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 (nla_len(attr) != sizeof(*label)) { 138 NL_SET_ERR_MSG_MOD(extack, "Invalid MPLS label length"); 139 return -EINVAL; 140 } 141 142 if (*label & ~MPLS_LABEL_MASK || *label == MPLS_LABEL_IMPLNULL) { 143 NL_SET_ERR_MSG_MOD(extack, "MPLS label out of range"); 144 return -EINVAL; 145 } 146 147 return 0; 148 } 149 150 static const struct nla_policy mpls_policy[TCA_MPLS_MAX + 1] = { 151 [TCA_MPLS_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_mpls)), 152 [TCA_MPLS_PROTO] = { .type = NLA_U16 }, 153 [TCA_MPLS_LABEL] = NLA_POLICY_VALIDATE_FN(NLA_BINARY, 154 valid_label), 155 [TCA_MPLS_TC] = NLA_POLICY_RANGE(NLA_U8, 0, 7), 156 [TCA_MPLS_TTL] = NLA_POLICY_MIN(NLA_U8, 1), 157 [TCA_MPLS_BOS] = NLA_POLICY_RANGE(NLA_U8, 0, 1), 158 }; 159 160 static int tcf_mpls_init(struct net *net, struct nlattr *nla, 161 struct nlattr *est, struct tc_action **a, 162 struct tcf_proto *tp, u32 flags, 163 struct netlink_ext_ack *extack) 164 { 165 struct tc_action_net *tn = net_generic(net, act_mpls_ops.net_id); 166 bool bind = flags & TCA_ACT_FLAGS_BIND; 167 struct nlattr *tb[TCA_MPLS_MAX + 1]; 168 struct tcf_chain *goto_ch = NULL; 169 struct tcf_mpls_params *p; 170 struct tc_mpls *parm; 171 bool exists = false; 172 struct tcf_mpls *m; 173 int ret = 0, err; 174 u8 mpls_ttl = 0; 175 u32 index; 176 177 if (!nla) { 178 NL_SET_ERR_MSG_MOD(extack, "Missing netlink attributes"); 179 return -EINVAL; 180 } 181 182 err = nla_parse_nested(tb, TCA_MPLS_MAX, nla, mpls_policy, extack); 183 if (err < 0) 184 return err; 185 186 if (!tb[TCA_MPLS_PARMS]) { 187 NL_SET_ERR_MSG_MOD(extack, "No MPLS params"); 188 return -EINVAL; 189 } 190 parm = nla_data(tb[TCA_MPLS_PARMS]); 191 index = parm->index; 192 193 /* Verify parameters against action type. */ 194 switch (parm->m_action) { 195 case TCA_MPLS_ACT_POP: 196 if (!tb[TCA_MPLS_PROTO]) { 197 NL_SET_ERR_MSG_MOD(extack, "Protocol must be set for MPLS pop"); 198 return -EINVAL; 199 } 200 if (!eth_proto_is_802_3(nla_get_be16(tb[TCA_MPLS_PROTO]))) { 201 NL_SET_ERR_MSG_MOD(extack, "Invalid protocol type for MPLS pop"); 202 return -EINVAL; 203 } 204 if (tb[TCA_MPLS_LABEL] || tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] || 205 tb[TCA_MPLS_BOS]) { 206 NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC or BOS cannot be used with MPLS pop"); 207 return -EINVAL; 208 } 209 break; 210 case TCA_MPLS_ACT_DEC_TTL: 211 if (tb[TCA_MPLS_PROTO] || tb[TCA_MPLS_LABEL] || 212 tb[TCA_MPLS_TTL] || tb[TCA_MPLS_TC] || tb[TCA_MPLS_BOS]) { 213 NL_SET_ERR_MSG_MOD(extack, "Label, TTL, TC, BOS or protocol cannot be used with MPLS dec_ttl"); 214 return -EINVAL; 215 } 216 break; 217 case TCA_MPLS_ACT_PUSH: 218 case TCA_MPLS_ACT_MAC_PUSH: 219 if (!tb[TCA_MPLS_LABEL]) { 220 NL_SET_ERR_MSG_MOD(extack, "Label is required for MPLS push"); 221 return -EINVAL; 222 } 223 if (tb[TCA_MPLS_PROTO] && 224 !eth_p_mpls(nla_get_be16(tb[TCA_MPLS_PROTO]))) { 225 NL_SET_ERR_MSG_MOD(extack, "Protocol must be an MPLS type for MPLS push"); 226 return -EPROTONOSUPPORT; 227 } 228 /* Push needs a TTL - if not specified, set a default value. */ 229 if (!tb[TCA_MPLS_TTL]) { 230 #if IS_ENABLED(CONFIG_MPLS) 231 mpls_ttl = net->mpls.default_ttl ? 