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                 ct_state |= OVS_CS_F_REPLY_DIR;
  79                 break;
  80         default:
  81                 break;
  82         }
  83
  84         switch (ctinfo) {
  85         case IP_CT_ESTABLISHED:
  86         case IP_CT_ESTABLISHED_REPLY:
  87                 ct_state |= OVS_CS_F_ESTABLISHED;
  88                 break;
  89         case IP_CT_RELATED:
  90         case IP_CT_RELATED_REPLY:
  91                 ct_state |= OVS_CS_F_RELATED;
  92                 break;
  93         case IP_CT_NEW:
  94                 ct_state |= OVS_CS_F_NEW;
  95                 break;
  96         default:
  97                 break;
  98         }
  99
 100         return ct_state;
 101 }
 102
 103 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 104 {
 105 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 106         return ct ? ct->mark : 0;
 107 #else
 108         return 0;
 109 #endif
 110 }
 111
 112 static void ovs_ct_get_labels(const struct nf_conn *ct,
 113                               struct ovs_key_ct_labels *labels)
 114 {
 115         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 116
 117         if (cl) {
 118                 size_t len = cl->words * sizeof(long);
 119
 120                 if (len > OVS_CT_LABELS_LEN)
 121                         len = OVS_CT_LABELS_LEN;
 122                 else if (len < OVS_CT_LABELS_LEN)
 123                         memset(labels, 0, OVS_CT_LABELS_LEN);
 124                 memcpy(labels, cl->bits, len);
 125         } else {
 126                 memset(labels, 0, OVS_CT_LABELS_LEN);
 127         }
 128 }
 129
 130 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 131                                 const struct nf_conntrack_zone *zone,
 132                                 const struct nf_conn *ct)
 133 {
 134         key->ct.state = state;
 135         key->ct.zone = zone->id;
 136         key->ct.mark = ovs_ct_get_mark(ct);
 137         ovs_ct_get_labels(ct, &key->ct.labels);
 138 }
 139
 140 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
 141  * previously sent the packet to conntrack via the ct action.
 142  */
 143 static void ovs_ct_update_key(const struct sk_buff *skb,
 144                               const struct ovs_conntrack_info *info,
 145                               struct sw_flow_key *key, bool post_ct)
 146 {
 147         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 148         enum ip_conntrack_info ctinfo;
 149         struct nf_conn *ct;
 150         u8 state = 0;
 151
 152         ct = nf_ct_get(skb, &ctinfo);
 153         if (ct) {
 154                 state = ovs_ct_get_state(ctinfo);
 155                 /* All unconfirmed entries are NEW connections. */
 156                 if (!nf_ct_is_confirmed(ct))
 157                         state |= OVS_CS_F_NEW;
 158                 /* OVS persists the related flag for the duration of the
 159                  * connection.
 160                  */
 161                 if (ct->master)
 162                         state |= OVS_CS_F_RELATED;
 163                 zone = nf_ct_zone(ct);
 164         } else if (post_ct) {
 165                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 166                 if (info)
 167                         zone = &info->zone;
 168         }
 169         __ovs_ct_update_key(key, state, zone, ct);
 170 }
 171
 172 /* This is called to initialize CT key fields possibly coming in from the local
 173  * stack.
