MTU-tests: Add tc_mtu_enforce.c

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>

MTU-tests/Makefile

@@ -2,6 +2,7 @@
 
 USER_TARGETS :=
 BPF_TARGETS := tc_bpf_inc_pkt_size
+BPF_TARGETS += tc_mtu_enforce
 EXTRA_DEPS := encap.h
 
 LIB_DIR = ../lib

MTU-tests/tc_mtu_enforce.c

@@ -0,0 +1,34 @@
+/* TC-BPF program that enforce MTU based on BPF-helper
+ */
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+
+/* Testing BPF-helper that is not upstream yet, requires us to define
+ * its call signature manually, which will only match our devel kernel.
+ *
+ * Generated signatur via command in kernel devel tree:
+ *   ./scripts/bpf_helpers_doc.py --header | grep bpf_check_mtu
+ */
+static int (*bpf_check_mtu)(void *ctx, __u32 ifindex, __u32 *mtu_len, __s32 len_diff, __u64 flags) = (void *) 163;
+
+enum  bpf_check_mtu_flags {
+        BPF_MTU_CHK_SEGS  = (1U << 0),
+};
+
+SEC("classifier") int tc_check_mtu(struct __sk_buff *skb)
+{
+	/* Main point behind this test is using flag BPF_MTU_CHK_SEGS,
+	 * because is will len check individual GSO/GRO segments in an SKB.
+	 * BPF and kernel often skip checks if "skb_is_gso()".
+	 */
+	__u64 flags = BPF_MTU_CHK_SEGS;
+	__u32 mtu_len = 0;
+	__s32 delta = 0;
+
+	if (bpf_check_mtu(skb, 0, &mtu_len, delta, flags)) {
+		return BPF_DROP;
+	}
+	return BPF_OK;
+}
+
+char _license[] SEC("license") = "GPL";