pythongh-106242: Fix path truncation in os.path.normpath (pythonGH-106816)

Author: finnagin

Include/internal/pycore_fileutils.h

@@ -244,7 +244,8 @@ extern int _Py_add_relfile(wchar_t *dirname,
                            const wchar_t *relfile,
                            size_t bufsize);
 extern size_t _Py_find_basename(const wchar_t *filename);
-PyAPI_FUNC(wchar_t *) _Py_normpath(wchar_t *path, Py_ssize_t size);
+PyAPI_FUNC(wchar_t*) _Py_normpath(wchar_t *path, Py_ssize_t size);
+extern wchar_t *_Py_normpath_and_size(wchar_t *path, Py_ssize_t size, Py_ssize_t *length);
 
 
 // Macros to protect CRT calls against instant termination when passed an

Lib/test/test_genericpath.py

@@ -460,6 +460,10 @@ def test_normpath_issue5827(self):
         for path in ('', '.', '/', '\\', '///foo/.//bar//'):
             self.assertIsInstance(self.pathmodule.normpath(path), str)
 
+    def test_normpath_issue106242(self):
+        for path in ('\x00', 'foo\x00bar', '\x00\x00', '\x00foo', 'foo\x00'):
+            self.assertEqual(self.pathmodule.normpath(path), path)
+
     def test_abspath_issue3426(self):
         # Check that abspath returns unicode when the arg is unicode
         # with both ASCII and non-ASCII cwds.

Misc/NEWS.d/next/Library/2023-08-14-23-11-11.gh-issue-106242.71HMym.rst

@@ -0,0 +1 @@
+Fixes :func:`os.path.normpath` to handle embedded null characters without truncating the path.

Modules/posixmodule.c

@@ -4552,7 +4552,9 @@ os__path_normpath_impl(PyObject *module, PyObject *path)
     if (!buffer) {
         return NULL;
     }
-    PyObject *result = PyUnicode_FromWideChar(_Py_normpath(buffer, len), -1);
+    Py_ssize_t norm_len;
+    wchar_t *norm_path = _Py_normpath_and_size(buffer, len, &norm_len);
+    PyObject *result = PyUnicode_FromWideChar(norm_path, norm_len);
     PyMem_Free(buffer);
     return result;
 }

Python/fileutils.c

@@ -2179,12 +2179,14 @@ _Py_find_basename(const wchar_t *filename)
    path, which will be within the original buffer. Guaranteed to not
    make the path longer, and will not fail. 'size' is the length of
    the path, if known. If -1, the first null character will be assumed
-   to be the end of the path. */
+   to be the end of the path. 'normsize' will be set to contain the
+   length of the resulting normalized path. */
 wchar_t *
-_Py_normpath(wchar_t *path, Py_ssize_t size)
+_Py_normpath_and_size(wchar_t *path, Py_ssize_t size, Py_ssize_t *normsize)
 {
     assert(path != NULL);
-    if (!path[0] || size == 0) {
+    if (!path[0] && size < 0 || size == 0) {
+        *normsize = 0;
         return path;
     }
     wchar_t *pEnd = size >= 0 ? &path[size] : NULL;
@@ -2233,11 +2235,7 @@ _Py_normpath(wchar_t *path, Py_ssize_t size)
                 *p2++ = lastC = *p1;
             }
         }
-        if (sepCount) {
-            minP2 = p2;      // Invalid path
-        } else {
-            minP2 = p2 - 1;  // Absolute path has SEP at minP2
-        }
+        minP2 = p2 - 1;
     }
 #else
     // Skip past two leading SEPs
@@ -2297,13 +2295,28 @@ _Py_normpath(wchar_t *path, Py_ssize_t size)
         while (--p2 != minP2 && *p2 == SEP) {
             *p2 = L'\0';
         }
+    } else {
+        --p2;
     }
+    *normsize = p2 - path + 1;
 #undef SEP_OR_END
 #undef IS_SEP
 #undef IS_END
     return path;
 }
 
+/* In-place path normalisation. Returns the start of the normalized
+   path, which will be within the original buffer. Guaranteed to not
+   make the path longer, and will not fail. 'size' is the length of
+   the path, if known. If -1, the first null character will be assumed
+   to be the end of the path. */
+wchar_t *
+_Py_normpath(wchar_t *path, Py_ssize_t size)
+{
+    Py_ssize_t norm_length;
+    return _Py_normpath_and_size(path, size, &norm_length);
+}
+
 
 /* Get the current directory. buflen is the buffer size in wide characters
    including the null character. Decode the path from the locale encoding.