Stephen Hemminger a écrit :

quoted

General/intuitive idea would be :

switch pointers to a newly allocated table (and zeroed counters)
wait one RCU grace period
collect/sum all counters of "old" table + (all cpus) into user provided table
restore previous table
wait one RCU grace period
disable_bh()
collect/sum all counters of "new and temporary" table (all cpus) and
reinject them into local cpu table (this cpu should not be interrupted)
enable_bh()

This way, "iptables -L" is not too expensive and doesnt block packet processing at all.

Pretty much what I said.

Cool then, sorry for misunderstanding your patch.