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/if_tunnel.h> 17 #include <linux/ip.h> 18 #include <linux/ipv6.h> 19 #include <linux/rhashtable.h> 20 #include <net/gre.h> 21 #include <net/netlink.h> 22 #include <net/pkt_sched.h> 23 #include <net/pkt_cls.h> 24 #include <net/act_api.h> 25 #include <net/ip.h> 26 #include <net/ipv6_frag.h> 27 #include <uapi/linux/tc_act/tc_ct.h> 28 #include <net/tc_act/tc_ct.h> 29 #include <net/tc_wrapper.h> 30 31 #include <net/netfilter/nf_flow_table.h> 32 #include <net/netfilter/nf_conntrack.h> 33 #include <net/netfilter/nf_conntrack_core.h> 34 #include <net/netfilter/nf_conntrack_zones.h> 35 #include <net/netfilter/nf_conntrack_helper.h> 36 #include <net/netfilter/nf_conntrack_acct.h> 37 #include <net/netfilter/ipv6/nf_defrag_ipv6.h> 38 #include <net/netfilter/nf_conntrack_act_ct.h> 39 #include <net/netfilter/nf_conntrack_seqadj.h> 40 #include <uapi/linux/netfilter/nf_nat.h> 41 42 static struct workqueue_struct *act_ct_wq; 43 static struct rhashtable zones_ht; 44 static DEFINE_MUTEX(zones_mutex); 45 46 struct zones_ht_key { 47 struct net *net; 48 u16 zone; 49 }; 50 51 struct tcf_ct_flow_table { 52 struct rhash_head node; /* In zones tables */ 53 54 struct rcu_work rwork; 55 struct nf_flowtable nf_ft; 56 refcount_t ref; 57 struct zones_ht_key key; 58 59 bool dying; 60 }; 61 62 static const struct rhashtable_params zones_params = { 63 .head_offset = offsetof(struct tcf_ct_flow_table, node), 64 .key_offset = offsetof(struct tcf_ct_flow_table, key), 65 .key_len = offsetofend(struct zones_ht_key, zone), 66 .automatic_shrinking = true, 67 }; 68 69 static struct flow_action_entry * 70 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action) 71 { 72 int i = flow_action->num_entries++; 73 74 return &flow_action->entries[i]; 75 } 76 77 static void tcf_ct_add_mangle_action(struct flow_action *action, 78 enum flow_action_mangle_base htype, 79 u32 offset, 80 u32 mask, 81 u32 val) 82 { 83 struct flow_action_entry *entry; 84 85 entry = tcf_ct_flow_table_flow_action_get_next(action); 86 entry->id = FLOW_ACTION_MANGLE; 87 entry->mangle.htype = htype; 88 entry->mangle.mask = ~mask; 89 entry->mangle.offset = offset; 90 entry->mangle.val = val; 91 } 92 93 /* The following nat helper functions check if the inverted reverse tuple 94 * (target) is different then the current dir tuple - meaning nat for ports 95 * and/or ip is needed, and add the relevant mangle actions. 96 */ 97 static void 98 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple, 99 struct nf_conntrack_tuple target, 100 struct flow_action *action) 101 { 102 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 103 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 104 offsetof(struct iphdr, saddr), 105 0xFFFFFFFF, 106 be32_to_cpu(target.src.u3.ip)); 107 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 108 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 109 offsetof(struct iphdr, daddr), 110 0xFFFFFFFF, 111 be32_to_cpu(target.dst.u3.ip)); 112 } 113 114 static void 115 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action, 116 union nf_inet_addr *addr, 117 u32 offset) 118 { 119 int i; 120 121 for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++) 122 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6, 123 i * sizeof(u32) + offset, 124 0xFFFFFFFF, be32_to_cpu(addr->ip6[i])); 125 } 126 127 static void 128 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple, 129 struct nf_conntrack_tuple target, 130 struct flow_action *action) 131 { 132 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 133 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3, 134 offsetof(struct ipv6hdr, 135 saddr)); 136 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 137 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3, 138 offsetof(struct ipv6hdr, 139 daddr)); 140 } 141 142 static void 143 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple, 144 struct nf_conntrack_tuple target, 145 struct flow_action *action) 146 { 147 __be16 target_src = target.src.u.tcp.port; 148 __be16 target_dst = target.dst.u.tcp.port; 149 150 if (target_src != tuple->src.u.tcp.port) 151 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 152 offsetof(struct tcphdr, source), 153 0xFFFF, be16_to_cpu(target_src)); 154 if (target_dst != tuple->dst.u.tcp.port) 155 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 156 offsetof(struct tcphdr, dest), 157 0xFFFF, be16_to_cpu(target_dst)); 158 } 159 160 static void 161 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple, 162 struct nf_conntrack_tuple target, 163 struct flow_action *action) 164 { 165 __be16 target_src = target.src.u.udp.port; 166 __be16 target_dst = target.dst.u.udp.port; 167 168 if (target_src != tuple->src.u.udp.port) 169 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 170 offsetof(struct udphdr, source), 171 0xFFFF, be16_to_cpu(target_src)); 172 if (target_dst != tuple->dst.u.udp.port) 173 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 174 offsetof(struct udphdr, dest), 175 0xFFFF, be16_to_cpu(target_dst)); 176 } 177 178 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct, 179 enum ip_conntrack_dir dir, 180 enum ip_conntrack_info ctinfo, 181 struct flow_action *action) 182 { 183 struct nf_conn_labels *ct_labels; 184 struct flow_action_entry *entry; 185 u32 *act_ct_labels; 186 187 entry = tcf_ct_flow_table_flow_action_get_next(action); 188 entry->id = FLOW_ACTION_CT_METADATA; 189 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 190 entry->ct_metadata.mark = READ_ONCE(ct->mark); 191 #endif 192 /* aligns with the CT reference on the SKB nf_ct_set */ 193 entry->ct_metadata.cookie = (unsigned long)ct | ctinfo; 194 entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL; 195 196 act_ct_labels = entry->ct_metadata.labels; 197 ct_labels = nf_ct_labels_find(ct); 198 if (ct_labels) 199 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE); 200 else 201 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE); 202 } 203 204 static int tcf_ct_flow_table_add_action_nat(struct net *net, 205 struct nf_conn *ct, 206 enum ip_conntrack_dir dir, 207 struct flow_action *action) 208 { 209 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple; 210 struct nf_conntrack_tuple target; 211 212 if (!