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
15 struct rpl_iptunnel_encap {
16 DECLARE_FLEX_ARRAY(struct ipv6_rpl_sr_hdr, srh);
17 };
18
19 struct rpl_lwt {
20 struct dst_cache cache;
21 struct rpl_iptunnel_encap tuninfo;
22 };
23
rpl_lwt_lwtunnel(struct lwtunnel_state * lwt)24 static inline struct rpl_lwt *rpl_lwt_lwtunnel(struct lwtunnel_state *lwt)
25 {
26 return (struct rpl_lwt *)lwt->data;
27 }
28
29 static inline struct rpl_iptunnel_encap *
rpl_encap_lwtunnel(struct lwtunnel_state * lwt)30 rpl_encap_lwtunnel(struct lwtunnel_state *lwt)
31 {
32 return &rpl_lwt_lwtunnel(lwt)->tuninfo;
33 }
34
35 static const struct nla_policy rpl_iptunnel_policy[RPL_IPTUNNEL_MAX + 1] = {
36 [RPL_IPTUNNEL_SRH] = { .type = NLA_BINARY },
37 };
38
rpl_validate_srh(struct net * net,struct ipv6_rpl_sr_hdr * srh,size_t seglen)39 static 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
rpl_build_state(struct net * net,struct nlattr * nla,unsigned int family,const void * cfg,struct lwtunnel_state ** ts,struct netlink_ext_ack * extack)67 static 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
rpl_destroy_state(struct lwtunnel_state * lwt)122 static void rpl_destroy_state(struct lwtunnel_state *lwt)
123 {
124 dst_cache_destroy(&rpl_lwt_lwtunnel(lwt)->cache);
125 }
126
rpl_do_srh_inline(struct sk_buff * skb,const struct rpl_lwt * rlwt,const struct ipv6_rpl_sr_hdr * srh,struct dst_entry * cache_dst)127 static int rpl_do_srh_inline(struct sk_buff *skb, const struct rpl_lwt *rlwt,
128 const struct ipv6_rpl_sr_hdr *srh,
129 struct dst_entry *cache_dst)
130 {
131 struct ipv6_rpl_sr_hdr *isrh, *csrh;
132 struct ipv6hdr oldhdr;
133 struct ipv6hdr *hdr;
134 unsigned char *buf;
135 size_t hdrlen;
136 int err;
137
138 memcpy(&oldhdr, ipv6_hdr(skb), sizeof(oldhdr));
139
140 buf = kcalloc(struct_size(srh, segments.addr, srh->segments_left), 2, GFP_ATOMIC);
141 if (!buf)
142 return -ENOMEM;
143
144 isrh = (struct ipv6_rpl_sr_hdr *)buf;
145 csrh = (struct ipv6_rpl_sr_hdr *)(buf + ((srh->hdrlen + 1) << 3));
146
147 memcpy(isrh, srh, sizeof(*isrh));
148 memcpy(isrh->rpl_segaddr, &srh->rpl_segaddr[1],
149 (srh->segments_left - 1) * 16);
150 isrh->rpl_segaddr[srh->segments_left - 1] = oldhdr.daddr;
151
152 ipv6_rpl_srh_compress(csrh, isrh, &srh->rpl_segaddr[0],
153 isrh->segments_left - 1);
154
155 hdrlen = ((csrh->hdrlen + 1) << 3);
156
157 err = skb_cow_head(skb, hdrlen + dst_dev_overhead(cache_dst, skb));
158 if (unlikely(err)) {
159 kfree(buf);
160 return err;
161 }
162
163 skb_pull(skb, sizeof(struct ipv6hdr));
164 skb_postpull_rcsum(skb, skb_network_header(skb),
165 sizeof(struct ipv6hdr));
166
167 skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
168 skb_reset_network_header(skb);
169 skb_mac_header_rebuild(skb);
170
171 hdr = ipv6_hdr(skb);
172 memmove(hdr, &oldhdr, sizeof(*hdr));
173 isrh = (void *)hdr + sizeof(*hdr);
174 memcpy(isrh, csrh, hdrlen);
175
176 isrh->nexthdr = hdr->nexthdr;
177 hdr->nexthdr = NEXTHDR_ROUTING;
178 hdr->daddr = srh->rpl_segaddr[0];
179
