xref: /openbmc/linux/samples/bpf/sockex2_kern.c (revision 82003e04)
1 #include <uapi/linux/bpf.h>
2 #include "bpf_helpers.h"
3 #include <uapi/linux/in.h>
4 #include <uapi/linux/if.h>
5 #include <uapi/linux/if_ether.h>
6 #include <uapi/linux/ip.h>
7 #include <uapi/linux/ipv6.h>
8 #include <uapi/linux/if_tunnel.h>
9 #define IP_MF		0x2000
10 #define IP_OFFSET	0x1FFF
11 
12 struct vlan_hdr {
13 	__be16 h_vlan_TCI;
14 	__be16 h_vlan_encapsulated_proto;
15 };
16 
17 struct bpf_flow_keys {
18 	__be32 src;
19 	__be32 dst;
20 	union {
21 		__be32 ports;
22 		__be16 port16[2];
23 	};
24 	__u16 thoff;
25 	__u8 ip_proto;
26 };
27 
28 static inline int proto_ports_offset(__u64 proto)
29 {
30 	switch (proto) {
31 	case IPPROTO_TCP:
32 	case IPPROTO_UDP:
33 	case IPPROTO_DCCP:
34 	case IPPROTO_ESP:
35 	case IPPROTO_SCTP:
36 	case IPPROTO_UDPLITE:
37 		return 0;
38 	case IPPROTO_AH:
39 		return 4;
40 	default:
41 		return 0;
42 	}
43 }
44 
45 static inline int ip_is_fragment(struct __sk_buff *ctx, __u64 nhoff)
46 {
47 	return load_half(ctx, nhoff + offsetof(struct iphdr, frag_off))
48 		& (IP_MF | IP_OFFSET);
49 }
50 
51 static inline __u32 ipv6_addr_hash(struct __sk_buff *ctx, __u64 off)
52 {
53 	__u64 w0 = load_word(ctx, off);
54 	__u64 w1 = load_word(ctx, off + 4);
55 	__u64 w2 = load_word(ctx, off + 8);
56 	__u64 w3 = load_word(ctx, off + 12);
57 
58 	return (__u32)(w0 ^ w1 ^ w2 ^ w3);
59 }
60 
61 static inline __u64 parse_ip(struct __sk_buff *skb, __u64 nhoff, __u64 *ip_proto,
62 			     struct bpf_flow_keys *flow)
63 {
64 	__u64 verlen;
65 
66 	if (unlikely(ip_is_fragment(skb, nhoff)))
67 		*ip_proto = 0;
68 	else
69 		*ip_proto = load_byte(skb, nhoff + offsetof(struct iphdr, protocol));
70 
71 	if (*ip_proto != IPPROTO_GRE) {
72 		flow->src = load_word(skb, nhoff + offsetof(struct iphdr, saddr));
73 		flow->dst = load_word(skb, nhoff + offsetof(struct iphdr, daddr));
74 	}
75 
76 	verlen = load_byte(skb, nhoff + 0/*offsetof(struct iphdr, ihl)*/);
77 	if (likely(verlen == 0x45))
78 		nhoff += 20;
79 	else
80 		nhoff += (verlen & 0xF) << 2;
81 
82 	return nhoff;
83 }
84 
85 static inline __u64 parse_ipv6(struct __sk_buff *skb, __u64 nhoff, __u64 *ip_proto,
86 			       struct bpf_flow_keys *flow)
87 {
88 	*ip_proto = load_byte(skb,
89 			      nhoff + offsetof(struct ipv6hdr, nexthdr));
90 	flow->src = ipv6_addr_hash(skb,
91 				   nhoff + offsetof(struct ipv6hdr, saddr));
92 	flow->dst = ipv6_addr_hash(skb,
93 				   nhoff + offsetof(struct ipv6hdr, daddr));
94 	nhoff += sizeof(struct ipv6hdr);
95 
96 	return nhoff;
97 }
98 
99 static inline bool flow_dissector(struct __sk_buff *skb, struct bpf_flow_keys *flow)
100 {
101 	__u64 nhoff = ETH_HLEN;
102 	__u64 ip_proto;
103 	__u64 proto = load_half(skb, 12);
104 	int poff;
105 
106 	if (proto == ETH_P_8021AD) {
107 		proto = load_half(skb, nhoff + offsetof(struct vlan_hdr,
