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 28 #include <net/netfilter/nf_flow_table.h> 29 #include <net/netfilter/nf_conntrack.h> 30 #include <net/netfilter/nf_conntrack_core.h> 31 #include <net/netfilter/nf_conntrack_zones.h> 32 #include <net/netfilter/nf_conntrack_helper.h> 33 #include <net/netfilter/nf_conntrack_acct.h> 34 #include <net/netfilter/ipv6/nf_defrag_ipv6.h> 35 #include <net/netfilter/nf_conntrack_act_ct.h> 36 #include <net/netfilter/nf_conntrack_seqadj.h> 37 #include <uapi/linux/netfilter/nf_nat.h> 38 39 static struct workqueue_struct *act_ct_wq; 40 static struct rhashtable zones_ht; 41 static DEFINE_MUTEX(zones_mutex); 42 43 struct tcf_ct_flow_table { 44 struct rhash_head node; /* In zones tables */ 45 46 struct rcu_work rwork; 47 struct nf_flowtable nf_ft; 48 refcount_t ref; 49 u16 zone; 50 51 bool dying; 52 }; 53 54 static const struct rhashtable_params zones_params = { 55 .head_offset = offsetof(struct tcf_ct_flow_table, node), 56 .key_offset = offsetof(struct tcf_ct_flow_table, zone), 57 .key_len = sizeof_field(struct tcf_ct_flow_table, zone), 58 .automatic_shrinking = true, 59 }; 60 61 static struct flow_action_entry * 62 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action) 63 { 64 int i = flow_action->num_entries++; 65 66 return &flow_action->entries[i]; 67 } 68 69 static void tcf_ct_add_mangle_action(struct flow_action *action, 70 enum flow_action_mangle_base htype, 71 u32 offset, 72 u32 mask, 73 u32 val) 74 { 75 struct flow_action_entry *entry; 76 77 entry = tcf_ct_flow_table_flow_action_get_next(action); 78 entry->id = FLOW_ACTION_MANGLE; 79 entry->mangle.htype = htype; 80 entry->mangle.mask = ~mask; 81 entry->mangle.offset = offset; 82 entry->mangle.val = val; 83 } 84 85 /* The following nat helper functions check if the inverted reverse tuple 86 * (target) is different then the current dir tuple - meaning nat for ports 87 * and/or ip is needed, and add the relevant mangle actions. 88 */ 89 static void 90 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple, 91 struct nf_conntrack_tuple target, 92 struct flow_action *action) 93 { 94 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 95 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 96 offsetof(struct iphdr, saddr), 97 0xFFFFFFFF, 98 be32_to_cpu(target.src.u3.ip)); 99 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 100 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 101 offsetof(struct iphdr, daddr), 102 0xFFFFFFFF, 103 be32_to_cpu(target.dst.u3.ip)); 104 } 105 106 static void 107 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action, 108 union nf_inet_addr *addr, 109 u32 offset) 110 { 111 int i; 112 113 for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++) 114 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6, 115 i * sizeof(u32) + offset, 116 0xFFFFFFFF, be32_to_cpu(addr->ip6[i])); 117 } 118 119 static void 120 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple, 121 struct nf_conntrack_tuple target, 122 struct flow_action *action) 123 { 124 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 125 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3, 126 offsetof(struct ipv6hdr, 127 saddr)); 128 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 129 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3, 130 offsetof(struct ipv6hdr, 131 daddr)); 132 } 133 134 static void 135 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple, 136 struct nf_conntrack_tuple target, 137 struct flow_action *action) 138 { 139 __be16 target_src = target.src.u.tcp.port; 140 __be16 target_dst = target.dst.u.tcp.port; 141 142 if (target_src != tuple->src.u.tcp.port) 143 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 144 offsetof(struct tcphdr, source), 145 0xFFFF, be16_to_cpu(target_src)); 146 if (target_dst != tuple->dst.u.tcp.port) 147 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 148 offsetof(struct tcphdr, dest), 149 0xFFFF, be16_to_cpu(target_dst)); 150 } 151 152 static void 153 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple, 154 struct nf_conntrack_tuple target, 155 struct flow_action *action) 156 { 157 __be16 target_src = target.src.u.udp.port; 158 __be16 target_dst = target.dst.u.udp.port; 159 160 if (target_src != tuple->src.u.udp.port) 161 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 162 offsetof(struct udphdr, source), 163 0xFFFF, be16_to_cpu(target_src)); 164 if (target_dst != tuple->dst.u.udp.port) 165 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 166 offsetof(struct udphdr, dest), 167 0xFFFF, be16_to_cpu(target_dst)); 168 } 169 170 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct, 171 enum ip_conntrack_dir dir, 172 struct flow_action *action) 173 { 174 struct nf_conn_labels *ct_labels; 175 struct flow_action_entry *entry; 176 enum ip_conntrack_info ctinfo; 177 u32 *act_ct_labels; 178 179 entry = tcf_ct_flow_table_flow_action_get_next(action); 180 entry->id = FLOW_ACTION_CT_METADATA; 181 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 182 entry->ct_metadata.mark = ct->mark; 183 #endif 184 ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED : 185 IP_CT_ESTABLISHED_REPLY; 186 /* aligns with the CT reference on the SKB nf_ct_set */ 187 entry->ct_metadata.cookie = (unsigned long)ct | ctinfo; 188 entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL; 189 190 act_ct_labels = entry->ct_metadata.labels; 191 ct_labels = nf_ct_labels_find(ct); 192 if (ct_labels) 193 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE); 194 else 195 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE); 196 } 197 198 static int tcf_ct_flow_table_add_action_nat(struct net *net, 199 struct nf_conn *ct, 200 enum ip_conntrack_dir dir, 201 struct flow_action *action) 202 { 203 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple; 204 struct nf_conntrack_tuple target; 205 206 if (!