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 /* Trim the skb to the length specified by the IP/IPv6 header,
 730  * removing any trailing lower-layer padding. This prepares the skb
 731  * for higher-layer processing that assumes skb->len excludes padding
 732  * (such as nf_ip_checksum). The caller needs to pull the skb to the
 733  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
 734  */
 735 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
 736 {
 737         unsigned int len;
 738
 739         switch (family) {
 740         case NFPROTO_IPV4:
 741                 len = skb_ip_totlen(skb);
 742                 break;
 743         case NFPROTO_IPV6:
 744                 len = sizeof(struct ipv6hdr)
 745                         + ntohs(ipv6_hdr(skb)->payload_len);
 746                 break;
 747         default:
 748                 len = skb->len;
 749         }
 750
 751         return pskb_trim_rcsum(skb, len);
 752 }
 753
 754 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
 755 {
 756         u8 family = NFPROTO_UNSPEC;
 757
 758         switch (skb_protocol(skb, true)) {
 759         case htons(ETH_P_IP):
 760                 family = NFPROTO_IPV4;
 761                 break;
 762         case htons(ETH_P_IPV6):
 763                 family = NFPROTO_IPV6;
 764                 break;
 765         default:
 766                 break;
 767         }
 768
 769         return family;
 770 }
 771
 772 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
 773 {
 774         unsigned int len;
 775
 776         len =  skb_network_offset(skb) + sizeof(struct iphdr);
 777         if (unlikely(skb->len < len))
 778                 return -EINVAL;
 779         if (unlikely(!pskb_may_pull(skb, len)))
 780                 return -ENOMEM;
 781
 782         *frag = ip_is_fragment(ip_hdr(skb));
 783         return 0;
 784 }
 785
 786 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
 787 {
 788         unsigned int flags = 0, len, payload_ofs = 0;
 789         unsigned short frag_off;
 790         int nexthdr;
 791
 792         len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
 793         if (unlikely(skb->len < len))
 794                 return -EINVAL;
 795         if (unlikely(!pskb_may_pull(skb, len)))
 796                 return -ENOMEM;
 797
 798         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 799         if (unlikely(nexthdr < 0))
 800                 return -EPROTO;
 801
 802         *frag = flags & IP6_FH_F_FRAG;
 803         return 0;
 804 }
 805
 806 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
 807                                    u8 family, u16 zone, bool *defrag)
 808 {
 809         enum ip_conntrack_info ctinfo;
 810         struct nf_conn *ct;
 811         int err = 0;
 812         bool frag;
 813         u16 mru;
 814
 815         /* Previously seen (loopback)? Ignore. */
 816         ct = nf_ct_get(skb, &ctinfo);
 817         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 818                 return 0;
 819
 820         if (family == NFPROTO_IPV4)
 821                 err = tcf_ct_ipv4_is_fragment(skb, &frag);
 822         else
 823                 err = tcf_ct_ipv6_is_fragment(skb, &frag);
 824         if (err || !frag)
 825                 return err;
 826
 827         skb_get(skb);
 828         mru = tc_skb_cb(skb)->mru;
 829
 830         if (family == NFPROTO_IPV4) {
 831                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 832
 833                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 834                 local_bh_disable();
 835                 err = ip_defrag(net, skb, user);
 836                 local_bh_enable();
 837                 if (err && err != -EINPROGRESS)
 838                         return err;
 839
 840                 if (!err) {
 841                         *defrag = true;
 842                         mru = IPCB(skb)->frag_max_size;
 843                 }
 844         } else { /* NFPROTO_IPV6 */
 845 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 846                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 847
 848                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 849                 err = nf_ct_frag6_gather(net, skb, user);
 850                 if (err && err != -EINPROGRESS)
 851                         goto out_free;
 852
 853                 if (!err) {
 854                         *defrag = true;
 855                         mru = IP6CB(skb)->frag_max_size;
 856                 }
 857 #else
 858                 err = -EOPNOTSUPP;
 859                 goto out_free;
 860 #endif
 861         }
 862
 863         if (err != -EINPROGRESS)
