org.glassfish.jersey.client.ChunkedInput Maven / Gradle / Ivy
Show all versions of jaxrs-ri Show documentation
/*
* Copyright (c) 2012, 2020 Oracle and/or its affiliates. All rights reserved.
*
* This program and the accompanying materials are made available under the
* terms of the Eclipse Public License v. 2.0, which is available at
* http://www.eclipse.org/legal/epl-2.0.
*
* This Source Code may also be made available under the following Secondary
* Licenses when the conditions for such availability set forth in the
* Eclipse Public License v. 2.0 are satisfied: GNU General Public License,
* version 2 with the GNU Classpath Exception, which is available at
* https://www.gnu.org/software/classpath/license.html.
*
* SPDX-License-Identifier: EPL-2.0 OR GPL-2.0 WITH Classpath-exception-2.0
*/
package org.glassfish.jersey.client;
import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.Closeable;
import java.io.IOException;
import java.io.InputStream;
import java.lang.annotation.Annotation;
import java.lang.reflect.Type;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.List;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.logging.Level;
import java.util.logging.Logger;
import jakarta.ws.rs.core.GenericType;
import jakarta.ws.rs.core.MediaType;
import jakarta.ws.rs.core.MultivaluedMap;
import jakarta.ws.rs.ext.ReaderInterceptor;
import org.glassfish.jersey.client.internal.LocalizationMessages;
import org.glassfish.jersey.internal.PropertiesDelegate;
import org.glassfish.jersey.message.MessageBodyWorkers;
/**
* Response entity type used for receiving messages in "typed" chunks.
*
* This data type is useful for consuming partial responses from large or continuous data
* input streams.
*
* @param chunk type.
* @author Marek Potociar
*/
@SuppressWarnings("UnusedDeclaration")
public class ChunkedInput extends GenericType implements Closeable {
private static final Logger LOGGER = Logger.getLogger(ChunkedInput.class.getName());
private final AtomicBoolean closed = new AtomicBoolean(false);
private ChunkParser parser = createParser("\r\n");
private MediaType mediaType;
private final InputStream inputStream;
private final Annotation[] annotations;
private final MultivaluedMap headers;
private final MessageBodyWorkers messageBodyWorkers;
private final PropertiesDelegate propertiesDelegate;
/**
* Create new chunk parser that will split the response entity input stream
* based on a fixed boundary string.
*
* @param boundary chunk boundary.
* @return new fixed boundary string-based chunk parser.
*/
public static ChunkParser createParser(final String boundary) {
return new FixedBoundaryParser(boundary.getBytes());
}
/**
* Create new chunk parser that will split the response entity input stream
* based on a fixed boundary sequence of bytes.
*
* @param boundary chunk boundary.
* @return new fixed boundary sequence-based chunk parser.
*/
public static ChunkParser createParser(final byte[] boundary) {
return new FixedBoundaryParser(boundary);
}
/**
* Create a new chunk multi-parser that will split the response entity input stream
* based on multiple fixed boundary strings.
*
* @param boundaries chunk boundaries.
* @return new fixed boundary string-based chunk parser.
