[CHERIoT]: Add cheriot_sealed attribute

Author: xdoardo

clang/include/clang/AST/Type.h

@@ -1140,6 +1140,10 @@ class QualType {
   /// Return true if this is a trivially relocatable type.
   bool isTriviallyRelocatableType(const ASTContext &Context) const;
 
+  /// Return true if this QualType has the attribute signaling that it
+  /// references a sealed type in CHERIoT.
+  bool hasCHERIoTSealedAttr() const;
+
   /// Returns true if it is a class and it might be dynamic.
   bool mayBeDynamicClass() const;
 

clang/include/clang/Basic/Attr.td

@@ -2906,6 +2906,14 @@ def CHERIOTSharedObject : DeclOrTypeAttr {
   let Subjects = SubjectList<[GlobalVar], ErrorDiag>;
 }
 
+def CHERIoTSealedType : TypeAttr {
+  let Spellings = [C23<"cheriot", "sealed">, CXX11<"cheriot", "sealed">,
+                   GNU<"cheriot_sealed">];
+  let Documentation = [CHERIoTSealedTypeDocs];
+  let Args = [StringArgument<"CompartmentName">,
+              StringArgument<"SealingTypeName">];
+}
+
 def CHERINoSubobjectBounds : DeclOrTypeAttr {
   let Spellings = [GNU<"cheri_no_subobject_bounds">,
                    CXX11<"cheri","no_subobject_bounds">,

clang/include/clang/Basic/AttrDocs.td

@@ -1540,6 +1540,19 @@ example, `"RWcm"` and `"mcWR"` are both valid and entail the same permissions.
   }];
 }
 
+def CHERIoTSealedTypeDocs : Documentation {
+  let Category = DocCatVariable;
+  let Heading = "cheriot_sealed";
+  let Content = [{
+Used to set a type as sealed: this means that values of the attributed type will be understood as sealed values. 
+The only operation valid on a sealed value is to take its address.
+
+**Usage**: ``__attribute__((cheriot_sealed("<compartment_name>",
+"<sealing_key_name>")))`` where `"<compartment_name>"` is the name of the
+compartment that is allowed to unseal the value and `<sealing_key_name>` is the
+name of the chosen sealing key exposed from the compartment.
+  }];
+}
 def ObjCMethodFamilyDocs : Documentation {
   let Category = DocCatFunction;
   let Content = [{

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -150,6 +150,11 @@ def warn_cheriot_global_cap_import_non_volatile : Warning<
   "global variable definition '%0' has attribute %1 but is not qualified as `volatile`">;
 def err_cheriot_global_cap_import_initialized: Error<
   "global variable definition '%0' with attribute %1 cannot have an initializer">;
+def err_cheriot_non_addr_of_expr_on_sealed
+    : Error<
+          "the only valid operation on a sealed value is to take its address">;
+def err_cheriot_invalid_sealed_declaration
+    : Error<"cannot declare a sealed variable as %0">;
 
 // C99 variable-length arrays
 def ext_vla : Extension<"variable length arrays are a C99 feature">,

clang/lib/AST/Type.cpp

@@ -120,6 +120,21 @@ const IdentifierInfo* QualType::getBaseTypeIdentifier() const {
   return nullptr;
 }
 
+bool QualType::hasCHERIoTSealedAttr() const {
+  const Type &Self = **this;
+  bool Return = false;
+
+  if (const auto *TT = Self.getAs<TagType>()) {
+    Return |= TT->getDecl()->hasAttr<CHERIoTSealedTypeAttr>();
+  }
+
+  if (const auto *TT = Self.getAs<TypedefType>()) {
+    Return |= TT->getDecl()->hasAttr<CHERIoTSealedTypeAttr>();
+  }
+
+  return Return;
+}
+
 bool QualType::mayBeDynamicClass() const {
   const auto *ClassDecl = getTypePtr()->getPointeeCXXRecordDecl();
   return ClassDecl && ClassDecl->mayBeDynamicClass();

