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 #include "dev.h" 27 28 DEFINE_STATIC_KEY_FALSE(nf_hooks_lwtunnel_enabled); 29 EXPORT_SYMBOL_GPL(nf_hooks_lwtunnel_enabled); 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_SEG6_LOCAL: 48 return "SEG6LOCAL"; 49 case LWTUNNEL_ENCAP_RPL: 50 return "RPL"; 51 case LWTUNNEL_ENCAP_IOAM6: 52 return "IOAM6"; 53 case LWTUNNEL_ENCAP_XFRM: 54 /* module autoload not supported for encap type */ 55 return NULL; 56 case LWTUNNEL_ENCAP_IP6: 57 case LWTUNNEL_ENCAP_IP: 58 case LWTUNNEL_ENCAP_NONE: 59 case __LWTUNNEL_ENCAP_MAX: 60 /* should not have got here */ 61 WARN_ON(1); 62 break; 63 } 64 return NULL; 65 } 66 67 #endif /* CONFIG_MODULES */ 68 69 struct lwtunnel_state *lwtunnel_state_alloc(int encap_len) 70 { 71 struct lwtunnel_state *lws; 72 73 lws = kzalloc(sizeof(*lws) + encap_len, GFP_ATOMIC); 74 75 return lws; 76 } 77 EXPORT_SYMBOL_GPL(lwtunnel_state_alloc); 78 79 static const struct lwtunnel_encap_ops __rcu * 80 lwtun_encaps[LWTUNNEL_ENCAP_MAX + 1] __read_mostly; 81 82 int lwtunnel_encap_add_ops(const struct lwtunnel_encap_ops *ops, 83 unsigned int num) 84 { 85 if (num > LWTUNNEL_ENCAP_MAX) 86 return -ERANGE; 87 88 return !cmpxchg((const struct lwtunnel_encap_ops **) 89 &lwtun_encaps[num], 90 NULL, ops) ? 0 : -1; 91 } 92 EXPORT_SYMBOL_GPL(lwtunnel_encap_add_ops); 93 94 int lwtunnel_encap_del_ops(const struct lwtunnel_encap_ops *ops, 95 unsigned int encap_type) 96 { 97 int ret; 98 99 if (encap_type == LWTUNNEL_ENCAP_NONE || 100 encap_type > LWTUNNEL_ENCAP_MAX) 101 return -ERANGE; 102 103 ret = (cmpxchg((const struct lwtunnel_encap_ops **) 104 &lwtun_encaps[encap_type], 105 ops, NULL) == ops) ? 0 : -1; 106 107 synchronize_net(); 108 109 return ret; 110 } 111 EXPORT_SYMBOL_GPL(lwtunnel_encap_del_ops); 112 113 int lwtunnel_build_state(struct net *net, u16 encap_type, 114 struct nlattr *encap, unsigned int family, 115 const void *cfg, struct lwtunnel_state **lws, 116 struct netlink_ext_ack *extack) 117 { 118 const struct lwtunnel_encap_ops *ops; 119 bool found = false; 120 int ret = -EINVAL; 121 122 if (encap_type == LWTUNNEL_ENCAP_NONE || 123 encap_type > LWTUNNEL_ENCAP_MAX) { 124 NL_SET_ERR_MSG_ATTR(extack, encap, 125 "Unknown LWT encapsulation type"); 126 return ret; 127 } 128 129 ret = -EOPNOTSUPP; 130 rcu_read_lock(); 131 ops = rcu_dereference(lwtun_encaps[encap_type]); 132 if (likely(ops && ops->build_state && try_module_get(ops->owner))) 133 found = true; 134 rcu_read_unlock(); 135 136 if (found) { 137 ret = ops->build_state(net, encap, family, cfg, lws, extack); 138 if (ret) 139 module_put(ops->owner); 140 } else { 141 /* don't rely on -EOPNOTSUPP to detect match as build_state 142 * handlers could return it 143 */ 144 NL_SET_ERR_MSG_ATTR(extack, encap, 145 "LWT encapsulation type not supported"); 146 } 147 148 return ret; 149 } 150 EXPORT_SYMBOL_GPL(lwtunnel_build_state); 151 152 int lwtunnel_valid_encap_type(u16 encap_type, struct netlink_ext_ack *extack) 153 { 154 const struct lwtunnel_encap_ops *ops; 155 int ret = -EINVAL; 156 157 if (encap_type == LWTUNNEL_ENCAP_NONE || 158 encap_type > LWTUNNEL_ENCAP_MAX) { 159 NL_SET_ERR_MSG(extack, "Unknown lwt encapsulation type"); 160 return ret; 161 } 162 163 ops = rcu_access_pointer(lwtun_encaps[encap_type]); 164 #ifdef CONFIG_MODULES 165 if (!ops) { 166 const char *encap_type_str = lwtunnel_encap_str(encap_type); 167 168 if (encap_type_str) { 169 request_module("rtnl-lwt-%s", encap_type_str); 170 ops = rcu_access_pointer(lwtun_encaps[encap_type]); 171 } 172 } 173 #endif 174 ret = ops ? 