1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _INET_ECN_H_ 3 #define _INET_ECN_H_ 4 5 #include <linux/ip.h> 6 #include <linux/skbuff.h> 7 #include <linux/if_vlan.h> 8 9 #include <net/inet_sock.h> 10 #include <net/dsfield.h> 11 12 enum { 13 INET_ECN_NOT_ECT = 0, 14 INET_ECN_ECT_1 = 1, 15 INET_ECN_ECT_0 = 2, 16 INET_ECN_CE = 3, 17 INET_ECN_MASK = 3, 18 }; 19 20 extern int sysctl_tunnel_ecn_log; 21 22 static inline int INET_ECN_is_ce(__u8 dsfield) 23 { 24 return (dsfield & INET_ECN_MASK) == INET_ECN_CE; 25 } 26 27 static inline int INET_ECN_is_not_ect(__u8 dsfield) 28 { 29 return (dsfield & INET_ECN_MASK) == INET_ECN_NOT_ECT; 30 } 31 32 static inline int INET_ECN_is_capable(__u8 dsfield) 33 { 34 return dsfield & INET_ECN_ECT_0; 35 } 36 37 /* 38 * RFC 3168 9.1.1 39 * The full-functionality option for ECN encapsulation is to copy the 40 * ECN codepoint of the inside header to the outside header on 41 * encapsulation if the inside header is not-ECT or ECT, and to set the 42 * ECN codepoint of the outside header to ECT(0) if the ECN codepoint of 43 * the inside header is CE. 44 */ 45 static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner) 46 { 47 outer &= ~INET_ECN_MASK; 48 outer |= !INET_ECN_is_ce(inner) ? (inner & INET_ECN_MASK) : 49 INET_ECN_ECT_0; 50 return outer; 51 } 52 53 static inline void INET_ECN_xmit(struct sock *sk) 54 { 55 inet_sk(sk)->tos |= INET_ECN_ECT_0; 56 if (inet6_sk(sk) != NULL) 57 inet6_sk(sk)->tclass |= INET_ECN_ECT_0; 58 } 59 60 static inline void INET_ECN_dontxmit(struct sock *sk) 61 { 62 inet_sk(sk)->tos &= ~INET_ECN_MASK; 63 if (inet6_sk(sk) != NULL) 64 inet6_sk(sk)->tclass &= ~INET_ECN_MASK; 65 } 66 67 #define IP6_ECN_flow_init(label) do { \ 68 (label) &= ~htonl(INET_ECN_MASK << 20); \ 69 } while (0) 70 71 #define IP6_ECN_flow_xmit(sk, label) do { \ 72 if (INET_ECN_is_capable(inet6_sk(sk)->tclass)) \ 73 (label) |= htonl(INET_ECN_ECT_0 << 20); \ 74 } while (0) 75 76 static inline int IP_ECN_set_ce(struct iphdr *iph) 77 { 78 u32 check = (__force u32)iph->check; 79 u32 ecn = (iph->tos + 1) & INET_ECN_MASK; 80 81 /* 82 * After the last operation we have (in binary): 83 * INET_ECN_NOT_ECT => 01 84 * INET_ECN_ECT_1 => 10 85 * INET_ECN_ECT_0 => 11 86 * INET_ECN_CE => 00 87 */ 88 if (!(ecn & 2)) 89 return !ecn; 90 91 /* 92 * The following gives us: 93 * INET_ECN_ECT_1 => check += htons(0xFFFD) 94 * INET_ECN_ECT_0 => check += htons(0xFFFE) 95 */ 96 check += (__force u16)htons(0xFFFB) + (__force u16)htons(ecn); 97 98 iph->check = (__force __sum16)(check + (check>=0xFFFF)); 99 iph->tos |= INET_ECN_CE; 100 return 1; 101 } 102 103 static inline int IP_ECN_set_ect1(struct iphdr *iph) 104 { 105 u32 check = (__force u32)iph->check; 106 107 if ((iph->tos & INET_ECN_MASK) != INET_ECN_ECT_0) 108 return 0; 109 110 check += (__force u16)htons(0x1); 111 112 iph->check = (__force __sum16)(check + (check>=0xFFFF)); 113 iph->tos ^= INET_ECN_MASK; 114 return 1; 115 } 116 117 static inline void IP_ECN_clear(struct iphdr *iph) 118 { 119 iph->tos &= ~INET_ECN_MASK; 120 } 121 122 static inline void ipv4_copy_dscp(unsigned int dscp, struct iphdr *inner) 123 { 124 dscp &= ~INET_ECN_MASK; 125 ipv4_change_dsfield(inner, INET_ECN_MASK, dscp); 126 } 127 128 struct ipv6hdr; 129 130 /* Note: 131 * IP_ECN_set_ce() has to tweak IPV4 checksum when setting CE, 132 * meaning both changes have no effect on skb->csum if/when CHECKSUM_COMPLETE 133 * In IPv6 case, no checksum compensates the change in IPv6 header, 134 * so we have to update skb->csum. 135 */ 136 static inline int IP6_ECN_set_ce(struct sk_buff *skb, struct ipv6hdr *iph) 137 { 138 __be32 from, to; 139 140 if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph))) 141 return 0; 142 143 from = *(__be32 *)iph; 144 to = from | htonl(INET_ECN_CE << 20); 145 *(__be32 *)iph = to; 146 if (skb->ip_summed == CHECKSUM_COMPLETE) 147 skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from), 148 (__force __wsum)to); 149 return 1; 150 } 151 152 static inline int IP6_ECN_set_ect1(struct sk_buff *skb, struct ipv6hdr *iph) 153 { 154 __be32 from, to; 155 156 if ((ipv6_get_dsfield(iph) & INET_ECN_MASK) != INET_ECN_ECT_0) 