datapath/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/kconfig.h>
  15 #include <linux/version.h>
  16
  17 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  18
  19 #include <linux/module.h>
  20 #include <linux/openvswitch.h>
  21 #include <linux/tcp.h>
  22 #include <linux/udp.h>
  23 #include <linux/sctp.h>
  24 #include <net/ip.h>
  25 #include <net/netfilter/nf_conntrack_core.h>
  26 #include <net/netfilter/nf_conntrack_helper.h>
  27 #include <net/netfilter/nf_conntrack_labels.h>
  28 #include <net/netfilter/nf_conntrack_seqadj.h>
  29 #include <net/netfilter/nf_conntrack_zones.h>
  30 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  31
  32 #ifdef CONFIG_NF_NAT_NEEDED
  33 #include <linux/netfilter/nf_nat.h>
  34 #include <net/netfilter/nf_nat_core.h>
  35 #include <net/netfilter/nf_nat_l3proto.h>
  36 #endif
  37
  38 #include "datapath.h"
  39 #include "conntrack.h"
  40 #include "flow.h"
  41 #include "flow_netlink.h"
  42 #include "gso.h"
  43
  44 struct ovs_ct_len_tbl {
  45         int maxlen;
  46         int minlen;
  47 };
  48
  49 /* Metadata mark for masked write to conntrack mark */
  50 struct md_mark {
  51         u32 value;
  52         u32 mask;
  53 };
  54
  55 /* Metadata label for masked write to conntrack label. */
  56 struct md_labels {
  57         struct ovs_key_ct_labels value;
  58         struct ovs_key_ct_labels mask;
  59 };
  60
  61 enum ovs_ct_nat {
  62         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  63         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  64         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  65 };
  66
  67 /* Conntrack action context for execution. */
  68 struct ovs_conntrack_info {
  69         struct nf_conntrack_helper *helper;
  70         struct nf_conntrack_zone zone;
  71         struct nf_conn *ct;
  72         u8 commit : 1;
  73         u8 nat : 3;                 /* enum ovs_ct_nat */
  74         u8 random_fully_compat : 1; /* bool */
  75         u16 family;
  76         struct md_mark mark;
  77         struct md_labels labels;
  78 #ifdef CONFIG_NF_NAT_NEEDED
  79         struct nf_nat_range range;  /* Only present for SRC NAT and DST NAT. */
  80 #endif
  81 };
  82
  83 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
  84
  85 static u16 key_to_nfproto(const struct sw_flow_key *key)
  86 {
  87         switch (ntohs(key->eth.type)) {
  88         case ETH_P_IP:
  89                 return NFPROTO_IPV4;
  90         case ETH_P_IPV6:
  91                 return NFPROTO_IPV6;
  92         default:
  93                 return NFPROTO_UNSPEC;
  94         }
  95 }
  96
  97 /* Map SKB connection state into the values used by flow definition. */
  98 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
  99 {
 100         u8 ct_state = OVS_CS_F_TRACKED;
 101
 102         switch (ctinfo) {
 103         case IP_CT_ESTABLISHED_REPLY:
 104         case IP_CT_RELATED_REPLY:
 105                 ct_state |= OVS_CS_F_REPLY_DIR;
 106                 break;
 107         default:
 108                 break;
 109         }
 110
 111         switch (ctinfo) {
 112         case IP_CT_ESTABLISHED:
 113         case IP_CT_ESTABLISHED_REPLY:
 114                 ct_state |= OVS_CS_F_ESTABLISHED;
 115                 break;
 116         case IP_CT_RELATED:
 117         case IP_CT_RELATED_REPLY:
 118                 ct_state |= OVS_CS_F_RELATED;
 119                 break;
 120         case IP_CT_NEW:
 121                 ct_state |= OVS_CS_F_NEW;
 122                 break;
 123         default:
 124                 break;
 125         }
 126
 127         return ct_state;
 128 }
 129
 130 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 131 {
 132 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 133         return ct ? ct->mark : 0;
 134 #else
 135         return 0;
 136 #endif
 137 }
 138
 139 static void ovs_ct_get_labels(const struct nf_conn *ct,
 140                               struct ovs_key_ct_labels *labels)
 141 {
 142         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 143
 144         if (cl) {
 145                 size_t len = cl->words * sizeof(long);
 146
 147                 if (len > OVS_CT_LABELS_LEN)
 148                         len = OVS_CT_LABELS_LEN;
 149                 else if (len < OVS_CT_LABELS_LEN)
 150                         memset(labels, 0, OVS_CT_LABELS_LEN);
 151                 memcpy(labels, cl->bits, len);
 152         } else {
 153                 memset(labels, 0, OVS_CT_LABELS_LEN);
 154         }
 155 }
 156
 157 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 158                                 const struct nf_conntrack_zone *zone,
 159                                 const struct nf_conn *ct)
 160 {
 161         key->ct.state = state;
 162         key->ct.zone = zone->id;
 163         key->ct.mark = ovs_ct_get_mark(ct);
 164         ovs_ct_get_labels(ct, &key->ct.labels);
 165 }
 166
 167 /* Update 'key' based on skb->nfct.  If 'post_ct' is true, then OVS has
 168  * previously sent the packet to conntrack via the ct action.  If
 169  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 170  * initialized from the connection status.
