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 <uapi/linux/netfilter/nf_nat.h> 36 37 static struct workqueue_struct *act_ct_wq; 38 static struct rhashtable zones_ht; 39 static DEFINE_MUTEX(zones_mutex); 40 41 struct tcf_ct_flow_table { 42 struct rhash_head node; /* In zones tables */ 43 44 struct rcu_work rwork; 45 struct nf_flowtable nf_ft; 46 refcount_t ref; 47 u16 zone; 48 49 bool dying; 50 }; 51 52 static const struct rhashtable_params zones_params = { 53 .head_offset = offsetof(struct tcf_ct_flow_table, node), 54 .key_offset = offsetof(struct tcf_ct_flow_table, zone), 55 .key_len = sizeof_field(struct tcf_ct_flow_table, zone), 56 .automatic_shrinking = true, 57 }; 58 59 static struct flow_action_entry * 60 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action) 61 { 62 int i = flow_action->num_entries++; 63 64 return &flow_action->entries[i]; 65 } 66 67 static void tcf_ct_add_mangle_action(struct flow_action *action, 68 enum flow_action_mangle_base htype, 69 u32 offset, 70 u32 mask, 71 u32 val) 72 { 73 struct flow_action_entry *entry; 74 75 entry = tcf_ct_flow_table_flow_action_get_next(action); 76 entry->id = FLOW_ACTION_MANGLE; 77 entry->mangle.htype = htype; 78 entry->mangle.mask = ~mask; 79 entry->mangle.offset = offset; 80 entry->mangle.val = val; 81 } 82 83 /* The following nat helper functions check if the inverted reverse tuple 84 * (target) is different then the current dir tuple - meaning nat for ports 85 * and/or ip is needed, and add the relevant mangle actions. 86 */ 87 static void 88 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple, 89 struct nf_conntrack_tuple target, 90 struct flow_action *action) 91 { 92 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 93 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 94 offsetof(struct iphdr, saddr), 95 0xFFFFFFFF, 96 be32_to_cpu(target.src.u3.ip)); 97 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 98 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4, 99 offsetof(struct iphdr, daddr), 100 0xFFFFFFFF, 101 be32_to_cpu(target.dst.u3.ip)); 102 } 103 104 static void 105 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action, 106 union nf_inet_addr *addr, 107 u32 offset) 108 { 109 int i; 110 111 for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++) 112 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6, 113 i * sizeof(u32) + offset, 114 0xFFFFFFFF, be32_to_cpu(addr->ip6[i])); 115 } 116 117 static void 118 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple, 119 struct nf_conntrack_tuple target, 120 struct flow_action *action) 121 { 122 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3))) 123 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3, 124 offsetof(struct ipv6hdr, 125 saddr)); 126 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3))) 127 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3, 128 offsetof(struct ipv6hdr, 129 daddr)); 130 } 131 132 static void 133 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple, 134 struct nf_conntrack_tuple target, 135 struct flow_action *action) 136 { 137 __be16 target_src = target.src.u.tcp.port; 138 __be16 target_dst = target.dst.u.tcp.port; 139 140 if (target_src != tuple->src.u.tcp.port) 141 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 142 offsetof(struct tcphdr, source), 143 0xFFFF, be16_to_cpu(target_src)); 144 if (target_dst != tuple->dst.u.tcp.port) 145 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP, 146 offsetof(struct tcphdr, dest), 147 0xFFFF, be16_to_cpu(target_dst)); 148 } 149 150 static void 151 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple, 152 struct nf_conntrack_tuple target, 153 struct flow_action *action) 154 { 155 __be16 target_src = target.src.u.udp.port; 156 __be16 target_dst = target.dst.u.udp.port; 157 158 if (target_src != tuple->src.u.udp.port) 159 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 160 offsetof(struct udphdr, source), 161 0xFFFF, be16_to_cpu(target_src)); 162 if (target_dst != tuple->dst.u.udp.port) 163 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP, 164 offsetof(struct udphdr, dest), 165 0xFFFF, be16_to_cpu(target_dst)); 166 } 167 168 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct, 169 enum ip_conntrack_dir dir, 170 struct flow_action *action) 171 { 172 struct nf_conn_labels *ct_labels; 173 struct flow_action_entry *entry; 174 enum ip_conntrack_info ctinfo; 175 u32 *act_ct_labels; 176 177 entry = tcf_ct_flow_table_flow_action_get_next(action); 178 entry->id = FLOW_ACTION_CT_METADATA; 179 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 180 entry->ct_metadata.mark = ct->mark; 181 #endif 182 ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED : 183 IP_CT_ESTABLISHED_REPLY; 184 /* aligns with the CT reference on the SKB nf_ct_set */ 185 entry->ct_metadata.cookie = (unsigned long)ct | ctinfo; 186 entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL; 187 188 act_ct_labels = entry->ct_metadata.labels; 189 ct_labels = nf_ct_labels_find(ct); 190 if (ct_labels) 191 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE); 192 else 193 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE); 194 } 195 196 static int tcf_ct_flow_table_add_action_nat(struct net *net, 197 struct nf_conn *ct, 198 enum ip_conntrack_dir dir, 199 struct flow_action *action) 200 { 201 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple; 202 struct nf_conntrack_tuple target; 203 204 if (!