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
  28 #include <net/netfilter/nf_flow_table.h>
  29 #include <net/netfilter/nf_conntrack.h>
  30 #include <net/netfilter/nf_conntrack_core.h>
  31 #include <net/netfilter/nf_conntrack_zones.h>
  32 #include <net/netfilter/nf_conntrack_helper.h>
  33 #include <net/netfilter/nf_conntrack_acct.h>
  34 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  35 #include <uapi/linux/netfilter/nf_nat.h>
  36
  37 static struct workqueue_struct *act_ct_wq;
  38 static struct rhashtable zones_ht;
  39 static DEFINE_MUTEX(zones_mutex);
  40
  41 struct tcf_ct_flow_table {
  42         struct rhash_head node; /* In zones tables */
  43
  44         struct rcu_work rwork;
  45         struct nf_flowtable nf_ft;
  46         refcount_t ref;
  47         u16 zone;
  48
  49         bool dying;
  50 };
  51
  52 static const struct rhashtable_params zones_params = {
  53         .head_offset = offsetof(struct tcf_ct_flow_table, node),
  54         .key_offset = offsetof(struct tcf_ct_flow_table, zone),
  55         .key_len = sizeof_field(struct tcf_ct_flow_table, zone),
  56         .automatic_shrinking = true,
  57 };
  58
  59 static struct flow_action_entry *
  60 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
  61 {
  62         int i = flow_action->num_entries++;
  63
  64         return &flow_action->entries[i];
  65 }
  66
  67 static void tcf_ct_add_mangle_action(struct flow_action *action,
  68                                      enum flow_action_mangle_base htype,
  69                                      u32 offset,
  70                                      u32 mask,
  71                                      u32 val)
  72 {
  73         struct flow_action_entry *entry;
  74
  75         entry = tcf_ct_flow_table_flow_action_get_next(action);
  76         entry->id = FLOW_ACTION_MANGLE;
  77         entry->mangle.htype = htype;
  78         entry->mangle.mask = ~mask;
  79         entry->mangle.offset = offset;
  80         entry->mangle.val = val;
  81 }
  82
  83 /* The following nat helper functions check if the inverted reverse tuple
  84  * (target) is different then the current dir tuple - meaning nat for ports
  85  * and/or ip is needed, and add the relevant mangle actions.
  86  */
  87 static void
  88 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
  89                                       struct nf_conntrack_tuple target,
  90                                       struct flow_action *action)
  91 {
  92         if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
  93                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
  94                                          offsetof(struct iphdr, saddr),
  95                                          0xFFFFFFFF,
  96                                          be32_to_cpu(target.src.u3.ip));
  97         if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
  98                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
  99                                          offsetof(struct iphdr, daddr),
 100                                          0xFFFFFFFF,
 101                                          be32_to_cpu(target.dst.u3.ip));
 102 }
 103
 104 static void
 105 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
 106                                    union nf_inet_addr *addr,
 107                                    u32 offset)
 108 {
 109         int i;
 110
 111         for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
 112                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
 113                                          i * sizeof(u32) + offset,
 114                                          0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
 115 }
 116
 117 static void
 118 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
 119                                       struct nf_conntrack_tuple target,
 120                                       struct flow_action *action)
 121 {
 122         if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
 123                 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
 124                                                    offsetof(struct ipv6hdr,
 125                                                             saddr));
 126         if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
 127                 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
 128                                                    offsetof(struct ipv6hdr,
 129                                                             daddr));
 130 }
 131
 132 static void
 133 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
 134                                      struct nf_conntrack_tuple target,
 135                                      struct flow_action *action)
 136 {
 137         __be16 target_src = target.src.u.tcp.port;
 138         __be16 target_dst = target.dst.u.tcp.port;
 139
 140         if (target_src != tuple->src.u.tcp.port)
 141                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 142                                          offsetof(struct tcphdr, source),
 143                                          0xFFFF, be16_to_cpu(target_src));
 144         if (target_dst != tuple->dst.u.tcp.port)
 145                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 146                                          offsetof(struct tcphdr, dest),
 147                                          0xFFFF, be16_to_cpu(target_dst));
 148 }
 149
 150 static void
 151 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
 152                                      struct nf_conntrack_tuple target,
 153                                      struct flow_action *action)
 154 {
 155         __be16 target_src = target.src.u.udp.port;
 156         __be16 target_dst = target.dst.u.udp.port;
 157
 158         if (target_src != tuple->src.u.udp.port)
 159                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
 160                                          offsetof(struct udphdr, source),
 161                                          0xFFFF, be16_to_cpu(target_src));
 162         if (target_dst != tuple->dst.u.udp.port)
 163                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
 164                                          offsetof(struct udphdr, dest),
 165                                          0xFFFF, be16_to_cpu(target_dst));
 166 }
 167
 168 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
 169                                               enum ip_conntrack_dir dir,
 170                                               struct flow_action *action)
 171 {
 172         struct nf_conn_labels *ct_labels;
 173         struct flow_action_entry *entry;
 174         enum ip_conntrack_info ctinfo;
 175         u32 *act_ct_labels;
 176
 177         entry = tcf_ct_flow_table_flow_action_get_next(action);
 178         entry->id = FLOW_ACTION_CT_METADATA;
 179 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 180         entry->ct_metadata.mark = ct->mark;
 181 #endif
 182         ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
 183                                              IP_CT_ESTABLISHED_REPLY;
 184         /* aligns with the CT reference on the SKB nf_ct_set */
 185         entry->ct_metadata.cookie = (unsigned long)ct | ctinfo;
 186         entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL;
 187
 188         act_ct_labels = entry->ct_metadata.labels;
 189         ct_labels = nf_ct_labels_find(ct);
 190         if (ct_labels)
 191                 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE);
 192         else
 193                 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE);
 194 }
 195
 196 static int tcf_ct_flow_table_add_action_nat(struct net *net,
 197                                             struct nf_conn *ct,
 198                                             enum ip_conntrack_dir dir,
