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 = 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 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
 307         __module_get(THIS_MODULE);
 308 out_unlock:
 309         params->ct_ft = ct_ft;
 310         params->nf_ft = &ct_ft->nf_ft;
 311         mutex_unlock(&zones_mutex);
 312
 313         return 0;
 314
 315 err_init:
 316         rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 317 err_insert:
 318         kfree(ct_ft);
 319 err_alloc:
 320         mutex_unlock(&zones_mutex);
 321         return err;
 322 }
 323
 324 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
 325 {
 326         struct flow_block_cb *block_cb, *tmp_cb;
 327         struct tcf_ct_flow_table *ct_ft;
 328         struct flow_block *block;
 329
 330         ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
 331                              rwork);
 332         nf_flow_table_free(&ct_ft->nf_ft);
 333
 334         /* Remove any remaining callbacks before cleanup */
 335         block = &ct_ft->nf_ft.flow_block;
 336         down_write(&ct_ft->nf_ft.flow_block_lock);
 337         list_for_each_entry_safe(block_cb, tmp_cb, &block->cb_list, list) {
 338                 list_del(&block_cb->list);
 339                 flow_block_cb_free(block_cb);
 340         }
 341         up_write(&ct_ft->nf_ft.flow_block_lock);
 342         kfree(ct_ft);
 343
 344         module_put(THIS_MODULE);
 345 }
 346
 347 static void tcf_ct_flow_table_put(struct tcf_ct_params *params)
 348 {
 349         struct tcf_ct_flow_table *ct_ft = params->ct_ft;
 350
 351         if (refcount_dec_and_test(&params->ct_ft->ref)) {
 352                 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 353                 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
 354                 queue_rcu_work(act_ct_wq, &ct_ft->rwork);
 355         }
 356 }
 357
 358 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry,
 359                                  struct nf_conn_act_ct_ext *act_ct_ext, u8 dir)
 360 {
 361         entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC;
 362         entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir];
 363 }
 364
 365 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
 366                                   struct nf_conn *ct,
 367                                   bool tcp)
 368 {
 369         struct nf_conn_act_ct_ext *act_ct_ext;
 370         struct flow_offload *entry;
 371         int err;
 372
 373         if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 374                 return;
 375
 376         entry = flow_offload_alloc(ct);
 377         if (!entry) {
 378                 WARN_ON_ONCE(1);
 379                 goto err_alloc;
 380         }
 381
 382         if (tcp) {
 383                 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 384                 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 385         }
 386
 387         act_ct_ext = nf_conn_act_ct_ext_find(ct);
 388         if (act_ct_ext) {
 389                 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
 390                 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
 391         }
 392
 393         err = flow_offload_add(&ct_ft->nf_ft, entry);
 394         if (err)
 395                 goto err_add;
 396
 397         return;
 398
 399 err_add:
 400         flow_offload_free(entry);
 401 err_alloc:
 402         clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 403 }
 404
 405 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
 406                                            struct nf_conn *ct,
 407                                            enum ip_conntrack_info ctinfo)
 408 {
 409         bool tcp = false;
 410
 411         if ((ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY) ||
 412             !test_bit(IPS_ASSURED_BIT, &ct->status))
 413                 return;
 414
 415         switch (nf_ct_protonum(ct)) {
 416         case IPPROTO_TCP:
 417                 tcp = true;
 418                 if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 419                         return;
 420                 break;
 421         case IPPROTO_UDP:
 422                 break;
 423 #ifdef CONFIG_NF_CT_PROTO_GRE
 424         case IPPROTO_GRE: {
 425                 struct nf_conntrack_tuple *tuple;
 426
 427                 if (ct->status & IPS_NAT_MASK)
 428                         return;
 429                 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 430                 /* No support for GRE v1 */
 431                 if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
 432                         return;
 433                 break;
 434         }
 435 #endif
 436         default:
 437                 return;
 438         }
 439
 440         if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 441             ct->status & IPS_SEQ_ADJUST)
 442                 return;
 443
 444         tcf_ct_flow_table_add(ct_ft, ct, tcp);
 445 }
 446
 447 static bool
 448 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
 449                                   struct flow_offload_tuple *tuple,
 450                                   struct tcphdr **tcph)
 451 {
 452         struct flow_ports *ports;
 453         unsigned int thoff;
 454         struct iphdr *iph;
 455         size_t hdrsize;
 456         u8 ipproto;
 457
 458         if (!pskb_network_may_pull(skb, sizeof(*iph)))
 459                 return false;
 460
 461         iph = ip_hdr(skb);
 462         thoff = iph->ihl * 4;
 463
 464         if (ip_is_fragment(iph) ||
