1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB 2 /* - 3 * net/sched/act_ct.c Connection Tracking action 4 * 5 * Authors: Paul Blakey <paulb@mellanox.com> 6 * Yossi Kuperman <yossiku@mellanox.com> 7 * Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> 8 */ 9 10 #include <linux/module.h> 11 #include <linux/init.h> 12 #include <linux/kernel.h> 13 #include <linux/skbuff.h> 14 #include <linux/rtnetlink.h> 15 #include <linux/pkt_cls.h> 16 #include <linux/ip.h> 17 #include <linux/ipv6.h> 18 #include <linux/rhashtable.h> 19 #include <net/netlink.h> 20 #include <net/pkt_sched.h> 21 #include <net/pkt_cls.h> 22 #include <net/act_api.h> 23 #include <net/ip.h> 24 #include <net/ipv6_frag.h> 25 #include <uapi/linux/tc_act/tc_ct.h> 26 #include <net/tc_act/tc_ct.h> 27 #include <net/tc_wrapper.h> 28 29 #include <net/netfilter/nf_flow_table.h> 30 #include <net/netfilter/nf_conntrack.h> 31 #include <net/netfilter/nf_conntrack_core.h> 32 #include <net/netfilter/nf_conntrack_zones.h> 33 #include <net/netfilter/nf_conntrack_helper.h> 34 #include <net/netfilter/nf_conntrack_acct.h> 35 #include <net/netfilter/ipv6/nf_defrag_ipv6.h> 36 #include <net/netfilter/nf_conntrack_act_ct.h> 37 #include <net/netfilter/nf_conntrack_seqadj.h> 38 #include <uapi/linux/netfilter/nf_nat.h> 39 40 static struct workqueue_struct *act_ct_wq; 41 static struct rhashtable zones_ht; 42 static DEFINE_MUTEX(zones_mutex); 43 44 struct tcf_ct_flow_table { 45 struct rhash_head node; /* In zones tables */ 46 47 struct rcu_work rwork; 48 struct nf_flowtable nf_ft; 49 refcount_t ref; 50 u16 zone; 51 52 bool dying; 53 }; 54 55 static const struct rhashtable_params zones_params = { 56 .head_offset = offsetof(struct tcf_ct_flow_table, node), 57 .key_offset = offsetof(struct tcf_ct_flow_table, zone), 58 .key_len = sizeof_field(struct tcf_ct_flow_table, zone), 59 .automatic_shrinking = true, 60 }; 61 62 static struct flow_action_entry * 63 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action) 64 { 65 int i = flow_action->num_entries++; 66 67 return &flow_action->entries[i]; 68 } 69 70 static void tcf_ct_add_mangle_action(struct flow_action *action, 71 enum flow_action_mangle_base htype, 72 u32 offset, 73 u32 mask, 74 u32 val) 75 { 76 struct flow_action_entry *entry; 77 78 entry = tcf_ct_flow_table_flow_action_get_next(action); 79 entry->id = FLOW_ACTION_MANGLE; 80 entry->mangle.htype = htype; 81 entry->mangle.mask = ~mask; 82 entry->mangle.offset = offset; 83 entry->mangle.val = val; 84 } 85 86 /* The following nat helper functions check if the inverted reverse tuple 87 * (target) is different then the current dir tuple - meaning nat for ports 88 * and/or ip is needed, and add the relevant mangle actions. 89 */ 90 static void 91 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple, 92 struct nf_conntrack_tuple target, 93 struct flow_action *action) 94 { 95 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 96 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 97 offsetof(struct iphdr, saddr), 98 0xFFFFFFFF, 99 be32_to_cpu(target.src.u3.ip)); 100 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 101 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 102 offsetof(struct iphdr, daddr), 103 0xFFFFFFFF, 104 be32_to_cpu(target.dst.u3.ip)); 105 } 106 107 static void 108 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action, 109 union nf_inet_addr *addr, 110 u32 offset) 111 { 112 int i; 113 114 for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++) 115 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6, 116 i * sizeof(u32) + offset, 117 0xFFFFFFFF, be32_to_cpu(addr->ip6[i])); 118 } 119 120 static void 121 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple, 122 struct nf_conntrack_tuple target, 123 struct flow_action *action) 124 { 125 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 126 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3, 127 offsetof(struct ipv6hdr, 128 saddr)); 129 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 130 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3, 131 offsetof(struct ipv6hdr, 132 daddr)); 133 } 134 135 static void 136 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple, 137 struct nf_conntrack_tuple target, 138 struct flow_action *action) 139 { 140 __be16 target_src = target.src.u.tcp.port; 141 __be16 target_dst = target.dst.u.tcp.port; 142 143 if (target_src != tuple->src.u.tcp.port) 144 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 145 offsetof(struct tcphdr, source), 146 0xFFFF, be16_to_cpu(target_src)); 147 if (target_dst != tuple->dst.u.tcp.port) 148 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 149 offsetof(struct tcphdr, dest), 150 0xFFFF, be16_to_cpu(target_dst)); 151 } 152 153 static void 154 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple, 155 struct nf_conntrack_tuple target, 156 struct flow_action *action) 157 { 158 __be16 target_src = target.src.u.udp.port; 159 __be16 target_dst = target.dst.u.udp.port; 160 161 if (target_src != tuple->src.u.udp.port) 162 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 163 offsetof(struct udphdr, source), 164 0xFFFF, be16_to_cpu(target_src)); 165 if (target_dst != tuple->dst.u.udp.port) 166 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 167 offsetof(struct udphdr, dest), 168 0xFFFF, be16_to_cpu(target_dst)); 169 } 170 171 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct, 172 enum ip_conntrack_dir dir, 173 enum ip_conntrack_info ctinfo, 174 struct flow_action *action) 175 { 176 struct nf_conn_labels *ct_labels; 177 struct flow_action_entry *entry; 178 u32 *act_ct_labels; 179 180 entry = tcf_ct_flow_table_flow_action_get_next(action); 181 entry->id = FLOW_ACTION_CT_METADATA; 182 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 183 entry->ct_metadata.mark = READ_ONCE(ct->mark); 184 #endif 185 /* aligns with the CT reference on the SKB nf_ct_set */ 186 entry->ct_metadata.cookie = (unsigned long)ct | ctinfo; 187 entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL; 188 189 act_ct_labels = entry->ct_metadata.labels; 190 ct_labels = nf_ct_labels_find(ct); 191 if (ct_labels) 192 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE); 193 else 194 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE); 195 } 196 197 static int tcf_ct_flow_table_add_action_nat(struct net *net, 198 struct nf_conn *ct, 199 enum ip_conntrack_dir dir, 200 struct flow_action *action) 201 { 202 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple; 203 struct nf_conntrack_tuple target; 204 205 if (!