net/core/lwtunnel.c

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