180 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
181 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
182
183 skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
184
185 kfree(buf);
186
187 return 0;
188 }
189
rpl_do_srh(struct sk_buff * skb,const struct rpl_lwt * rlwt,struct dst_entry * cache_dst)190 static int rpl_do_srh(struct sk_buff *skb, const struct rpl_lwt *rlwt,
191 struct dst_entry *cache_dst)
192 {
193 struct dst_entry *dst = skb_dst(skb);
194 struct rpl_iptunnel_encap *tinfo;
195
196 if (skb->protocol != htons(ETH_P_IPV6))
197 return -EINVAL;
198
199 tinfo = rpl_encap_lwtunnel(dst->lwtstate);
200
201 return rpl_do_srh_inline(skb, rlwt, tinfo->srh, cache_dst);
202 }
203
rpl_output(struct net * net,struct sock * sk,struct sk_buff * skb)204 static int rpl_output(struct net *net, struct sock *sk, struct sk_buff *skb)
205 {
206 struct dst_entry *orig_dst = skb_dst(skb);
207 struct dst_entry *dst = NULL;
208 struct rpl_lwt *rlwt;
209 int err;
210
211 rlwt = rpl_lwt_lwtunnel(orig_dst->lwtstate);
212
213 local_bh_disable();
214 dst = dst_cache_get(&rlwt->cache);
215 local_bh_enable();
216
217 err = rpl_do_srh(skb, rlwt, dst);
218 if (unlikely(err))
219 goto drop;
220
221 if (unlikely(!dst)) {
222 struct ipv6hdr *hdr = ipv6_hdr(skb);
223 struct flowi6 fl6;
224
225 memset(&fl6, 0, sizeof(fl6));
226 fl6.daddr = hdr->daddr;
227 fl6.saddr = hdr->saddr;
228 fl6.flowlabel = ip6_flowinfo(hdr);
229 fl6.flowi6_mark = skb->mark;
230 fl6.flowi6_proto = hdr->nexthdr;
231
232 dst = ip6_route_output(net, NULL, &fl6);
233 if (dst->error) {
234 err = dst->error;
235 goto drop;
236 }
237
238 /* cache only if we don't create a dst reference loop */
239 if (orig_dst->lwtstate != dst->lwtstate) {
240 local_bh_disable();
241 dst_cache_set_ip6(&rlwt->cache, dst, &fl6.saddr);
242 local_bh_enable();
243 }
244
245 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst_dev(dst)));
246 if (unlikely(err))
247 goto drop;
248 }
249
250 skb_dst_drop(skb);
251 skb_dst_set(skb, dst);
252
253 return dst_output(net, sk, skb);
254
255 drop:
256 dst_release(dst);
257 kfree_skb(skb);
258 return err;
259 }
260
rpl_input(struct sk_buff * skb)261 static int rpl_input(struct sk_buff *skb)
262 {
263 struct dst_entry *orig_dst = skb_dst(skb);
264 struct dst_entry *dst = NULL;
265 struct lwtunnel_state *lwtst;
266 struct rpl_lwt *rlwt;
267 int err;
268
269 /* We cannot dereference "orig_dst" once ip6_route_input() or
270 * skb_dst_drop() is called. However, in order to detect a dst loop, we
271 * need the address of its lwtstate. So, save the address of lwtstate
272 * now and use it later as a comparison.
273 */
274 lwtst = orig_dst->lwtstate;
275
276 rlwt = rpl_lwt_lwtunnel(lwtst);
277
278 local_bh_disable();
279 dst = dst_cache_get(&rlwt->cache);
280 local_bh_enable();
281
282 err = rpl_do_srh(skb, rlwt, dst);
283 if (unlikely(err)) {
284 dst_release(dst);
285 goto drop;
286 }
287
288 if (!dst) {
289 ip6_route_input(skb);
290
291 /* ip6_route_input() sets a NOREF dst; force a refcount on it
292 * before caching or further use.
