net/sched/act_ct.c

   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, &params->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         bool labels;
 703 };
 704
 705 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 706 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
 707                                    struct tcf_ct_params *p)
 708 {
 709         enum ip_conntrack_info ctinfo;
 710         struct nf_conn *ct;
 711
 712         ct = nf_ct_get(skb, &ctinfo);
 713         if (!ct)
 714                 return false;
 715         if (!net_eq(net, read_pnet(&ct->ct_net)))
 716                 goto drop_ct;
 717         if (nf_ct_zone(ct)->id != p->zone)
 718                 goto drop_ct;
 719         if (p->helper) {
 720                 struct nf_conn_help *help;
 721
 722                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 723                 if (help && rcu_access_pointer(help->helper) != p->helper)
 724                         goto drop_ct;
 725         }
 726
 727         /* Force conntrack entry direction. */
 728         if ((p->ct_action & TCA_CT_ACT_FORCE) &&
 729             CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 730                 if (nf_ct_is_confirmed(ct))
 731                         nf_ct_kill(ct);
 732
 733                 goto drop_ct;
 734         }
 735
 736         return true;
 737
 738 drop_ct:
 739         nf_ct_put(ct);
 740         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 741
 742         return false;
 743 }
 744
 745 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
 746 {
 747         u8 family = NFPROTO_UNSPEC;
 748
 749         switch (skb_protocol(skb, true)) {
 750         case htons(ETH_P_IP):
 751                 family = NFPROTO_IPV4;
 752                 break;
 753         case htons(ETH_P_IPV6):
 754                 family = NFPROTO_IPV6;
 755                 break;
 756         default:
 757                 break;
 758         }
 759
 760         return family;
 761 }
 762
 763 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
 764 {
 765         unsigned int len;
 766
 767         len =  skb_network_offset(skb) + sizeof(struct iphdr);
 768         if (unlikely(skb->len < len))
 769                 return -EINVAL;
 770         if (unlikely(!pskb_may_pull(skb, len)))
 771                 return -ENOMEM;
 772
 773         *frag = ip_is_fragment(ip_hdr(skb));
 774         return 0;
 775 }
 776
 777 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
 778 {
 779         unsigned int flags = 0, len, payload_ofs = 0;
 780         unsigned short frag_off;
 781         int nexthdr;
 782
 783         len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
 784         if (unlikely(skb->len < len))
 785                 return -EINVAL;
 786         if (unlikely(!pskb_may_pull(skb, len)))
 787                 return -ENOMEM;
 788
 789         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 790         if (unlikely(nexthdr < 0))
 791                 return -EPROTO;
 792
 793         *frag = flags & IP6_FH_F_FRAG;
 794         return 0;
 795 }
 796
 797 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
 798                                    u8 family, u16 zone, bool *defrag)
 799 {
 800         enum ip_conntrack_info ctinfo;
 801         struct nf_conn *ct;
 802         int err = 0;
 803         bool frag;
 804         u8 proto;
 805         u16 mru;
 806
 807         /* Previously seen (loopback)? Ignore. */
 808         ct = nf_ct_get(skb, &ctinfo);
 809         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 810                 return 0;
 811
 812         if (family == NFPROTO_IPV4)
 813                 err = tcf_ct_ipv4_is_fragment(skb, &frag);
 814         else
 815                 err = tcf_ct_ipv6_is_fragment(skb, &frag);
 816         if (err || !frag)
 817                 return err;
 818
 819         skb_get(skb);
 820         err = nf_ct_handle_fragments(net, skb, zone, family, &proto, &mru);
 821         if (err)
 822                 return err;
 823
 824         *defrag = true;
 825         tc_skb_cb(skb)->mru = mru;
 826
 827         return 0;
 828 }
 829
 830 static void tcf_ct_params_free(struct tcf_ct_params *params)
 831 {
 832         if (params->helper) {
 833 #if IS_ENABLED(CONFIG_NF_NAT)
 834                 if (params->ct_action & TCA_CT_ACT_NAT)
 835                         nf_nat_helper_put(params->helper);
 836 #endif
 837                 nf_conntrack_helper_put(params->helper);
 838         }
 839         if (params->ct_ft)
 840                 tcf_ct_flow_table_put(params->ct_ft);
 841         if (params->tmpl)
 842                 nf_ct_put(params->tmpl);
 843         kfree(params);
