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 
INET_ECN_is_ce(__u8 dsfield)22 static inline int INET_ECN_is_ce(__u8 dsfield)
23 {
24 	return (dsfield & INET_ECN_MASK) == INET_ECN_CE;
25 }
26 
INET_ECN_is_not_ect(__u8 dsfield)27 static inline int INET_ECN_is_not_ect(__u8 dsfield)
28 {
29 	return (dsfield & INET_ECN_MASK) == INET_ECN_NOT_ECT;
30 }
31 
INET_ECN_is_capable(__u8 dsfield)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  */
INET_ECN_encapsulate(__u8 outer,__u8 inner)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 
INET_ECN_xmit(struct sock * sk)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 
INET_ECN_dontxmit(struct sock * sk)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 
IP_ECN_set_ce(struct iphdr * iph)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 
IP_ECN_clear(struct iphdr * iph)103 static inline void IP_ECN_clear(struct iphdr *iph)
104 {
105 	iph->tos &= ~INET_ECN_MASK;
106 }
107 
ipv4_copy_dscp(unsigned int dscp,struct iphdr * inner)108 static inline void ipv4_copy_dscp(unsigned int dscp, struct iphdr *inner)
109 {
110 	dscp &= ~INET_ECN_MASK;
111 	ipv4_change_dsfield(inner, INET_ECN_MASK, dscp);
112 }
113 
114 struct ipv6hdr;
115 
116 /* Note:
117  * IP_ECN_set_ce() has to tweak IPV4 checksum when setting CE,
118  * meaning both changes have no effect on skb->csum if/when CHECKSUM_COMPLETE
119  * In IPv6 case, no checksum compensates the change in IPv6 header,
120  * so we have to update skb->csum.
121  */
IP6_ECN_set_ce(struct sk_buff * skb,struct ipv6hdr * iph)122 static inline int IP6_ECN_set_ce(struct sk_buff *skb, struct ipv6hdr *iph)
123 {
124 	__be32 from, to;
125 
126 	if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph)))
127 		return 0;
128 
129 	from = *(__be32 *)iph;
130 	to = from | htonl(INET_ECN_CE << 20);
131 	*(__be32 *)iph = to;
132 	if (skb->ip_summed == CHECKSUM_COMPLETE)
133 		skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from),
134 				     (__force __wsum)to);
135 	return 1;
136 }
137 
ipv6_copy_dscp(unsigned int dscp,struct ipv6hdr * inner)138 static inline void ipv6_copy_dscp(unsigned int dscp, struct ipv6hdr *inner)
139 {
140 	dscp &= ~INET_ECN_MASK;
141 	ipv6_change_dsfield(inner, INET_ECN_MASK, dscp);
142 }
143 
INET_ECN_set_ce(struct sk_buff * skb)144 static inline int INET_ECN_set_ce(struct sk_buff *skb)
145 {
146 	switch (skb_protocol(skb, true)) {
147 	case cpu_to_be16(ETH_P_IP):
148 		if (skb_network_header(skb) + sizeof(struct iphdr) <=
149 		    skb_tail_pointer(skb))
150 			return IP_ECN_set_ce(ip_hdr(skb));
151 		break;
152 
153 	case cpu_to_be16(ETH_P_IPV6):
154 		if (skb_network_header(skb) + sizeof(struct ipv6hdr) <=
155 		    skb_tail_pointer(skb))
156 			return IP6_ECN_set_ce(skb, ipv6_hdr(skb));
157 		break;
158 	}
159 
160 	return 0;
161 }
162 
163 /*
164  * RFC 6040 4.2
165  *  To decapsulate the inner header at the tunnel egress, a compliant
166  *  tunnel egress MUST set the outgoing ECN field to the codepoint at the
167  *  intersection of the appropriate arriving inner header (row) and outer
168  *  header (column) in Figure 4
169  *
170  *      +---------+------------------------------------------------+
171  *      |Arriving |            Arriving Outer Header               |
172  *      |   Inner +---------+------------+------------+------------+
173  *      |  Header | Not-ECT | ECT(0)     | ECT(1)     |     CE     |
174  *      +---------+---------+------------+------------+------------+
175  *      | Not-ECT | Not-ECT |Not-ECT(!!!)|Not-ECT(!!!)| <drop>(!!!)|
176  *      |  ECT(0) |  ECT(0) | ECT(0)     | ECT(1)     |     CE     |
177  *      |  ECT(1) |  ECT(1) | ECT(1) (!) | ECT(1)     |     CE     |
178  *      |    CE   |      CE |     CE     |     CE(!!!)|     CE     |
179  *      +---------+---------+------------+------------+------------+
180  *
181  *             Figure 4: New IP in IP Decapsulation Behaviour
182  *
183  *  returns 0 on success
184  *          1 if something is broken and should be logged (!!! above)
185  *          2 if packet should be dropped
186  */
INET_ECN_decapsulate(struct sk_buff * skb,__u8 outer,__u8 inner)187 static inline int INET_ECN_decapsulate(struct sk_buff *skb,
188 				       __u8 outer, __u8 inner)
189 {
190 	if (INET_ECN_is_not_ect(inner)) {
191 		switch (outer & INET_ECN_MASK) {
192 		case INET_ECN_NOT_ECT:
193 			return 0;
194 		case INET_ECN_ECT_0:
195 		case INET_ECN_ECT_1:
196 			return 1;
197 		case INET_ECN_CE:
198 			return 2;
199 		}
200 	}
201 
202 	if (INET_ECN_is_ce(outer))
203 		INET_ECN_set_ce(skb);
204 
205 	return 0;
206 }
207 
IP_ECN_decapsulate(const struct iphdr * oiph,struct sk_buff * skb)208 static inline int IP_ECN_decapsulate(const struct iphdr *oiph,
209 				     struct sk_buff *skb)
210 {
211 	__u8 inner;
212 
213 	switch (skb_protocol(skb, true)) {
214 	case htons(ETH_P_IP):
215 		inner = ip_hdr(skb)->tos;
216 		break;
217 	case htons(ETH_P_IPV6):
218 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
219 		break;
220 	default:
221 		return 0;
222 	}
223 
224 	return INET_ECN_decapsulate(skb, oiph->tos, inner);
225 }
226 
IP6_ECN_decapsulate(const struct ipv6hdr * oipv6h,struct sk_buff * skb)227 static inline int IP6_ECN_decapsulate(const struct ipv6hdr *oipv6h,
228 				      struct sk_buff *skb)
229 {
230 	__u8 inner;
231 
232 	switch (skb_protocol(skb, true)) {
233 	case htons(ETH_P_IP):
234 		inner = ip_hdr(skb)->tos;
235 		break;
236 	case htons(ETH_P_IPV6):
237 		inner = ipv6_get_dsfield(ipv6_hdr(skb));
238 		break;
239 	default:
240 		return 0;
241 	}
242 
243 	return INET_ECN_decapsulate(skb, ipv6_get_dsfield(oipv6h), inner);
244 }
245 #endif
246