kcm: do not attach PF_KCM sockets to avoid deadlock

syzkaller had no problem to trigger a deadlock, attaching a KCM socket
to another one (or itself). (original syzkaller report was a very
confusing lockdep splat during a sendmsg())

It seems KCM claims to only support TCP, but no enforcement is done,
so we might need to add additional checks.

Fixes: ab7ac4eb9832 ("kcm: Kernel Connection Multiplexor module")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 4 insertions, 0 deletions

diff --git a/net/kcm/kcmsock.c b/net/kcm/kcmsock.c
index da49191f7ad0..4abf6287d7e1 100644
--- a/net/kcm/kcmsock.c
+++ b/net/kcm/kcmsock.c
@@ -1383,6 +1383,10 @@ static int kcm_attach(struct socket *sock, struct socket *csock,
         if (!csk)
                 return -EINVAL;
 
+        /* We must prevent loops or risk deadlock ! */
+        if (csk->sk_family == PF_KCM)
+                return -EOPNOTSUPP;
+
         psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
         if (!psock)
                 return -ENOMEM;