net/core/lwtunnel.c

   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
  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
  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
  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
  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
 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
 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
 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                                 encap_type = nla_get_u16(nla_entype);
 196
 197                                 if (lwtunnel_valid_encap_type(encap_type,
 198                                                               extack) != 0)
 199                                         return -EOPNOTSUPP;
 200                         }
 201                 }
 202                 rtnh = rtnh_next(rtnh, &remaining);
 203         }
 204
 205         return 0;
 206 }
 207 EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type_attr);
 208
 209 void lwtstate_free(struct lwtunnel_state *lws)
 210 {
 211         const struct lwtunnel_encap_ops *ops = lwtun_encaps[lws->type];
 212
 213         if (ops->destroy_state) {
 214                 ops->destroy_state(lws);
 215                 kfree_rcu(lws, rcu);
 216         } else {
 217                 kfree(lws);
 218         }
 219         module_put(ops->owner);
 220 }
 221 EXPORT_SYMBOL_GPL(lwtstate_free);
 222
 223 int lwtunnel_fill_encap(struct sk_buff *skb, struct lwtunnel_state *lwtstate,
 224                         int encap_attr, int encap_type_attr)
 225 {
 226         const struct lwtunnel_encap_ops *ops;
 227         struct nlattr *nest;
 228         int ret;
 229
 230         if (!lwtstate)
 231                 return 0;
 232
 233         if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
 234             lwtstate->type > LWTUNNEL_ENCAP_MAX)
 235                 return 0;
 236
 237         nest = nla_nest_start_noflag(skb, encap_attr);
 238         if (!nest)
 239                 return -EMSGSIZE;
 240
 241         ret = -EOPNOTSUPP;
 242         rcu_read_lock();
 243         ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
 244         if (likely(ops && ops->fill_encap))
 245                 ret = ops->fill_encap(skb, lwtstate);
 246         rcu_read_unlock();
 247
 248         if (ret)
 249                 goto nla_put_failure;
 250         nla_nest_end(skb, nest);
 251         ret = nla_put_u16(skb, encap_type_attr, lwtstate->type);
 252         if (ret)
 253                 goto nla_put_failure;
 254
 255         return 0;
 256
 257 nla_put_failure:
 258         nla_nest_cancel(skb, nest);
 259
 260         return (ret == -EOPNOTSUPP ? 0 : ret);
 261 }
 262 EXPORT_SYMBOL_GPL(lwtunnel_fill_encap);
 263
 264 int lwtunnel_get_encap_size(struct lwtunnel_state *lwtstate)
 265 {
 266         const struct lwtunnel_encap_ops *ops;
 267         int ret = 0;
 268
 269         if (!lwtstate)
 270                 return 0;
 271
 272         if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
 273             lwtstate->type > LWTUNNEL_ENCAP_MAX)
 274                 return 0;
 275
 276         rcu_read_lock();
 277         ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
 278         if (likely(ops && ops->get_encap_size))
 279                 ret = nla_total_size(ops->get_encap_size(lwtstate));
 280         rcu_read_unlock();
 281
 282         return ret;
 283 }
 284 EXPORT_SYMBOL_GPL(lwtunnel_get_encap_size);
 285
 286 int lwtunnel_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
 287 {
 288         const struct lwtunnel_encap_ops *ops;
 289         int ret = 0;
 290
 291         if (!a && !b)
 292                 return 0;
 293
 294         if (!a || !b)
 295                 return 1;
 296
 297         if (a->type != b->type)
 298                 return 1;
 299
 300         if (a->type == LWTUNNEL_ENCAP_NONE ||
 301             a->type > LWTUNNEL_ENCAP_MAX)
 302                 return 0;
 303
 304         rcu_read_lock();
 305         ops = rcu_dereference(lwtun_encaps[a->type]);
 306         if (likely(ops && ops->cmp_encap))
 307                 ret = ops->cmp_encap(a, b);
 308         rcu_read_unlock();
 309
 310         return ret;
 311 }
 312 EXPORT_SYMBOL_GPL(lwtunnel_cmp_encap);
 313
 314 int lwtunnel_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 315 {
 316         struct dst_entry *dst = skb_dst(skb);
 317         const struct lwtunnel_encap_ops *ops;
 318         struct lwtunnel_state *lwtstate;
 319         int ret = -EINVAL;
 320
 321         if (!dst)
 322                 goto drop;
 323         lwtstate = dst->lwtstate;
 324
 325         if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
 326             lwtstate->type > LWTUNNEL_ENCAP_MAX)
 327                 return 0;
 328
 329         ret = -EOPNOTSUPP;
 330         rcu_read_lock();
 331         ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
 332         if (likely(ops && ops->output))
 333                 ret = ops->output(net, sk, skb);
 334         rcu_read_unlock();
 335
 336         if (ret == -EOPNOTSUPP)
 337                 goto drop;
 338
 339         return ret;
 340
 341 drop:
 342         kfree_skb(skb);
 343
 344         return ret;
 345 }
 346 EXPORT_SYMBOL_GPL(lwtunnel_output);
 347
 348 int lwtunnel_xmit(struct sk_buff *skb)
 349 {
 350         struct dst_entry *dst = skb_dst(skb);
 351         const struct lwtunnel_encap_ops *ops;
 352         struct lwtunnel_state *lwtstate;
 353         int ret = -EINVAL;
 354
 355         if (!dst)
 356                 goto drop;
 357
 358         lwtstate = dst->lwtstate;
 359
 360         if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
 361             lwtstate->type > LWTUNNEL_ENCAP_MAX)
 362                 return 0;
 363
 364         ret = -EOPNOTSUPP;
 365         rcu_read_lock();
 366         ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
 367         if (likely(ops && ops->xmit))
 368                 ret = ops->xmit(skb);
 369         rcu_read_unlock();
 370
 371         if (ret == -EOPNOTSUPP)
 372                 goto drop;
 373
 374         return ret;
 375
 376 drop:
 377         kfree_skb(skb);
 378
 379         return ret;
 380 }
 381 EXPORT_SYMBOL_GPL(lwtunnel_xmit);
 382
 383 int lwtunnel_input(struct sk_buff *skb)
 384 {
 385         struct dst_entry *dst = skb_dst(skb);
 386         const struct lwtunnel_encap_ops *ops;
 387         struct lwtunnel_state *lwtstate;
 388         int ret = -EINVAL;
 389
 390         if (!dst)
 391                 goto drop;
 392         lwtstate = dst->lwtstate;
 393
 394         if (lwtstate->type == LWTUNNEL_ENCAP_NONE ||
 395             lwtstate->type > LWTUNNEL_ENCAP_MAX)
 396                 return 0;
 397
 398         ret = -EOPNOTSUPP;
 399         rcu_read_lock();
 400         ops = rcu_dereference(lwtun_encaps[lwtstate->type]);
 401         if (likely(ops && ops->input))
 402                 ret = ops->input(skb);
 403         rcu_read_unlock();
 404
 405         if (ret == -EOPNOTSUPP)
 406                 goto drop;
 407
 408         return ret;
 409
 410 drop:
 411         kfree_skb(skb);
 412
 413         return ret;
 414 }
 415 EXPORT_SYMBOL_GPL(lwtunnel_input);