diff options
Diffstat (limited to 'applypatch/imgdiff.cpp')
-rw-r--r-- | applypatch/imgdiff.cpp | 651 |
1 files changed, 251 insertions, 400 deletions
diff --git a/applypatch/imgdiff.cpp b/applypatch/imgdiff.cpp index 880265260..59b600713 100644 --- a/applypatch/imgdiff.cpp +++ b/applypatch/imgdiff.cpp @@ -140,11 +140,12 @@ #include <android-base/logging.h> #include <android-base/memory.h> #include <android-base/unique_fd.h> -#include <ziparchive/zip_archive.h> - #include <bsdiff.h> +#include <ziparchive/zip_archive.h> #include <zlib.h> +#include "applypatch/imgdiff_image.h" + using android::base::get_unaligned; static constexpr auto BUFFER_SIZE = 0x8000; @@ -161,107 +162,16 @@ static inline bool Write4(int fd, int32_t value) { return android::base::WriteFully(fd, &value, sizeof(int32_t)); } -class ImageChunk { - public: - static constexpr auto WINDOWBITS = -15; // 32kb window; negative to indicate a raw stream. - static constexpr auto MEMLEVEL = 8; // the default value. - static constexpr auto METHOD = Z_DEFLATED; - static constexpr auto STRATEGY = Z_DEFAULT_STRATEGY; - - ImageChunk(int type, size_t start, const std::vector<uint8_t>* file_content, size_t raw_data_len) - : type_(type), - start_(start), - input_file_ptr_(file_content), - raw_data_len_(raw_data_len), - compress_level_(6), - source_start_(0), - source_len_(0), - source_uncompressed_len_(0) { - CHECK(file_content != nullptr) << "input file container can't be nullptr"; - } - - int GetType() const { - return type_; - } - size_t GetRawDataLength() const { - return raw_data_len_; - } - const std::string& GetEntryName() const { - return entry_name_; - } - - // CHUNK_DEFLATE will return the uncompressed data for diff, while other types will simply return - // the raw data. - const uint8_t * DataForPatch() const; - size_t DataLengthForPatch() const; - - void Dump() const { - printf("type: %d, start: %zu, len: %zu, name: %s\n", type_, start_, DataLengthForPatch(), - entry_name_.c_str()); - } - - void SetSourceInfo(const ImageChunk& other); - void SetEntryName(std::string entryname); - void SetUncompressedData(std::vector<uint8_t> data); - bool SetBonusData(const std::vector<uint8_t>& bonus_data); - - bool operator==(const ImageChunk& other) const; - bool operator!=(const ImageChunk& other) const { - return !(*this == other); - } - - size_t GetHeaderSize(size_t patch_size) const; - // Return the offset of the next patch into the patch data. - size_t WriteHeaderToFd(int fd, const std::vector<uint8_t>& patch, size_t offset) const; - - /* - * Cause a gzip chunk to be treated as a normal chunk (ie, as a blob - * of uninterpreted data). The resulting patch will likely be about - * as big as the target file, but it lets us handle the case of images - * where some gzip chunks are reconstructible but others aren't (by - * treating the ones that aren't as normal chunks). - */ - void ChangeDeflateChunkToNormal(); - bool ChangeChunkToRaw(size_t patch_size); - - /* - * Verify that we can reproduce exactly the same compressed data that - * we started with. Sets the level, method, windowBits, memLevel, and - * strategy fields in the chunk to the encoding parameters needed to - * produce the right output. - */ - bool ReconstructDeflateChunk(); - bool IsAdjacentNormal(const ImageChunk& other) const; - void MergeAdjacentNormal(const ImageChunk& other); - - /* - * Compute a bsdiff patch between |this| and the input source chunks. - * Store the result in the patch_data. - * |bsdiff_cache| can be used to cache the suffix array if the same |src| chunk is used - * repeatedly, pass nullptr if not needed. - */ - bool MakePatch(const ImageChunk& src, std::vector<uint8_t>* patch_data, saidx_t** bsdiff_cache); - - private: - int type_; // CHUNK_NORMAL, CHUNK_DEFLATE, CHUNK_RAW - size_t start_; // offset of chunk in the original input file - const std::vector<uint8_t>* input_file_ptr_; // ptr to the full content of original input file - size_t raw_data_len_; - - // --- for CHUNK_DEFLATE chunks only: --- - std::vector<uint8_t> uncompressed_data_; - std::string