PEP 475: handle EINTR in the ssl module

[vstinner]

BPO	23853
Nosy	@vstinner
Files	test_ssl_bug.patch ssl_timeout.patch

^{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 = <Date 2015-04-06.23:10:31.860>
created_at = <Date 2015-04-02.15:30:24.636>
labels = []
title = 'PEP 475: handle EINTR in the ssl module'
updated_at = <Date 2015-04-06.23:10:31.859>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2015-04-06.23:10:31.859>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2015-04-06.23:10:31.860>
closer = 'vstinner'
components = []
creation = <Date 2015-04-02.15:30:24.636>
creator = 'vstinner'
dependencies = []
files = ['38809', '38816']
hgrepos = []
issue_num = 23853
keywords = ['patch']
message_count = 5.0
messages = ['239924', '239927', '239967', '239982', '240182']
nosy_count = 2.0
nosy_names = ['vstinner', 'python-dev']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = None
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue23853'
versions = ['Python 3.5']

[vstinner]

The _ssl module uses an helper function check_socket_and_wait_for_timeout() to poll until the socket is ready (got data or became writable). check_socket_and_wait_for_timeout() always uses the socket timeout. If select() or poll() is interrupted by a signal (fails with EINTR), check_socket_and_wait_for_timeout() is restarted with the same timeout, which doesn't respect the contract of the timeout: the operation must timeout if it takes more than timeout seconds.

The code must be modified to recompute the timeout, as done in the new sock_call() function of Modules/socketmodule.c (issue bpo-23618). At least, the timeout must decreases when select()/poll() fails with EINTR.

Currently, the timeout is reset after each read/write/handshake operation. IMO the timeout must apply on the total duration of the ssl method, not be reset. But changing this may break backward compatibility :-/

Note: if the signal handler raises an exception, the ssl method fails with the exception. This issue is specific to signal handlers not raising an exception.

[vstinner]

test_ssl_bug.patch: Modify test_handshake_timeout() of test_ssl to show the bug: test_handshake_timeout() hangs with the patch (which sends a signal every millisecond).

[python-dev]

New changeset 753233baf27e by Victor Stinner in branch 'default':
Issue bpo-23853: Cleanup _ssl.c
https://hg.python.org/cpython/rev/753233baf27e

[vstinner]

Here is a patch to fix the issue: recompute the timeout.

It's unclear to me if we should reset the timeout after each successful read/write, or if the timeout is "global" (total duration of the ssl method). I asked the question on the python-dev mailing list.
https://mail.python.org/pipermail/python-dev/2015-April/139001.html

My patch uses a global timeout (never reset the timeout), whereas socket.sendall() resets the timeout at each send() success.

[python-dev]

New changeset cdc83da0b0f8 by Victor Stinner in branch 'default':
Issue bpo-23853: socket.socket.sendall() does no more reset the socket timeout
https://hg.python.org/cpython/rev/cdc83da0b0f8

New changeset 5983f3fdacdb by Victor Stinner in branch 'default':
Issue bpo-23853: Methods of SSL socket don't reset the socket timeout anymore each
https://hg.python.org/cpython/rev/5983f3fdacdb