1 : // Copyright 2012 Google Inc. All Rights Reserved.
2 : //
3 : // Licensed under the Apache License, Version 2.0 (the "License");
4 : // you may not use this file except in compliance with the License.
5 : // You may obtain a copy of the License at
6 : //
7 : // http://www.apache.org/licenses/LICENSE-2.0
8 : //
9 : // Unless required by applicable law or agreed to in writing, software
10 : // distributed under the License is distributed on an "AS IS" BASIS,
11 : // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 : // See the License for the specific language governing permissions and
13 : // limitations under the License.
14 :
15 : #include "syzygy/pe/new_decomposer.h"
16 :
17 : #include "pcrecpp.h" // NOLINT
18 : #include "base/bind.h"
19 : #include "base/stringprintf.h"
20 : #include "base/utf_string_conversions.h"
21 : #include "base/strings/string_split.h"
22 : #include "base/win/scoped_bstr.h"
23 : #include "base/win/scoped_comptr.h"
24 : #include "syzygy/core/disassembler_util.h"
25 : #include "syzygy/core/zstream.h"
26 : #include "syzygy/pdb/omap.h"
27 : #include "syzygy/pdb/pdb_byte_stream.h"
28 : #include "syzygy/pdb/pdb_constants.h"
29 : #include "syzygy/pdb/pdb_dbi_stream.h"
30 : #include "syzygy/pdb/pdb_file.h"
31 : #include "syzygy/pdb/pdb_reader.h"
32 : #include "syzygy/pdb/pdb_symbol_record.h"
33 : #include "syzygy/pdb/pdb_util.h"
34 : #include "syzygy/pe/dia_util.h"
35 : #include "syzygy/pe/find.h"
36 : #include "syzygy/pe/pe_file_parser.h"
37 : #include "syzygy/pe/pe_utils.h"
38 : #include "syzygy/pe/serialization.h"
39 : #include "third_party/cci/Files/CvInfo.h"
40 :
41 : namespace cci = Microsoft_Cci_Pdb;
42 :
43 : namespace {
44 :
45 : // A small helper struct for dumping block information to log messages.
46 : // TODO(chrisha): Move this to block_graph and reuse it everywhere!
47 : struct BlockInfo {
48 : enum AddressType {
49 : kNoAddress,
50 : kAbsoluteAddress,
51 : kFileOffsetAddress,
52 : kRelativeAddress,
53 : };
54 :
55 i : explicit BlockInfo(const block_graph::BlockGraph::Block* block)
56 : : block(block), type(kNoAddress) {
57 i : DCHECK(block != NULL);
58 i : }
59 :
60 i : BlockInfo(const block_graph::BlockGraph::Block* block,
61 : core::AbsoluteAddress address)
62 : : block(block), type(kAbsoluteAddress), abs_addr(address) {
63 i : DCHECK(block != NULL);
64 i : }
65 : BlockInfo(const block_graph::BlockGraph::Block* block,
66 : core::FileOffsetAddress address)
67 : : block(block), type(kFileOffsetAddress), file_addr(address) {
68 : DCHECK(block != NULL);
69 : }
70 : BlockInfo(const block_graph::BlockGraph::Block* block,
71 : core::RelativeAddress address)
72 : : block(block), type(kRelativeAddress), rel_addr(address) {
73 : DCHECK(block != NULL);
74 : }
75 :
76 : const block_graph::BlockGraph::Block* block;
77 : AddressType type;
78 :
79 : // Ideally these would be in a union but because they have non-trivial
80 : // constructors they are not allowed.
81 : core::AbsoluteAddress abs_addr;
82 : core::FileOffsetAddress file_addr;
83 : core::RelativeAddress rel_addr;
84 :
85 : private:
86 : DISALLOW_COPY_AND_ASSIGN(BlockInfo);
87 : };
88 :
89 : } // anonymous namespace
90 :
91 : // Pretty prints a BlockInfo to an ostream. This has to be outside of any
92 : // namespaces so that operator<< is found properly.
93 i : std::ostream& operator<<(std::ostream& os, const BlockInfo& bi) {
94 : os << "Block(id=" << bi.block->id() << ", name=\"" << bi.block->name()
95 i : << "\", size=" << bi.block->size();
96 i : if (bi.type != BlockInfo::kNoAddress) {
97 i : os << ", address=";
98 i : switch (bi.type) {
99 : case BlockInfo::kAbsoluteAddress: {
100 i : os << bi.abs_addr;
101 i : break;
102 : }
103 : case BlockInfo::kFileOffsetAddress: {
104 i : os << bi.file_addr;
105 i : break;
106 : }
107 : case BlockInfo::kRelativeAddress: {
108 i : os << bi.rel_addr;
109 : break;
110 : }
111 : default: break;
112 : }
113 : }
114 i : os << ")";
115 i : return os;
116 i : }
117 :
118 : namespace pe {
119 :
120 : // An intermediate reference representation used while parsing PE blocks.
121 : // This is necessary because at that point we haven't yet chunked the whole
122 : // image into blocks thus some references cannot be resolved.
123 : struct NewDecomposer::IntermediateReference {
124 : RelativeAddress src_addr;
125 : BlockGraph::ReferenceType type;
126 : BlockGraph::Size size;
127 : RelativeAddress dst_addr;
128 : };
129 :
130 : namespace {
131 :
132 : using base::win::ScopedBstr;
133 : using base::win::ScopedComPtr;
134 : using block_graph::BlockGraph;
135 : using builder::Callback;
136 : using builder::Opt;
137 : using builder::Or;
138 : using builder::Seq;
139 : using builder::Star;
140 : using core::AbsoluteAddress;
141 : using core::RelativeAddress;
142 :
143 : typedef BlockGraph::Block Block;
144 : typedef BlockGraph::BlockType BlockType;
145 : typedef BlockGraph::Offset Offset;
146 : typedef BlockGraph::Reference Reference;
147 : typedef BlockGraph::ReferenceType ReferenceType;
148 : typedef core::AddressRange<RelativeAddress, size_t> RelativeRange;
149 : typedef NewDecomposer::IntermediateReference IntermediateReference;
150 : typedef NewDecomposer::IntermediateReferences IntermediateReferences;
151 : typedef pcrecpp::RE RE;
152 : typedef std::vector<OMAP> OMAPs;
153 : typedef std::vector<pdb::PdbFixup> PdbFixups;
154 :
155 : const char kJumpTable[] = "<jump-table>";
156 : const char kCaseTable[] = "<case-table>";
157 :
158 : // Some helper functions for testing ranges.
159 : template<typename T1, typename T2, typename T3>
160 E : bool InRange(T1 value, T2 lower_bound_incl, T3 length_excl) {
161 E : T1 upper_bound_excl = static_cast<T1>(lower_bound_incl) + length_excl;
162 : return static_cast<T1>(lower_bound_incl) <= value &&
163 E : value < static_cast<T2>(upper_bound_excl);
164 E : }
165 : template<typename T1, typename T2, typename T3>
166 E : bool InRangeIncl(T1 value, T2 lower_bound_incl, T3 length_incl) {
167 E : T1 upper_bound_incl = static_cast<T1>(lower_bound_incl) + length_incl;
168 : return static_cast<T1>(lower_bound_incl) <= value &&
169 E : value <= upper_bound_incl;
170 E : }
171 :
172 : bool InitializeDia(const PEFile& image_file,
173 : const base::FilePath& pdb_path,
174 : IDiaDataSource** dia_source,
175 : IDiaSession** dia_session,
176 E : IDiaSymbol** global) {
177 E : DCHECK(*dia_source == NULL);
178 E : DCHECK(*dia_session == NULL);
179 E : DCHECK(*global == NULL);
180 :
181 E : if (!CreateDiaSource(dia_source))
182 i : return false;
183 E : DCHECK(*dia_source != NULL);
184 :
185 : // We create the session using the PDB file directly, as we've already
186 : // validated that it matches the module.
187 E : if (!CreateDiaSession(pdb_path, *dia_source, dia_session))
188 i : return false;
189 E : DCHECK(*dia_session != NULL);
190 :
191 E : HRESULT hr = (*dia_session)->get_globalScope(global);
192 E : if (hr != S_OK) {
193 i : LOG(ERROR) << "Failed to get the DIA global scope: "
194 : << com::LogHr(hr) << ".";
195 i : return false;
196 : }
197 :
198 E : return true;
199 E : }
200 :
201 : enum SectionType {
202 : kSectionCode,
203 : kSectionData,
204 : kSectionUnknown
205 : };
206 :
207 : // Determines the type of a section based on its attributes. This is used to
208 : // tag blocks with an appropriate type.
209 E : SectionType GetSectionType(const IMAGE_SECTION_HEADER* header) {
210 E : DCHECK(header != NULL);
211 E : if ((header->Characteristics & IMAGE_SCN_CNT_CODE) != 0)
212 E : return kSectionCode;
213 E : if ((header->Characteristics & kReadOnlyDataCharacteristics) != 0)
214 E : return kSectionData;
215 i : return kSectionUnknown;
216 E : }
217 :
218 : // Given a compiland, returns its compiland details.
219 : bool GetCompilandDetailsForCompiland(IDiaSymbol* compiland,
220 E : IDiaSymbol** compiland_details) {
221 E : DCHECK(compiland != NULL);
222 E : DCHECK(compiland_details != NULL);
223 E : DCHECK(IsSymTag(compiland, SymTagCompiland));
224 E : DCHECK(*compiland_details == NULL);
225 :
226 : // Get the enumeration of compiland details.
227 E : ScopedComPtr<IDiaEnumSymbols> enum_symbols;
228 : HRESULT hr = compiland->findChildren(SymTagCompilandDetails, NULL, 0,
229 E : enum_symbols.Receive());
230 E : DCHECK_EQ(S_OK, hr);
231 :
232 : // We expect there to be compiland details. For compilands built by
233 : // non-standard toolchains, there usually aren't any.
234 E : LONG count = 0;
235 E : hr = enum_symbols->get_Count(&count);
236 E : DCHECK_EQ(S_OK, hr);
237 E : if (count == 0) {
238 : // We don't log here because we see this quite often.
239 i : return false;
240 : }
241 :
242 : // We do sometimes encounter more than one compiland detail. In fact, for
243 : // import and export tables we get one compiland detail per table entry.
244 : // They are all marked as having been generated by the linker, so using the
245 : // first one is sufficient.
246 :
247 : // Get the compiland details.
248 E : ULONG fetched = 0;
249 E : hr = enum_symbols->Next(1, compiland_details, &fetched);
250 E : DCHECK_EQ(S_OK, hr);
251 E : DCHECK_EQ(1u, fetched);
252 :
253 E : return true;
254 E : }
255 :
256 : // Stores information regarding known compilers.
257 : struct KnownCompilerInfo {
258 : wchar_t* compiler_name;
259 : bool supported;
260 : };
261 :
262 : // A list of known compilers, and their status as being supported or not.
263 : KnownCompilerInfo kKnownCompilerInfos[] = {
264 : { L"Microsoft (R) Macro Assembler", false },
265 : { L"Microsoft (R) Optimizing Compiler", true },
266 : { L"Microsoft (R) LINK", false }
267 : };
268 :
269 : // Given a compiland, determines whether the compiler used is one of those that
270 : // we whitelist.
271 E : bool IsBuiltBySupportedCompiler(IDiaSymbol* compiland) {
272 E : DCHECK(compiland != NULL);
273 E : DCHECK(IsSymTag(compiland, SymTagCompiland));
274 :
275 E : ScopedComPtr<IDiaSymbol> compiland_details;
276 : if (!GetCompilandDetailsForCompiland(compiland,
277 E : compiland_details.Receive())) {
278 : // If the compiland has no compiland details we assume the compiler is not
279 : // supported.
280 i : ScopedBstr compiland_name;
281 i : if (compiland->get_name(compiland_name.Receive()) == S_OK) {
282 i : VLOG(1) << "Compiland has no compiland details: "
283 : << com::ToString(compiland_name);
284 : }
285 i : return false;
286 : }
287 E : DCHECK(compiland_details.get() != NULL);
288 :
289 : // Get the compiler name.
290 E : ScopedBstr compiler_name;
291 E : HRESULT hr = compiland_details->get_compilerName(compiler_name.Receive());
292 E : DCHECK_EQ(S_OK, hr);
293 :
294 : // Check the compiler name against the list of known compilers.
295 E : for (size_t i = 0; i < arraysize(kKnownCompilerInfos); ++i) {
296 E : if (::wcscmp(kKnownCompilerInfos[i].compiler_name, compiler_name) == 0) {
297 E : return kKnownCompilerInfos[i].supported;
298 : }
299 E : }
300 :
301 : // Anything we don't explicitly know about is not supported.
302 E : VLOG(1) << "Encountered unknown compiler: " << compiler_name;
303 E : return false;
304 E : }
305 :
306 : // Adds an intermediate reference to the provided vector. The vector is
307 : // specified as the first parameter (in slight violation of our coding
308 : // standards) because this function is intended to be used by Bind.
