[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 5 May 2020 16:26:30 -0700
From: Maciej Żenczykowski <zenczykowski@...il.com>
To: David Miller <davem@...emloft.net>
Cc: Linux NetDev <netdev@...r.kernel.org>,
Eric Dumazet <edumazet@...gle.com>,
Willem de Bruijn <willemb@...gle.com>,
Xin Long <lucien.xin@...il.com>,
Hannes Frederic Sowa <hannes@...essinduktion.org>
Subject: Re: [PATCH] Revert "ipv6: add mtu lock check in __ip6_rt_update_pmtu"
> > It's local system policy, how do I react to packets. If it doesn't
> > violate the min/max limits for ipv6 packets it emits onto the internet
> > I don't see this as something that can be seen as mandatory.
It does violate the max limit for ipv6 packets it emits onto the internet.
You're not allowed to emit > 1280 mtu packets without also supporting pmtu.
>
> And if you *truly* do want to violate internet standards you can
> indeed already achieve this behaviour by dropping incoming icmpv6
> packet too big errors (and there's lots of reasons why that is a bad
> idea...).
>
> I'll repeat what I said previously: this is a userspace visible
> regression in behaviour, of none or very questionable benefit.
>
> It results in TCP over IPv6 simply not working to destinations to
> which your locked mtu is higher then the real path mtu. This is why
> 'locked mtu' on IPv4 turns of the Don't Fragment bit - to allow
> fragmentation at intermediate routers. There is no such thing in
> IPv6.
> There is no DF bit, and there is no router fragmentation - all ipv6
> fragmentation is supposed to happen at the source host.
> This is why hosts must either use 1280 min guaranteed mtu or be
> responsive to pmtu errors. Otherwise things simply don't work.
Powered by blists - more mailing lists