datapath/conntrack.c

   1 /*
   2  * Copyright (c) 2015 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  */
  13
  14 #include <linux/kconfig.h>
  15 #include <linux/version.h>
  16
  17 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  18
  19 #include <linux/module.h>
  20 #include <linux/openvswitch.h>
  21 #include <linux/tcp.h>
  22 #include <linux/udp.h>
  23 #include <linux/sctp.h>
  24 #include <net/ip.h>
  25 #include <net/netfilter/nf_conntrack_core.h>
  26 #include <net/netfilter/nf_conntrack_helper.h>
  27 #include <net/netfilter/nf_conntrack_labels.h>
  28 #include <net/netfilter/nf_conntrack_seqadj.h>
  29 #include <net/netfilter/nf_conntrack_zones.h>
  30 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  31
  32 #ifdef CONFIG_NF_NAT_NEEDED
  33 #include <linux/netfilter/nf_nat.h>
  34 #include <net/netfilter/nf_nat_core.h>
  35 #include <net/netfilter/nf_nat_l3proto.h>
  36 #endif
  37
  38 #include "datapath.h"
  39 #include "conntrack.h"
  40 #include "flow.h"
  41 #include "flow_netlink.h"
  42 #include "gso.h"
  43
  44 struct ovs_ct_len_tbl {
  45         int maxlen;
  46         int minlen;
  47 };
  48
  49 /* Metadata mark for masked write to conntrack mark */
  50 struct md_mark {
  51         u32 value;
  52         u32 mask;
  53 };
  54
  55 /* Metadata label for masked write to conntrack label. */
  56 struct md_labels {
  57         struct ovs_key_ct_labels value;
  58         struct ovs_key_ct_labels mask;
  59 };
  60
  61 enum ovs_ct_nat {
  62         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  63         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  64         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  65 };
  66
  67 /* Conntrack action context for execution. */
  68 struct ovs_conntrack_info {
  69         struct nf_conntrack_helper *helper;
  70         struct nf_conntrack_zone zone;
  71         struct nf_conn *ct;
  72         u8 commit : 1;
  73         u8 nat : 3;                 /* enum ovs_ct_nat */
  74         u8 random_fully_compat : 1; /* bool */
  75         u16 family;
  76         struct md_mark mark;
  77         struct md_labels labels;
  78 #ifdef CONFIG_NF_NAT_NEEDED
  79         struct nf_nat_range range;  /* Only present for SRC NAT and DST NAT. */
  80 #endif
  81 };
  82
  83 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
  84
  85 static u16 key_to_nfproto(const struct sw_flow_key *key)
  86 {
  87         switch (ntohs(key->eth.type)) {
  88         case ETH_P_IP:
  89                 return NFPROTO_IPV4;
  90         case ETH_P_IPV6:
  91                 return NFPROTO_IPV6;
  92         default:
  93                 return NFPROTO_UNSPEC;
  94         }
  95 }
  96
  97 /* Map SKB connection state into the values used by flow definition. */
  98 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
  99 {
 100         u8 ct_state = OVS_CS_F_TRACKED;
 101
 102         switch (ctinfo) {
 103         case IP_CT_ESTABLISHED_REPLY:
 104         case IP_CT_RELATED_REPLY:
 105                 ct_state |= OVS_CS_F_REPLY_DIR;
 106                 break;
 107         default:
 108                 break;
 109         }
 110
 111         switch (ctinfo) {
 112         case IP_CT_ESTABLISHED:
 113         case IP_CT_ESTABLISHED_REPLY:
 114                 ct_state |= OVS_CS_F_ESTABLISHED;
 115                 break;
 116         case IP_CT_RELATED:
 117         case IP_CT_RELATED_REPLY:
 118                 ct_state |= OVS_CS_F_RELATED;
 119                 break;
 120         case IP_CT_NEW:
 121                 ct_state |= OVS_CS_F_NEW;
 122                 break;
 123         default:
 124                 break;
 125         }
 126
 127         return ct_state;
 128 }
 129
 130 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 131 {
 132 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 133         return ct ? ct->mark : 0;
 134 #else
 135         return 0;
 136 #endif
 137 }
 138
 139 static void ovs_ct_get_labels(const struct nf_conn *ct,
 140                               struct ovs_key_ct_labels *labels)
 141 {
 142         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 143
 144         if (cl) {
 145                 size_t len = cl->words * sizeof(long);
 146
 147                 if (len > OVS_CT_LABELS_LEN)
 148                         len = OVS_CT_LABELS_LEN;
 149                 else if (len < OVS_CT_LABELS_LEN)
 150                         memset(labels, 0, OVS_CT_LABELS_LEN);
 151                 memcpy(labels, cl->bits, len);
 152         } else {
 153                 memset(labels, 0, OVS_CT_LABELS_LEN);
 154         }
 155 }
 156
 157 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 158                                 const struct nf_conntrack_zone *zone,
 159                                 const struct nf_conn *ct)
 160 {
 161         key->ct.state = state;
 162         key->ct.zone = zone->id;
 163         key->ct.mark = ovs_ct_get_mark(ct);
 164         ovs_ct_get_labels(ct, &key->ct.labels);
 165 }
 166
 167 /* Update 'key' based on skb->nfct.  If 'post_ct' is true, then OVS has
 168  * previously sent the packet to conntrack via the ct action.  If
 169  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 170  * initialized from the connection status.
