backport hmac.compare_digest to partially implement PEP 466 (closes #21306)

Backport from Alex Gaynor.

Doc/library/hmac.rst

@@ -38,6 +38,13 @@ An HMAC object has the following methods:
    This string will be the same length as the *digest_size* of the digest given to
    the constructor.  It may contain non-ASCII characters, including NUL bytes.
 
+   .. warning::
+
+      When comparing the output of :meth:`digest` to an externally-supplied
+      digest during a verification routine, it is recommended to use the
+      :func:`compare_digest` function instead of the ``==`` operator
+      to reduce the vulnerability to timing attacks.
+
 
 .. method:: HMAC.hexdigest()
 
@@ -45,6 +52,13 @@ An HMAC object has the following methods:
    containing only hexadecimal digits.  This may be used to exchange the value
    safely in email or other non-binary environments.
 
+   .. warning::
+
+      When comparing the output of :meth:`hexdigest` to an externally-supplied
+      digest during a verification routine, it is recommended to use the
+      :func:`compare_digest` function instead of the ``==`` operator
+      to reduce the vulnerability to timing attacks.
+
 
 .. method:: HMAC.copy()
 
@@ -52,6 +66,25 @@ An HMAC object has the following methods:
    compute the digests of strings that share a common initial substring.
 
 
+This module also provides the following helper function:
+
+.. function:: compare_digest(a, b)
+
+   Return ``a == b``.  This function uses an approach designed to prevent
+   timing analysis by avoiding content-based short circuiting behaviour,
+   making it appropriate for cryptography.  *a* and *b* must both be of the
+   same type: either :class:`unicode` or a :term:`bytes-like object`.
+
+   .. note::
+
+      If *a* and *b* are of different lengths, or if an error occurs,
+      a timing attack could theoretically reveal information about the
+      types and lengths of *a* and *b*--but not their values.
+
+
+   .. versionadded:: 2.7.7
+
+
 .. seealso::
 
    Module :mod:`hashlib`

Lib/hmac.py

@@ -5,6 +5,9 @@ Implements the HMAC algorithm as described by RFC 2104.
 
 import warnings as _warnings
 
+from operator import _compare_digest as compare_digest
+
+
 trans_5C = "".join ([chr (x ^ 0x5C) for x in xrange(256)])
 trans_36 = "".join ([chr (x ^ 0x36) for x in xrange(256)])
 

Lib/test/test_hmac.py

@@ -302,12 +302,122 @@ class CopyTestCase(unittest.TestCase):
         self.assertTrue(h1.hexdigest() == h2.hexdigest(),
             "Hexdigest of copy doesn't match original hexdigest.")
 
