Re: [PATCH v1 net 1/3] udp: Update reuse->has_conns under reuseport_lock.

[Paolo Abeni <pabeni@redhat.com>]

On Mon, 2022-10-10 at 10:43 -0700, Kuniyuki Iwashima wrote:
> When we call connect() for a UDP socket in a reuseport group, we have
> to update sk->sk_reuseport_cb->has_conns to 1.  Otherwise, the kernel
> could select a unconnected socket wrongly for packets sent to the
> connected socket.
> 
> However, the current way to set has_conns is illegal and possible to
> trigger that problem.  reuseport_has_conns() changes has_conns under
> rcu_read_lock(), which upgrades the RCU reader to the updater.  Then,
> it must do the update under the updater's lock, reuseport_lock, but
> it doesn't for now.
> 
> For this reason, there is a race below where we fail to set has_conns
> resulting in the wrong socket selection.  To avoid the race, let's split
> the reader and updater with proper locking.
> 
>  cpu1                               cpu2
> +----+                             +----+
> 
> __ip[46]_datagram_connect()        reuseport_grow()
> .                                  .
> > - reuseport_has_conns(sk, true)   |- more_reuse = __reuseport_alloc(more_socks_size)
> >  .                               |
> >  |- rcu_read_lock()
> >  |- reuse = rcu_dereference(sk->sk_reuseport_cb)
> >  |
> >  |                               |  /* reuse->has_conns == 0 here */
> >  |                               |- more_reuse->has_conns = reuse->has_conns
> >  |- reuse->has_conns = 1         |  /* more_reuse->has_conns SHOULD BE 1 HERE */
> >  |                               |
> >  |                               |- rcu_assign_pointer(reuse->socks[i]->sk_reuseport_cb,
> >  |                               |                     more_reuse)
> >  `- rcu_read_unlock()            `- kfree_rcu(reuse, rcu)
> > 
> > - sk->sk_state = TCP_ESTABLISHED
> 
> Fixes: acdcecc61285 ("udp: correct reuseport selection with connected sockets")
> Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
> ---
>  include/net/sock_reuseport.h | 23 +++++++++++++++++------
>  net/ipv4/datagram.c          |  2 +-
>  net/ipv4/udp.c               |  2 +-
>  net/ipv6/datagram.c          |  2 +-
>  net/ipv6/udp.c               |  2 +-
>  5 files changed, 21 insertions(+), 10 deletions(-)
> 
> diff --git a/include/net/sock_reuseport.h b/include/net/sock_reuseport.h
> index 473b0b0fa4ab..fe9779e6d90f 100644
> --- a/include/net/sock_reuseport.h
> +++ b/include/net/sock_reuseport.h
> @@ -43,21 +43,32 @@ struct sock *reuseport_migrate_sock(struct sock *sk,
>  extern int reuseport_attach_prog(struct sock *sk, struct bpf_prog *prog);
>  extern int reuseport_detach_prog(struct sock *sk);
>  
> -static inline bool reuseport_has_conns(struct sock *sk, bool set)
> +static inline bool reuseport_has_conns(struct sock *sk)
>  {
>  	struct sock_reuseport *reuse;
>  	bool ret = false;
>  
>  	rcu_read_lock();
>  	reuse = rcu_dereference(sk->sk_reuseport_cb);
> -	if (reuse) {
> -		if (set)
> -			reuse->has_conns = 1;
> -		ret = reuse->has_conns;
> -	}
> +	if (reuse && reuse->has_conns)
> +		ret = true;
>  	rcu_read_unlock();
>  
>  	return ret;
>  }
>  
> +static inline void reuseport_has_conns_set(struct sock *sk)
> +{
> +	struct sock_reuseport *reuse;
> +
> +	if (!rcu_access_pointer(sk->sk_reuseport_cb))
> +		return;
> +
> +	spin_lock(&reuseport_lock);
> +	reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
> +					  lockdep_is_held(&reuseport_lock));
> +	reuse->has_conns = 1;
> +	spin_unlock(&reuseport_lock);
> +}

Since the above is not super critical, it's probably better move it
into  sock_reuseport.c file and export it (to fix the build issue)

Cheers,

Paolo