0 : -EOPNOTSUPP; 175 if (ret < 0) 176 NL_SET_ERR_MSG(extack, "lwt encapsulation type not supported"); 177 178 return ret; 179 } 180 EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type); 181 182 int lwtunnel_valid_encap_type_attr(struct nlattr *attr, int remaining, 183 struct netlink_ext_ack *extack) 184 { 185 struct rtnexthop *rtnh = (struct rtnexthop *)attr; 186 struct nlattr *nla_entype; 187 struct nlattr *attrs; 188 u16 encap_type; 189 int attrlen; 190 191 while (rtnh_ok(rtnh, remaining)) { 192 attrlen = rtnh_attrlen(rtnh); 193 if (attrlen > 0) { 194 attrs = rtnh_attrs(rtnh); 195 nla_entype = nla_find(attrs, attrlen, RTA_ENCAP_TYPE); 196 197 if (nla_entype) { 198 if (nla_len(nla_entype) < sizeof(u16)) { 199 NL_SET_ERR_MSG(extack, "Invalid RTA_ENCAP_TYPE"); 200 return -EINVAL; 201 } 202 encap_type = nla_get_u16(nla_entype); 203 204 if (lwtunnel_valid_encap_type(encap_type, extack)) 205 return -EOPNOTSUPP; 206 } 207 } 208 rtnh = rtnh_next(rtnh, &remaining); 209 } 210 211 return 0; 212 } 213 EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type_attr); 214 215 void lwtstate_free(struct lwtunnel_state *lws) 216 { 217 const struct lwtunnel_encap_ops *ops = lwtun_encaps[lws->type]; 218 219 if (ops->destroy_state) { 220 ops->destroy_state(lws); 221 kfree_rcu(lws, rcu); 222 } else { 223 kfree(lws); 224 } 225 module_put(ops->owner); 226 } 227 EXPORT_SYMBOL_GPL(lwtstate_free); 228 229 int lwtunnel_fill_encap(struct sk_buff *skb, struct lwtunnel_state *lwtstate, 230 int encap_attr, int encap_type_attr) 231 { 232 const struct lwtunnel_encap_ops *ops; 233 struct nlattr *nest; 234 int ret; 235 236 if (!lwtstate) 237 return 0; 238 239 if (lwtstate->type == LWTUNNEL_ENCAP_NONE || 240 lwtstate->type > LWTUNNEL_ENCAP_MAX) 241 return 0; 242 243 nest = nla_nest_start_noflag(skb, encap_attr); 244 if (!nest) 245 return -EMSGSIZE; 246 247 ret = -EOPNOTSUPP; 248 rcu_read_lock(); 249 ops = rcu_dereference(lwtun_encaps[lwtstate->type]); 250 if (likely(ops && ops->fill_encap)) 251 ret = ops->fill_encap(skb, lwtstate); 252 rcu_read_unlock(); 253 254 if (ret) 255 goto nla_put_failure; 256 nla_nest_end(skb, nest); 257 ret = nla_put_u16(skb, encap_type_attr, lwtstate->type); 258 if (ret) 259 goto nla_put_failure; 260 261 return 0; 262 263 nla_put_failure: 264 nla_nest_cancel(skb, nest); 265 266 return (ret == -EOPNOTSUPP ? 0 : ret); 267 } 268 EXPORT_SYMBOL_GPL(lwtunnel_fill_encap); 269 270 int lwtunnel_get_encap_size(struct lwtunnel_state *lwtstate) 271 { 272 const struct lwtunnel_encap_ops *ops; 273 int ret = 0; 274 275 if (!lwtstate) 276 return 0; 277 278 if (lwtstate->type == LWTUNNEL_ENCAP_NONE || 279 lwtstate->type > LWTUNNEL_ENCAP_MAX) 280 return 0; 281 282 rcu_read_lock(); 283 ops = rcu_dereference(lwtun_encaps[lwtstate->type]); 284 if (likely(ops && ops->get_encap_size)) 285 ret = nla_total_size(ops->get_encap_size(lwtstate)); 286 rcu_read_unlock(); 287 288 return ret; 289 } 290 EXPORT_SYMBOL_GPL(lwtunnel_get_encap_size); 291 292 int lwtunnel_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b) 293 { 294 const struct lwtunnel_encap_ops *ops; 295 int ret = 0; 296 297 if (!a && !b) 298 return 0; 299 300 if (!a || !b) 301 return 1; 302 303 if (a->type != b->type) 304 return 1; 305 306 if (a->type == LWTUNNEL_ENCAP_NONE || 307 a->type > LWTUNNEL_ENCAP_MAX) 308 return 0; 309 310 