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                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 489                 err = nf_ct_frag6_gather(net, skb, user);
 490                 if (err) {
 491                         if (err != -EINPROGRESS)
 492                                 kfree_skb(skb);
 493                         return err;
 494                 }
 495
 496                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 497                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 498 #endif
 499         } else {
 500                 kfree_skb(skb);
 501                 return -EPFNOSUPPORT;
 502         }
 503
 504         key->ip.frag = OVS_FRAG_TYPE_NONE;
 505         skb_clear_hash(skb);
 506         skb->ignore_df = 1;
 507         *OVS_CB(skb) = ovs_cb;
 508
 509         return 0;
 510 }
 511
 512 static struct nf_conntrack_expect *
 513 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 514                    u16 proto, const struct sk_buff *skb)
 515 {
 516         struct nf_conntrack_tuple tuple;
 517
 518         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 519                 return NULL;
 520         return __nf_ct_expect_find(net, zone, &tuple);
 521 }
 522
 523 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 524 static enum ip_conntrack_info
 525 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 526 {
 527         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 528
 529         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 530                 return IP_CT_ESTABLISHED_REPLY;
 531         /* Once we've had two way comms, always ESTABLISHED. */
 532         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 533                 return IP_CT_ESTABLISHED;
 534         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 535                 return IP_CT_RELATED;
 536         return IP_CT_NEW;
 537 }
 538
 539 /* Find an existing connection which this packet belongs to without
 540  * re-attributing statistics or modifying the connection state.  This allows an
 541  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 542  *
 543  * Must be called with rcu_read_lock.
 544  *
 545  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 546  * if there is no existing entry.
 547  */
 548 static struct nf_conn *
 549 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 550                      u8 l3num, struct sk_buff *skb, bool natted)
 551 {
 552         struct nf_conntrack_l3proto *l3proto;
 553         struct nf_conntrack_l4proto *l4proto;
 554         struct nf_conntrack_tuple tuple;
 555         struct nf_conntrack_tuple_hash *h;
 556         struct nf_conn *ct;
 557         unsigned int dataoff;
 558         u8 protonum;
 559
 560         l3proto = __nf_ct_l3proto_find(l3num);
 561         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 562                                  &protonum) <= 0) {
 563                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 564                 return NULL;
 565         }
 566         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 567         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 568                              protonum, net, &tuple, l3proto, l4proto)) {
 569                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 570                 return NULL;
 571         }
 572
 573         /* Must invert the tuple if skb has been transformed by NAT. */
 574         if (natted) {
 575                 struct nf_conntrack_tuple inverse;
 576
 577                 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
 578                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 579                         return NULL;
 580                 }
 581                 tuple = inverse;
 582         }
 583
 584         /* look for tuple match */
 585         h = nf_conntrack_find_get(net, zone, &tuple);
 586         if (!h)
 587                 return NULL;   /* Not found. */
 588
 589         ct = nf_ct_tuplehash_to_ctrack(h);
 590
 591         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 592          * select the other tuplehash to get the right 'ctinfo' bits for this
 593          * packet.
 594          */
 595         if (natted)
 596                 h = &ct->tuplehash[!h->tuple.dst.dir];
 597
 598         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 599         return ct;
 600 }
 601
 602 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 603 static bool skb_nfct_cached(struct net *net,
 604                             const struct sw_flow_key *key,
 605                             const struct ovs_conntrack_info *info,
 606                             struct sk_buff *skb)
 607 {
 608         enum ip_conntrack_info ctinfo;
 609         struct nf_conn *ct;
 610
 611         ct = nf_ct_get(skb, &ctinfo);
 612         /* If no ct, check if we have evidence that an existing conntrack entry
 613          * might be found for this skb.  This happens when we lose a skb->_nfct
 614          * due to an upcall.  If the connection was not confirmed, it is not
 615          * cached and needs to be run through conntrack again.