*/
public static ChunkParser createMultiParser(final String... boundaries) {
return new FixedMultiBoundaryParser(boundaries);
}
private abstract static class AbstractBoundaryParser implements ChunkParser {
@Override
public byte[] readChunk(final InputStream in) throws IOException {
final ByteArrayOutputStream buffer = new ByteArrayOutputStream();
byte[] delimiterBuffer = new byte[getDelimiterBufferSize()];
int data;
int dPos;
do {
dPos = 0;
while ((data = in.read()) != -1) {
final byte b = (byte) data;
byte[] delimiter = getDelimiter(b, dPos, delimiterBuffer);
// last read byte is part of the chunk delimiter
if (delimiter != null && b == delimiter[dPos]) {
delimiterBuffer[dPos++] = b;
if (dPos == delimiter.length) {
// found chunk delimiter
break;
}
} else if (dPos > 0) {
delimiter = getDelimiter(dPos - 1, delimiterBuffer);
delimiterBuffer[dPos] = b;
int matched = matchTail(delimiterBuffer, 1, dPos, delimiter);
if (matched == 0) {
// flush delimiter buffer
buffer.write(delimiterBuffer, 0, dPos);
buffer.write(b);
dPos = 0;
} else if (matched == delimiter.length) {
// found chunk delimiter
break;
} else {
// one or more elements of a previous buffered delimiter
// are parts of a current buffered delimiter
buffer.write(delimiterBuffer, 0, dPos + 1 - matched);
dPos = matched;
}
} else {
buffer.write(b);
}
}
} while (data != -1 && buffer.size() == 0); // skip an empty chunk
if (dPos > 0 && dPos != getDelimiter(dPos - 1, delimiterBuffer).length) {
// flush the delimiter buffer, if not empty - parsing finished in the middle of a potential delimiter sequence
buffer.write(delimiterBuffer, 0, dPos);
}
return (buffer.size() > 0) ? buffer.toByteArray() : null;
}
/**
* Selects a delimiter which corresponds to delimiter buffer. Method automatically appends {@code b} param on the
* {@code pos} position of {@code delimiterBuffer} array and then starts the selection process with a newly created array.
*
* @param b byte which will be added on the {@code pos} position of {@code delimiterBuffer} array
* @param pos number of bytes from the delimiter buffer which will be used in processing
* @param delimiterBuffer current content of the delimiter buffer
* @return delimiter which corresponds to delimiterBuffer
*/
abstract byte[] getDelimiter(byte b, int pos, byte[] delimiterBuffer);
/**
* Selects a delimiter which corresponds to delimiter buffer.
*
* @param pos position of the last read byte
* @param delimiterBuffer number of bytes from the delimiter buffer which will be used in processing
* @return delimiter which corresponds to delimiterBuffer
*/
abstract byte[] getDelimiter(int pos, byte[] delimiterBuffer);
/**
* Returns a delimiter buffer size depending on the selected strategy.
*
* If a strategy has multiple registered delimiters, then the delimiter buffer should be a length of the longest
* delimiter.
*
* @return length of the delimiter buffer
*/
abstract int getDelimiterBufferSize();
/**
* Tries to find an element intersection between two arrays in a way that intersecting elements must be
* at the tail of the first array and at the beginning of the second array.
*
* For example, consider the following two arrays:
*
* a1: {a, b, c, d, e}
* a2: {d, e, f, g}
*
* In this example, the intersection of tail of {@code a1} with head of {@code a2} is {d, e}
* and consists of 2 overlapping elements.
*
* The method takes the first array represented as a sub-array in buffer demarcated by an offset and length.
* The second array is a fixed pattern to be matched. The method then compares the tail of the
* array in the buffer with the head of the pattern and returns the number of intersecting elements,
* or zero in case the two arrays do not intersect tail to head.
*
* @param buffer byte buffer containing the array whose tail to intersect.
* @param offset start of the array to be tail-matched in the {@code buffer}.
* @param length length of the array to be tail-matched.
* @param pattern pattern to be head-matched.
* @return {@code 0} if any part of the tail of the array in the buffer does not match
* any part of the head of the pattern, otherwise returns number of overlapping elements.