 465             unlikely(thoff != sizeof(struct iphdr)))
 466                 return false;
 467
 468         ipproto = iph->protocol;
 469         switch (ipproto) {
 470         case IPPROTO_TCP:
 471                 hdrsize = sizeof(struct tcphdr);
 472                 break;
 473         case IPPROTO_UDP:
 474                 hdrsize = sizeof(*ports);
 475                 break;
 476 #ifdef CONFIG_NF_CT_PROTO_GRE
 477         case IPPROTO_GRE:
 478                 hdrsize = sizeof(struct gre_base_hdr);
 479                 break;
 480 #endif
 481         default:
 482                 return false;
 483         }
 484
 485         if (iph->ttl <= 1)
 486                 return false;
 487
 488         if (!pskb_network_may_pull(skb, thoff + hdrsize))
 489                 return false;
 490
 491         switch (ipproto) {
 492         case IPPROTO_TCP:
 493                 *tcph = (void *)(skb_network_header(skb) + thoff);
 494                 fallthrough;
 495         case IPPROTO_UDP:
 496                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 497                 tuple->src_port = ports->source;
 498                 tuple->dst_port = ports->dest;
 499                 break;
 500         case IPPROTO_GRE: {
 501                 struct gre_base_hdr *greh;
 502
 503                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 504                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 505                         return false;
 506                 break;
 507         }
 508         }
 509
 510         iph = ip_hdr(skb);
 511
 512         tuple->src_v4.s_addr = iph->saddr;
 513         tuple->dst_v4.s_addr = iph->daddr;
 514         tuple->l3proto = AF_INET;
 515         tuple->l4proto = ipproto;
 516
 517         return true;
 518 }
 519
 520 static bool
 521 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
 522                                   struct flow_offload_tuple *tuple,
 523                                   struct tcphdr **tcph)
 524 {
 525         struct flow_ports *ports;
 526         struct ipv6hdr *ip6h;
 527         unsigned int thoff;
 528         size_t hdrsize;
 529         u8 nexthdr;
 530
 531         if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
 532                 return false;
 533
 534         ip6h = ipv6_hdr(skb);
 535         thoff = sizeof(*ip6h);
 536
 537         nexthdr = ip6h->nexthdr;
 538         switch (nexthdr) {
 539         case IPPROTO_TCP:
 540                 hdrsize = sizeof(struct tcphdr);
 541                 break;
 542         case IPPROTO_UDP:
 543                 hdrsize = sizeof(*ports);
 544                 break;
 545 #ifdef CONFIG_NF_CT_PROTO_GRE
 546         case IPPROTO_GRE:
 547                 hdrsize = sizeof(struct gre_base_hdr);
 548                 break;
 549 #endif
 550         default:
 551                 return false;
 552         }
 553
 554         if (ip6h->hop_limit <= 1)
 555                 return false;
 556
 557         if (!pskb_network_may_pull(skb, thoff + hdrsize))
 558                 return false;
 559
 560         switch (nexthdr) {
 561         case IPPROTO_TCP:
 562                 *tcph = (void *)(skb_network_header(skb) + thoff);
 563                 fallthrough;
 564         case IPPROTO_UDP:
 565                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 566                 tuple->src_port = ports->source;
 567                 tuple->dst_port = ports->dest;
 568                 break;
 569         case IPPROTO_GRE: {
 570                 struct gre_base_hdr *greh;
 571
 572                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 573                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 574                         return false;
 575                 break;
 576         }
 577         }
 578
 579         ip6h = ipv6_hdr(skb);
 580
 581         tuple->src_v6 = ip6h->saddr;
 582         tuple->dst_v6 = ip6h->daddr;
 583         tuple->l3proto = AF_INET6;
 584         tuple->l4proto = nexthdr;
 585
 586         return true;
 587 }
 588
 589 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
 590                                      struct sk_buff *skb,
 591                                      u8 family)
 592 {
 593         struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
 594         struct flow_offload_tuple_rhash *tuplehash;
 595         struct flow_offload_tuple tuple = {};
 596         enum ip_conntrack_info ctinfo;
 597         struct tcphdr *tcph = NULL;
 598         struct flow_offload *flow;
 599         struct nf_conn *ct;
 600         u8 dir;
 601
 602         switch (family) {
 603         case NFPROTO_IPV4:
 604                 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
 605                         return false;
 606                 break;
 607         case NFPROTO_IPV6:
 608                 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
 609                         return false;
 610                 break;
 611         default:
 612                 return false;
 613         }
 614
 615         tuplehash = flow_offload_lookup(nf_ft, &tuple);
 616         if (!tuplehash)
 617                 return false;
 618
 619         dir = tuplehash->tuple.dir;
 620         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 621         ct = flow->ct;
 622
 623         if (tcph && (unlikely(tcph->fin || tcph->rst))) {
 624                 flow_offload_teardown(flow);
 625                 return false;
 626         }
 627
 628         ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
 629                                                     IP_CT_ESTABLISHED_REPLY;