 616          */
 617         if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
 618             !(key->ct_state & OVS_CS_F_INVALID) &&
 619             key->ct_zone == info->zone.id) {
 620                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 621                                           !!(key->ct_state
 622                                              & OVS_CS_F_NAT_MASK));
 623                 if (ct)
 624                         nf_ct_get(skb, &ctinfo);
 625         }
 626         if (!ct)
 627                 return false;
 628         if (!net_eq(net, read_pnet(&ct->ct_net)))
 629                 return false;
 630         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 631                 return false;
 632         if (info->helper) {
 633                 struct nf_conn_help *help;
 634
 635                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 636                 if (help && rcu_access_pointer(help->helper) != info->helper)
 637                         return false;
 638         }
 639         /* Force conntrack entry direction to the current packet? */
 640         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 641                 /* Delete the conntrack entry if confirmed, else just release
 642                  * the reference.
 643                  */
 644                 if (nf_ct_is_confirmed(ct))
 645                         nf_ct_delete(ct, 0, 0);
 646                 else
 647                         nf_conntrack_put(&ct->ct_general);
 648                 nf_ct_set(skb, NULL, 0);
 649                 return false;
 650         }
 651
 652         return true;
 653 }
 654
 655 #ifdef CONFIG_NF_NAT_NEEDED
 656 /* Modelled after nf_nat_ipv[46]_fn().
 657  * range is only used for new, uninitialized NAT state.
 658  * Returns either NF_ACCEPT or NF_DROP.
 659  */
 660 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 661                               enum ip_conntrack_info ctinfo,
 662                               const struct nf_nat_range *range,
 663                               enum nf_nat_manip_type maniptype)
 664 {
 665         int hooknum, nh_off, err = NF_ACCEPT;
 666
 667         nh_off = skb_network_offset(skb);
 668         skb_pull_rcsum(skb, nh_off);
 669
 670         /* See HOOK2MANIP(). */
 671         if (maniptype == NF_NAT_MANIP_SRC)
 672                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 673         else
 674                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 675
 676         switch (ctinfo) {
 677         case IP_CT_RELATED:
 678         case IP_CT_RELATED_REPLY:
 679                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 680                     skb->protocol == htons(ETH_P_IP) &&
 681                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 682                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 683                                                            hooknum))
 684                                 err = NF_DROP;
 685                         goto push;
 686                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 687                            skb->protocol == htons(ETH_P_IPV6)) {
 688                         __be16 frag_off;
 689                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 690                         int hdrlen = ipv6_skip_exthdr(skb,
 691                                                       sizeof(struct ipv6hdr),
 692                                                       &nexthdr, &frag_off);
 693
 694                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 695                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 696                                                                      ctinfo,
 697                                                                      hooknum,
 698                                                                      hdrlen))
 699                                         err = NF_DROP;
 700                                 goto push;
 701                         }
 702                 }
 703                 /* Non-ICMP, fall thru to initialize if needed. */
 704         case IP_CT_NEW:
 705                 /* Seen it before?  This can happen for loopback, retrans,
 706                  * or local packets.
 707                  */
 708                 if (!nf_nat_initialized(ct, maniptype)) {
 709                         /* Initialize according to the NAT action. */
 710                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 711                                 /* Action is set up to establish a new
 712                                  * mapping.