309 : bool AddIntermediateReference(IntermediateReferences* references,
310 : RelativeAddress src_addr,
311 : ReferenceType type,
312 : BlockGraph::Size size,
313 E : RelativeAddress dst_addr) {
314 E : DCHECK(references != NULL);
315 E : IntermediateReference ref = { src_addr, type, size, dst_addr };
316 E : references->push_back(ref);
317 E : return true;
318 E : }
319 :
320 : // Create a reference as specified. Ignores existing references if they are of
321 : // the exact same type.
322 : bool CreateReference(RelativeAddress src_addr,
323 : BlockGraph::Size ref_size,
324 : ReferenceType ref_type,
325 : RelativeAddress base_addr,
326 : RelativeAddress dst_addr,
327 E : BlockGraph::AddressSpace* image) {
328 E : DCHECK(image != NULL);
329 :
330 : // Get the source block and offset, and ensure that the reference fits
331 : // within it.
332 E : Block* src_block = image->GetBlockByAddress(src_addr);
333 E : if (src_block == NULL) {
334 i : LOG(ERROR) << "Unable to find block for reference originating at "
335 : << src_addr << ".";
336 i : return false;
337 : }
338 E : RelativeAddress src_block_addr;
339 E : CHECK(image->GetAddressOf(src_block, &src_block_addr));
340 E : Offset src_block_offset = src_addr - src_block_addr;
341 E : if (src_block_offset + ref_size > src_block->size()) {
342 i : LOG(ERROR) << "Reference originating at " << src_addr
343 : << " extends beyond block \"" << src_block->name() << "\".";
344 i : return false;
345 : }
346 :
347 : // Get the destination block and offset.
348 E : Block* dst_block = image->GetBlockByAddress(base_addr);
349 E : if (dst_block == NULL) {
350 i : LOG(ERROR) << "Unable to find block for reference pointing at "
351 : << base_addr << ".";
352 i : return false;
353 : }
354 E : RelativeAddress dst_block_addr;
355 E : CHECK(image->GetAddressOf(dst_block, &dst_block_addr));
356 E : Offset base = base_addr - dst_block_addr;
357 E : Offset offset = dst_addr - dst_block_addr;
358 :
359 E : Reference ref(ref_type, ref_size, dst_block, offset, base);
360 :
361 : // Check if a reference already exists at this offset.
362 : Block::ReferenceMap::const_iterator ref_it =
363 E : src_block->references().find(src_block_offset);
364 E : if (ref_it != src_block->references().end()) {
365 : // If an identical reference already exists then we're done.
366 E : if (ref == ref_it->second)
367 E : return true;
368 i : LOG(ERROR) << "Block \"" << src_block->name() << "\" has a conflicting "
369 : << "reference at offset " << src_block_offset << ".";
370 i : return false;
371 : }
372 :
373 E : CHECK(src_block->SetReference(src_block_offset, ref));
374 :
375 E : return true;
376 E : }
377 :
378 : // Loads FIXUP and OMAP_FROM debug streams.
379 : bool LoadDebugStreams(IDiaSession* dia_session,
380 : PdbFixups* pdb_fixups,
381 E : OMAPs* omap_from) {
382 E : DCHECK(dia_session != NULL);
383 E : DCHECK(pdb_fixups != NULL);
384 E : DCHECK(omap_from != NULL);
385 :
386 : // Load the fixups. These must exist.
387 : SearchResult search_result = FindAndLoadDiaDebugStreamByName(
388 E : kFixupDiaDebugStreamName, dia_session, pdb_fixups);
389 E : if (search_result != kSearchSucceeded) {
390 i : if (search_result == kSearchFailed) {
391 i : LOG(ERROR) << "PDB file does not contain a FIXUP stream. Module must be "
392 : "linked with '/PROFILE' or '/DEBUGINFO:FIXUP' flag.";
393 : }
394 i : return false;
395 : }
396 :
397 : // Load the omap_from table. It is not necessary that one exist.
398 : search_result = FindAndLoadDiaDebugStreamByName(
399 E : kOmapFromDiaDebugStreamName, dia_session, omap_from);
400 E : if (search_result == kSearchErrored) {
401 i : LOG(ERROR) << "Error trying to read " << kOmapFromDiaDebugStreamName
402 : << " stream.";
403 i : return false;
404 : }
405 :
406 E : return true;
407 E : }
408 :
409 : bool GetFixupDestinationAndType(const PEFile& image_file,
410 : const pdb::PdbFixup& fixup,
411 : RelativeAddress* dst_addr,
412 E : ReferenceType* ref_type) {
413 E : DCHECK(dst_addr != NULL);
414 E : DCHECK(ref_type != NULL);
415 :
416 E : RelativeAddress src_addr(fixup.rva_location);
417 :
418 : // Get the destination displacement from the actual image itself. We only see
419 : // fixups for 32-bit references.
420 E : uint32 data = 0;
421 E : if (!image_file.ReadImage(src_addr, &data, sizeof(data))) {
422 i : LOG(ERROR) << "Unable to read image data for fixup with source address "
423 : << "at" << src_addr << ".";
424 i : return false;
425 : }
426 :
427 : // Translate this to a relative displacement value.
428 E : switch (fixup.type) {
429 : case pdb::PdbFixup::TYPE_ABSOLUTE: {
430 E : *ref_type = BlockGraph::ABSOLUTE_REF;
431 E : *dst_addr = RelativeAddress(image_file.AbsToRelDisplacement(data));
432 E : break;
433 : }
434 :
435 : case pdb::PdbFixup::TYPE_PC_RELATIVE: {
436 E : *ref_type = BlockGraph::PC_RELATIVE_REF;
437 E : *dst_addr = RelativeAddress(fixup.rva_location) + sizeof(data) + data;
438 E : break;
439 : }
440 :
441 : case pdb::PdbFixup::TYPE_RELATIVE: {
442 E : *ref_type = BlockGraph::RELATIVE_REF;
443 E : *dst_addr = RelativeAddress(data);
444 E : break;
445 : }
446 :
447 : default: {
448 i : LOG(ERROR) << "Unexpected fixup type (" << fixup.type << ").";
449 i : return false;
450 : }
451 : }
452 :
453 E : return true;
454 E : }
455 :
456 : // Creates references from the @p pdb_fixups (translating them via the
457 : // provided @p omap_from information if it is not empty), all while removing the
458 : // corresponding entries from @p reloc_set. If @p reloc_set is not empty after
459 : // this then the PDB fixups are out of sync with the image and we are unable to
460 : // safely decompose.
461 : //
462 : // @note This function deliberately ignores fixup information for the resource
463 : // section. This is because chrome.dll gets modified by a manifest tool
464 : // which doesn't update the FIXUPs in the corresponding PDB. They are thus
465 : // out of sync. Even if they were in sync this doesn't harm us as we have no
466 : // need to reach in and modify resource data.
467 : bool CreateReferencesFromFixupsImpl(
468 : const PEFile& image_file,
469 : const PdbFixups& pdb_fixups,
470 : const OMAPs& omap_from,
471 : PEFile::RelocSet* reloc_set,
472 E : BlockGraph::AddressSpace* image) {
473 E : DCHECK(reloc_set != NULL);
474 E : DCHECK(image != NULL);
475 :
476 E : bool have_omap = !omap_from.empty();
477 E : size_t fixups_used = 0;
478 :
479 : // The resource section in Chrome is modified post-link by a tool that adds a
480 : // manifest to it. This causes all of the fixups in the resource section (and
481 : // anything beyond it) to be invalid. As long as the resource section is the
482 : // last section in the image, this is not a problem (we can safely ignore the
483 : // .rsrc fixups, which we know how to parse without them). However, if there
484 : // is a section after the resource section, things will have been shifted
485 : // and potentially crucial fixups will be invalid.
486 : const IMAGE_SECTION_HEADER* rsrc_header = image_file.GetSectionHeader(
487 E : kResourceSectionName);
488 E : RelativeAddress rsrc_start(0xffffffff);
489 E : RelativeAddress rsrc_end(0xffffffff);
490 E : if (rsrc_header != NULL) {
491 E : rsrc_start = RelativeAddress(rsrc_header->VirtualAddress);
492 E : rsrc_end = rsrc_start + rsrc_header->Misc.VirtualSize;
493 : }
494 :
495 : // Ensure the fixups are all valid.
496 E : for (size_t i = 0; i < pdb_fixups.size(); ++i) {
497 E : if (!pdb_fixups[i].ValidHeader()) {
498 i : LOG(ERROR) << "Unknown fixup header: "
499 : << base::StringPrintf("0x%08X.", pdb_fixups[i].header);
500 i : return false;
501 : }
502 :
503 : // For now, we skip any offset fixups. We've only seen this in the context
504 : // of TLS data access, and we don't mess with TLS structures.
505 E : if (pdb_fixups[i].is_offset())
506 E : continue;
507 :
508 : // All fixups we handle should be full size pointers.
509 E : DCHECK_EQ(Reference::kMaximumSize, pdb_fixups[i].size());
510 :
511 : // Get the original addresses, and map them through OMAP information.
512 : // Normally DIA takes care of this for us, but there is no API for
513 : // getting DIA to give us FIXUP information, so we have to do it manually.
514 E : RelativeAddress src_addr(pdb_fixups[i].rva_location);
515 E : RelativeAddress base_addr(pdb_fixups[i].rva_base);
516 E : if (have_omap) {
517 i : src_addr = pdb::TranslateAddressViaOmap(omap_from, src_addr);
518 i : base_addr = pdb::TranslateAddressViaOmap(omap_from, base_addr);
519 : }
520 :
521 : // If the reference originates beyond the .rsrc section then we can't
522 : // trust it.
523 E : if (src_addr >= rsrc_end) {
524 i : LOG(ERROR) << "Found fixup originating beyond .rsrc section.";
525 i : return false;
526 : }
527 :
528 : // If the reference originates from a part of the .rsrc section, ignore it.
529 E : if (src_addr >= rsrc_start)
530 E : continue;
531 :
532 : // Get the relative address/displacement of the fixup. This logs on failure.
533 E : RelativeAddress dst_addr;
534 E : ReferenceType type = BlockGraph::RELATIVE_REF;
535 : if (!GetFixupDestinationAndType(image_file, pdb_fixups[i], &dst_addr,
536 E : &type)) {
537 i : return false;
538 : }
539 :
540 : // Finally, create the reference. This logs verbosely for us on failure.
541 : if (!CreateReference(src_addr, Reference::kMaximumSize, type, base_addr,
542 E : dst_addr, image)) {
543 i : return false;
544 : }
545 :
546 : // Remove this reference from the relocs.
547 E : PEFile::RelocSet::iterator reloc_it = reloc_set->find(src_addr);
548 E : if (reloc_it != reloc_set->end()) {
549 : // We should only find a reloc if the fixup was of absolute type.
550 E : if (type != BlockGraph::ABSOLUTE_REF) {
551 i : LOG(ERROR) << "Found a reloc corresponding to a non-absolute fixup.";
552 i : return false;
553 : }
554 :
555 E : reloc_set->erase(reloc_it);
556 : }
557 :
558 E : ++fixups_used;
559 E : }
560 :
561 E : LOG(INFO) << "Used " << fixups_used << " of " << pdb_fixups.size() << ".";
562 :
563 E : return true;
564 E : }
565 :
566 E : bool GetDataSymbolSize(IDiaSymbol* symbol, size_t* length) {
567 E : DCHECK(symbol != NULL);
568 E : DCHECK(length != NULL);
569 :
570 E : *length = 0;
571 E : ScopedComPtr<IDiaSymbol> type;
572 E : HRESULT hr = symbol->get_type(type.Receive());
573 : // This happens if the symbol has no type information.
574 E : if (hr == S_FALSE)
575 E : return true;
576 E : if (hr != S_OK) {
577 i : LOG(ERROR) << "Failed to get type symbol: " << com::LogHr(hr) << ".";
578 i : return false;
579 : }
580 :
581 E : ULONGLONG ull_length = 0;
582 E : hr = type->get_length(&ull_length);
583 E : if (hr != S_OK) {
584 i : LOG(ERROR) << "Failed to retrieve type length properties: "
585 : << com::LogHr(hr) << ".";
586 i : return false;
587 : }
588 E : DCHECK_LE(ull_length, 0xFFFFFFFF);
589 E : *length = static_cast<size_t>(ull_length);
590 :
591 E : return true;
592 E : }
593 :
594 : bool ScopeSymTagToLabelProperties(enum SymTagEnum sym_tag,
595 : size_t scope_count,
596 : BlockGraph::LabelAttributes* attr,
597 E : std::string* name) {
598 E : DCHECK(attr != NULL);
599 E : DCHECK(name != NULL);
600 :
601 E : switch (sym_tag) {
602 : case SymTagFuncDebugStart: {
603 E : *attr = BlockGraph::DEBUG_START_LABEL;
604 E : *name = "<debug-start>";
605 E : return true;
606 : }
607 : case SymTagFuncDebugEnd: {
608 E : *attr = BlockGraph::DEBUG_END_LABEL;
609 E : *name = "<debug-end>";
610 E : return true;
611 : }
612 : case SymTagBlock: {
613 E : *attr = BlockGraph::SCOPE_START_LABEL;
614 E : *name = base::StringPrintf("<scope-start-%d>", scope_count);
615 E : return true;
616 : }
617 : default:
618 i : return false;
619 : }
620 i : return false;
621 E : }
622 :
623 : bool AddLabelToBlock(Offset offset,
624 : const base::StringPiece& name,
625 : BlockGraph::LabelAttributes label_attributes,
626 E : Block* block) {
627 E : DCHECK(block != NULL);
628 :
629 : // It is possible for labels to be attached to the first byte past a block
630 : // (things like debug end, scope end, etc). It is up to the caller to be more
631 : // strict about the offset if need be.
