1 // SPDX-License-Identifier: GPL-2.0 2 #include <limits.h> 3 #include <stddef.h> 4 #include <stdbool.h> 5 #include <string.h> 6 #include <linux/pkt_cls.h> 7 #include <linux/bpf.h> 8 #include <linux/in.h> 9 #include <linux/if_ether.h> 10 #include <linux/icmp.h> 11 #include <linux/ip.h> 12 #include <linux/ipv6.h> 13 #include <linux/tcp.h> 14 #include <linux/udp.h> 15 #include <linux/if_packet.h> 16 #include <sys/socket.h> 17 #include <linux/if_tunnel.h> 18 #include <linux/mpls.h> 19 #include "bpf_helpers.h" 20 #include "bpf_endian.h" 21 22 int _version SEC("version") = 1; 23 #define PROG(F) SEC(#F) int bpf_func_##F 24 25 /* These are the identifiers of the BPF programs that will be used in tail 26 * calls. Name is limited to 16 characters, with the terminating character and 27 * bpf_func_ above, we have only 6 to work with, anything after will be cropped. 28 */ 29 enum { 30 IP, 31 IPV6, 32 IPV6OP, /* Destination/Hop-by-Hop Options IPv6 Extension header */ 33 IPV6FR, /* Fragmentation IPv6 Extension Header */ 34 MPLS, 35 VLAN, 36 }; 37 38 #define IP_MF 0x2000 39 #define IP_OFFSET 0x1FFF 40 #define IP6_MF 0x0001 41 #define IP6_OFFSET 0xFFF8 42 43 struct vlan_hdr { 44 __be16 h_vlan_TCI; 45 __be16 h_vlan_encapsulated_proto; 46 }; 47 48 struct gre_hdr { 49 __be16 flags; 50 __be16 proto; 51 }; 52 53 struct frag_hdr { 54 __u8 nexthdr; 55 __u8 reserved; 56 __be16 frag_off; 57 __be32 identification; 58 }; 59 60 struct bpf_map_def SEC("maps") jmp_table = { 61 .type = BPF_MAP_TYPE_PROG_ARRAY, 62 .key_size = sizeof(__u32), 63 .value_size = sizeof(__u32), 64 .max_entries = 8 65 }; 66 67 static __always_inline void *bpf_flow_dissect_get_header(struct __sk_buff *skb, 68 __u16 hdr_size, 69 void *buffer) 70 { 71 void *data_end = (void *)(long)skb->data_end; 72 void *data = (void *)(long)skb->data; 73 __u16 thoff = skb->flow_keys->thoff; 74 __u8 *hdr; 75 76 /* Verifies this variable offset does not overflow */ 77 if (thoff > (USHRT_MAX - hdr_size)) 78 return NULL; 79 80 hdr = data + thoff; 81 if (hdr + hdr_size <= data_end) 82 return hdr; 83 84 if (bpf_skb_load_bytes(skb, thoff, buffer, hdr_size)) 85 return NULL; 86 87 return buffer; 88 } 89 90 /* Dispatches on ETHERTYPE */ 91 static __always_inline int parse_eth_proto(struct __sk_buff *skb, __be16 proto) 92 { 93 struct bpf_flow_keys *keys = skb->flow_keys; 94 95 switch (proto) { 96 case bpf_htons(ETH_P_IP): 97 bpf_tail_call(skb, &jmp_table, IP); 98 break; 99 case bpf_htons(ETH_P_IPV6): 100 bpf_tail_call(skb, &jmp_table, IPV6); 101 break; 102 case bpf_htons(ETH_P_MPLS_MC): 103 case bpf_htons(ETH_P_MPLS_UC): 104 bpf_tail_call(skb, &jmp_table, MPLS); 105 break; 106 case bpf_htons(ETH_P_8021Q): 107 case bpf_htons(ETH_P_8021AD): 108 bpf_tail_call(skb, &jmp_table, VLAN); 109 break; 110 default: 111 /* Protocol not supported */ 112 return BPF_DROP; 113 } 114 115 return BPF_DROP; 116 } 117 118 SEC("flow_dissector") 119 int _dissect(struct __sk_buff *skb) 120 { 121 struct bpf_flow_keys *keys = skb->flow_keys; 122 123 return parse_eth_proto(skb, keys->n_proto); 124 } 125 126 /* Parses on IPPROTO_* */ 127 static __always_inline int parse_ip_proto(struct __sk_buff *skb, __u8 proto) 128 { 129 struct bpf_flow_keys *keys = skb->flow_keys; 130 void *data_end = (void *)(long)skb->data_end; 131 struct icmphdr *icmp, _icmp; 132 struct gre_hdr *gre, _gre; 133 struct ethhdr *eth, _eth; 134 struct tcphdr *tcp, _tcp; 135 struct udphdr *udp, _udp; 136 137 keys->ip_proto = proto; 138 switch (proto) { 139 case IPPROTO_ICMP: 140 icmp = bpf_flow_dissect_get_header(skb, sizeof(*icmp), &_icmp); 141 if (!icmp) 142 return BPF_DROP; 143 return BPF_OK; 144 case IPPROTO_IPIP: 145 keys->is_encap = true; 146 return parse_eth_proto(skb, bpf_htons(ETH_P_IP)); 147 case IPPROTO_IPV6: 148 keys->is_encap = true; 149 return parse_eth_proto(skb, bpf_htons(ETH_P_IPV6)); 150 case IPPROTO_GRE: 151 gre = bpf_flow_dissect_get_header(skb, sizeof(*gre), &_gre); 152 if (!gre) 153 return BPF_DROP; 154 155 if (bpf_htons(gre->flags & GRE_VERSION)) 156 /* Only inspect standard GRE packets with version 0 */ 157 return BPF_OK; 158 159 keys->thoff += sizeof(*gre); /* Step over GRE Flags and Proto */ 160 if (GRE_IS_CSUM(gre->flags)) 161 keys->thoff += 4; /* Step over chksum and Padding */ 162 if (GRE_IS_KEY(gre->flags)) 163 keys->thoff += 4; /* Step over key */ 164 if (GRE_IS_SEQ(gre->flags)) 165 keys->thoff += 4; /* Step over sequence number */ 166 167 keys->is_encap = true; 168 169 if (gre->proto == bpf_htons(ETH_P_TEB)) { 170 eth = bpf_flow_dissect_get_header(skb, sizeof(*eth), 171 &_eth); 172 if (!eth) 173 return BPF_DROP; 174 175 keys->thoff += sizeof(*eth); 176 177 return parse_eth_proto(skb, eth->h_proto); 178 } else { 179 return parse_eth_proto(skb, gre->proto); 180 } 181 case IPPROTO_TCP: 182 tcp = bpf_flow_dissect_get_header(skb, sizeof(*tcp), &_tcp); 183 if (!tcp) 184 return BPF_DROP; 185 186 if (tcp->doff < 5) 187 return BPF_DROP; 188 189 if ((__u8 *)tcp + (tcp->doff << 2) > data_end) 190 return BPF_DROP; 191 192 keys->sport = tcp->source; 193 keys->dport = tcp->dest; 194 return BPF_OK; 195 case IPPROTO_UDP: 196 case IPPROTO_UDPLITE: 197 udp = bpf_flow_dissect_get_header(skb, sizeof(*udp), &_udp); 198 if (!udp) 199 return BPF_DROP; 200 201 keys->sport = udp->source; 202 keys->dport = udp->dest; 203 return BPF_OK; 204 default: 205 return BPF_DROP; 206 } 207 208 return BPF_DROP; 209 } 210 211 static __always_inline int parse_ipv6_proto(struct __sk_buff *skb, __u8 nexthdr) 212 { 213 struct bpf_flow_keys *keys = skb->flow_keys; 214 215 keys->ip_proto = nexthdr; 216 switch (nexthdr) { 217 case IPPROTO_HOPOPTS: 218 case IPPROTO_DSTOPTS: 219 bpf_tail_call(skb, &jmp_table, IPV6OP); 220 break; 221 case IPPROTO_FRAGMENT: 222 bpf_tail_call(skb, &jmp_table, IPV6FR); 223 break; 224 default: 225 return parse_ip_proto(skb, nexthdr); 226 } 227 228 return BPF_DROP; 229 } 230 231 PROG(IP)(struct __sk_buff *skb) 232 { 233 void *data_end = (void *)(long)skb->data_end; 234 struct bpf_flow_keys *keys = skb->flow_keys; 235 void *data = (void *)(long)skb->data; 236 struct iphdr *iph, _iph; 237 bool done = false; 238 239 iph = bpf_flow_dissect_get_header(skb, sizeof(*iph), &_iph); 240 if (!iph) 241 return BPF_DROP; 242 243 /* IP header cannot be smaller than 20 bytes */ 244 if (iph->ihl < 5) 245 return BPF_DROP; 246 247 keys->addr_proto = ETH_P_IP; 248 keys->ipv4_src = iph->saddr; 249 keys->ipv4_dst = iph->daddr; 250 251 keys->thoff += iph->ihl << 2; 252 if (data + keys->thoff > data_end) 253 return BPF_DROP; 254 255 if (iph->frag_off & bpf_htons(IP_MF | IP_OFFSET)) { 256 keys->is_frag = true; 257 if (iph->frag_off & bpf_htons(IP_OFFSET)) 258 /* From second fragment on, packets do not have headers 259 * we can parse. 