(ct->status & IPS_NAT_MASK)) 205 return 0; 206 207 nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple); 208 209 switch (tuple->src.l3num) { 210 case NFPROTO_IPV4: 211 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target, 212 action); 213 break; 214 case NFPROTO_IPV6: 215 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target, 216 action); 217 break; 218 default: 219 return -EOPNOTSUPP; 220 } 221 222 switch (nf_ct_protonum(ct)) { 223 case IPPROTO_TCP: 224 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action); 225 break; 226 case IPPROTO_UDP: 227 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action); 228 break; 229 default: 230 return -EOPNOTSUPP; 231 } 232 233 return 0; 234 } 235 236 static int tcf_ct_flow_table_fill_actions(struct net *net, 237 const struct flow_offload *flow, 238 enum flow_offload_tuple_dir tdir, 239 struct nf_flow_rule *flow_rule) 240 { 241 struct flow_action *action = &flow_rule->rule->action; 242 int num_entries = action->num_entries; 243 struct nf_conn *ct = flow->ct; 244 enum ip_conntrack_dir dir; 245 int i, err; 246 247 switch (tdir) { 248 case FLOW_OFFLOAD_DIR_ORIGINAL: 249 dir = IP_CT_DIR_ORIGINAL; 250 break; 251 case FLOW_OFFLOAD_DIR_REPLY: 252 dir = IP_CT_DIR_REPLY; 253 break; 254 default: 255 return -EOPNOTSUPP; 256 } 257 258 err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action); 259 if (err) 260 goto err_nat; 261 262 tcf_ct_flow_table_add_action_meta(ct, dir, action); 263 return 0; 264 265 err_nat: 266 /* Clear filled actions */ 267 for (i = num_entries; i < action->num_entries; i++) 268 memset(&action->entries[i], 0, sizeof(action->entries[i])); 269 action->num_entries = num_entries; 270 271 return err; 272 } 273 274 static struct nf_flowtable_type flowtable_ct = { 275 .action = tcf_ct_flow_table_fill_actions, 276 .owner = THIS_MODULE, 277 }; 278 279 static int tcf_ct_flow_table_get(struct tcf_ct_params *params) 280 { 281 struct tcf_ct_flow_table *ct_ft; 282 int err = -ENOMEM; 283 284 mutex_lock(&zones_mutex); 285 ct_ft = rhashtable_lookup_fast(&zones_ht, ¶ms->zone, zones_params); 286 if (ct_ft && refcount_inc_not_zero(&ct_ft->ref)) 287 goto out_unlock; 288 289 ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL); 290 if (!ct_ft) 291 goto err_alloc; 292 refcount_set(&ct_ft->ref, 1); 293 294 ct_ft->zone = params->zone; 295 err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params); 296 if (err) 297 goto err_insert; 298 299 ct_ft->nf_ft.type = &flowtable_ct; 300 ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD | 301 NF_FLOWTABLE_COUNTER; 302 err = nf_flow_table_init(&ct_ft->nf_ft); 303 if (err) 304 goto err_init; 305 306 __module_get(THIS_MODULE); 307 out_unlock: 308 params->ct_ft = ct_ft; 309 params->nf_ft = &ct_ft->nf_ft; 310 mutex_unlock(&zones_mutex); 311 312 return 0; 313 314 err_init: 315 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 316 err_insert: 317 kfree(ct_ft); 318 err_alloc: 319 mutex_unlock(&zones_mutex); 320 return err; 321 } 322 323 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work) 324 { 325 struct flow_block_cb *block_cb, *tmp_cb; 326 struct tcf_ct_flow_table *ct_ft; 327 struct flow_block *block; 328 329 ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table, 330 rwork); 331 nf_flow_table_free(&ct_ft->nf_ft); 332 333 /* Remove any remaining callbacks before cleanup */ 334 block = &ct_ft->nf_ft.flow_block; 335 down_write(&ct_ft->nf_ft.flow_block_lock); 336 list_for_each_entry_safe(block_cb, tmp_cb, &block->cb_list, list) { 337 list_del(&block_cb->list); 338 flow_block_cb_free(block_cb); 339 } 340 up_write(&ct_ft->nf_ft.flow_block_lock); 341 kfree(ct_ft); 342 343 module_put(THIS_MODULE); 344 } 345 346 static void tcf_ct_flow_table_put(struct tcf_ct_params *params) 347 { 348 struct tcf_ct_flow_table *ct_ft = params->ct_ft; 349 350 if (refcount_dec_and_test(¶ms->ct_ft->ref)) { 351 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params); 352 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work); 353 queue_rcu_work(act_ct_wq, &ct_ft->rwork); 354 } 355 } 356 357 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft, 358 struct nf_conn *ct, 359 bool tcp) 360 { 361 struct flow_offload *entry; 362 int err; 363 364 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status)) 365 return; 366 367 entry = flow_offload_alloc(ct); 368 if (!entry) { 369 WARN_ON_ONCE(1); 370 goto err_alloc; 371 } 372 373 if (tcp) { 374 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 375 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL; 376 } 377 378 err = flow_offload_add(&ct_ft->nf_ft, entry); 379 if (err) 380 goto err_add; 381 382 return; 383 384 err_add: 385 flow_offload_free(entry); 386 err_alloc: 387 clear_bit(IPS_OFFLOAD_BIT, &ct->status); 388 } 389 390 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft, 391 struct nf_conn *ct, 392 enum ip_conntrack_info ctinfo) 393 { 394 bool tcp = false; 395 396 if ((ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY) || 397 !test_bit(IPS_ASSURED_BIT, &ct->status)) 398 return; 399 400 switch (nf_ct_protonum(ct)) { 401 case