SupervisedUser SafeSites: Switch to the new SafeSearch API
[chromium-blink-merge.git] / courgette / disassembler_win32_x86.cc
blob8490fcba75afd5163acd569509f4ee23c9a1f44c
1 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 #include "courgette/disassembler_win32_x86.h"
7 #include <algorithm>
8 #include <string>
9 #include <vector>
11 #include "base/basictypes.h"
12 #include "base/logging.h"
14 #include "courgette/assembly_program.h"
15 #include "courgette/courgette.h"
16 #include "courgette/encoded_program.h"
18 namespace courgette {
20 DisassemblerWin32X86::DisassemblerWin32X86(const void* start, size_t length)
21 : Disassembler(start, length),
22 incomplete_disassembly_(false),
23 is_PE32_plus_(false),
24 optional_header_(NULL),
25 size_of_optional_header_(0),
26 offset_of_data_directories_(0),
27 machine_type_(0),
28 number_of_sections_(0),
29 sections_(NULL),
30 has_text_section_(false),
31 size_of_code_(0),
32 size_of_initialized_data_(0),
33 size_of_uninitialized_data_(0),
34 base_of_code_(0),
35 base_of_data_(0),
36 image_base_(0),
37 size_of_image_(0),
38 number_of_data_directories_(0) {
41 // ParseHeader attempts to match up the buffer with the Windows data
42 // structures that exist within a Windows 'Portable Executable' format file.
43 // Returns 'true' if the buffer matches, and 'false' if the data looks
44 // suspicious. Rather than try to 'map' the buffer to the numerous windows
45 // structures, we extract the information we need into the courgette::PEInfo
46 // structure.
48 bool DisassemblerWin32X86::ParseHeader() {
49 if (length() < kOffsetOfFileAddressOfNewExeHeader + 4 /*size*/)
50 return Bad("Too small");
52 // Have 'MZ' magic for a DOS header?
53 if (start()[0] != 'M' || start()[1] != 'Z')
54 return Bad("Not MZ");
56 // offset from DOS header to PE header is stored in DOS header.
57 uint32 offset = ReadU32(start(),
58 kOffsetOfFileAddressOfNewExeHeader);
60 if (offset >= length())
61 return Bad("Bad offset to PE header");
63 const uint8* const pe_header = OffsetToPointer(offset);
64 const size_t kMinPEHeaderSize = 4 /*signature*/ + kSizeOfCoffHeader;
65 if (pe_header <= start() ||
66 pe_header >= end() - kMinPEHeaderSize)
67 return Bad("Bad offset to PE header");
69 if (offset % 8 != 0)
70 return Bad("Misaligned PE header");
72 // The 'PE' header is an IMAGE_NT_HEADERS structure as defined in WINNT.H.
73 // See http://msdn.microsoft.com/en-us/library/ms680336(VS.85).aspx
75 // The first field of the IMAGE_NT_HEADERS is the signature.
76 if (!(pe_header[0] == 'P' &&
77 pe_header[1] == 'E' &&
78 pe_header[2] == 0 &&
79 pe_header[3] == 0))
80 return Bad("no PE signature");
82 // The second field of the IMAGE_NT_HEADERS is the COFF header.
83 // The COFF header is also called an IMAGE_FILE_HEADER
84 // http://msdn.microsoft.com/en-us/library/ms680313(VS.85).aspx
85 const uint8* const coff_header = pe_header + 4;
86 machine_type_ = ReadU16(coff_header, 0);
87 number_of_sections_ = ReadU16(coff_header, 2);
88 size_of_optional_header_ = ReadU16(coff_header, 16);
90 // The rest of the IMAGE_NT_HEADERS is the IMAGE_OPTIONAL_HEADER(32|64)
91 const uint8* const optional_header = coff_header + kSizeOfCoffHeader;
92 optional_header_ = optional_header;
94 if (optional_header + size_of_optional_header_ >= end())
95 return Bad("optional header past end of file");
97 // Check we can read the magic.
