[lld-macho] Make relative method lists work on x86-64 (#103905)

Local data is referenced in Objective-C metadata via section + offset
relocations on x86-64 rather than via symbols. Without this change, we
would crash on incorrect casts of the referents to `Defined`.

A basic test based on the existing `objc-relative-method-lists-simple.s`
adopted to x86-64 is added.

Author: BertalanD

lld/MachO/ObjC.cpp

@@ -186,28 +186,6 @@ ObjcCategoryChecker::ObjcCategoryChecker()
       roClassLayout(target->wordSize), listHeaderLayout(target->wordSize),
       methodLayout(target->wordSize) {}
 
-// \p r must point to an offset within a CStringInputSection or a
-// ConcatInputSection
-static StringRef getReferentString(const Reloc &r) {
-  if (auto *isec = r.referent.dyn_cast<InputSection *>())
-    return cast<CStringInputSection>(isec)->getStringRefAtOffset(r.addend);
-
-  auto *sym = cast<Defined>(r.referent.get<Symbol *>());
-  auto *symIsec = sym->isec();
-  auto symOffset = sym->value + r.addend;
-
-  if (auto *s = dyn_cast_or_null<CStringInputSection>(symIsec))
-    return s->getStringRefAtOffset(symOffset);
-
-  if (isa<ConcatInputSection>(symIsec)) {
-    auto strData = symIsec->data.slice(symOffset);
-    const char *pszData = reinterpret_cast<const char *>(strData.data());
-    return StringRef(pszData, strnlen(pszData, strData.size()));
-  }
-
-  llvm_unreachable("unknown reference section in getReferentString");
-}
-
 void ObjcCategoryChecker::parseMethods(const ConcatInputSection *methodsIsec,
                                        const Symbol *methodContainerSym,
                                        const ConcatInputSection *containerIsec,
@@ -219,7 +197,7 @@ void ObjcCategoryChecker::parseMethods(const ConcatInputSection *methodsIsec,
         methodLayout.nameOffset)
       continue;
 
-    CachedHashStringRef methodName(getReferentString(r));
+    CachedHashStringRef methodName(r.getReferentString());
     // +load methods are special: all implementations are called by the runtime
     // even if they are part of the same class. Thus there is no need to check
     // for duplicates.
@@ -251,14 +229,14 @@ void ObjcCategoryChecker::parseMethods(const ConcatInputSection *methodsIsec,
                          ->getReferentInputSection();
       nameReloc = roIsec->getRelocAt(roClassLayout.nameOffset);
     }
-    StringRef containerName = getReferentString(*nameReloc);
+    StringRef containerName = nameReloc->getReferentString();
     StringRef methPrefix = mKind == MK_Instance ? "-" : "+";
 
     // We should only ever encounter collisions when parsing category methods
     // (since the Class struct is parsed before any of its categories).
     assert(mcKind == MCK_Category);
     StringRef newCatName =
-        getReferentString(*containerIsec->getRelocAt(catLayout.nameOffset));
+        containerIsec->getRelocAt(catLayout.nameOffset)->getReferentString();
 
     auto formatObjAndSrcFileName = [](const InputSection *section) {
       lld::macho::InputFile *inputFile = section->getFile();
@@ -809,7 +787,7 @@ void ObjcCategoryMerger::parseCatInfoToExtInfo(const InfoInputCategory &catInfo,
   assert(extInfo.objFileForMergeData &&
          "Expected to already have valid objextInfo.objFileForMergeData");
 
-  StringRef catName = getReferentString(*catNameReloc);
+  StringRef catName = catNameReloc->getReferentString();
   extInfo.mergedContainerName += catName.str();
 
   // Parse base class

lld/MachO/Relocations.cpp

@@ -31,6 +31,31 @@ InputSection *Reloc::getReferentInputSection() const {
   }
 }
 
