xref: /openbmc/linux/include/net/inet_ecn.h (revision 4f6cce39)
1 #ifndef _INET_ECN_H_
2 #define _INET_ECN_H_
3 
4 #include <linux/ip.h>
5 #include <linux/skbuff.h>
6 
7 #include <net/inet_sock.h>
8 #include <net/dsfield.h>
9 
10 enum {
11 	INET_ECN_NOT_ECT = 0,
12 	INET_ECN_ECT_1 = 1,
13 	INET_ECN_ECT_0 = 2,
14 	INET_ECN_CE = 3,
15 	INET_ECN_MASK = 3,
16 };
17 
18 extern int sysctl_tunnel_ecn_log;
19 
20 static inline int INET_ECN_is_ce(__u8 dsfield)
21 {
22 	return (dsfield & INET_ECN_MASK) == INET_ECN_CE;
23 }
24 
25 static inline int INET_ECN_is_not_ect(__u8 dsfield)
26 {
27 	return (dsfield & INET_ECN_MASK) == INET_ECN_NOT_ECT;
28 }
29 
30 static inline int INET_ECN_is_capable(__u8 dsfield)
31 {
32 	return dsfield & INET_ECN_ECT_0;
33 }
34 
35 /*
36  * RFC 3168 9.1.1
37  *  The full-functionality option for ECN encapsulation is to copy the
38  *  ECN codepoint of the inside header to the outside header on
39  *  encapsulation if the inside header is not-ECT or ECT, and to set the
40  *  ECN codepoint of the outside header to ECT(0) if the ECN codepoint of
41  *  the inside header is CE.
42  */
43 static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner)
44 {
45 	outer &= ~INET_ECN_MASK;
46 	outer |= !INET_ECN_is_ce(inner) ? (inner & INET_ECN_MASK) :
47 					  INET_ECN_ECT_0;
48 	return outer;
49 }
50 
51 static inline void INET_ECN_xmit(struct sock *sk)
52 {
53 	inet_sk(sk)->tos |= INET_ECN_ECT_0;
54 	if (inet6_sk(sk) != NULL)
55 		inet6_sk(sk)->tclass |= INET_ECN_ECT_0;
56 }
57 
58 static inline void INET_ECN_dontxmit(struct sock *sk)
59 {
60 	inet_sk(sk)->tos &= ~INET_ECN_MASK;
61 	if (inet6_sk(sk) != NULL)
62 		inet6_sk(sk)->tclass &= ~INET_ECN_MASK;
63 }
64 
65 #define IP6_ECN_flow_init(label) do {		\
66       (label) &= ~htonl(INET_ECN_MASK << 20);	\
67     } while (0)
68 
69 #define	IP6_ECN_flow_xmit(sk, label) do {				\
70 	if (INET_ECN_is_capable(inet6_sk(sk)->tclass))			\
71 		(label) |= htonl(INET_ECN_ECT_0 << 20);			\
72     } while (0)
73 
74 static inline int IP_ECN_set_ce(struct iphdr *iph)
75 {
76 	u32 check = (__force u32)iph->check;
77 	u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
78 
79 	/*
80 	 * After the last operation we have (in binary):
81 	 * INET_ECN_NOT_ECT => 01
82 	 * INET_ECN_ECT_1   => 10
83 	 * INET_ECN_ECT_0   => 11
84 	 * INET_ECN_CE      => 00
85 	 */
86 	if (!(ecn & 2))
87 		return !ecn;
88 
89 	/*
90 	 * The following gives us:
91 	 * INET_ECN_ECT_1 => check += htons(0xFFFD)
92 	 * INET_ECN_ECT_0 => check += htons(0xFFFE)
93 	 */
94 	check += (__force u16)htons(0xFFFB) + (__force u16)htons(ecn);
95 
96 	iph->check = (__force __sum16)(check + (check>=0xFFFF));
97 	iph->tos |= INET_ECN_CE;
98 	return 1;
99 }
100 
101 static inline void IP_ECN_clear(struct iphdr *iph)
102 {
103 	iph->tos &= ~INET_ECN_MASK;
104 }
105 
106 static inline void ipv4_copy_dscp(unsigned int dscp, struct iphdr *inner)
107 {
108 	dscp &= ~INET_ECN_MASK;
109 	ipv4_change_dsfield(inner, INET_ECN_MASK, dscp);
110 }
111 
112 struct ipv6hdr;
113 
114 /* Note:
115  * IP_ECN_set_ce() has to tweak IPV4 checksum when setting CE,
116  * meaning both changes have no effect on skb->csum if/when CHECKSUM_COMPLETE
117  * In IPv6 case, no checksum compensates the change in IPv6 header,
118  * so we have to update skb->csum.
