from __future__ import absolute_import import collections import io import logging import sys import warnings import zlib from contextlib import contextmanager from socket import error as SocketError from socket import timeout as SocketTimeout brotli = None from . import util from ._collections import HTTPHeaderDict from .connection import BaseSSLError, HTTPException from .exceptions import ( BodyNotHttplibCompatible, DecodeError, HTTPError, IncompleteRead, InvalidChunkLength, InvalidHeader, ProtocolError, ReadTimeoutError, ResponseNotChunked, SSLError, ) from .packages import six from .util.response import is_fp_closed, is_response_to_head log = logging.getLogger(__name__) class DeflateDecoder(object): def __init__(self): self._first_try = True self._data = b"" self._unfed_data = b"" self._obj = zlib.decompressobj() def __getattr__(self, name): return getattr(self._obj, name) def decompress(self, data, max_length=-1): # CVE-2025-66471: bound the amount of data decompressed per call. # max_length < 0 -> no limit (backward-compatible default); # max_length == 0 -> decompress nothing now, keep the data for a # subsequent call (0 would mean "no limit" to # zlib, so it must not be forwarded); # max_length > 0 -> emit at most max_length bytes, retaining the # rest in the decoder so it is not lost. data = self._unfed_data + data self._unfed_data = b"" if not data and not self._obj.unconsumed_tail: return data original_max_length = max_length if original_max_length < 0: max_length = 0 elif original_max_length == 0: self._unfed_data = data return b"" # Subsequent calls always reuse self._obj. zlib requires passing the # unconsumed tail if decompression is to continue. if not self._first_try: return self._obj.decompress( self._obj.unconsumed_tail + data, max_length ) self._data += data try: decompressed = self._obj.decompress(data, max_length) if decompressed: self._first_try = False self._data = None return decompressed except zlib.error: self._first_try = False self._obj = zlib.decompressobj(-zlib.MAX_WBITS) try: return self.decompress(self._data, original_max_length) finally: self._data = None @property def has_unconsumed_tail(self): # CVE-2025-66471: True while the decoder still holds input it has not # turned into output yet (because a max_length cap was hit), so the # reader knows to keep draining it. return bool(self._unfed_data) or ( bool(self._obj.unconsumed_tail) and not self._first_try ) class GzipDecoderState(object): FIRST_MEMBER = 0 OTHER_MEMBERS = 1 SWALLOW_DATA = 2 class GzipDecoder(object): def __init__(self): self._obj = zlib.decompressobj(16 + zlib.MAX_WBITS) self._state = GzipDecoderState.FIRST_MEMBER self._unconsumed_tail = b"" def __getattr__(self, name): return getattr(self._obj, name) def decompress(self, data, max_length=-1): # CVE-2025-66471: bound the amount of data decompressed per call. # max_length < 0 -> no limit (backward-compatible default); # max_length == 0 -> decompress nothing now, keep the data for a # subsequent call (0 means "no limit" to zlib); # max_length > 0 -> emit at most max_length bytes, retaining the # rest in the decoder so it is not lost. ret = bytearray() if self._state == GzipDecoderState.SWALLOW_DATA: return bytes(ret) if max_length == 0: self._unconsumed_tail += data return b"" # zlib requires passing the unconsumed tail to the subsequent call if # decompression is to continue. data = self._unconsumed_tail + data if not data and self._obj.eof: return bytes(ret) while True: try: # Passing 0 to zlib means "no limit", which is exactly what we # want once the remaining budget reaches zero we break first. ret += self._obj.decompress(data, max(max_length - len(ret), 0)) except zlib.error: previous_state = self._state # Ignore data after the first error self._state = GzipDecoderState.SWALLOW_DATA self._unconsumed_tail = b"" if previous_state == GzipDecoderState.OTHER_MEMBERS: # Allow trailing garbage acceptable in other gzip clients return bytes(ret) raise self._unconsumed_tail = data = ( self._obj.unconsumed_tail or self._obj.unused_data ) if max_length > 0 and len(ret) >= max_length: break if not data: return bytes(ret) # When the end of a gzip member is reached, a new decompressor must # be created for unused (possibly future) data. if self._obj.eof: self._state = GzipDecoderState.OTHER_MEMBERS self._obj = zlib.decompressobj(16 + zlib.MAX_WBITS) return bytes(ret) @property def has_unconsumed_tail(self): # CVE-2025-66471: True while input the decoder has