net/sched/act_ct.c

   1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
   2 /* -
   3  * net/sched/act_ct.c  Connection Tracking action
   4  *
   5  * Authors:   Paul Blakey <paulb@mellanox.com>
   6  *            Yossi Kuperman <yossiku@mellanox.com>
   7  *            Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/init.h>
  12 #include <linux/kernel.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/rtnetlink.h>
  15 #include <linux/pkt_cls.h>
  16 #include <linux/ip.h>
  17 #include <linux/ipv6.h>
  18 #include <linux/rhashtable.h>
  19 #include <net/netlink.h>
  20 #include <net/pkt_sched.h>
  21 #include <net/pkt_cls.h>
  22 #include <net/act_api.h>
  23 #include <net/ip.h>
  24 #include <net/ipv6_frag.h>
  25 #include <uapi/linux/tc_act/tc_ct.h>
  26 #include <net/tc_act/tc_ct.h>
  27
  28 #include <net/netfilter/nf_flow_table.h>
  29 #include <net/netfilter/nf_conntrack.h>
  30 #include <net/netfilter/nf_conntrack_core.h>
  31 #include <net/netfilter/nf_conntrack_zones.h>
  32 #include <net/netfilter/nf_conntrack_helper.h>
  33 #include <net/netfilter/nf_conntrack_acct.h>
  34 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  35 #include <uapi/linux/netfilter/nf_nat.h>
  36
  37 static struct workqueue_struct *act_ct_wq;
  38 static struct rhashtable zones_ht;
  39 static DEFINE_MUTEX(zones_mutex);
  40
  41 struct tcf_ct_flow_table {
  42         struct rhash_head node; /* In zones tables */
  43
  44         struct rcu_work rwork;
  45         struct nf_flowtable nf_ft;
  46         refcount_t ref;
  47         u16 zone;
  48
  49         bool dying;
  50 };
  51
  52 static const struct rhashtable_params zones_params = {
  53         .head_offset = offsetof(struct tcf_ct_flow_table, node),
  54         .key_offset = offsetof(struct tcf_ct_flow_table, zone),
  55         .key_len = sizeof_field(struct tcf_ct_flow_table, zone),
  56         .automatic_shrinking = true,
  57 };
  58
  59 static struct flow_action_entry *
  60 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
  61 {
  62         int i = flow_action->num_entries++;
  63
  64         return &flow_action->entries[i];
  65 }
  66
  67 static void tcf_ct_add_mangle_action(struct flow_action *action,
  68                                      enum flow_action_mangle_base htype,
  69                                      u32 offset,
  70                                      u32 mask,
  71                                      u32 val)
  72 {
  73         struct flow_action_entry *entry;
  74
  75         entry = tcf_ct_flow_table_flow_action_get_next(action);
  76         entry->id = FLOW_ACTION_MANGLE;
  77         entry->mangle.htype = htype;
  78         entry->mangle.mask = ~mask;
  79         entry->mangle.offset = offset;
  80         entry->mangle.val = val;
  81 }
  82
  83 /* The following nat helper functions check if the inverted reverse tuple
  84  * (target) is different then the current dir tuple - meaning nat for ports
  85  * and/or ip is needed, and add the relevant mangle actions.
  86  */
  87 static void
  88 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
  89                                       struct nf_conntrack_tuple target,
  90                                       struct flow_action *action)
  91 {
  92         if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
  93                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
  94                                          offsetof(struct iphdr, saddr),
  95                                          0xFFFFFFFF,
  96                                          be32_to_cpu(target.src.u3.ip));
  97         if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
  98                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
  99                                          offsetof(struct iphdr, daddr),
 100                                          0xFFFFFFFF,
 101                                          be32_to_cpu(target.dst.u3.ip));
 102 }
 103
 104 static void
 105 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
 106                                    union nf_inet_addr *addr,
 107                                    u32 offset)
 108 {
 109         int i;
 110
 111         for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
 112                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
 113                                          i * sizeof(u32) + offset,
 114                                          0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
 115 }
 116
 117 static void
 118 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
 119                                       struct nf_conntrack_tuple target,
 120                                       struct flow_action *action)
 121 {
 122         if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
 123                 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
 124                                                    offsetof(struct ipv6hdr,
 125                                                             saddr));
 126         if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
 127                 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
 128                                                    offsetof(struct ipv6hdr,
 129                                                             daddr));
 130 }
 131
 132 static void
 133 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
 134                                      struct nf_conntrack_tuple target,
 135                                      struct flow_action *action)
 136 {
 137         __be16 target_src = target.src.u.tcp.port;
 138         __be16 target_dst = target.dst.u.tcp.port;
 139
 140         if (target_src != tuple->src.u.tcp.port)
 141                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 142                                          offsetof(struct tcphdr, source),
 143                                          0xFFFF, be16_to_cpu(target_src));
 144         if (target_dst != tuple->dst.u.tcp.port)
 145                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 146                                          offsetof(struct tcphdr, dest),
 147                                          0xFFFF, be16_to_cpu(target_dst));
 148 }
 149
 150 static void
 151 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
 152                                      struct nf_conntrack_tuple target,
 153                                      struct flow_action *action)
 154 {
 155         __be16 target_src = target.src.u.udp.port;
 156         __be16 target_dst = target.dst.u.udp.port;
 157
 158         if (target_src != tuple->src.u.udp.port)
 159                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 160                                          offsetof(struct udphdr, source),
 161                                          0xFFFF, be16_to_cpu(target_src));
 162         if (target_dst != tuple->dst.u.udp.port)
 163                 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
 164                                          offsetof(struct udphdr, dest),
 165                                          0xFFFF, be16_to_cpu(target_dst));
 166 }
 167
 168 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
 169                                               enum ip_conntrack_dir dir,
 170                                               struct flow_action *action)
 171 {
 172         struct nf_conn_labels *ct_labels;
 173         struct flow_action_entry *entry;
 174         enum ip_conntrack_info ctinfo;
 175         u32 *act_ct_labels;
 176
 177         entry = tcf_ct_flow_table_flow_action_get_next(action);
 178         entry->id = FLOW_ACTION_CT_METADATA;
 179 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 180         entry->ct_metadata.mark = ct->mark;
 181 #endif
 182         ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
 183                                              IP_CT_ESTABLISHED_REPLY;
 184         /* aligns with the CT reference on the SKB nf_ct_set */
 185         entry->ct_metadata.cookie = (unsigned long)ct | ctinfo;
 186
 187         act_ct_labels = entry->ct_metadata.labels;
 188         ct_labels = nf_ct_labels_find(ct);
 189         if (ct_labels)
 190                 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE);
