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 zones_ht_key { 45 struct net *net; 46 u16 zone; 47 }; 48 49 struct tcf_ct_flow_table { 50 struct rhash_head node; /* In zones tables */ 51 52 struct rcu_work rwork; 53 struct nf_flowtable nf_ft; 54 refcount_t ref; 55 struct zones_ht_key key; 56 57 bool dying; 58 }; 59 60 static const struct rhashtable_params zones_params = { 61 .head_offset = offsetof(struct tcf_ct_flow_table, node), 62 .key_offset = offsetof(struct tcf_ct_flow_table, key), 63 .key_len = sizeof_field(struct tcf_ct_flow_table, key), 64 .automatic_shrinking = true, 65 }; 66 67 static struct flow_action_entry * 68 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action) 69 { 70 int i = flow_action->num_entries++; 71 72 return &flow_action->entries[i]; 73 } 74 75 static void tcf_ct_add_mangle_action(struct flow_action *action, 76 enum flow_action_mangle_base htype, 77 u32 offset, 78 u32 mask, 79 u32 val) 80 { 81 struct flow_action_entry *entry; 82 83 entry = tcf_ct_flow_table_flow_action_get_next(action); 84 entry->id = FLOW_ACTION_MANGLE; 85 entry->mangle.htype = htype; 86 entry->mangle.mask = ~mask; 87 entry->mangle.offset = offset; 88 entry->mangle.val = val; 89 } 90 91 /* The following nat helper functions check if the inverted reverse tuple 92 * (target) is different then the current dir tuple - meaning nat for ports 93 * and/or ip is needed, and add the relevant mangle actions. 94 */ 95 static void 96 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple, 97 struct nf_conntrack_tuple target, 98 struct flow_action *action) 99 { 100 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 101 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 102 offsetof(struct iphdr, saddr), 103 0xFFFFFFFF, 104 be32_to_cpu(target.src.u3.ip)); 105 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 106 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 107 offsetof(struct iphdr, daddr), 108 0xFFFFFFFF, 109 be32_to_cpu(target.dst.u3.ip)); 110 } 111 112 static void 113 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action, 114 union nf_inet_addr *addr, 115 u32 offset) 116 { 117 int i; 118 119 for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++) 120 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6, 121 i * sizeof(u32) + offset, 122 0xFFFFFFFF, be32_to_cpu(addr->ip6[i])); 123 } 124 125 static void 126 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple, 127 struct nf_conntrack_tuple target, 128 struct flow_action *action) 129 { 130 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 131 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3, 132 offsetof(struct ipv6hdr, 133 saddr)); 134 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 135 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3, 136 offsetof(struct ipv6hdr, 137 daddr)); 138 } 139 140 static void 141 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple, 142 struct nf_conntrack_tuple target, 143 struct flow_action *action) 144 { 145 __be16 target_src = target.src.u.tcp.port; 146 __be16 target_dst = target.dst.u.tcp.port; 147 148 if (target_src != tuple->src.u.tcp.port) 149 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 150 offsetof(struct tcphdr, source), 151 0xFFFF, be16_to_cpu(target_src)); 152 if (target_dst != tuple->dst.u.tcp.port) 153 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 154 offsetof(struct tcphdr, dest), 155 0xFFFF, be16_to_cpu(target_dst)); 156 } 157 158 static void 159 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple, 160 struct nf_conntrack_tuple target, 161 struct flow_action *action) 162 { 163 __be16 target_src = target.src.u.udp.port; 164 __be16 target_dst = target.dst.u.udp.port; 165 166 if (target_src != tuple->src.u.udp.port) 167 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 168 offsetof(struct udphdr, source), 169 0xFFFF, be16_to_cpu(target_src)); 170 if (target_dst != tuple->dst.u.udp.port) 171 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 172 offsetof(struct udphdr, dest), 173 0xFFFF, be16_to_cpu(target_dst)); 174 } 175 176 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct, 177 enum ip_conntrack_dir dir, 178 enum ip_conntrack_info ctinfo, 179 struct flow_action *action) 180 { 181 struct nf_conn_labels *ct_labels; 182 struct flow_action_entry *entry; 183 u32 *act_ct_labels; 184 185 entry = tcf_ct_flow_table_flow_action_get_next(action); 186 entry->id = FLOW_ACTION_CT_METADATA; 187 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 188 entry->ct_metadata.mark = READ_ONCE(ct->mark); 189 #endif 190 /* aligns with the CT reference on the SKB nf_ct_set */ 191 entry->ct_metadata.cookie = (unsigned long)ct | ctinfo; 192 entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL; 193 194 act_ct_labels = entry->ct_metadata.labels; 195 ct_labels = nf_ct_labels_find(ct); 196 if (ct_labels) 197 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE); 198 else 199 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE); 200 } 201 202 static int tcf_ct_flow_table_add_action_nat(struct net *net, 203 struct nf_conn *ct, 204 enum ip_conntrack_dir dir, 205 struct flow_action *action) 206 { 207 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple; 208 struct nf_conntrack_tuple target; 209 210 if (!