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 IS_ENABLED(CONFIG_NF_CONNTRACK)
  18
  19 #include <linux/module.h>
  20 #include <linux/openvswitch.h>
  21 #include <net/ip.h>
  22 #include <net/netfilter/nf_conntrack_core.h>
  23 #include <net/netfilter/nf_conntrack_helper.h>
  24 #include <net/netfilter/nf_conntrack_labels.h>
  25 #include <net/netfilter/nf_conntrack_zones.h>
  26 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  27
  28 #include "datapath.h"
  29 #include "conntrack.h"
  30 #include "flow.h"
  31 #include "flow_netlink.h"
  32 #include "gso.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                 ct_state |= OVS_CS_F_REPLY_DIR;
  85                 break;
  86         default:
  87                 break;
  88         }
  89
  90         switch (ctinfo) {
  91         case IP_CT_ESTABLISHED:
  92         case IP_CT_ESTABLISHED_REPLY:
  93                 ct_state |= OVS_CS_F_ESTABLISHED;
  94                 break;
  95         case IP_CT_RELATED:
  96         case IP_CT_RELATED_REPLY:
  97                 ct_state |= OVS_CS_F_RELATED;
  98                 break;
  99         case IP_CT_NEW:
 100                 ct_state |= OVS_CS_F_NEW;
 101                 break;
 102         default:
 103                 break;
 104         }
 105
 106         return ct_state;
 107 }
 108
 109 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 110 {
 111 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 112         return ct ? ct->mark : 0;
 113 #else
 114         return 0;
 115 #endif
 116 }
 117
 118 static void ovs_ct_get_labels(const struct nf_conn *ct,
 119                               struct ovs_key_ct_labels *labels)
 120 {
 121         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 122
 123         if (cl) {
 124                 size_t len = cl->words * sizeof(long);
 125
 126                 if (len > OVS_CT_LABELS_LEN)
 127                         len = OVS_CT_LABELS_LEN;
 128                 else if (len < OVS_CT_LABELS_LEN)
 129                         memset(labels, 0, OVS_CT_LABELS_LEN);
 130                 memcpy(labels, cl->bits, len);
 131         } else {
 132                 memset(labels, 0, OVS_CT_LABELS_LEN);
 133         }
 134 }
 135
 136 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 137                                 const struct nf_conntrack_zone *zone,
 138                                 const struct nf_conn *ct)
 139 {
 140         key->ct.state = state;
 141         key->ct.zone = zone->id;
 142         key->ct.mark = ovs_ct_get_mark(ct);
 143         ovs_ct_get_labels(ct, &key->ct.labels);
 144 }
 145
 146 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
 147  * previously sent the packet to conntrack via the ct action.
 148  */
 149 static void ovs_ct_update_key(const struct sk_buff *skb,
 150                               const struct ovs_conntrack_info *info,
 151                               struct sw_flow_key *key, bool post_ct)
 152 {
 153         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 154         enum ip_conntrack_info ctinfo;
 155         struct nf_conn *ct;
 156         u8 state = 0;
 157
 158         ct = nf_ct_get(skb, &ctinfo);
 159         if (ct) {
 160                 state = ovs_ct_get_state(ctinfo);
 161                 /* All unconfirmed entries are NEW connections. */
 162                 if (!nf_ct_is_confirmed(ct))
 163                         state |= OVS_CS_F_NEW;
 164                 /* OVS persists the related flag for the duration of the
 165                  * connection.
 166                  */
 167                 if (ct->master)
 168                         state |= OVS_CS_F_RELATED;
 169                 zone = nf_ct_zone(ct);
 170         } else if (post_ct) {
 171                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 172                 if (info)
 173                         zone = &info->zone;
 174         }
 175         __ovs_ct_update_key(key, state, zone, ct);
 176 }
 177
 178 /* This is called to initialize CT key fields possibly coming in from the local
 179  * stack.