 171  */
 172 static void ovs_ct_update_key(const struct sk_buff *skb,
 173                               const struct ovs_conntrack_info *info,
 174                               struct sw_flow_key *key, bool post_ct,
 175                               bool keep_nat_flags)
 176 {
 177         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 178         enum ip_conntrack_info ctinfo;
 179         struct nf_conn *ct;
 180         u8 state = 0;
 181
 182         ct = nf_ct_get(skb, &ctinfo);
 183         if (ct) {
 184                 state = ovs_ct_get_state(ctinfo);
 185                 /* All unconfirmed entries are NEW connections. */
 186                 if (!nf_ct_is_confirmed(ct))
 187                         state |= OVS_CS_F_NEW;
 188                 /* OVS persists the related flag for the duration of the
 189                  * connection.
 190                  */
 191                 if (ct->master)
 192                         state |= OVS_CS_F_RELATED;
 193                 if (keep_nat_flags) {
 194                         state |= key->ct.state & OVS_CS_F_NAT_MASK;
 195                 } else {
 196                         if (ct->status & IPS_SRC_NAT)
 197                                 state |= OVS_CS_F_SRC_NAT;
 198                         if (ct->status & IPS_DST_NAT)
 199                                 state |= OVS_CS_F_DST_NAT;
 200                 }
 201                 zone = nf_ct_zone(ct);
 202         } else if (post_ct) {
 203                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 204                 if (info)
 205                         zone = &info->zone;
 206         }
 207         __ovs_ct_update_key(key, state, zone, ct);
 208 }
 209
 210 /* This is called to initialize CT key fields possibly coming in from the local
 211  * stack.
 212  */
 213 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 214 {
 215         ovs_ct_update_key(skb, NULL, key, false, false);
 216 }
 217
 218 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 219 {
 220         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 221                 return -EMSGSIZE;
 222
 223         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 224             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 225                 return -EMSGSIZE;
 226
 227         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 228             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 229                 return -EMSGSIZE;
 230
 231         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 232             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 233                     &key->ct.labels))
 234                 return -EMSGSIZE;
 235
 236         return 0;
 237 }
 238
 239 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 240                            u32 ct_mark, u32 mask)
 241 {
 242 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 243         enum ip_conntrack_info ctinfo;
 244         struct nf_conn *ct;
 245         u32 new_mark;
 246
 247         /* The connection could be invalid, in which case set_mark is no-op. */
 248         ct = nf_ct_get(skb, &ctinfo);
 249         if (!ct)
 250                 return 0;
 251
 252         new_mark = ct_mark | (ct->mark & ~(mask));
 253         if (ct->mark != new_mark) {
 254                 ct->mark = new_mark;
 255                 nf_conntrack_event_cache(IPCT_MARK, ct);
 256                 key->ct.mark = new_mark;
 257         }
 258
 259         return 0;
 260 #else
 261         return -ENOTSUPP;
 262 #endif
 263 }
 264
 265 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 266                              const struct ovs_key_ct_labels *labels,
 267                              const struct ovs_key_ct_labels *mask)
 268 {
 269         enum ip_conntrack_info ctinfo;
 270         struct nf_conn_labels *cl;
 271         struct nf_conn *ct;
 272         int err;
 273
 274         /* The connection could be invalid, in which case set_label is no-op.*/
 275         ct = nf_ct_get(skb, &ctinfo);
 276         if (!ct)
 277                 return 0;
 278
 279         cl = nf_ct_labels_find(ct);
 280         if (!cl) {
 281                 nf_ct_labels_ext_add(ct);
 282                 cl = nf_ct_labels_find(ct);
 283         }
 284         if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
 285                 return -ENOSPC;
 286
 287         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 288                                     OVS_CT_LABELS_LEN / sizeof(u32));
 289         if (err)
 290                 return err;
 291
 292         ovs_ct_get_labels(ct, &key->ct.labels);
 293         return 0;
 294 }
 295
 296 /* 'skb' should already be pulled to nh_ofs. */
 297 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 298 {
 299         const struct nf_conntrack_helper *helper;
 300         const struct nf_conn_help *help;
 301         enum ip_conntrack_info ctinfo;
 302         unsigned int protoff;
 303         struct nf_conn *ct;
 304         u8 nexthdr;
 305         int err;
 306
 307         ct = nf_ct_get(skb, &ctinfo);
 308         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 309                 return NF_ACCEPT;
 310
 311         help = nfct_help(ct);
 312         if (!help)
 313                 return NF_ACCEPT;
 314
 315         helper = rcu_dereference(help->helper);
 316         if (!helper)
 317                 return NF_ACCEPT;
 318
 319         switch (proto) {
 320         case NFPROTO_IPV4:
 321                 protoff = ip_hdrlen(skb);
 322                 break;
 323         case NFPROTO_IPV6: {
 324                 __be16 frag_off;
 325                 int ofs;
 326
 327                 nexthdr = ipv6_hdr(skb)->nexthdr;
 328                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 329                                        &frag_off);
 330                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 331                         pr_debug("proto header not found\n");
 332                         return NF_ACCEPT;
 333                 }
 334                 protoff = ofs;
 335                 break;
 336         }
 337         default:
 338                 WARN_ONCE(1, "helper invoked on non-IP family!");
 339                 return NF_DROP;
 340         }
 341
 342 #if LINUX_VERSION_CODE < KERNEL_VERSION(4,6,0)
 343         /* Linux 4.5 and older depend on skb_dst being set when recalculating
 344          * checksums after NAT helper has mangled TCP or UDP packet payload.
 345          * This dependency is avoided when skb is CHECKSUM_PARTIAL or when UDP
 346          * has no checksum.
 347          *
 348          * The dependency is not triggered when the main NAT code updates
 349          * checksums after translating the IP header (address, port), so this
 350          * fix only needs to be executed on packets that are both being NATted
 351          * and that have a helper assigned.
