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 LINUX_VERSION_CODE >= KERNEL_VERSION(3,10,0) && \
  18     IS_ENABLED(CONFIG_NF_CONNTRACK)
  19
  20 #include <linux/module.h>
  21 #include <linux/openvswitch.h>
  22 #include <net/ip.h>
  23 #include <net/netfilter/nf_conntrack_core.h>
  24 #include <net/netfilter/nf_conntrack_helper.h>
  25 #include <net/netfilter/nf_conntrack_labels.h>
  26 #include <net/netfilter/nf_conntrack_zones.h>
  27 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  28
  29 #include "datapath.h"
  30 #include "conntrack.h"
  31 #include "flow.h"
  32 #include "flow_netlink.h"
  33 #include "gso.h"
  34
  35 struct ovs_ct_len_tbl {
  36         size_t maxlen;
  37         size_t minlen;
  38 };
  39
  40 /* Metadata mark for masked write to conntrack mark */
  41 struct md_mark {
  42         u32 value;
  43         u32 mask;
  44 };
  45
  46 /* Metadata label for masked write to conntrack label. */
  47 struct md_labels {
  48         struct ovs_key_ct_labels value;
  49         struct ovs_key_ct_labels mask;
  50 };
  51
  52 /* Conntrack action context for execution. */
  53 struct ovs_conntrack_info {
  54         struct nf_conntrack_helper *helper;
  55         struct nf_conntrack_zone zone;
  56         struct nf_conn *ct;
  57         u8 commit : 1;
  58         u16 family;
  59         struct md_mark mark;
  60         struct md_labels labels;
  61 };
  62
  63 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
  64
  65 static u16 key_to_nfproto(const struct sw_flow_key *key)
  66 {
  67         switch (ntohs(key->eth.type)) {
  68         case ETH_P_IP:
  69                 return NFPROTO_IPV4;
  70         case ETH_P_IPV6:
  71                 return NFPROTO_IPV6;
  72         default:
  73                 return NFPROTO_UNSPEC;
  74         }
  75 }
  76
  77 /* Map SKB connection state into the values used by flow definition. */
  78 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
  79 {
  80         u8 ct_state = OVS_CS_F_TRACKED;
  81
  82         switch (ctinfo) {
  83         case IP_CT_ESTABLISHED_REPLY:
  84         case IP_CT_RELATED_REPLY:
  85         case IP_CT_NEW_REPLY:
  86                 ct_state |= OVS_CS_F_REPLY_DIR;
  87                 break;
  88         default:
  89                 break;
  90         }
  91
  92         switch (ctinfo) {
  93         case IP_CT_ESTABLISHED:
  94         case IP_CT_ESTABLISHED_REPLY:
  95                 ct_state |= OVS_CS_F_ESTABLISHED;
  96                 break;
  97         case IP_CT_RELATED:
  98         case IP_CT_RELATED_REPLY:
  99                 ct_state |= OVS_CS_F_RELATED;
 100                 break;
 101         case IP_CT_NEW:
 102         case IP_CT_NEW_REPLY:
 103                 ct_state |= OVS_CS_F_NEW;
 104                 break;
 105         default:
 106                 break;
 107         }
 108
 109         return ct_state;
 110 }
 111
 112 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 113 {
 114 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 115         return ct ? ct->mark : 0;
 116 #else
 117         return 0;
 118 #endif
 119 }
 120
 121 static void ovs_ct_get_labels(const struct nf_conn *ct,
 122                               struct ovs_key_ct_labels *labels)
 123 {
 124         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 125
 126         if (cl) {
 127                 size_t len = cl->words * sizeof(long);
 128
 129                 if (len > OVS_CT_LABELS_LEN)
 130                         len = OVS_CT_LABELS_LEN;
 131                 else if (len < OVS_CT_LABELS_LEN)
 132                         memset(labels, 0, OVS_CT_LABELS_LEN);
 133                 memcpy(labels, cl->bits, len);
 134         } else {
 135                 memset(labels, 0, OVS_CT_LABELS_LEN);
 136         }
 137 }
 138
 139 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 140                                 const struct nf_conntrack_zone *zone,
 141                                 const struct nf_conn *ct)
 142 {
 143         key->ct.state = state;
 144         key->ct.zone = zone->id;
 145         key->ct.mark = ovs_ct_get_mark(ct);
 146         ovs_ct_get_labels(ct, &key->ct.labels);
 147 }
 148
 149 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
 150  * previously sent the packet to conntrack via the ct action.
