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 <linux/tcp.h>
  17 #include <linux/udp.h>
  18 #include <linux/sctp.h>
  19 #include <net/ip.h>
  20 #include <net/netfilter/nf_conntrack_core.h>
  21 #include <net/netfilter/nf_conntrack_helper.h>
  22 #include <net/netfilter/nf_conntrack_labels.h>
  23 #include <net/netfilter/nf_conntrack_seqadj.h>
  24 #include <net/netfilter/nf_conntrack_zones.h>
  25 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  26
  27 #ifdef CONFIG_NF_NAT_NEEDED
  28 #include <linux/netfilter/nf_nat.h>
  29 #include <net/netfilter/nf_nat_core.h>
  30 #include <net/netfilter/nf_nat_l3proto.h>
  31 #endif
  32
  33 #include "datapath.h"
  34 #include "conntrack.h"
  35 #include "flow.h"
  36 #include "flow_netlink.h"
  37
  38 struct ovs_ct_len_tbl {
  39         int maxlen;
  40         int minlen;
  41 };
  42
  43 /* Metadata mark for masked write to conntrack mark */
  44 struct md_mark {
  45         u32 value;
  46         u32 mask;
  47 };
  48
  49 /* Metadata label for masked write to conntrack label. */
  50 struct md_labels {
  51         struct ovs_key_ct_labels value;
  52         struct ovs_key_ct_labels mask;
  53 };
  54
  55 enum ovs_ct_nat {
  56         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  57         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  58         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  59 };
  60
  61 /* Conntrack action context for execution. */
  62 struct ovs_conntrack_info {
  63         struct nf_conntrack_helper *helper;
  64         struct nf_conntrack_zone zone;
  65         struct nf_conn *ct;
  66         u8 commit : 1;
  67         u8 nat : 3;                 /* enum ovs_ct_nat */
  68         u8 force : 1;
  69         u8 have_eventmask : 1;
  70         u16 family;
  71         u32 eventmask;              /* Mask of 1 << IPCT_*. */
  72         struct md_mark mark;
  73         struct md_labels labels;
  74 #ifdef CONFIG_NF_NAT_NEEDED
  75         struct nf_nat_range range;  /* Only present for SRC NAT and DST NAT. */
  76 #endif
  77 };
  78
  79 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
  80
  81 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
  82
  83 static u16 key_to_nfproto(const struct sw_flow_key *key)
  84 {
  85         switch (ntohs(key->eth.type)) {
  86         case ETH_P_IP:
  87                 return NFPROTO_IPV4;
  88         case ETH_P_IPV6:
  89                 return NFPROTO_IPV6;
  90         default:
  91                 return NFPROTO_UNSPEC;
  92         }
  93 }
  94
  95 /* Map SKB connection state into the values used by flow definition. */
  96 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
  97 {
  98         u8 ct_state = OVS_CS_F_TRACKED;
  99
 100         switch (ctinfo) {
 101         case IP_CT_ESTABLISHED_REPLY:
 102         case IP_CT_RELATED_REPLY:
 103                 ct_state |= OVS_CS_F_REPLY_DIR;
 104                 break;
 105         default:
 106                 break;
 107         }
 108
 109         switch (ctinfo) {
 110         case IP_CT_ESTABLISHED:
 111         case IP_CT_ESTABLISHED_REPLY:
 112                 ct_state |= OVS_CS_F_ESTABLISHED;
 113                 break;
 114         case IP_CT_RELATED:
 115         case IP_CT_RELATED_REPLY:
 116                 ct_state |= OVS_CS_F_RELATED;
 117                 break;
 118         case IP_CT_NEW:
 119                 ct_state |= OVS_CS_F_NEW;
 120                 break;
 121         default:
 122                 break;
 123         }
 124
 125         return ct_state;
 126 }
 127
 128 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 129 {
 130 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 131         return ct ? ct->mark : 0;
 132 #else
 133         return 0;
 134 #endif
 135 }
 136
 137 /* Guard against conntrack labels max size shrinking below 128 bits. */
 138 #if NF_CT_LABELS_MAX_SIZE < 16
 139 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
 140 #endif
 141
 142 static void ovs_ct_get_labels(const struct nf_conn *ct,
 143                               struct ovs_key_ct_labels *labels)
 144 {
 145         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 146
 147         if (cl)
 148                 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
 149         else
 150                 memset(labels, 0, OVS_CT_LABELS_LEN);
 151 }
 152
 153 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
 154                                         const struct nf_conntrack_tuple *orig,
 155                                         u8 icmp_proto)
 156 {
 157         key->ct_orig_proto = orig->dst.protonum;
 158         if (orig->dst.protonum == icmp_proto) {
 159                 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
 160                 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
 161         } else {
 162                 key->ct.orig_tp.src = orig->src.u.all;
 163                 key->ct.orig_tp.dst = orig->dst.u.all;
 164         }
 165 }
 166
 167 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 168                                 const struct nf_conntrack_zone *zone,
 169                                 const struct nf_conn *ct)
 170 {
 171         key->ct_state = state;
 172         key->ct_zone = zone->id;
 173         key->ct.mark = ovs_ct_get_mark(ct);
 174         ovs_ct_get_labels(ct, &key->ct.labels);
 175
 176         if (ct) {
 177                 const struct nf_conntrack_tuple *orig;
 178
 179                 /* Use the master if we have one. */
 180                 if (ct->master)
 181                         ct = ct->master;
 182                 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 183
 184                 /* IP version must match with the master connection. */
 185                 if (key->eth.type == htons(ETH_P_IP) &&
 186                     nf_ct_l3num(ct) == NFPROTO_IPV4) {
 187                         key->ipv4.ct_orig.src = orig->src.u3.ip;
 188                         key->ipv4.ct_orig.dst = orig->dst.u3.ip;
 189                         __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
 190                         return;
 191                 } else if (key->eth.type == htons(ETH_P_IPV6) &&
 192                            !sw_flow_key_is_nd(key) &&
 193                            nf_ct_l3num(ct) == NFPROTO_IPV6) {
 194                         key->ipv6.ct_orig.src = orig->src.u3.in6;
 195                         key->ipv6.ct_orig.dst = orig->dst.u3.in6;
 196                         __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
 197                         return;
 198                 }
 199         }
 200         /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
 201          * original direction key fields.