(ct->status & IPS_NAT_MASK)) 211 return 0; 212 213 nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple); 214 215 switch (tuple->src.l3num) { 216 case NFPROTO_IPV4: 217 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target, 218 action); 219 break; 220 case NFPROTO_IPV6: 221 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target, 222 action); 223 break; 224 default: 225 return -EOPNOTSUPP; 226 } 227 228 switch (nf_ct_protonum(ct)) { 229 case IPPROTO_TCP: 230 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action); 231 break; 232 case IPPROTO_UDP: 233 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action); 234 break; 235 default: 236 return -EOPNOTSUPP; 237 } 238 239 return 0; 240 } 241 242 static int tcf_ct_flow_table_fill_actions(struct net *net, 243 struct flow_offload *flow, 244 enum flow_offload_tuple_dir tdir, 245 struct nf_flow_rule *flow_rule) 246 { 247 struct flow_action *action = &flow_rule->rule->action; 248 int num_entries = action->num_entries; 249 struct nf_conn *ct = flow->ct; 250 enum ip_conntrack_info ctinfo; 251 enum ip_conntrack_dir dir; 252 int i, err; 253 254 switch (tdir) { 255 case FLOW_OFFLOAD_DIR_ORIGINAL: 256 dir = IP_CT_DIR_ORIGINAL; 257 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ? 258 IP_CT_ESTABLISHED : IP_CT_NEW; 259 if (ctinfo == IP_CT_ESTABLISHED) 260 set_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags); 261 break; 262 case FLOW_OFFLOAD_DIR_REPLY: 263 dir = IP_CT_DIR_REPLY; 264 ctinfo = IP_CT_ESTABLISHED_REPLY; 265 break; 266 default: 267 return -EOPNOTSUPP; 268 } 269 270 err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action); 271 if (err) 272 goto err_nat; 273 274 tcf_ct_flow_table_add_action_meta(ct, dir, ctinfo, action); 275 return 0; 276 277 err_nat: 278 /* Clear filled actions */ 279 for (i = num_entries; i < action->num_entries; i++) 280 memset(&action->entries[i], 0, sizeof(action->entries[i])); 281 action->num_entries = num_entries; 282 283 return err; 284 } 285 286 static bool tcf_ct_flow_is_outdated(const struct flow_offload *flow) 287 { 288 return test_bit(IPS_SEEN_REPLY_BIT, &flow->ct->status) && 289 test_bit(IPS_HW_OFFLOAD_BIT, &flow->ct->status) && 290 !test_bit(NF_FLOW_HW_PENDING, &flow->flags) && 291 !test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags); 292 } 293 294 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft); 295 296 static void tcf_ct_nf_get(struct nf_flowtable *ft) 297 { 298 struct tcf_ct_flow_table *ct_ft = 299 container_of(ft, struct tcf_ct_flow_table, nf_ft); 300 301 tcf_ct_flow_table_get_ref(ct_ft); 302 } 303 304 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft); 305 306 static void tcf_ct_nf_put(struct nf_flowtable *ft) 307 { 308 struct tcf_ct_flow_table *ct_ft = 309 container_of(ft, struct tcf_ct_flow_table, nf_ft); 310 311 tcf_ct_flow_table_put(ct_ft); 312 } 313 314 static struct nf_flowtable_type flowtable_ct = { 315 .gc = tcf_ct_flow_is_outdated, 316 .action = tcf_ct_flow_table_fill_actions, 317 .get = tcf_ct_nf_get, 318 .put = tcf_ct_nf_put, 319 .owner = THIS_MODULE, 320 }; 321 322 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params) 323 { 324 struct zones_ht_key key = { .net = net, .zone = params->zone }; 325 struct tcf_ct_flow_table *ct_ft; 326 int err = -ENOMEM; 327 328 mutex_lock(&zones_mutex); 329 ct_ft = rhashtable_lookup_fast(&zones_ht, &key, zones_params); 330 if (ct_ft && refcount_inc_not_zero(&ct_ft->ref)) 331 goto out_unlock; 332 333 ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL); 334 if (!ct_ft) 335 goto err_alloc; 336 refcount_set(&ct_ft->ref, 1); 337 338 ct_ft->key = key; 339 err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params); 340 if (err) 341 goto err_insert; 342 343 ct_ft->nf_ft.type = &flowtable_ct; 344 ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD | 345 NF_FLOWTABLE_COUNTER; 346 err = nf_flow_table_init(&ct_ft->nf_ft); 347 if (err) 348 goto err_init; 349 write_pnet(&ct_ft->nf_ft.net, net); 350 351 __module_get(THIS_MODULE); 352 out_unlock: 353 params->ct_ft = ct_ft; 354 params->nf_ft = &ct_ft->nf_ft; 355 mutex_unlock(&zones_mutex); 356 357 return 0; 358 359 err_init: 360 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 361 err_insert: 362 kfree(ct_ft); 363 err_alloc: 364 mutex_unlock(&zones_mutex); 365 return err; 366 } 367 368 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft) 369 { 370 refcount_inc(&ct_ft->ref); 371 } 372 373 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work) 374 { 375 struct tcf_ct_flow_table *ct_ft; 376 struct flow_block *block; 377 378 ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table, 379 rwork); 380 nf_flow_table_free(&ct_ft->nf_ft); 381 382 block = &ct_ft->nf_ft.flow_block; 383 down_write(&ct_ft->nf_ft.flow_block_lock); 384 WARN_ON(!list_empty(&block->cb_list)); 385 up_write(&ct_ft->nf_ft.flow_block_lock); 386 kfree(ct_ft); 387 388 module_put(THIS_MODULE); 389 } 390 391 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft) 392 { 393 if (refcount_dec_and_test(&ct_ft->ref)) { 394 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 395 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work); 396 queue_rcu_work(act_ct_wq, &ct_ft->rwork); 397 } 398 } 399 400 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry, 401 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir) 402 { 403 entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC; 404 entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir]; 405 } 406 407 static void tcf_ct_flow_ct_ext_ifidx_update(struct flow_offload *entry) 408 { 409 struct nf_conn_act_ct_ext *act_ct_ext; 410 411 act_ct_ext = nf_conn_act_ct_ext_find(entry->ct); 412 if (act_ct_ext) { 413 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL); 414 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY); 415 } 416 } 417 418 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft, 