 151  */
 152 static void ovs_ct_update_key(const struct sk_buff *skb,
 153                               const struct ovs_conntrack_info *info,
 154                               struct sw_flow_key *key, bool post_ct)
 155 {
 156         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 157         enum ip_conntrack_info ctinfo;
 158         struct nf_conn *ct;
 159         u8 state = 0;
 160
 161         ct = nf_ct_get(skb, &ctinfo);
 162         if (ct) {
 163                 state = ovs_ct_get_state(ctinfo);
 164                 if (!nf_ct_is_confirmed(ct))
 165                         state |= OVS_CS_F_NEW;
 166                 if (ct->master)
 167                         state |= OVS_CS_F_RELATED;
 168                 zone = nf_ct_zone(ct);
 169         } else if (post_ct) {
 170                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 171                 if (info)
 172                         zone = &info->zone;
 173         }
 174         __ovs_ct_update_key(key, state, zone, ct);
 175 }
 176
 177 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 178 {
 179         ovs_ct_update_key(skb, NULL, key, false);
 180 }
 181
 182 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 183 {
 184         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 185                 return -EMSGSIZE;
 186
 187         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 188             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 189                 return -EMSGSIZE;
 190
 191         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 192             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 193                 return -EMSGSIZE;
 194
 195         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 196             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 197                     &key->ct.labels))
 198                 return -EMSGSIZE;
 199
 200         return 0;
 201 }
 202
 203 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 204                            u32 ct_mark, u32 mask)
 205 {
 206 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 207         enum ip_conntrack_info ctinfo;
 208         struct nf_conn *ct;
 209         u32 new_mark;
 210
 211
 212         /* The connection could be invalid, in which case set_mark is no-op. */
 213         ct = nf_ct_get(skb, &ctinfo);
 214         if (!ct)
 215                 return 0;
 216
 217         new_mark = ct_mark | (ct->mark & ~(mask));
 218         if (ct->mark != new_mark) {
 219                 ct->mark = new_mark;
 220                 nf_conntrack_event_cache(IPCT_MARK, ct);
 221                 key->ct.mark = new_mark;
 222         }
 223
 224         return 0;
 225 #else
 226         return -ENOTSUPP;
 227 #endif
 228 }
 229
 230 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 231                              const struct ovs_key_ct_labels *labels,
 232                              const struct ovs_key_ct_labels *mask)
 233 {
 234         enum ip_conntrack_info ctinfo;
 235         struct nf_conn_labels *cl;
 236         struct nf_conn *ct;
 237         int err;
 238
 239         /* The connection could be invalid, in which case set_label is no-op.*/
 240         ct = nf_ct_get(skb, &ctinfo);
 241         if (!ct)
 242                 return 0;
 243
 244         cl = nf_ct_labels_find(ct);
 245         if (!cl) {
 246                 nf_ct_labels_ext_add(ct);
 247                 cl = nf_ct_labels_find(ct);
 248         }
 249         if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
 250                 return -ENOSPC;
 251
 252         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 253                                     OVS_CT_LABELS_LEN / sizeof(u32));
 254         if (err)
 255                 return err;
 256
 257         ovs_ct_get_labels(ct, &key->ct.labels);
 258         return 0;
 259 }
 260
 261 /* 'skb' should already be pulled to nh_ofs. */
 262 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 263 {
 264         const struct nf_conntrack_helper *helper;
 265         const struct nf_conn_help *help;
 266         enum ip_conntrack_info ctinfo;
 267         unsigned int protoff;
 268         struct nf_conn *ct;
 269
 270         ct = nf_ct_get(skb, &ctinfo);
 271         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 272                 return NF_ACCEPT;
 273
 274         help = nfct_help(ct);
 275         if (!help)
 276                 return NF_ACCEPT;
 277
 278         helper = rcu_dereference(help->helper);
 279         if (!helper)
 280                 return NF_ACCEPT;
 281
 282         switch (proto) {
 283         case NFPROTO_IPV4:
 284                 protoff = ip_hdrlen(skb);
 285                 break;
 286         case NFPROTO_IPV6: {
 287                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 288                 __be16 frag_off;
