1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Authors:
4 * (C) 2020 Alexander Aring <alex.aring@gmail.com>
5 */
6
7#include <linux/rpl_iptunnel.h>
8
9#include <net/dst_cache.h>
10#include <net/ip6_route.h>
11#include <net/lwtunnel.h>
12#include <net/ipv6.h>
13#include <net/rpl.h>
14
15struct rpl_iptunnel_encap {
16	DECLARE_FLEX_ARRAY(struct ipv6_rpl_sr_hdr, srh);
17};
18
19struct rpl_lwt {
20	struct dst_cache cache;
21	struct rpl_iptunnel_encap tuninfo;
22};
23
24static inline struct rpl_lwt *rpl_lwt_lwtunnel(struct lwtunnel_state *lwt)
25{
26	return (struct rpl_lwt *)lwt->data;
27}
28
29static inline struct rpl_iptunnel_encap *
30rpl_encap_lwtunnel(struct lwtunnel_state *lwt)
31{
32	return &rpl_lwt_lwtunnel(lwt)->tuninfo;
33}
34
35static const struct nla_policy rpl_iptunnel_policy[RPL_IPTUNNEL_MAX + 1] = {
36	[RPL_IPTUNNEL_SRH]	= { .type = NLA_BINARY },
37};
38
39static bool rpl_validate_srh(struct net *net, struct ipv6_rpl_sr_hdr *srh,
40			     size_t seglen)
41{
42	int err;
43
44	if ((srh->hdrlen << 3) != seglen)
45		return false;
46
47	/* check at least one segment and seglen fit with segments_left */
48	if (!srh->segments_left ||
49	    (srh->segments_left * sizeof(struct in6_addr)) != seglen)
50		return false;
51
52	if (srh->cmpri || srh->cmpre)
53		return false;
54
55	err = ipv6_chk_rpl_srh_loop(net, srh->rpl_segaddr,
56				    srh->segments_left);
57	if (err)
58		return false;
59
60	if (ipv6_addr_type(&srh->rpl_segaddr[srh->segments_left - 1]) &
61	    IPV6_ADDR_MULTICAST)
62		return false;
63
64	return true;
65}
66
67static int rpl_build_state(struct net *net, struct nlattr *nla,
68			   unsigned int family, const void *cfg,
69			   struct lwtunnel_state **ts,
70			   struct netlink_ext_ack *extack)
71{
72	struct nlattr *tb[RPL_IPTUNNEL_MAX + 1];
73	struct lwtunnel_state *newts;
74	struct ipv6_rpl_sr_hdr *srh;
75	struct rpl_lwt *rlwt;
76	int err, srh_len;
77
78	if (family != AF_INET6)
79		return -EINVAL;
80
81	err = nla_parse_nested(tb, RPL_IPTUNNEL_MAX, nla,
82			       rpl_iptunnel_policy, extack);
83	if (err < 0)
84		return err;
85
86	if (!tb[RPL_IPTUNNEL_SRH])
87		return -EINVAL;
88
89	srh = nla_data(tb[RPL_IPTUNNEL_SRH]);
90	srh_len = nla_len(tb[RPL_IPTUNNEL_SRH]);
91
92	if (srh_len < sizeof(*srh))
93		return -EINVAL;
94
95	/* verify that SRH is consistent */
96	if (!rpl_validate_srh(net, srh, srh_len - sizeof(*srh)))
97		return -EINVAL;
98
99	newts = lwtunnel_state_alloc(srh_len + sizeof(*rlwt));
100	if (!newts)
101		return -ENOMEM;
102
103	rlwt = rpl_lwt_lwtunnel(newts);
104
105	err = dst_cache_init(&rlwt->cache, GFP_ATOMIC);
106	if (err) {
107		kfree(newts);
108		return err;
109	}
110
111	memcpy(&rlwt->tuninfo.srh, srh, srh_len);
112
113	newts->type = LWTUNNEL_ENCAP_RPL;
114	newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
115	newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
116
117	*ts = newts;
118
119	return 0;
120}
121
122static void rpl_destroy_state(struct lwtunnel_state *lwt)
123{
124	dst_cache_destroy(&rpl_lwt_lwtunnel(lwt)->cache);
125}
126
