net/openvswitch/conntrack.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2015 Nicira, Inc.
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/openvswitch.h>
   8 #include <linux/tcp.h>
   9 #include <linux/udp.h>
  10 #include <linux/sctp.h>
  11 #include <linux/static_key.h>
  12 #include <net/ip.h>
  13 #include <net/genetlink.h>
  14 #include <net/netfilter/nf_conntrack_core.h>
  15 #include <net/netfilter/nf_conntrack_count.h>
  16 #include <net/netfilter/nf_conntrack_helper.h>
  17 #include <net/netfilter/nf_conntrack_labels.h>
  18 #include <net/netfilter/nf_conntrack_seqadj.h>
  19 #include <net/netfilter/nf_conntrack_timeout.h>
  20 #include <net/netfilter/nf_conntrack_zones.h>
  21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  22 #include <net/ipv6_frag.h>
  23
  24 #if IS_ENABLED(CONFIG_NF_NAT)
  25 #include <net/netfilter/nf_nat.h>
  26 #endif
  27
  28 #include "datapath.h"
  29 #include "conntrack.h"
  30 #include "flow.h"
  31 #include "flow_netlink.h"
  32
  33 struct ovs_ct_len_tbl {
  34         int maxlen;
  35         int minlen;
  36 };
  37
  38 /* Metadata mark for masked write to conntrack mark */
  39 struct md_mark {
  40         u32 value;
  41         u32 mask;
  42 };
  43
  44 /* Metadata label for masked write to conntrack label. */
  45 struct md_labels {
  46         struct ovs_key_ct_labels value;
  47         struct ovs_key_ct_labels mask;
  48 };
  49
  50 enum ovs_ct_nat {
  51         OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
  52         OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
  53         OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
  54 };
  55
  56 /* Conntrack action context for execution. */
  57 struct ovs_conntrack_info {
  58         struct nf_conntrack_helper *helper;
  59         struct nf_conntrack_zone zone;
  60         struct nf_conn *ct;
  61         u8 commit : 1;
  62         u8 nat : 3;                 /* enum ovs_ct_nat */
  63         u8 force : 1;
  64         u8 have_eventmask : 1;
  65         u16 family;
  66         u32 eventmask;              /* Mask of 1 << IPCT_*. */
  67         struct md_mark mark;
  68         struct md_labels labels;
  69         char timeout[CTNL_TIMEOUT_NAME_MAX];
  70         struct nf_ct_timeout *nf_ct_timeout;
  71 #if IS_ENABLED(CONFIG_NF_NAT)
  72         struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
  73 #endif
  74 };
  75
  76 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
  77 #define OVS_CT_LIMIT_UNLIMITED  0
  78 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
  79 #define CT_LIMIT_HASH_BUCKETS 512
  80 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
  81
  82 struct ovs_ct_limit {
  83         /* Elements in ovs_ct_limit_info->limits hash table */
  84         struct hlist_node hlist_node;
  85         struct rcu_head rcu;
  86         u16 zone;
  87         u32 limit;
  88 };
  89
  90 struct ovs_ct_limit_info {
  91         u32 default_limit;
  92         struct hlist_head *limits;
  93         struct nf_conncount_data *data;
  94 };
  95
  96 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
  97         [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
  98 };
  99 #endif
 100
 101 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
 102
 103 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
 104
 105 static u16 key_to_nfproto(const struct sw_flow_key *key)
 106 {
 107         switch (ntohs(key->eth.type)) {
 108         case ETH_P_IP:
 109                 return NFPROTO_IPV4;
 110         case ETH_P_IPV6:
 111                 return NFPROTO_IPV6;
 112         default:
 113                 return NFPROTO_UNSPEC;
 114         }
 115 }
 116
 117 /* Map SKB connection state into the values used by flow definition. */
 118 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
 119 {
 120         u8 ct_state = OVS_CS_F_TRACKED;
 121
 122         switch (ctinfo) {
 123         case IP_CT_ESTABLISHED_REPLY:
 124         case IP_CT_RELATED_REPLY:
 125                 ct_state |= OVS_CS_F_REPLY_DIR;
 126                 break;
 127         default:
 128                 break;
 129         }
 130
 131         switch (ctinfo) {
 132         case IP_CT_ESTABLISHED:
 133         case IP_CT_ESTABLISHED_REPLY:
 134                 ct_state |= OVS_CS_F_ESTABLISHED;
 135                 break;
 136         case IP_CT_RELATED:
 137         case IP_CT_RELATED_REPLY:
 138                 ct_state |= OVS_CS_F_RELATED;
 139                 break;
 140         case IP_CT_NEW:
 141                 ct_state |= OVS_CS_F_NEW;
 142                 break;
 143         default:
 144                 break;
 145         }
 146
 147         return ct_state;
 148 }
 149
 150 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
 151 {
 152 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 153         return ct ? ct->mark : 0;
 154 #else
 155         return 0;
 156 #endif
 157 }
 158
 159 /* Guard against conntrack labels max size shrinking below 128 bits. */
 160 #if NF_CT_LABELS_MAX_SIZE < 16
 161 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
 162 #endif
 163
 164 static void ovs_ct_get_labels(const struct nf_conn *ct,
 165                               struct ovs_key_ct_labels *labels)
 166 {
 167         struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
 168
 169         if (cl)
 170                 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
 171         else
 172                 memset(labels, 0, OVS_CT_LABELS_LEN);
 173 }
 174
 175 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
 176                                         const struct nf_conntrack_tuple *orig,
 177                                         u8 icmp_proto)
 178 {
 179         key->ct_orig_proto = orig->dst.protonum;
 180         if (orig->dst.protonum == icmp_proto) {
 181                 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
 182                 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
 183         } else {
 184                 key->ct.orig_tp.src = orig->src.u.all;
 185                 key->ct.orig_tp.dst = orig->dst.u.all;
 186         }
 187 }
 188
 189 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
 190                                 const struct nf_conntrack_zone *zone,
 191                                 const struct nf_conn *ct)
 192 {
 193         key->ct_state = state;
 194         key->ct_zone = zone->id;
 195         key->ct.mark = ovs_ct_get_mark(ct);
 196         ovs_ct_get_labels(ct, &key->ct.labels);
 197
 198         if (ct) {
 199                 const struct nf_conntrack_tuple *orig;
 200
 201                 /* Use the master if we have one. */
 202                 if (ct->master)
 203                         ct = ct->master;
 204                 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 205
 206                 /* IP version must match with the master connection. */
 207                 if (key->eth.type == htons(ETH_P_IP) &&
 208                     nf_ct_l3num(ct) == NFPROTO_IPV4) {
 209                         key->ipv4.ct_orig.src = orig->src.u3.ip;
 210                         key->ipv4.ct_orig.dst = orig->dst.u3.ip;
 211                         __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
 212                         return;
 213                 } else if (key->eth.type == htons(ETH_P_IPV6) &&
 214                            !sw_flow_key_is_nd(key) &&
 215                            nf_ct_l3num(ct) == NFPROTO_IPV6) {
 216                         key->ipv6.ct_orig.src = orig->src.u3.in6;
 217                         key->ipv6.ct_orig.dst = orig->dst.u3.in6;
 218                         __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
 219                         return;
 220                 }
 221         }
 222         /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
 223          * original direction key fields.
 224          */
 225         key->ct_orig_proto = 0;
 226 }
 227
 228 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
 229  * previously sent the packet to conntrack via the ct action.  If
 230  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
 231  * initialized from the connection status.
 232  */
 233 static void ovs_ct_update_key(const struct sk_buff *skb,
 234                               const struct ovs_conntrack_info *info,
 235                               struct sw_flow_key *key, bool post_ct,
 236                               bool keep_nat_flags)
 237 {
 238         const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
 239         enum ip_conntrack_info ctinfo;
 240         struct nf_conn *ct;
 241         u8 state = 0;
 242
 243         ct = nf_ct_get(skb, &ctinfo);
 244         if (ct) {
 245                 state = ovs_ct_get_state(ctinfo);
 246                 /* All unconfirmed entries are NEW connections. */
 247                 if (!nf_ct_is_confirmed(ct))
 248                         state |= OVS_CS_F_NEW;
 249                 /* OVS persists the related flag for the duration of the
 250                  * connection.
 251                  */
 252                 if (ct->master)
 253                         state |= OVS_CS_F_RELATED;
 254                 if (keep_nat_flags) {
 255                         state |= key->ct_state & OVS_CS_F_NAT_MASK;
 256                 } else {
 257                         if (ct->status & IPS_SRC_NAT)
 258                                 state |= OVS_CS_F_SRC_NAT;
 259                         if (ct->status & IPS_DST_NAT)
 260                                 state |= OVS_CS_F_DST_NAT;
 261                 }
 262                 zone = nf_ct_zone(ct);
 263         } else if (post_ct) {
 264                 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
 265                 if (info)
 266                         zone = &info->zone;
 267         }
 268         __ovs_ct_update_key(key, state, zone, ct);
 269 }
 270
 271 /* This is called to initialize CT key fields possibly coming in from the local
 272  * stack.