98 if (size_of_optional_header_ < 2)
99 return Bad("optional header no magic");
101 uint16 magic = ReadU16(optional_header, 0);
103 if (magic == kImageNtOptionalHdr32Magic) {
104 is_PE32_plus_ = false;
105 offset_of_data_directories_ =
106 kOffsetOfDataDirectoryFromImageOptionalHeader32;
107 } else if (magic == kImageNtOptionalHdr64Magic) {
108 is_PE32_plus_ = true;
109 offset_of_data_directories_ =
110 kOffsetOfDataDirectoryFromImageOptionalHeader64;
111 } else {
112 return Bad("unrecognized magic");
115 // Check that we can read the rest of the the fixed fields. Data directories
116 // directly follow the fixed fields of the IMAGE_OPTIONAL_HEADER.
117 if (size_of_optional_header_ < offset_of_data_directories_)
118 return Bad("optional header too short");
120 // The optional header is either an IMAGE_OPTIONAL_HEADER32 or
121 // IMAGE_OPTIONAL_HEADER64
122 // http://msdn.microsoft.com/en-us/library/ms680339(VS.85).aspx
124 // Copy the fields we care about.
125 size_of_code_ = ReadU32(optional_header, 4);
126 size_of_initialized_data_ = ReadU32(optional_header, 8);
127 size_of_uninitialized_data_ = ReadU32(optional_header, 12);
128 base_of_code_ = ReadU32(optional_header, 20);
129 if (is_PE32_plus_) {
130 base_of_data_ = 0;
131 image_base_ = ReadU64(optional_header, 24);
132 } else {
133 base_of_data_ = ReadU32(optional_header, 24);
134 image_base_ = ReadU32(optional_header, 28);
136 size_of_image_ = ReadU32(optional_header, 56);
137 number_of_data_directories_ =
138 ReadU32(optional_header, (is_PE32_plus_ ? 108 : 92));
140 if (size_of_code_ >= length() ||
141 size_of_initialized_data_ >= length() ||
142 size_of_code_ + size_of_initialized_data_ >= length()) {
143 // This validation fires on some perfectly fine executables.
144 // return Bad("code or initialized data too big");
147 // TODO(sra): we can probably get rid of most of the data directories.
148 bool b = true;
149 // 'b &= ...' could be short circuit 'b = b && ...' but it is not necessary
150 // for correctness and it compiles smaller this way.
151 b &= ReadDataDirectory(0, &export_table_);
152 b &= ReadDataDirectory(1, &import_table_);
153 b &= ReadDataDirectory(2, &resource_table_);
154 b &= ReadDataDirectory(3, &exception_table_);
155 b &= ReadDataDirectory(5, &base_relocation_table_);
156 b &= ReadDataDirectory(11, &bound_import_table_);
157 b &= ReadDataDirectory(12, &import_address_table_);
158 b &= ReadDataDirectory(13, &delay_import_descriptor_);
159 b &= ReadDataDirectory(14, &clr_runtime_header_);
160 if (!b) {
161 return Bad("malformed data directory");
164 // Sections follow the optional header.
165 sections_ =
166 reinterpret_cast<const Section*>(optional_header +
167 size_of_optional_header_);
168 size_t detected_length = 0;
170 for (int i = 0; i < number_of_sections_; ++i) {
171 const Section* section = &sections_[i];
173 // TODO(sra): consider using the 'characteristics' field of the section
174 // header to see if the section contains instructions.
175 if (memcmp(section->name, ".text", 6) == 0)
176 has_text_section_ = true;
178 uint32 section_end =
179 section->file_offset_of_raw_data + section->size_of_raw_data;
180 if (section_end > detected_length)
181 detected_length = section_end;
184 // Pretend our in-memory copy is only as long as our detected length.