 864                 tc_skb_cb(skb)->mru = mru;
 865         skb_clear_hash(skb);
 866         skb->ignore_df = 1;
 867         return err;
 868
 869 out_free:
 870         kfree_skb(skb);
 871         return err;
 872 }
 873
 874 static void tcf_ct_params_free(struct tcf_ct_params *params)
 875 {
 876         if (params->helper) {
 877 #if IS_ENABLED(CONFIG_NF_NAT)
 878                 if (params->ct_action & TCA_CT_ACT_NAT)
 879                         nf_nat_helper_put(params->helper);
 880 #endif
 881                 nf_conntrack_helper_put(params->helper);
 882         }
 883         if (params->ct_ft)
 884                 tcf_ct_flow_table_put(params->ct_ft);
 885         if (params->tmpl)
 886                 nf_ct_put(params->tmpl);
 887         kfree(params);
 888 }
 889
 890 static void tcf_ct_params_free_rcu(struct rcu_head *head)
 891 {
 892         struct tcf_ct_params *params;
 893
 894         params = container_of(head, struct tcf_ct_params, rcu);
 895         tcf_ct_params_free(params);
 896 }
 897
 898 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
 899 {
 900 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 901         u32 new_mark;
 902
 903         if (!mask)
 904                 return;
 905
 906         new_mark = mark | (READ_ONCE(ct->mark) & ~(mask));
 907         if (READ_ONCE(ct->mark) != new_mark) {
 908                 WRITE_ONCE(ct->mark, new_mark);
 909                 if (nf_ct_is_confirmed(ct))
 910                         nf_conntrack_event_cache(IPCT_MARK, ct);
 911         }
 912 #endif
 913 }
 914
 915 static void tcf_ct_act_set_labels(struct nf_conn *ct,
 916                                   u32 *labels,
 917                                   u32 *labels_m)
 918 {
 919 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
 920         size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
 921
 922         if (!memchr_inv(labels_m, 0, labels_sz))
 923                 return;
 924
 925         nf_connlabels_replace(ct, labels, labels_m, 4);
 926 #endif
 927 }
 928
 929 static int tcf_ct_act_nat(struct sk_buff *skb,
 930                           struct nf_conn *ct,
 931                           enum ip_conntrack_info ctinfo,
 932                           int ct_action,
 933                           struct nf_nat_range2 *range,
 934                           bool commit)
 935 {
 936 #if IS_ENABLED(CONFIG_NF_NAT)
 937         int err, action = 0;
 938
 939         if (!(ct_action & TCA_CT_ACT_NAT))
 940                 return NF_ACCEPT;
 941         if (ct_action & TCA_CT_ACT_NAT_SRC)
 942                 action |= BIT(NF_NAT_MANIP_SRC);
 943         if (ct_action & TCA_CT_ACT_NAT_DST)
 944                 action |= BIT(NF_NAT_MANIP_DST);
 945
 946         err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit);
 947
 948         if (action & BIT(NF_NAT_MANIP_SRC))
 949                 tc_skb_cb(skb)->post_ct_snat = 1;
 950         if (action & BIT(NF_NAT_MANIP_DST))
 951                 tc_skb_cb(skb)->post_ct_dnat = 1;
 952
 953         return err;
 954 #else
 955         return NF_ACCEPT;
 956 #endif
 957 }
 958
 959 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 960                                  struct tcf_result *res)
 961 {
 962         struct net *net = dev_net(skb->dev);
 963         enum ip_conntrack_info ctinfo;
 964         struct tcf_ct *c = to_ct(a);
 965         struct nf_conn *tmpl = NULL;
 966         struct nf_hook_state state;
 967         bool cached, commit, clear;
 968         int nh_ofs, err, retval;
 969         struct tcf_ct_params *p;
 970         bool add_helper = false;
 971         bool skip_add = false;
 972         bool defrag = false;
 973         struct nf_conn *ct;
 974         u8 family;
 975
 976         p = rcu_dereference_bh(c->params);
 977
 978         retval = READ_ONCE(c->tcf_action);
 979         commit = p->ct_action & TCA_CT_ACT_COMMIT;
 980         clear = p->ct_action & TCA_CT_ACT_CLEAR;
 981         tmpl = p->tmpl;
 982
 983         tcf_lastuse_update(&c->tcf_tm);
 984         tcf_action_update_bstats(&c->common, skb);
 985
 986         if (clear) {
 987                 tc_skb_cb(skb)->post_ct = false;
 988                 ct = nf_ct_get(skb, &ctinfo);
 989                 if (ct) {
 990                         nf_ct_put(ct);
 991                         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 992                 }
 993
 994                 goto out_clear;
 995         }
 996
 997         family = tcf_ct_skb_nf_family(skb);
 998         if (family == NFPROTO_UNSPEC)
 999                 goto drop;
1000
1001         /* The conntrack module expects to be working at L3.