 180  */
 181 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 182 {
 183         ovs_ct_update_key(skb, NULL, key, false);
 184 }
 185
 186 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 187 {
 188         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 189                 return -EMSGSIZE;
 190
 191         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 192             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 193                 return -EMSGSIZE;
 194
 195         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 196             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 197                 return -EMSGSIZE;
 198
 199         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 200             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 201                     &key->ct.labels))
 202                 return -EMSGSIZE;
 203
 204         return 0;
 205 }
 206
 207 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 208                            u32 ct_mark, u32 mask)
 209 {
 210 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 211         enum ip_conntrack_info ctinfo;
 212         struct nf_conn *ct;
 213         u32 new_mark;
 214
 215         /* The connection could be invalid, in which case set_mark is no-op. */
 216         ct = nf_ct_get(skb, &ctinfo);
 217         if (!ct)
 218                 return 0;
 219
 220         new_mark = ct_mark | (ct->mark & ~(mask));
 221         if (ct->mark != new_mark) {
 222                 ct->mark = new_mark;
 223                 nf_conntrack_event_cache(IPCT_MARK, ct);
 224                 key->ct.mark = new_mark;
 225         }
 226
 227         return 0;
 228 #else
 229         return -ENOTSUPP;
 230 #endif
 231 }
 232
 233 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 234                              const struct ovs_key_ct_labels *labels,
 235                              const struct ovs_key_ct_labels *mask)
 236 {
 237         enum ip_conntrack_info ctinfo;
 238         struct nf_conn_labels *cl;
 239         struct nf_conn *ct;
 240         int err;
 241
 242         /* The connection could be invalid, in which case set_label is no-op.*/
 243         ct = nf_ct_get(skb, &ctinfo);
 244         if (!ct)
 245                 return 0;
 246
 247         cl = nf_ct_labels_find(ct);
 248         if (!cl) {
 249                 nf_ct_labels_ext_add(ct);
 250                 cl = nf_ct_labels_find(ct);
 251         }
 252         if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
 253                 return -ENOSPC;
 254
 255         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 256                                     OVS_CT_LABELS_LEN / sizeof(u32));
 257         if (err)
 258                 return err;
 259
 260         ovs_ct_get_labels(ct, &key->ct.labels);
 261         return 0;
 262 }
 263
 264 /* 'skb' should already be pulled to nh_ofs. */
 265 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 266 {
 267         const struct nf_conntrack_helper *helper;
 268         const struct nf_conn_help *help;
 269         enum ip_conntrack_info ctinfo;
 270         unsigned int protoff;
 271         struct nf_conn *ct;
 272
 273         ct = nf_ct_get(skb, &ctinfo);
 274         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 275                 return NF_ACCEPT;
 276
 277         help = nfct_help(ct);
 278         if (!help)
 279                 return NF_ACCEPT;
 280
 281         helper = rcu_dereference(help->helper);
 282         if (!helper)
 283                 return NF_ACCEPT;
 284
 285         switch (proto) {
 286         case NFPROTO_IPV4:
 287                 protoff = ip_hdrlen(skb);
 288                 break;
 289         case NFPROTO_IPV6: {
 290                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 291                 __be16 frag_off;
 292                 int ofs;
 293
 294                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 295                                        &frag_off);
 296                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 297                         pr_debug("proto header not found\n");
 298                         return NF_ACCEPT;
 299                 }
 300                 protoff = ofs;
 301                 break;
 302         }
 303         default:
 304                 WARN_ONCE(1, "helper invoked on non-IP family!");
 305                 return NF_DROP;
 306         }
 307
 308         return helper->help(skb, protoff, ct, ctinfo);
 309 }
 310
 311 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 312  * value if 'skb' is freed.