(ct->status & IPS_NAT_MASK)) 206 return 0; 207 208 nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple); 209 210 switch (tuple->src.l3num) { 211 case NFPROTO_IPV4: 212 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target, 213 action); 214 break; 215 case NFPROTO_IPV6: 216 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target, 217 action); 218 break; 219 default: 220 return -EOPNOTSUPP; 221 } 222 223 switch (nf_ct_protonum(ct)) { 224 case IPPROTO_TCP: 225 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action); 226 break; 227 case IPPROTO_UDP: 228 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action); 229 break; 230 default: 231 return -EOPNOTSUPP; 232 } 233 234 return 0; 235 } 236 237 static int tcf_ct_flow_table_fill_actions(struct net *net, 238 struct flow_offload *flow, 239 enum flow_offload_tuple_dir tdir, 240 struct nf_flow_rule *flow_rule) 241 { 242 struct flow_action *action = &flow_rule->rule->action; 243 int num_entries = action->num_entries; 244 struct nf_conn *ct = flow->ct; 245 enum ip_conntrack_info ctinfo; 246 enum ip_conntrack_dir dir; 247 int i, err; 248 249 switch (tdir) { 250 case FLOW_OFFLOAD_DIR_ORIGINAL: 251 dir = IP_CT_DIR_ORIGINAL; 252 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ? 253 IP_CT_ESTABLISHED : IP_CT_NEW; 254 if (ctinfo == IP_CT_ESTABLISHED) 255 set_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags); 256 break; 257 case FLOW_OFFLOAD_DIR_REPLY: 258 dir = IP_CT_DIR_REPLY; 259 ctinfo = IP_CT_ESTABLISHED_REPLY; 260 break; 261 default: 262 return -EOPNOTSUPP; 263 } 264 265 err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action); 266 if (err) 267 goto err_nat; 268 269 tcf_ct_flow_table_add_action_meta(ct, dir, ctinfo, action); 270 return 0; 271 272 err_nat: 273 /* Clear filled actions */ 274 for (i = num_entries; i < action->num_entries; i++) 275 memset(&action->entries[i], 0, sizeof(action->entries[i])); 276 action->num_entries = num_entries; 277 278 return err; 279 } 280 281 static bool tcf_ct_flow_is_outdated(const struct flow_offload *flow) 282 { 283 return test_bit(IPS_SEEN_REPLY_BIT, &flow->ct->status) && 284 test_bit(IPS_HW_OFFLOAD_BIT, &flow->ct->status) && 285 !test_bit(NF_FLOW_HW_PENDING, &flow->flags) && 286 !test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags); 287 } 288 289 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft); 290 291 static void tcf_ct_nf_get(struct nf_flowtable *ft) 292 { 293 struct tcf_ct_flow_table *ct_ft = 294 container_of(ft, struct tcf_ct_flow_table, nf_ft); 295 296 tcf_ct_flow_table_get_ref(ct_ft); 297 } 298 299 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft); 300 301 static void tcf_ct_nf_put(struct nf_flowtable *ft) 302 { 303 struct tcf_ct_flow_table *ct_ft = 304 container_of(ft, struct tcf_ct_flow_table, nf_ft); 305 306 tcf_ct_flow_table_put(ct_ft); 307 } 308 309 static struct nf_flowtable_type flowtable_ct = { 310 .gc = tcf_ct_flow_is_outdated, 311 .action = tcf_ct_flow_table_fill_actions, 312 .get = tcf_ct_nf_get, 313 .put = tcf_ct_nf_put, 314 .owner = THIS_MODULE, 315 }; 316 317 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params) 318 { 319 struct tcf_ct_flow_table *ct_ft; 320 int err = -ENOMEM; 321 322 mutex_lock(&zones_mutex); 323 ct_ft = rhashtable_lookup_fast(&zones_ht, ¶ms->zone, zones_params); 324 if (ct_ft && refcount_inc_not_zero(&ct_ft->ref)) 325 goto out_unlock; 326 327 ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL); 328 if (!ct_ft) 329 goto err_alloc; 330 refcount_set(&ct_ft->ref, 1); 331 332 ct_ft->zone = params->zone; 333 err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params); 334 if (err) 335 goto err_insert; 336 337 ct_ft->nf_ft.type = &flowtable_ct; 338 ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD | 339 NF_FLOWTABLE_COUNTER; 340 err = nf_flow_table_init(&ct_ft->nf_ft); 341 if (err) 342 goto err_init; 343 write_pnet(&ct_ft->nf_ft.net, net); 344 345 __module_get(THIS_MODULE); 346 out_unlock: 347 params->ct_ft = ct_ft; 348 params->nf_ft = &ct_ft->nf_ft; 349 mutex_unlock(&zones_mutex); 350 351 return 0; 352 353 err_init: 354 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 355 err_insert: 356 kfree(ct_ft); 357 err_alloc: 358 mutex_unlock(&zones_mutex); 359 return err; 360 } 361 362 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft) 363 { 364 refcount_inc(&ct_ft->ref); 365 } 366 367 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work) 368 { 369 struct tcf_ct_flow_table *ct_ft; 370 struct flow_block *block; 371 372 ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table, 373 rwork); 374 nf_flow_table_free(&ct_ft->nf_ft); 375 376 block = &ct_ft->nf_ft.flow_block; 377 down_write(&ct_ft->nf_ft.flow_block_lock); 378 WARN_ON(!list_empty(&block->cb_list)); 379 up_write(&ct_ft->nf_ft.flow_block_lock); 380 kfree(ct_ft); 381 382 module_put(THIS_MODULE); 383 } 384 385 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft) 