 171  */
 172 static void ovs_ct_update_key(const struct sk_buff *skb,
 173                               const struct ovs_conntrack_info *info,
 174                               struct sw_flow_key *key, bool post_ct,
 175                               bool keep_nat_flags)
 176 {
 177         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 178         enum ip_conntrack_info ctinfo;
 179         struct nf_conn *ct;
 180         u8 state = 0;
 181
 182         ct = nf_ct_get(skb, &ctinfo);
 183         if (ct) {
 184                 state = ovs_ct_get_state(ctinfo);
 185                 /* All unconfirmed entries are NEW connections. */
 186                 if (!nf_ct_is_confirmed(ct))
 187                         state |= OVS_CS_F_NEW;
 188                 /* OVS persists the related flag for the duration of the
 189                  * connection.
 190                  */
 191                 if (ct->master)
 192                         state |= OVS_CS_F_RELATED;
 193                 if (keep_nat_flags) {
 194                         state |= key->ct.state & OVS_CS_F_NAT_MASK;
 195                 } else {
 196                         if (ct->status & IPS_SRC_NAT)
 197                                 state |= OVS_CS_F_SRC_NAT;
 198                         if (ct->status & IPS_DST_NAT)
 199                                 state |= OVS_CS_F_DST_NAT;
 200                 }
 201                 zone = nf_ct_zone(ct);
 202         } else if (post_ct) {
 203                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 204                 if (info)
 205                         zone = &info->zone;
 206         }
 207         __ovs_ct_update_key(key, state, zone, ct);
 208 }
 209
 210 /* This is called to initialize CT key fields possibly coming in from the local
 211  * stack.
 212  */
 213 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 214 {
 215         ovs_ct_update_key(skb, NULL, key, false, false);
 216 }
 217
 218 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 219 {
 220         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 221                 return -EMSGSIZE;
 222
 223         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 224             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 225                 return -EMSGSIZE;
 226
 227         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 228             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 229                 return -EMSGSIZE;
 230
 231         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 232             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 233                     &key->ct.labels))
 234                 return -EMSGSIZE;
 235
 236         return 0;
 237 }
 238
 239 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 240                            u32 ct_mark, u32 mask)
 241 {
 242 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 243         enum ip_conntrack_info ctinfo;
 244         struct nf_conn *ct;
 245         u32 new_mark;
 246
 247         /* The connection could be invalid, in which case set_mark is no-op. */
 248         ct = nf_ct_get(skb, &ctinfo);
 249         if (!ct)
 250                 return 0;
 251
 252         new_mark = ct_mark | (ct->mark & ~(mask));
 253         if (ct->mark != new_mark) {
 254                 ct->mark = new_mark;
 255                 nf_conntrack_event_cache(IPCT_MARK, ct);
 256                 key->ct.mark = new_mark;
 257         }
 258
 259         return 0;
 260 #else
 261         return -ENOTSUPP;
 262 #endif
 263 }
 264
 265 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 266                              const struct ovs_key_ct_labels *labels,
 267                              const struct ovs_key_ct_labels *mask)
 268 {
 269         enum ip_conntrack_info ctinfo;
 270         struct nf_conn_labels *cl;
 271         struct nf_conn *ct;
 272         int err;
 273
 274         /* The connection could be invalid, in which case set_label is no-op.*/
 275         ct = nf_ct_get(skb, &ctinfo);
 276         if (!ct)
 277                 return 0;
 278
 279         cl = nf_ct_labels_find(ct);
 280         if (!cl) {
 281                 nf_ct_labels_ext_add(ct);
 282                 cl = nf_ct_labels_find(ct);
 283         }
 284         if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
 285                 return -ENOSPC;
 286
 287         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 288                                     OVS_CT_LABELS_LEN / sizeof(u32));
 289         if (err)
 290                 return err;
 291
 292         ovs_ct_get_labels(ct, &key->ct.labels);
 293         return 0;
 294 }
 295
 296 /* 'skb' should already be pulled to nh_ofs. */
 297 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 298 {
 299         const struct nf_conntrack_helper *helper;
 300         const struct nf_conn_help *help;
 301         enum ip_conntrack_info ctinfo;
 302         unsigned int protoff;
 303         struct nf_conn *ct;
 304         u8 nexthdr;
 305         int err;
 306
 307         ct = nf_ct_get(skb, &ctinfo);
 308         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 309                 return NF_ACCEPT;
 310
 311         help = nfct_help(ct);
 312         if (!help)
 313                 return NF_ACCEPT;
 314
 315         helper = rcu_dereference(help->helper);
 316         if (!helper)
 317                 return NF_ACCEPT;
 318
 319         switch (proto) {
 320         case NFPROTO_IPV4:
 321                 protoff = ip_hdrlen(skb);
 322                 break;
 323         case NFPROTO_IPV6: {
 324                 __be16 frag_off;
 325                 int ofs;
 326
 327                 nexthdr = ipv6_hdr(skb)->nexthdr;
 328                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 329                                        &frag_off);
 330                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 331                         pr_debug("proto header not found\n");
 332                         return NF_ACCEPT;
 333                 }
 334                 protoff = ofs;
 335                 break;
 336         }
 337         default:
 338                 WARN_ONCE(1, "helper invoked on non-IP family!");
 339                 return NF_DROP;
 340         }
 341
 342 #if LINUX_VERSION_CODE < KERNEL_VERSION(4,6,0)
 343         /* Linux 4.5 and older depend on skb_dst being set when recalculating
 344          * checksums after NAT helper has mangled TCP or UDP packet payload.
 345          * This dependency is avoided when skb is CHECKSUM_PARTIAL or when UDP
 346          * has no checksum.
 347          *
 348          * The dependency is not triggered when the main NAT code updates
 349          * checksums after translating the IP header (address, port), so this
 350          * fix only needs to be executed on packets that are both being NATted
 351          * and that have a helper assigned.
