On Wed, Mar 23, 2016 at 12:04:28PM +0000, Yigal Reiss (yreiss) wrote: > >> - seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n", + > >> seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %5u %5u %2d\n", > > Problematic since it changes layout of a file we unfortunately > > have to view as uapi. I would prefer if we could leave the proc > > file alone and not add any new stats counters for this, unless > > there is a good argument for doing so. > > So my arguments are that there in order to fine tune a system it is > required to know about the existence and number of packets that went > under the radar. As I wrote ENOBUF does not answer all these needs. > I understand it is problematic to change uapi. Tried to minimize > incompatibility by keeping the order of arguments. I'll probably use > a patch to proc any way. Please let me know if you think there is a > point in proposing this patch or is it a "no-no" from kernel's > perspective.
I'd suggest you extend the existing nfnetlink_queue infrastructure so we retrieve these statistics through netlink, ie. add the NFQNL_MSG_STATS message. Then, extend nft so we can list them via: # nft list queues ... [ stats here ] ...
