View Javadoc

1   /*
2    * Licensed to the Apache Software Foundation (ASF) under one
3    * or more contributor license agreements.  See the NOTICE file
4    * distributed with this work for additional information
5    * regarding copyright ownership.  The ASF licenses this file
6    * to you under the Apache License, Version 2.0 (the
7    * "License"); you may not use this file except in compliance
8    * with the License.  You may obtain a copy of the License at
9    *
10   *     http://www.apache.org/licenses/LICENSE-2.0
11   *
12   * Unless required by applicable law or agreed to in writing, software
13   * distributed under the License is distributed on an "AS IS" BASIS,
14   * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15   * See the License for the specific language governing permissions and
16   * limitations under the License.
17   */
18  package org.apache.hadoop.hbase.io.hfile;
19  
20  import java.io.ByteArrayOutputStream;
21  import java.io.DataInputStream;
22  import java.io.DataOutput;
23  import java.io.DataOutputStream;
24  import java.io.IOException;
25  import java.io.InputStream;
26  import java.nio.ByteBuffer;
27  import java.util.concurrent.locks.Lock;
28  import java.util.concurrent.locks.ReentrantLock;
29  
30  import org.apache.hadoop.fs.FSDataInputStream;
31  import org.apache.hadoop.fs.FSDataOutputStream;
32  import org.apache.hadoop.fs.Path;
33  import org.apache.hadoop.hbase.Cell;
34  import org.apache.hadoop.hbase.HConstants;
35  import org.apache.hadoop.hbase.classification.InterfaceAudience;
36  import org.apache.hadoop.hbase.fs.HFileSystem;
37  import org.apache.hadoop.hbase.io.ByteBufferInputStream;
38  import org.apache.hadoop.hbase.io.FSDataInputStreamWrapper;
39  import org.apache.hadoop.hbase.io.encoding.DataBlockEncoding;
40  import org.apache.hadoop.hbase.io.encoding.HFileBlockDecodingContext;
41  import org.apache.hadoop.hbase.io.encoding.HFileBlockDefaultDecodingContext;
42  import org.apache.hadoop.hbase.io.encoding.HFileBlockDefaultEncodingContext;
43  import org.apache.hadoop.hbase.io.encoding.HFileBlockEncodingContext;
44  import org.apache.hadoop.hbase.util.ByteBufferUtils;
45  import org.apache.hadoop.hbase.util.Bytes;
46  import org.apache.hadoop.hbase.util.ChecksumType;
47  import org.apache.hadoop.hbase.util.ClassSize;
48  import org.apache.hadoop.io.IOUtils;
49  
50  import com.google.common.base.Preconditions;
51  
52  /**
53   * Reading {@link HFile} version 1 and 2 blocks, and writing version 2 blocks.
54   * <ul>
55   * <li>In version 1 all blocks are always compressed or uncompressed, as
56   * specified by the {@link HFile}'s compression algorithm, with a type-specific
57   * magic record stored in the beginning of the compressed data (i.e. one needs
58   * to uncompress the compressed block to determine the block type). There is
59   * only a single compression algorithm setting for all blocks. Offset and size
60   * information from the block index are required to read a block.
61   * <li>In version 2 a block is structured as follows:
62   * <ul>
63   * <li>header (see Writer#finishBlock())
64   * <ul>
65   * <li>Magic record identifying the block type (8 bytes)
66   * <li>Compressed block size, excluding header, including checksum (4 bytes)
67   * <li>Uncompressed block size, excluding header, excluding checksum (4 bytes)
68   * <li>The offset of the previous block of the same type (8 bytes). This is
69   * used to be able to navigate to the previous block without going to the block
70   * <li>For minorVersions >=1, the ordinal describing checksum type (1 byte)
71   * <li>For minorVersions >=1, the number of data bytes/checksum chunk (4 bytes)
72   * <li>For minorVersions >=1, the size of data on disk, including header,
73   * excluding checksums (4 bytes)
74   * </ul>
75   * </li>
76   * <li>Raw/Compressed/Encrypted/Encoded data. The compression algorithm is the
77   * same for all the blocks in the {@link HFile}, similarly to what was done in
78   * version 1.
79   * <li>For minorVersions >=1, a series of 4 byte checksums, one each for
80   * the number of bytes specified by bytesPerChecksum.
81   * </ul>
82   * </ul>
83   */
84  @InterfaceAudience.Private
85  public class HFileBlock implements Cacheable {
86  
87    /**
88     * On a checksum failure on a Reader, these many suceeding read
89     * requests switch back to using hdfs checksums before auto-reenabling
90     * hbase checksum verification.
91     */
92    static final int CHECKSUM_VERIFICATION_NUM_IO_THRESHOLD = 3;
93  
94    public static final boolean FILL_HEADER = true;
95    public static final boolean DONT_FILL_HEADER = false;
96  
97    /**
98     * The size of block header when blockType is {@link BlockType#ENCODED_DATA}.
99     * This extends normal header by adding the id of encoder.
100    */
101   public static final int ENCODED_HEADER_SIZE = HConstants.HFILEBLOCK_HEADER_SIZE
102       + DataBlockEncoding.ID_SIZE;
103 
104   static final byte[] DUMMY_HEADER_NO_CHECKSUM =
105      new byte[HConstants.HFILEBLOCK_HEADER_SIZE_NO_CHECKSUM];
106 
107   public static final int BYTE_BUFFER_HEAP_SIZE = (int) ClassSize.estimateBase(
108       ByteBuffer.wrap(new byte[0], 0, 0).getClass(), false);
109 
110   // meta.usesHBaseChecksum+offset+nextBlockOnDiskSizeWithHeader
111   public static final int EXTRA_SERIALIZATION_SPACE = Bytes.SIZEOF_BYTE + Bytes.SIZEOF_INT
112       + Bytes.SIZEOF_LONG;
113 
114   /**
115    * Each checksum value is an integer that can be stored in 4 bytes.
116    */
117   static final int CHECKSUM_SIZE = Bytes.SIZEOF_INT;
118 
119   private static final CacheableDeserializer<Cacheable> blockDeserializer =
120       new CacheableDeserializer<Cacheable>() {
121         public HFileBlock deserialize(ByteBuffer buf, boolean reuse) throws IOException{
122           buf.limit(buf.limit() - HFileBlock.EXTRA_SERIALIZATION_SPACE).rewind();
123           ByteBuffer newByteBuffer;
124           if (reuse) {
125             newByteBuffer = buf.slice();
126           } else {
127            newByteBuffer = ByteBuffer.allocate(buf.limit());
128            newByteBuffer.put(buf);
129           }
130           buf.position(buf.limit());
131           buf.limit(buf.limit() + HFileBlock.EXTRA_SERIALIZATION_SPACE);
132           boolean usesChecksum = buf.get() == (byte)1;
133           HFileBlock ourBuffer = new HFileBlock(newByteBuffer, usesChecksum);
134           ourBuffer.offset = buf.getLong();
135           ourBuffer.nextBlockOnDiskSizeWithHeader = buf.getInt();
136           if (ourBuffer.hasNextBlockHeader()) {
137             ourBuffer.buf.limit(ourBuffer.buf.limit() - ourBuffer.headerSize());
138           }
139           return ourBuffer;
140         }
141 
142         @Override
143         public int getDeserialiserIdentifier() {
144           return deserializerIdentifier;
145         }
146 
147         @Override
148         public HFileBlock deserialize(ByteBuffer b) throws IOException {
149           return deserialize(b, false);
150         }
151       };
152   private static final int deserializerIdentifier;
153   static {
154     deserializerIdentifier = CacheableDeserializerIdManager
155         .registerDeserializer(blockDeserializer);
156   }
157 
158   /** Type of block. Header field 0. */
159   private BlockType blockType;
160 
161   /** Size on disk excluding header, including checksum. Header field 1. */
162   private int onDiskSizeWithoutHeader;
163 
164   /** Size of pure data. Does not include header or checksums. Header field 2. */
165   private final int uncompressedSizeWithoutHeader;
166 
167   /** The offset of the previous block on disk. Header field 3. */
168   private final long prevBlockOffset;
169 
170   /**
171    * Size on disk of header + data. Excludes checksum. Header field 6,
172    * OR calculated from {@link #onDiskSizeWithoutHeader} when using HDFS checksum.
173    */
174   private final int onDiskDataSizeWithHeader;
175 
176   /** The in-memory representation of the hfile block */
177   private ByteBuffer buf;
178 
179   /** Meta data that holds meta information on the hfileblock */
180   private HFileContext fileContext;
181 
182   /**
183    * The offset of this block in the file. Populated by the reader for
184    * convenience of access. This offset is not part of the block header.
185    */
186   private long offset = -1;
187 
188   /**
189    * The on-disk size of the next block, including the header, obtained by
190    * peeking into the first {@link HConstants#HFILEBLOCK_HEADER_SIZE} bytes of the next block's
191    * header, or -1 if unknown.
192    */
193   private int nextBlockOnDiskSizeWithHeader = -1;
194 
195   /**
196    * Creates a new {@link HFile} block from the given fields. This constructor
197    * is mostly used when the block data has already been read and uncompressed,
198    * and is sitting in a byte buffer.
199    *
200    * @param blockType the type of this block, see {@link BlockType}
201    * @param onDiskSizeWithoutHeader see {@link #onDiskSizeWithoutHeader}
202    * @param uncompressedSizeWithoutHeader see {@link #uncompressedSizeWithoutHeader}
203    * @param prevBlockOffset see {@link #prevBlockOffset}
204    * @param buf block header ({@link HConstants#HFILEBLOCK_HEADER_SIZE} bytes) followed by
205    *          uncompressed data. This
206    * @param fillHeader when true, parse {@code buf} and override the first 4 header fields.
