Source code for

# Copyright 2019-2020 by Vytautas Liuolia.
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# See the License for the specific language governing permissions and
# limitations under the License.

"""Multipart form media handler."""

import cgi
import re
from urllib.parse import unquote_to_bytes

from falcon import errors
from import BaseHandler
from import BoundedStream
from falcon.util import BufferedReader
from falcon.util import misc
from falcon.util.deprecation import deprecated_args

# TODO(vytas):
#   * Better support for form-wide charset setting
#   * Clean up, simplify, and optimize BufferedReader
#   * Better documentation


_FILENAME_STAR_RFC5987 = re.compile(r"([\w-]+)'[\w]*'(.+)")

_CRLF = b'\r\n'

[docs] class MultipartParseError(errors.MediaMalformedError): """Represents a multipart form parsing error. This error may refer to a malformed or truncated form, usage of deprecated or unsupported features, or form parameters exceeding limits configured in :class:`MultipartParseOptions`. :class:`MultipartParseError` instances raised in this module always include a short human-readable description of the error. The cause of this exception, if any, is stored in the ``__cause__`` attribute using the "raise ... from" form when raising. Args: source_error (Exception): The source exception that was the cause of this one. """ # NOTE(caselit): remove the description @property in MediaMalformedError description = None @deprecated_args(allowed_positional=0) def __init__(self, description=None, **kwargs): errors.HTTPBadRequest.__init__( self, title='Malformed multipart/form-data request media', description=description, **kwargs, )
# TODO(vytas): Consider supporting -charset- stuff. # Does anyone use that (?)
[docs] class BodyPart: """Represents a body part in a multipart form. Note: :class:`BodyPart` is meant to be instantiated directly only by the :class:`MultipartFormHandler` parser. Attributes: content_type (str): Value of the Content-Type header, or the multipart form default ``text/plain`` if the header is missing. data (bytes): Property that acts as a convenience alias for :meth:`~.get_data`. .. tabs:: .. tab:: WSGI .. code:: python # Equivalent to: content = part.get_data() content = .. tab:: ASGI The ``await`` keyword must still be added when referencing the property:: # Equivalent to: content = await part.get_data() content = await name(str): The name parameter of the Content-Disposition header. The value of the "name" parameter is the original field name from the submitted HTML form. .. note:: According to `RFC 7578, section 4.2 <>`__, each part MUST include a Content-Disposition header field of type "form-data", where the name parameter is mandatory. However, Falcon will not raise any error if this parameter is missing; the property value will be ``None`` in that case. filename (str): File name if the body part is an attached file, and ``None`` otherwise. secure_filename (str): The sanitized version of `filename` using only the most common ASCII characters for maximum portability and safety wrt using this name as a filename on a regular file system. If `filename` is empty or unset when referencing this property, an instance of :class:`MultipartParseError` will be raised. See also: :func:`~.secure_filename` stream: File-like input object for reading the body part of the multipart form request, if any. This object provides direct access to the server's data stream and is non-seekable. The stream is automatically delimited according to the multipart stream boundary. With the exception of being buffered to keep track of the boundary, the wrapped body part stream interface and behavior mimic :attr:`Request.bounded_stream <falcon.Request.bounded_stream>` (WSGI) and :attr:` <>` (ASGI), respectively: .. tabs:: .. tab:: WSGI Reading the whole part content: .. code:: python data = This is also safe: .. code:: python doc = yaml.safe_load( .. tab:: ASGI Similarly to :attr:`BoundedStream <falcon.asgi.BoundedStream>`, the most efficient way to read the body part content is asynchronous iteration over part data chunks: .. code:: python async for data_chunk in pass media (object): Property that acts as a convenience alias for :meth:`~.get_media`. .. tabs:: .. tab:: WSGI .. code:: python # Equivalent to: deserialized_media = part.get_media() deserialized_media = .. tab:: ASGI The ``await`` keyword must still be added when referencing the property:: # Equivalent to: deserialized_media = await part.get_media() deserialized_media = await text (str): Property that acts as a convenience alias for :meth:`~.get_text`. .. tabs:: .. tab:: WSGI .. code:: python # Equivalent to: decoded_text = part.get_text() decoded_text = part.text .. tab:: ASGI The ``await`` keyword must still be added when referencing the property:: # Equivalent to: decoded_text = await part.get_text() decoded_text = await part.text """ _content_disposition = None _data = None _filename = None _media = None _name = None def __init__(self, stream, headers, parse_options): = stream self._headers = headers self._parse_options = parse_options
[docs] def get_data(self): """Return the body part content bytes. The maximum number of bytes that may be read is configurable via :class:`MultipartParseOptions`, and a :class:`.MultipartParseError` is raised if the body part is larger that this size. The size limit guards against reading unexpectedly large amount of data into memory by referencing :attr:`data` and :attr:`text` properties that build upon this method. For large bodies, such as attached files, use the input :attr:`stream` directly. Note: Calling this method the first time will consume the part's input stream. The result is cached for subsequent access, and follow-up calls will just retrieve the cached content. Returns: bytes: The body part content. """ if self._data is None: max_size = self._parse_options.max_body_part_buffer_size + 1 self._data = if len(self._data) >= max_size: raise MultipartParseError(description='body part is too large') return self._data
