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 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 633 static bool skb_nfct_cached(struct net *net,
 634                             const struct sw_flow_key *key,
 635                             const struct ovs_conntrack_info *info,
 636                             struct sk_buff *skb)
 637 {
 638         enum ip_conntrack_info ctinfo;
 639         struct nf_conn *ct;
 640
 641         ct = nf_ct_get(skb, &ctinfo);
 642         /* If no ct, check if we have evidence that an existing conntrack entry
 643          * might be found for this skb.  This happens when we lose a skb->_nfct
 644          * due to an upcall.  If the connection was not confirmed, it is not
 645          * cached and needs to be run through conntrack again.
 646          */
 647         if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
 648             !(key->ct_state & OVS_CS_F_INVALID) &&
 649             key->ct_zone == info->zone.id) {
 650                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 651                                           !!(key->ct_state
 652                                              & OVS_CS_F_NAT_MASK));
 653                 if (ct)
 654                         nf_ct_get(skb, &ctinfo);
 655         }
 656         if (!ct)
 657                 return false;
 658         if (!net_eq(net, read_pnet(&ct->ct_net)))
 659                 return false;
 660         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 661                 return false;
 662         if (info->helper) {
 663                 struct nf_conn_help *help;
 664
 665                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 666                 if (help && rcu_access_pointer(help->helper) != info->helper)
 667                         return false;
 668         }
 669         /* Force conntrack entry direction to the current packet? */
 670         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 671                 /* Delete the conntrack entry if confirmed, else just release
 672                  * the reference.
 673                  */
 674                 if (nf_ct_is_confirmed(ct))
 675                         nf_ct_delete(ct, 0, 0);
 676
 677                 nf_conntrack_put(&ct->ct_general);
 678                 nf_ct_set(skb, NULL, 0);
 679                 return false;
 680         }
 681
 682         return true;
 683 }
 684
 685 #ifdef CONFIG_NF_NAT_NEEDED
 686 /* Modelled after nf_nat_ipv[46]_fn().
 687  * range is only used for new, uninitialized NAT state.
 688  * Returns either NF_ACCEPT or NF_DROP.
 689  */
 690 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 691                               enum ip_conntrack_info ctinfo,
 692                               const struct nf_nat_range *range,
 693                               enum nf_nat_manip_type maniptype)
 694 {
 695         int hooknum, nh_off, err = NF_ACCEPT;
 696
 697         nh_off = skb_network_offset(skb);
 698         skb_pull_rcsum(skb, nh_off);
 699
 700         /* See HOOK2MANIP(). */
 701         if (maniptype == NF_NAT_MANIP_SRC)
 702                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 703         else
 704                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 705
 706         switch (ctinfo) {
 707         case IP_CT_RELATED:
 708         case IP_CT_RELATED_REPLY:
 709                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 710                     skb->protocol == htons(ETH_P_IP) &&
 711                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 712                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 713                                                            hooknum))
 714                                 err = NF_DROP;
 715                         goto push;
 716                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 717                            skb->protocol == htons(ETH_P_IPV6)) {
 718                         __be16 frag_off;
 719                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 720                         int hdrlen = ipv6_skip_exthdr(skb,
 721                                                       sizeof(struct ipv6hdr),
 722                                                       &nexthdr, &frag_off);
 723
 724                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 725                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 726                                                                      ctinfo,
 727                                                                      hooknum,
 728                                                                      hdrlen))
 729                                         err = NF_DROP;
 730                                 goto push;
 731                         }
 732                 }
 733                 /* Non-ICMP, fall thru to initialize if needed. */
 734         case IP_CT_NEW:
 735                 /* Seen it before?  This can happen for loopback, retrans,
 736                  * or local packets.
 737                  */
 738                 if (!nf_nat_initialized(ct, maniptype)) {
 739                         /* Initialize according to the NAT action. */
 740                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 741                                 /* Action is set up to establish a new
 742                                  * mapping.
