Re: [PATCH v3 net-next RFC] Generic XDP

[Jesper Dangaard Brouer <brouer@redhat.com>]

On Fri, 14 Apr 2017 17:46:44 -0700
Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:

> But that's probably bad idea, since I was assuming that such ring
> reconfiguration can be made fast, which is unlikely.
> If it takes seconds to setup a ring, then drivers should just
> assume XDP_PACKET_HEADROOM, since at that time the program
> properties are unknown and in the future other programs will be loaded.
>
> Take a look at our current setup with slots for xdp_dump, ddos, lb
> programs. Only root program is attached at the begining.
> If the driver configures the ring for such empty program that would break
> dynamic addition of lb prog.
> The driver must not interrupt the traffic when user adds another
> prog to prog array. In such case XDP side doesn't even know that
> prog array is used. It's all happening purely on bpf side.

The bpf tail-call use-case is a very good example of why the verifier
cannot deduct the needed HEADROOM upfront.

Could we still make the verifier reject a program getting attached as a
tail-call when a too "low"/small HEADROOM have been setup? (to satisfy
programs needs)

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer