add -fret-clean option (amd64 and i386 only at first), defaulting to off.

This causes the caller to cleans the return address off the stack after
a callq completes.  The option is best used in low-level libraries (such as
libc), because libc contains low-level system call stubs.  The option
reduces hints (found on the stale parts of the stack) about libc.so's mapping
location, and together with random-relinking, relro got/pic, and xonly
makes some exploit methods more difficult.
ok mortimer, mlarkin, much discussion with kettenis, in snaps for 2 weeks.

diff --git a/gnu/llvm/llvm/lib/Target/X86/X86RetClean.cpp b/gnu/llvm/llvm/lib/Target/X86/X86RetClean.cpp
new file mode 100644 (file)
index 0000000..623bfed
--- /dev/null
+++ b/gnu/llvm/llvm/lib/Target/X86/X86RetClean.cpp
@@ -0,0 +1,115 @@
+//===-- X86RetClean.cpp - Clean Retaddr off stack upon function return ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file defines a function pass that clears the ret-address from
+/// the top of the stack, immediately upon return to the caller, the goal
+/// is remove this subtle but powerful info-leak which hints at the
+/// address space location of the lower level libraries.
+///
+//===----------------------------------------------------------------------===//
+
+#include "X86.h"
+#include "X86InstrBuilder.h"
+#include "X86InstrInfo.h"
+#include "X86MachineFunctionInfo.h"
+#include "X86Subtarget.h"
+#include "X86TargetMachine.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+#define RETCLEAN_DESC "X86 Ret Clean"
+#define RETCLEAN_NAME "x86-ret-clean"
+
+#define DEBUG_TYPE RETCLEAN_NAME
+
+// Toggle with cc1 option: -mllvm -x86-ret-clean=<true|false>
+static cl::opt<bool> RetClean(
+    "x86-ret-clean", cl::Hidden,
+    cl::desc("clean return address off stack after call"),
+    cl::init(false));
+
+namespace {
+class RetCleanPass : public MachineFunctionPass {
+
+public:
+  static char ID;
+
+  StringRef getPassName() const override { return RETCLEAN_DESC; }
+
+  RetCleanPass()
+      : MachineFunctionPass(ID) {}
+
+  /// Loop over all the instructions and replace ret with ret+clean
+  bool runOnMachineFunction(MachineFunction &MF) override;
+
+  MachineFunctionProperties getRequiredProperties() const override {
+    return MachineFunctionProperties().set(
+        MachineFunctionProperties::Property::NoVRegs);
+  }
+
+private:
+  bool fixupInstruction(MachineFunction &MF, MachineBasicBlock &MBB,
+                        MachineInstr &MI);
+};
+char RetCleanPass::ID = 0;
+} // namespace
+
+FunctionPass *llvm::createX86RetCleanPass() {
+  return new RetCleanPass();
+}
+
+bool RetCleanPass::fixupInstruction(MachineFunction &MF,
+                               MachineBasicBlock &MBB,
+                               MachineInstr &MI) {
+
+  const X86InstrInfo *TII = MF.getSubtarget<X86Subtarget>().getInstrInfo();
+  bool Is64Bit = MF.getTarget().getTargetTriple().getArch() == Triple::x86_64;
+  unsigned Opc = Is64Bit ? X86::MOV64mi32 : X86::MOV32mi;
+  unsigned Offset = Is64Bit ? -8 : -4;
+  Register SPReg = Is64Bit ? X86::RSP : X86::ESP;
+
+  // add "movq $0, -8(%rsp)" (or similar) in caller, to clear the
+  // ret-addr info-leak off the stack
+  addRegOffset(BuildMI(MBB, MI, MI.getDebugLoc(), TII->get(Opc)),
+    SPReg, false, Offset)
+    .addImm(0);
+  return true;
+}
+
+bool RetCleanPass::runOnMachineFunction(MachineFunction &MF) {
+  if (!RetClean)
+    return false;
+
+  bool modified = false;
+
+  for (auto &MBB : MF) {
+    std::vector<MachineInstr*> fixups;
+    bool foundcall = false;
+
+    for (auto &MI : MBB) {
+      if (MI.isCall()) {
+        foundcall = true;      // queue the insert before the next MI
+      } else if (foundcall) {
+         fixups.push_back(&MI);
+         foundcall = false;
+      }
+    }
+    for (auto *fixup : fixups)
+      modified |= fixupInstruction(MF, MBB, *fixup);
+  }
+  return modified;
+}