Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 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> |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 24 | #include <net/rtnh.h> |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 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"; |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 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"; |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 44 | case LWTUNNEL_ENCAP_IP6: |
| 45 | case LWTUNNEL_ENCAP_IP: |
| 46 | case LWTUNNEL_ENCAP_NONE: |
| 47 | case __LWTUNNEL_ENCAP_MAX: |
| 48 | /* should not have got here */ |
| 49 | WARN_ON(1); |
| 50 | break; |
| 51 | } |
| 52 | return NULL; |
| 53 | } |
| 54 | |
| 55 | #endif /* CONFIG_MODULES */ |
| 56 | |
| 57 | struct lwtunnel_state *lwtunnel_state_alloc(int encap_len) |
| 58 | { |
| 59 | struct lwtunnel_state *lws; |
| 60 | |
| 61 | lws = kzalloc(sizeof(*lws) + encap_len, GFP_ATOMIC); |
| 62 | |
| 63 | return lws; |
| 64 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 65 | EXPORT_SYMBOL_GPL(lwtunnel_state_alloc); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 66 | |
| 67 | static const struct lwtunnel_encap_ops __rcu * |
| 68 | lwtun_encaps[LWTUNNEL_ENCAP_MAX + 1] __read_mostly; |
| 69 | |
| 70 | int lwtunnel_encap_add_ops(const struct lwtunnel_encap_ops *ops, |
| 71 | unsigned int num) |
| 72 | { |
| 73 | if (num > LWTUNNEL_ENCAP_MAX) |
| 74 | return -ERANGE; |
| 75 | |
| 76 | return !cmpxchg((const struct lwtunnel_encap_ops **) |
| 77 | &lwtun_encaps[num], |
| 78 | NULL, ops) ? 0 : -1; |
| 79 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 80 | EXPORT_SYMBOL_GPL(lwtunnel_encap_add_ops); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 81 | |
| 82 | int lwtunnel_encap_del_ops(const struct lwtunnel_encap_ops *ops, |
| 83 | unsigned int encap_type) |
| 84 | { |
| 85 | int ret; |
| 86 | |
| 87 | if (encap_type == LWTUNNEL_ENCAP_NONE || |
| 88 | encap_type > LWTUNNEL_ENCAP_MAX) |
| 89 | return -ERANGE; |
| 90 | |
| 91 | ret = (cmpxchg((const struct lwtunnel_encap_ops **) |
| 92 | &lwtun_encaps[encap_type], |
| 93 | ops, NULL) == ops) ? 0 : -1; |
| 94 | |
| 95 | synchronize_net(); |
| 96 | |
| 97 | return ret; |
| 98 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 99 | EXPORT_SYMBOL_GPL(lwtunnel_encap_del_ops); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 100 | |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 101 | int lwtunnel_build_state(u16 encap_type, |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 102 | struct nlattr *encap, unsigned int family, |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 103 | const void *cfg, struct lwtunnel_state **lws, |
| 104 | struct netlink_ext_ack *extack) |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 105 | { |
| 106 | const struct lwtunnel_encap_ops *ops; |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 107 | bool found = false; |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 108 | int ret = -EINVAL; |
| 109 | |
| 110 | if (encap_type == LWTUNNEL_ENCAP_NONE || |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 111 | encap_type > LWTUNNEL_ENCAP_MAX) { |
| 112 | NL_SET_ERR_MSG_ATTR(extack, encap, |
| 113 | "Unknown LWT encapsulation type"); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 114 | return ret; |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 115 | } |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 116 | |
| 117 | ret = -EOPNOTSUPP; |
| 118 | rcu_read_lock(); |