 713                                  */
 714                                 ? nf_nat_setup_info(ct, range, maniptype)
 715                                 : nf_nat_alloc_null_binding(ct, hooknum);
 716                         if (err != NF_ACCEPT)
 717                                 goto push;
 718                 }
 719                 break;
 720
 721         case IP_CT_ESTABLISHED:
 722         case IP_CT_ESTABLISHED_REPLY:
 723                 break;
 724
 725         default:
 726                 err = NF_DROP;
 727                 goto push;
 728         }
 729
 730         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 731 push:
 732         skb_push(skb, nh_off);
 733         skb_postpush_rcsum(skb, skb->data, nh_off);
 734
 735         return err;
 736 }
 737
 738 static void ovs_nat_update_key(struct sw_flow_key *key,
 739                                const struct sk_buff *skb,
 740                                enum nf_nat_manip_type maniptype)
 741 {
 742         if (maniptype == NF_NAT_MANIP_SRC) {
 743                 __be16 src;
 744
 745                 key->ct_state |= OVS_CS_F_SRC_NAT;
 746                 if (key->eth.type == htons(ETH_P_IP))
 747                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 748                 else if (key->eth.type == htons(ETH_P_IPV6))
 749                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 750                                sizeof(key->ipv6.addr.src));
 751                 else
 752                         return;
 753
 754                 if (key->ip.proto == IPPROTO_UDP)
 755                         src = udp_hdr(skb)->source;
 756                 else if (key->ip.proto == IPPROTO_TCP)
 757                         src = tcp_hdr(skb)->source;
 758                 else if (key->ip.proto == IPPROTO_SCTP)
 759                         src = sctp_hdr(skb)->source;
 760                 else
 761                         return;
 762
 763                 key->tp.src = src;
 764         } else {
 765                 __be16 dst;
 766
 767                 key->ct_state |= OVS_CS_F_DST_NAT;
 768                 if (key->eth.type == htons(ETH_P_IP))
 769                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 770                 else if (key->eth.type == htons(ETH_P_IPV6))
 771                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 772                                sizeof(key->ipv6.addr.dst));
 773                 else
 774                         return;
 775
 776                 if (key->ip.proto == IPPROTO_UDP)
 777                         dst = udp_hdr(skb)->dest;
 778                 else if (key->ip.proto == IPPROTO_TCP)
 779                         dst = tcp_hdr(skb)->dest;
 780                 else if (key->ip.proto == IPPROTO_SCTP)
 781                         dst = sctp_hdr(skb)->dest;
 782                 else
 783                         return;
 784
 785                 key->tp.dst = dst;
 786         }
 787 }
 788
 789 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 790 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 791                       const struct ovs_conntrack_info *info,
 792                       struct sk_buff *skb, struct nf_conn *ct,
 793                       enum ip_conntrack_info ctinfo)
 794 {
 795         enum nf_nat_manip_type maniptype;
 796         int err;
 797
 798         if (nf_ct_is_untracked(ct)) {
 799                 /* A NAT action may only be performed on tracked packets. */
 800                 return NF_ACCEPT;
 801         }
 802
 803         /* Add NAT extension if not confirmed yet. */
 804         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 805                 return NF_ACCEPT;   /* Can't NAT. */
 806
 807         /* Determine NAT type.
 808          * Check if the NAT type can be deduced from the tracked connection.
 809          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 810          * when committing.
 811          */
 812         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 813             ct->status & IPS_NAT_MASK &&
 814             (ctinfo != IP_CT_RELATED || info->commit)) {
 815                 /* NAT an established or related connection like before. */
 816                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 817                         /* This is the REPLY direction for a connection
 818                          * for which NAT was applied in the forward
 819                          * direction.  Do the reverse NAT.
 820                          */
 821                         maniptype = ct->status & IPS_SRC_NAT
 822                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 823                 else
 824                         maniptype = ct->status & IPS_SRC_NAT
 825                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 826         } else if (info->nat & OVS_CT_SRC_NAT) {
 827                 maniptype = NF_NAT_MANIP_SRC;
 828         } else if (info->nat & OVS_CT_DST_NAT) {
 829                 maniptype = NF_NAT_MANIP_DST;
 830         } else {
 831                 return NF_ACCEPT; /* Connection is not NATed. */
 832         }
 833         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 834
 835         /* Mark NAT done if successful and update the flow key. */
 836         if (err == NF_ACCEPT)
 837                 ovs_nat_update_key(key, skb, maniptype);
 838
 839         return err;
 840 }
 841 #else /* !CONFIG_NF_NAT_NEEDED */
 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         return NF_ACCEPT;
 848 }
 849 #endif
 850
 851 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 852  * not done already.  Update key with new CT state after passing the packet
 853  * through conntrack.
 854  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 855  * set to NULL and 0 will be returned.
 856  */
 857 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 858                            const struct ovs_conntrack_info *info,
 859                            struct sk_buff *skb)
 860 {
 861         /* If we are recirculating packets to match on conntrack fields and
 862          * committing with a separate conntrack action,  then we don't need to
 863          * actually run the packet through conntrack twice unless it's for a
 864          * different zone.