 743                                  */
 744                                 ? nf_nat_setup_info(ct, range, maniptype)
 745                                 : nf_nat_alloc_null_binding(ct, hooknum);
 746                         if (err != NF_ACCEPT)
 747                                 goto push;
 748                 }
 749                 break;
 750
 751         case IP_CT_ESTABLISHED:
 752         case IP_CT_ESTABLISHED_REPLY:
 753                 break;
 754
 755         default:
 756                 err = NF_DROP;
 757                 goto push;
 758         }
 759
 760         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 761 push:
 762         skb_push(skb, nh_off);
 763         skb_postpush_rcsum(skb, skb->data, nh_off);
 764
 765         return err;
 766 }
 767
 768 static void ovs_nat_update_key(struct sw_flow_key *key,
 769                                const struct sk_buff *skb,
 770                                enum nf_nat_manip_type maniptype)
 771 {
 772         if (maniptype == NF_NAT_MANIP_SRC) {
 773                 __be16 src;
 774
 775                 key->ct_state |= OVS_CS_F_SRC_NAT;
 776                 if (key->eth.type == htons(ETH_P_IP))
 777                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 778                 else if (key->eth.type == htons(ETH_P_IPV6))
 779                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 780                                sizeof(key->ipv6.addr.src));
 781                 else
 782                         return;
 783
 784                 if (key->ip.proto == IPPROTO_UDP)
 785                         src = udp_hdr(skb)->source;
 786                 else if (key->ip.proto == IPPROTO_TCP)
 787                         src = tcp_hdr(skb)->source;
 788                 else if (key->ip.proto == IPPROTO_SCTP)
 789                         src = sctp_hdr(skb)->source;
 790                 else
 791                         return;
 792
 793                 key->tp.src = src;
 794         } else {
 795                 __be16 dst;
 796
 797                 key->ct_state |= OVS_CS_F_DST_NAT;
 798                 if (key->eth.type == htons(ETH_P_IP))
 799                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 800                 else if (key->eth.type == htons(ETH_P_IPV6))
 801                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 802                                sizeof(key->ipv6.addr.dst));
 803                 else
 804                         return;
 805
 806                 if (key->ip.proto == IPPROTO_UDP)
 807                         dst = udp_hdr(skb)->dest;
 808                 else if (key->ip.proto == IPPROTO_TCP)
 809                         dst = tcp_hdr(skb)->dest;
 810                 else if (key->ip.proto == IPPROTO_SCTP)
 811                         dst = sctp_hdr(skb)->dest;
 812                 else
 813                         return;
 814
 815                 key->tp.dst = dst;
 816         }
 817 }
 818
 819 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 820 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 821                       const struct ovs_conntrack_info *info,
 822                       struct sk_buff *skb, struct nf_conn *ct,
 823                       enum ip_conntrack_info ctinfo)
 824 {
 825         enum nf_nat_manip_type maniptype;
 826         int err;
 827
 828         /* Add NAT extension if not confirmed yet. */
 829         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 830                 return NF_ACCEPT;   /* Can't NAT. */
 831
 832         /* Determine NAT type.
 833          * Check if the NAT type can be deduced from the tracked connection.
 834          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 835          * when committing.
 836          */
 837         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 838             ct->status & IPS_NAT_MASK &&
 839             (ctinfo != IP_CT_RELATED || info->commit)) {
 840                 /* NAT an established or related connection like before. */
 841                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 842                         /* This is the REPLY direction for a connection
 843                          * for which NAT was applied in the forward
 844                          * direction.  Do the reverse NAT.
 845                          */
 846                         maniptype = ct->status & IPS_SRC_NAT
 847                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 848                 else
 849                         maniptype = ct->status & IPS_SRC_NAT
 850                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 851         } else if (info->nat & OVS_CT_SRC_NAT) {
 852                 maniptype = NF_NAT_MANIP_SRC;
 853         } else if (info->nat & OVS_CT_DST_NAT) {
 854                 maniptype = NF_NAT_MANIP_DST;
 855         } else {
 856                 return NF_ACCEPT; /* Connection is not NATed. */
 857         }
 858         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 859
 860         /* Mark NAT done if successful and update the flow key. */
 861         if (err == NF_ACCEPT)
 862                 ovs_nat_update_key(key, skb, maniptype);
 863
 864         return err;
 865 }
 866 #else /* !CONFIG_NF_NAT_NEEDED */
 867 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 868                       const struct ovs_conntrack_info *info,
 869                       struct sk_buff *skb, struct nf_conn *ct,
 870                       enum ip_conntrack_info ctinfo)
 871 {
 872         return NF_ACCEPT;
 873 }
 874 #endif
 875
 876 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 877  * not done already.  Update key with new CT state after passing the packet
 878  * through conntrack.
 879  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 880  * set to NULL and 0 will be returned.