 844 }
 845
 846 static void tcf_ct_params_free_rcu(struct rcu_head *head)
 847 {
 848         struct tcf_ct_params *params;
 849
 850         params = container_of(head, struct tcf_ct_params, rcu);
 851         tcf_ct_params_free(params);
 852 }
 853
 854 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
 855 {
 856 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 857         u32 new_mark;
 858
 859         if (!mask)
 860                 return;
 861
 862         new_mark = mark | (READ_ONCE(ct->mark) & ~(mask));
 863         if (READ_ONCE(ct->mark) != new_mark) {
 864                 WRITE_ONCE(ct->mark, new_mark);
 865                 if (nf_ct_is_confirmed(ct))
 866                         nf_conntrack_event_cache(IPCT_MARK, ct);
 867         }
 868 #endif
 869 }
 870
 871 static void tcf_ct_act_set_labels(struct nf_conn *ct,
 872                                   u32 *labels,
 873                                   u32 *labels_m)
 874 {
 875 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
 876         size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
 877
 878         if (!memchr_inv(labels_m, 0, labels_sz))
 879                 return;
 880
 881         nf_connlabels_replace(ct, labels, labels_m, 4);
 882 #endif
 883 }
 884
 885 static int tcf_ct_act_nat(struct sk_buff *skb,
 886                           struct nf_conn *ct,
 887                           enum ip_conntrack_info ctinfo,
 888                           int ct_action,
 889                           struct nf_nat_range2 *range,
 890                           bool commit)
 891 {
 892 #if IS_ENABLED(CONFIG_NF_NAT)
 893         int err, action = 0;
 894
 895         if (!(ct_action & TCA_CT_ACT_NAT))
 896                 return NF_ACCEPT;
 897         if (ct_action & TCA_CT_ACT_NAT_SRC)
 898                 action |= BIT(NF_NAT_MANIP_SRC);
 899         if (ct_action & TCA_CT_ACT_NAT_DST)
 900                 action |= BIT(NF_NAT_MANIP_DST);
 901
 902         err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit);
 903
 904         if (action & BIT(NF_NAT_MANIP_SRC))
 905                 tc_skb_cb(skb)->post_ct_snat = 1;
 906         if (action & BIT(NF_NAT_MANIP_DST))
 907                 tc_skb_cb(skb)->post_ct_dnat = 1;
 908
 909         return err;
 910 #else
 911         return NF_ACCEPT;
 912 #endif
 913 }
 914
 915 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 916                                  struct tcf_result *res)
 917 {
 918         struct net *net = dev_net(skb->dev);
 919         enum ip_conntrack_info ctinfo;
 920         struct tcf_ct *c = to_ct(a);
 921         struct nf_conn *tmpl = NULL;
 922         struct nf_hook_state state;
 923         bool cached, commit, clear;
 924         int nh_ofs, err, retval;
 925         struct tcf_ct_params *p;
 926         bool add_helper = false;
 927         bool skip_add = false;
 928         bool defrag = false;
 929         struct nf_conn *ct;
 930         u8 family;
 931
 932         p = rcu_dereference_bh(c->params);
 933
 934         retval = READ_ONCE(c->tcf_action);
 935         commit = p->ct_action & TCA_CT_ACT_COMMIT;
 936         clear = p->ct_action & TCA_CT_ACT_CLEAR;
 937         tmpl = p->tmpl;
 938
 939         tcf_lastuse_update(&c->tcf_tm);
 940         tcf_action_update_bstats(&c->common, skb);
 941
 942         if (clear) {
 943                 tc_skb_cb(skb)->post_ct = false;
 944                 ct = nf_ct_get(skb, &ctinfo);
 945                 if (ct) {
 946                         nf_ct_put(ct);
 947                         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 948                 }
 949
 950                 goto out_clear;
 951         }
 952
 953         family = tcf_ct_skb_nf_family(skb);
 954         if (family == NFPROTO_UNSPEC)
 955                 goto drop;
 956
 957         /* The conntrack module expects to be working at L3.
 958          * We also try to pull the IPv4/6 header to linear area
 959          */
 960         nh_ofs = skb_network_offset(skb);
 961         skb_pull_rcsum(skb, nh_ofs);
 962         err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
 963         if (err == -EINPROGRESS) {
 964                 retval = TC_ACT_STOLEN;
 965                 goto out_clear;
 966         }
 967         if (err)
 968                 goto drop;
 969
 970         err = nf_ct_skb_network_trim(skb, family);
 971         if (err)
 972                 goto drop;
 973
 974         /* If we are recirculating packets to match on ct fields and
 975          * committing with a separate ct action, then we don't need to
 976          * actually run the packet through conntrack twice unless it's for a
 977          * different zone.
