xref: /openbmc/linux/include/net/inet_ecn.h (revision 6c33a6f4)
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 
8 #include <net/inet_sock.h>
9 #include <net/dsfield.h>
10 
11 enum {
12 	INET_ECN_NOT_ECT = 0,
13 	INET_ECN_ECT_1 = 1,
14 	INET_ECN_ECT_0 = 2,
15 	INET_ECN_CE = 3,
16 	INET_ECN_MASK = 3,
17 };
18 
19 extern int sysctl_tunnel_ecn_log;
20 
21 static inline int INET_ECN_is_ce(__u8 dsfield)
22 {
23 	return (dsfield & INET_ECN_MASK) == INET_ECN_CE;
24 }
25 
26 static inline int INET_ECN_is_not_ect(__u8 dsfield)
27 {
28 	return (dsfield & INET_ECN_MASK) == INET_ECN_NOT_ECT;
29 }
30 
31 static inline int INET_ECN_is_capable(__u8 dsfield)
32 {
33 	return dsfield & INET_ECN_ECT_0;
34 }
35 
36 /*
37  * RFC 3168 9.1.1
38  *  The full-functionality option for ECN encapsulation is to copy the
39  *  ECN codepoint of the inside header to the outside header on
40  *  encapsulation if the inside header is not-ECT or ECT, and to set the
41  *  ECN codepoint of the outside header to ECT(0) if the ECN codepoint of
42  *  the inside header is CE.
43  */
44 static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner)
45 {
46 	outer &= ~INET_ECN_MASK;
47 	outer |= !INET_ECN_is_ce(inner) ? (inner & INET_ECN_MASK) :
48 					  INET_ECN_ECT_0;
49 	return outer;
50 }
51 
52 static inline void INET_ECN_xmit(struct sock *sk)
53 {
54 	inet_sk(sk)->tos |= INET_ECN_ECT_0;
55 	if (inet6_sk(sk) != NULL)
56 		inet6_sk(sk)->tclass |= INET_ECN_ECT_0;
57 }
58 
59 static inline void INET_ECN_dontxmit(struct sock *sk)
60 {
61 	inet_sk(sk)->tos &= ~INET_ECN_MASK;
62 	if (inet6_sk(sk) != NULL)
63 		inet6_sk(sk)->tclass &= ~INET_ECN_MASK;
64 }
65 
66 #define IP6_ECN_flow_init(label) do {		\
67       (label) &= ~htonl(INET_ECN_MASK << 20);	\
68     } while (0)
69 
70 #define	IP6_ECN_flow_xmit(sk, label) do {				\
71 	if (INET_ECN_is_capable(inet6_sk(sk)->tclass))			\
72 		(label) |= htonl(INET_ECN_ECT_0 << 20);			\
73     } while (0)
74 
75 static inline int IP_ECN_set_ce(struct iphdr *iph)
76 {
77 	u32 check = (__force u32)iph->check;
78 	u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
79 
80 	/*
81 	 * After the last operation we have (in binary):
82 	 * INET_ECN_NOT_ECT => 01
83 	 * INET_ECN_ECT_1   => 10
84 	 * INET_ECN_ECT_0   => 11
85 	 * INET_ECN_CE      => 00
86 	 */
87 	if (!(ecn & 2))
88 		return !ecn;
89 
90 	/*
91 	 * The following gives us:
92 	 * INET_ECN_ECT_1 => check += htons(0xFFFD)
93 	 * INET_ECN_ECT_0 => check += htons(0xFFFE)
94 	 */
95 	check += (__force u16)htons(0xFFFB) + (__force u16)htons(ecn);
96 
97 	iph->check = (__force __sum16)(check + (check>=0xFFFF));
98 	iph->tos |= INET_ECN_CE;
99 	return 1;
100 }
101 
102 static inline void IP_ECN_clear(struct iphdr *iph)
103 {
104 	iph->tos &= ~INET_ECN_MASK;
105 }
106 
107 static inline void ipv4_copy_dscp(unsigned int dscp, struct iphdr *inner)
108 {
109 	dscp &= ~INET_ECN_MASK;
110 	ipv4_change_dsfield(inner, INET_ECN_MASK, dscp);
111 }
112 
113 struct ipv6hdr;
114 
115 /* Note:
116  * IP_ECN_set_ce() has to tweak IPV4 checksum when setting CE,
117  * meaning both changes have no effect on skb->csum if/when CHECKSUM_COMPLETE
118  * In IPv6 case, no checksum compensates the change in IPv6 header,
119  * so we have to update skb->csum.