 352          */
 353         if (ct->status & IPS_NAT_MASK && skb->ip_summed != CHECKSUM_PARTIAL) {
 354                 u8 ipproto = (proto == NFPROTO_IPV4)
 355                         ? ip_hdr(skb)->protocol : nexthdr;
 356                 u16 offset = 0;
 357
 358                 switch (ipproto) {
 359                 case IPPROTO_TCP:
 360                         offset = offsetof(struct tcphdr, check);
 361                         break;
 362                 case IPPROTO_UDP:
 363                         /* Skip if no csum. */
 364                         if (udp_hdr(skb)->check)
 365                                 offset = offsetof(struct udphdr, check);
 366                         break;
 367                 }
 368                 if (offset) {
 369                         if (unlikely(!pskb_may_pull(skb, protoff + offset + 2)))
 370                                 return NF_DROP;
 371
 372                         skb->csum_start = skb_headroom(skb) + protoff;
 373                         skb->csum_offset = offset;
 374                         skb->ip_summed = CHECKSUM_PARTIAL;
 375                 }
 376         }
 377 #endif
 378         err = helper->help(skb, protoff, ct, ctinfo);
 379         if (err != NF_ACCEPT)
 380                 return err;
 381
 382         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 383          * FTP with NAT) adusting the TCP payload size when mangling IP
 384          * addresses and/or port numbers in the text-based control connection.
 385          */
 386         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 387             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 388                 return NF_DROP;
 389         return NF_ACCEPT;
 390 }
 391
 392 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 393  * value if 'skb' is freed.
 394  */
 395 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 396                             u16 zone, struct sk_buff *skb)
 397 {
 398         struct ovs_gso_cb ovs_cb = *OVS_GSO_CB(skb);
 399         int err;
 400
 401         if (!skb->dev) {
 402                 OVS_NLERR(true, "%s: skb has no dev; dropping", __func__);
 403                 return -EINVAL;
 404         }
 405
 406         if (key->eth.type == htons(ETH_P_IP)) {
 407                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 408
 409                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 410                 err = ip_defrag(net, skb, user);
 411                 if (err)
 412                         return err;
 413
 414                 ovs_cb.dp_cb.mru = IPCB(skb)->frag_max_size;
 415 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 416         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 417                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 418
 419                 skb_orphan(skb);
 420                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 421                 err = nf_ct_frag6_gather(net, skb, user);
 422                 if (err)
 423                         return err;
 424
 425                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 426                 ovs_cb.dp_cb.mru = IP6CB(skb)->frag_max_size;
 427 #endif /* IP frag support */
 428         } else {
 429                 kfree_skb(skb);
 430                 return -EPFNOSUPPORT;
 431         }
 432
 433         key->ip.frag = OVS_FRAG_TYPE_NONE;
 434         skb_clear_hash(skb);
 435         skb->ignore_df = 1;
 436         *OVS_GSO_CB(skb) = ovs_cb;
 437
 438         return 0;
 439 }
 440
 441 static struct nf_conntrack_expect *
 442 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 443                    u16 proto, const struct sk_buff *skb)
 444 {
 445         struct nf_conntrack_tuple tuple;
 446
 447         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 448                 return NULL;
 449         return __nf_ct_expect_find(net, zone, &tuple);
 450 }
 451
 452 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 453 static enum ip_conntrack_info
 454 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 455 {
 456         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 457
 458         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 459                 return IP_CT_ESTABLISHED_REPLY;
 460         /* Once we've had two way comms, always ESTABLISHED. */
 461         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 462                 return IP_CT_ESTABLISHED;
 463         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 464                 return IP_CT_RELATED;
 465         return IP_CT_NEW;
 466 }
 467
 468 /* Find an existing connection which this packet belongs to without
 469  * re-attributing statistics or modifying the connection state.  This allows an
 470  * skb->nfct lost due to an upcall to be recovered during actions execution.
 471  *
 472  * Must be called with rcu_read_lock.
 473  *
 474  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 475  * connection.  Returns NULL if there is no existing entry.
 476  */
 477 static struct nf_conn *
 478 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 479                      u8 l3num, struct sk_buff *skb)
 480 {
 481         struct nf_conntrack_l3proto *l3proto;
 482         struct nf_conntrack_l4proto *l4proto;
 483         struct nf_conntrack_tuple tuple;
 484         struct nf_conntrack_tuple_hash *h;
 485         enum ip_conntrack_info ctinfo;
 486         struct nf_conn *ct;
 487         unsigned int dataoff;
 488         u8 protonum;
 489
 490         l3proto = __nf_ct_l3proto_find(l3num);
 491         if (!l3proto) {
 492                 pr_debug("ovs_ct_find_existing: Can't get l3proto\n");
 493                 return NULL;
 494         }
 495         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 496                                  &protonum) <= 0) {
 497                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 498                 return NULL;
 499         }
 500         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 501         if (!l4proto) {
 502                 pr_debug("ovs_ct_find_existing: Can't get l4proto\n");
 503                 return NULL;
 504         }
 505         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 506                              protonum, net, &tuple, l3proto, l4proto)) {
 507                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 508                 return NULL;
 509         }
 510
 511         /* look for tuple match */
 512         h = nf_conntrack_find_get(net, zone, &tuple);
 513         if (!h)
 514                 return NULL;   /* Not found. */
 515
 516         ct = nf_ct_tuplehash_to_ctrack(h);
 517
 518         ctinfo = ovs_ct_get_info(h);
 519         if (ctinfo == IP_CT_NEW) {
 520                 /* This should not happen. */
 521                 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct);
 522         }
 523         skb->nfct = &ct->ct_general;
 524         skb->nfctinfo = ctinfo;
 525         return ct;
 526 }
 527
 528 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 529 static bool skb_nfct_cached(struct net *net,
 530                             const struct sw_flow_key *key,
 531                             const struct ovs_conntrack_info *info,
 532                             struct sk_buff *skb)
 533 {
 534         enum ip_conntrack_info ctinfo;
 535         struct nf_conn *ct;
 536
 537         ct = nf_ct_get(skb, &ctinfo);
 538         /* If no ct, check if we have evidence that an existing conntrack entry
 539          * might be found for this skb.  This happens when we lose a skb->nfct
 540          * due to an upcall.  If the connection was not confirmed, it is not
 541          * cached and needs to be run through conntrack again.
