/*
- * Copyright 2016 Facebook, Inc.
+ * Copyright 2017 Facebook, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#if FOLLY_HAVE_LIBLZ4
#include <lz4.h>
#include <lz4hc.h>
+#if LZ4_VERSION_NUMBER >= 10301
+#include <lz4frame.h>
+#endif
#endif
#include <glog/logging.h>
#if FOLLY_HAVE_LIBSNAPPY
-#include <snappy.h>
#include <snappy-sinksource.h>
+#include <snappy.h>
#endif
#if FOLLY_HAVE_LIBZ
-#include <zlib.h>
+#include <folly/io/compression/Zlib.h>
#endif
#if FOLLY_HAVE_LIBLZMA
#endif
#if FOLLY_HAVE_LIBZSTD
+#define ZSTD_STATIC_LINKING_ONLY
#include <zstd.h>
#endif
+#if FOLLY_HAVE_LIBBZ2
+#include <bzlib.h>
+#endif
+
+#include <folly/Bits.h>
#include <folly/Conv.h>
#include <folly/Memory.h>
#include <folly/Portability.h>
#include <folly/ScopeGuard.h>
#include <folly/Varint.h>
#include <folly/io/Cursor.h>
+#include <folly/io/compression/Utils.h>
+#include <algorithm>
+#include <unordered_set>
+
+using folly::io::compression::detail::dataStartsWithLE;
+using folly::io::compression::detail::prefixToStringLE;
-namespace folly { namespace io {
+namespace folly {
+namespace io {
Codec::Codec(CodecType type) : type_(type) { }
// Ensure consistent behavior in the nullptr case
std::unique_ptr<IOBuf> Codec::compress(const IOBuf* data) {
+ if (data == nullptr) {
+ throw std::invalid_argument("Codec: data must not be nullptr");
+ }
uint64_t len = data->computeChainDataLength();
- if (len == 0) {
- return IOBuf::create(0);
- } else if (len > maxUncompressedLength()) {
+ if (len > maxUncompressedLength()) {
throw std::runtime_error("Codec: uncompressed length too large");
}
return doCompress(data);
}
-std::unique_ptr<IOBuf> Codec::uncompress(const IOBuf* data,
- uint64_t uncompressedLength) {
- if (uncompressedLength == UNKNOWN_UNCOMPRESSED_LENGTH) {
+std::string Codec::compress(const StringPiece data) {
+ const uint64_t len = data.size();
+ if (len > maxUncompressedLength()) {
+ throw std::runtime_error("Codec: uncompressed length too large");
+ }
+
+ return doCompressString(data);
+}
+
+std::unique_ptr<IOBuf> Codec::uncompress(
+ const IOBuf* data,
+ Optional<uint64_t> uncompressedLength) {
+ if (data == nullptr) {
+ throw std::invalid_argument("Codec: data must not be nullptr");
+ }
+ if (!uncompressedLength) {
if (needsUncompressedLength()) {
throw std::invalid_argument("Codec: uncompressed length required");
}
- } else if (uncompressedLength > maxUncompressedLength()) {
+ } else if (*uncompressedLength > maxUncompressedLength()) {
throw std::runtime_error("Codec: uncompressed length too large");
}
if (data->empty()) {
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- uncompressedLength != 0) {
+ if (uncompressedLength.value_or(0) != 0) {
throw std::runtime_error("Codec: invalid uncompressed length");
}
return IOBuf::create(0);
return doUncompress(data, uncompressedLength);
}
+std::string Codec::uncompress(
+ const StringPiece data,
+ Optional<uint64_t> uncompressedLength) {
+ if (!uncompressedLength) {
+ if (needsUncompressedLength()) {
+ throw std::invalid_argument("Codec: uncompressed length required");
+ }
+ } else if (*uncompressedLength > maxUncompressedLength()) {
+ throw std::runtime_error("Codec: uncompressed length too large");
+ }
+
+ if (data.empty()) {
+ if (uncompressedLength.value_or(0) != 0) {
+ throw std::runtime_error("Codec: invalid uncompressed length");
+ }
+ return "";
+ }
+
+ return doUncompressString(data, uncompressedLength);
+}
+
bool Codec::needsUncompressedLength() const {
return doNeedsUncompressedLength();
}
return UNLIMITED_UNCOMPRESSED_LENGTH;
}
+std::vector<std::string> Codec::validPrefixes() const {
+ return {};
+}
+
+bool Codec::canUncompress(const IOBuf*, Optional<uint64_t>) const {
+ return false;
+}
+
+std::string Codec::doCompressString(const StringPiece data) {
+ const IOBuf inputBuffer{IOBuf::WRAP_BUFFER, data};
+ auto outputBuffer = doCompress(&inputBuffer);
+ std::string output;
+ output.reserve(outputBuffer->computeChainDataLength());
+ for (auto range : *outputBuffer) {
+ output.append(reinterpret_cast<const char*>(range.data()), range.size());
+ }
+ return output;
+}
+
+std::string Codec::doUncompressString(
+ const StringPiece data,
+ Optional<uint64_t> uncompressedLength) {
+ const IOBuf inputBuffer{IOBuf::WRAP_BUFFER, data};
+ auto outputBuffer = doUncompress(&inputBuffer, uncompressedLength);
+ std::string output;
+ output.reserve(outputBuffer->computeChainDataLength());
+ for (auto range : *outputBuffer) {
+ output.append(reinterpret_cast<const char*>(range.data()), range.size());
+ }
+ return output;
+}
+
+uint64_t Codec::maxCompressedLength(uint64_t uncompressedLength) const {
+ return doMaxCompressedLength(uncompressedLength);
+}
+
+Optional<uint64_t> Codec::getUncompressedLength(
+ const folly::IOBuf* data,
+ Optional<uint64_t> uncompressedLength) const {
+ auto const compressedLength = data->computeChainDataLength();
+ if (compressedLength == 0) {
+ if (uncompressedLength.value_or(0) != 0) {
+ throw std::runtime_error("Invalid uncompressed length");
+ }
+ return 0;
+ }
+ return doGetUncompressedLength(data, uncompressedLength);
+}
+
+Optional<uint64_t> Codec::doGetUncompressedLength(
+ const folly::IOBuf*,
+ Optional<uint64_t> uncompressedLength) const {
+ return uncompressedLength;
+}
+
+bool StreamCodec::needsDataLength() const {
+ return doNeedsDataLength();
+}
+
+bool StreamCodec::doNeedsDataLength() const {
+ return false;
+}
+
+void StreamCodec::assertStateIs(State expected) const {
+ if (state_ != expected) {
+ throw std::logic_error(folly::to<std::string>(
+ "Codec: state is ", state_, "; expected state ", expected));
+ }
+}
+
+void StreamCodec::resetStream(Optional<uint64_t> uncompressedLength) {
+ state_ = State::RESET;
+ uncompressedLength_ = uncompressedLength;
+ progressMade_ = true;
+ doResetStream();
+}
+
+bool StreamCodec::compressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) {
+ if (state_ == State::RESET && input.empty() &&
+ flushOp == StreamCodec::FlushOp::END &&
+ uncompressedLength().value_or(0) != 0) {
+ throw std::runtime_error("Codec: invalid uncompressed length");
+ }
+
+ if (!uncompressedLength() && needsDataLength()) {
+ throw std::runtime_error("Codec: uncompressed length required");
+ }
+ if (state_ == State::RESET && !input.empty() &&
+ uncompressedLength() == uint64_t(0)) {
+ throw std::runtime_error("Codec: invalid uncompressed length");
+ }
+ // Handle input state transitions
+ switch (flushOp) {
+ case StreamCodec::FlushOp::NONE:
+ if (state_ == State::RESET) {
+ state_ = State::COMPRESS;
+ }
+ assertStateIs(State::COMPRESS);
+ break;
+ case StreamCodec::FlushOp::FLUSH:
+ if (state_ == State::RESET || state_ == State::COMPRESS) {
+ state_ = State::COMPRESS_FLUSH;
+ }
+ assertStateIs(State::COMPRESS_FLUSH);
+ break;
+ case StreamCodec::FlushOp::END:
+ if (state_ == State::RESET || state_ == State::COMPRESS) {
+ state_ = State::COMPRESS_END;
+ }
+ assertStateIs(State::COMPRESS_END);
+ break;
+ }
+ size_t const inputSize = input.size();
+ size_t const outputSize = output.size();
+ bool const done = doCompressStream(input, output, flushOp);
+ if (!done && inputSize == input.size() && outputSize == output.size()) {
+ if (!progressMade_) {
+ throw std::runtime_error("Codec: No forward progress made");
+ }
+ // Throw an exception if there is no progress again next time
+ progressMade_ = false;
+ } else {
+ progressMade_ = true;
+ }
+ // Handle output state transitions
+ if (done) {
+ if (state_ == State::COMPRESS_FLUSH) {
+ state_ = State::COMPRESS;
+ } else if (state_ == State::COMPRESS_END) {
+ state_ = State::END;
+ }
+ // Check internal invariants
+ DCHECK(input.empty());
+ DCHECK(flushOp != StreamCodec::FlushOp::NONE);
+ }
+ return done;
+}
+
+bool StreamCodec::uncompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) {
+ if (state_ == State::RESET && input.empty()) {
+ if (uncompressedLength().value_or(0) == 0) {
+ return true;
+ }
+ return false;
+ }
+ // Handle input state transitions
+ if (state_ == State::RESET) {
+ state_ = State::UNCOMPRESS;
+ }
+ assertStateIs(State::UNCOMPRESS);
+ size_t const inputSize = input.size();
+ size_t const outputSize = output.size();
+ bool const done = doUncompressStream(input, output, flushOp);
+ if (!done && inputSize == input.size() && outputSize == output.size()) {
+ if (!progressMade_) {
+ throw std::runtime_error("Codec: no forward progress made");
+ }
+ // Throw an exception if there is no progress again next time
+ progressMade_ = false;
+ } else {
+ progressMade_ = true;
+ }
+ // Handle output state transitions
+ if (done) {
+ state_ = State::END;
+ }
+ return done;
+}
+
+static std::unique_ptr<IOBuf> addOutputBuffer(
+ MutableByteRange& output,
+ uint64_t size) {
+ DCHECK(output.empty());
+ auto buffer = IOBuf::create(size);
+ buffer->append(buffer->capacity());
+ output = {buffer->writableData(), buffer->length()};
+ return buffer;
+}
+
+std::unique_ptr<IOBuf> StreamCodec::doCompress(IOBuf const* data) {
+ uint64_t const uncompressedLength = data->computeChainDataLength();
+ resetStream(uncompressedLength);
+ uint64_t const maxCompressedLen = maxCompressedLength(uncompressedLength);
+
+ auto constexpr kMaxSingleStepLength = uint64_t(64) << 20; // 64 MB
+ auto constexpr kDefaultBufferLength = uint64_t(4) << 20; // 4 MB
+
+ MutableByteRange output;
+ auto buffer = addOutputBuffer(
+ output,
+ maxCompressedLen <= kMaxSingleStepLength ? maxCompressedLen
+ : kDefaultBufferLength);
+