 199                                             struct flow_action *action)
 200 {
 201         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 202         struct nf_conntrack_tuple target;
 203
 204         if (!(ct->status & IPS_NAT_MASK))
 205                 return 0;
 206
 207         nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
 208
 209         switch (tuple->src.l3num) {
 210         case NFPROTO_IPV4:
 211                 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target,
 212                                                       action);
 213                 break;
 214         case NFPROTO_IPV6:
 215                 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target,
 216                                                       action);
 217                 break;
 218         default:
 219                 return -EOPNOTSUPP;
 220         }
 221
 222         switch (nf_ct_protonum(ct)) {
 223         case IPPROTO_TCP:
 224                 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action);
 225                 break;
 226         case IPPROTO_UDP:
 227                 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action);
 228                 break;
 229         default:
 230                 return -EOPNOTSUPP;
 231         }
 232
 233         return 0;
 234 }
 235
 236 static int tcf_ct_flow_table_fill_actions(struct net *net,
 237                                           const struct flow_offload *flow,
 238                                           enum flow_offload_tuple_dir tdir,
 239                                           struct nf_flow_rule *flow_rule)
 240 {
 241         struct flow_action *action = &flow_rule->rule->action;
 242         int num_entries = action->num_entries;
 243         struct nf_conn *ct = flow->ct;
 244         enum ip_conntrack_dir dir;
 245         int i, err;
 246
 247         switch (tdir) {
 248         case FLOW_OFFLOAD_DIR_ORIGINAL:
 249                 dir = IP_CT_DIR_ORIGINAL;
 250                 break;
 251         case FLOW_OFFLOAD_DIR_REPLY:
 252                 dir = IP_CT_DIR_REPLY;
 253                 break;
 254         default:
 255                 return -EOPNOTSUPP;
 256         }
 257
 258         err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
 259         if (err)
 260                 goto err_nat;
 261
 262         tcf_ct_flow_table_add_action_meta(ct, dir, action);
 263         return 0;
 264
 265 err_nat:
 266         /* Clear filled actions */
 267         for (i = num_entries; i < action->num_entries; i++)
 268                 memset(&action->entries[i], 0, sizeof(action->entries[i]));
 269         action->num_entries = num_entries;
 270
 271         return err;
 272 }
 273
 274 static struct nf_flowtable_type flowtable_ct = {
 275         .action         = tcf_ct_flow_table_fill_actions,
 276         .owner          = THIS_MODULE,
 277 };
 278
 279 static int tcf_ct_flow_table_get(struct tcf_ct_params *params)
 280 {
 281         struct tcf_ct_flow_table *ct_ft;
 282         int err = -ENOMEM;
 283
 284         mutex_lock(&zones_mutex);
 285         ct_ft = rhashtable_lookup_fast(&zones_ht, &params->zone, zones_params);
 286         if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
 287                 goto out_unlock;
 288
 289         ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
 290         if (!ct_ft)
 291                 goto err_alloc;
 292         refcount_set(&ct_ft->ref, 1);
 293
 294         ct_ft->zone = params->zone;
 295         err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
 296         if (err)
 297                 goto err_insert;
 298
 299         ct_ft->nf_ft.type = &flowtable_ct;
 300         ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD |
 301                               NF_FLOWTABLE_COUNTER;
 302         err = nf_flow_table_init(&ct_ft->nf_ft);
 303         if (err)
 304                 goto err_init;
 305
 306         __module_get(THIS_MODULE);
 307 out_unlock:
 308         params->ct_ft = ct_ft;
 309         params->nf_ft = &ct_ft->nf_ft;
 310         mutex_unlock(&zones_mutex);
 311
 312         return 0;
 313
 314 err_init:
 315         rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 316 err_insert:
 317         kfree(ct_ft);
 318 err_alloc:
 319         mutex_unlock(&zones_mutex);
 320         return err;
 321 }
 322
 323 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
 324 {
 325         struct tcf_ct_flow_table *ct_ft;
 326
 327         ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
 328                              rwork);
 329         nf_flow_table_free(&ct_ft->nf_ft);
 330         kfree(ct_ft);
 331
 332         module_put(THIS_MODULE);
 333 }
 334
 335 static void tcf_ct_flow_table_put(struct tcf_ct_params *params)
 336 {
 337         struct tcf_ct_flow_table *ct_ft = params->ct_ft;
 338
 339         if (refcount_dec_and_test(&params->ct_ft->ref)) {
 340                 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 341                 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
 342                 queue_rcu_work(act_ct_wq, &ct_ft->rwork);
 343         }
 344 }
 345
 346 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
 347                                   struct nf_conn *ct,
 348                                   bool tcp)
 349 {
 350         struct flow_offload *entry;
 351         int err;
 352
 353         if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 354                 return;
 355
 356         entry = flow_offload_alloc(ct);
 357         if (!entry) {
 358                 WARN_ON_ONCE(1);
 359                 goto err_alloc;
 360         }
 361
 362         if (tcp) {
 363                 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 364                 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 365         }
 366
 367         err = flow_offload_add(&ct_ft->nf_ft, entry);
 368         if (err)
 369                 goto err_add;
 370
 371         return;
 372
 373 err_add:
 374         flow_offload_free(entry);
 375 err_alloc:
 376         clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 377 }
 378
 379 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
 380                                            struct nf_conn *ct,
 381                                            enum ip_conntrack_info ctinfo)
 382 {
 383         bool tcp = false;
 384
 385         if (ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY)
 386                 return;
 387
 388         switch (nf_ct_protonum(ct)) {
 389         case IPPROTO_TCP:
 390                 tcp = true;
 391                 if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 392                         return;
 393                 break;
 394         case IPPROTO_UDP:
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 401             ct->status & IPS_SEQ_ADJUST)
 402                 return;
 403
 404         tcf_ct_flow_table_add(ct_ft, ct, tcp);
 405 }
 406
 407 static bool
 408 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
 409                                   struct flow_offload_tuple *tuple,
 410                                   struct tcphdr **tcph)
 411 {
 412         struct flow_ports *ports;
 413         unsigned int thoff;
 414         struct iphdr *iph;
 415
 416         if (!pskb_network_may_pull(skb, sizeof(*iph)))
 417                 return false;
 418
 419         iph = ip_hdr(skb);
 420         thoff = iph->ihl * 4;
 421
 422         if (ip_is_fragment(iph) ||
 423             unlikely(thoff != sizeof(struct iphdr)))
 424                 return false;
 425
 426         if (iph->protocol != IPPROTO_TCP &&
 427             iph->protocol != IPPROTO_UDP)
 428                 return false;
 429
 430         if (iph->ttl <= 1)
 431                 return false;
 432
 433         if (!pskb_network_may_pull(skb, iph->protocol == IPPROTO_TCP ?
