From cf19586cfbe5aa379c8fdb046dc5a8c0fa1cebbb Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= Date: Sat, 6 May 2017 18:35:59 +0300 Subject: [PATCH] common/bitutil: fix decompression corner cases; fuzz, test & bench --- common/bitutil/compress.go | 137 +++++++++++++++++++--------- common/bitutil/compress_fuzz.go | 56 ++++++++++++ common/bitutil/compress_test.go | 152 ++++++++++++++++++++++++++++++++ 3 files changed, 302 insertions(+), 43 deletions(-) create mode 100644 common/bitutil/compress_fuzz.go create mode 100644 common/bitutil/compress_test.go diff --git a/common/bitutil/compress.go b/common/bitutil/compress.go index c6c139ab94..a806c0e8b2 100644 --- a/common/bitutil/compress.go +++ b/common/bitutil/compress.go @@ -16,78 +16,129 @@ package bitutil -/* -The compression algorithm implemented by CompressBytes and DecompressBytes is -optimized for "sparse" input data which contains a lot of zero bytes. Decompression -requires knowledge of the decompressed data length. Compression works as follows: +import "errors" -if data only contains zeroes, - CompressBytes(data) == nil -otherwise if len(data) <= 1, - CompressBytes(data) == data -otherwise: - CompressBytes(data) == append(CompressBytes(nonZeroBits(data)), nonZeroBytes(data)...) -where - nonZeroBits(data) is a bit vector with len(data) bits (MSB first): - nonZeroBits(data)[i/8] && (1 << (7-i%8)) != 0 if data[i] != 0 - len(nonZeroBits(data)) == (len(data)+7)/8 - nonZeroBytes(data) contains the non-zero bytes of data in the same order -*/ +var ( + // ErrMissingData is returned from decompression if the byte referenced by + // the bitset header overflows the input data. + ErrMissingData = errors.New("missing bytes on input") -// CompressBytes compresses the input byte slice + // ErrUnreferencedData is returned from decompression if not all bytes were used + // up from the input data after decompressing it. + ErrUnreferencedData = errors.New("extra bytes on input") + + // ErrExceededTarget is returned from decompression if the bitset header has + // more bits defined than the number of target buffer space available. + ErrExceededTarget = errors.New("target data size exceeded") + + // ErrZeroContent is returned from decompression if a data byte referenced in + // the bitset header is actually a zero byte. + ErrZeroContent = errors.New("zero byte in input content") +) + +// The compression algorithm implemented by CompressBytes and DecompressBytes is +// optimized for sparse input data which contains a lot of zero bytes. Decompression +// requires knowledge of the decompressed data length. +// +// Compression works as follows: +// +// if data only contains zeroes, +// CompressBytes(data) == nil +// otherwise if len(data) <= 1, +// CompressBytes(data) == data +// otherwise: +// CompressBytes(data) == append(CompressBytes(nonZeroBitset(data)), nonZeroBytes(data)...) +// where +// nonZeroBitset(data) is a bit vector with len(data) bits (MSB first): +// nonZeroBitset(data)[i/8] && (1 << (7-i%8)) != 0 if data[i] != 0 +// len(nonZeroBitset(data)) == (len(data)+7)/8 +// nonZeroBytes(data) contains the non-zero bytes of data in the same order + +// CompressBytes compresses the input byte slice according to the sparse bitset +// representation algorithm. func CompressBytes(data []byte) []byte { + // Empty slices get compressed to nil if len(data) == 0 { return nil } + // One byte slices compress to nil or retain the single byte if len(data) == 1 { if data[0] == 0 { return nil - } else { - return data } + return data } - - bitsLen := (len(data) + 7) / 8 - nonZeroBits := make([]byte, bitsLen) + // Calculate the bitset of set bytes, and gather the non-zero bytes + nonZeroBitset := make([]byte, (len(data)+7)/8) nonZeroBytes := make([]byte, 0, len(data)) + for i, b := range data { if b != 0 { nonZeroBytes = append(nonZeroBytes, b) - nonZeroBits[i/8] |= 1 << byte(7-i%8) + nonZeroBitset[i/8] |= 1 << byte(7-i%8) } } if len(nonZeroBytes) == 0 { return nil } - return append(CompressBytes(nonZeroBits), nonZeroBytes...) + return append(CompressBytes(nonZeroBitset), nonZeroBytes...) } -// DecompressBytes decompresses data with a known target size. -// In addition to the decompressed output, the function returns the length of -// compressed input data corresponding to the output. The input slice may be longer. -// If the input slice is too short, (nil, -1) is returned. -func DecompressBytes(data []byte, targetLen int) ([]byte, int) { - decomp := make([]byte, targetLen) - if len(data) == 0 { - return decomp, 0 +// DecompressBytes decompresses data with a known target size. In addition to the +// decompressed output, the function returns the length of compressed input data +// corresponding to the output as