parentINode = fsDir.rootDir;
616    for (long i = 0; i < numFiles; i++) {
617      pathComponents = FSImageSerialization.readPathComponents(in);
618      for (int j=0; j < pathComponents.length; j++) {
619        byte[] newComponent = renameReservedComponentOnUpgrade
620            (pathComponents[j], getLayoutVersion());
621        if (!Arrays.equals(newComponent, pathComponents[j])) {
622          String oldPath = DFSUtil.byteArray2PathString(pathComponents);
623          pathComponents[j] = newComponent;
624          String newPath = DFSUtil.byteArray2PathString(pathComponents);
625          LOG.info("Renaming reserved path " + oldPath + " to " + newPath);
626        }
627      }
628      final INode newNode = loadINode(
629          pathComponents[pathComponents.length-1], false, in, counter);
630
631      if (isRoot(pathComponents)) { // it is the root
632        // update the root's attributes
633        updateRootAttr(newNode.asDirectory());
634        continue;
635      }
636
637      namesystem.dir.addToInodeMap(newNode);
638      // check if the new inode belongs to the same parent
639      if(!isParent(pathComponents, parentPath)) {
640        parentINode = getParentINodeDirectory(pathComponents);
641        parentPath = getParent(pathComponents);
642      }
643
644      // add new inode
645      addToParent(parentINode, newNode);
646    }
647  }
648
649  private INodeDirectory getParentINodeDirectory(byte[][] pathComponents
650      ) throws FileNotFoundException, PathIsNotDirectoryException,
651      UnresolvedLinkException {
652    if (pathComponents.length < 2) { // root
653      return null;
654    }
655    // Gets the parent INode
656    final INodesInPath inodes = namesystem.dir.getExistingPathINodes(
657        pathComponents);
658    return INodeDirectory.valueOf(inodes.getINode(-2), pathComponents);
659  }
660
661  /**
662   * Add the child node to parent and, if child is a file, update block map.
663   * This method is only used for image loading so that synchronization,
664   * modification time update and space count update are not needed.
665   */
666  private void addToParent(INodeDirectory parent, INode child) {
667    FSDirectory fsDir = namesystem.dir;
668    if (parent == fsDir.rootDir) {
669        child.setLocalName(renameReservedRootComponentOnUpgrade(
670            child.getLocalNameBytes(), getLayoutVersion()));
671    }
672    // NOTE: This does not update space counts for parents
673    if (!parent.addChild(child)) {
674      return;
675    }
676    namesystem.dir.cacheName(child);
677
678    if (child.isFile()) {
679      updateBlocksMap(child.asFile());
680    }
681  }
682
683    public void updateBlocksMap(INodeFile file) {
684      // Add file->block mapping
685      final BlockInfo[] blocks = file.getBlocks();
686      if (blocks != null) {
687        final BlockManager bm = namesystem.getBlockManager();
688        for (int i = 0; i < blocks.length; i++) {
689          file.setBlock(i, bm.addBlockCollection(blocks[i], file));
690        } 
691      }
692    }
693
694    /** @return The FSDirectory of the namesystem where the fsimage is loaded */
695    public FSDirectory getFSDirectoryInLoading() {
696      return namesystem.dir;
697    }
698
699    public INode loadINodeWithLocalName(boolean isSnapshotINode, DataInput in,
700        boolean updateINodeMap) throws IOException {
701      return loadINodeWithLocalName(isSnapshotINode, in, updateINodeMap, null);
702    }
703
704    public INode loadINodeWithLocalName(boolean isSnapshotINode,
705        DataInput in, boolean updateINodeMap, Counter counter)
706        throws IOException {
707      byte[] localName = FSImageSerialization.readLocalName(in);
708      localName =
709          renameReservedComponentOnUpgrade(localName, getLayoutVersion());
710      INode inode = loadINode(localName, isSnapshotINode, in, counter);
711      if (updateINodeMap) {
712        namesystem.dir.addToInodeMap(inode);
713      }
714      return inode;
715    }
716  
717  /**
718   * load an inode from fsimage except for its name
719   * 
720   * @param in data input stream from which image is read
721   * @param counter Counter to increment for namenode startup progress
722   * @return an inode
723   */
724  @SuppressWarnings("deprecation")
725  INode loadINode(final byte[] localName, boolean isSnapshotINode,
726      DataInput in, Counter counter) throws IOException {
727    final int imgVersion = getLayoutVersion();
728    if (NameNodeLayoutVersion.supports(
729        LayoutVersion.Feature.SNAPSHOT, imgVersion)) {