 434                                         thoff + sizeof(struct tcphdr) :
 435                                         thoff + sizeof(*ports)))
 436                 return false;
 437
 438         iph = ip_hdr(skb);
 439         if (iph->protocol == IPPROTO_TCP)
 440                 *tcph = (void *)(skb_network_header(skb) + thoff);
 441
 442         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 443         tuple->src_v4.s_addr = iph->saddr;
 444         tuple->dst_v4.s_addr = iph->daddr;
 445         tuple->src_port = ports->source;
 446         tuple->dst_port = ports->dest;
 447         tuple->l3proto = AF_INET;
 448         tuple->l4proto = iph->protocol;
 449
 450         return true;
 451 }
 452
 453 static bool
 454 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
 455                                   struct flow_offload_tuple *tuple,
 456                                   struct tcphdr **tcph)
 457 {
 458         struct flow_ports *ports;
 459         struct ipv6hdr *ip6h;
 460         unsigned int thoff;
 461
 462         if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
 463                 return false;
 464
 465         ip6h = ipv6_hdr(skb);
 466
 467         if (ip6h->nexthdr != IPPROTO_TCP &&
 468             ip6h->nexthdr != IPPROTO_UDP)
 469                 return false;
 470
 471         if (ip6h->hop_limit <= 1)
 472                 return false;
 473
 474         thoff = sizeof(*ip6h);
 475         if (!pskb_network_may_pull(skb, ip6h->nexthdr == IPPROTO_TCP ?
 476                                         thoff + sizeof(struct tcphdr) :
 477                                         thoff + sizeof(*ports)))
 478                 return false;
 479
 480         ip6h = ipv6_hdr(skb);
 481         if (ip6h->nexthdr == IPPROTO_TCP)
 482                 *tcph = (void *)(skb_network_header(skb) + thoff);
 483
 484         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 485         tuple->src_v6 = ip6h->saddr;
 486         tuple->dst_v6 = ip6h->daddr;
 487         tuple->src_port = ports->source;
 488         tuple->dst_port = ports->dest;
 489         tuple->l3proto = AF_INET6;
 490         tuple->l4proto = ip6h->nexthdr;
 491
 492         return true;
 493 }
 494
 495 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
 496                                      struct sk_buff *skb,
 497                                      u8 family)
 498 {
 499         struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
 500         struct flow_offload_tuple_rhash *tuplehash;
 501         struct flow_offload_tuple tuple = {};
 502         enum ip_conntrack_info ctinfo;
 503         struct tcphdr *tcph = NULL;
 504         struct flow_offload *flow;
 505         struct nf_conn *ct;
 506         u8 dir;
 507
 508         /* Previously seen or loopback */
 509         ct = nf_ct_get(skb, &ctinfo);
 510         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 511                 return false;
 512
 513         switch (family) {
 514         case NFPROTO_IPV4:
 515                 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
 516                         return false;
 517                 break;
 518         case NFPROTO_IPV6:
 519                 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
 520                         return false;
 521                 break;
 522         default:
 523                 return false;
 524         }
 525
 526         tuplehash = flow_offload_lookup(nf_ft, &tuple);
 527         if (!tuplehash)
 528                 return false;
 529
 530         dir = tuplehash->tuple.dir;
 531         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 532         ct = flow->ct;
 533
 534         if (tcph && (unlikely(tcph->fin || tcph->rst))) {
 535                 flow_offload_teardown(flow);
 536                 return false;
 537         }
 538
 539         ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
 540                                                     IP_CT_ESTABLISHED_REPLY;
 541
 542         flow_offload_refresh(nf_ft, flow);
 543         nf_conntrack_get(&ct->ct_general);
 544         nf_ct_set(skb, ct, ctinfo);
 545         if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
 546                 nf_ct_acct_update(ct, dir, skb->len);
 547
 548         return true;
 549 }
 550
 551 static int tcf_ct_flow_tables_init(void)
 552 {
 553         return rhashtable_init(&zones_ht, &zones_params);
 554 }
 555
 556 static void tcf_ct_flow_tables_uninit(void)
 557 {
 558         rhashtable_destroy(&zones_ht);
 559 }
 560
 561 static struct tc_action_ops act_ct_ops;
 562 static unsigned int ct_net_id;
 563
 564 struct tc_ct_action_net {
 565         struct tc_action_net tn; /* Must be first */
 566         bool labels;
 567 };
 568
 569 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 570 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
 571                                    u16 zone_id, bool force)
 572 {
 573         enum ip_conntrack_info ctinfo;
 574         struct nf_conn *ct;
 575
 576         ct = nf_ct_get(skb, &ctinfo);
 577         if (!ct)
 578                 return false;
 579         if (!net_eq(net, read_pnet(&ct->ct_net)))
 580                 return false;
 581         if (nf_ct_zone(ct)->id != zone_id)
 582                 return false;
 583
 584         /* Force conntrack entry direction. */
 585         if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 586                 if (nf_ct_is_confirmed(ct))
 587                         nf_ct_kill(ct);
 588
 589                 nf_conntrack_put(&ct->ct_general);
 590                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 591
 592                 return false;
 593         }
 594
 595         return true;
 596 }
 597
 598 /* Trim the skb to the length specified by the IP/IPv6 header,
 599  * removing any trailing lower-layer padding. This prepares the skb
 600  * for higher-layer processing that assumes skb->len excludes padding
 601  * (such as nf_ip_checksum). The caller needs to pull the skb to the
 602  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
 603  */
 604 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
 605 {
 606         unsigned int len;
 607         int err;
 608
 609         switch (family) {
 610         case NFPROTO_IPV4:
 611                 len = ntohs(ip_hdr(skb)->tot_len);
 612                 break;
 613         case NFPROTO_IPV6:
 614                 len = sizeof(struct ipv6hdr)
 615                         + ntohs(ipv6_hdr(skb)->payload_len);
 616                 break;
 617         default:
 618                 len = skb->len;
 619         }
 620
 621         err = pskb_trim_rcsum(skb, len);
 622
 623         return err;
 624 }
 625
 626 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
 627 {
