Auto merge of #95548 - rcvalle:rust-cfi-2, r=nagisa

Add fine-grained LLVM CFI support to the Rust compiler

This PR improves the LLVM Control Flow Integrity (CFI) support in the Rust compiler by providing forward-edge control flow protection for Rust-compiled code only by aggregating function pointers in groups identified by their return and parameter types.

Forward-edge control flow protection for C or C++ and Rust -compiled code "mixed binaries" (i.e., for when C or C++ and Rust -compiled code share the same virtual address space) will be provided in later work as part of this project by identifying C char and integer type uses at the time types are encoded (see Type metadata in the design document in the tracking issue #89653).

LLVM CFI can be enabled with -Zsanitizer=cfi and requires LTO (i.e., -Clto).

Thank you again, `@eddyb,` `@nagisa,` `@pcc,` and `@tmiasko` for all the help!

Cargo.lock

@@ -4519,6 +4519,7 @@ dependencies = [
 name = "rustc_symbol_mangling"
 version = "0.0.0"
 dependencies = [
+ "bitflags",
  "punycode",
  "rustc-demangle",
  "rustc_data_structures",

compiler/rustc_codegen_gcc/src/builder.rs

@@ -784,16 +784,6 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         // TODO(antoyo)
     }
 
-    fn type_metadata(&mut self, _function: RValue<'gcc>, _typeid: String) {
-        // Unsupported.
-    }
-
-    fn typeid_metadata(&mut self, _typeid: String) -> RValue<'gcc> {
-        // Unsupported.
-        self.context.new_rvalue_from_int(self.int_type, 0)
-    }
-
-
     fn store(&mut self, val: RValue<'gcc>, ptr: RValue<'gcc>, align: Align) -> RValue<'gcc> {
         self.store_with_flags(val, ptr, align, MemFlags::empty())
     }

compiler/rustc_codegen_gcc/src/type_.rs

@@ -1,7 +1,7 @@
 use std::convert::TryInto;
 
 use gccjit::{RValue, Struct, Type};
-use rustc_codegen_ssa::traits::{BaseTypeMethods, DerivedTypeMethods};
+use rustc_codegen_ssa::traits::{BaseTypeMethods, DerivedTypeMethods, TypeMembershipMethods};
 use rustc_codegen_ssa::common::TypeKind;
 use rustc_middle::{bug, ty};
 use rustc_middle::ty::layout::TyAndLayout;
@@ -290,3 +290,14 @@ pub fn struct_fields<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLayout
 
     (result, packed)
 }
+
+impl<'gcc, 'tcx> TypeMembershipMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
+    fn set_type_metadata(&self, _function: RValue<'gcc>, _typeid: String) {
+        // Unsupported.
+    }
+
+    fn typeid_metadata(&self, _typeid: String) -> RValue<'gcc> {
+        // Unsupported.
+        self.context.new_rvalue_from_int(self.int_type, 0)
+    }
+}

compiler/rustc_codegen_llvm/Cargo.toml

@@ -18,7 +18,6 @@ rustc_middle = { path = "../rustc_middle" }
 rustc-demangle = "0.1.21"
 rustc_attr = { path = "../rustc_attr" }
 rustc_codegen_ssa = { path = "../rustc_codegen_ssa" }
-rustc_symbol_mangling = { path = "../rustc_symbol_mangling" }
 rustc_data_structures = { path = "../rustc_data_structures" }
 rustc_errors = { path = "../rustc_errors" }
 rustc_fs_util = { path = "../rustc_fs_util" }
@@ -30,6 +29,7 @@ rustc_metadata = { path = "../rustc_metadata" }
 rustc_query_system = { path = "../rustc_query_system" }
 rustc_session = { path = "../rustc_session" }
 rustc_serialize = { path = "../rustc_serialize" }
+rustc_symbol_mangling = { path = "../rustc_symbol_mangling" }
 rustc_target = { path = "../rustc_target" }
 smallvec = { version = "1.8.1", features = ["union", "may_dangle"] }
 rustc_ast = { path = "../rustc_ast" }

compiler/rustc_codegen_llvm/src/builder.rs

@@ -626,32 +626,6 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         }
     }
 