(ct->status & IPS_NAT_MASK)) 207 return 0; 208 209 nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple); 210 211 switch (tuple->src.l3num) { 212 case NFPROTO_IPV4: 213 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target, 214 action); 215 break; 216 case NFPROTO_IPV6: 217 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target, 218 action); 219 break; 220 default: 221 return -EOPNOTSUPP; 222 } 223 224 switch (nf_ct_protonum(ct)) { 225 case IPPROTO_TCP: 226 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action); 227 break; 228 case IPPROTO_UDP: 229 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action); 230 break; 231 default: 232 return -EOPNOTSUPP; 233 } 234 235 return 0; 236 } 237 238 static int tcf_ct_flow_table_fill_actions(struct net *net, 239 const struct flow_offload *flow, 240 enum flow_offload_tuple_dir tdir, 241 struct nf_flow_rule *flow_rule) 242 { 243 struct flow_action *action = &flow_rule->rule->action; 244 int num_entries = action->num_entries; 245 struct nf_conn *ct = flow->ct; 246 enum ip_conntrack_dir dir; 247 int i, err; 248 249 switch (tdir) { 250 case FLOW_OFFLOAD_DIR_ORIGINAL: 251 dir = IP_CT_DIR_ORIGINAL; 252 break; 253 case FLOW_OFFLOAD_DIR_REPLY: 254 dir = IP_CT_DIR_REPLY; 255 break; 256 default: 257 return -EOPNOTSUPP; 258 } 259 260 err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action); 261 if (err) 262 goto err_nat; 263 264 tcf_ct_flow_table_add_action_meta(ct, dir, action); 265 return 0; 266 267 err_nat: 268 /* Clear filled actions */ 269 for (i = num_entries; i < action->num_entries; i++) 270 memset(&action->entries[i], 0, sizeof(action->entries[i])); 271 action->num_entries = num_entries; 272 273 return err; 274 } 275 276 static struct nf_flowtable_type flowtable_ct = { 277 .action = tcf_ct_flow_table_fill_actions, 278 .owner = THIS_MODULE, 279 }; 280 281 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params) 282 { 283 struct tcf_ct_flow_table *ct_ft; 284 int err = -ENOMEM; 285 286 mutex_lock(&zones_mutex); 287 ct_ft = rhashtable_lookup_fast(&zones_ht, ¶ms->zone, zones_params); 288 if (ct_ft && refcount_inc_not_zero(&ct_ft->ref)) 289 goto out_unlock; 290 291 ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL); 292 if (!ct_ft) 293 goto err_alloc; 294 refcount_set(&ct_ft->ref, 1); 295 296 ct_ft->zone = params->zone; 297 err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params); 298 if (err) 299 goto err_insert; 300 301 ct_ft->nf_ft.type = &flowtable_ct; 302 ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD | 303 NF_FLOWTABLE_COUNTER; 304 err = nf_flow_table_init(&ct_ft->nf_ft); 305 if (err) 306 goto err_init; 307 write_pnet(&ct_ft->nf_ft.net, net); 308 309 __module_get(THIS_MODULE); 310 out_unlock: 311 params->ct_ft = ct_ft; 312 params->nf_ft = &ct_ft->nf_ft; 313 mutex_unlock(&zones_mutex); 314 315 return 0; 316 317 err_init: 318 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 319 err_insert: 320 kfree(ct_ft); 321 err_alloc: 322 mutex_unlock(&zones_mutex); 323 return err; 324 } 325 326 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work) 327 { 328 struct flow_block_cb *block_cb, *tmp_cb; 329 struct tcf_ct_flow_table *ct_ft; 330 struct flow_block *block; 331 332 ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table, 333 rwork); 334 nf_flow_table_free(&ct_ft->nf_ft); 335 336 /* Remove any remaining callbacks before cleanup */ 337 block = &ct_ft->nf_ft.flow_block; 338 down_write(&ct_ft->nf_ft.flow_block_lock); 339 list_for_each_entry_safe(block_cb, tmp_cb, &block->cb_list, list) { 340 list_del(&block_cb->list); 341 flow_block_cb_free(block_cb); 342 } 343 up_write(&ct_ft->nf_ft.flow_block_lock); 344 kfree(ct_ft); 345 346 module_put(THIS_MODULE); 347 } 348 349 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft) 350 { 351 if (refcount_dec_and_test(&ct_ft->ref)) { 352 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 353 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work); 354 queue_rcu_work(act_ct_wq, &ct_ft->rwork); 355 } 356 } 357 358 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry, 359 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir) 360 { 361 entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC; 362 entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir]; 363 } 364 365 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft, 366 struct nf_conn *ct, 367 bool tcp) 368 { 369 struct nf_conn_act_ct_ext *act_ct_ext; 370 struct flow_offload *entry; 371 int err; 372 373 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status)) 374 return; 375 376 entry = flow_offload_alloc(ct); 377 if (!entry) { 378 WARN_ON_ONCE(1); 379 goto err_alloc; 380 } 381 382 if (tcp) { 383 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 384 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 385 } 386 387 act_ct_ext = nf_conn_act_ct_ext_find(ct); 388 if (act_ct_ext) { 389 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL); 390 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY); 391 } 392 393 err = flow_offload_add(&ct_ft->nf_ft, entry); 394 if (err) 395 goto err_add; 396 397 return; 398 399 err_add: 400 flow_offload_free(entry); 401 err_alloc: 402 clear_bit(IPS_OFFLOAD_BIT, &ct->status); 403 } 404 405 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft, 406 struct nf_conn *ct, 407 enum ip_conntrack_info ctinfo) 408 { 409 bool tcp = false; 410 411 if ((ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY) || 412 !test_bit(IPS_ASSURED_BIT, &ct->status)) 413 return; 414 415 switch (nf_ct_protonum(ct)) { 416 case IPPROTO_TCP: 417 tcp = true; 418 if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED) 419 return; 420 break; 421 case IPPROTO_UDP: 422 break; 423 #ifdef CONFIG_NF_CT_PROTO_GRE 424 case IPPROTO_GRE: { 425 struct