net/openvswitch/conntrack.c

   1 /*
   2  * Copyright (c) 2015 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/openvswitch.h>
  16 #include <linux/tcp.h>
  17 #include <linux/udp.h>
  18 #include <linux/sctp.h>
  19 #include <net/ip.h>
  20 #include <net/netfilter/nf_conntrack_core.h>
  21 #include <net/netfilter/nf_conntrack_helper.h>
  22 #include <net/netfilter/nf_conntrack_labels.h>
  23 #include <net/netfilter/nf_conntrack_seqadj.h>
  24 #include <net/netfilter/nf_conntrack_zones.h>
  25 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  26
  27 #ifdef CONFIG_NF_NAT_NEEDED
  28 #include <linux/netfilter/nf_nat.h>
  29 #include <net/netfilter/nf_nat_core.h>
  30 #include <net/netfilter/nf_nat_l3proto.h>
  31 #endif
  32
  33 #include "datapath.h"
  34 #include "conntrack.h"
  35 #include "flow.h"
  36 #include "flow_netlink.h"
  37
  38 struct ovs_ct_len_tbl {
  39         int maxlen;
  40         int minlen;
  41 };
  42
  43 /* Metadata mark for masked write to conntrack mark */
  44 struct md_mark {
  45         u32 value;
  46         u32 mask;
  47 };
  48
  49 /* Metadata label for masked write to conntrack label. */
  50 struct md_labels {
  51         struct ovs_key_ct_labels value;
  52         struct ovs_key_ct_labels mask;
  53 };
  54
  55 enum ovs_ct_nat {
  56         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  57         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  58         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  59 };
  60
  61 /* Conntrack action context for execution. */
  62 struct ovs_conntrack_info {
  63         struct nf_conntrack_helper *helper;
  64         struct nf_conntrack_zone zone;
  65         struct nf_conn *ct;
  66         u8 commit : 1;
  67         u8 nat : 3;                 /* enum ovs_ct_nat */
  68         u16 family;
  69         struct md_mark mark;
  70         struct md_labels labels;
  71 #ifdef CONFIG_NF_NAT_NEEDED
  72         struct nf_nat_range range;  /* Only present for SRC NAT and DST NAT. */
  73 #endif
  74 };
  75
  76 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
  77
  78 static u16 key_to_nfproto(const struct sw_flow_key *key)
  79 {
  80         switch (ntohs(key->eth.type)) {
  81         case ETH_P_IP:
  82                 return NFPROTO_IPV4;
  83         case ETH_P_IPV6:
  84                 return NFPROTO_IPV6;
  85         default:
  86                 return NFPROTO_UNSPEC;
  87         }
  88 }
  89
  90 /* Map SKB connection state into the values used by flow definition. */
  91 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
  92 {
  93         u8 ct_state = OVS_CS_F_TRACKED;
  94
  95         switch (ctinfo) {
  96         case IP_CT_ESTABLISHED_REPLY:
  97         case IP_CT_RELATED_REPLY:
  98                 ct_state |= OVS_CS_F_REPLY_DIR;
  99                 break;
 100         default:
 101                 break;
 102         }
 103
 104         switch (ctinfo) {
 105         case IP_CT_ESTABLISHED:
 106         case IP_CT_ESTABLISHED_REPLY:
 107                 ct_state |= OVS_CS_F_ESTABLISHED;
 108                 break;
 109         case IP_CT_RELATED:
 110         case IP_CT_RELATED_REPLY:
 111                 ct_state |= OVS_CS_F_RELATED;
 112                 break;
 113         case IP_CT_NEW:
 114                 ct_state |= OVS_CS_F_NEW;
 115                 break;
 116         default:
 117                 break;
 118         }
 119
 120         return ct_state;
 121 }
 122
 123 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 124 {
 125 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 126         return ct ? ct->mark : 0;
 127 #else
 128         return 0;
 129 #endif
 130 }
 131
 132 static void ovs_ct_get_labels(const struct nf_conn *ct,
 133                               struct ovs_key_ct_labels *labels)
 134 {
 135         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 136
 137         if (cl) {
 138                 size_t len = sizeof(cl->bits);
 139
 140                 if (len > OVS_CT_LABELS_LEN)
 141                         len = OVS_CT_LABELS_LEN;
 142                 else if (len < OVS_CT_LABELS_LEN)
 143                         memset(labels, 0, OVS_CT_LABELS_LEN);
 144                 memcpy(labels, cl->bits, len);
 145         } else {
 146                 memset(labels, 0, OVS_CT_LABELS_LEN);
 147         }
 148 }
 149
 150 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 151                                 const struct nf_conntrack_zone *zone,
 152                                 const struct nf_conn *ct)
 153 {
 154         key->ct.state = state;
 155         key->ct.zone = zone->id;
 156         key->ct.mark = ovs_ct_get_mark(ct);
 157         ovs_ct_get_labels(ct, &key->ct.labels);
 158 }
 159
 160 /* Update 'key' based on skb->nfct.  If 'post_ct' is true, then OVS has
 161  * previously sent the packet to conntrack via the ct action.  If
 162  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 163  * initialized from the connection status.