+ // Compress the entire IOBuf chain into the IOBuf chain pointed to by buffer
+ IOBuf const* current = data;
+ ByteRange input{current->data(), current->length()};
+ StreamCodec::FlushOp flushOp = StreamCodec::FlushOp::NONE;
+ bool done = false;
+ while (!done) {
+ while (input.empty() && current->next() != data) {
+ current = current->next();
+ input = {current->data(), current->length()};
+ }
+ if (current->next() == data) {
+ // This is the last input buffer so end the stream
+ flushOp = StreamCodec::FlushOp::END;
+ }
+ if (output.empty()) {
+ buffer->prependChain(addOutputBuffer(output, kDefaultBufferLength));
+ }
+ done = compressStream(input, output, flushOp);
+ if (done) {
+ DCHECK(input.empty());
+ DCHECK(flushOp == StreamCodec::FlushOp::END);
+ DCHECK_EQ(current->next(), data);
+ }
+ }
+ buffer->prev()->trimEnd(output.size());
+ return buffer;
+}
+
+static uint64_t computeBufferLength(
+ uint64_t const compressedLength,
+ uint64_t const blockSize) {
+ uint64_t constexpr kMaxBufferLength = uint64_t(4) << 20; // 4 MiB
+ uint64_t const goodBufferSize = 4 * std::max(blockSize, compressedLength);
+ return std::min(goodBufferSize, kMaxBufferLength);
+}
+
+std::unique_ptr<IOBuf> StreamCodec::doUncompress(
+ IOBuf const* data,
+ Optional<uint64_t> uncompressedLength) {
+ auto constexpr kMaxSingleStepLength = uint64_t(64) << 20; // 64 MB
+ auto constexpr kBlockSize = uint64_t(128) << 10;
+ auto const defaultBufferLength =
+ computeBufferLength(data->computeChainDataLength(), kBlockSize);
+
+ uncompressedLength = getUncompressedLength(data, uncompressedLength);
+ resetStream(uncompressedLength);
+
+ MutableByteRange output;
+ auto buffer = addOutputBuffer(
+ output,
+ (uncompressedLength && *uncompressedLength <= kMaxSingleStepLength
+ ? *uncompressedLength
+ : defaultBufferLength));
+
+ // Uncompress the entire IOBuf chain into the IOBuf chain pointed to by buffer
+ IOBuf const* current = data;
+ ByteRange input{current->data(), current->length()};
+ StreamCodec::FlushOp flushOp = StreamCodec::FlushOp::NONE;
+ bool done = false;
+ while (!done) {
+ while (input.empty() && current->next() != data) {
+ current = current->next();
+ input = {current->data(), current->length()};
+ }
+ if (current->next() == data) {
+ // Tell the uncompressor there is no more input (it may optimize)
+ flushOp = StreamCodec::FlushOp::END;
+ }
+ if (output.empty()) {
+ buffer->prependChain(addOutputBuffer(output, defaultBufferLength));
+ }
+ done = uncompressStream(input, output, flushOp);
+ }
+ if (!input.empty()) {
+ throw std::runtime_error("Codec: Junk after end of data");
+ }
+
+ buffer->prev()->trimEnd(output.size());
+ if (uncompressedLength &&
+ *uncompressedLength != buffer->computeChainDataLength()) {
+ throw std::runtime_error("Codec: invalid uncompressed length");
+ }
+
+ return buffer;
+}
+
namespace {
/**
explicit NoCompressionCodec(int level, CodecType type);
private:
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
std::unique_ptr<IOBuf> doUncompress(
const IOBuf* data,
- uint64_t uncompressedLength) override;
+ Optional<uint64_t> uncompressedLength) override;
};
std::unique_ptr<Codec> NoCompressionCodec::create(int level, CodecType type) {
- return make_unique<NoCompressionCodec>(level, type);
+ return std::make_unique<NoCompressionCodec>(level, type);
}
NoCompressionCodec::NoCompressionCodec(int level, CodecType type)
- : Codec(type) {
+ : Codec(type) {
DCHECK(type == CodecType::NO_COMPRESSION);
switch (level) {
- case COMPRESSION_LEVEL_DEFAULT:
- case COMPRESSION_LEVEL_FASTEST:
- case COMPRESSION_LEVEL_BEST:
- level = 0;
+ case COMPRESSION_LEVEL_DEFAULT:
+ case COMPRESSION_LEVEL_FASTEST:
+ case COMPRESSION_LEVEL_BEST:
+ level = 0;
}
if (level != 0) {
throw std::invalid_argument(to<std::string>(
}
}
+uint64_t NoCompressionCodec::doMaxCompressedLength(
+ uint64_t uncompressedLength) const {
+ return uncompressedLength;
+}
+
std::unique_ptr<IOBuf> NoCompressionCodec::doCompress(
const IOBuf* data) {
return data->clone();
std::unique_ptr<IOBuf> NoCompressionCodec::doUncompress(
const IOBuf* data,
- uint64_t uncompressedLength) {
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- data->computeChainDataLength() != uncompressedLength) {
- throw std::runtime_error(to<std::string>(
- "NoCompressionCodec: invalid uncompressed length"));
+ Optional<uint64_t> uncompressedLength) {
+ if (uncompressedLength &&
+ data->computeChainDataLength() != *uncompressedLength) {
+ throw std::runtime_error(
+ to<std::string>("NoCompressionCodec: invalid uncompressed length"));
}
return data->clone();
}
return val;
}
-} // namespace
+} // namespace
#endif // FOLLY_HAVE_LIBLZ4 || FOLLY_HAVE_LIBLZMA
private:
bool doNeedsUncompressedLength() const override;
uint64_t doMaxUncompressedLength() const override;
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
bool encodeSize() const { return type() == CodecType::LZ4_VARINT_SIZE; }
std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
std::unique_ptr<IOBuf> doUncompress(
const IOBuf* data,
- uint64_t uncompressedLength) override;
+ Optional<uint64_t> uncompressedLength) override;
bool highCompression_;
};
std::unique_ptr<Codec> LZ4Codec::create(int level, CodecType type) {
- return make_unique<LZ4Codec>(level, type);
+ return std::make_unique<LZ4Codec>(level, type);
}
LZ4Codec::LZ4Codec(int level, CodecType type) : Codec(type) {
DCHECK(type == CodecType::LZ4 || type == CodecType::LZ4_VARINT_SIZE);
switch (level) {
- case COMPRESSION_LEVEL_FASTEST:
- case COMPRESSION_LEVEL_DEFAULT:
- level = 1;
- break;
- case COMPRESSION_LEVEL_BEST:
- level = 2;
- break;
+ case COMPRESSION_LEVEL_FASTEST:
+ case COMPRESSION_LEVEL_DEFAULT:
+ level = 1;
+ break;
+ case COMPRESSION_LEVEL_BEST:
+ level = 2;
+ break;
}
if (level < 1 || level > 2) {
throw std::invalid_argument(to<std::string>(
return LZ4_MAX_INPUT_SIZE;
}
+uint64_t LZ4Codec::doMaxCompressedLength(uint64_t uncompressedLength) const {
+ return LZ4_compressBound(uncompressedLength) +
+ (encodeSize() ? kMaxVarintLength64 : 0);
+}
+
std::unique_ptr<IOBuf> LZ4Codec::doCompress(const IOBuf* data) {
- std::unique_ptr<IOBuf> clone;
+ IOBuf clone;
if (data->isChained()) {
// LZ4 doesn't support streaming, so we have to coalesce
- clone = data->clone();
- clone->coalesce();
- data = clone.get();
+ clone = data->cloneCoalescedAsValue();
+ data = &clone;
}
- uint32_t extraSize = encodeSize() ? kMaxVarintLength64 : 0;
- auto out = IOBuf::create(extraSize + LZ4_compressBound(data->length()));
+ auto out = IOBuf::create(maxCompressedLength(data->length()));
if (encodeSize()) {
encodeVarintToIOBuf(data->length(), out.get());
}
int n;
+ auto input = reinterpret_cast<const char*>(data->data());
+ auto output = reinterpret_cast<char*>(out->writableTail());
+ const auto inputLength = data->length();
+#if LZ4_VERSION_NUMBER >= 10700
+ if (highCompression_) {
+ n = LZ4_compress_HC(input, output, inputLength, out->tailroom(), 0);
+ } else {
+ n = LZ4_compress_default(input, output, inputLength, out->tailroom());
+ }
+#else
if (highCompression_) {
- n = LZ4_compressHC(reinterpret_cast<const char*>(data->data()),
- reinterpret_cast<char*>(out->writableTail()),
- data->length());
+ n = LZ4_compressHC(input, output, inputLength);
} else {
- n = LZ4_compress(reinterpret_cast<const char*>(data->data()),
- reinterpret_cast<char*>(out->writableTail()),
- data->length());
+ n = LZ4_compress(input, output, inputLength);
}
+#endif
CHECK_GE(n, 0);
CHECK_LE(n, out->capacity());
std::unique_ptr<IOBuf> LZ4Codec::doUncompress(
const IOBuf* data,
- uint64_t uncompressedLength) {
- std::unique_ptr<IOBuf> clone;
+ Optional<uint64_t> uncompressedLength) {
+ IOBuf clone;
if (data->isChained()) {
// LZ4 doesn't support streaming, so we have to coalesce
- clone = data->clone();
- clone->coalesce();
- data = clone.get();
+ clone = data->cloneCoalescedAsValue();
+ data = &clone;
}
folly::io::Cursor cursor(data);
uint64_t actualUncompressedLength;
if (encodeSize()) {
actualUncompressedLength = decodeVarintFromCursor(cursor);
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- uncompressedLength != actualUncompressedLength) {
+ if (uncompressedLength && *uncompressedLength != actualUncompressedLength) {
throw std::runtime_error("LZ4Codec: invalid uncompressed length");
}
} else {
- actualUncompressedLength = uncompressedLength;
- if (actualUncompressedLength == UNKNOWN_UNCOMPRESSED_LENGTH ||
- actualUncompressedLength > maxUncompressedLength()) {
- throw std::runtime_error("LZ4Codec: invalid uncompressed length");
- }
+ // Invariants
+ DCHECK(uncompressedLength.hasValue());
+ DCHECK(*uncompressedLength <= maxUncompressedLength());
+ actualUncompressedLength = *uncompressedLength;
}
auto sp = StringPiece{cursor.peekBytes()};
return out;
}
-#endif // FOLLY_HAVE_LIBLZ4
+#if LZ4_VERSION_NUMBER >= 10301
+
+class LZ4FrameCodec final : public Codec {
+ public:
+ static std::unique_ptr<Codec> create(int level, CodecType type);
+ explicit LZ4FrameCodec(int level, CodecType type);
+ ~LZ4FrameCodec() override;
+
+ std::vector<std::string> validPrefixes() const override;
+ bool canUncompress(const IOBuf* data, Optional<uint64_t> uncompressedLength)
+ const override;
+
+ private:
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
+
+ std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
+ std::unique_ptr<IOBuf> doUncompress(
+ const IOBuf* data,
+ Optional<uint64_t> uncompressedLength) override;
+
+ // Reset the dctx_ if it is dirty or null.