not turned into # output yet (because a max_length cap was hit) is still buffered. return bool(self._unconsumed_tail) if brotli is not None: class BrotliDecoder(object): # Supports both 'brotlipy' and 'Brotli' packages # since they share an import name. The top branches # are for 'brotlipy' and bottom branches for 'Brotli' def __init__(self): self._obj = brotli.Decompressor() if hasattr(self._obj, "decompress"): self._decompress = self._obj.decompress else: self._decompress = self._obj.process def decompress(self, data, max_length=-1): # CVE-2025-66471: the bundled brotli bindings cannot limit their # output size, so max_length is accepted for API compatibility but # not enforced. The whole input is decoded in one call. return self._decompress(data) @property def has_unconsumed_tail(self): return False def flush(self): if hasattr(self._obj, "flush"): return self._obj.flush() return b"" class MultiDecoder(object): """ From RFC7231: If one or more encodings have been applied to a representation, the sender that applied the encodings MUST generate a Content-Encoding header field that lists the content codings in the order in which they were applied. """ # CVE-2025-66418: Maximum allowed number of chained HTTP encodings in # the Content-Encoding header. max_decode_links = 5 def __init__(self, modes): encodings = [m.strip() for m in modes.split(",")] if len(encodings) > self.max_decode_links: raise DecodeError( "Too many content encodings in the chain: " "%d > %d" % (len(encodings), self.max_decode_links) ) self._decoders = [_get_decoder(e) for e in encodings] def flush(self): return self._decoders[0].flush() def decompress(self, data, max_length=-1): # CVE-2025-66471: bound the output of the chain. With no limit, decode # straight through; otherwise loop over the decoders, taking at most # the remaining budget from each, until enough data is produced or no # decoder can produce more. if max_length <= 0: for d in reversed(self._decoders): data = d.decompress(data) return data ret = bytearray() while True: any_data = False for d in reversed(self._decoders): data = d.decompress(data, max_length - len(ret)) if data: any_data = True # Do not break on empty output: later decoders may still # produce data from input buffered on a previous call. ret += data if not any_data or len(ret) >= max_length: return bytes(ret) data = b"" @property def has_unconsumed_tail(self): return any(d.has_unconsumed_tail for d in self._decoders) def _get_decoder(mode): if "," in mode: return MultiDecoder(mode) if mode == "gzip": return GzipDecoder() if brotli is not None and mode == "br": return BrotliDecoder() return DeflateDecoder() class BytesQueueBuffer(object): """Memory-efficient bytes buffer To return decoded data in read() and still follow the BufferedIOBase API, we need a buffer to always return the correct amount of bytes. This buffer should be filled using calls to put() Our maximum memory usage is determined by the sum of the size of: * self.buffer, which contains the full data * the largest chunk that we will copy in get() The worst case scenario is a single chunk, in which case we'll make a full copy of the data inside get(). Backported from urllib3 2.x for the CVE-2025-66471 fix. """ def __init__(self): self.buffer = collections.deque() self._size = 0 def __len__(self): return self._size def put(self, data): self.buffer.append(data) self._size += len(data) def get(self, n): if n == 0: return b"" elif not self.buffer: raise RuntimeError("buffer is empty") elif n < 0: raise ValueError("n should be > 0") fetched = 0 ret = io.BytesIO() while fetched < n: remaining = n - fetched chunk = self.buffer.popleft() chunk_length = len(chunk) if remaining < chunk_length: left_chunk, right_chunk = chunk[:remaining], chunk[remaining:] ret.write(left_chunk) self.buffer.appendleft(right_chunk) self._size -= remaining break else: ret.write(chunk) self._size -= chunk_length fetched += chunk_length if not self.buffer: break return ret.getvalue() def get_all(self): buffer = self.buffer if not buffer: assert self._size == 0 return b"" if len(buffer) == 1: result = buffer.pop() else: ret = io.BytesIO() ret.writelines(buffer.popleft() for _ in range(len(buffer))) result = ret.getvalue() self._size = 0 return result class HTTPResponse(io.IOBase): """ HTTP Response container. Backwards-compatible with :class:`http.client.HTTPResponse` but the response ``body`` is loaded and decoded on-demand when the ``data`` property is accessed. This class