| 119 | ops = rcu_dereference(lwtun_encaps[encap_type]); |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 120 | if (likely(ops && ops->build_state && try_module_get(ops->owner))) |
| 121 | found = true; |
| 122 | rcu_read_unlock(); |
| 123 | |
| 124 | if (found) { |
| 125 | ret = ops->build_state(encap, family, cfg, lws, extack); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 126 | if (ret) |
| 127 | module_put(ops->owner); |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 128 | } else { |
| 129 | /* don't rely on -EOPNOTSUPP to detect match as build_state |
| 130 | * handlers could return it |
| 131 | */ |
| 132 | NL_SET_ERR_MSG_ATTR(extack, encap, |
| 133 | "LWT encapsulation type not supported"); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 134 | } |
| 135 | |
| 136 | return ret; |
| 137 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 138 | EXPORT_SYMBOL_GPL(lwtunnel_build_state); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 139 | |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 140 | int lwtunnel_valid_encap_type(u16 encap_type, struct netlink_ext_ack *extack) |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 141 | { |
| 142 | const struct lwtunnel_encap_ops *ops; |
| 143 | int ret = -EINVAL; |
| 144 | |
| 145 | if (encap_type == LWTUNNEL_ENCAP_NONE || |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 146 | encap_type > LWTUNNEL_ENCAP_MAX) { |
| 147 | NL_SET_ERR_MSG(extack, "Unknown lwt encapsulation type"); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 148 | return ret; |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 149 | } |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 150 | |
| 151 | rcu_read_lock(); |
| 152 | ops = rcu_dereference(lwtun_encaps[encap_type]); |
| 153 | rcu_read_unlock(); |
| 154 | #ifdef CONFIG_MODULES |
| 155 | if (!ops) { |
| 156 | const char *encap_type_str = lwtunnel_encap_str(encap_type); |
| 157 | |
| 158 | if (encap_type_str) { |
| 159 | __rtnl_unlock(); |
| 160 | request_module("rtnl-lwt-%s", encap_type_str); |
| 161 | rtnl_lock(); |
| 162 | |
| 163 | rcu_read_lock(); |
| 164 | ops = rcu_dereference(lwtun_encaps[encap_type]); |
| 165 | rcu_read_unlock(); |
| 166 | } |
| 167 | } |
| 168 | #endif |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 169 | ret = ops ? 0 : -EOPNOTSUPP; |
| 170 | if (ret < 0) |
| 171 | NL_SET_ERR_MSG(extack, "lwt encapsulation type not supported"); |
Googler | 0109c45 | 2022-10-13 17:50:39 +0800 | [diff] [blame] | 172 | |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 173 | return ret; |
| 174 | } |
| 175 | EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type); |
| 176 | |
| 177 | int lwtunnel_valid_encap_type_attr(struct nlattr *attr, int remaining, |
| 178 | struct netlink_ext_ack *extack) |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 179 | { |
| 180 | struct rtnexthop *rtnh = (struct rtnexthop *)attr; |
| 181 | struct nlattr *nla_entype; |
| 182 | struct nlattr *attrs; |
| 183 | u16 encap_type; |
| 184 | int attrlen; |
| 185 | |
| 186 | while (rtnh_ok(rtnh, remaining)) { |
| 187 | attrlen = rtnh_attrlen(rtnh); |
| 188 | if (attrlen > 0) { |
| 189 | attrs = rtnh_attrs(rtnh); |
| 190 | nla_entype = nla_find(attrs, attrlen, RTA_ENCAP_TYPE); |
| 191 | |
| 192 | if (nla_entype) { |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 193 | encap_type = nla_get_u16(nla_entype); |
| 194 | |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 195 | if (lwtunnel_valid_encap_type(encap_type, |
| 196 | extack) != 0) |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 197 | return -EOPNOTSUPP; |
| 198 | } |
| 199 | } |
| 200 | rtnh = rtnh_next(rtnh, &remaining); |
