xref: /openbmc/linux/include/net/inet_ecn.h (revision bbaf1ff0)
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 #include <net/checksum.h>
12 
13 enum {
14 	INET_ECN_NOT_ECT = 0,
15 	INET_ECN_ECT_1 = 1,
16 	INET_ECN_ECT_0 = 2,
17 	INET_ECN_CE = 3,
18 	INET_ECN_MASK = 3,
19 };
20 
21 extern int sysctl_tunnel_ecn_log;
22 
23 static inline int INET_ECN_is_ce(__u8 dsfield)
24 {
25 	return (dsfield & INET_ECN_MASK) == INET_ECN_CE;
26 }
27 
28 static inline int INET_ECN_is_not_ect(__u8 dsfield)
29 {
30 	return (dsfield & INET_ECN_MASK) == INET_ECN_NOT_ECT;
31 }
32 
33 static inline int INET_ECN_is_capable(__u8 dsfield)
34 {
35 	return dsfield & INET_ECN_ECT_0;
36 }
37 
38 /*
39  * RFC 3168 9.1.1
40  *  The full-functionality option for ECN encapsulation is to copy the
41  *  ECN codepoint of the inside header to the outside header on
42  *  encapsulation if the inside header is not-ECT or ECT, and to set the
43  *  ECN codepoint of the outside header to ECT(0) if the ECN codepoint of
44  *  the inside header is CE.
45  */
46 static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner)
47 {
48 	outer &= ~INET_ECN_MASK;
49 	outer |= !INET_ECN_is_ce(inner) ? (inner & INET_ECN_MASK) :
50 					  INET_ECN_ECT_0;
51 	return outer;
52 }
53 
54 static inline void INET_ECN_xmit(struct sock *sk)
55 {
56 	inet_sk(sk)->tos |= INET_ECN_ECT_0;
57 	if (inet6_sk(sk) != NULL)
58 		inet6_sk(sk)->tclass |= INET_ECN_ECT_0;
59 }
60 
61 static inline void INET_ECN_dontxmit(struct sock *sk)
62 {
63 	inet_sk(sk)->tos &= ~INET_ECN_MASK;
64 	if (inet6_sk(sk) != NULL)
65 		inet6_sk(sk)->tclass &= ~INET_ECN_MASK;
66 }
67 
68 #define IP6_ECN_flow_init(label) do {		\
69       (label) &= ~htonl(INET_ECN_MASK << 20);	\
70     } while (0)
71 
72 #define	IP6_ECN_flow_xmit(sk, label) do {				\
73 	if (INET_ECN_is_capable(inet6_sk(sk)->tclass))			\
74 		(label) |= htonl(INET_ECN_ECT_0 << 20);			\
75     } while (0)
76 
77 static inline int IP_ECN_set_ce(struct iphdr *iph)
78 {
79 	u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
80 	__be16 check_add;
81 
82 	/*
83 	 * After the last operation we have (in binary):
84 	 * INET_ECN_NOT_ECT => 01
85 	 * INET_ECN_ECT_1   => 10
86 	 * INET_ECN_ECT_0   => 11
87 	 * INET_ECN_CE      => 00
88 	 */
89 	if (!(ecn & 2))
90 		return !ecn;
91 
92 	/*
93 	 * The following gives us:
94 	 * INET_ECN_ECT_1 => check += htons(0xFFFD)
95 	 * INET_ECN_ECT_0 => check += htons(0xFFFE)
96 	 */
97 	check_add = (__force __be16)((__force u16)htons(0xFFFB) +
98 				     (__force u16)htons(ecn));
99 
100 	iph->check = csum16_add(iph->check, check_add);
101 	iph->tos |= INET_ECN_CE;
102 	return 1;
103 }
104 
105 static inline int IP_ECN_set_ect1(struct iphdr *iph)
106 {
107 	if ((iph->tos & INET_ECN_MASK) != INET_ECN_ECT_0)
108 		return 0;
109 
110 	iph->check = csum16_add(iph->check, htons(0x1));
111 	iph->tos ^= INET_ECN_MASK;
112 	return 1;
113 }
114 
115 static inline void IP_ECN_clear(struct iphdr *iph)
116 {
117 	iph->tos &= ~INET_ECN_MASK;
118 }
119 
120 static inline void ipv4_copy_dscp(unsigned int dscp, struct iphdr *inner)
121 {
122 	dscp &= ~INET_ECN_MASK;
123 	ipv4_change_dsfield(inner, INET_ECN_MASK, dscp);
124 }
125 
126 struct ipv6hdr;
127 
128 /* Note:
129  * IP_ECN_set_ce() has to tweak IPV4 checksum when setting CE,
130  * meaning both changes have no effect on skb->csum if/when CHECKSUM_COMPLETE
131  * In IPv6 case, no checksum compensates the change in IPv6 header,
132  * so we have to update skb->csum.
