[asan] Instrument comdat globals on COFF targets

Summary:
If we can use comdats, then we can make it so that the global metadata
is thrown away if the prevailing definition of the global was
uninstrumented. I have only tested this on COFF targets, but in theory,
there is no reason that we cannot also do this for ELF.

This will allow us to re-enable string merging with ASan on Windows,
reducing the binary size cost of ASan on Windows.

Reviewers: eugenis, vitalybuka

Subscribers: hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D47841

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334313 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/lib/Transforms/Instrumentation/AddressSanitizer.cpp
index 3c56869..18682c6 100644 (file)
--- a/lib/Transforms/Instrumentation/AddressSanitizer.cpp
+++ b/lib/Transforms/Instrumentation/AddressSanitizer.cpp
@@ -1655,14 +1655,6 @@ bool AddressSanitizerModule::ShouldInstrumentGlobal(GlobalVariable *G) {
   if (!Ty->isSized()) return false;
   if (!G->hasInitializer()) return false;
   if (GlobalWasGeneratedByCompiler(G)) return false; // Our own globals.
-  // Touch only those globals that will not be defined in other modules.
-  // Don't handle ODR linkage types and COMDATs since other modules may be built
-  // without ASan.
-  if (G->getLinkage() != GlobalVariable::ExternalLinkage &&
-      G->getLinkage() != GlobalVariable::PrivateLinkage &&
-      G->getLinkage() != GlobalVariable::InternalLinkage)
-    return false;
-  if (G->hasComdat()) return false;
   // Two problems with thread-locals:
   //   - The address of the main thread's copy can't be computed at link-time.
   //   - Need to poison all copies, not just the main thread's one.
@@ -1670,6 +1662,38 @@ bool AddressSanitizerModule::ShouldInstrumentGlobal(GlobalVariable *G) {
   // For now, just ignore this Global if the alignment is large.
   if (G->getAlignment() > MinRedzoneSizeForGlobal()) return false;
 
+  // If we know this global is defined only in this module (external or static),
+  // we can instrument it.
+  if (G->getLinkage() != GlobalVariable::ExternalLinkage &&
+      G->getLinkage() != GlobalVariable::PrivateLinkage &&
+      G->getLinkage() != GlobalVariable::InternalLinkage) {
+    // For targets that support comdats, we can instrument ODR globals in
+    // comdats. We simply add our global metadata to the group and hope that an
+    // instrumented global is selected.
+    // FIXME: This should work on ELF, enable it there too.
+    if (!TargetTriple.isOSBinFormatCOFF())
+      return false;
+    if (G->getLinkage() != GlobalVariable::WeakODRLinkage &&
+        G->getLinkage() != GlobalVariable::LinkOnceODRLinkage)
+      return false;
+    if (!G->hasComdat())
+      return false;
+  }
+
+  // If a comdat is present, it must have a selection kind that implies ODR
+  // semantics: no duplicates, any, or exact match.
+  if (Comdat *C = G->getComdat()) {
+    switch (C->getSelectionKind()) {
+    case Comdat::Any:
+    case Comdat::ExactMatch:
+    case Comdat::NoDuplicates:
+      break;
+    case Comdat::Largest:
+    case Comdat::SameSize:
+      return false;
+    }
+  }
+
   if (G->hasSection()) {
     StringRef Section = G->getSection();
 
@@ -2119,6 +2143,7 @@ bool AddressSanitizerModule::InstrumentGlobals(IRBuilder<> &IRB, Module &M, bool
         new GlobalVariable(M, NewTy, G->isConstant(), Linkage, NewInitializer,
                            "", G, G->getThreadLocalMode());
     NewGlobal->copyAttributesFrom(G);
+    NewGlobal->setComdat(G->getComdat());
     NewGlobal->setAlignment(MinRZ);
 
     // Move null-terminated C strings to "__asan_cstring" section on Darwin.

diff --git a/test/Instrumentation/AddressSanitizer/win-string-literal.ll b/test/Instrumentation/AddressSanitizer/win-string-literal.ll
new file mode 100644 (file)
index 0000000..f3ea1e9
--- /dev/null
+++ b/test/Instrumentation/AddressSanitizer/win-string-literal.ll
@@ -0,0 +1,46 @@
+; RUN: opt < %s -asan -asan-module -S | FileCheck %s
+
+; Generated like so:
+; $ clang -S -emit-llvm -Xclang -disable-llvm-passes -fsanitize=address -O1 t.cpp -o t.ll
+; const char *getstr() { return "asdf"; }
+
+; CHECK: $"??_C@_04JIHMPGLA@asdf?$AA@" = comdat any
+
+; CHECK: @"??_C@_04JIHMPGLA@asdf?$AA@" =
+; CHECK-SAME: linkonce_odr dso_local unnamed_addr constant { [5 x i8], [59 x i8] }
+; CHECK-SAME: { [5 x i8] c"asdf\00", [59 x i8] zeroinitializer }, comdat, align 32
+
+; CHECK: @"__asan_global_??_C@_04JIHMPGLA@asdf?$AA@" =
+; CHECK-SAME: private global { i64, i64, i64, i64, i64, i64, i64, i64 }
+; CHECK-SAME: { i64 ptrtoint ({ [5 x i8], [59 x i8] }* @"??_C@_04JIHMPGLA@asdf?$AA@" to i64),
+; CHECK-SAME:   i64 5, i64 64, i64 ptrtoint ([17 x i8]* @__asan_gen_.1 to i64),
+; CHECK-SAME:   i64 ptrtoint ([8 x i8]* @__asan_gen_ to i64), i64 0,
+; CHECK-SAME:   i64 ptrtoint ({ [6 x i8]*, i32, i32 }* @__asan_gen_.3 to i64), i64 0 },
+; CHECK-SAME:   section ".ASAN$GL", comdat($"??_C@_04JIHMPGLA@asdf?$AA@"), align 64
+
+; ModuleID = 't.cpp'
+source_filename = "t.cpp"
+target datalayout = "e-m:w-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-pc-windows-msvc19.13.26131"
+
+$"??_C@_04JIHMPGLA@asdf?$AA@" = comdat any
+
+@"??_C@_04JIHMPGLA@asdf?$AA@" = linkonce_odr dso_local unnamed_addr constant [5 x i8] c"asdf\00", comdat, align 1
+
+; Function Attrs: nounwind sanitize_address uwtable
+define dso_local i8* @"?getstr@@YAPEBDXZ"() #0 {
+entry:
+  ret i8* getelementptr inbounds ([5 x i8], [5 x i8]* @"??_C@_04JIHMPGLA@asdf?$AA@", i32 0, i32 0)
+}
+
+attributes #0 = { nounwind sanitize_address uwtable }
+
+!llvm.asan.globals = !{!0}
+!llvm.module.flags = !{!2, !3}
+!llvm.ident = !{!4}
+
+!0 = !{[5 x i8]* @"??_C@_04JIHMPGLA@asdf?$AA@", !1, !"<string literal>", i1 false, i1 false}
+!1 = !{!"t.cpp", i32 1, i32 31}
+!2 = !{i32 1, !"wchar_size", i32 2}
+!3 = !{i32 7, !"PIC Level", i32 2}
+!4 = !{!"clang version 7.0.0 "}