#endif
#if FOLLY_HAVE_LIBZSTD
+#define ZSTD_STATIC_LINKING_ONLY
#include <zstd.h>
#endif
return output;
}
+uint64_t Codec::maxCompressedLength(uint64_t uncompressedLength) const {
+ if (uncompressedLength == 0) {
+ return 0;
+ }
+ return doMaxCompressedLength(uncompressedLength);
+}
+
+Optional<uint64_t> Codec::getUncompressedLength(
+ const folly::IOBuf* data,
+ Optional<uint64_t> uncompressedLength) const {
+ auto const compressedLength = data->computeChainDataLength();
+ if (uncompressedLength == uint64_t(0) || compressedLength == 0) {
+ if (uncompressedLength.value_or(0) != 0 || compressedLength != 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;
+ doResetStream();
+}
+
+bool StreamCodec::compressStream(
+ ByteRange& input,
+ MutableByteRange& output,
+ StreamCodec::FlushOp flushOp) {
+ if (state_ == State::RESET && input.empty()) {
+ if (flushOp == StreamCodec::FlushOp::NONE) {
+ return false;
+ }
+ if (flushOp == StreamCodec::FlushOp::END &&
+ uncompressedLength().value_or(0) != 0) {
+ throw std::runtime_error("Codec: invalid uncompressed length");
+ }
+ return true;
+ }
+ 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;
+ }
+ bool const done = doCompressStream(input, output, flushOp);
+ // 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);
+ bool const done = doUncompressStream(input, output, flushOp);
+ // 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;
+ for (;;) {
+ 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));
+ }
+ bool const done = compressStream(input, output, flushOp);
+ if (done) {
+ DCHECK(input.empty());
+ DCHECK(flushOp == StreamCodec::FlushOp::END);
+ DCHECK_EQ(current->next(), data);
+ break;
+ }
+ }
+ 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;
+ for (;;) {
+ 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));
+ }
+ bool const done = uncompressStream(input, output, flushOp);
+ if (done) {
+ break;
+ }
+ }
+ 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,
};
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)
}
}
+uint64_t NoCompressionCodec::doMaxCompressedLength(
+ uint64_t uncompressedLength) const {
+ return uncompressedLength;
+}
+
std::unique_ptr<IOBuf> NoCompressionCodec::doCompress(
const IOBuf* data) {
return data->clone();
memcpy(&result[0], &prefix, n);
return result;
}
-
-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);
-}
} // namespace
#if FOLLY_HAVE_LIBLZ4
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<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) {
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) {
IOBuf clone;
if (data->isChained()) {
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());
}
public:
static std::unique_ptr<Codec> create(int level, CodecType type);
explicit LZ4FrameCodec(int level, CodecType type);
- ~LZ4FrameCodec();
+ ~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,
/* static */ std::unique_ptr<Codec> LZ4FrameCodec::create(
int level,
CodecType type) {
- return make_unique<LZ4FrameCodec>(level, type);
+ return std::make_unique<LZ4FrameCodec>(level, type);
}
static constexpr uint32_t kLZ4FrameMagicLE = 0x184D2204;
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(
prefs.compressionLevel = level_;
prefs.frameInfo.contentSize = uncompressedLength;
// Compress
- auto buf = IOBuf::create(LZ4F_compressFrameBound(uncompressedLength, &prefs));
+ auto buf = IOBuf::create(maxCompressedLength(uncompressedLength));
const size_t written = lz4FrameThrowOnError(LZ4F_compressFrame(
buf->writableTail(),
buf->tailroom(),