108 							h_vlan_encapsulated_proto));
109 		nhoff += sizeof(struct vlan_hdr);
110 	}
111 
112 	if (proto == ETH_P_8021Q) {
113 		proto = load_half(skb, nhoff + offsetof(struct vlan_hdr,
114 							h_vlan_encapsulated_proto));
115 		nhoff += sizeof(struct vlan_hdr);
116 	}
117 
118 	if (likely(proto == ETH_P_IP))
119 		nhoff = parse_ip(skb, nhoff, &ip_proto, flow);
120 	else if (proto == ETH_P_IPV6)
121 		nhoff = parse_ipv6(skb, nhoff, &ip_proto, flow);
122 	else
123 		return false;
124 
125 	switch (ip_proto) {
126 	case IPPROTO_GRE: {
127 		struct gre_hdr {
128 			__be16 flags;
129 			__be16 proto;
130 		};
131 
132 		__u64 gre_flags = load_half(skb,
133 					    nhoff + offsetof(struct gre_hdr, flags));
134 		__u64 gre_proto = load_half(skb,
135 					    nhoff + offsetof(struct gre_hdr, proto));
136 
137 		if (gre_flags & (GRE_VERSION|GRE_ROUTING))
138 			break;
139 
140 		proto = gre_proto;
141 		nhoff += 4;
142 		if (gre_flags & GRE_CSUM)
143 			nhoff += 4;
144 		if (gre_flags & GRE_KEY)
145 			nhoff += 4;
146 		if (gre_flags & GRE_SEQ)
147 			nhoff += 4;
148 
149 		if (proto == ETH_P_8021Q) {
150 			proto = load_half(skb,
151 					  nhoff + offsetof(struct vlan_hdr,
152 							   h_vlan_encapsulated_proto));
153 			nhoff += sizeof(struct vlan_hdr);
154 		}
155 
156 		if (proto == ETH_P_IP)
157 			nhoff = parse_ip(skb, nhoff, &ip_proto, flow);
158 		else if (proto == ETH_P_IPV6)
159 			nhoff = parse_ipv6(skb, nhoff, &ip_proto, flow);
160 		else
161 			return false;
162 		break;
163 	}
164 	case IPPROTO_IPIP:
165 		nhoff = parse_ip(skb, nhoff, &ip_proto, flow);
166 		break;
167 	case IPPROTO_IPV6:
168 		nhoff = parse_ipv6(skb, nhoff, &ip_proto, flow);
169 		break;
170 	default:
171 		break;
172 	}
173 
174 	flow->ip_proto = ip_proto;
175 	poff = proto_ports_offset(ip_proto);
176 	if (poff >= 0) {
177 		nhoff += poff;
178 		flow->ports = load_word(skb, nhoff);
179 	}
180 
181 	flow->thoff = (__u16) nhoff;
182 
183 	return true;
184 }
185 
186 struct pair {
187 	long packets;
188 	long bytes;
189 };
190 
191 struct bpf_map_def SEC("maps") hash_map = {
192 	.type = BPF_MAP_TYPE_HASH,
193 	.key_size = sizeof(__be32),
194 	.value_size = sizeof(struct pair),
195 	.max_entries = 1024,
196 };
197 
198 SEC("socket2")
199 int bpf_prog2(struct __sk_buff *skb)
200 {
201 	struct bpf_flow_keys flow;
202 	struct pair *value;
203 	u32 key;
204 
205 	if (!flow_dissector(skb, &flow))
206 		return 0;
207 
208 	key = flow.dst;
209 	value = bpf_map_lookup_elem(&hash_map, &key);
210 	if (value) {
211 		__sync_fetch_and_add(&value->packets, 1);
212 		__sync_fetch_and_add(&value->bytes, skb->len);
213 	} else {
214 		struct pair val = {1, skb->len};
215 
216 		bpf_map_update_elem(&hash_map, &key, &val, BPF_ANY);
217 	}
218 	return 0;
219 }
220 
221 char _license[] SEC("license") = "GPL";
222