1 //===- SymbolizableObjectFile.cpp -----------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // Implementation of SymbolizableObjectFile class.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/DebugInfo/Symbolize/SymbolizableObjectFile.h"
14 #include "llvm/ADT/STLExtras.h"
15 #include "llvm/ADT/Triple.h"
16 #include "llvm/BinaryFormat/COFF.h"
17 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
18 #include "llvm/Object/COFF.h"
19 #include "llvm/Object/ELFObjectFile.h"
20 #include "llvm/Object/ObjectFile.h"
21 #include "llvm/Object/SymbolSize.h"
22 #include "llvm/Support/Casting.h"
23 #include "llvm/Support/DataExtractor.h"
24 #include <algorithm>
25 
26 using namespace llvm;
27 using namespace object;
28 using namespace symbolize;
29 
30 Expected<std::unique_ptr<SymbolizableObjectFile>>
31 SymbolizableObjectFile::create(const object::ObjectFile *Obj,
32                                std::unique_ptr<DIContext> DICtx,
33                                bool UntagAddresses) {
34   assert(DICtx);
35   std::unique_ptr<SymbolizableObjectFile> res(
36       new SymbolizableObjectFile(Obj, std::move(DICtx), UntagAddresses));
37   std::unique_ptr<DataExtractor> OpdExtractor;
38   uint64_t OpdAddress = 0;
39   // Find the .opd (function descriptor) section if any, for big-endian
40   // PowerPC64 ELF.
41   if (Obj->getArch() == Triple::ppc64) {
42     for (section_iterator Section : Obj->sections()) {
43       Expected<StringRef> NameOrErr = Section->getName();
44       if (!NameOrErr)
45         return NameOrErr.takeError();
46 
47       if (*NameOrErr == ".opd") {
48         Expected<StringRef> E = Section->getContents();
49         if (!E)
50           return E.takeError();
51         OpdExtractor.reset(new DataExtractor(*E, Obj->isLittleEndian(),
52                                              Obj->getBytesInAddress()));
53         OpdAddress = Section->getAddress();
54         break;
55       }
56     }
57   }
58   std::vector<std::pair<SymbolRef, uint64_t>> Symbols =
59       computeSymbolSizes(*Obj);
60   for (auto &P : Symbols)
61     if (Error E =
62             res->addSymbol(P.first, P.second, OpdExtractor.get(), OpdAddress))
63       return std::move(E);
64 
65   // If this is a COFF object and we didn't find any symbols, try the export
66   // table.
67   if (Symbols.empty()) {
68     if (auto *CoffObj = dyn_cast<COFFObjectFile>(Obj))
69       if (Error E = res->addCoffExportSymbols(CoffObj))
70         return std::move(E);
71   }
72 
73   std::vector<SymbolDesc> &SS = res->Symbols;
74   // Sort by (Addr,Size,Name). If several SymbolDescs share the same Addr,
75   // pick the one with the largest Size. This helps us avoid symbols with no
76   // size information (Size=0).
77   llvm::stable_sort(SS);
78   auto I = SS.begin(), E = SS.end(), J = SS.begin();
79   while (I != E) {
80     auto OI = I;
81     while (++I != E && OI->Addr == I->Addr) {
82     }
83     *J++ = I[-1];
84   }
85   SS.erase(J, SS.end());
86 
87   return std::move(res);
88 }
89 
90 SymbolizableObjectFile::SymbolizableObjectFile(const ObjectFile *Obj,
91                                                std::unique_ptr<DIContext> DICtx,
92                                                bool UntagAddresses)
93     : Module(Obj), DebugInfoContext(std::move(DICtx)),
94       UntagAddresses(UntagAddresses) {}
95 
96 namespace {
97 
98 struct OffsetNamePair {
99   uint32_t Offset;
100   StringRef Name;
101 
102   bool operator<(const OffsetNamePair &R) const {
103     return Offset < R.Offset;
104   }
105 };
106 
107 } // end anonymous namespace
108 
109 Error SymbolizableObjectFile::addCoffExportSymbols(
110     const COFFObjectFile *CoffObj) {
111   // Get all export names and offsets.