632 E : DCHECK_LE(0, offset);
633 E : DCHECK_LE(offset, static_cast<Offset>(block->size()));
634 :
635 : // Try to create the label.
636 E : if (block->SetLabel(offset, name, label_attributes))
637 E : return true;
638 :
639 : // If we get here there's an already existing label. Update it.
640 E : BlockGraph::Label label;
641 E : CHECK(block->GetLabel(offset, &label));
642 :
643 : // Merge the names if this isn't a repeated name.
644 E : std::string name_str = name.as_string();
645 E : std::string new_name = label.name();
646 E : std::vector<std::string> names;
647 : base::SplitStringUsingSubstr(label.name(), NewDecomposer::kLabelNameSep,
648 E : &names);
649 E : if (std::find(names.begin(), names.end(), name_str) == names.end()) {
650 E : names.push_back(name_str);
651 E : new_name.append(NewDecomposer::kLabelNameSep);
652 E : new_name.append(name_str);
653 : }
654 :
655 : // Merge the attributes.
656 : BlockGraph::LabelAttributes new_label_attr = label.attributes() |
657 E : label_attributes;
658 :
659 : // Update the label.
660 E : label = BlockGraph::Label(new_name, new_label_attr);
661 E : CHECK(block->RemoveLabel(offset));
662 E : CHECK(block->SetLabel(offset, label));
663 :
664 E : return true;
665 E : }
666 :
667 : // Reads the linker module symbol stream from the given PDB file. This should
668 : // always exist as the last module.
669 : scoped_refptr<pdb::PdbStream> GetLinkerSymbolStream(
670 E : const pdb::PdbFile& pdb_file) {
671 : static const char kLinkerModuleName[] = "* Linker *";
672 :
673 : scoped_refptr<pdb::PdbStream> dbi_stream =
674 E : pdb_file.GetStream(pdb::kDbiStream);
675 E : if (dbi_stream.get() == NULL) {
676 i : LOG(ERROR) << "PDB does not contain a DBI stream.";
677 i : return false;
678 : }
679 :
680 E : pdb::DbiStream dbi;
681 E : if (!dbi.Read(dbi_stream.get())) {
682 i : LOG(ERROR) << "Unable to parse DBI stream.";
683 i : return false;
684 : }
685 :
686 E : if (dbi.modules().empty()) {
687 i : LOG(ERROR) << "DBI stream contains no modules.";
688 i : return false;
689 : }
690 :
691 : // The last module has always been observed to be the linker module.
692 E : const pdb::DbiModuleInfo& linker = dbi.modules().back();
693 E : if (linker.module_name() != kLinkerModuleName) {
694 i : LOG(ERROR) << "Last module is not the linker module.";
695 i : return false;
696 : }
697 :
698 : scoped_refptr<pdb::PdbStream> symbols = pdb_file.GetStream(
699 E : linker.module_info_base().stream);
700 E : if (symbols.get() == NULL) {
701 i : LOG(ERROR) << "Unable to open linker symbol stream.";
702 i : return false;
703 : }
704 :
705 E : return symbols;
706 E : }
707 :
708 : // Parses a symbol from a PDB symbol stream. The @p buffer is populated with the
709 : // data and upon success this returns the symbol directly cast onto the
710 : // @p buffer data. On failure this returns NULL.
711 : template<typename SymbolType>
712 : const SymbolType* ParseSymbol(uint16 symbol_length,
713 : pdb::PdbStream* stream,
714 E : std::vector<uint8>* buffer) {
715 E : DCHECK(stream != NULL);
716 E : DCHECK(buffer != NULL);
717 :
718 E : buffer->clear();
719 :
720 E : if (symbol_length < sizeof(SymbolType)) {
721 i : LOG(ERROR) << "Symbol too small for casting.";
722 i : return NULL;
723 : }
724 :
725 E : if (!stream->Read(buffer, symbol_length)) {
726 i : LOG(ERROR) << "Failed to read symbol.";
727 i : return NULL;
728 : }
729 :
730 E : return reinterpret_cast<const SymbolType*>(buffer->data());
731 E : }
732 :
733 : bool VisitNonControlFlowInstruction(const _DInst& instr,
734 : AbsoluteAddress block_addr,
735 : AbsoluteAddress instr_addr,
736 E : Block* block) {
737 E : DCHECK_NE(0u, block_addr.value());
738 E : DCHECK_NE(0u, instr_addr.value());
739 E : DCHECK_LE(block_addr, instr_addr);
740 E : DCHECK(block != NULL);
741 :
742 : // TODO(chrisha): We could walk the operands and follow references
743 : // explicitly. If any of them are of reference type and there's no
744 : // matching reference, this would be cause to blow up and die (we
745 : // should get all of these as relocs and/or fixups).
746 :
747 E : Offset instr_offset = instr_addr - block_addr;
748 : Block::ReferenceMap::const_iterator ref_it =
749 E : block->references().upper_bound(instr_offset);
750 : Block::ReferenceMap::const_iterator ref_end =
751 E : block->references().lower_bound(instr_offset + instr.size);
752 :
753 E : for (; ref_it != ref_end; ++ref_it) {
754 E : const Block* ref_block = ref_it->second.referenced();
755 :
756 : // We only care about inter-block references.
757 E : if (ref_block == block)
758 E : continue;
759 :
760 : // There should be no cross-block references to the middle of other
761 : // code blocks (to the top is fine, as we could be passing around a
762 : // function pointer). The exception is if the remote block is not
763 : // generated by cl.exe. In this case, there could be arbitrary labels
764 : // that act like functions within the body of that block, and referring
765 : // to them is perfectly fine.
766 : if (ref_block->type() == BlockGraph::CODE_BLOCK &&
767 : ref_it->second.base() != 0 &&
768 E : (block->attributes() & BlockGraph::BUILT_BY_UNSUPPORTED_COMPILER)) {
769 i : block->set_attribute(BlockGraph::ERRORED_DISASSEMBLY);
770 i : LOG(WARNING) << "Found a non-control-flow code-block to "
771 : << "middle-of-code-block reference from "
772 : << BlockInfo(block, block_addr) << " to "
773 : << BlockInfo(ref_block) << ".";
774 i : return true;
775 : }
776 E : }
777 :
778 E : return true;
779 E : }
780 :
781 : bool VisitPcRelativeControlFlowInstruction(bool create_missing_refs,
782 : const _DInst& instr,
783 : AbsoluteAddress image_addr,
784 : AbsoluteAddress block_addr,
785 : AbsoluteAddress instr_addr,
786 : BlockGraph::AddressSpace* image,
787 E : Block* block) {
788 E : DCHECK_NE(0u, image_addr.value());
789 E : DCHECK_NE(0u, block_addr.value());
790 E : DCHECK_NE(0u, instr_addr.value());
791 E : DCHECK_LT(image_addr, block_addr);
792 E : DCHECK_LE(block_addr, instr_addr);
793 E : DCHECK(image != NULL);
794 E : DCHECK(block != NULL);
795 :
796 E : int fc = META_GET_FC(instr.meta);
797 E : DCHECK(fc == FC_UNC_BRANCH || fc == FC_CALL || fc == FC_CND_BRANCH);
798 E : DCHECK_EQ(O_PC, instr.ops[0].type);
799 E : DCHECK_EQ(O_NONE, instr.ops[1].type);
800 E : DCHECK_EQ(O_NONE, instr.ops[2].type);
801 E : DCHECK_EQ(O_NONE, instr.ops[3].type);
802 : DCHECK(instr.ops[0].size == 8 ||
803 : instr.ops[0].size == 16 ||
804 E : instr.ops[0].size == 32);
805 :
806 : // Distorm gives us size in bits, we want bytes.
807 E : BlockGraph::Size size = instr.ops[0].size / 8;
808 :
809 : // Get the reference's address. Note we assume it's in the instruction's
810 : // tail end - I don't know of a case where a PC-relative offset in a branch
811 : // or call is not the very last thing in an x86 instruction.
812 E : AbsoluteAddress abs_src = instr_addr + instr.size - size;
813 : AbsoluteAddress abs_dst = instr_addr + instr.size +
814 E : static_cast<size_t>(instr.imm.addr);
815 E : RelativeAddress rel_dst(abs_dst.value() - image_addr.value());
816 E : Offset offset_src = abs_src - block_addr;
817 :
818 E : Block* dst_block = block;
819 E : RelativeAddress dst_block_addr(block_addr.value() - image_addr.value());
820 :
821 : // Is the reference to something outside this block?
822 E : if (abs_dst < block_addr || abs_dst >= block_addr + block->size()) {
823 : // Short PC-relative references should be to this block, otherwise this
824 : // block is not MSVC-like.
825 E : if (size < Reference::kMaximumSize) {
826 i : block->set_attribute(BlockGraph::ERRORED_DISASSEMBLY);
827 i : Offset offset_instr = instr_addr - block_addr;
828 i : LOG(WARNING) << "Found a " << size << "-byte PC-relative instruction to "
829 : << "an external " << abs_dst << " at offset "
830 : << offset_instr << " of " << BlockInfo(block, block_addr)
831 : << ".";
832 i : return true;
833 i : } else {
834 : // Long PC-relative references to other blocks should have been given to
835 : // us via FIXUPs, otherwise we risk breaking the world when moving blocks
836 : // around!
837 E : if (block->references().find(offset_src) == block->references().end()) {
838 i : LOG(ERROR) << "Missing fixup for a " << size << "-byte PC-relative "
839 : << "reference to " << abs_dst << " at offset "
840 : << offset_src << " of " << BlockInfo(block, block_addr)
841 : << ".";
842 i : return false;
843 : }
844 : }
845 :
846 : // Find the destination block and its address.
847 E : dst_block = image->GetContainingBlock(rel_dst, 1);
848 E : CHECK(image->GetAddressOf(dst_block, &dst_block_addr));
849 E : if (dst_block == NULL) {
850 i : LOG(ERROR) << "Found a " << size << "-byte PC-relative reference to a "
851 : << abs_dst << " outside of the image at offset "
852 : << offset_src << " of " << BlockInfo(block, block_addr) << ".";
853 i : return false;
854 : }
855 : }
856 :
857 : // Create the missing reference if need be. These are found by basic-block
858 : // disassembly so aren't strictly needed, but are useful debug information.
859 E : if (!create_missing_refs)
860 E : return true;
861 :
862 E : Offset offset_dst = rel_dst - dst_block_addr;
863 : Reference ref(BlockGraph::PC_RELATIVE_REF, size, dst_block, offset_dst,
864 E : offset_dst);
865 E : block->SetReference(offset_src, ref);
866 :
867 E : return true;
868 E : }
869 :
870 : bool VisitInstruction(bool create_missing_refs,
871 : const _DInst& instr,
872 : AbsoluteAddress image_addr,
873 : AbsoluteAddress block_addr,
874 : AbsoluteAddress instr_addr,
875 : BlockGraph::AddressSpace* image,
876 E : Block* block) {
877 E : DCHECK_NE(0u, image_addr.value());
878 E : DCHECK_NE(0u, block_addr.value());
879 E : DCHECK_NE(0u, instr_addr.value());
880 E : DCHECK_LT(image_addr, block_addr);
881 E : DCHECK_LE(block_addr, instr_addr);
882 E : DCHECK(image != NULL);
883 E : DCHECK(block != NULL);
884 :
885 E : int fc = META_GET_FC(instr.meta);
886 :
887 E : if (fc == FC_NONE) {
888 : return VisitNonControlFlowInstruction(
889 E : instr, block_addr, instr_addr, block);
890 : }
891 :
892 : if ((fc == FC_UNC_BRANCH || fc == FC_CALL || fc == FC_CND_BRANCH) &&
893 E : instr.ops[0].type == O_PC) {
894 : return VisitPcRelativeControlFlowInstruction(create_missing_refs,
895 E : instr, image_addr, block_addr, instr_addr, image, block);
896 : }
897 :
898 E : return true;
899 E : }
900 :
901 : bool DisassembleCodeBlockAndLabelData(bool create_missing_refs,
902 : AbsoluteAddress image_addr,
903 : AbsoluteAddress block_addr,
904 : BlockGraph::AddressSpace* image,
905 E : Block* block) {
906 E : DCHECK(image != NULL);
907 E : DCHECK(block != NULL);
908 E : DCHECK_EQ(BlockGraph::CODE_BLOCK, block->type());
909 :
910 : // We simultaneously walk through the block's references while disassembling
911 : // instructions. This is used to determine when (if) data starts. MSVC
912 : // always places jump tables first, which consist of absolute references.