[docs] def get_text(self): """Return the body part content decoded as a text string. Text is decoded from the part content (as returned by :meth:`~.get_data`) using the charset specified in the `Content-Type` header, or, if omitted, the :data:`default charset <MultipartParseOptions.default_charset>`. The charset must be supported by Python's ``bytes.decode()`` function. The list of standard encodings (charsets) supported by the Python 3 standard library can be found `here <>`__. If decoding fails due to invalid `data` bytes (for the specified encoding), or the specified encoding itself is unsupported, a :class:`MultipartParseError` will be raised when referencing this property. Note: As this method builds upon :meth:`~.get_data`, it will consume the part's input stream in the same way. Returns: str: The part decoded as a text string provided the part is encoded as ``text/plain``, ``None`` otherwise. """ content_type, options = cgi.parse_header(self.content_type) if content_type != 'text/plain': return None charset = options.get('charset', self._parse_options.default_charset) try: return except (ValueError, LookupError) as err: raise MultipartParseError( description='invalid text or charset: {}'.format(charset) ) from err
@property def content_type(self): # NOTE(vytas): RFC 7578, section 4.4. # Each part MAY have an (optional) "Content-Type" header field, which # defaults to "text/plain". value = self._headers.get(b'content-type', b'text/plain') return value.decode('ascii') @property def filename(self): if self._filename is None: if self._content_disposition is None: value = self._headers.get(b'content-disposition', b'') self._content_disposition = cgi.parse_header(value.decode()) _, params = self._content_disposition # NOTE(vytas): Supporting filename* as per RFC 5987, as that has # been spotted in the wild, even though RFC 7578 forbids it. match = _FILENAME_STAR_RFC5987.match(params.get('filename*', '')) if match: charset, value = match.groups() try: self._filename = unquote_to_bytes(value).decode(charset) except (ValueError, LookupError) as err: raise MultipartParseError( description='invalid text or charset: {}'.format(charset) ) from err else: value = params.get('filename') if value is None: return None self._filename = value return self._filename @property def secure_filename(self): try: return misc.secure_filename(self.filename) except ValueError as ex: raise MultipartParseError(description=str(ex)) from ex @property def name(self): if self._name is None: if self._content_disposition is None: value = self._headers.get(b'content-disposition', b'') self._content_disposition = cgi.parse_header(value.decode()) _, params = self._content_disposition self._name = params.get('name') return self._name
[docs] def get_media(self): """Return a deserialized form of the multipart body part. When called, this method will attempt to deserialize the body part stream using the Content-Type header as well as the media-type handlers configured via :class:`MultipartParseOptions`. .. tabs:: .. tab:: WSGI The result will be cached and returned in subsequent calls:: deserialized_media = part.get_media() .. tab:: ASGI The result will be cached and returned in subsequent calls:: deserialized_media = await part.get_media() Returns: object: The deserialized media representation. """ if self._media is None: handler, _, _ = self._parse_options.media_handlers._resolve( self.content_type, 'text/plain' ) try: self._media = handler.deserialize(, self.content_type, None) finally: if handler.exhaust_stream: return self._media
data = property(get_data) media = property(get_media) text = property(get_text)
class MultipartForm: def __init__(self, stream, boundary, content_length, parse_options): # NOTE(vytas): More lenient check whether the provided stream is not # already an instance of BufferedReader. # This approach makes testing both the Cythonized and pure-Python # streams easier within the same test/benchmark suite. if not hasattr(stream, 'read_until'): if isinstance(stream, BoundedStream): stream = BufferedReader(, content_length) else: stream = BufferedReader(, content_length) self._stream = stream self._boundary = boundary # NOTE(vytas): Here self._dash_boundary is not prepended with CRLF # (yet) for parsing the prologue. The CRLF will be prepended later to # construct the inter-part delimiter as per RFC 7578, section 4.1 # (see the note below). self._dash_boundary = b'--' + boundary self._parse_options = parse_options def __iter__(self): prologue = True delimiter = self._dash_boundary stream = self._stream max_headers_size = self._parse_options.max_body_part_headers_size remaining_parts = self._parse_options.max_body_part_count while True: # NOTE(vytas): Either exhaust the unused stream part, or skip # the prologue. try: stream.pipe_until(delimiter, consume_delimiter=True) if prologue: # NOTE(vytas): RFC 7578, section 4.1. # As with other multipart types, the parts are delimited # with a boundary delimiter, constructed using CRLF, # "--", and the value of the "boundary" parameter. delimiter = _CRLF + delimiter prologue = False separator = stream.read_until(_CRLF, 2, consume_delimiter=True) if separator == b'--': # NOTE(vytas): boundary delimiter + '--\r\n' signals the # end of a multipart