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 = cl->words * sizeof(long);
 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 || cl->words * sizeof(long) < 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                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 371                 err = nf_ct_frag6_gather(net, skb, user);
 372                 if (err)
 373                         return err;
 374
 375                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 376                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 377 #endif
 378         } else {
 379                 kfree_skb(skb);
 380                 return -EPFNOSUPPORT;
 381         }
 382
 383         key->ip.frag = OVS_FRAG_TYPE_NONE;
 384         skb_clear_hash(skb);
 385         skb->ignore_df = 1;
 386         *OVS_CB(skb) = ovs_cb;
 387
 388         return 0;
 389 }
 390
 391 static struct nf_conntrack_expect *
 392 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 393                    u16 proto, const struct sk_buff *skb)
 394 {
 395         struct nf_conntrack_tuple tuple;
 396
 397         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 398                 return NULL;
 399         return __nf_ct_expect_find(net, zone, &tuple);
 400 }
 401
 402 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 403 static enum ip_conntrack_info
 404 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 405 {
 406         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 407
 408         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 409                 return IP_CT_ESTABLISHED_REPLY;
 410         /* Once we've had two way comms, always ESTABLISHED. */
 411         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 412                 return IP_CT_ESTABLISHED;
 413         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 414                 return IP_CT_RELATED;
 415         return IP_CT_NEW;
 416 }
 417
 418 /* Find an existing connection which this packet belongs to without
 419  * re-attributing statistics or modifying the connection state.  This allows an
 420  * skb->nfct lost due to an upcall to be recovered during actions execution.
 421  *
 422  * Must be called with rcu_read_lock.
 423  *
 424  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 425  * connection.  Returns NULL if there is no existing entry.
 426  */
 427 static struct nf_conn *
 428 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 429                      u8 l3num, struct sk_buff *skb)
 430 {
 431         struct nf_conntrack_l3proto *l3proto;
 432         struct nf_conntrack_l4proto *l4proto;
 433         struct nf_conntrack_tuple tuple;
 434         struct nf_conntrack_tuple_hash *h;
 435         enum ip_conntrack_info ctinfo;
 436         struct nf_conn *ct;
 437         unsigned int dataoff;
 438         u8 protonum;
 439
 440         l3proto = __nf_ct_l3proto_find(l3num);
 441         if (!l3proto) {
 442                 pr_debug("ovs_ct_find_existing: Can't get l3proto\n");
 443                 return NULL;
 444         }
 445         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 446                                  &protonum) <= 0) {
 447                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 448                 return NULL;
 449         }
 450         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 451         if (!l4proto) {
 452                 pr_debug("ovs_ct_find_existing: Can't get l4proto\n");
 453                 return NULL;
 454         }
 455         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 456                              protonum, net, &tuple, l3proto, l4proto)) {
 457                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 458                 return NULL;
 459         }
 460
 461         /* look for tuple match */
 462         h = nf_conntrack_find_get(net, zone, &tuple);
 463         if (!h)
 464                 return NULL;   /* Not found. */
 465
 466         ct = nf_ct_tuplehash_to_ctrack(h);
 467
 468         ctinfo = ovs_ct_get_info(h);
 469         if (ctinfo == IP_CT_NEW) {
 470                 /* This should not happen. */
 471                 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct);
 472         }
 473         skb->nfct = &ct->ct_general;
 474         skb->nfctinfo = ctinfo;
 475         return ct;
 476 }
 477
 478 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 479 static bool skb_nfct_cached(struct net *net,
 480                             const struct sw_flow_key *key,
 481                             const struct ovs_conntrack_info *info,
 482                             struct sk_buff *skb)
 483 {
 484         enum ip_conntrack_info ctinfo;
 485         struct nf_conn *ct;
 486
 487         ct = nf_ct_get(skb, &ctinfo);
 488         /* If no ct, check if we have evidence that an existing conntrack entry
 489          * might be found for this skb.  This happens when we lose a skb->nfct
 490          * due to an upcall.  If the connection was not confirmed, it is not
 491          * cached and needs to be run through conntrack again.