 191         else
 192                 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE);
 193 }
 194
 195 static int tcf_ct_flow_table_add_action_nat(struct net *net,
 196                                             struct nf_conn *ct,
 197                                             enum ip_conntrack_dir dir,
 198                                             struct flow_action *action)
 199 {
 200         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 201         struct nf_conntrack_tuple target;
 202
 203         if (!(ct->status & IPS_NAT_MASK))
 204                 return 0;
 205
 206         nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
 207
 208         switch (tuple->src.l3num) {
 209         case NFPROTO_IPV4:
 210                 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target,
 211                                                       action);
 212                 break;
 213         case NFPROTO_IPV6:
 214                 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target,
 215                                                       action);
 216                 break;
 217         default:
 218                 return -EOPNOTSUPP;
 219         }
 220
 221         switch (nf_ct_protonum(ct)) {
 222         case IPPROTO_TCP:
 223                 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action);
 224                 break;
 225         case IPPROTO_UDP:
 226                 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action);
 227                 break;
 228         default:
 229                 return -EOPNOTSUPP;
 230         }
 231
 232         return 0;
 233 }
 234
 235 static int tcf_ct_flow_table_fill_actions(struct net *net,
 236                                           const struct flow_offload *flow,
 237                                           enum flow_offload_tuple_dir tdir,
 238                                           struct nf_flow_rule *flow_rule)
 239 {
 240         struct flow_action *action = &flow_rule->rule->action;
 241         int num_entries = action->num_entries;
 242         struct nf_conn *ct = flow->ct;
 243         enum ip_conntrack_dir dir;
 244         int i, err;
 245
 246         switch (tdir) {
 247         case FLOW_OFFLOAD_DIR_ORIGINAL:
 248                 dir = IP_CT_DIR_ORIGINAL;
 249                 break;
 250         case FLOW_OFFLOAD_DIR_REPLY:
 251                 dir = IP_CT_DIR_REPLY;
 252                 break;
 253         default:
 254                 return -EOPNOTSUPP;
 255         }
 256
 257         err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
 258         if (err)
 259                 goto err_nat;
 260
 261         tcf_ct_flow_table_add_action_meta(ct, dir, action);
 262         return 0;
 263
 264 err_nat:
 265         /* Clear filled actions */
 266         for (i = num_entries; i < action->num_entries; i++)
 267                 memset(&action->entries[i], 0, sizeof(action->entries[i]));
 268         action->num_entries = num_entries;
 269
 270         return err;
 271 }
 272
 273 static struct nf_flowtable_type flowtable_ct = {
 274         .action         = tcf_ct_flow_table_fill_actions,
 275         .owner          = THIS_MODULE,
 276 };
 277
 278 static int tcf_ct_flow_table_get(struct tcf_ct_params *params)
 279 {
 280         struct tcf_ct_flow_table *ct_ft;
 281         int err = -ENOMEM;
 282
 283         mutex_lock(&zones_mutex);
 284         ct_ft = rhashtable_lookup_fast(&zones_ht, &params->zone, zones_params);
 285         if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
 286                 goto out_unlock;
 287
 288         ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
 289         if (!ct_ft)
 290                 goto err_alloc;
 291         refcount_set(&ct_ft->ref, 1);
 292
 293         ct_ft->zone = params->zone;
 294         err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
 295         if (err)
 296                 goto err_insert;
 297
 298         ct_ft->nf_ft.type = &flowtable_ct;
 299         ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD;
 300         err = nf_flow_table_init(&ct_ft->nf_ft);
 301         if (err)
 302                 goto err_init;
 303
 304         __module_get(THIS_MODULE);
 305 out_unlock:
 306         params->ct_ft = ct_ft;
 307         params->nf_ft = &ct_ft->nf_ft;
 308         mutex_unlock(&zones_mutex);
 309
 310         return 0;
 311
 312 err_init:
 313         rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 314 err_insert:
 315         kfree(ct_ft);
 316 err_alloc:
 317         mutex_unlock(&zones_mutex);
 318         return err;
 319 }
 320
 321 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
 322 {
 323         struct tcf_ct_flow_table *ct_ft;
 324
 325         ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
 326                              rwork);
 327         nf_flow_table_free(&ct_ft->nf_ft);
 328         kfree(ct_ft);
 329
 330         module_put(THIS_MODULE);
 331 }
 332
 333 static void tcf_ct_flow_table_put(struct tcf_ct_params *params)
 334 {
 335         struct tcf_ct_flow_table *ct_ft = params->ct_ft;
 336
 337         if (refcount_dec_and_test(&params->ct_ft->ref)) {
 338                 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
 339                 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
 340                 queue_rcu_work(act_ct_wq, &ct_ft->rwork);
 341         }
 342 }
 343
 344 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
 345                                   struct nf_conn *ct,
 346                                   bool tcp)
 347 {
 348         struct flow_offload *entry;
 349         int err;
 350
 351         if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 352                 return;
 353
 354         entry = flow_offload_alloc(ct);
 355         if (!entry) {
 356                 WARN_ON_ONCE(1);
 357                 goto err_alloc;
 358         }
 359
 360         if (tcp) {
 361                 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 362                 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 363         }
 364
 365         err = flow_offload_add(&ct_ft->nf_ft, entry);
 366         if (err)
 367                 goto err_add;
 368
 369         return;
 370
 371 err_add:
 372         flow_offload_free(entry);
 373 err_alloc:
 374         clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 375 }
 376
 377 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
 378                                            struct nf_conn *ct,
 379                                            enum ip_conntrack_info ctinfo)
 380 {
 381         bool tcp = false;
 382
 383         if (ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY)
 384                 return;
 385
 386         switch (nf_ct_protonum(ct)) {
 387         case IPPROTO_TCP:
 388                 tcp = true;
 389                 if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
 390                         return;
 391                 break;
 392         case IPPROTO_UDP:
 393                 break;
 394         default:
 395                 return;
 396         }
 397
 398         if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 399             ct->status & IPS_SEQ_ADJUST)
 400                 return;
 401
 402         tcf_ct_flow_table_add(ct_ft, ct, tcp);
 403 }
 404
 405 static bool
 406 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
 407                                   struct flow_offload_tuple *tuple,
 408                                   struct tcphdr **tcph)
 409 {
 410         struct flow_ports *ports;
 411         unsigned int thoff;
 412         struct iphdr *iph;
 413
 414         if (!pskb_network_may_pull(skb, sizeof(*iph)))
 415                 return false;
 416
 417         iph = ip_hdr(skb);
 418         thoff = iph->ihl * 4;
 419
 420         if (ip_is_fragment(iph) ||
 421             unlikely(thoff != sizeof(struct iphdr)))
 422                 return false;
 423
 424         if (iph->protocol != IPPROTO_TCP &&
 425             iph->protocol != IPPROTO_UDP)
 426                 return false;
 427
 428         if (iph->ttl <= 1)
 429                 return false;
 430
 431         if (!pskb_network_may_pull(skb, iph->protocol == IPPROTO_TCP ?
