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 = sizeof(cl->bits);
 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 || sizeof(cl->bits) < 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                         if (err != -EINPROGRESS)
 375                                 kfree_skb(skb);
 376                         return err;
 377                 }
 378
 379                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 380                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 381 #endif
 382         } else {
 383                 kfree_skb(skb);
 384                 return -EPFNOSUPPORT;
 385         }
 386
 387         key->ip.frag = OVS_FRAG_TYPE_NONE;
 388         skb_clear_hash(skb);
 389         skb->ignore_df = 1;
 390         *OVS_CB(skb) = ovs_cb;
 391
 392         return 0;
 393 }
 394
 395 static struct nf_conntrack_expect *
 396 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 397                    u16 proto, const struct sk_buff *skb)
 398 {
 399         struct nf_conntrack_tuple tuple;
 400
 401         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 402                 return NULL;
 403         return __nf_ct_expect_find(net, zone, &tuple);
 404 }
 405
 406 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 407 static enum ip_conntrack_info
 408 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 409 {
 410         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 411
 412         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 413                 return IP_CT_ESTABLISHED_REPLY;
 414         /* Once we've had two way comms, always ESTABLISHED. */
 415         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 416                 return IP_CT_ESTABLISHED;
 417         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 418                 return IP_CT_RELATED;
 419         return IP_CT_NEW;
 420 }
 421
 422 /* Find an existing connection which this packet belongs to without
 423  * re-attributing statistics or modifying the connection state.  This allows an
 424  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 425  *
 426  * Must be called with rcu_read_lock.
 427  *
 428  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 429  * if there is no existing entry.
 430  */
 431 static struct nf_conn *
 432 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 433                      u8 l3num, struct sk_buff *skb, bool natted)
 434 {
 435         struct nf_conntrack_l3proto *l3proto;
 436         struct nf_conntrack_l4proto *l4proto;
 437         struct nf_conntrack_tuple tuple;
 438         struct nf_conntrack_tuple_hash *h;
 439         struct nf_conn *ct;
 440         unsigned int dataoff;
 441         u8 protonum;
 442
 443         l3proto = __nf_ct_l3proto_find(l3num);
 444         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 445                                  &protonum) <= 0) {
 446                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 447                 return NULL;
 448         }
 449         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 450         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 451                              protonum, net, &tuple, l3proto, l4proto)) {
 452                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 453                 return NULL;
 454         }
 455
 456         /* Must invert the tuple if skb has been transformed by NAT. */
 457         if (natted) {
 458                 struct nf_conntrack_tuple inverse;
 459
 460                 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
 461                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 462                         return NULL;
 463                 }
 464                 tuple = inverse;
 465         }
 466
 467         /* look for tuple match */
 468         h = nf_conntrack_find_get(net, zone, &tuple);
 469         if (!h)
 470                 return NULL;   /* Not found. */
 471
 472         ct = nf_ct_tuplehash_to_ctrack(h);
 473
 474         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 475          * select the other tuplehash to get the right 'ctinfo' bits for this
 476          * packet.
 477          */
 478         if (natted)
 479                 h = &ct->tuplehash[!h->tuple.dst.dir];
 480
 481         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 482         return ct;
 483 }
 484
 485 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 486 static bool skb_nfct_cached(struct net *net,
 487                             const struct sw_flow_key *key,
 488                             const struct ovs_conntrack_info *info,
 489                             struct sk_buff *skb)
 490 {
 491         enum ip_conntrack_info ctinfo;
 492         struct nf_conn *ct;
 493
 494         ct = nf_ct_get(skb, &ctinfo);
 495         /* If no ct, check if we have evidence that an existing conntrack entry
 496          * might be found for this skb.  This happens when we lose a skb->_nfct
 497          * due to an upcall.  If the connection was not confirmed, it is not
 498          * cached and needs to be run through conntrack again.