 202          */
 203         key->ct_orig_proto = 0;
 204 }
 205
 206 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
 207  * previously sent the packet to conntrack via the ct action.  If
 208  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 209  * initialized from the connection status.
 210  */
 211 static void ovs_ct_update_key(const struct sk_buff *skb,
 212                               const struct ovs_conntrack_info *info,
 213                               struct sw_flow_key *key, bool post_ct,
 214                               bool keep_nat_flags)
 215 {
 216         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 217         enum ip_conntrack_info ctinfo;
 218         struct nf_conn *ct;
 219         u8 state = 0;
 220
 221         ct = nf_ct_get(skb, &ctinfo);
 222         if (ct) {
 223                 state = ovs_ct_get_state(ctinfo);
 224                 /* All unconfirmed entries are NEW connections. */
 225                 if (!nf_ct_is_confirmed(ct))
 226                         state |= OVS_CS_F_NEW;
 227                 /* OVS persists the related flag for the duration of the
 228                  * connection.
 229                  */
 230                 if (ct->master)
 231                         state |= OVS_CS_F_RELATED;
 232                 if (keep_nat_flags) {
 233                         state |= key->ct_state & OVS_CS_F_NAT_MASK;
 234                 } else {
 235                         if (ct->status & IPS_SRC_NAT)
 236                                 state |= OVS_CS_F_SRC_NAT;
 237                         if (ct->status & IPS_DST_NAT)
 238                                 state |= OVS_CS_F_DST_NAT;
 239                 }
 240                 zone = nf_ct_zone(ct);
 241         } else if (post_ct) {
 242                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 243                 if (info)
 244                         zone = &info->zone;
 245         }
 246         __ovs_ct_update_key(key, state, zone, ct);
 247 }
 248
 249 /* This is called to initialize CT key fields possibly coming in from the local
 250  * stack.
 251  */
 252 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 253 {
 254         ovs_ct_update_key(skb, NULL, key, false, false);
 255 }
 256
 257 #define IN6_ADDR_INITIALIZER(ADDR) \
 258         { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
 259           (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
 260
 261 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 262                    const struct sw_flow_key *output, struct sk_buff *skb)
 263 {
 264         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 265                 return -EMSGSIZE;
 266
 267         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 268             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 269                 return -EMSGSIZE;
 270
 271         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 272             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 273                 return -EMSGSIZE;
 274
 275         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 276             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 277                     &output->ct.labels))
 278                 return -EMSGSIZE;
 279
 280         if (swkey->ct_orig_proto) {
 281                 if (swkey->eth.type == htons(ETH_P_IP)) {
 282                         struct ovs_key_ct_tuple_ipv4 orig = {
 283                                 output->ipv4.ct_orig.src,
 284                                 output->ipv4.ct_orig.dst,
 285                                 output->ct.orig_tp.src,
 286                                 output->ct.orig_tp.dst,
 287                                 output->ct_orig_proto,
 288                         };
 289                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 290                                     sizeof(orig), &orig))
 291                                 return -EMSGSIZE;
 292                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 293                         struct ovs_key_ct_tuple_ipv6 orig = {
 294                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
 295                                 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
 296                                 output->ct.orig_tp.src,
 297                                 output->ct.orig_tp.dst,
 298                                 output->ct_orig_proto,
 299                         };
 300                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 301                                     sizeof(orig), &orig))
 302                                 return -EMSGSIZE;
 303                 }
 304         }
 305
 306         return 0;
 307 }
 308
 309 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 310                            u32 ct_mark, u32 mask)
 311 {
 312 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 313         u32 new_mark;
 314
 315         new_mark = ct_mark | (ct->mark & ~(mask));
 316         if (ct->mark != new_mark) {
 317                 ct->mark = new_mark;
 318                 if (nf_ct_is_confirmed(ct))
 319                         nf_conntrack_event_cache(IPCT_MARK, ct);
 320                 key->ct.mark = new_mark;
 321         }
 322
 323         return 0;
 324 #else
 325         return -ENOTSUPP;
 326 #endif
 327 }
 328
 329 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 330 {
 331         struct nf_conn_labels *cl;
 332
 333         cl = nf_ct_labels_find(ct);
 334         if (!cl) {
 335                 nf_ct_labels_ext_add(ct);
 336                 cl = nf_ct_labels_find(ct);
 337         }
 338
 339         return cl;
 340 }
 341
 342 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 343  * since the new connection is not yet confirmed, and thus no-one else has
 344  * access to it's labels, we simply write them over.
 345  */
 346 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 347                               const struct ovs_key_ct_labels *labels,
 348                               const struct ovs_key_ct_labels *mask)
 349 {
 350         struct nf_conn_labels *cl, *master_cl;
 351         bool have_mask = labels_nonzero(mask);
 352
 353         /* Inherit master's labels to the related connection? */
 354         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 355
 356         if (!master_cl && !have_mask)
 357                 return 0;   /* Nothing to do. */
 358
 359         cl = ovs_ct_get_conn_labels(ct);
 360         if (!cl)
 361                 return -ENOSPC;
 362
 363         /* Inherit the master's labels, if any. */
 364         if (master_cl)
 365                 *cl = *master_cl;
 366
 367         if (have_mask) {
 368                 u32 *dst = (u32 *)cl->bits;
 369                 int i;
 370
 371                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 372                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 373                                 (labels->ct_labels_32[i]
 374                                  & mask->ct_labels_32[i]);
 375         }
 376
 377         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 378          * IPCT_LABEL bit is set in the event cache.