386 { 387 if (refcount_dec_and_test(&ct_ft->ref)) { 388 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 389 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work); 390 queue_rcu_work(act_ct_wq, &ct_ft->rwork); 391 } 392 } 393 394 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry, 395 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir) 396 { 397 entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC; 398 entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir]; 399 } 400 401 static void tcf_ct_flow_ct_ext_ifidx_update(struct flow_offload *entry) 402 { 403 struct nf_conn_act_ct_ext *act_ct_ext; 404 405 act_ct_ext = nf_conn_act_ct_ext_find(entry->ct); 406 if (act_ct_ext) { 407 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL); 408 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY); 409 } 410 } 411 412 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft, 413 struct nf_conn *ct, 414 bool tcp, bool bidirectional) 415 { 416 struct nf_conn_act_ct_ext *act_ct_ext; 417 struct flow_offload *entry; 418 int err; 419 420 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status)) 421 return; 422 423 entry = flow_offload_alloc(ct); 424 if (!entry) { 425 WARN_ON_ONCE(1); 426 goto err_alloc; 427 } 428 429 if (tcp) { 430 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 431 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 432 } 433 if (bidirectional) 434 __set_bit(NF_FLOW_HW_BIDIRECTIONAL, &entry->flags); 435 436 act_ct_ext = nf_conn_act_ct_ext_find(ct); 437 if (act_ct_ext) { 438 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL); 439 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY); 440 } 441 442 err = flow_offload_add(&ct_ft->nf_ft, entry); 443 if (err) 444 goto err_add; 445 446 return; 447 448 err_add: 449 flow_offload_free(entry); 450 err_alloc: 451 clear_bit(IPS_OFFLOAD_BIT, &ct->status); 452 } 453 454 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft, 455 struct nf_conn *ct, 456 enum ip_conntrack_info ctinfo) 457 { 458 bool tcp = false, bidirectional = true; 459 460 switch (nf_ct_protonum(ct)) { 461 case IPPROTO_TCP: 462 if ((ctinfo != IP_CT_ESTABLISHED && 463 ctinfo != IP_CT_ESTABLISHED_REPLY) || 464 !test_bit(IPS_ASSURED_BIT, &ct->status) || 465 ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED) 466 return; 467 468 tcp = true; 469 break; 470 case IPPROTO_UDP: 471 if (!nf_ct_is_confirmed(ct)) 472 return; 473 if (!test_bit(IPS_ASSURED_BIT, &ct->status)) 474 bidirectional = false; 475 break; 476 #ifdef CONFIG_NF_CT_PROTO_GRE 477 case IPPROTO_GRE: { 478 struct nf_conntrack_tuple *tuple; 479 480 if ((ctinfo != IP_CT_ESTABLISHED && 481 ctinfo != IP_CT_ESTABLISHED_REPLY) || 482 !test_bit(IPS_ASSURED_BIT, &ct->status) || 483 ct->status & IPS_NAT_MASK) 484 return; 485 486 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; 487 /* No support for GRE v1 */ 488 if (tuple->src.u.gre.key || tuple->dst.u.gre.key) 489 return; 490 break; 491 } 492 #endif 493 default: 494 return; 495 } 496 497 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) || 498 ct->status & IPS_SEQ_ADJUST) 499 return; 500 501 tcf_ct_flow_table_add(ct_ft, ct, tcp, bidirectional); 502 } 503 504 static bool 505 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb, 506 struct flow_offload_tuple *tuple, 507 struct tcphdr **tcph) 508 { 509 struct flow_ports *ports; 510 unsigned int thoff; 511 struct iphdr *iph; 512 size_t hdrsize; 513 u8 ipproto; 514 515 if (!pskb_network_may_pull(skb, sizeof(*iph))) 516 return false; 517 518 iph = ip_hdr(skb); 519 thoff = iph->ihl * 4; 520 521 if (ip_is_fragment(iph) || 522 unlikely(thoff != sizeof(struct iphdr))) 523 return false; 524 525 ipproto = iph->protocol; 526 switch (ipproto) { 527 case IPPROTO_TCP: 528 hdrsize = sizeof(struct tcphdr); 529 break; 530 case IPPROTO_UDP: 531 hdrsize = sizeof(*ports); 532 break; 533 #ifdef CONFIG_NF_CT_PROTO_GRE 534 case IPPROTO_GRE: 535 hdrsize = sizeof(struct gre_base_hdr); 536 break; 537 #endif 538 default: 539 return false; 540 } 541 542 if (iph->ttl <= 1) 543 return false; 544 545 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 546 return false; 547 548 switch (ipproto) { 549 case IPPROTO_TCP: 550 *tcph = (void *)(skb_network_header(skb) + thoff); 551 fallthrough; 552 case IPPROTO_UDP: 553 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 554 tuple->src_port = ports->source; 555 tuple->dst_port = ports->dest; 556 break; 557 case IPPROTO_GRE: { 558 struct gre_base_hdr *greh; 559 560 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff); 561 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0) 562 return false; 563 break; 564 } 565 } 566 567 iph = ip_hdr(skb); 568 569 tuple->src_v4.s_addr = iph->saddr; 570 tuple->dst_v4.s_addr = iph->daddr; 571 tuple->l3proto = AF_INET; 572 tuple->l4proto = ipproto; 573 574 return true; 575 } 576 577 static bool 578 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb, 579 struct flow_offload_tuple *tuple, 580 struct tcphdr **tcph) 581 { 582 struct flow_ports *ports; 583 struct ipv6hdr *ip6h; 584 unsigned int thoff; 585 size_t hdrsize; 586 u8 nexthdr; 587 588 if (!pskb_network_may_pull(skb, sizeof(*ip6h))) 589 return false; 590 591 ip6h = ipv6_hdr(skb); 592 thoff = sizeof(*ip6h); 593 594 nexthdr = ip6h->nexthdr; 595 switch (nexthdr) { 596 case IPPROTO_TCP: 597 hdrsize = sizeof(struct tcphdr); 598 break; 599 case IPPROTO_UDP: 600 hdrsize = sizeof(*ports); 601 break; 602 #ifdef CONFIG_NF_CT_PROTO_GRE 603 case IPPROTO_GRE: 604 hdrsize = sizeof(struct gre_base_hdr); 605 break; 606 #endif 607 default: 608 return false; 609 } 610 611 if (ip6h->hop_limit <= 1) 612 return false; 613 614 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 615 