Allow the GNU gold-style nonstandard SHT_GROUP section.

rui314 · rui314 · commit f9f69548a9cc · 2017-06-12T18:46:33.000Z
The ELF standard defines that the SHT_GROUP section as follows: - its sh_link has the symbol index, and - the symbol name is used to uniquify section groups. Object files created by GNU gold does not seem to comply with the standard. They have this additional rule: - if the symbol has no name and a STT_SECTION symbol, a section name is used instead of a symbol name. If we don't do anything for this, the linker fails with a mysterious error message if input files are generated by gas. It is unfortunate but I think we need to support it. Differential Revision: https://reviews.llvm.org/D34064 llvm-svn: 305218
diff --git a/lld/ELF/InputFiles.cpp b/lld/ELF/InputFiles.cpp
@@ -205,13 +205,27 @@ template <class ELFT>
 StringRef
 elf::ObjectFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> Sections,
                                             const Elf_Shdr &Sec) {
+  // Group signatures are stored as symbol names in object files.
+  // sh_info contains a symbol index, so we fetch a symbol and read its name.
   if (this->Symbols.empty())
     this->initSymtab(
         Sections,
         check(object::getSection<ELFT>(Sections, Sec.sh_link), toString(this)));
+
   const Elf_Sym *Sym = check(
       object::getSymbol<ELFT>(this->Symbols, Sec.sh_info), toString(this));
-  return check(Sym->getName(this->StringTable), toString(this));
+  StringRef Signature = check(Sym->getName(this->StringTable), toString(this));
+
+  // As a special case, if a symbol is a section symbol and has no name,
+  // we use a section name as a signature.
+  //
+  // Such SHT_GROUP sections are invalid from the perspective of the ELF
+  // standard, but GNU gold 1.14 (the neweset version as of July 2017) or
+  // older produce such sections as outputs for the -r option, so we need
+  // a bug-compatibility.
+  if (Signature.empty() && Sym->getType() == STT_SECTION)
+    return getSectionName(Sec);
+  return Signature;
 }
 
 template <class ELFT>
@@ -287,8 +301,7 @@ void elf::ObjectFile<ELFT>::initializeSections(
       check(this->getObj().sections(), toString(this));
   uint64_t Size = ObjSections.size();
   this->Sections.resize(Size);
-
-  StringRef SectionStringTable =
+  this->SectionStringTable =
       check(Obj.getSectionStringTable(ObjSections), toString(this));
 
   for (size_t I = 0, E = ObjSections.size(); I < E; I++) {
@@ -318,7 +331,7 @@ void elf::ObjectFile<ELFT>::initializeSections(
       // object files, we want to pass through basically everything.
       if (IsNew) {
         if (Config->Relocatable)
-          this->Sections[I] = createInputSection(Sec, SectionStringTable);
+          this->Sections[I] = createInputSection(Sec);
         continue;
       }
 
@@ -342,7 +355,7 @@ void elf::ObjectFile<ELFT>::initializeSections(
     case SHT_NULL:
       break;
     default:
-      this->Sections[I] = createInputSection(Sec, SectionStringTable);
+      this->Sections[I] = createInputSection(Sec);
     }
 
     // .ARM.exidx sections have a reverse dependency on the InputSection they
@@ -386,10 +399,8 @@ InputSectionBase *toRegularSection(MergeInputSection *Sec) {
 
 template <class ELFT>
 InputSectionBase *
-elf::ObjectFile<ELFT>::createInputSection(const Elf_Shdr &Sec,
-                                          StringRef SectionStringTable) {
-  StringRef Name = check(
-      this->getObj().getSectionName(&Sec, SectionStringTable), toString(this));
+elf::ObjectFile<ELFT>::createInputSection(const Elf_Shdr &Sec) {
+  StringRef Name = getSectionName(Sec);
 
   switch (Sec.sh_type) {
   case SHT_ARM_ATTRIBUTES:
@@ -521,6 +532,12 @@ elf::ObjectFile<ELFT>::createInputSection(const Elf_Shdr &Sec,
   return make<InputSection>(this, &Sec, Name);
 }
 
+template <class ELFT>
+StringRef elf::ObjectFile<ELFT>::getSectionName(const Elf_Shdr &Sec) {
+  return check(this->getObj().getSectionName(&Sec, SectionStringTable),
+               toString(this));
+}
+
 template <class ELFT> void elf::ObjectFile<ELFT>::initializeSymbols() {
   SymbolBodies.reserve(this->Symbols.size());
   for (const Elf_Sym &Sym : this->Symbols)
diff --git a/lld/ELF/InputFiles.h b/lld/ELF/InputFiles.h
@@ -194,15 +194,18 @@ template <class ELFT> class ObjectFile : public ELFFileBase<ELFT> {
   void initializeSymbols();
   void initializeDwarfLine();
   InputSectionBase *getRelocTarget(const Elf_Shdr &Sec);
-  InputSectionBase *createInputSection(const Elf_Shdr &Sec,
-                                       StringRef SectionStringTable);
+  InputSectionBase *createInputSection(const Elf_Shdr &Sec);
+  StringRef getSectionName(const Elf_Shdr &Sec);
 
   bool shouldMerge(const Elf_Shdr &Sec);
   SymbolBody *createSymbolBody(const Elf_Sym *Sym);
 
   // List of all symbols referenced or defined by this file.
   std::vector<SymbolBody *> SymbolBodies;
 
+  // .shstrtab contents.
+  StringRef SectionStringTable;
+
   // Debugging information to retrieve source file and line for error
   // reporting. Linker may find reasonable number of errors in a
   // single object file, so we cache debugging information in order to
diff --git a/lld/test/ELF/Inputs/sht-group-gold-r.elf b/lld/test/ELF/Inputs/sht-group-gold-r.elf
diff --git a/lld/test/ELF/Inputs/sht-group-gold-r.s b/lld/test/ELF/Inputs/sht-group-gold-r.s
@@ -0,0 +1,14 @@
+# sht-group-gold-r.elf is produced by
+#
+#   llvm-mc -filetype=obj -triple=x86_64-pc-linux sht-group-gold-r.s -o sht-group-gold-r.o
+#   ld.gold -o sht-group-gold-r.elf -r sht-group-gold-r.o
+
+.global foo, bar
+
+.section .text.foo,"aG",@progbits,group_foo,comdat
+foo:
+  nop
+
+.section .text.bar,"aG",@progbits,group_bar,comdat
+bar:
+  nop
diff --git a/lld/test/ELF/sht-group-gold-r.test b/lld/test/ELF/sht-group-gold-r.test
@@ -0,0 +1,17 @@
+# GNU gold 1.14 (the newest version as of July 2017) seems to create
+# non-standard-compliant SHT_GROUP sections when the -r option is given.
+#
+# Such SHT_GROUP sections use section names as their signatures
+# instead of symbols pointed by sh_link field. Since it is prevalent,
+# we accept such nonstandard sections.
+
+# RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t.o
+# RUN: ld.lld %p/Inputs/sht-group-gold-r.elf %t.o -o %t.exe
+# RUN: llvm-objdump -t %t.exe | FileCheck %s
+
+# CHECK: .text 00000000 bar
+# CHECK: .text 00000000 foo
+
+.globl _start
+_start:
+  ret