nf_conntrack_tuple *tuple; 426 427 if (ct->status & IPS_NAT_MASK) 428 return; 429 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; 430 /* No support for GRE v1 */ 431 if (tuple->src.u.gre.key || tuple->dst.u.gre.key) 432 return; 433 break; 434 } 435 #endif 436 default: 437 return; 438 } 439 440 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) || 441 ct->status & IPS_SEQ_ADJUST) 442 return; 443 444 tcf_ct_flow_table_add(ct_ft, ct, tcp); 445 } 446 447 static bool 448 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb, 449 struct flow_offload_tuple *tuple, 450 struct tcphdr **tcph) 451 { 452 struct flow_ports *ports; 453 unsigned int thoff; 454 struct iphdr *iph; 455 size_t hdrsize; 456 u8 ipproto; 457 458 if (!pskb_network_may_pull(skb, sizeof(*iph))) 459 return false; 460 461 iph = ip_hdr(skb); 462 thoff = iph->ihl * 4; 463 464 if (ip_is_fragment(iph) || 465 unlikely(thoff != sizeof(struct iphdr))) 466 return false; 467 468 ipproto = iph->protocol; 469 switch (ipproto) { 470 case IPPROTO_TCP: 471 hdrsize = sizeof(struct tcphdr); 472 break; 473 case IPPROTO_UDP: 474 hdrsize = sizeof(*ports); 475 break; 476 #ifdef CONFIG_NF_CT_PROTO_GRE 477 case IPPROTO_GRE: 478 hdrsize = sizeof(struct gre_base_hdr); 479 break; 480 #endif 481 default: 482 return false; 483 } 484 485 if (iph->ttl <= 1) 486 return false; 487 488 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 489 return false; 490 491 switch (ipproto) { 492 case IPPROTO_TCP: 493 *tcph = (void *)(skb_network_header(skb) + thoff); 494 fallthrough; 495 case IPPROTO_UDP: 496 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 497 tuple->src_port = ports->source; 498 tuple->dst_port = ports->dest; 499 break; 500 case IPPROTO_GRE: { 501 struct gre_base_hdr *greh; 502 503 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff); 504 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0) 505 return false; 506 break; 507 } 508 } 509 510 iph = ip_hdr(skb); 511 512 tuple->src_v4.s_addr = iph->saddr; 513 tuple->dst_v4.s_addr = iph->daddr; 514 tuple->l3proto = AF_INET; 515 tuple->l4proto = ipproto; 516 517 return true; 518 } 519 520 static bool 521 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb, 522 struct flow_offload_tuple *tuple, 523 struct tcphdr **tcph) 524 { 525 struct flow_ports *ports; 526 struct ipv6hdr *ip6h; 527 unsigned int thoff; 528 size_t hdrsize; 529 u8 nexthdr; 530 531 if (!pskb_network_may_pull(skb, sizeof(*ip6h))) 532 return false; 533 534 ip6h = ipv6_hdr(skb); 535 thoff = sizeof(*ip6h); 536 537 nexthdr = ip6h->nexthdr; 538 switch (nexthdr) { 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 (ip6h->hop_limit <= 1) 555 return false; 556 557 if (!pskb_network_may_pull(skb, thoff + hdrsize)) 558 return false; 559 560 switch (nexthdr) { 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 ip6h = ipv6_hdr(skb); 580 581 tuple->src_v6 = ip6h->saddr; 582 tuple->dst_v6 = ip6h->daddr; 583 tuple->l3proto = AF_INET6; 584 tuple->l4proto = nexthdr; 585 586 return true; 587 } 588 589 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p, 590 struct sk_buff *skb, 591 u8 family) 592 { 593 struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft; 594 struct flow_offload_tuple_rhash *tuplehash; 595 struct flow_offload_tuple tuple = {}; 596 enum ip_conntrack_info ctinfo; 597 struct tcphdr *tcph = NULL; 598 struct flow_offload *flow; 599 struct nf_conn *ct; 600 u8 dir; 601 602 switch (family) { 603 case NFPROTO_IPV4: 604 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph)) 605 return false; 606 break; 607 case NFPROTO_IPV6: 608 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph)) 609 return false; 610 break; 611 default: 612 return false; 613 } 614 615 tuplehash = flow_offload_lookup(nf_ft, &tuple); 616 if (!tuplehash) 617 return false; 618 619 dir = tuplehash->tuple.dir; 620 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]); 621 ct = flow->ct; 622 623 if (tcph && (unlikely(tcph->fin || tcph->rst))) { 624 flow_offload_teardown(flow); 625 return false; 626 } 627 628 ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED : 629 IP_CT_ESTABLISHED_REPLY; 630 631 flow_offload_refresh(nf_ft, flow); 632 nf_conntrack_get(&ct->ct_general); 633 nf_ct_set(skb, ct, ctinfo); 634 if (nf_ft->flags & NF_FLOWTABLE_COUNTER) 635 nf_ct_acct_update(ct, dir, skb->len); 636 637 return true; 638 } 639 640 static int tcf_ct_flow_tables_init(void) 641 { 642 return rhashtable_init(&zones_ht, &zones_params); 643 } 644 645 static void tcf_ct_flow_tables_uninit(void) 646 { 647 rhashtable_destroy(&zones_ht); 648 } 649 650 static struct tc_action_ops act_ct_ops; 651 652 struct tc_ct_action_net { 653 struct tc_action_net tn; /* Must be first */ 654 bool labels; 655 }; 656 657 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */ 658 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb, 659 struct tcf_ct_params *p) 660 { 661 enum ip_conntrack_info ctinfo; 662 struct nf_conn *ct; 663 664 ct = nf_ct_get(skb, &ctinfo); 665 if (!ct) 666 return false; 667 if (!net_eq(net, read_pnet(&ct->ct_net))) 668 goto drop_ct; 669 if (nf_ct_zone(ct)->id != p->zone) 670 goto drop_ct; 671 if (p->helper) { 672 struct nf_conn_help *help; 673 674 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); 675 if (help && rcu_access_pointer(help->helper) != p->helper) 676 goto drop_ct; 677 } 678 679 /* Force conntrack entry direction. */ 680 if ((p->ct_action & TCA_CT_ACT_FORCE) && 681 CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) { 682 if (nf_ct_is_confirmed(ct)) 683 nf_ct_kill(ct); 684 685 goto drop_ct; 686 } 687 688 return true; 689 690 drop_ct: 691 nf_ct_put(ct); 692 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 693 694 return false; 695 } 696 697 /* Trim the skb to the length specified by the IP/IPv6 header, 698 * removing any trailing lower-layer