 630
 631         flow_offload_refresh(nf_ft, flow);
 632         nf_conntrack_get(&ct->ct_general);
 633         nf_ct_set(skb, ct, ctinfo);
 634         if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
 635                 nf_ct_acct_update(ct, dir, skb->len);
 636
 637         return true;
 638 }
 639
 640 static int tcf_ct_flow_tables_init(void)
 641 {
 642         return rhashtable_init(&zones_ht, &zones_params);
 643 }
 644
 645 static void tcf_ct_flow_tables_uninit(void)
 646 {
 647         rhashtable_destroy(&zones_ht);
 648 }
 649
 650 static struct tc_action_ops act_ct_ops;
 651 static unsigned int ct_net_id;
 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         int err;
 700
 701         switch (family) {
 702         case NFPROTO_IPV4:
 703                 len = ntohs(ip_hdr(skb)->tot_len);
 704                 break;
 705         case NFPROTO_IPV6:
 706                 len = sizeof(struct ipv6hdr)
 707                         + ntohs(ipv6_hdr(skb)->payload_len);
 708                 break;
 709         default:
 710                 len = skb->len;
 711         }
 712
 713         err = pskb_trim_rcsum(skb, len);
 714
 715         return err;
 716 }
 717
 718 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
 719 {
 720         u8 family = NFPROTO_UNSPEC;
 721
 722         switch (skb_protocol(skb, true)) {
 723         case htons(ETH_P_IP):
 724                 family = NFPROTO_IPV4;
 725                 break;
 726         case htons(ETH_P_IPV6):
 727                 family = NFPROTO_IPV6;
 728                 break;
 729         default:
 730                 break;
 731         }
 732
 733         return family;
 734 }
 735
 736 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
 737 {
 738         unsigned int len;
 739
 740         len =  skb_network_offset(skb) + sizeof(struct iphdr);
 741         if (unlikely(skb->len < len))
 742                 return -EINVAL;
 743         if (unlikely(!pskb_may_pull(skb, len)))
 744                 return -ENOMEM;
 745
 746         *frag = ip_is_fragment(ip_hdr(skb));
 747         return 0;
 748 }
 749
 750 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
 751 {
 752         unsigned int flags = 0, len, payload_ofs = 0;
 753         unsigned short frag_off;
 754         int nexthdr;
 755
 756         len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
 757         if (unlikely(skb->len < len))
 758                 return -EINVAL;
 759         if (unlikely(!pskb_may_pull(skb, len)))
 760                 return -ENOMEM;
 761
 762         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 763         if (unlikely(nexthdr < 0))
 764                 return -EPROTO;
 765
 766         *frag = flags & IP6_FH_F_FRAG;
 767         return 0;
 768 }
 769
 770 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
 771                                    u8 family, u16 zone, bool *defrag)
 772 {
 773         enum ip_conntrack_info ctinfo;
 774         struct nf_conn *ct;
 775         int err = 0;
 776         bool frag;
 777         u16 mru;
 778
 779         /* Previously seen (loopback)? Ignore. */
 780         ct = nf_ct_get(skb, &ctinfo);
 781         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 782                 return 0;
 783
 784         if (family == NFPROTO_IPV4)
 785                 err = tcf_ct_ipv4_is_fragment(skb, &frag);
 786         else
 787                 err = tcf_ct_ipv6_is_fragment(skb, &frag);
 788         if (err || !frag)
 789                 return err;
 790
 791         skb_get(skb);
 792         mru = tc_skb_cb(skb)->mru;
 793
 794         if (family == NFPROTO_IPV4) {
 795                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 796
 797                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 798                 local_bh_disable();
 799                 err = ip_defrag(net, skb, user);
 800                 local_bh_enable();
 801                 if (err && err != -EINPROGRESS)
 802                         return err;
 803
 804                 if (!err) {
 805                         *defrag = true;
 806                         mru = IPCB(skb)->frag_max_size;
 807                 }
 808         } else { /* NFPROTO_IPV6 */
 809 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 810                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 811
 812                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 813                 err = nf_ct_frag6_gather(net, skb, user);
 814                 if (err && err != -EINPROGRESS)
 815                         goto out_free;
 816
 817                 if (!err) {
 818                         *defrag = true;
 819                         mru = IP6CB(skb)->frag_max_size;
 820                 }
 821 #else
 822                 err = -EOPNOTSUPP;
 823                 goto out_free;
 824 #endif
 825         }
 826
 827         if (err != -EINPROGRESS)
 828                 tc_skb_cb(skb)->mru = mru;
 829         skb_clear_hash(skb);
 830         skb->ignore_df = 1;
 831         return err;
 832
 833 out_free:
 834         kfree_skb(skb);
 835         return err;
 836 }
 837
 838 static void tcf_ct_params_free(struct rcu_head *head)
 839 {
 840         struct tcf_ct_params *params = container_of(head,
 841                                                     struct tcf_ct_params, rcu);
 842
 843         tcf_ct_flow_table_put(params);
 844
 845         if (params->tmpl)
 846                 nf_ct_put(params->tmpl);
 847         kfree(params);
 848 }
 849
 850 #if IS_ENABLED(CONFIG_NF_NAT)
 851 /* Modelled after nf_nat_ipv[46]_fn().