730      namesystem.getFSDirectory().verifyINodeName(localName);
731    }
732
733    long inodeId = NameNodeLayoutVersion.supports(
734        LayoutVersion.Feature.ADD_INODE_ID, imgVersion) ? in.readLong()
735        : namesystem.allocateNewInodeId();
736    
737    final short replication = namesystem.getBlockManager().adjustReplication(
738        in.readShort());
739    final long modificationTime = in.readLong();
740    long atime = 0;
741    if (NameNodeLayoutVersion.supports(
742        LayoutVersion.Feature.FILE_ACCESS_TIME, imgVersion)) {
743      atime = in.readLong();
744    }
745    final long blockSize = in.readLong();
746    final int numBlocks = in.readInt();
747
748    if (numBlocks >= 0) {
749      // file
750      
751      // read blocks
752      BlockInfo[] blocks = new BlockInfo[numBlocks];
753      for (int j = 0; j < numBlocks; j++) {
754        blocks[j] = new BlockInfo(replication);
755        blocks[j].readFields(in);
756      }
757
758      String clientName = "";
759      String clientMachine = "";
760      boolean underConstruction = false;
761      FileDiffList fileDiffs = null;
762      if (NameNodeLayoutVersion.supports(
763          LayoutVersion.Feature.SNAPSHOT, imgVersion)) {
764        // read diffs
765        fileDiffs = SnapshotFSImageFormat.loadFileDiffList(in, this);
766
767        if (isSnapshotINode) {
768          underConstruction = in.readBoolean();
769          if (underConstruction) {
770            clientName = FSImageSerialization.readString(in);
771            clientMachine = FSImageSerialization.readString(in);
772            // convert the last block to BlockUC
773            if (blocks.length > 0) {
774              BlockInfo lastBlk = blocks[blocks.length - 1]; 
775              blocks[blocks.length - 1] = new BlockInfoUnderConstruction(
776                  lastBlk, replication);
777            }
778          }
779        }
780      }
781
782      final PermissionStatus permissions = PermissionStatus.read(in);
783
784      // return
785      if (counter != null) {
786        counter.increment();
787      }
788
789      final INodeFile file = new INodeFile(inodeId, localName, permissions,
790          modificationTime, atime, blocks, replication, blockSize, (byte)0);
791      if (underConstruction) {
792        file.toUnderConstruction(clientName, clientMachine);
793      }
794        return fileDiffs == null ? file : new INodeFile(file, fileDiffs);
795      } else if (numBlocks == -1) {
796        //directory
797      
798      //read quotas
799      final long nsQuota = in.readLong();
800      long dsQuota = -1L;
801      if (NameNodeLayoutVersion.supports(
802          LayoutVersion.Feature.DISKSPACE_QUOTA, imgVersion)) {
803        dsQuota = in.readLong();
804      }
805
806      //read snapshot info
807      boolean snapshottable = false;
808      boolean withSnapshot = false;
809      if (NameNodeLayoutVersion.supports(
810          LayoutVersion.Feature.SNAPSHOT, imgVersion)) {
811        snapshottable = in.readBoolean();
812        if (!snapshottable) {
813          withSnapshot = in.readBoolean();
814        }
815      }
816
817      final PermissionStatus permissions = PermissionStatus.read(in);
818
819      //return
820      if (counter != null) {
821        counter.increment();
822      }
823      final INodeDirectory dir = new INodeDirectory(inodeId, localName,
824          permissions, modificationTime);
825      if (nsQuota >= 0 || dsQuota >= 0) {
826        dir.addDirectoryWithQuotaFeature(nsQuota, dsQuota);
827      }
828      if (withSnapshot) {
829        dir.addSnapshotFeature(null);
830      }
831      if (snapshottable) {
832        dir.addSnapshottableFeature();
833      }
834      return dir;
835    } else if (numBlocks == -2) {
836      //symlink
837      if (!FileSystem.areSymlinksEnabled()) {
838        throw new IOException("Symlinks not supported - please remove symlink before upgrading to this version of HDFS");
839      }
840
841      final String symlink = Text.readString(in);
842      final PermissionStatus permissions = PermissionStatus.read(in);
843      if (counter != null) {
844        counter.increment();
845      }
846      return new INodeSymlink(inodeId, localName, permissions,
847          modificationTime, atime, symlink);
848    } else if (numBlocks == -3) {
849      //reference
850      // Intentionally do not increment counter, because it is too difficult at
851      // this point to assess whether or not this is a reference that counts
852      // toward quota.