 432                                         thoff + sizeof(struct tcphdr) :
 433                                         thoff + sizeof(*ports)))
 434                 return false;
 435
 436         iph = ip_hdr(skb);
 437         if (iph->protocol == IPPROTO_TCP)
 438                 *tcph = (void *)(skb_network_header(skb) + thoff);
 439
 440         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 441         tuple->src_v4.s_addr = iph->saddr;
 442         tuple->dst_v4.s_addr = iph->daddr;
 443         tuple->src_port = ports->source;
 444         tuple->dst_port = ports->dest;
 445         tuple->l3proto = AF_INET;
 446         tuple->l4proto = iph->protocol;
 447
 448         return true;
 449 }
 450
 451 static bool
 452 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
 453                                   struct flow_offload_tuple *tuple,
 454                                   struct tcphdr **tcph)
 455 {
 456         struct flow_ports *ports;
 457         struct ipv6hdr *ip6h;
 458         unsigned int thoff;
 459
 460         if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
 461                 return false;
 462
 463         ip6h = ipv6_hdr(skb);
 464
 465         if (ip6h->nexthdr != IPPROTO_TCP &&
 466             ip6h->nexthdr != IPPROTO_UDP)
 467                 return false;
 468
 469         if (ip6h->hop_limit <= 1)
 470                 return false;
 471
 472         thoff = sizeof(*ip6h);
 473         if (!pskb_network_may_pull(skb, ip6h->nexthdr == IPPROTO_TCP ?
 474                                         thoff + sizeof(struct tcphdr) :
 475                                         thoff + sizeof(*ports)))
 476                 return false;
 477
 478         ip6h = ipv6_hdr(skb);
 479         if (ip6h->nexthdr == IPPROTO_TCP)
 480                 *tcph = (void *)(skb_network_header(skb) + thoff);
 481
 482         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 483         tuple->src_v6 = ip6h->saddr;
 484         tuple->dst_v6 = ip6h->daddr;
 485         tuple->src_port = ports->source;
 486         tuple->dst_port = ports->dest;
 487         tuple->l3proto = AF_INET6;
 488         tuple->l4proto = ip6h->nexthdr;
 489
 490         return true;
 491 }
 492
 493 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
 494                                      struct sk_buff *skb,
 495                                      u8 family)
 496 {
 497         struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
 498         struct flow_offload_tuple_rhash *tuplehash;
 499         struct flow_offload_tuple tuple = {};
 500         enum ip_conntrack_info ctinfo;
 501         struct tcphdr *tcph = NULL;
 502         struct flow_offload *flow;
 503         struct nf_conn *ct;
 504         u8 dir;
 505
 506         /* Previously seen or loopback */
 507         ct = nf_ct_get(skb, &ctinfo);
 508         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 509                 return false;
 510
 511         switch (family) {
 512         case NFPROTO_IPV4:
 513                 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
 514                         return false;
 515                 break;
 516         case NFPROTO_IPV6:
 517                 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
 518                         return false;
 519                 break;
 520         default:
 521                 return false;
 522         }
 523
 524         tuplehash = flow_offload_lookup(nf_ft, &tuple);
 525         if (!tuplehash)
 526                 return false;
 527
 528         dir = tuplehash->tuple.dir;
 529         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 530         ct = flow->ct;
 531
 532         if (tcph && (unlikely(tcph->fin || tcph->rst))) {
 533                 flow_offload_teardown(flow);
 534                 return false;
 535         }
 536
 537         ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
 538                                                     IP_CT_ESTABLISHED_REPLY;
 539
 540         flow_offload_refresh(nf_ft, flow);
 541         nf_conntrack_get(&ct->ct_general);
 542         nf_ct_set(skb, ct, ctinfo);
 543         nf_ct_acct_update(ct, dir, skb->len);
 544
 545         return true;
 546 }
 547
 548 static int tcf_ct_flow_tables_init(void)
 549 {
 550         return rhashtable_init(&zones_ht, &zones_params);
 551 }
 552
 553 static void tcf_ct_flow_tables_uninit(void)
 554 {
 555         rhashtable_destroy(&zones_ht);
 556 }
 557
 558 static struct tc_action_ops act_ct_ops;
 559 static unsigned int ct_net_id;
 560
 561 struct tc_ct_action_net {
 562         struct tc_action_net tn; /* Must be first */
 563         bool labels;
 564 };
 565
 566 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 567 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
 568                                    u16 zone_id, bool force)
 569 {
 570         enum ip_conntrack_info ctinfo;
 571         struct nf_conn *ct;
 572
 573         ct = nf_ct_get(skb, &ctinfo);
 574         if (!ct)
 575                 return false;
 576         if (!net_eq(net, read_pnet(&ct->ct_net)))
 577                 return false;
 578         if (nf_ct_zone(ct)->id != zone_id)
 579                 return false;
 580
 581         /* Force conntrack entry direction. */
 582         if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 583                 if (nf_ct_is_confirmed(ct))
 584                         nf_ct_kill(ct);
 585
 586                 nf_conntrack_put(&ct->ct_general);
 587                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 588
 589                 return false;
 590         }
 591
 592         return true;
 593 }
 594
 595 /* Trim the skb to the length specified by the IP/IPv6 header,
 596  * removing any trailing lower-layer padding. This prepares the skb
 597  * for higher-layer processing that assumes skb->len excludes padding
 598  * (such as nf_ip_checksum). The caller needs to pull the skb to the
 599  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
 600  */
 601 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
 602 {
 603         unsigned int len;
 604         int err;
 605
 606         switch (family) {
 607         case NFPROTO_IPV4:
 608                 len = ntohs(ip_hdr(skb)->tot_len);
 609                 break;
 610         case NFPROTO_IPV6:
 611                 len = sizeof(struct ipv6hdr)
 612                         + ntohs(ipv6_hdr(skb)->payload_len);
 613                 break;
 614         default:
 615                 len = skb->len;
 616         }
 617
 618         err = pskb_trim_rcsum(skb, len);
 619
 620         return err;
 621 }
 622
