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         const struct nf_conntrack_l3proto *l3proto;
 583         const struct nf_conntrack_l4proto *l4proto;
 584         struct nf_conntrack_tuple tuple;
 585         struct nf_conntrack_tuple_hash *h;
 586         struct nf_conn *ct;
 587         unsigned int dataoff;
 588         u8 protonum;
 589
 590         l3proto = __nf_ct_l3proto_find(l3num);
 591         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 592                                  &protonum) <= 0) {
 593                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 594                 return NULL;
 595         }
 596         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 597         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 598                              protonum, net, &tuple, l3proto, l4proto)) {
 599                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 600                 return NULL;
 601         }
 602
 603         /* Must invert the tuple if skb has been transformed by NAT. */
 604         if (natted) {
 605                 struct nf_conntrack_tuple inverse;
 606
 607                 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
 608                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 609                         return NULL;
 610                 }
 611                 tuple = inverse;
 612         }
 613
 614         /* look for tuple match */
 615         h = nf_conntrack_find_get(net, zone, &tuple);
 616         if (!h)
 617                 return NULL;   /* Not found. */
 618
 619         ct = nf_ct_tuplehash_to_ctrack(h);
 620
 621         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 622          * select the other tuplehash to get the right 'ctinfo' bits for this
 623          * packet.
 624          */
 625         if (natted)
 626                 h = &ct->tuplehash[!h->tuple.dst.dir];
 627
 628         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 629         return ct;
 630 }
 631
 632 static
 633 struct nf_conn *ovs_ct_executed(struct net *net,
 634                                 const struct sw_flow_key *key,
 635                                 const struct ovs_conntrack_info *info,
 636                                 struct sk_buff *skb,
 637                                 bool *ct_executed)
 638 {
 639         struct nf_conn *ct = NULL;
 640
 641         /* If no ct, check if we have evidence that an existing conntrack entry
 642          * might be found for this skb.  This happens when we lose a skb->_nfct
 643          * due to an upcall, or if the direction is being forced.  If the
 644          * connection was not confirmed, it is not cached and needs to be run
 645          * through conntrack again.
 646          */
 647         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 648                        !(key->ct_state & OVS_CS_F_INVALID) &&
 649                        (key->ct_zone == info->zone.id);
 650
 651         if (*ct_executed || (!key->ct_state && info->force)) {
 652                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 653                                           !!(key->ct_state &
 654                                           OVS_CS_F_NAT_MASK));
 655         }
 656
 657         return ct;
 658 }
 659
 660 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 661 static bool skb_nfct_cached(struct net *net,
 662                             const struct sw_flow_key *key,
 663                             const struct ovs_conntrack_info *info,
 664                             struct sk_buff *skb)
 665 {
 666         enum ip_conntrack_info ctinfo;
 667         struct nf_conn *ct;
 668         bool ct_executed = true;
 669
 670         ct = nf_ct_get(skb, &ctinfo);
 671         if (!ct)
 672                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 673
 674         if (ct)
 675                 nf_ct_get(skb, &ctinfo);
 676         else
 677                 return false;
 678
 679         if (!net_eq(net, read_pnet(&ct->ct_net)))
 680                 return false;
 681         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 682                 return false;
 683         if (info->helper) {
 684                 struct nf_conn_help *help;
 685
 686                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 687                 if (help && rcu_access_pointer(help->helper) != info->helper)
 688                         return false;
 689         }
 690         /* Force conntrack entry direction to the current packet? */
 691         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 692                 /* Delete the conntrack entry if confirmed, else just release
 693                  * the reference.
 694                  */
 695                 if (nf_ct_is_confirmed(ct))
 696                         nf_ct_delete(ct, 0, 0);
 697
 698                 nf_conntrack_put(&ct->ct_general);
 699                 nf_ct_set(skb, NULL, 0);
 700                 return false;
 701         }
 702
 703         return ct_executed;
 704 }
 705
 706 #ifdef CONFIG_NF_NAT_NEEDED
 707 /* Modelled after nf_nat_ipv[46]_fn().
 708  * range is only used for new, uninitialized NAT state.
 709  * Returns either NF_ACCEPT or NF_DROP.
