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 <net/netfilter/nf_conntrack_act_ct.h>
  36 #include <uapi/linux/netfilter/nf_nat.h>
  37
  38 static struct workqueue_struct *act_ct_wq;
  39 static struct rhashtable zones_ht;
  40 static DEFINE_MUTEX(zones_mutex);
  41
  42 struct tcf_ct_flow_table {
  43         struct rhash_head node; /* In zones tables */
  44
  45         struct rcu_work rwork;
  46         struct nf_flowtable nf_ft;
  47         refcount_t ref;
  48         u16 zone;
  49
  50         bool dying;
  51 };
  52
  53 static const struct rhashtable_params zones_params = {
  54         .head_offset = offsetof(struct tcf_ct_flow_table, node),
  55         .key_offset = offsetof(struct tcf_ct_flow_table, zone),
  56         .key_len = sizeof_field(struct tcf_ct_flow_table, zone),
  57         .automatic_shrinking = true,
  58 };
  59
  60 static struct flow_action_entry *
  61 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
  62 {
  63         int i = flow_action->num_entries++;
  64
  65         return &flow_action->entries[i];
  66 }
  67
  68 static void tcf_ct_add_mangle_action(struct flow_action *action,
  69                                      enum flow_action_mangle_base htype,
  70                                      u32 offset,
  71                                      u32 mask,
  72                                      u32 val)
  73 {
  74         struct flow_action_entry *entry;
  75
  76         entry = tcf_ct_flow_table_flow_action_get_next(action);
  77         entry->id = FLOW_ACTION_MANGLE;
  78         entry->mangle.htype = htype;
  79         entry->mangle.mask = ~mask;
  80         entry->mangle.offset = offset;
  81         entry->mangle.val = val;
  82 }
  83
  84 /* The following nat helper functions check if the inverted reverse tuple
  85  * (target) is different then the current dir tuple - meaning nat for ports
  86  * and/or ip is needed, and add the relevant mangle actions.
  87  */
  88 static void
  89 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
  90                                       struct nf_conntrack_tuple target,
  91                                       struct flow_action *action)
  92 {
  93         if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
  94                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
  95                                          offsetof(struct iphdr, saddr),
  96                                          0xFFFFFFFF,
  97                                          be32_to_cpu(target.src.u3.ip));
  98         if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
  99                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
 100                                          offsetof(struct iphdr, daddr),
 101                                          0xFFFFFFFF,
 102                                          be32_to_cpu(target.dst.u3.ip));
 103 }
 104
 105 static void
 106 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
 107                                    union nf_inet_addr *addr,
 108                                    u32 offset)
 109 {
 110         int i;
 111
 112         for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
 113                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
 114                                          i * sizeof(u32) + offset,
 115                                          0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
 116 }
 117
 118 static void
 119 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
 120                                       struct nf_conntrack_tuple target,
 121                                       struct flow_action *action)
 122 {
 123         if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
 124                 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
 125                                                    offsetof(struct ipv6hdr,
 126                                                             saddr));
 127         if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
 128                 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
 129                                                    offsetof(struct ipv6hdr,
 130                                                             daddr));
 131 }
 132
 133 static void
 134 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
 135                                      struct nf_conntrack_tuple target,
 136                                      struct flow_action *action)
 137 {
 138         __be16 target_src = target.src.u.tcp.port;
 139         __be16 target_dst = target.dst.u.tcp.port;
 140
 141         if (target_src != tuple->src.u.tcp.port)
 142                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 143                                          offsetof(struct tcphdr, source),
 144                                          0xFFFF, be16_to_cpu(target_src));
 145         if (target_dst != tuple->dst.u.tcp.port)
 146                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 147                                          offsetof(struct tcphdr, dest),
 148                                          0xFFFF, be16_to_cpu(target_dst));
 149 }
 150
 151 static void
 152 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
 153                                      struct nf_conntrack_tuple target,
 154                                      struct flow_action *action)
 155 {
 156         __be16 target_src = target.src.u.udp.port;
 157         __be16 target_dst = target.dst.u.udp.port;
 158
 159         if (target_src != tuple->src.u.udp.port)
 160                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
 161                                          offsetof(struct udphdr, source),
 162                                          0xFFFF, be16_to_cpu(target_src));
 163         if (target_dst != tuple->dst.u.udp.port)
 164                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
 165                                          offsetof(struct udphdr, dest),
 166                                          0xFFFF, be16_to_cpu(target_dst));
 167 }
 168
 169 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
 170                                               enum ip_conntrack_dir dir,
 171                                               struct flow_action *action)
 172 {
 173         struct nf_conn_labels *ct_labels;
 174         struct flow_action_entry *entry;
 175         enum ip_conntrack_info ctinfo;
 176         u32 *act_ct_labels;
 177
 178         entry = tcf_ct_flow_table_flow_action_get_next(action);
 179         entry->id = FLOW_ACTION_CT_METADATA;
 180 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 181         entry->ct_metadata.mark = READ_ONCE(ct->mark);
 182 #endif
 183         ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
 184                                              IP_CT_ESTABLISHED_REPLY;
 185         /* aligns with the CT reference on the SKB nf_ct_set */
 186         entry->ct_metadata.cookie = (unsigned long)ct | ctinfo;
 187         entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL;
 188
 189         act_ct_labels = entry->ct_metadata.labels;
 190         ct_labels = nf_ct_labels_find(ct);
 191         if (ct_labels)
 192                 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE);
 193         else
 194                 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE);
 195 }
 196