clang/lib/CodeGen/CodeGenModule.cpp

@@ -5749,6 +5749,94 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
   }
 
   llvm::Type* InitType = Init->getType();
+  if (ASTTy.hasCHERIoTSealedAttr()) {
+    // In this case, the global variable refers to a to-be-sealed value of type
+    // <T>. We need to:
+    // 1. Get (or create) the type of the sealed pointer, which has a fixed
+    // shape: `__Sealed_<T> {uint32_t sealing_key_ptr; uint32_t padding; <T>
+    // body}`;
+    // 2. Change the type of the global variable to `__Sealed_<T>`;
+    // 3. Create the initializer for the global variable.
+
+    auto *Module = &getModule();
+    auto *Ctxt = &getLLVMContext();
+    auto *Int32Ty = llvm::Type::getInt32Ty(*Ctxt);
+    CHERIoTSealedTypeAttr *Attr = nullptr;
+    std::string TypeName;
+
+    if (auto *TT = ASTTy->getAsTagDecl()) {
+      if (TT->hasAttr<CHERIoTSealedTypeAttr>()) {
+        Attr = TT->getAttr<CHERIoTSealedTypeAttr>();
+        TypeName = TT->getNameAsString();
+      }
+    }
+
+    if (const auto *TT = ASTTy->getAs<TypedefType>()) {
+      auto *TD = TT->getDecl();
+      if (TD->hasAttr<CHERIoTSealedTypeAttr>()) {
+        Attr = TD->getAttr<CHERIoTSealedTypeAttr>();
+        TypeName = TD->getNameAsString();
+      }
+    }
+
+    llvm::Comdat *C = Module->getOrInsertComdat(D->getName());
+    C->setSelectionKind(llvm::Comdat::Any);
+
+    std::vector<llvm::Type *> SealedStructElements = {Int32Ty, Int32Ty,
+                                                      Init->getType()};
+    auto *SealedStructType = llvm::StructType::create(
+        *Ctxt, SealedStructElements, ("struct.__Sealed_" + TypeName));
+
+    auto SealingKeyName =
+        ("__export.sealing_type." + Attr->getCompartmentName() + "." +
+         Attr->getSealingTypeName())
+            .str();
+
+    auto *SealingKeyRef = Module->getGlobalVariable(SealingKeyName);
+
+    if (!SealingKeyRef) {
+      llvm::GlobalVariable *SealingKeyVarDef = new llvm::GlobalVariable(
+          *Module, Int32Ty, false, llvm::GlobalValue::ExternalLinkage, nullptr,
+          SealingKeyName);
+      SealingKeyVarDef->setDSOLocal(true);
+      SealingKeyVarDef->setAlignment(llvm::Align(4));
+    }
+
+    SealingKeyRef = Module->getGlobalVariable(SealingKeyName);
+    auto *CastedPointer =
+        llvm::ConstantExpr::getPointerCast(SealingKeyRef, Int32Ty);
+
+    std::vector<llvm::Constant *> Values = {
+        CastedPointer, llvm::ConstantInt::get(Int32Ty, 0), Init};
+
+    // Create the initializer.
+    auto *Init = llvm::ConstantStruct::get(SealedStructType, Values);
+
+    llvm::Constant *Entry =
+        GetAddrOfGlobalVar(D, Init->getType(), ForDefinition_t(!IsTentative));
+    auto *GV = dyn_cast<llvm::GlobalVariable>(Entry);
+    GV->setSection(".sealed_objects");
+    GV->setDSOLocal(true);
+    GV->setLinkage(llvm::GlobalValue::LinkOnceODRLinkage);
+    GV->setAlignment(llvm::Align(4));
+    GV->setComdat(C);
+    GV->setInitializer(Init);
+    GV->addAttribute(llvm::CHERIoTSealedValueAttr::getAttrName());
+
+    addCompilerUsedGlobal(GV);
+
+    if (emitter)
+      emitter->finalize(GV);
+
+    SanitizerMD->reportGlobal(GV, *D, NeedsGlobalCtor);
+
+    // Emit global variable debug information.
+    if (CGDebugInfo *DI = getModuleDebugInfo())
+      if (getCodeGenOpts().hasReducedDebugInfo())
+        DI->EmitGlobalVariable(GV, D);
+    return;
+  }
+
   llvm::Constant *Entry =
       GetAddrOfGlobalVar(D, InitType, ForDefinition_t(!IsTentative));
 