112   std::vector<OffsetNamePair> ExportSyms;
113   for (const ExportDirectoryEntryRef &Ref : CoffObj->export_directories()) {
114     StringRef Name;
115     uint32_t Offset;
116     if (auto EC = Ref.getSymbolName(Name))
117       return EC;
118     if (auto EC = Ref.getExportRVA(Offset))
119       return EC;
120     ExportSyms.push_back(OffsetNamePair{Offset, Name});
121   }
122   if (ExportSyms.empty())
123     return Error::success();
124 
125   // Sort by ascending offset.
126   array_pod_sort(ExportSyms.begin(), ExportSyms.end());
127 
128   // Approximate the symbol sizes by assuming they run to the next symbol.
129   // FIXME: This assumes all exports are functions.
130   uint64_t ImageBase = CoffObj->getImageBase();
131   for (auto I = ExportSyms.begin(), E = ExportSyms.end(); I != E; ++I) {
132     OffsetNamePair &Export = *I;
133     // FIXME: The last export has a one byte size now.
134     uint32_t NextOffset = I != E ? I->Offset : Export.Offset + 1;
135     uint64_t SymbolStart = ImageBase + Export.Offset;
136     uint64_t SymbolSize = NextOffset - Export.Offset;
137     Symbols.push_back({SymbolStart, SymbolSize, Export.Name, 0});
138   }
139   return Error::success();
140 }
141 
142 Error SymbolizableObjectFile::addSymbol(const SymbolRef &Symbol,
143                                         uint64_t SymbolSize,
144                                         DataExtractor *OpdExtractor,
145                                         uint64_t OpdAddress) {
146   // Avoid adding symbols from an unknown/undefined section.
147   const ObjectFile &Obj = *Symbol.getObject();
148   Expected<StringRef> SymbolNameOrErr = Symbol.getName();
149   if (!SymbolNameOrErr)
150     return SymbolNameOrErr.takeError();
151   StringRef SymbolName = *SymbolNameOrErr;
152 
153   uint32_t ELFSymIdx =
154       Obj.isELF() ? ELFSymbolRef(Symbol).getRawDataRefImpl().d.b : 0;
155   Expected<section_iterator> Sec = Symbol.getSection();
156   if (!Sec || Obj.section_end() == *Sec) {
157     if (Obj.isELF()) {
158       // Store the (index, filename) pair for a file symbol.
159       ELFSymbolRef ESym(Symbol);
160       if (ESym.getELFType() == ELF::STT_FILE)
161         FileSymbols.emplace_back(ELFSymIdx, SymbolName);
162     }
163     return Error::success();
164   }
165 
166   Expected<SymbolRef::Type> SymbolTypeOrErr = Symbol.getType();
167   if (!SymbolTypeOrErr)
168     return SymbolTypeOrErr.takeError();
169   SymbolRef::Type SymbolType = *SymbolTypeOrErr;
170   if (Obj.isELF()) {
171     // Allow function and data symbols. Additionally allow STT_NONE, which are
172     // common for functions defined in assembly.
173     uint8_t Type = ELFSymbolRef(Symbol).getELFType();
174     if (Type != ELF::STT_NOTYPE && Type != ELF::STT_FUNC &&
175         Type != ELF::STT_OBJECT && Type != ELF::STT_GNU_IFUNC)
176       return Error::success();
177     // Some STT_NOTYPE symbols are not desired. This excludes STT_SECTION and
178     // ARM mapping symbols.
179     uint32_t Flags = cantFail(Symbol.getFlags());
180     if (Flags & SymbolRef::SF_FormatSpecific)
181       return Error::success();
182   } else if (SymbolType != SymbolRef::ST_Function &&
183              SymbolType != SymbolRef::ST_Data) {
184     return Error::success();
185   }
186 
187   Expected<uint64_t> SymbolAddressOrErr = Symbol.getAddress();
188   if (!SymbolAddressOrErr)
189     return SymbolAddressOrErr.takeError();
190   uint64_t SymbolAddress = *SymbolAddressOrErr;
191   if (UntagAddresses) {
192     // For kernel addresses, bits 56-63 need to be set, so we sign extend bit 55
193     // into bits 56-63 instead of masking them out.