is also compatible with the Python standard library's :mod:`io` module, and can hence be treated as a readable object in the context of that framework. Extra parameters for behaviour not present in :class:`http.client.HTTPResponse`: :param preload_content: If True, the response's body will be preloaded during construction. :param decode_content: If True, will attempt to decode the body based on the 'content-encoding' header. :param original_response: When this HTTPResponse wrapper is generated from an :class:`http.client.HTTPResponse` object, it's convenient to include the original for debug purposes. It's otherwise unused. :param retries: The retries contains the last :class:`~urllib3.util.retry.Retry` that was used during the request. :param enforce_content_length: Enforce content length checking. Body returned by server must match value of Content-Length header, if present. Otherwise, raise error. """ CONTENT_DECODERS = ["gzip", "deflate"] if brotli is not None: CONTENT_DECODERS += ["br"] REDIRECT_STATUSES = [301, 302, 303, 307, 308] def __init__( self, body="", headers=None, status=0, version=0, reason=None, strict=0, preload_content=True, decode_content=True, original_response=None, pool=None, connection=None, msg=None, retries=None, enforce_content_length=False, request_method=None, request_url=None, auto_close=True, ): if isinstance(headers, HTTPHeaderDict): self.headers = headers else: self.headers = HTTPHeaderDict(headers) self.status = status self.version = version self.reason = reason self.strict = strict self.decode_content = decode_content self.retries = retries self.enforce_content_length = enforce_content_length self.auto_close = auto_close self._decoder = None # CVE-2025-66471: holds decoded bytes produced beyond what a bounded # read(amt) returned, so the surplus is served on later calls instead # of being discarded. self._decoded_buffer = BytesQueueBuffer() self._body = None self._fp = None self._original_response = original_response self._fp_bytes_read = 0 self.msg = msg self._request_url = request_url if body and isinstance(body, (six.string_types, bytes)): self._body = body self._pool = pool self._connection = connection if hasattr(body, "read"): self._fp = body # Are we using the chunked-style of transfer encoding? self.chunked = False self.chunk_left = None tr_enc = self.headers.get("transfer-encoding", "").lower() # Don't incur the penalty of creating a list and then discarding it encodings = (enc.strip() for enc in tr_enc.split(",")) if "chunked" in encodings: self.chunked = True # Determine length of response self.length_remaining = self._init_length(request_method) # If requested, preload the body. if preload_content and not self._body: self._body = self.read(decode_content=decode_content) def get_redirect_location(self): """ Should we redirect and where to? :returns: Truthy redirect location string if we got a redirect status code and valid location. ``None`` if redirect status and no location. ``False`` if not a redirect status code. """ if self.status in self.REDIRECT_STATUSES: return self.headers.get("location") return False def release_conn(self): if not self._pool or not self._connection: return self._pool._put_conn(self._connection) self._connection = None def drain_conn(self): """ Read and discard any remaining HTTP response data in the response connection. Unread data in the HTTPResponse connection blocks the connection from being released back to the pool. """ try: self.read() except (HTTPError, SocketError, BaseSSLError, HTTPException): pass @property def data(self): # For backwards-compat with earlier urllib3 0.4 and earlier. if self._body: return self._body if self._fp: return self.read(cache_content=True) @property def connection(self): return self._connection def isclosed(self): return is_fp_closed(self._fp) def tell(self): """ Obtain the number of bytes pulled over the wire so far. May differ from the amount of content returned by :meth:``urllib3.response.HTTPResponse.read`` if bytes are encoded on the wire (e.g, compressed). """ return self._fp_bytes_read def _init_length(self, request_method): """ Set initial length value for Response content if available. """ length = self.headers.get("content-length") if length is not None: if self.chunked: # This Response will fail with an IncompleteRead if it can't be # received as chunked. This method falls back to attempt reading # the response before raising an exception. log.warning( "Received response with both