 174  */
 175 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 176 {
 177         ovs_ct_update_key(skb, NULL, key, false);
 178 }
 179
 180 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
 181 {
 182         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
 183                 return -EMSGSIZE;
 184
 185         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 186             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
 187                 return -EMSGSIZE;
 188
 189         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 190             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
 191                 return -EMSGSIZE;
 192
 193         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 194             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
 195                     &key->ct.labels))
 196                 return -EMSGSIZE;
 197
 198         return 0;
 199 }
 200
 201 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
 202                            u32 ct_mark, u32 mask)
 203 {
 204 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 205         enum ip_conntrack_info ctinfo;
 206         struct nf_conn *ct;
 207         u32 new_mark;
 208
 209         /* The connection could be invalid, in which case set_mark is no-op. */
 210         ct = nf_ct_get(skb, &ctinfo);
 211         if (!ct)
 212                 return 0;
 213
 214         new_mark = ct_mark | (ct->mark & ~(mask));
 215         if (ct->mark != new_mark) {
 216                 ct->mark = new_mark;
 217                 nf_conntrack_event_cache(IPCT_MARK, ct);
 218                 key->ct.mark = new_mark;
 219         }
 220
 221         return 0;
 222 #else
 223         return -ENOTSUPP;
 224 #endif
 225 }
 226
 227 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
 228                              const struct ovs_key_ct_labels *labels,
 229                              const struct ovs_key_ct_labels *mask)
 230 {
 231         enum ip_conntrack_info ctinfo;
 232         struct nf_conn_labels *cl;
 233         struct nf_conn *ct;
 234         int err;
 235
 236         /* The connection could be invalid, in which case set_label is no-op.*/
 237         ct = nf_ct_get(skb, &ctinfo);
 238         if (!ct)
 239                 return 0;
 240
 241         cl = nf_ct_labels_find(ct);
 242         if (!cl) {
 243                 nf_ct_labels_ext_add(ct);
 244                 cl = nf_ct_labels_find(ct);
 245         }
 246         if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
 247                 return -ENOSPC;
 248
 249         err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
 250                                     OVS_CT_LABELS_LEN / sizeof(u32));
 251         if (err)
 252                 return err;
 253
 254         ovs_ct_get_labels(ct, &key->ct.labels);
 255         return 0;
 256 }
 257
 258 /* 'skb' should already be pulled to nh_ofs. */
 259 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 260 {
 261         const struct nf_conntrack_helper *helper;
 262         const struct nf_conn_help *help;
 263         enum ip_conntrack_info ctinfo;
 264         unsigned int protoff;
 265         struct nf_conn *ct;
 266
 267         ct = nf_ct_get(skb, &ctinfo);
 268         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 269                 return NF_ACCEPT;
 270
 271         help = nfct_help(ct);
 272         if (!help)
 273                 return NF_ACCEPT;
 274
 275         helper = rcu_dereference(help->helper);
 276         if (!helper)
 277                 return NF_ACCEPT;
 278
 279         switch (proto) {
 280         case NFPROTO_IPV4:
 281                 protoff = ip_hdrlen(skb);
 282                 break;
 283         case NFPROTO_IPV6: {
 284                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 285                 __be16 frag_off;
 286                 int ofs;
 287
 288                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 289                                        &frag_off);
 290                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 291                         pr_debug("proto header not found\n");
 292                         return NF_ACCEPT;
 293                 }
 294                 protoff = ofs;
 295                 break;
 296         }
 297         default:
 298                 WARN_ONCE(1, "helper invoked on non-IP family!");
 299                 return NF_DROP;
 300         }
 301
 302         return helper->help(skb, protoff, ct, ctinfo);
 303 }
 304
 305 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 306  * value if 'skb' is freed.
 307  */
 308 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 309                             u16 zone, struct sk_buff *skb)
 310 {
 311         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 312         int err;
 313
 314         if (key->eth.type == htons(ETH_P_IP)) {
 315                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 316
 317                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 318                 err = ip_defrag(net, skb, user);
 319                 if (err)
 320                         return err;
 321
 322                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 323 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 324         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 325                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 326
 327                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 328                 err = nf_ct_frag6_gather(net, skb, user);
 329                 if (err)
 330                         return err;
 331
 332                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 333                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 334 #endif
 335         } else {
 336                 kfree_skb(skb);
 337                 return -EPFNOSUPPORT;
 338         }
 339
 340         key->ip.frag = OVS_FRAG_TYPE_NONE;
 341         skb_clear_hash(skb);
 342         skb->ignore_df = 1;
 343         *OVS_CB(skb) = ovs_cb;
 344
 345         return 0;
 346 }
 347
 348 static struct nf_conntrack_expect *
 349 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 350                    u16 proto, const struct sk_buff *skb)
 351 {
 352         struct nf_conntrack_tuple tuple;
 353
 354         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 355                 return NULL;
 356         return __nf_ct_expect_find(net, zone, &tuple);
 357 }
 358
 359 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 360 static enum ip_conntrack_info
 361 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 362 {
 363         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 364
 365         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 366                 return IP_CT_ESTABLISHED_REPLY;
 367         /* Once we've had two way comms, always ESTABLISHED. */
 368         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 369                 return IP_CT_ESTABLISHED;
 370         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 371                 return IP_CT_RELATED;
 372         return IP_CT_NEW;
 373 }
 374
 375 /* Find an existing connection which this packet belongs to without
 376  * re-attributing statistics or modifying the connection state.  This allows an
 377  * skb->nfct lost due to an upcall to be recovered during actions execution.