127static int rpl_do_srh_inline(struct sk_buff *skb, const struct rpl_lwt *rlwt,
128			     const struct ipv6_rpl_sr_hdr *srh)
129{
130	struct ipv6_rpl_sr_hdr *isrh, *csrh;
131	const struct ipv6hdr *oldhdr;
132	struct ipv6hdr *hdr;
133	unsigned char *buf;
134	size_t hdrlen;
135	int err;
136
137	oldhdr = ipv6_hdr(skb);
138
139	buf = kcalloc(struct_size(srh, segments.addr, srh->segments_left), 2, GFP_ATOMIC);
140	if (!buf)
141		return -ENOMEM;
142
143	isrh = (struct ipv6_rpl_sr_hdr *)buf;
144	csrh = (struct ipv6_rpl_sr_hdr *)(buf + ((srh->hdrlen + 1) << 3));
145
146	memcpy(isrh, srh, sizeof(*isrh));
147	memcpy(isrh->rpl_segaddr, &srh->rpl_segaddr[1],
148	       (srh->segments_left - 1) * 16);
149	isrh->rpl_segaddr[srh->segments_left - 1] = oldhdr->daddr;
150
151	ipv6_rpl_srh_compress(csrh, isrh, &srh->rpl_segaddr[0],
152			      isrh->segments_left - 1);
153
154	hdrlen = ((csrh->hdrlen + 1) << 3);
155
156	err = skb_cow_head(skb, hdrlen + skb->mac_len);
157	if (unlikely(err)) {
158		kfree(buf);
159		return err;
160	}
161
162	skb_pull(skb, sizeof(struct ipv6hdr));
163	skb_postpull_rcsum(skb, skb_network_header(skb),
164			   sizeof(struct ipv6hdr));
165
166	skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
167	skb_reset_network_header(skb);
168	skb_mac_header_rebuild(skb);
169
170	hdr = ipv6_hdr(skb);
171	memmove(hdr, oldhdr, sizeof(*hdr));
172	isrh = (void *)hdr + sizeof(*hdr);
173	memcpy(isrh, csrh, hdrlen);
174
175	isrh->nexthdr = hdr->nexthdr;
176	hdr->nexthdr = NEXTHDR_ROUTING;
177	hdr->daddr = srh->rpl_segaddr[0];
178
179	ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
180	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
181
182	skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
183
184	kfree(buf);
185
186	return 0;
187}
188
189static int rpl_do_srh(struct sk_buff *skb, const struct rpl_lwt *rlwt)
190{
191	struct dst_entry *dst = skb_dst(skb);
192	struct rpl_iptunnel_encap *tinfo;
193
194	if (skb->protocol != htons(ETH_P_IPV6))
195		return -EINVAL;
196
197	tinfo = rpl_encap_lwtunnel(dst->lwtstate);
198
199	return rpl_do_srh_inline(skb, rlwt, tinfo->srh);
200}
201
202static int rpl_output(struct net *net, struct sock *sk, struct sk_buff *skb)
203{
204	struct dst_entry *orig_dst = skb_dst(skb);
205	struct dst_entry *dst = NULL;
206	struct rpl_lwt *rlwt;
207	int err;
208
209	rlwt = rpl_lwt_lwtunnel(orig_dst->lwtstate);
210
211	err = rpl_do_srh(skb, rlwt);
212	if (unlikely(err))
213		goto drop;
214
215	preempt_disable();
216	dst = dst_cache_get(&rlwt->cache);
217	preempt_enable();
218
219	if (unlikely(!dst)) {
220		struct ipv6hdr *hdr = ipv6_hdr(skb);
221		struct flowi6 fl6;
222
223		memset(&fl6, 0, sizeof(fl6));
224		fl6.daddr = hdr->daddr;
225		fl6.saddr = hdr->saddr;
226		fl6.flowlabel = ip6_flowinfo(hdr);
227		fl6.flowi6_mark = skb->mark;
228		fl6.flowi6_proto = hdr->nexthdr;
229
230		dst = ip6_route_output(net, NULL, &fl6);
231		if (dst->error) {
232			err = dst->error;
233			dst_release(dst);
234			goto drop;
235		}
236
237		preempt_disable();
238		dst_cache_set_ip6(&rlwt->cache, dst, &fl6.saddr);
239		preempt_enable();
240	}
241