 164  */
 165 static void ovs_ct_update_key(const struct sk_buff *skb,
 166                               const struct ovs_conntrack_info *info,
 167                               struct sw_flow_key *key, bool post_ct,
 168                               bool keep_nat_flags)
 169 {
 170         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 171         enum ip_conntrack_info ctinfo;
 172         struct nf_conn *ct;
 173         u8 state = 0;
 174
 175         ct = nf_ct_get(skb, &ctinfo);
 176         if (ct) {
 177                 state = ovs_ct_get_state(ctinfo);
 178                 /* All unconfirmed entries are NEW connections. */
 179                 if (!nf_ct_is_confirmed(ct))
 180                         state |= OVS_CS_F_NEW;
 181                 /* OVS persists the related flag for the duration of the
 182                  * connection.
 183                  */
 184                 if (ct->master)
 185                         state |= OVS_CS_F_RELATED;
 186                 if (keep_nat_flags) {
 187                         state |= key->ct.state & OVS_CS_F_NAT_MASK;
 188                 } else {
 189                         if (ct->status & IPS_SRC_NAT)
 190                                 state |= OVS_CS_F_SRC_NAT;
 191                         if (ct->status & IPS_DST_NAT)
 192                                 state |= OVS_CS_F_DST_NAT;
 193                 }
 194                 zone = nf_ct_zone(ct);
 195         } else if (post_ct) {
 196                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 197                 if (info)
 198                         zone = &info->zone;
 199         }
 200         __ovs_ct_update_key(key, state, zone, ct);
 201 }
 202
 203 /* This is called to initialize CT key fields possibly coming in from the local
 204  * stack.
 205  */
 206 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 207 {
 208         ovs_ct_update_key(skb, NULL, key, false, false);
 209 }
 210
 211 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 212 {
 213         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 214                 return -EMSGSIZE;
 215
 216         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 217             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 218                 return -EMSGSIZE;
 219
 220         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 221             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 222                 return -EMSGSIZE;
 223
 224         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 225             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 226                     &key->ct.labels))
 227                 return -EMSGSIZE;
 228
 229         return 0;
 230 }
 231
 232 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 233                            u32 ct_mark, u32 mask)
 234 {
 235 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 236         enum ip_conntrack_info ctinfo;
 237         struct nf_conn *ct;
 238         u32 new_mark;
 239
 240         /* The connection could be invalid, in which case set_mark is no-op. */
 241         ct = nf_ct_get(skb, &ctinfo);
 242         if (!ct)
 243                 return 0;
 244
 245         new_mark = ct_mark | (ct->mark & ~(mask));
 246         if (ct->mark != new_mark) {
 247                 ct->mark = new_mark;
 248                 nf_conntrack_event_cache(IPCT_MARK, ct);
 249                 key->ct.mark = new_mark;
 250         }
 251
 252         return 0;
 253 #else
 254         return -ENOTSUPP;
 255 #endif
 256 }
 257
 258 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 259                              const struct ovs_key_ct_labels *labels,
 260                              const struct ovs_key_ct_labels *mask)
 261 {
 262         enum ip_conntrack_info ctinfo;
 263         struct nf_conn_labels *cl;
 264         struct nf_conn *ct;
 265         int err;
 266
 267         /* The connection could be invalid, in which case set_label is no-op.*/
 268         ct = nf_ct_get(skb, &ctinfo);
 269         if (!ct)
 270                 return 0;
 271
 272         cl = nf_ct_labels_find(ct);
 273         if (!cl) {
 274                 nf_ct_labels_ext_add(ct);
 275                 cl = nf_ct_labels_find(ct);
 276         }
 277         if (!cl || sizeof(cl->bits) < OVS_CT_LABELS_LEN)
 278                 return -ENOSPC;
 279
 280         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 281                                     OVS_CT_LABELS_LEN / sizeof(u32));
 282         if (err)
 283                 return err;
 284
 285         ovs_ct_get_labels(ct, &key->ct.labels);
 286         return 0;
 287 }
 288
 289 /* 'skb' should already be pulled to nh_ofs. */
 290 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 291 {
 292         const struct nf_conntrack_helper *helper;
 293         const struct nf_conn_help *help;
 294         enum ip_conntrack_info ctinfo;
 295         unsigned int protoff;
 296         struct nf_conn *ct;
 297         int err;
 298
 299         ct = nf_ct_get(skb, &ctinfo);
 300         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 301                 return NF_ACCEPT;
 302
 303         help = nfct_help(ct);
 304         if (!help)
 305                 return NF_ACCEPT;
 306
 307         helper = rcu_dereference(help->helper);
 308         if (!helper)
 309                 return NF_ACCEPT;
 310
 311         switch (proto) {
 312         case NFPROTO_IPV4:
 313                 protoff = ip_hdrlen(skb);
 314                 break;
 315         case NFPROTO_IPV6: {
 316                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 317                 __be16 frag_off;
 318                 int ofs;
 319
 320                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 321                                        &frag_off);
 322                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 323                         pr_debug("proto header not found\n");
 324                         return NF_ACCEPT;
 325                 }
 326                 protoff = ofs;
 327                 break;
 328         }
 329         default:
 330                 WARN_ONCE(1, "helper invoked on non-IP family!");
 331                 return NF_DROP;
 332         }
 333
 334         err = helper->help(skb, protoff, ct, ctinfo);
 335         if (err != NF_ACCEPT)
 336                 return err;
 337
 338         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 339          * FTP with NAT) adusting the TCP payload size when mangling IP
 340          * addresses and/or port numbers in the text-based control connection.
