diff options
author | dgarrett@chromium.org <dgarrett@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | 2011-10-26 00:50:20 +0000 |
---|---|---|
committer | dgarrett@chromium.org <dgarrett@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | 2011-10-26 00:50:20 +0000 |
commit | 423a381f4fd3efd99dfd7bc932777ea596cf7b17 (patch) | |
tree | fdbf4a4bc5f2b8d73b90020da470c40a22f4cc2a /courgette/disassembler_win32_x86.cc | |
parent | da1543a1a526aefd1114853cf737846eb5c29640 (diff) | |
download | chromium_src-423a381f4fd3efd99dfd7bc932777ea596cf7b17.zip chromium_src-423a381f4fd3efd99dfd7bc932777ea596cf7b17.tar.gz chromium_src-423a381f4fd3efd99dfd7bc932777ea596cf7b17.tar.bz2 |
Further refactoring, move ImageInfo into Disassembler/DisassemblerWin32X86.
This means that all PE specific knowledge is now contained in a single class
which leaves us in pretty good shape for supporting ELF 32.
There are still widespread assumptions about being 32 bit, but those can be
addressed at a much later date.
BUG=None
TEST=Unittests
Review URL: http://codereview.chromium.org/8166013
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@107260 0039d316-1c4b-4281-b951-d872f2087c98
Diffstat (limited to 'courgette/disassembler_win32_x86.cc')
-rw-r--r-- | courgette/disassembler_win32_x86.cc | 480 |
1 files changed, 419 insertions, 61 deletions
diff --git a/courgette/disassembler_win32_x86.cc b/courgette/disassembler_win32_x86.cc index fb12c22..d09d67d 100644 --- a/courgette/disassembler_win32_x86.cc +++ b/courgette/disassembler_win32_x86.cc @@ -14,7 +14,6 @@ #include "courgette/assembly_program.h" #include "courgette/courgette.h" #include "courgette/encoded_program.h" -#include "courgette/image_info.h" // COURGETTE_HISTOGRAM_TARGETS prints out a histogram of how frequently // different target addresses are referenced. Purely for debugging. @@ -22,16 +21,189 @@ namespace courgette { -DisassemblerWin32X86::DisassemblerWin32X86(PEInfo* pe_info) - : pe_info_(pe_info), - incomplete_disassembly_(false) { +DisassemblerWin32X86::DisassemblerWin32X86(const void* start, size_t length) + : Disassembler(start, length), + incomplete_disassembly_(false), + is_PE32_plus_(false), + optional_header_(NULL), + size_of_optional_header_(0), + offset_of_data_directories_(0), + machine_type_(0), + number_of_sections_(0), + sections_(NULL), + has_text_section_(false), + size_of_code_(0), + size_of_initialized_data_(0), + size_of_uninitialized_data_(0), + base_of_code_(0), + base_of_data_(0), + image_base_(0), + size_of_image_(0), + number_of_data_directories_(0) { +} + +// ParseHeader attempts to match up the buffer with the Windows data +// structures that exist within a Windows 'Portable Executable' format file. +// Returns 'true' if the buffer matches, and 'false' if the data looks +// suspicious. Rather than try to 'map' the buffer to the numerous windows +// structures, we extract the information we need into the courgette::PEInfo +// structure. +// +bool DisassemblerWin32X86::ParseHeader() { + if (length() < kOffsetOfFileAddressOfNewExeHeader + 4 /*size*/) + return Bad("Too small"); + + // Have 'MZ' magic for a DOS header? + if (start()[0] != 'M' || start()[1] != 'Z') + return Bad("Not MZ"); + + // offset from DOS header to PE header is stored in DOS header. + uint32 offset = ReadU32(start(), + kOffsetOfFileAddressOfNewExeHeader); + + if (offset >= length()) + return Bad("Bad offset to PE header"); + + const uint8* const pe_header = OffsetToPointer(offset); + const size_t kMinPEHeaderSize = 4 /*signature*/ + kSizeOfCoffHeader; + if (pe_header <= start() || + pe_header >= end() - kMinPEHeaderSize) + return Bad("Bad offset to PE header"); + + if (offset % 8 != 0) + return Bad("Misaligned PE header"); + + // The 'PE' header is an IMAGE_NT_HEADERS structure as defined in WINNT.H. + // See http://msdn.microsoft.com/en-us/library/ms680336(VS.85).aspx + // + // The first field of the IMAGE_NT_HEADERS is the signature. + if (!