On Mon, 20 Sep 2021 23:01:48 +0200 Toke Høiland-Jørgensen wrote:

quoted

In fact I don't think there is anything infra can do better for
flushing than the prog itself:

	bool mod = false;

	ptr = bpf_header_pointer(...);
	...
	if (some_cond(...)) {
		change_packet(...);
		mod = true;
	}
	...
	if (mod)  

to have an additional check like:

if (mod && ptr == stack)

(or something to that effect). No?

Good point. Do you think we should have the kernel add/inline this
optimization or have the user do it explicitly.

The draft API was:

void *xdp_mb_pointer_flush(struct xdp_buff *xdp_md, u32 flags, 
                           u32 offset, u32 len, void *stack_buf)

Which does not take the ptr returned by header_pointer(), but that's
easy to add (well, easy other than the fact it'd be the 6th arg).

BTW I drafted the API this way to cater to the case where flush()
is called without a prior call to header_pointer(). For when packet
trailer or header is populated directly from a map value. Dunno if
that's actually useful, either.