IPPROTO_TCP: 402 tcp = true; 403 if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED) 404 return; 405 break; 406 case IPPROTO_UDP: 407 break; 408 default: 409 return; 410 } 411 412 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) || 413 ct->status & IPS_SEQ_ADJUST) 414 return; 415 416 tcf_ct_flow_table_add(ct_ft, ct, tcp); 417 } 418 419 static bool 420 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb, 421 struct flow_offload_tuple *tuple, 422 struct tcphdr **tcph) 423 { 424 struct flow_ports *ports; 425 unsigned int thoff; 426 struct iphdr *iph; 427 428 if (!pskb_network_may_pull(skb, sizeof(*iph))) 429 return false; 430 431 iph = ip_hdr(skb); 432 thoff = iph->ihl * 4; 433 434 if (ip_is_fragment(iph) || 435 unlikely(thoff != sizeof(struct iphdr))) 436 return false; 437 438 if (iph->protocol != IPPROTO_TCP && 439 iph->protocol != IPPROTO_UDP) 440 return false; 441 442 if (iph->ttl <= 1) 443 return false; 444 445 if (!pskb_network_may_pull(skb, iph->protocol == IPPROTO_TCP ? 446 thoff + sizeof(struct tcphdr) : 447 thoff + sizeof(*ports))) 448 return false; 449 450 iph = ip_hdr(skb); 451 if (iph->protocol == IPPROTO_TCP) 452 *tcph = (void *)(skb_network_header(skb) + thoff); 453 454 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 455 tuple->src_v4.s_addr = iph->saddr; 456 tuple->dst_v4.s_addr = iph->daddr; 457 tuple->src_port = ports->source; 458 tuple->dst_port = ports->dest; 459 tuple->l3proto = AF_INET; 460 tuple->l4proto = iph->protocol; 461 462 return true; 463 } 464 465 static bool 466 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb, 467 struct flow_offload_tuple *tuple, 468 struct tcphdr **tcph) 469 { 470 struct flow_ports *ports; 471 struct ipv6hdr *ip6h; 472 unsigned int thoff; 473 474 if (!pskb_network_may_pull(skb, sizeof(*ip6h))) 475 return false; 476 477 ip6h = ipv6_hdr(skb); 478 479 if (ip6h->nexthdr != IPPROTO_TCP && 480 ip6h->nexthdr != IPPROTO_UDP) 481 return false; 482 483 if (ip6h->hop_limit <= 1) 484 return false; 485 486 thoff = sizeof(*ip6h); 487 if (!pskb_network_may_pull(skb, ip6h->nexthdr == IPPROTO_TCP ? 488 thoff + sizeof(struct tcphdr) : 489 thoff + sizeof(*ports))) 490 return false; 491 492 ip6h = ipv6_hdr(skb); 493 if (ip6h->nexthdr == IPPROTO_TCP) 494 *tcph = (void *)(skb_network_header(skb) + thoff); 495 496 ports = (struct flow_ports *)(skb_network_header(skb) + thoff); 497 tuple->src_v6 = ip6h->saddr; 498 tuple->dst_v6 = ip6h->daddr; 499 tuple->src_port = ports->source; 500 tuple->dst_port = ports->dest; 501 tuple->l3proto = AF_INET6; 502 tuple->l4proto = ip6h->nexthdr; 503 504 return true; 505 } 506 507 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p, 508 struct sk_buff *skb, 509 u8 family) 510 { 511 struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft; 512 struct flow_offload_tuple_rhash *tuplehash; 513 struct flow_offload_tuple tuple = {}; 514 enum ip_conntrack_info ctinfo; 515 struct tcphdr *tcph = NULL; 516 struct flow_offload *flow; 517 struct nf_conn *ct; 518 u8 dir; 519 520 /* Previously seen or loopback */ 521 ct = nf_ct_get(skb, &ctinfo); 522 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED) 523 return false; 524 525 switch (family) { 526 case NFPROTO_IPV4: 527 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph)) 528 return false; 529 break; 530 case NFPROTO_IPV6: 531 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph)) 532 return false; 533 break; 534 default: 535 return false; 536 } 537 538 tuplehash = flow_offload_lookup(nf_ft, &tuple); 539 if (!tuplehash) 540 return false; 541 542 dir = tuplehash->tuple.dir; 543 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]); 544 ct = flow->ct; 545 546 if (tcph && (unlikely(tcph->fin || tcph->rst))) { 547 flow_offload_teardown(flow); 548 return false; 549 } 550 551 ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED : 552 IP_CT_ESTABLISHED_REPLY; 553 554 flow_offload_refresh(nf_ft, flow); 555 nf_conntrack_get(&ct->ct_general); 556 nf_ct_set(skb, ct, ctinfo); 557 if (nf_ft->flags & NF_FLOWTABLE_COUNTER) 558 nf_ct_acct_update(ct, dir, skb->len); 559 560 return true; 561 } 562 563 static int tcf_ct_flow_tables_init(void) 564 { 565 return rhashtable_init(&zones_ht, &zones_params); 566 } 567 568 static void tcf_ct_flow_tables_uninit(void) 569 { 570 rhashtable_destroy(&zones_ht); 571 } 572 573 static struct tc_action_ops act_ct_ops; 574 static unsigned int ct_net_id; 575 576 struct tc_ct_action_net { 577 struct tc_action_net tn; /* Must be first */ 578 bool labels; 579 }; 580 581 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */ 582 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb, 583 u16 zone_id, bool force) 584 { 585 enum ip_conntrack_info ctinfo; 586 struct nf_conn *ct; 587 588 ct = nf_ct_get(skb, &ctinfo); 589 if (!ct) 590 return false; 591 if (!net_eq(net, read_pnet(&ct->ct_net))) 592 return false; 593 if (nf_ct_zone(ct)->id != zone_id) 594 return false; 595 596 /* Force conntrack entry direction. */ 597 if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) { 598 if (nf_ct_is_confirmed(ct)) 599 nf_ct_kill(ct); 600 601 nf_conntrack_put(&ct->ct_general); 602 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 603 604 return false; 605 } 606 607 return true; 608 } 609 610 /* Trim the skb to the length