419 struct nf_conn *ct, 420 bool tcp, bool bidirectional) 421 { 422 struct nf_conn_act_ct_ext *act_ct_ext; 423 struct flow_offload *entry; 424 int err; 425 426 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status)) 427 return; 428 429 entry = flow_offload_alloc(ct); 430 if (!entry) { 431 WARN_ON_ONCE(1); 432 goto err_alloc; 433 } 434 435 if (tcp) { 436 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 437 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 438 } 439 if (bidirectional) 440 __set_bit(NF_FLOW_HW_BIDIRECTIONAL, &entry->flags); 441 442 act_ct_ext = nf_conn_act_ct_ext_find(ct); 443 if (act_ct_ext) { 444 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL); 445 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY); 446 } 447 448 err = flow_offload_add(&ct_ft->nf_ft, entry); 449 if (err) 450 goto err_add; 451 452 return; 453 454 err_add: 455 flow_offload_free(entry); 456 err_alloc: 457 clear_bit(IPS_OFFLOAD_BIT, &ct->status); 458 } 459 460 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft, 461 struct nf_conn *ct, 462 enum ip_conntrack_info ctinfo) 463 { 464 bool tcp = false, bidirectional = true; 465 466 switch (nf_ct_protonum(ct)) { 467 case IPPROTO_TCP: 468 if ((ctinfo != IP_CT_ESTABLISHED && 469 ctinfo != IP_CT_ESTABLISHED_REPLY) || 470 !test_bit(IPS_ASSURED_BIT, &ct->status) || 471 ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED) 472 return; 473 474 tcp = true; 475 break; 476 case IPPROTO_UDP: 477 if (!nf_ct_is_confirmed(ct)) 478 return; 479 if (!test_bit(IPS_ASSURED_BIT, &ct->status)) 480 bidirectional = false; 481 break; 482 #ifdef CONFIG_NF_CT_PROTO_GRE 483 case IPPROTO_GRE: { 484 struct nf_conntrack_tuple *tuple; 485 486 if ((ctinfo != IP_CT_ESTABLISHED && 487 ctinfo != IP_CT_ESTABLISHED_REPLY) || 488 !test_bit(IPS_ASSURED_BIT, &ct->status) || 489 ct->status & IPS_NAT_MASK) 490 return; 491 492 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; 493 /* No support for GRE v1 */ 494 if (tuple->src.u.gre.key || tuple->dst.u.gre.key) 495 return; 496 break; 497 } 498 #endif 499 default: 500 return; 501 } 502 503 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) || 504 ct->status & IPS_SEQ_ADJUST) 505 return; 506 507 tcf_ct_flow_table_add(ct_ft, ct, tcp, bidirectional); 508 } 509 510 static bool 511 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb, 512 struct flow_offload_tuple *tuple, 513 struct tcphdr **tcph) 514 { 515 struct flow_ports *ports; 516 unsigned int thoff; 517 struct iphdr *iph; 518 size_t hdrsize; 519 u8 ipproto; 520 521 if (!pskb_network_may_pull(skb, sizeof(*iph))) 522 return false; 523 524 iph = ip_hdr(skb); 525 thoff = iph->ihl * 4; 526 527 if (ip_is_fragment(iph) || 528 unlikely(thoff != sizeof(struct iphdr))) 529 return false; 530 531 ipproto = iph->protocol; 532 switch (ipproto) { 533 case IPPROTO_TCP: 534 hdrsize = sizeof(struct tcphdr); 535 break; 536 case IPPROTO_UDP: 537 hdrsize = sizeof(*ports); 538 break; 539 #ifdef CONFIG_NF_CT_PROTO_GRE 540 case IPPROTO_GRE: 541 hdrsize = sizeof(struct gre_base_hdr); 542 break; 543 #endif 544 default: 545 return false; 546 } 547 548 if (iph->ttl <= 1) 549 return false; 550 551 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 552 return false; 553 554 switch (ipproto) { 555 case IPPROTO_TCP: 556 *tcph = (void *)(skb_network_header(skb) + thoff); 557 fallthrough; 558 case IPPROTO_UDP: 559 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 560 tuple->src_port = ports->source; 561 tuple->dst_port = ports->dest; 562 break; 563 case IPPROTO_GRE: { 564 struct gre_base_hdr *greh; 565 566 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff); 567 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0) 568 return false; 569 break; 570 } 571 } 572 573 iph = ip_hdr(skb); 574 575 tuple->src_v4.s_addr = iph->saddr; 576 tuple->dst_v4.s_addr = iph->daddr; 577 tuple->l3proto = AF_INET; 578 tuple->l4proto = ipproto; 579 580 return true; 581 } 582 583 static bool 584 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb, 585 struct flow_offload_tuple *tuple, 586 struct tcphdr **tcph) 587 { 588 struct flow_ports *ports; 589 struct ipv6hdr *ip6h; 590 unsigned int thoff; 591 size_t hdrsize; 592 u8 nexthdr; 593 594 if (!pskb_network_may_pull(skb, sizeof(*ip6h))) 595 return false; 596 597 ip6h = ipv6_hdr(skb); 598 thoff = sizeof(*ip6h); 599 600 nexthdr = ip6h->nexthdr; 601 switch (nexthdr) { 602 case IPPROTO_TCP: 603 hdrsize = sizeof(struct tcphdr); 604 break; 605 case IPPROTO_UDP: 606 hdrsize = sizeof(*ports); 607 break; 608 #ifdef CONFIG_NF_CT_PROTO_GRE 609 case IPPROTO_GRE: 610 hdrsize = sizeof(struct gre_base_hdr); 611 break; 612 #endif 613 default: 614 return false; 615 } 616 617 if (ip6h->hop_limit <= 1) 618 return false; 619 620 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 621 return false; 622 623 switch (nexthdr) { 624 case IPPROTO_TCP: 625 *tcph = (void *)(skb_network_header(skb) + thoff); 626 fallthrough; 627 case IPPROTO_UDP: 628 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 629 tuple->src_port = ports->source; 630 tuple->dst_port = ports->dest; 631 break; 632 case IPPROTO_GRE: { 633 struct gre_base_hdr *greh; 634 635 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff); 636 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0) 637 return false; 638 break; 639 } 640 } 641 642 ip6h = ipv6_hdr(skb); 643 644 tuple->src_v6 = ip6h->saddr; 645 tuple->dst_v6 = ip6h->daddr; 646 tuple->l3proto = AF_INET6; 647 tuple->l4proto = nexthdr; 648 649 return true; 650 } 651 652 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p, 