entry_name_; // used for zip entries - - // deflate encoder parameters - int compress_level_; - - size_t source_start_; - size_t source_len_; - size_t source_uncompressed_len_; - - const uint8_t* GetRawData() const; - bool TryReconstruction(int level); -}; +ImageChunk::ImageChunk(int type, size_t start, const std::vector<uint8_t>* file_content, + size_t raw_data_len, std::string entry_name) + : type_(type), + start_(start), + input_file_ptr_(file_content), + raw_data_len_(raw_data_len), + compress_level_(6), + entry_name_(std::move(entry_name)) { + CHECK(file_content != nullptr) << "input file container can't be nullptr"; +} const uint8_t* ImageChunk::GetRawData() const { CHECK_LE(start_ + raw_data_len_, input_file_ptr_->size()); @@ -290,20 +200,6 @@ bool ImageChunk::operator==(const ImageChunk& other) const { memcmp(GetRawData(), other.GetRawData(), raw_data_len_) == 0); } -void ImageChunk::SetSourceInfo(const ImageChunk& src) { - source_start_ = src.start_; - if (type_ == CHUNK_NORMAL) { - source_len_ = src.raw_data_len_; - } else if (type_ == CHUNK_DEFLATE) { - source_len_ = src.raw_data_len_; - source_uncompressed_len_ = src.uncompressed_data_.size(); - } -} - -void ImageChunk::SetEntryName(std::string entryname) { - entry_name_ = std::move(entryname); -} - void ImageChunk::SetUncompressedData(std::vector<uint8_t> data) { uncompressed_data_ = std::move(data); } @@ -316,18 +212,6 @@ bool ImageChunk::SetBonusData(const std::vector<uint8_t>& bonus_data) { return true; } -// Convert CHUNK_NORMAL & CHUNK_DEFLATE to CHUNK_RAW if the target size is -// smaller. Also take the header size into account during size comparison. -bool ImageChunk::ChangeChunkToRaw(size_t patch_size) { - if (type_ == CHUNK_RAW) { - return true; - } else if (type_ == CHUNK_NORMAL && (raw_data_len_ <= 160 || raw_data_len_ < patch_size)) { - type_ = CHUNK_RAW; - return true; - } - return false; -} - void ImageChunk::ChangeDeflateChunkToNormal() { if (type_ != CHUNK_DEFLATE) return; type_ = CHUNK_NORMAL; @@ -335,61 +219,6 @@ void ImageChunk::ChangeDeflateChunkToNormal() { uncompressed_data_.clear(); } -// Header size: -// header_type 4 bytes -// CHUNK_NORMAL 8*3 = 24 bytes -// CHUNK_DEFLATE 8*5 + 4*5 = 60 bytes -// CHUNK_RAW 4 bytes + patch_size -size_t ImageChunk::GetHeaderSize(size_t patch_size) const { - switch (type_) { - case CHUNK_NORMAL: - return 4 + 8 * 3; - case CHUNK_DEFLATE: - return 4 + 8 * 5 + 4 * 5; - case CHUNK_RAW: - return 4 + 4 + patch_size; - default: - CHECK(false) << "unexpected chunk type: " << type_; // Should not reach here. - return 0; - } -} - -size_t ImageChunk::WriteHeaderToFd(int fd, const std::vector<uint8_t>& patch, size_t offset) const { - Write4(fd, type_); - switch (type_) { - case CHUNK_NORMAL: - printf("normal (%10zu, %10zu) %10zu\n", start_, raw_data_len_, patch.size()); - Write8(fd, static_cast<int64_t>(source_start_)); - Write8(fd, static_cast<int64_t>(source_len_)); - Write8(fd, static_cast<int64_t>(offset)); - return offset + patch.size(); - case CHUNK_DEFLATE: - printf("deflate (%10zu, %10zu) %10zu %s\n", start_, raw_data_len_, patch.size(), - entry_name_.c_str()); - Write8(fd, static_cast<int64_t>(source_start_)); - Write8(fd, static_cast<int64_t>(source_len_)); - Write8(fd, static_cast<int64_t>(offset)); - Write8(fd, static_cast<int64_t>(source_uncompressed_len_)); - Write8(fd, static_cast<int64_t>(uncompressed_data_.size())); - Write4(fd, compress_level_); - Write4(fd, METHOD); - Write4(fd, WINDOWBITS); - Write4(fd, MEMLEVEL); - Write4(fd, STRATEGY); - return offset + patch.size(); - case CHUNK_RAW: - printf("raw (%10zu, %10zu)\n", start_, raw_data_len_); - Write4(fd, static_cast<int32_t>(patch.size())); - if (!android::base::WriteFully(fd, patch.data(), patch.size())) { - CHECK(false) << "failed to write " << patch.size() <<" bytes patch"; - } - return offset; - default: - CHECK(false) << "unexpected chunk type: " << type_; - return offset; - } -} - bool ImageChunk::IsAdjacentNormal(const ImageChunk& other) const { if (type_ != CHUNK_NORMAL || other.type_ != CHUNK_NORMAL) { return