 623 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
 624 {
 625         u8 family = NFPROTO_UNSPEC;
 626
 627         switch (skb_protocol(skb, true)) {
 628         case htons(ETH_P_IP):
 629                 family = NFPROTO_IPV4;
 630                 break;
 631         case htons(ETH_P_IPV6):
 632                 family = NFPROTO_IPV6;
 633                 break;
 634         default:
 635                 break;
 636         }
 637
 638         return family;
 639 }
 640
 641 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
 642 {
 643         unsigned int len;
 644
 645         len =  skb_network_offset(skb) + sizeof(struct iphdr);
 646         if (unlikely(skb->len < len))
 647                 return -EINVAL;
 648         if (unlikely(!pskb_may_pull(skb, len)))
 649                 return -ENOMEM;
 650
 651         *frag = ip_is_fragment(ip_hdr(skb));
 652         return 0;
 653 }
 654
 655 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
 656 {
 657         unsigned int flags = 0, len, payload_ofs = 0;
 658         unsigned short frag_off;
 659         int nexthdr;
 660
 661         len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
 662         if (unlikely(skb->len < len))
 663                 return -EINVAL;
 664         if (unlikely(!pskb_may_pull(skb, len)))
 665                 return -ENOMEM;
 666
 667         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 668         if (unlikely(nexthdr < 0))
 669                 return -EPROTO;
 670
 671         *frag = flags & IP6_FH_F_FRAG;
 672         return 0;
 673 }
 674
 675 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
 676                                    u8 family, u16 zone, bool *defrag)
 677 {
 678         enum ip_conntrack_info ctinfo;
 679         struct qdisc_skb_cb cb;
 680         struct nf_conn *ct;
 681         int err = 0;
 682         bool frag;
 683
 684         /* Previously seen (loopback)? Ignore. */
 685         ct = nf_ct_get(skb, &ctinfo);
 686         if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
 687                 return 0;
 688
 689         if (family == NFPROTO_IPV4)
 690                 err = tcf_ct_ipv4_is_fragment(skb, &frag);
 691         else
 692                 err = tcf_ct_ipv6_is_fragment(skb, &frag);
 693         if (err || !frag)
 694                 return err;
 695
 696         skb_get(skb);
 697         cb = *qdisc_skb_cb(skb);
 698
 699         if (family == NFPROTO_IPV4) {
 700                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 701
 702                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 703                 local_bh_disable();
 704                 err = ip_defrag(net, skb, user);
 705                 local_bh_enable();
 706                 if (err && err != -EINPROGRESS)
 707                         goto out_free;
 708
 709                 if (!err)
 710                         *defrag = true;
 711         } else { /* NFPROTO_IPV6 */
 712 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 713                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 714
 715                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 716                 err = nf_ct_frag6_gather(net, skb, user);
 717                 if (err && err != -EINPROGRESS)
 718                         goto out_free;
 719
 720                 if (!err)
 721                         *defrag = true;
 722 #else
 723                 err = -EOPNOTSUPP;
 724                 goto out_free;
 725 #endif
 726         }
 727
 728         *qdisc_skb_cb(skb) = cb;
 729         skb_clear_hash(skb);
 730         skb->ignore_df = 1;
 731         return err;
 732
 733 out_free:
 734         kfree_skb(skb);
 735         return err;
 736 }
 737
 738 static void tcf_ct_params_free(struct rcu_head *head)
 739 {
 740         struct tcf_ct_params *params = container_of(head,
 741                                                     struct tcf_ct_params, rcu);
 742
 743         tcf_ct_flow_table_put(params);
 744
 745         if (params->tmpl)
 746                 nf_conntrack_put(&params->tmpl->ct_general);
 747         kfree(params);
 748 }
 749
 750 #if IS_ENABLED(CONFIG_NF_NAT)
 751 /* Modelled after nf_nat_ipv[46]_fn().
 752  * range is only used for new, uninitialized NAT state.
 753  * Returns either NF_ACCEPT or NF_DROP.
 754  */
 755 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 756                           enum ip_conntrack_info ctinfo,
 757                           const struct nf_nat_range2 *range,
 758                           enum nf_nat_manip_type maniptype)
 759 {
 760         __be16 proto = skb_protocol(skb, true);
 761         int hooknum, err = NF_ACCEPT;
 762
 763         /* See HOOK2MANIP(). */
 764         if (maniptype == NF_NAT_MANIP_SRC)
 765                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 766         else
 767                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 768
 769         switch (ctinfo) {
 770         case IP_CT_RELATED:
 771         case IP_CT_RELATED_REPLY:
 772                 if (proto == htons(ETH_P_IP) &&
 773                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 774                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 775                                                            hooknum))
 776                                 err = NF_DROP;
 777                         goto out;
 778                 } else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) {
 779                         __be16 frag_off;
 780                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 781                         int hdrlen = ipv6_skip_exthdr(skb,
 782                                                       sizeof(struct ipv6hdr),
 783                                                       &nexthdr, &frag_off);
 784
 785                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 786                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 787                                                                      ctinfo,
 788                                                                      hooknum,
 789                                                                      hdrlen))
 790                                         err = NF_DROP;
 791                                 goto out;
 792                         }
 793                 }
 794                 /* Non-ICMP, fall thru to initialize if needed. */
 795                 /* fall through */
 796         case IP_CT_NEW:
 797                 /* Seen it before?  This can happen for loopback, retrans,
 798                  * or local packets.
