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 <net/ip.h>
  17 #include <net/netfilter/nf_conntrack_core.h>
  18 #include <net/netfilter/nf_conntrack_helper.h>
  19 #include <net/netfilter/nf_conntrack_labels.h>
  20 #include <net/netfilter/nf_conntrack_zones.h>
  21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  22
  23 #include "datapath.h"
  24 #include "conntrack.h"
  25 #include "flow.h"
  26 #include "flow_netlink.h"
  27
  28 struct ovs_ct_len_tbl {
  29         size_t maxlen;
  30         size_t minlen;
  31 };
  32
  33 /* Metadata mark for masked write to conntrack mark */
  34 struct md_mark {
  35         u32 value;
  36         u32 mask;
  37 };
  38
  39 /* Metadata label for masked write to conntrack label. */
  40 struct md_labels {
  41         struct ovs_key_ct_labels value;
  42         struct ovs_key_ct_labels mask;
  43 };
  44
  45 /* Conntrack action context for execution. */
  46 struct ovs_conntrack_info {
  47         struct nf_conntrack_helper *helper;
  48         struct nf_conntrack_zone zone;
  49         struct nf_conn *ct;
  50         u8 commit : 1;
  51         u16 family;
  52         struct md_mark mark;
  53         struct md_labels labels;
  54 };
  55
  56 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
  57
  58 static u16 key_to_nfproto(const struct sw_flow_key *key)
  59 {
  60         switch (ntohs(key->eth.type)) {
  61         case ETH_P_IP:
  62                 return NFPROTO_IPV4;
  63         case ETH_P_IPV6:
  64                 return NFPROTO_IPV6;
  65         default:
  66                 return NFPROTO_UNSPEC;
  67         }
  68 }
  69
  70 /* Map SKB connection state into the values used by flow definition. */
  71 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
  72 {
  73         u8 ct_state = OVS_CS_F_TRACKED;
  74
  75         switch (ctinfo) {
  76         case IP_CT_ESTABLISHED_REPLY:
  77         case IP_CT_RELATED_REPLY:
  78         case IP_CT_NEW_REPLY:
  79                 ct_state |= OVS_CS_F_REPLY_DIR;
  80                 break;
  81         default:
  82                 break;
  83         }
  84
  85         switch (ctinfo) {
  86         case IP_CT_ESTABLISHED:
  87         case IP_CT_ESTABLISHED_REPLY:
  88                 ct_state |= OVS_CS_F_ESTABLISHED;
  89                 break;
  90         case IP_CT_RELATED:
  91         case IP_CT_RELATED_REPLY:
  92                 ct_state |= OVS_CS_F_RELATED;
  93                 break;
  94         case IP_CT_NEW:
  95         case IP_CT_NEW_REPLY:
  96                 ct_state |= OVS_CS_F_NEW;
  97                 break;
  98         default:
  99                 break;
 100         }
 101
 102         return ct_state;
 103 }
 104
 105 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 106 {
 107 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 108         return ct ? ct->mark : 0;
 109 #else
 110         return 0;
 111 #endif
 112 }
 113
 114 static void ovs_ct_get_labels(const struct nf_conn *ct,
 115                               struct ovs_key_ct_labels *labels)
 116 {
 117         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 118
 119         if (cl) {
 120                 size_t len = cl->words * sizeof(long);
 121
 122                 if (len > OVS_CT_LABELS_LEN)
 123                         len = OVS_CT_LABELS_LEN;
 124                 else if (len < OVS_CT_LABELS_LEN)
 125                         memset(labels, 0, OVS_CT_LABELS_LEN);
 126                 memcpy(labels, cl->bits, len);
 127         } else {
 128                 memset(labels, 0, OVS_CT_LABELS_LEN);
 129         }
 130 }
 131
 132 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 133                                 const struct nf_conntrack_zone *zone,
 134                                 const struct nf_conn *ct)
 135 {
 136         key->ct.state = state;
 137         key->ct.zone = zone->id;
 138         key->ct.mark = ovs_ct_get_mark(ct);
 139         ovs_ct_get_labels(ct, &key->ct.labels);
 140 }
 141
 142 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
 143  * previously sent the packet to conntrack via the ct action.
