tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright (c) 2022 Meta */
3#include <stddef.h>
4#include <string.h>
5#include <linux/bpf.h>
6#include <linux/if_ether.h>
7#include <linux/if_packet.h>
8#include <linux/ip.h>
9#include <linux/ipv6.h>
10#include <linux/in.h>
11#include <linux/udp.h>
12#include <linux/tcp.h>
13#include <linux/pkt_cls.h>
14#include <sys/socket.h>
15#include <bpf/bpf_helpers.h>
16#include <bpf/bpf_endian.h>
17#include "test_iptunnel_common.h"
18#include "bpf_kfuncs.h"
19
20const size_t tcphdr_sz = sizeof(struct tcphdr);
21const size_t udphdr_sz = sizeof(struct udphdr);
22const size_t ethhdr_sz = sizeof(struct ethhdr);
23const size_t iphdr_sz = sizeof(struct iphdr);
24const size_t ipv6hdr_sz = sizeof(struct ipv6hdr);
25
26struct {
27 __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
28 __uint(max_entries, 256);
29 __type(key, __u32);
30 __type(value, __u64);
31} rxcnt SEC(".maps");
32
33struct {
34 __uint(type, BPF_MAP_TYPE_HASH);
35 __uint(max_entries, MAX_IPTNL_ENTRIES);
36 __type(key, struct vip);
37 __type(value, struct iptnl_info);
38} vip2tnl SEC(".maps");
39
40static __always_inline void count_tx(__u32 protocol)
41{
42 __u64 *rxcnt_count;
43
44 rxcnt_count = bpf_map_lookup_elem(&rxcnt, &protocol);
45 if (rxcnt_count)
46 *rxcnt_count += 1;
47}
48
49static __always_inline int get_dport(void *trans_data, __u8 protocol)
50{
51 struct tcphdr *th;
52 struct udphdr *uh;
53
54 switch (protocol) {
55 case IPPROTO_TCP:
56 th = (struct tcphdr *)trans_data;
57 return th->dest;
58 case IPPROTO_UDP:
59 uh = (struct udphdr *)trans_data;
60 return uh->dest;
61 default:
62 return 0;
63 }
64}
65
66static __always_inline void set_ethhdr(struct ethhdr *new_eth,
67 const struct ethhdr *old_eth,
68 const struct iptnl_info *tnl,
69 __be16 h_proto)
70{
71 memcpy(new_eth->h_source, old_eth->h_dest, sizeof(new_eth->h_source));
72 memcpy(new_eth->h_dest, tnl->dmac, sizeof(new_eth->h_dest));
73 new_eth->h_proto = h_proto;
74}
75
76static __always_inline int handle_ipv4(struct xdp_md *xdp, struct bpf_dynptr *xdp_ptr)
77{
78 __u8 eth_buffer[ethhdr_sz + iphdr_sz + ethhdr_sz];
79 __u8 iph_buffer_tcp[iphdr_sz + tcphdr_sz];
80 __u8 iph_buffer_udp[iphdr_sz + udphdr_sz];
81 struct bpf_dynptr new_xdp_ptr;
82 struct iptnl_info *tnl;
83 struct ethhdr *new_eth;
84 struct ethhdr *old_eth;
85 struct iphdr *iph;
86 __u16 *next_iph;
87 __u16 payload_len;
88 struct vip vip = {};
89 int dport;
90 __u32 csum = 0;
91 int i;
92
93 __builtin_memset(eth_buffer, 0, sizeof(eth_buffer));
94 __builtin_memset(iph_buffer_tcp, 0, sizeof(iph_buffer_tcp));
95 __builtin_memset(iph_buffer_udp, 0, sizeof(iph_buffer_udp));
96
97 if (ethhdr_sz + iphdr_sz + tcphdr_sz > xdp->data_end - xdp->data)
98 iph = bpf_dynptr_slice(xdp_ptr, ethhdr_sz, iph_buffer_udp, sizeof(iph_buffer_udp));
99 else
100 iph = bpf_dynptr_slice(xdp_ptr, ethhdr_sz, iph_buffer_tcp, sizeof(iph_buffer_tcp));
101
102 if (!iph)
103 return XDP_DROP;
104
105 dport = get_dport(iph + 1, iph->protocol);
106 if (dport == -1)
107 return XDP_DROP;
108
109 vip.protocol = iph->protocol;
110 vip.family = AF_INET;
111 vip.daddr.v4 = iph->daddr;
112 vip.dport = dport;
113 payload_len = bpf_ntohs(iph->tot_len);
114
115 tnl = bpf_map_lookup_elem(&vip2tnl, &vip);
116 /* It only does v4-in-v4 */
117 if (!tnl || tnl->family != AF_INET)
118 return XDP_PASS;
119
120 if (bpf_xdp_adjust_head(xdp, 0 - (int)iphdr_sz))
121 return XDP_DROP;
122
123 bpf_dynptr_from_xdp(xdp, 0, &new_xdp_ptr);
124 new_eth = bpf_dynptr_slice_rdwr(&new_xdp_ptr, 0, eth_buffer, sizeof(eth_buffer));
125 if (!new_eth)
126 return XDP_DROP;
127
128 iph = (struct iphdr *)(new_eth + 1);
129 old_eth = (struct ethhdr *)(iph + 1);
