[asan] Fix dead stripping of globals on Linux (compiler-rt)

[eugenis]

Summary

Runtime support for the new instrumentation of globals based on !associated, and a bunch of tests.

[eugenis]

+llvm-commits

[eugenis]

rebased

[pcc]

LGTM

compiler-rt/trunk/lib/asan/asan_globals.cc

 // This mirrors __asan_register_image_globals.
 void __asan_unregister_image_globals(uptr *flag) {
   if (!*flag)
     return;
   AsanApplyToGlobals(__asan_unregister_globals, flag);
   *flag = 0;
 }
 
+void __asan_register_elf_globals(uptr *flag, void *start, void *stop) {
+  if (*flag) return;
+  if (!start) return;
+  CHECK_EQ(0, ((uptr)stop - (uptr)start) % sizeof(__asan_global));
+  __asan_global *globals_start = (__asan_global*)start;
+  __asan_global *globals_stop = (__asan_global*)stop;
+  __asan_register_globals(globals_start, globals_stop - globals_start);
+  *flag = 1;
+}
+
+void __asan_unregister_elf_globals(uptr *flag, void *start, void *stop) {
+  if (!*flag) return;
+  if (!start) return;
+  CHECK_EQ(0, ((uptr)stop - (uptr)start) % sizeof(__asan_global));
+  __asan_global *globals_start = (__asan_global*)start;
+  __asan_global *globals_stop = (__asan_global*)stop;
+  __asan_unregister_globals(globals_start, globals_stop - globals_start);
+  *flag = 0;
+}
+
 // Register an array of globals.
 void __asan_register_globals(__asan_global *globals, uptr n) {
   if (!flags()->report_globals) return;
   GET_STACK_TRACE_MALLOC;
   u32 stack_id = StackDepotPut(stack);
   BlockingMutexLock lock(&mu_for_globals);
   if (!global_registration_site_vector)
     global_registration_site_vector =

compiler-rt/trunk/lib/asan/asan_interface.inc

 INTERFACE_FUNCTION(__asan_memset)
 INTERFACE_FUNCTION(__asan_poison_cxx_array_cookie)
 INTERFACE_FUNCTION(__asan_poison_intra_object_redzone)
 INTERFACE_FUNCTION(__asan_poison_memory_region)
 INTERFACE_FUNCTION(__asan_poison_stack_memory)
 INTERFACE_FUNCTION(__asan_print_accumulated_stats)
 INTERFACE_FUNCTION(__asan_region_is_poisoned)
 INTERFACE_FUNCTION(__asan_register_globals)
+INTERFACE_FUNCTION(__asan_register_elf_globals)
 INTERFACE_FUNCTION(__asan_register_image_globals)
 INTERFACE_FUNCTION(__asan_report_error)
 INTERFACE_FUNCTION(__asan_report_exp_load1)
 INTERFACE_FUNCTION(__asan_report_exp_load2)
 INTERFACE_FUNCTION(__asan_report_exp_load4)
 INTERFACE_FUNCTION(__asan_report_exp_load8)
 INTERFACE_FUNCTION(__asan_report_exp_load16)
 INTERFACE_FUNCTION(__asan_report_exp_load_n)

 INTERFACE_FUNCTION(__asan_store4_noabort)
 INTERFACE_FUNCTION(__asan_store8_noabort)
 INTERFACE_FUNCTION(__asan_store16_noabort)
 INTERFACE_FUNCTION(__asan_storeN_noabort)
 INTERFACE_FUNCTION(__asan_unpoison_intra_object_redzone)
 INTERFACE_FUNCTION(__asan_unpoison_memory_region)
 INTERFACE_FUNCTION(__asan_unpoison_stack_memory)
 INTERFACE_FUNCTION(__asan_unregister_globals)
+INTERFACE_FUNCTION(__asan_unregister_elf_globals)
 INTERFACE_FUNCTION(__asan_unregister_image_globals)
 INTERFACE_FUNCTION(__asan_version_mismatch_check_v8)
 INTERFACE_FUNCTION(__sanitizer_finish_switch_fiber)
 INTERFACE_FUNCTION(__sanitizer_print_stack_trace)
 INTERFACE_FUNCTION(__sanitizer_ptr_cmp)
 INTERFACE_FUNCTION(__sanitizer_ptr_sub)
 INTERFACE_FUNCTION(__sanitizer_start_switch_fiber)
 INTERFACE_FUNCTION(__sanitizer_unaligned_load16)
 INTERFACE_FUNCTION(__sanitizer_unaligned_load32)
 INTERFACE_FUNCTION(__sanitizer_unaligned_load64)
 INTERFACE_FUNCTION(__sanitizer_unaligned_store16)
 INTERFACE_FUNCTION(__sanitizer_unaligned_store32)
 INTERFACE_FUNCTION(__sanitizer_unaligned_store64)
 INTERFACE_WEAK_FUNCTION(__asan_default_options)
 INTERFACE_WEAK_FUNCTION(__asan_default_suppressions)
 INTERFACE_WEAK_FUNCTION(__asan_on_error)

compiler-rt/trunk/lib/asan/asan_interface_internal.h

extern "C" {
   // These functions can be called on some platforms to find globals in the same
   // loaded image as `flag' and apply __asan_(un)register_globals to them,
   // filtering out redundant calls.
   SANITIZER_INTERFACE_ATTRIBUTE
   void __asan_register_image_globals(uptr *flag);
   SANITIZER_INTERFACE_ATTRIBUTE
   void __asan_unregister_image_globals(uptr *flag);
 
