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                         if (err != -EINPROGRESS)
 375                                 kfree_skb(skb);
 376                         return err;
 377                 }
 378
 379                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 380                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 381 #endif
 382         } else {
 383                 kfree_skb(skb);
 384                 return -EPFNOSUPPORT;
 385         }
 386
 387         key->ip.frag = OVS_FRAG_TYPE_NONE;
 388         skb_clear_hash(skb);
 389         skb->ignore_df = 1;
 390         *OVS_CB(skb) = ovs_cb;
 391
 392         return 0;
 393 }
 394
 395 static struct nf_conntrack_expect *
 396 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 397                    u16 proto, const struct sk_buff *skb)
 398 {
 399         struct nf_conntrack_tuple tuple;
 400
 401         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 402                 return NULL;
 403         return __nf_ct_expect_find(net, zone, &tuple);
 404 }
 405
 406 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 407 static enum ip_conntrack_info
 408 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 409 {
 410         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 411
 412         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 413                 return IP_CT_ESTABLISHED_REPLY;
 414         /* Once we've had two way comms, always ESTABLISHED. */
 415         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 416                 return IP_CT_ESTABLISHED;
 417         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 418                 return IP_CT_RELATED;
 419         return IP_CT_NEW;
 420 }
 421
 422 /* Find an existing connection which this packet belongs to without
 423  * re-attributing statistics or modifying the connection state.  This allows an
 424  * skb->nfct lost due to an upcall to be recovered during actions execution.
 425  *
 426  * Must be called with rcu_read_lock.
 427  *
 428  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 429  * connection.  Returns NULL if there is no existing entry.
 430  */
 431 static struct nf_conn *
 432 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 433                      u8 l3num, struct sk_buff *skb)
 434 {
 435         struct nf_conntrack_l3proto *l3proto;
 436         struct nf_conntrack_l4proto *l4proto;
 437         struct nf_conntrack_tuple tuple;
 438         struct nf_conntrack_tuple_hash *h;
 439         struct nf_conn *ct;
 440         unsigned int dataoff;
 441         u8 protonum;
 442
 443         l3proto = __nf_ct_l3proto_find(l3num);
 444         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 445                                  &protonum) <= 0) {
 446                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 447                 return NULL;
 448         }
 449         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 450         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 451                              protonum, net, &tuple, l3proto, l4proto)) {
 452                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 453                 return NULL;
 454         }
 455
 456         /* look for tuple match */
 457         h = nf_conntrack_find_get(net, zone, &tuple);
 458         if (!h)
 459                 return NULL;   /* Not found. */
 460
 461         ct = nf_ct_tuplehash_to_ctrack(h);
 462
 463         skb->nfct = &ct->ct_general;
 464         skb->nfctinfo = ovs_ct_get_info(h);
 465         return ct;
 466 }
 467
 468 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 469 static bool skb_nfct_cached(struct net *net,
 470                             const struct sw_flow_key *key,
 471                             const struct ovs_conntrack_info *info,
 472                             struct sk_buff *skb)
 473 {
 474         enum ip_conntrack_info ctinfo;
 475         struct nf_conn *ct;
 476
 477         ct = nf_ct_get(skb, &ctinfo);
 478         /* If no ct, check if we have evidence that an existing conntrack entry
 479          * might be found for this skb.  This happens when we lose a skb->nfct
 480          * due to an upcall.  If the connection was not confirmed, it is not
 481          * cached and needs to be run through conntrack again.
 482          */
 483         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 484             !(key->ct.state & OVS_CS_F_INVALID) &&
 485             key->ct.zone == info->zone.id)
 486                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 487         if (!ct)
 488                 return false;
 489         if (!net_eq(net, read_pnet(&ct->ct_net)))
 490                 return false;
 491         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 492                 return false;
 493         if (info->helper) {
 494                 struct nf_conn_help *help;
 495
 496                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 497                 if (help && rcu_access_pointer(help->helper) != info->helper)
 498                         return false;
 499         }
 500
 501         return true;
 502 }
 503
 504 #ifdef CONFIG_NF_NAT_NEEDED
 505 /* Modelled after nf_nat_ipv[46]_fn().
 506  * range is only used for new, uninitialized NAT state.
 507  * Returns either NF_ACCEPT or NF_DROP.
