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(3,9,0) && \
  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
  34 struct ovs_ct_len_tbl {
  35         size_t maxlen;
  36         size_t minlen;
  37 };
  38
  39 /* Metadata mark for masked write to conntrack mark */
  40 struct md_mark {
  41         u32 value;
  42         u32 mask;
  43 };
  44
  45 /* Metadata label for masked write to conntrack label. */
  46 struct md_labels {
  47         struct ovs_key_ct_labels value;
  48         struct ovs_key_ct_labels mask;
  49 };
  50
  51 /* Conntrack action context for execution. */
  52 struct ovs_conntrack_info {
  53         struct nf_conntrack_helper *helper;
  54         struct nf_conntrack_zone zone;
  55         struct nf_conn *ct;
  56         u8 commit : 1;
  57         u16 family;
  58         struct md_mark mark;
  59         struct md_labels labels;
  60 };
  61
  62 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
  63
  64 static u16 key_to_nfproto(const struct sw_flow_key *key)
  65 {
  66         switch (ntohs(key->eth.type)) {
  67         case ETH_P_IP:
  68                 return NFPROTO_IPV4;
  69         case ETH_P_IPV6:
  70                 return NFPROTO_IPV6;
  71         default:
  72                 return NFPROTO_UNSPEC;
  73         }
  74 }
  75
  76 /* Map SKB connection state into the values used by flow definition. */
  77 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
  78 {
  79         u8 ct_state = OVS_CS_F_TRACKED;
  80
  81         switch (ctinfo) {
  82         case IP_CT_ESTABLISHED_REPLY:
  83         case IP_CT_RELATED_REPLY:
  84         case IP_CT_NEW_REPLY:
  85                 ct_state |= OVS_CS_F_REPLY_DIR;
  86                 break;
  87         default:
  88                 break;
  89         }
  90
  91         switch (ctinfo) {
  92         case IP_CT_ESTABLISHED:
  93         case IP_CT_ESTABLISHED_REPLY:
  94                 ct_state |= OVS_CS_F_ESTABLISHED;
  95                 break;
  96         case IP_CT_RELATED:
  97         case IP_CT_RELATED_REPLY:
  98                 ct_state |= OVS_CS_F_RELATED;
  99                 break;
 100         case IP_CT_NEW:
 101         case IP_CT_NEW_REPLY:
 102                 ct_state |= OVS_CS_F_NEW;
 103                 break;
 104         default:
 105                 break;
 106         }
 107
 108         return ct_state;
 109 }
 110
 111 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 112 {
 113 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 114         return ct ? ct->mark : 0;
 115 #else
 116         return 0;
 117 #endif
 118 }
 119
 120 static void ovs_ct_get_labels(const struct nf_conn *ct,
 121                               struct ovs_key_ct_labels *labels)
 122 {
 123         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 124
 125         if (cl) {
 126                 size_t len = cl->words * sizeof(long);
 127
 128                 if (len > OVS_CT_LABELS_LEN)
 129                         len = OVS_CT_LABELS_LEN;
 130                 else if (len < OVS_CT_LABELS_LEN)
 131                         memset(labels, 0, OVS_CT_LABELS_LEN);
 132                 memcpy(labels, cl->bits, len);
 133         } else {
 134                 memset(labels, 0, OVS_CT_LABELS_LEN);
 135         }
 136 }
 137
 138 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 139                                 const struct nf_conntrack_zone *zone,
 140                                 const struct nf_conn *ct)
 141 {
 142         key->ct.state = state;
 143         key->ct.zone = zone->id;
 144         key->ct.mark = ovs_ct_get_mark(ct);
 145         ovs_ct_get_labels(ct, &key->ct.labels);
 146 }
 147
 148 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
 149  * previously sent the packet to conntrack via the ct action.