 499          */
 500         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 501             !(key->ct.state & OVS_CS_F_INVALID) &&
 502             key->ct.zone == info->zone.id)
 503                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 504                                           !!(key->ct.state
 505                                              & OVS_CS_F_NAT_MASK));
 506         if (!ct)
 507                 return false;
 508         if (!net_eq(net, read_pnet(&ct->ct_net)))
 509                 return false;
 510         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 511                 return false;
 512         if (info->helper) {
 513                 struct nf_conn_help *help;
 514
 515                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 516                 if (help && rcu_access_pointer(help->helper) != info->helper)
 517                         return false;
 518         }
 519
 520         return true;
 521 }
 522
 523 #ifdef CONFIG_NF_NAT_NEEDED
 524 /* Modelled after nf_nat_ipv[46]_fn().
 525  * range is only used for new, uninitialized NAT state.
 526  * Returns either NF_ACCEPT or NF_DROP.
 527  */
 528 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 529                               enum ip_conntrack_info ctinfo,
 530                               const struct nf_nat_range *range,
 531                               enum nf_nat_manip_type maniptype)
 532 {
 533         int hooknum, nh_off, err = NF_ACCEPT;
 534
 535         nh_off = skb_network_offset(skb);
 536         skb_pull_rcsum(skb, nh_off);
 537
 538         /* See HOOK2MANIP(). */
 539         if (maniptype == NF_NAT_MANIP_SRC)
 540                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 541         else
 542                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 543
 544         switch (ctinfo) {
 545         case IP_CT_RELATED:
 546         case IP_CT_RELATED_REPLY:
 547                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 548                     skb->protocol == htons(ETH_P_IP) &&
 549                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 550                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 551                                                            hooknum))
 552                                 err = NF_DROP;
 553                         goto push;
 554                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 555                            skb->protocol == htons(ETH_P_IPV6)) {
 556                         __be16 frag_off;
 557                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 558                         int hdrlen = ipv6_skip_exthdr(skb,
 559                                                       sizeof(struct ipv6hdr),
 560                                                       &nexthdr, &frag_off);
 561
 562                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 563                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 564                                                                      ctinfo,
 565                                                                      hooknum,
 566                                                                      hdrlen))
 567                                         err = NF_DROP;
 568                                 goto push;
 569                         }
 570                 }
 571                 /* Non-ICMP, fall thru to initialize if needed. */
 572         case IP_CT_NEW:
 573                 /* Seen it before?  This can happen for loopback, retrans,
 574                  * or local packets.
 575                  */
 576                 if (!nf_nat_initialized(ct, maniptype)) {
 577                         /* Initialize according to the NAT action. */
 578                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 579                                 /* Action is set up to establish a new
 580                                  * mapping.
 581                                  */
 582                                 ? nf_nat_setup_info(ct, range, maniptype)
 583                                 : nf_nat_alloc_null_binding(ct, hooknum);
 584                         if (err != NF_ACCEPT)
 585                                 goto push;
 586                 }
 587                 break;
 588
 589         case IP_CT_ESTABLISHED:
 590         case IP_CT_ESTABLISHED_REPLY:
 591                 break;
 592
 593         default:
 594                 err = NF_DROP;
 595                 goto push;
 596         }
 597
 598         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 599 push:
 600         skb_push(skb, nh_off);
 601         skb_postpush_rcsum(skb, skb->data, nh_off);
 602
 603         return err;
 604 }
 605
 606 static void ovs_nat_update_key(struct sw_flow_key *key,
 607                                const struct sk_buff *skb,
 608                                enum nf_nat_manip_type maniptype)
 609 {
 610         if (maniptype == NF_NAT_MANIP_SRC) {
 611                 __be16 src;
 612
 613                 key->ct.state |= OVS_CS_F_SRC_NAT;
 614                 if (key->eth.type == htons(ETH_P_IP))