 628         u8 family = NFPROTO_UNSPEC;
 629
 630         switch (skb_protocol(skb, true)) {
 631         case htons(ETH_P_IP):
 632                 family = NFPROTO_IPV4;
 633                 break;
 634         case htons(ETH_P_IPV6):
 635                 family = NFPROTO_IPV6;
 636                 break;
 637         default:
 638                 break;
 639         }
 640
 641         return family;
 642 }
 643
 644 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
 645 {
 646         unsigned int len;
 647
 648         len =  skb_network_offset(skb) + sizeof(struct iphdr);
 649         if (unlikely(skb->len < len))
 650                 return -EINVAL;
 651         if (unlikely(!pskb_may_pull(skb, len)))
 652                 return -ENOMEM;
 653
 654         *frag = ip_is_fragment(ip_hdr(skb));
 655         return 0;
 656 }
 657
 658 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
 659 {
 660         unsigned int flags = 0, len, payload_ofs = 0;
 661         unsigned short frag_off;
 662         int nexthdr;
 663
 664         len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
 665         if (unlikely(skb->len < len))
 666                 return -EINVAL;
 667         if (unlikely(!pskb_may_pull(skb, len)))
 668                 return -ENOMEM;
 669
 670         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 671         if (unlikely(nexthdr < 0))
 672                 return -EPROTO;
 673
 674         *frag = flags & IP6_FH_F_FRAG;
 675         return 0;
 676 }
 677
 678 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
 679                                    u8 family, u16 zone, bool *defrag)
 680 {
 681         enum ip_conntrack_info ctinfo;
 682         struct qdisc_skb_cb cb;
 683         struct nf_conn *ct;
 684         int err = 0;
 685         bool frag;
 686
 687         /* Previously seen (loopback)? Ignore. */
 688         ct = nf_ct_get(skb, &ctinfo);
 689         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 690                 return 0;
 691
 692         if (family == NFPROTO_IPV4)
 693                 err = tcf_ct_ipv4_is_fragment(skb, &frag);
 694         else
 695                 err = tcf_ct_ipv6_is_fragment(skb, &frag);
 696         if (err || !frag)
 697                 return err;
 698
 699         skb_get(skb);
 700         cb = *qdisc_skb_cb(skb);
 701
 702         if (family == NFPROTO_IPV4) {
 703                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 704
 705                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 706                 local_bh_disable();
 707                 err = ip_defrag(net, skb, user);
 708                 local_bh_enable();
 709                 if (err && err != -EINPROGRESS)
 710                         return err;
 711
 712                 if (!err) {
 713                         *defrag = true;
 714                         cb.mru = IPCB(skb)->frag_max_size;
 715                 }
 716         } else { /* NFPROTO_IPV6 */
 717 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 718                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 719
 720                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 721                 err = nf_ct_frag6_gather(net, skb, user);
 722                 if (err && err != -EINPROGRESS)
 723                         goto out_free;
 724
 725                 if (!err) {
 726                         *defrag = true;
 727                         cb.mru = IP6CB(skb)->frag_max_size;
 728                 }
 729 #else
 730                 err = -EOPNOTSUPP;
 731                 goto out_free;
 732 #endif
 733         }
 734
 735         if (err != -EINPROGRESS)
 736                 *qdisc_skb_cb(skb) = cb;
 737         skb_clear_hash(skb);
 738         skb->ignore_df = 1;
 739         return err;
 740
 741 out_free:
 742         kfree_skb(skb);
 743         return err;
 744 }
 745
 746 static void tcf_ct_params_free(struct rcu_head *head)
 747 {
 748         struct tcf_ct_params *params = container_of(head,
 749                                                     struct tcf_ct_params, rcu);
 750
 751         tcf_ct_flow_table_put(params);
 752
 753         if (params->tmpl)
 754                 nf_conntrack_put(&params->tmpl->ct_general);
 755         kfree(params);
 756 }
 757
 758 #if IS_ENABLED(CONFIG_NF_NAT)
 759 /* Modelled after nf_nat_ipv[46]_fn().
 760  * range is only used for new, uninitialized NAT state.
 761  * Returns either NF_ACCEPT or NF_DROP.
 762  */
 763 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 764                           enum ip_conntrack_info ctinfo,
 765                           const struct nf_nat_range2 *range,
 766                           enum nf_nat_manip_type maniptype)
 767 {
 768         __be16 proto = skb_protocol(skb, true);
 769         int hooknum, err = NF_ACCEPT;
 770
 771         /* See HOOK2MANIP(). */
 772         if (maniptype == NF_NAT_MANIP_SRC)
 773                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 774         else
 775                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 776
 777         switch (ctinfo) {
 778         case IP_CT_RELATED:
 779         case IP_CT_RELATED_REPLY:
 780                 if (proto == htons(ETH_P_IP) &&
 781                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 782                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 783                                                            hooknum))
 784                                 err = NF_DROP;
 785                         goto out;
 786                 } else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) {
 787                         __be16 frag_off;
 788                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 789                         int hdrlen = ipv6_skip_exthdr(skb,
 790                                                       sizeof(struct ipv6hdr),
 791                                                       &nexthdr, &frag_off);
 792
 793                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 794                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 795                                                                      ctinfo,
 796                                                                      hooknum,
 797                                                                      hdrlen))
 798                                         err = NF_DROP;
 799                                 goto out;
 800                         }
 801                 }
 802                 /* Non-ICMP, fall thru to initialize if needed. */
 803                 fallthrough;
 804         case IP_CT_NEW:
 805                 /* Seen it before?  This can happen for loopback, retrans,
 806                  * or local packets.
