[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: sysctl options for Maximum segment size
> Hello!
>
> On Fri, Dec 01, 2000 at 07:42:33AM +0100, Rémi Guyomarch wrote:
> > [...]
>
> > 2) OpenBSD 2.7 doesn't have PMTU discovery, I believe 512 is just low
> > enough to cross most routers without getting into MTU
> > problems. OpenBSD 2.8 have PMTU discovery, but maybe the default
> > didn't get changed.
>
> OpenBSD 2.8 however does not respect
> net.inet.tcp.mssdflt, even if
> net.inet.ip.mtudisc = 0
>
> Neither does it respect the manual route MTU setting on the
> route used (default), but instead it derives a MSS from the
> interface MTU. That leads to problems, as I explicitly set
> those values (mssdflt and the MTU on the default route) in
> order to work around MTU problems (excess fragmentation or
> PMTU problems for other boxes which, for some reason, don't
> get the "need to fragment" ICMP packets).
>
> > [...]
>
> Kind regards, Hannah.
I find this rmx_mtu area very suspect too - in OpenBSD and in FreeBSD.
A quick glance at OpenBSD shows a lot of XXXs around rmx_mtu...
It seems that rmx_mtu (the route mtu) is only ever really used for
pmtu discovery. I don't believe it's used at all when determining
what can be shoved down an interface - the iface mtu is used there.
Interestingly enough, when packets with IP_DF set won't fit according
to the interface mtu and when the route mtu isn't locked, FreeBSD
will update the route mtu with the interface mtu value if the route
mtu is larger.... there are snippets of code around in both trees
that reduce rmx_mtu... but at different points in each.
Sooo, my reason for posting....
Despite the fact that I clearly don't understand what's going on here,
I believe that increasing the route MTU to a value greater than the
interface MTU is wrong, and will eventually be automatically fixed....
--
Brian <brian@Awfulhak.org> <brian@[uk.]FreeBSD.org>
<http://www.Awfulhak.org> <brian@[uk.]OpenBSD.org>
Don't _EVER_ lose your sense of humour !