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