false; @@ -402,15 +231,8 @@ void ImageChunk::MergeAdjacentNormal(const ImageChunk& other) { raw_data_len_ = raw_data_len_ + other.raw_data_len_; } -bool ImageChunk::MakePatch(const ImageChunk& src, std::vector<uint8_t>* patch_data, - saidx_t** bsdiff_cache) { - if (ChangeChunkToRaw(0)) { - size_t patch_size = DataLengthForPatch(); - patch_data->resize(patch_size); - std::copy(DataForPatch(), DataForPatch() + patch_size, patch_data->begin()); - return true; - } - +bool ImageChunk::MakePatch(const ImageChunk& tgt, const ImageChunk& src, + std::vector<uint8_t>* patch_data, saidx_t** bsdiff_cache) { #if defined(__ANDROID__) char ptemp[] = "/data/local/tmp/imgdiff-patch-XXXXXX"; #else @@ -424,8 +246,8 @@ bool ImageChunk::MakePatch(const ImageChunk& src, std::vector<uint8_t>* patch_da } close(fd); - int r = bsdiff::bsdiff(src.DataForPatch(), src.DataLengthForPatch(), DataForPatch(), - DataLengthForPatch(), ptemp, bsdiff_cache); + int r = bsdiff::bsdiff(src.DataForPatch(), src.DataLengthForPatch(), tgt.DataForPatch(), + tgt.DataLengthForPatch(), ptemp, bsdiff_cache); if (r != 0) { printf("bsdiff() failed: %d\n", r); return false; @@ -443,14 +265,7 @@ bool ImageChunk::MakePatch(const ImageChunk& src, std::vector<uint8_t>* patch_da } size_t sz = static_cast<size_t>(st.st_size); - // Change the chunk type to raw if the patch takes less space that way. - if (ChangeChunkToRaw(sz)) { - unlink(ptemp); - size_t patch_size = DataLengthForPatch(); - patch_data->resize(patch_size); - std::copy(DataForPatch(), DataForPatch() + patch_size, patch_data->begin()); - return true; - } + patch_data->resize(sz); if (!android::base::ReadFully(patch_fd, patch_data->data(), sz)) { printf("failed to read \"%s\" %s\n", ptemp, strerror(errno)); @@ -459,7 +274,6 @@ bool ImageChunk::MakePatch(const ImageChunk& src, std::vector<uint8_t>* patch_da } unlink(ptemp); - SetSourceInfo(src); return true; } @@ -470,8 +284,8 @@ bool ImageChunk::ReconstructDeflateChunk() { return false; } - // We only check two combinations of encoder parameters: level 6 - // (the default) and level 9 (the maximum). + // We only check two combinations of encoder parameters: level 6 (the default) and level 9 + // (the maximum). for (int level = 6; level <= 9; level += 3) { if (TryReconstruction(level)) { compress_level_ = level; @@ -483,10 +297,9 @@ bool ImageChunk::ReconstructDeflateChunk() { } /* - * Takes the uncompressed data stored in the chunk, compresses it - * using the zlib parameters stored in the chunk, and checks that it - * matches exactly the compressed data we started with (also stored in - * the chunk). + * Takes the uncompressed data stored in the chunk, compresses it using the zlib parameters stored + * in the chunk, and checks that it matches exactly the compressed data we started with (also + * stored in the chunk). */ bool ImageChunk::TryReconstruction(int level) { z_stream strm; @@ -529,55 +342,136 @@ bool ImageChunk::TryReconstruction(int level) { return true; } -// Interface for zip_mode and image_mode images. We initialize the image from an input file and -// split the file content into a list of image chunks. -class Image { - public: - explicit Image(bool is_source) : is_source_(is_source) {} - - virtual ~Image() {} - - // Create a list of image chunks from input file. - virtual bool Initialize(const std::string& filename) = 0; +PatchChunk::PatchChunk(const ImageChunk& tgt, const ImageChunk& src, std::vector<uint8_t> data) + : type_(tgt.GetType()), + source_start_(src.GetStartOffset()), + source_len_(src.GetRawDataLength()), + source_uncompressed_len_(src.DataLengthForPatch()), + target_start_(tgt.GetStartOffset()), + target_len_(tgt.GetRawDataLength()), + target_uncompressed_len_(tgt.DataLengthForPatch()), + target_compress_level_(tgt.GetCompressLevel()), + data_(std::move(data)) {} + +// Construct a CHUNK_RAW patch from the target data directly. +PatchChunk::PatchChunk(const