+
+class CompareDigestTestCase(unittest.TestCase):
+
+    def test_compare_digest(self):
+        # Testing input type exception handling
+        a, b = 100, 200
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = 100, b"foobar"
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = b"foobar", 200
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = u"foobar", b"foobar"
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = b"foobar", u"foobar"
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+
+        # Testing bytes of different lengths
+        a, b = b"foobar", b"foo"
+        self.assertFalse(hmac.compare_digest(a, b))
+        a, b = b"\xde\xad\xbe\xef", b"\xde\xad"
+        self.assertFalse(hmac.compare_digest(a, b))
+
+        # Testing bytes of same lengths, different values
+        a, b = b"foobar", b"foobaz"
+        self.assertFalse(hmac.compare_digest(a, b))
+        a, b = b"\xde\xad\xbe\xef", b"\xab\xad\x1d\xea"
+        self.assertFalse(hmac.compare_digest(a, b))
+
+        # Testing bytes of same lengths, same values
+        a, b = b"foobar", b"foobar"
+        self.assertTrue(hmac.compare_digest(a, b))
+        a, b = b"\xde\xad\xbe\xef", b"\xde\xad\xbe\xef"
+        self.assertTrue(hmac.compare_digest(a, b))
+
+        # Testing bytearrays of same lengths, same values
+        a, b = bytearray(b"foobar"), bytearray(b"foobar")
+        self.assertTrue(hmac.compare_digest(a, b))
+
+        # Testing bytearrays of diffeent lengths
+        a, b = bytearray(b"foobar"), bytearray(b"foo")
+        self.assertFalse(hmac.compare_digest(a, b))
+
+        # Testing bytearrays of same lengths, different values
+        a, b = bytearray(b"foobar"), bytearray(b"foobaz")
+        self.assertFalse(hmac.compare_digest(a, b))
+
+        # Testing byte and bytearray of same lengths, same values
+        a, b = bytearray(b"foobar"), b"foobar"
+        self.assertTrue(hmac.compare_digest(a, b))
+        self.assertTrue(hmac.compare_digest(b, a))
+
+        # Testing byte bytearray of diffeent lengths
+        a, b = bytearray(b"foobar"), b"foo"
+        self.assertFalse(hmac.compare_digest(a, b))
+        self.assertFalse(hmac.compare_digest(b, a))
+
+        # Testing byte and bytearray of same lengths, different values
+        a, b = bytearray(b"foobar"), b"foobaz"
+        self.assertFalse(hmac.compare_digest(a, b))
+        self.assertFalse(hmac.compare_digest(b, a))
+
+        # Testing str of same lengths
+        a, b = "foobar", "foobar"
+        self.assertTrue(hmac.compare_digest(a, b))
+
+        # Testing str of diffeent lengths
+        a, b = "foo", "foobar"
+        self.assertFalse(hmac.compare_digest(a, b))
+
+        # Testing bytes of same lengths, different values
+        a, b = "foobar", "foobaz"
+        self.assertFalse(hmac.compare_digest(a, b))
+
+        # Testing error cases
+        a, b = u"foobar", b"foobar"
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = b"foobar", u"foobar"
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = b"foobar", 1
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = 100, 200
+        self.assertRaises(TypeError, hmac.compare_digest, a, b)
+        a, b = "fooä", "fooä"
+        self.assertTrue(hmac.compare_digest(a, b))
+
+        # subclasses are supported by ignore __eq__
+        class mystr(str):
+            def __eq__(self, other):
+                return False
+
+        a, b = mystr("foobar"), mystr("foobar")
+        self.assertTrue(hmac.compare_digest(a, b))
+        a, b = mystr("foobar"), "foobar"
+        self.assertTrue(hmac.compare_digest(a, b))
+        a, b = mystr("foobar"), mystr("foobaz")
+        self.assertFalse(hmac.compare_digest(a, b))
+
+        class mybytes(bytes):
+            def __eq__(self, other):
+                return False
+
+        a, b = mybytes(b"foobar"), mybytes(b"foobar")
+        self.assertTrue(hmac.compare_digest(a, b))
+        a, b = mybytes(b"foobar"), b"foobar"
+        self.assertTrue(hmac.compare_digest(a, b))
+        a, b = mybytes(b"foobar"), mybytes(b"foobaz")
+        self.assertFalse(hmac.compare_digest(a, b))
+
+
 def test_main():
     test_support.run_unittest(
         TestVectorsTestCase,
         ConstructorTestCase,
         SanityTestCase,
-        CopyTestCase
+        CopyTestCase,
+        CompareDigestTestCase,
     )
 
 if __name__ == "__main__":

Misc/NEWS

@@ -49,6 +49,9 @@ Core and Builtins
 Library
 -------
 
+- Issue #21306: Backport hmac.compare_digest from Python 3. This is part of PEP
+  466.
+
 - Issue #21321: itertools.islice() now releases the reference to the source
   iterator when the slice is exhausted.  Patch by Anton Afanasyev.
 

Modules/operator.c

@@ -235,6 +235,132 @@ op_delslice(PyObject *s, PyObject *a)
 #define spam2o(OP,ALTOP,DOC) {#OP, op_##OP, METH_O, PyDoc_STR(DOC)}, \
                            {#ALTOP, op_##OP, METH_O, PyDoc_STR(DOC)},
 