194     SymbolAddress &= (1ull << 56) - 1;
195     SymbolAddress = (int64_t(SymbolAddress) << 8) >> 8;
196   }
197   if (OpdExtractor) {
198     // For big-endian PowerPC64 ELF, symbols in the .opd section refer to
199     // function descriptors. The first word of the descriptor is a pointer to
200     // the function's code.
201     // For the purposes of symbolization, pretend the symbol's address is that
202     // of the function's code, not the descriptor.
203     uint64_t OpdOffset = SymbolAddress - OpdAddress;
204     if (OpdExtractor->isValidOffsetForAddress(OpdOffset))
205       SymbolAddress = OpdExtractor->getAddress(&OpdOffset);
206   }
207   // Mach-O symbol table names have leading underscore, skip it.
208   if (Module->isMachO() && !SymbolName.empty() && SymbolName[0] == '_')
209     SymbolName = SymbolName.drop_front();
210 
211   if (Obj.isELF() && ELFSymbolRef(Symbol).getBinding() != ELF::STB_LOCAL)
212     ELFSymIdx = 0;
213   Symbols.push_back({SymbolAddress, SymbolSize, SymbolName, ELFSymIdx});
214   return Error::success();
215 }
216 
217 // Return true if this is a 32-bit x86 PE COFF module.
218 bool SymbolizableObjectFile::isWin32Module() const {
219   auto *CoffObject = dyn_cast<COFFObjectFile>(Module);
220   return CoffObject && CoffObject->getMachine() == COFF::IMAGE_FILE_MACHINE_I386;
221 }
222 
223 uint64_t SymbolizableObjectFile::getModulePreferredBase() const {
224   if (auto *CoffObject = dyn_cast<COFFObjectFile>(Module))
225     return CoffObject->getImageBase();
226   return 0;
227 }
228 
229 bool SymbolizableObjectFile::getNameFromSymbolTable(
230     uint64_t Address, std::string &Name, uint64_t &Addr, uint64_t &Size,
231     std::string &FileName) const {
232   SymbolDesc SD{Address, UINT64_C(-1), StringRef(), 0};
233   auto SymbolIterator = llvm::upper_bound(Symbols, SD);
234   if (SymbolIterator == Symbols.begin())
235     return false;
236   --SymbolIterator;
237   if (SymbolIterator->Size != 0 &&
238       SymbolIterator->Addr + SymbolIterator->Size <= Address)
239     return false;
240   Name = SymbolIterator->Name.str();
241   Addr = SymbolIterator->Addr;
242   Size = SymbolIterator->Size;
243 
244   if (SymbolIterator->ELFLocalSymIdx != 0) {
245     // If this is an ELF local symbol, find the STT_FILE symbol preceding
246     // SymbolIterator to get the filename. The ELF spec requires the STT_FILE
247     // symbol (if present) precedes the other STB_LOCAL symbols for the file.
248     assert(Module->isELF());
249     auto It = llvm::upper_bound(
250         FileSymbols,
251         std::make_pair(SymbolIterator->ELFLocalSymIdx, StringRef()));
252     if (It != FileSymbols.begin())
253       FileName = It[-1].second.str();
254   }
255   return true;
256 }
257 
258 bool SymbolizableObjectFile::shouldOverrideWithSymbolTable(
259     FunctionNameKind FNKind, bool UseSymbolTable) const {
260   // When DWARF is used with -gline-tables-only / -gmlt, the symbol table gives
261   // better answers for linkage names than the DIContext. Otherwise, we are
262   // probably using PEs and PDBs, and we shouldn't do the override. PE files
263   // generally only contain the names of exported symbols.