 378  *
 379  * Must be called with rcu_read_lock.
 380  *
 381  * On success, populates skb->nfct and skb->nfctinfo, and returns the
 382  * connection.  Returns NULL if there is no existing entry.
 383  */
 384 static struct nf_conn *
 385 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 386                      u8 l3num, struct sk_buff *skb)
 387 {
 388         struct nf_conntrack_l3proto *l3proto;
 389         struct nf_conntrack_l4proto *l4proto;
 390         struct nf_conntrack_tuple tuple;
 391         struct nf_conntrack_tuple_hash *h;
 392         enum ip_conntrack_info ctinfo;
 393         struct nf_conn *ct;
 394         unsigned int dataoff;
 395         u8 protonum;
 396
 397         l3proto = __nf_ct_l3proto_find(l3num);
 398         if (!l3proto) {
 399                 pr_debug("ovs_ct_find_existing: Can't get l3proto\n");
 400                 return NULL;
 401         }
 402         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 403                                  &protonum) <= 0) {
 404                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 405                 return NULL;
 406         }
 407         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 408         if (!l4proto) {
 409                 pr_debug("ovs_ct_find_existing: Can't get l4proto\n");
 410                 return NULL;
 411         }
 412         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 413                              protonum, net, &tuple, l3proto, l4proto)) {
 414                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 415                 return NULL;
 416         }
 417
 418         /* look for tuple match */
 419         h = nf_conntrack_find_get(net, zone, &tuple);
 420         if (!h)
 421                 return NULL;   /* Not found. */
 422
 423         ct = nf_ct_tuplehash_to_ctrack(h);
 424
 425         ctinfo = ovs_ct_get_info(h);
 426         if (ctinfo == IP_CT_NEW) {
 427                 /* This should not happen. */
 428                 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct);
 429         }
 430         skb->nfct = &ct->ct_general;
 431         skb->nfctinfo = ctinfo;
 432         return ct;
 433 }
 434
 435 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
 436 static bool skb_nfct_cached(struct net *net,
 437                             const struct sw_flow_key *key,
 438                             const struct ovs_conntrack_info *info,
 439                             struct sk_buff *skb)
 440 {
 441         enum ip_conntrack_info ctinfo;
 442         struct nf_conn *ct;
 443
 444         ct = nf_ct_get(skb, &ctinfo);
 445         /* If no ct, check if we have evidence that an existing conntrack entry
 446          * might be found for this skb.  This happens when we lose a skb->nfct
 447          * due to an upcall.  If the connection was not confirmed, it is not
 448          * cached and needs to be run through conntrack again.
 449          */
 450         if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
 451             !(key->ct.state & OVS_CS_F_INVALID) &&
 452             key->ct.zone == info->zone.id)
 453                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
 454         if (!ct)
 455                 return false;
 456         if (!net_eq(net, read_pnet(&ct->ct_net)))
 457                 return false;
 458         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 459                 return false;
 460         if (info->helper) {
 461                 struct nf_conn_help *help;
 462
 463                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 464                 if (help && rcu_access_pointer(help->helper) != info->helper)
 465                         return false;
 466         }
 467
 468         return true;
 469 }
 470
 471 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 472  * not done already.  Update key with new CT state after passing the packet
 473  * through conntrack.
 474  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
 475  * set to NULL and 0 will be returned.
