Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 20 Feb 2019 13:08:51 -0500
From:   Joel Fernandes <joel@joelfernandes.org>
To:     "Paul E. McKenney" <paulmck@linux.ibm.com>
Cc:     linux-kernel@vger.kernel.org, rcu@vger.kernel.org
Subject: Re: [RFC 1/5] net: rtnetlink: Fix incorrect RCU API usage
Message-ID: <20190220180851.GA97771@google.com>
References: <20190220040827.136184-1-joel@joelfernandes.org>
 <20190220040827.136184-2-joel@joelfernandes.org>
 <20190220164034.GM11787@linux.ibm.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190220164034.GM11787@linux.ibm.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Feb 20, 2019 at 08:40:34AM -0800, Paul E. McKenney wrote:
> On Tue, Feb 19, 2019 at 11:08:23PM -0500, Joel Fernandes (Google) wrote:
> > From: Joel Fernandes <joel@joelfernandes.org>
> > 
> > rtnl_register_internal() and rtnl_unregister_all tries to directly
> > dereference an RCU protected pointed outside RCU read side section.
> > While this is Ok to do since a lock is held, let us use the correct
> > API to avoid programmer bugs in the future.
> > 
> > This also fixes sparse warnings arising from not using RCU API.
> > 
> > net/core/rtnetlink.c:332:13: warning: incorrect type in assignment
> > (different address spaces) net/core/rtnetlink.c:332:13:    expected
> > struct rtnl_link **tab net/core/rtnetlink.c:332:13:    got struct
> > rtnl_link *[noderef] <asn:4>*<noident>
> > 
> > Signed-off-by: Joel Fernandes <joel@joelfernandes.org>
> 
> First, thank you for doing this!

No problem, it is my pleasure. It is just good to see these warnings/errors
show up (which I didn't anticipate when I first wrote the check) so we can
harden the kernel more fwiw.

> I was going to complain that these were update-side accesses, but it
> looks like rtnl_dereference() already handles both readers and updaters.
> 
> So looks good to me, but the maintainers of course have the final word.

Thanks!
Also my confidence level is a bit less for patches 4/5 and 5/5, could
you share your thoughts on those? The scheduler code seems to use
rcu_assign_pointer() in those where it seems a WRITE_ONCE() would just suffice.
In fact, in some cases I replaced with smp_store_release() just to be safe.
Speaking of which, do you feel those are legit uses of rcu_assign_pointer()
or would you expect rcu_assign_pointer() to be used only for RCU protected
pointers? I am hoping it is the latter since that is what the sparse check
expects (and RCU protected pointer being assigned to).

 - Joel


> 
> > ---
> >  net/core/rtnetlink.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
> > index 5ea1bed08ede..98be4b4818a9 100644
> > --- a/net/core/rtnetlink.c
> > +++ b/net/core/rtnetlink.c
> > @@ -188,7 +188,7 @@ static int rtnl_register_internal(struct module *owner,
> >  	msgindex = rtm_msgindex(msgtype);
> >  
> >  	rtnl_lock();
> > -	tab = rtnl_msg_handlers[protocol];
> > +	tab = rtnl_dereference(rtnl_msg_handlers[protocol]);
> >  	if (tab == NULL) {
> >  		tab = kcalloc(RTM_NR_MSGTYPES, sizeof(void *), GFP_KERNEL);
> >  		if (!tab)
> > @@ -329,7 +329,7 @@ void rtnl_unregister_all(int protocol)
> >  	BUG_ON(protocol < 0 || protocol > RTNL_FAMILY_MAX);
> >  
> >  	rtnl_lock();
> > -	tab = rtnl_msg_handlers[protocol];
> > +	tab = rtnl_dereference(rtnl_msg_handlers[protocol]);
> >  	if (!tab) {
> >  		rtnl_unlock();
> >  		return;
> > -- 
> > 2.21.0.rc0.258.g878e2cd30e-goog
> > 
>