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
 520         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 521                 return NULL;
 522         return __nf_ct_expect_find(net, zone, &tuple);
 523 }
 524
 525 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 526 static enum ip_conntrack_info
 527 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 528 {
 529         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 530
 531         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 532                 return IP_CT_ESTABLISHED_REPLY;
 533         /* Once we've had two way comms, always ESTABLISHED. */
 534         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 535                 return IP_CT_ESTABLISHED;
 536         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 537                 return IP_CT_RELATED;
 538         return IP_CT_NEW;
 539 }
 540
 541 /* Find an existing connection which this packet belongs to without
 542  * re-attributing statistics or modifying the connection state.  This allows an
 543  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 544  *
 545  * Must be called with rcu_read_lock.
 546  *
 547  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 548  * if there is no existing entry.
 549  */
 550 static struct nf_conn *
 551 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 552                      u8 l3num, struct sk_buff *skb, bool natted)
 553 {
 554         struct nf_conntrack_l3proto *l3proto;
 555         struct nf_conntrack_l4proto *l4proto;
 556         struct nf_conntrack_tuple tuple;
 557         struct nf_conntrack_tuple_hash *h;
 558         struct nf_conn *ct;
 559         unsigned int dataoff;
 560         u8 protonum;
 561
 562         l3proto = __nf_ct_l3proto_find(l3num);
 563         if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
 564                                  &protonum) <= 0) {
 565                 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
 566                 return NULL;
 567         }
 568         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 569         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
 570                              protonum, net, &tuple, l3proto, l4proto)) {
 571                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 572                 return NULL;
 573         }
 574
 575         /* Must invert the tuple if skb has been transformed by NAT. */
 576         if (natted) {
 577                 struct nf_conntrack_tuple inverse;
 578
 579                 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
 580                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 581                         return NULL;
 582                 }
 583                 tuple = inverse;
 584         }
 585
 586         /* look for tuple match */
 587         h = nf_conntrack_find_get(net, zone, &tuple);
 588         if (!h)
 589                 return NULL;   /* Not found. */
 590
 591         ct = nf_ct_tuplehash_to_ctrack(h);
 592
 593         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 594          * select the other tuplehash to get the right 'ctinfo' bits for this
 595          * packet.
 596          */
 597         if (natted)
 598                 h = &ct->tuplehash[!h->tuple.dst.dir];
 599
 600         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 601         return ct;
 602 }
 603
 604 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 605 static bool skb_nfct_cached(struct net *net,
 606                             const struct sw_flow_key *key,
 607                             const struct ovs_conntrack_info *info,
 608                             struct sk_buff *skb)
 609 {
 610         enum ip_conntrack_info ctinfo;
 611         struct nf_conn *ct;
 612
 613         ct = nf_ct_get(skb, &ctinfo);
 614         /* If no ct, check if we have evidence that an existing conntrack entry
 615          * might be found for this skb.  This happens when we lose a skb->_nfct
 616          * due to an upcall.  If the connection was not confirmed, it is not
 617          * cached and needs to be run through conntrack again.
 618          */
 619         if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
 620             !(key->ct_state & OVS_CS_F_INVALID) &&
 621             key->ct_zone == info->zone.id) {
 622                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 623                                           !!(key->ct_state
 624                                              & OVS_CS_F_NAT_MASK));
 625                 if (ct)
 626                         nf_ct_get(skb, &ctinfo);
 627         }
 628         if (!ct)
 629                 return false;
 630         if (!net_eq(net, read_pnet(&ct->ct_net)))
 631                 return false;
 632         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 633                 return false;
 634         if (info->helper) {
 635                 struct nf_conn_help *help;
 636
 637                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 638                 if (help && rcu_access_pointer(help->helper) != info->helper)
 639                         return false;
 640         }
 641         /* Force conntrack entry direction to the current packet? */
 642         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 643                 /* Delete the conntrack entry if confirmed, else just release
 644                  * the reference.
 645                  */
 646                 if (nf_ct_is_confirmed(ct))
 647                         nf_ct_delete(ct, 0, 0);
 648
 649                 nf_conntrack_put(&ct->ct_general);
 650                 nf_ct_set(skb, NULL, 0);
 651                 return false;
 652         }
 653
 654         return true;
 655 }
 656
 657 #ifdef CONFIG_NF_NAT_NEEDED
 658 /* Modelled after nf_nat_ipv[46]_fn().
 659  * range is only used for new, uninitialized NAT state.
 660  * Returns either NF_ACCEPT or NF_DROP.
 661  */
 662 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 663                               enum ip_conntrack_info ctinfo,
 664                               const struct nf_nat_range *range,
 665                               enum nf_nat_manip_type maniptype)
 666 {
 667         int hooknum, nh_off, err = NF_ACCEPT;
 668
 669         nh_off = skb_network_offset(skb);
 670         skb_pull_rcsum(skb, nh_off);
 671
 672         /* See HOOK2MANIP(). */
 673         if (maniptype == NF_NAT_MANIP_SRC)
 674                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 675         else
 676                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 677
 678         switch (ctinfo) {
 679         case IP_CT_RELATED:
 680         case IP_CT_RELATED_REPLY:
 681                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
 682                     skb->protocol == htons(ETH_P_IP) &&
 683                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 684                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 685                                                            hooknum))
 686                                 err = NF_DROP;
 687                         goto push;
 688                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
 689                            skb->protocol == htons(ETH_P_IPV6)) {
 690                         __be16 frag_off;
 691                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 692                         int hdrlen = ipv6_skip_exthdr(skb,
 693                                                       sizeof(struct ipv6hdr),
 694                                                       &nexthdr, &frag_off);
 695
 696                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 697                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 698                                                                      ctinfo,
 699                                                                      hooknum,
 700                                                                      hdrlen))
 701                                         err = NF_DROP;
 702                                 goto push;
 703                         }
 704                 }
 705                 /* Non-ICMP, fall thru to initialize if needed. */
 706         case IP_CT_NEW:
 707                 /* Seen it before?  This can happen for loopback, retrans,
 708                  * or local packets.
 709                  */
 710                 if (!nf_nat_initialized(ct, maniptype)) {
 711                         /* Initialize according to the NAT action. */
 712                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 713                                 /* Action is set up to establish a new
 714                                  * mapping.
 715                                  */
 716                                 ? nf_nat_setup_info(ct, range, maniptype)
 717                                 : nf_nat_alloc_null_binding(ct, hooknum);
 718                         if (err != NF_ACCEPT)
 719                                 goto push;
 720                 }
 721                 break;
 722
 723         case IP_CT_ESTABLISHED:
 724         case IP_CT_ESTABLISHED_REPLY:
 725                 break;
 726
 727         default:
 728                 err = NF_DROP;
 729                 goto push;
 730         }
 731
 732         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 733 push:
 734         skb_push(skb, nh_off);
 735         skb_postpush_rcsum(skb, skb->data, nh_off);
 736
 737         return err;
 738 }
 739
 740 static void ovs_nat_update_key(struct sw_flow_key *key,
 741                                const struct sk_buff *skb,
 742                                enum nf_nat_manip_type maniptype)
 743 {
 744         if (maniptype == NF_NAT_MANIP_SRC) {
 745                 __be16 src;
 746
 747                 key->ct_state |= OVS_CS_F_SRC_NAT;
 748                 if (key->eth.type == htons(ETH_P_IP))
 749                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 750                 else if (key->eth.type == htons(ETH_P_IPV6))
 751                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 752                                sizeof(key->ipv6.addr.src));
 753                 else
 754                         return;
 755
 756                 if (key->ip.proto == IPPROTO_UDP)
 757                         src = udp_hdr(skb)->source;
 758                 else if (key->ip.proto == IPPROTO_TCP)
 759                         src = tcp_hdr(skb)->source;
 760                 else if (key->ip.proto == IPPROTO_SCTP)
 761                         src = sctp_hdr(skb)->source;
 762                 else
 763                         return;
 764
 765                 key->tp.src = src;
 766         } else {
 767                 __be16 dst;
 768
 769                 key->ct_state |= OVS_CS_F_DST_NAT;
 770                 if (key->eth.type == htons(ETH_P_IP))
 771                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 772                 else if (key->eth.type == htons(ETH_P_IPV6))
 773                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 774                                sizeof(key->ipv6.addr.dst));
 775                 else
 776                         return;
 777
 778                 if (key->ip.proto == IPPROTO_UDP)
 779                         dst = udp_hdr(skb)->dest;
 780                 else if (key->ip.proto == IPPROTO_TCP)
 781                         dst = tcp_hdr(skb)->dest;
 782                 else if (key->ip.proto == IPPROTO_SCTP)
 783                         dst = sctp_hdr(skb)->dest;
 784                 else
 785                         return;
 786
 787                 key->tp.dst = dst;
 788         }
 789 }
 790
 791 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 792 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 793                       const struct ovs_conntrack_info *info,
 794                       struct sk_buff *skb, struct nf_conn *ct,
 795                       enum ip_conntrack_info ctinfo)
 796 {
 797         enum nf_nat_manip_type maniptype;
 798         int err;
 799
 800         /* Add NAT extension if not confirmed yet. */
 801         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 802                 return NF_ACCEPT;   /* Can't NAT. */
 803
 804         /* Determine NAT type.
 805          * Check if the NAT type can be deduced from the tracked connection.
 806          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 807          * when committing.
 808          */
 809         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 810             ct->status & IPS_NAT_MASK &&
 811             (ctinfo != IP_CT_RELATED || info->commit)) {
 812                 /* NAT an established or related connection like before. */
 813                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 814                         /* This is the REPLY direction for a connection
 815                          * for which NAT was applied in the forward
 816                          * direction.  Do the reverse NAT.
 817                          */
 818                         maniptype = ct->status & IPS_SRC_NAT
 819                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 820                 else
 821                         maniptype = ct->status & IPS_SRC_NAT
 822                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 823         } else if (info->nat & OVS_CT_SRC_NAT) {
 824                 maniptype = NF_NAT_MANIP_SRC;
 825         } else if (info->nat & OVS_CT_DST_NAT) {
 826                 maniptype = NF_NAT_MANIP_DST;
 827         } else {
 828                 return NF_ACCEPT; /* Connection is not NATed. */
 829         }
 830         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 831
 832         /* Mark NAT done if successful and update the flow key. */
 833         if (err == NF_ACCEPT)
 834                 ovs_nat_update_key(key, skb, maniptype);
 835
 836         return err;
 837 }
 838 #else /* !CONFIG_NF_NAT_NEEDED */
 839 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 840                       const struct ovs_conntrack_info *info,
 841                       struct sk_buff *skb, struct nf_conn *ct,
 842                       enum ip_conntrack_info ctinfo)
 843 {
 844         return NF_ACCEPT;
 845 }
 846 #endif
 847
 848 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 849  * not done already.  Update key with new CT state after passing the packet
 850  * through conntrack.
 851  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 852  * set to NULL and 0 will be returned.
 853  */
 854 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 855                            const struct ovs_conntrack_info *info,
 856                            struct sk_buff *skb)
 857 {
 858         /* If we are recirculating packets to match on conntrack fields and
 859          * committing with a separate conntrack action,  then we don't need to
 860          * actually run the packet through conntrack twice unless it's for a
 861          * different zone.
 862          */
 863         bool cached = skb_nfct_cached(net, key, info, skb);
 864         enum ip_conntrack_info ctinfo;
 865         struct nf_conn *ct;
 866
 867         if (!cached) {
 868                 struct nf_conn *tmpl = info->ct;
 869                 int err;
 870
 871                 /* Associate skb with specified zone. */
 872                 if (tmpl) {
 873                         if (skb_nfct(skb))
 874                                 nf_conntrack_put(skb_nfct(skb));
 875                         nf_conntrack_get(&tmpl->ct_general);
 876                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 877                 }
 878
 879                 err = nf_conntrack_in(net, info->family,
 880                                       NF_INET_PRE_ROUTING, skb);
 881                 if (err != NF_ACCEPT)
 882                         return -ENOENT;
 883
 884                 /* Clear CT state NAT flags to mark that we have not yet done
 885                  * NAT after the nf_conntrack_in() call.  We can actually clear
 886                  * the whole state, as it will be re-initialized below.
 887                  */
 888                 key->ct_state = 0;
 889
 890                 /* Update the key, but keep the NAT flags. */
 891                 ovs_ct_update_key(skb, info, key, true, true);
 892         }
 893
 894         ct = nf_ct_get(skb, &ctinfo);
 895         if (ct) {
 896                 /* Packets starting a new connection must be NATted before the
 897                  * helper, so that the helper knows about the NAT.  We enforce
 898                  * this by delaying both NAT and helper calls for unconfirmed
 899                  * connections until the committing CT action.  For later
 900                  * packets NAT and Helper may be called in either order.
 901                  *
 902                  * NAT will be done only if the CT action has NAT, and only
 903                  * once per packet (per zone), as guarded by the NAT bits in
 904                  * the key->ct_state.
 905                  */
 906                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
 907                     (nf_ct_is_confirmed(ct) || info->commit) &&
 908                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
 909                         return -EINVAL;
 910                 }
 911
 912                 /* Userspace may decide to perform a ct lookup without a helper
 913                  * specified followed by a (recirculate and) commit with one.
 914                  * Therefore, for unconfirmed connections which we will commit,
 915                  * we need to attach the helper here.
 916                  */
 917                 if (!nf_ct_is_confirmed(ct) && info->commit &&
 918                     info->helper && !nfct_help(ct)) {
 919                         int err = __nf_ct_try_assign_helper(ct, info->ct,
 920                                                             GFP_ATOMIC);
 921                         if (err)
 922                                 return err;
 923                 }
 924
 925                 /* Call the helper only if:
 926                  * - nf_conntrack_in() was executed above ("!cached") for a
 927                  *   confirmed connection, or
 928                  * - When committing an unconfirmed connection.
 929                  */
 930                 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
 931                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
 932                         return -EINVAL;
 933                 }
 934         }
 935
 936         return 0;
 937 }
 938
 939 /* Lookup connection and read fields into key. */
 940 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 941                          const struct ovs_conntrack_info *info,
 942                          struct sk_buff *skb)
 943 {
 944         struct nf_conntrack_expect *exp;
 945
 946         /* If we pass an expected packet through nf_conntrack_in() the
 947          * expectation is typically removed, but the packet could still be
 948          * lost in upcall processing.  To prevent this from happening we
 949          * perform an explicit expectation lookup.  Expected connections are
 950          * always new, and will be passed through conntrack only when they are
 951          * committed, as it is OK to remove the expectation at that time.
 952          */
 953         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
 954         if (exp) {
 955                 u8 state;
 956
 957                 /* NOTE: New connections are NATted and Helped only when
 958                  * committed, so we are not calling into NAT here.
 959                  */
 960                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
 961                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
 962         } else {
 963                 struct nf_conn *ct;
 964                 int err;
 965
 966                 err = __ovs_ct_lookup(net, key, info, skb);
 967                 if (err)
 968                         return err;
 969
 970                 ct = (struct nf_conn *)skb_nfct(skb);
 971                 if (ct)
 972                         nf_ct_deliver_cached_events(ct);
 973         }
 974
 975         return 0;
 976 }
 977
 978 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
 979 {
 980         size_t i;
 981
 982         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 983                 if (labels->ct_labels_32[i])
 984                         return true;
 985
 986         return false;
 987 }
 988
 989 /* Lookup connection and confirm if unconfirmed. */
 990 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
 991                          const struct ovs_conntrack_info *info,
 992                          struct sk_buff *skb)
 993 {
 994         enum ip_conntrack_info ctinfo;
 995         struct nf_conn *ct;
 996         int err;
 997
 998         err = __ovs_ct_lookup(net, key, info, skb);
 999         if (err)
1000                 return err;
1001
1002         /* The connection could be invalid, in which case this is a no-op.*/
1003         ct = nf_ct_get(skb, &ctinfo);
1004         if (!ct)
1005                 return 0;
1006
1007         /* Set the conntrack event mask if given.  NEW and DELETE events have
1008          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1009          * typically would receive many kinds of updates.  Setting the event
1010          * mask allows those events to be filtered.  The set event mask will
1011          * remain in effect for the lifetime of the connection unless changed
1012          * by a further CT action with both the commit flag and the eventmask
1013          * option. */
1014         if (info->have_eventmask) {
1015                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1016
1017                 if (cache)
1018                         cache->ctmask = info->eventmask;
1019         }
1020
1021         /* Apply changes before confirming the connection so that the initial
1022          * conntrack NEW netlink event carries the values given in the CT
1023          * action.
1024          */
1025         if (info->mark.mask) {
1026                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1027                                       info->mark.mask);
1028                 if (err)
1029                         return err;
1030         }
1031         if (!nf_ct_is_confirmed(ct)) {
1032                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1033                                          &info->labels.mask);
1034                 if (err)
1035                         return err;
1036         } else if (labels_nonzero(&info->labels.mask)) {
1037                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1038                                         &info->labels.mask);
1039                 if (err)
1040                         return err;
1041         }
1042         /* This will take care of sending queued events even if the connection
1043          * is already confirmed.
1044          */
1045         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1046                 return -EINVAL;
1047
1048         return 0;
1049 }
1050
1051 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1052  * value if 'skb' is freed.
1053  */
1054 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1055                    struct sw_flow_key *key,
1056                    const struct ovs_conntrack_info *info)
1057 {
1058         int nh_ofs;
1059         int err;
1060
1061         /* The conntrack module expects to be working at L3. */
1062         nh_ofs = skb_network_offset(skb);
1063         skb_pull_rcsum(skb, nh_ofs);
1064
1065         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1066                 err = handle_fragments(net, key, info->zone.id, skb);
1067                 if (err)
1068                         return err;
1069         }
1070
1071         if (info->commit)
1072                 err = ovs_ct_commit(net, key, info, skb);
1073         else
1074                 err = ovs_ct_lookup(net, key, info, skb);
1075
1076         skb_push(skb, nh_ofs);
1077         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1078         if (err)
1079                 kfree_skb(skb);
1080         return err;
1081 }
1082
1083 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1084                              const struct sw_flow_key *key, bool log)
1085 {
1086         struct nf_conntrack_helper *helper;
1087         struct nf_conn_help *help;
1088
1089         helper = nf_conntrack_helper_try_module_get(name, info->family,
1090                                                     key->ip.proto);
1091         if (!helper) {
1092                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1093                 return -EINVAL;
1094         }
1095
1096         help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1097         if (!help) {
1098                 module_put(helper->me);
1099                 return -ENOMEM;
1100         }
1101
1102         rcu_assign_pointer(help->helper, helper);
1103         info->helper = helper;
1104         return 0;
1105 }
1106
1107 #ifdef CONFIG_NF_NAT_NEEDED
1108 static int parse_nat(const struct nlattr *attr,
1109                      struct ovs_conntrack_info *info, bool log)
1110 {
1111         struct nlattr *a;
1112         int rem;
1113         bool have_ip_max = false;
1114         bool have_proto_max = false;
1115         bool ip_vers = (info->family == NFPROTO_IPV6);
1116
1117         nla_for_each_nested(a, attr, rem) {
1118                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1119                         [OVS_NAT_ATTR_SRC] = {0, 0},
1120                         [OVS_NAT_ATTR_DST] = {0, 0},
1121                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1122                                                  sizeof(struct in6_addr)},
1123                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1124                                                  sizeof(struct in6_addr)},
1125                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1126                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1127                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1128                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1129                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1130                 };
1131                 int type = nla_type(a);
1132
1133                 if (type > OVS_NAT_ATTR_MAX) {
1134                         OVS_NLERR(log,
1135                                   "Unknown NAT attribute (type=%d, max=%d).\n",
1136                                   type, OVS_NAT_ATTR_MAX);
1137                         return -EINVAL;
1138                 }
1139
1140                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1141                         OVS_NLERR(log,
1142                                   "NAT attribute type %d has unexpected length (%d != %d).\n",
1143                                   type, nla_len(a),
1144                                   ovs_nat_attr_lens[type][ip_vers]);
1145                         return -EINVAL;
1146                 }
1147
1148                 switch (type) {
1149                 case OVS_NAT_ATTR_SRC:
1150                 case OVS_NAT_ATTR_DST:
1151                         if (info->nat) {
1152                                 OVS_NLERR(log,
1153                                           "Only one type of NAT may be specified.\n"
1154                                           );
1155                                 return -ERANGE;
1156                         }
1157                         info->nat |= OVS_CT_NAT;
1158                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1159                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1160                         break;
1161