ImageChunk& tgt) + : type_(CHUNK_RAW), + source_start_(0), + source_len_(0), + source_uncompressed_len_(0), + target_start_(tgt.GetStartOffset()), + target_len_(tgt.GetRawDataLength()), + target_uncompressed_len_(tgt.DataLengthForPatch()), + target_compress_level_(tgt.GetCompressLevel()), + data_(tgt.DataForPatch(), tgt.DataForPatch() + tgt.DataLengthForPatch()) {} + +// Return true if raw data is smaller than the patch size. +bool PatchChunk::RawDataIsSmaller(const ImageChunk& tgt, size_t patch_size) { + size_t target_len = tgt.GetRawDataLength(); + return (tgt.GetType() == CHUNK_NORMAL && (target_len <= 160 || target_len < patch_size)); +} - // Look for runs of adjacent normal chunks and compress them down into a single chunk. (Such - // runs can be produced when deflate chunks are changed to normal chunks.) - void MergeAdjacentNormalChunks(); +// Header size: +// header_type 4 bytes +// CHUNK_NORMAL 8*3 = 24 bytes +// CHUNK_DEFLATE 8*5 + 4*5 = 60 bytes +// CHUNK_RAW 4 bytes + patch_size +size_t PatchChunk::GetHeaderSize() const { + switch (type_) { + case CHUNK_NORMAL: + return 4 + 8 * 3; + case CHUNK_DEFLATE: + return 4 + 8 * 5 + 4 * 5; + case CHUNK_RAW: + return 4 + 4 + data_.size(); + default: + CHECK(false) << "unexpected chunk type: " << type_; // Should not reach here. + return 0; + } +} - // In zip mode, find the matching deflate source chunk by entry name. Search for normal chunks - // also if |find_normal| is true. - ImageChunk* FindChunkByName(const std::string& name, bool find_normal = false); +// Return the offset of the next patch into the patch data. +size_t PatchChunk::WriteHeaderToFd(int fd, size_t offset) const { + Write4(fd, type_); + switch (type_) { + case CHUNK_NORMAL: + printf("normal (%10zu, %10zu) %10zu\n", target_start_, target_len_, data_.size()); + Write8(fd, static_cast<int64_t>(source_start_)); + Write8(fd, static_cast<int64_t>(source_len_)); + Write8(fd, static_cast<int64_t>(offset)); + return offset + data_.size(); + case CHUNK_DEFLATE: + printf("deflate (%10zu, %10zu) %10zu\n", target_start_, target_len_, data_.size()); + Write8(fd, static_cast<int64_t>(source_start_)); + Write8(fd, static_cast<int64_t>(source_len_)); + Write8(fd, static_cast<int64_t>(offset)); + Write8(fd, static_cast<int64_t>(source_uncompressed_len_)); + Write8(fd, static_cast<int64_t>(target_uncompressed_len_)); + Write4(fd, target_compress_level_); + Write4(fd, ImageChunk::METHOD); + Write4(fd, ImageChunk::WINDOWBITS); + Write4(fd, ImageChunk::MEMLEVEL); + Write4(fd, ImageChunk::STRATEGY); + return offset + data_.size(); + case CHUNK_RAW: + printf("raw (%10zu, %10zu)\n", target_start_, target_len_); + Write4(fd, static_cast<int32_t>(data_.size())); + if (!android::base::WriteFully(fd, data_.data(), data_.size())) { + CHECK(false) << "failed to write " << data_.size() << " bytes patch"; + } + return offset; + default: + CHECK(false) << "unexpected chunk type: " << type_; + return offset; + } +} - // Write the contents of |patch_data| to |patch_fd|. - bool WritePatchDataToFd(const std::vector<std::vector<uint8_t>>& patch_data, int patch_fd) const; +// Write the contents of |patch_chunks| to |patch_fd|. +bool PatchChunk::WritePatchDataToFd(const std::vector<PatchChunk>& patch_chunks, int patch_fd) { + // Figure out how big the imgdiff file header is going to be, so that we can correctly compute + // the offset of each bsdiff patch within the file. + size_t total_header_size = 12; + for (const auto& patch : patch_chunks) { + total_header_size += patch.GetHeaderSize(); + } - void DumpChunks() const; + size_t offset = total_header_size; - // Non const iterators to access the stored ImageChunks. - std::vector<ImageChunk>::iterator begin() { - return chunks_.begin(); + // Write out the headers. + if (!android::base::WriteStringToFd("IMGDIFF2", patch_fd)) { + printf("failed to write \"IMGDIFF2\": %s\n", strerror(errno)); + return false; } - std::vector<ImageChunk>::iterator end() { - return chunks_.end(); - } - // Return a pointer to the ith ImageChunk. - ImageChunk* Get(size_t i) { - CHECK_LT(i, chunks_.size()); - return &chunks_[i]; + Write4(patch_fd, static_cast<int32_t>(patch_chunks.size())); + for (size_t i = 0; i < patch_chunks.size(); ++i) { + printf("chunk %zu: ", i); + offset = patch_chunks[i].WriteHeaderToFd(patch_fd, offset); } - size_t NumOfChunks() const { - return chunks_.size(); + // Append each chunk's bsdiff patch, in order. + for (const auto& patch : patch_chunks) { + if (patch.type_ == CHUNK_RAW) { + continue; + } + if (!android::base::WriteFully(patch_fd, patch.data_.data(), patch.data_.size())) { + printf("failed to write %zu bytes patch to patch_fd\n", patch.data_.size()); + return false; + } } - protected: - bool ReadFile(const std::string& filename, std::vector<uint8_t>* file_content); + return true; +} + +ImageChunk& Image::operator[](size_t i) { + CHECK_LT(i, chunks_.size()); + return chunks_[i]; +} - bool is_source_; // True if it's for source chunks. - std::vector<ImageChunk> chunks_; // Internal storage of ImageChunk. - std::vector<uint8_t> file_content_; // Store the whole input file in memory. -}; +const ImageChunk& Image::operator[](size_t i) const { + CHECK_LT(i, chunks_.size()); + return chunks_[i]; +} void Image::MergeAdjacentNormalChunks() { size_t merged_last = 0, cur = 0; @@ -601,54 +495,6 @@ void Image::MergeAdjacentNormalChunks() { } } -ImageChunk* Image::FindChunkByName(const std::string& name, bool find_normal) { - if (name.empty()) { - return nullptr; - } - for (auto& chunk : chunks_) { - if ((chunk.GetType() == CHUNK_DEFLATE || find_normal) && chunk.GetEntryName() == name) { - return &chunk; - } - } - return nullptr; -} - -bool Image::WritePatchDataToFd(const std::vector<std::vector<uint8_t>>& patch_data, - int patch_fd) const { - // Figure out how big the imgdiff file header is going to be, so that we can correctly compute - // the offset of each bsdiff patch within the file. - CHECK_EQ(chunks_.size(), patch_data.size()); - size_t total_header_size = 12; - for (size_t i = 0; i < chunks_.size(); ++i) { - total_header_size += chunks_[i].GetHeaderSize(patch_data[i].size()); - } - - size_t offset = total_header_size; - - // Write out the headers. - if (!android::base::WriteStringToFd("IMGDIFF2", patch_fd)) { - printf("failed to write \"IMGDIFF2\": %s\n", strerror(errno)); - return false; - } - Write4(patch_fd, static_cast<int32_t>(chunks_.size())); - for (size_t i = 0; i < chunks_.size(); ++i) { - printf("chunk %zu: ", i); - offset = chunks_[i].WriteHeaderToFd(patch_fd, patch_data[i], offset); - } - - // Append each chunk's bsdiff patch, in order. - for (size_t i = 0; i < chunks_.size(); ++i) { - if (chunks_[i].GetType() != CHUNK_RAW) { - if (!android::base::WriteFully(patch_fd, patch_data[i].data(), patch_data[i].size())) { - printf("failed to write %zu bytes patch for chunk %zu\n", patch_data[i].size(), i); - return false; - } - } - } - - return true; -} - void Image::DumpChunks() const { std::string type = is_source_ ? "source" : "target"; printf("Dumping chunks for %s\n", type.c_str()); @@ -683,39 +529,6 @@ bool Image::ReadFile(const std::string& filename, std::vector<uint8_t>* file_con return true; } -class ZipModeImage : public Image { - public: - explicit ZipModeImage(bool is_source) : Image(is_source) {} - - bool Initialize(const std::string& filename) override; - - const ImageChunk& PseudoSource() const { - CHECK(is_source_); - CHECK(pseudo_source_ != nullptr); - return *pseudo_source_; - } - - // Verify that we can reconstruct the deflate chunks; also change the type to CHUNK_NORMAL if - // src and tgt are identical. - static bool CheckAndProcessChunks(ZipModeImage* tgt_image, ZipModeImage* src_image); - - // Compute the patches against the input image, and write the data into |patch_name|. - static bool GeneratePatches(ZipModeImage* tgt_image, ZipModeImage* src_image, - const std::string& patch_name); - - private: - // Initialize image chunks based on the zip entries. - bool InitializeChunks(const std::string& filename, ZipArchiveHandle handle); - // Add the a zip entry to the list. - bool AddZipEntryToChunks(ZipArchiveHandle handle, const std::string& entry_name, ZipEntry* entry); - // Return the real size of the zip file. (omit the trailing zeros that used for