rcu_read_lock(); 311 ops = rcu_dereference(lwtun_encaps[a->type]); 312 if (likely(ops && ops->cmp_encap)) 313 ret = ops->cmp_encap(a, b); 314 rcu_read_unlock(); 315 316 return ret; 317 } 318 EXPORT_SYMBOL_GPL(lwtunnel_cmp_encap); 319 320 int lwtunnel_output(struct net *net, struct sock *sk, struct sk_buff *skb) 321 { 322 const struct lwtunnel_encap_ops *ops; 323 struct lwtunnel_state *lwtstate; 324 struct dst_entry *dst; 325 int ret; 326 327 local_bh_disable(); 328 329 if (dev_xmit_recursion()) { 330 net_crit_ratelimited("%s(): recursion limit reached on datapath\n", 331 __func__); 332 ret = -ENETDOWN; 333 goto drop; 334 } 335 336 dst = skb_dst(skb); 337 if (!dst) { 338 ret = -EINVAL; 339 goto drop; 340 } 341 lwtstate = dst->lwtstate; 342 343 if (lwtstate->type == LWTUNNEL_ENCAP_NONE || 344 lwtstate->type > LWTUNNEL_ENCAP_MAX) { 345 ret = 0; 346 goto out; 347 } 348 349 ret = -EOPNOTSUPP; 350 rcu_read_lock(); 351 ops = rcu_dereference(lwtun_encaps[lwtstate->type]); 352 if (likely(ops && ops->output)) { 353 dev_xmit_recursion_inc(); 354 ret = ops->output(net, sk, skb); 355 dev_xmit_recursion_dec(); 356 } 357 rcu_read_unlock(); 358 359 if (ret == -EOPNOTSUPP) 360 goto drop; 361 362 goto out; 363 364 drop: 365 kfree_skb(skb); 366 367 out: 368 local_bh_enable(); 369 return ret; 370 } 371 EXPORT_SYMBOL_GPL(lwtunnel_output); 372 373 int lwtunnel_xmit(struct sk_buff *skb) 374 { 375 const struct lwtunnel_encap_ops *ops; 376 struct lwtunnel_state *lwtstate; 377 struct dst_entry *dst; 378 int ret; 379 380 local_bh_disable(); 381 382 if (dev_xmit_recursion()) { 383 net_crit_ratelimited("%s(): recursion limit reached on datapath\n", 384 __func__); 385 ret = -ENETDOWN; 386 goto drop; 387 } 388 389 dst = skb_dst(skb); 390 if (!dst) { 391 ret = -EINVAL; 392 goto drop; 393 } 394 395 lwtstate = dst->lwtstate; 396 397 if (lwtstate->type == LWTUNNEL_ENCAP_NONE || 398 lwtstate->type > LWTUNNEL_ENCAP_MAX) { 399 ret = 0; 400 goto out; 401 } 402 403 ret = -EOPNOTSUPP; 404 rcu_read_lock(); 405 ops = rcu_dereference(lwtun_encaps[lwtstate->type]); 406 if (likely(ops && ops->xmit)) { 407 dev_xmit_recursion_inc(); 408 ret = ops->xmit(skb); 409 dev_xmit_recursion_dec(); 410 } 411 rcu_read_unlock(); 412 413 if (ret == -EOPNOTSUPP) 414 goto drop; 415 416 goto out; 417 418 drop: 419 kfree_skb(skb); 420 421 out: 422 local_bh_enable(); 423 return ret; 424 } 425 EXPORT_SYMBOL_GPL(lwtunnel_xmit); 426 427 int lwtunnel_input(struct sk_buff *skb) 428 { 429 const struct lwtunnel_encap_ops *ops; 430 struct lwtunnel_state *lwtstate; 431 struct dst_entry *dst; 432 int ret; 433 434 DEBUG_NET_WARN_ON_ONCE(!in_softirq()); 435 436 if (dev_xmit_recursion()) { 437 net_crit_ratelimited("%s(): recursion limit reached on datapath\n", 438 __func__); 439 ret = -ENETDOWN; 440 goto drop; 441 } 442 443 dst = skb_dst(skb); 444 if (!dst) { 445 ret = -EINVAL; 446 goto drop; 447 } 448 lwtstate = dst->lwtstate; 449 450 if (lwtstate->type == LWTUNNEL_ENCAP_NONE || 451 lwtstate->type > LWTUNNEL_ENCAP_MAX) 452 return 0; 453 454 ret = -EOPNOTSUPP; 455 rcu_read_lock(); 456 ops = rcu_dereference(lwtun_encaps[lwtstate->type]); 457 if (likely(ops && ops->input)) { 458 dev_xmit_recursion_inc(); 459 ret = ops->input(skb); 460 dev_xmit_recursion_dec(); 461 } 462 rcu_read_unlock(); 463 464 if (ret == -EOPNOTSUPP) 465 goto drop; 466 467 return ret; 468 469 drop: 470 kfree_skb(skb); 471 472 return ret; 473 } 474 EXPORT_SYMBOL_GPL(lwtunnel_input); 475