 799                  */
 800                 if (!nf_nat_initialized(ct, maniptype)) {
 801                         /* Initialize according to the NAT action. */
 802                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 803                                 /* Action is set up to establish a new
 804                                  * mapping.
 805                                  */
 806                                 ? nf_nat_setup_info(ct, range, maniptype)
 807                                 : nf_nat_alloc_null_binding(ct, hooknum);
 808                         if (err != NF_ACCEPT)
 809                                 goto out;
 810                 }
 811                 break;
 812
 813         case IP_CT_ESTABLISHED:
 814         case IP_CT_ESTABLISHED_REPLY:
 815                 break;
 816
 817         default:
 818                 err = NF_DROP;
 819                 goto out;
 820         }
 821
 822         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 823 out:
 824         return err;
 825 }
 826 #endif /* CONFIG_NF_NAT */
 827
 828 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
 829 {
 830 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 831         u32 new_mark;
 832
 833         if (!mask)
 834                 return;
 835
 836         new_mark = mark | (ct->mark & ~(mask));
 837         if (ct->mark != new_mark) {
 838                 ct->mark = new_mark;
 839                 if (nf_ct_is_confirmed(ct))
 840                         nf_conntrack_event_cache(IPCT_MARK, ct);
 841         }
 842 #endif
 843 }
 844
 845 static void tcf_ct_act_set_labels(struct nf_conn *ct,
 846                                   u32 *labels,
 847                                   u32 *labels_m)
 848 {
 849 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
 850         size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
 851
 852         if (!memchr_inv(labels_m, 0, labels_sz))
 853                 return;
 854
 855         nf_connlabels_replace(ct, labels, labels_m, 4);
 856 #endif
 857 }
 858
 859 static int tcf_ct_act_nat(struct sk_buff *skb,
 860                           struct nf_conn *ct,
 861                           enum ip_conntrack_info ctinfo,
 862                           int ct_action,
 863                           struct nf_nat_range2 *range,
 864                           bool commit)
 865 {
 866 #if IS_ENABLED(CONFIG_NF_NAT)
 867         int err;
 868         enum nf_nat_manip_type maniptype;
 869
 870         if (!(ct_action & TCA_CT_ACT_NAT))
 871                 return NF_ACCEPT;
 872
 873         /* Add NAT extension if not confirmed yet. */
 874         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 875                 return NF_DROP;   /* Can't NAT. */
 876
 877         if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
 878             (ctinfo != IP_CT_RELATED || commit)) {
 879                 /* NAT an established or related connection like before. */
 880                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 881                         /* This is the REPLY direction for a connection
 882                          * for which NAT was applied in the forward
 883                          * direction.  Do the reverse NAT.
 884                          */
 885                         maniptype = ct->status & IPS_SRC_NAT
 886                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 887                 else
 888                         maniptype = ct->status & IPS_SRC_NAT
 889                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 890         } else if (ct_action & TCA_CT_ACT_NAT_SRC) {
 891                 maniptype = NF_NAT_MANIP_SRC;
 892         } else if (ct_action & TCA_CT_ACT_NAT_DST) {
 893                 maniptype = NF_NAT_MANIP_DST;
 894         } else {
 895                 return NF_ACCEPT;
 896         }
 897
 898         err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
 899         if (err == NF_ACCEPT &&
 900             ct->status & IPS_SRC_NAT && ct->status & IPS_DST_NAT) {
 901                 if (maniptype == NF_NAT_MANIP_SRC)
 902                         maniptype = NF_NAT_MANIP_DST;
 903                 else
 904                         maniptype = NF_NAT_MANIP_SRC;
 905
 906                 err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
 907         }
 908         return err;
 909 #else
 910         return NF_ACCEPT;
 911 #endif
 912 }
 913
 914 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 915                       struct tcf_result *res)
 916 {
 917         struct net *net = dev_net(skb->dev);
 918         bool cached, commit, clear, force;
 919         enum ip_conntrack_info ctinfo;
 920         struct tcf_ct *c = to_ct(a);
 921         struct nf_conn *tmpl = NULL;
 922         struct nf_hook_state state;
 923         int nh_ofs, err, retval;
 924         struct tcf_ct_params *p;
 925         bool skip_add = false;
 926         bool defrag = false;
 927         struct nf_conn *ct;
 928         u8 family;
 929
 930         p = rcu_dereference_bh(c->params);
 931
 932         retval = READ_ONCE(c->tcf_action);
 933         commit = p->ct_action & TCA_CT_ACT_COMMIT;
 934         clear = p->ct_action & TCA_CT_ACT_CLEAR;
 935         force = p->ct_action & TCA_CT_ACT_FORCE;
 936         tmpl = p->tmpl;
 937
 938         tcf_lastuse_update(&c->tcf_tm);
 939
 940         if (clear) {
 941                 ct = nf_ct_get(skb, &ctinfo);
 942                 if (ct) {
 943                         nf_conntrack_put(&ct->ct_general);
 944                         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 945                 }
 946
 947                 goto out;
 948         }
 949
 950         family = tcf_ct_skb_nf_family(skb);
 951         if (family == NFPROTO_UNSPEC)
 952                 goto drop;
 953
 954         /* The conntrack module expects to be working at L3.
 955          * We also try to pull the IPv4/6 header to linear area
 956          */
 957         nh_ofs = skb_network_offset(skb);
 958         skb_pull_rcsum(skb, nh_ofs);
 959         err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
 960         if (err == -EINPROGRESS) {
 961                 retval = TC_ACT_STOLEN;
 962                 goto out;
 963         }
 964         if (err)
 965                 goto drop;
 966
 967         err = tcf_ct_skb_network_trim(skb, family);
 968         if (err)
 969                 goto drop;
 970
 971         /* If we are recirculating packets to match on ct fields and
 972          * committing with a separate ct action, then we don't need to
 973          * actually run the packet through conntrack twice unless it's for a
 974          * different zone.