 542          */
 543         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 544             !(key->ct.state & OVS_CS_F_INVALID) &&
 545             key->ct.zone == info->zone.id)
 546                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 547         if (!ct)
 548                 return false;
 549         if (!net_eq(net, read_pnet(&ct->ct_net)))
 550                 return false;
 551         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 552                 return false;
 553         if (info->helper) {
 554                 struct nf_conn_help *help;
 555
 556                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 557                 if (help && rcu_access_pointer(help->helper) != info->helper)
 558                         return false;
 559         }
 560
 561         return true;
 562 }
 563
 564 #ifdef CONFIG_NF_NAT_NEEDED
 565 /* Modelled after nf_nat_ipv[46]_fn().
 566  * range is only used for new, uninitialized NAT state.
 567  * Returns either NF_ACCEPT or NF_DROP.
 568  */
 569 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 570                               enum ip_conntrack_info ctinfo,
 571                               const struct nf_nat_range *range,
 572                               enum nf_nat_manip_type maniptype)
 573 {
 574         int hooknum, nh_off, err = NF_ACCEPT;
 575
 576         nh_off = skb_network_offset(skb);
 577         skb_pull(skb, nh_off);
 578
 579         /* See HOOK2MANIP(). */
 580         if (maniptype == NF_NAT_MANIP_SRC)
 581                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 582         else
 583                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 584
 585         switch (ctinfo) {
 586         case IP_CT_RELATED:
 587         case IP_CT_RELATED_REPLY:
 588                 if (skb->protocol == htons(ETH_P_IP) &&
 589                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 590                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 591                                                            hooknum))
 592                                 err = NF_DROP;
 593                         goto push;
 594 #if IS_ENABLED(CONFIG_NF_NAT_IPV6)
 595                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
 596                         __be16 frag_off;
 597                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 598                         int hdrlen = ipv6_skip_exthdr(skb,
 599                                                       sizeof(struct ipv6hdr),
 600                                                       &nexthdr, &frag_off);
 601
 602                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 603                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 604                                                                      ctinfo,
 605                                                                      hooknum,
 606                                                                      hdrlen))
 607                                         err = NF_DROP;
 608                                 goto push;
 609                         }
 610 #endif
 611                 }
 612                 /* Non-ICMP, fall thru to initialize if needed. */
 613         case IP_CT_NEW:
 614                 /* Seen it before?  This can happen for loopback, retrans,
 615                  * or local packets.
 616                  */
 617                 if (!nf_nat_initialized(ct, maniptype)) {
 618                         /* Initialize according to the NAT action. */
 619                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 620                                 /* Action is set up to establish a new
 621                                  * mapping.
 622                                  */
 623                                 ? nf_nat_setup_info(ct, range, maniptype)
 624                                 : nf_nat_alloc_null_binding(ct, hooknum);
 625                         if (err != NF_ACCEPT)
 626                                 goto push;
 627                 }
 628                 break;
 629
 630         case IP_CT_ESTABLISHED:
 631         case IP_CT_ESTABLISHED_REPLY:
 632                 break;
 633
 634         default:
 635                 err = NF_DROP;
 636                 goto push;
 637         }
 638
 639         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 640 push:
 641         skb_push(skb, nh_off);
 642
 643         return err;
 644 }
 645
 646 static void ovs_nat_update_key(struct sw_flow_key *key,
 647                                const struct sk_buff *skb,
 648                                enum nf_nat_manip_type maniptype)
 649 {
 650         if (maniptype == NF_NAT_MANIP_SRC) {
 651                 __be16 src;
 652
 653                 key->ct.state |= OVS_CS_F_SRC_NAT;
 654                 if (key->eth.type == htons(ETH_P_IP))
 655                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 656                 else if (key->eth.type == htons(ETH_P_IPV6))
 657                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 658                                sizeof(key->ipv6.addr.src));
 659                 else
 660                         return;
 661
 662                 if (key->ip.proto == IPPROTO_UDP)
 663                         src = udp_hdr(skb)->source;
 664                 else if (key->ip.proto == IPPROTO_TCP)
 665                         src = tcp_hdr(skb)->source;
 666                 else if (key->ip.proto == IPPROTO_SCTP)
 667                         src = sctp_hdr(skb)->source;
 668                 else
 669                         return;
 670
 671                 key->tp.src = src;
 672         } else {
 673                 __be16 dst;
 674
 675                 key->ct.state |= OVS_CS_F_DST_NAT;
 676                 if (key->eth.type == htons(ETH_P_IP))
 677                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 678                 else if (key->eth.type == htons(ETH_P_IPV6))
 679                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 680                                sizeof(key->ipv6.addr.dst));
 681                 else
 682                         return;
 683
 684                 if (key->ip.proto == IPPROTO_UDP)
 685                         dst = udp_hdr(skb)->dest;
 686                 else if (key->ip.proto == IPPROTO_TCP)
 687                         dst = tcp_hdr(skb)->dest;
 688                 else if (key->ip.proto == IPPROTO_SCTP)
 689                         dst = sctp_hdr(skb)->dest;
 690                 else
 691                         return;
 692
 693                 key->tp.dst = dst;
 694         }
 695 }
 696
 697 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 698 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 699                       const struct ovs_conntrack_info *info,
 700                       struct sk_buff *skb, struct nf_conn *ct,
 701                       enum ip_conntrack_info ctinfo)
 702 {
 703         enum nf_nat_manip_type maniptype;
 704         int err;
 705
 706         if (nf_ct_is_untracked(ct)) {
 707                 /* A NAT action may only be performed on tracked packets. */
 708                 return NF_ACCEPT;
 709         }
 710
 711         /* Add NAT extension if not confirmed yet. */
 712         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 713                 return NF_ACCEPT;   /* Can't NAT. */
 714
 715         /* Determine NAT type.
