Messages in this thread | | | Date | Tue, 29 Jul 2014 18:42:26 +0200 | From | Pablo Neira Ayuso <> | Subject | Re: [PATCH v3 net-next] net: filter: cleanup sk_* and bpf_* names |
| |
On Tue, Jul 29, 2014 at 08:55:04AM -0700, Alexei Starovoitov wrote: > > I don't think this is the right moment to add this, but we have to > > keep in mind that something similar to this will need to be > > accomodated in struct sk_filter at some point to avoid sloppy changes > > that may result in reintroducing code later on. > > I thought in v1 series you were arguing exactly about introducing them now... > ok, I will drop callbacks and keep refcnt,rcu,filter_size and bpf_prog pointer > in there. Sounds good?
Agreed.
> > Next step should be to wrap the specific bpf fields in struct > > bpf_prog in some clean way IMO, which was partially the aim of this > > patch. > > it seems your only objection is 'rcu_head' still being there and rebasing > on top of yours will fix it...
Almost. I just wanted to leave in place struct sk_filter for the coming up generalization, that structure should contain the refcnt, rcu_head and the struct bpf_prog after some of your follow up patches. Please, also leave sk_filter_charge/uncharge/get_filter whatever will provide the room the generalization under net/core/filter.c, not need to move these to kernel/bpf/
After my patch (and your follow up), we don't have sloppy usage of rcu_head for unattached filter anymore and I guess Willem is going to same save bytes in his iptables/bpf rules given that he can directly use bpf_prog instead of sk_filter.
| |