the input slice may be longer. +func DecompressBytes(data []byte, target int) ([]byte, error) { + out, size, err := decompressBytes(data, target) + if err != nil { + return nil, err } - if targetLen == 1 { - return data[0:1], 1 + if size != len(data) { + return nil, ErrUnreferencedData } + return out, nil +} - bitsLen := (targetLen + 7) / 8 - nonZeroBits, ptr := DecompressBytes(data, bitsLen) - if ptr < 0 { - return nil, -1 +// decompressBytes decompresses data with a known target size. In addition to the +// decompressed output, the function returns the length of compressed input data +// corresponding to the output as the input slice may be longer. +func decompressBytes(data []byte, target int) ([]byte, int, error) { + // Sanity check 0 targets to avoid infinite recursion + if target == 0 { + return nil, 0, nil + } + // Handle the zero and single byte corner cases + decomp := make([]byte, target) + if len(data) == 0 { + return decomp, 0, nil + } + if target == 1 { + decomp[0] = data[0] // copy to avoid referencing the input slice + if data[0] != 0 { + return decomp, 1, nil + } + return decomp, 0, nil + } + // Decompress the bitset of set bytes and distribute the non zero bytes + nonZeroBitset, ptr, err := decompressBytes(data, (target+7)/8) + if err != nil { + return nil, ptr, err } - for i, _ := range decomp { - if nonZeroBits[i/8]&(1<= len(data) { + return nil, 0, ErrMissingData + } + if i >= len(decomp) { + return nil, 0, ErrExceededTarget + } + // Make sure the data is valid and push into the slot + if data[ptr] == 0 { + return nil, 0, ErrZeroContent } decomp[i] = data[ptr] ptr++ } } - return decomp, ptr + return decomp, ptr, nil } diff --git a/common/bitutil/compress_fuzz.go b/common/bitutil/compress_fuzz.go new file mode 100644 index 0000000000..2b7fe29775 --- /dev/null +++ b/common/bitutil/compress_fuzz.go @@ -0,0 +1,56 @@ +// Copyright 2017 The go-ethereum Authors +// This file is part of the go-ethereum library. +// +// The go-ethereum library is free software: you can redistribute it and/or modify +// it under the terms of the GNU Lesser General Public License as published by +// the Free Software Foundation, either version 3 of the License, or +// (at your option) any later version. +// +// The go-ethereum library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public License +// along with the go-ethereum library. If not, see . + +// +build gofuzz + +package bitutil + +import "bytes" + +// Fuzz implements a go-fuzz fuzzer method to test various compression method +// invocations. +func Fuzz(data []byte) int { + if len(data) == 0 { + return -1 + } + if data[0]%2 == 0 { + return fuzzCompress(data[1:]) + } + return fuzzDecompress(data[1:]) +} + +// fuzzCompress implements a go-fuzz fuzzer method to test the bit compression and +// decompression algorithm. +func fuzzCompress(data []byte) int { + proc, _ := DecompressBytes(CompressBytes(data), len(data)) + if !bytes.Equal(data, proc) { + panic("content mismatch") + } + return 0 +} + +// fuzzDecompress implements a go-fuzz fuzzer method to test the bit decompression +// and recompression algorithm. +func fuzzDecompress(data []byte) int { + blob, err := DecompressBytes(data, 1024) + if err != nil { + return 0 + } + if comp := CompressBytes(blob); !bytes.Equal(comp, data) { + panic("content mismatch") + } + return 0 +} diff --git a/common/bitutil/compress_test.go b/common/bitutil/compress_test.go new file mode 100644 index 0000000000..ef38bc7b38 --- /dev/null +++ b/common/bitutil/compress_test.go @@ -0,0 +1,152 @@ +// Copyright 2017 The go-ethereum Authors +// This file is part of the go-ethereum library. +// +// The go-ethereum library is free software: you can redistribute it and/or modify +// it under the terms of the GNU Lesser General Public License as published by +// the Free Software Foundation, either version 3 of the License, or +// (at your option) any later version. +// +// The go-ethereum library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public License +// along with the go-ethereum library. If not, see . + +package bitutil + +import ( + "bytes" + "math/rand" + "testing" + + "github.com/ethereum/go-ethereum/common/hexutil" +) + +// Tests that data compression and decompression works correctly. +func TestCompressCycle(t *testing.T) { + tests := []string{ + // Tests generated by go-fuzz to maximize code coverage + "0x000000000000000000", + "0xef0400", + "0xdf7070533534333636313639343638373532313536346c1bc33339343837313070706336343035336336346c65fefb3930393233383838ac2f65fefb", + "0x7b64000000", + "0x000034000000000000", + "0x0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000f0000000000000000000", + "0x4912385c0e7b64000000", + "0x000034000000000000000000000000000000", + "0x00", + "0x000003e834ff7f0000", + "0x0000", + "0x0000000000000000000000000000000000000000000000000000000000ff00", + "0x895f0c6a020f850c6a020f85f88df88d", + "0xdf7070533534333636313639343638373432313536346c1bc3315aac2f65fefb", + "0x0000000000", + "0xdf70706336346c65fefb", + "0x00006d643634000000", + "0xdf7070533534333636313639343638373532313536346c1bc333393438373130707063363430353639343638373532313536346c1bc333393438336336346c65fe", + } + for i, tt := range tests { + data := hexutil.MustDecode(tt) + + proc, err := DecompressBytes(CompressBytes(data), len(data)) + if err != nil { + t.Errorf("test %d: failed to decompress compressed data: %v", i, err) + continue + } + if !bytes.Equal(data, proc) { + t.Errorf("test %d: compress/decompress mismatch: have %x, want %x", i, proc, data) + } + } +} + +// Tests that data decompression works +func TestDecompress(t *testing.T) { + tests := []struct { + size int + input string + fail error + }{ + {size: 0, input: "0x"}, + + // Crashers generated by go-fuzz + {size: 0, input: "0x0020", fail: ErrUnreferencedData}, + {size: 0, input: "0x30", fail: ErrUnreferencedData}, + {size: 1, input: "0x00", fail: ErrUnreferencedData}, + {size: 2, input: "0x07", fail: ErrMissingData}, + {size: 1024, input: "0x8000", fail: ErrZeroContent}, + + // Tests generated by go-fuzz to maximize code coverage + {size: 29490, input: "0x343137343733323134333839373334323073333930783e3078333930783e70706336346c65303e", fail: ErrMissingData}, + {size: 59395, input: "0x00", fail: ErrUnreferencedData}, + {size: 52574, input: "0x70706336346c65c0de", fail: ErrExceededTarget}, + {size: 42264, input: "0x07", fail: ErrMissingData}, + {size: 52, input: "0xa5045bad48f4", fail: ErrExceededTarget}, + {size: 52574, input: "0xc0de", fail: ErrMissingData}, + {size: 52574, input: "0x"}, + {size: 29490, input: "0x34313734373332313433383937333432307333393078073034333839373334323073333930783e3078333937333432307333393078073061333930783e70706336346c65303e", fail: ErrMissingData}, + {size: 29491, input: "0x3973333930783e30783e", fail: ErrMissingData}, + + {size: 1024, input: "0x808080608080"}, + {size: 1024, input: "0x808470705e3632383337363033313434303137393130306c6580ef46806380635a80"}, + {size: 1024, input: "0x8080808070"}, + {size: 1024, input: "0x808070705e36346c6580ef46806380635a80"}, + {size: 1024, input: "0x80808046802680"}, + {size: 1024, input: "0x4040404035"}, + {size: 1024, input: "0x4040bf3ba2b3f684402d353234373438373934409fe5b1e7ada94ebfd7d0505e27be4035"}, + {size: 1024, input: "0x404040bf3ba2b3f6844035"}, + {size: 1024, input: "0x40402d35323437343837393440bfd7d0505e27be4035"}, + } + for i, tt := range tests { + data := hexutil.MustDecode(tt.input) + + orig, err := DecompressBytes(data, tt.size) + if err != tt.fail { + t.Errorf("test %d: failure mismatch: have %v, want %v", i, err, tt.fail) + } + if err != nil { + continue + } + if comp := CompressBytes(orig); !bytes.Equal(comp, data) { + t.Errorf("test %d: decompress/compress mismatch: have %x, want %x", i, comp, data) + } + } +} + +// Crude benchmark for compressing random slices of bytes. +func BenchmarkCompress1KBVerySparse(b *testing.B) { benchmarkCompress(b, 1024, 0.0001) } +func BenchmarkCompress2KBVerySparse(b *testing.B) { benchmarkCompress(b, 2048, 0.0001) } +func BenchmarkCompress4KBVerySparse(b *testing.B) { benchmarkCompress(b, 4096, 0.0001) } + +func BenchmarkCompress1KBSparse(b *testing.B) { benchmarkCompress(b, 1024, 0.001) } +func BenchmarkCompress2KBSparse(b *testing.B) { benchmarkCompress(b, 2048, 0.001) } +func BenchmarkCompress4KBSparse(b *testing.B) { benchmarkCompress(b, 4096, 0.001) } + +func BenchmarkCompress1KBDense(b *testing.B) { benchmarkCompress(b, 1024, 0.1) } +func BenchmarkCompress2KBDense(b *testing.B) { benchmarkCompress(b, 2048, 0.1) } +func BenchmarkCompress4KBDense(b *testing.B) { benchmarkCompress(b, 4096, 0.1) } + +func BenchmarkCompress1KBSaturated(b *testing.B) { benchmarkCompress(b, 1024, 0.5) } +func BenchmarkCompress2KBSaturated(b *testing.B) { benchmarkCompress(b, 2048, 0.5) } +func BenchmarkCompress4KBSaturated(b *testing.B) { benchmarkCompress(b, 4096, 0.5) } + +func benchmarkCompress(b *testing.B, bytes int, fill float64) { + // Generate a random slice of bytes to compress + random := rand.NewSource(0) // reproducible and comparable + + data := make([]byte, bytes) + bits := int(float64(bytes) * 8 * fill) + + for i := 0; i < bits; i++ { + idx := random.Int63() % int64(len(data)) + bit := uint(random.Int63() % 8) + data[idx] |= 1 << bit + } + // Reset the benchmark and measure compression/decompression + b.ResetTimer() + b.ReportAllocs() + for i := 0; i < b.N; i++ { + DecompressBytes(CompressBytes(data), len(data)) + } +}