653 struct sk_buff *skb, 654 u8 family) 655 { 656 struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft; 657 struct flow_offload_tuple_rhash *tuplehash; 658 struct flow_offload_tuple tuple = {}; 659 enum ip_conntrack_info ctinfo; 660 struct tcphdr *tcph = NULL; 661 bool force_refresh = false; 662 struct flow_offload *flow; 663 struct nf_conn *ct; 664 u8 dir; 665 666 switch (family) { 667 case NFPROTO_IPV4: 668 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph)) 669 return false; 670 break; 671 case NFPROTO_IPV6: 672 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph)) 673 return false; 674 break; 675 default: 676 return false; 677 } 678 679 tuplehash = flow_offload_lookup(nf_ft, &tuple); 680 if (!tuplehash) 681 return false; 682 683 dir = tuplehash->tuple.dir; 684 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]); 685 ct = flow->ct; 686 687 if (dir == FLOW_OFFLOAD_DIR_REPLY && 688 !test_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags)) { 689 /* Only offload reply direction after connection became 690 * assured. 691 */ 692 if (test_bit(IPS_ASSURED_BIT, &ct->status)) 693 set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags); 694 else if (test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags)) 695 /* If flow_table flow has already been updated to the 696 * established state, then don't refresh. 697 */ 698 return false; 699 force_refresh = true; 700 } 701 702 if (tcph && (unlikely(tcph->fin || tcph->rst))) { 703 flow_offload_teardown(flow); 704 return false; 705 } 706 707 if (dir == FLOW_OFFLOAD_DIR_ORIGINAL) 708 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ? 709 IP_CT_ESTABLISHED : IP_CT_NEW; 710 else 711 ctinfo = IP_CT_ESTABLISHED_REPLY; 712 713 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 714 tcf_ct_flow_ct_ext_ifidx_update(flow); 715 flow_offload_refresh(nf_ft, flow, force_refresh); 716 if (!test_bit(IPS_ASSURED_BIT, &ct->status)) { 717 /* Process this flow in SW to allow promoting to ASSURED */ 718 return false; 719 } 720 721 nf_conntrack_get(&ct->ct_general); 722 nf_ct_set(skb, ct, ctinfo); 723 if (nf_ft->flags & NF_FLOWTABLE_COUNTER) 724 nf_ct_acct_update(ct, dir, skb->len); 725 726 return true; 727 } 728 729 static int tcf_ct_flow_tables_init(void) 730 { 731 return rhashtable_init(&zones_ht, &zones_params); 732 } 733 734 static void tcf_ct_flow_tables_uninit(void) 735 { 736 rhashtable_destroy(&zones_ht); 737 } 738 739 static struct tc_action_ops act_ct_ops; 740 741 struct tc_ct_action_net { 742 struct tc_action_net tn; /* Must be first */ 743 bool labels; 744 }; 745 746 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */ 747 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb, 748 struct tcf_ct_params *p) 749 { 750 enum ip_conntrack_info ctinfo; 751 struct nf_conn *ct; 752 753 ct = nf_ct_get(skb, &ctinfo); 754 if (!ct) 755 return false; 756 if (!net_eq(net, read_pnet(&ct->ct_net))) 757 goto drop_ct; 758 if (nf_ct_zone(ct)->id != p->zone) 759 goto drop_ct; 760 if (p->helper) { 761 struct nf_conn_help *help; 762 763 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); 764 if (help && rcu_access_pointer(help->helper) != p->helper) 765 goto drop_ct; 766 } 767 768 /* Force conntrack entry direction. */ 769 if ((p->ct_action & TCA_CT_ACT_FORCE) && 770 CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) { 771 if (nf_ct_is_confirmed(ct)) 772 nf_ct_kill(ct); 773 774 goto drop_ct; 775 } 776 777 return true; 778 779 drop_ct: 780 nf_ct_put(ct); 781 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 782 783 return false; 784 } 785 786 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb) 787 { 788 u8 family = NFPROTO_UNSPEC; 789 790 switch (skb_protocol(skb, true)) { 791 case htons(ETH_P_IP): 792 family = NFPROTO_IPV4; 793 break; 794 case htons(ETH_P_IPV6): 795 family = NFPROTO_IPV6; 796 break; 797 default: 798 break; 799 } 800 801 return family; 802 } 803 804 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag) 805 { 806 unsigned int len; 807 808 len = skb_network_offset(skb) + sizeof(struct iphdr); 809 if (unlikely(skb->len < len)) 810 return -EINVAL; 811 if (unlikely(!pskb_may_pull(skb, len))) 812 return -ENOMEM; 813 814 *frag = ip_is_fragment(ip_hdr(skb)); 815 return 0; 816 } 817 818 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag) 819 { 820 unsigned int flags = 0, len, payload_ofs = 0; 821 unsigned short frag_off; 822 int nexthdr; 823 824 len = skb_network_offset(skb) + sizeof(struct ipv6hdr); 825 if (unlikely(skb->len < len)) 826 return -EINVAL; 827 if (unlikely(!pskb_may_pull(skb, len))) 828 return -ENOMEM; 829 830 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags); 831 if (unlikely(nexthdr < 0)) 832 return -EPROTO; 833 834 *frag = flags & IP6_FH_F_FRAG; 835 return 0; 836 } 837 838 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb, 839 u8 family, u16 zone, bool *defrag) 840 { 841 enum ip_conntrack_info ctinfo; 842 struct nf_conn *ct; 843 int err = 0; 844 bool frag; 845 u8 proto; 846 u16 mru; 847 848 /* Previously seen (loopback)? Ignore. */ 849 ct = nf_ct_get(skb, &ctinfo); 850 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED) 851 return 0; 852 853 if (family == NFPROTO_IPV4) 854 err = tcf_ct_ipv4_is_fragment(skb, &frag); 855 else 856 err = tcf_ct_ipv6_is_fragment(skb, &frag); 857 if (err || !frag) 858 return err; 859 860 err = nf_ct_handle_fragments(net, skb, zone, family, &proto, &mru); 861 if (err) 862 return err; 863 864 *defrag = true; 865 tc_skb_cb(skb)->mru = mru; 866 867 return 0; 868 } 869 870 static void tcf_ct_params_free(struct