 865          */
 866         bool cached = skb_nfct_cached(net, key, info, skb);
 867         enum ip_conntrack_info ctinfo;
 868         struct nf_conn *ct;
 869
 870         if (!cached) {
 871                 struct nf_conn *tmpl = info->ct;
 872                 int err;
 873
 874                 /* Associate skb with specified zone. */
 875                 if (tmpl) {
 876                         if (skb_nfct(skb))
 877                                 nf_conntrack_put(skb_nfct(skb));
 878                         nf_conntrack_get(&tmpl->ct_general);
 879                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 880                 }
 881
 882                 err = nf_conntrack_in(net, info->family,
 883                                       NF_INET_PRE_ROUTING, skb);
 884                 if (err != NF_ACCEPT)
 885                         return -ENOENT;
 886
 887                 /* Clear CT state NAT flags to mark that we have not yet done
 888                  * NAT after the nf_conntrack_in() call.  We can actually clear
 889                  * the whole state, as it will be re-initialized below.
 890                  */
 891                 key->ct_state = 0;
 892
 893                 /* Update the key, but keep the NAT flags. */
 894                 ovs_ct_update_key(skb, info, key, true, true);
 895         }
 896
 897         ct = nf_ct_get(skb, &ctinfo);
 898         if (ct) {
 899                 /* Packets starting a new connection must be NATted before the
 900                  * helper, so that the helper knows about the NAT.  We enforce
 901                  * this by delaying both NAT and helper calls for unconfirmed
 902                  * connections until the committing CT action.  For later
 903                  * packets NAT and Helper may be called in either order.
 904                  *
 905                  * NAT will be done only if the CT action has NAT, and only
 906                  * once per packet (per zone), as guarded by the NAT bits in
 907                  * the key->ct_state.
 908                  */
 909                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 910                     (nf_ct_is_confirmed(ct) || info->commit) &&
 911                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 912                         return -EINVAL;
 913                 }
 914
 915                 /* Userspace may decide to perform a ct lookup without a helper
 916                  * specified followed by a (recirculate and) commit with one.
 917                  * Therefore, for unconfirmed connections which we will commit,
 918                  * we need to attach the helper here.
 919                  */
 920                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 921                     info->helper && !nfct_help(ct)) {
 922                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 923                                                             GFP_ATOMIC);
 924                         if (err)
 925                                 return err;
 926                 }
 927
 928                 /* Call the helper only if:
 929                  * - nf_conntrack_in() was executed above ("!cached") for a
 930                  *   confirmed connection, or
 931                  * - When committing an unconfirmed connection.
 932                  */
 933                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 934                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 935                         return -EINVAL;
 936                 }
 937         }
 938
 939         return 0;
 940 }
 941
 942 /* Lookup connection and read fields into key. */
 943 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 944                          const struct ovs_conntrack_info *info,
 945                          struct sk_buff *skb)
 946 {
 947         struct nf_conntrack_expect *exp;
 948
 949         /* If we pass an expected packet through nf_conntrack_in() the
 950          * expectation is typically removed, but the packet could still be
 951          * lost in upcall processing.  To prevent this from happening we
 952          * perform an explicit expectation lookup.  Expected connections are
 953          * always new, and will be passed through conntrack only when they are
 954          * committed, as it is OK to remove the expectation at that time.
 955          */
 956         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 957         if (exp) {
 958                 u8 state;
 959
 960                 /* NOTE: New connections are NATted and Helped only when
 961                  * committed, so we are not calling into NAT here.
 962                  */
 963                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 964                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 965         } else {
 966                 struct nf_conn *ct;
 967                 int err;
 968
 969                 err = __ovs_ct_lookup(net, key, info, skb);
 970                 if (err)
 971                         return err;
 972
 973                 ct = (struct nf_conn *)skb_nfct(skb);
 974                 if (ct)
 975                         nf_ct_deliver_cached_events(ct);
 976         }
 977
 978         return 0;
 979 }
 980
 981 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 982 {
 983         size_t i;
 984
 985         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 986                 if (labels->ct_labels_32[i])
 987                         return true;
 988
 989         return false;
 990 }
 991
 992 /* Lookup connection and confirm if unconfirmed. */
 993 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 994                          const struct ovs_conntrack_info *info,
 995                          struct sk_buff *skb)
 996 {
 997         enum ip_conntrack_info ctinfo;
 998         struct nf_conn *ct;
 999         int err;
1000
1001         err = __ovs_ct_lookup(net, key, info, skb);
1002         if (err)
1003                 return err;
1004
1005         /* The connection could be invalid, in which case this is a no-op.*/
1006         ct = nf_ct_get(skb, &ctinfo);
1007         if (!ct)
1008                 return 0;
1009
1010         /* Apply changes before confirming the connection so that the initial
1011          * conntrack NEW netlink event carries the values given in the CT
1012          * action.