specified by the IP/IPv6 header, 611 * removing any trailing lower-layer padding. This prepares the skb 612 * for higher-layer processing that assumes skb->len excludes padding 613 * (such as nf_ip_checksum). The caller needs to pull the skb to the 614 * network header, and ensure ip_hdr/ipv6_hdr points to valid data. 615 */ 616 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family) 617 { 618 unsigned int len; 619 int err; 620 621 switch (family) { 622 case NFPROTO_IPV4: 623 len = ntohs(ip_hdr(skb)->tot_len); 624 break; 625 case NFPROTO_IPV6: 626 len = sizeof(struct ipv6hdr) 627 + ntohs(ipv6_hdr(skb)->payload_len); 628 break; 629 default: 630 len = skb->len; 631 } 632 633 err = pskb_trim_rcsum(skb, len); 634 635 return err; 636 } 637 638 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb) 639 { 640 u8 family = NFPROTO_UNSPEC; 641 642 switch (skb_protocol(skb, true)) { 643 case htons(ETH_P_IP): 644 family = NFPROTO_IPV4; 645 break; 646 case htons(ETH_P_IPV6): 647 family = NFPROTO_IPV6; 648 break; 649 default: 650 break; 651 } 652 653 return family; 654 } 655 656 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag) 657 { 658 unsigned int len; 659 660 len = skb_network_offset(skb) + sizeof(struct iphdr); 661 if (unlikely(skb->len < len)) 662 return -EINVAL; 663 if (unlikely(!pskb_may_pull(skb, len))) 664 return -ENOMEM; 665 666 *frag = ip_is_fragment(ip_hdr(skb)); 667 return 0; 668 } 669 670 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag) 671 { 672 unsigned int flags = 0, len, payload_ofs = 0; 673 unsigned short frag_off; 674 int nexthdr; 675 676 len = skb_network_offset(skb) + sizeof(struct ipv6hdr); 677 if (unlikely(skb->len < len)) 678 return -EINVAL; 679 if (unlikely(!pskb_may_pull(skb, len))) 680 return -ENOMEM; 681 682 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags); 683 if (unlikely(nexthdr < 0)) 684 return -EPROTO; 685 686 *frag = flags & IP6_FH_F_FRAG; 687 return 0; 688 } 689 690 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb, 691 u8 family, u16 zone, bool *defrag) 692 { 693 enum ip_conntrack_info ctinfo; 694 struct qdisc_skb_cb cb; 695 struct nf_conn *ct; 696 int err = 0; 697 bool frag; 698 699 /* Previously seen (loopback)? Ignore. */ 700 ct = nf_ct_get(skb, &ctinfo); 701 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED) 702 return 0; 703 704 if (family == NFPROTO_IPV4) 705 err = tcf_ct_ipv4_is_fragment(skb, &frag); 706 else 707 err = tcf_ct_ipv6_is_fragment(skb, &frag); 708 if (err || !frag) 709 return err; 710 711 skb_get(skb); 712 cb = *qdisc_skb_cb(skb); 713 714 if (family == NFPROTO_IPV4) { 715 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone; 716 717 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 718 local_bh_disable(); 719 err = ip_defrag(net, skb, user); 720 local_bh_enable(); 721 if (err && err != -EINPROGRESS) 722 return err; 723 724 if (!err) { 725 *defrag = true; 726 cb.mru = IPCB(skb)->frag_max_size; 727 } 728 } else { /* NFPROTO_IPV6 */ 729 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) 730 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone; 731 732 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm)); 733 err = nf_ct_frag6_gather(net, skb, user); 734 if (err && err != -EINPROGRESS) 735 goto out_free; 736 737 if (!err) { 738 *defrag = true; 739 cb.mru = IP6CB(skb)->frag_max_size; 740 } 741 #else 742 err = -EOPNOTSUPP; 743 goto out_free; 744 #endif 745 } 746 747 if (err != -EINPROGRESS) 748 *qdisc_skb_cb(skb) = cb; 749 skb_clear_hash(skb); 750 skb->ignore_df = 1; 751 return err; 752 753 out_free: 754 kfree_skb(skb); 755 return err; 756 } 757 758 static void tcf_ct_params_free(struct rcu_head *head) 759 { 760 struct tcf_ct_params *params = container_of(head, 761 struct tcf_ct_params, rcu); 762 763 tcf_ct_flow_table_put(params); 764 765 if (params->tmpl) 766 nf_conntrack_put(¶ms->tmpl->ct_general); 767 kfree(params); 768 } 769 770 #if IS_ENABLED(CONFIG_NF_NAT) 771 /* Modelled after nf_nat_ipv[46]_fn(). 772 * range is only used for new, uninitialized NAT state. 773 * Returns either NF_ACCEPT or NF_DROP. 774 */ 775 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct, 776 enum ip_conntrack_info ctinfo, 777 const struct nf_nat_range2 *range, 778 enum nf_nat_manip_type maniptype) 779 { 780 __be16 proto = skb_protocol(skb, true); 781 int hooknum, err = NF_ACCEPT; 782 783 /* See HOOK2MANIP(). */ 784 if (maniptype == NF_NAT_MANIP_SRC) 785 hooknum = NF_INET_LOCAL_IN; /* Source NAT */ 786 else 787 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */ 788 789 switch (ctinfo) { 790 case IP_CT_RELATED: 791 case IP_CT_RELATED_REPLY: 792 if (proto == htons(ETH_P_IP) && 793 ip_hdr(skb)->protocol == IPPROTO_ICMP) { 794 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo, 795 hooknum)) 796 err = NF_DROP; 797 goto out; 798 } else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) { 799 __be16 frag_off; 800 u8 nexthdr = ipv6_hdr(skb)->nexthdr; 801 int hdrlen = ipv6_skip_exthdr(skb, 802 sizeof(struct ipv6hdr), 803 &nexthdr, &frag_off); 804 805 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) { 806 if (!nf_nat_icmpv6_reply_translation(skb, ct, 807 ctinfo, 808 hooknum, 809 hdrlen)) 810 err = NF_DROP; 811 goto out; 812 } 813 } 814 /* Non-ICMP, fall thru to initialize if needed. */ 815 fallthrough; 816 case IP_CT_NEW: 817 /* Seen it before? This can happen for loopback, retrans, 818 * or local packets. 