""" Multi-part parsing for file uploads. Exposes one class, ``MultiPartParser``, which feeds chunks of uploaded data to file upload handlers for processing. """ import base64 import binascii import collections import html from django.conf import settings from django.core.exceptions import ( RequestDataTooBig, SuspiciousMultipartForm, TooManyFieldsSent, TooManyFilesSent, ) from django.core.files.uploadhandler import SkipFile, StopFutureHandlers, StopUpload from django.utils.datastructures import MultiValueDict from django.utils.encoding import force_str from django.utils.http import parse_header_parameters from django.utils.regex_helper import _lazy_re_compile __all__ = ("MultiPartParser", "MultiPartParserError", "InputStreamExhausted") class MultiPartParserError(Exception): pass class InputStreamExhausted(Exception): """ No more reads are allowed from this device. """ pass RAW = "raw" FILE = "file" FIELD = "field" FIELD_TYPES = frozenset([FIELD, RAW]) class MultiPartParser: """ An RFC 7578 multipart/form-data parser. ``MultiValueDict.parse()`` reads the input stream in ``chunk_size`` chunks and returns a tuple of ``(MultiValueDict(POST), MultiValueDict(FILES))``. """ boundary_re = _lazy_re_compile(r"[ -~]{0,200}[!-~]") def __init__(self, META, input_data, upload_handlers, encoding=None): """ Initialize the MultiPartParser object. :META: The standard ``META`` dictionary in Django request objects. :input_data: The raw post data, as a file-like object. :upload_handlers: A list of UploadHandler instances that perform operations on the uploaded data. :encoding: The encoding with which to treat the incoming data. """ # Content-Type should contain multipart and the boundary information. content_type = META.get("CONTENT_TYPE", "") if not content_type.startswith("multipart/"): raise MultiPartParserError("Invalid Content-Type: %s" % content_type) try: content_type.encode("ascii") except UnicodeEncodeError: raise MultiPartParserError( "Invalid non-ASCII Content-Type in multipart: %s" % force_str(content_type) ) # Parse the header to get the boundary to split the parts. _, opts = parse_header_parameters(content_type) boundary = opts.get("boundary") if not boundary or not self.boundary_re.fullmatch(boundary): raise MultiPartParserError( "Invalid boundary in multipart: %s" % force_str(boundary) ) # Content-Length should contain the length of the body we are about # to receive. try: content_length = int(META.get("CONTENT_LENGTH", 0)) except (ValueError, TypeError): content_length = 0 if content_length < 0: # This means we shouldn't continue...raise an error. raise MultiPartParserError("Invalid content length: %r" % content_length) self._boundary = boundary.encode("ascii") self._input_data = input_data # For compatibility with low-level network APIs (with 32-bit integers), # the chunk size should be < 2^31, but still divisible by 4. possible_sizes = [x.chunk_size for x in upload_handlers if x.chunk_size] self._chunk_size = min([2**31 - 4] + possible_sizes) self._meta = META self._encoding = encoding or settings.DEFAULT_CHARSET self._content_length = content_length self._upload_handlers = upload_handlers def parse(self): # Call the actual parse routine and close all open files in case of # errors. This is needed because if exceptions are thrown the # MultiPartParser will not be garbage collected immediately and # resources would be kept alive. This is only needed for errors because # the Request object closes all uploaded files at the end of the # request. try: return self._parse() except Exception: if hasattr(self, "_files"): for _, files in self._files.lists(): for fileobj in files: fileobj.close() raise def _parse(self): """ Parse the POST data and break it into a FILES MultiValueDict and a POST MultiValueDict. Return a tuple containing the POST and FILES dictionary, respectively. """ from django.http import QueryDict encoding = self._encoding handlers = self._upload_handlers # HTTP spec says that Content-Length >= 0 is valid # handling content-length == 0 before continuing if self._content_length == 0: return QueryDict(encoding=self._encoding), MultiValueDict() # See if any of the handlers take care of the parsing. # This allows overriding everything if need be. for handler in handlers: result = handler.handle_raw_input( self._input_data, self._meta, self._content_length, self._boundary, encoding, ) # Check to see if it was handled if result is not None: return result[0], result[1] # Create the data structures to be used later. self._post = QueryDict(mutable=True) self._files = MultiValueDict() # Instantiate the parser and stream: stream = LazyStream(ChunkIter(self._input_data, self._chunk_size)) # Whether or not to signal a file-completion at the beginning of the loop. old_field_name = None counters = [0] * len(handlers) # Number of bytes that have been read. num_bytes_read = 0 # To count the number of keys in the