 341          */
 342         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 343             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 344                 return NF_DROP;
 345         return NF_ACCEPT;
 346 }
 347
 348 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 349  * value if 'skb' is freed.
 350  */
 351 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 352                             u16 zone, struct sk_buff *skb)
 353 {
 354         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 355         int err;
 356
 357         if (key->eth.type == htons(ETH_P_IP)) {
 358                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 359
 360                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 361                 err = ip_defrag(net, skb, user);
 362                 if (err)
 363                         return err;
 364
 365                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 366 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 367         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 368                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 369
 370                 skb_orphan(skb);
 371                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 372                 err = nf_ct_frag6_gather(net, skb, user);
 373                 if (err)
 374                         return err;
 375
 376                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 377                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 378 #endif
 379         } else {
 380                 kfree_skb(skb);
 381                 return -EPFNOSUPPORT;
 382         }
 383
 384         key->ip.frag = OVS_FRAG_TYPE_NONE;
 385         skb_clear_hash(skb);
 386         skb->ignore_df = 1;
 387         *OVS_CB(skb) = ovs_cb;
 388
 389         return 0;
 390 }
 391
 392 static struct nf_conntrack_expect *
 393 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 394                    u16 proto, const struct sk_buff *skb)
 395 {
 396         struct nf_conntrack_tuple tuple;
 397
 398         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 399                 return NULL;
 400         return __nf_ct_expect_find(net, zone, &tuple);
 401 }
 402
 403 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 404 static enum ip_conntrack_info
 405 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 406 {
 407         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 408
 409         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 410                 return IP_CT_ESTABLISHED_REPLY;
 411         /* Once we've had two way comms, always ESTABLISHED. */
 412         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 413                 return IP_CT_ESTABLISHED;
 414         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 415                 return IP_CT_RELATED;
 416         return IP_CT_NEW;
 417 }
 418
 419 /* Find an existing connection which this packet belongs to without
 420  * re-attributing statistics or modifying the connection state.  This allows an
 421  * skb->nfct lost due to an upcall to be recovered during actions execution.
 422  *
 423  * Must be called with rcu_read_lock.
 424  *
 425  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 426  * connection.  Returns NULL if there is no existing entry.
 427  */
 428 static struct nf_conn *
 429 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 430                      u8 l3num, struct sk_buff *skb)
 431 {
 432         struct nf_conntrack_l3proto *l3proto;
 433         struct nf_conntrack_l4proto *l4proto;
 434         struct nf_conntrack_tuple tuple;
 435         struct nf_conntrack_tuple_hash *h;
 436         struct nf_conn *ct;
 437         unsigned int dataoff;
 438         u8 protonum;
 439
 440         l3proto = __nf_ct_l3proto_find(l3num);
 441         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 442                                  &protonum) <= 0) {
 443                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 444                 return NULL;
 445         }
 446         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 447         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 448                              protonum, net, &tuple, l3proto, l4proto)) {
 449                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 450                 return NULL;
 451         }
 452
 453         /* look for tuple match */
 454         h = nf_conntrack_find_get(net, zone, &tuple);
 455         if (!h)
 456                 return NULL;   /* Not found. */
 457
 458         ct = nf_ct_tuplehash_to_ctrack(h);
 459
 460         skb->nfct = &ct->ct_general;
 461         skb->nfctinfo = ovs_ct_get_info(h);
 462         return ct;
 463 }
 464
 465 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 466 static bool skb_nfct_cached(struct net *net,
 467                             const struct sw_flow_key *key,
 468                             const struct ovs_conntrack_info *info,
 469                             struct sk_buff *skb)
 470 {
 471         enum ip_conntrack_info ctinfo;
 472         struct nf_conn *ct;
 473
 474         ct = nf_ct_get(skb, &ctinfo);
 475         /* If no ct, check if we have evidence that an existing conntrack entry
 476          * might be found for this skb.  This happens when we lose a skb->nfct
 477          * due to an upcall.  If the connection was not confirmed, it is not
 478          * cached and needs to be run through conntrack again.
 479          */
 480         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 481             !(key->ct.state & OVS_CS_F_INVALID) &&
 482             key->ct.zone == info->zone.id)
 483                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 484         if (!ct)
 485                 return false;
 486         if (!net_eq(net, read_pnet(&ct->ct_net)))
 487                 return false;
 488         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 489                 return false;
 490         if (info->helper) {
 491                 struct nf_conn_help *help;
 492
 493                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 494                 if (help && rcu_access_pointer(help->helper) != info->helper)
 495                         return false;
 496         }
 497
 498         return true;
 499 }
 500
 501 #ifdef CONFIG_NF_NAT_NEEDED
 502 /* Modelled after nf_nat_ipv[46]_fn().
 503  * range is only used for new, uninitialized NAT state.
 504  * Returns either NF_ACCEPT or NF_DROP.