 492          */
 493         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 494             !(key->ct.state & OVS_CS_F_INVALID) &&
 495             key->ct.zone == info->zone.id)
 496                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 497         if (!ct)
 498                 return false;
 499         if (!net_eq(net, read_pnet(&ct->ct_net)))
 500                 return false;
 501         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 502                 return false;
 503         if (info->helper) {
 504                 struct nf_conn_help *help;
 505
 506                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 507                 if (help && rcu_access_pointer(help->helper) != info->helper)
 508                         return false;
 509         }
 510
 511         return true;
 512 }
 513
 514 #ifdef CONFIG_NF_NAT_NEEDED
 515 /* Modelled after nf_nat_ipv[46]_fn().
 516  * range is only used for new, uninitialized NAT state.
 517  * Returns either NF_ACCEPT or NF_DROP.
 518  */
 519 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 520                               enum ip_conntrack_info ctinfo,
 521                               const struct nf_nat_range *range,
 522                               enum nf_nat_manip_type maniptype)
 523 {
 524         int hooknum, nh_off, err = NF_ACCEPT;
 525
 526         nh_off = skb_network_offset(skb);
 527         skb_pull(skb, nh_off);
 528
 529         /* See HOOK2MANIP(). */
 530         if (maniptype == NF_NAT_MANIP_SRC)
 531                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 532         else
 533                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 534
 535         switch (ctinfo) {
 536         case IP_CT_RELATED:
 537         case IP_CT_RELATED_REPLY:
 538                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 539                     skb->protocol == htons(ETH_P_IP) &&
 540                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 541                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 542                                                            hooknum))
 543                                 err = NF_DROP;
 544                         goto push;
 545                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 546                            skb->protocol == htons(ETH_P_IPV6)) {
 547                         __be16 frag_off;
 548                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 549                         int hdrlen = ipv6_skip_exthdr(skb,
 550                                                       sizeof(struct ipv6hdr),
 551                                                       &nexthdr, &frag_off);
 552
 553                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 554                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 555                                                                      ctinfo,
 556                                                                      hooknum,
 557                                                                      hdrlen))
 558                                         err = NF_DROP;
 559                                 goto push;
 560                         }
 561                 }
 562                 /* Non-ICMP, fall thru to initialize if needed. */
 563         case IP_CT_NEW:
 564                 /* Seen it before?  This can happen for loopback, retrans,
 565                  * or local packets.
 566                  */
 567                 if (!nf_nat_initialized(ct, maniptype)) {
 568                         /* Initialize according to the NAT action. */
 569                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 570                                 /* Action is set up to establish a new
 571                                  * mapping.
 572                                  */
 573                                 ? nf_nat_setup_info(ct, range, maniptype)
 574                                 : nf_nat_alloc_null_binding(ct, hooknum);
 575                         if (err != NF_ACCEPT)
 576                                 goto push;
 577                 }
 578                 break;
 579
 580         case IP_CT_ESTABLISHED:
 581         case IP_CT_ESTABLISHED_REPLY:
 582                 break;
 583
 584         default:
 585                 err = NF_DROP;
 586                 goto push;
 587         }
 588
 589         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 590 push:
 591         skb_push(skb, nh_off);
 592
 593         return err;
 594 }
 595
 596 static void ovs_nat_update_key(struct sw_flow_key *key,
 597                                const struct sk_buff *skb,
 598                                enum nf_nat_manip_type maniptype)
 599 {
 600         if (maniptype == NF_NAT_MANIP_SRC) {
 601                 __be16 src;
 602
 603                 key->ct.state |= OVS_CS_F_SRC_NAT;
 604                 if (key->eth.type == htons(ETH_P_IP))
 605                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 606                 else if (key->eth.type == htons(ETH_P_IPV6))
 607                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 608                                sizeof(key->ipv6.addr.src));
 609                 else
 610                         return;
 611
 612                 if (key->ip.proto == IPPROTO_UDP)
 613                         src = udp_hdr(skb)->source;
 614                 else if (key->ip.proto == IPPROTO_TCP)
 615                         src = tcp_hdr(skb)->source;
 616                 else if (key->ip.proto == IPPROTO_SCTP)
 617                         src = sctp_hdr(skb)->source;
 618                 else
 619                         return;
 620
 621                 key->tp.src = src;
 622         } else {