120  */
121 static inline int IP6_ECN_set_ce(struct sk_buff *skb, struct ipv6hdr *iph)
122 {
123 	__be32 from, to;
124 
125 	if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph)))
126 		return 0;
127 
128 	from = *(__be32 *)iph;
129 	to = from | htonl(INET_ECN_CE << 20);
130 	*(__be32 *)iph = to;
131 	if (skb->ip_summed == CHECKSUM_COMPLETE)
132 		skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from),
133 				     (__force __wsum)to);
134 	return 1;
135 }
136 
137 static inline void ipv6_copy_dscp(unsigned int dscp, struct ipv6hdr *inner)
138 {
139 	dscp &= ~INET_ECN_MASK;
140 	ipv6_change_dsfield(inner, INET_ECN_MASK, dscp);
141 }
142 
143 static inline int INET_ECN_set_ce(struct sk_buff *skb)
144 {
145 	switch (skb->protocol) {
146 	case cpu_to_be16(ETH_P_IP):
147 		if (skb_network_header(skb) + sizeof(struct iphdr) <=
148 		    skb_tail_pointer(skb))
149 			return IP_ECN_set_ce(ip_hdr(skb));
150 		break;
151 
152 	case cpu_to_be16(ETH_P_IPV6):
153 		if (skb_network_header(skb) + sizeof(struct ipv6hdr) <=
154 		    skb_tail_pointer(skb))
155 			return IP6_ECN_set_ce(skb, ipv6_hdr(skb));
156 		break;
157 	}
158 
159 	return 0;
160 }
161 
162 /*
163  * RFC 6040 4.2
164  *  To decapsulate the inner header at the tunnel egress, a compliant
165  *  tunnel egress MUST set the outgoing ECN field to the codepoint at the
166  *  intersection of the appropriate arriving inner header (row) and outer
167  *  header (column) in Figure 4
168  *
169  *      +---------+------------------------------------------------+
170  *      |Arriving |            Arriving Outer Header               |
171  *      |   Inner +---------+------------+------------+------------+
172  *      |  Header | Not-ECT | ECT(0)     | ECT(1)     |     CE     |
173  *      +---------+---------+------------+------------+------------+
174  *      | Not-ECT | Not-ECT |Not-ECT(!!!)|Not-ECT(!!!)| <drop>(!!!)|
175  *      |  ECT(0) |  ECT(0) | ECT(0)     | ECT(1)     |     CE     |
176  *      |  ECT(1) |  ECT(1) | ECT(1) (!) | ECT(1)     |     CE     |
177  *      |    CE   |      CE |     CE     |     CE(!!!)|     CE     |
178  *      +---------+---------+------------+------------+------------+
179  *
180  *             Figure 4: New IP in IP Decapsulation Behaviour
181  *
182  *  returns 0 on success
183  *          1 if something is broken and should be logged (!!! above)
184  *          2 if packet should be dropped
185  */
186 static inline int __INET_ECN_decapsulate(__u8 outer, __u8 inner, bool *set_ce)
187 {
188 	if (INET_ECN_is_not_ect(inner)) {
189 		switch (outer & INET_ECN_MASK) {
190 		case INET_ECN_NOT_ECT:
191 			return 0;
192 		case INET_ECN_ECT_0:
193 		case INET_ECN_ECT_1:
194 			return 1;
195 		case INET_ECN_CE:
196 			return 2;
197 		}
198 	}
199 
200 	*set_ce = INET_ECN_is_ce(outer);
201 	return 0;
202 }
203 
204 static inline int INET_ECN_decapsulate(struct sk_buff *skb,
205 				       __u8 outer, __u8 inner)
206 {
207 	bool set_ce = false;
208 	int rc;
209 
210 	rc = __INET_ECN_decapsulate(outer, inner, &set_ce);
211 	if (!rc && set_ce)
212 		INET_ECN_set_ce(skb);
213 
214 	return rc;
215 }
216 
217 static inline int IP_ECN_decapsulate(const struct iphdr *oiph,
218 				     struct sk_buff *skb)
219 {
220 	__u8 inner;
221 
222 	if (skb->protocol == htons(ETH_P_IP))
223 		inner = ip_hdr(skb)->tos;
224 	else if (skb->protocol == htons(ETH_P_IPV6))
225 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
226 	else
227 		return 0;
228 
229 	return INET_ECN_decapsulate(skb, oiph->tos, inner);
230 }
231 
232 static inline int IP6_ECN_decapsulate(const struct ipv6hdr *oipv6h,
233 				      struct sk_buff *skb)
234 {
235 	__u8 inner;
236 
237 	if (skb->protocol == htons(ETH_P_IP))
238 		inner = ip_hdr(skb)->tos;
239 	else if (skb->protocol == htons(ETH_P_IPV6))
240 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
241 	else
242 		return 0;
243 
244 	return INET_ECN_decapsulate(skb, ipv6_get_dsfield(oipv6h), inner);
245 }
246 #endif
247