1162                 case OVS_NAT_ATTR_IP_MIN:
1163                         nla_memcpy(&info->range.min_addr, a,
1164                                    sizeof(info->range.min_addr));
1165                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1166                         break;
1167
1168                 case OVS_NAT_ATTR_IP_MAX:
1169                         have_ip_max = true;
1170                         nla_memcpy(&info->range.max_addr, a,
1171                                    sizeof(info->range.max_addr));
1172                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1173                         break;
1174
1175                 case OVS_NAT_ATTR_PROTO_MIN:
1176                         info->range.min_proto.all = htons(nla_get_u16(a));
1177                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1178                         break;
1179
1180                 case OVS_NAT_ATTR_PROTO_MAX:
1181                         have_proto_max = true;
1182                         info->range.max_proto.all = htons(nla_get_u16(a));
1183                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1184                         break;
1185
1186                 case OVS_NAT_ATTR_PERSISTENT:
1187                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1188                         break;
1189
1190                 case OVS_NAT_ATTR_PROTO_HASH:
1191                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1192                         break;
1193
1194                 case OVS_NAT_ATTR_PROTO_RANDOM:
1195                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1196                         break;
1197
1198                 default:
1199                         OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1200                         return -EINVAL;
1201                 }
1202         }
1203
1204         if (rem > 0) {
1205                 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1206                 return -EINVAL;
1207         }
1208         if (!info->nat) {
1209                 /* Do not allow flags if no type is given. */
1210                 if (info->range.flags) {
1211                         OVS_NLERR(log,
1212                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1213                                   );
1214                         return -EINVAL;
1215                 }
1216                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1217         } else if (!info->commit) {
1218                 OVS_NLERR(log,
1219                           "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1220                           );
1221                 return -EINVAL;
1222         }
1223         /* Allow missing IP_MAX. */
1224         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1225                 memcpy(&info->range.max_addr, &info->range.min_addr,
1226                        sizeof(info->range.max_addr));
1227         }
1228         /* Allow missing PROTO_MAX. */
1229         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1230             !have_proto_max) {
1231                 info->range.max_proto.all = info->range.min_proto.all;
1232         }
1233         return 0;
1234 }
1235 #endif
1236
1237 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1238         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1239         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1240         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1241                                     .maxlen = sizeof(u16) },
1242         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1243                                     .maxlen = sizeof(struct md_mark) },
1244         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1245                                     .maxlen = sizeof(struct md_labels) },
1246         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1247                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1248 #ifdef CONFIG_NF_NAT_NEEDED
1249         /* NAT length is checked when parsing the nested attributes. */
1250         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1251 #endif
1252         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1253                                     .maxlen = sizeof(u32) },
1254 };
1255
1256 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1257                     const char **helper, bool log)
1258 {
1259         struct nlattr *a;
1260         int rem;
1261
1262         nla_for_each_nested(a, attr, rem) {
1263                 int type = nla_type(a);
1264                 int maxlen = ovs_ct_attr_lens[type].maxlen;
1265                 int minlen = ovs_ct_attr_lens[type].minlen;
1266
1267                 if (type > OVS_CT_ATTR_MAX) {
1268                         OVS_NLERR(log,
1269                                   "Unknown conntrack attr (type=%d, max=%d)",
1270                                   type, OVS_CT_ATTR_MAX);
1271                         return -EINVAL;
1272                 }
1273                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1274                         OVS_NLERR(log,
1275                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1276                                   type, nla_len(a), maxlen);
1277                         return -EINVAL;
1278                 }
1279
1280                 switch (type) {
1281                 case OVS_CT_ATTR_FORCE_COMMIT:
1282                         info->force = true;
1283                         /* fall through. */
1284                 case OVS_CT_ATTR_COMMIT:
1285                         info->commit = true;
1286                         break;
1287 #ifdef CONFIG_NF_CONNTRACK_ZONES
1288                 case OVS_CT_ATTR_ZONE:
1289                         info->zone.id = nla_get_u16(a);