 150  */
 151 static void ovs_ct_update_key(const struct sk_buff *skb,
 152                               const struct ovs_conntrack_info *info,
 153                               struct sw_flow_key *key, bool post_ct)
 154 {
 155         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 156         enum ip_conntrack_info ctinfo;
 157         struct nf_conn *ct;
 158         u8 state = 0;
 159
 160         ct = nf_ct_get(skb, &ctinfo);
 161         if (ct) {
 162                 state = ovs_ct_get_state(ctinfo);
 163                 if (!nf_ct_is_confirmed(ct))
 164                         state |= OVS_CS_F_NEW;
 165                 if (ct->master)
 166                         state |= OVS_CS_F_RELATED;
 167                 zone = nf_ct_zone(ct);
 168         } else if (post_ct) {
 169                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 170                 if (info)
 171                         zone = &info->zone;
 172         }
 173         __ovs_ct_update_key(key, state, zone, ct);
 174 }
 175
 176 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 177 {
 178         ovs_ct_update_key(skb, NULL, key, false);
 179 }
 180
 181 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 182 {
 183         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 184                 return -EMSGSIZE;
 185
 186         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 187             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 188                 return -EMSGSIZE;
 189
 190         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 191             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 192                 return -EMSGSIZE;
 193
 194         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 195             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 196                     &key->ct.labels))
 197                 return -EMSGSIZE;
 198
 199         return 0;
 200 }
 201
 202 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 203                            u32 ct_mark, u32 mask)
 204 {
 205 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 206         enum ip_conntrack_info ctinfo;
 207         struct nf_conn *ct;
 208         u32 new_mark;
 209
 210
 211         /* The connection could be invalid, in which case set_mark is no-op. */
 212         ct = nf_ct_get(skb, &ctinfo);
 213         if (!ct)
 214                 return 0;
 215
 216         new_mark = ct_mark | (ct->mark & ~(mask));
 217         if (ct->mark != new_mark) {
 218                 ct->mark = new_mark;
 219                 nf_conntrack_event_cache(IPCT_MARK, ct);
 220                 key->ct.mark = new_mark;
 221         }
 222
 223         return 0;
 224 #else
 225         return -ENOTSUPP;
 226 #endif
 227 }
 228
 229 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 230                              const struct ovs_key_ct_labels *labels,
 231                              const struct ovs_key_ct_labels *mask)
 232 {
 233         enum ip_conntrack_info ctinfo;
 234         struct nf_conn_labels *cl;
 235         struct nf_conn *ct;
 236         int err;
 237
 238         /* The connection could be invalid, in which case set_label is no-op.*/
 239         ct = nf_ct_get(skb, &ctinfo);
 240         if (!ct)
 241                 return 0;
 242
 243         cl = nf_ct_labels_find(ct);
 244         if (!cl) {
 245                 nf_ct_labels_ext_add(ct);
 246                 cl = nf_ct_labels_find(ct);
 247         }
 248         if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
 249                 return -ENOSPC;
 250
 251         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 252                                     OVS_CT_LABELS_LEN / sizeof(u32));
 253         if (err)
 254                 return err;
 255
 256         ovs_ct_get_labels(ct, &key->ct.labels);
 257         return 0;
 258 }
 259
 260 /* 'skb' should already be pulled to nh_ofs. */
 261 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 262 {
 263         const struct nf_conntrack_helper *helper;
 264         const struct nf_conn_help *help;
 265         enum ip_conntrack_info ctinfo;
 266         unsigned int protoff;
 267         struct nf_conn *ct;
 268
 269         ct = nf_ct_get(skb, &ctinfo);
 270         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 271                 return NF_ACCEPT;
 272
 273         help = nfct_help(ct);
 274         if (!help)
 275                 return NF_ACCEPT;
 276
 277         helper = rcu_dereference(help->helper);
 278         if (!helper)
 279                 return NF_ACCEPT;
 280
 281         switch (proto) {
 282         case NFPROTO_IPV4:
 283                 protoff = ip_hdrlen(skb);
 284                 break;
 285         case NFPROTO_IPV6: {
 286                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 287                 __be16 frag_off;
 288                 int ofs;
 289
 290                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 291                                        &frag_off);
 292                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 293                         pr_debug("proto header not found\n");
 294                         return NF_ACCEPT;
 295                 }
 296                 protoff = ofs;
 297                 break;
 298         }
 299         default:
 300                 WARN_ONCE(1, "helper invoked on non-IP family!");
 301                 return NF_DROP;
 302         }
 303
 304         return helper->help(skb, protoff, ct, ctinfo);
 305 }
 306
 307 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 308  * value if 'skb' is freed.
 309  */
 310 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 311                             u16 zone, struct sk_buff *skb)
 312 {
 313         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 314
 315         if (!skb->dev) {
 316                 OVS_NLERR(true, "%s: skb has no dev; dropping", __func__);
 317                 return -EINVAL;
 318         }
 319
 320         if (key->eth.type == htons(ETH_P_IP)) {
 321                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 322                 int err;
 323
 324                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 325                 err = ip_defrag(skb, user);
 326                 if (err)
 327                         return err;
 328
 329                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 330 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 331         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 332                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 333                 struct sk_buff *reasm;
 334
 335                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 336                 reasm = nf_ct_frag6_gather(skb, user);
 337                 if (!reasm)
 338                         return -EINPROGRESS;
 339
 340                 if (skb == reasm) {
 341                         kfree_skb(skb);
 342                         return -EINVAL;
 343                 }
 344
 345                 /* Don't free 'skb' even though it is one of the original
 346                  * fragments, as we're going to morph it into the head.
