1 //===-- SymbolizableObjectFile.cpp ----------------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // Implementation of SymbolizableObjectFile class.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "SymbolizableObjectFile.h"
15 #include "llvm/Object/SymbolSize.h"
16 #include "llvm/Support/DataExtractor.h"
17 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
18 
19 namespace llvm {
20 namespace symbolize {
21 
22 using namespace object;
23 
24 static DILineInfoSpecifier
getDILineInfoSpecifier(FunctionNameKind FNKind)25 getDILineInfoSpecifier(FunctionNameKind FNKind) {
26   return DILineInfoSpecifier(
27       DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FNKind);
28 }
29 
30 ErrorOr<std::unique_ptr<SymbolizableObjectFile>>
create(object::ObjectFile * Obj,std::unique_ptr<DIContext> DICtx)31 SymbolizableObjectFile::create(object::ObjectFile *Obj,
32                                std::unique_ptr<DIContext> DICtx) {
33   std::unique_ptr<SymbolizableObjectFile> res(
34       new SymbolizableObjectFile(Obj, std::move(DICtx)));
35   std::unique_ptr<DataExtractor> OpdExtractor;
36   uint64_t OpdAddress = 0;
37   // Find the .opd (function descriptor) section if any, for big-endian
38   // PowerPC64 ELF.
39   if (Obj->getArch() == Triple::ppc64) {
40     for (section_iterator Section : Obj->sections()) {
41       StringRef Name;
42       StringRef Data;
43       if (auto EC = Section->getName(Name))
44         return EC;
45       if (Name == ".opd") {
46         if (auto EC = Section->getContents(Data))
47           return EC;
48         OpdExtractor.reset(new DataExtractor(Data, Obj->isLittleEndian(),
49                                              Obj->getBytesInAddress()));
50         OpdAddress = Section->getAddress();
51         break;
52       }
53     }
54   }
55   std::vector<std::pair<SymbolRef, uint64_t>> Symbols =
56       computeSymbolSizes(*Obj);
57   for (auto &P : Symbols)
58     res->addSymbol(P.first, P.second, OpdExtractor.get(), OpdAddress);
59 
60   // If this is a COFF object and we didn't find any symbols, try the export
61   // table.
62   if (Symbols.empty()) {
63     if (auto *CoffObj = dyn_cast<COFFObjectFile>(Obj))
64       if (auto EC = res->addCoffExportSymbols(CoffObj))
65         return EC;
66   }
67   return std::move(res);
68 }
69 
SymbolizableObjectFile(ObjectFile * Obj,std::unique_ptr<DIContext> DICtx)70 SymbolizableObjectFile::SymbolizableObjectFile(ObjectFile *Obj,
71                                                std::unique_ptr<DIContext> DICtx)
72     : Module(Obj), DebugInfoContext(std::move(DICtx)) {}
73 
74 namespace {
75 struct OffsetNamePair {
76   uint32_t Offset;
77   StringRef Name;
operator <llvm::symbolize::__anon3715c9e50111::OffsetNamePair78   bool operator<(const OffsetNamePair &R) const {
79     return Offset < R.Offset;
80   }
81 };
82 }
83 
addCoffExportSymbols(const COFFObjectFile * CoffObj)84 std::error_code SymbolizableObjectFile::addCoffExportSymbols(
85     const COFFObjectFile *CoffObj) {
86   // Get all export names and offsets.
87   std::vector<OffsetNamePair> ExportSyms;
88   for (const ExportDirectoryEntryRef &Ref : CoffObj->export_directories()) {
89     StringRef Name;
90     uint32_t Offset;
91     if (auto EC = Ref.getSymbolName(Name))
92       return EC;
93     if (auto EC = Ref.getExportRVA(Offset))
94       return EC;
95     ExportSyms.push_back(OffsetNamePair{Offset, Name});
96   }
97   if (ExportSyms.empty())
98     return std::error_code();
99 
100   // Sort by ascending offset.
101   array_pod_sort(ExportSyms.begin(), ExportSyms.end());
102 
103   // Approximate the symbol sizes by assuming they run to the next symbol.