 197 static int tcf_ct_flow_table_add_action_nat(struct net *net,
 198                                             struct nf_conn *ct,
 199                                             enum ip_conntrack_dir dir,
 200                                             struct flow_action *action)
 201 {
 202         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 203         struct nf_conntrack_tuple target;
 204
 205         if (!(ct->status & IPS_NAT_MASK))
 206                 return 0;
 207
 208         nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
 209
 210         switch (tuple->src.l3num) {
 211         case NFPROTO_IPV4:
 212                 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target,
 213                                                       action);
 214                 break;
 215         case NFPROTO_IPV6:
 216                 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target,
 217                                                       action);
 218                 break;
 219         default:
 220                 return -EOPNOTSUPP;
 221         }
 222
 223         switch (nf_ct_protonum(ct)) {
 224         case IPPROTO_TCP:
 225                 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action);
 226                 break;
 227         case IPPROTO_UDP:
 228                 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action);
 229                 break;
 230         default:
 231                 return -EOPNOTSUPP;
 232         }
 233
 234         return 0;
 235 }
 236
 237 static int tcf_ct_flow_table_fill_actions(struct net *net,
 238                                           const struct flow_offload *flow,
 239                                           enum flow_offload_tuple_dir tdir,
 240                                           struct nf_flow_rule *flow_rule)
 241 {
 242         struct flow_action *action = &flow_rule->rule->action;
 243         int num_entries = action->num_entries;
 244         struct nf_conn *ct = flow->ct;
 245         enum ip_conntrack_dir dir;
 246         int i, err;
 247
 248         switch (tdir) {
 249         case FLOW_OFFLOAD_DIR_ORIGINAL:
 250                 dir = IP_CT_DIR_ORIGINAL;
 251                 break;
 252         case FLOW_OFFLOAD_DIR_REPLY:
 253                 dir = IP_CT_DIR_REPLY;
 254                 break;
 255         default:
 256                 return -EOPNOTSUPP;
 257         }
 258
 259         err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
 260         if (err)
 261                 goto err_nat;
 262
 263         tcf_ct_flow_table_add_action_meta(ct, dir, action);
 264         return 0;
 265
 266 err_nat:
 267         /* Clear filled actions */
 268         for (i = num_entries; i < action->num_entries; i++)
 269                 memset(&action->entries[i], 0, sizeof(action->entries[i]));
 270         action->num_entries = num_entries;
 271
 272         return err;
 273 }
 274
 275 static struct nf_flowtable_type flowtable_ct = {
 276         .action         = tcf_ct_flow_table_fill_actions,
 277         .owner          = THIS_MODULE,
 278 };
 279
 280 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params)
 281 {
 282         struct tcf_ct_flow_table *ct_ft;
 283         int err = -ENOMEM;
 284
 285         mutex_lock(&zones_mutex);
 286         ct_ft = rhashtable_lookup_fast(&zones_ht, &params->zone, zones_params);
 287         if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
 288                 goto out_unlock;
 289
 290         ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
 291         if (!ct_ft)
 292                 goto err_alloc;
 293         refcount_set(&ct_ft->ref, 1);
 294
 295         ct_ft->zone = params->zone;
 296         err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
 297         if (err)
 298                 goto err_insert;
 299
 300         ct_ft->nf_ft.type = &flowtable_ct;
 301         ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD |
 302                               NF_FLOWTABLE_COUNTER;
 303         err = nf_flow_table_init(&ct_ft->nf_ft);
 304         if (err)
 305                 goto err_init;
 306         write_pnet(&ct_ft->nf_ft.net, net);
 307
 308         __module_get(THIS_MODULE);
 309 out_unlock:
 310         params->ct_ft = ct_ft;
 311         params->nf_ft = &ct_ft->nf_ft;
 312         mutex_unlock(&zones_mutex);
 313
 314         return 0;
 315
 316 err_init:
 317         rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 318 err_insert:
 319         kfree(ct_ft);
 320 err_alloc:
 321         mutex_unlock(&zones_mutex);
 322         return err;
 323 }
 324
 325 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
 326 {
 327         struct flow_block_cb *block_cb, *tmp_cb;
 328         struct tcf_ct_flow_table *ct_ft;
 329         struct flow_block *block;
 330
 331         ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
 332                              rwork);
 333         nf_flow_table_free(&ct_ft->nf_ft);
 334
 335         /* Remove any remaining callbacks before cleanup */
 336         block = &ct_ft->nf_ft.flow_block;
 337         down_write(&ct_ft->nf_ft.flow_block_lock);
 338         list_for_each_entry_safe(block_cb, tmp_cb, &block->cb_list, list) {
 339                 list_del(&block_cb->list);
 340                 flow_block_cb_free(block_cb);
 341         }
 342         up_write(&ct_ft->nf_ft.flow_block_lock);
 343         kfree(ct_ft);
 344
 345         module_put(THIS_MODULE);
 346 }
 347
 348 static void tcf_ct_flow_table_put(struct tcf_ct_params *params)
 349 {
 350         struct tcf_ct_flow_table *ct_ft = params->ct_ft;
 351
 352         if (refcount_dec_and_test(&params->ct_ft->ref)) {
 353                 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 354                 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
 355                 queue_rcu_work(act_ct_wq, &ct_ft->rwork);
 356         }
 357 }
 358
 359 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry,
 360                                  struct nf_conn_act_ct_ext *act_ct_ext, u8 dir)
 361 {
 362         entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC;
 363         entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir];
 364 }
 365
 366 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
 367                                   struct nf_conn *ct,
 368                                   bool tcp)
 369 {
 370         struct nf_conn_act_ct_ext *act_ct_ext;
 371         struct flow_offload *entry;
 372         int err;
 373
 374         if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 375                 return;
 376
 377         entry = flow_offload_alloc(ct);
 378         if (!entry) {
 379                 WARN_ON_ONCE(1);
 380                 goto err_alloc;
 381         }
 382
 383         if (tcp) {
 384                 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 385                 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 386         }
 387
 388         act_ct_ext = nf_conn_act_ct_ext_find(ct);
 389         if (act_ct_ext) {
 390                 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
 391                 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
 392         }
 393
 394         err = flow_offload_add(&ct_ft->nf_ft, entry);
 395         if (err)
 396                 goto err_add;
 397
 398         return;
 399
 400 err_add:
 401         flow_offload_free(entry);
 402 err_alloc:
 403         clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 404 }
 405
 406 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
 407                                            struct nf_conn *ct,
 408                                            enum ip_conntrack_info ctinfo)
 409 {
 410         bool tcp = false;
 411
 412         if ((ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY) ||
 413             !test_bit(IPS_ASSURED_BIT, &ct->status))
 414                 return;
 415
 416         switch (nf_ct_protonum(ct)) {
 417         case IPPROTO_TCP:
 418                 tcp = true;
 419                 if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 420                         return;
 421                 break;
 422         case IPPROTO_UDP:
 423                 break;
 424 #ifdef CONFIG_NF_CT_PROTO_GRE
 425         case IPPROTO_GRE: {
 426                 struct nf_conntrack_tuple *tuple;
 427
 428                 if (ct->status & IPS_NAT_MASK)
 429                         return;
 430                 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 431                 /* No support for GRE v1 */
 432                 if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
 433                         return;
 434                 break;
 435         }
 436 #endif
 437         default:
 438                 return;
 439         }
 440
 441         if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 442             ct->status & IPS_SEQ_ADJUST)
 443                 return;
 444
 445         tcf_ct_flow_table_add(ct_ft, ct, tcp);
 446 }
 447
 448 static bool
 449 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
 450                                   struct flow_offload_tuple *tuple,
 451                                   struct tcphdr **tcph)
 452 {
 453         struct flow_ports *ports;
 454         unsigned int thoff;
 455         struct iphdr *iph;
 456         size_t hdrsize;
 457         u8 ipproto;
 458
 459         if (!pskb_network_may_pull(skb, sizeof(*iph)))
 460                 return false;
 461
 462         iph = ip_hdr(skb);
 463         thoff = iph->ihl * 4;
 464
 465         if (ip_is_fragment(iph) ||
 466             unlikely(thoff != sizeof(struct iphdr)))
 467                 return false;
 468
 469         ipproto = iph->protocol;
 470         switch (ipproto) {
 471         case IPPROTO_TCP:
 472                 hdrsize = sizeof(struct tcphdr);
 473                 break;
 474         case IPPROTO_UDP:
 475                 hdrsize = sizeof(*ports);
 476                 break;
 477 #ifdef CONFIG_NF_CT_PROTO_GRE
 478         case IPPROTO_GRE:
 479                 hdrsize = sizeof(struct gre_base_hdr);
 480                 break;
 481 #endif
 482         default:
 483                 return false;
 484         }
 485
 486         if (iph->ttl <= 1)
 487                 return false;
 488
 489         if (!pskb_network_may_pull(skb, thoff + hdrsize))
 490                 return false;
 491
 492         switch (ipproto) {
 493         case IPPROTO_TCP:
 494                 *tcph = (void *)(skb_network_header(skb) + thoff);
 495                 fallthrough;
 496         case IPPROTO_UDP:
 497                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 498                 tuple->src_port = ports->source;
 499                 tuple->dst_port = ports->dest;
 500                 break;
 501         case IPPROTO_GRE: {
 502                 struct gre_base_hdr *greh;
 503
 504                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 505                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 506                         return false;
 507                 break;
 508         }
 509         }
 510
 511         iph = ip_hdr(skb);
 512
 513         tuple->src_v4.s_addr = iph->saddr;
 514         tuple->dst_v4.s_addr = iph->daddr;
 515         tuple->l3proto = AF_INET;
 516         tuple->l4proto = ipproto;
 517
 518         return true;
 519 }
 520
 521 static bool
 522 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
 523                                   struct flow_offload_tuple *tuple,
 524                                   struct tcphdr **tcph)
 525 {
 526         struct flow_ports *ports;
 527         struct ipv6hdr *ip6h;
 528         unsigned int thoff;
 529         size_t hdrsize;
 530         u8 nexthdr;
 531
 532         if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
 533                 return false;
 534
 535         ip6h = ipv6_hdr(skb);
 536         thoff = sizeof(*ip6h);
 537
 538         nexthdr = ip6h->nexthdr;
 539         switch (nexthdr) {
 540         case IPPROTO_TCP:
 541                 hdrsize = sizeof(struct tcphdr);
 542                 break;
 543         case IPPROTO_UDP:
 544                 hdrsize = sizeof(*ports);
 545                 break;
 546 #ifdef CONFIG_NF_CT_PROTO_GRE
 547         case IPPROTO_GRE:
 548                 hdrsize = sizeof(struct gre_base_hdr);
 549                 break;
 550 #endif
 551         default:
 552                 return false;
 553         }
 554
 555         if (ip6h->hop_limit <= 1)
 556                 return false;
 557
 558         if (!pskb_network_may_pull(skb, thoff + hdrsize))
 559                 return false;
 560
 561         switch (nexthdr) {
 562         case IPPROTO_TCP:
 563                 *tcph = (void *)(skb_network_header(skb) + thoff);
 564                 fallthrough;
 565         case IPPROTO_UDP:
 566                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 567                 tuple->src_port = ports->source;
 568                 tuple->dst_port = ports->dest;
 569                 break;
 570         case IPPROTO_GRE: {
 571                 struct gre_base_hdr *greh;
 572
 573                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 574                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 575                         return false;
 576                 break;
 577         }
 578         }
 579
 580         ip6h = ipv6_hdr(skb);
 581
 582         tuple->src_v6 = ip6h->saddr;
 583         tuple->dst_v6 = ip6h->daddr;
 584         tuple->l3proto = AF_INET6;
 585         tuple->l4proto = nexthdr;
 586
 587         return true;
 588 }
 589
 590 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
 591                                      struct sk_buff *skb,
 592                                      u8 family)
 593 {
 594         struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
 595         struct flow_offload_tuple_rhash *tuplehash;
 596         struct flow_offload_tuple tuple = {};
 597         enum ip_conntrack_info ctinfo;
 598         struct tcphdr *tcph = NULL;
 599         struct flow_offload *flow;
 600         struct nf_conn *ct;
 601         u8 dir;
 602
 603         switch (family) {
 604         case NFPROTO_IPV4:
 605                 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
 606                         return false;
 607                 break;
 608         case NFPROTO_IPV6:
 609                 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
 610                         return false;
 611                 break;
 612         default:
 613                 return false;
 614         }
 615
 616         tuplehash = flow_offload_lookup(nf_ft, &tuple);
 617         if (!tuplehash)
 618                 return false;
 619
 620         dir = tuplehash->tuple.dir;
 621         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 622         ct = flow->ct;
