tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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
26DEFINE_STATIC_KEY_FALSE(nf_hooks_lwtunnel_enabled);
27EXPORT_SYMBOL_GPL(nf_hooks_lwtunnel_enabled);
28
29#ifdef CONFIG_MODULES
30
31static const char *lwtunnel_encap_str(enum lwtunnel_encap_types encap_type)
32{
33 /* Only lwt encaps implemented without using an interface for
34 * the encap need to return a string here.
35 */
36 switch (encap_type) {
37 case LWTUNNEL_ENCAP_MPLS:
38 return "MPLS";
39 case LWTUNNEL_ENCAP_ILA:
40 return "ILA";
41 case LWTUNNEL_ENCAP_SEG6:
42 return "SEG6";
43 case LWTUNNEL_ENCAP_BPF:
44 return "BPF";
45 case LWTUNNEL_ENCAP_SEG6_LOCAL:
46 return "SEG6LOCAL";
47 case LWTUNNEL_ENCAP_RPL:
48 return "RPL";
49 case LWTUNNEL_ENCAP_IOAM6:
50 return "IOAM6";
51 case LWTUNNEL_ENCAP_IP6:
52 case LWTUNNEL_ENCAP_IP:
53 case LWTUNNEL_ENCAP_XFRM:
54 case LWTUNNEL_ENCAP_NONE:
55 case __LWTUNNEL_ENCAP_MAX:
56 /* should not have got here */
57 WARN_ON(1);
58 break;
59 }
60 return NULL;
61}
62
63#endif /* CONFIG_MODULES */
64
65struct lwtunnel_state *lwtunnel_state_alloc(int encap_len)
66{
67 struct lwtunnel_state *lws;
68
69 lws = kzalloc(sizeof(*lws) + encap_len, GFP_ATOMIC);
70
71 return lws;
72}
73EXPORT_SYMBOL_GPL(lwtunnel_state_alloc);
74
75static const struct lwtunnel_encap_ops __rcu *
76 lwtun_encaps[LWTUNNEL_ENCAP_MAX + 1] __read_mostly;
77
78int lwtunnel_encap_add_ops(const struct lwtunnel_encap_ops *ops,
79 unsigned int num)
80{
81 if (num > LWTUNNEL_ENCAP_MAX)
82 return -ERANGE;
83
84 return !cmpxchg((const struct lwtunnel_encap_ops **)
85 &lwtun_encaps[num],
86 NULL, ops) ? 0 : -1;
87}
88EXPORT_SYMBOL_GPL(lwtunnel_encap_add_ops);
89
90int lwtunnel_encap_del_ops(const struct lwtunnel_encap_ops *ops,
91 unsigned int encap_type)
92{
93 int ret;
94
95 if (encap_type == LWTUNNEL_ENCAP_NONE ||
96 encap_type > LWTUNNEL_ENCAP_MAX)
97 return -ERANGE;
98
99 ret = (cmpxchg((const struct lwtunnel_encap_ops **)
100 &lwtun_encaps[encap_type],
101 ops, NULL) == ops) ? 0 : -1;
102
103 synchronize_net();
104
105 return ret;
106}
107EXPORT_SYMBOL_GPL(lwtunnel_encap_del_ops);
108
109int lwtunnel_build_state(struct net *net, u16 encap_type,
110 struct nlattr *encap, unsigned int family,
111 const void *cfg, struct lwtunnel_state **lws,
112 struct netlink_ext_ack *extack)
113{
114 const struct lwtunnel_encap_ops *ops;
115 bool found = false;
116 int ret = -EINVAL;
117
118 if (encap_type == LWTUNNEL_ENCAP_NONE ||
119 encap_type > LWTUNNEL_ENCAP_MAX) {
120 NL_SET_ERR_MSG_ATTR(extack, encap,
121 "Unknown LWT encapsulation type");
122 return ret;
123 }
124
125 ret = -EOPNOTSUPP;
126 rcu_read_lock();
127 ops = rcu_dereference(lwtun_encaps[encap_type]);
128 if (likely(ops && ops->build_state && try_module_get(ops->owner)))
129 found = true;
130 rcu_read_unlock();
131
132 if (found) {
133 ret = ops->build_state(net, encap, family, cfg, lws, extack);
134 if (ret)
135 module_put(ops->owner);
136 } else {
137 /* don't rely on -EOPNOTSUPP to detect match as build_state
138 * handlers could return it
139 */
140 NL_SET_ERR_MSG_ATTR(extack, encap,
141 "LWT encapsulation type not supported");
142 }
143
144 return ret;
145}
146EXPORT_SYMBOL_GPL(lwtunnel_build_state);
147
148int lwtunnel_valid_encap_type(u16 encap_type, struct netlink_ext_ack *extack)
