1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) 2 /* Copyright (C) 2019 Netronome Systems, Inc. */ 3 4 #include <linux/if_arp.h> 5 #include <linux/init.h> 6 #include <linux/kernel.h> 7 #include <linux/module.h> 8 #include <linux/mpls.h> 9 #include <linux/rtnetlink.h> 10 #include <linux/skbuff.h> 11 #include <linux/tc_act/tc_mpls.h> 12 #include <net/mpls.h> 13 #include <net/netlink.h> 14 #include <net/pkt_sched.h> 15 #include <net/pkt_cls.h> 16 #include <net/tc_act/tc_mpls.h> 17 #include <net/tc_wrapper.h> 18 19 static struct tc_action_ops act_mpls_ops; 20 21 #define ACT_MPLS_TTL_DEFAULT 255 22 23 static __be32 tcf_mpls_get_lse(struct mpls_shim_hdr *lse, 24 struct tcf_mpls_params *p, bool set_bos) 25 { 26 u32 new_lse = 0; 27 28 if (lse) 29 new_lse = be32_to_cpu(lse->label_stack_entry); 30 31 if (p->tcfm_label != ACT_MPLS_LABEL_NOT_SET) { 32 new_lse &= ~MPLS_LS_LABEL_MASK; 33 new_lse |= p->tcfm_label << MPLS_LS_LABEL_SHIFT; 34 } 35 if (p->tcfm_ttl) { 36 new_lse &= ~MPLS_LS_TTL_MASK; 37 new_lse |= p->tcfm_ttl << MPLS_LS_TTL_SHIFT; 38 } 39 if (p->tcfm_tc != ACT_MPLS_TC_NOT_SET) { 40 new_lse &= ~MPLS_LS_TC_MASK; 41 new_lse |= p->tcfm_tc << MPLS_LS_TC_SHIFT; 42 } 43 if (p->tcfm_bos != ACT_MPLS_BOS_NOT_SET) { 44 new_lse &= ~MPLS_LS_S_MASK; 45 new_lse |= p->tcfm_bos << MPLS_LS_S_SHIFT; 46 } else if (set_bos) { 47 new_lse |= 1 << MPLS_LS_S_SHIFT; 48 } 49 50 return cpu_to_be32(new_lse); 51 } 52 53 TC_INDIRECT_SCOPE int tcf_mpls_act(struct sk_buff *skb, 54 const struct tc_action *a, 55 struct tcf_result *res) 56 { 57 struct tcf_mpls *m = to_mpls(a); 58 struct tcf_mpls_params *p; 59 __be32 new_lse; 60 int ret, mac_len; 61 62 tcf_lastuse_update(&m->tcf_tm); 63 bstats_update(this_cpu_ptr(m->common.cpu_bstats), skb); 64 65 /* Ensure 'data' points at mac_header prior calling mpls manipulating 66 * functions. 67 */ 68 if (skb_at_tc_ingress(skb)) { 69 skb_push_rcsum(skb, skb->mac_len); 70 mac_len = skb->mac_len; 71 } else { 72 mac_len = skb_network_header(skb) - skb_mac_header(skb); 73 } 74 75 ret = READ_ONCE(m->tcf_action); 76 77 p = rcu_dereference_bh(m->mpls_p); 78 79 switch (p->tcfm_action) { 80 case TCA_MPLS_ACT_POP: 81 if (skb_mpls_pop(skb, p->tcfm_proto, mac_len, 82 skb->dev && skb->dev->type == ARPHRD_ETHER)) 83 goto drop; 84 break; 85 case TCA_MPLS_ACT_PUSH: 86 new_lse = tcf_mpls_get_lse(NULL, p, !eth_p_mpls(skb_protocol(skb, true))); 87 if (skb_mpls_push(skb, new_lse, p->tcfm_proto, mac_len, 88 skb->dev && skb->dev->type == ARPHRD_ETHER)) 89 goto drop; 90 break; 91 case TCA_MPLS_ACT_MAC_PUSH: 92 if (skb_vlan_tag_present(skb)) { 93 if (__vlan_insert_inner_tag(skb, skb->vlan_proto, 94 skb_vlan_tag_get(skb), 95 ETH_HLEN) < 0) 96 goto drop; 97 98 skb->protocol = skb->vlan_proto; 99 __vlan_hwaccel_clear_tag(skb); 100 } 101 102 new_lse = tcf_mpls_get_lse(NULL, p, mac_len || 103 !eth_p_mpls(skb->protocol)); 104 105 if (skb_mpls_push(skb, new_lse, p->tcfm_proto, 0, false)) 106 goto drop; 107 break; 108 case TCA_MPLS_ACT_MODIFY: 109 if (!pskb_may_pull(skb, 110 skb_network_offset(skb) + MPLS_HLEN)) 111 goto drop; 112 new_lse = tcf_mpls_get_lse(mpls_hdr(skb), p, false); 113 if (skb_mpls_update_lse(skb, new_lse)) 114 goto drop; 115 break; 116 case TCA_MPLS_ACT_DEC_TTL: 117 if (skb_mpls_dec_ttl(skb)) 118 goto drop; 119 break; 120 } 121 122 if (skb_at_tc_ingress(skb)) 123 skb_pull_rcsum(skb, skb->mac_len); 124 125 return ret; 126 127 drop: 128 qstats_drop_inc(this_cpu_ptr(m->common.cpu_qstats)); 129 return TC_ACT_SHOT; 130 } 131 132 static int valid_label(const struct nlattr *attr, 133 struct netlink_ext_ack *extack) 134 { 135 const u32 *label = nla_data(attr); 136 137 if (*label & ~MPLS_LABEL_MASK || *label == MPLS_LABEL_IMPLNULL) { 138 NL_SET_ERR_MSG_MOD(extack, "MPLS label out of range"); 139 return -EINVAL; 140 } 141 142 return 0; 143 } 144 145 static const struct nla_policy mpls_policy[TCA_MPLS_MAX + 1] = { 146 [TCA_MPLS_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_mpls)), 147 [TCA_MPLS_PROTO] = { .type = NLA_U16 }, 148 [TCA_MPLS_LABEL] = NLA_POLICY_VALIDATE_FN(NLA_U32, valid_label), 149 [TCA_MPLS_TC] = NLA_POLICY_RANGE(NLA_U8, 0, 7), 150 [TCA_MPLS_TTL] = NLA_POLICY_MIN(NLA_U8, 1), 151 [TCA_MPLS_BOS] = NLA_POLICY_RANGE(NLA_U8, 0, 1), 152 }; 153 154 static int tcf_mpls_init(struct net *net, struct nlattr *nla, 155 struct nlattr *est, struct tc_action **a, 156 struct tcf_proto *tp, u32 flags, 157 struct netlink_ext_ack *extack) 158 { 159 struct tc_action_net *tn = net_generic(net, act_mpls_ops.net_id); 160 bool bind = flags & TCA_ACT_FLAGS_BIND; 161 struct nlattr *tb[TCA_MPLS_MAX + 1]; 162 struct tcf_chain *goto_ch = NULL; 163 struct tcf_mpls_params *p; 164 struct tc_mpls *parm; 165 bool exists = false; 166 struct tcf_mpls *m; 167 int ret = 0, err; 168 u8 mpls_ttl = 0; 169 u32 index; 170 171 if (!nla) { 172 NL_SET_ERR_MSG_MOD(extack, "Missing netlink attributes"); 173 return -EINVAL; 174 } 175 176 err = nla_parse_nested(tb, TCA_MPLS_MAX, nla, mpls_policy, extack); 177 if (err < 0) 178 return err; 179 180 if (!tb[TCA_MPLS_PARMS]) { 181 NL_SET_ERR_MSG_MOD(extack, "No MPLS params"); 182 return -EINVAL; 183 } 184 parm = nla_data(tb[TCA_MPLS_PARMS]); 185 index = parm->index; 186 187 /* Verify parameters against action type. */ 188 switch (parm->m_action) { 189 case TCA_MPLS_ACT_POP: 190 if (!tb[TCA_MPLS_PROTO]) { 191 NL_SET_ERR_MSG_MOD(extack, "Protocol must be set for MPLS pop"); 192 return -EINVAL; 193 } 194 if (!eth_proto_is_802_3(nla_get_be16(tb[TCA_MPLS_PROTO]))) { 195 