On 14/11/17(Tue) 14:20, Alexandr Nedvedicky wrote: > Hello, > > I would go for NET_RLOCK() instead of NET_LOCK(), which is currently alias > to NET_WLOCK(). My point is the icmp6_reflect(), which is the workhorse for > icmp6_error(), is a typical 'READER-user' of network stack. It does not > change any network configuration. So it should be fine to let it run > as a reader.
I'd prefer we do not use any NET_RLOCK() until PF_LOCK() is in and enabled. Such code path are not performance critical :) > I'm sure mpi@ can provide authoritative OK here as nothing should go wrong > with pure NET_LOCK() you have in your diff. I'm ok with the diff as is.