+ void resetDCtx();
+
+ int level_;
+ LZ4F_decompressionContext_t dctx_{nullptr};
+ bool dirty_{false};
+};
+
+/* static */ std::unique_ptr<Codec> LZ4FrameCodec::create(
+ int level,
+ CodecType type) {
+ return std::make_unique<LZ4FrameCodec>(level, type);
+}
+
+static constexpr uint32_t kLZ4FrameMagicLE = 0x184D2204;
+
+std::vector<std::string> LZ4FrameCodec::validPrefixes() const {
+ return {prefixToStringLE(kLZ4FrameMagicLE)};
+}
+
+bool LZ4FrameCodec::canUncompress(const IOBuf* data, Optional<uint64_t>) const {
+ return dataStartsWithLE(data, kLZ4FrameMagicLE);
+}
+
+uint64_t LZ4FrameCodec::doMaxCompressedLength(
+ uint64_t uncompressedLength) const {
+ LZ4F_preferences_t prefs{};
+ prefs.compressionLevel = level_;
+ prefs.frameInfo.contentSize = uncompressedLength;
+ return LZ4F_compressFrameBound(uncompressedLength, &prefs);
+}
+
+static size_t lz4FrameThrowOnError(size_t code) {
+ if (LZ4F_isError(code)) {
+ throw std::runtime_error(
+ to<std::string>("LZ4Frame error: ", LZ4F_getErrorName(code)));
+ }
+ return code;
+}
+
+void LZ4FrameCodec::resetDCtx() {
+ if (dctx_ && !dirty_) {
+ return;
+ }
+ if (dctx_) {
+ LZ4F_freeDecompressionContext(dctx_);
+ }
+ lz4FrameThrowOnError(LZ4F_createDecompressionContext(&dctx_, 100));
+ dirty_ = false;
+}
+
+LZ4FrameCodec::LZ4FrameCodec(int level, CodecType type) : Codec(type) {
+ DCHECK(type == CodecType::LZ4_FRAME);
+ switch (level) {
+ case COMPRESSION_LEVEL_FASTEST:
+ case COMPRESSION_LEVEL_DEFAULT:
+ level_ = 0;
+ break;
+ case COMPRESSION_LEVEL_BEST:
+ level_ = 16;
+ break;
+ default:
+ level_ = level;
+ break;
+ }
+}
+
+LZ4FrameCodec::~LZ4FrameCodec() {
+ if (dctx_) {
+ LZ4F_freeDecompressionContext(dctx_);
+ }
+}
+
+std::unique_ptr<IOBuf> LZ4FrameCodec::doCompress(const IOBuf* data) {
+ // LZ4 Frame compression doesn't support streaming so we have to coalesce
+ IOBuf clone;
+ if (data->isChained()) {
+ clone = data->cloneCoalescedAsValue();
+ data = &clone;
+ }
+ // Set preferences
+ const auto uncompressedLength = data->length();
+ LZ4F_preferences_t prefs{};
+ prefs.compressionLevel = level_;
+ prefs.frameInfo.contentSize = uncompressedLength;
+ // Compress
+ auto buf = IOBuf::create(maxCompressedLength(uncompressedLength));
+ const size_t written = lz4FrameThrowOnError(LZ4F_compressFrame(
+ buf->writableTail(),
+ buf->tailroom(),
+ data->data(),
+ data->length(),
+ &prefs));
+ buf->append(written);
+ return buf;
+}
+
+std::unique_ptr<IOBuf> LZ4FrameCodec::doUncompress(
+ const IOBuf* data,
+ Optional<uint64_t> uncompressedLength) {
+ // Reset the dctx if any errors have occurred
+ resetDCtx();
+ // Coalesce the data
+ ByteRange in = *data->begin();
+ IOBuf clone;
+ if (data->isChained()) {
+ clone = data->cloneCoalescedAsValue();
+ in = clone.coalesce();
+ }
+ data = nullptr;
+ // Select decompression options
+ LZ4F_decompressOptions_t options;
+ options.stableDst = 1;
+ // Select blockSize and growthSize for the IOBufQueue
+ IOBufQueue queue(IOBufQueue::cacheChainLength());
+ auto blockSize = uint64_t{64} << 10;
+ auto growthSize = uint64_t{4} << 20;
+ if (uncompressedLength) {
+ // Allocate uncompressedLength in one chunk (up to 64 MB)
+ const auto allocateSize = std::min(*uncompressedLength, uint64_t{64} << 20);
+ queue.preallocate(allocateSize, allocateSize);
+ blockSize = std::min(*uncompressedLength, blockSize);
+ growthSize = std::min(*uncompressedLength, growthSize);
+ } else {
+ // Reduce growthSize for small data
+ const auto guessUncompressedLen =
+ 4 * std::max<uint64_t>(blockSize, in.size());
+ growthSize = std::min(guessUncompressedLen, growthSize);
+ }
+ // Once LZ4_decompress() is called, the dctx_ cannot be reused until it
+ // returns 0
+ dirty_ = true;
+ // Decompress until the frame is over
+ size_t code = 0;
+ do {
+ // Allocate enough space to decompress at least a block
+ void* out;
+ size_t outSize;
+ std::tie(out, outSize) = queue.preallocate(blockSize, growthSize);
+ // Decompress
+ size_t inSize = in.size();
+ code = lz4FrameThrowOnError(
+ LZ4F_decompress(dctx_, out, &outSize, in.data(), &inSize, &options));
+ if (in.empty() && outSize == 0 && code != 0) {
+ // We passed no input, no output was produced, and the frame isn't over
+ // No more forward progress is possible
+ throw std::runtime_error("LZ4Frame error: Incomplete frame");
+ }
+ in.uncheckedAdvance(inSize);
+ queue.postallocate(outSize);
+ } while (code != 0);
+ // At this point the decompression context can be reused
+ dirty_ = false;
+ if (uncompressedLength && queue.chainLength() != *uncompressedLength) {
+ throw std::runtime_error("LZ4Frame error: Invalid uncompressedLength");
+ }
+ return queue.move();
+}
+
+#endif // LZ4_VERSION_NUMBER >= 10301
+#endif // FOLLY_HAVE_LIBLZ4
#if FOLLY_HAVE_LIBSNAPPY
private:
uint64_t doMaxUncompressedLength() const override;
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
std::unique_ptr<IOBuf> doUncompress(
const IOBuf* data,
- uint64_t uncompressedLength) override;
+ Optional<uint64_t> uncompressedLength) override;
};
std::unique_ptr<Codec> SnappyCodec::create(int level, CodecType type) {
- return make_unique<SnappyCodec>(level, type);
+ return std::make_unique<SnappyCodec>(level, type);
}
SnappyCodec::SnappyCodec(int level, CodecType type) : Codec(type) {
DCHECK(type == CodecType::SNAPPY);
switch (level) {
- case COMPRESSION_LEVEL_FASTEST:
- case COMPRESSION_LEVEL_DEFAULT:
- case COMPRESSION_LEVEL_BEST:
- level = 1;
+ case COMPRESSION_LEVEL_FASTEST:
+ case COMPRESSION_LEVEL_DEFAULT:
+ case COMPRESSION_LEVEL_BEST:
+ level = 1;
}
if (level != 1) {
throw std::invalid_argument(to<std::string>(
return std::numeric_limits<uint32_t>::max();
}
+uint64_t SnappyCodec::doMaxCompressedLength(uint64_t uncompressedLength) const {
+ return snappy::MaxCompressedLength(uncompressedLength);
+}
+
std::unique_ptr<IOBuf> SnappyCodec::doCompress(const IOBuf* data) {
IOBufSnappySource source(data);
- auto out =
- IOBuf::create(snappy::MaxCompressedLength(source.Available()));
+ auto out = IOBuf::create(maxCompressedLength(source.Available()));
snappy::UncheckedByteArraySink sink(reinterpret_cast<char*>(
out->writableTail()));
return out;
}
-std::unique_ptr<IOBuf> SnappyCodec::doUncompress(const IOBuf* data,
- uint64_t uncompressedLength) {
+std::unique_ptr<IOBuf> SnappyCodec::doUncompress(
+ const IOBuf* data,
+ Optional<uint64_t> uncompressedLength) {
uint32_t actualUncompressedLength = 0;
{
if (!snappy::GetUncompressedLength(&source, &actualUncompressedLength)) {
throw std::runtime_error("snappy::GetUncompressedLength failed");
}
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- uncompressedLength != actualUncompressedLength) {
+ if (uncompressedLength && *uncompressedLength != actualUncompressedLength) {
throw std::runtime_error("snappy: invalid uncompressed length");
}
}
#endif // FOLLY_HAVE_LIBSNAPPY
-#if FOLLY_HAVE_LIBZ
+#if FOLLY_HAVE_LIBLZMA
+
/**
- * Zlib codec
+ * LZMA2 compression
*/
-class ZlibCodec final : public Codec {
+class LZMA2StreamCodec final : public StreamCodec {
public:
- static std::unique_ptr<Codec> create(int level, CodecType type);
- explicit ZlibCodec(int level, CodecType type);
+ static std::unique_ptr<Codec> createCodec(int level, CodecType type);
+ static std::unique_ptr<StreamCodec> createStream(int level, CodecType type);
+ explicit LZMA2StreamCodec(int level, CodecType type);
+ ~LZMA2StreamCodec() override;
+
+ std::vector<std::string> validPrefixes() const override;
+ bool canUncompress(const IOBuf* data, Optional<uint64_t> uncompressedLength)
+ const override;
private:
- std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
- std::unique_ptr<IOBuf> doUncompress(
- const IOBuf* data,
- uint64_t uncompressedLength) override;
+ bool doNeedsDataLength() const override;
+ uint64_t doMaxUncompressedLength() const override;
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
+
+ bool encodeSize() const {
+ return type() == CodecType::LZMA2_VARINT_SIZE;
+ }
+
+ void doResetStream() override;
+ bool doCompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) override;
+ bool doUncompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) override;
+
+ void resetCStream();
+ void resetDStream();
- std::unique_ptr<IOBuf> addOutputBuffer(z_stream* stream, uint32_t length);
- bool doInflate(z_stream* stream, IOBuf* head, uint32_t bufferLength);
+ bool decodeAndCheckVarint(ByteRange& input);
+ bool flushVarintBuffer(MutableByteRange& output);
+ void resetVarintBuffer();
+
+ Optional<lzma_stream> cstream_{};
+ Optional<lzma_stream> dstream_{};
+
+ std::array<uint8_t, kMaxVarintLength64> varintBuffer_;
+ ByteRange varintToEncode_;
+ size_t varintBufferPos_{0};
int level_;
+ bool needReset_{true};
+ bool needDecodeSize_{false};
};
-std::unique_ptr<Codec> ZlibCodec::create(int level, CodecType type) {
- return make_unique<ZlibCodec>(level, type);
+static constexpr uint64_t kLZMA2MagicLE = 0x005A587A37FD;
+static constexpr unsigned kLZMA2MagicBytes = 6;
+
+std::vector<std::string> LZMA2StreamCodec::validPrefixes() const {
+ if (type() == CodecType::LZMA2_VARINT_SIZE) {
+ return {};
+ }
+ return {prefixToStringLE(kLZMA2MagicLE, kLZMA2MagicBytes)};
+}
+
+bool LZMA2StreamCodec::doNeedsDataLength() const {
+ return encodeSize();
+}
+
+bool LZMA2StreamCodec::canUncompress(const IOBuf* data, Optional<uint64_t>)
+ const {
+ if (type() == CodecType::LZMA2_VARINT_SIZE) {
+ return false;
+ }
+ // Returns false for all inputs less than 8 bytes.
+ // This is okay, because no valid LZMA2 streams are less than 8 bytes.