(ct->status & IPS_NAT_MASK)) 213 return 0; 214 215 nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple); 216 217 switch (tuple->src.l3num) { 218 case NFPROTO_IPV4: 219 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target, 220 action); 221 break; 222 case NFPROTO_IPV6: 223 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target, 224 action); 225 break; 226 default: 227 return -EOPNOTSUPP; 228 } 229 230 switch (nf_ct_protonum(ct)) { 231 case IPPROTO_TCP: 232 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action); 233 break; 234 case IPPROTO_UDP: 235 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action); 236 break; 237 default: 238 return -EOPNOTSUPP; 239 } 240 241 return 0; 242 } 243 244 static int tcf_ct_flow_table_fill_actions(struct net *net, 245 struct flow_offload *flow, 246 enum flow_offload_tuple_dir tdir, 247 struct nf_flow_rule *flow_rule) 248 { 249 struct flow_action *action = &flow_rule->rule->action; 250 int num_entries = action->num_entries; 251 struct nf_conn *ct = flow->ct; 252 enum ip_conntrack_info ctinfo; 253 enum ip_conntrack_dir dir; 254 int i, err; 255 256 switch (tdir) { 257 case FLOW_OFFLOAD_DIR_ORIGINAL: 258 dir = IP_CT_DIR_ORIGINAL; 259 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ? 260 IP_CT_ESTABLISHED : IP_CT_NEW; 261 if (ctinfo == IP_CT_ESTABLISHED) 262 set_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags); 263 break; 264 case FLOW_OFFLOAD_DIR_REPLY: 265 dir = IP_CT_DIR_REPLY; 266 ctinfo = IP_CT_ESTABLISHED_REPLY; 267 break; 268 default: 269 return -EOPNOTSUPP; 270 } 271 272 err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action); 273 if (err) 274 goto err_nat; 275 276 tcf_ct_flow_table_add_action_meta(ct, dir, ctinfo, action); 277 return 0; 278 279 err_nat: 280 /* Clear filled actions */ 281 for (i = num_entries; i < action->num_entries; i++) 282 memset(&action->entries[i], 0, sizeof(action->entries[i])); 283 action->num_entries = num_entries; 284 285 return err; 286 } 287 288 static bool tcf_ct_flow_is_outdated(const struct flow_offload *flow) 289 { 290 return test_bit(IPS_SEEN_REPLY_BIT, &flow->ct->status) && 291 test_bit(IPS_HW_OFFLOAD_BIT, &flow->ct->status) && 292 !test_bit(NF_FLOW_HW_PENDING, &flow->flags) && 293 !test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags); 294 } 295 296 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft); 297 298 static void tcf_ct_nf_get(struct nf_flowtable *ft) 299 { 300 struct tcf_ct_flow_table *ct_ft = 301 container_of(ft, struct tcf_ct_flow_table, nf_ft); 302 303 tcf_ct_flow_table_get_ref(ct_ft); 304 } 305 306 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft); 307 308 static void tcf_ct_nf_put(struct nf_flowtable *ft) 309 { 310 struct tcf_ct_flow_table *ct_ft = 311 container_of(ft, struct tcf_ct_flow_table, nf_ft); 312 313 tcf_ct_flow_table_put(ct_ft); 314 } 315 316 static struct nf_flowtable_type flowtable_ct = { 317 .gc = tcf_ct_flow_is_outdated, 318 .action = tcf_ct_flow_table_fill_actions, 319 .get = tcf_ct_nf_get, 320 .put = tcf_ct_nf_put, 321 .owner = THIS_MODULE, 322 }; 323 324 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params) 325 { 326 struct zones_ht_key key = { .net = net, .zone = params->zone }; 327 struct tcf_ct_flow_table *ct_ft; 328 int err = -ENOMEM; 329 330 mutex_lock(&zones_mutex); 331 rcu_read_lock(); 332 ct_ft = rhashtable_lookup(&zones_ht, &key, zones_params); 333 if (ct_ft && refcount_inc_not_zero(&ct_ft->ref)) { 334 rcu_read_unlock(); 335 goto out_unlock; 336 } 337 rcu_read_unlock(); 338 339 ct_ft = kzalloc_obj(*ct_ft); 340 if (!ct_ft) 341 goto err_alloc; 342 refcount_set(&ct_ft->ref, 1); 343 344 ct_ft->key = key; 345 err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params); 346 if (err) 347 goto err_insert; 348 349 ct_ft->nf_ft.type = &flowtable_ct; 350 ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD | 351 NF_FLOWTABLE_COUNTER; 352 err = nf_flow_table_init(&ct_ft->nf_ft); 353 if (err) 354 goto err_init; 355 write_pnet(&ct_ft->nf_ft.net, net); 356 357 __module_get(THIS_MODULE); 358 out_unlock: 359 params->ct_ft = ct_ft; 360 params->nf_ft = &ct_ft->nf_ft; 361 mutex_unlock(&zones_mutex); 362 363 return 0; 364 365 err_init: 366 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 367 err_insert: 368 kfree(ct_ft); 369 err_alloc: 370 mutex_unlock(&zones_mutex); 371 return err; 372 } 373 374 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft) 375 { 376 refcount_inc(&ct_ft->ref); 377 } 378 379 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work) 380 { 381 struct tcf_ct_flow_table *ct_ft; 382 struct flow_block *block; 383 384 ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table, 385 rwork); 386 nf_flow_table_free(&ct_ft->nf_ft); 387 388 block = &ct_ft->nf_ft.flow_block; 389 down_write(&ct_ft->nf_ft.flow_block_lock); 390 WARN_ON(!list_empty(&block->cb_list)); 391 up_write(&ct_ft->nf_ft.flow_block_lock); 392 kfree(ct_ft); 393 394 module_put(THIS_MODULE); 395 } 396 397 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft) 398 { 399 if (refcount_dec_and_test(&ct_ft->ref)) { 400 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 401 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work); 402 queue_rcu_work(act_ct_wq, &ct_ft->rwork); 403 } 404 } 405 406 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry, 