-    fn type_metadata(&mut self, function: &'ll Value, typeid: String) {
-        let typeid_metadata = self.typeid_metadata(typeid);
-        let v = [self.const_usize(0), typeid_metadata];
-        unsafe {
-            llvm::LLVMGlobalSetMetadata(
-                function,
-                llvm::MD_type as c_uint,
-                llvm::LLVMValueAsMetadata(llvm::LLVMMDNodeInContext(
-                    self.cx.llcx,
-                    v.as_ptr(),
-                    v.len() as c_uint,
-                )),
-            )
-        }
-    }
-
-    fn typeid_metadata(&mut self, typeid: String) -> Self::Value {
-        unsafe {
-            llvm::LLVMMDStringInContext(
-                self.cx.llcx,
-                typeid.as_ptr() as *const c_char,
-                typeid.as_bytes().len() as c_uint,
-            )
-        }
-    }
-
     fn store(&mut self, val: &'ll Value, ptr: &'ll Value, align: Align) -> &'ll Value {
         self.store_with_flags(val, ptr, align, MemFlags::empty())
     }

compiler/rustc_codegen_llvm/src/declare.rs

@@ -18,7 +18,9 @@ use crate::llvm;
 use crate::llvm::AttributePlace::Function;
 use crate::type_::Type;
 use crate::value::Value;
+use rustc_codegen_ssa::traits::TypeMembershipMethods;
 use rustc_middle::ty::Ty;
+use rustc_symbol_mangling::typeid::typeid_for_fnabi;
 use smallvec::SmallVec;
 use tracing::debug;
 
@@ -97,6 +99,12 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
             fn_abi.llvm_type(self),
         );
         fn_abi.apply_attrs_llfn(self, llfn);
+
+        if self.tcx.sess.is_sanitizer_cfi_enabled() {
+            let typeid = typeid_for_fnabi(self.tcx, fn_abi);
+            self.set_type_metadata(llfn, typeid);
+        }
+
         llfn
     }
 

compiler/rustc_codegen_llvm/src/type_.rs

@@ -19,7 +19,7 @@ use rustc_target::abi::{AddressSpace, Align, Integer, Size};
 use std::fmt;
 use std::ptr;
 
-use libc::c_uint;
+use libc::{c_char, c_uint};
 
 impl PartialEq for Type {
     fn eq(&self, other: &Self) -> bool {
@@ -289,3 +289,31 @@ impl<'ll, 'tcx> LayoutTypeMethods<'tcx> for CodegenCx<'ll, 'tcx> {
         ty.llvm_type(self)
     }
 }
+
+impl<'ll, 'tcx> TypeMembershipMethods<'tcx> for CodegenCx<'ll, 'tcx> {
+    fn set_type_metadata(&self, function: &'ll Value, typeid: String) {
+        let typeid_metadata = self.typeid_metadata(typeid);
+        let v = [self.const_usize(0), typeid_metadata];
+        unsafe {
+            llvm::LLVMGlobalSetMetadata(
+                function,
+                llvm::MD_type as c_uint,
+                llvm::LLVMValueAsMetadata(llvm::LLVMMDNodeInContext(
+                    self.llcx,
+                    v.as_ptr(),
+                    v.len() as c_uint,
+                )),
+            )
+        }
+    }
+
+    fn typeid_metadata(&self, typeid: String) -> &'ll Value {
+        unsafe {
+            llvm::LLVMMDStringInContext(
+                self.llcx,
+                typeid.as_ptr() as *const c_char,
+                typeid.len() as c_uint,
+            )
+        }
+    }
+}