 807                  */
 808                 if (!nf_nat_initialized(ct, maniptype)) {
 809                         /* Initialize according to the NAT action. */
 810                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 811                                 /* Action is set up to establish a new
 812                                  * mapping.
 813                                  */
 814                                 ? nf_nat_setup_info(ct, range, maniptype)
 815                                 : nf_nat_alloc_null_binding(ct, hooknum);
 816                         if (err != NF_ACCEPT)
 817                                 goto out;
 818                 }
 819                 break;
 820
 821         case IP_CT_ESTABLISHED:
 822         case IP_CT_ESTABLISHED_REPLY:
 823                 break;
 824
 825         default:
 826                 err = NF_DROP;
 827                 goto out;
 828         }
 829
 830         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 831 out:
 832         return err;
 833 }
 834 #endif /* CONFIG_NF_NAT */
 835
 836 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
 837 {
 838 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 839         u32 new_mark;
 840
 841         if (!mask)
 842                 return;
 843
 844         new_mark = mark | (ct->mark & ~(mask));
 845         if (ct->mark != new_mark) {
 846                 ct->mark = new_mark;
 847                 if (nf_ct_is_confirmed(ct))
 848                         nf_conntrack_event_cache(IPCT_MARK, ct);
 849         }
 850 #endif
 851 }
 852
 853 static void tcf_ct_act_set_labels(struct nf_conn *ct,
 854                                   u32 *labels,
 855                                   u32 *labels_m)
 856 {
 857 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
 858         size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
 859
 860         if (!memchr_inv(labels_m, 0, labels_sz))
 861                 return;
 862
 863         nf_connlabels_replace(ct, labels, labels_m, 4);
 864 #endif
 865 }
 866
 867 static int tcf_ct_act_nat(struct sk_buff *skb,
 868                           struct nf_conn *ct,
 869                           enum ip_conntrack_info ctinfo,
 870                           int ct_action,
 871                           struct nf_nat_range2 *range,
 872                           bool commit)
 873 {
 874 #if IS_ENABLED(CONFIG_NF_NAT)
 875         int err;
 876         enum nf_nat_manip_type maniptype;
 877
 878         if (!(ct_action & TCA_CT_ACT_NAT))
 879                 return NF_ACCEPT;
 880
 881         /* Add NAT extension if not confirmed yet. */
 882         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 883                 return NF_DROP;   /* Can't NAT. */
 884
 885         if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
 886             (ctinfo != IP_CT_RELATED || commit)) {
 887                 /* NAT an established or related connection like before. */
 888                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 889                         /* This is the REPLY direction for a connection
 890                          * for which NAT was applied in the forward
 891                          * direction.  Do the reverse NAT.
 892                          */
 893                         maniptype = ct->status & IPS_SRC_NAT
 894                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 895                 else
 896                         maniptype = ct->status & IPS_SRC_NAT
 897                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 898         } else if (ct_action & TCA_CT_ACT_NAT_SRC) {
 899                 maniptype = NF_NAT_MANIP_SRC;
 900         } else if (ct_action & TCA_CT_ACT_NAT_DST) {
 901                 maniptype = NF_NAT_MANIP_DST;
 902         } else {
 903                 return NF_ACCEPT;
 904         }
 905
 906         err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
 907         if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
 908                 if (ct->status & IPS_SRC_NAT) {
 909                         if (maniptype == NF_NAT_MANIP_SRC)
 910                                 maniptype = NF_NAT_MANIP_DST;
 911                         else
 912                                 maniptype = NF_NAT_MANIP_SRC;
 913
 914                         err = ct_nat_execute(skb, ct, ctinfo, range,
 915                                              maniptype);
 916                 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 917                         err = ct_nat_execute(skb, ct, ctinfo, NULL,
 918                                              NF_NAT_MANIP_SRC);
 919                 }
 920         }
 921         return err;
 922 #else
 923         return NF_ACCEPT;
 924 #endif
 925 }
 926
 927 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 928                       struct tcf_result *res)
 929 {
 930         struct net *net = dev_net(skb->dev);
 931         bool cached, commit, clear, force;
 932         enum ip_conntrack_info ctinfo;
 933         struct tcf_ct *c = to_ct(a);
 934         struct nf_conn *tmpl = NULL;
 935         struct nf_hook_state state;
 936         int nh_ofs, err, retval;
 937         struct tcf_ct_params *p;
 938         bool skip_add = false;
 939         bool defrag = false;
 940         struct nf_conn *ct;
 941         u8 family;
 942
 943         p = rcu_dereference_bh(c->params);
 944
 945         retval = READ_ONCE(c->tcf_action);
 946         commit = p->ct_action & TCA_CT_ACT_COMMIT;
 947         clear = p->ct_action & TCA_CT_ACT_CLEAR;
 948         force = p->ct_action & TCA_CT_ACT_FORCE;
 949         tmpl = p->tmpl;
 950
 951         tcf_lastuse_update(&c->tcf_tm);
 952
 953         if (clear) {
 954                 qdisc_skb_cb(skb)->post_ct = false;
 955                 ct = nf_ct_get(skb, &ctinfo);
 956                 if (ct) {
 957                         nf_conntrack_put(&ct->ct_general);
 958                         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 959                 }
 960
 961                 goto out_clear;
 962         }
 963
 964         family = tcf_ct_skb_nf_family(skb);
 965         if (family == NFPROTO_UNSPEC)
 966                 goto drop;
 967
 968         /* The conntrack module expects to be working at L3.