853      
854      final boolean isWithName = in.readBoolean();
855      // lastSnapshotId for WithName node, dstSnapshotId for DstReference node
856      int snapshotId = in.readInt();
857      
858      final INodeReference.WithCount withCount
859          = referenceMap.loadINodeReferenceWithCount(isSnapshotINode, in, this);
860
861      if (isWithName) {
862          return new INodeReference.WithName(null, withCount, localName,
863              snapshotId);
864      } else {
865        final INodeReference ref = new INodeReference.DstReference(null,
866            withCount, snapshotId);
867        return ref;
868      }
869    }
870    
871    throw new IOException("Unknown inode type: numBlocks=" + numBlocks);
872  }
873
874    /** Load {@link INodeFileAttributes}. */
875    public INodeFileAttributes loadINodeFileAttributes(DataInput in)
876        throws IOException {
877      final int layoutVersion = getLayoutVersion();
878      
879      if (!NameNodeLayoutVersion.supports(
880          LayoutVersion.Feature.OPTIMIZE_SNAPSHOT_INODES, layoutVersion)) {
881        return loadINodeWithLocalName(true, in, false).asFile();
882      }
883  
884      final byte[] name = FSImageSerialization.readLocalName(in);
885      final PermissionStatus permissions = PermissionStatus.read(in);
886      final long modificationTime = in.readLong();
887      final long accessTime = in.readLong();
888  
889      final short replication = namesystem.getBlockManager().adjustReplication(
890          in.readShort());
891      final long preferredBlockSize = in.readLong();
892
893      return new INodeFileAttributes.SnapshotCopy(name, permissions, null, modificationTime,
894          accessTime, replication, preferredBlockSize, (byte) 0, null);
895    }
896
897    public INodeDirectoryAttributes loadINodeDirectoryAttributes(DataInput in)
898        throws IOException {
899      final int layoutVersion = getLayoutVersion();
900      
901      if (!NameNodeLayoutVersion.supports(
902          LayoutVersion.Feature.OPTIMIZE_SNAPSHOT_INODES, layoutVersion)) {
903        return loadINodeWithLocalName(true, in, false).asDirectory();
904      }
905  
906      final byte[] name = FSImageSerialization.readLocalName(in);
907      final PermissionStatus permissions = PermissionStatus.read(in);
908      final long modificationTime = in.readLong();
909      
910      //read quotas
911      final long nsQuota = in.readLong();
912      final long dsQuota = in.readLong();
913  
914      return nsQuota == -1L && dsQuota == -1L ? new INodeDirectoryAttributes.SnapshotCopy(
915          name, permissions, null, modificationTime, null)
916        : new INodeDirectoryAttributes.CopyWithQuota(name, permissions,
917            null, modificationTime, nsQuota, dsQuota, null);
918    }
919  
920    private void loadFilesUnderConstruction(DataInput in,
921        boolean supportSnapshot, Counter counter) throws IOException {
922      FSDirectory fsDir = namesystem.dir;
923      int size = in.readInt();
924
925      LOG.info("Number of files under construction = " + size);
926
927      for (int i = 0; i < size; i++) {
928        INodeFile cons = FSImageSerialization.readINodeUnderConstruction(in,
929            namesystem, getLayoutVersion());
930        counter.increment();
931
932        // verify that file exists in namespace
933        String path = cons.getLocalName();
934        INodeFile oldnode = null;
935        boolean inSnapshot = false;
936        if (path != null && FSDirectory.isReservedName(path) && 
937            NameNodeLayoutVersion.supports(
938                LayoutVersion.Feature.ADD_INODE_ID, getLayoutVersion())) {
939          // TODO: for HDFS-5428, we use reserved path for those INodeFileUC in
940          // snapshot. If we support INode ID in the layout version, we can use
941          // the inode id to find the oldnode.
