diff --git a/llvm/test/tools/llvm-readobj/ELF/all.test b/llvm/test/tools/llvm-readobj/ELF/all.test --- a/llvm/test/tools/llvm-readobj/ELF/all.test +++ b/llvm/test/tools/llvm-readobj/ELF/all.test @@ -112,4 +112,5 @@ Sections: - Section: .note.gnu.build-id Symbols: [] -DynamicSymbols: [] +DynamicSymbols: + - Name: foo diff --git a/llvm/test/tools/llvm-readobj/ELF/dyn-symbols-size-from-hash-table.test b/llvm/test/tools/llvm-readobj/ELF/dyn-symbols-size-from-hash-table.test new file mode 100644 --- /dev/null +++ b/llvm/test/tools/llvm-readobj/ELF/dyn-symbols-size-from-hash-table.test @@ -0,0 +1,314 @@ +## This test shows how llvm-readobj uses the hash table section to derive the +## size of a dynamic symbol table. This allows dumping of the dynamic symbol +## table in the event of an object without section headers. + +## Case 1a) Table size is derived from hash table, with DT_SYMTAB before DT_HASH. +# RUN: yaml2obj --docnum=1 %s -o %t1a-64 -DBITS=64 \ +# RUN: -DTAG1=DT_SYMTAB -DTAG2=DT_HASH -DVAL1=0x400 -DVAL2=0x600 +# RUN: llvm-strip --strip-sections %t1a-64 +# RUN: llvm-readobj --dyn-symbols %t1a-64 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=LLVM1,STRIP --implicit-check-not=warning +# RUN: llvm-readelf --dyn-symbols %t1a-64 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=GNU1,GNU1-STRIP --implicit-check-not=warning +# RUN: yaml2obj --docnum=1 %s -o %t1a-32 -DBITS=32 \ +# RUN: -DTAG1=DT_SYMTAB -DTAG2=DT_HASH -DVAL1=0x400 -DVAL2=0x600 +# RUN: llvm-strip --strip-sections %t1a-32 +# RUN: llvm-readobj --dyn-symbols %t1a-32 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=LLVM1,STRIP --implicit-check-not=warning +# RUN: llvm-readelf --dyn-symbols %t1a-32 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=GNU1,GNU1-STRIP --implicit-check-not=warning + +## 1b) Table size is derived from hash table, with DT_HASH before DT_SYMTAB. +## We don't bother testing 32 and 64-bit here. The above cases show that reading +## the nchain value is correct for all formats, and other tests show the basic +## printing behaviour. +# RUN: yaml2obj --docnum=1 %s -o %t1b-64 -DBITS=64 \ +# RUN: -DTAG1=DT_HASH -DTAG2=DT_SYMTAB -DVAL1=0x600 -DVAL2=0x400 +# RUN: llvm-strip --strip-sections %t1b-64 +# RUN: llvm-readobj --dyn-symbols %t1b-64 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=LLVM1,STRIP --implicit-check-not=warning +# RUN: llvm-readelf --dyn-symbols %t1b-64 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=GNU1,GNU1-STRIP --implicit-check-not=warning + +# LLVM1: DynamicSymbols [ +# LLVM1-NEXT: Symbol { +# LLVM1-NEXT: Name: (0) +# LLVM1-NEXT: Value: 0x0 +# LLVM1-NEXT: Size: 0 +# LLVM1-NEXT: Binding: Local (0x0) +# LLVM1-NEXT: Type: None (0x0) +# LLVM1-NEXT: Other: 0 +# LLVM1-NEXT: Section: Undefined (0x0) +# LLVM1-NEXT: } +# LLVM1-NEXT: Symbol { +# LLVM1-NEXT: Name: foo (5) +# LLVM1-NEXT: Value: 0x100 +# LLVM1-NEXT: Size: 0 +# LLVM1-NEXT: Binding: Local (0x0) +# LLVM1-NEXT: Type: Function (0x2) +# LLVM1-NEXT: Other: 0 +# STRIP-NEXT: Section: (0x1) +# NOSTRIP-NEXT: Section: .text (0x1) +# LLVM1-NEXT: } +# LLVM1-NEXT: Symbol { +# LLVM1-NEXT: Name: bar (1) +# LLVM1-NEXT: Value: 0x200 +# LLVM1-NEXT: Size: 0 +# LLVM1-NEXT: Binding: Local (0x0) +# LLVM1-NEXT: Type: Object (0x1) +# LLVM1-NEXT: Other: 0 +# STRIP-NEXT: Section: (0x2) +# NOSTRIP-NEXT: Section: .data (0x2) +# LLVM1-NEXT: } +# LLVM1-NEXT: ] + +# GNU1-STRIP: Symbol table for image contains 3 entries: +# GNU1-NOSTRIP: Symbol table '.dynsym' contains 3 entries: +# GNU1-NEXT: Num: Value Size Type Bind Vis Ndx Name +# GNU1-NEXT: 0: {{0*}}00000000 0 NOTYPE LOCAL DEFAULT UND +# GNU1-NEXT: 1: {{0*}}00000100 0 FUNC LOCAL DEFAULT 1 foo +# GNU1-NEXT: 2: {{0*}}00000200 0 OBJECT LOCAL DEFAULT 2 bar +# GNU1-EMPTY: + +--- !ELF +FileHeader: + Class: ELFCLASS[[BITS]] + Data: ELFDATA2LSB + Type: ET_DYN + Machine: EM_X86_64 +Sections: + - Name: .text + Type: SHT_PROGBITS + - Name: .data + Type: SHT_PROGBITS + - Name: .dynsym + Type: SHT_DYNSYM + Flags: [ SHF_ALLOC ] + Address: 0x400 + AddressAlign: 0x400 + - Name: .hash + Type: SHT_HASH + Flags: [ SHF_ALLOC ] + Address: 0x600 + AddressAlign: 0x200 + Bucket: [ 1 ] + Chain: [ 1, 2, 3 ] + - Name: .dynstr + Type: SHT_STRTAB + Flags: [ SHF_ALLOC ] + Address: 0x800 + AddressAlign: 0x200 + - Name: .dynamic + Type: SHT_DYNAMIC + Flags: [ SHF_ALLOC ] + Address: 0xA00 + AddressAlign: 0x200 + Entries: + - Tag: DT_STRTAB + Value: 0x800 + - Tag: DT_STRSZ + Value: 9 + - Tag: [[TAG1]] + Value: [[VAL1]] + - Tag: [[TAG2]] + Value: [[VAL2]] + - Tag: DT_NULL + Value: 0 +DynamicSymbols: + - Name: foo + Type: STT_FUNC + Section: .text + Value: 0x100 + - Name: bar + Type: STT_OBJECT + Section: .data + Value: 0x200 +ProgramHeaders: + - Type: PT_LOAD + VAddr: 0 + Sections: + - Section: .text + - Section: .data + - Type: PT_LOAD + VAddr: 0x400 + Sections: + - Section: .dynsym + - Section: .hash + - Section: .dynstr + - Section: .dynamic + - Type: PT_DYNAMIC + VAddr: 0xA00 + Sections: + - Section: .dynamic + +## Case 2: Table size from DT_HASH does not match size from section header. +# RUN: yaml2obj --docnum=2 %s -o %t2-smaller -DCHAIN="[1, 2]" +# RUN: llvm-readobj --dyn-symbols %t2-smaller 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=LLVM2,WARN \ +# RUN: --implicit-check-not=warning -DNCHAIN=2 +# RUN: llvm-readelf --dyn-symbols %t2-smaller 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=GNU2,WARN \ +# RUN: --implicit-check-not=warning -DNCHAIN=2 + +# RUN: yaml2obj --docnum=2 %s -o %t2-larger -DCHAIN="[1, 2, 3, 4]" +# RUN: llvm-readobj --dyn-symbols %t2-larger 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=LLVM2,LLVM2-MORE,LLVM2-ALL,WARN \ +# RUN: --implicit-check-not=warning -DNCHAIN=4 +# RUN: llvm-readelf --dyn-symbols %t2-larger 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=GNU2,GNU2-MORE,GNU2-ALL,WARN \ +# RUN: --implicit-check-not=warning -DNCHAIN=4 + +## Show there's no warning if the sizes match +# RUN: yaml2obj --docnum=2 %s -o %t2-same -DCHAIN="[1, 2, 3]" +# RUN: llvm-readobj --dyn-symbols %t2-same 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=LLVM2,LLVM2-MORE --implicit-check-not=warning +# RUN: llvm-readelf --dyn-symbols %t2-same 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=GNU2,GNU2-MORE \ +# RUN: --implicit-check-not=warning -DNCHAIN=3 + +# WARN: warning: '{{.