compiler/rustc_codegen_ssa/src/mir/block.rs

@@ -20,7 +20,7 @@ use rustc_middle::ty::print::{with_no_trimmed_paths, with_no_visible_paths};
 use rustc_middle::ty::{self, Instance, Ty, TypeVisitable};
 use rustc_span::source_map::Span;
 use rustc_span::{sym, Symbol};
-use rustc_symbol_mangling::typeid_for_fnabi;
+use rustc_symbol_mangling::typeid::typeid_for_fnabi;
 use rustc_target::abi::call::{ArgAbi, FnAbi, PassMode};
 use rustc_target::abi::{self, HasDataLayout, WrappingRange};
 use rustc_target::spec::abi::Abi;
@@ -918,7 +918,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             // FIXME(rcvalle): Add support for generalized identifiers.
             // FIXME(rcvalle): Create distinct unnamed MDNodes for internal identifiers.
             let typeid = typeid_for_fnabi(bx.tcx(), fn_abi);
-            let typeid_metadata = bx.typeid_metadata(typeid);
+            let typeid_metadata = self.cx.typeid_metadata(typeid);
 
             // Test whether the function pointer is associated with the type identifier.
             let cond = bx.type_test(fn_ptr, typeid_metadata);

compiler/rustc_codegen_ssa/src/mir/mod.rs

@@ -3,7 +3,6 @@ use rustc_middle::mir;
 use rustc_middle::mir::interpret::ErrorHandled;
 use rustc_middle::ty::layout::{FnAbiOf, HasTyCtxt, TyAndLayout};
 use rustc_middle::ty::{self, Instance, Ty, TypeFoldable, TypeVisitable};
-use rustc_symbol_mangling::typeid_for_fnabi;
 use rustc_target::abi::call::{FnAbi, PassMode};
 
 use std::iter;
@@ -247,13 +246,6 @@ pub fn codegen_mir<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
     for (bb, _) in traversal::reverse_postorder(&mir) {
         fx.codegen_block(bb);
     }
-
-    // For backends that support CFI using type membership (i.e., testing whether a given  pointer
-    // is associated with a type identifier).
-    if cx.tcx().sess.is_sanitizer_cfi_enabled() {
-        let typeid = typeid_for_fnabi(cx.tcx(), fn_abi);
-        bx.type_metadata(llfn, typeid);
-    }
 }
 
 /// Produces, for each argument, a `Value` pointing at the

compiler/rustc_codegen_ssa/src/traits/builder.rs

@@ -160,8 +160,6 @@ pub trait BuilderMethods<'a, 'tcx>:
 
     fn range_metadata(&mut self, load: Self::Value, range: WrappingRange);
     fn nonnull_metadata(&mut self, load: Self::Value);
-    fn type_metadata(&mut self, function: Self::Function, typeid: String);
-    fn typeid_metadata(&mut self, typeid: String) -> Self::Value;
 
     fn store(&mut self, val: Self::Value, ptr: Self::Value, align: Align) -> Self::Value;
     fn store_with_flags(

compiler/rustc_codegen_ssa/src/traits/mod.rs

@@ -40,7 +40,8 @@ pub use self::intrinsic::IntrinsicCallMethods;
 pub use self::misc::MiscMethods;
 pub use self::statics::{StaticBuilderMethods, StaticMethods};
 pub use self::type_::{
-    ArgAbiMethods, BaseTypeMethods, DerivedTypeMethods, LayoutTypeMethods, TypeMethods,
+    ArgAbiMethods, BaseTypeMethods, DerivedTypeMethods, LayoutTypeMethods, TypeMembershipMethods,
+    TypeMethods,
 };
 pub use self::write::{ModuleBufferMethods, ThinBufferMethods, WriteBackendMethods};
 

compiler/rustc_codegen_ssa/src/traits/type_.rs

@@ -117,6 +117,13 @@ pub trait LayoutTypeMethods<'tcx>: Backend<'tcx> {
     ) -> Self::Type;
 }
 
+// For backends that support CFI using type membership (i.e., testing whether a given  pointer is
+// associated with a type identifier).
+pub trait TypeMembershipMethods<'tcx>: Backend<'tcx> {
+    fn set_type_metadata(&self, function: Self::Function, typeid: String);
+    fn typeid_metadata(&self, typeid: String) -> Self::Value;
+}
+
 pub trait ArgAbiMethods<'tcx>: HasCodegen<'tcx> {
     fn store_fn_arg(
         &mut self,
@@ -133,6 +140,12 @@ pub trait ArgAbiMethods<'tcx>: HasCodegen<'tcx> {
     fn arg_memory_ty(&self, arg_abi: &ArgAbi<'tcx, Ty<'tcx>>) -> Self::Type;
 }
 