 623                 __be16 dst;
 624
 625                 key->ct.state |= OVS_CS_F_DST_NAT;
 626                 if (key->eth.type == htons(ETH_P_IP))
 627                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 628                 else if (key->eth.type == htons(ETH_P_IPV6))
 629                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 630                                sizeof(key->ipv6.addr.dst));
 631                 else
 632                         return;
 633
 634                 if (key->ip.proto == IPPROTO_UDP)
 635                         dst = udp_hdr(skb)->dest;
 636                 else if (key->ip.proto == IPPROTO_TCP)
 637                         dst = tcp_hdr(skb)->dest;
 638                 else if (key->ip.proto == IPPROTO_SCTP)
 639                         dst = sctp_hdr(skb)->dest;
 640                 else
 641                         return;
 642
 643                 key->tp.dst = dst;
 644         }
 645 }
 646
 647 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 648 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 649                       const struct ovs_conntrack_info *info,
 650                       struct sk_buff *skb, struct nf_conn *ct,
 651                       enum ip_conntrack_info ctinfo)
 652 {
 653         enum nf_nat_manip_type maniptype;
 654         int err;
 655
 656         if (nf_ct_is_untracked(ct)) {
 657                 /* A NAT action may only be performed on tracked packets. */
 658                 return NF_ACCEPT;
 659         }
 660
 661         /* Add NAT extension if not confirmed yet. */
 662         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 663                 return NF_ACCEPT;   /* Can't NAT. */
 664
 665         /* Determine NAT type.
 666          * Check if the NAT type can be deduced from the tracked connection.
 667          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 668          * when committing.
 669          */
 670         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 671             ct->status & IPS_NAT_MASK &&
 672             (ctinfo != IP_CT_RELATED || info->commit)) {
 673                 /* NAT an established or related connection like before. */
 674                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 675                         /* This is the REPLY direction for a connection
 676                          * for which NAT was applied in the forward
 677                          * direction.  Do the reverse NAT.
 678                          */
 679                         maniptype = ct->status & IPS_SRC_NAT
 680                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 681                 else
 682                         maniptype = ct->status & IPS_SRC_NAT
 683                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 684         } else if (info->nat & OVS_CT_SRC_NAT) {
 685                 maniptype = NF_NAT_MANIP_SRC;
 686         } else if (info->nat & OVS_CT_DST_NAT) {
 687                 maniptype = NF_NAT_MANIP_DST;
 688         } else {
 689                 return NF_ACCEPT; /* Connection is not NATed. */
 690         }
 691         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 692
 693         /* Mark NAT done if successful and update the flow key. */
 694         if (err == NF_ACCEPT)
 695                 ovs_nat_update_key(key, skb, maniptype);
 696
 697         return err;
 698 }
 699 #else /* !CONFIG_NF_NAT_NEEDED */
 700 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 701                       const struct ovs_conntrack_info *info,
 702                       struct sk_buff *skb, struct nf_conn *ct,
 703                       enum ip_conntrack_info ctinfo)
 704 {
 705         return NF_ACCEPT;
 706 }
 707 #endif
 708
 709 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 710  * not done already.  Update key with new CT state after passing the packet
 711  * through conntrack.
 712  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 713  * set to NULL and 0 will be returned.
 714  */
 715 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 716                            const struct ovs_conntrack_info *info,
 717                            struct sk_buff *skb)
 718 {
 719         /* If we are recirculating packets to match on conntrack fields and
 720          * committing with a separate conntrack action,  then we don't need to
 721          * actually run the packet through conntrack twice unless it's for a
 722          * different zone.
 723          */
 724         bool cached = skb_nfct_cached(net, key, info, skb);
 725         enum ip_conntrack_info ctinfo;
 726         struct nf_conn *ct;
 727
 728         if (!cached) {
 729                 struct nf_conn *tmpl = info->ct;
 730                 int err;
 731
 732                 /* Associate skb with specified zone. */
 733                 if (tmpl) {
 734                         if (skb->nfct)
 735                                 nf_conntrack_put(skb->nfct);
 736                         nf_conntrack_get(&tmpl->ct_general);
 737                         skb->nfct = &tmpl->ct_general;
 738                         skb->nfctinfo = IP_CT_NEW;
 739                 }
 740
 741                 /* Repeat if requested, see nf_iterate(). */
 742                 do {
 743                         err = nf_conntrack_in(net, info->family,
 744                                               NF_INET_PRE_ROUTING, skb);
 745                 } while (err == NF_REPEAT);
 746
 747                 if (err != NF_ACCEPT)
 748                         return -ENOENT;
 749
 750                 /* Clear CT state NAT flags to mark that we have not yet done
 751                  * NAT after the nf_conntrack_in() call.  We can actually clear
 752                  * the whole state, as it will be re-initialized below.