913 E : const Block::ReferenceMap& ref_map(block->references());
914 E : Block::ReferenceMap::const_iterator ref_it = ref_map.begin();
915 :
916 : // We keep track of any self-references. If the block contains data these
917 : // are used as beginning points of tables. We rely on the sorted nature of
918 : // std::set when using these later on.
919 E : std::set<Offset> self_refs;
920 :
921 E : const uint8* data = block->data();
922 E : const uint8* data_end = block->data() + block->data_size();
923 :
924 : // If some of the data in this block is implicit then make it explicit for
925 : // ease of decoding.
926 E : std::vector<uint8> data_copy;
927 E : if (block->data_size() < block->size()) {
928 i : data_copy.resize(block->size(), 0);
929 i : ::memcpy(data_copy.data(), block->data(), block->data_size());
930 i : data = data_copy.data();
931 i : data_end = data + data_copy.size();
932 : }
933 :
934 : // Decode instructions one by one.
935 E : AbsoluteAddress addr(block_addr);
936 E : Offset offset = 0;
937 E : while (true) {
938 : // Stop the disassembly if we're at the end of the data.
939 E : if (data == data_end)
940 E : return true;
941 :
942 E : if (ref_it != ref_map.end()) {
943 : // Step past any references.
944 E : while (ref_it != ref_map.end() && ref_it->first < offset)
945 E : ++ref_it;
946 :
947 : // Stop the disassembly if the next byte is data. Namely, it coincides
948 : // with a reference.
949 E : if (ref_it->first == offset)
950 E : break;
951 : }
952 :
953 : // If we can't decode an instruction then we mark the block as not safe
954 : // for disassembly.
955 E : _DInst inst = { 0 };
956 : if (!core::DecodeOneInstruction(addr.value(), data, data_end - data,
957 E : &inst)) {
958 i : block->set_attribute(BlockGraph::ERRORED_DISASSEMBLY);
959 i : VLOG(1) << "Unable to decode instruction at offset " << offset
960 : << " of " << BlockInfo(block, block_addr) << ".";
961 i : return true;
962 : }
963 :
964 : // Visit the instruction itself. This validates that the instruction is of
965 : // a type we expect to encounter, and may also cause internal references to
966 : // be created.
967 : if (!VisitInstruction(create_missing_refs, inst, image_addr, block_addr,
968 E : addr, image, block)) {
969 i : return false;
970 : }
971 :
972 : // Step past the instruction.
973 E : addr += inst.size;
974 E : data += inst.size;
975 E : offset += inst.size;
976 :
977 : // References to data are by absolute pointer, for which we always receive
978 : // a reloc/fixup, thus no need to parse the instruction. Moreover, ref_it
979 : // points to the first reference after the beginning of the instruction at
980 : // this point.
981 E : if (ref_it != ref_map.end() && ref_it->first < offset) {
982 : // The reference should be wholly contained in the instruction.
983 E : if (static_cast<Offset>(ref_it->first + ref_it->second.size()) > offset) {
984 i : block->set_attribute(BlockGraph::ERRORED_DISASSEMBLY);
985 i : VLOG(1) << "Unexpected reference in instruction at offset "
986 : << ref_it->first << " of " << BlockInfo(block, block_addr)
987 : << ".";
988 i : return true;
989 : }
990 :
991 : // Store self-references to locations beyond our current cursor.
992 : if (ref_it->second.referenced() == block &&
993 E : ref_it->second.offset() > offset) {
994 E : self_refs.insert(ref_it->second.offset());
995 : }
996 :
997 E : ++ref_it;
998 : }
999 E : }
1000 :
1001 : // If we get here then we've encountered data. We need to label data
1002 : // sections as appropriate.
1003 :
1004 E : bool data_label_added = false;
1005 E : Offset end_of_code_offset = offset;
1006 :
1007 E : std::set<Offset>::const_iterator off_it = self_refs.begin();
1008 E : for (; off_it != self_refs.end(); ++off_it) {
1009 E : Offset referred_offset = *off_it;
1010 :
1011 : // References to data must be beyond the decoded instructions.
1012 E : if (referred_offset < end_of_code_offset)
1013 E : continue;
1014 :
1015 : // Determine if this offset points at another reference.
1016 E : bool ref_at_offset = false;
1017 E : if (ref_it != ref_map.end()) {
1018 : // Step past any references.
1019 E : while (ref_it != ref_map.end() && ref_it->first < referred_offset)
1020 E : ++ref_it;
1021 :
1022 : // Stop the disassembly if the next byte is data. Namely, it coincides
1023 : // with a reference.
1024 E : if (ref_it->first == referred_offset)
1025 E : ref_at_offset = true;
1026 : }
1027 :
1028 : // Build and set the data label.
1029 E : BlockGraph::LabelAttributes attr = BlockGraph::DATA_LABEL;
1030 E : const char* name = NULL;
1031 E : if (ref_at_offset) {
1032 E : name = kJumpTable;
1033 E : attr |= BlockGraph::JUMP_TABLE_LABEL;
1034 E : } else {
1035 E : name = kCaseTable;
1036 E : attr |= BlockGraph::CASE_TABLE_LABEL;
1037 : }
1038 E : if (!AddLabelToBlock(referred_offset, name, attr, block))
1039 i : return false;
1040 E : data_label_added = true;
1041 E : }
1042 :
1043 E : if (!data_label_added) {
1044 i : block->set_attribute(BlockGraph::ERRORED_DISASSEMBLY);
1045 i : VLOG(1) << "Disassembled into data but found no references to it for "
1046 : << BlockInfo(block, block_addr) << ".";
1047 i : return true;
1048 : }
1049 :
1050 E : return true;
1051 E : }
1052 :
1053 : bool JumpAndCaseTableAlreadyLabelled(const Block* block,
1054 : Offset offset,
1055 E : BlockGraph::LabelAttributes attr) {
1056 E : DCHECK(block != NULL);
1057 :
1058 : // We can't say anything about blocks that we were not able to disassemble.
1059 E : if (block->attributes() & BlockGraph::ERRORED_DISASSEMBLY)
1060 i : return true;
1061 :
1062 E : BlockGraph::Label label;
1063 E : if (!block->GetLabel(offset, &label)) {
1064 i : LOG(ERROR) << "Expected data label at offset " << offset << " of "
1065 : << BlockInfo(block) << ".";
1066 i : return false;
1067 : }
1068 :
1069 E : if ((label.attributes() & attr) == attr)
1070 E : return true;
1071 :
1072 i : LOG(ERROR) << "Label at offset " << offset << " of " << BlockInfo(block)
1073 : << " has attributes "
1074 : << BlockGraph::BlockAttributesToString(block->attributes())
1075 : << " but expected at least "
1076 : << BlockGraph::BlockAttributesToString(attr) << ".";
1077 :
1078 i : return false;
1079 E : }
1080 :
1081 : } // namespace
1082 :
1083 : // We use ", " as a separator between symbol names. We sometimes see commas
1084 : // in symbol names but do not see whitespace. Thus, this provides a useful
1085 : // separator that is also human friendly to read.
1086 : const char NewDecomposer::kLabelNameSep[] = ", ";
1087 :
1088 : // This is by CreateBlocksFromCoffGroups to communicate shared state to
1089 : // VisitLinkerSymbol via the VisitSymbols helper function.
1090 : struct NewDecomposer::VisitLinkerSymbolContext {
1091 : int current_group_index;
1092 : std::string current_group_prefix;
1093 : RelativeAddress current_group_start;
1094 :
1095 : // These are the set of patterns that indicate bracketing groups. They
1096 : // should match both the opening and the closing symbol, and have at least
1097 : // one match group returning the common prefix.
1098 : std::vector<RE> bracketing_groups;
1099 :
1100 E : VisitLinkerSymbolContext() : current_group_index(-1) {
1101 : // Matches groups like: .CRT$XCA -> .CRT$XCZ
1102 E : bracketing_groups.push_back(RE("(\\.CRT\\$X.)[AZ]"));
1103 : // Matches groups like: .rtc$IAA -> .rtc$IZZ
1104 E : bracketing_groups.push_back(RE("(\\.rtc\\$.*)(AA|ZZ)"));
1105 : // Matches exactly: ATL$__a -> ATL$__z
1106 E : bracketing_groups.push_back(RE("(ATL\\$__)[az]"));
1107 : // Matches exactly: .tls -> .tls$ZZZ
1108 E : bracketing_groups.push_back(RE("(\\.tls)(\\$ZZZ)?"));
1109 E : }
1110 :
1111 : private:
1112 : DISALLOW_COPY_AND_ASSIGN(VisitLinkerSymbolContext);
1113 : };
1114 :
1115 : NewDecomposer::NewDecomposer(const PEFile& image_file)
1116 : : image_file_(image_file), parse_debug_info_(true), image_layout_(NULL),
1117 E : image_(NULL), current_block_(NULL), current_scope_count_(0) {
1118 E : }
1119 :
1120 E : bool NewDecomposer::Decompose(ImageLayout* image_layout) {
1121 E : DCHECK(image_layout != NULL);
1122 :
1123 : // The temporaries should be NULL.
1124 E : DCHECK(image_layout_ == NULL);
1125 E : DCHECK(image_ == NULL);
1126 :
1127 : // We start by finding the PDB path.
1128 E : if (!FindAndValidatePdbPath())
1129 E : return false;
1130 E : DCHECK(!pdb_path_.empty());
1131 :
1132 : // Load the serialized block-graph from the PDB if it exists. This allows
1133 : // round-trip decomposition.
1134 E : bool stream_exists = false;
1135 : if (LoadBlockGraphFromPdb(
1136 E : pdb_path_, image_file_, image_layout, &stream_exists)) {
1137 E : return true;
1138 E : } else if (stream_exists) {
1139 : // If the stream exists but hasn't been loaded we return an error. At this
1140 : // point an error message has already been logged if there was one.
1141 i : return false;
1142 : }
1143 :
1144 : // At this point a full decomposition needs to be performed.
1145 E : image_layout_ = image_layout;
1146 E : image_ = &(image_layout->blocks);
1147 E : bool success = DecomposeImpl();
1148 E : image_layout_ = NULL;
1149 E : image_ = NULL;
1150 :
1151 E : return success;
1152 E : }
1153 :
1154 E : bool NewDecomposer::FindAndValidatePdbPath() {
1155 : // Manually find the PDB path if it is not specified.
1156 E : if (pdb_path_.empty()) {
1157 : if (!FindPdbForModule(image_file_.path(), &pdb_path_) ||
1158 E : pdb_path_.empty()) {
1159 i : LOG(ERROR) << "Unable to find PDB file for module: "
1160 : << image_file_.path().value();
1161 i : return false;
1162 : }
1163 : }
1164 E : DCHECK(!pdb_path_.empty());
1165 :
1166 E : if (!file_util::PathExists(pdb_path_)) {
1167 E : LOG(ERROR) << "Path not found: " << pdb_path_.value();
1168 E : return false;
1169 : }
1170 :
1171 E : if (!pe::PeAndPdbAreMatched(image_file_.path(), pdb_path_)) {
1172 i : LOG(ERROR) << "PDB file \"" << pdb_path_.value() << "\" does not match "
1173 : << "module \"" << image_file_.path().value() << "\".";
1174 i : return false;
1175 : }
1176 :
1177 E : return true;
1178 E : }
1179 :
1180 : bool NewDecomposer::LoadBlockGraphFromPdbStream(
1181 : const PEFile& image_file,
1182 : pdb::PdbStream* block_graph_stream,
1183 E : ImageLayout* image_layout) {
1184 E : DCHECK(block_graph_stream != NULL);
1185 E : DCHECK(image_layout != NULL);
1186 E : LOG(INFO) << "Reading block-graph and image layout from the PDB.";
1187 :
1188 : // Initialize an input archive pointing to the stream.
1189 E : scoped_refptr<pdb::PdbByteStream> byte_stream = new pdb::PdbByteStream();
1190 E : if (!byte_stream->Init(block_graph_stream))
1191 i : return false;
1192 E : DCHECK(byte_stream.get() != NULL);
1193 :
1194 E : core::ScopedInStreamPtr pdb_in_stream;
1195 : pdb_in_stream.reset(core::CreateByteInStream(
1196 E : byte_stream->data(), byte_stream->data() + byte_stream->length()));
1197 :
1198 : // Read the header.
1199 E : uint32 stream_version = 0;
1200 E : unsigned char compressed = 0;
1201 : if (!pdb_in_stream->Read(sizeof(stream_version),
1202 : reinterpret_cast<core::Byte*>(&stream_version)) ||
1203 : !pdb_in_stream->Read(sizeof(compressed),
1204 E : reinterpret_cast<core::Byte*>(&compressed))) {
1205 i : LOG(ERROR) << "Failed to read existing Syzygy block-graph stream header.";
1206 i : return false;
1207 : }
1208 :
1209 : // Check the stream version.