NL_SET_ERR_MSG_MOD(extack, "Invalid protocol type for MPLS pop"); 196 return -EINVAL; 197 } 198 if (tb[TCA_MPLS_LABEL] || tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] || 199 tb[TCA_MPLS_BOS]) { 200 NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC or BOS cannot be used with MPLS pop"); 201 return -EINVAL; 202 } 203 break; 204 case TCA_MPLS_ACT_DEC_TTL: 205 if (tb[TCA_MPLS_PROTO] || tb[TCA_MPLS_LABEL] || 206 tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] || tb[TCA_MPLS_BOS]) { 207 NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC, BOS or protocol cannot be used with MPLS dec_ttl"); 208 return -EINVAL; 209 } 210 break; 211 case TCA_MPLS_ACT_PUSH: 212 case TCA_MPLS_ACT_MAC_PUSH: 213 if (!tb[TCA_MPLS_LABEL]) { 214 NL_SET_ERR_MSG_MOD(extack, "Label is required for MPLS push"); 215 return -EINVAL; 216 } 217 if (tb[TCA_MPLS_PROTO] && 218 !eth_p_mpls(nla_get_be16(tb[TCA_MPLS_PROTO]))) { 219 NL_SET_ERR_MSG_MOD(extack, "Protocol must be an MPLS type for MPLS push"); 220 return -EPROTONOSUPPORT; 221 } 222 /* Push needs a TTL - if not specified, set a default value. */ 223 if (!tb[TCA_MPLS_TTL]) { 224 #if IS_ENABLED(CONFIG_MPLS) 225 mpls_ttl = net->mpls.default_ttl ? 226 net->mpls.default_ttl : ACT_MPLS_TTL_DEFAULT; 227 #else 228 mpls_ttl = ACT_MPLS_TTL_DEFAULT; 229 #endif 230 } 231 break; 232 case TCA_MPLS_ACT_MODIFY: 233 if (tb[TCA_MPLS_PROTO]) { 234 NL_SET_ERR_MSG_MOD(extack, "Protocol cannot be used with MPLS modify"); 235 return -EINVAL; 236 } 237 break; 238 default: 239 NL_SET_ERR_MSG_MOD(extack, "Unknown MPLS action"); 240 return -EINVAL; 241 } 242 243 err = tcf_idr_check_alloc(tn, &index, a, bind); 244 if (err < 0) 245 return err; 246 exists = err; 247 if (exists && bind) 248 return 0; 249 250 if (!exists) { 251 ret = tcf_idr_create(tn, index, est, a, 252 &act_mpls_ops, bind, true, flags); 253 if (ret) { 254 tcf_idr_cleanup(tn, index); 255 return ret; 256 } 257 258 ret = ACT_P_CREATED; 259 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 260 tcf_idr_release(*a, bind); 261 return -EEXIST; 262 } 263 264 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 265 if (err < 0) 266 goto release_idr; 267 268 m = to_mpls(*a); 269 270 p = kzalloc(sizeof(*p), GFP_KERNEL); 271 if (!p) { 272 err = -ENOMEM; 273 goto put_chain; 274 } 275 276 p->tcfm_action = parm->m_action; 277 p->tcfm_label = tb[TCA_MPLS_LABEL] ? nla_get_u32(tb[TCA_MPLS_LABEL]) : 278 ACT_MPLS_LABEL_NOT_SET; 279 p->tcfm_tc = tb[TCA_MPLS_TC] ? nla_get_u8(tb[TCA_MPLS_TC]) : 280 ACT_MPLS_TC_NOT_SET; 281 p->tcfm_ttl = tb[TCA_MPLS_TTL] ? nla_get_u8(tb[TCA_MPLS_TTL]) : 282 mpls_ttl; 283 p->tcfm_bos = tb[TCA_MPLS_BOS] ? nla_get_u8(tb[TCA_MPLS_BOS]) : 284 ACT_MPLS_BOS_NOT_SET; 285 p->tcfm_proto = tb[TCA_MPLS_PROTO] ? nla_get_be16(tb[TCA_MPLS_PROTO]) : 