207    * @param offset the file offset the block was read from
208    * @param onDiskDataSizeWithHeader see {@link #onDiskDataSizeWithHeader}
209    * @param fileContext HFile meta data
210    */
211   HFileBlock(BlockType blockType, int onDiskSizeWithoutHeader, int uncompressedSizeWithoutHeader,
212       long prevBlockOffset, ByteBuffer buf, boolean fillHeader, long offset,
213       int onDiskDataSizeWithHeader, HFileContext fileContext) {
214     this.blockType = blockType;
215     this.onDiskSizeWithoutHeader = onDiskSizeWithoutHeader;
216     this.uncompressedSizeWithoutHeader = uncompressedSizeWithoutHeader;
217     this.prevBlockOffset = prevBlockOffset;
218     this.buf = buf;
219     this.offset = offset;
220     this.onDiskDataSizeWithHeader = onDiskDataSizeWithHeader;
221     this.fileContext = fileContext;
222     if (fillHeader)
223       overwriteHeader();
224     this.buf.rewind();
225   }
226 
227   /**
228    * Copy constructor. Creates a shallow copy of {@code that}'s buffer.
229    */
230   HFileBlock(HFileBlock that) {
231     this.blockType = that.blockType;
232     this.onDiskSizeWithoutHeader = that.onDiskSizeWithoutHeader;
233     this.uncompressedSizeWithoutHeader = that.uncompressedSizeWithoutHeader;
234     this.prevBlockOffset = that.prevBlockOffset;
235     this.buf = that.buf.duplicate();
236     this.offset = that.offset;
237     this.onDiskDataSizeWithHeader = that.onDiskDataSizeWithHeader;
238     this.fileContext = that.fileContext;
239     this.nextBlockOnDiskSizeWithHeader = that.nextBlockOnDiskSizeWithHeader;
240   }
241 
242   /**
243    * Creates a block from an existing buffer starting with a header. Rewinds
244    * and takes ownership of the buffer. By definition of rewind, ignores the
245    * buffer position, but if you slice the buffer beforehand, it will rewind
246    * to that point. The reason this has a minorNumber and not a majorNumber is
247    * because majorNumbers indicate the format of a HFile whereas minorNumbers
248    * indicate the format inside a HFileBlock.
249    */
250   HFileBlock(ByteBuffer b, boolean usesHBaseChecksum) throws IOException {
251     b.rewind();
252     blockType = BlockType.read(b);
253     onDiskSizeWithoutHeader = b.getInt();
254     uncompressedSizeWithoutHeader = b.getInt();
255     prevBlockOffset = b.getLong();
256     HFileContextBuilder contextBuilder = new HFileContextBuilder();
257     contextBuilder.withHBaseCheckSum(usesHBaseChecksum);
258     if (usesHBaseChecksum) {
259       contextBuilder.withChecksumType(ChecksumType.codeToType(b.get()));
260       contextBuilder.withBytesPerCheckSum(b.getInt());
261       this.onDiskDataSizeWithHeader = b.getInt();
262     } else {
263       contextBuilder.withChecksumType(ChecksumType.NULL);
264       contextBuilder.withBytesPerCheckSum(0);
265       this.onDiskDataSizeWithHeader = onDiskSizeWithoutHeader +
266                                        HConstants.HFILEBLOCK_HEADER_SIZE_NO_CHECKSUM;
267     }
268     this.fileContext = contextBuilder.build();
269     buf = b;
270     buf.rewind();
271   }
272 
273   public BlockType getBlockType() {
274     return blockType;
275   }
276 
277   /** @return get data block encoding id that was used to encode this block */
278   public short getDataBlockEncodingId() {
279     if (blockType != BlockType.ENCODED_DATA) {
280       throw new IllegalArgumentException("Querying encoder ID of a block " +
281           "of type other than " + BlockType.ENCODED_DATA + ": " + blockType);
282     }
283     return buf.getShort(headerSize());
284   }
285 
286   /**
287    * @return the on-disk size of header + data part + checksum.
288    */
289   public int getOnDiskSizeWithHeader() {
290     return onDiskSizeWithoutHeader + headerSize();
291   }
292 
293   /**
294    * @return the on-disk size of the data part + checksum (header excluded).
295    */
296   public int getOnDiskSizeWithoutHeader() {
297     return onDiskSizeWithoutHeader;
298   }
299 
300   /**
301    * @return the uncompressed size of data part (header and checksum excluded).
302    */
303    public int getUncompressedSizeWithoutHeader() {
304     return uncompressedSizeWithoutHeader;
305   }
306 
307   /**
308    * @return the offset of the previous block of the same type in the file, or
309    *         -1 if unknown
310    */
311   public long getPrevBlockOffset() {
312     return prevBlockOffset;
313   }
314 
315   /**
316    * Rewinds {@code buf} and writes first 4 header fields. {@code buf} position
317    * is modified as side-effect.
318    */
319   private void overwriteHeader() {
320     buf.rewind();
321     blockType.write(buf);
322     buf.putInt(onDiskSizeWithoutHeader);
323     buf.putInt(uncompressedSizeWithoutHeader);
324     buf.putLong(prevBlockOffset);
325     if (this.fileContext.isUseHBaseChecksum()) {
326       buf.put(fileContext.getChecksumType().getCode());
327       buf.putInt(fileContext.getBytesPerChecksum());
328       buf.putInt(onDiskDataSizeWithHeader);
329     }
330   }
331 
332   /**
333    * Returns a buffer that does not include the header or checksum.
334    *
335    * @return the buffer with header skipped and checksum omitted.
336    */
337   public ByteBuffer getBufferWithoutHeader() {
338     ByteBuffer dup = this.buf.duplicate();
339     dup.position(headerSize());
340     dup.limit(buf.limit() - totalChecksumBytes());
341     return dup.slice();
342   }
343 
344   /**
345    * Returns the buffer this block stores internally. The clients must not
346    * modify the buffer object. This method has to be public because it is
347    * used in {@link org.apache.hadoop.hbase.util.CompoundBloomFilter} 
348    * to avoid object creation on every Bloom filter lookup, but has to 
349    * be used with caution. Checksum data is not included in the returned 
350    * buffer but header data is.
351    *
352    * @return the buffer of this block for read-only operations
353    */
354   public ByteBuffer getBufferReadOnly() {
355     ByteBuffer dup = this.buf.duplicate();
356     dup.limit(buf.limit() - totalChecksumBytes());
357     return dup.slice();
358   }
359 
360   /**
361    * Returns the buffer of this block, including header data. The clients must
362    * not modify the buffer object. This method has to be public because it is
363    * used in {@link org.apache.hadoop.hbase.io.hfile.bucket.BucketCache} to avoid buffer copy.
364    *
365    * @return the buffer with header and checksum included for read-only operations
366    */
367   public ByteBuffer getBufferReadOnlyWithHeader() {
368     ByteBuffer dup = this.buf.duplicate();
369     return dup.slice();
370   }
371 
372   /**
373    * Returns a byte buffer of this block, including header data and checksum, positioned at
374    * the beginning of header. The underlying data array is not copied.
375    *
376    * @return the byte buffer with header and checksum included
377    */
378   ByteBuffer getBufferWithHeader() {
379     ByteBuffer dupBuf = buf.duplicate();
380     dupBuf.rewind();
381     return dupBuf;
382   }
383 
384   private void sanityCheckAssertion(long valueFromBuf, long valueFromField,
385       String fieldName) throws IOException {
386     if (valueFromBuf != valueFromField) {
387       throw new AssertionError(fieldName + " in the buffer (" + valueFromBuf
388           + ") is different from that in the field (" + valueFromField + ")");
389     }
390   }
391 
392   private void sanityCheckAssertion(BlockType valueFromBuf, BlockType valueFromField)
393       throws IOException {
394     if (valueFromBuf != valueFromField) {
395       throw new IOException("Block type stored in the buffer: " +
396         valueFromBuf + ", block type field: " + valueFromField);
397     }
398   }
399 
400   /**
401    * Checks if the block is internally consistent, i.e. the first
402    * {@link HConstants#HFILEBLOCK_HEADER_SIZE} bytes of the buffer contain a
403    * valid header consistent with the fields. Assumes a packed block structure.
404    * This function is primary for testing and debugging, and is not
405    * thread-safe, because it alters the internal buffer pointer.
406    */
407   void sanityCheck() throws IOException {
408     buf.rewind();
409 
410     sanityCheckAssertion(BlockType.read(buf), blockType);
411 
412     sanityCheckAssertion(buf.getInt(), onDiskSizeWithoutHeader,
413         "onDiskSizeWithoutHeader");
414 
415     sanityCheckAssertion(buf.getInt(), uncompressedSizeWithoutHeader,
416         "uncompressedSizeWithoutHeader");
417 
418     sanityCheckAssertion(buf.getLong(), prevBlockOffset, "prevBlocKOffset");
419     if (this.fileContext.isUseHBaseChecksum()) {
420       sanityCheckAssertion(buf.get(), this.fileContext.getChecksumType().getCode(), "checksumType");
421       sanityCheckAssertion(buf.getInt(), this.fileContext.getBytesPerChecksum(), "bytesPerChecksum");
422       sanityCheckAssertion(buf.getInt(), onDiskDataSizeWithHeader, "onDiskDataSizeWithHeader");
423     }
424 
425     int cksumBytes = totalChecksumBytes();
426     int expectedBufLimit = onDiskDataSizeWithHeader + cksumBytes;
427     if (buf.limit() != expectedBufLimit) {
428       throw new AssertionError("Expected buffer limit " + expectedBufLimit
429           + ", got " + buf.limit());
430     }
431 
432     // We might optionally allocate HFILEBLOCK_HEADER_SIZE more bytes to read the next
433     // block's header, so there are two sensible values for buffer capacity.
