Re: [PATCH 10/34] netfilter: ipset: Missing rcu protection in mtype_list() fixed

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Fri, May 08, 2015 at 10:15:14PM +0200, Jozsef Kadlecsik wrote:
> On Fri, 8 May 2015, Pablo Neira Ayuso wrote:
> 
> > On Thu, May 07, 2015 at 08:19:18PM +0200, Pablo Neira Ayuso wrote:
> > [...]
> > > I will come back to review the rcu patches, I still have concerns
> > > there.
> > > 
> > > If you're OK, I'll take as much as I can that is independent your RCU
> > > changes, so you don't need to resend this large patchset, but you will
> > > need to rebase though.
> > 
> > Unless you have any concern, I'm going to start by applying this batch
> > of 15 patches:
> > 
> > netfilter: ipset: Use better include files in xt_set.c
> > netfilter: ipset: Improve preprocessor macros checks
> > netfilter: ipset: Fix hashing for ipv6 sets
> > netfilter: ipset: Fix ext_*() macros
> > netfilter: ipset: Check for comment netlink attribute length
> > netfilter: ipset: Return bool values instead of int
> > netfilter: ipset: Use HOST_MASK literal to represent host address CIDR len
> > netfilter: ipset: Check IPSET_ATTR_PORT only once
> > netfilter: ipset: Return ipset error instead of bool
> > netfilter: ipset: Preprocessor directices cleanup
> > netfilter: ipset: No need to make nomatch bitfield
> > netfilter: ipset: Properly calculate extensions offsets and total length
> > netfilter: ipset: make ip_set_get_ip*_port to use skb_network_offset
> > netfilter: ipset: Give a better name to a macro in ip_set_core.c
> > netfilter: ipset: Fix sparse warning
> > 
> > It would be good if other small patches come in first place if
> > possible. Otherwise, I'd suggest you focus on a next batch with the
> > rcu conversion, that will be already quite large.
> 
> OK. I'll work on the rebasing after the patches above are applied.

I just pushed out this, you can start preparing a new batch.

Please, do the following changes in the next batch:

1) Collapse the RCU fix to the master patch, where it belongs.
2) Add the rcu_barrier() to the patch that starts requiring this.
3) Remove the synchronize_rcu() after list_add_rcu() since it's not
   needed.

Regarding rhashtable, I understand your implementation is simpler, but
it would be good if you can give it a further look as you said, I
guess other fellows will probably ask the same as I did.

Let me know if I can help, thanks for your patience Jozsef!