*/
private static int matchTail(byte[] buffer, int offset, int length, byte[] pattern) {
if (pattern == null) {
return 0;
}
outer:
for (int i = 0; i < length; i++) {
final int tailLength = length - i;
for (int j = 0; j < tailLength; j++) {
if (buffer[offset + i + j] != pattern[j]) {
// mismatch - continue with shorter tail
continue outer;
}
}
// found the longest matching tail
return tailLength;
}
return 0;
}
}
private static class FixedBoundaryParser extends AbstractBoundaryParser {
private final byte[] delimiter;
public FixedBoundaryParser(final byte[] boundary) {
delimiter = Arrays.copyOf(boundary, boundary.length);
}
@Override
byte[] getDelimiter(byte b, int pos, byte[] delimiterBuffer) {
return delimiter;
}
@Override
byte[] getDelimiter(int pos, byte[] delimiterBuffer) {
return delimiter;
}
@Override
int getDelimiterBufferSize() {
return delimiter.length;
}
}
private static class FixedMultiBoundaryParser extends AbstractBoundaryParser {
private final List delimiters = new ArrayList();
private final int longestDelimiterLength;
public FixedMultiBoundaryParser(String... boundaries) {
for (String boundary: boundaries) {
byte[] boundaryBytes = boundary.getBytes();
delimiters.add(Arrays.copyOf(boundaryBytes, boundaryBytes.length));
}
Collections.sort(delimiters, new Comparator() {
@Override
public int compare(byte[] o1, byte[] o2) {
return Integer.compare(o1.length, o2.length);
}
});
byte[] longestDelimiter = delimiters.get(delimiters.size() - 1);
this.longestDelimiterLength = longestDelimiter.length;
}
@Override
byte[] getDelimiter(byte b, int pos, byte[] delimiterBuffer) {
byte[] buffer = Arrays.copyOf(delimiterBuffer, delimiterBuffer.length);
buffer[pos] = b;
return getDelimiter(pos, buffer);
}
@Override
byte[] getDelimiter(int pos, byte[] delimiterBuffer) {
outer:
for (byte[] delimiter: delimiters) {
if (pos > delimiter.length) {
continue;
}
for (int i = 0; i <= pos && i < delimiter.length; i++) {
if (delimiter[i] != delimiterBuffer[i]) {
continue outer;
} else if (pos == i) {
return delimiter;
}
}
}
return null;
}
@Override
int getDelimiterBufferSize() {
return this.longestDelimiterLength;
}
}
/**
* Package-private constructor used by the {@link ChunkedInputReader}.
*
* @param chunkType chunk type.
* @param inputStream response input stream.
* @param annotations annotations associated with response entity.
* @param mediaType response entity media type.
* @param headers response headers.
* @param messageBodyWorkers message body workers.
* @param propertiesDelegate properties delegate for this request/response.
*/
protected ChunkedInput(
final Type chunkType,
final InputStream inputStream,
final Annotation[] annotations,
final MediaType mediaType,
final MultivaluedMap headers,
final MessageBodyWorkers messageBodyWorkers,
final PropertiesDelegate propertiesDelegate) {
super(chunkType);
this.inputStream = inputStream;
this.annotations = annotations;
this.mediaType = mediaType;
this.headers = headers;
this.messageBodyWorkers = messageBodyWorkers;
this.propertiesDelegate = propertiesDelegate;
}
/**
* Get the underlying chunk parser.
*
* Note: Access to internal chunk parser is not a thread-safe operation and has to be explicitly synchronized
* in case the chunked input is used from multiple threads.
*
*
* @return underlying chunk parser.
*/
public ChunkParser getParser() {
return parser;
}
/**
* Set new chunk parser.
*
* Note: Access to internal chunk parser is not a thread-safe operation and has to be explicitly synchronized
* in case the chunked input is used from multiple threads.
*
*
* @param parser new chunk parser.
*/
public void setParser(final ChunkParser parser) {
this.parser = parser;
}
/**
* Get chunk data media type.
*
* Default chunk data media type is derived from the value of the response
* {@value jakarta.ws.rs.core.HttpHeaders#CONTENT_TYPE} header field.
* This default value may be manually overridden by {@link #setChunkType(jakarta.ws.rs.core.MediaType) setting}
* a custom non-{@code null} chunk media type value.
*
* Note: Access to internal chunk media type is not a thread-safe operation and has to
* be explicitly synchronized in case the chunked input is used from multiple threads.
*
*
* @return media type specific to each chunk of data.