 289                 int ofs;
 290
 291                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 292                                        &frag_off);
 293                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 294                         pr_debug("proto header not found\n");
 295                         return NF_ACCEPT;
 296                 }
 297                 protoff = ofs;
 298                 break;
 299         }
 300         default:
 301                 WARN_ONCE(1, "helper invoked on non-IP family!");
 302                 return NF_DROP;
 303         }
 304
 305         return helper->help(skb, protoff, ct, ctinfo);
 306 }
 307
 308 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 309  * value if 'skb' is freed.
 310  */
 311 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 312                             u16 zone, struct sk_buff *skb)
 313 {
 314         struct ovs_gso_cb ovs_cb = *OVS_GSO_CB(skb);
 315
 316         if (!skb->dev) {
 317                 OVS_NLERR(true, "%s: skb has no dev; dropping", __func__);
 318                 return -EINVAL;
 319         }
 320
 321         if (key->eth.type == htons(ETH_P_IP)) {
 322                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 323                 int err;
 324
 325                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 326                 err = ip_defrag(skb, user);
 327                 if (err)
 328                         return err;
 329
 330                 ovs_cb.dp_cb.mru = IPCB(skb)->frag_max_size;
 331 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 332         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 333                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 334                 struct sk_buff *reasm;
 335
 336                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 337                 reasm = nf_ct_frag6_gather(skb, user);
 338                 if (!reasm)
 339                         return -EINPROGRESS;
 340
 341                 if (skb == reasm) {
 342                         kfree_skb(skb);
 343                         return -EINVAL;
 344                 }
 345
 346                 /* Don't free 'skb' even though it is one of the original
 347                  * fragments, as we're going to morph it into the head.
 348                  */
 349                 skb_get(skb);
 350                 nf_ct_frag6_consume_orig(reasm);
 351
 352                 key->ip.proto = ipv6_hdr(reasm)->nexthdr;
 353                 skb_morph(skb, reasm);
 354                 skb->next = reasm->next;
 355                 consume_skb(reasm);
 356                 ovs_cb.dp_cb.mru = IP6CB(skb)->frag_max_size;
 357 #endif /* IP frag support */
 358         } else {
 359                 kfree_skb(skb);
 360                 return -EPFNOSUPPORT;
 361         }
 362
 363         key->ip.frag = OVS_FRAG_TYPE_NONE;
 364         skb_clear_hash(skb);
 365         skb->ignore_df = 1;
 366         *OVS_GSO_CB(skb) = ovs_cb;
 367
 368         return 0;
 369 }
 370
 371 static struct nf_conntrack_expect *
 372 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 373                    u16 proto, const struct sk_buff *skb)
 374 {
 375         struct nf_conntrack_tuple tuple;
 376
 377         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, &tuple))
 378                 return NULL;
 379         return __nf_ct_expect_find(net, zone, &tuple);
 380 }
 381
 382 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 383 static bool skb_nfct_cached(const struct net *net, const struct sk_buff *skb,
 384                             const struct ovs_conntrack_info *info)
 385 {
 386         enum ip_conntrack_info ctinfo;
 387         struct nf_conn *ct;
 388
 389         ct = nf_ct_get(skb, &ctinfo);
 390         if (!ct)
 391                 return false;
 392         if (!net_eq(net, read_pnet(&ct->ct_net)))
 393                 return false;
 394         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 395                 return false;
 396         if (info->helper) {
 397                 struct nf_conn_help *help;
 398
 399                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 400                 if (help && rcu_access_pointer(help->helper) != info->helper)
 401                         return false;
 402         }
 403
 404         return true;
 405 }
 406
 407 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 408                            const struct ovs_conntrack_info *info,
 409                            struct sk_buff *skb)
 410 {
 411         /* If we are recirculating packets to match on conntrack fields and
 412          * committing with a separate conntrack action,  then we don't need to
 413          * actually run the packet through conntrack twice unless it's for a
 414          * different zone.
