[Patch net] netrom: fix a potential NULL pointer dereference

[Cong Wang <xiyou.wangcong@gmail.com>]

It is possible that the skb gets removed between skb_peek() and
skb_dequeue(). So we should just check the return value of
skb_dequeue().  Otherwise, skb_clone() may get a NULL pointer.

Technically, this should be protected by sock lock, but netrom
doesn't use it correctly. It is harder to fix sock lock than just
fixing this.

Reported-by: syzbot+9fe8e3f6c64aa5e5d82c@syzkaller.appspotmail.com
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
---
 net/netrom/nr_out.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/net/netrom/nr_out.c b/net/netrom/nr_out.c
index 44929657f5b7..9491a0c02bce 100644
--- a/net/netrom/nr_out.c
+++ b/net/netrom/nr_out.c
@@ -131,9 +131,6 @@ void nr_kick(struct sock *sk)
 	if (nr->condition & NR_COND_PEER_RX_BUSY)
 		return;
 
-	if (!skb_peek(&sk->sk_write_queue))
-		return;
-
 	start = (skb_peek(&nr->ack_queue) == NULL) ? nr->va : nr->vs;
 	end   = (nr->va + nr->window) % NR_MODULUS;
 
@@ -151,6 +148,8 @@ void nr_kick(struct sock *sk)
 	 * Dequeue the frame and copy it.
 	 */
 	skb = skb_dequeue(&sk->sk_write_queue);
+	if (!skb)
+		return;
 
 	do {
 		if ((skbn = skb_clone(skb, GFP_ATOMIC)) == NULL) {
-- 
2.21.0