Re: [PATCH][next] mptcp: Avoid NULL dereference in mptcp_getsockopt_subflow_addrs()

[Mat Martineau]

On Mon, 20 Sep 2021, Mat Martineau wrote:

> On Mon, 20 Sep 2021, Tim Gardner wrote:
>
> > Coverity complains of a possible NULL dereference in
> > mptcp_getsockopt_subflow_addrs():
> >
> > 861        } else if (sk->sk_family == AF_INET6) {
> >    	3. returned_null: inet6_sk returns NULL. [show details]
> >    	4. var_assigned: Assigning: np = NULL return value from inet6_sk.
> > 862                const struct ipv6_pinfo *np = inet6_sk(sk);
> >
> > Fix this by checking for NULL.
> >
> > Cc: Mat Martineau <mathew.j.martineau@xxxxxxxxxxxxxxx>
> > Cc: Matthieu Baerts <matthieu.baerts@xxxxxxxxxxxx>
> > Cc: "David S. Miller" <davem@xxxxxxxxxxxxx>
> > Cc: Jakub Kicinski <kuba@xxxxxxxxxx>
> > Cc: netdev@xxxxxxxxxxxxxxx
> > Cc: mptcp@xxxxxxxxxxxxxxx
> > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > Signed-off-by: Tim Gardner <tim.gardner@xxxxxxxxxxxxx>
> >
> > [ I'm not at all sure this is the right thing to do since the final result is to
> > return garbage to user space in mptcp_getsockopt_subflow_addrs(). Is this one
> > of those cases where inet6_sk() can't fail ?]
>
> Hi Tim -
>
> Thanks for noticing this and proposing a fix.
>
> As you commented, this isn't the right change to merge since 
> mptcp_getsockopt_subflow_addrs() would copy garbage.
>
> This block of code already checks that CONFIG_IPV6 is enabled, so the 
> question is whether sk_fullsock() would return false because the subflow is 
> in TCP_TIME_WAIT or TCP_NEW_SYN_RECV. The caller is iterating over sockets in 
> the MPTCP socket's conn_list, which does not contain request_socks (so there 
> are no sockets in the TCP_NEW_SYN_RECV state).
>
> TCP subflow sockets are normally removed from the conn_list before they are 
> closed by their parent MPTCP socket, but I need to double-check for corner 
> cases. I created a github issue to track this: 
> https://github.com/multipath-tcp/mptcp_net-next/issues/231

Tim,

Could you submit a v2 of this patch? Paolo took a look and the condition 
should not happen, but adding the NULL check would be a good idea and 
returning early as your patch does is ok. The data copied after the early 
return will be zeroed and look like the address family is AF_UNSPEC.

Could you add a

Fixes: https://github.com/multipath-tcp/mptcp_net-next/issues/231

tag and make the one change below?

> > ---
> > net/mptcp/sockopt.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/net/mptcp/sockopt.c b/net/mptcp/sockopt.c
> > index 8137cc3a4296..c89f2bedce79 100644
> > --- a/net/mptcp/sockopt.c
> > +++ b/net/mptcp/sockopt.c
> > @@ -861,6 +861,9 @@ static void mptcp_get_sub_addrs(const struct sock *sk, 
> > struct mptcp_subflow_addr
> > 	} else if (sk->sk_family == AF_INET6) {
> > 		const struct ipv6_pinfo *np = inet6_sk(sk);
> >
> > +		if (!np)

This could be

if (WARN_ON_ONCE(!np))

(as suggested by Paolo) to make it clear the condition is unexpected.

> > +			return;
> > +
> > 		a->sin6_local.sin6_family = AF_INET6;
> > 		a->sin6_local.sin6_port = inet->inet_sport;
> >
> > --
> > 2.33.0

Thanks,

--
Mat Martineau
Intel