 710  */
 711 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 712                               enum ip_conntrack_info ctinfo,
 713                               const struct nf_nat_range *range,
 714                               enum nf_nat_manip_type maniptype)
 715 {
 716         int hooknum, nh_off, err = NF_ACCEPT;
 717
 718         nh_off = skb_network_offset(skb);
 719         skb_pull_rcsum(skb, nh_off);
 720
 721         /* See HOOK2MANIP(). */
 722         if (maniptype == NF_NAT_MANIP_SRC)
 723                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 724         else
 725                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 726
 727         switch (ctinfo) {
 728         case IP_CT_RELATED:
 729         case IP_CT_RELATED_REPLY:
 730                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 731                     skb->protocol == htons(ETH_P_IP) &&
 732                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 733                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 734                                                            hooknum))
 735                                 err = NF_DROP;
 736                         goto push;
 737                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 738                            skb->protocol == htons(ETH_P_IPV6)) {
 739                         __be16 frag_off;
 740                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 741                         int hdrlen = ipv6_skip_exthdr(skb,
 742                                                       sizeof(struct ipv6hdr),
 743                                                       &nexthdr, &frag_off);
 744
 745                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 746                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 747                                                                      ctinfo,
 748                                                                      hooknum,
 749                                                                      hdrlen))
 750                                         err = NF_DROP;
 751                                 goto push;
 752                         }
 753                 }
 754                 /* Non-ICMP, fall thru to initialize if needed. */
 755                 /* fall through */
 756         case IP_CT_NEW:
 757                 /* Seen it before?  This can happen for loopback, retrans,
 758                  * or local packets.
 759                  */
 760                 if (!nf_nat_initialized(ct, maniptype)) {
 761                         /* Initialize according to the NAT action. */
 762                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 763                                 /* Action is set up to establish a new
 764                                  * mapping.
 765                                  */
 766                                 ? nf_nat_setup_info(ct, range, maniptype)
 767                                 : nf_nat_alloc_null_binding(ct, hooknum);
 768                         if (err != NF_ACCEPT)
 769                                 goto push;
 770                 }
 771                 break;
 772
 773         case IP_CT_ESTABLISHED:
 774         case IP_CT_ESTABLISHED_REPLY:
 775                 break;
 776
 777         default:
 778                 err = NF_DROP;
 779                 goto push;
 780         }
 781
 782         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 783 push:
 784         skb_push(skb, nh_off);
 785         skb_postpush_rcsum(skb, skb->data, nh_off);
 786
 787         return err;
 788 }
 789
 790 static void ovs_nat_update_key(struct sw_flow_key *key,
 791                                const struct sk_buff *skb,
 792                                enum nf_nat_manip_type maniptype)
 793 {
 794         if (maniptype == NF_NAT_MANIP_SRC) {
 795                 __be16 src;
 796
 797                 key->ct_state |= OVS_CS_F_SRC_NAT;
 798                 if (key->eth.type == htons(ETH_P_IP))
 799                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 800                 else if (key->eth.type == htons(ETH_P_IPV6))
 801                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 802                                sizeof(key->ipv6.addr.src));
 803                 else
 804                         return;
 805
 806                 if (key->ip.proto == IPPROTO_UDP)
 807                         src = udp_hdr(skb)->source;
 808                 else if (key->ip.proto == IPPROTO_TCP)
 809                         src = tcp_hdr(skb)->source;
 810                 else if (key->ip.proto == IPPROTO_SCTP)
 811                         src = sctp_hdr(skb)->source;
 812                 else
 813                         return;
 814
 815                 key->tp.src = src;
 816         } else {
 817                 __be16 dst;
 818
 819                 key->ct_state |= OVS_CS_F_DST_NAT;
 820                 if (key->eth.type == htons(ETH_P_IP))
 821                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 822                 else if (key->eth.type == htons(ETH_P_IPV6))
 823                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 824                                sizeof(key->ipv6.addr.dst));
 825                 else
 826                         return;
 827
 828                 if (key->ip.proto == IPPROTO_UDP)
 829                         dst = udp_hdr(skb)->dest;
 830                 else if (key->ip.proto == IPPROTO_TCP)
 831                         dst = tcp_hdr(skb)->dest;
 832                 else if (key->ip.proto == IPPROTO_SCTP)
 833                         dst = sctp_hdr(skb)->dest;
 834                 else
 835                         return;
 836
 837                 key->tp.dst = dst;
 838         }
 839 }
 840
 841 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 842 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 843                       const struct ovs_conntrack_info *info,
 844                       struct sk_buff *skb, struct nf_conn *ct,
 845                       enum ip_conntrack_info ctinfo)
 846 {
 847         enum nf_nat_manip_type maniptype;
 848         int err;
 849
 850         /* Add NAT extension if not confirmed yet. */
 851         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 852                 return NF_ACCEPT;   /* Can't NAT. */
 853
 854         /* Determine NAT type.
