mptcp: fix possible deadlock in subflow_error_report

Christoph reported a possible deadlock while the TCP stack
destroys an unaccepted subflow due to an incoming reset: the
MPTCP socket error path tries to acquire the msk-level socket
lock while TCP still owns the listener socket accept queue
spinlock, and the reverse dependency already exists in the
TCP stack.

Note that the above is actually a lockdep false positive, as
the chain involves two separate sockets. A different per-socket
lockdep key will address the issue, but such a change will be
quite invasive.

Instead, we can simply stop earlier the socket error handling
for orphaned or unaccepted subflows, breaking the critical
lockdep chain. Error handling in such a scenario is a no-op.

Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Fixes: 15cc10453398 ("mptcp: deliver ssk errors to msk")
Cc: stable@vger.kernel.org
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/355
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

diff --git a/net/mptcp/subflow.c b/net/mptcp/subflow.c
index 4ae1a73..5070dc3 100644 (file)
--- a/net/mptcp/subflow.c
+++ b/net/mptcp/subflow.c
@@ -1432,6 +1432,13 @@ static void subflow_error_report(struct sock *ssk)
 {
        struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
 
+       /* bail early if this is a no-op, so that we avoid introducing a
+        * problematic lockdep dependency between TCP accept queue lock
+        * and msk socket spinlock
+        */
+       if (!sk->sk_socket)
+               return;
+
        mptcp_data_lock(sk);
        if (!sock_owned_by_user(sk))
                __mptcp_error_report(sk);