 144  */
 145 static void ovs_ct_update_key(const struct sk_buff *skb,
 146                               const struct ovs_conntrack_info *info,
 147                               struct sw_flow_key *key, bool post_ct)
 148 {
 149         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 150         enum ip_conntrack_info ctinfo;
 151         struct nf_conn *ct;
 152         u8 state = 0;
 153
 154         ct = nf_ct_get(skb, &ctinfo);
 155         if (ct) {
 156                 state = ovs_ct_get_state(ctinfo);
 157                 if (!nf_ct_is_confirmed(ct))
 158                         state |= OVS_CS_F_NEW;
 159                 if (ct->master)
 160                         state |= OVS_CS_F_RELATED;
 161                 zone = nf_ct_zone(ct);
 162         } else if (post_ct) {
 163                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 164                 if (info)
 165                         zone = &info->zone;
 166         }
 167         __ovs_ct_update_key(key, state, zone, ct);
 168 }
 169
 170 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 171 {
 172         ovs_ct_update_key(skb, NULL, key, false);
 173 }
 174
 175 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 176 {
 177         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 178                 return -EMSGSIZE;
 179
 180         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 181             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 182                 return -EMSGSIZE;
 183
 184         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 185             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 186                 return -EMSGSIZE;
 187
 188         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 189             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 190                     &key->ct.labels))
 191                 return -EMSGSIZE;
 192
 193         return 0;
 194 }
 195
 196 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 197                            u32 ct_mark, u32 mask)
 198 {
 199 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 200         enum ip_conntrack_info ctinfo;
 201         struct nf_conn *ct;
 202         u32 new_mark;
 203
 204
 205         /* The connection could be invalid, in which case set_mark is no-op. */
 206         ct = nf_ct_get(skb, &ctinfo);
 207         if (!ct)
 208                 return 0;
 209
 210         new_mark = ct_mark | (ct->mark & ~(mask));
 211         if (ct->mark != new_mark) {
 212                 ct->mark = new_mark;
 213                 nf_conntrack_event_cache(IPCT_MARK, ct);
 214                 key->ct.mark = new_mark;
 215         }
 216
 217         return 0;
 218 #else
 219         return -ENOTSUPP;
 220 #endif
 221 }
 222
 223 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 224                              const struct ovs_key_ct_labels *labels,
 225                              const struct ovs_key_ct_labels *mask)
 226 {
 227         enum ip_conntrack_info ctinfo;
 228         struct nf_conn_labels *cl;
 229         struct nf_conn *ct;
 230         int err;
 231
 232         /* The connection could be invalid, in which case set_label is no-op.*/
 233         ct = nf_ct_get(skb, &ctinfo);
 234         if (!ct)
 235                 return 0;
 236
 237         cl = nf_ct_labels_find(ct);
 238         if (!cl) {
 239                 nf_ct_labels_ext_add(ct);
 240                 cl = nf_ct_labels_find(ct);
 241         }
 242         if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
 243                 return -ENOSPC;
 244
 245         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 246                                     OVS_CT_LABELS_LEN / sizeof(u32));
 247         if (err)
 248                 return err;
 249
 250         ovs_ct_get_labels(ct, &key->ct.labels);
 251         return 0;
 252 }
 253
 254 /* 'skb' should already be pulled to nh_ofs. */
 255 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 256 {
 257         const struct nf_conntrack_helper *helper;
 258         const struct nf_conn_help *help;
 259         enum ip_conntrack_info ctinfo;
 260         unsigned int protoff;
 261         struct nf_conn *ct;
 262
 263         ct = nf_ct_get(skb, &ctinfo);
 264         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 265                 return NF_ACCEPT;
 266
 267         help = nfct_help(ct);
 268         if (!help)
 269                 return NF_ACCEPT;
 270
 271         helper = rcu_dereference(help->helper);
 272         if (!helper)
 273                 return NF_ACCEPT;
 274
 275         switch (proto) {
 276         case NFPROTO_IPV4:
 277                 protoff = ip_hdrlen(skb);
 278                 break;
 279         case NFPROTO_IPV6: {
 280                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 281                 __be16 frag_off;
 282                 int ofs;
 283
 284                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 285                                        &frag_off);
 286                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 287                         pr_debug("proto header not found\n");
 288                         return NF_ACCEPT;
 289                 }
 290                 protoff = ofs;
 291                 break;
 292         }
 293         default:
 294                 WARN_ONCE(1, "helper invoked on non-IP family!");
 295                 return NF_DROP;
 296         }
 297
 298         return helper->help(skb, protoff, ct, ctinfo);
 299 }
 300
 301 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 302  * value if 'skb' is freed.