133  */
134 static inline int IP6_ECN_set_ce(struct sk_buff *skb, struct ipv6hdr *iph)
135 {
136 	__be32 from, to;
137 
138 	if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph)))
139 		return 0;
140 
141 	from = *(__be32 *)iph;
142 	to = from | htonl(INET_ECN_CE << 20);
143 	*(__be32 *)iph = to;
144 	if (skb->ip_summed == CHECKSUM_COMPLETE)
145 		skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from),
146 				     (__force __wsum)to);
147 	return 1;
148 }
149 
150 static inline int IP6_ECN_set_ect1(struct sk_buff *skb, struct ipv6hdr *iph)
151 {
152 	__be32 from, to;
153 
154 	if ((ipv6_get_dsfield(iph) & INET_ECN_MASK) != INET_ECN_ECT_0)
155 		return 0;
156 
157 	from = *(__be32 *)iph;
158 	to = from ^ htonl(INET_ECN_MASK << 20);
159 	*(__be32 *)iph = to;
160 	if (skb->ip_summed == CHECKSUM_COMPLETE)
161 		skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from),
162 				     (__force __wsum)to);
163 	return 1;
164 }
165 
166 static inline void ipv6_copy_dscp(unsigned int dscp, struct ipv6hdr *inner)
167 {
168 	dscp &= ~INET_ECN_MASK;
169 	ipv6_change_dsfield(inner, INET_ECN_MASK, dscp);
170 }
171 
172 static inline int INET_ECN_set_ce(struct sk_buff *skb)
173 {
174 	switch (skb_protocol(skb, true)) {
175 	case cpu_to_be16(ETH_P_IP):
176 		if (skb_network_header(skb) + sizeof(struct iphdr) <=
177 		    skb_tail_pointer(skb))
178 			return IP_ECN_set_ce(ip_hdr(skb));
179 		break;
180 
181 	case cpu_to_be16(ETH_P_IPV6):
182 		if (skb_network_header(skb) + sizeof(struct ipv6hdr) <=
183 		    skb_tail_pointer(skb))
184 			return IP6_ECN_set_ce(skb, ipv6_hdr(skb));
185 		break;
186 	}
187 
188 	return 0;
189 }
190 
191 static inline int skb_get_dsfield(struct sk_buff *skb)
192 {
193 	switch (skb_protocol(skb, true)) {
194 	case cpu_to_be16(ETH_P_IP):
195 		if (!pskb_network_may_pull(skb, sizeof(struct iphdr)))
196 			break;
197 		return ipv4_get_dsfield(ip_hdr(skb));
198 
199 	case cpu_to_be16(ETH_P_IPV6):
200 		if (!pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))
201 			break;
202 		return ipv6_get_dsfield(ipv6_hdr(skb));
203 	}
204 
205 	return -1;
206 }
207 
208 static inline int INET_ECN_set_ect1(struct sk_buff *skb)
209 {
210 	switch (skb_protocol(skb, true)) {
211 	case cpu_to_be16(ETH_P_IP):
212 		if (skb_network_header(skb) + sizeof(struct iphdr) <=
213 		    skb_tail_pointer(skb))
214 			return IP_ECN_set_ect1(ip_hdr(skb));
215 		break;
216 
217 	case cpu_to_be16(ETH_P_IPV6):
218 		if (skb_network_header(skb) + sizeof(struct ipv6hdr) <=
219 		    skb_tail_pointer(skb))
220 			return IP6_ECN_set_ect1(skb, ipv6_hdr(skb));
221 		break;
222 	}
223 
224 	return 0;
225 }
226 
227 /*
228  * RFC 6040 4.2
229  *  To decapsulate the inner header at the tunnel egress, a compliant
230  *  tunnel egress MUST set the outgoing ECN field to the codepoint at the
