Add in-toto format with hashes of files as subjects

This converts model serialization manifests that record every model file hash into an in-toto payload that can then be passed to Sigstore's `sign_intoto` for signing to generate a Sigstore `Bundle` (if using Sigstore). This time, we record every hash as part of the subject instead of in the payload. We require verifiers to be aware of this and acknowledge that verifiers that only check subject by subject (that is, they check if the hash of a passed in argument is in the list of subjects and don't check if all the hashes are present), can fail to fully detect if the model integrity is compromised by renaming one file in the model, interchanging two file names, or deleting a file. The signing library will have additional checks for this, but verifying the signature with other tools might result in invalid results. Signed-off-by: Mihai Maruseac <[email protected]>
sigstore · Aug 1, 2024 · 5a7aa79 · 5a7aa79
1 parent fa08711
commit 5a7aa79
Show file tree

Hide file tree

Showing 9 changed files with 408 additions and 0 deletions.
diff --git a/model_signing/signing/in_toto.py b/model_signing/signing/in_toto.py
@@ -352,3 +352,132 @@ def from_manifest(cls, manifest: manifest_module.Manifest) -> Self:
             predicate_top_level_name="shards",
         )
         return cls(statement)
+
+
+def _convert_descriptors_to_direct_statement(
+    manifest: manifest_module.Manifest, predicate_type: str
+):
+    """Converts manifest descriptors to an in-toto statement, as subjects.
+
+    Args:
+        manifest: The manifest to extract the descriptors from. Assumed valid.
+        predicate_type: The predicate_type to use in the in-toto statement.
+    """
+    subjects = []
+    for descriptor in manifest.resource_descriptors():
+        subject = statement.ResourceDescriptor(
+            name=descriptor.identifier,
+            digest={"sha256": descriptor.digest.digest_hex},
+            annotations={"actual_hash_algorithm": descriptor.digest.algorithm},
+        )
+        subjects.append(subject.pb)
+
+    return statement.Statement(
+        subjects=subjects,
+        predicate_type=predicate_type,
+        # https://github.com/in-toto/attestation/issues/374
+        predicate={"unused":"Unused, just passed due to API requirements"},
+    )
+
+
+class DigestsIntotoPayload(IntotoPayload):
+    """In-toto payload where the subjects are the model files themselves.
+
+    This payload is supposed to be used for manifests where every file in the
+    model is matched with a digest. Because existing tooling only supports
+    established hashing algorithms, we annotate every subject with the actual
+    hash algorithm used to compute the file digest, and use "sha256" as the
+    algorithm name in the digest itself.
+
+    Example:
+    ```json
+    {
+      "_type": "https://in-toto.io/Statement/v1",
+      "subject": [
+        {
+          "name": "d0/d1/d2/d3/d4/f0",
+          "digest": {
+            "sha256": "6efa14..."
+          },
+          "annotations": {
+            "actual_hash_algorithm": "file-sha256"
+          }
+        },
+        {
+          "name": "d0/d1/d2/d3/d4/f1",
+          "digest": {
+            "sha256": "a9bc14..."
+          },
+          "annotations": {
+            "actual_hash_algorithm": "file-sha256"
+          }
+        },
+        {
+          "name": "d0/d1/d2/d3/d4/f2",
+          "digest": {
+            "sha256": "5f597e..."
+          },
+          "annotations": {
+            "actual_hash_algorithm": "file-sha256"
+          }
+        },
+        {
+          "name": "d0/d1/d2/d3/d4/f3",
+          "digest": {
+            "sha256": "eaf677..."
+          },
+          "annotations": {
+            "actual_hash_algorithm": "file-sha256"
+          }
+        }
+      ],
+      "predicateType": "https://model_signing/Digests/v0.1",
+      "predicate": {
+        "unused": "Unused, just passed due to API requirements"
+      }
+    }
+    ```
+
+    If the annotation for a subject is missing, or it does not contain
+    actual_hash_algorithm, it should be assumed that the digest is computed via
+    the algorithm listed in the digest dictionary (i.e., sha256).