 303  */
 304 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 305                             u16 zone, struct sk_buff *skb)
 306 {
 307         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 308
 309         if (key->eth.type == htons(ETH_P_IP)) {
 310                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 311                 int err;
 312
 313                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 314                 err = ip_defrag(net, skb, user);
 315                 if (err)
 316                         return err;
 317
 318                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 319 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 320         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 321                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 322                 struct sk_buff *reasm;
 323
 324                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 325                 reasm = nf_ct_frag6_gather(net, skb, user);
 326                 if (!reasm)
 327                         return -EINPROGRESS;
 328
 329                 if (skb == reasm) {
 330                         kfree_skb(skb);
 331                         return -EINVAL;
 332                 }
 333
 334                 /* Don't free 'skb' even though it is one of the original
 335                  * fragments, as we're going to morph it into the head.
 336                  */
 337                 skb_get(skb);
 338                 nf_ct_frag6_consume_orig(reasm);
 339
 340                 key->ip.proto = ipv6_hdr(reasm)->nexthdr;
 341                 skb_morph(skb, reasm);
 342                 skb->next = reasm->next;
 343                 consume_skb(reasm);
 344                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 345 #endif
 346         } else {
 347                 kfree_skb(skb);
 348                 return -EPFNOSUPPORT;
 349         }
 350
 351         key->ip.frag = OVS_FRAG_TYPE_NONE;
 352         skb_clear_hash(skb);
 353         skb->ignore_df = 1;
 354         *OVS_CB(skb) = ovs_cb;
 355
 356         return 0;
 357 }
 358
 359 static struct nf_conntrack_expect *
 360 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 361                    u16 proto, const struct sk_buff *skb)
 362 {
 363         struct nf_conntrack_tuple tuple;
 364
 365         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 366                 return NULL;
 367         return __nf_ct_expect_find(net, zone, &tuple);
 368 }
 369
 370 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 371 static bool skb_nfct_cached(const struct net *net, const struct sk_buff *skb,
 372                             const struct ovs_conntrack_info *info)
 373 {
 374         enum ip_conntrack_info ctinfo;
 375         struct nf_conn *ct;
 376
 377         ct = nf_ct_get(skb, &ctinfo);
 378         if (!ct)
 379                 return false;
 380         if (!net_eq(net, read_pnet(&ct->ct_net)))
 381                 return false;
 382         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 383                 return false;
 384         if (info->helper) {
 385                 struct nf_conn_help *help;
 386
 387                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 388                 if (help && rcu_access_pointer(help->helper) != info->helper)
 389                         return false;
 390         }
 391
 392         return true;
 393 }
 394
 395 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 396                            const struct ovs_conntrack_info *info,
 397                            struct sk_buff *skb)
 398 {
 399         /* If we are recirculating packets to match on conntrack fields and
 400          * committing with a separate conntrack action,  then we don't need to
 401          * actually run the packet through conntrack twice unless it's for a
 402          * different zone.
 403          */
 404         if (!skb_nfct_cached(net, skb, info)) {
 405                 struct nf_conn *tmpl = info->ct;
 406
 407                 /* Associate skb with specified zone. */
 408                 if (tmpl) {
 409                         if (skb->nfct)
 410                                 nf_conntrack_put(skb->nfct);
 411                         nf_conntrack_get(&tmpl->ct_general);
 412                         skb->nfct = &tmpl->ct_general;
 413                         skb->nfctinfo = IP_CT_NEW;
 414                 }
 415
 416                 if (nf_conntrack_in(net, info->family, NF_INET_PRE_ROUTING,
 417                                     skb) != NF_ACCEPT)
 418                         return -ENOENT;
 419
 420                 if (ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 421                         WARN_ONCE(1, "helper rejected packet");
 422                         return -EINVAL;
 423                 }
 424         }
 425
 426         ovs_ct_update_key(skb, info, key, true);
 427
 428         return 0;
 429 }
 430
 431 /* Lookup connection and read fields into key. */
 432 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 433                          const struct ovs_conntrack_info *info,
 434                          struct sk_buff *skb)
 435 {
 436         struct nf_conntrack_expect *exp;
 437
 438         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 439         if (exp) {
 440                 u8 state;
 441
 442                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 443                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 444         } else {
 445                 int err;
 446
 447                 err = __ovs_ct_lookup(net, key, info, skb);
 448                 if (err)
 449                         return err;
 450         }
 451
 452         return 0;
 453 }
 454
 455 /* Lookup connection and confirm if unconfirmed. */
 456 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 457                          const struct ovs_conntrack_info *info,
 458                          struct sk_buff *skb)
 459 {
 460         u8 state;
 461         int err;
 462
 463         state = key->ct.state;
 464         if (key->ct.zone == info->zone.id &&
 465             ((state & OVS_CS_F_TRACKED) && !(state & OVS_CS_F_NEW))) {
 466                 /* Previous lookup has shown that this connection is already
 467                  * tracked and committed. Skip committing.