 716          * Check if the NAT type can be deduced from the tracked connection.
 717          * Make sure expected traffic is NATted only when committing.
 718          */
 719         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 720             ct->status & IPS_NAT_MASK &&
 721             (!(ct->status & IPS_EXPECTED_BIT) || info->commit)) {
 722                 /* NAT an established or related connection like before. */
 723                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 724                         /* This is the REPLY direction for a connection
 725                          * for which NAT was applied in the forward
 726                          * direction.  Do the reverse NAT.
 727                          */
 728                         maniptype = ct->status & IPS_SRC_NAT
 729                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 730                 else
 731                         maniptype = ct->status & IPS_SRC_NAT
 732                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 733         } else if (info->nat & OVS_CT_SRC_NAT) {
 734                 maniptype = NF_NAT_MANIP_SRC;
 735         } else if (info->nat & OVS_CT_DST_NAT) {
 736                 maniptype = NF_NAT_MANIP_DST;
 737         } else {
 738                 return NF_ACCEPT; /* Connection is not NATed. */
 739         }
 740         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 741
 742         /* Mark NAT done if successful and update the flow key. */
 743         if (err == NF_ACCEPT)
 744                 ovs_nat_update_key(key, skb, maniptype);
 745
 746         return err;
 747 }
 748 #else /* !CONFIG_NF_NAT_NEEDED */
 749 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 750                       const struct ovs_conntrack_info *info,
 751                       struct sk_buff *skb, struct nf_conn *ct,
 752                       enum ip_conntrack_info ctinfo)
 753 {
 754         return NF_ACCEPT;
 755 }
 756 #endif
 757
 758 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 759  * not done already.  Update key with new CT state after passing the packet
 760  * through conntrack.
 761  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 762  * set to NULL and 0 will be returned.
 763  */
 764 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 765                            const struct ovs_conntrack_info *info,
 766                            struct sk_buff *skb)
 767 {
 768         /* If we are recirculating packets to match on conntrack fields and
 769          * committing with a separate conntrack action,  then we don't need to
 770          * actually run the packet through conntrack twice unless it's for a
 771          * different zone.
 772          */
 773         bool cached = skb_nfct_cached(net, key, info, skb);
 774         enum ip_conntrack_info ctinfo;
 775         struct nf_conn *ct;
 776
 777         if (!cached) {
 778                 struct nf_conn *tmpl = info->ct;
 779                 int err;
 780
 781                 /* Associate skb with specified zone. */
 782                 if (tmpl) {
 783                         if (skb->nfct)
 784                                 nf_conntrack_put(skb->nfct);
 785                         nf_conntrack_get(&tmpl->ct_general);
 786                         skb->nfct = &tmpl->ct_general;
 787                         skb->nfctinfo = IP_CT_NEW;
 788                 }
 789
 790                 /* Repeat if requested, see nf_iterate(). */
 791                 do {
 792                         err = nf_conntrack_in(net, info->family,
 793                                               NF_INET_FORWARD, skb);
 794                 } while (err == NF_REPEAT);
 795
 796                 if (err != NF_ACCEPT)
 797                         return -ENOENT;
 798
 799                 /* Clear CT state NAT flags to mark that we have not yet done
 800                  * NAT after the nf_conntrack_in() call.  We can actually clear
 801                  * the whole state, as it will be re-initialized below.
 802                  */
 803                 key->ct.state = 0;
 804
 805                 /* Update the key, but keep the NAT flags. */
 806                 ovs_ct_update_key(skb, info, key, true, true);
 807         }
 808
 809         ct = nf_ct_get(skb, &ctinfo);
 810         if (ct) {
 811                 /* Packets starting a new connection must be NATted before the
 812                  * helper, so that the helper knows about the NAT.  We enforce
 813                  * this by delaying both NAT and helper calls for unconfirmed
 814                  * connections until the committing CT action.  For later
 815                  * packets NAT and Helper may be called in either order.
 816                  *
 817                  * NAT will be done only if the CT action has NAT, and only
 818                  * once per packet (per zone), as guarded by the NAT bits in
 819                  * the key->ct.state.
 820                  */
 821                 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
 822                     (nf_ct_is_confirmed(ct) || info->commit) &&
 823                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 824                         return -EINVAL;
 825                 }
 826
 827                 /* Call the helper only if:
 828                  * - nf_conntrack_in() was executed above ("!cached") for a
 829                  *   confirmed connection, or
 830                  * - When committing an unconfirmed connection.