 753                  */
 754                 key->ct.state = 0;
 755
 756                 /* Update the key, but keep the NAT flags. */
 757                 ovs_ct_update_key(skb, info, key, true, true);
 758         }
 759
 760         ct = nf_ct_get(skb, &ctinfo);
 761         if (ct) {
 762                 /* Packets starting a new connection must be NATted before the
 763                  * helper, so that the helper knows about the NAT.  We enforce
 764                  * this by delaying both NAT and helper calls for unconfirmed
 765                  * connections until the committing CT action.  For later
 766                  * packets NAT and Helper may be called in either order.
 767                  *
 768                  * NAT will be done only if the CT action has NAT, and only
 769                  * once per packet (per zone), as guarded by the NAT bits in
 770                  * the key->ct.state.
 771                  */
 772                 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
 773                     (nf_ct_is_confirmed(ct) || info->commit) &&
 774                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 775                         return -EINVAL;
 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                 return __ovs_ct_lookup(net, key, info, skb);
 817
 818         return 0;
 819 }
 820
 821 /* Lookup connection and confirm if unconfirmed. */
 822 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 823                          const struct ovs_conntrack_info *info,
 824                          struct sk_buff *skb)
 825 {
 826         int err;
 827
 828         err = __ovs_ct_lookup(net, key, info, skb);
 829         if (err)
 830                 return err;
 831         /* This is a no-op if the connection has already been confirmed. */
 832         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 833                 return -EINVAL;
 834
 835         return 0;
 836 }
 837
 838 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 839 {
 840         size_t i;
 841
 842         for (i = 0; i < sizeof(*labels); i++)
 843                 if (labels->ct_labels[i])
 844                         return true;
 845
 846         return false;
 847 }
 848
 849 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 850  * value if 'skb' is freed.
 851  */
 852 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 853                    struct sw_flow_key *key,
 854                    const struct ovs_conntrack_info *info)
 855 {
 856         int nh_ofs;
 857         int err;
 858
 859         /* The conntrack module expects to be working at L3. */
 860         nh_ofs = skb_network_offset(skb);
 861         skb_pull(skb, nh_ofs);
 862
 863         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 864                 err = handle_fragments(net, key, info->zone.id, skb);
 865                 if (err)
 866                         return err;
 867         }
 868
 869         if (info->commit)
 870                 err = ovs_ct_commit(net, key, info, skb);
 871         else
 872                 err = ovs_ct_lookup(net, key, info, skb);
 873         if (err)
 874                 goto err;
 875
 876         if (info->mark.mask) {
 877                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 878                                       info->mark.mask);
 879                 if (err)
 880                         goto err;
 881         }
 882         if (labels_nonzero(&info->labels.mask))
 883                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 884                                         &info->labels.mask);
 885 err:
 886         skb_push(skb, nh_ofs);
 887         if (err)
 888                 kfree_skb(skb);
 889         return err;
 890 }
 891
 892 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 893                              const struct sw_flow_key *key, bool log)
 894 {
 895         struct nf_conntrack_helper *helper;
 896         struct nf_conn_help *help;
 897
 898         helper = nf_conntrack_helper_try_module_get(name, info->family,
 899                                                     key->ip.proto);
 900         if (!helper) {
 901                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 902                 return -EINVAL;
 903         }
 904
 905         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 906         if (!help) {
 907                 module_put(helper->me);
 908                 return -ENOMEM;
 909         }
 910
 911         rcu_assign_pointer(help->helper, helper);
 912         info->helper = helper;
 913         return 0;
 914 }
 915
 916 #ifdef CONFIG_NF_NAT_NEEDED
 917 static int parse_nat(const struct nlattr *attr,
 918                      struct ovs_conntrack_info *info, bool log)
 919 {
 920         struct nlattr *a;
 921         int rem;
 922         bool have_ip_max = false;
 923         bool have_proto_max = false;
 924         bool ip_vers = (info->family == NFPROTO_IPV6);
 925
 926         nla_for_each_nested(a, attr, rem) {
 927                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
 928                         [OVS_NAT_ATTR_SRC] = {0, 0},
 929                         [OVS_NAT_ATTR_DST] = {0, 0},
 930                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
 931                                                  sizeof(struct in6_addr)},
 932                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
 933                                                  sizeof(struct in6_addr)},
 934                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
 935                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
 936                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
 937                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
