[ubsan] Add preinit initializer for ubsan

[fjricci]

Summary

Now that ubsan does function interception (for signals), we
need to ensure that ubsan is initialized before any library
constructors are called. Otherwise, if a constructor calls
sigaction, ubsan will intercept in an unitialized state, which
will cause a crash.

This patch is a partial revert of r317757, which removed
preinit arrays for ubsan.

[fjricci]

Remove extra whitespace

[vitalybuka]

It was disabled by @pcc in D39827 because of Android issues, and now D42329 disabled signal handling on Android at all, So I guess it's safe to return preinit arrays.

lib/ubsan/CMakeLists.txt

14

This goes into both static and shared, but must go into STATIC only. Don't remember details but it will fails on Android and Apple bots.
PTAL at D39827. Maybe just restore relevant parts of D39827.

lib/ubsan/ubsan_init.h

34 ↗

(On Diff #130939)

I am not sure that we need this in API, I'd prefer D39827 approach.

[fjricci]

Great, thanks, I'll pull stuff in from that patch.

SANITIZER_CAN_USE_PREINIT_ARRAY is set to false when SANITIZER_ANDROID is defined, so we should be fine on the android front.

[fjricci]

incorporate D39827

[vitalybuka]

Oh, reverting env part of D39827 is necessary.
We can't use getenv() from preinit array.
@pcc If we revert getenv() part, we would not be able to use getenv/setenv for sanitizer options. but I guess we don't need them here as we don't need them for any other sanitizers.

[pcc]

@pcc If we revert getenv() part, we would not be able to use getenv/setenv for sanitizer options. but I guess we don't need them here as we don't need them for any other sanitizers.

No, we need it to support configuring the runtime using setenv for the reasons described in the review of D39827.

[fjricci]

No, we need it to support configuring the runtime using setenv for the reasons described in the review of D39827.

Afaik, those reasons are android app-specific. We could theoretically still use getenv() on android, since preinit arrays aren't supported there anyway.

I guess we don't need them here as we don't need them for any other sanitizers.

Are there any other sanitizers that we link in statically on android? That could be the difference?

[vitalybuka]

In D42389#984301, @fjricci wrote:

No, we need it to support configuring the runtime using setenv for the reasons described in the review of D39827.

Afaik, those reasons are android app-specific. We could theoretically still use getenv() on android, since preinit arrays aren't supported there anyway.

Could you update your patch to pick getenv vs GetEnv depending on SANITIZER_CAN_USE_PREINIT_ARRAY.

[fjricci]

Don't use getenv when using the preinit array

[vitalybuka]

lib/ubsan/ubsan_flags.cc

29

static here and above?

[fjricci]

lib/ubsan/ubsan_flags.cc

29

MaybeCallUbsanDefaultOptions is declared extern, but I'll make GetFlag static.

[fjricci]

+static

lib/ubsan/CMakeLists.txt

 # Build for the undefined behavior sanitizer runtime support library.
 
 set(UBSAN_SOURCES
   ubsan_diag.cc
   ubsan_init.cc
   ubsan_flags.cc
   ubsan_handlers.cc
   ubsan_value.cc
   )
 
 set(UBSAN_STANDALONE_SOURCES
   ubsan_diag_standalone.cc
   ubsan_init_standalone.cc
   ubsan_signals_standalone.cc
   )
 
 set(UBSAN_CXXABI_SOURCES
   ubsan_handlers_cxx.cc
   ubsan_type_hash.cc
   ubsan_type_hash_itanium.cc
   ubsan_type_hash_win.cc
   )
add_compiler_rt_object_libraries(RTUbsan_standalone
       ARCHS ${UBSAN_SUPPORTED_ARCH}
       SOURCES ${UBSAN_STANDALONE_SOURCES}
       CFLAGS ${UBSAN_STANDALONE_CFLAGS})
 