 379          */
 380         nf_conntrack_event_cache(IPCT_LABEL, ct);
 381
 382         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 383
 384         return 0;
 385 }
 386
 387 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 388                              const struct ovs_key_ct_labels *labels,
 389                              const struct ovs_key_ct_labels *mask)
 390 {
 391         struct nf_conn_labels *cl;
 392         int err;
 393
 394         cl = ovs_ct_get_conn_labels(ct);
 395         if (!cl)
 396                 return -ENOSPC;
 397
 398         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 399                                     mask->ct_labels_32,
 400                                     OVS_CT_LABELS_LEN_32);
 401         if (err)
 402                 return err;
 403
 404         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 405
 406         return 0;
 407 }
 408
 409 /* 'skb' should already be pulled to nh_ofs. */
 410 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 411 {
 412         const struct nf_conntrack_helper *helper;
 413         const struct nf_conn_help *help;
 414         enum ip_conntrack_info ctinfo;
 415         unsigned int protoff;
 416         struct nf_conn *ct;
 417         int err;
 418
 419         ct = nf_ct_get(skb, &ctinfo);
 420         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 421                 return NF_ACCEPT;
 422
 423         help = nfct_help(ct);
 424         if (!help)
 425                 return NF_ACCEPT;
 426
 427         helper = rcu_dereference(help->helper);
 428         if (!helper)
 429                 return NF_ACCEPT;
 430
 431         switch (proto) {
 432         case NFPROTO_IPV4:
 433                 protoff = ip_hdrlen(skb);
 434                 break;
 435         case NFPROTO_IPV6: {
 436                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 437                 __be16 frag_off;
 438                 int ofs;
 439
 440                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 441                                        &frag_off);
 442                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 443                         pr_debug("proto header not found\n");
 444                         return NF_ACCEPT;
 445                 }
 446                 protoff = ofs;
 447                 break;
 448         }
 449         default:
 450                 WARN_ONCE(1, "helper invoked on non-IP family!");
 451                 return NF_DROP;
 452         }
 453
 454         err = helper->help(skb, protoff, ct, ctinfo);
 455         if (err != NF_ACCEPT)
 456                 return err;
 457
 458         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 459          * FTP with NAT) adusting the TCP payload size when mangling IP
 460          * addresses and/or port numbers in the text-based control connection.
 461          */
 462         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 463             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 464                 return NF_DROP;
 465         return NF_ACCEPT;
 466 }
 467
 468 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 469  * value if 'skb' is freed.
 470  */
 471 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 472                             u16 zone, struct sk_buff *skb)
 473 {
 474         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 475         int err;
 476
 477         if (key->eth.type == htons(ETH_P_IP)) {
 478                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 479
 480                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 481                 err = ip_defrag(net, skb, user);
 482                 if (err)
 483                         return err;
 484
 485                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 486 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 487         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 488                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 489
 490                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 491                 err = nf_ct_frag6_gather(net, skb, user);
 492                 if (err) {
 493                         if (err != -EINPROGRESS)
 494                                 kfree_skb(skb);
 495                         return err;
 496                 }
 497
 498                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 499                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 500 #endif
 501         } else {
 502                 kfree_skb(skb);
 503                 return -EPFNOSUPPORT;
 504         }
 505
 506         key->ip.frag = OVS_FRAG_TYPE_NONE;
 507         skb_clear_hash(skb);
 508         skb->ignore_df = 1;
 509         *OVS_CB(skb) = ovs_cb;
 510
 511         return 0;
 512 }
 513
 514 static struct nf_conntrack_expect *
 515 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 516                    u16 proto, const struct sk_buff *skb)
 517 {
 518         struct nf_conntrack_tuple tuple;
 519         struct nf_conntrack_expect *exp;
 520
 521         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 522                 return NULL;
 523
 524         exp = __nf_ct_expect_find(net, zone, &tuple);
 525         if (exp) {
 526                 struct nf_conntrack_tuple_hash *h;
 527
 528                 /* Delete existing conntrack entry, if it clashes with the
 529                  * expectation.  This can happen since conntrack ALGs do not
 530                  * check for clashes between (new) expectations and existing
 531                  * conntrack entries.  nf_conntrack_in() will check the
 532                  * expectations only if a conntrack entry can not be found,
 533                  * which can lead to OVS finding the expectation (here) in the
 534                  * init direction, but which will not be removed by the
 535                  * nf_conntrack_in() call, if a matching conntrack entry is
 536                  * found instead.  In this case all init direction packets
 537                  * would be reported as new related packets, while reply
 538                  * direction packets would be reported as un-related
 539                  * established packets.
 540                  */
 541                 h = nf_conntrack_find_get(net, zone, &tuple);
 542                 if (h) {
 543                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 544
 545                         nf_ct_delete(ct, 0, 0);
 546                         nf_conntrack_put(&ct->ct_general);
 547                 }
 548         }
 549
 550         return exp;
 551 }
 552
 553 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 554 static enum ip_conntrack_info
 555 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 556 {
 557         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 558
 559         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 560                 return IP_CT_ESTABLISHED_REPLY;
 561         /* Once we've had two way comms, always ESTABLISHED. */
 562         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 563                 return IP_CT_ESTABLISHED;
 564         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 565                 return IP_CT_RELATED;
 566         return IP_CT_NEW;
 567 }
 568
 569 /* Find an existing connection which this packet belongs to without
 570  * re-attributing statistics or modifying the connection state.  This allows an
 571  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 572  *
 573  * Must be called with rcu_read_lock.
 574  *
 575  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 576  * if there is no existing entry.
 577  */
 578 static struct nf_conn *
 579 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 580                      u8 l3num, struct sk_buff *skb, bool natted)
 581 {
 582         struct nf_conntrack_l3proto *l3proto;
 583         struct nf_conntrack_l4proto *l4proto;
 584         struct nf_conntrack_tuple tuple;
 585         struct nf_conntrack_tuple_hash *h;
 586         struct nf_conn *ct;
 587         unsigned int dataoff;
 588         u8 protonum;
 589
 590         l3proto = __nf_ct_l3proto_find(l3num);
 591         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 592                                  &protonum) <= 0) {
 593                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 594                 return NULL;
 595         }
 596         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 597         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 598                              protonum, net, &tuple, l3proto, l4proto)) {
 599                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 600                 return NULL;
 601         }
 602
 603         /* Must invert the tuple if skb has been transformed by NAT. */
 604         if (natted) {
 605                 struct nf_conntrack_tuple inverse;
 606
 607                 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
 608                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 609                         return NULL;
 610                 }
 611                 tuple = inverse;
 612         }
 613
 614         /* look for tuple match */
 615         h = nf_conntrack_find_get(net, zone, &tuple);
 616         if (!h)
 617                 return NULL;   /* Not found. */
 618
 619         ct = nf_ct_tuplehash_to_ctrack(h);
 620
 621         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 622          * select the other tuplehash to get the right 'ctinfo' bits for this
 623          * packet.
