1 /* Copyright (c) 2016,2017 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 #include <stddef.h> 8 #include <string.h> 9 #include <linux/bpf.h> 10 #include <linux/if_ether.h> 11 #include <linux/if_packet.h> 12 #include <linux/ip.h> 13 #include <linux/ipv6.h> 14 #include <linux/in.h> 15 #include <linux/udp.h> 16 #include <linux/tcp.h> 17 #include <linux/pkt_cls.h> 18 #include <sys/socket.h> 19 #include "bpf_helpers.h" 20 #include "bpf_endian.h" 21 #include "test_iptunnel_common.h" 22 23 int _version SEC("version") = 1; 24 25 struct { 26 __u32 type; 27 __u32 max_entries; 28 __u32 *key; 29 __u64 *value; 30 } rxcnt SEC(".maps") = { 31 .type = BPF_MAP_TYPE_PERCPU_ARRAY, 32 .max_entries = 256, 33 }; 34 35 struct { 36 __u32 type; 37 __u32 max_entries; 38 struct vip *key; 39 struct iptnl_info *value; 40 } vip2tnl SEC(".maps") = { 41 .type = BPF_MAP_TYPE_HASH, 42 .max_entries = MAX_IPTNL_ENTRIES, 43 }; 44 45 static __always_inline void count_tx(__u32 protocol) 46 { 47 __u64 *rxcnt_count; 48 49 rxcnt_count = bpf_map_lookup_elem(&rxcnt, &protocol); 50 if (rxcnt_count) 51 *rxcnt_count += 1; 52 } 53 54 static __always_inline int get_dport(void *trans_data, void *data_end, 55 __u8 protocol) 56 { 57 struct tcphdr *th; 58 struct udphdr *uh; 59 60 switch (protocol) { 61 case IPPROTO_TCP: 62 th = (struct tcphdr *)trans_data; 63 if (th + 1 > data_end) 64 return -1; 65 return th->dest; 66 case IPPROTO_UDP: 67 uh = (struct udphdr *)trans_data; 68 if (uh + 1 > data_end) 69 return -1; 70 return uh->dest; 71 default: 72 return 0; 73 } 74 } 75 76 static __always_inline void set_ethhdr(struct ethhdr *new_eth, 77 const struct ethhdr *old_eth, 78 const struct iptnl_info *tnl, 79 __be16 h_proto) 80 { 81 memcpy(new_eth->h_source, old_eth->h_dest, sizeof(new_eth->h_source)); 82 memcpy(new_eth->h_dest, tnl->dmac, sizeof(new_eth->h_dest)); 83 new_eth->h_proto = h_proto; 84 } 85 86 static __always_inline int handle_ipv4(struct xdp_md *xdp) 87 { 88 void *data_end = (void *)(long)xdp->data_end; 89 void *data = (void *)(long)xdp->data; 90 struct iptnl_info *tnl; 91 struct ethhdr *new_eth; 92 struct ethhdr *old_eth; 93 struct iphdr *iph = data + sizeof(struct ethhdr); 94 __u16 *next_iph; 95 __u16 payload_len; 96 struct vip vip = {}; 97 int dport; 98 __u32 csum = 0; 99 int i; 100 101 if (iph + 1 > data_end) 102 return XDP_DROP; 103 104 dport = get_dport(iph + 1, data_end, iph->protocol); 105 if (dport == -1) 106 return XDP_DROP; 107 108 vip.protocol = iph->protocol; 109 vip.family = AF_INET; 110 vip.daddr.v4 = iph->daddr; 111 vip.dport = dport; 112 payload_len = bpf_ntohs(iph->tot_len); 113 114 tnl = bpf_map_lookup_elem(&vip2tnl, &vip); 115 /* It only does v4-in-v4 */ 116 if (!tnl || tnl->family != AF_INET) 117 return XDP_PASS; 118 119 if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct iphdr))) 120 return XDP_DROP; 121 122 data = (void *)(long)xdp->data; 123 data_end = (void *)(long)xdp->data_end; 124 125 new_eth = data; 126 iph = data + sizeof(*new_eth); 