3

I sometimes use PuTTY as a SOCKS proxy, and I've noticed that sometimes, when the web-page I'm trying to connect to (from the web-browser) does not exist, and requires a prolonged timeout, the shell session stalls (cannot type interactively in the terminal until the timeout is reached), and also all other web-page requests stall, too.

I've recently noticed that this appears to be related to DNS: right now, it would seem like the servers specified on the sshd side in /etc/resolv.conf are having some issues, and, as a result, it's nearly impossible to browse the internet through a PuTTY SOCKS proxy, and also the PuTTY terminal is stalled almost all the time when any web browsing is unsuccessfully attempted.

The following error frequently appears in PuTTY logs, after which the stalling appears to stop for a little while:

2014-01-11 17:12:03 Forwarded connection refused by server: Administratively prohibited [open failed]

Normally, this is what I see in the logs instead, which gives me the impression that my SOCKS-enabled browser doesn't even know which IP-address the SOCKS proxy will connect it to:

2014-01-11 17:18:11 Opening forwarded connection to superuser.com:80

Changing the DNS server around the ssh daemon would only be a temporary solution, which would not address the underlying issue with OpenSSH / PuTTY stalling in these situations. (Not using hostnames through SOCKS would seem to be wrong, too.)

Is there any way to mitigate the ssh stalling for good?

(At the very minimum, it would seem like sshd should be more aggressive in timing out DNS, and retrying with another server. I have several servers specified in /etc/resolv.conf, and dig seems to re-issue the request to the next server after exactly 1s, which seems more appropriate than what sshd appears to be doing.)

1

According to an OpenSSH developer on the official mailing list, http://lists.mindrot.org/pipermail/openssh-unix-dev/2014-January/031974.html, this is due to the fact that a standard libc resolver is used by OpenSSH, which is synchronous and blocking, so no other traffic is processed whilst a resolution is in progress.

Fixing this issue would require a fix around ssh/channels.c#connect_to in OpenSSH (bug #1357 confirms that it's in the codepath of SOCKS5 as used by Mozilla), to use a different resolver function than default getaddrinfo from libc.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.