1002          * We also try to pull the IPv4/6 header to linear area
1003          */
1004         nh_ofs = skb_network_offset(skb);
1005         skb_pull_rcsum(skb, nh_ofs);
1006         err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
1007         if (err == -EINPROGRESS) {
1008                 retval = TC_ACT_STOLEN;
1009                 goto out_clear;
1010         }
1011         if (err)
1012                 goto drop;
1013
1014         err = tcf_ct_skb_network_trim(skb, family);
1015         if (err)
1016                 goto drop;
1017
1018         /* If we are recirculating packets to match on ct fields and
1019          * committing with a separate ct action, then we don't need to
1020          * actually run the packet through conntrack twice unless it's for a
1021          * different zone.
1022          */
1023         cached = tcf_ct_skb_nfct_cached(net, skb, p);
1024         if (!cached) {
1025                 if (tcf_ct_flow_table_lookup(p, skb, family)) {
1026                         skip_add = true;
1027                         goto do_nat;
1028                 }
1029
1030                 /* Associate skb with specified zone. */
1031                 if (tmpl) {
1032                         nf_conntrack_put(skb_nfct(skb));
1033                         nf_conntrack_get(&tmpl->ct_general);
1034                         nf_ct_set(skb, tmpl, IP_CT_NEW);
1035                 }
1036
1037                 state.hook = NF_INET_PRE_ROUTING;
1038                 state.net = net;
1039                 state.pf = family;
1040                 err = nf_conntrack_in(skb, &state);
1041                 if (err != NF_ACCEPT)
1042                         goto out_push;
1043         }
1044
1045 do_nat:
1046         ct = nf_ct_get(skb, &ctinfo);
1047         if (!ct)
1048                 goto out_push;
1049         nf_ct_deliver_cached_events(ct);
1050         nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1051
1052         err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1053         if (err != NF_ACCEPT)
1054                 goto drop;
1055
1056         if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) {
1057                 err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC);
1058                 if (err)
1059                         goto drop;
1060                 add_helper = true;
1061                 if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) {
1062                         if (!nfct_seqadj_ext_add(ct))
1063                                 goto drop;
1064                 }
1065         }
1066
1067         if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) {
1068                 if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT)
1069                         goto drop;
1070         }
1071
1072         if (commit) {
1073                 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1074                 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1075
1076                 if (!nf_ct_is_confirmed(ct))
1077                         nf_conn_act_ct_ext_add(ct);
1078
1079                 /* This will take care of sending queued events
1080                  * even if the connection is already confirmed.
1081                  */
1082                 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1083                         goto drop;
1084         }
1085
1086         if (!skip_add)
1087                 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1088
1089 out_push:
1090         skb_push_rcsum(skb, nh_ofs);
1091
1092         tc_skb_cb(skb)->post_ct = true;
1093         tc_skb_cb(skb)->zone = p->zone;
1094 out_clear:
1095         if (defrag)
1096                 qdisc_skb_cb(skb)->pkt_len = skb->len;
1097         return retval;
1098
1099 drop:
1100         tcf_action_inc_drop_qstats(&c->common);
1101         return TC_ACT_SHOT;
1102 }
1103
1104 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1105         [TCA_CT_ACTION] = { .type = NLA_U16 },
1106         [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1107         [TCA_CT_ZONE] = { .type = NLA_U16 },
1108         [TCA_CT_MARK] = { .type = NLA_U32 },
1109         [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1110         [TCA_CT_LABELS] = { .type = NLA_BINARY,
1111                             .len = 128 / BITS_PER_BYTE },
1112         [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1113                                  .len = 128 / BITS_PER_BYTE },
1114         [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1115         [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1116         [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1117         [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1118         [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1119         [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1120         [TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN },
1121         [TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 },
1122         [TCA_CT_HELPER_PROTO] = { .type = NLA_U8 },
1123 };
1124
1125 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1126                                   struct tc_ct *parm,
1127                                   struct nlattr **tb,
1128                                   struct netlink_ext_ack *extack)
1129 {
1130         struct nf_nat_range2 *range;
1131
1132         if (!(p->ct_action & TCA_CT_ACT_NAT))
1133                 return 0;
1134
1135         if (!IS_ENABLED(CONFIG_NF_NAT)) {
1136                 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1137                 return -EOPNOTSUPP;
1138         }
1139
1140         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1141                 return 0;
1142
1143         if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1144             (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1145                 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1146                 return -EOPNOTSUPP;
1147         }
1148
1149         range = &p->range;
1150         if (tb[TCA_CT_NAT_IPV4_MIN]) {
1151                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1152
1153                 p->ipv4_range = true;
1154                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1155                 range->min_addr.ip =
1156                         nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1157
1158                 range->max_addr.ip = max_attr ?