return false; 616 617 switch (nexthdr) { 618 case IPPROTO_TCP: 619 *tcph = (void *)(skb_network_header(skb) + thoff); 620 fallthrough; 621 case IPPROTO_UDP: 622 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 623 tuple->src_port = ports->source; 624 tuple->dst_port = ports->dest; 625 break; 626 case IPPROTO_GRE: { 627 struct gre_base_hdr *greh; 628 629 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff); 630 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0) 631 return false; 632 break; 633 } 634 } 635 636 ip6h = ipv6_hdr(skb); 637 638 tuple->src_v6 = ip6h->saddr; 639 tuple->dst_v6 = ip6h->daddr; 640 tuple->l3proto = AF_INET6; 641 tuple->l4proto = nexthdr; 642 643 return true; 644 } 645 646 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p, 647 struct sk_buff *skb, 648 u8 family) 649 { 650 struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft; 651 struct flow_offload_tuple_rhash *tuplehash; 652 struct flow_offload_tuple tuple = {}; 653 enum ip_conntrack_info ctinfo; 654 struct tcphdr *tcph = NULL; 655 bool force_refresh = false; 656 struct flow_offload *flow; 657 struct nf_conn *ct; 658 u8 dir; 659 660 switch (family) { 661 case NFPROTO_IPV4: 662 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph)) 663 return false; 664 break; 665 case NFPROTO_IPV6: 666 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph)) 667 return false; 668 break; 669 default: 670 return false; 671 } 672 673 tuplehash = flow_offload_lookup(nf_ft, &tuple); 674 if (!tuplehash) 675 return false; 676 677 dir = tuplehash->tuple.dir; 678 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]); 679 ct = flow->ct; 680 681 if (dir == FLOW_OFFLOAD_DIR_REPLY && 682 !test_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags)) { 683 /* Only offload reply direction after connection became 684 * assured. 685 */ 686 if (test_bit(IPS_ASSURED_BIT, &ct->status)) 687 set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags); 688 else if (test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags)) 689 /* If flow_table flow has already been updated to the 690 * established state, then don't refresh. 691 */ 692 return false; 693 force_refresh = true; 694 } 695 696 if (tcph && (unlikely(tcph->fin || tcph->rst))) { 697 flow_offload_teardown(flow); 698 return false; 699 } 700 701 if (dir == FLOW_OFFLOAD_DIR_ORIGINAL) 702 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ? 703 IP_CT_ESTABLISHED : IP_CT_NEW; 704 else 705 ctinfo = IP_CT_ESTABLISHED_REPLY; 706 707 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 708 tcf_ct_flow_ct_ext_ifidx_update(flow); 709 flow_offload_refresh(nf_ft, flow, force_refresh); 710 if (!test_bit(IPS_ASSURED_BIT, &ct->status)) { 711 /* Process this flow in SW to allow promoting to ASSURED */ 712 return false; 713 } 714 715 nf_conntrack_get(&ct->ct_general); 716 nf_ct_set(skb, ct, ctinfo); 717 if (nf_ft->flags & NF_FLOWTABLE_COUNTER) 718 nf_ct_acct_update(ct, dir, skb->len); 719 720 return true; 721 } 722 723 static int tcf_ct_flow_tables_init(void) 724 { 725 return rhashtable_init(&zones_ht, &zones_params); 726 } 727 728 static void tcf_ct_flow_tables_uninit(void) 729 { 730 rhashtable_destroy(&zones_ht); 731 } 732 733 static struct tc_action_ops act_ct_ops; 734 735 struct tc_ct_action_net { 736 struct tc_action_net tn; /* Must be first */ 737 bool labels; 738 }; 739 740 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */ 741 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb, 742 struct tcf_ct_params *p) 743 { 744 enum ip_conntrack_info ctinfo; 745 struct nf_conn *ct; 746 747 ct = nf_ct_get(skb, &ctinfo); 748 if (!ct) 749 return false; 750 if (!net_eq(net, read_pnet(&ct->ct_net))) 751 goto drop_ct; 752 if (nf_ct_zone(ct)->id != p->zone) 753 goto drop_ct; 754 if (p->helper) { 755 struct nf_conn_help *help; 756 757 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); 758 if (help && rcu_access_pointer(help->helper) != p->helper) 759 goto drop_ct; 760 } 761 762 /* Force conntrack entry direction. */ 763 if ((p->ct_action & TCA_CT_ACT_FORCE) && 764 CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) { 765 if (nf_ct_is_confirmed(ct)) 766 nf_ct_kill(ct); 767 768 goto drop_ct; 769 } 770 771 return true; 772 773 drop_ct: 774 nf_ct_put(ct); 775 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 776 777 return false; 778 } 779 780 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb) 781 { 782 u8 family = NFPROTO_UNSPEC; 783 784 switch (skb_protocol(skb, true)) { 785 case htons(ETH_P_IP): 786 family = NFPROTO_IPV4; 787 break; 788 case htons(ETH_P_IPV6): 789 family = NFPROTO_IPV6; 790 break; 791 default: 792 break; 793 } 794 795 return family; 796 } 797 798 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag) 799 { 800 unsigned int len; 801 802 len = skb_network_offset(skb) + sizeof(struct iphdr); 803 if (unlikely(skb->len < len)) 804 return -EINVAL; 805 if (unlikely(!pskb_may_pull(skb, len))) 806 return -ENOMEM; 807 808 *frag = ip_is_fragment(ip_hdr(skb)); 809 return 0; 810 } 811 812 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag) 813 { 814 unsigned int flags = 0, len, payload_ofs = 0; 815 unsigned short frag_off; 816 int nexthdr; 817 818 len = skb_network_offset(skb) + sizeof(struct ipv6hdr); 819 if (unlikely(skb->len < len)) 820 return -EINVAL; 821 if (unlikely(!pskb_may_pull(skb, len))) 822 return -ENOMEM; 823 824 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags); 825 if (unlikely(nexthdr < 0)) 826 return -EPROTO; 827 828 *frag = flags & IP6_FH_F_FRAG; 829 return 0; 830 } 831 832 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb, 833 u8 family, u16 zone, bool *defrag) 834 { 835 enum ip_conntrack_info ctinfo; 836 struct nf_conn *ct; 837 int err = 