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 (key->eth.type == htons(ETH_P_IP)) {
 402                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 403
 404                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 405                 err = ip_defrag(net, skb, user);
 406                 if (err)
 407                         return err;
 408
 409                 ovs_cb.dp_cb.mru = IPCB(skb)->frag_max_size;
 410 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 411         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 412                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 413
 414                 skb_orphan(skb);
 415                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 416                 err = nf_ct_frag6_gather(net, skb, user);
 417                 if (err)
 418                         return err;
 419
 420                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 421                 ovs_cb.dp_cb.mru = IP6CB(skb)->frag_max_size;
 422 #endif /* IP frag support */
 423         } else {
 424                 kfree_skb(skb);
 425                 return -EPFNOSUPPORT;
 426         }
 427
 428         key->ip.frag = OVS_FRAG_TYPE_NONE;
 429         skb_clear_hash(skb);
 430         skb->ignore_df = 1;
 431         *OVS_GSO_CB(skb) = ovs_cb;
 432
 433         return 0;
 434 }
 435
 436 static struct nf_conntrack_expect *
 437 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 438                    u16 proto, const struct sk_buff *skb)
 439 {
 440         struct nf_conntrack_tuple tuple;
 441
 442         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 443                 return NULL;
 444         return __nf_ct_expect_find(net, zone, &tuple);
 445 }
 446
 447 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 448 static enum ip_conntrack_info
 449 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 450 {
 451         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 452
 453         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 454                 return IP_CT_ESTABLISHED_REPLY;
 455         /* Once we've had two way comms, always ESTABLISHED. */
 456         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 457                 return IP_CT_ESTABLISHED;
 458         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 459                 return IP_CT_RELATED;
 460         return IP_CT_NEW;
 461 }
 462
 463 /* Find an existing connection which this packet belongs to without
 464  * re-attributing statistics or modifying the connection state.  This allows an
 465  * skb->nfct lost due to an upcall to be recovered during actions execution.
 466  *
 467  * Must be called with rcu_read_lock.
 468  *
 469  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 470  * connection.  Returns NULL if there is no existing entry.
 471  */
 472 static struct nf_conn *
 473 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 474                      u8 l3num, struct sk_buff *skb)
 475 {
 476         struct nf_conntrack_l3proto *l3proto;
 477         struct nf_conntrack_l4proto *l4proto;
 478         struct nf_conntrack_tuple tuple;
 479         struct nf_conntrack_tuple_hash *h;
 480         enum ip_conntrack_info ctinfo;
 481         struct nf_conn *ct;
 482         unsigned int dataoff;
 483         u8 protonum;
 484
 485         l3proto = __nf_ct_l3proto_find(l3num);
 486         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 487                                  &protonum) <= 0) {
 488                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 489                 return NULL;
 490         }
 491         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 492         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 493                              protonum, net, &tuple, l3proto, l4proto)) {
 494                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 495                 return NULL;
 496         }
 497
 498         /* look for tuple match */
 499         h = nf_conntrack_find_get(net, zone, &tuple);
 500         if (!h)
 501                 return NULL;   /* Not found. */
 502
 503         ct = nf_ct_tuplehash_to_ctrack(h);
 504
 505         ctinfo = ovs_ct_get_info(h);
 506         if (ctinfo == IP_CT_NEW) {
 507                 /* This should not happen. */
 508                 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct);
 509         }
 510         skb->nfct = &ct->ct_general;
 511         skb->nfctinfo = ctinfo;
 512         return ct;
 513 }
 514
 515 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 516 static bool skb_nfct_cached(struct net *net,
 517                             const struct sw_flow_key *key,
 518                             const struct ovs_conntrack_info *info,
 519                             struct sk_buff *skb)
 520 {
 521         enum ip_conntrack_info ctinfo;
 522         struct nf_conn *ct;
 523
 524         ct = nf_ct_get(skb, &ctinfo);
 525         /* If no ct, check if we have evidence that an existing conntrack entry
 526          * might be found for this skb.  This happens when we lose a skb->nfct
 527          * due to an upcall.  If the connection was not confirmed, it is not
 528          * cached and needs to be run through conntrack again.
