Re: Disabling ip4 and IPV6 on F20RC1

----- Original Message -----
From: "Tore Anderson" <tore fud no>
To: "Pavel Simerda" <psimerda redhat com>
Cc: "Dan Winship" <danw redhat com>, networkmanager-list gnome org
Sent: Wednesday, December 18, 2013 12:10:15 AM
Subject: Re: Disabling ip4 and IPV6 on F20RC1

* Pavel Simerda

There's no correct handling of RA lifetimes until the standards are
fixed, anyway. That is something I feel much more motivated for, so
if you want to discuss that with me, feel free. A wiki page might be
useful for that.

The Linux kernel handles RA lifetimes correctly

Linux kernel will become irrelevant with new releases of NetworkManager as the Linux kernel router discovery 
was never designed to be used in multi-network and VPN scenarios and NetworkManager git master already does 
router discovery in user space.

RFC 4861 seems to describe this in a rather
straight-forward way to me, so I am therefore not sure how exactly you
feel that the standards need "fixing"?

It's only a bit sad that the whole handling of lifetimes is there
because of a (in my opinion shortsighted) decision to develop a
stateless autoconfiguration protocol for IPv6. Single-lifetime
contract-based protocol like DHCP seems to be a much better option in
the long term and this is one of the things that delays IPv6
deployment without any real advantages. But that's nothing more than
an opinion of mine.

An opinion you're entitled to, and I'm not necessarily disagreeing with
you either. However, the IPv6 standard is what it is,

1) We shouldn't pretend there is an IPv6 standard. IPv6 commonly refers to various sets of IETF standards 
which change in time and vary in their matureness and usability.

2) We should try our best to support the various IPv6 standards but we shouldn't do so blindly. After all the 
purpose of the standards is (or should be) to have a working network environment and that is also what our 
users expect.

NetworkManager deviates from the standard, there is no question about

All known implementations do. And currently all working implementations have to. Plus the standard*s* are 
sometimes unclear, sometimes ambiguous, sometimes apparently buggy. But that's not what we're talking about, 

This makes building networks that rely on RAs for high-availability
router service difficult, as Anders Blomdell recently found out.

Unfortunately that won't help much with the current code base.

The kernel IPv6 autoconfiguration was inherently wrong from the multi-network point of view. But the current 
code should be very close to what you actually want to see, but implemented in userspace instead of the 
kernel. That's why I and Jiří Pírko made the userspace implementation, after all, and why Dan Winship, Thomas 
Haller and Dan Williams are improving it.

I would say the current situation (in git master) is pretty good in that any bugs you find there can be fixed 
in a rather straightforward way. It's a pity we don't have it in a stable release, yet, but if anyone wants 
to test the development branch (I do), that will be great.

(I believe this is the only way to do HA with Linux-based routers, which aggravates the issue.)

It's been there for a while but apparently nobody wanted to touch the hacky implementation.



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]