 508  */
 509 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 510                               enum ip_conntrack_info ctinfo,
 511                               const struct nf_nat_range *range,
 512                               enum nf_nat_manip_type maniptype)
 513 {
 514         int hooknum, nh_off, err = NF_ACCEPT;
 515
 516         nh_off = skb_network_offset(skb);
 517         skb_pull_rcsum(skb, nh_off);
 518
 519         /* See HOOK2MANIP(). */
 520         if (maniptype == NF_NAT_MANIP_SRC)
 521                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 522         else
 523                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 524
 525         switch (ctinfo) {
 526         case IP_CT_RELATED:
 527         case IP_CT_RELATED_REPLY:
 528                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 529                     skb->protocol == htons(ETH_P_IP) &&
 530                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 531                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 532                                                            hooknum))
 533                                 err = NF_DROP;
 534                         goto push;
 535                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 536                            skb->protocol == htons(ETH_P_IPV6)) {
 537                         __be16 frag_off;
 538                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 539                         int hdrlen = ipv6_skip_exthdr(skb,
 540                                                       sizeof(struct ipv6hdr),
 541                                                       &nexthdr, &frag_off);
 542
 543                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 544                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 545                                                                      ctinfo,
 546                                                                      hooknum,
 547                                                                      hdrlen))
 548                                         err = NF_DROP;
 549                                 goto push;
 550                         }
 551                 }
 552                 /* Non-ICMP, fall thru to initialize if needed. */
 553         case IP_CT_NEW:
 554                 /* Seen it before?  This can happen for loopback, retrans,
 555                  * or local packets.
 556                  */
 557                 if (!nf_nat_initialized(ct, maniptype)) {
 558                         /* Initialize according to the NAT action. */
 559                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 560                                 /* Action is set up to establish a new
 561                                  * mapping.
 562                                  */
 563                                 ? nf_nat_setup_info(ct, range, maniptype)
 564                                 : nf_nat_alloc_null_binding(ct, hooknum);
 565                         if (err != NF_ACCEPT)
 566                                 goto push;
 567                 }
 568                 break;
 569
 570         case IP_CT_ESTABLISHED:
 571         case IP_CT_ESTABLISHED_REPLY:
 572                 break;
 573
 574         default:
 575                 err = NF_DROP;
 576                 goto push;
 577         }
 578
 579         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 580 push:
 581         skb_push(skb, nh_off);
 582         skb_postpush_rcsum(skb, skb->data, nh_off);
 583
 584         return err;
 585 }
 586
 587 static void ovs_nat_update_key(struct sw_flow_key *key,
 588                                const struct sk_buff *skb,
 589                                enum nf_nat_manip_type maniptype)
 590 {
 591         if (maniptype == NF_NAT_MANIP_SRC) {
 592                 __be16 src;
 593
 594                 key->ct.state |= OVS_CS_F_SRC_NAT;
 595                 if (key->eth.type == htons(ETH_P_IP))
 596                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 597                 else if (key->eth.type == htons(ETH_P_IPV6))
 598                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 599                                sizeof(key->ipv6.addr.src));
 600                 else
 601                         return;
 602
 603                 if (key->ip.proto == IPPROTO_UDP)
 604                         src = udp_hdr(skb)->source;
 605                 else if (key->ip.proto == IPPROTO_TCP)
 606                         src = tcp_hdr(skb)->source;
 607                 else if (key->ip.proto == IPPROTO_SCTP)
 608                         src = sctp_hdr(skb)->source;
 609                 else
 610                         return;
 611
 612                 key->tp.src = src;
 613         } else {
 614                 __be16 dst;
 615
 616                 key->ct.state |= OVS_CS_F_DST_NAT;
 617                 if (key->eth.type == htons(ETH_P_IP))
 618                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 619                 else if (key->eth.type == htons(ETH_P_IPV6))
 620                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 621                                sizeof(key->ipv6.addr.dst));
 622                 else
 623                         return;
 624
 625                 if (key->ip.proto == IPPROTO_UDP)
 626                         dst = udp_hdr(skb)->dest;
 627                 else if (key->ip.proto == IPPROTO_TCP)
 628                         dst = tcp_hdr(skb)->dest;
 629                 else if (key->ip.proto == IPPROTO_SCTP)
 630                         dst = sctp_hdr(skb)->dest;
 631                 else
 632                         return;
 633
 634                 key->tp.dst = dst;
 635         }
 636 }
 637
 638 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 639 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 640                       const struct ovs_conntrack_info *info,
 641                       struct sk_buff *skb, struct nf_conn *ct,
 642                       enum ip_conntrack_info ctinfo)
 643 {
 644         enum nf_nat_manip_type maniptype;
 645         int err;
 646
 647         if (nf_ct_is_untracked(ct)) {
 648                 /* A NAT action may only be performed on tracked packets. */
 649                 return NF_ACCEPT;
 650         }
 651
 652         /* Add NAT extension if not confirmed yet. */
 653         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 654                 return NF_ACCEPT;   /* Can't NAT. */
 655
 656         /* Determine NAT type.