request. num_post_keys = 0 # To count the number of files in the request. num_files = 0 # To limit the amount of data read from the request. read_size = None # Whether a file upload is finished. uploaded_file = True try: for item_type, meta_data, field_stream in Parser(stream, self._boundary): if old_field_name: # We run this at the beginning of the next loop # since we cannot be sure a file is complete until # we hit the next boundary/part of the multipart content. self.handle_file_complete(old_field_name, counters) old_field_name = None uploaded_file = True if ( item_type in FIELD_TYPES and settings.DATA_UPLOAD_MAX_NUMBER_FIELDS is not None ): # Avoid storing more than DATA_UPLOAD_MAX_NUMBER_FIELDS. num_post_keys += 1 # 2 accounts for empty raw fields before and after the # last boundary. if settings.DATA_UPLOAD_MAX_NUMBER_FIELDS + 2 < num_post_keys: raise TooManyFieldsSent( "The number of GET/POST parameters exceeded " "settings.DATA_UPLOAD_MAX_NUMBER_FIELDS." ) try: disposition = meta_data["content-disposition"][1] field_name = disposition["name"].strip() except (KeyError, IndexError, AttributeError): continue transfer_encoding = meta_data.get("content-transfer-encoding") if transfer_encoding is not None: transfer_encoding = transfer_encoding[0].strip() field_name = force_str(field_name, encoding, errors="replace") if item_type == FIELD: # Avoid reading more than DATA_UPLOAD_MAX_MEMORY_SIZE. if settings.DATA_UPLOAD_MAX_MEMORY_SIZE is not None: read_size = ( settings.DATA_UPLOAD_MAX_MEMORY_SIZE - num_bytes_read ) # This is a post field, we can just set it in the post if transfer_encoding == "base64": raw_data = field_stream.read(size=read_size) num_bytes_read += len(raw_data) try: data = base64.b64decode(raw_data) except binascii.Error: data = raw_data else: data = field_stream.read(size=read_size) num_bytes_read += len(data) # Add two here to make the check consistent with the # x-www-form-urlencoded check that includes '&='. num_bytes_read += len(field_name) + 2 if ( settings.DATA_UPLOAD_MAX_MEMORY_SIZE is not None and num_bytes_read > settings.DATA_UPLOAD_MAX_MEMORY_SIZE ): raise RequestDataTooBig( "Request body exceeded " "settings.DATA_UPLOAD_MAX_MEMORY_SIZE." ) self._post.appendlist( field_name, force_str(data, encoding, errors="replace") ) elif item_type == FILE: # Avoid storing more than DATA_UPLOAD_MAX_NUMBER_FILES. num_files += 1 if ( settings.DATA_UPLOAD_MAX_NUMBER_FILES is not None and num_files > settings.DATA_UPLOAD_MAX_NUMBER_FILES ): raise TooManyFilesSent( "The number of files exceeded " "settings.DATA_UPLOAD_MAX_NUMBER_FILES." ) # This is a file, use the handler... file_name = disposition.get("filename") if file_name: file_name = force_str(file_name, encoding, errors="replace") file_name = self.sanitize_file_name(file_name) if not file_name: continue content_type, content_type_extra = meta_data.get( "content-type", ("", {}) ) content_type = content_type.strip() charset = content_type_extra.get("charset") try: content_length = int(meta_data.get("content-length")[0]) except (IndexError, TypeError, ValueError): content_length = None counters = [0] * len(handlers) uploaded_file = False try: for handler in handlers: try: handler.new_file( field_name, file_name, content_type, content_length, charset, content_type_extra, ) except StopFutureHandlers: break for chunk in field_stream: if transfer_encoding == "base64": # We only special-case base64 transfer encoding # We should always decode base64 chunks by # multiple of 4, ignoring whitespace. stripped_chunk = b"".join(chunk.split()) remaining = len(stripped_chunk) % 4 while remaining != 0: over_chunk = field_stream.read(4 - remaining) if not over_chunk: break stripped_chunk += b"".join(over_chunk.split()) remaining = len(stripped_chunk) % 4 try: chunk = base64.b64decode(stripped_chunk) except Exception as exc: # Since this is only a chunk, any error is # an unfixable error. raise MultiPartParserError( "Could not decode base64 data." ) from exc for i, handler in enumerate(handlers): chunk_length = len(chunk) chunk = handler.receive_data_chunk(chunk, counters[i]) counters[i] += chunk_length if chunk is None: # Don't continue if the chunk received by # the handler is None. break except SkipFile: self._close_files() # Just use up the rest of this file... exhaust(field_stream) else: # Handle file upload completions on next iteration. old_field_name = field_name else: # If this is neither a FIELD nor a FILE, exhaust the field # stream. Note: There could be an error here at some point, # but there will be at least two RAW types (before and # after the other boundaries). This branch is usually not # reached at all, because a missing content-disposition # header will skip the whole boundary. exhaust(field_stream) except StopUpload as e: self._close_files() if not e.connection_reset: exhaust(self._input_data) else: if not uploaded_file: for handler