 615                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 616                 else if (key->eth.type == htons(ETH_P_IPV6))
 617                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 618                                sizeof(key->ipv6.addr.src));
 619                 else
 620                         return;
 621
 622                 if (key->ip.proto == IPPROTO_UDP)
 623                         src = udp_hdr(skb)->source;
 624                 else if (key->ip.proto == IPPROTO_TCP)
 625                         src = tcp_hdr(skb)->source;
 626                 else if (key->ip.proto == IPPROTO_SCTP)
 627                         src = sctp_hdr(skb)->source;
 628                 else
 629                         return;
 630
 631                 key->tp.src = src;
 632         } else {
 633                 __be16 dst;
 634
 635                 key->ct.state |= OVS_CS_F_DST_NAT;
 636                 if (key->eth.type == htons(ETH_P_IP))
 637                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 638                 else if (key->eth.type == htons(ETH_P_IPV6))
 639                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 640                                sizeof(key->ipv6.addr.dst));
 641                 else
 642                         return;
 643
 644                 if (key->ip.proto == IPPROTO_UDP)
 645                         dst = udp_hdr(skb)->dest;
 646                 else if (key->ip.proto == IPPROTO_TCP)
 647                         dst = tcp_hdr(skb)->dest;
 648                 else if (key->ip.proto == IPPROTO_SCTP)
 649                         dst = sctp_hdr(skb)->dest;
 650                 else
 651                         return;
 652
 653                 key->tp.dst = dst;
 654         }
 655 }
 656
 657 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 658 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 659                       const struct ovs_conntrack_info *info,
 660                       struct sk_buff *skb, struct nf_conn *ct,
 661                       enum ip_conntrack_info ctinfo)
 662 {
 663         enum nf_nat_manip_type maniptype;
 664         int err;
 665
 666         if (nf_ct_is_untracked(ct)) {
 667                 /* A NAT action may only be performed on tracked packets. */
 668                 return NF_ACCEPT;
 669         }
 670
 671         /* Add NAT extension if not confirmed yet. */
 672         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 673                 return NF_ACCEPT;   /* Can't NAT. */
 674
 675         /* Determine NAT type.
 676          * Check if the NAT type can be deduced from the tracked connection.
 677          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 678          * when committing.
 679          */
 680         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 681             ct->status & IPS_NAT_MASK &&
 682             (ctinfo != IP_CT_RELATED || info->commit)) {
 683                 /* NAT an established or related connection like before. */
 684                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 685                         /* This is the REPLY direction for a connection
 686                          * for which NAT was applied in the forward
 687                          * direction.  Do the reverse NAT.
 688                          */
 689                         maniptype = ct->status & IPS_SRC_NAT
 690                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 691                 else
 692                         maniptype = ct->status & IPS_SRC_NAT
 693                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 694         } else if (info->nat & OVS_CT_SRC_NAT) {
 695                 maniptype = NF_NAT_MANIP_SRC;
 696         } else if (info->nat & OVS_CT_DST_NAT) {
 697                 maniptype = NF_NAT_MANIP_DST;
 698         } else {
 699                 return NF_ACCEPT; /* Connection is not NATed. */
 700         }
 701         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 702
 703         /* Mark NAT done if successful and update the flow key. */
 704         if (err == NF_ACCEPT)
 705                 ovs_nat_update_key(key, skb, maniptype);
 706
 707         return err;
 708 }
 709 #else /* !CONFIG_NF_NAT_NEEDED */
 710 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 711                       const struct ovs_conntrack_info *info,
 712                       struct sk_buff *skb, struct nf_conn *ct,
 713                       enum ip_conntrack_info ctinfo)
 714 {
 715         return NF_ACCEPT;
 716 }
 717 #endif
 718
 719 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 720  * not done already.  Update key with new CT state after passing the packet
 721  * through conntrack.
 722  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 723  * set to NULL and 0 will be returned.
 724  */
 725 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 726                            const struct ovs_conntrack_info *info,
 727                            struct sk_buff *skb)
 728 {
 729         /* If we are recirculating packets to match on conntrack fields and
 730          * committing with a separate conntrack action,  then we don't need to
 731          * actually run the packet through conntrack twice unless it's for a
 732          * different zone.
