1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * lwtunnel Infrastructure for light weight tunnels like mpls
4 *
5 * Authors: Roopa Prabhu, <roopa@cumulusnetworks.com>
6 */
7
8 #include <linux/capability.h>
9 #include <linux/module.h>
10 #include <linux/types.h>
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/uaccess.h>
14 #include <linux/skbuff.h>
15 #include <linux/netdevice.h>
16 #include <linux/lwtunnel.h>
17 #include <linux/in.h>
18 #include <linux/init.h>
19 #include <linux/err.h>
20
21 #include <net/lwtunnel.h>
22 #include <net/rtnetlink.h>
23 #include <net/ip6_fib.h>
24 #include <net/rtnh.h>
25
26 #ifdef CONFIG_MODULES
27
lwtunnel_encap_str(enum lwtunnel_encap_types encap_type)28 static const char *lwtunnel_encap_str(enum lwtunnel_encap_types encap_type)
29 {
30 /* Only lwt encaps implemented without using an interface for
31 * the encap need to return a string here.
32 */
33 switch (encap_type) {
34 case LWTUNNEL_ENCAP_MPLS:
35 return "MPLS";
36 case LWTUNNEL_ENCAP_ILA:
37 return "ILA";
38 case LWTUNNEL_ENCAP_SEG6:
39 return "SEG6";
40 case LWTUNNEL_ENCAP_BPF:
41 return "BPF";
42 case LWTUNNEL_ENCAP_SEG6_LOCAL:
43 return "SEG6LOCAL";
44 case LWTUNNEL_ENCAP_RPL:
45 return "RPL";
46 case LWTUNNEL_ENCAP_IP6:
47 case LWTUNNEL_ENCAP_IP:
48 case LWTUNNEL_ENCAP_NONE:
49 case __LWTUNNEL_ENCAP_MAX:
50 /* should not have got here */
51 WARN_ON(1);
52 break;
53 }
54 return NULL;
55 }
56
57 #endif /* CONFIG_MODULES */
58
lwtunnel_state_alloc(int encap_len)59 struct lwtunnel_state *lwtunnel_state_alloc(int encap_len)
60 {
61 struct lwtunnel_state *lws;
62
63 lws = kzalloc(sizeof(*lws) + encap_len, GFP_ATOMIC);
64
65 return lws;
66 }
67 EXPORT_SYMBOL_GPL(lwtunnel_state_alloc);
68
69 static const struct lwtunnel_encap_ops __rcu *
70 lwtun_encaps[LWTUNNEL_ENCAP_MAX + 1] __read_mostly;
71
lwtunnel_encap_add_ops(const struct lwtunnel_encap_ops * ops,unsigned int num)72 int lwtunnel_encap_add_ops(const struct lwtunnel_encap_ops *ops,
73 unsigned int num)
74 {
75 if (num > LWTUNNEL_ENCAP_MAX)
76 return -ERANGE;
77
78 return !cmpxchg((const struct lwtunnel_encap_ops **)
79 &lwtun_encaps[num],
80 NULL, ops) ? 0 : -1;
81 }
82 EXPORT_SYMBOL_GPL(lwtunnel_encap_add_ops);
83
lwtunnel_encap_del_ops(const struct lwtunnel_encap_ops * ops,unsigned int encap_type)84 int lwtunnel_encap_del_ops(const struct lwtunnel_encap_ops *ops,
85 unsigned int encap_type)
86 {
87 int ret;
88
89 if (encap_type == LWTUNNEL_ENCAP_NONE ||
90 encap_type > LWTUNNEL_ENCAP_MAX)
91 return -ERANGE;
92
93 ret = (cmpxchg((const struct lwtunnel_encap_ops **)
94 &lwtun_encaps[encap_type],
95 ops, NULL) == ops) ? 0 : -1;
96
97 synchronize_net();
98
99 return ret;
100 }
101 EXPORT_SYMBOL_GPL(lwtunnel_encap_del_ops);
102
lwtunnel_build_state(struct net * net,u16 encap_type,struct nlattr * encap,unsigned int family,const void * cfg,struct lwtunnel_state ** lws,struct netlink_ext_ack * extack)103 int lwtunnel_build_state(struct net *net, u16 encap_type,
104 struct nlattr *encap, unsigned int family,
105 const void *cfg, struct lwtunnel_state **lws,
106 struct netlink_ext_ack *extack)
107 {
108 const struct lwtunnel_encap_ops *ops;
109 bool found = false;
110 int ret = -EINVAL;
111
112 if (encap_type == LWTUNNEL_ENCAP_NONE ||
113 encap_type > LWTUNNEL_ENCAP_MAX) {
114 NL_SET_ERR_MSG_ATTR(extack, encap,
115 "Unknown LWT encapsulation type");
116 return ret;
117 }
118
119 ret = -EOPNOTSUPP;
120 rcu_read_lock();
121 ops = rcu_dereference(lwtun_encaps[encap_type]);
122 if (likely(ops && ops->build_state && try_module_get(ops->owner)))
123 found = true;
124 rcu_read_unlock();
125
126 if (found) {
127 ret = ops->build_state(net, encap, family, cfg, lws, extack);
128 if (ret)
129 module_put(ops->owner);
130 } else {
