xref: /linux/samples/bpf/xdp_tx_iptunnel_kern.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /* Copyright (c) 2016 Facebook
2  *
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.
6  *
7  * This program shows how to use bpf_xdp_adjust_head() by
8  * encapsulating the incoming packet in an IPv4/v6 header
9  * and then XDP_TX it out.
10  */
11 #define KBUILD_MODNAME "foo"
12 #include <uapi/linux/bpf.h>
13 #include <linux/in.h>
14 #include <linux/if_ether.h>
15 #include <linux/if_packet.h>
16 #include <linux/if_vlan.h>
17 #include <linux/ip.h>
18 #include <linux/ipv6.h>
19 #include <bpf/bpf_helpers.h>
20 #include "xdp_tx_iptunnel_common.h"
21 
22 struct {
23 	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
24 	__type(key, __u32);
25 	__type(value, __u64);
26 	__uint(max_entries, 256);
27 } rxcnt SEC(".maps");
28 
29 struct {
30 	__uint(type, BPF_MAP_TYPE_HASH);
31 	__type(key, struct vip);
32 	__type(value, struct iptnl_info);
33 	__uint(max_entries, MAX_IPTNL_ENTRIES);
34 } vip2tnl SEC(".maps");
35 
36 static __always_inline void count_tx(u32 protocol)
37 {
38 	u64 *rxcnt_count;
39 
40 	rxcnt_count = bpf_map_lookup_elem(&rxcnt, &protocol);
41 	if (rxcnt_count)
42 		*rxcnt_count += 1;
43 }
44 
45 static __always_inline int get_dport(void *trans_data, void *data_end,
46 				     u8 protocol)
47 {
48 	struct tcphdr *th;
49 	struct udphdr *uh;
50 
51 	switch (protocol) {
52 	case IPPROTO_TCP:
53 		th = (struct tcphdr *)trans_data;
54 		if (th + 1 > data_end)
55 			return -1;
56 		return th->dest;
57 	case IPPROTO_UDP:
58 		uh = (struct udphdr *)trans_data;
59 		if (uh + 1 > data_end)
60 			return -1;
61 		return uh->dest;
62 	default:
63 		return 0;
64 	}
65 }
66 
67 static __always_inline void set_ethhdr(struct ethhdr *new_eth,
68 				       const struct ethhdr *old_eth,
69 				       const struct iptnl_info *tnl,
70 				       __be16 h_proto)
71 {
72 	memcpy(new_eth->h_source, old_eth->h_dest, sizeof(new_eth->h_source));
73 	memcpy(new_eth->h_dest, tnl->dmac, sizeof(new_eth->h_dest));
74 	new_eth->h_proto = h_proto;
75 }
76 
77 static __always_inline int handle_ipv4(struct xdp_md *xdp)
78 {
79 	void *data_end = (void *)(long)xdp->data_end;
80 	void *data = (void *)(long)xdp->data;
81 	struct iptnl_info *tnl;
82 	struct ethhdr *new_eth;
83 	struct ethhdr *old_eth;
84 	struct iphdr *iph = data + sizeof(struct ethhdr);
85 	u16 *next_iph_u16;
86 	u16 payload_len;
87 	struct vip vip = {};
88 	int dport;
89 	u32 csum = 0;
90 	int i;
91 
92 	if (iph + 1 > data_end)
93 		return XDP_DROP;
94 
95 	dport = get_dport(iph + 1, data_end, iph->protocol);
96 	if (dport == -1)
97 		return XDP_DROP;
98 
99 	vip.protocol = iph->protocol;
100 	vip.family = AF_INET;
101 	vip.daddr.v4 = iph->daddr;
102 	vip.dport = dport;
103 	payload_len = ntohs(iph->tot_len);
104 
105 	tnl = bpf_map_lookup_elem(&vip2tnl, &vip);
106 	/* It only does v4-in-v4 */
107 	if (!tnl || tnl->family != AF_INET)
108 		return XDP_PASS;
109 
110 	/* The vip key is found.  Add an IP header and send it out */
111 
112 	if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct iphdr)))
113 		return XDP_DROP;
114 
115 	data = (void *)(long)xdp->data;
116 	data_end = (void *)(long)xdp->data_end;
117 
118 	new_eth = data;
119 	iph = data + sizeof(*new_eth);