+
+
+/* compare_digest **********************************************************/
+
+/*
+ * timing safe compare
+ *
+ * Returns 1 of the strings are equal.
+ * In case of len(a) != len(b) the function tries to keep the timing
+ * dependent on the length of b. CPU cache locally may still alter timing
+ * a bit.
+ */
+static int
+_tscmp(const unsigned char *a, const unsigned char *b,
+        Py_ssize_t len_a, Py_ssize_t len_b)
+{
+    /* The volatile type declarations make sure that the compiler has no
+     * chance to optimize and fold the code in any way that may change
+     * the timing.
+     */
+    volatile Py_ssize_t length;
+    volatile const unsigned char *left;
+    volatile const unsigned char *right;
+    Py_ssize_t i;
+    unsigned char result;
+
+    /* loop count depends on length of b */
+    length = len_b;
+    left = NULL;
+    right = b;
+
+    /* don't use else here to keep the amount of CPU instructions constant,
+     * volatile forces re-evaluation
+     *  */
+    if (len_a == length) {
+        left = *((volatile const unsigned char**)&a);
+        result = 0;
+    }
+    if (len_a != length) {
+        left = b;
+        result = 1;
+    }
+
+    for (i=0; i < length; i++) {
+        result |= *left++ ^ *right++;
+    }
+
+    return (result == 0);
+}
+
+PyDoc_STRVAR(compare_digest__doc__,
+"compare_digest(a, b) -> bool\n"
+"\n"
+"Return 'a == b'.  This function uses an approach designed to prevent\n"
+"timing analysis, making it appropriate for cryptography.\n"
+"a and b must both be of the same type: either str (ASCII only),\n"
+"or any type that supports the buffer protocol (e.g. bytes).\n"
+"\n"
+"Note: If a and b are of different lengths, or if an error occurs,\n"
+"a timing attack could theoretically reveal information about the\n"
+"types and lengths of a and b--but not their values.\n");
+
+static PyObject*
+compare_digest(PyObject *self, PyObject *args)
+{
+    PyObject *a, *b;
+    int rc;
+
+    if (!PyArg_ParseTuple(args, "OO:compare_digest", &a, &b)) {
+        return NULL;
+    }
+
+    /* Unicode string */
+    if (PyUnicode_Check(a) && PyUnicode_Check(b)) {
+        rc = _tscmp(PyUnicode_AS_DATA(a),
+                    PyUnicode_AS_DATA(b),
+                    PyUnicode_GET_DATA_SIZE(a),
+                    PyUnicode_GET_DATA_SIZE(b));
+    }
+    /* fallback to buffer interface for bytes, bytesarray and other */
+    else {
+        Py_buffer view_a;
+        Py_buffer view_b;
+
+        if ((PyObject_CheckBuffer(a) == 0) & (PyObject_CheckBuffer(b) == 0)) {
+            PyErr_Format(PyExc_TypeError,
+                         "unsupported operand types(s) or combination of types: "
+                         "'%.100s' and '%.100s'",
+                         Py_TYPE(a)->tp_name, Py_TYPE(b)->tp_name);
+            return NULL;
+        }
+
+        if (PyObject_GetBuffer(a, &view_a, PyBUF_SIMPLE) == -1) {
+            return NULL;
+        }
+        if (view_a.ndim > 1) {
+            PyErr_SetString(PyExc_BufferError,
+                            "Buffer must be single dimension");
+            PyBuffer_Release(&view_a);
+            return NULL;
+        }
+
+        if (PyObject_GetBuffer(b, &view_b, PyBUF_SIMPLE) == -1) {
+            PyBuffer_Release(&view_a);
+            return NULL;
+        }
+        if (view_b.ndim > 1) {
+            PyErr_SetString(PyExc_BufferError,
+                            "Buffer must be single dimension");
+            PyBuffer_Release(&view_a);
+            PyBuffer_Release(&view_b);
+            return NULL;
+        }
+
+        rc = _tscmp((const unsigned char*)view_a.buf,
+                    (const unsigned char*)view_b.buf,
+                    view_a.len,
+                    view_b.len);
+
+        PyBuffer_Release(&view_a);
+        PyBuffer_Release(&view_b);
+    }
+
+    return PyBool_FromLong(rc);
+}
+
 static struct PyMethodDef operator_methods[] = {
 
 spam1o(isCallable,
@@ -318,6 +444,8 @@ spam2(ne,__ne__, "ne(a, b) -- Same as a!=b.")
 spam2(gt,__gt__, "gt(a, b) -- Same as a>b.")
 spam2(ge,__ge__, "ge(a, b) -- Same as a>=b.")
 
+    {"_compare_digest", (PyCFunction)compare_digest, METH_VARARGS,
+     compare_digest__doc__},
     {NULL,              NULL}           /* sentinel */
 
 };