 852  * range is only used for new, uninitialized NAT state.
 853  * Returns either NF_ACCEPT or NF_DROP.
 854  */
 855 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 856                           enum ip_conntrack_info ctinfo,
 857                           const struct nf_nat_range2 *range,
 858                           enum nf_nat_manip_type maniptype)
 859 {
 860         __be16 proto = skb_protocol(skb, true);
 861         int hooknum, err = NF_ACCEPT;
 862
 863         /* See HOOK2MANIP(). */
 864         if (maniptype == NF_NAT_MANIP_SRC)
 865                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 866         else
 867                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 868
 869         switch (ctinfo) {
 870         case IP_CT_RELATED:
 871         case IP_CT_RELATED_REPLY:
 872                 if (proto == htons(ETH_P_IP) &&
 873                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 874                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 875                                                            hooknum))
 876                                 err = NF_DROP;
 877                         goto out;
 878                 } else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) {
 879                         __be16 frag_off;
 880                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 881                         int hdrlen = ipv6_skip_exthdr(skb,
 882                                                       sizeof(struct ipv6hdr),
 883                                                       &nexthdr, &frag_off);
 884
 885                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 886                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 887                                                                      ctinfo,
 888                                                                      hooknum,
 889                                                                      hdrlen))
 890                                         err = NF_DROP;
 891                                 goto out;
 892                         }
 893                 }
 894                 /* Non-ICMP, fall thru to initialize if needed. */
 895                 fallthrough;
 896         case IP_CT_NEW:
 897                 /* Seen it before?  This can happen for loopback, retrans,
 898                  * or local packets.
 899                  */
 900                 if (!nf_nat_initialized(ct, maniptype)) {
 901                         /* Initialize according to the NAT action. */
 902                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 903                                 /* Action is set up to establish a new
 904                                  * mapping.
 905                                  */
 906                                 ? nf_nat_setup_info(ct, range, maniptype)
 907                                 : nf_nat_alloc_null_binding(ct, hooknum);
 908                         if (err != NF_ACCEPT)
 909                                 goto out;
 910                 }
 911                 break;
 912
 913         case IP_CT_ESTABLISHED:
 914         case IP_CT_ESTABLISHED_REPLY:
 915                 break;
 916
 917         default:
 918                 err = NF_DROP;
 919                 goto out;
 920         }
 921
 922         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 923         if (err == NF_ACCEPT) {
 924                 if (maniptype == NF_NAT_MANIP_SRC)
 925                         tc_skb_cb(skb)->post_ct_snat = 1;
 926                 if (maniptype == NF_NAT_MANIP_DST)
 927                         tc_skb_cb(skb)->post_ct_dnat = 1;
 928         }
 929 out:
 930         return err;
 931 }
 932 #endif /* CONFIG_NF_NAT */
 933
 934 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
 935 {
 936 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 937         u32 new_mark;
 938
 939         if (!mask)
 940                 return;
 941
 942         new_mark = mark | (ct->mark & ~(mask));
 943         if (ct->mark != new_mark) {
 944                 ct->mark = new_mark;
 945                 if (nf_ct_is_confirmed(ct))
 946                         nf_conntrack_event_cache(IPCT_MARK, ct);
 947         }
 948 #endif
 949 }
 950
 951 static void tcf_ct_act_set_labels(struct nf_conn *ct,
 952                                   u32 *labels,
 953                                   u32 *labels_m)
 954 {
 955 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
 956         size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
 957
 958         if (!memchr_inv(labels_m, 0, labels_sz))
 959                 return;
 960
 961         nf_connlabels_replace(ct, labels, labels_m, 4);
 962 #endif
 963 }
 964
 965 static int tcf_ct_act_nat(struct sk_buff *skb,
 966                           struct nf_conn *ct,
 967                           enum ip_conntrack_info ctinfo,
 968                           int ct_action,
 969                           struct nf_nat_range2 *range,
 970                           bool commit)
 971 {
 972 #if IS_ENABLED(CONFIG_NF_NAT)
 973         int err;
 974         enum nf_nat_manip_type maniptype;
 975
 976         if (!(ct_action & TCA_CT_ACT_NAT))
 977                 return NF_ACCEPT;
 978
 979         /* Add NAT extension if not confirmed yet. */
 980         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 981                 return NF_DROP;   /* Can't NAT. */
 982
 983         if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
 984             (ctinfo != IP_CT_RELATED || commit)) {
 985                 /* NAT an established or related connection like before. */
 986                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 987                         /* This is the REPLY direction for a connection
 988                          * for which NAT was applied in the forward
 989                          * direction.  Do the reverse NAT.