+  SANITIZER_INTERFACE_ATTRIBUTE
+  void __asan_register_elf_globals(uptr *flag, void *start, void *stop);
+  SANITIZER_INTERFACE_ATTRIBUTE
+  void __asan_unregister_elf_globals(uptr *flag, void *start, void *stop);
+
   // These two functions should be called by the instrumented code.
   // 'globals' is an array of structures describing 'n' globals.
   SANITIZER_INTERFACE_ATTRIBUTE
   void __asan_register_globals(__asan_global *globals, uptr n);
   SANITIZER_INTERFACE_ATTRIBUTE
   void __asan_unregister_globals(__asan_global *globals, uptr n);
 
   // These two functions should be called before and after dynamic initializers

compiler-rt/trunk/test/asan/CMakeLists.txt

macro(get_bits_for_arch arch bits)
   else()
     message(FATAL_ERROR "Unknown target architecture: ${arch}")
   endif()
 endmacro()
 
 set(ASAN_TEST_DEPS ${SANITIZER_COMMON_LIT_TEST_DEPS})
 if(NOT COMPILER_RT_STANDALONE_BUILD)
   list(APPEND ASAN_TEST_DEPS asan)
-  if(WIN32 AND COMPILER_RT_HAS_LLD_SOURCES)
+  if(NOT APPLE AND COMPILER_RT_HAS_LLD_SOURCES)
     list(APPEND ASAN_TEST_DEPS
       lld
     )
   endif()
 endif()
 set(ASAN_DYNAMIC_TEST_DEPS ${ASAN_TEST_DEPS})
 
 set(ASAN_TEST_ARCH ${ASAN_SUPPORTED_ARCH})

compiler-rt/trunk/test/asan/TestCases/Linux/global-overflow-bfd.cc

+// Test that gc-sections-friendly instrumentation of globals does not introduce
+// false negatives with the BFD linker.
+// RUN: %clangxx_asan -fuse-ld=bfd -Wl,-gc-sections -ffunction-sections -fdata-sections -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
+
+#include <string.h>
+int main(int argc, char **argv) {
+  static char XXX[10];
+  static char YYY[10];
+  static char ZZZ[10];
+  memset(XXX, 0, 10);
+  memset(YYY, 0, 10);
+  memset(ZZZ, 0, 10);
+  int res = YYY[argc * 10];  // BOOOM
+  // CHECK: {{READ of size 1 at}}
+  // CHECK: {{located 0 bytes to the right of global variable}}
+  res += XXX[argc] + ZZZ[argc];
+  return res;
+}

compiler-rt/trunk/test/asan/TestCases/Linux/global-overflow-lld.cc

+// Test that gc-sections-friendly instrumentation of globals does not introduce
+// false negatives with the LLD linker.
+// RUN: %clangxx_asan -fuse-ld=lld -Wl,-gc-sections -ffunction-sections -fdata-sections -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
+// REQUIRES: lld
+
+#include <string.h>
+int main(int argc, char **argv) {
+  static char XXX[10];
+  static char YYY[10];
+  static char ZZZ[10];
+  memset(XXX, 0, 10);
+  memset(YYY, 0, 10);
+  memset(ZZZ, 0, 10);
+  int res = YYY[argc * 10];  // BOOOM
+  // CHECK: {{READ of size 1 at}}
+  // CHECK: {{located 0 bytes to the right of global variable}}
+  res += XXX[argc] + ZZZ[argc];
+  return res;
+}

compiler-rt/trunk/test/asan/TestCases/Linux/globals-gc-sections-lld.cc

+// RUN: %clangxx_asan %s -o %t -Wl,--gc-sections -fuse-ld=lld -ffunction-sections -fdata-sections -mllvm -asan-globals=0
+// RUN: %clangxx_asan %s -o %t -Wl,--gc-sections -fuse-ld=lld -ffunction-sections -fdata-sections -mllvm -asan-globals=1
+
+// https://code.google.com/p/address-sanitizer/issues/detail?id=260
+// REQUIRES: lld
+
+int undefined();
+
+// On i386 clang adds --export-dynamic when linking with ASan, which adds all
+// non-hidden globals to GC roots.
+__attribute__((visibility("hidden"))) int (*unused)() = undefined;
+
+int main() {
+        return 0;
+}

compiler-rt/trunk/test/asan/TestCases/Linux/globals-gc-sections.cc

-// RUN: %clangxx_asan %s -o %t -Wl,--gc-sections -ffunction-sections -mllvm -asan-globals=0
-// RUN: %clangxx_asan %s -o %t -Wl,--gc-sections -ffunction-sections -mllvm -asan-globals=1
+// RUN: %clangxx_asan %s -o %t -Wl,--gc-sections -fuse-ld=bfd -ffunction-sections -fdata-sections -mllvm -asan-globals=0
+// RUN: %clangxx_asan %s -o %t -Wl,--gc-sections -fuse-ld=bfd -ffunction-sections -fdata-sections -mllvm -asan-globals=1
 
 // https://code.google.com/p/address-sanitizer/issues/detail?id=260
-// XFAIL: *
 
 int undefined();
 
-int (*unused)() = undefined;
+// bug in ld.bfd: with multiple "asan_globals" sections, __start_asan_globals is
+// treated as a strong GC reference to the first such section. As a result, the
+// first (for some definition of the word) global is never gc-ed.
+int first_unused = 42;
+
+// On i386 clang adds --export-dynamic when linking with ASan, which adds all
+// non-hidden globals to GC roots.
+__attribute__((visibility("hidden"))) int (*unused)() = undefined;
 
 int main() {
         return 0;
 }