 273  */
 274 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
 275 {
 276         ovs_ct_update_key(skb, NULL, key, false, false);
 277 }
 278
 279 int ovs_ct_put_key(const struct sw_flow_key *swkey,
 280                    const struct sw_flow_key *output, struct sk_buff *skb)
 281 {
 282         if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
 283                 return -EMSGSIZE;
 284
 285         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
 286             nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
 287                 return -EMSGSIZE;
 288
 289         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
 290             nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
 291                 return -EMSGSIZE;
 292
 293         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
 294             nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
 295                     &output->ct.labels))
 296                 return -EMSGSIZE;
 297
 298         if (swkey->ct_orig_proto) {
 299                 if (swkey->eth.type == htons(ETH_P_IP)) {
 300                         struct ovs_key_ct_tuple_ipv4 orig;
 301
 302                         memset(&orig, 0, sizeof(orig));
 303                         orig.ipv4_src = output->ipv4.ct_orig.src;
 304                         orig.ipv4_dst = output->ipv4.ct_orig.dst;
 305                         orig.src_port = output->ct.orig_tp.src;
 306                         orig.dst_port = output->ct.orig_tp.dst;
 307                         orig.ipv4_proto = output->ct_orig_proto;
 308
 309                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
 310                                     sizeof(orig), &orig))
 311                                 return -EMSGSIZE;
 312                 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 313                         struct ovs_key_ct_tuple_ipv6 orig;
 314
 315                         memset(&orig, 0, sizeof(orig));
 316                         memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
 317                                sizeof(orig.ipv6_src));
 318                         memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
 319                                sizeof(orig.ipv6_dst));
 320                         orig.src_port = output->ct.orig_tp.src;
 321                         orig.dst_port = output->ct.orig_tp.dst;
 322                         orig.ipv6_proto = output->ct_orig_proto;
 323
 324                         if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
 325                                     sizeof(orig), &orig))
 326                                 return -EMSGSIZE;
 327                 }
 328         }
 329
 330         return 0;
 331 }
 332
 333 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
 334                            u32 ct_mark, u32 mask)
 335 {
 336 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
 337         u32 new_mark;
 338
 339         new_mark = ct_mark | (ct->mark & ~(mask));
 340         if (ct->mark != new_mark) {
 341                 ct->mark = new_mark;
 342                 if (nf_ct_is_confirmed(ct))
 343                         nf_conntrack_event_cache(IPCT_MARK, ct);
 344                 key->ct.mark = new_mark;
 345         }
 346
 347         return 0;
 348 #else
 349         return -ENOTSUPP;
 350 #endif
 351 }
 352
 353 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
 354 {
 355         struct nf_conn_labels *cl;
 356
 357         cl = nf_ct_labels_find(ct);
 358         if (!cl) {
 359                 nf_ct_labels_ext_add(ct);
 360                 cl = nf_ct_labels_find(ct);
 361         }
 362
 363         return cl;
 364 }
 365
 366 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
 367  * since the new connection is not yet confirmed, and thus no-one else has
 368  * access to it's labels, we simply write them over.
 369  */
 370 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
 371                               const struct ovs_key_ct_labels *labels,
 372                               const struct ovs_key_ct_labels *mask)
 373 {
 374         struct nf_conn_labels *cl, *master_cl;
 375         bool have_mask = labels_nonzero(mask);
 376
 377         /* Inherit master's labels to the related connection? */
 378         master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
 379
 380         if (!master_cl && !have_mask)
 381                 return 0;   /* Nothing to do. */
 382
 383         cl = ovs_ct_get_conn_labels(ct);
 384         if (!cl)
 385                 return -ENOSPC;
 386
 387         /* Inherit the master's labels, if any. */
 388         if (master_cl)
 389                 *cl = *master_cl;
 390
 391         if (have_mask) {
 392                 u32 *dst = (u32 *)cl->bits;
 393                 int i;
 394
 395                 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
 396                         dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
 397                                 (labels->ct_labels_32[i]
 398                                  & mask->ct_labels_32[i]);
 399         }
 400
 401         /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
 402          * IPCT_LABEL bit is set in the event cache.
 403          */
 404         nf_conntrack_event_cache(IPCT_LABEL, ct);
 405
 406         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 407
 408         return 0;
 409 }
 410
 411 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
 412                              const struct ovs_key_ct_labels *labels,
 413                              const struct ovs_key_ct_labels *mask)
 414 {
 415         struct nf_conn_labels *cl;
 416         int err;
 417
 418         cl = ovs_ct_get_conn_labels(ct);
 419         if (!cl)
 420                 return -ENOSPC;
 421
 422         err = nf_connlabels_replace(ct, labels->ct_labels_32,
 423                                     mask->ct_labels_32,
 424                                     OVS_CT_LABELS_LEN_32);
 425         if (err)
 426                 return err;
 427
 428         memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
 429
 430         return 0;
 431 }
 432
 433 /* 'skb' should already be pulled to nh_ofs. */
 434 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
 435 {
 436         const struct nf_conntrack_helper *helper;
 437         const struct nf_conn_help *help;
 438         enum ip_conntrack_info ctinfo;
 439         unsigned int protoff;
 440         struct nf_conn *ct;
 441         int err;
 442
 443         ct = nf_ct_get(skb, &ctinfo);
 444         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 445                 return NF_ACCEPT;
 446
 447         help = nfct_help(ct);
 448         if (!help)
 449                 return NF_ACCEPT;
 450
 451         helper = rcu_dereference(help->helper);
 452         if (!helper)
 453                 return NF_ACCEPT;
 454
 455         switch (proto) {
 456         case NFPROTO_IPV4:
 457                 protoff = ip_hdrlen(skb);
 458                 break;
 459         case NFPROTO_IPV6: {
 460                 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 461                 __be16 frag_off;
 462                 int ofs;
 463
 464                 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 465                                        &frag_off);
 466                 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
 467                         pr_debug("proto header not found\n");
 468                         return NF_ACCEPT;
 469                 }
 470                 protoff = ofs;
 471                 break;
 472         }
 473         default:
 474                 WARN_ONCE(1, "helper invoked on non-IP family!");
 475                 return NF_DROP;
 476         }
 477
 478         err = helper->help(skb, protoff, ct, ctinfo);
 479         if (err != NF_ACCEPT)
 480                 return err;
 481
 482         /* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
 483          * FTP with NAT) adusting the TCP payload size when mangling IP
 484          * addresses and/or port numbers in the text-based control connection.
 485          */
 486         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 487             !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
 488                 return NF_DROP;
 489         return NF_ACCEPT;
 490 }
 491
 492 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
 493  * value if 'skb' is freed.
 494  */
 495 static int handle_fragments(struct net *net, struct sw_flow_key *key,
 496                             u16 zone, struct sk_buff *skb)
 497 {
 498         struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
 499         int err;
 500
 501         if (key->eth.type == htons(ETH_P_IP)) {
 502                 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
 503
 504                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 505                 err = ip_defrag(net, skb, user);
 506                 if (err)
 507                         return err;
 508
 509                 ovs_cb.mru = IPCB(skb)->frag_max_size;
 510 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 511         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 512                 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
 513
 514                 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 515                 err = nf_ct_frag6_gather(net, skb, user);
 516                 if (err) {
 517                         if (err != -EINPROGRESS)
 518                                 kfree_skb(skb);
 519                         return err;
 520                 }
 521
 522                 key->ip.proto = ipv6_hdr(skb)->nexthdr;
 523                 ovs_cb.mru = IP6CB(skb)->frag_max_size;
 524 #endif
 525         } else {
 526                 kfree_skb(skb);
 527                 return -EPFNOSUPPORT;
 528         }
 529
 530         /* The key extracted from the fragment that completed this datagram
 531          * likely didn't have an L4 header, so regenerate it.
 532          */
 533         ovs_flow_key_update_l3l4(skb, key);
 534
 535         key->ip.frag = OVS_FRAG_TYPE_NONE;
 536         skb_clear_hash(skb);
 537         skb->ignore_df = 1;
 538         *OVS_CB(skb) = ovs_cb;
 539
 540         return 0;
 541 }
 542
 543 static struct nf_conntrack_expect *
 544 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
 545                    u16 proto, const struct sk_buff *skb)
 546 {
 547         struct nf_conntrack_tuple tuple;
 548         struct nf_conntrack_expect *exp;
 549
 550         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
 551                 return NULL;
 552
 553         exp = __nf_ct_expect_find(net, zone, &tuple);
 554         if (exp) {
 555                 struct nf_conntrack_tuple_hash *h;
 556
 557                 /* Delete existing conntrack entry, if it clashes with the
 558                  * expectation.  This can happen since conntrack ALGs do not
 559                  * check for clashes between (new) expectations and existing
 560                  * conntrack entries.  nf_conntrack_in() will check the
 561                  * expectations only if a conntrack entry can not be found,
 562                  * which can lead to OVS finding the expectation (here) in the
 563                  * init direction, but which will not be removed by the
 564                  * nf_conntrack_in() call, if a matching conntrack entry is
 565                  * found instead.  In this case all init direction packets
 566                  * would be reported as new related packets, while reply
 567                  * direction packets would be reported as un-related
 568                  * established packets.