 978          */
 979         cached = tcf_ct_skb_nfct_cached(net, skb, p);
 980         if (!cached) {
 981                 if (tcf_ct_flow_table_lookup(p, skb, family)) {
 982                         skip_add = true;
 983                         goto do_nat;
 984                 }
 985
 986                 /* Associate skb with specified zone. */
 987                 if (tmpl) {
 988                         nf_conntrack_put(skb_nfct(skb));
 989                         nf_conntrack_get(&tmpl->ct_general);
 990                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 991                 }
 992
 993                 state.hook = NF_INET_PRE_ROUTING;
 994                 state.net = net;
 995                 state.pf = family;
 996                 err = nf_conntrack_in(skb, &state);
 997                 if (err != NF_ACCEPT)
 998                         goto out_push;
 999         }
1000
1001 do_nat:
1002         ct = nf_ct_get(skb, &ctinfo);
1003         if (!ct)
1004                 goto out_push;
1005         nf_ct_deliver_cached_events(ct);
1006         nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1007
1008         err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1009         if (err != NF_ACCEPT)
1010                 goto drop;
1011
1012         if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) {
1013                 err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC);
1014                 if (err)
1015                         goto drop;
1016                 add_helper = true;
1017                 if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) {
1018                         if (!nfct_seqadj_ext_add(ct))
1019                                 goto drop;
1020                 }
1021         }
1022
1023         if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) {
1024                 if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT)
1025                         goto drop;
1026         }
1027
1028         if (commit) {
1029                 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1030                 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1031
1032                 if (!nf_ct_is_confirmed(ct))
1033                         nf_conn_act_ct_ext_add(ct);
1034
1035                 /* This will take care of sending queued events
1036                  * even if the connection is already confirmed.
1037                  */
1038                 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1039                         goto drop;
1040         }
1041
1042         if (!skip_add)
1043                 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1044
1045 out_push:
1046         skb_push_rcsum(skb, nh_ofs);
1047
1048         tc_skb_cb(skb)->post_ct = true;
1049         tc_skb_cb(skb)->zone = p->zone;
1050 out_clear:
1051         if (defrag)
1052                 qdisc_skb_cb(skb)->pkt_len = skb->len;
1053         return retval;
1054
1055 drop:
1056         tcf_action_inc_drop_qstats(&c->common);
1057         return TC_ACT_SHOT;
1058 }
1059
1060 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1061         [TCA_CT_ACTION] = { .type = NLA_U16 },
1062         [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1063         [TCA_CT_ZONE] = { .type = NLA_U16 },
1064         [TCA_CT_MARK] = { .type = NLA_U32 },
1065         [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1066         [TCA_CT_LABELS] = { .type = NLA_BINARY,
1067                             .len = 128 / BITS_PER_BYTE },
1068         [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1069                                  .len = 128 / BITS_PER_BYTE },
1070         [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1071         [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1072         [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1073         [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1074         [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1075         [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1076         [TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN },
1077         [TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 },
1078         [TCA_CT_HELPER_PROTO] = { .type = NLA_U8 },
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, act_ct_ops.net_id);
1167         struct nf_conntrack_zone zone;
1168         int err, family, proto, len;
1169         struct nf_conn *tmpl;
1170         char *name;
1171
1172         p->zone = NF_CT_DEFAULT_ZONE_ID;
1173
1174         tcf_ct_set_key_val(tb,
1175                            &p->ct_action, TCA_CT_ACTION,
1176                            NULL, TCA_CT_UNSPEC,
1177                            sizeof(p->ct_action));
1178
1179         if (p->ct_action & TCA_CT_ACT_CLEAR)
1180                 return 0;
1181
1182         err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1183         if (err)
1184                 return err;
1185
1186         if (tb[TCA_CT_MARK]) {
1187                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1188                         NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1189                         return -EOPNOTSUPP;
1190                 }
1191                 tcf_ct_set_key_val(tb,
1192                                    &p->mark, TCA_CT_MARK,
1193                                    &p->mark_mask, TCA_CT_MARK_MASK,
1194                                    sizeof(p->mark));