 733          */
 734         bool cached = skb_nfct_cached(net, key, info, skb);
 735         enum ip_conntrack_info ctinfo;
 736         struct nf_conn *ct;
 737
 738         if (!cached) {
 739                 struct nf_conn *tmpl = info->ct;
 740                 int err;
 741
 742                 /* Associate skb with specified zone. */
 743                 if (tmpl) {
 744                         if (skb_nfct(skb))
 745                                 nf_conntrack_put(skb_nfct(skb));
 746                         nf_conntrack_get(&tmpl->ct_general);
 747                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 748                 }
 749
 750                 err = nf_conntrack_in(net, info->family,
 751                                       NF_INET_PRE_ROUTING, skb);
 752                 if (err != NF_ACCEPT)
 753                         return -ENOENT;
 754
 755                 /* Clear CT state NAT flags to mark that we have not yet done
 756                  * NAT after the nf_conntrack_in() call.  We can actually clear
 757                  * the whole state, as it will be re-initialized below.
 758                  */
 759                 key->ct.state = 0;
 760
 761                 /* Update the key, but keep the NAT flags. */
 762                 ovs_ct_update_key(skb, info, key, true, true);
 763         }
 764
 765         ct = nf_ct_get(skb, &ctinfo);
 766         if (ct) {
 767                 /* Packets starting a new connection must be NATted before the
 768                  * helper, so that the helper knows about the NAT.  We enforce
 769                  * this by delaying both NAT and helper calls for unconfirmed
 770                  * connections until the committing CT action.  For later
 771                  * packets NAT and Helper may be called in either order.
 772                  *
 773                  * NAT will be done only if the CT action has NAT, and only
 774                  * once per packet (per zone), as guarded by the NAT bits in
 775                  * the key->ct.state.
 776                  */
 777                 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
 778                     (nf_ct_is_confirmed(ct) || info->commit) &&
 779                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 780                         return -EINVAL;
 781                 }
 782
 783                 /* Userspace may decide to perform a ct lookup without a helper
 784                  * specified followed by a (recirculate and) commit with one.
 785                  * Therefore, for unconfirmed connections which we will commit,
 786                  * we need to attach the helper here.
 787                  */
 788                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 789                     info->helper && !nfct_help(ct)) {
 790                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 791                                                             GFP_ATOMIC);
 792                         if (err)
 793                                 return err;
 794                 }
 795
 796                 /* Call the helper only if:
 797                  * - nf_conntrack_in() was executed above ("!cached") for a
 798                  *   confirmed connection, or
 799                  * - When committing an unconfirmed connection.
 800                  */
 801                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 802                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 803                         return -EINVAL;
 804                 }
 805         }
 806
 807         return 0;
 808 }
 809
 810 /* Lookup connection and read fields into key. */
 811 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 812                          const struct ovs_conntrack_info *info,
 813                          struct sk_buff *skb)
 814 {
 815         struct nf_conntrack_expect *exp;
 816
 817         /* If we pass an expected packet through nf_conntrack_in() the
 818          * expectation is typically removed, but the packet could still be
 819          * lost in upcall processing.  To prevent this from happening we
 820          * perform an explicit expectation lookup.  Expected connections are
 821          * always new, and will be passed through conntrack only when they are
 822          * committed, as it is OK to remove the expectation at that time.
 823          */
 824         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 825         if (exp) {
 826                 u8 state;
 827
 828                 /* NOTE: New connections are NATted and Helped only when
 829                  * committed, so we are not calling into NAT here.
 830                  */
 831                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 832                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 833         } else {
 834                 struct nf_conn *ct;
 835                 int err;
 836
 837                 err = __ovs_ct_lookup(net, key, info, skb);
 838                 if (err)
 839                         return err;
 840
 841                 ct = (struct nf_conn *)skb_nfct(skb);
 842                 if (ct)
 843                         nf_ct_deliver_cached_events(ct);
 844         }
 845
 846         return 0;
 847 }
 848
 849 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 850 {
 851         size_t i;
 852
 853         for (i = 0; i < sizeof(*labels); i++)
 854                 if (labels->ct_labels[i])
 855                         return true;
 856
 857         return false;
 858 }
 859
 860 /* Lookup connection and confirm if unconfirmed. */
 861 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 862                          const struct ovs_conntrack_info *info,
 863                          struct sk_buff *skb)
 864 {
 865         int err;
 866
 867         err = __ovs_ct_lookup(net, key, info, skb);
 868         if (err)
 869                 return err;
 870
 871         /* Apply changes before confirming the connection so that the initial
 872          * conntrack NEW netlink event carries the values given in the CT
 873          * action.