 969          * We also try to pull the IPv4/6 header to linear area
 970          */
 971         nh_ofs = skb_network_offset(skb);
 972         skb_pull_rcsum(skb, nh_ofs);
 973         err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
 974         if (err == -EINPROGRESS) {
 975                 retval = TC_ACT_STOLEN;
 976                 goto out_clear;
 977         }
 978         if (err)
 979                 goto drop;
 980
 981         err = tcf_ct_skb_network_trim(skb, family);
 982         if (err)
 983                 goto drop;
 984
 985         /* If we are recirculating packets to match on ct fields and
 986          * committing with a separate ct action, then we don't need to
 987          * actually run the packet through conntrack twice unless it's for a
 988          * different zone.
 989          */
 990         cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
 991         if (!cached) {
 992                 if (tcf_ct_flow_table_lookup(p, skb, family)) {
 993                         skip_add = true;
 994                         goto do_nat;
 995                 }
 996
 997                 /* Associate skb with specified zone. */
 998                 if (tmpl) {
 999                         nf_conntrack_put(skb_nfct(skb));
1000                         nf_conntrack_get(&tmpl->ct_general);
1001                         nf_ct_set(skb, tmpl, IP_CT_NEW);
1002                 }
1003
1004                 state.hook = NF_INET_PRE_ROUTING;
1005                 state.net = net;
1006                 state.pf = family;
1007                 err = nf_conntrack_in(skb, &state);
1008                 if (err != NF_ACCEPT)
1009                         goto out_push;
1010         }
1011
1012 do_nat:
1013         ct = nf_ct_get(skb, &ctinfo);
1014         if (!ct)
1015                 goto out_push;
1016         nf_ct_deliver_cached_events(ct);
1017
1018         err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1019         if (err != NF_ACCEPT)
1020                 goto drop;
1021
1022         if (commit) {
1023                 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1024                 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1025
1026                 /* This will take care of sending queued events
1027                  * even if the connection is already confirmed.
1028                  */
1029                 nf_conntrack_confirm(skb);
1030         }
1031
1032         if (!skip_add)
1033                 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1034
1035 out_push:
1036         skb_push_rcsum(skb, nh_ofs);
1037
1038         qdisc_skb_cb(skb)->post_ct = true;
1039 out_clear:
1040         tcf_action_update_bstats(&c->common, skb);
1041         if (defrag)
1042                 qdisc_skb_cb(skb)->pkt_len = skb->len;
1043         return retval;
1044
1045 drop:
1046         tcf_action_inc_drop_qstats(&c->common);
1047         return TC_ACT_SHOT;
1048 }
1049
1050 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1051         [TCA_CT_ACTION] = { .type = NLA_U16 },
1052         [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1053         [TCA_CT_ZONE] = { .type = NLA_U16 },
1054         [TCA_CT_MARK] = { .type = NLA_U32 },
1055         [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1056         [TCA_CT_LABELS] = { .type = NLA_BINARY,
1057                             .len = 128 / BITS_PER_BYTE },
1058         [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1059                                  .len = 128 / BITS_PER_BYTE },
1060         [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1061         [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1062         [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1063         [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1064         [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1065         [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1066 };
1067
1068 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1069                                   struct tc_ct *parm,
1070                                   struct nlattr **tb,
1071                                   struct netlink_ext_ack *extack)
1072 {
1073         struct nf_nat_range2 *range;
1074
1075         if (!(p->ct_action & TCA_CT_ACT_NAT))
1076                 return 0;
1077
1078         if (!IS_ENABLED(CONFIG_NF_NAT)) {
1079                 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1080                 return -EOPNOTSUPP;
1081         }
1082
1083         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1084                 return 0;
1085
1086         if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1087             (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1088                 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1089                 return -EOPNOTSUPP;
1090         }
1091
1092         range = &p->range;
1093         if (tb[TCA_CT_NAT_IPV4_MIN]) {
1094                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1095
1096                 p->ipv4_range = true;
1097                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1098                 range->min_addr.ip =
1099                         nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1100
1101                 range->max_addr.ip = max_attr ?
1102                                      nla_get_in_addr(max_attr) :
1103                                      range->min_addr.ip;
1104         } else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1105                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1106
1107                 p->ipv4_range = false;
1108                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1109                 range->min_addr.in6 =
1110                         nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1111
1112                 range->max_addr.in6 = max_attr ?
1113                                       nla_get_in6_addr(max_attr) :
1114                                       range->min_addr.in6;
1115         }
1116
1117         if (tb[TCA_CT_NAT_PORT_MIN]) {
1118                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1119                 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1120
1121                 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1122                                        nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1123                                        range->min_proto.all;
1124         }
1125
1126         return 0;
1127 }
1128
1129 static void tcf_ct_set_key_val(struct nlattr **tb,
1130                                void *val, int val_type,
1131                                void *mask, int mask_type,
1132                                int len)
1133 {
1134         if (!tb[val_type])
1135                 return;
1136         nla_memcpy(val, tb[val_type], len);
1137
1138         if (!mask)
1139                 return;
1140
1141         if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1142                 memset(mask, 0xff, len);
1143         else
1144                 nla_memcpy(mask, tb[mask_type], len);
1145 }
1146
1147 static int tcf_ct_fill_params(struct net *net,
1148                               struct tcf_ct_params *p,
1149                               struct tc_ct *parm,
1150                               struct nlattr **tb,
1151                               struct netlink_ext_ack *extack)
1152 {
1153         struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1154         struct nf_conntrack_zone zone;
1155         struct nf_conn *tmpl;
1156         int err;
1157
1158         p->zone = NF_CT_DEFAULT_ZONE_ID;
1159
1160         tcf_ct_set_key_val(tb,