 352          */
 353         if (ct->status & IPS_NAT_MASK && skb->ip_summed != CHECKSUM_PARTIAL) {
 354                 u8 ipproto = (proto == NFPROTO_IPV4)
 355                         ? ip_hdr(skb)->protocol : nexthdr;
 356                 u16 offset = 0;
 357
 358                 switch (ipproto) {
 359                 case IPPROTO_TCP:
 360                         offset = offsetof(struct tcphdr, check);
 361                         break;
 362                 case IPPROTO_UDP:
 363                         /* Skip if no csum. */
 364                         if (udp_hdr(skb)->check)
 365                                 offset = offsetof(struct udphdr, check);
 366                         break;
 367                 }
 368                 if (offset) {
 369                         if (unlikely(!pskb_may_pull(skb, protoff + offset + 2)))
 370                                 return NF_DROP;
 371
 372                         skb->csum_start = skb_headroom(skb) + protoff;
 373                         skb->csum_offset = offset;
 374                         skb->ip_summed = CHECKSUM_PARTIAL;
 375                 }
 376         }
 377 #endif
 378         err = helper->help(skb, protoff, ct, ctinfo);
 379         if (err != NF_ACCEPT)
 380                 return err;
 381
 382         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 383          * FTP with NAT) adusting the TCP payload size when mangling IP
 384          * addresses and/or port numbers in the text-based control connection.
 385          */
 386         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 387             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 388                 return NF_DROP;
 389         return NF_ACCEPT;
 390 }
 391
 392 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 393  * value if 'skb' is freed.
 394  */
 395 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 396                             u16 zone, struct sk_buff *skb)
 397 {
 398         struct ovs_gso_cb ovs_cb = *OVS_GSO_CB(skb);
 399         int err;
 400
 401         if (!skb->dev) {
 402                 OVS_NLERR(true, "%s: skb has no dev; dropping", __func__);
 403                 return -EINVAL;
 404         }
 405
 406         if (key->eth.type == htons(ETH_P_IP)) {
 407                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 408
 409                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 410                 err = ip_defrag(net, skb, user);
 411                 if (err)
 412                         return err;
 413
 414                 ovs_cb.dp_cb.mru = IPCB(skb)->frag_max_size;
 415 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 416         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 417                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 418
 419                 skb_orphan(skb);
 420                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 421                 err = nf_ct_frag6_gather(net, skb, user);
 422                 if (err)
 423                         return err;
 424
 425                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 426                 ovs_cb.dp_cb.mru = IP6CB(skb)->frag_max_size;
 427 #endif /* IP frag support */
 428         } else {
 429                 kfree_skb(skb);
 430                 return -EPFNOSUPPORT;
 431         }
 432
 433         key->ip.frag = OVS_FRAG_TYPE_NONE;
 434         skb_clear_hash(skb);
 435         skb->ignore_df = 1;
 436         *OVS_GSO_CB(skb) = ovs_cb;
 437
 438         return 0;
 439 }
 440
 441 static struct nf_conntrack_expect *
 442 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 443                    u16 proto, const struct sk_buff *skb)
 444 {
 445         struct nf_conntrack_tuple tuple;
 446
 447         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 448                 return NULL;
 449         return __nf_ct_expect_find(net, zone, &tuple);
 450 }
 451
 452 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 453 static enum ip_conntrack_info
 454 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 455 {
 456         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 457
 458         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 459                 return IP_CT_ESTABLISHED_REPLY;
 460         /* Once we've had two way comms, always ESTABLISHED. */
 461         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 462                 return IP_CT_ESTABLISHED;
 463         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 464                 return IP_CT_RELATED;
 465         return IP_CT_NEW;
 466 }
 467
 468 /* Find an existing connection which this packet belongs to without
 469  * re-attributing statistics or modifying the connection state.  This allows an
 470  * skb->nfct lost due to an upcall to be recovered during actions execution.
 471  *
 472  * Must be called with rcu_read_lock.
 473  *
 474  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 475  * connection.  Returns NULL if there is no existing entry.
 476  */
 477 static struct nf_conn *
 478 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 479                      u8 l3num, struct sk_buff *skb)
 480 {
 481         struct nf_conntrack_l3proto *l3proto;
 482         struct nf_conntrack_l4proto *l4proto;
 483         struct nf_conntrack_tuple tuple;
 484         struct nf_conntrack_tuple_hash *h;
 485         enum ip_conntrack_info ctinfo;
 486         struct nf_conn *ct;
 487         unsigned int dataoff;
 488         u8 protonum;
 489
 490         l3proto = __nf_ct_l3proto_find(l3num);
 491         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 492                                  &protonum) <= 0) {
 493                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 494                 return NULL;
 495         }
 496         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 497         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 498                              protonum, net, &tuple, l3proto, l4proto)) {
 499                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 500                 return NULL;
 501         }
 502
 503         /* look for tuple match */
 504         h = nf_conntrack_find_get(net, zone, &tuple);
 505         if (!h)
 506                 return NULL;   /* Not found. */
 507
 508         ct = nf_ct_tuplehash_to_ctrack(h);
 509
 510         ctinfo = ovs_ct_get_info(h);
 511         if (ctinfo == IP_CT_NEW) {
 512                 /* This should not happen. */
 513                 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct);
 514         }
 515         skb->nfct = &ct->ct_general;
 516         skb->nfctinfo = ctinfo;
 517         return ct;
 518 }
 519
 520 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 521 static bool skb_nfct_cached(struct net *net,
 522                             const struct sw_flow_key *key,
 523                             const struct ovs_conntrack_info *info,
 524                             struct sk_buff *skb)
 525 {
 526         enum ip_conntrack_info ctinfo;
 527         struct nf_conn *ct;
 528
 529         ct = nf_ct_get(skb, &ctinfo);
 530         /* If no ct, check if we have evidence that an existing conntrack entry
 531          * might be found for this skb.  This happens when we lose a skb->nfct
 532          * due to an upcall.  If the connection was not confirmed, it is not
 533          * cached and needs to be run through conntrack again.