264   return FNKind == FunctionNameKind::LinkageName && UseSymbolTable &&
265          isa<DWARFContext>(DebugInfoContext.get());
266 }
267 
268 DILineInfo
269 SymbolizableObjectFile::symbolizeCode(object::SectionedAddress ModuleOffset,
270                                       DILineInfoSpecifier LineInfoSpecifier,
271                                       bool UseSymbolTable) const {
272   if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
273     ModuleOffset.SectionIndex =
274         getModuleSectionIndexForAddress(ModuleOffset.Address);
275   DILineInfo LineInfo =
276       DebugInfoContext->getLineInfoForAddress(ModuleOffset, LineInfoSpecifier);
277 
278   // Override function name from symbol table if necessary.
279   if (shouldOverrideWithSymbolTable(LineInfoSpecifier.FNKind, UseSymbolTable)) {
280     std::string FunctionName, FileName;
281     uint64_t Start, Size;
282     if (getNameFromSymbolTable(ModuleOffset.Address, FunctionName, Start, Size,
283                                FileName)) {
284       LineInfo.FunctionName = FunctionName;
285       LineInfo.StartAddress = Start;
286       if (LineInfo.FileName == DILineInfo::BadString && !FileName.empty())
287         LineInfo.FileName = FileName;
288     }
289   }
290   return LineInfo;
291 }
292 
293 DIInliningInfo SymbolizableObjectFile::symbolizeInlinedCode(
294     object::SectionedAddress ModuleOffset,
295     DILineInfoSpecifier LineInfoSpecifier, bool UseSymbolTable) const {
296   if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
297     ModuleOffset.SectionIndex =
298         getModuleSectionIndexForAddress(ModuleOffset.Address);
299   DIInliningInfo InlinedContext = DebugInfoContext->getInliningInfoForAddress(
300       ModuleOffset, LineInfoSpecifier);
301 
302   // Make sure there is at least one frame in context.
303   if (InlinedContext.getNumberOfFrames() == 0)
304     InlinedContext.addFrame(DILineInfo());
305 
306   // Override the function name in lower frame with name from symbol table.
307   if (shouldOverrideWithSymbolTable(LineInfoSpecifier.FNKind, UseSymbolTable)) {
308     std::string FunctionName, FileName;
309     uint64_t Start, Size;
310     if (getNameFromSymbolTable(ModuleOffset.Address, FunctionName, Start, Size,
311                                FileName)) {
312       DILineInfo *LI = InlinedContext.getMutableFrame(
313           InlinedContext.getNumberOfFrames() - 1);
314       LI->FunctionName = FunctionName;
315       LI->StartAddress = Start;
316       if (LI->FileName == DILineInfo::BadString && !FileName.empty())
317         LI->FileName = FileName;
318     }
319   }
320 
321   return InlinedContext;
322 }
323 
324 DIGlobal SymbolizableObjectFile::symbolizeData(
325     object::SectionedAddress ModuleOffset) const {
326   DIGlobal Res;
327   std::string FileName;
328   getNameFromSymbolTable(ModuleOffset.Address, Res.Name, Res.Start, Res.Size,
329                          FileName);
330   Res.DeclFile = FileName;
331 
332   // Try and get a better filename:lineno pair from the debuginfo, if present.
333   DILineInfo DL = DebugInfoContext->getLineInfoForDataAddress(ModuleOffset);
334   if (DL.Line != 0) {
335     Res.DeclFile = DL.FileName;
336     Res.DeclLine = DL.Line;
337   }
338   return Res;
339 }
340 
341 std::vector<DILocal> SymbolizableObjectFile::symbolizeFrame(
342     object::SectionedAddress ModuleOffset) const {
343   if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
344     ModuleOffset.SectionIndex =
345         getModuleSectionIndexForAddress(ModuleOffset.Address);
346   return DebugInfoContext->getLocalsForAddress(ModuleOffset);
347 }
348 
349 /// Search for the first occurence of specified Address in ObjectFile.
350 uint64_t SymbolizableObjectFile::getModuleSectionIndexForAddress(
351     uint64_t Address) const {
352 
353   for (SectionRef Sec : Module->sections()) {
354     if (!Sec.isText() || Sec.isVirtual())
355       continue;
356 
357     if (Address >= Sec.getAddress() &&
358         Address < Sec.getAddress() + Sec.getSize())
359       return Sec.getIndex();
360   }
361 
362   return object::SectionedAddress::UndefSection;
363 }
364