185 ReduceLength(detected_length);
187 if (!is_32bit()) {
188 return Bad("64 bit executables are not supported by this disassembler");
191 if (!has_text_section()) {
192 return Bad("Resource-only executables are not yet supported");
195 return Good();
198 bool DisassemblerWin32X86::Disassemble(AssemblyProgram* target) {
199 if (!ok())
200 return false;
202 target->set_image_base(image_base());
204 if (!ParseAbs32Relocs())
205 return false;
207 ParseRel32RelocsFromSections();
209 if (!ParseFile(target))
210 return false;
212 target->DefaultAssignIndexes();
214 return true;
217 ////////////////////////////////////////////////////////////////////////////////
219 bool DisassemblerWin32X86::ParseRelocs(std::vector<RVA> *relocs) {
220 relocs->clear();
222 size_t relocs_size = base_relocation_table_.size_;
223 if (relocs_size == 0)
224 return true;
226 // The format of the base relocation table is a sequence of variable sized
227 // IMAGE_BASE_RELOCATION blocks. Search for
228 // "The format of the base relocation data is somewhat quirky"
229 // at http://msdn.microsoft.com/en-us/library/ms809762.aspx
231 const uint8* relocs_start = RVAToPointer(base_relocation_table_.address_);
232 const uint8* relocs_end = relocs_start + relocs_size;
234 // Make sure entire base relocation table is within the buffer.
235 if (relocs_start < start() ||
236 relocs_start >= end() ||
237 relocs_end <= start() ||
238 relocs_end > end()) {
239 return Bad(".relocs outside image");
242 const uint8* block = relocs_start;
244 // Walk the variable sized blocks.
245 while (block + 8 < relocs_end) {
246 RVA page_rva = ReadU32(block, 0);
247 uint32 size = ReadU32(block, 4);
248 if (size < 8 || // Size includes header ...
249 size % 4 != 0) // ... and is word aligned.
250 return Bad("unreasonable relocs block");
252 const uint8* end_entries = block + size;
254 if (end_entries <= block ||
255 end_entries <= start() ||
256 end_entries > end())
257 return Bad(".relocs block outside image");
259 // Walk through the two-byte entries.
260 for (const uint8* p = block + 8; p < end_entries; p += 2) {
261 uint16 entry = ReadU16(p, 0);
262 int type = entry >> 12;
263 int offset = entry & 0xFFF;
265 RVA rva = page_rva + offset;
266 // Skip the relocs that live outside of the image. It might be the case
267 // if a reloc is relative to a register, e.g.:
268 // mov ecx,dword ptr [eax+044D5888h]
269 uint32 target_address = Read32LittleEndian(RVAToPointer(rva));
270 if (target_address < image_base_ ||
271 target_address > (image_base_ + size_of_image_)) {
272 continue;
274 if (type == 3) { // IMAGE_REL_BASED_HIGHLOW
275 relocs->push_back(rva);
276 } else if (type == 0) { // IMAGE_REL_BASED_ABSOLUTE
277 // Ignore, used as padding.
278 } else {
279 // Does not occur in Windows x86 executables.
280 return Bad("unknown type of reloc");
284 block += size;
287 std::sort(relocs->begin(), relocs->end());
289 return true;
292 const Section* DisassemblerWin32X86::RVAToSection(RVA rva) const {
293 for (int i = 0; i < number_of_sections_; i++) {
294 const Section* section = &sections_[i];
295 uint32 offset = rva - section->virtual_address;
296 if (offset < section->virtual_size) {
297 return section;
300 return NULL;
303 int DisassemblerWin32X86::RVAToFileOffset(RVA rva) const {
304 const Section* section = RVAToSection(rva);
305 if (section) {
306 uint32 offset = rva - section->virtual_address;
307 if (offset < section->size_of_raw_data) {
308 return section->file_offset_of_raw_data + offset;
309 } else {
310 return kNoOffset; // In section but not in file (e.g. uninit data).