286 htons(ETH_P_MPLS_UC); 287 288 spin_lock_bh(&m->tcf_lock); 289 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 290 p = rcu_replace_pointer(m->mpls_p, p, lockdep_is_held(&m->tcf_lock)); 291 spin_unlock_bh(&m->tcf_lock); 292 293 if (goto_ch) 294 tcf_chain_put_by_act(goto_ch); 295 if (p) 296 kfree_rcu(p, rcu); 297 298 return ret; 299 put_chain: 300 if (goto_ch) 301 tcf_chain_put_by_act(goto_ch); 302 release_idr: 303 tcf_idr_release(*a, bind); 304 return err; 305 } 306 307 static void tcf_mpls_cleanup(struct tc_action *a) 308 { 309 struct tcf_mpls *m = to_mpls(a); 310 struct tcf_mpls_params *p; 311 312 p = rcu_dereference_protected(m->mpls_p, 1); 313 if (p) 314 kfree_rcu(p, rcu); 315 } 316 317 static int tcf_mpls_dump(struct sk_buff *skb, struct tc_action *a, 318 int bind, int ref) 319 { 320 unsigned char *b = skb_tail_pointer(skb); 321 struct tcf_mpls *m = to_mpls(a); 322 struct tcf_mpls_params *p; 323 struct tc_mpls opt = { 324 .index = m->tcf_index, 325 .refcnt = refcount_read(&m->tcf_refcnt) - ref, 326 .bindcnt = atomic_read(&m->tcf_bindcnt) - bind, 327 }; 328 struct tcf_t t; 329 330 spin_lock_bh(&m->tcf_lock); 331 opt.action = m->tcf_action; 332 p = rcu_dereference_protected(m->mpls_p, lockdep_is_held(&m->tcf_lock)); 333 opt.m_action = p->tcfm_action; 334 335 if (nla_put(skb, TCA_MPLS_PARMS, sizeof(opt), &opt)) 336 goto nla_put_failure; 337 338 if (p->tcfm_label != ACT_MPLS_LABEL_NOT_SET && 339 nla_put_u32(skb, TCA_MPLS_LABEL, p->tcfm_label)) 340 goto nla_put_failure; 341 342 if (p->tcfm_tc != ACT_MPLS_TC_NOT_SET && 343 nla_put_u8(skb, TCA_MPLS_TC, p->tcfm_tc)) 344 goto nla_put_failure; 345 346 if (p->tcfm_ttl && nla_put_u8(skb, TCA_MPLS_TTL, p->tcfm_ttl)) 347 goto nla_put_failure; 348 349 if (p->tcfm_bos != ACT_MPLS_BOS_NOT_SET && 350 nla_put_u8(skb, TCA_MPLS_BOS, p->tcfm_bos)) 351 goto nla_put_failure; 352 353 if (nla_put_be16(skb, TCA_MPLS_PROTO, p->tcfm_proto)) 354 goto nla_put_failure; 355 356 tcf_tm_dump(&t, &m->tcf_tm); 357 358 if (nla_put_64bit(skb, TCA_MPLS_TM, sizeof(t), &t, TCA_MPLS_PAD)) 359 goto nla_put_failure; 360 361 spin_unlock_bh(&m->tcf_lock); 362 363 return skb->len; 364 365 nla_put_failure: 366 spin_unlock_bh(&m->tcf_lock); 367 nlmsg_trim(skb, b); 368 return -EMSGSIZE; 369 } 370 371 static int tcf_mpls_offload_act_setup(struct tc_action *act, void *entry_data, 372 u32 *index_inc, bool bind, 373 struct netlink_ext_ack *extack) 374 { 375 if (bind) { 376 struct flow_action_entry *entry = entry_data; 377 378 switch (tcf_mpls_action(act)) { 379 case TCA_MPLS_ACT_PUSH: 380 entry->id = FLOW_ACTION_MPLS_PUSH; 381 entry->mpls_push.proto = tcf_mpls_proto(act); 382 entry->mpls_push.label = tcf_mpls_label(act); 383 entry->mpls_push.tc = tcf_mpls_tc(act); 