 529          */
 530         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 531             !(key->ct.state & OVS_CS_F_INVALID) &&
 532             key->ct.zone == info->zone.id)
 533                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 534         if (!ct)
 535                 return false;
 536         if (!net_eq(net, read_pnet(&ct->ct_net)))
 537                 return false;
 538         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 539                 return false;
 540         if (info->helper) {
 541                 struct nf_conn_help *help;
 542
 543                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 544                 if (help && rcu_access_pointer(help->helper) != info->helper)
 545                         return false;
 546         }
 547
 548         return true;
 549 }
 550
 551 #ifdef CONFIG_NF_NAT_NEEDED
 552 /* Modelled after nf_nat_ipv[46]_fn().
 553  * range is only used for new, uninitialized NAT state.
 554  * Returns either NF_ACCEPT or NF_DROP.
 555  */
 556 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 557                               enum ip_conntrack_info ctinfo,
 558                               const struct nf_nat_range *range,
 559                               enum nf_nat_manip_type maniptype)
 560 {
 561         int hooknum, nh_off, err = NF_ACCEPT;
 562
 563         nh_off = skb_network_offset(skb);
 564         skb_pull(skb, nh_off);
 565
 566         /* See HOOK2MANIP(). */
 567         if (maniptype == NF_NAT_MANIP_SRC)
 568                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 569         else
 570                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 571
 572         switch (ctinfo) {
 573         case IP_CT_RELATED:
 574         case IP_CT_RELATED_REPLY:
 575                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 576                     skb->protocol == htons(ETH_P_IP) &&
 577                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 578                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 579                                                            hooknum))
 580                                 err = NF_DROP;
 581                         goto push;
 582                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 583                            skb->protocol == htons(ETH_P_IPV6)) {
 584                         __be16 frag_off;
 585                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 586                         int hdrlen = ipv6_skip_exthdr(skb,
 587                                                       sizeof(struct ipv6hdr),
 588                                                       &nexthdr, &frag_off);
 589
 590                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 591                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 592                                                                      ctinfo,
 593                                                                      hooknum,
 594                                                                      hdrlen))
 595                                         err = NF_DROP;
 596                                 goto push;
 597                         }
 598                 }
 599                 /* Non-ICMP, fall thru to initialize if needed. */
 600         case IP_CT_NEW:
 601                 /* Seen it before?  This can happen for loopback, retrans,
 602                  * or local packets.
 603                  */
 604                 if (!nf_nat_initialized(ct, maniptype)) {
 605                         /* Initialize according to the NAT action. */
 606                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 607                                 /* Action is set up to establish a new
 608                                  * mapping.
 609                                  */
 610                                 ? nf_nat_setup_info(ct, range, maniptype)
 611                                 : nf_nat_alloc_null_binding(ct, hooknum);
 612                         if (err != NF_ACCEPT)
 613                                 goto push;
 614                 }
 615                 break;
 616
 617         case IP_CT_ESTABLISHED:
 618         case IP_CT_ESTABLISHED_REPLY:
 619                 break;
 620
 621         default:
 622                 err = NF_DROP;
 623                 goto push;
 624         }
 625
 626         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 627 push:
 628         skb_push(skb, nh_off);
 629
 630         return err;
 631 }
 632
 633 static void ovs_nat_update_key(struct sw_flow_key *key,
 634                                const struct sk_buff *skb,
 635                                enum nf_nat_manip_type maniptype)
 636 {
 637         if (maniptype == NF_NAT_MANIP_SRC) {
 638                 __be16 src;
 639
 640                 key->ct.state |= OVS_CS_F_SRC_NAT;
 641                 if (key->eth.type == htons(ETH_P_IP))
 642                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 643                 else if (key->eth.type == htons(ETH_P_IPV6))