 623
 624         if (tcph && (unlikely(tcph->fin || tcph->rst))) {
 625                 flow_offload_teardown(flow);
 626                 return false;
 627         }
 628
 629         ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
 630                                                     IP_CT_ESTABLISHED_REPLY;
 631
 632         flow_offload_refresh(nf_ft, flow);
 633         nf_conntrack_get(&ct->ct_general);
 634         nf_ct_set(skb, ct, ctinfo);
 635         if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
 636                 nf_ct_acct_update(ct, dir, skb->len);
 637
 638         return true;
 639 }
 640
 641 static int tcf_ct_flow_tables_init(void)
 642 {
 643         return rhashtable_init(&zones_ht, &zones_params);
 644 }
 645
 646 static void tcf_ct_flow_tables_uninit(void)
 647 {
 648         rhashtable_destroy(&zones_ht);
 649 }
 650
 651 static struct tc_action_ops act_ct_ops;
 652
 653 struct tc_ct_action_net {
 654         struct tc_action_net tn; /* Must be first */
 655         bool labels;
 656 };
 657
 658 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 659 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
 660                                    u16 zone_id, bool force)
 661 {
 662         enum ip_conntrack_info ctinfo;
 663         struct nf_conn *ct;
 664
 665         ct = nf_ct_get(skb, &ctinfo);
 666         if (!ct)
 667                 return false;
 668         if (!net_eq(net, read_pnet(&ct->ct_net)))
 669                 goto drop_ct;
 670         if (nf_ct_zone(ct)->id != zone_id)
 671                 goto drop_ct;
 672
 673         /* Force conntrack entry direction. */
 674         if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 675                 if (nf_ct_is_confirmed(ct))
 676                         nf_ct_kill(ct);
 677
 678                 goto drop_ct;
 679         }
 680
 681         return true;
 682
 683 drop_ct:
 684         nf_ct_put(ct);
 685         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 686
 687         return false;
 688 }
 689
 690 /* Trim the skb to the length specified by the IP/IPv6 header,
 691  * removing any trailing lower-layer padding. This prepares the skb
 692  * for higher-layer processing that assumes skb->len excludes padding
 693  * (such as nf_ip_checksum). The caller needs to pull the skb to the
 694  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
 695  */
 696 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
 697 {
 698         unsigned int len;
 699
 700         switch (family) {
 701         case NFPROTO_IPV4:
 702                 len = ntohs(ip_hdr(skb)->tot_len);
 703                 break;
 704         case NFPROTO_IPV6:
 705                 len = sizeof(struct ipv6hdr)
 706                         + ntohs(ipv6_hdr(skb)->payload_len);
 707                 break;
 708         default:
 709                 len = skb->len;
 710         }
 711
 712         return pskb_trim_rcsum(skb, len);
 713 }
 714
 715 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
 716 {
 717         u8 family = NFPROTO_UNSPEC;
 718
 719         switch (skb_protocol(skb, true)) {
 720         case htons(ETH_P_IP):
 721                 family = NFPROTO_IPV4;
 722                 break;
 723         case htons(ETH_P_IPV6):
 724                 family = NFPROTO_IPV6;
 725                 break;
 726         default:
 727                 break;
 728         }
 729
 730         return family;
 731 }
 732
 733 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
 734 {
 735         unsigned int len;
 736
 737         len =  skb_network_offset(skb) + sizeof(struct iphdr);
 738         if (unlikely(skb->len < len))
 739                 return -EINVAL;
 740         if (unlikely(!pskb_may_pull(skb, len)))
 741                 return -ENOMEM;
 742
 743         *frag = ip_is_fragment(ip_hdr(skb));
 744         return 0;
 745 }
 746
 747 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
 748 {
 749         unsigned int flags = 0, len, payload_ofs = 0;
 750         unsigned short frag_off;
 751         int nexthdr;
 752
 753         len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
 754         if (unlikely(skb->len < len))
 755                 return -EINVAL;
 756         if (unlikely(!pskb_may_pull(skb, len)))
 757                 return -ENOMEM;
 758
 759         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 760         if (unlikely(nexthdr < 0))
 761                 return -EPROTO;
 762
 763         *frag = flags & IP6_FH_F_FRAG;
 764         return 0;
 765 }
 766
 767 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
 768                                    u8 family, u16 zone, bool *defrag)
 769 {
 770         enum ip_conntrack_info ctinfo;
 771         struct nf_conn *ct;
 772         int err = 0;
 773         bool frag;
 774         u16 mru;
 775
 776         /* Previously seen (loopback)? Ignore. */
 777         ct = nf_ct_get(skb, &ctinfo);
 778         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 779                 return 0;
 780
 781         if (family == NFPROTO_IPV4)
 782                 err = tcf_ct_ipv4_is_fragment(skb, &frag);
 783         else
 784                 err = tcf_ct_ipv6_is_fragment(skb, &frag);
 785         if (err || !frag)
 786                 return err;
 787
 788         skb_get(skb);
 789         mru = tc_skb_cb(skb)->mru;
 790
 791         if (family == NFPROTO_IPV4) {
 792                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 793
 794                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 795                 local_bh_disable();
 796                 err = ip_defrag(net, skb, user);
 797                 local_bh_enable();
 798                 if (err && err != -EINPROGRESS)
 799                         return err;
 800
 801                 if (!err) {
 802                         *defrag = true;
 803                         mru = IPCB(skb)->frag_max_size;
 804                 }
 805         } else { /* NFPROTO_IPV6 */
 806 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 807                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 808
 809                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 810                 err = nf_ct_frag6_gather(net, skb, user);
 811                 if (err && err != -EINPROGRESS)
 812                         goto out_free;
 813
 814                 if (!err) {
 815                         *defrag = true;
 816                         mru = IP6CB(skb)->frag_max_size;
 817                 }
 818 #else
 819                 err = -EOPNOTSUPP;
 820                 goto out_free;
 821 #endif
 822         }
 823
 824         if (err != -EINPROGRESS)
 825                 tc_skb_cb(skb)->mru = mru;
 826         skb_clear_hash(skb);
 827         skb->ignore_df = 1;
 828         return err;
 829
 830 out_free:
 831         kfree_skb(skb);
 832         return err;
 833 }
 834
 835 static void tcf_ct_params_free(struct rcu_head *head)
 836 {
 837         struct tcf_ct_params *params = container_of(head,
 838                                                     struct tcf_ct_params, rcu);
 839
 840         tcf_ct_flow_table_put(params);
 841
 842         if (params->tmpl)
 843                 nf_ct_put(params->tmpl);
 844         kfree(params);
 845 }
 846
 847 #if IS_ENABLED(CONFIG_NF_NAT)
 848 /* Modelled after nf_nat_ipv[46]_fn().
 849  * range is only used for new, uninitialized NAT state.
 850  * Returns either NF_ACCEPT or NF_DROP.