 505  */
 506 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 507                               enum ip_conntrack_info ctinfo,
 508                               const struct nf_nat_range *range,
 509                               enum nf_nat_manip_type maniptype)
 510 {
 511         int hooknum, nh_off, err = NF_ACCEPT;
 512
 513         nh_off = skb_network_offset(skb);
 514         skb_pull(skb, nh_off);
 515
 516         /* See HOOK2MANIP(). */
 517         if (maniptype == NF_NAT_MANIP_SRC)
 518                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 519         else
 520                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 521
 522         switch (ctinfo) {
 523         case IP_CT_RELATED:
 524         case IP_CT_RELATED_REPLY:
 525                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 526                     skb->protocol == htons(ETH_P_IP) &&
 527                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 528                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 529                                                            hooknum))
 530                                 err = NF_DROP;
 531                         goto push;
 532                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 533                            skb->protocol == htons(ETH_P_IPV6)) {
 534                         __be16 frag_off;
 535                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 536                         int hdrlen = ipv6_skip_exthdr(skb,
 537                                                       sizeof(struct ipv6hdr),
 538                                                       &nexthdr, &frag_off);
 539
 540                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 541                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 542                                                                      ctinfo,
 543                                                                      hooknum,
 544                                                                      hdrlen))
 545                                         err = NF_DROP;
 546                                 goto push;
 547                         }
 548                 }
 549                 /* Non-ICMP, fall thru to initialize if needed. */
 550         case IP_CT_NEW:
 551                 /* Seen it before?  This can happen for loopback, retrans,
 552                  * or local packets.
 553                  */
 554                 if (!nf_nat_initialized(ct, maniptype)) {
 555                         /* Initialize according to the NAT action. */
 556                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 557                                 /* Action is set up to establish a new
 558                                  * mapping.
 559                                  */
 560                                 ? nf_nat_setup_info(ct, range, maniptype)
 561                                 : nf_nat_alloc_null_binding(ct, hooknum);
 562                         if (err != NF_ACCEPT)
 563                                 goto push;
 564                 }
 565                 break;
 566
 567         case IP_CT_ESTABLISHED:
 568         case IP_CT_ESTABLISHED_REPLY:
 569                 break;
 570
 571         default:
 572                 err = NF_DROP;
 573                 goto push;
 574         }
 575
 576         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 577 push:
 578         skb_push(skb, nh_off);
 579
 580         return err;
 581 }
 582
 583 static void ovs_nat_update_key(struct sw_flow_key *key,
 584                                const struct sk_buff *skb,
 585                                enum nf_nat_manip_type maniptype)
 586 {
 587         if (maniptype == NF_NAT_MANIP_SRC) {
 588                 __be16 src;
 589
 590                 key->ct.state |= OVS_CS_F_SRC_NAT;
 591                 if (key->eth.type == htons(ETH_P_IP))
 592                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 593                 else if (key->eth.type == htons(ETH_P_IPV6))
 594                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 595                                sizeof(key->ipv6.addr.src));
 596                 else
 597                         return;
 598
 599                 if (key->ip.proto == IPPROTO_UDP)
 600                         src = udp_hdr(skb)->source;
 601                 else if (key->ip.proto == IPPROTO_TCP)
 602                         src = tcp_hdr(skb)->source;
 603                 else if (key->ip.proto == IPPROTO_SCTP)
 604                         src = sctp_hdr(skb)->source;
 605                 else
 606                         return;
 607
 608                 key->tp.src = src;
 609         } else {
 610                 __be16 dst;
 611
 612                 key->ct.state |= OVS_CS_F_DST_NAT;
 613                 if (key->eth.type == htons(ETH_P_IP))
 614                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 615                 else if (key->eth.type == htons(ETH_P_IPV6))
 616                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 617                                sizeof(key->ipv6.addr.dst));
 618                 else
 619                         return;
 620
 621                 if (key->ip.proto == IPPROTO_UDP)
 622                         dst = udp_hdr(skb)->dest;
 623                 else if (key->ip.proto == IPPROTO_TCP)
 624                         dst = tcp_hdr(skb)->dest;
 625                 else if (key->ip.proto == IPPROTO_SCTP)
 626                         dst = sctp_hdr(skb)->dest;
 627                 else
 628                         return;
 629
 630                 key->tp.dst = dst;
 631         }
 632 }
 633
 634 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 635 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 636                       const struct ovs_conntrack_info *info,
 637                       struct sk_buff *skb, struct nf_conn *ct,
 638                       enum ip_conntrack_info ctinfo)
 639 {
 640         enum nf_nat_manip_type maniptype;
 641         int err;
 642
 643         if (nf_ct_is_untracked(ct)) {
 644                 /* A NAT action may only be performed on tracked packets. */
 645                 return NF_ACCEPT;
 646         }
 647
 648         /* Add NAT extension if not confirmed yet. */
 649         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 650                 return NF_ACCEPT;   /* Can't NAT. */
 651
 652         /* Determine NAT type.
 653          * Check if the NAT type can be deduced from the tracked connection.
 654          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 655          * when committing.
 656          */
 657         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 658             ct->status & IPS_NAT_MASK &&
 659             (ctinfo != IP_CT_RELATED || info->commit)) {
 660                 /* NAT an established or related connection like before. */
 661                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 662                         /* This is the REPLY direction for a connection
 663                          * for which NAT was applied in the forward
 664                          * direction.  Do the reverse NAT.