 644                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 645                                sizeof(key->ipv6.addr.src));
 646                 else
 647                         return;
 648
 649                 if (key->ip.proto == IPPROTO_UDP)
 650                         src = udp_hdr(skb)->source;
 651                 else if (key->ip.proto == IPPROTO_TCP)
 652                         src = tcp_hdr(skb)->source;
 653                 else if (key->ip.proto == IPPROTO_SCTP)
 654                         src = sctp_hdr(skb)->source;
 655                 else
 656                         return;
 657
 658                 key->tp.src = src;
 659         } else {
 660                 __be16 dst;
 661
 662                 key->ct.state |= OVS_CS_F_DST_NAT;
 663                 if (key->eth.type == htons(ETH_P_IP))
 664                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 665                 else if (key->eth.type == htons(ETH_P_IPV6))
 666                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 667                                sizeof(key->ipv6.addr.dst));
 668                 else
 669                         return;
 670
 671                 if (key->ip.proto == IPPROTO_UDP)
 672                         dst = udp_hdr(skb)->dest;
 673                 else if (key->ip.proto == IPPROTO_TCP)
 674                         dst = tcp_hdr(skb)->dest;
 675                 else if (key->ip.proto == IPPROTO_SCTP)
 676                         dst = sctp_hdr(skb)->dest;
 677                 else
 678                         return;
 679
 680                 key->tp.dst = dst;
 681         }
 682 }
 683
 684 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 685 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 686                       const struct ovs_conntrack_info *info,
 687                       struct sk_buff *skb, struct nf_conn *ct,
 688                       enum ip_conntrack_info ctinfo)
 689 {
 690         enum nf_nat_manip_type maniptype;
 691         int err;
 692
 693         if (nf_ct_is_untracked(ct)) {
 694                 /* A NAT action may only be performed on tracked packets. */
 695                 return NF_ACCEPT;
 696         }
 697
 698         /* Add NAT extension if not confirmed yet. */
 699         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 700                 return NF_ACCEPT;   /* Can't NAT. */
 701
 702         /* Determine NAT type.
 703          * Check if the NAT type can be deduced from the tracked connection.
 704          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 705          * when committing.
 706          */
 707         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 708             ct->status & IPS_NAT_MASK &&
 709             (ctinfo != IP_CT_RELATED || info->commit)) {
 710                 /* NAT an established or related connection like before. */
 711                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 712                         /* This is the REPLY direction for a connection
 713                          * for which NAT was applied in the forward
 714                          * direction.  Do the reverse NAT.
 715                          */
 716                         maniptype = ct->status & IPS_SRC_NAT
 717                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 718                 else
 719                         maniptype = ct->status & IPS_SRC_NAT
 720                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 721         } else if (info->nat & OVS_CT_SRC_NAT) {
 722                 maniptype = NF_NAT_MANIP_SRC;
 723         } else if (info->nat & OVS_CT_DST_NAT) {
 724                 maniptype = NF_NAT_MANIP_DST;
 725         } else {
 726                 return NF_ACCEPT; /* Connection is not NATed. */
 727         }
 728         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 729
 730         /* Mark NAT done if successful and update the flow key. */
 731         if (err == NF_ACCEPT)
 732                 ovs_nat_update_key(key, skb, maniptype);
 733
 734         return err;
 735 }
 736 #else /* !CONFIG_NF_NAT_NEEDED */
 737 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 738                       const struct ovs_conntrack_info *info,
 739                       struct sk_buff *skb, struct nf_conn *ct,
 740                       enum ip_conntrack_info ctinfo)
 741 {
 742         return NF_ACCEPT;
 743 }
 744 #endif
 745
 746 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 747  * not done already.  Update key with new CT state after passing the packet
 748  * through conntrack.
 749  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 750  * set to NULL and 0 will be returned.
 751  */
 752 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 753                            const struct ovs_conntrack_info *info,
 754                            struct sk_buff *skb)
 755 {
 756         /* If we are recirculating packets to match on conntrack fields and
 757          * committing with a separate conntrack action,  then we don't need to
 758          * actually run the packet through conntrack twice unless it's for a
 759          * different zone.