130
131 set_ethhdr(new_eth, old_eth, tnl, bpf_htons(ETH_P_IP));
132
133 if (new_eth == eth_buffer)
134 bpf_dynptr_write(&new_xdp_ptr, 0, eth_buffer, sizeof(eth_buffer), 0);
135
136 iph->version = 4;
137 iph->ihl = iphdr_sz >> 2;
138 iph->frag_off = 0;
139 iph->protocol = IPPROTO_IPIP;
140 iph->check = 0;
141 iph->tos = 0;
142 iph->tot_len = bpf_htons(payload_len + iphdr_sz);
143 iph->daddr = tnl->daddr.v4;
144 iph->saddr = tnl->saddr.v4;
145 iph->ttl = 8;
146
147 next_iph = (__u16 *)iph;
148 for (i = 0; i < iphdr_sz >> 1; i++)
149 csum += *next_iph++;
150
151 iph->check = ~((csum & 0xffff) + (csum >> 16));
152
153 count_tx(vip.protocol);
154
155 return XDP_TX;
156}
157
158static __always_inline int handle_ipv6(struct xdp_md *xdp, struct bpf_dynptr *xdp_ptr)
159{
160 __u8 eth_buffer[ethhdr_sz + ipv6hdr_sz + ethhdr_sz];
161 __u8 ip6h_buffer_tcp[ipv6hdr_sz + tcphdr_sz];
162 __u8 ip6h_buffer_udp[ipv6hdr_sz + udphdr_sz];
163 struct bpf_dynptr new_xdp_ptr;
164 struct iptnl_info *tnl;
165 struct ethhdr *new_eth;
166 struct ethhdr *old_eth;
167 struct ipv6hdr *ip6h;
168 __u16 payload_len;
169 struct vip vip = {};
170 int dport;
171
172 __builtin_memset(eth_buffer, 0, sizeof(eth_buffer));
173 __builtin_memset(ip6h_buffer_tcp, 0, sizeof(ip6h_buffer_tcp));
174 __builtin_memset(ip6h_buffer_udp, 0, sizeof(ip6h_buffer_udp));
175
176 if (ethhdr_sz + iphdr_sz + tcphdr_sz > xdp->data_end - xdp->data)
177 ip6h = bpf_dynptr_slice(xdp_ptr, ethhdr_sz, ip6h_buffer_udp, sizeof(ip6h_buffer_udp));
178 else
179 ip6h = bpf_dynptr_slice(xdp_ptr, ethhdr_sz, ip6h_buffer_tcp, sizeof(ip6h_buffer_tcp));
180
181 if (!ip6h)
182 return XDP_DROP;
183
184 dport = get_dport(ip6h + 1, ip6h->nexthdr);
185 if (dport == -1)
186 return XDP_DROP;
187
188 vip.protocol = ip6h->nexthdr;
189 vip.family = AF_INET6;
190 memcpy(vip.daddr.v6, ip6h->daddr.s6_addr32, sizeof(vip.daddr));
191 vip.dport = dport;
192 payload_len = ip6h->payload_len;
193
194 tnl = bpf_map_lookup_elem(&vip2tnl, &vip);
195 /* It only does v6-in-v6 */
196 if (!tnl || tnl->family != AF_INET6)
197 return XDP_PASS;
198
199 if (bpf_xdp_adjust_head(xdp, 0 - (int)ipv6hdr_sz))
200 return XDP_DROP;
201
202 bpf_dynptr_from_xdp(xdp, 0, &new_xdp_ptr);
203 new_eth = bpf_dynptr_slice_rdwr(&new_xdp_ptr, 0, eth_buffer, sizeof(eth_buffer));
204 if (!new_eth)
205 return XDP_DROP;
206
207 ip6h = (struct ipv6hdr *)(new_eth + 1);
208 old_eth = (struct ethhdr *)(ip6h + 1);
209
210 set_ethhdr(new_eth, old_eth, tnl, bpf_htons(ETH_P_IPV6));
211
212 if (new_eth == eth_buffer)
213 bpf_dynptr_write(&new_xdp_ptr, 0, eth_buffer, sizeof(eth_buffer), 0);
214
215 ip6h->version = 6;
216 ip6h->priority = 0;
217 memset(ip6h->flow_lbl, 0, sizeof(ip6h->flow_lbl));
218 ip6h->payload_len = bpf_htons(bpf_ntohs(payload_len) + ipv6hdr_sz);
219 ip6h->nexthdr = IPPROTO_IPV6;
220 ip6h->hop_limit = 8;
221 memcpy(ip6h->saddr.s6_addr32, tnl->saddr.v6, sizeof(tnl->saddr.v6));
222 memcpy(ip6h->daddr.s6_addr32, tnl->daddr.v6, sizeof(tnl->daddr.v6));
223
224 count_tx(vip.protocol);
225
226 return XDP_TX;
227}
228
229SEC("xdp")
230int _xdp_tx_iptunnel(struct xdp_md *xdp)
231{
232 __u8 buffer[ethhdr_sz];
233 struct bpf_dynptr ptr;
234 struct ethhdr *eth;
235 __u16 h_proto;
236
237 __builtin_memset(buffer, 0, sizeof(buffer));
238
239 bpf_dynptr_from_xdp(xdp, 0, &ptr);
240 eth = bpf_dynptr_slice(&ptr, 0, buffer, sizeof(buffer));
241 if (!eth)
242 return XDP_DROP;
243
244 h_proto = eth->h_proto;
245
246 if (h_proto == bpf_htons(ETH_P_IP))
247 return handle_ipv4(xdp, &ptr);
248 else if (h_proto == bpf_htons(ETH_P_IPV6))
249
250 return handle_ipv6(xdp, &ptr);
251 else
252 return XDP_DROP;
253}
254
255char _license[] SEC("license") = "GPL";