1013          */
1014         if (info->mark.mask) {
1015                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1016                                       info->mark.mask);
1017                 if (err)
1018                         return err;
1019         }
1020         if (!nf_ct_is_confirmed(ct)) {
1021                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1022                                          &info->labels.mask);
1023                 if (err)
1024                         return err;
1025         } else if (labels_nonzero(&info->labels.mask)) {
1026                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1027                                         &info->labels.mask);
1028                 if (err)
1029                         return err;
1030         }
1031         /* This will take care of sending queued events even if the connection
1032          * is already confirmed.
1033          */
1034         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1035                 return -EINVAL;
1036
1037         return 0;
1038 }
1039
1040 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1041  * value if 'skb' is freed.
1042  */
1043 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1044                    struct sw_flow_key *key,
1045                    const struct ovs_conntrack_info *info)
1046 {
1047         int nh_ofs;
1048         int err;
1049
1050         /* The conntrack module expects to be working at L3. */
1051         nh_ofs = skb_network_offset(skb);
1052         skb_pull_rcsum(skb, nh_ofs);
1053
1054         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1055                 err = handle_fragments(net, key, info->zone.id, skb);
1056                 if (err)
1057                         return err;
1058         }
1059
1060         if (info->commit)
1061                 err = ovs_ct_commit(net, key, info, skb);
1062         else
1063                 err = ovs_ct_lookup(net, key, info, skb);
1064
1065         skb_push(skb, nh_ofs);
1066         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1067         if (err)
1068                 kfree_skb(skb);
1069         return err;
1070 }
1071
1072 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1073                              const struct sw_flow_key *key, bool log)
1074 {
1075         struct nf_conntrack_helper *helper;
1076         struct nf_conn_help *help;
1077
1078         helper = nf_conntrack_helper_try_module_get(name, info->family,
1079                                                     key->ip.proto);
1080         if (!helper) {
1081                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1082                 return -EINVAL;
1083         }
1084
1085         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1086         if (!help) {
1087                 module_put(helper->me);
1088                 return -ENOMEM;
1089         }
1090
1091         rcu_assign_pointer(help->helper, helper);
1092         info->helper = helper;
1093         return 0;
1094 }
1095
1096 #ifdef CONFIG_NF_NAT_NEEDED
1097 static int parse_nat(const struct nlattr *attr,
1098                      struct ovs_conntrack_info *info, bool log)
1099 {
1100         struct nlattr *a;
1101         int rem;
1102         bool have_ip_max = false;
1103         bool have_proto_max = false;
1104         bool ip_vers = (info->family == NFPROTO_IPV6);
1105
1106         nla_for_each_nested(a, attr, rem) {
1107                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1108                         [OVS_NAT_ATTR_SRC] = {0, 0},
1109                         [OVS_NAT_ATTR_DST] = {0, 0},
1110                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1111                                                  sizeof(struct in6_addr)},
1112                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1113                                                  sizeof(struct in6_addr)},
1114                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1115                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1116                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1117                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1118                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1119                 };
1120                 int type = nla_type(a);
1121
1122                 if (type > OVS_NAT_ATTR_MAX) {
1123                         OVS_NLERR(log,
1124                                   "Unknown NAT attribute (type=%d, max=%d).\n",
1125                                   type, OVS_NAT_ATTR_MAX);
1126                         return -EINVAL;
1127                 }
1128
1129                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1130                         OVS_NLERR(log,
1131                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
1132                                   type, nla_len(a),
1133                                   ovs_nat_attr_lens[type][ip_vers]);
1134                         return -EINVAL;
1135                 }
1136
1137                 switch (type) {
1138                 case OVS_NAT_ATTR_SRC:
1139                 case OVS_NAT_ATTR_DST:
1140                         if (info->nat) {
1141                                 OVS_NLERR(log,
1142                                           "Only one type of NAT may be specified.\n"
1143                                           );
1144                                 return -ERANGE;
1145                         }
1146                         info->nat |= OVS_CT_NAT;
1147                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1148                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1149                         break;
