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