 313  */
 314 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 315                             u16 zone, struct sk_buff *skb)
 316 {
 317         struct ovs_gso_cb ovs_cb = *OVS_GSO_CB(skb);
 318         int err;
 319
 320         if (!skb->dev) {
 321                 OVS_NLERR(true, "%s: skb has no dev; dropping", __func__);
 322                 return -EINVAL;
 323         }
 324
 325         if (key->eth.type == htons(ETH_P_IP)) {
 326                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 327
 328                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 329                 err = ip_defrag(net, skb, user);
 330                 if (err)
 331                         return err;
 332
 333                 ovs_cb.dp_cb.mru = IPCB(skb)->frag_max_size;
 334 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 335         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 336                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 337
 338                 skb_orphan(skb);
 339                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 340                 err = nf_ct_frag6_gather(net, skb, user);
 341                 if (err)
 342                         return err;
 343
 344                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 345                 ovs_cb.dp_cb.mru = IP6CB(skb)->frag_max_size;
 346 #endif /* IP frag support */
 347         } else {
 348                 kfree_skb(skb);
 349                 return -EPFNOSUPPORT;
 350         }
 351
 352         key->ip.frag = OVS_FRAG_TYPE_NONE;
 353         skb_clear_hash(skb);
 354         skb->ignore_df = 1;
 355         *OVS_GSO_CB(skb) = ovs_cb;
 356
 357         return 0;
 358 }
 359
 360 static struct nf_conntrack_expect *
 361 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 362                    u16 proto, const struct sk_buff *skb)
 363 {
 364         struct nf_conntrack_tuple tuple;
 365
 366         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 367                 return NULL;
 368         return __nf_ct_expect_find(net, zone, &tuple);
 369 }
 370
 371 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 372 static enum ip_conntrack_info
 373 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 374 {
 375         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 376
 377         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 378                 return IP_CT_ESTABLISHED_REPLY;
 379         /* Once we've had two way comms, always ESTABLISHED. */
 380         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 381                 return IP_CT_ESTABLISHED;
 382         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 383                 return IP_CT_RELATED;
 384         return IP_CT_NEW;
 385 }
 386
 387 /* Find an existing connection which this packet belongs to without
 388  * re-attributing statistics or modifying the connection state.  This allows an
 389  * skb->nfct lost due to an upcall to be recovered during actions execution.
 390  *
 391  * Must be called with rcu_read_lock.
 392  *
 393  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 394  * connection.  Returns NULL if there is no existing entry.
 395  */
 396 static struct nf_conn *
 397 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 398                      u8 l3num, struct sk_buff *skb)
 399 {
 400         struct nf_conntrack_l3proto *l3proto;
 401         struct nf_conntrack_l4proto *l4proto;
 402         struct nf_conntrack_tuple tuple;
 403         struct nf_conntrack_tuple_hash *h;
 404         enum ip_conntrack_info ctinfo;
 405         struct nf_conn *ct;
 406         unsigned int dataoff;
 407         u8 protonum;
 408
 409         l3proto = __nf_ct_l3proto_find(l3num);
 410         if (!l3proto) {
 411                 pr_debug("ovs_ct_find_existing: Can't get l3proto\n");
 412                 return NULL;
 413         }
 414         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 415                                  &protonum) <= 0) {
 416                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 417                 return NULL;
 418         }
 419         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 420         if (!l4proto) {
 421                 pr_debug("ovs_ct_find_existing: Can't get l4proto\n");
 422                 return NULL;
 423         }
 424         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 425                              protonum, net, &tuple, l3proto, l4proto)) {
 426                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 427                 return NULL;
 428         }
 429
 430         /* look for tuple match */
 431         h = nf_conntrack_find_get(net, zone, &tuple);
 432         if (!h)
 433                 return NULL;   /* Not found. */
 434
 435         ct = nf_ct_tuplehash_to_ctrack(h);
 436
 437         ctinfo = ovs_ct_get_info(h);
 438         if (ctinfo == IP_CT_NEW) {
 439                 /* This should not happen. */
 440                 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct);
 441         }
 442         skb->nfct = &ct->ct_general;
 443         skb->nfctinfo = ctinfo;
 444         return ct;
 445 }
 446
 447 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 448 static bool skb_nfct_cached(struct net *net,
 449                             const struct sw_flow_key *key,
 450                             const struct ovs_conntrack_info *info,
 451                             struct sk_buff *skb)
 452 {
 453         enum ip_conntrack_info ctinfo;
 454         struct nf_conn *ct;
 455
 456         ct = nf_ct_get(skb, &ctinfo);
 457         /* If no ct, check if we have evidence that an existing conntrack entry
 458          * might be found for this skb.  This happens when we lose a skb->nfct
 459          * due to an upcall.  If the connection was not confirmed, it is not
 460          * cached and needs to be run through conntrack again.