 990                          */
 991                         maniptype = ct->status & IPS_SRC_NAT
 992                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 993                 else
 994                         maniptype = ct->status & IPS_SRC_NAT
 995                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 996         } else if (ct_action & TCA_CT_ACT_NAT_SRC) {
 997                 maniptype = NF_NAT_MANIP_SRC;
 998         } else if (ct_action & TCA_CT_ACT_NAT_DST) {
 999                 maniptype = NF_NAT_MANIP_DST;
1000         } else {
1001                 return NF_ACCEPT;
1002         }
1003
1004         err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
1005         if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
1006                 if (ct->status & IPS_SRC_NAT) {
1007                         if (maniptype == NF_NAT_MANIP_SRC)
1008                                 maniptype = NF_NAT_MANIP_DST;
1009                         else
1010                                 maniptype = NF_NAT_MANIP_SRC;
1011
1012                         err = ct_nat_execute(skb, ct, ctinfo, range,
1013                                              maniptype);
1014                 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
1015                         err = ct_nat_execute(skb, ct, ctinfo, NULL,
1016                                              NF_NAT_MANIP_SRC);
1017                 }
1018         }
1019         return err;
1020 #else
1021         return NF_ACCEPT;
1022 #endif
1023 }
1024
1025 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
1026                       struct tcf_result *res)
1027 {
1028         struct net *net = dev_net(skb->dev);
1029         bool cached, commit, clear, force;
1030         enum ip_conntrack_info ctinfo;
1031         struct tcf_ct *c = to_ct(a);
1032         struct nf_conn *tmpl = NULL;
1033         struct nf_hook_state state;
1034         int nh_ofs, err, retval;
1035         struct tcf_ct_params *p;
1036         bool skip_add = false;
1037         bool defrag = false;
1038         struct nf_conn *ct;
1039         u8 family;
1040
1041         p = rcu_dereference_bh(c->params);
1042
1043         retval = READ_ONCE(c->tcf_action);
1044         commit = p->ct_action & TCA_CT_ACT_COMMIT;
1045         clear = p->ct_action & TCA_CT_ACT_CLEAR;
1046         force = p->ct_action & TCA_CT_ACT_FORCE;
1047         tmpl = p->tmpl;
1048
1049         tcf_lastuse_update(&c->tcf_tm);
1050         tcf_action_update_bstats(&c->common, skb);
1051
1052         if (clear) {
1053                 tc_skb_cb(skb)->post_ct = false;
1054                 ct = nf_ct_get(skb, &ctinfo);
1055                 if (ct) {
1056                         nf_ct_put(ct);
1057                         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1058                 }
1059
1060                 goto out_clear;
1061         }
1062
1063         family = tcf_ct_skb_nf_family(skb);
1064         if (family == NFPROTO_UNSPEC)
1065                 goto drop;
1066
1067         /* The conntrack module expects to be working at L3.
1068          * We also try to pull the IPv4/6 header to linear area
1069          */
1070         nh_ofs = skb_network_offset(skb);
1071         skb_pull_rcsum(skb, nh_ofs);
1072         err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
1073         if (err == -EINPROGRESS) {
1074                 retval = TC_ACT_STOLEN;
1075                 goto out_clear;
1076         }
1077         if (err)
1078                 goto drop;
1079
1080         err = tcf_ct_skb_network_trim(skb, family);
1081         if (err)
1082                 goto drop;
1083
1084         /* If we are recirculating packets to match on ct fields and
1085          * committing with a separate ct action, then we don't need to
1086          * actually run the packet through conntrack twice unless it's for a
1087          * different zone.
