[asan] Copy arguments passed by value into explicit allocas for ASan

  Summary:
  ASan determines the stack layout from alloca instructions. Since
arguments marked as "byval" do not have an explicit alloca instruction, ASan
does not produce red zones for them. This commit produces an explicit alloca
instruction and copies the byval argument into the allocated memory so that red
zones are produced.

  Submitted on behalf of @morehouse (Matt Morehouse)

  Reviewers: eugenis, vitalybuka

  Reviewed By: eugenis

  Subscribers: hiraditya, llvm-commits

  Differential Revision: https://reviews.llvm.org/D34789

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@308387 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/lib/Transforms/Instrumentation/AddressSanitizer.cpp
index 184940b..057f746 100644 (file)
--- a/lib/Transforms/Instrumentation/AddressSanitizer.cpp
+++ b/lib/Transforms/Instrumentation/AddressSanitizer.cpp
@@ -22,9 +22,11 @@
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/Triple.h"
+#include "llvm/ADT/Twine.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/Argument.h"
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/DIBuilder.h"
 #include "llvm/IR/DataLayout.h"
@@ -43,6 +45,7 @@
 #include "llvm/Support/DataTypes.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/Endian.h"
+#include "llvm/Support/ScopedPrinter.h"
 #include "llvm/Support/SwapByteOrder.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Instrumentation.h"
@@ -192,6 +195,11 @@ static cl::opt<uint32_t> ClMaxInlinePoisoningSize(
 static cl::opt<bool> ClUseAfterReturn("asan-use-after-return",
                                       cl::desc("Check stack-use-after-return"),
                                       cl::Hidden, cl::init(true));
+static cl::opt<bool> ClRedzoneByvalArgs("asan-redzone-byval-args",
+                                        cl::desc("Create redzones for byval "
+                                                 "arguments (extra copy "
+                                                 "required)"), cl::Hidden,
+                                        cl::init(true));
 static cl::opt<bool> ClUseAfterScope("asan-use-after-scope",
                                      cl::desc("Check stack-use-after-scope"),
                                      cl::Hidden, cl::init(false));
@@ -747,6 +755,9 @@ struct FunctionStackPoisoner : public InstVisitor<FunctionStackPoisoner> {
 
   bool runOnFunction() {
     if (!ClStack) return false;
+
+    if (ClRedzoneByvalArgs) copyArgsPassedByValToAllocas();
+
     // Collect alloca, ret, lifetime instructions etc.
     for (BasicBlock *BB : depth_first(&F.getEntryBlock())) visit(*BB);
 
@@ -763,6 +774,11 @@ struct FunctionStackPoisoner : public InstVisitor<FunctionStackPoisoner> {
     return true;
   }
 
+  // Arguments marked with the "byval" attribute are implicitly copied without
+  // using an alloca instruction.  To produce redzones for those arguments, we
+  // copy them a second time into memory allocated with an alloca instruction.
+  void copyArgsPassedByValToAllocas();
+
   // Finds all Alloca instructions and puts
   // poisoned red zones around all of them.
   // Then unpoison everything back before the function returns.
@@ -2528,6 +2544,28 @@ static int StackMallocSizeClass(uint64_t LocalStackSize) {
   llvm_unreachable("impossible LocalStackSize");
 }
 
+void FunctionStackPoisoner::copyArgsPassedByValToAllocas() {
+  BasicBlock &FirstBB = *F.begin();
+  IRBuilder<> IRB(&FirstBB, FirstBB.getFirstInsertionPt());
+  const DataLayout &DL = F.getParent()->getDataLayout();
+  for (Argument &Arg : F.args()) {
+    if (Arg.hasByValAttr()) {
+      Type *Ty = Arg.getType()->getPointerElementType();
+      unsigned Align = Arg.getParamAlignment();
+      if (Align == 0) Align = DL.getABITypeAlignment(Ty);
+
+      const std::string &Name = Arg.hasName() ? Arg.getName().str() :
+          "Arg" + llvm::to_string(Arg.getArgNo());
+      AllocaInst *AI = IRB.CreateAlloca(Ty, nullptr, Twine(Name) + ".byval");
+      AI->setAlignment(Align);
+      Arg.replaceAllUsesWith(AI);
+
+      uint64_t AllocSize = DL.getTypeAllocSize(Ty);
+      IRB.CreateMemCpy(AI, &Arg, AllocSize, Align);
+    }
+  }
+}
+
 PHINode *FunctionStackPoisoner::createPHI(IRBuilder<> &IRB, Value *Cond,
                                           Value *ValueIfTrue,
                                           Instruction *ThenTerm,

diff --git a/test/Instrumentation/AddressSanitizer/stack-poisoning-byval-args.ll b/test/Instrumentation/AddressSanitizer/stack-poisoning-byval-args.ll
new file mode 100644 (file)
index 0000000..32610ce
--- /dev/null
+++ b/test/Instrumentation/AddressSanitizer/stack-poisoning-byval-args.ll
@@ -0,0 +1,48 @@
+; This check verifies that arguments passed by value get redzones.
+; RUN: opt < %s -asan -asan-realign-stack=32 -S | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.A = type { [8 x i32] }
+
+declare i32 @bar(%struct.A*)
+
+; Test behavior for named argument with explicit alignment.  The memcpy and
+; alloca alignments should match the explicit alignment of 64.
+define void @foo(%struct.A* byval align 64 %a) sanitize_address {
+entry:
+; CHECK-LABEL: foo
+; CHECK: call i64 @__asan_stack_malloc
+; CHECK: alloca i8, i64 {{.*}} align 64
+; CHECK: [[copyPtr:%[^ \t]+]] = inttoptr i64 %{{[^ \t]+}} to %struct.A*
+; CHECK: [[copyBytePtr:%[^ \t]+]] = bitcast %struct.A* [[copyPtr]]
+; CHECK: [[aBytePtr:%[^ \t]+]] = bitcast %struct.A* %a
+; CHECK: call void @llvm.memcpy{{[^%]+}}[[copyBytePtr]]{{[^%]+}}[[aBytePtr]],{{[^,]+}}, i32 64
+; CHECK: call i32 @bar(%struct.A* [[copyPtr]])
+; CHECK: ret void
+
+  %call = call i32 @bar(%struct.A* %a)
+  ret void
+}
+
+; Test behavior for unnamed argument without explicit alignment.  In this case,
+; the first argument is referenced by the identifier %0 and the ABI requires a
+; minimum alignment of 4 bytes since struct.A contains i32s which have 4-byte
+; alignment.  However, the alloca alignment will be 32 since that is the value
+; passed via the -asan-realign-stack option, which is greater than 4.
+define void @baz(%struct.A* byval) sanitize_address {
+entry:
+; CHECK-LABEL: baz
+; CHECK: call i64 @__asan_stack_malloc
+; CHECK: alloca i8, i64 {{.*}} align 32
+; CHECK: [[copyPtr:%[^ \t]+]] = inttoptr i64 %{{[^ \t]+}} to %struct.A*
+; CHECK: [[copyBytePtr:%[^ \t]+]] = bitcast %struct.A* [[copyPtr]]
+; CHECK: [[aBytePtr:%[^ \t]+]] = bitcast %struct.A* %0
+; CHECK: call void @llvm.memcpy{{[^%]+}}[[copyBytePtr]]{{[^%]+}}[[aBytePtr]],{{[^,]+}}, i32 4
+; CHECK: call i32 @bar(%struct.A* [[copyPtr]])
+; CHECK: ret void
+
+  %call = call i32 @bar(%struct.A* %0)
+  ret void
+}