netrom: fix a memory leak in nr_rx_frame()
authorCong Wang <xiyou.wangcong@gmail.com>
Thu, 27 Jun 2019 21:30:58 +0000 (14:30 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 4 Aug 2019 07:33:34 +0000 (09:33 +0200)
[ Upstream commit c8c8218ec5af5d2598381883acbefbf604e56b5e ]

When the skb is associated with a new sock, just assigning
it to skb->sk is not sufficient, we have to set its destructor
to free the sock properly too.

Reported-by: syzbot+d6636a36d3c34bd88938@syzkaller.appspotmail.com
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/netrom/af_netrom.c

index 046ae1caecea88cf3f41bcb850087c015ecdf59e..699fd83fd587e5910c8bf809d89bca93b9c02dfd 100644 (file)
@@ -870,7 +870,7 @@ int nr_rx_frame(struct sk_buff *skb, struct net_device *dev)
        unsigned short frametype, flags, window, timeout;
        int ret;
 
-       skb->sk = NULL;         /* Initially we don't know who it's for */
+       skb_orphan(skb);
 
        /*
         *      skb->data points to the netrom frame start
@@ -969,6 +969,7 @@ int nr_rx_frame(struct sk_buff *skb, struct net_device *dev)
        window = skb->data[20];
 
        skb->sk             = make;
+       skb->destructor     = sock_efree;
        make->sk_state      = TCP_ESTABLISHED;
 
        /* Fill in his circuit details */