149{
150 const struct lwtunnel_encap_ops *ops;
151 int ret = -EINVAL;
152
153 if (encap_type == LWTUNNEL_ENCAP_NONE ||
154 encap_type > LWTUNNEL_ENCAP_MAX) {
155 NL_SET_ERR_MSG(extack, "Unknown lwt encapsulation type");
156 return ret;
157 }
158
159 rcu_read_lock();
160 ops = rcu_dereference(lwtun_encaps[encap_type]);
161 rcu_read_unlock();
162#ifdef CONFIG_MODULES
163 if (!ops) {
164 const char *encap_type_str = lwtunnel_encap_str(encap_type);
165
166 if (encap_type_str) {
167 __rtnl_unlock();
168 request_module("rtnl-lwt-%s", encap_type_str);
169 rtnl_lock();
170
171 rcu_read_lock();
172 ops = rcu_dereference(lwtun_encaps[encap_type]);
173 rcu_read_unlock();
174 }
175 }
176#endif
177 ret = ops ? 0 : -EOPNOTSUPP;
178 if (ret < 0)
179 NL_SET_ERR_MSG(extack, "lwt encapsulation type not supported");
180
181 return ret;
182}
183EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type);
184
185int lwtunnel_valid_encap_type_attr(struct nlattr *attr, int remaining,
186 struct netlink_ext_ack *extack)
187{
188 struct rtnexthop *rtnh = (struct rtnexthop *)attr;
189 struct nlattr *nla_entype;
190 struct nlattr *attrs;
191 u16 encap_type;
192 int attrlen;
193
194 while (rtnh_ok(rtnh, remaining)) {
195 attrlen = rtnh_attrlen(rtnh);
196 if (attrlen > 0) {
197 attrs = rtnh_attrs(rtnh);
198 nla_entype = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
199
200 if (nla_entype) {
201 if (nla_len(nla_entype) < sizeof(u16)) {
202 NL_SET_ERR_MSG(extack, "Invalid RTA_ENCAP_TYPE");
203 return -EINVAL;
204 }
205 encap_type = nla_get_u16(nla_entype);
206
207 if (lwtunnel_valid_encap_type(encap_type,
208 extack) != 0)
209 return -EOPNOTSUPP;
210 }
211 }
212 rtnh = rtnh_next(rtnh, &remaining);
213 }
214
215 return 0;
216}
217EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type_attr);
218
219void lwtstate_free(struct lwtunnel_state *lws)
220{
221 const struct lwtunnel_encap_ops *ops = lwtun_encaps[lws->type];
222
223 if (ops->destroy_state) {
224 ops->destroy_state(lws);
225 kfree_rcu(lws, rcu);
226 } else {
227 kfree(lws);
228 }
229 module_put(ops->owner);
230}
231EXPORT_SYMBOL_GPL(lwtstate_free);
232
233int lwtunnel_fill_encap(struct sk_buff *skb, struct lwtunnel_state *lwtstate,
234 int encap_attr, int encap_type_attr)
235{
236 const struct lwtunnel_encap_ops *ops;
237 struct nlattr *nest;
238 int ret;
239
240 if (!lwtstate)
241 return 0;
242
243 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
244 lwtstate->type > LWTUNNEL_ENCAP_MAX)
245 return 0;
246
247 nest = nla_nest_start_noflag(skb, encap_attr);
248 if (!nest)
249 return -EMSGSIZE;
250
251 ret = -EOPNOTSUPP;
252 rcu_read_lock();
253 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
254 if (likely(ops && ops->fill_encap))
255 ret = ops->fill_encap(skb, lwtstate);
256 rcu_read_unlock();
257
258 if (ret)
259 goto nla_put_failure;
260 nla_nest_end(skb, nest);
261 ret = nla_put_u16(skb, encap_type_attr, lwtstate->type);
262 if (ret)
263 goto nla_put_failure;
264
265 return 0;
266
267nla_put_failure:
268 nla_nest_cancel(skb, nest);
269
270 return (ret == -EOPNOTSUPP ? 0 : ret);
271}
272EXPORT_SYMBOL_GPL(lwtunnel_fill_encap);
273
274int lwtunnel_get_encap_size(struct lwtunnel_state *lwtstate)
275{
276 const struct lwtunnel_encap_ops *ops;