 855          * Check if the NAT type can be deduced from the tracked connection.
 856          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 857          * when committing.
 858          */
 859         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 860             ct->status & IPS_NAT_MASK &&
 861             (ctinfo != IP_CT_RELATED || info->commit)) {
 862                 /* NAT an established or related connection like before. */
 863                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 864                         /* This is the REPLY direction for a connection
 865                          * for which NAT was applied in the forward
 866                          * direction.  Do the reverse NAT.
 867                          */
 868                         maniptype = ct->status & IPS_SRC_NAT
 869                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 870                 else
 871                         maniptype = ct->status & IPS_SRC_NAT
 872                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 873         } else if (info->nat & OVS_CT_SRC_NAT) {
 874                 maniptype = NF_NAT_MANIP_SRC;
 875         } else if (info->nat & OVS_CT_DST_NAT) {
 876                 maniptype = NF_NAT_MANIP_DST;
 877         } else {
 878                 return NF_ACCEPT; /* Connection is not NATed. */
 879         }
 880         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 881
 882         /* Mark NAT done if successful and update the flow key. */
 883         if (err == NF_ACCEPT)
 884                 ovs_nat_update_key(key, skb, maniptype);
 885
 886         return err;
 887 }
 888 #else /* !CONFIG_NF_NAT_NEEDED */
 889 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 890                       const struct ovs_conntrack_info *info,
 891                       struct sk_buff *skb, struct nf_conn *ct,
 892                       enum ip_conntrack_info ctinfo)
 893 {
 894         return NF_ACCEPT;
 895 }
 896 #endif
 897
 898 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 899  * not done already.  Update key with new CT state after passing the packet
 900  * through conntrack.
 901  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 902  * set to NULL and 0 will be returned.
 903  */
 904 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 905                            const struct ovs_conntrack_info *info,
 906                            struct sk_buff *skb)
 907 {
 908         /* If we are recirculating packets to match on conntrack fields and
 909          * committing with a separate conntrack action,  then we don't need to
 910          * actually run the packet through conntrack twice unless it's for a
 911          * different zone.
 912          */
 913         bool cached = skb_nfct_cached(net, key, info, skb);
 914         enum ip_conntrack_info ctinfo;
 915         struct nf_conn *ct;
 916
 917         if (!cached) {
 918                 struct nf_conn *tmpl = info->ct;
 919                 int err;
 920
 921                 /* Associate skb with specified zone. */
 922                 if (tmpl) {
 923                         if (skb_nfct(skb))
 924                                 nf_conntrack_put(skb_nfct(skb));
 925                         nf_conntrack_get(&tmpl->ct_general);
 926                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 927                 }
 928
 929                 err = nf_conntrack_in(net, info->family,
 930                                       NF_INET_PRE_ROUTING, skb);
 931                 if (err != NF_ACCEPT)
 932                         return -ENOENT;
 933
 934                 /* Clear CT state NAT flags to mark that we have not yet done
 935                  * NAT after the nf_conntrack_in() call.  We can actually clear
 936                  * the whole state, as it will be re-initialized below.
 937                  */
 938                 key->ct_state = 0;
 939
 940                 /* Update the key, but keep the NAT flags. */
 941                 ovs_ct_update_key(skb, info, key, true, true);
 942         }
 943
 944         ct = nf_ct_get(skb, &ctinfo);
 945         if (ct) {
 946                 /* Packets starting a new connection must be NATted before the
 947                  * helper, so that the helper knows about the NAT.  We enforce
 948                  * this by delaying both NAT and helper calls for unconfirmed
 949                  * connections until the committing CT action.  For later
 950                  * packets NAT and Helper may be called in either order.