 657          * Check if the NAT type can be deduced from the tracked connection.
 658          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 659          * when committing.
 660          */
 661         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 662             ct->status & IPS_NAT_MASK &&
 663             (ctinfo != IP_CT_RELATED || info->commit)) {
 664                 /* NAT an established or related connection like before. */
 665                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 666                         /* This is the REPLY direction for a connection
 667                          * for which NAT was applied in the forward
 668                          * direction.  Do the reverse NAT.
 669                          */
 670                         maniptype = ct->status & IPS_SRC_NAT
 671                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 672                 else
 673                         maniptype = ct->status & IPS_SRC_NAT
 674                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 675         } else if (info->nat & OVS_CT_SRC_NAT) {
 676                 maniptype = NF_NAT_MANIP_SRC;
 677         } else if (info->nat & OVS_CT_DST_NAT) {
 678                 maniptype = NF_NAT_MANIP_DST;
 679         } else {
 680                 return NF_ACCEPT; /* Connection is not NATed. */
 681         }
 682         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 683
 684         /* Mark NAT done if successful and update the flow key. */
 685         if (err == NF_ACCEPT)
 686                 ovs_nat_update_key(key, skb, maniptype);
 687
 688         return err;
 689 }
 690 #else /* !CONFIG_NF_NAT_NEEDED */
 691 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 692                       const struct ovs_conntrack_info *info,
 693                       struct sk_buff *skb, struct nf_conn *ct,
 694                       enum ip_conntrack_info ctinfo)
 695 {
 696         return NF_ACCEPT;
 697 }
 698 #endif
 699
 700 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 701  * not done already.  Update key with new CT state after passing the packet
 702  * through conntrack.
 703  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 704  * set to NULL and 0 will be returned.
 705  */
 706 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 707                            const struct ovs_conntrack_info *info,
 708                            struct sk_buff *skb)
 709 {
 710         /* If we are recirculating packets to match on conntrack fields and
 711          * committing with a separate conntrack action,  then we don't need to
 712          * actually run the packet through conntrack twice unless it's for a
 713          * different zone.
 714          */
 715         bool cached = skb_nfct_cached(net, key, info, skb);
 716         enum ip_conntrack_info ctinfo;
 717         struct nf_conn *ct;
 718
 719         if (!cached) {
 720                 struct nf_conn *tmpl = info->ct;
 721                 int err;
 722
 723                 /* Associate skb with specified zone. */
 724                 if (tmpl) {
 725                         if (skb->nfct)
 726                                 nf_conntrack_put(skb->nfct);
 727                         nf_conntrack_get(&tmpl->ct_general);
 728                         skb->nfct = &tmpl->ct_general;
 729                         skb->nfctinfo = IP_CT_NEW;
 730                 }
 731
 732                 err = nf_conntrack_in(net, info->family,
 733                                       NF_INET_PRE_ROUTING, skb);
 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_rcsum(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         skb_postpush_rcsum(skb, skb->data, nh_ofs);
 905         if (err)
 906                 kfree_skb(skb);
 907         return err;
 908 }
 909
 910 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 911                              const struct sw_flow_key *key, bool log)
 912 {
 913         struct nf_conntrack_helper *helper;
 914         struct nf_conn_help *help;
 915
 916         helper = nf_conntrack_helper_try_module_get(name, info->family,
 917                                                     key->ip.proto);
 918         if (!helper) {
 919                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 920                 return -EINVAL;
 921         }
 922
 923         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 924         if (!help) {
 925                 module_put(helper->me);
 926                 return -ENOMEM;
 927         }
 928
 929         rcu_assign_pointer(help->helper, helper);
 930         info->helper = helper;
 931         return 0;
 932 }
 933
 934 #ifdef CONFIG_NF_NAT_NEEDED
 935 static int parse_nat(const struct nlattr *attr,
 936                      struct ovs_conntrack_info *info, bool log)
 937 {
 938         struct nlattr *a;
 939         int rem;
 940         bool have_ip_max = false;
 941         bool have_proto_max = false;
 942         bool ip_vers = (info->family == NFPROTO_IPV6);
 943
 944         nla_for_each_nested(a, attr, rem) {
 945                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
 946                         [OVS_NAT_ATTR_SRC] = {0, 0},
 947                         [OVS_NAT_ATTR_DST] = {0, 0},
 948                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
 949                                                  sizeof(struct in6_addr)},
 950                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
 951                                                  sizeof(struct in6_addr)},
 952                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
 953                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
 954                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
 955                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
 956                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
 957                 };
