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