 760          */
 761         bool cached = skb_nfct_cached(net, key, info, skb);
 762         enum ip_conntrack_info ctinfo;
 763         struct nf_conn *ct;
 764
 765         if (!cached) {
 766                 struct nf_conn *tmpl = info->ct;
 767                 int err;
 768
 769                 /* Associate skb with specified zone. */
 770                 if (tmpl) {
 771                         if (skb->nfct)
 772                                 nf_conntrack_put(skb->nfct);
 773                         nf_conntrack_get(&tmpl->ct_general);
 774                         skb->nfct = &tmpl->ct_general;
 775                         skb->nfctinfo = IP_CT_NEW;
 776                 }
 777
 778                 /* Repeat if requested, see nf_iterate(). */
 779                 do {
 780                         err = nf_conntrack_in(net, info->family,
 781                                               NF_INET_FORWARD, skb);
 782                 } while (err == NF_REPEAT);
 783
 784                 if (err != NF_ACCEPT)
 785                         return -ENOENT;
 786
 787                 /* Clear CT state NAT flags to mark that we have not yet done
 788                  * NAT after the nf_conntrack_in() call.  We can actually clear
 789                  * the whole state, as it will be re-initialized below.
 790                  */
 791                 key->ct.state = 0;
 792
 793                 /* Update the key, but keep the NAT flags. */
 794                 ovs_ct_update_key(skb, info, key, true, true);
 795         }
 796
 797         ct = nf_ct_get(skb, &ctinfo);
 798         if (ct) {
 799                 /* Packets starting a new connection must be NATted before the
 800                  * helper, so that the helper knows about the NAT.  We enforce
 801                  * this by delaying both NAT and helper calls for unconfirmed
 802                  * connections until the committing CT action.  For later
 803                  * packets NAT and Helper may be called in either order.
 804                  *
 805                  * NAT will be done only if the CT action has NAT, and only
 806                  * once per packet (per zone), as guarded by the NAT bits in
 807                  * the key->ct.state.
 808                  */
 809                 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
 810                     (nf_ct_is_confirmed(ct) || info->commit) &&
 811                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 812                         return -EINVAL;
 813                 }
 814
 815                 /* Userspace may decide to perform a ct lookup without a helper
 816                  * specified followed by a (recirculate and) commit with one.
 817                  * Therefore, for unconfirmed connections which we will commit,
 818                  * we need to attach the helper here.
 819                  */
 820                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 821                     info->helper && !nfct_help(ct)) {
 822                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 823                                                             GFP_ATOMIC);
 824                         if (err)
 825                                 return err;
 826                 }
 827
 828                 /* Call the helper only if:
 829                  * - nf_conntrack_in() was executed above ("!cached") for a
 830                  *   confirmed connection, or
 831                  * - When committing an unconfirmed connection.
 832                  */
 833                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 834                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 835                         return -EINVAL;
 836                 }
 837         }
 838
 839         return 0;
 840 }
 841
 842 /* Lookup connection and read fields into key. */
 843 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 844                          const struct ovs_conntrack_info *info,
 845                          struct sk_buff *skb)
 846 {
 847         struct nf_conntrack_expect *exp;
 848
 849         /* If we pass an expected packet through nf_conntrack_in() the
 850          * expectation is typically removed, but the packet could still be
 851          * lost in upcall processing.  To prevent this from happening we
 852          * perform an explicit expectation lookup.  Expected connections are
 853          * always new, and will be passed through conntrack only when they are
 854          * committed, as it is OK to remove the expectation at that time.
 855          */
 856         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 857         if (exp) {
 858                 u8 state;
 859
 860                 /* NOTE: New connections are NATted and Helped only when
 861                  * committed, so we are not calling into NAT here.
 862                  */
 863                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 864                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 865         } else
 866                 return __ovs_ct_lookup(net, key, info, skb);
 867
 868         return 0;
 869 }
 870
 871 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 872 {
 873         size_t i;
 874
 875         for (i = 0; i < sizeof(*labels); i++)
 876                 if (labels->ct_labels[i])
 877                         return true;
 878
 879         return false;
 880 }
 881
 882 /* Lookup connection and confirm if unconfirmed. */
 883 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 884                          const struct ovs_conntrack_info *info,
 885                          struct sk_buff *skb)
 886 {
 887         int err;
 888
 889         err = __ovs_ct_lookup(net, key, info, skb);
 890         if (err)
 891                 return err;
 892
 893         /* Apply changes before confirming the connection so that the initial
 894          * conntrack NEW netlink event carries the values given in the CT
 895          * action.