tcf_ct_params *params) 871 { 872 if (params->helper) { 873 #if IS_ENABLED(CONFIG_NF_NAT) 874 if (params->ct_action & TCA_CT_ACT_NAT) 875 nf_nat_helper_put(params->helper); 876 #endif 877 nf_conntrack_helper_put(params->helper); 878 } 879 if (params->ct_ft) 880 tcf_ct_flow_table_put(params->ct_ft); 881 if (params->tmpl) 882 nf_ct_put(params->tmpl); 883 kfree(params); 884 } 885 886 static void tcf_ct_params_free_rcu(struct rcu_head *head) 887 { 888 struct tcf_ct_params *params; 889 890 params = container_of(head, struct tcf_ct_params, rcu); 891 tcf_ct_params_free(params); 892 } 893 894 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask) 895 { 896 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 897 u32 new_mark; 898 899 if (!mask) 900 return; 901 902 new_mark = mark | (READ_ONCE(ct->mark) & ~(mask)); 903 if (READ_ONCE(ct->mark) != new_mark) { 904 WRITE_ONCE(ct->mark, new_mark); 905 if (nf_ct_is_confirmed(ct)) 906 nf_conntrack_event_cache(IPCT_MARK, ct); 907 } 908 #endif 909 } 910 911 static void tcf_ct_act_set_labels(struct nf_conn *ct, 912 u32 *labels, 913 u32 *labels_m) 914 { 915 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) 916 size_t labels_sz = sizeof_field(struct tcf_ct_params, labels); 917 918 if (!memchr_inv(labels_m, 0, labels_sz)) 919 return; 920 921 nf_connlabels_replace(ct, labels, labels_m, 4); 922 #endif 923 } 924 925 static int tcf_ct_act_nat(struct sk_buff *skb, 926 struct nf_conn *ct, 927 enum ip_conntrack_info ctinfo, 928 int ct_action, 929 struct nf_nat_range2 *range, 930 bool commit) 931 { 932 #if IS_ENABLED(CONFIG_NF_NAT) 933 int err, action = 0; 934 935 if (!(ct_action & TCA_CT_ACT_NAT)) 936 return NF_ACCEPT; 937 if (ct_action & TCA_CT_ACT_NAT_SRC) 938 action |= BIT(NF_NAT_MANIP_SRC); 939 if (ct_action & TCA_CT_ACT_NAT_DST) 940 action |= BIT(NF_NAT_MANIP_DST); 941 942 err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit); 943 944 if (action & BIT(NF_NAT_MANIP_SRC)) 945 tc_skb_cb(skb)->post_ct_snat = 1; 946 if (action & BIT(NF_NAT_MANIP_DST)) 947 tc_skb_cb(skb)->post_ct_dnat = 1; 948 949 return err; 950 #else 951 return NF_ACCEPT; 952 #endif 953 } 954 955 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a, 956 struct tcf_result *res) 957 { 958 struct net *net = dev_net(skb->dev); 959 enum ip_conntrack_info ctinfo; 960 struct tcf_ct *c = to_ct(a); 961 struct nf_conn *tmpl = NULL; 962 struct nf_hook_state state; 963 bool cached, commit, clear; 964 int nh_ofs, err, retval; 965 struct tcf_ct_params *p; 966 bool add_helper = false; 967 bool skip_add = false; 968 bool defrag = false; 969 struct nf_conn *ct; 970 u8 family; 971 972 p = rcu_dereference_bh(c->params); 973 974 retval = READ_ONCE(c->tcf_action); 975 commit = p->ct_action & TCA_CT_ACT_COMMIT; 976 clear = p->ct_action & TCA_CT_ACT_CLEAR; 977 tmpl = p->tmpl; 978 979 tcf_lastuse_update(&c->tcf_tm); 980 tcf_action_update_bstats(&c->common, skb); 981 982 if (clear) { 983 tc_skb_cb(skb)->post_ct = false; 984 ct = nf_ct_get(skb, &ctinfo); 985 if (ct) { 986 nf_ct_put(ct); 987 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 988 } 989 990 goto out_clear; 991 } 992 993 family = tcf_ct_skb_nf_family(skb); 994 if (family == NFPROTO_UNSPEC) 995 goto drop; 996 997 /* The conntrack module expects to be working at L3. 998 * We also try to pull the IPv4/6 header to linear area 999 */ 1000 nh_ofs = skb_network_offset(skb); 1001 skb_pull_rcsum(skb, nh_ofs); 1002 err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag); 1003 if (err) 1004 goto out_frag; 1005 1006 err = nf_ct_skb_network_trim(skb, family); 1007 if (err) 1008 goto drop; 1009 1010 /* If we are recirculating packets to match on ct fields and 1011 * committing with a separate ct action, then we don't need to 1012 * actually run the packet through conntrack twice unless it's for a 1013 * different zone. 1014 */ 1015 cached = tcf_ct_skb_nfct_cached(net, skb, p); 1016 if (!cached) { 1017 if (tcf_ct_flow_table_lookup(p, skb, family)) { 1018 skip_add = true; 1019 goto do_nat; 1020 } 1021 1022 /* Associate skb with specified zone. */ 1023 if (tmpl) { 1024 nf_conntrack_put(skb_nfct(skb)); 1025 nf_conntrack_get(&tmpl->ct_general); 1026 nf_ct_set(skb, tmpl, IP_CT_NEW); 1027 } 1028 1029 state.hook = NF_INET_PRE_ROUTING; 1030 state.net = net; 1031 state.pf = family; 1032 err = nf_conntrack_in(skb, &state); 1033 if (err != NF_ACCEPT) 1034 goto out_push; 1035 } 1036 1037 do_nat: 1038 ct = nf_ct_get(skb, &ctinfo); 1039 if (!ct) 1040 goto out_push; 1041 nf_ct_deliver_cached_events(ct); 1042 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 1043 1044 err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit); 1045 if (err != NF_ACCEPT) 1046 goto drop; 1047 1048 if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) { 1049 err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC); 1050 if (err) 1051 goto drop; 1052 add_helper = true; 1053 if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) { 1054 if (!nfct_seqadj_ext_add(ct)) 1055 goto drop; 1056 } 1057 } 1058 1059 if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) { 1060 if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT) 1061 goto drop; 1062 } 1063 1064 if (commit) { 1065 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask); 1066 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask); 1067 1068 if (!nf_ct_is_confirmed(ct)) 1069 nf_conn_act_ct_ext_add(skb, ct, ctinfo); 1070 1071 /* This will take care of sending queued events 1072 * even if the connection is already confirmed. 