434     int hdrSize = headerSize();
435     if (buf.capacity() != expectedBufLimit &&
436         buf.capacity() != expectedBufLimit + hdrSize) {
437       throw new AssertionError("Invalid buffer capacity: " + buf.capacity() +
438           ", expected " + expectedBufLimit + " or " + (expectedBufLimit + hdrSize));
439     }
440   }
441 
442   @Override
443   public String toString() {
444     StringBuilder sb = new StringBuilder()
445       .append("HFileBlock [")
446       .append(" fileOffset=").append(offset)
447       .append(" headerSize()=").append(headerSize())
448       .append(" blockType=").append(blockType)
449       .append(" onDiskSizeWithoutHeader=").append(onDiskSizeWithoutHeader)
450       .append(" uncompressedSizeWithoutHeader=").append(uncompressedSizeWithoutHeader)
451       .append(" prevBlockOffset=").append(prevBlockOffset)
452       .append(" isUseHBaseChecksum()=").append(fileContext.isUseHBaseChecksum());
453     if (fileContext.isUseHBaseChecksum()) {
454       sb.append(" checksumType=").append(ChecksumType.codeToType(this.buf.get(24)))
455         .append(" bytesPerChecksum=").append(this.buf.getInt(24 + 1))
456         .append(" onDiskDataSizeWithHeader=").append(onDiskDataSizeWithHeader);
457     } else {
458       sb.append(" onDiskDataSizeWithHeader=").append(onDiskDataSizeWithHeader)
459         .append("(").append(onDiskSizeWithoutHeader)
460         .append("+").append(HConstants.HFILEBLOCK_HEADER_SIZE_NO_CHECKSUM).append(")");
461     }
462     String dataBegin = null;
463     if (buf.hasArray()) {
464       dataBegin = Bytes.toStringBinary(buf.array(), buf.arrayOffset() + headerSize(),
465           Math.min(32, buf.limit() - buf.arrayOffset() - headerSize()));
466     } else {
467       ByteBuffer bufWithoutHeader = getBufferWithoutHeader();
468       byte[] dataBeginBytes = new byte[Math.min(32,
469           bufWithoutHeader.limit() - bufWithoutHeader.position())];
470       bufWithoutHeader.get(dataBeginBytes);
471       dataBegin = Bytes.toStringBinary(dataBeginBytes);
472     }
473     sb.append(" getOnDiskSizeWithHeader()=").append(getOnDiskSizeWithHeader())
474       .append(" totalChecksumBytes()=").append(totalChecksumBytes())
475       .append(" isUnpacked()=").append(isUnpacked())
476       .append(" buf=[ ").append(buf).append(" ]")
477       .append(" dataBeginsWith=").append(dataBegin)
478       .append(" fileContext=").append(fileContext)
479       .append(" ]");
480     return sb.toString();
481   }
482 
483   /**
484    * Called after reading a block with provided onDiskSizeWithHeader.
485    */
486   private void validateOnDiskSizeWithoutHeader(int expectedOnDiskSizeWithoutHeader)
487   throws IOException {
488     if (onDiskSizeWithoutHeader != expectedOnDiskSizeWithoutHeader) {
489       String dataBegin = null;
490       if (buf.hasArray()) {
491         dataBegin = Bytes.toStringBinary(buf.array(), buf.arrayOffset(), Math.min(32, buf.limit()));
492       } else {
493         ByteBuffer bufDup = getBufferReadOnly();
494         byte[] dataBeginBytes = new byte[Math.min(32, bufDup.limit() - bufDup.position())];
495         bufDup.get(dataBeginBytes);
496         dataBegin = Bytes.toStringBinary(dataBeginBytes);
497       }
498       String blockInfoMsg =
499         "Block offset: " + offset + ", data starts with: " + dataBegin;
500       throw new IOException("On-disk size without header provided is "
501           + expectedOnDiskSizeWithoutHeader + ", but block "
502           + "header contains " + onDiskSizeWithoutHeader + ". " +
503           blockInfoMsg);
504     }
505   }
506 
507   /**
508    * Retrieves the decompressed/decrypted view of this block. An encoded block remains in its
509    * encoded structure. Internal structures are shared between instances where applicable.
510    */
511   HFileBlock unpack(HFileContext fileContext, FSReader reader) throws IOException {
512     if (!fileContext.isCompressedOrEncrypted()) {
513       // TODO: cannot use our own fileContext here because HFileBlock(ByteBuffer, boolean),
514       // which is used for block serialization to L2 cache, does not preserve encoding and
515       // encryption details.
516       return this;
517     }
518 
519     HFileBlock unpacked = new HFileBlock(this);
520     unpacked.allocateBuffer(); // allocates space for the decompressed block
521 
522     HFileBlockDecodingContext ctx = blockType == BlockType.ENCODED_DATA ?
523       reader.getBlockDecodingContext() : reader.getDefaultBlockDecodingContext();
524 
525     ByteBuffer dup = this.buf.duplicate();
526     dup.position(this.headerSize());
527     dup = dup.slice();
528     ctx.prepareDecoding(unpacked.getOnDiskSizeWithoutHeader(),
529       unpacked.getUncompressedSizeWithoutHeader(), unpacked.getBufferWithoutHeader(),
530       dup);
531 
532     // Preserve the next block's header bytes in the new block if we have them.
533     if (unpacked.hasNextBlockHeader()) {
534       // Both the buffers are limited till checksum bytes and avoid the next block's header.
535       // Below call to copyFromBufferToBuffer() will try positional read/write from/to buffers when
536       // any of the buffer is DBB. So we change the limit on a dup buffer. No copying just create
537       // new BB objects
538       ByteBuffer inDup = this.buf.duplicate();
539       inDup.limit(inDup.limit() + headerSize());
540       ByteBuffer outDup = unpacked.buf.duplicate();
541       outDup.limit(outDup.limit() + unpacked.headerSize());
542       ByteBufferUtils.copyFromBufferToBuffer(
543           outDup,
544           inDup,
545           this.onDiskDataSizeWithHeader,
546           unpacked.headerSize() + unpacked.uncompressedSizeWithoutHeader
547               + unpacked.totalChecksumBytes(), unpacked.headerSize());
548     }
549     return unpacked;
550   }
551 
552   /**
553    * Return true when this buffer includes next block's header.
554    */
555   private boolean hasNextBlockHeader() {
556     return nextBlockOnDiskSizeWithHeader > 0;
557   }
558 
559   /**
560    * Always allocates a new buffer of the correct size. Copies header bytes
561    * from the existing buffer. Does not change header fields.
562    * Reserve room to keep checksum bytes too.
563    */
564   private void allocateBuffer() {
565     int cksumBytes = totalChecksumBytes();
566     int headerSize = headerSize();
567     int capacityNeeded = headerSize + uncompressedSizeWithoutHeader +
568         cksumBytes + (hasNextBlockHeader() ? headerSize : 0);
569 
570     // TODO we need consider allocating offheap here?
571     ByteBuffer newBuf = ByteBuffer.allocate(capacityNeeded);
572 
573     // Copy header bytes into newBuf.
574     // newBuf is HBB so no issue in calling array()
575     ByteBuffer dup = buf.duplicate();
576     dup.position(0);
577     dup.get(newBuf.array(), newBuf.arrayOffset(), headerSize);
578 
579     buf = newBuf;
580     // set limit to exclude next block's header
581     buf.limit(headerSize + uncompressedSizeWithoutHeader + cksumBytes);
582   }
583 
584   /**
585    * Return true when this block's buffer has been unpacked, false otherwise. Note this is a
586    * calculated heuristic, not tracked attribute of the block.
587    */
588   public boolean isUnpacked() {
589     final int cksumBytes = totalChecksumBytes();
590     final int headerSize = headerSize();
591     final int expectedCapacity = headerSize + uncompressedSizeWithoutHeader + cksumBytes;
592     final int bufCapacity = buf.capacity();
593     return bufCapacity == expectedCapacity || bufCapacity == expectedCapacity + headerSize;
594   }
595 
596   /** An additional sanity-check in case no compression or encryption is being used. */
597   public void assumeUncompressed() throws IOException {
598     if (onDiskSizeWithoutHeader != uncompressedSizeWithoutHeader +
599         totalChecksumBytes()) {
600       throw new IOException("Using no compression but "
601           + "onDiskSizeWithoutHeader=" + onDiskSizeWithoutHeader + ", "
602           + "uncompressedSizeWithoutHeader=" + uncompressedSizeWithoutHeader
603           + ", numChecksumbytes=" + totalChecksumBytes());
604     }
605   }
606 
607   /**
608    * @param expectedType the expected type of this block
609    * @throws IOException if this block's type is different than expected
610    */
611   public void expectType(BlockType expectedType) throws IOException {
612     if (blockType != expectedType) {
613       throw new IOException("Invalid block type: expected=" + expectedType
614           + ", actual=" + blockType);
615     }
616   }
617 
618   /** @return the offset of this block in the file it was read from */
619   public long getOffset() {
620     if (offset < 0) {
621       throw new IllegalStateException(
622           "HFile block offset not initialized properly");
623     }
624     return offset;
625   }
626 
627   /**
628    * @return a byte stream reading the data + checksum of this block
629    */
630   public DataInputStream getByteStream() {
631     ByteBuffer dup = this.buf.duplicate();
632     dup.position(this.headerSize());
633     return new DataInputStream(new ByteBufferInputStream(dup));
634   }
635 
636   @Override
637   public long heapSize() {
638     long size = ClassSize.align(
639         ClassSize.OBJECT +
640         // Block type, byte buffer and meta references
641         3 * ClassSize.REFERENCE +
642         // On-disk size, uncompressed size, and next block's on-disk size
643         // bytePerChecksum and onDiskDataSize
644         4 * Bytes.SIZEOF_INT +
645         // This and previous block offset
646         2 * Bytes.SIZEOF_LONG +
647         // Heap size of the meta object. meta will be always not null.
648         fileContext.heapSize()
649     );
650 
651     if (buf != null) {
652       // Deep overhead of the byte buffer. Needs to be aligned separately.
653       size += ClassSize.align(buf.capacity() + BYTE_BUFFER_HEAP_SIZE);
654     }
655 
656     return ClassSize.align(size);
657   }
658 
659   /**
660    * Read from an input stream. Analogous to
661    * {@link IOUtils#readFully(InputStream, byte[], int, int)}, but specifies a
662    * number of "extra" bytes that would be desirable but not absolutely
663    * necessary to read.
664    *
665    * @param in the input stream to read from
666    * @param buf the buffer to read into
667    * @param bufOffset the destination offset in the buffer
668    * @param necessaryLen the number of bytes that are absolutely necessary to
669    *          read
670    * @param extraLen the number of extra bytes that would be nice to read
671    * @return true if succeeded reading the extra bytes
672    * @throws IOException if failed to read the necessary bytes
673    */
674   public static boolean readWithExtra(InputStream in, byte buf[],
675       int bufOffset, int necessaryLen, int extraLen) throws IOException {
676     int bytesRemaining = necessaryLen + extraLen;
677     while (bytesRemaining > 0) {
678       int ret = in.read(buf, bufOffset, bytesRemaining);
679       if (ret == -1 && bytesRemaining <= extraLen) {
680         // We could not read the "extra data", but that is OK.