-pub trait TypeMethods<'tcx>: DerivedTypeMethods<'tcx> + LayoutTypeMethods<'tcx> {}
+pub trait TypeMethods<'tcx>:
+    DerivedTypeMethods<'tcx> + LayoutTypeMethods<'tcx> + TypeMembershipMethods<'tcx>
+{
+}
 
-impl<'tcx, T> TypeMethods<'tcx> for T where Self: DerivedTypeMethods<'tcx> + LayoutTypeMethods<'tcx> {}
+impl<'tcx, T> TypeMethods<'tcx> for T where
+    Self: DerivedTypeMethods<'tcx> + LayoutTypeMethods<'tcx> + TypeMembershipMethods<'tcx>
+{
+}

compiler/rustc_symbol_mangling/Cargo.toml

@@ -7,6 +7,7 @@ edition = "2021"
 doctest = false
 
 [dependencies]
+bitflags = "1.2.1"
 tracing = "0.1"
 punycode = "0.4.0"
 rustc-demangle = "0.1.21"

compiler/rustc_symbol_mangling/src/lib.rs

@@ -102,16 +102,16 @@ use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrs;
 use rustc_middle::mir::mono::{InstantiationMode, MonoItem};
 use rustc_middle::ty::query::Providers;
 use rustc_middle::ty::subst::SubstsRef;
-use rustc_middle::ty::{self, Instance, Ty, TyCtxt};
+use rustc_middle::ty::{self, Instance, TyCtxt};
 use rustc_session::config::SymbolManglingVersion;
-use rustc_target::abi::call::FnAbi;
 
 use tracing::debug;
 
 mod legacy;
 mod v0;
 
 pub mod test;
+pub mod typeid;
 
 /// This function computes the symbol name for the given `instance` and the
 /// given instantiating crate. That is, if you know that instance X is
@@ -150,11 +150,6 @@ fn symbol_name_provider<'tcx>(tcx: TyCtxt<'tcx>, instance: Instance<'tcx>) -> ty
     ty::SymbolName::new(tcx, &symbol_name)
 }
 
-/// This function computes the typeid for the given function ABI.
-pub fn typeid_for_fnabi<'tcx>(tcx: TyCtxt<'tcx>, fn_abi: &FnAbi<'tcx, Ty<'tcx>>) -> String {
-    v0::mangle_typeid_for_fnabi(tcx, fn_abi)
-}
-
 pub fn typeid_for_trait_ref<'tcx>(
     tcx: TyCtxt<'tcx>,
     trait_ref: ty::PolyExistentialTraitRef<'tcx>,

compiler/rustc_symbol_mangling/src/typeid.rs

@@ -0,0 +1,18 @@
+// For more information about type metadata and type metadata identifiers for cross-language LLVM
+// CFI support, see Type metadata in the design document in the tracking issue #89653.
+
+use rustc_middle::ty::{FnSig, Ty, TyCtxt};
+use rustc_target::abi::call::FnAbi;
+
+mod typeid_itanium_cxx_abi;
+use typeid_itanium_cxx_abi::TypeIdOptions;
+
+/// Returns a type metadata identifier for the specified FnAbi.
+pub fn typeid_for_fnabi<'tcx>(tcx: TyCtxt<'tcx>, fn_abi: &FnAbi<'tcx, Ty<'tcx>>) -> String {
+    typeid_itanium_cxx_abi::typeid_for_fnabi(tcx, fn_abi, TypeIdOptions::NO_OPTIONS)
+}
+
+/// Returns a type metadata identifier for the specified FnSig.
+pub fn typeid_for_fnsig<'tcx>(tcx: TyCtxt<'tcx>, fn_sig: &FnSig<'tcx>) -> String {
+    typeid_itanium_cxx_abi::typeid_for_fnsig(tcx, fn_sig, TypeIdOptions::NO_OPTIONS)
+}