selftests/bpf: fix vlan handling in flow dissector program
authorStanislav Fomichev <sdf@google.com>
Mon, 1 Apr 2019 20:57:30 +0000 (13:57 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Wed, 3 Apr 2019 14:49:48 +0000 (16:49 +0200)
When we tail call PROG(VLAN) from parse_eth_proto we don't need to peek
back to handle vlan proto because we didn't adjust nhoff/thoff yet. Use
flow_keys->n_proto, that we set in parse_eth_proto instead and
properly increment nhoff as well.

Also, always use skb->protocol and don't look at skb->vlan_present.
skb->vlan_present indicates that vlan information is stored out-of-band
in skb->vlan_{tci,proto} and vlan header is already pulled from skb.
That means, skb->vlan_present == true is not relevant for BPF flow
dissector.

Add simple test cases with VLAN tagged frames:
  * single vlan for ipv4
  * double vlan for ipv6

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
tools/testing/selftests/bpf/prog_tests/flow_dissector.c
tools/testing/selftests/bpf/progs/bpf_flow.c

index bcbd928..fc818bc 100644 (file)
@@ -39,6 +39,58 @@ static struct bpf_flow_keys pkt_v6_flow_keys = {
        .n_proto = __bpf_constant_htons(ETH_P_IPV6),
 };
 
+#define VLAN_HLEN      4
+
+static struct {
+       struct ethhdr eth;
+       __u16 vlan_tci;
+       __u16 vlan_proto;
+       struct iphdr iph;
+       struct tcphdr tcp;
+} __packed pkt_vlan_v4 = {
+       .eth.h_proto = __bpf_constant_htons(ETH_P_8021Q),
+       .vlan_proto = __bpf_constant_htons(ETH_P_IP),
+       .iph.ihl = 5,
+       .iph.protocol = IPPROTO_TCP,
+       .iph.tot_len = __bpf_constant_htons(MAGIC_BYTES),
+       .tcp.urg_ptr = 123,
+       .tcp.doff = 5,
+};
+
+static struct bpf_flow_keys pkt_vlan_v4_flow_keys = {
+       .nhoff = VLAN_HLEN,
+       .thoff = VLAN_HLEN + sizeof(struct iphdr),
+       .addr_proto = ETH_P_IP,
+       .ip_proto = IPPROTO_TCP,
+       .n_proto = __bpf_constant_htons(ETH_P_IP),
+};
+
+static struct {
+       struct ethhdr eth;
+       __u16 vlan_tci;
+       __u16 vlan_proto;
+       __u16 vlan_tci2;
+       __u16 vlan_proto2;
+       struct ipv6hdr iph;
+       struct tcphdr tcp;
+} __packed pkt_vlan_v6 = {
+       .eth.h_proto = __bpf_constant_htons(ETH_P_8021AD),
+       .vlan_proto = __bpf_constant_htons(ETH_P_8021Q),
+       .vlan_proto2 = __bpf_constant_htons(ETH_P_IPV6),
+       .iph.nexthdr = IPPROTO_TCP,
+       .iph.payload_len = __bpf_constant_htons(MAGIC_BYTES),
+       .tcp.urg_ptr = 123,
+       .tcp.doff = 5,
+};
+
+static struct bpf_flow_keys pkt_vlan_v6_flow_keys = {
+       .nhoff = VLAN_HLEN * 2,
+       .thoff = VLAN_HLEN * 2 + sizeof(struct ipv6hdr),
+       .addr_proto = ETH_P_IPV6,
+       .ip_proto = IPPROTO_TCP,
+       .n_proto = __bpf_constant_htons(ETH_P_IPV6),
+};
+
 void test_flow_dissector(void)
 {
        struct bpf_flow_keys flow_keys;
@@ -68,5 +120,21 @@ void test_flow_dissector(void)
              err, errno, retval, duration, size, sizeof(flow_keys));
        CHECK_FLOW_KEYS("ipv6_flow_keys", flow_keys, pkt_v6_flow_keys);
 
+       err = bpf_prog_test_run(prog_fd, 10, &pkt_vlan_v4, sizeof(pkt_vlan_v4),
+                               &flow_keys, &size, &retval, &duration);
+       CHECK(size != sizeof(flow_keys) || err || retval != 1, "vlan_ipv4",
+             "err %d errno %d retval %d duration %d size %u/%lu\n",
+             err, errno, retval, duration, size, sizeof(flow_keys));
+       CHECK_FLOW_KEYS("vlan_ipv4_flow_keys", flow_keys,
+                       pkt_vlan_v4_flow_keys);
+
+       err = bpf_prog_test_run(prog_fd, 10, &pkt_vlan_v6, sizeof(pkt_vlan_v6),
+                               &flow_keys, &size, &retval, &duration);
+       CHECK(size != sizeof(flow_keys) || err || retval != 1, "vlan_ipv6",
+             "err %d errno %d retval %d duration %d size %u/%lu\n",
+             err, errno, retval, duration, size, sizeof(flow_keys));
+       CHECK_FLOW_KEYS("vlan_ipv6_flow_keys", flow_keys,
+                       pkt_vlan_v6_flow_keys);
+
        bpf_object__close(obj);
 }
index 284660f..f177c7a 100644 (file)
@@ -119,10 +119,7 @@ static __always_inline int parse_eth_proto(struct __sk_buff *skb, __be16 proto)
 SEC("flow_dissector")
 int _dissect(struct __sk_buff *skb)
 {
-       if (!skb->vlan_present)
-               return parse_eth_proto(skb, skb->protocol);
-       else
-               return parse_eth_proto(skb, skb->vlan_proto);
+       return parse_eth_proto(skb, skb->protocol);
 }
 
 /* Parses on IPPROTO_* */
@@ -336,15 +333,9 @@ PROG(VLAN)(struct __sk_buff *skb)
 {
        struct bpf_flow_keys *keys = skb->flow_keys;
        struct vlan_hdr *vlan, _vlan;
-       __be16 proto;
-
-       /* Peek back to see if single or double-tagging */
-       if (bpf_skb_load_bytes(skb, keys->thoff - sizeof(proto), &proto,
-                              sizeof(proto)))
-               return BPF_DROP;
 
        /* Account for double-tagging */
-       if (proto == bpf_htons(ETH_P_8021AD)) {
+       if (keys->n_proto == bpf_htons(ETH_P_8021AD)) {
                vlan = bpf_flow_dissect_get_header(skb, sizeof(*vlan), &_vlan);
                if (!vlan)
                        return BPF_DROP;
@@ -352,6 +343,7 @@ PROG(VLAN)(struct __sk_buff *skb)
                if (vlan->h_vlan_encapsulated_proto != bpf_htons(ETH_P_8021Q))
                        return BPF_DROP;
 
+               keys->nhoff += sizeof(*vlan);
                keys->thoff += sizeof(*vlan);
        }
 
@@ -359,6 +351,7 @@ PROG(VLAN)(struct __sk_buff *skb)
        if (!vlan)
                return BPF_DROP;
 
+       keys->nhoff += sizeof(*vlan);
        keys->thoff += sizeof(*vlan);
        /* Only allow 8021AD + 8021Q double tagging and no triple tagging.*/
        if (vlan->h_vlan_encapsulated_proto == bpf_htons(ETH_P_8021AD) ||