314 // Small RVA values point into the file header in the loaded image.
315 // RVA 0 is the module load address which Windows uses as the module handle.
316 // RVA 2 sometimes occurs, I'm not sure what it is, but it would map into the
317 // DOS header.
318 if (rva == 0 || rva == 2)
319 return rva;
321 NOTREACHED();
322 return kNoOffset;
325 const uint8* DisassemblerWin32X86::RVAToPointer(RVA rva) const {
326 int file_offset = RVAToFileOffset(rva);
327 if (file_offset == kNoOffset)
328 return NULL;
329 else
330 return OffsetToPointer(file_offset);
333 std::string DisassemblerWin32X86::SectionName(const Section* section) {
334 if (section == NULL)
335 return "<none>";
336 char name[9];
337 memcpy(name, section->name, 8);
338 name[8] = '\0'; // Ensure termination.
339 return name;
342 CheckBool DisassemblerWin32X86::ParseFile(AssemblyProgram* program) {
343 // Walk all the bytes in the file, whether or not in a section.
344 uint32 file_offset = 0;
345 while (file_offset < length()) {
346 const Section* section = FindNextSection(file_offset);
347 if (section == NULL) {
348 // No more sections. There should not be extra stuff following last
349 // section.
350 // ParseNonSectionFileRegion(file_offset, pe_info().length(), program);
351 break;
353 if (file_offset < section->file_offset_of_raw_data) {
354 uint32 section_start_offset = section->file_offset_of_raw_data;
355 if(!ParseNonSectionFileRegion(file_offset, section_start_offset,
356 program))
357 return false;
359 file_offset = section_start_offset;
361 uint32 end = file_offset + section->size_of_raw_data;
362 if (!ParseFileRegion(section, file_offset, end, program))
363 return false;
364 file_offset = end;
367 #if COURGETTE_HISTOGRAM_TARGETS
368 HistogramTargets("abs32 relocs", abs32_target_rvas_);
369 HistogramTargets("rel32 relocs", rel32_target_rvas_);
370 #endif
372 return true;
375 bool DisassemblerWin32X86::ParseAbs32Relocs() {
376 abs32_locations_.clear();
377 if (!ParseRelocs(&abs32_locations_))
378 return false;
380 std::sort(abs32_locations_.begin(), abs32_locations_.end());
382 #if COURGETTE_HISTOGRAM_TARGETS
383 for (size_t i = 0; i < abs32_locations_.size(); ++i) {
384 RVA rva = abs32_locations_[i];
385 // The 4 bytes at the relocation are a reference to some address.
386 uint32 target_address = Read32LittleEndian(RVAToPointer(rva));
387 ++abs32_target_rvas_[target_address - image_base()];
389 #endif
390 return true;
393 void DisassemblerWin32X86::ParseRel32RelocsFromSections() {
394 uint32 file_offset = 0;
395 while (file_offset < length()) {
396 const Section* section = FindNextSection(file_offset);
397 if (section == NULL)
398 break;
399 if (file_offset < section->file_offset_of_raw_data)
400 file_offset = section->file_offset_of_raw_data;
401 ParseRel32RelocsFromSection(section);
402 file_offset += section->size_of_raw_data;
404 std::sort(rel32_locations_.begin(), rel32_locations_.end());
406 #if COURGETTE_HISTOGRAM_TARGETS
407 VLOG(1) << "abs32_locations_ " << abs32_locations_.size()
408 << "\nrel32_locations_ " << rel32_locations_.size()
409 << "\nabs32_target_rvas_ " << abs32_target_rvas_.size()
410 << "\nrel32_target_rvas_ " << rel32_target_rvas_.size();
412 int common = 0;
413 std::map<RVA, int>::iterator abs32_iter = abs32_target_rvas_.begin();
414 std::map<RVA, int>::iterator rel32_iter = rel32_target_rvas_.begin();
415 while (abs32_iter != abs32_target_rvas_.end() &&
416 rel32_iter != rel32_target_rvas_.end()) {
417 if (abs32_iter->first < rel32_iter->first)
418 ++abs32_iter;
419 else if (rel32_iter->first < abs32_iter->first)
420 ++rel32_iter;
421 else {
422 ++common;
423 ++abs32_iter;
424 ++rel32_iter;
427 VLOG(1) << "common " << common;
428 #endif
431 void DisassemblerWin32X86::ParseRel32RelocsFromSection(const Section* section) {
432 // TODO(sra): use characteristic.