1161                            &p->ct_action, TCA_CT_ACTION,
1162                            NULL, TCA_CT_UNSPEC,
1163                            sizeof(p->ct_action));
1164
1165         if (p->ct_action & TCA_CT_ACT_CLEAR)
1166                 return 0;
1167
1168         err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1169         if (err)
1170                 return err;
1171
1172         if (tb[TCA_CT_MARK]) {
1173                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1174                         NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1175                         return -EOPNOTSUPP;
1176                 }
1177                 tcf_ct_set_key_val(tb,
1178                                    &p->mark, TCA_CT_MARK,
1179                                    &p->mark_mask, TCA_CT_MARK_MASK,
1180                                    sizeof(p->mark));
1181         }
1182
1183         if (tb[TCA_CT_LABELS]) {
1184                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1185                         NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1186                         return -EOPNOTSUPP;
1187                 }
1188
1189                 if (!tn->labels) {
1190                         NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1191                         return -EOPNOTSUPP;
1192                 }
1193                 tcf_ct_set_key_val(tb,
1194                                    p->labels, TCA_CT_LABELS,
1195                                    p->labels_mask, TCA_CT_LABELS_MASK,
1196                                    sizeof(p->labels));
1197         }
1198
1199         if (tb[TCA_CT_ZONE]) {
1200                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1201                         NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1202                         return -EOPNOTSUPP;
1203                 }
1204
1205                 tcf_ct_set_key_val(tb,
1206                                    &p->zone, TCA_CT_ZONE,
1207                                    NULL, TCA_CT_UNSPEC,
1208                                    sizeof(p->zone));
1209         }
1210
1211         nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1212         tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1213         if (!tmpl) {
1214                 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1215                 return -ENOMEM;
1216         }
1217         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1218         nf_conntrack_get(&tmpl->ct_general);
1219         p->tmpl = tmpl;
1220
1221         return 0;
1222 }
1223
1224 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1225                        struct nlattr *est, struct tc_action **a,
1226                        int replace, int bind, bool rtnl_held,
1227                        struct tcf_proto *tp, u32 flags,
1228                        struct netlink_ext_ack *extack)
1229 {
1230         struct tc_action_net *tn = net_generic(net, ct_net_id);
1231         struct tcf_ct_params *params = NULL;
1232         struct nlattr *tb[TCA_CT_MAX + 1];
1233         struct tcf_chain *goto_ch = NULL;
1234         struct tc_ct *parm;
1235         struct tcf_ct *c;
1236         int err, res = 0;
1237         u32 index;
1238
1239         if (!nla) {
1240                 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1241                 return -EINVAL;
1242         }
1243
1244         err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1245         if (err < 0)
1246                 return err;
1247
1248         if (!tb[TCA_CT_PARMS]) {
1249                 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1250                 return -EINVAL;
1251         }
1252         parm = nla_data(tb[TCA_CT_PARMS]);
1253         index = parm->index;
1254         err = tcf_idr_check_alloc(tn, &index, a, bind);
1255         if (err < 0)
1256                 return err;
1257
1258         if (!err) {
1259                 err = tcf_idr_create_from_flags(tn, index, est, a,
1260                                                 &act_ct_ops, bind, flags);
1261                 if (err) {
1262                         tcf_idr_cleanup(tn, index);
1263                         return err;
1264                 }
1265                 res = ACT_P_CREATED;
1266         } else {
1267                 if (bind)
1268                         return 0;
1269
1270                 if (!replace) {
1271                         tcf_idr_release(*a, bind);
1272                         return -EEXIST;
1273                 }
1274         }
1275         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1276         if (err < 0)
1277                 goto cleanup;
1278
1279         c = to_ct(*a);
1280
1281         params = kzalloc(sizeof(*params), GFP_KERNEL);
1282         if (unlikely(!params)) {
1283                 err = -ENOMEM;
1284                 goto cleanup;
1285         }
1286
1287         err = tcf_ct_fill_params(net, params, parm, tb, extack);
1288         if (err)
1289                 goto cleanup;
1290
1291         err = tcf_ct_flow_table_get(params);
1292         if (err)
1293                 goto cleanup;
1294
1295         spin_lock_bh(&c->tcf_lock);
1296         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1297         params = rcu_replace_pointer(c->params, params,
1298                                      lockdep_is_held(&c->tcf_lock));
1299         spin_unlock_bh(&c->tcf_lock);
1300
1301         if (goto_ch)
1302                 tcf_chain_put_by_act(goto_ch);
1303         if (params)
1304                 call_rcu(&params->rcu, tcf_ct_params_free);
1305
1306         return res;
1307
1308 cleanup:
1309         if (goto_ch)
1310                 tcf_chain_put_by_act(goto_ch);
1311         kfree(params);
1312         tcf_idr_release(*a, bind);
1313         return err;
1314 }
1315
1316 static void tcf_ct_cleanup(struct tc_action *a)
1317 {
1318         struct tcf_ct_params *params;
1319         struct tcf_ct *c = to_ct(a);
1320
1321         params = rcu_dereference_protected(c->params, 1);
1322         if (params)
1323                 call_rcu(&params->rcu, tcf_ct_params_free);
1324 }
1325
1326 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1327                                void *val, int val_type,
1328                                void *mask, int mask_type,
1329                                int len)
1330 {
1331         int err;
1332
1333         if (mask && !memchr_inv(mask, 0, len))
1334                 return 0;
1335
1336         err = nla_put(skb, val_type, len, val);
1337         if (err)
1338                 return err;
1339
1340         if (mask_type != TCA_CT_UNSPEC) {
1341                 err = nla_put(skb, mask_type, len, mask);
1342                 if (err)
1343                         return err;
1344         }
1345
1346         return 0;
1347 }
1348
1349 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1350 {
1351         struct nf_nat_range2 *range = &p->range;
1352
1353         if (!(p->ct_action & TCA_CT_ACT_NAT))
1354                 return 0;
1355
1356         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1357                 return 0;
1358
1359         if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1360                 if (p->ipv4_range) {
1361                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1362                                             range->min_addr.ip))
1363                                 return -1;
1364                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1365                                             range->max_addr.ip))
1366                                 return -1;
1367                 } else {