1290                         break;
1291 #endif
1292 #ifdef CONFIG_NF_CONNTRACK_MARK
1293                 case OVS_CT_ATTR_MARK: {
1294                         struct md_mark *mark = nla_data(a);
1295
1296                         if (!mark->mask) {
1297                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1298                                 return -EINVAL;
1299                         }
1300                         info->mark = *mark;
1301                         break;
1302                 }
1303 #endif
1304 #ifdef CONFIG_NF_CONNTRACK_LABELS
1305                 case OVS_CT_ATTR_LABELS: {
1306                         struct md_labels *labels = nla_data(a);
1307
1308                         if (!labels_nonzero(&labels->mask)) {
1309                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1310                                 return -EINVAL;
1311                         }
1312                         info->labels = *labels;
1313                         break;
1314                 }
1315 #endif
1316                 case OVS_CT_ATTR_HELPER:
1317                         *helper = nla_data(a);
1318                         if (!memchr(*helper, '\0', nla_len(a))) {
1319                                 OVS_NLERR(log, "Invalid conntrack helper");
1320                                 return -EINVAL;
1321                         }
1322                         break;
1323 #ifdef CONFIG_NF_NAT_NEEDED
1324                 case OVS_CT_ATTR_NAT: {
1325                         int err = parse_nat(a, info, log);
1326
1327                         if (err)
1328                                 return err;
1329                         break;
1330                 }
1331 #endif
1332                 case OVS_CT_ATTR_EVENTMASK:
1333                         info->have_eventmask = true;
1334                         info->eventmask = nla_get_u32(a);
1335                         break;
1336
1337                 default:
1338                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1339                                   type);
1340                         return -EINVAL;
1341                 }
1342         }
1343
1344 #ifdef CONFIG_NF_CONNTRACK_MARK
1345         if (!info->commit && info->mark.mask) {
1346                 OVS_NLERR(log,
1347                           "Setting conntrack mark requires 'commit' flag.");
1348                 return -EINVAL;
1349         }
1350 #endif
1351 #ifdef CONFIG_NF_CONNTRACK_LABELS
1352         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1353                 OVS_NLERR(log,
1354                           "Setting conntrack labels requires 'commit' flag.");
1355                 return -EINVAL;
1356         }
1357 #endif
1358         if (rem > 0) {
1359                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1360                 return -EINVAL;
1361         }
1362
1363         return 0;
1364 }
1365
1366 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1367 {
1368         if (attr == OVS_KEY_ATTR_CT_STATE)
1369                 return true;
1370         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1371             attr == OVS_KEY_ATTR_CT_ZONE)
1372                 return true;
1373         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1374             attr == OVS_KEY_ATTR_CT_MARK)
1375                 return true;
1376         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1377             attr == OVS_KEY_ATTR_CT_LABELS) {
1378                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1379
1380                 return ovs_net->xt_label;
1381         }
1382
1383         return false;
1384 }
1385
1386 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1387                        const struct sw_flow_key *key,
1388                        struct sw_flow_actions **sfa,  bool log)
1389 {
1390         struct ovs_conntrack_info ct_info;
1391         const char *helper = NULL;
1392         u16 family;
1393         int err;
1394
1395         family = key_to_nfproto(key);
1396         if (family == NFPROTO_UNSPEC) {
1397                 OVS_NLERR(log, "ct family unspecified");
1398                 return -EINVAL;
1399         }
1400
1401         memset(&ct_info, 0, sizeof(ct_info));
1402         ct_info.family = family;
1403
1404         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1405                         NF_CT_DEFAULT_ZONE_DIR, 0);
1406
1407         err = parse_ct(attr, &ct_info, &helper, log);
1408         if (err)
1409                 return err;
1410
1411         /* Set up template for tracking connections in specific zones. */
1412         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1413         if (!ct_info.ct) {
1414                 OVS_NLERR(log, "Failed to allocate conntrack template");
1415                 return -ENOMEM;
1416         }
1417
1418         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1419         nf_conntrack_get(&ct_info.ct->ct_general);
1420
1421         if (helper) {
1422                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1423                 if (err)
1424                         goto err_free_ct;
1425         }
1426
1427         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1428                                  sizeof(ct_info), log);
1429         if (err)
1430                 goto err_free_ct;
1431
1432         return 0;
1433 err_free_ct:
1434         __ovs_ct_free_action(&ct_info);
1435         return err;
1436 }
1437
1438 #ifdef CONFIG_NF_NAT_NEEDED
1439 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1440                                struct sk_buff *skb)
1441 {
1442         struct nlattr *start;
1443
1444         start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1445         if (!start)
1446                 return false;
1447
1448         if (info->nat & OVS_CT_SRC_NAT) {