433 bool isCode = strcmp(section->name, ".text") == 0;
434 if (!isCode)
435 return;
437 uint32 start_file_offset = section->file_offset_of_raw_data;
438 uint32 end_file_offset = start_file_offset + section->size_of_raw_data;
439 RVA relocs_start_rva = base_relocation_table().address_;
441 const uint8* start_pointer = OffsetToPointer(start_file_offset);
442 const uint8* end_pointer = OffsetToPointer(end_file_offset);
444 RVA start_rva = FileOffsetToRVA(start_file_offset);
445 RVA end_rva = start_rva + section->virtual_size;
447 // Quick way to convert from Pointer to RVA within a single Section is to
448 // subtract 'pointer_to_rva'.
449 const uint8* const adjust_pointer_to_rva = start_pointer - start_rva;
451 std::vector<RVA>::iterator abs32_pos = abs32_locations_.begin();
453 // Find the rel32 relocations.
454 const uint8* p = start_pointer;
455 while (p < end_pointer) {
456 RVA current_rva = static_cast<RVA>(p - adjust_pointer_to_rva);
457 if (current_rva == relocs_start_rva) {
458 uint32 relocs_size = base_relocation_table().size_;
459 if (relocs_size) {
460 p += relocs_size;
461 continue;
465 //while (abs32_pos != abs32_locations_.end() && *abs32_pos < current_rva)
466 // ++abs32_pos;
468 // Heuristic discovery of rel32 locations in instruction stream: are the
469 // next few bytes the start of an instruction containing a rel32
470 // addressing mode?
471 const uint8* rel32 = NULL;
473 if (p + 5 <= end_pointer) {
474 if (*p == 0xE8 || *p == 0xE9) { // jmp rel32 and call rel32
475 rel32 = p + 1;
478 if (p + 6 <= end_pointer) {
479 if (*p == 0x0F && (*(p+1) & 0xF0) == 0x80) { // Jcc long form
480 if (p[1] != 0x8A && p[1] != 0x8B) // JPE/JPO unlikely
481 rel32 = p + 2;
484 if (rel32) {
485 RVA rel32_rva = static_cast<RVA>(rel32 - adjust_pointer_to_rva);
487 // Is there an abs32 reloc overlapping the candidate?
488 while (abs32_pos != abs32_locations_.end() && *abs32_pos < rel32_rva - 3)
489 ++abs32_pos;
490 // Now: (*abs32_pos > rel32_rva - 4) i.e. the lowest addressed 4-byte
491 // region that could overlap rel32_rva.
492 if (abs32_pos != abs32_locations_.end()) {
493 if (*abs32_pos < rel32_rva + 4) {
494 // Beginning of abs32 reloc is before end of rel32 reloc so they
495 // overlap. Skip four bytes past the abs32 reloc.
496 p += (*abs32_pos + 4) - current_rva;
497 continue;
501 RVA target_rva = rel32_rva + 4 + Read32LittleEndian(rel32);
502 // To be valid, rel32 target must be within image, and within this
503 // section.