1150
1151                 case OVS_NAT_ATTR_IP_MIN:
1152                         nla_memcpy(&info->range.min_addr, a,
1153                                    sizeof(info->range.min_addr));
1154                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1155                         break;
1156
1157                 case OVS_NAT_ATTR_IP_MAX:
1158                         have_ip_max = true;
1159                         nla_memcpy(&info->range.max_addr, a,
1160                                    sizeof(info->range.max_addr));
1161                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1162                         break;
1163
1164                 case OVS_NAT_ATTR_PROTO_MIN:
1165                         info->range.min_proto.all = htons(nla_get_u16(a));
1166                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1167                         break;
1168
1169                 case OVS_NAT_ATTR_PROTO_MAX:
1170                         have_proto_max = true;
1171                         info->range.max_proto.all = htons(nla_get_u16(a));
1172                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1173                         break;
1174
1175                 case OVS_NAT_ATTR_PERSISTENT:
1176                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1177                         break;
1178
1179                 case OVS_NAT_ATTR_PROTO_HASH:
1180                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1181                         break;
1182
1183                 case OVS_NAT_ATTR_PROTO_RANDOM:
1184                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1185                         break;
1186
1187                 default:
1188                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1189                         return -EINVAL;
1190                 }
1191         }
1192
1193         if (rem > 0) {
1194                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1195                 return -EINVAL;
1196         }
1197         if (!info->nat) {
1198                 /* Do not allow flags if no type is given. */
1199                 if (info->range.flags) {
1200                         OVS_NLERR(log,
1201                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1202                                   );
1203                         return -EINVAL;
1204                 }
1205                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1206         } else if (!info->commit) {
1207                 OVS_NLERR(log,
1208                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1209                           );
1210                 return -EINVAL;
1211         }
1212         /* Allow missing IP_MAX. */
1213         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1214                 memcpy(&info->range.max_addr, &info->range.min_addr,
1215                        sizeof(info->range.max_addr));
1216         }
1217         /* Allow missing PROTO_MAX. */
1218         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1219             !have_proto_max) {
1220                 info->range.max_proto.all = info->range.min_proto.all;
1221         }
1222         return 0;
1223 }
1224 #endif
1225
1226 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1227         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1228         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1229         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1230                                     .maxlen = sizeof(u16) },
1231         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1232                                     .maxlen = sizeof(struct md_mark) },
1233         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1234                                     .maxlen = sizeof(struct md_labels) },
1235         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1236                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1237 #ifdef CONFIG_NF_NAT_NEEDED
1238         /* NAT length is checked when parsing the nested attributes. */
1239         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1240 #endif
1241 };
1242
1243 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1244                     const char **helper, bool log)
1245 {
1246         struct nlattr *a;
1247         int rem;
1248
1249         nla_for_each_nested(a, attr, rem) {
1250                 int type = nla_type(a);
1251                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1252                 int minlen = ovs_ct_attr_lens[type].minlen;
1253
1254                 if (type > OVS_CT_ATTR_MAX) {
1255                         OVS_NLERR(log,
1256                                   "Unknown conntrack attr (type=%d, max=%d)",
1257                                   type, OVS_CT_ATTR_MAX);
1258                         return -EINVAL;
1259                 }
1260                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1261                         OVS_NLERR(log,
1262                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1263                                   type, nla_len(a), maxlen);
1264                         return -EINVAL;
1265                 }
1266
1267                 switch (type) {
1268                 case OVS_CT_ATTR_FORCE_COMMIT:
1269                         info->force = true;
1270                         /* fall through. */
1271                 case OVS_CT_ATTR_COMMIT:
1272                         info->commit = true;
1273                         break;
1274 #ifdef CONFIG_NF_CONNTRACK_ZONES