 347                  */
 348                 skb_get(skb);
 349                 nf_ct_frag6_consume_orig(reasm);
 350
 351                 key->ip.proto = ipv6_hdr(reasm)->nexthdr;
 352                 skb_morph(skb, reasm);
 353                 skb->next = reasm->next;
 354                 consume_skb(reasm);
 355                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 356 #endif /* IP frag support */
 357         } else {
 358                 kfree_skb(skb);
 359                 return -EPFNOSUPPORT;
 360         }
 361
 362         key->ip.frag = OVS_FRAG_TYPE_NONE;
 363         skb_clear_hash(skb);
 364         skb->ignore_df = 1;
 365         *OVS_CB(skb) = ovs_cb;
 366
 367         return 0;
 368 }
 369
 370 static struct nf_conntrack_expect *
 371 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 372                    u16 proto, const struct sk_buff *skb)
 373 {
 374         struct nf_conntrack_tuple tuple;
 375
 376         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, &tuple))
 377                 return NULL;
 378         return __nf_ct_expect_find(net, zone, &tuple);
 379 }
 380
 381 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 382 static bool skb_nfct_cached(const struct net *net, const struct sk_buff *skb,
 383                             const struct ovs_conntrack_info *info)
 384 {
 385         enum ip_conntrack_info ctinfo;
 386         struct nf_conn *ct;
 387
 388         ct = nf_ct_get(skb, &ctinfo);
 389         if (!ct)
 390                 return false;
 391         if (!net_eq(net, read_pnet(&ct->ct_net)))
 392                 return false;
 393         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 394                 return false;
 395         if (info->helper) {
 396                 struct nf_conn_help *help;
 397
 398                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 399                 if (help && rcu_access_pointer(help->helper) != info->helper)
 400                         return false;
 401         }
 402
 403         return true;
 404 }
 405
 406 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 407                            const struct ovs_conntrack_info *info,
 408                            struct sk_buff *skb)
 409 {
 410         /* If we are recirculating packets to match on conntrack fields and
 411          * committing with a separate conntrack action,  then we don't need to
 412          * actually run the packet through conntrack twice unless it's for a
 413          * different zone.
 414          */
 415         if (!skb_nfct_cached(net, skb, info)) {
 416                 struct nf_conn *tmpl = info->ct;
 417
 418                 /* Associate skb with specified zone. */
 419                 if (tmpl) {
 420                         if (skb->nfct)
 421                                 nf_conntrack_put(skb->nfct);
 422                         nf_conntrack_get(&tmpl->ct_general);
 423                         skb->nfct = &tmpl->ct_general;
 424                         skb->nfctinfo = IP_CT_NEW;
 425                 }
 426
 427                 if (nf_conntrack_in(net, info->family, NF_INET_FORWARD,
 428                                     skb) != NF_ACCEPT)
 429                         return -ENOENT;
 430
 431                 if (ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 432                         WARN_ONCE(1, "helper rejected packet");
 433                         return -EINVAL;
 434                 }
 435         }
 436
 437         ovs_ct_update_key(skb, info, key, true);
 438
 439         return 0;
 440 }
 441
 442 /* Lookup connection and read fields into key. */
 443 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 444                          const struct ovs_conntrack_info *info,
 445                          struct sk_buff *skb)
 446 {
 447         struct nf_conntrack_expect *exp;
 448
 449         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 450         if (exp) {
 451                 u8 state;
 452
 453                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 454                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 455         } else {
 456                 int err;
 457
 458                 err = __ovs_ct_lookup(net, key, info, skb);
 459                 if (err)
 460                         return err;
 461         }
 462
 463         return 0;
 464 }
 465
 466 /* Lookup connection and confirm if unconfirmed. */
 467 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 468                          const struct ovs_conntrack_info *info,
 469                          struct sk_buff *skb)
 470 {
 471         u8 state;
 472         int err;
 473
 474         state = key->ct.state;
 475         if (key->ct.zone == info->zone.id &&
 476             ((state & OVS_CS_F_TRACKED) && !(state & OVS_CS_F_NEW))) {
 477                 /* Previous lookup has shown that this connection is already
 478                  * tracked and committed. Skip committing.
