[PLUG] Re: SOLVED: TCP connect() times out

[Paul Heinlein]

The problem seems to have been that tcp congestion notification 
(net.ipv4.tcp_ecn) had been turned on, which doesn't work and play 
well with my internet connection. Turning it off got things working 
again.

My guess (and it's only a guess) is that /etc/sysctl.conf was tweaked 
when I updated the Gentoo sys-apps/baselayout package back on June 10. 
When I rebooted the machine yesterday morning, it got a new ECN 
setting, which hosed things.

Well, that blew a few hours of my time. Feh. :-)


On Sat, 9 Jul 2005, Paul Heinlein wrote:

> I got an odd new problem. The host in question (Gentoo, SPARC, 
> 2.4.30) is on a five-node LAN sitting behind a NAT gateway.
>
> I can
>
> a. ping any host on the LAN or Internet (i.e., ICMP works fine)
> b. query any DNS server on LAN or Internet (UDP works fine)
> c. receive TCP connections from LAN or Internet
> d. initiate TCP connections to hosts on LAN
>
> I cannot
>
> * initiate TCP connections to hosts on the Internet
>
> The problem isn't specific to any TCP protocol; they all fail: smtp, 
> telnet, ftp, http, ...
>
> All the other hosts on the LAN have no trouble initiating 
> connections to off-site servers.
>
> I'm working on isolating the interesting parts of strace output; 
> right now the only thing that leaps out at me is that connect() is 
> timing out, e.g.,
>
> connect(3, {sa_family=AF_INET, sin_port=htons(80), 
> sin_addr=inet_addr("129.95.36.100")}, 16) = -1 ETIMEDOUT (Connection 
> timed out)
>
> The problem just began in the past 24 hours. I've recently emerged 
> some pam packages, but core things like glibc and the kernel haven't 
> changed. Clues?

-- Paul Heinlein <heinlein at madboa.com>