 851  */
 852 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 853                           enum ip_conntrack_info ctinfo,
 854                           const struct nf_nat_range2 *range,
 855                           enum nf_nat_manip_type maniptype)
 856 {
 857         __be16 proto = skb_protocol(skb, true);
 858         int hooknum, err = NF_ACCEPT;
 859
 860         /* See HOOK2MANIP(). */
 861         if (maniptype == NF_NAT_MANIP_SRC)
 862                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 863         else
 864                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 865
 866         switch (ctinfo) {
 867         case IP_CT_RELATED:
 868         case IP_CT_RELATED_REPLY:
 869                 if (proto == htons(ETH_P_IP) &&
 870                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 871                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 872                                                            hooknum))
 873                                 err = NF_DROP;
 874                         goto out;
 875                 } else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) {
 876                         __be16 frag_off;
 877                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 878                         int hdrlen = ipv6_skip_exthdr(skb,
 879                                                       sizeof(struct ipv6hdr),
 880                                                       &nexthdr, &frag_off);
 881
 882                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 883                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 884                                                                      ctinfo,
 885                                                                      hooknum,
 886                                                                      hdrlen))
 887                                         err = NF_DROP;
 888                                 goto out;
 889                         }
 890                 }
 891                 /* Non-ICMP, fall thru to initialize if needed. */
 892                 fallthrough;
 893         case IP_CT_NEW:
 894                 /* Seen it before?  This can happen for loopback, retrans,
 895                  * or local packets.
 896                  */
 897                 if (!nf_nat_initialized(ct, maniptype)) {
 898                         /* Initialize according to the NAT action. */
 899                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 900                                 /* Action is set up to establish a new
 901                                  * mapping.
 902                                  */
 903                                 ? nf_nat_setup_info(ct, range, maniptype)
 904                                 : nf_nat_alloc_null_binding(ct, hooknum);
 905                         if (err != NF_ACCEPT)
 906                                 goto out;
 907                 }
 908                 break;
 909
 910         case IP_CT_ESTABLISHED:
 911         case IP_CT_ESTABLISHED_REPLY:
 912                 break;
 913
 914         default:
 915                 err = NF_DROP;
 916                 goto out;
 917         }
 918
 919         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 920         if (err == NF_ACCEPT) {
 921                 if (maniptype == NF_NAT_MANIP_SRC)
 922                         tc_skb_cb(skb)->post_ct_snat = 1;
 923                 if (maniptype == NF_NAT_MANIP_DST)
 924                         tc_skb_cb(skb)->post_ct_dnat = 1;
 925         }
 926 out:
 927         return err;
 928 }
 929 #endif /* CONFIG_NF_NAT */
 930
 931 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
 932 {
 933 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 934         u32 new_mark;
 935
 936         if (!mask)
 937                 return;
 938
 939         new_mark = mark | (READ_ONCE(ct->mark) & ~(mask));
 940         if (READ_ONCE(ct->mark) != new_mark) {
 941                 WRITE_ONCE(ct->mark, new_mark);
 942                 if (nf_ct_is_confirmed(ct))
 943                         nf_conntrack_event_cache(IPCT_MARK, ct);
 944         }
 945 #endif
 946 }
 947
 948 static void tcf_ct_act_set_labels(struct nf_conn *ct,
 949                                   u32 *labels,
 950                                   u32 *labels_m)
 951 {
 952 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
 953         size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
 954
 955         if (!memchr_inv(labels_m, 0, labels_sz))
 956                 return;
 957
 958         nf_connlabels_replace(ct, labels, labels_m, 4);
 959 #endif
 960 }
 961
 962 static int tcf_ct_act_nat(struct sk_buff *skb,
 963                           struct nf_conn *ct,
 964                           enum ip_conntrack_info ctinfo,
 965                           int ct_action,
 966                           struct nf_nat_range2 *range,
 967                           bool commit)
 968 {
 969 #if IS_ENABLED(CONFIG_NF_NAT)
 970         int err;
 971         enum nf_nat_manip_type maniptype;
 972
 973         if (!(ct_action & TCA_CT_ACT_NAT))
 974                 return NF_ACCEPT;
 975
 976         /* Add NAT extension if not confirmed yet. */
 977         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 978                 return NF_DROP;   /* Can't NAT. */
 979
 980         if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
 981             (ctinfo != IP_CT_RELATED || commit)) {
 982                 /* NAT an established or related connection like before. */
 983                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 984                         /* This is the REPLY direction for a connection
 985                          * for which NAT was applied in the forward
 986                          * direction.  Do the reverse NAT.
 987                          */
 988                         maniptype = ct->status & IPS_SRC_NAT
 989                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 990                 else
 991                         maniptype = ct->status & IPS_SRC_NAT
 992                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 993         } else if (ct_action & TCA_CT_ACT_NAT_SRC) {
 994                 maniptype = NF_NAT_MANIP_SRC;
 995         } else if (ct_action & TCA_CT_ACT_NAT_DST) {
 996                 maniptype = NF_NAT_MANIP_DST;
 997         } else {
 998                 return NF_ACCEPT;
 999         }
1000
1001         err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
1002         if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
1003                 if (ct->status & IPS_SRC_NAT) {
1004                         if (maniptype == NF_NAT_MANIP_SRC)
1005                                 maniptype = NF_NAT_MANIP_DST;
1006                         else
1007                                 maniptype = NF_NAT_MANIP_SRC;
1008
1009                         err = ct_nat_execute(skb, ct, ctinfo, range,
1010                                              maniptype);
1011                 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
1012                         err = ct_nat_execute(skb, ct, ctinfo, NULL,
1013                                              NF_NAT_MANIP_SRC);
1014                 }
1015         }
1016         return err;
1017 #else
1018         return NF_ACCEPT;
1019 #endif
1020 }
1021
1022 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
1023                       struct tcf_result *res)
1024 {
1025         struct net *net = dev_net(skb->dev);
1026         bool cached, commit, clear, force;
1027         enum ip_conntrack_info ctinfo;
1028         struct tcf_ct *c = to_ct(a);
1029         struct nf_conn *tmpl = NULL;
1030         struct nf_hook_state state;
1031         int nh_ofs, err, retval;
1032         struct tcf_ct_params *p;
1033         bool skip_add = false;
1034         bool defrag = false;
1035         struct nf_conn *ct;
1036         u8 family;
1037
1038         p = rcu_dereference_bh(c->params);
1039
1040         retval = READ_ONCE(c->tcf_action);
1041         commit = p->ct_action & TCA_CT_ACT_COMMIT;
1042         clear = p->ct_action & TCA_CT_ACT_CLEAR;
1043         force = p->ct_action & TCA_CT_ACT_FORCE;
1044         tmpl = p->tmpl;
1045
1046         tcf_lastuse_update(&c->tcf_tm);
1047         tcf_action_update_bstats(&c->common, skb);
1048
1049         if (clear) {
1050                 tc_skb_cb(skb)->post_ct = false;
1051                 ct = nf_ct_get(skb, &ctinfo);
1052                 if (ct) {
1053                         nf_ct_put(ct);
1054                         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1055                 }
1056
1057                 goto out_clear;
1058         }
1059
1060         family = tcf_ct_skb_nf_family(skb);
1061         if (family == NFPROTO_UNSPEC)
1062                 goto drop;
1063
1064         /* The conntrack module expects to be working at L3.
