On Tue, Aug 20, 2019 at 05:00:26PM +0100, Edward Cree wrote: > On 20/08/2019 15:44, Pablo Neira Ayuso wrote: > > It looks to me this limitation is coming from tc pedit. > > > > Four actions to mangle an IPv6 address consume more memory when making > > the translation, and if you expect a lot of rules. > > Your change means that now every pedit uses four hw entries, even if it > was only meant to be a 32-bit mangle.
It makes no sense to me that matching an IPv6 address takes _one_ action, while mangling an IPv6 address takes _four_ actions. A consistent model for drivers is good to have. I can update tc pedit to generate one single action for offset consecutive packet editions, if that is the concern, I'll send a v2.