1159                                      nla_get_in_addr(max_attr) :
1160                                      range->min_addr.ip;
1161         } else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1162                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1163
1164                 p->ipv4_range = false;
1165                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1166                 range->min_addr.in6 =
1167                         nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1168
1169                 range->max_addr.in6 = max_attr ?
1170                                       nla_get_in6_addr(max_attr) :
1171                                       range->min_addr.in6;
1172         }
1173
1174         if (tb[TCA_CT_NAT_PORT_MIN]) {
1175                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1176                 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1177
1178                 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1179                                        nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1180                                        range->min_proto.all;
1181         }
1182
1183         return 0;
1184 }
1185
1186 static void tcf_ct_set_key_val(struct nlattr **tb,
1187                                void *val, int val_type,
1188                                void *mask, int mask_type,
1189                                int len)
1190 {
1191         if (!tb[val_type])
1192                 return;
1193         nla_memcpy(val, tb[val_type], len);
1194
1195         if (!mask)
1196                 return;
1197
1198         if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1199                 memset(mask, 0xff, len);
1200         else
1201                 nla_memcpy(mask, tb[mask_type], len);
1202 }
1203
1204 static int tcf_ct_fill_params(struct net *net,
1205                               struct tcf_ct_params *p,
1206                               struct tc_ct *parm,
1207                               struct nlattr **tb,
1208                               struct netlink_ext_ack *extack)
1209 {
1210         struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1211         struct nf_conntrack_zone zone;
1212         int err, family, proto, len;
1213         struct nf_conn *tmpl;
1214         char *name;
1215
1216         p->zone = NF_CT_DEFAULT_ZONE_ID;
1217
1218         tcf_ct_set_key_val(tb,
1219                            &p->ct_action, TCA_CT_ACTION,
1220                            NULL, TCA_CT_UNSPEC,
1221                            sizeof(p->ct_action));
1222
1223         if (p->ct_action & TCA_CT_ACT_CLEAR)
1224                 return 0;
1225
1226         err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1227         if (err)
1228                 return err;
1229
1230         if (tb[TCA_CT_MARK]) {
1231                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1232                         NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1233                         return -EOPNOTSUPP;
1234                 }
1235                 tcf_ct_set_key_val(tb,
1236                                    &p->mark, TCA_CT_MARK,
1237                                    &p->mark_mask, TCA_CT_MARK_MASK,
1238                                    sizeof(p->mark));
1239         }
1240
1241         if (tb[TCA_CT_LABELS]) {
1242                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1243                         NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1244                         return -EOPNOTSUPP;
1245                 }
1246
1247                 if (!tn->labels) {
1248                         NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1249                         return -EOPNOTSUPP;
1250                 }
1251                 tcf_ct_set_key_val(tb,
1252                                    p->labels, TCA_CT_LABELS,
1253                                    p->labels_mask, TCA_CT_LABELS_MASK,
1254                                    sizeof(p->labels));
1255         }
1256
1257         if (tb[TCA_CT_ZONE]) {
1258                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1259                         NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1260                         return -EOPNOTSUPP;
1261                 }
1262
1263                 tcf_ct_set_key_val(tb,
1264                                    &p->zone, TCA_CT_ZONE,
1265                                    NULL, TCA_CT_UNSPEC,
1266                                    sizeof(p->zone));
1267         }
1268
1269         nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1270         tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1271         if (!tmpl) {
1272                 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1273                 return -ENOMEM;
1274         }
1275         p->tmpl = tmpl;