819 */ 820 if (!nf_nat_initialized(ct, maniptype)) { 821 /* Initialize according to the NAT action. */ 822 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS) 823 /* Action is set up to establish a new 824 * mapping. 825 */ 826 ? nf_nat_setup_info(ct, range, maniptype) 827 : nf_nat_alloc_null_binding(ct, hooknum); 828 if (err != NF_ACCEPT) 829 goto out; 830 } 831 break; 832 833 case IP_CT_ESTABLISHED: 834 case IP_CT_ESTABLISHED_REPLY: 835 break; 836 837 default: 838 err = NF_DROP; 839 goto out; 840 } 841 842 err = nf_nat_packet(ct, ctinfo, hooknum, skb); 843 out: 844 return err; 845 } 846 #endif /* CONFIG_NF_NAT */ 847 848 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask) 849 { 850 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) 851 u32 new_mark; 852 853 if (!mask) 854 return; 855 856 new_mark = mark | (ct->mark & ~(mask)); 857 if (ct->mark != new_mark) { 858 ct->mark = new_mark; 859 if (nf_ct_is_confirmed(ct)) 860 nf_conntrack_event_cache(IPCT_MARK, ct); 861 } 862 #endif 863 } 864 865 static void tcf_ct_act_set_labels(struct nf_conn *ct, 866 u32 *labels, 867 u32 *labels_m) 868 { 869 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) 870 size_t labels_sz = sizeof_field(struct tcf_ct_params, labels); 871 872 if (!memchr_inv(labels_m, 0, labels_sz)) 873 return; 874 875 nf_connlabels_replace(ct, labels, labels_m, 4); 876 #endif 877 } 878 879 static int tcf_ct_act_nat(struct sk_buff *skb, 880 struct nf_conn *ct, 881 enum ip_conntrack_info ctinfo, 882 int ct_action, 883 struct nf_nat_range2 *range, 884 bool commit) 885 { 886 #if IS_ENABLED(CONFIG_NF_NAT) 887 int err; 888 enum nf_nat_manip_type maniptype; 889 890 if (!(ct_action & TCA_CT_ACT_NAT)) 891 return NF_ACCEPT; 892 893 /* Add NAT extension if not confirmed yet. */ 894 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct)) 895 return NF_DROP; /* Can't NAT. */ 896 897 if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) && 898 (ctinfo != IP_CT_RELATED || commit)) { 899 /* NAT an established or related connection like before. */ 900 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) 901 /* This is the REPLY direction for a connection 902 * for which NAT was applied in the forward 903 * direction. Do the reverse NAT. 904 */ 905 maniptype = ct->status & IPS_SRC_NAT 906 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC; 907 else 908 maniptype = ct->status & IPS_SRC_NAT 909 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST; 910 } else if (ct_action & TCA_CT_ACT_NAT_SRC) { 911 maniptype = NF_NAT_MANIP_SRC; 912 } else if (ct_action & TCA_CT_ACT_NAT_DST) { 913 maniptype = NF_NAT_MANIP_DST; 914 } else { 915 return NF_ACCEPT; 916 } 917 918 err = ct_nat_execute(skb, ct, ctinfo, range, maniptype); 919 if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) { 920 if (ct->status & IPS_SRC_NAT) { 921 if (maniptype == NF_NAT_MANIP_SRC) 922 maniptype = NF_NAT_MANIP_DST; 923 else 924 maniptype = NF_NAT_MANIP_SRC; 925 926 err = ct_nat_execute(skb, ct, ctinfo, range, 927 maniptype); 928 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) { 929 err = ct_nat_execute(skb, ct, ctinfo, NULL, 930 NF_NAT_MANIP_SRC); 931 } 932 } 933 return err; 934 #else 935 return NF_ACCEPT; 936 #endif 937 } 938 939 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a, 940 struct tcf_result *res) 941 { 942 struct net *net = dev_net(skb->dev); 943 bool cached, commit, clear, force; 944 enum ip_conntrack_info ctinfo; 945 struct tcf_ct *c = to_ct(a); 946 struct nf_conn *tmpl = NULL; 947 struct nf_hook_state state; 948 int nh_ofs, err, retval; 949 struct tcf_ct_params *p; 950 bool skip_add = false; 951 bool defrag = false; 952 struct nf_conn *ct; 953 u8 family; 954 955 p = rcu_dereference_bh(c->params); 956 957 retval = READ_ONCE(c->tcf_action); 958 commit = p->ct_action & TCA_CT_ACT_COMMIT; 959 clear = p->ct_action & TCA_CT_ACT_CLEAR; 960 force = p->ct_action & TCA_CT_ACT_FORCE; 961 tmpl = p->tmpl; 962 963 tcf_lastuse_update(&c->tcf_tm); 964 tcf_action_update_bstats(&c->common, skb); 965 966 if (clear) { 967 qdisc_skb_cb(skb)->post_ct = false; 968 ct = nf_ct_get(skb, &ctinfo); 969 if (ct) { 970 nf_conntrack_put(&ct->ct_general); 971 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 972 } 973 974 goto out_clear; 975 } 976 977 family = tcf_ct_skb_nf_family(skb); 978 if (family == NFPROTO_UNSPEC) 979 goto drop; 980 981 /* The conntrack module expects to be working at L3. 982 * We also try to pull the IPv4/6 header to linear area 983 */ 984 nh_ofs = skb_network_offset(skb); 985 skb_pull_rcsum(skb, nh_ofs); 986 err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag); 987 if (err == -EINPROGRESS) { 988 retval = TC_ACT_STOLEN; 989 goto out_clear; 990 } 991 if (err) 992 goto drop; 993 994 err = tcf_ct_skb_network_trim(skb, family); 995 if (err) 996 goto drop; 997 998 /* If we are recirculating packets to match on ct fields and 999 * committing with a separate ct action, then we don't need to 1000 * actually run the packet through conntrack twice unless it's for a 1001 * different zone. 1002 */ 1003 cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force); 1004 if (!cached) { 1005 if (tcf_ct_flow_table_lookup(p, skb, family)) { 1006 skip_add = true; 1007 goto do_nat; 1008 } 1009 1010 /* Associate skb with specified zone. */ 1011 if (tmpl) { 1012 nf_conntrack_put(skb_nfct(skb)); 1013 nf_conntrack_get(&tmpl->ct_general); 1014 nf_ct_set(skb, tmpl, IP_CT_NEW); 1015 } 1016 1017 state.hook = NF_INET_PRE_ROUTING; 1018 state.net = net; 1019 state.pf = family; 1020 err = nf_conntrack_in(skb, &state); 1021 if (err != NF_ACCEPT) 1022 goto out_push; 1023 } 1024 1025 do_nat: 1026 ct = nf_ct_get(skb, &ctinfo); 1027 if (!ct) 1028 goto out_push; 1029 nf_ct_deliver_cached_events(ct); 1030 1031 err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit); 1032 if (err != NF_ACCEPT) 1033 goto drop; 1034 1035 if (commit) { 1036 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask); 1037 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask); 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 qdisc_skb_cb(skb)->post_ct = true; 1053 out_clear: 1054 if (defrag) 1055 qdisc_skb_cb(skb)->pkt_len = skb->len; 1056 return retval; 1057 1058 drop: 1059 tcf_action_inc_drop_qstats(&c->common); 1060 return TC_ACT_SHOT; 1061 } 1062 1063 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = { 1064 [TCA_CT_ACTION] = { .type = NLA_U16 }, 1065 [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)), 1066 [TCA_CT_ZONE] = { .type = NLA_U16 }, 1067 [TCA_CT_MARK] = { .type = NLA_U32 }, 1068 [TCA_CT_MARK_MASK] = { .type = NLA_U32 }, 1069 [TCA_CT_LABELS] = { .type = NLA_BINARY, 1070 .len = 128 / BITS_PER_BYTE }, 1071 [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY, 1072 .len = 128 / BITS_PER_BYTE }, 1073 [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 }, 1074 [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 }, 1075 [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1076 [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1077 [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 }, 1078 [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 }, 1079 }; 1080 1081 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p, 1082 struct tc_ct *parm, 1083 struct nlattr **tb, 1084 struct netlink_ext_ack *extack) 1085 { 1086 struct nf_nat_range2 *range; 1087 1088 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1089 return 0; 1090 1091 if (!IS_ENABLED(CONFIG_NF_NAT)) { 1092 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel"); 1093 return -EOPNOTSUPP; 1094 } 1095 1096 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1097 return 0; 1098 1099 if ((p->ct_action & TCA_CT_ACT_NAT_SRC) && 1100 (p->ct_action & TCA_CT_ACT_NAT_DST)) { 1101 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time"); 1102 return -EOPNOTSUPP; 1103 } 1104 1105 range = &p->range; 1106 if (tb[TCA_CT_NAT_IPV4_MIN]) { 1107 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX]; 1108 1109 p->ipv4_range = true; 1110 range->flags |= NF_NAT_RANGE_MAP_IPS; 1111 range->min_addr.ip = 1112 nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]); 1113 1114 range->max_addr.ip = max_attr ? 1115 nla_get_in_addr(max_attr) : 1116 range->min_addr.ip; 1117 } else if (tb[TCA_CT_NAT_IPV6_MIN]) { 1118 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX]; 1119 1120 p->ipv4_range = false; 1121 range->flags |= NF_NAT_RANGE_MAP_IPS; 1122 range->min_addr.in6 = 1123 nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]); 1124 1125 range->max_addr.in6 = max_attr ? 1126 nla_get_in6_addr(max_attr) : 1127 range->min_addr.in6; 1128 } 1129 1130 if (tb[TCA_CT_NAT_PORT_MIN]) { 1131 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED; 1132 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]); 1133 1134 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ? 1135 nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) : 1136 range->min_proto.all; 1137 } 1138 1139 return 0; 1140 } 1141 1142 static void tcf_ct_set_key_val(struct nlattr **tb, 1143 void *val, int val_type, 1144 void *mask, int mask_type, 1145 int len) 1146 { 1147 if (!tb[val_type]) 1148 return; 1149 nla_memcpy(val, tb[val_type], len); 1150 1151 if (!mask) 1152 return; 1153 1154 if (mask_type == TCA_CT_UNSPEC || !tb[mask_type]) 1155 memset(mask, 0xff, len); 1156 else 1157 nla_memcpy(mask, tb[mask_type], len); 1158 } 1159 1160 static int tcf_ct_fill_params(struct net *net, 1161 struct tcf_ct_params *p, 1162 struct tc_ct *parm, 1163 struct nlattr **tb, 1164 struct netlink_ext_ack *extack) 1165 { 1166 struct tc_ct_action_net *tn = net_generic(net, ct_net_id); 1167 struct nf_conntrack_zone zone; 1168 struct nf_conn *tmpl; 1169 int err; 1170 1171 p->zone = NF_CT_DEFAULT_ZONE_ID; 1172 1173 tcf_ct_set_key_val(tb, 1174 &p->ct_action, TCA_CT_ACTION, 1175 NULL, TCA_CT_UNSPEC, 1176 sizeof(p->ct_action)); 1177 1178 if (p->ct_action & TCA_CT_ACT_CLEAR) 1179 return 0; 1180 1181 err = tcf_ct_fill_params_nat(p, parm, tb, extack); 1182 if (err) 1183 return err; 1184 1185 if (tb[TCA_CT_MARK]) { 1186 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) { 1187 NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled."); 1188 return -EOPNOTSUPP; 1189 } 1190 tcf_ct_set_key_val(tb, 1191 &p->mark, TCA_CT_MARK, 1192 &p->mark_mask, TCA_CT_MARK_MASK, 1193 sizeof(p->mark)); 1194 } 1195 1196 if (tb[TCA_CT_LABELS]) { 1197 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) { 