+ return dataStartsWithLE(data, kLZMA2MagicLE, kLZMA2MagicBytes);
+}
+
+std::unique_ptr<Codec> LZMA2StreamCodec::createCodec(
+ int level,
+ CodecType type) {
+ return make_unique<LZMA2StreamCodec>(level, type);
+}
+
+std::unique_ptr<StreamCodec> LZMA2StreamCodec::createStream(
+ int level,
+ CodecType type) {
+ return make_unique<LZMA2StreamCodec>(level, type);
}
-ZlibCodec::ZlibCodec(int level, CodecType type) : Codec(type) {
- DCHECK(type == CodecType::ZLIB || type == CodecType::GZIP);
+LZMA2StreamCodec::LZMA2StreamCodec(int level, CodecType type)
+ : StreamCodec(type) {
+ DCHECK(type == CodecType::LZMA2 || type == CodecType::LZMA2_VARINT_SIZE);
switch (level) {
- case COMPRESSION_LEVEL_FASTEST:
- level = 1;
- break;
- case COMPRESSION_LEVEL_DEFAULT:
- level = Z_DEFAULT_COMPRESSION;
- break;
- case COMPRESSION_LEVEL_BEST:
- level = 9;
- break;
- }
- if (level != Z_DEFAULT_COMPRESSION && (level < 0 || level > 9)) {
- throw std::invalid_argument(to<std::string>(
- "ZlibCodec: invalid level: ", level));
+ case COMPRESSION_LEVEL_FASTEST:
+ level = 0;
+ break;
+ case COMPRESSION_LEVEL_DEFAULT:
+ level = LZMA_PRESET_DEFAULT;
+ break;
+ case COMPRESSION_LEVEL_BEST:
+ level = 9;
+ break;
+ }
+ if (level < 0 || level > 9) {
+ throw std::invalid_argument(
+ to<std::string>("LZMA2Codec: invalid level: ", level));
}
level_ = level;
}
-std::unique_ptr<IOBuf> ZlibCodec::addOutputBuffer(z_stream* stream,
- uint32_t length) {
- CHECK_EQ(stream->avail_out, 0);
+LZMA2StreamCodec::~LZMA2StreamCodec() {
+ if (cstream_) {
+ lzma_end(cstream_.get_pointer());
+ cstream_.clear();
+ }
+ if (dstream_) {
+ lzma_end(dstream_.get_pointer());
+ dstream_.clear();
+ }
+}
- auto buf = IOBuf::create(length);
- buf->append(length);
+uint64_t LZMA2StreamCodec::doMaxUncompressedLength() const {
+ // From lzma/base.h: "Stream is roughly 8 EiB (2^63 bytes)"
+ return uint64_t(1) << 63;
+}
- stream->next_out = buf->writableData();
- stream->avail_out = buf->length();
+uint64_t LZMA2StreamCodec::doMaxCompressedLength(
+ uint64_t uncompressedLength) const {
+ return lzma_stream_buffer_bound(uncompressedLength) +
+ (encodeSize() ? kMaxVarintLength64 : 0);
+}
- return buf;
+void LZMA2StreamCodec::doResetStream() {
+ needReset_ = true;
}
-bool ZlibCodec::doInflate(z_stream* stream,
- IOBuf* head,
- uint32_t bufferLength) {
- if (stream->avail_out == 0) {
- head->prependChain(addOutputBuffer(stream, bufferLength));
+void LZMA2StreamCodec::resetCStream() {
+ if (!cstream_) {
+ cstream_.assign(LZMA_STREAM_INIT);
+ }
+ lzma_ret const rc =
+ lzma_easy_encoder(cstream_.get_pointer(), level_, LZMA_CHECK_NONE);
+ if (rc != LZMA_OK) {
+ throw std::runtime_error(folly::to<std::string>(
+ "LZMA2StreamCodec: lzma_easy_encoder error: ", rc));
}
+}
- int rc = inflate(stream, Z_NO_FLUSH);
+void LZMA2StreamCodec::resetDStream() {
+ if (!dstream_) {
+ dstream_.assign(LZMA_STREAM_INIT);
+ }
+ lzma_ret const rc = lzma_auto_decoder(
+ dstream_.get_pointer(), std::numeric_limits<uint64_t>::max(), 0);
+ if (rc != LZMA_OK) {
+ throw std::runtime_error(folly::to<std::string>(
+ "LZMA2StreamCodec: lzma_auto_decoder error: ", rc));
+ }
+}
+static lzma_ret lzmaThrowOnError(lzma_ret const rc) {
switch (rc) {
- case Z_OK:
- break;
- case Z_STREAM_END:
- return true;
- case Z_BUF_ERROR:
- case Z_NEED_DICT:
- case Z_DATA_ERROR:
- case Z_MEM_ERROR:
- throw std::runtime_error(to<std::string>(
- "ZlibCodec: inflate error: ", rc, ": ", stream->msg));
- default:
- CHECK(false) << rc << ": " << stream->msg;
+ case LZMA_OK:
+ case LZMA_STREAM_END:
+ case LZMA_BUF_ERROR: // not fatal: returned if no progress was made twice
+ return rc;
+ default:
+ throw std::runtime_error(
+ to<std::string>("LZMA2StreamCodec: error: ", rc));
}
+}
- return false;
+static lzma_action lzmaTranslateFlush(StreamCodec::FlushOp flush) {
+ switch (flush) {
+ case StreamCodec::FlushOp::NONE:
+ return LZMA_RUN;
+ case StreamCodec::FlushOp::FLUSH:
+ return LZMA_SYNC_FLUSH;
+ case StreamCodec::FlushOp::END:
+ return LZMA_FINISH;
+ default:
+ throw std::invalid_argument("LZMA2StreamCodec: Invalid flush");
+ }
}
-std::unique_ptr<IOBuf> ZlibCodec::doCompress(const IOBuf* data) {
- z_stream stream;
- stream.zalloc = nullptr;
- stream.zfree = nullptr;
- stream.opaque = nullptr;
-
- // Using deflateInit2() to support gzip. "The windowBits parameter is the
- // base two logarithm of the maximum window size (...) The default value is
- // 15 (...) Add 16 to windowBits to write a simple gzip header and trailer
- // around the compressed data instead of a zlib wrapper. The gzip header
- // will have no file name, no extra data, no comment, no modification time
- // (set to zero), no header crc, and the operating system will be set to 255
- // (unknown)."
- int windowBits = 15 + (type() == CodecType::GZIP ? 16 : 0);
- // All other parameters (method, memLevel, strategy) get default values from
- // the zlib manual.
- int rc = deflateInit2(&stream,
- level_,
- Z_DEFLATED,
- windowBits,
- /* memLevel */ 8,
- Z_DEFAULT_STRATEGY);
- if (rc != Z_OK) {
- throw std::runtime_error(to<std::string>(
- "ZlibCodec: deflateInit error: ", rc, ": ", stream.msg));
+/**
+ * Flushes the varint buffer.
+ * Advances output by the number of bytes written.
+ * Returns true when flushing is complete.
+ */
+bool LZMA2StreamCodec::flushVarintBuffer(MutableByteRange& output) {
+ if (varintToEncode_.empty()) {
+ return true;
+ }
+ const size_t numBytesToCopy = std::min(varintToEncode_.size(), output.size());
+ if (numBytesToCopy > 0) {
+ memcpy(output.data(), varintToEncode_.data(), numBytesToCopy);
}
+ varintToEncode_.advance(numBytesToCopy);
+ output.advance(numBytesToCopy);
+ return varintToEncode_.empty();
+}
- stream.next_in = stream.next_out = nullptr;
- stream.avail_in = stream.avail_out = 0;
- stream.total_in = stream.total_out = 0;
+bool LZMA2StreamCodec::doCompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) {
+ if (needReset_) {
+ resetCStream();
+ if (encodeSize()) {
+ varintBufferPos_ = 0;
+ size_t const varintSize =
+ encodeVarint(*uncompressedLength(), varintBuffer_.data());
+ varintToEncode_ = {varintBuffer_.data(), varintSize};
+ }
+ needReset_ = false;
+ }
- bool success = false;
+ if (!flushVarintBuffer(output)) {
+ return false;
+ }
+ cstream_->next_in = const_cast<uint8_t*>(input.data());
+ cstream_->avail_in = input.size();
+ cstream_->next_out = output.data();
+ cstream_->avail_out = output.size();
SCOPE_EXIT {
- rc = deflateEnd(&stream);
- // If we're here because of an exception, it's okay if some data
- // got dropped.
- CHECK(rc == Z_OK || (!success && rc == Z_DATA_ERROR))
- << rc << ": " << stream.msg;
+ input.uncheckedAdvance(input.size() - cstream_->avail_in);
+ output.uncheckedAdvance(output.size() - cstream_->avail_out);
};
+ lzma_ret const rc = lzmaThrowOnError(
+ lzma_code(cstream_.get_pointer(), lzmaTranslateFlush(flushOp)));
+ switch (flushOp) {
+ case StreamCodec::FlushOp::NONE:
+ return false;
+ case StreamCodec::FlushOp::FLUSH:
+ return cstream_->avail_in == 0 && cstream_->avail_out != 0;
+ case StreamCodec::FlushOp::END:
+ return rc == LZMA_STREAM_END;
+ default:
+ throw std::invalid_argument("LZMA2StreamCodec: invalid FlushOp");
+ }
+}
- uint64_t uncompressedLength = data->computeChainDataLength();
- uint64_t maxCompressedLength = deflateBound(&stream, uncompressedLength);
-
- // Max 64MiB in one go
- constexpr uint32_t maxSingleStepLength = uint32_t(64) << 20; // 64MiB
- constexpr uint32_t defaultBufferLength = uint32_t(4) << 20; // 4MiB
+/**
+ * Attempts to decode a varint from input.
+ * The function advances input by the number of bytes read.
+ *
+ * If there are too many bytes and the varint is not valid, throw a
+ * runtime_error.
+ *
+ * If the uncompressed length was provided and a decoded varint does not match
+ * the provided length, throw a runtime_error.
+ *
+ * Returns true if the varint was successfully decoded and matches the
+ * uncompressed length if provided, and false if more bytes are needed.
+ */
+bool LZMA2StreamCodec::decodeAndCheckVarint(ByteRange& input) {
+ if (input.empty()) {
+ return false;
+ }
+ size_t const numBytesToCopy =
+ std::min(kMaxVarintLength64 - varintBufferPos_, input.size());
+ memcpy(varintBuffer_.data() + varintBufferPos_, input.data(), numBytesToCopy);
+
+ size_t const rangeSize = varintBufferPos_ + numBytesToCopy;
+ ByteRange range{varintBuffer_.data(), rangeSize};
+ auto const ret = tryDecodeVarint(range);
+
+ if (ret.hasValue()) {
+ size_t const varintSize = rangeSize - range.size();
+ input.advance(varintSize - varintBufferPos_);
+ if (uncompressedLength() && *uncompressedLength() != ret.value()) {
+ throw std::runtime_error("LZMA2StreamCodec: invalid uncompressed length");
+ }
+ return true;
+ } else if (ret.error() == DecodeVarintError::TooManyBytes) {
+ throw std::runtime_error("LZMA2StreamCodec: invalid uncompressed length");
+ } else {
+ // Too few bytes
+ input.advance(numBytesToCopy);
+ varintBufferPos_ += numBytesToCopy;
+ return false;
+ }
+}
- auto out = addOutputBuffer(
- &stream,
- (maxCompressedLength <= maxSingleStepLength ?
- maxCompressedLength :
- defaultBufferLength));
-
- for (auto& range : *data) {
- uint64_t remaining = range.size();
- uint64_t written = 0;
- while (remaining) {
- uint32_t step = (remaining > maxSingleStepLength ?