+StringRef Reloc::getReferentString() const {
+  if (auto *isec = referent.dyn_cast<InputSection *>()) {
+    const auto *cisec = dyn_cast<CStringInputSection>(isec);
+    assert(cisec && "referent must be a CStringInputSection");
+    return cisec->getStringRefAtOffset(addend);
+  }
+
+  auto *sym = dyn_cast<Defined>(referent.get<Symbol *>());
+  assert(sym && "referent must be a Defined symbol");
+
+  auto *symIsec = sym->isec();
+  auto symOffset = sym->value + addend;
+
+  if (auto *s = dyn_cast_or_null<CStringInputSection>(symIsec))
+    return s->getStringRefAtOffset(symOffset);
+
+  if (isa<ConcatInputSection>(symIsec)) {
+    auto strData = symIsec->data.slice(symOffset);
+    const char *pszData = reinterpret_cast<const char *>(strData.data());
+    return StringRef(pszData, strnlen(pszData, strData.size()));
+  }
+
+  llvm_unreachable("unknown reference section in getReferentString");
+}
+
 bool macho::validateSymbolRelocation(const Symbol *sym,
                                      const InputSection *isec, const Reloc &r) {
   const RelocAttrs &relocAttrs = target->getRelocAttrs(r.type);

lld/MachO/Relocations.h

@@ -69,6 +69,10 @@ struct Reloc {
         addend(addend), referent(referent) {}
 
   InputSection *getReferentInputSection() const;
+
+  // Must point to an offset within a CStringInputSection or a
+  // ConcatInputSection.
+  llvm::StringRef getReferentString() const;
 };
 
 bool validateSymbolRelocation(const Symbol *, const InputSection *,

lld/MachO/SyntheticSections.cpp

@@ -2010,11 +2010,8 @@ void ObjCMethListSection::setUp() {
     while (methodNameOff < isec->data.size()) {
       const Reloc *reloc = isec->getRelocAt(methodNameOff);
       assert(reloc && "Relocation expected at method list name slot");
-      auto *def = dyn_cast_or_null<Defined>(reloc->referent.get<Symbol *>());
-      assert(def && "Expected valid Defined at method list name slot");
-      auto *cisec = cast<CStringInputSection>(def->isec());
-      assert(cisec && "Expected method name to be in a CStringInputSection");
-      auto methname = cisec->getStringRefAtOffset(def->value);
+
+      StringRef methname = reloc->getReferentString();
       if (!ObjCSelRefsHelper::getSelRef(methname))
         ObjCSelRefsHelper::makeSelRef(methname);
 
@@ -2114,19 +2111,23 @@ void ObjCMethListSection::writeRelativeOffsetForIsec(
     uint32_t &outSecOff, bool useSelRef) const {
   const Reloc *reloc = isec->getRelocAt(inSecOff);
   assert(reloc && "Relocation expected at __objc_methlist Offset");
-  auto *def = dyn_cast_or_null<Defined>(reloc->referent.get<Symbol *>());
-  assert(def && "Expected all syms in __objc_methlist to be defined");
-  uint32_t symVA = def->getVA();
 
+  uint32_t symVA = 0;
   if (useSelRef) {
-    auto *cisec = cast<CStringInputSection>(def->isec());
-    auto methname = cisec->getStringRefAtOffset(def->value);
+    StringRef methname = reloc->getReferentString();
     ConcatInputSection *selRef = ObjCSelRefsHelper::getSelRef(methname);
     assert(selRef && "Expected all selector names to already be already be "
                      "present in __objc_selrefs");
     symVA = selRef->getVA();
     assert(selRef->data.size() == sizeof(target->wordSize) &&
            "Expected one selref per ConcatInputSection");
+  } else if (reloc->referent.is<Symbol *>()) {
+    auto *def = dyn_cast_or_null<Defined>(reloc->referent.get<Symbol *>());
+    assert(def && "Expected all syms in __objc_methlist to be defined");
+    symVA = def->getVA();
+  } else {
+    auto *isec = reloc->referent.get<InputSection *>();
+    symVA = isec->getVA(reloc->addend);
   }
 
   uint32_t currentVA = isec->getVA() + outSecOff;