 468                  */
 469                 return 0;
 470         }
 471
 472         err = __ovs_ct_lookup(net, key, info, skb);
 473         if (err)
 474                 return err;
 475         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 476                 return -EINVAL;
 477
 478         return 0;
 479 }
 480
 481 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 482 {
 483         size_t i;
 484
 485         for (i = 0; i < sizeof(*labels); i++)
 486                 if (labels->ct_labels[i])
 487                         return true;
 488
 489         return false;
 490 }
 491
 492 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 493  * value if 'skb' is freed.
 494  */
 495 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 496                    struct sw_flow_key *key,
 497                    const struct ovs_conntrack_info *info)
 498 {
 499         int nh_ofs;
 500         int err;
 501
 502         /* The conntrack module expects to be working at L3. */
 503         nh_ofs = skb_network_offset(skb);
 504         skb_pull(skb, nh_ofs);
 505
 506         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 507                 err = handle_fragments(net, key, info->zone.id, skb);
 508                 if (err)
 509                         return err;
 510         }
 511
 512         if (info->commit)
 513                 err = ovs_ct_commit(net, key, info, skb);
 514         else
 515                 err = ovs_ct_lookup(net, key, info, skb);
 516         if (err)
 517                 goto err;
 518
 519         if (info->mark.mask) {
 520                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 521                                       info->mark.mask);
 522                 if (err)
 523                         goto err;
 524         }
 525         if (labels_nonzero(&info->labels.mask))
 526                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 527                                         &info->labels.mask);
 528 err:
 529         skb_push(skb, nh_ofs);
 530         if (err)
 531                 kfree_skb(skb);
 532         return err;
 533 }
 534
 535 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 536                              const struct sw_flow_key *key, bool log)
 537 {
 538         struct nf_conntrack_helper *helper;
 539         struct nf_conn_help *help;
 540
 541         helper = nf_conntrack_helper_try_module_get(name, info->family,
 542                                                     key->ip.proto);
 543         if (!helper) {
 544                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 545                 return -EINVAL;
 546         }
 547
 548         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 549         if (!help) {
 550                 module_put(helper->me);
 551                 return -ENOMEM;
 552         }
 553
 554         rcu_assign_pointer(help->helper, helper);
 555         info->helper = helper;
 556         return 0;
 557 }
 558
 559 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
 560         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
 561         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
 562                                     .maxlen = sizeof(u16) },
 563         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
 564                                     .maxlen = sizeof(struct md_mark) },
 565         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
 566                                     .maxlen = sizeof(struct md_labels) },
 567         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
 568                                     .maxlen = NF_CT_HELPER_NAME_LEN }
 569 };
 570
 571 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
 572                     const char **helper, bool log)
 573 {
 574         struct nlattr *a;
 575         int rem;
 576
 577         nla_for_each_nested(a, attr, rem) {
 578                 int type = nla_type(a);
 579                 int maxlen = ovs_ct_attr_lens[type].maxlen;
 580                 int minlen = ovs_ct_attr_lens[type].minlen;
 581
 582                 if (type > OVS_CT_ATTR_MAX) {
 583                         OVS_NLERR(log,
 584                                   "Unknown conntrack attr (type=%d, max=%d)",
 585                                   type, OVS_CT_ATTR_MAX);
 586                         return -EINVAL;
 587                 }
 588                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
 589                         OVS_NLERR(log,
 590                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
 591                                   type, nla_len(a), maxlen);
 592                         return -EINVAL;
 593                 }
 594
 595                 switch (type) {
 596                 case OVS_CT_ATTR_COMMIT:
 597                         info->commit = true;
 598                         break;
 599 #ifdef CONFIG_NF_CONNTRACK_ZONES
 600                 case OVS_CT_ATTR_ZONE:
 601                         info->zone.id = nla_get_u16(a);
 602                         break;
 603 #endif
 604 #ifdef CONFIG_NF_CONNTRACK_MARK
 605                 case OVS_CT_ATTR_MARK: {
 606                         struct md_mark *mark = nla_data(a);
 607
 608                         if (!mark->mask) {
 609                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
 610                                 return -EINVAL;
 611                         }
 612                         info->mark = *mark;
 613                         break;
 614                 }
 615 #endif
 616 #ifdef CONFIG_NF_CONNTRACK_LABELS
 617                 case OVS_CT_ATTR_LABELS: {
 618                         struct md_labels *labels = nla_data(a);
 619
 620                         if (!labels_nonzero(&labels->mask)) {
 621                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
 622                                 return -EINVAL;
 623                         }