131 /* don't rely on -EOPNOTSUPP to detect match as build_state
132 * handlers could return it
133 */
134 NL_SET_ERR_MSG_ATTR(extack, encap,
135 "LWT encapsulation type not supported");
136 }
137
138 return ret;
139 }
140 EXPORT_SYMBOL_GPL(lwtunnel_build_state);
141
lwtunnel_valid_encap_type(u16 encap_type,struct netlink_ext_ack * extack)142 int lwtunnel_valid_encap_type(u16 encap_type, struct netlink_ext_ack *extack)
143 {
144 const struct lwtunnel_encap_ops *ops;
145 int ret = -EINVAL;
146
147 if (encap_type == LWTUNNEL_ENCAP_NONE ||
148 encap_type > LWTUNNEL_ENCAP_MAX) {
149 NL_SET_ERR_MSG(extack, "Unknown lwt encapsulation type");
150 return ret;
151 }
152
153 rcu_read_lock();
154 ops = rcu_dereference(lwtun_encaps[encap_type]);
155 rcu_read_unlock();
156 #ifdef CONFIG_MODULES
157 if (!ops) {
158 const char *encap_type_str = lwtunnel_encap_str(encap_type);
159
160 if (encap_type_str) {
161 __rtnl_unlock();
162 request_module("rtnl-lwt-%s", encap_type_str);
163 rtnl_lock();
164
165 rcu_read_lock();
166 ops = rcu_dereference(lwtun_encaps[encap_type]);
167 rcu_read_unlock();
168 }
169 }
170 #endif
171 ret = ops ? 0 : -EOPNOTSUPP;
172 if (ret < 0)
173 NL_SET_ERR_MSG(extack, "lwt encapsulation type not supported");
174
175 return ret;
176 }
177 EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type);
178
lwtunnel_valid_encap_type_attr(struct nlattr * attr,int remaining,struct netlink_ext_ack * extack)179 int lwtunnel_valid_encap_type_attr(struct nlattr *attr, int remaining,
180 struct netlink_ext_ack *extack)
181 {
182 struct rtnexthop *rtnh = (struct rtnexthop *)attr;
183 struct nlattr *nla_entype;
184 struct nlattr *attrs;
185 u16 encap_type;
186 int attrlen;
187
188 while (rtnh_ok(rtnh, remaining)) {
189 attrlen = rtnh_attrlen(rtnh);
190 if (attrlen > 0) {
191 attrs = rtnh_attrs(rtnh);
192 nla_entype = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
193
194 if (nla_entype) {
195 if (nla_len(nla_entype) < sizeof(u16)) {
196 NL_SET_ERR_MSG(extack, "Invalid RTA_ENCAP_TYPE");
197 return -EINVAL;
198 }
199 encap_type = nla_get_u16(nla_entype);
200
201 if (lwtunnel_valid_encap_type(encap_type,
202 extack) != 0)
203 return -EOPNOTSUPP;
204 }
205 }
206 rtnh = rtnh_next(rtnh, &remaining);
207 }
208
209 return 0;
210 }
211 EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type_attr);
212
lwtstate_free(struct lwtunnel_state * lws)213 void lwtstate_free(struct lwtunnel_state *lws)
214 {
215 const struct lwtunnel_encap_ops *ops = lwtun_encaps[lws->type];
216
217 if (ops->destroy_state) {
218 ops->destroy_state(lws);
219 kfree_rcu(lws, rcu);
220 } else {
221 kfree(lws);
222 }
223 module_put(ops->owner);
224 }
225 EXPORT_SYMBOL_GPL(lwtstate_free);
226
lwtunnel_fill_encap(struct sk_buff * skb,struct lwtunnel_state * lwtstate,int encap_attr,int encap_type_attr)227 int lwtunnel_fill_encap(struct sk_buff *skb, struct lwtunnel_state *lwtstate,
228 int encap_attr, int encap_type_attr)
229 {
230 const struct lwtunnel_encap_ops *ops;
231 struct nlattr *nest;
232 int ret;
233
234 if (!lwtstate)
235 return 0;
236
237 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
238 lwtstate->type > LWTUNNEL_ENCAP_MAX)
239 return 0;
240
241 nest = nla_nest_start_noflag(skb, encap_attr);
242 if (!nest)
243 return -EMSGSIZE;
244
245 ret = -EOPNOTSUPP;
246 rcu_read_lock();
247 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
248 if (likely(ops && ops->fill_encap))
249 ret = ops->fill_encap(skb, lwtstate);
250 rcu_read_unlock();
251
252 if (ret)
253 goto nla_put_failure;
254 nla_nest_end(skb, nest);
255 ret = nla_put_u16(skb, encap_type_attr, lwtstate->type);
256 if (ret)
257 goto nla_put_failure;
258
259 return 0;
260
261 nla_put_failure:
262 nla_nest_cancel(skb, nest);
263
264 return (ret == -EOPNOTSUPP ? 0 : ret);