 874          */
 875         if (info->mark.mask) {
 876                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 877                                       info->mark.mask);
 878                 if (err)
 879                         return err;
 880         }
 881         if (labels_nonzero(&info->labels.mask)) {
 882                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 883                                         &info->labels.mask);
 884                 if (err)
 885                         return err;
 886         }
 887         /* This will take care of sending queued events even if the connection
 888          * is already confirmed.
 889          */
 890         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 891                 return -EINVAL;
 892
 893         return 0;
 894 }
 895
 896 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 897  * value if 'skb' is freed.
 898  */
 899 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 900                    struct sw_flow_key *key,
 901                    const struct ovs_conntrack_info *info)
 902 {
 903         int nh_ofs;
 904         int err;
 905
 906         /* The conntrack module expects to be working at L3. */
 907         nh_ofs = skb_network_offset(skb);
 908         skb_pull_rcsum(skb, nh_ofs);
 909
 910         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 911                 err = handle_fragments(net, key, info->zone.id, skb);
 912                 if (err)
 913                         return err;
 914         }
 915
 916         if (info->commit)
 917                 err = ovs_ct_commit(net, key, info, skb);
 918         else
 919                 err = ovs_ct_lookup(net, key, info, skb);
 920
 921         skb_push(skb, nh_ofs);
 922         skb_postpush_rcsum(skb, skb->data, nh_ofs);
 923         if (err)
 924                 kfree_skb(skb);
 925         return err;
 926 }
 927
 928 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 929                              const struct sw_flow_key *key, bool log)
 930 {
 931         struct nf_conntrack_helper *helper;
 932         struct nf_conn_help *help;
 933
 934         helper = nf_conntrack_helper_try_module_get(name, info->family,
 935                                                     key->ip.proto);
 936         if (!helper) {
 937                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 938                 return -EINVAL;
 939         }
 940
 941         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 942         if (!help) {
 943                 module_put(helper->me);
 944                 return -ENOMEM;
 945         }
 946
 947         rcu_assign_pointer(help->helper, helper);
 948         info->helper = helper;
 949         return 0;
 950 }
 951
 952 #ifdef CONFIG_NF_NAT_NEEDED
 953 static int parse_nat(const struct nlattr *attr,
 954                      struct ovs_conntrack_info *info, bool log)
 955 {
 956         struct nlattr *a;
 957         int rem;
 958         bool have_ip_max = false;
 959         bool have_proto_max = false;
 960         bool ip_vers = (info->family == NFPROTO_IPV6);
 961
 962         nla_for_each_nested(a, attr, rem) {
 963                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
 964                         [OVS_NAT_ATTR_SRC] = {0, 0},
 965                         [OVS_NAT_ATTR_DST] = {0, 0},
 966                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
 967                                                  sizeof(struct in6_addr)},
 968                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
 969                                                  sizeof(struct in6_addr)},
 970                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
 971                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
 972                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
 973                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
 974                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
 975                 };
 976                 int type = nla_type(a);
 977
 978                 if (type > OVS_NAT_ATTR_MAX) {
 979                         OVS_NLERR(log,
 980                                   "Unknown NAT attribute (type=%d, max=%d).\n",
 981                                   type, OVS_NAT_ATTR_MAX);
 982                         return -EINVAL;
 983                 }
 984
 985                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
 986                         OVS_NLERR(log,
 987                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
 988                                   type, nla_len(a),
 989                                   ovs_nat_attr_lens[type][ip_vers]);
 990                         return -EINVAL;
 991                 }
 992
 993                 switch (type) {
 994                 case OVS_NAT_ATTR_SRC:
 995                 case OVS_NAT_ATTR_DST:
 996                         if (info->nat) {
 997                                 OVS_NLERR(log,
 998                                           "Only one type of NAT may be specified.\n"
 999                                           );
1000                                 return -ERANGE;
1001                         }
1002                         info->nat |= OVS_CT_NAT;
1003                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1004                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1005                         break;
1006