- maxSingleStepLength : remaining);
- stream.next_in = const_cast<uint8_t*>(range.data() + written);
- stream.avail_in = step;
- remaining -= step;
- written += step;
-
- while (stream.avail_in != 0) {
- if (stream.avail_out == 0) {
- out->prependChain(addOutputBuffer(&stream, defaultBufferLength));
- }
-
- rc = deflate(&stream, Z_NO_FLUSH);
-
- CHECK_EQ(rc, Z_OK) << stream.msg;
- }
+bool LZMA2StreamCodec::doUncompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) {
+ if (needReset_) {
+ resetDStream();
+ needReset_ = false;
+ needDecodeSize_ = encodeSize();
+ if (encodeSize()) {
+ // Reset buffer
+ varintBufferPos_ = 0;
}
}
- do {
- if (stream.avail_out == 0) {
- out->prependChain(addOutputBuffer(&stream, defaultBufferLength));
+ if (needDecodeSize_) {
+ // Try decoding the varint. If the input does not contain the entire varint,
+ // buffer the input. If the varint can not be decoded, fail.
+ if (!decodeAndCheckVarint(input)) {
+ return false;
}
+ needDecodeSize_ = false;
+ }
- rc = deflate(&stream, Z_FINISH);
- } while (rc == Z_OK);
+ dstream_->next_in = const_cast<uint8_t*>(input.data());
+ dstream_->avail_in = input.size();
+ dstream_->next_out = output.data();
+ dstream_->avail_out = output.size();
+ SCOPE_EXIT {
+ input.advance(input.size() - dstream_->avail_in);
+ output.advance(output.size() - dstream_->avail_out);
+ };
- CHECK_EQ(rc, Z_STREAM_END) << stream.msg;
+ lzma_ret rc;
+ switch (flushOp) {
+ case StreamCodec::FlushOp::NONE:
+ case StreamCodec::FlushOp::FLUSH:
+ rc = lzmaThrowOnError(lzma_code(dstream_.get_pointer(), LZMA_RUN));
+ break;
+ case StreamCodec::FlushOp::END:
+ rc = lzmaThrowOnError(lzma_code(dstream_.get_pointer(), LZMA_FINISH));
+ break;
+ default:
+ throw std::invalid_argument("LZMA2StreamCodec: invalid flush");
+ }
+ return rc == LZMA_STREAM_END;
+}
+#endif // FOLLY_HAVE_LIBLZMA
- out->prev()->trimEnd(stream.avail_out);
+#ifdef FOLLY_HAVE_LIBZSTD
- success = true; // we survived
+namespace {
+void zstdFreeCStream(ZSTD_CStream* zcs) {
+ ZSTD_freeCStream(zcs);
+}
- return out;
+void zstdFreeDStream(ZSTD_DStream* zds) {
+ ZSTD_freeDStream(zds);
+}
}
-std::unique_ptr<IOBuf> ZlibCodec::doUncompress(const IOBuf* data,
- uint64_t uncompressedLength) {
- z_stream stream;
- stream.zalloc = nullptr;
- stream.zfree = nullptr;
- stream.opaque = nullptr;
+/**
+ * ZSTD compression
+ */
+class ZSTDStreamCodec final : public StreamCodec {
+ public:
+ static std::unique_ptr<Codec> createCodec(int level, CodecType);
+ static std::unique_ptr<StreamCodec> createStream(int level, CodecType);
+ explicit ZSTDStreamCodec(int level, CodecType type);
- // "The windowBits parameter is the base two logarithm of the maximum window
- // size (...) The default value is 15 (...) add 16 to decode only the gzip
- // format (the zlib format will return a Z_DATA_ERROR)."
- int windowBits = 15 + (type() == CodecType::GZIP ? 16 : 0);
- int rc = inflateInit2(&stream, windowBits);
- if (rc != Z_OK) {
- throw std::runtime_error(to<std::string>(
- "ZlibCodec: inflateInit error: ", rc, ": ", stream.msg));
- }
+ std::vector<std::string> validPrefixes() const override;
+ bool canUncompress(const IOBuf* data, Optional<uint64_t> uncompressedLength)
+ const override;
- stream.next_in = stream.next_out = nullptr;
- stream.avail_in = stream.avail_out = 0;
- stream.total_in = stream.total_out = 0;
+ private:
+ bool doNeedsUncompressedLength() const override;
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
+ Optional<uint64_t> doGetUncompressedLength(
+ IOBuf const* data,
+ Optional<uint64_t> uncompressedLength) const override;
+
+ void doResetStream() override;
+ bool doCompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) override;
+ bool doUncompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) override;
+
+ void resetCStream();
+ void resetDStream();
+
+ bool tryBlockCompress(ByteRange& input, MutableByteRange& output) const;
+ bool tryBlockUncompress(ByteRange& input, MutableByteRange& output) const;
- bool success = false;
+ int level_;
+ bool needReset_{true};
+ std::unique_ptr<
+ ZSTD_CStream,
+ folly::static_function_deleter<ZSTD_CStream, &zstdFreeCStream>>
+ cstream_{nullptr};
+ std::unique_ptr<
+ ZSTD_DStream,
+ folly::static_function_deleter<ZSTD_DStream, &zstdFreeDStream>>
+ dstream_{nullptr};
+};
- SCOPE_EXIT {
- rc = inflateEnd(&stream);
- // If we're here because of an exception, it's okay if some data
- // got dropped.
- CHECK(rc == Z_OK || (!success && rc == Z_DATA_ERROR))
- << rc << ": " << stream.msg;
- };
+static constexpr uint32_t kZSTDMagicLE = 0xFD2FB528;
- // Max 64MiB in one go
- constexpr uint32_t maxSingleStepLength = uint32_t(64) << 20; // 64MiB
- constexpr uint32_t defaultBufferLength = uint32_t(4) << 20; // 4MiB
+std::vector<std::string> ZSTDStreamCodec::validPrefixes() const {
+ return {prefixToStringLE(kZSTDMagicLE)};
+}
- auto out = addOutputBuffer(
- &stream,
- ((uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- uncompressedLength <= maxSingleStepLength) ?
- uncompressedLength :
- defaultBufferLength));
-
- bool streamEnd = false;
- for (auto& range : *data) {
- if (range.empty()) {
- continue;
- }
+bool ZSTDStreamCodec::canUncompress(const IOBuf* data, Optional<uint64_t>)
+ const {
+ return dataStartsWithLE(data, kZSTDMagicLE);
+}
- stream.next_in = const_cast<uint8_t*>(range.data());
- stream.avail_in = range.size();
+std::unique_ptr<Codec> ZSTDStreamCodec::createCodec(int level, CodecType type) {
+ return make_unique<ZSTDStreamCodec>(level, type);
+}
- while (stream.avail_in != 0) {
- if (streamEnd) {
- throw std::runtime_error(to<std::string>(
- "ZlibCodec: junk after end of data"));
- }
+std::unique_ptr<StreamCodec> ZSTDStreamCodec::createStream(
+ int level,
+ CodecType type) {
+ return make_unique<ZSTDStreamCodec>(level, type);
+}
+
+ZSTDStreamCodec::ZSTDStreamCodec(int level, CodecType type)
+ : StreamCodec(type) {
+ DCHECK(type == CodecType::ZSTD);
+ switch (level) {
+ case COMPRESSION_LEVEL_FASTEST:
+ level = 1;
+ break;
+ case COMPRESSION_LEVEL_DEFAULT:
+ level = 1;
+ break;
+ case COMPRESSION_LEVEL_BEST:
+ level = 19;
+ break;
+ }
+ if (level < 1 || level > ZSTD_maxCLevel()) {
+ throw std::invalid_argument(
+ to<std::string>("ZSTD: invalid level: ", level));
+ }
+ level_ = level;
+}
+
+bool ZSTDStreamCodec::doNeedsUncompressedLength() const {
+ return false;
+}
+
+uint64_t ZSTDStreamCodec::doMaxCompressedLength(
+ uint64_t uncompressedLength) const {
+ return ZSTD_compressBound(uncompressedLength);
+}
+
+void zstdThrowIfError(size_t rc) {
+ if (!ZSTD_isError(rc)) {
+ return;
+ }
+ throw std::runtime_error(
+ to<std::string>("ZSTD returned an error: ", ZSTD_getErrorName(rc)));
+}
- streamEnd = doInflate(&stream, out.get(), defaultBufferLength);
+Optional<uint64_t> ZSTDStreamCodec::doGetUncompressedLength(
+ IOBuf const* data,
+ Optional<uint64_t> uncompressedLength) const {
+ // Read decompressed size from frame if available in first IOBuf.
+ auto const decompressedSize =
+ ZSTD_getDecompressedSize(data->data(), data->length());
+ if (decompressedSize != 0) {
+ if (uncompressedLength && *uncompressedLength != decompressedSize) {
+ throw std::runtime_error("ZSTD: invalid uncompressed length");
}
+ uncompressedLength = decompressedSize;
}
+ return uncompressedLength;
+}
+
+void ZSTDStreamCodec::doResetStream() {
+ needReset_ = true;
+}
- while (!streamEnd) {
- streamEnd = doInflate(&stream, out.get(), defaultBufferLength);
+bool ZSTDStreamCodec::tryBlockCompress(
+ ByteRange& input,
+ MutableByteRange& output) const {
+ DCHECK(needReset_);
+ // We need to know that we have enough output space to use block compression
+ if (output.size() < ZSTD_compressBound(input.size())) {
+ return false;
}
+ size_t const length = ZSTD_compress(
+ output.data(), output.size(), input.data(), input.size(), level_);
+ zstdThrowIfError(length);
+ input.uncheckedAdvance(input.size());
+ output.uncheckedAdvance(length);
+ return true;
+}
- out->prev()->trimEnd(stream.avail_out);
+void ZSTDStreamCodec::resetCStream() {
+ if (!cstream_) {
+ cstream_.reset(ZSTD_createCStream());
+ if (!cstream_) {
+ throw std::bad_alloc{};
+ }
+ }
+ // Advanced API usage works for all supported versions of zstd.
+ // Required to set contentSizeFlag.
+ auto params = ZSTD_getParams(level_, uncompressedLength().value_or(0), 0);
+ params.fParams.contentSizeFlag = uncompressedLength().hasValue();
+ zstdThrowIfError(ZSTD_initCStream_advanced(
+ cstream_.get(), nullptr, 0, params, uncompressedLength().value_or(0)));
+}
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- uncompressedLength != stream.total_out) {
- throw std::runtime_error(to<std::string>(
- "ZlibCodec: invalid uncompressed length"));
+bool ZSTDStreamCodec::doCompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) {
+ if (needReset_) {
+ // If we are given all the input in one chunk try to use block compression
+ if (flushOp == StreamCodec::FlushOp::END &&
+ tryBlockCompress(input, output)) {
+ return true;
+ }
+ resetCStream();
+ needReset_ = false;
+ }
+ ZSTD_inBuffer in = {input.data(), input.size(), 0};
+ ZSTD_outBuffer out = {output.data(), output.size(), 0};
+ SCOPE_EXIT {
+ input.uncheckedAdvance(in.pos);
+ output.uncheckedAdvance(out.pos);
+ };
+ if (flushOp == StreamCodec::FlushOp::NONE || !input.empty()) {
+ zstdThrowIfError(ZSTD_compressStream(cstream_.get(), &out, &in));
}
+ if (in.pos == in.size && flushOp != StreamCodec::FlushOp::NONE) {
+ size_t rc;
+ switch (flushOp) {
+ case StreamCodec::FlushOp::FLUSH:
+ rc = ZSTD_flushStream(cstream_.get(), &out);
+ break;
+ case StreamCodec::FlushOp::END:
+ rc = ZSTD_endStream(cstream_.get(), &out);
+ break;
+ default:
+ throw std::invalid_argument("ZSTD: invalid FlushOp");
+ }
+ zstdThrowIfError(rc);
+ if (rc == 0) {
+ return true;
+ }
+ }
+ return false;
+}
- success = true; // we survived
+bool ZSTDStreamCodec::tryBlockUncompress(
+ ByteRange& input,
+ MutableByteRange& output) const {
+ DCHECK(needReset_);
+#if ZSTD_VERSION_NUMBER < 10104
+ // We require ZSTD_findFrameCompressedSize() to perform this optimization.