padding. This prepares the skb 699 * for higher-layer processing that assumes skb->len excludes padding 700 * (such as nf_ip_checksum). The caller needs to pull the skb to the 701 * network header, and ensure ip_hdr/ipv6_hdr points to valid data. 702 */ 703 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family) 704 { 705 unsigned int len; 706 707 switch (family) { 708 case NFPROTO_IPV4: 709 len = ntohs(ip_hdr(skb)->tot_len); 710 break; 711 case NFPROTO_IPV6: 712 len = sizeof(struct ipv6hdr) 713 + ntohs(ipv6_hdr(skb)->payload_len); 714 break; 715 default: 716 len = skb->len; 717 } 718 719 return pskb_trim_rcsum(skb, len); 720 } 721 722 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb) 723 { 724 u8 family = NFPROTO_UNSPEC; 725 726 switch (skb_protocol(skb, true)) { 727 case htons(ETH_P_IP): 728 family = NFPROTO_IPV4; 729 break; 730 case htons(ETH_P_IPV6): 731 family = NFPROTO_IPV6; 732 break; 733 default: 734 break; 735 } 736 737 return family; 738 } 739 740 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag) 741 { 742 unsigned int len; 743 744 len = skb_network_offset(skb) + sizeof(struct iphdr); 745 if (unlikely(skb->len < len)) 746 return -EINVAL; 747 if (unlikely(!pskb_may_pull(skb, len))) 748 return -ENOMEM; 749 750 *frag = ip_is_fragment(ip_hdr(skb)); 751 return 0; 752 } 753 754 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag) 755 { 756 unsigned int flags = 0, len, payload_ofs = 0; 757 unsigned short frag_off; 758 int nexthdr; 759 760 len = skb_network_offset(skb) + sizeof(struct ipv6hdr); 761 if (unlikely(skb->len < len)) 762 return -EINVAL; 763 if (unlikely(!pskb_may_pull(skb, len))) 764 return -ENOMEM; 765 766 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags); 767 if (unlikely(nexthdr < 0)) 768 return -EPROTO; 769 770 *frag = flags & IP6_FH_F_FRAG; 771 return 0; 772 } 773 774 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb, 775 u8 family, u16 zone, bool *defrag) 776 { 777 enum ip_conntrack_info ctinfo; 778 struct nf_conn *ct; 779 int err = 0; 780 bool frag; 781 u16 mru; 782 783 /* Previously seen (loopback)? Ignore. */ 784 ct = nf_ct_get(skb, &ctinfo); 785 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED) 786 return 0; 787 788 if (family == NFPROTO_IPV4) 789 err = tcf_ct_ipv4_is_fragment(skb, &frag); 790 else 791 err = tcf_ct_ipv6_is_fragment(skb, &frag); 792 if (err || !frag) 793 return err; 794 795 skb_get(skb); 796 mru = tc_skb_cb(skb)->mru; 797 798 if (family == NFPROTO_IPV4) { 799 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone; 800 801 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 802 local_bh_disable(); 803 err = ip_defrag(net, skb, user); 804 local_bh_enable(); 805 if (err && err != -EINPROGRESS) 806 return err; 807 808 if (!err) { 809 *defrag = true; 810 mru = IPCB(skb)->frag_max_size; 811 } 812 } else { /* NFPROTO_IPV6 */ 813 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) 814 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone; 815 816 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm)); 817 err = nf_ct_frag6_gather(net, skb, user); 818 if (err && err != -EINPROGRESS) 819 goto out_free; 820 821 if (!err) { 822 *defrag = true; 823 mru = IP6CB(skb)->frag_max_size; 824 } 825 #else 826 err = -EOPNOTSUPP; 827 goto out_free; 828 #endif 829 } 830 831 if (err != -EINPROGRESS) 832 tc_skb_cb(skb)->mru = mru; 833 skb_clear_hash(skb); 834 skb->ignore_df = 1; 835 return err; 836 837 out_free: 838 kfree_skb(skb); 839 return err; 840 } 841 842 static void tcf_ct_params_free(struct tcf_ct_params *params) 843 { 844 if (params->helper) { 845 #if IS_ENABLED(CONFIG_NF_NAT) 846 if (params->ct_action & TCA_CT_ACT_NAT) 847 nf_nat_helper_put(params->helper); 848 #endif 849 nf_conntrack_helper_put(params->helper); 850 } 851 if (params->ct_ft) 852 tcf_ct_flow_table_put(params->ct_ft); 853 if (params->tmpl) 854 nf_ct_put(params->tmpl); 855 kfree(params); 856 } 857 858 static void tcf_ct_params_free_rcu(struct rcu_head *head) 859 { 860 struct tcf_ct_params *params; 861 862 params = container_of(head, struct tcf_ct_params, rcu); 863 tcf_ct_params_free(params); 864 } 865 866 #if IS_ENABLED(CONFIG_NF_NAT) 867 /* Modelled after nf_nat_ipv[46]_fn(). 868 * range is only used for new, uninitialized NAT state. 869 * Returns either NF_ACCEPT or NF_DROP. 870 */ 871 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct, 872 enum ip_conntrack_info ctinfo, 873 const struct nf_nat_range2 *range, 874 enum nf_nat_manip_type maniptype) 875 { 876 __be16 proto = skb_protocol(skb, true); 877 int hooknum, err = NF_ACCEPT; 878 879 /* See HOOK2MANIP(). */ 880 if (maniptype == NF_NAT_MANIP_SRC) 881 hooknum = NF_INET_LOCAL_IN; /* Source NAT */ 882 else 883 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */ 884 885 switch (ctinfo) { 886 case IP_CT_RELATED: 887 case IP_CT_RELATED_REPLY: 888 if (proto == htons(ETH_P_IP) && 889 ip_hdr(skb)->protocol == IPPROTO_ICMP) { 890 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo, 891 hooknum)) 892 err = NF_DROP; 893 goto out; 894 } else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) { 895 __be16 frag_off; 896 u8 nexthdr = ipv6_hdr(skb)->nexthdr; 897 int hdrlen = ipv6_skip_exthdr(skb, 898 sizeof(struct ipv6hdr), 899 &nexthdr, &frag_off); 900 901 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) { 902 if (!nf_nat_icmpv6_reply_translation(skb, ct, 903 ctinfo, 904 hooknum, 905 hdrlen)) 906 err = NF_DROP; 907 goto out; 908 } 909 } 910 /* Non-ICMP, fall thru to initialize if needed. */ 911 fallthrough; 912 case IP_CT_NEW: 913 /* Seen it before? This can happen for loopback, retrans, 914 * or local packets. 915 */ 916 if (!nf_nat_initialized(ct, maniptype)) { 917 /* Initialize according to the NAT action. */ 918 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS) 919 /* Action is set up to establish a new 920 * mapping. 