242	skb_dst_drop(skb);
243	skb_dst_set(skb, dst);
244
245	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
246	if (unlikely(err))
247		goto drop;
248
249	return dst_output(net, sk, skb);
250
251drop:
252	kfree_skb(skb);
253	return err;
254}
255
256static int rpl_input(struct sk_buff *skb)
257{
258	struct dst_entry *orig_dst = skb_dst(skb);
259	struct dst_entry *dst = NULL;
260	struct rpl_lwt *rlwt;
261	int err;
262
263	rlwt = rpl_lwt_lwtunnel(orig_dst->lwtstate);
264
265	err = rpl_do_srh(skb, rlwt);
266	if (unlikely(err)) {
267		kfree_skb(skb);
268		return err;
269	}
270
271	preempt_disable();
272	dst = dst_cache_get(&rlwt->cache);
273	preempt_enable();
274
275	if (!dst) {
276		ip6_route_input(skb);
277		dst = skb_dst(skb);
278		if (!dst->error) {
279			preempt_disable();
280			dst_cache_set_ip6(&rlwt->cache, dst,
281					  &ipv6_hdr(skb)->saddr);
282			preempt_enable();
283		}
284	} else {
285		skb_dst_drop(skb);
286		skb_dst_set(skb, dst);
287	}
288
289	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
290	if (unlikely(err))
291		return err;
292
293	return dst_input(skb);
294}
295
296static int nla_put_rpl_srh(struct sk_buff *skb, int attrtype,
297			   struct rpl_iptunnel_encap *tuninfo)
298{
299	struct rpl_iptunnel_encap *data;
300	struct nlattr *nla;
301	int len;
302
303	len = RPL_IPTUNNEL_SRH_SIZE(tuninfo->srh);
304
305	nla = nla_reserve(skb, attrtype, len);
306	if (!nla)
307		return -EMSGSIZE;
308
309	data = nla_data(nla);
310	memcpy(data, tuninfo->srh, len);
311
312	return 0;
313}
314
315static int rpl_fill_encap_info(struct sk_buff *skb,
316			       struct lwtunnel_state *lwtstate)
317{
318	struct rpl_iptunnel_encap *tuninfo = rpl_encap_lwtunnel(lwtstate);
319
320	if (nla_put_rpl_srh(skb, RPL_IPTUNNEL_SRH, tuninfo))
321		return -EMSGSIZE;
322
323	return 0;
324}
325
326static int rpl_encap_nlsize(struct lwtunnel_state *lwtstate)
327{
328	struct rpl_iptunnel_encap *tuninfo = rpl_encap_lwtunnel(lwtstate);
329
330	return nla_total_size(RPL_IPTUNNEL_SRH_SIZE(tuninfo->srh));
331}
332
333static int rpl_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
334{
335	struct rpl_iptunnel_encap *a_hdr = rpl_encap_lwtunnel(a);
336	struct rpl_iptunnel_encap *b_hdr = rpl_encap_lwtunnel(b);
337	int len = RPL_IPTUNNEL_SRH_SIZE(a_hdr->srh);
338
339	if (len != RPL_IPTUNNEL_SRH_SIZE(b_hdr->srh))
340		return 1;
341
342	return memcmp(a_hdr, b_hdr, len);
343}
344
345static const struct lwtunnel_encap_ops rpl_ops = {
346	.build_state	= rpl_build_state,
347	.destroy_state	= rpl_destroy_state,
348	.output		= rpl_output,
349	.input		= rpl_input,
350	.fill_encap	= rpl_fill_encap_info,
351	.get_encap_size	= rpl_encap_nlsize,
352	.cmp_encap	= rpl_encap_cmp,
353	.owner		= THIS_MODULE,
354};
355
356int __init rpl_init(void)
357{
358	int err;
359
360	err = lwtunnel_encap_add_ops(&rpl_ops, LWTUNNEL_ENCAP_RPL);
361	if (err)
362		goto out;
363
364	pr_info("RPL Segment Routing with IPv6\n");
365
366	return 0;
367
368out:
369	return err;
370}
371
372void rpl_exit(void)
373{
374	lwtunnel_encap_del_ops(&rpl_ops, LWTUNNEL_ENCAP_RPL);
375}
376