681         break;
682       }
683 
684       if (ret < 0) {
685         throw new IOException("Premature EOF from inputStream (read "
686             + "returned " + ret + ", was trying to read " + necessaryLen
687             + " necessary bytes and " + extraLen + " extra bytes, "
688             + "successfully read "
689             + (necessaryLen + extraLen - bytesRemaining));
690       }
691       bufOffset += ret;
692       bytesRemaining -= ret;
693     }
694     return bytesRemaining <= 0;
695   }
696 
697   /**
698    * @return the on-disk size of the next block (including the header size)
699    *         that was read by peeking into the next block's header
700    */
701   public int getNextBlockOnDiskSizeWithHeader() {
702     return nextBlockOnDiskSizeWithHeader;
703   }
704 
705   /**
706    * Unified version 2 {@link HFile} block writer. The intended usage pattern
707    * is as follows:
708    * <ol>
709    * <li>Construct an {@link HFileBlock.Writer}, providing a compression algorithm.
710    * <li>Call {@link Writer#startWriting} and get a data stream to write to.
711    * <li>Write your data into the stream.
712    * <li>Call {@link Writer#writeHeaderAndData(FSDataOutputStream)} as many times as you need to.
713    * store the serialized block into an external stream.
714    * <li>Repeat to write more blocks.
715    * </ol>
716    * <p>
717    */
718   public static class Writer {
719 
720     private enum State {
721       INIT,
722       WRITING,
723       BLOCK_READY
724     };
725 
726     /** Writer state. Used to ensure the correct usage protocol. */
727     private State state = State.INIT;
728 
729     /** Data block encoder used for data blocks */
730     private final HFileDataBlockEncoder dataBlockEncoder;
731 
732     private HFileBlockEncodingContext dataBlockEncodingCtx;
733 
734     /** block encoding context for non-data blocks */
735     private HFileBlockDefaultEncodingContext defaultBlockEncodingCtx;
736 
737     /**
738      * The stream we use to accumulate data in uncompressed format for each
739      * block. We reset this stream at the end of each block and reuse it. The
740      * header is written as the first {@link HConstants#HFILEBLOCK_HEADER_SIZE} bytes into this
741      * stream.
742      */
743     private ByteArrayOutputStream baosInMemory;
744 
745     /**
746      * Current block type. Set in {@link #startWriting(BlockType)}. Could be
747      * changed in {@link #finishBlock()} from {@link BlockType#DATA}
748      * to {@link BlockType#ENCODED_DATA}.
749      */
750     private BlockType blockType;
751 
752     /**
753      * A stream that we write uncompressed bytes to, which compresses them and
754      * writes them to {@link #baosInMemory}.
755      */
756     private DataOutputStream userDataStream;
757 
758     // Size of actual data being written. Not considering the block encoding/compression. This
759     // includes the header size also.
760     private int unencodedDataSizeWritten;
761 
762     /**
763      * Bytes to be written to the file system, including the header. Compressed
764      * if compression is turned on. It also includes the checksum data that
765      * immediately follows the block data. (header + data + checksums)
766      */
767     private byte[] onDiskBytesWithHeader;
768 
769     /**
770      * The size of the checksum data on disk. It is used only if data is
771      * not compressed. If data is compressed, then the checksums are already
772      * part of onDiskBytesWithHeader. If data is uncompressed, then this
773      * variable stores the checksum data for this block.
774      */
775     private byte[] onDiskChecksum;
776 
777     /**
778      * Valid in the READY state. Contains the header and the uncompressed (but
779      * potentially encoded, if this is a data block) bytes, so the length is
780      * {@link #uncompressedSizeWithoutHeader} + {@link org.apache.hadoop.hbase.HConstants#HFILEBLOCK_HEADER_SIZE}.
781      * Does not store checksums.
782      */
783     private byte[] uncompressedBytesWithHeader;
784 
785     /**
786      * Current block's start offset in the {@link HFile}. Set in
787      * {@link #writeHeaderAndData(FSDataOutputStream)}.
788      */
789     private long startOffset;
790 
791     /**
792      * Offset of previous block by block type. Updated when the next block is
793      * started.
794      */
795     private long[] prevOffsetByType;
796 
797     /** The offset of the previous block of the same type */
798     private long prevOffset;
799     /** Meta data that holds information about the hfileblock**/
800     private HFileContext fileContext;
801 
802     /**
803      * @param dataBlockEncoder data block encoding algorithm to use
804      */
805     public Writer(HFileDataBlockEncoder dataBlockEncoder, HFileContext fileContext) {
806       this.dataBlockEncoder = dataBlockEncoder != null
807           ? dataBlockEncoder : NoOpDataBlockEncoder.INSTANCE;
808       defaultBlockEncodingCtx = new HFileBlockDefaultEncodingContext(null,
809           HConstants.HFILEBLOCK_DUMMY_HEADER, fileContext);
810       dataBlockEncodingCtx = this.dataBlockEncoder
811           .newDataBlockEncodingContext(HConstants.HFILEBLOCK_DUMMY_HEADER, fileContext);
812 
813       if (fileContext.getBytesPerChecksum() < HConstants.HFILEBLOCK_HEADER_SIZE) {
814         throw new RuntimeException("Unsupported value of bytesPerChecksum. " +
815             " Minimum is " + HConstants.HFILEBLOCK_HEADER_SIZE + " but the configured value is " +
816             fileContext.getBytesPerChecksum());
817       }
818 
819       baosInMemory = new ByteArrayOutputStream();
820 
821       prevOffsetByType = new long[BlockType.values().length];
822       for (int i = 0; i < prevOffsetByType.length; ++i)
823         prevOffsetByType[i] = -1;
824 
825       this.fileContext = fileContext;
826     }
827 
828     /**
829      * Starts writing into the block. The previous block's data is discarded.
830      *
831      * @return the stream the user can write their data into
832      * @throws IOException
833      */
834     public DataOutputStream startWriting(BlockType newBlockType)
835         throws IOException {
836       if (state == State.BLOCK_READY && startOffset != -1) {
837         // We had a previous block that was written to a stream at a specific
838         // offset. Save that offset as the last offset of a block of that type.
839         prevOffsetByType[blockType.getId()] = startOffset;
840       }
841 
842       startOffset = -1;
843       blockType = newBlockType;
844 
845       baosInMemory.reset();
846       baosInMemory.write(HConstants.HFILEBLOCK_DUMMY_HEADER);
847 
848       state = State.WRITING;
849 
850       // We will compress it later in finishBlock()
851       userDataStream = new DataOutputStream(baosInMemory);
852       if (newBlockType == BlockType.DATA) {
853         this.dataBlockEncoder.startBlockEncoding(dataBlockEncodingCtx, userDataStream);
854       }
855       this.unencodedDataSizeWritten = 0;
856       return userDataStream;
857     }
858 
859     /**
860      * Writes the Cell to this block
861      * @param cell
862      * @throws IOException
863      */
864     public void write(Cell cell) throws IOException{
865       expectState(State.WRITING);
866       this.unencodedDataSizeWritten += this.dataBlockEncoder.encode(cell, dataBlockEncodingCtx,
867           this.userDataStream);
868     }
869 
870     /**
871      * Returns the stream for the user to write to. The block writer takes care
872      * of handling compression and buffering for caching on write. Can only be
873      * called in the "writing" state.
874      *
875      * @return the data output stream for the user to write to
876      */
877     DataOutputStream getUserDataStream() {
878       expectState(State.WRITING);
879       return userDataStream;
880     }
881 
882     /**
883      * Transitions the block writer from the "writing" state to the "block
884      * ready" state.  Does nothing if a block is already finished.
885      */
886     void ensureBlockReady() throws IOException {
887       Preconditions.checkState(state != State.INIT,
888           "Unexpected state: " + state);
889 
890       if (state == State.BLOCK_READY)
891         return;
892 
893       // This will set state to BLOCK_READY.
894       finishBlock();
895     }
896 
897     /**
898      * An internal method that flushes the compressing stream (if using
899      * compression), serializes the header, and takes care of the separate
900      * uncompressed stream for caching on write, if applicable. Sets block
901      * write state to "block ready".
902      */
903     private void finishBlock() throws IOException {
904       if (blockType == BlockType.DATA) {
905         BufferGrabbingByteArrayOutputStream baosInMemoryCopy =
906             new BufferGrabbingByteArrayOutputStream();
907         baosInMemory.writeTo(baosInMemoryCopy);
908         this.dataBlockEncoder.endBlockEncoding(dataBlockEncodingCtx, userDataStream,
909             baosInMemoryCopy.buf, blockType);
910         blockType = dataBlockEncodingCtx.getBlockType();
911       }
912       userDataStream.flush();
913       // This does an array copy, so it is safe to cache this byte array.
914       uncompressedBytesWithHeader = baosInMemory.toByteArray();
915       prevOffset = prevOffsetByType[blockType.getId()];
916 
917       // We need to set state before we can package the block up for
918       // cache-on-write. In a way, the block is ready, but not yet encoded or
919       // compressed.
920       state = State.BLOCK_READY;
921       if (blockType == BlockType.DATA || blockType == BlockType.ENCODED_DATA) {
922         onDiskBytesWithHeader = dataBlockEncodingCtx
923             .compressAndEncrypt(uncompressedBytesWithHeader);
924       } else {
925         onDiskBytesWithHeader = defaultBlockEncodingCtx
926             .compressAndEncrypt(uncompressedBytesWithHeader);
927       }
928       int numBytes = (int) ChecksumUtil.numBytes(
929           onDiskBytesWithHeader.length,
930           fileContext.getBytesPerChecksum());
931 
932       // put the header for on disk bytes
933       putHeader(onDiskBytesWithHeader, 0,
934           onDiskBytesWithHeader.length + numBytes,
935           uncompressedBytesWithHeader.length, onDiskBytesWithHeader.length);
936       // set the header for the uncompressed bytes (for cache-on-write)
937       putHeader(uncompressedBytesWithHeader, 0,
938           onDiskBytesWithHeader.length + numBytes,
939           uncompressedBytesWithHeader.length, onDiskBytesWithHeader.length);
940 
941       onDiskChecksum = new byte[numBytes];
942       ChecksumUtil.generateChecksums(
943           onDiskBytesWithHeader, 0, onDiskBytesWithHeader.length,
944           onDiskChecksum, 0, fileContext.getChecksumType(), fileContext.getBytesPerChecksum());
945     }
946 
947     public static class BufferGrabbingByteArrayOutputStream extends ByteArrayOutputStream {
948       private byte[] buf;
949 
950       @Override
951       public void write(byte[] b, int off, int len) {
952         this.buf = b;
953       }
954 
955       public byte[] getBuffer() {
956         return this.buf;
957       }
958     }
959 
960     /**
961      * Put the header into the given byte array at the given offset.