1276         if (tb[TCA_CT_HELPER_NAME]) {
1277                 name = nla_data(tb[TCA_CT_HELPER_NAME]);
1278                 len = nla_len(tb[TCA_CT_HELPER_NAME]);
1279                 if (len > 16 || name[len - 1] != '\0') {
1280                         NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name.");
1281                         err = -EINVAL;
1282                         goto err;
1283                 }
1284                 family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(tb[TCA_CT_HELPER_FAMILY]) : AF_INET;
1285                 proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP;
1286                 err = nf_ct_add_helper(tmpl, name, family, proto,
1287                                        p->ct_action & TCA_CT_ACT_NAT, &p->helper);
1288                 if (err) {
1289                         NL_SET_ERR_MSG_MOD(extack, "Failed to add helper");
1290                         goto err;
1291                 }
1292         }
1293
1294         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1295         return 0;
1296 err:
1297         nf_ct_put(p->tmpl);
1298         p->tmpl = NULL;
1299         return err;
1300 }
1301
1302 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1303                        struct nlattr *est, struct tc_action **a,
1304                        struct tcf_proto *tp, u32 flags,
1305                        struct netlink_ext_ack *extack)
1306 {
1307         struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id);
1308         bool bind = flags & TCA_ACT_FLAGS_BIND;
1309         struct tcf_ct_params *params = NULL;
1310         struct nlattr *tb[TCA_CT_MAX + 1];
1311         struct tcf_chain *goto_ch = NULL;
1312         struct tc_ct *parm;
1313         struct tcf_ct *c;
1314         int err, res = 0;
1315         u32 index;
1316
1317         if (!nla) {
1318                 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1319                 return -EINVAL;
1320         }
1321
1322         err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1323         if (err < 0)
1324                 return err;
1325
1326         if (!tb[TCA_CT_PARMS]) {
1327                 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1328                 return -EINVAL;
1329         }
1330         parm = nla_data(tb[TCA_CT_PARMS]);
1331         index = parm->index;
1332         err = tcf_idr_check_alloc(tn, &index, a, bind);
1333         if (err < 0)
1334                 return err;
1335
1336         if (!err) {
1337                 err = tcf_idr_create_from_flags(tn, index, est, a,
1338                                                 &act_ct_ops, bind, flags);
1339                 if (err) {
1340                         tcf_idr_cleanup(tn, index);
1341                         return err;
1342                 }
1343                 res = ACT_P_CREATED;
1344         } else {
1345                 if (bind)
1346                         return 0;
1347
1348                 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1349                         tcf_idr_release(*a, bind);
1350                         return -EEXIST;
1351                 }
1352         }
1353         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1354         if (err < 0)
1355                 goto cleanup;
1356
1357         c = to_ct(*a);
1358
1359         params = kzalloc(sizeof(*params), GFP_KERNEL);
1360         if (unlikely(!params)) {
1361                 err = -ENOMEM;
1362                 goto cleanup;
1363         }
1364
1365         err = tcf_ct_fill_params(net, params, parm, tb, extack);
1366         if (err)
1367                 goto cleanup;
1368
1369         err = tcf_ct_flow_table_get(net, params);
1370         if (err)
1371                 goto cleanup;
1372
1373         spin_lock_bh(&c->tcf_lock);
1374         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1375         params = rcu_replace_pointer(c->params, params,
1376                                      lockdep_is_held(&c->tcf_lock));
1377         spin_unlock_bh(&c->tcf_lock);
1378
1379         if (goto_ch)
1380                 tcf_chain_put_by_act(goto_ch);
1381         if (params)
1382                 call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1383
1384         return res;
1385
1386 cleanup:
1387         if (goto_ch)
1388                 tcf_chain_put_by_act(goto_ch);
1389         if (params)
1390                 tcf_ct_params_free(params);
1391         tcf_idr_release(*a, bind);
1392         return err;
1393 }
1394
1395 static void tcf_ct_cleanup(struct tc_action *a)
1396 {
1397         struct tcf_ct_params *params;
1398         struct tcf_ct *c = to_ct(a);
1399
1400         params = rcu_dereference_protected(c->params, 1);
1401         if (params)
1402                 call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1403 }
1404
1405 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1406                                void *val, int val_type,