1210 E : if (stream_version != pdb::kSyzygyBlockGraphStreamVersion) {
1211 E : LOG(ERROR) << "PDB contains an unsupported Syzygy block-graph stream"
1212 : << " version (got " << stream_version << ", expected "
1213 : << pdb::kSyzygyBlockGraphStreamVersion << ").";
1214 E : return false;
1215 : }
1216 :
1217 : // If the stream is compressed insert the decompression filter.
1218 E : core::InStream* in_stream = pdb_in_stream.get();
1219 E : scoped_ptr<core::ZInStream> zip_in_stream;
1220 E : if (compressed != 0) {
1221 E : zip_in_stream.reset(new core::ZInStream(in_stream));
1222 E : if (!zip_in_stream->Init()) {
1223 i : LOG(ERROR) << "Unable to initialize ZInStream.";
1224 i : return false;
1225 : }
1226 E : in_stream = zip_in_stream.get();
1227 : }
1228 :
1229 : // Deserialize the image-layout.
1230 E : core::NativeBinaryInArchive in_archive(in_stream);
1231 E : block_graph::BlockGraphSerializer::Attributes attributes = 0;
1232 : if (!LoadBlockGraphAndImageLayout(
1233 E : image_file, &attributes, image_layout, &in_archive)) {
1234 i : LOG(ERROR) << "Failed to deserialize block-graph and image layout.";
1235 i : return false;
1236 : }
1237 :
1238 E : return true;
1239 E : }
1240 :
1241 : bool NewDecomposer::LoadBlockGraphFromPdb(const base::FilePath& pdb_path,
1242 : const PEFile& image_file,
1243 : ImageLayout* image_layout,
1244 E : bool* stream_exists) {
1245 E : DCHECK(image_layout != NULL);
1246 E : DCHECK(stream_exists != NULL);
1247 :
1248 E : pdb::PdbFile pdb_file;
1249 E : pdb::PdbReader pdb_reader;
1250 E : if (!pdb_reader.Read(pdb_path, &pdb_file)) {
1251 i : LOG(ERROR) << "Unable to read the PDB named \"" << pdb_path.value()
1252 : << "\".";
1253 i : return NULL;
1254 : }
1255 :
1256 : // Try to get the block-graph stream from the PDB.
1257 E : scoped_refptr<pdb::PdbStream> block_graph_stream;
1258 : if (!pdb::LoadNamedStreamFromPdbFile(pdb::kSyzygyBlockGraphStreamName,
1259 : &pdb_file,
1260 : &block_graph_stream) ||
1261 E : block_graph_stream.get() == NULL) {
1262 E : *stream_exists = false;
1263 E : return false;
1264 : }
1265 E : if (block_graph_stream->length() == 0) {
1266 i : *stream_exists = false;
1267 i : LOG(WARNING) << "The block-graph stream is empty, ignoring it.";
1268 i : return false;
1269 : }
1270 :
1271 : // The PDB contains a block-graph stream, the block-graph and the image layout
1272 : // will be read from this stream.
1273 E : *stream_exists = true;
1274 : if (!LoadBlockGraphFromPdbStream(image_file, block_graph_stream.get(),
1275 E : image_layout)) {
1276 i : return false;
1277 : }
1278 :
1279 E : return true;
1280 E : }
1281 :
1282 E : bool NewDecomposer::DecomposeImpl() {
1283 : // Instantiate and initialize our Debug Interface Access session. This logs
1284 : // verbosely for us.
1285 E : ScopedComPtr<IDiaDataSource> dia_source;
1286 E : ScopedComPtr<IDiaSession> dia_session;
1287 E : ScopedComPtr<IDiaSymbol> global;
1288 : if (!InitializeDia(image_file_, pdb_path_, dia_source.Receive(),
1289 E : dia_session.Receive(), global.Receive())) {
1290 i : return false;
1291 : }
1292 :
1293 : // Copy the image headers to the layout.
1294 : CopySectionHeadersToImageLayout(
1295 : image_file_.nt_headers()->FileHeader.NumberOfSections,
1296 : image_file_.section_headers(),
1297 E : &(image_layout_->sections));
1298 :
1299 : // Create the sections in the underlying block-graph.
1300 E : if (!CreateBlockGraphSections())
1301 i : return false;
1302 :
1303 : // We scope the first few operations so that we don't keep the intermediate
1304 : // references around any longer than we have to.
1305 : {
1306 E : IntermediateReferences references;
1307 :
1308 : // First we parse out the PE blocks.
1309 E : if (!CreatePEImageBlocksAndReferences(&references))
1310 i : return false;
1311 :
1312 : // Now we parse the COFF group symbols from the linker's symbol stream.
1313 : // These indicate things like static initializers, which must stay together
1314 : // in a single block.
1315 E : if (!CreateBlocksFromCoffGroups())
1316 i : return false;
1317 :
1318 : // Next we parse out section contributions. Some of these may coincide with
1319 : // existing PE parsed blocks, but when they do we expect them to be exact
1320 : // collisions.
1321 E : if (!CreateBlocksFromSectionContribs(dia_session.get()))
1322 i : return false;
1323 :
1324 : // Flesh out the rest of the image with gap blocks.
1325 E : if (!CreateGapBlocks())
1326 i : return false;
1327 :
1328 : // Finalize the PE-parsed intermediate references.
1329 E : if (!FinalizeIntermediateReferences(references))
1330 i : return false;
1331 E : }
1332 :
1333 : // Parse the fixups and use them to create references.
1334 E : if (!CreateReferencesFromFixups(dia_session.get()))
1335 i : return false;
1336 :
1337 : // Disassemble code blocks and use the results to infer case and jump tables.
1338 E : if (!DisassembleCodeBlocksAndLabelData())
1339 i : return false;
1340 :
1341 : // Annotate the block-graph with symbol information.
1342 E : if (parse_debug_info_ && !ProcessSymbols(global.get()))
1343 i : return false;
1344 :
1345 E : return true;
1346 E : }
1347 :
1348 E : bool NewDecomposer::CreateBlockGraphSections() {
1349 : // Iterate through the image sections, and create sections in the BlockGraph.
1350 E : size_t num_sections = image_file_.nt_headers()->FileHeader.NumberOfSections;
1351 E : for (size_t i = 0; i < num_sections; ++i) {
1352 E : const IMAGE_SECTION_HEADER* header = image_file_.section_header(i);
1353 E : std::string name = pe::PEFile::GetSectionName(*header);
1354 : BlockGraph::Section* section = image_->graph()->AddSection(
1355 E : name, header->Characteristics);
1356 E : DCHECK(section != NULL);
1357 :
1358 : // For now, we expect them to have been created with the same IDs as those
1359 : // in the original image.
1360 E : if (section->id() != i) {
1361 i : LOG(ERROR) << "Unexpected section ID.";
1362 i : return false;
1363 : }
1364 E : }
1365 :
1366 E : return true;
1367 E : }
1368 :
1369 : bool NewDecomposer::CreatePEImageBlocksAndReferences(
1370 E : IntermediateReferences* references) {
1371 E : DCHECK(references != NULL);
1372 :
1373 : PEFileParser::AddReferenceCallback add_reference(
1374 E : base::Bind(&AddIntermediateReference, base::Unretained(references)));
1375 E : PEFileParser parser(image_file_, image_, add_reference);
1376 E : PEFileParser::PEHeader header;
1377 E : if (!parser.ParseImage(&header)) {
1378 i : LOG(ERROR) << "Unable to parse PE image.";
1379 i : return false;
1380 : }
1381 :
1382 E : return true;
1383 E : }
1384 :
1385 E : bool NewDecomposer::CreateBlocksFromCoffGroups() {
1386 E : pdb::PdbFile pdb_file;
1387 E : pdb::PdbReader pdb_reader;
1388 E : if (!pdb_reader.Read(pdb_path_, &pdb_file)) {
1389 i : LOG(ERROR) << "Failed to load PDB: " << pdb_path_.value();
1390 i : return false;
1391 : }
1392 :
1393 E : scoped_refptr<pdb::PdbStream> symbols = GetLinkerSymbolStream(pdb_file);
1394 :
1395 : // Process the symbols in the linker module symbol stream.
1396 E : VisitLinkerSymbolContext context;
1397 : pdb::VisitSymbolsCallback callback = base::Bind(
1398 : &NewDecomposer::VisitLinkerSymbol,
1399 : base::Unretained(this),
1400 E : base::Unretained(&context));
1401 E : if (!pdb::VisitSymbols(callback, symbols->length(), true, symbols.get()))
1402 i : return false;
1403 :
1404 : // Bail if we did not encounter a closing bracketing symbol where one was
1405 : // expected.
1406 E : if (context.current_group_index != -1) {
1407 i : LOG(ERROR) << "Unable to close bracketed COFF group \""
1408 : << context.current_group_prefix << "\".";
1409 i : return false;
1410 : }
1411 :
1412 E : return true;
1413 E : }
1414 :
1415 E : bool NewDecomposer::CreateBlocksFromSectionContribs(IDiaSession* session) {
1416 E : ScopedComPtr<IDiaEnumSectionContribs> section_contribs;
1417 : SearchResult search_result = FindDiaTable(session,
1418 E : section_contribs.Receive());
1419 E : if (search_result != kSearchSucceeded) {
1420 i : if (search_result == kSearchFailed)
1421 i : LOG(ERROR) << "No section contribution table found.";
1422 i : return false;
1423 : }
1424 :
1425 E : size_t rsrc_id = image_file_.GetSectionIndex(kResourceSectionName);
1426 :
1427 E : LONG count = 0;
1428 E : if (section_contribs->get_Count(&count) != S_OK) {
1429 i : LOG(ERROR) << "Failed to get section contributions enumeration length.";
1430 i : return false;
1431 : }
1432 :
1433 E : for (LONG visited = 0; visited < count; ++visited) {
1434 E : ScopedComPtr<IDiaSectionContrib> section_contrib;
1435 E : ULONG fetched = 0;
1436 E : HRESULT hr = section_contribs->Next(1, section_contrib.Receive(), &fetched);
1437 : // The standard way to end an enumeration (according to the docs) is by
1438 : // returning S_FALSE and setting fetched to 0. We don't actually see this,
1439 : // but it wouldn't be an error if we did.
1440 E : if (hr == S_FALSE && fetched == 0)
1441 i : break;
1442 E : if (hr != S_OK) {
1443 i : LOG(ERROR) << "Failed to get DIA section contribution: "
1444 : << com::LogHr(hr) << ".";
1445 i : return false;
1446 : }
1447 : // We actually end up seeing S_OK and fetched == 0 when the enumeration
1448 : // terminates, which goes against the publishes documentations.
1449 E : if (fetched == 0)
1450 i : break;
1451 :
1452 E : DWORD rva = 0;
1453 E : DWORD length = 0;
1454 E : DWORD section_id = 0;
1455 E : BOOL code = FALSE;
1456 E : ScopedComPtr<IDiaSymbol> compiland;
1457 E : ScopedBstr bstr_name;
1458 : if ((hr = section_contrib->get_relativeVirtualAddress(&rva)) != S_OK ||
1459 : (hr = section_contrib->get_length(&length)) != S_OK ||
1460 : (hr = section_contrib->get_addressSection(§ion_id)) != S_OK ||
1461 : (hr = section_contrib->get_code(&code)) != S_OK ||
1462 : (hr = section_contrib->get_compiland(compiland.Receive())) != S_OK ||
1463 E : (hr = compiland->get_name(bstr_name.Receive())) != S_OK) {
1464 i : LOG(ERROR) << "Failed to get section contribution properties: "
1465 : << com::LogHr(hr) << ".";
1466 i : return false;
1467 : }
1468 :
1469 : // Determine if this function was built by a supported compiler.
1470 : bool is_built_by_supported_compiler =
1471 E : IsBuiltBySupportedCompiler(compiland.get());
1472 :
1473 : // DIA numbers sections from 1 to n, while we do 0 to n - 1.
1474 E : DCHECK_LT(0u, section_id);
1475 E : --section_id;
1476 :
1477 : // We don't parse the resource section, as it is parsed by the PEFileParser.
1478 E : if (section_id == rsrc_id)
1479 E : continue;
1480 :
1481 E : std::string name;
1482 E : if (!WideToUTF8(bstr_name, bstr_name.Length(), &name)) {
1483 i : LOG(ERROR) << "Failed to convert compiland name to UTF8.";
1484 i : return false;
1485 : }
1486 :
1487 : // TODO(chrisha): We see special section contributions with the name
1488 : // "* CIL *". These are concatenations of data symbols and can very
1489 : // likely be chunked using symbols directly. A cursory visual inspection
1490 : // of symbol names hints that these might be related to WPO.
1491 :
1492 : // Create the block.
1493 : BlockType block_type =
1494 E : code ? BlockGraph::CODE_BLOCK : BlockGraph::DATA_BLOCK;
1495 : Block* block = CreateBlockOrFindCoveringPeBlock(
1496 E : block_type, RelativeAddress(rva), length, name);
1497 E : if (block == NULL) {
1498 i : LOG(ERROR) << "Unable to create block for compiland \"" << name << "\".";
1499 i : return false;
1500 : }
1501 :
1502 : // Set the block attributes.