form. break elif separator: raise MultipartParseError(description='unexpected form structure') except errors.DelimiterError as err: raise MultipartParseError( description='unexpected form structure' ) from err headers = {} try: headers_block = stream.read_until( _CRLF_CRLF, max_headers_size, consume_delimiter=True ) except errors.DelimiterError as err: raise MultipartParseError( description='incomplete body part headers' ) from err for line in headers_block.split(_CRLF): name, sep, value = line.partition(b': ') if sep: name = name.lower() # NOTE(vytas): RFC 7578, section 4.5. # This use is deprecated for use in contexts that support # binary data such as HTTP. Senders SHOULD NOT generate # any parts with a Content-Transfer-Encoding header # field. # # Currently, no deployed implementations that send such # bodies have been discovered. if name == b'content-transfer-encoding' and value != b'binary': raise MultipartParseError( description=( 'the deprecated Content-Transfer-Encoding ' 'header field is unsupported' ) ) # NOTE(vytas): RFC 7578, section 4.8. # Other header fields MUST NOT be included and MUST be # ignored. elif name in _ALLOWED_CONTENT_HEADERS: headers[name] = value remaining_parts -= 1 if remaining_parts < 0 < self._parse_options.max_body_part_count: raise MultipartParseError( description='maximum number of form body parts exceeded' ) yield BodyPart(stream.delimit(delimiter), headers, self._parse_options)
[docs] class MultipartFormHandler(BaseHandler): """Multipart form (content type ``multipart/form-data``) media handler. The ``multipart/form-data`` media type for HTML5 forms is defined in `RFC 7578 <>`_. The multipart media type itself is defined in `RFC 2046 section 5.1 <>`_. .. note:: Unlike many form parsing implementations in other frameworks, this handler does not consume the stream immediately. Rather, the stream is consumed on-demand and parsed into individual body parts while iterating over the media object. For examples on parsing the request form, see also: :ref:`multipart`. Attributes: parse_options (MultipartParseOptions): Configuration options for the multipart form parser and instances of :class:`` it yields. See also: :ref:`multipart_parser_conf`. """ _ASGI_MULTIPART_FORM = None def __init__(self, parse_options=None): self.parse_options = parse_options or MultipartParseOptions() def _deserialize_form( self, stream, content_type, content_length, form_cls=MultipartForm ): if not form_cls: raise NotImplementedError _, options = cgi.parse_header(content_type) try: boundary = options['boundary'] except KeyError: raise errors.HTTPInvalidHeader( 'No boundary specifier found in {!r}'.format(content_type), 'Content-Type', ) # NOTE(vytas): RFC 2046, section 5.1. # If a boundary delimiter line appears to end with white space, the # white space must be presumed to have been added by a gateway, and # must be deleted. boundary = boundary.rstrip() # NOTE(vytas): RFC 2046, section 5.1. # The boundary parameter consists of 1 to 70 characters from a set of # characters known to be very robust through mail gateways, and NOT # ending with white space. if not 1 <= len(boundary) <= 70: raise errors.HTTPInvalidHeader( 'The boundary parameter must consist of 1 to 70 characters', 'Content-Type', ) return form_cls(stream, boundary.encode(), content_length, self.parse_options) def deserialize(self, stream, content_type, content_length): return self._deserialize_form(stream, content_type, content_length) async def deserialize_async(self, stream, content_type, content_length): return self._deserialize_form( stream, content_type, content_length, form_cls=self._ASGI_MULTIPART_FORM ) def serialize(self, media, content_type): raise NotImplementedError('multipart form serialization unsupported')
# PERF(vytas): To avoid typos and improve storage space and speed over a dict. # Inspired by RequestOptions.
[docs] class MultipartParseOptions: """Defines a set of configurable multipart form parser options. An instance of this class is exposed via the :attr:`MultipartFormHandler.parse_options <>` attribute. The handler's options are also passed down to every :class:`BodyPart` it instantiates. See also: :ref:`multipart_parser_conf`. Attributes: default_charset (str): The default character encoding for :meth:`text fields <BodyPart.get_text>` (default: ``utf-8``). max_body_part_count (int): The maximum number of body parts in the form (default: 64). If the form contains more parts than this number, an instance of :class:`MultipartParseError` will be raised. If this option is set to 0, no limit will be imposed by the parser. max_body_part_buffer_size (int): The maximum number of bytes to buffer and return when the :meth:`BodyPart.get_data` method is called (default: 1 MiB). If the body part size exceeds this value, an instance of :class:`MultipartParseError` will be raised. max_body_part_headers_size (int): The maximum size (in bytes) of the body part headers structure (default: 8192). If the body part headers size exceeds this value, an instance of :class:`MultipartParseError` will be raised. media_handlers (Handlers): A dict-like object for configuring the media-types to handle. By default, handlers are provided for the ``application/json`` and ``application/x-www-form-urlencoded`` media types. """ _DEFAULT_HANDLERS = None __slots__ = ( 'default_charset', 'max_body_part_buffer_size', 'max_body_part_count', 'max_body_part_headers_size', 'media_handlers', ) def __init__(self): self.default_charset = 'utf-8' self.max_body_part_buffer_size = 1024 * 1024 self.max_body_part_count = 64 self.max_body_part_headers_size = 8192 self.media_handlers = self._DEFAULT_HANDLERS