| 201 | } |
| 202 | |
| 203 | return 0; |
| 204 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 205 | EXPORT_SYMBOL_GPL(lwtunnel_valid_encap_type_attr); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 206 | |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 207 | void lwtstate_free(struct lwtunnel_state *lws) |
| 208 | { |
| 209 | const struct lwtunnel_encap_ops *ops = lwtun_encaps[lws->type]; |
| 210 | |
| 211 | if (ops->destroy_state) { |
| 212 | ops->destroy_state(lws); |
| 213 | kfree_rcu(lws, rcu); |
| 214 | } else { |
| 215 | kfree(lws); |
| 216 | } |
| 217 | module_put(ops->owner); |
| 218 | } |
| 219 | EXPORT_SYMBOL_GPL(lwtstate_free); |
| 220 | |
| 221 | int lwtunnel_fill_encap(struct sk_buff *skb, struct lwtunnel_state *lwtstate, |
| 222 | int encap_attr, int encap_type_attr) |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 223 | { |
| 224 | const struct lwtunnel_encap_ops *ops; |
| 225 | struct nlattr *nest; |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 226 | int ret; |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 227 | |
| 228 | if (!lwtstate) |
| 229 | return 0; |
| 230 | |
| 231 | if (lwtstate->type == LWTUNNEL_ENCAP_NONE || |
| 232 | lwtstate->type > LWTUNNEL_ENCAP_MAX) |
| 233 | return 0; |
| 234 | |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 235 | nest = nla_nest_start_noflag(skb, encap_attr); |
| 236 | if (!nest) |
| 237 | return -EMSGSIZE; |
| 238 | |
Googler | 0109c45 | 2022-10-13 17:50:39 +0800 | [diff] [blame] | 239 | ret = -EOPNOTSUPP; |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 240 | rcu_read_lock(); |
| 241 | ops = rcu_dereference(lwtun_encaps[lwtstate->type]); |
| 242 | if (likely(ops && ops->fill_encap)) |
| 243 | ret = ops->fill_encap(skb, lwtstate); |
| 244 | rcu_read_unlock(); |
| 245 | |
| 246 | if (ret) |
| 247 | goto nla_put_failure; |
| 248 | nla_nest_end(skb, nest); |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 249 | ret = nla_put_u16(skb, encap_type_attr, lwtstate->type); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 250 | if (ret) |
| 251 | goto nla_put_failure; |
| 252 | |
| 253 | return 0; |
| 254 | |
| 255 | nla_put_failure: |
| 256 | nla_nest_cancel(skb, nest); |
| 257 | |
| 258 | return (ret == -EOPNOTSUPP ? 0 : ret); |
| 259 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 260 | EXPORT_SYMBOL_GPL(lwtunnel_fill_encap); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 261 | |
| 262 | int lwtunnel_get_encap_size(struct lwtunnel_state *lwtstate) |
| 263 | { |
| 264 | const struct lwtunnel_encap_ops *ops; |
| 265 | int ret = 0; |
| 266 | |
| 267 | if (!lwtstate) |
| 268 | return 0; |
| 269 | |
| 270 | if (lwtstate->type == LWTUNNEL_ENCAP_NONE || |
| 271 | lwtstate->type > LWTUNNEL_ENCAP_MAX) |
| 272 | return 0; |
| 273 | |
| 274 | rcu_read_lock(); |
| 275 | ops = rcu_dereference(lwtun_encaps[lwtstate->type]); |
| 276 | if (likely(ops && ops->get_encap_size)) |
| 277 | ret = nla_total_size(ops->get_encap_size(lwtstate)); |
| 278 | rcu_read_unlock(); |
| 279 | |
| 280 | return ret; |
| 281 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 282 | EXPORT_SYMBOL_GPL(lwtunnel_get_encap_size); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 283 | |
| 284 | int lwtunnel_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b) |
| 285 | { |
| 286 | const struct lwtunnel_encap_ops *ops; |
| 287 | int ret = 0; |
| 288 | |
| 289 | if (!a && !b) |
| 290 | return 0; |
| 291 | |
| 292 | if (!a || !b) |
| 293 | return 1; |
| 294 | |
| 295 | if (a->type != b->type) |
| 296 | return 1; |
| 297 | |