384 entry->mpls_push.bos = tcf_mpls_bos(act); 385 entry->mpls_push.ttl = tcf_mpls_ttl(act); 386 break; 387 case TCA_MPLS_ACT_POP: 388 entry->id = FLOW_ACTION_MPLS_POP; 389 entry->mpls_pop.proto = tcf_mpls_proto(act); 390 break; 391 case TCA_MPLS_ACT_MODIFY: 392 entry->id = FLOW_ACTION_MPLS_MANGLE; 393 entry->mpls_mangle.label = tcf_mpls_label(act); 394 entry->mpls_mangle.tc = tcf_mpls_tc(act); 395 entry->mpls_mangle.bos = tcf_mpls_bos(act); 396 entry->mpls_mangle.ttl = tcf_mpls_ttl(act); 397 break; 398 case TCA_MPLS_ACT_DEC_TTL: 399 NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"dec_ttl\" option is used"); 400 return -EOPNOTSUPP; 401 case TCA_MPLS_ACT_MAC_PUSH: 402 NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"mac_push\" option is used"); 403 return -EOPNOTSUPP; 404 default: 405 NL_SET_ERR_MSG_MOD(extack, "Unsupported MPLS mode offload"); 406 return -EOPNOTSUPP; 407 } 408 *index_inc = 1; 409 } else { 410 struct flow_offload_action *fl_action = entry_data; 411 412 switch (tcf_mpls_action(act)) { 413 case TCA_MPLS_ACT_PUSH: 414 fl_action->id = FLOW_ACTION_MPLS_PUSH; 415 break; 416 case TCA_MPLS_ACT_POP: 417 fl_action->id = FLOW_ACTION_MPLS_POP; 418 break; 419 case TCA_MPLS_ACT_MODIFY: 420 fl_action->id = FLOW_ACTION_MPLS_MANGLE; 421 break; 422 default: 423 return -EOPNOTSUPP; 424 } 425 } 426 427 return 0; 428 } 429 430 static struct tc_action_ops act_mpls_ops = { 431 .kind = "mpls", 432 .id = TCA_ID_MPLS, 433 .owner = THIS_MODULE, 434 .act = tcf_mpls_act, 435 .dump = tcf_mpls_dump, 436 .init = tcf_mpls_init, 437 .cleanup = tcf_mpls_cleanup, 438 .offload_act_setup = tcf_mpls_offload_act_setup, 439 .size = sizeof(struct tcf_mpls), 440 }; 441 442 static __net_init int mpls_init_net(struct net *net) 443 { 444 struct tc_action_net *tn = net_generic(net, act_mpls_ops.net_id); 445 446 return tc_action_net_init(net, tn, &act_mpls_ops); 447 } 448 449 static void __net_exit mpls_exit_net(struct list_head *net_list) 450 { 451 tc_action_net_exit(net_list, act_mpls_ops.net_id); 452 } 453 454 static struct pernet_operations mpls_net_ops = { 455 .init = mpls_init_net, 456 .exit_batch = mpls_exit_net, 457 .id = &act_mpls_ops.net_id, 458 .size = sizeof(struct tc_action_net), 459 }; 460 461 static int __init mpls_init_module(void) 462 { 463 return tcf_register_action(&act_mpls_ops, &mpls_net_ops); 464 } 465 466 static void __exit mpls_cleanup_module(void) 467 { 468 tcf_unregister_action(&act_mpls_ops, &mpls_net_ops); 469 } 470 471 module_init(mpls_init_module); 472 module_exit(mpls_cleanup_module); 473 474 MODULE_SOFTDEP("post: mpls_gso"); 475 MODULE_AUTHOR("Netronome Systems <oss-drivers@netronome.com>"); 476 MODULE_LICENSE("GPL"); 477 MODULE_DESCRIPTION("MPLS manipulation actions"); 478