1088          */
1089         cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
1090         if (!cached) {
1091                 if (tcf_ct_flow_table_lookup(p, skb, family)) {
1092                         skip_add = true;
1093                         goto do_nat;
1094                 }
1095
1096                 /* Associate skb with specified zone. */
1097                 if (tmpl) {
1098                         nf_conntrack_put(skb_nfct(skb));
1099                         nf_conntrack_get(&tmpl->ct_general);
1100                         nf_ct_set(skb, tmpl, IP_CT_NEW);
1101                 }
1102
1103                 state.hook = NF_INET_PRE_ROUTING;
1104                 state.net = net;
1105                 state.pf = family;
1106                 err = nf_conntrack_in(skb, &state);
1107                 if (err != NF_ACCEPT)
1108                         goto out_push;
1109         }
1110
1111 do_nat:
1112         ct = nf_ct_get(skb, &ctinfo);
1113         if (!ct)
1114                 goto out_push;
1115         nf_ct_deliver_cached_events(ct);
1116         nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1117
1118         err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1119         if (err != NF_ACCEPT)
1120                 goto drop;
1121
1122         if (commit) {
1123                 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1124                 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1125
1126                 if (!nf_ct_is_confirmed(ct))
1127                         nf_conn_act_ct_ext_add(ct);
1128
1129                 /* This will take care of sending queued events
1130                  * even if the connection is already confirmed.
1131                  */
1132                 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1133                         goto drop;
1134         }
1135
1136         if (!skip_add)
1137                 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1138
1139 out_push:
1140         skb_push_rcsum(skb, nh_ofs);
1141
1142         tc_skb_cb(skb)->post_ct = true;
1143         tc_skb_cb(skb)->zone = p->zone;
1144 out_clear:
1145         if (defrag)
1146                 qdisc_skb_cb(skb)->pkt_len = skb->len;
1147         return retval;
1148
1149 drop:
1150         tcf_action_inc_drop_qstats(&c->common);
1151         return TC_ACT_SHOT;
1152 }
1153
1154 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1155         [TCA_CT_ACTION] = { .type = NLA_U16 },
1156         [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1157         [TCA_CT_ZONE] = { .type = NLA_U16 },
1158         [TCA_CT_MARK] = { .type = NLA_U32 },
1159         [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1160         [TCA_CT_LABELS] = { .type = NLA_BINARY,
1161                             .len = 128 / BITS_PER_BYTE },
1162         [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1163                                  .len = 128 / BITS_PER_BYTE },
1164         [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1165         [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1166         [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1167         [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1168         [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1169         [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1170 };
1171
1172 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1173                                   struct tc_ct *parm,
1174                                   struct nlattr **tb,
1175                                   struct netlink_ext_ack *extack)
1176 {
1177         struct nf_nat_range2 *range;
1178
1179         if (!(p->ct_action & TCA_CT_ACT_NAT))
1180                 return 0;
1181
1182         if (!IS_ENABLED(CONFIG_NF_NAT)) {
1183                 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1184                 return -EOPNOTSUPP;
1185         }
1186
1187         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1188                 return 0;
1189
1190         if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1191             (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1192                 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1193                 return -EOPNOTSUPP;
1194         }
1195
1196         range = &p->range;
1197         if (tb[TCA_CT_NAT_IPV4_MIN]) {
1198                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1199
1200                 p->ipv4_range = true;
1201                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1202                 range->min_addr.ip =
1203                         nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1204
1205                 range->max_addr.ip = max_attr ?
1206                                      nla_get_in_addr(max_attr) :
1207                                      range->min_addr.ip;
1208         } else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1209                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1210
1211                 p->ipv4_range = false;
1212                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1213                 range->min_addr.in6 =
1214                         nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1215
1216                 range->max_addr.in6 = max_attr ?