942          oldnode = namesystem.dir.getInode(cons.getId()).asFile();
943          inSnapshot = true;
944        } else {
945          path = renameReservedPathsOnUpgrade(path, getLayoutVersion());
946          final INodesInPath iip = fsDir.getLastINodeInPath(path);
947          oldnode = INodeFile.valueOf(iip.getINode(0), path);
948        }
949
950        FileUnderConstructionFeature uc = cons.getFileUnderConstructionFeature();
951        oldnode.toUnderConstruction(uc.getClientName(), uc.getClientMachine());
952        if (oldnode.numBlocks() > 0) {
953          BlockInfo ucBlock = cons.getLastBlock();
954          // we do not replace the inode, just replace the last block of oldnode
955          BlockInfo info = namesystem.getBlockManager().addBlockCollection(
956              ucBlock, oldnode);
957          oldnode.setBlock(oldnode.numBlocks() - 1, info);
958        }
959
960        if (!inSnapshot) {
961          namesystem.leaseManager.addLease(cons
962              .getFileUnderConstructionFeature().getClientName(), path);
963        }
964      }
965    }
966
967    private void loadSecretManagerState(DataInput in)
968        throws IOException {
969      int imgVersion = getLayoutVersion();
970
971      if (!NameNodeLayoutVersion.supports(
972          LayoutVersion.Feature.DELEGATION_TOKEN, imgVersion)) {
973        //SecretManagerState is not available.
974        //This must not happen if security is turned on.
975        return; 
976      }
977      namesystem.loadSecretManagerStateCompat(in);
978    }
979
980    private void loadCacheManagerState(DataInput in) throws IOException {
981      int imgVersion = getLayoutVersion();
982      if (!NameNodeLayoutVersion.supports(
983          LayoutVersion.Feature.CACHING, imgVersion)) {
984        return;
985      }
986      namesystem.getCacheManager().loadStateCompat(in);
987    }
988
989    private int getLayoutVersion() {
990      return namesystem.getFSImage().getStorage().getLayoutVersion();
991    }
992
993    private boolean isRoot(byte[][] path) {
994      return path.length == 1 &&
995        path[0] == null;    
996    }
997
998    private boolean isParent(byte[][] path, byte[][] parent) {
999      if (path == null || parent == null)
1000        return false;
1001      if (parent.length == 0 || path.length != parent.length + 1)
1002        return false;
1003      boolean isParent = true;
1004      for (int i = 0; i < parent.length; i++) {
1005        isParent = isParent && Arrays.equals(path[i], parent[i]); 
1006      }
1007      return isParent;
1008    }
1009
1010    /**
1011     * Return string representing the parent of the given path.
1012     */
1013    String getParent(String path) {
1014      return path.substring(0, path.lastIndexOf(Path.SEPARATOR));
1015    }
1016    
1017    byte[][] getParent(byte[][] path) {
1018      byte[][] result = new byte[path.length - 1][];
1019      for (int i = 0; i < result.length; i++) {
1020        result[i] = new byte[path[i].length];
1021        System.arraycopy(path[i], 0, result[i], 0, path[i].length);
1022      }
1023      return result;
1024    }
1025    
1026    public Snapshot getSnapshot(DataInput in) throws IOException {
1027      return snapshotMap.get(in.readInt());
1028    }
1029  }
1030
1031  @VisibleForTesting
1032  public static final TreeMap<String, String> renameReservedMap =
1033      new TreeMap<String, String>();
1034
1035  /**
1036   * Use the default key-value pairs that will be used to determine how to
1037   * rename reserved paths on upgrade.
1038   */
1039  @VisibleForTesting
1040  public static void useDefaultRenameReservedPairs() {
1041    renameReservedMap.clear();
1042    for (String key: HdfsConstants.RESERVED_PATH_COMPONENTS) {
1043      renameReservedMap.put(
1044          key,
1045          key + "." + HdfsConstants.NAMENODE_LAYOUT_VERSION + "."
1046              + "UPGRADE_RENAMED");
1047    }
1048  }
1049
1050  /**
1051   * Set the key-value pairs that will be used to determine how to rename
1052   * reserved paths on upgrade.