alignment) - bool GetZipFileSize(size_t* input_file_size); - - // The pesudo source chunk for bsdiff if there's no match for the given target chunk. It's in - // fact the whole source file. - std::unique_ptr<ImageChunk> pseudo_source_; -}; - bool ZipModeImage::Initialize(const std::string& filename) { if (!ReadFile(filename, &file_content_)) { return false; @@ -736,9 +549,6 @@ bool ZipModeImage::Initialize(const std::string& filename) { return false; } - if (is_source_) { - pseudo_source_ = std::make_unique<ImageChunk>(CHUNK_NORMAL, 0, &file_content_, zipfile_size); - } if (!InitializeChunks(filename, handle)) { CloseArchive(handle); return false; @@ -834,14 +644,11 @@ bool ZipModeImage::AddZipEntryToChunks(ZipArchiveHandle handle, const std::strin ErrorCodeString(ret)); return false; } - ImageChunk curr(CHUNK_DEFLATE, entry->offset, &file_content_, compressed_len); - curr.SetEntryName(entry_name); + ImageChunk curr(CHUNK_DEFLATE, entry->offset, &file_content_, compressed_len, entry_name); curr.SetUncompressedData(std::move(uncompressed_data)); - chunks_.push_back(curr); + chunks_.push_back(std::move(curr)); } else { - ImageChunk curr(CHUNK_NORMAL, entry->offset, &file_content_, compressed_len); - curr.SetEntryName(entry_name); - chunks_.push_back(curr); + chunks_.emplace_back(CHUNK_NORMAL, entry->offset, &file_content_, compressed_len, entry_name); } return true; @@ -880,6 +687,28 @@ bool ZipModeImage::GetZipFileSize(size_t* input_file_size) { return false; } +ImageChunk ZipModeImage::PseudoSource() const { + CHECK(is_source_); + return ImageChunk(CHUNK_NORMAL, 0, &file_content_, file_content_.size()); +} + +const ImageChunk* ZipModeImage::FindChunkByName(const std::string& name, bool find_normal) const { + if (name.empty()) { + return nullptr; + } + for (auto& chunk : chunks_) { + if ((chunk.GetType() == CHUNK_DEFLATE || find_normal) && chunk.GetEntryName() == name) { + return &chunk; + } + } + return nullptr; +} + +ImageChunk* ZipModeImage::FindChunkByName(const std::string& name, bool find_normal) { + return const_cast<ImageChunk*>( + static_cast<const ZipModeImage*>(this)->FindChunkByName(name, find_normal)); +} + bool ZipModeImage::CheckAndProcessChunks(ZipModeImage* tgt_image, ZipModeImage* src_image) { for (auto& tgt_chunk : *tgt_image) { if (tgt_chunk.GetType() != CHUNK_DEFLATE) { @@ -907,40 +736,55 @@ bool ZipModeImage::CheckAndProcessChunks(ZipModeImage* tgt_image, ZipModeImage* } } - return true; -} - -bool ZipModeImage::GeneratePatches(ZipModeImage* tgt_image, ZipModeImage* src_image, - const std::string& patch_name) { // For zips, we only need merge normal chunks for the target: deflated chunks are matched via // filename, and normal chunks are patched using the entire source file as the source. tgt_image->MergeAdjacentNormalChunks(); tgt_image->DumpChunks(); - printf("Construct patches for %zu chunks...\n", tgt_image->NumOfChunks()); - std::vector<std::vector<uint8_t>> patch_data(tgt_image->NumOfChunks()); + return true; +} + +bool ZipModeImage::GeneratePatches(const ZipModeImage& tgt_image, const ZipModeImage& src_image, + const std::string& patch_name) { + printf("Construct patches for %zu chunks...\n", tgt_image.NumOfChunks()); + std::vector<PatchChunk> patch_chunks; + patch_chunks.reserve(tgt_image.NumOfChunks()); saidx_t* bsdiff_cache = nullptr; - size_t i = 0; - for (auto& tgt_chunk : *tgt_image) { - ImageChunk* src_chunk = (tgt_chunk.GetType() != CHUNK_DEFLATE) - ? nullptr - : src_image->FindChunkByName(tgt_chunk.GetEntryName()); + for (size_t i = 0; i < tgt_image.NumOfChunks(); i++) { + const auto& tgt_chunk = tgt_image[i]; + + if (PatchChunk::RawDataIsSmaller(tgt_chunk, 0)) { + patch_chunks.emplace_back(tgt_chunk); + continue; + } - const auto& src_ref = (src_chunk == nullptr) ? src_image->PseudoSource() : *src_chunk; + const ImageChunk* src_chunk = (tgt_chunk.GetType() != CHUNK_DEFLATE) + ? nullptr + : src_image.FindChunkByName(tgt_chunk.GetEntryName()); + + const auto& src_ref = (src_chunk == nullptr) ? src_image.PseudoSource() : *src_chunk; saidx_t** bsdiff_cache_ptr = (src_chunk == nullptr) ? &bsdiff_cache : nullptr; - if (!tgt_chunk.MakePatch(src_ref, &patch_data[i], bsdiff_cache_ptr)) { + std::vector<uint8_t> patch_data; + if (!ImageChunk::MakePatch(tgt_chunk, src_ref, &patch_data, bsdiff_cache_ptr)) { printf("Failed to generate patch, name: %s\n", tgt_chunk.GetEntryName().c_str()); return false; } - printf("patch %3zu is %zu bytes (of %zu)\n", i, patch_data[i].size(), + printf("patch %3zu is %zu bytes (of %zu)\n", i, patch_data.size(), tgt_chunk.GetRawDataLength()); - i++; + + if (PatchChunk::RawDataIsSmaller(tgt_chunk, patch_data.size())) { + patch_chunks.emplace_back(tgt_chunk); + } else { + patch_chunks.emplace_back(tgt_chunk, src_ref, std::move(patch_data)); + } } free(bsdiff_cache); + CHECK_EQ(tgt_image.NumOfChunks(), patch_chunks.size()); + android::base::unique_fd patch_fd( open(patch_name.c_str(), O_CREAT | O_WRONLY | O_TRUNC, S_IRUSR | S_IWUSR)); if (patch_fd == -1) { @@ -948,26 +792,9 @@ bool ZipModeImage::GeneratePatches(ZipModeImage* tgt_image, ZipModeImage* src_im return false; } - return tgt_image->WritePatchDataToFd(patch_data, patch_fd); + return PatchChunk::WritePatchDataToFd(patch_chunks, patch_fd); } -class ImageModeImage : public Image { - public: - explicit ImageModeImage(bool is_source) : Image(is_source) {} - - // Initialize the image chunks list by searching the magic numbers in an image file. - bool Initialize(const std::string& filename) override; - - // In Image Mode, verify that the source and target images have the same chunk structure (ie, the - // same sequence of deflate and normal chunks). - static bool CheckAndProcessChunks(ImageModeImage* tgt_image, ImageModeImage* src_image); - - // In image mode, generate patches against the given source chunks and bonus_data; write the - // result to |patch_name|. - static bool GeneratePatches(ImageModeImage* tgt_image, ImageModeImage* src_image, - const std::vector<uint8_t>& bonus_data, const std::string& patch_name); -}; - bool ImageModeImage::Initialize(const std::string& filename) { if (!ReadFile(filename, &file_content_)) { return false; @@ -1053,7 +880,7 @@ bool ImageModeImage::Initialize(const std::string& filename) { ImageChunk body(CHUNK_DEFLATE, pos, &file_content_, raw_data_len); uncompressed_data.resize(uncompressed_len); body.SetUncompressedData(std::move(uncompressed_data)); - chunks_.push_back(body); + chunks_.push_back(std::move(body)); pos += raw_data_len; @@ -1083,6 +910,18 @@ bool ImageModeImage::Initialize(const std::string& filename) { return true; } +bool ImageModeImage::SetBonusData(const std::vector<uint8_t>& bonus_data) { + CHECK(is_source_); + if (chunks_.size() < 2 || !chunks_[1].SetBonusData(bonus_data)) { + printf("Failed to set bonus data\n"); + DumpChunks(); + return false; + } + + printf(" using %zu bytes of bonus data\n", bonus_data.size()); + return true; +} + // In Image Mode, verify that the source and target images have the same chunk structure (ie, the // same sequence of deflate and normal chunks). bool ImageModeImage::CheckAndProcessChunks(ImageModeImage* tgt_image, ImageModeImage* src_image) { @@ -1097,7 +936,7 @@ bool ImageModeImage::CheckAndProcessChunks(ImageModeImage* tgt_image, ImageModeI return false; } for (size_t i = 0; i < tgt_image->NumOfChunks(); ++i) { - if (tgt_image->Get(i)->GetType() != src_image->Get(i)->GetType()) { + if ((*tgt_image)[i].GetType() != (*src_image)[i].GetType()) { printf("source and target don't have same chunk structure! (chunk %zu)\n", i); tgt_image->DumpChunks(); src_image->DumpChunks(); @@ -1106,26 +945,23 @@ bool ImageModeImage::CheckAndProcessChunks(ImageModeImage* tgt_image, ImageModeI } for (size_t i = 0; i < tgt_image->NumOfChunks(); ++i) { - auto& tgt_chunk = *tgt_image->Get(i); - auto& src_chunk = *src_image->Get(i); + auto& tgt_chunk = (*tgt_image)[i]; + auto& src_chunk = (*src_image)[i]; if (tgt_chunk.GetType() != CHUNK_DEFLATE) { continue; } - // Confirm that we can recompress the data and get exactly the same bits as are in the - // input target image. - if (!tgt_chunk.ReconstructDeflateChunk()) { - printf("failed to reconstruct target deflate chunk %zu [%s]; treating as normal\n", i, - tgt_chunk.GetEntryName().c_str()); - tgt_chunk.ChangeDeflateChunkToNormal(); - src_chunk.ChangeDeflateChunkToNormal(); - continue; - } - // If two deflate chunks are identical treat them as normal chunks. if (tgt_chunk == src_chunk) { tgt_chunk.ChangeDeflateChunkToNormal(); src_chunk.ChangeDeflateChunkToNormal(); + } else if (!tgt_chunk.ReconstructDeflateChunk()) { + // We cannot recompress the data and get exactly the same bits as are in the input target + // image, fall back to normal + printf("failed to reconstruct target deflate chunk %zu [%s]; treating as normal\n", i, + tgt_chunk.GetEntryName().c_str()); + tgt_chunk.ChangeDeflateChunkToNormal(); + src_chunk.ChangeDeflateChunkToNormal(); } } @@ -1144,29 +980,39 @@ bool ImageModeImage::CheckAndProcessChunks(ImageModeImage* tgt_image, ImageModeI // In image mode, generate patches against the given source chunks and bonus_data; write the // result to |patch_name|. -bool ImageModeImage::GeneratePatches(ImageModeImage* tgt_image, ImageModeImage* src_image, - const std::vector<uint8_t>& bonus_data, +bool ImageModeImage::GeneratePatches(const ImageModeImage& tgt_image, + const ImageModeImage& src_image, const std::string& patch_name) { - printf("Construct patches for %zu chunks...\n", tgt_image->NumOfChunks()); - std::vector<std::vector<uint8_t>> patch_data(tgt_image->NumOfChunks()); + printf("Construct patches for %zu chunks...\n", tgt_image.NumOfChunks()); + std::vector<PatchChunk> patch_chunks; + patch_chunks.reserve(tgt_image.NumOfChunks()); - for (size_t i = 0; i < tgt_image->NumOfChunks(); i++) { - auto& tgt_chunk = *tgt_image->Get(i); - auto& src_chunk = *src_image->Get(i); + for (size_t i = 0; i < tgt_image.NumOfChunks(); i++) { + const auto& tgt_chunk = tgt_image[i]; + const auto& src_chunk = src_image[i]; - if (i == 1 && !bonus_data.empty()) { - printf(" using %zu bytes of bonus data for chunk %zu\n", bonus_data.size(), i); - src_chunk.SetBonusData(bonus_data); + if (PatchChunk::RawDataIsSmaller(tgt_chunk, 0)) { + patch_chunks.emplace_back(tgt_chunk); + continue; } - if (!tgt_chunk.MakePatch(src_chunk, &patch_data[i], nullptr)) { + std::vector<uint8_t> patch_data; + if (!ImageChunk::MakePatch(tgt_chunk, src_chunk, &patch_data, nullptr)) { printf("Failed to generate patch for target chunk %zu: ", i); return false; } - printf("patch %3zu is %zu bytes (of %zu)\n", i, patch_data[i].size(), + printf("patch %3zu is %zu bytes (of %zu)\n", i, patch_data.size(), tgt_chunk.GetRawDataLength()); + + if (PatchChunk::RawDataIsSmaller(tgt_chunk, patch_data.size())) { + patch_chunks.emplace_back(tgt_chunk); + } else { + patch_chunks.emplace_back(tgt_chunk, src_chunk, std::move(patch_data)); + } } + CHECK_EQ(tgt_image.NumOfChunks(), patch_chunks.size()); + android::base::unique_fd patch_fd( open(patch_name.c_str(), O_CREAT | O_WRONLY | O_TRUNC, S_IRUSR | S_IWUSR)); if (patch_fd == -1) { @@ -1174,7 +1020,7 @@ bool ImageModeImage::GeneratePatches(ImageModeImage* tgt_image, ImageModeImage* return false; } - return tgt_image->WritePatchDataToFd(patch_data, patch_fd); + return PatchChunk::WritePatchDataToFd(patch_chunks, patch_fd); } int imgdiff(int argc, const char** argv) { @@ -1236,7 +1082,7 @@ int imgdiff(int argc, const char** argv) { } // Compute bsdiff patches for each chunk's data (the uncompressed data, in the case of // deflate chunks). - if (!ZipModeImage::GeneratePatches(&tgt_image, &src_image, argv[optind + 2])) { + if (!ZipModeImage::GeneratePatches(tgt_image, src_image, argv[optind + 2])) { return 1; } } else { @@ -1253,7 +1099,12 @@ int imgdiff(int argc, const char** argv) { if (!ImageModeImage::CheckAndProcessChunks(&tgt_image, &src_image)) { return 1; } - if (!ImageModeImage::GeneratePatches(&tgt_image, &src_image, bonus_data, argv[optind + 2])) { + + if (!bonus_data.empty() && !src_image.SetBonusData(bonus_data)) { + return 1; + } + + if (!ImageModeImage::GeneratePatches(tgt_image, src_image, argv[optind + 2])) { return 1; } } |