1198 NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled."); 1199 return -EOPNOTSUPP; 1200 } 1201 1202 if (!tn->labels) { 1203 NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length"); 1204 return -EOPNOTSUPP; 1205 } 1206 tcf_ct_set_key_val(tb, 1207 p->labels, TCA_CT_LABELS, 1208 p->labels_mask, TCA_CT_LABELS_MASK, 1209 sizeof(p->labels)); 1210 } 1211 1212 if (tb[TCA_CT_ZONE]) { 1213 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) { 1214 NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled."); 1215 return -EOPNOTSUPP; 1216 } 1217 1218 tcf_ct_set_key_val(tb, 1219 &p->zone, TCA_CT_ZONE, 1220 NULL, TCA_CT_UNSPEC, 1221 sizeof(p->zone)); 1222 } 1223 1224 nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0); 1225 tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL); 1226 if (!tmpl) { 1227 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template"); 1228 return -ENOMEM; 1229 } 1230 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status); 1231 nf_conntrack_get(&tmpl->ct_general); 1232 p->tmpl = tmpl; 1233 1234 return 0; 1235 } 1236 1237 static int tcf_ct_init(struct net *net, struct nlattr *nla, 1238 struct nlattr *est, struct tc_action **a, 1239 struct tcf_proto *tp, u32 flags, 1240 struct netlink_ext_ack *extack) 1241 { 1242 struct tc_action_net *tn = net_generic(net, ct_net_id); 1243 bool bind = flags & TCA_ACT_FLAGS_BIND; 1244 struct tcf_ct_params *params = NULL; 1245 struct nlattr *tb[TCA_CT_MAX + 1]; 1246 struct tcf_chain *goto_ch = NULL; 1247 struct tc_ct *parm; 1248 struct tcf_ct *c; 1249 int err, res = 0; 1250 u32 index; 1251 1252 if (!nla) { 1253 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed"); 1254 return -EINVAL; 1255 } 1256 1257 err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack); 1258 if (err < 0) 1259 return err; 1260 1261 if (!tb[TCA_CT_PARMS]) { 1262 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters"); 1263 return -EINVAL; 1264 } 1265 parm = nla_data(tb[TCA_CT_PARMS]); 1266 index = parm->index; 1267 err = tcf_idr_check_alloc(tn, &index, a, bind); 1268 if (err < 0) 1269 return err; 1270 1271 if (!err) { 1272 err = tcf_idr_create_from_flags(tn, index, est, a, 1273 &act_ct_ops, bind, flags); 1274 if (err) { 1275 tcf_idr_cleanup(tn, index); 1276 return err; 1277 } 1278 res = ACT_P_CREATED; 1279 } else { 1280 if (bind) 1281 return 0; 1282 1283 if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 1284 tcf_idr_release(*a, bind); 1285 return -EEXIST; 1286 } 1287 } 1288 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 1289 if (err < 0) 1290 goto cleanup; 1291 1292 c = to_ct(*a); 1293 1294 params = kzalloc(sizeof(*params), GFP_KERNEL); 1295 if (unlikely(!params)) { 1296 err = -ENOMEM; 1297 goto cleanup; 1298 } 1299 1300 err = tcf_ct_fill_params(net, params, parm, tb, extack); 1301 if (err) 1302 goto cleanup; 1303 1304 err = tcf_ct_flow_table_get(params); 1305 if (err) 1306 goto cleanup; 1307 1308 spin_lock_bh(&c->tcf_lock); 1309 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 1310 params = rcu_replace_pointer(c->params, params, 1311 lockdep_is_held(&c->tcf_lock)); 1312 spin_unlock_bh(&c->tcf_lock); 1313 1314 if (goto_ch) 1315 tcf_chain_put_by_act(goto_ch); 1316 if (params) 1317 call_rcu(¶ms->rcu, tcf_ct_params_free); 1318 1319 return res; 1320 1321 cleanup: 1322 if (goto_ch) 1323 tcf_chain_put_by_act(goto_ch); 1324 kfree(params); 1325 tcf_idr_release(*a, bind); 1326 return err; 1327 } 1328 1329 static void tcf_ct_cleanup(struct tc_action *a) 1330 { 1331 struct tcf_ct_params *params; 1332 struct tcf_ct *c = to_ct(a); 1333 1334 params = rcu_dereference_protected(c->params, 1); 1335 if (params) 1336 call_rcu(¶ms->rcu, tcf_ct_params_free); 1337 } 1338 1339 static int tcf_ct_dump_key_val(struct sk_buff *skb, 1340 void *val, int val_type, 1341 void *mask, int mask_type, 1342 int len) 1343 { 1344 int err; 1345 1346 if (mask && !memchr_inv(mask, 0, len)) 1347 return 0; 1348 1349 err = nla_put(skb, val_type, len, val); 1350 if (err) 1351 return err; 1352 1353 if (mask_type != TCA_CT_UNSPEC) { 1354 err = nla_put(skb, mask_type, len, mask); 1355 if (err) 1356 return err; 1357 } 1358 1359 return 0; 1360 } 1361 1362 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p) 1363 { 1364 struct nf_nat_range2 *range = &p->range; 1365 1366 if (!(p->ct_action & TCA_CT_ACT_NAT)) 1367 return 0; 1368 1369 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST))) 1370 return 0; 1371 1372 if (range->flags & NF_NAT_RANGE_MAP_IPS) { 1373 if (p->ipv4_range) { 1374 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN, 1375 range->min_addr.ip)) 1376 return -1; 1377 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX, 1378 range->max_addr.ip)) 1379 return -1; 1380 } else { 1381 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN, 1382 &range->min_addr.in6)) 1383 return -1; 1384 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX, 1385 &range->max_addr.in6)) 1386 return -1; 1387 } 1388 } 1389 1390 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) { 1391 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN, 1392 range->min_proto.all)) 1393 return -1; 1394 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX, 1395 range->max_proto.all)) 1396 return -1; 1397 } 1398 1399 return 