2 * lwtunnel Infrastructure for light weight tunnels like mpls
4 * Authors: Roopa Prabhu, <roopa@cumulusnetworks.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
13 #include <linux/capability.h>
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/slab.h>
18 #include <linux/uaccess.h>
19 #include <linux/skbuff.h>
20 #include <linux/netdevice.h>
21 #include <linux/lwtunnel.h>
23 #include <linux/init.h>
24 #include <linux/err.h>
26 #include <net/lwtunnel.h>
27 #include <net/rtnetlink.h>
28 #include <net/ip6_fib.h>
29 #include <net/nexthop.h>
33 static const char *lwtunnel_encap_str(enum lwtunnel_encap_types encap_type)
35 /* Only lwt encaps implemented without using an interface for
36 * the encap need to return a string here.
39 case LWTUNNEL_ENCAP_MPLS:
41 case LWTUNNEL_ENCAP_ILA:
43 case LWTUNNEL_ENCAP_SEG6:
45 case LWTUNNEL_ENCAP_BPF:
47 case LWTUNNEL_ENCAP_IP6:
48 case LWTUNNEL_ENCAP_IP:
49 case LWTUNNEL_ENCAP_NONE:
50 case __LWTUNNEL_ENCAP_MAX:
51 /* should not have got here */
58 #endif /* CONFIG_MODULES */
60 struct lwtunnel_state *lwtunnel_state_alloc(int encap_len)
62 struct lwtunnel_state *lws;
64 lws = kzalloc(sizeof(*lws) + encap_len, GFP_ATOMIC);
68 EXPORT_SYMBOL(lwtunnel_state_alloc);
70 static const struct lwtunnel_encap_ops __rcu *
71 lwtun_encaps[LWTUNNEL_ENCAP_MAX + 1] __read_mostly;
73 int lwtunnel_encap_add_ops(const struct lwtunnel_encap_ops *ops,
76 if (num > LWTUNNEL_ENCAP_MAX)
79 return !cmpxchg((const struct lwtunnel_encap_ops **)
83 EXPORT_SYMBOL(lwtunnel_encap_add_ops);
85 int lwtunnel_encap_del_ops(const struct lwtunnel_encap_ops *ops,
86 unsigned int encap_type)
90 if (encap_type == LWTUNNEL_ENCAP_NONE ||
91 encap_type > LWTUNNEL_ENCAP_MAX)
94 ret = (cmpxchg((const struct lwtunnel_encap_ops **)
95 &lwtun_encaps[encap_type],
96 ops, NULL) == ops) ? 0 : -1;
102 EXPORT_SYMBOL(lwtunnel_encap_del_ops);
104 int lwtunnel_build_state(struct net_device *dev, u16 encap_type,
105 struct nlattr *encap, unsigned int family,
106 const void *cfg, struct lwtunnel_state **lws)
108 const struct lwtunnel_encap_ops *ops;
111 if (encap_type == LWTUNNEL_ENCAP_NONE ||
112 encap_type > LWTUNNEL_ENCAP_MAX)
117 ops = rcu_dereference(lwtun_encaps[encap_type]);
118 if (likely(ops && ops->build_state && try_module_get(ops->owner))) {
119 ret = ops->build_state(dev, encap, family, cfg, lws);
121 module_put(ops->owner);
127 EXPORT_SYMBOL(lwtunnel_build_state);
129 int lwtunnel_valid_encap_type(u16 encap_type)
131 const struct lwtunnel_encap_ops *ops;
134 if (encap_type == LWTUNNEL_ENCAP_NONE ||
135 encap_type > LWTUNNEL_ENCAP_MAX)
139 ops = rcu_dereference(lwtun_encaps[encap_type]);
141 #ifdef CONFIG_MODULES
143 const char *encap_type_str = lwtunnel_encap_str(encap_type);
145 if (encap_type_str) {
147 request_module("rtnl-lwt-%s", encap_type_str);
151 ops = rcu_dereference(lwtun_encaps[encap_type]);
156 return ops ? 0 : -EOPNOTSUPP;
158 EXPORT_SYMBOL(lwtunnel_valid_encap_type);
160 int lwtunnel_valid_encap_type_attr(struct nlattr *attr, int remaining)
162 struct rtnexthop *rtnh = (struct rtnexthop *)attr;
163 struct nlattr *nla_entype;
164 struct nlattr *attrs;
169 while (rtnh_ok(rtnh, remaining)) {
170 attrlen = rtnh_attrlen(rtnh);
172 attrs = rtnh_attrs(rtnh);
173 nla = nla_find(attrs, attrlen, RTA_ENCAP);
174 nla_entype = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
