Re: [PATCH net 2/4] bonding: fix null pointer deref in bond_ipsec_offload_ok

[Hangbin Liu <liuhangbin@gmail.com>]

On Mon, Aug 19, 2024 at 10:25:37AM +0300, Nikolay Aleksandrov wrote:
> On 19/08/2024 05:53, Hangbin Liu wrote:
> > On Fri, Aug 16, 2024 at 02:48:11PM +0300, Nikolay Aleksandrov wrote:
> >> We must check if there is an active slave before dereferencing the pointer.
> >>
> >> Fixes: 18cb261afd7b ("bonding: support hardware encryption offload to slaves")
> >> Signed-off-by: Nikolay Aleksandrov <razor@blackwall.org>
> >> ---
> >>  drivers/net/bonding/bond_main.c | 2 ++
> >>  1 file changed, 2 insertions(+)
> >>
> >> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> >> index 85b5868deeea..65ddb71eebcd 100644
> >> --- a/drivers/net/bonding/bond_main.c
> >> +++ b/drivers/net/bonding/bond_main.c
> >> @@ -604,6 +604,8 @@ static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
> >>  	bond = netdev_priv(bond_dev);
> >>  	rcu_read_lock();
> >>  	curr_active = rcu_dereference(bond->curr_active_slave);
> >> +	if (!curr_active)
> >> +		goto out;
> >>  	real_dev = curr_active->dev;
> >>  
> >>  	if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
> >> -- 
> >> 2.44.0
> >>
> > 
> > BTW, the bond_ipsec_offload_ok() only checks !xs->xso.real_dev, should we also
> > add WARN_ON(xs->xso.real_dev != slave->dev) checking?
> > 
> > Thanks
> > Hangbin
> 
> We could, but not a warn_on() because I bet it can be easily triggered
> by changing the active slave in parallel. real_dev is read without a

OK, maybe a pr_warn or salve_warn()?

> lock here so we cannot guarantee a sane state if policies are changed
> under us. I think the callback should handle it by checking that the
> new device doesn't have the policy setup yet, because the case happens
> when an active slave changes which means policies are about to be
> installed on the new one.

Hmm, how to check if a device has policy setup except ipsec->xs->xso.real_dev?

Hangbin