*}}2-{{.*}}': hash table nchain ([[NCHAIN]]) differs from symbol count derived from SHT_DYNSYM section header (3) + +# LLVM2: DynamicSymbols [ +# LLVM2-NEXT: Symbol { +# LLVM2-NEXT: Name: (0) +# LLVM2-NEXT: Value: 0x0 +# LLVM2-NEXT: Size: 0 +# LLVM2-NEXT: Binding: Local (0x0) +# LLVM2-NEXT: Type: None (0x0) +# LLVM2-NEXT: Other: 0 +# LLVM2-NEXT: Section: Undefined (0x0) +# LLVM2-NEXT: } +# LLVM2-NEXT: Symbol { +# LLVM2-NEXT: Name: foo (9) +# LLVM2-NEXT: Value: 0x100 +# LLVM2-NEXT: Size: 0 +# LLVM2-NEXT: Binding: Local (0x0) +# LLVM2-NEXT: Type: Function (0x2) +# LLVM2-NEXT: Other: 0 +# LLVM2-NEXT: Section: .text (0x1) +# LLVM2-NEXT: } +# LLVM2-MORE-NEXT: Symbol { +# LLVM2-MORE-NEXT: Name: bar (5) +# LLVM2-MORE-NEXT: Value: 0x200 +# LLVM2-MORE-NEXT: Size: 0 +# LLVM2-MORE-NEXT: Binding: Local (0x0) +# LLVM2-MORE-NEXT: Type: Object (0x1) +# LLVM2-MORE-NEXT: Other: 0 +# LLVM2-MORE-NEXT: Section: .data (0x2) +# LLVM2-MORE-NEXT: } +# LLVM2-ALL-NEXT: Symbol { +# LLVM2-ALL-NEXT: Name: baz (1) +# LLVM2-ALL-NEXT: Value: 0x300 +# LLVM2-ALL-NEXT: Size: 0 +# LLVM2-ALL-NEXT: Binding: Local (0x0) +# LLVM2-ALL-NEXT: Type: Object (0x1) +# LLVM2-ALL-NEXT: Other: 0 +# LLVM2-ALL-NEXT: Section: .data (0x2) +# LLVM2-ALL-NEXT: } +# LLVM2-NEXT: ] + +# GNU2: Symbol table '.dynsym' contains [[NCHAIN]] entries: +# GNU2-NEXT: Num: Value Size Type Bind Vis Ndx Name +# GNU2-NEXT: 0: {{0*}}00000000 0 NOTYPE LOCAL DEFAULT UND +# GNU2-NEXT: 1: {{0*}}00000100 0 FUNC LOCAL DEFAULT 1 foo +# GNU2-MORE-NEXT: 2: {{0*}}00000200 0 OBJECT LOCAL DEFAULT 2 bar +# GNU2-ALL-NEXT: 3: {{0*}}00000300 0 OBJECT LOCAL DEFAULT 2 baz +# GNU2-EMPTY: + +## In this YAML, we define 4 dynamic symbols (including the null symbol), but +## constrain the .dynsym section header to say there are only 3. This means that +## when a size of 4 is derived from the hash table, we still have a valid symbol +## to dump. +--- !ELF +FileHeader: + Class: ELFCLASS64 + Data: ELFDATA2LSB + Type: ET_DYN + Machine: EM_X86_64 +Sections: + - Name: .text + Type: SHT_PROGBITS + - Name: .data + Type: SHT_PROGBITS + - Name: .dynsym + Type: SHT_DYNSYM + Flags: [ SHF_ALLOC ] + ShSize: 0x48 + Address: 0x400 + AddressAlign: 0x400 + - Name: .hash + Type: SHT_HASH + Flags: [ SHF_ALLOC ] + Address: 0x600 + AddressAlign: 0x200 + Bucket: [ 1 ] + Chain: [[CHAIN]] + - Name: .dynstr + Type: SHT_STRTAB + Flags: [ SHF_ALLOC ] + Address: 0x800 + AddressAlign: 0x200 + - Name: .dynamic + Type: SHT_DYNAMIC + Flags: [ SHF_ALLOC ] + Address: 0xA00 + AddressAlign: 0x200 + Entries: + - Tag: DT_SYMTAB + Value: 0x400 + - Tag: DT_HASH + Value: 0x600 + - Tag: DT_STRTAB + Value: 0x800 + - Tag: DT_STRSZ + Value: 13 + - Tag: DT_NULL + Value: 0 +DynamicSymbols: + - Name: foo + Type: STT_FUNC + Section: .text + Value: 0x100 + - Name: bar + Type: STT_OBJECT + Section: .data + Value: 0x200 + - Name: baz + Type: STT_OBJECT + Section: .data + Value: 0x300 +ProgramHeaders: + - Type: PT_LOAD + VAddr: 0 + Sections: + - Section: .text + - Section: .data + - Type: PT_LOAD + VAddr: 0x400 + Sections: + - Section: .dynsym + - Section: .hash + - Section: .dynstr + - Section: .dynamic + - Type: PT_DYNAMIC + VAddr: 0xA00 + Sections: + - Section: .dynamic + +## Case 3: DT_HASH is missing. +## Show that no warning occurs if there are section headers. +# RUN: yaml2obj --docnum=1 %s -o %t3 -DTAG1=DT_SYMTAB -DVAL1=0x400 -DTAG2=DT_NULL -DVAL2=0 -DBITS=64 +# RUN: llvm-readobj --dyn-symbols %t3 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=LLVM1,NOSTRIP --implicit-check-not=warning +# RUN: llvm-readelf --dyn-symbols %t3 2>&1 | \ +# RUN: FileCheck %s --check-prefixes=GNU1,GNU1-NOSTRIP --implicit-check-not=warning + +## Show that size is treated as zero, if no section headers are present. +# RUN: llvm-strip --strip-sections %t3 +# RUN: llvm-readobj --dyn-symbols %t3 2>&1 | \ +# RUN: FileCheck %s --check-prefix=LLVM3 --implicit-check-not=warning +# RUN: llvm-readelf --dyn-symbols %t3 2>&1 | \ +# RUN: FileCheck %s --implicit-check-not={{.}} --allow-empty + +# LLVM3: DynamicSymbols [ +# LLVM3: ] diff --git a/llvm/tools/llvm-readobj/ELFDumper.cpp b/llvm/tools/llvm-readobj/ELFDumper.cpp --- a/llvm/tools/llvm-readobj/ELFDumper.cpp +++ b/llvm/tools/llvm-readobj/ELFDumper.cpp @@ -283,7 +283,7 @@ DynRegionInfo DynRelaRegion; DynRegionInfo DynRelrRegion; DynRegionInfo DynPLTRelRegion; - DynRegionInfo DynSymRegion; + Optional DynSymRegion; DynRegionInfo DynamicTable; StringRef DynamicStringTable; std::string SOName = ""; @@ -322,7 +322,9 @@ } Elf_Sym_Range dynamic_symbols() const { - return DynSymRegion.getAsArrayRef(); + if (!DynSymRegion) + return Elf_Sym_Range(); + return DynSymRegion->getAsArrayRef(); } Elf_Rel_Range dyn_rels() const; @@ -667,8 +669,8 @@ StrTable = DynamicStringTable; Syms = dynamic_symbols(); SymtabName = DynSymtabName; - if (DynSymRegion.Addr) - Entries = DynSymRegion.Size / DynSymRegion.EntSize; + if (DynSymRegion) + Entries = DynSymRegion->Size / DynSymRegion->EntSize; } else { if (!DotSymtabSec) return; @@ -993,7 +995,7 @@ template Error ELFDumper::LoadVersionMap() const { // If there is no dynamic symtab or version table, there is nothing to do. - if (!DynSymRegion.Addr || !SymbolVersionSection) + if (!DynSymRegion || !SymbolVersionSection) return Error::success(); // Has the VersionMap already been loaded? @@ -1043,10 +1045,11 @@ return ""; } + assert(DynSymRegion && "DynSymRegion has not been initialised"); // Determine the position in the symbol table of this entry. size_t EntryIndex = (reinterpret_cast(Sym) - - reinterpret_cast(DynSymRegion.Addr)) / - sizeof(Elf_Sym); + reinterpret_cast(DynSymRegion->Addr)) / + sizeof(Elf_Sym); // Get the corresponding version index entry. const Elf_Versym *Versym = unwrapOrError( @@ -1980,8 +1983,12 @@ ScopedPrinter &Writer) : ObjDumper(Writer), ObjF(ObjF), DynRelRegion(ObjF->getFileName()), DynRelaRegion(ObjF->getFileName()), DynRelrRegion(ObjF->getFileName()), - DynPLTRelRegion(ObjF->getFileName()), DynSymRegion(ObjF->getFileName()), - DynamicTable(ObjF->getFileName()) { + DynPLTRelRegion(ObjF->getFileName()), DynamicTable(ObjF->getFileName()) { + if (opts::Output == opts::GNU) + ELFDumperStyle.reset(new GNUStyle(Writer, this)); + else + ELFDumperStyle.reset(new LLVMStyle(Writer, this)); + const ELFFile *Obj = ObjF->getELFFile(); typename ELFT::ShdrRange Sections = unwrapOrError(ObjF->getFileName(), Obj->sections()); @@ -1992,9 +1999,9 @@ DotSymtabSec = &Sec; break; case ELF::SHT_DYNSYM: - if (!DynSymRegion.Size) { + if (!DynSymRegion) { DynSymRegion = createDRIFrom(&Sec); - DynSymRegion.Context = + DynSymRegion->Context = ("section with index " + Twine(&Sec - &Sections.front())).str(); // This is only used (if Elf_Shdr present)for naming section in GNU // style @@ -2034,11 +2041,6 @@ } loadDynamicTable(Obj); - - if (opts::Output == opts::GNU) - ELFDumperStyle.reset(new GNUStyle(Writer, this)); - else - ELFDumperStyle.reset(new LLVMStyle(Writer, this)); } template @@ -2059,6 +2061,7 @@ uint64_t SONameOffset = 0; const char *StringTableBegin = nullptr; uint64_t StringTableSize = 0; + Optional DynSymFromTable; for (const Elf_Dyn &Dyn : dynamic_table()) { switch (Dyn.d_tag) { case ELF::DT_HASH: @@ -2077,26 +2080,13 @@ StringTableSize = Dyn.getVal(); break; case ELF::DT_SYMTAB: { - // Often we find the information about the dynamic symbol table - // location in the SHT_DYNSYM section header. However, the value in - // DT_SYMTAB has priority, because it is used by dynamic loaders to - // locate .dynsym at runtime. The location we find in the section header - // and the location we find here should match. If we can't map the - // DT_SYMTAB value to an address (e.g. when there are no program headers), we - // ignore its value. + // If we can't map the DT_SYMTAB value to an address (e.g. when there are + // no program headers), we ignore its value. if (const uint8_t *VA = toMappedAddr(Dyn.getTag(), Dyn.getPtr())) { - // EntSize is non-zero if the dynamic symbol table has been found via a - // section header. - if (DynSymRegion.EntSize && VA != DynSymRegion.Addr) - reportWarning( - createError( - "SHT_DYNSYM section header and DT_SYMTAB disagree about " - "the location of the dynamic symbol table"), - ObjF->getFileName()); - - DynSymRegion.Addr = VA; - DynSymRegion.EntSize = sizeof(Elf_Sym); - DynSymRegion.EntSizePrintName = ""; + DynSymFromTable.emplace(ObjF->getFileName()); + DynSymFromTable->Addr = VA; + DynSymFromTable->EntSize = sizeof(Elf_Sym); + DynSymFromTable->EntSizePrintName = ""; } break; } @@ -2176,6 +2166,47 @@ if (StringTableBegin) DynamicStringTable = StringRef(StringTableBegin, StringTableSize); SOName = getDynamicString(SONameOffset); + + if (DynSymRegion) { + // Often we find the information about the dynamic symbol