Re: [PATCH xfrm-next v7 6/8] xfrm: speed-up lookup of HW policies

[Leon Romanovsky <leon@kernel.org>]

On Mon, Nov 21, 2022 at 01:10:40PM +0100, Steffen Klassert wrote:
> On Mon, Nov 21, 2022 at 01:34:30PM +0200, Leon Romanovsky wrote:
> > 
> > Sorry, my bad. But why can't we drop all packets that don't have HW
> > state? Why do we need to add larval?
> 
> The first packet of a flow tiggers an acquire and inserts a larval
> state. On a traffic triggered connection, we need this to get
> a state with keys installed.
> 
> We need this larval state then, because that tells us we sent already an
> acquire to userspace. All subsequent packets of that flow will be
> dropped without sending another acquire. Otherwise each subsequent
> packet will generate another acquire until the keys are negotiated.
> If a flow starts sending on a high rate, this would be not so nice
> for userspace :)

The thing is that this SW acquire flow is a fraction case, as it applies
to locally generated traffic.

In my mind, there are other cases, like eswitch mode and tunnel mode. In
these cases, the packets are arrived to HW without even passing SW stack.

What we want to do is to catch in HW all TX packets, which don't have SAs
(applicable for all types of traffic), mark them and route back to the
driver. The driver will be responsible to talk with XFRM core to
generate acquire.

The same logic of rerouting packets is required for audit and will be done
later. Right now, we rely on *swan implementations which configure everything
in advance.

Also larval is default to 1 (drop) in all distros.

I hope that this larval/acquire is not must for this series to be merged.
And it is going to be implemented later as I'm assigned to work on this
offload feature till feature complete :).

Thanks