407 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir) 408 { 409 entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC; 410 entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir]; 411 } 412 413 static void tcf_ct_flow_ct_ext_ifidx_update(struct flow_offload *entry) 414 { 415 struct nf_conn_act_ct_ext *act_ct_ext; 416 417 act_ct_ext = nf_conn_act_ct_ext_find(entry->ct); 418 if (act_ct_ext) { 419 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL); 420 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY); 421 } 422 } 423 424 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft, 425 struct nf_conn *ct, 426 bool tcp, bool bidirectional) 427 { 428 struct nf_conn_act_ct_ext *act_ct_ext; 429 struct flow_offload *entry; 430 int err; 431 432 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status)) 433 return; 434 435 entry = flow_offload_alloc(ct); 436 if (!entry) { 437 WARN_ON_ONCE(1); 438 goto err_alloc; 439 } 440 441 if (tcp) { 442 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 443 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 444 } 445 if (bidirectional) 446 __set_bit(NF_FLOW_HW_BIDIRECTIONAL, &entry->flags); 447 448 act_ct_ext = nf_conn_act_ct_ext_find(ct); 449 if (act_ct_ext) { 450 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL); 451 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY); 452 } 453 454 err = flow_offload_add(&ct_ft->nf_ft, entry); 455 if (err) 456 goto err_add; 457 458 return; 459 460 err_add: 461 flow_offload_free(entry); 462 err_alloc: 463 clear_bit(IPS_OFFLOAD_BIT, &ct->status); 464 } 465 466 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft, 467 struct nf_conn *ct, 468 enum ip_conntrack_info ctinfo) 469 { 470 bool tcp = false, bidirectional = true; 471 472 switch (nf_ct_protonum(ct)) { 473 case IPPROTO_TCP: 474 if ((ctinfo != IP_CT_ESTABLISHED && 475 ctinfo != IP_CT_ESTABLISHED_REPLY) || 476 !test_bit(IPS_ASSURED_BIT, &ct->status) || 477 ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED) 478 return; 479 480 tcp = true; 481 break; 482 case IPPROTO_UDP: 483 if (!nf_ct_is_confirmed(ct)) 484 return; 485 if (!test_bit(IPS_ASSURED_BIT, &ct->status)) 486 bidirectional = false; 487 break; 488 #ifdef CONFIG_NF_CT_PROTO_GRE 489 case IPPROTO_GRE: { 490 struct nf_conntrack_tuple *tuple; 491 492 if ((ctinfo != IP_CT_ESTABLISHED && 493 ctinfo != IP_CT_ESTABLISHED_REPLY) || 494 !test_bit(IPS_ASSURED_BIT, &ct->status) || 495 ct->status & IPS_NAT_MASK) 496 return; 497 498 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; 499 /* No support for GRE v1 */ 500 if (tuple->src.u.gre.key || tuple->dst.u.gre.key) 501 return; 502 break; 503 } 504 #endif 505 default: 506 return; 507 } 508 509 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) || 510 ct->status & IPS_SEQ_ADJUST) 511 return; 512 513 tcf_ct_flow_table_add(ct_ft, ct, tcp, bidirectional); 514 } 515 516 static bool 517 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb, 518 struct flow_offload_tuple *tuple, 519 struct tcphdr **tcph) 520 { 521 struct flow_ports *ports; 522 unsigned int thoff; 523 struct iphdr *iph; 524 size_t hdrsize; 525 u8 ipproto; 526 527 if (!pskb_network_may_pull(skb, sizeof(*iph))) 528 return false; 529 530 iph = ip_hdr(skb); 531 thoff = iph->ihl * 4; 532 533 if (ip_is_fragment(iph) || 534 unlikely(thoff != sizeof(struct iphdr))) 535 return false; 536 537 ipproto = iph->protocol; 538 switch (ipproto) { 539 case IPPROTO_TCP: 540 hdrsize = sizeof(struct tcphdr); 541 break; 542 case IPPROTO_UDP: 543 hdrsize = sizeof(*ports); 544 break; 545 #ifdef CONFIG_NF_CT_PROTO_GRE 546 case IPPROTO_GRE: 547 hdrsize = sizeof(struct gre_base_hdr); 548 break; 549 #endif 550 default: 551 return false; 552 } 553 554 if (iph->ttl <= 1) 555 return false; 556 557 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 558 return false; 559 560 switch (ipproto) { 561 case IPPROTO_TCP: 562 *tcph = (void *)(skb_network_header(skb) + thoff); 563 fallthrough; 564 case IPPROTO_UDP: 565 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 566 tuple->src_port = ports->source; 567 tuple->dst_port = ports->dest; 568 break; 569 case IPPROTO_GRE: { 570 struct gre_base_hdr *greh; 571 572 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff); 573 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0) 574 return false; 575 break; 576 } 577 } 578 579 iph = ip_hdr(skb); 580 581 tuple->src_v4.s_addr = iph->saddr; 582 tuple->dst_v4.s_addr = iph->daddr; 583 tuple->l3proto = AF_INET; 584 tuple->l4proto = ipproto; 585 586 return true; 587 } 588 589 static bool 590 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb, 591 struct flow_offload_tuple *tuple, 592 struct tcphdr **tcph) 593 { 594 struct flow_ports *ports; 595 struct ipv6hdr *ip6h; 596 unsigned int thoff; 597 size_t hdrsize; 598 u8 nexthdr; 599 600 if (!pskb_network_may_pull(skb, sizeof(*ip6h))) 601 return false; 602 603 ip6h = ipv6_hdr(skb); 604 thoff = sizeof(*ip6h); 605 606 nexthdr = ip6h->nexthdr; 607 switch (nexthdr) { 608 case IPPROTO_TCP: 609 hdrsize = sizeof(struct tcphdr); 610 break; 611 case IPPROTO_UDP: 612 hdrsize = sizeof(*ports); 613 break; 614 #ifdef CONFIG_NF_CT_PROTO_GRE 615 case IPPROTO_GRE: 616 hdrsize = sizeof(struct gre_base_hdr); 617 break; 618 #endif 619 default: 620 return false; 621 } 622 623 if (ip6h->hop_limit <= 1) 624 return false; 625 626 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 627 return false; 628 629 switch (nexthdr) { 630 case IPPROTO_TCP: 631 *tcph = (void *)(skb_network_header(skb) + thoff); 632 fallthrough; 633 case IPPROTO_UDP: 634 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 635 tuple->src_port = ports->source; 636 tuple->dst_port = ports->dest; 637 break; 638 case IPPROTO_GRE: { 639 struct gre_base_hdr *greh; 640 641 