table + // location in the SHT_DYNSYM section header. However, the value in + // DT_SYMTAB has priority, because it is used by dynamic loaders to + // locate .dynsym at runtime. The location we find in the section header + // and the location we find here should match. + if (DynSymFromTable && DynSymFromTable->Addr != DynSymRegion->Addr) + ELFDumperStyle->reportUniqueWarning( + createError("SHT_DYNSYM section header and DT_SYMTAB disagree about " + "the location of the dynamic symbol table")); + + // According to the ELF gABI: "The number of symbol table entries should + // equal nchain". Check to see if the hash table nchain value conflicts with + // the number of symbols in the dynamic symbol table according to the + // section header. + if (HashTable && + HashTable->nchain != DynSymRegion->Size / DynSymRegion->EntSize) + ELFDumperStyle->reportUniqueWarning(createError( + "hash table nchain (" + Twine(HashTable->nchain) + + ") differs from symbol count derived from SHT_DYNSYM section " + "header (" + + Twine(DynSymRegion->Size / DynSymRegion->EntSize) + ")")); + } + + // Delay the creation of the actual dynamic symbol table until now, so that + // checks can always be made against the section header-based properties, + // without worrying about tag order. + if (DynSymFromTable) { + if (!DynSymRegion) { + DynSymRegion = DynSymFromTable; + } else { + DynSymRegion->Addr = DynSymFromTable->Addr; + DynSymRegion->EntSize = DynSymFromTable->EntSize; + DynSymRegion->EntSizePrintName = DynSymFromTable->EntSizePrintName; + } + } + + // Derive the dynamic symbol table size from the hash table, if present. + if (HashTable && DynSymRegion) + DynSymRegion->Size = HashTable->nchain * DynSymRegion->EntSize; } template @@ -2591,7 +2622,7 @@ ArrayRef Buckets = GnuHashTable->buckets(); W.printList("Buckets", Buckets); - if (!DynSymRegion.Addr) { + if (!DynSymRegion) { reportWarning(createError("unable to dump 'Values' for the SHT_GNU_HASH " "section: no dynamic symbol table found"), ObjF->getFileName()); @@ -3689,10 +3720,10 @@ size_t Entries, bool NonVisibilityBitsUsed) { if (!Name.empty()) - OS << "\nSymbol table '" << Name << "' contains " << Entries - << " entries:\n"; + OS << "\nSymbol table '" << Name << "'"; else - OS << "\n Symbol table for image:\n"; + OS << "\nSymbol table for image"; + OS << " contains " << Entries << " entries:\n"; if (ELFT::Is64Bits) OS << " Num: Value Size Type Bind Vis"; @@ -5881,7 +5912,12 @@ Expected SectionName = this->dumper()->getSymbolSectionName(Symbol, *SectionIndex); if (!SectionName) { - this->reportUniqueWarning(SectionName.takeError()); + // Don't report an invalid section name if the section headers are missing. + // In such situations, all sections will be "invalid". + if (!this->dumper()->getElfObject()->sections().empty()) + this->reportUniqueWarning(SectionName.takeError()); + else + consumeError(SectionName.takeError()); W.printHex("Section", "", *SectionIndex); } else { W.printHex("Section", *SectionName, *SectionIndex);