 461          */
 462         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 463             !(key->ct.state & OVS_CS_F_INVALID) &&
 464             key->ct.zone == info->zone.id)
 465                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 466         if (!ct)
 467                 return false;
 468         if (!net_eq(net, read_pnet(&ct->ct_net)))
 469                 return false;
 470         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 471                 return false;
 472         if (info->helper) {
 473                 struct nf_conn_help *help;
 474
 475                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 476                 if (help && rcu_access_pointer(help->helper) != info->helper)
 477                         return false;
 478         }
 479
 480         return true;
 481 }
 482
 483 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 484  * not done already.  Update key with new CT state after passing the packet
 485  * through conntrack.
 486  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 487  * set to NULL and 0 will be returned.
 488  */
 489 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 490                            const struct ovs_conntrack_info *info,
 491                            struct sk_buff *skb)
 492 {
 493         /* If we are recirculating packets to match on conntrack fields and
 494          * committing with a separate conntrack action,  then we don't need to
 495          * actually run the packet through conntrack twice unless it's for a
 496          * different zone.
 497          */
 498         if (!skb_nfct_cached(net, key, info, skb)) {
 499                 struct nf_conn *tmpl = info->ct;
 500
 501                 /* Associate skb with specified zone. */
 502                 if (tmpl) {
 503                         if (skb->nfct)
 504                                 nf_conntrack_put(skb->nfct);
 505                         nf_conntrack_get(&tmpl->ct_general);
 506                         skb->nfct = &tmpl->ct_general;
 507                         skb->nfctinfo = IP_CT_NEW;
 508                 }
 509
 510                 if (nf_conntrack_in(net, info->family, NF_INET_FORWARD,
 511                                     skb) != NF_ACCEPT)
 512                         return -ENOENT;
 513
 514                 ovs_ct_update_key(skb, info, key, true);
 515
 516                 if (ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 517                         WARN_ONCE(1, "helper rejected packet");
 518                         return -EINVAL;
 519                 }
 520         }
 521
 522         return 0;
 523 }
 524
 525 /* Lookup connection and read fields into key. */
 526 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 527                          const struct ovs_conntrack_info *info,
 528                          struct sk_buff *skb)
 529 {
 530         struct nf_conntrack_expect *exp;
 531
 532         /* If we pass an expected packet through nf_conntrack_in() the
 533          * expectation is typically removed, but the packet could still be
 534          * lost in upcall processing.  To prevent this from happening we
 535          * perform an explicit expectation lookup.  Expected connections are
 536          * always new, and will be passed through conntrack only when they are
 537          * committed, as it is OK to remove the expectation at that time.
 538          */
 539         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 540         if (exp) {
 541                 u8 state;
 542
 543                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 544                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 545         } else {
 546                 int err;
 547
 548                 err = __ovs_ct_lookup(net, key, info, skb);
 549                 if (err)
 550                         return err;
 551         }
 552
 553         return 0;
 554 }
 555
 556 /* Lookup connection and confirm if unconfirmed. */
 557 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 558                          const struct ovs_conntrack_info *info,
 559                          struct sk_buff *skb)
 560 {
 561         int err;
 562
 563         err = __ovs_ct_lookup(net, key, info, skb);
 564         if (err)
 565                 return err;
 566         /* This is a no-op if the connection has already been confirmed. */
 567         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 568                 return -EINVAL;
 569
 570         return 0;
 571 }
 572
 573 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 574 {
 575         size_t i;
 576
 577         for (i = 0; i < sizeof(*labels); i++)
 578                 if (labels->ct_labels[i])
 579                         return true;
 580
 581         return false;
 582 }
 583
 584 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 585  * value if 'skb' is freed.
