inet_connection_sock: clear inet_num out of destroy helper
authorPaolo Abeni <pabeni@redhat.com>
Thu, 4 Jun 2020 16:55:45 +0000 (18:55 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 4 Jun 2020 22:59:56 +0000 (15:59 -0700)
Clearing the 'inet_num' field is necessary and safe if and
only if the socket is not bound. The MPTCP protocol calls
the destroy helper on bound sockets, as tcp_v{4,6}_syn_recv_sock
completed successfully.

Move the clearing of such field out of the common code, otherwise
the MPTCP MP_JOIN error path will find the wrong 'inet_num' value
on socket disposal, __inet_put_port() will acquire the wrong lock
and bind_node removal could race with other modifiers possibly
corrupting the bind hash table.

Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Fixes: 729cd6436f35 ("mptcp: cope better with MP_JOIN failure")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/inet_connection_sock.h
net/ipv4/inet_connection_sock.c

index 2f1f8c3..e5b388f 100644 (file)
@@ -292,7 +292,6 @@ static inline void inet_csk_prepare_for_destroy_sock(struct sock *sk)
        /* The below has to be done to allow calling inet_csk_destroy_sock */
        sock_set_flag(sk, SOCK_DEAD);
        percpu_counter_inc(sk->sk_prot->orphan_count);
-       inet_sk(sk)->inet_num = 0;
 }
 
 void inet_csk_destroy_sock(struct sock *sk);
index f40b1b7..afaf582 100644 (file)
@@ -902,6 +902,7 @@ void inet_csk_prepare_forced_close(struct sock *sk)
        bh_unlock_sock(sk);
        sock_put(sk);
        inet_csk_prepare_for_destroy_sock(sk);
+       inet_sk(sk)->inet_num = 0;
 }
 EXPORT_SYMBOL(inet_csk_prepare_forced_close);