clang/lib/Sema/SemaDecl.cpp

@@ -14947,10 +14947,19 @@ Sema::DeclGroupPtrTy Sema::FinalizeDeclaratorGroup(Scope *S, const DeclSpec &DS,
   for (Decl *D : Group) {
     if (!D)
       continue;
+
+    auto *VD = dyn_cast<VarDecl>(D);
+
+    if (VD && VD->getType().hasCHERIoTSealedAttr()) {
+      if (VD->hasExternalStorage()) {
+        Diag(D->getLocation(), diag::err_cheriot_invalid_sealed_declaration)
+            << "external";
+      }
+    }
+
     // Check if the Decl has been declared in '#pragma omp declare target'
     // directive and has static storage duration.
-    if (auto *VD = dyn_cast<VarDecl>(D);
-        LangOpts.OpenMP && VD && VD->hasAttr<OMPDeclareTargetDeclAttr>() &&
+    if (LangOpts.OpenMP && VD && VD->hasAttr<OMPDeclareTargetDeclAttr>() &&
         VD->hasGlobalStorage())
       OpenMP().ActOnOpenMPDeclareTargetInitializer(D);
     // For declarators, there are some additional syntactic-ish checks we need

clang/lib/Sema/SemaDeclAttr.cpp

@@ -2227,6 +2227,30 @@ static void handleCHERIOTSharedObject(Sema &S, Decl *D, const ParsedAttr &Attr,
       S.Context, Attr, ObjectName, OwnedPermissions));
 }
 
+static void handleCHERIoTSealedType(Sema &S, Decl *D, const ParsedAttr &Attr,
+                                    Sema::DeclAttributeLocation DAL) {
+  auto *Tag = dyn_cast<TagDecl>(D);
+  auto *Tydef = dyn_cast<TypedefDecl>(D);
+  if (Tag || Tydef) {
+    StringRef CompartmentName;
+    SourceLocation CompartmentNameLiteralLoc;
+    if (!S.checkStringLiteralArgumentAttr(Attr, 0, CompartmentName,
+                                          &CompartmentNameLiteralLoc))
+      return;
+    StringRef SealingTypeName;
+    SourceLocation SealingTypeNameLiteralLoc;
+
+    if (!S.checkStringLiteralArgumentAttr(Attr, 1, SealingTypeName,
+                                          &SealingTypeNameLiteralLoc))
+      return;
+
+    // Here we simply copy the attribute.
+    D->addAttr(::new (S.Context) CHERIoTSealedTypeAttr(
+        S.Context, Attr, CompartmentName, SealingTypeName));
+  } else // Ignore if it's not a type definition.
+    S.Diag(Attr.getLoc(), diag::warn_attribute_ignored) << Attr.getAttrName();
+}
+
 static void handleCHERICompartmentName(Sema &S, Decl *D, const ParsedAttr &Attr,
                                        Sema::DeclAttributeLocation DAL) {
   // cheri_compartment is both:
@@ -7480,6 +7504,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
   case ParsedAttr::AT_CHERIOTSharedObject:
     handleCHERIOTSharedObject(S, D, AL, DAL);
     break;
+  case ParsedAttr::AT_CHERIoTSealedType:
+    handleCHERIoTSealedType(S, D, AL, DAL);
+    break;
   case ParsedAttr::AT_InterruptState:
     handleInterruptState(S, D, AL);
     break;

clang/lib/Sema/SemaExpr.cpp

@@ -704,6 +704,12 @@ ExprResult Sema::DefaultLvalueConversion(Expr *E) {
   QualType T = E->getType();
   assert(!T.isNull() && "r-value conversion on typeless expression?");
 
+  // CHERIoT-specific check.
+  if (T.hasCHERIoTSealedAttr() && !isUnevaluatedContext()) {
+    return ExprError(
+        Diag(E->getExprLoc(), diag::err_cheriot_non_addr_of_expr_on_sealed));
+  }
+
   // lvalue-to-rvalue conversion cannot be applied to types that decay to
   // pointers (i.e. function or array types).
   if (T->canDecayToPointerType())
@@ -4906,6 +4912,11 @@ ExprResult Sema::ActOnArraySubscriptExpr(Scope *S, Expr *base,
                                          MultiExprArg ArgExprs,
                                          SourceLocation rbLoc) {
 
+  // CHERIoT-specific check.
+  if (base->getType().hasCHERIoTSealedAttr() && !isUnevaluatedContext()) {
+    return ExprError(Diag(lbLoc, diag::err_cheriot_non_addr_of_expr_on_sealed));
+  }
+
   if (base && !base->getType().isNull() &&
       base->hasPlaceholderType(BuiltinType::ArraySection)) {
     auto *AS = cast<ArraySectionExpr>(base);
@@ -8535,6 +8546,21 @@ QualType Sema::CheckConditionalOperands(ExprResult &Cond, ExprResult &LHS,
                                         ExprObjectKind &OK,
                                         SourceLocation QuestionLoc) {
 