 569                  */
 570                 h = nf_conntrack_find_get(net, zone, &tuple);
 571                 if (h) {
 572                         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 573
 574                         nf_ct_delete(ct, 0, 0);
 575                         nf_conntrack_put(&ct->ct_general);
 576                 }
 577         }
 578
 579         return exp;
 580 }
 581
 582 /* This replicates logic from nf_conntrack_core.c that is not exported. */
 583 static enum ip_conntrack_info
 584 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
 585 {
 586         const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 587
 588         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
 589                 return IP_CT_ESTABLISHED_REPLY;
 590         /* Once we've had two way comms, always ESTABLISHED. */
 591         if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 592                 return IP_CT_ESTABLISHED;
 593         if (test_bit(IPS_EXPECTED_BIT, &ct->status))
 594                 return IP_CT_RELATED;
 595         return IP_CT_NEW;
 596 }
 597
 598 /* Find an existing connection which this packet belongs to without
 599  * re-attributing statistics or modifying the connection state.  This allows an
 600  * skb->_nfct lost due to an upcall to be recovered during actions execution.
 601  *
 602  * Must be called with rcu_read_lock.
 603  *
 604  * On success, populates skb->_nfct and returns the connection.  Returns NULL
 605  * if there is no existing entry.
 606  */
 607 static struct nf_conn *
 608 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
 609                      u8 l3num, struct sk_buff *skb, bool natted)
 610 {
 611         struct nf_conntrack_tuple tuple;
 612         struct nf_conntrack_tuple_hash *h;
 613         struct nf_conn *ct;
 614
 615         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
 616                                net, &tuple)) {
 617                 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
 618                 return NULL;
 619         }
 620
 621         /* Must invert the tuple if skb has been transformed by NAT. */
 622         if (natted) {
 623                 struct nf_conntrack_tuple inverse;
 624
 625                 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
 626                         pr_debug("ovs_ct_find_existing: Inversion failed!\n");
 627                         return NULL;
 628                 }
 629                 tuple = inverse;
 630         }
 631
 632         /* look for tuple match */
 633         h = nf_conntrack_find_get(net, zone, &tuple);
 634         if (!h)
 635                 return NULL;   /* Not found. */
 636
 637         ct = nf_ct_tuplehash_to_ctrack(h);
 638
 639         /* Inverted packet tuple matches the reverse direction conntrack tuple,
 640          * select the other tuplehash to get the right 'ctinfo' bits for this
 641          * packet.
 642          */
 643         if (natted)
 644                 h = &ct->tuplehash[!h->tuple.dst.dir];
 645
 646         nf_ct_set(skb, ct, ovs_ct_get_info(h));
 647         return ct;
 648 }
 649
 650 static
 651 struct nf_conn *ovs_ct_executed(struct net *net,
 652                                 const struct sw_flow_key *key,
 653                                 const struct ovs_conntrack_info *info,
 654                                 struct sk_buff *skb,
 655                                 bool *ct_executed)
 656 {
 657         struct nf_conn *ct = NULL;
 658
 659         /* If no ct, check if we have evidence that an existing conntrack entry
 660          * might be found for this skb.  This happens when we lose a skb->_nfct
 661          * due to an upcall, or if the direction is being forced.  If the
 662          * connection was not confirmed, it is not cached and needs to be run
 663          * through conntrack again.
 664          */
 665         *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
 666                        !(key->ct_state & OVS_CS_F_INVALID) &&
 667                        (key->ct_zone == info->zone.id);
 668
 669         if (*ct_executed || (!key->ct_state && info->force)) {
 670                 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
 671                                           !!(key->ct_state &
 672                                           OVS_CS_F_NAT_MASK));
 673         }
 674
 675         return ct;
 676 }
 677
 678 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
 679 static bool skb_nfct_cached(struct net *net,
 680                             const struct sw_flow_key *key,
 681                             const struct ovs_conntrack_info *info,
 682                             struct sk_buff *skb)
 683 {
 684         enum ip_conntrack_info ctinfo;
 685         struct nf_conn *ct;
 686         bool ct_executed = true;
 687
 688         ct = nf_ct_get(skb, &ctinfo);
 689         if (!ct)
 690                 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
 691
 692         if (ct)
 693                 nf_ct_get(skb, &ctinfo);
 694         else
 695                 return false;
 696
 697         if (!net_eq(net, read_pnet(&ct->ct_net)))
 698                 return false;
 699         if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
 700                 return false;
 701         if (info->helper) {
 702                 struct nf_conn_help *help;
 703
 704                 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
 705                 if (help && rcu_access_pointer(help->helper) != info->helper)
 706                         return false;
 707         }
 708         if (info->nf_ct_timeout) {
 709                 struct nf_conn_timeout *timeout_ext;
 710
 711                 timeout_ext = nf_ct_timeout_find(ct);
 712                 if (!timeout_ext || info->nf_ct_timeout !=
 713                     rcu_dereference(timeout_ext->timeout))
 714                         return false;
 715         }
 716         /* Force conntrack entry direction to the current packet? */
 717         if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
 718                 /* Delete the conntrack entry if confirmed, else just release
 719                  * the reference.
 720                  */
 721                 if (nf_ct_is_confirmed(ct))
 722                         nf_ct_delete(ct, 0, 0);
 723
 724                 nf_conntrack_put(&ct->ct_general);
 725                 nf_ct_set(skb, NULL, 0);
 726                 return false;
 727         }
 728
 729         return ct_executed;
 730 }
 731
 732 #if IS_ENABLED(CONFIG_NF_NAT)
 733 /* Modelled after nf_nat_ipv[46]_fn().
 734  * range is only used for new, uninitialized NAT state.
 735  * Returns either NF_ACCEPT or NF_DROP.
 736  */
 737 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
 738                               enum ip_conntrack_info ctinfo,
 739                               const struct nf_nat_range2 *range,
 740                               enum nf_nat_manip_type maniptype)
 741 {
 742         int hooknum, nh_off, err = NF_ACCEPT;
 743
 744         nh_off = skb_network_offset(skb);
 745         skb_pull_rcsum(skb, nh_off);
 746
 747         /* See HOOK2MANIP(). */
 748         if (maniptype == NF_NAT_MANIP_SRC)
 749                 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
 750         else
 751                 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
 752
 753         switch (ctinfo) {
 754         case IP_CT_RELATED:
 755         case IP_CT_RELATED_REPLY:
 756                 if (IS_ENABLED(CONFIG_NF_NAT) &&
 757                     skb->protocol == htons(ETH_P_IP) &&
 758                     ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 759                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 760                                                            hooknum))
 761                                 err = NF_DROP;
 762                         goto push;
 763                 } else if (IS_ENABLED(CONFIG_IPV6) &&
 764                            skb->protocol == htons(ETH_P_IPV6)) {
 765                         __be16 frag_off;
 766                         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 767                         int hdrlen = ipv6_skip_exthdr(skb,
 768                                                       sizeof(struct ipv6hdr),
 769                                                       &nexthdr, &frag_off);
 770
 771                         if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 772                                 if (!nf_nat_icmpv6_reply_translation(skb, ct,
 773                                                                      ctinfo,
 774                                                                      hooknum,
 775                                                                      hdrlen))
 776                                         err = NF_DROP;
 777                                 goto push;
 778                         }
 779                 }
 780                 /* Non-ICMP, fall thru to initialize if needed. */
 781                 fallthrough;
 782         case IP_CT_NEW:
 783                 /* Seen it before?  This can happen for loopback, retrans,
 784                  * or local packets.
 785                  */
 786                 if (!nf_nat_initialized(ct, maniptype)) {
 787                         /* Initialize according to the NAT action. */
 788                         err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
 789                                 /* Action is set up to establish a new
 790                                  * mapping.