120 	old_eth = data + sizeof(*iph);
121 
122 	if (new_eth + 1 > data_end ||
123 	    old_eth + 1 > data_end ||
124 	    iph + 1 > data_end)
125 		return XDP_DROP;
126 
127 	set_ethhdr(new_eth, old_eth, tnl, htons(ETH_P_IP));
128 
129 	iph->version = 4;
130 	iph->ihl = sizeof(*iph) >> 2;
131 	iph->frag_off =	0;
132 	iph->protocol = IPPROTO_IPIP;
133 	iph->check = 0;
134 	iph->tos = 0;
135 	iph->tot_len = htons(payload_len + sizeof(*iph));
136 	iph->daddr = tnl->daddr.v4;
137 	iph->saddr = tnl->saddr.v4;
138 	iph->ttl = 8;
139 
140 	next_iph_u16 = (u16 *)iph;
141 #pragma clang loop unroll(full)
142 	for (i = 0; i < sizeof(*iph) >> 1; i++)
143 		csum += *next_iph_u16++;
144 
145 	iph->check = ~((csum & 0xffff) + (csum >> 16));
146 
147 	count_tx(vip.protocol);
148 
149 	return XDP_TX;
150 }
151 
152 static __always_inline int handle_ipv6(struct xdp_md *xdp)
153 {
154 	void *data_end = (void *)(long)xdp->data_end;
155 	void *data = (void *)(long)xdp->data;
156 	struct iptnl_info *tnl;
157 	struct ethhdr *new_eth;
158 	struct ethhdr *old_eth;
159 	struct ipv6hdr *ip6h = data + sizeof(struct ethhdr);
160 	__u16 payload_len;
161 	struct vip vip = {};
162 	int dport;
163 
164 	if (ip6h + 1 > data_end)
165 		return XDP_DROP;
166 
167 	dport = get_dport(ip6h + 1, data_end, ip6h->nexthdr);
168 	if (dport == -1)
169 		return XDP_DROP;
170 
171 	vip.protocol = ip6h->nexthdr;
172 	vip.family = AF_INET6;
173 	memcpy(vip.daddr.v6, ip6h->daddr.s6_addr32, sizeof(vip.daddr));
174 	vip.dport = dport;
175 	payload_len = ip6h->payload_len;
176 
177 	tnl = bpf_map_lookup_elem(&vip2tnl, &vip);
178 	/* It only does v6-in-v6 */
179 	if (!tnl || tnl->family != AF_INET6)
180 		return XDP_PASS;
181 
182 	/* The vip key is found.  Add an IP header and send it out */
183 
184 	if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct ipv6hdr)))
185 		return XDP_DROP;
186 
187 	data = (void *)(long)xdp->data;
188 	data_end = (void *)(long)xdp->data_end;
189 
190 	new_eth = data;
191 	ip6h = data + sizeof(*new_eth);
192 	old_eth = data + sizeof(*ip6h);
193 
194 	if (new_eth + 1 > data_end ||
195 	    old_eth + 1 > data_end ||
196 	    ip6h + 1 > data_end)
197 		return XDP_DROP;
198 
199 	set_ethhdr(new_eth, old_eth, tnl, htons(ETH_P_IPV6));
200 
201 	ip6h->version = 6;
202 	ip6h->priority = 0;
203 	memset(ip6h->flow_lbl, 0, sizeof(ip6h->flow_lbl));
204 	ip6h->payload_len = htons(ntohs(payload_len) + sizeof(*ip6h));
205 	ip6h->nexthdr = IPPROTO_IPV6;
206 	ip6h->hop_limit = 8;
207 	memcpy(ip6h->saddr.s6_addr32, tnl->saddr.v6, sizeof(tnl->saddr.v6));
208 	memcpy(ip6h->daddr.s6_addr32, tnl->daddr.v6, sizeof(tnl->daddr.v6));
209 
210 	count_tx(vip.protocol);
211 
212 	return XDP_TX;
213 }
214 
215 SEC("xdp.frags")
216 int _xdp_tx_iptunnel(struct xdp_md *xdp)
217 {
218 	void *data_end = (void *)(long)xdp->data_end;
219 	void *data = (void *)(long)xdp->data;
220 	struct ethhdr *eth = data;
221 	__u16 h_proto;
222 
223 	if (eth + 1 > data_end)
224 		return XDP_DROP;
225 
226 	h_proto = eth->h_proto;
227 
228 	if (h_proto == htons(ETH_P_IP))
229 		return handle_ipv4(xdp);
230 	else if (h_proto == htons(ETH_P_IPV6))
231 
232 		return handle_ipv6(xdp);
233 	else
234 		return XDP_PASS;
235 }
236 
237 char _license[] SEC("license") = "GPL";
238