 534          */
 535         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 536             !(key->ct.state & OVS_CS_F_INVALID) &&
 537             key->ct.zone == info->zone.id)
 538                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 539         if (!ct)
 540                 return false;
 541         if (!net_eq(net, read_pnet(&ct->ct_net)))
 542                 return false;
 543         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 544                 return false;
 545         if (info->helper) {
 546                 struct nf_conn_help *help;
 547
 548                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 549                 if (help && rcu_access_pointer(help->helper) != info->helper)
 550                         return false;
 551         }
 552
 553         return true;
 554 }
 555
 556 #ifdef CONFIG_NF_NAT_NEEDED
 557 /* Modelled after nf_nat_ipv[46]_fn().
 558  * range is only used for new, uninitialized NAT state.
 559  * Returns either NF_ACCEPT or NF_DROP.
 560  */
 561 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 562                               enum ip_conntrack_info ctinfo,
 563                               const struct nf_nat_range *range,
 564                               enum nf_nat_manip_type maniptype)
 565 {
 566         int hooknum, nh_off, err = NF_ACCEPT;
 567
 568         nh_off = skb_network_offset(skb);
 569         skb_pull(skb, nh_off);
 570
 571         /* See HOOK2MANIP(). */
 572         if (maniptype == NF_NAT_MANIP_SRC)
 573                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 574         else
 575                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 576
 577         switch (ctinfo) {
 578         case IP_CT_RELATED:
 579         case IP_CT_RELATED_REPLY:
 580                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 581                     skb->protocol == htons(ETH_P_IP) &&
 582                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 583                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 584                                                            hooknum))
 585                                 err = NF_DROP;
 586                         goto push;
 587                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 588                            skb->protocol == htons(ETH_P_IPV6)) {
 589                         __be16 frag_off;
 590                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 591                         int hdrlen = ipv6_skip_exthdr(skb,
 592                                                       sizeof(struct ipv6hdr),
 593                                                       &nexthdr, &frag_off);
 594
 595                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 596                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 597                                                                      ctinfo,
 598                                                                      hooknum,
 599                                                                      hdrlen))
 600                                         err = NF_DROP;
 601                                 goto push;
 602                         }
 603                 }
 604                 /* Non-ICMP, fall thru to initialize if needed. */
 605         case IP_CT_NEW:
 606                 /* Seen it before?  This can happen for loopback, retrans,
 607                  * or local packets.
 608                  */
 609                 if (!nf_nat_initialized(ct, maniptype)) {
 610                         /* Initialize according to the NAT action. */
 611                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 612                                 /* Action is set up to establish a new
 613                                  * mapping.
 614                                  */
 615                                 ? nf_nat_setup_info(ct, range, maniptype)
 616                                 : nf_nat_alloc_null_binding(ct, hooknum);
 617                         if (err != NF_ACCEPT)
 618                                 goto push;
 619                 }
 620                 break;
 621
 622         case IP_CT_ESTABLISHED:
 623         case IP_CT_ESTABLISHED_REPLY:
 624                 break;
 625
 626         default:
 627                 err = NF_DROP;
 628                 goto push;
 629         }
 630
 631         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 632 push:
 633         skb_push(skb, nh_off);
 634
 635         return err;
 636 }
 637
 638 static void ovs_nat_update_key(struct sw_flow_key *key,
 639                                const struct sk_buff *skb,
 640                                enum nf_nat_manip_type maniptype)
 641 {
 642         if (maniptype == NF_NAT_MANIP_SRC) {
 643                 __be16 src;
 644
 645                 key->ct.state |= OVS_CS_F_SRC_NAT;
 646                 if (key->eth.type == htons(ETH_P_IP))
 647                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 648                 else if (key->eth.type == htons(ETH_P_IPV6))
 649                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 650                                sizeof(key->ipv6.addr.src));
 651                 else
 652                         return;
 653
 654                 if (key->ip.proto == IPPROTO_UDP)
 655                         src = udp_hdr(skb)->source;
 656                 else if (key->ip.proto == IPPROTO_TCP)
 657                         src = tcp_hdr(skb)->source;
 658                 else if (key->ip.proto == IPPROTO_SCTP)
 659                         src = sctp_hdr(skb)->source;
 660                 else
 661                         return;
 662
 663                 key->tp.src = src;
 664         } else {
 665                 __be16 dst;
 666
 667                 key->ct.state |= OVS_CS_F_DST_NAT;
 668                 if (key->eth.type == htons(ETH_P_IP))
 669                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 670                 else if (key->eth.type == htons(ETH_P_IPV6))
 671                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 672                                sizeof(key->ipv6.addr.dst));
 673                 else
 674                         return;
 675
 676                 if (key->ip.proto == IPPROTO_UDP)
 677                         dst = udp_hdr(skb)->dest;
 678                 else if (key->ip.proto == IPPROTO_TCP)
 679                         dst = tcp_hdr(skb)->dest;
 680                 else if (key->ip.proto == IPPROTO_SCTP)
 681                         dst = sctp_hdr(skb)->dest;
 682                 else
 683                         return;
 684
 685                 key->tp.dst = dst;
 686         }
 687 }
 688
 689 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 690 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 691                       const struct ovs_conntrack_info *info,
 692                       struct sk_buff *skb, struct nf_conn *ct,
 693                       enum ip_conntrack_info ctinfo)
 694 {
 695         enum nf_nat_manip_type maniptype;
 696         int err;
 697
 698         if (nf_ct_is_untracked(ct)) {
 699                 /* A NAT action may only be performed on tracked packets. */
 700                 return NF_ACCEPT;
 701         }
 702
 703         /* Add NAT extension if not confirmed yet. */
 704         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 705                 return NF_ACCEPT;   /* Can't NAT. */
 706
 707         /* Determine NAT type.