 975          */
 976         cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
 977         if (!cached) {
 978                 if (!commit && tcf_ct_flow_table_lookup(p, skb, family)) {
 979                         skip_add = true;
 980                         goto do_nat;
 981                 }
 982
 983                 /* Associate skb with specified zone. */
 984                 if (tmpl) {
 985                         ct = nf_ct_get(skb, &ctinfo);
 986                         if (skb_nfct(skb))
 987                                 nf_conntrack_put(skb_nfct(skb));
 988                         nf_conntrack_get(&tmpl->ct_general);
 989                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 990                 }
 991
 992                 state.hook = NF_INET_PRE_ROUTING;
 993                 state.net = net;
 994                 state.pf = family;
 995                 err = nf_conntrack_in(skb, &state);
 996                 if (err != NF_ACCEPT)
 997                         goto out_push;
 998         }
 999
1000 do_nat:
1001         ct = nf_ct_get(skb, &ctinfo);
1002         if (!ct)
1003                 goto out_push;
1004         nf_ct_deliver_cached_events(ct);
1005
1006         err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1007         if (err != NF_ACCEPT)
1008                 goto drop;
1009
1010         if (commit) {
1011                 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1012                 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1013
1014                 /* This will take care of sending queued events
1015                  * even if the connection is already confirmed.
1016                  */
1017                 nf_conntrack_confirm(skb);
1018         } else if (!skip_add) {
1019                 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1020         }
1021
1022 out_push:
1023         skb_push_rcsum(skb, nh_ofs);
1024
1025 out:
1026         tcf_action_update_bstats(&c->common, skb);
1027         if (defrag)
1028                 qdisc_skb_cb(skb)->pkt_len = skb->len;
1029         return retval;
1030
1031 drop:
1032         tcf_action_inc_drop_qstats(&c->common);
1033         return TC_ACT_SHOT;
1034 }
1035
1036 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1037         [TCA_CT_ACTION] = { .type = NLA_U16 },
1038         [TCA_CT_PARMS] = { .type = NLA_EXACT_LEN, .len = sizeof(struct tc_ct) },
1039         [TCA_CT_ZONE] = { .type = NLA_U16 },
1040         [TCA_CT_MARK] = { .type = NLA_U32 },
1041         [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1042         [TCA_CT_LABELS] = { .type = NLA_BINARY,
1043                             .len = 128 / BITS_PER_BYTE },
1044         [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1045                                  .len = 128 / BITS_PER_BYTE },
1046         [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1047         [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1048         [TCA_CT_NAT_IPV6_MIN] = { .type = NLA_EXACT_LEN,
1049                                   .len = sizeof(struct in6_addr) },
1050         [TCA_CT_NAT_IPV6_MAX] = { .type = NLA_EXACT_LEN,
1051                                    .len = sizeof(struct in6_addr) },
1052         [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1053         [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1054 };
1055
1056 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1057                                   struct tc_ct *parm,
1058                                   struct nlattr **tb,
1059                                   struct netlink_ext_ack *extack)
1060 {
1061         struct nf_nat_range2 *range;
1062
1063         if (!(p->ct_action & TCA_CT_ACT_NAT))
1064                 return 0;
1065
1066         if (!IS_ENABLED(CONFIG_NF_NAT)) {
1067                 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1068                 return -EOPNOTSUPP;
1069         }
1070
1071         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1072                 return 0;
1073
1074         if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1075             (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1076                 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1077                 return -EOPNOTSUPP;
1078         }
1079
1080         range = &p->range;
1081         if (tb[TCA_CT_NAT_IPV4_MIN]) {
1082                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1083
1084                 p->ipv4_range = true;
1085                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1086                 range->min_addr.ip =
1087                         nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1088
1089                 range->max_addr.ip = max_attr ?
1090                                      nla_get_in_addr(max_attr) :
1091                                      range->min_addr.ip;
1092         } else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1093                 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1094
1095                 p->ipv4_range = false;
1096                 range->flags |= NF_NAT_RANGE_MAP_IPS;
1097                 range->min_addr.in6 =
1098                         nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1099
1100                 range->max_addr.in6 = max_attr ?
1101                                       nla_get_in6_addr(max_attr) :
1102                                       range->min_addr.in6;
1103         }
1104
1105         if (tb[TCA_CT_NAT_PORT_MIN]) {
1106                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1107                 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1108
1109                 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1110                                        nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1111                                        range->min_proto.all;
1112         }
1113
1114         return 0;
1115 }
1116
1117 static void tcf_ct_set_key_val(struct nlattr **tb,
1118                                void *val, int val_type,
1119                                void *mask, int mask_type,
1120                                int len)
1121 {
1122         if (!tb[val_type])
1123                 return;
1124         nla_memcpy(val, tb[val_type], len);
1125
1126         if (!mask)
1127                 return;
1128
1129         if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1130                 memset(mask, 0xff, len);
1131         else
1132                 nla_memcpy(mask, tb[mask_type], len);
1133 }
1134
1135 static int tcf_ct_fill_params(struct net *net,
1136                               struct tcf_ct_params *p,
1137                               struct tc_ct *parm,
1138                               struct nlattr **tb,
1139                               struct netlink_ext_ack *extack)
1140 {
1141         struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1142         struct nf_conntrack_zone zone;
1143         struct nf_conn *tmpl;
1144         int err;
1145
1146         p->zone = NF_CT_DEFAULT_ZONE_ID;
1147
1148         tcf_ct_set_key_val(tb,
1149                            &p->ct_action, TCA_CT_ACTION,
1150                            NULL, TCA_CT_UNSPEC,
1151                            sizeof(p->ct_action));
1152
1153         if (p->ct_action & TCA_CT_ACT_CLEAR)
1154                 return 0;
1155
1156         err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1157         if (err)
1158                 return err;
1159
1160         if (tb[TCA_CT_MARK]) {