 791                                  */
 792                                 ? nf_nat_setup_info(ct, range, maniptype)
 793                                 : nf_nat_alloc_null_binding(ct, hooknum);
 794                         if (err != NF_ACCEPT)
 795                                 goto push;
 796                 }
 797                 break;
 798
 799         case IP_CT_ESTABLISHED:
 800         case IP_CT_ESTABLISHED_REPLY:
 801                 break;
 802
 803         default:
 804                 err = NF_DROP;
 805                 goto push;
 806         }
 807
 808         err = nf_nat_packet(ct, ctinfo, hooknum, skb);
 809 push:
 810         skb_push(skb, nh_off);
 811         skb_postpush_rcsum(skb, skb->data, nh_off);
 812
 813         return err;
 814 }
 815
 816 static void ovs_nat_update_key(struct sw_flow_key *key,
 817                                const struct sk_buff *skb,
 818                                enum nf_nat_manip_type maniptype)
 819 {
 820         if (maniptype == NF_NAT_MANIP_SRC) {
 821                 __be16 src;
 822
 823                 key->ct_state |= OVS_CS_F_SRC_NAT;
 824                 if (key->eth.type == htons(ETH_P_IP))
 825                         key->ipv4.addr.src = ip_hdr(skb)->saddr;
 826                 else if (key->eth.type == htons(ETH_P_IPV6))
 827                         memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
 828                                sizeof(key->ipv6.addr.src));
 829                 else
 830                         return;
 831
 832                 if (key->ip.proto == IPPROTO_UDP)
 833                         src = udp_hdr(skb)->source;
 834                 else if (key->ip.proto == IPPROTO_TCP)
 835                         src = tcp_hdr(skb)->source;
 836                 else if (key->ip.proto == IPPROTO_SCTP)
 837                         src = sctp_hdr(skb)->source;
 838                 else
 839                         return;
 840
 841                 key->tp.src = src;
 842         } else {
 843                 __be16 dst;
 844
 845                 key->ct_state |= OVS_CS_F_DST_NAT;
 846                 if (key->eth.type == htons(ETH_P_IP))
 847                         key->ipv4.addr.dst = ip_hdr(skb)->daddr;
 848                 else if (key->eth.type == htons(ETH_P_IPV6))
 849                         memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
 850                                sizeof(key->ipv6.addr.dst));
 851                 else
 852                         return;
 853
 854                 if (key->ip.proto == IPPROTO_UDP)
 855                         dst = udp_hdr(skb)->dest;
 856                 else if (key->ip.proto == IPPROTO_TCP)
 857                         dst = tcp_hdr(skb)->dest;
 858                 else if (key->ip.proto == IPPROTO_SCTP)
 859                         dst = sctp_hdr(skb)->dest;
 860                 else
 861                         return;
 862
 863                 key->tp.dst = dst;
 864         }
 865 }
 866
 867 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
 868 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 869                       const struct ovs_conntrack_info *info,
 870                       struct sk_buff *skb, struct nf_conn *ct,
 871                       enum ip_conntrack_info ctinfo)
 872 {
 873         enum nf_nat_manip_type maniptype;
 874         int err;
 875
 876         /* Add NAT extension if not confirmed yet. */
 877         if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
 878                 return NF_ACCEPT;   /* Can't NAT. */
 879
 880         /* Determine NAT type.
 881          * Check if the NAT type can be deduced from the tracked connection.
 882          * Make sure new expected connections (IP_CT_RELATED) are NATted only
 883          * when committing.
 884          */
 885         if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
 886             ct->status & IPS_NAT_MASK &&
 887             (ctinfo != IP_CT_RELATED || info->commit)) {
 888                 /* NAT an established or related connection like before. */
 889                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
 890                         /* This is the REPLY direction for a connection
 891                          * for which NAT was applied in the forward
 892                          * direction.  Do the reverse NAT.
 893                          */
 894                         maniptype = ct->status & IPS_SRC_NAT
 895                                 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
 896                 else
 897                         maniptype = ct->status & IPS_SRC_NAT
 898                                 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
 899         } else if (info->nat & OVS_CT_SRC_NAT) {
 900                 maniptype = NF_NAT_MANIP_SRC;
 901         } else if (info->nat & OVS_CT_DST_NAT) {
 902                 maniptype = NF_NAT_MANIP_DST;
 903         } else {
 904                 return NF_ACCEPT; /* Connection is not NATed. */
 905         }
 906         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
 907
 908         if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
 909                 if (ct->status & IPS_SRC_NAT) {
 910                         if (maniptype == NF_NAT_MANIP_SRC)
 911                                 maniptype = NF_NAT_MANIP_DST;
 912                         else
 913                                 maniptype = NF_NAT_MANIP_SRC;
 914
 915                         err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
 916                                                  maniptype);
 917                 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 918                         err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
 919                                                  NF_NAT_MANIP_SRC);
 920                 }
 921         }
 922
 923         /* Mark NAT done if successful and update the flow key. */
 924         if (err == NF_ACCEPT)
 925                 ovs_nat_update_key(key, skb, maniptype);
 926
 927         return err;
 928 }
 929 #else /* !CONFIG_NF_NAT */
 930 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
 931                       const struct ovs_conntrack_info *info,
 932                       struct sk_buff *skb, struct nf_conn *ct,
 933                       enum ip_conntrack_info ctinfo)
 934 {
 935         return NF_ACCEPT;
 936 }
 937 #endif
 938
 939 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
 940  * not done already.  Update key with new CT state after passing the packet
 941  * through conntrack.
 942  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
 943  * set to NULL and 0 will be returned.
 944  */
 945 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
 946                            const struct ovs_conntrack_info *info,
 947                            struct sk_buff *skb)
 948 {
 949         /* If we are recirculating packets to match on conntrack fields and
 950          * committing with a separate conntrack action,  then we don't need to
 951          * actually run the packet through conntrack twice unless it's for a
 952          * different zone.
 953          */
 954         bool cached = skb_nfct_cached(net, key, info, skb);
 955         enum ip_conntrack_info ctinfo;
 956         struct nf_conn *ct;
 957
 958         if (!cached) {
 959                 struct nf_hook_state state = {
 960                         .hook = NF_INET_PRE_ROUTING,
 961                         .pf = info->family,
 962                         .net = net,
 963                 };
 964                 struct nf_conn *tmpl = info->ct;
 965                 int err;
 966
 967                 /* Associate skb with specified zone. */
 968                 if (tmpl) {
 969                         if (skb_nfct(skb))
 970                                 nf_conntrack_put(skb_nfct(skb));
 971                         nf_conntrack_get(&tmpl->ct_general);
 972                         nf_ct_set(skb, tmpl, IP_CT_NEW);
 973                 }
 974
 975                 err = nf_conntrack_in(skb, &state);
 976                 if (err != NF_ACCEPT)
 977                         return -ENOENT;
 978
 979                 /* Clear CT state NAT flags to mark that we have not yet done
 980                  * NAT after the nf_conntrack_in() call.  We can actually clear
 981                  * the whole state, as it will be re-initialized below.
 982                  */
 983                 key->ct_state = 0;
 984
 985                 /* Update the key, but keep the NAT flags. */
 986                 ovs_ct_update_key(skb, info, key, true, true);
 987         }
 988
 989         ct = nf_ct_get(skb, &ctinfo);
 990         if (ct) {
 991                 bool add_helper = false;
 992
 993                 /* Packets starting a new connection must be NATted before the
 994                  * helper, so that the helper knows about the NAT.  We enforce
 995                  * this by delaying both NAT and helper calls for unconfirmed
 996                  * connections until the committing CT action.  For later
 997                  * packets NAT and Helper may be called in either order.
 998                  *
 999                  * NAT will be done only if the CT action has NAT, and only
1000                  * once per packet (per zone), as guarded by the NAT bits in
1001                  * the key->ct_state.
1002                  */
1003                 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1004                     (nf_ct_is_confirmed(ct) || info->commit) &&
1005                     ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1006                         return -EINVAL;
1007                 }
1008
1009                 /* Userspace may decide to perform a ct lookup without a helper
1010                  * specified followed by a (recirculate and) commit with one,
1011                  * or attach a helper in a later commit.  Therefore, for
1012                  * connections which we will commit, we may need to attach
1013                  * the helper here.
1014                  */
1015                 if (info->commit && info->helper && !nfct_help(ct)) {
1016                         int err = __nf_ct_try_assign_helper(ct, info->ct,
1017                                                             GFP_ATOMIC);
1018                         if (err)
1019                                 return err;
1020                         add_helper = true;
1021
1022                         /* helper installed, add seqadj if NAT is required */
1023                         if (info->nat && !nfct_seqadj(ct)) {
1024                                 if (!nfct_seqadj_ext_add(ct))
1025                                         return -EINVAL;
1026                         }
1027                 }
1028
1029                 /* Call the helper only if:
1030                  * - nf_conntrack_in() was executed above ("!cached") or a
1031                  *   helper was just attached ("add_helper") for a confirmed
1032                  *   connection, or
1033                  * - When committing an unconfirmed connection.
1034                  */
1035                 if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1036                                               info->commit) &&
1037                     ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1038                         return -EINVAL;
1039                 }
1040
1041                 if (nf_ct_protonum(ct) == IPPROTO_TCP &&
1042                     nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
1043                         /* Be liberal for tcp packets so that out-of-window
1044                          * packets are not marked invalid.
1045                          */
1046                         nf_ct_set_tcp_be_liberal(ct);
1047                 }
1048         }
1049
1050         return 0;
1051 }
1052
1053 /* Lookup connection and read fields into key. */
1054 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1055                          const struct ovs_conntrack_info *info,
1056                          struct sk_buff *skb)
1057 {
1058         struct nf_conntrack_expect *exp;
1059
1060         /* If we pass an expected packet through nf_conntrack_in() the
1061          * expectation is typically removed, but the packet could still be
1062          * lost in upcall processing.  To prevent this from happening we
1063          * perform an explicit expectation lookup.  Expected connections are
1064          * always new, and will be passed through conntrack only when they are
1065          * committed, as it is OK to remove the expectation at that time.
1066          */
1067         exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1068         if (exp) {
1069                 u8 state;
1070
1071                 /* NOTE: New connections are NATted and Helped only when
1072                  * committed, so we are not calling into NAT here.