+
+    See also https://github.com/sigstore/sigstore-python/issues/1018.
+    """
+
+    predicate_type: Final[str] = "https://model_signing/Digests/v0.1"
+
+    def __init__(self, statement: statement.Statement):
+        """Builds an instance of this in-toto payload.
+
+        Don't call this directly in production. Use `from_manifest()` instead.
+
+        Args:
+            statement: The DSSE statement representing this in-toto payload.
+        """
+        self.statement = statement
+
+    @classmethod
+    @override
+    def from_manifest(cls, manifest: manifest_module.Manifest) -> Self:
+        """Converts a manifest to the signing payload used for signing.
+
+        The manifest must be one where every model file is paired with its own
+        digest. Currently, this is only `FileLevelManifest`.
+
+        Args:
+            manifest: the manifest to convert to signing payload.
+
+        Returns:
+            An instance of `DigestOfDigestsIntotoPayload`.
+
+        Raises:
+            TypeError: If the manifest is not `FileLevelManifest`.
+        """
+        if not isinstance(manifest, manifest_module.FileLevelManifest):
+            raise TypeError("Only FileLevelManifest is supported")
+
+        statement = _convert_descriptors_to_direct_statement(
+            manifest, predicate_type=cls.predicate_type
+        )
+        return cls(statement)
diff --git a/model_signing/signing/in_toto_test.py b/model_signing/signing/in_toto_test.py
@@ -208,3 +208,59 @@ def test_only_runs_on_expected_manifest_types(self):
             match="Only ShardLevelManifest is supported",
         ):
             in_toto.DigestOfShardDigestsIntotoPayload.from_manifest(manifest)
+
+
+class TestDigestsIntotoPayload:
+
+    def _hasher_factory(self, path: pathlib.Path) -> file.FileHasher:
+        return file.SimpleFileHasher(path, memory.SHA256())
+
+    @pytest.mark.parametrize("model_fixture_name", test_support.all_test_models)
+    def test_known_models(self, request, model_fixture_name):
+        # Set up variables (arrange)
+        testdata_path = request.path.parent / "testdata"
+        test_path = testdata_path / "in_toto"
+        test_class_path = test_path / "TestDigestsIntotoPayload"
+        golden_path = test_class_path / model_fixture_name
+        should_update = request.config.getoption("update_goldens")
+        model = request.getfixturevalue(model_fixture_name)
+
+        # Compute payload (act)
+        serializer = serialize_by_file.ManifestSerializer(
+            self._hasher_factory, allow_symlinks=True
+        )
+        manifest = serializer.serialize(model)
+        payload = in_toto.DigestsIntotoPayload.from_manifest(manifest)
+
+        # Compare with golden, or write to golden (approximately "assert")
+        if should_update:
+            with open(golden_path, "w", encoding="utf-8") as f:
+                f.write(f"{json_format.MessageToJson(payload.statement.pb)}\n")
+        else:
+            with open(golden_path, "r", encoding="utf-8") as f:
+                json_contents = f.read()
+                expected_proto = json_format.Parse(
+                    json_contents, statement_pb2.Statement()
+                )
+
+            assert payload.statement.pb == expected_proto
+
+    def test_produces_valid_statements(self, sample_model_folder):
+        serializer = serialize_by_file.ManifestSerializer(
+            self._hasher_factory, allow_symlinks=True
+        )
+        manifest = serializer.serialize(sample_model_folder)
+
+        payload = in_toto.DigestsIntotoPayload.from_manifest(manifest)
+
+        payload.statement.validate()
+
+    def test_only_runs_on_expected_manifest_types(self):
+        digest = hashing.Digest("test", b"test_digest")
+        manifest = manifest_module.DigestManifest(digest)