 951                  *
 952                  * NAT will be done only if the CT action has NAT, and only
 953                  * once per packet (per zone), as guarded by the NAT bits in
 954                  * the key->ct_state.
 955                  */
 956                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 957                     (nf_ct_is_confirmed(ct) || info->commit) &&
 958                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 959                         return -EINVAL;
 960                 }
 961
 962                 /* Userspace may decide to perform a ct lookup without a helper
 963                  * specified followed by a (recirculate and) commit with one.
 964                  * Therefore, for unconfirmed connections which we will commit,
 965                  * we need to attach the helper here.
 966                  */
 967                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 968                     info->helper && !nfct_help(ct)) {
 969                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 970                                                             GFP_ATOMIC);
 971                         if (err)
 972                                 return err;
 973                 }
 974
 975                 /* Call the helper only if:
 976                  * - nf_conntrack_in() was executed above ("!cached") for a
 977                  *   confirmed connection, or
 978                  * - When committing an unconfirmed connection.
 979                  */
 980                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 981                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 982                         return -EINVAL;
 983                 }
 984         }
 985
 986         return 0;
 987 }
 988
 989 /* Lookup connection and read fields into key. */
 990 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 991                          const struct ovs_conntrack_info *info,
 992                          struct sk_buff *skb)
 993 {
 994         struct nf_conntrack_expect *exp;
 995
 996         /* If we pass an expected packet through nf_conntrack_in() the
 997          * expectation is typically removed, but the packet could still be
 998          * lost in upcall processing.  To prevent this from happening we
 999          * perform an explicit expectation lookup.  Expected connections are
1000          * always new, and will be passed through conntrack only when they are
1001          * committed, as it is OK to remove the expectation at that time.
1002          */
1003         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1004         if (exp) {
1005                 u8 state;
1006
1007                 /* NOTE: New connections are NATted and Helped only when
1008                  * committed, so we are not calling into NAT here.
1009                  */
1010                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1011                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1012         } else {
1013                 struct nf_conn *ct;
1014                 int err;
1015
1016                 err = __ovs_ct_lookup(net, key, info, skb);
1017                 if (err)
1018                         return err;
1019
1020                 ct = (struct nf_conn *)skb_nfct(skb);
1021                 if (ct)
1022                         nf_ct_deliver_cached_events(ct);
1023         }
1024
1025         return 0;
1026 }
1027
1028 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1029 {
1030         size_t i;
1031
1032         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1033                 if (labels->ct_labels_32[i])
1034                         return true;
1035
1036         return false;
1037 }
1038
1039 /* Lookup connection and confirm if unconfirmed. */
1040 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1041                          const struct ovs_conntrack_info *info,
1042                          struct sk_buff *skb)
1043 {
1044         enum ip_conntrack_info ctinfo;
1045         struct nf_conn *ct;
1046         int err;
1047
1048         err = __ovs_ct_lookup(net, key, info, skb);
1049         if (err)
1050                 return err;
1051
1052         /* The connection could be invalid, in which case this is a no-op.*/
1053         ct = nf_ct_get(skb, &ctinfo);
1054         if (!ct)
1055                 return 0;
1056
1057         /* Set the conntrack event mask if given.  NEW and DELETE events have
1058          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1059          * typically would receive many kinds of updates.  Setting the event
1060          * mask allows those events to be filtered.  The set event mask will
1061          * remain in effect for the lifetime of the connection unless changed
1062          * by a further CT action with both the commit flag and the eventmask
1063          * option. */
1064         if (info->have_eventmask) {
1065                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1066
1067                 if (cache)
1068                         cache->ctmask = info->eventmask;
1069         }
1070
1071         /* Apply changes before confirming the connection so that the initial
1072          * conntrack NEW netlink event carries the values given in the CT
1073          * action.
1074          */
1075         if (info->mark.mask) {
1076                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1077                                       info->mark.mask);
1078                 if (err)
1079                         return err;
1080         }
1081         if (!nf_ct_is_confirmed(ct)) {
1082                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1083                                          &info->labels.mask);
1084                 if (err)
1085                         return err;
1086         } else if (labels_nonzero(&info->labels.mask)) {
1087                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1088                                         &info->labels.mask);
1089                 if (err)
1090                         return err;
1091         }
1092         /* This will take care of sending queued events even if the connection
1093          * is already confirmed.
