Re: Linux-2.0.32..

Dean Gaudet (dgaudet-list-linux-kernel@arctic.org)
Tue, 18 Nov 1997 11:30:44 -0800 (PST)

Yeah I ran into this one while hacking on an alternate security model for
apache ... it is fixed in at least 2.1.29. In my case it was TCP and I
needed to know the local address for a socket which had been opened by
another program. When it's bound to INADDR_ANY, 2.0.x returns 127.0.0.1
... which is somewhat wrong.

Dean

On Tue, 18 Nov 1997, Olaf Titz wrote:

> > It also changes the behaviour of some stuff and may break some obscure kernel
> > things. I think not for 2.0.x but if the misbehaviour is in 2.1.x let me know
>
> I had reported the problem a few months ago and was told that it was
> already gone from 2.1. Didn't check myself though.
>
> > Oh and plenty of folks run linux socks5 so I would assume theres a small user
> > space mode
>
> It breaks UDP with socks5, TCP is not affected. Since most applications
> don't use UDP, it may well have gone unnoticed. The problem is that
> with the bug it is impossible to tell from user space whether a socket
> is unbound or bound to 127.0.0.1. (other than tracking the socket(2)
> call, which libsocks5 doesn't do).
>
> olaf
>