1073                  */
1074                 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1075                 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1076         } else {
1077                 struct nf_conn *ct;
1078                 int err;
1079
1080                 err = __ovs_ct_lookup(net, key, info, skb);
1081                 if (err)
1082                         return err;
1083
1084                 ct = (struct nf_conn *)skb_nfct(skb);
1085                 if (ct)
1086                         nf_ct_deliver_cached_events(ct);
1087         }
1088
1089         return 0;
1090 }
1091
1092 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1093 {
1094         size_t i;
1095
1096         for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1097                 if (labels->ct_labels_32[i])
1098                         return true;
1099
1100         return false;
1101 }
1102
1103 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1104 static struct hlist_head *ct_limit_hash_bucket(
1105         const struct ovs_ct_limit_info *info, u16 zone)
1106 {
1107         return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1108 }
1109
1110 /* Call with ovs_mutex */
1111 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1112                          struct ovs_ct_limit *new_ct_limit)
1113 {
1114         struct ovs_ct_limit *ct_limit;
1115         struct hlist_head *head;
1116
1117         head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1118         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1119                 if (ct_limit->zone == new_ct_limit->zone) {
1120                         hlist_replace_rcu(&ct_limit->hlist_node,
1121                                           &new_ct_limit->hlist_node);
1122                         kfree_rcu(ct_limit, rcu);
1123                         return;
1124                 }
1125         }
1126
1127         hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1128 }
1129
1130 /* Call with ovs_mutex */
1131 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1132 {
1133         struct ovs_ct_limit *ct_limit;
1134         struct hlist_head *head;
1135         struct hlist_node *n;
1136
1137         head = ct_limit_hash_bucket(info, zone);
1138         hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1139                 if (ct_limit->zone == zone) {
1140                         hlist_del_rcu(&ct_limit->hlist_node);
1141                         kfree_rcu(ct_limit, rcu);
1142                         return;
1143                 }
1144         }
1145 }
1146
1147 /* Call with RCU read lock */
1148 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1149 {
1150         struct ovs_ct_limit *ct_limit;
1151         struct hlist_head *head;
1152
1153         head = ct_limit_hash_bucket(info, zone);
1154         hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1155                 if (ct_limit->zone == zone)
1156                         return ct_limit->limit;
1157         }
1158
1159         return info->default_limit;
1160 }
1161
1162 static int ovs_ct_check_limit(struct net *net,
1163                               const struct ovs_conntrack_info *info,
1164                               const struct nf_conntrack_tuple *tuple)
1165 {
1166         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1167         const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1168         u32 per_zone_limit, connections;
1169         u32 conncount_key;
1170
1171         conncount_key = info->zone.id;
1172
1173         per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1174         if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1175                 return 0;
1176
1177         connections = nf_conncount_count(net, ct_limit_info->data,
1178                                          &conncount_key, tuple, &info->zone);
1179         if (connections > per_zone_limit)
1180                 return -ENOMEM;
1181
1182         return 0;
1183 }
1184 #endif
1185
1186 /* Lookup connection and confirm if unconfirmed. */
1187 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1188                          const struct ovs_conntrack_info *info,
1189                          struct sk_buff *skb)
1190 {
1191         enum ip_conntrack_info ctinfo;
1192         struct nf_conn *ct;
1193         int err;
1194
1195         err = __ovs_ct_lookup(net, key, info, skb);
1196         if (err)
1197                 return err;
1198
1199         /* The connection could be invalid, in which case this is a no-op.*/
1200         ct = nf_ct_get(skb, &ctinfo);
1201         if (!ct)
1202                 return 0;
1203
1204 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1205         if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1206                 if (!nf_ct_is_confirmed(ct)) {
1207                         err = ovs_ct_check_limit(net, info,
1208                                 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1209                         if (err) {
1210                                 net_warn_ratelimited("openvswitch: zone: %u "
1211                                         "exceeds conntrack limit\n",
1212                                         info->zone.id);
1213                                 return err;
1214                         }
1215                 }
1216         }
1217 #endif
1218
1219         /* Set the conntrack event mask if given.  NEW and DELETE events have
1220          * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1221          * typically would receive many kinds of updates.  Setting the event
1222          * mask allows those events to be filtered.  The set event mask will
1223          * remain in effect for the lifetime of the connection unless changed
1224          * by a further CT action with both the commit flag and the eventmask
1225          * option. */
1226         if (info->have_eventmask) {
1227                 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1228
1229                 if (cache)
1230                         cache->ctmask = info->eventmask;
1231         }
1232
1233         /* Apply changes before confirming the connection so that the initial
1234          * conntrack NEW netlink event carries the values given in the CT
1235          * action.
1236          */
1237         if (info->mark.mask) {
1238                 err = ovs_ct_set_mark(ct, key, info->mark.value,
1239                                       info->mark.mask);
1240                 if (err)
1241                         return err;
1242         }
1243         if (!nf_ct_is_confirmed(ct)) {
1244                 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1245                                          &info->labels.mask);
1246                 if (err)
1247                         return err;
1248         } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1249                    labels_nonzero(&info->labels.mask)) {
1250                 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1251                                         &info->labels.mask);
1252                 if (err)
1253                         return err;
1254         }
1255         /* This will take care of sending queued events even if the connection
1256          * is already confirmed.
1257          */
1258         if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1259                 return -EINVAL;
1260
1261         return 0;
1262 }
1263
1264 /* Trim the skb to the length specified by the IP/IPv6 header,
1265  * removing any trailing lower-layer padding. This prepares the skb
1266  * for higher-layer processing that assumes skb->len excludes padding
1267  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1268  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1269  */
1270 static int ovs_skb_network_trim(struct sk_buff *skb)
1271 {
1272         unsigned int len;
1273         int err;
1274
1275         switch (skb->protocol) {
1276         case htons(ETH_P_IP):
1277                 len = ntohs(ip_hdr(skb)->tot_len);
1278                 break;
1279         case htons(ETH_P_IPV6):
1280                 len = sizeof(struct ipv6hdr)
1281                         + ntohs(ipv6_hdr(skb)->payload_len);
1282                 break;
1283         default:
1284                 len = skb->len;
1285         }
1286
1287         err = pskb_trim_rcsum(skb, len);
1288         if (err)
1289                 kfree_skb(skb);
1290
1291         return err;
1292 }
1293
1294 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1295  * value if 'skb' is freed.
1296  */
1297 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1298                    struct sw_flow_key *key,
1299                    const struct ovs_conntrack_info *info)
1300 {
1301         int nh_ofs;
1302         int err;
1303
1304         /* The conntrack module expects to be working at L3. */
1305         nh_ofs = skb_network_offset(skb);
1306         skb_pull_rcsum(skb, nh_ofs);
1307
1308         err = ovs_skb_network_trim(skb);
1309         if (err)
1310                 return err;
1311
1312         if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1313                 err = handle_fragments(net, key, info->zone.id, skb);
1314                 if (err)
1315                         return err;
1316         }
1317
1318         if (info->commit)
1319                 err = ovs_ct_commit(net, key, info, skb);
1320         else
1321                 err = ovs_ct_lookup(net, key, info, skb);
1322
1323         skb_push(skb, nh_ofs);
1324         skb_postpush_rcsum(skb, skb->data, nh_ofs);
1325         if (err)
1326                 kfree_skb(skb);
1327         return err;
1328 }
1329
1330 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1331 {
1332         if (skb_nfct(skb)) {
1333                 nf_conntrack_put(skb_nfct(skb));
1334                 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1335                 ovs_ct_fill_key(skb, key);
1336         }
1337
1338         return 0;
1339 }
1340
1341 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1342                              const struct sw_flow_key *key, bool log)
1343 {
1344         struct nf_conntrack_helper *helper;
1345         struct nf_conn_help *help;
1346         int ret = 0;
1347
1348         helper = nf_conntrack_helper_try_module_get(name, info->family,
1349                                                     key->ip.proto);
1350         if (!helper) {
1351                 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1352                 return -EINVAL;
1353         }
1354
1355         help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1356         if (!help) {
1357                 nf_conntrack_helper_put(helper);
1358                 return -ENOMEM;
1359         }
1360
1361 #if IS_ENABLED(CONFIG_NF_NAT)
1362         if (info->nat) {
1363                 ret = nf_nat_helper_try_module_get(name, info->family,
1364                                                    key->ip.proto);
1365                 if (ret) {
1366                         nf_conntrack_helper_put(helper);
1367                         OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1368                                   name, ret);
1369                         return ret;
1370                 }
1371         }
1372 #endif
1373         rcu_assign_pointer(help->helper, helper);
1374         info->helper = helper;
1375         return ret;
1376 }
1377
1378 #if IS_ENABLED(CONFIG_NF_NAT)
1379 static int parse_nat(const struct nlattr *attr,
1380                      struct ovs_conntrack_info *info, bool log)
1381 {
1382         struct nlattr *a;
1383         int rem;
1384         bool have_ip_max = false;
1385         bool have_proto_max = false;
1386         bool ip_vers = (info->family == NFPROTO_IPV6);
1387
1388         nla_for_each_nested(a, attr, rem) {
1389                 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1390                         [OVS_NAT_ATTR_SRC] = {0, 0},
1391                         [OVS_NAT_ATTR_DST] = {0, 0},
1392                         [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1393                                                  sizeof(struct in6_addr)},
1394                         [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1395                                                  sizeof(struct in6_addr)},
1396                         [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1397                         [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1398                         [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1399                         [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1400                         [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1401                 };
1402                 int type = nla_type(a);
1403
1404                 if (type > OVS_NAT_ATTR_MAX) {
1405                         OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1406                                   type, OVS_NAT_ATTR_MAX);
1407                         return -EINVAL;
1408                 }
1409
1410                 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1411                         OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1412                                   type, nla_len(a),
1413                                   ovs_nat_attr_lens[type][ip_vers]);
1414                         return -EINVAL;
1415                 }
1416
1417                 switch (type) {
1418                 case OVS_NAT_ATTR_SRC:
1419                 case OVS_NAT_ATTR_DST:
1420                         if (info->nat) {
1421                                 OVS_NLERR(log, "Only one type of NAT may be specified");
1422                                 return -ERANGE;
1423                         }
1424                         info->nat |= OVS_CT_NAT;
1425                         info->nat |= ((type == OVS_NAT_ATTR_SRC)
1426                                         ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1427                         break;
1428
1429                 case OVS_NAT_ATTR_IP_MIN:
1430                         nla_memcpy(&info->range.min_addr, a,
1431                                    sizeof(info->range.min_addr));
1432                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1433                         break;
1434
1435                 case OVS_NAT_ATTR_IP_MAX:
1436                         have_ip_max = true;
1437                         nla_memcpy(&info->range.max_addr, a,
1438                                    sizeof(info->range.max_addr));
1439                         info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1440                         break;
1441
1442                 case OVS_NAT_ATTR_PROTO_MIN:
1443                         info->range.min_proto.all = htons(nla_get_u16(a));
1444                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1445                         break;
1446
1447                 case OVS_NAT_ATTR_PROTO_MAX:
1448                         have_proto_max = true;
1449                         info->range.max_proto.all = htons(nla_get_u16(a));
1450                         info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1451                         break;
1452
1453                 case OVS_NAT_ATTR_PERSISTENT:
1454                         info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1455                         break;
1456
1457                 case OVS_NAT_ATTR_PROTO_HASH:
1458                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1459                         break;
1460
1461                 case OVS_NAT_ATTR_PROTO_RANDOM:
1462                         info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1463                         break;
1464
1465                 default:
1466                         OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1467                         return -EINVAL;
1468                 }
1469         }
1470
1471         if (rem > 0) {
1472                 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1473                 return -EINVAL;
1474         }
1475         if (!info->nat) {
1476                 /* Do not allow flags if no type is given. */
1477                 if (info->range.flags) {
1478                         OVS_NLERR(log,
1479                                   "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1480                                   );
1481                         return -EINVAL;
1482                 }
1483                 info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1484         } else if (!info->commit) {
1485                 OVS_NLERR(log,
1486                           "NAT attributes may be specified only when CT COMMIT flag is also specified."