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff); 642 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0) 643 return false; 644 break; 645 } 646 } 647 648 ip6h = ipv6_hdr(skb); 649 650 tuple->src_v6 = ip6h->saddr; 651 tuple->dst_v6 = ip6h->daddr; 652 tuple->l3proto = AF_INET6; 653 tuple->l4proto = nexthdr; 654 655 return true; 656 } 657 658 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p, 659 struct sk_buff *skb, 660 u8 family) 661 { 662 struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft; 663 struct flow_offload_tuple_rhash *tuplehash; 664 struct flow_offload_tuple tuple = {}; 665 enum ip_conntrack_info ctinfo; 666 struct tcphdr *tcph = NULL; 667 bool force_refresh = false; 668 struct flow_offload *flow; 669 struct nf_conn *ct; 670 u8 dir; 671 672 switch (family) { 673 case NFPROTO_IPV4: 674 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph)) 675 return false; 676 break; 677 case NFPROTO_IPV6: 678 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph)) 679 return false; 680 break; 681 default: 682 return false; 683 } 684 685 tuplehash = flow_offload_lookup(nf_ft, &tuple); 686 if (!tuplehash) 687 return false; 688 689 dir = tuplehash->tuple.dir; 690 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]); 691 ct = flow->ct; 692 693 if (dir == FLOW_OFFLOAD_DIR_REPLY && 694 !test_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags)) { 695 /* Only offload reply direction after connection became 696 * assured. 697 */ 698 if (test_bit(IPS_ASSURED_BIT, &ct->status)) 699 set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags); 700 else if (test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags)) 701 /* If flow_table flow has already been updated to the 702 * established state, then don't refresh. 703 */ 704 return false; 705 force_refresh = true; 706 } 707 708 if (tcph && (unlikely(tcph->fin || tcph->rst))) { 709 flow_offload_teardown(flow); 710 return false; 711 } 712 713 if (dir == FLOW_OFFLOAD_DIR_ORIGINAL) 714 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ? 715 IP_CT_ESTABLISHED : IP_CT_NEW; 716 else 717 ctinfo = IP_CT_ESTABLISHED_REPLY; 718 719 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 720 tcf_ct_flow_ct_ext_ifidx_update(flow); 721 flow_offload_refresh(nf_ft, flow, force_refresh); 722 if (!test_bit(IPS_ASSURED_BIT, &ct->status)) { 723 /* Process this flow in SW to allow promoting to ASSURED */ 724 return false; 725 } 726 727 nf_conntrack_get(&ct->ct_general); 728 nf_ct_set(skb, ct, ctinfo); 729 if (nf_ft->flags & NF_FLOWTABLE_COUNTER) 730 nf_ct_acct_update(ct, dir, skb->len); 731 732 return true; 733 } 734 735 static int tcf_ct_flow_tables_init(void) 736 { 737 return rhashtable_init(&zones_ht, &zones_params); 738 } 739 740 static void tcf_ct_flow_tables_uninit(void) 741 { 742 rhashtable_destroy(&zones_ht); 743 } 744 745 static struct tc_action_ops act_ct_ops; 746 747 struct tc_ct_action_net { 748 struct tc_action_net tn; /* Must be first */ 749 }; 750 751 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */ 752 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb, 753 struct tcf_ct_params *p) 754 { 755 enum ip_conntrack_info ctinfo; 756 struct nf_conn *ct; 757 758 ct = nf_ct_get(skb, &ctinfo); 759 if (!ct) 760 return false; 761 if (!net_eq(net, read_pnet(&ct->ct_net))) 762 goto drop_ct; 763 if (nf_ct_zone(ct)->id != p->zone) 764 goto drop_ct; 765 if (p->helper) { 766 struct nf_conn_help *help; 767 768 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); 769 if (help && rcu_access_pointer(help->helper) != p->helper) 770 goto drop_ct; 771 } 772 773 /* Force conntrack entry direction. */ 774 if ((p->ct_action & TCA_CT_ACT_FORCE) && 775 CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) { 776 if (nf_ct_is_confirmed(ct)) 777 nf_ct_kill(ct); 778 779 goto drop_ct; 780 } 781 782 return true; 783 784 drop_ct: 785 nf_ct_put(ct); 786 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 787 788 return false; 789 } 790 791 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb) 792 { 793 u8 family = NFPROTO_UNSPEC; 794 795 switch (skb_protocol(skb, true)) { 796 case htons(ETH_P_IP): 797 family = NFPROTO_IPV4; 798 break; 799 case htons(ETH_P_IPV6): 800 family = NFPROTO_IPV6; 801 break; 802 default: 803 break; 804 } 805 806 return family; 807 } 808 809 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag) 810 { 811 unsigned int len; 812 813 len = skb_network_offset(skb) + sizeof(struct iphdr); 814 if (unlikely(skb->len < len)) 815 return -EINVAL; 816 if (unlikely(!pskb_may_pull(skb, len))) 817 return -ENOMEM; 818 819 *frag = ip_is_fragment(ip_hdr(skb)); 820 return 0; 821 } 822 823 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag) 824 { 825 unsigned int flags = 0, len, payload_ofs = 0; 826 unsigned short frag_off; 827 int nexthdr; 828 829 len = skb_network_offset(skb) + sizeof(struct ipv6hdr); 830 if (unlikely(skb->len < len)) 831 return -EINVAL; 832 if (unlikely(!pskb_may_pull(skb, len))) 833 return -ENOMEM; 834 835 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags); 836 if (unlikely(nexthdr < 0)) 837 return -EPROTO; 838 839 *frag = flags & IP6_FH_F_FRAG; 840 return 0; 841 } 842 843 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb, 844 u8 family, u16 zone, bool *defrag) 845 { 846 enum ip_conntrack_info ctinfo; 847 struct nf_conn *ct; 848 int err = 0; 849 bool frag; 850 u8 proto; 851 u16 mru; 852 853 /* Previously seen (loopback)? Ignore. */ 854 ct = nf_ct_get(skb, &ctinfo); 855 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED) 856 return 0; 857 858 if (family == NFPROTO_IPV4) 859 err = tcf_ct_ipv4_is_fragment(skb, &frag); 860 else 861 err = tcf_ct_ipv6_is_fragment(skb, &frag); 862 if (err || !frag) 863 return err; 864 865 err = nf_ct_handle_fragments(net, skb, zone, family, &proto, &mru); 866 if (err) 