flow_dissector: implements flow dissector BPF hook

Hello Petar Penkov,

This is a semi-automatic email about new static checker warnings.

The patch d58e468b1112: "flow_dissector: implements flow dissector BPF hook" from Sep 14, 2018, leads to the following Smatch complaint:

net/core/flow_dissector.c:797 __skb_flow_dissect()
error: we previously assumed 'skb' could be null (see line 758)

net/core/flow_dissector.c
757 rcu_read_lock();
758 attached = skb ? rcu_dereference(dev_net(skb->dev)->flow_dissector_prog)

                           ^^^

The patch adds a new check here.

759 : NULL;
760 if (attached) {
761 /* Note that even though the const qualifier is discarded
762 * throughout the execution of the BPF program, all changes(the
763 * control block) are reverted after the BPF program returns.
764 * Therefore, __skb_flow_dissect does not alter the skb.
765 */
766 struct bpf_flow_keys flow_keys = {};
767 struct bpf_skb_data_end cb_saved;
768 struct bpf_skb_data_end *cb;
769 u32 result;
770
771 cb = (struct bpf_skb_data_end *)skb->cb;
772
773 /* Save Control Block */
774 memcpy(&cb_saved, cb, sizeof(cb_saved));
775 memset(cb, 0, sizeof(cb_saved));
776
777 /* Pass parameters to the BPF program */
778 cb->qdisc_cb.flow_keys = &flow_keys;
779 flow_keys.nhoff = nhoff;
780
781 bpf_compute_data_pointers((struct sk_buff *)skb);
782 result = BPF_PROG_RUN(attached, skb);
783
784 /* Restore state */
785 memcpy(cb, &cb_saved, sizeof(cb_saved));
786
787 __skb_flow_bpf_to_target(&flow_keys, flow_dissector,
788 target_container);
789 key_control->thoff = min_t(u16, key_control->thoff, skb->len);
790 rcu_read_unlock();
791 return result = BPF_OK;
792 }
793 rcu_read_unlock();
794
795 if (dissector_uses_key(flow_dissector,
796 FLOW_DISSECTOR_KEY_ETH_ADDRS)) {

   797			struct ethhdr *eth = eth_hdr(skb);
                                                     ^^^

This old dereference isn't checked.

798 struct flow_dissector_key_eth_addrs *key_eth_addrs;
799

regards,
dan carpenter