1065          * We also try to pull the IPv4/6 header to linear area
1066          */
1067         nh_ofs = skb_network_offset(skb);
1068         skb_pull_rcsum(skb, nh_ofs);
1069         err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
1070         if (err == -EINPROGRESS) {
1071                 retval = TC_ACT_STOLEN;
1072                 goto out_clear;
1073         }
1074         if (err)
1075                 goto drop;
1076
1077         err = tcf_ct_skb_network_trim(skb, family);
1078         if (err)
1079                 goto drop;
1080
1081         /* If we are recirculating packets to match on ct fields and
1082          * committing with a separate ct action, then we don't need to
1083          * actually run the packet through conntrack twice unless it's for a
1084          * different zone.
1085          */
1086         cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
1087         if (!cached) {
1088                 if (tcf_ct_flow_table_lookup(p, skb, family)) {
1089                         skip_add = true;
1090                         goto do_nat;
1091                 }
1092
1093                 /* Associate skb with specified zone. */
1094                 if (tmpl) {
1095                         nf_conntrack_put(skb_nfct(skb));
1096                         nf_conntrack_get(&tmpl->ct_general);
1097                         nf_ct_set(skb, tmpl, IP_CT_NEW);
1098                 }
1099
1100                 state.hook = NF_INET_PRE_ROUTING;
1101                 state.net = net;
1102                 state.pf = family;
1103                 err = nf_conntrack_in(skb, &state);
1104                 if (err != NF_ACCEPT)
1105                         goto out_push;
1106         }
1107
1108 do_nat:
1109         ct = nf_ct_get(skb, &ctinfo);
1110         if (!ct)
1111                 goto out_push;
1112         nf_ct_deliver_cached_events(ct);
1113         nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1114
1115         err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1116         if (err != NF_ACCEPT)
1117                 goto drop;
1118
1119         if (commit) {
1120                 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1121                 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1122
1123                 if (!nf_ct_is_confirmed(ct))
1124                         nf_conn_act_ct_ext_add(ct);
1125
1126                 /* This will take care of sending queued events
1127                  * even if the connection is already confirmed.
1128                  */
1129                 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1130                         goto drop;
1131         }
1132
1133         if (!skip_add)
1134                 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1135
1136 out_push:
1137         skb_push_rcsum(skb, nh_ofs);
1138
1139         tc_skb_cb(skb)->post_ct = true;
1140         tc_skb_cb(skb)->zone = p->zone;
1141 out_clear:
1142         if (defrag)
1143                 qdisc_skb_cb(skb)->pkt_len = skb->len;
1144         return retval;
1145
1146 drop:
1147         tcf_action_inc_drop_qstats(&c->common);
1148         return TC_ACT_SHOT;
1149 }
1150
1151 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1152         [TCA_CT_ACTION] = { .type = NLA_U16 },
1153         [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1154         [TCA_CT_ZONE] = { .type = NLA_U16 },
1155         [TCA_CT_MARK] = { .type = NLA_U32 },
1156         [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1157         [TCA_CT_LABELS] = { .type = NLA_BINARY,
1158                             .len = 128 / BITS_PER_BYTE },
1159         [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1160                                  .len = 128 / BITS_PER_BYTE },
1161         [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1162         [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1163         [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1164         [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1165         [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1166         [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1167 };
1168
1169 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1170                                   struct tc_ct *parm,
1171                                   struct nlattr **tb,
1172                                   struct netlink_ext_ack *extack)
1173 {
1174         struct nf_nat_range2 *range;
1175
1176         if (!(p->ct_action & TCA_CT_ACT_NAT))
1177                 return 0;
1178
1179         if (!IS_ENABLED(CONFIG_NF_NAT)) {
1180                 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1181                 return -EOPNOTSUPP;
1182         }
1183
1184         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1185                 return 0;
1186
1187         if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1188             (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1189                 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1190                 return -EOPNOTSUPP;
1191         }
1192
1193         range = &p->range;
1194         if (tb[TCA_CT_NAT_IPV4_MIN]) {
1195                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1196
1197                 p->ipv4_range = true;
1198                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1199                 range->min_addr.ip =
1200                         nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1201
1202                 range->max_addr.ip = max_attr ?
1203                                      nla_get_in_addr(max_attr) :
1204                                      range->min_addr.ip;
1205         } else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1206                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1207
1208                 p->ipv4_range = false;
1209                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1210                 range->min_addr.in6 =
1211                         nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1212
1213                 range->max_addr.in6 = max_attr ?