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,
};
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) {
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()));
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,
}
}
+uint64_t ZlibCodec::doMaxCompressedLength(uint64_t uncompressedLength) const {
+ return deflateBound(nullptr, uncompressedLength);
+}
+
std::unique_ptr<Codec> ZlibCodec::create(int level, CodecType type) {
- return make_unique<ZlibCodec>(level, type);
+ return std::make_unique<ZlibCodec>(level, type);
}
ZlibCodec::ZlibCodec(int level, CodecType type) : Codec(type) {
private:
bool doNeedsUncompressedLength() const override;
uint64_t doMaxUncompressedLength() const override;
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
bool encodeSize() const { return type() == CodecType::LZMA2_VARINT_SIZE; }
}
std::unique_ptr<Codec> LZMA2Codec::create(int level, CodecType type) {
- return make_unique<LZMA2Codec>(level, type);
+ return std::make_unique<LZMA2Codec>(level, type);
}
LZMA2Codec::LZMA2Codec(int level, CodecType type) : Codec(type) {
return uint64_t(1) << 63;
}
+uint64_t LZMA2Codec::doMaxCompressedLength(uint64_t uncompressedLength) const {
+ return lzma_stream_buffer_bound(uncompressedLength) +
+ (encodeSize() ? kMaxVarintLength64 : 0);
+}
+
std::unique_ptr<IOBuf> LZMA2Codec::addOutputBuffer(
lzma_stream* stream,
size_t length) {
#ifdef FOLLY_HAVE_LIBZSTD
+namespace {
+void zstdFreeCStream(ZSTD_CStream* zcs) {
+ ZSTD_freeCStream(zcs);
+}
+
+void zstdFreeDStream(ZSTD_DStream* zds) {
+ ZSTD_freeDStream(zds);
+}
+}
+
/**
* ZSTD compression
*/
-class ZSTDCodec final : public Codec {
+class ZSTDStreamCodec final : public StreamCodec {
public:
- static std::unique_ptr<Codec> create(int level, CodecType);
- explicit ZSTDCodec(int level, CodecType type);
+ static std::unique_ptr<Codec> createCodec(int level, CodecType);
+ static std::unique_ptr<StreamCodec> createStream(int level, CodecType);
+ explicit ZSTDStreamCodec(int level, CodecType type);
std::vector<std::string> validPrefixes() const override;
bool canUncompress(const IOBuf* data, Optional<uint64_t> uncompressedLength)
private:
bool doNeedsUncompressedLength() const override;
- std::unique_ptr<IOBuf> doCompress(const IOBuf* data) override;
- std::unique_ptr<IOBuf> doUncompress(
- const IOBuf* data,
- Optional<uint64_t> uncompressedLength) 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;
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};
};
static constexpr uint32_t kZSTDMagicLE = 0xFD2FB528;
-std::vector<std::string> ZSTDCodec::validPrefixes() const {
+std::vector<std::string> ZSTDStreamCodec::validPrefixes() const {
return {prefixToStringLE(kZSTDMagicLE)};
}
-bool ZSTDCodec::canUncompress(const IOBuf* data, Optional<uint64_t>) const {
+bool ZSTDStreamCodec::canUncompress(const IOBuf* data, Optional<uint64_t>)
+ const {
return dataStartsWithLE(data, kZSTDMagicLE);
}
-std::unique_ptr<Codec> ZSTDCodec::create(int level, CodecType type) {
- return make_unique<ZSTDCodec>(level, type);
+std::unique_ptr<Codec> ZSTDStreamCodec::createCodec(int level, CodecType type) {
+ return make_unique<ZSTDStreamCodec>(level, type);
}
-ZSTDCodec::ZSTDCodec(int level, CodecType type) : Codec(type) {
+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_ = level;
}
-bool ZSTDCodec::doNeedsUncompressedLength() const {
+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;
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);
-
- 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);