 665                          */
 666                         maniptype = ct->status & IPS_SRC_NAT
 667                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 668                 else
 669                         maniptype = ct->status & IPS_SRC_NAT
 670                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 671         } else if (info->nat & OVS_CT_SRC_NAT) {
 672                 maniptype = NF_NAT_MANIP_SRC;
 673         } else if (info->nat & OVS_CT_DST_NAT) {
 674                 maniptype = NF_NAT_MANIP_DST;
 675         } else {
 676                 return NF_ACCEPT; /* Connection is not NATed. */
 677         }
 678         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 679
 680         /* Mark NAT done if successful and update the flow key. */
 681         if (err == NF_ACCEPT)
 682                 ovs_nat_update_key(key, skb, maniptype);
 683
 684         return err;
 685 }
 686 #else /* !CONFIG_NF_NAT_NEEDED */
 687 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 688                       const struct ovs_conntrack_info *info,
 689                       struct sk_buff *skb, struct nf_conn *ct,
 690                       enum ip_conntrack_info ctinfo)
 691 {
 692         return NF_ACCEPT;
 693 }
 694 #endif
 695
 696 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 697  * not done already.  Update key with new CT state after passing the packet
 698  * through conntrack.
 699  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 700  * set to NULL and 0 will be returned.
 701  */
 702 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 703                            const struct ovs_conntrack_info *info,
 704                            struct sk_buff *skb)
 705 {
 706         /* If we are recirculating packets to match on conntrack fields and
 707          * committing with a separate conntrack action,  then we don't need to
 708          * actually run the packet through conntrack twice unless it's for a
 709          * different zone.
 710          */
 711         bool cached = skb_nfct_cached(net, key, info, skb);
 712         enum ip_conntrack_info ctinfo;
 713         struct nf_conn *ct;
 714
 715         if (!cached) {
 716                 struct nf_conn *tmpl = info->ct;
 717                 int err;
 718
 719                 /* Associate skb with specified zone. */
 720                 if (tmpl) {
 721                         if (skb->nfct)
 722                                 nf_conntrack_put(skb->nfct);
 723                         nf_conntrack_get(&tmpl->ct_general);
 724                         skb->nfct = &tmpl->ct_general;
 725                         skb->nfctinfo = IP_CT_NEW;
 726                 }
 727
 728                 /* Repeat if requested, see nf_iterate(). */
 729                 do {
 730                         err = nf_conntrack_in(net, info->family,
 731                                               NF_INET_PRE_ROUTING, skb);
 732                 } while (err == NF_REPEAT);
 733
 734                 if (err != NF_ACCEPT)
 735                         return -ENOENT;
 736
 737                 /* Clear CT state NAT flags to mark that we have not yet done
 738                  * NAT after the nf_conntrack_in() call.  We can actually clear
 739                  * the whole state, as it will be re-initialized below.
 740                  */
 741                 key->ct.state = 0;
 742
 743                 /* Update the key, but keep the NAT flags. */
 744                 ovs_ct_update_key(skb, info, key, true, true);
 745         }
 746
 747         ct = nf_ct_get(skb, &ctinfo);
 748         if (ct) {
 749                 /* Packets starting a new connection must be NATted before the
 750                  * helper, so that the helper knows about the NAT.  We enforce
 751                  * this by delaying both NAT and helper calls for unconfirmed
 752                  * connections until the committing CT action.  For later
 753                  * packets NAT and Helper may be called in either order.
 754                  *
 755                  * NAT will be done only if the CT action has NAT, and only
 756                  * once per packet (per zone), as guarded by the NAT bits in
 757                  * the key->ct.state.
 758                  */
 759                 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
 760                     (nf_ct_is_confirmed(ct) || info->commit) &&
 761                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 762                         return -EINVAL;
 763                 }
 764
 765                 /* Userspace may decide to perform a ct lookup without a helper
 766                  * specified followed by a (recirculate and) commit with one.
 767                  * Therefore, for unconfirmed connections which we will commit,
 768                  * we need to attach the helper here.
 769                  */
 770                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 771                     info->helper && !nfct_help(ct)) {
 772                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 773                                                             GFP_ATOMIC);
 774                         if (err)
 775                                 return err;
 776                 }
 777
 778                 /* Call the helper only if:
 779                  * - nf_conntrack_in() was executed above ("!cached") for a
 780                  *   confirmed connection, or
 781                  * - When committing an unconfirmed connection.
 782                  */
 783                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 784                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 785                         return -EINVAL;
 786                 }
 787         }
 788
 789         return 0;
 790 }
 791
 792 /* Lookup connection and read fields into key. */
 793 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 794                          const struct ovs_conntrack_info *info,
 795                          struct sk_buff *skb)
 796 {
 797         struct nf_conntrack_expect *exp;
 798
 799         /* If we pass an expected packet through nf_conntrack_in() the
 800          * expectation is typically removed, but the packet could still be
 801          * lost in upcall processing.  To prevent this from happening we
 802          * perform an explicit expectation lookup.  Expected connections are
 803          * always new, and will be passed through conntrack only when they are
 804          * committed, as it is OK to remove the expectation at that time.
 805          */
 806         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 807         if (exp) {
 808                 u8 state;
 809
 810                 /* NOTE: New connections are NATted and Helped only when
 811                  * committed, so we are not calling into NAT here.