0; 838 bool frag; 839 u8 proto; 840 u16 mru; 841 842 /* Previously seen (loopback)? Ignore. */ 843 ct = nf_ct_get(skb, &ctinfo); 844 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED) 845 return 0; 846 847 if (family == NFPROTO_IPV4) 848 err = tcf_ct_ipv4_is_fragment(skb, &frag); 849 else 850 err = tcf_ct_ipv6_is_fragment(skb, &frag); 851 if (err || !frag) 852 return err; 853 854 err = nf_ct_handle_fragments(net, skb, zone, family, &proto, &mru); 855 if (err) 856 return err; 857 858 *defrag = true; 859 tc_skb_cb(skb)->mru = mru; 860 861 return 0; 862 } 863 864 static void tcf_ct_params_free(struct tcf_ct_params *params) 865 { 866 if (params->helper) { 867 #if IS_ENABLED(CONFIG_NF_NAT) 868 if (params->ct_action & TCA_CT_ACT_NAT) 869 nf_nat_helper_put(params->helper); 870 #endif 871 nf_conntrack_helper_put(params->helper); 872 } 873 if (params->ct_ft) 874 tcf_ct_flow_table_put(params->ct_ft); 875 if (params->tmpl) 876 nf_ct_put(params->tmpl); 877 kfree(params); 878 } 879 880 static void tcf_ct_params_free_rcu(struct rcu_head *head) 881 { 882 struct tcf_ct_params *params; 883 884 params = container_of(head, struct tcf_ct_params, rcu); 885 tcf_ct_params_free(params); 886 } 887 888 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask) 889 { 890 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 891 u32 new_mark; 892 893 if (!mask) 894 return; 895 896 new_mark = mark | (READ_ONCE(ct->mark) & ~(mask)); 897 if (READ_ONCE(ct->mark) != new_mark) { 898 WRITE_ONCE(ct->mark, new_mark); 899 if (nf_ct_is_confirmed(ct)) 900 nf_conntrack_event_cache(IPCT_MARK, ct); 901 } 902 #endif 903 } 904 905 static void tcf_ct_act_set_labels(struct nf_conn *ct, 906 u32 *labels, 907 u32 *labels_m) 908 { 909 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) 910 size_t labels_sz = sizeof_field(struct tcf_ct_params, labels); 911 912 if (!memchr_inv(labels_m, 0, labels_sz)) 913 return; 914 915 nf_connlabels_replace(ct, labels, labels_m, 4); 916 #endif 917 } 918 919 static int tcf_ct_act_nat(struct sk_buff *skb, 920 struct nf_conn *ct, 921 enum ip_conntrack_info ctinfo, 922 int ct_action, 923 struct nf_nat_range2 *range, 924 bool commit) 925 { 926 #if IS_ENABLED(CONFIG_NF_NAT) 927 int err, action = 0; 928 929 if (!(ct_action & TCA_CT_ACT_NAT)) 930 return NF_ACCEPT; 931 if (ct_action & TCA_CT_ACT_NAT_SRC) 932 action |= BIT(NF_NAT_MANIP_SRC); 933 if (ct_action & TCA_CT_ACT_NAT_DST) 934 action |= BIT(NF_NAT_MANIP_DST); 935 936 err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit); 937 938 if (action & BIT(NF_NAT_MANIP_SRC)) 939 tc_skb_cb(skb)->post_ct_snat = 1; 940 if (action & BIT(NF_NAT_MANIP_DST)) 941 tc_skb_cb(skb)->post_ct_dnat = 1; 942 943 return err; 944 #else 945 return NF_ACCEPT; 946 #endif 947 } 948 949 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a, 950 struct tcf_result *res) 951 { 952 struct net *net = dev_net(skb->dev); 953 enum ip_conntrack_info ctinfo; 954 struct tcf_ct *c = to_ct(a); 955 struct nf_conn *tmpl = NULL; 956 struct nf_hook_state state; 957 bool cached, commit, clear; 958 int nh_ofs, err, retval; 959 struct tcf_ct_params *p; 960 bool add_helper = false; 961 bool skip_add = false; 962 bool defrag = false; 963 struct nf_conn *ct; 964 u8 family; 965 966 p = rcu_dereference_bh(c->params); 967 968 retval = READ_ONCE(c->tcf_action); 969 commit = p->ct_action & TCA_CT_ACT_COMMIT; 970 clear = p->ct_action & TCA_CT_ACT_CLEAR; 971 tmpl = p->tmpl; 972 973 tcf_lastuse_update(&c->tcf_tm); 974 tcf_action_update_bstats(&c->common, skb); 975 976 if (clear) { 977 tc_skb_cb(skb)->post_ct = false; 978 ct = nf_ct_get(skb, &ctinfo); 979 if (ct) { 980 nf_ct_put(ct); 981 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 982 } 983 984 goto out_clear; 985 } 986 987 family = tcf_ct_skb_nf_family(skb); 988 if (family == NFPROTO_UNSPEC) 989 goto drop; 990 991 /* The conntrack module expects to be working at L3. 992 * We also try to pull the IPv4/6 header to linear area 993 */ 994 nh_ofs = skb_network_offset(skb); 995 skb_pull_rcsum(skb, nh_ofs); 996 err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag); 997 if (err) 998 goto out_frag; 999 1000 err = nf_ct_skb_network_trim(skb, family); 1001 if (err) 1002 goto drop; 1003 1004 /* If we are recirculating packets to match on ct fields and 1005 * committing with a separate ct action, then we don't need to 1006 * actually run the packet through conntrack twice unless it's for a 1007 * different zone. 1008 */ 1009 cached = tcf_ct_skb_nfct_cached(net, skb, p); 1010 if (!cached) { 1011 if (tcf_ct_flow_table_lookup(p, skb, family)) { 1012 skip_add = true; 1013 goto do_nat; 1014 } 1015 1016 /* Associate skb with specified zone. */ 1017 if (tmpl) { 1018 nf_conntrack_put(skb_nfct(skb)); 1019 nf_conntrack_get(&tmpl->ct_general); 1020 nf_ct_set(skb, tmpl, IP_CT_NEW); 1021 } 1022 1023 state.hook = NF_INET_PRE_ROUTING; 1024 state.net = net; 1025 state.pf = family; 1026 err = nf_conntrack_in(skb, &state); 1027 if (err != NF_ACCEPT) 1028 goto out_push; 1029 } 1030 1031 do_nat: 1032 ct = nf_ct_get(skb, &ctinfo); 1033 if (!ct) 1034 goto out_push; 1035 nf_ct_deliver_cached_events(ct); 1036 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 1037 1038 err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit); 1039 if (err != NF_ACCEPT) 1040 goto drop; 1041 1042 if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) { 1043 err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC); 1044 if (err) 1045 goto drop; 1046 add_helper = true; 1047 if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) { 1048 if (!nfct_seqadj_ext_add(ct)) 1049 goto drop; 1050 } 1051 } 1052 1053 if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) { 1054 if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT) 1055 goto drop; 1056 } 1057 1058 if (commit) { 1059 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask); 1060 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask); 1061 1062 if (!nf_ct_is_confirmed(ct)) 1063 nf_conn_act_ct_ext_add(skb, ct, ctinfo); 1064 1065 /* This will take care of sending queued events 1066 * even if the connection is already confirmed. 