1503 E : block->set_attribute(BlockGraph::SECTION_CONTRIB);
1504 E : if (!is_built_by_supported_compiler)
1505 E : block->set_attribute(BlockGraph::BUILT_BY_UNSUPPORTED_COMPILER);
1506 E : }
1507 :
1508 E : return true;
1509 E : }
1510 :
1511 E : bool NewDecomposer::CreateGapBlocks() {
1512 E : size_t num_sections = image_file_.nt_headers()->FileHeader.NumberOfSections;
1513 :
1514 : // Iterate through all the image sections.
1515 E : for (size_t i = 0; i < num_sections; ++i) {
1516 E : const IMAGE_SECTION_HEADER* header = image_file_.section_header(i);
1517 E : DCHECK(header != NULL);
1518 :
1519 E : BlockType type = BlockGraph::CODE_BLOCK;
1520 E : const char* section_type = NULL;
1521 E : switch (GetSectionType(header)) {
1522 : case kSectionCode:
1523 E : type = BlockGraph::CODE_BLOCK;
1524 E : section_type = "code";
1525 E : break;
1526 :
1527 : case kSectionData:
1528 E : type = BlockGraph::DATA_BLOCK;
1529 E : section_type = "data";
1530 E : break;
1531 :
1532 : default:
1533 i : continue;
1534 : }
1535 :
1536 E : if (!CreateSectionGapBlocks(header, type)) {
1537 i : LOG(ERROR) << "Unable to create gap blocks for " << section_type
1538 : << " section \"" << header->Name << "\".";
1539 i : return false;
1540 : }
1541 E : }
1542 :
1543 E : return true;
1544 E : }
1545 :
1546 : bool NewDecomposer::FinalizeIntermediateReferences(
1547 E : const IntermediateReferences& references) {
1548 E : for (size_t i = 0; i < references.size(); ++i) {
1549 : // This logs verbosely for us.
1550 : if (!CreateReference(references[i].src_addr,
1551 : references[i].size,
1552 : references[i].type,
1553 : references[i].dst_addr,
1554 : references[i].dst_addr,
1555 E : image_)) {
1556 i : return false;
1557 : }
1558 E : }
1559 E : return true;
1560 E : }
1561 :
1562 E : bool NewDecomposer::DisassembleCodeBlocksAndLabelData() {
1563 E : DCHECK(image_ != NULL);
1564 :
1565 : const BlockGraph::Block* dos_header_block =
1566 E : image_->GetBlockByAddress(RelativeAddress(0));
1567 E : DCHECK(dos_header_block != NULL);
1568 :
1569 : const BlockGraph::Block* nt_headers_block =
1570 E : GetNtHeadersBlockFromDosHeaderBlock(dos_header_block);
1571 E : if (nt_headers_block == NULL) {
1572 i : LOG(ERROR) << "Unable to get NT headers block for image.";
1573 i : return false;
1574 : }
1575 :
1576 : // GetNtHeadersBlockFromDosHeaderBlock sanity checks things so we can cast
1577 : // with impunity.
1578 : const IMAGE_NT_HEADERS* nt_headers =
1579 E : reinterpret_cast<const IMAGE_NT_HEADERS*>(nt_headers_block->data());
1580 E : core::AbsoluteAddress image_base(nt_headers->OptionalHeader.ImageBase);
1581 :
1582 : // Walk through the blocks and disassemble each one of them.
1583 E : BlockGraph::AddressSpace::RangeMapConstIter it = image_->begin();
1584 E : for (; it != image_->end(); ++it) {
1585 E : BlockGraph::Block* block = it->second;
1586 :
1587 E : if (block->type() != BlockGraph::CODE_BLOCK)
1588 E : continue;
1589 :
1590 E : core::AbsoluteAddress abs_addr(image_base + it->first.start().value());
1591 : if (!DisassembleCodeBlockAndLabelData(
1592 E : parse_debug_info_, image_base, abs_addr, image_, block)) {
1593 i : return false;
1594 : }
1595 E : }
1596 :
1597 E : return true;
1598 E : }
1599 :
1600 E : bool NewDecomposer::CreateReferencesFromFixups(IDiaSession* session) {
1601 E : DCHECK(session != NULL);
1602 :
1603 E : PEFile::RelocSet reloc_set;
1604 E : if (!image_file_.DecodeRelocs(&reloc_set))
1605 i : return false;
1606 :
1607 E : OMAPs omap_from;
1608 E : PdbFixups fixups;
1609 E : if (!LoadDebugStreams(session, &fixups, &omap_from))
1610 i : return false;
1611 :
1612 : // While creating references from the fixups this removes the
1613 : // corresponding reference data from the relocs. We use this as a kind of
1614 : // double-entry bookkeeping to ensure all is well and right in the world.
1615 : if (!CreateReferencesFromFixupsImpl(image_file_, fixups, omap_from,
1616 E : &reloc_set, image_)) {
1617 i : return false;
1618 : }
1619 :
1620 E : if (!reloc_set.empty()) {
1621 i : LOG(ERROR) << "Found reloc entries without matching FIXUP entries.";
1622 i : return false;
1623 : }
1624 :
1625 E : return true;
1626 E : }
1627 :
1628 E : bool NewDecomposer::ProcessSymbols(IDiaSymbol* root) {
1629 E : DCHECK(root != NULL);
1630 :
1631 : DiaBrowser::MatchCallback on_push_function_or_thunk_symbol(
1632 : base::Bind(&NewDecomposer::OnPushFunctionOrThunkSymbol,
1633 E : base::Unretained(this)));
1634 : DiaBrowser::MatchCallback on_pop_function_or_thunk_symbol(
1635 : base::Bind(&NewDecomposer::OnPopFunctionOrThunkSymbol,
1636 E : base::Unretained(this)));
1637 : DiaBrowser::MatchCallback on_function_child_symbol(
1638 : base::Bind(&NewDecomposer::OnFunctionChildSymbol,
1639 E : base::Unretained(this)));
1640 : DiaBrowser::MatchCallback on_data_symbol(
1641 E : base::Bind(&NewDecomposer::OnDataSymbol, base::Unretained(this)));
1642 : DiaBrowser::MatchCallback on_public_symbol(
1643 E : base::Bind(&NewDecomposer::OnPublicSymbol, base::Unretained(this)));
1644 : DiaBrowser::MatchCallback on_label_symbol(
1645 E : base::Bind(&NewDecomposer::OnLabelSymbol, base::Unretained(this)));
1646 :
1647 E : DiaBrowser dia_browser;
1648 :
1649 : // Find thunks.
1650 : dia_browser.AddPattern(Seq(Opt(SymTagCompiland), SymTagThunk),
1651 : on_push_function_or_thunk_symbol,
1652 E : on_pop_function_or_thunk_symbol);
1653 :
1654 : // Find functions and all data, labels, callsites, debug start/end and block
1655 : // symbols below them. This is done in one single pattern so that the
1656 : // function pushes/pops happen in the right order.
1657 : dia_browser.AddPattern(
1658 : Seq(Opt(SymTagCompiland),
1659 : Callback(Or(SymTagFunction, SymTagThunk),
1660 : on_push_function_or_thunk_symbol,
1661 : on_pop_function_or_thunk_symbol),
1662 : Star(SymTagBlock),
1663 : Or(SymTagData,
1664 : SymTagLabel,
1665 : SymTagBlock,
1666 : SymTagFuncDebugStart,
1667 : SymTagFuncDebugEnd,
1668 : SymTagCallSite)),
1669 E : on_function_child_symbol);
1670 :
1671 : // Global data and code label symbols.
1672 : dia_browser.AddPattern(Seq(Opt(SymTagCompiland), SymTagLabel),
1673 E : on_label_symbol);
1674 : dia_browser.AddPattern(Seq(Opt(SymTagCompiland), SymTagData),
1675 E : on_data_symbol);
1676 :
1677 : // Public symbols. These provide decorated names without any type info, but
1678 : // are useful for debugging.
1679 E : dia_browser.AddPattern(SymTagPublicSymbol, on_public_symbol);
1680 :
1681 E : return dia_browser.Browse(root);
1682 E : }
1683 :
1684 : bool NewDecomposer::VisitLinkerSymbol(VisitLinkerSymbolContext* context,
1685 : uint16 symbol_length,
1686 : uint16 symbol_type,
1687 E : pdb::PdbStream* stream) {
1688 E : DCHECK(context != NULL);
1689 E : DCHECK(stream != NULL);
1690 :
1691 E : if (symbol_type != cci::S_COFFGROUP)
1692 E : return true;
1693 :
1694 E : std::vector<uint8> buffer;
1695 : const cci::CoffGroupSym* coffgroup =
1696 E : ParseSymbol<cci::CoffGroupSym>(symbol_length, stream, &buffer);
1697 E : if (coffgroup == NULL)
1698 i : return false;
1699 :
1700 : // The PDB numbers sections starting at index 1 but we use index 0.
1701 : RelativeAddress rva(image_layout_->sections[coffgroup->seg - 1].addr +
1702 E : coffgroup->off);
1703 :
1704 : // We are looking for an opening symbol.
1705 E : if (context->current_group_index == -1) {
1706 E : for (size_t i = 0; i < context->bracketing_groups.size(); ++i) {
1707 E : std::string prefix;
1708 E : if (context->bracketing_groups[i].FullMatch(coffgroup->name, &prefix)) {
1709 E : context->current_group_index = i;
1710 E : context->current_group_prefix = prefix;
1711 E : context->current_group_start = rva;
1712 E : return true;
1713 : }
1714 E : }
1715 :
1716 : // No opening symbol was encountered. We can safely ignore this
1717 : // COFF group symbol.
1718 E : return true;
1719 : }
1720 :
1721 : // If we get here we've found an opening symbol and we're looking for the
1722 : // matching closing symbol.
1723 E : std::string prefix;
1724 : if (!context->bracketing_groups[context->current_group_index].FullMatch(
1725 E : coffgroup->name, &prefix)) {
1726 E : return true;
1727 : }
1728 :
1729 E : if (prefix != context->current_group_prefix) {
1730 : // We see another symbol open/close while already in an opened symbol.
1731 : // This indicates nested bracketing information, which we've never seen
1732 : // before.
1733 i : LOG(ERROR) << "Encountered nested bracket symbol \"" << prefix
1734 : << "\" while in \"" << context->current_group_prefix << "\".";
1735 i : return false;
1736 : }
1737 :
1738 E : RelativeAddress end = rva + coffgroup->cb;
1739 E : DCHECK_LT(context->current_group_start, end);
1740 :
1741 : // Create a block for this bracketed COFF group.
1742 : BlockGraph::Block* block = CreateBlock(
1743 : BlockGraph::DATA_BLOCK,
1744 : context->current_group_start,
1745 : end - context->current_group_start,
1746 E : base::StringPrintf("Bracketed COFF group: %s", prefix.c_str()));
1747 E : if (block == NULL) {
1748 i : LOG(ERROR) << "Failed to create bracketed COFF group \""
1749 : << prefix << "\".";
1750 i : return false;
1751 : }
1752 E : block->set_attribute(BlockGraph::COFF_GROUP);
1753 :
1754 : // Indicate that this block is closed and we're looking for another opening
1755 : // bracket symbol.
1756 E : context->current_group_index = -1;
1757 E : context->current_group_prefix.clear();
1758 E : context->current_group_start = RelativeAddress(0);
1759 :
1760 E : return true;
1761 E : }
1762 :
1763 : DiaBrowser::BrowserDirective NewDecomposer::OnPushFunctionOrThunkSymbol(
1764 : const DiaBrowser& dia_browser,
1765 : const DiaBrowser::SymTagVector& sym_tags,
1766 E : const DiaBrowser::SymbolPtrVector& symbols) {
1767 E : DCHECK(!symbols.empty());
1768 E : DCHECK_EQ(sym_tags.size(), symbols.size());
1769 E : DiaBrowser::SymbolPtr symbol = symbols.back();
1770 :
1771 E : DCHECK(current_block_ == NULL);
1772 E : DCHECK_EQ(current_address_, RelativeAddress(0));
1773 E : DCHECK_EQ(0u, current_scope_count_);
1774 :
1775 E : HRESULT hr = E_FAIL;
1776 E : DWORD location_type = LocIsNull;
1777 E : DWORD rva = 0;
1778 E : ULONGLONG length = 0;
1779 E : ScopedBstr name_bstr;
1780 : if (FAILED(hr = symbol->get_locationType(&location_type)) ||
1781 : FAILED(hr = symbol->get_relativeVirtualAddress(&rva)) ||
1782 : FAILED(hr = symbol->get_length(&length)) ||
1783 E : FAILED(hr = symbol->get_name(name_bstr.Receive()))) {
1784 i : LOG(ERROR) << "Failed to get function/thunk properties: " << com::LogHr(hr)
1785 : << ".";
1786 i : return DiaBrowser::kBrowserAbort;
1787 : }
1788 :
1789 : // We only care about functions with static storage. We can stop looking at
1790 : // things below this node, as we won't be able to resolve them either.
