[sanitizers] __sanitizer_get_module_and_offset_for_pc interface function

[aizatsky]

Summary

The function computes full module name and coverts pc into offset.

[aizatsky]

clang-format

[kcc]

lib/sanitizer_common/sanitizer_common_libcdep.cc

174

Hmmmm? Leftover from earlier experiments?

183

does this need to be null-terminated?

test/sanitizer_common/TestCases/get_module_and_offset_for_pc.cc

25

please add a loop which will call something like TestCallerPc w/o logging in a long loop,
to ensure that we are not insanely slow.
Ensure that this test really runs slow on the current implementation (with the leftover ListOfModules)

I would also add an extra DSO to the test.
(e.g. like in test/asan/TestCases/Linux/stack-trace-dlclose.cc)

[aizatsky]

review

[aizatsky]

ptal

lib/sanitizer_common/sanitizer_common_libcdep.cc

183

why not? null-terminated strings are easier to work with.

[kcc]

lib/sanitizer_common/sanitizer_common_libcdep.cc

183

probably I badly phrased the question.
Does this code guarantee that the module_name will be null terminated?

lib/tsan/go/buildgo.sh

28 ↗

(On Diff #80121)

Ar the changes in this file related to this patch?

[aizatsky]

lib/sanitizer_common/sanitizer_common_libcdep.cc

183

yes. that's how it is stored in LoadedModule.

lib/tsan/go/buildgo.sh

28 ↗

(On Diff #80121)

yes. sanitizer_common_libcdep.cc now depends on symbolizer.

[kcc]

lib/sanitizer_common/sanitizer_common_libcdep.cc

183

But what if module_name_len is less than the length of the module string stored in LoadedModule?

[aizatsky]

forcing 0-termination

[aizatsky]

ptal

lib/sanitizer_common/sanitizer_common_libcdep.cc

183

Got it. Sorry, a bit slow.

[kcc]

I am still very much confused on why we have to change build scripts, including things that touch Go and FreeBSD. Something is wrong here (or I can't get it).
Also, __sanitizer_symbolize_pc is defined in sanitizer_stacktrace_libcdep.cc
Wouldn't it be better to implement the new API there as well?

[aizatsky]

In D26820#612483, @kcc wrote:

I am still very much confused on why we have to change build scripts, including things that touch Go and FreeBSD. Something is wrong here (or I can't get it).

This is what happens without this change on check-all: https://gist.github.com/mikea/53f4dc2c43bab95e97e442be44d5cd72

Also, __sanitizer_symbolize_pc is defined in sanitizer_stacktrace_libcdep.cc
Wouldn't it be better to implement the new API there as well?

I find _stacktrace_libcdep.cc to be unexpected location for symbolization implementation, isn't it?

[kcc]

Also, __sanitizer_symbolize_pc is defined in sanitizer_stacktrace_libcdep.cc
Wouldn't it be better to implement the new API there as well?

I find _stacktrace_libcdep.cc to be unexpected location for symbolization implementation, isn't it?

stacktrace_libcdep deals with symbolizing stack traces, so an API about symbolizing makes sense there.
And I bet if you move your function there you won't have to change any build files.

[aizatsky]

In D26820#612518, @kcc wrote:

Also, __sanitizer_symbolize_pc is defined in sanitizer_stacktrace_libcdep.cc
Wouldn't it be better to implement the new API there as well?

I find _stacktrace_libcdep.cc to be unexpected location for symbolization implementation, isn't it?

stacktrace_libcdep deals with symbolizing stack traces, so an API about symbolizing makes sense there.
And I bet if you move your function there you won't have to change any build files.

Moved. It is accidental that it works: go needs a subset of runtime but doesn't actually specify the surface of that runtime. You will get into go runtime by merely putting your code into sanitizer_common_libcdep.cc.

[aizatsky]

moved code

[kcc]

LGTM with one nit.

lib/sanitizer_common/sanitizer_stacktrace_libcdep.cc

138 ↗

(On Diff #80166)

replace with uptr pc, remove NOLINT, remove the cast below.
Just like in __sanitizer_symbolize_pc.