     # Standalone UBSan runtimes.
     add_compiler_rt_runtime(clang_rt.ubsan_standalone
       STATIC
       ARCHS ${UBSAN_SUPPORTED_ARCH}
+      SOURCES ubsan_init_standalone_preinit.cc
       OBJECT_LIBS RTSanitizerCommon
               RTSanitizerCommonLibc
               RTUbsan
               RTUbsan_standalone
               RTInterception
       CFLAGS ${UBSAN_CFLAGS}
       PARENT_TARGET ubsan)
 

lib/ubsan/ubsan_flags.cc

 #include <stdlib.h>
 
 namespace __ubsan {
 
 const char *MaybeCallUbsanDefaultOptions() {
   return (&__ubsan_default_options) ? __ubsan_default_options() : "";
 }
 
+static const char *GetFlag(const char *flag) {
+  // We cannot call getenv() from inside a preinit array initializer
+  if (SANITIZER_CAN_USE_PREINIT_ARRAY) {
+    return GetEnv(flag);
+  } else {
+    return getenv(flag);
+  }
+}
+
 Flags ubsan_flags;
 
 void Flags::SetDefaults() {
 #define UBSAN_FLAG(Type, Name, DefaultValue, Description) Name = DefaultValue;
 #include "ubsan_flags.inc"
 #undef UBSAN_FLAG
 }
 
 void RegisterUbsanFlags(FlagParser *parser, Flags *f) {
 #define UBSAN_FLAG(Type, Name, DefaultValue, Description) \
   RegisterFlag(parser, #Name, Description, &f->Name);
 #include "ubsan_flags.inc"
 #undef UBSAN_FLAG
 }
 
 void InitializeFlags() {
   SetCommonFlagsDefaults();
   {
     CommonFlags cf;
     cf.CopyFrom(*common_flags());
     cf.print_summary = false;
-    cf.external_symbolizer_path = getenv("UBSAN_SYMBOLIZER_PATH");
+    cf.external_symbolizer_path = GetFlag("UBSAN_SYMBOLIZER_PATH");
     OverrideCommonFlags(cf);
   }
 
   Flags *f = flags();
   f->SetDefaults();
 
   FlagParser parser;
   RegisterCommonFlags(&parser);
   RegisterUbsanFlags(&parser, f);
 
   // Override from user-specified string.
   parser.ParseString(MaybeCallUbsanDefaultOptions());
   // Override from environment variable.
-  parser.ParseString(getenv("UBSAN_OPTIONS"));
+  parser.ParseString(GetFlag("UBSAN_OPTIONS"));
   InitializeCommonFlags();
   if (Verbosity()) ReportUnrecognizedFlags();
 
   if (common_flags()->help) parser.PrintFlagDescriptions();
 }
 
 }  // namespace __ubsan
 
 SANITIZER_INTERFACE_WEAK_DEF(const char *, __ubsan_default_options, void) {
   return "";
 }
 
 #endif  // CAN_SANITIZE_UB

lib/ubsan/ubsan_init_standalone_preinit.cc

+//===-- ubsan_init_standalone_preinit.cc ---------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Initialization of standalone UBSan runtime.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ubsan_platform.h"
+#if !CAN_SANITIZE_UB
+#error "UBSan is not supported on this platform!"
+#endif
+
+#include "sanitizer_common/sanitizer_internal_defs.h"
+#include "ubsan_init.h"
+#include "ubsan_signals_standalone.h"
+
+#if SANITIZER_CAN_USE_PREINIT_ARRAY
+
+namespace __ubsan {
+
+static void PreInitAsStandalone() {
+  InitAsStandalone();
+  InitializeDeadlySignals();
+}
+
+} // namespace __ubsan
+
+__attribute__((section(".preinit_array"), used)) void (*__local_ubsan_preinit)(
+    void) = __ubsan::PreInitAsStandalone;
+#endif // SANITIZER_CAN_USE_PREINIT_ARRAY