1487                           );
1488                 return -EINVAL;
1489         }
1490         /* Allow missing IP_MAX. */
1491         if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1492                 memcpy(&info->range.max_addr, &info->range.min_addr,
1493                        sizeof(info->range.max_addr));
1494         }
1495         /* Allow missing PROTO_MAX. */
1496         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1497             !have_proto_max) {
1498                 info->range.max_proto.all = info->range.min_proto.all;
1499         }
1500         return 0;
1501 }
1502 #endif
1503
1504 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1505         [OVS_CT_ATTR_COMMIT]    = { .minlen = 0, .maxlen = 0 },
1506         [OVS_CT_ATTR_FORCE_COMMIT]      = { .minlen = 0, .maxlen = 0 },
1507         [OVS_CT_ATTR_ZONE]      = { .minlen = sizeof(u16),
1508                                     .maxlen = sizeof(u16) },
1509         [OVS_CT_ATTR_MARK]      = { .minlen = sizeof(struct md_mark),
1510                                     .maxlen = sizeof(struct md_mark) },
1511         [OVS_CT_ATTR_LABELS]    = { .minlen = sizeof(struct md_labels),
1512                                     .maxlen = sizeof(struct md_labels) },
1513         [OVS_CT_ATTR_HELPER]    = { .minlen = 1,
1514                                     .maxlen = NF_CT_HELPER_NAME_LEN },
1515 #if IS_ENABLED(CONFIG_NF_NAT)
1516         /* NAT length is checked when parsing the nested attributes. */
1517         [OVS_CT_ATTR_NAT]       = { .minlen = 0, .maxlen = INT_MAX },
1518 #endif
1519         [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1520                                     .maxlen = sizeof(u32) },
1521         [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1522                                   .maxlen = CTNL_TIMEOUT_NAME_MAX },
1523 };
1524
1525 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1526                     const char **helper, bool log)
1527 {
1528         struct nlattr *a;
1529         int rem;
1530
1531         nla_for_each_nested(a, attr, rem) {
1532                 int type = nla_type(a);
1533                 int maxlen;
1534                 int minlen;
1535
1536                 if (type > OVS_CT_ATTR_MAX) {
1537                         OVS_NLERR(log,
1538                                   "Unknown conntrack attr (type=%d, max=%d)",
1539                                   type, OVS_CT_ATTR_MAX);
1540                         return -EINVAL;
1541                 }
1542
1543                 maxlen = ovs_ct_attr_lens[type].maxlen;
1544                 minlen = ovs_ct_attr_lens[type].minlen;
1545                 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1546                         OVS_NLERR(log,
1547                                   "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1548                                   type, nla_len(a), maxlen);
1549                         return -EINVAL;
1550                 }
1551
1552                 switch (type) {
1553                 case OVS_CT_ATTR_FORCE_COMMIT:
1554                         info->force = true;
1555                         fallthrough;
1556                 case OVS_CT_ATTR_COMMIT:
1557                         info->commit = true;
1558                         break;
1559 #ifdef CONFIG_NF_CONNTRACK_ZONES
1560                 case OVS_CT_ATTR_ZONE:
1561                         info->zone.id = nla_get_u16(a);
1562                         break;
1563 #endif
1564 #ifdef CONFIG_NF_CONNTRACK_MARK
1565                 case OVS_CT_ATTR_MARK: {
1566                         struct md_mark *mark = nla_data(a);
1567
1568                         if (!mark->mask) {
1569                                 OVS_NLERR(log, "ct_mark mask cannot be 0");
1570                                 return -EINVAL;
1571                         }
1572                         info->mark = *mark;
1573                         break;
1574                 }
1575 #endif
1576 #ifdef CONFIG_NF_CONNTRACK_LABELS
1577                 case OVS_CT_ATTR_LABELS: {
1578                         struct md_labels *labels = nla_data(a);
1579
1580                         if (!labels_nonzero(&labels->mask)) {
1581                                 OVS_NLERR(log, "ct_labels mask cannot be 0");
1582                                 return -EINVAL;
1583                         }
1584                         info->labels = *labels;
1585                         break;
1586                 }
1587 #endif
1588                 case OVS_CT_ATTR_HELPER:
1589                         *helper = nla_data(a);
1590                         if (!memchr(*helper, '\0', nla_len(a))) {
1591                                 OVS_NLERR(log, "Invalid conntrack helper");
1592                                 return -EINVAL;
1593                         }
1594                         break;
1595 #if IS_ENABLED(CONFIG_NF_NAT)
1596                 case OVS_CT_ATTR_NAT: {
1597                         int err = parse_nat(a, info, log);
1598
1599                         if (err)
1600                                 return err;
1601                         break;
1602                 }
1603 #endif
1604                 case OVS_CT_ATTR_EVENTMASK:
1605                         info->have_eventmask = true;
1606                         info->eventmask = nla_get_u32(a);
1607                         break;
1608 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1609                 case OVS_CT_ATTR_TIMEOUT:
1610                         memcpy(info->timeout, nla_data(a), nla_len(a));
1611                         if (!memchr(info->timeout, '\0', nla_len(a))) {
1612                                 OVS_NLERR(log, "Invalid conntrack timeout");
1613                                 return -EINVAL;
1614                         }
1615                         break;
1616 #endif
1617
1618                 default:
1619                         OVS_NLERR(log, "Unknown conntrack attr (%d)",
1620                                   type);
1621                         return -EINVAL;
1622                 }
1623         }
1624
1625 #ifdef CONFIG_NF_CONNTRACK_MARK
1626         if (!info->commit && info->mark.mask) {
1627                 OVS_NLERR(log,
1628                           "Setting conntrack mark requires 'commit' flag.");
1629                 return -EINVAL;
1630         }
1631 #endif
1632 #ifdef CONFIG_NF_CONNTRACK_LABELS
1633         if (!info->commit && labels_nonzero(&info->labels.mask)) {
1634                 OVS_NLERR(log,
1635                           "Setting conntrack labels requires 'commit' flag.");
1636                 return -EINVAL;
1637         }
1638 #endif
1639         if (rem > 0) {
1640                 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1641                 return -EINVAL;
1642         }
1643
1644         return 0;
1645 }
1646
1647 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1648 {
1649         if (attr == OVS_KEY_ATTR_CT_STATE)
1650                 return true;
1651         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1652             attr == OVS_KEY_ATTR_CT_ZONE)
1653                 return true;
1654         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1655             attr == OVS_KEY_ATTR_CT_MARK)
1656                 return true;
1657         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1658             attr == OVS_KEY_ATTR_CT_LABELS) {
1659                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1660
1661                 return ovs_net->xt_label;
1662         }
1663
1664         return false;
1665 }
1666
1667 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1668                        const struct sw_flow_key *key,
1669                        struct sw_flow_actions **sfa,  bool log)
1670 {
1671         struct ovs_conntrack_info ct_info;
1672         const char *helper = NULL;
1673         u16 family;
1674         int err;
1675
1676         family = key_to_nfproto(key);
1677         if (family == NFPROTO_UNSPEC) {
1678                 OVS_NLERR(log, "ct family unspecified");
1679                 return -EINVAL;
1680         }
1681
1682         memset(&ct_info, 0, sizeof(ct_info));
1683         ct_info.family = family;
1684
1685         nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1686                         NF_CT_DEFAULT_ZONE_DIR, 0);
1687
1688         err = parse_ct(attr, &ct_info, &helper, log);
1689         if (err)
1690                 return err;
1691
1692         /* Set up template for tracking connections in specific zones. */
1693         ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1694         if (!ct_info.ct) {
1695                 OVS_NLERR(log, "Failed to allocate conntrack template");
1696                 return -ENOMEM;
1697         }
1698
1699         if (ct_info.timeout[0]) {
1700                 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1701                                       ct_info.timeout))
1702                         pr_info_ratelimited("Failed to associated timeout "
1703                                             "policy `%s'\n", ct_info.timeout);
1704                 else
1705                         ct_info.nf_ct_timeout = rcu_dereference(
1706                                 nf_ct_timeout_find(ct_info.ct)->timeout);
1707
1708         }
1709
1710         if (helper) {
1711                 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1712                 if (err)
1713                         goto err_free_ct;
1714         }
1715
1716         err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1717                                  sizeof(ct_info), log);
1718         if (err)
1719                 goto err_free_ct;
1720
1721         __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1722         nf_conntrack_get(&ct_info.ct->ct_general);
1723         return 0;
1724 err_free_ct:
1725         __ovs_ct_free_action(&ct_info);
1726         return err;
1727 }
1728
1729 #if IS_ENABLED(CONFIG_NF_NAT)
1730 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1731                                struct sk_buff *skb)
1732 {
1733         struct nlattr *start;
1734
1735         start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1736         if (!start)
1737                 return false;
1738
1739         if (info->nat & OVS_CT_SRC_NAT) {
1740                 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1741                         return false;
1742         } else if (info->nat & OVS_CT_DST_NAT) {
1743                 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1744                         return false;
1745         } else {
1746                 goto out;
1747         }
1748
1749         if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1750                 if (IS_ENABLED(CONFIG_NF_NAT) &&
1751                     info->family == NFPROTO_IPV4) {
1752                         if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1753                                             info->range.min_addr.ip) ||
1754                             (info->range.max_addr.ip