921 */ 922 ? nf_nat_setup_info(ct, range, maniptype) 923 : nf_nat_alloc_null_binding(ct, hooknum); 924 if (err != NF_ACCEPT) 925 goto out; 926 } 927 break; 928 929 case IP_CT_ESTABLISHED: 930 case IP_CT_ESTABLISHED_REPLY: 931 break; 932 933 default: 934 err = NF_DROP; 935 goto out; 936 } 937 938 err = nf_nat_packet(ct, ctinfo, hooknum, skb); 939 if (err == NF_ACCEPT) { 940 if (maniptype == NF_NAT_MANIP_SRC) 941 tc_skb_cb(skb)->post_ct_snat = 1; 942 if (maniptype == NF_NAT_MANIP_DST) 943 tc_skb_cb(skb)->post_ct_dnat = 1; 944 } 945 out: 946 return err; 947 } 948 #endif /* CONFIG_NF_NAT */ 949 950 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask) 951 { 952 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 953 u32 new_mark; 954 955 if (!mask) 956 return; 957 958 new_mark = mark | (ct->mark & ~(mask)); 959 if (ct->mark != new_mark) { 960 ct->mark = new_mark; 961 if (nf_ct_is_confirmed(ct)) 962 nf_conntrack_event_cache(IPCT_MARK, ct); 963 } 964 #endif 965 } 966 967 static void tcf_ct_act_set_labels(struct nf_conn *ct, 968 u32 *labels, 969 u32 *labels_m) 970 { 971 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) 972 size_t labels_sz = sizeof_field(struct tcf_ct_params, labels); 973 974 if (!memchr_inv(labels_m, 0, labels_sz)) 975 return; 976 977 nf_connlabels_replace(ct, labels, labels_m, 4); 978 #endif 979 } 980 981 static int tcf_ct_act_nat(struct sk_buff *skb, 982 struct nf_conn *ct, 983 enum ip_conntrack_info ctinfo, 984 int ct_action, 985 struct nf_nat_range2 *range, 986 bool commit) 987 { 988 #if IS_ENABLED(CONFIG_NF_NAT) 989 int err; 990 enum nf_nat_manip_type maniptype; 991 992 if (!(ct_action & TCA_CT_ACT_NAT)) 993 return NF_ACCEPT; 994 995 /* Add NAT extension if not confirmed yet. */ 996 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct)) 997 return NF_DROP; /* Can't NAT. */ 998 999 if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) && 1000 (ctinfo != IP_CT_RELATED || commit)) { 1001 /* NAT an established or related connection like before. */ 1002 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) 1003 /* This is the REPLY direction for a connection 1004 * for which NAT was applied in the forward 1005 * direction. Do the reverse NAT. 1006 */ 1007 maniptype = ct->status & IPS_SRC_NAT 1008 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC; 1009 else 1010 maniptype = ct->status & IPS_SRC_NAT 1011 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST; 1012 } else if (ct_action & TCA_CT_ACT_NAT_SRC) { 1013 maniptype = NF_NAT_MANIP_SRC; 1014 } else if (ct_action & TCA_CT_ACT_NAT_DST) { 1015 maniptype = NF_NAT_MANIP_DST; 1016 } else { 1017 return NF_ACCEPT; 1018 } 1019 1020 err = ct_nat_execute(skb, ct, ctinfo, range, maniptype); 1021 if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) { 1022 if (ct->status & IPS_SRC_NAT) { 1023 if (maniptype == NF_NAT_MANIP_SRC) 1024 maniptype = NF_NAT_MANIP_DST; 1025 else 1026 maniptype = NF_NAT_MANIP_SRC; 1027 1028 err = ct_nat_execute(skb, ct, ctinfo, range, 1029 maniptype); 1030 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) { 1031 err = ct_nat_execute(skb, ct, ctinfo, NULL, 1032 NF_NAT_MANIP_SRC); 1033 } 1034 } 1035 return err; 1036 #else 1037 return NF_ACCEPT; 1038 #endif 1039 } 1040 1041 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a, 1042 struct tcf_result *res) 1043 { 1044 struct net *net = dev_net(skb->dev); 1045 enum ip_conntrack_info ctinfo; 1046 struct tcf_ct *c = to_ct(a); 1047 struct nf_conn *tmpl = NULL; 1048 struct nf_hook_state state; 1049 bool cached, commit, clear; 1050 int nh_ofs, err, retval; 1051 struct tcf_ct_params *p; 1052 bool add_helper = false; 1053 bool skip_add = false; 1054 bool defrag = false; 1055 struct nf_conn *ct; 1056 u8 family; 1057 1058 p = rcu_dereference_bh(c->params); 1059 1060 retval = READ_ONCE(c->tcf_action); 1061 commit = p->ct_action & TCA_CT_ACT_COMMIT; 1062 clear = p->ct_action & TCA_CT_ACT_CLEAR; 1063 tmpl = p->tmpl; 1064 1065 tcf_lastuse_update(&c->tcf_tm); 1066 tcf_action_update_bstats(&c->common, skb); 1067 1068 if (clear) { 1069 tc_skb_cb(skb)->post_ct = false; 1070 ct = nf_ct_get(skb, &ctinfo); 1071 if (ct) { 1072 nf_ct_put(ct); 1073 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 1074 } 1075 1076 goto out_clear; 1077 } 1078 1079 family = tcf_ct_skb_nf_family(skb); 1080 if (family == NFPROTO_UNSPEC) 1081 goto drop; 1082 1083 /* The conntrack module expects to be working at L3. 1084 * We also try to pull the IPv4/6 header to linear area 1085 */ 1086 nh_ofs = skb_network_offset(skb); 1087 skb_pull_rcsum(skb, nh_ofs); 1088 err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag); 1089 if (err == -EINPROGRESS) { 1090 retval = TC_ACT_STOLEN; 1091 goto out_clear; 1092 } 1093 if (err) 1094 goto drop; 1095 1096 err = tcf_ct_skb_network_trim(skb, family); 1097 if (err) 1098 goto drop; 1099 1100 /* If we are recirculating packets to match on ct fields and 1101 * committing with a separate ct action, then we don't need to 1102 * actually run the packet through conntrack twice unless it's for a 1103 * different zone. 1104 */ 1105 cached = tcf_ct_skb_nfct_cached(net, skb, p); 1106 if (!cached) { 1107 if (tcf_ct_flow_table_lookup(p, skb, family)) { 1108 skip_add = true; 1109 goto do_nat; 1110 } 1111 1112 /* Associate skb with specified zone. */ 1113 if (tmpl) { 1114 nf_conntrack_put(skb_nfct(skb)); 1115 nf_conntrack_get(&tmpl->ct_general); 1116 nf_ct_set(skb, tmpl, IP_CT_NEW); 1117 } 1118 1119 state.hook = NF_INET_PRE_ROUTING; 1120 state.net = net; 1121 state.pf = family; 1122 err = nf_conntrack_in(skb, &state); 1123 if (err != NF_ACCEPT) 1124 goto out_push; 1125 } 1126 1127 do_nat: 1128 ct = nf_ct_get(skb, &ctinfo); 1129 if (!ct) 1130 goto out_push; 1131 nf_ct_deliver_cached_events(ct); 1132 nf_conn_act_ct_ext_fill(skb, ct, ctinfo); 1133 1134 