1067 */ 1068 if (nf_conntrack_confirm(skb) != NF_ACCEPT) 1069 goto drop; 1070 } 1071 1072 if (!skip_add) 1073 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo); 1074 1075 out_push: 1076 skb_push_rcsum(skb, nh_ofs); 1077 1078 tc_skb_cb(skb)->post_ct = true; 1079 tc_skb_cb(skb)->zone = p->zone; 1080 out_clear: 1081 if (defrag) 1082 qdisc_skb_cb(skb)->pkt_len = skb->len; 1083 return retval; 1084 1085 out_frag: 1086 if (err != -EINPROGRESS) 1087 tcf_action_inc_drop_qstats(&c->common); 1088 return TC_ACT_CONSUMED; 1089 1090 drop: 1091 tcf_action_inc_drop_qstats(&c->common); 1092 return TC_ACT_SHOT; 1093 } 1094 1095 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = { 1096 [TCA_CT_ACTION] = { .type = NLA_U16 }, 1097 [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)), 1098 [TCA_CT_ZONE] = { .type = NLA_U16 }, 1099 [TCA_CT_MARK] = { .type = NLA_U32 }, 1100 [TCA_CT_MARK_MASK] = { .type = NLA_U32 }, 1101 [TCA_CT_LABELS] = { .type = NLA_BINARY, 1102 .len = 128 / BITS_PER_BYTE }, 1103 [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY, 1104 .len = 128 / BITS_PER_BYTE }, 1105 [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 }, 1106 [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 }, 1107 [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1108 [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1109 [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 }, 1110 [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 }, 1111 [TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN }, 1112 [TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 }, 1113 [TCA_CT_HELPER_PROTO] = { .type = NLA_U8 }, 1114 }; 1115 1116 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p, 1117 struct tc_ct *parm, 1118 struct nlattr **tb, 1119 struct netlink_ext_ack *extack) 1120 { 1121 struct nf_nat_range2 *range; 1122 1123 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1124 return 0; 1125 1126 if (!IS_ENABLED(CONFIG_NF_NAT)) { 1127 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel"); 1128 return -EOPNOTSUPP; 1129 } 1130 1131 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1132 return 0; 1133 1134 if ((p->ct_action & TCA_CT_ACT_NAT_SRC) && 1135 (p->ct_action & TCA_CT_ACT_NAT_DST)) { 1136 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time"); 1137 return -EOPNOTSUPP; 1138 } 1139 1140 range = &p->range; 1141 if (tb[TCA_CT_NAT_IPV4_MIN]) { 1142 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX]; 1143 1144 p->ipv4_range = true; 1145 range->flags |= NF_NAT_RANGE_MAP_IPS; 1146 range->min_addr.ip = 1147 nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]); 1148 1149 range->max_addr.ip = max_attr ? 1150 nla_get_in_addr(max_attr) : 1151 range->min_addr.ip; 1152 } else if (tb[TCA_CT_NAT_IPV6_MIN]) { 1153 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX]; 1154 1155 p->ipv4_range = false; 1156 range->flags |= NF_NAT_RANGE_MAP_IPS; 1157 range->min_addr.in6 = 1158 nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]); 1159 1160 range->max_addr.in6 = max_attr ? 1161 nla_get_in6_addr(max_attr) : 1162 range->min_addr.in6; 1163 } 1164 1165 if (tb[TCA_CT_NAT_PORT_MIN]) { 1166 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED; 1167 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]); 1168 1169 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ? 1170 nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) : 1171 range->min_proto.all; 1172 } 1173 1174 return 0; 1175 } 1176 1177 static void tcf_ct_set_key_val(struct nlattr **tb, 1178 void *val, int val_type, 1179 void *mask, int mask_type, 1180 int len) 1181 { 1182 if (!tb[val_type]) 1183 return; 1184 nla_memcpy(val, tb[val_type], len); 1185 1186 if (!mask) 1187 return; 1188 1189 if (mask_type == TCA_CT_UNSPEC || !tb[mask_type]) 1190 memset(mask, 0xff, len); 1191 else 1192 nla_memcpy(mask, tb[mask_type], len); 1193 } 1194 1195 static int tcf_ct_fill_params(struct net *net, 1196 struct tcf_ct_params *p, 1197 struct tc_ct *parm, 1198 struct nlattr **tb, 1199 struct netlink_ext_ack *extack) 1200 { 1201 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1202 struct nf_conntrack_zone zone; 1203 int err, family, proto, len; 1204 struct nf_conn *tmpl; 1205 char *name; 1206 1207 p->zone = NF_CT_DEFAULT_ZONE_ID; 1208 1209 tcf_ct_set_key_val(tb, 1210 &p->ct_action, TCA_CT_ACTION, 1211 NULL, TCA_CT_UNSPEC, 1212 sizeof(p->ct_action)); 1213 1214 if (p->ct_action & TCA_CT_ACT_CLEAR) 1215 return 0; 1216 1217 err = tcf_ct_fill_params_nat(p, parm, tb, extack); 1218 if (err) 1219 return err; 1220 1221 if (tb[TCA_CT_MARK]) { 1222 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) { 1223 NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled."); 1224 return -EOPNOTSUPP; 1225 } 1226 tcf_ct_set_key_val(tb, 1227 &p->mark, TCA_CT_MARK, 1228 &p->mark_mask, TCA_CT_MARK_MASK, 1229 sizeof(p->mark)); 1230 } 1231 1232 if (tb[TCA_CT_LABELS]) { 1233 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) { 1234 NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled."); 1235 return -EOPNOTSUPP; 1236 } 1237 1238 if (!tn->labels) { 1239 NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length"); 1240 return -EOPNOTSUPP; 1241 } 1242 tcf_ct_set_key_val(tb, 1243 p->labels, TCA_CT_LABELS, 1244 p->labels_mask, TCA_CT_LABELS_MASK, 1245 sizeof(p->labels)); 1246 } 1247 1248 if (tb[TCA_CT_ZONE]) { 1249 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) { 1250 NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled."); 1251 return -EOPNOTSUPP; 1252 } 1253 1254 tcf_ct_set_key_val(tb, 1255 &p->zone, TCA_CT_ZONE, 1256 NULL, TCA_CT_UNSPEC, 1257 sizeof(p->zone)); 1258 } 1259 1260 nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0); 1261 tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL); 1262 if (!tmpl) { 1263 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template"); 1264 return -ENOMEM; 1265 } 1266 p->tmpl = tmpl; 1267 if (tb[TCA_CT_HELPER_NAME]) { 1268 name = nla_data(tb[TCA_CT_HELPER_NAME]); 1269 len = nla_len(tb[TCA_CT_HELPER_NAME]); 1270 if (len > 16 || name[len - 1] != '\0') { 1271 NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name."); 1272 err = -EINVAL; 1273 goto err; 1274 } 1275 family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(tb[TCA_CT_HELPER_FAMILY]) : AF_INET; 1276 proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP; 1277 err = nf_ct_add_helper(tmpl, name, family, proto, 1278 p->ct_action & TCA_CT_ACT_NAT, &p->helper); 1279 if (err) { 1280 NL_SET_ERR_MSG_MOD(extack, "Failed to add helper"); 1281 goto err; 1282 } 1283 } 1284 1285 if (p->ct_action & TCA_CT_ACT_COMMIT) 1286 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status); 1287 return 0; 1288 err: 1289 nf_ct_put(p->tmpl); 1290 p->tmpl = NULL; 1291 return err; 1292 } 1293 1294 static int tcf_ct_init(struct net *net, struct nlattr *nla, 1295 struct nlattr *est, struct tc_action **a, 1296 struct tcf_proto *tp, u32 flags, 1297 struct netlink_ext_ack *extack) 1298 { 1299 struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id); 1300 bool bind = flags & TCA_ACT_FLAGS_BIND; 1301 struct tcf_ct_params *params = NULL; 1302 struct nlattr *tb[TCA_CT_MAX + 1]; 1303 struct tcf_chain *goto_ch = NULL; 1304 struct tc_ct *parm; 1305 struct tcf_ct *c; 1306 int err, res = 0; 1307 u32 index; 1308 1309 if (!nla) { 1310 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed"); 1311 return -EINVAL; 1312 } 1313 1314 err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack); 1315 if (err < 0) 1316 return err; 1317 1318 if (!tb[TCA_CT_PARMS]) { 1319 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters"); 1320 return -EINVAL; 1321 } 1322 parm = nla_data(tb[TCA_CT_PARMS]); 1323 index = parm->index; 1324 err = tcf_idr_check_alloc(tn, &index, a, bind); 1325 if (err < 0) 1326 return err; 1327 1328 if (!err) { 1329 err = tcf_idr_create_from_flags(tn, index, est, a, 1330 &act_ct_ops, bind, flags); 1331 if (err) { 1332 tcf_idr_cleanup(tn, index); 1333 return err; 1334 } 1335 res = ACT_P_CREATED; 1336 } else { 1337 if (bind) 1338 return 0; 1339 1340 if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 1341 tcf_idr_release(*a, bind); 1342 return -EEXIST; 1343 } 1344 } 1345 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 1346 if (err < 0) 1347 goto cleanup; 1348 1349 c = to_ct(*a); 1350 1351 params = kzalloc(sizeof(*params), GFP_KERNEL); 1352 if (unlikely(!params)) { 1353 err = -ENOMEM; 1354 goto cleanup; 1355 } 1356 1357 err = tcf_ct_fill_params(net, params, parm, tb, extack); 1358 if (err) 1359 goto cleanup; 1360 1361 err = tcf_ct_flow_table_get(net, params); 1362 if (err) 1363 goto cleanup; 1364 1365 spin_lock_bh(&c->tcf_lock); 1366 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 1367 params = rcu_replace_pointer(c->params, params, 1368 lockdep_is_held(&c->tcf_lock)); 1369 spin_unlock_bh(&c->tcf_lock); 1370 1371 if (goto_ch) 1372 tcf_chain_put_by_act(goto_ch); 1373 if (params) 1374 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1375 1376 return res; 1377 1378 cleanup: 1379 if (goto_ch) 1380 tcf_chain_put_by_act(goto_ch); 1381 if (params) 1382 tcf_ct_params_free(params); 1383 tcf_idr_release(*a, bind); 1384 return err; 1385 } 1386 1387 static void tcf_ct_cleanup(struct tc_action *a) 1388 { 1389 struct tcf_ct_params *params; 1390 struct tcf_ct *c = to_ct(a); 1391 1392 params = rcu_dereference_protected(c->params, 1); 1393 if (params) 1394 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1395 } 1396 1397 static int tcf_ct_dump_key_val(struct sk_buff *skb, 1398 void *val, int val_type, 1399 void *mask, int mask_type, 1400 int len) 1401 { 1402 int err; 1403 1404 if (mask && !memchr_inv(mask, 0, len)) 1405 return 0; 1406 1407 err = nla_put(skb, val_type, len, val); 1408 if (err) 1409 return err; 1410 1411 if (mask_type != TCA_CT_UNSPEC) { 1412 err = nla_put(skb, mask_type, len, mask); 1413 if (err) 1414 return err; 1415 } 1416 1417 return 0; 1418 } 1419 1420 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p) 1421 { 1422 struct nf_nat_range2 *range = &p->range; 1423 1424 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1425 return 0; 1426 1427 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1428 return 0; 1429 1430 if (range->flags & NF_NAT_RANGE_MAP_IPS) { 1431 if (p->ipv4_range) { 1432 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN, 1433 range->min_addr.ip)) 1434 return -1; 1435 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX, 1436 range->max_addr.ip)) 1437 return -1; 1438 } else { 1439 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN, 1440 &range->min_addr.in6)) 1441 return -1; 1442 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX, 1443 &range->max_addr.in6)) 1444 return -1; 1445 } 1446 } 1447 1448 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) { 1449 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN, 1450 range->min_proto.all)) 1451 return -1; 1452 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX, 1453 range->max_proto.all)) 1454 return -1; 1455 } 1456 1457 return 0; 1458 } 1459 1460 