 938                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
 939                 };
 940                 int type = nla_type(a);
 941
 942                 if (type > OVS_NAT_ATTR_MAX) {
 943                         OVS_NLERR(log,
 944                                   "Unknown NAT attribute (type=%d, max=%d).\n",
 945                                   type, OVS_NAT_ATTR_MAX);
 946                         return -EINVAL;
 947                 }
 948
 949                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
 950                         OVS_NLERR(log,
 951                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
 952                                   type, nla_len(a),
 953                                   ovs_nat_attr_lens[type][ip_vers]);
 954                         return -EINVAL;
 955                 }
 956
 957                 switch (type) {
 958                 case OVS_NAT_ATTR_SRC:
 959                 case OVS_NAT_ATTR_DST:
 960                         if (info->nat) {
 961                                 OVS_NLERR(log,
 962                                           "Only one type of NAT may be specified.\n"
 963                                           );
 964                                 return -ERANGE;
 965                         }
 966                         info->nat |= OVS_CT_NAT;
 967                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
 968                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
 969                         break;
 970
 971                 case OVS_NAT_ATTR_IP_MIN:
 972                         nla_memcpy(&info->range.min_addr, a,
 973                                    sizeof(info->range.min_addr));
 974                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
 975                         break;
 976
 977                 case OVS_NAT_ATTR_IP_MAX:
 978                         have_ip_max = true;
 979                         nla_memcpy(&info->range.max_addr, a,
 980                                    sizeof(info->range.max_addr));
 981                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
 982                         break;
 983
 984                 case OVS_NAT_ATTR_PROTO_MIN:
 985                         info->range.min_proto.all = htons(nla_get_u16(a));
 986                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 987                         break;
 988
 989                 case OVS_NAT_ATTR_PROTO_MAX:
 990                         have_proto_max = true;
 991                         info->range.max_proto.all = htons(nla_get_u16(a));
 992                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 993                         break;
 994
 995                 case OVS_NAT_ATTR_PERSISTENT:
 996                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
 997                         break;
 998
 999                 case OVS_NAT_ATTR_PROTO_HASH:
1000                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1001                         break;
1002
1003                 case OVS_NAT_ATTR_PROTO_RANDOM:
1004                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1005                         break;
1006
1007                 default:
1008                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1009                         return -EINVAL;
1010                 }
1011         }
1012
1013         if (rem > 0) {
1014                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1015                 return -EINVAL;
1016         }
1017         if (!info->nat) {
1018                 /* Do not allow flags if no type is given. */
1019                 if (info->range.flags) {
1020                         OVS_NLERR(log,
1021                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1022                                   );
1023                         return -EINVAL;
1024                 }
1025                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1026         } else if (!info->commit) {
1027                 OVS_NLERR(log,
1028                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1029                           );
1030                 return -EINVAL;
1031         }
1032         /* Allow missing IP_MAX. */
1033         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1034                 memcpy(&info->range.max_addr, &info->range.min_addr,
1035                        sizeof(info->range.max_addr));
1036         }
1037         /* Allow missing PROTO_MAX. */
1038         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1039             !have_proto_max) {
1040                 info->range.max_proto.all = info->range.min_proto.all;
1041         }
1042         return 0;
1043 }
1044 #endif
1045
1046 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1047         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1048         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1049                                     .maxlen = sizeof(u16) },
1050         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1051                                     .maxlen = sizeof(struct md_mark) },
1052         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1053                                     .maxlen = sizeof(struct md_labels) },
1054         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1055                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1056 #ifdef CONFIG_NF_NAT_NEEDED
1057         /* NAT length is checked when parsing the nested attributes. */
1058         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1059 #endif
1060 };
1061
1062 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1063                     const char **helper, bool log)
1064 {
1065         struct nlattr *a;
1066         int rem;
1067
1068         nla_for_each_nested(a, attr, rem) {