 881  */
 882 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 883                            const struct ovs_conntrack_info *info,
 884                            struct sk_buff *skb)
 885 {
 886         /* If we are recirculating packets to match on conntrack fields and
 887          * committing with a separate conntrack action,  then we don't need to
 888          * actually run the packet through conntrack twice unless it's for a
 889          * different zone.
 890          */
 891         bool cached = skb_nfct_cached(net, key, info, skb);
 892         enum ip_conntrack_info ctinfo;
 893         struct nf_conn *ct;
 894
 895         if (!cached) {
 896                 struct nf_conn *tmpl = info->ct;
 897                 int err;
 898
 899                 /* Associate skb with specified zone. */
 900                 if (tmpl) {
 901                         if (skb_nfct(skb))
 902                                 nf_conntrack_put(skb_nfct(skb));
 903                         nf_conntrack_get(&tmpl->ct_general);
 904                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 905                 }
 906
 907                 err = nf_conntrack_in(net, info->family,
 908                                       NF_INET_PRE_ROUTING, skb);
 909                 if (err != NF_ACCEPT)
 910                         return -ENOENT;
 911
 912                 /* Clear CT state NAT flags to mark that we have not yet done
 913                  * NAT after the nf_conntrack_in() call.  We can actually clear
 914                  * the whole state, as it will be re-initialized below.
 915                  */
 916                 key->ct_state = 0;
 917
 918                 /* Update the key, but keep the NAT flags. */
 919                 ovs_ct_update_key(skb, info, key, true, true);
 920         }
 921
 922         ct = nf_ct_get(skb, &ctinfo);
 923         if (ct) {
 924                 /* Packets starting a new connection must be NATted before the
 925                  * helper, so that the helper knows about the NAT.  We enforce
 926                  * this by delaying both NAT and helper calls for unconfirmed
 927                  * connections until the committing CT action.  For later
 928                  * packets NAT and Helper may be called in either order.
 929                  *
 930                  * NAT will be done only if the CT action has NAT, and only
 931                  * once per packet (per zone), as guarded by the NAT bits in
 932                  * the key->ct_state.
 933                  */
 934                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 935                     (nf_ct_is_confirmed(ct) || info->commit) &&
 936                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 937                         return -EINVAL;
 938                 }
 939
 940                 /* Userspace may decide to perform a ct lookup without a helper
 941                  * specified followed by a (recirculate and) commit with one.
 942                  * Therefore, for unconfirmed connections which we will commit,
 943                  * we need to attach the helper here.
 944                  */
 945                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 946                     info->helper && !nfct_help(ct)) {
 947                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 948                                                             GFP_ATOMIC);
 949                         if (err)
 950                                 return err;
 951                 }
 952
 953                 /* Call the helper only if:
 954                  * - nf_conntrack_in() was executed above ("!cached") for a
 955                  *   confirmed connection, or
 956                  * - When committing an unconfirmed connection.
 957                  */
 958                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 959                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 960                         return -EINVAL;
 961                 }
 962         }
 963
 964         return 0;
 965 }
 966
 967 /* Lookup connection and read fields into key. */
 968 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 969                          const struct ovs_conntrack_info *info,
 970                          struct sk_buff *skb)
 971 {
 972         struct nf_conntrack_expect *exp;
 973
 974         /* If we pass an expected packet through nf_conntrack_in() the
 975          * expectation is typically removed, but the packet could still be
 976          * lost in upcall processing.  To prevent this from happening we
 977          * perform an explicit expectation lookup.  Expected connections are
 978          * always new, and will be passed through conntrack only when they are
 979          * committed, as it is OK to remove the expectation at that time.
 980          */
 981         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 982         if (exp) {
 983                 u8 state;
 984
 985                 /* NOTE: New connections are NATted and Helped only when
 986                  * committed, so we are not calling into NAT here.