[aizatsky]

using uptr

lib/sanitizer_common/sanitizer_stacktrace_libcdep.cc

138 ↗

(On Diff #80166)

Done except NOLINT.
You need it otherwise clang-tidy complains:

/usr/local/google/home/aizatsky/src/llvm/projects/compiler-rt/lib/sanitizer_common/sanitizer_stacktrace_libcdep.cc:136:35: warning: invalid case style for function 'sanitizer_get_module_and_offset_for_pc' [readability-identifier-naming]
SANITIZER_INTERFACE_ATTRIBUTE int sanitizer_get_module_and_offset_for_pc(

^
SanitizerGetModuleAndOffsetForPc

[kcc]

Maybe don't use clang-tidy or tune it to accept this naming?
Anyway, up to you.

include/sanitizer/common_interface_defs.h

#endif
   // Note that the fake stack mechanism is disabled during fiber switch, so if a
   // signal callback runs during the switch, it will not benefit from the stack
   // use-after-return detection.
   void __sanitizer_start_switch_fiber(void **fake_stack_save,
                                       const void *bottom, size_t size);
   void __sanitizer_finish_switch_fiber(void *fake_stack_save,
                                        const void **bottom_old,
                                        size_t *size_old);
+
+  // Get full module name and calculate pc offset within it.
+  // Returns 1 if pc belongs to some module, 0 if module was not found.
+  int __sanitizer_get_module_and_offset_for_pc(void *pc, char *module_path,
+                                               size_t module_path_len,
+                                               void **pc_offset);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
 
 #endif  // SANITIZER_COMMON_INTERFACE_DEFS_H

lib/asan/asan_win_dll_thunk.cc

 INTERFACE_FUNCTION(__sanitizer_unaligned_load64)
 INTERFACE_FUNCTION(__sanitizer_unaligned_store16)
 INTERFACE_FUNCTION(__sanitizer_unaligned_store32)
 INTERFACE_FUNCTION(__sanitizer_unaligned_store64)
 INTERFACE_FUNCTION(__sanitizer_verify_contiguous_container)
 INTERFACE_FUNCTION(__sanitizer_install_malloc_and_free_hooks)
 INTERFACE_FUNCTION(__sanitizer_start_switch_fiber)
 INTERFACE_FUNCTION(__sanitizer_finish_switch_fiber)
+INTERFACE_FUNCTION(__sanitizer_get_module_and_offset_for_pc)
 
 // TODO(timurrrr): Add more interface functions on the as-needed basis.
 
 // ----------------- Memory allocation functions ---------------------
 WRAP_V_W(free)
 WRAP_V_W(_free_base)
 WRAP_V_WW(_free_dbg)
 

lib/sanitizer_common/sanitizer_common_libcdep.cc

if (!common_flags()->hard_rss_limit_mb &&
       !common_flags()->soft_rss_limit_mb &&
       !common_flags()->allocator_release_to_os &&
       !common_flags()->heap_profile) return;
   if (!&real_pthread_create) return;  // Can't spawn the thread anyway.
   internal_start_thread(BackgroundThread, nullptr);
 #endif
 }
 
+static int GetModuleAndOffsetForPc(uptr pc, char *module_name,
+                                   uptr module_name_len, uptr *pc_offset) {
+  ListOfModules modules;
+  modules.init();
+
+  const char *found_module_name = nullptr;
+  bool ok = Symbolizer::GetOrInit()->GetModuleNameAndOffsetForPC(
+      pc, &found_module_name, pc_offset);
+
+  if (!ok) return false;
+
+  internal_strncpy(module_name, found_module_name, module_name_len);
+  return true;
+}
+
 }  // namespace __sanitizer
 
 void NOINLINE
 __sanitizer_sandbox_on_notify(__sanitizer_sandbox_arguments *args) {
   __sanitizer::PrepareForSandboxing(args);
   if (__sanitizer::sandboxing_callback)
     __sanitizer::sandboxing_callback();
 }
+
+extern "C" SANITIZER_INTERFACE_ATTRIBUTE int
+__sanitizer_get_module_and_offset_for_pc(void *pc,  // NOLINT
+                                         char *module_name,
+                                         uptr module_name_len,
+                                         void **pc_offset) {
+  return __sanitizer::GetModuleAndOffsetForPc(
+      reinterpret_cast<uptr>(pc), module_name, module_name_len,
+      reinterpret_cast<uptr *>(pc_offset));
+}

test/sanitizer_common/TestCases/get_module_and_offset_for_pc.cc

+// RUN: %clangxx -v -O0 %s -o %t
+// RUN: %run %t 2>&1 | FileCheck %s
+// UNSUPPORTED: i386-darwin
+//
+// Tests __sanitizer_get_module_and_offset_for_pc.
+#include <sanitizer/common_interface_defs.h>
+#include <stdio.h>
+
+void Test(void *pc) {
+  char module_name[1024];
+  void *offset;
+  int ok = __sanitizer_get_module_and_offset_for_pc(
+      pc, module_name, sizeof(module_name), &offset);
+  if (!ok) {
+    printf("NOT FOUND: %p\n", pc);
+  } else {
+    printf("FOUND: %s %p\n", module_name, offset);
+  }
+}
+
+void TestCallerPc() { Test(__builtin_return_address(0)); }
+
+int main() {
+  Test(0);
+  TestCallerPc();
+}
+// CHECK: NOT FOUND: (nil)
+// CHECK-NEXT: FOUND: {{.*}}/get_module_and_offset_for_pc.cc.tmp