 624          */
 625         if (natted)
 626                 h = &ct->tuplehash[!h->tuple.dst.dir];
 627
 628         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 629         return ct;
 630 }
 631
 632 static
 633 struct nf_conn *ovs_ct_executed(struct net *net,
 634                                 const struct sw_flow_key *key,
 635                                 const struct ovs_conntrack_info *info,
 636                                 struct sk_buff *skb,
 637                                 bool *ct_executed)
 638 {
 639         struct nf_conn *ct = NULL;
 640
 641         /* If no ct, check if we have evidence that an existing conntrack entry
 642          * might be found for this skb.  This happens when we lose a skb->_nfct
 643          * due to an upcall, or if the direction is being forced.  If the
 644          * connection was not confirmed, it is not cached and needs to be run
 645          * through conntrack again.
 646          */
 647         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 648                        !(key->ct_state & OVS_CS_F_INVALID) &&
 649                        (key->ct_zone == info->zone.id);
 650
 651         if (*ct_executed || (!key->ct_state && info->force)) {
 652                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 653                                           !!(key->ct_state &
 654                                           OVS_CS_F_NAT_MASK));
 655         }
 656
 657         return ct;
 658 }
 659
 660 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 661 static bool skb_nfct_cached(struct net *net,
 662                             const struct sw_flow_key *key,
 663                             const struct ovs_conntrack_info *info,
 664                             struct sk_buff *skb)
 665 {
 666         enum ip_conntrack_info ctinfo;
 667         struct nf_conn *ct;
 668         bool ct_executed = true;
 669
 670         ct = nf_ct_get(skb, &ctinfo);
 671         if (!ct)
 672                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 673
 674         if (ct)
 675                 nf_ct_get(skb, &ctinfo);
 676         else
 677                 return false;
 678
 679         if (!net_eq(net, read_pnet(&ct->ct_net)))
 680                 return false;
 681         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 682                 return false;
 683         if (info->helper) {
 684                 struct nf_conn_help *help;
 685
 686                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 687                 if (help && rcu_access_pointer(help->helper) != info->helper)
 688                         return false;
 689         }
 690         /* Force conntrack entry direction to the current packet? */
 691         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 692                 /* Delete the conntrack entry if confirmed, else just release
 693                  * the reference.
 694                  */
 695                 if (nf_ct_is_confirmed(ct))
 696                         nf_ct_delete(ct, 0, 0);
 697
 698                 nf_conntrack_put(&ct->ct_general);
 699                 nf_ct_set(skb, NULL, 0);
 700                 return false;
 701         }
 702
 703         return ct_executed;
 704 }
 705
 706 #ifdef CONFIG_NF_NAT_NEEDED
 707 /* Modelled after nf_nat_ipv[46]_fn().
 708  * range is only used for new, uninitialized NAT state.
 709  * Returns either NF_ACCEPT or NF_DROP.
 710  */
 711 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 712                               enum ip_conntrack_info ctinfo,
 713                               const struct nf_nat_range *range,
 714                               enum nf_nat_manip_type maniptype)
 715 {
 716         int hooknum, nh_off, err = NF_ACCEPT;
 717
 718         nh_off = skb_network_offset(skb);
 719         skb_pull_rcsum(skb, nh_off);
 720
 721         /* See HOOK2MANIP(). */
 722         if (maniptype == NF_NAT_MANIP_SRC)
 723                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 724         else
 725                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 726
 727         switch (ctinfo) {
 728         case IP_CT_RELATED:
 729         case IP_CT_RELATED_REPLY:
 730                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 731                     skb->protocol == htons(ETH_P_IP) &&
 732                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 733                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 734                                                            hooknum))
 735                                 err = NF_DROP;
 736                         goto push;
 737                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 738                            skb->protocol == htons(ETH_P_IPV6)) {
 739                         __be16 frag_off;
 740                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 741                         int hdrlen = ipv6_skip_exthdr(skb,
 742                                                       sizeof(struct ipv6hdr),
 743                                                       &nexthdr, &frag_off);
 744
 745                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 746                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 747                                                                      ctinfo,
 748                                                                      hooknum,
 749                                                                      hdrlen))
 750                                         err = NF_DROP;
 751                                 goto push;
 752                         }
 753                 }
 754                 /* Non-ICMP, fall thru to initialize if needed. */
 755         case IP_CT_NEW:
 756                 /* Seen it before?  This can happen for loopback, retrans,
 757                  * or local packets.
 758                  */
 759                 if (!nf_nat_initialized(ct, maniptype)) {
 760                         /* Initialize according to the NAT action. */
 761                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 762                                 /* Action is set up to establish a new
 763                                  * mapping.
 764                                  */
 765                                 ? nf_nat_setup_info(ct, range, maniptype)
 766                                 : nf_nat_alloc_null_binding(ct, hooknum);
 767                         if (err != NF_ACCEPT)
 768                                 goto push;
 769                 }
 770                 break;
 771
 772         case IP_CT_ESTABLISHED:
 773         case IP_CT_ESTABLISHED_REPLY:
 774                 break;
 775
 776         default:
 777                 err = NF_DROP;
 778                 goto push;
 779         }
 780
 781         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 782 push:
 783         skb_push(skb, nh_off);
 784         skb_postpush_rcsum(skb, skb->data, nh_off);
 785
 786         return err;
 787 }
 788
 789 static void ovs_nat_update_key(struct sw_flow_key *key,
 790                                const struct sk_buff *skb,
 791                                enum nf_nat_manip_type maniptype)
 792 {
 793         if (maniptype == NF_NAT_MANIP_SRC) {
 794                 __be16 src;
 795
 796                 key->ct_state |= OVS_CS_F_SRC_NAT;
 797                 if (key->eth.type == htons(ETH_P_IP))
 798                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 799                 else if (key->eth.type == htons(ETH_P_IPV6))
 800                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 801                                sizeof(key->ipv6.addr.src));
 802                 else
 803                         return;
 804
 805                 if (key->ip.proto == IPPROTO_UDP)
 806                         src = udp_hdr(skb)->source;
 807                 else if (key->ip.proto == IPPROTO_TCP)
 808                         src = tcp_hdr(skb)->source;
 809                 else if (key->ip.proto == IPPROTO_SCTP)
 810                         src = sctp_hdr(skb)->source;
 811                 else
 812                         return;
 813
 814                 key->tp.src = src;
 815         } else {
 816                 __be16 dst;
 817
 818                 key->ct_state |= OVS_CS_F_DST_NAT;
 819                 if (key->eth.type == htons(ETH_P_IP))
 820                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 821                 else if (key->eth.type == htons(ETH_P_IPV6))
 822                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 823                                sizeof(key->ipv6.addr.dst));
 824                 else
 825                         return;
 826
 827                 if (key->ip.proto == IPPROTO_UDP)
 828                         dst = udp_hdr(skb)->dest;
 829                 else if (key->ip.proto == IPPROTO_TCP)
 830                         dst = tcp_hdr(skb)->dest;
 831                 else if (key->ip.proto == IPPROTO_SCTP)
 832                         dst = sctp_hdr(skb)->dest;
 833                 else
 834                         return;
 835
 836                 key->tp.dst = dst;
 837         }
 838 }
 839
 840 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 841 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 842                       const struct ovs_conntrack_info *info,
 843                       struct sk_buff *skb, struct nf_conn *ct,
 844                       enum ip_conntrack_info ctinfo)
 845 {
 846         enum nf_nat_manip_type maniptype;
 847         int err;
 848
 849         /* Add NAT extension if not confirmed yet. */
 850         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 851                 return NF_ACCEPT;   /* Can't NAT. */
 852
 853         /* Determine NAT type.
