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