962      * @param onDiskSize size of the block on disk header + data + checksum
963      * @param uncompressedSize size of the block after decompression (but
964      *          before optional data block decoding) including header
965      * @param onDiskDataSize size of the block on disk with header
966      *        and data but not including the checksums
967      */
968     private void putHeader(byte[] dest, int offset, int onDiskSize,
969         int uncompressedSize, int onDiskDataSize) {
970       offset = blockType.put(dest, offset);
971       offset = Bytes.putInt(dest, offset, onDiskSize - HConstants.HFILEBLOCK_HEADER_SIZE);
972       offset = Bytes.putInt(dest, offset, uncompressedSize - HConstants.HFILEBLOCK_HEADER_SIZE);
973       offset = Bytes.putLong(dest, offset, prevOffset);
974       offset = Bytes.putByte(dest, offset, fileContext.getChecksumType().getCode());
975       offset = Bytes.putInt(dest, offset, fileContext.getBytesPerChecksum());
976       Bytes.putInt(dest, offset, onDiskDataSize);
977     }
978 
979     /**
980      * Similar to {@link #writeHeaderAndData(FSDataOutputStream)}, but records
981      * the offset of this block so that it can be referenced in the next block
982      * of the same type.
983      *
984      * @param out
985      * @throws IOException
986      */
987     public void writeHeaderAndData(FSDataOutputStream out) throws IOException {
988       long offset = out.getPos();
989       if (startOffset != -1 && offset != startOffset) {
990         throw new IOException("A " + blockType + " block written to a "
991             + "stream twice, first at offset " + startOffset + ", then at "
992             + offset);
993       }
994       startOffset = offset;
995 
996       finishBlockAndWriteHeaderAndData((DataOutputStream) out);
997     }
998 
999     /**
1000      * Writes the header and the compressed data of this block (or uncompressed
1001      * data when not using compression) into the given stream. Can be called in
1002      * the "writing" state or in the "block ready" state. If called in the
1003      * "writing" state, transitions the writer to the "block ready" state.
1004      *
1005      * @param out the output stream to write the
1006      * @throws IOException
1007      */
1008     protected void finishBlockAndWriteHeaderAndData(DataOutputStream out)
1009       throws IOException {
1010       ensureBlockReady();
1011       out.write(onDiskBytesWithHeader);
1012       out.write(onDiskChecksum);
1013     }
1014 
1015     /**
1016      * Returns the header or the compressed data (or uncompressed data when not
1017      * using compression) as a byte array. Can be called in the "writing" state
1018      * or in the "block ready" state. If called in the "writing" state,
1019      * transitions the writer to the "block ready" state. This returns
1020      * the header + data + checksums stored on disk.
1021      *
1022      * @return header and data as they would be stored on disk in a byte array
1023      * @throws IOException
1024      */
1025     byte[] getHeaderAndDataForTest() throws IOException {
1026       ensureBlockReady();
1027       // This is not very optimal, because we are doing an extra copy.
1028       // But this method is used only by unit tests.
1029       byte[] output =
1030           new byte[onDiskBytesWithHeader.length
1031               + onDiskChecksum.length];
1032       System.arraycopy(onDiskBytesWithHeader, 0, output, 0,
1033           onDiskBytesWithHeader.length);
1034       System.arraycopy(onDiskChecksum, 0, output,
1035           onDiskBytesWithHeader.length, onDiskChecksum.length);
1036       return output;
1037     }
1038 
1039     /**
1040      * Releases resources used by this writer.
1041      */
1042     public void release() {
1043       if (dataBlockEncodingCtx != null) {
1044         dataBlockEncodingCtx.close();
1045         dataBlockEncodingCtx = null;
1046       }
1047       if (defaultBlockEncodingCtx != null) {
1048         defaultBlockEncodingCtx.close();
1049         defaultBlockEncodingCtx = null;
1050       }
1051     }
1052 
1053     /**
1054      * Returns the on-disk size of the data portion of the block. This is the
1055      * compressed size if compression is enabled. Can only be called in the
1056      * "block ready" state. Header is not compressed, and its size is not
1057      * included in the return value.
1058      *
1059      * @return the on-disk size of the block, not including the header.
1060      */
1061     int getOnDiskSizeWithoutHeader() {
1062       expectState(State.BLOCK_READY);
1063       return onDiskBytesWithHeader.length + onDiskChecksum.length - HConstants.HFILEBLOCK_HEADER_SIZE;
1064     }
1065 
1066     /**
1067      * Returns the on-disk size of the block. Can only be called in the
1068      * "block ready" state.
1069      *
1070      * @return the on-disk size of the block ready to be written, including the
1071      *         header size, the data and the checksum data.
1072      */
1073     int getOnDiskSizeWithHeader() {
1074       expectState(State.BLOCK_READY);
1075       return onDiskBytesWithHeader.length + onDiskChecksum.length;
1076     }
1077 
1078     /**
1079      * The uncompressed size of the block data. Does not include header size.
1080      */
1081     int getUncompressedSizeWithoutHeader() {
1082       expectState(State.BLOCK_READY);
1083       return uncompressedBytesWithHeader.length - HConstants.HFILEBLOCK_HEADER_SIZE;
1084     }
1085 
1086     /**
1087      * The uncompressed size of the block data, including header size.
1088      */
1089     int getUncompressedSizeWithHeader() {
1090       expectState(State.BLOCK_READY);
1091       return uncompressedBytesWithHeader.length;
1092     }
1093 
1094     /** @return true if a block is being written  */
1095     public boolean isWriting() {
1096       return state == State.WRITING;
1097     }
1098 
1099     /**
1100      * Returns the number of bytes written into the current block so far, or
1101      * zero if not writing the block at the moment. Note that this will return
1102      * zero in the "block ready" state as well.
1103      *
1104      * @return the number of bytes written
1105      */
1106     public int blockSizeWritten() {
1107       if (state != State.WRITING) return 0;
1108       return this.unencodedDataSizeWritten;
1109     }
1110 
1111     /**
1112      * Returns the header followed by the uncompressed data, even if using
1113      * compression. This is needed for storing uncompressed blocks in the block
1114      * cache. Can be called in the "writing" state or the "block ready" state.
1115      * Returns only the header and data, does not include checksum data.
1116      *
1117      * @return uncompressed block bytes for caching on write
1118      */
1119     ByteBuffer getUncompressedBufferWithHeader() {
1120       expectState(State.BLOCK_READY);
1121       return ByteBuffer.wrap(uncompressedBytesWithHeader);
1122     }
1123 
1124     /**
1125      * Returns the header followed by the on-disk (compressed/encoded/encrypted) data. This is
1126      * needed for storing packed blocks in the block cache. Expects calling semantics identical to
1127      * {@link #getUncompressedBufferWithHeader()}. Returns only the header and data,
1128      * Does not include checksum data.
1129      *
1130      * @return packed block bytes for caching on write
1131      */
1132     ByteBuffer getOnDiskBufferWithHeader() {
1133       expectState(State.BLOCK_READY);
1134       return ByteBuffer.wrap(onDiskBytesWithHeader);
1135     }
1136 
1137     private void expectState(State expectedState) {
1138       if (state != expectedState) {
1139         throw new IllegalStateException("Expected state: " + expectedState +
1140             ", actual state: " + state);
1141       }
1142     }
1143 
1144     /**
1145      * Takes the given {@link BlockWritable} instance, creates a new block of
1146      * its appropriate type, writes the writable into this block, and flushes
1147      * the block into the output stream. The writer is instructed not to buffer
1148      * uncompressed bytes for cache-on-write.
1149      *
1150      * @param bw the block-writable object to write as a block
1151      * @param out the file system output stream
1152      * @throws IOException
1153      */
1154     public void writeBlock(BlockWritable bw, FSDataOutputStream out)
1155         throws IOException {
1156       bw.writeToBlock(startWriting(bw.getBlockType()));
1157       writeHeaderAndData(out);
1158     }
1159 
1160     /**
1161      * Creates a new HFileBlock. Checksums have already been validated, so
1162      * the byte buffer passed into the constructor of this newly created
1163      * block does not have checksum data even though the header minor
1164      * version is MINOR_VERSION_WITH_CHECKSUM. This is indicated by setting a
1165      * 0 value in bytesPerChecksum.
1166      */
1167     public HFileBlock getBlockForCaching(CacheConfig cacheConf) {
1168       HFileContext newContext = new HFileContextBuilder()
1169                                 .withBlockSize(fileContext.getBlocksize())
1170                                 .withBytesPerCheckSum(0)
1171                                 .withChecksumType(ChecksumType.NULL) // no checksums in cached data
1172                                 .withCompression(fileContext.getCompression())
1173                                 .withDataBlockEncoding(fileContext.getDataBlockEncoding())
1174                                 .withHBaseCheckSum(fileContext.isUseHBaseChecksum())
1175                                 .withCompressTags(fileContext.isCompressTags())
1176                                 .withIncludesMvcc(fileContext.isIncludesMvcc())
1177                                 .withIncludesTags(fileContext.isIncludesTags())
1178                                 .build();
1179       return new HFileBlock(blockType, getOnDiskSizeWithoutHeader(),
1180           getUncompressedSizeWithoutHeader(), prevOffset,
1181           cacheConf.shouldCacheCompressed(blockType.getCategory()) ?