277 int ret = 0;
278
279 if (!lwtstate)
280 return 0;
281
282 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
283 lwtstate->type > LWTUNNEL_ENCAP_MAX)
284 return 0;
285
286 rcu_read_lock();
287 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
288 if (likely(ops && ops->get_encap_size))
289 ret = nla_total_size(ops->get_encap_size(lwtstate));
290 rcu_read_unlock();
291
292 return ret;
293}
294EXPORT_SYMBOL_GPL(lwtunnel_get_encap_size);
295
296int lwtunnel_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
297{
298 const struct lwtunnel_encap_ops *ops;
299 int ret = 0;
300
301 if (!a && !b)
302 return 0;
303
304 if (!a || !b)
305 return 1;
306
307 if (a->type != b->type)
308 return 1;
309
310 if (a->type == LWTUNNEL_ENCAP_NONE ||
311 a->type > LWTUNNEL_ENCAP_MAX)
312 return 0;
313
314 rcu_read_lock();
315 ops = rcu_dereference(lwtun_encaps[a->type]);
316 if (likely(ops && ops->cmp_encap))
317 ret = ops->cmp_encap(a, b);
318 rcu_read_unlock();
319
320 return ret;
321}
322EXPORT_SYMBOL_GPL(lwtunnel_cmp_encap);
323
324int lwtunnel_output(struct net *net, struct sock *sk, struct sk_buff *skb)
325{
326 struct dst_entry *dst = skb_dst(skb);
327 const struct lwtunnel_encap_ops *ops;
328 struct lwtunnel_state *lwtstate;
329 int ret = -EINVAL;
330
331 if (!dst)
332 goto drop;
333 lwtstate = dst->lwtstate;
334
335 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
336 lwtstate->type > LWTUNNEL_ENCAP_MAX)
337 return 0;
338
339 ret = -EOPNOTSUPP;
340 rcu_read_lock();
341 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
342 if (likely(ops && ops->output))
343 ret = ops->output(net, sk, skb);
344 rcu_read_unlock();
345
346 if (ret == -EOPNOTSUPP)
347 goto drop;
348
349 return ret;
350
351drop:
352 kfree_skb(skb);
353
354 return ret;
355}
356EXPORT_SYMBOL_GPL(lwtunnel_output);
357
358int lwtunnel_xmit(struct sk_buff *skb)
359{
360 struct dst_entry *dst = skb_dst(skb);
361 const struct lwtunnel_encap_ops *ops;
362 struct lwtunnel_state *lwtstate;
363 int ret = -EINVAL;
364
365 if (!dst)
366 goto drop;
367
368 lwtstate = dst->lwtstate;
369
370 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
371 lwtstate->type > LWTUNNEL_ENCAP_MAX)
372 return 0;
373
374 ret = -EOPNOTSUPP;
375 rcu_read_lock();
376 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
377 if (likely(ops && ops->xmit))
378 ret = ops->xmit(skb);
379 rcu_read_unlock();
380
381 if (ret == -EOPNOTSUPP)
382 goto drop;
383
384 return ret;
385
386drop:
387 kfree_skb(skb);
388
389 return ret;
390}
391EXPORT_SYMBOL_GPL(lwtunnel_xmit);
392
393int lwtunnel_input(struct sk_buff *skb)
394{
395 struct dst_entry *dst = skb_dst(skb);
396 const struct lwtunnel_encap_ops *ops;
397 struct lwtunnel_state *lwtstate;
398 int ret = -EINVAL;
399
400 if (!dst)
401 goto drop;
402 lwtstate = dst->lwtstate;
403
404 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
405 lwtstate->type > LWTUNNEL_ENCAP_MAX)
406 return 0;
407
408 ret = -EOPNOTSUPP;
409 rcu_read_lock();
410 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
411 if (likely(ops && ops->input))
412 ret = ops->input(skb);
413 rcu_read_unlock();
414
415 if (ret == -EOPNOTSUPP)
416 goto drop;
417
418 return ret;
419
420drop:
421 kfree_skb(skb);
422
423 return ret;
424}
425EXPORT_SYMBOL_GPL(lwtunnel_input);