(pe_header[0] == 'P' && + pe_header[1] == 'E' && + pe_header[2] == 0 && + pe_header[3] == 0)) + return Bad("no PE signature"); + + // The second field of the IMAGE_NT_HEADERS is the COFF header. + // The COFF header is also called an IMAGE_FILE_HEADER + // http://msdn.microsoft.com/en-us/library/ms680313(VS.85).aspx + const uint8* const coff_header = pe_header + 4; + machine_type_ = ReadU16(coff_header, 0); + number_of_sections_ = ReadU16(coff_header, 2); + size_of_optional_header_ = ReadU16(coff_header, 16); + + // The rest of the IMAGE_NT_HEADERS is the IMAGE_OPTIONAL_HEADER(32|64) + const uint8* const optional_header = coff_header + kSizeOfCoffHeader; + optional_header_ = optional_header; + + if (optional_header + size_of_optional_header_ >= end()) + return Bad("optional header past end of file"); + + // Check we can read the magic. + if (size_of_optional_header_ < 2) + return Bad("optional header no magic"); + + uint16 magic = ReadU16(optional_header, 0); + + if (magic == kImageNtOptionalHdr32Magic) { + is_PE32_plus_ = false; + offset_of_data_directories_ = + kOffsetOfDataDirectoryFromImageOptionalHeader32; + } else if (magic == kImageNtOptionalHdr64Magic) { + is_PE32_plus_ = true; + offset_of_data_directories_ = + kOffsetOfDataDirectoryFromImageOptionalHeader64; + } else { + return Bad("unrecognized magic"); + } + + // Check that we can read the rest of the the fixed fields. Data directories + // directly follow the fixed fields of the IMAGE_OPTIONAL_HEADER. + if (size_of_optional_header_ < offset_of_data_directories_) + return Bad("optional header too short"); + + // The optional header is either an IMAGE_OPTIONAL_HEADER32 or + // IMAGE_OPTIONAL_HEADER64 + // http://msdn.microsoft.com/en-us/library/ms680339(VS.85).aspx + // + // Copy the fields we care about. + size_of_code_ = ReadU32(optional_header, 4); + size_of_initialized_data_ = ReadU32(optional_header, 8); + size_of_uninitialized_data_ = ReadU32(optional_header, 12); + base_of_code_ = ReadU32(optional_header, 20); + if (is_PE32_plus_) { + base_of_data_ = 0; + image_base_ = ReadU64(optional_header, 24); + } else { + base_of_data_ = ReadU32(optional_header, 24); + image_base_ = ReadU32(optional_header, 28); + } + size_of_image_ = ReadU32(optional_header, 56); + number_of_data_directories_ = + ReadU32(optional_header, (is_PE32_plus_ ? 108 : 92)); + + if (size_of_code_ >= length() || + size_of_initialized_data_ >= length() || + size_of_code_ + size_of_initialized_data_ >= length()) { + // This validation fires on some perfectly fine executables. + // return Bad("code or initialized data too big"); + } + + // TODO(sra): we can probably get rid of most of the data directories. + bool b = true; + // 'b &= ...' could be short circuit 'b = b && ...' but it is not necessary + // for correctness and it compiles smaller this way. + b &= ReadDataDirectory(0, &export_table_); + b &= ReadDataDirectory(1, &import_table_); + b &= ReadDataDirectory(2, &resource_table_); + b &= ReadDataDirectory(3, &exception_table_); + b &= ReadDataDirectory(5, &base_relocation_table_); + b &= ReadDataDirectory(11, &bound_import_table_); + b &= ReadDataDirectory(12, &import_address_table_); + b &= ReadDataDirectory(13, &delay_import_descriptor_); + b &= ReadDataDirectory(14, &clr_runtime_header_); + if (!b) { + return Bad("malformed data directory"); + } + + // Sections follow the