 476  */
 477 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 478                            const struct ovs_conntrack_info *info,
 479                            struct sk_buff *skb)
 480 {
 481         /* If we are recirculating packets to match on conntrack fields and
 482          * committing with a separate conntrack action,  then we don't need to
 483          * actually run the packet through conntrack twice unless it's for a
 484          * different zone.
 485          */
 486         bool cached = skb_nfct_cached(net, key, info, skb);
 487         enum ip_conntrack_info ctinfo;
 488         struct nf_conn *ct;
 489
 490         if (!cached) {
 491                 struct nf_conn *tmpl = info->ct;
 492                 int err;
 493
 494                 /* Associate skb with specified zone. */
 495                 if (tmpl) {
 496                         if (skb->nfct)
 497                                 nf_conntrack_put(skb->nfct);
 498                         nf_conntrack_get(&tmpl->ct_general);
 499                         skb->nfct = &tmpl->ct_general;
 500                         skb->nfctinfo = IP_CT_NEW;
 501                 }
 502
 503                 /* Repeat if requested, see nf_iterate(). */
 504                 do {
 505                         err = nf_conntrack_in(net, info->family,
 506                                               NF_INET_PRE_ROUTING, skb);
 507                 } while (err == NF_REPEAT);
 508
 509                 if (err != NF_ACCEPT)
 510                         return -ENOENT;
 511
 512                 ovs_ct_update_key(skb, info, key, true);
 513         }
 514
 515         /* Call the helper only if:
 516          * - nf_conntrack_in() was executed above ("!cached") for a confirmed
 517          *   connection, or
 518          * - When committing an unconfirmed connection.
 519          */
 520         ct = nf_ct_get(skb, &ctinfo);
 521         if (ct && (nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 522             ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 523                 WARN_ONCE(1, "helper rejected packet");
 524                 return -EINVAL;
 525         }
 526
 527         return 0;
 528 }
 529
 530 /* Lookup connection and read fields into key. */
 531 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 532                          const struct ovs_conntrack_info *info,
 533                          struct sk_buff *skb)
 534 {
 535         struct nf_conntrack_expect *exp;
 536
 537         /* If we pass an expected packet through nf_conntrack_in() the
 538          * expectation is typically removed, but the packet could still be
 539          * lost in upcall processing.  To prevent this from happening we
 540          * perform an explicit expectation lookup.  Expected connections are
 541          * always new, and will be passed through conntrack only when they are
 542          * committed, as it is OK to remove the expectation at that time.
 543          */
 544         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 545         if (exp) {
 546                 u8 state;
 547
 548                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 549                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 550         } else {
 551                 int err;
 552
 553                 err = __ovs_ct_lookup(net, key, info, skb);
 554                 if (err)
 555                         return err;
 556         }
 557
 558         return 0;
 559 }
 560
 561 /* Lookup connection and confirm if unconfirmed. */
 562 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 563                          const struct ovs_conntrack_info *info,
 564                          struct sk_buff *skb)
 565 {
 566         int err;
 567
 568         err = __ovs_ct_lookup(net, key, info, skb);
 569         if (err)
 570                 return err;
 571         /* This is a no-op if the connection has already been confirmed. */
 572         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
 573                 return -EINVAL;
 574
 575         return 0;
 576 }
 577
 578 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 579 {
 580         size_t i;
 581
 582         for (i = 0; i < sizeof(*labels); i++)
 583                 if (labels->ct_labels[i])
 584                         return true;
 585
 586         return false;
 587 }
 588
 589 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 590  * value if 'skb' is freed.
