[3.6] bpo-35214: Initial clang MemorySanitizer support (GH-10479) (GH-10493)

Adds configure flags for msan and ubsan builds to make it easier to enable.
These also encode the detail that address sanitizer and memory sanitizer
should disable pymalloc.

Define MEMORY_SANITIZER when appropriate at build time and adds workarounds
to existing code to mark things as initialized where the sanitizer is otherwise unable to
determine that.  This lets our build succeed under the memory sanitizer.  not all tests
pass without sanitizer failures yet but we're in pretty good shape after this.

(cherry picked from commit 1584a00)

Contributed by Gregory P. Smith [Google LLC]

Also includes a whitespace fix from make patchcheck to _posixsubprocess.c - unrelated to the main change that makes the CI happy so I'm just doing it now rather than creating a separate PR.

Author: gpshead

Include/Python.h

@@ -53,6 +53,15 @@
 #include "pyport.h"
 #include "pymacro.h"
 
+/* A convenient way for code to know if clang's memory sanitizer is enabled. */
+#if defined(__has_feature)
+#  if __has_feature(memory_sanitizer)
+#    if !defined(MEMORY_SANITIZER)
+#      define MEMORY_SANITIZER
+#    endif
+#  endif
+#endif
+
 #include "pyatomic.h"
 
 /* Debug-mode build with pymalloc implies PYMALLOC_DEBUG.

Misc/NEWS.d/next/Core and Builtins/2018-11-12-11-38-06.bpo-35214.PCHKbX.rst

@@ -0,0 +1,4 @@
+The interpreter and extension modules have had annotations added so that
+they work properly under clang's Memory Sanitizer.  A new configure flag
+--with-memory-sanitizer has been added to make test builds of this nature
+easier to perform.

Modules/_ctypes/callproc.c

@@ -75,6 +75,10 @@
 #include <alloca.h>
 #endif
 
+#ifdef MEMORY_SANITIZER
+#include <sanitizer/msan_interface.h>
+#endif
+
 #if defined(_DEBUG) || defined(__MINGW32__)
 /* Don't use structured exception handling on Windows if this is defined.
    MingW, AFAIK, doesn't support it.
@@ -1136,6 +1140,13 @@ PyObject *_ctypes_callproc(PPROC pProc,
     rtype = _ctypes_get_ffi_type(restype);
     resbuf = alloca(max(rtype->size, sizeof(ffi_arg)));
 
+#ifdef MEMORY_SANITIZER
+    /* ffi_call actually initializes resbuf, but from asm, which
+     * MemorySanitizer can't detect. Avoid false positives from MSan. */
+    if (resbuf != NULL) {
+        __msan_unpoison(resbuf, max(rtype->size, sizeof(ffi_arg)));
+    }
+#endif
     avalues = (void **)alloca(sizeof(void *) * argcount);
     atypes = (ffi_type **)alloca(sizeof(ffi_type *) * argcount);
     if (!resbuf || !avalues || !atypes) {

Modules/_posixsubprocess.c

@@ -21,6 +21,10 @@
 #include <dirent.h>
 #endif
 
+#ifdef MEMORY_SANITIZER
+# include <sanitizer/msan_interface.h>
+#endif
+
 #if defined(__ANDROID__) && __ANDROID_API__ < 21 && !defined(SYS_getdents64)
 # include <sys/linux-syscalls.h>
 # define SYS_getdents64  __NR_getdents64
@@ -287,6 +291,9 @@ _close_open_fds_safe(int start_fd, PyObject* py_fds_to_keep)
                                 sizeof(buffer))) > 0) {
             struct linux_dirent64 *entry;
             int offset;
+#ifdef MEMORY_SANITIZER
+            __msan_unpoison(buffer, bytes);
+#endif
             for (offset = 0; offset < bytes; offset += entry->d_reclen) {
                 int fd;
                 entry = (struct linux_dirent64 *)(buffer + offset);
@@ -791,11 +798,11 @@ static PyMethodDef module_methods[] = {
 
 
 static struct PyModuleDef _posixsubprocessmodule = {
-	PyModuleDef_HEAD_INIT,
-	"_posixsubprocess",
-	module_doc,
-	-1,  /* No memory is needed. */
-	module_methods,
+        PyModuleDef_HEAD_INIT,
+        "_posixsubprocess",
+        module_doc,
+        -1,  /* No memory is needed. */
+        module_methods,
 };
 
 PyMODINIT_FUNC

Modules/faulthandler.c

@@ -1393,7 +1393,7 @@ void _PyFaulthandler_Fini(void)
 #ifdef HAVE_SIGALTSTACK
     if (stack.ss_sp != NULL) {
         /* Fetch the current alt stack */
-        stack_t current_stack;
+        stack_t current_stack = {};
         if (sigaltstack(NULL, &current_stack) == 0) {
             if (current_stack.ss_sp == stack.ss_sp) {
                 /* The current alt stack is the one that we installed.