104   // FIXME: This assumes all exports are functions.
105   uint64_t ImageBase = CoffObj->getImageBase();
106   for (auto I = ExportSyms.begin(), E = ExportSyms.end(); I != E; ++I) {
107     OffsetNamePair &Export = *I;
108     // FIXME: The last export has a one byte size now.
109     uint32_t NextOffset = I != E ? I->Offset : Export.Offset + 1;
110     uint64_t SymbolStart = ImageBase + Export.Offset;
111     uint64_t SymbolSize = NextOffset - Export.Offset;
112     SymbolDesc SD = {SymbolStart, SymbolSize};
113     Functions.insert(std::make_pair(SD, Export.Name));
114   }
115   return std::error_code();
116 }
117 
addSymbol(const SymbolRef & Symbol,uint64_t SymbolSize,DataExtractor * OpdExtractor,uint64_t OpdAddress)118 std::error_code SymbolizableObjectFile::addSymbol(const SymbolRef &Symbol,
119                                                   uint64_t SymbolSize,
120                                                   DataExtractor *OpdExtractor,
121                                                   uint64_t OpdAddress) {
122   SymbolRef::Type SymbolType = Symbol.getType();
123   if (SymbolType != SymbolRef::ST_Function && SymbolType != SymbolRef::ST_Data)
124     return std::error_code();
125   ErrorOr<uint64_t> SymbolAddressOrErr = Symbol.getAddress();
126   if (auto EC = SymbolAddressOrErr.getError())
127     return EC;
128   uint64_t SymbolAddress = *SymbolAddressOrErr;
129   if (OpdExtractor) {
130     // For big-endian PowerPC64 ELF, symbols in the .opd section refer to
131     // function descriptors. The first word of the descriptor is a pointer to
132     // the function's code.
133     // For the purposes of symbolization, pretend the symbol's address is that
134     // of the function's code, not the descriptor.
135     uint64_t OpdOffset = SymbolAddress - OpdAddress;
136     uint32_t OpdOffset32 = OpdOffset;
137     if (OpdOffset == OpdOffset32 &&
138         OpdExtractor->isValidOffsetForAddress(OpdOffset32))
139       SymbolAddress = OpdExtractor->getAddress(&OpdOffset32);
140   }
141   ErrorOr<StringRef> SymbolNameOrErr = Symbol.getName();
142   if (auto EC = SymbolNameOrErr.getError())
143     return EC;
144   StringRef SymbolName = *SymbolNameOrErr;
145   // Mach-O symbol table names have leading underscore, skip it.
146   if (Module->isMachO() && SymbolName.size() > 0 && SymbolName[0] == '_')
147     SymbolName = SymbolName.drop_front();
148   // FIXME: If a function has alias, there are two entries in symbol table
149   // with same address size. Make sure we choose the correct one.
150   auto &M = SymbolType == SymbolRef::ST_Function ? Functions : Objects;
151   SymbolDesc SD = { SymbolAddress, SymbolSize };
152   M.insert(std::make_pair(SD, SymbolName));
153   return std::error_code();
154 }
155 
156 // Return true if this is a 32-bit x86 PE COFF module.