1214                                       nla_get_in6_addr(max_attr) :
1215                                       range->min_addr.in6;
1216         }
1217
1218         if (tb[TCA_CT_NAT_PORT_MIN]) {
1219                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1220                 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1221
1222                 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1223                                        nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1224                                        range->min_proto.all;
1225         }
1226
1227         return 0;
1228 }
1229
1230 static void tcf_ct_set_key_val(struct nlattr **tb,
1231                                void *val, int val_type,
1232                                void *mask, int mask_type,
1233                                int len)
1234 {
1235         if (!tb[val_type])
1236                 return;
1237         nla_memcpy(val, tb[val_type], len);
1238
1239         if (!mask)
1240                 return;
1241
1242         if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1243                 memset(mask, 0xff, len);
1244         else
1245                 nla_memcpy(mask, tb[mask_type], len);
1246 }
1247
1248 static int tcf_ct_fill_params(struct net *net,
1249                               struct tcf_ct_params *p,
1250                               struct tc_ct *parm,
1251                               struct nlattr **tb,
1252                               struct netlink_ext_ack *extack)
1253 {
1254         struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1255         struct nf_conntrack_zone zone;
1256         struct nf_conn *tmpl;
1257         int err;
1258
1259         p->zone = NF_CT_DEFAULT_ZONE_ID;
1260
1261         tcf_ct_set_key_val(tb,
1262                            &p->ct_action, TCA_CT_ACTION,
1263                            NULL, TCA_CT_UNSPEC,
1264                            sizeof(p->ct_action));
1265
1266         if (p->ct_action & TCA_CT_ACT_CLEAR)
1267                 return 0;
1268
1269         err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1270         if (err)
1271                 return err;
1272
1273         if (tb[TCA_CT_MARK]) {
1274                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1275                         NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1276                         return -EOPNOTSUPP;
1277                 }
1278                 tcf_ct_set_key_val(tb,
1279                                    &p->mark, TCA_CT_MARK,
1280                                    &p->mark_mask, TCA_CT_MARK_MASK,
1281                                    sizeof(p->mark));
1282         }
1283
1284         if (tb[TCA_CT_LABELS]) {
1285                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1286                         NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1287                         return -EOPNOTSUPP;
1288                 }
1289
1290                 if (!tn->labels) {
1291                         NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1292                         return -EOPNOTSUPP;
1293                 }
1294                 tcf_ct_set_key_val(tb,
1295                                    p->labels, TCA_CT_LABELS,
1296                                    p->labels_mask, TCA_CT_LABELS_MASK,
1297                                    sizeof(p->labels));
1298         }
1299
1300         if (tb[TCA_CT_ZONE]) {
1301                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1302                         NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1303                         return -EOPNOTSUPP;
1304                 }
1305
1306                 tcf_ct_set_key_val(tb,
1307                                    &p->zone, TCA_CT_ZONE,
1308                                    NULL, TCA_CT_UNSPEC,
1309                                    sizeof(p->zone));
1310         }
1311
1312         nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1313         tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1314         if (!tmpl) {
1315                 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1316                 return -ENOMEM;
1317         }
1318         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1319         p->tmpl = tmpl;
1320
1321         return 0;
1322 }
1323
1324 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1325                        struct nlattr *est, struct tc_action **a,
1326                        struct tcf_proto *tp, u32 flags,
1327                        struct netlink_ext_ack *extack)
1328 {
1329         struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id);
1330         bool bind = flags & TCA_ACT_FLAGS_BIND;
1331         struct tcf_ct_params *params = NULL;
1332         struct nlattr *tb[TCA_CT_MAX + 1];
1333         struct tcf_chain *goto_ch = NULL;
1334         struct tc_ct *parm;
1335         struct tcf_ct *c;
1336         int err, res = 0;
1337         u32 index;
1338
1339         if (!nla) {
1340                 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1341                 return -EINVAL;
1342         }
1343
1344         err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1345         if (err < 0)
1346                 return err;
1347
1348         if (!tb[TCA_CT_PARMS]) {
1349                 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1350                 return -EINVAL;
1351         }
1352         parm = nla_data(tb[TCA_CT_PARMS]);
1353         index = parm->index;
1354         err = tcf_idr_check_alloc(tn, &index, a, bind);
1355         if (err < 0)
1356                 return err;
1357
1358         if (!err) {
1359                 err = tcf_idr_create_from_flags(tn, index, est, a,
1360                                                 &act_ct_ops, bind, flags);
1361                 if (err) {
1362                         tcf_idr_cleanup(tn, index);
1363                         return err;
1364                 }
1365                 res = ACT_P_CREATED;
1366         } else {
1367                 if (bind)
1368                         return 0;
1369
1370                 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1371                         tcf_idr_release(*a, bind);
1372                         return -EEXIST;
1373                 }
1374         }
1375         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1376         if (err < 0)
1377                 goto cleanup;
1378
1379         c = to_ct(*a);
1380
1381         params = kzalloc(sizeof(*params), GFP_KERNEL);
1382         if (unlikely(!params)) {
1383                 err = -ENOMEM;
1384                 goto cleanup;
1385         }
1386
1387         err = tcf_ct_fill_params(net, params, parm, tb, extack);
1388         if (err)
1389                 goto cleanup;
1390
1391         err = tcf_ct_flow_table_get(net, params);
1392         if (err)
1393                 goto cleanup_params;
1394
1395         spin_lock_bh(&c->tcf_lock);
1396         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1397         params = rcu_replace_pointer(c->params, params,
1398                                      lockdep_is_held(&c->tcf_lock));
1399         spin_unlock_bh(&c->tcf_lock);
1400
1401         if (goto_ch)
1402                 tcf_chain_put_by_act(goto_ch);
1403         if (params)
1404                 call_rcu(&params->rcu, tcf_ct_params_free);
1405
1406         return res;
1407
1408 cleanup_params:
1409         if (params->tmpl)
1410                 nf_ct_put(params->tmpl);
1411 cleanup:
1412         if (goto_ch)
1413                 tcf_chain_put_by_act(goto_ch);
1414         kfree(params);
1415         tcf_idr_release(*a, bind);
1416         return err;
1417 }
1418
1419 static void tcf_ct_cleanup(struct tc_action *a)
1420 {
1421         struct tcf_ct_params *params;
1422         struct tcf_ct *c = to_ct(a);
1423
1424         params = rcu_dereference_protected(c->params, 1);
1425         if (params)
1426                 call_rcu(&params->rcu, tcf_ct_params_free);
1427 }
1428
1429 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1430                                void *val, int val_type,
1431                                void *mask, int mask_type,
1432                                int len)
1433 {
1434         int err;
1435
1436         if (mask && !memchr_inv(mask, 0, len))
1437                 return 0;
1438
1439         err = nla_put(skb, val_type, len, val);
1440         if (err)
1441                 return err;
1442
1443         if (mask_type != TCA_CT_UNSPEC) {
1444                 err = nla_put(skb, mask_type, len, mask);
1445                 if (err)
1446                         return err;
1447         }
1448
