inet_ecn.h

  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
 13enum {
 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
 21extern int sysctl_tunnel_ecn_log;
 22
 23static inline int INET_ECN_is_ce(__u8 dsfield)
 24{
 25	return (dsfield & INET_ECN_MASK) == INET_ECN_CE;
 26}
 27
 28static inline int INET_ECN_is_not_ect(__u8 dsfield)
 29{
 30	return (dsfield & INET_ECN_MASK) == INET_ECN_NOT_ECT;
 31}
 32
 33static 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 */
 46static 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/* Apply either ECT(0) or ECT(1) */
 55static inline void __INET_ECN_xmit(struct sock *sk, bool use_ect_1)
 56{
 57	__u8 ect = use_ect_1 ? INET_ECN_ECT_1 : INET_ECN_ECT_0;
 58
 59	/* Mask the complete byte in case the connection alternates between
 60	 * ECT(0) and ECT(1).
 61	 */
 62	inet_sk(sk)->tos &= ~INET_ECN_MASK;
 63	inet_sk(sk)->tos |= ect;
 64	if (inet6_sk(sk)) {
 65		inet6_sk(sk)->tclass &= ~INET_ECN_MASK;
 66		inet6_sk(sk)->tclass |= ect;
 67	}
 68}
 69
 70static inline void INET_ECN_xmit(struct sock *sk)
 71{
 72	__INET_ECN_xmit(sk, false);
 73}
 74
 75static inline void INET_ECN_dontxmit(struct sock *sk)
 76{
 77	inet_sk(sk)->tos &= ~INET_ECN_MASK;
 78	if (inet6_sk(sk) != NULL)
 79		inet6_sk(sk)->tclass &= ~INET_ECN_MASK;
 80}
 81
 82#define IP6_ECN_flow_init(label) do {		\
 83      (label) &= ~htonl(INET_ECN_MASK << 20);	\
 84    } while (0)
 85
 86#define	IP6_ECN_flow_xmit(sk, label) do {				\
 87	if (INET_ECN_is_capable(inet6_sk(sk)->tclass))			\
 88		(label) |= htonl(INET_ECN_ECT_0 << 20);			\
 89    } while (0)
 90
 91static inline int IP_ECN_set_ce(struct iphdr *iph)
 92{
 93	u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
 94	__be16 check_add;
 95
 96	/*
 97	 * After the last operation we have (in binary):
 98	 * INET_ECN_NOT_ECT => 01
 99	 * INET_ECN_ECT_1   => 10
100	 * INET_ECN_ECT_0   => 11
101	 * INET_ECN_CE      => 00
102	 */
103	if (!(ecn & 2))
104		return !ecn;
105
106	/*
107	 * The following gives us:
108	 * INET_ECN_ECT_1 => check += htons(0xFFFD)
109	 * INET_ECN_ECT_0 => check += htons(0xFFFE)
110	 */
111	check_add = (__force __be16)((__force u16)htons(0xFFFB) +
112				     (__force u16)htons(ecn));
113
114	iph->check = csum16_add(iph->check, check_add);
115	iph->tos |= INET_ECN_CE;
116	return 1;
117}
118
119static inline int IP_ECN_set_ect1(struct iphdr *iph)
120{
121	if ((iph->tos & INET_ECN_MASK) != INET_ECN_ECT_0)
122		return 0;
123
124	iph->check = csum16_add(iph->check, htons(0x1));
125	iph->tos ^= INET_ECN_MASK;
126	return 1;
127}
128
129static inline void IP_ECN_clear(struct iphdr *iph)
130{
131	iph->tos &= ~INET_ECN_MASK;
132}
133
134static inline void ipv4_copy_dscp(unsigned int dscp, struct iphdr *inner)
135{
136	dscp &= ~INET_ECN_MASK;
137	ipv4_change_dsfield(inner, INET_ECN_MASK, dscp);
138}
139
140struct ipv6hdr;
141
142/* Note:
143 * IP_ECN_set_ce() has to tweak IPV4 checksum when setting CE,
144 * meaning both changes have no effect on skb->csum if/when CHECKSUM_COMPLETE
145 * In IPv6 case, no checksum compensates the change in IPv6 header,
146 * so we have to update skb->csum.