1791 E : if (location_type != LocIsStatic)
1792 i : return DiaBrowser::kBrowserTerminatePath;
1793 :
1794 E : RelativeAddress addr(rva);
1795 E : Block* block = image_->GetBlockByAddress(addr);
1796 E : CHECK(block != NULL);
1797 E : RelativeAddress block_addr;
1798 E : CHECK(image_->GetAddressOf(block, &block_addr));
1799 E : DCHECK(InRange(addr, block_addr, block->size()));
1800 :
1801 E : std::string name;
1802 E : if (!WideToUTF8(name_bstr, name_bstr.Length(), &name)) {
1803 i : LOG(ERROR) << "Failed to convert function/thunk name to UTF8.";
1804 i : return DiaBrowser::kBrowserAbort;
1805 : }
1806 :
1807 : // We know the function starts in this block but we need to make sure its
1808 : // end does not extend past the end of the block.
1809 E : if (addr + length > block_addr + block->size()) {
1810 i : LOG(ERROR) << "Got function/thunk \"" << name << "\" that is not contained "
1811 : << "by section contribution \"" << block->name() << "\".";
1812 i : return DiaBrowser::kBrowserAbort;
1813 : }
1814 :
1815 E : Offset offset = addr - block_addr;
1816 E : if (!AddLabelToBlock(offset, name, BlockGraph::CODE_LABEL, block))
1817 i : return DiaBrowser::kBrowserAbort;
1818 :
1819 : // Keep track of the generated block. We will use this when parsing symbols
1820 : // that belong to this function. This prevents us from having to do repeated
1821 : // lookups and also allows us to associate labels outside of the block to the
1822 : // correct block.
1823 E : current_block_ = block;
1824 E : current_address_ = block_addr;
1825 :
1826 : // Certain properties are not defined on all blocks, so the following calls
1827 : // may return S_FALSE.
1828 E : BOOL no_return = FALSE;
1829 E : if (symbol->get_noReturn(&no_return) != S_OK)
1830 E : no_return = FALSE;
1831 :
1832 E : BOOL has_inl_asm = FALSE;
1833 E : if (symbol->get_hasInlAsm(&has_inl_asm) != S_OK)
1834 E : has_inl_asm = FALSE;
1835 :
1836 E : BOOL has_eh = FALSE;
1837 E : if (symbol->get_hasEH(&has_eh) != S_OK)
1838 E : has_eh = FALSE;
1839 :
1840 E : BOOL has_seh = FALSE;
1841 E : if (symbol->get_hasSEH(&has_seh) != S_OK)
1842 E : has_seh = FALSE;
1843 :
1844 : // Set the block attributes.
1845 E : if (no_return == TRUE)
1846 E : block->set_attribute(BlockGraph::NON_RETURN_FUNCTION);
1847 E : if (has_inl_asm == TRUE)
1848 E : block->set_attribute(BlockGraph::HAS_INLINE_ASSEMBLY);
1849 E : if (has_eh || has_seh)
1850 E : block->set_attribute(BlockGraph::HAS_EXCEPTION_HANDLING);
1851 E : if (IsSymTag(symbol, SymTagThunk))
1852 E : block->set_attribute(BlockGraph::THUNK);
1853 :
1854 E : return DiaBrowser::kBrowserContinue;
1855 E : }
1856 :
1857 : DiaBrowser::BrowserDirective NewDecomposer::OnPopFunctionOrThunkSymbol(
1858 : const DiaBrowser& dia_browser,
1859 : const DiaBrowser::SymTagVector& sym_tags,
1860 E : const DiaBrowser::SymbolPtrVector& symbols) {
1861 : // Simply clean up the current function block and address.
1862 E : current_block_ = NULL;
1863 E : current_address_ = RelativeAddress(0);
1864 E : current_scope_count_ = 0;
1865 E : return DiaBrowser::kBrowserContinue;
1866 E : }
1867 :
1868 : DiaBrowser::BrowserDirective NewDecomposer::OnFunctionChildSymbol(
1869 : const DiaBrowser& dia_browser,
1870 : const DiaBrowser::SymTagVector& sym_tags,
1871 E : const DiaBrowser::SymbolPtrVector& symbols) {
1872 E : DCHECK(!symbols.empty());
1873 E : DCHECK_EQ(sym_tags.size(), symbols.size());
1874 :
1875 : // This can only be called from the context of a function, so we expect the
1876 : // parent function block to be set and remembered.
1877 E : DCHECK(current_block_ != NULL);
1878 :
1879 : // The set of sym tags here should match the pattern used in the DiaBrowser
1880 : // instance set up in ProcessSymbols.
1881 E : switch (sym_tags.back()) {
1882 : case SymTagData:
1883 E : return OnDataSymbol(dia_browser, sym_tags, symbols);
1884 :
1885 : case SymTagLabel:
1886 E : return OnLabelSymbol(dia_browser, sym_tags, symbols);
1887 :
1888 : case SymTagBlock:
1889 : case SymTagFuncDebugStart:
1890 : case SymTagFuncDebugEnd:
1891 E : return OnScopeSymbol(sym_tags.back(), symbols.back());
1892 :
1893 : case SymTagCallSite:
1894 E : return OnCallSiteSymbol(symbols.back());
1895 :
1896 : default:
1897 : break;
1898 : }
1899 :
1900 i : LOG(ERROR) << "Unhandled function child symbol: " << sym_tags.back() << ".";
1901 i : return DiaBrowser::kBrowserAbort;
1902 E : }
1903 :
1904 : DiaBrowser::BrowserDirective NewDecomposer::OnDataSymbol(
1905 : const DiaBrowser& dia_browser,
1906 : const DiaBrowser::SymTagVector& sym_tags,
1907 E : const DiaBrowser::SymbolPtrVector& symbols) {
1908 E : DCHECK(!symbols.empty());
1909 E : DCHECK_EQ(sym_tags.size(), symbols.size());
1910 E : DiaBrowser::SymbolPtr symbol = symbols.back();
1911 :
1912 E : HRESULT hr = E_FAIL;
1913 E : DWORD location_type = LocIsNull;
1914 E : DWORD rva = 0;
1915 E : ScopedBstr name_bstr;
1916 : if (FAILED(hr = symbol->get_locationType(&location_type)) ||
1917 : FAILED(hr = symbol->get_relativeVirtualAddress(&rva)) ||
1918 E : FAILED(hr = symbol->get_name(name_bstr.Receive()))) {
1919 i : LOG(ERROR) << "Failed to get data properties: " << com::LogHr(hr) << ".";
1920 i : return DiaBrowser::kBrowserAbort;
1921 : }
1922 :
1923 : // Symbols with an address of zero are essentially invalid. They appear to
1924 : // have been optimized away by the compiler, but they are still reported.
1925 E : if (rva == 0)
1926 E : return DiaBrowser::kBrowserTerminatePath;
1927 :
1928 : // We only care about functions with static storage. We can stop looking at
1929 : // things below this node, as we won't be able to resolve them either.
1930 E : if (location_type != LocIsStatic)
1931 i : return DiaBrowser::kBrowserTerminatePath;
1932 :
1933 : // Get the size of this datum from its type info.
1934 E : size_t length = 0;
1935 E : if (!GetDataSymbolSize(symbol, &length))
1936 i : return DiaBrowser::kBrowserAbort;
1937 :
1938 : // Reuse the parent function block if we can. This acts as small lookup
1939 : // cache.
1940 E : RelativeAddress addr(rva);
1941 E : Block* block = current_block_;
1942 E : RelativeAddress block_addr(current_address_);
1943 E : if (block == NULL || !InRange(addr, block_addr, block->size())) {
1944 E : block = image_->GetBlockByAddress(addr);
1945 E : CHECK(block != NULL);
1946 E : CHECK(image_->GetAddressOf(block, &block_addr));
1947 E : DCHECK(InRange(addr, block_addr, block->size()));
1948 : }
1949 :
1950 E : std::string name;
1951 E : if (!WideToUTF8(name_bstr, name_bstr.Length(), &name)) {
1952 i : LOG(ERROR) << "Failed to convert label name to UTF8.";
1953 i : return DiaBrowser::kBrowserAbort;
1954 : }
1955 :
1956 : // Zero-length data symbols mark case/jump tables, or are forward declares.
1957 E : BlockGraph::LabelAttributes attr = BlockGraph::DATA_LABEL;
1958 E : Offset offset = addr - block_addr;
1959 E : if (length == 0) {
1960 : // Jump and case tables come in as data symbols with no name. Jump tables
1961 : // are always an array of pointers, thus they coincide exactly with a
1962 : // reference. Case tables are simple arrays of integer values (themselves
1963 : // indices into a jump table), thus do not coincide with a reference.
1964 E : if (name.empty() && block->type() == BlockGraph::CODE_BLOCK) {
1965 E : if (block->references().find(offset) != block->references().end()) {
1966 E : name = kJumpTable;
1967 E : attr |= BlockGraph::JUMP_TABLE_LABEL;
1968 E : } else {
1969 E : name = kCaseTable;
1970 E : attr |= BlockGraph::CASE_TABLE_LABEL;
1971 : }
1972 :
1973 : // We expect jump and case tables to already have been discovered by
1974 : // the disassembly operation. If this is not the case then our decoding
1975 : // step is in error and its results can't be trusted.
1976 E : if (!JumpAndCaseTableAlreadyLabelled(block, offset, attr))
1977 i : return DiaBrowser::kBrowserAbort;
1978 E : } else {
1979 : // Zero-length data symbols act as 'forward declares' in some sense. They
1980 : // are always followed by a non-zero length data symbol with the same name
1981 : // and location.
1982 E : return DiaBrowser::kBrowserTerminatePath;
1983 : }
1984 : }
1985 :
1986 : // Verify that the data symbol does not exceed the size of the block.
1987 E : if (addr + length > block_addr + block->size()) {
1988 : // The data symbol can exceed the size of the block in the case of data
1989 : // imports. For some reason the toolchain emits a global data symbol with
1990 : // type information equal to the type of the data *pointed* to by the import
1991 : // entry rather than the type of the entry itself. Thus, if the data type
1992 : // is bigger than the entire IAT this symbol will exceed it. To complicate
1993 : // matters even more, a poorly written module can import its own export in
1994 : // which case a linker generated pseudo-import-entry block will be
1995 : // generated. This won't be part of the IAT, so we can't even filter based
1996 : // on that. Instead, we simply ignore global data symbols that exceed the
1997 : // block size.
1998 E : base::StringPiece spname(name);
1999 E : if (sym_tags.size() == 1 && spname.starts_with("_imp_")) {
2000 E : VLOG(1) << "Encountered an imported data symbol \"" << name << "\" that "
2001 : << "extends past its parent block \"" << block->name() << "\".";
2002 E : } else {
2003 i : LOG(ERROR) << "Received data symbol \"" << name << "\" that extends past "
2004 : << "its parent block \"" << block->name() << "\".";
2005 i : return DiaBrowser::kBrowserAbort;
2006 : }
2007 : }
2008 :
2009 E : if (!AddLabelToBlock(offset, name, attr, block))
2010 i : return DiaBrowser::kBrowserAbort;
2011 :
2012 E : return DiaBrowser::kBrowserContinue;
2013 E : }
2014 :
2015 : DiaBrowser::BrowserDirective NewDecomposer::OnPublicSymbol(
2016 : const DiaBrowser& dia_browser,
2017 : const DiaBrowser::SymTagVector& sym_tags,
2018 E : const DiaBrowser::SymbolPtrVector& symbols) {
2019 E : DCHECK(!symbols.empty());
2020 E : DCHECK_EQ(sym_tags.size(), symbols.size());
2021 E : DCHECK(current_block_ == NULL);
2022 E : DiaBrowser::SymbolPtr symbol = symbols.back();
2023 :
2024 E : HRESULT hr = E_FAIL;
2025 E : DWORD rva = 0;
2026 E : ScopedBstr name_bstr;
2027 : if (FAILED(hr = symbol->get_relativeVirtualAddress(&rva)) ||
2028 E : FAILED(hr = symbol->get_name(name_bstr.Receive()))) {
2029 i : LOG(ERROR) << "Failed to get public symbol properties: " << com::LogHr(hr)
2030 : << ".";
2031 i : return DiaBrowser::kBrowserAbort;
2032 : }
2033 :
2034 E : RelativeAddress addr(rva);
2035 E : Block* block = image_->GetBlockByAddress(addr);
2036 E : CHECK(block != NULL);
2037 E : RelativeAddress block_addr;
2038 E : CHECK(image_->GetAddressOf(block, &block_addr));
2039 E : DCHECK(InRange(addr, block_addr, block->size()));
2040 :
2041 E : std::string name;
2042 E : WideToUTF8(name_bstr, name_bstr.Length(), &name);
2043 :
2044 : // Public symbol names are mangled. Remove leading '_' as per
2045 : // http://msdn.microsoft.com/en-us/library/00kh39zz(v=vs.80).aspx
2046 E : if (name[0] == '_')
2047 E : name = name.substr(1);
2048 :
2049 E : Offset offset = addr - block_addr;
2050 E : if (!AddLabelToBlock(offset, name, BlockGraph::PUBLIC_SYMBOL_LABEL, block))
2051 i : return DiaBrowser::kBrowserAbort;
2052 :
2053 E : return DiaBrowser::kBrowserContinue;
2054 E : }
2055 :
2056 : DiaBrowser::BrowserDirective NewDecomposer::OnLabelSymbol(
2057 : const DiaBrowser& dia_browser,
2058 : const DiaBrowser::SymTagVector& sym_tags,
2059 E : const DiaBrowser::SymbolPtrVector& symbols) {
2060 E : DCHECK(!symbols.empty());
2061 E : DCHECK_EQ(sym_tags.size(), symbols.size());
2062 E : DiaBrowser::SymbolPtr symbol = symbols.back();
2063 :
2064 E : HRESULT hr = E_FAIL;
2065 E : DWORD rva = 0;
2066 E : ScopedBstr name_bstr;
2067 : if (FAILED(hr = symbol->get_relativeVirtualAddress(&rva)) ||
2068 E : FAILED(hr = symbol->get_name(name_bstr.Receive()))) {
2069 i : LOG(ERROR) << "Failed to get label symbol properties: " << com::LogHr(hr)
2070 : << ".";
2071 i : return DiaBrowser::kBrowserAbort;
2072 : }
2073 :
2074 : // If we have a current_block_ the label should lie within its scope.