+ return false;
+#else
+ // We need to know the uncompressed length and have enough output space.
+ if (!uncompressedLength() || output.size() < *uncompressedLength()) {
+ return false;
+ }
+ size_t const compressedLength =
+ ZSTD_findFrameCompressedSize(input.data(), input.size());
+ zstdThrowIfError(compressedLength);
+ size_t const length = ZSTD_decompress(
+ output.data(), *uncompressedLength(), input.data(), compressedLength);
+ zstdThrowIfError(length);
+ if (length != *uncompressedLength()) {
+ throw std::runtime_error("ZSTDStreamCodec: Incorrect uncompressed length");
+ }
+ input.uncheckedAdvance(compressedLength);
+ output.uncheckedAdvance(length);
+ return true;
+#endif
+}
- return out;
+void ZSTDStreamCodec::resetDStream() {
+ if (!dstream_) {
+ dstream_.reset(ZSTD_createDStream());
+ if (!dstream_) {
+ throw std::bad_alloc{};
+ }
+ }
+ zstdThrowIfError(ZSTD_initDStream(dstream_.get()));
}
-#endif // FOLLY_HAVE_LIBZ
+bool ZSTDStreamCodec::doUncompressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) {
+ if (needReset_) {
+ // If we are given all the input in one chunk try to use block uncompression
+ if (flushOp == StreamCodec::FlushOp::END &&
+ tryBlockUncompress(input, output)) {
+ return true;
+ }
+ resetDStream();
+ needReset_ = false;
+ }
+ ZSTD_inBuffer in = {input.data(), input.size(), 0};
+ ZSTD_outBuffer out = {output.data(), output.size(), 0};
+ SCOPE_EXIT {
+ input.uncheckedAdvance(in.pos);
+ output.uncheckedAdvance(out.pos);
+ };
+ size_t const rc = ZSTD_decompressStream(dstream_.get(), &out, &in);
+ zstdThrowIfError(rc);
+ return rc == 0;
+}
-#if FOLLY_HAVE_LIBLZMA
+#endif // FOLLY_HAVE_LIBZSTD
-/**
- * LZMA2 compression
- */
-class LZMA2Codec final : public Codec {
+#if FOLLY_HAVE_LIBBZ2
+
+class Bzip2Codec final : public Codec {
public:
static std::unique_ptr<Codec> create(int level, CodecType type);
- explicit LZMA2Codec(int level, CodecType type);
+ explicit Bzip2Codec(int level, CodecType type);
- private:
- bool doNeedsUncompressedLength() const override;
- uint64_t doMaxUncompressedLength() const override;
+ std::vector<std::string> validPrefixes() const override;
+ bool canUncompress(IOBuf const* data, Optional<uint64_t> uncompressedLength)
+ const override;
- bool encodeSize() const { return type() == CodecType::LZMA2_VARINT_SIZE; }
-
- std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
+ private:
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
+ std::unique_ptr<IOBuf> doCompress(IOBuf const* data) override;
std::unique_ptr<IOBuf> doUncompress(
- const IOBuf* data,
- uint64_t uncompressedLength) override;
-
- std::unique_ptr<IOBuf> addOutputBuffer(lzma_stream* stream, size_t length);
- bool doInflate(lzma_stream* stream, IOBuf* head, size_t bufferLength);
+ IOBuf const* data,
+ Optional<uint64_t> uncompressedLength) override;
int level_;
};
-std::unique_ptr<Codec> LZMA2Codec::create(int level, CodecType type) {
- return make_unique<LZMA2Codec>(level, type);
+/* static */ std::unique_ptr<Codec> Bzip2Codec::create(
+ int level,
+ CodecType type) {
+ return std::make_unique<Bzip2Codec>(level, type);
}
-LZMA2Codec::LZMA2Codec(int level, CodecType type) : Codec(type) {
- DCHECK(type == CodecType::LZMA2 || type == CodecType::LZMA2_VARINT_SIZE);
+Bzip2Codec::Bzip2Codec(int level, CodecType type) : Codec(type) {
+ DCHECK(type == CodecType::BZIP2);
switch (level) {
- case COMPRESSION_LEVEL_FASTEST:
- level = 0;
- break;
- case COMPRESSION_LEVEL_DEFAULT:
- level = LZMA_PRESET_DEFAULT;
- break;
- case COMPRESSION_LEVEL_BEST:
- level = 9;
- break;
+ case COMPRESSION_LEVEL_FASTEST:
+ level = 1;
+ break;
+ case COMPRESSION_LEVEL_DEFAULT:
+ level = 9;
+ break;
+ case COMPRESSION_LEVEL_BEST:
+ level = 9;
+ break;
}
- if (level < 0 || level > 9) {
- throw std::invalid_argument(to<std::string>(
- "LZMA2Codec: invalid level: ", level));
+ if (level < 1 || level > 9) {
+ throw std::invalid_argument(
+ to<std::string>("Bzip2: invalid level: ", level));
}
level_ = level;
}
-bool LZMA2Codec::doNeedsUncompressedLength() const {
- return !encodeSize();
+static uint32_t constexpr kBzip2MagicLE = 0x685a42;
+static uint64_t constexpr kBzip2MagicBytes = 3;
+
+std::vector<std::string> Bzip2Codec::validPrefixes() const {
+ return {prefixToStringLE(kBzip2MagicLE, kBzip2MagicBytes)};
}
-uint64_t LZMA2Codec::doMaxUncompressedLength() const {
- // From lzma/base.h: "Stream is roughly 8 EiB (2^63 bytes)"
- return uint64_t(1) << 63;
+bool Bzip2Codec::canUncompress(IOBuf const* data, Optional<uint64_t>) const {
+ return dataStartsWithLE(data, kBzip2MagicLE, kBzip2MagicBytes);
+}
+
+uint64_t Bzip2Codec::doMaxCompressedLength(uint64_t uncompressedLength) const {
+ // http://www.bzip.org/1.0.5/bzip2-manual-1.0.5.html#bzbufftobuffcompress
+ // To guarantee that the compressed data will fit in its buffer, allocate an
+ // output buffer of size 1% larger than the uncompressed data, plus six
+ // hundred extra bytes.
+ return uncompressedLength + uncompressedLength / 100 + 600;
+}
+
+static bz_stream createBzStream() {
+ bz_stream stream;
+ stream.bzalloc = nullptr;
+ stream.bzfree = nullptr;
+ stream.opaque = nullptr;
+ stream.next_in = stream.next_out = nullptr;
+ stream.avail_in = stream.avail_out = 0;
+ return stream;
}
-std::unique_ptr<IOBuf> LZMA2Codec::addOutputBuffer(
- lzma_stream* stream,
- size_t length) {
+// Throws on error condition, otherwise returns the code.
+static int bzCheck(int const rc) {
+ switch (rc) {
+ case BZ_OK:
+ case BZ_RUN_OK:
+ case BZ_FLUSH_OK:
+ case BZ_FINISH_OK:
+ case BZ_STREAM_END:
+ return rc;
+ default:
+ throw std::runtime_error(to<std::string>("Bzip2 error: ", rc));
+ }
+}
- CHECK_EQ(stream->avail_out, 0);
+static std::unique_ptr<IOBuf> addOutputBuffer(
+ bz_stream* stream,
+ uint64_t const bufferLength) {
+ DCHECK_LE(bufferLength, std::numeric_limits<unsigned>::max());
+ DCHECK_EQ(stream->avail_out, 0);
- auto buf = IOBuf::create(length);
- buf->append(length);
+ auto buf = IOBuf::create(bufferLength);
+ buf->append(buf->capacity());
- stream->next_out = buf->writableData();
+ stream->next_out = reinterpret_cast<char*>(buf->writableData());
stream->avail_out = buf->length();
return buf;
}
-std::unique_ptr<IOBuf> LZMA2Codec::doCompress(const IOBuf* data) {
- lzma_ret rc;
- lzma_stream stream = LZMA_STREAM_INIT;
-
- rc = lzma_easy_encoder(&stream, level_, LZMA_CHECK_NONE);
- if (rc != LZMA_OK) {
- throw std::runtime_error(folly::to<std::string>(
- "LZMA2Codec: lzma_easy_encoder error: ", rc));
- }
-
- SCOPE_EXIT { lzma_end(&stream); };
-
- uint64_t uncompressedLength = data->computeChainDataLength();
- uint64_t maxCompressedLength = lzma_stream_buffer_bound(uncompressedLength);
+std::unique_ptr<IOBuf> Bzip2Codec::doCompress(IOBuf const* data) {
+ bz_stream stream = createBzStream();
+ bzCheck(BZ2_bzCompressInit(&stream, level_, 0, 0));
+ SCOPE_EXIT {
+ bzCheck(BZ2_bzCompressEnd(&stream));
+ };
- // Max 64MiB in one go
- constexpr uint32_t maxSingleStepLength = uint32_t(64) << 20; // 64MiB
- constexpr uint32_t defaultBufferLength = uint32_t(4) << 20; // 4MiB
+ uint64_t const uncompressedLength = data->computeChainDataLength();
+ uint64_t const maxCompressedLen = maxCompressedLength(uncompressedLength);
+ uint64_t constexpr kMaxSingleStepLength = uint64_t(64) << 20; // 64 MiB
+ uint64_t constexpr kDefaultBufferLength = uint64_t(4) << 20;
auto out = addOutputBuffer(
- &stream,
- (maxCompressedLength <= maxSingleStepLength ?