optional header. + sections_ = + reinterpret_cast<const Section*>(optional_header + + size_of_optional_header_); + size_t detected_length = 0; + + for (int i = 0; i < number_of_sections_; ++i) { + const Section* section = §ions_[i]; + + // TODO(sra): consider using the 'characteristics' field of the section + // header to see if the section contains instructions. + if (memcmp(section->name, ".text", 6) == 0) + has_text_section_ = true; + + uint32 section_end = + section->file_offset_of_raw_data + section->size_of_raw_data; + if (section_end > detected_length) + detected_length = section_end; + } + + // Pretend our in-memory copy is only as long as our detected length. + ReduceLength(detected_length); + + if (!is_32bit()) { + return Bad("64 bit executables are not yet supported"); + } + + if (!has_text_section()) { + return Bad("Resource-only executables are not yet supported"); + } + + return Good(); } bool DisassemblerWin32X86::Disassemble(AssemblyProgram* target) { - if (!pe_info().ok()) + if (!ok()) return false; - target->set_image_base(pe_info().image_base()); + target->set_image_base(image_base()); if (!ParseAbs32Relocs()) return false; @@ -46,13 +218,159 @@ bool DisassemblerWin32X86::Disassemble(AssemblyProgram* target) { return true; } -static uint32 Read32LittleEndian(const void* address) { - return *reinterpret_cast<const uint32*>(address); +//////////////////////////////////////////////////////////////////////////////// + +bool DisassemblerWin32X86::ParseRelocs(std::vector<RVA> *relocs) { + relocs->clear(); + + size_t relocs_size = base_relocation_table_.size_; + if (relocs_size == 0) + return true; + + // The format of the base relocation table is a sequence of variable sized + // IMAGE_BASE_RELOCATION blocks. Search for + // "The format of the base relocation data is somewhat quirky" + // at http://msdn.microsoft.com/en-us/library/ms809762.aspx + + const uint8* relocs_start = RVAToPointer(base_relocation_table_.address_); + const uint8* relocs_end = relocs_start + relocs_size; + + // Make sure entire base relocation table is within the buffer. + if (relocs_start < start() || + relocs_start >= end() || + relocs_end <= start() || + relocs_end > end()) { + return Bad(".relocs outside image"); + } + + const uint8* block = relocs_start; + + // Walk the variable sized blocks. + while (block + 8 < relocs_end) { + RVA page_rva = ReadU32(block, 0); + uint32 size = ReadU32(block, 4); + if (size < 8 || // Size includes header ... + size % 4 != 0) // ... and is word aligned. + return Bad("unreasonable relocs block"); + + const uint8* end_entries = block + size; + + if (end_entries <= block || + end_entries <= start() || + end_entries > end()) + return Bad(".relocs block outside image"); + + // Walk through the two-byte entries. + for (const uint8* p = block + 8; p < end_entries; p += 2) { + uint16 entry = ReadU16(p, 0); + int type = entry >> 12; + int offset = entry & 0xFFF; + + RVA rva = page_rva + offset; + if (type == 3) { // IMAGE_REL_BASED_HIGHLOW + relocs->push_back(rva); + } else if (type == 0) { // IMAGE_REL_BASED_ABSOLUTE + // Ignore, used as padding. + } else { + // Does not occur in Windows x86 executables. + return Bad("unknown type of reloc"); + } + } + + block += size; + } + + std::sort(relocs->begin(), relocs->end()); + + return true; +} + +const Section* DisassemblerWin32X86::RVAToSection(RVA rva) const { + for (int i = 0; i < number_of_sections_; i++) { + const Section* section = §ions_[i]; + uint32 offset = rva - section->virtual_address; + if (offset < section->virtual_size) { + return section; + } + } + return NULL; +} + +int DisassemblerWin32X86::RVAToFileOffset(RVA