 831                  */
 832                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 833                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 834                         return -EINVAL;
 835                 }
 836         }
 837
 838         return 0;
 839 }
 840
 841 /* Lookup connection and read fields into key. */
 842 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 843                          const struct ovs_conntrack_info *info,
 844                          struct sk_buff *skb)
 845 {
 846         struct nf_conntrack_expect *exp;
 847
 848         /* If we pass an expected packet through nf_conntrack_in() the
 849          * expectation is typically removed, but the packet could still be
 850          * lost in upcall processing.  To prevent this from happening we
 851          * perform an explicit expectation lookup.  Expected connections are
 852          * always new, and will be passed through conntrack only when they are
 853          * committed, as it is OK to remove the expectation at that time.
 854          */
 855         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 856         if (exp) {
 857                 u8 state;
 858
 859                 /* NOTE: New connections are NATted and Helped only when
 860                  * committed, so we are not calling into NAT here.
 861                  */
 862                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 863                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 864         } else
 865                 return __ovs_ct_lookup(net, key, info, skb);
 866
 867         return 0;
 868 }
 869
 870 /* Lookup connection and confirm if unconfirmed. */
 871 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 872                          const struct ovs_conntrack_info *info,
 873                          struct sk_buff *skb)
 874 {
 875         int err;
 876
 877         err = __ovs_ct_lookup(net, key, info, skb);
 878         if (err)
 879                 return err;
 880         /* This is a no-op if the connection has already been confirmed. */
 881         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 882                 return -EINVAL;
 883
 884         return 0;
 885 }
 886
 887 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 888 {
 889         size_t i;
 890
 891         for (i = 0; i < sizeof(*labels); i++)
 892                 if (labels->ct_labels[i])
 893                         return true;
 894
 895         return false;
 896 }
 897
 898 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 899  * value if 'skb' is freed.
 900  */
 901 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 902                    struct sw_flow_key *key,
 903                    const struct ovs_conntrack_info *info)
 904 {
 905         int nh_ofs;
 906         int err;
 907
 908         /* The conntrack module expects to be working at L3. */
 909         nh_ofs = skb_network_offset(skb);
 910         skb_pull(skb, nh_ofs);
 911
 912         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 913                 err = handle_fragments(net, key, info->zone.id, skb);
 914                 if (err)
 915                         return err;
 916         }
 917
 918         if (info->commit)
 919                 err = ovs_ct_commit(net, key, info, skb);
 920         else
 921                 err = ovs_ct_lookup(net, key, info, skb);
 922         if (err)
 923                 goto err;
 924
 925         if (info->mark.mask) {
 926                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 927                                       info->mark.mask);
 928                 if (err)
 929                         goto err;
 930         }
 931         if (labels_nonzero(&info->labels.mask))
 932                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 933                                         &info->labels.mask);
 934 err:
 935         skb_push(skb, nh_ofs);
 936         if (err)
 937                 kfree_skb(skb);
 938         return err;
 939 }
 940
 941 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 942                              const struct sw_flow_key *key, bool log)
 943 {
 944         struct nf_conntrack_helper *helper;
 945         struct nf_conn_help *help;
 946
 947         helper = nf_conntrack_helper_try_module_get(name, info->family,
 948                                                     key->ip.proto);
 949         if (!helper) {
 950                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 951                 return -EINVAL;
 952         }
 953
 954         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 955         if (!help) {
 956                 module_put(helper->me);
 957                 return -ENOMEM;
 958         }
 959
 960         rcu_assign_pointer(help->helper, helper);
 961         info->helper = helper;
 962         return 0;
 963 }
 964
 965 #ifdef CONFIG_NF_NAT_NEEDED
 966 static int parse_nat(const struct nlattr *attr,
 967                      struct ovs_conntrack_info *info, bool log)
 968 {
 969         struct nlattr *a;
 970         int rem;
 971         bool have_ip_max = false;
 972         bool have_proto_max = false;
 973         bool ip_vers = (info->family == NFPROTO_IPV6);
 974
 975         nla_for_each_nested(a, attr, rem) {
 976                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
 977                         [OVS_NAT_ATTR_SRC] = {0, 0},
 978                         [OVS_NAT_ATTR_DST] = {0, 0},
 979                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
 980                                                  sizeof(struct in6_addr)},
 981                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
 982                                                  sizeof(struct in6_addr)},
 983                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
 984                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
 985                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
 986                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
 987                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
 988                 };
 989                 int type = nla_type(a);
 990
 991                 if (type > OVS_NAT_ATTR_MAX) {
 992                         OVS_NLERR(log,
 993                                   "Unknown NAT attribute (type=%d, max=%d).\n",
 994                                   type, OVS_NAT_ATTR_MAX);
 995                         return -EINVAL;
 996                 }
 997
 998                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
 999                         OVS_NLERR(log,
1000                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
1001                                   type, nla_len(a),
1002                                   ovs_nat_attr_lens[type][ip_vers]);
1003                         return -EINVAL;
1004                 }
1005
1006                 switch (type) {
1007                 case OVS_NAT_ATTR_SRC:
1008                 case OVS_NAT_ATTR_DST:
1009                         if (info->nat) {
1010                                 OVS_NLERR(log,
1011                                           "Only one type of NAT may be specified.\n"
1012                                           );
1013                                 return -ERANGE;
1014                         }
1015                         info->nat |= OVS_CT_NAT;
1016                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1017                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1018                         break;
1019
1020                 case OVS_NAT_ATTR_IP_MIN:
1021                         nla_memcpy(&info->range.min_addr, a,
1022                                    sizeof(info->range.min_addr));
1023                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1024                         break;
1025
1026                 case OVS_NAT_ATTR_IP_MAX:
1027                         have_ip_max = true;
1028                         nla_memcpy(&info->range.max_addr, a,
1029                                    sizeof(info->range.max_addr));
1030                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1031                         break;
1032
1033                 case OVS_NAT_ATTR_PROTO_MIN:
1034                         info->range.min_proto.all = htons(nla_get_u16(a));
1035                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1036                         break;
1037
1038                 case OVS_NAT_ATTR_PROTO_MAX:
1039                         have_proto_max = true;
1040                         info->range.max_proto.all = htons(nla_get_u16(a));
1041                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1042                         break;
1043
1044                 case OVS_NAT_ATTR_PERSISTENT:
1045                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1046                         break;
1047
1048                 case OVS_NAT_ATTR_PROTO_HASH:
1049                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1050                         break;
1051
1052                 case OVS_NAT_ATTR_PROTO_RANDOM:
1053 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1054                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1055 #else
1056                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1057                         info->random_fully_compat = true;
1058 #endif
1059                         break;
1060
1061                 default:
1062                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1063                         return -EINVAL;
1064                 }
1065         }
1066
1067         if (rem > 0) {
1068                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1069                 return -EINVAL;
1070         }
1071         if (!info->nat) {
1072                 /* Do not allow flags if no type is given. */
1073                 if (info->range.flags) {
1074                         OVS_NLERR(log,
1075                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1076                                   );
1077                         return -EINVAL;
1078                 }
1079                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1080         } else if (!info->commit) {
1081                 OVS_NLERR(log,
1082                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1083                           );
1084                 return -EINVAL;
1085         }
1086         /* Allow missing IP_MAX. */
1087         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1088                 memcpy(&info->range.max_addr, &info->range.min_addr,
1089                        sizeof(info->range.max_addr));
1090         }
1091         /* Allow missing PROTO_MAX. */
1092         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1093             !have_proto_max) {
1094                 info->range.max_proto.all = info->range.min_proto.all;
1095         }
1096         return 0;
1097 }
1098 #endif
1099
1100 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1101         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1102         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1103                                     .maxlen = sizeof(u16) },
1104         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1105                                     .maxlen = sizeof(struct md_mark) },
1106         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1107                                     .maxlen = sizeof(struct md_labels) },
1108         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1109                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1110 #ifdef CONFIG_NF_NAT_NEEDED
1111         /* NAT length is checked when parsing the nested attributes. */
1112         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1113 #endif
1114 };
1115
1116 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1117                     const char **helper, bool log)
1118 {
1119         struct nlattr *a;
1120         int rem;
1121
1122         nla_for_each_nested(a, attr, rem) {
1123                 int type = nla_type(a);
1124                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1125                 int minlen = ovs_ct_attr_lens[type].minlen;
1126
1127                 if (type > OVS_CT_ATTR_MAX) {
1128                         OVS_NLERR(log,
1129                                   "Unknown conntrack attr (type=%d, max=%d)",
1130                                   type, OVS_CT_ATTR_MAX);
1131                         return -EINVAL;
1132                 }
1133                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1134                         OVS_NLERR(log,
1135                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1136                                   type, nla_len(a), maxlen);
1137                         return -EINVAL;
1138                 }
1139
1140                 switch (type) {
1141                 case OVS_CT_ATTR_COMMIT:
1142                         info->commit = true;
1143                         break;
1144 #ifdef CONFIG_NF_CONNTRACK_ZONES
1145                 case OVS_CT_ATTR_ZONE:
1146                         info->zone.id = nla_get_u16(a);
1147                         break;
1148 #endif
1149 #ifdef CONFIG_NF_CONNTRACK_MARK
1150                 case OVS_CT_ATTR_MARK: {
1151                         struct md_mark *mark = nla_data(a);
1152
1153                         if (!mark->mask) {
1154                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1155                                 return -EINVAL;
1156                         }
1157                         info->mark = *mark;
1158                         break;
1159                 }
1160 #endif
1161 #ifdef CONFIG_NF_CONNTRACK_LABELS
1162                 case OVS_CT_ATTR_LABELS: {
1163                         struct md_labels *labels = nla_data(a);
1164
1165                         if (!labels_nonzero(&labels->mask)) {
1166                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1167                                 return -EINVAL;
1168                         }
1169                         info->labels = *labels;
1170                         break;
1171                 }
1172 #endif
1173                 case OVS_CT_ATTR_HELPER:
1174                         *helper = nla_data(a);
1175                         if (!memchr(*helper, '\0', nla_len(a))) {
1176                                 OVS_NLERR(log, "Invalid conntrack helper");
1177                                 return -EINVAL;
1178                         }
1179                         break;
1180 #ifdef CONFIG_NF_NAT_NEEDED
1181                 case OVS_CT_ATTR_NAT: {
1182                         int err = parse_nat(a, info, log);
1183
1184                         if (err)
1185                                 return err;
1186                         break;
1187                 }
1188 #endif
1189                 default:
1190                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1191                                   type);
1192                         return -EINVAL;
1193                 }
1194         }
1195
1196         if (rem > 0) {
1197                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1198                 return -EINVAL;
1199         }
1200
1201         return 0;
1202 }
1203
1204 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1205 {
1206         if (attr == OVS_KEY_ATTR_CT_STATE)
1207                 return true;
1208         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1209             attr == OVS_KEY_ATTR_CT_ZONE)
1210                 return true;
1211         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1212             attr == OVS_KEY_ATTR_CT_MARK)
1213                 return true;
1214         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1215             attr == OVS_KEY_ATTR_CT_LABELS) {