1275                 case OVS_CT_ATTR_ZONE:
1276                         info->zone.id = nla_get_u16(a);
1277                         break;
1278 #endif
1279 #ifdef CONFIG_NF_CONNTRACK_MARK
1280                 case OVS_CT_ATTR_MARK: {
1281                         struct md_mark *mark = nla_data(a);
1282
1283                         if (!mark->mask) {
1284                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1285                                 return -EINVAL;
1286                         }
1287                         info->mark = *mark;
1288                         break;
1289                 }
1290 #endif
1291 #ifdef CONFIG_NF_CONNTRACK_LABELS
1292                 case OVS_CT_ATTR_LABELS: {
1293                         struct md_labels *labels = nla_data(a);
1294
1295                         if (!labels_nonzero(&labels->mask)) {
1296                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1297                                 return -EINVAL;
1298                         }
1299                         info->labels = *labels;
1300                         break;
1301                 }
1302 #endif
1303                 case OVS_CT_ATTR_HELPER:
1304                         *helper = nla_data(a);
1305                         if (!memchr(*helper, '\0', nla_len(a))) {
1306                                 OVS_NLERR(log, "Invalid conntrack helper");
1307                                 return -EINVAL;
1308                         }
1309                         break;
1310 #ifdef CONFIG_NF_NAT_NEEDED
1311                 case OVS_CT_ATTR_NAT: {
1312                         int err = parse_nat(a, info, log);
1313
1314                         if (err)
1315                                 return err;
1316                         break;
1317                 }
1318 #endif
1319                 default:
1320                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1321                                   type);
1322                         return -EINVAL;
1323                 }
1324         }
1325
1326 #ifdef CONFIG_NF_CONNTRACK_MARK
1327         if (!info->commit && info->mark.mask) {
1328                 OVS_NLERR(log,
1329                           "Setting conntrack mark requires 'commit' flag.");
1330                 return -EINVAL;
1331         }
1332 #endif
1333 #ifdef CONFIG_NF_CONNTRACK_LABELS
1334         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1335                 OVS_NLERR(log,
1336                           "Setting conntrack labels requires 'commit' flag.");
1337                 return -EINVAL;
1338         }
1339 #endif
1340         if (rem > 0) {
1341                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1342                 return -EINVAL;
1343         }
1344
1345         return 0;
1346 }
1347
1348 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1349 {
1350         if (attr == OVS_KEY_ATTR_CT_STATE)
1351                 return true;
1352         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1353             attr == OVS_KEY_ATTR_CT_ZONE)
1354                 return true;
1355         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1356             attr == OVS_KEY_ATTR_CT_MARK)
1357                 return true;
1358         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1359             attr == OVS_KEY_ATTR_CT_LABELS) {
1360                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1361
1362                 return ovs_net->xt_label;
1363         }
1364
1365         return false;
1366 }
1367
1368 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1369                        const struct sw_flow_key *key,
1370                        struct sw_flow_actions **sfa,  bool log)
1371 {
1372         struct ovs_conntrack_info ct_info;
1373         const char *helper = NULL;
1374         u16 family;
1375         int err;
1376
1377         family = key_to_nfproto(key);
1378         if (family == NFPROTO_UNSPEC) {
1379                 OVS_NLERR(log, "ct family unspecified");
1380                 return -EINVAL;
1381         }
1382
1383         memset(&ct_info, 0, sizeof(ct_info));
1384         ct_info.family = family;
1385
1386         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1387                         NF_CT_DEFAULT_ZONE_DIR, 0);
1388
1389         err = parse_ct(attr, &ct_info, &helper, log);
1390         if (err)
1391                 return err;
1392
1393         /* Set up template for tracking connections in specific zones. */
1394         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1395         if (!ct_info.ct) {
1396                 OVS_NLERR(log, "Failed to allocate conntrack template");
1397                 return -ENOMEM;
1398         }
1399
1400         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1401         nf_conntrack_get(&ct_info.ct->ct_general);
1402
1403         if (helper) {
1404                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1405                 if (err)
1406                         goto err_free_ct;
1407         }
1408
1409         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1410                                  sizeof(ct_info), log);
1411         if (err)
1412                 goto err_free_ct;
1413
1414         return 0;
1415 err_free_ct:
1416         __ovs_ct_free_action(&ct_info);
1417         return err;
1418 }
1419
1420 #ifdef CONFIG_NF_NAT_NEEDED
1421 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1422                                struct sk_buff *skb)
1423 {
1424         struct nlattr *start;
1425
1426         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1427         if (!start)
1428                 return false;
1429
1430         if (info->nat & OVS_CT_SRC_NAT) {