- maxCompressedLength :
- defaultBufferLength));
-
- if (encodeSize()) {
- auto size = IOBuf::createCombined(kMaxVarintLength64);
- encodeVarintToIOBuf(uncompressedLength, size.get());
- size->appendChain(std::move(out));
- out = std::move(size);
- }
-
- for (auto& range : *data) {
- if (range.empty()) {
- continue;
- }
+ &stream,
+ maxCompressedLen <= kMaxSingleStepLength ? maxCompressedLen
+ : kDefaultBufferLength);
- stream.next_in = const_cast<uint8_t*>(range.data());
- stream.avail_in = range.size();
+ for (auto range : *data) {
+ while (!range.empty()) {
+ auto const inSize = std::min<size_t>(range.size(), kMaxSingleStepLength);
+ stream.next_in =
+ const_cast<char*>(reinterpret_cast<char const*>(range.data()));
+ stream.avail_in = inSize;
- while (stream.avail_in != 0) {
if (stream.avail_out == 0) {
- out->prependChain(addOutputBuffer(&stream, defaultBufferLength));
+ out->prependChain(addOutputBuffer(&stream, kDefaultBufferLength));
}
- rc = lzma_code(&stream, LZMA_RUN);
-
- if (rc != LZMA_OK) {
- throw std::runtime_error(folly::to<std::string>(
- "LZMA2Codec: lzma_code error: ", rc));
- }
+ bzCheck(BZ2_bzCompress(&stream, BZ_RUN));
+ range.uncheckedAdvance(inSize - stream.avail_in);
}
}
-
do {
if (stream.avail_out == 0) {
- out->prependChain(addOutputBuffer(&stream, defaultBufferLength));
+ out->prependChain(addOutputBuffer(&stream, kDefaultBufferLength));
}
-
- rc = lzma_code(&stream, LZMA_FINISH);
- } while (rc == LZMA_OK);
-
- if (rc != LZMA_STREAM_END) {
- throw std::runtime_error(folly::to<std::string>(
- "LZMA2Codec: lzma_code ended with error: ", rc));
- }
+ } while (bzCheck(BZ2_bzCompress(&stream, BZ_FINISH)) != BZ_STREAM_END);
out->prev()->trimEnd(stream.avail_out);
return out;
}
-bool LZMA2Codec::doInflate(lzma_stream* stream,
- IOBuf* head,
- size_t bufferLength) {
- if (stream->avail_out == 0) {
- head->prependChain(addOutputBuffer(stream, bufferLength));
- }
-
- lzma_ret rc = lzma_code(stream, LZMA_RUN);
-
- switch (rc) {
- case LZMA_OK:
- break;
- case LZMA_STREAM_END:
- return true;
- default:
- throw std::runtime_error(to<std::string>(
- "LZMA2Codec: lzma_code error: ", rc));
- }
-
- return false;
-}
-
-std::unique_ptr<IOBuf> LZMA2Codec::doUncompress(const IOBuf* data,
- uint64_t uncompressedLength) {
- lzma_ret rc;
- lzma_stream stream = LZMA_STREAM_INIT;
-
- rc = lzma_auto_decoder(&stream, std::numeric_limits<uint64_t>::max(), 0);
- if (rc != LZMA_OK) {
- throw std::runtime_error(folly::to<std::string>(
- "LZMA2Codec: lzma_auto_decoder error: ", rc));
- }
-
- SCOPE_EXIT { lzma_end(&stream); };
-
- // Max 64MiB in one go
- constexpr uint32_t maxSingleStepLength = uint32_t(64) << 20; // 64MiB
- constexpr uint32_t defaultBufferLength = uint32_t(4) << 20; // 4MiB
+std::unique_ptr<IOBuf> Bzip2Codec::doUncompress(
+ const IOBuf* data,
+ Optional<uint64_t> uncompressedLength) {
+ bz_stream stream = createBzStream();
+ bzCheck(BZ2_bzDecompressInit(&stream, 0, 0));
+ SCOPE_EXIT {
+ bzCheck(BZ2_bzDecompressEnd(&stream));
+ };
- folly::io::Cursor cursor(data);
- uint64_t actualUncompressedLength;
- if (encodeSize()) {
- actualUncompressedLength = decodeVarintFromCursor(cursor);
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- uncompressedLength != actualUncompressedLength) {
- throw std::runtime_error("LZMA2Codec: invalid uncompressed length");
- }
- } else {
- actualUncompressedLength = uncompressedLength;
- DCHECK_NE(actualUncompressedLength, UNKNOWN_UNCOMPRESSED_LENGTH);
- }
+ uint64_t constexpr kMaxSingleStepLength = uint64_t(64) << 20; // 64 MiB
+ uint64_t const kBlockSize = uint64_t(100) << 10; // 100 KiB
+ uint64_t const kDefaultBufferLength =
+ computeBufferLength(data->computeChainDataLength(), kBlockSize);
auto out = addOutputBuffer(
&stream,
- (actualUncompressedLength <= maxSingleStepLength ?
- actualUncompressedLength :
- defaultBufferLength));
-
- bool streamEnd = false;
- auto buf = cursor.peekBytes();
- while (!buf.empty()) {
- stream.next_in = const_cast<uint8_t*>(buf.data());
- stream.avail_in = buf.size();
-
- while (stream.avail_in != 0) {
- if (streamEnd) {
- throw std::runtime_error(to<std::string>(
- "LZMA2Codec: junk after end of data"));
+ ((uncompressedLength && *uncompressedLength <= kMaxSingleStepLength)
+ ? *uncompressedLength
+ : kDefaultBufferLength));
+
+ int rc = BZ_OK;
+ for (auto range : *data) {
+ while (!range.empty()) {
+ auto const inSize = std::min<size_t>(range.size(), kMaxSingleStepLength);
+ stream.next_in =
+ const_cast<char*>(reinterpret_cast<char const*>(range.data()));
+ stream.avail_in = inSize;
+
+ if (stream.avail_out == 0) {
+ out->prependChain(addOutputBuffer(&stream, kDefaultBufferLength));
}
- streamEnd = doInflate(&stream, out.get(), defaultBufferLength);
+ rc = bzCheck(BZ2_bzDecompress(&stream));
+ range.uncheckedAdvance(inSize - stream.avail_in);
}
-
- cursor.skip(buf.size());
- buf = cursor.peekBytes();
}
-
- while (!streamEnd) {
- streamEnd = doInflate(&stream, out.get(), defaultBufferLength);
+ while (rc != BZ_STREAM_END) {
+ if (stream.avail_out == 0) {
+ out->prependChain(addOutputBuffer(&stream, kDefaultBufferLength));
+ }
+ size_t const outputSize = stream.avail_out;
+ rc = bzCheck(BZ2_bzDecompress(&stream));
+ if (outputSize == stream.avail_out) {
+ throw std::runtime_error("Bzip2Codec: Truncated input");
+ }
}
out->prev()->trimEnd(stream.avail_out);
- if (actualUncompressedLength != stream.total_out) {
- throw std::runtime_error(to<std::string>(
- "LZMA2Codec: invalid uncompressed length"));
+ uint64_t const totalOut =
+ (uint64_t(stream.total_out_hi32) << 32) + stream.total_out_lo32;
+ if (uncompressedLength && uncompressedLength != totalOut) {
+ throw std::runtime_error("Bzip2 error: Invalid uncompressed length");
}
return out;
}
-#endif // FOLLY_HAVE_LIBLZMA
+#endif // FOLLY_HAVE_LIBBZ2
-#ifdef FOLLY_HAVE_LIBZSTD
+#if FOLLY_HAVE_LIBZ
+
+zlib::Options getZlibOptions(CodecType type) {
+ DCHECK(type == CodecType::GZIP || type == CodecType::ZLIB);
+ return type == CodecType::GZIP ? zlib::defaultGzipOptions()
+ : zlib::defaultZlibOptions();
+}
+
+std::unique_ptr<Codec> getZlibCodec(int level, CodecType type) {
+ return zlib::getCodec(getZlibOptions(type), level);
+}
+
+std::unique_ptr<StreamCodec> getZlibStreamCodec(int level, CodecType type) {
+ return zlib::getStreamCodec(getZlibOptions(type), level);
+}
+
+#endif // FOLLY_HAVE_LIBZ
/**
- * ZSTD compression
+ * Automatic decompression
*/
-class ZSTDCodec final : public Codec {
+class AutomaticCodec final : public Codec {
public:
- static std::unique_ptr<Codec> create(int level, CodecType);
- explicit ZSTDCodec(int level, CodecType type);
+ static std::unique_ptr<Codec> create(
+ std::vector<std::unique_ptr<Codec>> customCodecs);
+ explicit AutomaticCodec(std::vector<std::unique_ptr<Codec>> customCodecs);
+
+ std::vector<std::string> validPrefixes() const override;
+ bool canUncompress(const IOBuf* data, Optional<uint64_t> uncompressedLength)
+ const override;
private:
bool doNeedsUncompressedLength() const override;
- std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
+ uint64_t doMaxUncompressedLength() const override;
+
+ uint64_t doMaxCompressedLength(uint64_t) const override {
+ throw std::runtime_error(
+ "AutomaticCodec error: maxCompressedLength() not supported.");
+ }
+ std::unique_ptr<IOBuf> doCompress(const IOBuf*) override {
+ throw std::runtime_error("AutomaticCodec error: compress() not supported.");
+ }
std::unique_ptr<IOBuf> doUncompress(
const IOBuf* data,
- uint64_t uncompressedLength) override;
+ Optional<uint64_t> uncompressedLength) override;
- int level_;
-};
+ void addCodecIfSupported(CodecType type);
-std::unique_ptr<Codec> ZSTDCodec::create(int level, CodecType type) {
- return make_unique<ZSTDCodec>(level, type);
-}
+ // Throws iff the codecs aren't compatible (very slow)
+ void checkCompatibleCodecs() const;
-ZSTDCodec::ZSTDCodec(int level, CodecType type) : Codec(type) {
- DCHECK(type == CodecType::ZSTD);
- switch (level) {
- case COMPRESSION_LEVEL_FASTEST:
- level = 1;
- break;
- case COMPRESSION_LEVEL_DEFAULT:
- level = 1;
- break;
- case COMPRESSION_LEVEL_BEST:
- level = 19;
- break;
- }
- if (level < 1 || level > ZSTD_maxCLevel()) {
- throw std::invalid_argument(
- to<std::string>("ZSTD: invalid level: ", level));
+ std::vector<std::unique_ptr<Codec>> codecs_;
+ bool needsUncompressedLength_;
+ uint64_t maxUncompressedLength_;
+};
+
+std::vector<std::string> AutomaticCodec::validPrefixes() const {
+ std::unordered_set<std::string> prefixes;
+ for (const auto& codec : codecs_) {
+ const auto codecPrefixes = codec->validPrefixes();
+ prefixes.insert(codecPrefixes.begin(), codecPrefixes.end());
}
- level_ = level;
+ return std::vector<std::string>{prefixes.begin(), prefixes.end()};
}
-bool ZSTDCodec::doNeedsUncompressedLength() const {
- return false;
+bool AutomaticCodec::canUncompress(
+ const IOBuf* data,
+ Optional<uint64_t> uncompressedLength) const {
+ return std::any_of(
+ codecs_.begin(),
+ codecs_.end(),
+ [data, uncompressedLength](std::unique_ptr<Codec> const& codec) {
+ return codec->canUncompress(data, uncompressedLength);
+ });
}
-void zstdThrowIfError(size_t rc) {
- if (!ZSTD_isError(rc)) {
- return;
+void AutomaticCodec::addCodecIfSupported(CodecType type) {
+ const bool present = std::any_of(
+ codecs_.begin(),
+ codecs_.end(),
+ [&type](std::unique_ptr<Codec> const& codec) {
+ return codec->type() == type;
+ });
+ if (hasCodec(type) && !present) {
+ codecs_.push_back(getCodec(type));
}
- throw std::runtime_error(
- to<std::string>("ZSTD returned an error: ", ZSTD_getErrorName(rc)));
}
-std::unique_ptr<IOBuf> ZSTDCodec::doCompress(const IOBuf* data) {
- // Support earlier versions of the codec (working with a single IOBuf,
- // and using ZSTD_decompress which requires ZSTD frame to contain size,
- // which isn't populated by streaming API).