1094          */
1095         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1096                 return -EINVAL;
1097
1098         return 0;
1099 }
1100
1101 /* Trim the skb to the length specified by the IP/IPv6 header,
1102  * removing any trailing lower-layer padding. This prepares the skb
1103  * for higher-layer processing that assumes skb->len excludes padding
1104  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1105  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1106  */
1107 static int ovs_skb_network_trim(struct sk_buff *skb)
1108 {
1109         unsigned int len;
1110         int err;
1111
1112         switch (skb->protocol) {
1113         case htons(ETH_P_IP):
1114                 len = ntohs(ip_hdr(skb)->tot_len);
1115                 break;
1116         case htons(ETH_P_IPV6):
1117                 len = sizeof(struct ipv6hdr)
1118                         + ntohs(ipv6_hdr(skb)->payload_len);
1119                 break;
1120         default:
1121                 len = skb->len;
1122         }
1123
1124         err = pskb_trim_rcsum(skb, len);
1125         if (err)
1126                 kfree_skb(skb);
1127
1128         return err;
1129 }
1130
1131 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1132  * value if 'skb' is freed.
1133  */
1134 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1135                    struct sw_flow_key *key,
1136                    const struct ovs_conntrack_info *info)
1137 {
1138         int nh_ofs;
1139         int err;
1140
1141         /* The conntrack module expects to be working at L3. */
1142         nh_ofs = skb_network_offset(skb);
1143         skb_pull_rcsum(skb, nh_ofs);
1144
1145         err = ovs_skb_network_trim(skb);
1146         if (err)
1147                 return err;
1148
1149         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1150                 err = handle_fragments(net, key, info->zone.id, skb);
1151                 if (err)
1152                         return err;
1153         }
1154
1155         if (info->commit)
1156                 err = ovs_ct_commit(net, key, info, skb);
1157         else
1158                 err = ovs_ct_lookup(net, key, info, skb);
1159
1160         skb_push(skb, nh_ofs);
1161         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1162         if (err)
1163                 kfree_skb(skb);
1164         return err;
1165 }
1166
1167 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1168 {
1169         if (skb_nfct(skb)) {
1170                 nf_conntrack_put(skb_nfct(skb));
1171                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1172                 ovs_ct_fill_key(skb, key);
1173         }
1174
1175         return 0;
1176 }
1177
1178 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1179                              const struct sw_flow_key *key, bool log)
1180 {
1181         struct nf_conntrack_helper *helper;
1182         struct nf_conn_help *help;
1183
1184         helper = nf_conntrack_helper_try_module_get(name, info->family,
1185                                                     key->ip.proto);
1186         if (!helper) {
1187                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1188                 return -EINVAL;
1189         }
1190
1191         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1192         if (!help) {
1193                 nf_conntrack_helper_put(helper);
1194                 return -ENOMEM;
1195         }
1196
1197         rcu_assign_pointer(help->helper, helper);
1198         info->helper = helper;
1199         return 0;
1200 }
1201
1202 #ifdef CONFIG_NF_NAT_NEEDED
1203 static int parse_nat(const struct nlattr *attr,
1204                      struct ovs_conntrack_info *info, bool log)
1205 {
1206         struct nlattr *a;
1207         int rem;
1208         bool have_ip_max = false;
1209         bool have_proto_max = false;
1210         bool ip_vers = (info->family == NFPROTO_IPV6);
1211
1212         nla_for_each_nested(a, attr, rem) {
1213                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1214                         [OVS_NAT_ATTR_SRC] = {0, 0},
1215                         [OVS_NAT_ATTR_DST] = {0, 0},
1216                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1217                                                  sizeof(struct in6_addr)},
1218                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1219                                                  sizeof(struct in6_addr)},
1220                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1221                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1222                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1223                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1224                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1225                 };
1226                 int type = nla_type(a);
1227
1228                 if (type > OVS_NAT_ATTR_MAX) {
1229                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1230                                   type, OVS_NAT_ATTR_MAX);
1231                         return -EINVAL;
1232                 }
1233
1234                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1235                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1236                                   type, nla_len(a),
1237                                   ovs_nat_attr_lens[type][ip_vers]);
1238                         return -EINVAL;
1239                 }
1240
1241                 switch (type) {
1242                 case OVS_NAT_ATTR_SRC:
1243                 case OVS_NAT_ATTR_DST:
1244                         if (info->nat) {
1245                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1246                                 return -ERANGE;
1247                         }
1248                         info->nat |= OVS_CT_NAT;