2075 E : RelativeAddress addr(rva);
2076 E : Block* block = current_block_;
2077 E : RelativeAddress block_addr(current_address_);
2078 E : if (block != NULL) {
2079 E : if (!InRangeIncl(addr, current_address_, current_block_->size())) {
2080 i : LOG(ERROR) << "Label falls outside of current block \""
2081 : << current_block_->name() << "\".";
2082 i : return DiaBrowser::kBrowserAbort;
2083 : }
2084 E : } else {
2085 : // If there is no current block this is a compiland scope label.
2086 E : block = image_->GetBlockByAddress(addr);
2087 E : CHECK(block != NULL);
2088 E : CHECK(image_->GetAddressOf(block, &block_addr));
2089 E : DCHECK(InRange(addr, block_addr, block->size()));
2090 :
2091 : // TODO(chrisha): This label is in compiland scope, so we should be
2092 : // finding the block whose section contribution shares the same
2093 : // compiland.
2094 : }
2095 :
2096 E : std::string name;
2097 E : WideToUTF8(name_bstr, name_bstr.Length(), &name);
2098 :
2099 E : Offset offset = addr - block_addr;
2100 E : if (!AddLabelToBlock(offset, name, BlockGraph::CODE_LABEL, block))
2101 i : return DiaBrowser::kBrowserAbort;
2102 :
2103 E : return DiaBrowser::kBrowserContinue;
2104 E : }
2105 :
2106 : DiaBrowser::BrowserDirective NewDecomposer::OnScopeSymbol(
2107 E : enum SymTagEnum type, DiaBrowser::SymbolPtr symbol) {
2108 : // We should only get here via the successful exploration of a SymTagFunction,
2109 : // so current_block_ should be set.
2110 E : DCHECK(current_block_ != NULL);
2111 :
2112 E : HRESULT hr = E_FAIL;
2113 E : DWORD rva = 0;
2114 E : if (FAILED(hr = symbol->get_relativeVirtualAddress(&rva))) {
2115 i : LOG(ERROR) << "Failed to get scope symbol properties: " << com::LogHr(hr)
2116 : << ".";
2117 i : return DiaBrowser::kBrowserAbort;
2118 : }
2119 :
2120 : // The label may potentially lay at the first byte past the function.
2121 E : RelativeAddress addr(rva);
2122 E : DCHECK_LE(current_address_, addr);
2123 E : DCHECK_LE(addr, current_address_ + current_block_->size());
2124 :
2125 : // Get the attributes for this label.
2126 E : BlockGraph::LabelAttributes attr = 0;
2127 E : std::string name;
2128 E : CHECK(ScopeSymTagToLabelProperties(type, current_scope_count_, &attr, &name));
2129 :
2130 : // Add the label.
2131 E : Offset offset = addr - current_address_;
2132 E : if (!AddLabelToBlock(offset, name, attr, current_block_))
2133 i : return DiaBrowser::kBrowserAbort;
2134 :
2135 : // If this is a scope we extract the length and explicitly add a corresponding
2136 : // end label.
2137 E : if (type == SymTagBlock) {
2138 E : ULONGLONG length = 0;
2139 E : if (symbol->get_length(&length) != S_OK) {
2140 i : LOG(ERROR) << "Failed to extract code scope length for block \""
2141 : << current_block_->name() << "\".";
2142 i : return DiaBrowser::kBrowserAbort;
2143 : }
2144 E : DCHECK_LE(static_cast<size_t>(offset + length), current_block_->size());
2145 E : name = base::StringPrintf("<scope-end-%d>", current_scope_count_);
2146 E : ++current_scope_count_;
2147 : if (!AddLabelToBlock(offset + length, name,
2148 E : BlockGraph::SCOPE_END_LABEL, current_block_)) {
2149 i : return DiaBrowser::kBrowserAbort;
2150 : }
2151 : }
2152 :
2153 E : return DiaBrowser::kBrowserContinue;
2154 E : }
2155 :
2156 : DiaBrowser::BrowserDirective NewDecomposer::OnCallSiteSymbol(
2157 E : DiaBrowser::SymbolPtr symbol) {
2158 : // We should only get here via the successful exploration of a SymTagFunction,
2159 : // so current_block_ should be set.
2160 E : DCHECK(current_block_ != NULL);
2161 :
2162 E : HRESULT hr = E_FAIL;
2163 E : DWORD rva = 0;
2164 E : if (FAILED(hr = symbol->get_relativeVirtualAddress(&rva))) {
2165 i : LOG(ERROR) << "Failed to get call site symbol properties: "
2166 : << com::LogHr(hr) << ".";
2167 i : return DiaBrowser::kBrowserAbort;
2168 : }
2169 :
2170 E : RelativeAddress addr(rva);
2171 E : if (!InRange(addr, current_address_, current_block_->size())) {
2172 i : LOG(ERROR) << "Call site falls outside of current block \""
2173 : << current_block_->name() << "\".";
2174 i : return DiaBrowser::kBrowserAbort;
2175 : }
2176 :
2177 E : Offset offset = addr - current_address_;
2178 : if (!AddLabelToBlock(offset, "<call-site>", BlockGraph::CALL_SITE_LABEL,
2179 E : current_block_)) {
2180 i : return DiaBrowser::kBrowserAbort;
2181 : }
2182 :
2183 E : return DiaBrowser::kBrowserContinue;
2184 E : }
2185 :
2186 : Block* NewDecomposer::CreateBlock(BlockType type,
2187 : RelativeAddress address,
2188 : BlockGraph::Size size,
2189 E : const base::StringPiece& name) {
2190 E : Block* block = image_->AddBlock(type, address, size, name);
2191 E : if (block == NULL) {
2192 i : LOG(ERROR) << "Unable to add block \"" << name.as_string() << "\" at "
2193 : << address << " with size " << size << ".";
2194 i : return NULL;
2195 : }
2196 :
2197 : // Mark the source range from whence this block originates. This is assuming
2198 : // an untransformed image. To handle transformed images we'd have to use the
2199 : // OMAP information to do this properly.
2200 : bool pushed = block->source_ranges().Push(
2201 : Block::DataRange(0, size),
2202 E : Block::SourceRange(address, size));
2203 E : DCHECK(pushed);
2204 :
2205 E : BlockGraph::SectionId section = image_file_.GetSectionIndex(address, size);
2206 E : if (section == BlockGraph::kInvalidSectionId) {
2207 i : LOG(ERROR) << "Block \"" << name.as_string() << "\" at " << address
2208 : << " with size " << size << " lies outside of all sections.";
2209 i : return NULL;
2210 : }
2211 E : block->set_section(section);
2212 :
2213 E : const uint8* data = image_file_.GetImageData(address, size);
2214 E : if (data != NULL)
2215 E : block->SetData(data, size);
2216 :
2217 E : return block;
2218 E : }
2219 :
2220 : Block* NewDecomposer::CreateBlockOrFindCoveringPeBlock(
2221 : BlockType type,
2222 : RelativeAddress addr,
2223 : BlockGraph::Size size,
2224 E : const base::StringPiece& name) {
2225 E : Block* block = image_->GetBlockByAddress(addr);
2226 E : if (block != NULL) {
2227 E : RelativeAddress block_addr;
2228 E : CHECK(image_->GetAddressOf(block, &block_addr));
2229 :
2230 E : RelativeRange existing_block(block_addr, block->size());
2231 :
2232 : // If this is not a PE parsed or COFF group block that covers us entirely,
2233 : // then this is an error.
2234 : static const BlockGraph::BlockAttributes kCoveringAttributes =
2235 : BlockGraph::PE_PARSED | BlockGraph::COFF_GROUP;
2236 : if ((block->attributes() & kCoveringAttributes) == 0 ||
2237 E : !existing_block.Contains(addr, size)) {
2238 i : LOG(ERROR) << "Trying to create block \"" << name.as_string() << "\" at "
2239 : << addr.value() << " with size " << size << " that conflicts "
2240 : << "with existing block \"" << block->name() << " at "
2241 : << block_addr << " with size " << block->size() << ".";
2242 i : return NULL;
2243 : }
2244 :
2245 E : return block;
2246 : }
2247 E : DCHECK(block == NULL);
2248 :
2249 E : return CreateBlock(type, addr, size, name);
2250 E : }
2251 :
2252 : bool NewDecomposer::CreateGapBlock(BlockType block_type,
2253 : RelativeAddress address,
2254 E : BlockGraph::Size size) {
2255 : Block* block = CreateBlock(block_type, address, size,
2256 E : base::StringPrintf("Gap Block 0x%08X", address.value()).c_str());
2257 E : if (block == NULL) {
2258 i : LOG(ERROR) << "Unable to create gap block.";
2259 i : return false;
2260 : }
2261 E : block->set_attribute(BlockGraph::GAP_BLOCK);
2262 :
2263 E : return true;
2264 E : }
2265 :
2266 : bool NewDecomposer::CreateSectionGapBlocks(const IMAGE_SECTION_HEADER* header,
2267 E : BlockType block_type) {
2268 E : RelativeAddress section_begin(header->VirtualAddress);
2269 E : RelativeAddress section_end(section_begin + header->Misc.VirtualSize);
2270 : RelativeAddress image_end(
2271 E : image_file_.nt_headers()->OptionalHeader.SizeOfImage);
2272 :
2273 : // Search for the first and last blocks interesting from the start and end
2274 : // of the section to the end of the image.
2275 : BlockGraph::AddressSpace::RangeMap::const_iterator it(
2276 : image_->address_space_impl().FindFirstIntersection(
2277 : BlockGraph::AddressSpace::Range(section_begin,
2278 E : image_end - section_begin)));
2279 :
2280 : BlockGraph::AddressSpace::RangeMap::const_iterator end =
2281 E : image_->address_space_impl().end();
2282 E : if (section_end < image_end) {
2283 : end = image_->address_space_impl().FindFirstIntersection(
2284 : BlockGraph::AddressSpace::Range(section_end,
2285 E : image_end - section_end));
2286 : }
2287 :
2288 : // The whole section is missing. Cover it with one gap block.
2289 E : if (it == end)
2290 : return CreateGapBlock(
2291 i : block_type, section_begin, section_end - section_begin);
2292 :
2293 : // Create the head gap block if need be.
2294 E : if (section_begin < it->first.start()) {
2295 : if (!CreateGapBlock(
2296 i : block_type, section_begin, it->first.start() - section_begin)) {
2297 i : return false;
2298 : }
2299 : }
2300 :
2301 : // Now iterate the blocks and fill in gaps.
2302 E : for (; it != end; ++it) {
2303 E : const Block* block = it->second;
2304 E : DCHECK(block != NULL);
2305 E : RelativeAddress block_end = it->first.start() + block->size();
2306 E : if (block_end >= section_end)
2307 E : break;
2308 :
2309 : // Walk to the next address in turn.
2310 E : BlockGraph::AddressSpace::RangeMap::const_iterator next = it;
2311 E : ++next;
2312 E : if (next == end) {
2313 : // We're at the end of the list. Create the tail gap block.
2314 E : DCHECK_GT(section_end, block_end);
2315 E : if (!CreateGapBlock(block_type, block_end, section_end - block_end))
2316 i : return false;
2317 E : break;
2318 : }
2319 :
2320 : // Create the interstitial gap block.
2321 E : if (block_end < next->first.start())
2322 : if (!CreateGapBlock(
2323 E : block_type, block_end, next->first.start() - block_end)) {
2324 i : return false;
2325 : }
2326 E : }
2327 :
2328 E : return true;
2329 E : }
2330 :
2331 : } // namespace pe
|