 958                 int type = nla_type(a);
 959
 960                 if (type > OVS_NAT_ATTR_MAX) {
 961                         OVS_NLERR(log,
 962                                   "Unknown NAT attribute (type=%d, max=%d).\n",
 963                                   type, OVS_NAT_ATTR_MAX);
 964                         return -EINVAL;
 965                 }
 966
 967                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
 968                         OVS_NLERR(log,
 969                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
 970                                   type, nla_len(a),
 971                                   ovs_nat_attr_lens[type][ip_vers]);
 972                         return -EINVAL;
 973                 }
 974
 975                 switch (type) {
 976                 case OVS_NAT_ATTR_SRC:
 977                 case OVS_NAT_ATTR_DST:
 978                         if (info->nat) {
 979                                 OVS_NLERR(log,
 980                                           "Only one type of NAT may be specified.\n"
 981                                           );
 982                                 return -ERANGE;
 983                         }
 984                         info->nat |= OVS_CT_NAT;
 985                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
 986                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
 987                         break;
 988
 989                 case OVS_NAT_ATTR_IP_MIN:
 990                         nla_memcpy(&info->range.min_addr, a,
 991                                    sizeof(info->range.min_addr));
 992                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
 993                         break;
 994
 995                 case OVS_NAT_ATTR_IP_MAX:
 996                         have_ip_max = true;
 997                         nla_memcpy(&info->range.max_addr, a,
 998                                    sizeof(info->range.max_addr));
 999                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1000                         break;
1001
1002                 case OVS_NAT_ATTR_PROTO_MIN:
1003                         info->range.min_proto.all = htons(nla_get_u16(a));
1004                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1005                         break;
1006
1007                 case OVS_NAT_ATTR_PROTO_MAX:
1008                         have_proto_max = true;
1009                         info->range.max_proto.all = htons(nla_get_u16(a));
1010                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1011                         break;
1012
1013                 case OVS_NAT_ATTR_PERSISTENT:
1014                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1015                         break;
1016
1017                 case OVS_NAT_ATTR_PROTO_HASH:
1018                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1019                         break;
1020
1021                 case OVS_NAT_ATTR_PROTO_RANDOM:
1022                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1023                         break;
1024
1025                 default:
1026                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1027                         return -EINVAL;
1028                 }
1029         }
1030
1031         if (rem > 0) {
1032                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1033                 return -EINVAL;
1034         }
1035         if (!info->nat) {
1036                 /* Do not allow flags if no type is given. */
1037                 if (info->range.flags) {
1038                         OVS_NLERR(log,
1039                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1040                                   );
1041                         return -EINVAL;
1042                 }
1043                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1044         } else if (!info->commit) {
1045                 OVS_NLERR(log,
1046                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1047                           );
1048                 return -EINVAL;
1049         }
1050         /* Allow missing IP_MAX. */
1051         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1052                 memcpy(&info->range.max_addr, &info->range.min_addr,
1053                        sizeof(info->range.max_addr));
1054         }
1055         /* Allow missing PROTO_MAX. */
1056         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1057             !have_proto_max) {
1058                 info->range.max_proto.all = info->range.min_proto.all;
1059         }
1060         return 0;
1061 }
1062 #endif
1063
1064 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1065         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1066         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1067                                     .maxlen = sizeof(u16) },
1068         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1069                                     .maxlen = sizeof(struct md_mark) },
1070         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1071                                     .maxlen = sizeof(struct md_labels) },
1072         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1073                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1074 #ifdef CONFIG_NF_NAT_NEEDED
1075         /* NAT length is checked when parsing the nested attributes. */
1076         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1077 #endif
1078 };
1079
1080 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1081                     const char **helper, bool log)
1082 {
1083         struct nlattr *a;
1084         int rem;
1085
1086         nla_for_each_nested(a, attr, rem) {
1087                 int type = nla_type(a);
1088                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1089                 int minlen = ovs_ct_attr_lens[type].minlen;
1090
1091                 if (type > OVS_CT_ATTR_MAX) {
