flow_dissector: fix clamping of BPF flow_keys for non-zero nhoff
authorStanislav Fomichev <sdf@google.com>
Mon, 1 Apr 2019 20:57:32 +0000 (13:57 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Wed, 3 Apr 2019 14:49:48 +0000 (16:49 +0200)
Don't allow BPF program to set flow_keys->nhoff to less than initial
value. We currently don't read the value afterwards in anything but
the tests, but it's still a good practice to return consistent
values to the test programs.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
net/core/flow_dissector.c

index 9b84250039df17bfce4b8c70aac92b813667f07c..94a450b2191a9e25ca79534f9caa68a442daa80c 100644 (file)
@@ -717,7 +717,8 @@ bool __skb_flow_bpf_dissect(struct bpf_prog *prog,
        /* Restore state */
        memcpy(cb, &cb_saved, sizeof(cb_saved));
 
-       flow_keys->nhoff = clamp_t(u16, flow_keys->nhoff, 0, skb->len);
+       flow_keys->nhoff = clamp_t(u16, flow_keys->nhoff,
+                                  skb_network_offset(skb), skb->len);
        flow_keys->thoff = clamp_t(u16, flow_keys->thoff,
                                   flow_keys->nhoff, skb->len);