*/
public MediaType getChunkType() {
return mediaType;
}
/**
* Set custom chunk data media type.
*
* By default, chunk data media type is derived from the value of the response
* {@value jakarta.ws.rs.core.HttpHeaders#CONTENT_TYPE} header field.
* Using this methods will override the default chunk media type value and set it
* to a custom non-{@code null} chunk media type. Once this method is invoked,
* all subsequent {@link #read chunk reads} will use the newly set chunk media
* type when selecting the proper {@link jakarta.ws.rs.ext.MessageBodyReader} for
* chunk de-serialization.
*
* Note: Access to internal chunk media type is not a thread-safe operation and has to
* be explicitly synchronized in case the chunked input is used from multiple threads.
*
*
* @param mediaType custom chunk data media type. Must not be {@code null}.
* @throws IllegalArgumentException in case the {@code mediaType} is {@code null}.
*/
public void setChunkType(final MediaType mediaType) throws IllegalArgumentException {
if (mediaType == null) {
throw new IllegalArgumentException(LocalizationMessages.CHUNKED_INPUT_MEDIA_TYPE_NULL());
}
this.mediaType = mediaType;
}
/**
* Set custom chunk data media type from a string value.
*
* Note: Access to internal chunk media type is not a thread-safe operation and has to
* be explicitly synchronized in case the chunked input is used from multiple threads.
*
*
* @param mediaType custom chunk data media type. Must not be {@code null}.
* @throws IllegalArgumentException in case the {@code mediaType} cannot be parsed into
* a valid {@link MediaType} instance or is {@code null}.
* @see #setChunkType(jakarta.ws.rs.core.MediaType)
*/
public void setChunkType(final String mediaType) throws IllegalArgumentException {
this.mediaType = MediaType.valueOf(mediaType);
}
@Override
public void close() {
if (closed.compareAndSet(false, true)) {
if (inputStream != null) {
try {
inputStream.close();
} catch (final IOException e) {
LOGGER.log(Level.FINE, LocalizationMessages.CHUNKED_INPUT_STREAM_CLOSING_ERROR(), e);
}
}
}
}
/**
* Check if the chunked input has been closed.
*
* @return {@code true} if this chunked input has been closed, {@code false} otherwise.
*/
public boolean isClosed() {
return closed.get();
}
/**
* Read next chunk from the response stream and convert it to a Java instance
* using the {@link #getChunkType() chunk media type}. The method returns {@code null}
* if the underlying entity input stream has been closed (either implicitly or explicitly
* by calling the {@link #close()} method).
*
* Note: Access to internal chunk parser is not a thread-safe operation and has to be explicitly
* synchronized in case the chunked input is used from multiple threads.
*
*
* @return next streamed chunk or {@code null} if the underlying entity input stream
* has been closed while reading next chunk data.
* @throws IllegalStateException in case this chunked input has been closed.
*/
@SuppressWarnings("unchecked")
public T read() throws IllegalStateException {
if (closed.get()) {
throw new IllegalStateException(LocalizationMessages.CHUNKED_INPUT_CLOSED());
}
try {
final byte[] chunk = parser.readChunk(inputStream);
if (chunk == null) {
close();
} else {
final ByteArrayInputStream chunkStream = new ByteArrayInputStream(chunk);
// TODO: add interceptors: interceptors are used in ChunkedOutput, so the stream should
// be intercepted in the ChunkedInput too. Interceptors cannot be easily added to the readFrom
// method as they should wrap the stream before it is processed by ChunkParser. Also please check todo
// in ChunkedInput (this should be fixed together with this todo)
// issue: JERSEY-1809
return (T) messageBodyWorkers.readFrom(
getRawType(),
getType(),
annotations,
mediaType,
headers,
propertiesDelegate,
chunkStream,
Collections.emptyList(),
false);
}
} catch (final IOException e) {
Logger.getLogger(this.getClass().getName()).log(Level.FINE, e.getMessage(), e);
close();
}
return null;
}
}