 415          */
 416         if (!skb_nfct_cached(net, skb, info)) {
 417                 struct nf_conn *tmpl = info->ct;
 418
 419                 /* Associate skb with specified zone. */
 420                 if (tmpl) {
 421                         if (skb->nfct)
 422                                 nf_conntrack_put(skb->nfct);
 423                         nf_conntrack_get(&tmpl->ct_general);
 424                         skb->nfct = &tmpl->ct_general;
 425                         skb->nfctinfo = IP_CT_NEW;
 426                 }
 427
 428                 if (nf_conntrack_in(net, info->family, NF_INET_FORWARD,
 429                                     skb) != NF_ACCEPT)
 430                         return -ENOENT;
 431
 432                 if (ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 433                         WARN_ONCE(1, "helper rejected packet");
 434                         return -EINVAL;
 435                 }
 436         }
 437
 438         ovs_ct_update_key(skb, info, key, true);
 439
 440         return 0;
 441 }
 442
 443 /* Lookup connection and read fields into key. */
 444 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 445                          const struct ovs_conntrack_info *info,
 446                          struct sk_buff *skb)
 447 {
 448         struct nf_conntrack_expect *exp;
 449
 450         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 451         if (exp) {
 452                 u8 state;
 453
 454                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 455                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 456         } else {
 457                 int err;
 458
 459                 err = __ovs_ct_lookup(net, key, info, skb);
 460                 if (err)
 461                         return err;
 462         }
 463
 464         return 0;
 465 }
 466
 467 /* Lookup connection and confirm if unconfirmed. */
 468 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 469                          const struct ovs_conntrack_info *info,
 470                          struct sk_buff *skb)
 471 {
 472         u8 state;
 473         int err;
 474
 475         state = key->ct.state;
 476         if (key->ct.zone == info->zone.id &&
 477             ((state & OVS_CS_F_TRACKED) && !(state & OVS_CS_F_NEW))) {
 478                 /* Previous lookup has shown that this connection is already
 479                  * tracked and committed. Skip committing.
 480                  */
 481                 return 0;
 482         }
 483
 484         err = __ovs_ct_lookup(net, key, info, skb);
 485         if (err)
 486                 return err;
 487         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 488                 return -EINVAL;
 489
 490         return 0;
 491 }
 492
 493 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 494 {
 495         size_t i;
 496
 497         for (i = 0; i < sizeof(*labels); i++)
 498                 if (labels->ct_labels[i])
 499                         return true;
 500
 501         return false;
 502 }
 503
 504 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 505  * value if 'skb' is freed.