 708          * Check if the NAT type can be deduced from the tracked connection.
 709          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 710          * when committing.
 711          */
 712         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 713             ct->status & IPS_NAT_MASK &&
 714             (ctinfo != IP_CT_RELATED || info->commit)) {
 715                 /* NAT an established or related connection like before. */
 716                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 717                         /* This is the REPLY direction for a connection
 718                          * for which NAT was applied in the forward
 719                          * direction.  Do the reverse NAT.
 720                          */
 721                         maniptype = ct->status & IPS_SRC_NAT
 722                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 723                 else
 724                         maniptype = ct->status & IPS_SRC_NAT
 725                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 726         } else if (info->nat & OVS_CT_SRC_NAT) {
 727                 maniptype = NF_NAT_MANIP_SRC;
 728         } else if (info->nat & OVS_CT_DST_NAT) {
 729                 maniptype = NF_NAT_MANIP_DST;
 730         } else {
 731                 return NF_ACCEPT; /* Connection is not NATed. */
 732         }
 733         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 734
 735         /* Mark NAT done if successful and update the flow key. */
 736         if (err == NF_ACCEPT)
 737                 ovs_nat_update_key(key, skb, maniptype);
 738
 739         return err;
 740 }
 741 #else /* !CONFIG_NF_NAT_NEEDED */
 742 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 743                       const struct ovs_conntrack_info *info,
 744                       struct sk_buff *skb, struct nf_conn *ct,
 745                       enum ip_conntrack_info ctinfo)
 746 {
 747         return NF_ACCEPT;
 748 }
 749 #endif
 750
 751 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 752  * not done already.  Update key with new CT state after passing the packet
 753  * through conntrack.
 754  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 755  * set to NULL and 0 will be returned.
 756  */
 757 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 758                            const struct ovs_conntrack_info *info,
 759                            struct sk_buff *skb)
 760 {
 761         /* If we are recirculating packets to match on conntrack fields and
 762          * committing with a separate conntrack action,  then we don't need to
 763          * actually run the packet through conntrack twice unless it's for a
 764          * different zone.
 765          */
 766         bool cached = skb_nfct_cached(net, key, info, skb);
 767         enum ip_conntrack_info ctinfo;
 768         struct nf_conn *ct;
 769
 770         if (!cached) {
 771                 struct nf_conn *tmpl = info->ct;
 772                 int err;
 773
 774                 /* Associate skb with specified zone. */
 775                 if (tmpl) {
 776                         if (skb->nfct)
 777                                 nf_conntrack_put(skb->nfct);
 778                         nf_conntrack_get(&tmpl->ct_general);
 779                         skb->nfct = &tmpl->ct_general;
 780                         skb->nfctinfo = IP_CT_NEW;
 781                 }
 782
 783                 /* Repeat if requested, see nf_iterate(). */
 784                 do {
 785                         err = nf_conntrack_in(net, info->family,
 786                                               NF_INET_FORWARD, skb);
 787                 } while (err == NF_REPEAT);
 788
 789                 if (err != NF_ACCEPT)
 790                         return -ENOENT;
 791
 792                 /* Clear CT state NAT flags to mark that we have not yet done
 793                  * NAT after the nf_conntrack_in() call.  We can actually clear
 794                  * the whole state, as it will be re-initialized below.
 795                  */
 796                 key->ct.state = 0;
 797
 798                 /* Update the key, but keep the NAT flags. */
 799                 ovs_ct_update_key(skb, info, key, true, true);
 800         }
 801
 802         ct = nf_ct_get(skb, &ctinfo);
 803         if (ct) {
 804                 /* Packets starting a new connection must be NATted before the
 805                  * helper, so that the helper knows about the NAT.  We enforce
 806                  * this by delaying both NAT and helper calls for unconfirmed
 807                  * connections until the committing CT action.  For later
 808                  * packets NAT and Helper may be called in either order.
 809                  *
 810                  * NAT will be done only if the CT action has NAT, and only
 811                  * once per packet (per zone), as guarded by the NAT bits in
 812                  * the key->ct.state.
 813                  */
 814                 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
 815                     (nf_ct_is_confirmed(ct) || info->commit) &&
 816                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 817                         return -EINVAL;
 818                 }
 819
 820                 /* Userspace may decide to perform a ct lookup without a helper
 821                  * specified followed by a (recirculate and) commit with one.
 822                  * Therefore, for unconfirmed connections which we will commit,
 823                  * we need to attach the helper here.
 824                  */
 825                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 826                     info->helper && !nfct_help(ct)) {
 827                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 828                                                             GFP_ATOMIC);
 829                         if (err)
 830                                 return err;
 831                 }
 832
 833                 /* Call the helper only if:
 834                  * - nf_conntrack_in() was executed above ("!cached") for a
 835                  *   confirmed connection, or
 836                  * - When committing an unconfirmed connection.