1007                 case OVS_NAT_ATTR_IP_MIN:
1008                         nla_memcpy(&info->range.min_addr, a,
1009                                    sizeof(info->range.min_addr));
1010                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1011                         break;
1012
1013                 case OVS_NAT_ATTR_IP_MAX:
1014                         have_ip_max = true;
1015                         nla_memcpy(&info->range.max_addr, a,
1016                                    sizeof(info->range.max_addr));
1017                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1018                         break;
1019
1020                 case OVS_NAT_ATTR_PROTO_MIN:
1021                         info->range.min_proto.all = htons(nla_get_u16(a));
1022                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1023                         break;
1024
1025                 case OVS_NAT_ATTR_PROTO_MAX:
1026                         have_proto_max = true;
1027                         info->range.max_proto.all = htons(nla_get_u16(a));
1028                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1029                         break;
1030
1031                 case OVS_NAT_ATTR_PERSISTENT:
1032                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1033                         break;
1034
1035                 case OVS_NAT_ATTR_PROTO_HASH:
1036                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1037                         break;
1038
1039                 case OVS_NAT_ATTR_PROTO_RANDOM:
1040                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1041                         break;
1042
1043                 default:
1044                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1045                         return -EINVAL;
1046                 }
1047         }
1048
1049         if (rem > 0) {
1050                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1051                 return -EINVAL;
1052         }
1053         if (!info->nat) {
1054                 /* Do not allow flags if no type is given. */
1055                 if (info->range.flags) {
1056                         OVS_NLERR(log,
1057                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1058                                   );
1059                         return -EINVAL;
1060                 }
1061                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1062         } else if (!info->commit) {
1063                 OVS_NLERR(log,
1064                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1065                           );
1066                 return -EINVAL;
1067         }
1068         /* Allow missing IP_MAX. */
1069         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1070                 memcpy(&info->range.max_addr, &info->range.min_addr,
1071                        sizeof(info->range.max_addr));
1072         }
1073         /* Allow missing PROTO_MAX. */
1074         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1075             !have_proto_max) {
1076                 info->range.max_proto.all = info->range.min_proto.all;
1077         }
1078         return 0;
1079 }
1080 #endif
1081
1082 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1083         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1084         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1085                                     .maxlen = sizeof(u16) },
1086         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1087                                     .maxlen = sizeof(struct md_mark) },
1088         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1089                                     .maxlen = sizeof(struct md_labels) },
1090         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1091                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1092 #ifdef CONFIG_NF_NAT_NEEDED
1093         /* NAT length is checked when parsing the nested attributes. */
1094         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1095 #endif
1096 };
1097
1098 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1099                     const char **helper, bool log)
1100 {
1101         struct nlattr *a;
1102         int rem;
1103
1104         nla_for_each_nested(a, attr, rem) {
1105                 int type = nla_type(a);
1106                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1107                 int minlen = ovs_ct_attr_lens[type].minlen;
1108
1109                 if (type > OVS_CT_ATTR_MAX) {
1110                         OVS_NLERR(log,
1111                                   "Unknown conntrack attr (type=%d, max=%d)",
1112                                   type, OVS_CT_ATTR_MAX);
1113                         return -EINVAL;
1114                 }
1115                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1116                         OVS_NLERR(log,
1117                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1118                                   type, nla_len(a), maxlen);
1119                         return -EINVAL;
1120                 }
1121
1122                 switch (type) {
1123                 case OVS_CT_ATTR_COMMIT:
1124                         info->commit = true;
1125                         break;
1126 #ifdef CONFIG_NF_CONNTRACK_ZONES
1127                 case OVS_CT_ATTR_ZONE:
1128                         info->zone.id = nla_get_u16(a);
1129                         break;
1130 #endif
1131 #ifdef CONFIG_NF_CONNTRACK_MARK