1449                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1450                         return false;
1451         } else if (info->nat & OVS_CT_DST_NAT) {
1452                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1453                         return false;
1454         } else {
1455                 goto out;
1456         }
1457
1458         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1459                 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1460                     info->family == NFPROTO_IPV4) {
1461                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1462                                             info->range.min_addr.ip) ||
1463                             (info->range.max_addr.ip
1464                              != info->range.min_addr.ip &&
1465                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1466                                               info->range.max_addr.ip))))
1467                                 return false;
1468                 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1469                            info->family == NFPROTO_IPV6) {
1470                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1471                                              &info->range.min_addr.in6) ||
1472                             (memcmp(&info->range.max_addr.in6,
1473                                     &info->range.min_addr.in6,
1474                                     sizeof(info->range.max_addr.in6)) &&
1475                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1476                                                &info->range.max_addr.in6))))
1477                                 return false;
1478                 } else {
1479                         return false;
1480                 }
1481         }
1482         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1483             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1484                          ntohs(info->range.min_proto.all)) ||
1485              (info->range.max_proto.all != info->range.min_proto.all &&
1486               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1487                           ntohs(info->range.max_proto.all)))))
1488                 return false;
1489
1490         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1491             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1492                 return false;
1493         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1494             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1495                 return false;
1496         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1497             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1498                 return false;
1499 out:
1500         nla_nest_end(skb, start);
1501
1502         return true;
1503 }
1504 #endif
1505
1506 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1507                           struct sk_buff *skb)
1508 {
1509         struct nlattr *start;
1510
1511         start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1512         if (!start)
1513                 return -EMSGSIZE;
1514
1515         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1516                                             ? OVS_CT_ATTR_FORCE_COMMIT
1517                                             : OVS_CT_ATTR_COMMIT))
1518                 return -EMSGSIZE;
1519         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1520             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1521                 return -EMSGSIZE;
1522         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1523             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1524                     &ct_info->mark))
1525                 return -EMSGSIZE;
1526         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1527             labels_nonzero(&ct_info->labels.mask) &&
1528             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1529                     &ct_info->labels))
1530                 return -EMSGSIZE;
1531         if (ct_info->helper) {
1532                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1533                                    ct_info->helper->name))
1534                         return -EMSGSIZE;
1535         }
1536         if (ct_info->have_eventmask &&
1537             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1538                 return -EMSGSIZE;
1539
1540 #ifdef CONFIG_NF_NAT_NEEDED
1541         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1542                 return -EMSGSIZE;
1543 #endif
1544         nla_nest_end(skb, start);
1545
1546         return 0;
1547 }
1548
1549 void ovs_ct_free_action(const struct nlattr *a)
1550 {
1551         struct ovs_conntrack_info *ct_info = nla_data(a);
1552
1553         __ovs_ct_free_action(ct_info);
1554 }
1555
1556 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1557 {
1558         if (ct_info->helper)
1559                 module_put(ct_info->helper->me);
1560         if (ct_info->ct)
1561                 nf_ct_tmpl_free(ct_info->ct);
1562 }
1563
1564 void ovs_ct_init(struct net *net)
1565 {
1566         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1567         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1568
1569         if (nf_connlabels_get(net, n_bits - 1)) {
1570                 ovs_net->xt_label = false;
1571                 OVS_NLERR(true, "Failed to set connlabel length");
1572         } else {
1573                 ovs_net->xt_label = true;
1574         }
1575 }
1576
1577 void ovs_ct_exit(struct net *net)
1578 {
1579         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1580
1581         if (ovs_net->xt_label)
1582                 nf_connlabels_put(net);
1583 }