 837                  */
 838                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 839                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 840                         return -EINVAL;
 841                 }
 842         }
 843
 844         return 0;
 845 }
 846
 847 /* Lookup connection and read fields into key. */
 848 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 849                          const struct ovs_conntrack_info *info,
 850                          struct sk_buff *skb)
 851 {
 852         struct nf_conntrack_expect *exp;
 853
 854         /* If we pass an expected packet through nf_conntrack_in() the
 855          * expectation is typically removed, but the packet could still be
 856          * lost in upcall processing.  To prevent this from happening we
 857          * perform an explicit expectation lookup.  Expected connections are
 858          * always new, and will be passed through conntrack only when they are
 859          * committed, as it is OK to remove the expectation at that time.
 860          */
 861         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 862         if (exp) {
 863                 u8 state;
 864
 865                 /* NOTE: New connections are NATted and Helped only when
 866                  * committed, so we are not calling into NAT here.
 867                  */
 868                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 869                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 870         } else
 871                 return __ovs_ct_lookup(net, key, info, skb);
 872
 873         return 0;
 874 }
 875
 876 /* Lookup connection and confirm if unconfirmed. */
 877 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 878                          const struct ovs_conntrack_info *info,
 879                          struct sk_buff *skb)
 880 {
 881         int err;
 882
 883         err = __ovs_ct_lookup(net, key, info, skb);
 884         if (err)
 885                 return err;
 886         /* This is a no-op if the connection has already been confirmed. */
 887         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 888                 return -EINVAL;
 889
 890         return 0;
 891 }
 892
 893 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 894 {
 895         size_t i;
 896
 897         for (i = 0; i < sizeof(*labels); i++)
 898                 if (labels->ct_labels[i])
 899                         return true;
 900
 901         return false;
 902 }
 903
 904 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 905  * value if 'skb' is freed.
 906  */
 907 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 908                    struct sw_flow_key *key,
 909                    const struct ovs_conntrack_info *info)
 910 {
 911         int nh_ofs;
 912         int err;
 913
 914         /* The conntrack module expects to be working at L3. */
 915         nh_ofs = skb_network_offset(skb);
 916         skb_pull(skb, nh_ofs);
 917
 918         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 919                 err = handle_fragments(net, key, info->zone.id, skb);
 920                 if (err)
 921                         return err;
 922         }
 923
 924         if (info->commit)
 925                 err = ovs_ct_commit(net, key, info, skb);
 926         else
 927                 err = ovs_ct_lookup(net, key, info, skb);
 928         if (err)
 929                 goto err;
 930
 931         if (info->mark.mask) {
 932                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 933                                       info->mark.mask);
 934                 if (err)
 935                         goto err;
 936         }
 937         if (labels_nonzero(&info->labels.mask))
 938                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 939                                         &info->labels.mask);
 940 err:
 941         skb_push(skb, nh_ofs);
 942         if (err)
 943                 kfree_skb(skb);
 944         return err;
 945 }
 946
 947 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 948                              const struct sw_flow_key *key, bool log)
 949 {
 950         struct nf_conntrack_helper *helper;
 951         struct nf_conn_help *help;
 952
 953         helper = nf_conntrack_helper_try_module_get(name, info->family,
 954                                                     key->ip.proto);
 955         if (!helper) {
 956                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 957                 return -EINVAL;
 958         }
 959
 960         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 961         if (!help) {
 962                 module_put(helper->me);
 963                 return -ENOMEM;
 964         }
 965
 966         rcu_assign_pointer(help->helper, helper);
 967         info->helper = helper;
 968         return 0;
 969 }
 970
 971 #ifdef CONFIG_NF_NAT_NEEDED
 972 static int parse_nat(const struct nlattr *attr,
 973                      struct ovs_conntrack_info *info, bool log)
 974 {
 975         struct nlattr *a;
 976         int rem;
 977         bool have_ip_max = false;
 978         bool have_proto_max = false;
 979         bool ip_vers = (info->family == NFPROTO_IPV6);
 980
 981         nla_for_each_nested(a, attr, rem) {
 982                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
 983                         [OVS_NAT_ATTR_SRC] = {0, 0},
 984                         [OVS_NAT_ATTR_DST] = {0, 0},
 985                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
 986                                                  sizeof(struct in6_addr)},
 987                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
 988                                                  sizeof(struct in6_addr)},
 989                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
 990                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
 991                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
 992                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
 993                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
 994                 };
 995                 int type = nla_type(a);
 996
 997                 if (type > OVS_NAT_ATTR_MAX) {
 998                         OVS_NLERR(log,
 999                                   "Unknown NAT attribute (type=%d, max=%d).\n",
1000                                   type, OVS_NAT_ATTR_MAX);
1001                         return -EINVAL;
1002                 }
1003
1004                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1005                         OVS_NLERR(log,
1006                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
1007                                   type, nla_len(a),
1008                                   ovs_nat_attr_lens[type][ip_vers]);
1009                         return -EINVAL;
1010                 }
1011
1012                 switch (type) {
1013                 case OVS_NAT_ATTR_SRC:
1014                 case OVS_NAT_ATTR_DST:
1015                         if (info->nat) {
1016                                 OVS_NLERR(log,
1017                                           "Only one type of NAT may be specified.\n"
1018                                           );
1019                                 return -ERANGE;
1020                         }
1021                         info->nat |= OVS_CT_NAT;
1022                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1023                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1024                         break;
1025
1026                 case OVS_NAT_ATTR_IP_MIN:
1027                         nla_memcpy(&info->range.min_addr, a,
1028                                    sizeof(info->range.min_addr));
1029                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1030                         break;
1031
1032                 case OVS_NAT_ATTR_IP_MAX:
1033                         have_ip_max = true;