Content-Length and " "Transfer-Encoding set. This is expressly forbidden " "by RFC 7230 sec 3.3.2. Ignoring Content-Length and " "attempting to process response as Transfer-Encoding: " "chunked." ) return None try: # RFC 7230 section 3.3.2 specifies multiple content lengths can # be sent in a single Content-Length header # (e.g. Content-Length: 42, 42). This line ensures the values # are all valid ints and that as long as the `set` length is 1, # all values are the same. Otherwise, the header is invalid. lengths = set([int(val) for val in length.split(",")]) if len(lengths) > 1: raise InvalidHeader( "Content-Length contained multiple " "unmatching values (%s)" % length ) length = lengths.pop() except ValueError: length = None else: if length < 0: length = None # Convert status to int for comparison # In some cases, httplib returns a status of "_UNKNOWN" try: status = int(self.status) except ValueError: status = 0 # Check for responses that shouldn't include a body if status in (204, 304) or 100 <= status < 200 or request_method == "HEAD": length = 0 return length def _init_decoder(self): """ Set-up the _decoder attribute if necessary. """ # Note: content-encoding value should be case-insensitive, per RFC 7230 # Section 3.2 content_encoding = self.headers.get("content-encoding", "").lower() if self._decoder is None: if content_encoding in self.CONTENT_DECODERS: self._decoder = _get_decoder(content_encoding) elif "," in content_encoding: encodings = [ e.strip() for e in content_encoding.split(",") if e.strip() in self.CONTENT_DECODERS ] if len(encodings): self._decoder = _get_decoder(content_encoding) DECODER_ERROR_CLASSES = (IOError, zlib.error) if brotli is not None: DECODER_ERROR_CLASSES += (brotli.error,) def _decode(self, data, decode_content, flush_decoder, max_length=None): """ Decode the data passed in and potentially flush the decoder. ``max_length`` (CVE-2025-66471) bounds how many decoded bytes a single call may emit: ``None`` (or a flush) means no limit; ``0`` means decode nothing now and retain the input; a positive value caps the output. """ if not decode_content: return data if max_length is None or flush_decoder: max_length = -1 try: if self._decoder: data = self._decoder.decompress(data, max_length) except self.DECODER_ERROR_CLASSES as e: content_encoding = self.headers.get("content-encoding", "").lower() raise DecodeError( "Received response with content-encoding: %s, but " "failed to decode it." % content_encoding, e, ) if flush_decoder: data += self._flush_decoder() return data def _flush_decoder(self): """ Flushes the decoder. Should only be called if the decoder is actually being used. """ if self._decoder: buf = self._decoder.decompress(b"") return buf + self._decoder.flush() return b"" @contextmanager def _error_catcher(self): """ Catch low-level python exceptions, instead re-raising urllib3 variants, so that low-level exceptions are not leaked in the high-level api. On exit, release the connection back to the pool. """ clean_exit = False try: try: yield except SocketTimeout: # FIXME: Ideally we'd like to include the url in the ReadTimeoutError but # there is yet no clean way to get at it from this context. raise ReadTimeoutError(self._pool, None, "Read timed out.") except BaseSSLError as e: # FIXME: Is there a better way to differentiate between SSLErrors? if "read operation timed out" not in str(e): # SSL errors related to framing/MAC get wrapped and reraised here raise SSLError(e) raise ReadTimeoutError(self._pool, None, "Read timed out.") except (HTTPException, SocketError) as e: # This includes IncompleteRead. raise ProtocolError("Connection broken: %r" % e, e) # If no exception is thrown, we should avoid cleaning up # unnecessarily. clean_exit = True finally: # If we didn't terminate cleanly, we need to throw away our # connection. if not clean_exit: # The response may not be closed but we're not going to use it # anymore so close it now to ensure that the connection is # released back to the pool. if self._original_response: self._original_response.close() # Closing the response may not actually be sufficient to close # everything, so if we have a hold of the connection close that # too. if self._connection: self._connection.close() # If we hold the original response but it's closed now, we should # return the connection back to the pool. if self._original_response and self._original_response.isclosed(): self.release_conn() def _fp_read(self, amt): """ Read a response