isWin32Module() const157 bool SymbolizableObjectFile::isWin32Module() const {
158   auto *CoffObject = dyn_cast<COFFObjectFile>(Module);
159   return CoffObject && CoffObject->getMachine() == COFF::IMAGE_FILE_MACHINE_I386;
160 }
161 
getModulePreferredBase() const162 uint64_t SymbolizableObjectFile::getModulePreferredBase() const {
163   if (auto *CoffObject = dyn_cast<COFFObjectFile>(Module))
164     return CoffObject->getImageBase();
165   return 0;
166 }
167 
getNameFromSymbolTable(SymbolRef::Type Type,uint64_t Address,std::string & Name,uint64_t & Addr,uint64_t & Size) const168 bool SymbolizableObjectFile::getNameFromSymbolTable(SymbolRef::Type Type,
169                                                     uint64_t Address,
170                                                     std::string &Name,
171                                                     uint64_t &Addr,
172                                                     uint64_t &Size) const {
173   const auto &SymbolMap = Type == SymbolRef::ST_Function ? Functions : Objects;
174   if (SymbolMap.empty())
175     return false;
176   SymbolDesc SD = { Address, Address };
177   auto SymbolIterator = SymbolMap.upper_bound(SD);
178   if (SymbolIterator == SymbolMap.begin())
179     return false;
180   --SymbolIterator;
181   if (SymbolIterator->first.Size != 0 &&
182       SymbolIterator->first.Addr + SymbolIterator->first.Size <= Address)
183     return false;
184   Name = SymbolIterator->second.str();
185   Addr = SymbolIterator->first.Addr;
186   Size = SymbolIterator->first.Size;
187   return true;
188 }
189 
shouldOverrideWithSymbolTable(FunctionNameKind FNKind,bool UseSymbolTable) const190 bool SymbolizableObjectFile::shouldOverrideWithSymbolTable(
191     FunctionNameKind FNKind, bool UseSymbolTable) const {
192   // When DWARF is used with -gline-tables-only / -gmlt, the symbol table gives
193   // better answers for linkage names than the DIContext. Otherwise, we are
194   // probably using PEs and PDBs, and we shouldn't do the override. PE files
195   // generally only contain the names of exported symbols.
196   return FNKind == FunctionNameKind::LinkageName && UseSymbolTable &&
197          isa<DWARFContext>(DebugInfoContext.get());
198 }
199 
symbolizeCode(uint64_t ModuleOffset,FunctionNameKind FNKind,bool UseSymbolTable) const200 DILineInfo SymbolizableObjectFile::symbolizeCode(uint64_t ModuleOffset,
201                                                  FunctionNameKind FNKind,
202                                                  bool UseSymbolTable) const {
203   DILineInfo LineInfo;
204   if (DebugInfoContext) {
205     LineInfo = DebugInfoContext->getLineInfoForAddress(
206         ModuleOffset, getDILineInfoSpecifier(FNKind));
207   }
208   // Override function name from symbol table if necessary.
209   if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) {
210     std::string FunctionName;
211     uint64_t Start, Size;
212     if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
213                                FunctionName, Start, Size)) {
214       LineInfo.FunctionName = FunctionName;
215     }
216   }
217   return LineInfo;
218 }
219 
symbolizeInlinedCode(uint64_t ModuleOffset,FunctionNameKind FNKind,bool UseSymbolTable) const220 DIInliningInfo SymbolizableObjectFile::symbolizeInlinedCode(
221     uint64_t ModuleOffset, FunctionNameKind FNKind, bool UseSymbolTable) const {
222   DIInliningInfo InlinedContext;
223 
224   if (DebugInfoContext)
225     InlinedContext = DebugInfoContext->getInliningInfoForAddress(
226         ModuleOffset, getDILineInfoSpecifier(FNKind));
227   // Make sure there is at least one frame in context.
228   if (InlinedContext.getNumberOfFrames() == 0)
229     InlinedContext.addFrame(DILineInfo());
230 
231   // Override the function name in lower frame with name from symbol table.
232   if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) {
233     std::string FunctionName;
234     uint64_t Start, Size;
235     if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
236                                FunctionName, Start, Size)) {
237       InlinedContext.getMutableFrame(InlinedContext.getNumberOfFrames() - 1)
238           ->FunctionName = FunctionName;
239     }
240   }
241 
242   return InlinedContext;
243 }
244 
symbolizeData(uint64_t ModuleOffset) const245 DIGlobal SymbolizableObjectFile::symbolizeData(uint64_t ModuleOffset) const {
246   DIGlobal Res;
247   getNameFromSymbolTable(SymbolRef::ST_Data, ModuleOffset, Res.Name, Res.Start,
248                          Res.Size);
249   return Res;
250 }
251 
252 }  // namespace symbolize
253 }  // namespace llvm
254 
255