Re: [Exim] [BUG?] 3.32: IPv6 changes cause Exim to not start

A long time ago, in a galaxy far, far way, someone said...

> If I am not mistaken (and of-course by Philip's email), is this on
> Linux? What platforms are you using this on?

Um, yes, I forgot to mention the platform :)

Linux on Intel - stock Debian 2.2, with libssl from Woody (custom compile
and all that) since the libssl in a stock Potato install is a little too
old for the TLS support in Exim to work.

This happened on an number of PII and PIII systems so it's not just this
one computer.

I should probably try the same config on FreeBSD & NetBSD (I have my
kernels compiled without IPv6 support on those systems) and see what they
do... (runs off to compile 3.32 on a SPARC IPX :( )

> > On systems that do have IPv6 support, Exim consumes all CPU time by
> > writing this to the paniclog:
>
> > 50 accept() failures: Invalid argument
>
> Ok, this is bad. What platform(s)?

Stock Debian "potato" 2.2. With an updated libssl (Exim's TLS code won't
work with libssl 0.9.4) and an updated modutils to support the 2.4 kernel.

> > Granted, I probably should have configured IPv6 in this situation...
>
> ;-)
>
> > Exim 3.31 (and earlier releases) do not have this behavior and work
> > corretly. A workaround has been to not compile Exim with IPv6 support :)
>
> Yes, Exim-3.32 release had its IPv6 code re-written so as to support
> OS's such as Open/Net BSD which do not route IPv4 traffic to AF_INET6
> socket(s).

Yes, I saw that - I had the 3.32 changelog open as I was writing my first
email. I would think that Exim would not try to do accept() on an IPv6
socket if it got an error trying to create it.

Two possibilities:
1) Linux bug for not reporting the error
2) Exim but for not correctly testing for the error condition.

> I tested and reported that it worked on both these OS's, and I am
> pretty sure Philip/Sheldon tested this on Solaris and Free BSD
> respectively.
>
> As per Philip's email, he is going to address this issue after he
> attends the IETF meeting on IPv6.

I think that would be a good idea - people keep telling me that we'll be
using IPv6 Real Soon Now but it's little crap like this that's going to
make us stay with IPv4.


Phil