1431                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1432                         return false;
1433         } else if (info->nat & OVS_CT_DST_NAT) {
1434                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1435                         return false;
1436         } else {
1437                 goto out;
1438         }
1439
1440         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1441                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1442                     info->family == NFPROTO_IPV4) {
1443                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1444                                             info->range.min_addr.ip) ||
1445                             (info->range.max_addr.ip
1446                              != info->range.min_addr.ip &&
1447                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1448                                               info->range.max_addr.ip))))
1449                                 return false;
1450                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1451                            info->family == NFPROTO_IPV6) {
1452                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1453                                              &info->range.min_addr.in6) ||
1454                             (memcmp(&info->range.max_addr.in6,
1455                                     &info->range.min_addr.in6,
1456                                     sizeof(info->range.max_addr.in6)) &&
1457                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1458                                                &info->range.max_addr.in6))))
1459                                 return false;
1460                 } else {
1461                         return false;
1462                 }
1463         }
1464         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1465             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1466                          ntohs(info->range.min_proto.all)) ||
1467              (info->range.max_proto.all != info->range.min_proto.all &&
1468               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1469                           ntohs(info->range.max_proto.all)))))
1470                 return false;
1471
1472         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1473             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1474                 return false;
1475         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1476             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1477                 return false;
1478         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1479             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1480                 return false;
1481 out:
1482         nla_nest_end(skb, start);
1483
1484         return true;
1485 }
1486 #endif
1487
1488 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1489                           struct sk_buff *skb)
1490 {
1491         struct nlattr *start;
1492
1493         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1494         if (!start)
1495                 return -EMSGSIZE;
1496
1497         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1498                                             ? OVS_CT_ATTR_FORCE_COMMIT
1499                                             : OVS_CT_ATTR_COMMIT))
1500                 return -EMSGSIZE;
1501         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1502             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1503                 return -EMSGSIZE;
1504         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1505             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1506                     &ct_info->mark))
1507                 return -EMSGSIZE;
1508         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1509             labels_nonzero(&ct_info->labels.mask) &&
1510             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1511                     &ct_info->labels))
1512                 return -EMSGSIZE;
1513         if (ct_info->helper) {
1514                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1515                                    ct_info->helper->name))
1516                         return -EMSGSIZE;
1517         }
1518 #ifdef CONFIG_NF_NAT_NEEDED
1519         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1520                 return -EMSGSIZE;
1521 #endif
1522         nla_nest_end(skb, start);
1523
1524         return 0;
1525 }
1526
1527 void ovs_ct_free_action(const struct nlattr *a)
1528 {
1529         struct ovs_conntrack_info *ct_info = nla_data(a);
1530
1531         __ovs_ct_free_action(ct_info);
1532 }
1533
1534 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1535 {
1536         if (ct_info->helper)
1537                 module_put(ct_info->helper->me);
1538         if (ct_info->ct)
1539                 nf_ct_tmpl_free(ct_info->ct);
1540 }
1541
1542 void ovs_ct_init(struct net *net)
1543 {
1544         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1545         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1546
1547         if (nf_connlabels_get(net, n_bits - 1)) {
1548                 ovs_net->xt_label = false;
1549                 OVS_NLERR(true, "Failed to set connlabel length");
1550         } else {
1551                 ovs_net->xt_label = true;
1552         }
1553 }
1554
1555 void ovs_ct_exit(struct net *net)
1556 {
1557         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1558
1559         if (ovs_net->xt_label)
1560                 nf_connlabels_put(net);
1561 }