From: David Ahern <d...@cumulusnetworks.com> Date: Tue, 11 Apr 2017 12:57:59 -0600
> On 4/11/17 11:42 AM, David Miller wrote: >> David, if you have a specific case where it's absolutely impossible >> to resolve this when the library is converted to support extended >> ACKs, please mention it. > > I don't have a specific library in mind. It is more the disjoint nature > of a socket option and message parsing, and thinking through the API. Would you prefer it being passed via a control message at sendmsg() time? That would require more changes to userland programs, and also increase the per-request overhead in the kernel as it parses this CMSG every single time. We have a setsockopt for AF_PACKET that changes the entire layout of the packet mmap ring, nobody complains that libraries are hard to implement because of that right?
