Re: [PATCH net v3] mptcp: pm: Fix uaf in __timer_delete_sync

[Edward Adam Davis]

On Mon, 9 Sep 2024 15:07:21 +0200, Paolo Abeni wrote:
> On 9/5/24 14:27, Edward Adam Davis wrote:
> > There are two paths to access mptcp_pm_del_add_timer, result in a race
> > condition:
> > 
> >       CPU1				CPU2
> >       ====                               ====
> >       net_rx_action
> >       napi_poll                          netlink_sendmsg
> >       __napi_poll                        netlink_unicast
> >       process_backlog                    netlink_unicast_kernel
> >       __netif_receive_skb                genl_rcv
> >       __netif_receive_skb_one_core       netlink_rcv_skb
> >       NF_HOOK                            genl_rcv_msg
> >       ip_local_deliver_finish            genl_family_rcv_msg
> >       ip_protocol_deliver_rcu            genl_family_rcv_msg_doit
> >       tcp_v4_rcv                         mptcp_pm_nl_flush_addrs_doit
> >       tcp_v4_do_rcv                      mptcp_nl_remove_addrs_list
> >       tcp_rcv_established                mptcp_pm_remove_addrs_and_subflows
> >       tcp_data_queue                     remove_anno_list_by_saddr
> >       mptcp_incoming_options             mptcp_pm_del_add_timer
> >       mptcp_pm_del_add_timer             kfree(entry)
> > 
> > In remove_anno_list_by_saddr(running on CPU2), after leaving the critical
> > zone protected by "pm.lock", the entry will be released, which leads to the
> > occurrence of uaf in the mptcp_pm_del_add_timer(running on CPU1).
> > 
> > Keeping a reference to add_timer inside the lock, and calling
> > sk_stop_timer_sync() with this reference, instead of "entry->add_timer".
> > 
> > Fixes: 00cfd77b9063 ("mptcp: retransmit ADD_ADDR when timeout")
> > Cc: stable@xxxxxxxxxxxxxxx
> > Reported-and-tested-by: syzbot+f3a31fb909db9b2a5c4d@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Closes: https://syzkaller.appspot.com/bug?extid=f3a31fb909db9b2a5c4d
> > Signed-off-by: Matthieu Baerts (NGI0) <matttbe@xxxxxxxxxx>
> > Signed-off-by: Edward Adam Davis <eadavis@xxxxxx>
> > ---
> >   net/mptcp/pm_netlink.c | 14 ++++++++++----
> >   1 file changed, 10 insertions(+), 4 deletions(-)
> > 
> > diff --git a/net/mptcp/pm_netlink.c b/net/mptcp/pm_netlink.c
> > index 3e4ad801786f..7ddb373cc6ad 100644
> > --- a/net/mptcp/pm_netlink.c
> > +++ b/net/mptcp/pm_netlink.c
> > @@ -329,17 +329,21 @@ struct mptcp_pm_add_entry *
> >   mptcp_pm_del_add_timer(struct mptcp_sock *msk,
> >   		       const struct mptcp_addr_info *addr, bool check_id)
> >   {
> > -	struct mptcp_pm_add_entry *entry;
> >   	struct sock *sk = (struct sock *)msk;
> > +	struct timer_list *add_timer = NULL;
> > +	struct mptcp_pm_add_entry *entry;
> >   
> >   	spin_lock_bh(&msk->pm.lock);
> >   	entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
> > -	if (entry && (!check_id || entry->addr.id == addr->id))
> > +	if (entry && (!check_id || entry->addr.id == addr->id)) {
> >   		entry->retrans_times = ADD_ADDR_RETRANS_MAX;
> > +		add_timer = &entry->add_timer;
> > +	}
> >   	spin_unlock_bh(&msk->pm.lock);
> >   
> > -	if (entry && (!check_id || entry->addr.id == addr->id))
> > -		sk_stop_timer_sync(sk, &entry->add_timer);
> > +	/* no lock, because sk_stop_timer_sync() is calling del_timer_sync() */
> > +	if (add_timer)
> > +		sk_stop_timer_sync(sk, add_timer);
> >   
> >   	return entry;
> >   }
> > @@ -1430,8 +1434,10 @@ static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
> >   
> >   	entry = mptcp_pm_del_add_timer(msk, addr, false);
> >   	if (entry) {
> > +		spin_lock_bh(&msk->pm.lock);
> >   		list_del(&entry->list);
> >   		kfree(entry);
> > +		spin_unlock_bh(&msk->pm.lock);
> 
> I'm sorry for the late feedback.
> 
> I think this is not enough to fix races for good, i.e.
> 
> mptcp_nl_remove_subflow_and_signal_addr() -> mptcp_pm_remove_anno_addr()
> -> remove_anno_list_by_saddr()
> 
> could race with:
> 
> mptcp_pm_remove_addrs() -> remove_anno_list_by_saddr()
> 
> and both CPUs could see the same 'entry' returned by
> mptcp_pm_del_add_timer().
Yes, you are right.
> 
> I think the list_del() in remove_anno_list_by_saddr() should moved under
> the pm lock protection inside mptcp_pm_del_add_timer(), and no need to 
> add spin_lock_bh() around the kfree call.
Agreed, I will update the patch.

BR,
Edward