157 return 0; 158 159 from = *(__be32 *)iph; 160 to = from ^ htonl(INET_ECN_MASK << 20); 161 *(__be32 *)iph = to; 162 if (skb->ip_summed == CHECKSUM_COMPLETE) 163 skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from), 164 (__force __wsum)to); 165 return 1; 166 } 167 168 static inline void ipv6_copy_dscp(unsigned int dscp, struct ipv6hdr *inner) 169 { 170 dscp &= ~INET_ECN_MASK; 171 ipv6_change_dsfield(inner, INET_ECN_MASK, dscp); 172 } 173 174 static inline int INET_ECN_set_ce(struct sk_buff *skb) 175 { 176 switch (skb_protocol(skb, true)) { 177 case cpu_to_be16(ETH_P_IP): 178 if (skb_network_header(skb) + sizeof(struct iphdr) <= 179 skb_tail_pointer(skb)) 180 return IP_ECN_set_ce(ip_hdr(skb)); 181 break; 182 183 case cpu_to_be16(ETH_P_IPV6): 184 if (skb_network_header(skb) + sizeof(struct ipv6hdr) <= 185 skb_tail_pointer(skb)) 186 return IP6_ECN_set_ce(skb, ipv6_hdr(skb)); 187 break; 188 } 189 190 return 0; 191 } 192 193 static inline int INET_ECN_set_ect1(struct sk_buff *skb) 194 { 195 switch (skb_protocol(skb, true)) { 196 case cpu_to_be16(ETH_P_IP): 197 if (skb_network_header(skb) + sizeof(struct iphdr) <= 198 skb_tail_pointer(skb)) 199 return IP_ECN_set_ect1(ip_hdr(skb)); 200 break; 201 202 case cpu_to_be16(ETH_P_IPV6): 203 if (skb_network_header(skb) + sizeof(struct ipv6hdr) <= 204 skb_tail_pointer(skb)) 205 return IP6_ECN_set_ect1(skb, ipv6_hdr(skb)); 206 break; 207 } 208 209 return 0; 210 } 211 212 /* 213 * RFC 6040 4.2 214 * To decapsulate the inner header at the tunnel egress, a compliant 215 * tunnel egress MUST set the outgoing ECN field to the codepoint at the 216 * intersection of the appropriate arriving inner header (row) and outer 217 * header (column) in Figure 4 218 * 219 * +---------+------------------------------------------------+ 220 * |Arriving | Arriving Outer Header | 221 * | Inner +---------+------------+------------+------------+ 222 * | Header | Not-ECT | ECT(0) | ECT(1) | CE | 223 * +---------+---------+------------+------------+------------+ 224 * | Not-ECT | Not-ECT |Not-ECT(!!!)|Not-ECT(!!!)| <drop>(!!!)| 225 * | ECT(0) | ECT(0) | ECT(0) | ECT(1) | CE | 226 * | ECT(1) | ECT(1) | ECT(1) (!) | ECT(1) | CE | 227 * | CE | CE | CE | CE(!!!)| CE | 228 * +---------+---------+------------+------------+------------+ 229 * 230 * Figure 4: New IP in IP Decapsulation Behaviour 231 * 232 * returns 0 on success 233 * 1 if something is broken and should be logged (!!! above) 234 * 2 if packet should be dropped 235 */ 236 static inline int __INET_ECN_decapsulate(__u8 outer, __u8 inner, bool *set_ce) 237 { 238 if (INET_ECN_is_not_ect(inner)) { 239 switch (outer & INET_ECN_MASK) { 240 case INET_ECN_NOT_ECT: 241 return 0; 242 case INET_ECN_ECT_0: 243 case INET_ECN_ECT_1: 244 return 1; 245 case INET_ECN_CE: 246 return 2; 247 } 248 } 249 250 *set_ce = INET_ECN_is_ce(outer); 251 return 0; 252 } 253 254 static inline int INET_ECN_decapsulate(struct sk_buff *skb, 255 __u8 outer, __u8 inner) 256 { 257 bool set_ce = false; 258 int rc; 259 260 rc = __INET_ECN_decapsulate(outer, inner, &set_ce); 261 if (!rc) { 262 if (set_ce) 263 INET_ECN_set_ce(skb); 264 else if ((outer & INET_ECN_MASK) == INET_ECN_ECT_1) 265 INET_ECN_set_ect1(skb); 266 } 267 268 return rc; 269 } 270 271 static inline int IP_ECN_decapsulate(const struct iphdr *oiph, 272 struct sk_buff *skb) 273 { 274 __u8 inner; 275 276 switch (skb_protocol(skb, true)) { 277 case htons(ETH_P_IP): 278 inner = ip_hdr(skb)->tos; 279 break; 280 case htons(ETH_P_IPV6): 281 inner = ipv6_get_dsfield(ipv6_hdr(skb)); 282 break; 283 default: 284 return 0; 285 } 286 287 return INET_ECN_decapsulate(skb, oiph->tos, inner); 288 } 289 290 static inline int IP6_ECN_decapsulate(const struct ipv6hdr *oipv6h, 291 struct sk_buff *skb) 292 { 293 __u8 inner; 294 295 switch (skb_protocol(skb, true)) { 296 case htons(ETH_P_IP): 297 inner = ip_hdr(skb)->tos; 298 break; 299 case htons(ETH_P_IPV6): 300 inner = ipv6_get_dsfield(ipv6_hdr(skb)); 301 break; 302 default: 303 return 0; 304 } 305 306 return INET_ECN_decapsulate(skb, ipv6_get_dsfield(oipv6h), inner); 307 } 308 #endif 309