1182             getOnDiskBufferWithHeader() :
1183             getUncompressedBufferWithHeader(),
1184           FILL_HEADER, startOffset,
1185           onDiskBytesWithHeader.length + onDiskChecksum.length, newContext);
1186     }
1187   }
1188 
1189   /** Something that can be written into a block. */
1190   public interface BlockWritable {
1191 
1192     /** The type of block this data should use. */
1193     BlockType getBlockType();
1194 
1195     /**
1196      * Writes the block to the provided stream. Must not write any magic
1197      * records.
1198      *
1199      * @param out a stream to write uncompressed data into
1200      */
1201     void writeToBlock(DataOutput out) throws IOException;
1202   }
1203 
1204   // Block readers and writers
1205 
1206   /** An interface allowing to iterate {@link HFileBlock}s. */
1207   public interface BlockIterator {
1208 
1209     /**
1210      * Get the next block, or null if there are no more blocks to iterate.
1211      */
1212     HFileBlock nextBlock() throws IOException;
1213 
1214     /**
1215      * Similar to {@link #nextBlock()} but checks block type, throws an
1216      * exception if incorrect, and returns the HFile block
1217      */
1218     HFileBlock nextBlockWithBlockType(BlockType blockType) throws IOException;
1219   }
1220 
1221   /** A full-fledged reader with iteration ability. */
1222   public interface FSReader {
1223 
1224     /**
1225      * Reads the block at the given offset in the file with the given on-disk
1226      * size and uncompressed size.
1227      *
1228      * @param offset
1229      * @param onDiskSize the on-disk size of the entire block, including all
1230      *          applicable headers, or -1 if unknown
1231      * @param uncompressedSize the uncompressed size of the compressed part of
1232      *          the block, or -1 if unknown
1233      * @return the newly read block
1234      */
1235     HFileBlock readBlockData(long offset, long onDiskSize,
1236         int uncompressedSize, boolean pread) throws IOException;
1237 
1238     /**
1239      * Creates a block iterator over the given portion of the {@link HFile}.
1240      * The iterator returns blocks starting with offset such that offset <=
1241      * startOffset < endOffset. Returned blocks are always unpacked.
1242      *
1243      * @param startOffset the offset of the block to start iteration with
1244      * @param endOffset the offset to end iteration at (exclusive)
1245      * @return an iterator of blocks between the two given offsets
1246      */
1247     BlockIterator blockRange(long startOffset, long endOffset);
1248 
1249     /** Closes the backing streams */
1250     void closeStreams() throws IOException;
1251 
1252     /** Get a decoder for {@link BlockType#ENCODED_DATA} blocks from this file. */
1253     HFileBlockDecodingContext getBlockDecodingContext();
1254 
1255     /** Get the default decoder for blocks from this file. */
1256     HFileBlockDecodingContext getDefaultBlockDecodingContext();
1257   }
1258 
1259   /**
1260    * A common implementation of some methods of {@link FSReader} and some
1261    * tools for implementing HFile format version-specific block readers.
1262    */
1263   private abstract static class AbstractFSReader implements FSReader {
1264     /** Compression algorithm used by the {@link HFile} */
1265 
1266     /** The size of the file we are reading from, or -1 if unknown. */
1267     protected long fileSize;
1268 
1269     /** The size of the header */
1270     protected final int hdrSize;
1271 
1272     /** The filesystem used to access data */
1273     protected HFileSystem hfs;
1274 
1275     /** The path (if any) where this data is coming from */
1276     protected Path path;
1277 
1278     private final Lock streamLock = new ReentrantLock();
1279 
1280     /** The default buffer size for our buffered streams */
1281     public static final int DEFAULT_BUFFER_SIZE = 1 << 20;
1282 
1283     protected HFileContext fileContext;
1284 
1285     public AbstractFSReader(long fileSize, HFileSystem hfs, Path path, HFileContext fileContext)
1286         throws IOException {
1287       this.fileSize = fileSize;
1288       this.hfs = hfs;
1289       this.path = path;
1290       this.fileContext = fileContext;
1291       this.hdrSize = headerSize(fileContext.isUseHBaseChecksum());
1292     }
1293 
1294     @Override
1295     public BlockIterator blockRange(final long startOffset,
1296         final long endOffset) {
1297       final FSReader owner = this; // handle for inner class
1298       return new BlockIterator() {
1299         private long offset = startOffset;
1300 
1301         @Override
1302         public HFileBlock nextBlock() throws IOException {
1303           if (offset >= endOffset)
1304             return null;
1305           HFileBlock b = readBlockData(offset, -1, -1, false);
1306           offset += b.getOnDiskSizeWithHeader();
1307           return b.unpack(fileContext, owner);
1308         }
1309 
1310         @Override
1311         public HFileBlock nextBlockWithBlockType(BlockType blockType)
1312             throws IOException {
1313           HFileBlock blk = nextBlock();
1314           if (blk.getBlockType() != blockType) {
1315             throw new IOException("Expected block of type " + blockType
1316                 + " but found " + blk.getBlockType());
1317           }
1318           return blk;
1319         }
1320       };
1321     }
1322 
1323     /**
1324      * Does a positional read or a seek and read into the given buffer. Returns
1325      * the on-disk size of the next block, or -1 if it could not be determined.
1326      *
1327      * @param dest destination buffer
1328      * @param destOffset offset in the destination buffer
1329      * @param size size of the block to be read
1330      * @param peekIntoNextBlock whether to read the next block's on-disk size
1331      * @param fileOffset position in the stream to read at
1332      * @param pread whether we should do a positional read
1333      * @param istream The input source of data
1334      * @return the on-disk size of the next block with header size included, or
1335      *         -1 if it could not be determined
1336      * @throws IOException
1337      */
1338     protected int readAtOffset(FSDataInputStream istream,
1339         byte[] dest, int destOffset, int size,
1340         boolean peekIntoNextBlock, long fileOffset, boolean pread)
1341         throws IOException {
1342       if (peekIntoNextBlock &&
1343           destOffset + size + hdrSize > dest.length) {
1344         // We are asked to read the next block's header as well, but there is
1345         // not enough room in the array.
1346         throw new IOException("Attempted to read " + size + " bytes and " +
1347             hdrSize + " bytes of next header into a " + dest.length +
1348             "-byte array at offset " + destOffset);
1349       }
1350 
1351       if (!pread && streamLock.tryLock()) {
1352         // Seek + read. Better for scanning.
1353         try {
1354           istream.seek(fileOffset);
1355 
1356           long realOffset = istream.getPos();
1357           if (realOffset != fileOffset) {
1358             throw new IOException("Tried to seek to " + fileOffset + " to "
1359                 + "read " + size + " bytes, but pos=" + realOffset
1360                 + " after seek");
1361           }
1362 
1363           if (!peekIntoNextBlock) {
1364             IOUtils.readFully(istream, dest, destOffset, size);
1365             return -1;
1366           }
1367 
1368           // Try to read the next block header.
1369           if (!readWithExtra(istream, dest, destOffset, size, hdrSize))
1370             return -1;
1371         } finally {
1372           streamLock.unlock();
1373         }
1374       } else {
1375         // Positional read. Better for random reads; or when the streamLock is already locked.
1376         int extraSize = peekIntoNextBlock ? hdrSize : 0;
1377         int ret = istream.read(fileOffset, dest, destOffset, size + extraSize);
1378         if (ret < size) {
1379           throw new IOException("Positional read of " + size + " bytes " +
1380               "failed at offset " + fileOffset + " (returned " + ret + ")");
1381         }
1382 
1383         if (ret == size || ret < size + extraSize) {
1384           // Could not read the next block's header, or did not try.
1385           return -1;
1386         }
1387       }
1388 
1389       assert peekIntoNextBlock;
1390       return Bytes.toInt(dest, destOffset + size + BlockType.MAGIC_LENGTH) + hdrSize;
1391     }
1392 
1393   }
1394 
1395   /**
1396    * We always prefetch the header of the next block, so that we know its
1397    * on-disk size in advance and can read it in one operation.
1398    */
1399   private static class PrefetchedHeader {
1400     long offset = -1;
1401     byte[] header = new byte[HConstants.HFILEBLOCK_HEADER_SIZE];
1402     final ByteBuffer buf = ByteBuffer.wrap(header, 0, HConstants.HFILEBLOCK_HEADER_SIZE);
1403   }
1404 
1405   /** Reads version 2 blocks from the filesystem. */
1406   static class FSReaderImpl extends AbstractFSReader {
1407     /** The file system stream of the underlying {@link HFile} that
1408      * does or doesn't do checksum validations in the filesystem */
1409     protected FSDataInputStreamWrapper streamWrapper;
1410 
1411     private HFileBlockDecodingContext encodedBlockDecodingCtx;
1412 
1413     /** Default context used when BlockType != {@link BlockType#ENCODED_DATA}. */
1414     private final HFileBlockDefaultDecodingContext defaultDecodingCtx;
1415 
1416     private ThreadLocal<PrefetchedHeader> prefetchedHeaderForThread =
1417         new ThreadLocal<PrefetchedHeader>() {
1418           @Override
1419           public PrefetchedHeader initialValue() {
1420             return new PrefetchedHeader();
1421           }
1422         };
1423 
1424     public FSReaderImpl(FSDataInputStreamWrapper stream, long fileSize, HFileSystem hfs, Path path,
1425         HFileContext fileContext) throws IOException {
1426       super(fileSize, hfs, path, fileContext);
1427       this.streamWrapper = stream;
1428       // Older versions of HBase didn't support checksum.
1429       this.streamWrapper.prepareForBlockReader(!fileContext.isUseHBaseChecksum());
1430       defaultDecodingCtx = new HFileBlockDefaultDecodingContext(fileContext);
1431       encodedBlockDecodingCtx = defaultDecodingCtx;
1432     }
1433 
1434     /**
1435      * A constructor that reads files with the latest minor version.
1436      * This is used by unit tests only.
1437      */
1438     FSReaderImpl(FSDataInputStream istream, long fileSize, HFileContext fileContext)
1439     throws IOException {
1440       this(new FSDataInputStreamWrapper(istream), fileSize, null, null, fileContext);
1441     }
1442 
1443     /**
1444      * Reads a version 2 block (version 1 blocks not supported and not expected). Tries to do as
1445      * little memory allocation as possible, using the provided on-disk size.