Python/pymath.c

@@ -17,7 +17,9 @@ double _Py_force_double(double x)
 
 /* inline assembly for getting and setting the 387 FPU control word on
    gcc/x86 */
-
+#ifdef MEMORY_SANITIZER
+__attribute__((no_sanitize_memory))
+#endif
 unsigned short _Py_get_387controlword(void) {
     unsigned short cw;
     __asm__ __volatile__ ("fnstcw %0" : "=m" (cw));

Python/random.c

@@ -20,6 +20,10 @@
 #  endif
 #endif
 
+#ifdef MEMORY_SANITIZER
+#  include <sanitizer/msan_interface.h>
+#endif
+
 #ifdef Py_DEBUG
 int _Py_HashSecret_Initialized = 0;
 #else
@@ -143,6 +147,11 @@ py_getrandom(void *buffer, Py_ssize_t size, int blocking, int raise)
         else {
             n = syscall(SYS_getrandom, dest, n, flags);
         }
+#  ifdef MEMORY_SANITIZER
+        if (n > 0) {
+             __msan_unpoison(dest, n);
+        }
+#  endif
 #endif
 
         if (n < 0) {

configure

@@ -822,6 +822,8 @@ enable_optimizations
 with_lto
 with_hash_algorithm
 with_address_sanitizer
+with_memory_sanitizer
+with_undefined_behavior_sanitizer
 with_libs
 with_system_expat
 with_system_ffi
@@ -1520,7 +1522,10 @@ Optional Packages:
   --with-hash-algorithm=[fnv|siphash24]
                           select hash algorithm
   --with-address-sanitizer
-                          enable AddressSanitizer
+                          enable AddressSanitizer (asan)
+  --with-memory-sanitizer enable MemorySanitizer (msan)
+  --with-undefined-behavior-sanitizer
+                          enable UndefinedBehaviorSanitizer (ubsan)
   --with-libs='lib1 ...'  link against additional libs
   --with-system-expat     build pyexpat module using an installed expat
                           library
@@ -9926,6 +9931,44 @@ if test "${with_address_sanitizer+set}" = set; then :
 $as_echo "$withval" >&6; }
 BASECFLAGS="-fsanitize=address -fno-omit-frame-pointer $BASECFLAGS"
 LDFLAGS="-fsanitize=address $LDFLAGS"
+# ASan works by controlling memory allocation, our own malloc interferes.
+with_pymalloc="no"
+
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for --with-memory-sanitizer" >&5
+$as_echo_n "checking for --with-memory-sanitizer... " >&6; }
+
+# Check whether --with-memory_sanitizer was given.
+if test "${with_memory_sanitizer+set}" = set; then :
+  withval=$with_memory_sanitizer;
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $withval" >&5
+$as_echo "$withval" >&6; }
+BASECFLAGS="-fsanitize=memory -fsanitize-memory-track-origins=2 -fno-omit-frame-pointer $BASECFLAGS"
+LDFLAGS="-fsanitize=memory -fsanitize-memory-track-origins=2 $LDFLAGS"
+# MSan works by controlling memory allocation, our own malloc interferes.
+with_pymalloc="no"
+
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+fi
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for --with-undefined-behavior-sanitizer" >&5
+$as_echo_n "checking for --with-undefined-behavior-sanitizer... " >&6; }
+
+# Check whether --with-undefined_behavior_sanitizer was given.
+if test "${with_undefined_behavior_sanitizer+set}" = set; then :
+  withval=$with_undefined_behavior_sanitizer;
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $withval" >&5
+$as_echo "$withval" >&6; }
+BASECFLAGS="-fsanitize=undefined $BASECFLAGS"
+LDFLAGS="-fsanitize=undefined $LDFLAGS"
 
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5

configure.ac

@@ -2833,11 +2833,37 @@ esac
 AC_MSG_CHECKING(for --with-address-sanitizer)
 AC_ARG_WITH(address_sanitizer,
             AS_HELP_STRING([--with-address-sanitizer],
-                           [enable AddressSanitizer]),
+                           [enable AddressSanitizer (asan)]),
 [
 AC_MSG_RESULT($withval)
 BASECFLAGS="-fsanitize=address -fno-omit-frame-pointer $BASECFLAGS"
 LDFLAGS="-fsanitize=address $LDFLAGS"
+# ASan works by controlling memory allocation, our own malloc interferes.
+with_pymalloc="no"
+],
+[AC_MSG_RESULT(no)])
+
+AC_MSG_CHECKING(for --with-memory-sanitizer)
+AC_ARG_WITH(memory_sanitizer,
+            AS_HELP_STRING([--with-memory-sanitizer],
+                           [enable MemorySanitizer (msan)]),
+[
+AC_MSG_RESULT($withval)
+BASECFLAGS="-fsanitize=memory -fsanitize-memory-track-origins=2 -fno-omit-frame-pointer $BASECFLAGS"
+LDFLAGS="-fsanitize=memory -fsanitize-memory-track-origins=2 $LDFLAGS"
+# MSan works by controlling memory allocation, our own malloc interferes.
+with_pymalloc="no"
+],
+[AC_MSG_RESULT(no)])
+
+AC_MSG_CHECKING(for --with-undefined-behavior-sanitizer)
+AC_ARG_WITH(undefined_behavior_sanitizer,
+            AS_HELP_STRING([--with-undefined-behavior-sanitizer],
+                           [enable UndefinedBehaviorSanitizer (ubsan)]),
+[
+AC_MSG_RESULT($withval)
+BASECFLAGS="-fsanitize=undefined $BASECFLAGS"
+LDFLAGS="-fsanitize=undefined $LDFLAGS"
 ],
 [AC_MSG_RESULT(no)])