1249                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1250                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1251                         break;
1252
1253                 case OVS_NAT_ATTR_IP_MIN:
1254                         nla_memcpy(&info->range.min_addr, a,
1255                                    sizeof(info->range.min_addr));
1256                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1257                         break;
1258
1259                 case OVS_NAT_ATTR_IP_MAX:
1260                         have_ip_max = true;
1261                         nla_memcpy(&info->range.max_addr, a,
1262                                    sizeof(info->range.max_addr));
1263                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1264                         break;
1265
1266                 case OVS_NAT_ATTR_PROTO_MIN:
1267                         info->range.min_proto.all = htons(nla_get_u16(a));
1268                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1269                         break;
1270
1271                 case OVS_NAT_ATTR_PROTO_MAX:
1272                         have_proto_max = true;
1273                         info->range.max_proto.all = htons(nla_get_u16(a));
1274                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1275                         break;
1276
1277                 case OVS_NAT_ATTR_PERSISTENT:
1278                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1279                         break;
1280
1281                 case OVS_NAT_ATTR_PROTO_HASH:
1282                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1283                         break;
1284
1285                 case OVS_NAT_ATTR_PROTO_RANDOM:
1286                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1287                         break;
1288
1289                 default:
1290                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1291                         return -EINVAL;
1292                 }
1293         }
1294
1295         if (rem > 0) {
1296                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1297                 return -EINVAL;
1298         }
1299         if (!info->nat) {
1300                 /* Do not allow flags if no type is given. */
1301                 if (info->range.flags) {
1302                         OVS_NLERR(log,
1303                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1304                                   );
1305                         return -EINVAL;
1306                 }
1307                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1308         } else if (!info->commit) {
1309                 OVS_NLERR(log,
1310                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1311                           );
1312                 return -EINVAL;
1313         }
1314         /* Allow missing IP_MAX. */
1315         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1316                 memcpy(&info->range.max_addr, &info->range.min_addr,
1317                        sizeof(info->range.max_addr));
1318         }
1319         /* Allow missing PROTO_MAX. */
1320         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1321             !have_proto_max) {
1322                 info->range.max_proto.all = info->range.min_proto.all;
1323         }
1324         return 0;
1325 }
1326 #endif
1327
1328 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1329         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1330         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1331         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1332                                     .maxlen = sizeof(u16) },
1333         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1334                                     .maxlen = sizeof(struct md_mark) },
1335         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1336                                     .maxlen = sizeof(struct md_labels) },
1337         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1338                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1339 #ifdef CONFIG_NF_NAT_NEEDED
1340         /* NAT length is checked when parsing the nested attributes. */
1341         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1342 #endif
1343         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1344                                     .maxlen = sizeof(u32) },
1345 };
1346
1347 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1348                     const char **helper, bool log)
1349 {
1350         struct nlattr *a;
1351         int rem;
1352
1353         nla_for_each_nested(a, attr, rem) {
1354                 int type = nla_type(a);
1355                 int maxlen;
1356                 int minlen;
1357
1358                 if (type > OVS_CT_ATTR_MAX) {
1359                         OVS_NLERR(log,
1360                                   "Unknown conntrack attr (type=%d, max=%d)",
1361                                   type, OVS_CT_ATTR_MAX);
1362                         return -EINVAL;
1363                 }
1364
1365                 maxlen = ovs_ct_attr_lens[type].maxlen;
1366                 minlen = ovs_ct_attr_lens[type].minlen;
1367                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1368                         OVS_NLERR(log,
1369                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1370                                   type, nla_len(a), maxlen);
1371                         return -EINVAL;
1372                 }
1373
1374                 switch (type) {
1375                 case OVS_CT_ATTR_FORCE_COMMIT:
1376                         info->force = true;
1377                         /* fall through. */
1378                 case OVS_CT_ATTR_COMMIT:
1379                         info->commit = true;
1380                         break;