 812                  */
 813                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 814                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 815         } else {
 816                 struct nf_conn *ct;
 817                 int err;
 818
 819                 err = __ovs_ct_lookup(net, key, info, skb);
 820                 if (err)
 821                         return err;
 822
 823                 ct = (struct nf_conn *)skb->nfct;
 824                 if (ct)
 825                         nf_ct_deliver_cached_events(ct);
 826         }
 827
 828         return 0;
 829 }
 830
 831 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 832 {
 833         size_t i;
 834
 835         for (i = 0; i < sizeof(*labels); i++)
 836                 if (labels->ct_labels[i])
 837                         return true;
 838
 839         return false;
 840 }
 841
 842 /* Lookup connection and confirm if unconfirmed. */
 843 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 844                          const struct ovs_conntrack_info *info,
 845                          struct sk_buff *skb)
 846 {
 847         int err;
 848
 849         err = __ovs_ct_lookup(net, key, info, skb);
 850         if (err)
 851                 return err;
 852
 853         /* Apply changes before confirming the connection so that the initial
 854          * conntrack NEW netlink event carries the values given in the CT
 855          * action.
 856          */
 857         if (info->mark.mask) {
 858                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 859                                       info->mark.mask);
 860                 if (err)
 861                         return err;
 862         }
 863         if (labels_nonzero(&info->labels.mask)) {
 864                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 865                                         &info->labels.mask);
 866                 if (err)
 867                         return err;
 868         }
 869         /* This will take care of sending queued events even if the connection
 870          * is already confirmed.
 871          */
 872         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 873                 return -EINVAL;
 874
 875         return 0;
 876 }
 877
 878 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 879  * value if 'skb' is freed.
 880  */
 881 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 882                    struct sw_flow_key *key,
 883                    const struct ovs_conntrack_info *info)
 884 {
 885         int nh_ofs;
 886         int err;
 887
 888         /* The conntrack module expects to be working at L3. */
 889         nh_ofs = skb_network_offset(skb);
 890         skb_pull(skb, nh_ofs);
 891
 892         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 893                 err = handle_fragments(net, key, info->zone.id, skb);
 894                 if (err)
 895                         return err;
 896         }
 897
 898         if (info->commit)
 899                 err = ovs_ct_commit(net, key, info, skb);
 900         else
 901                 err = ovs_ct_lookup(net, key, info, skb);
 902
 903         skb_push(skb, nh_ofs);
 904         if (err)
 905                 kfree_skb(skb);
 906         return err;
 907 }
 908
 909 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 910                              const struct sw_flow_key *key, bool log)
 911 {
 912         struct nf_conntrack_helper *helper;
 913         struct nf_conn_help *help;
 914
 915         helper = nf_conntrack_helper_try_module_get(name, info->family,
 916                                                     key->ip.proto);
 917         if (!helper) {
 918                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 919                 return -EINVAL;
 920         }
 921
 922         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 923         if (!help) {
 924                 module_put(helper->me);
 925                 return -ENOMEM;
 926         }
 927
 928         rcu_assign_pointer(help->helper, helper);
 929         info->helper = helper;
 930         return 0;
 931 }
 932
 933 #ifdef CONFIG_NF_NAT_NEEDED
 934 static int parse_nat(const struct nlattr *attr,
 935                      struct ovs_conntrack_info *info, bool log)
 936 {
 937         struct nlattr *a;
 938         int rem;
 939         bool have_ip_max = false;
 940         bool have_proto_max = false;
 941         bool ip_vers = (info->family == NFPROTO_IPV6);
 942
 943         nla_for_each_nested(a, attr, rem) {
 944                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
 945                         [OVS_NAT_ATTR_SRC] = {0, 0},
 946                         [OVS_NAT_ATTR_DST] = {0, 0},
 947                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
 948                                                  sizeof(struct in6_addr)},
 949                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
 950                                                  sizeof(struct in6_addr)},
 951                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
 952                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
 953                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
 954                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
 955                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
 956                 };
 957                 int type = nla_type(a);
 958
 959                 if (type > OVS_NAT_ATTR_MAX) {
 960                         OVS_NLERR(log,
 961                                   "Unknown NAT attribute (type=%d, max=%d).\n",
 962                                   type, OVS_NAT_ATTR_MAX);
 963                         return -EINVAL;
 964                 }
 965
 966                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
 967                         OVS_NLERR(log,
 968                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
 969                                   type, nla_len(a),
 970                                   ovs_nat_attr_lens[type][ip_vers]);
 971                         return -EINVAL;
 972                 }
 973
 974                 switch (type) {
 975                 case OVS_NAT_ATTR_SRC:
 976                 case OVS_NAT_ATTR_DST:
 977                         if (info->nat) {
 978                                 OVS_NLERR(log,
 979                                           "Only one type of NAT may be specified.\n"
 980                                           );
 981                                 return -ERANGE;
 982                         }
 983                         info->nat |= OVS_CT_NAT;
 984                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
 985                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
 986                         break;
 987
 988                 case OVS_NAT_ATTR_IP_MIN:
 989                         nla_memcpy(&info->range.min_addr, a,
 990                                    sizeof(info->range.min_addr));
 991                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
 992                         break;
 993
 994                 case OVS_NAT_ATTR_IP_MAX:
 995                         have_ip_max = true;
 996                         nla_memcpy(&info->range.max_addr, a,
 997                                    sizeof(info->range.max_addr));
 998                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
 999                         break;
1000
1001                 case OVS_NAT_ATTR_PROTO_MIN:
1002                         info->range.min_proto.all = htons(nla_get_u16(a));
1003                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1004                         break;
1005
1006                 case OVS_NAT_ATTR_PROTO_MAX:
