net/smc: fix refcount non-blocking connect() -part 2

[ Upstream commit 6d6dd528d5af05dc2d0c773951ed68d630a0c3f1 ]

If an SMC socket is immediately terminated after a non-blocking connect()
has been called, a memory leak is possible.
Due to the sock_hold move in
commit 301428ea3708 ("net/smc: fix refcounting for non-blocking connect()")
an extra sock_put() is needed in smc_connect_work(), if the internal
TCP socket is aborted and cancels the sk_stream_wait_connect() of the
connect worker.

Reported-by: syzbot+4b73ad6fc767e576e275@syzkaller.appspotmail.com
Fixes: 301428ea3708 ("net/smc: fix refcounting for non-blocking connect()")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/net/smc/af_smc.c b/net/smc/af_smc.c
index 5aec060a8581147807dd642e03aca796fbe3eded..737b49909a7af7e8bdd58fbaa4bc73aa007ad564 100644 (file)
--- a/net/smc/af_smc.c
+++ b/net/smc/af_smc.c
@@ -796,6 +796,7 @@ static void smc_connect_work(struct work_struct *work)
                        smc->sk.sk_err = EPIPE;
                else if (signal_pending(current))
                        smc->sk.sk_err = -sock_intr_errno(timeo);
+               sock_put(&smc->sk); /* passive closing */
                goto out;
        }