1446      *
1447      * @param offset the offset in the stream to read at
1448      * @param onDiskSizeWithHeaderL the on-disk size of the block, including
1449      *          the header, or -1 if unknown
1450      * @param uncompressedSize the uncompressed size of the the block. Always
1451      *          expected to be -1. This parameter is only used in version 1.
1452      * @param pread whether to use a positional read
1453      */
1454     @Override
1455     public HFileBlock readBlockData(long offset, long onDiskSizeWithHeaderL,
1456         int uncompressedSize, boolean pread)
1457     throws IOException {
1458 
1459       // get a copy of the current state of whether to validate
1460       // hbase checksums or not for this read call. This is not
1461       // thread-safe but the one constaint is that if we decide
1462       // to skip hbase checksum verification then we are
1463       // guaranteed to use hdfs checksum verification.
1464       boolean doVerificationThruHBaseChecksum = streamWrapper.shouldUseHBaseChecksum();
1465       FSDataInputStream is = streamWrapper.getStream(doVerificationThruHBaseChecksum);
1466 
1467       HFileBlock blk = readBlockDataInternal(is, offset,
1468                          onDiskSizeWithHeaderL,
1469                          uncompressedSize, pread,
1470                          doVerificationThruHBaseChecksum);
1471       if (blk == null) {
1472         HFile.LOG.warn("HBase checksum verification failed for file " +
1473                        path + " at offset " +
1474                        offset + " filesize " + fileSize +
1475                        ". Retrying read with HDFS checksums turned on...");
1476 
1477         if (!doVerificationThruHBaseChecksum) {
1478           String msg = "HBase checksum verification failed for file " +
1479                        path + " at offset " +
1480                        offset + " filesize " + fileSize +
1481                        " but this cannot happen because doVerify is " +
1482                        doVerificationThruHBaseChecksum;
1483           HFile.LOG.warn(msg);
1484           throw new IOException(msg); // cannot happen case here
1485         }
1486         HFile.checksumFailures.incrementAndGet(); // update metrics
1487 
1488         // If we have a checksum failure, we fall back into a mode where
1489         // the next few reads use HDFS level checksums. We aim to make the
1490         // next CHECKSUM_VERIFICATION_NUM_IO_THRESHOLD reads avoid
1491         // hbase checksum verification, but since this value is set without
1492         // holding any locks, it can so happen that we might actually do
1493         // a few more than precisely this number.
1494         is = this.streamWrapper.fallbackToFsChecksum(CHECKSUM_VERIFICATION_NUM_IO_THRESHOLD);
1495         doVerificationThruHBaseChecksum = false;
1496         blk = readBlockDataInternal(is, offset, onDiskSizeWithHeaderL,
1497                                     uncompressedSize, pread,
1498                                     doVerificationThruHBaseChecksum);
1499         if (blk != null) {
1500           HFile.LOG.warn("HDFS checksum verification suceeded for file " +
1501                          path + " at offset " +
1502                          offset + " filesize " + fileSize);
1503         }
1504       }
1505       if (blk == null && !doVerificationThruHBaseChecksum) {
1506         String msg = "readBlockData failed, possibly due to " +
1507                      "checksum verification failed for file " + path +
1508                      " at offset " + offset + " filesize " + fileSize;
1509         HFile.LOG.warn(msg);
1510         throw new IOException(msg);
1511       }
1512 
1513       // If there is a checksum mismatch earlier, then retry with
1514       // HBase checksums switched off and use HDFS checksum verification.
1515       // This triggers HDFS to detect and fix corrupt replicas. The
1516       // next checksumOffCount read requests will use HDFS checksums.
1517       // The decrementing of this.checksumOffCount is not thread-safe,
1518       // but it is harmless because eventually checksumOffCount will be
1519       // a negative number.
1520       streamWrapper.checksumOk();
1521       return blk;
1522     }
1523 
1524     /**
1525      * Reads a version 2 block.
1526      *
1527      * @param offset the offset in the stream to read at
1528      * @param onDiskSizeWithHeaderL the on-disk size of the block, including
1529      *          the header, or -1 if unknown
1530      * @param uncompressedSize the uncompressed size of the the block. Always
1531      *          expected to be -1. This parameter is only used in version 1.
1532      * @param pread whether to use a positional read
1533      * @param verifyChecksum Whether to use HBase checksums.
1534      *        If HBase checksum is switched off, then use HDFS checksum.
1535      * @return the HFileBlock or null if there is a HBase checksum mismatch
1536      */
1537     private HFileBlock readBlockDataInternal(FSDataInputStream is, long offset,
1538         long onDiskSizeWithHeaderL, int uncompressedSize, boolean pread,
1539         boolean verifyChecksum)
1540     throws IOException {
1541       if (offset < 0) {
1542         throw new IOException("Invalid offset=" + offset + " trying to read "
1543             + "block (onDiskSize=" + onDiskSizeWithHeaderL
1544             + ", uncompressedSize=" + uncompressedSize + ")");
1545       }
1546 
1547       if (uncompressedSize != -1) {
1548         throw new IOException("Version 2 block reader API does not need " +
1549             "the uncompressed size parameter");
1550       }
1551 
1552       if ((onDiskSizeWithHeaderL < hdrSize && onDiskSizeWithHeaderL != -1)
1553           || onDiskSizeWithHeaderL >= Integer.MAX_VALUE) {
1554         throw new IOException("Invalid onDisksize=" + onDiskSizeWithHeaderL
1555             + ": expected to be at least " + hdrSize
1556             + " and at most " + Integer.MAX_VALUE + ", or -1 (offset="
1557             + offset + ", uncompressedSize=" + uncompressedSize + ")");
1558       }
1559 
1560       int onDiskSizeWithHeader = (int) onDiskSizeWithHeaderL;
1561       // See if we can avoid reading the header. This is desirable, because
1562       // we will not incur a backward seek operation if we have already
1563       // read this block's header as part of the previous read's look-ahead.
1564       // And we also want to skip reading the header again if it has already
1565       // been read.
1566       // TODO: How often does this optimization fire? Has to be same thread so the thread local
1567       // is pertinent and we have to be reading next block as in a big scan.
1568       PrefetchedHeader prefetchedHeader = prefetchedHeaderForThread.get();
1569       ByteBuffer headerBuf = prefetchedHeader.offset == offset? prefetchedHeader.buf: null;
1570 
1571       // Allocate enough space to fit the next block's header too.
1572       int nextBlockOnDiskSize = 0;
1573       byte[] onDiskBlock = null;
1574 
1575       HFileBlock b = null;
1576       if (onDiskSizeWithHeader > 0) {
1577         // We know the total on-disk size. Read the entire block into memory,
1578         // then parse the header. This code path is used when
1579         // doing a random read operation relying on the block index, as well as
1580         // when the client knows the on-disk size from peeking into the next
1581         // block's header (e.g. this block's header) when reading the previous
1582         // block. This is the faster and more preferable case.
1583 
1584         // Size that we have to skip in case we have already read the header.
1585         int preReadHeaderSize = headerBuf == null ? 0 : hdrSize;
1586         onDiskBlock = new byte[onDiskSizeWithHeader + hdrSize]; // room for this block plus the
1587                                                                 // next block's header
1588         nextBlockOnDiskSize = readAtOffset(is, onDiskBlock,
1589             preReadHeaderSize, onDiskSizeWithHeader - preReadHeaderSize,
1590             true, offset + preReadHeaderSize, pread);
1591         if (headerBuf != null) {
1592           // the header has been read when reading the previous block, copy
1593           // to this block's header
1594           // headerBuf is HBB
1595           assert headerBuf.hasArray();
1596           System.arraycopy(headerBuf.array(),
1597               headerBuf.arrayOffset(), onDiskBlock, 0, hdrSize);
1598         } else {
1599           headerBuf = ByteBuffer.wrap(onDiskBlock, 0, hdrSize);
1600         }
1601         // We know the total on-disk size but not the uncompressed size. Parse the header.
1602         try {
1603           // TODO: FIX!!! Expensive parse just to get a length
1604           b = new HFileBlock(headerBuf, fileContext.isUseHBaseChecksum());
1605         } catch (IOException ex) {
1606           // Seen in load testing. Provide comprehensive debug info.
1607           throw new IOException("Failed to read compressed block at "
1608               + offset
1609               + ", onDiskSizeWithoutHeader="
1610               + onDiskSizeWithHeader
1611               + ", preReadHeaderSize="
1612               + hdrSize
1613               + ", header.length="
1614               + prefetchedHeader.header.length
1615               + ", header bytes: "
1616               + Bytes.toStringBinary(prefetchedHeader.header, 0,
1617                   hdrSize), ex);
1618         }
1619         // if the caller specifies a onDiskSizeWithHeader, validate it.
1620         int onDiskSizeWithoutHeader = onDiskSizeWithHeader - hdrSize;
1621         assert onDiskSizeWithoutHeader >= 0;
1622         b.validateOnDiskSizeWithoutHeader(onDiskSizeWithoutHeader);
1623       } else {
1624         // Check headerBuf to see if we have read this block's header as part of
1625         // reading the previous block. This is an optimization of peeking into
1626         // the next block's header (e.g.this block's header) when reading the
1627         // previous block. This is the faster and more preferable case. If the
1628         // header is already there, don't read the header again.
1629 
1630         // Unfortunately, we still have to do a separate read operation to
1631         // read the header.
1632         if (headerBuf == null) {
1633           // From the header, determine the on-disk size of the given hfile
1634           // block, and read the remaining data, thereby incurring two read
1635           // operations. This might happen when we are doing the first read
1636           // in a series of reads or a random read, and we don't have access
1637           // to the block index. This is costly and should happen very rarely.
1638           headerBuf = ByteBuffer.allocate(hdrSize);
1639           // headerBuf is HBB
1640           readAtOffset(is, headerBuf.array(), headerBuf.arrayOffset(),
1641               hdrSize, false, offset, pread);
1642         }
1643         // TODO: FIX!!! Expensive parse just to get a length
1644         b = new HFileBlock(headerBuf, fileContext.isUseHBaseChecksum());
1645         onDiskBlock = new byte[b.getOnDiskSizeWithHeader() + hdrSize];
1646         // headerBuf is HBB
1647         System.arraycopy(headerBuf.array(), headerBuf.arrayOffset(), onDiskBlock, 0, hdrSize);
1648         nextBlockOnDiskSize =
1649           readAtOffset(is, onDiskBlock, hdrSize, b.getOnDiskSizeWithHeader()
1650               - hdrSize, true, offset + hdrSize, pread);
1651         onDiskSizeWithHeader = b.onDiskSizeWithoutHeader + hdrSize;
1652       }
1653 
1654       if (!fileContext.isCompressedOrEncrypted()) {
1655         b.assumeUncompressed();
1656       }
1657 
1658       if (verifyChecksum && !validateBlockChecksum(b, onDiskBlock, hdrSize)) {
1659         return null;             // checksum mismatch
1660       }
1661 
1662       // The onDiskBlock will become the headerAndDataBuffer for this block.