1755                              != info->range.min_addr.ip &&
1756                              (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1757                                               info->range.max_addr.ip))))
1758                                 return false;
1759                 } else if (IS_ENABLED(CONFIG_IPV6) &&
1760                            info->family == NFPROTO_IPV6) {
1761                         if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1762                                              &info->range.min_addr.in6) ||
1763                             (memcmp(&info->range.max_addr.in6,
1764                                     &info->range.min_addr.in6,
1765                                     sizeof(info->range.max_addr.in6)) &&
1766                              (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1767                                                &info->range.max_addr.in6))))
1768                                 return false;
1769                 } else {
1770                         return false;
1771                 }
1772         }
1773         if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1774             (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1775                          ntohs(info->range.min_proto.all)) ||
1776              (info->range.max_proto.all != info->range.min_proto.all &&
1777               nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1778                           ntohs(info->range.max_proto.all)))))
1779                 return false;
1780
1781         if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1782             nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1783                 return false;
1784         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1785             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1786                 return false;
1787         if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1788             nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1789                 return false;
1790 out:
1791         nla_nest_end(skb, start);
1792
1793         return true;
1794 }
1795 #endif
1796
1797 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1798                           struct sk_buff *skb)
1799 {
1800         struct nlattr *start;
1801
1802         start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1803         if (!start)
1804                 return -EMSGSIZE;
1805
1806         if (ct_info->commit && nla_put_flag(skb, ct_info->force
1807                                             ? OVS_CT_ATTR_FORCE_COMMIT
1808                                             : OVS_CT_ATTR_COMMIT))
1809                 return -EMSGSIZE;
1810         if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1811             nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1812                 return -EMSGSIZE;
1813         if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1814             nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1815                     &ct_info->mark))
1816                 return -EMSGSIZE;
1817         if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1818             labels_nonzero(&ct_info->labels.mask) &&
1819             nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1820                     &ct_info->labels))
1821                 return -EMSGSIZE;
1822         if (ct_info->helper) {
1823                 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1824                                    ct_info->helper->name))
1825                         return -EMSGSIZE;
1826         }
1827         if (ct_info->have_eventmask &&
1828             nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1829                 return -EMSGSIZE;
1830         if (ct_info->timeout[0]) {
1831                 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1832                         return -EMSGSIZE;
1833         }
1834
1835 #if IS_ENABLED(CONFIG_NF_NAT)
1836         if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1837                 return -EMSGSIZE;
1838 #endif
1839         nla_nest_end(skb, start);
1840
1841         return 0;
1842 }
1843
1844 void ovs_ct_free_action(const struct nlattr *a)
1845 {
1846         struct ovs_conntrack_info *ct_info = nla_data(a);
1847
1848         __ovs_ct_free_action(ct_info);
1849 }
1850
1851 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1852 {
1853         if (ct_info->helper) {
1854 #if IS_ENABLED(CONFIG_NF_NAT)
1855                 if (ct_info->nat)
1856                         nf_nat_helper_put(ct_info->helper);
1857 #endif
1858                 nf_conntrack_helper_put(ct_info->helper);
1859         }
1860         if (ct_info->ct) {
1861                 if (ct_info->timeout[0])
1862                         nf_ct_destroy_timeout(ct_info->ct);
1863                 nf_ct_tmpl_free(ct_info->ct);
1864         }
1865 }
1866
1867 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1868 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1869 {
1870         int i, err;
1871
1872         ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1873                                          GFP_KERNEL);
1874         if (!ovs_net->ct_limit_info)
1875                 return -ENOMEM;
1876
1877         ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1878         ovs_net->ct_limit_info->limits =
1879                 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1880                               GFP_KERNEL);
1881         if (!ovs_net->ct_limit_info->limits) {
1882                 kfree(ovs_net->ct_limit_info);
1883                 return -ENOMEM;
1884         }
1885
1886         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1887                 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1888
1889         ovs_net->ct_limit_info->data =
1890                 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1891
1892         if (IS_ERR(ovs_net->ct_limit_info->data)) {
1893                 err = PTR_ERR(ovs_net->ct_limit_info->data);
1894                 kfree(ovs_net->ct_limit_info->limits);
1895                 kfree(ovs_net->ct_limit_info);
1896                 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1897                 return err;
1898         }
1899         return 0;
1900 }
1901
1902 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1903 {
1904         const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1905         int i;
1906
1907         nf_conncount_destroy(net, NFPROTO_INET, info->data);
1908         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1909                 struct hlist_head *head = &info->limits[i];
1910                 struct ovs_ct_limit *ct_limit;
1911
1912                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1913                                          lockdep_ovsl_is_held())
1914                         kfree_rcu(ct_limit, rcu);
1915         }
1916         kfree(info->limits);
1917         kfree(info);
1918 }
1919
1920 static struct sk_buff *
1921 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1922                              struct ovs_header **ovs_reply_header)
1923 {
1924         struct ovs_header *ovs_header = info->userhdr;
1925         struct sk_buff *skb;
1926
1927         skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1928         if (!skb)
1929                 return ERR_PTR(-ENOMEM);
1930
1931         *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1932                                         info->snd_seq,
1933                                         &dp_ct_limit_genl_family, 0, cmd);
1934
1935         if (!*ovs_reply_header) {
1936                 nlmsg_free(skb);
1937                 return ERR_PTR(-EMSGSIZE);
1938         }
1939         (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1940
1941         return skb;
1942 }
1943
1944 static bool check_zone_id(int zone_id, u16 *pzone)
1945 {
1946         if (zone_id >= 0 && zone_id <= 65535) {
1947                 *pzone = (u16)zone_id;
1948                 return true;
1949         }
1950         return false;
1951 }
1952
1953 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1954                                        struct ovs_ct_limit_info *info)
1955 {
1956         struct ovs_zone_limit *zone_limit;
1957         int rem;
1958         u16 zone;
1959
1960         rem = NLA_ALIGN(nla_len(nla_zone_limit));
1961         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1962
1963         while (rem >= sizeof(*zone_limit)) {
1964                 if (unlikely(zone_limit->zone_id ==
1965                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1966                         ovs_lock();
1967                         info->default_limit = zone_limit->limit;
1968                         ovs_unlock();
1969                 } else if (unlikely(!check_zone_id(
1970                                 zone_limit->zone_id, &zone))) {
1971                         OVS_NLERR(true, "zone id is out of range");
1972                 } else {
1973                         struct ovs_ct_limit *ct_limit;
1974
1975                         ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1976                         if (!ct_limit)
1977                                 return -ENOMEM;
1978
1979                         ct_limit->zone = zone;
1980                         ct_limit->limit = zone_limit->limit;
1981
1982                         ovs_lock();
1983                         ct_limit_set(info, ct_limit);
1984                         ovs_unlock();
1985                 }
1986                 rem -= NLA_ALIGN(sizeof(*zone_limit));
1987                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1988                                 NLA_ALIGN(sizeof(*zone_limit)));
1989         }
1990
1991         if (rem)
1992                 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1993
1994         return 0;
1995 }
1996
1997 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1998                                        struct ovs_ct_limit_info *info)
1999 {
2000         struct ovs_zone_limit *zone_limit;
2001         int rem;
2002         u16 zone;
2003
2004         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2005         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2006
2007         while (rem >= sizeof(*zone_limit)) {
2008                 if (unlikely(zone_limit->zone_id ==
2009                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2010                         ovs_lock();
2011                         info->default_limit = OVS_CT_LIMIT_DEFAULT;
2012                         ovs_unlock();
2013                 } else if (unlikely(!check_zone_id(
2014                                 zone_limit->zone_id, &zone))) {
2015                         OVS_NLERR(true, "zone id is out of range");
2016                 } else {
2017                         ovs_lock();
2018                         ct_limit_del(info, zone);
2019                         ovs_unlock();
2020                 }