1092                         OVS_NLERR(log,
1093                                   "Unknown conntrack attr (type=%d, max=%d)",
1094                                   type, OVS_CT_ATTR_MAX);
1095                         return -EINVAL;
1096                 }
1097                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1098                         OVS_NLERR(log,
1099                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1100                                   type, nla_len(a), maxlen);
1101                         return -EINVAL;
1102                 }
1103
1104                 switch (type) {
1105                 case OVS_CT_ATTR_COMMIT:
1106                         info->commit = true;
1107                         break;
1108 #ifdef CONFIG_NF_CONNTRACK_ZONES
1109                 case OVS_CT_ATTR_ZONE:
1110                         info->zone.id = nla_get_u16(a);
1111                         break;
1112 #endif
1113 #ifdef CONFIG_NF_CONNTRACK_MARK
1114                 case OVS_CT_ATTR_MARK: {
1115                         struct md_mark *mark = nla_data(a);
1116
1117                         if (!mark->mask) {
1118                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1119                                 return -EINVAL;
1120                         }
1121                         info->mark = *mark;
1122                         break;
1123                 }
1124 #endif
1125 #ifdef CONFIG_NF_CONNTRACK_LABELS
1126                 case OVS_CT_ATTR_LABELS: {
1127                         struct md_labels *labels = nla_data(a);
1128
1129                         if (!labels_nonzero(&labels->mask)) {
1130                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1131                                 return -EINVAL;
1132                         }
1133                         info->labels = *labels;
1134                         break;
1135                 }
1136 #endif
1137                 case OVS_CT_ATTR_HELPER:
1138                         *helper = nla_data(a);
1139                         if (!memchr(*helper, '\0', nla_len(a))) {
1140                                 OVS_NLERR(log, "Invalid conntrack helper");
1141                                 return -EINVAL;
1142                         }
1143                         break;
1144 #ifdef CONFIG_NF_NAT_NEEDED
1145                 case OVS_CT_ATTR_NAT: {
1146                         int err = parse_nat(a, info, log);
1147
1148                         if (err)
1149                                 return err;
1150                         break;
1151                 }
1152 #endif
1153                 default:
1154                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1155                                   type);
1156                         return -EINVAL;
1157                 }
1158         }
1159
1160 #ifdef CONFIG_NF_CONNTRACK_MARK
1161         if (!info->commit && info->mark.mask) {
1162                 OVS_NLERR(log,
1163                           "Setting conntrack mark requires 'commit' flag.");
1164                 return -EINVAL;
1165         }
1166 #endif
1167 #ifdef CONFIG_NF_CONNTRACK_LABELS
1168         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1169                 OVS_NLERR(log,
1170                           "Setting conntrack labels requires 'commit' flag.");
1171                 return -EINVAL;
1172         }
1173 #endif
1174         if (rem > 0) {
1175                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1176                 return -EINVAL;
1177         }
1178
1179         return 0;
1180 }
1181
1182 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1183 {
1184         if (attr == OVS_KEY_ATTR_CT_STATE)
1185                 return true;
1186         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1187             attr == OVS_KEY_ATTR_CT_ZONE)
1188                 return true;
1189         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1190             attr == OVS_KEY_ATTR_CT_MARK)
1191                 return true;
1192         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1193             attr == OVS_KEY_ATTR_CT_LABELS) {
1194                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1195
1196                 return ovs_net->xt_label;
1197         }
1198
1199         return false;
1200 }
1201
1202 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1203                        const struct sw_flow_key *key,
1204                        struct sw_flow_actions **sfa,  bool log)
1205 {
1206         struct ovs_conntrack_info ct_info;
1207         const char *helper = NULL;
1208         u16 family;
1209         int err;
1210
1211         family = key_to_nfproto(key);
1212         if (family == NFPROTO_UNSPEC) {
1213                 OVS_NLERR(log, "ct family unspecified");
1214                 return -EINVAL;
1215         }
1216
1217         memset(&ct_info, 0, sizeof(ct_info));
1218         ct_info.family = family;
1219
1220         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1221                         NF_CT_DEFAULT_ZONE_DIR, 0);
1222
1223         err = parse_ct(attr, &ct_info, &helper, log);
1224         if (err)
1225                 return err;
1226
1227         /* Set up template for tracking connections in specific zones. */
1228         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1229         if (!ct_info.ct) {
1230                 OVS_NLERR(log, "Failed to allocate conntrack template");
1231                 return -ENOMEM;
1232         }
1233
1234         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1235         nf_conntrack_get(&ct_info.ct->ct_general);
1236
1237         if (helper) {
1238                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1239                 if (err)
1240                         goto err_free_ct;
1241         }
1242
1243         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1244                                  sizeof(ct_info), log);
1245         if (err)
1246                 goto err_free_ct;
1247