293 */
294 skb_dst_force(skb);
295 dst = skb_dst(skb);
296 if (unlikely(!dst)) {
297 err = -ENETUNREACH;
298 goto drop;
299 }
300
301 /* cache only if we don't create a dst reference loop */
302 if (!dst->error && lwtst != dst->lwtstate) {
303 local_bh_disable();
304 dst_cache_set_ip6(&rlwt->cache, dst,
305 &ipv6_hdr(skb)->saddr);
306 local_bh_enable();
307 }
308
309 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst_dev(dst)));
310 if (unlikely(err))
311 goto drop;
312 } else {
313 skb_dst_drop(skb);
314 skb_dst_set(skb, dst);
315 }
316
317 return dst_input(skb);
318
319 drop:
320 kfree_skb(skb);
321 return err;
322 }
323
nla_put_rpl_srh(struct sk_buff * skb,int attrtype,struct rpl_iptunnel_encap * tuninfo)324 static int nla_put_rpl_srh(struct sk_buff *skb, int attrtype,
325 struct rpl_iptunnel_encap *tuninfo)
326 {
327 struct rpl_iptunnel_encap *data;
328 struct nlattr *nla;
329 int len;
330
331 len = RPL_IPTUNNEL_SRH_SIZE(tuninfo->srh);
332
333 nla = nla_reserve(skb, attrtype, len);
334 if (!nla)
335 return -EMSGSIZE;
336
337 data = nla_data(nla);
338 memcpy(data, tuninfo->srh, len);
339
340 return 0;
341 }
342
rpl_fill_encap_info(struct sk_buff * skb,struct lwtunnel_state * lwtstate)343 static int rpl_fill_encap_info(struct sk_buff *skb,
344 struct lwtunnel_state *lwtstate)
345 {
346 struct rpl_iptunnel_encap *tuninfo = rpl_encap_lwtunnel(lwtstate);
347
348 if (nla_put_rpl_srh(skb, RPL_IPTUNNEL_SRH, tuninfo))
349 return -EMSGSIZE;
350
351 return 0;
352 }
353
rpl_encap_nlsize(struct lwtunnel_state * lwtstate)354 static int rpl_encap_nlsize(struct lwtunnel_state *lwtstate)
355 {
356 struct rpl_iptunnel_encap *tuninfo = rpl_encap_lwtunnel(lwtstate);
357
358 return nla_total_size(RPL_IPTUNNEL_SRH_SIZE(tuninfo->srh));
359 }
360
rpl_encap_cmp(struct lwtunnel_state * a,struct lwtunnel_state * b)361 static int rpl_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
362 {
363 struct rpl_iptunnel_encap *a_hdr = rpl_encap_lwtunnel(a);
364 struct rpl_iptunnel_encap *b_hdr = rpl_encap_lwtunnel(b);
365 int len = RPL_IPTUNNEL_SRH_SIZE(a_hdr->srh);
366
367 if (len != RPL_IPTUNNEL_SRH_SIZE(b_hdr->srh))
368 return 1;
369
370 return memcmp(a_hdr, b_hdr, len);
371 }
372
373 static const struct lwtunnel_encap_ops rpl_ops = {
374 .build_state = rpl_build_state,
375 .destroy_state = rpl_destroy_state,
376 .output = rpl_output,
377 .input = rpl_input,
378 .fill_encap = rpl_fill_encap_info,
379 .get_encap_size = rpl_encap_nlsize,
380 .cmp_encap = rpl_encap_cmp,
381 .owner = THIS_MODULE,
382 };
383
rpl_init(void)384 int __init rpl_init(void)
385 {
386 int err;
387
388 err = lwtunnel_encap_add_ops(&rpl_ops, LWTUNNEL_ENCAP_RPL);
389 if (err)
390 goto out;
391
392 pr_info("RPL Segment Routing with IPv6\n");
393
394 return 0;
395
396 out:
397 return err;
398 }
399
rpl_exit(void)400 void rpl_exit(void)
401 {
402 lwtunnel_encap_del_ops(&rpl_ops, LWTUNNEL_ENCAP_RPL);
403 }
404