in handlers: handler.upload_interrupted() # Make sure that the request data is all fed exhaust(self._input_data) # Signal that the upload has completed. # any() shortcircuits if a handler's upload_complete() returns a value. any(handler.upload_complete() for handler in handlers) self._post._mutable = False return self._post, self._files def handle_file_complete(self, old_field_name, counters): """ Handle all the signaling that takes place when a file is complete. """ for i, handler in enumerate(self._upload_handlers): file_obj = handler.file_complete(counters[i]) if file_obj: # If it returns a file object, then set the files dict. self._files.appendlist( force_str(old_field_name, self._encoding, errors="replace"), file_obj, ) break def sanitize_file_name(self, file_name): """ Sanitize the filename of an upload. Remove all possible path separators, even though that might remove more than actually required by the target system. Filenames that could potentially cause problems (current/parent dir) are also discarded. It should be noted that this function could still return a "filepath" like "C:some_file.txt" which is handled later on by the storage layer. So while this function does sanitize filenames to some extent, the resulting filename should still be considered as untrusted user input. """ file_name = html.unescape(file_name) file_name = file_name.rsplit("/")[-1] file_name = file_name.rsplit("\\")[-1] # Remove non-printable characters. file_name = "".join([char for char in file_name if char.isprintable()]) if file_name in {"", ".", ".."}: return None return file_name IE_sanitize = sanitize_file_name def _close_files(self): # Free up all file handles. # FIXME: this currently assumes that upload handlers store the file as 'file' # We should document that... # (Maybe add handler.free_file to complement new_file) for handler in self._upload_handlers: if hasattr(handler, "file"): handler.file.close() class LazyStream: """ The LazyStream wrapper allows one to get and "unget" bytes from a stream. Given a producer object (an iterator that yields bytestrings), the LazyStream object will support iteration, reading, and keeping a "look-back" variable in case you need to "unget" some bytes. """ def __init__(self, producer, length=None): """ Every LazyStream must have a producer when instantiated. A producer is an iterable that returns a string each time it is called. """ self._producer = producer self._empty = False self._leftover = b"" self.length = length self.position = 0 self._remaining = length self._unget_history = [] def tell(self): return self.position def read(self, size=None): def parts(): remaining = self._remaining if size is None else size # do the whole thing in one shot if no limit was provided. if remaining is None: yield b"".join(self) return # otherwise do some bookkeeping to return exactly enough # of the stream and stashing any extra content we get from # the producer while remaining != 0: assert remaining > 0, "remaining bytes to read should never go negative" try: chunk = next(self) except StopIteration: return else: emitting = chunk[:remaining] self.unget(chunk[remaining:]) remaining -= len(emitting) yield emitting return b"".join(parts()) def __next__(self): """ Used when the exact number of bytes to read is unimportant. Return whatever chunk is conveniently returned from the iterator. Useful to avoid unnecessary bookkeeping if performance is an issue. """ if self._leftover: output = self._leftover self._leftover = b"" else: output = next(self._producer) self._unget_history = [] self.position += len(output) return output def close(self): """ Used to invalidate/disable this lazy stream. Replace the producer with an empty list. Any leftover bytes that have already been read will still be reported upon read() and/or next(). """ self._producer = [] def __iter__(self): return self def unget(self, bytes): """ Place bytes back onto the front of the lazy stream. Future calls to read() will return those bytes first. The stream position and thus tell() will be rewound. """ if not bytes: return self._update_unget_history(len(bytes)) self.position -= len(bytes) self._leftover = bytes + self._leftover def _update_unget_history(self, num_bytes): """ Update the unget history as a sanity check to see if we've pushed back the same number of bytes in one chunk. If we keep ungetting the same number of bytes many times (here, 50), we're mostly likely in an infinite loop of some sort. This is usually caused by a maliciously-malformed MIME request. """ self._unget_history = [num_bytes] + self._unget_history[:49] number_equal = len( [ current_number for current_number in self._unget_history if current_number == num_bytes ] ) if number_equal > 40: raise SuspiciousMultipartForm( "The multipart parser got stuck, which shouldn't happen with" " normal uploaded files. Check for malicious upload activity;" " if there