1381 #ifdef CONFIG_NF_CONNTRACK_ZONES
1382                 case OVS_CT_ATTR_ZONE:
1383                         info->zone.id = nla_get_u16(a);
1384                         break;
1385 #endif
1386 #ifdef CONFIG_NF_CONNTRACK_MARK
1387                 case OVS_CT_ATTR_MARK: {
1388                         struct md_mark *mark = nla_data(a);
1389
1390                         if (!mark->mask) {
1391                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1392                                 return -EINVAL;
1393                         }
1394                         info->mark = *mark;
1395                         break;
1396                 }
1397 #endif
1398 #ifdef CONFIG_NF_CONNTRACK_LABELS
1399                 case OVS_CT_ATTR_LABELS: {
1400                         struct md_labels *labels = nla_data(a);
1401
1402                         if (!labels_nonzero(&labels->mask)) {
1403                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1404                                 return -EINVAL;
1405                         }
1406                         info->labels = *labels;
1407                         break;
1408                 }
1409 #endif
1410                 case OVS_CT_ATTR_HELPER:
1411                         *helper = nla_data(a);
1412                         if (!memchr(*helper, '\0', nla_len(a))) {
1413                                 OVS_NLERR(log, "Invalid conntrack helper");
1414                                 return -EINVAL;
1415                         }
1416                         break;
1417 #ifdef CONFIG_NF_NAT_NEEDED
1418                 case OVS_CT_ATTR_NAT: {
1419                         int err = parse_nat(a, info, log);
1420
1421                         if (err)
1422                                 return err;
1423                         break;
1424                 }
1425 #endif
1426                 case OVS_CT_ATTR_EVENTMASK:
1427                         info->have_eventmask = true;
1428                         info->eventmask = nla_get_u32(a);
1429                         break;
1430
1431                 default:
1432                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1433                                   type);
1434                         return -EINVAL;
1435                 }
1436         }
1437
1438 #ifdef CONFIG_NF_CONNTRACK_MARK
1439         if (!info->commit && info->mark.mask) {
1440                 OVS_NLERR(log,
1441                           "Setting conntrack mark requires 'commit' flag.");
1442                 return -EINVAL;
1443         }
1444 #endif
1445 #ifdef CONFIG_NF_CONNTRACK_LABELS
1446         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1447                 OVS_NLERR(log,
1448                           "Setting conntrack labels requires 'commit' flag.");
1449                 return -EINVAL;
1450         }
1451 #endif
1452         if (rem > 0) {
1453                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1454                 return -EINVAL;
1455         }
1456
1457         return 0;
1458 }
1459
1460 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1461 {
1462         if (attr == OVS_KEY_ATTR_CT_STATE)
1463                 return true;
1464         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1465             attr == OVS_KEY_ATTR_CT_ZONE)
1466                 return true;
1467         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1468             attr == OVS_KEY_ATTR_CT_MARK)
1469                 return true;
1470         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1471             attr == OVS_KEY_ATTR_CT_LABELS) {
1472                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1473
1474                 return ovs_net->xt_label;
1475         }
1476
1477         return false;
1478 }
1479
1480 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1481                        const struct sw_flow_key *key,
1482                        struct sw_flow_actions **sfa,  bool log)
1483 {
1484         struct ovs_conntrack_info ct_info;
1485         const char *helper = NULL;
1486         u16 family;
1487         int err;
1488
1489         family = key_to_nfproto(key);
1490         if (family == NFPROTO_UNSPEC) {
1491                 OVS_NLERR(log, "ct family unspecified");
1492                 return -EINVAL;
1493         }
1494
1495         memset(&ct_info, 0, sizeof(ct_info));
1496         ct_info.family = family;
1497
1498         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1499                         NF_CT_DEFAULT_ZONE_DIR, 0);
1500
1501         err = parse_ct(attr, &ct_info, &helper, log);
1502         if (err)
1503                 return err;
1504
1505         /* Set up template for tracking connections in specific zones. */
1506         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1507         if (!ct_info.ct) {
1508                 OVS_NLERR(log, "Failed to allocate conntrack template");
1509                 return -ENOMEM;
1510         }
1511
1512         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1513         nf_conntrack_get(&ct_info.ct->ct_general);
1514
1515         if (helper) {
1516                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1517                 if (err)
1518                         goto err_free_ct;
1519         }
1520
1521         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1522                                  sizeof(ct_info), log);
1523         if (err)
1524                 goto err_free_ct;
1525
1526         return 0;
1527 err_free_ct:
1528         __ovs_ct_free_action(&ct_info);
1529         return err;
1530 }
1531
1532 #ifdef CONFIG_NF_NAT_NEEDED
1533 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1534                                struct sk_buff *skb)
1535 {
1536         struct nlattr *start;
1537
1538         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1539         if (!start)