867 return err; 868 869 *defrag = true; 870 tc_skb_cb(skb)->mru = mru; 871 872 return 0; 873 } 874 875 static void tcf_ct_params_free(struct tcf_ct_params *params) 876 { 877 if (params->helper) { 878 #if IS_ENABLED(CONFIG_NF_NAT) 879 if (params->ct_action & TCA_CT_ACT_NAT) 880 nf_nat_helper_put(params->helper); 881 #endif 882 nf_conntrack_helper_put(params->helper); 883 } 884 if (params->ct_ft) 885 tcf_ct_flow_table_put(params->ct_ft); 886 if (params->tmpl) { 887 if (params->put_labels) 888 nf_connlabels_put(nf_ct_net(params->tmpl)); 889 890 nf_ct_put(params->tmpl); 891 } 892 893 kfree(params); 894 } 895 896 static void tcf_ct_params_free_rcu(struct rcu_head *head) 897 { 898 struct tcf_ct_params *params; 899 900 params = container_of(head, struct tcf_ct_params, rcu); 901 tcf_ct_params_free(params); 902 } 903 904 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask) 905 { 906 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 907 u32 new_mark; 908 909 if (!mask) 910 return; 911 912 new_mark = mark | (READ_ONCE(ct->mark) & ~(mask)); 913 if (READ_ONCE(ct->mark) != new_mark) { 914 WRITE_ONCE(ct->mark, new_mark); 915 if (nf_ct_is_confirmed(ct)) 916 nf_conntrack_event_cache(IPCT_MARK, ct); 917 } 918 #endif 919 } 920 921 static void tcf_ct_act_set_labels(struct nf_conn *ct, 922 u32 *labels, 923 u32 *labels_m) 924 { 925 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) 926 size_t labels_sz = sizeof_field(struct tcf_ct_params, labels); 927 928 if (!memchr_inv(labels_m, 0, labels_sz)) 929 return; 930 931 nf_connlabels_replace(ct, labels, labels_m, 4); 932 #endif 933 } 934 935 static int tcf_ct_act_nat(struct sk_buff *skb, 936 struct nf_conn *ct, 937 enum ip_conntrack_info ctinfo, 938 int ct_action, 939 struct nf_nat_range2 *range, 940 bool commit) 941 { 942 #if IS_ENABLED(CONFIG_NF_NAT) 943 int err, action = 0; 944 945 if (!(ct_action & TCA_CT_ACT_NAT)) 946 return NF_ACCEPT; 947 if (ct_action & TCA_CT_ACT_NAT_SRC) 948 action |= BIT(NF_NAT_MANIP_SRC); 949 if (ct_action & TCA_CT_ACT_NAT_DST) 950 action |= BIT(NF_NAT_MANIP_DST); 951 952 err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit); 953 if (err != NF_ACCEPT) 954 return err & NF_VERDICT_MASK; 955 956 if (action & BIT(NF_NAT_MANIP_SRC)) 957 qdisc_skb_cb(skb)->post_ct_snat = 1; 958 if (action & BIT(NF_NAT_MANIP_DST)) 959 qdisc_skb_cb(skb)->post_ct_dnat = 1; 960 961 return err; 962 #else 963 return NF_ACCEPT; 964 #endif 965 } 966 967 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a, 968 struct tcf_result *res) 969 { 970 struct net *net = dev_net(skb->dev); 971 enum ip_conntrack_info ctinfo; 972 struct tcf_ct *c = to_ct(a); 973 struct nf_conn *tmpl = NULL; 974 struct nf_hook_state state; 975 bool cached, commit, clear; 976 int nh_ofs, err, retval; 977 struct tcf_ct_params *p; 978 bool add_helper = false; 979 bool skip_add = false; 980 bool defrag = false; 981 struct nf_conn *ct; 982 u8 family; 983 984 p = rcu_dereference_bh(c->params); 985 986 retval = p->action; 987 commit = p->ct_action & TCA_CT_ACT_COMMIT; 988 clear = p->ct_action & TCA_CT_ACT_CLEAR; 989 tmpl = p->tmpl; 990 991 tcf_lastuse_update(&c->tcf_tm); 992 tcf_action_update_bstats(&c->common, skb); 993 994 if (clear) { 995 qdisc_skb_cb(skb)->post_ct = false; 996 ct = nf_ct_get(skb, &ctinfo); 997 if (ct) { 998 nf_ct_put(ct); 999 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 1000 } 1001 1002 goto out_clear; 1003 } 1004 1005 family = tcf_ct_skb_nf_family(skb); 1006 if (family == NFPROTO_UNSPEC) 1007 goto drop; 1008 1009 /* The conntrack module expects to be working at L3. 1010 * We also try to pull the IPv4/6 header to linear area 1011 */ 1012 nh_ofs = skb_network_offset(skb); 1013 skb_pull_rcsum(skb, nh_ofs); 1014 err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag); 1015 if (err) 1016 goto out_frag; 1017 1018 err = nf_ct_skb_network_trim(skb, family); 1019 if (err) 1020 goto drop; 1021 1022 /* If we are recirculating packets to match on ct fields and 1023 * committing with a separate ct action, then we don't need to 1024 * actually run the packet through conntrack twice unless it's for a 1025 * different zone. 1026 */ 1027 cached = tcf_ct_skb_nfct_cached(net, skb, p); 1028 if (!cached) { 1029 if (tcf_ct_flow_table_lookup(p, skb, family)) { 1030 skip_add = true; 1031 goto do_nat; 1032 } 1033 1034 /* Associate skb with specified zone. */ 1035 if (tmpl) { 1036 nf_conntrack_put(skb_nfct(skb)); 1037 nf_conntrack_get(&tmpl->ct_general); 1038 nf_ct_set(skb, tmpl, IP_CT_NEW); 1039 } 1040 1041 state.hook = NF_INET_PRE_ROUTING; 1042 state.net = net; 1043 state.pf = family; 1044 err = nf_conntrack_in(skb, &state); 1045 if (err != NF_ACCEPT) 1046 goto nf_error; 1047 } 1048 1049 do_nat: 1050 ct = nf_ct_get(skb, &ctinfo); 1051 if (!ct) 1052 goto out_push; 1053 nf_ct_deliver_cached_events(ct); 1054 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 1055 1056 err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit); 1057 if (err != NF_ACCEPT) 1058 goto nf_error; 1059 1060 if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) { 1061 err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC); 1062 if (err) 1063 goto drop; 1064 add_helper = true; 1065 if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) { 1066 if (!nfct_seqadj_ext_add(ct)) 1067 goto drop; 1068 } 1069 } 1070 1071 if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) { 1072 err = nf_ct_helper(skb, ct, ctinfo, family); 1073 if (err != NF_ACCEPT) 1074 goto nf_error; 1075 } 1076 1077 if (commit) { 1078 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask); 1079 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask); 1080 1081 if (!nf_ct_is_confirmed(ct)) 1082 nf_conn_act_ct_ext_add(skb, ct, ctinfo); 1083 1084 /* This will take care of sending queued events 1085 * even if the connection is already confirmed. 