1663       // If nextBlockOnDiskSizeWithHeader is not zero, the onDiskBlock already
1664       // contains the header of next block, so no need to set next
1665       // block's header in it.
1666       b = new HFileBlock(ByteBuffer.wrap(onDiskBlock, 0, onDiskSizeWithHeader),
1667         this.fileContext.isUseHBaseChecksum());
1668 
1669       b.nextBlockOnDiskSizeWithHeader = nextBlockOnDiskSize;
1670 
1671       // Set prefetched header
1672       if (b.hasNextBlockHeader()) {
1673         prefetchedHeader.offset = offset + b.getOnDiskSizeWithHeader();
1674         System.arraycopy(onDiskBlock, onDiskSizeWithHeader, prefetchedHeader.header, 0, hdrSize);
1675       }
1676 
1677       b.offset = offset;
1678       b.fileContext.setIncludesTags(this.fileContext.isIncludesTags());
1679       b.fileContext.setIncludesMvcc(this.fileContext.isIncludesMvcc());
1680       return b;
1681     }
1682 
1683     void setIncludesMemstoreTS(boolean includesMemstoreTS) {
1684       this.fileContext.setIncludesMvcc(includesMemstoreTS);
1685     }
1686 
1687     void setDataBlockEncoder(HFileDataBlockEncoder encoder) {
1688       encodedBlockDecodingCtx = encoder.newDataBlockDecodingContext(this.fileContext);
1689     }
1690 
1691     @Override
1692     public HFileBlockDecodingContext getBlockDecodingContext() {
1693       return this.encodedBlockDecodingCtx;
1694     }
1695 
1696     @Override
1697     public HFileBlockDecodingContext getDefaultBlockDecodingContext() {
1698       return this.defaultDecodingCtx;
1699     }
1700 
1701     /**
1702      * Generates the checksum for the header as well as the data and
1703      * then validates that it matches the value stored in the header.
1704      * If there is a checksum mismatch, then return false. Otherwise
1705      * return true.
1706      */
1707     protected boolean validateBlockChecksum(HFileBlock block,  byte[] data, int hdrSize)
1708         throws IOException {
1709       return ChecksumUtil.validateBlockChecksum(path, block, data, hdrSize);
1710     }
1711 
1712     @Override
1713     public void closeStreams() throws IOException {
1714       streamWrapper.close();
1715     }
1716 
1717     @Override
1718     public String toString() {
1719       return "hfs=" + hfs + ", path=" + path + ", fileContext=" + fileContext;
1720     }
1721   }
1722 
1723   @Override
1724   public int getSerializedLength() {
1725     if (buf != null) {
1726       // include extra bytes for the next header when it's available.
1727       int extraSpace = hasNextBlockHeader() ? headerSize() : 0;
1728       return this.buf.limit() + extraSpace + HFileBlock.EXTRA_SERIALIZATION_SPACE;
1729     }
1730     return 0;
1731   }
1732 
1733   @Override
1734   public void serialize(ByteBuffer destination) {
1735     ByteBufferUtils.copyFromBufferToBuffer(destination, this.buf, 0, getSerializedLength()
1736         - EXTRA_SERIALIZATION_SPACE);
1737     serializeExtraInfo(destination);
1738   }
1739 
1740   public void serializeExtraInfo(ByteBuffer destination) {
1741     destination.put(this.fileContext.isUseHBaseChecksum() ? (byte) 1 : (byte) 0);
1742     destination.putLong(this.offset);
1743     destination.putInt(this.nextBlockOnDiskSizeWithHeader);
1744     destination.rewind();
1745   }
1746 
1747   @Override
1748   public CacheableDeserializer<Cacheable> getDeserializer() {
1749     return HFileBlock.blockDeserializer;
1750   }
1751 
1752   @Override
1753   public boolean equals(Object comparison) {
1754     if (this == comparison) {
1755       return true;
1756     }
1757     if (comparison == null) {
1758       return false;
1759     }
1760     if (comparison.getClass() != this.getClass()) {
1761       return false;
1762     }
1763 
1764     HFileBlock castedComparison = (HFileBlock) comparison;
1765 
1766     if (castedComparison.blockType != this.blockType) {
1767       return false;
1768     }
1769     if (castedComparison.nextBlockOnDiskSizeWithHeader != this.nextBlockOnDiskSizeWithHeader) {
1770       return false;
1771     }
1772     if (castedComparison.offset != this.offset) {
1773       return false;
1774     }
1775     if (castedComparison.onDiskSizeWithoutHeader != this.onDiskSizeWithoutHeader) {
1776       return false;
1777     }
1778     if (castedComparison.prevBlockOffset != this.prevBlockOffset) {
1779       return false;
1780     }
1781     if (castedComparison.uncompressedSizeWithoutHeader != this.uncompressedSizeWithoutHeader) {
1782       return false;
1783     }
1784     if (ByteBufferUtils.compareTo(this.buf, 0, this.buf.limit(), castedComparison.buf, 0,
1785         castedComparison.buf.limit()) != 0) {
1786       return false;
1787     }
1788     return true;
1789   }
1790 
1791   public DataBlockEncoding getDataBlockEncoding() {
1792     if (blockType == BlockType.ENCODED_DATA) {
1793       return DataBlockEncoding.getEncodingById(getDataBlockEncodingId());
1794     }
1795     return DataBlockEncoding.NONE;
1796   }
1797 
1798   byte getChecksumType() {
1799     return this.fileContext.getChecksumType().getCode();
1800   }
1801 
1802   int getBytesPerChecksum() {
1803     return this.fileContext.getBytesPerChecksum();
1804   }
1805 
1806   /** @return the size of data on disk + header. Excludes checksum. */
1807   int getOnDiskDataSizeWithHeader() {
1808     return this.onDiskDataSizeWithHeader;
1809   }
1810 
1811   /**
1812    * Calcuate the number of bytes required to store all the checksums
1813    * for this block. Each checksum value is a 4 byte integer.
1814    */
1815   int totalChecksumBytes() {
1816     // If the hfile block has minorVersion 0, then there are no checksum
1817     // data to validate. Similarly, a zero value in this.bytesPerChecksum
1818     // indicates that cached blocks do not have checksum data because
1819     // checksums were already validated when the block was read from disk.
1820     if (!fileContext.isUseHBaseChecksum() || this.fileContext.getBytesPerChecksum() == 0) {
1821       return 0;
1822     }
1823     return (int)ChecksumUtil.numBytes(onDiskDataSizeWithHeader, this.fileContext.getBytesPerChecksum());
1824   }
1825 
1826   /**
1827    * Returns the size of this block header.
1828    */
1829   public int headerSize() {
1830     return headerSize(this.fileContext.isUseHBaseChecksum());
1831   }
1832 
1833   /**
1834    * Maps a minor version to the size of the header.
1835    */
1836   public static int headerSize(boolean usesHBaseChecksum) {
1837     if (usesHBaseChecksum) {
1838       return HConstants.HFILEBLOCK_HEADER_SIZE;
1839     }
1840     return HConstants.HFILEBLOCK_HEADER_SIZE_NO_CHECKSUM;
1841   }
1842 
1843   /**
1844    * Return the appropriate DUMMY_HEADER for the minor version
1845    */
1846   public byte[] getDummyHeaderForVersion() {
1847     return getDummyHeaderForVersion(this.fileContext.isUseHBaseChecksum());
1848   }
1849 
1850   /**
1851    * Return the appropriate DUMMY_HEADER for the minor version
1852    */
1853   static private byte[] getDummyHeaderForVersion(boolean usesHBaseChecksum) {
1854     if (usesHBaseChecksum) {
1855       return HConstants.HFILEBLOCK_DUMMY_HEADER;
1856     }
1857     return DUMMY_HEADER_NO_CHECKSUM;
1858   }
1859 
1860   /**
1861    * @return the HFileContext used to create this HFileBlock. Not necessary the
1862    * fileContext for the file from which this block's data was originally read.
1863    */
1864   public HFileContext getHFileContext() {
1865     return this.fileContext;
1866   }
1867 
1868   /**
1869    * Convert the contents of the block header into a human readable string.
1870    * This is mostly helpful for debugging. This assumes that the block
1871    * has minor version > 0.
1872    */
1873   static String toStringHeader(ByteBuffer buf) throws IOException {
1874     byte[] magicBuf = new byte[Math.min(buf.limit() - buf.position(), BlockType.MAGIC_LENGTH)];
1875     buf.get(magicBuf);
1876     BlockType bt = BlockType.parse(magicBuf, 0, BlockType.MAGIC_LENGTH);
1877     int compressedBlockSizeNoHeader = buf.getInt();;
1878     int uncompressedBlockSizeNoHeader = buf.getInt();;
1879     long prevBlockOffset = buf.getLong();
1880     byte cksumtype = buf.get();
1881     long bytesPerChecksum = buf.getInt();
1882     long onDiskDataSizeWithHeader = buf.getInt();
1883     return " Header dump: magic: " + Bytes.toString(magicBuf) +
1884                    " blockType " + bt +
1885                    " compressedBlockSizeNoHeader " +
1886                    compressedBlockSizeNoHeader +
1887                    " uncompressedBlockSizeNoHeader " +
1888                    uncompressedBlockSizeNoHeader +
1889                    " prevBlockOffset " + prevBlockOffset +
1890                    " checksumType " + ChecksumType.codeToType(cksumtype) +
1891                    " bytesPerChecksum " + bytesPerChecksum +
1892                    " onDiskDataSizeWithHeader " + onDiskDataSizeWithHeader;
1893   }
1894 }