1053   */
1054  @VisibleForTesting
1055  public static void setRenameReservedPairs(String renameReserved) {
1056    // Clear and set the default values
1057    useDefaultRenameReservedPairs();
1058    // Overwrite with provided values
1059    setRenameReservedMapInternal(renameReserved);
1060  }
1061
1062  private static void setRenameReservedMapInternal(String renameReserved) {
1063    Collection<String> pairs =
1064        StringUtils.getTrimmedStringCollection(renameReserved);
1065    for (String p : pairs) {
1066      String[] pair = StringUtils.split(p, '/', '=');
1067      Preconditions.checkArgument(pair.length == 2,
1068          "Could not parse key-value pair " + p);
1069      String key = pair[0];
1070      String value = pair[1];
1071      Preconditions.checkArgument(DFSUtil.isReservedPathComponent(key),
1072          "Unknown reserved path " + key);
1073      Preconditions.checkArgument(DFSUtil.isValidNameForComponent(value),
1074          "Invalid rename path for " + key + ": " + value);
1075      LOG.info("Will rename reserved path " + key + " to " + value);
1076      renameReservedMap.put(key, value);
1077    }
1078  }
1079
1080  /**
1081   * When upgrading from an old version, the filesystem could contain paths
1082   * that are now reserved in the new version (e.g. .snapshot). This renames
1083   * these new reserved paths to a user-specified value to avoid collisions
1084   * with the reserved name.
1085   * 
1086   * @param path Old path potentially containing a reserved path
1087   * @return New path with reserved path components renamed to user value
1088   */
1089  static String renameReservedPathsOnUpgrade(String path,
1090      final int layoutVersion) {
1091    final String oldPath = path;
1092    // If any known LVs aren't supported, we're doing an upgrade
1093    if (!NameNodeLayoutVersion.supports(Feature.ADD_INODE_ID, layoutVersion)) {
1094      String[] components = INode.getPathNames(path);
1095      // Only need to worry about the root directory
1096      if (components.length > 1) {
1097        components[1] = DFSUtil.bytes2String(
1098            renameReservedRootComponentOnUpgrade(
1099                DFSUtil.string2Bytes(components[1]),
1100                layoutVersion));
1101        path = DFSUtil.strings2PathString(components);
1102      }
1103    }
1104    if (!NameNodeLayoutVersion.supports(Feature.SNAPSHOT, layoutVersion)) {
1105      String[] components = INode.getPathNames(path);
1106      // Special case the root path
1107      if (components.length == 0) {
1108        return path;
1109      }
1110      for (int i=0; i<components.length; i++) {
1111        components[i] = DFSUtil.bytes2String(
1112            renameReservedComponentOnUpgrade(
1113                DFSUtil.string2Bytes(components[i]),
1114                layoutVersion));
1115      }
1116      path = DFSUtil.strings2PathString(components);
1117    }
1118
1119    if (!path.equals(oldPath)) {
1120      LOG.info("Upgrade process renamed reserved path " + oldPath + " to "
1121          + path);
1122    }
1123    return path;
1124  }
1125
1126  private final static String RESERVED_ERROR_MSG = 
1127      FSDirectory.DOT_RESERVED_PATH_PREFIX + " is a reserved path and "
1128      + HdfsConstants.DOT_SNAPSHOT_DIR + " is a reserved path component in"
1129      + " this version of HDFS. Please rollback and delete or rename"
1130      + " this path, or upgrade with the "
1131      + StartupOption.RENAMERESERVED.getName()
1132      + " [key-value pairs]"
1133      + " option to automatically rename these paths during upgrade.";
1134
1135  /**
1136   * Same as {@link #renameReservedPathsOnUpgrade(String)}, but for a single
1137   * byte array path component.
1138   */
1139  private static byte[] renameReservedComponentOnUpgrade(byte[] component,
1140      final int layoutVersion) {
1141    // If the LV doesn't support snapshots, we're doing an upgrade
1142    if (!NameNodeLayoutVersion.supports(Feature.SNAPSHOT, layoutVersion)) {
1143      if (Arrays.equals(component, HdfsConstants.DOT_SNAPSHOT_DIR_BYTES)) {
1144        Preconditions.checkArgument(
1145            renameReservedMap.containsKey(HdfsConstants.DOT_SNAPSHOT_DIR),
1146            RESERVED_ERROR_MSG);
1147        component =
1148            DFSUtil.string2Bytes(renameReservedMap
1149                .get(HdfsConstants.DOT_SNAPSHOT_DIR));
1150      }
1151    }
1152    return component;
1153  }
1154
1155  /**
1156   * Same as {@link #renameReservedPathsOnUpgrade(String)}, but for a single
1157   * byte array path component.