1034                         nla_memcpy(&info->range.max_addr, a,
1035                                    sizeof(info->range.max_addr));
1036                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1037                         break;
1038
1039                 case OVS_NAT_ATTR_PROTO_MIN:
1040                         info->range.min_proto.all = htons(nla_get_u16(a));
1041                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1042                         break;
1043
1044                 case OVS_NAT_ATTR_PROTO_MAX:
1045                         have_proto_max = true;
1046                         info->range.max_proto.all = htons(nla_get_u16(a));
1047                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1048                         break;
1049
1050                 case OVS_NAT_ATTR_PERSISTENT:
1051                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1052                         break;
1053
1054                 case OVS_NAT_ATTR_PROTO_HASH:
1055                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1056                         break;
1057
1058                 case OVS_NAT_ATTR_PROTO_RANDOM:
1059 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1060                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1061 #else
1062                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1063                         info->random_fully_compat = true;
1064 #endif
1065                         break;
1066
1067                 default:
1068                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1069                         return -EINVAL;
1070                 }
1071         }
1072
1073         if (rem > 0) {
1074                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1075                 return -EINVAL;
1076         }
1077         if (!info->nat) {
1078                 /* Do not allow flags if no type is given. */
1079                 if (info->range.flags) {
1080                         OVS_NLERR(log,
1081                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1082                                   );
1083                         return -EINVAL;
1084                 }
1085                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1086         } else if (!info->commit) {
1087                 OVS_NLERR(log,
1088                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1089                           );
1090                 return -EINVAL;
1091         }
1092         /* Allow missing IP_MAX. */
1093         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1094                 memcpy(&info->range.max_addr, &info->range.min_addr,
1095                        sizeof(info->range.max_addr));
1096         }
1097         /* Allow missing PROTO_MAX. */
1098         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1099             !have_proto_max) {
1100                 info->range.max_proto.all = info->range.min_proto.all;
1101         }
1102         return 0;
1103 }
1104 #endif
1105
1106 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1107         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1108         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1109                                     .maxlen = sizeof(u16) },
1110         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1111                                     .maxlen = sizeof(struct md_mark) },
1112         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1113                                     .maxlen = sizeof(struct md_labels) },
1114         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1115                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1116 #ifdef CONFIG_NF_NAT_NEEDED
1117         /* NAT length is checked when parsing the nested attributes. */
1118         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1119 #endif
1120 };
1121
1122 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1123                     const char **helper, bool log)
1124 {
1125         struct nlattr *a;
1126         int rem;
1127
1128         nla_for_each_nested(a, attr, rem) {
1129                 int type = nla_type(a);
1130                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1131                 int minlen = ovs_ct_attr_lens[type].minlen;
1132
1133                 if (type > OVS_CT_ATTR_MAX) {
1134                         OVS_NLERR(log,
1135                                   "Unknown conntrack attr (type=%d, max=%d)",
1136                                   type, OVS_CT_ATTR_MAX);
1137                         return -EINVAL;
1138                 }
1139                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1140                         OVS_NLERR(log,
1141                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1142                                   type, nla_len(a), maxlen);
1143                         return -EINVAL;
1144                 }
1145
1146                 switch (type) {
1147                 case OVS_CT_ATTR_COMMIT:
1148                         info->commit = true;
1149                         break;
1150 #ifdef CONFIG_NF_CONNTRACK_ZONES
1151                 case OVS_CT_ATTR_ZONE:
1152                         info->zone.id = nla_get_u16(a);
1153                         break;
1154 #endif
1155 #ifdef CONFIG_NF_CONNTRACK_MARK
1156                 case OVS_CT_ATTR_MARK: {
1157                         struct md_mark *mark = nla_data(a);
1158
1159                         if (!mark->mask) {
1160                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1161                                 return -EINVAL;
1162                         }
1163                         info->mark = *mark;
1164                         break;
1165                 }
1166 #endif
1167 #ifdef CONFIG_NF_CONNTRACK_LABELS
1168                 case OVS_CT_ATTR_LABELS: {
1169                         struct md_labels *labels = nla_data(a);
1170
1171                         if (!labels_nonzero(&labels->mask)) {
1172                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1173                                 return -EINVAL;
1174                         }
1175                         info->labels = *labels;
1176                         break;
1177                 }
1178 #endif
1179                 case OVS_CT_ATTR_HELPER:
1180                         *helper = nla_data(a);
1181                         if (!memchr(*helper, '\0', nla_len(a))) {
1182                                 OVS_NLERR(log, "Invalid conntrack helper");
1183                                 return -EINVAL;
1184                         }
1185                         break;
1186 #ifdef CONFIG_NF_NAT_NEEDED
1187                 case OVS_CT_ATTR_NAT: {
1188                         int err = parse_nat(a, info, log);
1189
1190                         if (err)
1191                                 return err;
1192                         break;
1193                 }
1194 #endif
1195                 default:
1196                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1197                                   type);
1198                         return -EINVAL;
1199                 }
1200         }
1201
1202         if (rem > 0) {
1203                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1204                 return -EINVAL;
1205         }
1206
1207         return 0;
1208 }
1209
1210 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1211 {
1212         if (attr == OVS_KEY_ATTR_CT_STATE)
1213                 return true;
1214         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1215             attr == OVS_KEY_ATTR_CT_ZONE)
1216                 return true;
1217         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1218             attr == OVS_KEY_ATTR_CT_MARK)
1219                 return true;
1220         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1221             attr == OVS_KEY_ATTR_CT_LABELS) {
1222                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1223