1086 */ 1087 err = nf_conntrack_confirm(skb); 1088 if (err != NF_ACCEPT) 1089 goto nf_error; 1090 1091 /* The ct may be dropped if a clash has been resolved, 1092 * so it's necessary to retrieve it from skb again to 1093 * prevent UAF. 1094 */ 1095 ct = nf_ct_get(skb, &ctinfo); 1096 if (!ct) 1097 skip_add = true; 1098 } 1099 1100 if (!skip_add) 1101 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo); 1102 1103 out_push: 1104 skb_push_rcsum(skb, nh_ofs); 1105 1106 qdisc_skb_cb(skb)->post_ct = true; 1107 tc_skb_cb(skb)->zone = p->zone; 1108 out_clear: 1109 if (defrag) 1110 qdisc_skb_cb(skb)->pkt_len = skb->len; 1111 return retval; 1112 1113 out_frag: 1114 if (err != -EINPROGRESS) 1115 tcf_action_inc_drop_qstats(&c->common); 1116 return TC_ACT_CONSUMED; 1117 1118 drop: 1119 tcf_action_inc_drop_qstats(&c->common); 1120 return TC_ACT_SHOT; 1121 1122 nf_error: 1123 /* some verdicts store extra data in upper bits, such 1124 * as errno or queue number. 1125 */ 1126 switch (err & NF_VERDICT_MASK) { 1127 case NF_DROP: 1128 goto drop; 1129 case NF_STOLEN: 1130 tcf_action_inc_drop_qstats(&c->common); 1131 return TC_ACT_CONSUMED; 1132 default: 1133 DEBUG_NET_WARN_ON_ONCE(1); 1134 goto drop; 1135 } 1136 } 1137 1138 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = { 1139 [TCA_CT_ACTION] = { .type = NLA_U16 }, 1140 [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)), 1141 [TCA_CT_ZONE] = { .type = NLA_U16 }, 1142 [TCA_CT_MARK] = { .type = NLA_U32 }, 1143 [TCA_CT_MARK_MASK] = { .type = NLA_U32 }, 1144 [TCA_CT_LABELS] = { .type = NLA_BINARY, 1145 .len = 128 / BITS_PER_BYTE }, 1146 [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY, 1147 .len = 128 / BITS_PER_BYTE }, 1148 [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 }, 1149 [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 }, 1150 [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1151 [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1152 [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 }, 1153 [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 }, 1154 [TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN }, 1155 [TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 }, 1156 [TCA_CT_HELPER_PROTO] = { .type = NLA_U8 }, 1157 }; 1158 1159 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p, 1160 struct tc_ct *parm, 1161 struct nlattr **tb, 1162 struct netlink_ext_ack *extack) 1163 { 1164 struct nf_nat_range2 *range; 1165 1166 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1167 return 0; 1168 1169 if (!IS_ENABLED(CONFIG_NF_NAT)) { 1170 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel"); 1171 return -EOPNOTSUPP; 1172 } 1173 1174 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1175 return 0; 1176 1177 if ((p->ct_action & TCA_CT_ACT_NAT_SRC) && 1178 (p->ct_action & TCA_CT_ACT_NAT_DST)) { 1179 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time"); 1180 return -EOPNOTSUPP; 1181 } 1182 1183 range = &p->range; 1184 if (tb[TCA_CT_NAT_IPV4_MIN]) { 1185 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX]; 1186 1187 p->ipv4_range = true; 1188 range->flags |= NF_NAT_RANGE_MAP_IPS; 1189 range->min_addr.ip = 1190 nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]); 1191 1192 range->max_addr.ip = 1193 nla_get_in_addr_default(max_attr, range->min_addr.ip); 1194 } else if (tb[TCA_CT_NAT_IPV6_MIN]) { 1195 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX]; 1196 1197 p->ipv4_range = false; 1198 range->flags |= NF_NAT_RANGE_MAP_IPS; 1199 range->min_addr.in6 = 1200 nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]); 1201 1202 range->max_addr.in6 = max_attr ? 1203 nla_get_in6_addr(max_attr) : 1204 range->min_addr.in6; 1205 } 1206 1207 if (tb[TCA_CT_NAT_PORT_MIN]) { 1208 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED; 1209 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]); 1210 1211 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ? 1212 nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) : 1213 range->min_proto.all; 1214 } 1215 1216 return 0; 1217 } 1218 1219 static void tcf_ct_set_key_val(struct nlattr **tb, 1220 void *val, int val_type, 1221 void *mask, int mask_type, 1222 int len) 1223 { 1224 if (!tb[val_type]) 1225 return; 1226 nla_memcpy(val, tb[val_type], len); 1227 1228 if (!mask) 1229 return; 1230 1231 if (mask_type == TCA_CT_UNSPEC || !tb[mask_type]) 1232 memset(mask, 0xff, len); 1233 else 1234 nla_memcpy(mask, tb[mask_type], len); 1235 } 1236 1237 static int tcf_ct_fill_params(struct net *net, 1238 struct tcf_ct_params *p, 1239 struct tc_ct *parm, 1240 struct nlattr **tb, 1241 struct netlink_ext_ack *extack) 1242 { 1243 struct nf_conntrack_zone zone; 1244 int err, family, proto, len; 1245 bool put_labels = false; 1246 struct nf_conn *tmpl; 1247 char *name; 1248 1249 p->zone = NF_CT_DEFAULT_ZONE_ID; 1250 1251 tcf_ct_set_key_val(tb, 1252 &p->ct_action, TCA_CT_ACTION, 1253 NULL, TCA_CT_UNSPEC, 1254 sizeof(p->ct_action)); 1255 1256 if (p->ct_action & TCA_CT_ACT_CLEAR) 1257 return 0; 1258 1259 err = tcf_ct_fill_params_nat(p, parm, tb, extack); 1260 if (err) 1261 return err; 1262 1263 if (tb[TCA_CT_MARK]) { 1264 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) { 1265 NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled."); 1266 return -EOPNOTSUPP; 1267 } 1268 tcf_ct_set_key_val(tb, 1269 &p->mark, TCA_CT_MARK, 1270 &p->mark_mask, TCA_CT_MARK_MASK, 1271 sizeof(p->mark)); 1272 } 1273 1274 if (tb[TCA_CT_LABELS]) { 1275 unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8; 1276 1277 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) { 1278 NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled."); 1279 return -EOPNOTSUPP; 1280 } 1281 1282 if (nf_connlabels_get(net, n_bits - 1)) { 1283 NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length"); 1284 return -EOPNOTSUPP; 1285 } else { 1286 put_labels = true; 1287 } 1288 1289 tcf_ct_set_key_val(tb, 1290 p->labels, TCA_CT_LABELS, 1291 p->labels_mask, TCA_CT_LABELS_MASK, 1292 sizeof(p->labels)); 1293 } 1294 1295 if (tb[TCA_CT_ZONE]) { 1296 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) { 1297 NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled."); 1298 return -EOPNOTSUPP; 1299 } 1300 1301 tcf_ct_set_key_val(tb, 1302 &p->zone, TCA_CT_ZONE, 1303 NULL, TCA_CT_UNSPEC, 1304 sizeof(p->zone)); 1305 } 1306 1307 nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0); 1308 tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL); 1309 if (!tmpl) { 1310 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template"); 1311 return -ENOMEM; 1312 } 1313 p->tmpl = tmpl; 1314 if (tb[TCA_CT_HELPER_NAME]) { 1315 name = nla_data(tb[TCA_CT_HELPER_NAME]); 1316 len = nla_len(tb[TCA_CT_HELPER_NAME]); 1317 if (len > 16 || name[len - 1] != '\0') { 1318 NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name."); 1319 err = -EINVAL; 1320 goto err; 1321 } 1322 family = nla_get_u8_default(tb[TCA_CT_HELPER_FAMILY], AF_INET); 1323 proto = nla_get_u8_default(tb[TCA_CT_HELPER_PROTO], 1324 IPPROTO_TCP); 1325 err = nf_ct_add_helper(tmpl, name, family, proto, 1326 p->ct_action & TCA_CT_ACT_NAT, &p->helper); 1327 if (err) { 1328 NL_SET_ERR_MSG_MOD(extack, "Failed to add helper"); 1329 goto err; 1330 } 1331 } 1332 1333 p->put_labels = put_labels; 1334 1335 if (p->ct_action & TCA_CT_ACT_COMMIT) 1336 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status); 1337 return 0; 1338 err: 1339 if (put_labels) 1340 nf_connlabels_put(net); 1341 1342 nf_ct_put(p->tmpl); 1343 p->tmpl = NULL; 1344 return err; 1345 } 1346 1347 static int tcf_ct_init(struct net *net, struct nlattr *nla, 1348 struct nlattr *est, struct tc_action **a, 1349 struct tcf_proto *tp, u32 flags, 1350 struct netlink_ext_ack *extack) 1351 { 1352 struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id); 1353 bool bind = flags & TCA_ACT_FLAGS_BIND; 1354 struct tcf_ct_params *params = NULL; 1355 struct nlattr *tb[TCA_CT_MAX + 1]; 1356 struct tcf_chain *goto_ch = NULL; 1357 struct tc_ct *parm; 1358 struct tcf_ct *c; 1359 int err, res = 0; 1360 u32 index; 1361 1362 if (!nla) { 1363 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed"); 1364 return -EINVAL; 1365 } 1366 1367 if (bind && !(flags & TCA_ACT_FLAGS_AT_INGRESS_OR_CLSACT)) { 1368 NL_SET_ERR_MSG_MOD(extack, 1369 "Attaching ct to a non ingress/clsact qdisc is unsupported"); 1370 return -EOPNOTSUPP; 1371 } 1372 1373 err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack); 1374 if (err < 0) 1375 return err; 1376 1377 if (!tb[TCA_CT_PARMS]) { 1378 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters"); 1379 return -EINVAL; 1380 } 1381 parm = nla_data(tb[TCA_CT_PARMS]); 1382 index = parm->index; 1383 err = tcf_idr_check_alloc(tn, &index, a, bind); 1384 if (err < 0) 1385 return err; 1386 1387 if (!err) { 1388 err = tcf_idr_create_from_flags(tn, index, est, a, 1389 &act_ct_ops, bind, flags); 1390 if (err) { 1391 tcf_idr_cleanup(tn, index); 1392 return err; 1393 } 1394 res = ACT_P_CREATED; 1395 } else { 1396 if (bind) 1397 return ACT_P_BOUND; 1398 1399 if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 1400 tcf_idr_release(*a, bind); 1401 return -EEXIST; 1402 } 1403 } 1404 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 1405 if (err < 0) 1406 goto cleanup; 1407 1408 c = to_ct(*a); 1409 1410 params = kzalloc_obj(*params); 1411 if (unlikely(!params)) { 1412 err = -ENOMEM; 1413 goto cleanup; 1414 } 1415 1416 err = tcf_ct_fill_params(net, params, parm, tb, extack); 1417 if (err) 1418 goto cleanup; 1419 1420 err = tcf_ct_flow_table_get(net, params); 1421 if (err) 1422 goto cleanup; 1423 1424 params->action = parm->action; 1425 spin_lock_bh(&c->tcf_lock); 1426 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 1427 params = rcu_replace_pointer(c->params, params, 1428 lockdep_is_held(&c->tcf_lock)); 1429 spin_unlock_bh(&c->tcf_lock); 1430 1431 if (goto_ch) 1432 tcf_chain_put_by_act(goto_ch); 1433 if (params) 1434 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1435 1436 return res; 1437 1438 cleanup: 1439 if (goto_ch) 1440 tcf_chain_put_by_act(goto_ch); 1441 if (params) 1442 tcf_ct_params_free(params); 1443 tcf_idr_release(*a, bind); 1444 return err; 1445 } 1446 1447 static void tcf_ct_cleanup(struct tc_action *a) 1448 { 1449 struct tcf_ct_params *params; 1450 struct tcf_ct *c = to_ct(a); 1451 1452 params = rcu_dereference_protected(c->params, 1); 1453 if (params) 1454 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1455 } 1456 1457 static int tcf_ct_dump_key_val(struct sk_buff *skb, 1458 const void *val, int val_type, 1459 const void *mask, int mask_type, 1460 int len) 1461 { 1462 int err; 1463 1464 if (mask && !memchr_inv(mask, 0, len)) 1465 return 0; 1466 1467 err = nla_put(skb, val_type, len, val); 1468 if (err) 1469 return err; 1470 1471 if (mask_type != TCA_CT_UNSPEC) { 1472 err = nla_put(skb, mask_type, len, mask); 1473 if (err) 1474 return err; 1475 } 1476 1477 return 0; 1478 } 1479 1480 static int tcf_ct_dump_nat(struct sk_buff *skb, const struct tcf_ct_params *p) 1481 { 1482 const struct nf_nat_range2 *range = &p->range; 1483 1484 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1485 return 0; 1486 1487 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1488 return 0; 1489 1490 if (range->flags & NF_NAT_RANGE_MAP_IPS) { 1491 if (p->ipv4_range) { 1492 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN, 1493 range->min_addr.ip)) 1494 return -1; 1495 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX, 1496 range->max_addr.ip)) 1497 return -1; 1498 } else { 1499 