1195         }
1196
1197         if (tb[TCA_CT_LABELS]) {
1198                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1199                         NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1200                         return -EOPNOTSUPP;
1201                 }
1202
1203                 if (!tn->labels) {
1204                         NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1205                         return -EOPNOTSUPP;
1206                 }
1207                 tcf_ct_set_key_val(tb,
1208                                    p->labels, TCA_CT_LABELS,
1209                                    p->labels_mask, TCA_CT_LABELS_MASK,
1210                                    sizeof(p->labels));
1211         }
1212
1213         if (tb[TCA_CT_ZONE]) {
1214                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1215                         NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1216                         return -EOPNOTSUPP;
1217                 }
1218
1219                 tcf_ct_set_key_val(tb,
1220                                    &p->zone, TCA_CT_ZONE,
1221                                    NULL, TCA_CT_UNSPEC,
1222                                    sizeof(p->zone));
1223         }
1224
1225         nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1226         tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1227         if (!tmpl) {
1228                 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1229                 return -ENOMEM;
1230         }
1231         p->tmpl = tmpl;
1232         if (tb[TCA_CT_HELPER_NAME]) {
1233                 name = nla_data(tb[TCA_CT_HELPER_NAME]);
1234                 len = nla_len(tb[TCA_CT_HELPER_NAME]);
1235                 if (len > 16 || name[len - 1] != '\0') {
1236                         NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name.");
1237                         err = -EINVAL;
1238                         goto err;
1239                 }
1240                 family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(tb[TCA_CT_HELPER_FAMILY]) : AF_INET;
1241                 proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP;
1242                 err = nf_ct_add_helper(tmpl, name, family, proto,
1243                                        p->ct_action & TCA_CT_ACT_NAT, &p->helper);
1244                 if (err) {
1245                         NL_SET_ERR_MSG_MOD(extack, "Failed to add helper");
1246                         goto err;
1247                 }
1248         }
1249
1250         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1251         return 0;
1252 err:
1253         nf_ct_put(p->tmpl);
1254         p->tmpl = NULL;
1255         return err;
1256 }
1257
1258 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1259                        struct nlattr *est, struct tc_action **a,
1260                        struct tcf_proto *tp, u32 flags,
1261                        struct netlink_ext_ack *extack)
1262 {
1263         struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id);
1264         bool bind = flags & TCA_ACT_FLAGS_BIND;
1265         struct tcf_ct_params *params = NULL;
1266         struct nlattr *tb[TCA_CT_MAX + 1];
1267         struct tcf_chain *goto_ch = NULL;
1268         struct tc_ct *parm;
1269         struct tcf_ct *c;
1270         int err, res = 0;
1271         u32 index;
1272
1273         if (!nla) {
1274                 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1275                 return -EINVAL;
1276         }
1277
1278         err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1279         if (err < 0)
1280                 return err;
1281
1282         if (!tb[TCA_CT_PARMS]) {
1283                 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1284                 return -EINVAL;
1285         }
1286         parm = nla_data(tb[TCA_CT_PARMS]);
1287         index = parm->index;
1288         err = tcf_idr_check_alloc(tn, &index, a, bind);
1289         if (err < 0)
1290                 return err;
1291
1292         if (!err) {
1293                 err = tcf_idr_create_from_flags(tn, index, est, a,
1294                                                 &act_ct_ops, bind, flags);
1295                 if (err) {
1296                         tcf_idr_cleanup(tn, index);
1297                         return err;
1298                 }
1299                 res = ACT_P_CREATED;
1300         } else {
1301                 if (bind)
1302                         return 0;
1303
1304                 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1305                         tcf_idr_release(*a, bind);
1306                         return -EEXIST;
1307                 }
1308         }
1309         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1310         if (err < 0)
1311                 goto cleanup;
1312
1313         c = to_ct(*a);
1314
1315         params = kzalloc(sizeof(*params), GFP_KERNEL);
1316         if (unlikely(!params)) {
1317                 err = -ENOMEM;
1318                 goto cleanup;
1319         }
1320
1321         err = tcf_ct_fill_params(net, params, parm, tb, extack);
1322         if (err)
1323                 goto cleanup;
1324
1325         err = tcf_ct_flow_table_get(net, params);
1326         if (err)
1327                 goto cleanup;
1328
1329         spin_lock_bh(&c->tcf_lock);
