On Tue, Oct 27, 2015 at 9:34 AM, Hannes Frederic Sowa
[off-list ref] wrote:

On Tue, Oct 27, 2015, at 17:04, Tom Herbert wrote:

quoted

On Tue, Oct 27, 2015 at 8:02 AM, Hannes Frederic Sowa
[off-list ref] wrote:

quoted

We cannot reliable calculate packet size on MSG_MORE corked sockets
and thus cannot decide if they are going to be fragmented later on,
so better not use CHECKSUM_PARTIAL in the first place.

MSG_MORE should be independent of checksum offload. If packet is
fragmented the fix in ip_output will ensure that skb_checksum_help is
properly called.

The probability is that we are going to fragment if MSG_MORE is set,
because exceeding link mtu is quite probable, see e.g. NFS use case. Why
not simply use the csum functions during copy-in in that case? It makes
much more sense to me.

For datagram sockets MSG_MORE means that more datagrams will be sent,
it's not used to incrementally add data to a datagram already queued
(SEQPACKET with EOR is for that).

I don't see a reason to test for fragment length at all, then.

Bye,
Hannes