 896          */
 897         if (info->mark.mask) {
 898                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 899                                       info->mark.mask);
 900                 if (err)
 901                         return err;
 902         }
 903         if (labels_nonzero(&info->labels.mask)) {
 904                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 905                                         &info->labels.mask);
 906                 if (err)
 907                         return err;
 908         }
 909         /* This will take care of sending queued events even if the connection
 910          * is already confirmed.
 911          */
 912         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 913                 return -EINVAL;
 914
 915         return 0;
 916 }
 917
 918 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 919  * value if 'skb' is freed.
 920  */
 921 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 922                    struct sw_flow_key *key,
 923                    const struct ovs_conntrack_info *info)
 924 {
 925         int nh_ofs;
 926         int err;
 927
 928         /* The conntrack module expects to be working at L3. */
 929         nh_ofs = skb_network_offset(skb);
 930         skb_pull(skb, nh_ofs);
 931
 932         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 933                 err = handle_fragments(net, key, info->zone.id, skb);
 934                 if (err)
 935                         return err;
 936         }
 937
 938         if (info->commit)
 939                 err = ovs_ct_commit(net, key, info, skb);
 940         else
 941                 err = ovs_ct_lookup(net, key, info, skb);
 942
 943         skb_push(skb, nh_ofs);
 944         if (err)
 945                 kfree_skb(skb);
 946         return err;
 947 }
 948
 949 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 950                              const struct sw_flow_key *key, bool log)
 951 {
 952         struct nf_conntrack_helper *helper;
 953         struct nf_conn_help *help;
 954
 955         helper = nf_conntrack_helper_try_module_get(name, info->family,
 956                                                     key->ip.proto);
 957         if (!helper) {
 958                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 959                 return -EINVAL;
 960         }
 961
 962         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 963         if (!help) {
 964                 module_put(helper->me);
 965                 return -ENOMEM;
 966         }
 967
 968         rcu_assign_pointer(help->helper, helper);
 969         info->helper = helper;
 970         return 0;
 971 }
 972
 973 #ifdef CONFIG_NF_NAT_NEEDED
 974 static int parse_nat(const struct nlattr *attr,
 975                      struct ovs_conntrack_info *info, bool log)
 976 {
 977         struct nlattr *a;
 978         int rem;
 979         bool have_ip_max = false;
 980         bool have_proto_max = false;
 981         bool ip_vers = (info->family == NFPROTO_IPV6);
 982
 983         nla_for_each_nested(a, attr, rem) {
 984                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
 985                         [OVS_NAT_ATTR_SRC] = {0, 0},
 986                         [OVS_NAT_ATTR_DST] = {0, 0},
 987                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
 988                                                  sizeof(struct in6_addr)},
 989                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
 990                                                  sizeof(struct in6_addr)},
 991                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
 992                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
 993                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
 994                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
 995                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
 996                 };
 997                 int type = nla_type(a);
 998
 999                 if (type > OVS_NAT_ATTR_MAX) {
1000                         OVS_NLERR(log,
1001                                   "Unknown NAT attribute (type=%d, max=%d).\n",
1002                                   type, OVS_NAT_ATTR_MAX);
1003                         return -EINVAL;
1004                 }
1005
1006                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1007                         OVS_NLERR(log,
1008                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
1009                                   type, nla_len(a),
1010                                   ovs_nat_attr_lens[type][ip_vers]);
1011                         return -EINVAL;
1012                 }
1013
1014                 switch (type) {
1015                 case OVS_NAT_ATTR_SRC:
1016                 case OVS_NAT_ATTR_DST:
1017                         if (info->nat) {
1018                                 OVS_NLERR(log,
1019                                           "Only one type of NAT may be specified.\n"
1020                                           );
1021                                 return -ERANGE;
1022                         }
1023                         info->nat |= OVS_CT_NAT;
1024                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1025                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1026                         break;
1027
1028                 case OVS_NAT_ATTR_IP_MIN:
1029                         nla_memcpy(&info->range.min_addr, a,
1030                                    sizeof(info->range.min_addr));