 506  */
 507 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 508                    struct sw_flow_key *key,
 509                    const struct ovs_conntrack_info *info)
 510 {
 511         int nh_ofs;
 512         int err;
 513
 514         /* The conntrack module expects to be working at L3. */
 515         nh_ofs = skb_network_offset(skb);
 516         skb_pull(skb, nh_ofs);
 517
 518         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 519                 err = handle_fragments(net, key, info->zone.id, skb);
 520                 if (err)
 521                         return err;
 522         }
 523
 524         if (info->commit)
 525                 err = ovs_ct_commit(net, key, info, skb);
 526         else
 527                 err = ovs_ct_lookup(net, key, info, skb);
 528         if (err)
 529                 goto err;
 530
 531         if (info->mark.mask) {
 532                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 533                                       info->mark.mask);
 534                 if (err)
 535                         goto err;
 536         }
 537         if (labels_nonzero(&info->labels.mask))
 538                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 539                                         &info->labels.mask);
 540 err:
 541         skb_push(skb, nh_ofs);
 542         if (err)
 543                 kfree_skb(skb);
 544         return err;
 545 }
 546
 547 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 548                              const struct sw_flow_key *key, bool log)
 549 {
 550         struct nf_conntrack_helper *helper;
 551         struct nf_conn_help *help;
 552
 553         helper = nf_conntrack_helper_try_module_get(name, info->family,
 554                                                     key->ip.proto);
 555         if (!helper) {
 556                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 557                 return -EINVAL;
 558         }
 559
 560         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 561         if (!help) {
 562                 module_put(helper->me);
 563                 return -ENOMEM;
 564         }
 565
 566         rcu_assign_pointer(help->helper, helper);
 567         info->helper = helper;
 568         return 0;
 569 }
 570
 571 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
 572         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
 573         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
 574                                     .maxlen = sizeof(u16) },
 575         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
 576                                     .maxlen = sizeof(struct md_mark) },
 577         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
 578                                     .maxlen = sizeof(struct md_labels) },
 579         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
 580                                     .maxlen = NF_CT_HELPER_NAME_LEN }
 581 };
 582
 583 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
 584                     const char **helper, bool log)
 585 {
 586         struct nlattr *a;
 587         int rem;
 588
 589         nla_for_each_nested(a, attr, rem) {
 590                 int type = nla_type(a);
 591                 int maxlen = ovs_ct_attr_lens[type].maxlen;
 592                 int minlen = ovs_ct_attr_lens[type].minlen;
 593
 594                 if (type > OVS_CT_ATTR_MAX) {
 595                         OVS_NLERR(log,
 596                                   "Unknown conntrack attr (type=%d, max=%d)",
 597                                   type, OVS_CT_ATTR_MAX);
 598                         return -EINVAL;
 599                 }
 600                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
 601                         OVS_NLERR(log,
 602                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
 603                                   type, nla_len(a), maxlen);
 604                         return -EINVAL;
 605                 }
 606
 607                 switch (type) {
 608                 case OVS_CT_ATTR_COMMIT:
 609                         info->commit = true;
 610                         break;
 611 #ifdef CONFIG_NF_CONNTRACK_ZONES
 612                 case OVS_CT_ATTR_ZONE:
 613                         info->zone.id = nla_get_u16(a);
 614                         break;
 615 #endif
 616 #ifdef CONFIG_NF_CONNTRACK_MARK
 617                 case OVS_CT_ATTR_MARK: {
 618                         struct md_mark *mark = nla_data(a);
 619
 620                         if (!mark->mask) {
 621                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
 622                                 return -EINVAL;
 623                         }
 624                         info->mark = *mark;
 625                         break;
 626                 }
 627 #endif
 628 #ifdef CONFIG_NF_CONNTRACK_LABELS
 629                 case OVS_CT_ATTR_LABELS: {
 630                         struct md_labels *labels = nla_data(a);
 631
 632                         if (!labels_nonzero(&labels->mask)) {
 633                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
 634                                 return -EINVAL;
 635                         }
 636                         info->labels = *labels;
 637                         break;
 638                 }
 639 #endif
 640                 case OVS_CT_ATTR_HELPER:
 641                         *helper = nla_data(a);
 642                         if (!memchr(*helper, '\0', nla_len(a))) {
 643                                 OVS_NLERR(log, "Invalid conntrack helper");
