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