127 old_eth = data + sizeof(*iph); 128 129 if (new_eth + 1 > data_end || 130 old_eth + 1 > data_end || 131 iph + 1 > data_end) 132 return XDP_DROP; 133 134 set_ethhdr(new_eth, old_eth, tnl, bpf_htons(ETH_P_IP)); 135 136 iph->version = 4; 137 iph->ihl = sizeof(*iph) >> 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 + sizeof(*iph)); 143 iph->daddr = tnl->daddr.v4; 144 iph->saddr = tnl->saddr.v4; 145 iph->ttl = 8; 146 147 next_iph = (__u16 *)iph; 148 #pragma clang loop unroll(full) 149 for (i = 0; i < sizeof(*iph) >> 1; i++) 150 csum += *next_iph++; 151 152 iph->check = ~((csum & 0xffff) + (csum >> 16)); 153 154 count_tx(vip.protocol); 155 156 return XDP_TX; 157 } 158 159 static __always_inline int handle_ipv6(struct xdp_md *xdp) 160 { 161 void *data_end = (void *)(long)xdp->data_end; 162 void *data = (void *)(long)xdp->data; 163 struct iptnl_info *tnl; 164 struct ethhdr *new_eth; 165 struct ethhdr *old_eth; 166 struct ipv6hdr *ip6h = data + sizeof(struct ethhdr); 167 __u16 payload_len; 168 struct vip vip = {}; 169 int dport; 170 171 if (ip6h + 1 > data_end) 172 return XDP_DROP; 173 174 dport = get_dport(ip6h + 1, data_end, ip6h->nexthdr); 175 if (dport == -1) 176 return XDP_DROP; 177 178 vip.protocol = ip6h->nexthdr; 179 vip.family = AF_INET6; 180 memcpy(vip.daddr.v6, ip6h->daddr.s6_addr32, sizeof(vip.daddr)); 181 vip.dport = dport; 182 payload_len = ip6h->payload_len; 183 184 tnl = bpf_map_lookup_elem(&vip2tnl, &vip); 185 /* It only does v6-in-v6 */ 186 if (!tnl || tnl->family != AF_INET6) 187 return XDP_PASS; 188 189 if (bpf_xdp_adjust_head(xdp, 0 - (int)sizeof(struct ipv6hdr))) 190 return XDP_DROP; 191 192 data = (void *)(long)xdp->data; 193 data_end = (void *)(long)xdp->data_end; 194 195 new_eth = data; 196 ip6h = data + sizeof(*new_eth); 197 old_eth = data + sizeof(*ip6h); 198 199 if (new_eth + 1 > data_end || old_eth + 1 > data_end || 200 ip6h + 1 > data_end) 201 return XDP_DROP; 202 203 set_ethhdr(new_eth, old_eth, tnl, bpf_htons(ETH_P_IPV6)); 204 205 ip6h->version = 6; 206 ip6h->priority = 0; 207 memset(ip6h->flow_lbl, 0, sizeof(ip6h->flow_lbl)); 208 ip6h->payload_len = bpf_htons(bpf_ntohs(payload_len) + sizeof(*ip6h)); 209 ip6h->nexthdr = IPPROTO_IPV6; 210 ip6h->hop_limit = 8; 211 memcpy(ip6h->saddr.s6_addr32, tnl->saddr.v6, sizeof(tnl->saddr.v6)); 212 memcpy(ip6h->daddr.s6_addr32, tnl->daddr.v6, sizeof(tnl->daddr.v6)); 213 214 count_tx(vip.protocol); 215 216 return XDP_TX; 217 } 218 219 SEC("xdp_tx_iptunnel") 220 int _xdp_tx_iptunnel(struct xdp_md *xdp) 221 { 222 void *data_end = (void *)(long)xdp->data_end; 223 void *data = (void *)(long)xdp->data; 224 struct ethhdr *eth = data; 225 __u16 h_proto; 226 227 if (eth + 1 > data_end) 228 return XDP_DROP; 229 230 h_proto = eth->h_proto; 231 232 if (h_proto == bpf_htons(ETH_P_IP)) 233 return handle_ipv4(xdp); 234 else if (h_proto == bpf_htons(ETH_P_IPV6)) 235 236 return handle_ipv6(xdp); 237 else 238 return XDP_DROP; 239 } 240 241 char _license[] SEC("license") = "GPL"; 242