svn commit: r323942 - head/sys/net

Jonathan Looney jonlooney at gmail.com
Sat Sep 23 23:33:28 UTC 2017


On Sat, Sep 23, 2017 at 4:37 AM, Bjoern A. Zeeb <
bzeeb-lists at lists.zabbadoz.net> wrote:

>
> Then this makes no sense as we don’t do LRO if forwarding is enabled on
> the machine;
> https://svnweb.freebsd.org/base/head/sys/netinet/tcp_lro.c?
> annotate=317390#l645


Yes, that is true. However, this change still makes a difference.
Previously, if LRO was not enabled or the packet was not eligible for LRO,
the iflib code would call ifp->if_input() once for each packet. Now, the
iflib code will build a chain of packets for which it couldn't do LRO and
call ifp->if_input() once for the entire chain.

(I agree that was not obvious from the rather short commit message and the
diff in the email. The lack of comments or meaningful variable names did
not help to alleviate the confusion. Nonetheless, when I looked at the diff
with enough surrounding context, it became clear.)

Jonathan


More information about the svn-src-all mailing list