 987                  */
 988                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 989                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 990         } else {
 991                 struct nf_conn *ct;
 992                 int err;
 993
 994                 err = __ovs_ct_lookup(net, key, info, skb);
 995                 if (err)
 996                         return err;
 997
 998                 ct = (struct nf_conn *)skb_nfct(skb);
 999                 if (ct)
1000                         nf_ct_deliver_cached_events(ct);
1001         }
1002
1003         return 0;
1004 }
1005
1006 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1007 {
1008         size_t i;
1009
1010         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1011                 if (labels->ct_labels_32[i])
1012                         return true;
1013
1014         return false;
1015 }
1016
1017 /* Lookup connection and confirm if unconfirmed. */
1018 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1019                          const struct ovs_conntrack_info *info,
1020                          struct sk_buff *skb)
1021 {
1022         enum ip_conntrack_info ctinfo;
1023         struct nf_conn *ct;
1024         int err;
1025
1026         err = __ovs_ct_lookup(net, key, info, skb);
1027         if (err)
1028                 return err;
1029
1030         /* The connection could be invalid, in which case this is a no-op.*/
1031         ct = nf_ct_get(skb, &ctinfo);
1032         if (!ct)
1033                 return 0;
1034
1035         /* Set the conntrack event mask if given.  NEW and DELETE events have
1036          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1037          * typically would receive many kinds of updates.  Setting the event
1038          * mask allows those events to be filtered.  The set event mask will
1039          * remain in effect for the lifetime of the connection unless changed
1040          * by a further CT action with both the commit flag and the eventmask
1041          * option. */
1042         if (info->have_eventmask) {
1043                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1044
1045                 if (cache)
1046                         cache->ctmask = info->eventmask;
1047         }
1048
1049         /* Apply changes before confirming the connection so that the initial
1050          * conntrack NEW netlink event carries the values given in the CT
1051          * action.
1052          */
1053         if (info->mark.mask) {
1054                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1055                                       info->mark.mask);
1056                 if (err)
1057                         return err;
1058         }
1059         if (!nf_ct_is_confirmed(ct)) {
1060                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1061                                          &info->labels.mask);
1062                 if (err)
1063                         return err;
1064         } else if (labels_nonzero(&info->labels.mask)) {
1065                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1066                                         &info->labels.mask);
1067                 if (err)
1068                         return err;
1069         }
1070         /* This will take care of sending queued events even if the connection
1071          * is already confirmed.
1072          */
1073         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1074                 return -EINVAL;
1075
1076         return 0;
1077 }
1078
1079 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1080  * value if 'skb' is freed.
1081  */
1082 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1083                    struct sw_flow_key *key,
1084                    const struct ovs_conntrack_info *info)
1085 {
1086         int nh_ofs;
1087         int err;
1088
1089         /* The conntrack module expects to be working at L3. */
1090         nh_ofs = skb_network_offset(skb);
1091         skb_pull_rcsum(skb, nh_ofs);
1092
1093         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1094                 err = handle_fragments(net, key, info->zone.id, skb);
1095                 if (err)
1096                         return err;
1097         }
1098
1099         if (info->commit)
1100                 err = ovs_ct_commit(net, key, info, skb);
1101         else
1102                 err = ovs_ct_lookup(net, key, info, skb);
1103
1104         skb_push(skb, nh_ofs);
1105         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1106         if (err)
1107                 kfree_skb(skb);
1108         return err;
1109 }
1110
1111 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1112                              const struct sw_flow_key *key, bool log)
1113 {
1114         struct nf_conntrack_helper *helper;
1115         struct nf_conn_help *help;
1116
1117         helper = nf_conntrack_helper_try_module_get(name, info->family,
1118                                                     key->ip.proto);
1119         if (!helper) {
1120                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1121                 return -EINVAL;
1122         }
1123
1124         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1125         if (!help) {
1126                 nf_conntrack_helper_put(helper);
1127                 return -ENOMEM;
1128         }
1129
1130         rcu_assign_pointer(help->helper, helper);
1131         info->helper = helper;
1132         return 0;
1133 }
1134
1135 #ifdef CONFIG_NF_NAT_NEEDED
1136 static int parse_nat(const struct nlattr *attr,
1137                      struct ovs_conntrack_info *info, bool log)
1138 {
1139         struct nlattr *a;
1140         int rem;
1141         bool have_ip_max = false;
1142         bool have_proto_max = false;
1143         bool ip_vers = (info->family == NFPROTO_IPV6);
1144
1145         nla_for_each_nested(a, attr, rem) {
1146                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1147                         [OVS_NAT_ATTR_SRC] = {0, 0},
1148                         [OVS_NAT_ATTR_DST] = {0, 0},
1149                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1150                                                  sizeof(struct in6_addr)},
1151                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1152                                                  sizeof(struct in6_addr)},
