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