1407                                void *mask, int mask_type,
1408                                int len)
1409 {
1410         int err;
1411
1412         if (mask && !memchr_inv(mask, 0, len))
1413                 return 0;
1414
1415         err = nla_put(skb, val_type, len, val);
1416         if (err)
1417                 return err;
1418
1419         if (mask_type != TCA_CT_UNSPEC) {
1420                 err = nla_put(skb, mask_type, len, mask);
1421                 if (err)
1422                         return err;
1423         }
1424
1425         return 0;
1426 }
1427
1428 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1429 {
1430         struct nf_nat_range2 *range = &p->range;
1431
1432         if (!(p->ct_action & TCA_CT_ACT_NAT))
1433                 return 0;
1434
1435         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1436                 return 0;
1437
1438         if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1439                 if (p->ipv4_range) {
1440                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1441                                             range->min_addr.ip))
1442                                 return -1;
1443                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1444                                             range->max_addr.ip))
1445                                 return -1;
1446                 } else {
1447                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1448                                              &range->min_addr.in6))
1449                                 return -1;
1450                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1451                                              &range->max_addr.in6))
1452                                 return -1;
1453                 }
1454         }
1455
1456         if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1457                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1458                                  range->min_proto.all))
1459                         return -1;
1460                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1461                                  range->max_proto.all))
1462                         return -1;
1463         }
1464
1465         return 0;
1466 }
1467
1468 static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper)
1469 {
1470         if (!helper)
1471                 return 0;
1472
1473         if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) ||
1474             nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) ||
1475             nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum))
1476                 return -1;
1477
1478         return 0;
1479 }
1480
1481 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1482                               int bind, int ref)
1483 {
1484         unsigned char *b = skb_tail_pointer(skb);
1485         struct tcf_ct *c = to_ct(a);
1486         struct tcf_ct_params *p;
1487
1488         struct tc_ct opt = {
1489                 .index   = c->tcf_index,
1490                 .refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1491                 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1492         };
1493         struct tcf_t t;
1494
1495         spin_lock_bh(&c->tcf_lock);
1496         p = rcu_dereference_protected(c->params,
1497                                       lockdep_is_held(&c->tcf_lock));
1498         opt.action = c->tcf_action;
1499
1500         if (tcf_ct_dump_key_val(skb,
1501                                 &p->ct_action, TCA_CT_ACTION,
1502                                 NULL, TCA_CT_UNSPEC,
1503                                 sizeof(p->ct_action)))
1504                 goto nla_put_failure;
1505
1506         if (p->ct_action & TCA_CT_ACT_CLEAR)
1507                 goto skip_dump;
1508
1509         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1510             tcf_ct_dump_key_val(skb,
1511                                 &p->mark, TCA_CT_MARK,
1512                                 &p->mark_mask, TCA_CT_MARK_MASK,
1513                                 sizeof(p->mark)))
1514                 goto nla_put_failure;
1515
1516         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1517             tcf_ct_dump_key_val(skb,
1518                                 p->labels, TCA_CT_LABELS,
1519                                 p->labels_mask, TCA_CT_LABELS_MASK,
1520                                 sizeof(p->labels)))
1521                 goto nla_put_failure;
1522
1523         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1524             tcf_ct_dump_key_val(skb,
1525                                 &p->zone, TCA_CT_ZONE,
1526                                 NULL, TCA_CT_UNSPEC,
1527                                 sizeof(p->zone)))
1528                 goto nla_put_failure;
1529