1031                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1032                         break;
1033
1034                 case OVS_NAT_ATTR_IP_MAX:
1035                         have_ip_max = true;
1036                         nla_memcpy(&info->range.max_addr, a,
1037                                    sizeof(info->range.max_addr));
1038                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1039                         break;
1040
1041                 case OVS_NAT_ATTR_PROTO_MIN:
1042                         info->range.min_proto.all = htons(nla_get_u16(a));
1043                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1044                         break;
1045
1046                 case OVS_NAT_ATTR_PROTO_MAX:
1047                         have_proto_max = true;
1048                         info->range.max_proto.all = htons(nla_get_u16(a));
1049                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1050                         break;
1051
1052                 case OVS_NAT_ATTR_PERSISTENT:
1053                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1054                         break;
1055
1056                 case OVS_NAT_ATTR_PROTO_HASH:
1057                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1058                         break;
1059
1060                 case OVS_NAT_ATTR_PROTO_RANDOM:
1061 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1062                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1063 #else
1064                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1065                         info->random_fully_compat = true;
1066 #endif
1067                         break;
1068
1069                 default:
1070                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1071                         return -EINVAL;
1072                 }
1073         }
1074
1075         if (rem > 0) {
1076                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1077                 return -EINVAL;
1078         }
1079         if (!info->nat) {
1080                 /* Do not allow flags if no type is given. */
1081                 if (info->range.flags) {
1082                         OVS_NLERR(log,
1083                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1084                                   );
1085                         return -EINVAL;
1086                 }
1087                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1088         } else if (!info->commit) {
1089                 OVS_NLERR(log,
1090                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1091                           );
1092                 return -EINVAL;
1093         }
1094         /* Allow missing IP_MAX. */
1095         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1096                 memcpy(&info->range.max_addr, &info->range.min_addr,
1097                        sizeof(info->range.max_addr));
1098         }
1099         /* Allow missing PROTO_MAX. */
1100         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1101             !have_proto_max) {
1102                 info->range.max_proto.all = info->range.min_proto.all;
1103         }
1104         return 0;
1105 }
1106 #endif
1107
1108 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1109         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1110         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1111                                     .maxlen = sizeof(u16) },
1112         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1113                                     .maxlen = sizeof(struct md_mark) },
1114         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1115                                     .maxlen = sizeof(struct md_labels) },
1116         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1117                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1118 #ifdef CONFIG_NF_NAT_NEEDED
1119         /* NAT length is checked when parsing the nested attributes. */
1120         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1121 #endif
1122 };
1123
1124 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1125                     const char **helper, bool log)
1126 {
1127         struct nlattr *a;
1128         int rem;
1129
1130         nla_for_each_nested(a, attr, rem) {
1131                 int type = nla_type(a);
1132                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1133                 int minlen = ovs_ct_attr_lens[type].minlen;
1134
1135                 if (type > OVS_CT_ATTR_MAX) {
1136                         OVS_NLERR(log,
1137                                   "Unknown conntrack attr (type=%d, max=%d)",
1138                                   type, OVS_CT_ATTR_MAX);
1139                         return -EINVAL;
1140                 }
1141                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1142                         OVS_NLERR(log,
1143                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1144                                   type, nla_len(a), maxlen);
1145                         return -EINVAL;
1146                 }
1147
1148                 switch (type) {
1149                 case OVS_CT_ATTR_COMMIT:
1150                         info->commit = true;
1151                         break;
1152 #ifdef CONFIG_NF_CONNTRACK_ZONES
1153                 case OVS_CT_ATTR_ZONE:
1154                         info->zone.id = nla_get_u16(a);
1155                         break;
1156 #endif
1157 #ifdef CONFIG_NF_CONNTRACK_MARK
1158                 case OVS_CT_ATTR_MARK: {