1540                 return false;
1541
1542         if (info->nat & OVS_CT_SRC_NAT) {
1543                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1544                         return false;
1545         } else if (info->nat & OVS_CT_DST_NAT) {
1546                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1547                         return false;
1548         } else {
1549                 goto out;
1550         }
1551
1552         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1553                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1554                     info->family == NFPROTO_IPV4) {
1555                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1556                                             info->range.min_addr.ip) ||
1557                             (info->range.max_addr.ip
1558                              != info->range.min_addr.ip &&
1559                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1560                                               info->range.max_addr.ip))))
1561                                 return false;
1562                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1563                            info->family == NFPROTO_IPV6) {
1564                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1565                                              &info->range.min_addr.in6) ||
1566                             (memcmp(&info->range.max_addr.in6,
1567                                     &info->range.min_addr.in6,
1568                                     sizeof(info->range.max_addr.in6)) &&
1569                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1570                                                &info->range.max_addr.in6))))
1571                                 return false;
1572                 } else {
1573                         return false;
1574                 }
1575         }
1576         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1577             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1578                          ntohs(info->range.min_proto.all)) ||
1579              (info->range.max_proto.all != info->range.min_proto.all &&
1580               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1581                           ntohs(info->range.max_proto.all)))))
1582                 return false;
1583
1584         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1585             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1586                 return false;
1587         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1588             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1589                 return false;
1590         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1591             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1592                 return false;
1593 out:
1594         nla_nest_end(skb, start);
1595
1596         return true;
1597 }
1598 #endif
1599
1600 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1601                           struct sk_buff *skb)
1602 {
1603         struct nlattr *start;
1604
1605         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1606         if (!start)
1607                 return -EMSGSIZE;
1608
1609         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1610                                             ? OVS_CT_ATTR_FORCE_COMMIT
1611                                             : OVS_CT_ATTR_COMMIT))
1612                 return -EMSGSIZE;
1613         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1614             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1615                 return -EMSGSIZE;
1616         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1617             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1618                     &ct_info->mark))
1619                 return -EMSGSIZE;
1620         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1621             labels_nonzero(&ct_info->labels.mask) &&
1622             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1623                     &ct_info->labels))
1624                 return -EMSGSIZE;
1625         if (ct_info->helper) {
1626                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1627                                    ct_info->helper->name))
1628                         return -EMSGSIZE;
1629         }
1630         if (ct_info->have_eventmask &&
1631             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1632                 return -EMSGSIZE;
1633
1634 #ifdef CONFIG_NF_NAT_NEEDED
1635         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1636                 return -EMSGSIZE;
1637 #endif
1638         nla_nest_end(skb, start);
1639
1640         return 0;
1641 }
1642
1643 void ovs_ct_free_action(const struct nlattr *a)
1644 {
1645         struct ovs_conntrack_info *ct_info = nla_data(a);
1646
1647         __ovs_ct_free_action(ct_info);
1648 }
1649
1650 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1651 {
1652         if (ct_info->helper)
1653                 nf_conntrack_helper_put(ct_info->helper);
1654         if (ct_info->ct)
1655                 nf_ct_tmpl_free(ct_info->ct);
1656 }
1657
1658 void ovs_ct_init(struct net *net)
1659 {
1660         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1661         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1662
1663         if (nf_connlabels_get(net, n_bits - 1)) {
1664                 ovs_net->xt_label = false;
1665                 OVS_NLERR(true, "Failed to set connlabel length");
1666         } else {
1667                 ovs_net->xt_label = true;
1668         }
1669 }
1670
1671 void ovs_ct_exit(struct net *net)
1672 {
1673         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1674
1675         if (ovs_net->xt_label)
1676                 nf_connlabels_put(net);
1677 }