Add dfsan_weak_hook_memcmp

[kcc]

Summary

Add a weak hook to be called from dfsan's custom memcmp.
The primary user will be lib/Fuzzer.
If this works well we'll add more hooks (strcmp, etc).

[kcc]

pass the caller PC to the memcmp hook as the first parameter

[pcc]

LGTM

include/sanitizer/dfsan_interface.h

 void dfsan_set_write_callback(dfsan_write_callback_t labeled_write_callback);
 
 /// Writes the labels currently used by the program to the given file
 /// descriptor. The lines of the output have the following format:
 ///
 /// <label> <parent label 1> <parent label 2> <label description if any>
 void dfsan_dump_labels(int fd);
 
+/// Whenever a dfsan's custom function is called the corresponding
+/// hook is called it non-zero. The hooks should be defined by the user.
+/// The primary use case is taint-guided fuzzing, where the fuzzer
+/// needs to see the parameters of the function and the labels.
+/// FIXME: implement more hooks.
+
+/// memcmp hook.
+void dfsan_weak_hook_memcmp(void *caller_pc, const void *s1, const void *s2,
+                            size_t n, dfsan_label s1_label,
+                            dfsan_label s2_label, dfsan_label n_label);
 #ifdef __cplusplus
 }  // extern "C"
 
 template <typename T>
 void dfsan_set_label(dfsan_label label, T &data) {  // NOLINT
   dfsan_set_label(label, (void *)&data, sizeof(T));
 }
 
 #endif
 
 #endif  // DFSAN_INTERFACE_H

lib/dfsan/dfsan_custom.cc

if (s[i] == c || s[i] == 0) {
         *ret_label = dfsan_union(dfsan_read_label(s, i + 1),
                                  dfsan_union(s_label, c_label));
       }
       return s[i] == 0 ? 0 : const_cast<char *>(s+i);
     }
   }
 }
 
+SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
+void
+dfsan_weak_hook_memcmp(uptr caller_pc, const void *s1, const void *s2, size_t n,
+                       dfsan_label s1_label, dfsan_label s2_label,
+                       dfsan_label n_label);
+
 SANITIZER_INTERFACE_ATTRIBUTE int __dfsw_memcmp(const void *s1, const void *s2,
                                                 size_t n, dfsan_label s1_label,
                                                 dfsan_label s2_label,
                                                 dfsan_label n_label,
                                                 dfsan_label *ret_label) {
+  if (dfsan_weak_hook_memcmp)
+    dfsan_weak_hook_memcmp(GET_CALLER_PC(), s1, s2, n, s1_label, s2_label,
+                           n_label);
   const char *cs1 = (const char *) s1, *cs2 = (const char *) s2;
   for (size_t i = 0; i != n; ++i) {
     if (cs1[i] != cs2[i]) {
       if (flags().strict_data_dependencies) {
         *ret_label = 0;
       } else {
         *ret_label = dfsan_union(dfsan_read_label(cs1, i + 1),
                                  dfsan_read_label(cs2, i + 1));