1132                 case OVS_CT_ATTR_MARK: {
1133                         struct md_mark *mark = nla_data(a);
1134
1135                         if (!mark->mask) {
1136                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1137                                 return -EINVAL;
1138                         }
1139                         info->mark = *mark;
1140                         break;
1141                 }
1142 #endif
1143 #ifdef CONFIG_NF_CONNTRACK_LABELS
1144                 case OVS_CT_ATTR_LABELS: {
1145                         struct md_labels *labels = nla_data(a);
1146
1147                         if (!labels_nonzero(&labels->mask)) {
1148                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1149                                 return -EINVAL;
1150                         }
1151                         info->labels = *labels;
1152                         break;
1153                 }
1154 #endif
1155                 case OVS_CT_ATTR_HELPER:
1156                         *helper = nla_data(a);
1157                         if (!memchr(*helper, '\0', nla_len(a))) {
1158                                 OVS_NLERR(log, "Invalid conntrack helper");
1159                                 return -EINVAL;
1160                         }
1161                         break;
1162 #ifdef CONFIG_NF_NAT_NEEDED
1163                 case OVS_CT_ATTR_NAT: {
1164                         int err = parse_nat(a, info, log);
1165
1166                         if (err)
1167                                 return err;
1168                         break;
1169                 }
1170 #endif
1171                 default:
1172                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1173                                   type);
1174                         return -EINVAL;
1175                 }
1176         }
1177
1178 #ifdef CONFIG_NF_CONNTRACK_MARK
1179         if (!info->commit && info->mark.mask) {
1180                 OVS_NLERR(log,
1181                           "Setting conntrack mark requires 'commit' flag.");
1182                 return -EINVAL;
1183         }
1184 #endif
1185 #ifdef CONFIG_NF_CONNTRACK_LABELS
1186         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1187                 OVS_NLERR(log,
1188                           "Setting conntrack labels requires 'commit' flag.");
1189                 return -EINVAL;
1190         }
1191 #endif
1192         if (rem > 0) {
1193                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1194                 return -EINVAL;
1195         }
1196
1197         return 0;
1198 }
1199
1200 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1201 {
1202         if (attr == OVS_KEY_ATTR_CT_STATE)
1203                 return true;
1204         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1205             attr == OVS_KEY_ATTR_CT_ZONE)
1206                 return true;
1207         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1208             attr == OVS_KEY_ATTR_CT_MARK)
1209                 return true;
1210         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1211             attr == OVS_KEY_ATTR_CT_LABELS) {
1212                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1213
1214                 return ovs_net->xt_label;
1215         }
1216
1217         return false;
1218 }
1219
1220 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1221                        const struct sw_flow_key *key,
1222                        struct sw_flow_actions **sfa,  bool log)
1223 {
1224         struct ovs_conntrack_info ct_info;
1225         const char *helper = NULL;
1226         u16 family;
1227         int err;
1228
1229         family = key_to_nfproto(key);
1230         if (family == NFPROTO_UNSPEC) {
1231                 OVS_NLERR(log, "ct family unspecified");
1232                 return -EINVAL;
1233         }
1234
1235         memset(&ct_info, 0, sizeof(ct_info));
1236         ct_info.family = family;
1237
1238         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1239                         NF_CT_DEFAULT_ZONE_DIR, 0);
1240
1241         err = parse_ct(attr, &ct_info, &helper, log);
1242         if (err)
1243                 return err;
1244
1245         /* Set up template for tracking connections in specific zones. */
1246         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1247         if (!ct_info.ct) {
1248                 OVS_NLERR(log, "Failed to allocate conntrack template");
1249                 return -ENOMEM;
1250         }
1251
1252         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1253         nf_conntrack_get(&ct_info.ct->ct_general);
1254
1255         if (helper) {
1256                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1257                 if (err)
1258                         goto err_free_ct;
1259         }
1260
1261         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1262                                  sizeof(ct_info), log);
1263         if (err)
1264                 goto err_free_ct;
1265
1266         return 0;
1267 err_free_ct:
1268         __ovs_ct_free_action(&ct_info);
1269         return err;
1270 }
1271
1272 #ifdef CONFIG_NF_NAT_NEEDED
1273 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1274                                struct sk_buff *skb)
1275 {
1276         struct nlattr *start;
1277
1278         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1279         if (!start)
1280                 return false;
1281
1282         if (info->nat & OVS_CT_SRC_NAT) {