1158   */
1159  private static byte[] renameReservedRootComponentOnUpgrade(byte[] component,
1160      final int layoutVersion) {
1161    // If the LV doesn't support inode IDs, we're doing an upgrade
1162    if (!NameNodeLayoutVersion.supports(Feature.ADD_INODE_ID, layoutVersion)) {
1163      if (Arrays.equals(component, FSDirectory.DOT_RESERVED)) {
1164        Preconditions.checkArgument(
1165            renameReservedMap.containsKey(FSDirectory.DOT_RESERVED_STRING),
1166            RESERVED_ERROR_MSG);
1167        final String renameString = renameReservedMap
1168            .get(FSDirectory.DOT_RESERVED_STRING);
1169        component =
1170            DFSUtil.string2Bytes(renameString);
1171        LOG.info("Renamed root path " + FSDirectory.DOT_RESERVED_STRING
1172            + " to " + renameString);
1173      }
1174    }
1175    return component;
1176  }
1177
1178  /**
1179   * A one-shot class responsible for writing an image file.
1180   * The write() function should be called once, after which the getter
1181   * functions may be used to retrieve information about the file that was written.
1182   *
1183   * This is replaced by the PB-based FSImage. The class is to maintain
1184   * compatibility for the external fsimage tool.
1185   */
1186  @Deprecated
1187  static class Saver {
1188    private static final int LAYOUT_VERSION = -51;
1189    private final SaveNamespaceContext context;
1190    /** Set to true once an image has been written */
1191    private boolean saved = false;
1192
1193    /** The MD5 checksum of the file that was written */
1194    private MD5Hash savedDigest;
1195    private final ReferenceMap referenceMap = new ReferenceMap();
1196
1197    private final Map<Long, INodeFile> snapshotUCMap =
1198        new HashMap<Long, INodeFile>();
1199
1200    /** @throws IllegalStateException if the instance has not yet saved an image */
1201    private void checkSaved() {
1202      if (!saved) {
1203        throw new IllegalStateException("FSImageSaver has not saved an image");
1204      }
1205    }
1206
1207    /** @throws IllegalStateException if the instance has already saved an image */
1208    private void checkNotSaved() {
1209      if (saved) {
1210        throw new IllegalStateException("FSImageSaver has already saved an image");
1211      }
1212    }
1213
1214
1215    Saver(SaveNamespaceContext context) {
1216      this.context = context;
1217    }
1218
1219    /**
1220     * Return the MD5 checksum of the image file that was saved.
1221     */
1222    MD5Hash getSavedDigest() {
1223      checkSaved();
1224      return savedDigest;
1225    }
1226
1227    void save(File newFile, FSImageCompression compression) throws IOException {
1228      checkNotSaved();
1229
1230      final FSNamesystem sourceNamesystem = context.getSourceNamesystem();
1231      final INodeDirectory rootDir = sourceNamesystem.dir.rootDir;
1232      final long numINodes = rootDir.getDirectoryWithQuotaFeature()
1233          .getSpaceConsumed().get(Quota.NAMESPACE);
1234      String sdPath = newFile.getParentFile().getParentFile().getAbsolutePath();
1235      Step step = new Step(StepType.INODES, sdPath);
1236      StartupProgress prog = NameNode.getStartupProgress();
1237      prog.beginStep(Phase.SAVING_CHECKPOINT, step);
1238      prog.setTotal(Phase.SAVING_CHECKPOINT, step, numINodes);
1239      Counter counter = prog.getCounter(Phase.SAVING_CHECKPOINT, step);
1240      long startTime = now();
1241      //
1242      // Write out data
1243      //
1244      MessageDigest digester = MD5Hash.getDigester();
1245      FileOutputStream fout = new FileOutputStream(newFile);
1246      DigestOutputStream fos = new DigestOutputStream(fout, digester);
1247      DataOutputStream out = new DataOutputStream(fos);
1248      try {
1249        out.writeInt(LAYOUT_VERSION);
1250        LayoutFlags.write(out);
1251        // We use the non-locked version of getNamespaceInfo here since
1252        // the coordinating thread of saveNamespace already has read-locked
1253        // the namespace for us. If we attempt to take another readlock
1254        // from the actual saver thread, there's a potential of a
1255        // fairness-related deadlock. See the comments on HDFS-2223.