1007                         have_proto_max = true;
1008                         info->range.max_proto.all = htons(nla_get_u16(a));
1009                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1010                         break;
1011
1012                 case OVS_NAT_ATTR_PERSISTENT:
1013                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1014                         break;
1015
1016                 case OVS_NAT_ATTR_PROTO_HASH:
1017                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1018                         break;
1019
1020                 case OVS_NAT_ATTR_PROTO_RANDOM:
1021                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1022                         break;
1023
1024                 default:
1025                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1026                         return -EINVAL;
1027                 }
1028         }
1029
1030         if (rem > 0) {
1031                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1032                 return -EINVAL;
1033         }
1034         if (!info->nat) {
1035                 /* Do not allow flags if no type is given. */
1036                 if (info->range.flags) {
1037                         OVS_NLERR(log,
1038                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1039                                   );
1040                         return -EINVAL;
1041                 }
1042                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1043         } else if (!info->commit) {
1044                 OVS_NLERR(log,
1045                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1046                           );
1047                 return -EINVAL;
1048         }
1049         /* Allow missing IP_MAX. */
1050         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1051                 memcpy(&info->range.max_addr, &info->range.min_addr,
1052                        sizeof(info->range.max_addr));
1053         }
1054         /* Allow missing PROTO_MAX. */
1055         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1056             !have_proto_max) {
1057                 info->range.max_proto.all = info->range.min_proto.all;
1058         }
1059         return 0;
1060 }
1061 #endif
1062
1063 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1064         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1065         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1066                                     .maxlen = sizeof(u16) },
1067         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1068                                     .maxlen = sizeof(struct md_mark) },
1069         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1070                                     .maxlen = sizeof(struct md_labels) },
1071         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1072                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1073 #ifdef CONFIG_NF_NAT_NEEDED
1074         /* NAT length is checked when parsing the nested attributes. */
1075         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1076 #endif
1077 };
1078
1079 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1080                     const char **helper, bool log)
1081 {
1082         struct nlattr *a;
1083         int rem;
1084
1085         nla_for_each_nested(a, attr, rem) {
1086                 int type = nla_type(a);
1087                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1088                 int minlen = ovs_ct_attr_lens[type].minlen;
1089
1090                 if (type > OVS_CT_ATTR_MAX) {
1091                         OVS_NLERR(log,
1092                                   "Unknown conntrack attr (type=%d, max=%d)",
1093                                   type, OVS_CT_ATTR_MAX);
1094                         return -EINVAL;
1095                 }
1096                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1097                         OVS_NLERR(log,
1098                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1099                                   type, nla_len(a), maxlen);
1100                         return -EINVAL;
1101                 }
1102
1103                 switch (type) {
1104                 case OVS_CT_ATTR_COMMIT:
1105                         info->commit = true;
1106                         break;
1107 #ifdef CONFIG_NF_CONNTRACK_ZONES
1108                 case OVS_CT_ATTR_ZONE:
1109                         info->zone.id = nla_get_u16(a);
1110                         break;
1111 #endif
1112 #ifdef CONFIG_NF_CONNTRACK_MARK
1113                 case OVS_CT_ATTR_MARK: {
1114                         struct md_mark *mark = nla_data(a);
1115
1116                         if (!mark->mask) {
1117                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1118                                 return -EINVAL;
1119                         }
1120                         info->mark = *mark;
1121                         break;
1122                 }
1123 #endif
1124 #ifdef CONFIG_NF_CONNTRACK_LABELS
1125                 case OVS_CT_ATTR_LABELS: {
1126                         struct md_labels *labels = nla_data(a);
1127
1128                         if (!labels_nonzero(&labels->mask)) {
1129                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1130                                 return -EINVAL;
1131                         }
1132                         info->labels = *labels;
1133                         break;
1134                 }
1135 #endif
1136                 case OVS_CT_ATTR_HELPER:
1137                         *helper = nla_data(a);
1138                         if (!memchr(*helper, '\0', nla_len(a))) {
1139                                 OVS_NLERR(log, "Invalid conntrack helper");
1140                                 return -EINVAL;
1141                         }
1142                         break;
1143 #ifdef CONFIG_NF_NAT_NEEDED
1144                 case OVS_CT_ATTR_NAT: {
1145                         int err = parse_nat(a, info, log);
1146
1147                         if (err)
1148                                 return err;
1149                         break;
1150                 }
1151 #endif
1152                 default:
1153                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1154                                   type);
1155                         return -EINVAL;
1156                 }
1157         }
1158
1159 #ifdef CONFIG_NF_CONNTRACK_MARK
1160         if (!info->commit && info->mark.mask) {
1161                 OVS_NLERR(log,
1162                           "Setting conntrack mark requires 'commit' flag.");
1163                 return -EINVAL;
1164         }
1165 #endif
1166 #ifdef CONFIG_NF_CONNTRACK_LABELS
1167         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1168                 OVS_NLERR(log,
1169                           "Setting conntrack labels requires 'commit' flag.");
1170                 return -EINVAL;
1171         }
1172 #endif
1173         if (rem > 0) {
1174                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1175                 return -EINVAL;
1176         }
1177
1178         return 0;
1179 }
1180
1181 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1182 {
1183         if (attr == OVS_KEY_ATTR_CT_STATE)
1184                 return true;
1185         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1186             attr == OVS_KEY_ATTR_CT_ZONE)
1187                 return true;
1188         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1189             attr == OVS_KEY_ATTR_CT_MARK)
1190                 return true;
1191         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1192             attr == OVS_KEY_ATTR_CT_LABELS) {
1193                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1194