1153                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1154                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1155                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1156                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1157                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1158                 };
1159                 int type = nla_type(a);
1160
1161                 if (type > OVS_NAT_ATTR_MAX) {
1162                         OVS_NLERR(log,
1163                                   "Unknown NAT attribute (type=%d, max=%d).\n",
1164                                   type, OVS_NAT_ATTR_MAX);
1165                         return -EINVAL;
1166                 }
1167
1168                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1169                         OVS_NLERR(log,
1170                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
1171                                   type, nla_len(a),
1172                                   ovs_nat_attr_lens[type][ip_vers]);
1173                         return -EINVAL;
1174                 }
1175
1176                 switch (type) {
1177                 case OVS_NAT_ATTR_SRC:
1178                 case OVS_NAT_ATTR_DST:
1179                         if (info->nat) {
1180                                 OVS_NLERR(log,
1181                                           "Only one type of NAT may be specified.\n"
1182                                           );
1183                                 return -ERANGE;
1184                         }
1185                         info->nat |= OVS_CT_NAT;
1186                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1187                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1188                         break;
1189
1190                 case OVS_NAT_ATTR_IP_MIN:
1191                         nla_memcpy(&info->range.min_addr, a,
1192                                    sizeof(info->range.min_addr));
1193                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1194                         break;
1195
1196                 case OVS_NAT_ATTR_IP_MAX:
1197                         have_ip_max = true;
1198                         nla_memcpy(&info->range.max_addr, a,
1199                                    sizeof(info->range.max_addr));
1200                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1201                         break;
1202
1203                 case OVS_NAT_ATTR_PROTO_MIN:
1204                         info->range.min_proto.all = htons(nla_get_u16(a));
1205                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1206                         break;
1207
1208                 case OVS_NAT_ATTR_PROTO_MAX:
1209                         have_proto_max = true;
1210                         info->range.max_proto.all = htons(nla_get_u16(a));
1211                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1212                         break;
1213
1214                 case OVS_NAT_ATTR_PERSISTENT:
1215                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1216                         break;
1217
1218                 case OVS_NAT_ATTR_PROTO_HASH:
1219                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1220                         break;
1221
1222                 case OVS_NAT_ATTR_PROTO_RANDOM:
1223                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1224                         break;
1225
1226                 default:
1227                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1228                         return -EINVAL;
1229                 }
1230         }
1231
1232         if (rem > 0) {
1233                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1234                 return -EINVAL;
1235         }
1236         if (!info->nat) {
1237                 /* Do not allow flags if no type is given. */
1238                 if (info->range.flags) {
1239                         OVS_NLERR(log,
1240                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1241                                   );
1242                         return -EINVAL;
1243                 }
1244                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1245         } else if (!info->commit) {
1246                 OVS_NLERR(log,
1247                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1248                           );
1249                 return -EINVAL;
1250         }
1251         /* Allow missing IP_MAX. */
1252         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1253                 memcpy(&info->range.max_addr, &info->range.min_addr,
1254                        sizeof(info->range.max_addr));
1255         }
1256         /* Allow missing PROTO_MAX. */
1257         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1258             !have_proto_max) {
1259                 info->range.max_proto.all = info->range.min_proto.all;
1260         }
1261         return 0;
1262 }
1263 #endif
1264
1265 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1266         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1267         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1268         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1269                                     .maxlen = sizeof(u16) },
1270         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1271                                     .maxlen = sizeof(struct md_mark) },
1272         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1273                                     .maxlen = sizeof(struct md_labels) },
1274         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1275                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1276 #ifdef CONFIG_NF_NAT_NEEDED
1277         /* NAT length is checked when parsing the nested attributes. */
1278         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1279 #endif
1280         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1281                                     .maxlen = sizeof(u32) },
1282 };
1283
1284 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1285                     const char **helper, bool log)
1286 {
1287         struct nlattr *a;
1288         int rem;
1289
1290         nla_for_each_nested(a, attr, rem) {
1291                 int type = nla_type(a);