 854          * Check if the NAT type can be deduced from the tracked connection.
 855          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 856          * when committing.
 857          */
 858         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 859             ct->status & IPS_NAT_MASK &&
 860             (ctinfo != IP_CT_RELATED || info->commit)) {
 861                 /* NAT an established or related connection like before. */
 862                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 863                         /* This is the REPLY direction for a connection
 864                          * for which NAT was applied in the forward
 865                          * direction.  Do the reverse NAT.
 866                          */
 867                         maniptype = ct->status & IPS_SRC_NAT
 868                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 869                 else
 870                         maniptype = ct->status & IPS_SRC_NAT
 871                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 872         } else if (info->nat & OVS_CT_SRC_NAT) {
 873                 maniptype = NF_NAT_MANIP_SRC;
 874         } else if (info->nat & OVS_CT_DST_NAT) {
 875                 maniptype = NF_NAT_MANIP_DST;
 876         } else {
 877                 return NF_ACCEPT; /* Connection is not NATed. */
 878         }
 879         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 880
 881         /* Mark NAT done if successful and update the flow key. */
 882         if (err == NF_ACCEPT)
 883                 ovs_nat_update_key(key, skb, maniptype);
 884
 885         return err;
 886 }
 887 #else /* !CONFIG_NF_NAT_NEEDED */
 888 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 889                       const struct ovs_conntrack_info *info,
 890                       struct sk_buff *skb, struct nf_conn *ct,
 891                       enum ip_conntrack_info ctinfo)
 892 {
 893         return NF_ACCEPT;
 894 }
 895 #endif
 896
 897 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 898  * not done already.  Update key with new CT state after passing the packet
 899  * through conntrack.
 900  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 901  * set to NULL and 0 will be returned.
 902  */
 903 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 904                            const struct ovs_conntrack_info *info,
 905                            struct sk_buff *skb)
 906 {
 907         /* If we are recirculating packets to match on conntrack fields and
 908          * committing with a separate conntrack action,  then we don't need to
 909          * actually run the packet through conntrack twice unless it's for a
 910          * different zone.
 911          */
 912         bool cached = skb_nfct_cached(net, key, info, skb);
 913         enum ip_conntrack_info ctinfo;
 914         struct nf_conn *ct;
 915
 916         if (!cached) {
 917                 struct nf_conn *tmpl = info->ct;
 918                 int err;
 919
 920                 /* Associate skb with specified zone. */
 921                 if (tmpl) {
 922                         if (skb_nfct(skb))
 923                                 nf_conntrack_put(skb_nfct(skb));
 924                         nf_conntrack_get(&tmpl->ct_general);
 925                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 926                 }
 927
 928                 err = nf_conntrack_in(net, info->family,
 929                                       NF_INET_PRE_ROUTING, skb);
 930                 if (err != NF_ACCEPT)
 931                         return -ENOENT;
 932
 933                 /* Clear CT state NAT flags to mark that we have not yet done
 934                  * NAT after the nf_conntrack_in() call.  We can actually clear
 935                  * the whole state, as it will be re-initialized below.
 936                  */
 937                 key->ct_state = 0;
 938
 939                 /* Update the key, but keep the NAT flags. */
 940                 ovs_ct_update_key(skb, info, key, true, true);
 941         }
 942
 943         ct = nf_ct_get(skb, &ctinfo);
 944         if (ct) {
 945                 /* Packets starting a new connection must be NATted before the
 946                  * helper, so that the helper knows about the NAT.  We enforce
 947                  * this by delaying both NAT and helper calls for unconfirmed
 948                  * connections until the committing CT action.  For later
 949                  * packets NAT and Helper may be called in either order.
 950                  *
 951                  * NAT will be done only if the CT action has NAT, and only
 952                  * once per packet (per zone), as guarded by the NAT bits in
 953                  * the key->ct_state.
 954                  */
 955                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 956                     (nf_ct_is_confirmed(ct) || info->commit) &&
 957                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 958                         return -EINVAL;
 959                 }
 960
 961                 /* Userspace may decide to perform a ct lookup without a helper
 962                  * specified followed by a (recirculate and) commit with one.
 963                  * Therefore, for unconfirmed connections which we will commit,
 964                  * we need to attach the helper here.
 965                  */
 966                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 967                     info->helper && !nfct_help(ct)) {
 968                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 969                                                             GFP_ATOMIC);
 970                         if (err)
 971                                 return err;
 972                 }
 973
 974                 /* Call the helper only if:
 975                  * - nf_conntrack_in() was executed above ("!cached") for a
 976                  *   confirmed connection, or
 977                  * - When committing an unconfirmed connection.