1248         return 0;
1249 err_free_ct:
1250         __ovs_ct_free_action(&ct_info);
1251         return err;
1252 }
1253
1254 #ifdef CONFIG_NF_NAT_NEEDED
1255 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1256                                struct sk_buff *skb)
1257 {
1258         struct nlattr *start;
1259
1260         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1261         if (!start)
1262                 return false;
1263
1264         if (info->nat & OVS_CT_SRC_NAT) {
1265                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1266                         return false;
1267         } else if (info->nat & OVS_CT_DST_NAT) {
1268                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1269                         return false;
1270         } else {
1271                 goto out;
1272         }
1273
1274         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1275                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1276                     info->family == NFPROTO_IPV4) {
1277                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1278                                             info->range.min_addr.ip) ||
1279                             (info->range.max_addr.ip
1280                              != info->range.min_addr.ip &&
1281                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1282                                               info->range.max_addr.ip))))
1283                                 return false;
1284                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1285                            info->family == NFPROTO_IPV6) {
1286                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1287                                              &info->range.min_addr.in6) ||
1288                             (memcmp(&info->range.max_addr.in6,
1289                                     &info->range.min_addr.in6,
1290                                     sizeof(info->range.max_addr.in6)) &&
1291                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1292                                                &info->range.max_addr.in6))))
1293                                 return false;
1294                 } else {
1295                         return false;
1296                 }
1297         }
1298         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1299             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1300                          ntohs(info->range.min_proto.all)) ||
1301              (info->range.max_proto.all != info->range.min_proto.all &&
1302               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1303                           ntohs(info->range.max_proto.all)))))
1304                 return false;
1305
1306         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1307             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1308                 return false;
1309         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1310             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1311                 return false;
1312         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1313             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1314                 return false;
1315 out:
1316         nla_nest_end(skb, start);
1317
1318         return true;
1319 }
1320 #endif
1321
1322 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1323                           struct sk_buff *skb)
1324 {
1325         struct nlattr *start;
1326
1327         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1328         if (!start)
1329                 return -EMSGSIZE;
1330
1331         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1332                 return -EMSGSIZE;
1333         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1334             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1335                 return -EMSGSIZE;
1336         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1337             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1338                     &ct_info->mark))
1339                 return -EMSGSIZE;
1340         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1341             labels_nonzero(&ct_info->labels.mask) &&
1342             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1343                     &ct_info->labels))
1344                 return -EMSGSIZE;
1345         if (ct_info->helper) {
1346                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1347                                    ct_info->helper->name))
1348                         return -EMSGSIZE;
1349         }
1350 #ifdef CONFIG_NF_NAT_NEEDED
1351         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1352                 return -EMSGSIZE;
1353 #endif
1354         nla_nest_end(skb, start);
1355
1356         return 0;
1357 }
1358
1359 void ovs_ct_free_action(const struct nlattr *a)
1360 {
1361         struct ovs_conntrack_info *ct_info = nla_data(a);
1362
1363         __ovs_ct_free_action(ct_info);
1364 }
1365
1366 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1367 {
1368         if (ct_info->helper)
1369                 module_put(ct_info->helper->me);
1370         if (ct_info->ct)
1371                 nf_ct_tmpl_free(ct_info->ct);
1372 }
1373
1374 void ovs_ct_init(struct net *net)
1375 {
1376         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1377         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1378
1379         if (nf_connlabels_get(net, n_bits - 1)) {
1380                 ovs_net->xt_label = false;
1381                 OVS_NLERR(true, "Failed to set connlabel length");
1382         } else {
1383                 ovs_net->xt_label = true;
1384         }
1385 }
1386
1387 void ovs_ct_exit(struct net *net)
1388 {
1389         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1390
1391         if (ovs_net->xt_label)
1392                 nf_connlabels_put(net);
1393 }