On 2019/08/15 2:07, Stanislav Fomichev wrote:

On 08/13, Toshiaki Makita wrote:

quoted

* Implementation

xdp_flow makes use of UMH to load an eBPF program for XDP, similar to
bpfilter. The difference is that xdp_flow does not generate the eBPF
program dynamically but a prebuilt program is embedded in UMH. This is
mainly because flow insertion is considerably frequent. If we generate
and load an eBPF program on each insertion of a flow, the latency of the
first packet of ping in above test will incease, which I want to avoid.

Can this be instead implemented with a new hook that will be called
for TC events? This hook can write to perf event buffer and control
plane will insert/remove/modify flow tables in the BPF maps (contol
plane will also install xdp program).

Why do we need UMH? What am I missing?

So you suggest doing everything in xdp_flow kmod?
I also thought about that. There are two phases so let's think about them separately.

1) TC block (qdisc) creation / eBPF load

I saw eBPF maintainers repeatedly saying eBPF program loading needs to be
done from userland, not from kernel, to run the verifier for safety.
However xdp_flow eBPF program is prebuilt and embedded in kernel so we may
allow such programs to be loaded from kernel? I currently don't have the will
to make such an API as loading can be done with current UMH mechanism.

2) flow insertion / eBPF map update

Not sure if this needs to be done from userland. One concern is that eBPF maps can
be modified by unrelated processes and we need to handle all unexpected state of maps.
Such handling tends to be difficult and may cause unexpected kernel behavior.
OTOH updating maps from kmod may reduces the latency of flow insertion drastically.

Alexei, Daniel, what do you think?

Toshiaki Makita