1256        out.writeInt(sourceNamesystem.unprotectedGetNamespaceInfo()
1257            .getNamespaceID());
1258        out.writeLong(numINodes);
1259        out.writeLong(sourceNamesystem.getGenerationStampV1());
1260        out.writeLong(sourceNamesystem.getGenerationStampV2());
1261        out.writeLong(sourceNamesystem.getGenerationStampAtblockIdSwitch());
1262        out.writeLong(sourceNamesystem.getLastAllocatedBlockId());
1263        out.writeLong(context.getTxId());
1264        out.writeLong(sourceNamesystem.getLastInodeId());
1265
1266
1267        sourceNamesystem.getSnapshotManager().write(out);
1268
1269        // write compression info and set up compressed stream
1270        out = compression.writeHeaderAndWrapStream(fos);
1271        LOG.info("Saving image file " + newFile +
1272                 " using " + compression);
1273
1274        // save the root
1275        saveINode2Image(rootDir, out, false, referenceMap, counter);
1276        // save the rest of the nodes
1277        saveImage(rootDir, out, true, false, counter);
1278        prog.endStep(Phase.SAVING_CHECKPOINT, step);
1279        // Now that the step is finished, set counter equal to total to adjust
1280        // for possible under-counting due to reference inodes.
1281        prog.setCount(Phase.SAVING_CHECKPOINT, step, numINodes);
1282        // save files under construction
1283        // TODO: for HDFS-5428, since we cannot break the compatibility of
1284        // fsimage, we store part of the under-construction files that are only
1285        // in snapshots in this "under-construction-file" section. As a
1286        // temporary solution, we use "/.reserved/.inodes/<inodeid>" as their
1287        // paths, so that when loading fsimage we do not put them into the lease
1288        // map. In the future, we can remove this hack when we can bump the
1289        // layout version.
1290        sourceNamesystem.saveFilesUnderConstruction(out, snapshotUCMap);
1291
1292        context.checkCancelled();
1293        sourceNamesystem.saveSecretManagerStateCompat(out, sdPath);
1294        context.checkCancelled();
1295        sourceNamesystem.getCacheManager().saveStateCompat(out, sdPath);
1296        context.checkCancelled();
1297        out.flush();
1298        context.checkCancelled();
1299        fout.getChannel().force(true);
1300      } finally {
1301        out.close();
1302      }
1303
1304      saved = true;
1305      // set md5 of the saved image
1306      savedDigest = new MD5Hash(digester.digest());
1307
1308      LOG.info("Image file " + newFile + " of size " + newFile.length() +
1309          " bytes saved in " + (now() - startTime)/1000 + " seconds.");
1310    }
1311
1312    /**
1313     * Save children INodes.
1314     * @param children The list of children INodes
1315     * @param out The DataOutputStream to write
1316     * @param inSnapshot Whether the parent directory or its ancestor is in
1317     *                   the deleted list of some snapshot (caused by rename or
1318     *                   deletion)
1319     * @param counter Counter to increment for namenode startup progress
1320     * @return Number of children that are directory
1321     */
1322    private int saveChildren(ReadOnlyList<INode> children,
1323        DataOutputStream out, boolean inSnapshot, Counter counter)
1324        throws IOException {
1325      // Write normal children INode.
1326      out.writeInt(children.size());
1327      int dirNum = 0;
1328      int i = 0;
1329      for(INode child : children) {
1330        // print all children first
1331        // TODO: for HDFS-5428, we cannot change the format/content of fsimage
1332        // here, thus even if the parent directory is in snapshot, we still
1333        // do not handle INodeUC as those stored in deleted list
1334        saveINode2Image(child, out, false, referenceMap, counter);
1335        if (child.isDirectory()) {
1336          dirNum++;
1337        } else if (inSnapshot && child.isFile()
1338            && child.asFile().isUnderConstruction()) {
1339          this.snapshotUCMap.put(child.getId(), child.asFile());
1340        }
1341        if (i++ % 50 == 0) {
1342          context.checkCancelled();
1343        }
1344      }
1345      return dirNum;
1346    }
1347
1348    /**
1349     * Save file tree image starting from the given root.
1350     * This is a recursive procedure, which first saves all children and
1351     * snapshot diffs of a current directory and then moves inside the
1352     * sub-directories.
1353     *
1354     * @param current The current node
1355     * @param out The DataoutputStream to write the image
1356     * @param toSaveSubtree Whether or not to save the subtree to fsimage. For
1357     *                      reference node, its subtree may already have been
1358     *                      saved before.
1359     * @param inSnapshot Whether the current directory is in snapshot
1360     * @param counter Counter to increment for namenode startup progress
1361     */
1362    private void saveImage(INodeDirectory current, DataOutputStream out,
1363        boolean toSaveSubtree, boolean inSnapshot, Counter counter)
1364        throws IOException {
1365      // write the inode id of the directory
1366      out.writeLong(current.getId());
1367
1368      if (!toSaveSubtree) {
1369        return;
1370      }
1371
1372      final ReadOnlyList<INode> children = current
1373          .getChildrenList(Snapshot.CURRENT_STATE_ID);
1374      int dirNum = 0;
1375      List<INodeDirectory> snapshotDirs = null;
1376      DirectoryWithSnapshotFeature sf = current.getDirectoryWithSnapshotFeature();
1377      if (sf != null) {
1378        snapshotDirs = new ArrayList<INodeDirectory>();
1379        sf.getSnapshotDirectory(snapshotDirs);
1380        dirNum += snapshotDirs.size();
1381      }
1382
1383      // 2. Write INodeDirectorySnapshottable#snapshotsByNames to record all
1384      // Snapshots
1385      if (current.isDirectory() && current.asDirectory().isSnapshottable()) {
1386        SnapshotFSImageFormat.saveSnapshots(current.asDirectory(), out);
1387      } else {
1388        out.writeInt(-1); // # of snapshots
1389      }
1390
1391      // 3. Write children INode
1392      dirNum += saveChildren(children, out, inSnapshot, counter);
1393
1394      // 4. Write DirectoryDiff lists, if there is any.
1395      SnapshotFSImageFormat.saveDirectoryDiffList(current, out, referenceMap);
1396
1397      // Write sub-tree of sub-directories, including possible snapshots of
1398      // deleted sub-directories
1399      out.writeInt(dirNum); // the number of sub-directories
1400      for(INode child : children) {
1401        if(!child.isDirectory()) {
1402          continue;
1403        }
1404        // make sure we only save the subtree under a reference node once
1405        boolean toSave = child.isReference() ?
1406            referenceMap.toProcessSubtree(child.getId()) : true;
1407        saveImage(child.asDirectory(), out, toSave, inSnapshot, counter);
1408      }
1409      if (snapshotDirs != null) {
1410        for (INodeDirectory subDir : snapshotDirs) {
1411          // make sure we only save the subtree under a reference node once
1412          boolean toSave = subDir.getParentReference() != null ?
1413              referenceMap.toProcessSubtree(subDir.getId()) : true;
1414          saveImage(subDir, out, toSave, true, counter);
1415        }
1416      }
1417    }
1418
1419    /**
1420     * Saves inode and increments progress counter.
1421     *
1422     * @param inode INode to save
1423     * @param out DataOutputStream to receive inode
1424     * @param writeUnderConstruction boolean true if this is under construction
1425     * @param referenceMap ReferenceMap containing reference inodes
1426     * @param counter Counter to increment for namenode startup progress
1427     * @throws IOException thrown if there is an I/O error
1428     */
1429    private void saveINode2Image(INode inode, DataOutputStream out,
1430        boolean writeUnderConstruction, ReferenceMap referenceMap,
1431        Counter counter) throws IOException {
1432      FSImageSerialization.saveINode2Image(inode, out, writeUnderConstruction,
1433        referenceMap);
1434      // Intentionally do not increment counter for reference inodes, because it
1435      // is too difficult at this point to assess whether or not this is a
1436      // reference that counts toward quota.
1437      if (!(inode instanceof INodeReference)) {
1438        counter.increment();
1439      }
1440    }
1441  }
1442}