231  *  intersection of the appropriate arriving inner header (row) and outer
232  *  header (column) in Figure 4
233  *
234  *      +---------+------------------------------------------------+
235  *      |Arriving |            Arriving Outer Header               |
236  *      |   Inner +---------+------------+------------+------------+
237  *      |  Header | Not-ECT | ECT(0)     | ECT(1)     |     CE     |
238  *      +---------+---------+------------+------------+------------+
239  *      | Not-ECT | Not-ECT |Not-ECT(!!!)|Not-ECT(!!!)| <drop>(!!!)|
240  *      |  ECT(0) |  ECT(0) | ECT(0)     | ECT(1)     |     CE     |
241  *      |  ECT(1) |  ECT(1) | ECT(1) (!) | ECT(1)     |     CE     |
242  *      |    CE   |      CE |     CE     |     CE(!!!)|     CE     |
243  *      +---------+---------+------------+------------+------------+
244  *
245  *             Figure 4: New IP in IP Decapsulation Behaviour
246  *
247  *  returns 0 on success
248  *          1 if something is broken and should be logged (!!! above)
249  *          2 if packet should be dropped
250  */
251 static inline int __INET_ECN_decapsulate(__u8 outer, __u8 inner, bool *set_ce)
252 {
253 	if (INET_ECN_is_not_ect(inner)) {
254 		switch (outer & INET_ECN_MASK) {
255 		case INET_ECN_NOT_ECT:
256 			return 0;
257 		case INET_ECN_ECT_0:
258 		case INET_ECN_ECT_1:
259 			return 1;
260 		case INET_ECN_CE:
261 			return 2;
262 		}
263 	}
264 
265 	*set_ce = INET_ECN_is_ce(outer);
266 	return 0;
267 }
268 
269 static inline int INET_ECN_decapsulate(struct sk_buff *skb,
270 				       __u8 outer, __u8 inner)
271 {
272 	bool set_ce = false;
273 	int rc;
274 
275 	rc = __INET_ECN_decapsulate(outer, inner, &set_ce);
276 	if (!rc) {
277 		if (set_ce)
278 			INET_ECN_set_ce(skb);
279 		else if ((outer & INET_ECN_MASK) == INET_ECN_ECT_1)
280 			INET_ECN_set_ect1(skb);
281 	}
282 
283 	return rc;
284 }
285 
286 static inline int IP_ECN_decapsulate(const struct iphdr *oiph,
287 				     struct sk_buff *skb)
288 {
289 	__u8 inner;
290 
291 	switch (skb_protocol(skb, true)) {
292 	case htons(ETH_P_IP):
293 		inner = ip_hdr(skb)->tos;
294 		break;
295 	case htons(ETH_P_IPV6):
296 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
297 		break;
298 	default:
299 		return 0;
300 	}
301 
302 	return INET_ECN_decapsulate(skb, oiph->tos, inner);
303 }
304 
305 static inline int IP6_ECN_decapsulate(const struct ipv6hdr *oipv6h,
306 				      struct sk_buff *skb)
307 {
308 	__u8 inner;
309 
310 	switch (skb_protocol(skb, true)) {
311 	case htons(ETH_P_IP):
312 		inner = ip_hdr(skb)->tos;
313 		break;
314 	case htons(ETH_P_IPV6):
315 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
316 		break;
317 	default:
318 		return 0;
319 	}
320 
321 	return INET_ECN_decapsulate(skb, ipv6_get_dsfield(oipv6h), inner);
322 }
323 #endif
324