err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit); 1135 if (err != NF_ACCEPT) 1136 goto drop; 1137 1138 if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) { 1139 err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC); 1140 if (err) 1141 goto drop; 1142 add_helper = true; 1143 if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) { 1144 if (!nfct_seqadj_ext_add(ct)) 1145 goto drop; 1146 } 1147 } 1148 1149 if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) { 1150 if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT) 1151 goto drop; 1152 } 1153 1154 if (commit) { 1155 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask); 1156 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask); 1157 1158 if (!nf_ct_is_confirmed(ct)) 1159 nf_conn_act_ct_ext_add(ct); 1160 1161 /* This will take care of sending queued events 1162 * even if the connection is already confirmed. 1163 */ 1164 if (nf_conntrack_confirm(skb) != NF_ACCEPT) 1165 goto drop; 1166 } 1167 1168 if (!skip_add) 1169 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo); 1170 1171 out_push: 1172 skb_push_rcsum(skb, nh_ofs); 1173 1174 tc_skb_cb(skb)->post_ct = true; 1175 tc_skb_cb(skb)->zone = p->zone; 1176 out_clear: 1177 if (defrag) 1178 qdisc_skb_cb(skb)->pkt_len = skb->len; 1179 return retval; 1180 1181 drop: 1182 tcf_action_inc_drop_qstats(&c->common); 1183 return TC_ACT_SHOT; 1184 } 1185 1186 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = { 1187 [TCA_CT_ACTION] = { .type = NLA_U16 }, 1188 [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)), 1189 [TCA_CT_ZONE] = { .type = NLA_U16 }, 1190 [TCA_CT_MARK] = { .type = NLA_U32 }, 1191 [TCA_CT_MARK_MASK] = { .type = NLA_U32 }, 1192 [TCA_CT_LABELS] = { .type = NLA_BINARY, 1193 .len = 128 / BITS_PER_BYTE }, 1194 [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY, 1195 .len = 128 / BITS_PER_BYTE }, 1196 [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 }, 1197 [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 }, 1198 [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1199 [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1200 [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 }, 1201 [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 }, 1202 [TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN }, 1203 [TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 }, 1204 [TCA_CT_HELPER_PROTO] = { .type = NLA_U8 }, 1205 }; 1206 1207 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p, 1208 struct tc_ct *parm, 1209 struct nlattr **tb, 1210 struct netlink_ext_ack *extack) 1211 { 1212 struct nf_nat_range2 *range; 1213 1214 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1215 return 0; 1216 1217 if (!IS_ENABLED(CONFIG_NF_NAT)) { 1218 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel"); 1219 return -EOPNOTSUPP; 1220 } 1221 1222 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1223 return 0; 1224 1225 if ((p->ct_action & TCA_CT_ACT_NAT_SRC) && 1226 (p->ct_action & TCA_CT_ACT_NAT_DST)) { 1227 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time"); 1228 return -EOPNOTSUPP; 1229 } 1230 1231 range = &p->range; 1232 if (tb[TCA_CT_NAT_IPV4_MIN]) { 1233 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX]; 1234 1235 p->ipv4_range = true; 1236 range->flags |= NF_NAT_RANGE_MAP_IPS; 1237 range->min_addr.ip = 1238 nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]); 1239 1240 range->max_addr.ip = max_attr ? 1241 nla_get_in_addr(max_attr) : 1242 range->min_addr.ip; 1243 } else if (tb[TCA_CT_NAT_IPV6_MIN]) { 1244 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX]; 1245 1246 p->ipv4_range = false; 1247 range->flags |= NF_NAT_RANGE_MAP_IPS; 1248 range->min_addr.in6 = 1249 nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]); 1250 1251 range->max_addr.in6 = max_attr ? 1252 nla_get_in6_addr(max_attr) : 1253 range->min_addr.in6; 1254 } 1255 1256 if (tb[TCA_CT_NAT_PORT_MIN]) { 1257 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED; 1258 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]); 1259 1260 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ? 1261 nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) : 1262 range->min_proto.all; 1263 } 1264 1265 return 0; 1266 } 1267 1268 static void tcf_ct_set_key_val(struct nlattr **tb, 1269 void *val, int val_type, 1270 void *mask, int mask_type, 1271 int len) 1272 { 1273 if (!tb[val_type]) 1274 return; 1275 nla_memcpy(val, tb[val_type], len); 1276 1277 if (!mask) 1278 return; 1279 1280 if (mask_type == TCA_CT_UNSPEC || !tb[mask_type]) 1281 memset(mask, 0xff, len); 1282 else 1283 nla_memcpy(mask, tb[mask_type], len); 1284 } 1285 1286 static int tcf_ct_fill_params(struct net *net, 1287 struct tcf_ct_params *p, 1288 struct tc_ct *parm, 1289 struct nlattr **tb, 1290 struct netlink_ext_ack *extack) 1291 { 1292 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1293 struct nf_conntrack_zone zone; 1294 int err, family, proto, len; 1295 struct nf_conn *tmpl; 1296 char *name; 1297 1298 p->zone = NF_CT_DEFAULT_ZONE_ID; 1299 1300 tcf_ct_set_key_val(tb, 1301 &p->ct_action, TCA_CT_ACTION, 1302 NULL, TCA_CT_UNSPEC, 1303 sizeof(p->ct_action)); 1304 1305 if (p->ct_action & TCA_CT_ACT_CLEAR) 1306 return 0; 1307 1308 err = tcf_ct_fill_params_nat(p, parm, tb, extack); 1309 if (err) 1310 return err; 1311 1312 if (tb[TCA_CT_MARK]) { 1313 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) { 1314 NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled."); 1315 return -EOPNOTSUPP; 1316 } 1317 tcf_ct_set_key_val(tb, 1318 &p->mark, TCA_CT_MARK, 1319 &p->mark_mask, TCA_CT_MARK_MASK, 1320 sizeof(p->mark)); 1321 } 1322 1323 if (tb[TCA_CT_LABELS]) { 1324 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) { 1325 NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled."); 1326 return -EOPNOTSUPP; 1327 } 1328 1329 if (!tn->labels) { 1330 NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length"); 1331 return -EOPNOTSUPP; 1332 } 1333 tcf_ct_set_key_val(tb, 1334 p->labels, TCA_CT_LABELS, 1335 p->labels_mask, TCA_CT_LABELS_MASK, 1336 sizeof(p->labels)); 1337 } 1338 1339 if (tb[TCA_CT_ZONE]) { 1340 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) { 1341 NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled."); 1342 return -EOPNOTSUPP; 1343 } 1344 1345 tcf_ct_set_key_val(tb, 1346 &p->zone, TCA_CT_ZONE, 1347 NULL, TCA_CT_UNSPEC, 1348 sizeof(p->zone)); 1349 } 1350 1351 nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0); 1352 tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL); 1353 if (!tmpl) { 1354 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template"); 1355 return -ENOMEM; 1356 } 1357 p->tmpl = tmpl; 1358 if (tb[TCA_CT_HELPER_NAME]) { 1359 name = nla_data(tb[TCA_CT_HELPER_NAME]); 1360 len = nla_len(tb[TCA_CT_HELPER_NAME]); 1361 if (len > 16 || name[len - 1] != '\0') { 1362 NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name."); 1363 err = -EINVAL; 1364 goto err; 1365 } 1366 family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(tb[TCA_CT_HELPER_FAMILY]) : AF_INET; 1367 proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP; 1368 err = nf_ct_add_helper(tmpl, name, family, proto, 1369 p->ct_action & TCA_CT_ACT_NAT, &p->helper); 1370 if (err) { 1371 NL_SET_ERR_MSG_MOD(extack, "Failed to add helper"); 1372 goto err; 1373 } 1374 } 1375 1376 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status); 1377 return 0; 1378 err: 1379 nf_ct_put(p->tmpl); 1380 p->tmpl = NULL; 1381 return err; 1382 } 1383 1384 static int tcf_ct_init(struct net *net, struct nlattr *nla, 1385 struct nlattr *est, struct tc_action **a, 1386 struct tcf_proto *tp, u32 flags, 1387 struct netlink_ext_ack *extack) 1388 { 1389 struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id); 1390 bool bind = flags & TCA_ACT_FLAGS_BIND; 1391 struct tcf_ct_params *params = NULL; 1392 struct nlattr *tb[TCA_CT_MAX + 1]; 1393 struct tcf_chain *goto_ch = NULL; 1394 struct tc_ct *parm; 1395 struct tcf_ct *c; 1396 int err, res = 0; 1397 u32 index; 1398 1399 if (!nla) { 1400 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed"); 1401 return -EINVAL; 1402 } 1403 1404 err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack); 1405 if (err < 0) 1406 return err; 1407 1408 if (!tb[TCA_CT_PARMS]) { 1409 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters"); 1410 return -EINVAL; 1411 } 1412 parm = nla_data(tb[TCA_CT_PARMS]); 1413 index = parm->index; 1414 err = tcf_idr_check_alloc(tn, &index, a, bind); 1415 if (err < 0) 1416 return err; 1417 1418 if (!err) { 1419 err = tcf_idr_create_from_flags(tn, index, est, a, 1420 &act_ct_ops, bind, flags); 1421 if (err) { 1422 tcf_idr_cleanup(tn, index); 1423 return err; 1424 } 1425 res = ACT_P_CREATED; 1426 } else { 1427 if (bind) 1428 return 0; 1429 1430 if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 1431 tcf_idr_release(*a, bind); 1432 return -EEXIST; 1433 } 1434 } 1435 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 1436 if (err < 0) 1437 goto cleanup; 1438 1439 c = to_ct(*a); 1440 1441 params = kzalloc(sizeof(*params), GFP_KERNEL); 1442 if (unlikely(!params)) { 1443 err = -ENOMEM; 1444 goto cleanup; 1445 } 1446 1447 err = tcf_ct_fill_params(net, params, parm, tb, extack); 1448 if (err) 1449 goto cleanup; 1450 1451 err = tcf_ct_flow_table_get(net, params); 1452 if (err) 1453 goto cleanup; 1454 1455 spin_lock_bh(&c->tcf_lock); 1456 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 1457 params = rcu_replace_pointer(c->params, params, 1458 lockdep_is_held(&c->tcf_lock)); 1459 spin_unlock_bh(&c->tcf_lock); 1460 1461 if (goto_ch) 1462 tcf_chain_put_by_act(goto_ch); 1463 if (params) 1464 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1465 1466 return res; 1467 1468 cleanup: 1469 if (goto_ch) 1470 tcf_chain_put_by_act(goto_ch); 1471 if (params) 1472 tcf_ct_params_free(params); 1473 tcf_idr_release(*a, bind); 1474 return err; 1475 } 1476 1477 static void tcf_ct_cleanup(struct tc_action *a) 1478 { 1479 struct tcf_ct_params *params; 1480 struct tcf_ct *c = to_ct(a); 1481 1482 params = rcu_dereference_protected(c->params, 1); 1483 if (params) 1484 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu); 1485 } 1486 1487 static int tcf_ct_dump_key_val(struct sk_buff *skb, 1488 void *val, int val_type, 1489 void *mask, int mask_type, 1490 int len) 1491 { 1492 int err; 1493 1494 if (mask && !memchr_inv(mask, 0, len)) 1495 return 0; 1496 1497 err = nla_put(skb, val_type, len, val); 1498 if (err) 1499 return err; 1500 1501 if (mask_type != TCA_CT_UNSPEC) { 1502 err = nla_put(skb, mask_type, len, mask); 1503 if (err) 1504 return err; 1505 } 1506 1507 return 0; 1508 } 1509 1510 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p) 1511 { 1512 struct nf_nat_range2 *range = &p->range; 1513 1514 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1515 return 0; 1516 1517 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1518 return 0; 1519 1520 if (range->flags & NF_NAT_RANGE_MAP_IPS) { 1521 if (p->ipv4_range) { 1522 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN, 1523 range->min_addr.ip)) 1524 return -1; 1525 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX, 1526 range->max_addr.ip)) 1527 return -1; 1528 } else { 1529 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN, 1530 &range->min_addr.in6)) 1531 return -1; 1532 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX, 1533 &range->max_addr.in6)) 1534 return -1; 1535 } 1536 } 1537 1538 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) { 1539 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN, 1540 range->min_proto.all)) 1541 return -1; 1542 