 479                  */
 480                 return 0;
 481         }
 482
 483         err = __ovs_ct_lookup(net, key, info, skb);
 484         if (err)
 485                 return err;
 486         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 487                 return -EINVAL;
 488
 489         return 0;
 490 }
 491
 492 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 493 {
 494         size_t i;
 495
 496         for (i = 0; i < sizeof(*labels); i++)
 497                 if (labels->ct_labels[i])
 498                         return true;
 499
 500         return false;
 501 }
 502
 503 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 504  * value if 'skb' is freed.
 505  */
 506 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 507                    struct sw_flow_key *key,
 508                    const struct ovs_conntrack_info *info)
 509 {
 510         int nh_ofs;
 511         int err;
 512
 513         /* The conntrack module expects to be working at L3. */
 514         nh_ofs = skb_network_offset(skb);
 515         skb_pull(skb, nh_ofs);
 516
 517         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 518                 err = handle_fragments(net, key, info->zone.id, skb);
 519                 if (err)
 520                         return err;
 521         }
 522
 523         if (info->commit)
 524                 err = ovs_ct_commit(net, key, info, skb);
 525         else
 526                 err = ovs_ct_lookup(net, key, info, skb);
 527         if (err)
 528                 goto err;
 529
 530         if (info->mark.mask) {
 531                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 532                                       info->mark.mask);
 533                 if (err)
 534                         goto err;
 535         }
 536         if (labels_nonzero(&info->labels.mask))
 537                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 538                                         &info->labels.mask);
 539 err:
 540         skb_push(skb, nh_ofs);
 541         if (err)
 542                 kfree_skb(skb);
 543         return err;
 544 }
 545
 546 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 547                              const struct sw_flow_key *key, bool log)
 548 {
 549         struct nf_conntrack_helper *helper;
 550         struct nf_conn_help *help;
 551
 552         helper = nf_conntrack_helper_try_module_get(name, info->family,
 553                                                     key->ip.proto);
 554         if (!helper) {
 555                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 556                 return -EINVAL;
 557         }
 558
 559         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 560         if (!help) {
 561                 module_put(helper->me);
 562                 return -ENOMEM;
 563         }
 564
 565         rcu_assign_pointer(help->helper, helper);
 566         info->helper = helper;
 567         return 0;
 568 }
 569
 570 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
 571         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
 572         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
 573                                     .maxlen = sizeof(u16) },
 574         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
 575                                     .maxlen = sizeof(struct md_mark) },
 576         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
 577                                     .maxlen = sizeof(struct md_labels) },
 578         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
 579                                     .maxlen = NF_CT_HELPER_NAME_LEN }
 580 };
 581
 582 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
 583                     const char **helper, bool log)
 584 {
 585         struct nlattr *a;
 586         int rem;
 587
 588         nla_for_each_nested(a, attr, rem) {
 589                 int type = nla_type(a);
 590                 int maxlen = ovs_ct_attr_lens[type].maxlen;
 591                 int minlen = ovs_ct_attr_lens[type].minlen;
 592
 593                 if (type > OVS_CT_ATTR_MAX) {
 594                         OVS_NLERR(log,
 595                                   "Unknown conntrack attr (type=%d, max=%d)",
 596                                   type, OVS_CT_ATTR_MAX);
 597                         return -EINVAL;
 598                 }
 599                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
 600                         OVS_NLERR(log,
 601                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
 602                                   type, nla_len(a), maxlen);
 603                         return -EINVAL;
 604                 }
 605
 606                 switch (type) {
 607                 case OVS_CT_ATTR_COMMIT:
 608                         info->commit = true;
 609                         break;
 610 #ifdef CONFIG_NF_CONNTRACK_ZONES
 611                 case OVS_CT_ATTR_ZONE:
 612                         info->zone.id = nla_get_u16(a);
 613                         break;
 614 #endif
 615 #ifdef CONFIG_NF_CONNTRACK_MARK
 616                 case OVS_CT_ATTR_MARK: {
 617                         struct md_mark *mark = nla_data(a);
 618
 619                         if (!mark->mask) {
 620                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
 621                                 return -EINVAL;
 622                         }
 623                         info->mark = *mark;
 624                         break;
 625                 }
 626 #endif
 627 #ifdef CONFIG_NF_CONNTRACK_LABELS
 628                 case OVS_CT_ATTR_LABELS: {
 629                         struct md_labels *labels = nla_data(a);
 630
 631                         if (!labels_nonzero(&labels->mask)) {
 632                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
 633                                 return -EINVAL;
 634                         }