rva) const { + const Section* section = RVAToSection(rva); + if (section) { + uint32 offset = rva - section->virtual_address; + if (offset < section->size_of_raw_data) { + return section->file_offset_of_raw_data + offset; + } else { + return kNoOffset; // In section but not in file (e.g. uninit data). + } + } + + // Small RVA values point into the file header in the loaded image. + // RVA 0 is the module load address which Windows uses as the module handle. + // RVA 2 sometimes occurs, I'm not sure what it is, but it would map into the + // DOS header. + if (rva == 0 || rva == 2) + return rva; + + NOTREACHED(); + return kNoOffset; +} + +const uint8* DisassemblerWin32X86::RVAToPointer(RVA rva) const { + int file_offset = RVAToFileOffset(rva); + if (file_offset == kNoOffset) + return NULL; + else + return OffsetToPointer(file_offset); +} + +std::string DisassemblerWin32X86::SectionName(const Section* section) { + if (section == NULL) + return "<none>"; + char name[9]; + memcpy(name, section->name, 8); + name[8] = '\0'; // Ensure termination. + return name; +} + +CheckBool DisassemblerWin32X86::ParseFile(AssemblyProgram* program) { + bool ok = true; + // Walk all the bytes in the file, whether or not in a section. + uint32 file_offset = 0; + while (ok && file_offset < length()) { + const Section* section = FindNextSection(file_offset); + if (section == NULL) { + // No more sections. There should not be extra stuff following last + // section. + // ParseNonSectionFileRegion(file_offset, pe_info().length(), program); + break; + } + if (file_offset < section->file_offset_of_raw_data) { + uint32 section_start_offset = section->file_offset_of_raw_data; + ok = ParseNonSectionFileRegion(file_offset, section_start_offset, + program); + file_offset = section_start_offset; + } + if (ok) { + uint32 end = file_offset + section->size_of_raw_data; + ok = ParseFileRegion(section, file_offset, end, program); + file_offset = end; + } + } + +#if COURGETTE_HISTOGRAM_TARGETS + HistogramTargets("abs32 relocs", abs32_target_rvas_); + HistogramTargets("rel32 relocs", rel32_target_rvas_); +#endif + + return ok; } bool DisassemblerWin32X86::ParseAbs32Relocs() { abs32_locations_.clear(); - if (!pe_info().ParseRelocs(&abs32_locations_)) + if (!ParseRelocs(&abs32_locations_)) return false; std::sort(abs32_locations_.begin(), abs32_locations_.end()); @@ -61,8 +379,8 @@ bool DisassemblerWin32X86::ParseAbs32Relocs() { for (size_t i = 0; i < abs32_locations_.size(); ++i) { RVA rva = abs32_locations_[i]; // The 4 bytes at the relocation are a reference to some address. - uint32 target_address = Read32LittleEndian(pe_info().RVAToPointer(rva)); - ++abs32_target_rvas_[target_address - pe_info().image_base()]; + uint32 target_address = Read32LittleEndian(RVAToPointer(rva)); + ++abs32_target_rvas_[target_address - image_base()]; } #endif return true; @@ -70,8 +388,8 @@ bool DisassemblerWin32X86::ParseAbs32Relocs() { void DisassemblerWin32X86::ParseRel32RelocsFromSections() { uint32 file_offset = 0; - while (file_offset < pe_info().length()) { - const Section* section = pe_info().FindNextSection(file_offset); + while (file_offset < length()) { + const Section* section = FindNextSection(file_offset); if (section == NULL) break; if (file_offset < section->file_offset_of_raw_data) @@ -114,12 +432,12 @@ void DisassemblerWin32X86::ParseRel32RelocsFromSection(const Section* section) { uint32 start_file_offset = section->file_offset_of_raw_data; uint32 end_file_offset = start_file_offset + section->size_of_raw_data; - RVA relocs_start_rva = pe_info().base_relocation_table().address_; + RVA relocs_start_rva = base_relocation_table().address_; - const