1283                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1284                         return false;
1285         } else if (info->nat & OVS_CT_DST_NAT) {
1286                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1287                         return false;
1288         } else {
1289                 goto out;
1290         }
1291
1292         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1293                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1294                     info->family == NFPROTO_IPV4) {
1295                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1296                                             info->range.min_addr.ip) ||
1297                             (info->range.max_addr.ip
1298                              != info->range.min_addr.ip &&
1299                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1300                                               info->range.max_addr.ip))))
1301                                 return false;
1302                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1303                            info->family == NFPROTO_IPV6) {
1304                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1305                                              &info->range.min_addr.in6) ||
1306                             (memcmp(&info->range.max_addr.in6,
1307                                     &info->range.min_addr.in6,
1308                                     sizeof(info->range.max_addr.in6)) &&
1309                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1310                                                &info->range.max_addr.in6))))
1311                                 return false;
1312                 } else {
1313                         return false;
1314                 }
1315         }
1316         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1317             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1318                          ntohs(info->range.min_proto.all)) ||
1319              (info->range.max_proto.all != info->range.min_proto.all &&
1320               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1321                           ntohs(info->range.max_proto.all)))))
1322                 return false;
1323
1324         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1325             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1326                 return false;
1327         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1328             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1329                 return false;
1330         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1331             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1332                 return false;
1333 out:
1334         nla_nest_end(skb, start);
1335
1336         return true;
1337 }
1338 #endif
1339
1340 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1341                           struct sk_buff *skb)
1342 {
1343         struct nlattr *start;
1344
1345         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1346         if (!start)
1347                 return -EMSGSIZE;
1348
1349         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1350                 return -EMSGSIZE;
1351         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1352             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1353                 return -EMSGSIZE;
1354         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1355             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1356                     &ct_info->mark))
1357                 return -EMSGSIZE;
1358         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1359             labels_nonzero(&ct_info->labels.mask) &&
1360             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1361                     &ct_info->labels))
1362                 return -EMSGSIZE;
1363         if (ct_info->helper) {
1364                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1365                                    ct_info->helper->name))
1366                         return -EMSGSIZE;
1367         }
1368 #ifdef CONFIG_NF_NAT_NEEDED
1369         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1370                 return -EMSGSIZE;
1371 #endif
1372         nla_nest_end(skb, start);
1373
1374         return 0;
1375 }
1376
1377 void ovs_ct_free_action(const struct nlattr *a)
1378 {
1379         struct ovs_conntrack_info *ct_info = nla_data(a);
1380
1381         __ovs_ct_free_action(ct_info);
1382 }
1383
1384 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1385 {
1386         if (ct_info->helper)
1387                 module_put(ct_info->helper->me);
1388         if (ct_info->ct)
1389                 nf_ct_tmpl_free(ct_info->ct);
1390 }
1391
1392 void ovs_ct_init(struct net *net)
1393 {
1394         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1395         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1396
1397         if (nf_connlabels_get(net, n_bits - 1)) {
1398                 ovs_net->xt_label = false;
1399                 OVS_NLERR(true, "Failed to set connlabel length");
1400         } else {
1401                 ovs_net->xt_label = true;
1402         }
1403 }
1404
1405 void ovs_ct_exit(struct net *net)
1406 {
1407         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1408
1409         if (ovs_net->xt_label)
1410                 nf_connlabels_put(net);
1411 }