1069                 int type = nla_type(a);
1070                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1071                 int minlen = ovs_ct_attr_lens[type].minlen;
1072
1073                 if (type > OVS_CT_ATTR_MAX) {
1074                         OVS_NLERR(log,
1075                                   "Unknown conntrack attr (type=%d, max=%d)",
1076                                   type, OVS_CT_ATTR_MAX);
1077                         return -EINVAL;
1078                 }
1079                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1080                         OVS_NLERR(log,
1081                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1082                                   type, nla_len(a), maxlen);
1083                         return -EINVAL;
1084                 }
1085
1086                 switch (type) {
1087                 case OVS_CT_ATTR_COMMIT:
1088                         info->commit = true;
1089                         break;
1090 #ifdef CONFIG_NF_CONNTRACK_ZONES
1091                 case OVS_CT_ATTR_ZONE:
1092                         info->zone.id = nla_get_u16(a);
1093                         break;
1094 #endif
1095 #ifdef CONFIG_NF_CONNTRACK_MARK
1096                 case OVS_CT_ATTR_MARK: {
1097                         struct md_mark *mark = nla_data(a);
1098
1099                         if (!mark->mask) {
1100                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1101                                 return -EINVAL;
1102                         }
1103                         info->mark = *mark;
1104                         break;
1105                 }
1106 #endif
1107 #ifdef CONFIG_NF_CONNTRACK_LABELS
1108                 case OVS_CT_ATTR_LABELS: {
1109                         struct md_labels *labels = nla_data(a);
1110
1111                         if (!labels_nonzero(&labels->mask)) {
1112                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1113                                 return -EINVAL;
1114                         }
1115                         info->labels = *labels;
1116                         break;
1117                 }
1118 #endif
1119                 case OVS_CT_ATTR_HELPER:
1120                         *helper = nla_data(a);
1121                         if (!memchr(*helper, '\0', nla_len(a))) {
1122                                 OVS_NLERR(log, "Invalid conntrack helper");
1123                                 return -EINVAL;
1124                         }
1125                         break;
1126 #ifdef CONFIG_NF_NAT_NEEDED
1127                 case OVS_CT_ATTR_NAT: {
1128                         int err = parse_nat(a, info, log);
1129
1130                         if (err)
1131                                 return err;
1132                         break;
1133                 }
1134 #endif
1135                 default:
1136                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1137                                   type);
1138                         return -EINVAL;
1139                 }
1140         }
1141
1142         if (rem > 0) {
1143                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1144                 return -EINVAL;
1145         }
1146
1147         return 0;
1148 }
1149
1150 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1151 {
1152         if (attr == OVS_KEY_ATTR_CT_STATE)
1153                 return true;
1154         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1155             attr == OVS_KEY_ATTR_CT_ZONE)
1156                 return true;
1157         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1158             attr == OVS_KEY_ATTR_CT_MARK)
1159                 return true;
1160         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1161             attr == OVS_KEY_ATTR_CT_LABELS) {
1162                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1163
1164                 return ovs_net->xt_label;
1165         }
1166
1167         return false;
1168 }
1169
1170 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1171                        const struct sw_flow_key *key,
1172                        struct sw_flow_actions **sfa,  bool log)
1173 {
1174         struct ovs_conntrack_info ct_info;
1175         const char *helper = NULL;
1176         u16 family;
1177         int err;
1178
1179         family = key_to_nfproto(key);
1180         if (family == NFPROTO_UNSPEC) {
1181                 OVS_NLERR(log, "ct family unspecified");
1182                 return -EINVAL;
1183         }
1184
1185         memset(&ct_info, 0, sizeof(ct_info));
1186         ct_info.family = family;
1187
1188         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1189                         NF_CT_DEFAULT_ZONE_DIR, 0);
1190
1191         err = parse_ct(attr, &ct_info, &helper, log);
1192         if (err)
1193                 return err;
1194
1195         /* Set up template for tracking connections in specific zones. */
1196         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1197         if (!ct_info.ct) {
1198                 OVS_NLERR(log, "Failed to allocate conntrack template");
1199                 return -ENOMEM;
1200         }
1201
1202         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1203         nf_conntrack_get(&ct_info.ct->ct_general);
1204
1205         if (helper) {
1206                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1207                 if (err)
1208                         goto err_free_ct;
1209         }
1210
1211         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1212                                  sizeof(ct_info), log);
1213         if (err)
1214                 goto err_free_ct;
1215
1216         return 0;
1217 err_free_ct:
1218         __ovs_ct_free_action(&ct_info);
1219         return err;
1220 }
1221
1222 #ifdef CONFIG_NF_NAT_NEEDED