 624                         info->labels = *labels;
 625                         break;
 626                 }
 627 #endif
 628                 case OVS_CT_ATTR_HELPER:
 629                         *helper = nla_data(a);
 630                         if (!memchr(*helper, '\0', nla_len(a))) {
 631                                 OVS_NLERR(log, "Invalid conntrack helper");
 632                                 return -EINVAL;
 633                         }
 634                         break;
 635                 default:
 636                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
 637                                   type);
 638                         return -EINVAL;
 639                 }
 640         }
 641
 642         if (rem > 0) {
 643                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
 644                 return -EINVAL;
 645         }
 646
 647         return 0;
 648 }
 649
 650 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
 651 {
 652         if (attr == OVS_KEY_ATTR_CT_STATE)
 653                 return true;
 654         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 655             attr == OVS_KEY_ATTR_CT_ZONE)
 656                 return true;
 657         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 658             attr == OVS_KEY_ATTR_CT_MARK)
 659                 return true;
 660         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 661             attr == OVS_KEY_ATTR_CT_LABELS) {
 662                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 663
 664                 return ovs_net->xt_label;
 665         }
 666
 667         return false;
 668 }
 669
 670 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
 671                        const struct sw_flow_key *key,
 672                        struct sw_flow_actions **sfa,  bool log)
 673 {
 674         struct ovs_conntrack_info ct_info;
 675         const char *helper = NULL;
 676         u16 family;
 677         int err;
 678
 679         family = key_to_nfproto(key);
 680         if (family == NFPROTO_UNSPEC) {
 681                 OVS_NLERR(log, "ct family unspecified");
 682                 return -EINVAL;
 683         }
 684
 685         memset(&ct_info, 0, sizeof(ct_info));
 686         ct_info.family = family;
 687
 688         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
 689                         NF_CT_DEFAULT_ZONE_DIR, 0);
 690
 691         err = parse_ct(attr, &ct_info, &helper, log);
 692         if (err)
 693                 return err;
 694
 695         /* Set up template for tracking connections in specific zones. */
 696         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
 697         if (!ct_info.ct) {
 698                 OVS_NLERR(log, "Failed to allocate conntrack template");
 699                 return -ENOMEM;
 700         }
 701
 702         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
 703         nf_conntrack_get(&ct_info.ct->ct_general);
 704
 705         if (helper) {
 706                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
 707                 if (err)
 708                         goto err_free_ct;
 709         }
 710
 711         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
 712                                  sizeof(ct_info), log);
 713         if (err)
 714                 goto err_free_ct;
 715
 716         return 0;
 717 err_free_ct:
 718         __ovs_ct_free_action(&ct_info);
 719         return err;
 720 }
 721
 722 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
 723                           struct sk_buff *skb)
 724 {
 725         struct nlattr *start;
 726
 727         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
 728         if (!start)
 729                 return -EMSGSIZE;
 730
 731         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
 732                 return -EMSGSIZE;
 733         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 734             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
 735                 return -EMSGSIZE;
 736         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
 737             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
 738                     &ct_info->mark))
 739                 return -EMSGSIZE;
 740         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 741             labels_nonzero(&ct_info->labels.mask) &&
 742             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
 743                     &ct_info->labels))
 744                 return -EMSGSIZE;
 745         if (ct_info->helper) {
 746                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
 747                                    ct_info->helper->name))
 748                         return -EMSGSIZE;
 749         }
 750
 751         nla_nest_end(skb, start);
 752
 753         return 0;
 754 }
 755
 756 void ovs_ct_free_action(const struct nlattr *a)
 757 {
 758         struct ovs_conntrack_info *ct_info = nla_data(a);
 759
 760         __ovs_ct_free_action(ct_info);
 761 }
 762
 763 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
 764 {
 765         if (ct_info->helper)
 766                 module_put(ct_info->helper->me);
 767         if (ct_info->ct)
 768                 nf_ct_put(ct_info->ct);
 769 }
 770
 771 void ovs_ct_init(struct net *net)
 772 {
 773         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
 774         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 775
 776         if (nf_connlabels_get(net, n_bits)) {
 777                 ovs_net->xt_label = false;
 778                 OVS_NLERR(true, "Failed to set connlabel length");
 779         } else {
 780                 ovs_net->xt_label = true;
 781         }
 782 }
 783
 784 void ovs_ct_exit(struct net *net)
 785 {
 786         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 787
 788         if (ovs_net->xt_label)
 789                 nf_connlabels_put(net);
 790 }