2021                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2022                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2023                                 NLA_ALIGN(sizeof(*zone_limit)));
2024         }
2025
2026         if (rem)
2027                 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2028
2029         return 0;
2030 }
2031
2032 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2033                                           struct sk_buff *reply)
2034 {
2035         struct ovs_zone_limit zone_limit;
2036
2037         zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2038         zone_limit.limit = info->default_limit;
2039
2040         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2041 }
2042
2043 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2044                                          struct nf_conncount_data *data,
2045                                          u16 zone_id, u32 limit,
2046                                          struct sk_buff *reply)
2047 {
2048         struct nf_conntrack_zone ct_zone;
2049         struct ovs_zone_limit zone_limit;
2050         u32 conncount_key = zone_id;
2051
2052         zone_limit.zone_id = zone_id;
2053         zone_limit.limit = limit;
2054         nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2055
2056         zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2057                                               &ct_zone);
2058         return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2059 }
2060
2061 static int ovs_ct_limit_get_zone_limit(struct net *net,
2062                                        struct nlattr *nla_zone_limit,
2063                                        struct ovs_ct_limit_info *info,
2064                                        struct sk_buff *reply)
2065 {
2066         struct ovs_zone_limit *zone_limit;
2067         int rem, err;
2068         u32 limit;
2069         u16 zone;
2070
2071         rem = NLA_ALIGN(nla_len(nla_zone_limit));
2072         zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2073
2074         while (rem >= sizeof(*zone_limit)) {
2075                 if (unlikely(zone_limit->zone_id ==
2076                                 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2077                         err = ovs_ct_limit_get_default_limit(info, reply);
2078                         if (err)
2079                                 return err;
2080                 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2081                                                         &zone))) {
2082                         OVS_NLERR(true, "zone id is out of range");
2083                 } else {
2084                         rcu_read_lock();
2085                         limit = ct_limit_get(info, zone);
2086                         rcu_read_unlock();
2087
2088                         err = __ovs_ct_limit_get_zone_limit(
2089                                 net, info->data, zone, limit, reply);
2090                         if (err)
2091                                 return err;
2092                 }
2093                 rem -= NLA_ALIGN(sizeof(*zone_limit));
2094                 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2095                                 NLA_ALIGN(sizeof(*zone_limit)));
2096         }
2097
2098         if (rem)
2099                 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2100
2101         return 0;
2102 }
2103
2104 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2105                                            struct ovs_ct_limit_info *info,
2106                                            struct sk_buff *reply)
2107 {
2108         struct ovs_ct_limit *ct_limit;
2109         struct hlist_head *head;
2110         int i, err = 0;
2111
2112         err = ovs_ct_limit_get_default_limit(info, reply);
2113         if (err)
2114                 return err;
2115
2116         rcu_read_lock();
2117         for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2118                 head = &info->limits[i];
2119                 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2120                         err = __ovs_ct_limit_get_zone_limit(net, info->data,
2121                                 ct_limit->zone, ct_limit->limit, reply);
2122                         if (err)
2123                                 goto exit_err;
2124                 }
2125         }
2126
2127 exit_err:
2128         rcu_read_unlock();
2129         return err;
2130 }
2131
2132 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2133 {
2134         struct nlattr **a = info->attrs;
2135         struct sk_buff *reply;
2136         struct ovs_header *ovs_reply_header;
2137         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2138         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2139         int err;
2140
2141         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2142                                              &ovs_reply_header);
2143         if (IS_ERR(reply))
2144                 return PTR_ERR(reply);
2145
2146         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2147                 err = -EINVAL;
2148                 goto exit_err;
2149         }
2150
2151         err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2152                                           ct_limit_info);
2153         if (err)
2154                 goto exit_err;
2155
2156         static_branch_enable(&ovs_ct_limit_enabled);
2157
2158         genlmsg_end(reply, ovs_reply_header);
2159         return genlmsg_reply(reply, info);
2160
2161 exit_err:
2162         nlmsg_free(reply);
2163         return err;
2164 }
2165
2166 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2167 {
2168         struct nlattr **a = info->attrs;
2169         struct sk_buff *reply;
2170         struct ovs_header *ovs_reply_header;
2171         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2172         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2173         int err;
2174
2175         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2176                                              &ovs_reply_header);
2177         if (IS_ERR(reply))
2178                 return PTR_ERR(reply);
2179
2180         if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2181                 err = -EINVAL;
2182                 goto exit_err;
2183         }
2184
2185         err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2186                                           ct_limit_info);
2187         if (err)
2188                 goto exit_err;
2189
2190         genlmsg_end(reply, ovs_reply_header);
2191         return genlmsg_reply(reply, info);
2192
2193 exit_err:
2194         nlmsg_free(reply);
2195         return err;
2196 }
2197
2198 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2199 {
2200         struct nlattr **a = info->attrs;
2201         struct nlattr *nla_reply;
2202         struct sk_buff *reply;
2203         struct ovs_header *ovs_reply_header;
2204         struct net *net = sock_net(skb->sk);
2205         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2206         struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2207         int err;
2208
2209         reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2210                                              &ovs_reply_header);
2211         if (IS_ERR(reply))
2212                 return PTR_ERR(reply);
2213
2214         nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2215         if (!nla_reply) {
2216                 err = -EMSGSIZE;
2217                 goto exit_err;
2218         }
2219
2220         if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2221                 err = ovs_ct_limit_get_zone_limit(
2222                         net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2223                         reply);
2224                 if (err)
2225                         goto exit_err;
2226         } else {
2227                 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2228                                                       reply);
2229                 if (err)
2230                         goto exit_err;
2231         }
2232
2233         nla_nest_end(reply, nla_reply);
2234         genlmsg_end(reply, ovs_reply_header);
2235         return genlmsg_reply(reply, info);
2236
2237 exit_err:
2238         nlmsg_free(reply);
2239         return err;
2240 }
2241
2242 static const struct genl_small_ops ct_limit_genl_ops[] = {
2243         { .cmd = OVS_CT_LIMIT_CMD_SET,
2244                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2245                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2246                                            * privilege. */
2247                 .doit = ovs_ct_limit_cmd_set,
2248         },
2249         { .cmd = OVS_CT_LIMIT_CMD_DEL,
2250                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2251                 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2252                                            * privilege. */
2253                 .doit = ovs_ct_limit_cmd_del,
2254         },
2255         { .cmd = OVS_CT_LIMIT_CMD_GET,
2256                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2257                 .flags = 0,               /* OK for unprivileged users. */
2258                 .doit = ovs_ct_limit_cmd_get,
2259         },
2260 };
2261
2262 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2263         .name = OVS_CT_LIMIT_MCGROUP,
2264 };
2265
2266 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2267         .hdrsize = sizeof(struct ovs_header),
2268         .name = OVS_CT_LIMIT_FAMILY,
2269         .version = OVS_CT_LIMIT_VERSION,
2270         .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2271         .policy = ct_limit_policy,
2272         .netnsok = true,
2273         .parallel_ops = true,
2274         .small_ops = ct_limit_genl_ops,
2275         .n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2276         .mcgrps = &ovs_ct_limit_multicast_group,
2277         .n_mcgrps = 1,
2278         .module = THIS_MODULE,
2279 };
2280 #endif
2281
2282 int ovs_ct_init(struct net *net)
2283 {
2284         unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2285         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2286
2287         if (nf_connlabels_get(net, n_bits - 1)) {
2288                 ovs_net->xt_label = false;
2289                 OVS_NLERR(true, "Failed to set connlabel length");
2290         } else {
2291                 ovs_net->xt_label = true;
2292         }
2293
2294 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2295         return ovs_ct_limit_init(net, ovs_net);
2296 #else
2297         return 0;
2298 #endif
2299 }
2300
2301 void ovs_ct_exit(struct net *net)
2302 {
2303         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2304
2305 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2306         ovs_ct_limit_exit(net, ovs_net);
2307 #endif
2308
2309         if (ovs_net->xt_label)
2310                 nf_connlabels_put(net);
2311 }