1159                         struct md_mark *mark = nla_data(a);
1160
1161                         if (!mark->mask) {
1162                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1163                                 return -EINVAL;
1164                         }
1165                         info->mark = *mark;
1166                         break;
1167                 }
1168 #endif
1169 #ifdef CONFIG_NF_CONNTRACK_LABELS
1170                 case OVS_CT_ATTR_LABELS: {
1171                         struct md_labels *labels = nla_data(a);
1172
1173                         if (!labels_nonzero(&labels->mask)) {
1174                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1175                                 return -EINVAL;
1176                         }
1177                         info->labels = *labels;
1178                         break;
1179                 }
1180 #endif
1181                 case OVS_CT_ATTR_HELPER:
1182                         *helper = nla_data(a);
1183                         if (!memchr(*helper, '\0', nla_len(a))) {
1184                                 OVS_NLERR(log, "Invalid conntrack helper");
1185                                 return -EINVAL;
1186                         }
1187                         break;
1188 #ifdef CONFIG_NF_NAT_NEEDED
1189                 case OVS_CT_ATTR_NAT: {
1190                         int err = parse_nat(a, info, log);
1191
1192                         if (err)
1193                                 return err;
1194                         break;
1195                 }
1196 #endif
1197                 default:
1198                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1199                                   type);
1200                         return -EINVAL;
1201                 }
1202         }
1203
1204 #ifdef CONFIG_NF_CONNTRACK_MARK
1205         if (!info->commit && info->mark.mask) {
1206                 OVS_NLERR(log,
1207                           "Setting conntrack mark requires 'commit' flag.");
1208                 return -EINVAL;
1209         }
1210 #endif
1211 #ifdef CONFIG_NF_CONNTRACK_LABELS
1212         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1213                 OVS_NLERR(log,
1214                           "Setting conntrack labels requires 'commit' flag.");
1215                 return -EINVAL;
1216         }
1217 #endif
1218         if (rem > 0) {
1219                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1220                 return -EINVAL;
1221         }
1222
1223         return 0;
1224 }
1225
1226 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1227 {
1228         if (attr == OVS_KEY_ATTR_CT_STATE)
1229                 return true;
1230         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1231             attr == OVS_KEY_ATTR_CT_ZONE)
1232                 return true;
1233         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1234             attr == OVS_KEY_ATTR_CT_MARK)
1235                 return true;
1236         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1237             attr == OVS_KEY_ATTR_CT_LABELS) {
1238                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1239
1240                 return ovs_net->xt_label;
1241         }
1242
1243         return false;
1244 }
1245
1246 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1247                        const struct sw_flow_key *key,
1248                        struct sw_flow_actions **sfa,  bool log)
1249 {
1250         struct ovs_conntrack_info ct_info;
1251         const char *helper = NULL;
1252         u16 family;
1253         int err;
1254
1255         family = key_to_nfproto(key);
1256         if (family == NFPROTO_UNSPEC) {
1257                 OVS_NLERR(log, "ct family unspecified");
1258                 return -EINVAL;
1259         }
1260
1261         memset(&ct_info, 0, sizeof(ct_info));
1262         ct_info.family = family;
1263
1264         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1265                         NF_CT_DEFAULT_ZONE_DIR, 0);
1266
1267         err = parse_ct(attr, &ct_info, &helper, log);
1268         if (err)
1269                 return err;
1270
1271         /* Set up template for tracking connections in specific zones. */
1272         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1273         if (!ct_info.ct) {
1274                 OVS_NLERR(log, "Failed to allocate conntrack template");
1275                 return -ENOMEM;
1276         }
1277
1278         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1279         nf_conntrack_get(&ct_info.ct->ct_general);
1280
1281         if (helper) {
1282                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1283                 if (err)
1284                         goto err_free_ct;
1285         }
1286
1287         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1288                                  sizeof(ct_info), log);
1289         if (err)
1290                 goto err_free_ct;
1291
1292         return 0;
1293 err_free_ct:
1294         __ovs_ct_free_action(&ct_info);
1295         return err;
1296 }
1297
1298 #ifdef CONFIG_NF_NAT_NEEDED
1299 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1300                                struct sk_buff *skb)
1301 {
1302         struct nlattr *start;
1303
1304         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1305         if (!start)
1306                 return false;
1307
1308         if (info->nat & OVS_CT_SRC_NAT) {
1309                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1310                         return false;