1217                                       nla_get_in6_addr(max_attr) :
1218                                       range->min_addr.in6;
1219         }
1220
1221         if (tb[TCA_CT_NAT_PORT_MIN]) {
1222                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1223                 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1224
1225                 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1226                                        nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1227                                        range->min_proto.all;
1228         }
1229
1230         return 0;
1231 }
1232
1233 static void tcf_ct_set_key_val(struct nlattr **tb,
1234                                void *val, int val_type,
1235                                void *mask, int mask_type,
1236                                int len)
1237 {
1238         if (!tb[val_type])
1239                 return;
1240         nla_memcpy(val, tb[val_type], len);
1241
1242         if (!mask)
1243                 return;
1244
1245         if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1246                 memset(mask, 0xff, len);
1247         else
1248                 nla_memcpy(mask, tb[mask_type], len);
1249 }
1250
1251 static int tcf_ct_fill_params(struct net *net,
1252                               struct tcf_ct_params *p,
1253                               struct tc_ct *parm,
1254                               struct nlattr **tb,
1255                               struct netlink_ext_ack *extack)
1256 {
1257         struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1258         struct nf_conntrack_zone zone;
1259         struct nf_conn *tmpl;
1260         int err;
1261
1262         p->zone = NF_CT_DEFAULT_ZONE_ID;
1263
1264         tcf_ct_set_key_val(tb,
1265                            &p->ct_action, TCA_CT_ACTION,
1266                            NULL, TCA_CT_UNSPEC,
1267                            sizeof(p->ct_action));
1268
1269         if (p->ct_action & TCA_CT_ACT_CLEAR)
1270                 return 0;
1271
1272         err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1273         if (err)
1274                 return err;
1275
1276         if (tb[TCA_CT_MARK]) {
1277                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1278                         NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1279                         return -EOPNOTSUPP;
1280                 }
1281                 tcf_ct_set_key_val(tb,
1282                                    &p->mark, TCA_CT_MARK,
1283                                    &p->mark_mask, TCA_CT_MARK_MASK,
1284                                    sizeof(p->mark));
1285         }
1286
1287         if (tb[TCA_CT_LABELS]) {
1288                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1289                         NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1290                         return -EOPNOTSUPP;
1291                 }
1292
1293                 if (!tn->labels) {
1294                         NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1295                         return -EOPNOTSUPP;
1296                 }
1297                 tcf_ct_set_key_val(tb,
1298                                    p->labels, TCA_CT_LABELS,
1299                                    p->labels_mask, TCA_CT_LABELS_MASK,
1300                                    sizeof(p->labels));
1301         }
1302
1303         if (tb[TCA_CT_ZONE]) {
1304                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1305                         NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1306                         return -EOPNOTSUPP;
1307                 }
1308
1309                 tcf_ct_set_key_val(tb,
1310                                    &p->zone, TCA_CT_ZONE,
1311                                    NULL, TCA_CT_UNSPEC,
1312                                    sizeof(p->zone));
1313         }
1314
1315         nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1316         tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1317         if (!tmpl) {
1318                 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1319                 return -ENOMEM;
1320         }
1321         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1322         p->tmpl = tmpl;
1323
1324         return 0;
1325 }
1326
1327 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1328                        struct nlattr *est, struct tc_action **a,
1329                        struct tcf_proto *tp, u32 flags,
1330                        struct netlink_ext_ack *extack)
1331 {
1332         struct tc_action_net *tn = net_generic(net, ct_net_id);
1333         bool bind = flags & TCA_ACT_FLAGS_BIND;
1334         struct tcf_ct_params *params = NULL;
1335         struct nlattr *tb[TCA_CT_MAX + 1];
1336         struct tcf_chain *goto_ch = NULL;
1337         struct tc_ct *parm;
1338         struct tcf_ct *c;
1339         int err, res = 0;
1340         u32 index;
1341
1342         if (!nla) {
1343                 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1344                 return -EINVAL;
1345         }
1346
1347         err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1348         if (err < 0)
1349                 return err;
1350
1351         if (!tb[TCA_CT_PARMS]) {
1352                 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1353                 return -EINVAL;
1354         }
1355         parm = nla_data(tb[TCA_CT_PARMS]);
1356         index = parm->index;
1357         err = tcf_idr_check_alloc(tn, &index, a, bind);
1358         if (err < 0)
1359                 return err;
1360
1361         if (!err) {
1362                 err = tcf_idr_create_from_flags(tn, index, est, a,
1363                                                 &act_ct_ops, bind, flags);
1364                 if (err) {
1365                         tcf_idr_cleanup(tn, index);
1366                         return err;
1367                 }
1368                 res = ACT_P_CREATED;
1369         } else {
1370                 if (bind)
1371                         return 0;
1372
1373                 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1374                         tcf_idr_release(*a, bind);
