net/openvswitch/conntrack.c

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