Merge PR #100 - avoid decoding of invalid Unicode output

Use PYTHONIOENCODING to enable utf-8 stdout for the nsjail pipe,
and handle the potential case where this is bypassable.

4 files changed, 28 insertions, 2 deletions

diff --git a/config/snekbox.cfg b/config/snekbox.cfg
index 257b5ca..73e36e1 100644
--- a/config/snekbox.cfg
+++ b/config/snekbox.cfg
@@ -15,6 +15,7 @@ envar: "MKL_NUM_THREADS=1"
 envar: "VECLIB_MAXIMUM_THREADS=1"
 envar: "NUMEXPR_NUM_THREADS=1"
 envar: "PYTHONPATH=/snekbox/user_base/lib/python3.9/site-packages"
+envar: "PYTHONIOENCODING=utf-8:strict"
 
 keep_caps: false
 
diff --git a/snekbox/api/resources/eval.py b/snekbox/api/resources/eval.py
index c567660..d96b811 100644
--- a/snekbox/api/resources/eval.py
+++ b/snekbox/api/resources/eval.py
@@ -41,7 +41,7 @@ class EvalResource:
         The return codes mostly resemble those of a Unix shell. Some noteworthy cases:
 
         - None
-            The NsJail process failed to launch
+            The NsJail process failed to launch or the output was invalid Unicode
         - 137 (SIGKILL)
             Typically because NsJail killed the Python process due to time or memory constraints
         - 255
diff --git a/snekbox/nsjail.py b/snekbox/nsjail.py
index 814b46c..9367cb2 100644
--- a/snekbox/nsjail.py
+++ b/snekbox/nsjail.py
@@ -208,7 +208,15 @@ class NsJail:
             except ValueError:
                 return CompletedProcess(args, None, "ValueError: embedded null byte", None)
 
-            output = self._consume_stdout(nsjail)
+            try:
+                output = self._consume_stdout(nsjail)
+            except UnicodeDecodeError:
+                return CompletedProcess(
+                    args,
+                    None,
+                    "UnicodeDecodeError: invalid Unicode in output pipe",
+                    None,
+                )
 
             # When you send signal `N` to a subprocess to terminate it using Popen, it
             # will return `-N` as its exit code. As we normally get `N + 128` back, we
diff --git a/tests/test_nsjail.py b/tests/test_nsjail.py
index 4d4676b..46193b2 100644
--- a/tests/test_nsjail.py
+++ b/tests/test_nsjail.py
@@ -100,6 +100,23 @@ class NsJailTests(unittest.TestCase):
         self.assertEqual(result.stdout, "ValueError: embedded null byte")
         self.assertEqual(result.stderr, None)
 
+    def test_print_bad_unicode_encode_error(self):
+        result = self.nsjail.python3("print(chr(56550))")
+        self.assertEqual(result.returncode, 1)
+        self.assertIn("UnicodeEncodeError", result.stdout)
+        self.assertEqual(result.stderr, None)
+
+    def test_unicode_env_erase_escape_fails(self):
+        result = self.nsjail.python3(dedent("""
+            import os
+            import sys
+            os.unsetenv('PYTHONIOENCODING')
+            os.execl(sys.executable, 'python', '-c', 'print(chr(56550))')
+        """).strip())
+        self.assertEqual(result.returncode, None)
+        self.assertEqual(result.stdout, "UnicodeDecodeError: invalid Unicode in output pipe")
+        self.assertEqual(result.stderr, None)
+
     @unittest.mock.patch("snekbox.nsjail.DEBUG", new=False)
     def test_log_parser(self):
         log_lines = (