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX, 1543 range->max_proto.all)) 1544 return -1; 1545 } 1546 1547 return 0; 1548 } 1549 1550 static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper) 1551 { 1552 if (!helper) 1553 return 0; 1554 1555 if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) || 1556 nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) || 1557 nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum)) 1558 return -1; 1559 1560 return 0; 1561 } 1562 1563 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a, 1564 int bind, int ref) 1565 { 1566 unsigned char *b = skb_tail_pointer(skb); 1567 struct tcf_ct *c = to_ct(a); 1568 struct tcf_ct_params *p; 1569 1570 struct tc_ct opt = { 1571 .index = c->tcf_index, 1572 .refcnt = refcount_read(&c->tcf_refcnt) - ref, 1573 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind, 1574 }; 1575 struct tcf_t t; 1576 1577 spin_lock_bh(&c->tcf_lock); 1578 p = rcu_dereference_protected(c->params, 1579 lockdep_is_held(&c->tcf_lock)); 1580 opt.action = c->tcf_action; 1581 1582 if (tcf_ct_dump_key_val(skb, 1583 &p->ct_action, TCA_CT_ACTION, 1584 NULL, TCA_CT_UNSPEC, 1585 sizeof(p->ct_action))) 1586 goto nla_put_failure; 1587 1588 if (p->ct_action & TCA_CT_ACT_CLEAR) 1589 goto skip_dump; 1590 1591 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && 1592 tcf_ct_dump_key_val(skb, 1593 &p->mark, TCA_CT_MARK, 1594 &p->mark_mask, TCA_CT_MARK_MASK, 1595 sizeof(p->mark))) 1596 goto nla_put_failure; 1597 1598 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1599 tcf_ct_dump_key_val(skb, 1600 p->labels, TCA_CT_LABELS, 1601 p->labels_mask, TCA_CT_LABELS_MASK, 1602 sizeof(p->labels))) 1603 goto nla_put_failure; 1604 1605 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 1606 tcf_ct_dump_key_val(skb, 1607 &p->zone, TCA_CT_ZONE, 1608 NULL, TCA_CT_UNSPEC, 1609 sizeof(p->zone))) 1610 goto nla_put_failure; 1611 1612 if (tcf_ct_dump_nat(skb, p)) 1613 goto nla_put_failure; 1614 1615 if (tcf_ct_dump_helper(skb, p->helper)) 1616 goto nla_put_failure; 1617 1618 skip_dump: 1619 if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt)) 1620 goto nla_put_failure; 1621 1622 tcf_tm_dump(&t, &c->tcf_tm); 1623 if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD)) 1624 goto nla_put_failure; 1625 spin_unlock_bh(&c->tcf_lock); 1626 1627 return skb->len; 1628 nla_put_failure: 1629 spin_unlock_bh(&c->tcf_lock); 1630 nlmsg_trim(skb, b); 1631 return -1; 1632 } 1633 1634 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 1635 u64 drops, u64 lastuse, bool hw) 1636 { 1637 struct tcf_ct *c = to_ct(a); 1638 1639 tcf_action_update_stats(a, bytes, packets, drops, hw); 1640 c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse); 1641 } 1642 1643 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data, 1644 u32 *index_inc, bool bind, 1645 struct netlink_ext_ack *extack) 1646 { 1647 if (bind) { 1648 struct flow_action_entry *entry = entry_data; 1649 1650 entry->id = FLOW_ACTION_CT; 1651 entry->ct.action = tcf_ct_action(act); 1652 entry->ct.zone = tcf_ct_zone(act); 1653 entry->ct.flow_table = tcf_ct_ft(act); 1654 *index_inc = 1; 1655 } else { 1656 struct flow_offload_action *fl_action = entry_data; 1657 1658 fl_action->id = FLOW_ACTION_CT; 1659 } 1660 1661 return 0; 1662 } 1663 1664 static struct tc_action_ops act_ct_ops = { 1665 .kind = "ct", 1666 .id = TCA_ID_CT, 1667 .owner = THIS_MODULE, 1668 .act = tcf_ct_act, 1669 .dump = tcf_ct_dump, 1670 .init = tcf_ct_init, 1671 .cleanup = tcf_ct_cleanup, 1672 .stats_update = tcf_stats_update, 1673 .offload_act_setup = tcf_ct_offload_act_setup, 1674 .size = sizeof(struct tcf_ct), 1675 }; 1676 1677 static __net_init int ct_init_net(struct net *net) 1678 { 1679 unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8; 1680 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1681 1682 if (nf_connlabels_get(net, n_bits - 1)) { 1683 tn->labels = false; 1684 pr_err("act_ct: Failed to set connlabels length"); 1685 } else { 1686 tn->labels = true; 1687 } 1688 1689 return tc_action_net_init(net, &tn->tn, &act_ct_ops); 1690 } 1691 1692 static void __net_exit ct_exit_net(struct list_head *net_list) 1693 { 1694 struct net *net; 1695 1696 rtnl_lock(); 1697 list_for_each_entry(net, net_list, exit_list) { 1698 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id); 1699 1700 if (tn->labels) 1701 nf_connlabels_put(net); 1702 } 1703 rtnl_unlock(); 1704 1705 tc_action_net_exit(net_list, act_ct_ops.net_id); 1706 } 1707 1708 static struct pernet_operations ct_net_ops = { 1709 .init = ct_init_net, 1710 .exit_batch = ct_exit_net, 1711 .id = &act_ct_ops.net_id, 1712 .size = sizeof(struct tc_ct_action_net), 1713 }; 1714 1715 static int __init ct_init_module(void) 1716 { 1717 int err; 1718 1719 act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0); 1720 if (!act_ct_wq) 1721 return -ENOMEM; 1722 1723 err = tcf_ct_flow_tables_init(); 1724 if (err) 1725 goto err_tbl_init; 1726 1727 err = tcf_register_action(&act_ct_ops, &ct_net_ops); 1728 if (err) 1729 goto err_register; 1730 1731 static_branch_inc(&tcf_frag_xmit_count); 1732 1733 return 0; 1734 1735 err_register: 1736 tcf_ct_flow_tables_uninit(); 1737 err_tbl_init: 1738 destroy_workqueue(act_ct_wq); 1739 return err; 1740 } 1741 1742 static void __exit ct_cleanup_module(void) 1743 { 1744 static_branch_dec(&tcf_frag_xmit_count); 1745 tcf_unregister_action(&act_ct_ops, &ct_net_ops); 1746 tcf_ct_flow_tables_uninit(); 1747 destroy_workqueue(act_ct_wq); 1748 } 1749 1750 module_init(ct_init_module); 1751 module_exit(ct_cleanup_module); 1752 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>"); 1753 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>"); 1754 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>"); 1755 MODULE_DESCRIPTION("Connection tracking action"); 1756 MODULE_LICENSE("GPL v2"); 1757