Multiple vulnerabilities in BaseHTTPRequestHandler enable HTTP response splitting attacks

[meitar]

BPO	29424
Nosy	@meitar
Files	python-3.6.0.patch: Proposed patch for Python 3.4+ python-2.7.13.patch: Proposed patch for Python 2.7+

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2017-02-02.21:33:01.182>
labels = ['type-security', '3.7', 'library']
title = 'Multiple vulnerabilities in BaseHTTPRequestHandler enable HTTP response splitting attacks'
updated_at = <Date 2017-02-13.03:28:26.836>
user = 'https://github.com/meitar'

bugs.python.org fields:

activity = <Date 2017-02-13.03:28:26.836>
actor = 'bgilbert'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2017-02-02.21:33:01.182>
creator = 'meitar'
dependencies = []
files = ['46497', '46498']
hgrepos = []
issue_num = 29424
keywords = ['patch']
message_count = 2.0
messages = ['286828', '286829']
nosy_count = 2.0
nosy_names = ['bgilbert', 'meitar']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'security'
url = 'https://bugs.python.org/issue29424'
versions = ['Python 2.7', 'Python 3.4', 'Python 3.5', 'Python 3.6', 'Python 3.7']

[meitar]

SUMMARY:

In the Python standard library, the BaseHTTPRequestHandler class’s send_header() method[0] does not correctly construct HTTP/1.1 message headers as described by Section 4.2 of RFC 2616[1] when it is given maliciously-crafted input, leaving applications that rely on it vulnerable to HTTP response splitting[2] if they do not take extra precautions, themselves. A similar vulnerability affects the BaseHTTPRequestHandler class’s send_response_only() method, as well, although this is not likely to be as exploitable in the wild. This second vulnerability can result in HTTP response splitting due to incorrect construction of the Reason-Phrase portion of an HTTP Status-Line.[3]

Since these APIs are designed to handle user-supplied input, it is reasonable to assume that developers will expect the standard library to consume arbitrary input safely. Unfortunately, the library fails to do that in these cases. According to a simple GitHub code search, slightly more than 100,000 repositories are directly using BaseHTTPRequestHandler,[4] so it is possible that a significant percentage of those applications are affected by this vulnerability.

PYTHON VERSIONS AFFECTED:

• Current development tip at time of writing: https://hg.python.org/cpython/file/tip/Lib/http/server.py#l511
• Current stable version 3.6 release (https://hg.python.org/cpython/file/3.6/Lib/http/server.py#l508)
• Current stable version 2.7.13 release (https://hg.python.org/cpython/file/2.7/Lib/BaseHTTPServer.py#l412)

DETAILS:

According to the HTTP specification, an HTTP message header field content *MUST NOT* contain the sequence CRLF (carriage return, line feed). The RFC defines a message header as:

message-header = field-name ":" [ field-value ]
field-name = token
field-value = *( field-content | LWS )
field-content = <the OCTETs making up the field-value
and consisting of either *TEXT or combinations
of token, separators, and quoted-string>

The RFC defines *TEXT to be the same as defined in RFC 822 section 3.3:[5]

text = <any CHAR, including bare ; => atoms, specials,
CR & bare LF, but NOT ; comments and
including CRLF> ; quoted-strings are
; NOT recognized.

However, the send_header() method does not perform any checking to ensure that a message header field-name nor field-content contains no CRLF sequences. The vulnerable Python 3.x code is in Lib/http/server.py on lines 507 and 508:

 self._headers_buffer.append(
 ("%s: %s\r\n" % (keyword, value)).encode('latin-1', 'strict'))

An impacted application is one that passes user-provided input into the send_header() method, such as is common for setting HTTP Cookies. An example of an affected application may run Python code such as:

def do_POST(self):
 # receive user-supplied data from a POST’ed HTTP request
 form_input = parse.unquote_plus(
 self.rfile.read(int(self.headers.get('content-length'))).decode('utf8')
 ).split('=')
 username = form_input[1] # extract a user-supplied value
 self.send_header('Set-Cookie', 'user={}'.format(username)) # use that value, assuming library will provide safety!
 self.end_headers()
 # ... send HTTP response body ...

Assuming the code above, this HTTP POST request…

POST https://victim.example/ HTTP/1.1
Host: victim.example
Content-Type: application/x-www-form-urlencoded
Content-Length: 10

user=alice

…would produce something like this (safe) HTTP response:

HTTP/1.0 200 OK
Server: BaseHTTP/0.6 Python/3.4.5
Date: Thu, 19 Jan 2017 22:58:44 GMT
Set-Cookie: user=alice

...HTTP RESPONSE BODY...

However, if an attacker supplies the following, maliciously-crafted HTTP POST payload…

POST https://victim.example/ HTTP/1.1
Host: victim.example
Content-Type: application/x-www-form-urlencoded
Content-Length: 46

user=%0d%0aContent-Length: 6%0d%0a%0d%0aHACKED

…then the application would serve a page that simply read “HACKED” as its output:

HTTP/1.0 200 OK
Server: BaseHTTP/0.6 Python/3.4.5
Date: Thu, 19 Jan 2017 22:58:44 GMT
Set-Cookie: user=
Content-Length: 6

HACKED

The remainder of the application’s original, intended HTTP response would be ignored by the client. This allows the attacker to submit arbitrary data to clients, including hijacking complete pages and executing XSS attacks.

SOLUTION:

To fix this issue, ensure that CRLF sequences in the relevant user-supplied values are replaced with a single SPACE character, as the RFC instructs.[1] For example, Python code referencing

value

should become

value.replace("\r\n", ' ')

Similar replacements should be made to the other relevant arguments in the affected methods.

Two patches are attached to this email, one for each associated affected version of Python. Apply each patch with:

cd Python-$VERSION
patch -p1 < python-$VERSION.patch

where $VERSION is the version of Python to be patched. The patch for Python 3.6.0 applies cleanly to the current development tip, as well.

EXTERNAL REFERENCES:

[0] https://hg.python.org/cpython/file/3.6/Lib/http/server.py#l502
[1] https://tools.ietf.org/html/rfc2616#section-4.2
[2] https://www.owasp.org/index.php/HTTP_Response_Splitting
[3] https://tools.ietf.org/html/rfc2616#section-6.1
[4] https://github.com/search?q=BaseHTTPRequestHandler+language%3Apython&type=Code
[5] https://tools.ietf.org/html/rfc822#section-3.3

SEE ALSO: https://bugs.python.org/issue17319

P.S. This is my first time reporting a bug to the Python Software Foundation. I can prioritize improving this report with feedback from you if you need any clarifications or further details.

[meitar]

A separate patch for Python 2.7+ that handles the issue described in the same way as the 3.4+ patch.