is none, report this to the Django developers." ) class ChunkIter: """ An iterable that will yield chunks of data. Given a file-like object as the constructor, yield chunks of read operations from that object. """ def __init__(self, flo, chunk_size=64 * 1024): self.flo = flo self.chunk_size = chunk_size def __next__(self): try: data = self.flo.read(self.chunk_size) except InputStreamExhausted: raise StopIteration() if data: return data else: raise StopIteration() def __iter__(self): return self class InterBoundaryIter: """ A Producer that will iterate over boundaries. """ def __init__(self, stream, boundary): self._stream = stream self._boundary = boundary def __iter__(self): return self def __next__(self): try: return LazyStream(BoundaryIter(self._stream, self._boundary)) except InputStreamExhausted: raise StopIteration() class BoundaryIter: """ A Producer that is sensitive to boundaries. Will happily yield bytes until a boundary is found. Will yield the bytes before the boundary, throw away the boundary bytes themselves, and push the post-boundary bytes back on the stream. The future calls to next() after locating the boundary will raise a StopIteration exception. """ def __init__(self, stream, boundary): self._stream = stream self._boundary = boundary self._done = False # rollback an additional six bytes because the format is like # this: CRLF[--CRLF] self._rollback = len(boundary) + 6 # Try to use mx fast string search if available. Otherwise # use Python find. Wrap the latter for consistency. unused_char = self._stream.read(1) if not unused_char: raise InputStreamExhausted() self._stream.unget(unused_char) def __iter__(self): return self def __next__(self): if self._done: raise StopIteration() stream = self._stream rollback = self._rollback bytes_read = 0 chunks = [] for bytes in stream: bytes_read += len(bytes) chunks.append(bytes) if bytes_read > rollback: break if not bytes: break else: self._done = True if not chunks: raise StopIteration() chunk = b"".join(chunks) boundary = self._find_boundary(chunk) if boundary: end, next = boundary stream.unget(chunk[next:]) self._done = True return chunk[:end] else: # make sure we don't treat a partial boundary (and # its separators) as data if not chunk[:-rollback]: # and len(chunk) >= (len(self._boundary) + 6): # There's nothing left, we should just return and mark as done. self._done = True return chunk else: stream.unget(chunk[-rollback:]) return chunk[:-rollback] def _find_boundary(self, data): """ Find a multipart boundary in data. Should no boundary exist in the data, return None. Otherwise, return a tuple containing the indices of the following: * the end of current encapsulation * the start of the next encapsulation """ index = data.find(self._boundary) if index < 0: return None else: end = index next = index + len(self._boundary) # backup over CRLF last = max(0, end - 1) if data[last : last + 1] == b"\n": end -= 1 last = max(0, end - 1) if data[last : last + 1] == b"\r": end -= 1 return end, next def exhaust(stream_or_iterable): """Exhaust an iterator or stream.""" try: iterator = iter(stream_or_iterable) except TypeError: iterator = ChunkIter(stream_or_iterable, 16384) collections.deque(iterator, maxlen=0) # consume iterator quickly. def parse_boundary_stream(stream, max_header_size): """ Parse one and exactly one stream that encapsulates a boundary. """ # Stream at beginning of header, look for end of header # and parse it if found. The header must fit within one # chunk. chunk = stream.read(max_header_size) # 'find' returns the top of these four bytes, so we'll # need to munch them later to prevent them from polluting # the payload. header_end = chunk.find(b"\r\n\r\n") if header_end == -1: # we find no header, so we just mark this fact and pass on # the stream verbatim stream.unget(chunk) return (RAW, {}, stream) header = chunk[:header_end] # here we place any excess chunk back onto the stream, as # well as throwing away the CRLFCRLF bytes from above. stream.unget(chunk[header_end + 4 :]) TYPE = RAW outdict = {} # Eliminate blank lines for line in header.split(b"\r\n"): # This terminology ("main value" and "dictionary of # parameters") is from the Python docs. try: main_value_pair, params = parse_header_parameters(line.decode()) name, value = main_value_pair.split(":", 1) params = {k: v.encode() for k, v in params.items()} except ValueError: # Invalid header. continue if name == "content-disposition": TYPE = FIELD if params.get("filename"): TYPE = FILE outdict[name] = value, params if TYPE == RAW: stream.unget(chunk) return (TYPE, outdict, stream) class Parser: def __init__(self, stream, boundary): self._stream = stream self._separator = b"--" + boundary def __iter__(self): boundarystream = InterBoundaryIter(self._stream, self._separator) for sub_stream in boundarystream: # Iterate over each part yield parse_boundary_stream(sub_stream, 1024)