504 if (IsValidRVA(target_rva) &&
505 start_rva <= target_rva && target_rva < end_rva) {
506 rel32_locations_.push_back(rel32_rva);
507 #if COURGETTE_HISTOGRAM_TARGETS
508 ++rel32_target_rvas_[target_rva];
509 #endif
510 p = rel32 + 4;
511 continue;
514 p += 1;
518 CheckBool DisassemblerWin32X86::ParseNonSectionFileRegion(
519 uint32 start_file_offset,
520 uint32 end_file_offset,
521 AssemblyProgram* program) {
522 if (incomplete_disassembly_)
523 return true;
525 if (end_file_offset > start_file_offset) {
526 if (!program->EmitBytesInstruction(OffsetToPointer(start_file_offset),
527 end_file_offset - start_file_offset)) {
528 return false;
532 return true;
535 CheckBool DisassemblerWin32X86::ParseFileRegion(
536 const Section* section,
537 uint32 start_file_offset, uint32 end_file_offset,
538 AssemblyProgram* program) {
539 RVA relocs_start_rva = base_relocation_table().address_;
541 const uint8* start_pointer = OffsetToPointer(start_file_offset);
542 const uint8* end_pointer = OffsetToPointer(end_file_offset);
544 RVA start_rva = FileOffsetToRVA(start_file_offset);
545 RVA end_rva = start_rva + section->virtual_size;
547 // Quick way to convert from Pointer to RVA within a single Section is to
548 // subtract 'pointer_to_rva'.
549 const uint8* const adjust_pointer_to_rva = start_pointer - start_rva;
551 std::vector<RVA>::iterator rel32_pos = rel32_locations_.begin();
552 std::vector<RVA>::iterator abs32_pos = abs32_locations_.begin();
554 if (!program->EmitOriginInstruction(start_rva))
555 return false;
557 const uint8* p = start_pointer;
559 while (p < end_pointer) {
560 RVA current_rva = static_cast<RVA>(p - adjust_pointer_to_rva);
562 // The base relocation table is usually in the .relocs section, but it could
563 // actually be anywhere. Make sure we skip it because we will regenerate it
564 // during assembly.
565 if (current_rva == relocs_start_rva) {
566 if (!program->EmitPeRelocsInstruction())
567 return false;
568 uint32 relocs_size = base_relocation_table().size_;
569 if (relocs_size) {
570 p += relocs_size;
571 continue;
575 while (abs32_pos != abs32_locations_.end() && *abs32_pos < current_rva)
576 ++abs32_pos;
578 if (abs32_pos != abs32_locations_.end() && *abs32_pos == current_rva) {
579 uint32 target_address = Read32LittleEndian(p);
580 RVA target_rva = target_address - image_base();
581 // TODO(sra): target could be Label+offset. It is not clear how to guess
582 // which it might be. We assume offset==0.
583 if (!program->EmitAbs32(program->FindOrMakeAbs32Label(target_rva)))
584 return false;
585 p += 4;
586 continue;
589 while (rel32_pos != rel32_locations_.end() && *rel32_pos < current_rva)
590 ++rel32_pos;
592 if (rel32_pos != rel32_locations_.end() && *rel32_pos == current_rva) {
593 RVA target_rva = current_rva + 4 + Read32LittleEndian(p);
594 if (!program->EmitRel32(program->FindOrMakeRel32Label(target_rva)))
595 return false;
596 p += 4;
597 continue;
600 if (incomplete_disassembly_) {
601 if ((abs32_pos == abs32_locations_.end() || end_rva <= *abs32_pos) &&
602 (rel32_pos == rel32_locations_.end() || end_rva <= *rel32_pos) &&
603 (end_rva <= relocs_start_rva || current_rva >= relocs_start_rva)) {
604 // No more relocs in this section, don't bother encoding bytes.
605 break;
609 if (!program->EmitByteInstruction(*p))
610 return false;
611 p += 1;
614 return true;
617 #if COURGETTE_HISTOGRAM_TARGETS
618 // Histogram is printed to std::cout. It is purely for debugging the algorithm
619 // and is only enabled manually in 'exploration' builds. I don't want to add
620 // command-line configuration for this feature because this code has to be
621 // small, which means compiled-out.