 978                  */
 979                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 980                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 981                         return -EINVAL;
 982                 }
 983         }
 984
 985         return 0;
 986 }
 987
 988 /* Lookup connection and read fields into key. */
 989 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 990                          const struct ovs_conntrack_info *info,
 991                          struct sk_buff *skb)
 992 {
 993         struct nf_conntrack_expect *exp;
 994
 995         /* If we pass an expected packet through nf_conntrack_in() the
 996          * expectation is typically removed, but the packet could still be
 997          * lost in upcall processing.  To prevent this from happening we
 998          * perform an explicit expectation lookup.  Expected connections are
 999          * always new, and will be passed through conntrack only when they are
1000          * committed, as it is OK to remove the expectation at that time.
1001          */
1002         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1003         if (exp) {
1004                 u8 state;
1005
1006                 /* NOTE: New connections are NATted and Helped only when
1007                  * committed, so we are not calling into NAT here.
1008                  */
1009                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1010                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1011         } else {
1012                 struct nf_conn *ct;
1013                 int err;
1014
1015                 err = __ovs_ct_lookup(net, key, info, skb);
1016                 if (err)
1017                         return err;
1018
1019                 ct = (struct nf_conn *)skb_nfct(skb);
1020                 if (ct)
1021                         nf_ct_deliver_cached_events(ct);
1022         }
1023
1024         return 0;
1025 }
1026
1027 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1028 {
1029         size_t i;
1030
1031         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1032                 if (labels->ct_labels_32[i])
1033                         return true;
1034
1035         return false;
1036 }
1037
1038 /* Lookup connection and confirm if unconfirmed. */
1039 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1040                          const struct ovs_conntrack_info *info,
1041                          struct sk_buff *skb)
1042 {
1043         enum ip_conntrack_info ctinfo;
1044         struct nf_conn *ct;
1045         int err;
1046
1047         err = __ovs_ct_lookup(net, key, info, skb);
1048         if (err)
1049                 return err;
1050
1051         /* The connection could be invalid, in which case this is a no-op.*/
1052         ct = nf_ct_get(skb, &ctinfo);
1053         if (!ct)
1054                 return 0;
1055
1056         /* Set the conntrack event mask if given.  NEW and DELETE events have
1057          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1058          * typically would receive many kinds of updates.  Setting the event
1059          * mask allows those events to be filtered.  The set event mask will
1060          * remain in effect for the lifetime of the connection unless changed
1061          * by a further CT action with both the commit flag and the eventmask
1062          * option. */
1063         if (info->have_eventmask) {
1064                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1065
1066                 if (cache)
1067                         cache->ctmask = info->eventmask;
1068         }
1069
1070         /* Apply changes before confirming the connection so that the initial
1071          * conntrack NEW netlink event carries the values given in the CT
1072          * action.
1073          */
1074         if (info->mark.mask) {
1075                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1076                                       info->mark.mask);
1077                 if (err)
1078                         return err;
1079         }
1080         if (!nf_ct_is_confirmed(ct)) {
1081                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1082                                          &info->labels.mask);
1083                 if (err)
1084                         return err;
1085         } else if (labels_nonzero(&info->labels.mask)) {
1086                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1087                                         &info->labels.mask);
1088                 if (err)
1089                         return err;
1090         }
1091         /* This will take care of sending queued events even if the connection
1092          * is already confirmed.
1093          */
1094         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1095                 return -EINVAL;
1096
1097         return 0;
1098 }
1099
1100 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1101  * value if 'skb' is freed.
1102  */
1103 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1104                    struct sw_flow_key *key,
1105                    const struct ovs_conntrack_info *info)
1106 {
1107         int nh_ofs;
1108         int err;
1109
1110         /* The conntrack module expects to be working at L3. */
1111         nh_ofs = skb_network_offset(skb);
1112         skb_pull_rcsum(skb, nh_ofs);
1113
1114         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1115                 err = handle_fragments(net, key, info->zone.id, skb);
1116                 if (err)
1117                         return err;
1118         }
1119
1120         if (info->commit)
1121                 err = ovs_ct_commit(net, key, info, skb);
1122         else
1123                 err = ovs_ct_lookup(net, key, info, skb);
1124
1125         skb_push(skb, nh_ofs);
1126         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1127         if (err)
1128                 kfree_skb(skb);
1129         return err;
1130 }
1131
1132 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1133                              const struct sw_flow_key *key, bool log)
1134 {
1135         struct nf_conntrack_helper *helper;
1136         struct nf_conn_help *help;
1137
1138         helper = nf_conntrack_helper_try_module_get(name, info->family,
1139                                                     key->ip.proto);
1140         if (!helper) {
1141                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1142                 return -EINVAL;
1143         }
1144
1145         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1146         if (!help) {
1147                 nf_conntrack_helper_put(helper);
1148                 return -ENOMEM;
1149         }
1150
1151         rcu_assign_pointer(help->helper, helper);
1152         info->helper = helper;
1153         return 0;
1154 }
1155
1156 #ifdef CONFIG_NF_NAT_NEEDED
1157 static int parse_nat(const struct nlattr *attr,
1158                      struct ovs_conntrack_info *info, bool log)
1159 {
1160         struct nlattr *a;
1161         int rem;
1162         bool have_ip_max = false;
1163         bool have_proto_max = false;
1164         bool ip_vers = (info->family == NFPROTO_IPV6);
1165
1166         nla_for_each_nested(a, attr, rem) {
1167                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1168                         [OVS_NAT_ATTR_SRC] = {0, 0},
1169                         [OVS_NAT_ATTR_DST] = {0, 0},
1170                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1171                                                  sizeof(struct in6_addr)},
1172                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1173                                                  sizeof(struct in6_addr)},
1174                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1175                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1176                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1177                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1178                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1179                 };
1180                 int type = nla_type(a);
1181
1182                 if (type > OVS_NAT_ATTR_MAX) {
1183                         OVS_NLERR(log,
1184                                   "Unknown NAT attribute (type=%d, max=%d).\n",
1185                                   type, OVS_NAT_ATTR_MAX);
1186                         return -EINVAL;
1187                 }
1188
1189                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1190                         OVS_NLERR(log,
1191                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
1192                                   type, nla_len(a),
1193                                   ovs_nat_attr_lens[type][ip_vers]);
1194                         return -EINVAL;
1195                 }
1196
1197                 switch (type) {
1198                 case OVS_NAT_ATTR_SRC:
1199                 case OVS_NAT_ATTR_DST:
1200                         if (info->nat) {
1201                                 OVS_NLERR(log,
1202                                           "Only one type of NAT may be specified.\n"
1203                                           );
1204                                 return -ERANGE;
1205                         }
1206                         info->nat |= OVS_CT_NAT;
1207                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1208                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1209                         break;
1210
1211                 case OVS_NAT_ATTR_IP_MIN:
1212                         nla_memcpy(&info->range.min_addr, a,
1213                                    sizeof(info->range.min_addr));
1214                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1215                         break;
1216
1217                 case OVS_NAT_ATTR_IP_MAX:
1218                         have_ip_max = true;
1219                         nla_memcpy(&info->range.max_addr, a,
1220                                    sizeof(info->range.max_addr));
1221                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1222                         break;
1223
1224                 case OVS_NAT_ATTR_PROTO_MIN:
1225                         info->range.min_proto.all = htons(nla_get_u16(a));
1226                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1227                         break;
1228
1229                 case OVS_NAT_ATTR_PROTO_MAX:
1230                         have_proto_max = true;
1231                         info->range.max_proto.all = htons(nla_get_u16(a));
1232                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1233                         break;
1234
1235                 case OVS_NAT_ATTR_PERSISTENT:
1236                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1237                         break;
1238
1239                 case OVS_NAT_ATTR_PROTO_HASH:
1240                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1241                         break;
1242
1243                 case OVS_NAT_ATTR_PROTO_RANDOM:
1244                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1245                         break;
1246
1247                 default:
1248                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1249                         return -EINVAL;
1250                 }
1251         }
1252
1253         if (rem > 0) {
1254                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1255                 return -EINVAL;
1256         }
1257         if (!info->nat) {
1258                 /* Do not allow flags if no type is given. */
1259                 if (info->range.flags) {
1260                         OVS_NLERR(log,
1261                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1262                                   );
1263                         return -EINVAL;
1264                 }
1265                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1266         } else if (!info->commit) {
1267                 OVS_NLERR(log,
1268                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1269                           );
1270                 return -EINVAL;
1271         }
1272         /* Allow missing IP_MAX. */
1273         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1274                 memcpy(&info->range.max_addr, &info->range.min_addr,
1275                        sizeof(info->range.max_addr));
1276         }
1277         /* Allow missing PROTO_MAX. */
1278         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1279             !have_proto_max) {
1280                 info->range.max_proto.all = info->range.min_proto.all;
1281         }
1282         return 0;
1283 }
1284 #endif
1285
1286 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1287         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1288         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1289         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1290                                     .maxlen = sizeof(u16) },
1291         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1292                                     .maxlen = sizeof(struct md_mark) },
1293         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1294                                     .maxlen = sizeof(struct md_labels) },
1295         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1296                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1297 #ifdef CONFIG_NF_NAT_NEEDED
1298         /* NAT length is checked when parsing the nested attributes. */
1299         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1300 #endif
1301         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1302                                     .maxlen = sizeof(u32) },
1303 };
1304
1305 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1306                     const char **helper, bool log)
1307 {
1308         struct nlattr *a;
1309         int rem;
1310
1311         nla_for_each_nested(a, attr, rem) {
1312                 int type = nla_type(a);
1313                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1314                 int minlen = ovs_ct_attr_lens[type].minlen;
1315
1316                 if (type > OVS_CT_ATTR_MAX) {
1317                         OVS_NLERR(log,
1318                                   "Unknown conntrack attr (type=%d, max=%d)",
1319                                   type, OVS_CT_ATTR_MAX);
1320                         return -EINVAL;
1321                 }
1322                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1323                         OVS_NLERR(log,
1324                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1325                                   type, nla_len(a), maxlen);
1326                         return -EINVAL;
1327                 }
1328
1329                 switch (type) {
1330                 case OVS_CT_ATTR_FORCE_COMMIT:
1331                         info->force = true;
1332                         /* fall through. */
1333                 case OVS_CT_ATTR_COMMIT:
1334                         info->commit = true;
1335                         break;
1336 #ifdef CONFIG_NF_CONNTRACK_ZONES
1337                 case OVS_CT_ATTR_ZONE:
1338                         info->zone.id = nla_get_u16(a);
1339                         break;
1340 #endif
1341 #ifdef CONFIG_NF_CONNTRACK_MARK
1342                 case OVS_CT_ATTR_MARK: {
1343                         struct md_mark *mark = nla_data(a);
1344
1345                         if (!mark->mask) {
1346                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1347                                 return -EINVAL;
1348                         }
1349                         info->mark = *mark;
1350                         break;
1351                 }
1352 #endif
1353 #ifdef CONFIG_NF_CONNTRACK_LABELS
1354                 case OVS_CT_ATTR_LABELS: {
1355                         struct md_labels *labels = nla_data(a);
1356
1357                         if (!labels_nonzero(&labels->mask)) {
1358                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1359                                 return -EINVAL;
1360                         }
1361                         info->labels = *labels;
1362                         break;
1363                 }
1364 #endif
1365                 case OVS_CT_ATTR_HELPER:
1366                         *helper = nla_data(a);
1367                         if (!memchr(*helper, '\0', nla_len(a))) {
1368                                 OVS_NLERR(log, "Invalid conntrack helper");
1369                                 return -EINVAL;
1370                         }
1371                         break;
1372 #ifdef CONFIG_NF_NAT_NEEDED
1373                 case OVS_CT_ATTR_NAT: {
1374                         int err = parse_nat(a, info, log);
1375
1376                         if (err)
1377                                 return err;
1378                         break;
1379                 }
1380 #endif
1381                 case OVS_CT_ATTR_EVENTMASK:
1382                         info->have_eventmask = true;
1383                         info->eventmask = nla_get_u32(a);
1384                         break;
1385
1386                 default:
1387                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1388                                   type);
1389                         return -EINVAL;
1390                 }
1391         }
1392
1393 #ifdef CONFIG_NF_CONNTRACK_MARK
1394         if (!info->commit && info->mark.mask) {
1395                 OVS_NLERR(log,
1396                           "Setting conntrack mark requires 'commit' flag.");
1397                 return -EINVAL;
1398         }
1399 #endif
1400 #ifdef CONFIG_NF_CONNTRACK_LABELS
1401         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1402                 OVS_NLERR(log,
1403                           "Setting conntrack labels requires 'commit' flag.");