177 encap_type = nla_get_u16(nla_entype);
179 if (lwtunnel_valid_encap_type(encap_type) != 0)
183 rtnh = rtnh_next(rtnh, &remaining);
188 EXPORT_SYMBOL(lwtunnel_valid_encap_type_attr);
190 void lwtstate_free(struct lwtunnel_state *lws)
192 const struct lwtunnel_encap_ops *ops = lwtun_encaps[lws->type];
194 if (ops->destroy_state) {
195 ops->destroy_state(lws);
200 module_put(ops->owner);
202 EXPORT_SYMBOL(lwtstate_free);
204 int lwtunnel_fill_encap(struct sk_buff *skb, struct lwtunnel_state *lwtstate)
206 const struct lwtunnel_encap_ops *ops;
213 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
214 lwtstate->type > LWTUNNEL_ENCAP_MAX)
218 nest = nla_nest_start(skb, RTA_ENCAP);
220 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
221 if (likely(ops && ops->fill_encap))
222 ret = ops->fill_encap(skb, lwtstate);
226 goto nla_put_failure;
227 nla_nest_end(skb, nest);
228 ret = nla_put_u16(skb, RTA_ENCAP_TYPE, lwtstate->type);
230 goto nla_put_failure;
235 nla_nest_cancel(skb, nest);
237 return (ret == -EOPNOTSUPP ? 0 : ret);
239 EXPORT_SYMBOL(lwtunnel_fill_encap);
241 int lwtunnel_get_encap_size(struct lwtunnel_state *lwtstate)
243 const struct lwtunnel_encap_ops *ops;
249 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
250 lwtstate->type > LWTUNNEL_ENCAP_MAX)
254 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
255 if (likely(ops && ops->get_encap_size))
256 ret = nla_total_size(ops->get_encap_size(lwtstate));
261 EXPORT_SYMBOL(lwtunnel_get_encap_size);
263 int lwtunnel_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
265 const struct lwtunnel_encap_ops *ops;
274 if (a->type != b->type)
277 if (a->type == LWTUNNEL_ENCAP_NONE ||
278 a->type > LWTUNNEL_ENCAP_MAX)
282 ops = rcu_dereference(lwtun_encaps[a->type]);
283 if (likely(ops && ops->cmp_encap))
284 ret = ops->cmp_encap(a, b);
289 EXPORT_SYMBOL(lwtunnel_cmp_encap);
291 int lwtunnel_output(struct net *net, struct sock *sk, struct sk_buff *skb)
293 struct dst_entry *dst = skb_dst(skb);
294 const struct lwtunnel_encap_ops *ops;
295 struct lwtunnel_state *lwtstate;
300 lwtstate = dst->lwtstate;
302 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
303 lwtstate->type > LWTUNNEL_ENCAP_MAX)
308 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
309 if (likely(ops && ops->output))
310 ret = ops->output(net, sk, skb);
313 if (ret == -EOPNOTSUPP)
323 EXPORT_SYMBOL(lwtunnel_output);
325 int lwtunnel_xmit(struct sk_buff *skb)
327 struct dst_entry *dst = skb_dst(skb);
328 const struct lwtunnel_encap_ops *ops;
329 struct lwtunnel_state *lwtstate;
335 lwtstate = dst->lwtstate;
337 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
338 lwtstate->type > LWTUNNEL_ENCAP_MAX)
343 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
344 if (likely(ops && ops->xmit))
345 ret = ops->xmit(skb);
348 if (ret == -EOPNOTSUPP)
358 EXPORT_SYMBOL(lwtunnel_xmit);
360 int lwtunnel_input(struct sk_buff *skb)
362 struct dst_entry *dst = skb_dst(skb);
363 const struct lwtunnel_encap_ops *ops;
364 struct lwtunnel_state *lwtstate;
369 lwtstate = dst->lwtstate;
371 if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
372 lwtstate->type > LWTUNNEL_ENCAP_MAX)
377 ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
378 if (likely(ops && ops->input))
379 ret = ops->input(skb);
382 if (ret == -EOPNOTSUPP)
392 EXPORT_SYMBOL(lwtunnel_input);