with the thought that reading the number of bytes larger than can fit in a 32-bit int at a time via SSL in some known cases leads to an overflow error that has to be prevented if `amt` or `self.length_remaining` indicate that a problem may happen. The known cases: * 3.8 <= CPython < 3.9.7 because of a bug https://github.com/urllib3/urllib3/issues/2513#issuecomment-1152559900. * urllib3 injected with pyOpenSSL-backed SSL-support. * CPython < 3.10 only when `amt` does not fit 32-bit int. """ assert self._fp c_int_max = 2 ** 31 - 1 if ( ( (amt and amt > c_int_max) or (self.length_remaining and self.length_remaining > c_int_max) ) and not util.IS_SECURETRANSPORT and (util.IS_PYOPENSSL or sys.version_info < (3, 10)) ): buffer = io.BytesIO() # Besides `max_chunk_amt` being a maximum chunk size, it # affects memory overhead of reading a response by this # method in CPython. # `c_int_max` equal to 2 GiB - 1 byte is the actual maximum # chunk size that does not lead to an overflow error, but # 256 MiB is a compromise. max_chunk_amt = 2 ** 28 while amt is None or amt != 0: if amt is not None: chunk_amt = min(amt, max_chunk_amt) amt -= chunk_amt else: chunk_amt = max_chunk_amt data = self._fp.read(chunk_amt) if not data: break buffer.write(data) del data # to reduce peak memory usage by `max_chunk_amt`. return buffer.getvalue() else: # StringIO doesn't like amt=None return self._fp.read(amt) if amt is not None else self._fp.read() def _raw_read(self, amt=None): """ Read ``amt`` bytes of the *raw* (still-encoded) response body off the socket, taking care of length bookkeeping and closing the connection once the body is exhausted. Returns ``None`` if there is no file object to read from. """ if self._fp is None: return None fp_closed = getattr(self._fp, "closed", False) with self._error_catcher(): data = self._fp_read(amt) if not fp_closed else b"" if amt is not None and amt != 0 and not data: # Platform-specific: Buggy versions of Python. # Close the connection when no data is returned # # This is redundant to what httplib/http.client _should_ # already do. However, versions of python released before # December 15, 2012 (http://bugs.python.org/issue16298) do # not properly close the connection in all cases. There is # no harm in redundantly calling close. self._fp.close() if self.enforce_content_length and self.length_remaining not in ( 0, None, ): # This is an edge case that httplib failed to cover due # to concerns of backward compatibility. We're # addressing it here to make sure IncompleteRead is # raised during streaming, so all calls with incorrect # Content-Length are caught. raise IncompleteRead(self._fp_bytes_read, self.length_remaining) if data: self._fp_bytes_read += len(data) if self.length_remaining is not None: self.length_remaining -= len(data) return data def read(self, amt=None, decode_content=None, cache_content=False): """ Similar to :meth:`http.client.HTTPResponse.read`, but with two additional parameters: ``decode_content`` and ``cache_content``. :param amt: How much of the content to read. If specified, caching is skipped because it doesn't make sense to cache partial content as the full response. :param decode_content: If True, will attempt to decode the body based on the 'content-encoding' header. :param cache_content: If True, will save the returned data such that the same result is returned despite of the state of the underlying file object. This is useful if you want the ``.data`` property to continue working after having ``.read()`` the file object. (Overridden if ``amt`` is set.) """ self._init_decoder() if decode_content is None: decode_content = self.decode_content if amt and amt < 0: # Negative numbers and `None` should be treated the same. amt = None if amt is not None: cache_content = False # CVE-2025-66471: a bounded read must not produce unbounded output. # First drain (bounded) any data the decoder has already buffered # internally, then serve from the decoded buffer if it can satisfy # the request without touching the socket. if self._decoder is not None and self._decoder.has_unconsumed_tail: decoded_data = self._decode( b"", decode_content, flush_decoder=False, max_length=amt - len(self._decoded_buffer), ) self._decoded_buffer.put(decoded_data) if len(self._decoded_buffer) >= amt: return self._decoded_buffer.get(amt) data = self._raw_read(amt) flush_decoder = amt is None or (amt != 0 and not data) if ( not data and len(self._decoded_buffer) == 0 and not (self._decoder