+
+        with pytest.raises(
+            TypeError,
+            match="Only FileLevelManifest is supported",
+        ):
+            in_toto.DigestsIntotoPayload.from_manifest(manifest)
diff --git a/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/deep_model_folder b/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/deep_model_folder
@@ -0,0 +1,45 @@
+{
+  "_type": "https://in-toto.io/Statement/v1",
+  "subject": [
+    {
+      "name": "d0/d1/d2/d3/d4/f0",
+      "digest": {
+        "sha256": "6efa14bb03544fcb76045c55f25b9315b6eb5be2d8a85f703193a76b7874c6ff"
+      },
+      "annotations": {
+        "actual_hash_algorithm": "file-sha256"
+      }
+    },
+    {
+      "name": "d0/d1/d2/d3/d4/f1",
+      "digest": {
+        "sha256": "a9bc149b70b9d325cd68d275d582cfdb98c0347d3ce54590aa6533368daed3d2"
+      },
+      "annotations": {
+        "actual_hash_algorithm": "file-sha256"
+      }
+    },
+    {
+      "name": "d0/d1/d2/d3/d4/f2",
+      "digest": {
+        "sha256": "5f597e6a92d1324d9adbed43d527926d11d0131487baf315e65ae1ef3b1ca3c0"
+      },
+      "annotations": {
+        "actual_hash_algorithm": "file-sha256"
+      }
+    },
+    {
+      "name": "d0/d1/d2/d3/d4/f3",
+      "digest": {
+        "sha256": "eaf677c35fec6b87889d9e4563d8bb65dcb9869ca0225697c9cc44cf49dca008"
+      },
+      "annotations": {
+        "actual_hash_algorithm": "file-sha256"
+      }
+    }
+  ],
+  "predicateType": "https://model_signing/Digests/v0.1",
+  "predicate": {
+    "unused": "Unused, just passed due to API requirements"
+  }
+}
diff --git a/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/empty_model_file b/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/empty_model_file
@@ -0,0 +1,18 @@
+{
+  "_type": "https://in-toto.io/Statement/v1",
+  "subject": [
+    {
+      "name": ".",
+      "digest": {
+        "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
+      },
+      "annotations": {
+        "actual_hash_algorithm": "file-sha256"
+      }
+    }
+  ],
+  "predicateType": "https://model_signing/Digests/v0.1",
+  "predicate": {
+    "unused": "Unused, just passed due to API requirements"
+  }
+}
diff --git a/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/empty_model_folder b/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/empty_model_folder
@@ -0,0 +1,7 @@
+{
+  "_type": "https://in-toto.io/Statement/v1",
+  "predicateType": "https://model_signing/Digests/v0.1",
+  "predicate": {
+    "unused": "Unused, just passed due to API requirements"
+  }
+}
diff --git a/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/model_folder_with_empty_file b/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/model_folder_with_empty_file
@@ -0,0 +1,18 @@
+{
+  "_type": "https://in-toto.io/Statement/v1",
+  "subject": [
+    {
+      "name": "empty_file",
+      "digest": {
+        "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
+      },
+      "annotations": {
+        "actual_hash_algorithm": "file-sha256"
+      }
+    }
+  ],
+  "predicateType": "https://model_signing/Digests/v0.1",
+  "predicate": {
+    "unused": "Unused, just passed due to API requirements"
+  }
+}
diff --git a/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/sample_model_file b/model_signing/signing/testdata/in_toto/TestDigestsIntotoPayload/sample_model_file
@@ -0,0 +1,18 @@
+{
+  "_type": "https://in-toto.io/Statement/v1",
+  "subject": [
+    {
+      "name": ".",
+      "digest": {
+        "sha256": "3aab065c7181a173b5dd9e9d32a9f79923440b413be1e1ffcdba26a7365f719b"
+      },
+      "annotations": {
+        "actual_hash_algorithm": "file-sha256"
+      }
+    }
+  ],
+  "predicateType": "https://model_signing/Digests/v0.1",
+  "predicate": {
+    "unused": "Unused, just passed due to API requirements"
+  }
+}