1073 */ 1074 if (nf_conntrack_confirm(skb) != NF_ACCEPT) 1075 goto drop; 1076 } 1077 1078 if (!skip_add) 1079 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo); 1080 1081 out_push: 1082 skb_push_rcsum(skb, nh_ofs); 1083 1084 tc_skb_cb(skb)->post_ct = true; 1085 tc_skb_cb(skb)->zone = p->zone; 1086 out_clear: 1087 if (defrag) 1088 qdisc_skb_cb(skb)->pkt_len = skb->len; 1089 return retval; 1090 1091 out_frag: 1092 if (err != -EINPROGRESS) 1093 tcf_action_inc_drop_qstats(&c->common); 1094 return TC_ACT_CONSUMED; 1095 1096 drop: 1097 tcf_action_inc_drop_qstats(&c->common); 1098 return TC_ACT_SHOT; 1099 } 1100 1101 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = { 1102 [TCA_CT_ACTION] = { .type = NLA_U16 }, 1103 [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)), 1104 [TCA_CT_ZONE] = { .type = NLA_U16 }, 1105 [TCA_CT_MARK] = { .type = NLA_U32 }, 1106 [TCA_CT_MARK_MASK] = { .type = NLA_U32 }, 1107 [TCA_CT_LABELS] = { .type = NLA_BINARY, 1108 .len = 128 / BITS_PER_BYTE }, 1109 [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY, 1110 .len = 128 / BITS_PER_BYTE }, 1111 [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 }, 1112 [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 }, 1113 [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1114 [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1115 [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 }, 1116 [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 }, 1117 [TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN }, 1118 [TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 }, 1119 [TCA_CT_HELPER_PROTO] = { .type = NLA_U8 }, 1120 }; 1121 1122 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p, 1123 struct tc_ct *parm, 1124 struct nlattr **tb, 1125 struct netlink_ext_ack *extack) 1126 { 1127 struct nf_nat_range2 *range; 1128 1129 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1130 return 0; 1131 1132 if (!IS_ENABLED(CONFIG_NF_NAT)) { 1133 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel"); 1134 return -EOPNOTSUPP; 1135 } 1136 1137 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1138 return 0; 1139 1140 if ((p->ct_action & TCA_CT_ACT_NAT_SRC) && 1141 (p->ct_action & TCA_CT_ACT_NAT_DST)) { 1142 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time"); 1143 return -EOPNOTSUPP; 1144 } 1145 1146 range = &p->range; 1147 if (tb[TCA_CT_NAT_IPV4_MIN]) { 1148 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX]; 1149 1150 p->ipv4_range = true; 1151 range->flags |= NF_NAT_RANGE_MAP_IPS; 1152 range->min_addr.ip = 1153 nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]); 1154 1155 range->max_addr.ip = max_attr ? 1156 nla_get_in_addr(max_attr) : 1157 range->min_addr.ip; 1158 } else if (tb[TCA_CT_NAT_IPV6_MIN]) { 1159 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX]; 1160 1161 p->ipv4_range = false; 1162 range->flags |= NF_NAT_RANGE_MAP_IPS; 1163 range->min_addr.in6 = 1164 nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]); 1165 1166 range->max_addr.in6 = max_attr ? 1167 nla_get_in6_addr(max_attr) : 1168 range->min_addr.in6; 1169 } 1170 1171 if (tb[TCA_CT_NAT_PORT_MIN]) { 1172 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED; 1173 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]); 1174 1175 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ? 1176 nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) : 1177 range->min_proto.all; 1178 } 1179 1180 return 0; 1181 } 1182 1183 static void tcf_ct_set_key_val(struct nlattr **tb, 1184 void *val, int val_type, 1185 void *mask, int mask_type, 1186 int len) 1187 { 1188 if (!tb[val_type]) 1189 return; 1190 nla_memcpy(val, tb[val_type], len); 1191 1192 if (!mask) 1193 return; 1194 1195 if (mask_type == TCA_CT_UNSPEC || !tb[mask_type]) 1196 memset(mask, 0xff, len); 1197 else 1198 nla_memcpy(mask, tb[mask_type], len); 1199 } 1200 1201 static int tcf_ct_fill_params(struct net *net, 1202 struct tcf_ct_params *p, 1203 struct tc_ct *parm, 1204 struct nlattr **tb, 1205 struct netlink_ext_ack *extack) 1206 { 1207 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1208 struct nf_conntrack_zone zone; 1209 int err, family, proto, len; 1210 struct nf_conn *tmpl; 1211 char *name; 1212 1213 p->zone = NF_CT_DEFAULT_ZONE_ID; 1214 1215 tcf_ct_set_key_val(tb, 1216 &p->ct_action, TCA_CT_ACTION, 1217 NULL, TCA_CT_UNSPEC, 1218 sizeof(p->ct_action)); 1219 1220 if (p->ct_action & TCA_CT_ACT_CLEAR) 1221 return 0; 1222 1223 err = tcf_ct_fill_params_nat(p, parm, tb, extack); 1224 if (err) 1225 return err; 1226 1227 if (tb[TCA_CT_MARK]) { 1228 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) { 1229 NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled."); 1230 return -EOPNOTSUPP; 1231 } 1232 tcf_ct_set_key_val(tb, 1233 &p->mark, TCA_CT_MARK, 1234 &p->mark_mask, TCA_CT_MARK_MASK, 1235 sizeof(p->mark)); 1236 } 1237 1238 if (tb[TCA_CT_LABELS]) { 1239 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) { 1240 NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled."); 1241 return -EOPNOTSUPP; 1242 } 1243 1244 if (!tn->labels) { 1245 NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length"); 1246 return -EOPNOTSUPP; 1247 } 1248 tcf_ct_set_key_val(tb, 1249 p->labels, TCA_CT_LABELS, 1250 p->labels_mask, TCA_CT_LABELS_MASK, 1251 sizeof(p->labels)); 1252 } 1253 1254 if (tb[TCA_CT_ZONE]) { 1255 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) { 1256 NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled."); 1257 return -EOPNOTSUPP; 1258 } 1259 1260 tcf_ct_set_key_val(tb, 1261 &p->zone, TCA_CT_ZONE, 1262 NULL, TCA_CT_UNSPEC, 1263 sizeof(p->zone)); 1264 } 1265 1266 nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0); 1267 tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL); 1268 if (!tmpl) { 1269 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template"); 1270 return -ENOMEM; 1271 } 1272 p->tmpl = tmpl; 1273 if (tb[TCA_CT_HELPER_NAME]) { 1274 name = nla_data(tb[TCA_CT_HELPER_NAME]); 1275 len = nla_len(tb[TCA_CT_HELPER_NAME]); 1276 if (len > 16 || name[len - 1] != '\0') { 1277 NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name."); 1278 err = -EINVAL; 1279 goto err; 1280 } 1281 family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(tb[TCA_CT_HELPER_FAMILY]) : AF_INET; 1282 proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP; 1283 err = nf_ct_add_helper(tmpl, name, family, proto, 1284 p->ct_action & TCA_CT_ACT_NAT, &p->helper); 1285 if (err) { 1286 NL_SET_ERR_MSG_MOD(extack, "Failed to add helper"); 1287 goto err; 1288 } 1289 } 1290 1291 if (p->ct_action & TCA_CT_ACT_COMMIT) 1292 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status); 1293 return 0; 1294 err: 1295 nf_ct_put(p->tmpl); 1296 p->tmpl = NULL; 1297 return err; 1298 } 1299 1300 static int tcf_ct_init(struct net *net, struct nlattr *nla, 1301 struct nlattr *est, struct tc_action **a, 1302 struct tcf_proto *tp, u32 flags, 1303 struct netlink_ext_ack *extack) 1304 { 1305 struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id); 1306 bool bind = flags & TCA_ACT_FLAGS_BIND; 1307 struct tcf_ct_params *params = NULL; 1308 struct nlattr *tb[TCA_CT_MAX + 1]; 1309 struct tcf_chain *goto_ch = NULL; 1310 struct tc_ct *parm; 1311 struct tcf_ct *c; 1312 int err, res = 0; 1313 u32 index; 1314 1315 if (!nla) { 1316 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed"); 1317 return -EINVAL; 1318 } 1319 1320 err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack); 1321 if (err < 0) 1322 return err; 1323 1324 if (!tb[TCA_CT_PARMS]) { 1325 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters"); 1326 return -EINVAL; 1327 } 1328 parm = nla_data(tb[TCA_CT_PARMS]); 1329 index = parm->index; 1330 err = tcf_idr_check_alloc(tn, &index, a, bind); 1331 if (err < 0) 1332 return err; 1333 1334 if (!err) { 1335 err = tcf_idr_create_from_flags(tn, index, est, a, 1336 &act_ct_ops, bind, flags); 1337 if (err) { 1338 tcf_idr_cleanup(tn, index); 1339 return err; 1340 } 1341 res = ACT_P_CREATED; 1342 } else { 1343 if (bind) 1344 return 0; 1345 1346 if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 1347 tcf_idr_release(*a, bind); 1348 return -EEXIST; 1349 } 1350 } 1351 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 1352 if (err < 0) 1353 goto cleanup; 1354 1355 c = to_ct(*a); 1356 1357 params = kzalloc(sizeof(*params), GFP_KERNEL); 1358 if (unlikely(!params)) { 1359 err = -ENOMEM; 1360 goto cleanup; 1361 } 1362 1363 err = tcf_ct_fill_params(net, params, parm, tb, extack); 1364 if (err) 1365 goto cleanup; 1366 1367 err = tcf_ct_flow_table_get(net, params); 1368 if (err) 1369 goto cleanup; 1370 1371 spin_lock_bh(&c->tcf_lock); 1372 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 1373 params = rcu_replace_pointer(c->params, params, 1374 lockdep_is_held(&c->tcf_lock)); 1375 spin_unlock_bh(&c->tcf_lock); 1376 1377 if (goto_ch) 1378 tcf_chain_put_by_act(goto_ch); 1379 if (params) 1380 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1381 1382 return res; 1383 1384 cleanup: 1385 if (goto_ch) 1386 tcf_chain_put_by_act(goto_ch); 1387 if (params) 1388 tcf_ct_params_free(params); 1389 tcf_idr_release(*a, bind); 1390 return err; 1391 } 1392 1393 static void tcf_ct_cleanup(struct tc_action *a) 1394 { 1395 struct tcf_ct_params *params; 1396 struct tcf_ct *c = to_ct(a); 1397 1398 params = rcu_dereference_protected(c->params, 1); 1399 if (params) 1400 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1401 } 1402 1403 static int tcf_ct_dump_key_val(struct sk_buff *skb, 1404 void *val, int val_type, 1405 void *mask, int mask_type, 1406 int len) 1407 { 1408 int err; 1409 1410 if (mask && !memchr_inv(mask, 0, len)) 1411 return 0; 1412 1413 err = nla_put(skb, val_type, len, val); 1414 if (err) 1415 return err; 1416 1417 if (mask_type != TCA_CT_UNSPEC) { 1418 err = nla_put(skb, mask_type, len, mask); 1419 if (err) 1420 return err; 1421 } 1422 1423 return 0; 1424 } 1425 1426 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p) 1427 { 1428 struct nf_nat_range2 *range = &p->range; 1429 1430 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1431 return 0; 1432 1433 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1434 return 0; 1435 1436 if (range->flags & NF_NAT_RANGE_MAP_IPS) { 1437 if (p->ipv4_range) { 1438 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN, 1439 range->min_addr.ip)) 1440 return -1; 1441 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX, 1442 range->max_addr.ip)) 1443 return -1; 1444 } else { 1445 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN, 1446 &range->min_addr.in6)) 1447 return -1; 1448 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX, 1449 &range->max_addr.in6)) 1450 return -1; 1451 } 1452 } 1453 1454 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) { 1455 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN, 1456 range->min_proto.all)) 1457 return -1; 1458 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX, 1459 range->max_proto.all)) 1460 return -1; 1461 } 1462 1463 return 0; 1464 } 1465 1466 static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper) 1467 { 1468 if (!helper) 1469 return 0; 1470 1471 if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) || 1472 nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) || 1473 nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum)) 1474 return -1; 1475 1476 return 0; 1477 } 1478 1479 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a, 1480 int bind, int ref) 1481 { 1482 unsigned char *b = skb_tail_pointer(skb); 1483 struct tcf_ct *c = to_ct(a); 1484 struct tcf_ct_params *p; 1485 1486 struct tc_ct opt = { 1487 .index = c->tcf_index, 1488 .refcnt = refcount_read(&c->tcf_refcnt) - ref, 1489 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind, 1490 }; 1491 struct tcf_t t; 1492 1493 spin_lock_bh(&c->tcf_lock); 1494 p = rcu_dereference_protected(c->params, 1495 lockdep_is_held(&c->tcf_lock)); 1496 opt.action = c->tcf_action; 1497 1498 if (tcf_ct_dump_key_val(skb, 1499 &p->ct_action, TCA_CT_ACTION, 1500 NULL, TCA_CT_UNSPEC, 1501 sizeof(p->ct_action))) 1502 goto nla_put_failure; 1503 1504 if (p->ct_action & TCA_CT_ACT_CLEAR) 1505 goto skip_dump; 1506 1507 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && 1508 tcf_ct_dump_key_val(skb, 1509 &p->mark, TCA_CT_MARK, 1510 &p->mark_mask, TCA_CT_MARK_MASK, 1511 sizeof(p->mark))) 1512 goto nla_put_failure; 1513 1514 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1515 tcf_ct_dump_key_val(skb, 1516 p->labels, TCA_CT_LABELS, 1517 p->labels_mask, TCA_CT_LABELS_MASK, 1518 sizeof(p->labels))) 1519 goto nla_put_failure; 1520 1521 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 1522 tcf_ct_dump_key_val(skb, 1523 &p->zone, TCA_CT_ZONE, 1524 NULL, TCA_CT_UNSPEC, 1525 sizeof(p->zone))) 1526 goto nla_put_failure; 1527 1528 if (tcf_ct_dump_nat(skb, p)) 1529 goto nla_put_failure; 1530 1531 if (tcf_ct_dump_helper(skb, p->helper)) 1532 goto nla_put_failure; 1533 1534 skip_dump: 1535 if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt)) 1536 goto nla_put_failure; 1537 1538 tcf_tm_dump(&t, &c->tcf_tm); 1539 if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD)) 1540 goto nla_put_failure; 1541 spin_unlock_bh(&c->tcf_lock); 1542 1543 return skb->len; 1544 nla_put_failure: 1545 spin_unlock_bh(&c->tcf_lock); 1546 nlmsg_trim(skb, b); 1547 return -1; 1548 } 1549 1550 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 1551 u64 drops, u64 lastuse, bool hw) 1552 { 1553 struct tcf_ct *c = to_ct(a); 1554 1555 tcf_action_update_stats(a, bytes, packets, drops, hw); 1556 c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse); 1557 } 1558 1559 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data, 1560 u32 *index_inc, bool bind, 1561 struct netlink_ext_ack *extack) 1562 { 1563 if (bind) { 1564 struct flow_action_entry *entry = entry_data; 1565 1566 if (tcf_ct_helper(act)) 1567 return -EOPNOTSUPP; 1568 1569 entry->id = FLOW_ACTION_CT; 1570 entry->ct.action = tcf_ct_action(act); 1571 entry->ct.zone = tcf_ct_zone(act); 1572 entry->ct.flow_table = tcf_ct_ft(act); 1573 *index_inc = 1; 1574 } else { 1575 struct flow_offload_action *fl_action = entry_data; 1576 1577 fl_action->id = FLOW_ACTION_CT; 1578 } 1579 1580 return 0; 1581 } 1582 1583 static struct tc_action_ops act_ct_ops = { 1584 .kind = "ct", 1585 .id = TCA_ID_CT, 1586 .owner = THIS_MODULE, 1587 .act = tcf_ct_act, 1588 .dump = tcf_ct_dump, 1589 .init = tcf_ct_init, 1590 .cleanup = tcf_ct_cleanup, 1591 .stats_update = tcf_stats_update, 1592 .offload_act_setup = tcf_ct_offload_act_setup, 1593 .size = sizeof(struct tcf_ct), 1594 }; 1595 1596 static __net_init int ct_init_net(struct net *net) 1597 { 1598 unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8; 1599 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1600 1601 if (nf_connlabels_get(net, n_bits - 1)) { 1602 tn->labels = false; 1603 pr_err("act_ct: Failed to set connlabels length"); 1604 } else { 1605 tn->labels = true; 1606 } 1607 1608 return tc_action_net_init(net, &tn->tn, &act_ct_ops); 1609 } 1610 1611 static void __net_exit ct_exit_net(struct list_head *net_list) 1612 { 1613 struct net *net; 1614 1615 rtnl_lock(); 1616 list_for_each_entry(net, net_list, exit_list) { 1617 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1618 1619 if (tn->labels) 1620 nf_connlabels_put(net); 1621 } 1622 rtnl_unlock(); 1623 1624 tc_action_net_exit(net_list, act_ct_ops.net_id); 1625 } 1626 1627 static struct pernet_operations ct_net_ops = { 1628 .init = ct_init_net, 1629 .exit_batch = ct_exit_net, 1630 .id = &act_ct_ops.net_id, 1631 .size = sizeof(struct tc_ct_action_net), 1632 }; 1633 1634 static int __init ct_init_module(void) 1635 { 1636 int err; 1637 1638 act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0); 1639 if (!act_ct_wq) 1640 return -ENOMEM; 1641 1642 err = tcf_ct_flow_tables_init(); 1643 if (err) 1644 goto err_tbl_init; 1645 1646 err = tcf_register_action(&act_ct_ops, &ct_net_ops); 1647 if (err) 1648 goto err_register; 1649 1650 static_branch_inc(&tcf_frag_xmit_count); 1651 1652 return 0; 1653 1654 err_register: 1655 tcf_ct_flow_tables_uninit(); 1656 err_tbl_init: 1657 destroy_workqueue(act_ct_wq); 1658 return err; 1659 } 1660 1661 static void __exit ct_cleanup_module(void) 1662 { 1663 static_branch_dec(&tcf_frag_xmit_count); 1664 tcf_unregister_action(&act_ct_ops, &ct_net_ops); 1665 tcf_ct_flow_tables_uninit(); 1666 destroy_workqueue(act_ct_wq); 1667 } 1668 1669 module_init(ct_init_module); 1670 module_exit(ct_cleanup_module); 1671 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>"); 1672 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>"); 1673 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>"); 1674 MODULE_DESCRIPTION("Connection tracking action"); 1675 MODULE_LICENSE("GPL v2"); 1676