1330         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1331         params = rcu_replace_pointer(c->params, params,
1332                                      lockdep_is_held(&c->tcf_lock));
1333         spin_unlock_bh(&c->tcf_lock);
1334
1335         if (goto_ch)
1336                 tcf_chain_put_by_act(goto_ch);
1337         if (params)
1338                 call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1339
1340         return res;
1341
1342 cleanup:
1343         if (goto_ch)
1344                 tcf_chain_put_by_act(goto_ch);
1345         if (params)
1346                 tcf_ct_params_free(params);
1347         tcf_idr_release(*a, bind);
1348         return err;
1349 }
1350
1351 static void tcf_ct_cleanup(struct tc_action *a)
1352 {
1353         struct tcf_ct_params *params;
1354         struct tcf_ct *c = to_ct(a);
1355
1356         params = rcu_dereference_protected(c->params, 1);
1357         if (params)
1358                 call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1359 }
1360
1361 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1362                                void *val, int val_type,
1363                                void *mask, int mask_type,
1364                                int len)
1365 {
1366         int err;
1367
1368         if (mask && !memchr_inv(mask, 0, len))
1369                 return 0;
1370
1371         err = nla_put(skb, val_type, len, val);
1372         if (err)
1373                 return err;
1374
1375         if (mask_type != TCA_CT_UNSPEC) {
1376                 err = nla_put(skb, mask_type, len, mask);
1377                 if (err)
1378                         return err;
1379         }
1380
1381         return 0;
1382 }
1383
1384 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1385 {
1386         struct nf_nat_range2 *range = &p->range;
1387
1388         if (!(p->ct_action & TCA_CT_ACT_NAT))
1389                 return 0;
1390
1391         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1392                 return 0;
1393
1394         if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1395                 if (p->ipv4_range) {
1396                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1397                                             range->min_addr.ip))
1398                                 return -1;
1399                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1400                                             range->max_addr.ip))
1401                                 return -1;
1402                 } else {
1403                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1404                                              &range->min_addr.in6))
1405                                 return -1;
1406                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1407                                              &range->max_addr.in6))
1408                                 return -1;
1409                 }
1410         }
1411
1412         if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1413                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1414                                  range->min_proto.all))
1415                         return -1;
1416                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1417                                  range->max_proto.all))
1418                         return -1;
1419         }
1420
1421         return 0;
1422 }
1423
1424 static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper)
1425 {
1426         if (!helper)
1427                 return 0;
1428
1429         if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) ||
1430             nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) ||
1431             nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum))
1432                 return -1;
1433
1434         return 0;
1435 }
1436
1437 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1438                               int bind, int ref)
1439 {
1440         unsigned char *b = skb_tail_pointer(skb);
1441         struct tcf_ct *c = to_ct(a);
1442         struct tcf_ct_params *p;
1443
1444         struct tc_ct opt = {
1445                 .index   = c->tcf_index,
1446                 .refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1447                 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1448         };
1449         struct tcf_t t;
1450
1451         spin_lock_bh(&c->tcf_lock);
1452         p = rcu_dereference_protected(c->params,
1453                                       lockdep_is_held(&c->tcf_lock));
1454         opt.action = c->tcf_action;
1455
1456         if (tcf_ct_dump_key_val(skb,
1457                                 &p->ct_action, TCA_CT_ACTION,
1458                                 NULL, TCA_CT_UNSPEC,
1459                                 sizeof(p->ct_action)))
1460                 goto nla_put_failure;
1461
1462         if (p->ct_action & TCA_CT_ACT_CLEAR)
1463                 goto skip_dump;
1464
1465         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1466             tcf_ct_dump_key_val(skb,
1467                                 &p->mark, TCA_CT_MARK,
1468                                 &p->mark_mask, TCA_CT_MARK_MASK,
1469                                 sizeof(p->mark)))
1470                 goto nla_put_failure;
1471
1472         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1473             tcf_ct_dump_key_val(skb,
1474                                 p->labels, TCA_CT_LABELS,