1216                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1217
1218                 return ovs_net->xt_label;
1219         }
1220
1221         return false;
1222 }
1223
1224 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1225                        const struct sw_flow_key *key,
1226                        struct sw_flow_actions **sfa,  bool log)
1227 {
1228         struct ovs_conntrack_info ct_info;
1229         const char *helper = NULL;
1230         u16 family;
1231         int err;
1232
1233         family = key_to_nfproto(key);
1234         if (family == NFPROTO_UNSPEC) {
1235                 OVS_NLERR(log, "ct family unspecified");
1236                 return -EINVAL;
1237         }
1238
1239         memset(&ct_info, 0, sizeof(ct_info));
1240         ct_info.family = family;
1241
1242         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1243                         NF_CT_DEFAULT_ZONE_DIR, 0);
1244
1245         err = parse_ct(attr, &ct_info, &helper, log);
1246         if (err)
1247                 return err;
1248
1249         /* Set up template for tracking connections in specific zones. */
1250         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1251         if (!ct_info.ct) {
1252                 OVS_NLERR(log, "Failed to allocate conntrack template");
1253                 return -ENOMEM;
1254         }
1255
1256         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1257         nf_conntrack_get(&ct_info.ct->ct_general);
1258
1259         if (helper) {
1260                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1261                 if (err)
1262                         goto err_free_ct;
1263         }
1264
1265         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1266                                  sizeof(ct_info), log);
1267         if (err)
1268                 goto err_free_ct;
1269
1270         return 0;
1271 err_free_ct:
1272         __ovs_ct_free_action(&ct_info);
1273         return err;
1274 }
1275
1276 #ifdef CONFIG_NF_NAT_NEEDED
1277 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1278                                struct sk_buff *skb)
1279 {
1280         struct nlattr *start;
1281
1282         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1283         if (!start)
1284                 return false;
1285
1286         if (info->nat & OVS_CT_SRC_NAT) {
1287                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1288                         return false;
1289         } else if (info->nat & OVS_CT_DST_NAT) {
1290                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1291                         return false;
1292         } else {
1293                 goto out;
1294         }
1295
1296         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1297                 if (info->family == NFPROTO_IPV4) {
1298                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1299                                             info->range.min_addr.ip) ||
1300                             (info->range.max_addr.ip
1301                              != info->range.min_addr.ip &&
1302                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1303                                               info->range.max_addr.ip))))
1304                                 return false;
1305 #if IS_ENABLED(CONFIG_NF_NAT_IPV6)
1306                 } else if (info->family == NFPROTO_IPV6) {
1307                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1308                                              &info->range.min_addr.in6) ||
1309                             (memcmp(&info->range.max_addr.in6,
1310                                     &info->range.min_addr.in6,
1311                                     sizeof(info->range.max_addr.in6)) &&
1312                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1313                                                &info->range.max_addr.in6))))
1314                                 return false;
1315 #endif
1316                 } else {
1317                         return false;
1318                 }
1319         }
1320         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1321             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1322                          ntohs(info->range.min_proto.all)) ||
1323              (info->range.max_proto.all != info->range.min_proto.all &&
1324               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1325                           ntohs(info->range.max_proto.all)))))
1326                 return false;
1327
1328         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1329             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1330                 return false;
1331         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1332             nla_put_flag(skb, info->random_fully_compat
1333                          ? OVS_NAT_ATTR_PROTO_RANDOM
1334                          : OVS_NAT_ATTR_PROTO_HASH))
1335                 return false;
1336 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1337         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1338             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1339                 return false;
1340 #endif
1341 out:
1342         nla_nest_end(skb, start);
1343
1344         return true;
1345 }
1346 #endif
1347
1348 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1349                           struct sk_buff *skb)
1350 {
1351         struct nlattr *start;
1352
1353         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1354         if (!start)
1355                 return -EMSGSIZE;
1356
1357         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1358                 return -EMSGSIZE;
1359         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1360             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1361                 return -EMSGSIZE;
1362         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1363             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1364                     &ct_info->mark))
1365                 return -EMSGSIZE;
1366         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1367             labels_nonzero(&ct_info->labels.mask) &&
1368             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1369                     &ct_info->labels))
1370                 return -EMSGSIZE;
1371         if (ct_info->helper) {
1372                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1373                                    ct_info->helper->name))
1374                         return -EMSGSIZE;
1375         }
1376 #ifdef CONFIG_NF_NAT_NEEDED
1377         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1378                 return -EMSGSIZE;
1379 #endif
1380         nla_nest_end(skb, start);
1381
1382         return 0;
1383 }
1384
1385 void ovs_ct_free_action(const struct nlattr *a)
1386 {
1387         struct ovs_conntrack_info *ct_info = nla_data(a);
1388
1389         __ovs_ct_free_action(ct_info);
1390 }
1391
1392 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1393 {
1394         if (ct_info->helper)
1395                 module_put(ct_info->helper->me);
1396         if (ct_info->ct)
1397                 nf_ct_tmpl_free(ct_info->ct);
1398 }
1399
1400 void ovs_ct_init(struct net *net)
1401 {
1402         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1403         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1404
1405         if (nf_connlabels_get(net, n_bits)) {
1406                 ovs_net->xt_label = false;
1407                 OVS_NLERR(true, "Failed to set connlabel length");
1408         } else {
1409                 ovs_net->xt_label = true;
1410         }
1411 }
1412
1413 void ovs_ct_exit(struct net *net)
1414 {
1415         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1416
1417         if (ovs_net->xt_label)
1418                 nf_connlabels_put(net);
1419 }
1420
1421 #endif /* CONFIG_NF_CONNTRACK */