 586  */
 587 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 588                    struct sw_flow_key *key,
 589                    const struct ovs_conntrack_info *info)
 590 {
 591         int nh_ofs;
 592         int err;
 593
 594         /* The conntrack module expects to be working at L3. */
 595         nh_ofs = skb_network_offset(skb);
 596         skb_pull(skb, nh_ofs);
 597
 598         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 599                 err = handle_fragments(net, key, info->zone.id, skb);
 600                 if (err)
 601                         return err;
 602         }
 603
 604         if (info->commit)
 605                 err = ovs_ct_commit(net, key, info, skb);
 606         else
 607                 err = ovs_ct_lookup(net, key, info, skb);
 608         if (err)
 609                 goto err;
 610
 611         if (info->mark.mask) {
 612                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 613                                       info->mark.mask);
 614                 if (err)
 615                         goto err;
 616         }
 617         if (labels_nonzero(&info->labels.mask))
 618                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 619                                         &info->labels.mask);
 620 err:
 621         skb_push(skb, nh_ofs);
 622         if (err)
 623                 kfree_skb(skb);
 624         return err;
 625 }
 626
 627 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 628                              const struct sw_flow_key *key, bool log)
 629 {
 630         struct nf_conntrack_helper *helper;
 631         struct nf_conn_help *help;
 632
 633         helper = nf_conntrack_helper_try_module_get(name, info->family,
 634                                                     key->ip.proto);
 635         if (!helper) {
 636                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 637                 return -EINVAL;
 638         }
 639
 640         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 641         if (!help) {
 642                 module_put(helper->me);
 643                 return -ENOMEM;
 644         }
 645
 646         rcu_assign_pointer(help->helper, helper);
 647         info->helper = helper;
 648         return 0;
 649 }
 650
 651 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
 652         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
 653         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
 654                                     .maxlen = sizeof(u16) },
 655         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
 656                                     .maxlen = sizeof(struct md_mark) },
 657         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
 658                                     .maxlen = sizeof(struct md_labels) },
 659         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
 660                                     .maxlen = NF_CT_HELPER_NAME_LEN }
 661 };
 662
 663 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
 664                     const char **helper, bool log)
 665 {
 666         struct nlattr *a;
 667         int rem;
 668
 669         nla_for_each_nested(a, attr, rem) {
 670                 int type = nla_type(a);
 671                 int maxlen = ovs_ct_attr_lens[type].maxlen;
 672                 int minlen = ovs_ct_attr_lens[type].minlen;
 673
 674                 if (type > OVS_CT_ATTR_MAX) {
 675                         OVS_NLERR(log,
 676                                   "Unknown conntrack attr (type=%d, max=%d)",
 677                                   type, OVS_CT_ATTR_MAX);
 678                         return -EINVAL;
 679                 }
 680                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
 681                         OVS_NLERR(log,
 682                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
 683                                   type, nla_len(a), maxlen);
 684                         return -EINVAL;
 685                 }
 686
 687                 switch (type) {
 688                 case OVS_CT_ATTR_COMMIT:
 689                         info->commit = true;
 690                         break;
 691 #ifdef CONFIG_NF_CONNTRACK_ZONES
 692                 case OVS_CT_ATTR_ZONE:
 693                         info->zone.id = nla_get_u16(a);
 694                         break;
 695 #endif
 696 #ifdef CONFIG_NF_CONNTRACK_MARK
 697                 case OVS_CT_ATTR_MARK: {
 698                         struct md_mark *mark = nla_data(a);
 699
 700                         if (!mark->mask) {
 701                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
 702                                 return -EINVAL;
 703                         }
 704                         info->mark = *mark;
 705                         break;
 706                 }
 707 #endif
 708 #ifdef CONFIG_NF_CONNTRACK_LABELS
 709                 case OVS_CT_ATTR_LABELS: {
 710                         struct md_labels *labels = nla_data(a);
 711
 712                         if (!labels_nonzero(&labels->mask)) {
 713                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
 714                                 return -EINVAL;
 715                         }
 716                         info->labels = *labels;
 717                         break;
 718                 }
 719 #endif
 720                 case OVS_CT_ATTR_HELPER:
 721                         *helper = nla_data(a);
 722                         if (!memchr(*helper, '\0', nla_len(a))) {
 723                                 OVS_NLERR(log, "Invalid conntrack helper");