0; 1400 } 1401 1402 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a, 1403 int bind, int ref) 1404 { 1405 unsigned char *b = skb_tail_pointer(skb); 1406 struct tcf_ct *c = to_ct(a); 1407 struct tcf_ct_params *p; 1408 1409 struct tc_ct opt = { 1410 .index = c->tcf_index, 1411 .refcnt = refcount_read(&c->tcf_refcnt) - ref, 1412 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind, 1413 }; 1414 struct tcf_t t; 1415 1416 spin_lock_bh(&c->tcf_lock); 1417 p = rcu_dereference_protected(c->params, 1418 lockdep_is_held(&c->tcf_lock)); 1419 opt.action = c->tcf_action; 1420 1421 if (tcf_ct_dump_key_val(skb, 1422 &p->ct_action, TCA_CT_ACTION, 1423 NULL, TCA_CT_UNSPEC, 1424 sizeof(p->ct_action))) 1425 goto nla_put_failure; 1426 1427 if (p->ct_action & TCA_CT_ACT_CLEAR) 1428 goto skip_dump; 1429 1430 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && 1431 tcf_ct_dump_key_val(skb, 1432 &p->mark, TCA_CT_MARK, 1433 &p->mark_mask, TCA_CT_MARK_MASK, 1434 sizeof(p->mark))) 1435 goto nla_put_failure; 1436 1437 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && 1438 tcf_ct_dump_key_val(skb, 1439 p->labels, TCA_CT_LABELS, 1440 p->labels_mask, TCA_CT_LABELS_MASK, 1441 sizeof(p->labels))) 1442 goto nla_put_failure; 1443 1444 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && 1445 tcf_ct_dump_key_val(skb, 1446 &p->zone, TCA_CT_ZONE, 1447 NULL, TCA_CT_UNSPEC, 1448 sizeof(p->zone))) 1449 goto nla_put_failure; 1450 1451 if (tcf_ct_dump_nat(skb, p)) 1452 goto nla_put_failure; 1453 1454 skip_dump: 1455 if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt)) 1456 goto nla_put_failure; 1457 1458 tcf_tm_dump(&t, &c->tcf_tm); 1459 if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD)) 1460 goto nla_put_failure; 1461 spin_unlock_bh(&c->tcf_lock); 1462 1463 return skb->len; 1464 nla_put_failure: 1465 spin_unlock_bh(&c->tcf_lock); 1466 nlmsg_trim(skb, b); 1467 return -1; 1468 } 1469 1470 static int tcf_ct_walker(struct net *net, struct sk_buff *skb, 1471 struct netlink_callback *cb, int type, 1472 const struct tc_action_ops *ops, 1473 struct netlink_ext_ack *extack) 1474 { 1475 struct tc_action_net *tn = net_generic(net, ct_net_id); 1476 1477 return tcf_generic_walker(tn, skb, cb, type, ops, extack); 1478 } 1479 1480 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index) 1481 { 1482 struct tc_action_net *tn = net_generic(net, ct_net_id); 1483 1484 return tcf_idr_search(tn, a, index); 1485 } 1486 1487 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 1488 u64 drops, u64 lastuse, bool hw) 1489 { 1490 struct tcf_ct *c = to_ct(a); 1491 1492 tcf_action_update_stats(a, bytes, packets, drops, hw); 1493 c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse); 1494 } 1495 1496 static struct tc_action_ops act_ct_ops = { 1497 .kind = "ct", 1498 .id = TCA_ID_CT, 1499 .owner = THIS_MODULE, 1500 .act = tcf_ct_act, 1501 .dump = tcf_ct_dump, 1502 .init = tcf_ct_init, 1503 .cleanup = tcf_ct_cleanup, 1504 .walk = tcf_ct_walker, 1505 .lookup = tcf_ct_search, 1506 .stats_update = tcf_stats_update, 1507 .size = sizeof(struct tcf_ct), 1508 }; 1509 1510 static __net_init int ct_init_net(struct net *net) 1511 { 1512 unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8; 1513 struct tc_ct_action_net *tn = net_generic(net, ct_net_id); 1514 1515 if (nf_connlabels_get(net, n_bits - 1)) { 1516 tn->labels = false; 1517 pr_err("act_ct: Failed to set connlabels length"); 1518 } else { 1519 tn->labels = true; 1520 } 1521 1522 return tc_action_net_init(net, &tn->tn, &act_ct_ops); 1523 } 1524 1525 static void __net_exit ct_exit_net(struct list_head *net_list) 1526 { 1527 struct net *net; 1528 1529 rtnl_lock(); 1530 list_for_each_entry(net, net_list, exit_list) { 1531 struct tc_ct_action_net *tn = net_generic(net, ct_net_id); 1532 1533 if (tn->labels) 1534 nf_connlabels_put(net); 1535 } 1536 rtnl_unlock(); 1537 1538 tc_action_net_exit(net_list, ct_net_id); 1539 } 1540 1541 static struct pernet_operations ct_net_ops = { 1542 .init = ct_init_net, 1543 .exit_batch = ct_exit_net, 1544 .id = &ct_net_id, 1545 .size = sizeof(struct tc_ct_action_net), 1546 }; 1547 1548 static int __init ct_init_module(void) 1549 { 1550 int err; 1551 1552 act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0); 1553 if (!act_ct_wq) 1554 return -ENOMEM; 1555 1556 err = tcf_ct_flow_tables_init(); 1557 if (err) 1558 goto err_tbl_init; 1559 1560 err = tcf_register_action(&act_ct_ops, &ct_net_ops); 1561 if (err) 1562 goto err_register; 1563 1564 static_branch_inc(&tcf_frag_xmit_count); 1565 1566 return 0; 1567 1568 err_register: 1569 tcf_ct_flow_tables_uninit(); 1570 err_tbl_init: 1571 destroy_workqueue(act_ct_wq); 1572 return err; 1573 } 1574 1575 static void __exit ct_cleanup_module(void) 1576 { 1577 static_branch_dec(&tcf_frag_xmit_count); 1578 tcf_unregister_action(&act_ct_ops, &ct_net_ops); 1579 tcf_ct_flow_tables_uninit(); 1580 destroy_workqueue(act_ct_wq); 1581 } 1582 1583 module_init(ct_init_module); 1584 module_exit(ct_cleanup_module); 1585 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>"); 1586 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>"); 1587 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>"); 1588 MODULE_DESCRIPTION("Connection tracking action"); 1589 MODULE_LICENSE("GPL v2"); 1590