 635                         info->labels = *labels;
 636                         break;
 637                 }
 638 #endif
 639                 case OVS_CT_ATTR_HELPER:
 640                         *helper = nla_data(a);
 641                         if (!memchr(*helper, '\0', nla_len(a))) {
 642                                 OVS_NLERR(log, "Invalid conntrack helper");
 643                                 return -EINVAL;
 644                         }
 645                         break;
 646                 default:
 647                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
 648                                   type);
 649                         return -EINVAL;
 650                 }
 651         }
 652
 653         if (rem > 0) {
 654                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
 655                 return -EINVAL;
 656         }
 657
 658         return 0;
 659 }
 660
 661 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
 662 {
 663         if (attr == OVS_KEY_ATTR_CT_STATE)
 664                 return true;
 665         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 666             attr == OVS_KEY_ATTR_CT_ZONE)
 667                 return true;
 668         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 669             attr == OVS_KEY_ATTR_CT_MARK)
 670                 return true;
 671         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 672             attr == OVS_KEY_ATTR_CT_LABELS) {
 673                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 674
 675                 return ovs_net->xt_label;
 676         }
 677
 678         return false;
 679 }
 680
 681 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
 682                        const struct sw_flow_key *key,
 683                        struct sw_flow_actions **sfa,  bool log)
 684 {
 685         struct ovs_conntrack_info ct_info;
 686         const char *helper = NULL;
 687         u16 family;
 688         int err;
 689
 690         family = key_to_nfproto(key);
 691         if (family == NFPROTO_UNSPEC) {
 692                 OVS_NLERR(log, "ct family unspecified");
 693                 return -EINVAL;
 694         }
 695
 696         memset(&ct_info, 0, sizeof(ct_info));
 697         ct_info.family = family;
 698
 699         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
 700                         NF_CT_DEFAULT_ZONE_DIR, 0);
 701
 702         err = parse_ct(attr, &ct_info, &helper, log);
 703         if (err)
 704                 return err;
 705
 706         /* Set up template for tracking connections in specific zones. */
 707         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
 708         if (!ct_info.ct) {
 709                 OVS_NLERR(log, "Failed to allocate conntrack template");
 710                 return -ENOMEM;
 711         }
 712         if (helper) {
 713                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
 714                 if (err)
 715                         goto err_free_ct;
 716         }
 717
 718         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
 719                                  sizeof(ct_info), log);
 720         if (err)
 721                 goto err_free_ct;
 722
 723         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
 724         nf_conntrack_get(&ct_info.ct->ct_general);
 725         return 0;
 726 err_free_ct:
 727         __ovs_ct_free_action(&ct_info);
 728         return err;
 729 }
 730
 731 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
 732                           struct sk_buff *skb)
 733 {
 734         struct nlattr *start;
 735
 736         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
 737         if (!start)
 738                 return -EMSGSIZE;
 739
 740         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
 741                 return -EMSGSIZE;
 742         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 743             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
 744                 return -EMSGSIZE;
 745         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
 746             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
 747                     &ct_info->mark))
 748                 return -EMSGSIZE;
 749         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 750             labels_nonzero(&ct_info->labels.mask) &&
 751             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
 752                     &ct_info->labels))
 753                 return -EMSGSIZE;
 754         if (ct_info->helper) {
 755                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
 756                                    ct_info->helper->name))
 757                         return -EMSGSIZE;
 758         }
 759
 760         nla_nest_end(skb, start);
 761
 762         return 0;
 763 }
 764
 765 void ovs_ct_free_action(const struct nlattr *a)
 766 {
 767         struct ovs_conntrack_info *ct_info = nla_data(a);
 768
 769         __ovs_ct_free_action(ct_info);
 770 }
 771
 772 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
 773 {
 774         if (ct_info->helper)
 775                 module_put(ct_info->helper->me);
 776         if (ct_info->ct)
 777                 nf_ct_tmpl_free(ct_info->ct);
 778 }
 779
 780 void ovs_ct_init(struct net *net)
 781 {
 782         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
 783         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 784
 785         if (nf_connlabels_get(net, n_bits)) {
 786                 ovs_net->xt_label = false;
 787                 OVS_NLERR(true, "Failed to set connlabel length");
 788         } else {
 789                 ovs_net->xt_label = true;
 790         }
 791 }
 792
 793 void ovs_ct_exit(struct net *net)
 794 {
 795         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 796
 797         if (ovs_net->xt_label)
 798                 nf_connlabels_put(net);
 799 }
 800
 801 #endif /* CONFIG_NF_CONNTRACK && LINUX > 3.9 */