 724                                 return -EINVAL;
 725                         }
 726                         break;
 727                 default:
 728                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
 729                                   type);
 730                         return -EINVAL;
 731                 }
 732         }
 733
 734         if (rem > 0) {
 735                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
 736                 return -EINVAL;
 737         }
 738
 739         return 0;
 740 }
 741
 742 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
 743 {
 744         if (attr == OVS_KEY_ATTR_CT_STATE)
 745                 return true;
 746         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 747             attr == OVS_KEY_ATTR_CT_ZONE)
 748                 return true;
 749         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 750             attr == OVS_KEY_ATTR_CT_MARK)
 751                 return true;
 752         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 753             attr == OVS_KEY_ATTR_CT_LABELS) {
 754                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 755
 756                 return ovs_net->xt_label;
 757         }
 758
 759         return false;
 760 }
 761
 762 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
 763                        const struct sw_flow_key *key,
 764                        struct sw_flow_actions **sfa,  bool log)
 765 {
 766         struct ovs_conntrack_info ct_info;
 767         const char *helper = NULL;
 768         u16 family;
 769         int err;
 770
 771         family = key_to_nfproto(key);
 772         if (family == NFPROTO_UNSPEC) {
 773                 OVS_NLERR(log, "ct family unspecified");
 774                 return -EINVAL;
 775         }
 776
 777         memset(&ct_info, 0, sizeof(ct_info));
 778         ct_info.family = family;
 779
 780         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
 781                         NF_CT_DEFAULT_ZONE_DIR, 0);
 782
 783         err = parse_ct(attr, &ct_info, &helper, log);
 784         if (err)
 785                 return err;
 786
 787         /* Set up template for tracking connections in specific zones. */
 788         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
 789         if (!ct_info.ct) {
 790                 OVS_NLERR(log, "Failed to allocate conntrack template");
 791                 return -ENOMEM;
 792         }
 793
 794         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
 795         nf_conntrack_get(&ct_info.ct->ct_general);
 796
 797         if (helper) {
 798                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
 799                 if (err)
 800                         goto err_free_ct;
 801         }
 802
 803         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
 804                                  sizeof(ct_info), log);
 805         if (err)
 806                 goto err_free_ct;
 807
 808         return 0;
 809 err_free_ct:
 810         __ovs_ct_free_action(&ct_info);
 811         return err;
 812 }
 813
 814 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
 815                           struct sk_buff *skb)
 816 {
 817         struct nlattr *start;
 818
 819         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
 820         if (!start)
 821                 return -EMSGSIZE;
 822
 823         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
 824                 return -EMSGSIZE;
 825         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 826             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
 827                 return -EMSGSIZE;
 828         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
 829             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
 830                     &ct_info->mark))
 831                 return -EMSGSIZE;
 832         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 833             labels_nonzero(&ct_info->labels.mask) &&
 834             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
 835                     &ct_info->labels))
 836                 return -EMSGSIZE;
 837         if (ct_info->helper) {
 838                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
 839                                    ct_info->helper->name))
 840                         return -EMSGSIZE;
 841         }
 842
 843         nla_nest_end(skb, start);
 844
 845         return 0;
 846 }
 847
 848 void ovs_ct_free_action(const struct nlattr *a)
 849 {
 850         struct ovs_conntrack_info *ct_info = nla_data(a);
 851
 852         __ovs_ct_free_action(ct_info);
 853 }
 854
 855 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
 856 {
 857         if (ct_info->helper)
 858                 module_put(ct_info->helper->me);
 859         if (ct_info->ct)
 860                 nf_ct_tmpl_free(ct_info->ct);
 861 }
 862
 863 void ovs_ct_init(struct net *net)
 864 {
 865         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
 866         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 867
 868         if (nf_connlabels_get(net, n_bits)) {
 869                 ovs_net->xt_label = false;
 870                 OVS_NLERR(true, "Failed to set connlabel length");
 871         } else {
 872                 ovs_net->xt_label = true;
 873         }
 874 }
 875
 876 void ovs_ct_exit(struct net *net)
 877 {
 878         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 879
 880         if (ovs_net->xt_label)
 881                 nf_connlabels_put(net);
 882 }
 883
 884 #endif /* CONFIG_NF_CONNTRACK */