1 /* Copyright (c) 2015 PLUMgrid, http://plumgrid.com
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
7 #include <uapi/linux/bpf.h>
8 #include <uapi/linux/in.h>
9 #include <uapi/linux/if.h>
10 #include <uapi/linux/if_ether.h>
11 #include <uapi/linux/ip.h>
12 #include <uapi/linux/ipv6.h>
13 #include <uapi/linux/if_tunnel.h>
14 #include <uapi/linux/mpls.h>
15 #include <bpf/bpf_helpers.h>
16 #include "bpf_legacy.h"
18 #define IP_OFFSET 0x1FFF
20 #define PROG(F) SEC("socket/"__stringify(F)) int bpf_func_##F
23 __uint(type, BPF_MAP_TYPE_PROG_ARRAY);
24 __uint(key_size, sizeof(u32));
25 __uint(value_size, sizeof(u32));
26 __uint(max_entries, 8);
27 } jmp_table SEC(".maps");
34 /* Protocol dispatch routine. It tail-calls next BPF program depending
35 * on eth proto. Note, we could have used ...
37 * bpf_tail_call(skb, &jmp_table, proto);
39 * ... but it would need large prog_array and cannot be optimised given
40 * the map key is not static.
42 static inline void parse_eth_proto(struct __sk_buff *skb, u32 proto)
47 bpf_tail_call(skb, &jmp_table, PARSE_VLAN);
51 bpf_tail_call(skb, &jmp_table, PARSE_MPLS);
54 bpf_tail_call(skb, &jmp_table, PARSE_IP);
57 bpf_tail_call(skb, &jmp_table, PARSE_IPV6);
64 __be16 h_vlan_encapsulated_proto;
67 struct flow_key_record {
77 static inline int ip_is_fragment(struct __sk_buff *ctx, __u64 nhoff)
79 return load_half(ctx, nhoff + offsetof(struct iphdr, frag_off))
80 & (IP_MF | IP_OFFSET);
83 static inline __u32 ipv6_addr_hash(struct __sk_buff *ctx, __u64 off)
85 __u64 w0 = load_word(ctx, off);
86 __u64 w1 = load_word(ctx, off + 4);
87 __u64 w2 = load_word(ctx, off + 8);
88 __u64 w3 = load_word(ctx, off + 12);
90 return (__u32)(w0 ^ w1 ^ w2 ^ w3);
94 struct flow_key_record flow;
98 __uint(type, BPF_MAP_TYPE_ARRAY);
100 __type(value, struct globals);
101 __uint(max_entries, 32);
102 } percpu_map SEC(".maps");
104 /* user poor man's per_cpu until native support is ready */
105 static struct globals *this_cpu_globals(void)
107 u32 key = bpf_get_smp_processor_id();
109 return bpf_map_lookup_elem(&percpu_map, &key);
112 /* some simple stats for user space consumption */
119 __uint(type, BPF_MAP_TYPE_HASH);
120 __type(key, struct flow_key_record);
121 __type(value, struct pair);
122 __uint(max_entries, 1024);
123 } hash_map SEC(".maps");
125 static void update_stats(struct __sk_buff *skb, struct globals *g)
127 struct flow_key_record key = g->flow;
130 value = bpf_map_lookup_elem(&hash_map, &key);
132 __sync_fetch_and_add(&value->packets, 1);
133 __sync_fetch_and_add(&value->bytes, skb->len);
135 struct pair val = {1, skb->len};
137 bpf_map_update_elem(&hash_map, &key, &val, BPF_ANY);
141 static __always_inline void parse_ip_proto(struct __sk_buff *skb,
142 struct globals *g, __u32 ip_proto)
144 __u32 nhoff = skb->cb[0];
154 __u32 gre_flags = load_half(skb,
155 nhoff + offsetof(struct gre_hdr, flags));
156 __u32 gre_proto = load_half(skb,
157 nhoff + offsetof(struct gre_hdr, proto));
159 if (gre_flags & (GRE_VERSION|GRE_ROUTING))
163 if (gre_flags & GRE_CSUM)
165 if (gre_flags & GRE_KEY)
167 if (gre_flags & GRE_SEQ)
171 parse_eth_proto(skb, gre_proto);
175 parse_eth_proto(skb, ETH_P_IP);
178 parse_eth_proto(skb, ETH_P_IPV6);
182 g->flow.ports = load_word(skb, nhoff);
184 g->flow.ip_proto = ip_proto;
185 update_stats(skb, g);
192 PROG(PARSE_IP)(struct __sk_buff *skb)
194 struct globals *g = this_cpu_globals();
195 __u32 nhoff, verlen, ip_proto;
202 if (unlikely(ip_is_fragment(skb, nhoff)))
205 ip_proto = load_byte(skb, nhoff + offsetof(struct iphdr, protocol));
207 if (ip_proto != IPPROTO_GRE) {
208 g->flow.src = load_word(skb, nhoff + offsetof(struct iphdr, saddr));
209 g->flow.dst = load_word(skb, nhoff + offsetof(struct iphdr, daddr));
212 verlen = load_byte(skb, nhoff + 0/*offsetof(struct iphdr, ihl)*/);
213 nhoff += (verlen & 0xF) << 2;
216 parse_ip_proto(skb, g, ip_proto);
220 PROG(PARSE_IPV6)(struct __sk_buff *skb)
222 struct globals *g = this_cpu_globals();
223 __u32 nhoff, ip_proto;
230 ip_proto = load_byte(skb,
231 nhoff + offsetof(struct ipv6hdr, nexthdr));
232 g->flow.src = ipv6_addr_hash(skb,
233 nhoff + offsetof(struct ipv6hdr, saddr));
234 g->flow.dst = ipv6_addr_hash(skb,
235 nhoff + offsetof(struct ipv6hdr, daddr));
236 nhoff += sizeof(struct ipv6hdr);
239 parse_ip_proto(skb, g, ip_proto);
243 PROG(PARSE_VLAN)(struct __sk_buff *skb)
249 proto = load_half(skb, nhoff + offsetof(struct vlan_hdr,
250 h_vlan_encapsulated_proto));
251 nhoff += sizeof(struct vlan_hdr);
254 parse_eth_proto(skb, proto);
259 PROG(PARSE_MPLS)(struct __sk_buff *skb)
265 label = load_word(skb, nhoff);
266 nhoff += sizeof(struct mpls_label);
269 if (label & MPLS_LS_S_MASK) {
270 __u8 verlen = load_byte(skb, nhoff);
271 if ((verlen & 0xF0) == 4)
272 parse_eth_proto(skb, ETH_P_IP);
274 parse_eth_proto(skb, ETH_P_IPV6);
276 parse_eth_proto(skb, ETH_P_MPLS_UC);
283 int main_prog(struct __sk_buff *skb)
285 __u32 nhoff = ETH_HLEN;
286 __u32 proto = load_half(skb, 12);
289 parse_eth_proto(skb, proto);
293 char _license[] SEC("license") = "GPL";