 644                                 return -EINVAL;
 645                         }
 646                         break;
 647                 default:
 648                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
 649                                   type);
 650                         return -EINVAL;
 651                 }
 652         }
 653
 654         if (rem > 0) {
 655                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
 656                 return -EINVAL;
 657         }
 658
 659         return 0;
 660 }
 661
 662 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
 663 {
 664         if (attr == OVS_KEY_ATTR_CT_STATE)
 665                 return true;
 666         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 667             attr == OVS_KEY_ATTR_CT_ZONE)
 668                 return true;
 669         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 670             attr == OVS_KEY_ATTR_CT_MARK)
 671                 return true;
 672         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 673             attr == OVS_KEY_ATTR_CT_LABELS) {
 674                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 675
 676                 return ovs_net->xt_label;
 677         }
 678
 679         return false;
 680 }
 681
 682 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
 683                        const struct sw_flow_key *key,
 684                        struct sw_flow_actions **sfa,  bool log)
 685 {
 686         struct ovs_conntrack_info ct_info;
 687         const char *helper = NULL;
 688         u16 family;
 689         int err;
 690
 691         family = key_to_nfproto(key);
 692         if (family == NFPROTO_UNSPEC) {
 693                 OVS_NLERR(log, "ct family unspecified");
 694                 return -EINVAL;
 695         }
 696
 697         memset(&ct_info, 0, sizeof(ct_info));
 698         ct_info.family = family;
 699
 700         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
 701                         NF_CT_DEFAULT_ZONE_DIR, 0);
 702
 703         err = parse_ct(attr, &ct_info, &helper, log);
 704         if (err)
 705                 return err;
 706
 707         /* Set up template for tracking connections in specific zones. */
 708         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
 709         if (!ct_info.ct) {
 710                 OVS_NLERR(log, "Failed to allocate conntrack template");
 711                 return -ENOMEM;
 712         }
 713         if (helper) {
 714                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
 715                 if (err)
 716                         goto err_free_ct;
 717         }
 718
 719         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
 720                                  sizeof(ct_info), log);
 721         if (err)
 722                 goto err_free_ct;
 723
 724         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
 725         nf_conntrack_get(&ct_info.ct->ct_general);
 726         return 0;
 727 err_free_ct:
 728         __ovs_ct_free_action(&ct_info);
 729         return err;
 730 }
 731
 732 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
 733                           struct sk_buff *skb)
 734 {
 735         struct nlattr *start;
 736
 737         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
 738         if (!start)
 739                 return -EMSGSIZE;
 740
 741         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
 742                 return -EMSGSIZE;
 743         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 744             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
 745                 return -EMSGSIZE;
 746         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
 747             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
 748                     &ct_info->mark))
 749                 return -EMSGSIZE;
 750         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 751             labels_nonzero(&ct_info->labels.mask) &&
 752             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
 753                     &ct_info->labels))
 754                 return -EMSGSIZE;
 755         if (ct_info->helper) {
 756                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
 757                                    ct_info->helper->name))
 758                         return -EMSGSIZE;
 759         }
 760
 761         nla_nest_end(skb, start);
 762
 763         return 0;
 764 }
 765
 766 void ovs_ct_free_action(const struct nlattr *a)
 767 {
 768         struct ovs_conntrack_info *ct_info = nla_data(a);
 769
 770         __ovs_ct_free_action(ct_info);
 771 }
 772
 773 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
 774 {
 775         if (ct_info->helper)
 776                 module_put(ct_info->helper->me);
 777         if (ct_info->ct)
 778                 nf_ct_tmpl_free(ct_info->ct);
 779 }
 780
 781 void ovs_ct_init(struct net *net)
 782 {
 783         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
 784         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 785
 786         if (nf_connlabels_get(net, n_bits)) {
 787                 ovs_net->xt_label = false;
 788                 OVS_NLERR(true, "Failed to set connlabel length");
 789         } else {
 790                 ovs_net->xt_label = true;
 791         }
 792 }
 793
 794 void ovs_ct_exit(struct net *net)
 795 {
 796         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 797
 798         if (ovs_net->xt_label)
 799                 nf_connlabels_put(net);
 800 }
 801
 802 #endif /* CONFIG_NF_CONNTRACK && LINUX > 3.10 */