1292                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1293                 int minlen = ovs_ct_attr_lens[type].minlen;
1294
1295                 if (type > OVS_CT_ATTR_MAX) {
1296                         OVS_NLERR(log,
1297                                   "Unknown conntrack attr (type=%d, max=%d)",
1298                                   type, OVS_CT_ATTR_MAX);
1299                         return -EINVAL;
1300                 }
1301                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1302                         OVS_NLERR(log,
1303                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1304                                   type, nla_len(a), maxlen);
1305                         return -EINVAL;
1306                 }
1307
1308                 switch (type) {
1309                 case OVS_CT_ATTR_FORCE_COMMIT:
1310                         info->force = true;
1311                         /* fall through. */
1312                 case OVS_CT_ATTR_COMMIT:
1313                         info->commit = true;
1314                         break;
1315 #ifdef CONFIG_NF_CONNTRACK_ZONES
1316                 case OVS_CT_ATTR_ZONE:
1317                         info->zone.id = nla_get_u16(a);
1318                         break;
1319 #endif
1320 #ifdef CONFIG_NF_CONNTRACK_MARK
1321                 case OVS_CT_ATTR_MARK: {
1322                         struct md_mark *mark = nla_data(a);
1323
1324                         if (!mark->mask) {
1325                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1326                                 return -EINVAL;
1327                         }
1328                         info->mark = *mark;
1329                         break;
1330                 }
1331 #endif
1332 #ifdef CONFIG_NF_CONNTRACK_LABELS
1333                 case OVS_CT_ATTR_LABELS: {
1334                         struct md_labels *labels = nla_data(a);
1335
1336                         if (!labels_nonzero(&labels->mask)) {
1337                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1338                                 return -EINVAL;
1339                         }
1340                         info->labels = *labels;
1341                         break;
1342                 }
1343 #endif
1344                 case OVS_CT_ATTR_HELPER:
1345                         *helper = nla_data(a);
1346                         if (!memchr(*helper, '\0', nla_len(a))) {
1347                                 OVS_NLERR(log, "Invalid conntrack helper");
1348                                 return -EINVAL;
1349                         }
1350                         break;
1351 #ifdef CONFIG_NF_NAT_NEEDED
1352                 case OVS_CT_ATTR_NAT: {
1353                         int err = parse_nat(a, info, log);
1354
1355                         if (err)
1356                                 return err;
1357                         break;
1358                 }
1359 #endif
1360                 case OVS_CT_ATTR_EVENTMASK:
1361                         info->have_eventmask = true;
1362                         info->eventmask = nla_get_u32(a);
1363                         break;
1364
1365                 default:
1366                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1367                                   type);
1368                         return -EINVAL;
1369                 }
1370         }
1371
1372 #ifdef CONFIG_NF_CONNTRACK_MARK
1373         if (!info->commit && info->mark.mask) {
1374                 OVS_NLERR(log,
1375                           "Setting conntrack mark requires 'commit' flag.");
1376                 return -EINVAL;
1377         }
1378 #endif
1379 #ifdef CONFIG_NF_CONNTRACK_LABELS
1380         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1381                 OVS_NLERR(log,
1382                           "Setting conntrack labels requires 'commit' flag.");
1383                 return -EINVAL;
1384         }
1385 #endif
1386         if (rem > 0) {
1387                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1388                 return -EINVAL;
1389         }
1390
1391         return 0;
1392 }
1393
1394 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1395 {
1396         if (attr == OVS_KEY_ATTR_CT_STATE)
1397                 return true;
1398         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1399             attr == OVS_KEY_ATTR_CT_ZONE)
1400                 return true;
1401         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1402             attr == OVS_KEY_ATTR_CT_MARK)
1403                 return true;
1404         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1405             attr == OVS_KEY_ATTR_CT_LABELS) {
1406                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1407
1408                 return ovs_net->xt_label;
1409         }
1410
1411         return false;
1412 }
1413
1414 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1415                        const struct sw_flow_key *key,
1416                        struct sw_flow_actions **sfa,  bool log)
1417 {
1418         struct ovs_conntrack_info ct_info;
1419         const char *helper = NULL;
1420         u16 family;
1421         int err;
1422
1423         family = key_to_nfproto(key);
1424         if (family == NFPROTO_UNSPEC) {
1425                 OVS_NLERR(log, "ct family unspecified");
1426                 return -EINVAL;
1427         }
1428
1429         memset(&ct_info, 0, sizeof(ct_info));
1430         ct_info.family = family;
1431
1432         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1433                         NF_CT_DEFAULT_ZONE_DIR, 0);
1434
1435         err = parse_ct(attr, &ct_info, &helper, log);
1436         if (err)
1437                 return err;
1438
1439         /* Set up template for tracking connections in specific zones. */
1440         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1441         if (!ct_info.ct) {
1442                 OVS_NLERR(log, "Failed to allocate conntrack template");
1443                 return -ENOMEM;
1444         }
1445
1446         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1447         nf_conntrack_get(&ct_info.ct->ct_general);
1448
1449         if (helper) {
1450                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1451                 if (err)
1452                         goto err_free_ct;
1453         }
1454
1455         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1456                                  sizeof(ct_info), log);