119  */
120 static inline int IP6_ECN_set_ce(struct sk_buff *skb, struct ipv6hdr *iph)
121 {
122 	__be32 from, to;
123 
124 	if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph)))
125 		return 0;
126 
127 	from = *(__be32 *)iph;
128 	to = from | htonl(INET_ECN_CE << 20);
129 	*(__be32 *)iph = to;
130 	if (skb->ip_summed == CHECKSUM_COMPLETE)
131 		skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from),
132 				     (__force __wsum)to);
133 	return 1;
134 }
135 
136 static inline void IP6_ECN_clear(struct ipv6hdr *iph)
137 {
138 	*(__be32*)iph &= ~htonl(INET_ECN_MASK << 20);
139 }
140 
141 static inline void ipv6_copy_dscp(unsigned int dscp, struct ipv6hdr *inner)
142 {
143 	dscp &= ~INET_ECN_MASK;
144 	ipv6_change_dsfield(inner, INET_ECN_MASK, dscp);
145 }
146 
147 static inline int INET_ECN_set_ce(struct sk_buff *skb)
148 {
149 	switch (skb->protocol) {
150 	case cpu_to_be16(ETH_P_IP):
151 		if (skb_network_header(skb) + sizeof(struct iphdr) <=
152 		    skb_tail_pointer(skb))
153 			return IP_ECN_set_ce(ip_hdr(skb));
154 		break;
155 
156 	case cpu_to_be16(ETH_P_IPV6):
157 		if (skb_network_header(skb) + sizeof(struct ipv6hdr) <=
158 		    skb_tail_pointer(skb))
159 			return IP6_ECN_set_ce(skb, ipv6_hdr(skb));
160 		break;
161 	}
162 
163 	return 0;
164 }
165 
166 /*
167  * RFC 6040 4.2
168  *  To decapsulate the inner header at the tunnel egress, a compliant
169  *  tunnel egress MUST set the outgoing ECN field to the codepoint at the
170  *  intersection of the appropriate arriving inner header (row) and outer
171  *  header (column) in Figure 4
172  *
173  *      +---------+------------------------------------------------+
174  *      |Arriving |            Arriving Outer Header               |
175  *      |   Inner +---------+------------+------------+------------+
176  *      |  Header | Not-ECT | ECT(0)     | ECT(1)     |     CE     |
177  *      +---------+---------+------------+------------+------------+
178  *      | Not-ECT | Not-ECT |Not-ECT(!!!)|Not-ECT(!!!)| <drop>(!!!)|
179  *      |  ECT(0) |  ECT(0) | ECT(0)     | ECT(1)     |     CE     |
180  *      |  ECT(1) |  ECT(1) | ECT(1) (!) | ECT(1)     |     CE     |
181  *      |    CE   |      CE |     CE     |     CE(!!!)|     CE     |
182  *      +---------+---------+------------+------------+------------+
183  *
184  *             Figure 4: New IP in IP Decapsulation Behaviour
185  *
186  *  returns 0 on success
187  *          1 if something is broken and should be logged (!!! above)
188  *          2 if packet should be dropped
189  */
190 static inline int INET_ECN_decapsulate(struct sk_buff *skb,
191 				       __u8 outer, __u8 inner)
192 {
193 	if (INET_ECN_is_not_ect(inner)) {
194 		switch (outer & INET_ECN_MASK) {
195 		case INET_ECN_NOT_ECT:
196 			return 0;
197 		case INET_ECN_ECT_0:
198 		case INET_ECN_ECT_1:
199 			return 1;
200 		case INET_ECN_CE:
201 			return 2;
202 		}
203 	}
204 
205 	if (INET_ECN_is_ce(outer))
206 		INET_ECN_set_ce(skb);
207 
208 	return 0;
209 }
210 
211 static inline int IP_ECN_decapsulate(const struct iphdr *oiph,
212 				     struct sk_buff *skb)
213 {
214 	__u8 inner;
215 
216 	if (skb->protocol == htons(ETH_P_IP))
217 		inner = ip_hdr(skb)->tos;
218 	else if (skb->protocol == htons(ETH_P_IPV6))
219 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
220 	else
221 		return 0;
222 
223 	return INET_ECN_decapsulate(skb, oiph->tos, inner);
224 }
225 
226 static inline int IP6_ECN_decapsulate(const struct ipv6hdr *oipv6h,
227 				      struct sk_buff *skb)
228 {
229 	__u8 inner;
230 
231 	if (skb->protocol == htons(ETH_P_IP))
232 		inner = ip_hdr(skb)->tos;
233 	else if (skb->protocol == htons(ETH_P_IPV6))
234 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
235 	else
236 		return 0;
237 
238 	return INET_ECN_decapsulate(skb, ipv6_get_dsfield(oipv6h), inner);
239 }
240 #endif
241