1161                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1162                         NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1163                         return -EOPNOTSUPP;
1164                 }
1165                 tcf_ct_set_key_val(tb,
1166                                    &p->mark, TCA_CT_MARK,
1167                                    &p->mark_mask, TCA_CT_MARK_MASK,
1168                                    sizeof(p->mark));
1169         }
1170
1171         if (tb[TCA_CT_LABELS]) {
1172                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1173                         NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1174                         return -EOPNOTSUPP;
1175                 }
1176
1177                 if (!tn->labels) {
1178                         NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1179                         return -EOPNOTSUPP;
1180                 }
1181                 tcf_ct_set_key_val(tb,
1182                                    p->labels, TCA_CT_LABELS,
1183                                    p->labels_mask, TCA_CT_LABELS_MASK,
1184                                    sizeof(p->labels));
1185         }
1186
1187         if (tb[TCA_CT_ZONE]) {
1188                 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1189                         NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1190                         return -EOPNOTSUPP;
1191                 }
1192
1193                 tcf_ct_set_key_val(tb,
1194                                    &p->zone, TCA_CT_ZONE,
1195                                    NULL, TCA_CT_UNSPEC,
1196                                    sizeof(p->zone));
1197         }
1198
1199         if (p->zone == NF_CT_DEFAULT_ZONE_ID)
1200                 return 0;
1201
1202         nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1203         tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1204         if (!tmpl) {
1205                 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1206                 return -ENOMEM;
1207         }
1208         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1209         nf_conntrack_get(&tmpl->ct_general);
1210         p->tmpl = tmpl;
1211
1212         return 0;
1213 }
1214
1215 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1216                        struct nlattr *est, struct tc_action **a,
1217                        int replace, int bind, bool rtnl_held,
1218                        struct tcf_proto *tp, u32 flags,
1219                        struct netlink_ext_ack *extack)
1220 {
1221         struct tc_action_net *tn = net_generic(net, ct_net_id);
1222         struct tcf_ct_params *params = NULL;
1223         struct nlattr *tb[TCA_CT_MAX + 1];
1224         struct tcf_chain *goto_ch = NULL;
1225         struct tc_ct *parm;
1226         struct tcf_ct *c;
1227         int err, res = 0;
1228         u32 index;
1229
1230         if (!nla) {
1231                 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1232                 return -EINVAL;
1233         }
1234
1235         err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1236         if (err < 0)
1237                 return err;
1238
1239         if (!tb[TCA_CT_PARMS]) {
1240                 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1241                 return -EINVAL;
1242         }
1243         parm = nla_data(tb[TCA_CT_PARMS]);
1244         index = parm->index;
1245         err = tcf_idr_check_alloc(tn, &index, a, bind);
1246         if (err < 0)
1247                 return err;
1248
1249         if (!err) {
1250                 err = tcf_idr_create_from_flags(tn, index, est, a,
1251                                                 &act_ct_ops, bind, flags);
1252                 if (err) {
1253                         tcf_idr_cleanup(tn, index);
1254                         return err;
1255                 }
1256                 res = ACT_P_CREATED;
1257         } else {
1258                 if (bind)
1259                         return 0;
1260
1261                 if (!replace) {
1262                         tcf_idr_release(*a, bind);
1263                         return -EEXIST;
1264                 }
1265         }
1266         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1267         if (err < 0)
1268                 goto cleanup;
1269
1270         c = to_ct(*a);
1271
1272         params = kzalloc(sizeof(*params), GFP_KERNEL);
1273         if (unlikely(!params)) {
1274                 err = -ENOMEM;
1275                 goto cleanup;
1276         }
1277
1278         err = tcf_ct_fill_params(net, params, parm, tb, extack);
1279         if (err)
1280                 goto cleanup;
1281
1282         err = tcf_ct_flow_table_get(params);
1283         if (err)
1284                 goto cleanup;
1285
1286         spin_lock_bh(&c->tcf_lock);
1287         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1288         params = rcu_replace_pointer(c->params, params,
1289                                      lockdep_is_held(&c->tcf_lock));
1290         spin_unlock_bh(&c->tcf_lock);
1291
1292         if (goto_ch)
1293                 tcf_chain_put_by_act(goto_ch);
1294         if (params)
1295                 call_rcu(&params->rcu, tcf_ct_params_free);
1296         if (res == ACT_P_CREATED)
1297                 tcf_idr_insert(tn, *a);
1298
1299         return res;
1300
1301 cleanup:
1302         if (goto_ch)
1303                 tcf_chain_put_by_act(goto_ch);
1304         kfree(params);
1305         tcf_idr_release(*a, bind);
1306         return err;
1307 }
1308
1309 static void tcf_ct_cleanup(struct tc_action *a)
1310 {
1311         struct tcf_ct_params *params;
1312         struct tcf_ct *c = to_ct(a);
1313
1314         params = rcu_dereference_protected(c->params, 1);
1315         if (params)
1316                 call_rcu(&params->rcu, tcf_ct_params_free);
1317 }
1318
1319 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1320                                void *val, int val_type,
1321                                void *mask, int mask_type,
1322                                int len)
1323 {
1324         int err;
1325
1326         if (mask && !memchr_inv(mask, 0, len))
1327                 return 0;
1328
1329         err = nla_put(skb, val_type, len, val);
1330         if (err)
1331                 return err;
1332
1333         if (mask_type != TCA_CT_UNSPEC) {
1334                 err = nla_put(skb, mask_type, len, mask);
1335                 if (err)
1336                         return err;
1337         }
1338
1339         return 0;
1340 }
1341
1342 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1343 {
1344         struct nf_nat_range2 *range = &p->range;
1345
1346         if (!(p->ct_action & TCA_CT_ACT_NAT))
1347                 return 0;
1348
1349         if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1350                 return 0;
1351
1352         if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1353                 if (p->ipv4_range) {
1354                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1355                                             range->min_addr.ip))
1356                                 return -1;
1357                         if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1358                                             range->max_addr.ip))
1359                                 return -1;
1360                 } else {
1361                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1362                                              &range->min_addr.in6))