1368                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1369                                              &range->min_addr.in6))
1370                                 return -1;
1371                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1372                                              &range->max_addr.in6))
1373                                 return -1;
1374                 }
1375         }
1376
1377         if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1378                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1379                                  range->min_proto.all))
1380                         return -1;
1381                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1382                                  range->max_proto.all))
1383                         return -1;
1384         }
1385
1386         return 0;
1387 }
1388
1389 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1390                               int bind, int ref)
1391 {
1392         unsigned char *b = skb_tail_pointer(skb);
1393         struct tcf_ct *c = to_ct(a);
1394         struct tcf_ct_params *p;
1395
1396         struct tc_ct opt = {
1397                 .index   = c->tcf_index,
1398                 .refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1399                 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1400         };
1401         struct tcf_t t;
1402
1403         spin_lock_bh(&c->tcf_lock);
1404         p = rcu_dereference_protected(c->params,
1405                                       lockdep_is_held(&c->tcf_lock));
1406         opt.action = c->tcf_action;
1407
1408         if (tcf_ct_dump_key_val(skb,
1409                                 &p->ct_action, TCA_CT_ACTION,
1410                                 NULL, TCA_CT_UNSPEC,
1411                                 sizeof(p->ct_action)))
1412                 goto nla_put_failure;
1413
1414         if (p->ct_action & TCA_CT_ACT_CLEAR)
1415                 goto skip_dump;
1416
1417         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1418             tcf_ct_dump_key_val(skb,
1419                                 &p->mark, TCA_CT_MARK,
1420                                 &p->mark_mask, TCA_CT_MARK_MASK,
1421                                 sizeof(p->mark)))
1422                 goto nla_put_failure;
1423
1424         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1425             tcf_ct_dump_key_val(skb,
1426                                 p->labels, TCA_CT_LABELS,
1427                                 p->labels_mask, TCA_CT_LABELS_MASK,
1428                                 sizeof(p->labels)))
1429                 goto nla_put_failure;
1430
1431         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1432             tcf_ct_dump_key_val(skb,
1433                                 &p->zone, TCA_CT_ZONE,
1434                                 NULL, TCA_CT_UNSPEC,
1435                                 sizeof(p->zone)))
1436                 goto nla_put_failure;
1437
1438         if (tcf_ct_dump_nat(skb, p))
1439                 goto nla_put_failure;
1440
1441 skip_dump:
1442         if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1443                 goto nla_put_failure;
1444
1445         tcf_tm_dump(&t, &c->tcf_tm);
1446         if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1447                 goto nla_put_failure;
1448         spin_unlock_bh(&c->tcf_lock);
1449
1450         return skb->len;
1451 nla_put_failure:
1452         spin_unlock_bh(&c->tcf_lock);
1453         nlmsg_trim(skb, b);
1454         return -1;
1455 }
1456
1457 static int tcf_ct_walker(struct net *net, struct sk_buff *skb,
1458                          struct netlink_callback *cb, int type,
1459                          const struct tc_action_ops *ops,
1460                          struct netlink_ext_ack *extack)
1461 {
1462         struct tc_action_net *tn = net_generic(net, ct_net_id);
1463
1464         return tcf_generic_walker(tn, skb, cb, type, ops, extack);
1465 }
1466
1467 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index)
1468 {
1469         struct tc_action_net *tn = net_generic(net, ct_net_id);
1470
1471         return tcf_idr_search(tn, a, index);
1472 }
1473
1474 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1475                              u64 drops, u64 lastuse, bool hw)
1476 {
1477         struct tcf_ct *c = to_ct(a);
1478
1479         tcf_action_update_stats(a, bytes, packets, drops, hw);
1480         c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1481 }
1482
1483 static struct tc_action_ops act_ct_ops = {
1484         .kind           =       "ct",
1485         .id             =       TCA_ID_CT,
1486         .owner          =       THIS_MODULE,
1487         .act            =       tcf_ct_act,
1488         .dump           =       tcf_ct_dump,
1489         .init           =       tcf_ct_init,
1490         .cleanup        =       tcf_ct_cleanup,
1491         .walk           =       tcf_ct_walker,
1492         .lookup         =       tcf_ct_search,
1493         .stats_update   =       tcf_stats_update,
1494         .size           =       sizeof(struct tcf_ct),
1495 };
1496
1497 static __net_init int ct_init_net(struct net *net)
1498 {
1499         unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1500         struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1501
1502         if (nf_connlabels_get(net, n_bits - 1)) {
1503                 tn->labels = false;
1504                 pr_err("act_ct: Failed to set connlabels length");
1505         } else {
1506                 tn->labels = true;
1507         }
1508
1509         return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1510 }
1511
1512 static void __net_exit ct_exit_net(struct list_head *net_list)
1513 {
1514         struct net *net;
1515
1516         rtnl_lock();
1517         list_for_each_entry(net, net_list, exit_list) {
1518                 struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1519
1520                 if (tn->labels)
1521                         nf_connlabels_put(net);
1522         }
1523         rtnl_unlock();
1524
1525         tc_action_net_exit(net_list, ct_net_id);
1526 }
1527
1528 static struct pernet_operations ct_net_ops = {
1529         .init = ct_init_net,
1530         .exit_batch = ct_exit_net,
1531         .id   = &ct_net_id,
1532         .size = sizeof(struct tc_ct_action_net),
1533 };
1534
1535 static int __init ct_init_module(void)
1536 {
1537         int err;
1538
1539         act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1540         if (!act_ct_wq)
1541                 return -ENOMEM;
1542
1543         err = tcf_ct_flow_tables_init();
1544         if (err)
1545                 goto err_tbl_init;
1546
1547         err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1548         if (err)
1549                 goto err_register;
1550
1551         static_branch_inc(&tcf_frag_xmit_count);
1552
1553         return 0;
1554
1555 err_register:
1556         tcf_ct_flow_tables_uninit();
1557 err_tbl_init:
1558         destroy_workqueue(act_ct_wq);
1559         return err;
1560 }
1561
1562 static void __exit ct_cleanup_module(void)
1563 {
1564         static_branch_dec(&tcf_frag_xmit_count);
1565         tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1566         tcf_ct_flow_tables_uninit();
1567         destroy_workqueue(act_ct_wq);
1568 }
1569
1570 module_init(ct_init_module);
1571 module_exit(ct_cleanup_module);
1572 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1573 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1574 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1575 MODULE_DESCRIPTION("Connection tracking action");
1576 MODULE_LICENSE("GPL v2");