is not None and self._decoder.has_unconsumed_tail) ): return data if amt is None: data = self._decode(data, decode_content, flush_decoder) if cache_content: self._body = data else: # Bound the decoded output to `amt` per decode call and buffer the # surplus, reading more raw data until we have `amt` decoded bytes # or the response is exhausted. decoded_data = self._decode( data, decode_content, flush_decoder, max_length=amt - len(self._decoded_buffer), ) self._decoded_buffer.put(decoded_data) while len(self._decoded_buffer) < amt and data: data = self._raw_read(amt) decoded_data = self._decode( data, decode_content, flush_decoder, max_length=amt - len(self._decoded_buffer), ) self._decoded_buffer.put(decoded_data) # The buffer may legitimately be empty (e.g. a body that decodes to # nothing); ``get`` raises on an empty buffer, so guard for it. if len(self._decoded_buffer) == 0: return b"" data = self._decoded_buffer.get(amt) return data def stream(self, amt=2 ** 16, decode_content=None): """ A generator wrapper for the read() method. A call will block until ``amt`` bytes have been read from the connection or until the connection is closed. :param amt: How much of the content to read. The generator will return up to much data per iteration, but may return less. This is particularly likely when using compressed data. However, the empty string will never be returned. :param decode_content: If True, will attempt to decode the body based on the 'content-encoding' header. """ if self.chunked and self.supports_chunked_reads(): for line in self.read_chunked(amt, decode_content=decode_content): yield line else: # CVE-2025-66471: keep streaming while the socket has data, the # decoded buffer still holds bytes, or the decoder is still holding # input it has not turned into output yet. Without the last two # conditions a bounded read() that left decoded/undecoded data # behind would be silently truncated once the socket closed. while ( not is_fp_closed(self._fp) or len(self._decoded_buffer) > 0 or (self._decoder is not None and self._decoder.has_unconsumed_tail) ): data = self.read(amt=amt, decode_content=decode_content) if data: yield data @classmethod def from_httplib(ResponseCls, r, **response_kw): """ Given an :class:`http.client.HTTPResponse` instance ``r``, return a corresponding :class:`urllib3.response.HTTPResponse` object. Remaining parameters are passed to the HTTPResponse constructor, along with ``original_response=r``. """ headers = r.msg if not isinstance(headers, HTTPHeaderDict): if six.PY2: # Python 2.7 headers = HTTPHeaderDict.from_httplib(headers) else: headers = HTTPHeaderDict(headers.items()) # HTTPResponse objects in Python 3 don't have a .strict attribute strict = getattr(r, "strict", 0) resp = ResponseCls( body=r, headers=headers, status=r.status, version=r.version, reason=r.reason, strict=strict, original_response=r, **response_kw ) return resp # Backwards-compatibility methods for http.client.HTTPResponse def getheaders(self): warnings.warn( "HTTPResponse.getheaders() is deprecated and will be removed " "in urllib3 v2.1.0. Instead access HTTPResponse.headers directly.", category=DeprecationWarning, stacklevel=2, ) return self.headers def getheader(self, name, default=None): warnings.warn( "HTTPResponse.getheader() is deprecated and will be removed " "in urllib3 v2.1.0. Instead use HTTPResponse.headers.get(name, default).", category=DeprecationWarning, stacklevel=2, ) return self.headers.get(name, default) # Backwards compatibility for http.cookiejar def info(self): return self.headers # Overrides from io.IOBase def close(self): if not self.closed: self._fp.close() if self._connection: self._connection.close() if not self.auto_close: io.IOBase.close(self) @property def closed(self): if not self.auto_close: return io.IOBase.closed.