232 net->mpls.default_ttl : ACT_MPLS_TTL_DEFAULT; 233 #else 234 mpls_ttl = ACT_MPLS_TTL_DEFAULT; 235 #endif 236 } 237 break; 238 case TCA_MPLS_ACT_MODIFY: 239 if (tb[TCA_MPLS_PROTO]) { 240 NL_SET_ERR_MSG_MOD(extack, "Protocol cannot be used with MPLS modify"); 241 return -EINVAL; 242 } 243 break; 244 default: 245 NL_SET_ERR_MSG_MOD(extack, "Unknown MPLS action"); 246 return -EINVAL; 247 } 248 249 err = tcf_idr_check_alloc(tn, &index, a, bind); 250 if (err < 0) 251 return err; 252 exists = err; 253 if (exists && bind) 254 return 0; 255 256 if (!exists) { 257 ret = tcf_idr_create(tn, index, est, a, 258 &act_mpls_ops, bind, true, flags); 259 if (ret) { 260 tcf_idr_cleanup(tn, index); 261 return ret; 262 } 263 264 ret = ACT_P_CREATED; 265 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 266 tcf_idr_release(*a, bind); 267 return -EEXIST; 268 } 269 270 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 271 if (err < 0) 272 goto release_idr; 273 274 m = to_mpls(*a); 275 276 p = kzalloc(sizeof(*p), GFP_KERNEL); 277 if (!p) { 278 err = -ENOMEM; 279 goto put_chain; 280 } 281 282 p->tcfm_action = parm->m_action; 283 p->tcfm_label = tb[TCA_MPLS_LABEL] ? nla_get_u32(tb[TCA_MPLS_LABEL]) : 284 ACT_MPLS_LABEL_NOT_SET; 285 p->tcfm_tc = tb[TCA_MPLS_TC] ? nla_get_u8(tb[TCA_MPLS_TC]) : 286 ACT_MPLS_TC_NOT_SET; 287 p->tcfm_ttl = tb[TCA_MPLS_TTL] ? nla_get_u8(tb[TCA_MPLS_TTL]) : 288 mpls_ttl; 289 p->tcfm_bos = tb[TCA_MPLS_BOS] ? nla_get_u8(tb[TCA_MPLS_BOS]) : 290 ACT_MPLS_BOS_NOT_SET; 291 p->tcfm_proto = tb[TCA_MPLS_PROTO] ? nla_get_be16(tb[TCA_MPLS_PROTO]) : 292 htons(ETH_P_MPLS_UC); 293 294 spin_lock_bh(&m->tcf_lock); 295 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 296 p = rcu_replace_pointer(m->mpls_p, p, lockdep_is_held(&m->tcf_lock)); 297 spin_unlock_bh(&m->tcf_lock); 298 299 if (goto_ch) 300 tcf_chain_put_by_act(goto_ch); 301 if (p) 302 kfree_rcu(p, rcu); 303 304 return ret; 305 put_chain: 306 if (goto_ch) 307 tcf_chain_put_by_act(goto_ch); 308 release_idr: 309 tcf_idr_release(*a, bind); 310 return err; 311 } 312 313 static void tcf_mpls_cleanup(struct tc_action *a) 314 { 315 struct tcf_mpls *m = to_mpls(a); 316 struct tcf_mpls_params *p; 317 318 p = rcu_dereference_protected(m->mpls_p, 1); 319 if (p) 320 kfree_rcu(p, rcu); 321 } 322 323 static int tcf_mpls_dump(struct sk_buff *skb, struct tc_action *a, 324 int bind, int ref) 325 { 326 unsigned char *b = skb_tail_pointer(skb); 327 struct tcf_mpls *m = to_mpls(a); 328 struct tcf_mpls_params *p; 329 struct tc_mpls opt = { 330 .index = m->tcf_index, 331 .refcnt = refcount_read(&m->tcf_refcnt) - ref, 332 .bindcnt = atomic_read(&m->tcf_bindcnt) - bind, 333 }; 334 struct tcf_t t; 335 336 spin_lock_bh(&m->tcf_lock); 337 opt.action = m->tcf_action; 338 p = rcu_dereference_protected(m->mpls_p, lockdep_is_held(&m->tcf_lock)); 339 opt.m_action = p->tcfm_action; 340 341 if (nla_put(skb, TCA_MPLS_PARMS, sizeof(opt), &opt)) 342 goto nla_put_failure; 343 344 if (p->tcfm_label != ACT_MPLS_LABEL_NOT_SET && 345 nla_put_u32(skb, TCA_MPLS_LABEL, p->tcfm_label)) 346 goto nla_put_failure; 347 348 if (p->tcfm_tc != ACT_MPLS_TC_NOT_SET && 349 nla_put_u8(skb, TCA_MPLS_TC, p->tcfm_tc)) 350 goto nla_put_failure; 351 352 if (p->tcfm_ttl && nla_put_u8(skb, TCA_MPLS_TTL, p->tcfm_ttl)) 353 goto nla_put_failure; 354 355 if (p->tcfm_bos != ACT_MPLS_BOS_NOT_SET && 356 nla_put_u8(skb, TCA_MPLS_BOS, p->tcfm_bos)) 357 goto nla_put_failure; 358 359 if (nla_put_be16(skb, TCA_MPLS_PROTO, p->tcfm_proto)) 360 goto nla_put_failure; 361 362 tcf_tm_dump(&t, &m->tcf_tm); 363 364 if (nla_put_64bit(skb, TCA_MPLS_TM, sizeof(t), &t, TCA_MPLS_PAD)) 