1375                         return -EEXIST;
1376                 }
1377         }
1378         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1379         if (err < 0)
1380                 goto cleanup;
1381
1382         c = to_ct(*a);
1383
1384         params = kzalloc(sizeof(*params), GFP_KERNEL);
1385         if (unlikely(!params)) {
1386                 err = -ENOMEM;
1387                 goto cleanup;
1388         }
1389
1390         err = tcf_ct_fill_params(net, params, parm, tb, extack);
1391         if (err)
1392                 goto cleanup;
1393
1394         err = tcf_ct_flow_table_get(params);
1395         if (err)
1396                 goto cleanup;
1397
1398         spin_lock_bh(&c->tcf_lock);
1399         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1400         params = rcu_replace_pointer(c->params, params,
1401                                      lockdep_is_held(&c->tcf_lock));
1402         spin_unlock_bh(&c->tcf_lock);
1403
1404         if (goto_ch)
1405                 tcf_chain_put_by_act(goto_ch);
1406         if (params)
1407                 call_rcu(&params->rcu, tcf_ct_params_free);
1408
1409         return res;
1410
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 int tcf_ct_walker(struct net *net, struct sk_buff *skb,
1561                          struct netlink_callback *cb, int type,
1562                          const struct tc_action_ops *ops,
1563                          struct netlink_ext_ack *extack)
1564 {
1565         struct tc_action_net *tn = net_generic(net, ct_net_id);
1566
1567         return tcf_generic_walker(tn, skb, cb, type, ops, extack);
1568 }
1569
1570 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index)
1571 {
1572         struct tc_action_net *tn = net_generic(net, ct_net_id);
1573
1574         return tcf_idr_search(tn, a, index);
1575 }
1576
1577 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1578                              u64 drops, u64 lastuse, bool hw)
1579 {
1580         struct tcf_ct *c = to_ct(a);
1581
1582         tcf_action_update_stats(a, bytes, packets, drops, hw);
1583         c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1584 }
1585
1586 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1587                                     u32 *index_inc, bool bind,
1588                                     struct netlink_ext_ack *extack)
1589 {
1590         if (bind) {
1591                 struct flow_action_entry *entry = entry_data;
1592
1593                 entry->id = FLOW_ACTION_CT;
1594                 entry->ct.action = tcf_ct_action(act);
1595                 entry->ct.zone = tcf_ct_zone(act);
1596                 entry->ct.flow_table = tcf_ct_ft(act);
1597                 *index_inc = 1;
1598         } else {
1599                 struct flow_offload_action *fl_action = entry_data;
1600
1601                 fl_action->id = FLOW_ACTION_CT;
1602         }
1603
1604         return 0;
1605 }
1606
1607 static struct tc_action_ops act_ct_ops = {
1608         .kind           =       "ct",
1609         .id             =       TCA_ID_CT,
1610         .owner          =       THIS_MODULE,
1611         .act            =       tcf_ct_act,
1612         .dump           =       tcf_ct_dump,
1613         .init           =       tcf_ct_init,
1614         .cleanup        =       tcf_ct_cleanup,
1615         .walk           =       tcf_ct_walker,
1616         .lookup         =       tcf_ct_search,
1617         .stats_update   =       tcf_stats_update,
1618         .offload_act_setup =    tcf_ct_offload_act_setup,
1619         .size           =       sizeof(struct tcf_ct),
1620 };
1621
1622 static __net_init int ct_init_net(struct net *net)
1623 {
1624         unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1625         struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1626
1627         if (nf_connlabels_get(net, n_bits - 1)) {
1628                 tn->labels = false;
1629                 pr_err("act_ct: Failed to set connlabels length");
1630         } else {
1631                 tn->labels = true;
1632         }
1633
1634         return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1635 }
1636
1637 static void __net_exit ct_exit_net(struct list_head *net_list)
1638 {
1639         struct net *net;
1640
1641         rtnl_lock();
1642         list_for_each_entry(net, net_list, exit_list) {
1643                 struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1644
1645                 if (tn->labels)
1646                         nf_connlabels_put(net);
1647         }
1648         rtnl_unlock();
1649
1650         tc_action_net_exit(net_list, ct_net_id);
1651 }
1652
1653 static struct pernet_operations ct_net_ops = {
1654         .init = ct_init_net,
1655         .exit_batch = ct_exit_net,
1656         .id   = &ct_net_id,
1657         .size = sizeof(struct tc_ct_action_net),
1658 };
1659
1660 static int __init ct_init_module(void)
1661 {
1662         int err;
1663
1664         act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1665         if (!act_ct_wq)
1666                 return -ENOMEM;
1667
1668         err = tcf_ct_flow_tables_init();
1669         if (err)
1670                 goto err_tbl_init;
1671
1672         err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1673         if (err)
1674                 goto err_register;
1675
1676         static_branch_inc(&tcf_frag_xmit_count);
1677
1678         return 0;
1679
1680 err_register:
1681         tcf_ct_flow_tables_uninit();
1682 err_tbl_init:
1683         destroy_workqueue(act_ct_wq);
1684         return err;
1685 }
1686
1687 static void __exit ct_cleanup_module(void)
1688 {
1689         static_branch_dec(&tcf_frag_xmit_count);
1690         tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1691         tcf_ct_flow_tables_uninit();
1692         destroy_workqueue(act_ct_wq);
1693 }
1694
1695 module_init(ct_init_module);
1696 module_exit(ct_cleanup_module);
1697 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1698 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1699 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1700 MODULE_DESCRIPTION("Connection tracking action");
1701 MODULE_LICENSE("GPL v2");