1224                 return ovs_net->xt_label;
1225         }
1226
1227         return false;
1228 }
1229
1230 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1231                        const struct sw_flow_key *key,
1232                        struct sw_flow_actions **sfa,  bool log)
1233 {
1234         struct ovs_conntrack_info ct_info;
1235         const char *helper = NULL;
1236         u16 family;
1237         int err;
1238
1239         family = key_to_nfproto(key);
1240         if (family == NFPROTO_UNSPEC) {
1241                 OVS_NLERR(log, "ct family unspecified");
1242                 return -EINVAL;
1243         }
1244
1245         memset(&ct_info, 0, sizeof(ct_info));
1246         ct_info.family = family;
1247
1248         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1249                         NF_CT_DEFAULT_ZONE_DIR, 0);
1250
1251         err = parse_ct(attr, &ct_info, &helper, log);
1252         if (err)
1253                 return err;
1254
1255         /* Set up template for tracking connections in specific zones. */
1256         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1257         if (!ct_info.ct) {
1258                 OVS_NLERR(log, "Failed to allocate conntrack template");
1259                 return -ENOMEM;
1260         }
1261
1262         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1263         nf_conntrack_get(&ct_info.ct->ct_general);
1264
1265         if (helper) {
1266                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1267                 if (err)
1268                         goto err_free_ct;
1269         }
1270
1271         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1272                                  sizeof(ct_info), log);
1273         if (err)
1274                 goto err_free_ct;
1275
1276         return 0;
1277 err_free_ct:
1278         __ovs_ct_free_action(&ct_info);
1279         return err;
1280 }
1281
1282 #ifdef CONFIG_NF_NAT_NEEDED
1283 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1284                                struct sk_buff *skb)
1285 {
1286         struct nlattr *start;
1287
1288         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1289         if (!start)
1290                 return false;
1291
1292         if (info->nat & OVS_CT_SRC_NAT) {
1293                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1294                         return false;
1295         } else if (info->nat & OVS_CT_DST_NAT) {
1296                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1297                         return false;
1298         } else {
1299                 goto out;
1300         }
1301
1302         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1303                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1304                     info->family == NFPROTO_IPV4) {
1305                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1306                                             info->range.min_addr.ip) ||
1307                             (info->range.max_addr.ip
1308                              != info->range.min_addr.ip &&
1309                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1310                                               info->range.max_addr.ip))))
1311                                 return false;
1312                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1313                            info->family == NFPROTO_IPV6) {
1314                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1315                                              &info->range.min_addr.in6) ||
1316                             (memcmp(&info->range.max_addr.in6,
1317                                     &info->range.min_addr.in6,
1318                                     sizeof(info->range.max_addr.in6)) &&
1319                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1320                                                &info->range.max_addr.in6))))
1321                                 return false;
1322                 } else {
1323                         return false;
1324                 }
1325         }
1326         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1327             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1328                          ntohs(info->range.min_proto.all)) ||
1329              (info->range.max_proto.all != info->range.min_proto.all &&
1330               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1331                           ntohs(info->range.max_proto.all)))))
1332                 return false;
1333
1334         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1335             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1336                 return false;
1337         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1338             nla_put_flag(skb, info->random_fully_compat
1339                          ? OVS_NAT_ATTR_PROTO_RANDOM
1340                          : OVS_NAT_ATTR_PROTO_HASH))
1341                 return false;
1342 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1343         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1344             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1345                 return false;
1346 #endif
1347 out:
1348         nla_nest_end(skb, start);
1349
1350         return true;
1351 }
1352 #endif
1353
1354 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1355                           struct sk_buff *skb)
1356 {
1357         struct nlattr *start;
1358
1359         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1360         if (!start)
1361                 return -EMSGSIZE;
1362
1363         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1364                 return -EMSGSIZE;
1365         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1366             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1367                 return -EMSGSIZE;
1368         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1369             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1370                     &ct_info->mark))
1371                 return -EMSGSIZE;
1372         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1373             labels_nonzero(&ct_info->labels.mask) &&
1374             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1375                     &ct_info->labels))
1376                 return -EMSGSIZE;
1377         if (ct_info->helper) {
1378                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1379                                    ct_info->helper->name))
1380                         return -EMSGSIZE;
1381         }
1382 #ifdef CONFIG_NF_NAT_NEEDED
1383         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1384                 return -EMSGSIZE;
1385 #endif
1386         nla_nest_end(skb, start);
1387
1388         return 0;
1389 }
1390
1391 void ovs_ct_free_action(const struct nlattr *a)
1392 {
1393         struct ovs_conntrack_info *ct_info = nla_data(a);
1394
1395         __ovs_ct_free_action(ct_info);
1396 }
1397
1398 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1399 {
1400         if (ct_info->helper)
1401                 module_put(ct_info->helper->me);
1402         if (ct_info->ct)
1403                 nf_ct_tmpl_free(ct_info->ct);
1404 }
1405
1406 void ovs_ct_init(struct net *net)
1407 {
1408         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1409         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1410
1411         if (nf_connlabels_get(net, n_bits - 1)) {
1412                 ovs_net->xt_label = false;
1413                 OVS_NLERR(true, "Failed to set connlabel length");
1414         } else {
1415                 ovs_net->xt_label = true;
1416         }
1417 }
1418
1419 void ovs_ct_exit(struct net *net)
1420 {
1421         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1422
1423         if (ovs_net->xt_label)
1424                 nf_connlabels_put(net);
1425 }
1426
1427 #endif /* CONFIG_NF_CONNTRACK */