 591  */
 592 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
 593                    struct sw_flow_key *key,
 594                    const struct ovs_conntrack_info *info)
 595 {
 596         int nh_ofs;
 597         int err;
 598
 599         /* The conntrack module expects to be working at L3. */
 600         nh_ofs = skb_network_offset(skb);
 601         skb_pull(skb, nh_ofs);
 602
 603         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
 604                 err = handle_fragments(net, key, info->zone.id, skb);
 605                 if (err)
 606                         return err;
 607         }
 608
 609         if (info->commit)
 610                 err = ovs_ct_commit(net, key, info, skb);
 611         else
 612                 err = ovs_ct_lookup(net, key, info, skb);
 613         if (err)
 614                 goto err;
 615
 616         if (info->mark.mask) {
 617                 err = ovs_ct_set_mark(skb, key, info->mark.value,
 618                                       info->mark.mask);
 619                 if (err)
 620                         goto err;
 621         }
 622         if (labels_nonzero(&info->labels.mask))
 623                 err = ovs_ct_set_labels(skb, key, &info->labels.value,
 624                                         &info->labels.mask);
 625 err:
 626         skb_push(skb, nh_ofs);
 627         if (err)
 628                 kfree_skb(skb);
 629         return err;
 630 }
 631
 632 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
 633                              const struct sw_flow_key *key, bool log)
 634 {
 635         struct nf_conntrack_helper *helper;
 636         struct nf_conn_help *help;
 637
 638         helper = nf_conntrack_helper_try_module_get(name, info->family,
 639                                                     key->ip.proto);
 640         if (!helper) {
 641                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
 642                 return -EINVAL;
 643         }
 644
 645         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
 646         if (!help) {
 647                 module_put(helper->me);
 648                 return -ENOMEM;
 649         }
 650
 651         rcu_assign_pointer(help->helper, helper);
 652         info->helper = helper;
 653         return 0;
 654 }
 655
 656 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
 657         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
 658         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
 659                                     .maxlen = sizeof(u16) },
 660         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
 661                                     .maxlen = sizeof(struct md_mark) },
 662         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
 663                                     .maxlen = sizeof(struct md_labels) },
 664         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
 665                                     .maxlen = NF_CT_HELPER_NAME_LEN }
 666 };
 667
 668 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
 669                     const char **helper, bool log)
 670 {
 671         struct nlattr *a;
 672         int rem;
 673
 674         nla_for_each_nested(a, attr, rem) {
 675                 int type = nla_type(a);
 676                 int maxlen = ovs_ct_attr_lens[type].maxlen;
 677                 int minlen = ovs_ct_attr_lens[type].minlen;
 678
 679                 if (type > OVS_CT_ATTR_MAX) {
 680                         OVS_NLERR(log,
 681                                   "Unknown conntrack attr (type=%d, max=%d)",
 682                                   type, OVS_CT_ATTR_MAX);
 683                         return -EINVAL;
 684                 }
 685                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
 686                         OVS_NLERR(log,
 687                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
 688                                   type, nla_len(a), maxlen);
 689                         return -EINVAL;
 690                 }
 691
 692                 switch (type) {
 693                 case OVS_CT_ATTR_COMMIT:
 694                         info->commit = true;
 695                         break;
 696 #ifdef CONFIG_NF_CONNTRACK_ZONES
 697                 case OVS_CT_ATTR_ZONE:
 698                         info->zone.id = nla_get_u16(a);
 699                         break;
 700 #endif
 701 #ifdef CONFIG_NF_CONNTRACK_MARK
 702                 case OVS_CT_ATTR_MARK: {
 703                         struct md_mark *mark = nla_data(a);
 704
 705                         if (!mark->mask) {
 706                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
 707                                 return -EINVAL;
 708                         }
 709                         info->mark = *mark;
 710                         break;
 711                 }
 712 #endif
 713 #ifdef CONFIG_NF_CONNTRACK_LABELS
 714                 case OVS_CT_ATTR_LABELS: {
 715                         struct md_labels *labels = nla_data(a);
 716
 717                         if (!labels_nonzero(&labels->mask)) {
 718                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
 719                                 return -EINVAL;
 720                         }
 721                         info->labels = *labels;
 722                         break;
 723                 }
 724 #endif
 725                 case OVS_CT_ATTR_HELPER:
 726                         *helper = nla_data(a);
 727                         if (!memchr(*helper, '\0', nla_len(a))) {
 728                                 OVS_NLERR(log, "Invalid conntrack helper");