265 }
266 EXPORT_SYMBOL_GPL(lwtunnel_fill_encap);
267
lwtunnel_get_encap_size(struct lwtunnel_state * lwtstate)268 int lwtunnel_get_encap_size(struct lwtunnel_state *lwtstate)
269 {
270 const struct lwtunnel_encap_ops *ops;
271 int ret = 0;
272
273 if (!lwtstate)
274 return 0;
275
276 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
277 lwtstate->type > LWTUNNEL_ENCAP_MAX)
278 return 0;
279
280 rcu_read_lock();
281 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
282 if (likely(ops && ops->get_encap_size))
283 ret = nla_total_size(ops->get_encap_size(lwtstate));
284 rcu_read_unlock();
285
286 return ret;
287 }
288 EXPORT_SYMBOL_GPL(lwtunnel_get_encap_size);
289
lwtunnel_cmp_encap(struct lwtunnel_state * a,struct lwtunnel_state * b)290 int lwtunnel_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
291 {
292 const struct lwtunnel_encap_ops *ops;
293 int ret = 0;
294
295 if (!a && !b)
296 return 0;
297
298 if (!a || !b)
299 return 1;
300
301 if (a->type != b->type)
302 return 1;
303
304 if (a->type == LWTUNNEL_ENCAP_NONE ||
305 a->type > LWTUNNEL_ENCAP_MAX)
306 return 0;
307
308 rcu_read_lock();
309 ops = rcu_dereference(lwtun_encaps[a->type]);
310 if (likely(ops && ops->cmp_encap))
311 ret = ops->cmp_encap(a, b);
312 rcu_read_unlock();
313
314 return ret;
315 }
316 EXPORT_SYMBOL_GPL(lwtunnel_cmp_encap);
317
lwtunnel_output(struct net * net,struct sock * sk,struct sk_buff * skb)318 int lwtunnel_output(struct net *net, struct sock *sk, struct sk_buff *skb)
319 {
320 struct dst_entry *dst = skb_dst(skb);
321 const struct lwtunnel_encap_ops *ops;
322 struct lwtunnel_state *lwtstate;
323 int ret = -EINVAL;
324
325 if (!dst)
326 goto drop;
327 lwtstate = dst->lwtstate;
328
329 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
330 lwtstate->type > LWTUNNEL_ENCAP_MAX)
331 return 0;
332
333 ret = -EOPNOTSUPP;
334 rcu_read_lock();
335 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
336 if (likely(ops && ops->output))
337 ret = ops->output(net, sk, skb);
338 rcu_read_unlock();
339
340 if (ret == -EOPNOTSUPP)
341 goto drop;
342
343 return ret;
344
345 drop:
346 kfree_skb(skb);
347
348 return ret;
349 }
350 EXPORT_SYMBOL_GPL(lwtunnel_output);
351
lwtunnel_xmit(struct sk_buff * skb)352 int lwtunnel_xmit(struct sk_buff *skb)
353 {
354 struct dst_entry *dst = skb_dst(skb);
355 const struct lwtunnel_encap_ops *ops;
356 struct lwtunnel_state *lwtstate;
357 int ret = -EINVAL;
358
359 if (!dst)
360 goto drop;
361
362 lwtstate = dst->lwtstate;
363
364 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
365 lwtstate->type > LWTUNNEL_ENCAP_MAX)
366 return 0;
367
368 ret = -EOPNOTSUPP;
369 rcu_read_lock();
370 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
371 if (likely(ops && ops->xmit))
372 ret = ops->xmit(skb);
373 rcu_read_unlock();
374
375 if (ret == -EOPNOTSUPP)
376 goto drop;
377
378 return ret;
379
380 drop:
381 kfree_skb(skb);
382
383 return ret;
384 }
385 EXPORT_SYMBOL_GPL(lwtunnel_xmit);
386
lwtunnel_input(struct sk_buff * skb)387 int lwtunnel_input(struct sk_buff *skb)
388 {
389 struct dst_entry *dst = skb_dst(skb);
390 const struct lwtunnel_encap_ops *ops;
391 struct lwtunnel_state *lwtstate;
392 int ret = -EINVAL;
393
394 if (!dst)
395 goto drop;
396 lwtstate = dst->lwtstate;
397
398 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
399 lwtstate->type > LWTUNNEL_ENCAP_MAX)
400 return 0;
401
402 ret = -EOPNOTSUPP;
403 rcu_read_lock();
404 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
405 if (likely(ops && ops->input))
406 ret = ops->input(skb);
407 rcu_read_unlock();
408
409 if (ret == -EOPNOTSUPP)
410 goto drop;
411
412 return ret;
413
414 drop:
415 kfree_skb(skb);
416
417 return ret;
418 }
419 EXPORT_SYMBOL_GPL(lwtunnel_input);
420