365 goto nla_put_failure; 366 367 spin_unlock_bh(&m->tcf_lock); 368 369 return skb->len; 370 371 nla_put_failure: 372 spin_unlock_bh(&m->tcf_lock); 373 nlmsg_trim(skb, b); 374 return -EMSGSIZE; 375 } 376 377 static int tcf_mpls_offload_act_setup(struct tc_action *act, void *entry_data, 378 u32 *index_inc, bool bind, 379 struct netlink_ext_ack *extack) 380 { 381 if (bind) { 382 struct flow_action_entry *entry = entry_data; 383 384 switch (tcf_mpls_action(act)) { 385 case TCA_MPLS_ACT_PUSH: 386 entry->id = FLOW_ACTION_MPLS_PUSH; 387 entry->mpls_push.proto = tcf_mpls_proto(act); 388 entry->mpls_push.label = tcf_mpls_label(act); 389 entry->mpls_push.tc = tcf_mpls_tc(act); 390 entry->mpls_push.bos = tcf_mpls_bos(act); 391 entry->mpls_push.ttl = tcf_mpls_ttl(act); 392 break; 393 case TCA_MPLS_ACT_POP: 394 entry->id = FLOW_ACTION_MPLS_POP; 395 entry->mpls_pop.proto = tcf_mpls_proto(act); 396 break; 397 case TCA_MPLS_ACT_MODIFY: 398 entry->id = FLOW_ACTION_MPLS_MANGLE; 399 entry->mpls_mangle.label = tcf_mpls_label(act); 400 entry->mpls_mangle.tc = tcf_mpls_tc(act); 401 entry->mpls_mangle.bos = tcf_mpls_bos(act); 402 entry->mpls_mangle.ttl = tcf_mpls_ttl(act); 403 break; 404 case TCA_MPLS_ACT_DEC_TTL: 405 NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"dec_ttl\" option is used"); 406 return -EOPNOTSUPP; 407 case TCA_MPLS_ACT_MAC_PUSH: 408 NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"mac_push\" option is used"); 409 return -EOPNOTSUPP; 410 default: 411 NL_SET_ERR_MSG_MOD(extack, "Unsupported MPLS mode offload"); 412 return -EOPNOTSUPP; 413 } 414 *index_inc = 1; 415 } else { 416 struct flow_offload_action *fl_action = entry_data; 417 418 switch (tcf_mpls_action(act)) { 419 case TCA_MPLS_ACT_PUSH: 420 fl_action->id = FLOW_ACTION_MPLS_PUSH; 421 break; 422 case TCA_MPLS_ACT_POP: 423 fl_action->id = FLOW_ACTION_MPLS_POP; 424 break; 425 case TCA_MPLS_ACT_MODIFY: 426 fl_action->id = FLOW_ACTION_MPLS_MANGLE; 427 break; 428 default: 429 return -EOPNOTSUPP; 430 } 431 } 432 433 return 0; 434 } 435 436 static struct tc_action_ops act_mpls_ops = { 437 .kind = "mpls", 438 .id = TCA_ID_MPLS, 439 .owner = THIS_MODULE, 440 .act = tcf_mpls_act, 441 .dump = tcf_mpls_dump, 442 .init = tcf_mpls_init, 443 .cleanup = tcf_mpls_cleanup, 444 .offload_act_setup = tcf_mpls_offload_act_setup, 445 .size = sizeof(struct tcf_mpls), 446 }; 447 448 static __net_init int mpls_init_net(struct net *net) 449 { 450 struct tc_action_net *tn = net_generic(net, act_mpls_ops.net_id); 451 452 return tc_action_net_init(net, tn, &act_mpls_ops); 453 } 454 455 static void __net_exit mpls_exit_net(struct list_head *net_list) 456 { 457 tc_action_net_exit(net_list, act_mpls_ops.net_id); 458 } 459 460 static struct pernet_operations mpls_net_ops = { 461 .init = mpls_init_net, 462 .exit_batch = mpls_exit_net, 463 .id = &act_mpls_ops.net_id, 464 .size = sizeof(struct tc_action_net), 465 }; 466 467 static int __init mpls_init_module(void) 468 { 469 return tcf_register_action(&act_mpls_ops, &mpls_net_ops); 470 } 471 472 static void __exit mpls_cleanup_module(void) 473 { 474 tcf_unregister_action(&act_mpls_ops, &mpls_net_ops); 475 } 476 477 module_init(mpls_init_module); 478 module_exit(mpls_cleanup_module); 479 480 MODULE_SOFTDEP("post: mpls_gso"); 481 MODULE_AUTHOR("Netronome Systems <oss-drivers@netronome.com>"); 482 MODULE_LICENSE("GPL"); 483 MODULE_DESCRIPTION("MPLS manipulation actions"); 484