1195                 return ovs_net->xt_label;
1196         }
1197
1198         return false;
1199 }
1200
1201 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1202                        const struct sw_flow_key *key,
1203                        struct sw_flow_actions **sfa,  bool log)
1204 {
1205         struct ovs_conntrack_info ct_info;
1206         const char *helper = NULL;
1207         u16 family;
1208         int err;
1209
1210         family = key_to_nfproto(key);
1211         if (family == NFPROTO_UNSPEC) {
1212                 OVS_NLERR(log, "ct family unspecified");
1213                 return -EINVAL;
1214         }
1215
1216         memset(&ct_info, 0, sizeof(ct_info));
1217         ct_info.family = family;
1218
1219         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1220                         NF_CT_DEFAULT_ZONE_DIR, 0);
1221
1222         err = parse_ct(attr, &ct_info, &helper, log);
1223         if (err)
1224                 return err;
1225
1226         /* Set up template for tracking connections in specific zones. */
1227         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1228         if (!ct_info.ct) {
1229                 OVS_NLERR(log, "Failed to allocate conntrack template");
1230                 return -ENOMEM;
1231         }
1232
1233         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1234         nf_conntrack_get(&ct_info.ct->ct_general);
1235
1236         if (helper) {
1237                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1238                 if (err)
1239                         goto err_free_ct;
1240         }
1241
1242         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1243                                  sizeof(ct_info), log);
1244         if (err)
1245                 goto err_free_ct;
1246
1247         return 0;
1248 err_free_ct:
1249         __ovs_ct_free_action(&ct_info);
1250         return err;
1251 }
1252
1253 #ifdef CONFIG_NF_NAT_NEEDED
1254 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1255                                struct sk_buff *skb)
1256 {
1257         struct nlattr *start;
1258
1259         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1260         if (!start)
1261                 return false;
1262
1263         if (info->nat & OVS_CT_SRC_NAT) {
1264                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1265                         return false;
1266         } else if (info->nat & OVS_CT_DST_NAT) {
1267                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1268                         return false;
1269         } else {
1270                 goto out;
1271         }
1272
1273         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1274                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1275                     info->family == NFPROTO_IPV4) {
1276                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1277                                             info->range.min_addr.ip) ||
1278                             (info->range.max_addr.ip
1279                              != info->range.min_addr.ip &&
1280                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1281                                               info->range.max_addr.ip))))
1282                                 return false;
1283                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1284                            info->family == NFPROTO_IPV6) {
1285                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1286                                              &info->range.min_addr.in6) ||
1287                             (memcmp(&info->range.max_addr.in6,
1288                                     &info->range.min_addr.in6,
1289                                     sizeof(info->range.max_addr.in6)) &&
1290                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1291                                                &info->range.max_addr.in6))))
1292                                 return false;
1293                 } else {
1294                         return false;
1295                 }
1296         }
1297         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1298             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1299                          ntohs(info->range.min_proto.all)) ||
1300              (info->range.max_proto.all != info->range.min_proto.all &&
1301               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1302                           ntohs(info->range.max_proto.all)))))
1303                 return false;
1304
1305         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1306             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1307                 return false;
1308         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1309             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1310                 return false;
1311         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1312             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1313                 return false;
1314 out:
1315         nla_nest_end(skb, start);
1316
1317         return true;
1318 }
1319 #endif
1320
1321 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1322                           struct sk_buff *skb)
1323 {
1324         struct nlattr *start;
1325
1326         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1327         if (!start)
1328                 return -EMSGSIZE;
1329
1330         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1331                 return -EMSGSIZE;
1332         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1333             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1334                 return -EMSGSIZE;
1335         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1336             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1337                     &ct_info->mark))
1338                 return -EMSGSIZE;
1339         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1340             labels_nonzero(&ct_info->labels.mask) &&
1341             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1342                     &ct_info->labels))
1343                 return -EMSGSIZE;
1344         if (ct_info->helper) {
1345                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1346                                    ct_info->helper->name))
1347                         return -EMSGSIZE;
1348         }
1349 #ifdef CONFIG_NF_NAT_NEEDED
1350         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1351                 return -EMSGSIZE;
1352 #endif
1353         nla_nest_end(skb, start);
1354
1355         return 0;
1356 }
1357
1358 void ovs_ct_free_action(const struct nlattr *a)
1359 {
1360         struct ovs_conntrack_info *ct_info = nla_data(a);
1361
1362         __ovs_ct_free_action(ct_info);
1363 }
1364
1365 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1366 {
1367         if (ct_info->helper)
1368                 module_put(ct_info->helper->me);
1369         if (ct_info->ct)
1370                 nf_ct_put(ct_info->ct);
1371 }
1372
1373 void ovs_ct_init(struct net *net)
1374 {
1375         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1376         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1377
1378         if (nf_connlabels_get(net, n_bits - 1)) {
1379                 ovs_net->xt_label = false;
1380                 OVS_NLERR(true, "Failed to set connlabel length");
1381         } else {
1382                 ovs_net->xt_label = true;
1383         }
1384 }
1385
1386 void ovs_ct_exit(struct net *net)
1387 {
1388         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1389
1390         if (ovs_net->xt_label)
1391                 nf_connlabels_put(net);
1392 }