| 298 | if (a->type == LWTUNNEL_ENCAP_NONE || |
| 299 | a->type > LWTUNNEL_ENCAP_MAX) |
| 300 | return 0; |
| 301 | |
| 302 | rcu_read_lock(); |
| 303 | ops = rcu_dereference(lwtun_encaps[a->type]); |
| 304 | if (likely(ops && ops->cmp_encap)) |
| 305 | ret = ops->cmp_encap(a, b); |
| 306 | rcu_read_unlock(); |
| 307 | |
| 308 | return ret; |
| 309 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 310 | EXPORT_SYMBOL_GPL(lwtunnel_cmp_encap); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 311 | |
| 312 | int lwtunnel_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
| 313 | { |
| 314 | struct dst_entry *dst = skb_dst(skb); |
| 315 | const struct lwtunnel_encap_ops *ops; |
| 316 | struct lwtunnel_state *lwtstate; |
| 317 | int ret = -EINVAL; |
| 318 | |
| 319 | if (!dst) |
| 320 | goto drop; |
| 321 | lwtstate = dst->lwtstate; |
| 322 | |
| 323 | if (lwtstate->type == LWTUNNEL_ENCAP_NONE || |
| 324 | lwtstate->type > LWTUNNEL_ENCAP_MAX) |
| 325 | return 0; |
| 326 | |
| 327 | ret = -EOPNOTSUPP; |
| 328 | rcu_read_lock(); |
| 329 | ops = rcu_dereference(lwtun_encaps[lwtstate->type]); |
| 330 | if (likely(ops && ops->output)) |
| 331 | ret = ops->output(net, sk, skb); |
| 332 | rcu_read_unlock(); |
| 333 | |
| 334 | if (ret == -EOPNOTSUPP) |
| 335 | goto drop; |
| 336 | |
| 337 | return ret; |
| 338 | |
| 339 | drop: |
| 340 | kfree_skb(skb); |
| 341 | |
| 342 | return ret; |
| 343 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 344 | EXPORT_SYMBOL_GPL(lwtunnel_output); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 345 | |
| 346 | int lwtunnel_xmit(struct sk_buff *skb) |
| 347 | { |
| 348 | struct dst_entry *dst = skb_dst(skb); |
| 349 | const struct lwtunnel_encap_ops *ops; |
| 350 | struct lwtunnel_state *lwtstate; |
| 351 | int ret = -EINVAL; |
| 352 | |
| 353 | if (!dst) |
| 354 | goto drop; |
| 355 | |
| 356 | lwtstate = dst->lwtstate; |
| 357 | |
| 358 | if (lwtstate->type == LWTUNNEL_ENCAP_NONE || |
| 359 | lwtstate->type > LWTUNNEL_ENCAP_MAX) |
| 360 | return 0; |
| 361 | |
| 362 | ret = -EOPNOTSUPP; |
| 363 | rcu_read_lock(); |
| 364 | ops = rcu_dereference(lwtun_encaps[lwtstate->type]); |
| 365 | if (likely(ops && ops->xmit)) |
| 366 | ret = ops->xmit(skb); |
| 367 | rcu_read_unlock(); |
| 368 | |
| 369 | if (ret == -EOPNOTSUPP) |
| 370 | goto drop; |
| 371 | |
| 372 | return ret; |
| 373 | |
| 374 | drop: |
| 375 | kfree_skb(skb); |
| 376 | |
| 377 | return ret; |
| 378 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 379 | EXPORT_SYMBOL_GPL(lwtunnel_xmit); |
Googler | af606d2 | 2022-10-26 21:40:12 -0700 | [diff] [blame] | 380 | |
| 381 | int lwtunnel_input(struct sk_buff *skb) |
| 382 | { |
| 383 | struct dst_entry *dst = skb_dst(skb); |
| 384 | const struct lwtunnel_encap_ops *ops; |
| 385 | struct lwtunnel_state *lwtstate; |
| 386 | int ret = -EINVAL; |
| 387 | |
| 388 | if (!dst) |
| 389 | goto drop; |
| 390 | lwtstate = dst->lwtstate; |
| 391 | |
| 392 | if (lwtstate->type == LWTUNNEL_ENCAP_NONE || |
| 393 | lwtstate->type > LWTUNNEL_ENCAP_MAX) |
| 394 | return 0; |
| 395 | |
| 396 | ret = -EOPNOTSUPP; |
| 397 | rcu_read_lock(); |
| 398 | ops = rcu_dereference(lwtun_encaps[lwtstate->type]); |
| 399 | if (likely(ops && ops->input)) |
| 400 | ret = ops->input(skb); |
| 401 | rcu_read_unlock(); |
| 402 | |
| 403 | if (ret == -EOPNOTSUPP) |
| 404 | goto drop; |
| 405 | |
| 406 | return ret; |
| 407 | |
| 408 | drop: |
| 409 | kfree_skb(skb); |
| 410 | |
| 411 | return ret; |
| 412 | } |
Googler | 9398cc3 | 2022-12-02 17:21:52 +0800 | [diff] [blame] | 413 | EXPORT_SYMBOL_GPL(lwtunnel_input); |