Re: [PATCH nf-next v2 1/1] netfilter: helper: Remove useless rcu lock when get expectfn

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

On Wed, Mar 29, 2017 at 06:29:10PM +0800, Gao Feng wrote:
> Hi Pablo,
> 
> > -----Original Message-----
> > From: Pablo Neira Ayuso [mailto:pablo@netfilter.org]
> > Sent: Wednesday, March 29, 2017 6:08 PM
> > To: gfree.wind@foxmail.com
> > Cc: netfilter-devel@vger.kernel.org; Gao Feng <fgao@ikuai8.com>
> > Subject: Re: [PATCH nf-next v2 1/1] netfilter: helper: Remove useless rcu
> lock
> > when get expectfn
> > 
> > On Wed, Mar 22, 2017 at 10:15:02AM +0800, gfree.wind@foxmail.com wrote:
> > > From: Gao Feng <fgao@ikuai8.com>
> > >
> > > Because these two functions return the nf_ct_helper_expectfn pointer
> > > which should be protected by rcu lock. So it should makes sure the
> > > caller should hold the rcu lock, not inside these functions.
> > >
> > > Signed-off-by: Gao Feng <fgao@ikuai8.com>
> > > ---
> > >  v2: Shorter subject, per Pablo
> > >  v1: Initial version
> > >
> > >  net/netfilter/nf_conntrack_helper.c | 6 ++----
> > >  1 file changed, 2 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/net/netfilter/nf_conntrack_helper.c
> > > b/net/netfilter/nf_conntrack_helper.c
> > > index 6dc44d9..bce3d1f 100644
> > > --- a/net/netfilter/nf_conntrack_helper.c
> > > +++ b/net/netfilter/nf_conntrack_helper.c
> > > @@ -311,38 +311,36 @@ void nf_ct_helper_expectfn_unregister(struct
> > > nf_ct_helper_expectfn *n)  }
> > > EXPORT_SYMBOL_GPL(nf_ct_helper_expectfn_unregister);
> > >
> > > +/* Caller should hold the rcu lock */
> > >  struct nf_ct_helper_expectfn *
> > >  nf_ct_helper_expectfn_find_by_name(const char *name)  {
> > >  	struct nf_ct_helper_expectfn *cur;
> > >  	bool found = false;
> > >
> > > -	rcu_read_lock();
> > >  	list_for_each_entry_rcu(cur, &nf_ct_helper_expectfn_list, head) {
> > >  		if (!strcmp(cur->name, name)) {
> > >  			found = true;
> > >  			break;
> > >  		}
> > >  	}
> > > -	rcu_read_unlock();
> > >  	return found ? cur : NULL;
> > >  }
> > >  EXPORT_SYMBOL_GPL(nf_ct_helper_expectfn_find_by_name);
> > 
> > nf_ct_helper_expectfn_find_by_name() is called from ctnetlink, via
> > ctnetlink_create_expect() and rcu read side lock is not held there.
> There are two reasons.
> 1. The rcu lock would be added in my patch " netfilter: helper: Add the rcu
> lock when call __nf_conntrack_helper_find" for nf
> https://patchwork.ozlabs.org/patch/741865/.

If you have interdependencies between two patches like this, it's
better to make it in one single go.

> So the ctnetlink_create_expect would hold the rcu lock after apply that
> patch.
> 
> 2. Because these two functions return one pointer which needs RCU lock, so
> the caller must hold rcu lock.
> Or it still meets one error even though there is one rcu lock in these two
> functions.
> Because the memory which the returned pointer point to would be freed
> already after rcu_read_unlock.
> So the rcu lock is unnecessary in these functions.

That's right. You're fixing up a real problem, no doubt.

I'm just questioning that I think that if you are fixing up rcu
locking, which seems to be the case, you just do it in one single
patch.

Thanks!