1475                                 p->labels_mask, TCA_CT_LABELS_MASK,
1476                                 sizeof(p->labels)))
1477                 goto nla_put_failure;
1478
1479         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1480             tcf_ct_dump_key_val(skb,
1481                                 &p->zone, TCA_CT_ZONE,
1482                                 NULL, TCA_CT_UNSPEC,
1483                                 sizeof(p->zone)))
1484                 goto nla_put_failure;
1485
1486         if (tcf_ct_dump_nat(skb, p))
1487                 goto nla_put_failure;
1488
1489         if (tcf_ct_dump_helper(skb, p->helper))
1490                 goto nla_put_failure;
1491
1492 skip_dump:
1493         if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1494                 goto nla_put_failure;
1495
1496         tcf_tm_dump(&t, &c->tcf_tm);
1497         if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1498                 goto nla_put_failure;
1499         spin_unlock_bh(&c->tcf_lock);
1500
1501         return skb->len;
1502 nla_put_failure:
1503         spin_unlock_bh(&c->tcf_lock);
1504         nlmsg_trim(skb, b);
1505         return -1;
1506 }
1507
1508 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1509                              u64 drops, u64 lastuse, bool hw)
1510 {
1511         struct tcf_ct *c = to_ct(a);
1512
1513         tcf_action_update_stats(a, bytes, packets, drops, hw);
1514         c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1515 }
1516
1517 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1518                                     u32 *index_inc, bool bind,
1519                                     struct netlink_ext_ack *extack)
1520 {
1521         if (bind) {
1522                 struct flow_action_entry *entry = entry_data;
1523
1524                 entry->id = FLOW_ACTION_CT;
1525                 entry->ct.action = tcf_ct_action(act);
1526                 entry->ct.zone = tcf_ct_zone(act);
1527                 entry->ct.flow_table = tcf_ct_ft(act);
1528                 *index_inc = 1;
1529         } else {
1530                 struct flow_offload_action *fl_action = entry_data;
1531
1532                 fl_action->id = FLOW_ACTION_CT;
1533         }
1534
1535         return 0;
1536 }
1537
1538 static struct tc_action_ops act_ct_ops = {
1539         .kind           =       "ct",
1540         .id             =       TCA_ID_CT,
1541         .owner          =       THIS_MODULE,
1542         .act            =       tcf_ct_act,
1543         .dump           =       tcf_ct_dump,
1544         .init           =       tcf_ct_init,
1545         .cleanup        =       tcf_ct_cleanup,
1546         .stats_update   =       tcf_stats_update,
1547         .offload_act_setup =    tcf_ct_offload_act_setup,
1548         .size           =       sizeof(struct tcf_ct),
1549 };
1550
1551 static __net_init int ct_init_net(struct net *net)
1552 {
1553         unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1554         struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1555
1556         if (nf_connlabels_get(net, n_bits - 1)) {
1557                 tn->labels = false;
1558                 pr_err("act_ct: Failed to set connlabels length");
1559         } else {
1560                 tn->labels = true;
1561         }
1562
1563         return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1564 }
1565
1566 static void __net_exit ct_exit_net(struct list_head *net_list)
1567 {
1568         struct net *net;
1569
1570         rtnl_lock();
1571         list_for_each_entry(net, net_list, exit_list) {
1572                 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1573
1574                 if (tn->labels)
1575                         nf_connlabels_put(net);
1576         }
1577         rtnl_unlock();
1578
1579         tc_action_net_exit(net_list, act_ct_ops.net_id);
1580 }
1581
1582 static struct pernet_operations ct_net_ops = {
1583         .init = ct_init_net,
1584         .exit_batch = ct_exit_net,
1585         .id   = &act_ct_ops.net_id,
1586         .size = sizeof(struct tc_ct_action_net),
1587 };
1588
1589 static int __init ct_init_module(void)
1590 {
1591         int err;
1592
1593         act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1594         if (!act_ct_wq)
1595                 return -ENOMEM;
1596
1597         err = tcf_ct_flow_tables_init();
1598         if (err)
1599                 goto err_tbl_init;
1600
1601         err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1602         if (err)
1603                 goto err_register;
1604
1605         static_branch_inc(&tcf_frag_xmit_count);
1606
1607         return 0;
1608
1609 err_register:
1610         tcf_ct_flow_tables_uninit();
1611 err_tbl_init:
1612         destroy_workqueue(act_ct_wq);
1613         return err;
1614 }
1615
1616 static void __exit ct_cleanup_module(void)
1617 {
1618         static_branch_dec(&tcf_frag_xmit_count);
1619         tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1620         tcf_ct_flow_tables_uninit();
1621         destroy_workqueue(act_ct_wq);
1622 }
1623
1624 module_init(ct_init_module);
1625 module_exit(ct_cleanup_module);
1626 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1627 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1628 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1629 MODULE_DESCRIPTION("Connection tracking action");
1630 MODULE_LICENSE("GPL v2");