1530         if (tcf_ct_dump_nat(skb, p))
1531                 goto nla_put_failure;
1532
1533         if (tcf_ct_dump_helper(skb, p->helper))
1534                 goto nla_put_failure;
1535
1536 skip_dump:
1537         if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1538                 goto nla_put_failure;
1539
1540         tcf_tm_dump(&t, &c->tcf_tm);
1541         if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1542                 goto nla_put_failure;
1543         spin_unlock_bh(&c->tcf_lock);
1544
1545         return skb->len;
1546 nla_put_failure:
1547         spin_unlock_bh(&c->tcf_lock);
1548         nlmsg_trim(skb, b);
1549         return -1;
1550 }
1551
1552 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1553                              u64 drops, u64 lastuse, bool hw)
1554 {
1555         struct tcf_ct *c = to_ct(a);
1556
1557         tcf_action_update_stats(a, bytes, packets, drops, hw);
1558         c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1559 }
1560
1561 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1562                                     u32 *index_inc, bool bind,
1563                                     struct netlink_ext_ack *extack)
1564 {
1565         if (bind) {
1566                 struct flow_action_entry *entry = entry_data;
1567
1568                 entry->id = FLOW_ACTION_CT;
1569                 entry->ct.action = tcf_ct_action(act);
1570                 entry->ct.zone = tcf_ct_zone(act);
1571                 entry->ct.flow_table = tcf_ct_ft(act);
1572                 *index_inc = 1;
1573         } else {
1574                 struct flow_offload_action *fl_action = entry_data;
1575
1576                 fl_action->id = FLOW_ACTION_CT;
1577         }
1578
1579         return 0;
1580 }
1581
1582 static struct tc_action_ops act_ct_ops = {
1583         .kind           =       "ct",
1584         .id             =       TCA_ID_CT,
1585         .owner          =       THIS_MODULE,
1586         .act            =       tcf_ct_act,
1587         .dump           =       tcf_ct_dump,
1588         .init           =       tcf_ct_init,
1589         .cleanup        =       tcf_ct_cleanup,
1590         .stats_update   =       tcf_stats_update,
1591         .offload_act_setup =    tcf_ct_offload_act_setup,
1592         .size           =       sizeof(struct tcf_ct),
1593 };
1594
1595 static __net_init int ct_init_net(struct net *net)
1596 {
1597         unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1598         struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1599
1600         if (nf_connlabels_get(net, n_bits - 1)) {
1601                 tn->labels = false;
1602                 pr_err("act_ct: Failed to set connlabels length");
1603         } else {
1604                 tn->labels = true;
1605         }
1606
1607         return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1608 }
1609
1610 static void __net_exit ct_exit_net(struct list_head *net_list)
1611 {
1612         struct net *net;
1613
1614         rtnl_lock();
1615         list_for_each_entry(net, net_list, exit_list) {
1616                 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1617
1618                 if (tn->labels)
1619                         nf_connlabels_put(net);
1620         }
1621         rtnl_unlock();
1622
1623         tc_action_net_exit(net_list, act_ct_ops.net_id);
1624 }
1625
1626 static struct pernet_operations ct_net_ops = {
1627         .init = ct_init_net,
1628         .exit_batch = ct_exit_net,
1629         .id   = &act_ct_ops.net_id,
1630         .size = sizeof(struct tc_ct_action_net),
1631 };
1632
1633 static int __init ct_init_module(void)
1634 {
1635         int err;
1636
1637         act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1638         if (!act_ct_wq)
1639                 return -ENOMEM;
1640
1641         err = tcf_ct_flow_tables_init();
1642         if (err)
1643                 goto err_tbl_init;
1644
1645         err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1646         if (err)
1647                 goto err_register;
1648
1649         static_branch_inc(&tcf_frag_xmit_count);
1650
1651         return 0;
1652
1653 err_register:
1654         tcf_ct_flow_tables_uninit();
1655 err_tbl_init:
1656         destroy_workqueue(act_ct_wq);
1657         return err;
1658 }
1659
1660 static void __exit ct_cleanup_module(void)
1661 {
1662         static_branch_dec(&tcf_frag_xmit_count);
1663         tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1664         tcf_ct_flow_tables_uninit();
1665         destroy_workqueue(act_ct_wq);
1666 }
1667
1668 module_init(ct_init_module);
1669 module_exit(ct_cleanup_module);
1670 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1671 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1672 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1673 MODULE_DESCRIPTION("Connection tracking action");
1674 MODULE_LICENSE("GPL v2");