1363                                 return -1;
1364                         if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1365                                              &range->max_addr.in6))
1366                                 return -1;
1367                 }
1368         }
1369
1370         if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1371                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1372                                  range->min_proto.all))
1373                         return -1;
1374                 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1375                                  range->max_proto.all))
1376                         return -1;
1377         }
1378
1379         return 0;
1380 }
1381
1382 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1383                               int bind, int ref)
1384 {
1385         unsigned char *b = skb_tail_pointer(skb);
1386         struct tcf_ct *c = to_ct(a);
1387         struct tcf_ct_params *p;
1388
1389         struct tc_ct opt = {
1390                 .index   = c->tcf_index,
1391                 .refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1392                 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1393         };
1394         struct tcf_t t;
1395
1396         spin_lock_bh(&c->tcf_lock);
1397         p = rcu_dereference_protected(c->params,
1398                                       lockdep_is_held(&c->tcf_lock));
1399         opt.action = c->tcf_action;
1400
1401         if (tcf_ct_dump_key_val(skb,
1402                                 &p->ct_action, TCA_CT_ACTION,
1403                                 NULL, TCA_CT_UNSPEC,
1404                                 sizeof(p->ct_action)))
1405                 goto nla_put_failure;
1406
1407         if (p->ct_action & TCA_CT_ACT_CLEAR)
1408                 goto skip_dump;
1409
1410         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1411             tcf_ct_dump_key_val(skb,
1412                                 &p->mark, TCA_CT_MARK,
1413                                 &p->mark_mask, TCA_CT_MARK_MASK,
1414                                 sizeof(p->mark)))
1415                 goto nla_put_failure;
1416
1417         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1418             tcf_ct_dump_key_val(skb,
1419                                 p->labels, TCA_CT_LABELS,
1420                                 p->labels_mask, TCA_CT_LABELS_MASK,
1421                                 sizeof(p->labels)))
1422                 goto nla_put_failure;
1423
1424         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1425             tcf_ct_dump_key_val(skb,
1426                                 &p->zone, TCA_CT_ZONE,
1427                                 NULL, TCA_CT_UNSPEC,
1428                                 sizeof(p->zone)))
1429                 goto nla_put_failure;
1430
1431         if (tcf_ct_dump_nat(skb, p))
1432                 goto nla_put_failure;
1433
1434 skip_dump:
1435         if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1436                 goto nla_put_failure;
1437
1438         tcf_tm_dump(&t, &c->tcf_tm);
1439         if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1440                 goto nla_put_failure;
1441         spin_unlock_bh(&c->tcf_lock);
1442
1443         return skb->len;
1444 nla_put_failure:
1445         spin_unlock_bh(&c->tcf_lock);
1446         nlmsg_trim(skb, b);
1447         return -1;
1448 }
1449
1450 static int tcf_ct_walker(struct net *net, struct sk_buff *skb,
1451                          struct netlink_callback *cb, int type,
1452                          const struct tc_action_ops *ops,
1453                          struct netlink_ext_ack *extack)
1454 {
1455         struct tc_action_net *tn = net_generic(net, ct_net_id);
1456
1457         return tcf_generic_walker(tn, skb, cb, type, ops, extack);
1458 }
1459
1460 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index)
1461 {
1462         struct tc_action_net *tn = net_generic(net, ct_net_id);
1463
1464         return tcf_idr_search(tn, a, index);
1465 }
1466
1467 static void tcf_stats_update(struct tc_action *a, u64 bytes, u32 packets,
1468                              u64 lastuse, bool hw)
1469 {
1470         struct tcf_ct *c = to_ct(a);
1471
1472         tcf_action_update_stats(a, bytes, packets, false, hw);
1473         c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1474 }
1475
1476 static struct tc_action_ops act_ct_ops = {
1477         .kind           =       "ct",
1478         .id             =       TCA_ID_CT,
1479         .owner          =       THIS_MODULE,
1480         .act            =       tcf_ct_act,
1481         .dump           =       tcf_ct_dump,
1482         .init           =       tcf_ct_init,
1483         .cleanup        =       tcf_ct_cleanup,
1484         .walk           =       tcf_ct_walker,
1485         .lookup         =       tcf_ct_search,
1486         .stats_update   =       tcf_stats_update,
1487         .size           =       sizeof(struct tcf_ct),
1488 };
1489
1490 static __net_init int ct_init_net(struct net *net)
1491 {
1492         unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1493         struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1494
1495         if (nf_connlabels_get(net, n_bits - 1)) {
1496                 tn->labels = false;
1497                 pr_err("act_ct: Failed to set connlabels length");
1498         } else {
1499                 tn->labels = true;
1500         }
1501
1502         return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1503 }
1504
1505 static void __net_exit ct_exit_net(struct list_head *net_list)
1506 {
1507         struct net *net;
1508
1509         rtnl_lock();
1510         list_for_each_entry(net, net_list, exit_list) {
1511                 struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1512
1513                 if (tn->labels)
1514                         nf_connlabels_put(net);
1515         }
1516         rtnl_unlock();
1517
1518         tc_action_net_exit(net_list, ct_net_id);
1519 }
1520
1521 static struct pernet_operations ct_net_ops = {
1522         .init = ct_init_net,
1523         .exit_batch = ct_exit_net,
1524         .id   = &ct_net_id,
1525         .size = sizeof(struct tc_ct_action_net),
1526 };
1527
1528 static int __init ct_init_module(void)
1529 {
1530         int err;
1531
1532         act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1533         if (!act_ct_wq)
1534                 return -ENOMEM;
1535
1536         err = tcf_ct_flow_tables_init();
1537         if (err)
1538                 goto err_tbl_init;
1539
1540         err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1541         if (err)
1542                 goto err_register;
1543
1544         return 0;
1545
1546 err_tbl_init:
1547         destroy_workqueue(act_ct_wq);
1548 err_register:
1549         tcf_ct_flow_tables_uninit();
1550         return err;
1551 }
1552
1553 static void __exit ct_cleanup_module(void)
1554 {
1555         tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1556         tcf_ct_flow_tables_uninit();
1557         destroy_workqueue(act_ct_wq);
1558 }
1559
1560 module_init(ct_init_module);
1561 module_exit(ct_cleanup_module);
1562 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1563 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1564 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1565 MODULE_DESCRIPTION("Connection tracking action");
1566 MODULE_LICENSE("GPL v2");