- if (!data->isChained()) {
- auto out = IOBuf::createCombined(ZSTD_compressBound(data->length()));
- const auto rc = ZSTD_compress(
- out->writableData(),
- out->capacity(),
- data->data(),
- data->length(),
- level_);
- zstdThrowIfError(rc);
- out->append(rc);
- return out;
- }
-
- auto zcs = ZSTD_createCStream();
- SCOPE_EXIT {
- ZSTD_freeCStream(zcs);
- };
-
- auto rc = ZSTD_initCStream(zcs, level_);
- zstdThrowIfError(rc);
+/* static */ std::unique_ptr<Codec> AutomaticCodec::create(
+ std::vector<std::unique_ptr<Codec>> customCodecs) {
+ return std::make_unique<AutomaticCodec>(std::move(customCodecs));
+}
- Cursor cursor(data);
- auto result = IOBuf::createCombined(ZSTD_compressBound(cursor.totalLength()));
-
- ZSTD_outBuffer out;
- out.dst = result->writableTail();
- out.size = result->capacity();
- out.pos = 0;
-
- for (auto buffer = cursor.peekBytes(); !buffer.empty();) {
- ZSTD_inBuffer in;
- in.src = buffer.data();
- in.size = buffer.size();
- for (in.pos = 0; in.pos != in.size;) {
- rc = ZSTD_compressStream(zcs, &out, &in);
- zstdThrowIfError(rc);
- }
- cursor.skip(in.size);
- buffer = cursor.peekBytes();
+AutomaticCodec::AutomaticCodec(std::vector<std::unique_ptr<Codec>> customCodecs)
+ : Codec(CodecType::USER_DEFINED), codecs_(std::move(customCodecs)) {
+ // Fastest -> slowest
+ addCodecIfSupported(CodecType::LZ4_FRAME);
+ addCodecIfSupported(CodecType::ZSTD);
+ addCodecIfSupported(CodecType::ZLIB);
+ addCodecIfSupported(CodecType::GZIP);
+ addCodecIfSupported(CodecType::LZMA2);
+ addCodecIfSupported(CodecType::BZIP2);
+ if (kIsDebug) {
+ checkCompatibleCodecs();
}
-
- rc = ZSTD_endStream(zcs, &out);
- zstdThrowIfError(rc);
- CHECK_EQ(rc, 0);
-
- result->append(out.pos);
- return result;
+ // Check that none of the codes are are null
+ DCHECK(std::none_of(
+ codecs_.begin(), codecs_.end(), [](std::unique_ptr<Codec> const& codec) {
+ return codec == nullptr;
+ }));
+
+ needsUncompressedLength_ = std::any_of(
+ codecs_.begin(), codecs_.end(), [](std::unique_ptr<Codec> const& codec) {
+ return codec->needsUncompressedLength();
+ });
+
+ const auto it = std::max_element(
+ codecs_.begin(),
+ codecs_.end(),
+ [](std::unique_ptr<Codec> const& lhs, std::unique_ptr<Codec> const& rhs) {
+ return lhs->maxUncompressedLength() < rhs->maxUncompressedLength();
+ });
+ DCHECK(it != codecs_.end());
+ maxUncompressedLength_ = (*it)->maxUncompressedLength();
}
-std::unique_ptr<IOBuf> ZSTDCodec::doUncompress(
- const IOBuf* data,
- uint64_t uncompressedLength) {
- auto zds = ZSTD_createDStream();
- SCOPE_EXIT {
- ZSTD_freeDStream(zds);
- };
-
- auto rc = ZSTD_initDStream(zds);
- zstdThrowIfError(rc);
-
- ZSTD_outBuffer out{};
- ZSTD_inBuffer in{};
-
- auto outputSize = ZSTD_DStreamOutSize();
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH) {
- outputSize = uncompressedLength;
- } else {
- auto decompressedSize =
- ZSTD_getDecompressedSize(data->data(), data->length());
- if (decompressedSize != 0 && decompressedSize < outputSize) {
- outputSize = decompressedSize;
+void AutomaticCodec::checkCompatibleCodecs() const {
+ // Keep track of all the possible headers.
+ std::unordered_set<std::string> headers;
+ // The empty header is not allowed.
+ headers.insert("");
+ // Step 1:
+ // Construct a set of headers and check that none of the headers occur twice.
+ // Eliminate edge cases.
+ for (auto&& codec : codecs_) {
+ const auto codecHeaders = codec->validPrefixes();
+ // Codecs without any valid headers are not allowed.
+ if (codecHeaders.empty()) {
+ throw std::invalid_argument{
+ "AutomaticCodec: validPrefixes() must not be empty."};
}
- }
-
- IOBufQueue queue(IOBufQueue::cacheChainLength());
-
- Cursor cursor(data);
- for (rc = 0;;) {
- if (in.pos == in.size) {
- auto buffer = cursor.peekBytes();
- in.src = buffer.data();
- in.size = buffer.size();
- in.pos = 0;
- cursor.skip(in.size);
- if (rc > 1 && in.size == 0) {
- throw std::runtime_error(to<std::string>("ZSTD: incomplete input"));
- }
+ // Insert all the headers for the current codec.
+ const size_t beforeSize = headers.size();
+ headers.insert(codecHeaders.begin(), codecHeaders.end());
+ // Codecs are not compatible if any header occurred twice.
+ if (beforeSize + codecHeaders.size() != headers.size()) {
+ throw std::invalid_argument{
+ "AutomaticCodec: Two valid prefixes collide."};
}
- if (out.pos == out.size) {
- if (out.pos != 0) {
- queue.postallocate(out.pos);
+ }
+ // Step 2:
+ // Check if any strict non-empty prefix of any header is a header.
+ for (const auto& header : headers) {
+ for (size_t i = 1; i < header.size(); ++i) {
+ if (headers.count(header.substr(0, i))) {
+ throw std::invalid_argument{
+ "AutomaticCodec: One valid prefix is a prefix of another valid "
+ "prefix."};
}
- auto buffer = queue.preallocate(outputSize, outputSize);
- out.dst = buffer.first;
- out.size = buffer.second;
- out.pos = 0;
- outputSize = ZSTD_DStreamOutSize();
}
- rc = ZSTD_decompressStream(zds, &out, &in);
- zstdThrowIfError(rc);
- if (rc == 0) {
- break;
- }
- }
- if (out.pos != 0) {
- queue.postallocate(out.pos);
- }
- if (in.pos != in.size || !cursor.isAtEnd()) {
- throw std::runtime_error("ZSTD: junk after end of data");
- }
- if (uncompressedLength != UNKNOWN_UNCOMPRESSED_LENGTH &&
- queue.chainLength() != uncompressedLength) {
- throw std::runtime_error("ZSTD: invalid uncompressed length");
}
+}
- return queue.move();
+bool AutomaticCodec::doNeedsUncompressedLength() const {
+ return needsUncompressedLength_;
}
-#endif // FOLLY_HAVE_LIBZSTD
+uint64_t AutomaticCodec::doMaxUncompressedLength() const {
+ return maxUncompressedLength_;
+}
-} // namespace
+std::unique_ptr<IOBuf> AutomaticCodec::doUncompress(
+ const IOBuf* data,
+ Optional<uint64_t> uncompressedLength) {
+ for (auto&& codec : codecs_) {
+ if (codec->canUncompress(data, uncompressedLength)) {
+ return codec->uncompress(data, uncompressedLength);
+ }
+ }
+ throw std::runtime_error("AutomaticCodec error: Unknown compressed data");
+}
-std::unique_ptr<Codec> getCodec(CodecType type, int level) {
- typedef std::unique_ptr<Codec> (*CodecFactory)(int, CodecType);
+using CodecFactory = std::unique_ptr<Codec> (*)(int, CodecType);
+using StreamCodecFactory = std::unique_ptr<StreamCodec> (*)(int, CodecType);
+struct Factory {
+ CodecFactory codec;
+ StreamCodecFactory stream;
+};
- static CodecFactory codecFactories[
- static_cast<size_t>(CodecType::NUM_CODEC_TYPES)] = {
- nullptr, // USER_DEFINED
- NoCompressionCodec::create,
+constexpr Factory
+ codecFactories[static_cast<size_t>(CodecType::NUM_CODEC_TYPES)] = {
+ {}, // USER_DEFINED
+ {NoCompressionCodec::create, nullptr},
#if FOLLY_HAVE_LIBLZ4
- LZ4Codec::create,
+ {LZ4Codec::create, nullptr},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBSNAPPY
- SnappyCodec::create,
+ {SnappyCodec::create, nullptr},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBZ
- ZlibCodec::create,
+ {getZlibCodec, getZlibStreamCodec},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBLZ4
- LZ4Codec::create,
+ {LZ4Codec::create, nullptr},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBLZMA
- LZMA2Codec::create,
- LZMA2Codec::create,
+ {LZMA2StreamCodec::createCodec, LZMA2StreamCodec::createStream},
+ {LZMA2StreamCodec::createCodec, LZMA2StreamCodec::createStream},
#else
- nullptr,
- nullptr,
+ {},
+ {},
#endif
#if FOLLY_HAVE_LIBZSTD
- ZSTDCodec::create,
+ {ZSTDStreamCodec::createCodec, ZSTDStreamCodec::createStream},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBZ
- ZlibCodec::create,
+ {getZlibCodec, getZlibStreamCodec},
#else
- nullptr,
+ {},
#endif
- };
- size_t idx = static_cast<size_t>(type);
+#if (FOLLY_HAVE_LIBLZ4 && LZ4_VERSION_NUMBER >= 10301)
+ {LZ4FrameCodec::create, nullptr},
+#else
+ {},
+#endif
+
+#if FOLLY_HAVE_LIBBZ2
+ {Bzip2Codec::create, nullptr},
+#else
+ {},
+#endif
+};
+
+Factory const& getFactory(CodecType type) {
+ size_t const idx = static_cast<size_t>(type);
if (idx >= static_cast<size_t>(CodecType::NUM_CODEC_TYPES)) {
- throw std::invalid_argument(to<std::string>(
- "Compression type ", idx, " not supported"));
+ throw std::invalid_argument(
+ to<std::string>("Compression type ", idx, " invalid"));
}
- auto factory = codecFactories[idx];
+ return codecFactories[idx];
+}
+} // namespace
+
+bool hasCodec(CodecType type) {
+ return getFactory(type).codec != nullptr;
+}
+
+std::unique_ptr<Codec> getCodec(CodecType type, int level) {
+ auto const factory = getFactory(type).codec;
if (!factory) {
- throw std::invalid_argument(to<std::string>(
- "Compression type ", idx, " not supported"));
+ throw std::invalid_argument(
+ to<std::string>("Compression type ", type, " not supported"));
}
auto codec = (*factory)(level, type);
- DCHECK_EQ(static_cast<size_t>(codec->type()), idx);
+ DCHECK(codec->type() == type);
return codec;
}
-}} // namespaces
+bool hasStreamCodec(CodecType type) {
+ return getFactory(type).stream != nullptr;
+}
+
+std::unique_ptr<StreamCodec> getStreamCodec(CodecType type, int level) {
+ auto const factory = getFactory(type).stream;
+ if (!factory) {
+ throw std::invalid_argument(
+ to<std::string>("Compression type ", type, " not supported"));
+ }
+ auto codec = (*factory)(level, type);
+ DCHECK(codec->type() == type);
+ return codec;
+}
+
+std::unique_ptr<Codec> getAutoUncompressionCodec(
+ std::vector<std::unique_ptr<Codec>> customCodecs) {
+ return AutomaticCodec::create(std::move(customCodecs));
+}
+} // namespace io
+} // namespace folly