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN, 1500 &range->min_addr.in6)) 1501 return -1; 1502 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX, 1503 &range->max_addr.in6)) 1504 return -1; 1505 } 1506 } 1507 1508 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) { 1509 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN, 1510 range->min_proto.all)) 1511 return -1; 1512 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX, 1513 range->max_proto.all)) 1514 return -1; 1515 } 1516 1517 return 0; 1518 } 1519 1520 static int tcf_ct_dump_helper(struct sk_buff *skb, 1521 const struct nf_conntrack_helper *helper) 1522 { 1523 if (!helper) 1524 return 0; 1525 1526 if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) || 1527 nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) || 1528 nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum)) 1529 return -1; 1530 1531 return 0; 1532 } 1533 1534 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a, 1535 int bind, int ref) 1536 { 1537 unsigned char *b = skb_tail_pointer(skb); 1538 const struct tcf_ct *c = to_ct(a); 1539 const struct tcf_ct_params *p; 1540 struct tc_ct opt = { 1541 .index = c->tcf_index, 1542 .refcnt = refcount_read(&c->tcf_refcnt) - ref, 1543 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind, 1544 }; 1545 struct tcf_t t; 1546 1547 rcu_read_lock(); 1548 p = rcu_dereference(c->params); 1549 opt.action = p->action; 1550 1551 if (tcf_ct_dump_key_val(skb, 1552 &p->ct_action, TCA_CT_ACTION, 1553 NULL, TCA_CT_UNSPEC, 1554 sizeof(p->ct_action))) 1555 goto nla_put_failure; 1556 1557 if (p->ct_action & TCA_CT_ACT_CLEAR) 1558 goto skip_dump; 1559 1560 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && 1561 tcf_ct_dump_key_val(skb, 1562 &p->mark, TCA_CT_MARK, 1563 &p->mark_mask, TCA_CT_MARK_MASK, 1564 sizeof(p->mark))) 1565 goto nla_put_failure; 1566 1567 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1568 tcf_ct_dump_key_val(skb, 1569 p->labels, TCA_CT_LABELS, 1570 p->labels_mask, TCA_CT_LABELS_MASK, 1571 sizeof(p->labels))) 1572 goto nla_put_failure; 1573 1574 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 1575 tcf_ct_dump_key_val(skb, 1576 &p->zone, TCA_CT_ZONE, 1577 NULL, TCA_CT_UNSPEC, 1578 sizeof(p->zone))) 1579 goto nla_put_failure; 1580 1581 if (tcf_ct_dump_nat(skb, p)) 1582 goto nla_put_failure; 1583 1584 if (tcf_ct_dump_helper(skb, p->helper)) 1585 goto nla_put_failure; 1586 1587 skip_dump: 1588 if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt)) 1589 goto nla_put_failure; 1590 1591 tcf_tm_dump(&t, &c->tcf_tm); 1592 if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD)) 1593 goto nla_put_failure; 1594 rcu_read_unlock(); 1595 1596 return skb->len; 1597 nla_put_failure: 1598 rcu_read_unlock(); 1599 nlmsg_trim(skb, b); 1600 return -1; 1601 } 1602 1603 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 1604 u64 drops, u64 lastuse, bool hw) 1605 { 1606 struct tcf_ct *c = to_ct(a); 1607 1608 tcf_action_update_stats(a, bytes, packets, drops, hw); 1609 c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse); 1610 } 1611 1612 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data, 1613 u32 *index_inc, bool bind, 1614 struct netlink_ext_ack *extack) 1615 { 1616 if (bind) { 1617 struct flow_action_entry *entry = entry_data; 1618 1619 if (tcf_ct_helper(act)) 1620 return -EOPNOTSUPP; 1621 1622 entry->id = FLOW_ACTION_CT; 1623 entry->ct.action = tcf_ct_action(act); 1624 entry->ct.zone = tcf_ct_zone(act); 1625 entry->ct.flow_table = tcf_ct_ft(act); 1626 *index_inc = 1; 1627 } else { 1628 struct flow_offload_action *fl_action = entry_data; 1629 1630 fl_action->id = FLOW_ACTION_CT; 1631 } 1632 1633 return 0; 1634 } 1635 1636 static struct tc_action_ops act_ct_ops = { 1637 .kind = "ct", 1638 .id = TCA_ID_CT, 1639 .owner = THIS_MODULE, 1640 .act = tcf_ct_act, 1641 .dump = tcf_ct_dump, 1642 .init = tcf_ct_init, 1643 .cleanup = tcf_ct_cleanup, 1644 .stats_update = tcf_stats_update, 1645 .offload_act_setup = tcf_ct_offload_act_setup, 1646 .size = sizeof(struct tcf_ct), 1647 }; 1648 MODULE_ALIAS_NET_ACT("ct"); 1649 1650 static __net_init int ct_init_net(struct net *net) 1651 { 1652 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1653 1654 return tc_action_net_init(net, &tn->tn, &act_ct_ops); 1655 } 1656 1657 static void __net_exit ct_exit_net(struct list_head *net_list) 1658 { 1659 tc_action_net_exit(net_list, act_ct_ops.net_id); 1660 } 1661 1662 static struct pernet_operations ct_net_ops = { 1663 .init = ct_init_net, 1664 .exit_batch = ct_exit_net, 1665 .id = &act_ct_ops.net_id, 1666 .size = sizeof(struct tc_ct_action_net), 1667 }; 1668 1669 static int __init ct_init_module(void) 1670 { 1671 int err; 1672 1673 act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0); 1674 if (!act_ct_wq) 1675 return -ENOMEM; 1676 1677 err = tcf_ct_flow_tables_init(); 1678 if (err) 1679 goto err_tbl_init; 1680 1681 err = tcf_register_action(&act_ct_ops, &ct_net_ops); 1682 if (err) 1683 goto err_register; 1684 1685 static_branch_inc(&tcf_frag_xmit_count); 1686 1687 return 0; 1688 1689 err_register: 1690 tcf_ct_flow_tables_uninit(); 1691 err_tbl_init: 1692 destroy_workqueue(act_ct_wq); 1693 return err; 1694 } 1695 1696 static void __exit ct_cleanup_module(void) 1697 { 1698 static_branch_dec(&tcf_frag_xmit_count); 1699 tcf_unregister_action(&act_ct_ops, &ct_net_ops); 1700 tcf_ct_flow_tables_uninit(); 1701 destroy_workqueue(act_ct_wq); 1702 } 1703 1704 module_init(ct_init_module); 1705 module_exit(ct_cleanup_module); 1706 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>"); 1707 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>"); 1708 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>"); 1709 MODULE_DESCRIPTION("Connection tracking action"); 1710 MODULE_LICENSE("GPL v2"); 1711