On Fri, 2023-06-23 at 10:08 +0100, David Howells wrote:

Jakub Kicinski [off-list ref] wrote:

quoted

If we can't reach instant agreement --
can you strategically separate out the minimal set of changes required
to just kill MSG_SENDPAGE_NOTLAST. IMHO it's worth getting that into
6.5.

Paolo Abeni [off-list ref] wrote:

quoted

Given all the above, and the late stage of the current devel cycle,
would you consider slicing down this series to just kill
MSG_SENDPAGE_NOTLAST, as Jakub suggested?

I could do that.

There is also another alternative.  I could just push the sendpage wrappers up
the stack into the higher-level callers.  Basically this:

int udp_sendpage(struct sock *sk, struct page *page, int offset,
		 size_t size, int flags)
{
	struct bio_vec bvec;
	struct msghdr msg = { .msg_flags = flags | MSG_SPLICE_PAGES };

	if (flags & MSG_SENDPAGE_NOTLAST)
		msg.msg_flags |= MSG_MORE;

	bvec_set_page(&bvec, page, size, offset);
	iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);
	return udp_sendmsg(sk, &msg, size);
}

and kill off sendpage and MSG_SENDPAGE_NOTLAST.

I'm unsure I follow the above ?!? I *thought* sendpage could be killed
even without patch 1/18 and 2/18, leaving some patches in this series
unmodified, and mangling those explicitly leveraging 1/18 to use
multiple sendmsg()s with different flags?

I haven't tried to code the above, but my wild guess/hope is that the
delta should be doable - ideally less then the other option.

Introducing slab support should still be possible later, with hopefully
less work.

Cheers,

Paolo