260 */ 261 done = true; 262 else 263 keys->is_first_frag = true; 264 } 265 266 if (done) 267 return BPF_OK; 268 269 return parse_ip_proto(skb, iph->protocol); 270 } 271 272 PROG(IPV6)(struct __sk_buff *skb) 273 { 274 struct bpf_flow_keys *keys = skb->flow_keys; 275 struct ipv6hdr *ip6h, _ip6h; 276 277 ip6h = bpf_flow_dissect_get_header(skb, sizeof(*ip6h), &_ip6h); 278 if (!ip6h) 279 return BPF_DROP; 280 281 keys->addr_proto = ETH_P_IPV6; 282 memcpy(&keys->ipv6_src, &ip6h->saddr, 2*sizeof(ip6h->saddr)); 283 284 keys->thoff += sizeof(struct ipv6hdr); 285 286 return parse_ipv6_proto(skb, ip6h->nexthdr); 287 } 288 289 PROG(IPV6OP)(struct __sk_buff *skb) 290 { 291 struct ipv6_opt_hdr *ip6h, _ip6h; 292 293 ip6h = bpf_flow_dissect_get_header(skb, sizeof(*ip6h), &_ip6h); 294 if (!ip6h) 295 return BPF_DROP; 296 297 /* hlen is in 8-octets and does not include the first 8 bytes 298 * of the header 299 */ 300 skb->flow_keys->thoff += (1 + ip6h->hdrlen) << 3; 301 302 return parse_ipv6_proto(skb, ip6h->nexthdr); 303 } 304 305 PROG(IPV6FR)(struct __sk_buff *skb) 306 { 307 struct bpf_flow_keys *keys = skb->flow_keys; 308 struct frag_hdr *fragh, _fragh; 309 310 fragh = bpf_flow_dissect_get_header(skb, sizeof(*fragh), &_fragh); 311 if (!fragh) 312 return BPF_DROP; 313 314 keys->thoff += sizeof(*fragh); 315 keys->is_frag = true; 316 if (!(fragh->frag_off & bpf_htons(IP6_OFFSET))) 317 keys->is_first_frag = true; 318 319 return parse_ipv6_proto(skb, fragh->nexthdr); 320 } 321 322 PROG(MPLS)(struct __sk_buff *skb) 323 { 324 struct mpls_label *mpls, _mpls; 325 326 mpls = bpf_flow_dissect_get_header(skb, sizeof(*mpls), &_mpls); 327 if (!mpls) 328 return BPF_DROP; 329 330 return BPF_OK; 331 } 332 333 PROG(VLAN)(struct __sk_buff *skb) 334 { 335 struct bpf_flow_keys *keys = skb->flow_keys; 336 struct vlan_hdr *vlan, _vlan; 337 338 /* Account for double-tagging */ 339 if (keys->n_proto == bpf_htons(ETH_P_8021AD)) { 340 vlan = bpf_flow_dissect_get_header(skb, sizeof(*vlan), &_vlan); 341 if (!vlan) 342 return BPF_DROP; 343 344 if (vlan->h_vlan_encapsulated_proto != bpf_htons(ETH_P_8021Q)) 345 return BPF_DROP; 346 347 keys->nhoff += sizeof(*vlan); 348 keys->thoff += sizeof(*vlan); 349 } 350 351 vlan = bpf_flow_dissect_get_header(skb, sizeof(*vlan), &_vlan); 352 if (!vlan) 353 return BPF_DROP; 354 355 keys->nhoff += sizeof(*vlan); 356 keys->thoff += sizeof(*vlan); 357 /* Only allow 8021AD + 8021Q double tagging and no triple tagging.*/ 358 if (vlan->h_vlan_encapsulated_proto == bpf_htons(ETH_P_8021AD) || 359 vlan->h_vlan_encapsulated_proto == bpf_htons(ETH_P_8021Q)) 360 return BPF_DROP; 361 362 keys->n_proto = vlan->h_vlan_encapsulated_proto; 363 return parse_eth_proto(skb, vlan->h_vlan_encapsulated_proto); 364 } 365 366 char __license[] SEC("license") = "GPL"; 367