1311         } else if (info->nat & OVS_CT_DST_NAT) {
1312                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1313                         return false;
1314         } else {
1315                 goto out;
1316         }
1317
1318         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1319                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1320                     info->family == NFPROTO_IPV4) {
1321                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1322                                             info->range.min_addr.ip) ||
1323                             (info->range.max_addr.ip
1324                              != info->range.min_addr.ip &&
1325                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1326                                               info->range.max_addr.ip))))
1327                                 return false;
1328                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1329                            info->family == NFPROTO_IPV6) {
1330                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1331                                              &info->range.min_addr.in6) ||
1332                             (memcmp(&info->range.max_addr.in6,
1333                                     &info->range.min_addr.in6,
1334                                     sizeof(info->range.max_addr.in6)) &&
1335                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1336                                                &info->range.max_addr.in6))))
1337                                 return false;
1338                 } else {
1339                         return false;
1340                 }
1341         }
1342         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1343             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1344                          ntohs(info->range.min_proto.all)) ||
1345              (info->range.max_proto.all != info->range.min_proto.all &&
1346               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1347                           ntohs(info->range.max_proto.all)))))
1348                 return false;
1349
1350         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1351             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1352                 return false;
1353         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1354             nla_put_flag(skb, info->random_fully_compat
1355                          ? OVS_NAT_ATTR_PROTO_RANDOM
1356                          : OVS_NAT_ATTR_PROTO_HASH))
1357                 return false;
1358 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1359         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1360             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1361                 return false;
1362 #endif
1363 out:
1364         nla_nest_end(skb, start);
1365
1366         return true;
1367 }
1368 #endif
1369
1370 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1371                           struct sk_buff *skb)
1372 {
1373         struct nlattr *start;
1374
1375         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1376         if (!start)
1377                 return -EMSGSIZE;
1378
1379         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1380                 return -EMSGSIZE;
1381         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1382             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1383                 return -EMSGSIZE;
1384         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1385             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1386                     &ct_info->mark))
1387                 return -EMSGSIZE;
1388         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1389             labels_nonzero(&ct_info->labels.mask) &&
1390             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1391                     &ct_info->labels))
1392                 return -EMSGSIZE;
1393         if (ct_info->helper) {
1394                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1395                                    ct_info->helper->name))
1396                         return -EMSGSIZE;
1397         }
1398 #ifdef CONFIG_NF_NAT_NEEDED
1399         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1400                 return -EMSGSIZE;
1401 #endif
1402         nla_nest_end(skb, start);
1403
1404         return 0;
1405 }
1406
1407 void ovs_ct_free_action(const struct nlattr *a)
1408 {
1409         struct ovs_conntrack_info *ct_info = nla_data(a);
1410
1411         __ovs_ct_free_action(ct_info);
1412 }
1413
1414 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1415 {
1416         if (ct_info->helper)
1417                 module_put(ct_info->helper->me);
1418         if (ct_info->ct)
1419                 nf_ct_tmpl_free(ct_info->ct);
1420 }
1421
1422 void ovs_ct_init(struct net *net)
1423 {
1424         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1425         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1426
1427         if (nf_connlabels_get(net, n_bits - 1)) {
1428                 ovs_net->xt_label = false;
1429                 OVS_NLERR(true, "Failed to set connlabel length");
1430         } else {
1431                 ovs_net->xt_label = true;
1432         }
1433 }
1434
1435 void ovs_ct_exit(struct net *net)
1436 {
1437         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1438
1439         if (ovs_net->xt_label)
1440                 nf_connlabels_put(net);
1441 }
1442
1443 #endif /* CONFIG_NF_CONNTRACK */