1457         if (err)
1458                 goto err_free_ct;
1459
1460         return 0;
1461 err_free_ct:
1462         __ovs_ct_free_action(&ct_info);
1463         return err;
1464 }
1465
1466 #ifdef CONFIG_NF_NAT_NEEDED
1467 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1468                                struct sk_buff *skb)
1469 {
1470         struct nlattr *start;
1471
1472         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1473         if (!start)
1474                 return false;
1475
1476         if (info->nat & OVS_CT_SRC_NAT) {
1477                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1478                         return false;
1479         } else if (info->nat & OVS_CT_DST_NAT) {
1480                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1481                         return false;
1482         } else {
1483                 goto out;
1484         }
1485
1486         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1487                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1488                     info->family == NFPROTO_IPV4) {
1489                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1490                                             info->range.min_addr.ip) ||
1491                             (info->range.max_addr.ip
1492                              != info->range.min_addr.ip &&
1493                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1494                                               info->range.max_addr.ip))))
1495                                 return false;
1496                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1497                            info->family == NFPROTO_IPV6) {
1498                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1499                                              &info->range.min_addr.in6) ||
1500                             (memcmp(&info->range.max_addr.in6,
1501                                     &info->range.min_addr.in6,
1502                                     sizeof(info->range.max_addr.in6)) &&
1503                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1504                                                &info->range.max_addr.in6))))
1505                                 return false;
1506                 } else {
1507                         return false;
1508                 }
1509         }
1510         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1511             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1512                          ntohs(info->range.min_proto.all)) ||
1513              (info->range.max_proto.all != info->range.min_proto.all &&
1514               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1515                           ntohs(info->range.max_proto.all)))))
1516                 return false;
1517
1518         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1519             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1520                 return false;
1521         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1522             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1523                 return false;
1524         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1525             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1526                 return false;
1527 out:
1528         nla_nest_end(skb, start);
1529
1530         return true;
1531 }
1532 #endif
1533
1534 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1535                           struct sk_buff *skb)
1536 {
1537         struct nlattr *start;
1538
1539         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1540         if (!start)
1541                 return -EMSGSIZE;
1542
1543         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1544                                             ? OVS_CT_ATTR_FORCE_COMMIT
1545                                             : OVS_CT_ATTR_COMMIT))
1546                 return -EMSGSIZE;
1547         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1548             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1549                 return -EMSGSIZE;
1550         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1551             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1552                     &ct_info->mark))
1553                 return -EMSGSIZE;
1554         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1555             labels_nonzero(&ct_info->labels.mask) &&
1556             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1557                     &ct_info->labels))
1558                 return -EMSGSIZE;
1559         if (ct_info->helper) {
1560                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1561                                    ct_info->helper->name))
1562                         return -EMSGSIZE;
1563         }
1564         if (ct_info->have_eventmask &&
1565             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1566                 return -EMSGSIZE;
1567
1568 #ifdef CONFIG_NF_NAT_NEEDED
1569         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1570                 return -EMSGSIZE;
1571 #endif
1572         nla_nest_end(skb, start);
1573
1574         return 0;
1575 }
1576
1577 void ovs_ct_free_action(const struct nlattr *a)
1578 {
1579         struct ovs_conntrack_info *ct_info = nla_data(a);
1580
1581         __ovs_ct_free_action(ct_info);
1582 }
1583
1584 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1585 {
1586         if (ct_info->helper)
1587                 nf_conntrack_helper_put(ct_info->helper);
1588         if (ct_info->ct)
1589                 nf_ct_tmpl_free(ct_info->ct);
1590 }
1591
1592 void ovs_ct_init(struct net *net)
1593 {
1594         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1595         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1596
1597         if (nf_connlabels_get(net, n_bits - 1)) {
1598                 ovs_net->xt_label = false;
1599                 OVS_NLERR(true, "Failed to set connlabel length");
1600         } else {
1601                 ovs_net->xt_label = true;
1602         }
1603 }
1604
1605 void ovs_ct_exit(struct net *net)
1606 {
1607         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1608
1609         if (ovs_net->xt_label)
1610                 nf_connlabels_put(net);
1611 }