Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

fix(files): normalize remote file Content-Type media types (#44677)

Winoren5f5f40afc235
diff --git a/api/core/file/remote_file_metadata.py b/api/core/file/remote_file_metadata.py
index d4d14860fcc..517a0f36d95 100644
--- a/api/core/file/remote_file_metadata.py
+++ b/api/core/file/remote_file_metadata.py
@@ -9,6 +9,8 @@ from uuid import uuid4
import httpx
+from libs.http_headers import content_type_media_type
+
try:
import magic
except ImportError:
@@ -67,7 +69,7 @@ def guess_file_info_from_response(response: httpx.Response) -> FileInfo:
mimetype, _ = mimetypes.guess_type(url)
if mimetype is None:
# If guessing fails, use Content-Type from response headers
- mimetype = response.headers.get("Content-Type", "application/octet-stream")
+ mimetype = content_type_media_type(response.headers.get("Content-Type", "application/octet-stream"))
# Use python-magic to guess MIME type if still unknown or generic
if mimetype == "application/octet-stream" and magic is not None:
diff --git a/api/libs/http_headers.py b/api/libs/http_headers.py
new file mode 100644
index 00000000000..35095e1dbee
--- /dev/null
+++ b/api/libs/http_headers.py
@@ -0,0 +1,9 @@
+"""Business-independent parsing of HTTP header values."""
+
+from werkzeug.http import parse_options_header
+
+
+def content_type_media_type(content_type: str) -> str:
+ """Return the lowercase media type without Content-Type parameters."""
+ media_type, _ = parse_options_header(content_type)
+ return media_type.lower()
diff --git a/api/tests/unit_tests/core/file/test_remote_file_metadata.py b/api/tests/unit_tests/core/file/test_remote_file_metadata.py
index 2e9425d65dc..a3659742f63 100644
--- a/api/tests/unit_tests/core/file/test_remote_file_metadata.py
+++ b/api/tests/unit_tests/core/file/test_remote_file_metadata.py
@@ -158,6 +158,48 @@ class TestGuessFileInfoFromResponse:
assert isinstance(info, FileInfo)
[email protected]("url", ["https://example.com/download", "https://example.com/"])
+ ("content_type", "expected_mimetype", "expected_extension"),
+ [
+ ("image/png; charset=binary", "image/png", ".png"),
+ ('application/pdf; name="a;b.pdf"', "application/pdf", ".pdf"),
+ ("Text/Plain; charset=UTF-8", "text/plain", ".txt"),
+ ("application/octet-stream; version=1", "application/octet-stream", ".bin"),
+ ("image/png", "image/png", ".png"),
+ (None, "application/octet-stream", ".bin"),
+ ],
+)
+def test_header_media_type_drives_extensionless_metadata(
+ monkeypatch: pytest.MonkeyPatch,
+ url: str,
+ content_type: str | None,
+ expected_mimetype: str,
+ expected_extension: str,
+) -> None:
+ # Metadata must work without optional content sniffing, including HEAD responses.
+ monkeypatch.setattr(remote_file_metadata, "magic", None)
+ headers = {"Content-Type": content_type} if content_type is not None else {}
+ response = httpx.Response(200, request=httpx.Request("HEAD", url), headers=headers)
+
+ info = guess_file_info_from_response(response)
+
+ assert info.mimetype == expected_mimetype
+ assert info.extension == expected_extension
+ assert info.filename.endswith(expected_extension)
+
+
+def test_filename_mimetype_keeps_priority_over_parameterized_header() -> None:
+ response = make_response(
+ url="https://example.com/report.pdf",
+ headers={"Content-Type": "text/plain; charset=utf-8"},
+ )
+
+ info = guess_file_info_from_response(response)
+
+ assert (info.filename, info.extension, info.mimetype) == ("report.pdf", ".pdf", "application/pdf")
+
+
class TestMagicImportWarnings:
@pytest.mark.parametrize(
("platform_name", "expected_message"),
diff --git a/api/tests/unit_tests/services/test_remote_file_service.py b/api/tests/unit_tests/services/test_remote_file_service.py
index 0b380b3936b..d96f39df1a6 100644
--- a/api/tests/unit_tests/services/test_remote_file_service.py
+++ b/api/tests/unit_tests/services/test_remote_file_service.py
@@ -371,3 +371,37 @@ def test_upload_rejects_file_that_exceeds_size_limit(
size_check.assert_called_once_with(extension=".pdf", file_size=1024)
upload.assert_not_called()
+
+
+def test_upload_parameterized_content_type_uses_media_extension_and_mimetype(
+ remote_file_service: RemoteFileService,
+ upload: MagicMock,
+ size_check: MagicMock,
+ storage: Storage,
+ sqlite_session: Session,
+) -> None:
+ url = "https://example.com/download"
+ content = b"%PDF-1.7\nremote document"
+ head = httpx.Response(
+ 200,
+ request=httpx.Request("HEAD", url),
+ headers={"Content-Type": "application/pdf; charset=binary", "Content-Length": str(len(content))},
+ )
+ body = httpx.Response(200, request=httpx.Request("GET", url), content=content)
+ with (
+ patch("services.remote_file_service.remote_fetcher.make_request", side_effect=[head, body]),
+ patch("services.remote_file_service.file_helpers.get_signed_file_url", return_value="signed-url"),
+ ):
+ result = remote_file_service.upload_from_url(url=url, user=_account())
+
+ size_check.assert_called_once_with(extension=".pdf", file_size=len(content))
+ assert upload.call_args.kwargs["filename"] == "download.pdf"
+ assert upload.call_args.kwargs["mimetype"] == "application/pdf"
+ upload_file = sqlite_session.get(UploadFile, result.id)
+ assert upload_file is not None
+ assert (upload_file.name, upload_file.extension, upload_file.mime_type) == (
+ "download.pdf",
+ "pdf",
+ "application/pdf",
+ )
+ assert storage.load_once(upload_file.key) == content