+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");
}
- cursor.skip(in.size);
- buffer = cursor.peekBytes();
+ uncompressedLength = decompressedSize;
}
+ return uncompressedLength;
+}
- rc = ZSTD_endStream(zcs, &out);
- zstdThrowIfError(rc);
- CHECK_EQ(rc, 0);
-
- result->append(out.pos);
- return result;
+void ZSTDStreamCodec::doResetStream() {
+ needReset_ = true;
}
-static std::unique_ptr<IOBuf> zstdUncompressBuffer(
- const IOBuf* data,
- Optional<uint64_t> uncompressedLength) {
- // Check preconditions
- DCHECK(!data->isChained());
- DCHECK(uncompressedLength.hasValue());
-
- auto uncompressed = IOBuf::create(*uncompressedLength);
- const auto decompressedSize = ZSTD_decompress(
- uncompressed->writableTail(),
- uncompressed->tailroom(),
- data->data(),
- data->length());
- zstdThrowIfError(decompressedSize);
- if (decompressedSize != uncompressedLength) {
- throw std::runtime_error("ZSTD: invalid uncompressed length");
+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;
}
- uncompressed->append(decompressedSize);
- return uncompressed;
+ 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;
}
-static std::unique_ptr<IOBuf> zstdUncompressStream(
- const IOBuf* data,
- Optional<uint64_t> uncompressedLength) {
- auto zds = ZSTD_createDStream();
+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)));
+}
+
+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 {
- ZSTD_freeDStream(zds);
+ input.uncheckedAdvance(in.pos);
+ output.uncheckedAdvance(out.pos);
};
-
- auto rc = ZSTD_initDStream(zds);
- zstdThrowIfError(rc);
-
- ZSTD_outBuffer out{};
- ZSTD_inBuffer in{};
-
- auto outputSize = uncompressedLength.value_or(ZSTD_DStreamOutSize());
-
- 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"));
- }
- }
- if (out.pos == out.size) {
- if (out.pos != 0) {
- queue.postallocate(out.pos);
- }
- auto buffer = queue.preallocate(outputSize, outputSize);
- out.dst = buffer.first;
- out.size = buffer.second;
- out.pos = 0;
- outputSize = ZSTD_DStreamOutSize();
+ 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");
}
- rc = ZSTD_decompressStream(zds, &out, &in);
zstdThrowIfError(rc);
if (rc == 0) {
- break;
+ return true;
}
}
- 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 && queue.chainLength() != *uncompressedLength) {
- throw std::runtime_error("ZSTD: invalid uncompressed length");
- }
+ return false;
+}
- return queue.move();
+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);
+ DCHECK_EQ(length, *uncompressedLength());
+ input.uncheckedAdvance(compressedLength);
+ output.uncheckedAdvance(length);
+ return true;
+#endif
}
-std::unique_ptr<IOBuf> ZSTDCodec::doUncompress(
- const IOBuf* data,
- Optional<uint64_t> uncompressedLength) {
- {
- // Read decompressed size from frame if available in first IOBuf.
- const auto decompressedSize =
- ZSTD_getDecompressedSize(data->data(), data->length());
- if (decompressedSize != 0) {
- if (uncompressedLength && *uncompressedLength != decompressedSize) {
- throw std::runtime_error("ZSTD: invalid uncompressed length");
- }
- uncompressedLength = decompressedSize;
+void ZSTDStreamCodec::resetDStream() {
+ if (!dstream_) {
+ dstream_.reset(ZSTD_createDStream());
+ if (!dstream_) {
+ throw std::bad_alloc{};
}
}
- // Faster to decompress using ZSTD_decompress() if we can.
- if (uncompressedLength && !data->isChained()) {
- return zstdUncompressBuffer(data, uncompressedLength);
+ zstdThrowIfError(ZSTD_initDStream(dstream_.get()));
+}
+
+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;
}
- // Fall back to slower streaming decompression.