uint8* start_pointer = pe_info().FileOffsetToPointer(start_file_offset); - const uint8* end_pointer = pe_info().FileOffsetToPointer(end_file_offset); + const uint8* start_pointer = OffsetToPointer(start_file_offset); + const uint8* end_pointer = OffsetToPointer(end_file_offset); - RVA start_rva = pe_info().FileOffsetToRVA(start_file_offset); + RVA start_rva = FileOffsetToRVA(start_file_offset); RVA end_rva = start_rva + section->virtual_size; // Quick way to convert from Pointer to RVA within a single Section is to @@ -133,7 +451,7 @@ void DisassemblerWin32X86::ParseRel32RelocsFromSection(const Section* section) { while (p < end_pointer) { RVA current_rva = static_cast<RVA>(p - adjust_pointer_to_rva); if (current_rva == relocs_start_rva) { - uint32 relocs_size = pe_info().base_relocation_table().size_; + uint32 relocs_size = base_relocation_table().size_; if (relocs_size) { p += relocs_size; continue; @@ -179,7 +497,7 @@ void DisassemblerWin32X86::ParseRel32RelocsFromSection(const Section* section) { RVA target_rva = rel32_rva + 4 + Read32LittleEndian(rel32); // To be valid, rel32 target must be within image, and within this // section. - if (pe_info().IsValidRVA(target_rva) && + if (IsValidRVA(target_rva) && start_rva <= target_rva && target_rva < end_rva) { rel32_locations_.push_back(rel32_rva); #if COURGETTE_HISTOGRAM_TARGETS @@ -193,39 +511,6 @@ void DisassemblerWin32X86::ParseRel32RelocsFromSection(const Section* section) { } } -CheckBool DisassemblerWin32X86::ParseFile(AssemblyProgram* program) { - bool ok = true; - // Walk all the bytes in the file, whether or not in a section. - uint32 file_offset = 0; - while (ok && file_offset < pe_info().length()) { - const Section* section = pe_info().FindNextSection(file_offset); - if (section == NULL) { - // No more sections. There should not be extra stuff following last - // section. - // ParseNonSectionFileRegion(file_offset, pe_info().length(), program); - break; - } - if (file_offset < section->file_offset_of_raw_data) { - uint32 section_start_offset = section->file_offset_of_raw_data; - ok = ParseNonSectionFileRegion(file_offset, section_start_offset, - program); - file_offset = section_start_offset; - } - if (ok) { - uint32 end = file_offset + section->size_of_raw_data; - ok = ParseFileRegion(section, file_offset, end, program); - file_offset = end; - } - } - -#if COURGETTE_HISTOGRAM_TARGETS - HistogramTargets("abs32 relocs", abs32_target_rvas_); - HistogramTargets("rel32 relocs", rel32_target_rvas_); -#endif - - return ok; -} - CheckBool DisassemblerWin32X86::ParseNonSectionFileRegion( uint32 start_file_offset, uint32 end_file_offset, @@ -233,8 +518,8 @@ CheckBool DisassemblerWin32X86::ParseNonSectionFileRegion( if (incomplete_disassembly_) return true; - const uint8* start = pe_info().FileOffsetToPointer(start_file_offset); - const uint8* end = pe_info().FileOffsetToPointer(end_file_offset); + const uint8* start = OffsetToPointer(start_file_offset); + const uint8* end = OffsetToPointer(end_file_offset); const uint8* p = start; @@ -251,12 +536,12 @@ CheckBool DisassemblerWin32X86::ParseFileRegion( const Section* section, uint32 start_file_offset, uint32 end_file_offset, AssemblyProgram* program) { - RVA relocs_start_rva = pe_info().base_relocation_table().address_; + RVA relocs_start_rva = base_relocation_table().address_; - const uint8* start_pointer = pe_info().FileOffsetToPointer(start_file_offset); - const uint8* end_pointer = pe_info().FileOffsetToPointer(end_file_offset); + const uint8* start_pointer = OffsetToPointer(start_file_offset); + const uint8* end_pointer = OffsetToPointer(end_file_offset); - RVA start_rva = pe_info().FileOffsetToRVA(start_file_offset); + RVA start_rva = FileOffsetToRVA(start_file_offset); RVA end_rva = start_rva + section->virtual_size; // Quick way to convert from Pointer to RVA within a single Section is to @@ -280,7 +565,7 @@ CheckBool DisassemblerWin32X86::ParseFileRegion( ok = program->EmitMakeRelocsInstruction(); if (!ok) break; - uint32 relocs_size = pe_info().base_relocation_table().size_; + uint32 relocs_size = base_relocation_table().size_; if (relocs_size) { p += relocs_size; continue; @@ -292,7 +577,7 @@ CheckBool DisassemblerWin32X86::ParseFileRegion( if (abs32_pos != abs32_locations_.end() && *abs32_pos == current_rva) { uint32 target_address = Read32LittleEndian(p); - RVA target_rva = target_address - pe_info().image_base(); + RVA target_rva = target_address - image_base(); // TODO(sra): target could be Label+offset. It is not clear how to guess // which it might be. We assume offset==0. ok = program->EmitAbs32(program->FindOrMakeAbs32Label(target_rva)); @@ -363,7 +648,7 @@ void DisassemblerWin32X86::HistogramTargets(const char* kind, std::cout << std::dec << p->first << ": " << count; if (count <= 2) { for (size_t i = 0; i < count; ++i) - std::cout << " " << pe_info().DescribeRVA(p->second[i]); + std::cout << " " << DescribeRVA(p->second[i]); } std::cout << std::endl; someSkipped = false; @@ -374,4 +659,77 @@ void DisassemblerWin32X86::HistogramTargets(const char* kind, } #endif // COURGETTE_HISTOGRAM_TARGETS + +// DescribeRVA is for debugging only. I would put it under #ifdef DEBUG except +// that during development I'm finding I need to call it when compiled in +// Release mode. Hence: +// TODO(sra): make this compile only for debug mode. +std::string DisassemblerWin32X86::DescribeRVA(RVA rva) const { + const Section* section = RVAToSection(rva); + std::ostringstream s; + s << std::hex << rva; + if (section) { + s << " ("; + s << SectionName(section) << "+" + << std::hex << (rva - section->virtual_address) + << ")"; + } + return s.str(); +} + +const Section* DisassemblerWin32X86::FindNextSection(uint32 fileOffset) const { + const Section* best = 0; + for (int i = 0; i < number_of_sections_; i++) { + const Section* section = §ions_[i]; + if (section->size_of_raw_data > 0) { // i.e. has data in file. + if (fileOffset <= section->file_offset_of_raw_data) { + if (best == 0 || + section->file_offset_of_raw_data < best->file_offset_of_raw_data) { + best = section; + } + } + } + } + return best; +} + +RVA DisassemblerWin32X86::FileOffsetToRVA(uint32 file_offset) const { + for (int i = 0; i < number_of_sections_; i++) { + const Section* section = §ions_[i]; + uint32 offset = file_offset - section->file_offset_of_raw_data; + if (offset < section->size_of_raw_data) { + return section->virtual_address + offset; + } + } + return 0; +} + +bool DisassemblerWin32X86::ReadDataDirectory( + int index, + ImageDataDirectory* directory) { + + if (index < number_of_data_directories_) { + size_t offset = index * 8 + offset_of_data_directories_; + if (offset >= size_of_optional_header_) + return Bad("number of data directories inconsistent"); + const uint8* data_directory = optional_header_ + offset; + if (data_directory < start() || + data_directory + 8 >= end()) + return Bad("data directory outside image"); + RVA rva = ReadU32(data_directory, 0); + size_t size = ReadU32(data_directory, 4); + if (size > size_of_image_) + return Bad("data directory size too big"); + + // TODO(sra): validate RVA. + directory->address_ = rva; + directory->size_ = static_cast<uint32>(size); + return true; + } else { + directory->address_ = 0; + directory->size_ = 0; + return true; + } +} + } // namespace courgette |