__get__(self) elif self._fp is None: return True elif hasattr(self._fp, "isclosed"): return self._fp.isclosed() elif hasattr(self._fp, "closed"): return self._fp.closed else: return True def fileno(self): if self._fp is None: raise IOError("HTTPResponse has no file to get a fileno from") elif hasattr(self._fp, "fileno"): return self._fp.fileno() else: raise IOError( "The file-like object this HTTPResponse is wrapped " "around has no file descriptor" ) def flush(self): if ( self._fp is not None and hasattr(self._fp, "flush") and not getattr(self._fp, "closed", False) ): return self._fp.flush() def readable(self): # This method is required for `io` module compatibility. return True def readinto(self, b): # This method is required for `io` module compatibility. temp = self.read(len(b)) if len(temp) == 0: return 0 else: b[: len(temp)] = temp return len(temp) def supports_chunked_reads(self): """ Checks if the underlying file-like object looks like a :class:`http.client.HTTPResponse` object. We do this by testing for the fp attribute. If it is present we assume it returns raw chunks as processed by read_chunked(). """ return hasattr(self._fp, "fp") def _update_chunk_length(self): # First, we'll figure out length of a chunk and then # we'll try to read it from socket. if self.chunk_left is not None: return line = self._fp.fp.readline() line = line.split(b";", 1)[0] try: self.chunk_left = int(line, 16) except ValueError: # Invalid chunked protocol response, abort. self.close() raise InvalidChunkLength(self, line) def _handle_chunk(self, amt): returned_chunk = None if amt is None: chunk = self._fp._safe_read(self.chunk_left) returned_chunk = chunk self._fp._safe_read(2) # Toss the CRLF at the end of the chunk. self.chunk_left = None elif amt < self.chunk_left: value = self._fp._safe_read(amt) self.chunk_left = self.chunk_left - amt returned_chunk = value elif amt == self.chunk_left: value = self._fp._safe_read(amt) self._fp._safe_read(2) # Toss the CRLF at the end of the chunk. self.chunk_left = None returned_chunk = value else: # amt > self.chunk_left returned_chunk = self._fp._safe_read(self.chunk_left) self._fp._safe_read(2) # Toss the CRLF at the end of the chunk. self.chunk_left = None return returned_chunk def read_chunked(self, amt=None, decode_content=None): """ Similar to :meth:`HTTPResponse.read`, but with an additional parameter: ``decode_content``. :param amt: How much of the content to read. If specified, caching is skipped because it doesn't make sense to cache partial content as the full response. :param decode_content: If True, will attempt to decode the body based on the 'content-encoding' header. """ self._init_decoder() # FIXME: Rewrite this method and make it a class with a better structured logic. if not self.chunked: raise ResponseNotChunked( "Response is not chunked. " "Header 'transfer-encoding: chunked' is missing." ) if not self.supports_chunked_reads(): raise BodyNotHttplibCompatible( "Body should be http.client.HTTPResponse like. " "It should have have an fp attribute which returns raw chunks." ) with self._error_catcher(): # Don't bother reading the body of a HEAD request. if self._original_response and is_response_to_head(self._original_response): self._original_response.close() return # If a response is already read and closed # then return immediately. if self._fp.fp is None: return while True: self._update_chunk_length() if self.chunk_left == 0: break chunk = self._handle_chunk(amt) decoded = self._decode( chunk, decode_content=decode_content, flush_decoder=False, max_length=amt, ) if decoded: yield decoded # CVE-2025-66471: if the max_length cap left undecoded input in # the decoder, drain it (bounded to `amt` per yield) so the # chunked body is not truncated. while ( decode_content and self._decoder is not None and self._decoder.has_unconsumed_tail ): decoded = self._decode( b"", decode_content=decode_content, flush_decoder=False, max_length=amt, ) if not decoded: break yield decoded if decode_content: # On CPython and PyPy, we should never need to flush the # decoder. However, on Jython we *might* need to, so # lets defensively do it anyway. decoded = self._flush_decoder() if decoded: # Platform-specific: Jython. yield decoded # Chunk content ends with \r\n: discard it. while True: line = self._fp.fp.readline() if not line: # Some sites may not end with '\r\n'. break if line == b"\r\n": break # We read everything; close the "file". if self._original_response: self._original_response.close() def geturl(self): """ Returns the URL that was the source of this response. If the request that generated this response redirected, this method will return the final redirect location. """ if self.retries is not None and len(self.retries.history): return self.retries.history[-1].redirect_location else: return self._request_url def __iter__(self): buffer = [] for chunk in self.stream(decode_content=True): if b"\n" in chunk: chunk = chunk.split(b"\n") yield b"".join(buffer) + chunk[0] + b"\n" for x in chunk[1:-1]: yield x + b"\n" if chunk[-1]: buffer = [chunk[-1]] else: buffer = [] else: buffer.append(chunk) if buffer: yield b"".join(buffer)