- return zstdUncompressStream(data, uncompressedLength);
+ 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;
}
-#endif // FOLLY_HAVE_LIBZSTD
+#endif // FOLLY_HAVE_LIBZSTD
#if FOLLY_HAVE_LIBBZ2
const override;
private:
+ uint64_t doMaxCompressedLength(uint64_t uncompressedLength) const override;
std::unique_ptr<IOBuf> doCompress(IOBuf const* data) override;
std::unique_ptr<IOBuf> doUncompress(
IOBuf const* data,
/* static */ std::unique_ptr<Codec> Bzip2Codec::create(
int level,
CodecType type) {
- return make_unique<Bzip2Codec>(level, type);
+ return std::make_unique<Bzip2Codec>(level, type);
}
Bzip2Codec::Bzip2Codec(int level, CodecType type) : Codec(type) {
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;
}
}
-static uint64_t bzCompressBound(uint64_t const uncompressedLength) {
- // 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 std::unique_ptr<IOBuf> addOutputBuffer(
bz_stream* stream,
uint64_t const bufferLength) {
};
uint64_t const uncompressedLength = data->computeChainDataLength();
- uint64_t const maxCompressedLength = bzCompressBound(uncompressedLength);
+ 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 <= kMaxSingleStepLength ? maxCompressedLength
- : kDefaultBufferLength);
+ maxCompressedLen <= kMaxSingleStepLength ? maxCompressedLen
+ : kDefaultBufferLength);
for (auto range : *data) {
while (!range.empty()) {
bool doNeedsUncompressedLength() const 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.");
}
/* static */ std::unique_ptr<Codec> AutomaticCodec::create(
std::vector<std::unique_ptr<Codec>> customCodecs) {
- return make_unique<AutomaticCodec>(std::move(customCodecs));
+ return std::make_unique<AutomaticCodec>(std::move(customCodecs));
}
AutomaticCodec::AutomaticCodec(std::vector<std::unique_ptr<Codec>> customCodecs)
throw std::runtime_error("AutomaticCodec error: Unknown compressed data");
}
-} // namespace
+using CodecFactory = std::unique_ptr<Codec> (*)(int, CodecType);
+using StreamCodecFactory = std::unique_ptr<StreamCodec> (*)(int, CodecType);
+struct Factory {
+ CodecFactory codec;
+ StreamCodecFactory stream;
+};
-typedef std::unique_ptr<Codec> (*CodecFactory)(int, CodecType);
-static constexpr CodecFactory
+constexpr Factory
codecFactories[static_cast<size_t>(CodecType::NUM_CODEC_TYPES)] = {
- nullptr, // USER_DEFINED
- NoCompressionCodec::create,
+ {}, // 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,
+ {ZlibCodec::create, nullptr},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBLZ4
- LZ4Codec::create,
+ {LZ4Codec::create, nullptr},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBLZMA
- LZMA2Codec::create,
- LZMA2Codec::create,
+ {LZMA2Codec::create, nullptr},
+ {LZMA2Codec::create, nullptr},
#else
- nullptr,
- nullptr,
+ {},
+ {},
#endif
#if FOLLY_HAVE_LIBZSTD
- ZSTDCodec::create,
+ {ZSTDStreamCodec::createCodec, ZSTDStreamCodec::createStream},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBZ
- ZlibCodec::create,
+ {ZlibCodec::create, nullptr},
#else
- nullptr,
+ {},
#endif
#if (FOLLY_HAVE_LIBLZ4 && LZ4_VERSION_NUMBER >= 10301)
- LZ4FrameCodec::create,
+ {LZ4FrameCodec::create, nullptr},
#else
- nullptr,
+ {},
#endif
#if FOLLY_HAVE_LIBBZ2
- Bzip2Codec::create,
+ {Bzip2Codec::create, nullptr},
#else
- nullptr
+ {},
#endif
};
-bool hasCodec(CodecType type) {
- size_t idx = static_cast<size_t>(type);
+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, " invalid"));
}
- return codecFactories[idx] != nullptr;
+ return codecFactories[idx];
+}
+} // namespace
+
+bool hasCodec(CodecType type) {
+ return getFactory(type).codec != nullptr;
}
std::unique_ptr<Codec> getCodec(CodecType type, int level) {
- size_t idx = static_cast<size_t>(type);
- if (idx >= static_cast<size_t>(CodecType::NUM_CODEC_TYPES)) {
+ auto const factory = getFactory(type).codec;
+ if (!factory) {
throw std::invalid_argument(
- to<std::string>("Compression type ", idx, " invalid"));
+ to<std::string>("Compression type ", type, " not supported"));
}
- auto factory = codecFactories[idx];
+ auto codec = (*factory)(level, type);
+ DCHECK(codec->type() == type);
+ return codec;
+}
+
+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 ", 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;
}