classification
Title: SSL support for asyncore
Type: enhancement Stage:
Components: Library (Lib) Versions: Python 3.3
process
Status: open Resolution:
Dependencies: Superseder:
Assigned To: giampaolo.rodola Nosy List: djc, giampaolo.rodola, pitrou
Priority: normal Keywords: patch

Created on 2010-10-13 13:09 by pitrou, last changed 2011-02-12 23:54 by pitrou.

Files
File name Uploaded Description Edit
asyncore_ssl_v1.patch giampaolo.rodola, 2011-02-12 19:36 review
Messages (5)
msg118519 - (view) Author: Antoine Pitrou (pitrou) * (Python committer) Date: 2010-10-13 13:09
It might be useful to make public the SSL support for asyncore which is currently implemented in various tests.
msg118520 - (view) Author: Antoine Pitrou (pitrou) * (Python committer) Date: 2010-10-13 13:11
(I'm posting this issue after having read this message:
http://mail.python.org/pipermail/python-list/2010-October/1257689.html
where the poster is clearly confused about SSL support for asyncore)
msg118552 - (view) Author: Giampaolo Rodola' (giampaolo.rodola) * (Python committer) Date: 2010-10-13 17:40
Problem with SSL dispatcher subclasses used in tests is that they are all similar to pyftpdlib's SSLConnection class ( http://code.google.com/p/pyftpdlib/source/browse/trunk/pyftpdlib/contrib/handlers.py?spec=svn743&r=729#73 ) and I'm not sure it's API is suitable for a general use case. 
It fits well for pyftpdlib, servers in general and stdlib tests but I'm not sure about other uses cases.
In details I'm thinking about clients, secure connections reverted back to clear-text (e.g FTP might need this) and recent issues about certificates validation.
Before writing anything we should agree on an API and make sure it is able to cover all use cases.
msg128459 - (view) Author: Giampaolo Rodola' (giampaolo.rodola) * (Python committer) Date: 2011-02-12 19:36
Initial draft of a patch including tests and a new ssl_dispatcher subclass.
asynchat needs to be changed as well, probably by using a mixin class.
msg128473 - (view) Author: Antoine Pitrou (pitrou) * (Python committer) Date: 2011-02-12 23:54
First comments:

- secure_connection() should be named ssl_something() like other
methods. ssl_start() perhaps?

- in ssl_shutdown():
+                elif err.args[0] == ssl.SSL_ERROR_SSL:
+                    pass

SSL_ERROR_SSL doesn't exist.  Perhaps you mean ssl.SSL_ERROR_EOF?

- in send(), you should handle SSL_ERROR_WANT_READ and
SSL_ERROR_WANT_WRITE as in recv(). Also:
+                if err.args[0] in (ssl.SSL_ERROR_EOF, ssl.SSL_ERROR_ZERO_RETURN):
+                    return 0

lacks a self.handle_close()?

- in recv(), you have "return ''" where it should be "return b''"

- in test_ssl_established(), I think it would be nice if you used e.g.
getpeercert() to check that we really are in SSL mode. Also, you could
make certificate checking mandatory using e.g.:

    ssl_context = ssl.SSLContext(ssl.PROTOCOL_SSLv23)
    ssl_context.verify_mode = ssl.CERT_REQUIRED
    cert_path = os.path.join(os.path.dirname(__file__), "keycert.pem")
    ssl_context.load_cert_chain(cert_path)
    ssl_context.load_verify_locations(cert_path)

- in addition to test_handle_read() and test_handle_write(), there
should be a test where a server and a client really send data to each
other, and receive at all

(also, I'm not sure why these tests can't be shared with non-SSL test
classes)

- test_create_socket() and test_bind() don't seem to test anything
SSL-related
History
Date User Action Args
2011-02-12 23:54:02pitrousetnosy: pitrou, giampaolo.rodola, djc
messages: + msg128473
2011-02-12 19:36:17giampaolo.rodolasetfiles: + asyncore_ssl_v1.patch

messages: + msg128459
keywords: + patch
nosy: pitrou, giampaolo.rodola, djc
2010-12-01 09:55:55djcsetnosy: + djc
2010-10-13 17:40:37giampaolo.rodolasetmessages: + msg118552
2010-10-13 13:11:42pitrousetmessages: + msg118520
2010-10-13 13:09:08pitroucreate