622 void DisassemblerWin32X86::HistogramTargets(const char* kind,
623 const std::map<RVA, int>& map) {
624 int total = 0;
625 std::map<int, std::vector<RVA> > h;
626 for (std::map<RVA, int>::const_iterator p = map.begin();
627 p != map.end();
628 ++p) {
629 h[p->second].push_back(p->first);
630 total += p->second;
633 std::cout << total << " " << kind << " to "
634 << map.size() << " unique targets" << std::endl;
636 std::cout << "indegree: #targets-with-indegree (example)" << std::endl;
637 const int kFirstN = 15;
638 bool someSkipped = false;
639 int index = 0;
640 for (std::map<int, std::vector<RVA> >::reverse_iterator p = h.rbegin();
641 p != h.rend();
642 ++p) {
643 ++index;
644 if (index <= kFirstN || p->first <= 3) {
645 if (someSkipped) {
646 std::cout << "..." << std::endl;
648 size_t count = p->second.size();
649 std::cout << std::dec << p->first << ": " << count;
650 if (count <= 2) {
651 for (size_t i = 0; i < count; ++i)
652 std::cout << " " << DescribeRVA(p->second[i]);
654 std::cout << std::endl;
655 someSkipped = false;
656 } else {
657 someSkipped = true;
661 #endif // COURGETTE_HISTOGRAM_TARGETS
664 // DescribeRVA is for debugging only. I would put it under #ifdef DEBUG except
665 // that during development I'm finding I need to call it when compiled in
666 // Release mode. Hence:
667 // TODO(sra): make this compile only for debug mode.
668 std::string DisassemblerWin32X86::DescribeRVA(RVA rva) const {
669 const Section* section = RVAToSection(rva);
670 std::ostringstream s;
671 s << std::hex << rva;
672 if (section) {
673 s << " (";
674 s << SectionName(section) << "+"
675 << std::hex << (rva - section->virtual_address)
676 << ")";
678 return s.str();
681 const Section* DisassemblerWin32X86::FindNextSection(uint32 fileOffset) const {
682 const Section* best = 0;
683 for (int i = 0; i < number_of_sections_; i++) {
684 const Section* section = &sections_[i];
685 if (section->size_of_raw_data > 0) { // i.e. has data in file.
686 if (fileOffset <= section->file_offset_of_raw_data) {
687 if (best == 0 ||
688 section->file_offset_of_raw_data < best->file_offset_of_raw_data) {
689 best = section;
694 return best;
697 RVA DisassemblerWin32X86::FileOffsetToRVA(uint32 file_offset) const {
698 for (int i = 0; i < number_of_sections_; i++) {
699 const Section* section = &sections_[i];
700 uint32 offset = file_offset - section->file_offset_of_raw_data;
701 if (offset < section->size_of_raw_data) {
702 return section->virtual_address + offset;
705 return 0;
708 bool DisassemblerWin32X86::ReadDataDirectory(
709 int index,
710 ImageDataDirectory* directory) {
712 if (index < number_of_data_directories_) {
713 size_t offset = index * 8 + offset_of_data_directories_;
714 if (offset >= size_of_optional_header_)
715 return Bad("number of data directories inconsistent");
716 const uint8* data_directory = optional_header_ + offset;
717 if (data_directory < start() ||
718 data_directory + 8 >= end())
719 return Bad("data directory outside image");
720 RVA rva = ReadU32(data_directory, 0);
721 size_t size = ReadU32(data_directory, 4);
722 if (size > size_of_image_)
723 return Bad("data directory size too big");
725 // TODO(sra): validate RVA.
726 directory->address_ = rva;
727 directory->size_ = static_cast<uint32>(size);
728 return true;
729 } else {
730 directory->address_ = 0;
731 directory->size_ = 0;
732 return true;
736 } // namespace courgette