1404                 return -EINVAL;
1405         }
1406 #endif
1407         if (rem > 0) {
1408                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1409                 return -EINVAL;
1410         }
1411
1412         return 0;
1413 }
1414
1415 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1416 {
1417         if (attr == OVS_KEY_ATTR_CT_STATE)
1418                 return true;
1419         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1420             attr == OVS_KEY_ATTR_CT_ZONE)
1421                 return true;
1422         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1423             attr == OVS_KEY_ATTR_CT_MARK)
1424                 return true;
1425         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1426             attr == OVS_KEY_ATTR_CT_LABELS) {
1427                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1428
1429                 return ovs_net->xt_label;
1430         }
1431
1432         return false;
1433 }
1434
1435 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1436                        const struct sw_flow_key *key,
1437                        struct sw_flow_actions **sfa,  bool log)
1438 {
1439         struct ovs_conntrack_info ct_info;
1440         const char *helper = NULL;
1441         u16 family;
1442         int err;
1443
1444         family = key_to_nfproto(key);
1445         if (family == NFPROTO_UNSPEC) {
1446                 OVS_NLERR(log, "ct family unspecified");
1447                 return -EINVAL;
1448         }
1449
1450         memset(&ct_info, 0, sizeof(ct_info));
1451         ct_info.family = family;
1452
1453         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1454                         NF_CT_DEFAULT_ZONE_DIR, 0);
1455
1456         err = parse_ct(attr, &ct_info, &helper, log);
1457         if (err)
1458                 return err;
1459
1460         /* Set up template for tracking connections in specific zones. */
1461         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1462         if (!ct_info.ct) {
1463                 OVS_NLERR(log, "Failed to allocate conntrack template");
1464                 return -ENOMEM;
1465         }
1466
1467         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1468         nf_conntrack_get(&ct_info.ct->ct_general);
1469
1470         if (helper) {
1471                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1472                 if (err)
1473                         goto err_free_ct;
1474         }
1475
1476         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1477                                  sizeof(ct_info), log);
1478         if (err)
1479                 goto err_free_ct;
1480
1481         return 0;
1482 err_free_ct:
1483         __ovs_ct_free_action(&ct_info);
1484         return err;
1485 }
1486
1487 #ifdef CONFIG_NF_NAT_NEEDED
1488 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1489                                struct sk_buff *skb)
1490 {
1491         struct nlattr *start;
1492
1493         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1494         if (!start)
1495                 return false;
1496
1497         if (info->nat & OVS_CT_SRC_NAT) {
1498                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1499                         return false;
1500         } else if (info->nat & OVS_CT_DST_NAT) {
1501                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1502                         return false;
1503         } else {
1504                 goto out;
1505         }
1506
1507         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1508                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1509                     info->family == NFPROTO_IPV4) {
1510                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1511                                             info->range.min_addr.ip) ||
1512                             (info->range.max_addr.ip
1513                              != info->range.min_addr.ip &&
1514                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1515                                               info->range.max_addr.ip))))
1516                                 return false;
1517                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1518                            info->family == NFPROTO_IPV6) {
1519                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1520                                              &info->range.min_addr.in6) ||
1521                             (memcmp(&info->range.max_addr.in6,
1522                                     &info->range.min_addr.in6,
1523                                     sizeof(info->range.max_addr.in6)) &&
1524                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1525                                                &info->range.max_addr.in6))))
1526                                 return false;
1527                 } else {
1528                         return false;
1529                 }
1530         }
1531         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1532             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1533                          ntohs(info->range.min_proto.all)) ||
1534              (info->range.max_proto.all != info->range.min_proto.all &&
1535               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1536                           ntohs(info->range.max_proto.all)))))
1537                 return false;
1538
1539         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1540             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1541                 return false;
1542         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1543             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1544                 return false;
1545         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1546             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1547                 return false;
1548 out:
1549         nla_nest_end(skb, start);
1550
1551         return true;
1552 }
1553 #endif
1554
1555 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1556                           struct sk_buff *skb)
1557 {
1558         struct nlattr *start;
1559
1560         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1561         if (!start)
1562                 return -EMSGSIZE;
1563
1564         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1565                                             ? OVS_CT_ATTR_FORCE_COMMIT
1566                                             : OVS_CT_ATTR_COMMIT))
1567                 return -EMSGSIZE;
1568         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1569             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1570                 return -EMSGSIZE;
1571         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1572             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1573                     &ct_info->mark))
1574                 return -EMSGSIZE;
1575         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1576             labels_nonzero(&ct_info->labels.mask) &&
1577             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1578                     &ct_info->labels))
1579                 return -EMSGSIZE;
1580         if (ct_info->helper) {
1581                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1582                                    ct_info->helper->name))
1583                         return -EMSGSIZE;
1584         }
1585         if (ct_info->have_eventmask &&
1586             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1587                 return -EMSGSIZE;
1588
1589 #ifdef CONFIG_NF_NAT_NEEDED
1590         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1591                 return -EMSGSIZE;
1592 #endif
1593         nla_nest_end(skb, start);
1594
1595         return 0;
1596 }
1597
1598 void ovs_ct_free_action(const struct nlattr *a)
1599 {
1600         struct ovs_conntrack_info *ct_info = nla_data(a);
1601
1602         __ovs_ct_free_action(ct_info);
1603 }
1604
1605 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1606 {
1607         if (ct_info->helper)
1608                 nf_conntrack_helper_put(ct_info->helper);
1609         if (ct_info->ct)
1610                 nf_ct_tmpl_free(ct_info->ct);
1611 }
1612
1613 void ovs_ct_init(struct net *net)
1614 {
1615         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1616         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1617
1618         if (nf_connlabels_get(net, n_bits - 1)) {
1619                 ovs_net->xt_label = false;
1620                 OVS_NLERR(true, "Failed to set connlabel length");
1621         } else {
1622                 ovs_net->xt_label = true;
1623         }
1624 }
1625
1626 void ovs_ct_exit(struct net *net)
1627 {
1628         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1629
1630         if (ovs_net->xt_label)
1631                 nf_connlabels_put(net);
1632 }