+  // CHERIoT-specific check.
+  if (!isUnevaluatedContext()) {
+    if (LHS.get()->getType().hasCHERIoTSealedAttr()) {
+      Diag(LHS.get()->getExprLoc(),
+           diag::err_cheriot_non_addr_of_expr_on_sealed);
+      return QualType();
+    }
+
+    if (RHS.get()->getType().hasCHERIoTSealedAttr()) {
+      Diag(RHS.get()->getExprLoc(),
+           diag::err_cheriot_non_addr_of_expr_on_sealed);
+      return QualType();
+    }
+  }
+
   ExprResult LHSResult = CheckPlaceholderExpr(LHS.get());
   if (!LHSResult.isUsable()) return QualType();
   LHS = LHSResult;
@@ -14717,6 +14743,9 @@ QualType Sema::CheckAddressOfOperand(ExprResult &OrigOp, SourceLocation OpLoc) {
   CheckAddressOfPackedMember(op);
 
   PointerInterpretationKind PIK = PointerInterpretationForBaseExpr(op);
+  if (op->getType().hasCHERIoTSealedAttr()) {
+    PIK = PIK_SealedCapability;
+  }
   return Context.getPointerType(op->getType(), PIK);
 }
 
@@ -15659,6 +15688,19 @@ ExprResult Sema::BuildBinOp(Scope *S, SourceLocation OpLoc,
   LHSExpr = LHS.get();
   RHSExpr = RHS.get();
 
+  // CHERIoT-specific check.
+  if (!isUnevaluatedContext()) {
+    if (LHSExpr->getType().hasCHERIoTSealedAttr()) {
+      return ExprError(Diag(LHS.get()->getExprLoc(),
+                            diag::err_cheriot_non_addr_of_expr_on_sealed));
+    }
+
+    if (RHSExpr->getType().hasCHERIoTSealedAttr()) {
+      return ExprError(Diag(RHS.get()->getExprLoc(),
+                            diag::err_cheriot_non_addr_of_expr_on_sealed));
+    }
+  }
+
   // We want to end up calling one of SemaPseudoObject::checkAssignment
   // (if the LHS is a pseudo-object), BuildOverloadedBinOp (if
   // both expressions are overloadable or either is type-dependent),
@@ -16097,6 +16139,12 @@ bool Sema::isQualifiedMemberAccess(Expr *E) {
 ExprResult Sema::BuildUnaryOp(Scope *S, SourceLocation OpLoc,
                               UnaryOperatorKind Opc, Expr *Input,
                               bool IsAfterAmp) {
+  // CHERIoT-specific check.
+  if (Input->getType().hasCHERIoTSealedAttr() && (Opc != UO_AddrOf) &&
+      !isUnevaluatedContext()) {
+    return ExprError(Diag(OpLoc, diag::err_cheriot_non_addr_of_expr_on_sealed));
+  }
+
   // First things first: handle placeholders so that the
   // overloaded-operator check considers the right type.
   if (const BuiltinType *pty = Input->getType()->getAsPlaceholderType()) {

clang/lib/Sema/SemaExprCXX.cpp

@@ -6999,6 +6999,21 @@ QualType Sema::CXXCheckConditionalOperands(ExprResult &Cond, ExprResult &LHS,
                                            ExprResult &RHS, ExprValueKind &VK,
                                            ExprObjectKind &OK,
                                            SourceLocation QuestionLoc) {
+  // CHERIoT-specific check.
+  if (!isUnevaluatedContext()) {
+    if (LHS.get()->getType().hasCHERIoTSealedAttr()) {
+      Diag(LHS.get()->getExprLoc(),
+           diag::err_cheriot_non_addr_of_expr_on_sealed);
+      return QualType();
+    }
+
+    if (RHS.get()->getType().hasCHERIoTSealedAttr()) {
+      Diag(RHS.get()->getExprLoc(),
+           diag::err_cheriot_non_addr_of_expr_on_sealed);
+      return QualType();
+    }
+  }
+
   // FIXME: Handle C99's complex types, block pointers and Obj-C++ interface
   // pointers.