147 */
148static inline int IP6_ECN_set_ce(struct sk_buff *skb, struct ipv6hdr *iph)
149{
150	__be32 from, to;
151
152	if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph)))
153		return 0;
154
155	from = *(__be32 *)iph;
156	to = from | htonl(INET_ECN_CE << 20);
157	*(__be32 *)iph = to;
158	if (skb->ip_summed == CHECKSUM_COMPLETE)
159		skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from),
160				     (__force __wsum)to);
161	return 1;
162}
163
164static inline int IP6_ECN_set_ect1(struct sk_buff *skb, struct ipv6hdr *iph)
165{
166	__be32 from, to;
167
168	if ((ipv6_get_dsfield(iph) & INET_ECN_MASK) != INET_ECN_ECT_0)
169		return 0;
170
171	from = *(__be32 *)iph;
172	to = from ^ htonl(INET_ECN_MASK << 20);
173	*(__be32 *)iph = to;
174	if (skb->ip_summed == CHECKSUM_COMPLETE)
175		skb->csum = csum_add(csum_sub(skb->csum, (__force __wsum)from),
176				     (__force __wsum)to);
177	return 1;
178}
179
180static inline void ipv6_copy_dscp(unsigned int dscp, struct ipv6hdr *inner)
181{
182	dscp &= ~INET_ECN_MASK;
183	ipv6_change_dsfield(inner, INET_ECN_MASK, dscp);
184}
185
186static inline int INET_ECN_set_ce(struct sk_buff *skb)
187{
188	switch (skb_protocol(skb, true)) {
189	case cpu_to_be16(ETH_P_IP):
190		if (skb_network_header(skb) + sizeof(struct iphdr) <=
191		    skb_tail_pointer(skb))
192			return IP_ECN_set_ce(ip_hdr(skb));
193		break;
194
195	case cpu_to_be16(ETH_P_IPV6):
196		if (skb_network_header(skb) + sizeof(struct ipv6hdr) <=
197		    skb_tail_pointer(skb))
198			return IP6_ECN_set_ce(skb, ipv6_hdr(skb));
199		break;
200	}
201
202	return 0;
203}
204
205static inline int skb_get_dsfield(struct sk_buff *skb)
206{
207	switch (skb_protocol(skb, true)) {
208	case cpu_to_be16(ETH_P_IP):
209		if (!pskb_network_may_pull(skb, sizeof(struct iphdr)))
210			break;
211		return ipv4_get_dsfield(ip_hdr(skb));
212
213	case cpu_to_be16(ETH_P_IPV6):
214		if (!pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))
215			break;
216		return ipv6_get_dsfield(ipv6_hdr(skb));
217	}
218
219	return -1;
220}
221
222static inline int INET_ECN_set_ect1(struct sk_buff *skb)
223{
224	switch (skb_protocol(skb, true)) {
225	case cpu_to_be16(ETH_P_IP):
226		if (skb_network_header(skb) + sizeof(struct iphdr) <=
227		    skb_tail_pointer(skb))
228			return IP_ECN_set_ect1(ip_hdr(skb));
229		break;
230
231	case cpu_to_be16(ETH_P_IPV6):
232		if (skb_network_header(skb) + sizeof(struct ipv6hdr) <=
233		    skb_tail_pointer(skb))
234			return IP6_ECN_set_ect1(skb, ipv6_hdr(skb));
235		break;
236	}
237
238	return 0;
239}
240
241/*
242 * RFC 6040 4.2
243 *  To decapsulate the inner header at the tunnel egress, a compliant
244 *  tunnel egress MUST set the outgoing ECN field to the codepoint at the
245 *  intersection of the appropriate arriving inner header (row) and outer
246 *  header (column) in Figure 4
247 *
248 *      +---------+------------------------------------------------+
249 *      |Arriving |            Arriving Outer Header               |
250 *      |   Inner +---------+------------+------------+------------+
251 *      |  Header | Not-ECT | ECT(0)     | ECT(1)     |     CE     |
252 *      +---------+---------+------------+------------+------------+
253 *      | Not-ECT | Not-ECT |Not-ECT(!!!)|Not-ECT(!!!)| <drop>(!!!)|
254 *      |  ECT(0) |  ECT(0) | ECT(0)     | ECT(1)     |     CE     |
255 *      |  ECT(1) |  ECT(1) | ECT(1) (!) | ECT(1)     |     CE     |
256 *      |    CE   |      CE |     CE     |     CE(!!!)|     CE     |
257 *      +---------+---------+------------+------------+------------+
258 *
259 *             Figure 4: New IP in IP Decapsulation Behaviour
260 *
261 *  returns 0 on success
262 *          1 if something is broken and should be logged (!!! above)
263 *          2 if packet should be dropped
264 */
265static inline int __INET_ECN_decapsulate(__u8 outer, __u8 inner, bool *set_ce)
266{
267	if (INET_ECN_is_not_ect(inner)) {
268		switch (outer & INET_ECN_MASK) {
269		case INET_ECN_NOT_ECT:
270			return 0;
271		case INET_ECN_ECT_0:
272		case INET_ECN_ECT_1:
273			return 1;
274		case INET_ECN_CE:
275			return 2;
276		}
277	}
278
279	*set_ce = INET_ECN_is_ce(outer);
280	return 0;
281}
282
283static inline int INET_ECN_decapsulate(struct sk_buff *skb,
284				       __u8 outer, __u8 inner)
285{
286	bool set_ce = false;
287	int rc;
288
289	rc = __INET_ECN_decapsulate(outer, inner, &set_ce);
290	if (!rc) {
291		if (set_ce)
292			INET_ECN_set_ce(skb);
293		else if ((outer & INET_ECN_MASK) == INET_ECN_ECT_1)
294			INET_ECN_set_ect1(skb);
295	}
296
297	return rc;
298}
299
300static inline int IP_ECN_decapsulate(const struct iphdr *oiph,
301				     struct sk_buff *skb)
302{
303	__u8 inner;
304
305	switch (skb_protocol(skb, true)) {
306	case htons(ETH_P_IP):
307		inner = ip_hdr(skb)->tos;
308		break;
309	case htons(ETH_P_IPV6):
310		inner = ipv6_get_dsfield(ipv6_hdr(skb));
311		break;
312	default:
313		return 0;
314	}
315
316	return INET_ECN_decapsulate(skb, oiph->tos, inner);
317}
318
319static inline int IP6_ECN_decapsulate(const struct ipv6hdr *oipv6h,
320				      struct sk_buff *skb)
321{
322	__u8 inner;
323
324	switch (skb_protocol(skb, true)) {
325	case htons(ETH_P_IP):
326		inner = ip_hdr(skb)->tos;
327		break;
328	case htons(ETH_P_IPV6):
329		inner = ipv6_get_dsfield(ipv6_hdr(skb));
330		break;
331	default:
332		return 0;
333	}
334
335	return INET_ECN_decapsulate(skb, ipv6_get_dsfield(oipv6h), inner);
336}
337#endif