1449         return 0;
1450 }
1451
1452 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1453 {
1454         struct nf_nat_range2 *range = &p->range;
1455
1456         if (!(p->ct_action & TCA_CT_ACT_NAT))
1457                 return 0;
1458
1459         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1460                 return 0;
1461
1462         if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1463                 if (p->ipv4_range) {
1464                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1465                                             range->min_addr.ip))
1466                                 return -1;
1467                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1468                                             range->max_addr.ip))
1469                                 return -1;
1470                 } else {
1471                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1472                                              &range->min_addr.in6))
1473                                 return -1;
1474                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1475                                              &range->max_addr.in6))
1476                                 return -1;
1477                 }
1478         }
1479
1480         if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1481                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1482                                  range->min_proto.all))
1483                         return -1;
1484                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1485                                  range->max_proto.all))
1486                         return -1;
1487         }
1488
1489         return 0;
1490 }
1491
1492 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1493                               int bind, int ref)
1494 {
1495         unsigned char *b = skb_tail_pointer(skb);
1496         struct tcf_ct *c = to_ct(a);
1497         struct tcf_ct_params *p;
1498
1499         struct tc_ct opt = {
1500                 .index   = c->tcf_index,
1501                 .refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1502                 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1503         };
1504         struct tcf_t t;
1505
1506         spin_lock_bh(&c->tcf_lock);
1507         p = rcu_dereference_protected(c->params,
1508                                       lockdep_is_held(&c->tcf_lock));
1509         opt.action = c->tcf_action;
1510
1511         if (tcf_ct_dump_key_val(skb,
1512                                 &p->ct_action, TCA_CT_ACTION,
1513                                 NULL, TCA_CT_UNSPEC,
1514                                 sizeof(p->ct_action)))
1515                 goto nla_put_failure;
1516
1517         if (p->ct_action & TCA_CT_ACT_CLEAR)
1518                 goto skip_dump;
1519
1520         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1521             tcf_ct_dump_key_val(skb,
1522                                 &p->mark, TCA_CT_MARK,
1523                                 &p->mark_mask, TCA_CT_MARK_MASK,
1524                                 sizeof(p->mark)))
1525                 goto nla_put_failure;
1526
1527         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1528             tcf_ct_dump_key_val(skb,
1529                                 p->labels, TCA_CT_LABELS,
1530                                 p->labels_mask, TCA_CT_LABELS_MASK,
1531                                 sizeof(p->labels)))
1532                 goto nla_put_failure;
1533
1534         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1535             tcf_ct_dump_key_val(skb,
1536                                 &p->zone, TCA_CT_ZONE,
1537                                 NULL, TCA_CT_UNSPEC,
1538                                 sizeof(p->zone)))
1539                 goto nla_put_failure;
1540
1541         if (tcf_ct_dump_nat(skb, p))
1542                 goto nla_put_failure;
1543
1544 skip_dump:
1545         if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1546                 goto nla_put_failure;
1547
1548         tcf_tm_dump(&t, &c->tcf_tm);
1549         if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1550                 goto nla_put_failure;
1551         spin_unlock_bh(&c->tcf_lock);
1552
1553         return skb->len;
1554 nla_put_failure:
1555         spin_unlock_bh(&c->tcf_lock);
1556         nlmsg_trim(skb, b);
1557         return -1;
1558 }
1559
1560 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1561                              u64 drops, u64 lastuse, bool hw)
1562 {
1563         struct tcf_ct *c = to_ct(a);
1564
1565         tcf_action_update_stats(a, bytes, packets, drops, hw);
1566         c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1567 }
1568
1569 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1570                                     u32 *index_inc, bool bind,
1571                                     struct netlink_ext_ack *extack)
1572 {
1573         if (bind) {
1574                 struct flow_action_entry *entry = entry_data;
1575
1576                 entry->id = FLOW_ACTION_CT;
1577                 entry->ct.action = tcf_ct_action(act);
1578                 entry->ct.zone = tcf_ct_zone(act);
1579                 entry->ct.flow_table = tcf_ct_ft(act);
1580                 *index_inc = 1;
1581         } else {
1582                 struct flow_offload_action *fl_action = entry_data;
1583
1584                 fl_action->id = FLOW_ACTION_CT;
1585         }
1586
1587         return 0;
1588 }
1589
1590 static struct tc_action_ops act_ct_ops = {
1591         .kind           =       "ct",
1592         .id             =       TCA_ID_CT,
1593         .owner          =       THIS_MODULE,
1594         .act            =       tcf_ct_act,
1595         .dump           =       tcf_ct_dump,
1596         .init           =       tcf_ct_init,
1597         .cleanup        =       tcf_ct_cleanup,
1598         .stats_update   =       tcf_stats_update,
1599         .offload_act_setup =    tcf_ct_offload_act_setup,
1600         .size           =       sizeof(struct tcf_ct),
1601 };
1602
1603 static __net_init int ct_init_net(struct net *net)
1604 {
1605         unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1606         struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1607
1608         if (nf_connlabels_get(net, n_bits - 1)) {
1609                 tn->labels = false;
1610                 pr_err("act_ct: Failed to set connlabels length");
1611         } else {
1612                 tn->labels = true;
1613         }
1614
1615         return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1616 }
1617
1618 static void __net_exit ct_exit_net(struct list_head *net_list)
1619 {
1620         struct net *net;
1621
1622         rtnl_lock();
1623         list_for_each_entry(net, net_list, exit_list) {
1624                 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1625
1626                 if (tn->labels)
1627                         nf_connlabels_put(net);
1628         }
1629         rtnl_unlock();
1630
1631         tc_action_net_exit(net_list, act_ct_ops.net_id);
1632 }
1633
1634 static struct pernet_operations ct_net_ops = {
1635         .init = ct_init_net,
1636         .exit_batch = ct_exit_net,
1637         .id   = &act_ct_ops.net_id,
1638         .size = sizeof(struct tc_ct_action_net),
1639 };
1640
1641 static int __init ct_init_module(void)
1642 {
1643         int err;
1644
1645         act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1646         if (!act_ct_wq)
1647                 return -ENOMEM;
1648
1649         err = tcf_ct_flow_tables_init();
1650         if (err)
1651                 goto err_tbl_init;
1652
1653         err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1654         if (err)
1655                 goto err_register;
1656
1657         static_branch_inc(&tcf_frag_xmit_count);
1658
1659         return 0;
1660
1661 err_register:
1662         tcf_ct_flow_tables_uninit();
1663 err_tbl_init:
1664         destroy_workqueue(act_ct_wq);
1665         return err;
1666 }
1667
1668 static void __exit ct_cleanup_module(void)
1669 {
1670         static_branch_dec(&tcf_frag_xmit_count);
1671         tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1672         tcf_ct_flow_tables_uninit();
1673         destroy_workqueue(act_ct_wq);
1674 }
1675
1676 module_init(ct_init_module);
1677 module_exit(ct_cleanup_module);
1678 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1679 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1680 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1681 MODULE_DESCRIPTION("Connection tracking action");
1682 MODULE_LICENSE("GPL v2");