1223 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1224                                struct sk_buff *skb)
1225 {
1226         struct nlattr *start;
1227
1228         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1229         if (!start)
1230                 return false;
1231
1232         if (info->nat & OVS_CT_SRC_NAT) {
1233                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1234                         return false;
1235         } else if (info->nat & OVS_CT_DST_NAT) {
1236                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1237                         return false;
1238         } else {
1239                 goto out;
1240         }
1241
1242         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1243                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1244                     info->family == NFPROTO_IPV4) {
1245                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1246                                             info->range.min_addr.ip) ||
1247                             (info->range.max_addr.ip
1248                              != info->range.min_addr.ip &&
1249                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1250                                               info->range.max_addr.ip))))
1251                                 return false;
1252                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1253                            info->family == NFPROTO_IPV6) {
1254                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1255                                              &info->range.min_addr.in6) ||
1256                             (memcmp(&info->range.max_addr.in6,
1257                                     &info->range.min_addr.in6,
1258                                     sizeof(info->range.max_addr.in6)) &&
1259                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1260                                                &info->range.max_addr.in6))))
1261                                 return false;
1262                 } else {
1263                         return false;
1264                 }
1265         }
1266         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1267             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1268                          ntohs(info->range.min_proto.all)) ||
1269              (info->range.max_proto.all != info->range.min_proto.all &&
1270               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1271                           ntohs(info->range.max_proto.all)))))
1272                 return false;
1273
1274         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1275             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1276                 return false;
1277         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1278             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1279                 return false;
1280         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1281             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1282                 return false;
1283 out:
1284         nla_nest_end(skb, start);
1285
1286         return true;
1287 }
1288 #endif
1289
1290 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1291                           struct sk_buff *skb)
1292 {
1293         struct nlattr *start;
1294
1295         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1296         if (!start)
1297                 return -EMSGSIZE;
1298
1299         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1300                 return -EMSGSIZE;
1301         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1302             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1303                 return -EMSGSIZE;
1304         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1305             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1306                     &ct_info->mark))
1307                 return -EMSGSIZE;
1308         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1309             labels_nonzero(&ct_info->labels.mask) &&
1310             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1311                     &ct_info->labels))
1312                 return -EMSGSIZE;
1313         if (ct_info->helper) {
1314                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1315                                    ct_info->helper->name))
1316                         return -EMSGSIZE;
1317         }
1318 #ifdef CONFIG_NF_NAT_NEEDED
1319         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1320                 return -EMSGSIZE;
1321 #endif
1322         nla_nest_end(skb, start);
1323
1324         return 0;
1325 }
1326
1327 void ovs_ct_free_action(const struct nlattr *a)
1328 {
1329         struct ovs_conntrack_info *ct_info = nla_data(a);
1330
1331         __ovs_ct_free_action(ct_info);
1332 }
1333
1334 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1335 {
1336         if (ct_info->helper)
1337                 module_put(ct_info->helper->me);
1338         if (ct_info->ct)
1339                 nf_ct_put(ct_info->ct);
1340 }
1341
1342 void ovs_ct_init(struct net *net)
1343 {
1344         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1345         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1346
1347         if (nf_connlabels_get(net, n_bits)) {
1348                 ovs_net->xt_label = false;
1349                 OVS_NLERR(true, "Failed to set connlabel length");
1350         } else {
1351                 ovs_net->xt_label = true;
1352         }
1353 }
1354
1355 void ovs_ct_exit(struct net *net)
1356 {
1357         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1358
1359         if (ovs_net->xt_label)
1360                 nf_connlabels_put(net);
1361 }