static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper) 1461 { 1462 if (!helper) 1463 return 0; 1464 1465 if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) || 1466 nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) || 1467 nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum)) 1468 return -1; 1469 1470 return 0; 1471 } 1472 1473 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a, 1474 int bind, int ref) 1475 { 1476 unsigned char *b = skb_tail_pointer(skb); 1477 struct tcf_ct *c = to_ct(a); 1478 struct tcf_ct_params *p; 1479 1480 struct tc_ct opt = { 1481 .index = c->tcf_index, 1482 .refcnt = refcount_read(&c->tcf_refcnt) - ref, 1483 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind, 1484 }; 1485 struct tcf_t t; 1486 1487 spin_lock_bh(&c->tcf_lock); 1488 p = rcu_dereference_protected(c->params, 1489 lockdep_is_held(&c->tcf_lock)); 1490 opt.action = c->tcf_action; 1491 1492 if (tcf_ct_dump_key_val(skb, 1493 &p->ct_action, TCA_CT_ACTION, 1494 NULL, TCA_CT_UNSPEC, 1495 sizeof(p->ct_action))) 1496 goto nla_put_failure; 1497 1498 if (p->ct_action & TCA_CT_ACT_CLEAR) 1499 goto skip_dump; 1500 1501 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && 1502 tcf_ct_dump_key_val(skb, 1503 &p->mark, TCA_CT_MARK, 1504 &p->mark_mask, TCA_CT_MARK_MASK, 1505 sizeof(p->mark))) 1506 goto nla_put_failure; 1507 1508 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1509 tcf_ct_dump_key_val(skb, 1510 p->labels, TCA_CT_LABELS, 1511 p->labels_mask, TCA_CT_LABELS_MASK, 1512 sizeof(p->labels))) 1513 goto nla_put_failure; 1514 1515 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 1516 tcf_ct_dump_key_val(skb, 1517 &p->zone, TCA_CT_ZONE, 1518 NULL, TCA_CT_UNSPEC, 1519 sizeof(p->zone))) 1520 goto nla_put_failure; 1521 1522 if (tcf_ct_dump_nat(skb, p)) 1523 goto nla_put_failure; 1524 1525 if (tcf_ct_dump_helper(skb, p->helper)) 1526 goto nla_put_failure; 1527 1528 skip_dump: 1529 if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt)) 1530 goto nla_put_failure; 1531 1532 tcf_tm_dump(&t, &c->tcf_tm); 1533 if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD)) 1534 goto nla_put_failure; 1535 spin_unlock_bh(&c->tcf_lock); 1536 1537 return skb->len; 1538 nla_put_failure: 1539 spin_unlock_bh(&c->tcf_lock); 1540 nlmsg_trim(skb, b); 1541 return -1; 1542 } 1543 1544 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 1545 u64 drops, u64 lastuse, bool hw) 1546 { 1547 struct tcf_ct *c = to_ct(a); 1548 1549 tcf_action_update_stats(a, bytes, packets, drops, hw); 1550 c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse); 1551 } 1552 1553 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data, 1554 u32 *index_inc, bool bind, 1555 struct netlink_ext_ack *extack) 1556 { 1557 if (bind) { 1558 struct flow_action_entry *entry = entry_data; 1559 1560 if (tcf_ct_helper(act)) 1561 return -EOPNOTSUPP; 1562 1563 entry->id = FLOW_ACTION_CT; 1564 entry->ct.action = tcf_ct_action(act); 1565 entry->ct.zone = tcf_ct_zone(act); 1566 entry->ct.flow_table = tcf_ct_ft(act); 1567 *index_inc = 1; 1568 } else { 1569 struct flow_offload_action *fl_action = entry_data; 1570 1571 fl_action->id = FLOW_ACTION_CT; 1572 } 1573 1574 return 0; 1575 } 1576 1577 static struct tc_action_ops act_ct_ops = { 1578 .kind = "ct", 1579 .id = TCA_ID_CT, 1580 .owner = THIS_MODULE, 1581 .act = tcf_ct_act, 1582 .dump = tcf_ct_dump, 1583 .init = tcf_ct_init, 1584 .cleanup = tcf_ct_cleanup, 1585 .stats_update = tcf_stats_update, 1586 .offload_act_setup = tcf_ct_offload_act_setup, 1587 .size = sizeof(struct tcf_ct), 1588 }; 1589 1590 static __net_init int ct_init_net(struct net *net) 1591 { 1592 unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8; 1593 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1594 1595 if (nf_connlabels_get(net, n_bits - 1)) { 1596 tn->labels = false; 1597 pr_err("act_ct: Failed to set connlabels length"); 1598 } else { 1599 tn->labels = true; 1600 } 1601 1602 return tc_action_net_init(net, &tn->tn, &act_ct_ops); 1603 } 1604 1605 static void __net_exit ct_exit_net(struct list_head *net_list) 1606 { 1607 struct net *net; 1608 1609 rtnl_lock(); 1610 list_for_each_entry(net, net_list, exit_list) { 1611 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1612 1613 if (tn->labels) 1614 nf_connlabels_put(net); 1615 } 1616 rtnl_unlock(); 1617 1618 tc_action_net_exit(net_list, act_ct_ops.net_id); 1619 } 1620 1621 static struct pernet_operations ct_net_ops = { 1622 .init = ct_init_net, 1623 .exit_batch = ct_exit_net, 1624 .id = &act_ct_ops.net_id, 1625 .size = sizeof(struct tc_ct_action_net), 1626 }; 1627 1628 static int __init ct_init_module(void) 1629 { 1630 int err; 1631 1632 act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0); 1633 if (!act_ct_wq) 1634 return -ENOMEM; 1635 1636 err = tcf_ct_flow_tables_init(); 1637 if (err) 1638 goto err_tbl_init; 1639 1640 err = tcf_register_action(&act_ct_ops, &ct_net_ops); 1641 if (err) 1642 goto err_register; 1643 1644 static_branch_inc(&tcf_frag_xmit_count); 1645 1646 return 0; 1647 1648 err_register: 1649 tcf_ct_flow_tables_uninit(); 1650 err_tbl_init: 1651 destroy_workqueue(act_ct_wq); 1652 return err; 1653 } 1654 1655 static void __exit ct_cleanup_module(void) 1656 { 1657 static_branch_dec(&tcf_frag_xmit_count); 1658 tcf_unregister_action(&act_ct_ops, &ct_net_ops); 1659 tcf_ct_flow_tables_uninit(); 1660 destroy_workqueue(act_ct_wq); 1661 } 1662 1663 module_init(ct_init_module); 1664 module_exit(ct_cleanup_module); 1665 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>"); 1666 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>"); 1667 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>"); 1668 MODULE_DESCRIPTION("Connection tracking action"); 1669 MODULE_LICENSE("GPL v2"); 1670