 729                                 return -EINVAL;
 730                         }
 731                         break;
 732                 default:
 733                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
 734                                   type);
 735                         return -EINVAL;
 736                 }
 737         }
 738
 739         if (rem > 0) {
 740                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
 741                 return -EINVAL;
 742         }
 743
 744         return 0;
 745 }
 746
 747 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
 748 {
 749         if (attr == OVS_KEY_ATTR_CT_STATE)
 750                 return true;
 751         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 752             attr == OVS_KEY_ATTR_CT_ZONE)
 753                 return true;
 754         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 755             attr == OVS_KEY_ATTR_CT_MARK)
 756                 return true;
 757         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 758             attr == OVS_KEY_ATTR_CT_LABELS) {
 759                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 760
 761                 return ovs_net->xt_label;
 762         }
 763
 764         return false;
 765 }
 766
 767 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
 768                        const struct sw_flow_key *key,
 769                        struct sw_flow_actions **sfa,  bool log)
 770 {
 771         struct ovs_conntrack_info ct_info;
 772         const char *helper = NULL;
 773         u16 family;
 774         int err;
 775
 776         family = key_to_nfproto(key);
 777         if (family == NFPROTO_UNSPEC) {
 778                 OVS_NLERR(log, "ct family unspecified");
 779                 return -EINVAL;
 780         }
 781
 782         memset(&ct_info, 0, sizeof(ct_info));
 783         ct_info.family = family;
 784
 785         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
 786                         NF_CT_DEFAULT_ZONE_DIR, 0);
 787
 788         err = parse_ct(attr, &ct_info, &helper, log);
 789         if (err)
 790                 return err;
 791
 792         /* Set up template for tracking connections in specific zones. */
 793         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
 794         if (!ct_info.ct) {
 795                 OVS_NLERR(log, "Failed to allocate conntrack template");
 796                 return -ENOMEM;
 797         }
 798
 799         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
 800         nf_conntrack_get(&ct_info.ct->ct_general);
 801
 802         if (helper) {
 803                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
 804                 if (err)
 805                         goto err_free_ct;
 806         }
 807
 808         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
 809                                  sizeof(ct_info), log);
 810         if (err)
 811                 goto err_free_ct;
 812
 813         return 0;
 814 err_free_ct:
 815         __ovs_ct_free_action(&ct_info);
 816         return err;
 817 }
 818
 819 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
 820                           struct sk_buff *skb)
 821 {
 822         struct nlattr *start;
 823
 824         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
 825         if (!start)
 826                 return -EMSGSIZE;
 827
 828         if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
 829                 return -EMSGSIZE;
 830         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 831             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
 832                 return -EMSGSIZE;
 833         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
 834             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
 835                     &ct_info->mark))
 836                 return -EMSGSIZE;
 837         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 838             labels_nonzero(&ct_info->labels.mask) &&
 839             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
 840                     &ct_info->labels))
 841                 return -EMSGSIZE;
 842         if (ct_info->helper) {
 843                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
 844                                    ct_info->helper->name))
 845                         return -EMSGSIZE;
 846         }
 847
 848         nla_nest_end(skb, start);
 849
 850         return 0;
 851 }
 852
 853 void ovs_ct_free_action(const struct nlattr *a)
 854 {
 855         struct ovs_conntrack_info *ct_info = nla_data(a);
 856
 857         __ovs_ct_free_action(ct_info);
 858 }
 859
 860 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
 861 {
 862         if (ct_info->helper)
 863                 module_put(ct_info->helper->me);
 864         if (ct_info->ct)
 865                 nf_ct_put(ct_info->ct);
 866 }
 867
 868 void ovs_ct_init(struct net *net)
 869 {
 870         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
 871         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 872
 873         if (nf_connlabels_get(net, n_bits)) {
 874                 ovs_net->xt_label = false;
 875                 OVS_NLERR(true, "Failed to set connlabel length");
 876         } else {
 877                 ovs_net->xt_label = true;
 878         }
 879 }
 880
 881 void ovs_ct_exit(struct net *net)
 882 {
 883         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 884
 885         if (ovs_net->xt_label)
 886                 nf_connlabels_put(net);
 887 }